From 9f8ff5eb12e1ac572af407284f105783179c4187 Mon Sep 17 00:00:00 2001 From: Juan C Rodriguez Date: Wed, 14 Jul 2021 09:28:16 -0300 Subject: [PATCH] Merge Rahul's changes (#173) * Create optimization_ideas.Rmd (#156) Creates optimization_ideas vignette * DCE: examples and Idea section from GCI-2019 DCE: examples and Idea section from GCI-2019 * A PR combining the PRs of all Google Code-in students related to loop-invariate optimizer (#163) thanks! * GCI PR for constant folding optimizer (#164) Co-authored-by: Stephanie <20eristoffs@asij.ac.jp> Co-authored-by: Umang Majumder <54509868+ZeroDawn0D@users.noreply.github.com> Co-authored-by: udaydatar7 <54199456+udaydatar7@users.noreply.github.com> Co-authored-by: Eric Liu Co-authored-by: Pratish Co-authored-by: Rahul Saxena * GCI PRs for opt-common-subexpr (#165) thanks! Co-authored-by: Adrian Serapio Co-authored-by: azamlynny Co-authored-by: Umang Majumder <54509868+ZeroDawn0D@users.noreply.github.com> Co-authored-by: Jack-horwell <36800721+Jack-horwell@users.noreply.github.com> Co-authored-by: Pratish Co-authored-by: udaydatar7 <54199456+udaydatar7@users.noreply.github.com> Co-authored-by: Juan Cruz Rodriguez Co-authored-by: Rahul Saxena * All GCI PRs for constant propagation optimizer (#166) thanks! Co-authored-by: Pratish Co-authored-by: udaydatar7 <54199456+udaydatar7@users.noreply.github.com> Co-authored-by: Umang Majumder <54509868+ZeroDawn0D@users.noreply.github.com> Co-authored-by: Juan Cruz Rodriguez Co-authored-by: Rahul Saxena * All GCI PRs for dead expression optimizer. (#167) thanks! Co-authored-by: Adrian Serapio Co-authored-by: Umang Majumder <54509868+ZeroDawn0D@users.noreply.github.com> Co-authored-by: Pratish Co-authored-by: Eric Liu Co-authored-by: Rahul Saxena * All GCI PRs for Dead Store Optimizer (#168) thanks! Co-authored-by: Umang Majumder <54509868+ZeroDawn0D@users.noreply.github.com> Co-authored-by: Eric Liu Co-authored-by: Pratish Co-authored-by: Adrian Serapio Co-authored-by: Rishi R <56316487+Rishi0812@users.noreply.github.com> Co-authored-by: Rahul Saxena Co-authored-by: Juan Cruz Rodriguez * Conditional Thread Optimizer PR (#162) * Explained Dead Code Elimination * Update opt-dead-code.R * Update opt-dead-code.R * Create opt-dead-code-example.R * Update opt-dead-code.R * Added examples Examples of unoptimized and optimized code to explain dead code elimination * Added #' to example * Non-technical description of dead code elimination * Update to definition in ##Idea GCI user ID Rishi0812 * Added example of Dead Code Elimination * Added Idea Section to Dead Code Elimination Vignette * Updating NEWS.md * Changing \donttest for \dontrun, it seems 4.0.0 will check donttest examples. * Rebuilding site. * Removing renv. * Removing a separate example file for DCE This was introduced when I merged a PR from a student's fork of the `rco`. * Finalizing the Idea section of the DCE * Selected final examples for DCE * Final example selected * Resolved spacing issue * Resolved Spacing Issue - 2 * Resolved Spacing Issue - 3 * Improved the Idea section * Fixed the performance typo * Revert "Fixed the performance typo" This reverts commit c791bef18894110192fd05ba10e2c4c832d45ea7. * Update opt-dead-code.Rmd typo fix * Added a few functions * Added a few functions * Added logic for duplicate expr * Meh * Added merge logic * The entire process of merging implemented. Phhew!! * Ran into the else parse bug * First iteration complete * Final commit for cond_thread * Meh * Resolved duplicate vignette * Solved another issue in vignette build * I'm sleepy now, this is last * Missed a comma :( * I dunno what's happening anymore * Fixed the >= function logic * Last ditch * Ok I'm officially crying now T_T * Now no test would fail * Seriously, this is the last time * Bhak * erfg * Update parse.R * Update opt-constant-folding.R * Update opt-dead-code.R * Dude c'mmon * Update optimize-files.R * Update opt-dead-code.R * plixx * Added more tests to increase coverage * Another test for extra coverage * Implemented suggested changes * Some missing edits * Improved Tests * Fixed boolean condition * Improved indentations * documented * Workaround for R CMD check failed * Version bump & code styling * Examples edit * Fixed the indentation bug * Code Styiling * Provided another example in vignette * Improving vignette * Styling test cases * Fixed unwanted changes due to `;` * Ran all checks, works fine Co-authored-by: Pratish Co-authored-by: Ian Chang Co-authored-by: Jack-horwell <36800721+Jack-horwell@users.noreply.github.com> Co-authored-by: Rishi R <56316487+Rishi0812@users.noreply.github.com> Co-authored-by: Adrian Serapio Co-authored-by: Juan Cruz Rodriguez Co-authored-by: Rahul Saxena * Adding the Memory Allocation Optimizer (#169) * Initial Commit * Second commit - firm logic in place * First iteration of a working prototype * Just another test text * Tests and Vignettes for mem-alloc opt * Final edits in diff files for the new optimizer * Removed unnecessary files * Reverted changes in man/opt_dead * Reverted changes in man/opt_dead * Reverted changes in man/const_prop * Reverted changes in man/common_subexpr Co-authored-by: Rahul Saxena Co-authored-by: Juan C Rodriguez * Possible fix for Issue#107 (#170) * Possible fix for issue#107 * Reverting unrelated changes * Reverting unrelated changes * Reverting unrelated changes Co-authored-by: Rahul Saxena * Added the reporting functions (#171) * Added the reporting functions * Added comments Co-authored-by: Rahul Saxena * Please add this doc in the Article section of the official rco website (#172) * Please add this in the Article section of the official rco website * Fixed some blockquote issues * Trying new changes * Delete Document.dcf Co-authored-by: Rahul Saxena Co-authored-by: Juan C Rodriguez * Checking merge Co-authored-by: Rahul Saxena Co-authored-by: Stephanie <20eristoffs@asij.ac.jp> Co-authored-by: Umang Majumder <54509868+ZeroDawn0D@users.noreply.github.com> Co-authored-by: udaydatar7 <54199456+udaydatar7@users.noreply.github.com> Co-authored-by: Eric Liu Co-authored-by: Pratish Co-authored-by: Rahul Saxena Co-authored-by: Adrian Serapio Co-authored-by: azamlynny Co-authored-by: Jack-horwell <36800721+Jack-horwell@users.noreply.github.com> Co-authored-by: Rishi R <56316487+Rishi0812@users.noreply.github.com> Co-authored-by: Ian Chang --- NAMESPACE | 5 + NEWS.md | 5 + R/opt-common-subexpr.R | 8 + R/opt-cond-thread.R | 695 ++++++++++++++++++ R/opt-constant-propagation.R | 8 + R/opt-dead-expr.R | 29 +- R/opt-dead-store.R | 12 + R/opt-memory-alloc.R | 461 ++++++++++++ R/optimizers.R | 12 +- R/reporting-fun-files.R | 62 ++ R/reporting-fun-folder.R | 31 + R/reporting-fun-text.R | 16 + README.Rmd | 4 +- README.md | 42 +- _pkgdown.yml | 6 + docs/404.html | 13 +- docs/CODE_OF_CONDUCT.html | 13 +- docs/CONTRIBUTING.html | 13 +- docs/articles/contributing-an-optimizer.html | 131 ++-- .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/docker-readme.html | 34 +- .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/gsoc2020-final-report.html | 344 +++++++++ .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../cond_thread_thread_benchmark-1.png | Bin 0 -> 54612 bytes .../figure-html/pre_allocate_benchmark-1.png | Bin 0 -> 50907 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/index.html | 21 +- docs/articles/opt-common-subexpr.html | 79 +- .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../figure-html/unnamed-chunk-6-1.png | Bin 15016 -> 30928 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/opt-cond-thread.html | 256 +++++++ .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../figure-html/unnamed-chunk-4-1.png | Bin 0 -> 36788 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/opt-constant-folding.html | 95 ++- .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../figure-html/unnamed-chunk-4-1.png | Bin 14135 -> 34666 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/opt-constant-propagation.html | 129 ++-- .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../figure-html/unnamed-chunk-6-1.png | Bin 12959 -> 44030 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/opt-dead-code.html | 115 +-- .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../figure-html/unnamed-chunk-8-1.png | Bin 15627 -> 39933 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/opt-dead-expr.html | 89 ++- .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../figure-html/unnamed-chunk-5-1.png | Bin 14082 -> 34969 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/opt-dead-store.html | 164 +++-- .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../figure-html/unnamed-chunk-8-1.png | Bin 15209 -> 38322 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/opt-loop-invariant.html | 152 ++-- .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../figure-html/unnamed-chunk-5-1.png | Bin 15224 -> 32955 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/opt-memory-alloc.html | 255 +++++++ .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../figure-html/unnamed-chunk-4-1.png | Bin 0 -> 45642 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/articles/potential-optimizers.html | 365 +++++++++ .../anchor-sections-1.0/anchor-sections.css | 4 + .../anchor-sections-1.0/anchor-sections.js | 33 + .../figure-html/col_ext_benchmark-1.png | Bin 0 -> 49512 bytes .../inline_expansion_benchmark-1.png | Bin 0 -> 48960 bytes .../figure-html/pre_allocate_benchmark-1.png | Bin 0 -> 45396 bytes .../figure-html/val_ext_benchmark-1.png | Bin 0 -> 69002 bytes .../figure-html/vectorized_benchmark-1.png | Bin 0 -> 40570 bytes .../header-attrs-2.5/header-attrs.js | 12 + docs/authors.html | 13 +- docs/index.html | 187 ++--- docs/news/index.html | 66 +- docs/pkgdown.css | 4 +- docs/pkgdown.yml | 10 +- docs/reference/Rplot001.png | Bin 0 -> 1011 bytes docs/reference/all_optimizers.html | 21 +- docs/reference/generate_files_opt_report.html | 220 ++++++ .../reference/generate_folder_opt_report.html | 235 ++++++ docs/reference/generate_text_opt_report.html | 220 ++++++ docs/reference/index.html | 43 +- docs/reference/max_optimizers.html | 21 +- docs/reference/opt_common_subexpr.html | 37 +- docs/reference/opt_cond_thread.html | 246 +++++++ docs/reference/opt_constant_folding.html | 24 +- docs/reference/opt_constant_propagation.html | 36 +- docs/reference/opt_dead_code.html | 24 +- docs/reference/opt_dead_expr.html | 24 +- docs/reference/opt_dead_store.html | 43 +- docs/reference/opt_loop_invariant.html | 24 +- docs/reference/opt_memory_alloc.html | 231 ++++++ docs/reference/optimize_files.html | 25 +- docs/reference/optimize_folder.html | 29 +- docs/reference/optimize_text.html | 15 +- docs/reference/rco_gui.html | 22 +- man/all_optimizers.Rd | 4 +- man/generate_files_opt_report.Rd | 16 + man/generate_folder_opt_report.Rd | 27 + man/generate_text_opt_report.Rd | 16 + man/max_optimizers.Rd | 4 +- man/opt_common_subexpr.Rd | 8 + man/opt_cond_thread.Rd | 34 + man/opt_constant_propagation.Rd | 8 + man/opt_dead_store.Rd | 12 + man/opt_memory_alloc.Rd | 25 + tests/testthat/test-opt_cond_thread.R | 421 +++++++++++ tests/testthat/test-opt_dead_expr.R | 195 +++++ tests/testthat/test-opt_memory_alloc.R | 231 ++++++ vignettes/.gitignore | 1 + vignettes/gsoc2020-final-report.Rmd | 185 +++++ vignettes/opt-common-subexpr.Rmd | 4 + vignettes/opt-cond-thread.Rmd | 88 +++ vignettes/opt-constant-folding.Rmd | 4 + vignettes/opt-constant-propagation.Rmd | 4 + vignettes/opt-dead-code.Rmd | 4 + vignettes/opt-dead-expr.Rmd | 4 + vignettes/opt-dead-store.Rmd | 4 + vignettes/opt-loop-invariant.Rmd | 4 + vignettes/opt-memory-alloc.Rmd | 98 +++ vignettes/potential-optimizers.Rmd | 235 ++++++ 137 files changed, 7113 insertions(+), 644 deletions(-) create mode 100644 R/opt-cond-thread.R create mode 100644 R/opt-memory-alloc.R create mode 100644 R/reporting-fun-files.R create mode 100644 R/reporting-fun-folder.R create mode 100644 R/reporting-fun-text.R create mode 100644 docs/articles/contributing-an-optimizer_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/contributing-an-optimizer_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/contributing-an-optimizer_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/docker-readme_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/docker-readme_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/docker-readme_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/gsoc2020-final-report.html create mode 100644 docs/articles/gsoc2020-final-report_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/gsoc2020-final-report_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/gsoc2020-final-report_files/figure-html/cond_thread_thread_benchmark-1.png create mode 100644 docs/articles/gsoc2020-final-report_files/figure-html/pre_allocate_benchmark-1.png create mode 100644 docs/articles/gsoc2020-final-report_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/opt-common-subexpr_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/opt-common-subexpr_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/opt-common-subexpr_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/opt-cond-thread.html create mode 100644 docs/articles/opt-cond-thread_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/opt-cond-thread_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/opt-cond-thread_files/figure-html/unnamed-chunk-4-1.png create mode 100644 docs/articles/opt-cond-thread_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/opt-constant-folding_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/opt-constant-folding_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/opt-constant-folding_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/opt-constant-propagation_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/opt-constant-propagation_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/opt-constant-propagation_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/opt-dead-code_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/opt-dead-code_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/opt-dead-code_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/opt-dead-expr_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/opt-dead-expr_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/opt-dead-expr_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/opt-dead-store_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/opt-dead-store_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/opt-dead-store_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/opt-loop-invariant_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/opt-loop-invariant_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/opt-loop-invariant_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/opt-memory-alloc.html create mode 100644 docs/articles/opt-memory-alloc_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/opt-memory-alloc_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/opt-memory-alloc_files/figure-html/unnamed-chunk-4-1.png create mode 100644 docs/articles/opt-memory-alloc_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/articles/potential-optimizers.html create mode 100644 docs/articles/potential-optimizers_files/anchor-sections-1.0/anchor-sections.css create mode 100644 docs/articles/potential-optimizers_files/anchor-sections-1.0/anchor-sections.js create mode 100644 docs/articles/potential-optimizers_files/figure-html/col_ext_benchmark-1.png create mode 100644 docs/articles/potential-optimizers_files/figure-html/inline_expansion_benchmark-1.png create mode 100644 docs/articles/potential-optimizers_files/figure-html/pre_allocate_benchmark-1.png create mode 100644 docs/articles/potential-optimizers_files/figure-html/val_ext_benchmark-1.png create mode 100644 docs/articles/potential-optimizers_files/figure-html/vectorized_benchmark-1.png create mode 100644 docs/articles/potential-optimizers_files/header-attrs-2.5/header-attrs.js create mode 100644 docs/reference/Rplot001.png create mode 100644 docs/reference/generate_files_opt_report.html create mode 100644 docs/reference/generate_folder_opt_report.html create mode 100644 docs/reference/generate_text_opt_report.html create mode 100644 docs/reference/opt_cond_thread.html create mode 100644 docs/reference/opt_memory_alloc.html create mode 100644 man/generate_files_opt_report.Rd create mode 100644 man/generate_folder_opt_report.Rd create mode 100644 man/generate_text_opt_report.Rd create mode 100644 man/opt_cond_thread.Rd create mode 100644 man/opt_memory_alloc.Rd create mode 100644 tests/testthat/test-opt_cond_thread.R create mode 100644 tests/testthat/test-opt_memory_alloc.R create mode 100644 vignettes/gsoc2020-final-report.Rmd create mode 100644 vignettes/opt-cond-thread.Rmd create mode 100644 vignettes/opt-memory-alloc.Rmd create mode 100644 vignettes/potential-optimizers.Rmd diff --git a/NAMESPACE b/NAMESPACE index b19674c..efff424 100644 --- a/NAMESPACE +++ b/NAMESPACE @@ -1,14 +1,19 @@ # Generated by roxygen2: do not edit by hand export(all_optimizers) +export(generate_files_opt_report) +export(generate_folder_opt_report) +export(generate_text_opt_report) export(max_optimizers) export(opt_common_subexpr) +export(opt_cond_thread) export(opt_constant_folding) export(opt_constant_propagation) export(opt_dead_code) export(opt_dead_expr) export(opt_dead_store) export(opt_loop_invariant) +export(opt_memory_alloc) export(optimize_files) export(optimize_folder) export(optimize_text) diff --git a/NEWS.md b/NEWS.md index 504d3a3..832b6a2 100644 --- a/NEWS.md +++ b/NEWS.md @@ -1,3 +1,8 @@ +# rco 1.0.2 + + - Adding Memory Allocation optimizer. + - Adding Conditional Threading optimizer. + # rco 1.0.1 - Bug fixing. diff --git a/R/opt-common-subexpr.R b/R/opt-common-subexpr.R index 1d9cd3e..3e5acf1 100644 --- a/R/opt-common-subexpr.R +++ b/R/opt-common-subexpr.R @@ -16,6 +16,14 @@ #' sep = "\n" #' ) #' cat(opt_common_subexpr(list(code))$codes[[1]]) +#' +#' heron_formula <- paste( +#' "area <- (a/2 + b/2 +c/2) * (a/2 + b/2 + c/2 - a) * (a/2 + b/2 + c/2 - b) *", +#' " (a/2 + b/2 + c/2 - c)", +#' "area <- sqrt(area)", +#' sep = '\n' +#' ) +#' cat(opt_common_subexpr(list(heron_formula))$codes[[1]]) #' @export #' opt_common_subexpr <- function(texts, n_values = 2, in_fun_call = FALSE) { diff --git a/R/opt-cond-thread.R b/R/opt-cond-thread.R new file mode 100644 index 0000000..f0d58c6 --- /dev/null +++ b/R/opt-cond-thread.R @@ -0,0 +1,695 @@ +#' Optimizer: Conditional Threading. +#' +#' Performs one conditional threading pass. +#' Carefully examine the results after running this function! +#' +#' @param code A list of character vectors with the code to optimize. +#' +#' @examples +#' code <- paste( +#' "num <- sample(100, 1)", +#' "even_sum <- 0", +#' "odd_sum_a <- 0", +#' "odd_sum_b <- 0", +#' "if (num %% 2 == 1) {", +#' " odd_sum_a <- odd_sum_a + num", +#' "}", +#' "if (num %% 2 == 1) {", +#' " odd_num_b <- odd_num_b + num", +#' "}", +#' "if (!(num %% 2 == 1)) {", +#' " even_sum <- even_sum + num", +#' "}", +#' sep = "\n" +#' ) +#' cat(opt_cond_thread(list(code))$codes[[1]]) +#' @export + +opt_cond_thread <- function(code) { + res <- list() + res$codes <- lapply(code, ct_one_file) + return(res) +} + +# Executes conditional threading on one text of code. +# +# @param code A character vector with code to optimize. +# +ct_one_file <- function(code) { + parsed_dataset <- parse_text(code) + flatten_pd <- flatten_leaves(parsed_dataset) + result_flatten_pd <- ct_one_flatten_pd(parsed_dataset, flatten_pd) + return(deparse_data(result_flatten_pd)) +} + +# Checks if the given `node_id` is an `IF` statement +# +# @param fpd A flatten parsed data data.frame. +# @param node_id A numeric indicating the node ID of the function def expression. +# +check_if <- function(fpd, node_id) { + return("IF" %in% fpd[fpd$parent == node_id, "token"]) +} + +# Checks if the given `node_id` has consequent `IF` statements +# +# @param fpd A flatten parsed data data.frame. +# @param node_id A numeric indicating the node ID of the function def expression. +# @param exam_nodes A list consisting of all the nodes that needs to be checked for `IF` statements +# +check_if_next <- function(fpd, node_id, exam_nodes) { + if (check_if(fpd, node_id)) { + # Checks whether the given `node_id` is an `IF` statement itself + row_num <- which(exam_nodes$id == node_id, arr.ind = TRUE) + curr_parents <- get_children(fpd, node_id)$id + if (row_num != nrow(exam_nodes)) { + # Checks whether there exists another node after the given `node_id` in the `exam_nodes` + for (i in exam_nodes[-seq_len(row_num), "id"]) { + # Eliminating nodes with `node_id` and nodes before that + if (exam_nodes[exam_nodes$id == i, "parent"] >= 0 & !(i %in% curr_parents)) { + # Checking parent for positivity to be to able to avoid comments and second condition is to avoid nested `IF` + return(check_if(fpd, i)) + } + } + # If the function doesnot broke till now, return false + return(FALSE) + } + else { + # If the given node if the last node + return(FALSE) + } + } else { + # If the give node with node_id itself isn't an IF statement + return(FALSE) + } +} + +# Returns `id` of the `IF` condition of the node_id +# +# @param fpd A flatten parsed data data.frame. +# @param node_id A numeric indicating the node ID of the function def expression. +# +first_if_expr <- function(fpd, node_id) { + # `pos_id` of the keyword `IF` + if_pos <- get_children(fpd, node_id)[get_children(fpd, node_id)$token == "IF", "pos_id"] + # `pos_id` of the opening bracket of the `IF` condition + start_bracket <- get_children(fpd, node_id)[get_children(fpd, node_id)$pos_id > if_pos & + get_children(fpd, node_id)$token == "'('", "pos_id"][1] + # `parent` of the opening bracket + start_bracket_parent <- get_children(fpd, node_id)[get_children(fpd, node_id)$pos_id > if_pos & + get_children(fpd, node_id)$token == "'('", "parent"][1] + # `pos_id` of the closing bracket of the `IF` condition + end_bracket <- get_children(fpd, node_id)[get_children(fpd, node_id)$pos_id > if_pos & + get_children(fpd, node_id)$token == "')'" & + get_children(fpd, node_id)$parent == start_bracket_parent, "pos_id"] + # `id` of the expression found between the two brackets + if_expr_id <- get_children(fpd, node_id)[get_children(fpd, node_id)$pos_id > start_bracket & + get_children(fpd, node_id)$pos_id < end_bracket & + get_children(fpd, node_id)$token == "expr", "id"][1] + return(if_expr_id) +} + +# Used to find the condition of the consecutive IF block of the given node_id in the given fpd. +# +# @param fpd A flatten parsed data data.frame. +# @param node_id A numeric indicating the node ID of the function def expression. +# @param exam_nodes A list consisting of all the nodes that needs to be checked for `IF` statements +# +consecutive_if_expr <- function(fpd, node_id, exam_nodes) { + flag <- FALSE + # Find the row number of given node in `exam_nodes` + row_num <- which(exam_nodes$id == node_id, arr.ind = TRUE) + curr_parents <- get_children(fpd, node_id)$id + for (i in exam_nodes[-seq_len(row_num), "id"]) { + # Iterating for all nodes after the given node with the `node_id` + if (exam_nodes[exam_nodes$id == i, "parent"] >= 0 & !(i %in% curr_parents)) { + # Marking the immediate next node, that is not the sub-node of the given `node_id` + start_id <- i + flag <- TRUE + break + } + } + if (!flag) { # If no such node was found + return(flag) + } + + # Check for `IF` statement in the marked node + if ("IF" %in% get_children(fpd, start_id)$token) { + # If `IF` statement was found in this node too, find its `IF` condition too + first_if_expr(fpd, start_id) + } else { + return(FALSE) + } +} + +# Checks for exact negations in the conditions of the consecutive IF statements for the given node_id in the given fpd. +# For example: if(a) {//do A}; if(!a) {//do B} +# +# @param fpd A flatten parsed data data.frame. +# @param node1 A numeric indicating the node ID of the first IF expression. +# @param node2 A numeric indicating the node ID of the second IF expression. +# +check_negation <- function(fpd, node1, node2) { + # Assume second expr has the `!` symbol + first_expr_a <- fpd[fpd$id == node1, "text"] + second_expr_a <- fpd[fpd$id == node2, "text"] + + # Assume first expr has the `!` symbol + first_expr_b <- fpd[fpd$id == node2, "text"] + second_expr_b <- fpd[fpd$id == node1, "text"] + + # `gsub` is used to remove extra spaces from either expressions, + # so that difference does not arise due to spacing + + # The case when the second expr has the `!` symbol + check_first1a <- gsub(" ", "", paste("!", "(", first_expr_a, ")", sep = ""), fixed = TRUE) # Takes care of `if(!(a)) {}` + check_first2a <- gsub(" ", "", paste("!", first_expr_a, sep = ""), fixed = TRUE) # Takes care of `if(!a){}` + check_second_a <- gsub(" ", "", second_expr_a, fixed = TRUE) + + # The case when the first expr has the `!` symbol + check_first1b <- gsub(" ", "", paste("!", "(", first_expr_b, ")", sep = ""), fixed = TRUE) # Takes care of `if(!(a)){}` + check_first2b <- gsub(" ", "", paste("!", first_expr_b, sep = ""), fixed = TRUE) # Takes care of `if(!a){}` + check_second_b <- gsub(" ", "", second_expr_b, fixed = TRUE) + + return(check_first1a == check_second_a | check_first2a == check_second_a | check_first1b == check_second_b | check_first2b == check_second_b) +} + +# Checks for negations in the equality conditions of the consecutive IF statements for the given node_id in the given fpd. +# Example: if(a == b) {//do A} ; if(a != b) {// do B} +# +# @param fpd A flatten parsed data data.frame. +# @param node1 A numeric indicating the node ID of the first IF expression. +# @param node2 A numeric indicating the node ID of the second IF expression. +# +check_not_equal <- function(fpd, node1, node2) { + first_expr <- get_children(fpd, node1, include_father = FALSE) + second_expr <- get_children(fpd, node2, include_father = FALSE) + if (nrow(first_expr) != nrow(second_expr)) { + # This if condition ensures that the `IF` conditions are not too different + return(FALSE) + } + + # This for loop ensures that the only difference between the two conditions if of the token `EQ` and `NE` + for (i in nrow(first_expr)) { + if (first_expr[i, "text"] != second_expr[i, "text"]) { + if (!((first_expr[i, "token"] == "EQ" & second_expr[i, "token"] == "NE") | (first_expr[i, "token"] == "NE" & second_expr[i, "token"] == "EQ"))) { + return(FALSE) + } + } + } + return(TRUE) +} + +# Checks for exact duplicates conditions of the consecutive IF statements for the given node_id in the given fpd. +# Example: if(a) {//do A} if(a) {// do B} +# +# @param fpd A flatten parsed data data.frame. +# @param node1 A numeric indicating the node ID of the first IF expression. +# @param node2 A numeric indicating the node ID of the second IF expression. +# +check_duplicate_expr <- function(fpd, node1, node2) { + first_expr <- fpd[fpd$id == node1, "text"] + second_expr <- fpd[fpd$id == node2, "text"] + + # `gsub` is used to remove extra spaces from either expressions, + # so that difference does not arise due to spacing + first_expr <- gsub(" ", "", first_expr, fixed = TRUE) + second_expr <- gsub(" ", "", second_expr, fixed = TRUE) + + return(first_expr == second_expr) +} + +# Checks for exact negations in the greater than equal to logic in the given nodes +# Example if(a >= b) {// do A} ; if(a < b) {//do B} +# +# @param fpd A flatten parsed data data.frame. +# @param node1 A numeric indicating the node ID of the first IF expression. +# @param node2 A numeric indicating the node ID of the second IF expression. +# +check_comparsion_logic_ge <- function(fpd, node1, node2) { + first_expr <- fpd[fpd$id == node1, "text"] + second_expr <- fpd[fpd$id == node2, "text"] + + # Check for the GE symbol in the first_expr + if (length(grep(">=", first_expr)) > 0) { + # Remove spacing from first expression + first_expr <- gsub(" ", "", first_expr, fixed = T) + # Convert the `>=` symbol to `<` + first_expr <- gsub(">=", "<", first_expr) + # Remove spacing from second expression + second_expr <- gsub(" ", "", second_expr, fixed = TRUE) + # Compare first and second expressions + return(first_expr == second_expr) + } + # This is for the case when the GE symbol is in the second_expr + else if (length(grep(">=", second_expr)) > 0) { + second_expr <- gsub(" ", "", second_expr, fixed = TRUE) + first_expr <- gsub(" ", "", first_expr, fixed = TRUE) + second_expr <- gsub(">=", "<", second_expr) + return(first_expr == second_expr) + } + else { + # The case where no expression contains the GE symbol + return(FALSE) + } +} + +# Checks for exact negations in the lesser than equal to logic in the given nodes +# Example if(a <= b) {// do A} ; if(a > b) {//do B} +# +# @param fpd A flatten parsed data data.frame. +# @param node1 A numeric indicating the node ID of the first IF expression. +# @param node2 A numeric indicating the node ID of the second IF expression. +# +check_comparsion_logic_le <- function(fpd, node1, node2) { + first_expr <- fpd[fpd$id == node1, "text"] + second_expr <- fpd[fpd$id == node2, "text"] + # Check for the LE symbol in the first_expr + if (length(grep("<=", first_expr)) > 0) { + # Remove spacing from first expression + first_expr <- gsub(" ", "", first_expr, fixed = TRUE) + # Convert the `<=` symbol to `>` + first_expr <- gsub("<=", ">", first_expr) + # Remove spacing from first expression + second_expr <- gsub(" ", "", second_expr, fixed = TRUE) + # Compare first and second expressions + return(first_expr == second_expr) + } + # Check for the LE symbol in the second expression + else if (length(grep("<=", second_expr)) > 0) { + first_expr <- gsub(" ", "", first_expr, fixed = TRUE) + second_expr <- gsub(" ", "", second_expr, fixed = TRUE) + second_expr <- gsub("<=", ">", second_expr) + return(first_expr == second_expr) + } + else { + # Neither expression had the LE symbol + return(FALSE) + } +} + +# Checks for function calls in the IF statement conditions +# +# @param fpd A flatten parsed data.frame. +# @param node1 A numeric indicating the node ID of the first IF expression. +# @param node2 A numeric indicating the node ID of the second IF expression. +# +has_func_calls <- function(fpd, node1, node2) { + return("SYMBOL_FUNCTION_CALL" %in% get_children(fpd, node1)$token || + "SYMBOL_FUNCTION_CALL" %in% get_children(fpd, node2)) +} + +# Retrieves the `IF` block expression from the node index(itr) in exam_nodes +# +# @param fpd A flatten parsed data.frame +# @param exam_nodes A list of all nodes that need to be examined for the optimization +# @param itr The position of the iterator in exam_nodes +# +get_if_block_expr <- function(fpd, exam_nodes, itr) { + expr_id <- exam_nodes[itr, "id"] + + # This function is similar to `first_if_expr()` + if_pos <- get_children(fpd, expr_id)[get_children(fpd, expr_id)$token == "IF", "pos_id"] + start_bracket <- get_children(fpd, expr_id)[get_children(fpd, expr_id)$pos_id > if_pos & + get_children(fpd, expr_id)$token == "'('", "pos_id"][1] + start_bracket_parent <- get_children(fpd, expr_id)[get_children(fpd, expr_id)$pos_id > if_pos & + get_children(fpd, expr_id)$token == "'('", "parent"][1] + end_bracket <- get_children(fpd, expr_id)[get_children(fpd, expr_id)$pos_id > start_bracket & + get_children(fpd, expr_id)$token == "')'" & + get_children(fpd, expr_id)$parent == start_bracket_parent, "pos_id"] + + # Return the text of the node just after the closing bracket of `IF` statement + return(fpd[fpd$pos_id == (end_bracket + 1) & fpd$token == "expr", "text"]) +} + +# Retrieves the modified IF block expression from the index of node in modified exam_nodes +# +# @param fpd A flatten parsed data.frame +# @param exam_nodes A list of all nodes that need to be examined for the optimization +# @param index The position of the iterator in exam_nodes +# +get_modified_if_expr <- function(fpd, exam_nodes, index) { + # A new parsed data.frame for when the code-block under `IF` changes + modified_fpd <- flatten_leaves(parse_text(exam_nodes[index, "text"])) + + # Retrieve the code-block under the new `IF` statement + if_pos <- modified_fpd[modified_fpd$token == "IF", "pos_id"] + start_bracket <- modified_fpd[modified_fpd$pos_id > if_pos & + modified_fpd$token == "'('", "pos_id"][1] + start_bracket_parent <- modified_fpd[modified_fpd$pos_id > if_pos & + modified_fpd$token == "'('", "parent"][1] + end_bracket <- modified_fpd[modified_fpd$pos_id > start_bracket & + modified_fpd$token == "')'" & + modified_fpd$parent == start_bracket_parent, "pos_id"] + return(modified_fpd[modified_fpd$pos_id == (end_bracket + 1) & modified_fpd$token == "expr", "text"]) +} + +# Determines the nodes that have to be removed based on the index (itr) of the exam_nodes +# For example, if two IF statements are merged then the node of one IF must be removed from exam_nodes +# +# @param fpd A flatten parsed data.frame +# @param exam_nodes A list of all nodes that need to be examined for the optimization +# @param itr The position of the iterator in exam_nodes +# +node_removal_fun <- function(itr, exam_nodes, fpd) { + return_id <- res_list <- NULL + curr_id <- exam_nodes[itr, "id"] + curr_parents <- get_children(fpd, curr_id)$id + for (i in exam_nodes[-seq_len(itr), "id"]) { + if (exam_nodes[exam_nodes$id == i, "parent"] >= 0 & !(i %in% curr_parents)) { + return_id <- i + break + } + } + curr_parents2 <- get_children(fpd, return_id)$id + for (i in exam_nodes$id) { + if (i %in% curr_parents2) { + res_list <- append(res_list, which(i == exam_nodes$id, arr.ind = TRUE)) + } + } + rest_list <- NULL + itr_parents <- get_children(fpd, exam_nodes[itr, "id"])$id + for (i in exam_nodes$id) { + if (i %in% itr_parents & exam_nodes[exam_nodes$id == i, "parent"] >= 0) { + rest_list <- append(rest_list, which(i == exam_nodes$id, arr.ind = TRUE)) + } + } + rest_list <- rest_list[-1] + return(append(res_list, rest_list)) +} + + +# Executes conditional threading of a flatten_pd. +# +# @param flatten_pd A flatten parsed data data.frame. +# +ct_one_flatten_pd <- function(parsed_dataset, flatten_pd) { + pd <- parsed_dataset + fpd <- flatten_pd + # Procedure to include nodes from Functions into exam nodes + fun_ids <- fpd$id[fpd$token == "FUNCTION"] + fun_prnt_ids <- fpd$parent[fpd$id %in% fun_ids] + + fun_exam_nodes <- NULL + for (i in fun_prnt_ids) { + fun_exam_nodes <- rbind(fun_exam_nodes, get_children(fpd, i, FALSE)[get_children(fpd, i, FALSE)$token == "expr", ]) + } + + # Procedure to include nodes from Loops into exam nodes + loop_ids <- fpd[fpd$token %in% loops, "id"] + loop_parent_ids <- fpd[fpd$id %in% loop_ids, "parent"] + loop_exam_nodes <- NULL + for (i in loop_parent_ids) { + loop_exam_nodes <- rbind(loop_exam_nodes, get_children(fpd, i, FALSE)[get_children(fpd, i, FALSE)$token == "expr", ]) + } + + exam_nodes <- get_roots(pd) + exam_nodes <- rbind(exam_nodes, fun_exam_nodes) + exam_nodes <- rbind(exam_nodes, fun_exam_nodes, loop_exam_nodes) + + # Procedure to include nodes from nested IF loops + stray_nodes <- fpd[fpd$token == "IF", "parent"] + stray_exam_nodes <- NULL + for (i in stray_nodes) { + stray_exam_nodes <- rbind(stray_exam_nodes, fpd[fpd$id == i, ]) + } + for (i in stray_exam_nodes$id) { + if (!(i %in% exam_nodes$id)) { + exam_nodes <- rbind(exam_nodes, fpd[fpd$id == i, ]) + } + } + + # Removing duplicate nodes from exam_nodes + e_nodes <- NULL + for (i in unique(exam_nodes$id)) { + e_nodes <- rbind(e_nodes, unique(exam_nodes[exam_nodes$id == i, ])) + } + + exam_nodes <- e_nodes + exam_nodes_copy <- exam_nodes # Creating a copy as exam_nodes would be edited later + + # Inspection of the exam_nodes begin + + # First we will handle all the statements that have to be merged. + to_change_node <- to_remove_node <- merge_to <- merge_from <- NULL + + for (itr in seq_len(length(exam_nodes$id))) { + # For each node in exam_nodes, check if its an `IF` statement and if yes, then also check the immediate next node + i <- exam_nodes[itr, "id"] + if (check_if(fpd, i)) { + if (check_if_next(fpd, i, exam_nodes)) { + node1 <- first_if_expr(fpd, i) + node2 <- consecutive_if_expr(fpd, i, exam_nodes) + # Check that no `IF` statement conditions have a function call and then check whether the two IF statements are identical + if (!has_func_calls(fpd, node1, node2)) { + if (check_duplicate_expr(fpd, node1, node2)) { + # If found to be duplicate, make arrangements to merge them + merge_to <- append(merge_to, itr) + merge_from <- append(merge_from, node_removal_fun(itr, exam_nodes = exam_nodes, fpd = fpd)[1]) + to_change_node <- append(to_change_node, itr) + to_remove_node <- append(to_remove_node, node_removal_fun(itr, exam_nodes = exam_nodes, fpd = fpd)) + } + } + } + } + } + + # deletion_node contains all the nodes that need to be deleted after merging IF statements with duplicate conditions + deletion_nodes <- vector(mode = "numeric") + + for (i in to_remove_node) { + deletion_nodes <- append(deletion_nodes, exam_nodes[i, "id"]) + } + + for (i in to_change_node) { + deletion_nodes <- append(deletion_nodes, exam_nodes[i, "id"]) + } + + # The following sequence is used to change the text of exam_nodes text where merging has to take place + # to_expr refers to where the merging will take place and from_expr represents from where the code-block for merging will come + to_expr <- from_expr <- NULL + + for (i in seq_len(length(merge_from))) { + to_expr[i] <- get_if_block_expr(fpd, exam_nodes, merge_to[i]) + + if (grepl("{", to_expr[i], fixed = TRUE)) { + to_expr[i] <- gsub("{", "", to_expr[i], fixed = TRUE) + to_expr[i] <- gsub("}", "", to_expr[i], fixed = TRUE) + to_expr[i] <- trimws(to_expr[i]) + } + + from_expr[i] <- get_if_block_expr(fpd, exam_nodes, merge_from[i]) + + # Case where the IF code block starts from the same line + if (grepl("{", from_expr[i], fixed = TRUE)) { + from_expr[i] <- gsub("{", "", from_expr[i], fixed = TRUE) + from_expr[i] <- gsub("}", "", from_expr[i], fixed = TRUE) + from_expr[i] <- trimws(from_expr[i]) + } + } + + # Merging takes place here + for (i in seq_len(length(to_change_node))) { + if_cond <- fpd[fpd$id == first_if_expr(fpd, exam_nodes[to_change_node[i], "id"]), "text"] + string1 <- paste("if", "(", if_cond, ") ", "{\n", sep = "") + string2 <- paste(to_expr[i], "\n", from_expr[i], "\n}", sep = " ") + exam_nodes[to_change_node[i], "text"] <- paste0(string1, string2) + } + + # Removing the not-required nodes from exam_nodes + if (length(to_remove_node) > 0) { + exam_nodes <- exam_nodes[-(to_remove_node), ] + } + + # Here conversion of marked `IF` statements to `ELSE` takes place + to_change_node <- to_remove_node <- convert_to_else <- NULL + + for (itr in seq_len(length(exam_nodes$id))) { + # For each node in exam_nodes, check if its an `IF` statement and if yes, then also check the immediate next node + i <- exam_nodes[itr, "id"] + if (check_if(fpd, i)) { + if (check_if_next(fpd, i, exam_nodes)) { + node1 <- first_if_expr(fpd, i) + node2 <- consecutive_if_expr(fpd, i, exam_nodes) + if (!has_func_calls(fpd, node1, node2)) { + if (check_negation(fpd, node1, node2) | + check_not_equal(fpd, node1, node2) | + check_comparsion_logic_ge(fpd, node1, node2) | + check_comparsion_logic_le(fpd, node1, node2)) { + # Check all the conditions that can convert two `IF` statements to one `If-Else` block + convert_to_else <- append(convert_to_else, node_removal_fun(itr, exam_nodes = exam_nodes, fpd = fpd)[1]) + to_change_node <- append(to_change_node, itr) + to_remove_node <- append(to_remove_node, node_removal_fun(itr, exam_nodes = exam_nodes, fpd = fpd)) + } + } + } + } + } + + for (i in to_remove_node) { + deletion_nodes <- append(deletion_nodes, exam_nodes[i, "id"]) + } + + for (i in to_change_node) { + deletion_nodes <- append(deletion_nodes, exam_nodes[i, "id"]) + } + + else_expr <- NULL + + for (i in seq_len(length(convert_to_else))) { + else_expr[i] <- get_modified_if_expr(fpd, exam_nodes, convert_to_else[i]) + + if (grepl("{", else_expr[i], fixed = TRUE)) { + # For cases in which the statement starts from the same line + else_expr[i] <- gsub("{", "", else_expr[i], fixed = TRUE) + else_expr[i] <- gsub("}", "", else_expr[i], fixed = TRUE) + else_expr[i] <- trimws(else_expr[i]) + } + } + # Conversion of an `IF` statement to `ELSE` takes place here + for (i in seq_len(length(to_change_node))) { + string1 <- exam_nodes[to_change_node[i], "text"] + exam_nodes[to_change_node[i], "text"] <- paste0(string1, paste("else", "{\n", else_expr[i], " \n}", sep = " ")) + } + + # Removing the redundant nodes after conversion to else + if (length(to_remove_node) > 0) { + exam_nodes <- exam_nodes[-(to_remove_node), ] + } + + exam_nodes <- exam_nodes[order(exam_nodes$pos_id), ] + + # Segregating the nodes that are not to be edited + not_to_edit <- c() + + for (i in exam_nodes_copy$id) { + if (!(i %in% deletion_nodes)) { + not_to_edit <- rbind(not_to_edit, exam_nodes_copy[exam_nodes_copy$id == i, ]) + } + } + + # Tidying up the not_to_edit list + for (i in seq_len(nrow(not_to_edit))) { + not_to_edit <- rbind(not_to_edit, get_children(pd, not_to_edit[i, "id"])) + } + + not_to_edit <- not_to_edit[order(not_to_edit$pos_id), ] + + # Removing entries with duplicate IDs. + not_to_edit_final <- NULL + for (i in unique(not_to_edit$id)) { + not_to_edit_final <- rbind(not_to_edit_final, unique(not_to_edit[not_to_edit$id == i, ])) + } + + # Creating a list that consists of all the nodes to be changed + final_exam_nodes <- c() + for (i in to_change_node) { + final_node_id <- exam_nodes_copy[i, "id"] + final_exam_nodes <- rbind(final_exam_nodes, exam_nodes[exam_nodes$id == final_node_id, ]) + } + + # Creating new properties for the new_fpd + new_fpd <- NULL + if (length(final_exam_nodes$id) > 0) { + for (itr in seq_len(nrow(final_exam_nodes))) + { + act_fpd <- final_exam_nodes[itr, ] + new_act_fpd <- parse_text(act_fpd$text) + + # Setting new ids for the newly edited and parsed codes + new_act_fpd$id <- paste0(act_fpd$id, "_", new_act_fpd$id) + + # Keeping old parents for new fpd + new_act_fpd$parent[new_act_fpd$parent != 0] <- paste0(act_fpd$id, "_", new_act_fpd$parent[new_act_fpd$parent != 0]) + new_act_fpd$parent[new_act_fpd$parent == 0] <- act_fpd$parent + + # Calling a pre-wriiten rco::function.... + new_act_fpd$pos_id <- create_new_pos_id(act_fpd, nrow(new_act_fpd), act_fpd$id) + + # Fixing the next_spaces section of new_fpd + new_act_fpd$next_spaces[nrow(new_act_fpd)] <- act_fpd$next_spaces + + # Fixing the next_lines section of new_fpd + new_act_fpd$next_lines[nrow(new_act_fpd)] <- act_fpd$next_lines + + # Fixing the prev_spaces section of new_fpd + new_act_fpd$prev_spaces[which(new_act_fpd$terminal)[[1]]] <- act_fpd$prev_spaces + + # Merging the new_fpd and the act_fpd(obtained upon iteration) + new_fpd <- rbind(new_fpd, new_act_fpd) + + # Ordering the new_fpd according to the pos_id + new_fpd <- new_fpd[order(new_fpd$pos_id), ] + } + } + + resultant_fpd <- rbind(not_to_edit_final, new_fpd) + resultant_fpd <- resultant_fpd[order(resultant_fpd$pos_id), ] + + test_fpd <- flatten_leaves(resultant_fpd) + + # Removing the nodes that were in deletion_nodes + inspection_ids <- c() + remove_indices <- c() + for (i in deletion_nodes) { + inspection_ids <- get_children(fpd, i)$id + for (j in inspection_ids) { + if (length(which(test_fpd$id == j, arr.ind = T)) == 1) { + remove_indices <- append(remove_indices, which(test_fpd$id == j, arr.ind = T)) + } + } + } + + if (length(remove_indices) > 0) { + test_fpd <- test_fpd[-remove_indices, ] + } + + # Inserting appropriate values in the new_fpd + comments_ids <- NULL + curr_new_line_ids <- NULL + next_line_ids <- NULL + correction_nodes <- NULL + correction_ids <- NULL + + correction_nodes <- test_fpd[test_fpd$parent == 0 & is.na(test_fpd$next_spaces) & is.na(test_fpd$next_lines) & is.na(test_fpd$prev_spaces), ] + + for (i in correction_nodes$id) { + correction_ids <- append(correction_ids, which(test_fpd$id == i, arr.ind = TRUE)) + } + + test_fpd[which(is.na(test_fpd$next_lines)), "next_lines"] <- 0 + test_fpd[which(is.na(test_fpd$next_spaces)), "next_spaces"] <- 0 + test_fpd[which(is.na(test_fpd$prev_spaces)), "prev_spaces"] <- 0 + + comments_ids <- which(test_fpd$parent < 0) + correction_ids <- append(correction_ids, comments_ids) + + for (i in correction_ids) { + if ((i - 1) > 0) { + test_fpd[i - 1, "next_lines"] <- 1 + } else { + next + } + } + + #Copying the behaviour of "';'" from fpd to test_fpd + if("';'" %in% test_fpd$token) { + semiColon_indices <- which("';'" == fpd$token) + for(i in seq_len(length(semiColon_indices))) { + fpd_nextSpaces <- fpd[semiColon_indices[i], "next_spaces"] + fpd_nextLines <- fpd[semiColon_indices[i], "next_lines"] + fpd_prevSpaces <- fpd[semiColon_indices[i], "prev_spaces"] + + fpd_id <- fpd[semiColon_indices[i], "id"] + testFpd_index <- which(fpd_id == test_fpd$id) + + test_fpd[testFpd_index, "next_spaces"] <- fpd_nextSpaces + test_fpd[testFpd_index, "next_lines"] <- fpd_nextLines + test_fpd[testFpd_index, "prev_spaces"] <- fpd_prevSpaces + } + } + + test_fpd <- test_fpd[!(test_fpd$parent == 0 & test_fpd$terminal == FALSE), ] + + return(test_fpd) +} diff --git a/R/opt-constant-propagation.R b/R/opt-constant-propagation.R index c8aa3fd..c99e167 100644 --- a/R/opt-constant-propagation.R +++ b/R/opt-constant-propagation.R @@ -16,6 +16,14 @@ #' sep = "\n" #' ) #' cat(opt_constant_propagation(list(code))$codes[[1]]) +#' +#' hemisphere_vol <- paste( +#' "pi <- 3.141593 ", +#' "radius <- 25 ", +#' "hemis_vol <- 2/3 * pi * radius ^ 3 ", +#' sep = "\n" +#' ) +#' cat(opt_constant_propagation(list(hemisphere_vol))$codes[[1]]) #' @export #' opt_constant_propagation <- function(texts, in_fun_call = FALSE) { diff --git a/R/opt-dead-expr.R b/R/opt-dead-expr.R index 5603140..52c2e98 100644 --- a/R/opt-dead-expr.R +++ b/R/opt-dead-expr.R @@ -88,9 +88,18 @@ get_unassigned_exprs <- function(pd, id) { act_prnt_pd <- get_children(act_pd, act_parent) act_sblngs <- act_prnt_pd[act_prnt_pd$parent == act_parent, ] if (act_sblngs$token[[1]] == "'{'") { - new_visit <- c(new_visit, act_sblngs$id[act_sblngs$token == "expr"]) + new_visit <- c(new_visit, act_sblngs$id[act_sblngs$token == "expr" | act_sblngs$token == "exprlist"]) + } else if (act_sblngs$token[[1]] == "exprlist") { + exprlist_ids <- act_prnt_pd[act_prnt_pd$token == "exprlist", "id"] + for (exprlist_idx in exprlist_ids) { + separate_exprs <- act_prnt_pd[act_prnt_pd$parent == exprlist_idx & act_prnt_pd$token == "expr", "id"] + new_visit <- c(new_visit, separate_exprs) + } + new_visit <- unique(new_visit) + } else if (all(act_sblngs$token %in% c("expr", "';'"))) { + new_visit <- c(new_visit, act_sblngs[act_sblngs$token == "expr", "id"]) } else if (all(act_prnt_pd$token %in% - c(constants, ops, precedence_ops, "expr", "SYMBOL"))) { + c(constants, ops, precedence_ops, "expr", "exprlist","SYMBOL"))) { # it is an expression exprs_ids <- c(exprs_ids, act_parent) } else if (any(c(loops, "IF") %in% act_sblngs$token)) { @@ -121,6 +130,8 @@ get_unassigned_exprs <- function(pd, id) { act_pd$parent[act_pd$id == act_id], ] any(assigns %in% act_sblngs$token) # the expr is being assigned })] + + unique(exprs_ids) } # Returns the IDs of the exprs that can return in a function. @@ -142,10 +153,16 @@ get_fun_last_exprs <- function(pd, id) { act_sblngs <- act_prnt_pd[act_prnt_pd$parent == act_parent, ] if (act_sblngs$token[[1]] == "'{'") { # has multiple exprs, check only the last one - new_visit <- c( - new_visit, - utils::tail(act_sblngs$id[act_sblngs$token == "expr"], 1) - ) + if (act_sblngs[(nrow(act_sblngs)-1), "token"] == "exprlist") { + exprlist_id <- act_sblngs[(nrow(act_sblngs)-1), "id"] + exprlist_expr_ids <- act_prnt_pd[act_prnt_pd$parent == exprlist_id & act_prnt_pd$token == "expr", "id"] + new_visit <- c(new_visit, utils::tail(exprlist_expr_ids, n = 1)) + } else { + new_visit <- c( + new_visit, + utils::tail(act_sblngs$id[act_sblngs$token == "expr"], 1) + ) + } } else if (any(loops %in% act_sblngs$token)) { next } else if ("IF" %in% act_sblngs$token) { diff --git a/R/opt-dead-store.R b/R/opt-dead-store.R index c9f30d5..be7ef18 100644 --- a/R/opt-dead-store.R +++ b/R/opt-dead-store.R @@ -14,6 +14,18 @@ #' sep = "\n" #' ) #' cat(opt_dead_store(list(code))$codes[[1]]) +#' +#' code <- paste( +#' "sinpi <- function() {", +#' " pi <- 3.1415", +#' " e <- 2.718", +#' " phi <- 1.618", +#' " sin(pi)", +#' "}", +#' sep = "\n" +#' ) +#' cat(opt_dead_store(list(code))$codes[[1]]) +#' #' @export #' opt_dead_store <- function(texts) { diff --git a/R/opt-memory-alloc.R b/R/opt-memory-alloc.R new file mode 100644 index 0000000..fbf265f --- /dev/null +++ b/R/opt-memory-alloc.R @@ -0,0 +1,461 @@ +#' Optimizer: Memory Allocation. +#' +#' Performs one memory allocation pass. +#' Carefully examine the results after running this function! +#' +#' @param code A list of character vectors with the code to optimize. +#' +#' @examples +#' code <- paste( +#' "v <- NULL", +#' "for (i in 1:5) {", +#' " v[i] <- i^2", +#' "}", +#' sep = "\n" +#' ) +#' cat(opt_memory_alloc(list(code))$codes[[1]]) +#' @export + +opt_memory_alloc <- function(code) { + res <- list() + res$codes <- lapply(code, ma_one_file) + return(res) +} + +# Executes memory allocation on one text of code. +# +# @param code A character vector with code to optimize. +# +ma_one_file <- function(code) { + parsed_dataset <- parse_text(code) + flatten_pd <- flatten_leaves(parsed_dataset) + if(nrow(parsed_dataset) == 0) { + return (deparse_data(flatten_pd)) + } + result_flatten_pd <- ma_one_flatten_pd(parsed_dataset, flatten_pd) + return(deparse_data(result_flatten_pd)) +} + +# Restricts the operations of the optimizer to the body of a single `FOR` loop in code snippets containing nested `FOR` loops +# +# @param body_fpd A parsed data.frame containing the `body` of a `FOR` loop +# +restrict_body_to_one_loop <- function(body_fpd) { + list_of_fors <- body_fpd[body_fpd$token == "FOR", "pos_id"] + if(length(list_of_fors) > 1) { + body_fpd <- body_fpd[body_fpd$pos_id < list_of_fors[[2]], ] + } + body_fpd +} + +# Restricts the operations of the optimizer to the condition of a single `FOR` loop in code snippets containing nested `FOR` loops +# +# @param cond_fpd A parsed data.frame containing the `condition` of a `FOR` loop +# +restrict_condition_to_one_loop <- function(cond_fpd) { + list_of_conds <- cond_fpd[cond_fpd$token == "forcond", "pos_id"] + if(length(list_of_conds) > 1) { + cond_fpd <- cond_fpd[cond_fpd$pos_id < list_of_conds[[2]], ] + } + cond_fpd +} + +# Returns the name of the vector being assigned in the given expression. Returns FALSE if no vector is being assigned. +# +# @param node_id A numeral indicating the node ID of the expression from which the vector's name is to be extracted +# @param fpd A flattened parsed data.frame of the original code snippet given +# +extract_vector_name <- function(fpd, node_id) { + assignment_pd <- get_children(fpd, node_id) + assignment_index <- which(assignment_pd$token %in% assigns) + if(length(assignment_index) > 0) { + assignment_type <- assignment_pd[assignment_index, "token"] + if(assignment_type == "EQ_ASSIGN" || assignment_type == "LEFT_ASSIGN") { + vector_name <- assignment_pd[(assignment_index - 1), "text"] + } else { + vector_name <- assignment_pd[(assignment_index + 1), "text"] + } + return (vector_name) + } else { + return (FALSE) + } +} + +# Checks whether a number is explicitly mentioned in the `condition` of a `FOR` loop. +# Returns the number mentioned if present, else returns `FALSE` +# +# @param cond_fpd A parsed data.frame containing the `condition` of a `FOR` loop +# @param index_name A string representing the name of the iterator of the `FOR` loop +# @param in_pos A numeric indicating the `pos_id` of `in` from the condition of a `FOR` loop +# +check_forcond_num_declaration <- function(cond_fpd, index_name, in_pos) { + forcond_number_flag <- FALSE + if("':'" %in% cond_fpd[cond_fpd$pos_id > in_pos, "token"]) { + colon_pos <- which(cond_fpd$token == "':'", arr.ind = TRUE) + colon_pos_id <- cond_fpd[colon_pos, "pos_id"] + if("NUM_CONST" %in% cond_fpd[cond_fpd$pos_id > colon_pos_id, "token"]) { + forcond_number_flag <- TRUE + cond_fpd <- cond_fpd[cond_fpd$pos_id > colon_pos_id, ] + memory_alloc_number <- as.integer(cond_fpd[cond_fpd$token == "NUM_CONST", "text"]) + } + } + else if("SYMBOL_FUNCTION_CALL" %in% cond_fpd[cond_fpd$pos_id > in_pos & + cond_fpd$text == "seq_len", "token"]) { + seqLen_pos_id <- cond_fpd[cond_fpd$pos_id > in_pos & + cond_fpd$token == "SYMBOL_FUNCTION_CALL" & + cond_fpd$text == "seq_len", "pos_id"] + if("NUM_CONST" %in% cond_fpd[cond_fpd$pos_id > seqLen_pos_id, "token"]) { + forcond_number_flag <- TRUE + cond_fpd <- cond_fpd[cond_fpd$pos_id > seqLen_pos_id, ] + memory_alloc_number <- as.integer(cond_fpd[cond_fpd$token == "NUM_CONST", "text"]) + } + forcond_number_flag <- FALSE + } + else if("SYMBOL_FUNCTION_CALL" %in% cond_fpd[cond_fpd$pos_id > in_pos & + cond_fpd$text == "c", "token"]) { + c_pos_id <- cond_fpd[cond_fpd$pos_id > in_pos & + cond_fpd$token == "SYMBOL_FUNCTION_CALL" & + cond_fpd$text == "c", "pos_id"] + if("NUM_CONST" %in% cond_fpd[cond_fpd$pos_id > c_pos_id, "token"]) { + forcond_number_flag <- TRUE + cond_fpd <- cond_fpd[cond_fpd$pos_id > c_pos_id, ] + memory_alloc_number <- 0 + for(i in cond_fpd$token) { + if(i == "NUM_CONST") { + memory_alloc_number = memory_alloc_number + 1 + } + } + } + forcond_number_flag <- FALSE + } + if(forcond_number_flag == TRUE) { + return (memory_alloc_number) + } else { + return (FALSE) + } +} + +# Checks for presence of functions in the body of the `FOR` loop +# +# @param body_fpd A parsed data.frame containing the `body` of a `FOR` loop +# +check_fun_call <- function(body_fpd) { + return ("SYMBOL_FUNCTION_CALL" %in% body_fpd$token) +} + +# Checks whether the index is explicitly mentioned at the time of assignment inside the loop +# Returns list of vectors that have the index, if none are present empty list is returned +# +# @param body_fpd A parsed data.frame containing the `body` of a `FOR` loop +# @param index_name A string representing the name of the iterator of the `FOR` loop +# +check_index_assignment <- function(body_fpd, index_name) { + assign_operator_list <- list() + assign_operator_list <- body_fpd[body_fpd$token %in% assigns, ] + check_index_assignment_list <- failed_list <- NULL + for(i in seq_len(length(assign_operator_list$id))) { + check_pos_id <- assign_operator_list[i, "pos_id"] + check_assign_op <- assign_operator_list[i, "token"] + check_parent_id <- assign_operator_list[i, "parent"] + + open_bracket_flag <- close_bracket_flag <- FALSE + name_of_vector <- vector(mode = "character", length = 1) + ##If = or <- is used, then we will search the expr before the assignment operator + if(check_assign_op == "EQ_ASSIGN" | check_assign_op == "LEFT_ASSIGN") { + assigned_vector_node_id <- body_fpd[body_fpd$token == "expr" & + body_fpd$parent == check_parent_id & + body_fpd$pos_id < check_pos_id, "id"] + + if(length(assigned_vector_node_id) == 0) { + return (check_index_assignment_list) + } + + parsed_assigned_expr <- get_children(body_fpd, assigned_vector_node_id) + + index_name_idx <- which(parsed_assigned_expr$text == index_name & + parsed_assigned_expr$token == "SYMBOL", arr.ind = TRUE) + + open_bracket_flag <- (parsed_assigned_expr[(index_name_idx - 1), "text"] == "[") + close_bracket_flag <- (parsed_assigned_expr[(index_name_idx + 1), "text"] == "]") + + open_bracket_index <- which(parsed_assigned_expr$token == "'['", arr.ind = TRUE) + name_of_vector <- parsed_assigned_expr[(open_bracket_index-1), "text"] + + } else { + assigned_vector_node_id <- body_fpd[body_fpd$token == "expr" & + body_fpd$parent == check_parent_id & + body_fpd$pos_id > check_pos_id, "id"] + + if(length(assigned_vector_node_id) == 0) { + return (check_index_assignment_list) + } + + parsed_assigned_expr <- get_children(body_fpd, assigned_vector_node_id) + + index_name_idx <- which(parsed_assigned_expr$text == index_name & + parsed_assigned_expr$token == "SYMBOL", arr.ind = TRUE) + + open_bracket_flag <- (parsed_assigned_expr[(index_name_idx - 1), "text"] == "[") + close_bracket_flag <- (parsed_assigned_expr[(index_name_idx + 1), "text"] == "]") + + open_bracket_index <- which(parsed_assigned_expr$token == "'['", arr.ind = TRUE) + name_of_vector <- parsed_assigned_expr[(open_bracket_index-1), "text"] + } + + if(open_bracket_flag & close_bracket_flag) { + check_index_assignment_list <- c(check_index_assignment_list, name_of_vector) + } else { + check_index_assignment_list <- c(check_index_assignment_list, name_of_vector) + failed_list <- c(failed_list, name_of_vector) + } + } + return (unique(check_index_assignment_list[!(check_index_assignment_list %in% failed_list)])) +} + +ma_one_flatten_pd <- function(parsed_dataset, flatten_pd) { + pd <- parsed_dataset + fpd <- flatten_pd + fpd <- eq_assign_to_expr(fpd) ## eq_assign_to_expr helps to convert the `EQ_ASSIGN` token to `expr` + + ## taking a different approach from simply calling get_children(fpd) and including all `expr`. + exam_nodes <- fpd[fpd$token == "expr", ] + exam_nodes <- rbind(exam_nodes, get_roots(fpd)) + exam_nodes <- unique(exam_nodes) + + ## custom_exam_nodes + ## custom_exam_nodes will contain only loops and the assignment nodes from the entire exam_nodes + null_parents <- fpd[fpd$token == "NULL_CONST", "parent"] + na_parents <- fpd[fpd$token == "NUM_CONST" & fpd$text == "NA", "parent"] + init_list <- c("c()", "numeric()", "logical()", "double()", "factor()", "integer()") + alternate_initialization_parents <- fpd[fpd$token == "expr" & fpd$text %in% init_list, "parent"] + + null_parents <- c(null_parents, c(alternate_initialization_parents, na_parents)) + + edit_nodes_list <- NULL + for(i in seq_len(length(alternate_initialization_parents))) { + alt_init_pd <- get_children(fpd, alternate_initialization_parents[i], FALSE) + expr_nodes <- alt_init_pd[alt_init_pd$token == "expr", "id"] + edit_nodes_list <- c(edit_nodes_list, expr_nodes) + } + + custom_exam_nodes <- NULL + for(i in null_parents) { + custom_exam_nodes <- rbind(custom_exam_nodes, fpd[fpd$id == i, ]) + } + for(i in exam_nodes$id) { + if(nrow(get_children(fpd, i)) > 1) { + if(get_children(fpd, i, FALSE)[1, "token"] == "FOR") { + custom_exam_nodes <- rbind(custom_exam_nodes, fpd[fpd$id == i, ]) + } + } + } + # Only order a *custom_exam_nodes* that is not NULL + if(!(is.null(custom_exam_nodes))) { + custom_exam_nodes <- custom_exam_nodes[order(custom_exam_nodes$pos_id), ] + } + + ## Here we analyse all the nodes of custom_exam_nodes that consists of assignemnts and FOR loops + vector_initialization_list <- list() + initializations_to_change <- NULL + vectors_memory_list <- list() + + ## if condition handles the loops from custom_exam_nodes and the else part handles the vector assignment from custom_exam_nodes + for(i in seq_len(length(custom_exam_nodes$id))) { + if(get_children(fpd, custom_exam_nodes[i, "id"], F)[1, "token"] == "FOR") { + i <- custom_exam_nodes[i, "id"] + sub_fpd <- get_children(fpd, i) + + cond_fpd <- get_children(sub_fpd, sub_fpd[sub_fpd$token=="forcond", "id"]) + cond_fpd <- restrict_condition_to_one_loop(cond_fpd) + + body_fpd <- sub_fpd[!(sub_fpd$id %in% cond_fpd$id), ] + body_fpd <- restrict_body_to_one_loop(body_fpd) + + in_pos <- cond_fpd[cond_fpd$token == "IN", "pos_id"] + index_name <- cond_fpd[cond_fpd$pos_id < in_pos & + cond_fpd$token == "SYMBOL", "text"] + + ## Checking condition 1: No function calls inside the loop, to counter the possibility of use of `<<-` in that particular function + fun_call_flag <- check_fun_call(body_fpd) + + ## Checking condition 2: A number being explicitly mentioned in the for condition, and isolating that number if condition is true + forcond_number_flag <- check_forcond_num_declaration(cond_fpd, index_name, in_pos) + if(typeof(forcond_number_flag) != "logical") { + memory_alloc_num <- forcond_number_flag + } + + ## Checking condition 3: The index should be mentioned specifically of the vector to which something is being assigned + passing_vectors <- check_index_assignment(body_fpd, index_name) + index_assignment_flag <- (length(passing_vectors) == 0) + + if(fun_call_flag | typeof(forcond_number_flag) == "logical" | index_assignment_flag) { + next + } else { + initializations_to_change <- c(initializations_to_change, passing_vectors) + initializations_to_change <- unique(initializations_to_change) + for(vec in passing_vectors) { + if(vec %in% names(vectors_memory_list)) { + if(vectors_memory_list[[vec]] < memory_alloc_num) { + vectors_memory_list[[vec]] <- memory_alloc_num + } else { + next + } + } else { + vectors_memory_list[[vec]] <- memory_alloc_num + } + } + } + } else{ + # Here we associate all the vectors (of the same name too) with the IDs with which they were initialized. + initialized_vector_name <- extract_vector_name(fpd, custom_exam_nodes[i, "id"]) + if(initialized_vector_name %in% names(vector_initialization_list)) { + vector_initialization_list[[initialized_vector_name]] <- c(vector_initialization_list[[initialized_vector_name]], custom_exam_nodes[i, "id"]) + } else { + vector_initialization_list[[initialized_vector_name]] <- custom_exam_nodes[i, "id"] + } + } + } + +# Now the id of the nodes that have to be edited is stored in vector_initialization_list, and when we remove those nodes from exam_nodes we have the not_to_edit fpd + + for(i in seq_len(length(vector_initialization_list))) { + if(names(vector_initialization_list)[i] %in% names(vectors_memory_list)) { + edit_nodes_list <- c(edit_nodes_list, vector_initialization_list[[i]]) + } + } + + not_to_edit <- exam_nodes[!(exam_nodes$id %in% edit_nodes_list), ] + not_to_edit_copy <- not_to_edit + + for(i in seq_len(length(not_to_edit_copy$id))) { + not_to_edit <- rbind(not_to_edit, get_children(fpd, not_to_edit_copy[i, "id"])) + } + + not_to_edit <- not_to_edit[order(not_to_edit$pos_id), ] + + not_to_edit <- unique(not_to_edit) + +# Creating the final_exam_nodes list from the `vector_initialization_list` and `vectors_memory_list` #### + + final_exam_nodes <- exam_nodes[exam_nodes$id %in% edit_nodes_list, ] + + to_remove_indices <- NULL + + ## Now we iterate over all the entries of final_exam_nodes and change their text + for(i in seq_len(length(final_exam_nodes$id))) { + name_of_vector <- extract_vector_name(fpd, final_exam_nodes[i, "id"]) + if(typeof(name_of_vector) == "logical") { + to_remove_indices <- c(to_remove_indices, i) + } else { + final_exam_nodes[i, "text"] <- sprintf("%s <- vector(length = %d)", name_of_vector, vectors_memory_list[[name_of_vector]]) + } + } + if(!(is.null(to_remove_indices))) { + final_exam_nodes <- final_exam_nodes[-(to_remove_indices), ] + } + +# Merging the final_exam_nodes and the not_to_edit and subsequent cleaning operations #### + +final_exam_nodes_ids <- final_exam_nodes$id + + new_fpd <- NULL + for(itr in seq_len(length(final_exam_nodes$id))) { + act_fpd <- final_exam_nodes[itr, ] + new_act_fpd <- parse_text(act_fpd$text) + + #Setting new ids for the newly edited and parsed codes + new_act_fpd$id <- paste0(act_fpd$id, "_", new_act_fpd$id) + + #Keeping old parents for new fpd + new_act_fpd$parent[new_act_fpd$parent != 0] <- paste0(act_fpd$id, "_", new_act_fpd$parent[new_act_fpd$parent != 0]) + new_act_fpd$parent[new_act_fpd$parent == 0] <- act_fpd$parent + + #Calling a pre-wriiten rco::function.... + new_act_fpd$pos_id <- create_new_pos_id(act_fpd, nrow(new_act_fpd), act_fpd$id) + + #Fixing the next_spaces section of new_fpd + new_act_fpd$next_spaces[nrow(new_act_fpd)] <- act_fpd$next_spaces + + #Fixing the next_lines section of new_fpd + new_act_fpd$next_lines[nrow(new_act_fpd)] <- act_fpd$next_lines + + #Fixing the prev_spaces section of new_fpd + new_act_fpd$prev_spaces[which(new_act_fpd$terminal)[[1]]] <- act_fpd$prev_spaces + + #Merging the new_fpd and the act_fpd(obtained upon iteration) + new_fpd <- rbind(new_fpd, new_act_fpd) + + #Ordering the new_fpd according to the pos_id + new_fpd <- new_fpd[order(new_fpd$pos_id), ] + } + +# Final Steps of molding the not_to_edit and final_exam_nodes into the same fpd #### + + resultant_fpd <- rbind(not_to_edit, new_fpd) + resultant_fpd <- resultant_fpd[order(resultant_fpd$pos_id), ] + + test_fpd <- flatten_leaves(resultant_fpd) + + comments_ids <- curr_new_line_ids <- next_line_ids <- correction_nodes <- correction_ids <- NULL + + correction_nodes <- test_fpd[test_fpd$parent == 0 & is.na(test_fpd$next_spaces) & is.na(test_fpd$next_lines) & is.na(test_fpd$prev_spaces), ] + + for(i in correction_nodes$id){ + correction_ids <- append(correction_ids, which(test_fpd$id == i, arr.ind = TRUE)) + } + + test_fpd[which(is.na(test_fpd$next_lines)), "next_lines"] <- 0 + test_fpd[which(is.na(test_fpd$next_spaces)), "next_spaces"] <- 0 + test_fpd[which(is.na(test_fpd$prev_spaces)), "prev_spaces"] <- 0 + + comments_ids <- which(test_fpd$parent < 0) + correction_ids <- append(correction_ids, comments_ids) + + for(i in correction_ids){ + if((i-1) > 0){ + test_fpd[i-1, "next_lines"] <- 1 + }else{ + next + } + } + + test_fpd <- test_fpd[!(test_fpd$parent == 0 & test_fpd$terminal == FALSE), ] + deletion_nodes <- NULL + for(i in final_exam_nodes_ids){ + if(i %in% test_fpd$id){ + deletion_nodes <- rbind(deletion_nodes, get_children(test_fpd, i)) + } + } + + next_line_ids <- as.double(test_fpd[test_fpd$id %in% deletion_nodes$id & test_fpd$next_lines == 1, "pos_id"]) + + test_fpd <- test_fpd[!(test_fpd$id %in% deletion_nodes$id), ] + + for(i in next_line_ids){ + curr_new_line_ids <- append(curr_new_line_ids, + max(which(test_fpd$pos_id < i))) + } + + for (i in curr_new_line_ids) { + test_fpd[i, ]$next_lines <- 1 + } + + #Copying the behaviour of "';'" from fpd to test_fpd + if("';'" %in% test_fpd$token) { + semiColon_indices <- which("';'" == fpd$token) + for(i in seq_len(length(semiColon_indices))) { + fpd_nextSpaces <- fpd[semiColon_indices[i], "next_spaces"] + fpd_nextLines <- fpd[semiColon_indices[i], "next_lines"] + fpd_prevSpaces <- fpd[semiColon_indices[i], "prev_spaces"] + + fpd_id <- fpd[semiColon_indices[i], "id"] + testFpd_index <- which(fpd_id == test_fpd$id) + + test_fpd[testFpd_index, "next_spaces"] <- fpd_nextSpaces + test_fpd[testFpd_index, "next_lines"] <- fpd_nextLines + test_fpd[testFpd_index, "prev_spaces"] <- fpd_prevSpaces + } + } + + return(test_fpd) +} diff --git a/R/optimizers.R b/R/optimizers.R index a00e23f..fd68b05 100644 --- a/R/optimizers.R +++ b/R/optimizers.R @@ -2,6 +2,7 @@ #' #' List of all the optimizer functions: #' \itemize{ +#' \item Conditional Threading \code{\link{opt_cond_thread}} #' \item Constant Folding \code{\link{opt_constant_folding}} #' \item Constant Propagation \code{\link{opt_constant_propagation}} #' \item Dead Code Elimination \code{\link{opt_dead_code}} @@ -9,18 +10,21 @@ #' \item Dead Expression Elimination \code{\link{opt_dead_expr}} #' \item Common Subexpression Elimination \code{\link{opt_common_subexpr}} #' \item Loop-invariant Code Motion \code{\link{opt_loop_invariant}} +#' \item Memory Allocation \code{\link{opt_memory_alloc}} #' } #' #' @export #' all_optimizers <- list( + "Conditional Threading" = opt_cond_thread, "Constant Folding" = opt_constant_folding, "Constant Propagation" = opt_constant_propagation, "Dead Code Elimination" = opt_dead_code, "Dead Store Elimination" = opt_dead_store, "Dead Expression Elimination" = opt_dead_expr, "Common Subexpression Elimination" = opt_common_subexpr, - "Loop-invariant Code Motion" = opt_loop_invariant + "Loop-invariant Code Motion" = opt_loop_invariant, + "Memory Allocation" = opt_memory_alloc ) #' Max optimizers list. @@ -29,6 +33,7 @@ all_optimizers <- list( #' enabled. #' Note that using this optimizers could change the semantics of the program! #' \itemize{ +#' \item Conditional Threading \code{\link{opt_cond_thread}} #' \item Constant Folding \code{\link{opt_constant_folding}} #' \item Constant Propagation \code{\link{opt_constant_propagation}} #' \item Dead Code Elimination \code{\link{opt_dead_code}} @@ -36,11 +41,13 @@ all_optimizers <- list( #' \item Dead Expression Elimination \code{\link{opt_dead_expr}} #' \item Common Subexpression Elimination \code{\link{opt_common_subexpr}} #' \item Loop-invariant Code Motion \code{\link{opt_loop_invariant}} +#' \item Memory Allocation \code{\link{opt_memory_alloc}} #' } #' #' @export #' max_optimizers <- list( + "Conditional Threading" = opt_cond_thread, "Constant Folding" = function(texts) { opt_constant_folding(texts, fold_floats = TRUE, in_fun_call = TRUE) }, @@ -53,5 +60,6 @@ max_optimizers <- list( "Common Subexpression Elimination" = function(texts) { opt_common_subexpr(texts, in_fun_call = TRUE) }, - "Loop-invariant Code Motion" = opt_loop_invariant + "Loop-invariant Code Motion" = opt_loop_invariant, + "Memory Allocation" = opt_memory_alloc ) diff --git a/R/reporting-fun-files.R b/R/reporting-fun-files.R new file mode 100644 index 0000000..70d2b83 --- /dev/null +++ b/R/reporting-fun-files.R @@ -0,0 +1,62 @@ +#' Report possible optimizations in `.R` files. +#' +#' @param files A character vector with paths to files to optimize. +#' @param optimizers A named list of optimizer functions. +#' +#' @export +#' +generate_files_opt_report <- function(files, optimizers = rco:::all_optimizers) { + # An introductory message, as the function could take quite some time depending on the number and size of files + message("Sit back, this may take some time...") + # Initializations + final_result <- codes <- list() + # Read the contents of a file. 3 files will form a list/vector/df of size 3 + codes <- lapply(files, read_code_file) + # Naming each entry in the list + names(codes) <- files + # Setting up the progress bar + pb_outer <- utils::txtProgressBar(1, length(codes)*length(optimizers), style=3) + pb_itr <- 0 + # Probing the contents of each file: + for (i in seq_len(length(codes))) { + # Original code from the file + og_code <- codes[[i]] + # Remove whitespaces for comparsion purposes + og_code_no_ws <- gsub(" ", "", og_code) + # Initialize an empty vector + opt_used_in_i <- vector(mode = "character") + # Running each file through each of the optimizers specified + for(j in seq_along(optimizers)) { + #Incrementing the progress bar + pb_itr <- pb_itr + 1 + utils::setTxtProgressBar(pb_outer, pb_itr) + # act_optim is the current optimizer + act_optim <- optimizers[[j]] + # act_optim_name is the name of the current optimizer + act_optim_name <- names(optimizers)[[j]] + # Optimized code + opt_j_code <- act_optim(og_code) + # Remove whitespaces for comparsion purposes + opt_j_code_no_ws <- gsub(" ", "", opt_j_code[[1]]) + # If code was optimized, then append the optimizer name to the file + if(!isTRUE(all.equal(og_code_no_ws, opt_j_code_no_ws))) + opt_used_in_i <- append(opt_used_in_i, act_optim_name) + } + # Setting the name for the files in the resultant list + code_file_name <- names(codes)[i] + final_result[[code_file_name]] <- opt_used_in_i + } + # Removing all entries with `character(0)` or no optimizations + final_result <- final_result[lapply(final_result, length) > 0] + # If no optimization required + if(length(final_result) == 0) { + message("No more optimizations are required in your script(s)") + } else { #If optimization is required + cat("\n") + message(sprintf("`rco` found %d files for optimization", as.integer(length(final_result)))) + message("Use the `rco::rco_gui()` function for detailed comparsion") + message("Here are the file(s) that could be optimized along with the required optimizers:") + cat("\n") + print(final_result) + } +} diff --git a/R/reporting-fun-folder.R b/R/reporting-fun-folder.R new file mode 100644 index 0000000..1c08e9b --- /dev/null +++ b/R/reporting-fun-folder.R @@ -0,0 +1,31 @@ +#' Report possible optimizations in folder containing `.R` files. +#' +#' @param folder Path to a directory with files to optimize. +#' @param optimizers A named list of optimizer functions. +#' @param pattern An optional regular expression. Only file names which match +#' the regular expression will be optimized. +#' @param recursive A logical value indicating whether or not files +#' in subdirectories of `folder` should be optimized as well. +#' +#' @export +#' +generate_folder_opt_report <- function(folder, optimizers = all_optimizers, + pattern = "\\.R$", recursive = TRUE) { + #Check if given folder exists or not + if (!dir.exists(folder)) { + stop("The specified folder does not exist.") + } + #List out the files from the specified folder + to_opt_files <- list.files( + path = folder, pattern = pattern, full.names = TRUE, recursive = recursive + ) + #Ensure folder is non-empty + if (length(to_opt_files) == 0) { + return() + } + message("Files to check for optimization:") + message(paste(to_opt_files, collapse = "\n")) + #Run the files in the `generate_files_opt_report` function + generate_files_opt_report(files = to_opt_files, + optimizers = optimizers) +} \ No newline at end of file diff --git a/R/reporting-fun-text.R b/R/reporting-fun-text.R new file mode 100644 index 0000000..f5ec24d --- /dev/null +++ b/R/reporting-fun-text.R @@ -0,0 +1,16 @@ +#' Report possible optimizations in `.R` code snippet. +#' +#' @param text A character vector with the code to optimize. +#' @param optimizers A named list of optimizer functions. +#' +#' @export +#' +generate_text_opt_report <- function(text, optimizers = all_optimizers) { + #Convert the given text into a temporary file + input_code_snippet <- tempfile() + write_code_file(text, input_code_snippet) + message("Ignore the name that will be assigned to your text/code snippet") + cat("\n") + #Run the temporary file in the `generate_files_opt_report` function + generate_files_opt_report(input_code_snippet, optimizers) +} diff --git a/README.Rmd b/README.Rmd index 0b1c404..87b83a1 100644 --- a/README.Rmd +++ b/README.Rmd @@ -6,7 +6,7 @@ output: github_document [![CRAN status](https://www.r-pkg.org/badges/version/rco)](https://CRAN.R-project.org/package=rco) -[![Lifecycle: stable](https://img.shields.io/badge/lifecycle-stable-brightgreen.svg)](https://www.tidyverse.org/lifecycle/#stable) +[![Lifecycle: stable](https://img.shields.io/badge/lifecycle-stable-brightgreen.svg)](https://lifecycle.r-lib.org/articles/stages.html#stable) [![Travis build status](https://travis-ci.org/jcrodriguez1989/rco.svg?branch=master)](https://travis-ci.org/jcrodriguez1989/rco) [![AppVeyor build status](https://ci.appveyor.com/api/projects/status/github/jcrodriguez1989/rco?branch=master&svg=true)](https://ci.appveyor.com/project/jcrodriguez1989/rco) [![Coverage status](https://codecov.io/gh/jcrodriguez1989/rco/branch/master/graph/badge.svg)](https://codecov.io/github/jcrodriguez1989/rco?branch=master) @@ -17,7 +17,7 @@ Make your R code run faster! The `rco` project, from its start to version 1.0.0, was made possible by a [Google Summer of Code 2019 project](https://summerofcode.withgoogle.com/archive/2019/projects/6300906386096128/). -Thanks to the kind mentorship of [Dr. Yihui Xie](https://yihui.name/en/) and [Dr. Nicolás Wolovick](https://cs.famaf.unc.edu.ar/~nicolasw/). +Thanks to the kind mentorship of [Dr. Yihui Xie](https://yihui.org/en/) and [Dr. Nicolás Wolovick](https://cs.famaf.unc.edu.ar/~nicolasw/). ```{r echo=FALSE} library("rco") diff --git a/README.md b/README.md index 2a006f7..16b6362 100644 --- a/README.md +++ b/README.md @@ -6,7 +6,7 @@ [![CRAN status](https://www.r-pkg.org/badges/version/rco)](https://CRAN.R-project.org/package=rco) [![Lifecycle: -stable](https://img.shields.io/badge/lifecycle-stable-brightgreen.svg)](https://www.tidyverse.org/lifecycle/#stable) +stable](https://img.shields.io/badge/lifecycle-stable-brightgreen.svg)](https://lifecycle.r-lib.org/articles/stages.html#stable) [![Travis build status](https://travis-ci.org/jcrodriguez1989/rco.svg?branch=master)](https://travis-ci.org/jcrodriguez1989/rco) [![AppVeyor build @@ -15,7 +15,7 @@ status](https://ci.appveyor.com/api/projects/status/github/jcrodriguez1989/rco?b status](https://codecov.io/gh/jcrodriguez1989/rco/branch/master/graph/badge.svg)](https://codecov.io/github/jcrodriguez1989/rco?branch=master) -Make your R code run faster\! `rco` analyzes your code and applies +Make your R code run faster! `rco` analyzes your code and applies different optimization strategies that return an R code that runs faster. @@ -23,7 +23,7 @@ The `rco` project, from its start to version 1.0.0, was made possible by a [Google Summer of Code 2019 project](https://summerofcode.withgoogle.com/archive/2019/projects/6300906386096128/). -Thanks to the kind mentorship of [Dr. Yihui Xie](https://yihui.name/en/) +Thanks to the kind mentorship of [Dr. Yihui Xie](https://yihui.org/en/) and [Dr. Nicolás Wolovick](https://cs.famaf.unc.edu.ar/~nicolasw/). ## Installation @@ -46,39 +46,35 @@ Or install the development version from GitHub: `rco` can be used in three ways: - - Using the RStudio Addins - +- Using the RStudio Addins + 1. `Optimize active file`: Optimizes the file currently open in RStudio. It will apply the optimizers present in `all_optimizers`. - + 2. `Optimize selection`: Optimizes the code currently highlited in the RStudio Source Pane. It will apply the optimizers present in `all_optimizers`. - - Using the `shiny` GUIs - +- Using the `shiny` GUIs + 1. `rco_gui("code_optimizer")` opens a `shiny` interface in a browser. This GUI allows to easily optimize chunks of code. - + 2. `rco_gui("pkg_optimizer")` opens a `shiny` interface in a browser. This GUI allows to easily optimize R packages that are hosted at CRAN or GitHub. - - Using the R functions - +- Using the R functions + 1. Optimize some `.R` code files - - - + ``` r optimize_files(c("file_to_optimize_1.R", "file_to_optimize_2.R")) ``` - - 2. Optimize some code in a character vector - - - + + 1. Optimize some code in a character vector + ``` r code <- paste( "code_to_optimize <- 8 ^ 8 * 1918", @@ -87,11 +83,9 @@ Or install the development version from GitHub: ) optimize_text(code) ``` - - 3. Optimize all `.R` code files into a folder - - - + + 1. Optimize all `.R` code files into a folder + ``` r optimize_folder("~/myfolder_to_optimize", recursive = FALSE) ``` diff --git a/_pkgdown.yml b/_pkgdown.yml index 766647c..c0b0782 100644 --- a/_pkgdown.yml +++ b/_pkgdown.yml @@ -12,6 +12,8 @@ navbar: menu: - text: "Common Subexpression Elimination" href: articles/opt-common-subexpr.html + - text: "Conditional Threading" + href: articles/opt-cond-thread.html - text: "Constant Folding" href: articles/opt-constant-folding.html - text: "Constant Propagation" @@ -24,10 +26,14 @@ navbar: href: articles/opt-dead-store.html - text: "Loop-invariant Code Motion" href: articles/opt-loop-invariant.html + - text: "Memory Allocation" + href: articles/opt-memory-alloc.html - text: "Articles" menu: - text: "Contributing an Optimizer" href: articles/contributing-an-optimizer.html + - text: "Potential Optimizers" + href: articles/potential-optimizers.html - text: "Docker Image" href: articles/docker-readme.html - text: "News" diff --git a/docs/404.html b/docs/404.html index 691351c..35f1247 100644 --- a/docs/404.html +++ b/docs/404.html @@ -71,7 +71,7 @@ rco - 1.0.1 + 1.0.2 @@ -96,6 +96,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -114,6 +117,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -126,6 +132,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -178,7 +187,7 @@

    Contents

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/CODE_OF_CONDUCT.html b/docs/CODE_OF_CONDUCT.html index 91b61ea..97f5771 100644 --- a/docs/CODE_OF_CONDUCT.html +++ b/docs/CODE_OF_CONDUCT.html @@ -71,7 +71,7 @@ rco - 1.0.1 + 1.0.2 @@ -96,6 +96,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -114,6 +117,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -126,6 +132,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -228,7 +237,7 @@

    Contents

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/CONTRIBUTING.html b/docs/CONTRIBUTING.html index 8f21dd8..a00e315 100644 --- a/docs/CONTRIBUTING.html +++ b/docs/CONTRIBUTING.html @@ -71,7 +71,7 @@ rco - 1.0.1 + 1.0.2 @@ -96,6 +96,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -114,6 +117,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -126,6 +132,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -217,7 +226,7 @@

    Contents

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/contributing-an-optimizer.html b/docs/articles/contributing-an-optimizer.html index 44fea48..fa7dfcd 100644 --- a/docs/articles/contributing-an-optimizer.html +++ b/docs/articles/contributing-an-optimizer.html @@ -31,7 +31,7 @@ rco - 1.0.1 + 1.0.2 @@ -56,6 +56,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -74,6 +77,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -86,6 +92,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -112,7 +121,8 @@ -
    + +

    Contributing an optimizer

    @@ -135,37 +145,38 @@

    Writing the code

    All the code developed for the new optimizer will go in a file named R/opt-**optimizer-name**.R.

    An optimizer will be a single function (one exported), which should be called opt_**optimizer_name**. This function will take as input a texts parameter that contains a string list (character vector) with the code to optimize. An example of this, with just one code to optimize, would be:

    -
    texts <- list(
-  paste(
+
    +texts <- list(
+  paste(
     "i <- 320 * 200 * 32",
     "x <- i * 20 + 100",
-    sep = "\n"
-  )
-)
    
+    sep = "\n"
+  )
+)

    All other parameters that the function opt_**optimizer_name** takes as input must have an associated default parameter, for example, opt_**optimizer_name** <- function(texts, param1 = FALSE) { ... }.

    As output, the developed function must return a list with a field codes that must have the same format as the texts input but with the code optimized.

    To document code we use the roxygen2 package. You should give a brief description of the new optimizer, of the input parameters, and an example of use.

    As skeleton to use to develop the new optimizer we propose to follow:

    -
    #' Optimizer: **New Optimizer Name**
-#' 
-#' **New optimizer description**
-#' Carefully examine the results after running this function!
-#' 
-#' @param texts A list of character vectors with the code to optimize.
-#' **Other parameters description**
-#' 
-#' @examples
-#' **New optimizer example of use**
-#' 
-#' @export
-#' 
-opt_**optimizer_name** <- function(texts) {
-  # todo: **list of possible improvements for the optimizer**
-  res <- list()
-  **new optimizer code**
-  res$codes <- **...**
-  return(res)
-}
    +
    #' Optimizer: **New Optimizer Name**
+#' 
+#' **New optimizer description**
+#' Carefully examine the results after running this function!
+#' 
+#' @param texts A list of character vectors with the code to optimize.
+#' **Other parameters description**
+#' 
+#' @examples
+#' **New optimizer example of use**
+#' 
+#' @export
+#' 
+opt_**optimizer_name** <- function(texts) {
+  # todo: **list of possible improvements for the optimizer**
+  res <- list()
+  **new optimizer code**
+  res$codes <- **...**
+  return(res)
+}

    For the parsing and tokenization of the code we suggest to use the functions developed in R/parse.R.

    @@ -174,39 +185,39 @@

    For testing, we use the testthat package. All the code for testing the new optimizer will go in a file named tests/testthat/test-opt_**optimizer_name**.R.

    Please create a test suite that covers a large percentage of possible use and border cases.

    As skeleton to use to develop the new optimizer’s test suite we propose to follow:

    -
    context("opt_**optimizer_name**")
-
-test_that("**test name**", {
-  **...**
-})
-
-**more test cases**
    +
    context("opt_**optimizer_name**")
+
+test_that("**test name**", {
+  **...**
+})
+
+**more test cases**

    Writing the vignette

    When developing a new optimizer, we create a vignette to explain it. This documentation will not only be a vignette of the package but will also be part of the rco website.

    For writing vignettes, we use the knitr and rmarkdown packages. The new vignette will go in a file named vignettes/opt-**optimizer-name**.Rmd, and must follow the skeleton:

    -
    ---
-output: rmarkdown::html_vignette
-title: **New Optimizer Name**
-vignette: >
-  %\VignetteIndexEntry{**New Optimizer Name**}
-  %\VignetteEngine{knitr::rmarkdown}
-  %\VignetteEncoding{UTF-8}
----
-
-# **New Optimizer Name**
-
-## Background
-
-## Example
-
-## Implementation
-
-## **Additional headings** (optional)
-
-## To-Do (optional)
    +
    ---
+output: rmarkdown::html_vignette
+title: **New Optimizer Name**
+vignette: >
+  %\VignetteIndexEntry{**New Optimizer Name**}
+  %\VignetteEngine{knitr::rmarkdown}
+  %\VignetteEncoding{UTF-8}
+---
+
+# **New Optimizer Name**
+
+## Background
+
+## Example
+
+## Implementation
+
+## **Additional headings** (optional)
+
+## To-Do (optional)

    The Background section must introduce the reader to why this optimization provides improvements, and what it does. The Example section must give a real example to be optimized, and show the improvements it gave in terms of execution speed, memory usage, or others. The Implementation section must show the idea beneath the optimizer coding, this section intends to ease the understanding of the developed code if it is needed to be edited or improved. Then, as many sections as necessary can be included, where questions and challenges related to the optimizer are explained or commented. Finally, if a list of possible improvements for the optimizer were detailed, each of them should be discussed in the To-Do section.

    @@ -223,11 +234,11 @@

    NEWS.md

    A entry to the NEWS.md file should be added:

    -
    # rco **x.y.(z+1)**
-
-  - Adding **New Optimizer Name** optimizer.
-
-**...**
    +
    # rco **x.y.(z+1)**
+
+  - Adding **New Optimizer Name** optimizer.
+
+**...**

    @@ -260,7 +271,7 @@

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/contributing-an-optimizer_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/contributing-an-optimizer_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/contributing-an-optimizer_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/contributing-an-optimizer_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/contributing-an-optimizer_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/contributing-an-optimizer_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/contributing-an-optimizer_files/header-attrs-2.5/header-attrs.js b/docs/articles/contributing-an-optimizer_files/header-attrs-2.5/header-attrs.js new file mode 100644 index 0000000..dd57d92 --- /dev/null +++ b/docs/articles/contributing-an-optimizer_files/header-attrs-2.5/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/docker-readme.html b/docs/articles/docker-readme.html index 8f0f3b7..3b410dd 100644 --- a/docs/articles/docker-readme.html +++ b/docs/articles/docker-readme.html @@ -31,7 +31,7 @@ rco - 1.0.1 + 1.0.2
    @@ -56,6 +56,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -74,6 +77,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -86,6 +92,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -112,7 +121,8 @@ -
    + +

    Docker file

    @@ -133,24 +143,24 @@

    Installation

    First, you need to install Docker on your computer. And then, from a terminal, the rco Docker image can be pulled from Docker Hub with:

    - +
    docker pull jcrodriguez1989/rco

    Usage

    The docker image basic usage will optimize a random CRAN package, this is done with:

    - +
    docker run jcrodriguez1989/rco

    We can also specify a package to optimize, using the RCO_PKG environment variable, for example, if we want to optimize the rflights package:

    - +
    docker run -e RCO_PKG=rflights jcrodriguez1989/rco

    Also, if we want to get the resulting optimized files we can set a docker shared folder, with:

    - +
    # Replace DEST_FOLDER path, with your desired output path
+DEST_FOLDER=/tmp/rco_dock_res
+docker run -v $DEST_FOLDER:/rco_results jcrodriguez1989/rco

    In summary, if we want to optimize the rflights package and save its results, we can do:

    - +
    docker run -e RCO_PKG=rflights -v $DEST_FOLDER:/rco_results jcrodriguez1989/rco

    And in the DEST_FOLDER we will have files as:

    - +
    ls $DEST_FOLDER
+## rflights  rflights_0.1.0.tar.gz  rflights_opt
    @@ -170,7 +180,7 @@

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/docker-readme_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/docker-readme_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/docker-readme_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/docker-readme_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/docker-readme_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/docker-readme_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/docker-readme_files/header-attrs-2.5/header-attrs.js b/docs/articles/docker-readme_files/header-attrs-2.5/header-attrs.js new file mode 100644 index 0000000..dd57d92 --- /dev/null +++ b/docs/articles/docker-readme_files/header-attrs-2.5/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/gsoc2020-final-report.html b/docs/articles/gsoc2020-final-report.html new file mode 100644 index 0000000..a0555db --- /dev/null +++ b/docs/articles/gsoc2020-final-report.html @@ -0,0 +1,344 @@ + + + + + + + +GSoC 2020 Final Report • rco + + + + + + + + + + +
    +
    + + + + +
    +
    +
    +
    +

    GSoC 2020 Final Report

    +

    Rahul Saxena

    + + + Source: vignettes/gsoc2020-final-report.Rmd + + +
    + + + +
    +

    +My GSoC 2020 work in a nutshell

    +

    Phew… So it’s already time for the end of GSoC 2020. Time sure flew by this summer. Now it’s the time for me to take a step back and ponder over my summer with The R Project for Statistical Computing. So, let’s touch upon my summer in a chronological fashion.

    +
    +

    +Humble Beginnings:

    +

    Around the time when The R Project for Statistical Computing announced their participation in GSoC 2020, a list of proposed projects were announced. The rco or the R Code Optimizer caught my attention at the very first moment and I didn’t look back since then.

    +

    Although the exsisting rco repository was extremely well maintained and documented, me being a newbie still faced quite a lot of difficulties and also, most probably gave my mentors, Dr. JC Rodriguez and Mauricio “Pachá” Vargas Sepúlveda a hard time.

    +

    As part of the tests to be completed for being considered as a potential student developer for R with the rco project, I was required to build a completely functional and working optimizer.

    +

    I decided to work on a suggested optimizer that made column extraction faster and more efficient. After the benign guidance of my mentors, I was able to come up with an column extraction optimizer, that performed according to its expectation. This got me so excited that I worked on and built another value extraction optimizer without any explicit incentive :D

    + +
    +
    +

    +Post GSoC selection:

    +

    It was only a few days before my acceptance into the GSoC 2020 program, that my mentors gave their insights into both these optimizers and explained, how these optimizers can be broken, owing to the extreme flexibility that the R Language offers. Since, .subset2( ) is not a reserved keyword in the R Language, it could be over-written. Hence we can never be sure while optimizing, that the functions that are being called has not been re-defined by the user. However, this served as a great primer as to what to expect when GSoC officially began.

    +
    +
    +

    +Getting up to speed:

    +

    The next tasks that I were assigned were not too glorious/dramatic or blinge-y but they were essential steps for the upkeep of a package on CRAN. Firstly, I designed a vignette that enlisted all the future potential optimizers that can be implemented in the rco package. Next, I went on to collate the efforts put in the rco library by the Google Code-In students, by creating PRs that such that every student gets credit as a contributor and at the same time hand-picking the best of examples and explanations to include in the official rco documentation. It was a riveting experience for me, as I had to learn and implement some really advanced concepts of git and github to be able to pull this off, including, but not limited to, cherry-picking, resolving merge conficts, adding remotes, etc.

    + +
    +
    +

    +Unexpected Turn of Events:

    +

    As Dwight D. Eisenhower once said,

    +

    In preparing for battle I have always found that plans are useless, but planning is indispensable.

    +

    I had listed several optimization techniques that could have been implemented in rco in my GSoC proposal as well the potential optimizers vignette but when we got down to discussing the design of these optimizers, it became quite clear to me and my mentors that we were walking on thin ice. With the extreme constraints that are imposed due to R’s flexible nature, only a few of our optimization strategies seemed bullet-proof.

    +
    +

    +Conditional Threading Optimizer:

    +

    I went on ahead with the idea of Jump/Conditional Threading and started designing the optimizer. The objective was to replace if statments with else wherever possible and also to group together the code blocks of different if blocks if the if conditionals were same. This concept is succintly covered over here.

    +

    To see the speed-ups provided by this optimization strategy, have a look at this example:

    +
    +

    +Unoptimized Code

    +
    +cond_thread <- function(n) {
+  evens <- 0
+  evens_sum <- 0
+  odds <- 0
+  for (i in seq_len(n)) {
+    if (i %% 2 == 0) { # same logical as next if condition (can be merged)
+      evens <- evens + 1
+    }
+    if (i %% 2 == 0) {
+      evens_sum <- evens_sum + i
+    }
+    if (!(i %% 2 == 0)) { # exact negation as previous if (can be an else)
+      odds <- odds + 1
+    }
+  }
+}
    +
    +
    +

    +Proposed Optimized Code

    +
    +cond_thread_opt <- function(n) {
+  evens <- 0
+  evens_sum <- 0
+  odds <- 0
+  for (i in seq_len(n)) {
+    if (i %% 2 == 0) { # merged
+      evens <- evens + 1
+      evens_sum <- evens_sum + i
+    } else { # converted to else
+      odds <- odds + 1
+    }
+  }
+}
    +
    +
    +

    +Benchmark

    +
    +n <- 100000
+autoplot(microbenchmark(cond_thread(n), cond_thread_opt(n)))
    +

    + +
    +
    +
    +

    +Memory Allocation Optimizer:

    +

    Next, I set my eyes on the Memory Pre-Allocation optimization technique. The objective here was to save the R Programmers from the sin of growing a vector in loops. In our experience, we’ve seen that vector, lists, etc are initialized often with NULL or c() out of convenience. Bothering about the size of the vectors may seem trivial to the programmer, but it has a huge impact on the performance of the script.

    +

    To see the speed-ups provided by this optimization strategy, consider this example:

    +
    +

    +Unoptimized Code

    +
    +mem_alloc <- function(n) {
+  vec <- NULL
+  for (i in seq_len(n)) {
+    vec[i] <- i
+  }
+}
    +
    +
    +

    +Proposed Optimized Code

    +
    +mem_alloc_opt <- function(n) {
+  vec <- vector(length = n)
+  for (i in seq_len(n)) {
+    vec[i] <- i
+  }
+}
    +
    +
    +

    +Benchmark

    +
    +n <- 100000
+autoplot(microbenchmark(mem_alloc(n), mem_alloc_opt(n)))
    +

    + +
    +
    +
    +
    +

    +Pivot and Push:

    +

    As mentioned in the last segment, we had pretty much exhausted all the possible optimization strategies, that could be implemented without breaking. Now, it was onto us to decide whether, to work on new optimizer, that had a non-zero chance of breaking or pivot and change objectives. Following the lead of my mentors, I decided to put on hold, the optimizers that could break and focus on further bolstering the rco package.

    +
    +

    +Fixing a critical issue:

    +

    Owing to the fantastic upkeep of the rco repository, it had almost no bugs in the optimizers, and that is no mean feat. However an issue reporting a bug was reported. The bug was that an optimizer, namely opt-dead-expr() did not function when the user used ;. Ideally, the optimizer shouldn’t have been affected by the usage of ;, but here it was, a bug, laying dormant for over 6 months.

    +

    I decided to tackle this bug head on. I created a separate branch for this bug and reproduced the bug. After lots of trials and tribulations and lots of reverse-engineering, I zeroed in one the problem. It was a problem that existed in very low-level R code. When a R script without the use of ; is parsed, then they get tokenized as something called expr but when ; was used, the parser tokenized them as an exprlist. The exsisting optimizer did not handle the case of exprlist, so I appended code that handled exprlist and then the optimizer started functioning normally irrespective of the usage of ;. This is an issue which could arise in other optimizers too, but given that, it has been solved in one optimizer, solving them in other optimizers would not prove challenging. Also, in the process of making the opt_dead_expr() optimizer work, I further bolstered the robustness of the optimizer by doubling the number of test cases that the optimizer must pass, including tests with ;.

    + +
    +
    +

    +Call for Actions:

    +

    To reduce dependencies on 3rd party applications and to promote homogeneity, we decided to go for an in-house solution for the CI/CD needs of our package, ie Github Actions. Earlier the popular choice for carrying out CI/CD operations were 3rd party applications such as Travis CI or codecov.io, but lately more and more developers and organizations had been migrating to Github Actions and we decided to follow suit. Github Marketplace wasn’t much help, as there were not much support for the R Language as compared to other languages such as Rust, Ruby, Typescript, etc. So, I went through several documentations and scarce examples and created a branch that renders a website and completes the testing that Travis used to do. We left out the code coverage as Github does not yet support badges that shows the percentage of coverage.

    +
    +
    +

    +Acting on users’ recommendations:

    +

    One of the users of rco opened an issue, with a feature request. The request that the user asked for, is as follows:

    +
    +

    I think it would be nifty to be able to create a report out of rco optimisers. The idea would be to analyse the whole code and return some kind of markdown / bookdown that will list all the results from the optimisers, without changing the > original code. It might be useful to analyse your own code and learn what you could do better, but also it could be use in the industry to analyse the quality of code.

    +
    +

    While actually implementing this idea, would have been a tall order, me and my mentors discussed that, as the first iteration we could work on creating a function, which could be called to check, exactly how many files from a folder or a group of files can be optimized and which optimizer is being used there. If interested, the user could run that specific file along with the optimizers named in the output, in the rco_gui() function and see the code difference.

    + +
    +
    +
    +

    +And, my GSoC ’20 journey comes to an end:

    +

    Well, I never thought that 3 months could pass in a jiffy, but these months when I was working under the R Language in this program, had been the golden months of my life. Never to be forgotten and to be re-lived time and again. I am proud and happy to state that, my GSoC experience and project was mostly a hit. Yes, we did face obstacles, but each and every time we emerged stronger. I have now become a life-long fan of open-source.

    +
    +

    Open source is simply put, magic. And, no one can ever get enough of magic, right >.<

    +
    +
    +
    +
    + + + +
    + + + +
    +

    Developed by Juan Cruz Rodriguez.

    +
    + +
    +

    Site built with pkgdown 1.6.1.

    +
    + +
    +
    + + + + + + diff --git a/docs/articles/gsoc2020-final-report_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/gsoc2020-final-report_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/gsoc2020-final-report_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/gsoc2020-final-report_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/gsoc2020-final-report_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/gsoc2020-final-report_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/gsoc2020-final-report_files/figure-html/cond_thread_thread_benchmark-1.png b/docs/articles/gsoc2020-final-report_files/figure-html/cond_thread_thread_benchmark-1.png new file mode 100644 index 0000000000000000000000000000000000000000..5fce67c999341aecf5faa7be12447766d8aabc8a GIT binary patch literal 54612 zcmeFZc{J7U+cvy4Pnty(4I4#K8B>NPwJSsBjFplpBqVd1Luw;sh$8clkj$D$l6fd| zl8`C$w7ti*`Q7V&pZ8tsUGG}Y^WSat`+a@;?)~{(!+D+OaUSP!UiXw0kFqY|TtcBx zSZT)&t57KO?J1PGXBREN-{{S$(8oU(oI7#!F#d_3u|t`z_{U!sP|FL+t z^%uDo`(rbiU+YZ@5Sc9IEEVvdYLadR^i%bfk4BG<68Z_doW$hmE4 zoW9%>!wq^3$}U}7+N2(_4oBvi!8zbj8|`1erL;;EmWFfq>meW z-g=L_cOMqCw%p)y6IsW}8PnHiMN_$aIZ0Q@A1{ltY-wr1le3g4?hhL*G}Y96snIDZ zqM>seGhQ5-J3Gn$%~StBwx<8{;Q#Y>`#)>^|Fy>A$#5j030eAy$4{J4*U;b#ERJ~c zgyVa+U2+lA@r0HU4>U*K#3Z^p(IAn>b@J4{t37IVc5mt}u6hkn`S@ZC3m!;KPY&3Q zwfPH5OY5!k_!NGh>1K?z;CkHnEK8nAkB_!`ry3Opw9aWT5Zty+`M`ne;^wuX=~vnX z_U=6`(P$qwKG+fV@#9DN7!Gof-SQ8>#Tmc0zH&v(ao9XwFY`@xg%6$X8x$0G_wJ%= z`?=ZHuH7amc!JxzKvGFj@xeMF-KZlT>pI4}5^J(Y8e=OfEAumEoTMmmhH^fwLk>fq z*bV+PTy63zYTNVX&aGR_Pute$MHUHVp1geda)DlToAvgsTMu>BB-ThYIfVu9yLy}+ z6BCnSS|#8-P%P0|mzLO*JMHo*KVSTZ^UMNm)`Isg`MZS@8W(qR)v8r=YGg(9moLYC z1l>;uDMs=s21R*sNpRM|&)G~*jl>${y4=5aZ*N#v(bun) zjU}V)vCjEa?XGy2C;UP}irn5ud3kw_{O4@SJ1j3Ruz$ZUT}@Y4x3%a9<=tj+aSi=M zO~uFx_}7#;jxe&W-4GfafDqj{L$8<*qE`n1fJhpo98QQ&dZYP z!!PdaEQuc<@?tgnLZ&B=Zaw-y^nC-o^we01Yjtd5jze`z0+pAK&-B;lN8^9OyepG0 zDj7~x$LkE?Upn!D)lJu!t2T!fu`16dP{OkMPg$7PBmYZvZ>qAc4Gh*bG2S#<#_DC- zIkb97h@O2vFYJ!xkt4Sv6oX^@4O~vYeED+7C6MXsMb~0q-J)H_JNUge?yHvEFzmim zsC1$?H*s)qa6zVm$EO#8mN}l$g3L?IM|7?s!w_{G z{#*9;a_cXm{rq{ZqLaCl&F9GgyuP##nBW`oA$2!z&dcu4=Yo;#PT#YA&mNUE z{EC&B_=O}bDovI#8g_dHPr}s1NTBXU<_rq3u@upm64`0p>B)9kUYcE|o)4dbN>_D! z@y-}dEF;U@1;$@|x98ve85FcZdhD0akGGe@3aglhT$wCkXxZN%sV9;zmWWq}xQv_F zcK@jQcwaG2Gf|&dmWrw>FRiVujdPz}+~b`(d?6(lt!q7Z;8VH{KILUx+zy&b zrd{`7{gsmS&K7~MYiqT#$9@N9bR}eZUYefBosKOFJ1R29Nxqn8L!8ork*SfUE=$*m zKx&~;az~rLf%tgcf;pKD3HW4JbmTL7mBU|yc8pfkY1FE_F0G{gC;2v|n3GKRz}{d* zgFEgn(hhu`+~?q4Gq!-a*K=*d)4jZ{=We?B~XI?(A-_m5OQ&4uK9#tCGXSUZES zHl8|r_Bpi!wyHeMm{Lgi^H~OfoFI__cGkX-8zM1UN zlj1;$u3p#a%y7d?%k#asA|F0n=@TpE$SABy$mkR>to!-NW9X-caJfIz@0Ug~-}%1V zu1nZ;%p*z0s#(tVPwBgl_d7LDp7g|GCM@K(&#N8&utL`OPwDd?InJ5){Fu8Mu!i`| z8(JAQ(XW$}c{g(PbN-26Z)%vTEr#__2$C#&A}S;+YXA>WiI=7UCY4odcjd#stEIziD@hH7YUe|=O!}GA=8Qo zh4G%uUMlU;qeq-FPHzj@#GaLwDtmEDYy0pkX*fA$QKMhHI4ENDwe<6&t*MqxGFO() zBjcX8k<2zbIwqz(?sT%}(7PK81vhWTI(;q6ylT1g^y|Yjhp#Tpvc}L`I^uM)5rAjq z$4>HP%DbDeL}g2jIIZV6RVU~_(kv4ld(sPj-T&(I@L_0;zM}$8pxU`}FTxF3Mqghj z8Z14u!es1Un>^P{fa)X|URn6~@#86H-pw&|{&;uQ>+j}#8kM5;Rv%c$fB7g?CK||| zjxS(zuVZI7!|`_*ZMicvk~^)3@b^oAxntL79zwi0d;_iAPg6honY3eo?x~7ZY-}%2 zg~@kdn=9Egb=2BU!hK2Q}5-qf!`+ZB~rav1D@URb!9<&uaLNC_WSG8FU7@T zNyo`xdvazs%*!rKp)Ox+WW{H zDS3>hX=-PG0Z+|nf2&uXGn}3S)=6Dg*L>Dp-abrD>&{fKM%KlPcPDL`)Naag7IPlA zxpn8x3A$ZxV?2p@bf$xJ=RnYEPH~#(PRsw(fRwqxx@hO`bvL#3E`MSpr~1W<7u1bQ zui=cGP+}4Xi&V%us^TzczqNXv#OV95qR6z(;Fzjy*%9_Cj>61Kb>fHur}PtRetjWx zp$vH`wc{HxjgJ0hnb)72)g%P$zxKM!=PD*e4rHM^(@k}9k)edUG%!EbdS5W1v^L4m zYsgz>{Mhtpf$Yh+MdWG9O`|uLQzWB1v#-{rnic$Mpj4z;8c0v}eaIh&=kk5})a2X| z@WH=R{)~KSP@%SDab7j7*BP$~*6Ot(3t0&*Pz&o=i<3x?#)nID$>(?p( zWb~RYv9xXE@p8e0y?HPI3?qnah3-qIzkmS+=&tJY_;jx9$!6+rPOVpYTCRVd-#MRFrx3S@)dYFKeQ$4n z>ZOJZbwSMp(ey3k3wciom&M^S+On{_K3$cW)zw+%HvGC@`tY2gp_uirV}?^W)as{B zJ^Ih;kX&A_yg%&f!etzJe;SOMAM03@A;@i~!CgB5WGf^vEys$F94dM^p!6PxwDqsG zb~yB;xBs$|q|z{eN}NFR**&Ttdvs>dLZs%+B^~L}<{SCr$QJ+t1O}EcjU`AN=h}^= zy`G-SXe!5#FA2LE_?$5{A;`;ngbo{w7S5e4|8<{Ci4u7OK|EkD>~L(LovUl^!sX}j z=(S+65PH{l5$ppdYj!q@3o#qfD^1QL!FxnRcxlnm(HTS4I+%h{M^&u9!|>nj9kn;j zs%{XPk|I|@*0;De@o4XIPSIoHfyE#1pLs)4G%EElJ>}e|CEX2~HM8U3Cy=_h7U3q< zb=rldqOBb;V|lxD@M}-@h(4_#hw?<)@eShZ?4eX72qs{=cR|?vrRzAlZY*5!xav#p z{9`y#uYR_+^3q72mSNjU%2v46rJ^rsqcyaGfS z&Hv2)y>U~L)0lObr3m?Cs0s2jHzO)W;yTr3z)M@_apc2oR+02AqP7UdB+H=Ekf;L) zJez%Z3{omLqw`4n;DYGkIGZs5yudDUyX^bU965aW7`-|{FE+%L@qKpNT;OF&Po6h2 zGHv_fN2lKT@O<;3+fQ^?%{g-LpgXry-vLAC;lJPEmVUD_LDnurD>;$oficsY)+3FM ziF6+yg@h&7e*qL`Yp0s>(tacv-qA?ZzhupHlaza8wzm+`y9#(*eW}Q9yPn_S-jlv% zUO;JSDW`N`lk2o|x~|X~L8j>+!*Es()?1`CUTfcI=nUX+|Ni|wNn7T{=vbs5X%BJD zp>7uzE_w251zDjp+*tiZ)24%({<2Ip4{8Cd8wxF)wvzxwS(5laoHFg&&6$M0igwaWm#7B5~*4~IK6 zEe;R^!%e74jtDr)`=S{Q2`SRxa_cS|K=Y0)m3G#u~s+ z#s4FIffO-BA(AwRO8? zMhD-}pRy0XVphh*X~aIe_IthI6v0Ix=4O11+yHR|2hQV=j>C&nX@HcRVy3>BeXX8- z%dL#o2m7cri!>diLucMIxeVjOWY?)nmMjr9tKPxPL-MIwa)&s@&7Yh*caBQqlCU`c z-#_kQdC$8SfOl zRQu`xeJudpZo~Y~=aQc2vIY3;>v?+!#E&19ZQVN4Z$W$iGxxbS#E+HA%7fb z7LyWOBY5x%m@+YkLE~FDZyuxPjQ(7TttA5wV~y1 z5{9*zi!3*n7#KxOSvgeGz*!wR41I8-X}VQ^QvK=Tt1p^Rg>(^I{?H2kN;?BRef zMMXH-O40M5SK9uqJw^8o42)ebVg%nc%q6iOkC_PAlmk zPPIQ35J}<5t^6MsWKN+%qn+PM9jc}Gf4sdQ;rLGAm zY-LKkR6_Q!HcbU|UcNU^;(wtsw`M~2vYY`OLWZNJuFgyAtcdhR!eD8wo)>+3$2vlc zZQs6~9uBT2n=#%6U^-Vq)s;H|s+ob1t`9>DdFsGy(Jg>yz<>{#bf=*JSt&&MZ^ zTr7j&wEv`sGl5Ueigx7hWS_4S`oqZIV z^4fe7JLDcRzZR2~r3+67s{nNG@9z(gan3-rEZ(^+DvS`1Hx{t&scv%U+6m;JKNANx z^%1LF^UFG2)S_PUQd5o|eeJ4M&vB*+BfW=DM3Hc=Zh6wftMzX$r_2y;Qp`!e zc>7ih32NLM^Y2?aYCks=`tZjh=f%(~bHTMJJl%7oA@i!g&5mErs0XB)f8PTS>k~_e zPPaU=%?ffLpyfR|o)b(}@IH||0E{diAQ}WFnJ||bzXQyu#Qebpl~+cqw%KJWBrGQr zm_uB_6LaJ#4nObl`}+Cu(h!%*J=cqHqyktIsLzRG7elU0PpT`WV zC3gl>psb14&(;o;_bNwz#7o<$^QO|VH#^cpIH&g47q@)FX-(bxM{pvuL7U*ux%(_Mt{4|U)#mfG#R7vG7&>-PJ?8U}~n`nTfmCf&N zTuA*F%%$7_fGBqP=V1ids@g_EOk6J5nDO^@k?CZ^3`Y`$7@2fp{>tVpTeq@ta|_V~ zgoGZ$RRY2f?fic&Jn$|GrxTKVjA?=*Ebvns8OSP#Vf<6;|17kL`z%X$9b}I^f%eB>RXM%@fP>J%?(Iz zB2GSifrI&}*6!VWcH{Rq7avzp5KG##$!deNgBp*^=yMsys39_bo~sUUeL}gD)_OK& zM@*Y?vZZYPgu-Cg2+Uu3cai<|X4FaM0WaWP3Ws;?gwt+E0XHDjQe+5hT%Wq1Na#J96WUBLh^T0 zU4FL+z&R4Eh}uWMi|#i*DQz~sc(82CB6?jI&Ny`3mW+^b@Fw{_n> zohGLNFUC-!t4Y!0o&W6j8~NL8z(`Rz0%ySW&KfI(?2CC^&7H zddso~i1VE{-m5<^3vMLf?wQ5*A|fJUSAL(o^!@dxTBB6$p*ymZ2arAdet0?bI`J|% zHRZ010iN#Kq(dO+Re(f zL!OR?Mg+(iomh?dm{r>dgmGYuwc8@O3((qFx*x}D<9>VPJIgn!z`^_*7_c4ua|*x& zW=xDcCN9pmsEuCGy8DN)){b?C0CJDOZ2l0^im&GMn0BIju3L_^2VBP|)exZ&cGbcA z!M!AB$U?$!Tv1U2EKF5|LLlb}bMqbXn;5c4_FlZ$0<_WzOadz)S_QDUo)VKn{wMGU z56m}vGu>||xnT(_tGbrf6EK+s^!)i_j9T4xoFbqy`Pdc1*ZSa^+<91pS7t4;=e>QM zA}9R>*q1O0aij(Xc78l@dR%yU+Jcimtt5S)JxdrIv`Iekh>B7uiV+)O?jTc1Y2n|$ zf4(6hiKx`H{rZLP9g+fLAMc+S6GNcO>({Rz?Ay6z%lkhPe|g60GHlIw`T0S1Z;sNx zgMfj0tC=1aPhDAgQ({eBv+d2BH@C^in;7bfMB>8Hv}ukT^}D(F)5`NTH5$ZR} zIMr%ltbV98fQQBo1ZamDK$C)@n`JRV?tW;AN=h^|G*rr1xd^eR`6Nk~q%%=KSbUYG z^67^bVkoAp*98Rxm;C_VJlAxt#11FzP0~zYkw5#3}CC}i?dV-IK3*y=u*kExr zcqF*kNR?Q>ukZOC3}rEZld`h1B`a1)^c+G8ghx>eqpI4hNu}b#g$u|Ud@4G8D>~G` zg_MoI%U!#6ZIDK`kbe=!1t+P$c8*@F@?gvI+j>JDCZYg7nFxg9Mui(gjpvk1SHBx<*B z$@JaU_wNseNgqbZzCTlT-f(YC#iflccA4O8BeSxye4Yt!*>VuTFkgnIpr8)cv=Vd* z3gp!VddQFDD4Uuuxig(kXJs_0qrwUQOU_7O*S9}FGP89%w5;Br;szFZyPSJJso`(m zvEy%?g|KpEyNrc-Z>fr5y!ZFBELmqCn@yxriRvgBvW5)$B5faN{KR}*&z2jZ#!9wWw4HC36qxpSQd zvYWMQ)`Sgr*M$sbG444(*s%B94#ugX)G2)}t!>@~SIWC~2nwni8AX7u@!q!+n`UBS zLc&OSc{w?Z%Qs4KthtA!-(gAUNcU7GyS9q}bfOdsi8y3=*kPK!A6_`XgK`!m`)V8E8Bm0dMcJpbHYrTX z?r#)aL1^O$j{Ca@hyv#jtdKe?6Q2kk)gikFs~3Br=rQ4wjY^&sFFZt9w0ilS4}T1n z4yf^Lbb3u23sSu3wRjAli*%kCHUWhXU%S6&5=XHdync#Zx3-~bizXf+fO_AM%&q}6a2~cUh zMygDv<|7W-m}fPJffUCPiwoz^8`mW0jiVs)V4r(5(kR$$Io!JECGkLoOFvRhfhYTs zY8DCX6om+WoJKQQeTf^+5%9)q@kEUGp2WB~QTZnm-cJ9#ccp3VAV9I$$OlmkK^;Nb zYaKu1(XX>-&)&a#w|S{KW$0ViPz=J5Vx%gfYW|}=@C^A~{&n5r4YG zZa&GEz|d)Qab7IXkJsc9mI7g^96Va$^jOGr*?e|gtM{DQ`iYEwZ`VJ-yvhg( zFkzbt1&NN^TozIO&Q|ct#47WjPmVOXx_VU9Tk4UhSlFw}`A~a?**gASB^NIipTncAI^+K4E%M6{z}M9B4b*cNP}f z`MOA`0vUM%#5z7bH*Pl;xY%A}91|PSqtWg$#}Aqw4p24CqJdOEQ3hV$o>*ikb+t#A zR*nqD=?iD)zx_Sc@qtuQA0`7)%n5_j2}{zBm)};0;;o@0Krx`uOGtT4VIv4Jr7#3! zV5Yi_t^FKf*ImhJhBq+%{`%$WHEaB40^O&NAGb+KX%nyp*Zf{Ju$a%|(-!s54h<4_BQ!VtD1*s;3-Kl=^1gP=2aQ2{z*Q%xENjWdM#Lboch4hGk=DWGrRXG6V#|I3O68i2!-69@7i^; zz8UrG(=Ma1*DXQX=XrxCo=w2;m#rZ}Rs{@YGv7zVfId#!EX6 zN5RNb8(oI5#kD6?aI#1?0D>oaMrM`z4Ai42D+E^=$otKl#&p!xU&5`d4|AG&-rMrw zj#>7DnZSz@Nm$Ys@gsl-tG)&uKYaMGS6)pnC@MiAp>Lz>;V>UQd?+8qE$3{N*+5Vn zB5l8evX++dj;as3X>|=}Y%9f|@=?^-)mVCgUZRo09fn2*33v*Q9N(P}o3S+rlFm~@ zH3eP~P9TC)Oe*+zf!hGU^Y0e^{cE0$JdRMZM8|~Hu5t(@ltfR~fFd|qCCq}9+Hzg8 z2|JTL@NHj4M?~O*3HfOp58}8gBNVtGazav(N=J1Se(}`YHpU9+Wo(0ps66SO!5hn_1W-@LV4^ereG#mB zam1sw)5OzxNAT3ZQ999+m?fmkAG*El6tuY{J4P1UF;rtvgW^tnf?lS8nAngfPX&`( zgJ2~_wb48I+F!?^&X@SBZK!1V>sl9Xw}dJ}bSEIG{0jYP_M-BnO%fs^s*WQ)RkN@W zuxLB*3s8MTrD^ha61k!Hfq+gBC;8N=Q-rGK>mTsu2JVKCB7f1jj_I6TBli0dKl49c z$M--J?UrA}&c@b(<0fVO>$8d1x?yh_24aGu1IgZ^qPDfl96oYneb}?DTef(HFeV6= zLZ;Np7-fgbc``*ie(1I3bvezcUev<4JaW6Q1J}VZf<5)jb#`@}=bFhr{b(*$9f3Gu zq$UH81cKr6u^0!>pz!irZUOzqbVFbtoz5%W; zm(w5YR~bCsW*X7Ot>xk}3wj(75EHE$*$E!I0*s-hG;}4YisII!QVk_xz60PSDoN3S z78%Ua7dF6jgLtovR4S5HP^QNJeih1_&B1K1eA|1J5GN?ybe{Fz(CYf(s?^knZ-Y5% zHSpouU3TfZ4p2X`adE|huBqXmaiXjR(x3-ilCK|{)G5;E;8Ba|COr^Y{Ld>*sTkCFY4~PhV*ox?dFKdb6b4pkQU1n4` zpS;>rznvB-mj2`W{$Z46MJ#&Hv-BcN4?^M@(lR&71QXztlt?U>wUY{#x72l&)rH@N_eLIpI_b>>s7IbZe=eU}2cCXw-{74z z=s)sHU`tSbh&dS4195HKYke5$bivL$D|j?O5H9DDiTnBUr>eexN&74?*U-?|s2$@{;ZB+}sJ%jNa*Lz?;--14a#q)&;@s5Vb@P zWjKO$_Y>6YM4l(+o`izZX}V3p#;cW_XI@4`#6bI)|EDP5CSv4e#+sqIq#z8+-en}S zDfpE_9N%wTUyfaW#|2_)DF~#Vuo%UPt?yQdjrppB3Jv( ziC(p#D)IMYHQ#6n2LXY_V`lkP23rusXHiczuyCwJDWf*Y;L!g z^_y(Kn>XqEW3Hs@ekm!@M2?QAd|^Y2@1sX&Q-tQOe2e_b=G)%(C?qQYavngO#y8)x zO4A#s+c?}*leObn9BqoeS8fzdT$y>b_c2xPO0ig^^HUh;@p9TklD<>_CvC{p@Kkj( zIZd5c{DsEbl~Gshx`e+Px|#Mp?C1*BIgKLO!c}93atsD!g`iighW5g-ye{$5YHuXW zMB+>MRwN?!x&~pW;_zI3n*XCJf z(Z5qs^ylkWuawLC*W9f7T@xv+^XBB0-(UBJ6&>?R^X_UL+u9L+XE~LZ*JPxpfuB`N z>7y*yaEIsZc<%y++MU(9SafpSmFykBk?f}FKwcEwll;_0dO{M3L0+462LSnxT-V&? z8ze)K?h62@kVLbn$gGrhYjV>0YU|IRo{w1>yg*m*)v3d;Y~ZEqy2a%Gsr8m&zo4D| z((2GR?r1sg*f-9vDun`1cB61?m3EkmD^JVk&!0K7Z%%EtAL7ok%WN>VJ)2fX1R3+% z850v-pQdwqYPxg4n(bJ3!b^GFW=>~k@0!Iao1^6uyDA!bnto3=OjqPKG<P8vb6= zHN;sr*~Jjw#c0eVWR3p`MhILmGAjZ*kq-e5crr3^K=avE86lAaK?$Hy;08dlC_3a2 zbO)EdMtGcGZ!q^ZA`q8>wfeMRaVqx z;CFE+$ocs8@2@X-P!Zx3FB+Urt*g(ljTHip(k@AMnHqOUkB?{$y7KLjO`K-W(|K78 zMv}MmuvYv;$N8`$T!=F@1MM)y)Zw)Fn3z33xP#1n`|jcMdx<;`nVeFr!Xdq}L*~VZqf|zr%7^#wA#Cwl6+!nbyUO_9 z4cQ_yv;P51czE}-OwF18oW>N`j)&t(h=OZVl?MM~0UV(=R|R3Nu~ z^Yzx-2T)%(?I?dq4~JTz>O?Tox$EJE-3y^Jx?2DFYgX3&t@Uv$7!zbj?PnV=hKWSm^!YoqgwKLaFqds87|a^@ zR)MVEX0)%RV%#xGaC$RMFT=*BumbTq%Pz>+f~}*r>oRdY0&h`?0;MkYMz>$@EzZR)y5xpX?dKAnn zKkygCMrdR)#sEnRjo-6pLLJlcH&FXlCG-c4w%dVkr$LS`kc3prCGl#Rgv53wtT}<& zq)>>w0f(v_I-SsaYxs}TA=7-s*^UjY=!lRllb&w0g zy-kS-ZD!cEADLGZ!DuOETAD+$4pinvQYf;E?>~ItdOKcm5$0~UaHzOxL_@0ji$~)) zrPBwrxq}%$KYyU!a<@%uMOhN5q;>=kUjcmj{Bmo4mBTYG%dDjQaU#=33@`<*L4No7 za=Vg}QjJK-e$_x>ArD2F^nS2)cUsgxPymP7!8=y!1g zw}i{&sCL~P(VIj}1JcHWUAP)`BqjSCQr{YtMz;X6hkGs%sA}{0^M^R8Kn}jI#a0Z9 z=G|TbXiKEXU0sQ}GCe6Qa~Evexidch4RGcZ3pzH;sv?y_L~<-6yNHmF+JVaNC1UJP zK7hJG!b!Zc{80^_vk-=OF1CpArl)A_$PZUyoT7v0$)~C?3ST}sJz0rWAmm0{d%GY% z|4~A)6&99(+p_6vKClrbCr`e8nAR<$Qg)h|dQ22V*i7gCHxEn=x~mNLoqAd5NGYiYM} z`5j^>IF=m)C9*`BLzo^cY*)38Wfj^fI=kOMHZ+7@Aav)*lp~)#FY0Ug^>zgWP7u_E zkkb*bo328P8LLjTY=BbH$jdvwCtvhp$m4UZ`Cb*W(_>1+N|BYyuwZY> zD-^o-+5Ig|Z+})7>g)oA|A#i~^=q;>$T(>su;!oN%~*UME(H`!IF`p}I4e;YglB%# zS7?_(v_%sn4ulFh)XQJL&Z8FnT0ZlI97++UC#zNKdaEWRtA{RINp3(iP=#~{r;qTs z+oYwl8~Ij%MwF{K(=S#K-R8!326Yq?8;QF9)vH$v(!C%Y?jq_AWXO*4KJ~!~wq4aG zs6P@8>RSfuVa%^g=2M4F3<@qX$IV4S%@E0#AWi%A`Fp$jw^Bs;Kp=q_AVn*n9zw*T zI%aKcjh2WokO|2r;?^KFTenl2W6Hoa8y#q0Bwu{m0T@Bx^ML~gyk3q|_c31P+!zQC zry3f7agawguMV$V33ZJ8u@F9OQOY}EW~H`gcoiF&T<69BNzb zSt@M`dHo4J)tfu*lD#ks+#->&k&5MXL1XrvBYMB5d0dP0&TQJW>EN@XO2JF9dIBPX zMXKC;-V}@O&*a{dsvef|@1b8n!2PESOYi?ver$CAO|{DR&Z{D+ReK}^8`kze+_L)m znIIz<*S>+|6Om%`d&a)*=c#E+9bhc`nsq+Y?b@{x=QM1N3J({hugE2<$rMh303E2p z{(QW*rrp`TIuT99F{hIwhENRq_~Tqlz1>nQ!IEXmQW!z|vW^Q(uSZ1!Isw?<3+PI) zN<{YXYWW8NG#k3O)Sc0d0|DRltI#K4&xIw9!&9G; zN>_!`*5n9Lglg00p7Y+Y_XUzgNDltxV*3@z z4VV~K5Va*=)Wh&SQocve;02x2s3-#Ioi^yYVlWYs>D*&@GVvtg;77umQ?J7~k>>`D z6ikCeu7qQa+4ZQmzb)}Hr26W)F3yPj#_(Q#!*4+(Rf1AUK&U?McyixFuk_tWg+S!+ z;vJ8Ua-V|tVvU&79FH?>J3eRv>9wGsVC?(nr-64X003PF?kj@}L_ca(qJaSpm^%32 zc(!=dOi@7}_~B~5WYMBm!!1)UcUq9zrM0!dhCOx(mC~1g-c$^cnev$5=Z*3xs94P( zb$w=4F)xwlgOE8r?U=CN8zmFOJoVbw7l`6ffNQ>Rb1N!(@Q8v_m)pNSRqgOLxtQ%R zv}&8G0a7}q7N8sosd#MOjKCf!G#b(sXbL|nEw`ViEWMxGJ7=2VNcbtFJ;eJnkLq=d zmN+TUkw`91>nu6e*0itW(%zLTSDu;LE3q$F(X1^J9T5esNpE-&m(-DX0Ls`5|2=KC z&;5$!5c+kpqtXVB9zLu!5eGf80a#UK$csP&b*V0VH1!1KK^vQt7hz#TnN3qY{3FN6)L@yy7f>hOPQ9Jvs45l;uHnVFvVg9juM6Ls8; z4oZ_h3QSO69W-zr@*_zgG%Fxc!jBKj$D6`XiJ(P2QC}Kc+9Y6&M%xlu4i~vPmBx{|TsqWf60o?$mWxJ&juU)NU_+2?Pw9^1x}grI`sU<#<3Q@I&03p9$+{`b zmeqb9#USu+M0*@>M)drmCUl~~yirJT+y{q0Ti!zoGSyjd`_?U@5=LVs;WusXYR%MZ zQIQSL%ag-uYTr2>To*Z9|CkjfKq|M`Q48#ana%>-d~m6KQ=voPw0Bb{4ikM4r zlZ{c2eGHctAv@KfRGo^V<56^8g~A6faI9XrGL%#+zCPIvZ}w14-->mklp5SPmaIx}=S~I04rQSB@PF!=q>eFbkZV)yz;gWi+V|GzcH#bL?Vd5N2Q_AQUTJQ^fhrR zGW${+aUv*La=N-f(AFBptL=B0d`mjq5qvlm7pQH7v7uwH17yRiga(%hl47Z#MEmYM zdHoq49HeI6gttGJ6bJ9qgkPqASrPc+*R4dTi}1;E2)T^(P8>T{3VbQxUvy#Ddh)f} zrI@(=BP6Y6s3cT(egX_4#RyayC4!bk=>j+cv-ydwK*|I!NB_{W>j_f;`}YnR>`8NT z^IP;0#7|E672DCZXRe~EL0W9YR-h{oE_=@tW>dgl5Atzk;7ZaZj3kdnG=D>=ki>+B z=2`7&{%H=~E*Pxp`bZ@n@U9hSm@?p35-__y719Y#gw=3|G&)I|klK*8PkizHUkFCf zQ~J1|6>`(AY6T+r4w@OG7kQ}`1csuvooeWDJw#7IBO7eDLEtkb^E2{+t={|LnYPf% zQ5s1^+)ip|8k8=Tyw+M%|JqA#*U8I!gWH;3OIXxLp&-YhI6o*CxJG)o)({&>r;;PS zhVP^@TwTk|R=8p-&depX`@5^Q#$9{oKK}>!l6Qym#&ojmlR)b1q(OsJMXHCOtv*DP zjmVQwhWRMeBTZ#?=^$8=_VlMupWaqw$+bY*rk&%IF6NEqC`?jpW>uZY?$v%qc05FM z%XBp;nSr$kXO3^l_6~Mb5OHSQ-p!j2pjme4ymGr=oq*4qZO9hT+!5WA* zC~mzkpbTX;jN26LOglyg%o2qzX zI5ZXed^d8!-N{zcC_f?Pe;6 zfyodO7v$-#tn?i=`eT=8}u@r9zYTT zwV?T~Rg<4k|3!z?cA9DZ+azf8xIf$@ugG&H7PkU|XKyqkP-%o=MA~{ra+Om9>AVDS zWaffCJ2nmu<6TVs`Y1a&6`Mh(u8knH@@USr1(UyPhY!XLmX_7bM{Nbw>XuC9=HKT_Ga1- zIBp0)&SPK-fW{E^niV6rCV~ImVKS5E5QLSyg|}|sCIbBG{b)A@lf!Xxo&nfNhbbJ5 z$JDA=O(js9X)XWl8%NQ-te-p4fHrlam_Rnza=4ETB_;?A-~7hy6$a3D&G?}{1L9u8 z07?PTcq5Z1b*8|p5aD$4^X4@G2Y!Kpg0%4uJd9BMeVG;0dCh$31ChGg_BQINj6I8n zhr7MzT@S8rLafrD(QzvYu7D7_^Wm%{8=1PgTgD9e96v!%8QlFt|%yc2Zy8#lzx z5+OvNv4s=k>i_NhP-!*F_P;~CE%C1u9m~EJYXO@ zb>Ef-kh?94Kp@PVC`q#2lQ*Fge0BdSTCUI#f{lIULD&-Hob+(mR@a+43z3=AWVM`} zrbmeLF$lKwFPaxn7O%A0VxNh!5h`uk1Mr7!HvhVqH+tAF)ql?77P1~R&YZb(d1=s! z!O%s67B!k98lxrJmrC^9sDQ^V<7c|LNoL3qfkPrRJ}>FZK@`NU=@5PZiItCAvY+XG!cvL_uEVlM{)5a6IKQ)Mx)Nz z;<%rmyq&?75jLC(#obBN(Zpxw+INL)u2KUSQ@Ac4s`Fsk(I3edy3n0As`s;n%GQJK?JXazI@gWa&MAhXyGmfW>Y`d zigIvu=OkC1oPdT%o3;y$h+vElHP=67XEv+2AY6%b*)>>_O^4b*~2l`Uz;cjje!mrOR4TI@o zu7HbvFVF~Edv(8N1ztjw)Yu0)xxpp&z>NinZe+g9PrE^b6vvEWg!2TzU9Whv^9@8F z-QSasGW(jaLbIMllAt7tM*ny^_<7vq(iCuql`8`4b%W^=DeVNNY@(r%?gRNXFAdJj zkS&qtG!*ejX{j`{38i8+FV2{HvlAd6h3c316HT9KV#)VwWSv0jSHCm8e*OA*kpW}N zf1T@@kM02Pcm~fS0s{EwxG~s@I^fANz8TBYAfqet9{CA^e|xOGbFE-v)(fz<$LS!j zBW0Y&Ic6N>XgthEDj`E_W^7J~Xdj>1|6Cj!*ZVy=l5{7{03!-J32c4IxOmQCsPh6&ORB_-;Isoc zn;s57KqMBk79q3_C!m>~v>l;p^qA_5PJL{cf6L*w%2;)hK~T4&a18hk4f>wh`6iKJ z<^E|O{&{OEJTg*HOspyZ9Sdix;&p0fyrA5L=WuJ2F+_nw?x7&+MUDO+Uzz2hp-RBBQcwzK(!z28F0Xo$PJd#y)}Jnp6-joJiP z(9y)ZQF>~y0!ql;LCm`il)sa)wBJ(C8t#LJ$rJ=Cq<0UgV62RhX+Tpg&NcHvUgihe zU?gY}J-3V1dyZCK`p37Jp(jYSUnAK#$S>TOHdtQL5iVB$gWqb#5uh}zK#x>Ia z4=Trx39Jz>O)H}wwb+m^y2Z%Dh({F=yNtTqGtkWsuA#wd%e3F4N2EvB7#F!zrxNT1 z;ez^qrABU@2bo{f71G2Sz?aAUj5k8LinQ*a*1$t9)2JRop9QWFQilj-LB-54nfoHq zg5M%ES=UAeeOWHVzz!7&R==DO=OUP_={)4k2KN0f+&|>u}5~2BzmqcS} zl3~Fht{T!o9g4^?I<#ayVGKZy5am{(7%f;|^Y*eKk&BTV0iHA0iegCs)iOurP@{Q12F5^!oDNOb>V}jHya?r)MWTYxVPY1b^CYZ{#V8ga;CPT^7 z0Z0TDiYo_V0HM0jOjm~Cq*ydakq~3EqYfrsiVY#xvVqZj{oWb%PShkKiJlHyy#6>7 z0{4l-wYe?*Z6hTHsWrK-f@rSs4I#Rzhn+6tIf? z7hf>FtP3|WzZa4lmXnKX!j5`BrV^55r2dLGSUXn!0e!3hNnufljl?Q-Wq^_;H;{mh zk3k({!5Zdc{<_00S(O1WB|uWA=Q=&nLZhn0U(imseu*3=t_xcM zZO{#Pv6l;2IXO-;7v>cgPLSNoQF1p1DafE$LTVsT`c!6JyHb&qAo=v{5FCB5u0Op6(gC$EW|;TX?~j!Mve$c(78<0$?jzp<38 z`m`6G9p)JBI@!uoojT(M{dO|RAxaf=;bg0D~TG0KNV--D8ga}Zq;LrWWKyPl2}z$f?}`~p?l!qsWj4- zfbAkR@(WT|etVKGZ|F2XciWwy6DAqg!^mmsUcQy2OD0a4=YZa;Z;-9b^=w_rLnMA@hq_r$R~4nnvS^PTTL_rno189*+AuRS{qV zcAa&fGdGrT)ECtO$m4N0&bdpgTF`NA^pyhz_oOc&Jmev69)yM2MY`F5Mu~EeY62Ms zX%xW})IeznI5v`gTL3L0ldz3H`{6dZO8{l-*8+~rTaqXOYjhAKJTpPrh(gb5mU$bg zKr~X~aiB7RsdTvDl=FqFhhVq|({YiJCRV_0hW$MwNF12`Epp2bDLjA}M`lGXmBGCP zW|dK=Isxy4zIjYY(11MA?1v;)nO`yJRl-quc~k5QI<$s9F_;NFCCYBOcj49|mgZ@} z9XrTP8r1E$tpqlP4APf0-Qfz3a%^+F8DW4yMiIAS=GQw0@1uc~8mKt|AJcf7NILta zY)~kAFTlI+D;ijofsmj_F_c6Xo7{Og~0_2yLI_HYK(M3Cl)ZxEtd0k^8vz4|sTZ z455d%6H!eK14D-A@=uc7F^4POPVNQb@L}S58qX6g5POdBUs<()LQw!oVBH*v1lktiqALekP6xc&^uTU}0jPm(V4tLYCbV`wF|E5`(0OBC2Hif9R2fVLf#4?k1ai{}X}&>Ya17@f=D|{SCI{`N-5K)S92XU)lLDYdJx#(9siLsfC7MJjeP z|8m!TuSLwj1zZ(Dp3MCDpI`rH9sal2!yE-B9HZDvAYxQRDs_SW$E6gcMhdE_LksbZ z@C)#_2Bfo$DEx^Eoz&{<+0jn`yc>(KL0IDFK$UKQIQUV1hq@uT>lyBJ>SsX^R11LD zvpOAHG{z;z;B#|l&G|9Guq69J_p@y=F7rS?pc;I6Tfe;xD5#AP!ezKZY%ty~I~@&e zP^v}4kHa`JlO%7)j~p&R@s!#yshZwEt)U zz$>DYqytxER)S@(M6cQ24G%gQf=C(sNoA1yd?j@-Ma2Z=9$b^-MH~n+k>(KJOgd>= zAzfGS$B;iHCkPXjAG`45y6%L>f8j#*!G++b?+WSiLS7q}j;KN)7HFaYS=lrYd!!>B z8DhK{k%+i#Zf<4$ynWUqc@(r-+ef(6<5CbPiy7!U}*SsZPc1 zM1OEOR(xd~hRuxVa1%;kmNZc-NcjzVobVj%llDYc2A3Cq4P5CDT<&<}0?=$O0X_Jc zBn3)n%`Ce8>f&`ISb1b7U+|>iqHDI}KgoTcaJuYMZ}En9LfT-LLkR1OK9`w-58{6~ z%?DSs3AFX|;HoQ-D2jA7D0xgOBGAT>sKU)fp{Nl`7(Xt2sELw0Kbfqh$i;7{KPkVI z&H>O=Rg!+mrx0+A4$T!h5eWLL?yKXY@K)LDC&oJ&Be)xb>_r zAvG;lv!3keaXkv=g{Qjw;PJqRh}PS5Hr zT)F~QChNyDjLLwp7xkKKL;HUWpU=`^^ zG~>qYSL!hV)lf)~9FLB?AQhd9%XQ~cn&-?K<_bW3A3$1O+IT~+OC-6^%Id{^!YeV! zWYYZydmwctH|g20{9o<8XH=E<*7m#9sEKvE)u0$9G0PZ1K~ajRU}9bNLQ$#4f>Kq$ zf>Mmp6g8Tqhz-R~5wHPb!xA-!U;|W8zzC=yA|fEt-rrn{AGFtkE;M*@5MiKGLRrXa@vFzir@+XzWK2{4BHh0bh1dqnq77~zgS`}>Ot;) zZANI_)wZldcjG@@cIO#V5ys$bS4yH=cBwX2^BgUjj#@Bz zz)Ro)^ttL3)$*{_3yw1PlFokvo83) zngJC?O056>@*bKLf8kRMHPvJhI5MAJ_zJ{}o7`W8!p8|QUQ%!fUeWaVd*+;99^-eO zuu;ISok#f=LlIETy1-?S=5i^0=Z7`j+YOlFO2qEQtEpJg1MR}+yFWO(+3RAZt?K@+ z!LILdsyo~B`8xkRfA)#Fa<`N!-pmGM{n==nuLljfmu>ff=gTCEX)haazUf@1b%B$~ zVNEC!-L*+G1F-Azu=WQ?TMR)m{VvYC3${i;A?VlE>5(S6QFQvJHHcPnT|Jd}g0yJ@` zRBCdrqx9Ha0sDcY-8$YabAlSK+%byKnh);q&j(vMT`Zw`e>Bn(E+@f8gjDrH^v+fR zerPV=s;1xU+(`LR20Bbs1MB(Gh}Uh3^+opRICc106?HWgXv(;WH7II2my?g%C#Ii! z8f|C9_N-fcyBmGbKp#db(L#A^(oAt#ujx12&v$jzh`%d^u}6c<&yPoLAWs}6(itWa zKOnXjBbAq(lG*ZOdSdDGu`sq&8mo%U?mlJy6nsf0iKuD@{TS%j89_{wWnb8|O`)}v z{wWiED_(_}rIZX+5?%9ZtmUjB89FH8iJ_QP6Flw46nw@%x%P+Tr+dJr8tgMR*9&&p8c zmU|&7_WT=*YrkIj8L9=!rY~i%ft#y| zas(4Gp^j5?SCPjK-+FD*k#2yjD`B5WzU_H1-+ z3?4>s3X@|;j|#=t;Y$h>pEW(z0W6XGp0$YWxclh`3Ky1&keOz`7N+lwq=Ke-pgB#u zuz~z!_7Yg{cnPr8r3lYg{REd zqXMzp)N}XV^mn1>h3-!mhKC#r8!3~^UyauefbmWfrqj+F`=nMT(w4hrC`(Al0y)$F z{_Pz5*gd(#4MHEuS#!ahf zvT+lSJ6+5^_cS>us3d%#!&$a@5yGZf!>JU7nJ?8)5A=1C2guU|>pX(V1K9tyMqrL= z@?!Hp%DREH$w|~0M4@+S`9X@2g-tHW=q=3@vl46R-2kCnw|I?kQ7MBQ&_oX)REI=* zSq|(!e{l_|8BsM*sqwl}Hn{fdnyHXX*y_cetN)-S0Y>3X@k;*$jNKAY`Jz0nx-WBX z=q%zd9#E-SEkjVG{C=FCu576w{`(gqy#y(7q*|z|n!*7lFISAL+57qWc%K3UouuVpAv_ zx@phJz%K!C$#}L~U-G<6fZ4w9J1f>NQ$zGZlAkj_Q0G(1K2nO}r_}{2{#pf8X|38x^FJCzXBvIC%<%78 zsP30PzbS)>#5|-RIAf)K|7mF=>#Mn3qL{-vb(>O<2l%h4Mq&WehY2!X!uu1p;J+bW z=C^2lK!pyqzT{t>TphfAsKgq?LNTl0QAsIT?C4%S?FvMUWuAa7A{EM4)U;hHV|89p zF4PE~lWq#7b;`csi)E4j$un%u}{B{+Xup}msfy8X3OVYM86YJGQjvfcD(IZ-=0Q6#O4#4kcT z4Bp?A1lYXh|D9kP+^ThJ_rJD&wW0H~HSNAIpT^xft$al`Cn}sora~VKmr@Hq5#M84 zL=e=ZW4D2qv-7&<3P^yN5LqwfD?0L|(@r+zx2DHxqoKxt`YNTLyqMhQ+2{`gXeK&B zBs8bC+>2&-jJk>ZjRr^O#OOm*m@&8V4fM+{^dk@KWca7aLkCInXIDt`p#b`XbQ%U6 z4>hvuW8x>RgM)&tNoKCDFV0I&rXzoVOlHV{Xf5G_>}1pN>LXDaEWkNF__RvXphsJ@ zY43v4{dJ`Ai%HI9d7F;DO>cDdG=&}Vn90J$-216cG%w^29(=q8X!ArBBKMaCu9Rh( zkHMTt@~VVqDc%44dPpHXd8azu5*=L-YL1F5><75tn!ga~0@2hq>Btt)lrV7Try-ui zB;vA_(gOjCupP}k)+s-(X>I@O`Q+?$Q6f}E0%Ax7=kdr0U%Gi@=inAlMgS>=QM!m1 zzwE4yen$2g3PDQ*B-14R5|nw}uhVRV<)$@zNcPg_BFRU?Z(7Tuzm_7VkexxHwG2UG zjUH3hN6AnsDbA{`>4=aqJLq?C_1`wxuxX!(2p0M7M78uoK>CH-6IHat{QaMurwZ(< zJ_8AW#0CcHzkjsk4;nve+q90Beo#UQ5|vEpaEGINqKJ4>=+tcfKi{$B4+!XoN&HpF z1?_!vsn!doOPH7pxE|xx*ux<>B|qVoS=rDZ~!8sJO-)=H9iZqsom?`E6r5E==9J4SfHH#*NWqu5U z>C~l%kB#o5GJCRQok*Boi*(UU!?-h|dc1~AiQ#Lq7H|EUo~@baK9LCE$3~5+WMs`L zlNM456AGkSP`Ha#+3u1tt?8K@Am4P^Tl2qafDeYl8JL=8x%6JS5n7a}Pi56fS0eq} z;=tGR)Vid)R#tUy*9j%NB&M2HpqaG4w?FoySrSwNzQ;&aXupa(^RtXFuxauyf$Koo zUqOdgABJyULiNl1rwzc=Ejyqobw`)s-8KbRt8+`f8!Snj5r}QISn)!8_)vx!L6K6Ji zH9)!jry=)im4lP5J}l_fd%)J8w!I5!AoXm_>{xy{M{X z%a>;{*QcPg)N_e`MjzaG^xdduX=Ztc3Gu_Tr)hm`>LXZ%mLFe!UAq#YV@+*w(og54 zRwd&ulqy^d~d(S{?h|v9P#U#Bj3hO~j)B zW*B2V!Y#v59Ofrk_{)MO(v+*#XerU+oo^jcLN@uHet-!Uc{!`7auE$^{x0^A_V08~ zJofO1G42P)Lp4KUxjXv?xVpNo+jhWm<-3%=BI=;xw%!+W` zgLGK8ou5`t2P1k9AvgP`fZU|SYcX?v-CJsG@K#2BXl6_yUZ)=7`%=vCU%XmeuYmW{ z<2@eFP>Vhx<*PgSqZPsDerqndPjD+EA`XFq$R3m`%49N|?Mmdd{cw^O)4W%en=X^$ zgaJ)ZIMQ*n=CwwyL+j=!M0_m^j@UD8-w1nZ-L3kL3V2A(X+*{C!BypX4OM1vVs&xb z({!e5_WEwP@P0I8)0DKd$-GBM9QFOkyVKgy2u@72T?v@Jm@QyqHS5twyBFTy_-CBrKmX?<4!vHTfANG8r$amA8kwsgj?F`=aLra1Dgt+C> zdk9U`--*z5zy;KlA=Tc!EX(mz({l8MK_l4)Nr^I&+bwefc)G1ZdzMyc8OP&B1zILl z$-y$z9<1>1pw0^EL!0f82a8nBu1Vi;Ko0Z(arZ)}Uo(RP+}8lzC|OKT?HL)lZZ7Ke zm{OW4L0y_RA0p}IxpTTb+HEc1lu!tN+}Mkjw^$1C-SME)a+=oCrT?5W<4Jh7DTfXl zMk`P540`vjLPAX`%SI zm@G)0e#c@z%8fL7kyPi?}KH1^PoG$c&$m8=1u3kJGxY$ zpnoXk^ismuwyuTu`f0UD=Lj)v>!Cx3G?n_U*_$ZEXqRN#Q~23yTj8euE z%YDo+q=Cj$vzp&)Z^A|KthIMg^4ui0-bFXuO?ZTqWGCORdCU!;`jB3u_Og`i_ioZK zE)*tdd3kv+KOGrjZf^cEE~q{lK2!{#&17OU>84~3rHAlBdd6iVc@NxT0EvnViB(cx z|3M@sd3Og#$HF&n4E6N%=ov`4dGqJoY=v%r?Yk&4ni(1z!UfqU3?MKrv3JCs8PYaD z@9UxGj)^p(3Z=Yt$6guaFe3I$crrUCYl7Z24^sme_w;BYd1yy}9*95`T%`LedW^i- zWY>Z%uBN6=looLT-ASk>(5a?N03sb$!p`c6tp9@t@hDM}lud>g9D4nS$ou7-195j` zM#a)Xdtv}_lrWrx;>Q_?#x~!sYB?=Vdkhre9m=b#9|D(2BtgY%rpPG35s@w&Lbn?) zSB6SMt?V3lmJH8ghJj(y{*?JMU^WXD1q(3c;X}JOZ{FaGcBIT_0sj$4=Ey>)5ZU7a z14`95Crz3pv-z@)&;DAqY{d#YSa%t*j>AMj6;jMM+akyAAEwAq{eL(yqW&czECeDNVfIJSsr-J5LoRH>sA>TWp| z1P<(7(rZ5GC7%GuJAQpjoj#F#s3;#GC9hPGL2DFPnOFg`Jk-L%DX()&#kzl2ic~8@ zs77={_*9v;hPW~Vc6~#2P~*e9rk~&yP3kY{i+Ol~xoVgm5giKxj2prQB%YZPKjtmAAJy8Yqq;El#pB*`T$;uesJ6 zDT=@4WXiZwC=aJ_A4=6nuAYQ?q6P;=-E7DfR9nF;BN*T)P4Hb4hMJp4fES*_Nx?HZ z*28pWR!)w+Wa}~;C~W_$^AMAy!&mr4^h%c2(gv`I8lIs4!ge6`gJNUPN|`AQ6b>D? zl`fm?t;lZ$O?2z|^XFx3YvR?u#foX&FI~DsJB#?xd(#ye{}mAt)DV>_0i>Wm5^7|8 z!O%bfOT>oOTN)V|35Ty3Li*R*Q>r21crV{oe(Rn5q^TNvJ@tdNfqlb?e(MQDv4)tg`)6Svn=VjQ< zj5hqsx+@Er%x!CbkE)&4M2$?1FL=4x*KYwj& z5lO`**wrSve~(Ix(5a}Sh$5GcXGglw_%E_nG^Tgo`k%{X)ANro9C!z=EIbNx|6xR% z3yUI1V&z*YmpllR`vgd9;_>K&zK(xIMMWi0$zskCS{V6bau|mhCzWKRRbwBVcAUD` z`~Y3cwv5?2K5bxV`kmjQi=H=O$YjmOkGk9c6FhpY-NFCU7ygAdvoyfrukC2=?V9iu z)6H_9m%{3+AHJus9avv!*Q6lNHI6*-G$lm^Qgok|8$Y`BtaB z1Q1Z_GYp#?O$@PCZjDOmlRjJY5MP0G-Rn!W<3C)0#ZXY`-i>&5Y~Hv05zTxhtdvOI z8hH3DSt5PIGUHNYCBlCid5wiw7FG0xM5c=H9dE0uvV_RpT5ruXDQsM-De6Ib`8bim zCHjK~mGA!j)tfi5QhoN$yCji+nAf!>115bV-6A5%i1Ani1McIUpoT-IP8p*#B?m(w zulJ(I57af2>dd`Anhp8S?fjP03#8EG?U9gD!f&d#E{O%_sT3WZ-uF6+i5IdM&#}OGSGx(HW(3{NU(iYet}EJa zz*)UH`@DmQ1F6#!mzbli;+L+YQB(QGf2)fS% zTaun{;(>{3Azh;gD@_n|6U|^X-AeuX!*nB4MXy1GTSPAd{RmiZ7%S;mL)8?k|JY*( zjIt+BpAud6!EA97Z{_7V!-|$>>to0Ij`7TI^+PM8;5lo%UrCG3JR#LkVY2gg5DD z`j)!u{hGha=$wpatQj(OLb>@!?rCdehI>_jtK`vnQW+Svh?k^DXGmwC=g>&s6?{45Kih zc+V||=a`jZ`~!uO@8kE?{@QonU4hX=85o=)(;os4m{scDurOiww+Dcd@x0RR#j$*x zcV06R0nVINLwQG$%oG|t)%`<&NVl$Cdwwo`7eqw1jq@U8J}I!$4iJ!e6Bipr6zz8Z z@UK86UD0Yl-9dsaoDwfWz+qay$$3>el2lCvEw6Ph#qz@0df z&L-aI&yd*oINnLL*qOEv>nkactmwOU>d??oj7@3cyM)dJ7VpZ-=l2A^5FPP3?d!<_QSdjUhm zf^L4x#yTI>42&8@)PrCP{l5HiBze;^KfjQ5>((hn0G*d(2V_(r)*s1I(cpA6*>d{y zX=2+*I)`4#utn1M;{=hr(u&-I-Vm|^nog&|MDFzejPVef7E3@*7{9-I4wE)!_unz$ z#lSsj6rrc13k$Kd3(efja}ezsq!t>6HTN1K@0+lq%n6vR__6E$n5ubuCVeAjK+tTd zkg#dCwL|5d!!Lb`Cu4N9z;RN&6V+q_jl;y2wvUYZ^@FCY%3PMtl-p>caUr4+A|i`F zpzpCBGur^5&Lq--nPYgKKx8^vKr}Kn&5^)JNa)482``rLjMQF+|H3C~lQ+Md&@)-4HGGP#-LAZaa(b1gLj_tSO4LASp)ve8i_*pDA zQR*QPdByUG>6GFiDH#v!cqW<2O%k)Azag9<+h`k@t~;yE_^t1IgonCsC; zB^o7weQ~=r{6k#}5P^7vMY#K!+-5(5@OsKkAMsRl53ZdMZ}VXUNQ}wi`=cT7*x?WZ zcEoPkGPV8ib;a{c_t2nln2OyfpGO|1Y$wGq2}YjUT@H!#u2s^b4FJ?rXE3(Ow!ZRki^|fFkn>snWs%)qgqf5p1GK?QyYobE`VIe zkG}D)Nma9BA=8EUUV1CkY#}()X(lRY6t*a$7v|@GsiUK_%HN;nPhIcgM-8R>dN~yS49<4=or`^ zb}EwMOjWPH{OYS`bIUP$k8^yt=iS4Okh1h^J;mB(Xp zu)C$(F0!n z<;sll9SieLKlZ3$dXVDj6SVk(hLIw%v zs~6LQ(6uVE*u)asptNyY8Xx6wnqu{%J1dSfFXR3&iX}VMD+qFDsnxJy!^|E{eX4M< z!o;$|<(|zI%gz37PncgwFeq4wXysn5M>eMFpm&W&9ygDi$@!hRZFJ_PZygr^cR9p6 zEsHGW!homDSU)Q|)8DaW&cvmDe(^?Aq>r?=f010t4I6BC4q0IWM$xs+1#k(!<()nN zk(31XZ<#x%2s@%aj4^hOH!CUAdq5ZE#Hxo&eS98>xF=LBH?VY^*26KiPvFso4quKn zkfu?LOTzbS%ODG=4>RkpS{>ke?b@|WW=z%gzu)84oc*cmxhPjA<_2-niCXImje?Rq z&adNMPEbR-b9!~jh=M$}D&$64+Y2@qZxjaUguGf{9pY}=P#J<5b1eS?GVyM!poRsB z&a+&WFI}26XTIX>;di}_6ypUFjbFEa|9;Y_ID8uhA4iW0Uj{H?>Bhv6#Gvt$C&#=m zH#j}=$!26>7ACV0Q4}7lc61?AF|qy0WKWu3Rb{e%^0?e)na1cOg(?Mr2??A8)T4Af zsaN8e`1GiBnQ%eR&tF@;Z1clZ`!-#4be!MqrCUFAgez;&uHRdy5}%~Ih5n`68=V-u z6b>83sl1Ly*LfNW+g{k<#sLEi906~} z-|xEm!4B{94xLL<>nHjp)%y7O#Q9G-WLt6}>&3DIsqrn5`QGYVe$YGgU2UoOM#@Bo z-2FFMSGqe`#wQ!cbclMs8>j8CG+DeN6^qcPUMFY09=xJ0T6=Tm_-FdGxV-+>>R*`g z(7k8NkprfO#91^~OzTO#<8{EQRBzrHKkX4>-|H;#_=5*CNqLA3w~=bCuWDttda85o z!Dm;?*KZl16#~Cl?S%>}R>+_-@FJ`jURUZqcwWO@7cYz$RTH!3hBx;$HgfW{^_c0B z*CjzV#b0(=>E4~RQ`Km9CT7#S<}OZ7xeKaO?%au6QWu5IxllYbu5f6^^1?eS0KG-T z|A=1bjvYs(p@HKs&EgWdnwgoco;sJijo)x8-}`X&=CH6Zq#BoH{^)BHqSxp)!U$uI zqF!>USMrIG4!r;M+1C_H`ted#ji0qtM7nuzS18VYPIO5N`83=#5K?K|G7 z`7(kO2Y{V`d_39eyXxvhBjcjy&%fr`@gKVPci&!kIubQ{v9NR*O4JO;o@O;V^sj5^Yb`UQil2?U{W`KAO=!7$Je+@pHE`_bo3PGZ@?qGyF@Dor2oBC0(` z@~c;C;-0wod%L6hN&sa;!&PCiLrUVK9-gJy4Y**&dXZ|5UE2jl+)|}giL~=56 zT)N>TLlba^P?%tV*nQcOC0j{)0Ox=mju8)X4+R9q`BtTjj_6K2C7aIm5cEIHGGGq3 z*cy#*lgitER1F$)Z|QV1*q@}^rHgfR&=!tAZ_J51SG?Pbh^{F&6vTodm&At>^>C2v z_MX{+n82n@o1$Z@*HsX27=w-PP>M>CbT3O~w|*xItxocP1r)g^0piPeRq5D$%f!UBj$^(Eh!`s#*6QQ z%G;J2)! zT@9*eUvl=`Ik!K5>GToks}lF?PH;}{_s6zv+X!hntOCpe66Z98tOUD|1^{X5LAOhI zd~CnUBCmHo41Ad`4s7*%bl|xevvYHE1>~mO6+!ScJ&tB?3U$gb=Yg-O@5bz?*zOL~ zMPhUOgMff$br1U6+1Uw&;UkRu14RAq(#i4buP3X2VrY#lgvi0kg3+Swp0~}6sCPrEWb_N0; zCU?TK{(EMtYJ{e#ZU*je%HCP! zWsJ?y9sO0cKXKYiI7H_P64RKkM$q(n@G5jED42SdwRHGb6*Yx;TD1PQZR_09DWeyq zfH*J|nN&law^CjEo-=UugEU*o?c=z@5iCAfzXu98LfF>s>BO$txic4X1iB^wEksbn z1-zIYzgMs>K@){>gI##6Qw7oJV;`%QQJ=v8nF%uvgrvg2@N-lALG2xOWhMb^JZc~l z8=R!3M{ZPh>Soj}6GNl7B&v-zuu%CK0;`$Z;n)#}x$nVB}kf>jO2h>m3k;9QAOz?&jE`Bun>-;M=y9I^Q&E97+kB!Ho}x4?AtTWa-lB*dy|#*QH(L zi$OzBGtSr=c(!sEuqdb#ku;M?S6vTqbaZTAld!_`sE=xjig75Ijy{4Cq5KjKqklsV_n0Dp8&lT^<$!=45m1>-Ij8gTHwS}hz-;)ds zEfSQRJE3Y@aM1D?DIg7Tr%kpjcsZi-`;${qr_3r^0`_mHLpT2f+Y!pl1mw+(0! zOC{!Zg#Y`kN=ogVo)SSzsZhYyLTw^)8Dj`p{psdW>`~0Wj z{uE$~2XR)VixXyu1r*x#Rh?`Ec@hO|h*@hU3zaxx9rjcD#sz3*@#MneQ%Q0Qo;^Dt zaTh9j1VT7J*Q)1THG`|+DNwv)=8ao&-h5-J+JEELO@woJ4tKPWrDHU@^sH#hh`t3_hi%Rvi3Ku_19MAhNk$||$%v|%P4QBk zyxdA+Du&J1t3a_$_MY*2gjukp+h}9fxIueloXgLc0ZYJ>N8!E?ZxCcu<~k`67HwDS$)Gd2Jc?Vx&=VA5=%hn6TblLz}Yc^YYfC`yY-$#D(9pQNs)ji!EW3CGjh}@7Ge{HVd{=^ecQ>QSu{|j zfbcp~0ZNh=aS8X(FLMN+fvU?S@CP}+nQl@z7BrA@v54jacEsGB2{lABNF&$3WPrpK zSPJd-5%Q>G3t{|WJ5GX?|M!p=RZvo+PVa>LI*m#l5wOG?`%9`{a@6yc*XaTL+g zo_qD`Vl_&dB9RrrI)xajTa281KufC|Il|@i<1@6mU1VdVSmzr!`FHUiB0Cxv=N-TI zCdbI>bx-J9V8x9u>03R0r$X-zA9B9M^2s_HFWtX9~@dsYy7+7ljjr5YzQW5%C47EI) z=|dunq3IQ)=}`u4%CQn$TslIDtWxJ2XU^K^z94=M-ojEwqzOY6kuAtZTk-@jgd_MfT`}kg%GWaD>4qt-y%Z#sh_+og?@oLeLLo~>p?h2^}%X|+} zEg`ixQA;{SFR19=$^jAiLqP(bCS%b)KG@$NUDNtX<>9f$%|`k)u_{OQ1Ummro*}n3 z3p$bXPrmk|*&7O*KS%f8D!x z^8O2W5K@eiTTHajRMW#*pVkRAx_MV+H zG;N9xlNvYGA)tCGb&c3TlED?E-$5#i!j+e*DaEydvN9ta5luA-vcVAs|Pv(NuI zgTbiuiC&-#cvi9>#UiufbNy-FU{G2Lg=8wCQktC1>5@7Or3Gry_|ZOL-g)^jGn!%} z0%`nR#0Q0oc@vxOBFwuUJ)1?SNs)z%56+`j1_cEX)=vc4xX~*5J8F}Z<3&0f=M!%0 zGtfdjfFk23&QtB3iWhetCV1wwo7BR=3d98&ElrD{lC8Wbl6CFdo0xZ3gqt=IK%In} z1(~Yft2Jvq$T=m3favI+2opTUIoauc&FmNgwCV5|HDtg6O}xrbay?2XLp6;E+bJPa z&T%3kYt^@fzBe9i4(gu)qO1E#L}=-*3s2A*&+uFd(Ed>Okrw}2MMkU^l_db-b=#WR zt$6)t5cyK3WiMcTWs zBHQ}imf`PyxkBG#d?)ReRXUNuI0e=$7;3&ZsI6wME23qC&ATpHT(~hxDlck~pQBX8 zvKyJybM`-Rnv6*ICnUv$KUGyOkv@JQ z{f*#eEqpcY24qz7wOnW@upg|Tl4j;LHms=qK)nMB;P$0!dB`T)*8Y~Lo3nWu^&!!F z0(9Zewqm@8EWWSgD5_=<$Y#=Ov<`l}+c52+(d|NI@M7J&RE^xN2z`;VS(?1-7lpSiMdm+yH=ApCNG6Rx13IFE;s+=0l8QPtnH$UJcErUXM zRH;c-k(+mEbZMptb|?Im!)16El{JYUs>;clgxTMs^8rijpeVd7)U%Sgdw@C`j(tz< zdHfJ2j2g&L>h$UfJ$YYX4EFe=3my@pv`q=6#vSmj6}>kZe%iOmUqO~rd4HmjD8H&j z!R~X%Q6U29VT;wuU12t|^ zAQB7`i*QM6gA}3ngnAE%QB@9yEF3mHT?2lZgi2_Xex;HbScnGxfIj%hfbX=65F0L} zY&1HR5;{>ku_cax{~rp~?>m68KYTBlTm#1~%)gwol=kZUt{TmP*uTqf>IpQURKfez zOvj@^jXi3~|0xY(0mDV|;OC!v2=_lvd(?xYp?}~`)fX;a?4yz!EGmHVN6Fr4ONfYd z(VhPHmnVD*pY^BDI=_B?eWQ*$)D4cfXCk_c5__r(UcWw$m$|o_w%*c}3nVQGo*?mn zJY{gX<|u}X*}^fMS)2Sy{6g?GH_oW zZQjcDnn!aN&qxB0W!Zq<11E(&^ynA_8j4hKb9EohyhebIS4&rICE6K0r)ItW3J20e+_Tq&92F2Y3<6XP*+12%wD+U_=$7?4$TTB=B z$u0Ph#gv7mxdatVZ6lG}U35Q$0j!>Xa+b`L0OMRY^T0RK6u=6~JSF9;sX-sNQ6l|8 zS}-I7lvRjIdA`NU(An8pC6yeJlU>*Oz&E0_2NiF-lKODTF4uhrOpuS*(6m39e-g1G zGJ0mq3#Jc{5woL5>j;+xc`kj<319|bs@Uv*vODWTdS){x{uO)e=mDaFH7oYCE_O{ZITAXRHXd*+-# zwPYfFkG-GT(8k!0uqoA2W-5~FOE96(#u5s*pU9BMZt6ByFPZ*ef^Q5`)IkbF_=_2Y zYMRz$$qXgvQ8b-}LEtUZ;O!_?SX+T)ul%~?VnbD8@++9UnkY?1+!N4ot`$vNtNT1Z zKb+G@^G4u@1K{}WCjC@$5<|awX)Y8}p+8atmm_DsMSD&&!f{u^U{fEit6PK7Y&16S z6g!4MiIi`lUPnDk2P&eizR~mtYg$02UE@u@YW;JCck-o=#Ril!u}C4)!C<-YLIu*a z=Sk6A&M95RO@Ads653}}YJ7ksrJNmOh@HnY;^|;*CQJ}neV~^jIQvICfio0mo5|b- z>hY@klFq>gVV&(xEiav;sph0tKMi2EM!}00uABsg;aKe<^->VO($zIyEj9p*q}$$4 zLp#$oK;uoCW#G#LKTG!@*H;mo+H?lYNLfY!FEPggtfaoAF=7ND*$^l};QJHQUdi># zMZ7`Q*`X z?r;H>O(;w&P$*mGxhPds>_j{N%CE(V0)a`xh%yC7C!sl@t=z7&flX`Xf{uq&)s(LQ zWu#l2zdFWY5(;`%IVdV8Oq{JEs>v~R>klwbu^?2C)&We-$RAyK2abu<-D*DsHM|It zDJ{C9+2E#au2Q(IMR+c|_Oc=P2Ue+7-zY2C2C@ji>f_Z=%qs7W>L@*N=&WcjiQ8I; zmdGy6R!cG^3K)FNO?;VTt*RHU6X{Nv-l4;Vo~;xa9WNXd?>utAQaSEHPE=T^WUXCX z{4?Z7Hxw7t!eJ)LQc?fY9Dt8aXKc>j5MhN+s+s8_a1(Ez0N+JxC#6G<86?tPHmzQ} zJ0k}yd)7afP$qTZ>Fzs^z7e@+QLwSLR;pywhqOm%&gB36KB_@XLP46E&=jwcmOt9> z0czhd8d)?@07j0wJO9}M*RSnfRAls~xF85cKpmn6oO~~do+Z(ZbB7*m*4ZeO(t!)mNpcsh!8%=RNT)9GR#CSd%6u?n$?%$YKW=`$M9Leku@!Ov(f8r&tdg;B;XY z{Z3mo-k$X%Mo}otIy9}+b;G^a&!cQc?%3i%h9y+ByJQ-#pAVXM1Kzy&dBu}iO&|Zc z2xe)Nm__GK3PMW`W?!Rm6tIqJs`@Z*S3;yrh^)+_Fv*iXg253^^5He}WTn7HL z(}k2ov7TT1w@b3M~T*~CI zminkjbOkDsCmNXE$oS}WhxN>p@v3X#(#u64USVWa-U5Do7{B zPzH7Ftb0t{(LaUqy}hH=i@iSgAt>wGexiu)BLU*KzigXg(eKENeyU|ynuT0&lh{S6 zd*~~Gz8I!V`L1=(?en&bNVw6p*+|dNqz${x514Jn_e@jIM@)a0spp){2HB_3?SA|O zyj`XK`0=B|%4h2{FZWO+*fa;T2yWhe=(72KH7Ja(Vy0&M=*P?W50IJt6OO2r-A&xA zks43bwQH$}@2h#Ft>Epsq&LD0=s0HX7{!Uo z|7h0f_VjT*Lpbh3RmMI=w{Pzqdiq9JzvD6io5fP8kxdy$s1m#MfBs@pKtRA0i&uw> zKf))oz2;_-1!Hh4sRkkF-Md?Xtn)T!D)L)=A`wu_A!|Oo^)1UCp?b&EQA1`ZEZl71 z78Ej}PhM?qb(EPcT7vU6!@Z*vE+xG+uN{s8auG@G)KlMlT92D{dj4B>UR+>hbn*Mn zs1ePB-aPa9Z~Xe#nqQ9q&>6}xRjDWKF&#hx!NZoivVk@Cwgg%`LQV6i{B40EzwuMe zuDKlxTYihkN3A49z14HYlj%7>X5T|w^wg4PSCuJ17NPTgA9|F1zMQ2JOEt{}5@D<% z1968+*VY@OY|S$rUv3Eu;oMQGI*Al@;BGE}IQLF5(2mCKc^2{)qFwPb0XTXnQNoQMeK^ z;YfV!-(f=*&yeHBFH_vPbm7C_8&zuV@B*YLO=3S)Bm`;>=ahWhWg5S+Pq_+nS%H>7 zRzi3hs+hky1l3n^c+!XVZ4a_}{8Pnm-^s$<{VXFhvvB*&DUq>x^y5WZ>zf>3RPluW zI{VC`>FR>;bscVX8u!I?_A0!a<{iTRbZ)G4cIDov)D-3lh0(>OVes1R6tcn!rTfu6 z=EPew^+Vhox+=P?(=5}ri0Ej0cf(x3vZVjml{e(Kwi-_%N?qXHSmk}105|h&Z^gGA zH1FKk7$)-jl7~jBci>PSm3HOnI|qH8%>0OLO3dLMyG!pvvOoVs(Nc4EUXCsf{Cu7uo)Li8u?1PwA-^~e{0N&uQ9>nWK8e2f>}=^$z9G*C(zSqme8Z{<>0* z|I43>i;I^P?Q*r%yaKJ9nI{f5BC&91K4HB|P0PVv?xt+?Zxc1UrfT17xH<8tvwA)Q zAZxB01+qg~PiB>Z4<3)*(tYqz3+I7~6E)f$vyQaB1;!IyLZgVgm_l9K?N7KQHPmgk zD_`X$2g!AO;I9aOsr`dTl>2s~|7xr{#nE|!q_Ii#0xnDvBD7H@I!-GO^oVXYvWLcm zS|pJQ+pfzH<8jBTvF`J@12ZBEhUL{v@#~dOQLtd}{+ipL$F)#c^w6&7!{ngGO>-6frq?g;RxYSri zqG7P+_u=pUT5i=*;ZdU5mYkg|=X^3gmHOz>qjfx$J1rddy-JhwiilOz%GIw=j!Oss zUGA=@_+X{62%eMJdwF8S_LM4cvTdMh3!s5`dYyMLd;wUsobu8xSns%6n1;t-czb&F zP|VU;HVr}$89DLGDR*{ubN$~SaLg42CJG(rx z`@er2&m~y!E}uIEz=Jf3sje z&7n$|`*Y9bvkWi1y;}T{DrLoH?=@coAm)CPc#;FT30b(9CDcGp4u!#&9dD~}an!!; zlh69@J&`r}Zm(4Zf2^rVbxsF|B%m(?ydzuC^43%kmxq`8gXuvTT4Zjt#b({ zOc8NW91u*b^p1gV*C8Kh0a?i{-Te8cHPof=(1ms?FR<3;*s^a!Q&}!o;{~_2*CE6j zn{W?zDSoZHsmH9CeXSH8J`+L`ZijcgI|b6gA4mi#RZ#qP%bh;kJ6z_aZ=*rSm1ra0 zu;sTqag#^!6Dpe1cN`}3z0ox@OTvQa*mL?+sGLj6X>79GUsup~W~8+-yiC7b@7C?Q zZS`LByvJj3?1cGACsYYM8w1DsJDxj}bwp8$(x}%fzC=BrkyW#kmMy@?yj-5W_}CWT zy$0exBeFro2QU5m&nE;c67=levuB%9`W{40U`1Gcj~+b+(}%Nn*{yzyJv^S=FKm9< zbgCGh%X26S?+g0ybI@O`$vYAVSKf#zDRxnFz3Cu7&4(;RWkgm3p*qLszcq8o(Zpsh z7v;zkP0t6#&x}h9op9tmJRJ)?BgHg(**t}5RBWsv`AOK z$P?hpr)hp233<(3B2L^cw-EOouDK=zEShg59{g4;4guA7nnU&ff5iX8wTSpQM@O+$ zNdz@Xe7ALD%ST@j1rnlUEl{+iN#@oN6O%rJI;>5=5sI+ zydxo=!jwtq8{Z*Nrc_y0o!oK+31>+tI*U_MC?>dVG3W3dSB?J8y36){Xp!;re|<=; z17g}LEaH|W?Ue!I;_LPinp^;`TD)k{UzWA6w-K^_L!3S(pngN%1_>49?SnpC(>uio zbI1~wHBVkxhC=W;X07jez_q!;W3!kR%84fQ{QL!5sJ6d-LVqCPTcBa4uRbJf8oY_h zx2IcJci*FT-px3I+dB7YxX(H2+Z9BL@#ik$ZFk`ym*Wj9y@J>Vv@1goA33rh%#-%> zsSNXp+I;D83q{xiv7wI_gvL~SKr!%K?oF%X1D>~=*F~}D{G5fY!b(p&TyE>TldOC` zu*Pl4wBY$1=HjI%47lGM(@d+Q$G*C>5W;Ih;fmqB4fP>GwOd4tL4J zlv1C_y%N$~5vH=g(f!7^iW3Wbtry;%ZI#;c!pO-G;R_HwnyP+ZmKPM`2{9#=vk|If zKOD1G*M)z#PTi%Xy$)_nyX7vaFC%V;pAOuitBSnWkW zwly5DF7Bznf@`u#&&rYjx6z^&*g>g!SH30aLZ)e6&^m87w+TmP%yzx%rwFcW`lRf< z9k<(Al(;7K(+%!m7imow@Qi>%Q`9hs3JFmhN={arn3?rWu?RX+8_qsihcd=@m~C5y z??TV+uH6nPM%L!2S=ajhD_R=eeMZK)Rc=ViZnK+^(S{{HSvHX(_IBAPs^d@RIB$-K zbFRpu+7o4)QT1l!7pF`{xW|;v*7f@+H3O2*g}8s%j5eCn>&S%B5)W&(3(}boCufUDGKttj$WHTzrhUNnz2*kqwocx7f#`K{d^4C55sCS}r_NnmX70)hB8HRV)f3>CcRt<@FcWYzLKO--|g-o1PHqxA5*ROD?Cz z>bJBxZmzNa!TSF%M;zf@i2|q62+xtIc$aGQ@Z8=3Zzv5Nw`}a#Fr0UZHyn_1qxRUL z#F?v+(jCulk{!pLS)D_5*|Dswk*auhy@TTJ=auo&shd49$zck*D%nrc3E-BILLQqD zc{nL)sA@)}b@mgld&H5&6^(D7+ppVPIdz=(u_|K2Xse!?4M&H!Qp8KTie=dHc1&;! z#mMOrnT2%f}Qlk0Pm7mzMY;Q=blf-202akDL)E1fT` zS=CTs)i`+E7fVNT&_e{9V}#&j^iz^l@yI0T16)jQpNmB65+UjQukFH zmF$KBvzY^&HnK4Go)c=hB$E9s%bM+~`7&1Z)U^)73Iu|&s()Zr2)8i>kq%S;*s9_? zEY5HkCjtna0`53Mj9&9+U&}wl=~>mJOpOY7up8^Tkc@a}am1nfNI$=(JE{AiA4Uhl zuM{w|$MEF=$t`%z+bftW|HYEa&0awTpeVGnB=)=^%%Z%*FL=8OE6tbizbzz*`V0AI zA5bmNtj!JC(OVQQ(Kc{oJpY^;_-@+-^#t`TB;!urtEa|o_q6dysM8$RTF^h|r#AS4(cW zhnXM81oi4Lv|y`k+!V*NOXgZqVoc0Vhd(qJ!pvr;W~dWqe6ixmQj;XrqaFSO+&qC0 z1lQe@y(q3>FDTvbBdQ)vzGPf=&G2mgk85gk@^bSnfd+>n*-&x0MMsY_7qMKf-7S0< zMdySXIqzeL!i7wOrN9gC>;BeBUR3!(AOt7?PhQ&Omfjo3C6IV?u(CFODgPXby>dgVT10+=unq1e1pi7P?6tL5{L#2um?>?BlaX!D(30wr+ zr&3F3Be)lKF>J?<9o&0@_d-Tm++mfrDYuNMn)N2Rv0*?(^EH=utmI8$L`&+cR{s3O zvhnmw)f{(7GA>+}l&u4AD^+|DQ+rlk^gLr1M6t=tHO$Y}ubx>=X$%bC&hfX_4pugq%~pukOcZn=|NhFyxrw zRgdPyWZmr*@QGsmv9ByAR{n(;M|BV}SWJXRu-CTMb ziW96jJ|!{U^RlC@5Rkm>Hoy#??!t%V4MSS1^g*J$=R7StxY01%W&fJDE5daW3!SrK)(p1Q8)PS5L6(=-or^;Gr#+=g#GQJi2HRJsn@p9Kr!vCZfS-*6B68 z{rR!i`O>WOr6BKdh8?>bE3p8N{Hux7o=T}inL-wEJ}ZI!ega`wrKY7$aI2WC8Tuhn z&WS--xr7hx4_kbUf21p}IWjJ8s(LsC-aP?pR#(1&5nlmjI7IahOLwPk;H z8dpFE=TZhS>;i=Gc>e)w&4?`PP@#iLs+CpoA-`P5(f3}fYKhkogbbUXHl$5R``cv~ z4uqS}?mV`y#D`h_Yv3_u76ONQo!E>S_!=|m?YG*#7TETC!7fT&=GnPi@+SVyb{R%c zHBs&7WeK6Au9&_|A6gLhQJ1z2LWfE0=MlUzmczYg-jgR6X5TpyBfBz3ZaQ~mPC5s2 z(tgVkY{C@F%E+k;aOHO-g2nE~W-hsS@nZ$1Io426vpw}+(BNY2m@BoTbB(Rn6>%MH zt4p4quWpsID%qP8@YHRDfJ|m{?zmc3>}v1jJyEY&lYj5&Q5>>u&+7GswFy~UeFEeu zcE9E>CNa`dB0tFq5|isaFR@&%*zTkP^~5N%skJ9q$Bs^8aVzW7l*0<%-0KW(`@$03G4a|eKI z*|qj&N|h`>Ul|)8#DQ#`vkK7C?ED*&P1{iC)ur*9-!x)#Y(i0JIrsQQ&Q6L$C(qSE zQT6g|OxtC@!sp-JXnB`D`r>!2lGC*bkGfDcBu8@ctLN41-B)WqyoT5?`SH!NQ>liT zg>ZztmgE%`6#-lHRH;XG79*>l{NQ6FKt(@f`;Y8mG#Ujr_h&n z@@B6+$|U^gU?O4LYGI#5`xH4*meh;PogDc>bD_S;4`ULFe|5CAeobw1V{Q9{%>TJS z-;t(WIyan}g9BU*YKwoiBtQ;VmXgc(<;D z#kbts>A>DIul+VGnL3aw>|H4bak?;h_a8mlNv*f9m?*VUU>|r}_z`vkotv8v+XQ+u zt{JvRFENCX))VNV#arxR@0IBB%z80)A*S!`9J^~$GM1dC@)3^EL{d?|T8^Foz-!*? zcz%t}`*tZ#r|kZ6Iq}ge7y^}*kEkbN0JjwIU(%@I#^0;TzCTVe?GphwzuhWV^*S@qt`KOJG)RZ-oPU15Ji`v&9D&oXA;X)#RhKbIm{F#gC*h& z*FXi!bFt@XiEn<$rMfOKNKo~Ppit_VM=o8mq!}U=_ e*NAmv$JbT)-}DT)HAAvi)tD)xW50L#@qYoz;9Ua% literal 0 HcmV?d00001 diff --git a/docs/articles/gsoc2020-final-report_files/figure-html/pre_allocate_benchmark-1.png b/docs/articles/gsoc2020-final-report_files/figure-html/pre_allocate_benchmark-1.png new file mode 100644 index 0000000000000000000000000000000000000000..5d3196a6fe9bcb42838608564eb7d04fbce408d7 GIT binary patch literal 50907 zcmeFZWn5KV^fh|yZV3?^6p^q1MG;#MC?XO{TA(!2AY}()96(eMQ9w#SK{^doN~J^^ zgO-v;;vI9t^S}3gydU2C<-X_lJU>y+-h1t}=9+WNF~?jt6czT(n94Vm#bV81@87M& zVok7RvBsU4JQ4q*HMT$-e@r|nzi&7G#HVjpk~998Vz&R71&g)l8vQ@!PEETmi?x)+ z-n~QF*5^-!o$Y+9#P0T_E$lPbW-tG@QrO#Q%Iu{s_fFjNyS`!ZggxyO_RZ$+(dD<| z-@EARI1`^aW@E-k35+}8wtUNF!7;nJ$1GdEbnFF}yPg5=Ej_aoiW)An#>^Fd>>yiQ zq!x6)#XfXgLuiuophs@AbC>h(K6&IPR(0TH+c6_Q zr_0Z28$0r&i`Vk%|863_C;h*BXy0_<8NDuf{uuGmA8l9vzq#?>m_!zbV%eBTuPIZe95yi#b6%5j zJu)(~&-r%MqAVpv#k+qyI{Kupte-c3zQB?tOW5O+D_05$?Z0y6O5jY}$oZmtqx10p z&HVjuTbTcI}am?RL6g>&3^1O{{C{u!-o$8?GIEW zS``J#^Hf-u85`sWu=}_tGn-R{8~|70=f_(lj7v8*m&PH7+s1C6VDa}i+p+D0o`JlPh0Y0@l%FTOIRul3!#TPh5g#hh(}u?0Z_m=d9QptR7=fleUYwwVx)O`1T2kM~?>I+19_m zqwb4T-GoV#ntohc^389@rQfZs&d0u}#hGZ)nBHBTQSkQ4w64rmPaf=Ojnd8b;BcOO zTzvk(t5>gd6PGYItS*hx&|@$%)MCOmLjZ7)78^!G~F?ET@m=s6Nl%^U))9y|NMGDV~Yh zJ1VeZ#V+yWwu{|=N>;G34#k(J%-iVDYgS?uYhI$~zt?S}Zq|I6o^p%Yvz4!3Zoz6la2An^(nFm=ssPxg=px zdq6@ze>dm1rF7TlEv;v>JXQ;fh=}xzxc%)VcqJFZoCeFrCBwbfu=Hyh$}L1;tDuW+xI0O6Du1&VCo`9CgG4xYY=I zu~;`(V7kq_VX^h;7shb{VEPR@!~I@=x^E8F#o2c?KH6#*m1xzpl5L)#Y2TXCGk?hv z)6plVAIVVQWu<2um%e6C;}*FqP||Fj(pT#pdf5MJhKi@c_U$f}PXCO%TC1vjq^ug1 z4?fwZWfE8RHj8<(%LRa+tW2!s&CZI!jwiz}VvWnxkLJ~MbafRySa*tcuSMmHW6_;& zCN23?V}9)D0_M8A09bzI2Oe!?2OWBHUm(V~Om&v*z{?CStp!OH5p}-j74B^PdwsKM z-m^WA)}FXHb?Q`+7t@(5$pvjb8fWrYHBwuYv2oeqzC*w*M>UH)qf3nVHk$|dI`!*Y zRNmxT)A?bc>8A!u_0gxQEW{@9cy;U6^n*T|@_d{I+CLV(Jh5xq8sROwu1?Q`ncH+M zKetfIq%d1N?S+pIzv~*tv-%kzM!BqKCOGHe22Q}IPx~v*x2ABbDaTHn$sc^~*N4pF zVl~cd$A8v?U6IZbJtJ?axrJ8sDyHOlYwAA*&Tq3#H*Z;oAHa$P+Eh8J0q1c79RGDU zl$?2@qOSfZ!;?XsiZRNmg=1JEKP{zQU+dZS{^G}otyOHf zLx+l^zibOzw8i?U=Ezu8#?jndfLCyYY1Vea_Jw3+Wv8u4v^Gg@-+S<}bP<2JkEds# zm$p6I@Y{=5qc$%r3!p1BZi!RxQ&_Jcd+=Y6nzFJ7R#JnkwlL&KQKFS0yRKB_)}k${ z(PzHgFKdpCmuA*dSRB{J%OSp{e2df(XHT6u^PBsjZTjc-d<%y8k-yL|NyzY#LVQZ; zsXPzx%B{+n{Pr3G;A&4_zEq?qZGhn1>D7E1nKq=pV8H@T0HSW#`JNV7P`hPwsmcbu97Xq)2XSR@X>a?l zuAzUe-aLWK6Ip7vvBx6P+kbuZynA;kJF3LUzA+@|baaVGMCeKKCwe&#o@$t;6#4Yz zN>$b0Smpcm!njEyBT0@+mJ4(Btpyw4s|DHPCQko)P^#kexjkGOW@axx+*uO2>WM@C zpX~G1>0@MudVf8VGWFFI^UC02ykYfYSXA};=BhNEHL5S}3iSQ^*Oa$!Nlg6m*5lD7 zoz<6Skx{8WmbMajY*?}u(S_{4D9?YNXvA!0javz2c>zB&Wqi0W8F&EqNkbz7ui#0B&&Gi;!7lo-6e|@@3 zG~&~iD_5^Jl8>u&=s1Y1gUx~UNx(nJbbmip`s&oR|6X7bZ3z!V5@04h1=)eNcT=ZK zdG&I6YjKi|X|At~#$bQfhB^uYUu1`qKipmt6VHv~NqU|Tg7<~a`* zJ0H_LTr+#|;s~iq+g%iqE=`<~(}xX_x8ly$XKeSCl*PTB%53D|czXDJOsC#wAAJSc z{(67&P5g)keed7DKZ_vEhd4iW+&DotqOAJM;}@>ZSonO_w%+af`uf~zQ)mjdXY9oM z1-^9XGhZZaxu0V>*i{_nJUj%$*}R#5=8PrVG&^c@3us`OA3yG+fE)uwp&YS1y?tbcA4Bdux^xEaIEU_PDrfCuy<6avS$;q7WejWcT)F05S z|6kQ`L|;dpVO#YZy{3t5xq}Cb-rtxfVcT|`LW-P%!u^7V*DuwX8%4NUIc-hoqsGDze+mLC_T4VLIAI#XRph&Cv*e|Fwji{5!hQmgp&}^n zS+{N-q0fQALD2|)x~3n3x)aBr~YdxfHAy2 z{$^~rW`Yus8Yci~6L|MP_I*+JCr_R%U$LU-)v0}Z_U#K(cr4AWHjD1Oy@aCdv`@LY za5T|54f~cjdgLW>U9K=utNZ-8D8b^4uCA_rQK$+bDi0r@7f8$5b@r=*0s@2t1@~Zg zD`7-59z+k|It6?++IEC5h_@%7GzP#_ZSUTz0wET4pKsp085O_WLjdWyP5HTNE%Co* z0wNj!!lAl!p4MXg zj(iT>;zE=79r$;taaCXmT)UYYyX4c1%*_D;N20+AfU$*HHGSh&yPOTS0iJ&q_6+;b zw!Q$uN*Ld2%!9X->9>0;_T^Y1gW*!Mgg=N9PsIq46 zmKikjEJ1qC$@LpHNZ7P!TG;>obdOv0Qj}7(-odV>;zQWL?gEQ8oA1W0B~6hFNEM3; z34Oi2Xv^zUA7|$JY(Dbz=~J?j1-u1m3{|6em{7zzP#eJE_XRROdS5st~0rr<1o1F)fv8>HmU{R5ia{zgX9FQmnXc#W{#e)aD8Smc* zTQ`?-w=e+KFBBtY&BCn6?ccvPXhDjG#CIb}34_A@ZhR8?Plh^IC1{E@S0t%rW@eVY zJaNgX_s1lz`SbTn7k_@kTk#Kh38i<5HTN~3topL7*&y%P74_F=pI%#Rzkl=Q%`|E_*%<3)by!}OrcG;QN@w;0 z_D$iykEIC~LHEUVA`sk~zP~zEYyaU>CHX9DU&S=1LBgoCi^C?-9hLPk-nKM&_1wIq zl=D5`W=NGEKes3N`XB40(pcj#LM0(VzWH}#hk^+r$=?^r%XD8Q$VR@4fb$g?uF!AC zxC*C#&5Z>?cMx1fBAB5nmVAd#FKb-RvVGFoX&ELjbOcE9+pbwM0m;c*F#vMy#X8O@ zMY?ZX<2A(s0|L$hOsc;+c?E&ORip<=YNU3GW!+<&P&n;x^X;xIU%nh+-PnmWwskfD z7IKEi+Z;=YbHi^0*a$NHgFRKE5fi=q;W9D*$XIR{1^RjL&%73@61I|!f07W7nx+}f zCt(2W^u+my&<3pe-8*;g1Z3HWX_w#@AwVpYm8(~6;!f7Q1uJ>HFvZ);YZbe(IC9Ak zF)?rW6E88XLL}_Er%%_z&lZ&@SRmCCW@lw*kM4G^1q(uwT2nO;68sz)lu#WClMen1 zcNI8c8uxx_o_v|VpH^_IjW<;bSC=5Lnh1wV05#*|N5hgRT@Z|S1&k{aRW#yER1hkx z0D9`bX_$IOikoKA(z#@xiA!X$UX~)Ots&@e?%X+p!r*%uTu-B3mKYfnM?9{y{WU{W zJNf8}y|)x?ZQ~OLkQx$PzdCD41V-Ug{EabM7kOYN#lXOzAj9^b8Jy2PUlrcmjVa3mfS*kp$tFPG762A+v?RETeW$T&O zc8!}fvk=4ULc&F!r#7k}=#U|ThkgCiYa0!}iCkpJ-E={ugTeU(F>mB0#dpK@PM$i| zg{W?TycJ(}PY^-kjB5#8OtEJM&4W1NC%0j5>%>Wu!Vud)wuXMviw-&U@it{s@CMJZ z?JbV!={p}M>ZOLUn{tBWvHKD`deY<0Q@K?c2=Q_*bA{%=?q&%9w1d7|j~SQ6 zO4J1f2I{}PJcUr#w-?8DePsx3mD%)V?Po`V6dp-ig|1Za2?nk6pu>5vOU(@c{*Fxd z1DeRVGUF4T^qww;HxOM|;wPg^=Fgkw|7F|11HV4rQA32y)lD1t>#BXeS%PpSoYAs+ za*i&-gh*N90I-G8XAb2)l8_(nFHR#EZgI@OAXMA2OGT!u@JMO0-G&`ZcvLtAW0J4Ti zW0=XL$_Gnm8jVGHr<0RYf5DTX03t7t5(%$=)&sgEFaF%G&N|uQV^t0vZaxY@S3H?O z88MN0(t|mjD+L7VL5b|U=lY5w79|Xe?dzA3vpRO{7`H0(jQF|CGsYk$@~u7YGK9@? zwmkkE_R7PAKM1RjB#j>}c4*t-h|Ew-=u0&E0tCJTN4ez#0V&W2E(wA2KV4I7SNf%znj9%wuD>4s~>DF7RD zt1`xU_iP@XdJ}Ow51u&k&810(2>y>`&d0(eO~tCf1c5$#_VVT0u7-krKo%+>^F&|p z(pW>q;6qr7%V9)>Q;>U60!{QB0BaMI)x!a-0TjXG35tNM@kjQ#2SAEjZ6@79g>pDn ze89vRd`y<*e`g!~zUpd_rPz(@y%1hCl5ASP{myhJIM~%vK@R#Z0Ob`Ga?;ytm=~AJ z{b3etT9w)en1#3|C3^Jbc2{muB@h4tnGKf0t5>t}nL1;J3W76jkqU>7+^91B!C5p2 zT{qf2n)iU!46$cu_ez>pzS!f+t)Fp|OZt5G3(!k?@&^tm0~jK36{eiGA%Fuzy*#o< z_97`^rK`YBZr;AFYH9ge+o63AAsRATj=Ot}d4d4Dp`n2?L3pa5>(?j4w;rVU4$L$8 zJh`E2VT2o&3zdHY01_!9k+y2|xG$>l94Yr%fXP~Ex&=@~?UyH!SP?5^xiBz;s#F=7 zkz1^%z3O5<2e}l&lN$2-$o%~L*G47lMRig&L90m>Z+7f|O^Lc=N3PV=>C=@#ED+IL zUw;QAa`i@WxZh6%Di*6)87P~QmKn(DgA^X;6Ns|!GB7TWKTO$zfLC@P zm>B)y2(3rk9204h3G4p(e)!(Kd!VYKMo_)qN!pOC`%`AkDh4elVOG7p2E%}7Bm~dl za4bQjTciNJM0;KWMnGy4ICGJ-kg%{Ea?2Gw24pjey38z1UAA_uVqCeot4Tr)H(|$; z=zTB}?96;vGmxqv;S|Q@|I`2Ux%xXe6(~{a9wr%ydotXc8BO zNhCIX?=9Y^@b~&a0R%?jA%j85RA@>eC{khz-Wi*`q;T;SQ#M5{2^rzdi&C(c79n#Il5Ya}F8 zLPA1r-nruiH?AIQ6oQ;$&HD992rKT9;@5W35R-Y%Y=oVJ!I6Xj?KFN1_r@MMU52Uz z{2I~wV7)9VE?yiP?c54V_j!8ycEZP4&6~GwDPvl6qf1_ta?!owjmS6qDc!h!T~Slh z2Ze~=BWc2;3$55JR!oKfce`Qo{Qy{fw6iH}7M@$b$b>H69>UDN93~*-u>W4dQ!T0g zOlU>R;=W9Gh%>1O#nl#r1Z3cV)f3DE1wsH&kq#;$+baUPh6x^vSwnYIT+aA3-S5rT z5~h`!D1{V{o3(XmP(cWYy|XYhWZd82&atC&x@Mu01iW;%r96QuG?XgBYAMHM@)iav z_%Jo$P^9Ft(Tm=Fy0^j;gdY-Pq0O6*EtE9+3i`?*0%-M2tVJ)lpTYq4JlaI4eH=!X zq#IWzT0V~krou&@GJP0S%V8s<7eHHr?0vWS^T2PsHmQ&X*Br9{$7Jb17HdQcuw~LI z%T}QEa7-C8DWR1s_hJkBd(sYYF}zA1uX-#vi68>KU{2$RD1SFgR%pamN?X?7O{j#` zdp>v&jA(+|PMExxSYu^M5}?uk?-V?k8W4-+aq-;`MfkjD{{CG#8@|*7@To`Zxe=uR zP?XthEzgOHj*bD{LQd$yg$sy?^a`;1-%Z(1F%XtZRK_t-MCn{-v3iSbtH;Q+rH>=; z_boyz%yW&JFeVG>3&+FT+wf`%viu|D$HUabV~k3aB)xNg%HSp}#RX&uEX}~SHZ!Vk zVQG|ARqq3d`6CzAN8r-??tSmxQ|uN(&7oRJu@sSjL_GFERUyPLqO2$ z5cdi^N-qO1Oss>LmpsSg{{83RwluM465s&CQ6CK5cV{vD>FaoV8cHQI!G1oMuU<6( zY9h=*2|MLuH*VaZDrCZESW%2codSVJazn8D!el!O^-XlW5Eq%Di=iLVqC{L}S>@QH zKRh%DH+zrBBVS*!2F;8?-iQG^9^L0K8GR?~RI3Vk+Pbb>sP zlP(Ukd2^8=_dtRAQ~rV|2=k4OH>bH%Wnx6IA4*{nlNQ@<@2PT5qj)|%P$x@V7q~+K zeQbPk?Qh?{EsWHW&B(|Ip)IRML1zrBt(pf{+wLfbQaT3>&E<1FXiOG9nfWa}vr+;HYLx0q%y z=n^(zn{EB|D;)o3t|HKZS;fZu6ijR8$6GAD~OnT=hH*NA{#i~@tqBv6>-7{wliHD3edI%oU4}1%o z1I`#5GUj}b2@!6<%)}1>bp*kLs6IA6m1b!VV`)pN(4$IR5v=*K%z0I?VCqTM;ebRQ zzP=_AUI;osoJ76X{(=)ZQE8IcuoC#@7y~)Bd+tnkm8V4JwQe@|D8Cn|eRa6wZ??E0 zKw1Bv=r8l<&Na{MsxXQ3wf#8-<+ELpDc$N1i`sv>if1)g8X)gQ9Ywc0ndOqTYU$GL zsO{f$bMwruQ9=ev3-tT&cA3CIFZ=Z-sRX*(m z3s44MiR@Mld@>bbl!1Zw?A|T3b*mN1puF$L#Ow!JV_ z50nIPYF+Q=9rEb2C^M3jL4Af#%EX7#`WUc6IdNlH!S)W0j=4igR|f_h2C>9cRX{v@ zfrv@v?tW}FvEFd*xUz5(%A^esC3Pbfh8p8`6DCd+-n8lHjvYHFdpLgF@8ZRa^JmXK zUO8#}c%ijxzYdgGB=$NuJ3DV14_kLB4ptu13a=l5Vnb6{WU7*hNknbVLw_(%LR+?I zJlAy$eIr;GBo{H1l52q{;i>rW>eNRMKoJj*`5?W*c3heW;Lt@cUQ5(URQ`Q3T3;>yr z!54y|Sbk?S0F$r7Uv|gtZq(&Pq6T7$wT~c&y_w^UghqfCT<}}Vv14ecDmGP{CEiKH0s2VxuleRmoWZVF7( z2A_jBYk+12d_@5mKC7SaZ*-K(Xx=JoRt1X+#Yxr+Ywz5ZSc(mQBzB%URk~q%_w*}n z8jv60uBh+8VXeY%QS$=+&HQ2Z2?ckMFn_zcJn!5&^zce13>;_&wg@*%Ucp&B0TrP0N)oH?g@98D3%H zm8<8Dq|7e5thv_kXI z!8zytLzPv+e@%>ejU4H#0G{&z!L6X*wCE!@VAr8X6ddBww$=o=w$y>Y@{Xtn?x}4>eU|Dq zC|*Y(k8VwlEc7~a(5neG{sJJ8n>ky+3lJ28QG#x*zYQGH*u1~4`yZTHF))YDX(u_k zo5+R=pY553hs3@gXNnDl61T5f6UsFnC=On}Jf5Thfr&?t9s8y?Am+~DY&`e#{qk(` z+b|Kts?}m*5uZ|RZz~CJI~me@;}F28(@@`;J$v?OwFSHnK82LT9*q0Vt#x>2B-k%) zl6g#=&a3$I*Wln_5gtC+`|A$PrlO+ajixay#l;Q|4pah=Ktca~&EDhZ8w*!cHG@Cz zJaVnan=SUDT9K8LL*Uu&2_nazKYy0e{q|fU*ln7_%E)^ zu`+kE8Y)n*jR z{(m>4G!P{$Zx+{zyy>^U>KD78wc>2$A)^WlNJg@ldGFbfH*J z)x<qh66`?rnTzQHIt_{V)PnZ`H zLs0f(wh5Uo3VN%N7!8DKyM~1vaOhh77#0U)6Sq)AnMN62$%dZBoh zb1*^<_4>@0#Z)vys+{&a7P+`8aQDSam$>-(9ZFK_A1efzTUb!WiVPYUr3sY*)c3^J z+Y78(weJi9nCa;o>$C=bf2c9{EU5H7^eN@2_UJkxlP})nA9NfpTeKI?V<~R7 zO%>-}5@O(R*Zbi^eT65;7CzOh{?d^C@C6>L&T-2}UePAIR=XT`Z||nAkITUKs(Kv{ zkz3IAAtR#!Md=Mei)NyX05%u}RVVk_s! zPk?=_8wx=f36}1XNJ#Eiq*XG*108#u)CkzZx=Fo9%gDDu@4&r}ygbecg**BnnGs+I zNqYdl zUvIav6zzDI0b~?q{n-p>=%Y|)cfqWE-ilw99ltX>^n8rCOzjktGcKI_tLO7)fducZ zyyU&rXU4nu{_s0vTVKNWwnOi=+vLfU*Oo2gmHN^>*Ed_{No0W8U%AR7Ztll^S+yCh zK9Q{F8+oR^xN5_D@ogMHfU;!ILb0FbyjCd9X%=-LcRIG9c+cc1Q`V(P92xN5@ZnmZ zZ`{x3h^=?jw#v?zxNWgjw$lN6*Vm^r=dzVha7PL!;>7j&41bvC^Vl%`&8n-%Rmu*v z->J(*oB`QXNPh8OOH$-}KT%{^GAnsiThTW? zUP+^)H`2lu%h=aHwyDgcCj%a>Y5b>GsP^b^A>xSXjSkR6q1dO52;Dr>ZGo;u8_Ls~ zUEv#UfoW6w83p>nzHAIjSjIb7&-b*4uGuY%TkacWWz)>du9h`RzN`N8#|XlR27n~# zb3ZSG;k{Xoy|EL@*&1T?E8l|+2M!gij7Vyc#u7i+VFI8QI4<}WEB=}AL1VR!ewA{W zw5dFD8E$<%Q@`B0)cx~a{*fa`1lda*JMW>G?e1#w4?wUu;Q1PmuDD1Bg7o zoUeogS-VTpsjr5It%UqWEM5Vxv=Lz$ULT<6>It9@>~2AAk@p>8cRr{s zm+wzJ-R|mP*PWm8{YutWjU~K4&bH?z`L3{)DH}UKtzLueZ!J@I{=@D<73uc zEbeFu5bPS;$)DTQ7o8WjdDW`A(0!e#VsZVht7~fe@eXwgvm4nm`x^PAPK}Q|(|Udd zYv#8B6vrR<R7w)*$VAr(sS5l8mm~=RH)5i9be>?ts*d9WXPi3`^>CL*Kivc zoN&+3;bfUz?UFNmQ%<^__kX;b&2znDdbG?glP}!ddzWe-es|P=m{uC_4nEwqvsBb+ zO@l}8Xw``BlwyIvO1wEXb2Y6aTjN-ozaPI3E zr5K``@_@%ga;R#YE|mm<{sz2-_hk%6X#O$1Sy%P?=N-`oQ;qhtD<8~%$-D@7;&%Sm zyJKH`J2*LgAFjpk?f%*4boRCW-#;0K$+G?b{$g_9I+|B z<~h|i(xd%nYJX!GrRf<(nh&8Es|WP=eD~ukX-tvvaV^-^YMx+;5a^j1d&{NscmU&&b z++8{76H>W8&H`m76-7lq5XjzfP`VOAg5aPwm)BGO*4Iw{OuN7(@jo3tOnEM4v59|= zJ&SvZigNP4;}Cfu(9TCY4^Xy+G;-pZ*M23B;RDgLQC7&s2JJ(&eHS7l#d%)sdN>7% zos-kH;pd8aps-K)dItps>2!72IW0um+M3)n-PUQ>^Ij?4Te16mdvxt`tO7ssqngPj z?Jv|?I@wnK$8&EJj%9>wc-+L~y{5&at_Rs(Cn}yj?y&!rH>d{`A)}pJT$1XTB+vj6 zugLmg)C(Y3KvM&T;P$XMe6qt|wA$x+trVMYsA zoRSgO(>uUpy6i_n>bjkuY{gD((ak-(=CH&37j=dP$t@q$q-8E0UGYF-tmc=7o8?}! zK$@bp>tSoSE|jcar(cja@IQU|gKy7=;FDisRee9M{IKrYIR}e%C?hI@YREpVlgzbm zbKqA})aE0z3Pa&6PE*`kRr}VD9v zf#|+Eb5SVqipR9nG`vrs#0|tYVE5#M39>{EDm!HzV)K9Y6-Bzvh%-eI6cXuu5Y|7- z>8YT!O^}U_4WT$=1U{s9h)yb$b%>`8JcFBs#A;VR@(JgW4gt$h&+eqU+QYT|53P(oDYg zvfWA^)TS0NJEYF`8J8-=GB(w=C97VT(_e~KhDA&3L}5kCV5FoFM2RmLh59DIOhpu% zpgj|KyoDtTKSc=*T1@~S3V+`Ga^_Z{&wAaP#4y8AP){9jAvEt_)pQ(H^vb|i6dA`= zPyw?Z<*RSN$4shUr@kk7sJ{vQ4a%05opwB4pu)bwFH_gdV2!wQ9;DL21bBBfgPRP8 zMl}o&#>+@t@mtIS^ge`N)pi zb3xvS@`*~MHZqnI|2OL8>({T#dD`3fLl(@Qy?wu}8}FcwBPBy8!FWW~+T|Q}*uLoV zvTD&ll<%kz4wCggR2wnVB6_NZ2dj!-hUOchU_*sEXsvfZJ}7P4cjS;vAXRL0VteZS z?=)gM`T;%U%e9f>b-kS??$P@lOmQQ?JyiiUAjt->^Pn670016PqYkyxmZF54K^I#mySxCfuZ~7D=}1kymLnlX*t*tqGeHy?!D+A*92`Wu$m5J3Q*8_ zLlr6bxaxEgHZ^I%V@<0fkhhpGnBfl214&&l3jYpnNTdxwdtj1|AiF{lH3qpE=HL z;rQx_5BD0B{=9hV(Q}1BH{A_I-&>UQ(+*o0F0kM$n9pN2?OY~nPQd~19Kj>!*rIb3 zXV3KM`1h{C;o9Ltfx{TBjGc2d-~Uzi?}btTs?gK2>7n zJcT_={yBi;ut%~D`I5(ayI>tKpwAT^hoZZoqG;Xh!1mt?``SLOxI-Krq72PS-Z@tK zQ_Xm@9bmULUwkF?H=mB{rU4`+1|7&^Pn$- zCnt$lA4iEz6d{z7`atbA^>9^^=}?9SzoQ-+IfEs%1ZzfK00i?vl0KOvM$qb^5>LVe zK!VkqHuXL~lS9={p)K>1T`sMAcz{?@5}oDb{~|bT)8a!K8EfS+EF&mF{R0Bf zv?SPoy(p#mtDzwTwz?WjFmk8EW@c-v;8IzvN!#Bvou(WPySDOgKic*5k-Y0Z_{3W? z*z^^m=c14!j8F77O+981i3SMG!J(csHCT>(v&0zIXH;h7Np?UAYN$RIMoa;-iaI_5 zjvwS7f2%j(Dd}Slw~d|o>GXR2p@F`@p~TkY4uz1km8^6-=$}`kW)D`gA5^Z2rlv#1 zBfNYFG7GA!V6%M^)1o`PIou|_z|_VLQSL8YQVl#e@mdV8O*aDemSf~*ZmSY|;%fy-t3vfR80C^)&euMrWU zh)L>nxHfI6CA_@BC<>A82V~Mjf${jRqN)%T*L({9tJ-KKbtgt!X=!SbECb)Za_w4n zQBjef2i-YqE_3I3ua1gPiw@MUP$w)8D^fq13iD`0!c6=CO3HWvt@T2y>6-I?^qyv8 z!6$%aQbg4bwqQ}=06LfPZD5r_i?~A`F=jOtEx#{ho^dHkJsd9m38cAXiw|>a=XRiG zK-D=+6DSoUMLD|t-e=71*OLAM;t=wGaB*5vRqzzh6$wMF;_K@>^0L)J%**npM80V!k-tA(PC?po(E3*^sA$CIY~)LVK53RKo`C^w&CG{m6tfAzB99& z55)${`gn}G6`VWK;ZzBSQ;pD=hsp%9?L(j|j>nM>&P8Jm`Ag0_*3sSZssJ%#Qg`gEc6)XL0|$zjmy`2_`@P`99a33OPMnHIt!aJ3p$ z9#Q)AVLF8Taui8JP<}b%WF9hRPlyGehM|TZZr7TthP!s{qOvy1|0IHCb{6Z;MOwOQ z*#Yce0wVMnU*8B+>?_S4hlPbv4S+;r5JXX#*n-i+GFHl^iN@g6p?)eK$|Db_=xtUI z15%0#3-ci%LSY+yEFW*P$yBTDX}bJ&uJg|AM^Tlki=>H0t}iSmmij}Thlhu{P?4fj z?FsW&17}?Ko>s=pg1J;P`lyIBbVq2teT}P$EzkP$C9Rz&ZwOYOQ1{lY4t!B`zAxZJ3g50&GV=M*Ddl6H~N_*6ox zL1Uzs11U`!D3io*+1c_`TZJJN0=9$TF#AL6#0p|uj*kLpGU|hX3=Fj*%J-$Tnp#>S zKDaF`n_2-tmypoDf?BCu{$5Ce;;OpMNyd7asOUlJO#^TZ1ZT4_LgL{8LVWPj)NMy% zAZ$u&1*T=Pdk2k_^&`Q+S)}|mux_z)dx70=AuHV#;hCYDe0z+<>|h~ zmcwvBlW~B$914`AHf~fVjD-u%o)tPX@?~S@%Yo#57|nrkX#kXQ6=nyKeUYY<8d?Au zqAYge)+`}e+H2`rNtO+$fp8#gO7z|)6b@X{m>=kh`1}H(zHIX{`mV=0aRI$mswUk!US2hR%)T@hRpr$h^z2(@F&L`R|Az6v9GA!bb*`KkwU`HblO0| zL%V*=6qCY9UHY z3l;(sqW7s0&+f9D>}@CS}B$T=g@5Pt5Pq69x ztBWznLGV|AwUp0M2~}p3fEUs$QqL(e^PQeczDjWi(%c=$d_oYrtWYtVYk_c!s_B!H zCr?tX8SU|Czo_yHpm}%kCLG{*F*7Hpkm{w-K2WT)z~(QyG;3Qhnz0h%<9R)pZbh>& zTF~I7SefT^QUD~DQAb5$n^wBBfzR{l`@lVX4@SP6L|+!Fo4I7$3u=IY8x}oRk(+xE z^edHGsjo&+n#Pb`L965ELprre=V3>9@Ravf5bTponS@`U!PfntZ?dqRc_50dg&zPq z$+OCch>MFOk=YF1wc9Z%>sIp9*ZhW8{CD}w?g%Y$*On;xQ9tqBUeLS&1|nwmyQEdZQIAPv#(j!M;rwT$Vc zYts~&%|$t>@MnfARW^YDN+aL{)_~romT#1HXB=P-q#ysG$XA{Eefx79y+M6bwm>s8*sf0Feq}OLL;Dt1EO+5~n}ShTL<($VwgG z$lUn=YWCX(JCD%!y~K55wWJ;%ATcO)U4jDQ9r^MIh%uWDzypO#306RWc0D8kYFOCm za!g;7*={^tc%yr!0Gq}LfK`JzH(Rc}8!)UN#eHb~qUP>fw_gcKBE)i_F$_jl#AOBc zwc=CSjY!2`ll-`H$pV*?NG$hXOi6d2RxnDsW( zqd82bPEKgUbsXBS81uA(SJ1T-*|ZMM29%O8EBD7NgfsqW7jsh+rR5|42h_l&l zw;c|Ib9MTHXOKa7e9zE`F;MY8^(Ip@*_Oe1kb!~tyVq1pg)&2E!{}AYD=KE}X9gi; zirLt%gq1A}*N~!&44AFx;}Yi=D35)pKWa7>^ganNAh`*6{rdIXSqI$(q@-{XLLr-k zf=KFw-7`-sj`aN4k?R|lw5(U8qAy$r^_$W1L-&EVBB>OW6(A4=c!fwI-UNO1R+c!C zL3riLs~aAZD`)8vG9D6zZUvQzV&qiHbZ&|4&{32p)=P8qzlVb)Scmr&LV*IwY;kx9 z&?%|FfYDxQFw1guI4z7%>2 z%$hYT#A?MwTMqF90`!d_%N(3bvWbvzWk? zAL?)%bTiV-;J^`otX;`w24Jy@VZiCm3)Y{$PR$xXY~fg4+FlUfx!t~^q!**!1r8T% zP-;E_9FQ2u?EgJ)f_#nS-K3L2gNEQ1D`vW8UZClc$HkPWL6ntcB-w=4jC_c#j$4Y9 z3&NC0-9CDt4nZJ7s+K7TjU*TcxFNkus8*8XnMTGJ^3Q}XEKjHQ4j{KM#4$=Xug_UQ z@+$vRKO{-U51kpMXb}khxTSZ5m5q%Wfds%W00cRVIy#9q6u(V=f>xx))YFpQBo6UpNIdORCAN4uF@q^~wZ<^c=OKCHOM|)CmvGR4}+o+P#a<|5;ZQsZ&*dl`a1e;7bIa;w`Porz~Gg4T_IH z3W(AF+x?IwTF#ql4RJOSKo}%aa*!Z)&tkT?%MNVuYN_db8~1~DK!;udO?9O6Alb|Y zY1J#oCB2ER19~U=Q?Pccj0~{}lo--7W}aRzBuKBv`T!{5QdQm5M3OFjXqh#*BN7zFg#GYp{or zEl^cH?^RYHHW7v;4|FGbF6G3#tM=fLxaQ1xlfWdLtO3g2%g17Ok<(`mY@x=rCC&ru zK92QxMtm^1;(RcW@|v@zO)CJ7C_+bgY9;dm13Z|}_&oqtK<{CeLj!hb*RmnhZQogc z!crvY4Bi30oD>M~d$CP$bEH8+3dnv1P>1kyuuUq0W#$SGDuKFWnJwB0kc>&Lz1Oh? z<=w^5D?pK1pxR<@XQ$s;pGT(}ApY!57`b|wymj1DlAlkRa!=s#Ry!r^EeXlH?r@eA zH>vFg$!C>y3RElo=rP!cwwZ*aq=h5FnN`Q!;7vxx9wg1L-FP4{hEp9kZ}Sr!r@lS& zRv!3@CBz95LhPwv>8Zp<=at}XH*6Z2|Ie3b{tL(_!OHh1s;W96TuxkNzL*XnKz;oM zTvllW;c;rpLioOr$;_RZ0__v;11sI>*rA2u$D(%24i7|MxWWGfMqP!oW86So0_k`N zr~ryNqiwHipyNw+adC0=W#$f>R?+N_1GfrlnIoW;Vy}V)T)XMsFEUDi`$)1TMBn_F z$SRQB6p>E7V?0RpZ?fT;Zv30`yO+Ksegq4PL2o_t*(Z;L!l1xQBd8JK(PoS16k`vJ zoRiW=^JOE11C!7k;APo7`67yySq3qDK3V68Rs}{V01R8oHn*`U#?dkxod#?*)-mmF ztQN)(UO}53x^r|mB)=g|Z1wKju*(@lNTO8&hjN6f6kyUJJvTO6&%9_BV`kGTD=Ycd zoxE&30QA;~Oz6Xe^$##rr0QRNFuk}qRP>d(xc7hlo)-(fhoc{D?sN?K;e&PF2u$H!R=266|gyw2UXfPi6R_jRi9=? zt1JQ=R=kN9S$W{WQlK(+K8}SlI#Y0;r^@MHICZ?zHsUCwlucd;PJH90k@JDLX2J1z zL_x=kf6S3|f(R50sp0Ww`O%;};3q#@ckafCG6*h#_|m1(nO#RSyMTA&*}EwF2sD&l z`b4KmU7A;sBA(j26Z=&W3Qd6!@X#a-r04#3=2bjI(a8`^-|#k^01yj+ zUn-FBN{(JgE3%$c>DXZk97`mnDQ$130i;h1N|ZqC!FClom%JO~YbUs`Y%s#TIzuP7STk^g#d`-X zDeEZqQI}WRz#b%<`7tYIL0W&@EYy1qjK`p4QXpXx{zNW!CUAvat>Gdbh`Qmf;H9h zuz}f1l(_YNlx-a7>zKI8orL40H3iW$asZpRVU31Q_PPX?;CklQ196%U+w2mbp^tuU?2qR+>u+f&|C2O0RsZaKZ4F&P-~4H zj>ZW@6HQsoU>bT)Db50_#Ar1^i-@w5Fp5G+BdgA`p_QAC6yq6+@fyf}H~|!aaUQ}~ z3?ciE44`l*9&d<9NUga7nH3dU&S)I(M5xQIs7T4F1f_?M2hmz<75DGIj{T#%#l7cm zO*WmRfaY28=n~D)RvZ8kO8w3N9kDAGz?lZ&O`w{4u3W`LR@Qe~P2LyuCg_I-1uI6> zML+lfiqjBW)`hnhF3H4Cg6-!_MMP^hciGcDyK0UFS^5rKVV0Bwa}EsYK}Z#q{aaHy4eKUOy( z!I|jfJm#rw5Ne_1~D28??rSR>X5&in~nczx@J6sjHL+ygO0_a>MYu6 z=0>&xfW^me>Vp0(A|k?i5Iq9;xI~>mNki3t7t3rL1?IdtSrSNcpjxJreIPfJ(wsMY z_EID{4|Da51fM6r zci$W1$fY`F)||Fdvt z>^`&S=jRNBQz-eRQX+><(}qNBlfKjBe8S`q8T4eEBNeOwmC4^lA(oC;M=DJf~I-oZ$zv$7D)?t?->Ran`7?+kSA zqL6t2k8(yEf+KA?)zRX?f z=;=r!q8XLF3Fhe0&wYXu%Fx(jWr4b!3eI~2og)MesOI!b`pOL zzew~W8gid3!ifvK*GIO9FbnRix7St_iGyBriM0$m`l(?T5)&s-Q`x82YxN``WT`@% zhV(56K9Nn1WZOX7F5{A@J3sf8GlZL2}#u~!ckVXZ?>s7O6 z-5Ktq!+8lw;}OJYL$`7kMHWr~TyWEs`nqe@pkBhis-3%b0cn})qrDV-Nvr^s8<71H{FnbT6W~;r0wO+HAxO{aAD;2e zL%JcrMkYX!1ZstvB&Y4_04U%`*Mt53Xe|mdjE<}o(@AwugF1p`vS`GQc@d&!=~$NVwBf;46sW~wdfK|XzkweLrPCV4#EMPCsk@y{V+Ax5j_HA$k=h5Jicd`r z@Ol^8{-`5|L!Bg~%)nVAP9bSt07biSViW;{vNDbQ7z?{6=UX-dPkZ+N8^dzpjG7NW z$(`wx46Ow7Jtr}OhSS2*2v=dXi|~P=sS|abLaAaEs4E8`y89c02zY$ zxdiL7hX_!B(hDC=!DXWc?(?0FZaVXszpNK0(JE8sd}HVk5;X~Mz-6`&YM}XR!`eAm z6Urs<*%X4NEIL`#I@K8c9Vm^RF$-PvB&LVO8aWq$gQHdg0DH5U-OI80PTBevy65KL9-)Q`N1miL?6ULR3Mz7on#)W!u0}Sh0 zNMMiL4G@1FC^yOozr#_IhYU0X%V^Hsu zy!>uxSk1;O9)SeoO$dhPG-3kPr>SoPCk(QcuvfaipSrWar~VsGc2YzN^vrkw-4qSz zm!7zCEGxZlBrdwl%l!1|D<)iky?F6rIx+iqW21pts77(np(WJm&w=EDdU?P^^7%|< zvFey&G731$(cgwPqE~Jo-4+IcTK_e+;wkl{qkwQ%0Clqq9f-m4(?&-oK&Kf~HH||@ zIpFk^_tm>?J`@xj0qjx(G#6k~Apj}jbcG2l7mLwIy~+${`7nqO`A%oPy!hs#yy?TD zIH?dmD%OohRGmD`LLRVh|An!cpHIG`lP8$G3kyp6I!yR^>bt#h-t+NeQ8q~t=(q+vDiw~{5GQZP zw!_2GA()y=QRkmLYQ4drl|5}=mZHmU4Qc>pp;7Dfa(GEi$LR#~bQU@sAkv9JSl9N^ zb-hHC7V2dpY(AvK)WdG}Eb3Qn?Q_c8P_pL)KuoKIbVFhjjvKi>>P`8-;>-p>Y_W#Y z%`r7KZw!XzQLyO5`4!(kpmyL(=?-lS{w1R$<3*^!v{gwihSrjbCC8WlTY%yV4k-|1 zgHRxn2qBBHQq56YI8H@-5_ocm=B7s6iL0WPV6MT{e4pBkP(>$UF>gbuIp)k|90@bf zuV?%VXUHnSQYWsAO2gqPDZM{kW3k^|iO1)A06X|NYM(PtgE@+|IO7f9hjPLKpHIBh zDg?z560ejhuobX`OGn1As%=MnZ~6{&K^Q`G7Tb=}kRXrO>j5l=?%6#0dcH`Re^= zJuKhBjKLAkm_G*#Q^DEvOUb?a8WiaK!_*?>IH+qDA%5XWzEBL%5!RBvIN|KGuZ$tH`r}4cpSPYQdi(d^6+mJ8Y|evq z;0ZNn$#mX@AVGu;`AacUausFe)8r9vjV{bYs#HOpjOL;9J*vn50w*e{RW?NiQejjo$XJd$kF3gZ7XY*dq@SDs2p)h5zu+MmNULb0LI5YA4@0AB z0pl$ON3HlV@e4R5BzE1I*2;%KZ%v+R+OOPzBtmgKaaR0xys*p23;z+pSzp*coB(u9 zp_Q=^_g{lz1B^P_?F^1#25G&3?ET-UtK34SI|k3MFd|E{+RJE z86E#VbWAG4)C5-J2qU21^_%W>SXfvTV8I}l7h)6Cq?Wv6;i^QFDRz%et3r-mGz`-C z7g}}s9z1uz?s!DhAEFhTL;D&hVeukPxI~SjBI_tL6H_xihx9+8$Th(p>QFfhh^kU@ z=pPR9MRrbm_y4-)IEn+P;FsCLBYwED`XeRIpeT=WPyv>3?6*X7MIo7o$D$FR`tlOA zM$~`8Kqj0^!0Qt{K<5qMHH6rpe7;i1#$!G+SIro?O6U}bwrnr)IJbSbc$dMi9)I>X zx2Q;kLK5(=z-k=mLFe=lR|-D#e*n|^SesVuLGT@%fdAFrnMdWE|9}6o<Z3bE(nvHJ->%!m|07cz zSi*sVT&jzKJXF#xd?Jy7w#a;Yap13R*Nvp{_BGFO!6%mDt)M?pX4=W=S)H#5Z(A%v z&;tlfrA_=(7X*081yW1>+u5mtsK6G8N-GtzmMnQa7T_`1>%f7fMz1acOTe-nZ1CNt z$H&IE7T{Rz%18m05{Zdc#M!H~>o%%8s3~kP6k`^|sEZdzkNsTq*h|Qj!r){^}kXzY+gkSXa70=>94Io6|Gqt{;bl?AgPI z*{WljNLuKj2vzH*8|dk!CJg=9KmK!OiTs(UzDKTLRUcChNQeffI3d*%eu8x%*im}$ z(WU}>a_pL~67fPOq8;=18%$V;4e4mK_7u#CmK0eM?CZZ~!Nu9_n5&?bc;f~oL(D)_ z633PnqeqCi_LpGY98Tt3_b<-U;jlJ`bg7O63#0fp$IKbXMk&H&7%qa3musa4CA=0` zeXE{wO^;L2u!550pYn}jwSi_aCy3knyzZ7)cStBh2DU7=fJY_T)*}9hOIAg^%{xp- zhe#(ujZKwWRCQO)$!`LWRP=ncLmYCGgh6$6`|9BWVk0{$QE}EdCg!AHiqmffLS->F!OEe!dMy*II6ofpEl~jU6bQnBQ z^|qFrck%8z>s~yD(%klKVN&2o7<#4|EPm;~7pfjL&|Hv+Lq@LePfCRYZ@3(4yBzH` zKQ~acYgp&7GPXi_CVf28eKks!FpF0vSV_P6=0*5{*tj?;jS2@!Bs7HIcktkHqdk>Cxj-rq^IoJjx6$!KITxUY(Pj}C|y!z)Kox*w3sCMt}FyvQ=Dk$ux9TWZ! zp$z`dSDOT0XV9RjRNp0Lr_}rxMky4ix@qrb3InZu6;I#I*pNf@@~pRr1Dl~PTw$>~x%4HlSqJj6v;K=~lycpAP8 z$uUuE;zdxY>ge^oOW$tYM#E!?amd|qf2x@&19on4SHB0Fy;+CNMr~;N`LxM5FZ9-` zL*0bVGl`Ghz7o-QhxYNeK?{VJ4pa*y-Gr_S5p0uvSe2r=$^ES^nMvCo*b5uIq_2xY z@zMTGc9YfPKZbWVTXSa1Q^T`$|E!0u|10NI^?11t)?N~CczJM`lwb>!q=g^o*O}8s zp93A5D11TDMB+}%>)WU;onv)Yy~%c-`{DW3B8Tw-YWT3-Ga`G6M?Rq zT4b9%Zm4_B2I=Pqus)Z9V11zkX$0vJf}N0TO@}1B85qLzkIu1*;;-U|HP&sl_T7ly znt!%ofVmx$SULx1XG=~DKf64)I{XB<0uOg8nxk~Q(Ki7tQ^@Q>a8xrf{&k4vmkhYHqDda4TLJR3b|Hc#!5QQHz;vdAH_;{Fv6a>qr&!7^{ zx2TT_*OdRCbZ-yOZc&a6MXMU7;2xfd2G!MeW{(?7_T_7EXAVdom?w`6C?> zC}+{s)|jNDe*PDP>tKQl8xf7Gs2FB2OcB`c-3nuIQkR8CG|~ z*n|pbIZ`N+hKXs%Cru^vUggNf3l~V_Bk%;MZ&t(%*+5s&_MRQ(k+O%sfdEDkXrEK~ z*rNN#wq4!saj5+L`$x!92EzT7F&hCJKlsrrK?0(!d>-!}k66=bm{F`)C~<@^Wq8SK zQ_b>ky3Wb8{k)DgLE{U}O|LE*9D+QeTKo&HjTjtCFy~VOs3djW@NTovB85bO=AHh) z{(Owzsp(;l3%*tO-$gQNrBh?Y+VAPo(J+JybS||7`Qa!tAI+BvPWrb**@e%q#{xmO zFx_G!QYj+)ryT;%0N$2#a$F7ZFj@#>2n38LkvU6Goy&2QqVXzXrJmiRf2k|_NEwL_ zb?AYy;oPV$wRh2{MhIPobh5a1NlEJUqy5?s_snY4V&LCD9rf3KnslfC!wJ*&HZqTQmr%5A>oJMTF( zc+ucWK+)?RmFsO^RQ@p5KYhuO%E&J6uS$M**-t}mh2m^&WDq2<~8PW=9XJ^q&k$q>K4Zt1D&ucV^2s>CEnX`K_ zS`EaVsoMv!gGPCHF_~8;q;~2Bwq4B>t9#T%FRR;7koMfDyngMP#i^NtwzO$bbsc6A z-sAz5H#z()K*AtjxQ!fM@qto!054oT&s7_7I2V3A4S{Fb!$Ouf_J}J=E5MyWzi|SSpQHj>#UfU zH@n0^JOxtlZDHD6u~h>xDK7PO@8`JQ4vY8Otne6b!(t}^4$7QKGK*W zazteaAAAV{)l6D6SgDe2VTDQ5roqSDPw&0_%_z;^beqw*S+f;HwQ@E^M;sX)S{zxp zIQoRqhCh!a(&j!50$119C8J+6_u>J8S_LG_5V#+nr6S`d9Ci9W!YKO)!MnKz?^Ag7 z7Ot8EUt*A;guMDO3!1M$7aNgRLh(``qIug5_0Mznz@(pqj>(A;uoo4%>;rf_Kh&R5 z!5SMJ9Ngi+_loA4?NoT>XJlkZ*RBi(A(JS3_YR_tt*x905U+b8t0qhfbA`^#x&y5k zeWkRl%$$%yTs#N~6(OZcZ*J^Kuc0bHG+oNOGyGDq%h6J>TI$pBRfbjuajn#vLB&q@t)o}DAvEyDh8WJ zRpUOcM7V+2GRGnMigD_;OFm!cTUk*N5e2=&qq>|bxR|yQ+wUXVyV7+{bB$M~-Ef2c zik$kBWEPmA~Ns1qO41u^;770A!FvF6D4e}+ffr{K{3D(h$U*2ySuDGcIw%>#7oiE6^|*@54?xibPB=C$a+=U%xWZ# zSxT5iZs{-NWny8-JU{ZaHC^hzcJE6Nbof;?z^~zw?8hxZk?2%TBbb^&-ri=zR7pS8 z$?G=?cMYtrXsd*f6)1p>zz<^jq8)NOIikHKZLpM7)oj#fXScl7QW*4#T*^N1Gw!*} zIKZZoNBt}X3cT3`P+gQTQ3#Q6Nl-j05M-mrMP!SrM!_>hx9#mTObC@WsQQT`t`meD zP0tcSXzf9$4f0xdF+M3u&CtY zj7~tck)~#mU!&@<5cf1rQ$|%J^q<-92YFO!%%cEcNurjRLGez&c@ct^M(pVBT8dxJ zG^-1R%YtY|!>s%{5R&cO&>hwme3D z7vk6AzOFL}(WyPCQy*wbc(uTR^-kp0I07`exMU%qO2c8?Pl&lPUdF&mp-W&{2sVQa z`gmN^K*EY{t=smUerM`?Xm&e2N4l=kKP0-KB)A|)Q`^QVioe#}`_WzG>d>Ags+`gs zi{-1EJjQAQu|49L{(S$N?1=hB$=M>c6bkNA!{;gk@&*-+D`6m-Oobr4I5&QPhtZfi z$V%b0M`OFx>glM-Pk)c|m?==H%;O=S+On^q&Q}^|qiFFhggPWTmTN5C%1j3O2}~v_ zn`pq1MC)D7v|Bw}ky=mkNfcHG(FkSD+{2)N32(gZUgHYM8Cj=?{)ji8> zlt&x*QjfIkWB(Gm7RHNY#4}p@dRRwr-zOCBoFN)E+ShRPPxU)y`ZuD3m_^hu{AIYN z`i$#Y$4IefDk@d3AU_w;Qvl}GRyA{N#Zif1k%)cpXEZVC7P_Wg=8@#Tsw^T` z&Jv0x_|QSD+bOZ89lDP05|}-1Zz&aYsRSKFuLczyJGa!Y&}f!(j3&Twnz?eGuW!cU zz;FIi;$90%o^-KXYI2G)UplaX@=n77={s_yHI<_$kO?JZhrMv4^pXTcTtAUJ*7nhv zIz|*YCEWKJU^5r1Fk}1oy>kA!w(6kA?`-?Hn^e}leSDk~9e;gBULc@1?lI@}dIqS3 ziz$-cC?cs+tC7Ry>l1#YOIWQ>DJUE9w=isk&VomIO>mah#qHDkWdcRv7M*tilBATm=%Z&no1irHv7>L_ z8K#K#Jc4YE?=sjlT6ip@yT8KOKRZ1`C2A)qC`;Qs=in203F92jjm+oaq!nkuf`2C8 zEu`bx!$X03_bu=rIi|JHf&~0paAQZvhN`ktjhnUr95(Rd*b=Tdy271z{E`7zmm?`P zEWo!DnCIM2A_uNeq#naYiN+_TH0?q*K%MIHJJeD8f`y41G=Y`fdQXr!asg8L!U3n{ zSkXr}bVhljk8W*aO=V1sm^7L2Nl9b{7FD29n(ov!TC^7FN%syaavuojtZuhsV2a?4 z+?SI)wfp+rh=Rz(FCxSdkXWRX6!`)(#x$Lx0NmYAbZ#HY#BdqQzzUIc?I#`)*nwT^ zYhYk7Bf@bDGU9k`DoipB5TDGqO6rJfu1>c+10)T#cCU!p@NKetKzlNa?v>+*@lXQx zYs)boJ~FFo{0~GD(ibmio$wCGkf-CNEjR#k@$he?A?gJ;9m$& z9u%V^BXbBq7TPf{M<5#+FtgsxO~{pL`cvwM2+`sY!YUZ1uHv{++#&?&oBnZk)PhM6 zL_q73&JG(r*EVLFN%gxez%Bb$y#cWl!5-KjBp<;9SDEF`K}GS!(z-`5ldOLQt!5gg zgxdB14ep%0{pE9zEfzky;W?z`#*ilycs!xnm;3hZ+XK>8E>jGM1wBmO`%1rImXcdBy9jJ3LbaUFaI6=#T^(giRCrmZql2nJ47NbdOz5?RQpuQzL3t|npY$=A%o^5>z(=EMs~v^KB4ydj zm)))`ikVgmaa@K)^8&X^Pya5z2wzV+b5XgJ9zETuNB z1Wn{zNv^XTp1AAFt?i2oN6=*~W0S~K$O){h7H=bLg`Lx$PWX7!r+y?JlwHKTN|TScz_=}H&8U! zhQhq*&)jn+FD`lBU3Jyr)e09Q8w3n@UB1$)$>}zD3v16oWG6C9P8!xYo=y+jrvp3D z0CL_NFeuz9gCiqfxc1Gua*Wo~gx z`mHE9lh!{+Y*4a?SJBii;*O2nY;Rn$cDle+Sx(hLm!utmkxjSf{18iPwko3b$j82J zAKu&tz_lCEp5YU$U6s9KL!C|p2=<>+TN#+lZQK@${r1y`uj06}iwVSe#qOh`n(W!R zGr7OpC6gH6x0AVV2jb&vZ9KC=8^1@+TwnrQEg5&BE!8g(=(CyonXJA_RcOZ{Sw?Ch zUKL8`{#UU;$yf4MbdpKd?_oB)C>tar`cSG7b*7a~N_O9Mrje?o;2{0}zCr2Fy=rWO z%ew}+5zEXORe5uvomFeKomDhx)5N}k)55K9y80Gmc~ZGoKFyk4o2bRW)2x*I>BldB z2P{FJ@g;X*@zB~7>91D7&UH*q%s6lvSShkpD=|uUzJ2aZhYd4kJdNM@Cj5o&MkD*b zuB@qf|Nedah717{2wgIq^ceC6z2Z~)iUv9#32&Tu7&q7CZ_5I1ETF{nDcz*#rIlTE z<~1bx`u9)C9y%@^L$cdbp=iQ<2HiPsx#wukyn8?D-H6zEfF>jg@!|0^tWcDRIQ`4j zIe)BkbDOB<4$`i0Eqka-J|Rv{*i@FQ5;-XYWxF|243`3oF>RWCaA;p!q`5C(@jVkGO^C^3yt`&IvE-E%6 zKMp}xvH&4=Yi62(fD{*GhD13I>7$6=SrXIZZ@du1BsCrk1yVpOs^l)Q&OPvn(j!Rr zZvkJ!@}$|-?7s~&XRa9m6_5Y*GhgWNX26f2pdE(T<~+a1aTASuA@E9rvow72P|BDq zmVZmbAYpP>)VS@r4LoK6={WHGO@T&JbW-&s=|Vz@FyzQZW^(ZTfb_?~6p`jM}+nv~#%WIBrd%i3$0J8dNiz;8#a z)!grS(lAN=cp?nM$JiHG8mkkBm%gVm2j<+4_8E8g+LaK?tB@prT3Q<)qcjwZYR=!F zqtYt3{6%cExCGQy3b9zAo@V$23m^yKXb`UJk9a*9lCCsWajY!J_FDmZh_5OAyN#d) zBK?uG{A$tb*P(=9%bLU10f#&pQ7ranwV?-p+G6 zzu19e24t&9Xw>>7P`dQSQ&kM&Xors7{yo|6ofA2V;{C*L>APDs$gY*(S@ftO>v;Db zem$KqN1_+Nj;x3J--6WLzG0#8pGC4Lp(&k(&{JArI;8GQ1U7s`^WucvZrZR13P~@T zp6K&j#)-ig8>^BLj&UEJub?8}F-I5(vclA!gv}PU!5;_3X*)?`Ctm%EV&d~le~z|Sk-HzIbsG~#e2;oNjwBluQ| zm66)&b+Zp<<}lBVjS>Q*!<}XnBIrsGaLa<)$`9`qC$K#+{o-+{nzDep(eHL&9k$h7tWDuU}8`b zl);EU?7Sf#r(6GV#$wW>a}y%nE&XCX5qCB1O~XsQ>p`9M-N<>#y&jSXoZs!{=9awt zBGcb201MX7A(Pn0Qtr@JxMU*`;PO4daim+c;zxrz|*o65|9jSIX%+ojtDo8(nQB=q-9; zWUNncC+AWiddk##7Wmk+CSDr091SYdps5P5PpY#No1T_O8uS?HOp4>Tu|$D?D0XzL zUz+&6bVd@~3zx39ssAngd?Z6jXG9fNreM+6h&^aAp?ToRdJOW?R-QoWM5KdAvZv^` z$L_G8vf1P?!*#lQ9R2r0w`P%q?t;f$4(>{JmzKva=pcFm0HjftTwRX^dkB?cDf8&$ zLa=z}Q?SK?W3j}@LWgn+smH3BS0@JdIbWjX!iv(D7UwUEuS z)3B!$Z4H?rOq~9hDB~iwyHokqt5-#pj-Wd?H+S*qF^Wqg>&p|j4Dg^*^P#n-fMRP} z9^WlJKA7JksKmaIlXw*IifUNPV+>HpMhVYlrX03>r6aaoJ@B6F3D9WE_YD&=>z8n= zFo}hVfzMxxb2OASIVq#acrIz@)^6Q!6iGNB5(&*P~)DJ}pvS5C38 z{>q(Liz=oNDb&_SF0UL<&OP{>K5An>YPy0H1|s~aHU|{j)OQoPV^SxSWeBF5{?u6U zs^0#bARRj@oHqa~qFjevS!B`NIdSk<$`(>+6dexvG1v-{EL}H(q|m7I=0}kWC6wo# zgmX7<3=O@SW0N4pRHk~1QUMWc;w8XNVJ)|xyY6S;9FZL}7@#@5^S;8Nk>HAoA(hx| z=TgLW6Ca~m7rg81W}tnVmAd63c^u{R3m^+CUtxA1N_7yA+R?j(qPd7&$&Z)4OzI+9 zT|pai+RbF|KF6q7O1L=KZxU_)tCE@^{rc2kKgO>Sk4GyNztE%Qfdi*abX6i3P8&= z9D`--x;65_FEu1qg~M*VESX6K^j{IpOv^jM%rE@sN!d*e= zbK+r`Rz*matltIBjcgI7!}pNZ<3@^{UNzozt*$PMc=S)$c89i?xzW+Za<0Hr{kgLTiFJ%!en9<;&; zbSlHK;pgFpmpJuH_!cx>f;}(*oWk2*&!%%bu=gD>Xgl=l_WNTSXm|kp6EJ}LSxN9l zsRtfg*r1B7TrnFBtmBjXjzTP0bf~<%d>Y(u{O$D}){24kM3+AvFSmqEA$rd+a-{-7 zI)a9mS3X8xf*(39`@$Scw<0bStxKWd!JTf}tFfY5Q$@-6gFbCwB*Kd@vJ<%}R4NDZ zAxmURE1C#>AbnD^Ov@8=7y64dSBa71L)0n_9LS8{c@hdL@S22JE1Eb|Tz^e$C}%{l z-Q!aNBFM3sOT1Vubt?}8#BxjgnYY3tUb7ETQZ=FkD>?+wJnlj0eNVQUTre|XWHJhi z0NnzXlJ*&;``=z9x@v5VRE0aTva8jKi4-yiI8l#;sJ&O4N^c0shS(Z|cB@^0C1qLY zc@!z8G?8L@%$!*XhN!;BV_p?5nrc+d7|ARM7{hJi=kodjD?fsh!2Y-_!Y#|ylS&+d z*(*gLQe<1XepgNADn>WZ+&%Z6kn~_!`f6kECPTOM7UFD(NBn;3oDyME+of=HYmg1` zE}iqQ`HuaV91-1&79l5oYJO!;B_5_Dt^n6Z>3bU1YnNw}3=7n^go;NU33}h6LqiUw z5OMDXaR?(%6U|AlLz<(iZw%JvY3U_KkS<#hvR$p4Jpv2Hohc$SeZ~c# z!Sa}|)l{LQY0l5qCzBK5HetnRF9Vu(>CVAG?L=pGR-nMfHV;NpgChJjfdzh*d9$ zr#OPH#nJgv&)Q<%P*JAgc4QnU0a^}VHiDvrOk>5Sp5$>*2>fnM=|AHrO9)*#o-o<7 z3jTQrZd8T?gZIn&{;Huhr8B$zI1#EClLr-+v29~-O(f6|77AAjC-1lLVgz20kX|%l z3TdZYJkC(j_UAgEvAWrjjUQ&o_5y@T8BX#|)<~=KH(T%E9BDc|K_4gemp+Lb2yukM zhp6x$hmJQG{hcO6a7o}Tc{({#+RH$UMOLGw>DFuA;I&K3H;N}0vDTf+0f3g@3*5|g z6*OXcQA34xokdLOObKSU90oFM-aHHrhVI3`@n9d2H1Zu`4MD?gc#UcT=2R-}a`a)S=C7N?Z%eB& zA#Et4l%oGiolQ+eSlymz8M>LaQieB(DHjvy-#qmVg;}9+5V$x&{ZYS{6h)~5gh~^F zMn|^o(E88eoE*a@7QvN$A}7uMg|UFgzb5{^PYGR}(<01c;ESG#v6uyDJq)5;D2{_% z!UF^Id;QCqlktnt?L-w1d>l6@F>JQQz=K#|88pIOBHW(@Bz3#=0Jlhn%gPZIiN()sE?jk?Kw8%(F+S1`HSZQzAe!BGKKRO|}FD2TL|96NaHZIo?5V#)|Z1Iy^Vf zw&-ef<8-_=7gLW2nY2fSBCFm5u$=?~6?fnJmrXLyO6(@#J$dF5x!$74OM`dnl^BJG zvF1rC0|9FUKX3+R5I`(F%3h;&xM4N1mo#d?-C+7mvi=P2kSL+>B^u4RKpo9JaLWb~ z0h3t{{I3dvNM-)+2L~wi((z9=a%7c+0%R4IBtet_lP1`~D69Wlbn53wS|zhni9Ve_ zzMJpUf6kMFnYR`T&5}m(FiJp6=A118^#V|lKoOfML0#K%D{Iu?P~|OhF@inMR|p(; zIM6h@KSc_b%p7Znb1Kqt>Hot{e1UvJ9F(r%(%2o}nw9-%p=H?fDD%PtH}1MyR7Q<1=e zDj}9Wk!Wf47=^;2F7NcPdY@<+ORC_K;q^^we@gL8LGb+5vr~w)fns06|$Vxk13omU+zX;2R>VI4Fgm7AJC~Y|zgUH7mMax{yx5(2;u~>YO*G{r1 zTz84V5r_jk>*nq6W(fENOGcMuL`6AFq!i;AEdzu#)Bs7@dN!29ECEH%EmA5)KSqr8 zkN#f`KTlUpTv<|I(dE$YgJcCpU+O+r`PNIK94X(LEL=xa1`|tZ7i}?V%OQ^Cec>lj z+g3ghR9eKoMD17oXBsIFFSTW+@;Xs9onZm&dtCqw0?QU<64E+ zs+6feq|7Z$EiOoA&Z%Ws6+7jt84zqQxo+9v`=3W~=fnl1M zDETNk#Aaw~$hb|mkE__v7XG9ZYXl0A7)dwWn5_ZJ)ti+_@jeI}Fi|BvhFnK2|C8?~ zGT@zho%EP1fw6Z$ywM>~kv2guP0-e@t2#v4LZddk zyjtO%W7k!?45nW71=|tfFS<8XLE;ywj?!nMP*iD>+cHp0;WuNv=mC zJbnflI2Zm;XYkF<%bNcQ*icCQ-MQ2Fv11DWwg-!H$W0BV*N9+z#Pm{5`E;;7ecGg{ zI~^2BT|Dtb)e+a%efSNvG{(ZZ3+Ay|>iQRFcTQY*bJw7OyB(<@mRz9hr8Gag;D*%9 z&y{eqZfV|j3xA%(*j1G<$cS66)9JkgBuT1>uRhYNt|BYjyd45_tGeSiZy<>#q|n4t zXh@5r(pyf-u_DP6P6LNsY0i-$g6yx>Ht3|;g|A`V%qB72r;QbvETUysxBDj|G%QV$ zN^{7lZNK|&{^}kOpGDn`AA<%!&^VmiYpU^@XLihZKq!kIS#KpyM0mk)1m#k!28@^A z&(@aAk)eJ{wfvEu^>Wk@q$L|`>)Z(^ZH=E9C@Y()!7lv&;K&j~c1{$ZM&i|C(xOhJ zArx*4@PjHb9brVQYK?-bx+Ck+hW`A$NxlOHC`VA9fG5(@bkQ#1>#1`qg33Z!v7O%x znBPG0amc0-0_l0QffsjsZJV7+X~W??SH2I z5-#j-D)`~je3F%&78wWWEv`E<>w9ynaH}xd_h}2B2%Ioj3`DAig6wZWT@3)XcVa2q-l@sb`E4M{ zv+|e+U%kU`e|2OJpJD|?qOB-{%7MudrEih*PyzV% zB2Z7_Tdu=8VaRU`>??IElt^pHF{|%gNVrXuEW}aU^C&ek4~e8z%m4L@$@J}zsO{B; z-YjKFhcz!3VSL+Jy{Uu8Jo3|IIQ0yB<^)_)09?3%5Sun(A z12ulK!F(`LqGaU*lDDDs&!Q!I7_{v*T_1$dEHfolkd`5`%y1gi*QV{aJQywj3#HN) z1O*}>@-suoBc*0Xd8(HZ3Z_-$p4^i7@J5=`)^@b8$e{3K8+0*VcZNM} zImu}SuP~LPNwesBOtAa=li=-z(!vFpL^FJNcVx?VD#Woj#X(f*OQf>7 zjLqAFU6$Iep~Q3cTe zf0AJP&qWOsfoC)kz?cMbCXtq6661GD(KisS2hkL&I;b92FDkcwetil*z5ZwTi*?t} zZf#7!VO@7s%zj(9Ze4*%ZPgy?#J!XVk-r99zTy*$PW@B~ruOdbt`u^wD7bMwbCW9= z7qUaM=iipH(6F`SfGfF~s=2O(!Nyq;^WkH|ZFi4!aG{Ta0{X3opa^D6nNjNv+!l5iE~*nbpG2RtrKef?t94Ca9p$ zFyOYY+E%LvKn;-^sbJwi-ckJdyocCnugM+0F49DL7MIQt}4qaLy@B?H!00wzV89_}D zl5a4BRT8}sh$?$5$%X5M0C#*Vtvp3fr%-e|h*qoAsQ`tPPow4;peF$Fd{c71aGJ?~ zrU}5k0-Uk82Gk!Xe2Pb;a5801Nbc0jD!4^5Q$qhxN~r;QSHk3$8V1o*IQIQt?Jj?e z|4=;g*W}j;T|c_~={NtXCwFTID2;LyYI(MPrqTka({4kua0MrMw*3KH4e%++`;xmR@YM{ zd-nRh9NehHcu2sM!MBH$IHs<@`?*s<&+?S@cN>i!-a|SR3vQ5A9KiW5!F4NDum+dV zPPEL<&J>gN@U%^Q^U7}(1^@EkZ9UE%Jb1807voD0G7-rza}G9=9abcq{jRPW*w$Nq zOQ8r1Y^?TLiYC3}8!)Ll4b%pI^v_?{b)+?)kf1)ZW@Zi5MI4XczyJRq@qc42hWQjL z6!+pTUbqlsp=%kV)hXCKWKl+QskSOxWw^B6=@zk$O-!n^LcnON+b>fVj0y}RcN+5e z@sp)L+~0n%b?5|J;D>@AC@1)8ucwtDpjnr7f!E?TAEC=|aaG>%ct7pvyn`#sTGzg7 z-LCk%(dnbg{qE+Vi8xVZOwp}fam&oTACUn0KRXAOJg7yXDSvev@qkMp4&O?S%33Jakx z0f#Tf_iTGan7=B5h-K*T;l4I*B!p!E?2CEi=-mBtMzq-WS5zOL3E{=QACfL4j%?=S z1Nu8^c&5|C?x#z#;%n=_aFk`RXH5xFsZ$aO#E_NG&iM0L9U$$5hJ`(R@IXbHxEMvz zS+GpNM~)a3Upn@9$3gJ)#lRrghg7874)=|?)Z^mCi)mLjx?QP^E6L@*rQj?4?OCXH z)lT=~PFY0{&ieeKvTV2ghMtV8Puvq4;nu2P0*OKD=}i@3R;#Zzk;TMd`j%|SH!=jD zPH(ZP^%K9UTUJku(nRjT^B4uWB{2Thm$+pe8ny9LP-v(wLLayLobXps9yOGIXEFBz zL0<@sqQ!%ySwfxHZ_vVDU%!5BZfWU5r8#f^96KeD{5kOUm4wP;0$RkJURd?%tUM1T zzO`3MciYQ`$|8KKEOGza*=R^tzhu|mV!bSWn>=~4snOdD+P4Ycl27|i2+lyFLaBnx z(8>HXJcwu}FYeQ(s%g?-s2da=Z72hs_G$M%>||b5K~$1Y z{xw0>qU)&Glo)y^ru@@-XhCyM%;kbR8y4vm#l1NRajEyr5Q`+<(vKbJnUa#?XXt1P zPBmv#EwGs3?ad~sp<(HB_YGR`_rJ7n<2F6rxBWG*3xQw1xv>UP=Y{d-h^Nn`Q!d4( z`i9MgSug(EvzdB8*v=`ksdYuKY5B(I%T6qYB4Ai_>ileDc<}(!?QzpBp}>XPL6Zvwd9dHpm>eB(JCAtP*ho9&G|G zbbYpDM$w2lh~_tQrgdcMu75r-ct5Ky5b_0{38mcx@%a zE-4nJWw;iF%N-r;Qk-$;$jHlBmo>E@AJZ-$tf8WI8|1 zrPb?B`}mJHT67+9!RqFJEZDYF3>W-*y0!ITScLgEwe0WAT5{Njc-XlQ5|!EioFeZQyl`H#1YDLg)O z*s#T{Q+SuQP2$qHYRG|Aiaup{aTJHoxcttLs>|`8FKgd!(%#CeGB$tYwnzW*HWrr4 z?WYmW22SHR?!CLQ*s=5e4nvok+0n7dIFK?gN6J+Wh|gu(#t}*nG5L74b&}1EKVH}o z^~lxQwm8)=-+x2p<*do8R;@aepXoSy#+iVrEV9-%Pu6|$C8z&XVvHn>yBpYgAk-`aa&dQwvF=MsTZRI(6y{$+Y2M{y68We z!Tz|tGCi_?*XrzM<=dxh=+X7_vqa-B9be4n^3H-=D`g~lB!`heO*b7?mE-U7V!fZi zngyfYZEoEuvLzePuA`P#a(XYPoa2Y=l8ErL@}lb=Xt$F)5%nJ0P)(+o=cE6CFO58mdZ*cIclxMSbjeM1-Er$7}6l zV&CG(F0OA)dEjC);}Sfh4o-04AzPgY4(?S^QE}LA2^my0x6@qL=fR^#-rAhe!*)?i z<>Cg^9zN~TY@1d<&kpesOSdFpNOWsEo&K;VXPBa(Si;Y?V`elSa7|00aP27OJ1&*N zQ26s|6OMC8cz8GsI)>l9dKJI<#VR0_@SE=PT~q`6KI^G4So;B!6__3t5n&o)vAl8# zRmt3Q?8b_y418tZk9}fJaC9fQG|)}ms2IIhf?Aze9F6`Sve*;lr)~Jw!+I zC}fLZ0jJT?p(s4c7N*_!jQBtQ{L`2=gJ{4n$D}?dg6Joa-xv&6>eSr07{{(PKl&FG zt+`ZM*nE6h7yO0TW=E*q3U4LUN`-{pemAO0mM%)6!n$^Yk15rYEcm}t}wYT=BOWF3>nfZProR=5J z6h$C06m>mj)$Upd#AO3&%LW*Jeslg|ynF2~t>JnggDI5vJUtf6;;Z(%5h~1Kl5^4g&yl?31CIuco^OoT4yH)z z)3IYm&zdn`0)RL!WK5qn?cx070QHg?W>iHiT}XHryl={==<92KGe>4X3 z#@AHLPgj~#$II&OJZJXo3ANRCYcuA@Vc;K7Nq@ev*2S}yH1@vT9GE~l;V7|Rzxxmp zW6bN&u;;55QYZ<=fQfeN1my%8BaMT0NBkJD1s59b9W-=tl6Ozt){0EE9>KlM0p$46 z|Lo}dd}J-GhWv=j#Yv>?gE4ArXB@ZV5CtIqXSCroA;@qN;1O?hZ$Uf|A0K|}musdu z6(@sWV%cIh6RqOIZaTm&^)p^yHZ^?L$Y=5@V&LAMnKW(Mv{6;Jr`SRq+P10$e>d(Y zQ~zhhRUoiA$GU!5x-D!qoei(6%Y5TkR#w})e@?gJGd?>GX#jCf3}QmttC@L_!h+q#C^1rG1+^-Z|RV z^s|-#T2-aTe@WNAO~`V7>HVWCGh>Fav(Si?D0zS{+?eDyxR+7_IAi3c6KfmBUf&+m z#YYH&#wc$pRl=(z{=7r)EHJ-$mz@v@@-{c?a*Jr9RY;cQ!f1*pR&H-E=4PE{U+7W$ ziDQppk$@J2lfvqb4KKg3F$k)G?GPXVYi?HUm#o^6z+3x!0^rV z9`Hjq(yihTNJYcTNFMdJFnED^{GFd5*~j3vCY!+cb;DPaFMG5YBd4uczIL_kyLQ9-sMx?K~B`Hx_&qx89ZdpQ2&1wO)c!- z5t6#is7ee_NXiGJ2bK$~0Ozk{>m9w?|T!Wkqr&NEawJ-b%@ zb_Eggu9haa#BxO4dnrK}y&-N~ob$n^w!$Vk?6d?#&d-WTD6eCIf>IpmSWT6rB`2ZL zqZc86{nZ`b$7I!p59b3?(v}2d+8V)ii7rnyN`L#ylAFyGu06z|IAyCQ>Mg6xNbkPb zq1&0)w~%OisePM*bSuLsLeH`z!B4FmCkwX1dGK|AT~sv9#wH*pCWh#2F%5*dF`Ouq zgL@t>>-~?b#cYy`VmMdYx3Oko`SX^Te4fRag{n8;@YiE7&m@b#R(-u58dbGO`pJ>GpMcHs-HKS6`%dZgS)H=7A%WNSapsl1kDZyrDXHz{Zg4YvsPZ zr}mTLc(_AXMZtD)llL}X4^P;T?)IYmNUL<*!xWW%*HIVAHOtmu8>m8x0+x?ECiB6) zdtS$nOL%s1`LU3Y!he}eoMG_Y(_=;l@Bnnh96Igh_%oPa(;>>SKU-{kyJ8mQP1%fm zjCt~9)JI8ljPV4khFFz^p3Pt+XVp4<()i3;uCLD?6DvuKjMt>OqZ61=?)wamo?n9 z{K#~>iM#-^jb62%PS+;29G>CDo8;1sJiui1+j%T#v(rc4zPV_u`ebi^|6OY3sQr=8 zA5OiTST;9tFt@Occ=f<*t>(UsfwDx-)3athiR_H4M=D!!9t;bE_Bq_{<-3SPDUB4# z?(*A6?c40bc?^t0T13a>E$QO-J!9^IVfODz_wk1_xfEv$%huq?_k8-|`8PKI96Fw94N6m_`(9PXxQ@!BwTPNFp z@80l%v|yE$!UZS`x8K-T(&ITJPlXZy63*PFj^f2h?7VjIPl>iv6a^a?GXL5i#uOsb z$QIVuKI1^fl6FTU1CyL9mL@6`BVUF3M*M~iDZ`%T5Sb(s1w6>e@Xs0jEjmswjEPJ9 za72%IG;h1XJ8?viA z0^Z@;%_45|p@Y|#<@BR!lv!uDAATh#M(amIhedQl%&hqW-f$ag>JcIY3rovj4QK%^ z#SzfP782`v$U=Rey*eW0LcgKQ%*BxqM#?V}IZ~NQ6&ywT_T?895qQ2}%Q(&z{kj(| z-x;~rRQs>F*#Y{b($dlr%x1{E1P^vzpMcybr@O7B85h6*An(srN^*v#?Z->{-WAe+ zA*GCbz320Hbu}49fAo4)qxo-`DDnWfw``c%+E9UGop#O^=UrL38}x{OEnABl>{r*p>gai0frjl{7artl^c6Z6dHBCY!FGBk?upOYE5FGM235-0GLtPR%^q zgLv-_{9ggot=6V3jz$i%oX%k~gow;f*2B1#(=2>mRE_LH_&a_MVfd(ukSVbLW&sx^x3)Me zH1r%1)C8BW6`p?o9hP6*T7$Zut0PPQKL+d{)izogTw*)OVX&oSsmgJtV :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/index.html b/docs/articles/index.html index 6021931..1427a0a 100644 --- a/docs/articles/index.html +++ b/docs/articles/index.html @@ -71,7 +71,7 @@ rco - 1.0.1 + 1.0.2
    @@ -96,6 +96,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -114,6 +117,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -126,6 +132,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -167,8 +176,12 @@

    All vignettes

    Docker file
    +
    GSoC 2020 Final Report
    +
    Common Subexpression Elimination
    +
    Conditional Threading Optimizer
    +
    Constant Folding
    Constant Propagation
    @@ -181,6 +194,10 @@

    All vignettes

    Loop-invariant Code Motion
    +
    Memory Allocation Optimizer
    +
    +
    Potential Optimizers
    +
    @@ -193,7 +210,7 @@

    All vignettes

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/opt-common-subexpr.html b/docs/articles/opt-common-subexpr.html index 89745e8..bd61d76 100644 --- a/docs/articles/opt-common-subexpr.html +++ b/docs/articles/opt-common-subexpr.html @@ -31,7 +31,7 @@ rco - 1.0.1 + 1.0.2 @@ -56,6 +56,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -74,6 +77,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -86,6 +92,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -112,7 +121,8 @@ -
    + +

    Common Subexpression Elimination

    @@ -128,32 +138,40 @@

    Common Subexpression Elimination

    Common Subexpression Elimination

    +
    +

    +Idea

    +

    The main idea behind the common subexpression elimination optimizer is to separate/isolate those parts of an expression (sub-expressions) that are being calculated again and again. By isolating a common sub-expression and replacing it with a variable, we can limit the number of computations. Thus improving the performance of our script.

    +

    Background

    Common Subexpression Elimination is an optimization that searches for instances of identical expressions, and replaces them with a single variable holding the computed value.

    For instance, consider the following code:

    -
    a <- 1 / (8 + 8 + 1 + 9 * 1 ^ 8)
-b <- (8 + 8 + 1 + 9 * 1 ^ 8) * 2
    +
    +a <- 1 / (8 + 8 + 1 + 9 * 1 ^ 8)
+b <- (8 + 8 + 1 + 9 * 1 ^ 8) * 2

    This code computes twice 8 + 8 + 1 + 9 * 1 ^ 8, this could be evaluated once, assigned to a new variable, and used twice. Like, for example:

    -
    cs_1 <- (8 + 8 + 1 + 9 * 1 ^ 8)
-a <-  1 / cs_1
-b <- cs_1 * 2
    +
    +cs_1 <- (8 + 8 + 1 + 9 * 1 ^ 8)
+a <-  1 / cs_1
+b <- cs_1 * 2

    Example

    Consider the following example:

    -
    code <- paste(
+
    +code <- paste(
   "a <- b <- c <- 1",
   "for (i in 1:1000) {",
   "  a <- a ^ i ^ c",
   "  b <- b * i ^ c",
   "  c <- c + i ^ c",
   "}",
-  sep = "\n"
-)
-cat(code)
    
+  sep = "\n"
+)
+cat(code)
    ## a <- b <- c <- 1
 ## for (i in 1:1000) {
 ##   a <- a ^ i ^ c
@@ -161,8 +179,9 @@ 
    
 ##   c <- c + i ^ c
 ## }

    Then, the automatically optimized code would be:

    -
    opt_code <- opt_common_subexpr(list(code))
-cat(opt_code$codes[[1]])
    +
    +opt_code <- opt_common_subexpr(list(code))
+cat(opt_code$codes[[1]])
    ## a <- b <- c <- 1
 ## for (i in 1:1000) {
 ##   cs_1 <- i ^ c
@@ -171,16 +190,18 @@ 
    
 ##   c <- c + cs_1
 ## }

    And if we measure the execution time of each one, and the speed-up:

    -
    bmark_res <- microbenchmark({
-  eval(parse(text = code))
-}, {
-  eval(parse(text = opt_code))
-})
-autoplot(bmark_res)
    +
    +bmark_res <- microbenchmark({
+  eval(parse(text = code))
+}, {
+  eval(parse(text = opt_code))
+})
+autoplot(bmark_res)

    -
    speed_up(bmark_res)
    -
    ##            Min.  1st Qu.   Median   Mean  3rd Qu.     Max.
-## Expr_2 70.96235 65.62056 64.95742 67.424 58.25943 148.3645
    +
    +speed_up(bmark_res)
    +
    ##            Min.  1st Qu.   Median     Mean  3rd Qu.    Max.
+## Expr_2 73.61738 74.05437 71.04682 64.59769 55.53048 52.0365

    @@ -195,12 +216,14 @@

    Consider functions as common subexpressions?

    This can have an issue if the common function call returns random values.

    For example:

    -
    a <-rnorm(1) * 8
-b <-rnorm(1) * 18
    +
    +a <-rnorm(1) * 8
+b <-rnorm(1) * 18

    Will be wrongly optimized to:

    -
    cs_1 <- rnorm(1)
-a <-cs_1 * 8
-b <-cs_1 * 18
    +
    +cs_1 <- rnorm(1)
+a <-cs_1 * 8
+b <-cs_1 * 18

  • Use information about which functions modify their parent env?

    @@ -226,7 +249,7 @@

    • -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/opt-common-subexpr_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/opt-common-subexpr_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/opt-common-subexpr_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/opt-common-subexpr_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/opt-common-subexpr_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/opt-common-subexpr_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/opt-common-subexpr_files/figure-html/unnamed-chunk-6-1.png b/docs/articles/opt-common-subexpr_files/figure-html/unnamed-chunk-6-1.png index fbf466afbe631a20fe71ba6fae67adb5daf3fe7a..915bdc8a7758baa17a1db1e40d185c2584d7f914 100644 GIT binary patch literal 30928 zcmeFacT|+=(=Cc(81py=R0KwCML|JDKqQ&94PZhWRB{eF=72ULBGHIQ1_8+! z5lM>VtR%_UpScIan2v-uJd!vEE#Ct_X+h>?b@~LnXCI$c1)YRU@{*c z-!$gVtq1t{#@q1m@gM(f0=}a?_Nxy5n()`|9b54?{`7B2bjDwk&hFGO=i?K)O#hB? ztNVG1kMB=D=GIMzYb6x&=_V@jux{9df(tB5BCPrNdkT|&FOHyZ=`leaG&7M%ZE~}+VWT-gq z@$=uLTH>pJ9x)Dy?az%^3P>@~-Of z^!M`Rt^)LTK)^Vf51-5VKmM0L@FKK}Z+31=Nwm1)SkwLL>YfiCJYZzxJ>C7awDjn= zt1dZD!onof)YLQ*EY3XLF=oS^7cX8|w3bRU^GZvb6vv)E@!tM#XI(zir<*K%oQ4NN z3u{8EFIsLa3e*1j`Sa~dmoBlrSEfy_NU+@P=;-(~oo|B|9>5+|wrHF?Qh)27hh`}U2Mq`x~#Py1KfJU&uQi$^9)U;+mg* zTYR~BuA_V=aqRZL!e zyEIfM>1>-m_%b$D@6+8CUTaV7avJ#RU!Q+aHQOYyHFj>`IQ+^nM@Gk+M4Q?>JJ`PM z4x*h6xsoEm!NHjli7hNw{)x{JF2Yj=41}GBekS#l8ZuexI<`ySUY?QrdZLJryQ-2L z9KOc_5qi0$Pudjm{bC+wgROB5mz9T(_a8x!NEs*yum8BjZBMvPa%p+$)2B-@Xn*s! z(`0Ke3h3-)m1F$2-!2N6;op?j5j=?Rv+xUXMLr9erw3NZRymk^ze;QlORSGAGs$F% z`o7FNqIvM3yJK&Kta5*bs@bO%flm_b`r`_fM}E3F43<^pq>c`^$H*wtlbUC^3RHe? zP#(2ydsTCb-MYk0VQrDFv!hJON%vj{BV%LZl#%Z6hJ@y@-6u|*VAW~LOcmO&E?Pgg zEy9ap^%3PsNgZ=jv67jiiZ&h9^iyeR>GozhrJhK)x_WQZoC$(Pw;5h*PHfLn9v#9D zW12(g%^_9ebZo9~yFPa*Gr_9$!7NpUH3+-G-3t5@rhc5^)N!G!wM>?|)^VUq%i_zn zGbSb`aT=UArs9p$OKaYo9vdMrLpAaGQ0i_kp07A9APB^T$eBL>R}G zEQ9H+FzxI}4I5Xe8WA{(H}dMKUIzN#7$PA{Z32RSbZ_3 zw&U%L+;5jf{1|79jkAAzc`T)$<1<-RO%6*Qt(OgRImj+b9g%j%da^%>hd+sliOG9R z>(Vw1&~UiulhoKccG?~ap_CzRxC43 zbzG`Lj%#D!(vHpJ9dN#KzU&b3oiuA1mP}e*>)f`m)RBR8GBStlURoW**=SGht^Xis zy2sqyyuZ(748K{|mmT`4!#$df&Dv2XnImk^)LMPzL5-1--c(Vhj$PwZAIF{&xu)lj z%~~SUpCy#`e8cPbu3it>^nU96wTv>O=)J5uakcQ|p3=L!o=#RRP@22y=-oR@cIXZD z{j7Uc7H3+ey|>flTZB$n0nK&dKV@VqEDnCSCFH#-ws>&{d)VF?!`JKU z_;$v?;HY6ZEn$GvESra$6D%58c;GX`qA(`w>MYUvk-_%xjzfkfJuM}2%^}h3>I*aO zxwWQ_j>Om8l4oC(bU5yzUUgVat?r=zz@cS(?%#gwrj+2|z^=`ZtGOXX2K92Vg{HA> z30BQGzV049=2Kia)+%X;ojJ$nbaqnT_v`fOwU$l!jLaoHS7t6TD>>!)R6o}*xZ6iN z(JEU*-`Atvp+P<L~aEm?K;!_!5S6jYV^PbT`&a08Z))A=#ZuDi)& zw`B6mq`Yo{)V^)P_AQZ8D{B%$${QEZ8xkXZ_Vdfi0j zzD-ue$1gK7)fIKmz8+8R{<05c9-KXNFNVYHj0txPb+FI6zs3t8V!WuGsk^9Zj`MJl z^Qp$Xr%aZwRq2YV(UIP-u$k$Krn!EK<`tGjrviPgcCnrB^oor3?YQhmJ|-XF zU6)LG;>T(D?gRK`10_T(Q6?SNDUwO6 z_wL<0$Gagz)IY1^+uO`@)|Z8>e#|~+eQ}yh@t-y6Zn^LpJ`688b3F<)S-$ddZ`{ct zTP$=H*boF?dSijKn73()E$ifm8}p4N3}R&zNBU~!D>>NMYF-kuDqd9c;pT2PVa39M zmF;?8_sV8>7wP(^I1Vmjk`u%No*EhIpW`p{O;g2>diDF~`p%kuAZ%Zk- ztkm`JhM;U%lHxMi4&e70X=`zfFk(^x1AO(`;Lx?d>^p$@bc?#5OSz`L54fq%eG>%rV-z1z)U6&jiZsA3!k7{`_!D z=Rk9~y3JY)(4}Wl$7a`sW*-;Dcf6T$P&;`@vp3v%*f2+-Gm9KtD|?8JUQnRlgA8|3 zPd8tf<@_~&Z8b13D6Ly)Se7H-UaqHkRY2Ea4TbK`rr_|y+y_gGc5HsV(4pgX{P%0( z$m=~BnHTCNiTd1zjRhw8z8REu9vQkJBrnt8@9b2c<>k(Rd(2f+*Y=pDvi@O`^w#~Y zFTbkvBf9wOB-<@z>Ns{6_?-JNpZ1r#QE3lwxLx2eb8BI8S=$AT?=m!Aojp!V?c>1C zpop*DZ%d~R@lR3I=j`UT#P$^SPvf`>nYcYqvzkyQ{Ox z7FNL`*c7z@T{UG&Pd}CsKP!MS1o}H|a9_6flmF;YgN%J&VPRptq3+;OmG$5L&Mznk zjVX$h$$lU$uJ3bhyB{JcJ95hm#dW%)14Tt8MN@CyycrY}lu;gUF0MHF^VLJ@>Vc+J zDNI&(SW!sOU z@G1jEmFhajSvK;X2ZHWa4b{uW12AWni6?2`)J46XAowI9VFTiVZlfdXWUgXwR%4*Q zUubOUi^n^{UR|6bP=9`k#7CM2JCcxHgO9n2OqE0DWR(d*TdKz&-d;Vv@@CqUq*<2_`N)J>{d*fyK*PX0)UOsmD13+8Ty|CHc=+>HBi-mvZ|H_!9H^wraR0A z{wT)u_4)B+aGk{`+z1hp8>QXQwklalOGQvkTIq6{rb6buRl5xh4dDn&ThcSlI_tAD zJ=AsjV|(2ytg94t`#-5iAzG-NUVB<(tYpO6v63Ao<8-Qhe@hb5x8yf9me1Y4f7Pm0 z!Ck=x%}wFDl3K4;r!O?mD!Y~IBPW$Q+-llcRy87nXuKmwu3_OFQ^kXti7Jl4afvuZ zb?i*ZT-S-h_r=4L)`lmscd=?C(}gonxGCOqbEsV+2hpXS+nC{E!;1IGrQ@Wn+a+~a% zgI~NuM=!3zO4F(Dt;jPV6ehdtt7-!*N(IKo4pMa#^Y42n8r7L2AGpI!c#hvZy!uF0 zsJm-ig57YEl@?`Ym{n@4jkdP7Rh46}7NF*Q8KWakLp34|NDtz!2^@a3+fzcWr}%_< z-?z&Rx8&PC<~sF6XW=wuu%{GKLXz3;)SHM?qhFoo`U^RCwO1*78^>w@$<*0XR(iy+ zFe7cyX_)+KOp%T^?VW(u+1F*?v)){sV*dTAs2<=)eB(3kQmwlNA8#+R?XUBcI{xPG zl}94}w63((r#P_^8kjEYNv$#8ke)TO{uEmq+CQHu(9DwdVC%6awLNj1yCH zTH~sem*zjRb9zJ|nkvpdO+Q7iTLhJH#a`pF50A~*^MJ8e@Z1>|Kd#+??Xs9jn zB{nuD##Re!GSJiN4Jh*L;lo80iPox~650neHOuNRtKyTCw@01Iq+MDkD;tQM7J;zt z1nh^>Det&l;bCDaNQ?~(4G)5Y4bx5}?A%zHY~R$}{1r(~*N-ntrcImHSP-nfKv3{& zuzGk`Q-QdmeUFx@Gaos}W-(XU41%GQd)fl|vyo6QLg>=%H%E@^jcg6#u>wSNBmCp3^_0=vON^K7>3xAR>()wC&j={GMO&)EJ3HglJ zF|U-JUY22OHRX@Zdq|vE#S2+?=oiJ^|T~(wX?R^U|hiss2uh$nQ24d~8rnay?3c6a(!`mL}JrHz?b3)iq09{KW!K`$`LZAN2NYHHe);D^W*ix=`&9biJbx}q9Df} z>db8bR0LK?VgUzZR&qi;7amB1xyZ}UKUAq{qXvHEDYDhZ(ilU^eI&FKh&D%dSm%); zet3eqnwp36{X_UXNcEAO_eReezW;e3=+MZKhc8};*|)~1-%=X*VP#iqLro!TD3!=BNlDe2j)Uns+>C9c9 zyXq5b>F&Jp^73T^Va0mK-|xVI@OCRLEzQr*M-)pKa*ADiBdoaeI1a7`(pJK-fJ=2Z zWh*}d@qLSe1ecNYl6ciuP&c~_AFDg>ESr8y)+ga$0(_=ihHDUlT z>4(-l^jJ6M%6>9T=x90Fry{j_vmSaNYM9@hwrAc`=4X){0d{DLpYB zmtG@f7FXAGwDhz2es!g0Y>RPflLL-gO3`kO6F6i?L{eWf+cf{fLkiJqX>m_4+^x0nTdrNoU zd-?TcFvYb>{S7YRD>_u*?~``$iFcb^59CYQ@zP&9UN2`uzw-Z^n^$S<{2JpJc~QXzd=lE+DDXIJpwTP9ACi9a31IYdv`U z$kLMS6g9=h|C{1Zk!l+t^8Vxck4b0Z856|%x6EA5R9 zMc8nQ=E7A>Y{Fv<3)L;WSrD-H_0TdNSlW7p2&+SwVZPEpw-X~B;Y5(b3SKf{+?4G4*&y}vBtw4U6mA5gUPwcAe*oC$3= zP27yolE-tVhyX^N`iS6p#l=S|O+(>?j3uH9VYOxsIzsJp?A47neRF&^-2Y&#X-{GA zxb~T8-pT!nuLt=%R4VPh8HsJFc65hh*DH=V&Sd?pNpA>L9zH|an(lCuIvumB?nnZM z2)Hv%7xTD}u;*FR zdhK}`LPf!_SxDrZ&f-ubSoINvs-INO`Xyu$<~28E^@_Bu zvnMKbxLP~3ID@m%2X|pQCCE*Iy_0nkyu{I(DN91HVXrvH9MQ_3oFTIJS=Ps;L5J<} zQ;?&SgdCyA$kwRXSQ$;@7&RhdMurFVqqAi)DZfD${46r^+|99iK#+2Lsr}{Rv0IL~ z&v}VKrX_5t9vwE0dZgCb z?mQ|_thxKj8HaQS@;oblru<^br~6O3MuY-wQzx;^a+RP1hI`5+DPxW*IIO8Lhe#5O z1sK|1xZ6arqnesh4t*Vr3`(K_x#B7scIyrtAfSa@g}(%Oe)M2lVrr|Be}9&K6iN-X zm})LK!UxELI8)!hL&#$c|Adf|>l)L&|gJcB_AXfpI@;+t~>u=Cop*fGSJ;|CKY0e$@j}aXL7XwIY-xk$KB=j`QdcM zuFpj53L3xMPbus!x9Fpk?vtk_UXIw;q>+K<6&}0JoId^WnYYOU%5g7DnDzCzx0LYoz1}jb2zX`7R{0=pks)tUblu|zqM$M*j>FwW`tK#z zfXA*!nvn~9rtz&2592%Rn^rby6|ngZA)`_yzsk|pl+n?_R+!>}xKerPnOWz?nqb43TOl;!98+TB{)3gsFD@Hsy=cUBm$0Af27e{^ zeoLQ%;q32CR(||Hj;SgSTTtc7v&IC$Vy#`*4>%#C7|FY_$Z_i)7=~s1Oc(Jd?MLPk z_PKN+Payi|6?)-!HKO(Ep17-srB|3#BoMhW)0!sp2Hf06qOJ*bQ#Kh@qUU|8r8u(H z?05tov(PesKY4)YFWUu|YNW`v&)4_4L;UVd|I`CO&|b7MlF3~MP>Prj@ahXUoXq7X zL*&5KeFs|5zch6|3kMwJ?!`9+>bkx_@%*iJegsSB%bI9WNru71IVTcWWT<-LiqcyW zbiia&e%>U~jLAYIp$I^krZ|sO{N3gupAP&V)j~L|gi&&LA%`1Fv~Hg+rwbBW3bbWQ zT$NMZ3FW@mGt9aR)CHEYbG)QP1AWYPdB3u+^HA4A>K)Q@jO?g&XZ1yuFs#9rOH(s| zh3-*4;fyiGmkzduy#Nv+`;hy7Wyz;>DsAk4y!|PvV_R~QpApzSsH!N90K1#*e9rp% zs@UfpCpiz@jfWyzJSEqj)cNUQ6izi!os=|z8V7{&!O@oc_V=t$>1|1!ImCYW%CIGX z#c33j)tu}%sQvPI$Mr-!E^Px7k86RgXOq)pP+}>O+?ChWTx1%Wdk)F4UPtxox-ukJ z)-d-d6#-iSJRhlC1^m;whs8-Rqz&#G6EEvxc@JcHzNu5EHxG;d4NsWxCo6pOp8Su13#D5tby4M_lEM2WrpRUxweAoyLIvuAWKzc)BUHoY_!4mt@OV8)tnj#{zSx%YDn z71^<|vGv66zz=zz>_0r0NI9jSS5-p_XxQ*N)nJ>Flcj&vod_?Y}oDtIO+foMG*pY;GaQt(R%2C03qyC>LHLO8Z81>Wy zhd~sz>`>$Jo~+VUiP~L+%$;8F(4SCWxqmA2p5+f)O-)U6EHm*#D@6wDWmCNX*iyIT z((-*S0gFEgiOotP;z6~zGbnU{3^^lnwaEJr497t zT@EWvm{GpgfhOzg1w%=jCR=_Rl^GExOUZGdHY7S{7x;jSfk!QA@d)j+bzaX<>N|#6 zo1R2&jT1h8%3O|f9UaOYz3b-K>0OltZ<-g@K=k|a$yB9-yUJY^SAHNx5mhJ4)nMV;P*XgG@hpL8EHN`O(; znv}nqI&XErTW%qgRSFGa^JA0*l+?8?)%ts?Utgf|RK)pNF{y18i5OT;2H@gPkOfL$ zPInX9>JHYeQ%@b>faOgWCY%Zmjok&B1$N?d=IgUMMEQ@936v_R17VrHnIu{paACTb z?njO~!BwCzff#MfA``&Wo#8*y6&O7_GN^9z;uZ-2^a%1& z-UeVp(_6|TedNxJZ;&PLTRv7YWE;T9k{!3Qz-!-UK)lqH{@fiW&s zRrt9zrPH_luBktUgOh74wtoL9W6&N{1rb-IPi}k|e{^QZc*)cuvrE&39^AWkkCB0K zcwU&k;rW|`bSCy$kh~+_yrEvRYTz(IQU*PW9ct>BIhRx%d?9NfYoZK)MD}{|xfx zG6{1c|6S6im`1|q-Wwgn0sjQ&*PG&h0@WnB#wYG}1!6W65nsc)ely$#do}?sK!AI=H(zM=v5i!tn!0fPYLK+H00f)YUBxgBpt68@?MMW?ljNrZjxyVTaL&xi67N5`M zi`vWpU_8jd#z=hPaaN=dEj%j2k!=z^d~5ypD=|?993{{=og?G%8aYH=5@jr+#2*O2 z9blK#DM-h22#6z@gH4C!J*KUX5T0rC&GGq*7rT%~eob{A-Q`B#<_ikzqQ9_9!vFhK z%m3@T=l?u`_U36NEz)j)MI7HOww5P_bSG?-#-2^Zf3;! z)B9%}dLJwCrSNdb)A5?i-$#998_x8&HivP|C4X1s`KN08j;mY5>4eyKA1;?PiEW+bP^GenK?=AUpv7I1 z$(`WK6_C;w*Y^WG&t5(4t{bOuV&icXJdS9Ojw;n6qa3-CMjL5kaP2)NYe5 z9|f@zgi(NMc*~;?=ktE&be1gNs;a8mU2i~gqAV{d5`8i>bL%aqOSV9}OG94dhZ>nt zWs)4z4B_-C0v!dQ?(WLIDJ}PTA3-8^KwAV16O&aEt*`DMeL=uVrJ5rb{2h#aXqbNX z1CQGN&)!|r($adk{T;KvL)a7&X3yXL^2}Ef05vs3&w0BfU?p$DKY=RLicsd;$s)C6 zQFBqas=dAaqUcQ;+!m62?WYGaAiDHr5T62NMR=1TCJ+*4xMX$2Dvm|DBp8~Ss_xja zgLM)b0Z?|&Q>?08Cmt?Jo;j~riT4qEg#O@W{^{!jat%Sfg@^fL#tGTuPBFOqVx8$J z86MV)`npnh!Gx;P-=`Yw=1#7SNSvml)fW{-c^?%&9P&J@7ulauI1$9)%$a$zS8c zEup0+1PxBC{tP)U?Togig~droxR@*$o^h{YOro8qq^^!9X>+`Yx10q?vVxn1TB}196JGBvJoc~8ww<163LvyV zFz#7!sZi=hU2dSHZRKx4 zUWs*cNx+5$cclPj<-?+IOP)xPV`{k1^Q z+V2^l@p0bbd?_jMx^?RoBhgA>3!ak0d&;)fX@%o-=FQ`MYQLb3sEW_4-TuIH7kR_o z*503(5!PNUgcAg~>W&mhk@xXP(H#&$t~&{h4U>g>)3d-p{e5~DqIjl!T~Y4QBBpeQ zm9@1#SX6!`fxBP>ICcrwrd|XER%u0`0VPM@cIcfE3-WosZS!S3x%f{aA z^v-cQeS3esG|;tZx=t~28O~1qP<%wt8by=Dq4-jvB-E3HI^qMeem$dinc*x^l|3hF z!t{giVK7 z3t>GI3=a5EzhuE5t^x+!g`b+h8SYd_HEE)oWN+(ZaXJ)u z|5;QelyopqB{WPuTu1Rwsn`3sFJ1j(!dkNK+<#^xt04&a1a{7PuoKg4<^AB0qN1Wk zs*@v2u`Jn}L28ZF$2b4$?JsT$%3?o7GyEo2*7j>Bk#E)4Z*a+)aEdqN=L7a5CXx*u zZ(_M?JweU5+IQWVJYn5HnT{}>xHx_ zn!MS*0BhD{7jWzJkr)HDRK4>LH*;5PU%`aRk(TNTc&K1QoM_%5+ax_tRs%ugpQrI# zy2q?I`0PpKIm4Vk{`Ika@;2TN9zb;Q;AHd6X^0~g=g zA}O17>-ei*BTQao1x-xBlDR#HT5lFE{EBQ9TfukXd2q0}kdP2_09FzX&#G1=t{&FX z+a1utTZz3m^_GgiyT(~fY`-=~e85q0xjb**4hq@~nqP>~iIKMv`ts|6r%`<@=*C$d za7+fV7?@zSm7jM)nQ$Yi;ZTP19O~0Am(-bnDEqS6oY2?opaH2nExe^^<0+a?jx*I1WGV zNeyaA9UUo%Ny|LTbC+Tb1@?Rg>bQuRe=?&$ z_*z(4%vP8=dk6Oo7LX6?!2Vso@s7R)NfMg|AZ``Ilff=yoi{duw||^4*JgeWcU^bs zgHGTKW*PT{ZNLcwLH>T)*zvr{`0|8IOqQs>!&=xlEe8+CJVha6DRYCod?!s;VcKjc zB=MSnyVK=rZ#NBqoLY`|^MKKcNlCn8d;p^ABbLL^K)gs}2CU~z&f?>u%Vcbof7r|q z;|=dj{`36>Cl(wIJ?a6GmVm;{e1Gml&ImzmR?v;y{Ae{}-&C$CwctDn;=G=<$&qWscMrorNVv*i0_&(cvE9#$ zo*&>DKnzMO@d-K+$2PgTZye_RLMa-e?t)!mWL%h`bRR*1eu3L&ygVlIM0&kixSbm7 zB$$i-aIi7|`e`zVfanBJ(20n(lYJ&M@piL|s??kcH$hT`-YREj8X(t#@9IG8lCz;X zq^Sly4AO|g5+;8>pT%8+Mc<~G+H9fqXJmkn^~NV+s;?gF;2qFOprGQ>`th~+m9%XI zxXXUDG-!7mTK$T<;+Jkva~2W(-9L$ACdP2Q%D4+$@O5k_r_lt81SMu(MTIYFZ5ya( z=k>wY)m)R9cL06I>^SGaIBITS5mcfAC&I%1j+3i9crWomxu&j0WYneysEv_9dR9Tx zq{!-^W7WKILT$O1BU1!Vyo^a@;VggURD6I3BBp9c+Ggx#dph@f7Nq+ReVj6ckjUv}4r*P-dVrY(G|k7bq_! zQH2iNu|DewJT`Teh&9uc=feJD#iVlQ;-{hh&dv)WLjl-%1XXgeQ{~Cr zN7$01Q3E!}fk!OJ^kqqi+fYmSK=+USP2TC?Eo3+K#yeXDa%bV__FK2&g z0aP8E4kBbQSu1r@6sQr#ms%<**R$F13va!xgk*^wpX(@rbYl<8!rHoFif-R#R9uo2 zAkTU~eWK(>j`MM05XDRzp41&`8mbhBym6>NcUX;DAfFuwn!R4gi0ef7M)gr!jql3a zZQhwJs}I>TdWA}&mQq6&^^4)_r`3}lN^&-p&*?1{bMZPUQ=7MfFLE8J$q_VYtCo%q z+&ist@hv*+?IUgNW?!}mJcTdYKKRAW>27dX!mir7ZCr=sB6J!m258+od!ePFb^}%| zX+)IRcig5n*DP>?W?LVrE~z9DXbyI*e_MLx5mn9jTm%xUMp~vNwboDuPOe0er`v91?NG-XVHICzeQwV zpY)HB2ImodklV6GRuLoV&O!v<55$&FG9?BIK22fIwtGi2jU-Yx>Vs&+O)e2Vi zZ726tjsmhqqLuNTB{KwXo#sri;w)xek6Mal7-WHrjJTBUP?(_-ap{o19PF^*ZzmPY zNWcGR{6HUyeU4OerIIN#3Yk|q82os&se+Qf#}Q7El7Mviwx~ZnIB`pE8Mx>iXy~eHHG22{^WS6goZ9r5jqocKi|zVZ+z-G zS8)Nxns&Pbo>P(e;*K>$g#E#(1HsfnDu?cg>kExHI${TMarloq4()L6+Cl0Gszy47 zf$>G7+VM*ClQd9c2WleTC!23olGvA|P2I`F55v~YP-LqAdzQ*h^e7Ie++SpX(YCd< z!SLJj1~0wvA8yfiblMY8bT%b;361BN4Ghu`EswP?g2Q%!xW2ba^B$c(5@2ewmXh0(lai?`C5i13zS8c-W^|3#Nr&6+ z<%BSl14x;QBQX;8Ba-&`(i7_0#BU#?og)_*bVw||J!Q}gwY5{Io=}(NS|ssc;H5#4 zgtr}OZE4B6G)*wCveHl5u<%K8vLY8d>m$}sjg5M%sVsuB(^~YTCOBym0dlFCVTO z@5R?zwG?j}F1_?%(R>^G_LJ)u5&)57e_RsmrXF%+fAu<=+m@=CI~HaQH%HHHG ze0_bHv2BO7QyltTh8Dm(e>BwTe2~-=6Ta8bu-Qe+e21UYKqE95A|EkDmQm|ok%7;b ztrFu=+Bk*6ErrqU_@15~k_3~bvY{M{ursH^o*qCAIS{>EKi{|02qb%PIy^>j0W~_?&iZa!WAx>DDLXsmDE5|9t5(@?IW*SntEvuo^rVGcO1UISy2@ z-BuM!Z8pccpkf2lRMPOU{ zNq(6RsSbP)5#UM>4md_f`T6;Al$z91+cdqF&@3s9WfYA9TCl`&z2)ZSVAweQ|84k2 zIjU6PJj&{m>F|^JCt=lu?rMD6Q@uA2QbLq-*X#RQa;lu;;2?3O!i2Lgk3hhQ-RkOcIu%~qyXR?wYNn^DEEq}lCNI!WuP z!zym@DExCJT6c-$6Fo;H62|(h2>_vJ$(b)9p7a3;=L@`lV7)Q$~WG zI~vJ0e#ERdA(I%7< z`?x9zq3JXyUXkcU6xTfrbps`2-QRP6uvWvagJ;sE8w_1wNFZfGCJ8GE>HR zbxN~#dufU1fh(xlH$Yfi*=7jeZo7Z`nOOEj(TF8mx+OPn9EG#%7an6&lry3|&+1~; z4*p=ep`!1y^L=a7@#)yn{z8qWT-;`F-8+x^FbDzJX2Y948d}l$a3M&5lg+oMCV6U^ zF?G;q*7ikzw0g$KpQJ%Zfg*evv-0OH>smC)S2i|4EEiwV02%|mp_gk(Ux8pJWy`Kr z3NJe0&8c2*$;2L$9!!zo=MS$5jk2UCS8tH!w14DbiPk6_dJ$NkWMB zyat^gsoP_qSE7fKfHd1s(&FycJ0xLn0Fo$gGX&8=Jdv;H_lu-2vRnBUvUHuZ6XjPA zq1$;G2(kk61U^=%@}rW$6P_R-Qg-k;f{v%~+xX+hM+Ek3!k1Z{`dMkzXp#bNQBM-3 z4I4JZ+Bza*rZ%DO+K?-~_&#|fiT zN~dj!XLdK(w_*i{BcXyF8cZFv@BT@hVgLqufn|BJn9N&|o4Uz%mMpX3oo{gLQ!BjR zEX{bK791-WydnoaHrO=|3XX?08%020;6SmH1%EuYI*5XY8#+itkbIKH5Io_4r38av zazpxekO$FdcyWLS&wxLTm2>KH;h+8y+(1(jnyzvoQHs(OoSc}L*ct6VypqIp7`4b! zPTTU(Pm$4cO5H)Kr+QjS_y=NI_8^m|`inkwCjfVlI#XS@8Oi|bd#%6Q7+P3ak#Kc4 zw;5T;y8KRmX9Kd`whZseu4e*6EihV=tD1H?SX*0D<%a4ULuoTePm^|?oHV&8QanL& zpNW|hC_w}z(iSUEEoo@5@j&`|3c@^M0I3gX*HacExWaJP*M6vy?@;$`S0m z(+WDR2X|M8o}=3Wzz+9!Dvx&3O&syn=ch!S{r0BZt~pc=H?UZ}dMwQ(1T7~rSezc) zf4LFydnXH}sKA7`e^B6lcm|hl^ivj(iyf+%_V$A=y3V-BLJl)92SV3_ZkgY%wZ=_Z z$sbZ^64sxD^!<#fX|C?jd)-?9kw$;IRAtO?3{VxRMhJrw$kXY#avl|$S_&b%b=a&i zzQ>>cZZn#Tgq6&I6t4ZgRsFsUP3I@wOGNs2Qi4ZPF9KAFZ~31IVEHKq!P(J81YhB9 zX$bweU*!Qq4moafPYkBOw@(*kp^ub`g6FhO3s;0{lClsEB&5qtaM_D+hwkH|{Gfy8 ztuaOQ5F+Lvsf|X#gCI+MYi!xtbKl=NX(xgKK=m+(6tWJan!~hHvh7KyOTC~Z_@cS) z@mEr$kW?e|>FsNrq9QHxY?a&v{b+I28on8c$5I0j-kLKWy# zVc|W?=E76RbV`zJ%?Vlqy%7i}SV?llBLAsXkm9nx&2U~%In3dZBX{Wb1^SxQ3~I4u z5LFawe=nU9Ru0}ph%{$sh&G_}xvE|<9PXa(U#X`on65j~BMe)TsdLXdsYZyzfTJn* zD_%YZ_U#w)Jw=EtEsU}Y?4=p{SPuIOO6l#vv*T)SnmU|FwN9EFMEAp>`HUzRqw-g4 zlm|<4Nikc8%BJJ%n;uvq=$SwIT331@eTqovKWk6-G>~32+`gi@Jn@Cd0Zt|I?QqrTaMQH5%&-GNCqdx!LpiVO*oMxu z>_LDEY&{I3*Q)~3jjQ`zcu<`}=8!Zqo%nXrajMZ}9i8%k&Hk3`{qvIM1coBKQFch-@)S7qUL& zFHc?>oxM{_q^zwAaE;X0f&#U!d2hMZ1irG!zusRXSvESFoDMj)k?F%d^g>n;nU?XZ z*qi?x`45s^KLW2v{GKaS*>8#YmSXNHAiyQSMsh!Sh?_&Erw_uTpn=+@dQ(S-NsKtz zXKF0B(&54&5`*Xfq;9bT<;0myk$CkdXu|O-6Xdy7hSe>Oy@^%X^Qu~h5#y+t`fzTABiiTgBY%1Z-t-soDh_FaF0}$N zsAI^yP(y!?{O?^T)RB9Nh(s`>OD+cblG`0k;VMAN{-S$}p@5fX6Rc-alavqcIRV`} z%ps|Gl41Veqiiy!Eq%ks(^Z0<%kE*^JFv?qyBoYM! z^e|aB7Rs$9&gb>4*J(TgKaa{eaPhAOY|sfxZKkgHA-9?_FOTpCof9UDWZ^iqt85m} z<5h6B6ykgSMBOr3z@CU^7^ zQ)Dl#t*uNJF><7-rVnt9Iqh~HBn_$k-l;@DyEBlcW>Z#!t6^p<%%93D8I`~@xtd+g&8RB%Y@&0ylNl2uU`1ATzE<2#EH1H&K*ZK|i@~{03?7)uT9-NIu(6R!G6mhxPS=kWb^WL4MBtI5Nk{hPW08lMjO*rF z!9D{tL#qO-vTEmq+iXLFj7G=SeY|4A(;niA5}e+P>+DCB%$ZkcQwuu8Ml7jfvQYJ0 zh|4tSaooDezN5s15s|{6J3cs+>(+h03Z>ofU@sGt5C2O-k=wb?-j$cEk%Ws(uwqXL3oK{rTgb*%qN-mDq{cSrzCxBl$~yNk-+Io=~rfo(o zG_YClt8N=axkEv>(qPL$JcwuJ2O1Uq7DZkV;Qj7Xl*SQXX={E}mdQdbjVDUJ_2n)} zP`>b&2#~q7oWCm5cKq#VHCcN)K`M!`OQx zUIp^`1hI~U*9&-3vt#?FefTe04SvfB;Y$`A)8vg0+Az%6$}Kz^tKMXX+;RB9q)R;U z)Fw6ity`SOnzsIzCqVB|R$AKAw*l|um1p;Xyp#ZXWKPhV*beanwTg}Ln)ZoTAl!$p z@I%%=+s;&1fX$ULDZl@&P5b#T^$z436>xe~e`@b(Ff@&j!BszFyr=y7?au|3r;Tyv zbmDM-X`10gNd>M29i#7urjmq+Cviv%@P=N>GFi0YU#hUxKU%6UzUGNnei-Cd$6OGY zz*PpM1(>@_&s)e_36nqFPTz;<#U*Rp7F@|S9Rlm7a$D3Q+7vyh)bj7OS{-uA5pY{omG|A*;{gny-7E&!LNlt72 z*SF?f`Ilg7pTO|6B=pkk*}Xf{C3pNOo&-a{-fZg@SaSE3ODHho_LbxRR)9gwsF(df zdteWe1%gHE{l>)dJflQU?UhTVoz|jzAFQjE`lZRq-{#C8E|m0-^lctOG660|fw`si zvmVEPtS7((7GP{@8txZ5xEY%zz#ETja{4;B(w8wYtzE2+0}TvgIP zy%M)pRJ@kQWuoh3c4>5Y4{_RXHdeH#RBMAU12CcK@|0fzCI+T=W22WWC_gssm+;lZ zs$ps+a*{*jXfDXi`lXJmfZyAfxW(GuF1Kp$!Y_RCT+g*n7MWTO?%-mzI7~D(CX?jB z*nluQ>lz$L8Il@)8#Hh`eRhH*uWYVlpn0Y+Rm-t?kJp#rYN1lrPE zRmep@>!1H6{|hedw6LCS1HoxZMK3>3u9C0TI=xaz=t~E3Bef!s9lZg$<4*Cyxq8eF ziI!c}mJnj{=vJ}dQsLiC`S|ATq5nrfGlM~cp-R?3 zIYo}kehGOuQ zYvFPs?pxk0L=_i>Ew)T9wbPB?!(YeS#ru#BCpGRhZR#wC*6Tiqtnh!S=euziUQEJ? zUP|6=8+_FPOof+R6Q%^k<(qNs$2*(`aR^3%&t@x5zk7|#JR~i@k{H2L1-two+?2o< z@c)C(iT^Ew+&){8@3S)u=M!`y?hG^nk5}bSDn?MiKabM+bWYmY+2s`$9@gd`TIPJ2 zFE%iC6GuFAW+&^?wf4e&c2v_gLmn*3qy|7<&cT;d9I6$+mTH8o+EbGD8n{O3-g(in ze4na``K3GN@OgYMspXWUxXofXZu+zD0N18BGP&`XZz58^T!|g`Kz|Se)YS@a*3Uj zw)4;4bnn`iN2kxv*|ucN%VST@&)yy6AyJsK+T7Xs&)%XujnW(;J&7dU3`U&!J^MDf zxa_xYeWf7$nLUXk0K(`ZR$M+lZ+pnZF@LVb)d!FgUSOQTMHIL&?E23uP+%r?+9;I9 z8hbNVtXj1f*9u&v@45(CWNn%%tZ)RW!5ZP2J2|)-4|CZYxs7V3^yk04;|~r;hwmOc z$l(6pN&u6dX+;Y2om~vBwa_$m7<8FTBLsQ}@Mbe}^KS4SR>6+=4~QwB;v%r4s533t zVDTK+m>Ja7)n$A;yEPQow_K4({P5d52&A{_GR~bBIx+71^^S#?j?aq}9(Q(eY1FCf zm`P%iGVZ!?X%{iZ9o7%loVetmr0W@l%LokMi_GH3j*6%-jK}b8v?!6@{$~nQUIUA- z&T5&rcW*D)``ZNn%?~rO^~OFNlF$$d=nBC zwRRfT1XnlNfpuDC3M}|2_tu)RKf#-&Z#S(OCr=}-*l}Lq${ncx8t?27-mjzMw^mBZ zgT=~gjh!=R4!hssQ*N&MxpR@_CUFa=U&ha*akJc*BQp7?mReft?VurZkL-@n)Xk&ILz9p1_o~032Zk zbhW3B?qUt}wu?lln3#k;33IMpdTkH^?_0y z;b#vYzQ+hkNGOep#j>Tm7vrv9O3500@nilUNiNxt*DzVQI_dY9DhCd*7#TG`Kdzp- zyL2p{OW)k!d9Ho*gTS{GX0oOTv>81aZx2 z-&@YbSgfVAhYr`(Zfta*FQQN2i!V~JH*ex}mom{iJnd-1$L+OOuU_@-d2#g#Ev0GX zwV78S>b_!Xbpv16)xS%zOS`mo?OFyCw8*C3`?v_#lTlY&d!%qKAK$yUBkwuZ_D+_w z1&UkF3q*Lme7P>Z!i1TJ%c1xSOJ1IFXEfq+i$FKQ24FsW%0o#QhbGop?^SrzRw#hm;5G$mlzUKXR zyP$M=rqPlqeEaKu+8ol-G8$>=9n5w-Ma$bO;yiRt8)a$jQna#c0Ua^5xEJ>ebiJ7( zd0#|7BI>_CUu`ghcURJD)+n3&zU$1HAXE&6B0_?Km`JtrQJi>s>CyyW@S_R1`o$8p ziHpM%GIr92|A;i_?v2H^1){}dBEPwf;|S_G{^E=Y`3?MuSn=Qci+}l&3ctaMV(rdJT|DAi1y25W~j^T6J zSbPDq{&TyJO5oJDzYQFXIOeJm_w3oS)5&0W-gk(7cYpisO84(BHPchoVZoxD*PZ*+ zXfZ*7?iFxZ(&vfFQ`)&lVPQE#_Xd6Ct*us!qP&&G#%-L;JlM!XQ9ql_a&mIJR8@<> zUIszhZe6k~ASOl)emfSV+hd%7X(CEmxSSvg>1Tp;Y)_9BV&wslAqBYf=i%MEv+3us zu>4r#vc;)=!q?tw{jKSLOF`I8OiXN@ap^j8of#P!|4(gK8q`#pMGISOYh0#VaTyFK zEwn6xV6()?rh|mw2nxz-Py~g5AYuZtGqeh{LLd;u$l_=OlqCXgfPy1xXm*eU6aklD zWS<5CLO|y{oT{E5Q}btPs!~N2RY|?~z3;pC+E_3Lp57!OYy2u5?}uzp1TdbS zmUdkt@sMLfz7s zwcQzIS$9P?!lMUTfsP>|_uzI=fGXD?=Fny{W0nQt#aL1;BIliC;EK$xohw{-zI^e0b7~|C6i?}HPODhFdV+_s zl?y~L34mMD5%H9K?m5VuRbwPXI%`ClmJ(w_!?loVg?E-D!>mIR?!sbEgp3sO*w;r3 zFmdRZMZdMPO>t>s)Gl>3J*oQhI`7O?x&e$~{} zTkzpSVdhD_8VMrSVa2 zB@Z7SBy2_gQgvp?rEyniI+x2po!1>*@lyu)Y$`6B-ZT2CU5c}m?~@(&qql{Jz!@xc z5;n|&i|q9l9xzb)K1p;dLcOiL?1(uf9Q2D(^z5greRymYkV*e+p17Sgus;eSy&YXK z@HyatrR>}!EOLPY^}F8raFQ9T8a1)+=+UDL>>`-jrg^UYK<#1=O}RTbTzm3lC!{fc zP}0qym)bWa?;D?+0)SZyqa5N19L1 zMTeQH2J}|Z6%`dZd=#p_@P8K`TRRpzgtiRZiFvd?dVj!{?a*+1spzDq>u1jGaRf2^8;XDC}2#DbZhJB z8P}fxx2>1>6swOWtUZSh=EzSzIZWJK@2E*fXm3>%SPV83W=o4mc?YI!_@{iHMUnz= zkn3t|TR@4ozj~!z-77$nL2h@0-lSNafp5Kn9_P|M%x{Q6W@xDN#6&>0;ujb=(k=uy zmW#0950tV)c=(wzj3?G5ynRuFgRxkR=WfW)%~k&PkE%;M?Ok16^VoqgP_QcuV)-&y zKvWg!7N85xsb7b~l{L0q%SriMv=>^S=hVI6N$Dy0DWfN_HE8pPC;CqW%^9To=AKS~@tAY9np>Wm^A2 zlYdHjy7w%oQ3W~3Y{S9E%xz`6!=j_BT&ocBW1~JrntOrsJod=a-O#1PT4f&6{+76j zz^!(c;OcpVQaJV~u@%Ei17DNrnCDg)%mNXLnEO7#D=eDSQ!<3OXmevcg9R;wx(2T2w%45j)kNlem^y2#v^zenm2)R_w8w6=qod=BkPuzgnlD;T%7>@a zLR~g93E9jng+K;mg|iWEV+oFS+cdl_kX1PCedn(sNH%y0Itozu;qYA+AV(n$)DOz) z24x|Bqq6V_rf%vllLwhXyuzyCeP(8+>U|5hkdTm;Qk;IuKDpa6q;Yo zsx7?mSy@2L*+DyE_BK4OxW$iO_MidJp6kE4tkkB1h-K3$R zSR`dqSa@p%1Cl-rttCUFGl6LU8_tjR*@ejXcCw-h>JM3!zKzrqNGHvvJ!L-joik=H z-S6l5gNPaW>pUFWe2L3Fsv3jVkpU^ujF~AQ-U@I_IsohKFjKg>rm!A)2w;V*O;bkK zmz{0PB9ki61#8UdD0F>s^YwhF-|jTtdgV8yi?fB55mSK((33L<-x7(q^P733=b~p| zun#FDPn`iqI4e8c%31HyZs@iv)fo_yxS@&F-o#B$=R;Ox!x0zL3s8D07RjOU;zc|) zhZZ!eF8cGd4?26HYR#gaocA_e%VYm7(y$WSB~+ zz^3E(>#%wTjMDHPvt#{`6?6ba^z`q(0VD{k)4Qxer^YKLTZUGq3(MfJ)ni%(?%XhH zhKZP1D;=j%GF#;M{{+NaxA=U%za#x@nCzO>CXA9u(&65kmA9)M|-Xqy^U1K2or9bx+nXTx5?P{>zX@H#l zwkF;m2XF?eWDS*UHjk&gCqsV+m&;X?mmI;7vYg#W1c%ss;88L3h7CEeNG~Efu0R*T zWD-8MHa0|9AZ&oKuD&pKHz1unz6@4kj$^@4UnrAvAUyP3 z)Cyas&i?*p3<2nvnRT1m17*RR>hQ%a^ospJd9x25AS)x!CkyG|^F8I~z5o5z zw6wL`(D&XRby6t~9Yy&aKan;-xAojXcmyV+cfP7LxxTLrf|5pWFY0ycq&B$8a+>e5- zAC6Y-u;jmqX-a!c-z8&(h_&%;*#JU>hUv=}7+;bhg$u(U7?_-VXQg6#dM`$%T9Nk) z3kx?R0RXr3L-{3w9Pxm=dx1nENoab{q!{@$k%nKEQNgDcVl^6BTAG?IxUT{zwdrv4 z-O<&HStg){cFgPHooPxqv-Wm*j==a8$o3QHeKTNzsSL}j+7duTtF57_P4OaJAb8eX z0-ZpbiDb7IYio!L({(Mi0xY=LHN5;~JeN4|?=ICONXiIrJ||8rqC`Flb5?!VM~xF9s1{tt!Z|MbJ{UlcY=l5}pzx}*@lL)&F%RlM_X+&=-M CGx2T! literal 15016 zcmeHuc~lcww=as_iZX~pgEA@3BPy++5pck{4T`oC0@@CUC{Y=b5Fu(yzzIZA+SY)A z8gaq_J3uO+1O=f5F)C3u zz4zIt>ewpk5{v%B`M~~jNkNfuZZig*e)Y;iNu8!(;`3&XUs_Zpa%asXd68aoku2{RElA1I}L|Hy(kjcc{HD-$_iyF<*@!tu* zgY!LdzVD9jspEe_dJGr`(a`4Z0H@IoW+vx8WFAfDI!Cw+hWUKgWH-`M)*zVg)7+Wh zG>Ww7G6 zjDi9^U#_lOxuo~JG4zR7cJam!^o01K43~m3L9u{+I()Q=UPlw5H-mQY>hGZ>hP}P5 zP_7-EL0;=_>yHRE_bn%ol!^I>2xa~vfsI~BNktP!b-MLBUjJUlIlyRM2!q~RpX9zW z5vAj5m=NE2S)`7pTcDl>ygjq*Vbju!c+sqMJiuT#DtZMjN+Q-$Q!LSMQ%IUbJw}A; zMIlbMu!EtzMtwvya#Lyw;kfj<%}Q-&FAFqMUyxTjhB8tD9J&@nhr?}nnhejqAv?F z_#{uinGSL2qx3M|uylnI!|B8EMt0aRjU#_As2f?s4b?^fbFHXF!QkWx$VK z_eT;zIh&&jd`9--nT6)4NXliMBT#V$4|$oRQ3g!xcQKqak2!P?L*!UagC>PCib+1P zwil@{URh6ym?NSA=uHJxO7eaCGiAQ3FUi#(Mu-HB+w9&dD=1UX9>|NksVkMZ034jMxcl2qfW0OR5)%6 z7r$FYVibRZl8y|;A zpZ%^P@j#|QcFfzP1?e=%k9mW%@D6ZbUL`GnLx6;NnY2(j4PxEPAuT+|7CHszuHngT zv`DEC6_U*90~U%NkQUktm5Qh7tpQO}lb;%b-iAYWI}g2N$#YlzWT+FHY+=oYRq%?6 z*ax=ylS2_8eV$k+Y~MWogdwd@G)M1#3l|%a`r5+LU00x#)B!uD>w~LbEF!%~Gk)xs z+h%Ciq*?UF!C!t!nZun3p>@IwMgrH?Tvo!&U310PsP+9(y4*m?#$yo6_G5DI@;(LbbZs1s#X|+!m{%F*AfIKlq*~s4ZXCklT^+a650-q2((n zi$)~$L)Ep<5g`@*b)k!FsqlwH8q3_HN_!w*k~#FUub7$?=TNB5%+P6-UnUl!e1)gB z%;~ACe`9#5GQjhZss)wHtUF6NNlWB_Z&(m-cMo5?NKiW`_0_47#;+0?O&UGd14{FVo#+}JG6G}QBe_DtlTbDP#Wy>Ecof!t~TYlMEja6pYy$HP6thN zHf>ZNXtO9U`wE=e(x5ZKFjREygS6i$5a^A47?r^RBAEMLiPY2dYomEIOZm@Ic4%DU53)hr$O>sg~+BsXY3-f`LMvb`{~7 z;tL}6wXC!>m9Tp_yfD|~S`Q9331+CUApAs7-fMX_qVfGuHYJlne{2`$m{I>eu>1wL zwAaL$EklB<$3SVGw~%h9JaLsEt$85EaJGmwDLsYd%TcZUAkDPJOkW)r?U|@re{k5- z`id4!cPA@rUJ>StXG++zp^h~{#2OS+ojC}5v@$M;-RtR@Vs;O&u0obe8aXD%7g@>@ zRce=ETJ4uF<9cFnDnzGkB8O2@ynmruD@pN_jjopz7pL`zdZ3JH+w|mG{eq3=`UcAR zzKR{TSHeGe60>qGqUjj(p1gCmgPrrYitQRn-eFW**uP8eI%q06Y>y`?wcTHvv`u+c zJ+l4;-NCUfWVM%qt(27J)l2m%?8@XH^-aN@Y6>KY$AphY{FM<_WP*4=aB55-%}6-d z-rNlx5v0N(oCc}yDJ7GLNdcBS3SZGe&f+D0wYf;p_e24A1JPQ{Fh?yX>t-bosqh7s zw?0bhiI&S-qzwI>MDt4}7+Ht}NyXX_Yk~|UM7!ybK3_C&5NT0{E+U&K1P5&4DRaLC zUi$^_AD4nG`?yv&^CR(4B*@I(OeZu{OiEiK<=vf&=og8Em+ly^bo7OXzPpo0B)o&R zaKvQw)YE*bt}^bDqdw>LIcg1mX}wz(`i5XQq)2&p>s~TLXHbF4&(kBwYw067s;%qt zA$3gmEl2uJLte7>WdxN9MhboX||u+Lmo4B4XPqW5#RIq995+seBVf1 zm^u1X8EUHc+KVR~)#281qaaJ)#;`ya>!GuD*vW93h;A_o;yE9cQ>SDh(jWDi?xv{) zYRZSsO81hsPP1)*NMNnsG27#Y!%yoRTF6^4yjVFLK4l0sN;Tk#?lO`?$#R6>*}){3 z`&qd`C-OrXp9``(<`embS&eA!Ln0&jtz`MheeK~N>bMHB{Hw*v&~GCWiPKsVc7prX z^yrHbV5|m2a~Be)y;Ay+wtG=uCzw`BF^(UrIgr;;72szPwdn{6~WF|I{75r4cL&PdEipTc)p@po6P+T}UD|9ckLV=iVp$t+wbDr0! zgeRcWjW96*=lp_X)ny|C4+AA5%2+Rrf;Z{1jTu_s>iE_u!lVO^szR7nO=K;vDux!? z)kRYj9I{TnsyT=TSZTEBGFiuz0V%2yHF@tZm4+Nu`7mYL?W%i5dNl@&2Hl@W4tXCW z<*heEy?912T~+KTR9B>RD86|b%|o(=z=Mf{==YiS_ksWSOAVrr`+no|Kl4`0<(^og zb8o!Ryo5Ap{z8vpEllR`;w|MO4r8#HlaeA7PA~N%-Q0C@%~5we%T?05`MD`45bkK- zAjXnqSbN+}<* z`7vb}{ChlZBLobJ|Cc|CtCX{hGG6_lOuqM)o7|gy}!Ak`9&DX8UslfPa>TOxfW#_RL&K1?E*xUg_rX5fqR-F zm4`-P(X-G%bO1mzet&~rhO;6ZJ@+FZu{V&!1K(^rhCjZKrv@@c6> z6;aCqYi2^GMM(xY?u%Oi9KCRc-qrHM7#j0%7X2t5FM)#Og<)_Q!LtyA(q^3Dv3nrw zY;bj)o7jv8Z1#k}C^r+i{M;OqS#8boIht_@dc3#tht?(}1R39_GI-ug7qQiQdGua@7X4O2_ zi+^1JG`8FSQ4iovnfyt`SXZcG9M1eJ#p6BDI2j3t4#n&3k+HWi?$5_?}Toyn-(LEGbJbMNd-IG@cS$E86by%X` z|K6BrFHQu4Zpo#;a9uaq!FgD^;t>k2i{*dmfoP8rkv&0#8E8ooYDx3McD{(Ps;SVA z(GyFLa7h?Z`FkT&f(W4)Dh~;I%mW*ZQMrwyA}&>SoWp#$+gL$kSqX%!I~Fv5Ou@6p z0+tyIxQYwtaq%l-1cPx|*<+wzG1#dj0hTmL59}{-qTn@*i&!DI&&CnhRPSgY2I~=yX{b=hkQ??PN_!atnorw9ol;9J>2PDh_k@OHSX&R$k&hWFdBbAo2sXmpS6E0DZ$*S%FB#YSI#dqU zW6@!W@>p<$RKsv#UB`GGk3jsx8dku|j9K+ZhYFyqmh_ZVZq2}i{%Fg^I%7UoXnnh) zqx`Sq0a2012fL4k=z-VbgQ8wqdn939-fo$E7$R)0mX_5^;ICQpHjkyh@9_EwDOkvm z$-bfWY9 z=~_8g+)^Ad>+{s)*@u?5P-Y%gsoSUiE!nY?_m%(2>F75{mzhwUN|ng&R;#9fq2SBd z8U0s*fJsnlSL8>1*-|sLLM9ix=GoiBoqHbKT0i8jBEm(b?cdVqloR&N`9Z3JeJ)=( z*fT*@|5xHCDxJ16DV`h#fx*;{U&xyMSzzSsRa=k8rqPSj5Ic!lAbWGiFQ=Lx#? z@|^q_k&O3OVl5S1!{qYW?E61MXweC0N#Nz#cGyT87#YfRKhz*qrKi*@+8+tMJTnAG zKhIV$WXFQ?G@r_!opniGXfp8;`g#y-x6$v<53=NaV+iXflV9}mB81`|ARD6YJfw}EFZn+H{$j;x~f@bNo zt{k^NpjFo~5$z|H&u;am2>DGsD1CPd!Y%&^ZgTPhcOy0QR;Mq&@SR@2`y+qV*q&#P zDxdkwJ<&Z68?FWM+>bcqJE5w;83%RT51lXmiHZq5lRr@9bV{v7-RIWOnEUO2uBf2y zHy0dLY4|Su{AJ}*`hi-v+6#{1-ac0wurW38ja0<)FMU(zuUW2$1}~zBs?`O#f3^56 ze&*`Y5TCjf8}WfuKxbqmhrAMKU0&vAfRMGqcLuQHb6-aDwz(O3y0u_y|Az1RNK5q8 z2G0$!^n^MNY&7r1h3cmD(By-O#!+m6zS!y6dO-B>f#0JwmyQuxa3-WmzaSUJOh=k1 zjKz_+B9E_;_FSisB0RW431@x&HAt2}5dAn?x))?4QjTzdX`bcLo{rjV#jKVg#k&dB z%u%a1-;&>d6PAV1=(7y)vp79ib+o%rGJb#;sA1Hc@>_?_FAGvO%GY%70^rf z2;+ytXPVcw3|T@d0UZ7q!LO93!^kDz&hyYW)cqxdhHDtC9vqFTNyZQ7eF1yY+Hg1{ zP4^qnZVgcd{0b%EA_>S@#uY8WsG|Lf86QRvNeWr*57~$QU8U@4g@W*a$@YZ&UA_Q+ z^d(w~S+S~yak_q>%FHLMdxK07Ai$%f;=R`lt*p?D1bY8(_6Nax84_*ionUf9l06by z;`Ox{t7Jr*fOUdP2nE-1_=)aYrYtyZC)coyq73_u(+_;{W&A7QKH)i@@ZKQY+ zNQ-*U10eDq5mb#pWRVkArT@zfUw_(G%w=N2Wmv-&jKkd1GG`Ma{#5=HGh-tNDx#MS z-oZ$Wd7rrE|4rxY44n(~ss!OnbnTmLK=^*`55l0$*sOnf5)eyj>t;eeW6KS#FgSkL z!{u07z|S|`px>7h=Nk<;TZ9wmn;2eP;HQ~T#aR3VqXEZ-|2xN{Mj6;6r`-4H!>Cte z)NSY*mdx-2+aaLa^Z3Dr^ZFLxwHf$X4;(MAfIAP~;Bkf{=|9KEV|phayC)un>;*SJ zRMHU=MCu~aarEDuKF#inO#yfn;}Q%9!+4-gX8b6FHox)uY9|fhsJe&)>wVCd55}ea zr}Og^_MIn6&ouV~dW)w=ZnB5pW8Mfdm{3ApQ5M4J8}Dk6vIE%cSo60&QoY~F_rCMU zLL5ev8?;L1!so%|o8>^9$KO;MyX&(o55_ph890|ySCob@#t}n?U|t#!MOp~u(To>p zK=#l0e5ALceI>Zt>_c97al!%{K!ocJvZl{~iU;8$KY>_UMNCINJXbm{$0s9T9>cnG z1zfUFo*ITW$8rIFXWSzU@X+(ZfI>@%DNi_1XvDu73ccipa`?FzKYVVL2l$=F=T@uf zdB0*KA$x(o;M@SATuY3dB6xmjt9CMih?Phn&laC#6=FjuVv_a0n2~|EPTa8hlR;Px zC3*r4+{qII&12YD?KnQ{%K_SCkS2j^&I7L!b36vGAB|2XH~aUqLcp6CB$iO3r)D8s z=YKJkbVk99ZY2I=J2ef23Xjl5D2`y12|$s=hxZL&FO2jye4ud@2CiK|`tfd|1eRV0 z?*{8j^a3>eYufNxP#8{9gpth0Nt(Hyp&+>I@j>Mi5K4WCQS3qB2@}$fQC75QBQWyC z>J#7%(bG7b4HS(fdXYR0?xq&dm}(<4I=Na4myaBNu+e~9gX9D~v?6_ga#Uvaifm2| zS_EzazkjY`7&3qVmgoQT@0BR)8DAiw&_Od~doRW7`6#%tNrng1J=({!i~<`mqdj1J zRxBmrCCFF~*HjCbQ$A$=4ucdhUT9Mqf>0gWL26tg7p@8zKySPZZX77f(b(2aIcxd~ zK~h^C<&s=g$eq(5eg5yh71MuG7ComR#$VG~RqU4rpRV2T^XU+W;#yNKdkQu+1@B3; ziR})ISn=b5JEly=WZ6TFaq z05l)dp$EPH5B_%qvW(HS%PnS~`N1^zZPMI1t3;bPD!DCub{oSd0Xa8xv;V9Dt(Fno zx!BRA@W|oKO-mgeZfwPD_3WUj-$u9;o&k&V935|L!4~gMJ>-)jl4qS-zqNAS-3a?c zt8ne3|L#Y5ExqTsn40ISIlc$57=|)9n6y8eazVjcaUJS6y)6Fx;l4rw7KoH_#^l^FmQLLrfLsb zpzUl+EI2Co+9^rw?(FKm-&GN#TJ6YEYdq@HWUR3qRqTG|Y)g^gQ*xhpAnHSZn)YB5 zz3q$2F+yFD&wgEzxLMa$RWEG6+J^&vw}r2b8#8TlKtgTD3JWk@P_fHtI`J@=<%@ZB`_x}pu`11-G)JcTi&vbgP3;D`oa~XDMQt4fo-W?QK7D+BIIentCtIH*qcML1WGaYKxSxm)(2} zz!f4jCE$W|2lApgM*ByX+}{t*ANp3tj;bhB=1&GnZ)Fsu9)tQ!brELA-O_ZG?cN*^ z6>PO$-Inqu2Wo%ic(-^?reOO$6y}->Jd|zs^Y!TJ_?t(2x{sUg|I1B&) diff --git a/docs/articles/opt-common-subexpr_files/header-attrs-2.5/header-attrs.js b/docs/articles/opt-common-subexpr_files/header-attrs-2.5/header-attrs.js new file mode 100644 index 0000000..dd57d92 --- /dev/null +++ b/docs/articles/opt-common-subexpr_files/header-attrs-2.5/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/opt-cond-thread.html b/docs/articles/opt-cond-thread.html new file mode 100644 index 0000000..eba9b93 --- /dev/null +++ b/docs/articles/opt-cond-thread.html @@ -0,0 +1,256 @@ + + + + + + + +Conditional Threading Optimizer • rco + + + + + + + + + + +
    +
    + + + + +
    +
    +
    +
    +

    Conditional Threading Optimizer

    + + + Source: vignettes/opt-cond-thread.Rmd + + +
    + + + +
    +

    +Conditional Threading Optimizer

    +
    +

    +Background

    +

    Conditional Threading is an optimization that searches for separate blocks of IF statements but with exactly the same conditions. When several conditionals present the same boolean condition, merging their inside actions would avoid computing the condition several times. Also, in the cases where a second IF statement has been used with the exact negation of the condition of the first IF, the second IF could then be incorporated in an ELSE statement.

    +
    +
    +

    +Example

    +

    Consider the following example:

    +
    +code <- paste(
+  "n_evens <- 0",
+  "n_odds <- 0",
+  "odds_sum <- 0",
+  "evens_sum <- 0",
+  "",
+  "for (i in sample(10000, replace = TRUE)) {",
+  "  if (i %% 2 == 1) {",
+  "    n_odds <- n_odds + 1",
+  "  }",
+  "  if (i %% 2 == 1) {",
+  "    odds_sum <- odds_sum + i",
+  "  }",
+  "  if (!(i %% 2 == 1)) {",
+  "    n_evens <- n_evens + 1",
+  "  }",
+  "  if (!(i %% 2 == 1)) {",
+  "    evens_sum <- evens_sum + i",
+  "  }",
+  "}",
+  sep = "\n"
+)
+cat(code)
    +
    ## n_evens <- 0
+## n_odds <- 0
+## odds_sum <- 0
+## evens_sum <- 0
+## 
+## for (i in sample(10000, replace = TRUE)) {
+##   if (i %% 2 == 1) {
+##     n_odds <- n_odds + 1
+##   }
+##   if (i %% 2 == 1) {
+##     odds_sum <- odds_sum + i
+##   }
+##   if (!(i %% 2 == 1)) {
+##     n_evens <- n_evens + 1
+##   }
+##   if (!(i %% 2 == 1)) {
+##     evens_sum <- evens_sum + i
+##   }
+## }
    +

    Then, the automatically optimized code would be:

    +
    +opt_code <- opt_cond_thread(list(code))
+cat(opt_code$codes[[1]])
    +
    ## n_evens <- 0
+## n_odds <- 0
+## odds_sum <- 0
+## evens_sum <- 0
+## for (i in sample(10000, replace = TRUE)) {
+## if(i %% 2 == 1) {
+## n_odds <- n_odds + 1
+##  odds_sum <- odds_sum + i
+## }else {
+##  n_evens <- n_evens + 1
+##  evens_sum <- evens_sum + i
+## }}
    +

    And if we measure the execution time of each one, and the speed-up:

    +
    +bmark_res <- microbenchmark({
+  eval(parse(text = code))
+}, {
+  eval(parse(text = opt_code))
+})
+autoplot(bmark_res)
    +

    +
    +speed_up(bmark_res)
    +
    ##            Min.  1st Qu.   Median     Mean  3rd Qu.     Max.
+## Expr_2 212.6976 206.8032 195.1441 201.1962 210.8124 190.4774
    +
    +
    +

    +Implementation

    +

    The opt-cond-thread optimizer simply looks for IF statements in the code snippet given as the input and if it finds an IF statement, it checks the immediate next block of code. If it is another IF statement, the optimizer merges the two IF blocks if the IF conditions were same or converts the second IF statement to ELSE if the conditions were negation of each other.

    +
    +
    +
    + + + +
    + + + +
    +

    Developed by Juan Cruz Rodriguez.

    +
    + +
    +

    Site built with pkgdown 1.6.1.

    +
    + +
    +
    + + + + + + diff --git a/docs/articles/opt-cond-thread_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/opt-cond-thread_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/opt-cond-thread_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/opt-cond-thread_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/opt-cond-thread_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/opt-cond-thread_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/opt-cond-thread_files/figure-html/unnamed-chunk-4-1.png b/docs/articles/opt-cond-thread_files/figure-html/unnamed-chunk-4-1.png new file mode 100644 index 0000000000000000000000000000000000000000..1f4e38d8c0c71af6720761f315022505a073e550 GIT binary patch literal 36788 zcmeFad036>`!?QYw{2zk*oMq)ETNHPh*UOVr2$Q(!O%nsl?Iw*V-sRE4=OZB8Z=5N zu}LK*Dh(>MN}^F}HU7@)Sf1Pz{ z)t@tNoIf@pdYangU2{VQ&aR%e;qYPqfp3Ac4O8_?7Mx=(+^8vYaW3oeoIkc5*qU;_ zcjQ{tN#O+_-M`li4i2~FRvDJ|7>QaZU6;xnjFa?Nk7kCn3u~=N^S7_R~ZsQYn?S3o zH$`o1=UMgWUADQr=D8k2UEVTIzD=#M5iiFt3A?(!{r;`-ubJ!F>UWOuu6X-4$XhOV zZD0MVqr7|V@3Ji;+`boP<_&+P>d$Ycz6nBpdcC<8}nJl|p z5qDE!gRPjk6s~ln1}OZpxye7*5@*AQaK|IX=lGs%}H?8=an>h*8&Tc6|r2-3l>? z>*7O=GY%(2pJJwolhpW474T5m6opkct=!mJahMu0@6+CJ8 zop;ZkJqvTHci~~g`iJ%LEze4gd{OBGc-vW@Zh9B>EV{CG;b41q$8YoY>goBv)><#q zFi@GvVRht*b;NoyNDk0j^+ z!se`La9!zCcf7#P@@ITw-%flZde^c*laq&B{J)x<`s#0!`OY!xWp#{ZXqanvA(ux3 z?5>Bg4L|RLjzrUo#tW|9B}X z7?2gWxilg|W$686E>GLN+xN|zH+PeJ#7}*>$L*5ItmE){T9T6O$Kumzjv^*jTv~iD z!W37sQ@Wnbu`J!tXLO^b!TICm)RNV;tUdQHVnuZr-)I?!FIbgK%PG5960%Ba%{vDj z7iHJ}PF;hfesR(B;_l8auEjE$QTOg0KWj=Ck6g8@^A9P{LvPV5zE7e z4~Kjk82pl|?c&?zxm8)@-3OoV zusN|2?(49^wR|n#?X@(}gAB+c*wwl`>DxO_Y)@4nNhtXTHz8^%<^D*3dNru6T(VUcej6P*U!1 zJo6ek^Kr_Nv{zeNU&m^5%s#(~<28&9H#|xmrX<-~grVcDPKmVlFjd&-O?}c7{H_p#jM|3)Uh^{B;i<{r4 zhopJrL~Zq%U+pnEEUJ}OW@w$@uerLK#g?~kev#ApSX6ik;gj1+Sj_ndrdxdG8;6S@ z|9cAU^5dtK%>#`cm-&dh_rD+GOPAJ_z)<&1U83~@(KLwFib?wZ&Z)s`s?6v04LmIu zCHEocS_S%~^kRH+$Ha*fz3;RWUub z5Fhs}y4iSk#%FAok~eWWzO09*`d-pYJ$AFSuTU@N$q9E$CBe*EY74CXED7Qr)}9vNy;X<o(B^z=tCBl2ZnEs!S08%6 zRIdy@lEUVgHNObv_2rEk3&wB*V`#p-rv5qNN&qYF&YhydB@toRKwdg-_p}o~xFH9L zYS0_ou?LIB3#-HEWNUO>Qj#`j!n}0~b8OOf=EvpM#bzE_>aQ*wLyPA2T#RGmvrDsl zEyGk&M2Mh4mh(=h>GSzz%Z=vo8TF_%Y!uzmMn zdXQD?m-`>DaPQg7#)X9fE@7TM&z(DGbSz&uE+IjaGhw#E(oN?kEL&SqUcLYPyrPfe zr%Fr{qu3egglx2wkQ>0LoCRu8lX@n+a zkEE8AC+y91?aUwN^1a$}rRB@8yD3GIY6Ub+RvFWBl7FL3-2HqJ@1yu=(5#gfuFX8` z$7bB&MN^@#JvLM9-b~wB%lE1NGI6#aBDb_nt?bAx9u_0_qEzDZCzDl-+>ew+Zk(lP z*R(6}?-I?>)i&~oCUq_?*AH1)J^$#V5`NvWChpU2+)}6{z)MU&*`=qMd(&IiFv@dy zH$o8Bko>9dUoevHkK(SK8{2aS+X*KSdRX1y+H)kOP+|`v?rS`Xf`Wn>5_hnYi{1Q_ zpKiRVsPOlhCvUdxrBcs}G=k#w4MiDGCarIJxxPaqgkLPf1RZ(UTR96Ws<|XI5O8hv@IbHBx*52o zYHC#D+3At@?yYDEKV_R$_q9BFyKGhy=SW7i#de1w%Jg`%YXEnSGT+6vD^(eF1xN^( ztla;*O>Kr|Y4`>j&V;#Zm%n>(VsG%#*kiRv^W2{Kh)0;cUp-NTicHT^EdS|^=jPZx z$1?GiHoATa@W$e`<{IM|`xBqvEUP_|s)<{MPM>w}<`DpqS*s4~a+?2#1xS{7}+V$^IGcvgR}P5P0Xu4ilxQXHkgWrAIYr{nH0 zFbEWt%Sm=UlZNP}Lg_U>&hXsVjM_w{@U_PeoMzL-L*#_zq|b*Q%~jNi)zakL#m9I_ zqXS7^Q`6_J|B}4PW$5iepXvMnFiWTNvHZ!F>rUN)c|*}k;wuhpUgz*>$G)2Z(LmLE zEU%y7uto>-MnkbfriJ~At4H#7;w!TeGicgbv#_#2Z5> z@2xU5oiC0R9~T$5Y`JD}Cbo%RfgmBE{tS40Y1T@4+q(NKcC6NQyT;SwW`y15wl(B= ztaJPl#3D$$62-9BR0>uK;vy!yMKbG8TF;&{$CtI~{3Ojx+jk;bZ@N!>DqAr2mkHZA z(z}y?*H-448CnkPV9JSl@Sx`NJBe?FHnWxMM`zL}@cVl!4n8xh z&v9XM5HgloCuUjXUoiB^KQk^CVTH|M`$|OT=H?0>>I{agio`wwI1|`HSYx?nTWXPH zycvO5&Wr2o*HDfZ+SDEWl=l$aW ztmkpNkp6}d&Zh^(oTx$0*k$~%b896~X9yAt|2<68&KEWPfMr;E&}TyctFp&|$@xkGN-nKv%RUX}S_NCMLE5}*Hr2jVn)MZIC-XWKkV>{#HChI$L>|KVQFY=id z#3Ut`9i5NGkSVh|atDigfYW0iDKJB|nT@S@G?W5NWCa$km@~G|O4JH0c7(E9Ikqk? zXS?C#x1~p4FGpNe8vgd0pF5XgDds*3T5+)GBLGHiNkzzMW2blNN`Nru1*uyJuW5_1 zv9U`;)#uNjC(M$MYTV2c%{*lFI!2RowpIZFSMV5Xu>dY%M$8c)*&IB`YP#jxDQwC( zN3y?WusJm~HUC|NPqASR)B$U8y9m7O`%U?)Z+sl$eLf#PP1~L?8gcOZmugyQzATU? zFTd9w1p-n&H25`ByXEC|gqcGdkirCSeiMBjwHC6>HjZn{`UW;ow20Q%?HQ_a&7ku3b+B`*xnN+>Tdw;^TY7N4vziJj|Q^K0-{MG0}~} zlJ%&_eYLKyTg*nj3;b`iJG9jK^hE<59p8uFYoyraU<~%!W<~Qiq9!IH?blIW#Axv6 z&u3N2DSf{PSryO!tyF}UnAH@@U0)FW*_;Ot9@L=-g{*&#NIiH?!5M^G8Q{Lb^mZ(A3NUOGoB$Edo6=AT3eh058N>Bpj(6XW=68dcb zs5*GX!7V6&({p&up>gr?*5z$fHzsT+lI89;1lJcR2|}QymYT zz!~Kslq%oaa=B=6pb-ZMTBb*_?ek|lEc6yR58ig9wH@v`_AM`~k-AclfSTtt^lt%F zD=Gz-OM}XKe{POl!8|9F<|{fL_L*$CH1it0j&XC3JzAJVRq5{8;Fn}7JKsM^1%6+` zCe(gy(N+sY_+p2|*zl8*1#|5+M!vkt)D777ab34p^pfy<8?If>j!KQOf8MU^>D+=cN>gn$>zwojmRy$&-%g4jWOr;TE zK?+y6!J!PjTSCMnQf)78ZvXzQSd81?cVy%U59K9c$NoTt;^FrGI91HDc|+Y)`V(5@ zb=5yS)q*wjLe9O*TkU(5@h(8C5^axx*F$(*85KtW$d;61+3j(mX7v}n)P?SZ$%zDGx?j|+; z3di<3MJg^Pp(+O!w#`~Oan5S#dFvdW-`;tPBW+Yzduyu?Ex2MgTih_RcSq-#gvWqd z{;zIrZ9x_+j!()BVFfE3+9jbG^6hK^0+}V2U=h6|@ZWA{EoQHCuu?|FI#fD3q|IqU z94w(2gE@0KYeExYm38EmP}*!Tn1zUkfk=wqxA&ENMOysc;4xZqp}?^WIPP2s$%q-BOge#UQAzt*LijB<}WY zC+5rlTl>59dym{x8|dqZn1uS09fMLs8ELBc%w**tgqfiaVtLO|qz8aGTT|GJckIc8 zjc9BybL_to>}Nv&kkIkabFd`R85U-6t>#p2uwi* z2b+U(uIJ@?IZ$k}Yl+4zGlC2d5)4?lprQZS8ZeK!gSJ#Gf+d9OGdn~>!y!QLsgQXNAHSj z)Wro(o)5iFkfNO8Pk}zxAdZySfm$#9JfVf)P?dX%LSEf57t~R|MolS z_AxOr5-nl&T6f~bMt;p}OU&*DaGSxuf)}nZ#s(FQ;z<9;;vR3=2TPOkMoyk4Z0osj zsq;^gZ-^lSR9&*?zIX=_*1)>S{HyDLrPjTTz#Mj>a?c)QCSMj%Xep9iAaJ^uFFl~9 zTE6Ixe58H5d||OD;Au9e(054-YI`aE1nbfeZJ$hB^8#E$aZjynwhZ1Le}Va8_R@qv zRd3ffb?arZZES4viRkf+xrvpx;DO^;Lk9*_Uicr;o6=?0J-YY}Z&R5#-N4+$qYb$< zk5gHPPh!)QfI>-9>iT=COM6xek!!>>R)f_^MHz~qtE{$_1}CI{b%tK&Y6XR;8#ne} zFsb^jqO!6%^30@x7avzzym*e>=XYsuJglU~HdhKC>(jrolhw~*8>Z%JnrqPWZ( zADvb!So-&*^-Yr*I<5{psJK0+Z#C0@)O#pnErf&bA0^X zrrLpB1U-pXBDMn+@OBjRxLG;a&m&pf0NztM_iL;ykUbC*MYM<=7m`k@tp3xzXw&2k zem6FqZ^phDsr!(M@I7w6^FE;1Jm4YgYb{!9-~UNe7=Xlq*3|`r8rBDiqne)Y+Fd(;FS_lF9C!XN>F$7gI6S z?Vsejm~Ds9%FSvCFUA|bj+)%|Ud#O^&UWI2=0YnVQfKz;+49Ff{7uwc7KD#f#1f0I z=~W{yj*A&P>^wJN76&1GEd(OtxHT%F^5*IsGw>aV0JyuVXxT#1di&B1Yf$m*`O-to zN(PX++Wa=qvPib3lr0rMURv2r3zA?{;{6X79a_s5BQO?;nWSRDDRTXEiiDAoZB{yW zHiA=X0&o37e5CXIrMZgTD+_Ee&x#0Efg_r-(E|emL!S-u)Hx)hm<|ja|M-kckhDd) z3QK}P3^#ZozW#esHtDq1Kg=EtSccGKLuBU7gQ{wWk&+OIeKTzKHYK%|nE`vdiRH@= z^k+Y}{57Do++7UoT(&3O@XOIFLjn>JVcQ)o{fvWC-h_ndeR0Pj_7(r_!B{QWClr;o5Q*+{RQ6fhmczU$1!1gOF zmdhK;Dg{kbG;?d#4=+HW_`1ko>)MP2ufWHcU0%2+w`t|i#RX1Nq*Wpo>{i}BXsm1e z@Jpt;^|gpf63MU;jbaW?hJK)gS2b(B;}MS8+vNScnyTFI(FF-u%MpXlco8Z2zszYO z72`{F>T9fOweXIL_$qm>(Jts!Vy3RgiS3-{_!ioR-E;Ma^J$HhIT)JdXDeNH!rsdO zw$`}5ZWTKP;X>}!anrQNPKe&6#MMD_@@USIt5HAT0#r&YD2;=tN?)yYhb|a=`-)=3 zyY5K4Z#`(Q99%$3{%~r_{ps~P*!YRDl=GooBqZeheFR+PU`HP6&D$EZ>A}J(4|P7d zc}rEzkaVAZSugG@0Rf4MS8eO*i)nWo)bI$jPGhVyo_jgD5KWzhs-w7vwlhF;d*F=4 zNJrWQSLUhoW%(4B$<;RRR~xurHxE^32ADmy@8e8m4bp_Uy8B0zn;F)Te;3cJ5u#tdW`mY>@6gWCM0v!t8c0-Bo4SDmN3*JD6TN) z4;K(B^}5fi3@t!<8hw0l6j0U8R%#SP*<$wUcMh)!RYo_BV0eBgWE4PR6b9_I4aR6T zWif$J#?Py7bfx=HzY;)VHHm!%^G~tZHJL{v09=-^?VioDL{;v7uON>q2*ib5d-sN# zy)Sx4$qpLZcFx=rZ+1%Y<_(Q<+pz&6WHR0^T(srV?!@80C{21cC#bdOBM-E=x0V#! zA}n+lRqjc(;12baY6BjvJn;8lYWg{eV5Stmm2Dydn_zY^G^zfqmi1Vu_B?K2%jt3R z`(M~o>?L)b>%4V)Gx9|t-H_$-IW9H@m7vV`^gvU0m&)A-FZ+j&r}wto1*=0(TKv1 zJs!W^;hkJ|bIPacR7+dX$=a!#hJjk5Gm z=Xn*Q3k)^q+5DU#b%i`d%!NDb3}>y`-d2~D8X8S6T*$QO$1hQH(+_U%_}~4&YQg8W zHdt-&M_~m(pJ?X?Z#mM9wvq0d(MSTAst)jq?tKhH?#cPda;2F7GYC1;`B+j?#Kcpi z68h_FZIunk{M$HpX*PkoM73xD3eyQo4t_oA5lU1&oA7v;OLH)BZ~1Zf0E2pN?;w3U zF?Zl|_Yfs2lq$yi7UH*&alk;5WtevV(H@TxZ5QPuZausFu z-GAZaMnH;EWIJCL6?gD;ggbt%;TH-^LjY8RL)V=+>^|IP8rVVwb|uJ=F$`;4au&ax z6`zz;`qz|2d=rD`Ph%;rit)<~Sv-BoTLkN1Dsb852-g%Bp!NyM3?o3@5WugKlK7Qu z>G{dx;ZUYNjG?7c0%13DR>+1^jzk3j*|G{Dm(XJ8WB8KGG&Zo@^l-!lut@Q}vi8GY zg5O`G-)8|c5G}@!8kA@tqzf=GQ{-GD{)8Z9aSstdh;1f(7~Tw9N0|okzF)t#g|GMM$_khh_+u%KJq=N*uHIx!F_-l9DY^f@=ur&Ij4@8QkYG1LG5V;U_=GJ8Gd&R#naNVt0_%2!LhqyC}6=T&%(o&g)dr_CnZOhpXlrenM5An2eL zGVmPr7_6lTNvsU$=nqcm6jQcbnY$-^w7_F%D;1g3-REHpk>;p65*|YU362nUb`!H>mmY-21eu?Cki=H5!sU%)}z7lr~sGDqFe)N!EcW3 z{RMKl{RcPxc8kbCMU)5x5V%F(2+lxDJ3ty0i&vSMOST-F=)1oWnQADjIi~Dp zN%)5K1>zpAS#`$9&vsz{EJ3^*f2n_fbXx6LYlHfENCJ$%1K3{5U)q2uODcrg*YveE$ZxwWw-^3r1sZ;NsZh z;tumd(wjL!>4#oRP=N_+mkv}$a9*p9KJ3+}oQ*o{?!9NT3El>PZ@QbtX!~sp5sl5b znkY~Xd`PPc8}rYEeyJh*5A@NXiEHYT3lj-Hq}jH&k4>9N5_ckq-R_~=x2~kjqe&^d zNu&9QVt+eIca9nKbp3Mrg$$^Ct0|O*=tU{1Aa#gp(55R>px1kh*jh;NP`Z2D;eaze z@;iR#DFH{GqzT+)3*(*oK)a23jh_lB5cS_hW(#PlSn$@6#JU=9GG3j=AB+l;|MHh7;blh0#ZP7 zLfJ{HKN@vN#$^?Pa1KPUogQ{hj@Z~b>^n9G8Ot1VCMCGouFl7?aA@3C8o93&jQA;X8{+Ju1Zlv2H@#vEYzZH<;^F@6%Dhn4A?yIMlkhM3fi6ju2RQ`T9CBKaK?<@Q?Jp+MWgs+=3q+s^ zB5CJ7T5`0$7mm+_M=aMbsF@a9#&viIDGBCbY0v5Yb;Q!=u{q@9A?gM8gDcv$rOBWI zqwpw?82ercj2i))V0VCBY4pBQUjGO09`Hxgf^KvB&GM~J5k_GQGTbyIJWEJqh|lro zpy;$P{xLoC_=JQKGDfk00k*HMcm604XSK-z=AVd%Q{feA%R`PJ;)6J>D;vJQHG+Yt zwK}y${<=NM$mS$u)JQT2nr~ZU>_J$0itvWo z!))G6-UTfBaHJ6>2p{#oB`C}ItGl2|?m?YGz>r*k58uwo=*4t_BJuc+S}N^mgS%Ru zzs39a=f!1xY?wEn?5r34G<{q0O!&;C*Zq;&i_dXMRvIVndX*KInD~+^WYTj?hd&#L zx!AyGz<|%8{w|_zU|z}4tuh2qTWQzy{K$jW7%6EF`!ZC5#*Wp|Id(jR9gRR)YOPmv?HflgOH%#zJ1_yK}*Txd4NRYW?gV2kQ*nhkeFB7h8b_&`#-eDgsk5J?s}r z9Z09-@Kv;Lo;FWdvb@_>%+2*nc?!&FkpOxsL_-2o2O}YFK%CD4L1_nXlR2r;6o{}C zNGD3l9XL!lBdZ0|VoCFO7pxao<&$QF@&+H35$71K>(J9sj9o#*njo&M1Z`ARC%44^ zGhun~VZN+T+Xshy{j+K1`ftk@m3m8h8Z28Z0AVsZ5*4lXDXuPko$b^yVZPE@ zLMT|--eO_R#GrU%;xyb0Tvzl zB`kpv-thu-LxggbZC^W2BKVN6|3?sC#(ScM>x zDU3D)_~sWmuh&E^E0Z$X4jEd9ZzpI*5SH4RWb)7s>VDH-i>^E;jAvbW7dAgQ$D`oY zE6(|lL4`tVlrI&URB^C5#j@GbfK!g3@&c#xW`w*UnvcBhY~m54EK9=$F|!^f)6MHn zeD;H!myp~;EMD@HNyT_BKPnE{9OzV1q(g#^OrP#}J?2|sMRtZIz36_s3bk=U0mk?a zN)K^!#xLXw2&f2_m2iibugsi51{+BOH0jBg>TS!?fTic?FMUWL=Zqlh_w9r0Rs zv(ixXX~KuZZBqh^b_X6#kaKXj`}BeCJOe`~;drYXB9(3ipYJ;w8Gh+Mn5Lh$XkoPN z{pm6bFJ9C?F8nilTle6Z)_F7IL|BShg@eOYyzo;UJmq_u?ZvUN5vgQ3MSV;nvh>=D zltH}z0IGZoZ@!vADk0w{QP#c8`Ns55Al0>0hQI0R$qe}Ce!Q}tk>vd;^jlT(Q$Xmi9W1fU_qd2bEFcJQp^NGV zp<>{Qgxv_#;H-5`OrZ@TWXUam#eo)~;^*T8QR`psayx>XSk6g{*)iG?;fSlrjhnq) zF=B4l1oQ{2V66E26ftxD@c^`denuvy*jF+iWb5IHU&>?zM3s7HCP{IT@>lPp1cmJ3 z%hdiI9;1T?#xCA&_7;J`Ta@a99r<|tUxpg8oiU}xxu-IIrHm8~D+R0YaC|}lnm!NA!3joLf#sO!f$Y!1aw6d)e^_snQG^~m$5xI5jkq!;gVvC@mYd>bPKPg>Ll z`+?>A$7f5zZ{4`D0PLwuQeMyX4h4Q9t^5Pm+=fM*3W{Md;|3q9@)$NqwR8pn*Nv)Y z37aHclZPi){reknx&tqy>tR8&DCwl4o%<2BaKfHGdT=WnJkc}Pw!IBSRHB)^dh&9H zoKa{EEhvV1B7+nx4y7d-)Uu#zjjXvEA&!eSre#SUq^dwcFG=YGPBEyDhXJ^?GGTA( z7vjLn46!38)&S_H7D|M&*cK%r{t(?$3;Rf7IEv!7`y@;ihr2frN-~6TAZ;URuk-U9 zce}Vzl0k0Qe9(f@AsE{q&-U%pHNcO%>^`ddtVA*LEjGxFj1k5RAExXS;*dAm?=!ko z993vN7&grZ;xz8PG{0XU6=~~L+*BD0N_i{sG!oS<;2MfydGuu=yxQf`+b&ebH-v1! z(bXhF{|_jxM4eAOr)Eki?BGVqBlUf`fr$C5|MkiHko%*I?o%HJ(dM&z2dG?TQUn`o zjS|e3hjd}py)hVyE$KEu*1Ju7<|@jkr>9?b9LD4`v`F{64)-7Yi){B^d0(KW>q`G6LQT^Z_`2eem~8TCYO0 z0i~C&|9tyakyCXp@E-Nw1+=XrF-=&^Xom^e7-A>*1(g*l`C$$ zb!D9-NNB5+F#3+obuY30OTaMXhN`x@bs5pOxrB__fmBu_D@3?>Y}&Mm8ViWEV_qLN zPl**1G$djgg#+#YYs^x1b*c_k>sBcM*REwp}p z3R@g*Uv>aGYEuI`(d^-9kU5?U*hDZN7M*UQ#84`+IRvwb+-0nsXa;R>j)o@LWs<;jHG8d8HC6(gcS$HZE7P7jFq{wi_DkL715H5L-9>A#ZCF@3(?sVet(Z|#b ztp`eioqn6p-#DB9?)V7dP>yEoy{UfpQXJ_;7-IqDxn6H_=^fM}fb zKZTTW1YenAf(A7zlo_x$x9rJG3c?vD5(GY+x=N4$K|LO--VhKCt*!#yuuyJo$wWEvwj;0+3 zf#ZS*3$y&X6GdpsZZJD_om707_Iz0g&?RJ(lc*I90fR=P`D2)X69I7+nSjyfR%v!B z0Ugy0spSS0vI(pDP|7f)k>VRmH9$wZj}A5jR#D~Z+mQt&lBo9WOm|0>k-esG$9S@? z;lrx?nTA59DVO~Jv{_QWoI; zpuHxh7(9}(O~&wzd;Doy`Zq|22x{46S_{6wB~bEUG;gRl&kSxj4G#YQ`YeQr&=Ve3 zXb!|kghZ$%JWxC_gOq`2fBnOi!?kG9hK7`oL*)_Gp%x`VQtJGHFPM)Gi0&@TgxFoHS}@(h2rlN2Lr^yl_c3XX-7@zLf^u3(w#qbF5be6&Yp z)c!O(APui7(DJ7|{0IJiw8u~b60?K!u~~NW?k}F6gH=scc~KjMMslm2{_m$&lR1_8 z#aKMJqC#Mfk8Iq$y#{hqadZ!KD0@`XU{{r)3Npi`PUG$_@Y`X#;&k_gM6ZAU=hSjU za~9*a;I|lzz$rlUR?>{l3wxe3^jpd0g(?|9AkCg*CvD?EXR!SC>6P#jFAtO$_@ruT z5D0KcUs{+Y7h zumNPq1mfa0+*=ik)}lxM`ZTO`JC*Cv_9LRv&81TXXwH3EMEX#1hCVQ5@YggJUu)=z zHyhu9{1nmX;R1$Q!k5ME)m-z=51O1<(0@O93O8Db)yFCXnHAb!jD{F(&e|>ScKZ>AG>Le3ZZ}^51Rvzq*-@U zA>O!};L>qqsb|xj*8lE7E@0C~Y5pA{9=_!&@~e+Tkg^6r;ETPfX6 z0x4B8EYgSDfyKS1p5`wQiH!cWXRsb-1erZ00Pm-xwJZr$L}d(hTBs>=enwRuuwbb^a%K?uU z$(AJGZtcUh(R%$5q2>HH8}38dg~O_v!VhNVkb_sLkieS zuItXP=*0eO5S5gj*#6WNzkA=p8WH~EUT%k`C?|dt1Q4D= z8Ow?BlCsz^-VuQbujLZq*qfw`(}dn`etA8V^CvwW5Oi3wMms-rG*5ONY)+9FSrdbF+t+#h> zY4Uw)^`S=iq$FfTP`s&+HvjX93g!Tf<9nr@eVa^2drecZO(odMpbRg~T(J{hBLF&=Hj`|{(gCJI)D$j4dA4LFJ;99H?j zDg&eAY6O5jjCm#bFe{od&q9qK zf|jJ12M^Y<5fyyQPz_=pck$ocDny&W*IyZp`OH|g60m{ zY>;5)&#x}v_5vRS!6h;+=r8pjm!4Ug{DGt!xYL!HK0I2sl-%>dzGFFD|(RF+>duR>`K`?NfGNHzzl zS245BZaZeal|S>+nRsQ2*d}ZaonN4*_c>-dHOw@l3r>bX^wcYg{pR5Qk@oLyTW^&K zc&)Pw2s?-mf7;6JMIRH*?u?+n0{AfCkzJ2xddzWTEpr5KO%CZ1tOyaI(tg^*@!PXo4dq;0Yo(`uqpaECrJ#O9cp{(9m%k>QV{1P=(DA zt%}vejQwB$VKEQ`GUG%{aWwxdvt%O^=l5s3?{!VkJV>1yC^!sE$WV|4LxZTs=I#9W zy)P>$uU3f^4syhEdvR=wFBpn6JUoj(5dp;?oH}iZ(Uv$gjBbWxX3}PqD9;C5LVhxO zIKfCQ$WT%bX^q8(+yX{Ep0s%>*z6!x6T`tOMD|#TB8uM$e9KC3kKnEeB}pW}j1KjH z42fWZKE-RMY1eETgjUSMBYC{y1GM(Yt4(HXmbZF6^)CU|n&uZ$a}7S`X`bPnpN%?b zdQj9?vSQN7No2WU?=m%wfa6^0hXSm%)t(>JlQn3Xh^J$7aKeTZ$v*iPYQm}g5j~lO z)F!MMbN70h@$ss|=?#`YGU; z7FvnL#IkZfQWOvRGp`M=svR|lE6O6dh(YpllF1nVeMwzkpp{}t8}L&Q{;^Zg#5M2q zD4EB}U`BLp!2Q&dz%!`XPuYPcYbJS&b}S(*(5AvM(f85>EJ&F%8LtW65W#APQCdF! zb3Avw6Pz@`)TT(jHn0X9s#}1FY;Zb;FFM}D!os*Wn4sdR59T@=T@S=~(K=zeWzl^H z0wT$;p$Bv#b1(zBr&xP{yvU|Ssx@hhdvzTGV9TV}V0aIt@p?LuiHd(5&M=3H&mNe= z%>fkYP=UM-viXx>hm^FK+>%7Z5Jtl)lUcsOlpnB!Z2rcTLrDlGJ%utTtxR&vNoWLJ zh1oZLC>MCf9BmT>T8W-ELuXI>XPX=RSW@;2^>_oUXd|E5Nb<%jaFS5SB1ilE;~%NmrjthSrisRs z-}lIqPUqNTosfR6jEY#|8<ikHgVu&Zym)%r@OT)a5EbOE2MS z+PUvR2Ba){E7F$Z$*RTl7eW4G6XK&5zh|oyeiK|KiVi^`J0*og zP7_X^DN5e|M9Dc}ir}?MWWw>%L~!N&q`oECFcAvTq>X^s?N>#VqdlCk(}QlO@YA2CMLGga^QT)rz_mbdN{IKu5_gRF zX&|zCxW7Qe&Q%SWf7VdCu;#baTOi6oHs+%@xb*S3}_CG~rgpp!{}Y`1}YZBvJuSEcH%vd#M|p zQH^ig3Et4-d35F$Vi(5@F_U2-+uB66D!K(v_ZTaQY3wB18>^6-VGtMN8+FN;%*@gh zregxJm)lE68Nz=y1700|5)zVmMn@ff0Xk$r5xr%KW$+3KR-d6Cn1B^vQEfR*2+|a& zcf;?*ul+%d68VNm9*V@%%O%Eb2(1ZH_`cxj4Z{fU&@qk0ZQzq`JyMYQv1G52HUH+K zG#m*+1N9X7|J7naAN}7qVf>%}L`Ci=q06mQhMa z>T*kDIk|OP7SOSZIQb$9+7ZcA-MGCuI?VXTR|$ov#!I6JsA;!)Psg^=p=B<^kRJKw zRUs>No6PaXFFeqjVTt5H?artM;R$Ymr<@LxByq8uB&4ny3dZ0eo6&Jn0=gs+sY{tj zbm*j~21+3E-Jip~J=$TCdO@rXNX~WS!U2W;hNZHEhF|s|t!TY-U#Ju!U8lMoEG9l*K0Pl{PrZ!vy|3oG%mjq7#p< zC^iS(MO}+oD5et@XsM&_w<5c(El4x_US;*zKf7?tax=FMf_5I(4t497f( z!M0_Hgw(KTw-)w)u6N`VKlR0VTa&{^MvQLB4a_^-kgKP!o^}j%|Lvf>j^Y7hT^*g& zx>M;j4bN~mfJc8#dGi+urn?5`3I0wfisCpOslevY;R|3*BC;Dm|7E<%K1#e8vG2qh zrDDmJ4*fJ&!KzGbc#Jb`KP+1b6{__mD-k!jG~5T)Zv7*Hf-ni9gWUKQEcCC5Z6oCbb1rL5i2{ zbwz7n3MgYsz?@+8*Nn7Aj2ya9RRvm!Tg1#xnWSia-e|-4fWB0sQpnK9ss^IVebb`c zF{uH7kiHAQYQCk(V9w6p)M~I|>|i~X=}kw5#1>X83Hvp~1ok={3J^#D_|wuI*Qh); zdo|iU1duq5jB_YOQ3!V6oMc}dZG0RZ4>;O*6F7j;tPZ#uZDKZe=1Jc0HLUNrc~(?R zwl5ru#&`L)Bju=q@b|`_L|0;gt-fnpmxG7pW#|CJPU(=%TM3Y121tBW{ ztkHwW)3DYZAOQab8G_C6*Ed{7RVRnVv|8_(Ng@r*>EZBcFW%$O(oUvO&=C@BG7L(L zFjMQ95u^8#*?D+t=2ZHZio&WjnJ9s~Y31tFv=I=r=q)UCL!Y*)6rm8{0E5e`47X|r zo_t6>QY6$+EGo@$MBh#ZhpsS`w2zzskY9S`ktzAS3|A(?Pm z*qsysmbA!3@D4lvY?cc;R(x6HL$)tk78`X(-8@Tr=rwV-LLqvHXmm^j7;+`%Kws*h z+5-$Ps-^ED7)B}_SfNFe^+2XZ6y&NE^g$DAvkpUf@G)w zkIo=Y6zFNE$z&X|2pl23eCH21s@V>?x(K@rVK#zA2X9g+z_kT0s4R+V+!nkT>A-AM z*Tz~PQFQgi%$B@vCv0m0kd zLX9C!5O9bC*rnV@c6#)ut3&%-zNbw9hi5;fV={<8qKbJJ1&d$G zY-ET>{o(Rx6z$GJ{%WyQ8QinF#_HA>-*lu_de8f^Ho(PBqap&YI}6}`9ibx9=J6^1 zQI$gbf=j>(BAA>Y+6HAkI!gg8>4*suwZU+UXuVx6cm%lwSP<#VI9b@d$b7=>F~#Y_ zv`9#8mR@dL%jfV=Ws@5>(8!KWd|Y=Hu0z3)hr?I>V3Qd0I$e`L%0~N=pcRV<43TUo zecT`Io%eSeG_V(Nm0SlcB1h%g z#$|hOE;PN+w2jDV7G%eu!-YV)JqHJ%+!IcGLc?HvADsgh1J9OlRFgpDi=22H&N<*1 z#$rVRGEGHvZW5-IfCwbf0FnzSbB6hu|6ru<;4PCnNSX$nn++q~p3j5i=B7C$Vh5IU zYV5YqD(dDh2~{Z_u<8^Y$Tb1&EcZKufaMBEz$IavXMB#qQ~xH-CAJzJ z8c(e%+jA-r%;*poI-rvdIdy;+BhpVnuy2bK5)%_Y{W|1BZXATrQXtx_zj;aExvAfl zXdoPGoe=3jE$H;Vip{##tKe+P03g~KLFZVr(8WO2I}lZ1DV>0!2xBDO2MXkv*RTBe z!9yBNg4I#3L)gPjfPIq7HEZ?pXtZG|yGP&_duh{?3ynIjm(MFfOE7NtNS+Qw3nmp2 zJ{M}fL^!WoVCacHh9)$*8y{e8!h5qrTZaYgCd-!3Dq)|U%zUvG5rukUScQ1qLGa~1 zY}{;9N{tyFgP9&uf!QSaLjWNTVX0H?Ou_RV=~^)nWSusLM=$3`l~ zD{RsFyc$gl%pG((XM6;Fc?tX9bw;QwPq(h3;t3P9Z0#(y6=$dz_cZ_RA)dD7HMNeR zYb^|TVac9v%hN?=@BhvuvRP_& z>kuHPjts~aIQbmo6_O5ox_yiA(bvwPJ?znP6Iext0qo23W2V0Pq3p=gQKd|K95s8O zY1x3p0`QQG-H8HuHGDy-)L9vJF@Ue$J+gyOm%Cyjb*w;!7u9-m(iPPOrQDTBV|;@E zUZ^rjfo5q6f=3J;m#``q7Bc2^GvHuf7D@0g@zfN&!!88~=EJThGW}T}HEj@0&mp~j z77lNtj!b9=KlCkIoQ9}@(_TS5;XNX3iD>Co$7lp|aKP2=9A%i!%3*S$Clx3uw$#>2 z)&-^k(dDjCu`S4P>YJe5$>b~g#~*NuAKzHe+?jYEmN4SJ>2OV77M!Th5&x!6*83)rsqjH%*eiUlpBv1t~&j-@^Y2%#a~qqU4_o8!V6G zLuEFh?~)o<$ZN-@15=uzqu@}y1RPYsjQZfD4oh@@nNznLmOAX={63dBuGQR;P{J~* zx2kbKM{$8jeG|x>m$aRk)9GhfhFt^80&`3n?7;Y-(CiX8ry{5ZW7BcG{G)`Qy3mkO z2>VyjQ93jB9@&W@nE}Y5T5sI!~cJ7W9pFarS z-0^$S*M+~|&KGuE@mtwb;h%L?7yk6y?;O=XEP`4a#HEeLXBBF!yPjjJ7xy{5-F=`| z+v(1e^Wm`_8TTc1)qZAT6Kw2K=MG)1_ZW2>hHJy0rM7iz8xH3uL7 z9(^o^D@|X3!9y{yZ^a~lgs%-Z*ijeS%2Ly96`0q)SVX{WiJ;Rs&I(ob>@srqKMwS zpO_bqVJE^m+3av)=J_=nHW-5%TGO)i%G@}#$?V=e<&NkWW=7GYQiQI_&Y5uVpwrFZ{~AtW)cSe;P)Q`Z&2I3`8JO5;zXsTDGUQM%`lj@*3`A@ zEdHeg7F%TU?tJl^WQ*VmRbfC`)6x~t-VoYa;hZE13j&}nYPBF zQh%J-^oWr?ggiZ;&*_Bstm3iym=@GsC2JI(;CNk&<7&5Uf4qYK{0BQ`^?Av+4^CM7 z&pGdev%w&LCb8I%3}>&g4VydZJdJjNV6>{Ln>TO%1lrS|6&)KZ#k}UBCF|GkW!}g6 ziF0Ctf@abf)czSm7HFc9)}|&iQK_yZ2wrWF?m1C7bmJ3b^tAg6bq#*3onM1Myyb@_ zO8(S=gXiwRv9z1hXk;t3Jx#Z$P8L-bV{cv<-9YeifA(^Uv$shxD^D z7T(46qEgEu0DG2Vad6(;R1Yy(D0zJD|26xzI-x|~_;duU^g2|G8=FkHLC+U%0Rn)#_sU*8yIG|H1xq&`&gqgW#ghK@W%BJ2(ku z82sxiBCqZp{Qug!_Hd}zu06G@y%X7$L+N0rnaHUq6$|7u80(alB5_RD#z^*-~A}<_xA*Z< z78|FX1AK+g6xllU7y?Jo-?~a(eg`YtzCx^x&$qBLdw~b5i+#(mI zmOxZ$<3=;=NEj8rEUT*e;%#8yAd>#8)0j?9DOc7xVCS`3zCVZ^C?ls`hhODAyQdeJ zdwj?x)%=$3^LztuaQby(GerLQ{<$~6J@5^FBuo0V{Z?N5m|&T~mMy0R%Pvw82MsW? zBkF$u>xCdf@j37vQB@ zZgBU;l8c$JnqegafHar;P}%UiS5~Hw@{`P>DL*93c1c)mDETV7=QnAeM>(N3c zA4k;?FX$g}4XRLi`vV8`Sg54M0^(a*%`;G&vQ>_&_^dk`*Pmut41R1*e-I?D&$_lN z2_%KRxq_DYlpV$1dvO0g$G+kci{0PPg!ijtsR@=+IrXr%=a7QSVY^x3moIZZeE1;Hwi9(!3f8G%hSsrKTiG>nOYhBt10SgEH6BB1T*?^$lgOjJK=~yenOEF(-JNm=)^Hb;a z64(6J4fGz;=xWCYQy|@2c=5}^fH>%`WY(_T!>TPlaU5=G)_xl;ohgUl*EVbJ5}7k) z%jp)mPEpFr%45ak{x8Xwc}EhctJKx4aPn1#W;cjVoj-Zgl9MOEkNC2vZLO^b!NQz$ zo&e8KgRYk+!$^;=u0Hz1hv2D4ins^;id<5kl~vBM&-<2Y_%y8U=JWZEz@FHMP^w4w zOJgk4pnjttZ)%AQR-u%kMPwJ$$8dhAZ^|6K+WxTV3ly3(I#`Zm9F2$7f*j@V|AYeo z5W~*Sj>W!ze=AgA6SAhqp(?Bgqd}u9M(QKC4<_viJ!+@QPk(6q5<(RcR9f`t-s!GK1k2u7&#+?O8In2~;urL= z=%>76CSM94*?qDE+AfEmBE6$ufh}0N%seTImdCpRZys!u{v3kJ1O!QhP)clplAtz` zT~L!T{1Q&r-2X%5a*1zevGm52MgqH zc<1dFjzG(_f-V3ojF2Rs_7laRskKPlKOlgM04#r%dC&-}}i`qqu&~ z%E}_eALkcYI#4k4_6$jtts%RXAqPK6o)<-ElUoNWnQuy`5q*M5r145sRaL&E5|J@s za8T@Nz}bKNQ5B;I?HloDq9eTwZO&)Hhv_m6vKd!k@Mi{JqYdzXNv` z>_M8$KJYYV?+xj63MJ9B>7$Og-mYDBcpneEK^NGNBlMKC0?l?X^Y;GV-uFNNcp^MsA-!Ji`cu=+ z0pO^3oggZmv_E91uUp3ezk|tM*lv3+O-5!r?vej77GC>|{Np1XJN zNB?(|pwoB}87)@uRRA8kKqwi`WUM zrq%iA$&+yEUX&7b(%dz!c@#D&3no-YCStf~C&Emus`^^?f6SD4+ho z>yLmH3l9rRK>Np!Y+Z_$EhF*xJZJ*hVA<4uI2S-K3TKq&l*<-tl#RgX-moCWA zxSn3^Ayl@eew7n(K67f1p!)?m7y#=2)OMToKQCEg#l2Y9!Q{f>0Q|5J1Cmi;l2cX9 zzu!%vh~AWa6?947JU%#Ba1+!&I^)t*tPEFRD8BC)SV%v|mCDLStbQw6fsl#wR4u==eVs+m zo8H8n6m8=?@#qQR`!Gzg2L4!>iMhGC#2Q;aY6mi~I#E+ALoR~jDO0?8`{xPI)tL6oxG3Y++D*jCVOsTMhU2$gdVDj@R?!@8gYnK&Z%9b5ztf{1&Msdqk4405bb zi#}YKk|M8Qp0yP#TDjpuM126wc}OyFaI1W%yLYy=h*)btP~ z>v%G5)=b~R;tGoMagaCc$jr>NR~77TDl|n-Aj34rm;sd&Bg{TI{q|^eFC#MBCbL^t zM~CwAjH!cA^pFq*bo;#Cbu%0(;O$rf0iQ79P4ymd1J|MNeCN&iE1 zHTjQm&q7ooix9bd`SKB@qN$2W_fq#3)~S!*(FFF(DOg9pCOJbF!e`O>^Wy>ThKssW zQc||~CLxFVICNqf#V<(wYi7lZc-p?l%*-sm@f^Y*x?$`W2B~{Om1{@Cc z@j065g51&J&9=wXm-R^-bVu?rk==IbButw#yh3KV6Il)JYVsF;vhnD&QeQNp>2li= zA|&jq5Q5g)Id$5&+8SIDP8ttk?qL>PAYw@fwrg}R}dsB=3vaLW52R@H86ydIjBpP^aM0|hHtaDx!FL=a4c zRPs?uKKnPlojZAu&!vD%-mKzSTa9$x)V2^2AN1Q!AcR{|D^?gM?OHnUvDpS;aF3^_ zXQqc93)T$dUxyTy9BD8K$e1so*vBDOckbLd5<4=N`tDsJUUPK3Gvr>?W^eBT+@NZo z3_ z0aR_)5p-wIuh;bXJ0pJ_fKWib8G4X>bp?e=Gi2$izAdD!L@2%f(82&aB*G7W5*A*~ zEP_Dg$wrXlnI5#2D<$U5o2S#G&zdJGdAOC&x3=2$rHC*(XOaf1E=Fbqr?fIa_3y#Kyb zatG+3&);1V`3=Q=roL9k_NHQ|9_1oL2P1lkzqs`9of1laF;cfb)+UFwcov?|%iE~U zQi9fTOXKHyRb|1lvmqfHsEd%L)6$SM$my`<-T;|{*?8gMyJt}G#R2e(6x_kcJ{%GS)70?--SWnxMO)m1H}th~YA67Cb`Th=61r=&F~KmfjX% zciF~SQ}gvl^ugDb`1(d`oUlN#HU`9!Fo2g=k#+a$B1^<*78NkC1D`l^Ff=A64IgjG zVibBhE9@;GfDQx({6kn{H&KU(h>J_pUuM|-VUljWS83_RZ94rr4j2XDh$dA27z8&7 z*cqU~iZz1x0uZ*ct5(s;<)gW&NM7Znq_$K^&BFKw!UHJ;0N>W7q_-4SmuULqC99U6f?R}7t00}h)~`f zqv>0WWUe>o2u8T78C{HcF=YEYA-Op@*a3 zlOQizO(#$uUMnc&L5u5UoAb?ZiT))YR#l%cl)zc$VU8upLofH-w;pA8pI1H=nC~o!-d1!3e28O(kPS~Y+bH~ zh8otDt%}qd!a+|=jH*Ev6kgUMftxplRh z*T26EJBjwR5Mxh*oXNAY1n$Um;vx-xHhc=sc|=&)Z{u*dvKuy-E(_VVEJV6_a9K#w zj)z9pr+R{(!#`hEZ9@8_eAqV*%)Ds+2e?!|8Rn?(BzHKY)3dNop{u#1gp^TG&QA?c-~qa)f_HQd#PP@q#+6TA73`CVk- zb!J6{JMKC0+ca4{U}z_vDF3|gI<@HX#Tr@iERRO19L+&cpn5sh4uc*iw@;`0Yh-%k zNOVD@07k>m(UA)Y`C`H)gRt-hSgedp+WW|+{;cL`mE;%0*A{_?`t}XEXA&pa@WUkc zI;QxsChhU#$4DG&HAyh`kFZ>E*!K&7!L%k*IFHH>gontyaoGKn4h+wgs&`k5E&y@O zOV}n%f-*LU6bq5*&k{0%I=*<|>$^_pNs8Y@7S3nt>FJ&8inMoX%8FS53?w}1eLoeV z*6i*L;hC7pK4GuXF0^^fd0&j)#6!DEz_dI_>-v0b($g5BFjL;o-~!-d=PHMBcH1m_ zCXc(R;g=&y1Eh(r!9#eNItSs~roUK^1nJUclc?{_Q9>!eVNh2A_ciYck-?ws0NN#! zT*MW~A4>;CB_z0r^>h|&Di!`tB)YE;#`pMylhE~X3wl@P`V*(^pH*x8VofUcQ%1i3EUtR`$u#*rWe=B8pyIqeQ=_g6o5JSC-PS=sVeyCLIS_j}b6_u;n>g5Yz` zZ)wDNGD1NZ`EeUlt<80#2=F@mF)-^+78}4vR#7pnH22#EeGvA2k5$5C07ZKH>4v}m zKIVVs%+)W?@Cd+NB*`j3zmmRFy80wE`1Wp>@q5m%@2EBhB^YFr{8P>}jp6F*S~{e1 ze9#FNmf7He3PnE@Em3yAS)(hc8ncQ-=gl*8T1-YcU`~p8>zv19lU85eC^??H|M7g{ z{vNdvu_QQ2W2cGHE*uZP@~`#3X@pH(s_M3N0;AXmaFJ;;it7Rb0!CZ6{s}2Z3Rn_H zZ1u~`-_Bqdjn~bJ94o{}W^XGUQk(6#Va0|(8OSO~%@*C+-heLQSFUg~GBVt{A9gSu z(XWGm_zYuHIIzm7cABE%F0}feLeY>3`xVRo0va_5@$dxXg9T9LG61u|dJ#YkiCZkt zryO{;_t2Bc=ZXPJw{;Zu4xTB&Mv_y2O?eKbzR1$0sbH>g+IEAdeh#mJ-7#`nJg!cF zbFbUNwqX~dcrc(Nnb1XbDU{iBsC)C~&5#Z-6yQRhL#8V#E}jHt@?NJK?0Di6AuK4| zYj=LjOEbUq7yi<+yu4i5Hw3B9CB?mt9&~*2&y!~ozelm%^DjRJe{DI%U$1|#;G4+l j-+$v@e))gdj}yWrX97x|{JdD3ycS(&*VeQxwx|CKN;K|{ literal 0 HcmV?d00001 diff --git a/docs/articles/opt-cond-thread_files/header-attrs-2.5/header-attrs.js b/docs/articles/opt-cond-thread_files/header-attrs-2.5/header-attrs.js new file mode 100644 index 0000000..dd57d92 --- /dev/null +++ b/docs/articles/opt-cond-thread_files/header-attrs-2.5/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/opt-constant-folding.html b/docs/articles/opt-constant-folding.html index 2cd1d48..25374b7 100644 --- a/docs/articles/opt-constant-folding.html +++ b/docs/articles/opt-constant-folding.html @@ -31,7 +31,7 @@ rco - 1.0.1 + 1.0.2
    @@ -56,6 +56,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -74,6 +77,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -86,6 +92,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -112,7 +121,8 @@ -
    + +

    Constant Folding

    @@ -128,6 +138,11 @@

    Constant Folding

    Constant Folding

    +
    +

    +Idea

    +

    The basic idea behind this optimizer is to reduce as many calculations as possible. This implies that any expression that can be calculated before running the script, must be calculated and reduced or folded to a single constant value.

    +

    Background

    @@ -141,16 +156,17 @@

    Example

    A simple example would be to have to convert the unit of many temporary samples from hours to miliseconds miliseconds <- 1000 * 60 * 60 * hours.

    -
    code <- paste(
+
    +code <- paste(
   "hours_vector <- runif(1000, 0, 24)",
   "ms_vector <- numeric(1000)",
   "# of course it would be much efficient to do vectorized operations xP",
   "for (i in seq_along(hours_vector)) {",
   "  ms_vector[i] <- 1000 * 60 * 60 * hours_vector[i]",
   "}",
-  sep = "\n"
-)
-cat(code)
    
+  sep = "\n"
+)
+cat(code)
    ## hours_vector <- runif(1000, 0, 24)
 ## ms_vector <- numeric(1000)
 ## # of course it would be much efficient to do vectorized operations xP
@@ -158,8 +174,9 @@ 
    
 ##   ms_vector[i] <- 1000 * 60 * 60 * hours_vector[i]
 ## }

    Then, the automatically optimized code would be:

    -
    opt_code <- opt_constant_folding(list(code))
-cat(opt_code$codes[[1]])
    +
    +opt_code <- opt_constant_folding(list(code))
+cat(opt_code$codes[[1]])
    ## hours_vector <- runif(1000, 0, 24)
 ## ms_vector <- numeric(1000)
 ## # of course it would be much efficient to do vectorized operations xP
@@ -167,16 +184,18 @@ 
    
 ##   ms_vector[i] <- 3600000 * hours_vector[i]
 ## }

    And if we measure the execution time of each one, and the speed-up:

    -
    bmark_res <- microbenchmark({
-  eval(parse(text = code))
-}, {
-  eval(parse(text = opt_code))
-})
-autoplot(bmark_res)
    +
    +bmark_res <- microbenchmark({
+  eval(parse(text = code))
+}, {
+  eval(parse(text = opt_code))
+})
+autoplot(bmark_res)

    -
    speed_up(bmark_res)
    -
    ##            Min.  1st Qu.  Median     Mean  3rd Qu.     Max.
-## Expr_2 31.70547 31.69347 29.5478 32.35429 29.28767 108.0261
    +
    +speed_up(bmark_res)
    +
    ##            Min.  1st Qu.   Median     Mean  3rd Qu.     Max.
+## Expr_2 31.10212 30.68826 27.86423 29.55408 27.34598 30.09918

    @@ -191,44 +210,50 @@

    Floating-point precision

    When folding, we can have floating-point precision issues, for instance, consider the following code:

    -
    x <- 1 / (2 + 1)
-y <- 1 / (2 + 1)
-z <- 1 / (2 + 1)
-x + y + z == 1
    +
    +x <- 1 / (2 + 1)
+y <- 1 / (2 + 1)
+z <- 1 / (2 + 1)
+x + y + z == 1
    ## [1] TRUE

    If we fold it, we would have:

    -
    code <- paste(
+
    +code <- paste(
   "x <- 1/(2+1)",
   "y <- 1/(2+1)",
   "z <- 1/(2+1)",
   "x + y + z == 1",
-  sep = "\n"
-)
-opt_code <- opt_constant_folding(list(code), fold_floats = TRUE)$codes[[1]]
-cat(opt_code)
    
+  sep = "\n"
+)
+opt_code <- opt_constant_folding(list(code), fold_floats = TRUE)$codes[[1]]
+cat(opt_code)
    ## x <- 0.333333333333333 
 ## y <- 0.333333333333333 
 ## z <- 0.333333333333333 
 ## x + y + z == 1

    However, this code is not equivalent due to precision:

    -
    eval(parse(text = code))
    +
    +eval(parse(text = code))
    ## [1] TRUE
    -
    eval(parse(text = opt_code))
    +
    +eval(parse(text = opt_code))
    ## [1] FALSE

    In this case, we can use the parameter fold_floats. If set to FALSE, then the optimizer will fold every expression except those which will lose precision:

    -
    opt_code <- opt_constant_folding(list(code), fold_floats = FALSE)$codes[[1]]
-cat(opt_code)
    +
    +opt_code <- opt_constant_folding(list(code), fold_floats = FALSE)$codes[[1]]
+cat(opt_code)
    ## x <- 1/ 3 
 ## y <- 1/ 3 
 ## z <- 1/ 3 
 ## x + y + z == 1

    Consider this example where a sub-expression folding causes precision loss, but as it is not important in overall, then it can be folded:

    -
    opt_code <- opt_constant_folding(list(paste(
+
    +opt_code <- opt_constant_folding(list(paste(
   "x <- 1/(2+1)", # will not fold it because we lose precision
   "y <- 1/(2+1) > 3", # however, folded or not, it is not > 3, so folds it
-  sep = "\n"
-)), fold_floats = FALSE)$codes[[1]]
-cat(opt_code)
    
+  sep = "\n"
+)), fold_floats = FALSE)$codes[[1]]
+cat(opt_code)
    ## x <- 1/ 3 
 ## y <- FALSE
    @@ -266,7 +291,7 @@

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/opt-constant-folding_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/opt-constant-folding_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/opt-constant-folding_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/opt-constant-folding_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/opt-constant-folding_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/opt-constant-folding_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/opt-constant-folding_files/figure-html/unnamed-chunk-4-1.png b/docs/articles/opt-constant-folding_files/figure-html/unnamed-chunk-4-1.png index 82b53402343dbfa8c592e7f0c780b94b104f5ae2..5b24cfb2e0f1886f8177c4ec64d45964687eefe7 100644 GIT binary patch literal 34666 zcmeEvXIPYH*DY$)BnrmYAW>S3Bd7>O5fBhCmgtBAR-{`%M4Hl(YK$5yIv|Kt3pSck zrK3iXA|N10N2N&b3`0F@&y4x<{XXYh=g0ZJ>v}(Zg?Z+A?q}b7uf5jV_j^HCd#m90 zi@z5T5D;W;+q_#qV5X~pz>GuR&ceSKPcJjUAF~c?ZQYDN@qe66>E8I`yA#{=9Rvi# z&(nXV`8M~N3kduyz})=XUe}P$I=7HOr_8CT*13fOFMnNrXIA_rspS$z3FYObgZ?(L zVh+spmSKZU0fx`+-K@zo+9fk8wbJrfP;Q3Xsz-BOPO98`cG~B5#9z@f6mklC=6QKt z{d3Nnxsr0v8tjy9m18>wQe%s)GCW&mXgq82+_zR)S$Q_yna_(ytJw4x0RfLS)8*;U z4ZnStL4OL&T-H0C{`}EDXAb?TtNXwI3pNGn0o(at-H60bY5wsR;IM& z>V5n6J$m?X+mRzjQnwxwf0t}ilW3E^N_dq?#yYJ)IoW~D(LK20ZhifrbLY<4`%brP z$ns>}d-UkhzRAw6>ZA^7Rqe*Je?&z`@3674Ia(Yh+q>IYT-9;;%(+vWa;hUXofpE5 z{ikesx{^)xz2syiX8f^tzw|H-c8m>o?a|j?(OY`>aa7cvJ$w8ZLCSVv*~Pn@aapZ> z`)=fDRqflm*MCwY#Vn7N)Lv|&wU(Cf|L9u&@4t=zxdOeD{}XZjKP&k^EBXI;cfU`r zogvWtuDsm8t?YqCQHV<0ho?uBYOE6yEz9oL9+`$0^yK?*Ob!$mhN>xNPjuN0Xig1i zy0PnAw8qq&+WqA{KF(RR#%!BX?TMFbw7;BUe?VQtF&*>`y8) zTiyl-HcU;jIc)DK_GMA!*c}U~PCQwOi_b%Ai{=YPzcv&Lhj_wpiz`$nI+pJc6iC(!Y%rKM%O(?YYL$BUg_nkDZ1VTnb^ z$=^qZI`pcQZK@N;M{+fbj`*qvFczuUn<4KG48?hSUMZf-z0;F3<)c5_W)4*}f9l?~ zRs2RrT((#0vtUNBO|pJ$YRfILQ(taJdMekR*d*aS-lwP}gHH)W{oZ)i=>wL@p+Gjp zKUrtp$3H_{KV67y{%KO*+}vDStt-)EAh=;id}n(@fn543g`toF!ZkHTOk58wls5@sR5=u@JG}T#I(o1V zi|! zSY?+zvAsC_*Fzh%H8@Q>f|YEp6mfDjCs%uq)ikge@87>q9W2#Cc^$Y=l*F!gZygAq z@?hP|%F@UlZ<`bGBXcN^wJf#qk2%|)I=>WDE_`!+=|ETa~3cv@-Gx?J{qa>1e9kWpP>-2MxScZYca~{?hF=55i6zA7%GU zT`1Puwd;DBwwf$%f*J7Yv!02Yc~a{)_`V2pyqW*}er55E)^KkP=59m7V8oinjM+lt zIj!CC1#+DCab7oU>(Z9#Cr7{3w+MaTULd;4IBjj9v9ta{Sp$CtXS`K2e1_1vgGt3T zneJ?{lYh^T?7F^mYwWNJZPH_lrptR=#`|1yl~ z!bi&=uoz_zbc*rX^e)E5d>pQ^EB^Dl1uTYfnq%IRLoavl+V$?2R*m+DcY%}w8F>oD z>;3R-kGB*NlB4(YaVeoBdMw~9KgL$Elbbczsl_qnPwUd03Y;p91Jlyd7%4)P;lFzysY@7BdQQi8(lik;BzEf0Gy0`o!uQO6Z`BJz&0l9yu_*HDYf=mmx?}=lw3RX@|<~FX(W?n z=QR)8ZBJ%aWQ|uS{p)+CKZ1_P#iOQ{CP9cV7wPYK@NQgzd_r zhQeoN&nr@C4cSvIp?=D7ayTVB>CI)P{lrkEMSoKslZAh6YVy;ued>^H>g42fPDtFK zI;%Or(C8H{hwf1<_+O?|qm}y4P8?#H#bplXW9tuIC?0B-_g0~OYdO|mAYU96mz7qn zpfS>TCbc!poyE9uC z^Li@`?V9V|dv?aJ3JF7oiSjI3puKKMdRMX`>)z9+tFh5cjuwS*E)=szJK|a&XiivK zj*WcwEvEck?cn8)eJpXoeW3ty?;7qg#D4A&&M_I+aHRc22P4Yt3x zN-EX|;JqZ5wtQlANQ;#*P^7j<$@)M`ad>uq>n&wFlOoNj@lxT};bXO}0dwMF+N?~+ z4BBSd7Hf=In_#sS`_t&YqqjAl^OXr8x6gMlbX@{*> z!9v)sD387T`t@t06VGP`DcKmX?%_qJ_P&_2M?dvwR%lI$yd2{45u1QYzo|kc2W`|Bd?pfm>>)EvqOxCSiw~S7{ z5^k$ab+}YCa#Be3OGK$b_B{1L-4#j{A5V^cm(i;;i47L$)nHH{F;{HeuX;u&x1d`1 zt-6$f>}h|0M^OLmTv`=M!2<{jDScEz*kwE&h#T|87_fk_;T#y-*QEJGgQd^yD6U^XzmYlk&;?Ux6L-FD> zw6Df~o`6-9;?w(Gh(dqU@lPTuC=vsxsz=cTEkcCbB5?4kWo&Aa-=YL<*ZQ5oG&VMt z%O3CVeDYE#Mx$YCm6_kpc%#NSIefn!@|ChVIQp|3=w9b?U zFDo-QZa-@De2!HNBBH#aBKAFSi8Ef6_0v}inJfUF=Du7>`2lYK-7)X9$N^c@Tuwzw z*1qa20FiwDBW!7$xxXfBtlQ8)d4$s%UW^2HpydL0edbMoihCI(l{-0Y<^Nwl1gvR3_r%(w{K%V&BUh; z%=pG6;+2^5heN1@+ga=J(;yXxqe0S?F%P7Cm$BhemFwp#12^J4mNDb8pKFoY`AbwM z)jM4lsr>48^X7Qp0Duc+Y|9n=FO7MU;djGozI^iCi0#j`kl@H3_{6}Dao3SBErTf|E-ROQ9sZ*021u_!(8Xf>?sd+`%F)2a~4eheVAH!&&o2K>hn#N)JACtCU? z6sgJ}0+=kd?q^HmGWxgkzIZua*MX(81Q#dP3RcqgS*xTJ;IgujLd^u|*60u$r7JdF zURS9u-KF@SqE4taf5ii{_VP#OVT3j#e!?n71NelC+CE!?|DXyYC{)_+`Srh9>g(^e z2EXK!1f=C(7hq@`1h8^6;jNMutGu$Zb*tv+uHP^JbbZUEg*V((xxW~gGbdX3)KL2k zfQ1zR(R(Z#fM7$h@#K6pCtVjwQ?5SX{#{0n!I3k9N^j`Sfb5lS+$hipLzTFWwN%yw zR2|5diSu{c$i3(#Ty)<`85xJpr@1Xau=|p0?F&jL`#$Hk{=Gn_5ar(w#YhfY&OGEl z_{71k66%ONuAQDSyMig&F;*enk3i^q`9ZXl2$hV zr7-R-YG<&LO4GGX+pYR4-oL*g5xmxNH!bztMQa#=M)pjKQEY=}m&+)Kqun~ng)^wb zm~8c)Y&{aLe4;VO$no74tAlqG?Iy=LoUw+fFaq$!>TDxa`yfH(+FxsIQjGD|vj;mW z<7$__SlNNhmo*UD(`48%x_@#kdrA!%&;7o3=q-jNfH}ctU|(-$?p5)DpkfcH^v-y# z*+^ZKZ&$j*($0a*yNUiVvL^Pku4ENa1J`l1zaoyDre!rB0E+ zJtgU1mqdpl4O54u0&GMjwIw#aJD=_q`$>wqaMgipPmkmt?fmfcQFL@6&`p~)n#o0L zkLe~?ACE$gy8X)HxUFqbVUWT=wN;#X={=@fS5oB~WDs-(T?>qN83B%W*QGx|4XHXF z6%&t#w7^5^pl2}faHjhn!F`W~DHN?+n%^(UU?m*ogN=Y$mnHFnR z`XA0O{`@Wo4NKe2`b>Aao3@%1JBoAwe)G>)l(n6lhH|rx0<_e7jdkq^*HA}oC_sY{ zmggsJk?+3(JfX=YPa8xjK_zgV0f23BOqP(isu>D#QRDM734nCCWsOJCi7$lkXiu)r zsvlDHC@(K>^53Jaf5;p7*}3*)u954x28FD1kiuQ52;F5CmRpHs{w*{8)z}Fh}P$Z9x$xM zr@t@Y39Ic}4FIac&|9wOH2E=E=qjv%UKXu)J%P0AZ<2Y6#SB5J zDnJ~2K>Fw(wzqe5I25kVC77H&@=8Ab<2lg)2{mlr0blR&rL=nVpyL4jZ}0kQ+!`Gn zwa%UxRv9eWRF>}2BWyA-kt=KT>{qI01HchSk=Rm4K3z(9zuz!vq*4W&lu;1-gVB?N zzv)*>pRLeELnrUqro|}@%<{)#?Fd=7`g5*tF||&T zYc!EmeG*VsYDDO#Uw;mt9DJaqx7?+xdfy-rgU^spPOzD;`tX}G^E7T!)|j`xcf&Ax zh|(hnp42aQF2x^w{znKfXu+$xrm2bvalB^APR$bT@;PE4IXIQx6P5GUcLVel-Ow{h ztvy0r+_>;$LuG&14bFAPh*>yO~a{rgFmYJ zKG)c-`3846#k@v7;#Y0n%XCX=humaf4N`InPCbI$0Jv-nY@Z(8djNHWI(^*oT1CZ3 zR1wO$&GAiKcuWC}T5J^gtdUQ?8K1|iEW$%W5dx{6T{S7iXGKhIBFq$VJ!jiz zAJ0)KNChA-LsA%ToUJ+Xrn0j#&eW^-jHLfYUED~*Sa-Sw+V0d(7iIcEuB->8G5LLN z*R#)8RVGJ1>rGrT+JeQ-TNY@GZ`>{>F*wmN05a$s6p-NT-#z3UN;_q$5uv8br%0G6{AxAum(gbNF#kD?`)JvERrkFcfnljpM{yV%~t zD6pQ5``$f~%4COmOpX1R|8c@Q`(bSCaM-<9BO!MqGp8Puq0ufyf4D|_hDPfFwIrbY zR}FqcLqp*sADq_oN)-fb82MFlp;+>V!Mk?G6Ml*Ss7cmHLD&( z)WqpSC(xZ03+owO2Qjh|PpSWLYtTH+@m}At-TMs@d2*>utili#1!@Hm=3`HD3xk#A zfgST_YgF9cG+UD;XmuUplr@l071huhP3-CnPD?PoI1|sI_v7g7-VoJ<#z`6Xi2}X@ z7pATcjRoagaW!X>o>W1Z|H| z@$vSVnJa#c$!!&jdEU@^Ab2F?hz$`wpL+7%bH+ZZa_UaLUSwfbY3wqwIadNLhuh%0 ztqRVa9~gO{4dZcxpddX*2RS(tU5``V?o-48!jbmKjb^;=1h zZ5Z&Dh^m;;Qybh+)Q=*)yR}6Iu0guk0U97Kd-4wP{35lU z5EL0HCIOurs@->Edr@qxXo=RYU5ou#0cIn@j8f9ZrFX) zp@)r&G-2;r=IyFn>2)vUmz!!XwzOf0Ky@3gvKT1Ul8<>J)N@PM*f_Jc^*27m5ix0hghPUYm*Uy_teDcq!WO1^ut&)2)qLUXLDxnj1-B7C*vsinz}AFrJ6 z|AIIg6ZftpT35S_zTfLlb;M90qT=rK^z>Z3cyTG-40c&D8WrbMCrD7&KJXX*)_ob&6d$;*wO^ zJ9mD;a}in!kTbr8mrCdnI&Hy(7ns&a(;i3+IzSO|^Fr^YrdCzZLnN5OHru58uHepvvR*Xff4{ru{7kcvZ>*g@`?mVKReb+mq6#*P9 z`F<==&e*`9AbHONZCTC1*{(!&tr{eFszJJ?mU&-uz zm)rNRzgzv)3*gc7S{dbJ;)C~du*2g83j8{ZPj_CfKY#L)UV~ zrbG!QqP$T)4rr5FIunYEK&xvG3c1N=kJ}T_@Mu+QU?OfPM5qhH z2HeS_f3zHm`#5pZWvnZu7+iGi?VzC-L@N=qkIf`&ym|^PvkWa>q+LT+q>Pa=*ebLf zk%=D9$Fg%pz*YgpkxZ3v^ld(JMkpy-Xye_``ztuLmkU=gh^qq2q}Ab?LaB4SAbVUZ z8RRIQ??4GLF0AOgr`ew_$xnU=4x9iulznxb(lD{`$gTjuqYOmu){L}ItEpmxXUh()-JS)%M75u1B{LQZCdHBROeV`o%DWx6%v91E`~^` z(fojC6_7a~J;RQ5Bg?@Z(8FuwrKA+oYw)gQGB@q`5AanbM$sHU8aEjl;q6t-Y6WoSu9Ll4f{z8S**=H0k6$* zLCO{7`gL__PTU=cu17KXs4^Z%*IM|kRhewV_g^$8KWp}Lkv{6VLDTE)TYd#@u=g!_ zoCW9)CTK~pbv&~>lU;A`yD4De%qx-zK~S``JqS1X*hhUC`t3r1&g;mHhG`L(QL}d& z8;6=z>1-W2FO?sIyb7B~eumpXJ|%dz*GdMH;2US4HPG#s3Sml*Py<^EgZ6aqIpXtE-VXn$c{_vX@7T`$N91h$eAGbM&3>8mo%v(kwaSv?jdS z0yOV6k2~6ltoZQ)*c3&8Yx_E<%I5pQwkx}$P~@J^mJHb>q+*t--%4EPLG$l2qAWq; z%6b*ycZI!HiB~w`E4j>h@5_0C!-D8sNHjyu+4VXP9}%c90pH||crF47;@tw^x4cUf zF?Ds_r@u&vzU;^4?dphe!jk$S^fZW&W8H%lh^Q-P+!2qudNK?SF8&b+hzrtT2QH?g z))kVjd@#p{&uz`ccP&H;qBdBoV2vZB2_j|)0e)q_ZZt~>9MCj&c&kW#FzX&R9XPww z&xk^Pa`79JgI7e=oULuq!<7y%vIw~j_ArgdE&k|-HQEJ<@F)oorNdxB0w96M@PA4q z*J}Fio{WJt=s+2HNc({==V@jX?%}d?++U#GMXev>QB+jHJV5Wg`$vDGH!yO~U-;k^ zSCAMEXvG7Pp~X)rpb1*>nM+!~lJ1>u^>=fiY1r90B2r8&ie5)IeJ&9?#4#t5JS04N zpED={XR$MT_`Iak8IaSDUtJPumDc1iEyutb$FC)}ZNnVmV4Wbj97U~VvalB{i1g>P z$U?ZN;y`P0aWPnr(&4AM5@1%U{ZIaakNP-)dP}^8%mkGJ)EQ-oDl7(KA3x1W1Bq_% zDq25b4T7L_xa9UGzCe(MI5v0%DLEeW6C;myAcWS8LLPDJPK5Z|TZ%HIao%;XcsP9nd5SPmY^1Tj$+n>kFI$_ZhAEIQaBTShrY{|O! z>9|XQm~FV39Y{Q+os^n@S!3~eRJXgnMloWwdJwum|-(V6%?QjsRu4Vn~O?7w7Or{1TVj#XD82O9Gq zMSi+`^h`lL2; zA(HVb5X{B#-0prSji_~6* zIF>TG51w)#t z0b2InR$mG8Alh~ZJ&#_PH6pFn`D7EzyyH_nQbbJuVg-w$LXe847t<-PdySTbAEk(p z>&|LGl?{f~VkKUj4hxK(jC9C}jrE{h_XqXeU8Nt-NU1+%WjWS(W}eAs1$=?P&%H0c z7g8TADedu|=rFZ_yGHiZBvt`UbVD%`E<6G0&Tn<#FBxnrYdzu%46)MzT!p#cP9-J_ z^r|c+T}$y=#pcPe9{E>opv#Yfi8Q$zM=Em;T%qHBUK69u3gH}d(e)W_q(EK@O$X+D zc}}nZ@7bT5+|cZV=yf>5C{G>;%b6qDNOX4_o3=c5cG)kjXa%hX@RU;e$rU}2=B6%Glytq1(koK!h%U7k zy1Ui4h*0>#puW%RK{^M4nzy`)jbwf$LyndO^4G^1BrEa+7U}~4U-ia9ux;$%-q=^I z(Un+6yH>{|OB^j*>JOeD&x!;D>^vrm)`yr4{b~)6|GV|uj&OY+M*fRNe%LDRw0DIG zal1k5-TI|9Dyg~n`0{*U@;tm_C=(JiYh^~Km7M!8Kdn{vd)uiTIT}e z%yg_EAPrsg$?O4d*e(n`_?4u!^u8Ls~m_cN)*o^nhoB-|@w z18tSqV`>`ir=y{wqdDQox(7v*96;!SgA=Pm8pi6`@oo2mrI?gv3CznncLPeP`AO~E z&Km5j;wl4%owf8B9`@w8!FH#uAL$Axk})CID&kFldDxwpL zR@N1EUEeB^hot)~zZGKj8{}%ei!aGKKnZ+ds4y<<^ByjrmF*B4$hjACs{IjlJbOr3 zZbQ#Ul&z6{-ZHY7BmijNuz3v%x)s{U3L9UfWOB=`4E>xv*^%83ieed)Bnx`3Ef%zq z4-O^CvITTElRJ(iaL7T@08#pBj+QG1V`ZI@>BWGouxvY6P2&T_I;fLmjq>b#O3V(- zi4CxJt$j&gE)o6fzE*C9d7t`+$o@U%=FJ8oE)5P~+$=!A?PO8PrcL$ws`%LT+**kD zsc?X7uYtGxEtbyEiECohSae#X;t*IMmCu(I)_@*n0*pyrC;B?)l1h0Ps0 z>(knuRUi_=sjIW{AjwW(|J7xS4~pxdr;(iavXv}JV!hBl--ciB&JT3gW*iehc?nT?7xvS zry~gEl$?%vFo5>6v!~PvR6@=qSdWdFQ2Uv#%(bfO*x+haC|CvRMDYtmJ)a&-fBs(? zP52)RH?8ddBQyM@HwIw*-wy>zPr3;IIJWzTnKS)&?iUuRtriaa=}5r-)??ca=Q5;L z7qxYtWUXs^HyC2Vx~624rdYQ+>}DBLHPnoCq3q7nyNg7AoVMk;uvE*x>mN#2=|yMR~*I-*(>9;ysf`q0I2jgL=jrlqX=#WZh2{D zY+Czcs`*?&D@zQ-TP1>^d81N`*hI*2JuxM<5YiN9sBOfVg6MTfqRIY zXHwN5t(r;zS*;j($fac;J{%}qcv|k~MygTEH(e5um%D7S4#<~=qR>)NQQppiC@DKf zL~hm5OB+Of;0}HH-%eI=h9=px(X;QBh)6p+e@9E54V4S{KQ%815N?O1L8Ki3So6k> z^k2eZ2T6NUAhK>PZxiL0kn4d#hL2|_KY#D5r>93gogFOes(AC1pMP36_$_y{up2W+xKuQ z3UZ|tx!}oWS0&t=oS5jywA)bmAKpS;4+sc|HcnUK@&j{!`t@L3r64@1xsu);! z>~Wuw7RHO%%h9IUS-O%i8~3zUWhfkb(i$?YPA@F4H z7uihP@(b^a7N|LQZo_9;W?dla zr#EK#48r)8u4%VJnTo#1D^SqP-H}M`n+YlIpLkIsD;>cu($Jq!STxa1Kyk!)-dP8?rKAp8DqN zJDm%k9x>(2{?qe38Y6mW(Wwn#vM?73oh4<{=s&A9OUbU@=;6GhleUJEqb29R9|A#J zAL~Bt6@T~4i21yTHQa^81b3sxwV0Dv2(3m}nxyV}{+gC3HQMwD)k8ER{-4Q$&at>m zzORG;BW5@9YXcE`isum;dU$+x|u{q9x@d<5*^=o zo{r`3hLf_gve+dqECy7y?eI^XuN#9;S*9FhYR{9g>BjsZ`~(C2{-452b`4->qwPIA z3Jl*>;yxo(hqor53|xAo#Ff2M-a|992a_PhM8dmqMO4}O%f#s-{_39$J(*)*Xc6kP z|F#JPmh%mgU-(P@pUP7?2Ju$2?6xlG-N@mmpJ&NS7ptC{WyhjX3}orUcW7t@w=AAD zt3-vD4VJ}r{dKPB!*g(+4J0`;<6-&%oqPXfrGP3wSpQC{dtS3>njMRpv9qHf^}qNl z`dpRfZPb&&%64+)p(o$q`*&=L3N-af-BTf;%1i7WvdGhqhOHk|!s9WT)S zQvo#wb(#@MD+goi_erY-btU*u7M+}&tc$|q(r$ocxlAj-vE5)pE`M`Bspy-_Z0%`O zAN_K7j@2sDtn~udkO@YV3-hQDJ~V-ljhnXm4Y6)}|^zne8ec@)_nA!Y7yy zc%bXzoD+QShhC{XD0UTX9n-q_!EtE1Nn<82ptc{-yLxOJ3;vgd$EV5rYw#ZMyN-78 z40OnEsPr-1AO?WcXlS#VKARsEFVaq7r|h`3_K|M(six-{mTUZm=HjMap4x(#$`V2q&HPyF}s7`%m+mKG~kw7BH4lZYLl)Baeasz|FD zWBePwa}fa=1vyxxy}2OJrL8(=lD||BK}v3{%l4jg)w7dFuB%NS{+)ZR z920DoSY6RFfk-C*ABfmHYmhVNpMO9cr`tE@e^Q*@>Bsw`%;TZYgTHmXy_c6aK17jS zmhD*-Ox6lTlb&n z#Gs~_y??xKOXoKqMEM_Xicnl`^Uv=K)q~CydAi`B9q)y9?m;*>>~d0V|VOe}F$Bo8X(xE~koRXrxSomVS- z*yXBsmN3b7@Qd^GuXkKu_T}B(IZ5Ifx&~jb3)4swjTx{Av0yW7>HhW@FYh#ZCKY|O zfu8|)bFo&iWIJzt$BwI=ci}zcG5%MN9j|*HeDmd;NFIRA8N;o_nq;Q^6)IrETVYpI zKs32>=iZHNyq(yaMgDdGf@Mq+iQz6|X7hF$2{gSt} zeP)#6%P%ihT*-_Fz=pkSVak$qmwB&Y-9cCkR+T&VZS3P;_aL~@mE~|68q8PwuS+Po z^*3#lZ9Myl_d-tO&@fYezVSPjj|d-k=1l8=TmgnFm{IlQ(57VTs@+Xj8+#et>d+WL z_R6!TFuH`jAY{tFWoS#t)wl$&KIZ&}6p)1hu>bZuVC)shpPL6~CbcqjRbFasi~!!I znLce6xzS)sq6O6CU%VQcgTX^nkG@w3`18`dC7*2r;gegtmLHZr;mX@KCrZd)14asx zKWryZn14)Nx-NsiBFBi4gP+u|^VUnN9Wxwdc~|PPN9Q5Z4!H;0cilMemyaR*dO)X_ zmX_}R+qSWn7a|eHSJ(Hy;gwa0D60&Uqu+=>|Hymj4e7@4p`RTEI9x@y#9QGKTTBq1 zyzK=D0IX|7*IL3|$%s1A_lcQivIIqy%wQ3vhn`n=sgC+pf%E6cD^Wx?OWptuz&G3T zEs_=2k`F&b5&8pUm=aQbNK>##nREUd-ZP1RBEuT;<#kZh>+#I`Nyqp7epM{BxP$yS zq^Q!ctgWvmjMdy(l|XM!AbhReoyoVO7rmn!*l_*};()9{45|_|F4HwoJ=T-0NqXb( zw=RZnp`MW~APklX?KWW{VcrW<<<0~^pOR+wgQ_MEO}ctElCYh?F#zOwn zcb!}+$=T)wgMaItOLF-@T>X!9JXg9vWj}=>RmRteY{xctAN#|zs2vm#KaMoz)}JR$ zc5?u1b-EZvNpCrIR0zjH0(cWX*%sbHG8s5kXZHz2cz=wYrwmQMvl!X1uTVQgW7fu_ z8WCb2o|6S(!}-T3Xblr}4K(#|Hn;j*I@z=6MSwlG^*xdDp41ckJ+`+`bNr>I8GOW{ zm23T1@ZXzlj?Xl-%kErNh2$xG`5L&JjPNiV_@ILU)5kn}`XFejKqu0Pm-%^;UR{x` zX#UmtG|4C6coAK9i0J3XlDAc_@yJONDS94g&80S z?YANg+pwBw7Ko;Q=|F)up--VcaanF?MAN#WW8rE+cBBq21a48%30KMpimv;K7kS(3 zAt*S!->1i&&V>v`_5hMjb%2$g8mg_51-GKf$m$ z)+}3O5@(dU29DC=H8}mDPNIj@Qh4}DprWxx6n0HchlaUa==;WkxvTtA_S8$r@Gtibnt;P zj7jPe4M0ls(}+I(vJU3rJkzNm(^hy1wgDrO@@)!1_$$4V?@zI+DGr7+)ggoy@Z&vs0@i0XdHxA7_ho5V~RL z&A7)&taq~~FE%!{^P92t{$i|YJs;O!4~#hn~} zb8X4&A&rKPiDH=UNxg>#hO-vZi><~mrie*=RX8JUaIKO<>hwmu`=z~EXup&|aTeh% zWw%0F^2F?r-@RYs*w6;so+!=}K5v%e9zlTuU?+}OyQiO0p}RFZ7A7+q`2ZgR0!BUk z?c_*;=2!x``N;|pgc~O9j;6P09;yA&gO&y1Bb%u}=H9X~O+hqrlMa54Z4?Oc&L)lK{W{hi)riv6l^p4USkwg9{Dv0vj|yZbv@i!N?-{( zGe_NoO-sn>J+B!;!H$iru!`%{)FfJ&2K31#+62GkHWnS3;D+Ksi-HAvfRQcu4yYcp zRZbm5$slN}U|ai43@j#%I9wH1iwBsou1;J5_uqB|0nHkq38y=vBLEO-H2h=s;mIMY z(GQwbY4E&NbBg29c|@`RJ8puqHTRqbygJQ1auZVhq3LO=k7)9Qh1A@k&=cdI0Y-y6 zb4Gxw_Ss2a#dC8P5l73QLZOXmg*$(yie+qC(-?~Jdy|yplr$XLK?l?H3?Cr7Op5nd z3J192mZ5FxGAg(U$Vjz#TV&uPODEgN0luzO?BY2MZxADNU`il>hDVQkLxC?W(wP#@ zxV;C>uSMv@wuz1eidmL@Nhlk<1u4HV{11pPi z577GHL(GYrR_?!VW_QKa z#hz-RfZt6=$5C>{7}&Go$}4Eqz@E9$IT4qCT74&~TKyOuKQh^3Hz~#44Db|Ca$q`N zk-dprL+a48WX15~mX=km6KO6zVRPi3{Y4!{*hhC{80F2;{<)AEv_IOW?xk3H+V>FDnC^xWYx3|LjkU1I9f&eaH z!)QueubNBu4t96CcRqT>YPP$~mZxwmjUy&&I~))#a(TL=Cn|C5%i$D(^gc!A3QqCV z& z&@kr&mZRwALm>%5O=`knCr6UiH&BW{J3Z6fJ|y&JK<2J>Wbp3wHc0FZN465^>?VZ0y|x=JK(KFGQh$Vj$w58;%of)hfmA z<94%eH;@H32BGOFlA+^;Dy$_X@ma;IrxQg1P{XmbERw(y@I6Z|*323>q*M(yU4A3wy zw|XLTwtzP;Bru6zb9rTWEb#Er-9R*Lm`7w8`^``tbx%;&w9}fBNx2kbkhF5Db!t5P zMJ^@@3;|{OvJ8iD?99_*&EOA-mU~$JqXPj*?L9-f&&(?4{gHY0{f}Ea#Inzwn>E$9 z;9Kt@vddQ$v`X*CheOJ zTz#o-W1b;ERTgkP6{54b)JYwB@h1@|#6$w(fb0wcUK3HeL$nQ?iqwkI!|%~ni9GSf z03$m60J{r?^lLKQ9Dwz~T0E;+VzpJ$)A8Iq$uLTuKsMAiCq}q}wcsMk7u(+Lr>2h@ zJvx<&?0i1cW;nZh0Si(GfY$3Qbqoa54*mHOv*>Hc%Y#%x#`igF6=+o3?LD?YVS48! z2U5z=z+iiO>s_CeD+Fj@Ok=Vn(MxlUxR^+2Jw03EO|Hi9Q?^GJ$^i}dboaReLQ?u; znD+%!ZT7K$iulH1aDr~IgR(lm|41`V9-kJ3;|VqmKg&WFL4#%JAZV~T`N@h%{0lT= zGn)2xW|I=@*i)}g!yPoMgoHqotf(F27PLnr-HJ_K-E+kUlShT*pGD8ntz_uZ^yWHQ zl&E)M&%J>e;ZTe*QFl!Ls$~7)8d4&IJI)O;RJ!WM1$E$Rnl<)gNRH&pUgI@BLgQ65 zL4&esUykG8#--0{FKe{LyQUunyqAunK;`FhN&(uInkvkv=Kh`iUQw{hCCV|H2aE@pwjhm5$ z?|fH)?XhEE(O=s7W@4;*Apji3MQtE<%bc$(|jLwzt|Q>tv*ReeS1eY42eRZyfD`GK#K=%@Ms$ zgGaz}7BpggA5A%p$N(7!W{;(3Yq6*&>WpVoJAPuh74iqo@x3^7oKOO_&|vSavwK#< z(q)7;KXtsn06XEPu*`n0s(@Ac+D*Vf_w_7*e{~c11sXq~G>U2a{p7bLRD#?>rpz)7 zx*^wLoRem5>Wa^kF44?r*Q5YR<(@$)@Du%N= zS`;`{#a;p4qYu3E(efu&j*+D^6;ux$y3*NKOTY>q^LX)|V^fN2*EUhd@$qv->KfX1 zu^2>t@tF+M+W2<4w(syNB9_)+r3g~_buPZZ3GMM{^?6Y@6^)q}$|<0g*&U;BeRR20 zP!2~^L*n{W5|P1fDq`ay0_zIGMQ5zY1+eze}i>+s|VhjlJ%t zSkwzHb)p^eA=XJ~AKJ5Ev+Q}pG&u=B^DPFa11`=NZLe)60;l>-iPu<1d#Z=YqO8I(o$BxwPQCjH&FCKT=DsM- zB=0dL=IiOlvmwNOWv0k)3KkL>UnD&y^Ar^FevFE&#K2^Zgu^9ioC!&%D`FpQ$DI(g zL7}3W3QGg2di5RyZ=*qHVgC8GiSp0;F&;!7W;&$}bXkH~QQ~6|uCq0IPxI6`v@JrW zpCQ|=dsrzRuf}y2t#gD&qDifRqrIe79p{Arjk($EIW_`Hp0mzAj&di?9@ zsvd&ks%wey?QhfZR;5mlj$Xk^9LhHJCaMm!0QK?qh+px+?tR~6EPDh*=QEknNSz(g zLoSE6QmDQ{J@%vH{(Ac;5y*vW71YzksW$W@aVXJjI2G)nj!M#R5ot8wLLwqPAm$G+ z0gwl3TPS$f17n`tx`CMv_c}2%jJ$hHzkVdP#5dx7;UsZ`cTkTwQ}N}e>GjT|&*yoS zd}Mbb8i2B^KC-#?v2&0|%rL*n$(gO8yY8l&_gfr1O6SUjnN|Mw8B@@H@8*`A5;Q(K^x$vRBt&H29VV^x_LO!Vo*4QL2JuH&yj@(0MlksOEHO*CSXh!YaB+=UpzHlOauYTcuDu;eNhC;uIZI0)<2Cp9XEIxAp#vZo`c)_^`vk*f^r|HNV zCUb|ks3!-Sf8g%&PB=rdVHt%$q9LjU(Z(-wB{^@l4tF>qKV#&^9;eCH>c>1^@|O2< zs%jVypYJq*VX=b$(QNR3}YN_VLMr6=u61MuuJV{ z9A(D)-w41g)^CRJ{M-&(W4@gfTE2nd3|aK)xLEE{t733G{96jmk2fzs;}^~#+ z6;K_Iau098489AkRv=%8r`ZnS6=)zv9xD8$TEVY|0mN;zuJ2;S$F5fd znEdbdf%kxu{*Ul)SmRUq*R16{myf_$jAA3saVf+Ukztdq0PkMLA^}xEmG5MUp+Hc2 zg3p9^wa!55F^_*<^xwP;m}y`5FE2ysu#X~)M6>u;dnIJDT%8xk!`$=67<3c}dQOuj zTOYm`!|NdqIpVU#FVD}@0b6yccK$W~Ld<)uu)&gVCCM_#RbXJ%mhrAvuZKlMrCw>W zG>4yDGvXk2)8qWD{f{jruEHk;%^Dr!30sN*-|V4-Lo%tLv#8%{12>MR{O4yf`=gha z3mrJa_fs^gXMz#r`gOMO3?}aOIg9=IFR$m3O^(%U`L9Nxo<~lbH(>+edG_NG`R16; zJ^a>zd(a3k-v9Yo8TP{WNquv5Px04zPYoxX|6nG(`~C2sf|Gs$rkIQW{EAQIzn1oK zWMt$Iy+os93`W*|YmT=RSDQZ=2is%_@`-d_Y+d?|8+RDX6|-cB{QnR|Ug_flF1Pu{ z=nqk-B@1b~iOfHy{O{R^HKvu|nI;a#RnN*mHCb_3Iwy<&hBi@u4xc)MU>(?Ryu2sM z{5Spt0xPIJ##48Z?LplA5wSTbG4aQCaGi3-HPfvAvBW)wo$hQJ{#mQi3**FdhQK+c zJK7wa;ywTNy)~&R0FZ5^Nv0UP-e|LV27moGTxbVZvr@+8m9nvTZ*q5I44^M;Yi0j7 zf59eC;Dcc-+6!87HGF~EESi$V*6oQ;Iz8j7FAk?=@1VmY=^#(7inZP%`Fx|{pGmM< zS9X0eMx%UtVedxEuRgoOVu{J1#I0yB%QrNWZJpWn5pMs0m zKGb#Dk~0S4>c@6ygN``*XpHwRwjaRTGG2X#4ZT2)5|$_TJ^1{m;)a4N#Dh6%u8`2l zuApr^@0ccci*EQX3fZhYH1muL=s)fNAkMYcpq2TzpO9FzczeA+%c>My$-u(R#@fbn z0L!=z@c-<`>CQiul~hgZL3gJ65QoB-KeC!8{?%p*yA;;d0;M-JSsPghu{)3dNji^m zW1YT(Yl?Lh3#KM1CelMzO{ZyuBYz=`YMvw|yj}|x^@%R9Ltkk8tdc%nfWjpzR1fXR zIe#(rLgko-Jp3Sfn&W!j!(9-CBzU_!dcGhXZpkfC|Nd!5=b{YigW==xW6-?QQmKRbGjR_}i&Ps2-ey85$%e3#X6TrnqHqXyb>}Gn_pbiO!grD;Yn`E$`?^ zVm9>%I`Yk#7ou=U_>9p@zr#r=wTIYVAg$&M;gpDXhuAaIct|&5ZsJwTV>|`)d~E_) z<6=#mQg4P%iw=rBUn}xOmG=VHlnh5CUA^^E4IIo1+eSL75{crM<;q!BT$OM`&6x_- z<3c?!T$?Ntlku>Y zG4kN}2?H=o91Hte;r&lv4SPO@nU&kRIO&bkl_mMlWI@UGX?8dsf!HgvO4oC^ySv8h z8DA$0W~oAi0rWhy{w1;USh}PRuMK?XSbh#xX^xY-MXv_TL1#~)JM2mm`Nv;hf)J-e zi5&Nz6tNH zI`?b)n-6LB_Q<8zUifx?@724f|JKU*X3HN7Z`@jPe|OA*$1O+K`W@YUe^19{rA=jy z){nJ=RsD8b9sg^cbLmRgMe=(WtTuVP#Ndj-lj%wkZ^Spt$;20QkByxfopi376X^M| zZbo=7!d6lzUkA> zE1f_7jleEwMxN<9IYv|fya2N7A#`iWhmXv7%-^Ix)OnT|bexaci2>lGq&rKNt2VqM z&^zQE7z`@t>Mg-sLs znHTEQIbVLW!q5r2c|+%@x4AP+8gqCiw&m9<*dL8EPYubXc=u$o`x|hkCevBtUKb8f zD%TJdeZnt9-<#>osD~RM%R}ZormXKL?-%=K_J107rQvuE>?>vIxloBQFxtNW;gxdQn>P0S;A4XmE_p1-I!!D^)$nZW&-K8mK% zKF1KgB8<~Gc?$}DDCR#houcYr00aB~YVXXWqB_$wennd+YNXQ!;~0f*g+U>b*fuWE zK-qSJMMM-q6b0N-kVV$WQV`9FhS-uRAS7ym-2nj+N|CL=2#y+BhypGwRUjD^Q8o>$ zWwG@0Tr{0C=giFMGc*59{_@0Q)vbHK`+e{7d!ErvNy+Z-7iENKyuq<8{r8eVdH@y5HBCRySl=2igjU(J~m{8xBRj3Us=m2~xT6@s^@UTdQ z?g__ZW9k*Xfidh_GEMm)O+d93#3#&XB2Me8$ZdarB!?- znII=uiA4@HPLfA$kkI)|eeI{4YKi(#kKZgAV8h)gjY#FBxoGTZd3a){>C1;|1s@4E z(FzXaAO@)466~1t!Gz8++&5NF_ubM<|BCjcsgw?3$p44lM?qv$^NMjgM~?!UDGKHH zI#7y189oI?4RVs9q4y1!ef9hp|8|+cYL%;aZ*`A$i<2(8 z==8(V39*w^7_icVdcnU)(S=MvZnz;_BG_D|WgIO}rN2cIPs$;?=ZIhxMhKQeMdV_! z2@Va}zn>eMvaO>oht+JkF^vE)BA~l{TBbqK9xj=GWJS0*X1SQTM*dQmJyQmQ*xl$# zN#*Qs5W`5apHQ<)p?3`h;qG0uH$=B)j;sOqV}h=R6E+Caa|AYNyzJpoZ)ZVh zkG3ofTiIJgTGsZt*U)wjy+S!Vx^Bj)U!k5ZXUl+SLND(5ZrnGLY=m{1eY|okYhJuI zDS2+{g-yy>hh;OX8^enHcuTtK-Vp*NXz#qpp!24pP9Lp3IxE9PO1*6u!7z)Q=_-}V zqZuWJV90;MUsQb%uGVw;H;QLb>@12LhC;XfS$}+X-K;&spB!>pSKWBlO=2?pXxqNX zTN}!!1b#u6NOH*NL7UL@1;u-mNAr*p($fgK-tSSOp|w+?uH_mVV#sLdL(5heeP(E| zhM(KJG^TecTfOXg4{1|J50nmet=Sywz^OPQDvh6h6W4hGr~;9#mK?7cX9^~|+x`0w zE?0w>bQr8%FiEpqlTsL}?-_fjiQZ+TQ4Y@u+en9W+)d6bAzL*AaZnd0SdAH z=u|A;fB!0#YV5XS@skFk>lmViN)`+u@7a*QP?9b>G+h9m$2Kp2F`5|RrG5o?7_kKhZz=@oRC#6T2NZV}Df8qaLTQ!gx8FTbet5Sk*?d=NOiN z^8TDcBP5S4(7B zU-a$0a08Te>drR(lWO&`(Xv2vnndlc>)C}8r?ui5%dboL`KL|(XdFZ8MWJA`LcCjR6Zza~roXxW^`;Yr z{OAui73DPoCE7IyVQoYxWlZy?e7(ZH`Sj7H`EP%A z-{OM!3?mNC&HXs?cAddRAs=DuRwV;3wga*W2;u{|`LCRZ$#1Kq{m8`>Zfv&D7aeDJe7AChL8 z!nZr+)Dhnw?&TXi)2m6p*?o8k;y*F7$pa}dtw{U$t)nE~41A~5On32pJzu7$c(S)` zU!D0W)A{#j6)*Ms#YG|HjEh+-_={C?0RDiy1)= z55e`^K zJH6_C&8t-(_-sM@-=F4Q&2eY%rNP1vu|;SZMYWoHJqbEN*8sX6z>B7K+c0fv*t6Og z4giClgoUFAq$V0Ymp=eTfd#GmNH{Wdt*19$VzaW5=BKiy>>!s8h%Z@S;DR20UFIi; zemv>tq{fn-7KBL{)k>I|eUI)EyZ;=xMK>dMqJzLQGH(p%9e-gS%1bv_>f0m|O(nBL zLthcrh8W;NmFe7YEP`-LOmrEYkZVqag45Wxd5rB7TjHX)#!5Q2+IVZ4xliD~*n>S@ zyZDFy^}win+;$vwtYK4N0))*aa+$7~ss6}NP8oAaAN_@&4PUvm2nRRU6qz;idxJ=z ztcFsKqw{gCWs1-cDIRa-mr;i_0o~gbX1D(I>g(*mVp|cjWruX+dVFHOa?}Zr4TN*H&Cw;D zMsd4gJmq(+al!i24`B|cynjT9Jys0;ha?GNGVCUyFD#+1?kU7l>BBsWymta`dQ&ujlFg3OasV{ z`5Ld^E_}DP8Oj(4sOEL~o78~%;NM+Gs!3HCg*009;lIL>)>YR~7|@bJ+fz4>P*-ZM z=3?j%WgkGs@4*lxMQvL?adVytBmoCcNBhWGYy&`BxVG4u*UHzJf{0$J%m zaC3ta1H0kdb~hjr%fqwt7+9aRV`splT<6}}THB39{vI^ur^mDd&NI>r)H;F9YsjhY zrJ6zoD+@Y{(Zq{H4Xq%%Yp;wqFNhan4bMyvB=I5W&Lznc<%w+HdH9#1LDMsUw9Cg0 zR9`@1M}4lx<_OG|y|-18 z@iZyzci5tCLH<7I|CfRv4MbXbPu z`}RUsb>?C;wty$fk#=QCR6O?QTe$hwb2?DDcK|PR)uJK-fP09_&Mfq3XIGP}kFBk2 z3t+7EiUCU$lW!qTvhK>p*M(d&a0^k(fK{ZUEdwAd!9nUksQd7+i;odv>|5Zy@V@|x zs#^<%@{tFVJ%!jqHEfA8&{5Rc;oMe%*tvkCg!Hw-Q&ECceCXcOVb6?Vk?>!(|GtHC z!^y?oYp6qTY5mPkov%e$1wfe~7@>6!+*K%ebaT2@n6l_ndoKZl5!k7fRLSs`G0oeB z3fG)EFO0(@yjS#{d)AJ$(qOmGw0pYv9EtOXxYMnhhlRrPvq&mX#scuELWf6J-rQ)Td>mGDLq| zgwUrg9#huCM1;!F(m}2=xSfp5m%=po3z8?;Y@HAb@}HndPuUFa9tM4IUc}I@xE&Cp()Z_26VMk$R#T zKGbVB8hk(p#aPT=5p`CjJ|R;J6;@S$%(F=`H5RLypdqFIz43MW@$!=JgRz1>JTQv7 z4iORwk*8@Dn_k`5=f!}gEI0&`bPrNwfFJBWbGR8Xa^vgh&J*;@_vIi*(Y-XUcSM8U zx)Xx?Tc_ihepB6m0>;;CD%O4$JwbR=VxTmfD4d$2>zVO7TJ!T^7?*WLhA`&tBd4Mm zkOiT|&8WpyYM`+|nFLrum$Up93VfIJ+~3(a8F|houh0C^Iye9FlXpx&V5ypl0;KODi5M zOfm~U8WP@Zw#X>mSrc6YqWl&xv-)jzfGFPy2**)el${FWMU@N{l{o{oP>>!+8Tvr! zG5=cptUJdy|4xupKOB28NWc^5e0;7O@S<6AHRWO`zhE1{v)?U-MZcIbBR|U*p#=dT zW16b0f<==%DR+t1-n+&y^MrmI8MfFPt(IYAk+Po#nQR z124t+Yg`{VtA@RHF@xbTZkOV;q@y%<{xowD@nJ|h@lMBAY#e*MS=#5tG&SNxAM3Mk zV{ni-PO)v}9e1$Nk=Uh@jYYqDHqNiu67R$=7`pt^?(TVfk_tCJwZsas&>?Gc646TQfVbEBgpO*c9g5)$O^$$qfCOqn znK-Bo@yor|ndYJ51GO|Vs&`#`vUBP;2|KUF8Tfg6;ayclfmsS^tN=%o9Jb8STM#Cy zrGH%E0#q{8y``C}^bjgKKm0mBJKMyr2OpeQCF7+VcNxA}fTbFwf>;k#M^+uQz6cly zA7Goh1!dOh==JMw8%BTGt*#F~Z;d_Q%K}|(pLk776IcRTSkndCI!WJ_rm> zl<+5jxDc5SlvdkLv<&SgILmn}?EI4bTr*zhFlSz~qSWE0y)Sixd8zTolCe5MtWw|t zygT`UwyDjwepht(z#GEasffI=M|XR+78kk6yB+_!X`wE{BFw(Se1-XswulfO4&Z;^63TfFO05|XR|9b45JlxiDo z)tn2wicXzGl1WoH0`BHT7++_Si!w%G<0a>cRpD*^UNLOlR9Kfc7b(0 zksw-q0%ULLjg{##vj zsV4+SwlmvE#x{Qf;5chCt%DTJaf(cjEdeteDw@9YXunfg`p=MUQ6UMAI4f$j0UdDK zrqNrrQj-b|P}g_sUL7ZyX>|{lqs$!4;cqZ>U!50hi#YsC;;=aDI`UE-INsQ&mId^s#Vf{HNKg%NSD}iO@EdrC%=<^OJ7BWmGu;z99`Sqz!bu$v zB=PJe(2W_f-}wst2R>cQbb|Za3hF(Xb|tlV1EOFqts0>LH^M~gPG21E+)&w*!AJ00 zxo3gL!gv*x@pV&harBSz(ERjLnsCStee&gs6`cr~sLt$(A~{5TBaNhJT3p(xwATHT z@L0In!_@?BR_O9DYg&?-hy1m0tmK?{+m|vw;-gy=E2)aAP`yX(drXPVQ zH)JG7BYmhNf|zeol!6n9lh9a^QVs-WcSF{A6De*Gx=WArN1y30Mze@yDGDm$OgBbJ z=5D#G5%rh$+rqaB3v+WGUe_iVjk2tFpbYnG0Y%@mG8+QlD9h*q^Z%|ls)nK=I^s}z zlEGK*20d1>M?ZY4?D_koGrl=yb*AdPOZ&_xH=9DkZOvHl=g%C{_R=PoXDT%>78u~g zaFeVy+bI^h%!mBlxyPNuM+#vK5OD++<)S8Eu867ggo#n>8<#((II|js0*+l8s)!k~ zFE~727&1c5MF=e48?9fDf3zWT+1gczKo-Op-Mw>%gWvBO{z|AB_a6)Vl{PP{+m)!V k{9mNJ_{@K#gg#=l{!i@zwuv*H4(Dc2Hyg$f5Xig1k5MT9BRJnrWdLph~>NVyhi&Ll)h znFlAMi-g=Fn$Z*~VjhZ-VdmY>V`iB1JN@3@Iq&;<-+%g}PuAXReb?G+U)I`dKXZDA z?KYeYUPekv3TL_9oFFAN1Co-Owr}<{(Bk!Kxi|Q4?v5XKS%Cj_baZTNYX zX0;Y*l0LfK$zMuJISTzZCAPEwpp=xVl%@IRT~JDID%(A2&eZ9(kGSFQcnbSZYO1PD zA$xRfYrC_vU`|coi_g6F`|*UQxucx-+E?NUhjrns_yQdqNh9myj|>+p4WkC+BH?Mx zUu%XPA=48Fx){pR9vV;996B7m%hHu0`}6JZ4$RllHNSB!mp$WmCz*mXi}7l^3C+mV z*K21uJ)DEPu!eR7nsW5w$(!yJ%JzgtJy^=ndb7hiGqY<}-FTu@N|tK&rK`&bD+6%a zQZHB+`?~lq&G-gF|lU}CNQ@lB=qCQPTE-f9{uPC#V*>QBKn0HlV zB>4_K{S}oQ!?-MtYbWF9vuGc~GCnY0AJr1`F0qIt=hC~#UX9$F?K4Q zm#kirr&2Rh6x-@Et|2B+Ij`~xtgTWrOVmurxPzE%s;~=mGf9~%LdY`P%yw;LwhxpB z87>!xO$x*=xyWr|EPf!0w`<}RSy@cb_>i&Tfm2fdBG>6?fIcn14mRy%cBhmYrDQFR zt+UvRhTxeGBeaQAk#^S4wlvFm^soKFZYUNQikN zi)I4#X}==5LFK`v@#LyF-gV4ew%Yn>Sjs_V&xki*7rQ=@U3g(JmP8oOW4pGwS_LWs zT{p|3Utz$VzeFYWyPKhw4KBQ(gfVW?V7msJSJ-1i=+Ra)!0d(Ivn|FEM`E7fF#XCQ zkhHcws@Exe-hIsWf(B%2+Q6zmHerSz7_;b@iTa*q%aC1-c|)voBaXyKt`H$Bq*Ayk zZj$=_@hrRYD%1gD97|gU&w`;;E)pZ8P`EK4$6fcr&D z)%>vexpSSZtl%uk=7nc1qdt>#9-#qYop+Mt^Te}U@I;b$gwJ*1E$)(5rFudMze^&m z2S^tjC6Nx|S;224=jQZ}45Ka1l4xf=p~tf|B`0hhjFMZKuXKCX&-))wZ?Z{%Mc6=8nX`U;d<>~$h^Q;+0+qO%Z=jjP~R6_L3PAjiAogWhyiC+KDDfaA*ir z8yBupL~GcF+APA;^6n=v^x#C5G9K?!-7{Im2-Ir&AGlV-!zT)QNTcS*k)ODmMz|gZ z+)PjB{)Zv$A)ViI-?EjKS!pcNm*JGPy-u5zH(bN49!&fhkvqlU06GU=~|ByqDZrrA|^QD1AgtKqJ(6YA*{$*bp#Sb3%jHltJo!hjtnGj)yT##hoS(B9vqe6fCdA< zN8Z>~&Zp&FC@kn`TkYH4797r_IjTDN2Gt^^+m4%BPGSof?jc>-W^uWZYeQ^NK0o=8JYD%hP+i!}uY8h_d_tNt+#X*Y0V%pBOE6VhGU&hF zPb&>P!CQ28IvPNxXR%$&-=>t30!&AQO)Cy2B77c~G!npErfD#C3~Dc=#-L*ppK9h= z?9hl%F~)&|JxJ-Jr#9h>iGq(q>+sU3n?G2Ao%s(jPXyTb%hl+izivH=#=iP`q(;7; zX|h%E0UAnuJRz!drLRin1?()prwhkb950HkGsi*_I3J`5#6!a?CU3_2L<%|OiddaA zIM1whk$H=3uwkKR)bebgZWX3Zp#eE3|3a(|9CRGZ9-!_trtXO@Os)tJt2>A%uFMmQ zg7b7uAhUPB?mTQuT&6hA{wS6N^@E}nC&ZG6@hl}ZdfVf}YUl>c(#3JkKkFM>moyu7 zh@}m~NcD~w*SP_ao82o$dZ7!KT{B^e^eBK(71bh^rP(loR~Ct7{TdLWC!{D2Mt}}e zd?g-Ui6^AJf-JV`D4r!-DIS3W8a&^t!y?z50kZ#Oie)Q3A=An}u~{9uaM$nzb5aBH zZUXo?o;5mQqAeqtKb5mhG>6@~@E1EsG#>|MsSy7mo}J@Vb~Q(=qwhJx8Y&lgZ6%tP zI1$ao%<&S;1LhIrfWXBQ$k%2m6Ss?*@2D?_HLGMlh{1=>GpqwAz)fBw?!_WiStiTa zt~-ggV(E@qO2i#v>a+Svxb|XVqu4i7PExqp^eOs@7tr*qFy@Igx+Dmyhafu~X?32f zxh=Mwri0}2@<15b;P-dy4Xt&SX2xJC7*37CZDmR&SG`_a4gIQxWt--!)!~?`;19l( z>Vx83S|&1*do0kdomqYB{v1(mTz~Znq80R%*T9^EA_(hB%dBKMA%WqymE8RoixTFQ z7MASLJ*<6X9dE5uqSDlz)ob8Soh6mN(UTP6XHdv6tS+g2>A^^H0Ze|0%b!qAxONK( zOXRL9p}KXL!(oPV#d&yleL1gR@nxRTrqoph5>^BpV~-i0)j6Sd*Dz(8*tUQ`wz4&+ z;&jWqqnx=D)h&0()VI~hUrpcWRWd`A(GwQpSqIhKYrmD-z7-F@l|sMOkH1yJ#UbMV z^a;WLW*UG0kRpMirMgA<#*wst@bho5hS@^-Wx|J!v-v?Ajn|N#JYQ)pdg0iAQy-qD zCVJw~a4^jbPSYkmdTt@+u)i60f*axsX-{g=ht0$?%N(?ssoJE9=Lf|MW&qUSE0oSn zBE1bEMxPhHw;bBz1CdVuIkk0m`nBT=d7jav++>4>PV!IDCOBkU%BbwR}+XREX z?+KE8-KwyTUEnl|*2>LZ3?svN!AtwL8(J?1=|cq53*#$zEQb|P z#2!o>ISo+32{+;b!1Mne+3Hfl@nsy*9+b^*MW5aZi&~L$ueJUb8Bi1?z%ks|%9?rH z=hPn(gEoWd0*`jgoO8#?-5R1s;8LtnAKrh-x)3Db(GH}kK zcLeR|M)W<8MsK2!=lIhWntP!ShmFkcUC_3tJ;t9kCQD?3E3LSZDhbmd0aM{Q4g*#e z<$j~=n>*+WWVdf2Tbc1w_l#qhBlLiHM)HoHVSkMo?m?&3(yus6EEH-fQIe-M*+m();uGscnF@m@fJCERmPMEOAK z{RtcY5rX;QY>vrnj%%P#lu2w};2Hjb^~l?=fX~ zpbP^zE(FkrCXb{gn*$8>ie8GK;cbd)c{xdrE%IquaAh;G5ujyQ+5xQ0IM*?4@$T|!j! zGYl#ZK;d0zX6EBi%>e3bPUA8)Ayj}9G^;Z}xp?X%mhO{Cni6q;#X(H@i|&8+>BBYA zRRHd^c0!Gm)Pi#KTL2GQ4uGENNHf5?{vw&(^tkLy@p)Yr-Gazw7GXT$ON*9oL~ahM++fghJlkoMWubfZzS-`c2LhW!oIRha*; zJ1mUizCw!C)P!WZ42K+O8kMHCg!fQjoJN>2Z+ub8SENpsbO2TN z=Ek_O0h!44adkI@>TrzYJ?Qv;6Dnk`fv;lYa~FwJAu}(cnAe4v2C^K7PGell>po%y zXwDEew|@^)8(b1l&Wslk-1eVwxd~H@(xZ*%l2@c;U3VywP1#k!h)>9P;R_x7># z41hTu@xJn`o+vlda6aPFRJCn6`8m=j?E%d4VHmCE_!p*h=d#|$g!c}8u5LH$juG@M z_c$I03Y=pfKwaCZB;-zqt(JFC#?p!L8pv^6S0{$eQ+%mRuPW?y_2(I!A88ddhNw~f zV-oSX?2^%8W_x-pp2c9V)hjFxgvy{*q~==Z5VB@BedOIN>(I_{kFX23zr1Oy%q{Cz ztbgFo{M6`n6=$+JC3o1}D!+@PHTK1s)&H2B7v&WADbtX2*ees^({X))$Qe#Y>Kv0r zYc*#)ZYPe^Ms&{QguYpA*UncBD{XB)@YBn8uDBw*u<-CY9Mc zpRo10r*1bkcE4b|a=-J@FUiw)j2(E^+rFjSh{6R%GVHvd_Ng^s^4lonfCbd5;9zo| z{`;343rLAZTEGf#Z_ge2)RSK8DP#kao3O?^X4?;L;nNMJScl?B>0}uKwdwPJZZ%m%;8j>jc-;WhVI@c3(lW6o^c2< zoVRHxw&O=usIZ$_7DKZ;l}ONZwf7g-LitIL-pMng-6M zVQUnPCft&A#5cZ@{4_Uv*Jg9DMJDbK2`Qm+0Y}&v=AZ@-| zD7}@93oSpgX1BZ}7JGoZBf0NahMxp3{7Gb$BC8RDxgdgZf?}5rp!>(%VH^g;xELUD zKq#RP-~M{1uL7&3F9{4lQr26fZ62$|AnrcEb`+av9Q4SI4F>2cv=PTy3wC&&$ehj4 zDLOg?H;L|kVfRQ7bV}e=z59l0x#%R5=F~0c%Y8jIBr=qkuJ*K(izGuwzQj@el}<;7 z(%7N(jv%ZFg+sMqs%%l2|L2G@^xPE3KpD=5Yn`}o%=s}w%WEUJy05w9b3}xSI&i3xD<3r_%y}nsV%SyjDE~&V@uy*IH;$>Sfup0ClV8SH7Z&+kKW)Xpn zITcKj{l6h=N^`EV1GbGb=Ky9!L+i24{~KgY$t2{G9NQ=^--x2wG)Ht%1Xq`_1H!xC ze|B3`Vja7gMVyz3xr&tCKstin=Eh0?YrcMc%;<=&UL!Fr*$>4)MA z?SDrGbo=0NoI1Dip6GgPKIl#T$hi{QRYJAA=qew_E)neFT(#S$HF(kxyTKNtcV4v4 zkb+GB6ZjM>DU9FL#U%w1fUu+>ss+>`=SndJqLKp`))$ciQ04*Iu~@-1Nd;f2S!AM1 zDEZdzaqixna}nfG8(e}HWxP()>Y_ZC)8dgdH*_YM<`&(A4YUND2Gjp@@i_G7t7)M0Lmsvs zJ3LV>JGol}e8!{-)-Un#Xx_LN*KVvnhbipO|&P#j$L z=TwsWARWeW7S0oX3}63Ub|O*}P(-LcKVC7kzey~O%3zmFDm;%YX5A;H|F+9OINOV| zsj@oh^VT1i7NaZyi%WYF6`Nzx`n0|MBM+Tp3PUiY+y>ETUjTA%1&CBc*7? z&g@@SSouKK3V*SolTs8v%A=o-S`SUlRzdz(D>%v|dbhg_Up z&z%!7(wu66VVK}eH)%Nd;i*Uy&E`{j8*`ZYFpGan7v8B0SB{qI>ki|ZT)m->xBbpS z1Zq(mrMXQdiDGbuVvs;Fi23V(%-)mG>)U>zkOQ^Q#59^nI zane_O_toYCQvv<$C+-VY?dSB;0X^T!7RU91!o;Iqr627WZAdxoi{BCka#NNwV#c$M@d{=c$7jEdfZjLchxC+|UeFaFj4=I*L;skZR z&A&zDANR6g#AIQd+^%b?zkPWV6MXMg8NVPYNxSr3RMl9Ib~f!Rja4mf_wK(QElf;A tIsD<2vY~ :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/opt-constant-propagation.html b/docs/articles/opt-constant-propagation.html index 007869d..350cf22 100644 --- a/docs/articles/opt-constant-propagation.html +++ b/docs/articles/opt-constant-propagation.html @@ -31,7 +31,7 @@ rco - 1.0.1 + 1.0.2
    @@ -56,6 +56,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -74,6 +77,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -86,6 +92,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -112,7 +121,8 @@ -
    + +

    Constant Propagation

    @@ -128,25 +138,33 @@

    Constant Propagation

    Constant Propagation

    +
    +

    +Idea

    +

    The basic idea behind the constant propagation optimizer is to replace as many variables with their corresponding constant values, as possible. When we replace a variable with its corresponding constant value or propagate the constant, the computation time of the script decreases and we get a speed-up.

    +

    Background

    Constant propagation is the process of substituting the values of known constants in expressions. Constant propagation eliminates cases in which values are copied from one location or variable to another, in order to simply assign their value to another variable.

    For example, consider the following code:

    -
    x <- 14
-y <- 7 - x / 2
-z <- y * (28 / x + 2) - x
    +
    +x <- 14
+y <- 7 - x / 2
+z <- y * (28 / x + 2) - x

    Here, x is assigned a constant, and thus, can be propagated (three times). Propagating yields:

    -
    x <- 14
-y <- 7 - 14 / 2
-z <- y * (28 / 14 + 2) - 14
    +
    +x <- 14
+y <- 7 - 14 / 2
+z <- y * (28 / 14 + 2) - 14

    Constant propagation enables the code to assign static values, which is faster than looking up and copying the value of a variable, and also saves time by eliminating assigning a value to a variable that is itself subsequently used only to propagate that value throughout the code. In some cases, copy propagation itself may not provide direct optimizations, but simply facilitates other transformations, such as constant folding, code motion, and dead code elimination.

    Example

    A simple example would be a code that converts the unit of many temporary samples, from hours to miliseconds miliseconds <- 1000 * 60 * 60 * hours.

    -
    code <- paste(
+
    +code <- paste(
   "n <- 1000",
   "hours_vector <- runif(n, 0, 24)",
   "ms_vector <- numeric(n)",
@@ -157,9 +175,9 @@ 
    
   "for (i in 1:n) {",
   "  ms_vector[i] <- secs_to_ms * mins_to_secs * hs_to_mins * hours_vector[i]",
   "}",
-  sep = "\n"
-)
-cat(code)
    
+  sep = "\n"
+)
+cat(code)
    ## n <- 1000
 ## hours_vector <- runif(n, 0, 24)
 ## ms_vector <- numeric(n)
@@ -171,8 +189,9 @@ 
    
 ##   ms_vector[i] <- secs_to_ms * mins_to_secs * hs_to_mins * hours_vector[i]
 ## }

    Then, the automatically optimized code would be:

    -
    opt_code <- opt_constant_propagation(list(code))
-cat(opt_code$codes[[1]])
    +
    +opt_code <- opt_constant_propagation(list(code))
+cat(opt_code$codes[[1]])
    ## n <- 1000
 ## hours_vector <- runif(n, 0, 24)
 ## ms_vector <- numeric(n)
@@ -184,16 +203,18 @@ 
    
 ##   ms_vector[i] <- 1000 * 60 * 60 * hours_vector[i]
 ## }

    And if we measure the execution time of each one, and the speed-up:

    -
    bmark_res <- microbenchmark({
-  eval(parse(text = code))
-}, {
-  eval(parse(text = opt_code))
-})
-autoplot(bmark_res)
    +
    +bmark_res <- microbenchmark({
+  eval(parse(text = code))
+}, {
+  eval(parse(text = opt_code))
+})
+autoplot(bmark_res)

    -
    speed_up(bmark_res)
    -
    ##            Min.  1st Qu.   Median    Mean  3rd Qu.     Max.
-## Expr_2 28.91359 28.39406 27.04191 29.6329 26.53973 72.75947
    +
    +speed_up(bmark_res)
    +
    ##            Min.  1st Qu.   Median     Mean  3rd Qu.     Max.
+## Expr_2 29.11166 27.10578 27.91726 28.48276 30.15566 47.18368

    @@ -230,19 +251,21 @@

    If it is a loop ( repeat, while, or for ): it will get which variables are being assigned in the loop and remove them from values, and then propagate in COND and BODY.

    In-loop assigned variables are removed from values as, if not, it would try to propagate them, and these variables might change in next execution of BODY. For instance:

    -
    x <- 3
-y <- 1 + 1
-while (y < 5) {
-  y <- x
-  x <- x + 1
-}
    +
    +x <- 3
+y <- 1 + 1
+while (y < 5) {
+  y <- x
+  x <- x + 1
+}

    Would try to propagate to:

    -
    x <- 3
-y <- 1 + 1
-while (y < 5) {
-  y <- 3
-  x <- 3 + 1
-}
    +
    +x <- 3
+y <- 1 + 1
+while (y < 5) {
+  y <- 3
+  x <- 3 + 1
+}

    which is not equivalent.

    @@ -277,30 +300,36 @@

  • Recognize functions that modify the environment?

    When opt_constant_propagation finds a function call, it deletes the previously found constant-assigned values. This is mainly done to avoid having

    -
    x <- 1
-assign("x", 3)
-y <- x
    +
    +x <- 1
+assign("x", 3)
+y <- x

    Propagated to

    -
    x <- 1
-assign("x", 3)
-y <- 1
    +
    +x <- 1
+assign("x", 3)
+y <- 1

    In R we can list which base functions modify the environment, for example, assign, rm, etc. However, this would not be a solution, as we can define new functions that wrap the base ones, for instance:

    -
    rm_all <- function() {
-env <- .GlobalEnv
-rm(list = ls(envir = env), envir = env)
-}
    +
    +rm_all <- function() {
+env <- .GlobalEnv
+rm(list = ls(envir = env), envir = env)
+}

    A possible solution would be to let the user identify which functions edit the environment.

    • @@ -324,7 +353,7 @@

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/opt-constant-propagation_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/opt-constant-propagation_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/opt-constant-propagation_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/opt-constant-propagation_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/opt-constant-propagation_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/opt-constant-propagation_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/opt-constant-propagation_files/figure-html/unnamed-chunk-6-1.png b/docs/articles/opt-constant-propagation_files/figure-html/unnamed-chunk-6-1.png index 6525bbf6f8b36c49dd1e9072d11476fee9ee69c0..8a49e3fc2ba61ddc3c95376da1773c1f6cb04a0b 100644 GIT binary patch literal 44030 zcmeFZ2Uk>CyER(sLA%9px>amI(bhr`B}h^b0kegIl0-5pIZG6YYO8HTi<}V&k|ha9 z)>bkoQ6wr!vVi3D%~e33_xyl6#=UoZj`1G*2(@ePwVp7aIp^Bcx=%aU3{Bzb78r9qI?_uSuThI1S^VI#Z&-kCaVSgv> zT5a;nxK`K)HueK`hW%k5?poCR_a~iOVY_5no2OiU9QgJAo#ZToH3@z7k_KI6nQQJg zG&t@Ej*3y@@mfjlMEZR+pE&ss3dQ<#0W49mc(>zYzIIxgdXqXeFC%Yd^j=YorcK=5I5?rF#4N6jfAIY&P@cmko)~ z6jW5YZQq#=vqp4Y(tUQat5-tky5EozzWMmQ4bchXBl_dy%G>74M# zXEtxYxAv=$OlP&*3bKAm+)VuBZ|OS`FQ(b(TOJO^Bj)- zNso3&&6^CVC@Xu3)Lq&wHC22*`RiTwNQ7&-@9+48n>W&!$yea=v4p(&X^F|U+tG=M z3w1L-G7;J1UrZbZ((03cR-Nr`yS=gEMqhmWIvNRH!{W`^AIpMWrhC?H7Z(n3op+eJ zgxsssavYfbep;$TW?#IgT#$G~LIU5yWTDi^bYEh3F;_@gREUc|l~qXpoZR7o)LHyQ zt~J-WI_YX*STHG$6eh(V$AoYLmC7obURguASx#}0k$bZGlh2Mf7=-Lkt5MK)XnAsY z(06)zT4H{(J#(aF-)Mc_ya;V1N@{Me>r|<>Rmt8cT`%Q}M&8}ziqfSDnl_Y6|9vi4 zLM$ryM(_U4LGivgk4Ez>UammPxAopxLTSdeJnE)NC(ScEy?^%A^P4rkmZdxW{B&Au zFr$5>$+}X+$c0==Z4VQ_aMJxvTwNU==PUmHZ8yE_(vzH{eQ9KX`4?_?Ea;5J^{;mtOjF9~W!w(;*0Iukv~EPpbL_xp-Mr5i zFw2H2@#ua3_`EYY4b?Sox;8PnLcCA7*H$gvJSe47@-Vwa_BA>yduEVrovOo7&IH>S zqY0#*TdQTbk`Y-qzCWi5)s-LV4zy;=X$p$QF5Od1 zuQ{;^mx;L4eEHt?WPD7xH0`i$uO~BK#?FB+X=mx#*!aw;u7H5fFZcvy^YH@a$Kgsm zowu0zinv0Y=u}nPT4n2R_qiftVqzKu<6ga8D>jf?UD6(4kt=4AJ$x!fld&9K5p8!) zjr@4Gu{*On_$+;Y>c?YZcqVK}W51YaqiM>|!REuXs*fL^T2~%9=&?uEczhfALg6Eb zgAdbg zfB#0KKiPh`E8IiQxX#u-uYxBwj(&WDSh&Re#P4k6m#1#u+S*m%qwV5X%9j-4JMjWc>81J=s@7rCpe3b?C`;ScOWF0+%=;)%S4x&Dmy zCrn)a%CfSuBsi5+Dkq~{CYtTLEnH{R3!?EUisAmXTE1ry6k3BwP%6#V*7l*Q|5vH_ z*I$equDTzgQ`fIw&njwqNhU&J=@aYl3FZiey|_6CHa0do6?N(KMKz;{TZ+y(lFrk* zU+-*{qi16~9cG6MnV0VF!jHEa*N5M49WB|HNTrF}4@#8k>+{Obj1O4X$e;Eqtk=XTO@(9+l1Jf8N`spyU6c8t`k|^|qUx!|M+=>9CM4FMS6okv=&$acOmGss=%}YvpPtUpLz|?oTNb zq|xPHoa#h7;ayyjl(PlQ+Ap3z|1-VWPTfHwue^m(&em+Vx!~YbEql=KJ;A*f{@k4X zLnPRFyx|e6&>mWH2~Vu<+*m#Cx4J<;z)auoy$LUY@opudZFdqc@OW7jZnyU=oEglr zKm}TRt|k?+;5Qt4h&uvlv?a>g#=@#Cq*-yZS@YvtV768N&m4B@0| z934TYTI9{yoqF_#yd@*fb06Q@n)^1$zD0HZMq*2exQrzmE%KaMw4H{j#YN9{kM_nUvIH{r+XDN& zV2OP9F-RE99<4A!Gby<5e;$iGS{*WodgPcL0H5tw8_z+t3=S{(L}&!5x#QdpaCd7^B9C^o^lq+Q)2CoNwX z*Q7XY!tOoBdWG%YtdH$D8zmP+r}Ah2ICknWoAS~EJh-)Wk{?YVK+oYp>-=~+QzXfS zwm$CbHxp~J?1zBA`4-{8)en#MF7ELHt$^7n0*C~i+KxsS78dS!-Dtk=KYkdGJUmRM z3%bPg_ZO@O%{3O|dD4EZR+|YgBE12@=s`ufBbl>Ur1J9=f2?f%4(MSN6KjyIyl#i2 z1p3xestN7Ae$yfq;3-NaQHw^8R*dRG$>dx7QpOL{7UpMs@^XiAX9SI^!oR&GYjEMB zq753Sxy5k#vR@L`$;o~*(0 z-TD+{X}IIWpsX(oE=$^q&g<=#`70by^F64f^uBJ;XJ5$4)4`G!@uoM?$r4mYoe$}J z4e%3(6r38g&kMkoifX>Gpy z19c6@bsZK>n@_uf$rlR*Y|hR%?Q2LesJnFaarXRl-_-XP?jw`!?qu6e7OL$OLA2al z!%;L=uN}glH*4kH$ar!tsVSPvZ2OKV>+Xuy=|kl4?rz)>{Po>EgA&iZG`iIMu)70_ zvsM0zHM&^?ug}`19T7vQhH{QRM8_ivq8#2guTfr*nbbW)wkNi}#Mp2reUWuoMRWT>) zABKT^uy4X9L!gcUhIg-0O>g{kWd zkUzPLk;yuB3t7x7H}A8o^WZvP0aQ1Noe(GUJAHq0(JnDy$I)2rIpg35%?qiYuRQGZ z$(z-ykeuna`Th|NtS2g@>#t36WT!qE)|0gE3PQA94KROWlzCwr`gE{Em*1^T<*gqc zv3k6CB5v&6P{n}mSYOjOJ(03@&8VN92lJ+69X^cz?5P&I{-do4L@DnxLHc~j*UW#7!G zNcZ$s7e|7Z5zAb(It}acrt9sGRn5|B*A^>jY6f}8iSH$?X6iEu(yYNZu}>v%wR|8i zodmO-u{u?|20ahzi&Kx2SuA#Qd&vMpm{{S4eB8~F-m_%lE~-%cR(h0w)=31Vk>Z)H zqH*mchs)&Gm5SJh{9C{#WockrIzw-7;Bk+Cf7k*@PSEu8t0t7i&Z}hRK{XHh^5x4) zkGaw^x;@JGz8zUWpbS9kOU64E4ctz!DQ+hv?)^-S;(0VWL6Ti)3ngfc7bNYzKfl)n z40_hqw)949LQG7_uUvsok^7jDw6#Ny-d?Tu^ZO?zE_5I_rt_u7>)jqy$0)~cvnu+t za~L4UkY#9_F$ShPTeW=eaol+6L;!7R!I~XSaRJfg_fa+iF|(?o3VBcg=1cfph93eu zJ#W*+?Y+`Lr5uez*7vmJIMO?AZ6Z-{f`=jII z%Q2U!Xx($@|D41sYLFYuvTfE74tE2eodc-#pe6#j?q~mpguN^0DuPe{2XV~0aFfN~e6a$W)X z_CB<&)$7?jdwW@1wFc4RB~XEwqJR|LrhIN~HMBXs?b-=4^->)Wd4k3@FK=zyGDLbB zKy`pZv?rBxee1FVh2f~zZMu&wfR8H>PZ}KBmIX{x{VA1e&yitjlpl}x-CNB#x*wmU z(ecqz{=94rQ@+1jkg0*!KCpt;I8P@<3I$rn!Mktvd;Y48Wn}827XS6@Cad|WZvKyn zWcLkuMlG@^n|tx|b7;hL5*fyXY{AW4t$8Uw^`75FzQxp(Y)a_S+f~SbTOK0uetpBQ zEm|dMf>-|eyUBH7Zs|GQ@tnU8G^SVnc6ysUhxWDiTl@s|WNzN~Z}mNJRZu72?VP_b z#)O<9XMR*CEn^cCX1O{EZ?ux1L#=rdxtO$t^vir#l93~S7A&x^-#~h+#KJweIZXFpUmGXaI)62vx42*uS34V7#wC@1hBseMt?pZ*ed(G~_L4)KqMK+3Zng6xC5<>>z@`EJpZfz2IyjL-@ z!&6bN7=SE*N(N27+{thIk%~-Q`&uyOd6E^P2|7>z40N5J!p`$E9#%;bQPLp}B7-qr z0-@MS0&T9`B$5key7OoC`APtC4=OWn>V>--w)-nzRE=oU&%2 z!gdpn$by!mCyMBz7IzeTSYw8+ICxeBhSH>C#sZ61SCEBvt@RndL-=P`83nQaX_j=jb&xH67r>WGjP@ zAU*`mfEWlkp_fgj1cW>w1R3`&+?D9V{mEyhCx)~+v~dT7JbJ(^>`tykh$Ylc4SG1p zM?$hB7p!LA92lo}c>MF#l%ib^sf>-B1*TgB=1P26tvJvMv8HLY7is#N!_k#1JohTD zl@saKuV}(7X5c{uNFv2ov_aS7%57ygVEj&yBs}hk=z>m@U+*#PA=zFM(Z=N83>4@a z;jP9)6TI>B`#N^f{SgY$ZAvK?P!PoCepZ(X*&BFg5j;^)%v1P*-69ZRg3~YQXix;9 zSab#v4f^$9D+BS>goC9^fIirg!Z8$1^ANOb%YXF8fnCLq^jKGKIRTJ z@585wB+i?^;lilmWY*$AT5|t`(JT1s6R}8%Ty)@(FD8jwe7`q7-XSGfxWm;+%XMyy zO2h0Wq+9gV6jLJ;SL`{>EW2{_#0wnSAh@P*ab`yS@j_m1*2m;liMb|#N+?u6gJ3@( z@X$6TZL>1p17ZlYMXIz4Sa+AWHIYu1cBijpd>t${()Vc1Qq2!oEWdapVrSIsg1L)h)H)r~6^gWkQCz{Cu%v zOJLjxQk3k33{(?3b@!=9i@lTw&;D}TTCLCvm_IVxZZVexBoMxuKqbr(%yOddke$9H z?J=O9r0W)Q`jmmb@c@1OqD^kS7NSA*TUUMkXtI2RgLD6Vw7Sf0SFz zp?v_(@!#9wa@|2F4HfL?K;uF5QkllrW&xaqw?GVdmpnVop~|t5T{JFRnCOT}If{nDoWxW+p1pO4!EK$d27z&+dI> z{PS)?2~ACc8u2FM0->c|Xt`SuTvRUF7$*@7{`;m8ZacIycCR=kUquz-n^mo{N+*g; zd9!^hS~B<*GVb-idbgng7`{5wx;xWlcDTK;+dgkbleBZRu4t$1vp#%c5Cf`XRrpiv zn5*8{_J!@nm8IVUy6|2^4J11ehdGN?;<_aU(vY6&sBWO?GF;S$v?WqWb7z`|DlW1f zMuW-*z#>fKC<<~^LPCO*I;MWR7KiR_*Z%pf^KlEVO;Hm^ar$P4tn9()NXkDcTtucM zj6WFyCx3iZ1rMJBq^%#}b&;_lz}wENlgjScQkS z8EVOKW9S6}6WG_2-mY|dh=0_4Rldf?y7SPcT7VQAuccW+}Qm|Ac0t%!=&8|GPd1IdGL`t(6} zmG22KrH>=rj$gs2-6CJKG4WIZqXuHJ4w4xo7pp^53k(w_=zKz#-~I(YB^;_r&zCe4 zZ&iK2IN&u^qX;hoOu}9ewn&%_**Pq3^Zlu9z2-rro%ttu{4>R0+4d7@Xy6SGuO;*u z#ceGrHTjR17ykeB4wSltG(kMyLsUVP!IF6T*E=0|QowN{SdASjDvO8Iqeqbs=G5s> zDKUDEG!<2ID0PM%29t^1f{Lg)|9;`x7Airj*o0)@DuRktP2$stHnWFD2qO$IM@U8x zVJy-4(evZ?^xv&NSO($C8x{hR(_yzr*>bJYU}I2uuwGfXp7$N&!DdN^2>T85m{;+lBkts$qGQ%w&OfUyK%PwSWk% zK@^owc?2Fup(N0bPDP9EPj6X(457%E*+Eatj<@ZrBdi0Pi1tRKMuwy+ zU&EOJ-hxkHz1LdC!=h2qp6i9y3X~@LPM9M z2SV8dV<{y9t$$`fJkiYu~L?*&Lm4*(GQYEv`xz7P@ z_pJ(tMQw~S5f`)ZIHa+o&xw`}0IW=EbOvgIBqB;Kk$ZGKWJQZ}X9gR?87eHrzY;4_ zakd>#6-4UqP~L(IUAHejJve?e*XZuhu7~^S_N~qb6{F8=CTFAb^v;BdlL@N*OL3LSu|pN}bo#Y~B_!l($>D^uV=m|FF!s zR3nFodnUwrT=kRu%l`&?mBsEt#Ukx5to?$U!`!1|o$Bb59%Y}E8YVXCt6<_KJTBOS zOxS4Lz$n@De?GvVH_rtgN@_cyv@%3q?Jjl%=}}gjIjUE1n4eJLhX*rWNN@=1o&o%x zjeq|gaE&NWiP>iQV#8*o62g&worCW-N(B;n!cWh+twCN>b1uz9A`P(dq)(_&ikD0T zOVFlCJFg!@84=wt^hVyOzEXbc$SzxI;+FYUx-Mwr){G6s1a**7 zgN&i+mo+(Kd>kp&o)@H9K}|b2EHtT-Ia9RTT^AOBV!;0aU!q7;Mw6zj*0&VgUgO%= zr3a!(#X-N1u>CME5Jhv^#rqPo=};)fOokkLB7F4kb9x(&oFiW?UG2xu+@kwo z-Is<>kmCtSOS78piHbr@E!c+OGa{rElhiTI_<9$M((T%8KmZNKsvrV@iH=0N6W0uRmYOoM>@mpS#BB(cZ-?;~CfbAAej|EGJvnR`887NNIza zc?9ZJLYVJijBb?F6Z*>rFrn++@`|S3bqSh>10oo}sI_F>|95c_P!Gj$-6k%%G9uJH z0xlsF0+Fhamo@`mL`wIkoFj-)No6L8A;!3Zg544gd7{#F-kR(snLxk+)*NifL)}*e zCP^VuG1*syS?f859m)U6t*tM)QNGI2B|U)VId>536LJue2${*r^!e)YyDG@^gl{K; zZ53epo^$K4Qy*;zzDDmZ!oU~;x?dR4IPBV-i=1WvcZIXOUhrNJ&lF+KFiUtafIFax z2hC%T*N~J|M-f59giRX9=+ckcYlKP7T9OmjcXNnrV^Yh;;MGkC9^i7((qZspL|#(G z0G$FI#JT+{(IoNNax`TRG+zlqdeNOgageFi9COC(Y&RNSqq#{u35d`uAR|~}u~IFL zL*-BZJWi&$k>MfdA;$*P@{~F?6U=kQuNjV{*n~I6WZvwDjNpK=*&0M&7YYoG{)qh` zUsRBt+8}N@5b_`an;sHe;wDiN%eiJnth_{MI)AFh@Cq*oebeV*e1`9iAC zjq!dzj92@JC4<3QFce^-s6rCFglE{1;=m00`@s`jn6l-#^@`L(M6{e9A2@tAGYf^Y z3!7O9;!OB_=b)<^bPZ&kPX~}q1sOebPi0Dg!rKNpLOZjCU$zfDn$NWv`~HPrezpxQ*O93rL}fIQ~evE#s%6XRGs1V|Vy zq14r}%g9K#g{&!Ddkl>r8>PYk3MW5Yt4P}oeqn-P>Hv@C`z_zqUF7VQ&gsEC2 zfF~kbsQ8hsy*nyZP0i;uo3Y5mc)+cMwgbkw#AJ{lFN45qr!K1MdEN3TzqD%n<%=gW z@Mb7z9%&yKfdEqknmq8lK>Qz?ZFZ)_q>g1tu4TK*Zj=T zdBLm4({8R>6M5R5gDGm-2PKyn76{w@4E9;jgW%$~K_Mmv!h)Wzjf8EQpP$I%qIn~l z2=Tu7>?P!?qqC!x2H=7^hp`P>MgCbSS20YLB2%iM#8ZpGSXg~Lx~7xiY%1ny#76m% z%2E92)&lNP7Q7oXgRb#{V`lR=^so}Q6|0C7j0zK_3w_{qV?vu!YyaWaDY*(LCPd$) z(cPuy58c~lbp~2cR~fMip+os-ufl>_J6t`CWz-obB%%t$EP}-eXrdD`1x)eYL$&=x zrn80+V1Vd9$H;!MpC6Bpg81DEMh&u`2+>0pNii&j3d0idQUXELpo!8rH5#%INbLde zkrVF;e=uz^gpe}82XAU;bVQTup2 z2pvp{a&%O%nwGUKzPZR9?BFfM@*FZ-DU3V;(}^c9hOn<~ne07qm)*Qbdw#@+T1>Q` z+Z)7RF4AzBsL&~yn0kaTH&ctDg$EHQ4P@2?a6eO>{H}fg7UB#68C0|66jW+LQa@%B z;UPMvgA*KMde85$@GNqLqTp`!w_xOL!vAP=!mst!TjZ)*r@_0;xEv6f1Z@P9VJAT? zsK*+(zX;4eq*&tj>C~JXVqwAtxR5G?ZEHViU zEC#F!8sU`TKMjTS<|k~VP*D6}f>2t?!taQf4FYGYFX5THHX=A^b0jb#KKC`I4>_d> zzf(cW)w4of_9-oe{@>+$^+_>;b*)#?O@Y@8FF%o1hjo3{0!lrtKxEbUwJ_3{zB<3Z$E+pm@VFnMh3oEay1-LH( z4wXhH)>V=&#AM{oal{I+H4;`*+6V@WE>bTNk7tKohvr7cdoPB66wHW<#!ft$ZzU0D z95NgS)>PIokoW#ijVlw z7>*&Gc`#;d&BU@p6hH~dj;F^-Ls_|Li%FwddU>VGc*@8mX?LhRjgWqF!15kKSFoUm zphtAUr9h)AftM#De&W%rPzDGmZJ-d&>>Cc7-b}rRX~MdS z#-)db1<7TW$jb~Hr_MZxIksA2b;jV#NdxSCNJz@I6TGJ}-J+ETK!88QV1-F(1M#!> zkZSwXSfVR~-lT!&_Sptxg)phv;wu&-+&c;u>bdE$ef;Y3 zC@m1m^x%GF;tH^Lml!!ifDl!vq`2qP(V%!-0q7h!vfbts64}t3ca6BCE6*VsJvyEcRYq1 zQUM5kM4c%{-nilm^lZef(d#Ve9YqBGGRL4biaDPP#@ zGAdFD?1^4OPThxTbYRI&j9hUq{jv#-f#q=pZ%NPtC`9Z968|6bR@3)3+sVv=!Gp|yR8XW7h+xJ>yO z_&Gd?N=?kdEg1BZKOSR-`aVLOIAl^JeE1$3aVx_(*o6|#bEx5978c36Uelg)sy@<~ zhyh8@aWS102>GsJ!e1h;xoC)XZ$^7O;bo*>eArsHg{YgPk&tY=<>8zLcbI{3UH3eN71VCqpHbLB@l#+Z_0EBn^fsf?h;F>)akm!W`U1N-Fq~ zo5lLqi4y=~9soI&Cfe#Wx(&0$VxTdREACqmbF_z@{|rG0=Y$swo8QromX)S z1|%r3IoYo?kk&&uaR8$ADP+DqwY9lkSTAY$S)~nSUf9D-3CDExbaVY7(PT80ZFZk8 zi{S(Y@gGwiFp(#N`O^iZQ0^ZApbxRI0u;VHpFLp&0OF16k%M&Fm_vEdh#l*ixkFqH2ZC9UJ;t7n&fLZ!}Y-q z%vWIW3}u3LV%?usuT@ImG?l1S#6U~DS8#ddkYgXEyy*~g(enrqNwD!kVN1}x`s;h< zcPs>Ycx?IYAFb!!JTZ!A&qn(h-K)EJ_vx1yZW*uF+7JAVdDE16Kh=4^_TFX88qage z!!9z-A|lex_*YJE`kAL5ful3Kr7C|p`n02h){@v);$9oRj;P38AuF8e{QJ$v?) zcFjR!uR$}2S4%4quX|%_0(hq$#~?u{JMV?IFrAvGHhYYu`2+xeJ3Fw}gZS zKtb{ov({YA%a<=d`1|jVUi(x8%@R5~be+*=dp{?=*s$%;PMX!BZN2r0A((86gGt^| zC>=XzJvX!^^fzaz^wlT&P`7YjkS-56crcKA=*Pr_3Y5k%G+=H4feUm`PtS0G)R&kT z)sl7Z)9LiVsaf-MbyO^6$f27#ILw1$4L^mkNx4YuOJmIWMK>qe)hYI!K!oY$>BhAs zU)Wh1x{-DYW4*Ooe*5kD%a`1=DnB6uVY~iI<$+>~yy~x&QuND}5;QW`1@Rv^Adf%O zNWU}y0^x}xqh&(9%~|PdH*-I2;Ac?Z6y23jR9NjMe_VE6n46%7ySY))3_rcK49kwy zDY(TZpyTE*ViNiH-@gpBd%u5rVQ-sD=RC`^fB!k`*0*EEj0;_-xZGIt z^iQty!8N_Lv79tXNlD|{7_YAdHl`EUKZ;v1At2fAKTK~&0A0dk+ zfP{*{&oxf0XIyCGu>$6u$Ggh>cKeia(u|ujlJG?9xQ^Z6xZbuR2QU+@-^XUWUlo=A z{OKp#RP2-ee{bCyg^npB)7&?PX-a_M(8d?lBQZ1L;X_rpfB!x`9MZB3eogd${W#*E ze@>h{bxN(>Zr}?m!>)~ao*mne5usP=y_b(qj;=)Z^W}wtMN>6}q5(gG^czH~utkf6 zNmH(~<87kO;?GkK+)mFlW)R!1GFDm@RGBcWJ1gUu#`1M(GBOYJD}&7*H#ia-Bg83# zFl3;Ih)LadxNhj-54P+hTX2_gZ~;F)f`YYv+lG789tfLnYQu zxH7rHboJ8jwp|3!V~6kQxI(WeleY7i7k+oBbzwdo())fIk7DE_IGbtj80QZt6w7_& z(lB+IHEU&9KBs$ltdA_#LiQo}GDXbm6N38fav78&a#~6FVE%gCSPYMH%o#d4vIcUz z#WJiv1^U$hQk||8cH-ALfEW>W0;lmO{bT`i_-svDtz}qTI_Bp&n7_QJTlQT%2PD|K zj&UpnwLAzr-1Dk&z-atikdVQX0%6t40lR&>cD-f0cjr!VV`JmuaTE$g_rI91Qij&5E(7Xn{dL37ZxcY0;Pax$vX$1uZHvkcruPuEF*G_Ni>x=c~Raa|AP6jZ0 z+`-LbYBGPUn#0iNC~d^{)QI|YXLfFmlZL_qbY%oTCNfZzGOnv1_ovaji=q2WvA8>n3h zNN3?d?$KG$Dbp~oa?(ydWKo9BvVM6vno-IJLoYl(R`;4|@7}%BnCsVbm2Bhl*;n3@ z1*TkgjE`|)w|iJ?!M15;cV{@YGOuj0P*U<7A8O^NVQ)RzL`@Bs^J@@?M4qm|($tMJ zu86N*&-F&CxVYF~+%`sJuqiVIky)_xa2Z=MeXiDYU6G8P`VMmE?xyA}l16*>oT4k` zIyuZiQTjLkCSZap3T$@u>GIh+-IIYe_nSTE_17_ z=GB{;gD_+pwPZWcsR8AYU=~aDQ&<_)1+vo-(b^ON`s5{QJUtu@%LL&1?k2q^7^%>w zsI=+E*7-=JC);vD&m6@^I9**`!J8Gco%a`}VP#ft(kla|i9@cGH z?tax)cJ^3cOVxJJmZkrh87v4$$)CNor95X<5;USQ2C&I_dLtY&uC=3aanoY7^KxPP zy2|k7;g$q(Sj%#Uj3@VV@NbveaYU%6#B1N`O%YqRY>9wC*!TP$&z;U31vD7unswRUbID$m;W>%BbH62Q}D zJ#cqF=xqQ3v8 z*RHW0vJ6F^pwh@`+Zxd7rsa#DbL+=5>V_rjmG%NRrhLE=fV!^ofE2w!PtVQkSp`oV zIBHg6Vb9j=a0-3Q zeo%%!J3A|E^Zf$TfBlzk&*Do6C<2(;i&la}lK0|KrV>`93PeC8&O#k2?Rp4!hwtO0 zi3Z3*n^MZO>P$A3T)f+kbx3QLALHY7cm$;wrM)x-wD}ujy%ESJy=e{Lo~klGq4;ay zC6quMp9>*2X8aM`i0W~2EzACwp!8$UJljPNhr;n4Ww8!yhxguY`%DdZ?c9GtAiY%h z_U+pm%tPnIq!vGIMK+X~bOY2l#jkJgaMIxRxrk6n#i%wNp0<2Uw(WvgN}4-#MwY;o)HdR57T<#i}H>j7_;rb#6uG(ms9q zRF?$ZA_g1|x~ZQx`{Esl;ZVra^)lZ3qksIcBv;tcqzD7IFp^|DtYTnhkSZ$>_Is~>;iuRFp0JUI@qN$dBJqfWa7c3DC1wsWX zN)p>a^ElA&FP)}FcrLvPMK2x$Hvt%|ISp1e_42#bnHqe3P6Sj!w2dQG$VqweHU;iu zV^3|kcO2tYHPpUnu(rJz<>}!7o#jD?RWSJGw2NiuCCSvg4D=e&p4$$zF9D>ImvF%L!I)68ttDrg( zs5S>e`b)CjLl9|E2qsN{t&x!th3Cia{KQeGUT|!+7@GGWp{tXYqWCs?tC(~#owGLV zBy;tF#p||R?vGr&C;b0|>la`7fd9OXmbv&FlphzC-l#!PBsasifuH0bN3mc04Yzx! ze@nl-5%qlOKRIv`Zs|WIHviuX|G#!AtbCUWtKc3^&XZ7zs_rmv?HCw{|D96?#OOCivq1E31&keq5swe7p8nQ6sYJ_mq*5d=p~oT|_kUbUJFG8DJH%OALafwBr> zMk?|_U9t-mT@9$$n@=+aCkuwiJd3KCe1MaZPsR%J$0`&fukn{uBP;Qsx&u;DQp$;1 znw&D8A90S!wk>}JqNWhm7;|ph{sCAAQh_ZvG4ATg z2b<;5%4-1uP@KaZX2B}|1UfdH8tDW^5qB5~hf#g~dN$!Rn9M2%Cn;{G`Ukf!Zdo{f zV;WB8YR@e29dYE=@h*UX5=|P9Nq@XW0`C680T_eI>T|A=nNq~Ml%)dwuwD5J z>OBx4KdJ@zb`EWOi9&ZyD{SN-Q4d&2|xE&y_1z z3J%Po2(k(3Kk6T^*|Cr})x{3~>)z^$jT8Y4gydtCRe^E_{?7JUwtbNZO8sdGsPT)q z4X*diTetf3jN%x_(pKcVA!yXqg*^GSwGcX`$(hiMW8nBuqT76Z%LE)ORNA$YqRYP& zi^&b3G*T_U9y1}Q8vx-Bmv)Vgj8vf~ceJ#l_XuY~&E8ucBBkEdTOq9J-IS8*JK32s<{(9%_u4Ev>mpnIE%ndMSC_If{3`?isS%71U=Q946Uy;`^dtXH!k z2PHpec_klim$-fbj?#EX*U>S%pZ4``Mu-OqWq9vzHSoP+4RFU-*;j( zxNS5a1w$W?YH~Q$2-Me?SC?;N8g$EAYERNE;?C2lII_e=qvz%2eaf-7AQcZMw=T8A z#$D2BMMmTd_2r8{#}TIz9$Cz-(fRq(K=#IUael{+9i!5~zkGrllZ=B+>zAYE3{cX= zL@i!ga^pbn$LIL!gNPcLl&wS9fJ->A&+*sNM(!VhsiJ43vy>j5H$Rq0M!~NqdcZ$- zljNaQ!Kt0*Fse^bJ?p>sZkef}mg@wv0AovmaC$eg5*9Jl_M)OQna$?6|JUzgAiO6R zc<4_WUWsy#Ol+)}du08EX$^ILUAImN1(HhR^Eq4U%_{^3$@1{h zM|VCZfXe)Q{5q_eD+r?s@I?9X0663U+=L%&( zczU-JOMp%=-#k4QD7@}Es>14O3MFCr@mZ^~9jvEjX3o-K42wnaAl%zh!KJV&gi5av zG5exK_oUOqz&QV}AF{Bp=r?Nt*E49a6_n+LSr>cZ`7PwwUK-S}?7Bp45>g#zOWWU; zpy8U7W|2GfK?m@7DQq+LkyD$P;yG!sd{?2vm#tpuCh=%_xwhPU_Z;VL)Z0Om`~lL@ zfqnUdCZOo)&B1E>PMkiy@4x}F=S2a_A)&cXE%hZ1HyY8ZI!k>pZR?RckzKx%;J@zN zQBBXA8&_Wpik)~z#Ud7-2nX53qn(>g?Ezy~vjD&7r+Slel_ycZJ@WQzt6 z^E_f=4LWR;BK74bagP4aH5D~)B7{@N_T?wZ)x?_O5ZD*2>3@8P*p8IK&-_NKf+xbJ zqv$uVDm&emcXy)vt0gzMMz+LDNpH!w9bV3rNJDVn+1c4t!q3T^IRk>&?@uthj^q8G zP+CY)8)!&5je$M#>WayQZA(*#Thnt8dpBT9A;iW$l>}}c9vNr}!rPfA+OUq(ctu6( zwf0l*|M@3;#ut*wOH{raunH(c_cxXM<=-fOYJ|~t^Oh}~G*T)_7Bvjau1Jdd4ElyJ zSXid!;iN+M9rf^KrClGsQX8#!8eUChGK5P~-Gu*rkZAcAcX=+xy@XA@7w~m@syhT> z%v+sgx#rhje~Nj7DZMBLdNbYr*-D;k54nUdUGkKC>oS@2oAFYHn_B zgZK6#yby#@bfhDb8cf$RGgss3IBC%H)-zdbNOzntf6OX$%Ag`J=?N1P(_Tmo;Ko#< z6(Xvfwa%S;q!#p+YoL#E zaQ5p<#r!10> z_@zJO{xZ?v^C46MCJwbiytFU|6mjgaJ8LJLTCfQwAQ^8!6OVJiSUm>V(rT|yNk6q* zz%)Tg)rKBj#bD^5!k8WUExe}67KRJ86#d7xHG%l*Wtd;c&|ilj-$OvnytR&p>1Gx zeF@vYeFmfpj-9+Y7qBXSVYv+KAI4-63;uylj9Ab?iv;n`8)Y4xP^uBY;)DD54?sJ_0=r|y zO&{a^X5j#(2h_X90e~^lfv(b{RRML0SaqD?D>ShvYGDVkS6bDNWp1b?cNkZ^jsVWDO3^#koIf(=NhK65D5gNt+@e-WEH|7Y zE9Dp^d3*~ejc~D;gDp~SS|t%u3umk9CgFVjgjbAQ7WlOAdB6?>)H%$r&!_{fdD;k! zyNWA<;dNxBi0(s3LaOOt6?xJd%>-lzp#jDb865@pDRrr$i3IY*hlcLNGBvTCK6T0% za2V3ne_VrTbPx=@sN2P?PGG@>0=XydV`WhAsWj|O9h?)7G?WJ9dh6n2U%e7wz!AI> z;PdNM8K=dDgcxs}qjbpFWgRK!X~8Z)ahjQhcvOo{7B@Vm{%6MtNM_0~qyR6wbJPMh z0xwJ8ZAy%%C=$h5EOLz)3p&n^3KgT(irn~V2=o>t1et4Ic^%PeT+aUIu)DLfE3+!9NrX=!P(iCKk1U|1hjk^uva zEQpDxR3b7ngufGR2LIknjP%Ww2a z95yE7TnwI>!~-5{nF)}7&Aeg5MF4mz4Vnkhv%j7QHUP!@;K>tj&o}AACg>KQ&_l9LZXQcuh~+^BMEKg z-n4nI!U=HRNgqh<1QuDhXQjldrQH??Q;Jp~m5f^4V`Du0?fuw`D-DV1i-8v`Mi9xh~1VOVgB2{IGmh8gQ35DAjDj;95T_QkDy8Ulgb-HoY6ivIDe zaI_rhYyDW=fXv~LxCCaMo@&#hK#r_f^6WGy-5xT>xI+GJuf2+=K-+thfJyG5#)sr9UctZ=ua%=lhY>^A^s|6TnzgwK`g`o8y?7A%wm-QC$T5s7Sp9Y7{ex&= zFNU-N6%Qa{6ov+2qbjON(~Hv^i4j02tiL|${0`!3p{MGVvVaFFw%oPz-*gi@g}sXt zT$X^>aS58>B}{d7>&Q!`v~Kip;vuv_j}sJPBH=u3gI07Ko3YGzb^LMj=FQ$OX|W=~ zC>nqn1t+>pzogPk>f)YJ9UwNrs}0l5T}8j&Ft*Bg@zCfl3v>1$?rU%x-OGSjOoo&z zBG**m7}6F3GUUe}K74pKMdTf9USUK+IH&9F!#2amTlZfIg>bN zDiF4kUc8#agBm2|szo}-COw>fThK!4cu}G18bLxyZj831Mr-G}Fb0lz&8&;?+8+%3 zMZi6Akrm!q2zW}&H>BwH;5`6Qa7CLRQm#7;0tL3RO(uA`s1$_NKhAlkZ01!{!l~C+ zAJE7OS9LDX~ z7DGG;hknz+iyRi?{*(Vfiv*Nl8vWb1Z^BnUtb?PVp1>hEph)AY!h9ZzxBL7d=QWyx z*O=BZ)uWser4V$4FGTFc;1`A{PFglY$vk=bbhp1Cj3zyJ`j0?xK+pSBD&&O+v2e}#=BCvKR$YvujL^~@~$8E+M*p}P?NR9UyZ z0_0pQ@X95+4Z2Kuh*T>AcV=0|asP;lMXE1@2)%RfUMYw>IolyoiwYnnk zta}*f52qC39)2RC)xC{7L<83rX)V{p-2|}!kh!?H(8ICcMj(#wDmDlB`u>5}Xw>n% zgcs{20+UUy(~poX1WrV&YvDAdDjE~&N&{Yx!+56akDfmL?=kpF+0F-vD}!b`u=P6! zdQBhB?I?m@qef%CCVBEzDL5XcdW>kk46ct*{$NUymkZPnZt`-fF%0Rl^bQ+wf>vYq zm-hfhf+C6lhkk}m)J90V#a}J|kn3td_x}mDSG^s){&%b=(9-|h)n;rLOq8cFfp}91 z%Z(U(j&ku;H@({~YYPWQ0$R-(8D35p)IU0kM&DDLp+Z#2RgZUz&Kcc2CpoJfCd_-PkLv?8{RXovs`2YtEid6Zs0MQrLm zmI7ICa(La&^kv?HBNn%~SP0QaUYj=?3w$ca_!>zHGuuRegChE&C@x`A^YR&n8*T>V zz}t{Rk<0I*-Qt$Zx`Z0+*YPWM39ty6`XemYJbTB^ zkG48=ZwKS@XJg8tAUPsc5Yfdp4t)uhf}Ne_qJ-r;cKkReVI(oEDIYs_>#LG<*YYTw zc?{e%-jT99w!>CDh`>Y)>3r#pLA8Y{~}(8$qiv0VpZ>k7hEWkSKu&? z#H`)8vGTP$)6T5r(NJTzcCHfpnaJ{vMFXzu#6*njV~ z5E2A-gtQ6{Q*!Rvj%cy){th zXbM(V$xG8+0lDXYFApMsKm?Gp-Y^#JrV&vNuYH6YlHX{&zd;YNl?lNx>XxXILvFz|Ot4i5+;uLQ#(mo80Oqse3()d;|j?4crMs zNev7ZBChjJ5#izAab*$@EWB#rY$C=l@#ZDGFlRm-Hp^o3^I04fu)v9mUL<13YitaL zwx-k&3uLg!^&dK3{MwR-wb=%5L28hk{c+rOut^oZ2+Z*5BwmQ+rvKSG_vNYA>Ijq( z@U zO8rTJefksTVKU32AH}PtoFcwZ;FEKwR}Hc*w=tm-Z{AoGzaG5WOD)y#8FY&veW{l4 zaO!p`m)Y0d#DYRz|8wW=-Cp>4qCSraz`pSbChG`%ibjX6GJe;|2QMLf-t|~9ad)YF zcZgvYsD>e847+5ETmO>nkTO5hy3mVLUDe=*o^>SwaS?YUf+7<9Xe=f*!gxRLC|U;? zpbVC)t1^V2<}YkSC$0#bm8`N9Y#Uq-V+qJ{>S!^_2+jDba#7V^=mi)ML9a!A`;(cl zRpQoNj|iBCVF#)usU7$#?;gW^&7{_;KNmOK;a+-rI&cF}`iabdhh%I(`(h}vYW;Y7 zQP~%F)*qXf+8d?F1BQ+NpBw zPK|5(I+^v- z3I~k0M}0yY)9XYSCK^LxDc#Uc89sNw3KBxaLN7!>G`b~3!v*eBUeC1GdErVPiw6MfmH&FV7quBCh7!Qz z#M-y!Z?SLr7iul5Ei}6hkzb+?1mUCamyMTVcJDqE`_!Ug?c6MAWit^XNKgz%m8T&$ z5U^0v$Lzi>O^hU!M-gg1u|VdjZdiQP*r0ZWKuf-_pYb~QjF+Gu&>?SM~!-s1dm zR!1^KIH7>q0rei4Sc1lR01z7%Cw<43241@s1_K^SU$Vh@2<|9?`1$#@06)@Z=bOSW zwtRWEOJ84KQUeMjTD(vYpb;~&opmY4#iP5ZpkUXT^x(>zw4I5PPM@j>MnlYxLs7R7 zz|fv$O=v1eA$Akf4xok(T;vye+=CIr2c*GA$ADGH>1V)%#6XY^uu2;i3>Od%UCV>2 zl|NA~!T94-unoaVZiaB{E+co2)(2dnn!#O2R-se5kA4<#OTzm+Uu!H>it1l`>f>-c zk2rE*efzU|o7~)7kd*5RzL-e3WM*W%^~5Fw2V?ZN-z`?hDb9{dK}X0DL_!e47bcp zt%?25-8>*+i^EWJWST8xsfOui5?ls*5+Fy2{FZ^tVhw}+V?F7~=jf0L{0S+sN)iY0 zw_?Q#7p!j4s55&XJ$kgor9TGG%CY&c8T0_?|G0jQCy%f{{`f<(z%d~Z3LS&Z>Q?`x z*oNXF?_$vDwWHtSOBeL+haM(DBU*9{@yr|T+~=VMghiBv3{`_TF*J>Ceq-^vso%OmPWXDAazLJNG6ALh$|iwt$Cqw6s8)Qag}P3-vIt z5C8Jc1E@W)SK|*=YnUQs;7Y)0bY&K_Y9y^3+#7f5aS5()o5av^#DYK-FXE5OFy@ zIli%^#J95a^|db9#KFG48%Sm^HucU!^9$q)Vd&o7Y*pj&JyOeLo#!^*ZCaRKUtE1& z;bmT4p5tW3(qd##2yCFq*}O!w)fqc^uW!daI}`t0$bH^+CGN zi1kU9!3KOM3TB2NGt!Y#NMyYQeZL;RQd?Y&p4ri3Q5kp^s#Sp|=-JzU*`ufOvEqJt zu&MW)X|ooW0;?*|1e&X??*CR6nwO1BCN1sQ8`_n9#lA~+{_8E@Eg0u~`LbrUERkFc zjrft2TFaE$@^hd30tFhr88EI%2sB`j@IFX{!Tvf(0geL69DyQIl!eu&8|V((G!&inc2u)qHLt2VJ^YGRtZ z(NrayYx_}&0L809r?^}F{H4_7>a5!Cw)d-jn_cJ41B|8hfj>AtUAF824Fk+yZ(0HDquls#H{zzo?%frD@Dl{$O*QJEC$9; z7fE$7>3B)!*Ik#U4f$EDK~kHP-V=?tLGlipaK=oa?uboY>V6m=-1U9@VnT-k_fEL=NaAA&W_f(VLo*IrsqYKc^1 z+3sCebmL+&e_yyuynNQ38LB-i7TSNRMXQT=fCXvw&E*PeRN^Hd z%l`BtsWo%D(|8w(29gJof|JJM*e#zvJy0GWF1fS8`~nc_dsxHyB@ddk{gJiNozo7X zWPRpLGVB(2(NRL|NOS`?|2Z;I!+M?;%pnW4CucvI5^sgVG{JbC-lOCJ2;5LB%o7&A zhsq)bI4VDoE3^ti^X5fjEQPpk)O3hk&hDs{i_E8YpK*QEFnTzpRwqhB2tDcbXs^rz zX@X8jwXQkqdr{vMj{ll^>fHqxGdzEd1^}L%C4KX3Ae5E??2t(V2|hmMSOx-hjiA-S zWgDDS(FEnEd;~@1c7JsC?~p{IhAR9zP`nBm0|8sMeP7__J#>=P{fdc7O><5VupXLQ zLWn>{{BwJCQf+sv-h&W8ATfkoatHh{FCJc6s@eC(AWSabGF#&Q-#@9P8?Vd6=s!CW zEU1K-m{{W$*vXdSc|QToPg-~avr~#;1n>X_>v6;|=yvp6T%Mt7N@q$Cjlessmvd+u=C}Jyh&hqOa+9ypeVwY6B;A>yc2_%y-v>w_chgtjL{dBRU>DMw3P-%4R=paPA|{Cr+-C#2P9*J<@q$x_cvl|t5NyJ=HFYy* z%qYVQHAkmvyg*ao$Iq{!KZ~HITktGwV<8rZ!+XzZ&i8Cy#eJ?r5AbSeq!ftMb-%vo z5w!N{P(?~63ih;}lIzFB3{A_j?L+EB>n_T!Nvxv zu^2jn;$aaqk#0cSwsi{gr`V?%6SaYdBE}-z9>IG0*>^X&cdK&ILc|M_!w0klP+D9o zXnpo>GblmM_JUEQlyRyiAonWtvRna*-}cH=2$dctAe;S|!y_Y9lTojot1~IZi~G(F zh+sca+`w}BBGvc#^M15<)Va@kb1ft!5w#&tP&qH+|(UkYa<45C)W0-Z)`IY1**Y(Alq zMgL8tG1??zF__{qGCFVY68v^WzOU31|U^ zCRP#JCdIGLr>wDZ!jN?bJ-IP-grUuRXKGaQ7ranbh zgedg``G7{4#RdSY`r0TTTv0r(L0_x*MjhCKjM`VGQ}ch}-g#3chE}NqB?=OMBONp1 zk5Plz9y5Vg(Ocl0Md2$*&fAG`@`I^fTGxOUp=!%Lp$w~$%jrJ$KHT-}l+fkj<7vlheJJr=;22c+ z6$%=ZL00B!p6zqEqf`%Wpx8z3#!6zbfmPiFUH5l$thX?jMkG67r}H15(+qSG4JQY}C#(Z@w}tqeBLF3zjgNg3W!2Qw5TQ)220n3X zm(iOW8ggQ3K{!mq>OKe`)3}MLcP&nU^B>lF3O){?-<~t+_4p`2Hk~+8M+a+sVyqF< z7KQUKysLss&<76~92b3swLx*IV9bUe4_ofs&_m*8k`#e_)PO$5T_a2t*q`@6$jluQdf^xYh+xR z06|PF@kvU~8E8rYc)x0n%_s-)@&tpd9HBo9lEuLcVU26d=G`7YXb>$9p$Gt)MRj01gW4wzx@e+$eLl$2dt5;egl${W+Q=ynt~Vc<&sraoe?u<&vS~ z(gpmLt8>#BT^J4IX|NwZeoUQB7jnF?UaKO_P|{=vy8?|ev7B^k>fBq)e-a-rNfVZf z?}WpB7nR(T)U|^TYfiX;i5b8FRT>;96@FHKes=O%0a6h@TvuxVMn<=`w;#cZlKF96 z$@Z-pU?g|5;p&lyWM*i5-F3>;sWe~J=9B@eQ}V7mY^P%xXfEP?<;u_`;k`>>aiQ&*PYXsQ6r?t?;vb0ZZgxNj^+*SF>6#JeOxOK=pt zgQ5PMgHpTzx|-4;VSx0GoLvOq{ng^_x%OBzPg7$b(+si(i*=m| zoy5-Biv8Deg(g+;wyfn|bT9BE6iQMA*#-2t{-w8uIxZbC>TohraV&i3jr8}|d_i+` z3=JjQp6@fN=z_dUP69%U#<`>7x08Smj)J9QN75uIYFJ^t11_{+J35YXlycbk>Xuek zhQ>Ds&=f%gVlm683}*qT?bxAjEObzjcVqXA)9hg`ASGQ3KiU{<3q*@7*hmbXeze7=PL8ur9{+aua3rKEV&X@f zE>QcjYE@2iZ2V`kl987;U^~KX=|0rX!g`bami+i6eW3efo^c=G^0W6>w@N0yHiTib zHu@1!7IN_c!1TGeJB*)!<0?GSXCbqE{g@1=@b;8m0k@bySzCW5AW&2!FPpvNfdI)M zW^61q#sK=oMX~1K4XrxHv9s1nOTWe5d3=#q?99L;#JmKXInY_Iju-zhHRACK$dRAJ zFskrt=ve<6(1=BN*CiAf;CR8(($u_0>-FR|l?6MvTTn6poi5(PP7d6K_T%qn?|grB z4|h~$zGEM|pgosLeBFn;q`vc3Hwv6Mu!h3All}H^pF_4ijl4vE8|aEgudA=Wr>3tU zd?EV8D4Hf}j#aP*a*pviHnZ}9nHfqfdWb)MSqq`wgZgKd7TRMOjLrKYm zEjCIKCJ6(Yy%i4OfwsxzpxG%y@qN-xO{GSS_bB>kLg+~ZdTTV53A`XahN_DslNG47ftK}sCOLD^uEXOL))=b@bzLcj0(7blB(urz= z4u~{x&gg{CN~=|jmsyrR&|q_tJY5u;WVwtRj~@Poa{mG7gPm-Yq|PpbtKn~}g--f9 zlh_z%Q>}KAEcmzQ9>O}J`{k3NgoRO-b@F)sjNV-A7T#)Id zp{8=qnhR|}Q|@t;BwM{0Yd_Iq+ z)jzJBju7wL_ZhCk+bGaAduJFS!9kX?JV_C?E48>Uy={uogcrKdj2lgp8P91l`EmVe zP?e&AT*Jcq2IBT8HR6jooHnUuzzJCt%kg?1JWN2&GhBkF}yGFw&or1 z6&*_~tpvG@{%N+k$}2pst`I2gd}{`-+CLIz@Y1eGorC6cMdg=&EdlZv0-yO$SW z4ji_$Jci9<Ul3$vI}w)DKWL}pEV^b5qybeYaPEaJmgu90E3?yi z1)uLXc=Ut3hpNvcJ(*@L?9^KI^!0C{usNnQ2LOv9LU>VgtUWX-RD5wNS!C=O=r4rz zrZ~xRS4Wc_sEtk0FwH0leebG|7)Y9%B*i1ei#Hxj(L%hZwL{xbGR>Cb>!qAK|DpX0MC5ii zoJYxrEXs76=F%TeCGFhsW0a(lEC8N5P?hEi=|}>LzoY1y%o(pCtu^+TXh4C^aTzK& z;pVuXQw!UjMgMAeJa1b5R^F5O70dqof{%o=p48WC4 zEf9lpn1r-lzgGS3RENh{yGK7NZ5Ep<<)S;5SEmVPyZ!j27WcG;`#$;41TaUFCN*>% zx(wdBrKM$Bw+MilGVOheQ3S_n75`lTR$micQvDf0#yD?Vrq%6%)+!mogT0ggLiBl$Zua{FkSo#%uhDrspbP_#kT^8)Ynr4%TBrGpG4)9Y=_Wrm{g zijSf#YAYxvuz2Q@EeEhnpSQZflS7-n1`6HfEa*>5F)lEDaMJ9)q-UIU$lta0Y%R)-lH=jCR`I4l-UW8T^r#!NgS}`o9SmsC0g_4Fs;R;IirQ4@ZcNgvPN=J?8R*aS zzKhgo`0(I)gwkl}I`c6+Cn2#55w8;eM+ik45cL zng-r<`1Q}&zBhH>FvfVjbt*!wbzvv80Vw|j;$9R{>ZPvf%2Gu`n%#rQ?jEQYF-1ki z5*O@A;$(p{Jiq}dF`(Rpbp$=R;XYLjBrBk~^(=S;>41<d^4(8_B<{}!q zc4-?L-eKNHB@b}}|2sv^-*|B(MIVX*`Fa#N=hnAfIE%c48M#A4Q^GD={AJBq9-0;l z7A&A4g~%Y4Uf8vup-{k zzW}dg*@?NCSe;$Svrp*rBq3l8({ZznZ;ajnBiSQ^=$O1&Y;H2m@ap$4_|8NeUUXS7Itc zltthrPLHs(Q)6g;=4EuF$5urP9iCAV(B)$Y&F9`$>M@iqRDvAHDVI;V+gfEoit10Hy2r z@rxE41yy0|v$>Y4SKXf&%efG5?pK@`CAMM(9HTHQiLjE-xp7_*mYYnMzXb{gC8Zq# zjDw;s<`b_}k_Mt+$G~Kv9v}tyQG6YPR{=@FLhgXBu0N-*0B=Nh$Kc(1IC{qfv*!&b z)Ya9E^fA@Fh5Ku#RN<^sh2BM!MI)4mIN0Zm<-B{h(9d*vDqjz%Oh951qvbbl)$8rp z-?1N_<$CN$R4j2wgtUNCd6?=vDnHsBG2YQa9k&d_vyQh?!nL%u$vKO~2C$0$A0;s3 z+9KY5}4h&ISC!=ZLeCh?>8|R}%218YLB=BY<96x>u z%`C(HPwr-GIV9qx8?SGkjL>RJm-CYMi|1gcF92=UPBv!H6TssbVugk+=dRG@p+FK# zf^f`kC|35*j4;WFvr{?m!g;Z>yD_@Y=b|j|B^3VX3;B8Q%NKnhXY!+v^-FN_JcFUe zTh{lmX+O}U0P_t-#yS1MWQG_&_X%u(_)}wVv^KjU!AKS=tV*nmfB!B!5{M(Yx^T?m zKm-AaV$@CN{_fJI_kAAE?P0=gsL%RL(+Zq==tnh)7c__a))khcF$`)*wYv+o#>L7y zeUt}e!IC@;s%LsY&D_0~{S)f=)9{edV3RczX~mI^pJ3a4fNqH$iHw42Ry3guM((zW z#o;;a?-QFLqbWJ@0OqbQk9!l@fx+ZE-8fbIFKlFn?C;$7GP@HaC7{1JMp$H1m@6c7 z0&0bc0WKI@(IgZ-SZ`hgp8W`(5#`aoHKva&E?%shC%JK>HWa+cHA{gKU#w&>exNc- zE07P+hp~vA14@piqUS!ki3+Q?m)yHZ-`~nkO&;qj938C&8$%Ddf2T8=F7iJ4A)t>A zB+(m8%fpBL{)~3g#umU_dy*pHi7xL+j5=xVYteYI%2&Z+?{k2VN8(BKj` zo?1A=dV_;x1MY9FjLg#t>kWdcq)@~vV8WP31KD8Rz!rF2ao?qi_%TomWUIqTkp7Ye zS|4OkRH)-jsN0+{3dx1k0BrJlxJzF}oI3(2$`R|M>e$u&H`bv`>rK7Fuh7BH`{frc zRG^GhUb`9)&BcZo?k){&Dk=TN$7t&Ac*7yx#VonJX3ZE*N{T=!>$~I= z5I3rQhAp1b5jP`%*SjCRpTziqehNF`vf$slf(4+gj3^u5CD!orfP%x1N%|vzV1s`@ zVlbk3>r-H8Xy_;iaA(uhWRG$8lx0T%#*tFdxezB???r84qdbd|y^wnw`Yj-@OX;O2 zW0?IuxU2M&<51?2*0g3RDpu^i`eYA==KyyxJuId9Y0R0kw9wDeC?mLv=FxjmQG_?k z@NQHQ9w!mWVxlwE6dBLWv^I7M#@CW10`%L_=mx0Ico&j2*79261a38`TbO#SpEf^}txD`a$xpV+yyQt)_Sj&*gPMnr!NWz(68U|8nzrii! z@w3H@)>8295jZOSH#T{SIxJKfe)aQ&KluN_O7tt*Aei$3UP&4xy5Yem{T zg_h2q)N`oSZOY&(mEq`=leX~c#r+z4$?oa zM*>Aa{(Rev=!f=3oC+dag38vE`ZIwvK8nwN18p7#%E)CEC>$t=iOlr=$@NG`m_s_|${1PcsnoKXh5XhF8xmZqk z0Wd3r*ro+;+$48_ZnQBOP=yXI25b5r63`o{?I0^eCsYpIWh?$7;O!5%Jezu9m6{3+ zJ{ZNbu(ek6*seS|HB@L&sLq$s>YGxpt)C{v7(@)!>=_y!CQUCpk`he; z$TGZIQ5Iy;`6(v>D5DZx!y>JDZSoY&F@4UPj4GMS`J~rHu}bH&ri&>fMEh+8yA3Wh zl*yvbGwCBpEg~41_c0qV#ikL9fi-NqgI?X2JkFaEa|=m*(BwW@#YT5vM!%ZM@Qa<* zjQp878~pCvA!V)D)~!bL9#m9RLQbzU%@7lBU;S5Jbkq2IOzj0QmngbaoUG*&^WZ#X z&$sX2eBSj9L z3O}XS8wC3-*NWB%s|+9FJS`{PWA%Ecl=+s+#*IuKSsC{I;J>CaOtF)(7z3aG!Fmhp z*?w>|t+&qxgT9SCx38(Ct*gs=FLZ|~Cj`R2KZ;NeU&f~p_k^bnl(K zp7B5zBP-PTYZdRwK|CO@1YjS#QE^tntM1G9H`hYV_ND~fzHQK<12hF(VJHkhVPnBv}68# ziR%Sb(hk8X3p3^{EzE(0SmIa=jwQtF4VU>+lrfONr1tDTAAkFw|I6dgk9zsaiQJhu zj`0}7os#G<8jym<@!3fuEJLBu8KXZRoO!nKe|pY9Gc|~Rm=E&_LD{aiB zF^{1`9Zo(iw-|;7jLaQb5!Ua_*=CG1gUU6f^Cz{ElH!2_Kor{g zsDk&SG?Q1xnj$zATMPt1Sax*7Dm?eOvZ^1^D~W_^B!-~MCAj**ek^)Ae5bX4k@=>XId zFb=`DfCQeO`Vr)91^gyXjJ>(OHD!wCSGmQrFf#c11lEg#q&y^%4r7?TaD;PzOKrnM zC|f4Mc`^Sblo3iveG&~PkgNw?5H^G!C}Z@u@u;ZV(*9ig8sg!8jLdbQy9glVHtr0V zkty1`@|lfa4B5CNDa`bIpPLi0J5$GZ^Um|gUdei)?`KdGiq^GNo_h`V>; zCG-Hk%>nP(k1fUpmppJ>BAEGb|U;yD{IkRr$Q5DlxLR}Qk8l)qbqdIN&-JXLu+7B8J-t5OojN?Qw{s+u*h%W%19zI)q8gD zmUPTSrEfSSp1(S<$m;YAm9(7a5)4NEYRc@M()oFL(b*h%Z0glUI&(4qPneS2vQ1Nl z<3Uy{4HBn)N9)Xv#BMhrssZghN(&Oe9EZGb*z1>6@Q9=Us>g{|v>nj}-Xo_mAi24* zQb0P0tD#f}KmkJOomjZ#V2n<~se6nHo{Ewyam4T9uRNC>I-$I0*OAZGze?eYPN zS(4QxwJmbH2W0`BltTWBWC{`3_GVh~*GDhht8KYaxQc?&G@6vK9t{m)F67ju=k zc&Ara=$6y@#^Zqdhy9ny+y7;Z0SJo|b*W`1#=ooqPW-yKc(Y*61~OWKvhRNO0A=m> z$iHe;gl(*nlS?f<#FEC~)@59+(boOr0|3YGd~?_mtd?lad4Op)cm%QvX6(!Ovt79y zl;vf}L0fCvK$<*F%PMG<2LEW#@+%mpNBF^ZYel#)?gtyWsA>dC=y}eHbpuP%DjnJr z`#u+Zss!vUfDCo&A#CD4ZyyA_c?CdNWNCNpQ|luLE_c!Tc`*-xFrI>kSB(U>YW3<@ z*ADl8tm*UW;$!4#Mp-Zz0v4D`sZ+ybde_p@@=N81#7{d}CBa*Z)op+`h2YSg^VL3EaMd)0#83!J54DGh;Is3XL)tE=~JIRSMLX`N(2?9z)=w(#B%x`)8@Sg`wGdQ1((5t4WpAYEd@L|ITUYvRpi9SVY~8iTF=OmL%g)@M0#vtMwq z6ix(!mK+3344x@FlIksD+Jfg^+jT6+r+Ft#e6Vtgz5;NGz%mBIgr`kkfpZ3k7(fN2 z(~?C}D-2uEpdC>pP|E;A;LJ~vqaBUNIE+H0O_gTb5dbG|X*{C!`DnLo-;_TIqQ{=a z$wY8?q{B2WUlzoD^o@kz_y+PZ#e)}!FUKZfhip9?+G;?Jm){`?Iglr-isTm zn5XC8LDhMoNKp$BJmTOfC9@M zYHADEmqBed^g0N)>0UZ2cE!P5L^ewaPj*wldee-dsiDaoCCoQ+RIGHFjwiXZUXe|7 zhYMsOV*L(Kf4?;Tiqo(f9W%Y&WpP)`CG(cdsX%RhJAVB5`EO|$N=a)#`0A7TRFx2z z_8qQsQm2-oI3=&5!U6t;5igchi^+N&MG!dAJcEKP|51Q%;LD^=8NfZGs1vkNEVzd$ ziF(i7-9Xl{*x>PN6$&0FC6Q?*iij9+L$LyBkj1Kc<8eG)?g3_`H_?% z7)b%TbjZdP-yS$BC)(VJ4URIdtsg_#;H>D@U`r9Ep9eB{`}J=#z$Qb7FWM@&=wQpj z@v4%NeJHvEnDE!>La|l}+Vm``eSrJEooVe@eKs<(z1XDveSkN=cT~^@L!tt19~lEC zThs}JXgb( zTY|}nOA_bVo zko%oOTy{W%=Z+FQm3a(q?xZo#yDEeUQ6$P;Y;pjBhIRh49#V!nT>cwahIGkek~xsH z+ed*^ee!N7j};1#X2m%BO=>o+7#aLA*6?NM5&Gb!g_IwvkQHFc5FHiHn(p-|E4mqo%M{#XmaC881pu9-l< z2)Ra(dp)sSoc$Jz7je{>r+It*E6!9Srjzs{4FmeF@1A-_3?dYPglreuf8Di?UFTE3R z0#m)hyH;eo%7^BvFw&sEiPc0(1*9t|J8VWQjnklwl`bg{B^Mtk-kpO`haQJmxa|w@ zYt)@F*=|c*4p0fm0OEl06R{ji&Llzb^ zCEa%p-gla~EiKd`mB`?WdLA%2g@PMr3-+QU5h3555}&)8snF=}2DAm%cyQ}tO-kr;2UL+Vfvw2dT#rztJ>ec6m^85NX_8l7A zw5-j|!Nuhcreq4Ok}|Tg+P;20)A+up-bsOm&F|cMb6A>XQhh&m?P-%En-*Fsg&xT? zZO2JJab1q-^}D0XpPUR#Pfr(|`g=*c^x|0#9v-1ZRyUH8lE#;MxVZ)6zcIRQY-)P1 zx+|)-`(ERx_O`awB_$=tYB#T0v&O9WaopL%fvzdtqklh*js4Qwdks!|#wI4MLql@V z1uXE*J9cMvT;tG$;t(} zxoW~bYF;_!At51O6=yMeFDYXbZ^&+}A?I|OagXa_vA5s8eQQ5k(dqbDU-GNsFO1&V ze4M#PmGH@&F)z>UHNSJF0K+uLCbSV^eiJ0Jrr3Gz5B}?K%$S%hy!gw+*dM>>gaCD( f#{cODe3-whv0Ybt%C&>wiTl?yL@4i3p^9LW_-e=8g?X}ll zdnY4u-W;0&LkCbO6r0eHpam34e}F>iyTYmMx-;L2QoY=gM!8*N$E0lE=>bwG z7PEjplrPfI&nUwN!u=^z2g|G}3FAKmSThJpDX1(wo6*p8gEr!k?iij=nJJag@;2l`~75bVXZ)wljB?Ff;m9^dDkrMN>4>tgQalX)0&O3b5C(YUEp6% zZwpXoW^AFI_X2-ilvT=`uutus$d@1bMEsLkK1Iu?Mf54R{Qt^|R2s;Va_+?>PRxi> z2H4)rI=(IG3^z1p8KBgoN4cuG4#Gl?>qfz>*7muY{kN{8g_4qkzD*4|^?fum7G+gn zJB771@-liJ#}NZmvjss$z}u*vo4>1h1e4cNh0w9R=?J!uEP;7iM~E#G`?1AYCyb1k z0E518x)FP;Td)!P9GaZydmO|?(%h;rL9PenG}f!9z7-PnwR5flY#Yn|63SO^C=)N; zWnjL=ZMG4;8fE^>EhnjBG8%1Nzo58@Vyyv(kS5;KzN_qu_p>T&X|oO@zjqENRJK~7CCSr6afy?lKB`w~N(vBHdr|>agzV`6}HOb?&XAQ1vQ9BxNn;$zf z|7EJ-fIVcvT!*F@6ra>c-)|6yUIHifp7N$ofO)UWIN7)**S_ElH~5n4Dbqf8WXe?K z7qxQ@CTp`}M&qIScg0$RB-ees!7N^&${=$2t%_3yU)sVvT}+oJ8SvCufACptI2I{K z5(S|KpSa4xfN|zgUNh=v?6b6IaoeF5ui;9ph|LTs``Ql|EnYa=#g$Mhux|1up-UjA(_I}~|l{WZ-JSv)7D8*LpeD~%S5lu6%l zpPdgES|iG4bu;)wWkfa!-1}>q6;d+BqFbNP)Y*u``wm2cw)xcfH*q$k6yS@J2%{sM zk}w+qtZQ$INa-$2tVy*hLkn$R`U6V!67{P?=ck=KLWKFc@&n3wZ5?~0_7b2#OYdsA zmS1u5G`D(u$_9P)IJfq{39DaSbJ>zW8UHF6)+M@BX)>ddSiGYZu~bKh5!+Y*Li$CF z6ecC1X0q~Wg6_3OSviV+0s2<-XUxz%YDC_U0P?jKq!;N9gJ~J zdeD-O=inz`wX#N{i%MfDne?lHswZcnQso!K0qqazvJAsAzr+az7)l+zbIzs>kO{^q zRX!SnIa(iuZ;e+tttp!uZ>CJ#WQYMP;Sx3nJ;vi3)8hBDfbNygFpg{8;2=X1dtnDf zROO+~)&5#bQ$nrVi|RXD9!8LMfT@aEwSA`eFjHD5vo?itA(w z4?IAGFW*N*EH*4q)E2SZ?B^M>Z^jp}H+YB$Qf!piNVP4zmX+XX$f|$HY;k0Hug(Ij zgx*9f{L5&e8}lVu(k;DGDmJ>cm25NrM!1zHE8B^@^U4juGrj|!z+JCwUr*+q6D2HQ}Rk=9AVA=B#93Ym56z<0p21jEI&(?cZ zh@>_VA=@|^JM7~{f_Z0lbbz$6ahH1mrI>)&H*5{m-$q0N$Y@~-*&UxP8p_nLytU!v z8Ue2_{A1(3{RcTH`Zo+j29UK1ja&enEfdSEZOrS$=7J3phddRy<8`I0hGSS>Ha}V+ zNgU-ffb6LS$Dg5v@9DZB_%iM(qH@uP`lc77$!zXLeglYVdD_C@?^7%EQ-}Efq;c-c;d1@!+|Szebm5~x zNpQArPvc(X4yrs%{de6sOJrM!sYjoZDSFpD+uYUHf5m494mUODrTyU;TOw*%C~J$- z3R!7J>?;$y6TM62c9gn{xFqm(2dzB`UN)A!6K$ggl1b=CnKy`Bohxe zlbQ)+2N8D|F;toJQmE&C18E)tRhSelV8RbKBJWBcDCrY=Iw8>W_ru{MQ%fM3LN_#W{e^rjz4oa3d9t z=})jMwSyL$eO5ja+(-sJ91b`Nu$AN#Z@k%Zn*~G#UpL;BB@cvS2K_$UnEtR<-Qpl; z(`DbB>9(euc@( z*t!hnlL*)o1bmBg7UIS-spZvH?A(g$-0o?(?$m8}R(ONzT{M%XSa$R(G--;Z&D{hT zMJ68qD${71zLtpWFA=u0WX?Ea0x5cHx7HCmZcFTAj3O&HT@&-=!+sAo2HrS6drgp& zQBzTpGjtG-Rx_wSIKyDEaYiv&9^e?SUyWi)Q_+Im&rRAKLW-RiU|d(4M2#-AVm&nI zPfS=M_jipYQMpQ@@|%guS2FQL&%Z*b6{B}?L*5`Ey~3Im_)T;)&Si{=OUD#&6z=~E zBu?i@NGAxSLR(e{9u}iW;+-KO9VQ{IB#>}U1i}$()^sv~BbKbY2IUFh5yj$U9AKo0 z!x1Z15I*)Ham+JuTr(OJV5iCIH#?>L5Y34UlV)|r+dHuDHMm%_*;PJ zI$3^gVAs0}?9P3xDICjI1B^1QCW;&T`pEe1@19`%h7)obyk(XL zI6$#Z0SPV6>npb1=e?eRoy}MrcTjX++n#C9(D}_i#kJqO6iD|~NP8w}B~ zY|WsGR7{vd*JS?5S3wfs6?D&IZKcBEKDXtJe}oY6dL#*%<_(I43&>)94GNaMN5|x` zr7IIQK91+G^O#Br{hyNGfEieqtP81EGLjaz)?6y`aR-$P@u3*yL|+|WoATyPtYMymRt9G4HC`Q%)kn{YI-dA!)`$^#U0bhKu)A#DmTXpNB#fEjb{qNryut#?CeFRStJ_1I z4ve zTp%rI)5ud&uq%}v2C~Zl;i~q<6A6rg*T0wvt%nejrP%-1-jz@;jjajy*|{=|wEffk@8f?!|e# z>La%LK9+pzHyo5E-z$@GNlV~}HIl@R$KK8yHIb`5j=Cnlcfv~juO)t9#g(2lwZ;z= zNVyN*rQLE0H|76{gP_FC+Tc^%rXbT6a19RZx-P^l1DG=1p(E&s02q_^1R?K4Fx{=K zHt3eqe1eKSw95xPqAvN;w0D%JYS3$iSbq9oio`dL59L&B^Pk&|a?ra;&a5-sE5v@POxK{B z^#ieT=PQ$C?4Y0RqfKU;2}I3aPbBC@v<|$J^Om`Vm=Ix(ILrt-#r@#!3@XOb;MUm$ z^zBRN=Mo$}$%KB}7Y^*)%fNDVlS!C9eD;bvrp?{L4obC;HHzVe71Of`GzuIu_%9JQ zp2Th%hY?`*S~$F6!x^r*+QG`DpuZ}Q?rZ=K=reL7H7kjlXhW)|+EG6oGZh(kl}Sdt zv?6M=JL;63JJ1(ica8DLSnUHY!PgT}+u=3qzf~N56vB{nUzOlDt3=vEf~_xD>WZXe zE037L=}aY+v9HKw>?C6x+YU+dy(B)O4jyU@h;%z>jr}^4mUx+d%H%-2Oy5To&?nI| zy4DQLT15H1RmOb>C+A`djF`bo<;esOot$cgI1uBk$~0fZ8^RWr33yc+9Qgh_3oJ$a zXO17_^U8Xeb9+3N3yAl7Zj%Ne$9?>F^qiHmJiz>(bJ4V+d;wRBUZ0NVS)?vTQ-2dxiBTF;t}8+SCxCu12XBK9}oZR1jt?Xb2dud zNyzh&hjMChP9>Je%xZnyiM`jM+zGa6#8wq?vefkY_@^g!a?c=3i2hw~A=hIuOY~Y? zqu@X^#KNJc{qg7hgO6hf>c-LF7%M_dwg(ty_AQmq4mBUE|Fj>HU4h=!FSO~27-g*| zSnBd7lH@mQIL6__EChHipr#W3(UKcMf-JNydEqRQ5j&ir;ys(kliD;N*G0lk>gJA% za~GABfbG&mn5Fy=+FUzDwtM`_;hwj$|4mnIpMn%g${lDyI-#tnSIh3I7)$>|Q{>Yd zTl3=k^z_)=yZzV~*~g-QRFOz@EsNswgO)k8%8_}wpGNZa4Igm1m72RQAp$6JsM0*D z&O>*2-#|_m1;YKq5|m*(xiafR)rHxgYxquMi+D5}^`&BK%3jOAcuD>h6f31Bcc@gC z@_pTRk5VoBY(*6-@Qnj%g9u}t*ey?;_uk!|Nd4h!^{;tbtPZDYR2y{dPX!_&wEgS9 zU)*eUYNM$2T-fbVU{p7^t=>t%?x@weXM|#O<_D|{Y;bS~6R3QVy^Pz|hcC?DOofi8 zx$j0UZMJ&7@oekGud0?r=bYGhjoT>Orv$CRq+|o6Gt2I*N7Pwz>pgq^_i_cpU zT=|;ko44B>PHzFG1myEG(H+}4Q*E(gW7_S{%ZoYdW4~{2U-fH6YznVSWed+1xhWe| z^<^&`5J~G|HCNj*NmZ(uHxtOjs9V6tZf3jNVOYCm`HUlu2*%bLlN{p#QVOJSkK-!PQY zBtSY?p`Q%vI{oCV_}I~dm3pzLwZ#!yIo`H@2B19dh>@%6UQpcLCLDm&M=KSZlC(%M zSJz?1nt&rFlH-azOj`z)TgHx1zap^fS2mJ(D8@U6PE-o%NMx0e%^AKnx7um+EQG+-1XU- z_ZA!SFE>}h>&Dxvu5JRqq5B8e)P3YnGyWI}fBX$c2V?Eq{Q`ODvT@kYN+_Yh^MVRy HF5mHA)oq6D diff --git a/docs/articles/opt-constant-propagation_files/header-attrs-2.5/header-attrs.js b/docs/articles/opt-constant-propagation_files/header-attrs-2.5/header-attrs.js new file mode 100644 index 0000000..dd57d92 --- /dev/null +++ b/docs/articles/opt-constant-propagation_files/header-attrs-2.5/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/opt-dead-code.html b/docs/articles/opt-dead-code.html index c850e14..379a75b 100644 --- a/docs/articles/opt-dead-code.html +++ b/docs/articles/opt-dead-code.html @@ -31,7 +31,7 @@ rco - 1.0.1 + 1.0.2
    @@ -56,6 +56,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -74,6 +77,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -86,6 +92,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -112,7 +121,8 @@ -
    + +

    Dead Code Elimination

    @@ -128,49 +138,59 @@

    Dead Code Elimination

    Dead Code Elimination

    +
    +

    +Idea

    +

    Dead code are pieces of code that do not affect the output of a program. The removal of these unnecessary code lines can be defined as Dead Code Elimination. These code lines do not serve any purpose in the main program flow because these codes will never be executed mainly because the condition for their execution is logically impossible/infeasible. This removal results in performance enhancements.

    +

    Background

    Dead Code Elimination is an optimization that removes code which does not affect the program results. You might wonder why someone would write this type of source code, but it can easily creep into large, long-lived programs even at the source code level. Removing such code has several benefits: it shrinks program size and it allows the running program to avoid executing irrelevant operations, which reduces its running time. It can also enable further optimizations by simplifying program structure.

    For example, consider the following code:

    -
    foo <- function() {
-  a <- 24
-  if (a > 25) {
-    return(25)
-    a <- 25 # dead code
-  }
-  return(a)
-  b <- 24 # dead code
-  return(b) # dead code
-}
    +
    +foo <- function() {
+  a <- 24
+  if (a > 25) {
+    return(25)
+    a <- 25 # dead code
+  }
+  return(a)
+  b <- 24 # dead code
+  return(b) # dead code
+}

    In functions, after calling return, the following code would not be executed, so it is dead code and can be eliminated. In this example, resulting in:

    -
    foo <- function() {
-  a <- 24
-  if (a > 25) {
-    return(25)
-  }
-  return(a)
-}
    +
    +foo <- function() {
+  a <- 24
+  if (a > 25) {
+    return(25)
+  }
+  return(a)
+}

    Also, after constant propagating and folding we would get:

    -
    foo <- function() {
-  a <- 24
-  if (FALSE) { # dead code
-    return(25) # dead code
-  } # dead code
-  return(a)
-}
    +
    +foo <- function() {
+  a <- 24
+  if (FALSE) { # dead code
+    return(25) # dead code
+  } # dead code
+  return(a)
+}

    So it could be reduced to:

    -
    foo <- function() {
-  a <- 24
-  return(a)
-}
    +
    +foo <- function() {
+  a <- 24
+  return(a)
+}

    This dead code optimizer also removes code after next or break calls.

    Example

    Consider the following example:

    -
    code <- paste(
+
    +code <- paste(
   "i <- 0",
   "n <- 1000",
   "while (i < n) {",
@@ -180,9 +200,9 @@ 
    
   "    i <- i - 1",
   "  }",
   "}",
-  sep = "\n"
-)
-cat(code)
    
+  sep = "\n"
+)
+cat(code)
    ## i <- 0
 ## n <- 1000
 ## while (i < n) {
@@ -193,24 +213,27 @@ 
    
 ##   }
 ## }

    Then, the automatically optimized code would be:

    -
    opt_code <- opt_dead_code(list(code))
-cat(opt_code$codes[[1]])
    +
    +opt_code <- opt_dead_code(list(code))
+cat(opt_code$codes[[1]])
    ## i <- 0
 ## n <- 1000
 ## while (i < n) {
 ##   i <- i + 1
 ## }

    And if we measure the execution time of each one, and the speed-up:

    -
    bmark_res <- microbenchmark({
-  eval(parse(text = code))
-}, {
-  eval(parse(text = opt_code))
-})
-autoplot(bmark_res)
    +
    +bmark_res <- microbenchmark({
+  eval(parse(text = code))
+}, {
+  eval(parse(text = opt_code))
+})
+autoplot(bmark_res)

    -
    speed_up(bmark_res)
    -
    ##            Min.  1st Qu.   Median     Mean  3rd Qu.     Max.
-## Expr_2 81.07803 76.37139 70.88553 68.57435 54.16527 186.2943
    +
    +speed_up(bmark_res)
    +
    ##            Min.  1st Qu.  Median     Mean  3rd Qu.     Max.
+## Expr_2 81.79659 85.27682 57.0721 63.33898 54.17745 45.12142

    @@ -250,7 +273,7 @@

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/opt-dead-code_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/opt-dead-code_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/opt-dead-code_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/opt-dead-code_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/opt-dead-code_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/opt-dead-code_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/opt-dead-code_files/figure-html/unnamed-chunk-8-1.png b/docs/articles/opt-dead-code_files/figure-html/unnamed-chunk-8-1.png index e61ef1a464aaacb1c3caf4fcd74170b848a8834b..5c992755cfa502f354c773b9ef4009887140125e 100644 GIT binary patch literal 39933 zcmeFZcT|o7#J8f zP|lx`XJA-q$H1`S>Y7#fH=4_9weZKPYZuR*!Jqgwb2`Tvf2_T6{<0+l1Iq*Qd)d?O zQ9TBRKN%=zPAb|347S?ZDe4t1jvn<0l3I6_`sW*l&Hqxp;_af!Ii-0cetq`$ZwEsE z%%kj=SBQGK?Ih#+Oz!@Eceg>c6N+b){l4=U${jIWx5rngM&HS~%X@rB-?OeRw)QIP z3du?_%bu2d9B*6fd+fzq%rewzL&E=*CHu1c^Y^{y=MU*_kjam zW!-{en_VoEn@q<>N58heb6Kf4@2^U|aPgv&nwqgi=$ERmH#BPO8FLw~EL&V${IFi< z=1qYmpOwa9tgNgnSFMsCS-!PnV1UQj*_lGE?NhvYGx?vd9JD#AxB1F4EiJ8-<{@g> z-Mh>44PrMd2>qHo7s*Q_%jtjr?_d8r4-Eg^1pnOx|J?-t_hv+UQa0lvLPAc=O%2t) zJpJIs^XD(;=Wl1Vn$672lzhGMy_6@hu3kku{@JsCrM{v~gv*g9cbPM#va+&fM~hh} z$E%K)^;wk;+3@`N=ZOtFMHI(-Dy-i#seIJeNy@G7?&|vPAuCbcQ{kUxP_;WtBVjyf z%a$##%v!Vi8r02|baiz%g#AVfvdIlA`ERTh6cn7OKF8YU&Y%BD_V|KT<=p-I_hT-4 zAIR>D@Sk{aeEv>Kf5^7n%qEmE=`kE|t2^r{~kl(+-pMDw8w4!P=kJRJOahxz#L8wO!=gP&jO6h8@^%52THt}VQ?g5Tkc`T6sIqsp^WE?>UvJ+Z{<;@3GmoLE>`=&<0o zR#o!Km5}*`1$K%|q{936?>Y7_zj2_gt!>@fwU=ycrX)8NGRURgc64N7WMqrJ9rG5& z+G9(-)5Zs#)XG;K?0XwHLpC{S=Is*k>KQV_x#Lg9#>XG)Bq2zI)C} zVPtv3sW6Lknjf@-~n86@fo#&Gmv1nYv!f7?h*d0bcN4~DxupvHuKJ&)$rTJ8Bpvb&i(>vyF z-@aX!VXi$n&?tNL>Q(QFOl(w#p|r-=4wL;C_zj1eGaCO|yM=N0Zl|{5kdBpcu6LP} zHm=huR09+o2GopgM_TC?Br0h%2V;NkCNUgK$hCK zM_uLRdWNe^k%bK9|4^h(>t%R zwb=LXA9%azxX{$fl`DFpW-In@f zRpkF}KD^aNYS@+)JV)g353&oD&zk$V@9B5MMZutj3ODP@`Hp*=j4S+vjT;jJ!^W!f zNR0jQ#~+G{idIdHg{~zP6-pW!8sQex_jj6o8b~bAy560@)_QxqLM>JQ@e!-DYvSn#8!6nZchauBS&lf90xnf@2;`9VGwd$)bol=A z!6Qeq5FAoIAprpa+<9ANyuG|41av-p#}|1iO1L?E|MXHV*DiU7kfGdA&{tw-U?mmMoB_wL<$3hBc7vBtoYqdz{?${xR7E^RsZ zS>V{{yDdksp&n}#NX#IDgv?qF>lqkaZNXe1`LT~5&wF?*1n)G@y|kYBfKbH1=Pb)e zYz$SFZM3HbFs&yj5j*BEAa5m(3+YWDH(lDJh3+Ja_Xb%q~zpe zwfx&zckkVkma)mk*o@n9?LVAvL&h=wTI6bxrNOa(1?`z>oAJ!`nm6AwsU=@?6`+b9 zJErNuAzf4Av0qw7|5$T|d7|PEo4y)J6bvP-R!K=^!ma5a7tWtQ*weMp%rDCfm0uMGx7_la1x@jT5yI(h6w& zN=l3ikXaSg!Up|J5-wR-q#(EBg}?sDpEG;BORVOz424o%u(V*lWy}8w$Xs|>VqKjgs>XInJ0w`ofB@5n&vObUN8WB4c(a;SDBYkkAlf{C z=EJ|oh8@WpBJD4NTrNxJY(Bt zAIVnIZNViVa0wesSj<_*j zDs0l=6F~SD1_mEeo&H2-$3pha4kUWBdA@jI_~o_Hq}kW_(9peDE@>b$4>pt&E_Qaw z>gsB|t(IoCRdpD}PG4%3#36yrw8tGJ#pM!`sNYfB1UzLQrIePP0|XBpl9o|E zbm77UA(IAPN7M`|8`uH9%PD8De=*96%;EgGiHw4UiqAnv5%LNO(o%F-R(=9SOBJ!? z#e4Z0Z?0!hP(o(r;qIOuS4eC-?ZuGq$0o)N`YwqZpom*yVcU5Q0IlW4NPpJ03lKs)AaHEEMG8!ho8+bLGiZLqmKNM2G)YUR5~*Hwo{uaE!(1?fnhZQf5DY3CC=+ zf(08758uR_ogc_n1WsY2Ad%N249qRKX~UG;!fAKsNEZqJQ>X5lrjMBq0plvVGxMk4 zE>Rb-n11zCLU2kt@Q9N0;&`yMlv2)X^LDM;NLkVJc838zg7P>x=-r6$9b_&Yv#jo| z*drC67ZLZKRO;T0=IhsDE~==h#n_Mc^Z~qxOiB9+N=~=kN_-e?9C>`9Pu5|6wD|E5 zi5;X%(O)|!%8X2)Rd1lJaZMQI!_@5L03LTO>v=9aJ3G?nrwD2PM~(`SGWFVvW6a~H z>MfogoAUXQ*}0L@y)SsnAAh_C0{2j$=dFG~!MaP1vlG>sH#;8W8TP*1YL#tN$N z-C}_<|EzD!Sq$^$rehP&fG@6?I?wh+k`lLW{d#FBYn#=ZcGrEce)ZUB<^25oi-3T* zX7hs9F5QMdcI}D_zv#^^F0Mhf>iW1@I#@yzqZa&UCtkor9vdlKvB~hmqwOh8r3Kw6 z#{^lO^7Zva@R|-Z#P2wy|EQ+6mYwqQ z!eR71bA5v5QIFco>SbPhX@=E%I}4CWn#msb^z`%%_?s$0`ST{I3Atkz6a?#qQMr**1kc_u zZ_B+*wsHC0TeRnP3}UA{L}=#NIP`^k*QMxPq@Yua2BLoP?3rAr`whp%ncf}Zc1bov zU-SkC8WV$q9QwIdu3p_}=#~BB!%p+Ma^FMUU%&35%zl6MlvM2tgotKf;3c#00B9nP zSZ#Dwb$s`|RU2y1BE&ihnV&s-U%$dH{I9=$Lz22c9kj{msR~Z4es!*_wl-BQ zQCoYmzy2aYN9o6ZcC7fHkBvk#CVXqk6v?=xs!EpnA}}!i=TAG5n>*gzQ*x?62bMQK zR_1^XmEIG@nN1GckKwNF;Icr>~GnZiu@)i;T}slIJIsQ0vaeqWyPP@0OF2D%aE8zkmNZY?RtBX(o?1=TI5{ z^{&@x)*E3opcA90L#I|If zu*1vMU|?xRwHzd3kx8Sia#a$wb_iVi3lEAJ6zljFr^DBz`O~E9P@9hR+YB5Yg4X3XKJz%apaO;iw*RB!|K$DD#sb+JYW5*U!c9&t? z&a<__q$y1^u8(p1BQ(aeHCqXATm=mx8<&ufA~Ja7U#kzSptpYi^n$lWIqfd#^w0V$ zt!pdAuh))qHrGEVog_tLw6mE1du~U^6%ujB940(nM>m@#62l*GQf@6J^qvpnvWpll!Abb*OxaO{oS0_clp9Vf@`a=f-R?~j8{Pdgk3{mfZ zM>mO&xM*ratSC%>2UBQ}5a-+Ee>Q z$q(Cu1fZBS#8rbp>hA5`PeE8zBT6TqI?tcx;^G4R;#0y!3;HN$qCPRd1EaEH)$gD1 zL)OE2)1~@6&TH|5)!Ehq6Uc24d}Pq$u5?ePeMs&!QH~grwmx|K66I>4lXdCSdj8xH zwEnT+1E{o3(w?sg+}q!lmy>NfoPZQsQs(ZxlQznjdk>n2Z2Vc4H&A_jjcoohrco_X zQRg>r-mF&Ol-G3AMU5Yrm&8y*f+kO7t^JqL!om|sN!%hL$}C5$f1pCo4L3{o<=A+?K)+;Ln}l1w3Jqlh z?}d(4^6c4|t8edH9NvX8uV82x1$M5&PK`Eob;eNCT(}81m2jMmLaqhEtRorldb7yT zm$Zw==SSWeu}1P7IM8rs#cHdL!aJ!F5v^t_Ao|}-rnLDw5{oKBvytgw@Q5ck;KoAd*f}1czd9$s5@tb0bFFkTLh+~(KjzxPzG8s z{uFYD(0SwqIXFHH{>{+89jM)U=!+5xtEZw?PeepS|J-mvBajgRU@YQxa{Fb2#Ctot zy1TjXK!j3G)IhUo`UO{WNwHmwjJk{;klfLlDq+wdP5#VjmEod6fuz5VD2;e|dT!^* z3LbW{zJ$fw4F;xmDl$C0@A%TpQy|hM65VLamcL)gPz!7og<*ge(hOe!8%%co6V~Gi z+DMo0q@8dY_lKcVtUT%Y|Yz zpa(K{9pK`+1l$1fxzhf4PU`^#A2y5Yg9ki-d0@ue{{CC;%$dK6ii%Xl5uFSSa`FAT zZ`8HhrB?s;8=;FLho+~dlzS@!34S1m8%=Nnfy!q`mnZE)>!+hQ^d3z}X<6B!b>9XD zO;PyGvKGViPnR+R~hrk&TriK;57b3 zXVcLuKnBo3tuMJdeeEAA5(_@M*lB0-y>d6M; z1Szk~EJI~Rm?0z`+~SAj*O`lkY#;#a*k^x8Q1FG9*P083hyL*bL`+OrbykrLVQ+75 z;P@ydMKC-(oQscdC~kUmdKwG;3~L>6Zw+D*2(B*Vxc$Z~jej%IeOjOs?(*wDpgrEs` zWSY}F;GL{Fx{axKY4!+{TAFdVNn)PYrE)SIHy(1@DJ^n175%lgzrCQq>wsnq;pj=s zu*GW^Xtg`d#b;afh3-D)5HKk~{xf0HKR(+7u^%b27B3Y&Gu~?gCYH46U`u3U)AO-Vd z?g$twP(0&W#h^j}S;fS}lolAJykEQsd%|+mkA=qC@kYyR&81K%V-pkN1Eb(vzM}z- zeq9BKooZ0kutiq)zjDSC&6ex7Iw3&q(rRu z#lMsZ>+$K6T&j*@v|FE0|CjOQ{h`{j&I+T^YZ%g0-p{3lgoLcuI>GEB=G2<_b7n^1 z^N9Gmb?Xkyv>drSazRjHodbu#t<7Ks!*|e3x*$x=YDjE%+2tF$F*7mOeCEbFn!5jLpYg%QS}I-#qbv9ALE zUXS!&X;?axKRusv-I`#f!J0-s!Kkv@&1$r z(spT8mCDVVqk0At6B92I6B7kio_XJhCwNXNNp1K#G<3ZsH#e7Cdeha^rU@u^eOL}^ zX_;sCY__~gFteOWpvki}E?-=SRSMNiMRuCF1r&GBFVvOTvY-!AL~Y>jYk9ES;XDDg zh*WlpY%ncc@6fSCn%vsFpta#~CwNY`*c*YaRKLBW7 zGH<>DMk8v2#>m79{CM2!&n<~cFE&~;v5qg3Vne`=-7ohWbI=jVlOH49^%-Y%8bFs0VLAYSk3Lo9*VmUH7>Dg-L zt(c!FbWgIf0(rq%n>iMN?SpI>#c)vVUX`^bZbP&~(%~q1kPdGn2oP_wKMt$kg2AG-rEiz5Cs16-<|~UMVA` zCPlAo0;_5ap_s4?_-KM<&@Ir)Xh{}dFW|dAXr!YRRAQa|-AR*1=Y!+F_Vpk`<{69y zD-3LL@NBd4qr&)Yu5)Gl1!#arQWA)?O2-m%(8CCLy1AYXbO<^F8UWYllt4>KY?nrq zso-<~OO7r6>;&ia@a)sw@yIX(S-n9DD2m99l@32*lwyUP7w2o+ZkVH`qt?AVz1sfP zEfo`|=-j{UNCZOxkG1TrgbuWhoq=Kgep$CXgaaTAq(zEOP%|cJS0(|v2EECeOqb#-Mcf7 zw+WnrGEohL?A%bFq@bXXX4NOZVb@`~Wy_Y4^p2D(1bT;31MxX_L2N#|$kdwUdP*~+A!s&%j+MD@?Ay2RapLJyr+(_u zT8a?0x{|x1TwHy*?%sI|Q_UazhX?@+i3WA;5|REyzoG}PCt6j!R-PI{X9qgg?X2b~ zJVe(KG5vhwOEN^WD1Z#_(MJcQkbfi=f9gkqG)ai6*b&?JT~AF-_4M#~g#v3_rIXsk_{Sfi_c!b;{d$9wmp88FW3$_-X;Vut z+APYo0-f=AJ{`>d7xphmMNhVh0I~5$tj~v9b28FRn?fJ&6pcYj>Tk&!H=53Z1YDo2 zbB;*CB;P24rUbtbZQh>m_V8g=?taU)+*{L^xqOMr%v2?U;QxTz4>qwjpjlS}vvs|3 za%AKgf*38qjT=d1(_xHYInd2;3klWvjEnH{O6}RRr+2}14@pZ5`eQdz|7z9Se#9#3 z38Y(MX+ZYSFOJ@fEu()fN_o1)@+o?9#?6~wq5<@0Ssk?{B{elor+6djN7`?XKL%3| zQ=kLosa|T6ZbmcN=2%FW00HjK(~-#sKKvEA=z}BAlS9@UjPL$`=4Q0V3idCn`yF5g z$&^AxMZ?tmhnlgr`Hi2UP#iId&gdS9l`sIX0rVpUA>%tlOry>|-m!hW5z9~F@k7)X zw^QEKt>nF3ENGpUCFoP)Dp;DSa<&49PHp1{Spu=B4h13{+z!u?BWhSFuf#<24G}BJ zCktme?Mz;Rhl5%_eRtRyK7#MCi^P98!*Ig}YmJ`2#@{?Y8hABuR05IK|2g>;MCcGS z$y|H)hSuEmB9CF{r~y&JaDW19WZSLbW7xN|k6u~pT)S5H>f9!hMctk}iTFhO%8(r} zGejWCDb47t`t7&xt+ws)W5p|7V&s&SBhXy*H5mkB3Dm-M&{@r*UymXl5~9?w(-f4D zMM6htBqiq0URb}C_Y;H$o*{FTJcsEY@xW&#&CShWRo1jd2@FqR371}bCQyM$;G>Cts6DZC9fyXa{*jj z#$fOQQ3E#gdP`>Y=ic7l$by9-^K2*rzHYQ0%S8<;9>WbkBtIh80UwA?5t3uyxhIOc zx?vCL6ntSK>A8Q$S`+dkRMNf6B8!az*^S5v^|31a5jwP22=Ri>c%vG-)EcXc$mN1V zMt-yZINeQVlGr=4}_=eZ565*h$8 z92~TKz|adtqT>_wPX`YS}+>i`eUZfh)^0;Nb&t$Mm?^|$x#^%2rLvA&K35Tjzp z3J4!~4^5r0T`Xc2XQ;ZVO+t1fS6VU0kKWN1BgE@f$4a?}%|i{q%oD(+aIj3i?gY?4 zPE@k8dcAAo15{cNDldzQPGLU!ASxWW(cyxyeq=fIqcxWB{E+qtE;_yG7#q4D583qw z&;9&j08@BqS|`|F2#4}4M{WKQyfXiVo$@=B$9m{xlXxM;FRb_}Y^P|n=m7U`?azG$ z^y3y5R$5vZb|xe)&?LMTO)!mA^#ge}hMc@t7b(kX@8F=^JpLxjac(GMhnPi}g{37s z1??upGq9q*S#M}dQn(WD8G=-51ZkW2ZMtl79xs4X?ng#7%bQf5nVr=d=|piRv1i7_1*G2rV>F6-++!+fK(Ix);pP_JK z?V)_hFb{_28{)JuYc$gFhPOxQG(0xZXy4%CQJ{vw$iWHmmZl%mHwkl@z6}FI&yAc` z;#?tm@rxX+Bhj%+J$bA#Jt{L3Yx&e(%R#&&3O4c^g{pDDTRBlH9#K62pr+(-EWiwe z#13KO7ngpd8`Yk*>?wcQLeE`rqu+6e0v88IK<(X?>lM}1xT9|$xepmclNZy#(D?P z`kAn?=)+Vh`LOP*e(e{9*pv}1?yKaWJvxbPA*b%p$1SD0mG& zDA8!F^vw>^9xK;PMrw>n@GNa@U-)6`<##I$9tw!z%1BOZC4E-LfN8Tm{^l?a&E-Kz zaTu=XFm?L*Nidkk7(PtG_QUraMHKi@)9=0xMwBb2eZPL%8<7 z4GLHfx3&J`L^*;bw=O5bw^3y)UXo|eo;McVX0AX=m^BqB zJXA%ku_xOE?mo{oNU|N)B6{4A%|$BFJ|};)Hg_q~h8044&_|M=O+Oz*3rLX%SAoF1 z-ei$+N>1(>7LCU<7aPPtrcIk7E87TF;nAh$z@xT=80rGY{m*&^l3Ch|Tnjy40jy=% zn3ySSs(goff6ROynmXi_8yy|ZvHy@Z6dmvmI`bLkNR1IvzI#FK{i7%M6Fqqu?1&_b z0?#3~e|!OM2mFIq!WgVD_Wm}@v7VkDW7yV65hy7s?}Pyp%w7KjR;Oo#_cn&<0bJ6- zfAr+FkfYucR|rJeS=3ARg?6h|05*`4ccV#y{3?%h!cIX6+(B?8%Tav`K}%lsbSYpV zDjV7r6L=-aP7`IM?=%5ouz}B^!bjk=w&SEHTL}~jP^$9S5G5rsgw94J)n(q)p~xe$ z?*G0$6g(b8`Z$7i!o9Tw(~BQ%e}%p3$GnF2$hyNYSQZh!3GT!bq)IX}Ge-ir_yol0 z|K;k6mVg5Gk{SjD2GMktu%7Uod`>g*MxuE<1U+94$=vhs}AA@Q+X}7L_5z?6>=022e-D`Wt-@kijh!hPm!$VQR@eQ1)it_T; zEK*|7AcVpJ+Ycd`$_9FQ`yqX>gN-d$Xm4C7MAjrMB7*oE&fc^Kz%ar*!5({1%y$Os z_X^&>GW3qr*B~;9K?FzwJ75R!7#IQKOJOU4#7b6CRn_NL?R4=a^LW^4K9gi>*t%iH z=>h57-PdPik@DpgloCI2`xF7qH(+K6aZNM7Xy$wJsPqE`c`D|IwBg5@wrtV)Wnj4? zf`wjAECHbE31v+U`~KY!y_Lt)p5QkdL|Z~&#Mu7IuQ6Ud5(W)$IQ{{HH87vvBH~)| zF{|&N95O<7)86^;Ay^N*ERO_|9QtIJ*k1TV>mj#;RrL>uo67F?QWpYmtuq&&(IyYK z9Q(yKv<75$KTN@-8J9Dh-~3NX5kZ17PDWcm)(oLaB~1{_pWJ)*7Io>SwWEuE`R3m0 zoy~*+Cq|}d>NMH}8n?*BvG}w2xH5MF4Gl9O6Q~h{Ks*}xGd(*g-R{Q7cW6bv!2NL( z3SK`N4&ijtv>1as_2RUj9%_!f^y+LhDm%3fgd_s%t&cX{1$ly40>j-bY{Ae|+04z& zVduO^^IZ8(wm2^>kb;`L0GZMdSv<|*YZ`Gxkr1h<@MME+;}r12a|m7kIAN|s2(1%j zsueVCPL~D(I0_plc#}4atsWlH3+%&yTnW%-)Gm+hhUSUKDJv^`g^jJ81mI@ew$1RL zoVOEcIPLG+zPgdGZ_^+>Q@Xo(5D9{Ns;n|2$HiJn{@k+*KQQ8J1_k(Ubvl1pD;pkAfv9pn&qvZD7^z zZ(=EXm+r&)^p%w6M$ z4<9~iiC<;gYV8e4nUm(O1NfLE=H-vi_eJQDPF~3H)1SmxOC^R8FyB1b53i-@A-4l7 z-M(hS2H||CuIJ~lGhZ$hx80G5)dzJIlC3$Mt-@3__4Zg){4SRA1lCC|x z?wweBFOB!&#vCyTvB59HofEvtIlBjLeoXoNZ0pMx`zex=GJy&@IJ!7t>1<+YXsrL^ zheO_CcEUh@;($YyPx-OoIZJ1W4Gsl}pze2DWXs{HTsK{S(N<(NyFj5&)PQA%plqrx z+A%TEQ-`j-`}=IwSo_jK`}V^(uE5R`%m$g`qSLs4JEtwStxI=khpY3=A6dN-a0F7e z^4nRXE9(WL9z8vGd#cGGt3M{z;>}8CT^RK6pQEf7CL0Y<=M$&rl3#<(&<8?px^iy= zl{aX~4j?gWD7|@y-e_7(OpJ!(WPR3PN}0T-W*}Q1I=zIf9>3QBBX5R@8XLEpF?WO? zQoSeQD-ul%g^IxMiA$|4^;FTiyiG4FMGNip{P%OmOW|A4KAhh9_GZe0G{iMp~yeguQwka#k;mR%?;U)yji`zm>?|$88PN(^we1~Y9UXQGWgJnjc5PtF zt!wg|C^nXsmWF$%4Nc}T*QeF{Dd;mS2a}8Sk&|R3CEtr`82}^LcidS$2D(^SNlD4I z6D?yoFmoV!rIeFn3Zkf(A5nr1-I^LKN=CT!Lhh-8IVh`W@c{P&;ctz33SLMGo>V?p(oqb2qxt`+NCsC+qtDHqdQ zZJMoyz6fKhNCTmeugt|#Y(Hva8sxkNS>%BkV^ zk^lT@9oZ3H9T)8!*HBm2HSD}}3x4Y#Qo$0At_jtzjcNg$CB%Asbjs6m+-||ktH!I>f^?vgU?uOKJUR-eS|A50ACUr7sm^{Uxv>LI?eygg&}Oa#BDp!RT*_N zJFmaKgwBzC=A{fy=>G2q(?=VO?z_J?+apfLk^OmvyNgtsbUmcU0*7z!N2{rgn%=yI zD|(Z#u&_OtWXkQlb0IeQ9=v)A^n@3X6EFq`%1y76{{JBz2MoPvS}%2@Zq{Xrjb zX2549*s4Z*Z0ZV|=Ih?*CHapIy{^5$*AUU+T7!gB^YY(+#^*5c*$o&N*{PmRGi?%t zRnjpXV&t`RFPC{jX$O(o2eT<6Owg<)^U+Sx1K_bskoW}Q8?0++h(HH?8Cjwhv;d4v z`_!+eDM9&E8vIG3+*s<(db!o4Zo>aw3%dB)IZ~?6+b7fe_M~gzLdY-n3*g1@OD^=k z|Kk6M`EXt_;`@&m;J-`qf5ejf-?$7hM)XV&S3)w3Ug--(C8cnjEHlDE%;bH`*m1gv zWQE@VrAv-Yw|Ia{6Gv|2+%*hi<`Y=`CI*{Q|L${^ac$l%oqC2i#!d6HV$D1X<6(5c0eQo?~ALM+a&c+PE1U98}@7^$3IOf2OI*$qP@< zl>06L%rsM`y(s@8w=hPQ{^e`5fT6@^x;!XUP|)*J83od)^92)G4|Z9cuY}L0Xc0wn z?h(t;^h*}Y!zJmTDvXJZwS@0=>y+%V$vUfnh9Q$GF<7(5^M)PyDG;GNViU5tF{y^1 z0#5Xy4Zs+Ml0aGuLMxO>v1@B(o$pv(X5EIFQt&DI3QOomI5yj^Y`;?zMxXT)sB~EX z2=jtirm2!_tbP5~48q5#KF zC%ratUWdyQh(G1*Ib?%!YZ`yM;Qwpfxj(L95)z)0__TkSPV=4%EH*P~2hSyF-qMoCAFhCr#0GZWpuA zrG`P@a?P?FI1O8vC6tt{Q}7^8p)@x%Pu@YXCvg#F07*y+u#&2q(Q0!MP^^D)_8hCD z6`|VM?9on)CMGYBMu=ZnfV-dk0Lo~hk%dX?Kn`m1a*HZ~r^VzN_k zWGV&Sl1Tr;YT!*YAZ#TFDjzuBGcOHqV*K^hyK{ltXnQBjcDT(5bn>gV0w_dc-$TGd zVb)=HxV2+)@x)Cc6*4gNzrn<6eQrY|~YfO zkWo!osNS+2LBD=l+E0iJ(SDI?itGm_g;ULIvO%4n^k3Al^mLJ>nc$_%uX#_rrH5Nw z82{N}4hm4bn~!d?&eLNem@ykFfz;bY%`~2GS+;?mxss@DcAbxlrefqq(#=|rAV^KG zF5$pC@3kEvd+4*TT=O|u$AhiAySvFaHr_2jKN*@)INBBp6=|O@ZOyKvS^C=BF5_du$iTVxKuOmZ$}vBg-^?e4g&qDO%0+KZLuId~Cffs&Lbhrj)8 zn)GHBWvlGIQDA;XvCVSr^g!gfT5{Lc|9U#tHL9MHqGA_#0p7uOY}T`*MaJMkFri6* zc=6h)c+S{*Tj5+ZIa!05Win;tkJI|~jPEdpRDp4%vJp`or#IXJ3@?&HJ!mX{B9tNFN-lfz&pr7(Y!zvl1m;O<)LGMCNpr$-5bV{G{z#H-|#ahsDUJZ$RZj9JV`z8qP)`S5lw}+Pi;N|J<&d zXUH_O*@NR40}IY9wmB)~*Lz=qcORS|g*pM*_M?%PEOOX?>QcdcH@~FJZ*jc8MyhJD zu^IkIMgdAIX)U|E-FXO^sfW-YKlb}SKbF$bYEj1L?h|GP`@pljsV_W~6tq*n?-ap& zxp@9s-(?rK=Bk?}OK$e4Yu)5Da<6C#M6jZd@p8^%^tnEQ9e?TSoiz9lJl`!hxd^$` z_T`Hp6mC+7_wHY={DPh&o}dURZktrQFh1<`5*zCri?!m_B}jh@~pppJ&VL|kxg_(!JI2}&Mutbm!aYuOc(4*{1$t! zNdEIU%p4xaXXnv{^Mm}Ro50XrGm*PXpIB=&gbJK?*yw>cIL0hi?bd9XT9Hz(tjkvRk%HNd#42yZ6oxhL~70oWAsMN8ho5d_-SFx3l3WDRx zNqB|sa9ROSl|i1WL`8eS)&$Stddo(yY^(~IYW_iTbol-H6?G5k2}>;AY<&1YpUaAk zoTfaq?wgFptCqUbJj1Nj%_nSeaPwQsYuU;Li8!V?}px zlb@fESeI!@PLuxL=1CFd1vy~|+}NarqV@cY0&7RC>GR)yMVi+x11_D{s0frXP$4)h zFNWy9y`4T@QB_1g_PQH$*k;&L*Z}u#QNyycj{pGig19TAFGT(u;5rXB!ug4SyV)CQ zB;NDYRVkRZLpn^#!SsQxHll`s1UG_+^L&eNP@v5hF63ki6<%yH>}GuHJP-X_8&$XT zwlAClu1cwuIm%7}7EgV&LpTOA>`dRFKX;f|yz%#mSf*@Ef62sTjsr>QVPM+$NFob^ zGXMln1U|t|uWk7snAgX#oAj({13xE<_?4#WLSQ6d%{^d^DwRG|9soUka9$KDq%J)i zMnzfsrfT7LA7?CI+MRUym45Pz2xS9ilFj|xFpi2aiCwxmKOmk9;uev?`)g^K^0(#r z$NAPCx=pUbaZPlFngZhL0$Dx!KD+J8-rKIZg;`adYP91@Jsto5Tz7V4@%Yja=@m)F z|NgP{ZNiSf4R2L9eg0VF3nzu}ZGaT7@P723Zp4#JQqk0rfjEVCZ0}W|bkG*u+;5et zFny6!4pRrZM6Pp> z2MX!}#^SNqtC~t*k|+<*$Qels<%g=UC+S}hB17o1I6zMdwNcyo7_M-kJm2!hkB<6I z1d@V{L4l#gG2N!U`|k>S(^k&qNHa8~_04F?6gPVm`bpEE!ihkrpy~#>&5wG-od3Ib7!7wJ5(*{15@3zn#D@)>RfbW&;?|e&9#NUK2o@js14~~ zjAPkK&{I5X)VhVRzfwUD`zx+279aPx4xxxL39OCtrNhOvo>yU(fyUiy40mm=2%!r#PBiI55O zqt~7*74%YZC7h7)QLmnG58_PkJ)}_+P}O9Oc!uz%<)r(e)k%C*t?SG`P(GY{%6;t~){Vn4luJCSDy+`mzLq>xsU2;c{r zi(90>d!0WH-b*0YqG#Rhz=@)A0PmjnYxza}&9aU1#{-AiC@}e5IiO{D<=1|yFWDcg zvF-gF%C(<6^NwfqC3dI>oFjFTLPgHzU7FNh0-{QqrI%f~VHq^!*dJ^ZT0iq^vHO25 zcHRI|^}ivy6WT%`r??>cIEDTWp6}WZ6e~r|9VuGBUD#NWyy7|fKDnm~ZOwA5WIyv1 z?h)w$USty39Ov}wqdvH*;0V0*6k2;mt}Afgxkee1E}8WmQfQ7X0M)im4P*1?;@$#D zso!Fm=@Yxbm~>0aR_}Vob{XI(p4!hq!;W{0uwDRU1U`V4Q;f(eP{3;lE<))e-D0dPk3Cdj?O{5j3%q zGAqM>P4z1%6lMHQqwe%DhAl-On&SWb0TU1{2eQA17A=Au@+3%DVDk$O`lLuEgDmI; z3X`QOs;Km$Jt&7m%q?Ku1$X*ehl&9p)IwWhDnpBQ@B#{{i5Lgzd#So01ZHp# zS0Lay6INuFz9Nj8WbU%MsIN^L4^YO?KXpa;n_WOYdhQYLhe#dAR2l%B`l8UofulKi zo{>^}jdsY^h3&FIptS|==IDViCWB59wrwxI^v`<zKPsN_qPyhmMe%S8104Yuva)G4OYhwa1DsfyZz`n zUdhLTpn-ne^H{X)Bh;s0Oh-6$EgU*8#f4w~fNtW5?+q5i-6b~gAnRhK(SmU8BBwZU zvq%kAA5Szu_8>EY162_mvf`s>$T_6ncO5=>YcToTrIgyk^x2J*2OzWf>dLH-pC|el zW-t(keh(tiUT?tcIpkE4yi0EHm;nXk1tr_Ldz|zoBzdjgb>jqdG-dPYC5}V->726L{6qr6F5?G*;f%K61vZmD$*!sRJ8d$P5rXUsr75q9lupEEFzZ?z821;X;nfLujH1LRAE~ zjUp(N+$M$3a2ztIEXQ!$IH%!hbSBk;jc`jlx`TL!z|T@})yw*BvVZ+>)74(mss<)1 zfx!g7co;2*0oYmf!#E8Sq-oQ{o->rLE&`)Hl12}N#|pT251D>Wdc=C%DSCj=f@09b z&ve0f(*(PGDit7jO7ggCK)CqIRr}V|i!_UvJ1vNhlYWF(l({KlV$bx8C3*CKH zNnRLc+5jscLoE+JzO?hA6&2oGXV*%F5qX0}Ro4%Z{}$PZ=%#6jtI9+o18vk^zQB1% zM-r2!xEis{@iCk?-NzR@j|*Nc`l~1zYu$rvQiXO$WgFLgB4YFq)Za~4B?-4Daz?chVoJht1|K6aM*(s#q(7^RueJxp* zdAf_->)Nu_tM4{4ZXkI(t_H`keG!#mNJC&J0_ReL7kFU_!<}m`gsa9`?ijN67d>-G zw!zf}e&9H!d9a5@Zx@0W)FC9&{Cw|mSvVl~yv5e{Bo%a0z{vv6g{*IGJYL+k(~uYE zf7lkNf3*;$*nBaHB(goty}$+s@J+&D>z!p+$bQrjUis+k-46!8LAl_l zq;)t{C4LARB-gHHqPY?hn~-lb9~J3Bwcs%z9JxJ$(rb37@vFxor9?(2*8;c-O8Fez zL*jwf6D5;f0-|2O0o^!Lb`J?uTpK9_{?z8HD)yQsGI2y)oT+QVOk#v=a1eG;-bfyC z1~^T|YKMcCo3vPP*89!8)i?Cc+0Lg*hPT&gaq{6e3;j>3F1nNLxZn8b zOF{H3db>?v#mB;B5?02jI`sVoJ=LTuCy58*+;00Yx(5OXITaLCW1kv-?#Lt_sVQ-(FCTAfpB^PDHv^=aBp6T$#0e(5XbX4TpEov|WDrSqE}X#1eU+|rUC$hmot;gKOI%g;8W+$6 z&`0sVh$g;&QVaW-@Af9J z_3~)d5*xKyaKneOYo&Vp*^xG|L!{CajM~GTUf%7k-L~eTn+}$WWPDr-=xY${D1y$f z0_)@Byfp2c=smFW@o$f-)?jtdj-2W{c$fKt`{vyFSmcM?iJFV{z*>hu&=Z>qpao43 z$Kim%-0{jLgo#S#VfEd_X@S5^s+-Rz!Vs1(05L1-Z<<26>gS4vI2Q?iJJQf@T-ZUJ zKdzRjS2-F^Gu^IP7CUQzDF`m>!2)ycItl^D60ZPvjFZ=(;N6TL%2vM(mlo(HuY+Qx zm0;VV(rep|qZgB8uc-Fe=`4IB<-w4baHUrtu%w{;F!G%S6t)Dt-GgK0RV=nG&#R6v z5q<6BicR9rR`0g^6I3c6@3$!owhO>mtAGBN6d;CZ8i?0eplMuirC1_M_rOjgO>My$ z3}1133lhsrK^J|#sm8Bs;}1Y~Mo!gqV|m3%b{`?-NZ2BYxx-GvrT>*=A;ZJyueTq( za$+MA&+pp>*@$*<)i{rA4fItdg1u-K^uVc9;O?6sm^ejjx`+{{V75N72E~bF6ABgk zs*8AINalk(UBrPyx!m~jh)6>YOKd3aTL}kC@45i4lu>}>zCu484{4DSFrL1Uu=q*; zLR>Yuh&xwVWl_L$*bVJzkOs|C6E>l{A31`Di$Ttl>BDPvqs1&E*(0^`2oJKJAyFn# zc0?|CVO_~}HDqIvG-+pFI zt7^gHoDdUak&RLft3&4(rwgi(Xw~gZ;q|hIJ~LK)d{Kbx)!?Q1-bdR6_7drVxIPHV zhhJK&C+K2O2I$XukioXZC4{X0C!+EKz|M~O0df>3WbryGo7iNX0xjbqDA5F5NU?MD zBxydW(mX5r$hb*^W!04pIvqxceEYaz@P^cKoDKb8${&YBnLx4GL!}N z!ENn^sTCq+NEd-fFznc%w#XNP1P@`Nh>SvvW(K%gugVKQaEg5ru{MT!srQRh|L}xG zz1OvopyS@T1lYPSi}#k`bNnM>7W`6aB{odwpmmmDtp+TN0DKR#A-k-r;mc(v$TLgsDagj++rc>vW#AU8m2n-Q#1N9W^p z#K79|aUSB1Ny0YGaT0@H@Oji01iU!9Lw}18EDkEv?JUQ!K!)*)J&^17KOnD3PEJI?pw?k3rphbucC)L8RrzTOz#?hbP&H z#D)bV5L+v(+qy`IE3?$vt+Rfzvk?*+8hDxCYV9T|jPlQIQLfKU}c=h;P#Gw@L!!kI(TITIjP zgGofEnXuat+Tp&tQr*S>+9{|x1{X@AQiQ~0iXg}c%x?{V3v*asvB3!;r-F$xwCBEF`S$#@}=61r+qcF;SCYIkrhMz*r=g)tD=2bAZZ!bye)SB6k zDlm%|SY66wn?W;_|2l5f4jHh*>sAh9A#*?erKEvlc0f2krrBS(ut&;aqpL;F47=Q) zeqWr$%%@goX#R1V#v4C!8Yc_7$szl(F1_6Yzc{?2y(TEg6+}BC_v9hP;`-il;Cb{- z0pLOjE+(>_C+x;c6#$A}*$sqofKhcG%z!i%|3}|fvH@|bVCMw-uJN<>sW&-)p+)Y# zyZ*p=Dyg}IrSio2#a;+lm3toS8cXXXW?npYQm{BSHg^56MI18GBrEMEu>LmyeSSQN zxq>LtUGl}a63^$}#v`1BcNh75<(_%23^^o-pgYAZ0{*$&UzDG8x~QqXI5fQvmsvSL zsCiw7OPDK4|K7Y$5RPewIlqSA*0r(!zFV9aZA!Qz1E?q_JSX>5>AJmw@CK?@6PK!}J+Q*=@wy~geU(h3ElG*J)&0tx~G0{5GHlk@Jl z@1FbXjrab%jEs?Ulq&XKYp&UR-#6zXFO?cJ<_6mA8lA;~W1vUx;3T)BXQd1&+r+tN zAu4nxKeY#qJE7ah=o0@1v9o{t{VU`u|Jc+qvfI{QiJ~p{Z2NBkNEb+B1qJOECJBL- z2|w)!H-)orBePC_jTKx4hRFrI{-T-kiQlzjk&kQwWa~?j=&bFh{Ru(lw)uMwQ__e! zykwD6)cg3l#mL!+U2^;>O|BRs)JxDOr#>{x{6ts<5|P}riAS2zpkSUN^)%fa2t9kN zRr82xoCEQX=3h|`odbWc5gOA}|8hty`XL!_0{~KLYULr=SiX4$Qelr3H$O4xN2y}o zI#S#GVU8Te3vcI$D$n%;1EJwXq;h;SEitzy;)HS;5_D_CTOQwbtkcY;HnDa?#V~TH z7!-k81ic>t(BclgOsS?58cg({O7McgIAJvBW$mCqatNd-fJ2iXdQ?+o7SOFPtIE^F zL#L@HAW1=@Knf~g8b2&uq=Afr0y}w{2=yi$Hg{s&;8^$sXf4tX^xt>k9%VzM>%sB5 zh8KISect#=Itde98|eNDYLa-O3X+Z55h%2ravb{|I=(0zjsxI0Fh8!ZkB~LWaS9q= z+b*!Uilr~GIpHgUNu*XDPwJk*VAG}sIL|ovd#B1JI zVAQ?p5u ztKd|kYomOsdFZlm1K42&!3+f7H^IW6z7SXZDq%5lo6Dad_B-#>z*GMM#*L@}h;13%hh@&awB z*Dz3iJ>~t~J7Zr<(|co|5SBMKLyWcUuqSxU)E!dcXn4y=(~ z!9d#PZI}R{$>3x!Z3un_deMr(-v+?UBGszWHt95k&h%n#T?W zetfV@QFO0fm7pNPtYBDahCL}Fu_fR(eZrH4hB>3B@&DziYV>+IZ@X0J7F5#&2KJ$> zOnK*9^ex~vM)2l!6pZRq)VG4Xq?u=vcRNY{dh)T;1?ANW(iW$x_p$%O3LTAFn%JQu ze-t75GcnghR>MB7&Lw;`1(iM93cpkARV+KxL@uk_ej_!M0qcaQhR|Q zTtFS84a>5!Cd zV%3e5F{AcmJ2>6~fuQ40U=9TWOq4H1+Gw(H_zD-YFJ4jLnAg+wPm*hHH>OfA1nra3 z4h5Q5ivmEdd9-P$F!S%@8>sig6G`hQl5sA1L17Y$q@=w4@b|mliK9az{~PeENx1^( zlNS_?2hcdk<0CO^MMSho(S2ID3)!127Wea|>YI=tSoHK$8x!EUEvSQR5=NE6=%hwk zd-48gSK^V6k3vZgPyC?n+9-TVI}&wY^q!e-&O-G^El-wz_BQqzy%RT{kiy^3wmY|b~JP~c^ch}&k^3lMZgQBqy*u{brQbba92`tbd z&hggaMcBn=sAN~PIOmFHai z(t~NEik?~2%7#kU=75wt4Par(k9N3Gn>X)?xNnd0-UVp*`K1ft3-9MGpJZ}G--5*Y zF6RZ}3$?t>vCM#`ARcc@@{&`H_PUr;?MO1G?7{)zOAs@JYwf$A39QkP{B{&2KmMRJ zD=uu8y$nq0ap;Z}tUNHSA+Y?7+9n37+zxe2_(fC~TOYYZh=91rSJNe@bjfY>a-gar~l-n~x zmT!Ep*+?8Ly<)t?>*1Ao2QKd0J@5P@iR;@|&a;^I_-}S0GE?@+{cXmoZ>DBk>G5f9 zUbd)*%Z(O#wtwgjyq0KOaH~SCcerWjs36h z55IlU2EoC)yNRJf{4DHBA@}oHre;2rC~^+rCZu+v{rO}BNNW#wc5X9ETR+YW%Q3u7 zi#Scz5mXIg(6Foz;&#vVC#$#~&O=5wXhvLYUj>8d|ZhA$8qa zv7Hh%X+`TxT-rpf(dQM?*7IwmB^Z+a^d)GWs40)jLnEMfmF|LgjM8aIjLn+CIP>~W z%GU|Q6m*kkBckBs{H-2eGQzVo6H#VnEB$he4!pos1ExV#(|mmYD)#%WYEi{U0juNk z&}@2%j^G-B-ZZmmBHlgXjPIE;Su$K&iu=TC?Qwf_a2ewHU$D>bKEJH2Y#4;dWk_qd z#Yo~QEtbOft4x0XBUFP|P{OG5HsF8~OgYVLs>p!(a-BRpd!1LAa-DJ3#2?GXzoE5H zTi&sR{qoEtube1KMdi&>cJP;=OP%z5_f598f6Z>B4USR3WwapjZ-A~Gu8u4;KBQZD zoLw1b)zV0%3MBt~pKIZeZ8+3&Pf|XlTzB#}Q#MRJV32KSy*o)90iG-RY%mFQgO997 zz6i)hfyXyWxL7pyqG+d0^yXQK#m%e9CK?_QnECkeW9E%BKt;~Cp68>*Zg}EB96-Xn z_*??N;V4Ypw^s6IGTU`K7rg+9)v{n#K|~naDOVn!{EI{~qT^`959d-_CvFFrl`=zJ ziCsk#8cl855HnJ9s~UioWYWpx*9iNXt6$w z=dAqnT+ZZQkjAzZ5Rna^IvB4tofhh_-)BN^Ef|#3dOy>x_jl9Dg)Wm;+8c$7WaN05 z`WHd;1QyHY6uVd|kPsT-)|<;*HKa7;hC}2-Qns?+cQrFJ(*cq`K{t}0oO`aZZu1z+ ztXxG76P0y1m$KSHxKaAl_miKZ9uL7F3LSI1m6ZdNroNwwak%~{rq`Af{`3)cL+?pn z1txyw>eX;A>7wCLEno_w$xkkC^HPIi!_0Bie%Yw`$y%K{E@MGf;OSoK_4c9pCgN-nMu48o|5)(U1NPtTn^GgpN6L+O|vfp=<6g@3=B~ zAhO6I5cb4pWT=SmR(KF)c9-i>33YqQdl>Y7cX;yWBVb0FkWcXVBO@bAj@0j?m5pRq zwrm4{pLMo{nmj7nA<0c`ITPD8d9*F4gGP@oq*|BuCA)j~R=K&mKSk{1b6uRBaS)dn z(IQO#ZU@BC4$IPwA9@<|@I7u{u`Z$*ZR~}5qa<=o`|@jcchY`AS=Wa@lH+xDbtzPE zmIkv=w6hQLYX@*cJpPoqo4-arj(2B&dD_g9Hm|CcQa#`@k~6>1c$+3EFVV!hT<1qu z3~no0-%Y!xccu<*nTSiCk(pT}MC3R+=hzjLAH&4(N3UkBG0(Ahg%6O2m;W#)p+ zc5GWCPWj>9?4|53*-b&wXz6nFk)sT?vmgSml=*UgQ&zVdgMw-m5;w)lOIdOqHZfa>2w?mOKS zE51d{d^JnLl6}ScpOtMh#U|oEJYYgdGQzoY=FC~Ne*d=n^Ch}gFsG&d=ahx-sBg-m z(Em8UHTKk8J-8M48}nt%w%q#x>0)MeATrO{$qrpaTgZ2V21~xD7d(eGEcb`IG`cV5 z=GLy)GdmWNjdO%5%8-8pahLR?cbCU1GhQ_KoHGm6&e^z0ZZ=gqJ9$;%F*0n z6uvQ-E49FSUZo8C%U79uFKsF>-0^@~>0Oo~WKJ`cbpMsHTf484bLRFUKHA!?>`sc0 zcN7jBot9sj&hEX+{VN9Vj=b?b29WPE(>O!Sr#o_YoJIGa|N4gEsh&=AkM%OA<>wWl ztJA);+rt{J%6;O4zW30u6JVBdwIv+ zzoNMWJU*`L^ankmaCiCt`k6faO$4kSWrU&uxK7@e%;cucO(j)8#z;(_>FstHLar{KJixO!@+%<$W?Ps!|5 z!{m{dq|iILmc?(V7e~IEg8SZ->B>9U4VxzRuWA6$JuBIvpD75KYzuypX;4X-;XIq& zJZba(nN60nj(LJ{(HljO*nvVFmgy`%`S#0G{XV&aC@ceMl3kw_fhB+>eiZWq*v7KH4U32dgB#f3@)L-lH-wx7Lcqg?Z{`$e(Nj-A|EIi1SR+ckM9BWh+rB% z7xXNNHDTS(O6<{7k->zB_%U2x``bM|Q?OiJ)%EqLg`konO#n8pfz8Te&rv@kRbd#H z{>;GX(V6+DYf0)u`UW1^R)znYji+lRGqoq5L7*4!|L)YX9sEtIs!rf!PS^GUsQi`* zkq9NP7j1C(ai56O%~m6Z(`c4P{1zlN-?;+PUO;4)9FH&~(j}c&4;F4`4u#<~x9RyL zS^P$sI+)i)mh$*C5QS*YZ+tmcVT@76&Z5W1W)ljI@qbw+soK~~P2hq&K1jR+7@K7^ zFru@bH8&G@Ag_B8pPhZUzMR8Z?7`c%)tJk}?9{Bz4Resbg5q>^JOMZtgkrO4o|fSf z_9X>`8Vd%2Q)e~!Vw_w$+rLc22tdRp*ShIv*i&_zuRNs*%{B)GZU|8fq2W0C>&9{# zFq2&(I{AtDfaQ2dp@;RCdk{YxPH72)DTv>S@Dc&Yt1{!WD zp*QI|0rC{?(2CEEH(En%Pm);G@C5tw-7Tma5;AckiM1#&6*ww#xK%vTHju|~4VWTg zZLMSY~%e>Al)^7z0%9l;-a zq(f>ipYt26PFx;8dDy+O&y4+%i%rlDLNgD2Ym-s#hKx87Rf4!Z8QSBs2!d-Yzum zNy<-GS2tivL3dqDL@OUE94aUY&*uRxO{IP)!J%w`?!#!~h(Zpe5pR@m+vucB%S=MI zf`#s&@);!&9Jf62_qb((&jpTCM!+NO0VWnrGn-i>$(ox4>Pc#xXooSE16aR<(*hDj z+t)Yn51D!&*33dM&@{W_%V3$dn7`9&aa=DlJI;yyfpt><{}28juSe|y8w9CrpFgW> zYK7yfIT)uLk|d#W(^~PCwc-Ms>1i_waAXiN0MWBjrq?g@URM)}>4~P}6Czx#(c-F} zZS*{7k%T%VoEl0@gChySpQDmOa==ml2gAk@ILReNxHrU7($@1hK643pDMVc&$1D8! zNa+nx=J)6O?u~F}EamoY#)LS*GHlf2{11O%vNp1`?fO}c-2)oBwazAY6K?~A1~@`e z%x@y;1T#?XIUbCZaL7f1%lAPtL{&830mDoxFa*oceQWfkzxKo8I)fVj4~A+SP&JL? zbATKtlsFe@;0j41F(LE{1~9*QbylhAMn6dQD@3T1Ml1sqETg7}qb8jDqF($FjRhR_ z5XM%>i-1Bu^#0|QZlrw+CE4uCdlM-{Zg+q++V zlyYF|L*)gGjZ8$RD6~uJY&T5mzpb_~I5ak?EY2oiT(oV9js|hbp~&<7yWMO{AqzZD zj`ga44@NMr4Sz-#rjkIBn#c)fbRgL=JjB(Bk6{zHM{+ooyK7!)y3U;*8`q)Ne|y;M z-t5YQz`~3C+vIIYGqwQ&AJk_j!YiWD0tx7j1iqvnZ_GPt0K5gm6q$)%6x$J5FG5LS zGGNW=SnYaPR{66jmAif7kJo&XZ<*F{2*?wWq1aS z3bRzxMJi!eg3J&Oe7;4AT4p;jQpYWZ$z?a$Q?~BVI5Ux!!^yCQdGd|7MfeqYSArMB zo;r^04o|)o)!8YYi1`yrelCw14ilqibE3EIW>&8}R{Umnf3K_L-r%_hF#nRrC!u() zJ*ALb7kBp^m?vF=g;^w}9CqcDM91f@90Oe@?{>!rx? znXX$F_r{K+e#-Ld_OhZRct7<^zzMmnrdzSYh z8FLeEtcmcJh~2o_m07|?H4U1s<2FL?z_r0LDnQ7&#ZVSl+AO1qn>;?aE0ce>d~oqQ zQR0QuBo5o&^$Y~v3fYXX;orOU1 zh;|}G*aCCU_^pVRV63hAU>r!D4yA2k=l}LQSH*j|%{Jy#u0qmCf=tIe000@Q453|L ztBqYRx)$c&x}qy;0+o4N5{{bRXcM$ZeHa#$-IuhwR`b6eu+IKeSV1(>G}Ay6hwtfp zQBQqWV>ZfrszB%v6N>|zkduo#^c@vN=L*N}kUdqo9;fv-%uLxFcBGAt0y!(gIQ66c zOm|4-FaM|GT5R|wfJ>P6dktuBJEVI`xlzI%K)osc}tb9*+GMfMG@|IB*rj zQw$VFuCvH!D%ykQWZ{U26RconODvd;36dltDc=Or76O!2Miv7WjmP)OdHE*LZ72S| zt-olfed$)SBJIpDY)whbU}w;t!08+*N@NXVU8k=6U7Ew`$ff)@`XL-aZ*k>BElNG0 zBG*gJ&yIdCq|(N+`8m4(+S|1!j-Q+_!RfM}vzHSbQd3(SZ>^X;(hePO(h?2u#-?F; zfInt^l3Kb4XCQZKhtH=1$>30`R4cbajVR2<%Dz*axP`2U2T@#VU&JV-anWO#CbYwG z0>}uT5>3l-Oixk6Y-@dR?R`~Yft!jDT^s)_o2F!kR}pMn?d!U38A;g#Jj5rKd0B%k6G-ox3|YUY(Aw=gkDi65R6#u zR+N_>Dbe-H7LV>AI+Nsa{t#H~B6JrroK){7FeHoso=_p^Gl*TG9fPvZK0?X1@Vpy@{Kq_~;kZs7dFFz;)u3 zNZD>?eOQ+&RbC9_ZtX!{^o%(GL_0%TS`GuW8d14lb zu6tGUI>K&ykPIZ%=^4pF&UKt>5*5PAlB_sC?1l${9%#9NQ4_(lx1XON}xG%5s4wzj%Y>`wP&UnXZ^^jp*YLt z3kpl~T+cl@QDlbOQ+1L820$|--%!eKFvEzCVE_%D1D_C>BjB-05v=|3uBe*LJ+*6d zx3_uaXhL!5#qWcdx~R%tB3St3E7-S9CBbN>9AWLH$VW!Z2mJf|>eBtQw2lpJ$LR zLa6o2C@NWXU_=#`Tj)cjARJXjT8}CB%;T`Qcy26>VI#E)^{UL5H@#u4Xcj0MYa9qv zy4w;$R!FcIgrVUmwKY(9qsS|#`T~t2t9aB<2}?}%J_kg-6*ZMUpQ>1D(jjRoU;;3M zH`3GBlbM8SkALJAQ;-MA=ou|x{y@C&>`o$KsTLA+lkkMlR++_%_b5g_d-*a7<);(I zMyPcq-%+Wrqi{g-6)F}fkwQ8On3Ni~d3|3CWwFf)qLjP}vH`g8Eg1 z?(8}bh4k+vozdeHnnDQW9at3}(72};TX(@0O+GYJRBYuWDqP_Ki~04*2^(NUG;^5x z@-S0=9aew>6eR_30j%@*B)^4Z!>0WuY^og|&mhBZtQ#%vYmod|B!!WRg`7=4!luab zds=$KDYN|j7IK2;I9IHcfzoA9AcCa$1=ExvPCbZw%;LJX9E2wp*i$p5?;xq(+qi8| z{71rfSGzXiDcZt!sV79;$j8_Oj}P8i2lOMI>t*bBD|&t!_wVtJNqSoZ;j?O=i|VJ# z{EO7WWEvJQR5ucSk+==`RdgFzj#&dQPIycFyh>h^SB_uvg{)8;<;QW>dq-t~+sAzZ zMlZ~!=3kT0(N43X-12eYvSx!7C;$rr0JuelfuJop++}tBH<=;loygCFLyOtN0cd{= zMPE?Cs1|?*vj&v6KBN>Jkj;i1?*On}9v{PQQ$(roZB#7tTTzVefd8`+#v|bgR33*S zqoGsjgljJj3c3|a2#e$n|8u(*U&;TE*;gB%T%ib^kNaJ0dkayAP{d=AV?4ZYu%D1$ z!L3U{KqQhyo^Wyo0RaL0R!kXq`Bo-rV5v(N`N^g(P-=nXO$T}wW2>2|a!Iz=vu>i2 zc>@m0fDgzqrWkEe*-csy3VC68v}f8+SxqZAu_{CZF@6x}dl-*yVdIfDUpz z8j6rS&`+Jn8Ce~eMSYoCBhcfVib-CCym3r#0_5upN%A&ws=lHn+&#tbnen7(@ka6(~vCB+z zGLD1;($mGlbof7PtLX9T0u?DKxugdj+~atIx{+ao$FE1KJ5qM3+kO(*o0610%-zq>+gHn~xYr+unb=EXag#=oKbgDq!W}`TU1ldrPd_XSKrT`StUQX;Eu!0&tU(+5YtMovPQW zP9BYA1=~3~9)AO>W9SPHo{$Maf>=^MVUAm99jRGFT_V~#y;0n#^SmV;AtVQQe>fKk z;g$ni1eR+Beep^G2^j$dpoxkYD`^zVgefGD!oR*H9T#@FQl{{zuA%;~6%9^w183W= zHI&08NHATzCoDM$z+KGPTc>cET3&{W+G`Hh(YOV~wQw$VG0FygdT$yw_E@!Wx5bLV z-ih(vx&~Ds%`~TBalRO3j(j&IJHh}YFPXk?6P0Cat5M%ALFYWr^s{_Yu=$0_N0ttw zGrH)0VS`^7C%A!RE3r1#0N@3KXnE*EZ^jm2%_Wvf0#_H8VyNmiz`fDL^u+-o85>6< zgN7sZMj--wN@yFx)-5=m;!Yh;T)juy=JT@^@7f9N4tfU+vgD3+m}R;8(;-qZ+yrg^2bSNkD}+vRpT5?g9RGIx}bKoTgih5|_K z-MuN^7(;Pzu=K9j2Q) z?x@&+Bt?Uo^!OyDp!xiDYT2xhQGOC9LnSt3Kv>zYj}ViAekw;zdKOAiN%{(EIlXGU z!3%rHK=3+wr0J(dDDl}i=SGx(kxZKb!~sKYz__(bbQjoU-nR#+NJA6#>hIp9=2V)$ zfkp$B`X@A_{MI$U#M%XX6ZW)p9W#DGjkpj>i*hHvUp8V$@q12akmz5TkFyyrH!30d zJ<|t6zm}JtcTF#D;UOTxzjIGa38)exrGON) zeiid^8{G{+?gI|q{h2St3eZE2z)Oe4*@XgRn5I3Y$RcBiGNaiqT2R?%`Heei^R0^q*7g_*Yf#u+i5pNoc<&hRaN#BMe9!tYqlTuTehGX{*p zC+@wZ>7I;y9!%T@uv~~Q0jC0t`(-6KOL0wfM02+!CwT6DFbl&E<+Lt145W&14GSa| zK8jdFSIBuU(eNHpNuN>tjPud<8XdD|7T$B}kbFwn%&oZr!#5t61uhR{D7}Nv2_7IP zgnw#QGx#q@=Um2gM%buxt`9dP;vb!JtdQV8{mB2_Ke#>dg{r|#3EQ?zHF_SN?yhYq I-`R!y2R)`9(EtDd literal 15627 zcmeHOdsq`^)=yhq6vTA3%B>O=0SS~~y;L-idO;9G6tPH)iDD65Tq9_8LT_VcT4_j&f)%^!It=biU@ z&hPhL&Y5@bvZdk9j_!^)9L{-h#KKq{Za9F$y|Z@2J6Ot}?t5**9{6QT;=-`!^y$;1 zqodo~+ffwV<^MwJ0pQZNHe zk$H@`GTu^fIQ&+0BmTSZTm>^qdJa0bqUP0+e(IAZu>CFU8B+*-AkKDZfOBGNGB*NI ze(EbTq~dJt8h2%EBqPX~?k6T6<4Lk(J(tKoi}b^P^S@+Y1(`*=46`R|$9%c4WXh$) zn7{KH5*9sir;fL$*tKCtf}I6+7T8%}XMvpsb{5!KU}u4y1^)poXsR@--TeKTQ(hX& zc`SYSKgJ`jE#A1jv5V6$<*#urDm(C_cNQXkdRZJ#ef!WT@E$8#j9(pbynfPW!zws%5p0wHcnr1v#- z##b|QYd`!ddyI02s%P$S_O$XCPo!8^W7Ftr{klrydIs%>P9{Dv+-Bg#HKMV~MkmAi z?I`D|LxpXdah6lCEb)dYw5a!{1IN+!WvrzbwwSUU2_>Cn4rd=gGhGUc>X}2%oy)Re zOV2O}qE95U$$JpO(@Ao@ZlqI0DC;Epr#1hj#)Hgj$elP|C~<~7+LVry%zpzQ9Jo_ln9g#l=QWxe7*dQ8Z%G0vFtU9oZm<`{cb77KFQ zgS6x)9zsy=(a^Ug*x2IxYC0$Q4kGF!6~u(zHXEzzL?KMd^rXrlG&GUbLN8P!tIuwW z^y9DblX64wovIjW7rXrd3q45(955$=n!R0i8qzy}6b}3eAmYr}DOkn7&dH*y}ILNMCu{{E(A^sKs6fs9!Si`b#6HL{?QNf7|@nOBqi!jT4NnG$wf4 zqU`N-3@V{*KE#kYQc>Te(>8>-c}M_<8c*|zz&3$QW~drnP`^NKyi_{!dmAXa-O&vUfaO93g<`o7KXIYhfrpgQcH1^KUwenUv&oB(?kt|N3t@&_2xlu(KwBr zIx2u91PUiC&QSH`L17BloL26_=zPhvd#*(SAgWW5%wECtP>yL|u`UW?x22Bew563J z#{w6{h><Qr!HAW(*S*i;Vtl?1=3Z`H`AGTFD%uFRL_GhMyt$YckKb z_t;=*Y7JqWQrNBMWDs&zm)bI8h36P0Xfd?+G=3qPOk@X0OIy41bfJOeRNGz@n}Q+s zbP@d&vXjthscX04#9=`PZ1F6pcc?UO4xug9tSft|2P2{)0ZQ++a}uQ@iuMqG{4~Rs zcl+KOtHiU?pH?8#>y4Gdk6EF%1V1{t(zu<$Jt?A5=X*phw1H}I>>>bk=@ujUR6Y4j zCn~j7OoHP`&0lFb6Nl-7ULI4Bco%}rnZx74Y$}Xcy%Z3!P8mA#CUEYb_Oh87PX$hu zma;ZqKVANbt$uzC%R;hrfq(8;YYVaxCnpV>FWIqfE&t_Ks6fSG8+OJv!sid!Q=Ts* zc5Mt@R_)rbYvb7l%Z>y)66{E@v%t;*I}7YA_#IhLbyckplIZ9k2!b8c=0wRAi~5fR z<2-~s;kAg>a>e|g7fBjHW4LZ%w@{_urqs;bmjU7w#i3ndTuX7=q@ zNMPAYnuh@C3XYOH4LCu@xd?eKHP%*sn@E7zXQZ=;bIb3f;BfEf^nVMWY2KQ{b2&J0 zP7VxoCS@PK%oAq4x#$x}bEi)A`ofvyG~k?ClITt){&G=9h6Jjn0_|IyYPH~Ce|)h1 z2(0~iMMN9#fgaS!zr2+<>eA~o%XT9gp37CMWyWG~7-FbbMDEwd9df53=H446k4m;$ z|II{2BWzh9cUt$W%X1-ig*1>z`0h~vnhHPzhts5{dN|gtBo#=oi(|w|Ob0BBuM+Wa%a*$>zpwjn?i(k@%fDYe`%vmZj5Av#D(s< zmLMB~=t?o9{`VI#RI0TZ#jFuW%UDC^)huk0wTpv$j4Z_{&3<*yGX1+gDf%JHl@qpl zVG*Xm&CDB$AIXhs)kow(;1umbN_`{P>~SkM?g$e;TT3AA3P6;ND1=-^`$_GjCAC^X z+OaZN*0;8Z-&<$~?;+x8A@O~wc2!kmqO zLvv6(#MWmf+tjfgK-ukSkXfhwI#=!A9OMhJdsaHyWXZsq!E1_!brm7{!;}qc4^@l8 zM2z>>Pm-!sCWD!xG8I3ONT^uILgP=il;{`>OIBT!x-01q%p~bOb)KgC9d2#SL_9@7 zKR_N`hPlc)EQG5fU&Bg;YM07R68Dzlv~W?qFSd?AS}H(kU7!>x+k^xHjcc0c$PZO$ zmS6>kMqNIU8~w?K65a8Ewp^7Y%4K6u2PkYXs&-3g`IhS-w>OnrTU)KfhbrjN))$1XH8fWrEzB-}E4lM&91Fc&;TxoC3%C5G?s!YXo#0!s z0Gi=>9~4QOj8En=v{%or@Z;HKJ^&UDXz8b-`qlw0Z3RFPOf@F(`P(!LTz~p|qzH?vTa})gBoGqE4R4rI} zclAy0fiTLdi!-E&)^BfL5wBzP6uQ2k^i@@j@N)|IeTLA#|LN=v_~^ioGo=9o z;l6WQHUgLa{Y8eSQdoYk?0%4S^ZMUz%in+!xjKR2K^FZMOY;1T#=DdzJK zfh`2PFbJNrLh!#p)N4;kZ8rNpfs{()r#qNA9@a_i7&gbn*~RDAD~??>xhP4_r2W<- zkxyWZ+*m_TTdv2=O3~8>{Z6m=FJAB6D22?eC*H^(^7{I?C+@*b))V9_o(XM;v=m+8 z9HMZx^6d>>POtaxy#DR4a#y|N%tC!6P}LAO!av23c*rU1^(%F;le6%S}S=|~i=tbFtO_e@Y(``yrwr{RWAL2}St#@_EH*wETEcTRv zF9hw{uzo4`;xuc=#_x%Z3I5Xla7d$9wx})PkgEubX{M!qD!V!7wl-JLg^$p>ka z5$GIeO-Qp|YyKgHo?YE_n5mnMZxP~?w&d}UTteL;I>mr=*t}}$f0dJh7U@+47Ppsd zzNB>x1^Aw=%cnY|U0f<>Vy>hDDC;JiQ`c1OzaoZ{6ndvd4xpYPwF~rBnX3|upo$_W z`6gVs0Z^pefX{CL*3OWsr*8lfW=MnFm0GtM(g1g=t213c3e%1==>Si{`UAEa^IR6m zKMT#|3B%-0*gZ?p6;=Vu*m=i8y_5A2tK3U}D|5DOG|9%FSj_w{7LrN8+)=iu#6 z1{u}(TJ*s^W~t?`;Phr6Qna@Bsdp|_M$YF+AhWzpt(XuaFU*QZZ-r~`PRSjq$m~yY z)~sut>YAZXuh#IYy%U}3ZaS^{AP+gW^}SIQtV_+ULU$Sa>kcEIYclX-Dl2q-LPAfo zv521sB4;S-My17)-&zXJS}Zp@G$@2~-Ie@JCBy2me^f_}$s!iz@wpw#N$A~s{0#K4 z0io2HZ$=A(um;2B#^?X46{MLalQ|Ww(9Xq~_&fK~@77Mma3kRcWiCFK}ZcK;?Ud%CizzZL>s)2(T3 zR*l96R9#3z1wyX5LZf1uQnv~z2cKn?ofs31)zr5E0IO2hBkQ0%KHHK~OVSwfBs9l% z361?j0c2kHOyzZR{kX#_SY1^iiJjdp$EmWuCeJ`bzkdmd#!ABrji#_~ge2L;-3qp4 za)ZK!Mm6j@;Z$2DRP`s%u2wL5wfA(|Uvc?&5G}8{E38L}Y3%(MEQuSjE6ljXVM`aD JoxkR*{{g(FmSO+^ diff --git a/docs/articles/opt-dead-code_files/header-attrs-2.5/header-attrs.js b/docs/articles/opt-dead-code_files/header-attrs-2.5/header-attrs.js new file mode 100644 index 0000000..dd57d92 --- /dev/null +++ b/docs/articles/opt-dead-code_files/header-attrs-2.5/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/opt-dead-expr.html b/docs/articles/opt-dead-expr.html index de4175a..b7d5d54 100644 --- a/docs/articles/opt-dead-expr.html +++ b/docs/articles/opt-dead-expr.html @@ -31,7 +31,7 @@ rco - 1.0.1 + 1.0.2
    @@ -56,6 +56,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -74,6 +77,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -86,6 +92,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -112,7 +121,8 @@ -
    + +

    Dead Expression Elimination

    @@ -128,18 +138,24 @@

    Dead Expression Elimination

    Dead Expression Elimination

    +
    +

    +Idea

    +

    The basic idea behind the dead expression elimination optimizer is to remove all those expressions whose result is neither used in the program, nor the removal of which affects the result of the code. Therefore this dead expression elimination frees up memory, saves computation time, and provides a speed boost.

    +

    Background

    Dead Expression Elimination intends to remove those expressions that do not modify the environment.

    In R, when evaluating an expression which is not assigned to any variable, its result will be seen in the console. However, when the expression is within a function definition, the value will not be printed, the environment will not be modified, and anyway the computational cost will be incurred.

    For example, in the code:

    -
    foo <- function() {
-  val <- rnorm(1)
-  e <- exp(1)
-  val
-  res <- e ^ val
-}
    +
    +foo <- function() {
+  val <- rnorm(1)
+  e <- exp(1)
+  val
+  res <- e ^ val
+}

    The unassigned val expression is a dead expression, and could be removed.

    This optimization strategy is closely related to dead store elimination, since when a dead store is deleted, it is possible that a dead expression is generated.

    @@ -147,7 +163,8 @@

    Example

    Consider the following example:

    -
    code <- paste(
+
    +code <- paste(
   "foo <- function(n) {",
   "  i <- 0",
   "  res <- 0",
@@ -159,9 +176,9 @@ 
    
   "  res",
   "}",
   "foo(10000)",
-  sep = "\n"
-)
-cat(code)
    
+  sep = "\n"
+)
+cat(code)
    ## foo <- function(n) {
 ##   i <- 0
 ##   res <- 0
@@ -174,8 +191,9 @@ 
    
 ## }
 ## foo(10000)

    Then, the automatically optimized code would be:

    -
    opt_code <- opt_dead_expr(list(code))
-cat(opt_code$codes[[1]])
    +
    +opt_code <- opt_dead_expr(list(code))
+cat(opt_code$codes[[1]])
    ## foo <- function(n) {
 ##   i <- 0
 ##   res <- 0
@@ -187,16 +205,18 @@ 
    
 ## }
 ## foo(10000)

    And if we measure the execution time of each one, and the speed-up:

    -
    bmark_res <- microbenchmark({
-  eval(parse(text = code))
-}, {
-  eval(parse(text = opt_code))
-})
-autoplot(bmark_res)
    +
    +bmark_res <- microbenchmark({
+  eval(parse(text = code))
+}, {
+  eval(parse(text = opt_code))
+})
+autoplot(bmark_res)

    -
    speed_up(bmark_res)
    -
    ##            Min.  1st Qu.   Median     Mean  3rd Qu.     Max.
-## Expr_2 11.67878 11.61415 11.42088 14.09386 10.84379 171.2217
    +
    +speed_up(bmark_res)
    +
    ##            Min. 1st Qu.   Median     Mean  3rd Qu.     Max.
+## Expr_2 14.72117 14.5723 14.25465 14.64141 13.40819 33.05681

    @@ -215,17 +235,20 @@

  • Check if dead expressions have never-assigned vars?

    The optimizer is currently eliminating any dead expression, however, the code:

    -
    foo <- function() {
-  x
-  return(8818)
-}
    +
    +foo <- function() {
+  x
+  return(8818)
+}

    is not equivalent to

    -
    foo <- function() {
-  return(8818)
-}
    +
    +foo <- function() {
+  return(8818)
+}

    Both functions will return the same value, however, the first one will give an error similar to:

    -
    foo()
-## Error in foo() : object 'x' not found
    +
    +foo()
+## Error in foo() : object 'x' not found

    • @@ -247,7 +270,7 @@

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/opt-dead-expr_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/opt-dead-expr_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/opt-dead-expr_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/opt-dead-expr_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/opt-dead-expr_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/opt-dead-expr_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/opt-dead-expr_files/figure-html/unnamed-chunk-5-1.png b/docs/articles/opt-dead-expr_files/figure-html/unnamed-chunk-5-1.png index 8007b97b7e6d672609bb6c5a01a338a5e4c8ec06..b67eb1b8f6ecdcd021422e8b9efc28b756d1aac4 100644 GIT binary patch literal 34969 zcmeFaXIPb2w>7-gpi!beiG^5@ByI~J5J5#iz!J*_1r!jFW}%2QLlKaw#uyvwrip-H z6cD9X=~$7Dpddv+rAif~m;H`;Z$0nx{r|q}`hJ{qUC$wbyR3Vyx#k>m%rRFkYN+j= zJ7?(}7K=5Pv*(WkES7*Xi#6lO_cQSu!|9bq_%ZXS>h3@A6aP25BhwQt>XTN`?A1}8{D1l>CkWR zq z_56Xp>GTg;YyZFa#r<1$v%>m|Lln~Lf0Vy>>(;H;uO|jo6QiO;B_t$F)xLAkI=}tb z5P3(HW5gaS)8ykz zLlo9W(2xvdYz7S5-y04)=64WV@cQw4}%7;c*q&uESb*q{5u? zGurr>lh*M%N;CZ7w{j_n@CbL3cQJ2n`w2E~5wktoc|81Fd^8WN=(_m)txA*rw zSyTS{wQu7@t8DsUm41>*0oTvlI=#hTuK(pq8N23}(&6OL4!a+_I8%3r;No>JAN~2! z?wwWO1*%fnzLjbU64geo*5p*K{|DE44PIT66+iP+XQzFwMPy5fY~#AC-HAn# zH6G*Ln#RVVjl8L0m+|g2ANGPZYQ>qoIYEDZd=dVjuuOw2U)O12R^NK~FS{x|bHBoO zw-sGmCkD#O+F#vOZCmczl@R#*k?pa$Q)BFcbMuvMW@KczwJ9`m?r5x4wW0gvl_~|_ zA2V?6PI2i758$#DPk(t?<~cQhN3o+GJ}ly_3Go$~>bnrWeYnB3x5Uzh9s;5iylF{Q0zs4aGZ=I^kzX5W(& z8#Zoi!RLv0i+GH@AAKT})0a>!>`=a~#*;Vc!;XoGk>gDaRF;>=m)ep4kDoe#nRp;{QX10`uycdyQ&w>IUjtg)?$g0n={v>G+1u*xs*r3f8BIJH_?RSG1+Mx zFr=MxX1C4HZAl`xxD}{@)`^}E2Q*H*s1=;&dZBe zaZWY3v~4_e>;1W1SAM>+^4gZfWY5u-fXMW8ml)IUa6L;v%5Ce89Xom+ee|xFn!P5t zFRIt^*^j$TUS9u|!?)4>->I3PkP(!b$W?xjfTmNl3uf6xyhNv&YE4?>=kZH7y3^`SJbNw;ygrjxi zGTW@rGL1B7)=vUb(MP8yMmbzW7W3wpH!NFjuL*Ut>o{XMTyL*#R&p!lU=M%2_B|GS zxL-#nt*U1u#<<+M=?sVK@9$4*Z`s{I4~x!Sm{q>ZTdwt{LwD^)PJjE|Ypri|j3w-U zqrZxUoE56w6RPBH_wv^ZzC&@DJsJJyXN+UA*c8v z-FDw(j?wWQhMA{Ie%!r$e{Jqit)=F=#YH#2$KSkEBAU)`8cK3`Up6(~8_LFal+sq@4>#Efi?Z7r zcQ|xbNpSi{K3;1JG`}tS^;Q|&y3&1os3a|Svb$nxezM8giR&>|Rmu?Bar=V1n6R zLt7sUE;?K!kygxnU~W%FXMg#Q`QJVew~IZ4=MG3XmhV3#mC^q1-rHN!uP{4dt*n5Q z)^#6sB{|PwCr<`P7LYGep89;vR80U+-+{3@|*U7n{{z2>0qICrsB_b1>5TnvrVhBOfj& z_2!KGzv#&tD7~^wRuZ-=c1)Tss^wz9TGrOqrPva|C!fp~T{wrs)iH8el`rgMV3R#q zwLw-kK<-z%w(}OQ)xD$f5;GDg9{cFG$+1CIF8T9M593Pz%oh&G7xCQk{Afb4C>ge{ z4kmlrl@*He6u1yQ43`GU*_$yhvMUsI`EYT~!fbl3@b$Ao^8dhpwb56@*ks&B`YJ|> zq;i8P^vJ#cYk?haa@h4h%W(15S=jsY%)Ye0O7DmWvTF@AH|uM99R8Vhd3yW5m-}OE z-Nj`a-R$8^bzk_>^P}g^F>3dE@bF<$qsMp)jQ%A|J@D5ZfBSlugeogn8#)=zQyOV9 zI8|dBhygC~AxrGj!&1z67!y-Mt6680DFIvjbJx|S+VWLE09cNeo5#Eha(UyqZGJjN zqQ{VYA$z%syn-3w8zq1#IC>_401x9)-rSsx-(;U?NmzTJEO7r zmYMFIjeGKLY4klgeIn0mnq0>{AG~IkyRWU_&|XB~5#259s0G`OO-@bqmlMTJERlOn z0$lc2M}7l_Co@aL?c|9QQyG5@e|Uab#$#e6*sZ@HP|5vtpe8eg9&?NiuTpd!Di)0l z=IYy|uR_SMw3t9hDIr|Z8Jj7&Z{I%Cz2xlk=FBl_+A(*DG@JKXgm;5%n+XZY$;r*F zLAIrI18`GJ@odIv_Dz4Ny>?+DJZW2%zHN!}WTmp-3u8F*i8$V1uH8V1tbULX9vgOL zVOC!Lr>YozHDG{=i8kf3Hl^Wowr5C*D`1&v{i8qOz|;a4(14>uaq-e+8&7E2)v_|nBNKHfcRL2?)O!j&z2N>9K#=dm;uq}=se<>hj za9vO;>-V{H=bEa~cg0U>;9?5hG$`}75GTp2ihX%AM%z56+OS{6zdn<>>`<=9q#bZr z>uXK*eb|)T=9sjG!l=mGpQc&iSuATfAGg(rGD{RrSrMxHVcDj21w2ny9h}i^;ElAb zX*50ZzY9pkA0+fjuwvfkCa;RO7}a&Mg!#%7Cr8iHRSt)-J~18j>Gq|YKVC6#et*8B zCdr&@^7+Ga!eI6QK&J1XoacW>dBR%fGbuLL6o+2I84jG6a=Wu(D3muEDhB{0?b6>Y zy_%43yDp|se*NJ*F9VO!mL;+_20&U|_CQ-%nVxxwpP=NUJS$=)yi)Kf(K7gy(HF8g zH-c<3L|5+iUUyX)vn)5(es|!4^2D!jC0Cf`76UZXZTowk@U4e&Ir5F8rA{hIE<%z9 zH!n-&Y;K(FaT*xwj0=$c1rKI;1M7buJP^CE;`H&=YQZ;O-Q6AV!Wdsi?k)YpCGi8s z#>OxH!nG{F-H!#6qpohMU8`g8=T46=cSXXPu!B8N?>_ua?!!MTrT_f!Y@KmV7T>3d zojWUZ+#bPtyemoFCUxPSlsHz6sy#SR`N93ilnL$h29udlJ`U8YkR z*726UZT3clbA1apfVL&d9&YoLC;IB)JM~hn^p-H&CfRt7QCQdCKP=(W*Hs(X?<23X zG*s5sX!YKk*PqT)SSMS5a>wFzx*B}{)_o0wl1;5K3)*}0_g_EeUn_&_(O+u(=;RTC zzikLlB|C)-eRYi8Q?4+sx%L{lDd*hncdlc0+H844wgX*BWv$rnj9+2`^W<=2=y-3g za*%bZ-cXHsXcxRzM@6`5pe8M}?*+Ux0OSM)=ePgvu9lQ)qoO#E!H4y(z1bF4R(rX7 zL*zFi`jq@NdjSF7E?lrg&d%h)gNdWG+2W_xF1-BwX!;Hfz1GqY{hseTUz4W=YBF!i zI}^CU;z0VxJ|#+_fU^9^0R!6@wTyU8n~|8m!qtQmMQ~8S5bZC=20H`7RRbdK+__V~-)r6jjkWiTb3Hgj z;5haNhfPh6dp_t~nB{hKj!|I@!m9d~&wo(V+%nPkROFBb`KMzqt`rF6jy(TiiS!>e zg1vW}1ZvM>E>y9Z3@yt9;#Mv7x}>SC?T1WaJpdM=`kR!*A&R?Fx=|q~V^;j~s%*ne z;P~~+tm3yJ@C0-vSE|*x4U|ZH#N4nPem3jtzZm#f)pZcgua#0S3Wa+^l_qzTb-uslt!fZeHaZx+H6>{_p2760^SPorhd+KYrNMvOu%%!0ZV1(A%_<^SAVE zHG+}OSB%8|fI~LzIATM~d*Z~z(HX*__zFSTo`k6L8OME(RWJP86kbt$bNxzk87o|x zZ)|~Au&}nCEFZ{l=-|RN$z#tJZ>qEM*(dvMLrO}Db@1XUMFFYu#S1YMI7y5`(s`fu z4EZcX+vfPE^NNmK5cVG%8{0B9^3eqbMUJC5R5{n70fsdIuh)W2qS9Z@yXg(*rEi_0 zy{XY{fS)TpU5tr2YzTMS>1ULE+O#)&xD*o;LIEpilZ4M2^-^G~K;_8}b-fG+i;nt; zh=?peAD7S1g;b9Ls}z|9S^HCDg}1T5sFS|*yvtl}%K$Gpqg(`S>7&p7A-8!!_>}Ig zprC;^Wu9D_(zw;smq_InEMO6!xlV>zLzYV*ZAbOn21`5Z(rh=?nD|&F7ObvusN9!? zxZGwt)oOb{Tl-CwEd^-ZvbE0Brt}f(Xp>qcm24WVO3U#(nsAfS#4=@`Tg2W#uBc{2 zGxmt8s%nu+H31WHn)o8UrJZomf6ra!51VS^DwLG#WVhaIZIO7tUF-$x4^5 z@f^9}xrYmnwIwcR?1a(jj_|CB{vsEouFA<@4B=!tM#cv_tEF~am{(GMXNQc{+jy(w zSG$%-o9xMO>TNLVr39;eGkC?&tspn0!D(DsE|<8W>h0T`9hH%7ln+|dw*GhL<%8)S z&wrXAL<)i!9mx5i(tbOOo*8jfD=FyQ&Zx5ey>#+|lYI=N5Q(r%`w2{KGMj$?}xY2&dzZkk1#L0T{VuW@f!3!D|A?xriLz_zakiZ zBxOTGeSPo}88Z_^?C@^nQmPHubQ?BpinR;HK*0H!_0*>aVxOKceh56&@#dk<=`Za! zU@4LMX*T5l$o9eWMk2u^f*2Y6i4!NV*nY5|g>%kVC?3pe81KzIng#oFKoGZvYpOto z>d;ZK3g~Gcm|dWE9eEMV$MI#t51J-x$~^qZLx{LG9Mre=voBLjoBLBr_Vdd^ZO?^_ zi#631{e+C1f3FzvxEftsp{Sy&Hc`?zvCkr6045}JvT3X3JOrM+HRX5aG%O-Cr#*)| z_#CMEHgY_fr+?2?#a##o6=2OGTm9T24#uj*I9hQYxj3{3(Lm{I5qyE*mt$c2ga-O7ba%!6-W zPyIS;`AILmtn$aNk%0K`TpIv`1?(sel9hsKlwf9koElu#*yc{2YCtG0zOAT$kg2v~ z$I}yWgdSQ;f&v1JPFtkJRmSM2k@HLLIazjFk^ZfsD#fz8V(h7Ss#=JAcx-HJa*VF( zql}ExhLb22vGU}rF+%lE?=Ogys(bYNSf#x?B{xtkqa&)cZ0SM=`JXWI5V8e)1Eepu3fr!87n#k zhGXN5jWwJW$1;kH%HuXHj;gd}RdiW!Iu;%q1)Qe1;vYSeb>l@%AVUx6)zi-Gsh6X? z1{+G|#Ot-nn_*{RC8Ewgl*1wJ?e3?@1QU9R=&3D#aNS~iaCv5UH~?^JYIIR2FN%yqZv&f z;WfT|dLy2FdQfb#sbdnRCxfj~5#6{b$qPirJCvzp2U<&tCPoL6j=%c%nt{>Leb`nJ z2??K%1^AC_ZX9nIP6R@7Z{32m>f@|+9;i-ypV>HRL+Fz7+L}{!)-nL)4CmSZL{Q2E z2{hbb@rm9)F=kUB){@WpBBA8w!oycxQ=W2`%<6k$;L_rUh-IDlc;>QbyLaIMMz`FD zYSvK7ieQnWX6~5{I*-Ezl_utV=F~&X@#OfhI&W%nV6@EBBc&}RIr&~rjuKEA?GZvk zu!evt5gVjRaHUCD`*r%M(aAP_*iqn^d;$L=VM$oPhZ8lgUw^q+^byfO)EP+nxV>;< zlOo)QlHI=NGNqeWdXA@gW>)C7m4*!56`5Si>zCqj>G#UxHD&#!3cV&z{+iuEnV&*; z@-?@T_fIdh0c43UN`@Vbeh|vNN$jUx!K(C64;~VI9%J`zo;{CkSyViD)ZJ1zjHF?-b363cpE-cmM9rL65QF4J$@)qU~b=$2A-leV&ZeR zUDN%d5aF<$SIbPZ{ulc%#?v+_!~9M3#d&&tt35+%Qsf_wmvs@6iSwN7MCC!>4zayW z-E&Nn+YJ7MgP^U{KM2zJpZO9qX-BqlRV5it78v*S>*xt}5Ae5c_4=d5m!2OjRpw2Q zc^Qw_Rwn|vcy?`k==#K+i-j@u8=h8v1GWiwb1x!dHFk($w+li{_*tjthjRR$&1m&X zMchkiQ2Kt!aRntyuhhg3yepQK=A#o2qInl%?^LQQD`p}vL}r{qyr2!f!maNyyFfPQ z^HqXNBChX%;FTP*GBV@|lTqwHFDMi%U1?Pwwk=ew?UvZ$%cBMA9$avvzkv#Iou~Mv z+;QwQOZeTTjjjk#83$MNQ%WAX^TIl3Wz~;{2*U)6UtC$j;UX%zUX2~Uza$4nS(a>3 zndmm!uk9UERHOO)=z7A515M}Vo008c~8EB0ux!--sSm*D; zO4NAa(fz8LtYNuuDk`15c)LURMrWOMnej&@8Wwf-c$`#hm@oM%9)aPF(iLl9g+&83 zk3LjiL~N+6uQ%w~ymgkgzJ3sDG^;uO68iqDRQ-MBek{K+*j<|AeLG~N+ocJixhOA`#?NNu({uZS00~c24e-@d# zrr&zR`q2KKsrU0eUy|KdjOAG(89tF>DfFhQN^IL1$5`xdkbySH?nCeh1s9m6j8jiN zV7PMbOHqq&S^z?yNaX?q1Q%lq=NS6V1RoQ9v+TuiTeNTS)|Szxw##M9!X!qbr`2vx z(6!PDm7UWtjLff?!WElxTPFJ?XV~j8oar|B(0KBtrFrqq8#h+&j+Jhxs5o$_C>0pX zQu)gTXU}v6;H-(x+^L7NE==oIv$V9NXaGXO{Hu#^INcdTF%3xOU<14&F;fa=2*Li6vv}P!gTxk6U9JWsj}IG|bF`5+?%8(FM~y)T zphZrnsK0a_e4|4#h^R>p_$e}5;tYihPHy8iqLrsRn>0;Z{Nhgfn5$!#mdD>#8ijIrLL6-hc`< z5)%N<&z=BgQ-e43=sPD8t+n_EVfhV|j@dkV|Mvvau#$o`p%Zt86_7#NxA^I_60&P~ zOO!OCkjLLn$!266^{6ItxaOhmn`W(WZaPtx4W>=Txv$Cb_H13(rOTE9oCK5g6ZuE< z!y%tj&rc`g!wT^wRMtI}vZ8zzN{sY9=Vr|}GF`UexW?E}cjqiE3_X6PyvgGrldGQH zX&?y*yZf;5vQt>zj6ylQ^6(1OUa7ED<5+E&1kNVft`xtz(Vbc+nbvUH^0xDbi-aCg z*exS>_=AYjL{fI3I^elStFimcS|dy_gX4kS5vSM2>{9@ashF`$d>ctaWNgNVdqbzq zm}vsCWC?~w*%1CUxgc>P##jDo(WP#eOF9CmMlF!GZ!HGFEI~|P`p4vpNOJrYp2(+w z4L8M_F4=tY@alcHw0br>Y=y(Y#RUN2GDzq^se2Z5#^v6wHuh9R3=i#S0M|t!)k30{ zQXs^s@Ie!uXi@$DNq^o*{@FKm43|a4#;PHVBR4we@=_OLuDrYZ8m&KI&VhzhdwNo3 zq&oRV)Rw;aMkMyD7~e=Zf$@znAZ(U>sI;cvWeIED(bxl$l*=wDDbb5R{P>XPRX%4U z-}Qr$*ANi6z_%)f<^&`M{@9jrDk^A*+X|8QBVmC`~4G^9Ec3Dxi&CU>uH zDkS(y={za}#0DTCHeW&R3jCmobK9|{yoG*QSsxUVj4UvQHr~~mj@CXg<@w~ zF-=;zjfI5QJW#Nr>Pn*~Z~$kvK%e0PCPC>-77UjPGry`Xm3LW_>s4LvqRJp%z zy6`S=Ft<_4sE4z4X}@zJ5ukg%qHTNmoh6E=9VjicN_lfIDkjGI<2~HOvIG};|2X!J zRPKZwRr%@0KxNMi?28(k z%l$@9qbNsl>YY;~0j|pjh-+!@ z!^~x-0rOBjhe~3^{b_sKg6^q7E;=Z zod+Wrs=b?-b6MB2muFVn2t*-$uPARgXlE7?a~?u)GZgd{@5;yXq`AC+U|*PJ>dW*S32l}#6ZvM zw)smM-UPV*5M%m!piDWGGMl0x+Z?^6%it5# z{L)SY>XIN+N`ZZDVmJD~7;g0BEyrG?l=+xE zJ`GE)b~~#dfNr4k2>_X!V}h5-4tMO6rlDd2}?F? zf}3;eb!{vHiBNR&2R!fa|X7hBC|^}a$l+hC+6{x6(RsG$-k z5%7E?AjM4H?psvWMj}O}{tZ=_yyduZv_BxXYugEojb)0;sgf38B4HQ<5|vn4t-E?{ z=|(XsWl0||50j1{+xrF1M?_=Y;(@#w^P9n^6?N4lseU9$78FlD1ZT~<$#oc- z2Nf^UF}MNVIC?`4Os)lbozyBxT(DK)E){qabc@NiK8(!}BL$R7KB=>odro|=0hBKC zorf#=rC=?fZE~vyD4@E$KQDApeiO)o7>>jNN#bB%l}g z6Ma-42g;FiL7S`((WaZ28qQS*235%?BtK*RmfxvrcH)HFa$L&NT6%S!gT~k5Qh(pm z`EeVJ)-LQvQSw(&l-Sfw?-;YS^(+z9!=0)8V7NkPENmH zS1xag#2|$?88Kc8%tQDOOb2i%VN~5quRZ`xIFu(b2 z7c&Z?6}tEwDz}olsso8A-%8?Rp(S{t2*f&y&kAL1=@hEZZsT>0rFmy&@J-kYH5b7} zR&b3Z*dX$eP4(EEfskGklsp~V(u=B}goG!MDvRg0-??MB8m@FF#|x?YbEH+^2M~(rlV>T{O5}$h>{*$6L|U(2qh0_w-iz2I_w_#>)BHT zQ;JAS#-^DpMU?>(KEMBx=_CXgd&n(Vy?1SQ=;=>UOk`ygRpqxy7X{#1NF`^u7OSyil`i7LpxwTXb0ha!G?tMH9W_! zk%9sj!E8Yj#!q62)JWR!i6MR+6mRHx(+Q~{KKu|>9+kiiPTJ+aDFe5*+-W8|Y?+~o~ zst63kGs{tZ@Sy~WLm+~RWk zc?j1vtrMU9>d&C#qMe*XOz!h~eb$cKt4 zig5r#(J5Ex2H4&F^@xlUWxP>kMi-pk_e2oN(dv;gI5v^V_T8vq9pRcFE)?Ac)>KQR zspZRsNtJAjq|IYX>4x%O#06 zhyhMNGU%EL0w}C#nK?!y-$Od`@x|35up?Y{T<+9ql3*5tgz&Q$foYLK1uCLFT*@y< z%vHOAX@=FI%ITRFRZ0a8od4RlBu08CWUEkwb=0G~?pbS8?HMWqLOSPI9 zR=S$8^JW@?g9E;E21!CuOAUox@p4Cgyix7JY+B-V>%=uarqXQ_Q%$mwv)R#sTkKbdmN}MeEE@Z#?b-00~knk+l{iE+L5v-42MeQ7n zN}8w^u1To-Pyr=v&?KDonZ|na@_VT8yh=o2ltoCY z?jM}PToYy>+4kUJY#~x|ANJL&SEFLwKZ;Ub5MVsjAF%Vq4AuiEEr8l&oT1yM;v$;^ zC>m-!`ALU#F52l6D8?0|+;{oXES8oiwe#hforJzvD^*>-;~oVOVoXVc%PzRINSqX` z3m462-9Nw#U>-`0ai;yv1sv|Sb2D2=tR^0CRYW-_(H-F(Zq_lW4PywPF@&8XZKYo~ z`Zc8C+x#X;VGD<=%B9eV!rq8k8O9(*W7s-NETD+0yJF*=-VN5n!=Aihziv_uCN)oI zWz6JXa~d48v~5Ecy8seSY2=-025m^s_9C0((u|#ZUd8fqLO6gMJVUH*;Sd<~41sN-jiKlIr!g--dGpWZwq zI*rX~%yB33*y>9Lf|xA{QP?h+jpFQOQ(M_X2mtXGfgOq(_bu8=CEfqE6fu4C!)l4 zh?})j^Sy<~y{j9Mzl+u3Eluwryuu&7pc5eaaW5AF0tFPEyA>MDh$KWw5+Mzmqs3B2 z^>^r9;2!WU7GbJ`aYiBR8u0Rsr>5RHb|bCuvmU_ndI35xanqoGI#5HredA!a(!~4u zI;k+&ntiusv)WHHK3Vwi^Gl2Tx{o~lJG-+kMnNa;$1GN!0)M>mXn%5YfA>|7wwDX( z%hN6Ra~;z`>LA_e{lOhc?VP>SSq`U|2~#se=zd_Ojs}}NHfyUnf21aawF^^WN6rdg zF9Rz}xv~$Nn7xBBhqO=r#tKXN`dpNeYrqL{+2IF7iAH^Vk>yp&gp%^v-I#H|?gF1R zLN~u-wHGtr5TY^ueZw;v*$e(U-$6z5G-Q@hiD*}6KTjUNHi_# zM^Wm(|M(i&tWnKvX;YOmod0qs<)+-(hq>Zz!oE}{?I%m2XojLHo^ zqn#%jd=X5bpd0BP-zYly^J}*Y2C|Glf@T2LOe!7}EL-WU8h(J8>h1g+QP+gLbUlab zC$3|L!mj|DW8?*y0W5k8lM=6MeJ{YOWrqGBFJquj%z2s@-wwp$|9g%W@bCE^SeYP4NMet<5Vnw=2ej1gFX<%a#M1ng+q{m zYaITG6?UBez$Ea`*lZZ|bRQNADGTwU?N=P;qZwDH3G`9;=dO_R_y(n)eTobB6$yo3;LInwA(D z*xP#km>)n#LW{vQ0WVEOAu^g0{-i%AlBRHIf%=yHG#TIXC*xGpnOO%@7iF>HTNuxx zm4E@70={y%q|~Qei+KirLZ5?SpjJ4oArWYQK0INug3uv{RV$o)V?3#CIu6Q16bluf z&Yy;Q_)32W7FTnqDPwe?jdt`~1O8xRkrq+tG39|1mi&D#D_)MdE5nOwhl!IADn!6R z+?@IC4#PZp+>a}Fu{lP_-6s&eoS4CFV|IlAL>tYTp<`O6V8(l;=z-~sy=44D`Ynp# z77f%9ioBL)UYH-jc9L@#h34!Q+dj-A?wb}sFWt^`iXc6+xA?XxNKCg_9m6?PRUJHx z?QoYrdu)2^Enk&z0$M?!>tGU> z>kuC803{+)hpt9?p>2W}NTLYQfFgl4GgwN@83%cbn^ua9?vZ@dbhdDc)ICJU?&_bu z)sNYhR=;|qiWTfwrRO)v8@a}N3G(-dnh`J`W4#g#Nle<&L>@J6qM%en8+ES$D+~E8 z(4Iu`{@ucUXhU50F2hlk3$|L0h)ZAz?UT~1rfkOTTAfk)d{Mm_0Y{w?&r_QtwM$Vb z3;u!SjB*^plm2};R#_@!Q#FB2dTU_3tS#u=pzmprAa{_r*oxY;u$hzaQZ}87U{Cr9 zEP?3eB|tljL8hK0K6p5nH%WCM4%aR{cp0>gpjBPrX&-m6ywaE$@Cy-XX*q*2RFm*K z28Mk%(dn4y>6m9*lqTSSxc6dXC83o)LLEDW2~ZQrQW*#OjX7xDvARNw;dKdkfoOaZ zFpB3Br9;UrgzXZ)FJZ<1N^`|(_+TG8ZXk*NJ!~X4hpPDoZl9m)rJT@yT)^^jWGww{ z7~xE69buYWNR=Z)EDRKFO(r1WX%bk=98oW9(Qh0(2;Q8_CYd@l{gKDRxGypza>AF- zYHpnBZmdAWWCD-fizVjG)L4d@3ZtFVBzzUD29@y$Tr+)vL~mYPux2sVCDn>QY(Ih& zv@X<^!KQsHncmE;ha6PYPemYdhL;2?8<+zpYhX5}g&kU;gWm5DpX4B|BczMz%=2el zT%KzMWGTJq7$40C3s;yxj2`hlc3|$8mI$(fK&s(k!_k;^VtdK(DFziR)7<0+CZfb5 zGQ^)MHcfxA;!PRn-j~B{)(&;gTW~**?7CV4j7upqp=K9CS!X&}VV#V@_nr9xQe>tK zmoxSWb=qjV2`YL}A&JN+sI*ni%$^p;#|wRDXrEu-F|UAhGSnFkCGP`+A);`u$dodj zSqvPC_|7Yqv6j5mq(_x6jh%y)g_1K2sQ%tR?V9SKiG?y`tS4}Dqm5#in{vIPUxSSntE1q z`qn`vLdDlm`z4%P!8KT47fLM52BPLXAw!4#KomwNx3lg`G7sOB$ruMV))ZRo670IL zuM5bApmd{z@4;neHMhTpKf@Fd^BBTt)u?|YsZ?H^+D11Ev$RkefMK7fj&G*MhYh@k zi`K?AXkAM|oX=64$qGR?2~6V&n%mV#rAVn0y8w`cX)7D*s)3*IDrKV1*2v${@L3Oj znx90FFq;S}xLY!9c_v$9g}i{$Vj#QMT?$74@Iwo2^h1XZF|XsBwVElq@2MVL5oess zE`UVB?)f5}dk`&RLza&+4yWCL^a6B}0iDLeZAR-{DI^!y?a~e42HW`o*+GSR6{*@B z3F#xJztfl}PYw0(`(gIX)R$Wbqtq$~I3hkym9X3BDn?A&+Q!7IFinUw%(z*H)5s1OeHx@oY?9>+ zl-PR6gdd2)=V+{BmO=XtcsHh>1X^%5bt(^h6yZH;()ynZj-lzW6dkB<3a99Fg9{=a z6yLtx-h%R38DOSGeE5n zxK^gq55fqp2^tHSW@EdSLVQGCYdATsFtI_1o1wBaWE>ZjT!>{rzM6`!i03cL{X-D& z7eQ6)1M+bVhbEJ%9OSfMCnY#*g5V`Y2JikSiawgJG;-J`qa(t%sD-91pK%%*gVdVt z)*@~k)dhOsT{;w3zF8~>7iQ8k{(?}pl1^L@8z%v%Jr4R1bD7p&!JqCW#v@OJNO_Tv z=!pTh{JqTP_cc;1LkKTm%aSedB?fWLq}i~CS(%jF+9(h)-OC%LXMCnAb-hz#STHmJ ze#eMwX1dRHB)w5-L_c>ECIem3D!s_&EhhQ4$kKaK^^NfOO?{T?s? zK5Qzv(~bbPmw+vkAD^$rOi92d6keznRVta;#bsl@h`F~1M?mM8Fc+d{9&;BUiw!sd zj8qJsT~lrdE!(Z+ukqJXS-l&CXeGeRGDhdr9o>oZoqMiOgz*x94Viw62&I z`I|zu#MEY}`b!2&4SUclx#mA8bBy^f^*rGM`pOj#s; zkjaTi%|@EiuN_<@VWdtC$l#W_#nvK2Bf%oNU#LAn2DG;ouq2(tQVD-DU1@<=X^iTk ztZu?OR5;$5zxEHHN~N(^*IEg>A?QSAVHY9fs$TMjQItmk0j*?|>1a!y(y@VXiKL0Z zeZU(fu+IPxJ5p5JCTTN0lf(tDZYYGKK$>|pgsNIl!7A51SgKGjlwd5WyXdr-0_cYP2=7tl0`r2Zr{>B260Dmk)IZcRPZg+! zafA`jkhyK+R%ri(-BM!_$`l02K!Fh!ArTCwJ6uaFyX92(T`v(pIW~uw4qzl|5GCc4 zG=c%-kOm!sR{Jwd9*-MAqcN$IgHs^P0Xs_3$@;}#`8AZPbT9xhVG~WWlwYE8N;|y^ zeHtYwr`UsBG{KsVqWlsFgalpwD%2TE2>&9b%WgRq>zWC=fDydFW0L-v>2*LAPGMpI zit?tpP%D;J=M1F;oTPkq%fuQP&cIyKun1yn)u0drF)kteQ0m|BB+UJjet2wR@^-R^>n<6dwLj?`rwL+p;JQq(D`QvV+{hSg&AxmI=X9Cml)y*Cu;dZ z(U^45fOl|HeyBCxgCZP4t(d*rUfm|;9Cg8kzdaqd@=wgH3>NPS^g+~L`|`%hfjjfv z*U~?SZ9Au=@tK+kNHYqoLwa!WeEma*sFRk_aH3Uv09er+l7-H9sd$0(u{YS}P;~88F>z5V*)5`|v5OsbpE) z--LUaqe9?_m57~5g{+@lVCyjuL{)qG9Q*QZv#h@;(PA;mKr|AN3;=tHx`vr%9W+9i zQU5m-uZ?z@bdE&m$g?&5=s8k?3vc*WZ2){`do;4LemaC|v?zC?h$OTK{+f(18^u~g z-cR}}{{Wkz!-4uO&No|&oT*d#8OJIz%`1zX*$l zJ2nBwN8y|4vE|c4J!S=!piiltl@6o=4jb_=LX})Y^ua9H6EW0us0CR%;t4XPYKNah zG_c-4C)z{1t<8k1s3(n*^*WDYkhoOtjaiudF2k`1;NT{fgOqIPqpd-* zkKWdR;*$nDGd#sP6^|Z6Q4+Syl(IyO1GLNxT;S;5Id#DH`028G~z^E*bNInH|ZL+4cXx`3iicZsTZM#Gi~Js+o9%i6MQ z-yTb_@pl?)s9AVzC*TXYy2s~cnWK}yPv$phlLp@PdZbAk?t(v8c+OjhcLi<9SD2i1 zRxnI#+IPUEN!H|8<4uLG_#H}UwQHsGW1JvUn?O>p!li7W*ZI(Mv=in&7P&{#JvzgJ zgt7$r?JlWhQi9(5C2HXZibBJ{n{$n-V6JLy|y(Ex?|;i;crAF8YS$m5++9JKFPnV# zmSnmU&1*Orryh@Q^AK_<;3a+okG-9#`OEG1(h(&jJ>u9wQ$goBxaQZnx9df{f=3KFz{dl$XPLEPxQNZ zG|q^Ip_3OHTrup}==>0s2^5c4Q>aHY*|NG|nSEGpz;m%t5g5f}*n;$4s_N;~R4G3H z*q!z6`$oW^Rw8#$%xtvNGoO6%4vFpn&U&*x7m%X3=AUb->)I@bWszv|S+)1ALptzq3^V z?`%5F4JnQ6*m`_<2-;k&9saOJY!J8q7^|RNm?5|-)pLHd!p^5?NopLfP6Uod0C>1% z3NFvM*YfkJYm=FiMsLfl8@nAF-GXE-bFTsWz~@mqLvR97r9U-Ski00m%-}@_B?1Vs z;KT&mz<=K(SKZknztiaeST9chi^j7^lB_DmJu-XWJ5{U_Y?%K6h`fk4YvB}5cq<`< z+&iLr0)A(i;+qtA?&OI>iuT|Uml-USbUAiB%k!XYHllvv@Sbt8OqMWaOA^gKDwdWH zg?zl`CH08Lb=wQsSY-WEYqC5uoP5UBXKE zn{B!q6q>er$E_h*r{f;)%m|=Ng38p3^{6FJ253cXefMKJJpmBMj5;+SoHxB)jh3i; z;735h1)Q&vy@B}wI*UttSqtkS<7QGU9GQTkb=Ipor)OnH;T%Ymq0iyTE@_8HpxGSd_5RI8CCaAVNyk-kwjC}qpqkCZJK6? z2|>wuCqxt{fhpU-lNN!n3?fd}=N->g8z(90g;@fJE|GUQPPh^UdQ;54_Qp@?wJE&i zlnS6Vpti{oJe@SWOrhWstW}h69wyDR%+wHVT{s!xCWRlYv0r}fq`SFMaQsS`=|K-6 z(&U}1rs_htv#qD4=Q`x;@vpY6#Re!tckAmLpC9_4brPodLjAO)kS7X};m%2#jKf}n zaS)%4!ylFm2TtIzp+Dae8fy+0p#SM_phj3B_0)4g%^5CG(#h&dBe0tzgG5pJ7$c?x z?NoC8*BuSR+s~3?b#ZOnVrzp1gMTrNEgW)ja|NVF=bCn5{l1N7Pr*G<@62iYe@bs{ z6sM39EN8#TdF&*uF-WZ}RAOz@qtPXvecrcrx6C|bX+)tSWtQFi`~&Raf1)yI^_Rbn=}Y zI4riy$2T{rmPPtVDvl5ZRo^bsO(#3jHlpepupecWq#H#mq%E66zbqL_$J+Gu_!YGf z?q`}munmSJ0*29&Niu&NDv{eo%5I0TY@8`X#R8z~dXwQ%568}`7(201=sB_*Oi6zO zx&-)3Pt?7xQC3#gwr>jd_^p$!zC9oLw0VY8?``Z73aW*{PL?G705KvBevL?=U8Vgw z)Cdm=CaLmK_B!AYEK!?rH^F5XEQKp-dpRG)Z_+V@;&I7nbXZG=1XFv*E{U(PhEvXyL z4F&Yr<)Jvwmrmm<;m!u44i1!$(~Pv_HN9I)XwIv$f8i)YyLtjx0rk%eDo z*roz3Vbvb?Z$fdFaePQr6eRynMxcu}Nbff^ZO4~UM;4Vk{nAj{SjC|X9w~Uv!yj{4 zhWi56Ir})}hHZoJjGCy}={T@eTzoRNQzmqDFOyl(`E3vplMo3?DV)o2oK!UvUFOO~ zbUG90ZMTu9BBfBoXTOr$EnoOTY1r|a)9XutSnM}ph`wlKu_h8~(3=jT*}cNWL<#3( zF55OFiJUG*ycqQ&Q_2J$ zu&}*eBK!EPx7Q`L`)ER`8q-*5jl!}${Bu?}o#SJNlO3`?y3;vdk+)BO${c&t`(=G& z9OxOU-w4|^(lBG8h{H!Qhd?x_^w29vyu4$+;_ETdAPoeKL>%+1_w($q8IDmz9kVvk zVIiaFk?<$>-7XqO!H-_RQB4!?8Yc}=7o4+q7ONM1V#G4rBCnJK9ZaNDVFHxIEyQYp zk`Y7+Q{>9ov;`+DEDA7zld*B0;4)hOTAxQHBppP#c)rp=U|BQ9 zxCi@?xrpKE%T2BB*DHd->(Av4^CW*qT?EZV_z1)8=>~LCZHiI4D`@OiI!bE}irB)8 zPH1buNooru{l6&>rKKV|RQT|3taj(sBEZC9pC;*u$NA_TFPviVjU-F|JAR#{+bMV4 zkR<2_hrrFZGgE)&BJ1z~Q_j&yfvD%!>;Z@#776PN{Fkq07H4q?;~6aKA!CW><#bBpLq;u>W%huNV7z7Ulsc@h2)O0nMp;!wyz zO@qV5-r7J%l2ZMRzhLE2bhsPU--!5U7oekq4p0wAsQRy3-v$Cz_&6nS0IFPKY3Yzy zIsuYueZl;J&OtI>7?ATNoPRnu&Vrx?BBMsFKWu<8YK=!Yba3685QhT|3H1bWIlk%)~e%Wl{QTVEKG=dMHtxDUw z8yv|fon{RT!4zWAPYh2`(nH5BQJ)v``@#e&Y2fhJO9F3LO^Qq{J-&vZ1QD10)kOh%ymVuVuFVEY=(T_PRjz1`^QF zF^+nfr>t9?kyX-hpTANbOT7e#QBY=dr>srsND5nVQYUlHWSs2LG@SM0IUwulbIeIf~36FNhjYwsmJ+vqolK?AnRR?Iy0RKNUS`Ry^GL+mw~2Q%%X@c-a{?X zbOUn2sO53!gjoY+irY|9XbGT?AP$W+8q%PbFOq$}cg?;l3sZ%6JoH1YaRMI9$Y+S}5u#9Nj6I{iT8qef2v= z^&nC*z!%fx2>q!E{nbYRM;XyjqE9R?{o=@j{}flS`ixIIoP z(k^^Qh6Tl>X3zAkkFPY5KE!&GI%#vKI&+~ETf-r_A4yhP>jb>iaM&vJIJ?j^qg)7S zpmB{dVzeFPrQraHd0OAJDz^J21MESrw)*SaL@igq$J;}Of&xw~Dwug0%x;MWztutq zKBJPPEjJXU?k+~V;)JC{%XC%3D`a^#WdEOy&0Nb~L!V z6F{XjicQ56J4ioCRLU1KX+nR=vN{6mp#!n&(H-Y^j7|Y1n}fT7V20I;lg;TpK*aXA zx0RB;mbYk_r0__mk_zB>+=lsN-5|7pALyKSly;(>3^#zcnQ!q14my!4FB4UC~|Ct|BB{FJ4rr=(_lN_U8H+fiE)re`VO^h67tB{76UzNBs z9p?W3kZDqx5Ms|Y97vOWP62}$OEN&8FzQ=EzoLS)zB%j96wH~%-pq5qwViIuY zX3hakO`qJ6%eiZz`u-{~eef3l#UVOwj$Hs^GMLiDCWs$f8MWs9qf|CCbp;724kfDw zgQT`D(obum9V19;UdFp?gn#8@`1t~Izf|TSjDqt10Xjq%n}%8^=^Tov~nf{maR?62D0FU>wY zW7lC36K$8q`4jG_Xot?cGiTSadxh67brg`huYOUo81#5qfQobMfoxY<=)Q_6qAIQ6+@%=Bqt(rLHRFmDPd_V1S zmt?$aK;|r?xIp31>^7Iu$@;e7&Mi5LuBAnOsSll|HN9R%zIbgInx?}ma*#)z(8au| zPAC24SycUru;+^uAswy5zYkNr!Geyu{;%uKST*J-2Fcp0 zaVfx1pkrRkVP5-7Yy+h63z)aO1dk-%4Fvo;9+GIO|1iJYGcN8dkj3(1p8pG~DtzBT z^<`vKOLe~c_U>P(#RG1ieId3%!cS#m?DyXWx{Xe4ib)1cHGZnX^2*uHy!k)f|EIkx z4T|zix1SnKOm1Y1$;2p2Zp6{Z2vNg?MFB?*(=KQfXhony)DSQTB8#j|oJw@EKx`$L zi7aXen;;+!LJRGb3jr5IR%rpFARsnGz~%x%-*djcnVPzH?yb62Q#F64DnG1L^DXcD ze((D{=Q-y*Q`(Rj=YN(c2r-|Emlv=`SU)xIX*wgr*Rq8}P~jy0T%eaJQt`v)v?i}p z%k{kX&{T0?CkN@hgtU2ve{aPRn!Xsb&}Ntl0i$(urZ{9~pz~7>1Aqa}DrC~~F}p3E zA*D6u4!I$Sw9Kne%mPQ>-xJaZ>2s7ujM|j=LLf1ZbK89%KX>vq{Eg>WTsRgrhr4^X z;eRy)kU5J;!MqBa4GXS~=denx#F-qTWYi;FtNSAfr#BMUEe$?|eXAHM_5BSXKqE`8< zLiZtYXcPS9`c~OUqAbg2zThMiN2dE!ED#0v7SsG`Lda^4^37+<5fyGL+2f_y30Rk> ztvCp;8Xh%`WF3^^t4?T^Vb2vOV2wEF4w>Uq8c{ynBu?;p!?lz~`&YdGA_OglC-6ht zGob1Uie*dh9(chpyoM_ugQ@XO$Vw#&pmp@-zVgZ>AturAha=q6){cRgmM1>^0&y00 zv^;xWd~)&oxp@#-6`40A+jVgGlEmRFhozeK*cfU*)?7e}4wi#SZD?Ecf?zXUoZQI2 zU^Oq&el_!z&(12k$4}b_DZ2_Uc=D}WMPG^d&5E^DTJ|i5DHr(Yd zz=7+i%`417dAR{FGrtbJ}d-o9E3!5YEdRa#Gs5Q^#2NhprK+udYGhr>x-pIG(y)LdhB=_ckTTf#{kuZiF~f9y40U|s-7qXT;Iu0htEBy4UQYz1x1;p8 zS=!}TC&caW?ZloCH2DtHUx~W&5rc(5wu$N?k>026m-`ssvRdquESOzI{kvA4_M!A6z0zQu=c)^I>Yl|y4!8xskyHS9_66+26D@6wQZgDLNd}mo z(R`LiM|TB|H?lS)3JA_R=o#~s#i-*RgYN$SF~gagOFb={%AR}uc5%RV(3}+JBjDg2 zLv-7OQz))agUUfHGWGcq^5FpFlyE~2kzpG`hyrs^u;#8H9=QMcM8Yo%@yvcN#pTxlTz zL!QV>#e`!KUBhajQg8Odf*rVmbYr8XMB8Z9=pY0=w0ot=s&ZN5;yGytv9Tw~#+0(F zX6R@`pEL@sgV*1mVQjN$c)@y;Wzry=s-eld{^!t3I`A!`@6Ad@ zI{|di@<$KEvK;r`S9hBUX{yi9#rE9TI?1a30`_?X4gKa4I`qsKqt{&3VcAg1ms5rA zJda%2=@=}CMfXP2?eX-Rrcf`YFdjYG*bw^E+88LBVkhclZkU&wUP=%3)-ge7b-VOr zUHFyf_n(*+*58(F4i?`&J1ttO9;52T1z$<^R$Mc?Vf1p#v$)9{d9t!)eEgdT?>h7q zUz_lK(_2wOOf2B#a-MvlG{W`Qc&bBuPshFLp9Y*t0&PpZU#d>lw>iu8fBD^Au%Tp% zoiI2})l8q-kG0(z{kq8L;ve~q>-mNyTc_kvyB1z z{4JFbsc;YYt>6^i;4dzf5b4Y&tJc>n?tVIlt(X7-(@0fx3qxRX#9S4otwuePU2CD~ zqL-eV$$uOCtB>OsC)%sNof5Pj<&|Y%dsg=Yyibh!9_;8d1EZuZ!&K~HegVpCS5bJX z`zaP(ES^Sdcgu1-ztMZIc;np zq664qK|;)L$-0RjeYu-U$z{U|FO|7+I~T}>tR7b{wQ(<&A9oC=l_TwU_>X`1A$CFX zfu=ZsJXR_fP^~O)>^vTXN}FfrV2M5g#B_fzkC!SBwqZB@l18x!!FqdD#yGnOUEEmv4yX(sNx}k?b_=2V zJ6=4-6~8_hfCE2##A>P#cjOWl_M(lYfKHndnnhL|JdaBjNMZrHbZwmI*vLvLDWXIm zQ=kSLJ+yuLBTb1F;9hwuJN1z}%05IOW2&!#9TcBc4D^ga6^ZBU3rD^(bxRo86oNJ@ z^gmZ-{d(WtqWg8;=*hr6j@Y`GBBL&86n%Hf2zV9WvZ%Isu-2gtECA7alJW!6+SWn+ zc4#k-s4DQ#zNW01ybvBr^hAU}w+b#%s?*D;O&p|MA!UWMjRbE*L>{SuvWk8D1?e=% ze&fqh{gdaleyj4o?(AIo{+LcH;`T_IX(9%fhYv_Z##{FvbEu&;2Ok#%h-&udf}x`= zo2Besu{__S64P?3S`daL6v*02Y$XvuL7cT}WCd&WI0#G&VnDP>)Y9&H1hTi&Pbb<^ z@zh6gh+gdsE^drOQ+aXRwRbk+8Sx<{)nIqp`^$v8m-Yeyy6|Q<&$j^>zZceNMRqr6 zGf1=NQEHhX&DaA(L;ZNJy0#~lHD{c8otaV#OV4v}(H3g=z^@8tS-)Dzwx}feE|Im~ z7Uf=0Dm<8ZrbPX4kR^ygU%AfLt1komP_g$aTq}OkQF)5h((_(!gPn7vU3kN>1Rce4 zOK75(p&8a{lIXctXR$XBH4<7-_J%yAL0QKS-|ew`2=`W-cLk^R-ypNHnTmi!R!#Oc zsWX@?P8Txv9q%}aYUsaQ?1pjM|0HGu+Ozf z^}||arD)e!hQpJrc?T|;*oW-|xlp1Y^`1#EzHc~5xZ75M=6;=EHF=>XjoWhfeV?&U z`qjw(Gt^U1eL{;rydv~IixCQ610!x9UIb;wJ*|HRHN|vE80c&fPajtvF)q{V*jEFn zvvF|(@mw~G9Ypau#AB6p{7hk!V0}JvHlXUE2aD&R<&S3@!svrzWZnA6%f|cBQqe@G z!N!+f1B!J*^90-29shng!0rJk9qn>e2(Fid4i~H#08So1YHMHDy8Z3rhe{VFCB#OP z?do1E@7>;RSDm39T{izsFvv>lD*7k3$rS?$=pW}YhB}+7u0{zqUv)7JS}VN*OFhf~ z>!lYa897dag-$00;5|W(_Thb}hg3+0_giO+&rrO1GviE-3 zvJdH7_jjJC35|neU2pZoPTaGaI2ruPU^|-i7#9cBXTC=zh=N#5rPmJS zjl(H7yp#{_m$O*P9ko9Q-hEu(l@p!7}dd*!NvWc+ir% zg6%y6Q5JXRU5~JI6Nc{({eiE&pUOWi0OohnH}{+yu5ClBnvnJV*5IgF4l~x|ZVjYs z+EGobz8LW$L2SiNz>=jnm{53+UO*BQz`g{Jzj@8F=RZHP$|pqeaohQk{3h_Ty#3_p zg4G*rMm-U?>(;iq9WdqiBl4cRFNYN6%x{0PPCc~FtB)9gB>2)FPvW3;_bWjY2oS<- z(f6n`16d+Aa{4o(IhfCV5IW;~&knhY?*VkdLJQc;bof7p=fKlCLxXepJ-<}>#E5gN zbnJa4)EM~zB+Mcq5evDb!=tg;2rG0E=pB<__sRjla-I54fWn*puGy_2coK zIb#Ux`|HN#C%jR0+9~UW>husbJ1{iqwVJFr8!rkhl;+* z)7@?z?&(`HS@?V@IkhkrGD$;n%7s3*QRW>!-k znlSv!YolPf>8MFaBQNtO9Evj?Y$o%99TGw@vK`DC}TB9o+S#j zSN%CZ$JBN+ef_$^V;4vY!2+kpLb=WxuSfBTqJvy8*u38HYiBVq_s%?>(ZpS&85$iM z){Hz(kl*Qy0vEuo1sm~-kGKEwMV|5?HYa|DFaMmky(I5qELcI<=GB_cipa<>g}4t` z0k~;FF4o#8y3)caV4z(;-hOa}D|#y~M(}er+bw4)XiBHNS4J!ylf%)&t%VBDU?`41 z#72ao-iF|XH#I=yU^tXHhFRNjQ$$l|da4w2fg9XUS zC?EkhwkO!pa6S`dOOsMyh5nM6Ph-(tnC(>7;nv8P@E>z`BxoG~ovF1u3#@&PyXST` zDE9fO6L8>C&~K^#FlbltiQJ$SGmse5;fv>5RG@YksYaO>lYC~;kG4{ocvqO(Ko(I0 zSkCTF6BH>Fx6pqc{XD>OVWM&6tGP+k&q#v7Wue~I`ODl!+go*|IS-;;FQQ@nz@Iio zzfpDK^)T$vgi&@muoWaa$WaF|ThktelNqr&_3*g`H+rotE4s zDmyH@)M7-d;a)wIjl{AGd-FA>0-`(wDILr)Q;~pnAsDJ*TQDkO8nWF)_qpES-Ftvt z5%gj}_p~u^+?~)Ss`+R>xRib^WXgc8v{=lb+A=^b=lA@(fR>|lyR5;;EUM&q!BD&6 zNV0wSgLgngu{Y7Px{hO3G{R;JVzB+02QkKH?|6|`y5YOu(LQKxjSR4%Et6N@m!f`iP zD`)nu2@0l4iu@P$sYEeD@7WuSrf;33k+|4t`+q8fGRY%p%BMxh9IJ;jAs|8|uC#>v tnOJ!HTU4ym6eG-BSN!XLJ|3$#i?aQ3#}{|Tl>PqzR7 literal 14082 zcmeHO4OA0X7LGqu5CN4cRfK2}aVr=t)mTAN!4{|@9;%D-tMzX!O0WtUOvoTcvOQI) zKLycBwkwuZTcUL-90*AXRYa??g)UlThLJxZzzjqflF4Ltf(fMSc8_Pzo}OKI&UraG z=e;-Y-Fx4?_kHia`#xH?Hq>e2%!w=(%PBl;RTPV52e4S%y)Jqwa_cK@`~lY{e&V1zUOy#abyxh>zY+tSn-YCkm%ZG3pm_1mq`_^Q zS~NFR(s1HH!z(>v((&~zK_`~0lfo;Lj-;tfcKM6g;T}i{Gc6-zMx_`7V%!De9GM-6 z*>4$i!CiYj6I_!@@@%h;sZu-?gj~d7=uEr7`ga01z4mg zg-aw>G#kGfP=P**(jTorS?ho%EV}ZnkihvDJBN`KIC8IYN6AP=)U6_`<-@sMC;53{#>*F~Odf!X# zlWV1{>7<&d4dL%M%z#jNleZ$V$V!kP7T^Q-h-fUlgl?V&p{nwLjlY~A93lH!`0|c- ziW)HG_Di-bU##<26S_X|O~rd{m1MEEReGPjq$Y~bQ%20-tIo*_$}sX(spso904T>+ z8js$X3UM{FjJTAmflWYXT@BTSnGC+t3iy4zGHbu(3{e>28o9(c3C&?sJ*YfyVZrI4 zMkIqYel6@067j38>+?pFzH^&=3^@?gE5}n+;85Q+fjo3vdi7!|$f$$US=<6Amh|

    $CB719JfPCH+>6FHX-J{hdYy=T#wJu@B93e5)qAr0!(7RkC(&e!=M z9tS*|ZTjpzqYs4m{o7wgd8IgB0}f95?&w;!RM6GgV51lHoqTZ2^=MD(7EOdy@Yru@ z-v=M=vK-l#JBfIZmNP>sHl_fu_&~$`4)<>Op@j+^mtIpINIgE=8n{To=Uq+f)hPz~ z$reG+ut%ri8Mkk`pA$Qxw>JwMmvXI~brK(SG8zwIYfVL&dgco+=Rx>0W{MY2p`HRjd9} z3C?GZvCN^JiHtCDASUX?#Q2zqCKJD9g5)p4m4Ka?!{G=5%f(eru27B1!P|0S;b^o+ zmrSrv=AN?LZ}?(fiiTOU%bM0%u0yVVn}a3n%WO^yv0P)^7|T-&?~KW6bYcb>m5C~^ z;#RDxIoko!RsHQ=la0$q$p~TjD<(n?nKlo95G$#m=@K zWFwXBtg4=DW{_W(iAd%44?3t9OBzyawDcV%8(Ctj$H&o7z{ytATZt(N)wXKa+VvdpTn3kVQOo(r^IG;|CMRQTX=kmIZ4t=)TQn&_o9y`DGO)C@|>W5khCX*l4RT z2H=SvUxSyi%|XA>v6}WEmB`A(1Kt{Oiztk~v$xqxG!xMp)R#vb)y<(}Bs5|E3a&Ww zt5L!t9nd08RNzR5kzLnGf_HpO5qcRr;j*E{TYWY{30GjIbdi-?j4#0(wUHEwMEo99 zJoz*_7?6c3VM8C<$usyM`G{_- zeO$NrJ#%i468<ppgiVDsPaS9Sas@ux7w$<6=;L)gqx2)H^ECC!Mhkz9uxDmvh8XoBj4VcF zf%hMZwbDR|v)zxJMH%QbdM^LJmPM~NnPPA<6UF)MBU^BMQ%_<_ zwFJKk1~U;!>yGvxiPAEe?@XTmi~Xu$`(f^hyu7@fGM1n!OMCI`Cs<6^Ql9hlxX31n zq#>$QzxPV>fv$OT6}jbiat_%wyvLn`l+^DL&+;uRD0gs>Ur5%qY_1xpj|c<&QqQs{ zI2IQr2dk4VHX4_0Oj(iYN@r0$#{(7IOPL|m-R8?VmBKNv>VHg|G^19=FHH+m6H{LA z*0{lytbEb-fb8#Eb4nWn<|1Tf!|y<27o2C?5F0ob*#ME~JrL~M*Ir6F^$$Iw2fFFk z=;wI?drczXn2Vd(Pv6VVGF{FLGp0?-zr{{2O*cM_7H$J5J&@K3CeY5?~7_eD`fX7AOWUAvMwmVv}J2PvqI- zKeX_~^;!*|eHk3=;_I*Ffyx~usaZHjF`=67ZjVMo@2<1IU9B6MuQ2FBsSRxE!TBIl zr8ZO_3Ym1M037hM+fGmCq9%SSefd-s{ZWwk5`xQk1n(}-R5ugSnC`+MLefWNB6{0$ z_B7Al=vgoU`H31UH4~)tgrb8E7}o%aQ~}N*P3x;_as(W$*2hYKV3{k;L0?DY5|MG} zBAswqyQ06PhQO2{DZw`nk3}{)9LuD64_vB{bzm)pkb4Yna6qe`l6BRs-pmR;5zf=L z$g3~&R6zzEtRKRgG`)SYCZ`%R(e`Ku-~BIU35u2|c#keTAY*`mt0=XWd{Qh-FwOsx zC&v99)k0ri7tJBeWLkRAYjgVj+k+{4RP5yJwk@2->Jn01HX~PBx!a{-kU(~9^c?(p zMO-5)J-dbO7|Kp*RJS)G$>!=e6VCFb66dG)@VD2l JDh}S9^>++bxWxbf diff --git a/docs/articles/opt-dead-expr_files/header-attrs-2.5/header-attrs.js b/docs/articles/opt-dead-expr_files/header-attrs-2.5/header-attrs.js new file mode 100644 index 0000000..dd57d92 --- /dev/null +++ b/docs/articles/opt-dead-expr_files/header-attrs-2.5/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/opt-dead-store.html b/docs/articles/opt-dead-store.html index 17f6581..c5b4e0a 100644 --- a/docs/articles/opt-dead-store.html +++ b/docs/articles/opt-dead-store.html @@ -31,7 +31,7 @@ rco - 1.0.1 + 1.0.2

    @@ -56,6 +56,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -74,6 +77,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -86,6 +92,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -112,7 +121,8 @@ -
    + +

    Dead Store Elimination

    @@ -128,50 +138,60 @@

    Dead Store Elimination

    Dead Store Elimination

    +
    +

    +Idea

    +

    The removal of a value allocated to a variable when it is not being read by any following instruction is called Dead Store Elimination. This Optimization is necessary as it saves processor time and memory which ultimately results in the faster execution of the program.

    +

    Background

    Dead Store Elimination is an optimization that intends to remove an assignation of a variable that is not read by any subsequent instruction.

    For instance, consider the following code:

    -
    foo <- function(x) {
-  i <- 8818
-  return(x ^ 3)
-}
    +
    +foo <- function(x) {
+  i <- 8818
+  return(x ^ 3)
+}

    Variable i is never used, so this assignation could be removed, resulting in:

    -
    foo <- function(x) {
+
    +foo <- function(x) {
   8818 # this line can be removed by Dead Expression Elimination
-  return(x ^ 3)
-}
    
+  return(x ^ 3)
+}

    After applying other optimizations, such as Constant Propagation, some variables become dead stores.

    For example, consider:

    -
    foo <- function(x) {
-  i <- 0
-  n <- 8818
-  res <- 0
-  while (i < n) {
-    res <- res + i
-    i <- i + 1
-  }
-  return(res)
-}
    +
    +foo <- function(x) {
+  i <- 0
+  n <- 8818
+  res <- 0
+  while (i < n) {
+    res <- res + i
+    i <- i + 1
+  }
+  return(res)
+}

    After Constant Propagation we would get:

    -
    foo <- function(x) {
-  i <- 0
-  n <- 8818
-  res <- 0
-  while (i < 8818) {
-    res <- res + i
-    i <- i + 1
-  }
-  return(res)
-}
    +
    +foo <- function(x) {
+  i <- 0
+  n <- 8818
+  res <- 0
+  while (i < 8818) {
+    res <- res + i
+    i <- i + 1
+  }
+  return(res)
+}

    And thus, n would become a dead store.

    Example

    Consider the following example:

    -
    code <- paste(
+
    +code <- paste(
   "foo <- function(n) {",
   "  i <- 0",
   "  res <- 0",
@@ -183,9 +203,9 @@ 
    
   "  res",
   "}",
   "foo(10000)",
-  sep = "\n"
-)
-cat(code)
    
+  sep = "\n"
+)
+cat(code)
    ## foo <- function(n) {
 ##   i <- 0
 ##   res <- 0
@@ -198,8 +218,9 @@ 
    
 ## }
 ## foo(10000)

    Then, the automatically optimized code would be:

    -
    opt_code <- opt_dead_store(list(code))
-cat(opt_code$codes[[1]])
    +
    +opt_code <- opt_dead_store(list(code))
+cat(opt_code$codes[[1]])
    ## foo <- function(n) {
 ##   i <- 0
 ##   res <- 0
@@ -212,16 +233,18 @@ 
    
 ## }
 ## foo(10000)

    And if we measure the execution time of each one, and the speed-up:

    -
    bmark_res <- microbenchmark({
-  eval(parse(text = code))
-}, {
-  eval(parse(text = opt_code))
-})
-autoplot(bmark_res)
    +
    +bmark_res <- microbenchmark({
+  eval(parse(text = code))
+}, {
+  eval(parse(text = opt_code))
+})
+autoplot(bmark_res)

    -
    speed_up(bmark_res)
    -
    ##            Min.  1st Qu.   Median     Mean  3rd Qu.     Max.
-## Expr_2 16.78983 17.87406 17.42409 20.89991 18.14949 149.5256
    +
    +speed_up(bmark_res)
    +
    ##            Min. 1st Qu.   Median     Mean  3rd Qu.     Max.
+## Expr_2 20.26309 20.3466 19.20069 20.22119 18.21587 39.05164

    @@ -237,38 +260,41 @@

    Intelligent dead store?

    If within a function, a variable is assigned multiple times, but just the last assignation is read, then the optimizer could keep just the last one.

    For example:

    -
    foo <- function() {
-  a <- 8
-  a <- 8818
-  return(a ^ 2)
-}
    +
    +foo <- function() {
+  a <- 8
+  a <- 8818
+  return(a ^ 2)
+}

    Would be equivalent to:

    -
    foo <- function() {
+
    +foo <- function() {
   8
-  a <- 8818
-  return(a ^ 2)
-}
    
+  a <- 8818
+  return(a ^ 2)
+}

  • Remove variables that do not affect the returned value?

    Eliminate all those variables that are assigned, read or not, but that do not affect the value returned by the function.

    For example:

    -
    foo <- function() {
-  a <- 8818
-  b <- 0
-  c <- 1000
-  res <- 0
-  for (b < c) {
-    b <- b + 1
-    res <- res + b
-  }
-  return(a ^ 2)
-}
    +
    foo <- function() {
+  a <- 8818
+  b <- 0
+  c <- 1000
+  res <- 0
+  for (b < c) {
+    b <- b + 1
+    res <- res + b
+  }
+  return(a ^ 2)
+}

    Would be equivalent to:

    -
    foo <- function() {
-  a <- 8818
-  return(a ^ 2)
-}
    +
    +foo <- function() {
+  a <- 8818
+  return(a ^ 2)
+}

    • @@ -290,7 +316,7 @@

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/opt-dead-store_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/opt-dead-store_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/opt-dead-store_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/opt-dead-store_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/opt-dead-store_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/opt-dead-store_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/opt-dead-store_files/figure-html/unnamed-chunk-8-1.png b/docs/articles/opt-dead-store_files/figure-html/unnamed-chunk-8-1.png index 2786cedbda65bd02d2992e83b7f147808aa2960f..0522eb2145eb8e4dc60c5eb93771688efbd4604c 100644 GIT binary patch literal 38322 zcmeFZ_g7U}vo(CwrtKEPHPW_$Ru2M_6chv$X!8*SBSDgZppq1jAj#H7+lU?%M6!yg zfFMz_36YF|Yjt`4|(CE?n z$Mp#Q{`&Rn8>OW^GrF0PoFw<%6(W}^SEo;1;0yt9@95%+VuRzi@jW2 zHxhV7qA13sB->ZMWZovnz;EBa6%`k+}aA`uzEG4!7vb7yLZmi|5Y|bobtuJOAw-o)I+BcTM1TW=S*eUE-rZ z1gHqxO`|{j_b>jt9{)WK|G!wndC~JMO^cdi1hdOr}b(4(C|(KIR`6iZaSscVHW1!1pil9sPVy_)u@xT)igBA@IqC~j!Y{QBP_ zc7coGuVe)M9W8(V{dX?=^3QWJ)A;mtL~Y%lJbJYH{ndGS_cipq6r4ub9HX3v_ZO;e zUe%R4>CCaDkKM1gC@FuIczgplspsP|4)=??wwSTMlg*hkXA}lYkEm-2;JeRV=>K+~ zZQl`7Vp*(Z926fPpVK>?ZuY~?@1E}S_K@#SVglxsOX*%VXL+6^w)Kt7pbH z`+4X3$_sKB}J8$KD^?u170UjPK+A+o-fvV*8DOmS8ljb z!bPhxWvn-=uRvKM#+r$o<@0$Yq5~Ydl4Y=FYTVGI-pn%F8YjNi;WfK%eo7t7N@a7J za=blj#+qbKPfg?l{d|bVE5Ge&!DS%;Z?JqC2~0=%Y+zl}&|Mv3-I~$3rY$;A^M;FB-;(31Zy& zxZ~o$&bPc`0rBy&CHgK-=SJ^rpUc^~Gwp=GMPH6gUia7c?(8{p=3s%9JvOhQpXhMI ziFwnI;okQxBz!Grq`%EmEUwb>TAqFOhV&<~vC^F0H#6oneBRBkTW3YfoS!0O7sylT z*q4)*jyJPiC*nD#ZQ{CE{Sywv!OE1DM>iy0CdLF5B$&rvMO4erl1TMQU+EyqDTp%8 zYmaz8SInlf%c;wuwIa@f+kgX+(GgcueWIB_O2C+~-SJ=Xk!ES>iCrlUFG51J+ba=1 zWt$$Hyk4Q9?{vP}WpaFpydpinao<~B<$zbOHk?eqy^niXUA-x>X59Qn-mA1F+6JHQ ztm1GJ5)#IS^IXh(Wa)Y#wOM+a)9cgSB=im>;+iM6UDaWdd2TLa^>qm;GBu7?M!7y^ z>-y+gI*k!Obtd)Y*40`X5R800tOct!|>x?PU zYb}d1<{Eq}dT!qmc(vSsupP@o6(2Y|Z|k6rQTBtoY%Pb5n5cnrlemnen3x#Z@A=#D zgz=h5-{HRIxQt4=L(~bM`CDi5V0B&n{589xTPxv?Oz0r6?QyoO(pT6r%C{&G0$bf%G;Q32>4Eq|LDfA?^s!^(d8TphoLx}p>|Mf z_;XzMKAEa}IMc~&ExJ}{1&@U2B{lK9{Ra-*J^ud6ebKCPlWMTe%`;WCS)a|F1!O|%+?09fpdCmeT4!_0f!CPG00W;lTJ9WCvzbi`J? z{F|ddhjzJR_uI4%_=i9GnsP*rucmh>=VB8NT3K1Gw5$EAZXmY8T>IU|$!3R|Q9Wk$ zyf&QNW&Zl>l?8͉ICtlHX_E#$_evVA`b&;2vbvyW5OJALI?zy4AC@ws{ z>310&a)p|^=<_ZgYFPRlB1ld?}sR#?QlS+ZaMUU0%`YnUMfVC-<;8%Be=LeJ7DZ#o`mMG&4D9vywh{?WZ?U#TV|1aa z@}WbAe!MStQBt=6Kyh!;AvFPxhqigVDK<<%HM*Ou7T#Y3R>R?@^Sf-})3rV@%XQl| zcjiM6(v1{GHp<8da<0r>@h~&9cf^AnnyPH$y}HN+an%<>ykFAA(Q6vNsQTq5Iy{?eBiqY&X&xd`LsXlf87~sr?DIU3zsp#Dd{9WP3g=j4IJ{Se*Xx_6qOE zZVGY)m40OwIN}-oB4b$f0})%^2OoiJ$EnWkN$2;6yIjlBbL>f5e(uBk`2J$;!pcgW zy1(&*EYaCK5{`QV_n)$D>vF!^d_f^G)*Y&rvlqHf6m!G9O z90hPk=J(dZz9w%A1VF%^>SPDc%?H}b;jyd$GnQDSc(BQj%~c%h);ar@ha8FWgSoi* zqQv7`S2Z%co^uW<7Y5xgd~dlcmAiROu)p(#3j^(5?9=Avxd^CSr@<;2Pkm?mYQ)S1 z!9U^N+CqGy@H~Z~V?9066{d&YvCTr@@hU7*oybz!fmT}c1GjS<-p^CaK3}5i2@fsr zGW4Fm@}D`l{h070{Iq@jEg3F5ODs;OSo7qq1;M8ssuIp;skw|4jsdr1AmAUuIbh>( zWW#PA&X~7lJ%Hh)->(bT8dN9Q3#d{~9bww0`HdA3{o=)ob?5))9yN>kVu=lHMBvYU zcHpsPv#;$_q`sWq(vxnDuhLg4!=$6I?8X&laXB(6J-1FQr8A{eah6A-JBWOryQ-Wp z48UXnUn2@Qk>&);jE9S{{Qh~jaq9R$7M|Ptj#q-R>-8IO^P9-`i<91LYyA?UaeH&! zKg+H2|D>2{j+?f?Pp3%^`cIA-PXGorQ^dG1*y$@4Te^DtmDw31ZQ=fzdsCIkO`MuD zOYhiS`A zU_OPZE|aE_8EeZ7-1ilP>nmyu(hG@BqpLFMan@xYmV zIMPcja*KgIPZvHlERR0Dbkmu`uo1s$Gk!M65BRgpuKrdbPVa(LqI53yKC0z8T3M1h z(ND*t%qnlIlyR<)%EOJqfHZjsW4%~$KDQ9$6*zF701OhGf?zd&QyeIIRMh8P%YyLH zK^K0-<5-4&zU7U6VP2JBINA}Hhgr_k2lq7gWoIHEL`bKs?Pr9Jx<<7~E%O?c2+J9TqO~ z1X6HtpXA^tG*z+Sc zz(}F6gy+wm3H5~FDXddlXXyoqI$jvB9;|kdpxX(0srYQ(0Jq8Ej*N^{+Bf^g_vOkL z##z*~1AUg6Cw3gv*4D1mp!@TC7bPiaeZ%P`r`wKr<|S`yPCvWhv zdg|x$@afzENZmp?37+A>_%(vTN?ADFhT%`2uBC(wP6EHy)qp_xiBJ}{N08H$>+3x^ z>_3?Su;ybHTecP_-B?}A@cx2e^6Z+H_k-rL^0^c;9VbU?^2k9ci^sDo%oEMli6~wO zF@fVl5OKdePhqpMpS=;_=z+Ru`U2&Ti<6WR55q70HDjg`4w3zMe~A^KCS|vKPo6vp zgh>(Z<}w*^di8nE-IqmE$9jWqlb4D3q=Q52AQGNr={ykS-x%&PZbGQpy0cO?W4Ffl zYUj}oGBc|$M>GsOE8_)J$#No2;S@+dM*c1P?b%~QuQ-E-Pj_2Odmg)d$j{HeL&!(A z`^^k1Y;7=7$`2em&r5D!;?r#U-d-k!?8TIv1_1J+bUekEGCRshdXY(ctY3wT-@L=3 zW`!aM$_D#b?tO`z&@}qT zf6@1EUqmb&~(Jp_uYj%c@ ztd(W8eRHFr;xNb0Wny?LwrJkD#8U1 zX(qqp{&!5!oZS7Ew_BEcln>mbEbRYAmNbA(Gj=y)+RWv5&eeYmA{tZx?-CJusC&BO zU*IxL;SA!xURf52gG_X3TsBf*8DpPwyKXB^3?(7!5KtvIBbtS`*AIk6FsqQWsl?#% zE$MZ)6_NGIA{(?!TdeWq)vH&`x?D5_g>h_ESWNNJ`0oYEeQo-aa$7Ha-^0!JmXQEh zFq6dv z5SXQRF)jb`Za1H^?|vPw=bwClG(UK=0$>mYocPX!HcB3TT+M^41tRc)O)o-TW>QRL z_}AZ;Ewh3*Y3_bMFR$KBQI7#0N)AtJc{I1JkhA{e=g{eeBx3^suXAHxH1u;M0PcO9 zM_Ro$nW+D|kbFr8J~F^H^mW~JStKS_UtRf(aQ-E3*I)%@TLLayBCH!tj133z{(`Hc zMzP;^ZQzc@f~&Lj1p8PXPtq zn2yIMmLR)`t z8IOe0GBDbi^MG8KcR827;zqqi%CHe}Nn;IyiW%Ux_C6rqUq}YuDP4?k=i}I$!R*h( z_$XhxzkQR}a*$R882jpONKDz=esLTtfk^C+QPHylw9^HkQ0j&d1ufxq8|gC%<|SOj zqu(4;_ zL%~Fq#Uq`MLcE4sk9}=X@PtE*jTHi06S*uWC#PTCXdwkd2sBXk75U5v(3+g{W90MZQ-?7^hqPMxDRX}l`ABzh68|V@7C(Tp zM@1XfPcc445x(dtU_3m*9z#Pz!Z+9+Dkvo+$ob0Kiu*;Gl)P+3C{8%v=5b&B zsBrRy{xj~1%kb8k9`x1+ZXxcaj#hQi!5I6dkvgK%qouF!KV$CZKgiD7BeMwYEFN3| zWW~^+*sCUhtvt9tsnZ2ditxIKs>5+EqjBFmDh8#joO|tQD=Cz7+}9d{EtDPqUKj;; z7dzC}UX$v=B#4OAdjNp79p?u0pT{9gY>!m7DKqp$5v&2IKZ|H(+CLm92f%|-*F1br zr)Q^yFus5Ivzzlh;Hda7&9H=vhF+IsGUAY{Rw5J=0<^^L8X#bZbc!sRMNY2|Oi=?8 z3SH3I6Bbo9$VeIhcC?Sxq)(cTfFsNT9vO!1OW-2flV?paor+gsU^Lykg}_L z%{Zk^V2zwnj&y}+#_leDDWmL9HzZOXXz1Asgr`nisJ_q^mhHDyzAJf94`1yLbcL@G z?e@zrKrRo^6kZDo<)Zr2+ML}Q569b$6|+)CJG}viB$k;)A5hmhm&PV;(vR%5yY}r; zIqMcazk0l97Gk3pqOquMqRl=aE<(#bh$ivf?|7r&lb)*iEBM=26Xm?ZwsHe#I1Ny? zxG@$tk+{sJ5*kF_0cQplBa_v@QJ?YJlZir_hB6wlxHtQx>XTV1f23ZY&Gy(ai z^D3YN1XXPLnGn=D8$tMF5I?yPl#Wzh1(F?d9%3UmbaoK4z0}y}F55p*wJZ1*zoG&k z(&gq(m|iA7u*Z^|xQ(x*H)LIc=Q(`jNDvHTXP&&nK>LZe709X_J52OVKKJ!~QHY=G zWM*cD$g=mrm4T-EaEY3V=x8-(n=ZJFU)>|3L!Ld`)12$88}W{Bq`^2)TsHwsb?o1# zH6J>S*}COza`bZRQZ8JgFNnptFf5~Ib|>#M>FrO`cpe(6 zV`LGU^N8xgU%OT;*`@91E1FSb*3la>)T}sZA6t6jM%snW!PLodO6?eG`{r_4Nt2iH zO&SUoKnjzUJ#$yc82}|^QpErqH$bY@U*ibldZ9kNAt=|_rjF{aRrB+qAVEBRNbXFW z2v)Q&RzsXqudAiy_4e}2X24Yr7vbkbpx>aO#?VX9Z}t znE=u82UU7#kx<2XWtW{~-&fx6kXG=`Jxem%OG*&VO+a;rLjh<->(+v_L20p`!s{Jri9< zD{3TBJ>V&xsrMTogZ}H6RKA~A6|o2wiokB$n<)}bd1#)tMe=>hPm#i90g4}lpGNh6 zJwAmD8H`y0sHvp*a>3M8#gw<@=`vd0_$`@3dTicKdPFSiSNBni|B_igCbD9__No|s^8P4@o7^EV~ z&&y1^y?Wp|>W;>O1~M7W_pf(v?4)e=hW^;QTkiu_5HkQ)&_mGAZnc`?xHZ)zg6iV( zf#%GgQ$a{6F}Qa8Z>wBcK^*4{s_{|T4OU?7XIFQP(B{(3XOAI!?{m#duDdQmH8`iI zdwf9{Fr2n&PvRFBKO7XCOGThrTWRni%U#BLGIdW~g;Vr{p&N}2_1J?H)NY8o#xG&f z*W<8j=gyry=OCY;svVJvSnuaAUsCc7FK~VGd|PqYPz@H{XVvyAqUYA)MQgK!?Iuh9 z{&k@w(7R)wcTKjce8)XD?#czi^8m<_B#$>SnmRek+x2h|Fs_OG_MbCX+$%3Hhbsf) zFFYJRITFs!z+V=93;&A6dh^$ezAXKTf-1wn74~37tiW+Kg4)l*_e{E`pZ``Q>@+&C z&tJhQ62%Ndq%JCXep?HXQvkv|RrVBTfWeCw*Mq|Je?o?$WPiq!r;9mlG7`(cc)ZHD z!vV4ch4d7NAQdznpOD87;GkBqi%Dxq#4=A93=NWV~s-dJ0aYOvA z$kELf$^Gh`h*j(i+Rm*5rS-gS$rO92#ByQnv5b8nM<`z+ZWYBI7o*AEED3+gxG&F< z{zR-DRfC=5i6hDShg>9om{a0SgM`blGC3N=*4e*r6k4I6gin0?o#3JE6Ur#6Hg~x%#u0i9{XO-i@d2S&#_+{$E2y&hHqI2aR zLbamyt_VERfpiwSnLXip$(Gn4Pr12S4{+WUXMpwzy#GnKrn?ks&kyV1 zThjE^bKz)+pN6A3c<7KjLPzvy-t_3?aHkHFw4i*Hs*;Opol5|feWRjPPN3N-t> zIHNcn`(|jCNa!6Ds=_qeM*@jU8G3+gK9FsY%^>g6KQ@N6_rk8*%e@fIrC|#Kd&zl= z<}6B54#YYm7+40n)qr!cdNXZi^oi90o05=(#}C!rutHtw2?&hMDKJhWEMX>4`1MI0 zDsy8Bq*?H!JJ<6|Z4h;*- z0e@Xk*UYom0tQ3uKp%+PMd^S-ce)&#<%i$SbG`!(@&5W1xOcxfvNE$;5(p6Xp;F>s zJpl6?sT9iaM*?)>GQIH4i*tVD^y10i3m_Ya+guHKlnbO9wL|R~%PGe88`YnIfZ-1se+#_<-jlXy+_?me#qn#me`H zIyz-RS3LohO~r9>05R+?U%CcZEvh6eU4Kkz>PI{7;_Y~d{FJS~DMb{|l5~K5%B5;7 z8LB%3T#x8q#COoXQr5)t@#^fQ-0kzWZbT3kKOl+UJEn!)T61NJ4>_R6d{Ntu+FVVgOj=(Qpg<>0 zrQMq9Na|BQgoLL&l*-#8MHPv2p~#33SZsUW^G~dYIq{a1|Bjka^wx z`0-=i5De0r{4MZ_(bNJu?f|3oD}J3=1PkXr@1%*5wzM8%nu+`e5qnTeYl?tiG3-;S z9_-BrF3@Z9xw`4~L`xF!v~uf(dqPX`J#U)9I%Oqc9dwYA5K)7Y-{Iw5CE@zFxAcA@ zigoH0yVHp1GeK>Gs3}XNFO7)JIOJ``Vx#R*YBs0?WKqF^rM!tpA|jjWGerIJN+hkv zxe>ap0WJwPoC~w|?W{`l{10yLCf2$V)=niQDqe_2Sb(A@zJ}6V&$X0!zB#eKdwNGT za$TzuU0Xvi6C`$-K3a^RO{5-0&uayAvPBh888H8X5;!=7OyWRNQEbbps&!8U< zi!INTOT>jrWgwnAP}KT_oD*KTPmy7V0U+K?Aa7u77tZ)`DOH0rAtOA+rw3Y|svdDw zHrlkZdF@@u^1xZRp+dOh6%icK%fLgMCIiXD!*x)EX`m>TJpoj18uLXRE2kTYb@up! zNF}8}LI)*Fj)%BhFXY(s;L8R10>7x3hFnwM*ai?dHvoTo- zardYlQpknx9;!U)01*!{{3t3@J>lSi1D`*hX70lWXr4jt1HNJnQS?NYfDhT+h3L(< zvIe9$RXLHa*cT(b1u^EI4RaLeM*+jQw;_Y__La^tt>m*4DzlV0Z-MW?4S*rNw=5jj%=mfsCm0qj!qbzxEwphDi@)$5 zE3A`{A2Ap+SK9pnxYy11^07APZ0GS2LlAelbQ8`yn1ZLb3W*oGHxFA~htS#w?xPW5 zJddg~;MrZiz9Qx=;LppH>=Ru~%)qj>w39*>*r1(1a0)^0kGGb2BI9uCJcqnH7c8*P zpO=3&bl)ePK?NachUdN*OsCP0F=} zxVY?3ID!r}D@G0pfL) z-tzkQ#)LbG5z7UXAZbz@8j3_F=yuUlxIpaROCL~Y!^o}T-=u(979n8~iuT`joyP4f zWz^^#spw;c&tJ=-%1%274*}I88$AGChPs75WO_Yu0t1y+d9+c)_8cHr%Tj;_$)DDU zkGC4|ReT|j1r!kz--8t!;UrX0&^YG^JsHEK-;yqW#&~kp&}!8B@6C~^To0&DxdYT- zZU(tNvP4QgKN<6o(h872M}Vz#mbW4Eox7#`&Fp2;qR2?5eBB$*&y{b22`<-+6vuJd z#19}XxVA9HkE)4;p8_gZNSkVkgUT_55bK3~xwcU7Dck}4SJ3S*Cx9Ii zi2g|HAZ~85hkx=exUJ`ZKc@xKJp8iOHSaU>YcfKHwx*gc*4OeSC~%(4D&lWQ*l?fx z9m{afTVDNR5Ufckm`h}B)PclJc5LPc{aQw@{sZASc zwF=gqDq--`ktwaERCkACi+i1nqCj))<$0=rYrh^MFB8Gej0?M@UJUCC1;pfV<7%Ac zQA-qqazX|Pg^y*Xa9W41P1(+Pa+djXA+u19zH0b8!8HA`IzGe=Rjz4Njx?!=re3?_ z;w8S655c_E+XDB$+js+^UL4e;3P@OYHcs*J+o%^s>YX9`xj6Z_@h^$b00N;YBi}e& zsM{qc1K0#ykuDk&d;vPG!G@OQ8Q4~bsmaG_0a zRGg?+EC|z%`=PKA5Qf5rsNS^n#)I7xKrVUU?MCYc4(wiWBxwP~)Qbb#vCe_|sm`r{ zvRrcixyZvwg|i8iADz9XJb!;SmC>QJQLj6Un^@65!&qFw;J^z%d7T`spEe5Oc$Gx* zcO-y~_cd~lUtOVIuEbpWJIRgIUcN`n9Q0V9;pxzuAxMgfwH%V0z{cPqpMGG@$SE2-?go*}w zOx=X3m;aTnM1nIVf<*9Y1VJiv?vWc}6CHI%a6?vAcI@(;y*^BRJ~p7%)fLp%eL9dF zE`+p(aF7D3-|tc>exabM3fCGy1aafowS{yJ^VRVMzbM43ysw!Uh+D9Bcn{n~E{LbP z0U{y^|EwonC(DfUmiNXnRJiwn_;P0xNW;NWe%K7zW-gY^kKGC^L?Yk+RIwo+;qyFw z3hEC89F0g(D1|do8P=;*80&ry`d*u=2FJ7>&AGI5Bm78R{Xy}bWfK)b zMQ-W5Zm2WG0Sh^is1pYPE13DwDqJieiDN0iz;(AS$aAE9N5KwfqmsjA<7no6y1S;r z+8GzJJXVstk%}HE12N$OoTn;YTtc?{n#=&?NbgfOpy$wWpYpo@lsD2Dp?s6WMXE@l zWM32f$mBzycLhcvL4|ZPNk+4(EQpH57RHQ z^yZog5DRXgYNZyulslWu0CIrx7k#=QY3vbCxNQ6%y`Ehb^!lQS*zN~F-ja?(-E{-V zi;B)z<$$2p&^fpB+~^M%6_P_>Tge{lz2G;v?{i*%!@Y{|J9hsWl5O>QCtnbFUydX; zjBsz5BuPqHjR>&H0YsPuEuh`4MA1)7Ass2DKd_Q4hZ<8*BuGtbrmF+Ckjl0LkSZ;m zk|AIdsBEz_B)9JIp|nl?=;@&URN0;aZ=hLgWa7Sd-US|=JCIR>r(*QS;Vif?a9S~ zH5V=D3x*UOjQZ=>PDKYM{gWN$6()#w|2*OQ*V1E~`FVJZckExfqv)r7I}1ajv-gT7 zR>xJvUcP@{+p|%<*rrre{D~QS#8%eE5~bfsoeK2s5X;WhJEaZ&ek?zmvp<6N$owBlM{% zCMIp-Ze>aKRto*k)f%DeSK$)lfgJH{urc|FfM{X6KzzjgMo=-Th*?}xqh|KP+go9lb(j;Gy*o_-i4GS4HpF~9C__jb-tt#q3?_|O7b~< z9^$$WNeN<)8Z(zo6{iBvlzW%w$~{x=qma7);RQZimmFffw+|ZuwfKM^-3R* zVvjX&Z)s9K56Rnf>>KZEC0L(cBC2@@n3z7AbSabs%wD@5(?jz2X+ENVWd-c`(Y3Yr-!`uQ{sprz|6Px%F1G*2kHfKfK$!+Tb!ljBI)t;W z#C>A>inu`~;n$G5^g^Dq8FdtbH45kgXhP~NW+TeFuvv zdu~}YP`tkDJ}tkHv5jMV>MwP?Kx`$PO@y|UQ&3#&fx3U>{*mcQ8=#WGWB4Z8QYS+8 z*ZD!86Axv(<%jvZ*<}YONH>~xbEeX~;95Agi@utv&Q7ErW|Ityycpj^@)u&ZkOcFI ztY&S6oD1)6Q-@4wFQdSlV_~U zeOz*fS;dxtm$u0v>#OCm_*e-Kr=E8VPMYu&*WklH^IexFYWqR}LL?H`02oax(ZA(Wj#6_DOuv&mUh<|B51!}aj+K>j#G*Ka;+>X&Drz$1LqIQzk? z_su>CgRZa6+f5S!y2qoEwkazsQ=Qo$OQi}02AivS-qL1AsZ3#zB{@f1#$d+9m(ZeO zWi&LSPaTjpExp5*1%Bog=oEoe+Xw|B4j1I$mWiQyT?EPn%wtW-^$Xv0LmsEAF!=XE zRiT?evlXYkSjToVFV#H@l&1Y;jK;uhlPbj6m5e` zJY!;FJ3K~04FG*ktY}9B$QRYTUs&4A#s~NsW_c{$TMkwtC^FL0GfzT9Bn#ReuBq_V zio6M)i@A_%;pJ_psP)bctdxVb%INtE72!3!EMMgF__YpE&4_^Y3UJEStu_!NPE_|1|9VgpqBqT= ze^$}}y9o>BE*4QI@c~B?x}1BSsUxY9aT0d_d4Yhk$D)tO5h_Ehq9yG7q33eS{fU+M zm|1pq-j+U<<6CfVvRpdUyOL*$p0TxO)uSF@yJI7(RG3-1Cnt@%*9x0P7nhem+<9Y3 zhm|<1bTRWU1{F(_uA=)w&)?YOt#|3EITv#oGYv>Lfo1?%30EvjboTPiXT$Tpq}0aD zmH4rzN^XNF`3DCD32-Ex$A+BbO0bx->Vy3*obk`)`SErE>Dq|Q$CJW&x>+88&} z(gRsQ{JSBv+jWVDSFSyYl@HRDJ}NBAs5^FQyXMMEu%f3{ zb$3O=+oJ`q4)6=f|EmR%vcpPKN@0+-U=FNug^f9@lki&x&PI{lK ztEZ=@c{L7I1IXx2uOd;1oqEBOVY-RZ%cM>B5Q+>A4GrC%=K`tFl8@AEan*0x53jLR z(>N?O@yO_|*43^0+N{w%J%7rbe%I`f(QLc9G>LS>kVCn% z$*iEyyQ)9qq90?|c)yg|4qU<^C9FVMWMBofFlE9etbpbpuVp4?mz9|*X7|> zgxvhe-E5i`ux9F1+)X$!Ud1N%g^C#L3^2TXC^p%Y$FN%LfBc;8k?yq>;)UI)2-|$a z-)sH*_gdX6oxXo|?SX)$`gzIp0At3&^Vgg||5jTsr7^tLO=08|D&2DZ7g*nyGhXJx z6SzeIB44CnajMDcKd6giuAp`ddG%MRw{JHe+W#J=|o%WkV@^_PC9iu^b8r zy!8-$J}C^Jsg#G}7K;^-UMA-tg9?R>E3%`-A1NpF)LD436M9QH3b#V%@hZq5#N(Sb zq%S}8Kp3guykHcRlwa~H4KhwRApNv{x6PN?)!+-3qt3_SZZauufEcxv+crHvnepc> z=gK8MBkij)EJ2-CU;8&m&*HM^5kXTgaa!?moF-3Kds$RsdjiW^f9iAZ0lmMQT@Aw9 z7=;>Lq%q~?R6OAHqA|_*{gpYoHfF2@r2Fu$TQji+kzj1ty7CS(oXo5&E2MIQWd}DZ z4%J@Pz2L-3;Gep1?`Y#vP#_6oBc$^_J_f2ZC@XxP(%=u{!sn10OxJn51SAEK=3L@+ zUCTbOB4no4R!F;~W?zEywJQv3qq?Q%13Lr7Vbj>3-zZFNKQ_SBwb#M-lJ7H@}Z&?N=;(#QTO0boPgM|_3s(#dfUdpL;+kmx?Jdt*todNYH z)!gnaCqir)I*HlTw>JSr_!5Dcte_v;^~$s?5)MZ|xVm??b}|@QiJQ>sHbqBAFWNAN z74*ZxKRaKJ!Gr5~IR7?wa3URtdn-XQ?aY~>boz%Knd-X82v5wej*N^P>!_J9`#OM( zk2lJ0n%$SFW%v#H4%c(9;3^r!f)c)98Mgz zijlfkKTm3TzlR9$DVZi;Az2-kCGbo^|^XRtPf@?3q2Y)9xT6j zLh>@RCdXYQ?A*oLdZ-^N54gr=$PSx{j(aj<1IVx#Ip$!>U~)VvlMr*|r=7-|Jeab{ z2t|p{`FdD;k{BC6W*!gu54AuVVp$1|BfLbxn(5W>5uDh2DKgb|C#j8B!l_jm5*zVe z5mtc6)DfFc8X`7_M|%MW)oC2jtbBn zG8HIr^)+}gp(V;2u|WA9ln~$*I9zHtc=YJer)vZI0QRU=(h)E@FAp6b+m0s#>M{Q> zsk$smXfT5OU!vPw`l8~%^5_OtO57{ zQmvEJi=BZ;*b9}^qfgJVQ;hpaNFHFq+;zSu3vZ#AN_`q%2asX3iRvZ2{^#2NUI*pJ zl8*o+r{=p`zhp)lTqsH8vaxoK@Rrt`2V1Az`Ex zB0oC=ipskvJtcm^lc!K&LJS~u650!Xq)uN4c={`!~cBo}C>UCMxaTmx{ue<)W2qISwX=85cpVgV%k2`4Nd3z%Nck3hj!c-5#EvR2F@vl~9gy&8 zF)jZqV`J9ZkmqQnINEa!K+pr_^R~Xe`~w9Yf+!Wwh@EYcCq@;Y7?s#zzvj#?*!ZEiiKn~(PkoF#^ zRAO2G#zclH>e33V4Fm5D)=RTtM_PZbZA#R7s=oNsA$U41;PI3L_fQYjybC1{Tf4ZX zDq+W;YUjb+HxmCFY1n!lwOgEhR?9D2M<>l;8P?i2mWZvpIS(x*tJjt(-n)OFG~BKg z;4Mw&7qapXOs$-9M@fw@Q8S{rW~^vNrGK_=I>s9 z-~)Tqjq1TzFDEXVa5hk&xw~*$MEH+KYV%h2_V$wW{?q9J{w}+8`C0Yo`ARb~VF^H* zgAgAV1!SCrUL6jX4pW5avmtx@JZ_ z<)(9pGejodVpGmXGYWR9M7qd9slm*{(3aK?2@jW`hj>gqB-rV#JWBPn40BeCXx7xVDy7nRJxhWV6de@IR+nrP@=!cYi?Y zz~ZFDf#SH7O-S81T+&r*#hL439lOs=eVlrBZIu<|2f2(s8Y4k6Cr4c-!?DSBF;gqO z4Bo=^QXHyr%#;S7Q?2Pp4rAt9>dj`+gQ*Y2PegqxISeYBK@`;pmpZIsF7}tS2+u zg{t3%Y1bFisqv)9bGhg!ur`K;VNwBd|`^6=wX@UJh zDf-($oPj5@IaKwgavMV@P-_Z3vXil*iSiE6AOP;}mESnbH zt_vD-Wg8G^A`P>8QF=N*LJ;1rL9- z6%_nV)t~dH!s0~@LsE!Bv`AfCQ!7mL+FrejtEbLScsZK%tx)j^f{TthOM<1yEb%Cv z8A^Q6hG7dF8gEMFegYbA+?L^hmWbvk|8Y}@DKr`kSX(iN2A7nY&@=n+>&X#1?+N5x zQ*&HRY{a?C)4u;t@(q$OvV_d&^#45P_j}aQ3YeG;b=V!~h*R@+EA-x_(l9C4YK)8^kEDClZUGcV&H2!A z-KQ=q{?XEON$6}(?5;dXi6;8eNX}-mPdM0z`bcaJ9GoHm z^Zz9AgFo6>s7c5YK@2ip(uh-mhT83@y_34??;h%`EC^C&w9OEC*d1y6B`!AB2%862{XR+k6Ngqw(C<;9!-U!u>6BiQ`qh3`jsBx+9nPk`C z=TJ)bh1}RY>hJ9UB{TsGIv>;0h*gO1R69qkq)ATr03&FNNMwM#bWpYnNw5??xe^i| z5T6q^ZWg}24m36ZsAlPg6FaC47EhwtF{oHEIxIL1HitM!0x$6(k1f%{&B z5fDMo9jW1|e+mU2Xhq+sx}7#L$-2H+gnC9vhynPZx=MR=jP_T{*!MohXBoyxozp&g zl$oJ~IR!YhmF{dP;6AyGrMom>AW06{80kF(l9z0Vpyqchv+oJ?8V$pR`Z5q6TCzTv zsDg-kXlWC50aYvxbvOfJ7!7Xci3Bj~U>zI#YPn=1vGFW zHZEBVRVr0<^GBu2(&__qZUEXiu|(|XJ@Djm@UI$72pLA>*+b+2jiBM}=5DUNh+ew? zvy@*@kncmo*GB%~PrIonjNzgqj{Qk#X%bZ4q|c5tl(qm={dw;42c%O&M;h4yh+%hx zez4}j;18)M)k8wx541$kt5@d|+ro~OD}QW?cE~I`;BiTfVl)uupq^feT6jNn z7&+8ewgp7OFamKia_5B{F8U)IsHvG0vohqsyQvnvxxWUc3XslQ z_s+A;@MHdfbjHY)Xs!~SP3lr$>Y?qYum|6dH8wVqlo0x>%2mxKW@b55OUGc3gzh{b z1x$a4dtWERkSWxh2#L3#&)IEMi$(;r-s3lL8-kKCY+?>MR>v-fnk?WAl1`a~CrcB~ zQb1AB7~Jy;XWfA2R&^c=knV_$11N zs%_RFSuM_iYuQ_%2Vxp%A`S$@oqeKJ8txVViLxEhg;bR!Ri83fR5xKGjrSn{+c<-N zNC{r5Ou)X>7Po7~TN+Y_X(+a!tt4C`beZmt7JcMNG^+uTSV=0TbwO2cV!M?8unN`C zPoxW7*Gs-YV|B8A74*47x_xq|Db&{>zXa=O?FIZu@=$1cC2a1oGoVLjWVzHxMr{aN z0QROZ6C)&~KnMM;9?p$s!;_?g=+Ie-jiC`YR0pAh3f#O^E*9z$YW4twOC-c{m;tGp z1qGi9e)Hh5&=_B9?(#j;Eg{LW=K@|MU^DFu`3I`V`ca%@GF{BHK`@ph1VOT!Bv2lC zB#m1UlMWx21uu!yOijIn!@QLxlOAeO~wD;v8?h>pCeN1*5J+!ITb=h2s>jz#nUzP=Aw;0CKN2+hBP?tIcSvX)ZN(aPp(!aT`6~oC*lrpb zLXvrW6Jpwjl;r&J{Nh{lsncEbXtWRzUf%aj+Zi5+OZA1XY^L5s%n^yjaN4zH z2j`*_{8l_P)|9CsL0olpg^eb}p^S%Gxm6AnJoV^YA6qw|NbUutN;)?*xe{TJnTAK7 zn&lx@(fA7MKpp&+7(o>T>d4>dfEzZ+!tVXD-YYUmt2oq_O}$byLIK@`+CxVLwWVvQ zPy@+V4RUdtK5FB~3_=>ML#Gy!zWZ?Mppj05OP!D1=PlhxGf^_{i$sAB-2y!+6}>26 zAwp8ae0UxaOLJAqovgDi4Z^f)8oev1(-p>TW~EOGnlItboBT(Lf-PHvJex>(hh1hm zB*sSv^8`F~BC||xqwAGCY+yhKXni!I+(qG)U`|IiNFu}bto6YienCIihjC^@Z& zV{dKJV_o>G%9UZ*L1AHL;cK~OTVq4o+b?MM#ks5qU!K}KteyN&ZTCDd<^*YAd*x&k zz6polp&{6cFc!*bYUH8GP+_q&(}>r*%d8*E*G$U|$>i%QdzX#R7WLx9$W-pmJPvA^80i|={S9lf~uY9OSrz`uW=m=&~zTJ zI6ADwRV>(M=ym6T5(b zp_l3c@SKb?*5FH$&&<~vXf!Doh#5v5@3 zptEmZ&fKmpjM$shv{hyy9>>~`U7RtuG{grsuQV8QKCB>&-=fCTEhZ|Ax@L*9 zB`$y_5X@YDRtcHW;UX1iGsuFznoPjGEZU@IOt2QoEoNV(zEi}BMW*4S#@~4v`_!3b zp-bQe-1ocaA#@p8x!tb#BKjOV&B%^{RpT5?UB;Wx(0^v4)qmnN$m5M0HT#}i#)B58 z@B!S{p1jsxL|&|@DEj^2!acELlc(CUgQyc85)9e-RUA;ofS2{nEIA*4*V01{y;1on zswAwRZ=HntG1F#4-SISr%`>k9%Rt?MT;AE)DldCE-8ZfO~8^c7ne z@wky?bAR-{@=vD8;UEMflHT5~mOwKu4en~fP}CfBKW5$Y^facKAV8w>231vC`;mEI z`r5EEiC#(R?(ltr7|gHluTmQY#%SbAD_cf`;PEghEjyVgZ)B^yk(w#c2|$n@ZER76 zUEW~=G{8mobvSyls00V!%G)-_4UOOwathTHb?x*tY!6bQA8bYxICIt1;*_zfFzqEt zqbESD4$}}|?-sP)n|aaRTw+E&ij@ym-NoV{(2BWJF#&Yvses?5bqNKSmwTL?U~ zhm`jLo@6)|UCNaK*@5>OwYcMpc&!mw+n>B`w}^xUW*ay;>VN)PE|C>s-)F=lAs-{C z$UEG)Ht1Pcm>5<>zZbc;C{+4YU#%z+M(|w@M$Yy8-rcvCYuyXt*}D_ojktsdJjrPG z2BE&0ioTSs+vy=)WFQ~3Bjh!SpC`W~i+((6f;NW9t>6ZOtRx$!DZ(gzo5eMmI{^RC zNtXg0Y_>NK%kBB0ga#)dqsSyzO$Sw4i=qZ4>m3)|NTC8QwBla8e8HE6el$-6XtrZe zadMqq1?A=B|JlP(PWlsIpyf)?MYJtGC3^Z1g0skAFGY+agN4vV$BMHha8Z>zQ-b)AYDp?3FQYgMt{SG-e=_F}r z0QJKl;g`r>Z-$ixcO!+TjUfdz=86eW{QS@S7sW!DJ_%Pzy-9cjT#U`|bDDs!;;*n3 zKsF1R!vz%!&;nlQWihJzaIK;e9)CnsJ61HyQz8!sCK4%fsD3KV*gz#ehj=)mxc59C zw3t;rNmM1Z1InUvr>{vSRW`%J;x(0r;xPl6fe1j8J*qAaR9b7K8B#Qm4M-Z?A|=ej zxW|hfUwer+p*;rc!x{~SkumdQ3DlS!nxLu^Y1*xziXMFM$in3S*a^VYqN1V(-!4aE zLc*XSg{W1ZLK>|2w0CnVxcG{yI6)f3h@P7!ns{t&5afmQY5H^LE0!thn6^ z&Y76g#HyrrnbhHZ$74HS>3uX05ymw9jxQx5N3;@)#uRXwo^YC;(VL;>Y~hw{c6J9Pd&)f4*NK>Cx5_egT@NXUvRMNtl($R2Qz*?`@jU~4pe5kyUu zpYivW&t&yhQ0_H=sn15My3?)JMO6cdFM@&$-TdfBFf+1`!zFGK!>GzTWs%fnQ*t2R zl_W*O%Bj_^cWwgSQAMH!1(*N46>`xLvSppphr0GqvPePzC_L7dz_(eH&YzKntBXM` zN}>7y-PU@vUefd=K!!pP_!x*pEC)d)gxF6L}8dibcGf0=k8EMCX*Dg3)G)qAWO_Achf6lf>dnMK);7|@`^iyhK1=gW6_8yiY2hWd! z>qt&0MrS7F_Xi&DGNcAb_*A?TQyT|Mf;>(u`qXD>lV{XA#I!6)U_3X|Gme~TD-E6s z5kC&8ACs1%L4>*kcn3W_J)0r($;G=>fA6Ys3D@I$lPhF5xSeMC!19yR^IR$_s+mHn z?*p0urOQb#xqvjt*|$cnVh zdAMm9Z1No%BAW{eB-TC@j4%Iq%Wf%&aO6SRC}A#|mPX7j7%(FY?4mS?ARe04h*`lh5Q9X}B04tK-0Bvs4|1+i zOH61xdV5;S;aSyAMN7Sd06EKqpxD32{f_*m8Dsx*pn+9?Z2=CGrpF>=#Tnqw4%U8Q zCOw_4eS_*KpA+3j1i*01ZpDYxuS3n(R5`Pwq_dBf!7W#Y;u_{&Q-R^apn4+9H&32= z1M*i zinSlb^MjZg*c*jJ5?)Zn1g<5+qo9wgES_NkBV8?6ArEqg=rf1vK0lB!uyO&K`$NZ` z&J%J<>cJ<*8suztm7%PryP4w$e#+*&W&g&1oEGV56jBLJvX27YUpfVv(a{li+K-Q$v^2|p; z5z`;>_|oV9Uwc;?71f!hFKWc4g-)VqR3sfMYCurq0)hgGQ7nw2h=L$)5fv4Pf(sy0 zt!>PR!B`3;xPXx|K{io9WD{c(ql85e5DIV$0 z)m^^j{hsZGCSWx--$T3hAkiWm^nu>&5q%0~#TmGk(_5jk_(;wZ71KIqt>}OM*j^i& zWp@9B8Bf#Lqjqe^Su~J950x!RauoV+2sy?&C4L|m13_d(5|ESmPqG7nQ+NST4dKzWCZ951u{(aIF^ET7bGREFzGdz1cy-(xN+6VF?)KkzYUnJEJw9<9`KETyc(9A-J@P~xJ2-ZBq zqXt;tAdl#c!|P`>wK58mnM?S;%MmIPZfXG2j+l14rv)+qIJ>qfe+PYff71zb!S*1> zr^2%aTFog(M`T!(0XXE-Ii&XT&2`}e&y;kRE_`D|5M{TqA&0d!l#QHyx5pia7M2pC zqmODajJ<4i3X9-4I!4>#NN4)ca*fwFmUoM9mE= zEtzbKiw}@4UQIrVc$3T*2f5{IL2Q4fIfmNA6hzxXC8F1EKiw*+T)KMOo9Eg#q~-U2 zt$A-#>cQ1Ro*%F_oV544VoThF@^h=BSFWCwcPBZeDamu>FJi2 zU7iuO9saLdpXBYyvJ&c);l~~$lF$7uqx#i^6YSMDON5IN3cetddQfkZlfx#jvcF+# zevBsa>u*XDl#TDzG(OrYl-Olc!xFY}KuZrHCREEZjxVO+4{;n|?J`)@-6%PRw}NGO zIU+X7n_giYA{QkrE)WW54w62>VglRo1d-s!nR$q#!zwP?i*y}N0iMFa%eDM~Flne3 zl@PA>=ax=dT)Y%BYDLV-6{$bm2}dSew8twG>$1O*MUKqfYH7Qx{IXk!QPj_@cIh#e zEAaz-V}Bl@ zwRA~8?m20(DA!aMHWgR}q&Ey7hP;&C6HV%?6Xi^)0w~~^`oW*yRKh?UY}%se_~e27 z<_*ZU7rB_C2JmT)_|b}k(!-AHoG}dN!i04n>bEeVlzBN~a3G=resth>E|#ut7Ptd@ z4YC-2!bf9}HiXlWO+E8i44sAvP4oLcg$!Ezbc?*38l2Z_#L}yFELfLB+(_ZL0W=5s zxRXkcu5!JAHgQ{iCGA<%fRLRFuP)6N()2Q+#ur>GAhtRIpSVoyh%m&lcp57S93&^o z#X1EJl@alS6z$SuT?CC_F3+(t@;E6T>Xnafp+g$*aa&Q)RZxvnn9jVV-# z_USQp#m`yQmLpJ3`8(<%Zqv&j9TB$JNl!ZiPdlW-)i?YrVwmIzlu@YLI-x9H!;g9O zee9{OvrFQJ`87k_Z6L~L80l(1dOY1w2abpxqzHardhy%FJ>Q-4(P$=bh=+EdFJ->d@T3dwp z8aaI^=u4yT>Qp=z3O#Z*Oe2#8Ris0`(m4rk8XoaFbEfP+j%s}qGvS4Z@+!0FBNB6*Xk#RWh@!gAmxp-yc9bg(ZMyM?HCdk*=>R# zkY0R}IyC!lZ;Ow?qftL69bh4#maLyk8m9qYViN+nlDeVg5pPZ&R%w0tYu7h~46fcS zfBMJ+vjMr5&VJpFfo+c5gWfXg7g-@ao>}1qH`VX;s|I8`H6f$bR!K&v^MqCB2`+lc zeLp$F;D!d}+KNlcx!x^3I8RK<%mUVk_;o|4w}q00;v~I#_xlCUvF}|sE}2R~5M}s? zIBmrhi$=xr<`-$5d?GD_!#gim~DM?n-JL`O|LjSp>!~1I##INUM^zqqsW`bbgj#zdvhUjS$}R5 z5sM=^8xHHILy!ogVZsR+&RL{i%^U9vK8vNvAA(4f(K+cJ>Xhx3sfjm(fw|gkR;PH# zm28%R5SHdkU-U1WDQ!r*4cj7 zkE`_kXKar?tb%Zz;DzixL?)1K(V1-`xSkRhZ_ginv&Ea4{S@bbRa?okV(HM|xmrk@ zB|W$$FYjK_pE>^~6fw)TMxb$={?5;ZG|RsK2o$FEf@vmtpn%x|^qg3A>>9{Bdzb8? z*4LzQ7us&Q#RIzsiO%$K+ki0AHl~z;2c2{%7fBz|rK*DW6;=?7vDdJ=E&D>NFthIq z3`aaa3t*+mZ?cd!Qu<+gRljO5KS>RVLfsl^J}%N9@w~Yr*$C;dEhoeVd)NVrJ#Y1k z)JS97co?uEECTWh(}#qpui){@Zi60OK52MGlat_M)psRK7gM_GT~8;3)6`q=)8=Dv4QQvLF|UQkZaplW>G?>OHZ zYNiQw)zY=PA{Kr}p+C56}>DwFP+Q;2c@9U`|`PHb=7NmPG zh017AQ~RhJ>5#A5`(qZnJ+Km_XO%MWFQdLaA8c-T`|ROZtS=$Edht^%5Mwn16`zU2 z#q{c);VtO*N1egtragEyzCcfujf0h-$?GmRjSGIx(t;al5Es)LB~V zrbrj!?Eh9q(8wt!$$8g@TKh{6Ic}!z-moca*Q9j&Wses0KamFLZqO596*XgT;3QfQ zT7lvQMEdCj-)s$- zk0jqfK#{ge#@7IN^QblulB5T9wQS#!uQGLQEbQqKxC}r==6~P`R-F|0K%sE7^bo4S zcAY`~p3K7l@5*nWu2cWWOliWl6I~K53^cGo=4S9Bt9?nqw=!lioZ8l3Q7XQ)_vt~h zZAv-%*cB$!=m210SEDBgW~0O(kWpe3oeBTZ_zIgur^(+ujEjj;Pn=nnCZ^yvJeW%1ayH_men|PshF}^Y2%U!u4kLEdW}} zw??35bBh@!IE#y!LWbZIp5~PFv|>-6(-?riUbjVjDX-5zefb=DGOsSB#3dvYK$ao3 zwFGZ^$9aq}_G><6p=V3WplX@wQv1?w9uSdJc0vZk*bXCUn}_h_`+Z-9)g|%XehfpM zZEA5k&6?X^&=*TWA#etsq1Xh3iIW$cCXZ`P0e9PC+XQ1&SmrFj?62aJ>RN^o;U4}K z!k&mZTShfu)YxiyVIKyWuJ)YS&)Qr3xh)rb%{X`FRcwM3o4FnC8oa^X(Zhslaf1`` zW<*$Zz-l);%0tg&wbRxJ_`ux~2LJI~Hql09uUTp%ZGy>fWe7ep@fv1K2k0YO>F9QX zKJt7?TTsr1v+()K<1uKMbz@{{YH?6yB?)zQP#kE&oC z4wEea9}@m&J5k#P(rhzl_V^C7c6&XBV6*cMrkG?awB*V|Lq|R(Xb4W|gbdFlAIC?- zRpW%J)8dUgPs7yO2QUWNT=#3cF}Av(9S%TEwCvm+!J2ELpuLMak)dl}e@Y3##)6D= zYVCM5aY%(JE4{*&(?=*-hiA%^w<{DpkX)EgTghR; zR^-_h-TLO;$V`WZ?%=Fp=3yz-4YD`yDnf^&?oTY5Z+*19u2c|4d?5eq6Nj08;0(PH zg4*2C10GiQb0Z zxHv*+QaQKmdF4^qnN+i8;@t=#X+ic-n=J^8>8!U4497}m1#*k}`&JyeUwqqh;|fYQ z3L+`oV0KeSW=jC?g;j}aO29{u8s@@6e3If(w<^854D!-EV^Y1$-Y;PK0Fr?~HKNi3 z+|6mQl}vwEZ;tx9XTTP_lD*f@dmYTLU)Zwio$kTeH@-VKGPn>K^sUi$0fu-~`S_M1 zb_PPp%-Ph*YJwV#!E9PeZsN%Lpj;5A`OPcm5UC&QCkFjmD zdo$}N6Ycae`KyV-;ZL)N>Td@TE?3f`m$R$A`W}o(KRQhg9rtlzsK?)*(eA$Q@w*WV zl|{wj1TjFOu)TmLHs;T-tLxIK6$jARA9ua3HrOlmRT*c0L%pNIbcJ@zTJ4J}X0>MQ~v2=l4Cc1$gWhNSaZ~fC_l- zIFoszv#g|~$=kb_Hwj-JGjeOZBQ&e&@F50+^Zg{_e|v*r2dl(cIq#hI0i9W((Pg(m z@?z=LP0fc8g(S;H_V%`kg7(CRhjdn= zdmT1vu`OM@%Njtygwy3aaS#sW033lr21J2eiDtcVud;du)WsT>Sk6Go|N=P-Rol{vAi#!Ry zFSZiA1Mxn!$=wUSuKV*Y%_CAO#Rqlpqv0;$pAX!!0vZPBv6lH&VaEzT6^#T%74%=7 z1+@rARC{Siyn`0}O{*gSeGz_VAYHxVfjif*bds|!TkX9#rhz~Uqgtl}GVBhmw^Ov+ z$8}h|nN_#C!cV*}=kX5@jduXW@FwI;DG9^xk#nvU=7A~<^-dm^?w2rg*RVNNI#%e6 zwm7f3b<3SFsPHp=t^j&bDNe#+ojn`90XzTh;pOj=j|}C2pHG?=VaYrfE-snZlG>yt zQ#{}y98cBd$b7^9X*c`d_q+f06V@XWalYpIMuZis^4j{H=9N$*IlSr03uKFA+py4o zww()r?8lfNp_J@$i$4})?XKxFp_;#JFjO$x$?`#nOvh1@-b{5y1mO2te0Du{88{S_ zGHf<^nmF3)R0X}ibXXj0-XE~V0A?~(M0-W=xeRT=AsBmOCHO9#~#>&_I6YbjWuf#IqCXOOgdY;Yf$&p zcDVFltcQexytNczqgn&7(KPpQw9rXSn@5XjNHz{Ct;9zuPoda>>Y7gY_Rb<_bgaDx zpz1rIWW)}c;OrJdiO!yK9F?DAkE(2kOxkdViO%kTaG0eF&{k(Bbr;}<;DX9zwm?Qp zKua#Y>K8lLMYvPL(HeIBG>;-Ut>hv1H|xP}h}noEzobWuqAMPp@p44pQhSryC2qQ$ zi$78-WPh|3K~}|PY}Odc8^nY*HffD_P1i|_*co6G{Cj0c!@jX~qbZ=Hc#;AUv?DVm z#LVor?dp`KZ!HU0fDyW>0b=vETYZVQdFxHrO-oJFN zB#no-4bb{I&snK_49GkPkJ$Y3yKOd%)pZZVn2L*c`48_Mmm02=F>~VJ*#ub*UPjXE z@2OBd4fWW5^=Y3XxvK3{c=!=;?Y;M(O@m7A2==ONiKTczaeI;YHa4jg4$`QShGp?B zzaBrEzVpm6)BOo-W}(+nf^qQ}K?*LFju6nlpxh%!Z5KWx+6k+~X~>|b$zMP-X#Q~j zon7Bx4(z~K&5`9m&^OMMpkO*kMIyAM~I*3-2DGdW3C*j?DQ-?D_nhbE)We<_#6wxYoM-Q0B(M_3~zai>X0d zC+)Gluk3p7J=@Q85$fP@ixv-C>5RpYabF%+^Cmc{tuo}vz8M|#l+gSTlzY`f9ZJMF z=XfgWmR;<^lb+KQh&DaHqv^d_-LEfpt{#s`L>|Ovo|wCWL|pPG<%wU|;CAaB)gAdA z{^-)R8S1n06b~2A%(hZ^bpqMnmLn09%ME8&Il+RBz&AAdUqFG<8Fc-1kc^%DK8A0L zOD9!DSX8jgGpIxiG;O7N8wT;l@7^u^+~#!9yhM}{9E3PwLr6!>n^a~n;rx`TS3bi* z@YAU!0aaMLDFdSpYkO7NOe{wzXefl20<`JtJXzZNi)q$z$kswFi=q9YXy z03b}f&ZF>~hyks0fa_~KG}u9&!s5N;^5&i<-V12fTuTjcf^fD|WYHNk^W>A!TEdFR z1LqHG30vm-FS%RD!3Z^sPHF^>M&1!?s~Ei(kljYz$?xomQ?|1g$@`!EW0QzQ61 zOEtf3w!HG76+*C+p;Mfb>J%C<)|1!x+?}s#)0r`U-t72g-WRqAiu~Jy)_{zvJB3CE z&<^a)cH;qYQ4>Aa2z~C+DjS0!4{~HOYWOU3Ii#Q%v5-@%2UXtn9A^~a^^Q+QVu^jE z>%XzlLI=? zd21kAECUTUYYdpN2fePw7A4Z;x3Ysh$?0m6w|_on>GoN@IL>-d35rB+q#m=Gir77d zG!NDb?YP9%500lu3*c7cNlV)QaaT@pbwD+|Fagn|fs@mY;UF9ZYk{$eSHsMH!bS4# zkz>s7Z@HaGVP|q$1TJ{u?ftq(oWWg3SD_HZncMAFu)Zn+T^QZ)r06@IH}9sU+H}T@ zv89{cW7;i89^L>ZR@Dwsxvsy)bTnQ-P%a|ElkGJKpfsz?3fGPS*r1sVMA?~M}t)wUX1>;Nw>!9%l%}3+pJ3JxMnsy&*nhr6Ek!*Hj8T0sd4iEQMPbr_S-SU z*BBX0IUe`u2ff`7_Tl;GfjQg>rOe53XWJKGwLSTgt8G$pXlc7|)UDY?+Xe^@+5{`y z%JwBSFhyuX;@)4a(9$V(T}{Z{u6=U>t2Oz>;`Dj40tyDJx;J)~og1U8@DBCLAW*nX zXrB{cPZ2U}wj@$cVzG}Pne)+U?Zsosz+!PDA zJ$nJHf6FLZC$9hwhKB1Y4T)?Wu=N29XmRmsyfVa3SjZX2Y*6~sJ%q9IQzn`F;dy9Z zwMP0O8~ji-7oF|9sz6k3!is0JFZQ4-HVkvZ`jgUuG>R!6N;DWT)*SWQ7zn zn^#@S8t*Z8PmQ@jjY!gBZIQPR$be{J?n6Hu69YTL0?VIeRrq})D4r3ljdX>i^U6I- zsI6YY&1XrGyQOXjO4&Qapn<@{6V`yxzx2z9V!w8ebn|z|Aaz`Y>JsbDiN?XNQoJ6V z1r*Yoi{3KV?WXQY&4Q9Av;vB*JM2FSBqubOL>2{-AGzMd1Hj(*MMMC=^bZlU`5PPT zuz?6pNvh5-khFanwweq?pHGepgx;;$(WdXN%nWMoWt^oH(+UIpGEN^d|NL7(QAEcA zN;uD5RyC;$1nXLo(vhx5(V7?{8I#POJC0dOKodVg&;=+1|7TxrHaMngp5X>SiH)3g zxKWp(`9~PFxl~ud`vb~}Wi_*Hu-`G=;pYBPj^3Nk>?*Z9Z)Meb{Mbgo&8jE-E&et4 zP8grgV4WKGM=V(cvi8GfG(psbp(bj+>d|+9a|~*x!X%Rc@MrPm ziC>?Afftq2a&rSV!+|~hG)IJ&zgn%6c0GkP$V=8Bg`Tv{(uN$NZFBcBboF0Q77g2J zS@W$l^a5x_$K;ffU0A!t_LKMTS7N6=#T}N_K#ax=nE#c%FCOa|ELat5MsL!wibZAXk; zJ#$wTiKFC`c~gjm0&eKh0`BV@OU)mnR^9pLeh^Gm$QBV0P+8oHijd+)E%in#mcc+`u?hrBRIE$c z-0D&l)LOAH5Rs&~f|Vjz%K$?G5p@^=Cy?z<5@s?Puhq8qJLkLS^be1RcX^)Q`|R)Y zPEtdI{f7>=7;Io*Ff=g0cZq?)0KmY&DC#{UeB_hgY(B;RV}veR=7<09?CcyA6x80{ zu2QMKoO;gz9~!hdVC7Z=1IuLfKSNsdG^oq#+6^4#<;9hWg^uF zFHQKtV#6of@0}N<1GQJLH{7lG&Na(qz6x9F>zi|1RqVDS3_>4=E8|G$wo$7R>%jV61@E< z8vN?C>@>g_|6oq;$I}QmNS_RCy(|0pKo4o_=|H_HVoClf;dbl|?(x`KY98y`@90pU zZFZ>A;||D}|HyzKnrwn(!o|kKpXDQv5DC|Y|J)1k z`GX}~ig%CqI9blR^)bOV8}4H86o^eTMbako{&$` zyl>(>Z#xidZSw$T&wk2Cl&xA}qgbDr_sRa^$Q2;tFlB|6cEyQj{rV33nRQxR9oWA3 zb3z<0YH?f{y`CVkdE_ea4r@v@h3{k#ICQsYJh>z^Lv6BYIvTSsNs4+`mFcjTElNQ1 zs5qh~Ymn-2Y)DFrLx0#!k=bEW+}l>|x%_^gE>i;4%9ZGZde~tCG@vuj|IvUx+Y_Hu z$BZ>PDKCme^NJr{)FJKIPebY$;SFM$6ah6(q5+G%krvmc2kamX*vE8WuPtaB0T(}} z3VAvzDSD}={skKK&nM4vo_3b91vSHw;+6;@DUTI&wqN7&*i~ckP$$$SOK|VpID@np zEBfKyzMwD#Kt7<@sn3G2;fR9$k|OpjI@CbgSzccdW`ZPA&*G|7>kpm~T;ErH0G6*{ANLSK^ZFbT@MM{(x+ zKW`IrEA@mlmL{Z&`)FP?Hew-IKn+b%HZ&p4eoX@NX~$a^I|+r9k7FpI$UGtlEI2}S zPHP%jhr$B@XEs%^el)VnXj1d>s~Le@pi1p9T_mXvlcBLxkqoDw*W@)|#}*2(aB3{_ zh@FE%T~rtM@u@LI;;5WIezF}dZq~E6d6%&c znIW9c@nXGuXQf-GA{>DReP^Q*+mC|!i_IlkNsfo?l}!w zhO|2FuH4?0(J=0b&%??pJW1}KJ){w*B2r=bmEYI4^+$P1hh)G6zLLx`uK2=hox2}FZ?97wQto&l@QN~JE2%g0Qes%hEYZAbU@2XMB zx3a z65ZJ{wuudNtr1Ul4Bb4e%c{L~Bc5#o$L@})0M0f>s1dVOO6;6txzoL!^}(-{yHyi=8)NEnn~~haUXQa^cR8teQrI7Qb?cKV9$JxMllop- zhFI40{boK7kYMLRI+X<(Vs01b%n#wkopbj2t_!c|$=hAYmMeQQ^TjgsTw@!ofZvCV z?59SW%=!H!$5tl8?v$8t6u%&jWmrnPJ2EXtxo1y=;-W@5PWnal-~?t54I&9%p3^D= z_+tKiw1>-8DWxaECg7{BA)$pry5u>K>`+Q&U)ZK(*!`fq@8ig@FHPh z(v~lC%VpjCney)Pa_%U1PY1~9S*0)^%js%})+6)N96u1@C|4@>W;wFFpzmupt_1x> z{;x!K>yXHXiv14H4W&sZhx&L%tB%A76*q56g45tc_m>VE5u?6JZExi+3HMCIFOYFx zbr+@fqH|#*DCxX6UI6gr(k=4B7yhH%FQj01%r{eUB zqNoY#Xqo`aKA0SM438 zixI;Lj{_o=$q-5)S4XURFKX@frGp(bme*2CDG}$Hn;{srT8TQ=o{AmJy+ln;_nXDs zDuxSN2OM~^9}l8b;2yM)U*;1w64}HT{HkN-<7++=`HG*=tQ#uxH8({z>G7v$o5_U- z0Ztc1X`%GQU>30eEZBbN3Wn=j)4aJxPcCVBlc7u|rARIF!_)XZcE|o#Uc~)w3@r}t z{aGUB%BZ#Q%_sBVV#WKB4|L}8dd58nAbQsDS{hep`tqqmPZl$XYYDyPh>f*Jp-!rH z-DoA{u@iSvu$(Z904|fb>Qq&+And z4Vu--v=pr55D=(?wlB{Ha@UhSy%5_&H=Y2z?#^~ZDe9R-Q%~rF^>|rHRnK3jRq55P zx**0fsbbB@vxbWFu)jl{Jshdi!){N*zW0Hjn5&{<51`lC2WZgVzM!oiT&&W={&k)e zw2OMHyhyLJr_z939Ki(1*7Kp>3A`AX-n0R8lb z@+LsvQjx!Yh5T7k)V07!7s`jzufsDU=7KSLrc1y5uEB4=&%Yi>acBC94KRK`rbfBD z;!y}}4pKZMh~`AUd})q6rKaNi@E1IN?idZQ}RVZdUY%jML4u1N@y0= z5xpw8uNmS4O?sY4v}lqvnpW&AAVQ|MH{J@0qj>tU;jrQ{D8R<*ccBuxbd9-F08=lR zYa4F~(t$*O#SQPkBnB`N6vH(@^QP-%U~hV9!RXF&=X}2~+|B41W3}5*^-qy>|}a5M85cZ?Q}W7c+<> zCXYnM=y`9oiG+LGXS+OO57|6b`*WAnb`06Y)btY3keL<)72^^3-C_!qNu;N3rf3QO z7WX!CVV^ErRvSvV75Z&iHRaeF(MN_59lf2YSe9Onda5>jy;d%Yw{m^lnyQA7`1Xlq zrvMK5XOvhba*Pkz{_PjKaDl(g;5{j4uX`U67Gk*ojE<++vUbH%ugcZr3V%?Q)f%Dm zb*+*rdVJRs{lje&(43q9TP^&5o}vCQ4iO)7c!J5zmlRE~OIU3&g8&Yi#3A`77bKQIy*nh714E$AR>OjAZ{n(-C z2m=ZMZ^=Rf>Sm4w%CGJ<|NR z0>%DA`Vyu|>I^Vw2L<_2c#^@pi9F4M& zQ7ws&OF{kZH2Ig7i)HN*o%ilAO{i*Mnj><~%qD%Kd5eXl1JRi@(RfXS3hI$1VE#Zo z;$|H5;g&zfPg9NGz#Q$d$ z_idN1HZZX0Qhy8ZZf4yPjPgD^4`=uL10SL**9cBr3NEF3Ln+5ULAC(oL#~f#YU>z< z^4z%mak6Tr%xkbRh3cC$x_{5&3f1phqp9-T><4oA#Xe}sD5_h7;ULQl@t`iGo7Psb z`(yesicC~5^I*&Oo>>31k@%kBn!5CKH@@$;r|MK+l;|))m5QH^Xtd+HNlyj3MOuEk zX~Z1i+#0V+=+bJ{^;2RWb2!50WDe&0;w;Jt7Ja+PMWQj6I`ZIbV;or&flMpeakGhU zE7D+3IV08fhT2)0E)o%L^>9SRn|Tt>?vL+XslPXC7O|HtT(Lb2O*g`Ko+o#XLh3o( zH7&9RwWn>#BXVh83U(T{9Eg+qaY^4tA6h6`QwehC;OS78_%2qe!Lpj9ID+#9oHz+L zkJbgm*CUz;q$AFo13$+fh#m2u-9+y5SO{ilJ)~&`K}#FC)4%SE_1X~g-s_Zyw@Tcrv}}eEZ2Bn9)u;Hb)1?^_ z_Xk>v6&d=X^U$aG#%YgmaAcp)x$198r%@!QJt9D(zAYc?8KV)OTh zv9jVQWt;25{1F_5T3e{-p6pe!^L9mJo(VJH=dV@V4H4qx9Pg}?OYvumO14s^ zO1S?NVb{G3muvG(dUfG1?5OUjA>z6ud3 z6-byM&RUv;mk9Bd-Ll7pt%dDZO@vDlm`xovxf|-on4~gY>>yTQxU;TLoBjHJ^$LfY>b7fm@^zQ1aT@JiTnxswJjrN3WSfbx0IQ0dl^{>Xi|UR6{)$5}K^28t#l~W^U#n8M(R|B6{KY|*drpC2 zDuuHkZi+NXd48b!PT+r5#yWbfCqpizTgL)hxd)W5%4lo~SpuS@VRe7%XfdZG3-bT<_^U0!ekg58PFs$kkzvb7pteG##tW#A{)@y^(7PoLvh(1m3tkB%I zwN-q1U|-T~cm!u-;W1&O)tR1jXxv3QrzRL_oU8MserOIM@M%LcIo+`i4Ngey!(c~X zXidX`B2je2qF$ow1=`AKu|6hp zQD5!zrs<9q;MA7oV7?)Kb~504SZoOdokeFGwbCSGcyt=phH`djPV&Y;Ps;;^ivjcTOSJ29 z{gW>xNOZ)Y(Ar3e%uQ|7MEy{jKYtlmgkL4Md`B?Xn(S#p7TUW1jnfSq;q2b>5c$?! z>gyy~)s;F+WWUtLUDuS6Wa%fL4;EyGp(Vrdqpc-z-7zskCjU`k)@Q(Ps@wkFw6r^= zAT#b_oTHQ4Y2H{EYfX)uzBUnplU?nguzu76TPr%UL~hW0s_BHwyMmPWdbC8*ebshD z2}6@Dql1XZ;9_$ScaAExamDCs@O$EJvDq6GjXMYU8{#}`33cVAVbAkjf_G^=g}h>1 zvx5@K@yB9~7r*&(GWL>FK5Y4imlF)9U$BlP!f$G(xZ23O)aF2t|Jq=meZT2ywRt6& zTaVY~gKr3D{VxgBe7D#4~8+6@=)IW*;}r1jhl zTy69wyHRgpLu&{DrvaAzhiEV!P5)w(Ox?mz<~6DmVol)CVb2gh9*cTwiuyN#bJ+vP zPp_gkP7>jv657~XPKEb!`1Ric?4i^SN6rMO4Jy{eWb)?;S~Gbg9*uwM`jcMMpn^8m z*~Y>8{l2xZBHCE%6}na;xLzz|DHo%`{i&iv%#eD-0?~$MU5k?;z?c$uy;h`$HyHF| zg(m4z08D`_HTgrgl?IWf&p&rQ$PU6-8|R_vdYCyDTQXR1>wEuqjhwe=6PgXnawne$ zqVce(J2;k#4ccPy-;@U4XnvKc{J>Jb&sJI_e#|^I;7^^tApP1UZS2?S4)Qn-TMnTb zQ{ATE%-$(t&pOIxwg6zTK;$pWX4}r}KMx2i%ytGa>&lOf~c3B3R{Z{Pw z<=-8Cm!H8Q?krw{^f8mq>uxijc_8!Q*4xK^6G5+*PJ!aq{DQH0y}K4QAMDQhY)V{y zf$9MMVX5-2RE8)O-KW_%m;GyHrTyv8ujRcIB4_wa;k|eZl6)?9Kk%j|gA?xuHj{N|wgvgeyGCk;I8jN0tSa zR5)y29hoD9mxJcz&CS~qOe?ASlJC>5ndbL* zaLo5XyQTOOI;QAVIJe4#`|Y6j zn_j+jGexF{1ny8%TLAXdJiX9^`ZJe{kal6O(xyh$8}}kS_rUed@$r@d?4j)>r`3^> z{y2krzGpALqMeU_%Fu<#!>_Pa83iI$5xZv7@AGh4!z7uUeQ6Ci`G1)s%A z%;k4u3kP(U^D(>$gR4`sP_{HL_0FC4YpQQ|_cm3IKxFwPvYWk4B6Fm*G9OnyJ{Jn| zeeAF`3I%jxlmNS|MTI%< z!#Q4dEJ};PN@Ytba0eiF1q%4rQ@Az;Ilg~;9osO)0*&w)tqb3Z&W;L z#xcEGA07WZD`Nal&3?W)dtft@_?dz*-`{W9MVjohJpugHK{>0{<)G2 :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/opt-loop-invariant.html b/docs/articles/opt-loop-invariant.html index a227cfa..0db6bf7 100644 --- a/docs/articles/opt-loop-invariant.html +++ b/docs/articles/opt-loop-invariant.html @@ -31,7 +31,7 @@ rco - 1.0.1 + 1.0.2
    @@ -56,6 +56,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -74,6 +77,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -86,6 +92,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -112,7 +121,8 @@ -
    + +

    Loop-invariant Code Motion

    @@ -128,28 +138,35 @@

    Loop-invariant Code Motion

    Loop-invariant Code Motion

    +
    +

    +Idea

    +

    Invariant code is a piece of code that could be run outside of a loop without affecting the actual outcome of a program. This may come in the form of variables that are not modified after going through the loop and as such are called “invariants” or non-changing. If these pieces of code were removed from the loops, this would speed up the script being run.

    +

    Background

    Loop-invariant code consists of statements or expressions which can be moved outside the body of a loop without affecting the semantics of the program. Loop-invariant code motion is a compiler optimization which performs this movement automatically.

    For example, consider the following code:

    -
    i <- 0
-n <- 1000
-y <- rnorm(1)
-z <- rnorm(1)
-a <- c()
-while (i < n) {
-  x <- y + z
-  a[i] <- 6 * i + x * x
-  i <- i + 1
-}
    +
    +i <- 0
+n <- 1000
+y <- rnorm(1)
+z <- rnorm(1)
+a <- c()
+while (i < n) {
+  x <- y + z
+  a[i] <- 6 * i + x * x
+  i <- i + 1
+}

    Here, x is assigned a value y + z that is constant in each loop execution. In this example, the assignment would be executed n times. The same result, but much faster, would be obtained by doing the assign outside the loop.

    Example

    Consider the previous example to be automatically optimized.

    -
    code <- paste(
+
    +code <- paste(
   "i <- 0",
   "n <- 1000",
   "y <- rnorm(1)",
@@ -160,9 +177,9 @@ 
    
   "  a[i] <- 6 * i + x * x",
   "  i <- i + 1",
   "}",
-  sep = "\n"
-)
-cat(code)
    
+  sep = "\n"
+)
+cat(code)
    ## i <- 0
 ## n <- 1000
 ## y <- rnorm(1)
@@ -174,8 +191,9 @@ 
    
 ##   i <- i + 1
 ## }

    Then, the automatically optimized code would be:

    -
    opt_code <- opt_loop_invariant(list(code))
-cat(opt_code$codes[[1]])
    +
    +opt_code <- opt_loop_invariant(list(code))
+cat(opt_code$codes[[1]])
    ## i <- 0
 ## n <- 1000
 ## y <- rnorm(1)
@@ -189,16 +207,18 @@ 
    
 ##   i <- i + 1
 ## }

    And if we measure the execution time of each one, and the speed-up:

    -
    bmark_res <- microbenchmark({
-  eval(parse(text = code))
-}, {
-  eval(parse(text = opt_code))
-})
-autoplot(bmark_res)
    +
    +bmark_res <- microbenchmark({
+  eval(parse(text = code))
+}, {
+  eval(parse(text = opt_code))
+})
+autoplot(bmark_res)

    -
    speed_up(bmark_res)
    -
    ##            Min.  1st Qu.   Median     Mean  3rd Qu.    Max.
-## Expr_2 61.23334 59.35936 51.22635 57.88994 49.62913 119.566
    +
    +speed_up(bmark_res)
    +
    ##            Min.  1st Qu.   Median    Mean  3rd Qu.     Max.
+## Expr_2 63.28935 62.14527 54.43887 57.8912 51.62791 58.50693

    @@ -219,55 +239,61 @@

  • if/else code motion?

    Actually, the opt_loop_invariant does not consider if/else expressions to move. In this sense, the code:

    -
    while (i < n) {
-  if (n > 3) {
-    something_without_i
-  }
-}
    +
    +while (i < n) {
+  if (n > 3) {
+    something_without_i
+  }
+}

    Will not be optimized to its equivalent code:

    -
    if (i < n) {
-  if (n > 3) {
-    something_without_i
-  }
-}
-while (i < n) {
-}
    +
    +if (i < n) {
+  if (n > 3) {
+    something_without_i
+  }
+}
+while (i < n) {
+}

  • Invariant subexpressions motion?

    Actually, the opt_loop_invariant considers only full expressions, it is not working on subexpressions, for instance, the code:

    -
    while (i < n) {
-  i <- (x * y) + 1
-}
    +
    +while (i < n) {
+  i <- (x * y) + 1
+}

    as x and y are invariant, could be optimized to:

    -
    is_1 <- (x * y)
-while (i < n) {
-  i <- is_1 + 1
-}
    +
    +is_1 <- (x * y)
+while (i < n) {
+  i <- is_1 + 1
+}

  • Include repeat optimization?

    Since determining the conditional that causes a repeat loop to stop is not that simple, it is not easy to remove invariant variables and place them in an if.

    For example, the code:

    -
    y <- 1
-repeat{
-  if (y > 4) {
+
    +y <- 1
+repeat{
+  if (y > 4) {
     break
-  }
-  x <- 8 * 8
-  y <- y + 1
-}
    
+  }
+  x <- 8 * 8
+  y <- y + 1
+}

    Must be optimized to:

    -
    y <- 1
-if (y <= 4) {
-  x <- 8 * 8
-}
-repeat{
-  if (y > 4) {
+
    +y <- 1
+if (y <= 4) {
+  x <- 8 * 8
+}
+repeat{
+  if (y > 4) {
     break
-  }
-  y <- y + 1
-}
    
+  }
+  y <- y + 1
+}

    • @@ -289,7 +315,7 @@

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/articles/opt-loop-invariant_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/opt-loop-invariant_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/opt-loop-invariant_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/opt-loop-invariant_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/opt-loop-invariant_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/opt-loop-invariant_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/opt-loop-invariant_files/figure-html/unnamed-chunk-5-1.png b/docs/articles/opt-loop-invariant_files/figure-html/unnamed-chunk-5-1.png index e8e51c976f8683958926eb29f1756c9d7d1e17c0..53864572851b658955199a16dfebb946336b7251 100644 GIT binary patch literal 32955 zcmeFacTkk+);)?k${2u|F>nmPs4ZZE5mb_Zaj-#x5hacaq9PzU2X)jjfwlz67!V^U zIb#?kgQAj?M3KNe)#`IU#cbK3!jp!RCVpo9Iclih>!q-V^_et3TF?5mbLZftuzzRFTFRyA<* zjYE%h7jE3U(ssuStp&S&{iAAj;zv>SjaN22-}L(pwR?|FUjKZj?w3uf*H`})Z6TeX z&lYg25=q?or+Udb(w|yd(Shu86^&*Li|jqAOhyY#MUX=H0foN2O6>%Pvt9#U<^ z;c9sQ;m!>nQs3U+-O)NYOG)65-uQvuupO8An}lbI-}e+!@O86zvAJ)%+Aj{%XwGv~ z>t@oQ;^OB7>CdpRA82=ZJkI~`2mW^*{?{J<_gwt{$Ui)KVkN+{Yoqf(`&grxa!zlN za;SZ+tBj&kU$)1z8CFAIE}LWt*{>T!u;Sr4Bl?5E(0>+5=-qj|KjK7lzJHsdl6_}Y zYG1M1aJQFx&;5?vgDo=EKEmp8QbLiROTa5Bk5y4iK zW(me$-`{PmPO(Sqh_1G)Z0h@{Su~hFm{zF{M7Nwzs9k^J@EI)l{2 z&nGCJ|Ixd=>rH6)a|_qC*3J1$)|r@s%}E0lrpi6dr>c`}^K47?{nA}WMUPm|h+n<` z;e(eiUuL#Y%=n{sMc13l4(%_qdS1;CU@)INeJX97>wPAv$#-e2MNx>N_2+}y;qmdC znLU_6i}GZzvhJCZc%$z6Ovm9Szuq@-7LQfG3T zYUSI@0!7$ZrIHPHRhC)T7fXvCe|Me9%D=N=C6?S`q`$57Sxo%Qwc3&Sn(iYXtK28Y zhsW9O88z0fBYio`FE3+Bx)0~Iez>pVFZu(%;AAM`rbu0hu0_F{i*s)?X7KXo-(0?9 z^EX-deriNw0|XVUshm@Gj&3}0#e!}2;hD?o@->Te&)mATQl-{S01Cx`JuZt)vWsEgRs(E$9l^QHoa&ef&zwFM{t5=H~GA75_(pau7eeZWL zOP;B0IrjHodjd8-es4&I?h*fLVQQDKWNO!T)+TLsh*j?D0;~KT{BjIlVTV6hM#DW# z*lyi`ro<*KL&H$ZhIc;o=DT+7D%{eWUuBiIzT70#wd7h#PC(pW5qtbuSLTa7v~DS6 zvPuoz6<4Zwu3LYB4&JMGKRbL;EHOx{#M11$d4_-G@7H&$tE;mb&+!V2Uu;})>dcw^ z#+=)|`O-yPJ9F%5oF^pTx6<#W>{!=Z>xLULzUHw5mzBqsAJ6hE$n{yP>tasBA5g8k zc!+a|a3m$n4Bf~3FQ$AcgDFYD*cr^r{6bQ&*=0^&?=DT9>`Y;Yw0?|;&`mIBvTDaW z?UTlO^5zcO;!Qj!*ZyE|au+T7NMGx1hGT2EUr%r7WNPXR*LfL3nUXzkMJ8Uvy22&2 zc+B7*D$6Q&YzZby@`*g*hBIO{+L0hr8APuNoPSPGXK#c?u$0SCSMdM_vrk)F``wkz z`9X5lR`f{xWxl1pFh`5N?tvO-_H`kJ_}KAawu@Mi(rCMA!H;;!3%Oq$?)nJ14Sgv; z?xo0LSY(XbkyZLR-cyc>jFgx!rm^n1Ps-%PXp5=gXoaZAVD?HcvUb+9BZCe-jX6hR zPks;;A<*}S^@^_TYo9)QmJjEpTQM;)m=T(JP0-Kso-(^R(4w#{DkFqZej?D!3X^tg zVwn`Z)-Qs73A z-(z4m-q2baW#}pY?GXoKo8v4>X~w3X`uMP^+P+qug^dwjy!p)GYr$>o$q9CEQjx0! zGx|(<^qye(&-?gjN=vumE%F_)ANT3%1~UBRtWTUM3ej)AQgR1>6~_8Am@rj|w7y~+ z`0dO>?Y(<%;fQ*tFN!%)kh{`rOYyN5I!Y%ce=rcgZ;@Kj^hq}^u3m=sBj z`aG>2DM^Sv`{fP8)O?nVQTEM43WI()>hqS#{>G%ASK)h}+i?!V3Rq0OyzMvdDEINU z@D5#zv`tulWB(>BLY|~s_caDH%Tvhu!!AJ|^B+o=tXaEO`b7Sn1KQeROq`7LEt6xt z&+Xi3W$*p$aFOD-NL&6wzS>A4+xreBV{ zR1iEzct=}gibQ{Z|8pf?yk+apKc>E)b>PsU=CR%)NwZqJN;7{M3xlVHzuA8KR9rj{ z?$rLNL_wE!(Y`ZpadLK7+-;Z&Mw6;Y1T+%aYRJrdzP4wyX#6PR51-U(%cg z+WoBy&R5cDF#}HE3Vq9CaK383)`ats)yo{%U|D0e&n7)1Wrni^))|wf* z4L2Q0JbN@o$h{`j(O=f`80L7Jk+sZv55HXF%@Yc4)=wTi%D%JV_&%&TW4OG?%@-i+ zVg7#3^!PBny+}q2FbIDJy+~9&z#9p%x>D!Wg=?kEs}k4g#ELVeU{?LU+fH6Xi1*vC zsTddQTCk}!_G`u8G^Sn4yS6Vqh&VfYzUTua!OW6%kIy0mm=|woYipA}`sUA6;1?!~ z&RS2Nq@=!3rEjoJF|*v*PoKp={3(FT&E4`dUbS_2cIx{}W&lb_z=dj+ll_L1J_?d9f2I?+#!uhwh_Gwl2L{vF*jQ8}=uX-| z#nU4uW~T_UX~;Es7Y{7d*1Kh#=j&s+0d$#k&%UeUzsoM&{cQH zKV=o1fyvlVcew2Z5tpvlaxr6X~t4;nZ&`Q!&|TC3(KvA!b%7~({(_v&(-`Me>CfGiD#4(~5- zFSlG>8@ZcBr!A!;UN;`cQc9D6(EDV}OTDs1{S6`-{Ec@r2~MACN07C5`*H#FOiNIH)h|WL&$0*OpCz=jt|ym zKKc`gq6M7f09d^X3>L^VT#YR8&orgp4pcIwuY-+=a9I;}(1Sf5HuTkc(x z$ffK@N|W&Y`{*AG!t@c-cBFUTurTsc9N5iTXPCa_zKU1j+Njih%{llGkB;Z7xv@** z;^HW0=zXl|x`r8Tl*78dOs?39;^)>BuxcHugPqmVW)-pRu^%;j(s0g z^0q9d5B1J>IA<539c7Tz{cb5IsF;>C2LE<+`^0$dq~eVmH;hC0@wu?!A2H}pa9Vmm za&mPZGgQ5hA(<^!bh1H;W%!OWqyR4rNtcFHe|4=@F;mP(*cm63ymQOMC|eV2u{eUZ zsm$hJlW>@vKHinYP9Ih3v7Iq%UiP1}=j$B3SCU}UT0Az8kWoOm1;<0Fx1~rErX+o~ z?iIzH*4hmBB$Yo89R!ExKGEV{2ut@-lT6#F=DRKjRGc+-yczMSpfr6Tu06RevPpP8 zzVna!u}@!BzJGrcUXRa7-#ON0xQ7gc?d;N7m6QaVux6(XHUnC^{O-BTA2u$v=f@`N z?B$nf2eOB{8+!kqugPEngapSFDJg`GR<%gMvfHEcB#c9T!8g=3oYFpL|9tQL-<#+y zNm=68DAzR{dLtN(BVlxHef(}f^g~ew!7y=&`J#KWfgpVq*s0PgLbeBzR4yJWO|fU| z;o#D$SoMF__A9A(9qZ2O9cu70dF3%}@fL@(1GVmxHs3xy&i_cHs!Ob0(ku&O;}jQN zk;h2+V50^UoNd@OiatI*j~_kS4X81TZ;3ZfoftEHyu(7**}LEn#^EVj|5p{TvfOyz z$B%KQrN!1jHia@7!v*seOECar^LGd+dS9F?=xeEiZ}L!%(l>ZLYRSpUH1ps)7k6-g*DI$v~=_8*Q9dGFSIS(&OjP z3n;zE(rs2A`QX)aL1gUL-sS=|)-$~%DX%R(yM-MaZ!ws7cQle!I<<77KPFhzo_U$M zsA-u9CiCl+SAh(Cj=xPSHvC>RNwk zljaK3D1CMk{zj~ZT*GyE+*>Mcb#7kD6AqE-Bj5T%OwXJ-(`-L9S(~S`Y15|lR{Y`t z>*LE-5RlT?(*5_AfwJRC7?d<}piiGa`xp?P(0D_pCOd7gCZ)y=$j~+(=^HXgX;1)a zrIGp{ldM~8+RHY4{`{HfXn+dU(E|q#fDPDy(|PLTN%3*S%g^}BZXB=Hn4=xm07&7X zwWBMUt;39P*OGL%j+Z49hn+*<-%Rk)$5Bz5uCC)wMQQfV)sCf#r4@rQ11bpXh#~tC z+!10CZ6XkOF$H48i8}`oaU%QX^45~)x@waq@KPQHmtF;Kx^)bWP!7J`8u{%BoS#DY z!(ax7ivhSaL440cDmipFD-m8QhLR#gxZN}_oJ6q@hZTN?ZjKK%Lqmv1C zZ5RyrE+rOpc6+j2N5wrEWoMj)+s~4g?5(@lZwq`yNiYQla*Jd$^EJ2JGMwn~bAOAsZleb1b~5>gj&Qht9Oe{T*hREH#|2;ldm0mC&OaE7Hd?}CqAf%6aY*pbtU1% zQ~|#h$Q$Ywvu*_+dj#XYds#RDCvPLOr%-|2n2<4UX5iEsPRPm;SgR*Xd16DJ@A|d9 z57ysLPL>yvw>ivKL7Y6L&$fN;{JASxdzQ?`jV^UB1=s6%3oAOszNO_c?(yx-ErUM$uftkKd>wqcr?nJtdEEAA_4nEqijKJQgvy zEP-*64vZz}xTl<`^3ZDmD`2lCfERU^o?W>yOr#-Hv37MtP|vxeUC6<+eY6cg+c}DY zRM0s1dBm`1zJ~0V`!%_urZ~qk2x|GE%E|=EiCyxlGQn02S0nzNE3)mZ8g^yRsEApg zP!(-s%T6ZGfvS!PfnI47jfp|`Ng2e0BG5=N$QGX+ez^j9P57$$B@)Ij>}FrUg6{hd zywdSg%nNZvQTp{qJBmzKwk5x`?PzFwR|eZIwk(nGdNJT&>h8u~X6in^?PvSr8#ix$ zh}0dP(h;$WfD~i3D_=!m+@v~>i!cv!qJ5^ytLDs>X-iM(Ej2ruTst39hY6c?GMdG=w>0ir(oNrSR0`k zdLzJk2R5S#SSSZr+lPV$j>NHRAp#XokvI&xYIh+)mm!8>d1Ryh(nAV1U%S%FQaX#; zE;R|HoKLuS?^xOR*m}V8ti^KULs{u z7{ti=S!LPFZKLB!Yh4m5PCq5o22}WYG{iI~BZvI#r^owy-|rAGo*W-6!Z}WPt=%ky zsM}V3##|qmSEB|Yqj_1Jdbj4|-7Lr6=B6uRA=wnFk;H7qAr8$|EEXkh@nWgZT5Vr5 zPprA!;P3|o7a;M|w~(yGA{M1U-#$OhQ{>kY1&nA# zC}3@x-0oBz=GH23`Y;|*zwj$yEGEcb(0lxBefRu)u=DxqfEG%V#K1cg4!|ll<~2XQS7-x zC)wi@gN{6bQMMVAu8Vk>C8IIRVi%mV7lK8SA7U9Hml&=YO!uQO1WusPFTF2ZIfV`0 zDjIM!jol4YQ>+uOd*s6B8s6W<)*XcBVj=6jzlxSbd}Pyf{++Hb|$r0eN>k*4s{*KK_WDHcXVPjW71h7 zW2|TJ;`JjjCsr|)u<8k_Kc0rc5zh^O{_BNVe7Rpg+|OJ%#DfUt*pnk>J>Fm1LV#p6 zg#9r>bD~FbQi}NJfhy}FfW+{5`{O<1+S=kTA11&Xxb)oMB;^>yXSKt02`zCFNWwB* zU*1YP^?fC7RK#g;5Mf6S3ulIWC-zfG$-}#MmuHN8j*jE0`(+4{mGZ!nqUOu?eh1A+Io-l0W3Ry>%A^(%DJy8xIXL z+yc8iH|K-?4yH9?_7Iyd_wCVrrM@^UNzvxYGau`vx9K7=4JBR)#L;2)LqP!bLPVKR zO<%+BT7Y0c&}XqfDPfFTe~yJ~e|BWlZU;OcsHP1})G(45m=W8#V6TdhPnL-L76HXR zEdrbKmTdg$8U>PJhZGR_Sl4CceDYRJd44AFQ3ddK#To9C>dw+jk9P0{cx2wZ(r+6^XlZLKj&*?-%=4i}@{#=}W z;t#F7rh7ASY?AHSAq+~Z5kYjCuXtDydh!xhj6V6{9#SqphywOd0e__Q?lnP44Sjum zf3bVw@o;YubjH?{;`i@g7*st6ud~=B=vKyfWybnmBm9=9 zK(nrTC-QvLf#W45C6R+gx###kmX+N>78JX{e(l<|))T`mEprWrwoj$Si>FJzHz!L| z9?sEZJxj8(|NF0>aV;!{< zX6d8V?1w0!&AT1{3?ZT!ZbhB-z(wQ2N?XYjDl}wyN$TIw1+VQsmgSyL9ws6_-Y%^aNyJm)?E~om<^(LgMfBPC1pmor z6AKNjLN*|#Y9*1CNUlSB_-HG%M14au`~%_0+DL}~=>ZyQAf z+vV;O)c3w7nJ!CVg0)Fa!3E4@L3??OuI)hcC)hhmVC#^1nzM`stL=*r*bg^9#>;tf zchp^+?f}-n8dRQ7yny`IustDL9547XhT2V(1BqRxZD(P@WKDu?4;i@je_j-ru&Ua+ zNRhKVj8?!3N~`&Wo&;FTUdKRMjJLAajWaV<`0?u7qUyPvA}$3O&n-RKTrNh-fS6O$3(bhM zECxs^v){dh$6rQPxkdq78`b}Pg{qva#&AD_T(F$JzzHY(lU4z4DCZ%EVKE596ry;9 zobCI>VFovnDZxj+B+9AY+O?Mz7Zh*a6{_TFg(^Te(h4U1wiydh*4PmL0Jt$KVdGoM zd*KrOxHFF5R(EmpNw`dkE~iGp@Teit4eS^qG6f!EZ$&!LT`!nenab|KN=mEXpk$cQ zaH%>06s+ndi9rv*f{Y51Qw&Fn{Q}09UEmccq%tl8$zl4*S^XB10~VAKu^6FHC0OTg$ntjNaJTWhur5Y$t;4mGp7A21{ef=GJ5FRTh)Q19hFc6_2id17TwK-2s}`L<_=bWQCr zU@n^t#vu=RPWhUa@!tSfh@C_Q`!)kL1qY;RQPG7Q54i}m^zVAs1)Cl}eX7QyT5~z@ zj&4U4@@+ZPb=0)8Yg`AbiILOo7-RQ;44v3W1>@Tck^o@8b>fd;X?QA7gR1Z#*+kmX zr+h;4v7b%8qQIsh-}<0<00AgB`oPtOvm`~C4v)g{+u1&ff%6#xZ`>)luD#w*;Krv z>OaZ{C>w$llIwakBX7%xC5FSf+R?DBRZL1rTpE_vp6s>;R^LPgOO_F4n-2(7UEe|* zAuf+W7=3wj(SO77-#KT+R66fxD%Jc`@&-d$gsmt&*5ZsLkONDYq!R>)lf-VVw9|1E02{A$$TnDvg{r(StTFZcM z|6QxSgEHo7Z+_t|vKYbo-kl=PqNAl~TCf5Z1LeS|(v;Oz4@XFRGs@0TX71D14+cx= zGcF3*D9Soig+_6zr3b)MlqqlD8H9xWZvOmnM*>#aw^piE)(#ii_aKuxbK7e)uy#z8 zXq!``ke6s8iP$`xNblj6*N-1RPBv9MfK>^EkMy>@$0vRAeHd}A;T;rx0D3KCMjWUf z^w-ZT3Au8pIKma`AsnjKdL>4b3^yeO$q|DI9MDU2dj5Q~M2u`Xr8cYwNZ65b+8KaE z_Hqr{b#0OHP}diKm^91ivXEUi`FA*==4G5pd(5wR;}Jv$Zjlt*ggV*&cUP1`a&vR_ z9KKv6R+&WxLUd>^4zO9*2CA`37s}p`CHmPKarMFe2#xhp(wOU3XEA7S4o4d?akc_G z9Ec$zGi2mYg>8nwmKWuo)jFq5O(|j;2b#r0JpSutdM?%fm}vS zBBebV{w`vd2}f~K^n-ze7bRrO=wq2ztKc-C{~$370-!S_A#dEq+YBY!SwGMgHTfBa zAQDRC1pK4~PsT}7>NBr_2?HVEGFn+%h**9{8Poui6{(+c0%40ZPdc`Cn2N{1!+S_f zA6M^#vH0Sm~gk zD3R32O6__xH$01SX5unXx^_ll--8NGGwLdZ=V#7UjUo?1HT%A{(x_#}cf_=Tl_aE3 zr8nmVVS50@)H1XW|K9`mV!nN-yB^ozlYGIB7*{Sa_8UlG3we17`=z5~{4!h-@?}xbc}(KE=GQA3mf0qEI`m(TM^-pH5+Y z*aiOFBPPK<;BD_Me0ZT`dDrymk}rH;&1FoNTxxNAM%1b$b&4nnS(O|y2?yR~S^+Dy zkf(Wce4NS_pmPsbAmYbVs#mF(jGePXCC49v0)C*T;?njsQUK-L7Hct_Xfpg;pElq` z;~WE}_yM<>Y@OQ4i9u=Tj2KLqTyGlNRT5cU|15TfLw9}9**CKt+ny(|7!;mqatYbb z%rZfY_dwFsNL4u4H1Wseky^`ROp;VCENYlz04dQcoyfYGzi4+wcW)k;Q}g6R$c%~` z_q~+lHI;^qsGsNOHCmUBKT8boEaS`j_^su@9A z`95lr#BZ!wTk7(ZRAlrYu!jC~Q6$8o+yQV*lH-Qqu)>iJKi^Epk0(&zoS@R1SLbF} z!*4yjcW(ugv~Z26&~TE|IJLtv$ZUzGpBQTB?RqO9cEma+9^{1BWXt4OXmkMu$E~9$ z_zcQ;;4Kb=`=+`P@c>b213Dz5BEVmHCPuvm#Ya*ahL^(DTHwO-fQM-0 zvaOs756WR6qE-(hf_pPaPXWZ5P}4>_hDPYavT?rmd2XPG%C7&I+O?rQeYBH`ZoO22 zKrmz)psr0sB%C)DUHO&Ax?H}@HKh7Yk_t!nC;e;Jo2&zH-bBO_{-=XOxd>6W2YoEW z@Nymd*fYvyK87hb%) zPHU04PF!SQxOd$4P0`a2ZCo)+YxkZXnb|jtc8d-=OWpSG9bK6rr?F={%b!{3FIUvk z@cM0$T#Ivhul$cQQk$jOC~KbQHeFoa$L&1n)0$&I%i$rc`kK?D@_xr^Zl8ve_)KoI zia|I`hRjwzTQRQ6)iglzWrzG?9blvoHbTq18kn)7FH0X3#)8C zs)NW|W4WfzA?5-nLAdPr<)Yx#CUX`QSBHm-+zsCB-1=CYvwKc=z$G<$w2)fx7sfPI zbs&@mS3_sgkt%37zF_!~L?)%T~Z3gc|_*KI&@Nekv&WaL!V48*7Q z2D8_t4?qa(y=IZt3XZvW2vgp_CK93T7JzNyP@=1eq?X)iRUsO@$4VKCnp6f8xdSrL z)Or8;?N`!0LzXV(J~7r_og%Yo(;AUS3+y;1*!R;?8S_1ey(}YWM2Wg4xhn-MW^q+j z5Yho(sTF5~xK`=(7vS!~Qc4Fcs|~tDNBMPUthvmw>9EU(fpRTfk-`X;3gy*+IS?t` zv-Q;X2R$1EAMWwrr>7Y#AD8F``o+ssG%S{D(u#jWzDM<#$MH$i4*q-ZH;ETU)1w>j9YjJW=WJ zYXt~b>ysOv8HCyIpBhC}#)Ok4W%4r&*oV99sygQOnNMv?7>K?^rPc?x`5jy#D_Xqa z*v{^X`1tq+Up{dm&UxM{@IL)Q0-_{}ze-;>b=+U7qoV`sv#E-n7Ji>IG;wR1C%P0c zPZci>b@hU6QEjq;_JF}ztCni%aKB3`0c3uC_C-{nyKk1-E~^f!yO_(hAaQHx!&XN^ zKX0AgCNG@+@L#_XYH5Wb4B&~X*Of{!f?un-TVQP7Q4yEfzqF%HNXA^x^7+3XvG4K} zRvfum3iZ7t$*ZC*p6wW|G6);wIx+7g&zWpIsO}bXEg?L773^ut!z7Pl?s|DA*|wFe zVgip}Rj?7ryff?3x~T~q5?^q2rUgh`6ew!CM(~n=O{SiwHiz|14e1QZ%`gFG&%0jx z%GC~cqq&DGw+te|Yi+<~LIWc&k7q3^g(Of>)Ze3ryZz$FI~o?V8bSWcw)fk0Z8zS_ z**Nc)_=a2Ubse%1rThk8$G?9`!@C=q{kImRqBNL!Bn{btWFdQ|lNfim7PrY>7~K8) zMn;DJTV3%51C`U}^(jq_zvN||kepTH&uN4(GvhY!y-(prlp;1jH1FDlp5JuhcsW|UCs zKd=1s?#9w*3w%2_&5wJw_t=Kl34yP+ zG~<`-GnktHXP@&4t~+=O#XJ^98|wY~!0lwvLGj z9ggJ0LtWe154gv7M~8v~1P4SbXjC63^%R&Vmg`K5yINo4UA8(-S%LrdyISiURIdN$ z7S4u7ttrwMCQCoX&Ukd7qnafkE;2Qb7?3OEw3#g}?D#f^_Z2uL$u^QFH{N-(>H70ZYm* zPWxvGq_a-Xiu#%F{pdemQdrjk!n9i$XH6Xi1iwyRZ$a+D`nM)tNO-z(39B~$i`l7D zM~Yuf#5Z0DVfKpccLf3AIAx7qT;^SgKN>UN&u{jY%*cYDNj zR=idKGf}_JP{HA@{nUcoI{xxP0-`xJcd(2qET>(CcYB-If+PKz9pZ`20Icj*n@Sfwln-5-xmG> zhA}hlyr7c+_X}of{bMeVcy?Tx$8>ddB@4mAUYS3wdK>rHEGNM*gFGO@^QpHy)!V0j zuAwyCSnq@YxoXdJo2iAncn3_dWtvq}?k}R#lCN>sWxLHf2gE&t^@_3P7OliFJjY^r{=W{ z9L%*82YYw}vfLvsnvY2|0NXSBHjh4c%^&?nESzQ9AFt2TjJfz(>!|#q zROoah@rB<`((=jE!Gk-APeJ%NSikethp9z-fdUaJd{tS{C9Nd=lxonAvP-ybWzOI^0hh2u z;M2=7fE&L&oaQ<8(l^0?pCmJ5F#=^Q3hmQ-gRw`R+|0$p1sz9{ks>I#V7~EKh?U%x z{1v4a4dT}rOgQvTa121OzeG&uSLu9{^60x3fMdT~S&c=LV@1-yz>FN~*MlnjDK`5p z*FtY{YCpu%*Edkg>qOk~S&Khb~l2@JCq&$L4u>hoCKP$eHLS`phP~(1_oGk3ohC@t3^;7BC>rhV+9s& z((o;7G57?o%awhQklM7KI60OPO7;AD7o@l$r|u?gF&aM}Bs&@2=z~(`A&5~-%8s*J zJT>p3a$~KHOkBY%E-q$SxQ|w@()b~E>N^jM3}>sLqs)&%;*}WagRo)0aOb1)2dv+b z?OYqkAkT`97aZ{2+_hYG_n!hPs=;HI^;5NGP5tH;nYAL@QIqAG;y}BNH%|TPB_4{D zy*7v%Q7(i^c-@*njyZa)+zfF!XXyqy>c{=uhy8vLJD+{ zTDShd`%$jRd|gD(acULRJww8LZ`q&w+@@ZfiGo58`KHXQ$-&8q+Oe-H%4?3U_>OLE z0!hV0Y~MAn$*jpC5q~HnX3yWo=59)uZZ$MBa7St^#`^W^hpSqY{782#yRc7V>I)8n z1yq_GE+VB6lhs}kw>F{1$rJ?&t$L&H=-{i}_(*>qT84a}KU~j@MuBDQv*Gwb%42^K zQy9RE0;x<=7$iqIbM)!YPh#WZeB>9U9Z-Kvm0X_p3Ve`$57xT7o2L&quRHwWSJYo} zT9gKp$lny9%wX`%q1ybzywh zG@~G+h9e9;!o+ekpA@J-y$$J=ulzB;=Y|$iWFA|31KAPYjNW?v=p#;4aL0?M(A(D}i3{E&M&>ntI%-MZ zo;;;Cf$_-NM9bq7kfN+7Z%5FIfkw&$Zct1%L;PwVnf&bung_cQ-DWxQUVcXG+*v(1 z(I)7l{-8F%`pD1C&CM=24Mm~oOlk(lp9d_g8r0K8UG6AXH;nqE2_BIr}=oj zQZ_?=LHPlbjvH&g!s!uW(@DoQvBZ?|2{c%rGPU z)6c5rPzjN43bWm8UUm4JESQ zj6!Y26n#m^NG)EO!Y1%RB!;&1W6u@~x(SrK3Th|1J3$&ziR2)(LQ@SH&>YzeTYrq{ zdvZRZIs!^u5@USYpn4B=A##U3kZH+5xTFKkUkKIlR@<(_6g_qzOjA3a6x3bFr%%8k z9YF&ul|w*|m>}^Ye7V`8CJ6s263b(L=FA?{qufw%7iA(}^@V_O6IJQKJ4qqyC3zXa zf9P)poRJ@9cUGswAXi5@rWvxCg0bOVX>fTn>>4%En8>-a0L)y3?&LDk)KU4TN6@`n zkm4dW2^m4HKB^xB97tdkyO+P5O`R1WIQu#(Wk@Fg){MLYSvlrRVi>S;#p`FvHr>@; zlZlUw381}1>kUnvJODiL32ZIC8M163 z3^8<>$kmN^r0Cyv&1eOR>U`^YNDi!64XzharEtX~-#qp;kz(W#qIDyDam(U~r^M__@gBP!_pB zYgEIVG7t>iUTI3&byoS4@1Y}HQJU_H4nsl1(HP~?7;4VsROHUorYmwp1{lDJmcn2X z$MgaqNTN+jw$9Yfr51&2(k?ec^Hu=BV@(%!pbsi*%$j0X-%cuu;&+_pENZ5^%>Z?g zOR^PGigE4@fOv12l~EJk6pj4(ty?fknZ`gbblpHNNdgls);xpaiIiNVh095=LK2*Z z4q03YbqCrmB975HSb%!wwOzJA*bt51Cb9r7kvDCS77#gE6OM)|L_w(0j#0Pja9ZtX z8p|*3;`ZfBH=oG{aS)9ig6>X?#B%73_>T!@Y~w{U)iI9HHOjyh#A zB+5_@*u9dyyzd(%6XZUFEZ;3w84L0C^`+$@&a=}g1fIzNJ<CUQf@sVyo1KilquPmq1fh-gKh3FW4^;1=)<%($_m{2UWgG3? zTI{8$KO=`UnmChtLX{LDi|Ag?wxr)v4VxB$8eZe#HYo!#SyPN5aSt?Wx6zp{)Rj(d z%g?~+R3zGfA#cuZC)J6ExfZZHQsqNq7R=I;5>CFl>GUTz>kAOJhZv643sQGTFQSRy z+|QUTevpw(-kXsFVIlcY6WAlUa`OI|u!b=ZNI6XtEq%v^Ct!X=r(br3pv9RMKDAZd z;@hXv&H^Xn*Uqr#v@e?4bXCHPiDRVh;s8FnK?bG$0Fr;hriPIrWYHx9fc5Ufmkh1Z z#Wb9kF_w31n=ZEOnnber^XJc#AX@7E9Oc#@b~%{BflRf{#}Dq=$Is9mFqnkEhz^c4 zX<-_I9;L=kPU^uu;a&-ESw1b*4&MI^8SwZ#ZHPKO;>P3ie0+2p;;DkZ+tz4ysMy(O zd`mWpTe!CA_Cn9K2}hc`u*vOl`snCnl&;rL-bfG+3WMa@(aMY#+xFdVx}&Le;M&F- zcK5Msl_@YX#e%&7^cIG)VRupE_RoQDG%^xO-Lg)anJG4!ULLu2ic0FYm4|OrGBSMG zedu!PQ2jL~OA@8=%-5BdtbyrBR?bc5VyYBZ--rYzc>yo|YJGx2&PQ z2<70#(kE=brIqSgtA_yNP~_YEt#j!^CA2t-8$fgyVyH;`FI z4bmP~FX1pqtB>MCIePId#{*D%Q3cRIG_%`!`Q6?A-17-*#g-gGpEwFfBz$Kz3b7rs zeokZGup8QpID;-LUd6Z8j!5qpz{mwOt9SU(Vv^szT@HP{bqa%Dj(!~Qaq`*HGHY;**}mex=MP?DY4vnjm@i7UeR(kAvRY0>(^hDf0c~~?hObG zgbNJ;UjkV208$g)F`-S=Cki>QfPAqH@(c%H2^wI!GAx!O%%z@xvaV?SC2KNCIcVFtsne5Y? zw8_4`jv6xIEySp!gxC#$coNJF_xM2qNaTQ5r3s~dUvIz~F-l*$tcgqs!d_bEv60dY zr9UFuUL$Kh0j=X=81=|NMRd%zyG3MDbAU_T3=#3d zT4i5^CC%UL%Y zAwdCA)dftL5x~ZUER;p|#3!V6yfE2p`;NLa-6sd#i%`(ryQQ@CY=wa1GiVag8QbkX zid#{r?H!zitMfdsX@#qq=#tPQT(d;M@?_YU?aeM zju-AEA+>4d@UKGp=+vNFL^w@h)Z9ua86C#IY8$%R;(C@(aRS!f{Qcz$O71USmt!e- z!r#MLQfCNVktGced*2PMd*jw(o2_ZklvfF!@55qp8?>oo-W4g&6p^i%tZ#k?^J+361u1h4VJ6v!U9|{ zL-338IIN?BsIj!-j)U7imaY;REz2TCiQ@D^Y2F9pV%k1uxb(rA!VA;Z&>BOJavN!NM4w*}Z3gurl;7AJ=4@-XY*$n@mpLESL42KP zwwP%45K7_W<0C5hA#NXn&1}Gb(!_8ub?-eGCDSA}k$Mv8)|T=#XZy`(%5D-{GyIBI z=_VMY)<<@HlHl41zmN!UJVXS_3!>7-z?Iz{j#>6Im!-`5Oq8ihtD0dh64Yx&!2sTl z;JrdI(6BT1PZ0cDa2rJ)NuTLnAv!T|CF!sLYS%8HJ|lvt77kwT-n}EeWu|^b2VG1+ z9SXx2l!q?74Ou@^XthJ$djxlDs6#}$``J#VDOR}YXxZDHWQ6*qV7{YVn%073Qn^~h zhkN>H@2PL4|Bf7a)&gzx7r3E(B-M?5e&zw-o+rLjpFL(?JabFm&l0bYly>yU#3ripYdERnP`2sQ^LL(k~Cgr|ODl_h2IbeH3h zs?iO{Nl=3d8N9EW&_x3c?uke^4QEdEmY{9AU>vDo1D3GfQtGSg6E&_Js)iy)5yCh3 z{zt)#p1iVMhMKzcKBz)X;umk9Ll41N1(?W7wq;hYYA4qd|)lKsQc=OE;g zmRnH7o$v~Ea-gulRaK#|fZ!W~u(yg?2zY|>k;Ur&78c~6BgRm_EQ>lI={Q_J;nA?0 zyEGAsXjq}@F)I~~4@GdxYD<4ybDcXa4Gy^Mx*)<}((NfgRK6&62v3zds3GH_mN+Vt7E=+ITYGiwc3 zL*j8i8k>$IKcC5>He{ln|6Kt}N4toi3ERLjbD}5nD%_Fqx7LE1Cyyk?Bt9k;-J%j8 zvsZFw$Qy^Hssg#3nML3YI|b7iGMjowVe^n}{ohS(Ki3oL* zXHugy-DNc;VVo*QmT7{Uq&C?423%^efv!E7J%8F|?tH^=CCGZZYlu86BL}1>UAYCz zJ}b`s`l}}jgZzo%VzQv^X#`!cAz)hW)Tv?QK&MC1ZK23y>Ewu^G$Q>kqNoOQO_PEH zy161y@?uioC^gNk(wMo-lWUwHE~QCP8Nwhvq7p9KfpMN?a2Ir8B-m^?r^zEOW2j;O ziaNl0lvsCjEoMy>@O2FMY(@?s169p^@aTi7M?2jo*^~Nu8Ms?d6V7$^M;?_a?bT~y zdZ`-2q#A-A1_h&7#>d%)tuxV>l6AGsKD9vz70mTVm(-n{`nChuTr$u424bTBRhr*k z01%9?V6t#Ug(8Z4IJN&g^@;rUf>NZPme4bN*CJ>k3zUc?44CmJ^S=TEL`=#xn4J4z z(1{Vw_k+z<&N_unFYuR+NKB_IVeb6!b3a6>M>`p_=0H4W9eolt2eX7 z;6dIWocbMI6F>|wQwoUnfcDc%VG+Id8xVZAUZfr?OGTt&|JKvsa;9gAAz;m0&+}CO z&M{`K<5msqH&Hg4cns{3Jh#TnKF) z*PG>LqaX5G8<3JsrK=(EPrVb;@_*@)j2jOxOcjE>bE!;ZW#?RPU#?kQ$2|ptrBFh$ zeVyau;*23?VzPjbJ7i`}|2pMf@o0OUeuf(dTmS7a>#G^}h#5a<;IHXi<4{TJTAaY- zrjUOWZ?}lrK_!8?AobN3?g7fBE)&Z=NDP+l6JL<*qB--w@+|@w95o5p=MPP&hu{4! zIQ8#TI%x(*Dl=|JN!WK$m;dpDU4Rxcu(OVy@qfgi((vz@w%T)RAgm;19CP4?$s^S5 zit!PyrxTa1PP&lL;BX0g8T(d)t;1Y-fww;dO(51&Y@$ewlt@<){14q-)_HrRBKpaG z|B`9$vG+XrHSvNrq-kE<#mu*9;jh5~tFQQL)>qf5MLGdhAZa0r`xm3K_m6T9yT1m* z#Sl-PG&ODOOM`9_mjauB^!c|U@RLCoZ40uM;N&b~6;K{wPPpAAhZY3nU~=rsqzLdtO{oow~g8y}RQc zagdu%RlvjM*{ZxwL0sWX=lg52|C&{|d&=gt?AzN3B9)!4} z^T4vt&I*OKwy&1_ON=Tq0Nc_k0b`tN^M;@{WMTl-%-dJL7zf>QnR z*QPsX{P#_kyljZCJ-?-ONIO7VPNB=ZVCMHB`hRQZuK@{vYsCOWAA zD!v~B*t}_zQ1>XnKFpbQCu43}FxL$B*xK<%4X0?yO|ASB6!xhgSv-86)X3jeyU$$r zyd^FLZ0AZR?8J9*{q;>7H!djc;EfQ{K+NVGf36%!>z#bz7scZ9rSlA?)FvKM5GJD& zQU2awEW7S`WMqv6?{D9w8o6aK6X=LnJ0NgTp|IW=B4O)%h|i}ao5zLh>*56yER3KB zhhj*Lg)3D??leu$y)fk*Y}Yw>8S%1lUD7pE-e=p`|2(I%aaj(34QWlK&_R5hHTxe4 z>t?j!QUQ_~4a<3JV1ZrXifXnleu{!BM@vOr2-eoUOATr*d7~Dme%HU`>Sb-Xh5igG3LKRrj~1bo&$oNkks3%kz}PKjQJohIQ34YaBpU_;b)RaV9Xf;~{+F z>!&QjH1u^~YWFtNjuiX(4U5C}P1&hlmQbpYz}`_ajyuir6ms4pJ*{cVjW^5hCwJL- z=fpoNl8?Qi>tdPF&>HP97(@^6!qVS2ZZ)0~`qE`N0o3Fxe?A3^0} ziv+qGxQeAdo(^eSkl@R+HpU93QmGa>lPC1fHe8EQ+?wT|EM$Ui+h}98_%r4EQsa|# zI%fF1s%?es@o;{PYeMAf`r#NBmZJ8pEaCrMgx|+QQdi3~An+Lm+K|M5|4tzR^)`n) zIw{aBEh?po=f9sJrXM*jYuvsqoQGpand>?Bfg6wZQx)4e6kGO%xUb zZmrtFK=QkA-RUn7Rf|f6?AxKJmcTU`OjeLfPYzg+qHfPw-hXk8>ar~8`)h5S<`P^q zB2K~HMNfL#l)U!qa?nx-^d$8XUnJ7q0hX@c9!CiOkM5GD4!YyIeuUE(gR8Eb{{8eU z0pTDFbQO~b{ToYV9wsFvL8MzZS&D+qse`5NJd<1UI<1aq{rSVjP0Sf$zg_(4jq@-3 zr_ZcWb5L8g!TRbctDq0d)&#!$`?1Z&cNqY>Vv-mjv9!ruu%U~^xF?tJgT-I zF0mavIyg4*Ye;Qm`%F6#*=3GtC+wQLTr$uCRx}t?%`Q!7k<;9@i!KvAW%qMg-VdCH zuEkbBmy6;Jp*bAFKbo)@5%FR)>laIbcFnPOL`EBB!?x#Cf;2U(N7irpa$cFT_rq#V zADvw_V*LHs*o}w-i@@!O1XjOPz-0>yoM%0FEyR6)F>0*HE#|LSu_ClcP`Kzf1Xe2Z z1?bP%_bVs4$-YCBlU0&lOey_a5#CNS9xyroX41_Ka8jP&_I?LlcK_M4b)*((&nxv9 zzih(ksxW#Ff=37`!i~TFijI!%dSo5kob=UH75Z>EdmEVp6>KjS9N zv|U>T2~Az)Sndpsjn4P~L7M~)n!JFGZgkFe?qI`jXk z?MkDXI@9pgo~qTUHOi(4+JJRIz!Vn@1=L0sQ7eP8ifE`B6mVi8AOg~!VMswpWpL;b zaRI8e2pZXx)q_A1VH@2teOgnSt$BchGo&(o=zx&;Hd7kHe zzNKt7a8P6Cg+kt4`>D)>LIA|xC87R2UNrc$WY{FE>M~WkF8y`0Nd&sP_z%18i=YT* z_!1)d$jd6N30aI0g(3w)Xtd_gPD19&3bn}GN~q(K(AaUHTQj0>2fdaBM5iXVc85y( zW-(z~J-cIjr!hS@8@|>0oE^_I{(6aPYL4QKHe(q7&jtD=fBFIhTJgx3%o1%MaVP?T zpzR77j|pL0<{T7MuWih+8K26S{JrJYqtYo1H{WLHgaKWa{a)6`lP(bEch(1Jj!u9U z5rgNhg)_j-`f<=c2AaKuRd7JkK$4`H0|jFuP2)omXFxBg>E-JT*DC3ws_@sCu z+LaBHt#7IpKi*%(fof<3;>$&tNlO7^%!8+y=-4BO&v;;i>8KXOXuYnku2c10St#;# z_eysHnVJL{SU)^~xuCpZ`_gY6v!OdIOw-u;!x6>p*;DrjoKwv?P`-)97-4v@y$iCI zWcW9&W3vI^?ktR%5*L=UM)zx6-LKzA(CDC*AAu@|&CUX2YS}s;Z-J+pa4w*QdFpbw zza*523;R%rzb`DGVqK^NkS`YZoml0z#aDBhOgDHuYUTug{NsTru30Q$+rES%-^(DX zSuFQ}fJLsUdiw*I#U?@buqw=FDZ15Jz2tGc=*OW^@rc&(q$N3O>?_02%CbMcFokM1 z8~eyNuleV<_FJV}+^u9ba37q(PQaCQJ3@eP*=Gyz&Wz1R@0HiuuHHL@;Ci63^69s~ z|LME!@gnjJV6(xe%AeUBZoL+;h{b}f;fjvm|1@2@2h#bp^mJ!l4GLY~t3wamtT*#& z$ad@BtTw$<^d@XJc?)3W0#~qhemi!4C&b#xgpJTAORoE8!&`lXP|}x-C#Pk$8!~M%KIe273*_3p zy4Z$a^BF&y0!Q5cI-fnB$4B|?zGsgqJ}5Qd>&uGFiheMeE6TN(97(?0;h!|4JHb91 zrVV%N>y-cyt)t`+kXUNA@AC9CnDk~g$SBq`ag)2yF_1H0b01W_wTOLlTvK@t=OtpX zxhq!@+6o@P;GtF)@W@a#fJ>U8Mu;8WIbJrnfceZ6X?R0oA}ei%ADqm!l$}TZYP4!^ zoZb7{%axT5FtK^i?_|^PmQOQ^j04{X1G%Atg_aM`U<991iI9Uo=1_gTI0!*Yd?_>) zf})22<2|Ac$1m-+N*5VEo+lg7XH*$f0*rr|nwn}*g3?CxJOF2HoS%b_&oa7x62N|i z$7sfpN}px;-eY*Ya!|c$A`+p})q@vJB2hzX_>~X*yzFE2Un3a5m`7DFd0PK%w#fo= zw}Nh1ls*Q!%8=x##Jj6WwvBp;1=t`N=L)PPTc3uiQAJvvcVR^AnXj&0lOV>`;4mU3 z!c7YBqFQ*$n8Hzc4*|mQ4ka(oVf@$xN+%C8NCJo;fiAQ1dUGsRBn5@846_?^(X^g~ ze#Qxg$;5xDagE~kMzI(}qLI)@1VVqqL6_8n6tXbv{zeYC&Jj@5hJe0K!b?XaU*5Xr z!6~UlYh+>AZ2G|I|0BP2x_8W6ieX4)2?ez1=_DlK791dXcugC0*8;Tlx#I~$bp^ z9d?d(1nl;uc=5>bGq3V8%-`PuBqI+x-x$;XQT!N;#lInfa;?PO)AJt~1Gvyqi8B{C zWEd&$>xH#k(@&$zunt3b1dZFiBRe?mw{i+nXOh_403y9gcWNhlXfq9UGQ z;8>X(FnK7qkK_UXtDE~3iMaD5NTh8}TfZ&$3zS-*x79}>tmrDU^F#i6$=kK7W&5>t z6}>}b^FzAspn}Y3heb5DM$c9|2#+T->>b6GN`u}7PgQODf$yCe12JNF@T|}&w+kN4 zLSTZ$QA_(F30p|0wVy%@?#&Op;zfm>EJ>Kc#KmNwhTiKK23d|XC`Bde@nmo!E~z=s zn}$sq$1@>phx@P-z-V%FQbP&PLqx#qtik6rAM;8l#bQkB%b+$DaqH+&VH+bBi;?7O zz{T?c&Nbb?KL;;|a^R72wbM!CbyjTlIn;N+CYKI3Fsr5O>iAwb_>3QWdj9kOk#L<9vUT?2SLFL zQBV9J9=$9xLOy&vQT(B<7*MFC8nFu|U>u8lA>zD}S9uL~k_LA%9a5p_Ka#39@XJ<8 z%^6+a1Htkjyq!vFMNn7m{4xUNdH+ra;4_;>kxc!9vmvTl56FJz3*RW-`sCtv&W?Wc%0~;n1y*;)8 zXl4#>(^Iaq*0`B83MZ&t1vj7-<$+Q_{D;c2Od;d@n`N4m@snjTg-~W9yuWdJk#{SG zac%KTZ2oxqmuq$*+EWx2i^*yj^xFfz@scG=gdnGi+7lYip*bV|O-SD|p4tas=k!j&&N1 zb{HIuE|?q|od~T+Z|G=7Sc7V0N1N$t!Trzytq=~Mf}&tB^6^x0eWfTqGc6Zk_TF9i zSOsDvKnR#fhG{4zB*!OVd1*z5B@ou|0}gG;n>Jiq>cq7Z)h?7=0#?>?z|Jl|9b2NP z#MAn{Ql}@x4}8%qujKp%SdO zYLTA~?29zt$SaJLJ%gZQJ_%V+iF?Dla02vU3XM17@syY`SRQ?>L#E%mVEIoJWx4Y7 z;to}hkQ;>n8axMkeMi`~0Z|Oj?m{5)*Q|$_$U}9EdnV<3Dm%DsJWmT- zIfG(|-Z~CdpjfemKXoT+TnS}*swUyNg1rXRs}2xpMvFDlRl8K~p=@#osAm~`4c_yL zv9h6Hg~HyM*|2s7@Qr1qkZ>9nFTyEOddE}W+$QUVNTJr$-V=<~OgVdAVa_{e54$4B0Bs)8gp#!+OwihF~W$qpx%})Z0B7S>H z)E`2SqA2$vXKhvGJBvJheHYLv5%}hy!{lJ_&R%pClL~bhoiSr0zMm5l2iQ^y$%tfT zqZksfD*07afOgEQ9ML?)G#5i^*DDa<4($Ks^)8U`$;1Avs3V9awSOFgGTrxTy16U) z0|po5{r1hNSeP}k4wgXjNlKa#1mb3u+ZPDrv@_wLy#Pv3!OPiqKe$0o`tf~4*aL!^ zi!zc*Cvlr;PYgzw)o9A?A%yB!`MKPm%=Y#j$Ji?e>ee>Q<2i&>sB`3Na?KTiMo7z9{D)Vo)3zL1N6DN>Ig;}>F7Ly&j7Lgc=T2e zLNCRU6#L;eGSQ@Alv>lvz1Q!|K5VcDW0E%T>NA20n))u3&v5FrVQ3Z$U8WF}v-T{hX?nhL1yn0Nc ztk9-^r%K8eD0X@I>*LHK@3hmW`_zGC!Yssz9YsEu(e%bDAGA? zadXRtj!_65k`?z%I~dBJVtBnW%cyigJtNz7D=JUwS`O>xlPuAqyZ2u`o`9? zwj6ZkhKS2azanv3?xr-$lRuf9Py+&0oOYZ9Yer;9nN5yp21V{|5NgPR|PaWS^bhfKH0;x_LVjBVcmZ3c@iIG+1uB(X;`d44h4G`C=pQ`aVVj>Sb0vkl5t*}J8(f0d;G}>~87cF6+ ztT%0m)&FX!_KFOX2JDmbR@}U~dtCR~?kA^!NB7^;INFasx3sVSywq0=c-V3cn4Glj zvjbBe>^(AGHX-$yAN=EB*IBg;pPZliB0Ihv(}n7G z>sq5bAHrpV@{qehau&;G;KI?#hus6BW~kz@;;dqRgWlu8&Rs(yduGm^Z}%xAxfN2( zuhY-!7|U>lBogrfG)>J-eA`G~{TPW?G)U&h!ue%ui8e9*OVz=?7KU7#mqS&I{~$2MJOabBpBFHen-o&ew8B;Fsys>l*+HsLzEBZI#yomL-ZKl+Q= z?#p}G!Fsz$&u|p=s9DX=O4>HV=`^A5j9Q)36e)_KF70ZUk`(8b$hBC|_;FXJgKlYN$M z%bdIQku&5v{_a7IU%@p4`7RTJYEP=^`7$w z6ZsnIgO{z4K-E2}#v}GJY{y2k2lrbc1#abL8jtvc*B4)irCRBi2>h_ZOnQYC67T%X zOnRR`Xm@i##C?s{3T5r$5CJe}Ug~!Yc*wWr27N5!9r()3#h$f}h|ybnRW&qvd3$e_ zp_QFE@q#{T0|hbtxlbXFh}f{>(!tP|W(uCxFp>B2vt!J8Hn+77A=n{MSxfR~HXhf)LrhSP0f?d@y0A&hTm!)663jR7*u zOSA`mG@hzDU|70#ZV?-a$6-2djIFz2Hj>d*_Y6Wf*c8bGZ$lJP zkhFkK1$fTwDKe}v&%V|p1gU=#vFWn+Bkq8RgtdY_l#OuYhZ4RM%xYE)g9H(9iA35q zKuBl|KnnJMgHNSK**C^E?%}a9n6mspojAq!))jmgRo6aiJr9)Nx0@Oy8Okffgdaxi z--r*Pu9PRSQgW_wQ=QYTE4sGk3BZ}9CEr9FA z#m$wm9XqwB6~g)Sao!-OGo5=Nl-L84sgDi2;Q|SaI(54u$}M_%>{;MD>ffiEHrP|D znq`h7Ao2RJr7 zgwYIRotcLUXGmGilT;P8cBHpmHY!8F=I9AzAj+H}kqpu9^Q~0E)G&reHdFuRKCL2pwwsD@l=`n9}qHvrrOLHfuX%< zX8a=i(L>`daB@uv)Fo>*@+yeoc@jmvn8#k^9w(*uRl^x#@c2)tLM=l@0wO*CM!#ep zDdyF&??R{0hK%SvV+sP_RHF6s&H-Y6?)4H@1P_~Y*KSpu?OTyUTZ(zD6J(w%eOFc0 zzpZ~0a)jt@ba*+Ch(6!`Ts!`431^Q|EzrF3^*Pi(Qf;a?zp*-6_XVKn$y!pqI+@0R zduUvsw6_YMEEA%MF)X;^*mm)wj@s|S83hl5W|Jm}nnNTriA&?KTBAlhxub5U0?9G3 zW-Y1oolLV8>n{rPQgyd=iobNcWs%j zbYXg>>q?bfWy16h-Blw0{@q1-43EbvJgN_%E?Dzki6qQ7f_i|Ly#?4VBpvX@^iqB9 zLOKAfb{8syUWXc=Rrx=8V9piq(er&=}x zrzxP^@IshB$0SnXzj+(1ifd$>kM@y;2T;6VhzC%dm;`=Jfx!}4M1kR53evTeA+qqv z61VXWsYu0EM<o zg^DyXtbaoV$ShF48Zz;go&Bv}r6k%`+=C1tdlI;u0-yzwNdd6CEc^@d3iM-$2mYyo zP$8OP4zQau$D0grb^>tkH-PC&)J7W3!P0aW;J1VKm~N`iUCI0M1EFFY8Ek8CFjbr5 z$ow>55M{w=swj#M3!!2=MTaxTmu!$@0#NbqDQF4=<>BUd(~zqaFln-S3YeA*u>%=Q z`Xq2IWi|eO3KCFN+#-z?4uW9276^!OiO zwv!AyWU6nkzUa8#@Oot*Z|`IeOc8s+q4OGyHL;QA`gXDn^vlB_wKq|EDTCT!P_BM2 zv79T-L`%nzM2X-(wx;vOjcHF#FC%3yVl(WzWZQ8=C&sk$Yg{{+jAqmBdQP@IqZr$0 za)>f$C&RXhnu!2jW+r-@EK1~yy=1pNQ_v4gS#2YEkieqOo{B~3=VlItn>nN~bLb4o zp~3)!DI&`3-jk$3la~w3Og(94>U=Y+PLQn%9soxkAh4$negr>?yG=OcLl!-ghI$=9 z8yd1p`ex3}Bu^3~!z9ca;s8tyd9Il0PSy_;V@ZH5y#4LXDC=*Ev5+gu#*wBd@FBr* zc`%fG5KWF2__gOcGR?FE_k)$%y=X=>D}u3AzIQ~Adn`ib$Q_vm0<70Q$ba-CQfUE8j zXb4M!XmlCir_mL97gBEzPdI73V3U?{@nQQD=)%Byu1bOAs5Nf_IwMr4eT ziFWoX&^9b(2`V$Dm#D8Q&lqf<13d<+c0^mznD=s-Pnj=U&Mjx@e|hK%0TjN zL;`zWWxN9y$!~YA`o3g&w}di&V`9V6=(WS@Zhi%GYmiO)S^RSn2fx}pdq~iXVQxaZ z7Xx+W)@3~1ORr$;^d4)ywuvpT$dXq1D+Ajgi8Ox0a7g6Wk~Ug|+rQy9n-k^H7X`=NWwD-0{VMjdhd_dOebhxR3vV4SL~p&;cqkS15LJhG!)}hFiEyZlzCEUha$>x&D?S<4lu$vT@QPKnNty zWEow@1vyn;_aa4aFvta)D<9-K`Vfg&_YglA&c)nZGddtx7XorGorWp}?P}Js<)jU` zZZSxYb;6@7BO&=jKC!FG+f7>~dCKf1!z}Bbna)yndG+Rgcje8c%4!RV3_FZAp*id8 z_i*T$b7RRlg^3h+!(q~dPkeG(^ed3~?VXMv(&wC;hLnLKV`3h$TUU^cnL-$2t#rcR zaDAYl?wZuIGfIW{(Hk5?SX%Sk6QJHW6urlC3-;>wG~v zaqzy9ob8BuM}6!wD2$-?_s%s>{HsVdn=088v&8>rIX)pxWe8Um^L&XkLoOIEG6AO+ zXksb>$@|n5Qwa$ngFzI^tK^l)kYRk~x4is4r80 zzuufS0G?+&x|2BWC-WSIJrL&0LA+sRRKY)>Zx#}LbKVT9H-Ej^a4~ia$NHJ0W=)`0 z5j*mqhuXVdZ3`zL%+t|(s1(ot{hMimWFR0Ju zsOaj333MRLXm6I*e#eEgiMtNea@PZ~r*K*QW#JSOdmck?XEM2_@6u>WoUjHaqa2`8`l|1^Xy)n825aqzbI24p>0oeft*+l^9pzt*$4O7 z3~afgYe}+59FvBk8H9BQ%CL54XmY&Z)(ZU=X73j#beg0-?Z5Ij6bb!pVTo)ZZxSe` zt43HN1xz8@>JKh+h7{JOFOoj5MTsx_HUnIOn^yzubAyLy9*w z_5e2Y9kcX-oBI6g*-O|PmN!_uRH$}|UTMiLk5>-0y{ zh_3t&%W0tif6m(iq=m*jwM4Qnbl*&ft)Bt5SLJY@5LY35N=a@{@h@_fOx(u@!Pnb}p;jX%@Z%ZfMv@?t?`!bY9 zxhf9V3MC)!OPFh9CnLyCaY~A>^Zkr2zH=(A z{)tj>*0)Fi!-^XB*8?XEwJgm*lp!7T>$Oqy6W<=IQ(|7B@2 N0s^O=nY?7nzW}34{C@xd diff --git a/docs/articles/opt-loop-invariant_files/header-attrs-2.5/header-attrs.js b/docs/articles/opt-loop-invariant_files/header-attrs-2.5/header-attrs.js new file mode 100644 index 0000000..dd57d92 --- /dev/null +++ b/docs/articles/opt-loop-invariant_files/header-attrs-2.5/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/opt-memory-alloc.html b/docs/articles/opt-memory-alloc.html new file mode 100644 index 0000000..5796606 --- /dev/null +++ b/docs/articles/opt-memory-alloc.html @@ -0,0 +1,255 @@ + + + + + + + +Memory Allocation Optimizer • rco + + + + + + + + + + +
    +
    + + + + +
    +
    +
    +
    +

    Memory Allocation Optimizer

    + + + Source: vignettes/opt-memory-alloc.Rmd + + +
    + + + +
    +

    +Memory Allocation Optimizer

    +
    +

    +Background

    +

    As a general rule of thumb, in any programming language we should undertake the memory management as much as possible. When we grow a vector inside a loop, the vector asks the processor for extra space in between the running program and then proceeds once it gets the required memory. This process is repeated for every iteration of the loop resulting in massive delays. Thus we should pre-allocate the required memory to a vector to avoid such delays.

    +

    This Memory Allocation Optimizer checks the code for vectors that have been initialized without proper memory allocation and whenever possible, changes that allocation to an allocation with a fixed size, taken from the the loops where the vector is being called.

    +
    +
    +

    +Example

    +

    Consider the following example:

    +
    +code <- paste(
+  "square_vec <- NULL",
+  "mul_vec <- NULL",
+  "for(i in 1:100) {",
+  "  square_vec[i] <- i^2",
+  "}",
+  "for(i in 1:100) {",
+  "  mul_vec[i] <- i * i",
+  "}",
+  "identical(square_vec, mul_vec)",
+  sep = "\n"
+)
+cat(code)
    +
    ## square_vec <- NULL
+## mul_vec <- NULL
+## for(i in 1:100) {
+##   square_vec[i] <- i^2
+## }
+## for(i in 1:100) {
+##   mul_vec[i] <- i * i
+## }
+## identical(square_vec, mul_vec)
    +

    Then, the automatically optimized code would be:

    +
    +opt_code <- opt_memory_alloc(list(code))
+cat(opt_code$codes[[1]])
    +
    ## square_vec <- vector(length = 100)
+## mul_vec <- vector(length = 100)
+## for(i in 1:100) {
+##   square_vec[i] <- i^2
+## }
+## for(i in 1:100) {
+##   mul_vec[i] <- i * i
+## }
+## identical(square_vec, mul_vec)
    +

    And if we measure the execution time of each one, and the speed-up:

    +
    +bmark_res <- microbenchmark({
+  eval(parse(text = code))
+}, {
+  eval(parse(text = opt_code))
+})
+autoplot(bmark_res)
    +

    +
    +speed_up(bmark_res)
    +
    ##            Min.  1st Qu.   Median     Mean  3rd Qu.     Max.
+## Expr_2 50.85415 48.00791 50.40635 50.52503 47.73553 136.3603
    +
    +
    +

    +Implementation

    +

    The memory-allocation-optimizer looks for vector assignments in the entire code snippet, that have been assigned without proper allocation of memory. For instance, consider a vector vec, now this vector can be initialized without proper memory allocation in the following ways:

    +
      +
    • vec <- NULL
      • +
    • vec = c()
      • +
    • vec <- NA
      • +
    • logical() -> vec()
      • +
    • and so on….
      • +
    +

    When we spot these improper initializations of vectors, we change the intialization to include a suitable memory allocation in the following way, if possible:

    +
      +
    • vec <- vector(length = 10)
      • +
    +
    +
    +

    +Limitations of the Optimizer

    +

    While the highly flexible and accomodating nature of the R Language is a boon for programmers, that very nature has limited the scope of this optimizer to a certain extent. In this optimizer the following points have been kept in mind:

    +
      +
    • Only FOR loops have been considered, owing to the fact that their sizes, most of the times, can be predicted accurately which is not the case with while and repeat loops.
      • +
    • Since most of the in-built functions of base-R can be overwritten, we had to ignore the cases where any function was used in either the body or conidition of the FOR loop.
      • +
    • Also to be sure of the size of the vector, we have only proceeded with the cases where the index was explicitly mentioned with the vectors inside the loop.
      • +
    +
    +
    +
    + + + +
    + + + +
    +

    Developed by Juan Cruz Rodriguez.

    +
    + +
    +

    Site built with pkgdown 1.6.1.

    +
    + +
    +
    + + + + + + diff --git a/docs/articles/opt-memory-alloc_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/opt-memory-alloc_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/opt-memory-alloc_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/opt-memory-alloc_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/opt-memory-alloc_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/opt-memory-alloc_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/opt-memory-alloc_files/figure-html/unnamed-chunk-4-1.png b/docs/articles/opt-memory-alloc_files/figure-html/unnamed-chunk-4-1.png new file mode 100644 index 0000000000000000000000000000000000000000..d853b77aec3f7cbe53d3986c790157c649b43ca7 GIT binary patch literal 45642 zcmeGEXH=Bg76poy+UQY4k8MCfwABJ6t0Yl^wy{V`&Y+Srl96a`BU(g}WFSWgB3ZN* zNs^-^C5b2?K{DLA3q1FZ`}dAF#vAY58si*ksQPMud#}CLoO8|n-8!pyie(Mk8VZHN zLOXq2nL=T5rBGI!|IbSNMsInoKK@yGLGILX{EOGrv0QKbv)b;omIH;tag+RSnQzZe zV+!Rj3hnq&RoBq(ZKk1WJ^9ju?reISn3&m?-4lJm?ypHXtoih))UV*J8>N&C>zdAp z9962;RC+<(|NhLM3SQyIFEH_oY|yD|k`qz?Z`|z~lLdoemz}A*%gaNgbQ=@Kze~9% zxv$@Me?oFpPc2G0ZWCiTyoyfM%NRdt9#}($O7U3})yYJDh&}uNkzZt`Kte`e=p6BjOyj(F!WVf3Jv(kC(lqgW{A)A$+Aq##^l+n zG$tXD(2$T2$FEgI>$#p>cKTYii*~@aT@t^K%*n~guxwN;l6!aHDP!Iu|K@RJ{J`sf zg_TU(|ND#oFPV?}EgTe|N(}^sX@sX(rR1o*^z86kRSk_kOM@BhmrWP$V1=K2#b%`( zDcY1rsW=;_TADx6{`j#?4v%f&+^?8$+xmF5>G6S}W9uZSWdd2c%T{BleR_`+F;|tB zhe`;u>bsujk`Erv-YIPQF?4or$h$ID>6v|td5m$_J#m{i*T}UUo^&#qoox4c^84jGx*&+8o;v`l**nTD5O z137%Y&KeyZ&GpH)&{OFr8TxWnR!j=h9@dYC1`_SFb8BSzdc+H-6n?cjRGuhp^Bw@`6m$>i`Y?N4WD{w3-)&#cMeZi3}vBe`GC) zR!dOdCSrEht#Pn5+ri~x4gz(WSkG-zT)`(zQ)^0oSJWu*plVSQ) z?_=*MhYX~OtE<*4<~skNQ+*{zPG|2UIagbZN$AMs86$29jlJYIfmD3 z7)Zs0+RM6mwmCH#f2)gQp>H6ADsBCU#qEhUW zl^=Dg^Xhv0x$#Ev(~9^=47NwrnuFFFfy8izSUN+NdkMYKQB&tD#v)JQ4`bvVno* z+>b1cMe?n)`|PcN-UF~r!sf}ii%e|$lD_t_$gAiic&86Ku% zlO2^eKeRAEq^hPSI_ZS5r6j(1&qkvM8F)$H+s5GjIO$dhe###(~BZ z19ctiOlrw!rA!2Oxa(7;lvQ}Uh!mz~qZ{I5WH5J#ize1$_Uz=R{FxUBjqR7dKIs{M z({q4a_5gw*LN4+FN35+v zCw5IaUrHXGI@Qux5ZIpYE?(mFDZ|_kn=&hJex_rgo!Ur>UdRn{kFCz1tb4jvte4zZ zU#~FHzJ7r*=~IoWxYM*sEWlc}%NFSD8f z7p%_V9~ml)!bbMBpZ-qr0ZH-!=e{tP=z;?2k}Hp1{bzmGI&i)3vslzUv6j8~FB^JyQdWEE#xs&ir`O@Vk>@eZY}J!LyM(P< zGN>h41>;GMD47qotEs8o@h@{}y?APu1m5U#16AGZI5TV9%=h#&mud7i=T8FGpI$1^ z$>1MJdtQ3tM&1?`gSSyPax0d1`O1V-`|ILVrhom=F$;J8(PE5|E7Hp!iwhyH`9;WB zQ^&|yyRl}1JBJEp2ICSE`1VyScOS}knVQ6J72{NS_py*SW{@{V;=bSBTrZSTI2tyz zDY$pRDx>;jKbC{ujrA+LBUGSi^%vfKE)L7gR)pJp=+faSUUEe_6PbdETRSp7o)>GY z;U;zA-+z|Z;;j@pIRx^5wzTKx_JzA|3GS|)o*0Vw_KnG4ekgaSpginwRwsd8uj$ya zBB@&GCdTE#`)PFbB<%oOs|H=NlXPtJo#;lr<&WGZKi_0lBzu*g2LZW)UD!k(zxJc{ z*T$R)#74QY(NlY2ey*%|qTSv0G7WF^i6ZI!9GQ`wLb}hzxbxj76tUm@sAT7~doRpg z*+}q#+}{|c57(LSQ0@qi1GKa+xA~$qQuLX)r>)2`bQE~{Q3>v8dSSRUg4JXMIha+X zLOM0~dzu`w<{HJZDO4m4=>q&=jr{We4ORQqUcFaGtnXj?_5G1z*db0@xa*f&c^D=W zH+fS5FS(iu(|0O*dSPKM?b&%^(d44McgQ^g# z*NbXwwuF970O%{l?W*d3ddG}61{^D5PW9(?n;0zK zBIwHMC4=7`8I8vGtikv6aHK@EuGXPROGKal^6#z1$=q`D3hOx@$?-ZBMkX{$uMm^# z`xwTom|=u@O6$6RII!`(2)T|i>Nfv%B>6~>ywQ(`S%Z!G1@dNG5rK;nxFrZzT3rz# zQ>c4QT|+}CZNA!gs9r>#H8T|4Q!Lr*$+?C08@pU&UpC@o$b&nULH=JJT`l~sTi zS2l{?y(iv&kS{4incht>19tmj@Of-t5_iG0}j}MUnw=w z>K7B?btp=?U(O(XcBIU-n8T&qrC>BHF|CwW;X#@)38|@9QTI$IHL)9ot(sD)C9f`a zA&gK)dL|Vi0jGAjQFnTdDdDD79QnzjZrqf2_PlJAGL|dMG9_Qk zkNb>%cXoifHV?U}-p97#NQ{T|avTYuAUvgyOj7L`nXvta@4Kr{?No?UmG$)W9DLPv z|LzW{Lkn{q3vCz5N^jKF6bpMK0+LPr$|?-ou;-)^3W{6z!&7NLUVhrbK_Y+h4@}Sl z9+l|s8abhgYpXYIamzEU43|Jbs7=sFQPuFCZoh>RRfdcaS|Rp{|K@i4)9sfAl3E>W z6|%=0bjU&xjK!*JcN9SM#de4C{pM{sPAd6Pw+WOQxhThh_2+PI+O_B8-P82D+k{UR zk#DZ3ih2B3q!tSMP5U~fQv%9^UZamY%5Cxgx=!_K%_fhnrPEvI_yMU1;0;CQ<&<*& z#V;f6F*zLSSVMbsz*zQbpY7`2ipN>S2}E36?e*2~Q9E1c~S8ts!=$ewKi zatnEETOfg&TX|uqX>K>FvIMfHFqp^4y#_AD(BO!L**>??3h|)nS%DHjjK}^Ocf+ie`g#tzca&k^erg(enMEmC{n3wmT?%LP z7SR<7aFB|k1Sbl))v%@7Q}Ey3qE$QP=^(`!lSm2!ZMiN}t5&am>ot)bnV%eKBkSvFg_M;+PKO)8I?_hbs zKZASL+EJNWudWl|ayZK|?2uC_nE)zDk_1sNt_d@Ux9})1X3xJH&+re0>Ycy7*&cOC ztogJL?nU(2&25JUq}^ZUIQup73C z+vw7tzkDg?IX%WBc6OuAkUniTtysL^)&iMy~F+>t*EWW8doeko5QuD%EORXV#|`tz@Ta z*{vu5rp{gjB*mNZs@9UKz_>nvZ_K4_BnKoUrpZ3Z#=Rtv$V@Fs`z-kmbY0ZSDsy7T&S&*7p~ z5~Jn980-+0rjF{Q>p5QE!yYQSibV8}jM}sYZA(tAVZQ0I8p-ybUB#wQUTE|mSsl_y zAbbav_LD@uL+TRdV9#sucGqW&IXLpP^ZFY=H1WPr2St5E9JOQ(U*dy=a)RM2&&x8T|uBeFK186Sv$O9za!uovMJ_c|I*~uidDN>cUXQWK1 zk^x`Mq4Nk9&W;dB+-VeHo0Qp5TIuyGePlX5A-VM5TN|0>>n2UggCCN_OsgsIoUuYJ zXX55m$cQY-DW2=NZp(8yA4t1FpauZPFg!4~CZ1QUc z*1%kHjU!bXk)Rwux{AY}u$$M%4ccV+l0Zo%qVfvnrw4@1Yo0IPD0=_aM($pNg=r1? zbHL7K`!WG$#X8CFC?h>x)z40`1``x~8X;;#=Eq=uBjk;E8z z;B}gU;W7onpe2U#MoWd|av_Bry}DA(%3KqApUN32QI127r|FWQ#SBmy#|AP{`r+Z8 zvVk_2j{Mgm6>?Y!RAE1(qP~)f1`Ib+GoV@U==CY+FcHN&y(JkdI*Vd081yV4UM@&b z0%fvv06F6j^cwzsN^jk;7^BkzDcrOMa7M@&hnTo6^tbG&jby;h3)of<@3D&uX0_VO zxy>aQ%<&>`3JIQF(Si({jylCjgP<7LF*oRf4{k-3+q=kvz*F`qj1r;3^!jFX+{RS5 zA7NPxY<;dds`L8Ev0tBT`<~tr!2vyGx|YzejK zCoD$NJ1ZyW(A@MmcGIYeii#him5YYty~yXmx1T&@Y^88_8-g{c{g6oSANm%H-IABt zpv{Mktx0DR)=B|rEC=H3RF$)to&4FOv=M|{EyuB6+3snt`RDyC2wk3it1+MapRoR? zlc81Rjn&A}=pVxlW21hIYkC~-`DoAmgS%0m73g_j6jQ}A^GbG4IbERc1o=)tOVykPS~et-3=Y=yjp#6Sjg8ziBJhC?ZVrug`vfFgTKw$f zJ>JFp&&B=w5K+!`xHE5IeulPXmThLD5S&g$Gz{$Yh3@V9+a(U1ks1eS+G618Ik&rO z+pl)_q2;xzH*7rzbQqEZ|JK@lC?88Q$HMvX6lw`_Rm9~Lt6nx`=SD;k z8R*MDdrAkORU8=Yt>D{N{l$H(`jjwe6tzUW;OD)i5S_QMv`l}y?l4vB>;qqp!!H@x zH1)fda|pMY5wNcA{P0Z~|3=d=Cv~#0=bm05DvFljJyAL}rC{3P1`9i3UJjsQLB6Zj zZasUBUYa%>e1kBR*Ho-gs+BVIaYlW>y^qW`AeV>}9kD3xb~WzKHv z?ftkEN`buSRP5OgC=$AMPfK6)1nNYiF4O2S;g{lf+a?pCCjhyWr4#s}jNlAve?yWE zBb#j8JU}uFV>=(6YHPK9P@^0)bINlc79vqnh#@RQpK)9iug14@ImOK?S;N`dHXhajH^CD|IF$#Ux12nP+DQ zisL&bqK_IsxWXFv1~}Coi6N*_X1dy!RPIf|u~V9;;=EsOSR%p=NJ^yk*C%K!GT72| zpf3~PB}KX=zef44Je9>Mr5j1 zAWTp=RU&CYgbQq3wcYN>WpiE)C#UhhPX(wQlS| z(5OH#3K5)!+OMtO0c>7j;nHX5D;20=;3-P0L52nxNoYE<*@{8m*%`8l0^(5a9Eb=( zNN%337mEX2;?rWorQ8EuTs}S$<|cp`#H1*0%PTBoY4NI^9(M5wo-|$M0dO(5ymth8 zc`wX#FgUS~@l~-E1y7aZ{C6F@;otYt=Eb`$%@EPoDAwnN0mbxJJvl}?Iz|#=bE2%UQZ!+-E$?4Kmkb2-)Cq4 zB-HS})UT$IX%xJ2LMr9`XZz21^X}XUsn;B`{bh6Q8$^=u@IX0=Dkvym6vRCG}Cgt{XFoL)EeMoHUSNiNmSvAJ{At z9}>n)JK#3bJ+h9BF!JYK*Ux;fkONt_RqQ+W+ULqfwfo<`eG4TLGWuI||y2- zdu*6M-}Q&SMrPP{6g=22pu~`p7MXdM%J7>DY?vIv1gQkLc>TO%u(e`%GtuOMPmJ6= zOE`nNudbIit66(SX3lYk*W_m^t@u8NL)Fn$qW$AKIYcPY{o)H6xYKL$JfqY^8rLco zT?htsxv_Sevi?pmG%Y|kW_i`@b%(#FTBR1PBuQ-n$h+RCQ7}L@8xWyv01&oyux(`n z1b!>zXd2yDdOB9C^(;Xx+6S0~_-_!Pib2ZJ35RON24YrJ%|=#oK#>n}aU*FFdDMb# zNz@`z`TeL!YWDL3h4TrtBciEWIHXbFPQ)a;8~5d+2mn6_1x z!{$<##y-4fw*}%tvKnnN^xe!zmD*fPLLN5=6{O1v(x4mAx1-NPhK^l&VElWt!$Y^c zV>i~8?D3VX`Kylv1HyU~wTg?@04}lJsZNQ--lm`x1x`=e=IfUohz3ny*oe>2{Z-~i zBNkAP*6hGE0}M5+yz==c&;?Zd$=i+mHz; zxX{7sv?owZvM0aXDdYTpbEj7@fuZ2Yof^`OC~x+_41s-vgD;AANObzFU=tC#)0f*Z zXEWsr49}oTK7lk$$M8=x9}I7HEJjA)U`1M#=gcpXWwX1(PSLRytUz^arY4{HN-tuQ40$p$zP5=C zQF1{fpLB|ph2Q5mY=WLRCbsa~F&lw9)dqM0r^xLT3lW%v@`GfCykg2{W+Hg8WKc7?-e8{-w zIgK>UNPr0_&ksWWC*!RFI(XVn8gWOSp@mhUj1IrobZrDty9v}EA#htORb@iRd#bFj zAQ^mL;`6nQX{ZQnW%v{&!*z{*q}BjNXOrC%xRzavpRBz;nmd)VuW*OGhdZVeN%A5J z;4KxRZf}C*DdG+~Fg55>_}KGTGxq{+ex%#)OfM+tM^N6HIF*>htoBRIX3y@Gw~$s( zpaj0=0$N^CI$1UZhg2((Zx6Zt4Z0+Gr5Z7$KMG|tlyLHkvrqp2^B3_izfxGt5m(tW zgSmxcU1F_vZ$sS&vlSl1_U8i?yG{4UjDFbT+YJCx3bDNwb8Ypq*@Wn!L>Ji1K?He< z*di)7JKT*8*c!abkk=di&CkC|;CD@A&M{SX~3IcVaAW5hm6+hJO&1zEZ)1U;$fLDQYqgt1t&_6<|;-cB){d|wU z#+6zUUKBaYW~;wtJ`)`o`_#{1?wdx0md8gje)MlwuHE_ow=6i)N_HEGK9ZLSV7G4loKYhC zGd^r!F;E+7&nKFy(>BpmEV6DPd2h{|$7c%Sw@isoKqo4JmY#k*Me1wh)dSE5TcLE&=quJ8{0q_iE2tnD zdoE(i6Vt?oQ**%)k|%*PrqUBX7xsBI2j9Fgb}{=f zG(*rY%ZF31?>x)NnRc-K7mdCwO-{2oRpr3tcZNkT_m+2@F>@{0dF_2y#}*OEh2&UO zxq|vkrVyY7MZ`a~1n4{f$&_JYn*ZtR?a^;3)BbaYhI@5tl=Lsn=fSgc&XBNS3iPq}cVu2HdNmQ9;Ij7mMgy5$dc9kT}* z@$Q|vFFpO0$kWJ@-7(=aMeDsD*tF#cygRHdi7jGt^6u7$REF6HWUX?piKu}PoGqT2 zGEzO;nLGThe*C}ME)5I}IK1CxN~ICR77c6~F&(^Y#Y!WRK0#2^u`x|BDYJFOV?REL zKem3U1(CFnaNI|tmg>Y`w%zXaGA$&ez`+Xs%gS-x_unyY?oDRwl?iT{DVo=aUxY^# zWKs=A!g=FMTGM3XM)hf<4d#HRi(rFO;r9@^3lJARA_5vZ`^fZDLbz9GGm`cH={n4e z-M}L7E>SC;6dV3u#om2~J4!a3+Ft$7E%FJU-Vf`R{GGcq+ z3!a{ymJi;yi$wfNcB==%m6F25^RO|R2abjw6QRZH_AQ zxRR)>enIPVnqO&gW~Ns(5s+Cxz9WW_vkE;V(MipNWWf}zsWn)aAP_Z30(qK zK_a@v z6jB}*tcc*oh=<|n8s0`S(_&KxeYZYTGf4SmVwc9g>Vn+#jAOGtgydiy_fZRYmY^wi ztJI@(BZ#5|w_vFYRr`#8q92j08*dBchOm}Jm8VB(_>HM1uHpbd8PG`Xju=jtZ_Lz9 zDV)Fb5whfnU_S)2J%I4GS3`4?lLZN1KIHiI0rkLS_1s+DDqX#R4PXA>k78WZQwt>b~M$l7+6Q$j?i)aJLUKXjA2oKKU?GEy8 zp3TNV+v=1Uj!Sa0)se;1a@!qbn+z{6YGc%=NJY zsWcdC2;7T-4pNj-izcNNNHUEMBUcY>3)aJysjhoq|=kb(ZWr-6oJU#oi_$EzXhX4EQ{q2XAyw-NiG^bGRxqFRi2?8fe}^tXJ` z8STtmnbZzq!Yof}ssM0m!M|Ada9fWqb{d`7C*W{&D!RqnfnJ5MUFYkO=?BNw@p#{7 zN5@Ie9-hWPvLLtElSS#ofkO0m`g4#@Ey(k?Sx7n%2*yQ^gTIpf1!~PnWB56Uu&)}B zu!Qk|MCeo!5lb4H6zUi8uF&Yj-wY+qmOdb8NtAttg@Da5nm8s?k;wUP5a;X79hc19 zJRxB)^36V_9=81#j=Ke3goNfm_!ipxv}8REjUWhSd3ihFYXwp3pf>8OLto`aPv;Ds zNoIR9+(e6x2Qm{|YC*BmK;5`#NLbw`^%#iiy#1Y2knY0kzz;c8}4DR&1c_-Dc)!@TrHNX+!W58J*E-sT$#d=-Yx^ zoCbZ8n+79?nd;jM!RU~p`H{y6fRb+sDf*h>2(0a{otJyYk=q|2H@+Fk*hEy%p&RYk z>a~p4FgDwfvKS097G?w_z?$997Q0&Uy+4RMn~Vmwdi$>bEa9_nPwwJ8V4!`-6anpLW~S$ZN&@`{0=F~%i7`b0d#8ybt$kM!(+%7|CaBL1H2qTcuwIc0au-RI4a;YWIEFFM0SVQAA zg$ifw=V3hI)*7CY=nXau8JBOkUMxadMQf&0h{FK|o+kwnt#^*-nVP>uI^Hw!o+jxA zm@%-C@JW4-pYN7$+zh`<&lP6Dy+O~xA5W28ctyqYMGDA^78;>`R48(a@9%Ty;J$x? zN%oC58l0gcL+&w{X=z;C=w(}Q^5ucF=d=-8$k1Am4NOBXo`iQ?&n`aN1*_GAWGP2a zOntLP0*350iW#oXfFUD4;f&KpXOSSHkr!>b7-|&^!&N@_aaWAgwWZbd+hX<%X+VSN zft*l&2E!{JypH40reL$5*6?2UBpMdqcS_HDca}F`3y*77hu6$fOAKNwODvE%sYRuu zCPY5HyImr!Q&pia#4@=Ag(rwQ^Q3&dLG*Nkeu0spd!TBBw`iS?BY|1kkQkHX7G^B z%U_(J6-v_WHn~3*O}S-)n6oPYb}|Gu(Yx_eCGqAk&L_qHMB|yTMn2Z3#LWx@po9x1 zgS~4(r0}ESM%3K`_eH4l1}xeS*Y0(FvVrgfa;YFi%IPKcQ2V#g?d3-_y!U2VhDsCg#P)>mFJf@j;;1428Xh=yf%)7g9^UH>b1m9MD2M z3)qhyVepMj=t|zRT9})tgpY0U&=*My0nF91;^*I7YenmWMu*SucHWB?Q!Q8dx5Y!g z3!}uX2k3d4o(P{qKRoq}O0wv(AW9AK*!BGR9dG%t%`{Z0V3$_ZG>(%Ufazdl8ko># z|A`J_wS+>ji#CG4EZT^bMlVbV;{HT-uF*GJPLwSn$mc{DppD>Dz0kdc4%XK&PJpnQ zBBcp4+JB;<=xu-!X5|&2slhB?J4+;#o-XT`&C!hBx(|$Hr}1_Na%YVM<|1rfF5f7S zO(caWvtc8~;?iK?O)Ly~aF)$17!I|+bxan&yH4r(-uc!Esmfv}rymmz9S+`wDOxr436*&=>TZuh zLla??->TTXqbZ+3I(*#W7S^?@iad&@Ll`wKGV%xrQx%eD`Hh@4a6qJwfbXk8c@brg z^>4Bg!(qH4ZQLJ3C9yb7)`sfB+k4&XwM?jIhWj7#u}iq_kub&$g~ zawmm}182>D{Z8>PgGQtIjj#2ImiLG@7@&&jvYI3{J3UE7o@igmm&SP zx5{}7+_TWCN^7=S$!NG!D7UV#?u+)A8bK0>e)HxHm3DLe&IV$^>gnxeQ~V{T7^4Ea zL_Jh7P8yCoRl|3=Hpc!Q468bg>|$2?SpQ}D_uV_K#_|q^2Jl)6=JJ(lm>yx1^0Vj8 zMN)ZqcxYt$-rZqk;m)lC5_vW`=}J?^-@?77&8McOwkh_op^YBUrayM72i#Zf?Adb% zA8v!zIt{o&rETU@uY=5B!f3uxC=Xyc@A}q|lmL{gdA%S0A6OyJWGj^9S~dz+JQlCi zhWSJtqK|0shs@F-p%i?j3f=MZ8zv&1+3gla*NrQtmWO81oED{aX|b=8J}K__nnlE> zRg(~}?yjyqG}+Uq?e2u-ai0mVgSFY3L$D#=-GwZ*8P~?!zUED@EsdRZK#iT-q6CPfWwz-t)R`+_+H`+N-gtDSL?nzJ9Jp$Xn^sr5q~Hfdk5i2ojbz%dAZv7BlNmGG+vG5P&N^>b4zzTiBlJyCs8SqXntHJjg1=g4AaUT zQa~Vwyyx@e9J;%^zoN@f1@|1423PUbaO|oB!Y&iu0el*YNZvtI9zH&~2(Rh5&&QJk z9z2MA^X6~biG^KXVmcxiLw}8y=Ie?Ki+Wku`}1b!<~V5pXjSk>IH3##Ny>J#$I1c)ShyHE5M=?UChU&kv`iLg z6B8K>eGEYUD>}2aoF{b5KYaYy0LRjK4r*^-Ukut6@yOv()4FHuKz1|_2WMG-iq6ZE ze0xiG4>$Lz!!CpVRHl_HwGTgHT;9hWhPc)`s0PWXV6n0Ry!6|pjqvrz;&h#0z_wKQ z#_QqDOVUd4^76u-Su?7>U|?sLimspy?L^_0TeoiA*qHqu5tMWQ6m;mNbKqh$OrU@K z>IGo%H*gd+0JX%%Kp}%C@; z%h5eNJ?j&-gx@k0HOlJ+oJNuKX|rjB05XOm#JhX<-}G*~gBkWcvXF>8Lu9-Q)T;Dv zAq3w2`|nqv&pNoCIB&k)MT^(VeW~v~=XB!a$-vTy&;3`fTuG!0lbLt^zArS^(Gwks z(hnaVP(>~3PVd^a>jvWx5yfYr$MO)xhp#d6=n4VN)FZ?qH;(dlYTeCa&CR6mn?-De z3D}Yr9R>Mp;x>tx+ESZ^slw0QJw5#pbFkOW%FEvcgLf#&cOU;21#wr{aNH%s`cp$l zVJ@m0*?4wOYgA^>76uCFurWNa6pA1Yy$tU=dh{rpg#B~)o}z)nbXKokO~38jW7AQf z2LK*KW%}cfSo@yRk%mtjwYKVVa&g52IYt`>yg~<_7@_IU4-dXN)>9TF&M-t! zD6?$XjBo2>m5$+!_~)6wzyI^+^J9-1*{BGi#3wy(Q6J3Ffyr_2dkirn}>qsE;r z0;-u7YPHo;>W{^&6T@l|iN8v(xL`LnIN_Ma+NCX7h$BzKe_}saG$cN9`cE)bB1}`=EA}C8u;G+qQ$Gk+~2)>_wT|Lgl0IUF;d~NSi?BD>sl@zw-+y7 zG>CR>V_e$K7Ld}8#fz=>uaZM{6*`-QT%ko+Gh8@06C%a4B5O6Iabti za&r%$`q6K%8^#TWG#^McKdd zqMad&y#GK|B>V{k=->OB4=z6nXC&gaVI1=B1;G!Zd!^N%g5&)D&a3#&BdcMFd&s~L z0Cv?#V=UY`_Bgwck$k6e<6?;%^u?fMeNs|VL~|QH+qssXw}e-$_TipWRGM16suHI0 zfA@Ely!8WIHb#lC2 z+8^9Kf)lDyu(H$WDCd#vwt3aRzf*wm);0j&v`=^DUE5O6UmY18W#OKV`qH)SHY=k8 zLV2CexX%RL>@;Leep-&*=PO^jx-hS*&dx6{7#ecY$SStwWPU!rc)!|+-y=L3^s^yLKG~;<~YM&I*gr z;Pi7hA@LN7>kh`-06IzTrj~rR`w9*q#{PAIOeKv*qd&jBnJ*r|$6Den!>bYSfFesn)R7gToj7cYBMc7D+_3!ww^g3a z^H~goP78WUGe)tOq_xiFXjY^p2MHMr)|jT~<*Gk=^oaf(D41{e z?l)1D#GXu{M4`-4bR$tf;}Htfe%KT0@I-_k^u9ye*ClDbc%u#G_KCBS0l_KBWB~*_ z2L{|r1lwU+DCVrh`L_hca7i9Ui64Zz5+uoZE(qn-N>JQaryU(Lu3Wv!NxOLQBC!j^ zNmuqeTJ+aMHRM1n16zoimS+s4Q=1HVH)?;wGE-0kxj=rN`aYmd)0RT-LMjii2BL?r zzHe!Pu^1HAOHiA;hlYe`+J#>EilGOUcq=a+3Ik4ln2BQ}5eNv`9xcY@UvI}u+dUcS zDNBHkgv5SC=at?0I7w6J?TcIh7*btTRW(j>abki=W)f^D^!8?~wwP5@Bp$vH_;rqB z-A$M$Et6m)|4T-u@K4Xh1+rUD#(x3Dwh?B_b4h*WkM_jWYF(IEnC$X(njAh)t06}j z843c0lD8F;v$ZP4Kqf%QK%W7+-@Shy_`x$hEo}^rF`WTuHZR;4XHtlqB3$_g6H`Az zqc@TWRtOMKUk%G=Fx!@#9pq}_7=;G?%E;G=?rQdT5@dYNZjfr z>OX}tNVeTtcNj<=LH1A6Yieu9aXN1Ao;_W+WeU#~Ew^q?J3;RTa0lWvcJE-a7rSkHjYy+@GA&r(0B`qw;KAJ$`uUS&EQXz zSC@g~&7t)m3DBvY5FcUFn|}_5KKre)ue;j}N8dxlZNtY~j;)t^4(OgsW%Pt8i%ZLe zPd>v*3?0rPcs9q$2@qlj&jkIQ!xA65_~n{Uu? z%XP5@6k>h8B2Vu6n$3hh_Gc|jt?c|ZIW@&2DXELINzc|WGb%zA9h{#x}q*wPX_^{Q5i1PgALuB0+A{lQ}7Jq zj<9D$p3i}ABNfXiuY(^v_%E7ihTpKKfE}VCA@l9oa|{I{UOztvk0*)(&b)eI=`;7S zxUHI~p7k+jc8>q}q!GZYrfgyoZSUZqP|WDIQ*5qmV3@ryF`ZA4*ccN;d3_9FyZEEe zO%jZYuk~anF1~i$#f3HHPD3pe0Ow3!pqRCb0YFl)L>Gtj0H86Q^ z+OoyuFAj>oUk4d1W5w_QS6T3WhrvIY3T^5>U#uq{5M@(2l?A)8rS zJ48-+#Sz-?c-~=-S-b?VPvSU`UOgm^j#+uYB(n3smeuIw2_5-(ZiM~8?NK3{DYDE< zaXx%dLPFvaY+Okfet}x6LZ43ln*QX;6Y{&wY;3$VvNo6&IRQS3KLO(8)p;?HU+byA z|0bF;zz|fZR|(S|a6gX0K~bSy2-M#&M|th{znini?OGNm`TF`sD~9oUdoLsuoI#1K zk5l2Mofvgx70_0FDzWVK=A}*RQ*3u|JK_DIiZN7k=oXl}LD3Fd9JE8)31sjQ9-g@4 z_RD<|{`Valuvn<2vUIpkPqOTkp3NCVVlu`&)@Tdq=LN55q_}SSJ^l^I))_d{WhrG+ zz7-LmH{m%mF_Z$DFH5KU`C;L`+a~^?Xh2wH-0+U`UGMKrqfp}H>Og}U#C*B#0%R(} z$P>rYjLVwdF~2taJ(jGXh=>$Kf#v{gkYzSP+1k&5sfk)a?I)P6z|;L!8#_goR5|jp za|iitU`UoMY$9jD@Z`qkopbeY5;C;`*1T}12kewr!oLUK05%P?>IzFdzvVy8Rxkhb zRs<bB1^LE04vVBaL8X+= zlEXd>?GALV9vc?F zqSz(<9*hnYCAA9|o>F=C?mdlTCXOWze5$dLv9Wt;%3fXtz)qd?zU8ko8K+D!i#9x5 zTqhxe(CeO^T7Uh8&1LcQ`^v}B`a7aEQzx?4Ng#h%E(&@t0k{jVPCF; zad_S{`sem-Wr%io+|3yovQZyyrwr@-@1~}JlSp-^DZjhD87At(dnABllioI<+-4e= zZP_D1>CF1y{H8%+ADBjC*)Jj@;srkgdW+k%?v1H6=VxXJFN`nm6eFL$*mcrUFYlb_ z6CD3SBnMLa(GET2F_~H|Nf|EuJ%Ue7`Owr~X+tGc&A;UI%*{&dF|({Qu0-M~} zxH%56AE-x=2KlWDZEBEYpNQXeIP1TxtT9Bs1T}W&pr@6 zuv>S#0CU1FKQe=Jr5E&U+j8omyH0xlzVG0~ME)O5FOJcH<<4;!^(F%% zrSPi7pQhtP*i$?$1@j-7_6ZHv? z0I9D}GnUKRxvP`*<;$1E0YSZ-55>jB=R){Q)0^wE9ZUtZGfpd43$_CQgs?D0mM@hy ze;S(LVMNXT_uu0Z+wYfo>0wX(`RAYh&?jM)h(qom2$EMdw)0I!dvU)bub5ba_Nh~+ z*ySC@pFVvm$m4o*?Y7;ID@+b<_DTC^@wU%K8=0C~me4uTqzDUD?Bg6_TD59`+~AwH zZ+q+xI`)=t^11v ze0Y{8JI;{E0jGng`j5BcLStYaeEauT8+Y6kMPX*9nkg$>O2JshvOngc=7W4!GwL3=s*bsi}Mg8Uz{Oz_%V;bUCCt(-0f?VI)fi zJ60andjhM5a$jXc5JRVjthPV&YYWJq0>trLDi0^;yB096WIQX0#M!Vmay-7P9=Ir} z@#(W?nW*DH5`7^#u z<;n?y2O7i6#r6J@HBQKe9CC_>q?tT3|M~N0V)Plu*;TC+oclNhKqAdKWc#6_BJ-p+ zcoGCW#Ty&-BEC#@b#+Z;uIjX3+P^8JdtjhEs2IHOC~+xOzs*Cv6Y-jLy|+WE<&rff z^LzIk3)(}fm1+lT zl{18BUhI0WHGt;aX+Di){>(nNG{|uzr|qMa@JN&POZ&M$DaLpb6`8x&2~}obe~V@s zwg$wovkglHD(|GM?Cp89Pjr}cAj#m+;)Wop1;C^R^hlD%0g1<8g~|OsH7F=3_;EV( z*6rJL6td69IlCZ6s-p^X(jemDSv3Dk`Y8{9onW*AZAT?fJ%0?%i|pg3gY!EcS487k zhi_V{wa%W6gfvY8!+2Y()=!QKC}k`C{3US+cpusYN~KBBb@8TsyjLaiOED4&qO%fi3mvN*1=QimGW_kv^9Yqfaq; z`p_6eK@DZ4qC=Jq{B3ZyydU}Fk3YV`BdpFK$gtG>;B>fL_Vi1xi~HEu&oyE7$^YH3 z;R&Q*S@PJ281&IlxgrynDvfIf$_5^X^-0@z>;At<+q3BS39>)(advX#=g)^!>(*CF z5Tr#yOYdH}vK(5Dwxun5<*?ruLpEeulYOgw(w6dg=L=wEea~?<^yQ8!SGTr7>Di_@ zo=$qnI3j3}IF8PZ)9mEMtGCLV(oKD9YHFx71YSMlBddc-{a{sonZgp%(&77AD7s6a zrt<~l$V`VmMH&G#*swM~f}L3Ltd=Trwv?9R`tvFAu=Mu+oF8&sY3qc8SFwmk`tzOM zvr5pop%om_LTi#N$na@%b93#~bC)xBh}(q5t0!6Pt4>L8wz>o@&Z$;GoZ(H2#v#R8 zJKjdxnKO-WdGf}^UK^@_mGH;jLfd+=yv}6dEyE9 z{_CkX%K}bfBmgpQMWefnXwegk{?l$ zliD6$zI-`E(k)GNrcSR7Zjou6$F)i_@R~BVbWY6zJZ^#TojF*a5L6_Rtn}i-2~q17 zRpsiz9^|kZXyxrhry%jCG%_+G;xedfDP>W&7d!jixu<;M9+P&WW>rU&ahw2hsGed> ztzCSL1IQ71vKw6$2!6nQ!Wj=Ldyx$3*h26>J_j!Uz8;Z*7>o-!oFt`fijF-btys5+ zSjr#$`|m4QE0!2uDLfrCiySZd@hdk`P8O>G9fpr4D=)7K&TXa^+lto$n4#qnIWkAE(sSsb0M$dx~hm?)7Nn!+|Q!{^2@WCUvOgG-YV| z1dySV>$*fve2O*660@h_oa&C5Jqxfq7s7qGgQg6v4`j5<=@}c!zyIUMd~s%{ERm~O zxb2e=&iExdew{y-(%p{ppja&cs?VQt(DBsPSLl3wG)eD=i(k#mES72xeJN=B$=-fM zz4ZK~p(QD}?DB>u(F5JQb?e@^uM;*H9l71jOSk(2xYhFWo9hkJu=A)*jr6c^<#>^Y zD8a*Oo5U)p8;i8NRbJH|rn%?OPWj66veBy1zC$x#2m}O4+fQ|DMjU04Eb|gW===ru zcnlDk939@9B(zbHI1c2v3?*ZK%JPP24G27AYe?3!)FUVO(>|4%%vB}nh!2DyoMq*j zXE$zP1j~M+RRT|F51?ID)3r`LcRR8jQ6zPq@8AT3p9n3h7$owfE*Po zc{#-%E0EwT_O&>T0JX4)r^X?15&7lw@m{i3&@T3)a!6fL1;Z6)|Bp}O5+i>V1+LEv za{+zR_U(yyG9TtW#|BT6GPrATnnv3n{Pi2U!Q_D++VHM^IL9_|ei6+lWmz>}z|&N1 zm*9lmVYCHsek(c?X6;zi3bTC=EnP=qZvd=l!&?-{XrKc0AoA~_&1-0AxL5=`t$)u7 zaji{5UgzA*@wXT-A|mFkNjjboMZ@j-q2hAt9jI*xBd2^D#!(o1y1ESa-E)sa5138V zg32%m)9G}yo;wwOU-KRMRxp01HK2qMGE;OsSq*B2Lg>M2?LrsZYxG>~M-tJux*IN{B)9nkhaa&%nMSOUF|&pWMB>mh5xp->dlnniz_UVG`@%yM`X8UH zrDFK%9%up|C+7*eBU~mr*EO|N3q0KL;2MX9OVv+Lkai93*9QBx;iw&+CJ+grj@)7J z&gB`{+Y#FDxVR3_f`CCmt2UnCp)K~f;@s|0Vno6b<0Euc%^y75WPf`BG-Jv0CJEsQ z{bmXdNsdoEf7+Idz!D4nkS}Upa~wU10EX-35U@N+Cd-zRi2#m>okT&SKZn~>4kj+~ z&<|2~*~Jy%G5+Ano&U865<@6Al?r(2;3J6q;sz5|Z$-0rG*NZFtKPV(fz0 zI1V{N)6#ipo*&r;sVKu_L-D`=n&B+Q4HdL4X(SsmI(=Zc2SPsx<7|xzC^<(uL<5HR z?T^>BOAPz24xa6gOLEfij25N0H&`%HLg+)*qJfe08weP>K}JIAFiJv{Cj;pjx=GTA zz7DR<0O+D$mQ#{{XU|wb6#{}AE&(L3tbLV&LwuU=>ui6IfuiV+bNiB2ES>(j9S} zkI0`V?Oo#|dU@q)fJRAbXITpElS#VSHPidYkz?wUbQ`GaFCC8~;K|2v za_E$T6%WBn!2y7eFtbqG1S|fJ_TD@!=lt*cKW(!p%vj29h8m^FE@T<|S(cGK*9mntf{eFMm9oKan z^W}8*S>DU*_1xZ>xu6OQ+nrfXk0n}Cmb(30YIex!V_80;*WP1oB8jdCZ&1WY7y*6b zbK1AZ!XZJVWM`|1EmZv<+m#}GA zsE`&PWTqmytYVHGxv{hJw0tNzBN|QpJ6pGHOQ$l{S5i56;IZki{k}ENI=M@psKIz_ zGHtrlPMaTxR2HrUx%x@@p|tb?OpJEt4{fPlMO76|Y|t6j0rS#5)WjZ`>5iA;J3sFF z<2etSkvv*3Jwy6@Hh7db-bZR%n3jz|uUbYR2v<$q41~Y}W(Im?L}{ zg-57a7>hKmr`4!YK6<(AHrdmon-)rYlg^z-FF#c;;elo{K81`5kE~m#PMy|+J=uKs zCoMcX9}86t2#vxD^eB~1(Y$i1mTtHElx>|t=QJ(^ns+%n$wFqyMbC;T{XksFO7F&&fwTF(Ygob1i{i>gy%dA;>cw`?|eW|c`+ls#Q zOyfEfihFy^@Ijzs{TltTOiiFdKFLuz&%b zhFxzdnl5t{{AO?0mm^T2Ih9?Yzz7?;S!ZfKBi9~PLEO8Fm$R8K_vipRD$U1_e+aPS zjG}RZW(f`DQ&ZAe9wxiHpS#ri`N3i98rq#(e6VAuPAvWZ_HoM|i}bkK>epW{*<=VI zO?jiH{_Ma7`;Y+EG2V*7+N9z!za4be4#@c~(9ZF6mhhttbho2+h6sM| z;!rwX%2(6Z>ggn^z%+1Mwrp?BxZ93%BN6W!XOWHvMx&MHmt`53uY$l|_q@fphlokSP>@qI=FtF{3X6!zrU#KS%e3ka?01Lp0~(b4$_!xXT@NL4UN zGxM%(Y2`Di&f*od%fIi>Xr(r7*wDRQJ^vna?*2q~waT0phy_7H8K{F2cn%A<{8&qG zF6^kMH$qg_CUnSuWmUo-lPO?SQ zrX7|4(%({+&o*tLSRv99`N_;??I?l3o0`v;q z=smag4Jq0u)Kqjdu6y+aDD}Whj!IR#bG_+JHMg+-yGD&3kQ8751tLS{5XJXGu3~5H z)~7HVL-W>W<%gN`FR*CS&$s3t^=kG8g`!EbcJ0DJ>840*kCCC_4c^Q@ng`$>L{_z+ z>#AU&fN}c<*f2x8GpNHfu)6zCvCmE*B?wcu;LI_ReX_Ur4b=8F`ud@x-rfo@hiSZ~ zoq8@VF1bXO-(#t5)VW?}m#=qwoW4VrF4)daxjCvb4w~pEU-R{9&eg6^H4q)FB+y%+ zgo~KA@=pCa)*A4(O9>V@s@@BL{4bD&CHh)bF-UBU@xS-6;>B>~h1OmNX zKrr3K*tmD#4_|9VMWGc1uv+0ShrX!v09!?T@tH;Z7@^NC$A9}Ds>?l6m}xwM_ar>- z!g9_t0?o4%NKSNc|I}N2;p@%V4;|`_oAD@-TQ)$PyVQdUbrQ*RTPPJ~oBr%QUn4)T z93?9D_wm&iS*Ql*M^i7)2eS3RH}AW5@4W8qyyt7iu%*vm>(+^5La0%Cq=gsSL!_|j z&3_!_f@H4G?jOGH;0lAaf{n0{Ou#j`n+;Y^#Gz6G0Z09VHhu)#^TwrfUWb1vu-2tjg%!C(}@er|;akQ{Xt!+d$O3ebeCSceRsAVN7ob zTKfRS;Ssi$Py`_lSJ7-dVDjbCi~{ty&kjnPXil9yiKaBX%*x8DgdsZz+L_SXst^|S z-C7NOCu&C6CkJTxOR|&$$u>H-u!8=#0n^gf>9)3z_PqxG-Ux>8@YFl=;E$xJSB*o zrt)1nLic%~DBZ@)97PcaQ4P`G?RRwJi137^5razFJ@v2i_NN`2mm0irZ|}K>7|e61 zgzRFMhew?{rLPo?E}R;!6Ikp7cKTZf0NSLYz-hd zh}Z`cM_`cd+iP@bz6}^8lIGT%odj9_93J|BC9gIFA8J2~O8ei)8SB}Eh4JAE$8+() zZyT6<+qQ@svIh=>SLrgq=SQAn3q>4l1vxXj9cqaVOKS!TSgQqE8gGwGRdy(yWdP-}Zhsl+vfd7Xpi7HXdhl)bf(Y z`=X={=`aPSe@kpgVLOjy_cO6Be*5-?OM^;$?u|(&d&emqci==rU z%`Z#Ci1ZKDIIfCAV;AKW_oh6KL|#JHMa}j##$C1 z{M2vF{)zy|I?_T4RAp-h^jSg+$uNHsv1TD-g@7J-LYTd#RgG&TTeq~DG_fVA>j5e? zu4^vD4SowLOP3;z&Ch4O4ZmF084;Rjo0&5q?&ah#_dgCL_@SMdOE_lkY%)Xs1W9r|hMRg?p(CTfa{< z6ZFmZA9(Eyd5f_Lgyl)ZMG?o)2vp^*O04#3qOCo%Ur)=+6s6@m^n&WQsID+HHGMR_ zX+mseN!esfcObkkdM}1_bu5oKq&Tq_IUz7=c4>D~$chl*iP3?w_xxO%}xMJ;$Zhz?2kGk=8!Sjfm7p}98tUi-) z?!efKr$z&@r<|SHTYI?tembSip2rWZiMZ)dzm3n&>#sOhmA#yyo1RRJT!Gd=UB{T? zCreFItR7|g&pL1V&UG$q?nLO;sZpaxjud+xydMpKCGb{C>L9>e8h%;YsEXi->EeTV z9Y}`XpyElVtZ;dk?o=KB?DZ0d;xlVjP44=3?Xf!@AO9<(73H> z5NsFea>Hd?5ZMBdy3WcBA7@{W7jsw8ejT%mJ_bF1E9@|!R@86N-PjJ$oniXiWmZCqrIxvU?>f>3Nc@~ z3-;e>(s=c{=^drlr;iPFLm#bYCydXU(10OALGTnAoaSD>~Eei?;7tHD~?i zWY&yZMk}-`8m)5Q=A;vsSYGPFdIn4;x1tILBX2%8_VS5BK&&Wf`Hmkyp1i9(@7oKM zC=xt;F&W%#8B#@_ppVbqdI7U%&Ma)!Bf$c&O}w4pA;|^Hr?vaGu#N0h)$s>|+}e1D z>7sn3qNZrt`g81dlPa%o68r+uV2CwM2)Hk#3J3s5ak44;PEb%#vQPc*7(zAkgFU^t zUB`nbi#$x)o%lIExAM%53mF#cr(Iw4-G<7jSN+VcCRS&^dUc&UHZn?XyQ&{+GxuHo zani^a%VYgiswFn>hmRQX7aEG`Ti3Py@ki&Xvorv*#FRTya~iU=AfeCxp%Dv=ywCIT zmj@We8r^P`TN#zvbt{m_JZ?F=k`oSei4T}&+tzcSkx^Je@*R_{b=*Rn+j}b6i;6E8 z2I}F$IF#U}y`(ztT=MG=dC!{u5%u=BO92eSl+H60{>fkR9V(yIjmvr2q|%|A?vTRK z7eCo6&RMfrv3UFg0y77S2Q=7g$2Hd#EOex3AkqnEWfafcG3!tj|2cE(7dy~n$|W7s zrw&u5>}I9hY~~a3Y^7QIQ*TqggF(1skBRnz1!y6uw_wM>y}@%YQHThZons?0V;?)y|yl5OJTtK`T2!#z2~1T zEI#XP)_z<0v<_fESn9?r2lsh$DDz8%Bs~0!0~uS;g6I2sn3_%z&PBQnP{cgiOOwKd zlEv|W0E-~7k}j-)6hwl_6V?2?GL`MnqTqZ@DfDR*D9v1&9%*M-YPP>0(4vdA<$?vr zrRQOx5XoXlU{n1>9x4_O&=py29=KMjn9~w)(PEKfUl5++c~5}S)La(JEIZXvGusyw zUX!F}x>)(aVjL0<>bnIOE82k23eyNk_BqOh)b+nGBu?Ej+j=QO1o*f|j)`O4csMVT z#;k*^(7ljv+w3FU=!Qu9JEm7CnjNTct+Tan7Y@{bO!W`Z?Je7)5&ks z;5qS0S@-bF7%PQ-4$PTlq)V{$B$NR6xk0Hu#IfTao<=)?=xD6yiBNZ8t3QZ@jz+@w z^Xseza3|YzU4m*IL}9K#<|%u6jRvoUXv=d+#zPcJHyTEhwXX5H?}Oe{9?JwYK8h~X z!{!O>8Dl>3?lKlG37;~0$4DJTd1I978nN`&{1z1FNH%v}zgSb(baQFxgJ9_prQbY0 z99AZDZl$q4EG+Sd+D|ilCdd|KZR-j>mWkQyyX79Q3!jea`4Bt{1 zUepH9)4m&8RCKmxJ}E}-;_iTy1Uo~jdE?Y^LICnoLpHy=Q07PnGR5}?15}o)EoqEU zhi!_hqWCxhOW_NOB0rW7*&cGqPI@TC0_C7t@(EJCcwNJbii+55qYZwx?5pVwgNCvw zNFyh@SEL57ranC9I)g>b0-$kwTJC8+Oe3i?N%_>ctC?9~n~<8vV8LhAmXE=!V}LJ; zjbm8^#7c{%WQuR&gpryyvio`aZi2%&c^0jsRNOgxD*UPN`k~}3>V(u=Y^Z^Su6Ng# z29-<#OGIet2|nSMZ}{K4!wEfJL31g`m%K`U<~ zjg;BoFd-!2cwP&mR@Hgtp7nQY^tcSGSrdCIMt(&luDP4F?`nU}vV&*gA818xB_S_5 zPUB>VFcR6q<-Zi0G$Pp`Z89P^h!2~Ql;SD$6~weP+cYg;*RkJb&eJ?p-#R|dbJ!?6 zAnV2DR$RnzZPadnHMRRGK{_6CGbQht7Oy~C7Myn`b{ux3B=%&q^vyM3= zO!)e_R{qH$qRc>o-K&FGS4bBO-Mzb|*3LSG8eg?yh-&ZNV*j|WR*b5V852Y2v0w*}N zU{c?DIa6W&6zc&l82heh^QA`wQr29X2KM z5Z^QUyq0ed&5iY)(b7i{9(FW~n8m}RA%Au{xAN{;D>+ak%#xK84%S}ut3g6K00Nu= z2mJ~kKDy3wRSHbEr366=>SK&biG->^faIz;l~RC!nT(Mv4LO-+1htvFuGl_;jy z<8Q>V#!&96l+B9iHPf|=VzLGWKluk5GVwC1f@uiFOhLP-bENL{KALIOp^O!F4#!bu zS8`}Y7*)d*5a9^TlPYv(`w@!Z>B6&mq3kzOZ6<~)J~jOM1$~9^p)gYg-}Ez@&lx2; zF!&al>F8+<04&YCTNP~9WIR%Gk6*#CPe;i%uvM!b1tMWpISJR~RLeCfGS~K9`1B|z zQ2eTD-oE__4As-%{7h=8E=6GvI(5M*=iSDtm355Qv~Hu~Q!UM?P{>i*ZQGf-TdJ&2 z%^D|=lU$pGI;h{V=NRfKk?y#yAvFem=VW(&_gC0lgDg{MQ1`nVQgo7~0c?f=DL}{>!d`FmG|UqMB+xpqdLQD+&`H|WD+cSHHK|)4kn>>yJA(RSCuUku$HC^g7$#9 zp?)`Xo)-B#h@hy+1vtj`tbWV(6B`AMIaiBt4q>Dd-xndLDHtxRhgg<1^%*&GQf*07 z2D@AEEY_BMry)zx)+tk|TzW&`45EcPi-bRP9(-|mfgS#tr^X##td>L)qz-yYG`GT$bLj(XmoDF9*RPLp^8ho|`K8E2 zRx$57_2fw_tzv zig)Sdh>y#8k&W7$O+gW&3YIg;L@LOFa!;99z2r_{K=b14^&z_H<4GPLpWU9bc%Z zcyadMHfisxhm2ZDwBi{A(iheomwXS%^=#MCf45E?mmZf1;F${Cq_1S_okSLgT=dr3 zi{9mqhqk#eX?Bsnt)t6bKyY_IAZ=YF|(M{Q~Nt1M>BrTdXo76GM&9tp(IBi1A zTd0qUDM~(eFp<$vE^-U_Be?+u+tD? z>RnP`RhZw{5!uAn5=3O2^Xt_)NeG~l=Kr6 ztJxX#6z9Y@EXF{L#&tUVC0XMMrB17KASFfQhHIG@mEDt(#j%&vZU$9`QU=V$%g4xFi zRQok#eH9am>@2zrCg8Mj6 z*?qmAbgX1)Y}e!0<%vd_QK&x^G00)yyfG>%uq2R5b>)Q2zwsHQE$ATAX`&efu^>b# zP8CQOV(DM`Z5eWk;v}T7A^nG7JS#4Z`V)yDrAtb==(SN}N0}`rdG<%OjnSIwU-}!F z&KjOOSxgp8%f30C#|K_diLJ^9Q52DVN;wA!Skp4<{1OvH{Y!?4zmSu2ad0x}*JDQy zNCO->^R~+ZMU@0?@RQk$WKButF;~I1gdsk<{c^;^UfSvu8Wu}^CA`Bex7WuXNVN>@ zsnqppX9~%XxY2VOnpV+1-CjXuak;mmvO%q|t8@KtzR5$;&?|?1b%rG6?7r^7{16MI zzC+8fPXh9Sx+I2Hi6UP_+re7>{>bX`{IYNp#`yuzElg+Gt%uas5}&q^su9ZntAb_q z0+QyQ##0>-{^{Vh2FgGnfpr@{_;24neb;Si&?xh`O51mLiW7fptJoyD6LQDB07H?@ zEV}K*jFT*_bsLYrC~_APMlFY?9}b0<%6MsA9!3So^D)l7Y!e6Icgl3fBGC*!jBql^ zz5L*jC`(w(y=n{TLODMtK%LnkYnkjOu7^^~z|-HT-ET665K;JDd2!Oz<`P1nPDhG| zC3au*+>A28j)`VWxktgI>SiC>zG>6n7@JiZ5&0xKNOTm_Z%^1aAbM;IAEsieh+l4# z%AZzKxttJx8Oxua9{N>B*ugWVJKo<|?bgRvek!e=7JfVMYw*> zwD;QN*DAf}MH!A{V4N|vxo1W~ZFUJ|>5ReOjuA}Z>q3VzqALsCf!ccC(@K_Gr-}o8 z=I_EH*>$}22ktx+Qh1CCXi_Z~2t4|RI{Pp>ccpsHp!=E@czuPjL(Kg^i3gd`u5PV7bt^I-R>rWjVgj;*nBL%GxMe8OFGWI3N2D!sgq`$Jx1cMrM(_ ztbcBUaY@K%)YvsW(%P}CRpE`j8*Rk#m9B&U3OPnV(ZA@ zG9JJ5cZQjGNN#(vR_gpy)ad;B*_n9`3Vn?jJX4g%t~0ME^lvRi5f;eQiLpze&F%1DC#;?%Y&SRa@R>7IW{k za|>=yKs^B`t@BHliv&H|i2k~NHD@M#N@=klq{xyYOi#%=c9)|s89ATnl27BnRQGUX zg*eL;sBMv_Yq1hpdwRy{-FUqum)O{_#EmB$(5I7?@q6QxAE?@3&+Gzk#)?+nEylQU z+DY)Z;+wC@$`Cq4-knB?*;nS$^8{MmW#UEPn^It*53=DYRg%M(lQ+{#YAL3yxk?ln z6-Fl_9!hzc3)(YHx4w5qZIkq96tzVo)msF+YNvU;1FwRpzzEobT) zpc_nG7qCNoEP=;f$vH*YC4A}YB6s7llRS_u4{YW0vr>|<&?`tqqOpuF(EsVDp>&En zue{hUuXP-RVXCaHkT<^B`MXRr#Xzv2b z^r1Y$oq9mR$Pcc?9rTO%j5oIrjg#88DFPCP#N4^uARL;WgTeW< z)Kp`fM!aU1@yJCdkMI0I%UhKgFQMZK#WeKeX`Fd5L%rS)?r%L@#?02Xb9FeFySG}1 z;(9E#8DIwYIYvdNDuN@|Ln^;j?mp}{Q+L-NI<{HmfMwc>=7EM9K(N)e*m$|BVx5zP6v-SKMU@u^Gnp$v|KR+q^$J1u&GS6 zTt`~%uQPORZD;bx5s7roxR#O?fOz`U{oo!VCWq_UzSP-ZBb=_?t|Z-8-Q4}VPWxqB z(hTR;?*|eNK_y|N6p@!2T;WzVg@K8BS7{LiZz2H>P|A*Z{plJ-u^`}#N~wfD-{Qc7-x zG>ENyCcaGSt34bbtyhwBBcz(pW5uNeOC@1Z9UtP6CH;&~eTFh@6my5rwn_vWY0uLZR5_ed{)*Ar;#zVg3PLM*p z{E_6^N@+Adsj#2uQzh<73RPC{lcp1kker*ysMJ@owof0^SP`Nza-F5F{@@#@rDLp) z7v%~3_P&BE_ISCmvmA6#94QH&Sb)2Hy!Zuc;`WuzdTvWPEb$M4N8OiX*oU)VY2Q%< z-LHi~EA2osIQ`4;lsi;b04I<->7p$>0LVNqKz-TaF*>~R7_ zi{m+Ia)cvnh(|8z@@Y(RlQuDzx@)&1^pX8Cy1iGWx{b!=+S^}pmnAG3{nr<(Rqd-1 zUMzlf{`-9ni_Tno5UcR~RRbOwbmbz_NG<5YE-mTB?koP^}c|Kr>CcNxy%nVFfXo8J1P!8t)w_9o|m*F#BYV)qum$Jqgs zcN9FoFiq_c60HrW@da=wUz=L!qNkv=VwL0wl4A^V!6Jph-HIj+64*jTnY2_4llYxv zmO!9)n&#k=mrssz5NDoFWv2{(9GaQ3y0lxCNxhS`P;k&#Eofi9>j5lKapc>#3jpB| z229y~t6EB6$;@4{WQovuy#32Mwxink9He;MTg-JGzq+lRdf7QQV*2!-UN1DaF19 zSk}1~GW?`00hs;W_0VNDSGP(B1a83ecJe*MOshX_+zkFmy=!r`AD+hD(iJHdNSUvm zXsWnkU^4Hh+DWLGA{nIFC8Xzc9lOLuUAPgIy2ydl1e01kcDmHwhV&7t(^N6-Oc%2` zm%E*}7fqF(;ZHv$#C>S_9|YO>_0j>wPZ#`71c1&zhL42S4LNeeLJfGfYBiTlP6GrF z4iZeD1Ur+uPT3j%^h&Q+-Gb|Ug{u>m2!=v+qOTMe5xVBYjUPzLN>dM6s&}FI7#f*% zYp*o38MdWZymwkV?Gb%m-1($nmmWTAJ3dRh?5$P16M9dWL!LLpq`nfMM}?{c4&h}^*#Bq0GSf?n z%&*sYXE7Ko9Z&xuOj)h8G1zVIF;DcBM6&OMP<5W$UU5#7s#R!66W@3JuTx@{hC&;6 zO;)Yiy#WVRa#UM|-I%;N6%!Xxta-o*`K)^A#cU>V1f`e_e$%f8hk z$8DRIw&G%~ms;_uExR(9d3s*{t^a)l_RmWQ@2qofp(fZE$d>uKzl z^~co=4D;`SLPL=DEan-6QqTHvoo&bJlA+9+~Ho&V5i z2OffdW7@J+EhS}}n>{vSYu1wUy=k!AE&`J2FCpsYYerNWeZLx$A`s+h+fL~L;>cG! zn3*e5p+|J6r-;*#SD)f2fZ$7XPH+p+JA-zy*)OD!%nr4Xa%~Joh7= z>$J`6x7%Li`|vCspHx%N){tm)`H&X*(X0PUK-QVNte3J*r>={1Cq*&K@5?NhT0ENB zoY#4P4MG6u`zUzk>7$d%u|^u?@&44aqq&}THau)qbzNIh!A>BF=+|IZgl*9(|C zv!`4XU;9;NO5r981?+{PJ;{ES;+6ty+ajKEnJ~|1q^I zp5O=Aw(?3et7k6Njek;>;fIe%MpG$;7Ts=F^CgvbY0li=MOg9Bw}W?%l5@U}pvH8Jmy z0+U}(trHZinT=0GdHfZ(L?K32^km)p>8(yoZ zwrneY67h5aLn%`w?t`=>X>lY(3xW%*D}t_L?~@TDN6zKILD7KNMAk_#`uaSDz-rxl zOphKtL^DI>y^P$Kl)(PIGD=x-!-c@$??M8bE8-*wSbiSu2lNza0fPW}wNf;VF8P8I z!sEuOhRbJQ+Dgfl~#*yLL*rSnJkY$-Gb$V;I#U2Xs`pYy35Z1F@mV zYWkpuBIs&=&7G1i)wlLhbLhyV|NEcvBL44lkf-qf&vW==^0(ez{W+)V_JJWLN)+Ic zrj0V4w6AP;ZaFzgQE5)%QsDSC&l~yHzdhA^1$BUs!!O_KS~mT(s=+|FmY#pU7%816 zsem1oa@dE+yr`iR4L>H-Yf>Yf1e~=C{6^vZC;wJ$+?U8Wkv$9VonX88Mosi9GfY!0 z{2%_O@0IS_SMq)+>>z-vnkhm0GD&!M(!z#bV$l+fno{;hWbjOB9N}i9cxaME@TQHhG4NOpz3>`BG{touycPkpL~j&MrPnUdX2Wv#q%^Y=Q6kR<1pKF-U&n-_xZJ=0a-X*RT+9&zN zV)cW%j?ezPueI-WrF#kV!x$B0gWKZ|C=j0k46A}!&^>;i#>o0DQGIto zKTTde`(Tp9apz|7h=mu?3q{>he!`7sZxK^@)hUoCX|SP1&M&v;G6JWR?%Vi& zp2G5sW#P-ch%85?b|2I4l|u333FU}it=H|mn6{^SbWJP4{YU(+r5O6_anpXTjGE?G zkG&u3Xd{?%ELx{@Z_Wt6;gs=ZqVJhM)t@KAn_R zbn^^{`x(hA{-8hL*@Y8Z*!V7Y~Fs3@D!0gB8DA468Hr z@#Xld=Vn2|3Pv7WH^~7t-3~n&{_~&I<$H~8j|=3iod%J41Dj<$Py{VFn3>qe2r00- z5n)R&KXD3~qcRmQTyFG#|6JKyMJ}i64OW}HD;Bj|K)(qp1l>me{R2%*g)ikFiY^c- z;3VKqJz>iakNR6hE!b|w(}TR%mCOFO@4sE+!$H3;vz!GBPD)!+V41sF_7n?w&{;@$ zrO#dd_@mH4Fh>X8U!;c)!+i0ivPFkQ3$+x5PRq^IZ@AhesaWO5qcDe5c(vsGEvHL{ zw;1L34~^XH>5qnRZdGNuM1d~-@Jjh0!pTXFi=lvDofvyeN9LVZRDR98>m0kkX5M+! z790d{{qt!@o0@g||Mfcef4n?;9~fh9E(uP&K`?^BVOhR0&{zR?hLjx;I;-8drth1W zkD|>JQ70n5m8&x-W>U)DTk~h2Z^VZb)RbM8lUe)`w(G1K0tA#wSBC6U6jC?3Z~5(g z`a;qER+_ASs_R=?_O1Svg5?2T=JTpYuH+zxDUu1VNEr6qz}$fMr|^zavziGLu3o*m z*yW8EdPpK3r?%_V1-&w#Rje>}2ac+xuW%q9Y8oWnRm}_4#D7*t8v+zU( zF8CJbp3@!~d+J;yO?XPXI)wu(FVs~e$gvt6#X(f&p7&~Lkp0zu5?DE8hp+F<=dagL z_4{|t0FgOwrcND!xE6P+WY5r}iMjwy)|;q<^(DCr4e(g{9V`49572>r0S2;}HEY(6 zC?TmC4<4^}jFvRbb?euQ{Z&;(H_z1*844yyq8U$bFT$LQJfL2yTU45rz&LDK-Le;C&27=4{xoO_YYGqbzX-v3CD*F>=ZdZyBtHB$QM zWdViS+E>%WW`Aa;pI=Z;Gk)te>%=uW?OWv*)l1($e|G)GjUL)px1f4=TE{Tpvcu-W0~x9oS5`(9)oUVVDtNkO-z4n~ffjV=t){%@&ZIE! zJ{_;w#Vlvi_Gm!HOZj8Hbo*!+n2_eQ&P^OU=V@~}9uV}@%@3zXU z%;o?xp$CkkoMB~I%VUXkw(aT=(>voeSw0A$j3-a#rEaSDcV$-T_21bik+T{tE=MH2 z6kJ!wUH=^&@*z&jJ(f^_VcP2$liD_J$jp=QEIgHO67x( zo3p+&CfYzSWxlQCM&|?K{rcc;Mrn8tJem}IB>8keCCopE#Jz`u3@09kAHWZ0Q3Y zD0!rS=AlmUYu3G+k6Tc9O;jpjWcIQCAc+E=wou0s(E>iQug1-kMyH#S0|flDEdKUY z)SdQ|_4e}~)MqO7S@hhVzE_ZMS);1j6SyXUl@x;?QgXxZjB#}uxG1i@zr>wD{hxL( z`VE9-JX||MOF_R{_g!G16pJdlO!voUKMT|*1KhF#ASWJ|KPZqWkn$5{JjFfvWB2rY9rb#+ zk9H5eRvZVq&z0Fy*NtH<8CyPj!r9q+?r(7Z;m@tjv!c6iF^V4|4etQgg*&N=C@onU2}p0~=_oM({_W6dA1iL{Q#QC$;_8C9B9E4%o|pZkr+3wq z5qq(sKk|_C&)FQk9)IA(JGXKczT#(+{&o+J&h0{_&z!<*t9;vt&@wYXt&t4UmXjYr z9`?MCQe$vHbGE?1#SRf5IqE!xf~V(u-H9SyWHr)DRq*%Cei2@%!7NOQ*%8;ObB0tK zmIe&+E15sBIZs;RH}!Hk)H@AcF~Xz6d9hi7LnW`+!$`stt*=C^k37g^@emGO-MX?x zG4pfx?5O~i>Uv4kur{+x9?lLv7qRcG<2DP)3qfS^+r?yFtN2xwx(g5Onuhfe`l74S zgG?hivdXjU#nJ&JGrStst9jWVHja;KX2tRwHq46&F1nk|bBA5B-YrYB3aMw2*|r0w zu9w?RD?uwXiCqEUdtuf0|8eZGk&?Uq;OR%IT-%DJn?iz}HXj-@eth2}qqnrgAL#m^ z;(9pqapZu9@`j#GpnUWwd)GDh^2s`*P82OdY^r9Wt=NGDha;)VD-9x^sj)n9%k$4$ znwQC1?izY|({H)MT6I+d6Q&fqBaH1&+LZUOvWZMl@$j9CJ15iq{`#&=fKV0DQ*k^K1F`#V95NghwH*ZpK?Rrvh`_zRe1s9LySlR1;v1^`4 zmi?_zv>t4t3(sK}aa_z2X6sZ@w?>Z)K= zM}+A1xbwN@Z5f58XW7a2om~Q@9&xVf!Itn<(f`x0h92)f2vND#&PT)FrROt&yYf59 zjts2gB^O4St|VWSj{SroiJZ!+Li-QPWF#Sb!e7c^GI#&B9fwC?dj=aelVXz zyD^A{VzXyBVrDpiv7 z&sL^;?j3~mysbZ0a9Bc*J5fSW0a_ZCOkucWmrzZ7FfP_iyC#do-^zN6#=)K85b)ju zl9(eO4VF43uHME#lw&s)il7l>rri#_h-I)3l5GU*-S3PGVS$#Ki}iON5-QucKT{8U z@mjT_N$Uq^eCmAK=lV;_!%a_)`K$Rizjkfz^~3t+lbe43SJ3vO2ESzuJ$C=hpP3U5 z47+PR*|p;ab7Q@Z->F6|I33xr{>ZQ&I(FRfbeelB?VR1o&ZAVT-S^LHvg&>G;ET~m zRz4n*9^tfTp;OFK!%~kyT3TA=MOmX&2mW}@AJd2gj2%1nc*E?fP*&nTXXi)Tam+i5 zk$<+?fgxVwFE}nb;T&z3!d5x?@5rbq6=C?Sgr)T`F@cw&3 zf|eqo!@Va@0!AIaSGk&c@U5I$G_OyHl_CGfEW62`zX=e+#fA3wy;T%QlVUee#e!GUdH%rBigR~IFnTf;#VFKkwLJpF@y0p7X_?;-PeE1Ym~ zlT?GAMBU=p46zCg4hu6Par-t!5KpBT*2tbp{z9_D3XTSt>gafzsso#4!2bRFt-JV@ zl$6YLcQ@sr^xK3_(Fteo$~N>*K2J-mqpRzNBc2D49i!c-(QB@Bs-4|Upr|#gs-W3U zoj&c$zsZ59cuae7^Ys#T_??iMIaCi*?d``pIc4mPKFz&^tdM)@;%Cn7aMCD;v94RX zV#SeDr+RTod-m_Y4I`d*?_Okf%L4mOis@Nhz8vwJ9Sat?b~n!(5wz;SwrJZfe#=gL zSLGn?GAcW{B5I_3hsKJ+A@9iU&veR3UvcjC?b|0-^tT*iWMq_5J&_~mZ{51(;a?xE zpu8UZ9KU{FV;X6AlLLO_Pa58ASEG$?pkH$WfPumP`cZSnD?gT&b!)WQe>Rob4$|Sv kYv3m+sQ&-=uc*>Gr#G{zQn&tU`AX%;@gq(QoBrqj044?n-~a#s literal 0 HcmV?d00001 diff --git a/docs/articles/opt-memory-alloc_files/header-attrs-2.5/header-attrs.js b/docs/articles/opt-memory-alloc_files/header-attrs-2.5/header-attrs.js new file mode 100644 index 0000000..dd57d92 --- /dev/null +++ b/docs/articles/opt-memory-alloc_files/header-attrs-2.5/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/articles/potential-optimizers.html b/docs/articles/potential-optimizers.html new file mode 100644 index 0000000..bc451fd --- /dev/null +++ b/docs/articles/potential-optimizers.html @@ -0,0 +1,365 @@ + + + + + + + +Potential Optimizers • rco + + + + + + + + + + +
    +
    + + + + +
    +
    +
    +
    +

    Potential Optimizers

    + + + Source: vignettes/potential-optimizers.Rmd + + +
    + + + +
    +

    +Potential optimizers

    +
    +

    +Inline Expansion

    +
    +

    +Idea

    +

    Replacing a function call with the body of the called function is called “inline expansion”. This eliminates the function calling overhead and also the overhead of return call from a function. It also saves the overhead of variables push/pop on the stack while function calling.

    +
    +
    +

    +Code Examples

    +
    +

    +Unoptimized Code

    +
    +cubed <- function(x) {
+  x * x * x
+}
+
+inline <- function(n) {
+  to_cubes <- 0
+  for (i in seq_len(n)) {
+    to_cubes <- to_cubes + cubed(i)
+  }
+}
    +
    +
    +

    +Proposed Optimized Code

    +
    +inline_opt <- function(n) {
+  to_cubes <- 0
+  for (i in seq_len(n)) {
+    to_cubes <- to_cubes + (i * i * i) # function inlined
+  }
+}
    +
    +
    +
    +

    +Benchmark

    +
    +n <- 1000
+autoplot(microbenchmark(inline(n), inline_opt(n)))
    +

    +
    +
    +
    +

    +Memory Pre-Allocation

    +
    +

    +Idea

    +

    As a general rule of thumb, in any programming language, we should undertake memory management as much as possible. When we grow a vector inside a loop, the vector asks the processor for extra space in between the running program and then proceeds, once it gets the required memory. This process is repeated for every iteration of the loop. Thus we should pre-allocate the required memory to a vector to avoid such delays.

    +
    +
    +

    +Code Examples

    +
    +

    +Unoptimized Code

    +
    +mem_alloc <- function(n) {
+  vec <- NULL
+  for (i in seq_len(n)) {
+    vec[i] <- i
+  }
+}
    +
    +
    +

    +Proposed Optimized Code

    +
    +mem_alloc_opt <- function(n) {
+  vec <- vector(length = n)
+  for (i in seq_len(n)) {
+    vec[i] <- i
+  }
+}
    +
    +
    +
    +

    +Benchmark

    +
    +n <- 100000
+autoplot(microbenchmark(mem_alloc(n), mem_alloc_opt(n)))
    +

    +
    +
    +
    +

    +Vectorization

    +
    +

    +Idea

    +

    A golden rule in R programming is to access the underlying C/Fortran routines as much as possible; the fewer R function calls required to achieve this, the better. Many R functions are therefore vectorized, that is, the function’s inputs and/or outputs naturally work with vectors, reducing the number of function calls required.

    +
    +
    +

    +Code Examples

    +
    +

    +Unoptimized Code

    +
    +non_vectorized <- function(n) {
+  v1 <- seq_len(n)
+  v2 <- length(seq.int(n + 2, n + 1000, 2))
+  res <- vector(length = length(v1))
+  for (i in seq_len(n)) {
+    res[i] <- v1[i] + v2[i]
+  }
+}
    +
    +
    +

    +Proposed Optimized Code

    +
    +vectorized <- function(n) {
+  v1 <- seq_len(n)
+  v2 <- length(seq.int(n + 2, n + 1000, 2))
+  res <- v1 + v2
+}
    +
    +
    +
    +

    +Benchmark

    +
    +n <- 10000
+autoplot(microbenchmark(non_vectorized(n), vectorized(n)))
    +

    +
    +
    +
    +

    +Efficient Column Extraction

    +
    +

    +Idea

    +

    The idea would be to replace the different one-column extraction alternatives by the much faster .subset2 call alternative.

    +
    +
    +

    +Benchmark

    +
    +autoplot(microbenchmark(
+  mtcars[, 11],
+  mtcars$carb,
+  mtcars[[c(11)]],
+  mtcars[[11]],
+  .subset2(mtcars, 11)
+))
    +

    +
    +
    +

    +Drawbacks

    +
      +

    1. For some R classes, the [[ ]] operator and .subset work differently. For instance, they seem to be equivalent for data.frame but are not the same for matrix class.

      2. +

    2. Moreover, both [[ ]] and .subset2 are functions and in R, any function can be overwritten. Thus the above optimization can be made to fail just by redefining, say, the the .subset2 function.

      3. +
    +
    +
    +
    +

    +Efficient Value Extraction

    +
    +

    +Idea

    +

    The idea would be to replace the different one-value extraction alternatives by the much faster .subset2 call alternative.

    +
    +
    +

    +Benchmark

    +
    +autoplot(microbenchmark(
+  mtcars[32, 11],
+  mtcars$carb[32],
+  mtcars[[c(11, 32)]],
+  mtcars[[11]][32],
+  .subset2(mtcars, 11)[32],
+  times = 1000L
+))
    +

    +
    +
    +

    +Drawback

    +
      +

    1. For some R classes, the [[ ]] operator and .subset work differently. For instance, they seem to be equivalent for data.frame but are not the same for matrix class.

      2. +

    2. Moreover, both [[ ]] and .subset2 are functions and in R, any function can be overwritten. Thus the above optimization can be made to fail just by redefining, say, the .subset2 function.

      3. +
    + +
    +
    +
    +
    + + + +
    + + + +
    +

    Developed by Juan Cruz Rodriguez.

    +
    + +
    +

    Site built with pkgdown 1.6.1.

    +
    + +
    +
    + + + + + + diff --git a/docs/articles/potential-optimizers_files/anchor-sections-1.0/anchor-sections.css b/docs/articles/potential-optimizers_files/anchor-sections-1.0/anchor-sections.css new file mode 100644 index 0000000..07aee5f --- /dev/null +++ b/docs/articles/potential-optimizers_files/anchor-sections-1.0/anchor-sections.css @@ -0,0 +1,4 @@ +/* Styles for section anchors */ +a.anchor-section {margin-left: 10px; visibility: hidden; color: inherit;} +a.anchor-section::before {content: '#';} +.hasAnchor:hover a.anchor-section {visibility: visible;} diff --git a/docs/articles/potential-optimizers_files/anchor-sections-1.0/anchor-sections.js b/docs/articles/potential-optimizers_files/anchor-sections-1.0/anchor-sections.js new file mode 100644 index 0000000..570f99a --- /dev/null +++ b/docs/articles/potential-optimizers_files/anchor-sections-1.0/anchor-sections.js @@ -0,0 +1,33 @@ +// Anchor sections v1.0 written by Atsushi Yasumoto on Oct 3rd, 2020. +document.addEventListener('DOMContentLoaded', function() { + // Do nothing if AnchorJS is used + if (typeof window.anchors === 'object' && anchors.hasOwnProperty('hasAnchorJSLink')) { + return; + } + + const h = document.querySelectorAll('h1, h2, h3, h4, h5, h6'); + + // Do nothing if sections are already anchored + if (Array.from(h).some(x => x.classList.contains('hasAnchor'))) { + return null; + } + + // Use section id when pandoc runs with --section-divs + const section_id = function(x) { + return ((x.classList.contains('section') || (x.tagName === 'SECTION')) + ? x.id : ''); + }; + + // Add anchors + h.forEach(function(x) { + const id = x.id || section_id(x.parentElement); + if (id === '') { + return null; + } + let anchor = document.createElement('a'); + anchor.href = '#' + id; + anchor.classList = ['anchor-section']; + x.classList.add('hasAnchor'); + x.appendChild(anchor); + }); +}); diff --git a/docs/articles/potential-optimizers_files/figure-html/col_ext_benchmark-1.png b/docs/articles/potential-optimizers_files/figure-html/col_ext_benchmark-1.png new file mode 100644 index 0000000000000000000000000000000000000000..a70b82ac9fbba6b5139ae0701242143600e764f7 GIT binary patch literal 49512 zcmeFabyU^q8$AkkGbTzpBZz{4K`2OA=n(;>B~_#aq&pO2Mnn_^R76m^ML2Y+BPd88 zIMOImhgL%R?icv};;wb?zrS_Y{kq15Gs5|N-g=(>?7g3N{yDFBdi@%nUeW zDl;)HcVc2%dU@3f{1@FNHG256;)?v~llY2X6DM*!@n!Y(GZ$@{m^kl}-^_j;!-h;u zJDDgakEuGn=xcV?46vV_8>WAIQ+mbAR_VUB({VJ5}Cr&YoKjbVQWJ$L5JSJPuB7T&c z{_*3H@yYy(tkI_2@~&C8ePY37yXRkwiOH+y@R97rKQqaW{xP&<;a|Kc8~;!J@rWc| zC?s<5m~;nb;9 z{QLIV+3hu1($v&+HAR!>p9*!_+9_j!E_TL=&*oY0-yfIOz95_#7_ZCB%v`JGw=I{c z(km-7Gv;Ed(F38S3b%ZGd~$Phe{^*4YG`Pj4)Wf8^Vj;(Sp4nJpFh8+ZXGR}tzSc1 zX>4qq)Ny|Bxv2SWadGj}K~+Egk4@8mA53=Y|G|Cw-vwA$fcf((`yn7GDEh~btDKS! z=XUJaQS#->lgE!AH#IlczBQDmn$s2W7f<5j52cdXR%iK6LKjj zDQC}|Ie+ZfyY)ei*47tHOkyTSd!uhGU0KrDnDdf<{%4q$r4kYN=%*54TIIr}S=?eO(5F87fO%U7C(b0f1BNcV^K*g6wWB>j4WPfvx#>I;l{gZCz1&dqN zb(RI!<~UmUvWqK*NIE7Bo|uo4VV`%B4`x)pI-!+${nglDtG-!waBwh}+t4vYq*&vM zmv6Na#hVl_TzHf}H6Z!%-+#M%lMB>6Jo0%=2V3$QavYN<#|M-3XBZl0HBky7oNMRb zS0Uh|QEEMh?lLCYNf(#waQT=+TwGi~M@CfTvohVdXLCkL^P zhoX_Jr>3a*z+2$BO55OI@{1Q7hQ`K{*;_=IJv=;YY;63?jN)r6)WiN7tEek-%e!~y zLs?nbc9va|lEEar%E-ucbaeDEYU3XWMP697YE?{~3zq0y@&Wz*bfd!Hr%z)d<%9Ue z#L{VY_+F`*ks7umHWAnp=DFNGT8A^Q)gKhpN{gZ0j=X#KE|+WH0jbIE@Sj6NF#+zc z#{L+3zj2k*g;4x%UHJNK)v{cwj~A=#T2HiHHA&mApv-6hD6aXj=S$ zv!tX%`Q*u4Q3}E8#=VNcVhT7hlcoBzfBAe=ZZ|N&Ak|hoCF^9p&dfaA-$0dj9Ul;^ zy`nx@pq?5o=WFw|;2-amEDUcJ={+@bR;fm1Bb6sudRPm7-)=~^=zZ8ATe!A(XWJ?F zzwmq4{MXyN{uruDVywt5r|PV{Utgobwp%t_rvL`1;BB?^hxtBljLSaN#ZsxBBtpHy?0G#r@1r?X7&}E$7P?856VT zu4fM|QCqw$SX>FCtCg9XP$CD zoB0Y`ONr&^kQc&?u$7h7#eLhbn;3VwJt{+INBGE;#cL$jeR{O3RJYYAwI|$9TDdXZ z!Zd)+bTLuA*>i4&Q4{T#ZP)XlN zq$`FYq4+M?6>c!K9$kxkz<+DFQ)ZlDH+%K zCCtM7KYDs%gY;Ztzu2*|vfjRXx0l89x3`}As2Yys8N8TD7PB%YQ%Em2rggegeDuP> zxESpWOGaH{@bL8X#IMn}NguCNrB+DC10?le^lMhHjvdMNtxYi~CaZ{@-Q8E8Dq-7s zUoliVQ9vU})rDP9J6#ofaUf?<`qasjpUTP-HwYwgHs?5Mj1M&1kM&(l)Xh-@6AU^dIP%Wj}4OWA#V^7rqUQ$E{%j*kcV`S}r~HZUOM=;#Ot_X(Kr^Jj{pqT*Z4 zeQ)cp75UjEj;11Ie7nY4$BG@c`F#BP5*dl$$?xC)!kDwNvr7=qz7|ABM^`1v zVQ2s7>uYLzZ%?L3w!Ql5620KEsNcVxEE>`%cQg8cH?dKuL3Fx<#Xi!iFNDj*>gG65 zm5nK}v9S$rE_-9oKY95wNz9_bq^NL!|rE*j_=_Z|A(8@klgpPcBm4Q9 ze)w10Do^e4ZiQ-u+0k}C88Za02C8|`rp=p6_qleIY}L)OIYpsRHpHZvk!9@ZFSX(a9F=jVP| zH6Y$wK+Vk5M6&J}bKXmlbR3`|FqNKb#A({C@I2zvr%zh3HAbnXJ=Kv~`EJg!J)b{+ zKC7r$hZLMtpK8pm%kxz`NzLv>0HsEK$JMWV6@<43%aCw`jKu_sSJp%&+Wa@`)8!a+ou>;In z^YgUQOfEOC7(W30;~1@U@1rFi1VDAeJ*0^k7%K)sv1F# zMc#Nzjm+#YOS5xVNTimldPX^Ng89VH4)R$Z<2kLdSRNoBWni=`1_q<_f>|};g6X%? zbJMnn7jZbd_Z~c`qs6M&{~o$5c=7GO+f7sX`x3jbPcMs-$%Cjb5SDym7<>-X&;Qm*NL?bR;EiSoYQ>ZoP9Ac$k6V=swX*#wi3lPu%+z&^Uj&a5yRDB7uLd5;Ox4!xK=K+qsW05n_KW6*%Iks-yx?}27 zMTLf*-b))>+koKU_`bet9Y20-e`f;tJ^VN%MA!1L)>iBzyeb*1SsLeQ1a`SfSNHC| zzCN8e_H|RPYCrSd$mGxF3HrG|E171GD|sv4D5sENJCn|zTGm^kk(#2FxYgqZ6D{3e z|N4RL!VK@3U-wbazw^kapqi&LCe5q-=`YmdYfbhmxBT3yYu`I2vo;@*R*@z=E03N)$Ia2vQ`uo-6)o$Fl0sMC@q)+_tVa?>+pZBbP9+8wZ zFZ5!@R<3TH9TRt-9xjet5gDf#GBGhRIrM?0qqjG{V0JV)QmZ68E6WtYo?l3)ZfdX~ z87HcB}DA=cuOZRV|zPF=-l?FTXmFCt=^9)K(RKTI1{bjd>f9w}J60&MGO<4g93kqgZrufML&8uh)Lt;Jf|6>n|dFe0;!q7)3Rw!h+#u{gs^&gaDGxI; zGH@L3x9aKX{Tv$;`lVvCW6IC{B~ENO%Sw`g*k1VW&{nD8(Ha^XXx7HIvF-d)^-{`p z3zqT$r({WSakQe9r)6utdsa=Ndc1@~UrhLEKSoJ_fLx45h=d=P6n{^{PK8jnpP$$F z9`TrVqWXO-M_xcW0I;cXA8*z$VoLSzOBC#H-rjTu<(<{xyw?xE78jpZ9(|tDsGl=C zGo|6&j2&UooUJxIJbacy*_%2SygY5}(l3_>5*nNzJuFpe2Qwl7+8EzI`gD}{eoYBa z(#z9|r`GaK3G!htHe}n!SJ9JDG(VWo z<1mrEjs;WA>*pTTTYX6U{Amz`LymTg0-|3a&&1@Ux@jIVbT{yD&bNc&u89IJ3LN%H z68C28Kb4o~_q_dZXFZ;Taz55-Jl*qek`7Sx6pd^X)QUmQLD4~q2HzWX2)*<@D;nbO zPfxpXO1pjSj<1*gq~+q`@_3y*j=t})+P3SoTn?C-` zP(LXq<_snonE1d7&D3#oH}_VpG3L6yg5*JcdTnj3lZQK}7th4hlmH)Jv~uWjSp`I> z9dmOC8i%tvM3?~zz{=_F8}`p;K7LVAHN>h5=g!>^+;{P{aQ^S(Yq#y&uT`yHSN8rk=5~#CTfOj}WR@H8 zTG4qdCE~4aD&M|+Q;RF$@B21@8ZXgxAS+t6_5D_BdwVrJ%A50iVasF&e|TebWMmH} z_A%%qlJD&Ynl6G6kvo4uM6m?=p8(@tbo$r1RRIyG`^4&fc9FBb|MBk4Lk%eSP+jlc zv*(QWiuKiiBBCe=5;aqfZ{|}J2^83&w~duGsIj{HYl?w*6CEH57)ag71(k}meD+Ku zl~Gq8tHMX2Pr4<>du3eBbVadY-jHS*zz_;kiB^n+vWA&=4|}{rW5VWq1Sn%3obwtA^x4%B<$*JXKk97QNY4~7{r3L0%V6Gk z$@FC_?!LZq$-#t?OGmcWnug|Ex64Tl|GRk&$S2xZ#GLsI_&4ouCL`qNx((d`Ba<*R zbUkm95g)LDKii={4tandMGO*nKvH~qQ{V5td(cmJkX1v&Q8w z_S&+(Lr$_E>Av7TJJI#?*RPs~f1cQ~b!#LF9(va|RZy7z1W=-`jBN_cZfR*DMfLIH z$7v}BrwRB#1-@U+X+4=DSzR+#&p!vjG3=7NU&%f9yI8q~o~Rv-l0rA%t$x+XYhuAd zy4evd(!XW>Y1oek8`%$D{hOyp*mJ@Me{nUz5tVe_&k_OgewUqK2C(X)TK3zvZrv*+ zq&S%S>j6(sR@cjT>ku4!7rZVnFc@QKTwLb+YG-x%=GAxZK@3tOl#F z9u1*Qo?1f95D5{tdI?g%BJF2sd3hbHzS}DR+S-^-#*3vm!E6Vv+>*P${c^VANWXWL z=ntmqM@S$vq+ie!0^l+$q^A;ETzUoFe}3lE%CHQPo*p`(@a*6z6ew>3dxK=D>$K(sR0k z-9!wE?ysSH;;hR{6`)Y0k_uHP^E^f&vLg5S0fRfJ#GV8OMxj`d_a{s@=w9{Ae)6$~ zOLn^!=9trkPRYrsXIkoJ)elpucljN?vaGEpT1nVZ5<4So_%L89U?v@DOw_bG;$Z3Q zSVloD=$Uw3TlxPo@;Ym#%4~INx8jQ!*TFn8P1Sf`MEfI%zj5+G zLN48}?h^Wi{hK}oY4169QC*U5JRqwUwoUfTSfeON3#p;PrJyDo*N?km_r-%Q)5@}m zBs@z<2;uxq$pUZKkVKUb+~!mCfHNLUo~FL%3>Du3y3HCvrzgG9Ql;XhRBeWpJ`Q9$ z%_UKbTK(dPEuoTU0OpCj5n-c!qv7=Vs$Nf*&%?QeIS4m z@09iZ0(J+ zH}3tOm;f6?5erUi2u+k#`zoW&{%e$mgXgg&U$uBe z|Jjy%{RPO=c;{c8lPIHBJN14-{HI~V*l6DtS4F4-Y&q|NP@a{8v)zw?LIZbPz6PN=n$km@Z3p-g|f{K zjTT^K1iYi>Q&QVb2yI?#P-@8rn9kk%?;puj z{C6G}HpGAD;r~nXP*T5es7>4uqc?CI4nryTiB8vm^b>^S&8dszEgOEKkWl1d>V`%} z3?k-j+O+AuDP%FW{g=F6x-j;lGClG}0ywMawh$W`K8+1T8PvOgHIBTo&_O_b(UE4h za&ig=mvx4%L8xJ4T|&%kOjIM_w8`12L0-zsmoJ|mb$-j`{9~iDlhYZ0P!}Urr@4DT zyzu@&?1@=Ud>TwlW@Y!b3fwOy>?pzD;w?@mdg;k@(0snmPy>^5VxAUz|h6 zL+77={_&3hD(@u<^MPb!X_u+~3{ik7g9fDC1aSYcExpT_m`u*CLfFP2?)5iju>0Lz z&%T>4y4uqcz)EVo$(BvpEi=M=<|i0ePr@eT)vab?s(iX&aEsXr!%0#5T`rfHCPPr7 zdd?kNsaCCe{&q;Az$pC_^LajgJ4@za^Bdb2`q6OaBcL%dDzId@oUT^an zK790Ou5$0Wmq#yxBJ@u~NvjLlOwo+I!k5KY_zKG2w^+($B!(uDsGU*w=JE#~VtqmR zqZ>kl?XdFcf*&d7(U&jhNF`cLopqdljQc4@^aY@4_QuA1 zch|(aRsYMYrQg554hWaxPl{r22{-@bi&zj(|mT>?uzP=Ljm010cRirgrxa%^$$Y3>VTOZ04Y zAK&3;U|?Xbi4Tyi0SdwAgdGFOxeEr1hp0}#pVrq;;XkkEUKsm|cNh<}(EuJo>VeGN zkc5GRecRPW5^?3Fk-^qhCk|HDuEMkU_b)f*6S1tfb%oUUcJ~QDQ4RPp-)+SX61lc; za@g(@KgX+Aulyr03n;xMGjUG5?)WbL=<$W&El%pO9XQ{^PAFG4B;Z{9j|Yd@f$M_c z{)7jhiGSCw4-3!GS$xJ8kQt!8*(x0S8#)Tj@M>&g3mbTSdDt=U`^8X0?Jn_>fwC_S zG#ZE#bjD*I&coMbvA7EZy=>!P$1H1@H`ZW2*%KX_<}`c-n!}Oo7ZQiA{QI|Z1yts^ zu9vP!C8cMSjf|rFA3p|r5)C`p9xx~yH*S0m?OUNYW+0oWqEsAs<(j&Liz*$4WCe_< zTBGorm_nih$U_b714~Co_gnJ1982_`q6U9-@7_HQUAd5)+3mpE@zirDKM65$ux9Pr zwN_du6zT-`?aS?%QhkO`WbzpH(Ko@T`A5u6cgeUw&gl3-bFw=A<;xeqjnS@>=2L#cC3PSBkQ1&fZZ`5h|lE+}qH+ z??qDILF|AY=w*cZXmR_cK&gIx%7*}3P?IY))LjoAIyBlPGj|a?UH6$YG*g1nu?dNW z#CYjCa3CoZp)Ly0e%`nxLGp))$-^RF_UJ%C?Yhzf7*sL5K8Y^FA6X#*zeZ8Fmyho> z0n9j6cB2e6f{|%47eL`rdB|TKIB)>o3s?b)|4ftE(Z3K%Zn(^10=K&;xbapNmQKlZ5%0SIPbKGMK;Am%HXnwrA6r=MyOrFk1s*eY zZH3IIPM?l{^X5Dux9;yaM1u$n!DKHeOqBc`mZ`|8#6E~xTGZ|Lb*b6k?LUWyW0Q$} zaD!~2kCx#epE#Tog9UT>h>QoUpa77WnneKvomL(dS}qg{-wAx$JwHD>%I&ReY-Z;& z>aRskn&;&Y*7;ApslNur{t1vrhO#kS^8gx-wTwDY3zY<=7 zc~;`oDK)2r9LK>r5J2)}spJ2cr`C@Yv0I_=c}1H>Naf|_HH9k;mWQmPM~{|VA8KmYFCJ5*KiSXb2!^?Gn%a0CS)ih#sqr-fWb;d$i95ttE@9{*a& zzm;dr(*p(~SW$dRJgQ^)z1)8vMdMcj!}qMeZpR zZ&k&vCJywThUM)Tx`=J(xiUlmgiH?49mOuX3 z3H=jN{9efS*r1l@S;N=Y`>t@p0kfnnO+wJ4VSbhHc66XLHOMT45@0lDBoAbDae=6c z!Ca$8tps06s4GqYH=M?zpaDcNP-V*S0>KA%nvkRmZjQRzc#Qz~oIoz~<9I|XhLBv4 zVx^>{G+)}nej#DkO-?4_ZfRqqwxy+|jf2n(U%*U5dCRJIEt-FV$XMjti695ws9uu~ zO9*O$SdZi_;xRmMvp3X}^%9q_-L^+iu$rl(g3Yka|LTL@`#bjJ? zEZ__aNJwab?20GeN^B6qDgZ0NaZ4P_`?ks7rbpsjV{gU zOOHuY;-AB8sK7u??4|ABO8`#5Q^U2Qj&OSL;6beZd9&a#rsUUa--1abJI5ckwI*d< z0YGQ0Es5QRb%=)7pP6D6U}eiODt|$|f)v7%f-%(|@c8iI10j-ay6T0S^xcL(D(fE5 zg!Z*!g?gsK6Hypils#7kh(HHWczHDIr9hlMea@AM0ryMUec?Q>?9QH^y;%CwK{w3e zx_!nQ#Bcy3;g-Xz`$vE|F^BpTpp^ju0wb@(E+Z@~f4?~P08E~B5dGqHYz>l>t{tgw(^8op>lxk$uxZDb2L-3y* z>({W^xRus?X!cJ~ zj~~eWr6>i(p-9(Ed=m#p0?bT)08m;M0ooSClmh7l*vCE(o>dn_>#Z9%e(QeAIHjO1 zzLe?wjTZH=!jnJ}{1>Ml)BC$h}@TdPOkCQkweh7;34h+|6?r(=`#2SXL$BZJ}% z;mZe6zo{5qzmh{B(rB&5BotZplPA1kkz#MZEFwCVb2&5cgnOITh6jaIo&MO*qFb4T z839tX1fN-DVg3Al6u^SSY8Ugmgev1ucZUy#mUhP+!kPa>F8kly084WIhZ|r^hO9<) z&S1Xww_G$H>E=46aK*>UEZFX^r4`@3^RcY<(QI#~N_J5PXc*z3m(v)Q@6mVR&2&MS}%e6Nr;T-+eRysMijL*_pw) z*<^^p7EW zdM52ugFqD=Y@U01OxnrY+q)KaQ_zxi^4@Evb`rUaAml3HVQha`r!4F?7E?JG-cV_) zEH;ED#t7qxBI3md^)jp8_hR+ElQ(MhnVHpcH#FK%_pxU>FgrDvC&xmzgym+>Xkr(L zY;zs@|EuXBhGyZ6Go(-pumShHaF3SHr|}k7C1m@^Kwzm05HV|$^$XTyv_r?nulj3Z za6gqTbSq9deBsS+gnT~!1*r)~j}#uFaRow$@i7$@N|?fq z{N1ANWz0-5>z3$=uo1gAk0}a5@!U7UpZ_;!K)Web3m7E=iFQ4k$Qgo8;drH9`M5^m z#o?5?F%l>)?c{YwNBZyK_EFTh4ERE+ zXsC#SW3wjnV?P$GA>TcR%j5TDNR?;(fhTz=M4qhD5BhlJ0#+9qALimMTWzjovjmwH zK!gC`TWk5SJYi7^HmZg%8CInxj-xqfrJ`Rblti;{n1#G}F#>rh4#yWhK9AVz;s8Am zjH_ViE!CNVF^KV*FO&v~BH+oB2oP(;C!Y<+9!K33FdTw=%YQpw@%p>*l(=G|m+1yzOETlQ1!?Fu7U4(1p%9ZhWWJ7^x!MqW$p6rF^QmzhE zb43cJ3T-v~FIuA%*pUs@T&KkSLqo0BF?P{w`+ z+6-V3Igf)$3JT!p+C6OuwXX;A#`Q=8$JF>B9gbnr6G7DbS9iIpp-)>ZtF?a>6Vn}4 zBcJaLM7^46&h9e zU&x>~UC%C-@P2|*!mbj|VJofEz3!n8OOgr^NMe{H5 zj%{a_9N&Ly5A%uRXLhdsRH_S5De>u|}{IqZ4$FnYG zFNb^Yp5d*S(j;BS`Yt#-XYrV#Aqv}R?L0Bozgf!Zh64lS#bcgk+X^Wh1ijR8Q*%-# zf!Wp0Wnp5_GtZN0Gp|<) zSC)$y^)-$il^z%^x@g%a&BWAJgkAF|w*47@_(k3gt;0L54_i4;tvY<9yK4A@c;=p` z-5O*TxUK#Ym~ebMfIL>b*@SI5=C{D9A7>yOe8|dq%57|sa?82k>a@aT z5OurGBJ1biO@1)Ubb%mETWe@&2ronc55gExf)`(VkpmR!1IwqTrsf}k1P-EN`ND@U zJljj(>FCRCoO2yP`fQ~LmR6TVF%Iz+3ol#-->>Put{*={ayT;@6G3d8^}i;@-cH^{ zWYYpMXXNASD;jv{;KAq3if7N>L(9${3fd_jmUfs~6S`S^$$vhJiz^9ZllOm{nyRL* zUYFoNd`Jcp4}x@(St*h5TrNfi-!A zt}%lnQc`)Ji>=^3Kv#o*1R~dgqX?w3n-}PIlSd8AgYf&+L>T1hB@;$|M{=}6aAyYD z+m_1~2uZE0m0`!;Y6%xm&&bO^#*;@e=qF3-yh-J zBhY|?62==)s`pxQTwFCs11e+)iZYV2p3}GG1}4OjEW)Xzj4w>nug&R6XV0B`!h^$O z(oqsnln(4kjC6#4;`wYwWkd$)DZ?bWL|FjqczWZOZcIL=pM!X;(3i$TA#80xKtLoa zXh2%FDU^SCL+{su$~hNS0v@`1|H6qbv_gCq1vNspQ!P<;Zsj2P)pE}ctweEnsd=0l6+W&&-%NJG9rPG%mBKeEifl-U}7rN zSc0;KRP$<(fZq4JRp-e?Ux|3Q-bu+wUgk=S{Li*xsL1D-S;teY;hoi5P!6Xi#zMY)@Rl!REKPj9yf(#lAGKI^j zjoBTV6^S7TI2v7qO@mMsv^)eiH5dhd(lVtbDk>^y2@l_Y=Iys` z|DRjT#&a!JN7uxHLpb1xY!gxu+*u|L+#Zh!eM+Cfvtbrob|FLrQQ zcyFcxxYX5PCF9xq*I&os{T|Y8r1F^?a1b}f!bEuxpMil}6UG9WU?Q>q>C^4#HkS7% zCRX$nOePXUcW42Ov}r)^5eu|J6L>+oMT#aPEz7D^x*#1;{F2z>%M|rHu%F;twsNI8 zEQcO4#Q+wT`(QmU9qh;QJzc%{Vn-bMo&sm^Qb-Fic;?oM17@^65d9XU8-M>z1)=!< z!hMSXML|t1AccdIvzfnZ)tWWcEFMEkDgMRnuUIk8%vZ~oXEQx5#<8{k@o6Ke^Mrx~ zDm&8RASKm0V=p@rBgm=8V>G6cT7jL$EjHVg8+A%~X1`2Z0j|<5X_GDh-IL7g= zwr{i?NFnFE$ueUWGH>aX6dl`8VrKW-o>f3lOXW`oQYDZM->D?HP)gvKnh94KgDN@LM>M@R|s!|<#G$Wfd94W@8WN_V#j5!BG~ zUoV>M`tIg(B7~TknI$ngnx430L~K{6TA~n&;0HKY-J!H}1~AmvD<06x#C2J&tR!)l zgJUDT06CpwP?bm#Q`Ae~basCrhk1(TNNVWJx#YaD=P1HOuYP*CV%*}KD?Yhwv~Vd) zH)*hG?5~4FEaEzLZM3icP^IBS0KHcYeV+IM(&ss!NZx<=Kqdh;>{vAEFq*OsRj4yD zHHEEml6(rTRvqSRG>KEOzu;5Wf({v*mPUHj<9}6Z>**z-kTyY#A$khL%qYUC;R1x@ znod|heOm(kN4mrP>gw}Ns{UHfEj=Bl1-Ak;sNJ}W02+ZljxC$G9a{mI^GVn?e0s!2sXR||}Lk?J!eF!b!@s4n8 zz)!$+a2Amsn%0XIEL{6?np-YOa`~MmLK^}wieA2+pY9KE6M?Gg6aAzeOiaI7nXOi~ zv-4pY^5(AtNv4}$gAXSs-KsT-92HbA$ayNKPTfwC!iZ6qYk?NWv>PRiyX4sy$>-Vh z(Hgpj^7PRC03W`#u{#zU06!Xj@SYDWG9tUpw6knF)TY!`RZXyl!%_G=^$`ltY%^Nx?M<=V2fsf-9kqaNI6HMP}pVM~~nhNN}!gML}Yv{IM}ubzL!qNG5ts?1htTMkQtx_8u8QJr6J4{ zFFt`Oq~`=by%z?u7F(;YZ;qj?3qbWm*l}Dnu@_w*ddV8N!T~*;<{>NFQ2Y0SW+0+D zy5DwF;^N}SE++na@SvukTS(`3UY8rLOb~IOv?JsoFofI`%s$9pDs!_lRJ2n9MfzrXtw zy;wr}vxqCMCUjO6!#B5+#Ji$@stW2}z>60N&{%P&lsMxnj9o;Ho|zo20gt=|LOH@+ z`;i-)zfPAH?{?leZ1V`O!;)z$$=ksxFk5Mc2; zhV?~FTcsf0Oyt5C5)h$%#@mlD{h^5sg;QNraTABX8}JolsYW`aUy5RWx)CQ=6w6cg z>`<{2C-ea#aT8-CteC_KBQ@2RTI8*H{`^B?s=(X{AoyXaG~Aa#$~$Z3)I=*OIcxCR z2lRQ}4!w6*QuZKvU}q=UegQ$5Y*q|%1^zum>^d0IlhJlWycS90xo~HKgq~IFEr2tL z5_H!lrn$5b{u+>%@5Hp@j*K7oCme0h^T+X2R)nJfyR5qOHBLBCap=S_w+sg78g=ej z^=|t#QCvdk^YQ)CRSmbwj1C2IZaAGYy@gqV3slD44;Q+oSUR`;tpp=yg%p}?Xb8u= z6bQ|?^RaovBt0))eD;9j$ZPl)MS2d$*Tw;9lGBS?|7~MG{vpy0JQgd~Z4U!Mg-!vH z>@utgIJnriZ#C$)s0Yr8hcVzgk%i`gNW;d-mXxDFHIp4R^!K%vRre;NfoKiF;{QIaL{f z7{u1F$vdE@?TsP{YbTu3oU~Oy{dZzB%vp?RnVHx-8#vAEwhfKGkaGSRfsw_4j1kRh z#tkG=V_yxPgN7PjE#`@frEDZZ@DEaSD)ee|w}W4LM;WAj6nY+Mcmyj5OIenV5Td;T z$vlwkd;2Sgff~)oJtb(F&^KSs?ru|~M4BOu`CyX5Wu1aSV2u$ zxjMlCj45g8BAP3xC7its9ibpBs$oXs%S?9IaB8rKcIhY=@In03&k~L7gm)Ud3_r@LyLX!g9u1zK+&S0y*J;hBqF=d0+ zJu#Us1i}%tJh6W`0>M+^xd;WEY%0fwyTj`^*x2~vYx(_(!|&Z&S>fsQ@aK41n4rSnz@|RNagZWd6NPQw><|b!W-u>_jv6KdbyK>Ru3j@`WK7{0l0#2F(5h>z-&Blz0>Ku7xr-FG!*3480 z4YqFo>0UqylC+>kh=*NtKFVbFI$tzF z2P_aU{a;f~RCk0{!KKH*1V_xtsGhX-K7mbj_rssbbk{POH@}Y$3e@*I7|eR#x^>HF zeU~jNHWcdYm%RpA?&t4sVLh*ZbRsNbr{}YJZZGs{^;Su zKM-~90IDjt%4N-Bf5E8vJG zzAw1493;=3Fg1;hD5au!T1IZ8%%5ZkP=Y?B0Uu$<31NSuG#yEe>vId_ftU&5;HDYO= zyD2#2=#sky-~&t`w?-8e0dbV9l5e)mAW<1sQaxP$K#nc}qG(>nX+|5W8Sa|_;ZIA< zXyFekt^L7fW}BrWXc>OMZFB9xi-Ok;aHk27@P3KQ%$ysUh1tr+|4>J{G`O^WzS98I zTcv@Yl+k@Pb#8sD8v)!njot^1sQ@(ku(oj3doh#&KEcbKTlCWFy zn0|WrCvy^DdDACS)Iy$bA4T+^Iap<70p<(mVaz;uo6o7GE7nyKmq2rLT-;fVfIxJ} zH9!+6X>^Q3;m8hOdIBlkGP`H@kq|{}B>V%26bsA?mB7_oO26MBr%F*pVmD}%G1%;V z#bCUGO|LgTIlr2u=839Q?4L5sDcUAqdQ6n+6PcqL^8S`BTdul&ng2#VkC9yv@z|tf zTb;W&A-2krtGp1D06EYDm<**pBc>Cz`Z>Tl6|5slT3^2(1kVv7ke33?Ygu)KQA2b6 z*1lFj30Jvcb<@vU2zCUIl($+rpk{vl7dX#WkEui+a!F`5%vwuz9?gMyVVCmY1}RAi zDn30v%Jbmwzhe%~!h@b^4d-bU@qVFl)*`heI>&wWn9xpG4cgRqhFqu@a>P~+hr3|3 zL?BrwH4$d?GyVks+`6@l^QhB#2rC6<1>~+#av|DtU0YyLbO<-QWFT>CBA#>MfMy_9Nn>SzBO{yLB{We4l#I>igB;DDWWNT|RD<)r}E(3=S1 zx{Y*r8%)4$U{&Yz#^}49#A69)rvyU*f!KXPYkQD26JqQqq|3+RIrrcj*Yw7(wM?h9ZD4qOfZ`L8m7S&*W1D*Eofu zzo{8lVl?ErW)M&VgpEbS4NHnL)8AOYD}UXsK&w<$$YPa7uqHQJfLlZPlk{A$K01H( zIkV(stGd=e4(*EzyKFdcVJ5W8#XLLBhoVXvC+(gCXvu&sObkz}th1V*`0+ian=Kr#4~- znZ%<0d;hzA1aK?q;lqcEZ>7QVddjc*nd=Rq$PFOQ&}*+jL^nhYj&2*U zQIcKkYvM%FFJ7-1H4bA#=A$ZaR+B>P2R=$RXf_x#a*YOSRB_} zu8P{3bbaNt&M0p4>3BZ1@8a8&&Cbdhn&}xxvyo?NO=j~~ZBAABa zK{;#!oj`I1--}4;0mp<+wB|3@!k5LyWxIZ$|abhj@`RzDVWYv$f^z=U@ay(~# z-3CK#Ga#a(pRfqc)M2f?E-ujZLpi{;ptjEVr#TMlIm3jIL3%PeIy+%eMb+5%&lwEi zKfJt0mz{(tic5+zL>y2q+D%6NEjjBd@GaN>qbgi*U3I(R+cn!-F2+0YBJjI~z7G~y z|M;Pn2f*1vl3rs&_c#@ak@O;NCuk zVKsFxx-HiDeEgoFmrE}GO>yf@%t((wko4Mcd+}L2K=P&J0=61a5w%U16P&e>^2P-zF9jZ8Q^i5Q@4yfu)<{`FWv zT-t2)xKT65Au+|RE5va!Q4CLhu&8aslkzDg*R&>TH)k-ewuar4QccQx@U~M_eCPkr zBjU&f8AxE1AU=5kK|u?eHOLPbWDe+6JU#GXUaSw=IQ0Fe-LC(50Wf|h6E&jMpoZp& zMV4e2x6BuNJV_8|Pd34Y^I^nN!&Etp8+FylWbqJEEgLn6A+@gG zN?A=U64!f}fz<){tHwL}S`_^LDEw3h^VG(+IM><89|X`7m;!)A^_X!fdvQdiy5o`~ zDf7@G0vhdlZ3?+(gp~Gmyx90pal9!a=?!?f_L^5}mx zbTSkF^;9hEhyTvQ|Ig;(*g`D00ZE+TR{+;|5X4B}C@+b^fV&)N;ip+HCXUR9k3&6`n;K<@?9;BW&-~yX{tc! zZd>aBh7b-^Q&^)(lMqVnxcGR|8bz*)1B1gaEF8oGq8#>xi%?Ii0AokL8u-N$yys3K zpF9hzE=j*7@DHGYhol6MPin@&r5)9{hu{RcY?tV292-vO^=CF)5@r;+iw4pR6T}O6 z3daVy>>2jrblA)RcUl^VlEc)tW?M47cfH+4rpiqy^==5gss;T7U=Q|aPi17=ia(&B z1VM_1a3tz-?fnbF36u6Y(v+X-ftq&`H`&C2eiPfysEfvmkTzBH&Ot3OO&CPkzippp z7%Y7vv3c6 zYEi4JqU|kAzUL1VB+u=k;xg6req#BuEZ0OH1;PpV{cx3@6uI-Nd=eBjVHC(URp?CA zc6$lJodG~jgQUht2o~*+B!0^+e3jkVHk#bIB6nRCrlgWtRZXo2 zw~~^MLnsZG6Z~)?F_M@(?r$y=ewU;KF>7?Hbxw~6j1Ug&;Rc{olMAb{SEPP@IYjV7 zYdNt4wl%DRQCA4wV%R=v4pki8+zs2!Xv_lLK-z$zb&x(P;3NE>ET9CDW=XKY#MnVD zjl~teR$XIGM~V5JT%QK5VtX&=hPzyz)7FGd1~oa{IFUpy@du~X1Cf{xAD{8L@-f^h zYzCDKVuJMJH%R(SiGa3IZy+wg26GZF_G=@jr6t#3Z!j1P*`WWcy)zHXIsgCt9R@SR z7&De4Te4@(t_%htyKE_BNV}y(Du!WZl(_9hwrs5ot=foavZvCbB1t8(RFvvGUU!-E zJHP8(=UnIfan5zF>&#`o)0f+Q-=ELt{eHb)ujljmdcGd^6>?k}u)t}I)!QmMe^06= zf_JK#)g&+;T93LyssiJJa?D-6{1gu6UJf-t{!+V%izm!|v31o$Pgh@25x}$98D|=S z^olF8C0*M>z>2A}q4@qd(A!)6qKCmlO!wnNYO6T$6YfDjLaEQ4GpF^dK@AcD#{Zi0 zs@7I9@+^u6v3O)tS*1P^!*mf*;wPm8F9G==q67vh%(k|E?XYv!a9W6kVHNsry5|x@ zktUWw^7WzjHqqg8JQ&DFs`(DO*TeYF;0n!WP(G8;EbMz~+j6q-`eP@N z%Ul@`TLdkh4EcJ1;t>oNmv`|wlf~u$;Yp9mH<8m6;a#w-o&ejmB5T!HMl|3Z z*Aw>WEk?=i0}uA6*Ox_Kk{!JmcZ=EV#!G>9;-aBDhKKvG?_Ws^F%hky)@xH?&;?Lo z(uK?|DNV!>@JkT^7uc*&-%{=lb-NaNkWTIlI3_sZ$oSpBK=nL{exnpQkw z)JU_Slexb`cuO+)l;N?S1b#4LN8lL|9f2Lq11cY~>7#|{W2K4YbZ`YusmUy1X?;#J zbvA;l*ozKB&I>pO{3M9wQ~^^dcT&WiY06L6rIvawk z%WMm@Z5M#h)jN)yOPjB#92jSSr)G-INvPQaxAF$=_z9$VXzJ&fpsZVA$1_U~a<8B$ z*s<qMF)OvmiI+T;FeoG_~o4AX@^)wFNK#4BET#@yM==RcM?tkUVm64fXzWWhb z^JI%zI?Ck2fOYD$K1cW_wN66#Vcy1(Z3PHt_58-4X#wX(RGiS``AATN@|ho)$@m1Z zf(Oh?b2dV?&zZ)cfj;;lP#$yJW`P1qGXObD9iaz&F2-^)v%<;A=|qz4AnRiajiJd#ZZliG_ zqG02YNZ%yCipx^J^>L{UGHF}2tk^R&qJfO zv(u6(8kE4IVUW-$z6)ZF6kTLw2+CUaKB>PIcsR7p zzV)AtR{nRS7im@aFjx==H@(<9>BZUz_jL>r0U$6Ea1524htFhG0~FAYkXi+O*13QP zqs^`}*E}mz&t0f{_{1(tvHLy&Pt;5Wbp!(!vai(mF+GvQ!+%& zb-SkB8kV}2&b%#su6dkr^c?4XH=$z5m_>tgIrr)Nx7%5LL}!Ul(>HivA9DrfwvyQ# zocgFh&mm?_Nxxs>*r(3-LCx=q1CLn*5}btqkPne7wfdCW;(XZ*mADJmCTyV~Eq zyn5hpxKx?dRiBen(*r3gzZV5G*mG>gkdO`R2J-i0XyPxaE8zk;yBzPXQl<*X**zxb zLF(q%;w_=ycjz!JuDHp3M<`=>r9Jn}VL(z_aQ zZOezfle8jUR+4NR;vnL7Yhj8pi~2_ zZGlFU$G05gi@s|~3d)C{t(LgiK#N>`xd}Hn5_6fr2X8vDr)xjT(D}=kKi3OmBVLG# z`V#2?C^&P^Ye(mu(fU{V$Gjb8nDFULc*cP|_tT#_MZ}fKf`e!zwq$0`JEjb}KF=+z zS#5iKTF9)Ptc)#?PV!YfUBgcGEA6{8bZ(XT`_V}!1L1Sqedu)J*LJ7W&gMEFw)JR# zslRCJaJZkdZ#S2D4ASoEPkJ$s41OBq5ffEe{Wy z|K7txs2DlOKpBOvL3}H2q4A(bwbc=1P@mP|^^sEo>}*c5VnX~`|qcw23qeTD?M ziDrwd_9m3~>-;)%>=U2kB@v7Iw%J)=8=H;tC$@iPUiw6x%m(MZbQaQ!IryN>zU!hB zIsfaQ%+m3hY94(rUFWaMivpuUPtAH_y7;s6c=Lhw+KtDMNqH~S+V!xDE7eg`;R1z``wo3>ZGg1`tyLDM8nh^;lxC!maYg&bQ{ zo|4~lTw*n`QE&GmbvkcFVF~^r%@dR+vdTGt*3-R-k}31qCEtwN78-e~BWcC*8 zs#upTmGO!3*0r|f&n|G$VnGExygfX0&tv_*Wxj4RR+a8(3E`k>T`c_$&|RsG8~P+f zvsQt-FO>zLO8HwnK2~2HY*6wMi>f-dUp$~}(*OPi#f-EdpFxk1S~;3lcVP|0t)-*> z<|kcp1B>6D7m1D!tx~ojp0SM(1Iubud|>(iag<(VsnhX0BiV(bFe+ zsFj05YiH7dT0gr8gMW2#{uyh+zrvBKbr&A+sCAUVQvaoC*$5lHeF5Y%!%(GN+%vO{d(vywjYz zaC{dqi=)>ia&5lIRys6z1YF_r>Df-Hs}N=gubjG3IeR`ijD#71Vh|sVur3Dpk;#BG zg<^7M%$0BU_>oIK&y+@|?O<2E&^se}m`Tt*6Fi`C3 zOq?~a@sbP)Y3e@grOLiiDrCaZTu_x?2OXIUmIj+9Q>N7YHX<4j$2B2^(6wUba2q8p z9#T^9XkUODZgK1I-IXqAz#^eTc6+LWu=t3AI_1eNbH`!Y^yGOPUL4`?InZqVq?|9+ZxmJ9R0?2(3k;>|%7}y8hacoY3+csF z))M%E8E1$`8M0{q(f27#f9<>B^^;lB7$)+3R(<577zFy#lQ!qA7qevapr|dw!myY^ z1sZV0m7*jgYU5?!;^;z7Dd`JFN7%j!U$%*)b_aF`CcTYIxu27bPpN7+muRBT*vuGH zwja%N!TwunTM{1B4~Z1bA6N=aQa)4mO)){Am*De?uRIf^NpESKg2x$bY-2f1XWzp& zVwdokj6i%LqO5u!|yc6;-dr^Vm(+r-o!21 zauyO%s|)KhBlF#40t7yMg79FZ(-XK$L^G&YI%JIECBn){ zX($OWWj>-Gu`L(im4rg`M|%vYW{Gf-LhWN&npq!A;Sl!DN1H32`pn&kcyuQjGi{MS zNnIzHGyz`3(%NF6C#oaBfz#pP;k`Ru){SjA-@wh8M>>ZJ0$9$H*Gsu0GHpRQfMiDI z4^*5OD264d$izL2x<$IPgleNt?HGlp%o5^)q}>M%HH`xM>%|MSkj!C;S=oQ0%y6#n zESAYK%Qt`Tgt5Aew2)2_;^xiX&L^&3M`R;T=;*oI70c*Lsae5Ork$4*E&Pwf%J^Ph zYFRTyvM77OYtXNMU7Z@SEPb54-3o7;XutLaYO9(xZ$3o75#LfcZtp(hmTczE$ndT* zcGU2IPJm?sXOLta8f^=lL?@m8)s)2ffg zwvvz(@3-^Uggg8$rS^8U$tyxxXyLXKg=RG*zswuQp zqcNyFV!6)Li$WAB~lXOzw*Hbvg|CN~%%NXq3gbZtAX*@EG= zTQvofJ`u`F-O;?g_MHJ36lY@Y zN6wKfntd=!WM+8SxE!d_L4v13e5o7De1@P-@c5H|v3ZdHz8M==EdHooeeGhF>E!-m z-Zi((m8?T1#L`;z=)^D=Y6{gD3Z6IKJHbEeq0Ab{znAL%d%a^*r(?SeQ@KSmPZL6< zm9EmQ_HC`$@l@v2R+}++Hr9Aq42gxzewJv4K5_2lm01(zz7A{f<_>VU*I_cTda&z2 zy0tM#fpr%*V8_l1!wEmnl^y~atcF4ozQU4un&ZZgM@k78IPwDJxVl!L&`fckRUE$8 z07mslb3BK)*ZzggZz_Kr_qofNB>yY1Urp!#{~N1>Byv~&|E>66zZEB1K`mgL7Jij% zBDZ_p8a&a8Oo{4ns`Pu|MkxNv;a9bJ!8c!DV2y7HDZWzsn7+#*q`m9b{UVudoi)VR zZck_sh`&#!cVZr*6^j9NE8hMR*od@=wJ6p#iwS_#3ct{mckNgAu5mZT$xRUdMQ)^D z!{+i=qw1Su!=H$XCLeR)c1f~lPSkIgxSMp^yXRB0w@DA_Xl` z87t$Y5$P{iP=ASWzu#GTxz&q_1T^pr(io*LA(mtI$+3LGM9aEvXoapguuZ#|%1LQB zo9lHi0gYHUX(87-weDKCG)oU#2JegFlzxqbk#eOEzOUQXx#Q5=Cj4#l!j?Ma5_Dkj zLn~rYcAXAKVRGAbdr)A7EtvZyT}H~#yw8^d+5E^Xx!*DM&*+!e+ZI=ioxk>V{8R}?MU0J? zy=S70c7H+5vN+9-9c}v3OBwUfQZBVlIi@hV5%bGQD78Gb za=pvPpu`nxpJ;(?Qvpp?#m=&?kJ6sx29(}#wu~q&!Psem8B7lCUlRjc5|x|Of6i9c zfkqM}+`KpDIfv3*-H5_q8j|cN+Fm?9{BBH6&={RX+|qxS+R~i)VMEpWyKoI-UA}&) z$2z$o^$%}mr~3QaK(%xSetpT4CZXj&2g!kt2l>~J2zPpSN;?r3Lz&_CrnRP~Ro{aA ze7z}neZ^g^|C{@$-pwc2lGJV9?B`kxB}Apu*-*6s%Idrw6w{4o&g^_Sz3+v}nybnc zgO0e*iUM9BgSOgV$>&wnsr3{+{EcdtcuVT($t9gvufGd}s$3%t+LUHgJYy81!$KoI zd{o`3v}Ip<<=ZI^n%C-{fBi{xG(fkX2Eu!fU*!#E&ac1V73FHdF3on}g8wvQ!95J# z|I<5p-smoOHmFWFrkK96nNveh0WB@lycn(@Ei4gc)P{TcA^%~`OqV`Ds>7*~`dhsc zSNN@Q+n4AI2RB(6WD;=-_qxa^913#B)UQ~T3wSj0EMYXKk%)=AmxtROg%}@Y3p#8wzgO5y;QOepEpbhbW*7e5_$*F#ZguN08 z@sYbL4R_nLwLA^~nD8EuulUBA8nBa5L2t3DRa310W{b$apweX`+gs_0rD=#J)6Vc5lI?RG zI-|-4LMw^IGNMzN#x_6LRAj~Y1rIsY5dpFZ(9_LjPeCm-K8_|xWZiY0edX}s!<&x9 zlRZaoDwAPr^QfB9S{H)@tTq1A^ckaQrIAl;gR%1fXqZ8Z#~WQDYFL-t-zN?7zRRyS zy7Bq0pE?{H^IIN34IIqGGiBu7>k2RNzg9ogy7FmdYhY2OrcnAqu>Me*kw_!=Ti1$F z=7UBV%16l5H~OSCK`wnAUr$#$(*m=au&FS)MKXV(CRuWI0BGGpEv2?`LUknC-Gr-k) zjTu8|7oLG;3$ee!+V90nOU?TzIp#wVO;s{qj^Ts-rPo`43T+u{i9K=PTJ7MQ%lEy!b0T{)%q7zbB0r4Fp}VtW0DEr^J0zocE|er3GCoLh7i4Dl;$X zvxk@CdkNkumh-Y7<44>-I>Oxj%YI*4g|n7b(GgQL;KN02u{##rB+TwO?Lsn_m=duF zpiECICt>*|6`l))EFDhB68zC9$QU$ybGBMKJ6D%R1T7AmP6Ng~SY+X07(+JiQJMEO zl<-2KNrwx?w2TTvh9NRA6lOuS{X4KAO!Lrhq}jQC&@_BR4|{K$FuFXX?Hs#DqDs)E zB&+pgGJ{^)moh+;PK!W*XCc+lT?sZHQB^zp1xvf(0!y;6_KNP$E@cypwvLIvPsoo)G6Qe3RZVIcp=;>c=jv1QF9G0H! zczr1#QMfSF;xuH=h-r^PS0pN@R2_nS6Rif)n1d9t$D+lt{A zESN%eRg26S)O_GyG6_Q(Mp+@fj}eO$MoDyNt;emjm|}Q9s+i`4q?slIp_j?spg~oH zU|3v>A|v(oKVPk{i092}Q>Z15!7YSL@V%-EB=#9y?lVM!_HU~h?sKzaT zX{Ipl%-jVmQ$}|3*RL(WWx*FJNY#2J^T){y+h=mi#V#GC$V8I9zues)(|?C!!OQ(Q zhD)UC1-+DxI4p`~8a+dO6Z?fY38XC-fQP^=d;5#h(dTU&dEr{lWR^ZMLo$&46 z*`niY{uQeR;6CPwPg;3-p$NFZU@XY~6$-<&ZAs!dK@-n>I-YP130RYtK)?PXZsSAR zt`9I2liN})^FU?x2(WL;h`AvP`T`KiKo1d}QG1QFZmF)GGQ;(KYxVYnCLoF#Hqj5X z?iJjR?28%%hkm}kH-^7NnK%oTmPjDPAs=G^(`9x7UUwM2^kdoqU+&sM&oSi-&g7(k z%!+9bcW)|R8%T5glXD>YBA10n@bEkW`pC+y!w)Ym zhB*MWLFOdBOA~}ND&YHa`pgawCdiBDtBfB6oCk#p`0&rr;1{0tAo)N z6%H7_l4uM8=q-4aa7*NMQEfkOr4*DVV_>QjZYq%3n~TvvO@iqPV?j&vqFRqb2M)-P zdX?_a_NnF34Jm%}WF~B;ej`P0`ZVLJE}mOji?js43VDdqZnV>et%EZ9g|I^pjl zDl)^gRR%$c+rMp%&Tm$JsMzbw(Eq&RSW(%J@Q>6Y?Y?aht}JxS*-M}_L6_^d>`s4d&G=HcF8^));u0lA@{b)DTcjR?P}rd|)2ooWggUW2Al zBH-ZEBh$p9=hcx_K4LUZb5RK4nddl#?I5r)oVD0z(VT=^{gng_(gGis@`cg@CEfQ| zRwpyoCHHZ`pJzp_6XF`Y;HNkFxh=AiM$F)ljE>zK-_$hRMA#e$P)R|eS&k=;v+d{Y zIE4B9PMIh|L1p}%liMAdqO=XevexdJbA5^7M1$PNj~`D_3%<)<@gJj)8P}sS{iW&~ zAKLnnZvkJQ=^X~Bxv{!Y+w7Vz_Ia1wwkg6j-0bX|%y6?YqOAs5?P-|7q{4zX7?}z( zO^8n`42R6>L&`w3b=+j6{7BR0T+KIAltf-}k!NI+Snl%?$zqiGbj_`f3Hrh*5nuer z+#&;U&%QP9GZ=A@n8he145~(|46`)}shWg{3+!Jt13h_;)ewbJ8$%b0aB zH3%5;=;=b{JRfzmc^>kU$Dgso3M_8F-Ls{wt~f<&q|Hxz6IY(HpTcvbf!Hf`PdsdL zQkS(!l`&NEoxy`IKgx4MiJ79M*}CEwauG zy_&$jWghMB-0EkW+Z)Ij zDggFZs|>1+NctiHE^tjP4>^0lvmkBZEZ1)MR_ks`?ZJEPzDbPkprKi@_FnRyoNj9U zvl7a$=l%7tb!f0*`V9A(nsOFkcx{AhKt**bQmDNr;c=B(;9rTmH+C|@b{M4Q4wkS23ZR*cJ~M!#=dA#{iE9a)hT`l zSe@;Rc1)x@b=l?vkADgdEZ?_r#CYM1bWJQi=DI*ZN`A%}Cu9g|m2!P_+gEHKe^IV4 zyc`*QlSIe<)iq~6a%Tz7bgDki6}P`u&5h$Nt^(OW!{SLc@DuhiZjtT$Si%iIsn*wrefAKiZiwb>sRXt)T^G(`qTkd4@i}{}%Gep#PUfXw^{H5X2 zTSF@=sCGj1b^g0@hhZbjtCwre{vT@{Qg%zj*#i_^2ujqBZ`VYBbPn_b^PGUoG( zeKLoGicIE|q4O2OT)jo}h7B7!P-6LUHdS(v%j6-^(e^P8u(voOQ`E>P+CK~O^z=kw zbDBRD$qC0Ug?+*V8$IsE0~FtZ^pNbd>zWN;BjgWgg|?xXtzGJe?8fa=d?)wNwNfhb z6;(Pk6uUq*b5^YI#iiq*W$e>e)*KY53S{P_=*r1EWR?rX%Uo0dj2Uqw!FXqMU4sT= z6Y=Vx{ZI7Gk_3>qF&2lnHjOo?`T)pkDn$-@XCxkT3e=`3mVy}tBPr@(s2VlK<)#+3 zTbz?kbn=H`JPbCb=XDhbPZ;2 z$gGs&j&Evak}L1$Sby>2#mJ6K-abV!nb}KkZg+isQz4%bQt;v8=Rut$esP@Mn+Sm+ zc)O5l-DNS0;Q?E1+7j%%P)GJ4V`P&_)+BB#Bzg5(TM?+Bm`2X!Kj?5dH4YahCS+1K z;DQIH^oZtz6JqVefQJ@U-qUo%X-D1!db_uWyR9{)P(q`7=P|+>;L&d2_3j1I@qNtI zRAB@_Gg?{F=LT^);qPc*g~>`$P|H#A@Zm#)P50=DHuNjEvI%-U|5yKEg9j)7>IrnE zxPOVTX+8pX0_n9L@2tiFFF{twv%Izijb*ZgWK%Hf=*diEKv_-cP6G=mX-Ri4`x_6Q`*2UP>vD!)cx}Aje7f-^&${MZf@?T5FaqAbc??mdalqsd6kT0O0=(U zv01?;0-d&@SahgT#tG<^#0dj~x#9u6hp`NmA7R|_!%MQTU4pbibVmP5D@2jbopbjJ zH1-$lp8GLF`Iu23Bac^*42h*@T%-Z;BhP>G73G=9Qd=>E$5{%(fN8Ysr@0}}KUm}kVgmlm+Z zBXPbOxa_y;rpDlwQlQ@Wqm?{l5pID|i>U~oC+X~X=O(-aOr{9ODY5Pr$`wtgv<(AU z854}RHONNpWXaxu(?2Bg0*VShk^(_=-o#YdQY@rd{kX-ZYXtg)RR2Ubhi4b7D~l#J z9`<4Z7c|!fEGgkVWnk8VM2{YrlZn|z$jZAKV&H~o0u9Jq20vb%+E5W^GtKx>+uR+$ zYW{waz81^)OnI#|yy@XhdD`Lkb2v(fE<$@XVudn&&#OHG?azQrqbSK*w*R}_-e|HC znuuc><;I$IzgT<+UwwxDmUGiremA!VIUrdiJ70IBjSF&Bv2`_k#C5TDyhk2C&Bl-( z1&(f5*^ssQ|Mc!r{oAUG(V?3@RQI;+K)8{iRO_QtItK?@H0QvV5DJ!ab?UqC*wSSq zcaQoeE?zJ^a17BxU7gB$wRprfSl+i2W;~_^!%^b13c4GTOZH&tDGdI)`b}KJ(JU|x zJB3<*HL2|Jt_`GY@$;fUHYd;Bt-A z8nH1tjd4@F7$>|Z&KIu_sIHkNeeIU@l+Mu``iI5;q-gzX zR+!)D{{}LqBD3ndYQ>7jV8RDmly}?8=UUctI@5p9)a=vgc(L8(7>c=(9j!W*hFYcF zyWqfU-hPG0pBEI+HaOgH_VnqyDM73pjv}ROkj+GPR!2m@L@s*CSHN86E$1WIOD#GK z>|abak?<~bRvZ^#u;Z%SNripa%%-M>9EJfY1ItS~ASBEC&p^J@<-h;7bZ9QX^}n3b zDGFW9q>zr+oS8#uegvJuohMaawE+KtB$!2JQaNCr-AiQ^aKKb z!h5vcmJaUKlg(88N87J|#v~%k^uN#e7I$+BG`8}P+{?U?9km!Mc!_B@5N)mcm#*N< zQcADL652cyT%^T{$85N}q$Ag7Sx*^4v)E|7uKM@`Bks3Eeq|}}FdTG{VLw(`hvcz- zr38no+#@Xde*J**ei}sp#G5K0=Ke+S7m{LJ`eLqT$8)%3-KjY^>|-X|$W0x7#AR3G zDn(dtd4m&O>Yqj;l~ylu0ihOML#klZy5*{43OIJ`*wc?$!BO=y1co0WC1QJbHx$hO z=JE?`V@a{mQcPX`k@o)cs@_(foZ8|ztD)YidavTr+xu2IsVtKdq4l5k&FhAlMLLUA zX4C(bZ_~WiXI{=u#$n!Hpp(lcEvp+gtk9k5-6|^d+f?mdg9bgySJ5%`Y?$s+gWu9z zeS^kSm8q^q|KInYRGO-j>#=*c<3{0js{ZBX;C6z3+2(>;v(>ZrDpIJ6d=w<$AAkI@ zCvk=HV_r8WX5V#TM&3k!j2)a>)^lfSl6o?U=uug#C}aPWWzK{kvXSm1c9-)8=bP2P ztkvR9*;uB)tq-@pY?m3=$cpR!mu9u}vbbBbs;Zg~Ec&gUW4+=o--45>{TI3=EoR9!m*WtJ|E}q_e|J)CN<+EGxQ+Es z>-T>d#@zae&bi|l+c@xGY;BFR;cm6y>ZW2X{&i!tg+1FF92~4IK6T2P{;v-B5^^@| z=P5*erOhQVg&KuYi)6wZ)EpPuvgLeW&QVG11+97sk4XS0Pr`dxFnq^!6EC^&Xa91& zyC?)7F384Z=~5QD(hd9h!ze};sqUnVPIb?x{AibiYV%D`wun=!`s=?W2SDq<^vf+> zS@cC;bpd@lcOJ|6vtdEoHf<*E;&wk+TKB9UG?f#5=W(FP-kmySE8)y{JhNpxV}g5) z6p*rJtl?}xRm7YsYnZ7QOqnX_hu(h!?+w zkg{~x&WSg3G&P!4thI0QCf~NXPM=!blEyv~Rk+~1%8%fIR(B2$v}JD6Q7g`C=9TXS zRHA3JxMHrAQ)og;d|m4Itjr z;^QLC`_kEtePZ57k39|T%`J-3!aE7SD6VV*D1%5?Y%fv^*GP*fX5RP+nT^)UzwuH} znGq5`OWGtc{ar2g%^);jYfp`*;w#RVqmR0LWXB3T*%PgY7E67EXKq)-Ho(UM^>M=v z5k_T(AR0(a=cMVDa6`23PA0&sq#UA>CzdLCA?DNcJiiS%_w9*RdSzBeqbZ7`X~!3A z-+{_Z@5Kd0_*VO4S9iC3uBOsj#Kj9?TJz#(eEdpDIwplD>k?^2fs5Yoh{RXj%m{$89x11gHo;RwI1GdHsX)0YFIeA zTw$~|%%ngOu2NZO;n8Uc$}Bx1)WH78KEXf*Ii@G)0id)4LqHFu+B0_@9rcvRY}=E) ziQ@a`Rt)=w0zf7O$jFKywRsInCuFbp7)9m446HLK){}B_az6jwpmcKf`eO}qO=SX8 z+<0kF!G1xgrH?#^UQ}A_Sk(DxU`=IEQ^mixHGM#H`18NVK7)@|2E}$MeRrX9U7pIU zO;pzo(`kzUXf4X~G&l%i2s7~`xbM;5)SQ~zMI`X`xn&ClazqomWOfh5{TKbxMi@$M zDek8RMSC5&!ni5bq41?j#(|C@c{|kUh#uMax9Yvq}QJ# z(7~`HLzeeyLjs-6aIt=7&EE?!wmL8-1I;-ud@BqGtz2%)W59u49a!aV`mYVcuWd}$ zdxLHKRUwDC_6XmJ3ZF#ppmj>~Pu}%oW?;1ific}w*qy~RGGeWw+yYyUElq^A!`7n= zqvw%PT`sPw8u^qPw#0an?8lVh*HdeD#6o*zW>}wWOmwyfTby^mfqp#1d_!;-;1P0pfIL^6IgOQwVK2^ zJH_jXtAQXz2xUx5b}m+U>DqA{Ck6{@w{@r@ZD}J? zFRj}{yRG4ElJBk7+&k;g-j#TeB?R0f(L>nh|Bm@L7uFf2l)e5Y_i0q2#{A4#j^)J) zCj-04dnqq~iIl#jrKKY~H9Ju*xo7EE#4UTY$u$iN?jPT2YB905os?fg^xB4afbZ}ZjD!ol|z%dFM+v7Y*sla(AamI1pwJvY3&tY+Ws z-Hzfc&j_BI%+R~wdfH9xv6f_yT0N0+#(v4~-(9+0XLCkTN_mtJeYgrS^<-@?S(F5t z*%_FKv&;(QY@%BjDGdkVc)~L$tNocj^}x}jkwoa`kJ|@(-xUg)?qVT(nJk`ab)%`` zzI;os0jsqYJY1?Y;rk@_4?jfrc^U`(5OlzXk2u^QO9~$qpb66Hlk-gjK%sw8v&DVe zKBWaykfVZ;2h%iYc<-AMm-tFOfivqK}#`=bil*ks2`9xs+HBpEYX~(SBjcP>!KWn%<=GNZ=!s0MZ1Z2NNe6hIzYug8C_b8OVt(U3R|JW!IdpB zH4=@8H02Rqbb~u8MDD8kC(iuctig3Z{-5k47^_aJcmCVA{mBW$()x|=t$w7tE55XN zU7JY*AmCupuS~bEjg5Bpw27>&9bQxPI(ie$Ya8ijju_)|{6mfVo({-_CXEPS%&?-7(zbQF%=uq0$%i)ep*4o`dt!bYx zcdOJAn75Nk`X3BI=67P&cFtk0ZBiBcTS;1m0>N?ZNp?4 zV#ctI+jpc4G@MONAlU%B1H#CkuZ3&a17KVSPCT8Qw;w*sR5ld(fWU@;7lKqv6P4IU z!J1)B(QL$yc@Y&(qk-5_telkr={!RrH0apt-KZdg$QrZ zQbD7Kt{l_u!CxKJ6c2P&o<{~N2#hy+*8g0fBH%cD2j}-FfTkkhTcPVfzlzS1Mzcys*G}qW1r|Pn_@d>RN zKTG>_qmFqbED18x=tUGZ2HSQooA+kJo2>K~!RQ7Rv=tdzs3g0Od5e|y_=W`Q)L>#(uSh6s~Qxm%lf zh4>HJL*B@MrZT^mPcjaRx4vZbT@2#!rej7iPpat?A54gd>pJG@34U2j))gPDyBY1B zNAyX}`=QU9L*$HwXlR;e)Bv5Y##qPb%^o|O-9Yd&epE%cWe`;!~?AW-UH|tBkBelx`748%e@cOU&knfLM&lugsP=sFy6M1ai zkKgYRo6{d7Es9YoV3lGF$l%>*mo_K1(lja)S@0O!mz1x}d7FmvOd7pKSf}(wqz2jB z|5ZDDxx}hn>ehEoFDk0?`Q><3CczT14gLCO{jPEAufyeUbAr+J-1xHb-SeB`nzU{M z6&}M_)2-d8{(^dH`;6K!9kA-EHqT)pVv+rsB|{5%OOdprf0OEL{*9&c-XtX@9XN7i zbi$hiu2P+XWvA1!=>ihSR%2(3zi#AI`LTdkCCZ6oa#?KW9L2`UM8s47n#} zTWD9sp!nxJ3)M~D&@s|Y)xWsP#%02Ix=v={xOUNO`t<3igE!Q|;;2;-T8_9NMG3Ny z7LG27G9q}^;>DUn9Ak6r-06GXv*II}U1CJzb7t^NDra^sup3$3b^?pWvRTTCP4{P}~lRbMVa+jjTx>M{On4H%p@csKOK&`~Yo zFRmOD5M{<%yra)ORjc~6S^tCUw~qI(44rf+?~lfcv5U=B6R4M-lmcu+vTRs>m99uE@j8oZq0Qr$#2Ff}4IiPp0o>{|q>cC8m98!|$OcMk#)=b}AIZO3 zChUSp(+ta6-mvV2(+4YJvJxvT@1eUA^WMD@dm3%N(2-F)8-Avr`p(mgPG~Yuhf+U@ z&+sEuni(si)E6G8qBN84ZKbd1x!G$r#vPOn@`&CXii{iyd@@CH=paNOO*#nKMYZ;= z#6W%3N+#QnRT##9X#NO*sV9Xt?I|yj z04^eWV#CBQ2-tx+iy2vwRd{8}`BTT|nwjkP1#tQO_53+=WQH$`P$q5>^7RS|?1$eO z?72ku5Hrx8=nTX{Ej8-iKmU9;Vj>+SjLFizQ1e7_`da&N?49wG+^-E4XhF-bG?&T} zCuiB&AsDiC|_w;|By)rl0|A;U2Y-6DroS6JCJKPT`VnmpCT9tr+H&cINWZ%-JNTcZ;F^_0nSz zk4Eh#j%D;`#Ze)#TTavH5-0)^=zZ06AS4G_udU|2(cHa`kgZej!{o(*lcq}f9;J@~ z$-Zjub)mNF);7Ww@r0@gb^BhV4~Ns}$}f(aoMCKaTeuJQ()+|4G~7)%{XFtIbDY4cpKQEMvDsVU7>#KGaccJhfH+l z!=Lk_Z__Iv*I5<)Y651CXn-?dT^e?MGNctlUV0fa_Q3G;vAt>gcC1R+k@0RLkWDlK zARDb2tYk5GHxv=AD_rsk7c=J)y$?s9pb8J2mB|)x?prCtEK@7|Z4*VyY@os2r zdl{-U`|H+ukJ&M=c-a9Nn1J+Yw#(KEJX_VO+V41t`C{)E3j!1A!0A|*DSp3_eCKX4 z8Xsn4POSuV)Eu+KGi`|%9~on+LG2fgW0*-9M|Nih*NhT3%fdY)W09y{1Z#vpz&6v|r&Dp+ zCYkezH$=i~Od?K9d>hugLi|uRj$xMw#t0Pg{`2OSn=utXLH_`v*M%P6c^L=oJfM*< zcxjwM2K^F6gMBOYP-KW`8$n6CwMSE5Hf8F^#I2H2G#!l6mvV2gn=5j($LD5r5_L50 za4Rnt?$ETINxmsF#rdd0c5BCWYd#Pm$JBoLp!6l-pm3LQXhXe?fH!69i5 zc>TzqV8Mb6d9|o$9-f|Lg(kyjcbaOqXV-=1Ww)^p<`o`Vx(?Ba_wE^UPCh87!jmin zP`sp#s^frGj^AIKhh>Ir{IK%!lP6f6{6+CZcHDNYJL=6MjvqD8a~FG8fWRiBqN1Em ze?D;0RV`!A{A9$=+d^A*7(SDPq*+9b@e8^66Y6gM9C_lviy_pC6#Fy802Bu}5>kXX z21dA0aCviRyOQaTyIX{%Q#QamhWwh;<_unUy1Ms9S2^GeCBq4k#boH6{o{|OG|&|D zp_}M)kdadB?jjjLb=Z5}{Q1Rrh<#gG!zcPaeVeG&>81BG>;WOM>pp(qlffivcvvEyFJ)o!}`O$RIKS3r>HE7t+Wlx95uXa=grg5f71&f19E^_ zG9NypbQ}L#Rj}EoX!w*ZnmO|UznONFjq}0o22=H*Y)_AGpfD+@d$PRo*u$s(?XF+; zm#Tc**4ja8zc}#)kk9Ts?UE|5mY=Wxs0z&TW79cxxhp?*o;FMMsrlAfOIG-D$*ql zqEdn=CH;-L7VhtQ&v~xv`}6(&c)X5Bx32YzJLVj7j4^L-C>_{4bK25r3i!z&FkmRT;TdQx6*>`N?&G7FB={VdR``Gx}ZS$nl`l}aZZJaTV{b&q^_MOu2e}2!1-dR6s?9aJ&|KIxUSCd|b zQ)EtlzPR00^`F0fvECfNcWiM)RFsI_a>k0gUS3|yXI=kpl~NJNXg<1r(XG_fR9>d2 zsOUvsM!~lB@~J5`G2Y%wpJ$INc$t`#^gb(VC+oMA@?L#?{YdV;^^=m<^6>B+iB=yk z?8IMTELPEO^=hWmCz4 z4>z|!`@!DQs=5SIJ(s1T@+aQjE)3Xy<>b@DmF5@69_w1>^Z>t8Ii3bJ$(Fg?jV&$f z9334Igw!-NBBP=dn)3ZkE#5n?-g~F{-M!TVzZ+c_&Yk;cnMvaIL!k%McJ10_mo^q{ zbS7Wo+9neo9-b$s6|Mf}?OVgw|IA$~ZT;l<@#7_Fb{F~w2E-)hCxnFTnKWtAhrGP9 z!oa=pXY#p&gMyk%!Vl%0IB#y=^D~>L&F!FUVRK=iWnaw&K{2u3n`x$;SN5GrH;6ye zpt*k0+3G~)apT5u^+@yHVsuhOY`gVkIx{C=V$NvYp3S~W0pHyq zjE?kW^mJy;#vFc1G9Bz~x9xm6rKe2GG(0u6`t>?aa?_r2i7RtqQ)6SGr1`hzTf*rD zJ7>%A%E`$w?aG%c*SS!` z7unM!%(Stwv26_x?f#V`92pl^@@4EEQbOZ}8O4((PSmeWwb4vEA3S+B_xI}bV_B1P zeYa}Z^>v7zlVV@n>CP=IgsjCP`ghf%4nL|J?znLN!f5}}ySQ-P?=~|D1;06*hd8ma zb9}QB)@gU7-mpB+#m-BUXDt4kd0t|~$3=bN; zxw$a8$+NC>6w6L$o zh*^&T_Yyhrc1GGjYh>Q7*cDr?6kKI`?;r4$I`!)$=R+E^E^#^KEZuVYcCBsO25bi= zix&T<6Rs^-`1}VsgJO0HgOV7Bji1()xAo@j+wQyz7cDZxM9LpLC~mWv^P0WD5B0wu z2L|?Zr2G)z=ikFp)z*$$<7d5tX7|EI&Zn&Q)K4l{clG*pqyArCmr9rg@M0xuPhYfN zNmJANy&Gp}<9Gl-4y<0gcAsflkLTylpLv-c9v-x;;*83L+}2EBUrV0{W8r(#-`~I1 zx^d>4?CkyQ2{3Eek@tmz)O_b*ONBOKVVCPU>SQ_3bn7<%Vn5IlO56S4%V^PMSh=T* zi-Pz2p82J&uHM=b>R*PXqUgt&oPc&(6`EKHv1z^DG0Ck0OqJccuhA0g?zZ^%Ld)nv z-mb2$oY0QNoz3YuUWIoo}J2Buy7$C^Fw}q z`TT>@kIb9$9!M`?e`;GBJ^i|4$Bx-{<%%!l<`!apc>lg|sJ{yk&iLOO-l0pi6b4?W z(VR4SvOEisJmlJRt_XyRKQ|Pj(C1YWdXOirx6vy&m?Ov$0Q4FSh;e$<8YhasrZqhuNyaQQjX0SNk+E% zv#LDj;@NHH^D7jLj{F#v`Vp&_p)05!wej34&U(5(??(Px<*)Z%UrV5jJgO*Wswf3K7^xF{& zq+sI1+>qe?UIJ25nk-c%r2tHPlaGR;{hAWd+ID z^)}}_jyyLsG*lSd0T!pQwiEZAJ9o~u;TqSMjt(IvqEZoyZ$Ad(-w&zLhbB^Tf4ut^ zk~=SRnYc;Gudg|$7CVnkyHm_Rcq*^vkcTEkSb@HMa)aNu*#B<0&VPUtsFMW{x8h#4 zC)CAYs`~N<1vO$yxb!X_8=H^-!Bv$+F^#WK_M^^#j(UwmBkv`~C zRTirkd&b{hec{4|-)3?WrS|2NiV2Ku8fTs!?jEcg)qVN$rR`S_4FEylJ`vMe=?$83 z`z~F&blUYVi+#?=!h(W!oyqr8Q&zrr^X5%eb@h;6zs{rP#{VRj&syTV68E#QwtknB zvmg8N*Uz8#Vq;@3U%tFx`SSIqrl!y9R?vj$FT_fQ$*mgHR zYHrrwy?3u5GC2^rky|-?b@Xn-1kO77>&yFTGiR2baFq!G$N~syE!@jfdU?`xk>jub zdY(O@Qpdp|eHOQL81mz()2CZI8?poh1PuGiH=w#e%)7%aqiLE_YN)2F>OOha^04h! zCL46tzicg!^#`D4caYwPN|k5NHq$H&o?IjI09uSG--H3AWx|U=B4ClaAeTvZ!9>&Ye4L`t(CeN_WSPAOEGXaWzKIvaL)J zvz|9cOGjq|Xpx&YrvYPxV@gnIXl&np>DRAc?hhYE0)0%5Z7`Ft`22dRuf(|>55-MA zd2Q_MjG7A`Efq7qKX2YVN{1u5w0kzvl9LL+$femP4~; zy6;L$OMm2^tM7r=KvvXm%JY?1R6M(>+1A!J!f|w@wJsxr$}G%Uq(-bRO@_3LOqIn8 z7Hgibw2jJKjTO1IOl1p-)^&orYq&)5mfm)DN)m;qSGhiLcZ%gOm`uJU3DZqQ9V zY*Zd&X|fDZ+|Rak-L-4i9!M-nz8nznH~!kg-Cet^>&v%qhA8?f)9q7=ii%VZ9V%`< zhg>;f@(iVm7gI5Ih)|LD?=LR)+PPLhpy<+sDej)0Av!5m>o#mq#KLpoZfj_GGcaOo{5p}*&s-S=$XpvDOAUWnG~X5y?fmxTi++yhh5=w^i|?PF?s27%oOrI$0@2;@ zV*M5TA6Ga`QgKdrOOG|-k86pl~O5+>BG;tga{7uVbkQU z>=Xl@yMHzFmf$*J;e%b@a*MH`e5>}{Ui>JV$FaKqidccV5P#<);CQj+1hSCuF6AdYFNdrxO0trP>}=|ep-g&Qp+T1w|K2};C`Q{Gen&K zhuh|wBp3nx4SaDs=u_#ZW`42efg@k7sB~-LGEtLY^8fk9 z)1MooCZ|LVDCAppI@ddR@JEIo_B!4Au}Qt?jGarCEUA2Q?8*z{ilXKtX&^U}`@cQ3 z1Ufd#bQ-6#xm7GUAYdE2c7DU9w^mRv^5qE7TCR`ep~h@iuauOOsq@Es{r6i>-$FYS z#Y|Yj$(<JkOT}Ven8Daop5#-@boB4*JI!6oj+*Q1`tiokTMdUBc}5FTex$bzEi}VMCdBJ!*KzOq_3Q8FzbXYmL(v8cVPItNhuWb-hfFg{ z4NK$8btAP*)6@WPR-Cz=pmOZkqx9dOr&En2Wj|n%W{+V@{_3e^G1#6UxOsC?(9}yI z|9$^L<)GkTf-_)KtW4T!QW8c6Ge)(ZK7ES1TH_v~C;(bB3K^4bBQ1Yfd#XwmABrii zT)C28;ge)(NeO!oV1;0zshIlX!LCl7Z-Ivj*UNC=SNuz47Yd4bs29G?v0eszqOhZW z_g)oU-55aR*5ZS5mL|^+uTKWAVOeM2_rAQm{B=vgTm&h;HEWE92YTe0%))DRi#{2P z)mRiOdP|yzATk}+Nbp@DVYV$&J-P@fW~sLOSuD@wwwRO!JUsbL<3a%Z^|6b$ng5mQ zg5UG6?0mc$`Z0SxXenszH%P$}BlMatfACL#&RYeZHGW_%BPj68p zPyT~`>mTNxesxu+3fZZr;W|$%ruPuYj5Pb5&NB)@xQU+svTsBvO9cCN1F)(7tV;>! zChBJLPSmwf#TKw_^1aZFIV;!NY5DH%O7E=~V!(48{cN#uCull~Uvyu)-tPua$`$6xGD}_`MI1b7UtMb@)D_@yI{9BFPkF=%YCd;`JvKC?}|7%z;{O&v^ zGdk3Tti&ExZ)uxn#6mrJvT=UsPP;8Lc7iV%z&7jwMvsV#i+i~V0kq_7LQM!}*l&7> zr4pWc;PWkEdTLaw3AO2nJx0CXE4(Gnt$M_zG2_2OWlZMA8TLJ;8n*b^HY`p_;mGsT zEqPmixMyT!+?fA38rN(AR4v3*S$uhSZ}Q;IojXObL}^nZXo(!#5$6o(O~Ha{m)d#j zSty8u?K+b7idUOm0jf~&o})AAf4FrJ*u@`deSWgP>nC8#NhVZbouG6`ZhukbGwOD& zEj%Ws43RwEpqL*>H9R>vxz+(Y>}O-PQCZZHWI)9mPC`+r4wi`-(+-Rb-0NCaRSv4L z#ynq!fM<10jiUD)-xdGE-PekXYY@v(7Ol1BdR9YqM_XGYUflo*+^4W@m7&+sPpLM~ z(}#PteB~U|Uv4TaETjc&lGanE?wI9t=Y4}&GmXpEAE|*mug@y2;sPy-Db_%AUj2yc zES2^D5g;|JzPy_|Jp9A4T)gh$A=4D7MkYyL7tuT4{O6yI22vtx)~wl!o#acmT_NX? z`qBaAC=jC&o+=r1(y6D174FZ|xS7>Z0PGO^HOEU(MkcK`{hxpS%Jq}ca$9rI6LE(* z|M7EL(EqWCw(L82u)NANBO1(TNtH>wrl?7^1WhBV%UwI%Ktl_H2f*!huKcrR_~Osh zd#m@h%8fM2S!B%=M|o6rb=@)ktc~vX?}sJSIcQCp%^j5>4B8bHVS-6=3;zybp=JL) z(7x|canOovBSIy?;tDS)JN?Qy1DYSmS&H6I+CYEacKF>eg^!=#urA%+g}WI0g#r~| z*4Nuc`vxStO=^xPX`1B@fnN|KF_w$@PF9Adn6?#iL34bLlPbp6;&%c zd-iM{yY_hR&8KcU&*l*V5-4)_bvw3L%3&yx5C)IqkTTMf3pdRpCiX}|_4!-a@R=2* z`tHXTSH3v=u-0MlY&(K%ZF+tE(Hfg3UvIIq-cb6mz++HD{>OCR%8QDNQ*A4YE^aN2 z1jH>p>S$*6g#Um~*wp#_`T*UpYt^4swIf*)bsuLxX!Z5W7ZF=+ef=<#)Sr7OVkWmn zMEO_|~QP)6)|Rxwx`p9mJux60TmiZXE^JIN*0|LQz}elP7~r5*sEfdW$eM^z22I zxxC(NoYDU2=}rv42B?WNdj|&x+6t%heAa!<-+Ez(|_rUgK5^y@b&V*SsrZzO}G6V)&H+f#2_%06Tf zxsji)ll9ut($Xy3q7zzQRGAz?f@T&!3j80kh2U2EfjBYK+9)g`gM2?3MfRNj{%x&e zS*o^F$|4L^R$+MHsUt@o%vvE8gqpb(Ae1Npl!M;VwWpsOAktBMu>=~l>i_$Wpb;)( z3q8W){{6$IDcEDkFHxgI`J;&Q?(`Ta>;GfFjhEYsWt~1cVuxidASs#jo^J{-u!P=~ z$uo6aCGcxD?ixK&>+NRKQpTU$iedar%fv)&+ocEn@9ch>5ym7pFgzsK~}!I za;FGkWK?g+Z|e=l6`wkJve|!hC=^I8D7lUsiipcH@dFoXZ9+lt8-KpqybqXIa|iOw z^6{pPf2BGx|4QAqlaL*gsjQ;Hvg)V?adCSQPh)#~G-%E+)I4~56<2PVlEA&QcklA- zPkYN?aJ&G!Bu2AMCueOWVwnopFJDNok+S&}geSBd`hCvB)APme<=oudH(gvxQFD6S zy&FK7e?WFzt|T^Ct>dVDv{u4_($dnN-ZyXFw9xVe(|u%i@~!`#!0{WlI*z1T{iuyD zDN&`Rfly=IJELxT|L@;x*RNLqe3>?Dmip170aRB0_#uo{2*C=)VGc@m6wPixiYWKM zgq@7n{AwlvA^=oVL}VmZcrf@u5F}W2s=>j*B)XkD_mn8!=V$U;NrqdyR>{F(gfCZs ziS(vwY#c?^6c-oiZ#K5Jx~iyxtJ`G@STKL2rVV|9LogDW7 z6=?L4=X(Gp?m@gdWa{nNp8n{$&hA&^g@(BwI zlN-WV4AnExr3M7QX)|W1B0}!iv17r!d8;m57}1!zPcD{E=4_>iqOX)1f)0sL^XAV7 z^$BL8x4kMJk#+Z;J*^PmO5N8T@)8h8eo`o6k)aW&}jiLROE}jgB`e7n8VB zTvMY1<*yVgT-5l}#=O?{lM+`@W77YbOI%4f=kpSRr;{Lf6?s?Lfv|6RzLA6kN5N!T zbk=WYs#sda;{gyV!jPA#wnAQ3N2F8qTZ&sZYh!1EWLb}eS{!xcxrq5U1zHloG=-2$ zqQONmRZL7`)T1?n)gsjm2zvqw@+;iWoHP|AqLUb?AiTjtP^l0y|8btNw#Y*XN!-0Ynf(y@<{&f)Z&_5GO=y9>tC?Wp zn?CewzpteEo}&Ey62?_86u~pN!E8d4PtXe_JM;(sgCu+=&H@LvE-09JnV4WIo&lK!kcCl_DA4#~`!Ad~@A$87Twq%^_(=vMbKx_g3*n%RSB>9`Sk%)EYQFv#z)Dh(p zkOky}_q*Bk>sL__(@n&nh! z<1$fR9W^OLZT%+06yqDB6P`;Ib7lTk7Z2d?Fu<&H98IGMc5`c#j!8`Xl;nPe7mIqG ztgJ4g^UYhg4j~@=m5#4BpI`5MZ1L&Yh3Y5?B|)AYv>hGp1JWw*$gkjzEw9*c_{p}x zz8~cuJOsbCwN-t83IYYSiaIDWP>3a`w@8Bt21>B~`G%|N%k@m~vtsvKKTMxB%d^zb zIXf(UC#R){=D?1QcEmC01fhqg>eQ%3TU{3&JP}akQenm)~_Zb`K&NQbg3t zD^rZ>GcWz>?M=Tt|H2B;KG<6BylU#|5qPC9Hy9Qo>XB;TtFTwqQSekWf)Y$kOZy5M zkBP8(PN=-J^qRBti`H|#tuty-2&0_Te%fU0THf{@FhOYGgr|=Wpx1g3$gzrNtVsZW zx7t;|JVs}!>_zK`vf|)Sw}Q07f~EB%dM0l<0<)l~sM`DY?<FPZD#1S)Mjbyd2r1oks881fL)6ITM<@~VKrYE zKOp$#;4z=U;A%k5JJj$XnW^Mdm*9UTxe*;OW(`uBF!XqnEe(?p>P*23t?0%d^4 z$B!RVO~ox^8wP$ALa-{LE0)6Vy~pqX7W;e!I>qAoJK!D&f}g}5sR>I)^Z98Ps#39! zj1WVTAt8%eG#AY3u?;=wUxs?k1Jg4!bg}1q#i99Y4wP(2s4hBv#tU2s@}54bcT%^J zHl1g!hyc(QSh@1id!FO1>5izy=-QyBh&l(HtFiB^X{|}Qke0n%&Ow{#Vsjg3y`j%s zJ-xyp-ZgYAJ-S9v8m(bc4Ws4rNR+1$Ut<>CTqp?L{_{6ECG0UiYa1I$;be8lZAw#S zEmzvJ#~BO2|~BGyBMkU*t_XAXTS3oe;_AHV~s`$3AU0T5DJSiU?8@EjQV zonicow!yz>ZG;_~R2x$EI$EdC64Ar9|EpKq3O5$(vAg>Ma&maf+9$Wfo>}Zn;lTda z`!zi;c!0w8-dSEah38@b-+oW+-e;4Lplo|9&akNxyYuItvE1B55qgHd3On9QvCA6W zWmuO1kYm_V6x`EWs#{?TH;%WIrD0lMWJh+xD_5o8; zftG1Je^rKfeO4~~?i=|$op*isCa z#_sHr;B>c(kR}QK&^kbBfDwdfTR+DU8!B0dSWGafQLs+~9B%;|tPCcqyE)J`0$T`V zI3y4i08(As7QW$z*^UJm@xKoq%vaCv1q-p3k8d}rIo}?f@i|lNCuJ$0FFP1T0*8jK z&2EU!)^7PvNsHo<(*?Znzq`fb2uPin28Dintv z6hv@XbI?!K%IXCe{9qrSspRf%-65g5bJ-xl^7y=*DwA56> zH3ttKZ1R_pmj2Sx(sHp4q4y$)mz5VTTu4eu$;r))uJvD*a%BCYBsfJ4K2&1(+Z9{W2wW%&4IqjrDcYO975%LdN| z~ZYNVJ&z&{cAd{a6GI_=j7 z!DNUq1b_5~TQ#sV&vx4k^tMN``S08ez&l`>0N7#EhJNK}reCzVVtLcZ@$Azip2gVk zd@YajRMLI&4J<6;x_Z2)r=k`rJ9JQ{^vekL;aXnapf^XhdMEqeN*P1-)T-elK8tYK1ALhI9*kxUZ z{!^%l!;+H+SdL$xSfyj#EOJf5OCzoa`O9_Wg0mM5F!wqxZVaZ=cQYP~K{!aY2%2LV1je zSJ>10>zIIxEQeV(r_*%-o_JfpM3zx$e%gN!#6q-R8$cZ^2)vP)&Bsy=z39?NjkbbE zI~6C_-oJl;PeMvkQaGfx55Qi~UZE=q8>Cm>*Bm@&nD_PK0V7v~b+tfSC5Uh#sj1S0 zLC@s*ob!7LlJ&^K1q&n^O%Uqcj0VOtoMJ+sJ$sg0Rea6x#M$=q`?=oMPM=^}lU-8t zwr?ZYX@vVc1>@K)i-dc6^8923!y6{OUs_x=@(p>Esp4i-Ush9I9Xr+5#zr(r7}N-; z)#!m&0IGZ~O-j|QoPM7&S;zgMnX|d256g}GKH)Y9y$gPjg3oQ1B2X0AB3@R1ou7rx z=-F|Mf+cVlO60$YmT43Ng+~b%$V4lCVpcFd>Ec%bTV*&XEEvB7UrMD~me)qhxMk<& z=JuD>)YK&9nbZu%SGA5XtQ%}oZ}A z`i$z9IxGJ;8=YL$Cfo69(9$(SY5(BEt?8S%x|VO#K+U^w@!~L4jZOL1XB&XPDQsnTf-Bqb`zp=&Dx>l{(J6Ogx<`dxtMw(mEvt&QOEGh) zscaNVEj5WvQ8#TLPBt|%<gb*RqrAppQ`--*1b)*qziocM$~V$NC33b_CfTy> zoAl2Z6xN^_&JGO4yxTYfp3fJpw4JOh;{9L4)YR&26l0K39`2F=7@^30XzJN;bL4kXCgpO>v-MqJV&^Jg(WMtE;sWA}oZ7AIo^u%+bv!m!)TxdRpQ9sw zv0Nto>GT3otCv#Dnd9xuFJ(M+l(-v&2t?)@!*Ua|2Dx4JMRTpU-X(hn3oxNdw6jbo z`mVQk1fEySRsohfUfKwXviYc}mg&^J2nl(c^nJVenW^HZpo9?+1TQFT_`ta=!sSyu z&jmOnC5eg0W&&@Zfz7vh^Je1X)qRx_6=_e8ihb+*DAd0dF0WQfbIECAK8{CCJc5HzOPBk@FcjkoIBGw8S z4#RW7+hl-#e&!$sc0X}S`}X~%aSmT{+P|{XY9l1aVqgVwJ{!TmrO4fv zkTKl;Ot(oqqu&L?6>nN63$Jd%%Mqj)GJodEW>+Vf_dq%}gseETx5jaFnBe!7b`o4< z2bxz$AAY(6W<4|AxD&Wy-dAqJ2|uWm28`GJHE8t|#Va2ZG^AkynM=+R;vVxc^gz@j z*JLiY>C?khbAWG;gcjIZl@LogJRl%|omm(K>3^lJxc%mMAq*Fh4^5KEai$7T44BG~ z983llI1Ir!ABxpY6Kb<0yRcWr&~DhnH8UL2U|ImqBw7v6M83TZx}O@ImzflRJnR7- z4ZidUJP1oS9lv@y->>FN#T(F6BvR2|8iGs7n|{7TN1+O^7uEoEY$(7;&D6{zMhGOW z^)Ih17138piRnUtCUg*d3`DR%V)#)+9)jWHIakA9)20;>tx8#zqA000FeBGWB<&+gsC9zcj8_Q2ZO`qt<^mz<`Kju<2Xrrc zgl+IIM4+}ow2FN8tmykWhQ&tEjd}Asa3N&WW50H-*l84-{bosO^If7c@a43@j+C?9 z)}WV3sKQ2@xny&YqW>}mBkll5CK8(Lp}#{NM+ej1z-;gs2ERKSP@yM&ZfpcWbN<>I z;#H&;jTe^Uz4hiXP?X`%&u^s8eSUpPn5m3_;YrrzP`Tk7ThLOlu@H%m6gzSlpjbWI zy%-P}CgC;&{=$FeuBOs_WOODjS+c#}@*E}^V5)aFj~r+)Dwa*MZ&c>N+d+#jV|4h( zK$rh$_38XU2%nJ${bl9%?bCnW1j`~<6Ksr?iRS`|u0jEyQg{NfUgY%0#V$*>?19(M zG=c9CO=q4{=B-&^beeO;>x{da?1@Pkk{rj_R!p{W@IgdO%MEp2LE*4;`TV6z*#aZk zVnD9J{xa1Kn1m5?J+tkzvDm}17#+Ladgc|+PHA-a;Oa&n-n}b;yO7Oz?cTk6X7BZ@ zSDO)6i-8|Qw_lk&*OE9pbtohtr*gr@8o=d3^)r};x}zDeKO?AEHoAyI>xo&_O84@* z0keNjJ?00S>Qcyuz zM<*H*u>d@J;NO8lOYk@z{{E)D`hXKe{KI~Rd{xlA30b1GIMkH_H)NV7m(E*xZ#k0M za9>?HShZ!}c_1V~0t*L`^uET|bOTMZpv#X};}{x9!9HnAMN+mZMf6b&{hH zi1?jJg1BT3)Cg*n0Qa{w0f>=eub7xRQZs2?$4{Jq-^;RvXJh8cS=V=xx=j5Y?6x2Z zFfbe$LQP+~@u(AXKCQmHV1}KnR1O~wgD0)MI_VIl2EY`pi?@N7Kt-`rS##k|q#O;1 zg{xMrTCivlKT3ANjs5vEvXo#mr0neF6^c5!4I-tKWvhFax{gjw{U(4ktmGmHi+oIk zjbiwxL_o67zD(PF=}JfDl3^6S5t69>XItr2Yc$ z9H0oUs(aKmZE%+BGx_RK$9&%T`&M3Bykp0dbt*bHOPu!o^~8A2)+mpNn7F^?jsFh& z`$^|G)x&cqYTsR`y{+(?(yPe@yLW~iQL;TUb;S;;bzAD>eoJQLUY4IK*~R;#+|owJ z&)n~Q^_4kUGH3FhoBETY2=<`uI*-HMoUa3K+9phxpq5~wQyP9~FQfx?vw zo112shTK6s3U%Avy*p>Z)aW&+QgQ#p(9Fywk<(cUNR z9nco;X=RWa$rpaa;Es3BWyVTD6_@S*G9Qcjalu0oz;8@mNu#S=%u-Rq#R|mN)~hQa zURMKr^?dcxGX#d!sM(m?U={`@n2fowQGNV5J%NeLfL#!Qo;(3m-9S- zK4FoI7-Ew?@-5+Tger1ZW3Aq&eY*LYj_(CL4+2NnAj;d%B zo5>H9!BAST7P*yT7J4q?^b1$%BwK_RXhMRt?fo7Otte7E>Adqi&jV1}&<%MN(U9IA z=nx!lGfYY5j<03mK&EaAh&*cHDttYNop2L;@rHLC=rJ%>& z)eHEh=`z;7Ie-2<`R$<<1r)fX^M*SN57>UWx3?rlXDeiaCG!9H4_};Y3TR8la(Ltq zpb?UwC9L4QEr`F~Ffe_~_m72RE)2P|wK6^uP-#Z!-g6RAK2tso_xD}p#g)SMdkGD| z%?KKEBj*9CXDvu&C{6oS%oXs*3qakfu=kcgatwmBz@=#fZ>nuyNA9K>+)>6A&yaWB z5b%TG>LBqLYiR(LO-(uPnl*v|^~46#O!TzAnLCTG1lfoD`S!y-~ls#M7g-Q&&G4qfKpTsu>RUJ6p`Qz~@>@1s0?X_=^j` zM3C)-ZU6P?vIWmfOjIOd9d@`ETICilU1}&HX;O-`#;&^%1BE@U&t%6U{)R?r6Aa8{P&F;Hzv=29JWIC;$ggAQfr10 z{K+Mdr=TAfC~aOK8d@BVXEDT%A;a%9-z)(g$y?B+du{74O({vYx3jV9a)|X}h;L`?LCr=?lHP^szD`YIuqnV6&;|f;bV)@08 zHz^XM&Ze$i80@J&gEk?btimv@O+5k&zB)t}TKKwYw&8}Q5zcOnb_EzELa~Jz923~vcNA-n!ZRM2CHjjA3QfiLYI>n z>dcBzdUTnSaTI9j%xgxiOjfgz{(Q&eCiJvWrzCm|E z(*VNjBqb}-{BZf#uXlkuZkg@)`wF*p{kYN?ovNdsi%mooD<*Y@*x>2AVcNk$!lwX-G8|Eit z_p6Q|oYOM#BNI&KiXXWWnrPN=ndo-tZGApaN3ga+F;aIyo9-AWUq?a|T z*673(VTSu=(kn<10dPuc0u2!W;}ELqB?`{0S0)LOcs0XMPaXj~xC5YJm=_GlB#p(m0hWb|xbj}f8)Dmj((PcU3_!A93#x&Iou2T&vj(YDRX_+JoS-=-fcJ0dafK z3C`t%eAokHG_vT|8T>m92n9q#F7ZhDG z-+B=FTmWmmcuS*>973Kb2C9mN1t=UpVJ-;K-LsM$3$O@~lMs`t4^pl}=ffzST7kZ= z)=y6Z(WceBDVEBUZSQwsp19E`zm0Y!+V{TsMQSC)P`4C2xlPsve=7(3BT@OPD4*ej zh$Jx`;a#KMRtB9{R6Q*eR8Qp5qM9pKPY-WiNOOUl9(l3THU>Oir=QfN2n_PsRmD~hMTJ+*?GpOc^6((0p zgU!S^?t&kd`eQsm;!`e%i-qhW5aN}a2T-D6`1j*{2mukEIdcL4^vIXsA)uxL23-vj zr-!hfE*3bZ`p|>Tr>GS!rA7$eZul>-Knbo=W&G|Pb>sgrBjS`0a;AYIsopC;nJ-q6D#16=2N*kf=F%{58r+r_3I943xzK8Q2brAQ zaHPpm&nLB)0ncZ>cdClT*u(H=HtI0ipLmtZMG#uleT@gV0Q0otO#h-_EMhSHkytzt z%7M;QAR_PP)YXl?kmkW%7}Wwa4|4$Fsfi9vc6>TX4=Ta!lp=&=y=QDe;wBP^o7HpePnRe0`(Ip8D)KzDz#obJv*R<{KaSAE7&JzQlsR*lBt!}l9bI-NoJ-~}xSJc)r6 z9k9h%{HEa?cx>57&8bWkn=qy8}W@Ts&78^VX*>* z3Dn1 zyVZd_TE7fNxht-7^Bz_{5xgFP8aQL);xDEQ`Nk@YCy$-%Ac*45>9`6^nACazp1C6l z^<@jJW@7i*Z^Ssf1U2!3D=4qYd`At4`Jjq-p-`E8h;ye%!c$dI*u$o&Kb?_7)-Z7g zqUceC8A>c+Kh6^=3}vEAj9SN?<&r7@Gn>Xv&~TrKQo`)v(8cvxj$SEr*ar65%r%@N zCHSt@STGoCZ}=PteS2p`o2R+2%^Akscj)!O z$B@~NT8#U&A-Dj<%m|^q!C>$$r1zTZl|#-|)GDVsl1MZ{5E*+fC)RDc#5*5J^KOU> z$=Ni%0EsjC$1bpfE^vvoOQ4POu81m;@Te3VUB$k>(o&?~%nk4{BvBVLH^QJ6%9l;| z2?jD4?u+QPr&9()p9G{s`Hm-9)yOw{<))qd)E#f-7x z37&*UsJ=>37E%ccrMtYq2W$EH_gHPMQ)kc4wx5q~0Qbc?jyj;ff|^@cB~eGV;4BEh zyc{!5JGD|JJ?qThXw(VBAgo1Q%A#&B12jB>0d?jq7N??rFd#9-V4KL_qyq|rr5d#7ILI`8mzSrGX@T3GmkDmL z2re7gcaM*)nYHc6pY}Ia&d~ zE1mv;y~|WVy^Hw33(9)zg)`#`G6g}oGK7SS33B5Ff33NC&+WT4c?nGR`_?Dv3o+bn z$%6*^4Ym|MCUmlUbWyl?myMm9)7gf1aW{k$sRW{zCCUb-3Q`eu_E7%%^QkKOlqQ#u zcENPv)BqsweJn&%QOM=+$^H41nLT|<9)%?-A0yCR_OUt=8)qpZ$1N04-8eSy^#geL zwU)47h#*V0*JbFjRAaOyalp=YB)rXkKJ^w$ZLJzfLq+?A4X`U%ykINIrv>Wk-)p_4 zYl#v9!X&Ll`w=A<5j;KW_586lUU%U=C6^FB7=72r6Z1oWUU^2lfvW z?n%N<^YU9P$TdetHu1hUNoe!X&{YO6_-7*}ieh%;MNp4pnG(2XVOB!ua0qaf|2|Y1 zJ=9YyHEPy@letJn9H|e3`{t+s`(93u>4|oMG9YKRr?6g!pB-d!I(wk@-;aDf3wm!s zy#Ixc5z{0)sC~=EBEM2Mjq_SL|D?gX?9I|>`5Q|c0lN^Tm5yT{7~vhD1e)hRHRr@Z z_mlMCSCQT5gan8{X8AK=O@X=~!o`W-^Ky-M z1||foLkqf%E%FI){z&b*125+ceolJWcm)r{ckk$8raw0Ax zAMri&1WL&^2<64Vy4B_&cBs(fou@~7D;%r#pLKl9izuCcMuFn7li>J0Y1%_NTc+_4^e8{Q*%_0PfuK@4u#jR zU+*x2(5-~YLL51^mVJf=!^x7isTiDwBrF1#hoH{l(u@Q*OT`-XGm1UTRocP1su4!m zObZnSnYgJ{fX?CoN3lqAHcTg=E*Gp=k$aAFb`wJ!StL^w0gzZ3!{QJZDrQnTlJ`Sf z!^+GUNQ7w;DNTXg-tdcpHQ$Pk934v?ED>(B&C*E`z;0^@l^>R*+4RxcIN$ zZ(3Bqf~|=T&>5%nE=)O)4nHAI04`y{I8PABK!I(*v*p7*WWEbx&PVsWI-P4VYnD*! zJI=5KFGA?q0ni@`heiZAA)>e;SkP%#!%cFdg|2q!Boh?v=&Z>91${N9c_s}#{yIJCGwvRoHzEa-hY4g*|NDTpDNf#$#n z=&GB7+27>hd@~AUkO3sT%2b;!H6zli(QZNFB!H{{s(Vr;V2gz$CCEN14k;rZ=Xwy~ zfpaR*(WZYcgW_gA7qACpkJsmtw8=tP4V+BEVTL%;mzJIbY8|m^JQj1pj63> z)FQ!F5;L{hr z0cTc#z`0<-0@a|3=vQbHw-@^f@Qj8R6C4e8@I*q@LbPPyJR)$i6&A}G+;7pz2QGv~ z3AS%pYcS&dJ4Km0XDP+4-9+pMMrx73g{@ZK!w~<;C@HSmA+G`w6EHdm*N((!cl3kE zq2@R#i)N*Jr|8QjlR1-dmB=fAX)~`Fo9Nl_#Y0e3=k z*-y{sw0OjC0opSdo)^%COWt=1<3WOZZ+}E9y#AQ)M^z&-G)AnSK< zc#uvb+hmE&!>vBt09hvTSSR(K2u*3!}8y3z3SUq7&dz*|$% zg7;w8d;z=w4tPmE2{0)@bRaHSL2x4(qUpi;(85Jno_O~6Kn+D6HE(f@%*JJuiTS7y zWH8n@;2xm%D*2f36qi8K6@*NPds;!qX^l>(J?|Ns!A{w_hAphRQ4c9 zS32?-6-1~BUSXQ#$@EnR0_h}Ylb8#_GHDyGw^o@wu})RrlbZYvL38J0QoTuSUtT$< zICmb;zH_};S8p%fED)V~@~rg!Z_A(*7n%v+Dr}Yr!?-C-H!T@;#5eAIbUb4>h_Lw- zzH#iQQBvTx`7*ap@)c^;<)!!@pc06C42^lmPcZqh0Jm();OK6Y33$SlbkZoYk1!Jl{KPF1($;4moEXKK z!;P;&RMHWg-+Kk+JY(VaLH#1exjZpMxO~ZzV>`*Q!TMq$C5nT5uw%2r@fSxIa&B-e zUv4xNr*ZKyiT4KEJ!1`GgOeHnK0f74U9K=wSh=D%DHcsBoQE6x?i5&*++bbcWug&+ zj(2R>h3`D!%va_F9N=9U;d9H66=OnV{n*h-Og#h8yXy zOKGw)Ay>ZyhaE~K5{sJbp)`ErJkPwbr*#1X${sW12Lo_K(ChJ-qzvQR+%X7MWUIE5 zM(Z3Vlg$X$Z{m#8bZd0Hkr9Mh1m+!v-!tshBF$O{2&6|=5YjGeBbmm3*z|5 zFLZ3%Iwl|j=xKg&ZKTJwZ#a*3s|6?JvDX z%y!PhyKf|P%#Nd@Fsmy=CwWObJCT!xo#xT8ACNU3@UDXvZ2s-zCeFi6;5_?yoB%#x=tJm76B6jJMBXq%V%%G1(P`Av!L_m&qC-(HG4%bqV1P7JNDNT)yBSF z7^b{hrt9ODj-OCxL9?FXOu$&kzc~ZF0ocACyacKoaXdI@e4{~rVI1Y>Qh)$2&K@D! z=J4;ZN2vZZ=N*F;;QApNz0VqAr<7f96lVgCjNQH)7CQ0-u+cDqlLMi}?1WXlcI*X! z)X6=K5{Hios$>AKJ0lbkB-4eB8yNE8X#Roc{jH2{oc=vPcQzIM@83QJJ>K#(o7;`H z40Qv5NDyUjs`8vF4qJ4ZPGllM6P-LD@&Zs`7=V0#R#h=uWgbkDEdYMx;U!Bm^m53; zIUI2f-52muKu3Fw35Wpsf(=~!bQ~koa_rG0F+7CC=rnbRdhsW+XKm0zo+!H?y@m={X&{Ne{}R<0I&p6+}dFpBba5qo|RIDl_oVL2#EE={A9miB#V} z^#JCvbj5}0N4Pvyrm$wwc`IlTK)H&xaT4w6CF!(2M;9QBXq^;wC~ayuTmXU-sSKdw z7nJ8I z@-|`-kTnz$Rqvkc1*q4=Ho^g*j~TdDhNHuJ@b4wG+}q2I<9SW*$qGl|=u2`R02ES` zpt|-p>Au6md`$BDBRcuK+k(cn>XOE_)|a#Z+X#DK3k|3jZG?J4hu{td4nXlF`n*n%z-is8RgVvwTAzMiQ3Bs#5-Ac79W9TtT4HeA$z))8{ zACqbX49S8eOAI8Ayzio=j3_?<<#}3e0&Fgh1I7L&;8l0NcQ}HUG&)v^Oxbj_8oduT zb1onbwJo5N0ZkyMrTD2d8@tyk$T!6H!or(Iu6WFNgu~Z%;n&QApw*V@a1O@D-QwB z&s-s;3b%x~&&NtAR4|c7u&+GED^zh1{mz7+CW-5CpwcOg8lS=uM%K{xJ-sW@Q{T!22Hm zPz9YIhP_jYW4Z#*e6PI;RkfMUwnU9uZ4SDi4B;RQdc-n?$uQoW%=KSPOKw4mlM{m= zQ-n^+3tpU2v7fp_U~Ds%cwk8*MdmZ$9QbShpZ3l@Ea$v!_cskQlZ=PQCX5VjBch1L zHq0O@Ls24a>^Bvnq-?|33`X2C*%z{(%4X6|$~I+4p%@|CByC8Nl=ZoOzdY}|j`w}m zv5vL=S${0Y9LM9Csr&wIzTfM6UFUV4=T)?)m>39Y>LBAfgxGAc?f8r@zWmaYGhP-} z!r}gSlZ!71&4QLAzT{ifbR#$3@vz;GfM|OO(CsD1lgtTz-t=DEr<)+br~kp;2>fY! zX4YZ*LG&SSQ3XTlm~EOnYXY5G3~~yj3Do-TCbToK0UbXvUc=y!o+j+6dg6KyLN;!( z=$%*jhzDesGV6eHXZ4nVgM!`FID6ht=@v}XJhk< zg(=QDb0Qc4$Qyj$xxZK~pga}(wDKJ8^etp`lMWqHq!sk}(|=29T_oW0A<|52Ut9Nv*I zaip9}B|ZK@Nf|Uf8H6(LL+;83Y`XkC<4G)OWRC5$U|JCjHt`z4bt-uC<~I2@Cn{uY z0*gUNnU8*%#`#@V{Lmd9;UkxZn$!mACJ}9?^D$MBS}hHvurD z)jw9nVgtudi@7IFN@1gLWaz@dQ9OYAMDq8@H%1h^{SoGicJtxV%7{g0iI~CsKLr5=Vz=n8s?#e8M7Caub*$Ix1)>?>Ucn0E;IYauS zS;9vkT^=S3GSS1qVJ+CRo@S_kIy}Ma6Hbl~2@Q=BdO9*&iMa`Xp2qJ9Vz}rMl+3hr z>g#(#tq>Rx!hJ*md4mdNhCGvT;$vV7N7_|@9MV)O`n)-iQT7!f)W0c zNuq>mDIrT*jC@#*C%oTmj;^xHE$4`%WY=w(@lXZD%S9N7$^Z%$NI*Q!t*C>nuE(~L7y!H%GM@-zr2VK^qU^fe=g1Okq8;5P zuO^Y7+;wVdE)Nql^B8ctPOCC>3J6|ZTLOhQm`?Z7^ky54NoDcy$@xMAL3c> z+^CnVC!vOelJu%;t2zgA0&D4sRlL}7=pOZmC668*y5!yV{y6#o{>SABa*A?mEXO;2 z>aN?49~S@ur;{wy0>N9hjGhsJW8uQ0xID={*a(BZwv7}v6jKfes($GLg2UK=FV!>c z-zFNa0o}`%eR3b~yh=$k&aCgHG;Uok@x-Vso#57rW^mcZC#^}ZNPn_LfHEbu5!_@~ zx(-ReQ@>65ihwtiY%@N-?7Z?PjqDPUdtkS=isGbW^2IgYqRmuJ)drUOD^b!TOhGcNC%pn53+F zvI%RQJWg`HBXpb>7jaN(oJwk6y4d85Ce0)=cU3~J{)0Ete*mAq4F%jUD`N+O&lm-o z_}6mz+Ns2?1>efHDlsHcxJh3|h7ZUT6>!*7Qm5w+_*4-j1QwzHksvz~_Js`v=QVXw z#|J@O#eR{Nj+hHWy=E7T42gf<9?GWR6T&ISkY&>De`cplF`o$xjfg3O_h4-OA!R^y z8&VF)Q+6QlB}r2bBeK?EfBG^jLRgc+aKMvH zJS-9Z;QMB>{!&|70i+mW(})?ED3ip<1#8V#a5bU#obhUktV40y;%p|?wLi6e(WsW! zxIpCExVUApkhIpHKIRR8I3`xvZyo}ysYR%_IP6Q8Mi@IVy*?oEvex=0WN2l5Ne68xLRDYt+E_UkbR0R8y3vgIz-mT zKn@e%KVHwbnD>IKa9J(Sa(~yKRzm6JkXf;Fwv^<+q**j^q?92+mst<68PGMkE8&I&HL;Jhx-V&`O!qe~%x$R-0e5aanJDZWm=NkD1mc0_l^6iXMiS%v#4((ok!BsEn1;w5#t_p01M`&z2H58r>5yP{AeIJNIu|lLP@H3x&vI2iD{$UkYD|b7n|VI% ztwGi#U5NI{8?h2Dq1{4tduX0nL1YlQ3V^B7#*is>p7aPV*X1UI)w5WWwzZ_hn|9(b z;X`PNz$AAsfrq-uN!ngb&op*LFrSBrZ8#Nkdu#f8l7~oGqJ6aTCSN-5^Ox$N?Bd-D z1*kz>79_36=G~{^gVubT0m-)t@M?@>ur-Br+k=cN`){~j`$F6nS}&`vIJ8_IZ1c~z zDfM2QuCSK%h|*;aY7MF8q6Q+DS`5a2%;NxnAiuYfv&4|XZ8}4Qw&)r^hc)t*XPULDy0j6k{nes zF2J)2HJDOWZ2Xokcjd#ziJCz3nw6PsAu=H`P8NJ|-G&Xm5v_5FrJl_=KJmg^sw8DC zH`3TOtl2-_t%y@aXDeqffjun5KlYJJr8rzm;D8d4O}`C;uW1uTX$A&}aE3nvyWs>T zJaB@-G3+fzKV2uXAGt(2{t~>9DJ9px%MM((BD=n9Nh)+lD=Vw*-=6Am-Kh~YNh>a& z*cS3)E(4CDC|8$nA1j}26O2C$7y5^!J5qZSC?5Td3vwB!`%Jtqm4THqte{)B&7rm$ z%_XaI1*?f&%xAp%YTJRiXLbJoRDqxCc_Zw{_6u`3c4yE(g)O$|S0ky`G}=_JAC9r| zohfppj1tTU_C;hB_=9NV#IzPcgeP(A0~(wAZ}QAFftCAnSB)3)X3sf$#J!+lL;HOx zD0)uQB@#92((xjQ4s+nd7#Vfu%{n;Siv-2wF?w#ULdTQIvl0gm(VJqk;GDQHnzH$N zx(PF)Cw1$iF@1-h?C%Aot|X6s;=;$y>-2i^PEJv5oyo_uX`dpFTG?oTk{&jGjvuH< ziU0!zO(IN+RU!P$?6O@zmq@?dfEhxx_q{LF7Zxy8_+fV1G3X%pi&-sU4fhuvE(A;~p9IR|>N@Z>m#ah5zt3O5sdxuDSU38G$xuGoIwSzWJeULa049%Kgv2ub2! zUv4RE7b?FaS8Rh#B}KI!1q@{xmlS(ff7(P@AlyjO;}m%fn)ad#k_=V!h4@B@rwUi> zgz6KZr653Fhwudv4dj2kmc#x%fT;Yla{|XQ4W+3+l-70)(=ssy1xF{2Vpxd?4Cs8! zPuv|6(qZQ{zl?jdG5L33;*kSr>L717W6n^*eYdCf$H#bIU$N$9e&;vFcih8DwpNA2 zr#+lv?ppr$Ru})stBL1x&zD`@eUFR~H>|S$oE+>|{vj~(kMx|bFDo>;1E22X*v=k* z|9OP1ROI==Fs-}3TTkBu+hgmveT z0*TAlmCB}P1|K0YKky+SmUJb0L9rW8?T4UM~@| zH%knqe@O|Bm+f9#qips++S16OaJvw-i^7Lf<`})zqYg)OV87zE0!f&4BiPy5vuAm` z*^8rFrzNd8W-3dFl_l$ET~@uQd0mX{>lv+zsi&X4BBowQbe&3ln;a0x{R4bs$&W_Z zl~kMQyEVAY2`ME>2e1}ubK%F3Q(t9hh5o+?KjPmB}-g`R=9gLp5$|2d> zJ~1-2aNak;V8}U>Ci+$mjWI7B*p4OQFRA!z07bp~-`t`fK_F9D)E#Zw@ zD!X$pTx)kffq!3RxZ81p4$b@0E?$3{OD(I%N4sg>4}dJ!Z0lL2#ZPS@9tL#bM3_gn zws7IDegSXd)5rM_|6)RTu4+3>1Ds_RdF)!fInB=MSDnvm5*)iKX365k@7*q}%vo3v z5xF88D?*pD=9em*ua5LGLeTRQgI(qtl#-@4^6jTp0hi$%%zT5<=j5ILlVciEVnH*i`6FP&i=Li_K4#vo=l%U z-Tifd)Q$8-qx^Rc>REMfTwO}(F}GD7_5ntB zQ(Q(c6JmanSsgPQ7i#wQ)pdAma^j$XrTFCd$#?y z$L;z(bHZ3tKuN}oWEN{j?A!TvfRr#|o^aiUxc$u_pY>ZGdi`L6{UbH@Vl(@5y}sSp zmZU3xyo*7?V2roYS+530pWD%=~NF`f(J zbV`d2U3I+v1MYns_4}wXW3n87n;Gi{t{GV>-JQS;bbb3Irc=9%W&oAgr#DNeqBwsZ zEhK~f8GgThg9c8VKCV$YdTHzKdil8)I{&?E#PrRz-}5*lAD5qR8#}@MQJ(wQ7Vcwv z#-?r0rcIhm_5C8|$8+l&F^F%nbtZbbLq_=ek}PDut(I{-(hY%hGKg#PG;&Wj~mvz*VG6`T=v|GOCT22!xy#FGa#Bu#03H0JY-N1ZWiHai*1>(e~lr zXB%(M9MyH*zOa(SlA}+FP;H-d(!V^m@rFnVJ%?*0+%IY|$nJe%y_6=yKJ`4GJ8J(o zkwfP$eKdMh)KT;8!z1mRl0mQ@Ev&*PI1G8Txk;UNo%JufkW-5<7znZP=6=s0?bwL- zf=p@<%L?(=1Hz}p)k!!wc}c$#v+bGNNK)pXF~>z;Kbsx{aE;Z0-%BYGbYhpX_3z%Y z>%+RDdZU4xFKTh|zrK6Sw@cIYq+U@(xq{!v&$g#OZs&_kayI4Nz>U$odjgC)t z_sEm4)+myTvmaiX2)-4P*wLA|JD&k{;uyl)x8Hu|Vk=EJ|TzONI?0%K}12}Oc z#!bUvamUVB25vDQjp&QWgdzmBn`(V+8%iCB$~oIlo?spe@va&70qVn zGk3(1*e*u*jm-Rak2oar&d?Jb1v6~FkkVVDY2pNuBmFi^5{9#fhrFW?FShq&-&;g= zZ1)ot8?Zpk6!Y!kmnMi6HY;5+`nEFyzo8@d(*#MMOiv;; zS|}9WqBEr<7!t9vQEiQ6wXqQ(X@@~*#URLy;aH9O76Dy|EQK6(-^GpcX0jb5WVE zf5l?hjag6x$cQ1fQ+64jYgd_c`L@Q=bV!z9$I_@k3R}0mp2oqUrkek_3I&`rF`aQ? zf{ox=B||vr-2;pfJ_BrzaB5O|`=lJMo(=lH52MqY_S-G+O!P%^n|6C; zER3|5c()y`sX#??$itl>U7|?QOYUW|BU#J!QM!E?(kBx zof&8(BSN-cnTKSe>$BM$Mj(>L z3u!c4 znAo=lVe0p{bd5xTO|GQqzomK1TnOc--p<`3oLEg@3T4Aq5fW$nu}Lh*oCykgqj^Yv zH4n)sA0J?b{_gUuB+6y7IH#N_*>6fP$aS|=KcfO0Ad3Y>F4^zERicD*?KQ0tlV>K? zAddbb>J-sD|9L@#BE;;!xYsj2<0lpI%WX*eLAa;1euhCTyh(*6_VeH!EcFp2CSd>K z_e2p#85<7E%`wW&7404Jn~gjuima(`*Q_WIJ~XpqKqPKOzoP-~!E8OwW?34^z0|;w zBjy*v*n%JE310vcoM9(|s8}LckB8yAd0=8D>P^zBtplBdf8g3d@re0cRr>e8S2;Z!EvQFu!0A~IP+f6u++{P%%7kF$7@{z6D zlYgwQY!lWPphI2{&TzgFmRM3roLGJYyY2?+MH@Wx?IZ0e8KUv2_*2yYM)=tFH*4GW zC!xo2Y)Zw?5|E2NN?J}GKJ|y4{1OcysSRYH8BmUZ(4vzBs+RJQ<0W$P-%R!bDIR~? zOXl5^h#X}H9Aih5VRk+?vQ}g=MI(^3IRALgf>kT7d&;%7uDLcQb1Sqo@Lw3Q#X~BO zIULzXEhauYfG7EmO(Kb<>`+=$V>texB8;1o#DR-Pw4`x0-Y()>M%@wL6XB~!ve1c( zx002{@FE#M1_wg?@tj4eFXxwX1!n3Mqy^I@q;=baH3-mh{7QdP93cyeX`%=gIXH5lO6^v4h z!CgGb$8KScLeJ-s|r?0Yc3CCpxLl;j-Ki@fG7ySTvp z(}xh?9mgr|*dKlzs2}4MY^6Inb7blAVyny%jlk z-+7HK;>RfVV&Y3FXq-ZglzDOj-wC|UlAYWA^rT)W!sNVA z3Uma)LVMxWwh+^UaurbzyC@ws@i#g4$Chvzh(Z?=&rb8McgwnMAC8&jDsbmWdd@PY zTNZ>ipTj{gGeL6*qy?lwKlF;olfj8{Fo;Gvv%(|+5!+EL9Od}bl&k4W*kEO*7{Ud- zZ@T1vbUZF9lRf^AzaH0lPgT3MtbU0+GBOj2#!c9IE(fNNj7thesYwPmk@g;6HhmcF zaNN0RUvyGrPM%STR`%a$k{#!TCMf)|k{K*dKI|rkfy=YmtN>ZahxRN?dvf*udGdXh z?IEddfSkd-biUO-$Dx}DATK^r_h3p zqWxb_%66u-l%wVTysKU6esAImSRkfr^4~10-#>|L)EC<#p%0-r7Z0tKqvH>eL(SkM zM)R%HN{t~_>hN>qxZb0tEdx*N!m5?zoujSlsdE9AeUeeoJ=5E=bvH3IpOFFL2TD}& z%>F?3Q%3&@la>jrOLVcCD&#Et-MJBPC`}^+9!EJ$j<3%?k(9!&t<>P~ZC*-oc-{Sy3_U0St?u4y`)gT59J?YzCo$8(Q64U}ohVIU z^3L==e{uod(IS;`Us`d%GlH(8^v!^xrC}I2T}^kruL*vL^rQgvM;A_5A&Lr>!lKvq zN&x4DoNH1ew4K~ioVplv@q&@&Vi{9^_*g&otynB}Bz@I^0*i4J4Y5Cda?7u;8EF03 zeU$9X?C`K0OZJnZTH|#IFY&~=90$nIlB5dx50ZrLdivffnQ1KtJ1_RxLaz(LHsTwN zf#P(G|BQn3dYIc#lAOggG{M`Fz=0BXf?bwbvHg@R7n-dGBc4|aJ6fI)2q9RoqLC4i zgLE%EH_i>%3}#KW=t<8Luzag;b@5fvs?oB#p9^0;xaI_mOyh@ypjHK_t-BsFUnrX7 zcjr?3ts!ZV@W^fu0c&KLklQ&%vLHLoM+l+!&)+vd!;@DeJu1%#o+~E!r@6nWlG>-W zRRi~=#~u8ktt*Ro)hxxa~-MQK6S=M zRc>d`m`E4tDaV~gw=#t2^H=iVCWAmNESO^;vk#=d%=*(>VpZ( z%{*N~3L0>j9gC2qEZDq6nknR?;mOI#=IM2)QN`Dd?`lh)ma(CEcb9vj>n7c-KiuI2 zlpK*Th_e{!Oj&A&&=(3+M4Q5xWD1sCwqApNVMKXpQjrfI*dgs~nS%*Gdml@(@!e+3 zSOA3(Mfm~oM{!YG`vV(N^$0@c`MiLUZii_7wAgiYlKIYSVw@-RxGJ1(vRGgY`7%~I zG{TXm3l6~wY3m5Zx0-}+($pxDG>N3V7NK(3Gh<+caip+ zoheyCUC`bdzR?XRIqol=);vz5Op|gJZjmP$qRhDebK-@%PG2vh%SK8+QqCppSq{@* zJSGWfPV5P`_tKv~RY3=Q3sh!;<)ri@=goUFB>ij9Np!EbPD~y;E+VT@WiON^(Q4fL zF8|~09ikX#I)j-tFi{4tu474#oTMU1f$fOXl`;cO2C4`Zc#n?^ZRU(7k!OA;?(AYr zC|A>RcI=N7UwCG1<=0W`BY_eBK>DL(tR?%7`PO5%#ffn&cjhCG!?SB-AKIDlGGvy6 z)|w8A@DZq=&mHmZEhAopw@&ZYq+>^G>Nf-|JHd<{o{IO6n1vGN16=dspl|q!LYWU4 zBW%vLfEmU=s{2HyeG|~HMq47DSC85Aw^<`KLwO{4Ay1JrOh{auG<-og8KUd$6CfG% z_5`aC>k@Gzl;gId9Sil6%64_|rp6GtIi39W)?#BRcerx&qlJVQY1RwbIJ|U!PnuSe z0Pf~k=F%nDUU($DDnfoURC~T)L>AwW1HItj%!=V*f?X_glR$ z_&rZGiE|hTvs8{n{Hs)F{_4DC)hW3S+-0ozoWt~pKrfAsTNrF6S|TU)bDOkb04W`X&`nRz&^D3!VBdM8;FcHQ1NoqXdJ%dM3g3z zR_x8>@ILn|UtlMmL-O4xSpG}Uvnd%cNHcQ+Tc}FOns=b#9RPpLpH*IC@~+6*Y-%g% zmR+qgaMnbJtQ99`DelydZ=~$RzwZsOGSFHr?9)!rU>141E?fo!0aC}RpRTL<^q11~wv(0P7>*@`BBZ$`rg!v}QxhOpJmH*_$hxJg8wP$~kNRyzn z&3*@t4zX~SA|wuOsy_d3X?h7+Tsg6%?j(TJSDt^Ss$>XmCqsTg=h|Ci?<{&C^%i#2 z+yd~!Al^`9A+c5Lo{|T1`Np%a?*75+R$_*k`-r3R;K)$;1Rbpx!ZZ}^jIw-!RZ(N|1){!~-EP)mESyj1 z@1aO=(^S!9?m6&66`H|0PQeS?^0C2UCXc{jrAWS59lCH*8dq8g7V0mn;@{l80>+C6 zvrY((+s_0CGK*rL{p_82(}0nPTMDn85+c$szh~ zD>W_aC!^2UHkR4%ATj;xwHT@ydP!G8OqY3+qeZZxJTfa4OzL-qa3`YxZ2Byl9LL{2z6Cjv+AU!e!_LDM43dsccYbbpC6_sDuMHM%xh~0DOOQ<$gD)JKaAJ z6^Y0qBMV^{UCGYQR$is|m&&UQ{Bj%6c=gB9k=9BfmC|Z~`YwyaV;88#2Xh$b6O&eH z0m*IQzdgI`!N1<5es!!Ezk|q)rgGBVI$+xj!koHGi_|dhJz1m+r{_F0wH7ik<&1$M zKVP6O!dRw2b`}?c=qf4^!3flA>#1JbP9Z&UzGQv5H(VjI^A8F18PG-p|dX8E2V(%KJElpLHSZq7>x34Xh`BYT&;s$brOtr3} z)J{;hc+N18)(1-~XJx+=fjDvnTMz!xK@kBLsc*--9jVBI`(uS&FH>V{ z?ofUDvA^Q6F8EevMayF%m!H=C;YqrAxo_Uj7Nxzn9*+*1k^DXxz5stppvWy!Q-+=; zED8xHDm6;@Wa{USR9B;6XJpNy4|Tc6%Mr%JatWik9o1i7DHa(B2sUZ&``w#`EA4J& zmw*1xdNRO3-jG(zp*ib=Q>Un5sGlnCp0J7JnO@BO==)@_a0*%~cfO~_)8PtgFa`b@ z1$|aby_FD^d{Yfkb?=}gmo|;Hj@}8Wj@kN4lazPhuR3&uZ2(CU=hx562RQqoas-tO zRs&7*@w7qIOofyN26fHNtH)6AUeZMwDHmBNzz(!HB}@l)P*a_>weV=fs9fOrOvx3| zB(+oyks3D!<92wv$zWowHH;#x%r~izNA+?DG(YZ{Z7R)}$ii-uFaK+K8aGg5dvDN8?4-Y@?^YN_P28h5o#hmlC5HBIs>3lS!M&_cSz z!{;p#v{emDC&uE)U^#B#lrqV6zqWzFgbY`)0HyW9u;ayrZ5w z>FN^@p1KeoVS!m^^1C^+jf$yr-jXh(Sh4NnZnFTBLuCaS*ZfL1(G>x_r+Gx%=;RSo z-CfCb$m8^SDLyXrCACL5<3txj?@<`y^jg)OZS)MF%G5jvMZm+`n2&f(XxkA(MX}H2 z$pa(Rt0%44V{7>xjZHn};O=)_7TQpwfWeB52^APg=7mk2VWtTJhBCCBVj~GQ$RKa8 zVMhF_LuX~Rmkbj+jMIK500XRNo`*VJE;3ah{ z1U<#&62r$AT(N2{1oAa?%$sT041pTSTuUuI5NSQpLGQSwR2T*7f!J>wAgV}D8#ZjC zxK;qx*)DOpz(W(#xXdET{IoTJ^40iush~5>ln@y>v>xOy39#+w@DN2TDY{69>*DoK zJ3(ep2m79#vWINe(}mPiWB0rI7GIOg_MnJl@i~E9oRPBiVfDM73|lE=L+%u`wB3NhA3}v@Nd^5E^ ziVr)A#KIl5z?d|Kcn|VF&sKBS0>f~E$7g}(h-Wb{#m_K&lMX(?Dn1{x6h{HzC?&EN z0>(<;>c)*5jjjrL)}_IgoB6&iTGcL@f|jLQpY!uciT`WM{a;XdtSN@fw*_Y*+9zaD z?+Hpn>MJZk%CL@a`jZUOf42j@1nKHa`{KZrYDxql9tbeBeX&7N+N2XQ&`K8$Mh*uA)3o4Us1P9EF2 zt-Q_bU(|Qu2CO995z*5N6hOhpGOP}QVac1V_B4I102^OjUCA^~o-Wh=3{h&7$I zUoJ@bQ)8AVfA?2CD&(&vcFHD|zvk{$7P9;`c~DJflBlspO&#Q4*VIY!SEJ7zapZ;n zVMqD@KUj&D@__!IeiUhw0N*b6;s9mGQnh5*YhpW6wj8=S)7L;V+n6>qF=wk*u)>=f zZ>Pm(wQ}dN4~4h=YjJy<6{)FC1`jL? zHMu`lFI!aJdc5P3Pjf%chqPro{?EokkLg99YtqiEZClqNdZW74Iq=eW%*7+aH*}92 zr`7hlIm>jtUiVhp8x5(|@!WCEvo#-Ibd4O|V`8g~&A;xdZ`J4crBNOS2Ku|%zcu=1Nq^nWUGGYtEWTd(?ZSrvK1Ff6ckfos-O#qdi2osqda73{Fq?zB zyP37v??=YJ?6lMAsaIpge}RXPyw`v5=l$BNaUCfK<;ib8#BU7i9k8vP#b1Up`i?l3p~NTa$l$_12>$ zR$myEXoEVu{Zx6eds74*t4I6PdBaK6DCO6R+5uKm`t7Uby@}qnw01w-j#_$P-8y|~ zsdWn~eD$iwr!c8zvF%xiB#h~Jdj)fgj*gb8v0fFZP&V}%@mG^8WS+x1A9_~rY4Z&G z+&bS*p<5{#4b~N5lvoG{FZDZp^r+49fdeV18HBnuyI||pXRsfC=jtHYYi0IJmo8m? z-$*$8^~pt-<^A-y=XuU&2kvRus*c+PO;>|KU~yZtEaP_`Bdx77$aB84miZ`CC)d&h zcE7XLWYct>`?r2)^0rLBxVpAkSIwU>F$LHOZqd>l={zM=^6Bmi+zDsAJ^nT@^-XVQ$eYu5_$|_y;rv^^Ouo0j8bv68RN@vzg-Gq6{ z``C41WBKAY8}8WU)~TxrN@-{P^5skYdxt7pn`23p%9GAD@O!5Jlzv;UHLZ#(o;E+R zH@TjH1_1ql_Q=_@{k=j<8yJloF{0?T*AD;x?izmcar)-FY4oaf;g>I8KAs*t)58No z^s{HYx9Okil}S`Mp7GsG8qfvrpz0D}TNjvRu+g`Sb06`QWh! zPn@uKi*n*0O26nuy}yn@FkF zl&0T{`GyW}?j*`5-|1jzKo)FK6oAo_%uZ{*x=j+Wy}3cVqJZ7s8yf6LS@x*dtd3sQ z$FNWDtCp7Db}vnN!<4+bZIwW%x2wsJM*3a&q(DpMhLX*xLvosb7|=g-BrOm>v;ALQ zZ?J0yd?L%pdpcbsv8lY=1cu(sN6jUY>TSDM`1?e|iS9uEJu0(q-(FQ112Ee%=y4{K zJYKog_8#~6A+OR>c}4@r&W`E(^auGmTfK3h@tV#z@7$TO^gpo zfZY#eQOhD|dx^`xw8fy-WEI%gZq=&QLRPqNwn%WL?vc4gx9{FSgck&$bC#?D{j;`Qo}wL=cS?`dHnI%XN=AXh6h8Q?^`eDPv! z#kOi5z|5kYI2j8kbK3Ewpw>zp>Hgh}OTAfqZG@FD(2xq;T-#i?fhuWe;WD}jrXi+Ejng&nO=XpRWrsdS8Q0STTGPn5HP+g^PhW_)qbE<^ zZ$EC_UC*2bOH1@lE$eF7V<}LRRnAB_PWFzDP5ctCUAuPKBHQJk^Api+e%*JCWA@N~ zCr-6(+RRPddi47W?{_Z#x@%LOlvn;n_k7Lnb^``HEU8#*WpBT^O{za*K7YDC`BLvb zQQ_eiH%$Si8&X(LlRPlFm8`$xm!x8X7}M3O63kTedlo%=dg8Y`v3U3_506Z`cg^Oy zb1;pjEqkkPQz{psF5h`GWXi7_hHBB?BR};BEh;jYjv02Rjve#MDvK5@=?_{c6L_1= zH6p~(=RU@InNo=n#H|}QyfX^^eEDH*Pror2_a(42o#fW;J;=z|3!ZHG*fo|$c!qoR zn~>v$;zmF}^smDUGE)|9|$cVY@!o=`t%VW z{v|L}_3N(G|Dv|!KT0Qt&CX2hdU2r#4}LFW1$5yY-;QI(h%k(wjQ>U3_CG0{ciP^(2wCxZBE3+ zjT6rit>SphRd8=j4C-DhUBtSSo-ll2+b9d?`AaUg>}xVSSr?AP!~+s*o0MYwDMbG1Thu&sot{wVkX(_F&orQA}%f{R)8`}4KYbyj`ch2Sw z$TCZPfA3*+)6?#>@9sPikLSDZ0A$|!(%&PJH*|+>dL_Ww+4--f zG}phnd+7M_8?Su8`!+49y*#iRqUclZhllpKxM-v4#da+;nisz+2454_XS*%>*~w+@ zkSnKIhCA?Q~~@un(xseKg6tPE^FX8Dh=Uf{o_Q?%7IHzsC3nHTq0P)h55{#_rr zzYAXLKX@J|w=xpQ_uj0S{3D0khh$N?wjRq(iG zJh=1`YEAi!tEs7*Q=V+rO5eNa=7&Gx`#=6|x8sS}8_=Z04CEKDt4xzeB(L3FSv$Jk z{Q2{b#K$}96w(U0W}CdZ3HJ%Eju+=&^pSnwtDM0T)+4%IGpnzAc<}WmA#IyOoYUru z>-TLxSR$nr`XZY&M@tx=b5-jRYWtq0soWv6xlsIAcctXyuwOKLg)XTiuW zzFTiO6un{}<@7&3ArrKpAA=%zqG?)$I4-(%17KgSN8ohwzhSfVrQWS zi9gbn;__phfI>}-=0S5)9@QT!o;N!@fBNmRiBmbyRyFbbwMX%#qJl}+WixJ6@+s5t z_ly0i7z8)P(CV+KhV88G#J;&$U~~*Ot9O;vq@jNmLc1MOm;|8X)0eZyYe3As2o`~} zEOoq@cr!y))YgbY&LU>yu3inJm$~nWxEJyt?K8t|08diR&S}MA79n+R#``_RGa$ek9il&^80T4`{pH`lx ztwU#1zn?iJ8de$sHj#WZON<(pM*%u`sFL4>M8IVbPM$8DF)VLmOB-4S^fRh@duYH` zZ6R}kiB|d9RWK%Iz>L?0!;+YR5Awvy>gPR^x`qC;a=rVAO>*Lw1?`r!^RO^>3clwL&(AqFs>J zWaQqN=zE)k3A*MCh|2cG`_8z?kk_s;trxxOCd3ieBPFKbn|$KgWsVo&sR{o(tTwlzH3sM@0OLRhk{+DuvPv#cW%8P+6riVpSNh zr#{gKShOLg=CnNTaX(uq{7P2|T64Hhw zjsG*<{o?$+<7dYmNVy*>d>L@;YfHdCXP_vYRtblQdulo_k`0E_TqlmD*)ik!Hj;Gl z4S_e***gC5Iui#6hg*|(eco(iaLpH<(RcYSZBB=W4@P@tK3T`y91<$w&tjf9{Z@tP zyKkw1uP8Zo?n$RnqlA9+_WfhyU*6&PC!#%E4@O+67Lyj3c~-BUy!Mw=_SBs__k|RS z_r`%khlcKY0wo~}QsSVc9k(l2u)b&0D*B4`VVq67^G-yIoTWjA%S>bkeifuV%DTT$ zqb(sz{(xdZaTsJb2d6v{Gi0ImQp<^pNgpf?-Jo(Mgm@mK)eA>Ums{Zdk`0yva-s@c#ZZmUyB41=ZWW?aepC+vMA4U`mr2qf` literal 0 HcmV?d00001 diff --git a/docs/articles/potential-optimizers_files/figure-html/pre_allocate_benchmark-1.png b/docs/articles/potential-optimizers_files/figure-html/pre_allocate_benchmark-1.png new file mode 100644 index 0000000000000000000000000000000000000000..4a8e076b34ca7e4ce248a8550cff844e029eacf9 GIT binary patch literal 45396 zcmeFZWmJ{x+BQ7ZW!JJmkyz>k0|b>&Kw?=qr4$j6G?4}=X|Mwjod^hspwg*yg9TC& z64I!IG}6s`TyyGv_V@D}<9o;Wemsulmd(89eP40LaUAD)-SJ=mbP!NFL(ud5K8w;Zhin#~qPDe+QBf$lKh#S2fB#G=JNQ@T@1I*l$qJC!S$E(yaf& z<4O`OIuvFmN}W~UA8$;WGHvaCzPKxN_4|=^ybBoDX)S~c3LREg_j~RzNE2m$sNM~y z7%Ak93r|m1_vDe^7a_*Du1BBt3B zp`(3`hAQ_NR|mQN)3VToXR^L|HuT-QLo}re7hbl14v3049oE(6qa$*xS1dq5Vs3gU zGCX`wf->_71qdC|>)6Gu<#Zf=-KG|lmMS_=ehFOv_uv1};?-m9zTe+ond9tCQ&LcP z$g*4gl6nmDSwq{%v&IMBuWmC>qta5Eoigl3d(L*-Nl8gbj8vYESL34xZ{Stgv3+}3 zWA|T-FKI0x^RDm}ypnkS{URDYCnu*bbLg|3>3++fC*^(j9i_LL9T=XS>QB*dXFkPn z3!ZXJJ;OQ%p@mAr(kLA{a^z-8R)6+XYj-yn5+AX-2O&UTSZ-xp?lGn zQ2C6O-dC?)jlANTIbKn_w8mRNrzv!V5j0(5?;|3zeapSRK+LM^301!==wG&jHUmdf z(uO{Zx3#r#(K582&QIh_OE_cr+YUBkz?vVv zy}#=f7$`tv6ES*q6f?Si$nwm2Ui*V_tT}% zpMSB6LyX$l6Rf{<=~DS7nd8UrWzWs@%hD0hMNAsLM*f+*@tisQqO|BX< zFT_mBC_Y6Y9OBl;#l+9nJ>7fJFH0b*DiY%@Pt7!GOlPTC%(zq@shG8?EjkmsG5q zn;T32I>uF&#^lj1jTzTA7rm14F!?ph$Fkg*jrpi3q)(ScJ+%q>Y9{V3s&pQOXL@|5 zUteDrVQy{EQDog9PU{S| z^3mjQ=a;Vhr+d!dcq8Xe01vLxB*ya`kBNHu@(}&W)2BP4l16lP>3?B-Oi?G^`K$vT zk!KDw^k;u%Kh}4wdvTo*<8zmGvNS_I^(m!lCQ(?L*qm9ri@tx>MHK7OafXKJy*{%gLC2l)X5dH`wy_X8p8HN3#CcOk2S}`{hYyF1m{IpdMd}Os>6?{~e3% zHq7s)ZY0moPn*~OXv^Wcwx@~u%z2iOs6N+bcEGad>s}hYME5ktkikCI;BpNi<$WWZy~KwPQ!wa|!#pwp7MVbB7Q{bDcRP9lok@ z3hAB9;I#k#;eOs=p8Mwdk*~~4EEtyrY#vy}e&Ds1hlJC#&BYG`q4NI9r%!*0&1Y`- zb7C*EMa63}A3uJ4;=~C$bps2FK~1byubl(qHZ8tbXlXiv+~aLBRND3H*B^Lzn2xcT zq<&j;(H6m}*Xj3#w&X%36O-5zva;N?nW-_Ks=8SL5A44W{!lxM1agFU9DeWB60Gy_ zyq{VTGgJNPw{PF>94r#>p?Z0Hcln!U?+sbSd^Xu+X^w3zUcs%cmoC~irdx$pF)`>t zVm!wSKYiNcy?DjObLt#5+V(NPSKKrNt7G*`7~gl9#BR9+!-Qq&d@Qb$WBsP;9GF3R zUwSu{w(9O-9jA#9{UqkK4?SF1rF~SjwPR0*DbVQOfBe|+_uuCXBuIMnmYV7jT*u3N zGzZ5#0T=%MuWv5q#ryVdYRG%Y!RuebL&LOBWA+6bm;wJGAh5L0HIq3l!Fs_|Aj{6Y z&8APNX`1|wzdo|^s4`QO*e)22mbT1t8*cr<)`C3q6DJ;M^84Jc`aLMdO={LGXqvZR z(Z-}_2W^$$?Sg4r86TIe!506l!1WI~C3evUhlWl&IAo?Robma1p z&bPcYRV^#8bgOP2T3wcXio)}Q`sPO&GYHy-gj*J3HSAJTU!PHZ$uQ7qGU7jzze85g z+s9`+ZP|(yPfBzJ)BGlpwm*^t@c8|=J%48Qm~1muKtLctR_`QMfIp*WA595~CVR4r z*OQky91a44Y+Sfze#+zxZVz<0aDn@OhCo@p$E4xF#rUMeM4`*Ny8l^+C756C0vmgK z!>-DRyvxUaG4Dm3=3ri;{+CNh&mC)LdlE{OwY5!t&nedkaj#{|%U7?2aZj4k*|WY# z)`EY=p$i~GS#91OlfqWS-Ff-KeWQ);>ynf;_|UfYktaj z>OE|&aZ01ErgoZn$nTY6e(^3%K|uw{l&ard9V2Et82sB3DG6kG?k-)jq^vUHG&c<} zVQ#V}NAAyUZqi>)09N=qVo4KzVGfC;0k|lf_n>}eT(p4o(w;cX?%D6eZvGoCC%cSiI>0GUw zq;5F!eWpWdPXB^1b1$>V3f}cUXd7=N;!oAjf5;(bGZ6HjCmteC{9IF`n$l?Pf7P}& z{_>x()~$neSk9N4{pGP?lVtsXhX1Ueu{)Mm?EcT6dh&ikGy*si&D&1=KRB$o0aDTYBUbRN+S^za7I$AE!Veo$1LS~2&A(@S^R<2OAr?su45C_ zb-S~QV;8NkprHKD%FTbwC2=7({O2t`7TK6nnX*3oUSB$e-uw)jHTEK$Q3x#b0x2hTh{v%_FVOE#)$Q-7?W)@d%yC5ySo>)w>d{A z!@76hpIGO|7}n6}Xtmb0#><-8@t5xZ2^7`Dk^&k^_BZp9KwOt*5e7hb<)?Zbs0#5f9t`6nn>`TRM7y1-5SZoe~ecaHY&|J z=ovVDq^YTi9{l9Vdf=-+AyscF8Py>YJ!}Ur(dh5Ll?}EP{fPmIgmPH!m73{PB;Goq zlc4MV;K74Gn|2A&l;iy|9yYI!eH(kB(Cs zH3H<(usVA0pM6s%AKwlTe{c=%_FvJ#k4Qx6ughP_@b276qg%EYZ~E`RQnsFa@8eAA z+_cA=Ov_HC-?6vlUt$NYn_-^E+%|5Z6_%E&B4_@YVjcl`{GTtp`3W8)$S-g<{?;j` zf@QM=3+uw4Gi}bz!*haeZecM}dnM5KcP7(YK|TdY;IwI(_ha8HrgIOhz#u5J3eb=9 zZr@7^3SO`>wJc5o{r`#R+tKvVl~wQoL+f%`^$@>2m-_tyuh4}LOK3`<53@7<&VP0` z^>6Zsw;0;n#au%C0_us!H*DCzqrwCN{I~GslSOb=%gzv=6LfHnr2h&OE0HHDl^5Qq z4S(9^vC+q8w?-&}^t#`Wi)5Old{X3ly>IGowP*P0zuAYV{vot>TPDJB608*KR@!}atS z_0}fHG&@bjR8&-04z@y@Sd}2l$R0L3OcOw}EscQJCO--|y$E}5cY;3rS-fAb=j$7w zn0;ycN*@IV#(`ya0yCC3W!ZzUX?KV~nq%8@&V_&g$A0_f(_?*&^5N@=a?noY=70DTMBvsm6ZGj(}^0ot9mmijw8yzkUeomHm?-+ieE2 z5N6}x=Jib@cR4XmtTE-sWa%E~*$#WfbMVm}GALM>=B-zhA71_(BNvMb7J zZf1;EOe{UCpTVbc4U~%$qDwbb5JvIV*Wsm8wqF+-(0`VSavoPled6moK*h zpeY*G#Kt5^QmID@3JUxVTzz?GH5b2tK%7{8x+XuiPdZ99t~&Bel(Lf2Rz|K%Hm(n~ z?@N=}-)QNH*q+hAH0h3WbYY70t48O!nFp?}o>UGAJ7rmKzP)KLAD%~SONT6Ca63c9 zUmoijULAe*71q1&CKWH(xS#eJYEKVN37!NNl& zRx^U^#=Lp+kUPKS{U5kEM4rxp$c@*&DwepPcCsyRp9AG~BbIJ1&Rcr|7cz zncnxEmIeiMZo8g0`A_`)d1-?qBUKkYBtju#)!S&Q%bJ_VOOCfJSjm3iBwblcD`tO_ z&0(@cAlT|7YRu(1OqORKmZ!v5P$oM&n;yJ!vxpv&K96eED1yaEtEXaDty&`90jEAlyT`7I;C*_+NcGVEmiFQNJX(DcvA{5IG{tCrk(o>2x`ecmes zuY9iT{1V1Zh^5`^M8`-*^UUXs9~(Fkigk;hzF^gHC_m5#>Ny;xGjrnZYA%x$rc}v6 zl!~CEx}Tp#$O>dB)0I!1dNDUPHya2lWMFbCCO9Za8p`F781EIt3Kw%Z1Ytv(64Qy+ z)E#iG>2Zm!r=|5^QuFL+$_aUS;TRsW&@O~>Qg(y=z4Fnvqb*pp;t>^5TED&6SIqL~ zqXMraIRk@85)FI5zIm@U_JeCqZEieYsxsk<;yPY&-o1NI!6#zNIT7v-IRjcG8P&$~ z3x4BZ+#J!OY#CN=7>05brs-`d(&|V|$*Nn$lh-gkK7L;oJl{{)&>LBuN_+eEt=+fV ztiu4@*;*R=H03-vC4@|y4!U!QtI%7`)5yxs)i*m7oiO$p1R zMr3AcDoV{HvhZt$qfFu%W%vwu1lH>bLocvHz;k<^2}0q@vYX-PZf+}Z`>zF_+eaV} zc6&HazXB`erUB^Kjeol zPoe~DBzt`3_MJOjz;=}!*BZ`8*XKCr5Npi(mAARDq(t~ldL5nse>UyBzL>MDM#tGW znoG=pifsrb+eir*7uBJBpe2? zm5w3+lY}hXNDKNAos^4&4XgF?+&0?t}61X#tQZEk_iS`yiPq} zx(+2H#fxK>U&0i98ckHO{?#!WSMV4m8tpz2)I(%_s+Y39fA`L^_Og#Y!s`h+Ic21^ z%Wu@`f*&?CXd2hXA0=}7DEz)#3`7`7QGTr1IcMwEt#z&Ay(Mkt`Un+v7zP-;ZnTH1 z>leg|7{l1KV=9J2IUPyrG4ZuVyZHH4E?>TE_{7axH=!0Db*g(}Ug_U|{~Z(*^bhTd z&P8A4^4~=;)hnN$?af#(^mwX2kczBb+iT6JsI|3)+`G2~|BqzyqJOuqTf25E6netO zx7?Hx3;OiwldiE$?((3sOKi%O&z?PN*Pk^u{41Yp|HfmtZruu343zAuit@1={`unR z(~Yn`d5AuzZER9eX5+VGhkgs#695@R#)Gr!`hvy7m_{d(t(8ePT$XFPWwGDhi+8?% z|E})9CH@isgSbas>eY)eBUi6p_1ky(L7oR!TyAbIPrr{#4OBEc{`uOaGP&KrG3C|x z>%GKAoIsvlr1p|CN=h&1raPnvb+YdHS{^L2vhU=!J$p_f01u9idRG~ov9e0q#mjp} zUA-=R|NA#@-b{X5?HrDH7om}2vO7zb^BZ&%k_uvpDX8~%cSrHYrXC_{6`f8;)uSwY z27|y8OONJuPb{&K0EDk&!{4>@O!I3RTxv4Bx}~=dhe^%Q2u>nC{OHl6V>)-|hCZxQ zB=7_a`25Drn*w(BSRI|&L7xZr?v=r|HDVr;TnNZLITSe8pd0pf0qZ(;_Gl7}>k{>M z?b=nbzuDmx^l-h>=MrLOEx8NdzZ)DhN6Be_cdFy$aJiSNc0@lg6q~5|CD*!jHZOVO zGL;C(CwfLnpp?$tRUC9GN~rqp@2n)$J~Jyz_|m7p0k91QTHbaw=j0%OwC{~$1*TlM za3Ms@M=)?(+q7SAKOv7J<|A`K4h=RE(eSZfMbYv__9|+lR~d{vt$_^e0yzKiDC)XV zp`oD#HQ(kcD=URMuhykoMLpyY4+YprHsDx&aKOcBBua+>EC>yjQGJ{B9+s7zbmx-N zCgP*z=mHzmHOLp%ZCCgS4FqxH3~FVcpNVeEXe96gQ>e<2($UemefKUegkzXW*OxGl zMXNcZ&%eK8<@-3D^+J-mFiL9jPxt&wu*HK7you1ve*B7#ja5Nm4fxf~7?GXifmfH0 z!8wsNLcsbbrl$POrGN*B`g{4(rH!Rt7{v}?vGQ&29333ib8*Gpy7kw}jeDdI-&t`A z1gy2UL}Od-IxqUgfCCf?vlqa}hj-hy!<#m3f+i$1>$IMQB{DHFv3kb=_|}i4UW9Vh z=4?m#Xm-Suw(<1~EcYNN69|DCSp;h@^&r2@6(JOwsmnZ_zIFqjaIA+|CdP>K6 zCJ9N(pfNc$RSgjNgrZ{HH66e)1d=2t1Rl4*fdMn<4msSsJRD7I;z*!))J}= z4KU(j{==Puf~V8c(srlMRn(Se_wD&PR4V0TDKboSi||vggzSd(M!IV}PGw6khFON8 zhGGa?#iMFa?5X_v*#ZnVGCEoryU8M;9W6EeZMCuHZVE-Z9z0O&DX#BEM4oIDaIWo| zN><}bA8WE7X|O2O2ed@_%`A#*f!=Dch7!MxE39ClJ4Hp$0x_G51&BF}zs9CMdHy^W zfPcrfZ9i@p?=37UvI0v%{26?iKltkxX$+XZU_o1dzW_D@m42wZ+Y$wMc~8TFJg45N zV}bMXA`z>wQc8$o@Kg8u6DV{+YIiFeN8o}_geQ3v7M_5SJkr%T`H=&Y#x0 zmYbV<@3PE*1MI6{pb;q50)yH$tqf~@SR$!{Raz5gTe|cy^oq>POoB3ti{*YyB2p&r+61dwkeFV^w}S zF+L#ypqS0LG5riMAnF7)saJnk{&3%4ExuRYzOoXv$kS6}{WS<75ePeQ0>$&^&!@N2 zWi7wR?@yXsF$k;G*Vk8vs)4m{wG{Ju`ZOHUKNdo3c!9dLb#isARusz6$6!Ve+}tW5 z!dtyxxqqX_S^H}1zW(^&;3Hg)!;h6tpN5(p8XQcO50umb(27lxM+KckiA`Md`_sE) z)oWl!8i>hpNqq!0Q{CA`4P~NRx@52AlAjtrPjPOkqR(Lvlx-WzElO^=ruDA=-PE=HsL>v4PKtx$h?Q8gq zIQyyqQH#){p37h`MBYTHL=}Et8K)yP0;%uUuV3=3Y6$;#o^x^>YrF=K=40L0aF$n% zA2A#JBngp+rWB{0C1g;!!+qo4L-+6BN6^CY%b|3=DzyFdz zi}`dvJK-z|V*!sK)(+Y0#fulqR<1mwruOugDUb(3*!uPBgTYM^xHfz4OGof6`ewFK zXV(*N??`z1$B!S8IZ#@A++&^b%M=J@uW#+`WB2Xs?Rj_XIF9{XwtBS>id@RdPs*#Q zs37j$yEm_uCp@sIjg*%wzI=hg848fBfb7rOz=JCg_{xL+z4_$Pqaf`pyPYB;$xQ+a zEG#U9^(%stcX1Kxw2E}8Y3vXJJC1`W9J9X#8uu3@7Y33hFuCRWVuO+cna1@Z#`Vea z7VX95ug|~l>Zu3vDSvr%5`6=PF;eEO_a3r~z63JTiOU*Q4>;XbA0HY@AV{La#GraN zW|jy8jyaofjG!kT#pey*e;mKkKSqe3+YR%2B%WlG#>lrf7aJmbjEss|rfI6Dnw<|& zPl%6Cyzp^ZjM|$+0K*2AmoCSg^`E+yW;>)yplVWVJrrdT)28UVoOZ_%Vv_ob(z33V z_P@8Ez54-}2BkbTu+EjE*DY&)0qy=I$u~)H>{9}PjG*PGv+AcNCiIb4>k_UYU=(g# zxU}U4XcDBt4ICWC4*yz8jf!jhrB}noShvS7G{V=CQ?We)CBdoH@?2U zq$B~^fJ(!rvI*%uTC|Ei9NBB6uTf*onl)OZfA$~GZx)%0Sq%6n6qV|Sho{RB=5xWVNx za9t%@?L}?E71;03uBfa z=y|t0eT^9@DiC*x5(7*y*>ZFB`n78-&Kcuzo>VQTi8j_OE`h9JdZ?W5OHglKz*)p) zlFpwy&)R08km0xI{Gprc{?W215w1%y4MhDo=+vuYTEl7-vn^29x|*^w(y#S5h1ZJ* zC_uvcg!nKoa%%lGyLvqDmW zt|Z9w3YO+iFvH^zIPLXc#Rg9eYcco`M90jP1VW ztxD7xVPRpQW02>QsL>%q^VePYAVGITX0VC{4^&ApX>{vQMA{}JTT_U^c9CIj+lI>U zKVe^X;y+W<{#C!)4#~^A!&2PxUnM3gBUuq5MZCbF0I?sA0EW7`N__v8E&uvxn&0Qo z8df5y{RSr_N%nw!E%FKq)v#U@%~1$LXoor-qnW0J%t0D-oPN3LY&5-*1!RqSK#Y3Q z34)*;CoN-6QQ#nULnZrBu@5Dvy!z7#?11~`{R-Gmgvm%!{v@f{So^d^1L`VP)mSjX zAN>TcJmhp7J{1xi{1ZhCz4EKTBf|yT1CP@Q^#^F%RD-b5NyJ64?iZkw`XInbeX4+n zV5GdAMvv@F>X*2Fe2Hj+9NYM$u3_!kwZr2-JZdJjh~Y#=M3jRryu8+AE0dsG5}D*a zwP5PBjG;-EO@r}ls|5@NBANA4t)e8{D0VKX;1klx&WXzgf`r_su% zCx*JS_>rwhg;7BLicHO#_Zxx|64YZjuZ~JX{6pVE7`l&WV)p7|kTMf5=C48}T2|&E zdqjdfHkW9yHy5u?zLv(}=i%W&=w)i^0ZcIh*evhfnQM<~D<9XwTH@<_cLA@C{NUWF zt*IFeMdKgkd$d-V-pZ~Cwq_aJ$q$)6#my-O1D1~2x^MF zwsKKrRaF&AXQvbt6g(f+`i@Ud5=z9j_rfjob40+MlFkNo_x8pN3=D|Wj@GZ-uNl=@wRnF4v%@Sycl+xN}OV z12xvbe6&r=&^~p*5jqP2l^c3`8#7b^s8H+FxEWTGq1Z1cFOSS=-0P?ApHbQmkwJrf zAYJB_8<$kJic1#@cs{9)tEXE=fb70lx{jx^b`(v5MkCL2cgK_%!vaQTV@WGu)?guT zUAAKe*urLqSfSADJhzwGAQs(P&L20Z?%?J=8ippZJuOxq@%2b93q8)K)Nfv~6D%*G zZyUuXl35Kc_}O?C1iQCBrTd46R5h}>Tx{$`v^kfd3L<3Pqee&(AiU{_KqXo`fU=u^ zKK{|$dEr$ZwgY$yEj(LZ9aUDHQwqX1rX$T-&5YFWZrSppvFT#!_5#7>{UHbaYfHH- zO*$4`mDXqxt^SqRI%6I-ZCF{B65cnm^7a+rAG&hK@VmozSJCK1Y{`K>$a-)Tb*cK0 z&~_53ph3Ps2`bVoo zYhx2|=!k3Lh=)Xv1y?4h32WflZQ-11ky4iWXmWm+MM_L`w2_&{m~r~#T^dDi8Av=> z>*j4XUr-W3-TRni<$DR4w|BXmb|86=kB_TM^%dknUQ}m)n5xMS0<>wHO{3?_s{J== zW-`oE{`xgGrh4w&xtOK(4X>=Bk$;FK0@PN zSp}-mhDE?OdIJp+AYv*cR{(Yic0ZYCdbq)~!r$*6>-zP_BgN0FIPzON8dg=Z^QuIb zk#hO+rqGCynlJzjI{#P0YR7;^lenZN$Fvs%6~%Q8{6a%k;P>eGvwUkda;PE9P^8^I z@s6aWNc;P6`#ILlvr<-F$}#*Ze#O^`vm9itKR*KgVs z0sj<}uxae~=<0sk-M7FfrCy`gb-)6B<-B}+#p74UT2?Muvd77k#r302&*$E^-A+a* zPWjDT%gfIXQ(4_%F?B%Z$dMgew;lpD&a?Q?$odtP8k?Rjdr8{dnD)2nj?%O)`_E}L z8B?>>m6d47O|KrxpJ@=ySsXWYRn(&04Tg@YfkuCS3@8~Y;uW!48E#e}n1ndFac0Jd z8q|L?dphWp*I1PH`t|EZ4|nf3>M_TP{=WPEyhAoc-4*Ru@7})${6C|mWo#|_^Ae$< zayL8F@_(480TMMt8)qoz)@$}H9*YQjUJ>G>ZD1~3;t9e{6l_whSf_aK1=O}(=xHBs z0WTB2T=+yq)LZ=8QHNIntm5>}E{1)ma zkWN0pP0&eIWaj!-p}8TDJJO?87k}9mm_ooV^stu{QFAyrVi2`%-?}9<9%2k|a#aLPNcTY)>Vor+HW4)ffDmnje0YK>HZ(AB`={m5wp#+1KvW5JfQGZVD)dEK1J|>LQe#T$Q?&F1 zh*=db>1)_F0Hop`=HEKT82uQA+)y1(Rx!*x$2fy zRbVNQ*qVE5wh0-1-LrDjzTlG22YylE6MCjlD^GUVK}ElDV?KF|%_?xyjirIW%({US z-yw*;e!b`7t;CFnEiLB=Dc!TDY~#wF-dn-CGlE2TIRh$gWKr?k} z)kCYkA6DSxHJh}x&z_BdGJgU(F1@SK`bVC7I;a{9rP07DDN!V4;c}LraHYth7h6DE z>F5c#C$=r6As-DAsV!zxTMpkLx+^3l!;vW`{;{2T04OF27bydBz@?{#p)E z^CM_vI*m9N#Z@F>UgKCe_VL)?%F4=mXmPgUxwm53G9~=90ZJl34fbg#I!^$j^Sa-l z85ULw1{MP;oI}b<6Nsq%{<`h_HpQSX7oW6zdc4iwu~1nRy;K_z?I6!D+CCQ@9!_*x zL`aP83<_&qU0v!D7Qmp$!Ew+A$nh4`mf_aM+Bs;%d#7{FxKD)z4hBm=F$Y#8Xs~?c zO0S$8sd*RC%E|)A+sV(bz|>=lV@Wf#RLA0MfZP)_+9|-kH~jznBTf9x%}p6P#Pswu zM92_7KfgsC#n_kuw2`>C;OUDwZKR#2>YIlNqeoSaAgW(4ULwJTDF#*p(?GH*vy+K? zS|Xt|1XQGQ=+F(oJ#>%1M0JheXJk8*q>J$6pLaH!ltEk>@o%0!28^nN>KZk`c*Cz^ z!BWyA?4xCUHh?Aw=wVeVF{U>=JwaM1Vu`3uo(i!WQYkc@a8ijy`)!F>NFsMq8Y!=v zX8yW^GLYoLMT;VFC!*XFu1Twao}+rj-)Ux&H0cq^9NqMFE%!LCwfOn^Mtu12fu}hq z&s;HdKje53@kNxNEC#FZFX{$Zn-li-@Js3q+^^TtFs1{s| z&sqNKmh~>|5$oaH`r~Mqr8k_>YRnT~yT6-d5##@z?v4p1)<=?Qi1(wyto znNue!OcX~y8e%G8m5HcQ+j)>HL0+pfdLe^Eqq^ldSg;)>QI)KewAsmSd5~aEyPsQT zM&g{SIs%=wJJLzx0ulKVX6>7&o)aM$zNh#0#=>Jd_2B6CO=Fg1Q<1n&g(m7oMDmy(W?N)oLQTk+^H*+7xiHaC#z z)vuePmI7=Q4D{LwS0?fd`V;xZ!VbfShtBg_=03GcF(vKcnpf7Xojhc{QHD8QQjy5_4l+a9%5Nm~|B!j{yJ0IM?uV3X- zgCliBVnZS#dKs|_P)t0x=z5@8K4s)eHz>AlJIZ0uPaSVK~WE%M&|g@B$ldL4le+0lcE zULQlOF(|iz=V?V=Ufv|rX33RoLgDgL)AQZEJU#zZwdznU#`;utt=pZhn2^&I8ATGL zRcD72h60Shb%~)s*oiLA+Q#j8aoX8YQx+}sRpruq^+TjQqvKa z8mP87#|BBJ0@9y`Rden~WmQ!}PXm~tZC_m<+>HbH}q0~2g>t>Ar1gw1FH(( zI1UPhvIwvVTb2fN*(Z-5zkmz<{8|1g@7@GHaYUmDImlH7ZrNy6qc!<fqz6Y0SRWcXJ z&^WU8+-Xw(9eW~tIs|>4SrhI4P--JB9Dj@Tq)QU$Ex`qT6WSpJlA}SX6B4huK7%4==K?Y0gaYbH~wI zmHK{k&P8N9P9fe!RE`eCx{a@EvcCoLa!PdF{np)atJ< z-<+S(U#oQ@TGQ*VsSOuR!@p{L;9Ox4yUf7k9#8m|3$b5Mbx{@w=18TAs=6&SYw;Hx zI%s$~ZR}dRL$mnn8_%za*uVGj71CEiZS@(ew$1tL-KbK@1E?i5TGwCpI-_#-yC@MI z;o@P?7$c&hieU6!2qig*5+BgO>nZs8z&7BwPmrsgK{@98B`2$N`0!yI51Q1pZSz`g zl>d5Ww7%of;Z-ywo8@vi=pubuY2c6`8X6v>`2Mn>UmO^LCuh6EV<@eBfL+?>oJ%36 zY{yfaU)VQSS67GXtJ9PM1*ZRd?ZsXwE7 z!Lp-DRok|W-Ff=gM0HB`nRENFA>M!8jh^TFaf_kD<2!fmkk+$Jk`6YV>d;~>pwnE* z=eTBMt?`eLh)6r!`?GM3rY9*pSZ1NKTJ(pTV0@dm2Fp{E@@Z5s3k0AAIt;vB9AcQ! zo2;=)yyrEK6nn)M!?sfVE&88-`Se$vy+||vq&}3AdN*1)PO+BqBx!PT@|j&%#PTN@ z8b)u9#-?{g_TVUFNo<3Bd)(80gl6RVE*xZ#47pj~1e008eqax)rrE!MSf|Yj<3`I4 zuWoz&eig4*MiF0v+@h6iPtd`ZTu?-g;rV&AkI^=}#4GFQ1d;{>TO8s%Q8U`?JogM% zcJ}=34a?+`&FYek{co<8JV0Po_1P-{6(jW~8M_yASe}H6;Jp56pR2HtP^;A3kd#H< zrYlcWY%RO1caw{=hXVT7Kf{qA7jF5t%X4e0%gP@8dvj46R!m{vT%5q{ch;Ow5ayft zRC}F#aZXVjDl2UiWj&=)qz{S3>}JwVV;RjGo|*g$m{+(n>=}KtN$LyBUh(D$1%7n2 zhM3!s2Ca|0af*!;A#rzi_f2A!dgxFrpbix^eo+wLjaI{pZwpv-tgNh7iTwJg{8sb-&8z)vSpln9IqaPzVUJ8b5StCIt$ z4OjCyLMCY21D@a{M{@2F`7p>=n?#J%lAG+hE0O?l|H0pLFEm8tp6=O#)-|(&^6aTU zy#O#WFY3_pPMzuRtfB+C3yMrWqE&*T7o>9+2qqe|TLcr@kxwA1xI>ELAV1q*&UP2A*d^d4fVvSG9e_ePvHK6Xfbg7=7<2{ ziQep82+89q7i-7@`8`(MRiRd;yRDrC07Eg^^LQesy(q#eTxPsMJF2;|e5J4h*H) z@*I@|bFzS@AcaRAWXT@Q!yYYfqs#Ia01SM? z3%vTH!j}suWezY4LBMt1lGWlpd!poE5J8(x4a^_7jXv?`=;hu)t3W%Z5!Pzqb^nLtWv+$(i43weD*~?xx)?qZ1Kp8x6K9AhfN9g_Kp%HTc ziT-`tt;9jh0OUhp?X-#<=ecN{x?qkp`yO_RPQpIv{-p;eCyRg{UU^GiS>Pf7Qw(Z_ zJ?(NwA4(vY!TP79xqf^X?ntK+HhRpR&5NoJanWh~Jr@ge8q7~sbsVkq51z|)Ni=QF z&Ltr?_ua{X`H%1{$Lm!Y5WoZukmHlD_pHyPv|=lElZTj&kUe^`bfJJW@BZ`eZXHB- zJB`6(g0Xzd;V$!idl4)}A^G6u3-*EG*wL@~jb@lxRiI{RujH=2~T>CQmV!;Alw8 zLcI`^^v+Ncif9#aIyvY}d&3;bOEQu}D1)Ex%bA;{(ol9~jwFKaYqlK=j*aD~RiGLa4eY_( zD;A#{tU()mMJ;qux<2LC5A?Axl7DC%MSv)c5j<|L6xqSpk^w6+itET%?V47g!D^jk1I;a9*JREq5Qx2^c{gEkcxJcjN!l#^51$^V%&QBmT9}N zn1HrfV`k{DgBv z7C1J*O#?$8M2mbX(@^Na*rR{3*N>T2geYv;*O2yKA8tWEzb`<3f>4SbIf0@sN!IZy z0_-J*A)r0zGRP|*)6s&-uDSr(HrRE?giIhYo;gQ?(m8#^vP8;iscAE`fzMEB5ccA~ z-Pz1^XEip=?s6=`?|+$3HEil+J0V3RE7RhUz;#`BCo|3Gbt3$&ken~nWFyV~3YSBq z>WCjEmCdI-c>~zbjfM&(dN4W~YXK>k0gmqs3}oMEXVx_mYoKH@El$=S8!3mwpgR>M zCoVApRP$8R*l*u8Cf^Z;R4gPUM5Un_#UD5$nYlt0p)Ct{92V3q{)CL-!|*d?9p)F; zV=qtMAzoXI&`yhob{z%|!Su5SLHME|LdB-N7mn!9;`JoqjQAz}n5+bYd0?;URAd`% zk+0`&GL7me7V(KMFnWwo6Rqn zdVoY#eZZ9rR)&%4@sm5Ykah`@Ca7o8bnpd9i#hm`>saL#4OpmJD>0k!lJK!PW= z-Z(|eTg$rk_j7PNy{c+Wi;{|Qgkk&ZG)WyaGT|^%Q$;2?isxQeC69`;>1sdi^icCEJg@U7l!AlGQ?9ICJ6$;%U`4a^^I5{`ySmsN`yO zL$AD&n3fo@VwB__QzK$yty77~515h*s?n*Hh!(hl=^H6fv09GIbpX&h5~rdP1kx#u zj$(?7Gj=vcom8uWk{}nhgO%+#P(Ei12qF6;z`8CXmx{}zEs!eIv)WVT-xJ!6)-WEL zH{O_F^yVWAbF=-nVOm|n`8URJaEjlZ9Onqd4l+0V_qSCBL_ltlh<25r6{O?$Cir62 zMkKWca6GtsA&+lA+LsLR>XWBbf>sE^{k^(gY5&tedI3!08 zMs2*-AS%YYYY4H(We}b>t`o>_13@ZBIZy(WgQQ4W1qz}LKQ=e3MV69ReiVA%U~O@B z9P)4GbN8`+JN5OVhgTjN70yPM5+qcJl-%MBOWloT7`!nFJ#VtTX^;m>oo8yCxoHN5 zh6&Il_>-=XLo;n~+<_eOJMb?a8jrl+6Cyu*b)Y^VHZn3oqZ=pJ@_>2Cwx!{6X|l8q zg=oLkU=xTjX5)lbQjrRkN4CwEd|ZS$HU31AHIwKpKI21>=~e2Uzkongi%s0@)8y5%hXhq(&@nB8`PYeb_wDCUQ(sPYV$<@D;ifi6ld(q8O*B zrKRQBFoqMW1?apFq@D$_Oy~^t7b$QgYh6tv_ai4ou+Y>3G&1@baW&!P8T*FXOJwPU zA24T}&4OXEeg%|}{Q+(kLF=197qxd@y=ji4XTv)3@GLxBCkUe=hk-2vaR5sa<&WvQ zF@jmu_ml6dkq{0wl3IYi@qIW%9%q$P8ru{TJuI__?8As2Drm10w3{dhamR8<})#JvVLp*P15!Cr_t=cd(;sTB`FjAK{WaqRV`s)vc8(n#N3$| z8YiFCp9aqt775iU?zw%tE}qqTc|JNKVaI9mIt}8{mL^x1dX93po+4 z-z7m@duFUT@Y3uz0|FF%ju2+^`$AL4ZjDX|2lPug|6vQ%oHGbeYM}_C`u*fJ zE4LqEJ$jDV-{mjT)5VBXMh^zOA}HI(X$Db9&|!R_!h!hO%L!b|u&dXPz^+x{Pc?HI~eW$aq3h z;FLax#K^FPL16f90L6s~WslYw3R-nlF4Q(PGAc&-Oi5K0Zy~~btI>-Z!W^^&2JPCS z(LAXlQm6NdT0mqk6y}4GJclO%aRm`YsWf9_n-#cVZQb^2UtMYl2LK^DqXq`-D zw7lhK8c)qe@DEJ)NRw1HMo3l^+C z_=pHj*W+q2|9EuYvaTU-7t=dxR8D!lF?%elzwO?(hsSY3sUzE(?AfDbQuufvD^aAEFqJ4jkFDWiT#37;WI&GfIQ2<@|#S7qj3k0nWh?_`^jU7(% z5Mddq1F?h@ZWgWB_y~!5KaiJJq5eFVkMJ&h@j8L7@LU1Z+uG;XsA^vQXq5Wx4#~3n z5P>prq}vuRbMP!~&OVQq@=d(wntepXb%HhK72x^I+nh)7uyX~6m=I>iuT>8Lmb@f} z>TeWJLkd1ovl)pFX{Z%wqXZ4J&-3!~P)raP?F{xbM_>p8-8ky7oRWJel8*#KihKtW zN}=;~zVnm9M&vHtudmNfplc!k{ZRZT`*=tQLwtLT2)v(&SPTg23kZ#~XLL=ZIejA| z&AR4Al-G|KfnxUuFyLo6GaM?^Y+Y|`0lKJLv^zyJd~EDz`8J#-wE$n$NVkTj-hrTZ za&W%1AaiBs5zY@CHbkfyYPfm%QPvRDC|oTAi4Mkj@PO1*ZRv>`BHn;V37^oW=-pvF z@p2!)aPi9h4JZBPknBrH1x#6mze$FO?d-O#k40z4f5_JKffW*}r?Y<-rR)Rqn?2BK zN^k_|#L$9O2mV1Eo=(ZKX3d|snElbKhhmXJIbisMsMua&wq(Kd zf+_`^3gOv(&2yq&_?cw|jigCB72OgAsCjuo6Bh2YptyX55=o)#5JWwR)cN+IpORXD zqc(xWj6eruTW#^bqDZ!1Vl+}vg=!;d>BWol_Ybb7sM$cSq`2&>fB=JEw_;o5=tQ$- z)v`Q=YDCJBK=fb~>Ehf>NCzS5pw2aG9eF4NRm{a{#jEIwNd<^XG;03Lfp^ehY$Fb4 z;L!3WG{wz73zx}Je!i08009Q>rSj5b* zDaydobIj#jM3z(2;n&BI39^6yb}aNZU}wJGd~uxY(!{5?TPYOAkV*&w@=5qC#>=1$ z{ixT8LVQ|YbXr5hA8iOetE>nEM^l1^8Ye}sg!!efcm>-&y|N%V-+kP)cyP%M?5M)! z%Wt?kNO=J-0Lc`}`T5dy9|+F;b@LcF3H};NY8D_-g5+$sCzUrQQHekO=SxWWEYFOx z!G=>47Y`695EYvq3GJn95)csD+zS9cVyRMLk`k3d%N%~&gd>v80{yNmmO7!kO_1Q z`khVC-SBh?MV*zg_JT?%s{)-%;fRtEJs2%AbHBLe9+QH{!tGNR2zZEaa3V&CLzF~@ zQ(w*e-YEbcQtD!`#fi2)Vt*XgUthFb(uoC5LkH6?j%jNEmNvXzcE^DYl*CmGyV7le zdQmhto8oUVrqc+zYaMQ?(!Ckr>DspO$( z`_oMHKG5g{62S~RaYo_ITbA<2GMv5)6peDA`$6C~2njCUiVW|2gjTJNl$;r3Qe1{z za??~0NQspY(Dt^I1qFO&l@Z?Ig3i?K>f3!SG8Ln9#}jVLEIIQQ&}cAG0W z9Ll$+wgxu{BqA%Vy?>_z3I`iDW0?++Wy)0&f-OEn^+~ih@yj5x3e`8KxetPcA9jBi zv5mJ&bFu?rDiJX;@^qXqcHYs;yMiJl!TkCj^fPIKUerQq9I8C6=oK3WGfu-RCe15P z$)FJ>GSx*s94F{RK~GK$q1JW~Cccmd^Ohy^e*G8Gmob?V;v_`-RJKeutl{7O%} zeQyZNuUekSqi18Z2t|besx91*q$yUi)r@SHyLR8z?;oaN%%w+JwVvP*9F=4YQW$9f z+ae=74WegO$o&fY4*F)U?}Xsp4%{!WXALD$j>^4F~? z4d$|BN1(v4JT>e53MQTa0>wikPDD^Hagf2pVy+DUm+LR#w~QQvf={ev3|^&4(o;MN zu%x(~Ybo2jbo;YMU`AJPp)KA8L#l4NIK(6%#zINlsDjt%P=p>sx!tqYrJB78UJ|PN+cK zx$u625gf$2g^j|ug)!NVxPu>OlQ_X9-8@ zqTktJKo}fo8*v)14CqXZUJYcX*DaSyiV)(x@faRo=axDU%rhu`=xt&Q%ryc;2wjR! z`Wko0Gl+dxt6g-za8vj+*SndmKCQgT?<2&`@#buk*TF%9&~nGRwFzbZ6Xn z4OrskPmi|f>**b*_dq78MeTpN!Wm9U`<6Pwzz99hJwiFz#2o)m@+vHj$Yda(j$7*# zb{sfmTWge15)W~gn^83loSO&)P~b!SctT41l-#Z3cngp>+7|XOUVaJXOB`Tq=Yca5 zBwaHJK8{BY;SDjwhKVADfo&mfz$iA0NT93X3mQ6aoc0PMi-uIZyCe=JKIpMoWNgnF zV_V({+3_=8bVsK~MaQ3T8chh?@GLwNhi#+?W}>$OBz;DCLKWMwLyEEOw~4nlkctO7 zd1wDmdv6|AbN>JTA7dSvVP+`IVsLDwvV|?Xa>S3?znLp0}6 z!at1h4w!<7j8dtS(~ftgcRjE`eWWUnVmbjLoveu}*1o(;&^Ms4zsOoUYff2_OA(~9 zxkbj(Up+D=PNl}CR+hw-T13WuOag*6Ev2ePdGs5Bs8HS6sVucA(ZDVPCngJvRu)-k zSeUtb2#2*y*J}C1tTgvLX!fx~$sIaN91IB8GOup#PzZ8Lze>&UA6r0P6086(}3!n>= zK~BSdbyFsFC9IA^r;E=#M+DqCXPb8=u$a$YIX=M8^p!5i+)kCTPrPSZLsw8G1mg6Z3i|nCu@F8`I z;WJ>9$3sd}`Zf;cJs^kjo+A$Nm%*TdWl!y=ENy8&7&-WJclzfjRrIhOkcb{x6jQnF zx>&K+$h+#RBqAy`lCpcPw@!+as)roSJQH=fp690Te%^zURzEqMf*bn2&`hzKtYDjx zVpcGNW6r*jCc z{po`ze9uxOymV#L)75+BbK=mzC9}2-tNGfZT^}V;koDDX8_am~@g=B*uM)Q7jP~u@ ztCcc$fd4HQcS;z;LJ~^Z0&qOz>QCvVWNF=0Qm%(O8U%CvAxFhou*;L2wVx+82{hA8 zt5Z*T-XAsiAVW0aPiI?O(54g?+frhKygkzv-R2L}%G&DO+Cw6wXQA#%@GQ$BjS1Nk z0L~e~7j=1;CCnqR#>ub<>4y;B2lFzSL}wpuvO~j9U9fZZ@1^KxBO6U%=4v=_npvY_ zyEyl6d)L+nH*c^am?itrCDvUbDG|jiL+FDB@g@YQRqxz%hNKtAVm|2&6qS zbXG&tU7_Gvnvj*58JYv+wDj(7HmArS0pcPTY=GwQZ$C!dLtDMNXu0J*N&QkP;%cP24$tV{r- zmV)N8yq$l{rS3Wo)RngJv_^kDuSI;sX+s<9TZ9_~sE+Qx*zJRuySGOBIst~7pji9h z=2Y5L(`Ut}YKFvWwFq107Y~5-vW5;0gMmLd{=wI*Dl7D-uG>W=KrTv9ujFKgxdXrF zV46cRIlJiIl8H;eK?$WmwsomZ6Ak6VFY8g=N>VQ{)|JB8`^5Iop15*mZfC_Bisr=o znGSH9(pEz6So*t-mOMa)>LkTlzpAH%^l=2Sp0F`XslMz@dM+F7@zGn82+0nWta&wf z?C(a3o9BL$- z5_Y5cZ#0*4uaOiIr}I>%dSqHr0YOjCV(`23;|}yc^=yvf{2ZAuius+SCxb%4VICLN zB|sL5C1iu@)-W&#sF%$2Cp&IvDY|H)cB4|$1fIWE%U}%CMZWH@y0803AYa*N{QXOZ zEguF)>k=CUXO-#Lx`FI6gMq!vhHD+Loxn>fhn!6HZ$G_WczwosJ4Tw9(k!DCewzH& zxo&71{mQ|}1=!c+!Ki!oB^qBlxxu|jnmu&Jx2)Ba{qqj*2Y2bJ671Zuy3l4n5rSOe zw{@SgKH53Gs}OFjA%H7y!$j7QHd!BXvyU|~!i*laLegUy$<{MgEftM+s2G7;$?zL% zHoPm7&a%zR)UzOdP%8}+zRr>iA3{NE9`#ZUBI6r0pND$4{MC^oGQV#B!8VH06@02Q zjQ?)xx{QBMCw8UV#6{+%GQBSm0{X`(diK%$vh}m3!@dr;0Afq45|~Na`STjj*r~bY zz~9MH9JdWB3`wh!xlT;%Gy%?*W;rUFyVC$vz?N)yhoJoFbc!329RC@r=&qxA@ZGZs z><-V~W_+@r56k7FoEzX2dcE2x@=Y}}_0AQ5p@TB9En>M0 zm*nztkjZLWcKlgTW?KfDgZ%xa)wW&_UCr2#Q)QEmUaAb7Y-To3q%XI zMF*21!WDS)NGfaBzaQLFNb3^4jN=oJ@rnDmX=H{j-im6YsYTP5y$ni>(HpPrR2$-e zGs@wbXPo%Lq6%=J;?oD(IvSnbMcP0ubjSgZ7mpM=vc#u%=I(4RC-8{A>CldNm`<{o z$!VUR_)<(&eb9>HgZ%4MF|Tn7q$&DlHQ@gXAPYQrC$d5y@j%>{4)go~yl+F#81Ia_m_?M$8JxcfrvF<$^*e zX(8wKNp=c38P9@|o%E{w>=>Zq+qlQd14o${{WYhf5Yt(CmpGqG9c#-R_xGMX`obfZ z@V89YpgW(73JX8Nhj*e;UZ#i?M}Jv-rGeZ^!l!oN4~vA$LU%()+4iibEsS1(2!k-C z#s)_lyc?$#PEttqHC0Ff!g!Vak4WUw0`TL90GU!w?|*w)jqn@kK}n{yt1pFMmaq_F z%}G+v0xObHzZjcaOXyf2@ByL;ANlztd%MsA(kT3;MLi4_L3Ndyt;f5#+N65*M8G6D z;RF=oC|8rh_7Z}m)V~Z<&hfEbREyYH#6_obYGQFTKdA zD_2Lcw&tBz?$>_@$N-Z%6P9)oNRVoj9O4uJgQr%1mk4GH@xeqd;`9hKnM|>)ikztSUG28l4zmv zunAZsRIQHUaQ-9|7jO`fT_6>LfIXMuXwbWkwMLOqQHIr}=Z5v88MphB%P;`|cl4`U zNh6&Qq0e1wt~W=HIpO$Au)r(C@j;hke75v^P=o9XWGlEM(2`;G`!|k>10)uJtn0A& z(X8_H)9osx9Nix^(hgH2_ zL-RrHqf}LaPwuOO%Pbt|I%yo42<+&|Hj_qPeUPgFg7XyY!Djh(7>}L#O6uQlht^kPo#C;}-uJzs`}*jf2rGxYk)#icfJRf${;? z^(%sd+dTELt381x@>w3gy#eyEfN>@KXifd2_w0Y?zy>=~S;m93d$!DOnXn}#1tezN zKXAzlsxLe~shHQMK0UPQ53AgF9TMlMZy(;V{MXYWxAj5?>D|N`TbbqptU`lZ&sQv|MBdd9ma{nRu=T@ zICy_vp0Xs%BY##w#F72i56&qrzxzs*J6+OF{8b)R5*+jA*pQz^Wdw~_n%gw}#av%X z4r*&%5=Othi|E2kH~L+9u$@bD37qMzx?A|8R__)g*#Y}*M+$0Q^J-<6E*a^nB!C&juOT8_#eag-i-SXJWU{KN z)<`F?bVIIfto2DcWz0tJ#l4|619a>--Ty9F$i%X%q0o@O{4!)P1Jtc7ML?jY=1U^_ zY$0|oVIf>;k?1_|l#sVrtWxK^d%$9gB4&3%qdyjKHA7DQcYmmX0j)%$owz_mi30|> z8N3-WX7Iebzx+C6rDeV9uicy{h%^>~CSoI*#uClEv}_)o{I`3xkVZwW*X*z#T~_9? z@83+Gigcbat(s$)&Nisd#W_SJ(3vYTB*+cgz#m7gYSP+LSvMf9ZFp`VP3Po-KhY$EhVXRBt-n}OxMag#1XPSQ7jYoZb6(}-_r~uvSD7_RU zias6MJ@hrBfpO=)D0Kf4{rcz7*RI!}{d~%H$IP|`n@3&mm9o>Xan8`Zf&w~#$BJN8 zO3D-(VMV+yAi&P0)vK~qcYc@ld$WQ+Mvo5TTy@DleH2BsKu8VYAm#MQPDGr5{JeFW zHdb#d9WsV&oO$cr@&mPhh^izCco`W}=))DQ9Oly?bUyAL9_s3GX-}^fBk7qnM-qx> z;y5RvWaJNyuTeu9x~%nL+@R<+4n}hciL~xYWuE%n&)aVvN5%!y8eOzB5l<{|9mHoM zAI3*meRr{<)pNqQ->_5Mq`5=#H#|Xj_6m=Me87TM~?1D zUCAg%FMnqzqMbc@^w?l>bjjr>B8o0ofcU(2`;P5VQ{!MsDk^h9V^3^1$mOi^jqJAw z9qZP41RXryW=|A3G_u6=Na2af;-pGrLJ=a8ZmLyMBRSNasdduKPurP2OxiMy2$4r@ z*FznWsk2L|?+_PbnbU03b&+gBSuF0BNlOgmRQa(nk6@%9Bf1Fq((s#a#G^(!vkwe# zdjK%uPQKduaxv+{qT;CYzF&ZyxWaj7N?TmNhqP!?#80O0@&>0iZQS^N=hTx(a!xnU z$;~@`bgg3|f(m^qr@ctaybyK%STSOtE}3EecF`_Q9zDFDFTd=+d-sWd{u#AQITvjK zRIkLifKhr>HK_8&n&)YpRknpD%@2>%YGQX|LuTXPHdWTHOOmw*#6LS9kn!X3V+Q%X zTYp*OKhG<-w7Bwg;8~PrQhd|DG%&?R1O+_TjWt-j6>R#Mxml%n`NCRD$H#W!;@Far z-#J!W1gF0p{?7H4<@=iho@9mF#lG*j(QS3Y;cRxCbhZqC^h)W|@7rB}VDkqrKeVpQ z*7EK?Y0ui~b7KPx+YH@UgRv7c(&@M3qJM=pDVtG2Z-mkA6Ud(8ohqG2_p+-?9wklh zzRUMwlb?xtcRJJDO)8egvlbPLSw1r%m`nz~+Xlx`Yr{5BnRrH(sbv2mtJV!?FqMeZ zZ24|wwf(Nd@`|cr6ywqwlps~oMaB7J@Og3|MI&@VMsWa&b3-QenNCH^Sx_T}{^?cZ zKkw5ALQ|`^pRna~CX^6%=3fomNXwO>*%Qi|qgL#@$H3n1ZCrM?EsAH6DT5khEJE|b zdatkf`}j~_FBX08ij3MUpOY)7|P-^f0Uz@mr%F`d%i&S0#x-WdwPNxzC{RU+!qqxJiW!$lBT&y1)Xg3P6I|EX@~lhbTkt|=|LO=-UJ71M&E4+a@%*O^asCT`o%&N7onq}4p1aiQ z2!k`NF6Nx#PD@`bO}L+!u=oiEZT0@qj*?+3%o4Zl4G7Q;U3280DW7%) zj6L;c(i$KYyGu)*LX*F~+;tLp1C?&sPbRr>eygj#7TcV!^>Z8O?U_(KJxvukq}#CE z&*#pa8*3j4X+3}x>`W<-s!!j($&Yn>AKk`>gC=r>g1sx>f1jzx&%bRC&S=?aVqYHf z{^PGazpMDeDdER&v;Umg_T$PSq6;52sKkBHo!{aT9Ab?&%xudB-mlW5BEh=Y^?s`k z9UNU&Z~WM-MGMk}-6Hga^d))Xwf!0xpm}JkG)@N zEtZ_pGoO~L^VMnIJe|Tfg_Xmmt5E?!S0(PBKV3OmTpq_PIa-sF5LL^mpzVDB+3MQ2 z;Wx9!KFT_FH7n=QfIB+26*WE;qidQpYP2st+;GwTXRhhbT=RdgzG$|tq$GN+etS-V zaE=C)Ds)zDozhcFUK*{ z9avsVFF&!Wv1*q4Ik}?ZnrykBi{fpUF70zvVY#XN?&U`7KeyrPY3^TEwcr*CUkwaz z-WwKH;bs^Zdjd5f+P^iYIDxiIvreu-jho_InynR4ZV*?X<=MQ+wFHNw3D1I`Z9g&M zcsH+^HYgmoTVGzY@%@Go$n9;Y@Ut%PXvesGvekcD@`^u=9Fw*2hn=!-=-{~h5x)&10WvGtYt$*alim6G9aWDmuZWUny|;Zd zYMs^8p?dbK24mi#6xTrHm;bo*yFi$m9Rv&^miRu=bXcTtOZbY4Xu zuD7K1^r>r(D_K_lYW0g(&8wqk*yb1IPCdP19^p@Bmt$KHmfnEi`H&%T8AVTGo-8Uo zHXFPyjdJ@PzcNJJtylK49{!WRg{ZZuq95wI&G+-|1f`0)tHJUoA{a=b zVnqXEa&#(HdUG0VY_N^Ely#{!!Ib)hl%B*w}f`WkK&PWdHFZ`|*0vg0Ip`48TrL`Q&Iy&>b$6uCR&(H7W zi_P7wEG;cVbH{jGej-5@TCnv>;Yez8i5ASExmwEYH)ns|5Fe8{;_8Fu3D(~m>$BeB$VUsco&&W0_%b(!G z>g^gnaa;D?TemDx)kSNY5dV($SYdAYf3H5*@a|@hF${4H4N8TC_V+1__FMReCB5Ei#WZi?(0E;M#|cXu2X97H z45wc6VG~TJqBNGWym81woEFO)v@{=mW!g=O1df!9GAY8^zqoNu`G#unG9zm%Z@z01 zD18}O*!-%mAIu8Kr>~0#OVx!$k+Z>tH-j+7?};PYI1KOkSy|{k1^)QcPtl8>3=&D)L0{`Sa+M)#x z8sr&NIS(b0bFT~(SAJ`@vZyVQFY#;`ghk2^=LnKcfgOou-CoDa2mitbuHyoDUjpKb zApI^nn{03{He;6YFW+-PW8rjVGU!C&y+zZN<(5u%mw_inVv{&DQ$z?ESMP99v0@|s zxGLU-rAZzWY+xU|`)kwPSkQH-dEb;AZ$eR4RF=deXq(7q!3_}Q@-64RHluG(oF+Ov zET?7e(ZdkMc>@;C^ILeRwz&|GEz>=R{3aGn44pcw$SBxx$&EImq{WUmHz#UtLq*l~ znYIt7H#PkMU@#K^0kB|#)$JSa=}Zig>=eAu-=*Qsg9qQ<3{zxpO@A|mr9zZY^eNNj zT~CB-Bt(ce?GYxzRplMu@tzdd*3Z&JpOX^yPNnmQ7dJ${_(q!sI{)0f=?Vf-fpWwh zXLNeRw~)TruGqV1g^#pq#=%Pl$9b&By7Mp&aSG#Lis`p^ReaCU@4qpU>YJznV&!qC zq(RiowyVHf<;0cs<^4p|fIad(sMVTILf9;C=-r7RXtX3o(p65ns>1vcxoR%g#G zL`o(2F3!hEEJUz+E8B^8(xP2$kT)7fM3qS#S#M>u26kpynJxY%(feA z(6S)Gu_`!E{mvDDhcoU>WG{v3R%{hBOwrN?`pd0Kc*5m9EyA~B@5s6e=3&;5k_eOE2N0HjIm$-UJPmW@Saa2|~lGi8itv*4u_6-=o(dhPyz;R-) z$IJE~Qb@454cDGfk#fayTid#^;`C4Wu*j?D46c%+P=xvW1dETRG|u;oXs(zsz#?2K zS8VhKb_r`PJ@u|!SX#r**Nm}gBesydOpR7k6~lux0$=AyXG! zoThe%Z21QM7}kxaWBMq35B(s8DhBOFpFVv)ffwuQ>yK9>qpk;JJCco3+Xjm5;QD(F z+i~;dea_A@@-bY9U%Ws@Z2>}#3zVr zm;C;S1RbJpl2XZGo9^aa`m>B&hIReWe7#RsfgCazSdLYT*9fql?esQ^zzg_q$VXh6 zAkYjaXAeRmb;i+=O~)S^WM$FKEwO>ZYF)0h^}sK=O52CHs~j4-+PZFJ4jNA;{e6+7 zP(-(@e@y<8XIIk#h-sZ#846M-j&Q`4ixCQ`oT0c_Np$+lFe4U zv@D9)g~$p|?%2^vvuLd9hjjOI+D_`_&|EB*4LT&2B~%Vp+J|szi3CxO z7p4(tRzokj59g7#sxI4azTErn8-?Z6`kU^@QAl0g_3GNj>Ju~>AMnlE_73y9tJ#8 za@y9-6sDP->sFD9T3r@WT|IaPIJr#$q?-g2F~UIW#z3E%I1#b~#>+aWjdcfv@Yk?d zr*O3!FiW?LkLic7oI){%^2;RPRi%BiR$3c3g&3oKE4z{QAd|$3%qTIh5G-*0Wrgpc zx-nAJq!K8klr6cG-0Gl$6&{RxhZYM@PN?ad<;DY5%pV7?JJa0 zvM31~z8SGEhQP>AZ|wvW_n^)PLd72wdrv)+FtWYLnwWzYPal1wC^Ycal1B6URQ^lAo zP-rZRbrine)XnB+ohXB0oqkX$va%Hb`8^{;7Ks(r9z(}PPklf;1oxKxF|eSc=DGR$ zHEPnt1!`GAH5s*??f56|2u-5Bk{&ks*IlCyb$jvdPBgbr9_U{pWqfVx$*V=5&o0qBUJOyiHxl>v^P0;M zV>VvhuhVRRICl@#qcgKNiO23VzGP|fV`HT9QsaD>uwiWa+F0xQdbQG=zNYOEWfbAw z1792wV)cx7(9#RDXQeN=Fn50X$*st8S@`kju9(z6i(jbq2a@;Wcw1Yun#BF?i!Cmj z#CnKg?Vn0iiY+NfP6S)HVbwq}qeJ~vB+VO3PTRM{&>_wiEl+LhIX=H_7e^4I5-7{# z^zF`{j{IGrSg^Tnu#&`ppfLN)ab*vz+~8^)yV2%h4mjRXs%GYDLB@&w?PR<@iv=jJ ze0HEuR~oj&J6olemiYPTWog?uO~5$F4!37d9|U&_N}JQnUH?RMpNX*(`~1wOM~qyx zPqo**)vHA{4A?z6@AfD$Z^`+k9bPKB8|$t~_ceqLfiXL(B#$R{)VtoRXJnFzD0HGk zIYtd+*1W7geXMQjuc=LIvRF>!{rZi?%fC5rqgt@vcyb=I)7Fhtqf8Q21YUnJGnfGjaK%+M2T!Gi_lmN90_lKbZS<^f6_wddKU8LraQcxH@&Ku3no-I6 zt!~CVYl@MkShE6gE8mP5yOxgHLQ)bN0OlK+Q;Xnaq*F&!+Q*z0diLBU|M}i-V2DE?p zJj)F_wcU8b%DScI9rY)VukO9$f3zfL7)NotPcWR#5zPAg;)IZQ>xZH(;azLV9?nx} zR86=2PT?CSzVf`Rlt6H7KUp|;WD;z1kR4Imh0RC^A5xg<(b zR_p4XahR>7-wd;rNL~VzcTtgR5P?T-{QlQ4uEciw(OPj?(f~aq)Jta?DHLJN>YuYqS(S2nc-)c5Q3n> z7>!donU9$JM#sc%89R1tIu-aCPiqm@388+gIdMMMM+$xd=ZL!xSU^t|5DZq{_i${A z(2rQFt#5Zi0@+FAf0;xYvfl;PztOk-?*wvCD<*Qd-Eqd_`F@hWo_<7FmE_)>?_<=! zCtQybQvq@aMTU3%jyJMVqek-qh{OlU_?`F=NfM?Nt~(7mR9Os96B)deJ%|7Vy40X)@&4kneCm`s5#W_@0rRDzWwR!lkX_ExccSmOX;5 zD=-${@RTPwdt0yNWH;1ch{Yr+``_>QooJEnC{v^`b|}eEieSlP^5Ew$ajF%KV5zY8 zYM~N|*-Ri+zi@_QQT@r?9&c?iyptKAq<%k^O3`WNk$&XcwX$SDcXudZPB~S;+I7_| z%AxfGvW{qzGk&swYD!LQbMeN0yy#+(hw!#2cpZciOi)kWxZ;whszIGy6`qLhlc zctJC&;nD2$?;Y#k)tRQqO?WLBs)*1?ifc|VwXKsA=ues^W~&9&GUkGW?eS`;@O@Et zbG};vo5ehnsCiJ)X*@Esw7c#Gjhm~)Lek7Qp zw&eKbW!b7Sx)ERVZw21altRg<5{p6Srnj$O$f1CaZxFZKgC(g_w0TJ>cV`uqC4c_z z`aPJy=RxXKYB&!-Rhi(Gi3{+moAO)TR+OC$E&LQr;OQ2xHe6i62^D&HRbh{FHpM{< zoJco+Vh94##mra?)+Xg})iq{y`)B_w^rOe`?{gzA5xdQy6zHjG#h-WwfF_V0lc;yF zj3&Es*n&`=X=;7gq8-EN^>a#UP0prDV^cQ*oWEFNT<~WoVow*a9awND)IsWFbnu&8A9j^D%RPV7GghsqoiK$8<2uB)= zG%RkpLb2HSn^PK~O7U!%5I2A=KG6W6M@U1FLQW|0h9~d~1$4z9aKPTw+qdt7Cqgp{L~#M5jecGbM1En!d0O+9>yn87ta$$tRNXRvBME<0s z6D&^PV*7f~Vo(JkUDNvYJeHv>qux!5a9qGwu>wMjbUSc*iW3J^CJAPBA_AV0mo#qj zu|Kie;O7Gd$d_g;1z;7^Fd)vZZ${)k!G-~zPwT>)T+`e7G^=nQ#c+Xm>}8!b4A9kp z4?r2?>qdaj`JW$UH2-HD>X5|$8HfKFhyQ1bp|PE)*PiZnz1N(!8#A4r{kWv+tsZ4o zP4ZLfn5`8xtdi4c5Iz*1KZi6w*ul~YpaHunJrzDl7h~Fg-F5w0itI<(|Dq1GA|4j0 z`hr+IHhiTYITL_D3t=?Z&@sb6KVItzJXT2uOHQ zXmU@x^J4%9x-T(_8m6!kUu=MZOtPkEtV5-!4A+0%$q7dsBFpraZF^BQx25KI^1Z>K zl_>o74JRePg&6ce3Up8e^{MvYn{^}7hsqzZgc%u2!FduRYq6uDE*lO;;@_@SLl6I% zZ8ta!?xH{WM+YB8m^4g(A~0&b-y6%DrKcSCFKhj3K;`1LLDeTEbGOV#tjs<$%F$75 zB4*8=oe$fzY?TKzS@}w-#ZxSZ#!$D@aVOY^{V>QQ`W9O`$q|Tpsg{4jW9KzAS4+mP zj3y}CPp^Fg76N2QwV@wQmok!(_hRE}6jP$X0g5PxSfR?B$smi55v1WsdkyTjeCMqd zmSVi6^g#T*b%GI`rh;?boqqCg6yos_-9cF!j7D%2G)KEb=pk>!xkC4BF$Sc&AppQW za|OnNxkbUol6oSDZ!lRTPis7MO6?(iu9z6)hySa&5|SY)hRcc0jbprNUF=c!I1Dw9 z!}L3CfDk@mteW&z>rUSmBL!ktI`a)?55D{U`v)MZO4VQrwJ*&7jNOpc!8ahH&#Uft z-RjYuj2S#7kl`arl}Lp4D}hWw=;yD5e{njyOv9~wmtp+Z^sZunE~7t5Fsq z{!2!TI#eMYyRHObX&VMOTFm<}djPx*rAitqGqqb}Kc>bjRcjef2)?nD@H2@3`KQ?I ze9!6zA-8okGi_M?t?k5^2Y}UF?Rle+TXbc1#L{Zy*Yn8EogGybHw?WezUc3D^faK- z%(i-dK>*H0~KX*%AzZf?~rNS9_1Z)aq^qWUFl zAZF+^+Ku`e^{^6M)Hy&6}I4CSk5E^=mk(F3cu)G&?a> zvYS7Y$?S4A>EQ8!i4R+9*0$wHjBUmdYx}!DzcUFQpU|7d5{mj?`Y|)V=3Z8PdcD7P zq!ioiE{oEO$^tI7vv1g1>{43q*M=6t$h-k7eIQDo;@3{$L&zGJTCiq~raIl7crf?v zCcg;TFO7Ckon{>vL5Prwuyi~SyoQfR#&2hWx-}fM7t7b^sNkb?`Fy{jWgOEni~@;z z%jf-hWlaOShHIvE^9R}m|SHjd;ZT0{{j5uv~QIT?oFu@;n z!}$y>d&d;_-R=K#j5Ju!X-i51vX@A;n-gh3@T*s^^e8})uH6Uy#>r;V+wS}6HX6y- z2s?xgWB#I-bbUxRnD=2NTj|9SI-9;(JRIr@1(Z|k*RLNDH4L^QSpEEK>%Uc>fQPDU zU=Vq$g7up*$KUvmCEn+BHcgz=7r+zLshAIw6n$QPdQ%8r#!l5t>XRi-SPm(+zCSl*(E$r}WF429&IhcQ|Lbh&{)+n&PSy zK^*5)*UiGFnprsG4373*v`C8fgZJ0xP03`X^oT5WxH}Rqa*}R%`cIo~pzK0gxZsu> z3E#pV+lRues3SDzjE-FN-R@4t##6;B2Eo5j+GK`9eQE$-D8m z`=@Ye*c0lucWyeqp9kb-y|-)V+n19MzkIcQ;T1cC?%-VQz_0Axmxc4@&eg7L-?rBb z-&}g*Os%xe$I7J#_5>-!oY(*9T_7ECMM?U^C z%x6Kv2kb5d6RZ{{4oL@4d85gpXK2+lb74YwzS(nwpf9cGf951I(5Nfe?DzWJYraC)qo)OPW-??!2|VkvR~; zWRPmCFrt`jF10n){;S{b4U9dq{p_~w!*z!^l@b{Ck^arqLUAOeZPv>-n*UR(I^ ztQi|*Q1MO2n`!TsyRyl(?>qNz@A70vUH%d!hW(iPL{RYxxEZ8`SuE(b5NqU;Xgz4p za}$?orv2X7MN4iD>(N7nv{3}hAsE&cyl)TL!-+$as`nuf?7FpG-_dZqtzqM^NuETU z*1vL=Pt8CR^6TcL%!(Py^lT#gDch<4{;8YA?n_nNR7yTQ0#oMHP#8-o=g9w8wvMNX zU*}!%Z7>?73Es;R+=Dy!>vujopT3P5%5L5EPR*f*=5HKdfBe$zcH2SjAv^c(wfWGk zur|Um@#*6|H(&NxR$l0Bk~GM~>CTs%{Q&^}ri8y|)i26+Ad$=dNnSBh|f^=H2R)^Zw?C9#W&+sCk$9K8j+d98 zl0gLs3!&K5W!t#%=H_W&hz^Ab$S(g%k>_&C#2r$~g^lNW+w59cY5D0is__x*nPVMt z9d9nBLJj;eVZ-QuwZ$cFC?X3%1~1uVfn9|ok@BvZQd<`9nf{`Ro)Zt~`WX{rAmyh? zP3A?F6h=mFY|+DG_X?-&ErDFVtvhzNaQ@27)bW%;JQRN{n*h#hg*hE z>nEzW1#v>Syj-33wa8Y)dFTEr`eqx7-bWAIF{g^|vYtzX@bABBNb}!nVXAf}q{x1E zA}DBoGF+bMY%y0^Xi&=2X>)lTOY1(Rg=LrgruCch`D^P<+MKh4XE&1inB|rmemWQ; zhGlya(GPI@0^boPivIFTtQkEDM(R)UA^VRIm7}R|ranLC_K$2DyKT4QF`Bp(C6OAJ z#LZJYhIFuYg@pq&-9-fhD)e4#&7eI55LskD2C7&Zr;X^AZJtT5PBieTO59QLdMo*H zuNSau%Ka>nH4DwL2la+$yX6q!)F~(+=CO+#MEk*8bf(U&K0%Y@AApH^%Lh}DVT0*; zW@FMX0K6`pHSlm?Yrt3CC5H`9>^5j*9rQ+OxI!a0D}cLyVCYzunuBy{;SycNCMeu{ z*dub@K`^^JeSp6HsZK3})52bspSGhahf#r-D$Tb+L)pBgz6Ak%^61f{LyIbkY`b>t z8gQ}0>bJeg8=DW^xVC6QK)=uor*-Srkrwn)>DYB|8}w(DGUzua&;OaOi^tnZy^hsztaVY@;0s9Xu znWWi5eyHC^S{IjFQx6udG;!HNIqNmSINS0>N&>X3`@EqcUrhMuYo6s` zdg^z+eeLo2dX>}n%=2>e+vxYjt4EHm$~h-~x0ij=hX(NJ>X3+@){D;7)$Y!JoE-jX=DO}{e^+kq z+TVG~7{|^(=*5omXnM6^rQ33g{Gk`(6LaS;d>ZX=I384_w0K2z_z-V80zHS=#P{?! zH>NK|odmZX;j1qBC(c=ky>GVI6_+f1fXDW~1v+qVZN zCe|7rNy&RPZo-6lm@halTzH54sV_;1x6fO2@=IB-Z#fx;_yC~veUNHpNCG+Sw+_Zj zl~0~L@sy4cuZrDyuB|HId=K%i8WeZz$ZuogdtbC3VY%=6>QM8_Q zW*>isAW?+<<^Eqvawn*(NIRU}+>V@C6uIhQcJ{!|Uw>`eb_|jW=0TI1Zy2%V#w-!+ zcS_##>C2betd4W(mpuGIxnRYL6%d&m9CDBe;gL90rGDnIwAVbAuCcMPA34O^ZJPKY)pgOLyBr&Q zY2siv3NM}jC<9t{IS}sf`O~L7_6`KZDniS23>{KH7ZL^sPtnb%$29+0na{$ND+eAr zbm*KzE^KwS^xj1^sb&mY$GTxV5gly=zy2;9FFJE~5f}zd(G5cD1Iq)8dGqE~R95yi zF*!_3le*%P)}%AGDHM+S7#r`0#{PV}5;Ko^4%oQiQGaf!;L#HGW?$Or2lxQsjopz*yP_s;;@PmcrZXVy%tW+zR5Z%|5 zr`CSpwA6Zp-~6CKgVbxP53C8D67Xuv4DWYOE)VG3xie2UmCU!`@-K?79S;)<`qs|& z@iu(>`t|EIdmGknhFkfuB*C&fq^&s#W54k|(p3{KDncyl3Y)&o^M9=)Ubjj5ss0D^ ox+;^{RjIag9Blr_{d zCsinvB@PtIqKnIy;{VZGSfh{MmR^!SeG|d z%3l=f$z!Sx&%d`gI-I9975+F(4d@q$NC;6qr?uhSO3Q!uF63Xn=)uCbA?5!jukI8r z7tej08FFLCza{$x+i!o2+^!F7}{(`BcQsV`ZwB>TK}c7(#v%$u)gPv{uSGlo;! zDrjAy65U_FN)($=`+p7n+H0A?*f;lB6pBZ$rmNokpD7`R%N-W{{^oV~|8IZ%<$?zi zragQ1EMM>iHAOoinbF5DDENfYH%gOuxZ#gK{_^~=J%f?(<;x`lzeUuIo~ITpSYR#A zZJni?FSv~U-7Y7--siH&_|#QF(Ex>+nHec7zD~!yzo-!r5&Zl1ja)zVkoU-uBaP;w z`a;5g1U0Opf3=lVQ&IWSdGD|3f!V;&(Cr3(H)bVU%>I7*^eJbb&$?WS2Va?C${t~1 zRndU5hKTlp8A#Cfm1{eqj)*^-~L4?bM_LE96( z6z|n(U%vjrn=w2Ze$Yj4=}f6<8fU?-zyJOlPdwW-Fi^p|wY2c3yPU=L_9gSX;q$+L zoLh#6-;bC`D#noS)O^_n7wf_Q~Cw=bmt(KILeQ-Efa`-cX6aWKWp4D!zK{ z+}zxf8_mi2%xTeQ$;r#7SujK`80iW^dBgO?oPNm`m!D;FXU;^khJHT;Hy$E{N}l@g zKyH|bS^Z=h`6MDZ9KDDj$NN^e^xg?FYig5eL+8Y2?>eGE%S7Jdj%&tUCBE8O))9%>O#bP9`+}NN9C}|pPb?|)adhpj4cb+J7?#-~o7W>w1z)eUyMQ`6Rtef)TX z>#rY0lfQme;9Gtwd0T7iWT)Yt7dide^82(?qx~-bey1&S@ARh6&TM*e#9z!RC^;p? ztIIA&Ppr))r9ZQC?S~H^>Wtg4Qq0!8oNW7n#Q6BVd1I~4+S#@-LqkKnR4y?~RlNA^ z)tf`x@Ii)q=CVXBCt33_`|-hxWzPO5w~28Oh%TC(LT+WN4=#-H4|>>lfGsN_gL5B`wYBx| z_mm3eqtjcf9_={I{F?uJKbMjHRBp<$iQ-dy9zF51n=8+5$M)^Rh!EyTgMr4>Xdf%$ zL3H$P1%F}cDh@$aKS90wUHC9Hnb$j0ZF^P1j1ZOhsyp?16~$Q9+Qw!Ci_Us+XKVF~ z6L;=+-CV|2kL?={X!}*TD`A~rcF*>*z{48eM`kj&Z{N-v$bIqGw*V{V%+x@`%=8p1 zkB_{V$Hcn9g%r*uex(sgN2XQn+gvl$6SS}HwW^Et)pzV-mmd9m`QH8e!xwMOvZiv( z8xOAG7O&LIcb|1@aURHX!biSeRY+clqPheh@bz&1^r+5suabN9_r?_8fB*sN(AZeR z=U120t`z@CW7U@xu{(L*zh7G$b6&)G)ciwHk-X&Qb3$6lC)e*%tnF{fan;>MUhVZ> zQg|cN@J0&H4xaMiQ>gxX(X#tp8v2eAPXzU1v0o*Y8B;Ic*02yOt*&EwYj|XuHSKyT zw@pt}xYy&9#G-=ns4pY0wtvwP0}cx1D6Nl?q2$RSRR3h3PVSQo^F|dmsX-5?vHm_E znH-UI+}!bZc-%A*oi#p6GBE~j6Q?XQbV~h%8oreU&<66xL^IcYB(LXj`=$p6<-$sw zx;=aLOt#--OLd*HOQLmjbYxHU=g_b$qCM+!qI{k{jovB$q^C~9pbn>a2Ua;XRhWFB zJg1Rr?CXXtTl!cB=K^HSY7!vFNQrW(N zgF`?nW74X;m8->BT8&8{L#{!Me%s%6e{7UKOvElIr-= zuHwY8^9S&C8=FMPy5L1FPp3p zl5?uU?kru)9lcfBNqDjs<~M=tHsjd^UGsgqUM#@(S(cd zZpWOsvlclcUl7qnp`249K|Asm8LsHTmdWu>?=(D??~^A_7<>l~C@)&N`dnMSD{4>w zO@1y>cXxL$ZV8o}OV?E2Sh6bB;m4&g-wz1P)T^Z%^=zw7aS9thB%dH8&_j4!-sR4c|(mer78k=vG>^Nwj%ap z^cy#CvTxcH{r$VSf!{(Z;HR>(GF?&8|GA%EH{fNUi@_h9oJ~D#_WsnPM?K!Year8% zhGYHu^I@(BZ(x6RA-DNHejFYi&Wk5-N?5gWWi?V{(zX+~qfvBIH$=bq`0>p24{!I{ zK>OAhb^&$c-s%WRm+_apG11XyNXRs$nW3VGzkh#H&uyZMSsmed>GEaAPvoS9Jo*dh zvNm>j*fwu=W(tMzYStsS?qK))+iX0E0hZa+ZQHiV#J>E2)vStE4zp!gWbDDw;c**( zk^gH*9#3{gQli?pCQ_j()x?K4#_{Jj)3*FPL}Z+(MN>4g+`W7EwhzzMru78A*4}$} zhs_L+sGVb%@LVyfAv-QHTdrWLWx%#hO=I2m`fXTlvbx)AxX7kPgoiua z>f5||bIHe#m3{_pCVN+{SfO0{WFI>ZPZH3}hxhMy;|M=`^k^j;n}(+5Q=sZA8H@_1 zgl$d>Aimo5>#y)6s=B)7zjQS-CPacTMmCq)5BkKue~EDC-?Q5`2Apc{0{h! zf?i#%lBFw%U!A&pr|Nd0$-u9GGCur@tK&z|PE1S`TV|BDwHct{Sh;$&Dz@;(jT;-V z+yMQZfByMr=*yS;Z{4~@8ijT1)CP{XZl{E5>?Z{)+)YIVLB-g1?6P%@} zmZUmUwF?(Q&{Hg2xbSOtx6wc0ThZ7p*9cUMTKC%N?dQTKn}hrJpTmO`?&Rk$VO7U# z9qPQ4!En_RL@;>B-XIxa#~n7FxtakulSClNiQB8^Qt8Hh+T+oZk4ZEmF+`vC-bxR@ zabsa$W6EjZz$&&Qze1*`##q3YYX18RyR30qkD~pg>DhTX%6@aYVdVMY3;R)L*|8^U zUYPB$73AkvQC5CbU0q$CX@MO>rjLC3bkm2zLVi(E^|AmlF}DYI?>;IsJk0VZCtlcX zaW1XB)p?*UsbD7dgC~c;{{58=U8<%-UyBjqejE2{hE!HouFR0Y;kN5^KrIqZzz{luChf_szm2BGAgMX2#IQxU$d$G{A#|PHy``?MJ-ykcK6Pmrwm+v z+){il{z4!rR|OR}Hie-GEV6UwPJam-!w7?s{E2S?d5LV?;^)SiZM+)}qMYHw3*aZ) zTz-1toN*w&;s8#WHBsJ)`IDxXZ~r(SSe7-NBEPCD!Y| z7Zr{myR}%C?gKKj^)i|{IlX-CTJ=kpLjPR9-s>ugFF>uN_1Aw@&Y$-U3Q9ymxoTu2 zuxr=ptgI}O&w$ar{;KJ?7M+?poI2KJ;@WQF+@7^~=~7M(rB(lggoId^N*(dDc-EeU zgcTNXqx(`yqiOB&w_Vno_wkgZ-`u@-ukgaYU%yl>RmNF3R;#Ae|lx%*O1v!;|Qn9aplq z;--*ATfW};$g_Jgex3chtg+RuIrDlpa?}W_dMw~MyQK}&Mx&}H8;3_LPCa2r``ZNn zWPO4Ta&M%B&9~hMaI=6E=?r;!d5H|$Wbe2l>9OW8hGmWd;$J3yxzW2zuQnwt7R>re z=Wi>odGH1c;c6*8ZBe%2sFSwR^S)&%Saf_)S@9rW-w3kahKBry4xR7q?L`(k=j4=) z08#FB(-rhX5I=tMBy!7PyT>{M0|N#Z$GJE-qL7zPojh4Nz=_o`-oD)7x^|xPm^zLs zZDyidVDH|uXjXu`C;M{-ND!DIixT`~Te>ep)xjBjhM&F@wL8#0s2qUFDRHY2cC0o^ ziARrCr)Jua80a=LVI(~>-br&~{H!=)@aokor``xZVf+5(n%jQ}sK-a!x|*BEZ*^#2 zKx5VxZz}s{o%Pt?KNgE)++7|x$ugLYMZWfB6g56)pgSz+FPL?c2rReW;S~`2@Zld` zTw>)7S6`nlQB=R}<9*t}A@l9KcjVE7?HD-s5m8YBA|iDxYxH+pq@7evOiX;;o}yz4 zMTTdgC1OuRg}&$Mc}QEvfv?=GH8<<}MHO+ZMacEP+D}EGFpdDFWi?wPk)lv2D9KEO zZ*}~-E^m62G1+J3u5)pD2B0Js_3T)Sy@BiF`=;4n>GHXrn!v7%mkm=s}Fd*b>M!UuO;+IAB^I z|12~tj2);|M1p($Tawq;==eBl^xNg6(OBpl(&TdN2h^$5yJ)Op1QIikBTjsMoc5h zlX37zdZU8*kz>GwNh0sm#c2o;5V&&X%7f{LZF?Rbm^??fESR=Snm!vSsTL?niyd~8 z63idJ6!z`cuU{*tn)65Mb?D*pLV2lgT_5*pkWzTH^oa=yX-P#zgRh2u%(@-2n|y?E zh?v0rRRnH6di3JKmcz_ZhLr}ZJ|S;Z%coObL7^^IEp9k?_Sfl{n3#Ka?^cnY<+99t z(afuiRshwzE>a;N6^;GXD_3@J-+mmaunwhxbk%Y^Dd)ZIKip#{4YIbjf22}y;%FZu zxK82%Ur=lC&+dwb;ocfE*I7VO&;XU_c4DPBtE!b-0QLbvbw9UxP%hEUP3vn)Yj(Qx zu7iEjNib;L>&C{$s1|qREMTTQCld|c-Mi1=tdpGrt|IIlDaL?mXdW{E`RAYA;^Lae zjvXs3aG?+!Uz^AGBbL!;#~N^-%}KXv>rZo;otbVba4#TTK!m)X)2$GdiLD_~L!a{d zdLyBY6o?R1lpbKSs#QX{7gEn$ z1Lsj>$Gom?Cl$q?qjxII8SQ*3db^!`eCm@+D{6L$0bn7q<~GPQB^Pb3U(t6qiJQxu_)!(>6%o}hSOhIfMWzU zC$4(g1wbDDy?a$nO-<9xhR4Rn{JE1t?+%qHKDB2qcDkA&-SVNWtxZ)r7^lXgsg`$P zE5gGBZNF(tb{d~jP&|NuR)V@jpyJi5S53rHq=zd{I3cUv z7rRNNQk}*-?xY!2Y~81udG6xHi*+*jDC6>U96A2gZ?L&LsM*Zexbn;-trSC(uOA{1m#VY&blmR-{hMi`c?qGA@zqOQfi2co3lSM~nUSzbnF&xZeZCrv~ zhVG)H;X|^U?h#%Z~RhvFwH(l{TXj7hZ298l= zmQ6RaL?B7^=eIHe=~fPXhhu1pRe4NBjrP<{|UsuTS#cYd~JlB;+lV@j|79(?up?s_S18V ztd|1XWQ}h0{9s0ZbAHU7nJw=m*6?;Z9QtHv%hiSk1&1HJuG@v&<;jZOD% z^gLii@m79EV>Zy4r4$J!%R}TXAmxLq-q45)C)oR5*G)UbB4p zxg`C(uhv<^pqR|C1EgPk)%vq2gHLEH!hPh^p^7sqxAM;JG(j1nt40*pk){(1%-PUq zm(!p7Q_&R8hxx&b#m3k zHeKcmTR~_Te|dkGJ|lMN@V0HoDx`kiud;j*8*6RZ@P4SPG|g#*xw1jezJ&&is+7>8 z;c9Z0N0g$VcIM2#XrIc(w>>ZIQqk8BLgQ@vYy3JO;HG_Bk*#dMykF5tN9O?)fX;GL zm^->sv`)-nO|((mYftC>Q%tF>M0S=3yy28_$xM=u;Yd|Ab3hSt(C6F~#JFgrNny?5MzO#zDYYvCQd9mt*{; z>)7Q>5C8i0EC0ZlE&lHy$8Vbf&Y1N&R{I(jHCzU|q1!e+ir-`=JW%v#R^2#m}BA|WQ# zEvJCmp@M9PoO6({Zmgt&#Ee!ac|z|1TDlju-mjC$FuQ~&XI8#Aaf=1YIuV!c^o>pJd#P2YZ&|~MD_P6+mrhsJnc|SE(12rAt zLHGB)YNOmQI`%lb`Auh;GyH3&gZ%W=8z+-Zt0OyV+nfW+BCEnqvkJMaguBc>F;St= z62*;%mqc@^$JIR*icC#a1&=3_xV*Y};~Kl1ysBC0{p*njmMmTj3j8%|V5ZS!e*=mU z8nLqABL=HB?yUsMxnJdrF8JN;H8ml}ZpLmnuHMi)=~npueGj&4c=D!(kEHYt4W+(C zYLA%ruf+e$=&OG&VKa;sC)LVn@Xa!%snB;n)jr*8y`bi6G)p^hz2rw{(ayl{?%G=j zp&DS7@7QEfKCG!R`gEMBJHlSkc`iWcBf<+jjfhD#?{wSjbTCRp#O$x`BZX&1@2QS8 zC*BQ`c8)cw2x3pfhN+@-CA9KT5FLQa@y#dAf>zX5*VOnvdUU-$iLLzo`s^Ud#-^C{ z@xFQbRVU4(^vAv@eWxi26-NGb@L=`$D7dbT7hl}{A1X^WFQJ1gs9Ql&Vj_I=jX(WL z>v{|>Ywz9IXr3)K7K{@~^||HTD%91YTv zWy`=poO|=;4T$~j0e)b}b9emDzGh){YLl8D zo>J5AJo9V9|G;J0tk)k8HEJPEL+Ex_r#U@I(#=vq+o*B&>{;K@UDH;>R-j2EQubuG zSWf>8o+UCsZGw)N^w7s0YHMFh+V+G59=7M-vEu}qUbV!y^yj_fQr;~b1}-lS_h*L5 zi8!^M(GY#na+=3gwZ^)l$F@6G3r(Xuy(z=WfHh^8(?S|;-k7&++U4JxZ;Rz`TwQxV zYSseHNY(KU;3|E;lwM~dF@FB6nIw-gOcEYIg249ep^Dmnk!CO2z>B}V1+1rG`SX}W zy&Th$$9oEgK)~)s>DVG^dk&ly&M?Xop-|jseg@OS@z0(x+PuVCqe*Lj$f&$F%v&ZF z3ky1(^w6loM=rDeCX61iXLfgx^9FA2y-_tWpuDta$BT~eQs?NVQIVKKh`i21uNopr z9GUsyO@M426giv&>aNqPjILhYgGLc6@(RrC<>t{Fs|3%<$=wq*{~UsZlhquMo4i01 zAT;vqGnE9f8$p%*NbID)j&;dtvX1OrwR&|eut9`atHTNR$(mqcBqCaWvle0vvMqR^ zDp0ew(<2RB`xZ;>P%wZ%mS$2T3z737fXlICf79u75cywHVzcGZsp?A z>0*0CHtwlT*h&XILBHpUMpLfVf3W1)>GfcbmAHpqG>0oK3V{@Z)JJ-C*%(080BH;ene^7#^*V znTl^1k#U_Qs;U1G_xx`a!B%G-L}*M^P%>0Juy%@}7q1DV&FroxI{YnTfaCFa?4+wf z9`@jaIQL&T8j+D8U?eF+*h~@|jbomzc0=?QXCu@ztY;j6VbT z9iR*2KtdELDKD=_g(FLW1ausYBv%$r$@_71`{2=vW<$v(!8eDBG2pF4Lh z%U#+pxfK8Tm^zMX1xV|+SY{w!gTf650}fsy_BX9e_zHL!?#ut~_q~dCfeIxQsoO7y zg!|@fYuAphv{IV(V^hVJQ_gtpZBO`w!PsZ7Kn*9gbH#Dr=1U>`eBTdjj_d zllCG3&e;_P-?0H9n-W}!vs1Z7tT`m#bqWnCEh?`KTN;fn(+Oz)F;)~n7-WQ6MvGlQ z!_)Q(^fOeW`1e6%sK&u{M_8R)Wbasw>@f2c0+^+#6 zZM0Z~J#P;!r=@P~x_9s1`96FY0#UQH#X51EOt>xm+O6(>>j7={q5g>m*@Q`}KKe9&Tjq3Z1N&!5Eu6^W2Z-Ho{ z<0FC=N+DP#@9oB|1rudfyn*IARp%IrnRB2)l@6bSYUlJ7L&JJ?stlm)T5_a`jM^{+ z#d*!ni4VPgt)@FThA6coA+$S+dTd@$vC| zKc?z5{|_*my1)KyXvd(G6!-MsU?zX=@!i2;T&C*Ro-UrCM(? zKvfWmU~(+b>wn}$jlit564O6-ICLEL)3bYqWH|!R{KsJZBO*s5EcI`Nz2)KukJVt{ zt_O~&;RA=kcb4_FCz3Q~xMH9rk(-DS1ri@`?^n-k#H|L=h9^f@Q4pCo`z#6Vm*8G& z(VD9j>^_qvYT4{iSe%VLMX=MrB5m@y=J>W4!F3BH92zJfsFll?N85?_Zr-P(qO$$Y zyBCPVurO8Cj=<&e`%C+f`=oxw7!9gp* z;d=Ye3V#{bY$!N>))w2Z*+HWm?Qe-QsfiT%tenU$;V>A2yfOUwvUlLCeyFf#=$6^N ze4qjoWl9c*k$U^KwMjt_8GU4|U{-G#+t%8gf&4f$>BU{v_|@AsJCiXpJ>VXP-iWZf zlcRk)0~R{|3$fl5CG24Ho-6=dLi0n;Q$`oU8-oUu)sWPHW55pOmsAVVFgK+M^evt##nux77Tmk4 zrLb+3VjwX|F;RiLe-J;02j4$)ehhL%a0Ch6-hj;~wmU*EgHWZ1L!3iGhXPE!e*HRe zHxTz1k!3Tj+Zyv-bBLp)CCA=*-vH9LtoPQ0wOb^|qYsjx@DQrH$w6_suYYD2kP)%J zfDWiZi-pXpZ5cqEbm(x;dvBHY|FpXfsH!YTS_k?5Rl^Wo#Q4+G1I*8_!o2z84iC3m zIP|<>3|8K3$nAM!rikyK;`^7C^V@;#fdh^Gk3T~6^PFNpwh$SgAah_m$r-hYat-`LPGu@w=4-y;0UI z*x|6?bvrhk56ffyp3TCGVXb{kZV?)MJTAY|MQAC8e{H&m$fRv(+ zcXz~MFDESD;lQ<3c2HF6e!GZ}?4bX8e9wpP_uj{+*@RD5`(b%&mOvL)@nWSqOFkbi{wok++wDi22 zvr``(sSI5JqZI7N%(j98k}GkrO#lp=UR}P!yGL4D51_H`%LmVq$`focIvJKSiRqIh z_n>Hfd%<{U5{a>TB7_QR!DyJ4t-R5Yg}f&w#{E5Wd69E0Ug0}=+| z!J(?z<*}YvQFOS=^&1PQy{RKY>JPieXtb%H~Lu zSyC@rB6>LFCZbUi_aByWTR_>jPC0>eR7dHFuRfC3+C5(Fg>p}n#B94h=DQN|f)vAJ zhzE}tU_IisA!|ZJJ6Lb1Drgd5MvQ>CS+l8waSpBt<8PmyWwR1 zFnk#<^M<;!#GB+L2QF9SYRLxaAMe-8vgt#?C$i1wmlw4Q-164N*GCZ@^UX4z+7oQj z;qa2;>W>N$cl8;7kL(O*wK;Ok#KZ(u@voR~W#rw0L+NHHTIpi8=};aXF0pC~o3 zEfZ5tM+wkV;3<4NuCSl*7|Cz%906-8yB~SRaH+JZcYG%X4;i>93?u}ZSNw`Ey6u#9*$a8pYJ|qi)N8~M}-%4I`*x@>;0#Z(eo|XVL z3HmU77>uYX%AI!dmAkyfF#bzP>7ohb-}AqJBrEa1>+t_cJj}U} z@ybUPpihEpJz#gQE(khkP|pX; zxYic9yT2uPgOn|N?e~IDej3geQn>_l2duMFoyRQuTJv-Ua(_M`km%UWrNyQZC$_Dw zM%utvWjpD1z~5=W@WHzW&H^+=8J2k?>R^ev7{o4%0SBwxAGgtc?xZ3|RR_-u?;?~n z>A{qk%P1jF?6Zpnp(bf;asee;o@5QAw57yWMi3C}j610Sh7HYGHZV-O7cu}VbUZiiQ^CUS zq$0r`GW=9&6bxH|$Rf(Q;Q2wXH7ej%f*iiA0Q*a&6aIj_<^KIIOQwwB<{_SVq8`!} z)2}(;F~M{&C1W;IJSJs-Z-((Tke{T&lj2EgpYY|v)s20lrUPh=%_fI?ImNGqy!wu& zqJh-u1Y{l;l7;bA+3qshXjI5+0)K`Ra{Gpj8`V3fF`z~KPJ~w3Nd;B}7^#6*)GXrv z{*xi$u$V!9FB~BTc9;NVSkoX36kv$K68P}~uJzpA6|`S=#UjcD+5eX6UCoY$8v6{0 z&A`>?68hRIDwK#|3MZ=om9U+i-bXB(d4Ok5%zwE3gi5u_pOB?*zy!_gRKcvzhf6Va zQ*iCo!|D+NQI}ectRD{ALEkbXM}1%!mgeaEPyYe9lB<14YVh?U@M?)ZQ}?{Q>h5no z^mSlBsQ1dtUjPvG#jXn|=luTLjJH6BzJY=Js7&l)Lb_q}8T1gtXmiRkX8_R-^m0;O z+s^-E8GGm%`wa4RtBZ>(`4s$l$mr8z)LKeFGO)9Hg1a_$S1d4FAK!xvhL=6!hp_$Do$<{MZG8=k+dF@(9za z%LAwiUxSOf_lHYW;Zz^=&#S1_fGK&h9E&KGe~_ApFPBSr2Z#y}d1KJ0XZJ=1xad|D zZ+)#KJ@OcngI0cUJ`4klPGmyk;h#@6(e~)Yv|nEM`}(xE`>!yNR%`;3!LVYm;O7%f zJOaT3MCj#y4G_+5PKZF+k8aI#o=Oink6o8Vv3|l!oNnYTacaEdkpHB9 z7vQ6LpigVD`(7*zhCId$;l)#3I0_`9!JN>;32VdSGWhrZ^M@uuU=j+JTFtJZOKcv^ zMeJ(dVQQ!|m>J^eoW!I689d-oUragbIDZVBd;$XE0JADGWv+iY$)j##2v;U?EbgR& zc$s&)xacBF&CVZ#XfaTrK?r7Es_W!%)lNTM3{5H$mJ+sH#I!FjnaO~qYwWrjh4PCH z2ozr-tu9aw3M*h+7`(VsjROY=O%A8)4G<2lNuAij|H6Kxu-BIJ$Sz*q@t1UFZ+$|5 z+w4>uSA!8cUGOIpiGNWjGoHU!*MonPnC0_19f};-lnNOE@Y$WQ6pF%@-)~D%Ywzxk z0+mDT*mTktLaz8&q_%+KFF6l(Vx!RnI02G+u`qZPPQpEEJ`|&Gqcc&2sp2s&(z*1{ zPls?)e&SfeuMvjRctZb;v};`(&D3%t_k{4 zakv@9!)ZRSj;=v}qYH89F?v>B6No&duXM$#hs10H7$)m?n=o;t*DV~ub}6B0hNG2I zSvy}buWcmb8z3$@8!)+Z6-luC4m&>OpZ`@|kH!{I`*{AEd+f6nt${_Sv<9Ym+$&b+ z?7gv^pZ{N8!Dx?6RqbaM{(X*ejDDu-=pjG#!=~R}=`wAjeASCK3Me^dA3o)zkr8Q9 zbdY`BA4}FPQQpva@`{0xoBQ~M2g{>B9t=t9?mWQOot-#9b3M{<{}ZjuD8Hdwdk1_y z1=D?2AS`PzQeeJCJLkNbTEwSMXNgk~6d#5a&!IIWY_GENwv@)bVO6`0jEtZ-7FSg% z18yS$rl7-&MQtvrs;a8bSwKm+z4-S;n8KC)n}p5k_ku12XQps4pp3ZS_yq)L2snmh zu4t*bC<{&dyswa)0G!9T*jQ$0hROa-Us9pn}uY=2294IM)Uf%nNcwqVS3=xO{FnJDR6oVh>2+wdGj>DIe{K95@)K|T(qI7e_p?t%kE#hx*`CZ z$e{syiUAUN&%C9SO1fadjycdp+4r%vRUdGo8cB+M^X5HNGTcFwJao#X3iWbT0k#d=$q9PGnVClLL2xJ;U zhD|V{z{e*GyJDP3i5Pqf^Q7NSD#o{oN(Y}b2*B;m&Qpfo>nZ+=R{ZX%v_0NjsNCT# z6-p$0n{S_pN(wxihH)pd&OW=8)Ir@?Y%vVOR`7$O!#mW#=JELH*p`oH|MLPobtsq2 zhO?(~-mOc_wSt0z37h)dGjnoS{$`0)A`QAbs2Gq}hy9yOLZG_f<+9YLKv)85Ga&#z zLX3bMxS}vVIW=W)aD@}4k7tE`Dp!h81)*_aU*4&qp`ml9<`SCLMC}?PHLq8as_GFX z!b3D(2(Ad?8wLceo+m|u#U6O-*~1c9@2@UmDkyJuHeglb>6FHJOp*`}1=bAxYH~(~ z#Sq2!&-Xtwr*fDw@DjdOR@i!j>60W$)v zg=jAHa4_G5ddh}GMMGCiI5W`)$!>tFB(ov5d?8wW{dxt|TwRQnGao(?1k1C8Il)XG zyblfc)){0Hev>#mC9K1tW)%C7xl|}KsUWtt9C2%IN>5H!A$T3>hL;LYwF>%dLcGEm zQX)Zlx#!PZW3kvcc+B)HSo^h`8>cke?$+E&z3c7W0HU4wDJeEq(rUGWRtAzx?g0KK z>!Ir}P;&8pN_D#h1kV3MquBc()nx;?q zC3oL~hvR7KJq}Hbj~ipTCo#uL%)dxMAI(L{MDZjrwTSuW{bZ1uI93(p7N}}yu%Leu zn>(=xV_1){Q`nF0v)|;RSUqd!DSmKgwJ=FMI~%EfFL%Ph$%z>mU|Fpo*20UtjXANt z&#z7}GOQW^bYOzLWZ?|FQNXEK?RVb{27A`7!dtBQ(FlgpBu}&okx(p%3vLcvns*{V$=oEl2SQ9tZ1d`+YHDhP1SQ}s-LfSXHCYh^u9D9e zB1)KU@{(sEc6%c%%E0deunQ_P+w`kM>qkiv-O4Bqw_ z9t(0w60(0JKv6?X8}>Pr42`3_qZ*#aslbs8Cw6y(0yk^$cpa^j%vwF|;MbJ+^86P^ z?dQa+)95B>P@DjgC*6>7YOk)E@& zGv-iw;0~?<=o^NZ*8_-ZxnI%L5@FPS$-%*)hcsb?=OoM`(K9Yyd=Xxw5n$1ne2Nq~ zTPNFipb=R9YT+?KHYU@a*4M9-G;)^D=aXRyIRU6D@M}U}V8PIYdD(N&d{BCbPXozK zR8rl*z<`*h&>a(LrNC_(uhwnoD&4f`@ZrNq=KLrzgeU`^+yL_pS*1PQ=YS5d;i#j= zAo>q-dXbw3CdUSnhzO5{4aA#s4tLk;5R2_^JPnDrf>>!V-K!3Dkj!9{&KvL*MpTD{ zCKUwTB$*AHPZofN0EZ}0$^;(r0B_$=* z(C^6UqLMZq=e|D4fQ;|OAj5!7j~wT{be>h5C(bU7!7!d3+e0U3Kpm zE@~L3Cr%LlpAnO_-GYq1!GS~ufylW;8LJn?=ObJRzC4TmgbD+}8pl!xrHKjSa&t?I zPHNj|NlS|!JjFLHGdq?7RgeXP;fjpIAbK!`w>4Z8yw3tx6FUw8kOnCT7;tYgOf+#L9CIGlc7rTl} zN|rP5!nET^O2vQ9{@bBl9H06gTqJc1v+u8m#?R%=kvfoge5OlJyCG6 zpbf7>)%I4mDGLRRnh`=DR(w@0tr(0vs9>##!2;NpXim~dhIk1z8Z8Z~UKD(2L=32e z@ODx9DBK8UXmQDS7_1H9jE;ve<+~Z_$ z5~nm4RuMwohu17)5Pw?7+gn5}059Fze)SBwF9SCl5T&W4t}d;9bp87EWHc98nmE~p zMn|7i8LQ&i$V?@~a_he=4Mhe<;Li)k4v}jf9z1y9)g)Uc#DYFn_M|-_F)`6sT*i5` zpAmA@QAs^Ta~NtkMNA)u*K}u)kw@5#ku-flenC$GV2aJn)y0D^y?G32(%RPcF&0x- zPY?K=j4#DvAQ8oll*E90x~@vg*mjO(94yyCJPkjSV!Q8D+-n{B-O7 z_DwWa3Heh3anRb@s-69=yj+pbhbkLZHMxJ&>a?vQZOW!IR+@XCXnWuO78qlwBjOsQ zbtv0#@{@zDr9% z-y6t@#5;I78F`7Aafa%+2MG(GCC%+xQ)45!$b(>Q zWF@FLcO`kAi!8f)TTu?0K-#4bE2-769_YlS-gtB2VMayo<(dzhIess;cVvP+=Vj=dF0V>@VZuP zly?~=sPxs*X8}2NIfV`?(d6uk5iYQpHGT(_$;sWc@-5~ z+-E_KGLbwmerafEm~e0lM<@_d6)WTipo4yt`4#6kjx>hNsD$qM#QMM9cy7E7>8$RV zkuWFvK{8#8WAv)^WFHxnA(l!+x!24f^~{-vykw|yO?{a~yuZ_Hq{20PmqDq#S=x{! zas3FU>A`3D@8qkc*==T{!eW*QcO&c>E^iFB12>TQFu+!F4G4{SY!N}LHsXh_AG!2? zIqMg&97(Te(Mjb?04#)|=E9*>+|m+Q z8s<-O7H)JA=(U~h#Phih74ITS+b{;)bTD@wLwk146^{=cehV;oFL0FE=*{$`q$FGe zqBn4mhbLHag=wmwPw{)_%G=K;!d4H+Gn8%oMsJsth0yABo%ny)SGR6jz0|BYv6MY{ zM`klh3GvGjPZtIh&;t{wIXpbPAkapd( zk!$`sWq((`dl#n6g0yG~3re@SO(NS{8?UuAJG**tsEI`BM-@hqqT6O)`2hGM46mPA zi-9m37y=$1p1fZ}rIV!wGodipkqI{7e5-3)09(n3BN48F@Pk}O8x56SS72-m$THeq z3+=OFkW^#*bjRC92BY-dyAx2u5zomq@ygyfp8K+4dpE9f=d$>x${0kuAGt(8S@|OC zT1d?ie@0`VjBiCn#jcC>55&dANg@X)Yg)~bHTvDUWx(^IXaF4e*rByrW`jYnLgJ;# zWdtY(pA3IGnhiQRJC_3EBG8-9mMsNTN`-u?E#iQA8bK##q^WXPDer`Zg>iFpf5qk} z80i>HH~DKP8_k??NM99G;5>Su`x9p_n$-HAnf}gRwY42-IO#Lac6lw6-|F7Ge+x4- zD32G=HL&Qo&Fl@D{`JW<^PwHRsbwMt)erfRq1Wtp@bqaiMJ~dU;9jAxd(c#NR`lD^ zXFzS3WGg}XAx$B~manjdC>#waGbEC`!UNJ$`%5_{Bl(cKX{K?<$0Qp5kqY;j^RW6C zU)8&r5P|3*|C;Es_`12G)&FZhF%Eb6YA^vruh@#u61jB%>jSgBUsTkt?zd&_K;C5ZAN&VlhY@J2v@42>sVLXVxH9N# zXJ;7PYQ?3c@C0l59sn_*P1a(43>rK0~kfQnEDQ=M@|^Ev0Z`St1=f@d(d!yU&~ky0uJq8|3Nc z+G1^JZ4VED6Z~V0uXq2%rxUcmi0oXChKp%X+f$7Vh$R-+$-p7R+-qZRuK{By8r?d0 zsR&d9k=kLH+Ktnj1TPP%DszQ8?NY_g2nJo&oR(FNJh&-~UMLJE# z6Z|TMurSsbyOfccmDO7SsjaKM{hXPbqT&NsOLtR=bcVYfYT8A~>^u#F*N!)^lCQ5O z;07FWM&WQZ^BPBbsKq@~#D`Mc(2#KoaAUQ#y}e&fP7ZipvwbPxXQ3OA@gG19#3dP^ zz^c+LrCh&%K0wr&Bh$k^G#=1^(M*MabDy;1 zGF+5aFYP!KPOe0mMtPVUf-!yn?>gdr#hrRh-%h`J;_FNJ8sdW^6buGCmlh0&OGuau zw!fL@Q$bZXB>FT`nvUH3@MJS5h1QG+rz`r|8~5;b9B=yUk1zoSnxmdA!Wg>zQx4nU zAq)`U)m;qyjy7vItVm#B($_ANoW7b78D54}Hs7+Ace9r);WOYqMJVEL>BzLk`;^F# zYrss;y~69)u02GEAaRvDpZF|KdfKZBYx_UH>ZB)~GDQOmr&I4uHkp_C-Do_BNr^T) zJywcCUbs&GIaqecgk&<9%yMHYsM!iyy?ggZ|H{6GB!hr6A2USIcc6~Z!-s~fSs{BrKlxq2Ct4eClQy-iETPp#+{$Xe zvxyH5N$efPcmZg$xqCivavLRua5Tvki>Mst9GZtvE!1E_r-fUAbceYZ*P=xMB4b!V zli^=x$SPx~$7^2HBh`%f)sG%M>e;qF42?3TW=VvB?jS857G^19YJ8j+x#1C3*3{Gt z;Q4!X`w%d#aTvMFkXoIP(Ha|t3)(^%sp;rqi=4F7&o|W?CrvE}jajOLx&U0J1)By# za)20L!8Bi|KA5pZmVCJFgfL8BME8duLxV8dn1sZ20fefpPX})OA7KFGhKR%91!CPr zi>d}b4WfeJq+CAy#+L8$|V1B=vxq^poHZ7nM+bE?S%k;__>xgbRPFBtjob@ULq z+vf3}^R__h)F}CbX9Gk*fZF?H10_Zy2NJZe~n}E**$4xH*P3A%+iJ z^TJOBfmlxr9~(Csc7AMnfKp5-G;+>TjJ(l@aK*bbba8J&JYWd9IunxmUoRL7@rt6# zH#7QPEd8yhSzx&I1*XI0tX&LCNh1C|~kuyUCml6s5Xm$aWS>=cN&U+{ksDTzWy=F@cKxg8I3dKo{xP zuUG_wn;>RcR6?z8EIARafE#dT3B!n~A`l=r*E@ZRdBN`!KE+s=bo<~O)B!p+JzLeD zd}3a^K(bLq5~7->+`Mei@CT+7<_JEmTr`izT2 zh$$fujrZ~m<~ zcFV}5PjfaSXT2?}gw|#4v0-J0ZPNIU%W&D_I#!mrydlgUlxfq+N+upFlYpw;7fW1npibiDC@9 zUh4d}JRvq#j_7H~W4tDWCc+FfS-JKj^DC$Ag&^AM)U%B>jEfHv#P*qaBDcLjpGc*n zU4jC?=jNLWA>EzBN*g_h`Lnpx^_rh0@qF8}yymjl=^i|8D z+t16jsr)|0;itDczL%Uvu*4=OSKnS&r-r#Ta2ndDIh4TQ?U23y@;V06YES(R$Vw); z8Vub4m5x2#uz9md=f9@TkOR;*Zle;EABldl^IuGxLYu`fy@)>{p@z8g!StGM zs&bFV#k>TYgx84NM2EBw>>=0bC6uGn^Pk^*`)?@ngcp%b?DC=ydWMa^;06=zC?4nU z2bL^d+JtDNlzaUSk=J~ym!3#a&Lw~k`Q3b@5~w@kIYT0)l9X&)w8j4O-mPeE7XN-O zkFw!DhQ^#Zf*7pO$faxTb%$32B|?ZgG&ooZPt;$C)iR1ly4MbDJ^)%Ix$XpQ7`3`! zW<&^xE6r&&S{-sfFKrU{(mafcGKwo&KskDOKJY$+nhwT9xe%2I``NZ8aihUVow(8J z7RqmOy2xYL|(u@!!uhFJhA-!Vt9@5tc|sTGS>%FJg)@B|go4YPk1&oZx;6 z#bQ4GJ<75mqr$D7tXP2Cl47HyjeoecIS)t!2%Vv$V#B!bRB*Y^b z#7R>;0$Ui=08LNaoOo1z`}|KFBKL~G)FDR)KQRmy6u-r4O^KrwAm!6+1w>8Rp1+rE zHLjOI@!vOpyevwv8-i9=R@7=_`qWDw?in(02Xj8&JQjwcGXxKip_t8HF49XWsY={$ z$mUQi?xAHN!#}KH4C$t!pz>K|08kU-80MsTNm~tHk6%iPjZelJibukK$3OzPk6it_ zlU&43h6sSD!Db-JoxVLYdi?b1ZYm*?abMqGV3-W=!d9U(pCyfbNJNm9@#ei99kpQ33kfyzk`#it+XN80fp1M@BV(DXFY4pIoDiJwQQuc zp72>ye{)0jSNf~=veJ`^n>(WP8&Hb)mu>`pDyK&2VZ}P%UmD~{+tSsBLh7TeY7 zA;^r~2=leUcPr~rcmskSSKtDQV2hPCP|}qKf~df!$Oe&@jNpWei)Y(L0G>%_0FNmk zhqjj+mxtp?{$=UYZ(oVzNS4lc^|i+T&!2#Ikp3BP%}k0`ZE4@RAksR;2uSYmrWF6j z?f)?toNJVB;8{deYuP@7wNLQJeXW~%gQ6cStuN@)e+(HKM9C8cbsF)TTb(K{ih&3a zNT!IA^zZ|Yow}Y4OV-^G;NM(G;$-V*>4?Ge`K|nOldMs3Q&5U4)NY)IqheS$mv{u` zSGIP}1(-VK{*II7{nEe|Gx4yaT2VyB8z*$BmwVa#;1i(J4FLfUd!$IT#Mcv8nQ4Np z*qvEk2-LwmMt?cW@UQx)Z%o9btM$;74>OuAJ|n~?b9Jw0ZnrGtt1Q2BdVS-3)9)7- z&GsV>Uae=&X8xgMAMSR+H)EHoLH`BFvhksdT9LYmha%zw(j%p(7l6I`&$cya$;*Bz zec8R%h=ZWKCs=H6ZiN$C40i9g13l}l8X)l;CMHJ`M#CWjl%)jb@GOm`&ob4YCbBWu z`56Oc#lz25T)gK}T33(~eLN8*J685YG}(Im9Z8t0R2JqglN9Y4vfaYUNRD~T`@wIo z(rv;zRjIhx_4}9o?*aW7&CnwuV^_U$HJN~WbxGO9;@ct9_|QZMuHj?Hjy?Kje3yD% zX*Git9b{x3%F7;G=TmnI5a+foyVg}g#;8V&5^h+Wc=qfBRKOD|UR$Ie&8;M^Qx7gj zb*=2r42#Z}o#wSs+ytph&o}|m8aYMXRqYBQlHm}bX4Y}`nzhDA1R&kYlsAkoJ6`9q z*_0}Co|sL6vm=7LZ#-F8SSa6&(O!Sls$2Jf+zssWD{_XEU1cnUxIImR#y}~Q(Z9PT z6_I6+3c1E}84Z2ynN)`?vwGQSiqnH@NZa2yfz*EGwbe*zX17tY>(juWx$!f4QH17KohoEH0XfE@VGe_{oE zl}zOumMG&tT)V3_qiRYh`|5%6L8F=2Xoko{vR#s2%cAU4MT6*gU2jIcl}YEm$m##6 zwo(JdeoVC(WGs5U-|`YP&-qUZ^^L|9=R5d5y*=Ex0dod=BGXL8RcRf!`>GZ20@$jV z71+RaGOuY5f6fx~?CO_mpZywT!(LLUJO6X-w|mH8$F|Zkg(lJ+9rWdwUrwzv{H*%$ zEEykQTCvz2hfv0yA70MTzU9Ap9gF?DhLy4b%nP2Mx_|x?|1^9vrTny#)`Yc;CPRhR zzg6n>Nr3WdLyj0ajLs84af_8t-XRN(@P5#b>TIA<0uCg@=S-heyX=$BgJ8hmcCxYs z$m$B~R{z_X-$9A0p`)avDj-Y$uS6AC*3GXE9t(9>W}{CAE|_Z4D>rRQFiI5O{doR#E@ZIhDFrsg2etFV4NtN<+WF zX)S_Lw7*^5u78e7A=+YcLIZNkJKeu9oZh>rG)HQ81%}QmE_V|&jF}Z?APCDhf#G^K zsB73Acl4-l$!X+V4J_<6_Sy2r7(x1vi-DpJ3iyGdY|t(5M7PXe0Gn=vdV()vpXuxr zPGxVu5rx;pj&%rD`~yU5CQWu!)I|_2(R8n0iymbudBzPgTyI-;tF3jsNTM!vD_{+v zMDFHO@A~7dww7EW?5xLU1TtBbQDsY>x*shfDA?u8X!0f=&3%}a)rOP&ZGOIVIFcW| zdNl$JOED#2-HM&%j#V#Rq;CBohZ8xC_D|k9cL1O(?y{yv944GxB+E>G5E0G7E0q-s zaK(TTo#^P;`+&4W0rZez310vfOX$Q<3IQ9W=cp>ps#jlAYfWWCy%nYRJZvgHL7JDM z8j6}&x&(T^Qe?)2D(!a9*jk%mX21&NN>9#HyUQ2HS<)IBfqFUVz=)QmU?4yrC5>_M z;>8?NW(T69owyR+aN5Y=nnf%L@S-^mT_1GIv_loABUbdS@N#QxTwD@v(7Bn3-_iym z2{qcG?L|Kt`hHd#UQbt)o=`IZP+O=T0rHlMWwsbF5X}igM!J5)?bk!W%+8JTlr`ve zC=BV0AUDlMA*sNXN7ENtw`{q8(wYzeV0?>89kAt?4BRIeRQy9?ALQg@Tq1JK5XlUe zEj~CVGD@`wLB@8Szr*0&zVwkDS_eG=AwL;Uu%)UjP&P3mME8siiNP`LD~*LgXK)DT z*%0v>h?e4VoSmJ?{`SC*v;oy{aQ5jRl61<~$7eW~NleOW+Y7n@`I{VT5dX?Cp=($B zNk{Lbo}X$l`rCX8myt%fi9wCs&VTXx_0)}i<-Fe_9$aC;hR_1R+!!rYgm3qDe*u04 zckPYyu2R8O(-LcN{89q1-@ku9Xx^B~b3sSBbe`H?>^2aM4V1i?EbjYs&u$a%tw3$R z5t%1Mt}I^Q=_snOY)NL3`$B?WP6#hrgWR~O)JJ@8tZ{~Oh7#hWSQ$O-M1K}qS5k=cAJdI|8phY?~TKLXK z98k_*OUEF6H{oWeVg-bx4?kqAfuBF7Ehcl@?%rb;KQ4~^Gc)r5?kpxwhZryg4p#jV zNYTZ?g6leEC$0Rj2qw;wuoPOm#93KGhFA!0NreXQOpp1ycKpD4MN6%NZU?jxc+ZUQvL#<2*Bgc5Sg6EZ3Y_|ix;qHbebhyXkq$aiX z&yh1Qr@GVeY5oGyc(d~+_^m5qB>Y{ z_QkvPWWlsz#R$}ckRsW)a(b7ltYM5VgGN)_C;#NdU7YCUbw4$xT*jvy0#SX@+4k_N z8d`ZuqGM@Jr%_R$7q>=DBPE1kz_qI%ds}Mu*txITyQQHbZw;NVb5|?}Nm>b&fFc>k z!FXjP!jN%18t$L+>c1yhX-x6-vy#I=WN)?udJ%%sEPzukUr%3;TujG>}gV&|(CQ#jIQjo=p1|w-R(NNmHsPZKI=k6WE;mVEJEAtZ|c&mTjbL;YH zY~T8|S(ltJs_JsvH!kbfKs|FZL%VR_7L3vS=R%-yg3LqyB~+qhe8T=@tKEb{S9CR=B* zOLOPXKkn%4iBfE{QeEv}4WZ~#`eY0QZs%4S@Vetnz)8e+vA@X55&GJ8d~kIo~n#n86ffu&@Gc? z1)>3W&sV6>l5eaqVx3(*$*Ir6qVHLKhmvE#|1}d%9+W85IFP!M!AW44T;#Fk>Lhu6 zDQrX9amf9N5)*mElgPAbD0=R{pOa0dA_;6L$XJsG=Q1-UZ*?lG4wp_+}OC zTj~1VD+o6bAYvV652Y87%txIHOnlC^_lb+xgL7JQD}mRTEtjUvuOVrTLl;g)K%i?hlc=4oZp(L;&KD&?e>G+j>VbW)ZGRUQ8fqsEO9^2I zBtURKLpncUq5vU z{ND(!KRN#25?l`%;>>r@R4fvkh)w$;RuM`FdLff;hqkI`npg0Ox$Ri(I{99_cwsR& z@rO+|E-qmNjzLCR_m~WH?cJ%A>z7ueZDXm1lqkMvym<7-V(llVEte)}D46t&@}skS z_hPXW1ZZh4r|v^e#At`r9gqF(#BCI^^rEN(Mmpt5 zUdH5FV2-dIbg|pWsK$4Zg0>iCPKKTV0aQimmQs6BL}VhZ4q5}q^AK$hhfIXGY%ZA- z)t?mGvu6Eu`Rj}?PPcq9@ejt5BugBDR|*ua700H;aCE~RKrW-Bn}8;_&o+>h_6`mn zI=a9<1SI>>JfFPyNBS!y?vOJmzw~N)f<}6s(xy}!15l~V`fZQ`u_7q#B&4EN+naD^ zABC@YO>U(}T#Os~LK%d}pq?oUGT??WbN1Tx|5T+NdHc-sajq+vJ!N$hZo$cHUu584 zzW-3E+VL>0*0-P`^YHLA{N&fo%rAa&V!_o$2(&@OlzK6Xr^UIdJj-Jh;nu4M%TC2=S{lW31;FzB(UuhDixVY7dPwHq5I5 zksfBa4Tk2N3%=8S^5jPN`}?JTA-yE|$6752MD$Odv{LIrsbCvvhEB?@gZba&WoO17 zT5YgQiQUFBtVWt4_Ae)3-;k>huS^NJ;pkPR>$R#7U~jv2x^?Wp);=-!w&wgWsBWe7 z_}>w^ccpI2K1Rl_VI_WOJyfiq=Mb4wc%4sAM3yMuE=ggvB-~pjeZRZg>#s)9A44)e z(8!Fwdro_6lJOHV)wz~5>fpYPgiH2G69)El zNqBI@whD(Cot;&u9X-%KFGA$9SF$uQ5X{KefGk`*BF7!CDq%!#XUOTbFYgU>{tSY-l|p5sBzIB*Di3n6`FG0 z?NQ#dr_&o7WtgwjrF-|=`FmzJNgIVyp2Ax8|H%jH`&rNlWuoKSw|}wXW0o|y@~xQA z;NayyS=H}b^=RZTr&dKSiJ5nN$UA=+x!LswG~N9YEpn;tU$)Kl_lN;92xQ!<(<_M! znInLNMA8U=Q+24}mO_m`^Y1^c>i4g*BkxknUGgN2xV+#G%Y;75Jf@!5(Rp1)Ve!*@ z^PM;NrWe0_Q>Du1(?9+*aCfUlFLq2$yf`>M`D#0t&LI)e#}_Y)NRIAy^+-F9#ADOi z4V%5sWADi5(IK-ZHQ&2+-;{3})n2o^_0KyVeLi{1+rkH4lfLTv{tsJkq`Z2$W!aUk zDGxe#&TxLU^ZEBDyZua7IUTez$KsvD&kyGeZ^;%H8@u422Om&iSzO$`1S&ootUYp*&q#T< z1Cb?3a|Jkr8vd}~?Vmixb5Me%P;UrS6&?>BV;|y2kMCtXDTJ-jn#P0YP?r*l^~^U5~xr4O>|;G2Eq>)lHH$2}hliERsL_x%G&Bw_8*R3hFb=RcH=p-u3G? z0&QgtNoDz3Jp?kjKmKr_@BFr=#9RgTf(1psY5Vr{beMN^M5nvT`PmY~L)KM5z-?@A z+RKXq(DBL-85NcWvN_K({LJa1@$4UeJhpN^i?VU2&It| znvhQmc?w@J+LU>6o2_b>KPN{|#L?h#wzwVCd)g@~-Ld>VEY)}hS@4d-Cr$dR=JCo* zj3GxW9XUlYh)!)FD2_X1O*1>!&K?7+?0)&li(eB$j~1$o>oxyq#4ma5gx6SdBl&!? zyx>?u9q`1{3?c0A5+B#pBaVcyYUK0OFKZdqe+%D_;=3^TxM6O;$7NaopILwiI1y|% zL$wr_npPX2U5FHYd-F&zJqo)XHu*4-#)ij@CFhO-E1QaJ1UjykQK_O>Yv+EYI z^od-YVKxpbgwd^IgUbI;@L;MBYA?+PknK2NqZQ{MwEl0RV%^^IcLvg1(k~#7x~r z4t{@cBF$D$D`I4KY_Hl~=>R;-;VL&oqIH5!XZ&?K(&S45UVQ#g;06rsp_Y=u8q6|?! zT6dZd7X}E5BiBu&^;3I)<2pa5Yns(*S{OoDBl1T_v=Thaa`s)HWjT+50@d=W#ziDuHON1bjM?AWfcX(~lXD^;%+QexcG)ry{!}rc;KFE!FY1gPF z<+RAity`e!QpBXjoqNylxc})DJ(*8?11n=q?8_qDMpn$)oUo%iu#+%7Rc*?2Xb4ez zHhL%Bsaxr1#&&c=mzFO>2cL)P9MRY=;W=Uh*REab@zI=?L$(D9Pr{l`v+2N|3DyTV zUDanLd(L)2;a(AbZ39=VKMHfLd8}=ijP>apr=%LkMmoEWc=#-OnumYKaCMR9XKke2 zFCEtvTFvO3Y!^8pHY{uYUPy=~7)yq|N48l?_=d&iu9^=JZ6F7#Z6~&JJpQ zkyrwPyv3)_gi2Pm1rWUXp{knH%u4`)uoDuZ5}D;WC0A4uG26L7!wfU+v<*tPl( zH-73pj*FDz?`xJ`FT2w*n4icBl@^n{_K$)QaD_3Ug+ol^qvPV$T`=`=5YhH%yULIm zSPwqG^*9u$S`E9}dU6UO00-Pc8Tc8i2BO$x?=Ms19iSU^|Q> z3|ck;M${a7IdzKDvSU82K9!$JLfQ`0WA8b%uEEW%LhNaK^;$HylFuT#6355B?(u9N zED`i*mFZw&lN7(;AieA-lxLz6?VeOAo_V3-8o{sjo{z4Lh#7Mg*IAYhN2~Fryqgn= zF!kFfq%rM?&}JZI`_$t-j>?OdF?s<5R@u^DCCequhc(s+MMqIeC*lF1*(q9lFIfko&i`{_D0s}v@A$rv&N$-b!PCOPQ9L@IN zP@aCjfHip7S2xIj{B)(ez^im~X^^J)+xeTD)vmqS!`*6Z2XrIDAuQBJIDa$7c8luo zpesgdNfa@6Hpy96r1;s~)UaWkG>O#B!fVsaJh`<=kbS0e-4_mdD>r}mfn8qxNlw-1 z#(4#L+hqNv3ESfXLPH0G*vk4Avbeo-)NY99o=H!$vVytc2iUPvcM`F^Hs3gBu=#A> zi}x1!->7`)ED+FXrL%CwA*s-4#KOv`060|H7H)PGp@GdbtP~)-;o;$MkPd z&^VC31?;RWlmU|vDt6{wVb9_K&qW6UYQ$S^zTM5|z3AD3f0?;ua`nsjT`KjqjyA3D z5=evbocZoTaD&C5xnLauqG>k)u$j`D`FWvW>iglJ!E9vm!bXlpJyg>&up0&|aNB4R z;9cYegDVGlku!TilNZ2!a=xBq@U|#%&@*%~s-^?ni_GR|>GI&&M5fON0fK9s)y6bH z4jB^gLFCMINDL<2@eeF888-MpO#5$tDBrjLyACLm;H#SbL(Gtsb(;8DXD#)H=Fxq) z4^FKNQF~tAE0K!UyM3y-(2GYJKg*9`CPe?K!^eh<=&wrLCi|9$)=iP^U)nbtfsk$? z$Uw09)VZ!9V_x5~pCRntpbq0uYHcMRg5PhwQrTeRHK)I+n`rkBZ5A{5nU#g+O=Q zPS2W~e9=!gA(>B$`N^q(P^RP8I)B)G7us)nIUqUNS!(&%U;rt4&SM_dox{ddW<<+E zXHrZNm_Z0ftLd7WX-Yf4y063!=3;WfSRN}t>JESN=$?cLjUDBYl{l+{RX^4D98h{Y zz3Y%g-N3Z9i4~{s4L0jj=Nn0*;^swg?l-q05tZ&QGY04I8nLKu`PJYu-e65lLUpNx zNG;P&{2EW;nuxH0=Hq0uxeVDG1RhkvRuM8VP`0IlE#jZy1V8$(t>*Zi~6h91U z=7=-KKosGcgq?uL;WV?-HM*=bs5%tfK;x|E>0z{d@uE05xDjM?#z4BEa7nn4fM4LW zE@iJsJT_%S?|(8DN6Hm(14H1(_qV4FSa$G6wcTjk5F-CWS`GE9maK-;O&GXn%b8#d z1Rj{e%9s`Y#8P@m!WIj<(fq+25l6@O=Os1EUP7kUJMZ@R$ZScu(2KC#anG}ZiTjn1 zP+~F(U0rgo`2!C^Hgco`tIf4=;XRTqH}1`meWeaD?1? z^@e6fd)id%z1~8XO2yKrGDY@7e1%Oj4|jM*+XQuU{XZRr!eWVxFIZD6O=kDK*85GD zi-%DWnwOYqE*kMRaaWNUThHnyCw?xeby7j61%9~6ANuqADj9PS*Q>&(a>he{!a15aGT^4RM3D_sda0Tab{m>q!G zS3em&00-xu0;iLgib=GK;ORPKIvR($R|{wXXj1B#x#K4I4j%mVt!!y;&`42^>|1je zO28@VmzZs33kdgDg1urq$N}1=>!Z~ZOyZ{4BZvE#y5E_XuI4!M= z1VSZ%ZLkJ&_26SqmIk67BY_gs%%Rpmql^#bhQl~zhX_B`Kj+tOyekeII%H$=8HP+U zH!}-7lcl#PyK*Pe1oI`af&~GmgXG~g>F9}6Ptb7rpW$zmiF4B7cF*aHN3?Qq(V|5I z4m30@@_6L_+;mhLqDCTkx=FVnUR7;5725#XE57x4_u^|6ETry|N*{rX56-1A78e6gc$ z)|vR8!KN&*r>v6NmxV8Bil{IR$cvgrk1ERIK6q;VG(5&&*k8Kqqr&AaL9eAcgq19E zu*tuNbWE*(j5ks7Mh12_eJ{@US4&Til;(&SNC$(7NqOzx*SRs<#uO63fvrj(3c1tz z(gTUVYFYY;{B~%m(<{H-QKR(G&%JkeZMYAK6Y#&g-6n}2cdB`c*16vZuGIsa1?`k)t6Np_i;uYzpf3b%D*qV z#kf(^h!0Zyjho(}XOx#FejdPQ!354svPqTXt+;YcBY`1N^RBwcQYDn?-pO>arz;6$ zQyhQ`@QjzG#(v8}D}bfYBSnR8?jK}>y3r6UV?47=$-hBPg(V5yzTr*|+qbY&Dtne+ z4tPR8;Wef-baYfU)+NNUY{5ttcOx|SUF}J|i{HJHt)Llrj$nA8Q@0Ix+8b<6m29S# z!(kgxbV_RzIS~>wMbc#2DqNC86rNw*Fhp8uxM^b`KoiIKD9|*|wb8uCb??&mutD|9 zJoikn_=t7|XB!vkK)-nZ{;A}GYycFXN;AMvlTmnA;@7CQo92bo>~qb7L0`1A#2LDKkPLX}MMrmGlzALS*^}Uc9M>?3 zvgBe2<`bpSGPo}tND)M{`@52WTuoLhBx{LOzozU{U>RzGw9hYWr#8axsa@u{K|ix> zI^!d6SoB%R>T57qY%&ArO`Wo=YMF0>$;H|EHB;k)Xe2QPio(HEYX0Je)%noe z#U;1tb9Mu8`?iFgz0IvyK_+o7pm0UYStNtV{u-qH?kSZHY3RL-3tGVUN&;}TrX+H@ zTpIe!tUz2eXwifegzP~~H}8*MT24)=&C!Ul@_=46I~v@Vm3*cF98UuSOdaql59s%; zNw3h8h%eD^MlOhj&SRGAPSmjV<5clGYqF5RFm^I3MEPaz2e~kv^U_$-2XuRHqktPh z%D|9;7z45Y)xerr2UHnsE_p7uhXYrrQ_^(M)4Z?;mOltAeO4Ik@Ia2^;GBGx!7)tc zdSSQD$lQ{rGHW^lIjI#XH;sW*&r<-ZsO=+WOsvf@CU-ZZY*)F*`0t-p5^vaDVkCaL zntA~N-@X?05eP&Z=6JXtUedF|LywT#spK@xVbmNiL|?PQKjMN*lGDD;3Xp)Qk0=N+ zukjXBcl2T)JYK9OX6|}w{>w(x-ZFPr@nu+4lA&K8NKGvhd(FqD1~T4a00)rtm0G)d z{>qR&ecFJB=Z;sWGQSBUGUYM$>ju+=AcJAmuVnJYuPloL)HSh5kxP{vHF1vPoum`2 z+uU0thTSpce2o554Q4QWYI;23orRjj$4j{Gxgj(TU4eiZxo&#iA?;P~3msHfS8q&8 zI9uH^lvp&{NSdlY&j#`}ef^B{f0g7ri%fVMg~D)5?yfUCgLdohD$iOqAV5L`%;*6F zG|0&0QJ2(r{B7W8!tH*x-{>LAD%q`xZe4HHCY7Ld3=-V+r)!+6nx=63er03H=mm@sDA~*ovvU%-4h^{jbQ)ZmG zQ!H9_@u?jZhI z7>6uwaP+V<^#)G+*Gbb#DoFwYe8#Zoy)UnwWpNYwjhZv>X!U6R0 z^3P>ALRDu&14Z1RAIxgkuFYbzGQy2PB~R85&;t*T{DG{sFJhICKuB% z#0cr8{yk=#j|`IrYX>=)h9HUAM!ry#A-N=X^M) z;hJU@Vr45t8ogorc8}3?rwx({9p{{H!hTF#xw4A-2T?tYj|f=A*O2$gz3K7o)NSPK z(xG_oaol({enFE<5iF0${3ed5B!cP|`nizsYj?CXG_(eq&T|xXUl%|INQ1P)UtIX+ zboLG|;Zih$1;5aIr;KXXqBA&c`5 z7T;uZA3QK%vW|Vkw= zb*je)IX66Y+F`%|iDm>;0N}Ay_sgP`!yf}6zUO*DP*qA3SGSxMi&9?l zXN~5+AIs*{2ye|wOFIH&odpXt5?D7oSAL87ezeN95F?#X@DiUSoraY6uAE)Dj(cVh zf|i-4On+y{4KKQ(`~hv8!B1*#$`& z{mIl!xsQL9J!2;4k-A#|!H>2nw+~5x5L3>5a~NFU@{`ZcdAspKjp4NOZiOGt%FBr{4tx?5!92-BH&7I8=oTPmYU>+O$B`&*k&`8E9m>h)Q!{39~ zf%z*H|K%SIhpL>0(B&df5zknI11Qy)3RmXHgmY=O@W-S%km(>NHs5#Lb3-T ze3CHk+_}KM06Fn+fI`shgM_G7geP(?o3UeetBVIrj77zjz~a*!^DSOsbcx>NA`yso zDg>4b?=)vh>ahdr!V$+MQJas?fosdoO!=f_sr}b|@Gh@@B}x7mNjQJ=j}?Eb9OlPQ zHR8@Z{IHxC{)_6~>Y93Z_(D)ekqtZv&4rA}>XjBcs&n9uCJV~60Jy>1>O%SM{l5NM zoh0ZhmQ`AKb<$B@tH{KdETHbx1(B}Tkj{;`<|6hVSu5StX$OYd?|G9}0KQ7n7Klecmyj2c>!+C&|u@{fb?y-s~ zyWKsXyoPTCM6e>LhvjZR1nK1vU8Fb+^*&+mL~;wlff!nlBU*r3k(P+#RJ~7^%Bs(P zR_NmK$ovhSS3|nEEF$wD9Uz*DINo#f(7a|=41^3yDl88)i^?ega7~_S8tug{HhIdD z=I~K$jxe7zRswASaxJr;RvGg@N5?JbIHoo^DKxBl1Uu6a_Zv8}@wt0dch3z(JpqJ(j`^^Zs1-ErC3Q`pJ#ZA+|N=?khw)shsVa zCdEV27mUt{Ok;{)$h?Eh^S#CFnjQV1u*iOLT$bMJp53>=V{Ur;d*RXDhRWQI4jJ=y z*dyAaU(WRP0T$9IC*e1Hs?wAH4mxcko3WMUa+`)jZ)XC z8nS6`M3W)D8WXC3T%Xv8nB-UrFv&UP%vgWDOhE zjh+onUPcflRS8EB8GsunDSEce@^0@m{=8j}XS7$yb)F#QPy$At#o zXaIu~FtY|SrPS%z*$FIdDYr>q0j`u5hM1eQHcV7l6X!~BmTI!s?4h8Qa**LkXMgXk z=FYcol>(!&aiA_Ve?KyL#hNwOeEs@-OwS;X-A-N6GN>cw5z(K3XGi*?dyqszo)YBF zWH}B9BGg!^C`ObY`q?!0d?;c$uQqw3+c81w6dps8kn(Yw=pyHL%315^?w!t1sc7@c zMiY)FQ56IS<;c{>Jh!RMnm5lRwF{qFAS#uw<=`}_;glC}-sU9!R84v}jrL4?8ODLa zW}md(xhZc&!6>V=1#e>RGdt&X@bl-AB0;okM{3dungC-*ZNGKML zhT#FCX=HTF+>H#?YuQ(x(_=w=xt!LYf8Lh^V#bzBWbNLZBthMh7lhCYc|(%#171&g z8m4hwb#l8{Fy;oeBqhwR zZQ|9{MsDDN_hr}g{n6lrDJC?AgE}ZFvgX2r;l7*PK_VK{S`xV~GRP7^82t=3BsXeT zqchqAITwbVyJU8J<;oT7?Ao){Y{ zc0~Qr-dHa9c|jM2o`oeZa2fBfRt^mZ`5 z)%*Oj8~IUlRFg#ZEV4(XDX;0H(!g_0jr@m=i(65R5db~7Cr&57KcQ+_MgGaGT9nf) zLw8Oc34Z%9aRuR@!~DA%LQu>TLj3lcsdO5-MMPj^6RG|t!no5T^O@ioGE&yI22!-U zdG1D|Z?T4Btj@9T-%e+GR>l#pDzYyZM*(4qQh9Su9kPO+b6)J027~?EhP6e65Pm?D zBs)I1(0UB#DQF?>SssPAQ}Rk4R7pqUULH}}&R5KLK>w631N{(ST<2NGygePCXD_d5 z&H@ZK%?DXC5!fei#i@n4HD5HM5sXlul+FMuY}`h5xbRiZsBfphW~Tq%luNg5n60(l13n-Z>xgc4pqwQ4PPGgQ201MjNnwd>OG&?>Jgw zq#^@WJ+{gl+3B4xmNhQNEB>B*I%?b;=T2L~=S)M_-X`w6%e+GODohiTcl3@&#YdOs z@95Bd>GD&X2L!BpeC{j9`_Er|-@RIPkKDbFqY28s_aAq)uV6QxwW#92Gv(ghyRYf% z$to^-UM#~qVn8M(#Dk(yJ}9E;x45=GM&jQ1*mAir+jax=xFcRVr6C zwP|$08>UT~{H^Ra;Z7wK1#jckIzC3rl|(ggz2_tkeqHDO{b!r_u+N5n3B0kZ2*>!_heIMwsL@J(Iqb^y7W+R9M8Q3AiKIc#2bL8qu>ViC?o7uA=lR52WCJb>6iD13S(-XlNyRPOo}bfP#l?u_rGk*;C;Y)=L{ zm^v0LEJmU_uwT7!?;;DR(bPvAVC#uja+hQeDp*ph7S^0RdK!HM!em)-ii44lKyY@c zYnw7xp6K>@!&5Ho-t%l>f5?Xevrh0PD%&7tsO?hwy-by8ibKD-%hTfRm^0w+lEp!P z3lKaPICRWQkE!HqYMKNLLC9=7vEL~VoHus2X_gp9mw6b4z`Wj#!gtE-WsT zZ}VKgcg3MO9p}V(S4n>5p7}>N^S9e)S4qz6^zPJy?#SY?7T*WuVm-rfv zYm(oe+GC&w;QdEOiFw?*uqm4!fvvdVT;Ozc(khAq&t~7TsntM<4CBc8gFlU)|F9O? z8nQ3-%xgv+R{NT?ql^wyn-uFNUM`ImFO4m(=`zhshTzhIIy--VuF$zERgB14@Mfuc zWcvVcKMKp(AfNiAFjO2*#Q!jGToV?Y-h3TnykGcicGvVx>()K!IHn-xlTj^gf%jtf z)wa9Um|~Mt2xPO!qJai0AkgPLua|q(34_>xhL!u?1G3H}ThatO9uPgIfgY)2&Q-b= z`dR3x4(t#pugj;BT|<*ECAaT(X);%8?w90H%lS0zFu|AOOkNs5D|^#p zR;^08u%bkOb>`uN&$+`>pBK_kF~_yRPh7y@7N1>PR{n6_Oq_lW)_CDs`8`+X+VqBS ztR9@VojvbIpyY%~B>hx}d)Fy9f3Kx!2&e+Hwqbt>G`0A{qMjvYK z5FYz9$X*q8)-5X~m{bfh2G!}r2B1)pe%LP4@n(``#Z^AVVN_bk#D?d1CU4!8U63t9 z9CU{_X*1@TjTnF0qw4HabcCu2+xFmCQl>(NOJa!`S^Xekm^^Ooy@Tc9_9$pH7fL-0 zIol^}e)#yYQeK(;sviL{J9ro)FU?89;YIHJ*7MsPeXjtc&pPSi=s3^@+PqbT*r@gl zmmF%;tVE%U4iT#&Cg-4)m|FOQdGNL2U(2f)cX{$w?7V|X(8+fC=RdYAK0uo1lDh!q zsqz<9R#I}*c)=2PyHc=_dPtoLmw^SQr%jlNBr12-mm%hU_fC$BN`qElu`TWfhTeTE zq13IulaXf!_!E589EZoiJ&G1c1ODo9S#l%t%WeDGS&Bwxj!Ak z*dulQSx!YOqk0~&dHsf^>Tq{|#IO-V#vI7UbT`ZCSk)Q2=|TYcq6oVkcowcnH8q-L2v4gY*%XqHDm{?q^0o$o)xt_B=c-E9ZZ$gPq3!l zTY0Q01f5N5FXQ*2yZVIT7vPb#hn|^G5>9}MF-iIC1-_&TYfvdF$f?C6W0@a&H*@h= z_9fuB(o{8=xx2eZWRr>Z({`6!owTOoh+fb#D;L!sy61msz-So{v~QZ2fQl~r!2X># zaWayNGhLqmI2tskxnBS!Y85gWdYVo7Z#QAkgiC&nl;;K(O3gJn_`tCt023^1NNF z=v3Sox@!Lt${tUOx3c~`IR=Z%53znVf0;J%Dj(PK*m!sASLE8Dpz&R zgNNkUrYtBVMP?nVU6tI?JR(2^TZvN-vM_J^uMT&en!t2c84NnpTuO0utyTqQVk7%; z6!ciB(s1NHc;QU9e^s|jr0^9=`Z0dF8`ZXv(^IM$4%V6A`sQuhmPi3=e)8hC*0Q@- z`2ljhSE9wCKaEk(#*-_eeu;08Ve3I+>~j{#aXsxC!fiQ^%Z8tX)?w5eS^o5T9y{Xk zHilU<#p7~4uj=G9R1YC-*RVW*1p-&ZJ-{SNZRG$<36Tk3GZ2suw7(N+vF59tjv`x# zG3E=m=+~$?WsGkR5_fqQ3SJG>io|?2PO8QH#JuqT*LKB?S0ULqy=RvMI;N2{v)W) zigV+(PlfL8vJ(8p^Io()qW)VzuHa!KLvnZrr09dx{mlRXGG#m@HbNeZ^77nC(x~$m zDu*uR_;3%Z9BLZt9}ec;AGEMNO({N>*fJ$s&LN-MIE{{(;Xnl{_R^Yn?06{M$(js7 zp7m;VIq4Up-l?3y=?Lm?ssn1m#kKXWeT78RuWs9yTkNomvp5M*Qct^)wTt zS?0Kq%HQU!^03(6RStAFNdU8B8+-g*y?}94lD16YL6_=9haY4Ba1cHt4{HXX(8PqPfLOyTKGZ8q=PE=#ZM#yemfl6wDC$dEzO3ysnN zaUTJZ4i;>V0h(*xo9n6J8m*~9nwt@*B;)+>M1r2FjfQ6uOc;N+dJt*K(cD_?+qSiX zFZVb9(M)b* z*{WZ3N}`Z_4dg1AqwCPNz6WT77A-6;uH}R#&cnt%F<@%PS@7nG|x)SC9FzKI-DAsq{Ud7$u9wZvJNgR48fOS zYP?f~``!;4e{UmIaEQ5AZ}+pHNGy(x3$De-$9o2Rk4z_wcrkkWGTd3*nPj>mU>_yq z>*Cun&tC*xepmSL&ER>}KOPuVF0l1i)^6nj175djx_p((wx55uKfF71McX6K9ESHD z?mWEQS8MAn|7P0;!-_8br@UA3Oy4$UP2O*@^=;w0F_S*)*Yd%YwQsVcqrQB6<;sjc z>wUiHJbUNcFZpivq~&(|hc-+K8|}dQnw|9w5zD9FRChIWtP{6(W&i5%y!<=fu6p|S zGEM9;W1pgPK_`eN^{oN@A3yd&j(Hk+Ma;35$aebNjcx32kuQQ!gz#fn2%q|+JAp&m zTaFi|wD0RzSdl$1|2*h{i`T9_-{mc{lV9QH|o8ad?xK&hlJ#%D0#n@OX49gQE| zrN(BK5&*7rS|dYns~aV0#mkqiBbPj}?p)g0Qmgd;cN(9Z83ajP@!^tWM-wrh^0H3D z7k{O3bO8O0eDY==2Za>tJVR;W9Q>QRAH5?_3dL`!2gy*DJ6~W;A|bKm3ENR$Y@c@6 z`{f_~g+&95w60-c7i42{mP9bMV8p{7m5Ktl3*7eXANb1^+IT<^!gDqJ)O`u~qzL09Iy*f$yL7LL{LKxO| zZ)jtKvYqGG4I{c5@q;fw&PnjV&`F;?_svCW089F&TycwV3KJwusxr|s8Q2@3Wyh>4 zDSiJh6%5b%#a^#h#_WIl(VSN@{G(zaM?SfC8_CJjqrPAF>u0^YpC-)Y%j8Z@dU@kD zri&u4u>7H;NA64&#!b39$>p(9BlVRzPpRXx_l@p&_al;c_I3^T??i^6 zD-Wivtoec>UwtmZ&Zf; zssPe&f2Yv;H?u+JW!-nGn z@4yO#aaFb7&fD1xU!j#6a{|g^t=UaMEq8r?Gefeol2)9%O>j1ljXhy^tYw}Y8_;!C zySFCy=H5Zl?9b#}v9hHsmtywrx<2Q5X_Q5YTDO`Ce(4G5j!Ro3(HkuA^xJFSaCjur zh1=YULfT~c&m1^!dtSNLq%mF8D}kJt7^0B}PcH!DXjUcQO&!Nk!zFJgLN{%F@EKLi z{l7oD#_L)wFbT&OmBU~|dI3=3gauQ%DotQ%?-D&>?8M?3oDvk4mrfQ_;RG|Pt)fXy z7+)fqtf++Fb-~+-Td#)iXu7OS3ozH~>hY;}ewD2hmH=IjwP!`~vF4pSk1%xLSiN|*8~?Yc z{G5y*K9C&rA}T2(a6xDlct86+mgc0ULHR!A!{vb%F6b_G$0u zqM4Bu4>Dr4Q#tp2d=D3~kEHf&pl}50qh!k3DEDqPS2b_G_nQRzSbNgAkaE%zqfq#) zjTcYohJIh7H~v5a!93_)(HLZtEvJcD;&85!Jy!+L!gmiF=M}0UisG_e@(fd2S4cO$DAC)jt@%iY6fdOfJu49 zBTh_zWM)M^W%^8b()-C#iixN4-dpXP5VT`g1PZKEmDJ z%9c!3NGm4umnI#(_j(S30F7d%W^HZNX2iag^xraob=SDC`nmZE9e(md!#lWsqsJ(sv^R7nSs6)3#p1wXZytD$w`kR)WHWsB64fx6|7+Uv($-+z zW762kQU0)F_`K1aN796WPT109Y$g&}F-jtk=$X9D27^OMSn{4hI&Yeo0AUQ(5bL$Rm7SR2W*>b05>7B)3XHLI3u#;=?o2~5*Dt^ zSIgzuSQ~*%DO1~04QFpV?#HeW088?o{);WQ*QQ^o0=*U@G|>>EI1x65kw&?`%GW(7 zyH)((I(T0%yXJh9^+%=$>RtOo`dyJCA@PWyIa^c_4^^G$WWl7Ha z(hqgFKP~eKKThFo_Wpe5NlB{@|8ab&=?U?>|I@|t|6xN<_yAKAC(r9CkOon=`Sz>( zc$D-6Tzz6RZ1}XSmFPIrjnro$W|bDK<}!YhvH)mn@b${Oe}^B?vovJX3IsN1+K`Ho zoCo1%QUrj-h(m?Coiuid!MeQZU?Zw+r~RtAa0?jKb!IS?CehgFNcc&knJkN>P{1n1 zZNdS=v*~4*(oJ0%!W{x*H&r2m-2xVr8?E2Fmlr6fH=3ynhGU}R5j(@|9{+yUA@pOI z-pw{jL>x!6priwZe}h;}Ay7^Cy*fHy7X9)xY7+5YR6$uE*4@uJtzgO1-h8{V&K|&~ zrYRnGqYk6L9)EuqDLec4*}zo>YdYgicVJO{=%u|#hOCXG9$};ce+Q?qR}Ntq`?-~* zC>#;b|ML}V0i0R|`f-nPJnMx2KGhKRh1jI&W+we+jx3^B`^H0o(rMYMVFIuF-*X`5X5NHa*Mki}T{ElM?y3oPuHW zH4IrZ+`ChxT1JqiIl7lOl-&)Fe~_u77BIIGLXYOmYu-w5s9=s6V<40U16B5PA5_6+ z`M>Xb7HM-%8Zg+gH;Hkh*7VNX;@1mpn8o&-NRR(|-;jZW2FW=Ke+;OwYJhk4t5 z{A+FGP&mNrYYgJCbSt}%klQfGEON60$WACWV;&tQ&uPSrrGf=#ww^I~hx~gRPER@t zCYpNk{&BDSX^oBo%|Z|u&E%flwQFwVg$g#&|6c7G%Cpxr*$grwm)nObAw6#ksvA+- z&8?tfL1FHBuBo(pW;4?XPG0xpj}Me;qS0PH@h&3;?;OX0bkx+Ur#+jL{m5YE%F0;F zYHr*NugN|U@ya3I+?}z;EK}GE$M4})Rd>Js_vf!jNNCC-3C~E|R(MWbf5|Rr*K?3O zes9<-m8*W8o~XG9_n%Re9bR*vx&|!W+K9o;uUKl2v;1Ju%^ynR^6o{{I+D&9|3q8@ zW>Tx52XKj8lM3ESne_Cr!KP=)8DucjIE2fcOYEHC#@BG67nyDQO+q4NP(dEO^JAJm zHh8}&`~I!QYLVr@V{E)Vm&D61+_{I9ycCfi-1)Ru?L{%Q%PzK9Ws3YQj4(tjcBhTE zPr#Xp11Km0%CUaxY%v9GH=kk9foJ^%ia*ke8uK-$%eaBPNutOC@(@h@6K-= zx%RXrovB&QyT!jh&6*kkyJV(OM54P6zhBO~&A)$}O|bOq#i}}OZRlRV>+on^L)pM+Bm|WUoY{)QRZPT1>9DE1arUImQf5VP) zrMGQS`5>D;KZ3?Lvr_2<@lZAV!$qA-odgWtA48cQ@}gPXbSZF*CEx7>@cKQ7ORSBh zfOtLUPYmCdv=RVbZU$Z6!xK-c3*?;{t z3<^A3#B(suG3?NaBY%=}9k}@oX}qMNhpF8;x zTZ=^~NkWBG<3`|_vK06#{V=qLx8|MVYZ`+w$K>z`k>|$nl+Jm(4 z95q#eOqSZh(QVdqZA-dmi0i|I9Kt3%f9>*B>wg#GX+o&daxl?BMhPZ(j;(-WZvXbI zDHxJq9z11yNdWq7#+b=QqweFu9J3G{P1~WQbD3@f8_^UW~-f{K$ znO@-TG7lo17_i5%fcV2?!!2NC7alGh_wNd?facWPWQr=MC4U^Ul&KnlaJ~$M)H8dOo9hR-tC|v9o!V)IhO)S`qc6pf z67qr~xw7M4Ro;C1P@p-G7+n-tM3a!DHdeuGVU$F_0x{WY15c#RC%gVO_` zKA~YqOSA<{A6Ta@R`g$kP&yk0eGI{BlCIgP@evu%N=*|C8530cBSbDJ!c^iIGG-n} zTR@6qAH;Bf50fHwXzkU)v* zwf2t9AvaX2>mL23Rw;xOi6)AfWtYfBck#|RJ??q4?6yVkUJa869?`Mg%`_!l!q`bc zs|JT3B9F-eTDvZZD648PaY|hwMa;mGnSz|%0pC}(|7_^Lk+>3pKx$iB%Yeo=noZ%3 z^kfbRnQAjSbr+k*c7`cArYU$Je@XQr9DU@vB_|2O#F-512NzK_2dleC$`MQ(9%W~H%f@|YH*>ak=XmX4()960F znJtK)`rH?rw{_*D7UfIlhwqDwuf*(dVx0QJcQn*O%bj!jk)--YjeRqndioZWpLE)s4fKt?Oq}mYOSGKQN?dL_c_c{S?avrQ||8RBGy9G{R$xTzXQY!F* z$_?Wa5@Z`6ftCfrZVFEa_+s#$>sBLYX*H4P<&(5FqTiB_w3Jy~cCfHq;i#~YsAooU>#mDW9_*6@WYH5`a~p zATuj@Qv%V7Y9{ywfeOhU42u1X{n>$}flr|&9sk?{q6zkFLS9tjoHEpYTKtfPnNx*1 zUj3(KvpvAAvt;FuJ3~0QyT)mpvsiJQN`c*V($BIo~cuwoQZ{`e4tEX_n zs;B(qyT1p){9EUb>G%Y-y_6Y;WLf0O+3^b$64Hz+CSm{L=mrl|n08znIvq#+L6 z0h#B4|0{r!qz~BBsra9V)eEW&ZteBx#*biqLF7L*dd@9(WX~Za2%FDKFICu`=8$pf}R*=n?A<9^A}${%U^Ioi*Kxq8)PI*Ot_Av zRe)Nd7N22b%7lh)??YqmL$q+bIZ&5tR9=bn?Gns4I<`Eo*-l|2Q5~q7B9x1xpwvLT zDor#1v3X?q*gL7vLmFpA-V|)R*E-nw#1T;*LT+IDBG1nrfHX|_Xc1?(JH|>bF@p`pM(+o3^8kZ*EpD?J*+5Gs;YSg7isHC1=@U$4cr%~|cHzV53sm9(CDSXM; zd`b9cm_<&E-z0(7Hp_(z_FSJ%dwBowUH<-IL{~SG0huN-)CfzY7>Y?r9hbB}>J$di zC3Yjn+~coate_|%Jr|_NUkAE7;O837cIhN(G%vX5>hOnB5f89Z- zMxYF8yZ}v_<8we^lPLBD{1#@9DZWXH`n1ERpt({XaExdUk~k2DLh}spl?V*|U^@iV z*+M5#;teNGve(hhGSW*=H<_%(Y8xRg@ zm4KL2ue~H{WpQ4OqySOU1ru>;s)b~d0QM+j{0^V2^Vj`rN@@6fsr#lyuPcBIYW}G3 zZlyVTW*G_GiV-F1s9u~L(tOC1_ch(a#?r#U z_kmukdEy=tIJ-Z$p!ZWrhmtOvCV=e}FZ*F@o(8`;&P(_kyLAuI<8BX(9a-)D(duLf z;uKD4@M?h40URxjMcrb0Ade!~IDC9`)Nm3?k)38PhlhDh=|r|S-zzX4?Z5!6NB?RV z{^8rdcNNoEc06}!t#kV`p8th?U_IxmfyKRtk#Nu1+5P7q8#`Uz(0csX=24^GUfci0 z#LF+fpWgI+%MN(CADJGv7;RJYzZWz?Hsg3LmxG~hUO9xi|{Jh(4l;Y zp7Ziop83QGCqOZVy^h;;b*p7!BI;@6GZl_k)P6*|L;^h35;6B4^f;NyUZ7poBKlt) z_vj;HGHb0o-n4l8<@ZndxEP7rfkgJlt9N9NtnqyZ8=C;Sk$%G2%hJO5o!4n!?`+jmbXr^DxbS$s8~ z=fU5GjXRJzee>}{YXiZfpP)#ElnDo^nl@(#Mkq>XF`o5a=X!Ck6&GOd@Z;5s7hCx* zbeZ|h+r}SfIL?{Yz1Gf_tABa?%ZxwX{uw*xX@g-t#f{pAopRu~ff#@5u^_QVLi2?d z9_Zo1=lO%o6@pgf)BlH6<>WCRcAyIoMy)4q0~{0|*pmJ1?kH(9uFp;h!=`<`nW_G;1EYw4q_T93ejYIab|OaFvj_pjx-10x=s6H8S> zzxWMIfrHpRvUCCF+{9_IB4^bkWUTa^H*fviqdpxJ`6QCLfW7V{1lzdrz6(dCPPNe7 zdg%$t^u|nR5_9u>e0+AWXJ15~|D(Du536~9*IpSzVjHs&vD&0GNP`T;7M5)6QqinT z%|&U_VA?hzmJ(`96e<-`D@{v8XcV=Y2bAVCYd-h0lCzz2{yOjb{_(n8*YEnZQ|tTv z4A1l2&wby|wNx|ND{s=wD%BIc5M4v)p4sR4jR6+oS3cI6)$8uAI5rWg`8>5QkS zah(&d>BBaxyZ_Lp2@Vp ze6*5TF3Tl}RR(@yao^yq@u(~&&RkZDW@J7IGSND#M+&kFyxJ(&xwuH@$R9X_@`VJ3 zM$T2IHB~eA!i4&U?B3@yWimYNPs^bfNYiAHhEYS_P}sNokFhaDQUO`7j=We;`sL;N zIskySE4Y2Kfm`uXU0xEW4J#}>BsOm76K-oPB8b7wak@9s|G6?0&%%V+0&DjRvtBOo z6uS?Ow(0!twz0396#>E^QzmCELiNB4#Sz>J9U*TZ&~-W99JNm0UR@r6tFzClvcEhL ziQY2flr`(^Ca|_|m;P|F8>uxNw0HZ9s=G6jlejh>u;f=SZF6gym9#x|;KZbO!;}^{ zJr5IUDuPuvgHlGEqjN1)QHpYnu7b`5_irkVZ~=~6zV@8%B=7l-$aMgF#Bg6y9f|AV zGDt~lz2fvD5%5~9c{G8J>^nhzrabNmnad!8kX*1!NRNMi?-&z!u|ak1p@G zb3*z_GA%GB4C-GPQbs1LiE@lWH-KPrv?~1Fo&`5rDk(9Y$0x!l-AXaySWjE{?nk!~ zG|u(ozL9?NS~*G@tX|^PLH>asA7wdM8vPsB^D&ppB_m!`ye*N9)z$l3Ry9K@vF}|> z#_t6gvsWkfFYqyc1Gq7&iki8;JZ>l~Pm~-2`csm@GbVbCw}DaMFGtD4fi2$taXb6mb0^V1g?i*l zx+vDPQRbkBZ`>*0cmWp|u}qQR!=_xBHa>1CvKBxc?A{thf}Xe&uL1C(h2!us?X0Ac z(IZoJJiZ1B0WH*y=-8PIbv2cdyAqWw+rH+SA@S>+apV zq;h44piLPdss!>owyieOf9$bUeco<32NW>jA7uc1+Yh38e0uHLxUnzFARMBxr14=l zU*jzi6v;FPN<8KTy>&Zpx;+4<)B=G%eG!uu>d0_*$*F=SQk5$NNlc$UogY7t z;Hnp_w{vpRg{+e5eJ~woy9I1gMvK+gg1})6fA7HMTL&lm$4YKN$cEjoTEJ+5hL0i~ z*FiR$@KG5TWSlMU!5%H-qIoFI0b1N&#Jh0*{0GoR?J^Yb5A+}*>(Q2p)(kyIrN{eE zNCe6L6A*H8y(2-3z<5!aCFiyx%b?2N)AL`wdk(1wqiUZhSe**+naZ%juG9Y&TcK={Y$bb*Gx2Q?e$@d@>g zbX>1pJ7oLCj$Rr*CZSXiCQOW-@}$U$#3;nQM#UR};*W@TaLbI=yY~rp)puS{p z4BhmtEf41+{{-IT3q zBq4zC5L)X9WXA*`5{pwKf<11`7-9g=Abh8br=$V;P#x;=SvS;wgoFrW<`-CV&?Hum9L(T`mz=wSu28y_ zy#z)RbTV4<)SVSYTV%X6gTcZFlWdK0F;AQ3uTUw2gfv#h>WdISL4fa)EEHV{)%LZg z_er6CgkyFs$Pl=R zZG^gqs!3cp4bmo)(k~AO`Kwr6p(sJka#I61qn$qS@S`34pw_}AiQSeGxeCRQ>3nWH zs-`~x+-La9a28lXPlnEh+vbDL#JS#feFZxN<%W?{YV1c-FmgfYBa1N_dxTcXIfHwE zTL@l;sK9lzMZ!#^4@6gjlE6|G(sDpXOW?fo`CP@R{ zj*>0Kb-xQ23!vNgwVHH+JTot}Qp72D17q|1G_d?!(L=SMyOP8R z)y+X;6B}r)=#vmckB-S1b6v4&_md6)-g~i*lLo|SX?f_%^aP%tiFA5o$W^re%FWj;e&MW? zlDQVlFYruAzau4;Ir0)3F?>C@R~pFD^3E!o^&o|V!W=}P9oz$+mmheEJ&mwVBUAt@ z3Yq(xAtaK-41VNzbBJUOpb1i!Sssn0-)Wrpi)Wp_Fn_fUU{HkwzWsdYF6^#X zJ|NaEAtWBv?ym-ZW>i$;Qu+X^MY2?&*a55+Qby7q;B(eNRKPWh1jl#UN_+)-;=RWM z>=r_Dk=QnGnH=9Id-8i*mxbjVfagLk3OObV;qpD&nAcie<`XQ0*g^h`3O1#M9>-`6 zL4KjRagB|Q(^hj4=+N9C-P?w$X85$-LW|4SZZ_xioFv*QFgP|vT|r_2<{OORp+U&V zmW86;y6gfQg%AizCJ+?~p(WUZ1W$mZNwNsSjp3tlHR$G3ZHsipLbV2tOThQvUle$G z;>T}Ti1Kb{MHww8amUXE67q~&y93_iVaPYTygJeJBbPcFfsJNwM_kSkU?5p=`k+cK z*iH%b%^;87FtB~M#Cw~j3VrXQFOfwmgB*+XcG3jo=T>&=*;zn>DB2Sl9y}uQf@HVN zdguna5_R4PDnxdf5@vto@Zl%ecermg0CsnMDladmcL_(ptq;9TM;xFlD~b0B>G;;S z`pQ`3vkw^eUH~~CYXqqo>Wtb{fXs7ppCP+9w;5Z2} zvEbbO_%Hb=S_zQIa~dBC+bsS**~7mpx&s3))=ZK3Pt z7pDE_=iahX44{43H;8H-gJ<@esE3|J?oJcJ2)w8Hz0~lBtOy-;gHA<&IpqFB###CI z|NJDZ$JDRLRiA z4mT|fIS?_F-ulCyd;GqbqK=@U{`3?e+CCZieoNDr;u1vYm+~EV#!PX9YA}N$Fi|Wu z%NxspmGjvcKYF}g@JFz$5S2Ay#5f5U{4%m&CszR>jkB$gmat8gacvnD|LYHSM+~mJ zzN3J?F-2vs$-kKvqgfzW;7BrA$YF`VO|+JMBg7TqeK3T@)@3uvo`}lDpZ_z3VO&{V z1E3VmC6dbb=@Vd#P|1bFo25zZ2ab*5Ny$-_wIhx2d1=U|8%uDsWFJoZ7v)8O2`vOX z_t;E;b~qQ=mFF1BBaA`D_=o5PBf7E3=@0(-=06c%QW1ZN%wGYNCgONL<6q>Q66nOJ z;mXnk-N6CJQ3OPkgClKVlZPi_oEEs&ij9^OsI(gSh(eGGj-{?4o4S_BxcL8Og!^D} ztA#sc1l=5_@ zN1#nA?y;TW^n*P(6pc!BML_#UHkuZG&ad_Dx@_>QjbFxj&3$Q0#At9Kh9K(QK`bXG zS@J-!D1KBLT^*#Fh20B|eNFv|Mx`czlo^B{@M+SrW6=JE+GHymU_I{B(EswCcsXoh zq|>MjFhQ1HZbxV+;#^OjRH-vUd7DvLT}@P9mdm6_C`lg`=FsRUC;&Cs=mir0YewHH zhOh9*Q~TFpkfnPcycs0BTDg7AtC5tkG*OfQ5!e2SMP@*xcvKPK5`o(Ql;Wjg9Kr>W zL;7&){tM~tJPkSG*vd%ATrvd7=FtUr%mpA!mmTufPcj1ivX_dAUU|c=%z%|AVi%iM z7@;+8rKtbtd?J9KzM}{jd?KC!BXs%*?QAC0jvBZO{706Gv-SX5Ft}VLC2X(SrB0p0 zuo?T~o3Cwy+{)fx#@M68ENV~8Cg7QTGh__G0u}Jz1VoJNyaTjfoJHyj6t8v}W<^$t zlxCz<02nBVn(~7t6151`JW0@*5EoYzm|7A`n^1Cowgf7UcjzWY=-iPHp)e!RJQ1o} zeIqMtbOO7YP+UEV|JBQMWP{sHrZyn_5kit)7T{dEyf;gG^~wTQCYwo`9Q0y#TaMZ3 zIJV#qOvW8q@_DAgQVZ zwwLBe3D`^}CC-Mua*R^x@3)Xw0~K{t8TBD{)=jQB3d|cJ+>eR!KR`B+o1zMS5vrVu%fAZv{$LL(1p_<%K?Ch2U&*DP0kKA- zP(l62BoF&tI0ci#W!SQShNLq>kS+y&6s)7EIc|hLB`(#zCofIfV zmx!;9sMMzzI67y^7s?djEWMSDlzVd2xUsFr4$lJ=*Y)LWzJ)^kuQSrx(^_YAo0&dt zU5fWRJ`ZE(K{x>KPYWw1pD^#47!9W(%$8{mun;2(!KtiljzjyBR-PRK`cVUB3a*S_ z^sIFrY`=h)v6l-y8rb&5p=?1<`}{%UM&S8ZyFJXiWus*mGl9L z|A^#{9fBZHwjc)hv(r*r@G9|MG@1}$?@|v16{&@l)rV_xX5?o^a`0K+;!@TpiSO1&qYPEM(3y{ z4^Ca8Hr{)$$fl6-_)NWA6haorFv`{iq0-68Nu(Yc$TG_~C1sIR6%co{LUm1VvEnuN zQSPk)75$sQJ@~4B6TloixhIRRDLm-H{KzbfQ;S#s)VYLhEhjrUtznA&sQv}hkgrr; z4MaNYiDq(V<;f`)_=(}+;dzqLNryE&O*pZ!<`9D%PAt=lw+lvaxP#7uf`UTbLDjzN zL-Td7dtA++SH&YS=`c|3FEOq{K zmkPf-I&gNMzL3G_rrzHJ8RhZgQIW)p<1oAjDegWR?*wDZ$+ccB(b=`@b<&+MLp7gNOn}%TfT_2&V7R(;}KkLV`*@ zkLP4L_pI`SxaD@MHMGH?79lmwewvm3Ew1;g_(i|{BJ5P$c}7_vVfZK2qiHcbsSarTx`urR z4{Fp8YB(F0;GHcrD7E3N>#bePBBP1qw~KyRk{UO*H}TMUNz-XncJ~ZD|DJwpb+aN8 z!sqoMU;2G$ofMy6-#2s>R1_u)jnGM~5|ge8r*7;Op_yV&fTTYyv!Ti#wWX2cPd_i% z;hquy=h4D1e4wYN$7JEnb?bkrSul6|J*~g?UYjyhYsvq48yp>D(bF2sS`pN|7NCcp zukRta8JE=$Z`lr~cOtrJ%o7Szqg2x-}{K|#s1_DP@a!2ek% z_KIK&)hyt#cWG(e!FzOdKdLpnCp~9huvg!t~cyQcqKy15L^KSln zS^%|Whk)1qE%HeF`ZV==L4p_ztRxpzyCan8^Hn`cBewta8GV~S z(Drah-?56p5R5=ikELmC9VztU>z;clDMoB?u4l62QX8iW&W4f*psfZQ*GG-^EdDOb ztxz{J_~nY=nbrKQUUr!@Aba8cBb8Vt%?gi*Xx~GtK>@ zRUpNZS1KDOk>!!4bq*HT3+xb3c>W5ZXL|lpR*3lRY$7K|Ki-nYdT4!oQ2N(T=dNv3 zE;Z*Z_&BEa;ust^!9?UQAVAfkRv|wIxGIXVENpBFkX&Ouec{Xs-11WF?H30Zm;M&X%r&n4V3DGkdFbOoE zol^_Ruoh@KIc>v_W#|2>VF$Csv~tdUdaiG(Ob|?-Ew@+sr-oNGY7b-wpjdy5c@P?^ zPJJ4%v`m&BGzEQ!Z-j2dh7=iwn!`hNSNarcPgiV)O^D&ee_>Nibt3Y=&={$y7X5LsFlTJsjU$Pt!(UWhxsm;l zB^+>ZcTZ;40Szm_Y>eOGNh*OxpK)I?4|8>I+HhR$l;`!9wRqga;aLV91>9YE)b5Xq zi_`yAAJ-6)?NE?KEFiL{ZZ>*j5HPL)45aeyWQ(}*m)ki0=riO12~{SG2LRE;!<#3( z*=JzYJYh07(8DGqY!pg3`VR1RF_f?Xm7ZIY`=X#@M^TD7C2<0~N5S>1-lV1NBp(xfQ#LKa^*$Ye6VJ#g|E zx{N0FTh_+)yPVJpn+uOEfpnr@FUCB-mK5e3_&DK?cdOX`VF>!Wbup^`88aW+@Q&D{ zyyu$16nX^|PqiEweEDmkO$x}O2bAYsELrgzc3=OubX8&Ho5)7)+`j$XcEN=U7ha+G z=@In-1&wFBAULoObfA0b_%+tn)_Yei2p+tCUDY$Cq6<{ZzuwoaM7cSc^$PvPV_nnF ztxCYq!`w^zAbjuMJ)2>7&~z*v$EoUMIBp4vwnqk0It^|!KJ(E$LX9k0_bLS=oRpZ};M@2Q;KNvi9xLFA#OcbHzZEp*sHrTIInRL{H z#5byNhW02yTdoh>6&oMP&nN#`%wVYSvkk_^%QGc)mM&h5=EHnVef-Gd8) z7`%!uy{O7XK0?!K-+pREN=3trZ3jXmHO;*db7#!uu>DIC#Ac zhf-SP2v9@)38yD!9e`w@G8Ij$1pbkNBql5@j69m=37=Ub>Vp^(jMl0-5(ak2jT^rr zYx-cj01HmVPJMm-X9tz)R_D*H&L0X2?jpm5bLI}s%^#(Ass@X^2i1zElvB7M7OsJ{brtckgbZF*a#urzc(z z^F3m0Xc&nZZ2>KSYoqaj-qblDtwY#~ee&c!9v3YMhuztmbyG8Q$bAR?5v=*$1?7+T zd+X`zC-*vo(?sJaAW<%C5jjCGIpJ(=?_TVUO5qvV`YV~@;&Y&&Ue{NIG=#@X9D-x_ zw+Fh~+WQOe6)jF17?S0e_6f~dk#@-1x^uM+wr(b7CtunJe~PH|PaXH7IfwrC_Le97 zI;iu;jT@P+z-k)W3-J==09$w&bEf29v+;;DX18*<16eKCATt@rt^lhz>oHB(p1rvk z&pCu)R@cq;0DFy2pLVW&uSjq}5GI$z^rMTnuFnp+SB^I~=(@?5`;s>AKwbX6y~(86i%XGZ6st==1_ zI<+2bf?V82?pIL@^H~WhqJQV>qK?F>`7sBeWDzd87Py@Z0%ce*vXj>p?rY019?Ua{ zBnoSB1j2(Bl4*7axCR}{<=Gonbt~DFlAuwXHD%3LniW`NbU}RAmpc(D**kC_qcO6S zMa^7TKr08i?~c>~J}Flza^gBx=$IHktZuvva@M!yiPxX_sjzn*CnsEX2uSCy?&Z0; zCgq+EK!G6A)&WPP2e%}2d`ZJ$1IB|^an=?V^-Wb{#CPWHsNW*BH4kl=Q3aMrim2lW zJ5>ZET;n&=4{}Pf+(cK;6E2IlUo2&YG)|ra=KtufUqFL~X$Y%6-rDP3ooGCs4|HUI zCVDT2AVa72I*lpCJ{;4i*!rw3AYEjsL?4crBt`_F@t21l2?62w3D?tkog!{qzjirp!^K+D1*B?p zm)IN0r}Yjh$#0P?I3PnNhyuCb2l6u^dOW}hO)#y1^nH?_v*7Jh+e3$BIWdnPyRNwz zwZR2<*USE6Rg;Ws?bjIhywx3bwcGI`gYotBa2!{+aTMQYtGliIgF;nFP-mxzEqT^s zOnHtB;9Ii+=mD;kps@0201QMOJyHq5l@p~@2l#4R`Qpy{dW#sF zrC(Bx7y~n^ZFNqV$m^+%KuEt<0{FQ3k%w(_WJv(x?6cX!iQzURm^A>jx${#mpVEb) zo(uOqP(w=Z;Sz}5*%H^=hOA@nf*p9nPm__4csUwM`(wp^W5I7oGg>E9PimGsravfD zVDc`nRJgj-skuRy+p8U2QW4eu&cD^F_=$u`8II(eD!UFBG`k`SEn!wgn9MfD+&XHAk0zDTj^~-CJM2;u)-iI_%5KoKU%vvmx4KDVRP`{K}*z%Xt3T`b+@=!Z~30bX@NmV z9zd~j>3mmQ98TGcoo_Rk0| zp9DD=PEqDDE*L4qf+W;+w-o+jAkm-b?-M;AKN)qF_=&~MYXSL^a+bXdF&cqV66=DS z792nvKfW*Er&p5)D#xqFFE=Cc0Ixm=vJ^N6y_`n{6Q$HwZDB?Xs@A$b{9Y7~H*wJu zBm2?reVh^ksODoz5!-=h=p>=4S`lHm^oz&0uT@5GDu4+df~*RdaE%-$beQLbu;5AC z)U_~X=S`oTuOW9sCV_P|g@D)Z(j}9dSI~E_Lu4SsJ7jMKW}fc+5IS7<+zLQxL5V>_ zm8R1T_%5T*^~Tl(XOO6CfSgMd?;`iL#Pn5x2G9^d`pej+cvR7)Z6E0xL0%K+6H#%v z54llOd)EwPd^pS7(d>nhw^dN_c(-*MB2lU$>UZjUkaP;Jh5|XBOR>wmvWGYRR~n^- z-a|f(UZ8&ss4p6Q1QL>h4?qMl>@*V3!S%~;%M=a|A%cV{K>s9FRgLi(YLbn4XubdS z*ElR*TinGr7O@yev=QgL0ALeYS>0^MAeH3K0)0qP@|03qCv%&EAlJhwF()H3B&BFn zFG5!jF@^A8SBz_O=xN(e7XB^Dh=m}>(WazvL$>Y>`e@gYeH|KAB-12eDM|<25mS14 z1QoD7Ff|H-b_>9EWE_gu2~uGw+647DBf!lB`ULam2{7!gysYhAD?(|0`~}!rcc4kF zQU9$zG3$W$*T21LPJsWL(>k5M#^&Sg>I+!i1LH{_5I zU?+*zi3uM|8{jfxPXf(QuQJ~rWe?Y;Ns{kX5)_S&)4IYlX~z1CcFj?qW&eGIQD${(D|v4Vra zV9aG6+NaE5OtWV&rW~6&9lxVHxkL~DntuH7!F~8Qe)aE7bH%@AnIF=!VlY-;rTe#u}o%ssz!>}%QU*|W#K0^I>RCy{rGW`+QxB5Lws zCH2=BR!=lH3Mf6k)MlHuYTnhNFN{#7$NOe!ZCoPuP(CCybcMt`S)T3Nw<`?N6D0W+ z7}od_?_mL6-plxs!OhL>;NZYywb|}IzciOOs!TviN{W|zH=uyIe8}nCI7m0dzakUFDWT0p!GC(&z?OEjg2RI{$>Ztzy)i*Pj-8;SvNJ< zdGKd*{fmp&Z{93@>C>L;83wRE%2`%LG->Z(Zfn8??+Rz~5w@a6rD zOAfnu%w=DJV>!}O^`5Wq9+q@9<|kTwIS>*YZ0Pjh4_=w@bz02umba8R<*7m*}Zr5)fQp4|01~`XaUo058|HhItHK`&8f3g## z;w2pN`X#?w{{45E#GXfYktRb9`<$tKdg5tSdq+DL7uT7h$Cv%>KhF{|sZhj!iuUf5 z4^zqU;4Ysp^*gY)wOmj@Kp_8Y%ZHe#cMo?9ii=l&Z(MoK*}1W`RgftoE310uOyr|( zhS48?{PFnC9Zp_e-sKV;M_0ru{Ch~*#H4Mm(5sCg~iBP!2JS z*K9D~$+>c+anDZS+wb4M7jHjUKh0caX1dsozyE$;8lkdnwU%*0WlyP!LUYPje-+=Y zCpa1`9?FN!TfHl|EJ~yGfm!C8KhB>ikBu4m{Z%SbKg;F3xQpom`}67RzE2i)&z?;+S6>BaQajSOo9lLrs~{dZrCUWWGJyBpth zjkNeXbrj7ON)J3!7FqMuL1$*VbnQa>^DA+KT~|i0>j2Dibmn9;qoSgs6{4no?yJN; z*Q&>8i`-haWlwKQ**25SX*IDT#vi_UDM%LecYm#^+O*^bHfxA!O=?ZCS3Iuwf4fhC zxbuA%%bD+b zsQuLN8lQrHG{6?KSl6y5g!k}We(Z|+X3M6pU&CrMoQ(Us8XbRqx>{4E;J5S9`6)Bb zbTz!`&>TBfLO50pw_kpG|JHX~B6fb?bF5nP7Y`}*=HigKejEGpe1zic>YO#WU4A_h zP>=prn|bWz^o(DhZww3%cD_<>XFnRJ9zEI=9K8sYurmesMLNzFO7!`?_TY}g0bRCJ ztG3#8WJo*JIzrE-|3n|_k6E*3Iri1qiJbdo{&2V5yZTvdY$|PwFkyZXD3Bh>ENkdP zMhx~*OI1?b#w9n-d)2IZP!NjM(Lx|mdwFGUO_g0|Z4CW}{<$BQotBS+WW}8eaF(a< zd?@dX$NrdX5#&^ba8;IQtT81lPamha7!NCA*>Gsu)TxJn7Soyv9)!6J4K#l%3YxAk zll?^4WDN5uO>cdJ%rg65J3CZ7Ip)~pdvcCn_)*P`OC*MWebLT5m!h3y+Ft9@^G&I@ zy-FkDWTCpH<XQ*a9&+DaFnX?*spGY>5cN1?n$fx^(LIuOIh!*+k}f z3$AbO`eU)6My#p_m&CKPRXZ-OpGtQSXpdtfz4`aAFCO=|opc|xEcBNsi%?lVWqoQQ zmwo-DO-G)~Vzb;XbLoD*>+~}-uV*2*hP8AjB*sqXNU#p)jU^7@vQkdmUGR*bq$ODR)OJ^HghD z-~6$9x9!IKUVnf8NR=@6j?6(*%AIz#4q=0G_AZBHWR&q(n{|={D=IW!y?V7lQc@4m zAl15=rL3%c=l1PKJw4VA_V&3YB`TSjnH)1`DjquYH_k~IWo^KW{#{EwHKyv-zowA= z=l97Dd+Yb>A$u=-96nVqtR$kZ)=8nxbtHo)LMO#iJKaA1!i5WS{G-1iSbofM%~Frk zi@CK@v}svbw0hJ(|NNsGe|*2UpoaXTM~^r-I9LGz0q4%0t3EimYt=kjHeTNSONI1Q z_w2cV^+9@!qPzb0Uw%Hm*UsC27Ib0lkmA+i^dv4{y&8oeGF5-s1r(z@CkuIsPSsmu zHZwg}5HWTyr+0Q5BBz|cxK?rFI+?EsIlF9HmWitsdEdVu8Q?msB{lpj=($Pd<4c#O z1|3X21{9XsnK@Yc^u*yB=f0X%&9{yq+5JCU_;U8jrAxyo>2`2y&zQR+SpMOz<>G2T zKAG>d{`UUQ_K%5&o%%Y|3=M;U7<`2KRxeu?{`5qCq;+$#fJ*318K2GVl52nE1zkA1 z;{n#hPd-dpMTI~i9+fuME{Cq*$B!Rdf3JVBY9!qUYo#F3dk49b*UYT*obT$~S?8A~ zTZ}%dbnOEB+ju*SEBqYi`@ROaeyE5y;APGi|L!LFt9}ZK`io%#wFDxusap|-uV1^S zq@&}vSuZ_2IG86SB!o8J%a?m=YiqlH{7{mXcKbavw0PC3Xza-i+qbLYH9f5~>EGd1 z(p*}Za5B+((DrO?x>nWmv-w>I)(Z31ixBO_ycvb6d z)^+AXG_!isRGGUQBUXr<2|sXa#Q~4cM)8`QOO}NGY_Ey}%1yQZ@py-6b+}OGuRRDP ztl(e2>^v=@#@CXR!X8l2X0lD zKa_ISqj=@{=U!_d#3^Cy0MBWIWn!3Kcy+u;qOUq=`8K0d0Fwv>&j(WpVV7y>mto8Q z+9^yQ5GakL#>id&>C>nCQm&b}+mQPDqfBLO?faCZ{@i=z+dr3Qk7Y0ZGl;c}hILV8 zk*>qt0ow7$FLODyZRz{@_un&n+bUJk>^gUt@1n~}D zapi#fstpPX3e4h9FRzxx=}SG_Ws_ImOy9^bB2KG*iR14dIU)wH=k%qf1L!$ociLcw zUmUyILiW`*ii_)v40XDyWu9|jDjz@om}0+#)6Zq%%jrtlCbTxo0xcW!Xdye&yXBe6 zGBVfk;Sp!6o-ZH!E?M?>@$m5MqqJ^spUksi(Ni70Gx$_ zCBov%YgRCDnA|<#zaH#3yVZUxeUsaAdU3+{l9G}L)F9b1SpYb-E`w*+vI)2PRQf2+ z$pC*qJkPmYejQ`(W3zJ79rUC8Jn8A_+bur-yYV(1s2!u1l|Du*61d<;PfzKkS&K^3 z9jur>>pJlzVF-g6?Dgfu0bnpVmY3g7pD+0<5Xo6@rIbs$ZTrU)-woL}_YXn}L;6qD zc_B%4E=T(6T(j5@8g$U(!|>4H+|dVByvM$gV3QpWYMuLZYpe<-SUkYfx7QsM*t)g2 z{taD=ad{f8|DLy1RYz1pWt9Nsd6;kByjdZ6E*artY|Rvxu`kttl10R(MYYD^>qA!X zjio|ozP*2hczEZ97kdNUJbITXc{+0w{xh_~x$OrHU6nU4i&ohVUQ6Q+myu zH}AvE<=b#>`IrbNMJvtHAKkmRY;0F712xESyLM-@uJ*Q*?^IE?@Gz-J*lPGU?Zol# zI#1vA=8llQ9szP>9G+%Wc=TO}`(yWR9z{%$@FOIl(*PMrO`bffrCfA?L`!hs#}><6 zko{c|fL*%W&ONt(|NhOxJm|qy_WjdKL{-DFl|M3iUq~K@30h-4c$oxNwj7^&n@#Fo zpM|u+;L}Jl)kTh#LqKE&Uz{=Dl83!(I2X~Yz3w)B?h3=}=~*MgX$NjDdjxbfwt)lJ z&|UDa{paONtaGXC!LE5LcgSA5c5Q6!`1#pC@9ET$au5g8%frJXIN!Ky9+$Z0j~_qA zziB4>o9+Q2$gbDBas7I6;9nOx7cBzPqV;$CO>Zf1Ccxi$qj!EATwPt61jo(8+g9lp^%k@fd(#Mv~Q=fedjuB93diOA@$}~-m z#mUVb_t(Yg&E@*8B4hjBg&l+bRf|+*2Au1<){*>W9#a`bFX2=$62ByA{={LSiwL)v`UVxF-e+F*=Io(c!D9l zMLj~9m&gT5^fgt_&FW&gBpgrW-`zA_o_!i}=hA+8<%c&(*`X-tV19YI273t+nNGJ= z#HX5ny6|GOIq>4H?NnTHhnriusTKM;`F)7SuR68D3k;@Z7RrS7cGJIYrn76KVfLN z%;gV%yqK3Klew^oZr;3UeBeWSRZ{D}*A_W8hxl~$#yj7$W>U{1GX`Pu-t|NnW_a&U3Q&{xlq^)p$>0}|rnRwKKdt3G{ z7F2)R6aVeEVu;LQ?+q#kwi&&x=htJ1zQy|_^5H|QZq%eH)4x5hGPTn8@$?MD61D&? zs8I!J|Ks@n{ri8|vFm%zHAL}sRbl_%+27->W?Ug$v^BVQIIiXi3&(A*C-Q*9*zh2(&{pVLV7Oz+l zNs$9+1Vm3J8)0*?pJU$SxLLs=ArZ)wr0GZwwr$hec-O?t4tbiQ_sFjtsZwN@jpYHm z?c`fO#FPMQEnc)J2;||nM_0LurG|gZ@Bj6y;ny$X`jC{OP)~eA0mde+EW}B>-z-UWU#d-Os?hF-^6hn5`gC#RutxKpst7%ar%Mjtb}J=w zB%KF{y&GstD7@q67cG?aTN&_C#oGEApS<63Xi_4(3m ztpy|%{o(D5q_2&MRy$V>e7eg0%(h+op_tVHyaRO{#b{#&)RK_Mw$QaHm zUkl^_C{zmMKmqMzqH(M5Nq;548_V>O@88Xt#TTZ{rVQWuZVE#z9u;G4ddhuKvsgfn z7SL)%cq;49|IX?=d;IwEp7x}=7UULhp39dnN1(<{?S8l0CTF!>IMP!2Z##C~?zs)G z!QR24`Na(GQm~%kcpc3@tIP_2+D2)_spD{ixp;Uw4Inx!z$FqCq|ryN#fwA2!}&Mq zrSHsNUaUK~Nhetm<#_J1p|;8dpnY$AR)~VskSb&}%UutBKV)P$&aD=d@_&GuE#4sK z3=pxeX#xTAx=dY!V!|a2&A$B}VaCpNF-MYGF3y+CwSM^Wleud^$%^z?8#A+gdirTN zS94cvKa}IX+RE(4&6}#IQ?S}$PMsN|IXSYjvZ%FlIJvm|8%4}5GM4PNZF}GF#%r5I zTuqvtF+P&cUnyRfNuLn*8^TQl>U>nYOcamh&YKwZj>KRs)wBN#KcJziddKD0rx{c- z%$T<-*E$WD;2X*iHcNB_G_2aAJ=4^}75_rG(@3#wR1K9~e)-Ckyuw1E9DP#|JAuET zAIiwd*|?%Qgs^z>qft^-O%N&NGs6}fkfrPY8)5WGO8SKo8Lnt`)h)=2Hpl{5ff?F6G+wf0|d9<2`%#cJKt=1m+JR!9(t7e$L@A%eQrt<}5L zug+Ut41NW1KhUIar2gXk(ongT5`%3CqCKHmfHjp|l^dBvzb)CM9SjoRXqTgCTwD#- z6sUg-EhWODDpt19dKdl9tyN-M4-01&KiYehhlyv)?64vxu(!cW0epcxXb^9(kzi9J znO=OS-}#Aw91ls)$VjN^qs@#Tv;Ot{82F9)RF7$lloRoqsu_+pD8O}5r9J-q`SUY_ zoO$ht=xMg?dDhl(hjFgKi6tVZZ!E7g*)eU#48ff{HP4?vPxvh>D+`$)2!1`pDj2HTVJ zjS`xX5(1DXc!h-GWXn5i(ll7~sIq1{?Y3#53IVa%+0q#!!-Kx=GdJPhxE#OziCrb< zyLCVIiduJ50k7`7*9aGPc?c^WJ9eyN2nb})-o2mMiBaz(N~++^qY$1J5gRy5(D3YT52=W z2A^KNdKD$&V}!sKWWW+s4qG?fvcO^pq;PjSp0`8sIe^@okP(Z|g1S*xsu{UuyL?WK%YwP`UZROrPc_LgEtySV2+)<_zr-V}-x7!tgFQLX| zWT>x|76wF+uR%S?-DsZ1Wu|e~13jl#i&^qiwXWEH%JWOB!4mAn10D=c!AX-Qkv4@z z{alL>WP@*PuS__YH9Qby+ff|>5{ZqP+-4c`)p3>#U-=?n>!Q<;rtWUmjl@A) zt5PqA(2Uopnt85&pvaYnWbXA{C4PTh~Y{oROhqW0y`YA;qeSOLCs4F##A=K;x1&NoBulBuU ztawl2&mAl2w}RzGnqo19PK`FP%VY0yA3XB0-8N%*ptY;3s~D-ur`@cyYv5zb z@T8OvR_PyF&m>jxHg-5PJ>xuJ`oYJ##?|zA9x#Vh`eyL1kUNt*oH>m|2RtO*R@x!Z z5(iDe0p$d6v(3xCV~N^}7c8hBCQf}Yd^QEAU%zq@oKOaKh}eUv6PJ)cXsmw_O>n!7DZNGrI=cbcXtR6HQzH;N%Q0Emh@RzrOCf@k~yQ z;o;#9-ghX0J6HK!n&jz=0}A=d``*1%x!Pg-)oeSNy|*uIys?XjIDB_-R2%6pw{^<#VOb&cEgGN#8F7t{zUTxl8@&P%j2 zdsSBhqIt;wx3-`BSCP6IS1(=4Ws}PbuG}_;wh1`l@DayzD;~V~Ah&Zc$*CUVU4Mg! z9K57^YlTRjrBgZgP3*PEV=u3yxcC7}GxPd9a#A0EMcHkYUI6^EwZOBEXcfoSO~hG3 zGLd(0jce*nv}(7D`D!0iUS^b#QdqC$snZwmp!Z~)`1$HGETPq69tNk->C>k_rdWmT zFsqFwoPrlsUXxQ-r;qR`pr4rzVm}HYTLAPU0v3@2;3dpj3=Z$VEU5ZJn~`F7qF(Os zVdym}4WB;!i85I1;X9rELV>Wud%Cp*nZXp>t`(4QFie3(y1DU>?9Nz$yyLxs*=pg zNe)4N(d}NUyI4R~5wC`wdt%a zDGOKHi`eKD3o_rd2L_|b%EPAc$yf)?c6UW{&5bKq|@4fxeU{G1$xGqi?8;OUB zN}Ea;)yMm;$(9F5xz;+bO>BcSau~SpJ+_#4Yq^Bby8tO6Cj1IU=Ow1Qeg1RTE^X4H z?~52$yC&(xAUSc&pRWNJuLdw-yxD+Fs^v;ZR-N2~qU4UXn0l;sqD=9~{#Q|`iKBpq zSiu_l&T6~uI^Oma-3J%=5Y#Uss+y{*YUmKOx6RC!nzSTasFpy9Odlz|1)%Pso}bjp zDzjYzEd2Wx5M}~(;Dnoe^hCFC=Jq;2LrLuva?NK;qUX4!&)3#hPqKpi zSJ}~xJ#Ccw`}@oJtqFyaB`AQN9epvm?^-&lxG$2&e|l@^6=9#+)MrluW4$fkSv^W! z4kYmbcdMo{=(M~{D(`R#tkP@18dSHgz*%p_+73p4k5%kYhN_cjQmLHpBUJb^Q->`@ zyi@!8T@)7PFzXC%;i$)?z;71NgmCbQK?n9dnUV==l<+1P8JS!%QMw&*%QoHvY8F3N zjx@&4gqqhbVrCYvSsHS9Y2ukO8Nxxt#vu7b!Ri>P8o>hWE(>NZ0#x6U?}GaB>BncM zg8*r@u`x=amQvzD($IiH`sV4sdytBVdvgBqYAI=*e=J(WdqPjO8u@q;pX_^o38!bK zHM>x|Ly-Fxo7SNWav2X0G+Fmzkxf%!H4gB}gv#y6wR94QZ~$r!f@0Q+@>CsFLgI=1 zyQ{djVOvm*vWhKlf@Sw`<&kme41*W2gEVE>)|=hCeEJ1vg8=Q}m1qi&Ow zJK={_GX@f11*87@vua)b`Oj@2FiPPjfrIC@Y}MX9dx)=*_*r4_`{}nOpazw2*6*HR zF&Z`s3%_?h?q5++VdC@06i;vDrwbP^!ZqP}I%OB5Ts=~C16f2@&3pCwHQOSZ^wl+W z#li2BFDrt};Y7AmknHEFn-Yf_H{GFIHP!kw^yc0Aohzk_Sg_pyW}dE1cc428;t z#JL);;=Y(wD10TX(?dO|rNW`v38+Qh8DrsMLA7#hxbf;SsMZ0 zop1c_-wUr_ziw$Zd-iOVx=d#^z^pNarQs5rJN#vC5ey7bu1q6bEx~ZAv=2) z7;PJ4BT$lG09qjJb)2P0 zgUjVR(P)nUJY0?8d4$a_`(jf9$udz_ zvx48hSIp>ddJw4;bY)BlFeNjx+wp&}_g~JJDh0h#T}37%(jNHvbF)Xa2lD?g0+&6h zYKNAm8tN4RFUy{x!9GacKIcaz1=Hztp|zVq4NRFn=e_&t-PKi8jSmghjTmL<(j5tl z5rw*6^W3>~yq_T4y$lPA=s;g*9LSF`W_^tmdlep&gpIQaJC+Q~5XPvc2WBm`PMs4x zHL;tDQz-$&rFUINow9*RXpUGaw(vX3D@(B{;Nwtc!vLX)ieLo;-^lsxR6<%19Vfl9 z9T|$3{3M*z>0OJ5Cq#eBMlp%ejtTx9m62 zhb<*x9`0Nc0?e+yzVcwE7hD&-V?sOJJ4gMqmRp2N6R66jR>aYTlWFL`0&6VPWBVnChm?a zO~iy}PX%@)1u(#6fXM=gdjzPX0$NIVOvJm*J{y+6K?w8ZSLUwdVS&Lb) zjcaXyfn=;n70QCw3LF06`9*9sfCCOXe9C(;q9SIs;(d*P0Flrirpr&IPvF1FzSFIs z1WG{HSNkC4L_=?1wsRr9yc%@)VB0mL0TLOdsJix1ci6fcu08 z*1FRW98KzOR&K^x5RV`mpdPWC}%U^@Yzlp`I& z10n!+qZCd@G4eqzx$$bu6Ssd7g{^b8Fqu(M82s68WhRKDNvbw@;T5C$RA4wQg{K1e zFM3}@0ix*2$N;n@z&UyHz>TSF(vDuFEQkOq5Z{u;qy0#gm4 zFo=A`e0%~p&Fs7bbss=EKPTrP%1^TPC@OlA&A3@hcqARU9H?6MfvDLB6#j9!xw#vY z9KOGp0sD|Bg3#hz_S!LYASVV^L@9?{9~=bB91d_p$<<=Y;=Le6$yJ5)89Hb#!v9IKNanUISPYHiW>OhSF&af%Rj$ zLju4hZ(5DNS>I}aeyq%D$#cK%CnZ#xbp8B!7l{gN8a%i#Vc9en^MXIt>9a6RYu`a2 z6Re3Um^qhQ^Xd3W$KoYQD<_%OSOXsqp@zSH<3#-hZ~@$zLdR z60sb11yxrrT?&UGNPW{qmH()fvl3)F07@|&tR?8xo*Tdl6aGmTuHh;1ZWm64uaigG}rzro9~{`?HX zL~YP4;UHr!^05PMb~R$Qxi{itGduzVjXwI=100pfrFbkp=lZ?)@zK_mc|4b1w8iym}cNSk`7CF=3;(%OGZ4Tra4z zb?eq=*3HT!#zH4D&(TG7A{;!2BLY~l5Mq3&CL-o>+6gRlxJ02tfFUPaIQiu(%B1OUmB*sE%$Jj z8f&{>x?u3W%zNB-idaRD`X(uFQh>Y_Ix@>3TEXJ$(6rhN`7=Yf5 z(r_!Y_}7mw6c~_cmy1`%P)-H|(mWDx7jRyw)Tt^bu12+(fAWt<7nSGiS+Iw{IQU|C z;Ikn6ebW|tIR5J>K4W2o%)Q%RWugmHS@IX7md~ zE=S}5o8r~ZjL&#bI=9+Cb^LK370DZ5f!I1Y0Ox0v87jYi2hH;O{LR{%7bRAZ~$ybBx zP;S&i%;2}56v!Y0CtLtgaC5MThm`0oRe=*$L`>@kQ$d04EHKPn&|aHSRpAOr#28jM z^pNk??KEugsl6YL@G^aUeFyroMq&WB@)|9D{bK9k z`ROS>EYytk`2g8$GZW%k>R?Bj2y$zpY0bxmOYCjAMwBz1p?Ih<&$|RD_GVW}qa_)G z?u$D-MGAb=+_fGJHFO+Ew}#8-MTI_$qEz6}u7%O==b-2*thBPD64L!Y7>0{Gd9DN^hcMEgKo`9619ZF7~q=wqH;t z$*6JVWY`uJ21kC=)6X?-jO1IO=M&Z?KZHy{A{M#NFzNbi))gr0og^3q`lYy}1bFIzIw)Rnd!9%2UAO^CbH70-3#=eH}Teue#)V{-IFvQ1bB^-b}Q3Rl1CB78l1S)Qn zSzTsVceh~tobo*{W^-~)6H+wiOqc$o2vTA1iyQwuMzabBN3HR07~TlygFR`i(1(mk z79r<>9wiuOZk&jD`&JgI_%tkHWHhjb#=EMJa2iYxsbBq_IF~Kc+-SUH39fFBtG~d; z#?@zY&YLF0apS^Lw+;OEafR5CV1VxLwu?oGzWfnbp$XO=hy5)lEJ^4cBKrEIB}Kav!jiNO9F9v z1qElo<^&wfxpb*U{TC+(hX{aR^5BD8ivxZ1`6e4fF{&3JzTMf1juON+WC}8}g6$zB z)lrjX1Gn=z55ez17^u%mpB&pu{%+VjUstf>D?@Py-lp3%fO?n{V#&n;V;5PHZMqtm zRFOdXO+DBD6laGV%;2L|nkN4R`|@j@QShi}i*;4}8HSPuY?~D8IuKhpd^qM7%1}sLlGb(KA+i7lYj)WBn zh4)+csr4KPrtDwIc>qAvO%wxB3w-qJVMH~Fp7(n>X3yTXWwo2h%~PY_C_N7Xe$gTY z@kijewjN9oWRk7{Z7vMyTXVBde()v6Qwu~__OChQAUZ6RK`w6@>JeDGZXJ2^&sIL& z4{L^HijFoK#*m=5qazOn9WRwBCzg)?Fb|VlfP!)9$M2+p7h!w5ADE^xS=1;*ZHZ{L zy~9KNglM%n%7Yb#+D3uV2K&Yy8P2EPfz~+c1ZD-pUPFxk0o*Tu79nAOhoBEm96}w+_55>%@KHCOH%O~3r4cfR4+|znGlr3E z*^e4{^RSFe@l|e@Q{o_VpCKdJ7WCaj@q|*$lLwY)bu^Xn7u{r7B7L%dLh&`K$L!?E zeeSzE6!EBWgIrW^K~S)DQqOn>&K@jA^ggS}$t%WgO?tWQ?_U;#I*MISwl!(sp{crM zYHd4*o+6F4%O^G58o!-6)QSh#!*d$2v3zyGcMVw8Rmj{<^*xg2rk^(%T?fYRe?w;E)DUO%OHX_|NQt+}j(wSf9V!4+vu*s~L|fKb z1GDHn;24wTlXO;2-1Hq5>oExJ!z|!ZYnWY@4rl^WsUf}2rR?|(P5egI_!`v>xt0UA@UsJ%EmYHPW-!QDlr zIS=sM8Qj-?Nj|sMpz*64eE#x{V;yDdF<1Uxq_VHZY((>b{@b+I?<<*){&ZTpN`FZTz^WoD=-9m749s=St2cyCt}ZK4iFq_*P^XJ*Q~5UX^Cd3 zjkj$W04$^5yBs~p!g(L4$GEC0ZQq9o$StTU_?WxwI{ctz>aA^M1dgxRHkX0s^#*BR zZh2Ebcz@xBIaA}Y@uU;WC*G|m*bJyUhgeiO=o~wuj29=sM%}Pmf_qkd+&byR;)!ox zuzK}b6FYdLZJpkH*hBU%u-iNvBqVep4+#|Yx;?!D4)WW=GYZ(?; zWW~>4mOt(9?29`z;9lg_5aZC@bl^uiDmVCy9{A0CvUy@nKED&y%x^+l z^xNsM7>RA2!oO_d&(Io5PN73ASiTl7Sz=`QI9*pP^cYwvKE6rO z*!q)V{j!X`FKFb6UG>>Amt9MTrN-Zb`E2qEKyL|- zevkvg5YJ`q`kz09vyrNFc0(yb6yN&E)#EG1s6GHI*UfgEgzgKBC4Y@AHzTPC_7rwE z(s2U=Z(rZjmTR*!#y3g!S&)1Yuz9e8F$e`<4#&gXY~u3-u^3$*V~fEs4M7rkZ3}G> zoD&|SZyXdcl9UtAsEL?XYqG3i)LboY{~lX@aD1=2Ri%7!-+i+kbO+98$N0%%n1+o9 zZtU#;aWR!B^-X|53M^}YxFK|8&|{9D3`X@I8b2nNj~{eKlJ$uf)H6<39*39=(+D(~ z(}*3O=G@!ss5cw$qi;MTGLm3LAZOB-Ex44%IP7t&YWbKL6+Z;(8!_EnVPX}IAghtG z!-IU4J9li$Fsko@$f2Z{qw{KZ!}uX#B-x&LF$j+4{?e)DdnYbo3|dlyZMnlSD?!_t zT)~v&MdSN5Sp?w@V0f8Fd7Qr4@fFz_n&Vdrp{)z%_G_=rsiCbUFTA3cOmyrTZ`^l@|+9S?TWqfUOj zVk}tq#4gziRmH#r6IdwvJ5B7~6KI|P3)%~dJsOFR@4nwm;fbnb<4c@vAf7uskV0Z~ z(>~zyQYJoc1L~5b_#s%z(K>zq-t_vS@}iSv9`?PyFEI|6*k9BcxB$AP{Nqy{u|u2QLVP$L1^( zmPX|B;&BG6PPI#VMnqclzm1j@Dan0}eob%Kp>Pp$Xae7=;0?TW|Mv6Ob+#VCaO0(6 z5@?;G`-L}cF|;Ji{JFP!c7HvT92hX2;R?h?H1v;41GO!JwmQTndE4DDXIj408ifUJ zCr|?AR=mr+zqF$r%ytp>ZtuI@ojM8C-jp~`!vN$6yI|bl$EVi?$FRgvEu>C$Kv}rH z0zm?v0idSUtBm)Yjl4<@u2275MgjSH7^15dq>_?NN!5?h7eDkPhnpS_%B|U*REVPN zY89iO)KN}tZ(WznJH9_P@1ilAY?7eCWmuWNKF=QdDmAi2Z1J!FxO?<57eN`hL&*40+SjUp|?@DOjNCk`j*F|kI%UK0Ue|4 zaWm-oqagrZi$2MK@&p%e?QYqJx~m+{rc(4Ah~LmgXEoU5R_e>d$J`ns1pyulOPy&t zcPI93F<@0$l9@hJ%YZ@Ch%DC{0w=S`UIbjo%uD_3SeFKQXjkz)&0`H1KfN^j-DYs4+?Lm0aa`KnJKvOq6vKosA>K-*G;yrBs z9*;7fif1%*R#zm!xQPi!sTEz?;~<~0F%EE(Y(5mu%@@bLL0BcG8zti*7WGfSEfoov zb6GO9z^H`$qL^T?Y!F?<5mF<+6LKA5ZjRp<*C&wQcWeCB>@Y}&4Eo!|lqyNUBbq{4 zeamjx^B|E&=H?_;nm9rbDhvE;3R6pC7WXdeaQx=h@mFns35wF!rXf>UJz4q2iSSmhgsLkV&9m?A ze`6cv125dm?M9hH<|5Wp(@Q&!{WDc9M!T#*^~99%05AToZe-YQ)F5uxsRx+yu3?|Q zAgpnKfZMpFx^mY+9@xG`sd60JFg#Qcfk?G`{RIm^B0V2pm+YtmNVKWr%GIl_@i|=E z#MJ)BkCaAf2$4q13Py|D#g(z%ZZrc2p_mmM7`WyZw-nmc^!UOAj!Z20ZNLc{5fx;v zgH}tI{>G(VDh}@G7ez~?!|N|J>Io-q8*XeOD|yiH6{4wLe<9Nr$SvI19H3T`s!*Z@ zM@NOAJNKAntdkWr&V}vYu*d`C`1UY#_XPWqf6&7Ho{&>Bz7o3%^UwGR@ra#z? znMvqX;Y*W<;y8l4Lo6_~6lkTeh^`oc4aRd`CtG}h^VAR@HIc8mghrKbQ7(+|(S=IQ z$M-Z8fP_SB$Q~@Vo&^+-WvCio21Pz`ynzGDeLy!}F_p5mXEN=g>OSPRr(5iXi!gkGF&TNXYY-`;_;WNtukG86X~HS@PPArRYy z&5ihv667!y)`5_~9{6!oPCQN%Wg&FhZ0AzIQ-etm*>Zc3G%4f}9wu=g2wUqO6c||w z7>+|0XOeSx|esV3uHVjFvSMV=IcgL+9SP?WJDap!Ss{YSfp!cv0e@b;4Gf%() zaOOzU$*$0lkkg2yFbaF~kQImgA#Wl)SB{63JEzf#MLB?q+zrM$FVUXb`)a9vE?Y|O zL6kt8lf%Wlc+g3D1}^ERRS|#GEOZqZP15|`aA$3|YLcZXDHWoKMHFTMV*WJ@=|saF zCH>%NK|dhiNI2c^`*ApoR|||gYDV&qPvK`3{)rH*O+(jU{x&%Nv^W8Y846s5y|`n~ zDac#C_hyz)pv2$Y2x>IMjXJ<@-kwT55!85%WAUQFn%5YJfd&xV%z^a!1OpDv9$(uq zfziyT0n%X$(oiuCG{o>QX?`PWY8vQ)uGo8iPkxbI5B(+SL_S6UoQ-Z*Pww#-B)wVt z*(qu(+yIAMMIW5K-w-`Y5N^errkjwQ2hR5~^jRiZBSmc6G)*Io7XBU`B?ea-tV~ZS zp8!w2p(zK}!@nKK9!GOaP#ww1Yi=h_8Z~Y-duUlqQf(yK304Ly?wG**$j| zVE9cblXzDqi-~ob2 zZ>Q_X5jJRq7pxWY)jq&Z(F-hctgya{AR4nd4}4%R?7|y%?yP9A=AY2{)&yYZ z5)MSBl%*rnR7x8VIEUtr^}v`!8g)Z(%3&t62nH4rbebFJ6B?n5!4RE(lz;p9)(muG z_de^NQ-VUR1d06S5Fl_V2SnOeMr{K0P)QkCMh1VFHZ=w;eGE0mqtOSolflHa<95h;rY- zgdAQ%Zv)NucO1Dmg(hzCZ&)J(0hwBW2+4@mgSfN2`)nUPd;nR^$ge73Fv}k8oq-Mr zao%PJ8rVK7DIQpp_9)QR;o(*s~pxC0p z0F<-{aF2A33`wE!SkLMq^bxYuBY=96FN5YwV8e*%86;JatqzUT%8-wFyU!n6F|m4= zD%x#fNYO?HSH}<(8dgOnJ9yGbz-c9Vls&VoVpS(KIlxyyjiAKMYH2jl*=g5zLba zzZ4n3$5!*t#k8WzN%dNog*6fm_k_c+VZ?S>09(9>w5B;L5{8gxTFT=BFn*2o)fiZn zY)-e9Zwu8dA13P++!peieWElyMtL5YD<~SE45_AAo*)nfR1;Qx`TT{YuUcR|BGo@) z`POaQN!v~ zaz`ZC&QqwGUb&nbO*VIs(Lem48Lo)e188E#bc`Mt4#}}Y#yF6Hp-^$c(VgRvS=VSu z)A?vNB|0Jy$}w3rS}(&f0&oR}s8^XrE|_jWo)!4c=wem{e=-q@*asc05ogesq9Ngm z&SXOFpt)&2E%rZ6PaN}<{D#&e?*3EDX7Qna2&sSd3igfvKAuwz~k!X~Gr^Pt3ggFVZju%eklMpqNZ8@aidu zO%5*aOrR@y*Z6mzY7Ie%hEXDds31zlJ~)PnW~^Y0J<z%dMJPXg4dxkRkR7<$?dN{&6#s!xJdp$`h)lKgln5T{rRij{Z@>oO3Z(Z2fdU7o4S%y9Yr;~;Fonn1Evv2lYfQE}G& zT_@l+T9@O3oq==Hj50JoeuTIm(8ZGd4T~~hV$--a**ryu8bx8rEM7H;j!!$vAV{7i zC@cX=d3^G{za)&cIE}G*qmEbMFX34wRSVj+D{8<~H~v{DNo| zm=3B_$&Cu21T!Or4yvCDe77CMihE^cxw`kgeITk}VPQetbFbZ3M>G>=$A z<4CChCY>6sVf&x~zaGpS`gvp%6e=t~(GO+m4puY;0vjK-kks$*7cD~0!X!n_Qv{c& zJ#WcooqJ#ZzVz6?HL2F|Fw8+T*AEMP=j-AESCQlY{;d;dayJ;V<~Q8s2N+==erc)s z$3Klg1qJ}A;{a?29Hpruqbv+sB;TrC{^1Qmc|8+`&t4lT6{XA$vMW&rggS|bsI(~` z74NXBxfhb4SBn|>IKtcnt%-syiKggqXj-<^{L??v$p6pk0li3*a$Z{@J(pub8tJaa zP=Lv_sfMP*wQ*XC*k}ouJ%${zYv>F2h^L2#g8Nsywu(QQEp>P=|7-&bLUji+2YjAA zt28bRj4Eb$k@vS4T%k@POr>RNUH$>FfR~AYqY5N&s}D*(9c8GjsRAc$#qMh~Yp?pc z3%+pJUEdBGE`E>m-UDbvUO6^7nf2uBW*GUGw zaMb5<*JFI8`^tjJ{da~scS%WQiMysP=r}LMea8l9=kCbx&UNd8-goxqW=(>Z^j_;Q zYP?ej!>~)Lx);;98CEc4wg&!H3D9-rp|1u|2L%nh!A=(Jx<3s-M9~K^gSrE7Sz9ryuG3lJ-?no%Vuz3fK6fBOt;2gzkccTzSLv%Tg5mjFmhMd#idAP3Q23B9pGgD zg}%HWLDL#4Pc4pxuJGAi_sK{K&l_b)U@XS6uuED2h62rkY^dAFJGMhi8A zqLWoM=)hcR!^>=m{DPCxR#@@ubcwir9p1b6>{@EW8|W;5m&n=Jx_MBfvUL=7Wjl&L ztuHVZm)0A^Oj_Ozg}(6`k~AT4N&fxI7GG|OM#3TcF~cbdvqevFTEjrf3IOq->H<8uKu<{UybvcnppT7Uo=v-ZQa^~u~2ycF~}(Sy&QCK@mS3A;hW{V;wpX6 zH(|AeZsQk%UAc4nIuG&37<$t*po2o_X5pL$9}JM)lC%pGpAbhu?nkeKL4V%WAWawh zId>K#0KV>%M7eZy-xN;;c(Vad&Se=*+TUMhn!N~fv$QfF@aNSfaA|#CqH4R)4zOQb z-wM4POiUd(6k(W`j{$sC1e`%i!c^h}KoyIZv1pZv(&I*qXCju|T)TE{SBCu0uWvUS z-ng;b&lprGHKh};GH@A z!l)60>I>e!Z7`n6?*@W+IBgO~Ze3j+zdhFJdXJ!|wAN=B67VOeXgKzXY7~#SOs&?1 zg~E4uxWZ+~$37DW=MW2y0{^eZz_h#@T8&N`+%r zoTmLX^nodhIzc#jJJhN_=@g<>0mCsq0I|f0bv&Y*-}X8u=WG9cWFdXKnzqMj!G_#Y zP7M@mpkF1=B zyaP7DqgvN$nAM3rM6>ZT>wKJh&p0{`*mU2sh|L;k5d_t*j_{SQiYm{=X)UXCjYY5*D6$#7F-1#$dvCS-Kfe|2Wo?IN`k}ak zGU5UL!9q~>u|cXQb1qs<_`9Mvoj8U@Ix&6+X2qvuP(@m z`=A47Eq<5=yh5a8eQ~C_89X$qM-^CP%WUHLVQ(4$j)=Wqci*|O6I9Th*0@Yxhu9>b zcOn;Q;38xdo0{^aKdUjCJRE!rL0B5$jShqr0YBh%c7zW&HpSHT(*W$Yi<=CZsxWee zhl!-l|DVUZ_hQf$qjxRHV@59w7$wdz6M)eXo$YdmY!lhD&VZ14nG%jYxxIYHDrtBk z2wxS{J|94yb)BohtSL+JOhmbxldWLiPst3UT*p;sk+Yg0W+KRDvP|8xL9Mg%%XO`U zcfCZ)F||`HLl9HFzqS$?mnP8)j3;4-L?4V2HyUqk*?x!T)NBO$Y2fn>AxjB$+ZQVr zFZPaGoBD(bK_;YyW7pWjW6-F{+K-2UDPcQETMF+qzYU$1FqGMV7(thj9thGqXrA63 z3OUmpAe~8Lym*=+YNR){Vn9nr)`$zW!_YKExHG8V6%;yx<6)9PK{co$uSV7ORTy|g z;Bo`V)`^%w1Ia_4+Vp|kkhX;kc=kTWKvN|e4GWt8G5&akop5@W{41jXV?esjNj-Q> zoMP(cq#JNbCB5OBIh5Ls4DFtHg*4c*Q6S@q(Ck^JgW7U60?j*$mcwGgU0A-g}dNAN&&^TX+)|} zb3ZzRsY4JxD>`b@KP*yoWWbMbyZ(LzgCOQ6utf(N3IN1FZR2NMT30ne?5p-fp@e-Z z0{U__=AlF`T(IAQDp5@I17jdCGHN8uo|~+*nW|IT*1*Iz4Y>>1?S=9WQ#%&-Agn`r zyGu+476;95k2Rg9CQEVDD&NIyZzP?sgS}))<3d65hwN0lvQ!&mnpif{Ej1DZf_F)k z#YvD^aHK?{w0dPL%MKWd!U)-F&?{*!4O+I~AAGKu!<3`HS7h*+S)C-=9$C;DHc{g& zwGE(d`d{sRXIPZy*6yfLBeoqQ8W1JMLE<0~F^DJv#uyz?P(bQXRC<%5AWG9j-J4hk z2NXewiu5YI#||hc3J9X0QX()QAP^YJcdwZ-XJ6O1^PThKT-W(?@-NKz&htKHt$W?; zUh4^2yN-nIEjK}=zeW_WPpb#MG-Uje@R2Q}dbk4ubLa?ji{4e$hpeqpy(o@hO>Asp zkX=9qnbeg;N5FPaSmhZiaHYaZYT+G#icEmi?1_otQesWUR;Hc0)}m-fu1()i_$!Q- z1}ct+aE=PYcHxn!X^^=WyuW#cDplesfsHafL_o^{!TeJ^LEbH0@1adqAk+XIB%X9N z!FA}tYTR;P-E_F>ck`i|#VNsG#AiqYC@3NHz<~nMP_w7n1Z`+2a0Hu2JOPqz3y-8= zxv?_Fa0EO|{%$NW2j--fvN^aUMkwN-F5U?s1t@idaxl&xgDqB{$c5m2EnfLl#|qMS zI}As0=Ll*6$wGEM`R!+S5Ks%2T;6rde&;51GNIZ&xEvPMJwS}dy@AdW4Ag-s=QT;OHFQvq;?eL*QQf7h7kr5oodozQYpi+(b2?@y^I^p-w*9g$8 z78)8#poIK(&@+u`pS-UOz0H@fTC7kMs22c^G12z3us z$qdec>OdeUx_i-Yps(*K2ofr~*omA8QaMBjQ$t=PxKO(Ly!)Tue*K98`&yQ^46t$+ z0xY#5IQTAn%OqMe0I7}@fG}Tv*=`N)oO-UT*?*rJ`qNfEIbSvv#Ts4rmmK(h<1&fh zSrwwKTc&y=F3zRm;;PBg;*IArlMTh$#s-gi<5y_};XeHxv39=SJft~@24`PuU}U8w*Q^ixA}8nzT8HAqVHT#>n~sGJp@H- z>X(ARY6O3YTwAq%1!%g}=>@8}$j^7c`K2b83XtDwtF)f*Lh~C#54&X66d|{?`VQQ% z*X>#IttS>+ZRb9(t7}IxvKkc8RUBJMj%DS@AKSDdAoDP}-3g$i=@50u7~IzE)YE&G z3k6)k_Byc!O8Ko^|1OJWoh@6oaOLLjE^OuM_BT`q=CH!|n_hmr?&a#mmIrr}+?44D zlE(Y#*2S?LrNta6g%zTQOp+{nU@$Fc6DolM+Op9iW`)*Bxqc?{iIS!bjk$P+EnRUz)`5$CqOu#<->fu3%G(Ob+h^PCuO zp4=oWc=3SC^lpDtu2#}FcfO@d>}Ns&ewGF7aTO{Kui{v)WK;V^*|CuL1etr+qx8Hm~utyx?a^4Gtt(huwpZGtIfhiUBMAD^xC5QDRH-JSA$_qL@xoM=Djh`3PVy_|i(!+SUOaM3Az>!Mp zMI6cA1w!5O6Zax8F!>*D&sKQ(K*oCq@KWhig~WL- z2=+ldyu2gQTaPA(DfPYMF1__{emS0$xD2V4NAlQEOx?QgCfe0XFT?RlawMk;%dHwu z$f8fN9~dVt6MBa#gt$?S=6x?_dZKNH0|cM?T+GELt+6-rXQzJNY(u7_u?UX59ifX( zrtQuK?<#e9VBBx>5<3m8t)uCG2L}h)To_Y(5P{Q8YD;DbPp#uKG&{f2>))}F#|K}e zxv@Ka%Q4A4xe&6b;9d6F39kLBQ~~lvA~`J6gjqmiAgvLzSn|&rxBW1SX{YP zwLt;>J@mO6$fS=%PmV+%PjZeHFBls>wF3KC;brv1$GB&?2b$kXIPYKAa?~TQ^=K)>hlR9}LGsd0`pl*_3QC)Kiab327|h z=e$&k7BqFRUXpFHI{fAY-g2AxXI7b5`Po4-6UkI>J-;QheC3seoZSMC)(hM0arF_K zi;I+6v3aY2MkP`c`QnwbM-vD++U?vM0JrLB*C&x0KOUcw@|d<1L_YHDiIM6cdl<4+ zJe}z#bvfc)u{N;pS8Jn2Wa47s-p|;g&vTzwgO*6Ytqy z0HIq^i=oRs*j1;vZJD2K#`J7M3%Be&B@=^bCOJWZm(A@bm$*A-Oe|TlWFWkBEH=~9 z;$_K-s-f}sgqt1LlGo7@KwU>aCXswI0CfVUnu3OE3r0E01oG3Puc zM2=m9h~i^c{ed%r6noJ)5?_Px#ncajV?p4J%x+Oj$a=iF*OSdN_vBCkat9ihTkt2y zO%#2&T_IhWot>@TTQooqQ%%i|5Y0jSky6-nM985`G~S3W^tCt^5fMQfP70$En%~>A zl^-EIvCKD4N~D91ZVU##Q-b&|)|FsQEwb93nL=+8llVANeJxTPy4>gW^&N)Mqej0= z_N6v9Hu}F?()*e0{vkB`-;L~!Aavk;q+;d%K1 zB0psl%R_t4!kuZLhcaqh3JQc{KyTx*-=15|?Fj(gX>)A#QT;CUTL7OTYg)ZML!@! zoDr+|Kq+2#j+6>>+tNcuC~Gm%j;-A+G8lX4YL@}{Q+f8f2HVhXWV@%E+)YeONGL}W z5fz)V;73l;IKBZ^=GL8g8*c!bZw+tk6fcd=G=>zdJBq3t0ZWE$F*PI4>+Inzy<}*c0o?9Mn0P{y2Je2UZwuJU=3+l;E^$rYOpp3Tbku8|@hBVziR91B z#hx$40457}EEzVu22~`HJmaqqY7F%^@o*;ylkM|j@65`zWh!Dc#lzKw)=o$Bq*q@N z&8~4Awm8O-;@XQTo^CuXMGP^F9S|&B)(kkpwZ#!)qaOmz(x+e68#tB3$3CNV`t+m+ z{d3Zt#cqW~G&8QYJkP|Y-MV6I#&UlFZ(rSE$m69)sN707ZCM|1vQTjgd zSI!`TvzEXT*A{uk$7ESOwmvBk&!Orv{|~=ktZZYJBg`|jP$)Sr*{A#&Y<;lt^^ws+ z#DR*b|A=28e%1-(-P>^ogAVr~m=1VOQg^M&{Yw0SS>!Yx^R_~5)U9ph@q%!IFK)U$ntv7Yd3?`mr_DBB3F z?^K0joua@c-it7^vrG70joO@fy^!rf7IF^05P$?BcLq*zfoP8hd6BEw~cTUXQKM+UQ`TY!Bz$920V%&o}DCm~~BHeN=5pI~_x5U1| zO&fYOvaEpck1mH2|AK!}g^k-Tns>LmXhdr6L_$%lqI1N|BJp&Kwr=pceOQ;oELj9g zQx(ke;VV}FAhNk1dtPl6d790KM6f>V3s6x=%Ug1plY-L8D|q?i3g#ssB8(c*Uk`NG z&SSOOgP*<($odfT5Xp3MbbJV14+dT7;5WljO0>Dw*>2?$M-uybF1XQPdPkLFEuykq7HeGD3X%d;VPz6>W+ z*b?$OZKQt=9UknjK=BXXK;(|nOc(AABUT&s;;OuBu{@pFifIQEs}L71?PUMB*{IIh z#pM*Rp|>Hmxz~S2Lh)|c-xh1tAN4UA~dj>Zt_tXRy`=T7_FUfG_&T42kVIV zmp6p!SBd0SYY_dhz+tly#?9%uE<5f>lOzJLS(|a#Zy=WX)L5D>#ouc+k>M z4?4BfXFD|6p-f~RwhMWHZ}VaKS#DF5v*w83&?t`AXIs)aO5#3|xF?SW^j`C(+!|rY=rV{4IgV?ZN3^${J znJJt8Z25V*F}Dfr7I{Mee`YpvPjW=3t1%wvS{$lfS*NI6{Lx_fA!uUdA#K$YBIJ$y zEW_o8v@OKv=MHKljhyr^efQgW>6JS!orlyyjty#xJ32c0LuYW#cHxPCauV%O^J zEyX~aEIxwRr1R1-DdO$97x7B~6mS=qe%%=lRB$Qu{J}9Oz^jqF3T1+Br(>=8d|-=( z$f`QtK>)JL?B-qs$RoIPXB77a0Jb@~y6(ud&HpDVv;yNhPUVQK&6E2m#LlWI!SrfA z_r?^x;tjzgjsJFaN#iu}RI33fRs8&z%znkvY?qf%%7vR>{cU7gb8wh=cHa>!2aBRu z&Q`*Je)hm#pjFqAs7M zCR%?V8&t#-1v4E;AixaG1I#xy!>05jhL9Ih7@z@?YX>$Tl@odq$C*TH04uAqRZx|Y zU9#g2{)Y$5z#PHFKvw||vyW&xripem|HlEl1wBB5*+)_lI2$(t**2SRTAbw|{`Gx8 zb-ayE=rH*jh%lKqK;f3q&bSu!-ERP{udnc9h^D3xl-0RzAaXVrEuuQ$ORN<-@)7f~ z_|UOxsj8~Fv}pv3Ufj~DgmW7_n0XQPq0=eFhw3fC)6BjIqJ4bYNoF2a6$ z7)iy(NystR;;n?XI}@#navUeiYQ(QX9ut~?XeK+s$h>o}qvBrNc)=S2>~FdjFs)a- z9ULb3%@Zj#B|z+kES~v|W{T;A$HX=L_ZdM%aU#3>9xngizP)q<0(%8Ma!H)Q*LUKw&+C6gXdaBoTZ&W4H293FxB-K zElV{vGz^8VFV!=kEb$8=agk-X;-}Lz5!_aow`xeOO_HJ^=TX85(dTw!6$u2HC;wz~ zPdt|AP$8R`Dfr@e-V9{AG>YRE4ia8}Y%q7dAi3;4UCG7rT#$HU!{F2fLBb=R*7@$EJStXmR10 zQaekG0b`;zwqM#`gA0_H_z&FzS8eGH@#Ru}{@ie7E}iY*No(-oSc!41cTMF#i$xuCIk4X|gS%rj@sj8vGE za1m#>1qk3uIR1V2@^Mr{nczj#!D_L({+ueSB;4LqGFxs^I$G+|8OR~Fe&YzyOU*xL6oUlxMpl!t~!lv`BYeS`(y*L2F)W?zFu`^4t5Y+`GFuA z;HCM?k-x^@qBI5`t#yLPv^$7cFz6Vk8_&u6nb_9g5E$DIuC7u!;ZHxLZk$F+;mvjO zC{65->%?+7@aSCy_cMQ8%BnZn$jbN0<|UtNW36PTjoOUHY8i|1fa$S#&#g`yQ?>BI5?6If2-M zbGkv;S3{y1%A{QdX@RB3Rz{W`PLuDV)u!j?sx^J*+vK_JY1@-|rlP{U7#&fd#7Mj- zw5g2$;SWYT5nU{{{EWe`gykAQTN!ya9blwn5V$iOE9uG~LKy6t(x6GmnSu4`uZQ3M z>!MC9cqa#><5w5RG)@tnSZ!XWs7;MFPlO&AMyo!3^63u=1=nUXVhZS~%0(Z3px^)n z3txjvd?hn&w1gsA^30!crmg3lR=cL1U{aBYy$&89YA*khlRjU|@2hd9qTq zo)ZcZ+J<-5BHdLh zSF_;tFx@cLgstHR#>V{P-#^`$SJ?(8doG(xS>AwNUvg*QUAeYz@BT{>v*%{ATR3xW ziTC~PMWPpv6{9W@lWq@n7Mpu!U8Cb=!JK6p>&pc}eWs!Nr#KddL|8o1Ww;rtRw=YI zBVgInjf=s~vBSn{c3Tv7JH;V4%jSYy%DIQ>eXo2KH=?`*|4z~CZ(i|GotZqlAME04-)LXidAqEwOP`PKcKHuqPgP=& zo&jkal+TTN)dxTs`V-d{aiv?v-wYVam`y?^8$fMIWs?J%ak9B!MX&$ejO-D?iIr@) zrh$@SqOhSD)MRG@hv&Tb&oA3G59t&TRILbBD^QywSZ$v5ExR8Eg;Iy`Wfj0kw{T}r z_8H;tK1}%@67Y#Rj$0`si9?5@ zhg5V2cJ^ab=blHqLOZZ`m78ZSy+}20vHFKBX&*`t$ ze1qkD)S)??zo_>pL=BZ=Y9xkUq_vole3r}@BW9e%(?RXdfb{TMuhxegwAx-3UWrxE z;R2(m_T)-ay!=(s@L*k;>OMC|K#j5FlnIkIRmQi+VV;A3$o_HrH;z*nF~wvbpV-3; zKd}7mF=q27gc5(KguRvC?S|+p%$BA$ac<7{+tnJKNnd&2}5MLyz+~hqLb1} z3Nm||8#bpKS8Vu&UpP2n8go2J|FI$gkFP`{iO z{{BWNVSef_LDYoz-7mJaqD19TY&$RDCTP_T$`>jXc5|WgvIq1$((DPct6H3ejzh>| zhKXNGI`yOcY$OOKB)WERalcOo9TK*D)J1Up>mW0$Lf~};yR%KfdPE0wX!OaGCnblT z8Z@7)mC6m1IcI12Gr>&!oDLL}G;~TM;QnAIWPNjWqThoADW~V5!gXT%gC;tHQa{K5 zbGIQdP|U5v`Q(AnddI6PK^`qfIt#hw!s21f& zn#hBMA030st^}%Y!y!+Pyy4&E`9SejbXeX`y{L|~RcDWsO`FeoU&8pX*JCEM82mcQ z=937dGGLPYJGja#WE;S6S@gV0z%UN&%O~voViW! zDCCqu?U6#UxBt$xFBnq zL7j%tN374U2mFFs;!BhzgI}TDn+XOIkEs|0BWL0b@tbYErEh@s8O6bI$A5Xc{HyP9 z8YSXM!+@O5T9^xbo4geI61XBKaf(@v0}%m^p``Z(F30V_7~rOJd-=mIL6da)bKI`wm@wYrjbDRd^r zL;39yF0Y3toZ_O4^$bv!GgygZWaiEXN$ii`=Y?KGfnI4%b%91w;g5Gtty$F8Uy=3{ zoP??0Bsyh`=xFjov<$iZkih_}G>_k!V7B4U)s?S$Pp_>q?dhNkdRe{2WBrw~d9UfH zywBrRDz07D)?ZT}10`J>)jC}sFM?6yjaO~xN**rOf<&M~ppR|?Dgsliu^*JVyz#K? zAMZq1T^}acoN{I(Rj}v`rg+ z+5~+}CXA5sKrygDlV3*^n+woMA#f$Xl~83V^|@p zy>nf>&M;!6$;~_F5_{BEGVH$V^6FWajBSzqXLE}H4|Iz)v0@dC*GS`pBF(EX{KkPK zt)feXP;&SlI}2iRQ}rSEn$Z|phcn6i3{y#BQep$D)H@JdwbNU|Ugzs{Y(e zq(!};DVbk)vpkj_I4PX08M2?|H(&5}|JSd@2y)};7$phc(P>4JB42JiZ;6eL1I z<&GWX76Ce9Y(V@vg4x!8-j)!&KynndTH$Rar7P;zN_dCQG@m$_8M+)@ZF~>$kpdrX zYiOG6q!lceSY!d?%6KG1fq*V} zo83yqA8YJp42jQXd7on{idGc5J6`oxqJ}Qaq8dISr{)N1YSM2Hn+!`&(N_Hj0TK6< z&Zx6VN9Esl{t~&>?B)IpRz&lx#swW<;O?;axvY*OzU<@7%A);fcUCg;UOy-T_Lc-h zAaf320V45+Un9q2-!{eharX1!>#+|~LWnOh0J!8MB;RGR&jNkehI$zHF2}SDOv4&!iEAd5zBS4*292_MVJ_qb%1>DW09jE%vnlLBdrE=T3Au zLu;CaYiw$D3dMd+b(vx;ERax7A~&U%=(&9Bd`=bFHOZmuzcwkO?ivXm8%Z>V(;rUd zsJfes5{+0iaKFxkI4_7Yl0VL?$muR@9(RuoAOR&b7*#--w_tggx>JjZbw<5gwUk3- zhS7J!n*$~V&1qRP=P!$dh2H=;mlD~5ZC{Pz4Nm?^qWIui!aMf^OOtSo!g;K+JajU8 zcH=K-2Z8J+U(XWsW>zZcE^P)h3;QZJ?5lHO&x@`i*|>%*D=FV>%P+x;s*pMcW40n2 zVezcb(Mj#qLBq4?*8-~sC+h&-M?q#1sE+YOC}e8cf~!29kfEl$m>2enJ$_Z!($W%g z|FD#tMixo6Xuil%<+mK49io)$fwY<_G@DUGOpJVt(;}@{93-HGq?VtO;lkUh0M_dx z^;Z9f20|4nG?Jo3-dKImxC#1Maq^W24y6`!CX#0zX_eQ?Z+fRSaJjk_UZ8^F6B)oE ziUlYzBFqmPy|^Fh&{dQyqZMweBebu&G*@-$xiDjB2GMzW70N#

    U0evL2XnC3Wa* zyl{U)E7*5xa$@pgn=VeVV~TTyOTXGqNCA^n+?(=bydEwG{b6<#?~U~0OY$vt$bitg zdLt-WXgQA|a9n?HJoX~I$f^P+!!+c-(R8Xp86{$10tEQ) k|A5K+kD>1WwfK2FXr<-8j)I(hd>RgWtKN^PKiC}qFH4tx@Bjb+ literal 0 HcmV?d00001 diff --git a/docs/articles/potential-optimizers_files/header-attrs-2.5/header-attrs.js b/docs/articles/potential-optimizers_files/header-attrs-2.5/header-attrs.js new file mode 100644 index 0000000..dd57d92 --- /dev/null +++ b/docs/articles/potential-optimizers_files/header-attrs-2.5/header-attrs.js @@ -0,0 +1,12 @@ +// Pandoc 2.9 adds attributes on both header and div. We remove the former (to +// be compatible with the behavior of Pandoc < 2.8). +document.addEventListener('DOMContentLoaded', function(e) { + var hs = document.querySelectorAll("div.section[class*='level'] > :first-child"); + var i, h, a; + for (i = 0; i < hs.length; i++) { + h = hs[i]; + if (!/^h[1-6]$/i.test(h.tagName)) continue; // it should be a header h1-h6 + a = h.attributes; + while (a.length > 0) h.removeAttribute(a[0].name); + } +}); diff --git a/docs/authors.html b/docs/authors.html index afbda59..6894aeb 100644 --- a/docs/authors.html +++ b/docs/authors.html @@ -71,7 +71,7 @@ rco - 1.0.1 + 1.0.2

    @@ -96,6 +96,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -114,6 +117,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -126,6 +132,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -185,7 +194,7 @@

    Authors

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/index.html b/docs/index.html index d4d8af5..180f890 100644 --- a/docs/index.html +++ b/docs/index.html @@ -32,7 +32,7 @@ rco - 1.0.1 + 1.0.2
    @@ -57,6 +57,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -75,6 +78,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -87,6 +93,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -122,17 +131,18 @@

    Make your R code run faster! rco analyzes your code and applies different optimization strategies that return an R code that runs faster.

    The rco project, from its start to version 1.0.0, was made possible by a Google Summer of Code 2019 project.

    -

    Thanks to the kind mentorship of Dr. Yihui Xie and Dr. Nicolás Wolovick.

    +

    Thanks to the kind mentorship of Dr. Yihui Xie and Dr. Nicolás Wolovick.

    Installation

    Install the current released version of rco from CRAN:

    - +

    Or install the development version from GitHub:

    - +
    if (!require("remotes")) {
+  install.packages("remotes")
+}
+remotes::install_github("jcrodriguez1989/rco", dependencies = TRUE)

    @@ -158,20 +168,23 @@

    1. Optimize some .R code files
    -
    optimize_files(c("file_to_optimize_1.R", "file_to_optimize_2.R"))
    +
    +optimize_files(c("file_to_optimize_1.R", "file_to_optimize_2.R"))
    1. Optimize some code in a character vector
    -
    code <- paste(
+
    +code <- paste(
   "code_to_optimize <- 8 ^ 8 * 1918",
   "cto <- code_to_optimize * 2",
-  sep = "\n"
-)
-optimize_text(code)
    
+  sep = "\n"
+)
+optimize_text(code)
    1. Optimize all .R code files into a folder
    -
    optimize_folder("~/myfolder_to_optimize", recursive = FALSE)
    +
    +optimize_folder("~/myfolder_to_optimize", recursive = FALSE)

    @@ -179,7 +192,8 @@

    Example

    Suppose we have the following code:

    -
    code <- paste(
+
    +code <- paste(
   "# I want to know my age in seconds!",
   "years_old <- 29",
   "days_old <- 365 * years_old # leap years don't exist",
@@ -191,79 +205,84 @@ 
    
   "} else {",
   '  print("Meh!")',
   "}",
-  sep = "\n"
-)
    
+  sep = "\n"
+)

    We can automatically optimize it by doing:

    -
    opt_code <- optimize_text(code, iterations = 1)
    - +
    +opt_code <- optimize_text(code, iterations = 1)
    +
    ## Optimization number 1
+
+## # I want to know my age in seconds!
+## years_old <- 29
+## days_old <- 365 * 29 # leap years don't exist
+## hours_old <- 24 * days_old
+## seconds_old <- 3600 * hours_old
+## 
+## if (seconds_old > 10e6) {
+##   print("Whoa! More than a million seconds old, what a wise man!")
+## } else {
+##   print("Meh!")
+## }

    After one optimization pass we can see that it has only propagated the years_old variable. Another pass:

    -
    opt_code <- optimize_text(opt_code, iterations = 1)
    - +
    +opt_code <- optimize_text(opt_code, iterations = 1)
    +
    ## Optimization number 1
+
+## # I want to know my age in seconds!
+## years_old <- 29
+## days_old <- 10585 # leap years don't exist
+## hours_old <- 24 * 10585
+## seconds_old <- 3600 * hours_old
+## 
+## if (seconds_old > 10e6) {
+##   print("Whoa! More than a million seconds old, what a wise man!")
+## } else {
+##   print("Meh!")
+## }

    Now, it has folded the days_old variable, and then propagated it. Another pass:

    -
    opt_code <- optimize_text(opt_code, iterations = 1)
    - +
    +opt_code <- optimize_text(opt_code, iterations = 1)
    +
    ## Optimization number 1
+
+## # I want to know my age in seconds!
+## years_old <- 29
+## days_old <- 10585 # leap years don't exist
+## hours_old <- 254040
+## seconds_old <- 3600 * 254040
+## 
+## if (seconds_old > 10e6) {
+##   print("Whoa! More than a million seconds old, what a wise man!")
+## } else {
+##   print("Meh!")
+## }

    It has folded the hours_old variable, and then propagated it. Another pass:

    -
    opt_code <- optimize_text(opt_code, iterations = 1)
    - +
    +opt_code <- optimize_text(opt_code, iterations = 1)
    +
    ## Optimization number 1
+
+## # I want to know my age in seconds!
+## years_old <- 29
+## days_old <- 10585 # leap years don't exist
+## hours_old <- 254040
+## seconds_old <- 914544000
+## 
+## if (914544000 > 10e6) {
+##   print("Whoa! More than a million seconds old, what a wise man!")
+## } else {
+##   print("Meh!")
+## }

    It has folded the seconds_old variable, and then propagated it into the if condition. Another pass:

    -
    opt_code <- optimize_text(opt_code, iterations = 1)
    - +
    +opt_code <- optimize_text(opt_code, iterations = 1)
    +
    ## Optimization number 1
+
+## # I want to know my age in seconds!
+## years_old <- 29
+## days_old <- 10585 # leap years don't exist
+## hours_old <- 254040
+## seconds_old <- 914544000
+## 
+## print("Whoa! More than a million seconds old, what a wise man!")

    Now, it has folded the if condition, and as it was TRUE it just kept its body, as testing the condition or the else clause were dead code. So, optimize_text function has automatically detected constant variables, constant foldable operations, and dead code. And returned an optimized R code.

    @@ -317,7 +336,7 @@

    Developers

    Dev status

    • CRAN status
      • -
    • Lifecycle: stable
      • +
    • Lifecycle: stable
    • Travis build status
    • AppVeyor build status
    • Coverage status
      • @@ -332,7 +351,7 @@

      Dev status

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/news/index.html b/docs/news/index.html index 4e20c9b..dee1cf1 100644 --- a/docs/news/index.html +++ b/docs/news/index.html @@ -71,7 +71,7 @@ rco - 1.0.1 + 1.0.2
    @@ -96,6 +96,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -114,6 +117,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -126,6 +132,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -159,25 +168,34 @@

    Changelog

    Source: NEWS.md
    -
    +
    +

    +rco 1.0.2 2021-04-17 +

    +
      +
    • Adding Memory Allocation optimizer.
      • +
    • Adding Conditional Threading optimizer.
      • +
    +
    +

    -rco 1.0.1 2020-02-08 +rco 1.0.1 2020-02-08

    • Bug fixing.
    -
    +

    -rco 1.0.0 2020-02-06 +rco 1.0.0 2020-02-06

    • Version bump to submit to CRAN.
    -
    +

    -rco 0.1.8 Unreleased +rco 0.1.8 Unreleased

    • Adding Shiny GUIs, called with the rco_gui function.

      • @@ -186,65 +204,65 @@

    • Adding new optimizers list.

    -
    +

    -rco 0.1.7 Unreleased +rco 0.1.7 Unreleased

    • Adding Loop-invariant Code Motion optimizer.
    -
    +

    -rco 0.1.6 Unreleased +rco 0.1.6 Unreleased

    • Adding Dead Expression Elimination optimizer.
    -
    +

    -rco 0.1.5 Unreleased +rco 0.1.5 Unreleased

    • Adding Common Subexpression Elimination optimizer.
    -
    +

    -rco 0.1.4 Unreleased +rco 0.1.4 Unreleased

    • Adding Dead Store Elimination optimizer.
    -
    +

    -rco 0.1.3 Unreleased +rco 0.1.3 Unreleased

    • Adding Dead Code Elimination optimizer.
    -
    +

    -rco 0.1.2 Unreleased +rco 0.1.2 Unreleased

    • Adding Constant Propagation optimizer.
    -
    +

    -rco 0.1.1 Unreleased +rco 0.1.1 Unreleased

    • Adding Constant Folding optimizer.
    -
    +

    -rco 0.1.0 Unreleased +rco 0.1.0 Unreleased

    • Creating basic rco package skeleton.
      • @@ -267,7 +285,7 @@

      Contents

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/pkgdown.css b/docs/pkgdown.css index c01e592..1273238 100644 --- a/docs/pkgdown.css +++ b/docs/pkgdown.css @@ -244,14 +244,14 @@ nav[data-toggle='toc'] .nav .nav > .active:focus > a { .ref-index th {font-weight: normal;} -.ref-index td {vertical-align: top;} +.ref-index td {vertical-align: top; min-width: 100px} .ref-index .icon {width: 40px;} .ref-index .alias {width: 40%;} .ref-index-icons .alias {width: calc(40% - 40px);} .ref-index .title {width: 60%;} .ref-arguments th {text-align: right; padding-right: 10px;} -.ref-arguments th, .ref-arguments td {vertical-align: top;} +.ref-arguments th, .ref-arguments td {vertical-align: top; min-width: 100px} .ref-arguments .name {width: 20%;} .ref-arguments .desc {width: 80%;} diff --git a/docs/pkgdown.yml b/docs/pkgdown.yml index f72d023..97a28d3 100644 --- a/docs/pkgdown.yml +++ b/docs/pkgdown.yml @@ -1,15 +1,19 @@ -pandoc: 2.3.1 -pkgdown: 1.5.1 +pandoc: 2.11.2 +pkgdown: 1.6.1 pkgdown_sha: ~ articles: contributing-an-optimizer: contributing-an-optimizer.html docker-readme: docker-readme.html + gsoc2020-final-report: gsoc2020-final-report.html opt-common-subexpr: opt-common-subexpr.html + opt-cond-thread: opt-cond-thread.html opt-constant-folding: opt-constant-folding.html opt-constant-propagation: opt-constant-propagation.html opt-dead-code: opt-dead-code.html opt-dead-expr: opt-dead-expr.html opt-dead-store: opt-dead-store.html opt-loop-invariant: opt-loop-invariant.html -last_built: 2020-05-03T14:07Z + opt-memory-alloc: opt-memory-alloc.html + potential-optimizers: potential-optimizers.html +last_built: 2021-07-14T11:56Z diff --git a/docs/reference/Rplot001.png b/docs/reference/Rplot001.png new file mode 100644 index 0000000000000000000000000000000000000000..17a358060aed2a86950757bbd25c6f92c08c458f GIT binary patch literal 1011 zcmeAS@N?(olHy`uVBq!ia0y~yV0-|=9Be?5+AI5}0x7m6Z+90U4Fo@(ch>_c&H|6f zVg?3oArNM~bhqvg0|WD9PZ!6KiaBo&GBN^{G%5UFpXcEKVvd5*5Eu=C0SJK)8A6*F U7`aXvEC5;V>FVdQ&MBb@00SN#Z2$lO literal 0 HcmV?d00001 diff --git a/docs/reference/all_optimizers.html b/docs/reference/all_optimizers.html index 476c4af..46304c1 100644 --- a/docs/reference/all_optimizers.html +++ b/docs/reference/all_optimizers.html @@ -41,6 +41,7 @@ @@ -81,7 +83,7 @@ rco - 1.0.1 + 1.0.2
    @@ -106,6 +108,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -124,6 +129,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -136,6 +144,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -172,6 +183,7 @@

    All optimizers list.

    List of all the optimizer functions:

    - 
    all_optimizers
    + 
    all_optimizers

    Format

    -

    An object of class list of length 7.

    +

    An object of class list of length 9.

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/reference/generate_files_opt_report.html b/docs/reference/generate_files_opt_report.html new file mode 100644 index 0000000..f6740a6 --- /dev/null +++ b/docs/reference/generate_files_opt_report.html @@ -0,0 +1,220 @@ + + + + + + + + +Report possible optimizations in `.R` files. — generate_files_opt_report • rco + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
    +
    + + + + +
    + +
    +
    +
    +

    Report possible optimizations in `.R` files.

    + Source: R/reporting-fun-files.R + +
    + +
    +

    Report possible optimizations in `.R` files.

    +
    + + 
    generate_files_opt_report(files, optimizers = rco:::all_optimizers)
    + +

    Arguments

    + + + + + + + + + + +
    files

    A character vector with paths to files to optimize.

    optimizers

    A named list of optimizer functions.

    + + +
    + +
    + + +
    +
    +

    Developed by Juan Cruz Rodriguez.

    +
    + +
    +

    Site built with pkgdown 1.6.1.

    +
    + +
    +
    + + + + + + + + diff --git a/docs/reference/generate_folder_opt_report.html b/docs/reference/generate_folder_opt_report.html new file mode 100644 index 0000000..bc2bb9a --- /dev/null +++ b/docs/reference/generate_folder_opt_report.html @@ -0,0 +1,235 @@ + + + + + + + + +Report possible optimizations in folder containing `.R` files. — generate_folder_opt_report • rco + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
    +
    + + + + +
    + +
    +
    +
    +

    Report possible optimizations in folder containing `.R` files.

    + Source: R/reporting-fun-folder.R + +
    + +
    +

    Report possible optimizations in folder containing `.R` files.

    +
    + + 
    generate_folder_opt_report(
+  folder,
+  optimizers = all_optimizers,
+  pattern = "\\.R$",
+  recursive = TRUE
+)
    + +

    Arguments

    + + + + + + + + + + + + + + + + + + +
    folder

    Path to a directory with files to optimize.

    optimizers

    A named list of optimizer functions.

    pattern

    An optional regular expression. Only file names which match +the regular expression will be optimized.

    recursive

    A logical value indicating whether or not files +in subdirectories of `folder` should be optimized as well.

    + + +
    + +
    + + +
    +
    +

    Developed by Juan Cruz Rodriguez.

    +
    + +
    +

    Site built with pkgdown 1.6.1.

    +
    + +
    +
    + + + + + + + + diff --git a/docs/reference/generate_text_opt_report.html b/docs/reference/generate_text_opt_report.html new file mode 100644 index 0000000..4d2df8a --- /dev/null +++ b/docs/reference/generate_text_opt_report.html @@ -0,0 +1,220 @@ + + + + + + + + +Report possible optimizations in `.R` code snippet. — generate_text_opt_report • rco + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
    +
    + + + + +
    + +
    +
    +
    +

    Report possible optimizations in `.R` code snippet.

    + Source: R/reporting-fun-text.R + +
    + +
    +

    Report possible optimizations in `.R` code snippet.

    +
    + + 
    generate_text_opt_report(text, optimizers = all_optimizers)
    + +

    Arguments

    + + + + + + + + + + +
    text

    A character vector with the code to optimize.

    optimizers

    A named list of optimizer functions.

    + + +
    + +
    + + +
    +
    +

    Developed by Juan Cruz Rodriguez.

    +
    + +
    +

    Site built with pkgdown 1.6.1.

    +
    + +
    +
    + + + + + + + + diff --git a/docs/reference/index.html b/docs/reference/index.html index 38c79cf..87caddd 100644 --- a/docs/reference/index.html +++ b/docs/reference/index.html @@ -71,7 +71,7 @@ rco - 1.0.1 + 1.0.2
    @@ -96,6 +96,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -114,6 +117,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -126,6 +132,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -186,6 +195,24 @@

    generate_files_opt_report()

    + +

    Report possible optimizations in `.R` files.

    + + + +

    generate_folder_opt_report()

    + +

    Report possible optimizations in folder containing `.R` files.

    + + + +

    generate_text_opt_report()

    + +

    Report possible optimizations in `.R` code snippet.

    + +

    max_optimizers

    @@ -198,6 +225,12 @@

    opt_cond_thread()

    + +

    Optimizer: Conditional Threading.

    + +

    opt_constant_folding()

    @@ -234,6 +267,12 @@

    opt_memory_alloc()

    + +

    Optimizer: Memory Allocation.

    + +

    optimize_files()

    @@ -275,7 +314,7 @@

    Contents

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/reference/max_optimizers.html b/docs/reference/max_optimizers.html index 158bd24..40a0ca8 100644 --- a/docs/reference/max_optimizers.html +++ b/docs/reference/max_optimizers.html @@ -43,6 +43,7 @@ @@ -83,7 +85,7 @@ rco - 1.0.1 + 1.0.2
    @@ -108,6 +110,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -126,6 +131,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -138,6 +146,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -176,6 +187,7 @@

    Max optimizers list.

    List of all the optimizer functions, with maximum optimization techniques enabled. Note that using this optimizers could change the semantics of the program!

    - 
    max_optimizers
    + 
    max_optimizers

    Format

    -

    An object of class list of length 7.

    +

    An object of class list of length 9.

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/reference/opt_common_subexpr.html b/docs/reference/opt_common_subexpr.html index ffea2b5..99ce145 100644 --- a/docs/reference/opt_common_subexpr.html +++ b/docs/reference/opt_common_subexpr.html @@ -73,7 +73,7 @@ rco - 1.0.1 + 1.0.2
    @@ -98,6 +98,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -116,6 +119,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -128,6 +134,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -167,7 +176,7 @@

    Optimizer: Common Subexpression Elimination.

    Carefully examine the results after running this function!

    - 
    opt_common_subexpr(texts, n_values = 2, in_fun_call = FALSE)
    + 
    opt_common_subexpr(texts, n_values = 2, in_fun_call = FALSE)

    Arguments

    @@ -190,14 +199,26 @@

    Arg

    Examples

    - 
    code <- paste(
+    
    code <- paste(
   "a <- b * c + g",
   "d = b * c * e",
-  sep = "\n"
-)
-cat(opt_common_subexpr(list(code))$codes[[1]])
    #> cs_1 <- b * c
+  sep = "\n"
+)
+cat(opt_common_subexpr(list(code))$codes[[1]])
+
    #> cs_1 <- b * c
 #> a <- cs_1 + g
-#> d = cs_1 * e
    
+#> d = cs_1 * e
    
+heron_formula <- paste(
+  "area <- (a/2 + b/2 +c/2) * (a/2 + b/2 + c/2 - a) * (a/2 + b/2 + c/2 - b) *",
+  "  (a/2 + b/2 + c/2 - c)",
+  "area <- sqrt(area)",
+  sep = '\n'
+)
+cat(opt_common_subexpr(list(heron_formula))$codes[[1]])
+
    #> cs_1 <- a/2 + b/2 +c/2
+#> area <- (cs_1) * (cs_1 - a) * (cs_1 - b) *
+#>   (cs_1 - c)
+#> area <- sqrt(area)
    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/reference/opt_cond_thread.html b/docs/reference/opt_cond_thread.html new file mode 100644 index 0000000..fa066ff --- /dev/null +++ b/docs/reference/opt_cond_thread.html @@ -0,0 +1,246 @@ + + + + + + + + +Optimizer: Conditional Threading. — opt_cond_thread • rco + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +
    +
    + + + + +
    + +
    +
    +
    +

    Optimizer: Conditional Threading.

    + Source: R/opt-cond-thread.R + +
    + +
    +

    Performs one conditional threading pass. +Carefully examine the results after running this function!

    +
    + + 
    opt_cond_thread(code)
    + +

    Arguments

    +
    + + + + + +
    code

    A list of character vectors with the code to optimize.

    + + +

    Examples

    + 
    code <- paste(
+  "num <- sample(100, 1)",
+  "even_sum <- 0",
+  "odd_sum_a <- 0",
+  "odd_sum_b <- 0",
+  "if (num %% 2 == 1) {",
+  "  odd_sum_a <- odd_sum_a + num",
+  "}",
+  "if (num %% 2 == 1) {",
+  "  odd_num_b <- odd_num_b + num",
+  "}",
+  "if (!(num %% 2 == 1)) {",
+  "  even_sum <- even_sum + num",
+  "}",
+  sep = "\n"
+)
+cat(opt_cond_thread(list(code))$codes[[1]])
+
    #> num <- sample(100, 1)
+#> even_sum <- 0
+#> odd_sum_a <- 0
+#> odd_sum_b <- 0
+#> if(num %% 2 == 1) {
+#> odd_sum_a <- odd_sum_a + num
+#>  odd_num_b <- odd_num_b + num
+#> }else {
+#>  even_sum <- even_sum + num
+#> }
    +
    + +
    + + +
    +
    +

    Developed by Juan Cruz Rodriguez.

    +
    + +
    +

    Site built with pkgdown 1.6.1.

    +
    + +
    +
    + + + + + + + + diff --git a/docs/reference/opt_constant_folding.html b/docs/reference/opt_constant_folding.html index 4b40f83..6f53d01 100644 --- a/docs/reference/opt_constant_folding.html +++ b/docs/reference/opt_constant_folding.html @@ -73,7 +73,7 @@ rco - 1.0.1 + 1.0.2
    @@ -98,6 +98,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -116,6 +119,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -128,6 +134,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -167,7 +176,7 @@

    Optimizer: Constant Folding.

    Carefully examine the results after running this function!

    - 
    opt_constant_folding(texts, fold_floats = FALSE, in_fun_call = FALSE)
    + 
    opt_constant_folding(texts, fold_floats = FALSE, in_fun_call = FALSE)

    Arguments

    @@ -190,12 +199,13 @@

    Arg

    Examples

    - 
    code <- paste(
+    
    code <- paste(
   "i <- 320 * 200 * 32",
   "x <- i * 20 + 100",
-  sep = "\n"
-)
-cat(opt_constant_folding(list(code))$codes[[1]])
    #> i <- 2048000
+  sep = "\n"
+)
+cat(opt_constant_folding(list(code))$codes[[1]])
+
    #> i <- 2048000
 #> x <- i * 20 + 100
    
   
    
   
 
 
    
-  
    Site built with pkgdown 1.5.1.
    
+  
    Site built with pkgdown 1.6.1.
    
 
    
 
       
diff --git a/docs/reference/opt_constant_propagation.html b/docs/reference/opt_constant_propagation.html
index c4b016e..bfa1826 100644
--- a/docs/reference/opt_constant_propagation.html
+++ b/docs/reference/opt_constant_propagation.html
@@ -73,7 +73,7 @@
       
       
         rco
-        1.0.1
+        1.0.2
       
     
 
@@ -98,6 +98,9 @@
     
  • 
       Common Subexpression Elimination
     
    • 
+    
  • 
+      Conditional Threading
+    
    • 
     
  • 
       Constant Folding
     
    • 
@@ -116,6 +119,9 @@
     
  • 
       Loop-invariant Code Motion
     
    • 
+    
  • 
+      Memory Allocation
+    
    • 
   
 
 
  • 
@@ -128,6 +134,9 @@
     
  • 
       Contributing an Optimizer
     
    • 
+    
  • 
+      Potential Optimizers
+    
    • 
     
  • 
       Docker Image
     
    • 
@@ -167,7 +176,7 @@ 
    Optimizer: Constant Propagation.
    
 Carefully examine the results after running this function!
    
     
 
-    
    opt_constant_propagation(texts, in_fun_call = FALSE)
    
+    
    opt_constant_propagation(texts, in_fun_call = FALSE)
    
 
     
    Arguments
    @@ -185,17 +194,28 @@

    Arg

    Examples

    - 
    code <- paste(
+    
    code <- paste(
   "i <- 170",
   "x <- -170",
   "y <- x + 124",
   "z <- i - 124",
-  sep = "\n"
-)
-cat(opt_constant_propagation(list(code))$codes[[1]])
    #> i <- 170
+  sep = "\n"
+)
+cat(opt_constant_propagation(list(code))$codes[[1]])
+
    #> i <- 170
 #> x <- -170
 #> y <- -170 + 124
-#> z <- 170 - 124
    
+#> z <- 170 - 124
    
+hemisphere_vol <- paste(
+  "pi <- 3.141593 ",
+  "radius <- 25 ",
+  "hemis_vol <- 2/3 * pi * radius ^ 3 ",
+  sep = "\n"
+)
+cat(opt_constant_propagation(list(hemisphere_vol))$codes[[1]])
+
    #> pi <- 3.141593
+#> radius <- 25
+#> hemis_vol <- 2/3 * 3.141593 * 25 ^ 3
    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/reference/opt_dead_code.html b/docs/reference/opt_dead_code.html index 88d4bbb..8593376 100644 --- a/docs/reference/opt_dead_code.html +++ b/docs/reference/opt_dead_code.html @@ -73,7 +73,7 @@ rco - 1.0.1 + 1.0.2 @@ -98,6 +98,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -116,6 +119,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -128,6 +134,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -167,7 +176,7 @@

    Optimizer: Dead Code Elimination.

    Carefully examine the results after running this function!

    - 
    opt_dead_code(texts)
    + 
    opt_dead_code(texts)

    Arguments

    @@ -180,14 +189,15 @@

    Arg

    Examples

    - 
    code <- paste(
+    
    code <- paste(
   "while (TRUE) {",
   "  break",
   "  dead_code()",
   "}",
-  sep = "\n"
-)
-cat(opt_dead_code(list(code))$codes[[1]])
    #> while (TRUE) {
+  sep = "\n"
+)
+cat(opt_dead_code(list(code))$codes[[1]])
+
    #> while (TRUE) {
 #>   break
 #> }
    
   
    
@@ -205,7 +215,7 @@ 
    Contents
    
 
 
 
    
-  
    Site built with pkgdown 1.5.1.
    
+  
    Site built with pkgdown 1.6.1.
    
 
    
 
       
diff --git a/docs/reference/opt_dead_expr.html b/docs/reference/opt_dead_expr.html
index 668e4eb..18eb9d8 100644
--- a/docs/reference/opt_dead_expr.html
+++ b/docs/reference/opt_dead_expr.html
@@ -73,7 +73,7 @@
       
       
         rco
-        1.0.1
+        1.0.2
       
     
 
@@ -98,6 +98,9 @@
     
  • 
       Common Subexpression Elimination
     
    • 
+    
  • 
+      Conditional Threading
+    
    • 
     
  • 
       Constant Folding
     
    • 
@@ -116,6 +119,9 @@
     
  • 
       Loop-invariant Code Motion
     
    • 
+    
  • 
+      Memory Allocation
+    
    • 
   
 
 
  • 
@@ -128,6 +134,9 @@
     
  • 
       Contributing an Optimizer
     
    • 
+    
  • 
+      Potential Optimizers
+    
    • 
     
  • 
       Docker Image
     
    • 
@@ -167,7 +176,7 @@ 
    Optimizer: Dead Expression Elimination.
    
 Carefully examine the results after running this function!
    
     
 
-    
    opt_dead_expr(texts)
    
+    
    opt_dead_expr(texts)
    
 
     
    Arguments
    @@ -180,14 +189,15 @@

    Arg

    Examples

    - 
    code <- paste(
+    
    code <- paste(
   "foo <- function(x) {",
   "  x ^ 3",
   "  return(x ^ 3)",
   "}",
-  sep = "\n"
-)
-cat(opt_dead_expr(list(code))$codes[[1]])
    #> foo <- function(x) {
+  sep = "\n"
+)
+cat(opt_dead_expr(list(code))$codes[[1]])
+
    #> foo <- function(x) {
 #>   return(x ^ 3)
 #> }
    
   
    
@@ -205,7 +215,7 @@ 
    Contents
    
 
 
 
    
-  
    Site built with pkgdown 1.5.1.
    
+  
    Site built with pkgdown 1.6.1.
    
 
    
 
       
diff --git a/docs/reference/opt_dead_store.html b/docs/reference/opt_dead_store.html
index bb875aa..f91e7e8 100644
--- a/docs/reference/opt_dead_store.html
+++ b/docs/reference/opt_dead_store.html
@@ -73,7 +73,7 @@
       
       
         rco
-        1.0.1
+        1.0.2
       
     
 
@@ -98,6 +98,9 @@
     
  • 
       Common Subexpression Elimination
     
    • 
+    
  • 
+      Conditional Threading
+    
    • 
     
  • 
       Constant Folding
     
    • 
@@ -116,6 +119,9 @@
     
  • 
       Loop-invariant Code Motion
     
    • 
+    
  • 
+      Memory Allocation
+    
    • 
   
 
 
  • 
@@ -128,6 +134,9 @@
     
  • 
       Contributing an Optimizer
     
    • 
+    
  • 
+      Potential Optimizers
+    
    • 
     
  • 
       Docker Image
     
    • 
@@ -167,7 +176,7 @@ 
    Optimizer: Dead Store Elimination.
    
 Carefully examine the results after running this function!
    
     
 
-    
    opt_dead_store(texts)
    
+    
    opt_dead_store(texts)
    
 
     
    Arguments
    @@ -180,17 +189,35 @@

    Arg

    Examples

    - 
    code <- paste(
+    
    code <- paste(
   "foo <- function() {",
   "  x <- 128 ^ 2",
   "  return(TRUE)",
   "}",
-  sep = "\n"
-)
-cat(opt_dead_store(list(code))$codes[[1]])
    #> foo <- function() {
+  sep = "\n"
+)
+cat(opt_dead_store(list(code))$codes[[1]])
+
    #> foo <- function() {
 #>   128 ^ 2
 #>   return(TRUE)
-#> }
    
+#> }
    
+code <- paste(
+  "sinpi <- function() {",
+  "  pi <- 3.1415",
+  "  e <- 2.718",
+  "  phi <- 1.618",
+  "  sin(pi)",
+  "}",
+  sep = "\n"
+)
+cat(opt_dead_store(list(code))$codes[[1]])
+
    #> sinpi <- function() {
+#>   pi <- 3.1415
+#>   2.718
+#>   1.618
+#>   sin(pi)
+#> }
    
+
    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/reference/opt_loop_invariant.html b/docs/reference/opt_loop_invariant.html index e50b12b..786137c 100644 --- a/docs/reference/opt_loop_invariant.html +++ b/docs/reference/opt_loop_invariant.html @@ -73,7 +73,7 @@ rco - 1.0.1 + 1.0.2 @@ -98,6 +98,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -116,6 +119,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -128,6 +134,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -167,7 +176,7 @@

    Optimizer: Loop-invariant Code Motion.

    Carefully examine the results after running this function!

    - 
    opt_loop_invariant(texts)
    + 
    opt_loop_invariant(texts)

    Arguments

    @@ -180,16 +189,17 @@

    Arg

    Examples

    - 
    code <- paste(
+    
    code <- paste(
   "i <- 0",
   "while (i < n) {",
   "  x <- y + z",
   "  a[i] <- 6 * i + x * x",
   "  i <- i + 1",
   "}",
-  sep = "\n"
-)
-cat(opt_loop_invariant(list(code))$codes[[1]])
    #> i <- 0
+  sep = "\n"
+)
+cat(opt_loop_invariant(list(code))$codes[[1]])
+
    #> i <- 0
 #> if (i < n) {
 #>   x <- y + z
 #> } 
@@ -212,7 +222,7 @@ 
    Contents
    
 
    
 
 
    
-  
    Site built with pkgdown 1.5.1.
    
+  
    Site built with pkgdown 1.6.1.
    
 
    
 
       
diff --git a/docs/reference/opt_memory_alloc.html b/docs/reference/opt_memory_alloc.html
new file mode 100644
index 0000000..755d689
--- /dev/null
+++ b/docs/reference/opt_memory_alloc.html
@@ -0,0 +1,231 @@
+
+
+
+  
+  
+
+
+
+Optimizer: Memory Allocation. — opt_memory_alloc • rco
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+  
+
+  
+    
    
+      
    
+      
+
+      
+
+      
    
+
+
    
+  
    
+    
    
+    
    Optimizer: Memory Allocation.
    
+    Source: R/opt-memory-alloc.R
+    
+    
    
+
+    
    
+    
    Performs one memory allocation pass.
+Carefully examine the results after running this function!
    
+    
    
+
+    
    opt_memory_alloc(code)
    
+
+    
    Arguments
    
+
    + + + + + +
    code

    A list of character vectors with the code to optimize.

    + + +

    Examples

    + 
    code <- paste(
+  "v <- NULL",
+  "for (i in 1:5) {",
+  "  v[i] <- i^2",
+  "}",
+  sep = "\n"
+)
+cat(opt_memory_alloc(list(code))$codes[[1]])
+
    #> v <- vector(length = 5)
+#> for (i in 1:5) {
+#>   v[i] <- i^2
+#> }
    +
    + +
    + + +
    +
    +

    Developed by Juan Cruz Rodriguez.

    +
    + +
    +

    Site built with pkgdown 1.6.1.

    +
    + +
    +
    + + + + + + + + diff --git a/docs/reference/optimize_files.html b/docs/reference/optimize_files.html index d61d51f..a4212f0 100644 --- a/docs/reference/optimize_files.html +++ b/docs/reference/optimize_files.html @@ -76,7 +76,7 @@ rco - 1.0.1 + 1.0.2
    @@ -101,6 +101,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -119,6 +122,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -131,6 +137,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -173,12 +182,12 @@

    Optimize `.R` files.

    information to rco.

    - 
    optimize_files(
-  files,
-  optimizers = all_optimizers,
-  overwrite = FALSE,
-  iterations = Inf
-)
    + 
    optimize_files(
+  files,
+  optimizers = all_optimizers,
+  overwrite = FALSE,
+  iterations = Inf
+)

    Arguments

    @@ -219,7 +228,7 @@

    Contents

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/reference/optimize_folder.html b/docs/reference/optimize_folder.html index 44fa6af..2094704 100644 --- a/docs/reference/optimize_folder.html +++ b/docs/reference/optimize_folder.html @@ -76,7 +76,7 @@ rco - 1.0.1 + 1.0.2 @@ -101,6 +101,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -119,6 +122,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -131,6 +137,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -173,14 +182,14 @@

    Optimize a folder with `.R` files.

    information to rco.

    - 
    optimize_folder(
-  folder,
-  optimizers = all_optimizers,
-  overwrite = FALSE,
-  iterations = Inf,
-  pattern = "\\.R$",
-  recursive = TRUE
-)
    + 
    optimize_folder(
+  folder,
+  optimizers = all_optimizers,
+  overwrite = FALSE,
+  iterations = Inf,
+  pattern = "\\.R$",
+  recursive = TRUE
+)

    Arguments

    @@ -231,7 +240,7 @@

    Contents

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/reference/optimize_text.html b/docs/reference/optimize_text.html index d025a94..551b65b 100644 --- a/docs/reference/optimize_text.html +++ b/docs/reference/optimize_text.html @@ -73,7 +73,7 @@ rco - 1.0.1 + 1.0.2 @@ -98,6 +98,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -116,6 +119,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -128,6 +134,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -167,7 +176,7 @@

    Optimize text containing code.

    Carefully examine the results after running this function!

    - 
    optimize_text(text, optimizers = all_optimizers, iterations = Inf)
    + 
    optimize_text(text, optimizers = all_optimizers, iterations = Inf)

    Arguments

    @@ -203,7 +212,7 @@

    Contents

    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/docs/reference/rco_gui.html b/docs/reference/rco_gui.html index f83c56b..a3c8594 100644 --- a/docs/reference/rco_gui.html +++ b/docs/reference/rco_gui.html @@ -72,7 +72,7 @@ rco - 1.0.1 + 1.0.2 @@ -97,6 +97,9 @@
  • Common Subexpression Elimination
    • +
  • + Conditional Threading +
  • Constant Folding
    • @@ -115,6 +118,9 @@
  • Loop-invariant Code Motion
    • +
  • + Memory Allocation +
  • @@ -127,6 +133,9 @@
  • Contributing an Optimizer
    • +
  • + Potential Optimizers +
  • Docker Image
    • @@ -165,7 +174,7 @@

    rco GUI selector.

    Starts the selected rco Graphical User Interface (GUI).

    - 
    rco_gui(option)
    + 
    rco_gui(option)

    Arguments

    @@ -183,9 +192,10 @@

    Arg

    Examples

    ## Start the GUI
-if (FALSE) {
-rco_gui("code_optimizer")
-}
    +if(FALSE){ +rco_gui("code_optimizer") +} +
    -

    Site built with pkgdown 1.5.1.

    +

    Site built with pkgdown 1.6.1.

    diff --git a/man/all_optimizers.Rd b/man/all_optimizers.Rd index 71f57e4..0bac778 100644 --- a/man/all_optimizers.Rd +++ b/man/all_optimizers.Rd @@ -5,7 +5,7 @@ \alias{all_optimizers} \title{All optimizers list.} \format{ -An object of class \code{list} of length 7. +An object of class \code{list} of length 9. } \usage{ all_optimizers @@ -13,6 +13,7 @@ all_optimizers \description{ List of all the optimizer functions: \itemize{ + \item Conditional Threading \code{\link{opt_cond_thread}} \item Constant Folding \code{\link{opt_constant_folding}} \item Constant Propagation \code{\link{opt_constant_propagation}} \item Dead Code Elimination \code{\link{opt_dead_code}} @@ -20,6 +21,7 @@ List of all the optimizer functions: \item Dead Expression Elimination \code{\link{opt_dead_expr}} \item Common Subexpression Elimination \code{\link{opt_common_subexpr}} \item Loop-invariant Code Motion \code{\link{opt_loop_invariant}} + \item Memory Allocation \code{\link{opt_memory_alloc}} } } \keyword{datasets} diff --git a/man/generate_files_opt_report.Rd b/man/generate_files_opt_report.Rd new file mode 100644 index 0000000..7036581 --- /dev/null +++ b/man/generate_files_opt_report.Rd @@ -0,0 +1,16 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/reporting-fun-files.R +\name{generate_files_opt_report} +\alias{generate_files_opt_report} +\title{Report possible optimizations in `.R` files.} +\usage{ +generate_files_opt_report(files, optimizers = rco:::all_optimizers) +} +\arguments{ +\item{files}{A character vector with paths to files to optimize.} + +\item{optimizers}{A named list of optimizer functions.} +} +\description{ +Report possible optimizations in `.R` files. +} diff --git a/man/generate_folder_opt_report.Rd b/man/generate_folder_opt_report.Rd new file mode 100644 index 0000000..93e3cf3 --- /dev/null +++ b/man/generate_folder_opt_report.Rd @@ -0,0 +1,27 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/reporting-fun-folder.R +\name{generate_folder_opt_report} +\alias{generate_folder_opt_report} +\title{Report possible optimizations in folder containing `.R` files.} +\usage{ +generate_folder_opt_report( + folder, + optimizers = all_optimizers, + pattern = "\\\\.R$", + recursive = TRUE +) +} +\arguments{ +\item{folder}{Path to a directory with files to optimize.} + +\item{optimizers}{A named list of optimizer functions.} + +\item{pattern}{An optional regular expression. Only file names which match +the regular expression will be optimized.} + +\item{recursive}{A logical value indicating whether or not files +in subdirectories of `folder` should be optimized as well.} +} +\description{ +Report possible optimizations in folder containing `.R` files. +} diff --git a/man/generate_text_opt_report.Rd b/man/generate_text_opt_report.Rd new file mode 100644 index 0000000..9c30219 --- /dev/null +++ b/man/generate_text_opt_report.Rd @@ -0,0 +1,16 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/reporting-fun-text.R +\name{generate_text_opt_report} +\alias{generate_text_opt_report} +\title{Report possible optimizations in `.R` code snippet.} +\usage{ +generate_text_opt_report(text, optimizers = all_optimizers) +} +\arguments{ +\item{text}{A character vector with the code to optimize.} + +\item{optimizers}{A named list of optimizer functions.} +} +\description{ +Report possible optimizations in `.R` code snippet. +} diff --git a/man/max_optimizers.Rd b/man/max_optimizers.Rd index 2c61ffa..ddc426c 100644 --- a/man/max_optimizers.Rd +++ b/man/max_optimizers.Rd @@ -5,7 +5,7 @@ \alias{max_optimizers} \title{Max optimizers list.} \format{ -An object of class \code{list} of length 7. +An object of class \code{list} of length 9. } \usage{ max_optimizers @@ -15,6 +15,7 @@ List of all the optimizer functions, with maximum optimization techniques enabled. Note that using this optimizers could change the semantics of the program! \itemize{ + \item Conditional Threading \code{\link{opt_cond_thread}} \item Constant Folding \code{\link{opt_constant_folding}} \item Constant Propagation \code{\link{opt_constant_propagation}} \item Dead Code Elimination \code{\link{opt_dead_code}} @@ -22,6 +23,7 @@ Note that using this optimizers could change the semantics of the program! \item Dead Expression Elimination \code{\link{opt_dead_expr}} \item Common Subexpression Elimination \code{\link{opt_common_subexpr}} \item Loop-invariant Code Motion \code{\link{opt_loop_invariant}} + \item Memory Allocation \code{\link{opt_memory_alloc}} } } \keyword{datasets} diff --git a/man/opt_common_subexpr.Rd b/man/opt_common_subexpr.Rd index 3327c2d..e61c693 100644 --- a/man/opt_common_subexpr.Rd +++ b/man/opt_common_subexpr.Rd @@ -26,4 +26,12 @@ code <- paste( sep = "\n" ) cat(opt_common_subexpr(list(code))$codes[[1]]) + +heron_formula <- paste( + "area <- (a/2 + b/2 +c/2) * (a/2 + b/2 + c/2 - a) * (a/2 + b/2 + c/2 - b) *", + " (a/2 + b/2 + c/2 - c)", + "area <- sqrt(area)", + sep = '\n' +) +cat(opt_common_subexpr(list(heron_formula))$codes[[1]]) } diff --git a/man/opt_cond_thread.Rd b/man/opt_cond_thread.Rd new file mode 100644 index 0000000..fc08791 --- /dev/null +++ b/man/opt_cond_thread.Rd @@ -0,0 +1,34 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/opt-cond-thread.R +\name{opt_cond_thread} +\alias{opt_cond_thread} +\title{Optimizer: Conditional Threading.} +\usage{ +opt_cond_thread(code) +} +\arguments{ +\item{code}{A list of character vectors with the code to optimize.} +} +\description{ +Performs one conditional threading pass. +Carefully examine the results after running this function! +} +\examples{ +code <- paste( + "num <- sample(100, 1)", + "even_sum <- 0", + "odd_sum_a <- 0", + "odd_sum_b <- 0", + "if (num \%\% 2 == 1) {", + " odd_sum_a <- odd_sum_a + num", + "}", + "if (num \%\% 2 == 1) {", + " odd_num_b <- odd_num_b + num", + "}", + "if (!(num \%\% 2 == 1)) {", + " even_sum <- even_sum + num", + "}", + sep = "\n" +) +cat(opt_cond_thread(list(code))$codes[[1]]) +} diff --git a/man/opt_constant_propagation.Rd b/man/opt_constant_propagation.Rd index 3e0c48b..6ca60c1 100644 --- a/man/opt_constant_propagation.Rd +++ b/man/opt_constant_propagation.Rd @@ -25,4 +25,12 @@ code <- paste( sep = "\n" ) cat(opt_constant_propagation(list(code))$codes[[1]]) + +hemisphere_vol <- paste( + "pi <- 3.141593 ", + "radius <- 25 ", + "hemis_vol <- 2/3 * pi * radius ^ 3 ", + sep = "\n" +) +cat(opt_constant_propagation(list(hemisphere_vol))$codes[[1]]) } diff --git a/man/opt_dead_store.Rd b/man/opt_dead_store.Rd index da04e8f..652ef43 100644 --- a/man/opt_dead_store.Rd +++ b/man/opt_dead_store.Rd @@ -22,4 +22,16 @@ code <- paste( sep = "\n" ) cat(opt_dead_store(list(code))$codes[[1]]) + +code <- paste( + "sinpi <- function() {", + " pi <- 3.1415", + " e <- 2.718", + " phi <- 1.618", + " sin(pi)", + "}", + sep = "\n" +) +cat(opt_dead_store(list(code))$codes[[1]]) + } diff --git a/man/opt_memory_alloc.Rd b/man/opt_memory_alloc.Rd new file mode 100644 index 0000000..f13d5c7 --- /dev/null +++ b/man/opt_memory_alloc.Rd @@ -0,0 +1,25 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/opt-memory-alloc.R +\name{opt_memory_alloc} +\alias{opt_memory_alloc} +\title{Optimizer: Memory Allocation.} +\usage{ +opt_memory_alloc(code) +} +\arguments{ +\item{code}{A list of character vectors with the code to optimize.} +} +\description{ +Performs one memory allocation pass. +Carefully examine the results after running this function! +} +\examples{ +code <- paste( + "v <- NULL", + "for (i in 1:5) {", + " v[i] <- i^2", + "}", + sep = "\n" +) +cat(opt_memory_alloc(list(code))$codes[[1]]) +} diff --git a/tests/testthat/test-opt_cond_thread.R b/tests/testthat/test-opt_cond_thread.R new file mode 100644 index 0000000..6738a47 --- /dev/null +++ b/tests/testthat/test-opt_cond_thread.R @@ -0,0 +1,421 @@ +context("opt_cond_thread") + +test_that("General Case without comments", { + code <- paste( + "num <- sample(1:100, 1)", + "even_sum <- 0", + "odd_sum_a <- 0", + "odd_sum_b <- 0", + "if (num %% 2 == 1) {", + " odd_sum_a <- odd_sum_a + num", + "}", + "if (num %% 2 == 1) {", + " odd_sum_b <- odd_sum_b + num", + "}", + "if (!(num %% 2 == 1)) {", + " even_sum <- even_sum + num", + "}", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "num <- sample(1:100, 1)", + "even_sum <- 0", + "odd_sum_a <- 0", + "odd_sum_b <- 0", + "if(num %% 2 == 1) {", + "odd_sum_a <- odd_sum_a + num", + " odd_sum_b <- odd_sum_b + num", + "}else {", + " even_sum <- even_sum + num", + "}", + sep = "\n" + )) +}) + +test_that("General Case without comments (Alternative)", { + code <- paste( + "num <- sample(1:100, 1)", + "even_sum_a <- 0", + "odd_sum <- 0", + "even_sum_b <- 0", + "if (!(num %% 2 == 1)) {", + " even_sum_a <- even_sum_a + num", + "}", + "if (!(num %% 2 == 1)) {", + " even_sum_b <- even_sum_b + num", + "}", + "if (num %% 2 == 1) {", + " odd_sum <- odd_sum + num", + "}", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "num <- sample(1:100, 1)", + "even_sum_a <- 0", + "odd_sum <- 0", + "even_sum_b <- 0", + "if(!(num %% 2 == 1)) {", + "even_sum_a <- even_sum_a + num", + " even_sum_b <- even_sum_b + num", + "}else {", + " odd_sum <- odd_sum + num", + "}", + sep = "\n" + )) +}) + +test_that("General Case with comments", { + code <- paste( + "num <- sample(1:100, 1)", + "even_sum_a <- 0", + "even_sum_b <- 0", + "# Hello, I am a comment.", + "odd_sum_a <- 0", + "odd_sum_b <- 0", + "if (num %% 2 == 1) { # I'm another comment, and I won't go easily", + " odd_sum_a <- odd_sum_a + num", + "}", + "# This is another comment", + "if (num %% 2 == 1) {", + " odd_sum_b <- odd_sum_b + num", + "}", + "if (!(num %% 2 == 1)) {", + " even_sum_a <- even_sum_a + num", + "}", + "if (!(num %% 2 == 1)) {", + " even_sum_b <- even_sum_b + num", + "}", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "num <- sample(1:100, 1)", + "even_sum_a <- 0", + "even_sum_b <- 0", + "# Hello, I am a comment.", + "odd_sum_a <- 0", + "odd_sum_b <- 0", + "if(num %% 2 == 1) {", + "# I'm another comment, and I won't go easily", + " odd_sum_a <- odd_sum_a + num", + " odd_sum_b <- odd_sum_b + num", + "}else {", + " even_sum_a <- even_sum_a + num", + " even_sum_b <- even_sum_b + num", + "}", + "# This is another comment", + "", + sep = "\n" + )) +}) + +test_that("Code snippet without any IF-ELSE blocks", { + code <- paste( + "for(i in 1:100) {", + " print(i)", + "}", + "rnorm(1)", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, code) +}) + +test_that("Code Blocks containing IF-ELSE blocks within loops", { + code <- paste( + "num_vec <- sample(1:5, 10, replace = TRUE)", + "ones <- 0", + "others <- 0", + "for (i in num_vec) {", + " if (i == 1) {", + " ones <- ones + 1", + " }", + " if (i != 1) {", + " others <- others + 1", + " }", + "}", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "num_vec <- sample(1:5, 10, replace = TRUE)", + "ones <- 0", + "others <- 0", + "for (i in num_vec) {", + "if (i == 1) {", + " ones <- ones + 1", + " }else {", + " others <- others + 1", + "}}", + sep = "\n" + )) +}) + +test_that("Code block containing IF-ELSE blocks inside a function", { + code <- paste( + "is_even <- function(a) {", + " if (a %% 2 == 1) {", + " return (FALSE)", + " }", + " if (!(a %% 2 == 1)) {", + " return (TRUE)", + " }", + " if (!(a %% 2 == 1)) {", + " print(\"Yaay\")", + " }", + "}", + "is_even(150)", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "is_even <- function(a) {", + "if (a %% 2 == 1) {", + " return (FALSE)", + " }else {", + " return (TRUE)", + " print(\"Yaay\")", + "}}", + "is_even(150)", + sep = "\n" + )) +}) + +test_that("Code Block containing a nested IF-ELSE block", { + code <- paste( + "num <- sample(1:100, 1)", + "even_sum <- 0", + "odd_sum <- 0", + "count <- 0", + "if (even_sum + odd_sum < 10) {", + " if (num %% 2 == 1) {", + " odd_sum <- odd_sum + num", + " }", + " if (!(num %% 2 == 1)) {", + " even_sum <- even_sum + num", + " }", + " count <- count + 1", + "}", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "num <- sample(1:100, 1)", + "even_sum <- 0", + "odd_sum <- 0", + "count <- 0", + "if (even_sum + odd_sum < 10) {", + "if (num %% 2 == 1) {", + " odd_sum <- odd_sum + num", + " }else {", + " even_sum <- even_sum + num", + "} count <- count + 1", + "}", + sep = "\n" + )) +}) + +test_that("Conversion to else due to Greater than equal to logic", { + code <- paste( + "num <- sample(1:100, 1)", + "big <- 0", + "small <- 0", + "if (num >= 50) {", + " big <- big + 1", + "}", + "if (num < 50) {", + " small <- small + 1", + "}", + "print(big)", + "print(small)", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "num <- sample(1:100, 1)", + "big <- 0", + "small <- 0", + "if (num >= 50) {", + " big <- big + 1", + "}else {", + " small <- small + 1", + "}", + "print(big)", + "print(small)", + sep = "\n" + )) +}) + +test_that("Conversion to else due to Greater than equal to logic(alternative)", { + code <- paste( + "num <- sample(1:100, 1)", + "big <- 0", + "small <- 0", + "if (num < 50) {", + " small <- small + 1", + "}", + "if (num >= 50) {", + " big <- big + 1", + "}", + "print(big)", + "print(small)", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "num <- sample(1:100, 1)", + "big <- 0", + "small <- 0", + "if (num < 50) {", + " small <- small + 1", + "}else {", + " big <- big + 1", + "}", + "print(big)", + "print(small)", + sep = "\n" + )) +}) + +test_that("Conversion to else due to Lesser than equal to logic", { + code <- paste( + "num <- sample(1:100, 1)", + "big <- 0", + "small <- 0", + "if (num <= 50) {", + " small <- small + 1", + "}", + "if (num > 50) {", + " big <- big + 1", + "}", + "print(big)", + "print(small)", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "num <- sample(1:100, 1)", + "big <- 0", + "small <- 0", + "if (num <= 50) {", + " small <- small + 1", + "}else {", + " big <- big + 1", + "}", + "print(big)", + "print(small)", + sep = "\n" + )) +}) + +test_that("Conversion to else due to Lesser than equal to logic (Alternative)", { + code <- paste( + "num <- sample(1:100, 1)", + "big <- 0", + "small <- 0", + "if (num > 50) {", + " big <- big + 1", + "}", + "if (num <= 50) {", + " small <- small + 1", + "}", + "print(big)", + "print(small)", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "num <- sample(1:100, 1)", + "big <- 0", + "small <- 0", + "if (num > 50) {", + " big <- big + 1", + "}else {", + " small <- small + 1", + "}", + "print(big)", + "print(small)", + sep = "\n" + )) +}) + +test_that("Conversion to else due to inequality", { + code <- paste( + "num <- sample(1:10, 1)", + "lottery <- FALSE", + "if (num == 5) {", + " lottery <- TRUE", + "}", + "if (num != 5) {", + " lottery <- FALSE", + "}", + "print(lottery)", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "num <- sample(1:10, 1)", + "lottery <- FALSE", + "if (num == 5) {", + " lottery <- TRUE", + "}else {", + " lottery <- FALSE", + "}", + "print(lottery)", + sep = "\n" + )) +}) + +test_that("No changes when none required, `;` version", { + code <- "var1 <- 3; var2 <- var1 * 5" + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, code) +}) + +test_that("No changes when none required, `;` version - 2", { + code <- paste("var1 <- 3;", + "var2 <- var1 * 5", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, code) +}) + +test_that("Case when the IF condtions have function calls", { + code <- paste( + "a <- 0", + "b <- 0", + "if (rnorm(1) > 0) {", + " a <- a + 1", + "}", + "if (rnorm(1) > 0) {", + " print(\"Yayy\")", + "}", + "if (!(rnorm(1) > 0)) {", + " b <- b + 1", + "}", + "print(a)", + "print(b)", + sep = "\n" + ) + opt_code <- opt_cond_thread(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "a <- 0", + "b <- 0", + "if (rnorm(1) > 0) {", + " a <- a + 1", + "}", + "if (rnorm(1) > 0) {", + " print(\"Yayy\")", + "}", + "if (!(rnorm(1) > 0)) {", + " b <- b + 1", + "}", + "print(a)", + "print(b)", + sep = "\n" + )) +}) + diff --git a/tests/testthat/test-opt_dead_expr.R b/tests/testthat/test-opt_dead_expr.R index 5d7437a..768e2c3 100644 --- a/tests/testthat/test-opt_dead_expr.R +++ b/tests/testthat/test-opt_dead_expr.R @@ -243,3 +243,198 @@ test_that("eliminate empty in one side of `ifelse`", { sep = "\n" )) }) + +test_that("Remove DE also if `;` and `\n` both are included", { + code <- paste( + "f2 <- function(x){", + "1;", + "2;", + "3;", + "4;", + "5;", + "}", + sep = "\n" + ) + opt_code <- opt_dead_expr(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "f2 <- function(x){", + ";", + ";", + ";", + ";", + "5;", + "}", + sep = "\n" + )) +}) + +test_that("dont eliminate assigns", { + code <- paste( + "bar <- function(x) {", + " x <- 3;", + " x <- x + 3;", + " if (x == 0) {", + " x <- 3;", + " x + 8818;", + " } else if (x == 1) {", + " x <- 3;", + " x + 8818;", + " } else {", + " x <- 3;", + " x + 8818;", + " }", + " x <- 3;", + " x + 8818;", + "}", + sep = "\n" + ) + opt_code <- opt_dead_expr(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "bar <- function(x) {", + " x <- 3;", + " x <- x + 3;", + " if (x == 0) {", + " x <- 3;", + ";", + " } else if (x == 1) {", + " x <- 3;", + ";", + " } else {", + " x <- 3;", + ";", + " }", + " x <- 3;", + " x + 8818;", + "}", + sep = "\n" + )) +}) + +test_that("dont eliminate DE in if/else with `;` ending", { + code <- paste( + "bar <- function(x) {", + " if (x == 0) {", + " x + 8818;", + " } else if (x == 1) {", + " x + 8818;", + " } else {", + " x + 8818;", + " }", + "}", + sep = "\n" + ) + opt_code <- opt_dead_expr(list(code))$codes[[1]] + expect_equal(opt_code, code) +}) + +test_that("dont eliminate DE in parent env ending with `;`", { + code <- paste( + "8 + 8 + 1918;", + sep = "\n" + ) + opt_code <- opt_dead_expr(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "8 + 8 + 1918;", + sep = "\n" + )) +}) + +test_that("eliminate DE in fun", { + code <- paste( + "8 + 8 + 1918;", + "foo <- function() 8818;", + "bar <- function() {", + " 8818;", + " 8818;", + "}", + sep = "\n" + ) + opt_code <- opt_dead_expr(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "8 + 8 + 1918;", + "foo <- function() 8818;", + "bar <- function() {", + ";", + " 8818;", + "}", + sep = "\n" + )) +}) + +test_that("eliminate DE in if/else", { + code <- paste( + "bar <- function(x) {", + " if (x == 0) {", + " x + 8818;", + " } else if (x == 1) {", + " x + 8818;", + " } else {", + " x + 8818;", + " }", + " x + 8818;", + "}", + sep = "\n" + ) + opt_code <- opt_dead_expr(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "bar <- function(x) {", + " if (x == 0) {", + ";", + " } else if (x == 1) {", + ";", + " } else {", + ";", + " }", + " x + 8818;", + "}", + sep = "\n" + )) +}) + +test_that("dont eliminate part of exprs,`;` case", { + code <- paste( + "bar <- function(x) {", + " tp <- ip[startsWith(ip, token)];", + " completions <- lapply(tp, function(package) NULL);", + " x + 8818;", + "}", + sep = "\n" + ) + opt_code <- opt_dead_expr(list(code))$codes[[1]] + expect_equal(opt_code, code) +}) + +test_that("dont eliminate empty `if` or loop, `;` version", { + code <- paste( + "foo <- function() {", + " if (cond) NULL;", + " if (cond) NULL else NULL;", + " while (cond) NULL;", + " for (i in cond) NULL;", + " 8818;", + "}", + sep = "\n" + ) + opt_code <- opt_dead_expr(list(code))$codes[[1]] + expect_equal(opt_code, code) +}) + +test_that("eliminate empty in one side of `ifelse`", { + code <- paste( + "foo <- function() {", + " if (cond) x else NULL;", + " if (cond) NULL else x;", + " 8818;", + "}", + sep = "\n" + ) + opt_code <- opt_dead_expr(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "foo <- function() {", + " if (cond) {} else NULL;", + " if (cond) NULL else {};", + " 8818;", + "}", + sep = "\n" + )) +}) diff --git a/tests/testthat/test-opt_memory_alloc.R b/tests/testthat/test-opt_memory_alloc.R new file mode 100644 index 0000000..bd54c48 --- /dev/null +++ b/tests/testthat/test-opt_memory_alloc.R @@ -0,0 +1,231 @@ +context("opt_memory_alloc") + +test_that("Code snippet with semicolon and nothing to optimize", { + code <- "var1 <- 3; var2 <- var1 * 5" + opt_code <- rco::opt_memory_alloc(list(code))$codes[[1]] + expect_equal(opt_code, "var1 <- 3; var2 <- var1 * 5") +}) + +test_that("Empty code snippet", { + code <- paste( + "", + sep = "\n" + ) + opt_code <- rco::opt_memory_alloc(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "", + sep = "\n" + )) +}) + +test_that("Code snippet with assignments within loops and presence of functions in the loop conditionals", { + code <- paste( + "v <- NULL", + "for(i in 1:10) {", + " v[i] <- i*i", + "}", + "fun <- function(n) {", + " num <- NULL", + " for(i in c(2010,2011)) {", + " num[i] <- i", + " }", + " no_fun_num <- NULL", + " for(i in 2010:2011) {", + " no_fun_num[i] <- i", + " }", + " sum(num)", + " sum(no_fun_sum)", + "}", + sep = "\n" + ) + opt_code <- rco::opt_memory_alloc(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "v <- vector(length = 10)", + "for(i in 1:10) {", + " v[i] <- i*i", + "}", + "fun <- function(n) {", + " num <- NULL", + " for(i in c(2010,2011)) {", + " num[i] <- i", + " }", + "no_fun_num <- vector(length = 2011)", + " for(i in 2010:2011) {", + " no_fun_num[i] <- i", + " }", + " sum(num)", + " sum(no_fun_sum)", + "}", + sep = "\n" + )) +}) + +test_that("Code snippet showcasing different types of assignments to vectors", { + code <- paste( + "v = numeric()", + "for(i in 1:5) {", + " v[i] <- i^2", + "}", + "logical() -> w", + "for(j in 1:10) {", + " w[j] <- FALSE", + "}", + "c() -> x", + "for(k in 1:6) {", + " x[k] <- k*2", + "}", + "y = c()", + "for(l in seq_len(5)) {", + " y[l] <- l*2", + "}", + sep = "\n" + ) + opt_code <- rco::opt_memory_alloc(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "v <- vector(length = 5)", + "for(i in 1:5) {", + " v[i] <- i^2", + "}", + "w <- vector(length = 10)", + "for(j in 1:10) {", + " w[j] <- FALSE", + "}", + "x <- vector(length = 6)", + "for(k in 1:6) {", + " x[k] <- k*2", + "}", + "y = c()", + "for(l in seq_len(5)) {", + " y[l] <- l*2", + "}", + sep = "\n" + )) +}) + +test_that("Code snippet ", { + code <- paste( + "x <- NULL", + "v <- NULL", + "z <- NULL", + "for(i in 1:5) {", + " v[i] = i*i", + " i -> x[i]", + " v[i+5] <- 2*i", + " z[i] <- v[i]", + "}", + sep = "\n" + ) + opt_code <- rco::opt_memory_alloc(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "x <- vector(length = 5)", + "v <- NULL", + "z <- vector(length = 5)", + "for(i in 1:5) {", + " v[i] = i*i", + " i -> x[i]", + " v[i+5] <- 2*i", + " z[i] <- v[i]", + "}", + sep = "\n" + )) +}) + +test_that("memory allocation empty code", { + code <- paste( + "break_it <- function() {", + " i <<- 1", + "}", + "x <- c()", + "for(i in 1:10) {", + " break_it()", + " x[i] <- i^2", + " print(i)", + "}", + sep = "\n" + ) + opt_code <- rco::opt_memory_alloc(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "break_it <- function() {", + " i <<- 1", + "}", + "x <- c()", + "for(i in 1:10) {", + " break_it()", + " x[i] <- i^2", + " print(i)", + "}", + sep = "\n" + )) +}) + +test_that("memory allocation empty code", { + code <- paste( + "v <- NULL", + "if(5 == 5) { #Comment1", + " if(3 == 3) {", + " equal <- NULL", + " for(i in 1:10) {", + " #Comment2", + " equal[i] <- i", + " }", + " }", + "}", + "really <- function() {", + " v <- c()", + " for(i in 1:10) {", + " v[i] <- i", + " }", + "}", + "#Comment3", + "really()", + "for(i in 1:10) {", + " v <- NULL", + " #Comment4", + " for(j in 1:5) {", + " v[j] <- i", + " }", + "}", + sep = "\n" + ) + opt_code <- rco::opt_memory_alloc(list(code))$codes[[1]] + expect_equal(opt_code, paste( + "v <- vector(length = 10)", + "if(5 == 5) { #Comment1", + " if(3 == 3) {", + "equal <- vector(length = 10)", + " for(i in 1:10) {", + " #Comment2", + " equal[i] <- i", + " }", + " }", + "}", + "really <- function() {", + "v <- vector(length = 10)", + " for(i in 1:10) {", + " v[i] <- i", + " }", + "}", + "#Comment3", + "really()", + "for(i in 1:10) {", + "v <- vector(length = 10)", + " #Comment4", + " for(j in 1:5) {", + " v[j] <- i", + " }", + "}", + sep = "\n" + )) +}) + + + + + + + + + + + + diff --git a/vignettes/.gitignore b/vignettes/.gitignore index 097b241..5cbb9dd 100644 --- a/vignettes/.gitignore +++ b/vignettes/.gitignore @@ -1,2 +1,3 @@ *.html *.R +*.Rhistory \ No newline at end of file diff --git a/vignettes/gsoc2020-final-report.Rmd b/vignettes/gsoc2020-final-report.Rmd new file mode 100644 index 0000000..b692079 --- /dev/null +++ b/vignettes/gsoc2020-final-report.Rmd @@ -0,0 +1,185 @@ +--- +author: "Rahul Saxena" +email: rahulSaxena.hinduBale@gmail.com +output: rmarkdown::html_vignette +title: GSoC 2020 Final Report +vignette: > + %\VignetteIndexEntry{GSoC 2020 Final Report} + %\VignetteEngine{knitr::rmarkdown} + %\VignetteEncoding{UTF-8} +--- + +```{r setup, include=FALSE} +library("ggplot2") +library("microbenchmark") +knitr::opts_chunk$set(echo = TRUE) +``` + +# My GSoC 2020 work in a nutshell + +Phew... So it's already time for the end of *GSoC 2020*. Time sure flew by this summer. Now it's the time for me to take a step back and ponder over my summer with *The R Project for Statistical Computing*. So, let's touch upon my summer in a chronological fashion. + +## Humble Beginnings: + +Around the time when *The R Project for Statistical Computing* announced their participation in *GSoC 2020*, a list of proposed projects were announced. The [`rco`](https://cran.r-project.org/web/packages/rco/index.html) or the **R Code Optimizer** caught my attention at the very first moment and I didn't look back since then. + +Although the exsisting [`rco`](https://github.com/jcrodriguez1989/rco) repository was extremely well maintained and documented, me being a newbie still faced quite a lot of difficulties and also, most probably gave my mentors, [Dr. JC Rodriguez](https://jcrodriguez1989.github.io/) and [Mauricio "Pachá" Vargas Sepúlveda](https://pacha.dev/) a hard time. + +As part of the [tests](https://github.com/rstats-gsoc/gsoc2020/wiki/rco%3A-The-R-Code-Optimizer#tests) to be completed for being considered as a potential student developer for R with the `rco` project, I was required to build a completely functional and working optimizer. + +I decided to work on a suggested optimizer that made [column extraction](https://rpubs.com/IACCancu/462502) faster and more efficient. After the benign guidance of my mentors, I was able to come up with an column extraction optimizer, that performed according to its expectation. This got me so excited that I worked on and built another [value extraction optimizer](https://rpubs.com/IACCancu/462501) without any explicit incentive :D + +* [**PR for Column Extraction Optimizer**](https://github.com/jcrodriguez1989/rco/pull/152) +* [**PR for Value Extraction Optimizer**](https://github.com/jcrodriguez1989/rco/pull/155) + +## Post GSoC selection: + +It was only a few days before my acceptance into the GSoC 2020 program, that my mentors gave their insights into both these optimizers and explained, how these optimizers can be broken, owing to the extreme flexibility that the R Language offers. Since, `.subset2( )` is not a reserved keyword in the R Language, it could be over-written. Hence we can never be sure while optimizing, that the functions that are being called has not been re-defined by the user. However, this served as a great primer as to what to expect when GSoC officially began. + +## Getting up to speed: + +The next tasks that I were assigned were not too glorious/dramatic or blinge-y but they were essential steps for the upkeep of a package on `CRAN`. Firstly, I designed a vignette that enlisted all the [future potential optimizers](https://rpubs.com/hinduBale/strategies_ideas_rco) that can be implemented in the `rco` package. Next, I went on to collate the efforts put in the `rco` library by the *Google Code-In* students, by creating PRs that such that every student gets credit as a contributor and at the same time hand-picking the best of examples and explanations to include in the official `rco` documentation. It was a riveting experience for me, as I had to learn and implement some really advanced concepts of `git` and `github` to be able to pull this off, including, but not limited to, `cherry-picking`, resolving `merge conficts`, adding `remotes`, etc. + +* [**PR for the Potential Optimizers Vignette**](https://github.com/jcrodriguez1989/rco/pull/156) +* [**PR #1 related to Google Code-In**](https://github.com/jcrodriguez1989/rco/pull/161) +* [**PR #2 related to Google Code-In**](https://github.com/jcrodriguez1989/rco/pull/163) +* [**PR #3 related to Google Code-In**](https://github.com/jcrodriguez1989/rco/pull/164) +* [**PR #4 related to Google Code-In**](https://github.com/jcrodriguez1989/rco/pull/165) +* [**PR #5 related to Google Code-In**](https://github.com/jcrodriguez1989/rco/pull/166) +* [**PR #6 related to Google Code-In**](https://github.com/jcrodriguez1989/rco/pull/167) +* [**PR #7 related to Google Code-In**](https://github.com/jcrodriguez1989/rco/pull/168) + +## Unexpected Turn of Events: + +As **Dwight D. Eisenhower** once said, + +*In preparing for battle I have always found that plans are useless, but planning is indispensable.* + +I had listed several optimization techniques that could have been implemented in `rco` in my [GSoC proposal](https://drive.google.com/file/d/1HvF0oE9gW4BYt2Lo-1zRl1dcbsG1wEPg/view) as well the [*potential optimizers* vignette](https://rpubs.com/hinduBale/strategies_ideas_rco) but when we got down to discussing the design of these optimizers, it became quite clear to me and my mentors that we were walking on thin ice. With the extreme constraints that are imposed due to R's flexible nature, only a few of our optimization strategies seemed bullet-proof. + +### Conditional Threading Optimizer: +I went on ahead with the idea of [Jump/Conditional Threading](https://beza1e1.tuxen.de/articles/jump_threading.html) and started designing the optimizer. The objective was to replace `if` statments with `else` wherever possible and also to group together the code blocks of different `if` blocks if the `if conditionals` were same. This concept is succintly covered over [here](https://developers.redhat.com/blog/2019/03/13/intro-jump-threading-optimizations/). + +To see the speed-ups provided by this optimization strategy, have a look at this example: + +#### Unoptimized Code + +```{r cond_thread_thread_code_og, echo=TRUE, warning=FALSE} +cond_thread <- function(n) { + evens <- 0 + evens_sum <- 0 + odds <- 0 + for (i in seq_len(n)) { + if (i %% 2 == 0) { # same logical as next if condition (can be merged) + evens <- evens + 1 + } + if (i %% 2 == 0) { + evens_sum <- evens_sum + i + } + if (!(i %% 2 == 0)) { # exact negation as previous if (can be an else) + odds <- odds + 1 + } + } +} +``` + +#### Proposed Optimized Code + +```{r cond_thread_thread_code_op, echo=TRUE, warning=FALSE} +cond_thread_opt <- function(n) { + evens <- 0 + evens_sum <- 0 + odds <- 0 + for (i in seq_len(n)) { + if (i %% 2 == 0) { # merged + evens <- evens + 1 + evens_sum <- evens_sum + i + } else { # converted to else + odds <- odds + 1 + } + } +} +``` + +#### Benchmark + +```{r cond_thread_thread_benchmark, warning=FALSE, message=FALSE} +n <- 100000 +autoplot(microbenchmark(cond_thread(n), cond_thread_opt(n))) +``` + + + +* [**PR for the Conditional Threading Optimizer**](https://github.com/jcrodriguez1989/rco/pull/162) + +### Memory Allocation Optimizer: +Next, I set my eyes on the [Memory Pre-Allocation optimization technique](https://insightr.wordpress.com/2018/08/23/growing-objects-and-loop-memory-pre-allocation/). The objective here was to save the *R Programmers* from the sin of growing a vector in loops. In our experience, we've seen that vector, lists, etc are initialized often with `NULL` or `c()` out of convenience. Bothering about the size of the vectors may seem trivial to the programmer, but it has a huge impact on the performance of the script. + +To see the speed-ups provided by this optimization strategy, consider this example: + +#### Unoptimized Code + +```{r pre_allocation_code_og} +mem_alloc <- function(n) { + vec <- NULL + for (i in seq_len(n)) { + vec[i] <- i + } +} +``` + +#### Proposed Optimized Code + +```{r pre_allocation_code_op} +mem_alloc_opt <- function(n) { + vec <- vector(length = n) + for (i in seq_len(n)) { + vec[i] <- i + } +} +``` + +#### Benchmark + +```{r pre_allocate_benchmark, echo = TRUE, warning = FALSE, message= FALSE} +n <- 100000 +autoplot(microbenchmark(mem_alloc(n), mem_alloc_opt(n))) +``` + + + +* [**PR for the Memory Allocation Optimizer**](https://github.com/jcrodriguez1989/rco/pull/169) + +## Pivot and Push: + +As mentioned in the last segment, we had pretty much exhausted all the possible optimization strategies, that could be implemented without breaking. Now, it was onto us to decide whether, to work on new optimizer, that had a non-zero chance of breaking or pivot and change objectives. Following the lead of my mentors, I decided to put on hold, the optimizers that could break and focus on further bolstering the `rco` package. + +### Fixing a critical issue: + +Owing to the fantastic upkeep of the `rco` repository, it had almost no bugs in the optimizers, and that is no mean feat. However [an issue reporting a bug](https://github.com/jcrodriguez1989/rco/issues/107) was reported. The bug was that an optimizer, namely `opt-dead-expr()` did not function when the user used `;`. Ideally, the optimizer shouldn't have been affected by the usage of `;`, but here it was, a bug, laying dormant for over 6 months. + +I decided to tackle this bug head on. I [created a separate branch](https://github.com/hinduBale/rco/tree/issue%23107) for this bug and reproduced the bug. After lots of trials and tribulations and lots of reverse-engineering, I zeroed in one the problem. It was a problem that existed in very low-level R code. When a R script without the use of `;` is parsed, then they get tokenized as something called `expr` but when `;` was used, the parser tokenized them as an `exprlist`. The exsisting optimizer did not handle the case of `exprlist`, so I appended code that handled `exprlist` and then the optimizer started functioning normally irrespective of the usage of `;`. This is an issue which could arise in other optimizers too, but given that, it has been solved in one optimizer, solving them in other optimizers would not prove challenging. Also, in the process of making the `opt_dead_expr()` optimizer work, I further bolstered the robustness of the optimizer by **doubling the number of test cases** that the optimizer must pass, including tests with `;`. + +* [**PR for solution of issue #107**](https://github.com/jcrodriguez1989/rco/pull/170) + +### Call for Actions: + +To reduce dependencies on 3rd party applications and to promote homogeneity, we decided to go for an in-house solution for the CI/CD needs of our package, ie `Github Actions`. Earlier the popular choice for carrying out CI/CD operations were 3rd party applications such as *Travis CI* or *codecov.io*, but lately more and more developers and organizations had been migrating to Github Actions and we decided to follow suit. *Github Marketplace* wasn't much help, as there were not much support for the **R Language** as compared to other languages such as *Rust*, *Ruby*, *Typescript*, etc. So, I went through several documentations and scarce examples and [created a branch](https://github.com/hinduBale/rco/tree/ghActions) that [renders a website](https://hindubale.github.io/rco/) and completes the testing that Travis used to do. We left out the *code coverage* as *Github* [does not yet support badges that shows the percentage](https://github.com/r-lib/actions/issues/156#event-3643325125) of coverage. + +### Acting on users' recommendations: + +One of the users of `rco` opened an issue, with a [feature request](https://github.com/jcrodriguez1989/rco/issues/72). The request that the user asked for, is as follows: + +> I think it would be nifty to be able to create a report out of `rco` optimisers. +> The idea would be to analyse the whole code and return some kind of markdown / bookdown that will list all the results from the optimisers, without changing the > original code. +> It might be useful to analyse your own code and learn what you could do better, but also it could be use in the industry to analyse the quality of code. + +While actually implementing this idea, would have been a tall order, me and my mentors discussed that, as the first iteration we could work on creating a function, which could be called to check, exactly how many files from a folder or a group of files can be optimized and which optimizer is being used there. If interested, the user could run that specific file along with the optimizers named in the output, in the `rco_gui()` function and see the code difference. + +* [**PR for the reporting function**](https://github.com/jcrodriguez1989/rco/pull/171) + +## And, my GSoC '20 journey comes to an end: + +Well, I never thought that 3 months could pass in a jiffy, but these months when I was working under the *R Language* in this program, had been the golden months of my life. Never to be forgotten and to be re-lived time and again. I am proud and happy to state that, my GSoC experience and project was mostly a hit. Yes, we did face obstacles, but each and every time we emerged stronger. I have now become a life-long fan of open-source. + +> Open source is simply put, magic. And, no one can ever get enough of magic, right >.< \ No newline at end of file diff --git a/vignettes/opt-common-subexpr.Rmd b/vignettes/opt-common-subexpr.Rmd index 8ac85d4..ca07a90 100644 --- a/vignettes/opt-common-subexpr.Rmd +++ b/vignettes/opt-common-subexpr.Rmd @@ -30,6 +30,10 @@ speed_up <- function(obj) { # Common Subexpression Elimination +## Idea + +The main idea behind the common subexpression elimination optimizer is to separate/isolate those parts of an expression (sub-expressions) that are being calculated again and again. By isolating a common sub-expression and replacing it with a variable, we can limit the number of computations. Thus improving the performance of our script. + ## Background Common Subexpression Elimination is an optimization that searches for instances of identical expressions, and replaces them with a single variable holding the computed value. diff --git a/vignettes/opt-cond-thread.Rmd b/vignettes/opt-cond-thread.Rmd new file mode 100644 index 0000000..94280c5 --- /dev/null +++ b/vignettes/opt-cond-thread.Rmd @@ -0,0 +1,88 @@ +--- +output: rmarkdown::html_vignette +title: Conditional Threading Optimizer +vignette: > + %\VignetteIndexEntry{Conditional Threading Optimizer} + %\VignetteEngine{knitr::rmarkdown} + %\VignetteEncoding{UTF-8} +--- + +```{r echo=FALSE, message=FALSE} +library("rco") +library("microbenchmark") +library("ggplot2") +autoplot.microbenchmark <- function(obj) { + levels(obj$expr) <- paste0("Expr_", seq_along(levels(obj$expr))) + microbenchmark:::autoplot.microbenchmark(obj) +} +speed_up <- function(obj) { + levels(obj$expr) <- paste0("Expr_", seq_along(levels(obj$expr))) + obj <- as.data.frame(obj) + summaries <- do.call(rbind, by(obj$time, obj$expr, summary)) + res <- c() + for (i in seq_len(nrow(summaries) - 1) + 1) { + res <- rbind(res, summaries[1, ] / summaries[i, ]) + } + rownames(res) <- levels(obj$expr)[-1] + return(res) +} +``` + +# Conditional Threading Optimizer + +## Background + +Conditional Threading is an optimization that searches for separate blocks of `IF` statements but with exactly the same conditions. When several conditionals present the same boolean condition, merging their inside actions would avoid computing the condition several times. Also, in the cases where a second `IF` statement has been used with the exact negation of the condition of the first `IF`, the second `IF` could then be incorporated in an `ELSE` statement. + +## Example + +Consider the following example: + +```{r} +code <- paste( + "n_evens <- 0", + "n_odds <- 0", + "odds_sum <- 0", + "evens_sum <- 0", + "", + "for (i in sample(10000, replace = TRUE)) {", + " if (i %% 2 == 1) {", + " n_odds <- n_odds + 1", + " }", + " if (i %% 2 == 1) {", + " odds_sum <- odds_sum + i", + " }", + " if (!(i %% 2 == 1)) {", + " n_evens <- n_evens + 1", + " }", + " if (!(i %% 2 == 1)) {", + " evens_sum <- evens_sum + i", + " }", + "}", + sep = "\n" +) +cat(code) +``` + +Then, the automatically optimized code would be: + +```{r} +opt_code <- opt_cond_thread(list(code)) +cat(opt_code$codes[[1]]) +``` + +And if we measure the execution time of each one, and the speed-up: + +```{r message=FALSE} +bmark_res <- microbenchmark({ + eval(parse(text = code)) +}, { + eval(parse(text = opt_code)) +}) +autoplot(bmark_res) +speed_up(bmark_res) +``` + +## Implementation + +The `opt-cond-thread` optimizer simply looks for `IF` statements in the code snippet given as the input and if it finds an `IF` statement, it checks the immediate next block of code. If it is another `IF` statement, the optimizer merges the two `IF` blocks if the `IF` conditions were same or converts the second `IF` statement to `ELSE` if the conditions were negation of each other. diff --git a/vignettes/opt-constant-folding.Rmd b/vignettes/opt-constant-folding.Rmd index 9afa172..8c2d4b8 100644 --- a/vignettes/opt-constant-folding.Rmd +++ b/vignettes/opt-constant-folding.Rmd @@ -30,6 +30,10 @@ speed_up <- function(obj) { # Constant Folding +## Idea + +The basic idea behind this optimizer is to reduce as many calculations as possible. This implies that any expression that can be calculated before running the script, must be calculated and *reduced* or *folded* to a single constant value. + ## Background Constant folding is an optimization technique that eliminates expressions that calculate a value that can already be determined before code execution. These are typically calculations that only reference constant values or expressions that reference variables whose values are constant. diff --git a/vignettes/opt-constant-propagation.Rmd b/vignettes/opt-constant-propagation.Rmd index f350614..fc144b2 100644 --- a/vignettes/opt-constant-propagation.Rmd +++ b/vignettes/opt-constant-propagation.Rmd @@ -30,6 +30,10 @@ speed_up <- function(obj) { # Constant Propagation +## Idea + +The basic idea behind the constant propagation optimizer is to replace as many variables with their corresponding constant values, as possible. When we replace a variable with its corresponding constant value or *propagate the constant*, the computation time of the script decreases and we get a speed-up. + ## Background Constant propagation is the process of substituting the values of known constants in expressions. Constant propagation eliminates cases in which values are copied from one location or variable to another, in order to simply assign their value to another variable. diff --git a/vignettes/opt-dead-code.Rmd b/vignettes/opt-dead-code.Rmd index 96c2df6..8879fa2 100644 --- a/vignettes/opt-dead-code.Rmd +++ b/vignettes/opt-dead-code.Rmd @@ -30,6 +30,10 @@ speed_up <- function(obj) { # Dead Code Elimination +## Idea + +Dead code are pieces of code that do not affect the output of a program. The removal of these unnecessary code lines can be defined as Dead Code Elimination. These code lines do not serve any purpose in the main program flow because these codes will never be executed mainly because the condition for their execution is logically impossible/infeasible. This removal results in performance enhancements. + ## Background Dead Code Elimination is an optimization that removes code which does not affect the program results. You might wonder why someone would write this type of source code, but it can easily creep into large, long-lived programs even at the source code level. Removing such code has several benefits: it shrinks program size and it allows the running program to avoid executing irrelevant operations, which reduces its running time. It can also enable further optimizations by simplifying program structure. diff --git a/vignettes/opt-dead-expr.Rmd b/vignettes/opt-dead-expr.Rmd index e654de2..62c39d3 100644 --- a/vignettes/opt-dead-expr.Rmd +++ b/vignettes/opt-dead-expr.Rmd @@ -30,6 +30,10 @@ speed_up <- function(obj) { # Dead Expression Elimination +## Idea + +The basic idea behind the dead expression elimination optimizer is to remove all those expressions whose result is neither used in the program, nor the removal of which affects the result of the code. Therefore this *dead expression elimination* frees up memory, saves computation time, and provides a speed boost. + ## Background Dead Expression Elimination intends to remove those expressions that do not modify the environment. diff --git a/vignettes/opt-dead-store.Rmd b/vignettes/opt-dead-store.Rmd index 47ceac5..f5dc649 100644 --- a/vignettes/opt-dead-store.Rmd +++ b/vignettes/opt-dead-store.Rmd @@ -30,6 +30,10 @@ speed_up <- function(obj) { # Dead Store Elimination +## Idea + +The removal of a value allocated to a variable when it is not being read by any following instruction is called Dead Store Elimination. This Optimization is necessary as it saves processor time and memory which ultimately results in the faster execution of the program. + ## Background Dead Store Elimination is an optimization that intends to remove an assignation of a variable that is not read by any subsequent instruction. diff --git a/vignettes/opt-loop-invariant.Rmd b/vignettes/opt-loop-invariant.Rmd index ed1af36..d4f48a6 100644 --- a/vignettes/opt-loop-invariant.Rmd +++ b/vignettes/opt-loop-invariant.Rmd @@ -30,6 +30,10 @@ speed_up <- function(obj) { # Loop-invariant Code Motion +## Idea + +Invariant code is a piece of code that could be run outside of a loop without affecting the actual outcome of a program. This may come in the form of variables that are not modified after going through the loop and as such are called "invariants" or non-changing. If these pieces of code were removed from the loops, this would speed up the script being run. + ## Background Loop-invariant code consists of statements or expressions which can be moved outside the body of a loop without affecting the semantics of the program. Loop-invariant code motion is a compiler optimization which performs this movement automatically. diff --git a/vignettes/opt-memory-alloc.Rmd b/vignettes/opt-memory-alloc.Rmd new file mode 100644 index 0000000..1d02aa8 --- /dev/null +++ b/vignettes/opt-memory-alloc.Rmd @@ -0,0 +1,98 @@ +--- +output: rmarkdown::html_vignette +title: Memory Allocation Optimizer +vignette: > + %\VignetteIndexEntry{Memory Allocation Optimizer} + %\VignetteEngine{knitr::rmarkdown} + %\VignetteEncoding{UTF-8} +--- + +```{r echo=FALSE, message=FALSE} +library("rco") +library("microbenchmark") +library("ggplot2") +autoplot.microbenchmark <- function(obj) { + levels(obj$expr) <- paste0("Expr_", seq_along(levels(obj$expr))) + microbenchmark:::autoplot.microbenchmark(obj) +} +speed_up <- function(obj) { + levels(obj$expr) <- paste0("Expr_", seq_along(levels(obj$expr))) + obj <- as.data.frame(obj) + summaries <- do.call(rbind, by(obj$time, obj$expr, summary)) + res <- c() + for (i in seq_len(nrow(summaries) - 1) + 1) { + res <- rbind(res, summaries[1, ] / summaries[i, ]) + } + rownames(res) <- levels(obj$expr)[-1] + return(res) +} +``` + +# Memory Allocation Optimizer + +## Background + +As a general rule of thumb, in any programming language we should undertake the memory management as much as possible. When we grow a vector inside a loop, the vector asks the processor for extra space in between the running program and then proceeds once it gets the required memory. This process is repeated for every iteration of the loop resulting in massive delays. Thus we should pre-allocate the required memory to a vector to avoid such delays. + +This *Memory Allocation Optimizer* checks the code for vectors that have been initialized without proper memory allocation and whenever possible, changes that allocation to an allocation with a fixed size, taken from the the loops where the vector is being called. + +## Example + +Consider the following example: + +```{r} +code <- paste( + "square_vec <- NULL", + "mul_vec <- NULL", + "for(i in 1:100) {", + " square_vec[i] <- i^2", + "}", + "for(i in 1:100) {", + " mul_vec[i] <- i * i", + "}", + "identical(square_vec, mul_vec)", + sep = "\n" +) +cat(code) +``` + +Then, the automatically optimized code would be: + +```{r} +opt_code <- opt_memory_alloc(list(code)) +cat(opt_code$codes[[1]]) +``` + +And if we measure the execution time of each one, and the speed-up: + +```{r message=FALSE} +bmark_res <- microbenchmark({ + eval(parse(text = code)) +}, { + eval(parse(text = opt_code)) +}) +autoplot(bmark_res) +speed_up(bmark_res) +``` + +## Implementation + +The `memory-allocation-optimizer` looks for vector assignments in the entire code snippet, that have been assigned without proper allocation of memory. For instance, consider a vector `vec`, now this vector can be initialized without proper memory allocation in the following ways: + +* `vec <- NULL` +* `vec = c()` +* `vec <- NA` +* `logical() -> vec()` +* and so on.... + +When we spot these *improper* initializations of vectors, we change the intialization to include a suitable memory allocation in the following way, if possible: + +* `vec <- vector(length = 10)` + +## Limitations of the Optimizer + +While the highly flexible and accomodating nature of the **R Language** is a boon for programmers, that very nature has limited the scope of this optimizer to a certain extent. In this optimizer the following points have been kept in mind: + +* Only `FOR` loops have been considered, owing to the fact that their sizes, most of the times, can be predicted accurately which is not the case with `while` and `repeat` loops. +* Since most of the in-built functions of base-R can be overwritten, we had to ignore the cases where any function was used in either the body or conidition of the `FOR` loop. +* Also to be sure of the size of the vector, we have only proceeded with the cases where the index was explicitly mentioned with the vectors inside the loop. diff --git a/vignettes/potential-optimizers.Rmd b/vignettes/potential-optimizers.Rmd new file mode 100644 index 0000000..036085f --- /dev/null +++ b/vignettes/potential-optimizers.Rmd @@ -0,0 +1,235 @@ +--- +output: rmarkdown::html_vignette +title: Potential Optimizers +vignette: > + %\VignetteIndexEntry{Potential Optimizers} + %\VignetteEngine{knitr::rmarkdown} + %\VignetteEncoding{UTF-8} +--- + +# Potential optimizers + +```{r setup, include=FALSE} +library("ggplot2") +library("microbenchmark") + +knitr::opts_chunk$set(echo = TRUE) +``` + +## Inline Expansion + +### Idea + +Replacing a function call with the body of the called function is called "inline expansion". This eliminates the function calling overhead and also the overhead of return call from a function. It also saves the overhead of variables push/pop on the stack while function calling. + +### Code Examples + +#### Unoptimized Code + +```{r inline_expansion_code_og, echo = TRUE, warning = TRUE} +cubed <- function(x) { + x * x * x +} + +inline <- function(n) { + to_cubes <- 0 + for (i in seq_len(n)) { + to_cubes <- to_cubes + cubed(i) + } +} +``` + +#### Proposed Optimized Code + +```{r inline_expansion_code_op, echo = TRUE, warning = TRUE} +inline_opt <- function(n) { + to_cubes <- 0 + for (i in seq_len(n)) { + to_cubes <- to_cubes + (i * i * i) # function inlined + } +} +``` + +### Benchmark + +```{r inline_expansion_benchmark, echo = TRUE, warning = FALSE, message= FALSE} +n <- 1000 +autoplot(microbenchmark(inline(n), inline_opt(n))) +``` + +## Memory Pre-Allocation + +### Idea + +As a general rule of thumb, in any programming language, we should undertake memory management as much as possible. When we grow a vector inside a loop, the vector asks the processor for extra space in between the running program and then proceeds, once it gets the required memory. This process is repeated for every iteration of the loop. Thus we should pre-allocate the required memory to a vector to avoid such delays. + +### Code Examples + +#### Unoptimized Code + +```{r pre_allocation_code_og} +mem_alloc <- function(n) { + vec <- NULL + for (i in seq_len(n)) { + vec[i] <- i + } +} +``` + +#### Proposed Optimized Code + +```{r pre_allocation_code_op} +mem_alloc_opt <- function(n) { + vec <- vector(length = n) + for (i in seq_len(n)) { + vec[i] <- i + } +} +``` + +### Benchmark + +```{r pre_allocate_benchmark, echo = TRUE, warning = FALSE, message= FALSE} +n <- 100000 +autoplot(microbenchmark(mem_alloc(n), mem_alloc_opt(n))) +``` + +## Vectorization + +### Idea + +A golden rule in R programming is to access the underlying C/Fortran routines as much as possible; the fewer R function calls required to achieve this, the better. Many R functions are therefore vectorized, that is, the function's inputs and/or outputs naturally work with vectors, reducing the number of function calls required. + +### Code Examples + +#### Unoptimized Code + +```{r vectorization_code_og} +non_vectorized <- function(n) { + v1 <- seq_len(n) + v2 <- length(seq.int(n + 2, n + 1000, 2)) + res <- vector(length = length(v1)) + for (i in seq_len(n)) { + res[i] <- v1[i] + v2[i] + } +} +``` + +#### Proposed Optimized Code + +```{r vectorization_code_op} +vectorized <- function(n) { + v1 <- seq_len(n) + v2 <- length(seq.int(n + 2, n + 1000, 2)) + res <- v1 + v2 +} +``` + +### Benchmark + +```{r vectorized_benchmark, echo = TRUE, warning = FALSE, message= FALSE} +n <- 10000 +autoplot(microbenchmark(non_vectorized(n), vectorized(n))) +``` + +## Efficient Column Extraction + +### Idea + +The idea would be to replace the different one-column extraction alternatives by the much faster `.subset2` call alternative. + +### Benchmark + +```{r col_ext_benchmark, warning = FALSE, message = FALSE} +autoplot(microbenchmark( + mtcars[, 11], + mtcars$carb, + mtcars[[c(11)]], + mtcars[[11]], + .subset2(mtcars, 11) +)) +``` + +### Drawbacks + +1. For some R classes, the `[[ ]]` operator and `.subset` work differently. For instance, they seem to be equivalent for `data.frame` but are not the same for `matrix` class. + +2. Moreover, both `[[ ]]` and `.subset2` are functions and in R, any function can be overwritten. Thus the above optimization can be made to fail just by redefining, say, the the `.subset2` function. + +## Efficient Value Extraction + +### Idea + +The idea would be to replace the different one-value extraction alternatives by the much faster `.subset2` call alternative. + +### Benchmark + +```{r val_ext_benchmark, warning = FALSE, message = FALSE} +autoplot(microbenchmark( + mtcars[32, 11], + mtcars$carb[32], + mtcars[[c(11, 32)]], + mtcars[[11]][32], + .subset2(mtcars, 11)[32], + times = 1000L +)) +``` + +### Drawback + +1. For some R classes, the `[[ ]]` operator and `.subset` work differently. For instance, they seem to be equivalent for `data.frame` but are not the same for `matrix` class. + +2. Moreover, both `[[ ]]` and `.subset2` are functions and in R, any function can be overwritten. Thus the above optimization can be made to fail just by redefining, say, the `.subset2` function. + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + +