pre-commit: PR183145

[zyw-bot]

Link: llvm/llvm-project#183145
Requested by: @SavchenkoValeriy

[zyw-bot]

Diff mode

runner: ariselab-64c-docker
baseline: llvm/llvm-project@796e7eb
patch: llvm/llvm-project#183145
sha256: aef2ac5bb5a7e49c695b99b573fb133837b907e591d7534b2e9ad9ab966d1f65
commit: 374d5eb

548 files changed, 315535 insertions(+), 315845 deletions(-)

Improvements:
  licm.NumAddSubHoisted 1907 -> 3515 +84.32%
  licm.NumHoisted 5264744 -> 5266600 +0.04%
  sccp.NumInstReplaced 131997 -> 132027 +0.02%
  correlated-value-propagation.NumAddNSW 215589 -> 215620 +0.01%
  correlated-value-propagation.NumNNeg 95915 -> 95926 +0.01%
  gvn.NumGVNEqProp 426001 -> 426047 +0.01%
  instcombine.NegatorNumTreesNegated 47836 -> 47841 +0.01%
  correlated-value-propagation.NumAnd 44240 -> 44244 +0.01%
  correlated-value-propagation.NumAddNW 426447 -> 426483 +0.01%
  instcombine.NegatorNumInstructionsNegatedSuccess 59297 -> 59302 +0.01%
Regressions:
  instsimplify.NumReassoc 813650 -> 799400 -1.75%
  instcombine.NegatorTotalNegationsAttempted 20572557 -> 20569277 -0.02%
  instcombine.NegatorNumValuesVisited 21488198 -> 21484937 -0.02%
  instcount.NumSubInst 1915424 -> 1915177 -0.01%
  instcount.NumPtrToIntInst 2995807 -> 2995486 -0.01%
  memdep.NumCacheNonLocal 19738 -> 19736 -0.01%
  gvn.NumGVNPRE 149385 -> 149375 -0.01%
  aggressive-instcombine.NumExprsReduced 15622 -> 15621 -0.01%
  aggressive-instcombine.NumInstrsReduced 45390 -> 45388 -0.00%
  instcount.NumAddInst 7317367 -> 7317141 -0.00%

+9 abc/wlcBlast.ll
+6 delta-rs/2ounwovke18sd889.ll
+6 rocksdb/block_cache_tier.ll
+6 rocksdb/histogram_windowing.ll
+6 zed-rs/13pr1pzdsvcd15h9bn5oaps4z.ll
+4 abc/cbaBlast.ll
+4 opencv/dwt.ll
+3 assimp/BlenderCustomData.ll
+3 assimp/MMDPmxParser.ll
+3 ceres/schur_eliminator_2_d_d.ll
+3 coreutils-rs/3x959pa9na58uqov.ll
+3 ffmpeg/dsp.ll
+3 folly/json.ll
+3 gromacs/selvalue.ll
+3 grpc/legacy_channel_idle_filter.ll
+3 icu/caniter.ll
+3 icu/coll.ll
+3 icu/measfmt.ll
+3 just-rs/bulvmt5gknxyhsd.ll
+3 llvm/CoverageMapping.ll
+3 luau/ltable.ll
+3 mitsuba3/jitallocator.ll
+3 rustfmt-rs/4gk399kploc9gcsb.ll
+3 slurm/bitstring.ll
+3 typst-rs/avdrw26ojy6f9qt.ll
+3 wireshark/packet-sip.ll
+3 z3/euf_bv_plugin.ll
+2 syn/2tga7oe2tfdpj05w.ll
+2 typst-rs/xrijsyrxk2ciyk9.ll
+1 fish-rs/8zxt2gjt68ikxu18thf3foa45.ll
+1 llvm/SemaTemplateDeduction.ll
+1 openmpi/pmix_mca_base_var.ll
+1 quiche-rs/7uu0743ctz557px8xpqxzn9zi.ll
+1 ruff-rs/bat91tmhz9zqtv9nbte3nknzg.ll
+1 zed-rs/dx9yxhk8z0nuczbnxw7yidd6v.ll
+0 darktable/collect.ll
+0 delta-rs/11f8x98axanecwnw.ll
+0 draco/mesh_edgebreaker_decoder_impl.ll
+0 draco/mesh_edgebreaker_encoder_impl.ll
+0 foundations-rs/208outlkq19x3hykb3fof84u4.ll
+0 g2o/solver_cholmod.ll
+0 git/commit.ll
+0 hermes/String.ll
+0 libquic/quic_stream_sequencer_buffer.ll
+0 lief/ssl_tls13_server.ll
+0 llvm/IVDescriptors.ll
+0 luau/BytecodeBuilder.ll
+0 minetest/voxelalgorithms.ll
+0 openvdb/Filter.ll
+0 openvdb/TopologyToLevelSet.ll
+0 quantlib/lfmprocess.ll
+0 regex-rs/6c2onrqlphpgxx0.ll
+0 rocksdb/statistics.ll
+0 rust-analyzer-rs/5a391sezfv8ojps9.ll
+0 tev/Common.ll
+0 uv-rs/afb3uo40tnjfqrj7louv4l8in.ll
+0 wasmtime-rs/37pex3k1sj15o95m.ll
+0 wireshark/range.ll
+0 zed-rs/9jkx32jki2vcljeo0xctxc8uw.ll
-1 abc/bmcMaj3.ll
-1 abc/sclLibScl.ll
-1 csmith/ArrayVariable.ll
-1 ffmpeg/magicyuvenc.ll
-1 hwloc/hwloc-calc.ll
-1 linux/pagelist.ll
-1 llvm/CGCleanup.ll
-1 lodepng/pngdetail.ll
-1 minetest/mapgen.ll
-1 openjdk/c1_LinearScan.ll
-1 openspiel/catch.ll
-1 pola-rs/eno41myg76h8nryc673zxzdw7.ll
-1 ruff-rs/3962cj1utxtdqcwyobmnzow9t.ll
-2 lvgl/lv_label.ll
-2 minetest/string.ll
-2 openjdk/instanceStackChunkKlass.ll
-2 quinn-rs/em94uyxz0wp9gccfab963606p.ll
-3 openspiel/breakthrough.ll
-4 boost/debug.ll
-4 openusd/pathPattern.ll
-4 pola-rs/bg2ex4k819k713rsl5rc2q2v3.ll
-5 openjdk/zRelocate.ll
-5 openspiel/cliff_walking.ll
-6 grpc/compression_internal.ll
-8 graphviz/emit.ll

[github-actions]

This patch introduces several loop optimization improvements across multiple benchmarks, primarily focused on loop invariant code motion and loop condition simplification. The key changes are:

1. Loop Invariant Code Motion: Introduces invariant.op variables to precompute loop termination conditions (e.g., %invariant.op = add i32 %4, -1) instead of recalculating them in each iteration. This replaces expressions like %624 = add nuw nsw i64 %.idx.i, 168; %625 = icmp eq i64 %624, %614 with %626 = icmp eq i64 %.idx.i, %invariant.op.i.

2. Loop Condition Simplification: Converts loop exit conditions from icmp eq i64 %indvars.iv.next, %trip.count to icmp eq i64 %indvars.iv, %invariant.op, eliminating the need for increment operations before the comparison.

3. Phi Node Updates: Adjusts phi node initial values and predecessor blocks to accommodate new loop structures, such as changing %indvars.iv71 = phi i64 [ 0, %.preheader.us.preheader ], [ %indvars.iv.next72, %._crit_edge.us ] to use %indvars.iv.next73.

4. Pointer Arithmetic Optimization: Replaces pointer arithmetic sequences (e.g., ptrtoint + sub + icmp) with direct pointer comparisons using inttoptr and add, improving readability and potentially enabling further optimizations.

5. Boolean Logic Refinement: Converts conditional selects (e.g., select i1 %.not54, i1 %69, i1 false) to more efficient boolean operations like and i1 %.not54, %69.

These changes collectively improve loop efficiency by reducing redundant computations and simplifying control flow, while maintaining functional equivalence.

model: qwen-plus-latest
CompletionUsage(completion_tokens=403, prompt_tokens=108078, total_tokens=108481, completion_tokens_details=None, prompt_tokens_details=None)

[nikic]

Regression