Skip to content

Refactor the dataflow mode logic #121

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ShangkunLi merged 3 commits into from
Aug 25, 2025
Merged

Refactor the dataflow mode logic #121

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c05bd41
refactor the dataflow mode logic
ShangkunLi Aug 24, 2025
4357d61
rename some functions
ShangkunLi Aug 25, 2025
be3f8a8
refactor the output format
ShangkunLi Aug 25, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
137 changes: 58 additions & 79 deletions test/neura/interpreter/loop_convert_controlflow_to_dataflow.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,95 +7,74 @@
// RUN: --fold-constant \
// RUN: %s -o %t_dataflow.mlir

// RUN: neura-interpreter %t_dataflow.mlir --verbose --dataflow > %t_output.txt 2>&1
// RUN: neura-interpreter %t_dataflow.mlir --verbose --dataflow > %t_output.txt

// RUN: FileCheck %s --check-prefix=DATAFLOW_IR --input-file=%t_dataflow.mlir

// RUN: FileCheck %s --check-prefix=INTERPRETER_OUTPUT --input-file=%t_output.txt

func.func @loop_sum() -> f32 {
%init_i = arith.constant 0.0 : f32
%init_sum = arith.constant 0.0 : f32
%step = arith.constant 1.0 : f32
%add_val = arith.constant 3.0 : f32
%limit = arith.constant 5.0 : f32
%c0 = "neura.constant"() <{predicate = true, value = 0.000000e+00 : f32}> : () -> f32 // init_i / init_sum
%c1 = "neura.constant"() <{predicate = true, value = 1.000000e+00 : f32}> : () -> f32 // step
%c2 = "neura.constant"() <{predicate = true, value = 3.000000e+00 : f32}> : () -> f32 // add_val
%c3 = "neura.constant"() <{predicate = true, value = 5.000000e+00 : f32}> : () -> f32 // limit

%v_i = "neura.reserve"() : () -> (f32)
%v_sum = "neura.reserve"() : () -> (f32)
// Entry: jump to loop head with initial i and sum
neura.br %c0, %c0 : f32, f32 to ^loop_head

"neura.ctrl_mov"(%init_i, %v_i) : (f32, f32) -> ()
"neura.ctrl_mov"(%init_sum, %v_sum) : (f32, f32) -> ()
^loop_head(%i: f32, %sum: f32): // predecessors: entry, ^loop_body
%cond = "neura.fcmp"(%i, %c3) <{cmpType = "lt"}> : (f32, f32) -> i1
// If cond true, go to body carrying i and sum; else go to exit with sum
neura.cond_br %cond : i1 then %i, %sum : f32, f32 to ^loop_body else %sum : f32 to ^loop_exit

"neura.br"() [^loop_head] {operandSegmentSizes = array<i32: 0>} : () -> ()
^loop_body(%i_in: f32, %sum_in: f32): // pred: ^loop_head
%new_sum = "neura.fadd"(%sum_in, %c2) : (f32, f32) -> f32
%new_i = "neura.fadd"(%i_in, %c1) : (f32, f32) -> f32
// loop back with updated i and sum
neura.br %new_i, %new_sum : f32, f32 to ^loop_head

^loop_head:
%i = "neura.phi"(%v_i, %init_i) : (f32, f32) -> f32
%sum = "neura.phi"(%v_sum, %init_sum) : (f32, f32) -> f32

%cond = "neura.fcmp"(%i, %limit) {cmpType = "lt"} : (f32, f32) -> i1

"neura.cond_br"(%cond) [^loop_body, ^loop_exit] {operandSegmentSizes = array<i32: 1, 0, 0, 0>} : (i1) -> ()

^loop_body:
%new_sum = "neura.fadd"(%sum, %add_val) : (f32, f32) -> f32

%new_i = "neura.fadd"(%i, %step) : (f32, f32) -> f32

"neura.ctrl_mov"(%new_i, %v_i) : (f32, f32) -> ()
"neura.ctrl_mov"(%new_sum, %v_sum) : (f32, f32) -> ()

"neura.br"() [^loop_head] {operandSegmentSizes = array<i32: 0>} : () -> ()

^loop_exit:
return %sum : f32
^loop_exit(%ret_sum: f32): // pred: ^loop_head
"neura.return"(%ret_sum) : (f32) -> ()
}

// DATAFLOW_IR: module {
// DATAFLOW_IR: func.func @loop_sum() -> f32 attributes {accelerator = "neura"} {
// DATAFLOW_IR: %cst = arith.constant 0.000000e+00 : f32
// DATAFLOW_IR: %cst_0 = arith.constant 1.000000e+00 : f32
// DATAFLOW_IR: %cst_1 = arith.constant 3.000000e+00 : f32
// DATAFLOW_IR: %cst_2 = arith.constant 5.000000e+00 : f32
// DATAFLOW_IR: %0 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR: %1 = "neura.grant_once"(%0) : (!neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR: %2 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR: %3 = "neura.grant_once"(%2) : (!neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR: neura.ctrl_mov %cst -> %0 : f32 !neura.data<f32, i1>
// DATAFLOW_IR: neura.ctrl_mov %cst -> %2 : f32 !neura.data<f32, i1>
// DATAFLOW_IR: %4 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR: %5 = "neura.phi"(%4, %cst_2) : (!neura.data<f32, i1>, f32) -> !neura.data<f32, i1>
// DATAFLOW_IR: %6 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR: %7 = "neura.phi"(%6, %3) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR: %8 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR: %9 = "neura.phi"(%8, %cst) : (!neura.data<f32, i1>, f32) -> !neura.data<f32, i1>
// DATAFLOW_IR: %10 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR: %11 = "neura.phi"(%10, %1) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR: %12 = "neura.phi"(%11, %9) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR: %13 = "neura.phi"(%7, %9) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR: %14 = "neura.fcmp"(%12, %5) <{cmpType = "lt"}> : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<i1, i1>
// DATAFLOW_IR: %15 = neura.grant_predicate %13, %14 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR: %16 = neura.grant_predicate %cst_1, %14 : f32, !neura.data<i1, i1> -> f32
// DATAFLOW_IR: %17 = neura.grant_predicate %12, %14 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR: %18 = neura.grant_predicate %cst_0, %14 : f32, !neura.data<i1, i1> -> f32
// DATAFLOW_IR: %19 = neura.grant_predicate %1, %14 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR: %20 = neura.grant_predicate %3, %14 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR: %21 = neura.grant_predicate %cst, %14 : f32, !neura.data<i1, i1> -> f32
// DATAFLOW_IR: %22 = neura.grant_predicate %cst_2, %14 : f32, !neura.data<i1, i1> -> f32
// DATAFLOW_IR: %23 = "neura.not"(%14) : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
// DATAFLOW_IR: %24 = neura.grant_predicate %13, %23 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR: %25 = "neura.fadd"(%15, %16) : (!neura.data<f32, i1>, f32) -> !neura.data<f32, i1>
// DATAFLOW_IR: %26 = "neura.fadd"(%17, %18) : (!neura.data<f32, i1>, f32) -> !neura.data<f32, i1>
// DATAFLOW_IR: neura.ctrl_mov %26 -> %19 : !neura.data<f32, i1> !neura.data<f32, i1>
// DATAFLOW_IR: neura.ctrl_mov %25 -> %20 : !neura.data<f32, i1> !neura.data<f32, i1>
// DATAFLOW_IR: neura.ctrl_mov %19 -> %10 : !neura.data<f32, i1> !neura.data<f32, i1>
// DATAFLOW_IR: neura.ctrl_mov %21 -> %8 : f32 !neura.data<f32, i1>
// DATAFLOW_IR: neura.ctrl_mov %20 -> %6 : !neura.data<f32, i1> !neura.data<f32, i1>
// DATAFLOW_IR: neura.ctrl_mov %22 -> %4 : f32 !neura.data<f32, i1>
// DATAFLOW_IR: "neura.return"(%24) : (!neura.data<f32, i1>) -> ()
// DATAFLOW_IR: }
// DATAFLOW_IR: }
// DATAFLOW_IR: func.func @loop_sum() -> f32 attributes {accelerator = "neura"} {
// DATAFLOW_IR-NEXT: %0 = "neura.grant_once"() <{constant_value = 0.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %1 = "neura.grant_once"() <{constant_value = 1.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %2 = "neura.grant_once"() <{constant_value = 3.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %3 = "neura.grant_once"() <{constant_value = 5.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %4 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %5 = "neura.phi"(%4, %1) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %6 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %7 = "neura.phi"(%6, %2) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %8 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %9 = "neura.phi"(%8, %3) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %10 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %11 = "neura.phi"(%10, %0) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %12 = neura.reserve : !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %13 = "neura.phi"(%12, %0) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %14 = "neura.fcmp"(%13, %9) <{cmpType = "lt"}> : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<i1, i1>
// DATAFLOW_IR-NEXT: %15 = neura.grant_predicate %13, %14 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %16 = neura.grant_predicate %11, %14 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %17 = neura.grant_predicate %7, %14 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %18 = neura.grant_predicate %5, %14 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %19 = neura.grant_predicate %9, %14 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %20 = "neura.not"(%14) : (!neura.data<i1, i1>) -> !neura.data<i1, i1>
// DATAFLOW_IR-NEXT: %21 = neura.grant_predicate %11, %20 : !neura.data<f32, i1>, !neura.data<i1, i1> -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %22 = "neura.fadd"(%16, %17) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: %23 = "neura.fadd"(%15, %18) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: neura.ctrl_mov %23 -> %12 : !neura.data<f32, i1> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: neura.ctrl_mov %22 -> %10 : !neura.data<f32, i1> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: neura.ctrl_mov %19 -> %8 : !neura.data<f32, i1> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: neura.ctrl_mov %17 -> %6 : !neura.data<f32, i1> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: neura.ctrl_mov %18 -> %4 : !neura.data<f32, i1> !neura.data<f32, i1>
// DATAFLOW_IR-NEXT: "neura.return"(%21) : (!neura.data<f32, i1>) -> ()
// DATAFLOW_IR-NEXT: }

// INTERPRETER_OUTPUT: [neura-interpreter] Executing neura.return:
// INTERPRETER_OUTPUT: [neura-interpreter] ├─ Return values:
// INTERPRETER_OUTPUT: [neura-interpreter] │ └─15.000000, [pred = 1]
// INTERPRETER_OUTPUT: [neura-interpreter] └─ Execution terminated successfully
// INTERPRETER_OUTPUT: [neura-interpreter] DFG Iteration 5 | Topological Level 6 | ready_to_execute_ops 3
// INTERPRETER_OUTPUT-NEXT: [neura-interpreter] ========================================
// INTERPRETER_OUTPUT-NEXT: [neura-interpreter] Executing operation: "neura.return"(%21) : (!neura.data<f32, i1>) -> ()
// INTERPRETER_OUTPUT-NEXT: [neura-interpreter] Executing neura.return:
// INTERPRETER_OUTPUT-NEXT: [neura-interpreter] ├─ Return values:
// INTERPRETER_OUTPUT-NEXT: [neura-interpreter] │ └─15.000000, [pred = 1]
// INTERPRETER_OUTPUT-NEXT: [neura-interpreter] └─ Execution terminated successfully
// INTERPRETER_OUTPUT-NEXT: [neura-interpreter] Termination signal received
Loading