Skip to content

Avoid unnecessary phi #35

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
tancheng merged 1 commit into from
Jun 15, 2025
Merged

Avoid unnecessary phi #35

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
33 changes: 33 additions & 0 deletions lib/NeuraDialect/Transforms/TransformCtrlToDataFlowPass.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -113,6 +113,39 @@ void createPhiNodesForBlock(Block *block, OpBuilder &builder,

assert(!phi_operands.empty());

// Puts all operands into a set to ensure uniqueness. Specifically, following
// case is handled:
// ---------------------------------------------------------
// ^bb1:
// "neura.br"(%a)[^bb3] : (!neura.data<f32, i1>) -> ()
//
// ^bb2:
// "neura.br"(%a)[^bb3] : (!neura.data<f32, i1>) -> ()
//
// ^bb3(%x: !neura.data<f32, i1>):
// ...
// ---------------------------------------------------------
// In above case, %a is used in both branches of the control flow, so we
// don't need a phi node, but we still need to replace its uses with the
// result of the phi node.
// This ensures that we only create a phi node if there are multiple unique
// operands.
llvm::SmallDenseSet<Value, 4> unique_operands(phi_operands.begin(), phi_operands.end());

if (unique_operands.size() == 1) {
// No phi needed, but still replace
Value single = *unique_operands.begin();
SmallVector<OpOperand *, 4> uses;
for (OpOperand &use : live_in.getUses()) {
uses.push_back(&use);
}
for (OpOperand *use : uses) {
use->set(single);
}
value_map[live_in] = single;
continue;
}

// Creates the phi node with dynamic number of operands.
auto phi_op = builder.create<neura::PhiOp>(loc, predicated_type, phi_operands);

Expand Down
65 changes: 65 additions & 0 deletions test/neura/ctrl/branch.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,65 @@
// RUN: mlir-neura-opt %s \
// RUN: --assign-accelerator \
// RUN: --lower-llvm-to-neura \
// RUN: --leverage-predicated-value \
// RUN: | FileCheck %s

// RUN: mlir-neura-opt %s \
// RUN: --assign-accelerator \
// RUN: --lower-llvm-to-neura \
// RUN: --leverage-predicated-value \
// RUN: --transform-ctrl-to-data-flow \
// RUN: | FileCheck %s -check-prefix=CTRL2DATA

func.func @test(%in: i64) -> f32 {
%c0 = llvm.mlir.constant(0 : i64) : i64
%c1 = llvm.mlir.constant(1.0 : f32) : f32
%c2 = llvm.mlir.constant(2.0 : f32) : f32
%c3 = llvm.mlir.constant(3.0 : f32) : f32
%c4 = llvm.mlir.constant(4.0 : f32) : f32
%cond = llvm.icmp "eq" %in, %c0 : i64
llvm.cond_br %cond, ^bb2(%c3, %c4 : f32, f32), ^bb1(%c1, %c2 : f32, f32)

^bb1(%ca: f32, %cb: f32):
%a = llvm.fadd %ca, %cb : f32
llvm.br ^bb3(%a : f32)

^bb2(%cc: f32, %cd: f32):
%b = llvm.fmul %cc, %cd : f32
llvm.br ^bb3(%b : f32)

^bb3(%v: f32):
return %v : f32
}


// CHECK: func.func @test(%arg0: i64) -> f32 attributes {accelerator = "neura"} {
// CHECK-NEXT: %0 = "neura.constant"() <{predicate = true, value = 0 : i64}> : () -> !neura.data<i64, i1>
// CHECK-NEXT: %1 = "neura.constant"() <{predicate = true, value = 1.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// CHECK-NEXT: %2 = "neura.constant"() <{predicate = true, value = 2.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// CHECK-NEXT: %3 = "neura.constant"() <{predicate = true, value = 3.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// CHECK-NEXT: %4 = "neura.constant"() <{predicate = true, value = 4.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// CHECK-NEXT: %5 = "neura.icmp"(%arg0, %0) <{cmpType = "eq"}> : (i64, !neura.data<i64, i1>) -> !neura.data<i1, i1>
// CHECK-NEXT: neura.cond_br %5 : !neura.data<i1, i1> then %3, %4 : !neura.data<f32, i1>, !neura.data<f32, i1> to ^bb2 else %1, %2 : !neura.data<f32, i1>, !neura.data<f32, i1> to ^bb1
// CHECK-NEXT: ^bb1(%6: !neura.data<f32, i1>, %7: !neura.data<f32, i1>): // pred: ^bb0
// CHECK-NEXT: %8 = "neura.fadd"(%6, %7) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CHECK-NEXT: neura.br %8 : !neura.data<f32, i1> to ^bb3
// CHECK-NEXT: ^bb2(%9: !neura.data<f32, i1>, %10: !neura.data<f32, i1>): // pred: ^bb0
// CHECK-NEXT: %11 = "neura.fmul"(%9, %10) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CHECK-NEXT: neura.br %11 : !neura.data<f32, i1> to ^bb3
// CHECK-NEXT: ^bb3(%12: !neura.data<f32, i1>): // 2 preds: ^bb1, ^bb2
// CHECK-NEXT: "neura.return"(%12) : (!neura.data<f32, i1>) -> ()
// CHECK-NEXT: }

// CTRL2DATA: func.func @test(%arg0: i64) -> f32 attributes {accelerator = "neura"} {
// CTRL2DATA-NEXT: %0 = "neura.constant"() <{predicate = true, value = 0 : i64}> : () -> !neura.data<i64, i1>
// CTRL2DATA-NEXT: %1 = "neura.constant"() <{predicate = true, value = 1.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %2 = "neura.constant"() <{predicate = true, value = 2.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %3 = "neura.constant"() <{predicate = true, value = 3.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %4 = "neura.constant"() <{predicate = true, value = 4.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %5 = "neura.icmp"(%arg0, %0) <{cmpType = "eq"}> : (i64, !neura.data<i64, i1>) -> !neura.data<i1, i1>
// CTRL2DATA-NEXT: %6 = "neura.fadd"(%1, %2) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %7 = "neura.fmul"(%3, %4) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %8 = "neura.phi"(%6, %7) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: "neura.return"(%8) : (!neura.data<f32, i1>) -> ()
// CTRL2DATA-NEXT: }
12 changes: 4 additions & 8 deletions test/neura/ctrl/branch_no_arg.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,12 +57,8 @@ func.func @test(%in: i64) -> f32 {
// CTRL2DATA-NEXT: %3 = "neura.constant"() <{predicate = true, value = 3.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %4 = "neura.constant"() <{predicate = true, value = 4.000000e+00 : f32}> : () -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %5 = "neura.icmp"(%arg0, %0) <{cmpType = "eq"}> : (i64, !neura.data<i64, i1>) -> !neura.data<i1, i1>
// CTRL2DATA-NEXT: %6 = "neura.phi"(%1) : (!neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %7 = "neura.phi"(%2) : (!neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %8 = "neura.fadd"(%6, %7) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %9 = "neura.phi"(%3) : (!neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %10 = "neura.phi"(%4) : (!neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %11 = "neura.fmul"(%9, %10) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %12 = "neura.phi"(%8, %11) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: "neura.return"(%12) : (!neura.data<f32, i1>) -> ()
// CTRL2DATA-NEXT: %6 = "neura.fadd"(%1, %2) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %7 = "neura.fmul"(%3, %4) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: %8 = "neura.phi"(%6, %7) : (!neura.data<f32, i1>, !neura.data<f32, i1>) -> !neura.data<f32, i1>
// CTRL2DATA-NEXT: "neura.return"(%8) : (!neura.data<f32, i1>) -> ()
// CTRL2DATA-NEXT: }