Skip to content

[mlir][arith] Add mulf(x, 0) -> 0 to mulf folder #161395

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
python3kgae merged 8 commits into from
Oct 2, 2025
Merged

[mlir][arith] Add mulf(x, 0) -> 0 to mulf folder #161395

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
7e4a0d4
[mlir][arith] Add mulf(x, 0) -> 0 to mulf folder
Sep 30, 2025
8d49f3f
mulf(NaN, x) -> NaN
Sep 30, 2025
8c72844
Limit to fastmask nnan | ninf
Sep 30, 2025
e86191a
Update per comment.
Sep 30, 2025
bdb0f61
Add nsz.
Sep 30, 2025
18e9504
Remove ninf which is not needed.
Oct 1, 2025
f43001f
Add test for neg zero.
Oct 1, 2025
4019c36
Use CHECK_DAG.
Oct 1, 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
3 changes: 3 additions & 0 deletions mlir/lib/Dialect/Arith/IR/ArithOps.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1281,6 +1281,9 @@ OpFoldResult arith::MulFOp::fold(FoldAdaptor adaptor) {
// mulf(x, 1) -> x
if (matchPattern(adaptor.getRhs(), m_OneFloat()))
return getLhs();
// mulf(x, 0) -> 0
if (matchPattern(adaptor.getRhs(), m_AnyZeroFloat()))
return getRhs();

return constFoldBinaryOp<FloatAttr>(
adaptor.getOperands(),
Expand Down
10 changes: 10 additions & 0 deletions mlir/test/Dialect/Arith/canonicalize.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2216,6 +2216,16 @@ func.func @test_mulf1(%arg0 : f32, %arg1 : f32) -> (f32) {
return %2 : f32
}

// CHECK-LABEL: @test_mulf2(
func.func @test_mulf2(%arg0 : f32, %arg1 : f32) -> (f32, f32) {
// CHECK-NEXT: %[[C0:.+]] = arith.constant 0.000000e+00 : f32
// CHECK-NEXT: return %[[C0]], %[[C0]]
%c0 = arith.constant 0.0 : f32
%0 = arith.mulf %arg0, %c0 : f32
%1 = arith.mulf %c0, %arg1 : f32
return %0, %1 : f32, f32
}
Comment on lines 2219 to 2229
Copy link
Contributor

@ThomasRaoux ThomasRaoux Sep 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

that's not correct for Nan

Copy link
Collaborator

@joker-eph joker-eph Sep 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Also does not preserve the sign of the operand.

We could do all this with fast-math flags.

Copy link
Contributor Author

@python3kgae python3kgae Sep 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed by fold NaN before 0.

Copy link
Contributor

@ThomasRaoux ThomasRaoux Sep 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

that's still not correct, the value may dynamically be Nan and not be a constant Nan

Copy link
Contributor Author

@python3kgae python3kgae Sep 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I see.
Updated with fast-math flags.

Copy link
Collaborator

@joker-eph joker-eph Sep 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is nnan/ninf enough to lose the sign of the input?

Copy link
Contributor Author

@python3kgae python3kgae Sep 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't think so :( .
Will FastMathFlags::nsz cover it or we'll have to go FastMathFlags::fast?

Copy link
Contributor Author

@python3kgae python3kgae Sep 30, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Added FastMathFlags::nsz.


// -----

// CHECK-LABEL: @test_divf(
Expand Down
12 changes: 6 additions & 6 deletions mlir/test/Dialect/SCF/loop-pipelining.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -930,31 +930,31 @@ func.func @dynamic_loop_result(%A: memref<?xf32>, %result: memref<?xf32>, %lb: i
// CHECK-DAG: %[[C0:.*]] = arith.constant 0 : index
// CHECK-DAG: %[[C1:.*]] = arith.constant 1 : index
// CHECK-DAG: %[[C3:.*]] = arith.constant 3 : index
// CHECK-DAG: %[[CST0:.*]] = arith.constant 0.000000e+00 : f32
// CHECK-DAG: %[[CST10:.*]] = arith.constant 1.000000e+01 : f32
// CHECK-DAG: %[[CST2:.*]] = arith.constant 2.000000e+00 : f32
// Prologue:
// CHECK: %[[L0:.*]] = memref.load %[[A]][%[[C0]]] : memref<?xf32>
// Kernel:
// CHECK-NEXT: %[[L1:.*]]:2 = scf.for %[[IV:.*]] = %[[C0]] to %[[C3]]
// CHECK-SAME: step %[[C1]] iter_args(%[[ARG0:.*]] = %[[CST2]], %[[ARG1:.*]] = %[[L0]]) -> (f32, f32) {
// CHECK-NEXT: %[[ADD0:.*]] = arith.addf %[[ARG1]], %[[ARG0]] : f32
// CHECK-NEXT: %[[MUL0:.*]] = arith.mulf %[[ADD0]], %[[CST0]] : f32
// CHECK-NEXT: %[[MUL0:.*]] = arith.mulf %[[ADD0]], %[[CST10]] : f32
// CHECK-NEXT: memref.store %[[MUL0]], %[[A]][%[[IV]]] : memref<?xf32>
// CHECK-NEXT: %[[IV1:.*]] = arith.addi %[[IV]], %[[C1]] : index
// CHECK-NEXT: %[[L2:.*]] = memref.load %[[A]][%[[IV1]]] : memref<?xf32>
// CHECK-NEXT: scf.yield %[[CST0]], %[[L2]] : f32
// CHECK-NEXT: scf.yield %[[CST10]], %[[L2]] : f32
// CHECK-NEXT: }
// Epilogue:
// CHECK-NEXT: %[[ADD1:.*]] = arith.addf %[[L1]]#1, %[[CST0]] : f32
// CHECK-NEXT: %[[MUL1:.*]] = arith.mulf %[[ADD1]], %[[CST0]] : f32
// CHECK-NEXT: %[[ADD1:.*]] = arith.addf %[[L1]]#1, %[[CST10]] : f32
// CHECK-NEXT: %[[MUL1:.*]] = arith.mulf %[[ADD1]], %[[CST10]] : f32
// CHECK-NEXT: memref.store %[[MUL1]], %[[A]][%[[C3]]] : memref<?xf32>
// CHECK-NEXT: return %[[L1]]#0 : f32

func.func @yield_constant_loop(%A: memref<?xf32>) -> f32 {
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%c4 = arith.constant 4 : index
%cf0 = arith.constant 0.0 : f32
%cf0 = arith.constant 10.0 : f32
%cf2 = arith.constant 2.0 : f32
%r = scf.for %i0 = %c0 to %c4 step %c1 iter_args(%arg0 = %cf2) -> f32 {
%A_elem = memref.load %A[%i0] { __test_pipelining_stage__ = 0, __test_pipelining_op_order__ = 3 } : memref<?xf32>
Expand Down
Loading