This is an archive of the discontinued LLVM Phabricator instance.

Differential D137456

[mlir][arith] Convert fastmath to LLVM dialect for some arith ops.
ClosedPublic

Authored by vzakhari on Nov 4 2022, 12:16 PM.

Details

Reviewers
jfurtek
ftynse
rriddle
dcaballe
Commits
rGe72fb692104e: [mlir][arith] Convert fastmath to LLVM dialect for some arith ops.
Summary

This is a follow-up on D126305 and D136225.
We can now preserve fastmath for arith::MaxFOp,MinFOp,RemFOp during
ArithToLLVM conversion.

Diff Detail

Event Timeline

vzakhari created this revision.Nov 4 2022, 12:16 PM
Herald added a project: Restricted Project. · View Herald TranscriptNov 4 2022, 12:16 PM
Herald added subscribers: Moerafaat, zero9178, bzcheeseman and 20 others. · View Herald Transcript
vzakhari requested review of this revision.Nov 4 2022, 12:16 PM
Herald added a reviewer: dcaballe. · View Herald TranscriptNov 4 2022, 12:16 PM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: stephenneuendorffer, nicolasvasilache. · View Herald Transcript
Harbormaster completed remote builds in B196199: Diff 473313.Nov 4 2022, 12:45 PM
rriddle accepted this revision.Nov 7 2022, 6:44 PM
rriddle added inline comments.
mlir/test/Conversion/ArithToLLVM/arith-to-llvm.mlir
473

I know the others have it, but can we drop the leading {{.*}} = ? I don't think it adds anything to these tests.

This revision is now accepted and ready to land.Nov 7 2022, 6:44 PM
vzakhari updated this revision to Diff 473862.Nov 7 2022, 7:55 PM

Thank you for the review, River! I changed the test.

This revision was landed with ongoing or failed builds.Nov 7 2022, 7:57 PM
Closed by commit rGe72fb692104e: [mlir][arith] Convert fastmath to LLVM dialect for some arith ops. (authored by vzakhari). · Explain Why
This revision was automatically updated to reflect the committed changes.
vzakhari added a commit: rGe72fb692104e: [mlir][arith] Convert fastmath to LLVM dialect for some arith ops..
Harbormaster completed remote builds in B196626: Diff 473862.Nov 7 2022, 8:11 PM
tblah mentioned this in D138907: [flang] Remove warnings that fast-math options are unimplemented.Nov 29 2022, 3:40 AM
tblah mentioned this in rG6841c43f36ee: [flang] Remove warnings that fast-math options are unimplemented.Nov 29 2022, 9:17 AM
Dinistro mentioned this in D143912: [mlir] Drop unused arith conversion class (NFC).Feb 13 2023, 6:49 AM
Dinistro mentioned this in rG544831ab9f35: [mlir] Drop unused arith conversion class (NFC).Feb 14 2023, 1:22 AM

Revision Contents

PathSize
mlir/
lib/
Conversion/
ArithToLLVM/
18 lines
test/
Conversion/
ArithToLLVM/
33 lines

Diff 473863

mlir/lib/Conversion/ArithToLLVM/ArithToLLVM.cpp

Show First 20 Lines • Show All 46 Lines • ▼ Show 20 Lines
using ExtSIOpLowering = using ExtSIOpLowering =
VectorConvertToLLVMPattern<arith::ExtSIOp, LLVM::SExtOp>; VectorConvertToLLVMPattern<arith::ExtSIOp, LLVM::SExtOp>;
using ExtUIOpLowering = using ExtUIOpLowering =
VectorConvertToLLVMPattern<arith::ExtUIOp, LLVM::ZExtOp>; VectorConvertToLLVMPattern<arith::ExtUIOp, LLVM::ZExtOp>;
using FPToSIOpLowering = using FPToSIOpLowering =
VectorConvertToLLVMPattern<arith::FPToSIOp, LLVM::FPToSIOp>; VectorConvertToLLVMPattern<arith::FPToSIOp, LLVM::FPToSIOp>;
using FPToUIOpLowering = using FPToUIOpLowering =
VectorConvertToLLVMPattern<arith::FPToUIOp, LLVM::FPToUIOp>; VectorConvertToLLVMPattern<arith::FPToUIOp, LLVM::FPToUIOp>;
// TODO: Add LLVM intrinsic support for fastmath using MaxFOpLowering =
using MaxFOpLowering = VectorConvertToLLVMPattern<arith::MaxFOp, LLVM::MaxNumOp, VectorConvertToLLVMPattern<arith::MaxFOp, LLVM::MaxNumOp,
arith::AttrDropFastMath>; arith::AttrConvertFastMathToLLVM>;
using MaxSIOpLowering = using MaxSIOpLowering =
VectorConvertToLLVMPattern<arith::MaxSIOp, LLVM::SMaxOp>; VectorConvertToLLVMPattern<arith::MaxSIOp, LLVM::SMaxOp>;
using MaxUIOpLowering = using MaxUIOpLowering =
VectorConvertToLLVMPattern<arith::MaxUIOp, LLVM::UMaxOp>; VectorConvertToLLVMPattern<arith::MaxUIOp, LLVM::UMaxOp>;
// TODO: Add LLVM intrinsic support for fastmath using MinFOpLowering =
using MinFOpLowering = VectorConvertToLLVMPattern<arith::MinFOp, LLVM::MinNumOp, VectorConvertToLLVMPattern<arith::MinFOp, LLVM::MinNumOp,
arith::AttrDropFastMath>; arith::AttrConvertFastMathToLLVM>;
using MinSIOpLowering = using MinSIOpLowering =
VectorConvertToLLVMPattern<arith::MinSIOp, LLVM::SMinOp>; VectorConvertToLLVMPattern<arith::MinSIOp, LLVM::SMinOp>;
using MinUIOpLowering = using MinUIOpLowering =
VectorConvertToLLVMPattern<arith::MinUIOp, LLVM::UMinOp>; VectorConvertToLLVMPattern<arith::MinUIOp, LLVM::UMinOp>;
using MulFOpLowering = using MulFOpLowering =
VectorConvertToLLVMPattern<arith::MulFOp, LLVM::FMulOp, VectorConvertToLLVMPattern<arith::MulFOp, LLVM::FMulOp,
arith::AttrConvertFastMathToLLVM>; arith::AttrConvertFastMathToLLVM>;
using MulIOpLowering = VectorConvertToLLVMPattern<arith::MulIOp, LLVM::MulOp>; using MulIOpLowering = VectorConvertToLLVMPattern<arith::MulIOp, LLVM::MulOp>;
using NegFOpLowering = using NegFOpLowering =
VectorConvertToLLVMPattern<arith::NegFOp, LLVM::FNegOp, VectorConvertToLLVMPattern<arith::NegFOp, LLVM::FNegOp,
arith::AttrConvertFastMathToLLVM>; arith::AttrConvertFastMathToLLVM>;
using OrIOpLowering = VectorConvertToLLVMPattern<arith::OrIOp, LLVM::OrOp>; using OrIOpLowering = VectorConvertToLLVMPattern<arith::OrIOp, LLVM::OrOp>;
// TODO: Add LLVM intrinsic support for fastmath using RemFOpLowering =
using RemFOpLowering = VectorConvertToLLVMPattern<arith::RemFOp, LLVM::FRemOp, VectorConvertToLLVMPattern<arith::RemFOp, LLVM::FRemOp,
arith::AttrDropFastMath>; arith::AttrConvertFastMathToLLVM>;
using RemSIOpLowering = using RemSIOpLowering =
VectorConvertToLLVMPattern<arith::RemSIOp, LLVM::SRemOp>; VectorConvertToLLVMPattern<arith::RemSIOp, LLVM::SRemOp>;
using RemUIOpLowering = using RemUIOpLowering =
VectorConvertToLLVMPattern<arith::RemUIOp, LLVM::URemOp>; VectorConvertToLLVMPattern<arith::RemUIOp, LLVM::URemOp>;
using SelectOpLowering = using SelectOpLowering =
VectorConvertToLLVMPattern<arith::SelectOp, LLVM::SelectOp>; VectorConvertToLLVMPattern<arith::SelectOp, LLVM::SelectOp>;
using ShLIOpLowering = VectorConvertToLLVMPattern<arith::ShLIOp, LLVM::ShlOp>; using ShLIOpLowering = VectorConvertToLLVMPattern<arith::ShLIOp, LLVM::ShlOp>;
using ShRSIOpLowering = using ShRSIOpLowering =
▲ Show 20 Lines • Show All 331 Lines • Show Last 20 Lines

mlir/test/Conversion/ArithToLLVM/arith-to-llvm.mlir

Show First 20 Lines • Show All 447 Lines • ▼ Show 20 Linesfunc.func @minmaxf(%arg0 : f32, %arg1 : f32) -> f32 {
%1 = arith.maxf %arg0, %arg1 : f32 %1 = arith.maxf %arg0, %arg1 : f32
return %0 : f32 return %0 : f32
} }
// ----- // -----
// CHECK-LABEL: @fastmath // CHECK-LABEL: @fastmath
func.func @fastmath(%arg0: f32, %arg1: f32, %arg2: i32) { func.func @fastmath(%arg0: f32, %arg1: f32, %arg2: i32) {
// CHECK: {{.*}} = llvm.fadd %arg0, %arg1 {fastmathFlags = #llvm.fastmath<fast>} : f32 // CHECK: llvm.fadd %arg0, %arg1 {fastmathFlags = #llvm.fastmath<fast>} : f32
// CHECK: {{.*}} = llvm.fmul %arg0, %arg1 {fastmathFlags = #llvm.fastmath<fast>} : f32 // CHECK: llvm.fmul %arg0, %arg1 {fastmathFlags = #llvm.fastmath<fast>} : f32
// CHECK: {{.*}} = llvm.fneg %arg0 {fastmathFlags = #llvm.fastmath<fast>} : f32 // CHECK: llvm.fneg %arg0 {fastmathFlags = #llvm.fastmath<fast>} : f32
// CHECK: {{.*}} = llvm.fadd %arg0, %arg1 : f32 // CHECK: llvm.fadd %arg0, %arg1 : f32
// CHECK: {{.*}} = llvm.fadd %arg0, %arg1 {fastmathFlags = #llvm.fastmath<nnan, ninf>} : f32 // CHECK: llvm.fadd %arg0, %arg1 {fastmathFlags = #llvm.fastmath<nnan, ninf>} : f32
%0 = arith.addf %arg0, %arg1 fastmath<fast> : f32 %0 = arith.addf %arg0, %arg1 fastmath<fast> : f32
%1 = arith.mulf %arg0, %arg1 fastmath<fast> : f32 %1 = arith.mulf %arg0, %arg1 fastmath<fast> : f32
%2 = arith.negf %arg0 fastmath<fast> : f32 %2 = arith.negf %arg0 fastmath<fast> : f32
%3 = arith.addf %arg0, %arg1 fastmath<none> : f32 %3 = arith.addf %arg0, %arg1 fastmath<none> : f32
%4 = arith.addf %arg0, %arg1 fastmath<nnan,ninf> : f32 %4 = arith.addf %arg0, %arg1 fastmath<nnan,ninf> : f32
return return
} }
// -----
// CHECK-LABEL: @ops_supporting_fastmath
func.func @ops_supporting_fastmath(%arg0: f32, %arg1: f32, %arg2: i32) {
// CHECK: llvm.fadd %arg0, %arg1 {fastmathFlags = #llvm.fastmath<fast>} : f32
rriddleUnsubmitted
Not Done
ReplyInline Actions

I know the others have it, but can we drop the leading {{.*}} = ? I don't think it adds anything to these tests.

rriddle: I know the others have it, but can we drop the leading `{{.*}} = `? I don't think it adds…
%0 = arith.addf %arg0, %arg1 fastmath<fast> : f32
// CHECK: llvm.fdiv %arg0, %arg1 {fastmathFlags = #llvm.fastmath<fast>} : f32
%1 = arith.divf %arg0, %arg1 fastmath<fast> : f32
// CHECK: llvm.intr.maxnum(%arg0, %arg1) {fastmathFlags = #llvm.fastmath<fast>} : (f32, f32) -> f32
%2 = arith.maxf %arg0, %arg1 fastmath<fast> : f32
// CHECK: llvm.intr.minnum(%arg0, %arg1) {fastmathFlags = #llvm.fastmath<fast>} : (f32, f32) -> f32
%3 = arith.minf %arg0, %arg1 fastmath<fast> : f32
// CHECK: llvm.fmul %arg0, %arg1 {fastmathFlags = #llvm.fastmath<fast>} : f32
%4 = arith.mulf %arg0, %arg1 fastmath<fast> : f32
// CHECK: llvm.fneg %arg0 {fastmathFlags = #llvm.fastmath<fast>} : f32
%5 = arith.negf %arg0 fastmath<fast> : f32
// CHECK: llvm.frem %arg0, %arg1 {fastmathFlags = #llvm.fastmath<fast>} : f32
%6 = arith.remf %arg0, %arg1 fastmath<fast> : f32
// CHECK: llvm.fsub %arg0, %arg1 {fastmathFlags = #llvm.fastmath<fast>} : f32
%7 = arith.subf %arg0, %arg1 fastmath<fast> : f32
return
}