Skip to content

[SimplifyLibCalls] Add f128 type for merging sqrt into the power of exp #112373

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

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[SimplifyLibCalls] Add f128 type for merging sqrt into the power of exp #112373

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
3 changes: 3 additions & 0 deletions llvm/lib/Transforms/Utils/SimplifyLibCalls.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2720,6 +2720,9 @@ Value *LibCallSimplifier::mergeSqrtToExp(CallInst *CI, IRBuilderBase &B) {
ExpLb = LibFunc_exp;
Exp2Lb = LibFunc_exp2;
Exp10Lb = LibFunc_exp10;
} else if (CI->getType()->getScalarType()->isFP128Ty()) {
ExpLb = LibFunc_expl;
Copy link
Contributor

@arsenm arsenm Oct 15, 2024

Choose a reason for hiding this comment

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

This intrinsic->intrinsic transform really shouldn't depend on the libcall availability...

Exp2Lb = LibFunc_exp2l;
} else
return nullptr;
} else
Expand Down
22 changes: 22 additions & 0 deletions llvm/test/Transforms/InstCombine/sqrt.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -201,6 +201,25 @@ define <2 x float> @sqrt_exp_vec(<2 x float> %x) {
ret <2 x float> %res
}

define dso_local fp128 @sqrt_exp_long_double(fp128 %x, fp128 %y) {
Copy link
Contributor

@arsenm arsenm Oct 15, 2024

Choose a reason for hiding this comment

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

Don't need dso_local

; CHECK-LABEL: @sqrt_exp_long_double(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[MERGED_SQRT:%.*]] = fmul fast fp128 [[X:%.*]], 0xL00000000000000003FFE000000000000
; CHECK-NEXT: [[TMP0:%.*]] = call fast fp128 @llvm.exp.f128(fp128 [[MERGED_SQRT]])
; CHECK-NEXT: [[MERGED_SQRT1:%.*]] = fmul fast fp128 [[Y:%.*]], 0xL00000000000000003FFE000000000000
; CHECK-NEXT: [[TMP1:%.*]] = call fast fp128 @llvm.exp2.f128(fp128 [[MERGED_SQRT1]])
; CHECK-NEXT: [[ADD:%.*]] = fadd fast fp128 [[TMP0]], [[TMP1]]
; CHECK-NEXT: ret fp128 [[ADD]]
;
entry:
%0 = call fast fp128 @llvm.exp.f128(fp128 %x)
%1 = call fast fp128 @llvm.sqrt.f128(fp128 %0)
%2 = call fast fp128 @llvm.exp2.f128(fp128 %y)
%3 = call fast fp128 @llvm.sqrt.f128(fp128 %2)
%add = fadd fast fp128 %1, %3
Comment on lines +215 to +219
Copy link
Contributor

@arsenm arsenm Oct 15, 2024

Choose a reason for hiding this comment

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

Test the different cases in different functions.

Use minimal fast math flags, this looks like it only requires reassoc. Add an additional test which shows the preservation behavior of other fast math flags.

Also test the same situation with equivalent library calls instead of intrinsics (and the mixed intrinsic + libcall case)

ret fp128 %add
}

declare i32 @foo(double)
declare double @sqrt(double) readnone
declare float @sqrtf(float)
Expand All @@ -212,3 +231,6 @@ declare double @exp2(double)
declare double @exp10(double)
declare <2 x float> @llvm.exp.v2f32(<2 x float>)
declare <2 x float> @llvm.sqrt.v2f32(<2 x float>)
declare fp128 @llvm.exp.f128(fp128)
declare fp128 @llvm.sqrt.f128(fp128)
declare fp128 @llvm.exp2.f128(fp128)
Loading