Skip to content

[X86][Inline] Check correct function for target feature check #152515

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 2 commits into
base: main
Choose a base branch
from
Open

[X86][Inline] Check correct function for target feature check #152515

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
05221b4
[X86][Inline] Check correct function for target feature check
nikic Aug 7, 2025
26016e2
fix typo
nikic Aug 7, 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
18 changes: 4 additions & 14 deletions llvm/lib/Target/X86/X86TargetTransformInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6525,8 +6525,8 @@ bool X86TTIImpl::areInlineCompatible(const Function *Caller,

for (const Instruction &I : instructions(Callee)) {
if (const auto *CB = dyn_cast<CallBase>(&I)) {
// Having more target features is fine for inline ASM.
if (CB->isInlineAsm())
// Having more target features is fine for inline ASM and intrinsics.
if (CB->isInlineAsm() || CB->getIntrinsicID() != Intrinsic::not_intrinsic)
continue;

SmallVector<Type *, 8> Types;
Expand All @@ -6542,19 +6542,9 @@ bool X86TTIImpl::areInlineCompatible(const Function *Caller,
if (all_of(Types, IsSimpleTy))
continue;

if (Function *NestedCallee = CB->getCalledFunction()) {
// Assume that intrinsics are always ABI compatible.
if (NestedCallee->isIntrinsic())
continue;

// Do a precise compatibility check.
if (!areTypesABICompatible(Caller, NestedCallee, Types))
return false;
} else {
// We don't know the target features of the callee,
// assume it is incompatible.
// Do a precise compatibility check.
if (!areTypesABICompatible(Caller, Callee, Types))
return false;
}
}
}
return true;
Expand Down
38 changes: 32 additions & 6 deletions llvm/test/Transforms/Inline/X86/call-abi-compatibility.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,8 +34,7 @@ define i64 @callee_not_avx(<4 x i64> %arg) noinline {
ret i64 %v
}

; This call also shouldn't be inlined, as we don't know whether callee_unknown
; is ABI compatible or not.
; This call also shouldn't be inlined, as caller_not_avx2 is not ABI compatible.
define void @caller_avx2() "target-features"="+avx" {
; CHECK-LABEL: define {{[^@]+}}@caller_avx2
; CHECK-SAME: () #[[ATTR0]] {
Expand All @@ -55,27 +54,54 @@ define internal void @caller_not_avx2() {
ret void
}

; Should be inlined, as caller_avx4 is ABI compatible. The fact that we don't
Copy link
Collaborator

Choose a reason for hiding this comment

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

caller_avx4 -> caller_avx7 ?

; know anything about callee_unknown doesn't matter, as it is the caller that
; determines the ABI as far as target features are concerned.
define void @caller_avx6() "target-features"="+avx" {
; CHECK-LABEL: define {{[^@]+}}@caller_avx6
; CHECK-SAME: () #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = call i64 @callee_unknown(<4 x i64> <i64 0, i64 1, i64 2, i64 3>)
; CHECK-NEXT: ret void
;
call void @caller_avx7()
ret void
}

define void @caller_avx7() "target-features"="+avx" {
; CHECK-LABEL: define {{[^@]+}}@caller_avx7
; CHECK-SAME: () #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = call i64 @callee_unknown(<4 x i64> <i64 0, i64 1, i64 2, i64 3>)
; CHECK-NEXT: ret void
;
call i64 @callee_unknown(<4 x i64> <i64 0, i64 1, i64 2, i64 3>)
ret void
}

declare i64 @callee_unknown(<4 x i64>)

; This call should get inlined, because we assume that intrinsics are always
; ABI compatible.
define void @caller_avx3() "target-features"="+avx" {
; CHECK-LABEL: define {{[^@]+}}@caller_avx3
; CHECK-SAME: () #[[ATTR0]] {
; CHECK-NEXT: [[TMP1:%.*]] = call i64 @llvm.some_intrinsic(<4 x i64> <i64 0, i64 1, i64 2, i64 3>)
; CHECK-NEXT: [[V_I:%.*]] = load <4 x i64>, ptr @g, align 32
; CHECK-NEXT: [[V2_I:%.*]] = call <4 x i64> @llvm.abs.v4i64(<4 x i64> [[V_I]], i1 false)
; CHECK-NEXT: store <4 x i64> [[V2_I]], ptr @g, align 32
; CHECK-NEXT: ret void
;
call void @caller_not_avx3()
ret void
}

@g = external global <4 x i64>

define internal void @caller_not_avx3() {
call i64 @llvm.some_intrinsic(<4 x i64> <i64 0, i64 1, i64 2, i64 3>)
%v = load <4 x i64>, ptr @g
%v2 = call <4 x i64> @llvm.abs(<4 x i64> %v, i1 false)
store <4 x i64> %v2, ptr @g
ret void
}

declare i64 @llvm.some_intrinsic(<4 x i64>)

; This call should get inlined, because only simple types are involved.
define void @caller_avx4() "target-features"="+avx" {
; CHECK-LABEL: define {{[^@]+}}@caller_avx4
Expand Down
Loading