Skip to content

[Clang][CodeGen] Emit fake uses before musttail calls #136867

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
merged 5 commits into from
Apr 25, 2025
Merged

[Clang][CodeGen] Emit fake uses before musttail calls #136867

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
90c09c8
[Clang] Emit Fake Uses before musttail calls
SLTozer Apr 23, 2025
2765fb7
Simplify test
SLTozer Apr 24, 2025
d0e4e68
Use extern C to prevent name mangling
SLTozer Apr 24, 2025
c8c371f
Reduce comment, further simplify test
SLTozer Apr 25, 2025
bc98cd0
Set explicit triple
SLTozer Apr 25, 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
12 changes: 11 additions & 1 deletion clang/lib/CodeGen/CGCall.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6001,8 +6001,18 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
for (auto it = EHStack.find(CurrentCleanupScopeDepth); it != EHStack.end();
++it) {
EHCleanupScope *Cleanup = dyn_cast<EHCleanupScope>(&*it);
if (!(Cleanup && Cleanup->getCleanup()->isRedundantBeforeReturn()))
// Fake uses can be safely emitted immediately prior to the tail call; we
// choose to emit the fake use before the call rather than after, to avoid
// forcing variable values from every call on the "stack" to be preserved
// simultaneously.
Copy link
Contributor

@rjmccall rjmccall Apr 25, 2025

Choose a reason for hiding this comment

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

The second half of this comment is odd. There isn't really a choice about this; the fake uses must be emitted prior to the tail call because there cannot be anything after it.

if (Cleanup && Cleanup->isFakeUse()) {
CGBuilderTy::InsertPointGuard IPG(Builder);
Builder.SetInsertPoint(CI);
Cleanup->getCleanup()->Emit(*this, EHScopeStack::Cleanup::Flags());
} else if (!(Cleanup &&
Cleanup->getCleanup()->isRedundantBeforeReturn())) {
CGM.ErrorUnsupported(MustTailCall, "tail call skipping over cleanups");
}
}
if (CI->getType()->isVoidTy())
Builder.CreateRetVoid();
Expand Down
56 changes: 56 additions & 0 deletions clang/test/CodeGenCXX/fake-use-musttail.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,56 @@
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
// RUN: %clang_cc1 -emit-llvm -fextend-variable-liveness -o - %s | FileCheck %s

/// Tests that when we have fake uses in a function ending in a musttail call,
/// we emit the fake uses and their corresponding loads immediately prior to the
/// tail call.

extern "C" char *bar(int *, const char *, int *, const int *, unsigned long);

// CHECK-LABEL: define dso_local ptr @foo(
// CHECK-SAME: ptr noundef [[E:%.*]], ptr noundef [[F:%.*]], ptr noundef [[G:%.*]], ptr noundef [[H:%.*]], i64 noundef [[I:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK-NEXT: [[ENTRY:.*:]]
// CHECK-NEXT: [[E_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT: [[F_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT: [[G_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT: [[H_ADDR:%.*]] = alloca ptr, align 8
// CHECK-NEXT: [[I_ADDR:%.*]] = alloca i64, align 8
// CHECK-NEXT: store ptr [[E]], ptr [[E_ADDR]], align 8
// CHECK-NEXT: store ptr [[F]], ptr [[F_ADDR]], align 8
// CHECK-NEXT: store ptr [[G]], ptr [[G_ADDR]], align 8
// CHECK-NEXT: store ptr [[H]], ptr [[H_ADDR]], align 8
// CHECK-NEXT: store i64 [[I]], ptr [[I_ADDR]], align 8
// CHECK-NEXT: [[TMP0:%.*]] = load ptr, ptr [[E_ADDR]], align 8
// CHECK-NEXT: [[TMP1:%.*]] = load ptr, ptr [[F_ADDR]], align 8
// CHECK-NEXT: [[TMP2:%.*]] = load ptr, ptr [[G_ADDR]], align 8
// CHECK-NEXT: [[TMP3:%.*]] = load ptr, ptr [[H_ADDR]], align 8
// CHECK-NEXT: [[TMP4:%.*]] = load i64, ptr [[I_ADDR]], align 8
// CHECK-NEXT: [[FAKE_USE:%.*]] = load i64, ptr [[I_ADDR]], align 8
// CHECK-NEXT: notail call void (...) @llvm.fake.use(i64 [[FAKE_USE]]) #[[ATTR3:[0-9]+]]
Copy link
Contributor

@rjmccall rjmccall Apr 25, 2025

Choose a reason for hiding this comment

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

If you're not testing the attributes, you can just leave this off; matches don't have to match the entire line.

// CHECK-NEXT: [[FAKE_USE1:%.*]] = load ptr, ptr [[H_ADDR]], align 8
// CHECK-NEXT: notail call void (...) @llvm.fake.use(ptr [[FAKE_USE1]]) #[[ATTR3]]
// CHECK-NEXT: [[FAKE_USE2:%.*]] = load ptr, ptr [[G_ADDR]], align 8
// CHECK-NEXT: notail call void (...) @llvm.fake.use(ptr [[FAKE_USE2]]) #[[ATTR3]]
// CHECK-NEXT: [[FAKE_USE3:%.*]] = load ptr, ptr [[F_ADDR]], align 8
// CHECK-NEXT: notail call void (...) @llvm.fake.use(ptr [[FAKE_USE3]]) #[[ATTR3]]
// CHECK-NEXT: [[FAKE_USE4:%.*]] = load ptr, ptr [[E_ADDR]], align 8
// CHECK-NEXT: notail call void (...) @llvm.fake.use(ptr [[FAKE_USE4]]) #[[ATTR3]]
// CHECK-NEXT: [[CALL:%.*]] = musttail call ptr @bar(ptr noundef [[TMP0]], ptr noundef [[TMP1]], ptr noundef [[TMP2]], ptr noundef [[TMP3]], i64 noundef [[TMP4]])
// CHECK-NEXT: ret ptr [[CALL]]
// CHECK: [[BB5:.*:]]
// CHECK-NEXT: [[FAKE_USE5:%.*]] = load i64, ptr [[I_ADDR]], align 8
// CHECK-NEXT: notail call void (...) @llvm.fake.use(i64 [[FAKE_USE5]]) #[[ATTR3]]
// CHECK-NEXT: [[FAKE_USE6:%.*]] = load ptr, ptr [[H_ADDR]], align 8
// CHECK-NEXT: notail call void (...) @llvm.fake.use(ptr [[FAKE_USE6]]) #[[ATTR3]]
// CHECK-NEXT: [[FAKE_USE7:%.*]] = load ptr, ptr [[G_ADDR]], align 8
// CHECK-NEXT: notail call void (...) @llvm.fake.use(ptr [[FAKE_USE7]]) #[[ATTR3]]
// CHECK-NEXT: [[FAKE_USE8:%.*]] = load ptr, ptr [[F_ADDR]], align 8
// CHECK-NEXT: notail call void (...) @llvm.fake.use(ptr [[FAKE_USE8]]) #[[ATTR3]]
// CHECK-NEXT: [[FAKE_USE9:%.*]] = load ptr, ptr [[E_ADDR]], align 8
// CHECK-NEXT: notail call void (...) @llvm.fake.use(ptr [[FAKE_USE9]]) #[[ATTR3]]
// CHECK-NEXT: ret ptr undef
//
extern "C" const char *foo(int *e, const char *f, int *g, const int *h,
unsigned long i) {
[[clang::musttail]] return bar(e, f, g, h, i);
}
Loading