Skip to content

[KeyInstr][Clang] Atomic ops atoms #141624

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 1 commit into from
Jun 24, 2025
Merged

[KeyInstr][Clang] Atomic ops atoms #141624

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
17 changes: 13 additions & 4 deletions clang/lib/CodeGen/CGAtomic.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -411,7 +411,9 @@ static void emitAtomicCmpXchg(CodeGenFunction &CGF, AtomicExpr *E, bool IsWeak,

CGF.Builder.SetInsertPoint(StoreExpectedBB);
// Update the memory at Expected with Old's value.
CGF.Builder.CreateStore(Old, Val1);
auto *I = CGF.Builder.CreateStore(Old, Val1);
CGF.addInstToCurrentSourceAtom(I, Old);

// Finally, branch to the exit point.
CGF.Builder.CreateBr(ContinueBB);

Expand Down Expand Up @@ -591,7 +593,8 @@ static void EmitAtomicOp(CodeGenFunction &CGF, AtomicExpr *E, Address Dest,
Load->setAtomic(Order, Scope);
Load->setVolatile(E->isVolatile());
CGF.maybeAttachRangeForLoad(Load, E->getValueType(), E->getExprLoc());
CGF.Builder.CreateStore(Load, Dest);
auto *I = CGF.Builder.CreateStore(Load, Dest);
CGF.addInstToCurrentSourceAtom(I, Load);
return;
}

Expand All @@ -606,6 +609,7 @@ static void EmitAtomicOp(CodeGenFunction &CGF, AtomicExpr *E, Address Dest,
llvm::StoreInst *Store = CGF.Builder.CreateStore(LoadVal1, Ptr);
Store->setAtomic(Order, Scope);
Store->setVolatile(E->isVolatile());
CGF.addInstToCurrentSourceAtom(Store, LoadVal1);
return;
}

Expand Down Expand Up @@ -731,7 +735,8 @@ static void EmitAtomicOp(CodeGenFunction &CGF, AtomicExpr *E, Address Dest,
CGF.Builder.getInt8(1), Order, Scope, E);
RMWI->setVolatile(E->isVolatile());
llvm::Value *Result = CGF.Builder.CreateIsNotNull(RMWI, "tobool");
CGF.Builder.CreateStore(Result, Dest);
auto *I = CGF.Builder.CreateStore(Result, Dest);
CGF.addInstToCurrentSourceAtom(I, Result);
return;
}

Expand All @@ -740,6 +745,7 @@ static void EmitAtomicOp(CodeGenFunction &CGF, AtomicExpr *E, Address Dest,
CGF.Builder.CreateStore(CGF.Builder.getInt8(0), Ptr);
Store->setAtomic(Order, Scope);
Store->setVolatile(E->isVolatile());
CGF.addInstToCurrentSourceAtom(Store, nullptr);
return;
}
}
Expand All @@ -762,7 +768,8 @@ static void EmitAtomicOp(CodeGenFunction &CGF, AtomicExpr *E, Address Dest,
if (E->getOp() == AtomicExpr::AO__atomic_nand_fetch ||
E->getOp() == AtomicExpr::AO__scoped_atomic_nand_fetch)
Result = CGF.Builder.CreateNot(Result);
CGF.Builder.CreateStore(Result, Dest);
auto *I = CGF.Builder.CreateStore(Result, Dest);
CGF.addInstToCurrentSourceAtom(I, Result);
}

// This function emits any expression (scalar, complex, or aggregate)
Expand Down Expand Up @@ -845,6 +852,8 @@ static void EmitAtomicOp(CodeGenFunction &CGF, AtomicExpr *Expr, Address Dest,
}

RValue CodeGenFunction::EmitAtomicExpr(AtomicExpr *E) {
ApplyAtomGroup Grp(getDebugInfo());

QualType AtomicTy = E->getPtr()->getType()->getPointeeType();
QualType MemTy = AtomicTy;
if (const AtomicType *AT = AtomicTy->getAs<AtomicType>())
Expand Down
162 changes: 162 additions & 0 deletions clang/test/DebugInfo/KeyInstructions/atomic.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,162 @@
// RUN: %clang_cc1 -triple x86_64-linux-gnu -gkey-instructions -x c++ -std=c++17 %s -debug-info-kind=line-tables-only -emit-llvm -o - -gno-column-info \
// RUN: | FileCheck %s --implicit-check-not atomGroup --implicit-check-not atomRank
// RUN: %clang_cc1 -triple x86_64-linux-gnu -gkey-instructions -x c %s -debug-info-kind=line-tables-only -emit-llvm -o - -gno-column-info \
// RUN: | FileCheck %s --implicit-check-not atomGroup --implicit-check-not atomRank

// Check that atomic handling code gets Key Instruction annotations.

_Atomic(unsigned int) x;
unsigned int y;
void fun() {
unsigned int r1 = __c11_atomic_fetch_add(&x,- 1, __ATOMIC_RELAXED);
// CHECK: store i32 -1, ptr %.atomictmp, align 4, !dbg [[LINE11_G2R1:!.*]]
// CHECK-NEXT: %0 = load i32, ptr %.atomictmp, align 4, !dbg [[LINE11:!.*]]
// CHECK-NEXT: %1 = atomicrmw add ptr @x, i32 %0 monotonic, align 4, !dbg [[LINE11_G2R2:!.*]]
// CHECK-NEXT: store i32 %1, ptr %atomic-temp, align 4, !dbg [[LINE11_G2R1]]
Copy link
Member

Choose a reason for hiding this comment

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

So it's deliberate + desirable that both this and the earlier store of -1 are group-two-rank-one? (I suppose the buoy-based KIs will float the breakpoint upwards)

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Yeah it's deliberate. Possibly could be argued that the first store is "setup" and doesn't need to be key at all. I don't have a strong opinion either way, both should produce identical output if the instructions remain grouped like this (as you point out, due to the Buoys). The difference is if they split up over multiple blocks; the current impl has you stepping on the line twice, whereas alternative impls may have you step on this one.

Do you have a strong opinion?

Copy link
Member

Choose a reason for hiding this comment

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

I've no strong opinion; I suspect due to the stricter rules on atomic regions we wouldn't see this split over multiple blocks, but I wouldn't want to be quoted on that.

// CHECK-NEXT: %2 = load i32, ptr %atomic-temp, align 4, !dbg [[LINE11_G1R2:!.*]]
// CHECK-NEXT: store i32 %2, ptr %r1, align 4, !dbg [[LINE11_G1R1:!.*]]

unsigned int r2 = __c11_atomic_load(&x, __ATOMIC_RELAXED);
// CHECK-NEXT: %3 = load atomic i32, ptr @x monotonic, align 4, !dbg [[LINE19_G4R2:!.*]]
// CHECK-NEXT: store i32 %3, ptr %atomic-temp1, align 4, !dbg [[LINE19_G4R1:!.*]]
// CHECK-NEXT: %4 = load i32, ptr %atomic-temp1, align 4, !dbg [[LINE19_G3R2:!.*]]
// CHECK-NEXT: store i32 %4, ptr %r2, align 4, !dbg [[LINE19_G3R1:!.*]]

__c11_atomic_store(&x, 2, __ATOMIC_RELAXED);
// CHECK-NEXT: store i32 2, ptr %.atomictmp2, align 4, !dbg [[LINE25_G5R1:!.*]]
// CHECK-NEXT: %5 = load i32, ptr %.atomictmp2, align 4, !dbg [[LINE25_G5R2:!.*]]
// CHECK-NEXT: store atomic i32 %5, ptr @x monotonic, align 4, !dbg [[LINE25_G5R1:!.*]]

int r3 = __atomic_test_and_set(&x, __ATOMIC_RELAXED);
// CHECK-NEXT: %6 = atomicrmw xchg ptr @x, i8 1 monotonic, align 4, !dbg [[LINE30:!.*]]
// CHECK-NEXT: %tobool = icmp ne i8 %6, 0, !dbg [[LINE30_G7R2:!.*]]
// CHECK-NEXT: store i1 %tobool, ptr %atomic-temp3, align 1, !dbg [[LINE30_G7R1:!.*]]
// CHECK-NEXT: %7 = load i8, ptr %atomic-temp3, align 1, !dbg [[LINE30_G6R4:!.*]]
// CHECK-NEXT: %loadedv = trunc i8 %7 to i1, !dbg [[LINE30_G6R3:!.*]]
// CHECK-NEXT: %conv = zext i1 %loadedv to i32, !dbg [[LINE30_G6R2:!.*]]
// CHECK-NEXT: store i32 %conv, ptr %r3, align 4, !dbg [[LINE30_G6R1:!.*]]

__atomic_clear(&x, __ATOMIC_RELAXED);
// CHECK-NEXT: store atomic i8 0, ptr @x monotonic, align 4, !dbg [[LINE39_G8R1:!.*]]

int r4 = __c11_atomic_exchange(&x, 2,__ATOMIC_RELAXED);
// CHECK-NEXT: store i32 2, ptr %.atomictmp4, align 4, !dbg [[LINE42_G10R1:!.*]]
// CHECK-NEXT: %8 = load i32, ptr %.atomictmp4, align 4, !dbg [[LINE42:!.*]]
// CHECK-NEXT: %9 = atomicrmw xchg ptr @x, i32 %8 monotonic, align 4, !dbg [[LINE42_G10R2:!.*]]
// CHECK-NEXT: store i32 %9, ptr %atomic-temp5, align 4, !dbg [[LINE42_G10R1:!.*]]
// CHECK-NEXT: %10 = load i32, ptr %atomic-temp5, align 4, !dbg [[LINE42_G9R2:!.*]]
// CHECK-NEXT: store i32 %10, ptr %r4, align 4, !dbg [[LINE42_G9R1:!.*]]

int r5 = __atomic_compare_exchange(&y, &y, &y, 0, __ATOMIC_RELAXED, __ATOMIC_RELAXED);
// CHECK-NEXT: %11 = load i32, ptr @y, align 4, !dbg [[LINE50:!.*]]
// CHECK-NEXT: %12 = load i32, ptr @y, align 4, !dbg [[LINE50]]
// CHECK-NEXT: %13 = cmpxchg ptr @y, i32 %11, i32 %12 monotonic monotonic, align 4, !dbg [[LINE50]]
// CHECK-NEXT: %14 = extractvalue { i32, i1 } %13, 0, !dbg [[LINE50_G12R2:!.*]]
// CHECK-NEXT: %15 = extractvalue { i32, i1 } %13, 1, !dbg [[LINE50_G12R3:!.*]]
// CHECK-NEXT: br i1 %15, label %cmpxchg.continue, label %cmpxchg.store_expected, !dbg [[LINE50]]
// CHECK: cmpxchg.store_expected:
// CHECK-NEXT: store i32 %14, ptr @y, align 4, !dbg [[LINE50_G12R1:!.*]]
// CHECK-NEXT: br label %cmpxchg.continue, !dbg [[LINE50]]
// CHECK: cmpxchg.continue:
// CHECK-NEXT: %storedv = zext i1 %15 to i8, !dbg [[LINE50_G12R2]]
// CHECK-NEXT: store i8 %storedv, ptr %cmpxchg.bool, align 1, !dbg [[LINE50_G12R1]]
// CHECK-NEXT: %16 = load i8, ptr %cmpxchg.bool, align 1, !dbg [[LINE50_G11R4:!.*]]
// CHECK-NEXT: %loadedv6 = trunc i8 %16 to i1, !dbg [[LINE50_G11R3:!.*]]
// CHECK-NEXT: %conv7 = zext i1 %loadedv6 to i32, !dbg [[LINE50_G11R2:!.*]]
// CHECK-NEXT: store i32 %conv7, ptr %r5, align 4, !dbg [[LINE50_G11R1:!.*]]

int r6 = __c11_atomic_compare_exchange_strong(&x, &y, 42, __ATOMIC_RELAXED, __ATOMIC_RELAXED);
// CHECK-NEXT: store i32 42, ptr %.atomictmp8, align 4, !dbg [[LINE68_G14R1:!.*]]
// CHECK-NEXT: %17 = load i32, ptr @y, align 4, !dbg [[LINE68:!.*]]
// CHECK-NEXT: %18 = load i32, ptr %.atomictmp8, align 4, !dbg [[LINE68]]
// CHECK-NEXT: %19 = cmpxchg ptr @x, i32 %17, i32 %18 monotonic monotonic, align 4, !dbg [[LINE68]]
// CHECK-NEXT: %20 = extractvalue { i32, i1 } %19, 0, !dbg [[LINE68_G14R2:!.*]]
// CHECK-NEXT: %21 = extractvalue { i32, i1 } %19, 1, !dbg [[LINE68_G14R3:!.*]]
// CHECK-NEXT: br i1 %21, label %cmpxchg.continue11, label %cmpxchg.store_expected10, !dbg [[LINE68]]
// CHECK: cmpxchg.store_expected10:
// CHECK-NEXT: store i32 %20, ptr @y, align 4, !dbg [[LINE68_G14R1:!.*]]
// CHECK-NEXT: br label %cmpxchg.continue11, !dbg [[LINE68]]
// CHECK: cmpxchg.continue11:
// CHECK-NEXT: %storedv12 = zext i1 %21 to i8, !dbg [[LINE68_G14R2]]
// CHECK-NEXT: store i8 %storedv12, ptr %cmpxchg.bool9, align 1, !dbg [[LINE68_G14R1:!.*]]
// CHECK-NEXT: %22 = load i8, ptr %cmpxchg.bool9, align 1, !dbg [[LINE68_G13R4:!.*]]
// CHECK-NEXT: %loadedv13 = trunc i8 %22 to i1, !dbg [[LINE68_G13R3:!.*]]
// CHECK-NEXT: %conv14 = zext i1 %loadedv13 to i32, !dbg [[LINE68_G13R2:!.*]]
// CHECK-NEXT: store i32 %conv14, ptr %r6, align 4, !dbg [[LINE68_G13R1:!.*]]

int r7 = __c11_atomic_compare_exchange_weak(&x, &y, 43, __ATOMIC_RELAXED, __ATOMIC_RELAXED);
// CHECK-NEXT: store i32 43, ptr %.atomictmp15, align 4, !dbg [[LINE87_G16R1:!.*]]
// CHECK-NEXT: %23 = load i32, ptr @y, align 4, !dbg [[LINE87:!.*]]
// CHECK-NEXT: %24 = load i32, ptr %.atomictmp15, align 4, !dbg [[LINE87]]
// CHECK-NEXT: %25 = cmpxchg weak ptr @x, i32 %23, i32 %24 monotonic monotonic, align 4, !dbg [[LINE87]]
// CHECK-NEXT: %26 = extractvalue { i32, i1 } %25, 0, !dbg [[LINE87_G16R2:!.*]]
// CHECK-NEXT: %27 = extractvalue { i32, i1 } %25, 1, !dbg [[LINE87_G16R3:!.*]]
// CHECK-NEXT: br i1 %27, label %cmpxchg.continue18, label %cmpxchg.store_expected17, !dbg [[LINE87]]
// CHECK: cmpxchg.store_expected17:
// CHECK-NEXT: store i32 %26, ptr @y, align 4, !dbg [[LINE87_G16R1]]
// CHECK-NEXT: br label %cmpxchg.continue18, !dbg [[LINE87]]
// CHECK: cmpxchg.continue18:
// CHECK-NEXT: %storedv19 = zext i1 %27 to i8, !dbg [[LINE87_G16R2]]
// CHECK-NEXT: store i8 %storedv19, ptr %cmpxchg.bool16, align 1, !dbg [[LINE87_G16R1]]
// CHECK-NEXT: %28 = load i8, ptr %cmpxchg.bool16, align 1, !dbg [[LINE87_G15R4:!.*]]
// CHECK-NEXT: %loadedv20 = trunc i8 %28 to i1, !dbg [[LINE87_G15R3:!.*]]
// CHECK-NEXT: %conv21 = zext i1 %loadedv20 to i32, !dbg [[LINE87_G15R2:!.*]]
// CHECK-NEXT: store i32 %conv21, ptr %r7, align 4, !dbg [[LINE87_G15R1:!.*]]
}

// CHECK: [[LINE11_G2R1]] = !DILocation(line: 11, scope: ![[#]], atomGroup: 2, atomRank: 1)
// CHECK: [[LINE11]] = !DILocation(line: 11, scope: ![[#]])
// CHECK: [[LINE11_G2R2]] = !DILocation(line: 11, scope: ![[#]], atomGroup: 2, atomRank: 2)
// CHECK: [[LINE11_G1R2]] = !DILocation(line: 11, scope: ![[#]], atomGroup: 1, atomRank: 2)
// CHECK: [[LINE11_G1R1]] = !DILocation(line: 11, scope: ![[#]], atomGroup: 1, atomRank: 1)

// CHECK: [[LINE19_G4R2]] = !DILocation(line: 19, scope: ![[#]], atomGroup: 4, atomRank: 2)
// CHECK: [[LINE19_G4R1]] = !DILocation(line: 19, scope: ![[#]], atomGroup: 4, atomRank: 1)
// CHECK: [[LINE19_G3R2]] = !DILocation(line: 19, scope: ![[#]], atomGroup: 3, atomRank: 2)
// CHECK: [[LINE19_G3R1]] = !DILocation(line: 19, scope: ![[#]], atomGroup: 3, atomRank: 1)

// CHECK: [[LINE25_G5R1]] = !DILocation(line: 25, scope: ![[#]], atomGroup: 5, atomRank: 1)
// CHECK: [[LINE25_G5R2]] = !DILocation(line: 25, scope: ![[#]], atomGroup: 5, atomRank: 2)

// CHECK: [[LINE30]] = !DILocation(line: 30, scope: ![[#]])
// CHECK: [[LINE30_G7R2]] = !DILocation(line: 30, scope: ![[#]], atomGroup: 7, atomRank: 2)
// CHECK: [[LINE30_G7R1]] = !DILocation(line: 30, scope: ![[#]], atomGroup: 7, atomRank: 1)
// CHECK: [[LINE30_G6R4]] = !DILocation(line: 30, scope: ![[#]], atomGroup: 6, atomRank: 4)
// CHECK: [[LINE30_G6R3]] = !DILocation(line: 30, scope: ![[#]], atomGroup: 6, atomRank: 3)
// CHECK: [[LINE30_G6R2]] = !DILocation(line: 30, scope: ![[#]], atomGroup: 6, atomRank: 2)
// CHECK: [[LINE30_G6R1]] = !DILocation(line: 30, scope: ![[#]], atomGroup: 6, atomRank: 1)

// CHECK: [[LINE39_G8R1]] = !DILocation(line: 39, scope: ![[#]], atomGroup: 8, atomRank: 1)

// CHECK: [[LINE42_G10R1]] = !DILocation(line: 42, scope: ![[#]], atomGroup: 10, atomRank: 1)
// CHECK: [[LINE42]] = !DILocation(line: 42, scope: ![[#]])
// CHECK: [[LINE42_G10R2]] = !DILocation(line: 42, scope: ![[#]], atomGroup: 10, atomRank: 2)
// CHECK: [[LINE42_G9R2]] = !DILocation(line: 42, scope: ![[#]], atomGroup: 9, atomRank: 2)
// CHECK: [[LINE42_G9R1]] = !DILocation(line: 42, scope: ![[#]], atomGroup: 9, atomRank: 1)

// CHECK: [[LINE50]] = !DILocation(line: 50, scope: ![[#]])
// CHECK: [[LINE50_G12R2]] = !DILocation(line: 50, scope: ![[#]], atomGroup: 12, atomRank: 2)
// CHECK: [[LINE50_G12R3]] = !DILocation(line: 50, scope: ![[#]], atomGroup: 12, atomRank: 3)
// CHECK: [[LINE50_G12R1]] = !DILocation(line: 50, scope: ![[#]], atomGroup: 12, atomRank: 1)
// CHECK: [[LINE50_G11R4]] = !DILocation(line: 50, scope: ![[#]], atomGroup: 11, atomRank: 4)
// CHECK: [[LINE50_G11R3]] = !DILocation(line: 50, scope: ![[#]], atomGroup: 11, atomRank: 3)
// CHECK: [[LINE50_G11R2]] = !DILocation(line: 50, scope: ![[#]], atomGroup: 11, atomRank: 2)
// CHECK: [[LINE50_G11R1]] = !DILocation(line: 50, scope: ![[#]], atomGroup: 11, atomRank: 1)

// CHECK: [[LINE68_G14R1]] = !DILocation(line: 68, scope: ![[#]], atomGroup: 14, atomRank: 1)
// CHECK: [[LINE68]] = !DILocation(line: 68, scope: ![[#]])
// CHECK: [[LINE68_G14R2]] = !DILocation(line: 68, scope: ![[#]], atomGroup: 14, atomRank: 2)
// CHECK: [[LINE68_G14R3]] = !DILocation(line: 68, scope: ![[#]], atomGroup: 14, atomRank: 3)
// CHECK: [[LINE68_G13R4]] = !DILocation(line: 68, scope: ![[#]], atomGroup: 13, atomRank: 4)
// CHECK: [[LINE68_G13R3]] = !DILocation(line: 68, scope: ![[#]], atomGroup: 13, atomRank: 3)
// CHECK: [[LINE68_G13R2]] = !DILocation(line: 68, scope: ![[#]], atomGroup: 13, atomRank: 2)
// CHECK: [[LINE68_G13R1]] = !DILocation(line: 68, scope: ![[#]], atomGroup: 13, atomRank: 1)

// CHECK: [[LINE87_G16R1]] = !DILocation(line: 87, scope: ![[#]], atomGroup: 16, atomRank: 1)
// CHECK: [[LINE87]] = !DILocation(line: 87, scope: ![[#]])
// CHECK: [[LINE87_G16R2]] = !DILocation(line: 87, scope: ![[#]], atomGroup: 16, atomRank: 2)
// CHECK: [[LINE87_G16R3]] = !DILocation(line: 87, scope: ![[#]], atomGroup: 16, atomRank: 3)
// CHECK: [[LINE87_G15R4]] = !DILocation(line: 87, scope: ![[#]], atomGroup: 15, atomRank: 4)
// CHECK: [[LINE87_G15R3]] = !DILocation(line: 87, scope: ![[#]], atomGroup: 15, atomRank: 3)
// CHECK: [[LINE87_G15R2]] = !DILocation(line: 87, scope: ![[#]], atomGroup: 15, atomRank: 2)
// CHECK: [[LINE87_G15R1]] = !DILocation(line: 87, scope: ![[#]], atomGroup: 15, atomRank: 1)