Skip to content

[ARM] Fix musttail calls #109943

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

Closed
wants to merge 10 commits into from
Closed

[ARM] Fix musttail calls #109943

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
9b26468
[ARM] Re-generate a test
ostannard May 9, 2024
07a5a3d
[ARM] Fix comment typo
ostannard Sep 24, 2024
7773f66
[ARM] Add debug trace for tail-call optimisation
ostannard May 9, 2024
4ec65db
[ARM] Tail-calls do not require caller and callee arguments to match
ostannard May 9, 2024
4ef5bf0
[ARM] Allow functions with sret returns to be tail-called
ostannard Sep 24, 2024
4179471
[ARM] Allow tail calls with byval args
ostannard Sep 24, 2024
898c17f
[Clang] Always forward sret parameters to musttail calls
kiran-isaac Aug 19, 2024
e5ae451
[ARM] Avoid clobbering byval arguments when passing to tail-calls
ostannard Sep 26, 2024
12ba763
[ARM] Optimise byval arguments in tail-calls
ostannard Sep 26, 2024
3900ce2
fixup! [ARM] Avoid clobbering byval arguments when passing to tail-calls
ostannard Oct 8, 2024
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
6 changes: 5 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 @@ -5112,7 +5112,11 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
RawAddress SRetAlloca = RawAddress::invalid();
llvm::Value *UnusedReturnSizePtr = nullptr;
if (RetAI.isIndirect() || RetAI.isInAlloca() || RetAI.isCoerceAndExpand()) {
if (IsVirtualFunctionPointerThunk && RetAI.isIndirect()) {
// For virtual function pointer thunks and musttail calls, we must always
// forward an incoming SRet pointer to the callee, because a local alloca
// would be de-allocated before the call. These cases both guarantee that
// there will be an incoming SRet argument of the correct type.
if ((IsVirtualFunctionPointerThunk || IsMustTail) && RetAI.isIndirect()) {
SRetPtr = makeNaturalAddressForPointer(CurFn->arg_begin() +
IRFunctionArgs.getSRetArgNo(),
RetTy, CharUnits::fromQuantity(1));
Expand Down
84 changes: 84 additions & 0 deletions clang/test/CodeGen/musttail-sret.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,84 @@
// RUN: %clang_cc1 -triple=arm %s -emit-llvm -O3 -o - | FileCheck %s --check-prefix=CHECK-ARM
// RUN: %clang_cc1 -triple=arm64 %s -emit-llvm -O3 -o - | FileCheck %s --check-prefix=CHECK-ARM64
// RUN: %clang_cc1 -triple=i686 %s -emit-llvm -O3 -o - | FileCheck %s --check-prefix=CHECK-X86
// RUN: %clang_cc1 -triple=x86_64 %s -emit-llvm -O3 -o - | FileCheck %s --check-prefix=CHECK-X64

// Sret tests
struct Big {
int a, b, c, d, e, f, g, h;
};

struct Big F1(signed short P0);

struct Big F2(signed short P0) {
signed short P1 = 20391;
[[clang::musttail]] return F1(P1);
}

// CHECK-NOT: alloca
// CHECK-ARM: musttail call arm_aapcscc void @_Z2F1s(ptr dead_on_unwind writable sret(%struct.Big) align 4 %agg.result, i16 noundef signext 20391)
// CHECK-ARM64: musttail call void @_Z2F1s(ptr dead_on_unwind writable sret(%struct.Big) align 4 %agg.result, i16 noundef 20391)
// CHECK-X86: musttail call void @_Z2F1s(ptr dead_on_unwind writable sret(%struct.Big) align 4 %agg.result, i16 noundef signext 20391)
// CHECK-X64: musttail call void @_Z2F1s(ptr dead_on_unwind writable sret(%struct.Big) align 4 %agg.result, i16 noundef signext 20391)

struct ReallyBig {
int a[100];
};

// Indirect sret tests
// Function pointer for testing indirect musttail call.
struct FunctionPointers {
ReallyBig (*F3)(int, int, int, int, float, double);
ReallyBig (*F4)(int, int, int, char, float, double);
};

struct ReallyBig F3(int P0, int P1, int P2, int P3, float P4, double P5);
struct ReallyBig F4(int P0, int P1, int P2, char P3, float P4, double P5);

static struct FunctionPointers FP = {F3, F4};

struct ReallyBig F5 (int P0, int P1, int P2, int P3, float P4, double P5) {
[[clang::musttail]] return FP.F3(P0, P1, P2, P3, P4, P5);
}

// CHECK-NOT: alloca
// CHECK-ARM: musttail call arm_aapcscc void @_Z2F3iiiifd(ptr dead_on_unwind writable sret(%struct.ReallyBig) align 4 %agg.result, i32 noundef %P0, i32 noundef %P1, i32 noundef %P2, i32 noundef %P3, float noundef %P4, double noundef %P5)
// CHECK-ARM64: musttail call void @_Z2F3iiiifd(ptr dead_on_unwind writable sret(%struct.ReallyBig) align 4 %agg.result, i32 noundef %P0, i32 noundef %P1, i32 noundef %P2, i32 noundef %P3, float noundef %P4, double noundef %P5)
// CHECK-X86: musttail call void @_Z2F3iiiifd(ptr dead_on_unwind writable sret(%struct.ReallyBig) align 4 %agg.result, i32 noundef %P0, i32 noundef %P1, i32 noundef %P2, i32 noundef %P3, float noundef %P4, double noundef %P5)
// CHECK-X64: musttail call void @_Z2F3iiiifd(ptr dead_on_unwind writable sret(%struct.ReallyBig) align 4 %agg.result, i32 noundef %P0, i32 noundef %P1, i32 noundef %P2, i32 noundef %P3, float noundef %P4, double noundef %P5)

struct ReallyBig F6 (int P0, int P1, int P2, char P3, float P4, double P5) {
[[clang::musttail]] return FP.F4(P0, P1, P2, P3, P4, P5);
}

// Complex and BitInt. Special cases for sret.
// CHECK-NOT: alloca
// CHECK-ARM: musttail call arm_aapcscc void @_Z2F4iiicfd(ptr dead_on_unwind writable sret(%struct.ReallyBig) align 4 %agg.result, i32 noundef %P0, i32 noundef %P1, i32 noundef %P2, i8 noundef signext %P3, float noundef %P4, double noundef %P5)
// CHECK-ARM64: musttail call void @_Z2F4iiicfd(ptr dead_on_unwind writable sret(%struct.ReallyBig) align 4 %agg.result, i32 noundef %P0, i32 noundef %P1, i32 noundef %P2, i8 noundef %P3, float noundef %P4, double noundef %P5)
// CHECK-X86: musttail call void @_Z2F4iiicfd(ptr dead_on_unwind writable sret(%struct.ReallyBig) align 4 %agg.result, i32 noundef %P0, i32 noundef %P1, i32 noundef %P2, i8 noundef signext %P3, float noundef %P4, double noundef %P5)
// CHECK-X64: musttail call void @_Z2F4iiicfd(ptr dead_on_unwind writable sret(%struct.ReallyBig) align 4 %agg.result, i32 noundef %P0, i32 noundef %P1, i32 noundef %P2, i8 noundef signext %P3, float noundef %P4, double noundef %P5)

double _Complex F7(signed short P0);

double _Complex F8(signed short P0) {
signed short P1 = 20391;
[[clang::musttail]] return F7(P1);
}

// CHECK-NOT: alloca
// CHECK-ARM: musttail call arm_aapcscc void @_Z2F7s(ptr dead_on_unwind writable sret({ double, double }) align 8 %agg.result, i16 noundef signext 20391)
// CHECK-ARM64: musttail call noundef { double, double } @_Z2F7s(i16 noundef 20391)
// CHECK-X86: musttail call void @_Z2F7s(ptr dead_on_unwind writable sret({ double, double }) align 4 %agg.result, i16 noundef signext 20391)
// CHECK-X64: musttail call noundef { double, double } @_Z2F7s(i16 noundef signext 20391)

signed _BitInt(100) F9(float P0, float P1, double P2, char P3);

signed _BitInt(100) F10(float P0, float P1, double P2, char P3) {
[[clang::musttail]] return F9(P0, P1, P2, P3);
}

// CHECK-NOT: alloca
// CHECK-ARM: musttail call arm_aapcscc void @_Z2F9ffdc(ptr dead_on_unwind writable sret(i128) align 8 %agg.result, float noundef %P0, float noundef %P1, double noundef %P2, i8 noundef signext %P3)
// CHECK-ARM64: musttail call noundef i100 @_Z2F9ffdc(float noundef %P0, float noundef %P1, double noundef %P2, i8 noundef %P3)
// CHECK-X86: musttail call void @_Z2F9ffdc(ptr dead_on_unwind writable sret(i128) align 4 %agg.result, float noundef %P0, float noundef %P1, double noundef %P2, i8 noundef signext %P3)
// CHECK-X64: musttail call noundef { i64, i64 } @_Z2F9ffdc(float noundef %P0, float noundef %P1, double noundef %P2, i8 noundef signext %P3)
Loading