Fix #104794: Fixed sret bugs with clang::musttail on 32 bit targets

[kiran-isaac]

Fix #104794

This fix adds test cases for clang::musttail, the final two of which would fail without my small change to CGCall.cpp

[github-actions]

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be
notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write
permissions for the repository. In which case you can instead tag reviewers by
name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review
by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate
is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

[llvmbot]

@llvm/pr-subscribers-clang-codegen

@llvm/pr-subscribers-clang

Author: Kiran (kiran-isaac)

Changes

Fix #104794

This fix adds test cases for clang::musttail, the final two of which would fail without my small change to CGCall.cpp

---

Full diff: https://github.com/llvm/llvm-project/pull/104795.diff

2 Files Affected:

• (modified) clang/lib/CodeGen/CGCall.cpp (+1-1)
• (added) clang/test/CodeGen/musttail-sret.cpp (+84)

diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
index e4f221ae55eefa..47150c101a300b 100644
--- a/clang/lib/CodeGen/CGCall.cpp
+++ b/clang/lib/CodeGen/CGCall.cpp
@@ -5085,7 +5085,7 @@ 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()) {
+    if ((IsVirtualFunctionPointerThunk && RetAI.isIndirect()) || IsMustTail) {
       SRetPtr = makeNaturalAddressForPointer(CurFn->arg_begin() +
                                                  IRFunctionArgs.getSRetArgNo(),
                                              RetTy, CharUnits::fromQuantity(1));
diff --git a/clang/test/CodeGen/musttail-sret.cpp b/clang/test/CodeGen/musttail-sret.cpp
new file mode 100644
index 00000000000000..ca67c218cd67f6
--- /dev/null
+++ b/clang/test/CodeGen/musttail-sret.cpp
@@ -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)
\ No newline at end of file

[efriedma-quic]

Do any of the tests cover the case of RetAI.isInAlloca() or RetAI.isCoerceAndExpand()?

[efriedma-quic]

Waiting for an answer here.

[kiran-isaac]

Apologies for slow replies on both of my PRs, our fuzzer and some of our downstream testing identified some issues and I had to verify they were unrelated. No none of the tests cover these cases. I am in the process of trying to add some that do

[kiran-isaac]

tagged wrong PR in revert commit, see #102896

[ostannard]

Continued in #109943.