[clang] fix emitvaarg when struct is null

[Jolyon0202]

Fix bug when emit null struct with attribute aligned(16) and ICE of debugbuild.

[llvmbot]

@llvm/pr-subscribers-backend-aarch64
@llvm/pr-subscribers-clang

@llvm/pr-subscribers-clang-codegen

Author: Jolyon (Jolyon0202)

Changes

Fix bug when emit null struct with attribute aligned(16) and ICE of debugbuild.

---

Full diff: https://github.com/llvm/llvm-project/pull/72624.diff

1 Files Affected:

• (modified) clang/lib/CodeGen/Targets/AArch64.cpp (+2-2)

diff --git a/clang/lib/CodeGen/Targets/AArch64.cpp b/clang/lib/CodeGen/Targets/AArch64.cpp
index be5145daa00b7f5..49dff102b7d554a 100644
--- a/clang/lib/CodeGen/Targets/AArch64.cpp
+++ b/clang/lib/CodeGen/Targets/AArch64.cpp
@@ -526,7 +526,7 @@ Address AArch64ABIInfo::EmitAAPCSVAArg(Address VAListAddr, QualType Ty,
   llvm::Type *BaseTy = CGF.ConvertType(Ty);
   if (IsIndirect)
     BaseTy = llvm::PointerType::getUnqual(BaseTy);
-  else if (AI.getCoerceToType())
+  else if (AI.canHaveCoerceToType() && AI.getCoerceToType())
     BaseTy = AI.getCoerceToType();
 
   unsigned NumRegs = 1;
@@ -594,7 +594,7 @@ Address AArch64ABIInfo::EmitAAPCSVAArg(Address VAListAddr, QualType Ty,
   // Integer arguments may need to correct register alignment (for example a
   // "struct { __int128 a; };" gets passed in x_2N, x_{2N+1}). In this case we
   // align __gr_offs to calculate the potential address.
-  if (!IsFPR && !IsIndirect && TyAlign.getQuantity() > 8) {
+  if (!IsFPR && !IsIndirect && TyAlign.getQuantity() > 8 && !TySize.isZero()) {
     int Align = TyAlign.getQuantity();
 
     reg_offs = CGF.Builder.CreateAdd(

[Jolyon0202]

demo：

#include <stdio.h>
extern void *memset (void *__s, int __c, size_t __n);
typedef __builtin_va_list va_list;

struct S94 {
  struct __attribute__((aligned(16))) {
  } a;
};

struct S94 s94;

void check94va(int z, ...) {
  va_list ap;
  __builtin_va_start(ap, z);
  struct S94 arg1 =  __builtin_va_arg(ap, struct S94);
  long long int tmp = __builtin_va_arg(ap, long long);
  printf("result = %lld\n", tmp);
  if (tmp != 2LL) {
    printf("Fails!!!!\n");
  }
  __builtin_va_end(ap);
}

int main(void) {
  memset(&s94, '\0', sizeof(s94));
  printf("sizeof(s94) = %ld\n", sizeof(s94));
  check94va(1, s94, 2LL);
  return 0;
}

found two bug need to fix:
1 If the alignment of S94(null struct) is less than 16, the result is correct.
2 ICE of AI.getCoerceToType() in debugbuild.(AI can not getcoerceTotype when it's null)

[Jolyon0202]

@efriedma-quic

[efriedma-quic]

Please fix the pull request description to say this is aarch64-specific.

Missing regression test in clang/test/CodeGen/

How does this interact with #72197?