[clang] [llvm] [PowerPC] Diagnose musttail instead of crash inside backend (PR #93267)

Chen Zheng via llvm-commits llvm-commits at lists.llvm.org
Wed May 29 20:07:09 PDT 2024


================
@@ -743,6 +743,20 @@ bool Sema::checkMustTailAttr(const Stmt *St, const Attr &MTA) {
     CallerType.Func = CallerDecl->getType()->getAs<FunctionProtoType>();
   }
 
+  if (Context.getTargetInfo().getTriple().isPPC()) {
+    if (Context.getTargetInfo().getTriple().isOSAIX())
+      return Diag(St->getBeginLoc(), diag::err_aix_musttail_unsupported);
+    else if (!Context.getTargetInfo().hasFeature("pcrelative-memops")) {
+      if (Context.getTargetInfo().hasFeature("longcall"))
+        return Diag(St->getBeginLoc(), diag::err_ppc_impossible_musttail) << 0;
+      else if (!CE->getDirectCallee())
+        return Diag(St->getBeginLoc(), diag::err_ppc_impossible_musttail) << 1;
+      else if (isa_and_nonnull<FunctionDecl>(CE->getCalleeDecl()) &&
----------------
chenzheng1030 wrote:

Thanks for comments. @efriedma-quic 

I tried below two cases, 
```
$ cat 1.c
int func2(int i);
int external_call2(int i) {
  [[clang::musttail]] return func2(i);
}

int func2(int i) {
  return 0;
}
$ cat 2.c

__attribute__((weak)) int func2(int i) {
  return 0;
}
int external_call2(int i) {
  [[clang::musttail]] return func2(i);
}
```

- `1.c` is good to compile without this patch, but now it is mistakenly reported as invalid to do tail call because the callee is treated as undefined...
- `2.c` can not be detected by this patch.

Will try to fix this. If you have any idea to fix these two issues in your mind, please let me know!

https://github.com/llvm/llvm-project/pull/93267


More information about the llvm-commits mailing list