[llvm] [AArch64] Don't tail call memset if it would convert to a bzero. (PR #98969)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-aarch64

Author: Amara Emerson (aemerson)

<details>
<summary>Changes</summary>

Well, not quite that simple. We can tc memset since it returns the first
argument but bzero doesn't do that and therefore we can end up miscompiling.

rdar://131419786


---
Full diff: https://github.com/llvm/llvm-project/pull/98969.diff


2 Files Affected:

- (modified) llvm/lib/CodeGen/Analysis.cpp (+6-1) 
- (added) llvm/test/CodeGen/AArch64/no-tail-call-bzero-from-memset.ll (+20) 


``````````diff
diff --git a/llvm/lib/CodeGen/Analysis.cpp b/llvm/lib/CodeGen/Analysis.cpp
index 7fc18639e5852..2a3015866da5f 100644
--- a/llvm/lib/CodeGen/Analysis.cpp
+++ b/llvm/lib/CodeGen/Analysis.cpp
@@ -677,6 +677,8 @@ bool llvm::returnTypeIsEligibleForTailCall(const Function *F,
   // will be expanded as memcpy in libc, which returns the first
   // argument. On other platforms like arm-none-eabi, memcpy may be
   // expanded as library call without return value, like __aeabi_memcpy.
+  // Similarly, llvm.memset can be expanded to bzero, which doesn't have a
+  // return value either.
   const CallInst *Call = cast<CallInst>(I);
   if (Function *F = Call->getCalledFunction()) {
     Intrinsic::ID IID = F->getIntrinsicID();
@@ -685,7 +687,10 @@ bool llvm::returnTypeIsEligibleForTailCall(const Function *F,
          (IID == Intrinsic::memmove &&
           TLI.getLibcallName(RTLIB::MEMMOVE) == StringRef("memmove")) ||
          (IID == Intrinsic::memset &&
-          TLI.getLibcallName(RTLIB::MEMSET) == StringRef("memset"))) &&
+          TLI.getLibcallName(RTLIB::MEMSET) == StringRef("memset") &&
+          (!isa<ConstantInt>(Call->getOperand(1)) ||
+           !cast<ConstantInt>(Call->getOperand(1))->isZero() ||
+           !TLI.getLibcallName(RTLIB::BZERO)))) &&
         (RetVal == Call->getArgOperand(0) ||
          isPointerBitcastEqualTo(RetVal, Call->getArgOperand(0))))
       return true;
diff --git a/llvm/test/CodeGen/AArch64/no-tail-call-bzero-from-memset.ll b/llvm/test/CodeGen/AArch64/no-tail-call-bzero-from-memset.ll
new file mode 100644
index 0000000000000..90e641cd4fe3d
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/no-tail-call-bzero-from-memset.ll
@@ -0,0 +1,20 @@
+; RUN: llc -o - %s | FileCheck %s
+; RUN: llc -global-isel -o - %s | FileCheck %s
+target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
+target triple = "arm64-apple-macosx15.0.0"
+
+define ptr @test()  {
+; CHECK-LABEL: test:
+; CHECK-NOT: b _bzero
+  %1 = tail call ptr @fn(i32 noundef 1) #3
+  tail call void @llvm.memset.p0.i64(ptr noundef nonnull align 1 dereferenceable(1000) %1, i8 noundef 0, i64 noundef 1000, i1 noundef false) #3
+  ret ptr %1
+}
+
+declare ptr @fn(i32 noundef)
+
+; Function Attrs: nocallback nofree nounwind willreturn memory(argmem: write)
+declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i1 immarg) #2
+
+attributes #2 = { nocallback nofree nounwind willreturn memory(argmem: write) }
+attributes #3 = { nounwind optsize }

``````````

</details>


https://github.com/llvm/llvm-project/pull/98969