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

[Amara Emerson via llvm-commits]

================
@@ -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)))) &&
----------------
aemerson wrote:

I tried to implement this. It didn't turn out as well as I'd hoped, memset is ok to do and we the logic for memset. However for memmove we have a user in XCore which is computing a tail-call-position using a DAG node without any reference to a `CallInst` so we can't do the same thing there. From what I could tell memcpy usages seemed ok to convert too but I didn't check all of them since there were so many.

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