[llvm] [Asan][RISCV] Enhance getTgtMemIntrinsic() to allow Asan instrument t… (PR #135198)

[Hank Chang via llvm-commits]

================
@@ -494,8 +494,9 @@ unsigned GCNTTIImpl::getMaxInterleaveFactor(ElementCount VF) {
   return 8;
 }
 
-bool GCNTTIImpl::getTgtMemIntrinsic(IntrinsicInst *Inst,
-                                       MemIntrinsicInfo &Info) const {
+bool GCNTTIImpl::getTgtMemIntrinsic(
+    IntrinsicInst *Inst, MemIntrinsicInfo &Info,
+    SmallVectorImpl<InterestingMemoryOperand> *Interesting) const {
----------------
HankChang736 wrote:

Thank you for the feedback, after few days of thinking. I'd like to rewrite this patch since this `getTgtMemIntrinsic()` return `true` means it fill MemIntrinsicInfo. However, I break this rule and it might cause further error in `EarlyCSE` where `getTgtMemIntrinsic` is used if we fill InterestingMemoryOperand. I will try to find a way to combine these, maybe change the return value to optional value or another better way.

https://github.com/llvm/llvm-project/pull/135198