jmciver wrote:
@dtcxzyw I was not aware I should be using a fuzzer. What tool would you
recommend?
>From a correctness standpoint the match is relatively narrow and I have
>discussed with @nlopes extensively. The patch has been tested with bootstrap
>build of LLVM running all regressions. Add
@@ -4813,15 +4813,22 @@ Instruction *InstCombinerImpl::visitFreeze(FreezeInst
&I) {
// TODO: This could use getBinopAbsorber() / getBinopIdentity() to avoid
// duplicating logic for binops at least.
auto getUndefReplacement = [&I](Type *Ty) {
-Constant *BestVal
@@ -3124,6 +3124,19 @@ inline auto m_c_LogicalOp(const LHS &L, const RHS &R) {
return m_LogicalOp(L, R);
}
+struct GuaranteedNotToBeUndefOrPoison_match {
+ template bool match(ITy *V) {
+if (auto *AsValue = dyn_cast(V))
+ return isGuaranteedNotToBeUndefOrPoison(As
jmciver wrote:
@nikic and @dtcxzyw thanks for the feedback.
This patch does bypasses the need for freeze poison -> null canonicalization
removal in InstCombine. To provide context I am seeing the lack of store of
select freeze poison folding in uninitialized memory semantics work that I am
do
https://github.com/jmciver ready_for_review
https://github.com/llvm/llvm-project/pull/129776
___
llvm-branch-commits mailing list
llvm-branch-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
