[llvm] [VPlan] Directly use masks on recipes in dropPoisonGen (PR #193978)
Ramkumar Ramachandra via llvm-commits
llvm-commits at lists.llvm.org
Mon Apr 27 00:18:04 PDT 2026
================
@@ -27,16 +27,16 @@ define void @vector_reverse_i32(ptr noalias %A, ptr noalias %B) {
; RV64-NEXT: [[AVL:%.*]] = phi i64 [ 1023, %[[VECTOR_PH]] ], [ [[AVL_NEXT:%.*]], %[[VECTOR_BODY]] ]
; RV64-NEXT: [[TMP19:%.*]] = call i32 @llvm.experimental.get.vector.length.i64(i64 [[AVL]], i32 4, i1 true)
; RV64-NEXT: [[OFFSET_IDX:%.*]] = sub i64 1023, [[INDEX]]
-; RV64-NEXT: [[TMP7:%.*]] = add nsw i64 [[OFFSET_IDX]], -1
-; RV64-NEXT: [[TMP8:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[TMP7]]
+; RV64-NEXT: [[TMP1:%.*]] = add i64 [[OFFSET_IDX]], -1
+; RV64-NEXT: [[TMP8:%.*]] = getelementptr i32, ptr [[B]], i64 [[TMP1]]
----------------
artagnon wrote:
Yeah, I'm not sure what the cause of this is, as I haven't worked with header masks in the past. I've been meaning to ask: why isn't guarding the transform with CM.foldTailByMasking() equivalent?
https://github.com/llvm/llvm-project/pull/193978
More information about the llvm-commits
mailing list