[llvm] [LV] Header masked condition of FindLastSelect when optimizeFindIVReductions when folding tail. (PR #191166)

[Florian Hahn via llvm-commits]

================
@@ -1213,11 +1216,14 @@ define i64 @findlast_expr_flipped_select_anyof(ptr %a, ptr %b, i64 %rdx.start, i
 ; TF-NEXT:    [[TMP7:%.*]] = getelementptr i64, ptr [[TMP6]], i64 -3
 ; TF-NEXT:    [[WIDE_MASKED_LOAD8:%.*]] = call <4 x i64> @llvm.masked.load.v4i64.p0(ptr align 8 [[TMP7]], <4 x i1> [[REVERSE]], <4 x i64> poison)
 ; TF-NEXT:    [[REVERSE9:%.*]] = shufflevector <4 x i64> [[WIDE_MASKED_LOAD8]], <4 x i64> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; TF-NEXT:    [[TMP8:%.*]] = icmp sle <4 x i64> [[REVERSE6]], [[REVERSE9]]
+; TF-NEXT:    [[TMP16:%.*]] = icmp sgt <4 x i64> [[REVERSE6]], [[REVERSE9]]
----------------
fhahn wrote:

Just to note that when the operands here, we apply the header mask to the wrong condition I think

https://github.com/llvm/llvm-project/pull/191166