[llvm] [InstCombine] Support trunc to i1 in foldSelectICmpAnd (PR #127905)

[Yingwei Zheng via llvm-commits]

================
@@ -903,8 +902,9 @@ define i8 @neg_select_trunc_bittest_to_shl_extra_use(i8 %x) {
 
 define i16 @select_trunc_nuw_bittest_or(i8 %x) {
 ; CHECK-LABEL: @select_trunc_nuw_bittest_or(
-; CHECK-NEXT:    [[TMP1:%.*]] = trunc nuw i8 [[X:%.*]] to i1
-; CHECK-NEXT:    [[RES:%.*]] = select i1 [[TMP1]], i16 20, i16 4
+; CHECK-NEXT:    [[TMP1:%.*]] = zext i8 [[X:%.*]] to i16
+; CHECK-NEXT:    [[SELECT:%.*]] = shl nuw nsw i16 [[TMP1]], 4
+; CHECK-NEXT:    [[RES:%.*]] = or disjoint i16 [[SELECT]], 4
----------------
dtcxzyw wrote:

It is not a regression. The codegen without select is generally better: https://godbolt.org/z/1W6ocqdT3


https://github.com/llvm/llvm-project/pull/127905