[llvm] [DAGCombiner] Fold smax(X, -1)/smin(X, 0) to bitwise shift forms (PR #206242)

[via llvm-commits]

================
@@ -319,8 +319,9 @@ define i32 @or_neg_no_smin_but_zero(i32 %x, i32 %y) {
 ; CHECK-LABEL: or_neg_no_smin_but_zero:
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    bic w8, w0, w0, asr #31
-; CHECK-NEXT:    cmn w1, w8
-; CHECK-NEXT:    cset w0, lt
+; CHECK-NEXT:    neg w8, w8
----------------
Andarwinux wrote:

This looks like a regression?

https://github.com/llvm/llvm-project/pull/206242