[llvm] [CGP]: Optimize mul.overflow. (PR #148343)

[Alexis Engelke via llvm-commits]

================
@@ -261,21 +261,33 @@ define i128 @u128_mul(i128 %x, i128 %y) {
 
 define { i128, i8 } @u128_checked_mul(i128 %x, i128 %y) {
 ; CHECK-LABEL: u128_checked_mul:
-; CHECK:       // %bb.0:
+; CHECK:       // %bb.0: // %overflow.entry
+; CHECK-NEXT:    orr x8, x1, x3
+; CHECK-NEXT:    cbz x8, .LBB17_2
+; CHECK-NEXT:  // %bb.1: // %overflow
 ; CHECK-NEXT:    mul x9, x3, x0
 ; CHECK-NEXT:    cmp x1, #0
 ; CHECK-NEXT:    ccmp x3, #0, #4, ne
-; CHECK-NEXT:    umulh x8, x1, x2
-; CHECK-NEXT:    umulh x10, x3, x0
+; CHECK-NEXT:    umulh x10, x1, x2
+; CHECK-NEXT:    umulh x8, x3, x0
 ; CHECK-NEXT:    madd x9, x1, x2, x9
-; CHECK-NEXT:    ccmp xzr, x8, #0, eq
-; CHECK-NEXT:    umulh x11, x0, x2
 ; CHECK-NEXT:    ccmp xzr, x10, #0, eq
+; CHECK-NEXT:    umulh x11, x0, x2
+; CHECK-NEXT:    ccmp xzr, x8, #0, eq
 ; CHECK-NEXT:    mul x0, x0, x2
 ; CHECK-NEXT:    cset w8, ne
 ; CHECK-NEXT:    adds x1, x11, x9
 ; CHECK-NEXT:    csinc w8, w8, wzr, lo
-; CHECK-NEXT:    eor w2, w8, #0x1
+; CHECK-NEXT:    b .LBB17_3
+; CHECK-NEXT:  .LBB17_2: // %overflow.no
+; CHECK-NEXT:    umulh x8, x0, x2
+; CHECK-NEXT:    madd x8, x0, x3, x8
----------------
aengelke wrote:

overflow.res is executed in both paths and therefore also on the fast path -- for which the instruction is unnecessary.

https://github.com/llvm/llvm-project/pull/148343