[llvm] ecad53c - [InstCombine] Don't fold `uadd.sat` to `or` if it increase instruction count

[Noah Goldstein via llvm-commits]

Author: Noah Goldstein
Date: 2023-04-29T12:38:41-05:00
New Revision: ecad53c3f442df32149967a49803d22d98b2d2c7

URL: https://github.com/llvm/llvm-project/commit/ecad53c3f442df32149967a49803d22d98b2d2c7
DIFF: https://github.com/llvm/llvm-project/commit/ecad53c3f442df32149967a49803d22d98b2d2c7.diff

LOG: [InstCombine] Don't fold `uadd.sat` to `or` if it increase instruction count

In the `(cmp eq/ne (uadd.sat X, Y),0)` case, we where missing a
`hasOneUse` check.

Differential Revision: https://reviews.llvm.org/D149425

Added: 
    

Modified: 
    llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
    llvm/test/Transforms/InstCombine/cmp-intrinsic.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
index bd3000ada1f38..aeb1c53c555fe 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -3386,7 +3386,7 @@ Instruction *InstCombinerImpl::foldICmpEqIntrinsicWithConstant(
 
   case Intrinsic::uadd_sat: {
     // uadd.sat(a, b) == 0  ->  (a | b) == 0
-    if (C.isZero()) {
+    if (C.isZero() && II->hasOneUse()) {
       Value *Or = Builder.CreateOr(II->getArgOperand(0), II->getArgOperand(1));
       return new ICmpInst(Pred, Or, Constant::getNullValue(Ty));
     }

diff  --git a/llvm/test/Transforms/InstCombine/cmp-intrinsic.ll b/llvm/test/Transforms/InstCombine/cmp-intrinsic.ll
index 529f107fafffc..078158729ef1b 100644
--- a/llvm/test/Transforms/InstCombine/cmp-intrinsic.ll
+++ b/llvm/test/Transforms/InstCombine/cmp-intrinsic.ll
@@ -872,8 +872,7 @@ define i1 @uadd_sat_ne_zero_fail_multiuse(i8 %x, i8 %y) {
 ; CHECK-LABEL: @uadd_sat_ne_zero_fail_multiuse(
 ; CHECK-NEXT:    [[M:%.*]] = call i8 @llvm.uadd.sat.i8(i8 [[X:%.*]], i8 [[Y:%.*]])
 ; CHECK-NEXT:    call void @use8(i8 [[M]])
-; CHECK-NEXT:    [[TMP1:%.*]] = or i8 [[X]], [[Y]]
-; CHECK-NEXT:    [[R:%.*]] = icmp ne i8 [[TMP1]], 0
+; CHECK-NEXT:    [[R:%.*]] = icmp ne i8 [[M]], 0
 ; CHECK-NEXT:    ret i1 [[R]]
 ;
   %m = call i8 @llvm.uadd.sat.i8(i8 %x, i8 %y)