[InstCombine] enhance fold for copysign with known sign arg

This is another optimization suggested in PRPR44153: https://bugs.llvm.org/show_bug.cgi?id=44153
llvm · Dec 22, 2019 · 9cdcd81 · 9cdcd81
1 parent dc5b614
commit 9cdcd81
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 14 deletions.
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp b/llvm/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -2286,18 +2286,22 @@ Instruction *InstCombiner::visitCallInst(CallInst &CI) {
     break;
   }
   case Intrinsic::copysign: {
-    const APFloat *C;
-    if (match(II->getArgOperand(1), m_APFloat(C))) {
-      // If we know the sign bit of the sign argument, reduce to FABS/FNABS:
-      // copysign X, PosC --> fabs X
-      // copysign X, NegC --> fneg (fabs X)
+    if (SignBitMustBeZero(II->getArgOperand(1), &TLI)) {
+      // If we know that the sign argument is positive, reduce to FABS:
+      // copysign X, Pos --> fabs X
       Value *Fabs = Builder.CreateUnaryIntrinsic(Intrinsic::fabs,
                                                  II->getArgOperand(0), II);
-      if (C->isNegative())
-        Fabs = Builder.CreateFNegFMF(Fabs, II);
-
       return replaceInstUsesWith(*II, Fabs);
     }
+    // TODO: There should be a ValueTracking sibling like SignBitMustBeOne.
+    const APFloat *C;
+    if (match(II->getArgOperand(1), m_APFloat(C)) && C->isNegative()) {
+      // If we know that the sign argument is negative, reduce to FNABS:
+      // copysign X, Neg --> fneg (fabs X)
+      Value *Fabs = Builder.CreateUnaryIntrinsic(Intrinsic::fabs,
+                                                 II->getArgOperand(0), II);
+      return replaceInstUsesWith(*II, Builder.CreateFNegFMF(Fabs, II));
+    }
     break;
   }
   case Intrinsic::fabs: {

diff --git a/llvm/test/Transforms/InstCombine/copysign.ll b/llvm/test/Transforms/InstCombine/copysign.ll
@@ -45,9 +45,8 @@ define <3 x double> @negative_sign_arg_vec_splat(<3 x double> %x) {
 
 define float @known_positive_sign_arg(float %x, float %y) {
 ; CHECK-LABEL: @known_positive_sign_arg(
-; CHECK-NEXT:    [[FABS:%.*]] = call float @llvm.fabs.f32(float [[Y:%.*]])
-; CHECK-NEXT:    [[R:%.*]] = call ninf float @llvm.copysign.f32(float [[X:%.*]], float [[FABS]])
-; CHECK-NEXT:    ret float [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = call ninf float @llvm.fabs.f32(float [[X:%.*]])
+; CHECK-NEXT:    ret float [[TMP1]]
 ;
   %fabs = call float @llvm.fabs.f32(float %y)
   %r = call ninf float @llvm.copysign.f32(float %x, float %fabs)
@@ -56,9 +55,8 @@ define float @known_positive_sign_arg(float %x, float %y) {
 
 define <3 x double> @known_positive_sign_arg_vec(<3 x double> %x, <3 x i32> %y) {
 ; CHECK-LABEL: @known_positive_sign_arg_vec(
-; CHECK-NEXT:    [[YF:%.*]] = uitofp <3 x i32> [[Y:%.*]] to <3 x double>
-; CHECK-NEXT:    [[R:%.*]] = call arcp <3 x double> @llvm.copysign.v3f64(<3 x double> [[X:%.*]], <3 x double> [[YF]])
-; CHECK-NEXT:    ret <3 x double> [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = call arcp <3 x double> @llvm.fabs.v3f64(<3 x double> [[X:%.*]])
+; CHECK-NEXT:    ret <3 x double> [[TMP1]]
 ;
   %yf = uitofp <3 x i32> %y to <3 x double>
   %r = call arcp <3 x double> @llvm.copysign.v3f64(<3 x double> %x, <3 x double> %yf)