Push fneg into RHS instead of LHS

This is better because we canonicalize fneg to the right, so this
will cancel out two fnegs.

Author: nikic

llvm/lib/Transforms/InstCombine/InstCombineAddSub.cpp

@@ -2719,8 +2719,10 @@ Instruction *InstCombinerImpl::hoistFNegAboveFMulFDiv(Value *FNegOp,
                                                       Instruction &FMFSource) {
   Value *X, *Y;
   if (match(FNegOp, m_FMul(m_Value(X), m_Value(Y)))) {
+    // Push into RHS which is more likely to simplify (const or another fneg).
+    // FIXME: It would be better to invert the transform.
     return cast<Instruction>(Builder.CreateFMulFMF(
-        Builder.CreateFNegFMF(X, &FMFSource), Y, &FMFSource));
+        X, Builder.CreateFNegFMF(Y, &FMFSource), &FMFSource));
   }
 
   if (match(FNegOp, m_FDiv(m_Value(X), m_Value(Y)))) {

llvm/test/Transforms/InstCombine/fast-basictest.ll

@@ -549,9 +549,8 @@ define float @test16_reassoc(float %a, float %b, float %z) {
 
 define float @test17(float %a, float %b, float %z) {
 ; CHECK-LABEL: @test17(
-; CHECK-NEXT:    [[C:%.*]] = fmul fast float [[Z:%.*]], -4.000000e+01
-; CHECK-NEXT:    [[TMP1:%.*]] = fneg fast float [[A:%.*]]
-; CHECK-NEXT:    [[F:%.*]] = fmul fast float [[C]], [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fmul fast float [[Z:%.*]], 4.000000e+01
+; CHECK-NEXT:    [[F:%.*]] = fmul fast float [[A:%.*]], [[TMP1]]
 ; CHECK-NEXT:    ret float [[F]]
 ;
   %d = fmul fast float %z, 4.000000e+01
@@ -563,9 +562,8 @@ define float @test17(float %a, float %b, float %z) {
 
 define float @test17_unary_fneg(float %a, float %b, float %z) {
 ; CHECK-LABEL: @test17_unary_fneg(
-; CHECK-NEXT:    [[C:%.*]] = fmul fast float [[Z:%.*]], -4.000000e+01
-; CHECK-NEXT:    [[TMP1:%.*]] = fneg fast float [[A:%.*]]
-; CHECK-NEXT:    [[F:%.*]] = fmul fast float [[C]], [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fmul fast float [[Z:%.*]], 4.000000e+01
+; CHECK-NEXT:    [[F:%.*]] = fmul fast float [[A:%.*]], [[TMP1]]
 ; CHECK-NEXT:    ret float [[F]]
 ;
   %d = fmul fast float %z, 4.000000e+01
@@ -577,9 +575,8 @@ define float @test17_unary_fneg(float %a, float %b, float %z) {
 
 define float @test17_reassoc_nsz(float %a, float %b, float %z) {
 ; CHECK-LABEL: @test17_reassoc_nsz(
-; CHECK-NEXT:    [[C:%.*]] = fmul reassoc nsz float [[Z:%.*]], -4.000000e+01
-; CHECK-NEXT:    [[TMP1:%.*]] = fneg reassoc nsz float [[A:%.*]]
-; CHECK-NEXT:    [[F:%.*]] = fmul reassoc nsz float [[C]], [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fmul reassoc nsz float [[Z:%.*]], 4.000000e+01
+; CHECK-NEXT:    [[F:%.*]] = fmul reassoc nsz float [[A:%.*]], [[TMP1]]
 ; CHECK-NEXT:    ret float [[F]]
 ;
   %d = fmul reassoc nsz float %z, 4.000000e+01
@@ -609,9 +606,8 @@ define float @test17_reassoc(float %a, float %b, float %z) {
 
 define float @test17_unary_fneg_no_FMF(float %a, float %b, float %z) {
 ; CHECK-LABEL: @test17_unary_fneg_no_FMF(
-; CHECK-NEXT:    [[C:%.*]] = fmul float [[Z:%.*]], -4.000000e+01
-; CHECK-NEXT:    [[TMP1:%.*]] = fneg float [[A:%.*]]
-; CHECK-NEXT:    [[F:%.*]] = fmul float [[C]], [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fmul float [[Z:%.*]], 4.000000e+01
+; CHECK-NEXT:    [[F:%.*]] = fmul float [[A:%.*]], [[TMP1]]
 ; CHECK-NEXT:    ret float [[F]]
 ;
   %d = fmul float %z, 4.000000e+01
@@ -623,9 +619,8 @@ define float @test17_unary_fneg_no_FMF(float %a, float %b, float %z) {
 
 define float @test17_reassoc_unary_fneg(float %a, float %b, float %z) {
 ; CHECK-LABEL: @test17_reassoc_unary_fneg(
-; CHECK-NEXT:    [[C:%.*]] = fmul reassoc float [[Z:%.*]], -4.000000e+01
-; CHECK-NEXT:    [[TMP1:%.*]] = fneg reassoc float [[A:%.*]]
-; CHECK-NEXT:    [[F:%.*]] = fmul reassoc float [[C]], [[TMP1]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fmul reassoc float [[Z:%.*]], 4.000000e+01
+; CHECK-NEXT:    [[F:%.*]] = fmul reassoc float [[A:%.*]], [[TMP1]]
 ; CHECK-NEXT:    ret float [[F]]
 ;
   %d = fmul reassoc float %z, 4.000000e+01