[DAGCombiner] exclude x*2.0 from normal negation profitability rules

[spatel]

Summary

This is the codegen part of fixing:
https://bugs.llvm.org/show_bug.cgi?id=32939

Even with the optimal/canonical IR that is ideally created by D65954, we would reverse that transform in DAGCombiner and end up with the same asm on AArch64 or x86.

I see 2 options for trying to correct this:

1. Limit isNegatibleForFree() by special-casing the fmul pattern (this patch).
2. Avoid creating (fmul X, 2.0) in the 1st place by adding a special-case transform to SelectionDAG::getNode() that matches the transform done by DAGCombiner.

This seems like the less intrusive patch, but let me know if there's some other reason to prefer 1 option over the other.

[cameron.mcinally]

LGTM. Less invasive is good.

But I do see value in not creating something just to immediately undo it. Option #2 would be similar to the special case in SelectionDAGBuilder::visitFSub(...), which is also good.

I don't have a strong opinion either way...

llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

static char isNegatibleForFree(SDValue Op, bool LegalOperations,
 
   case ISD::FMUL:
   case ISD::FDIV:
     // fold (fneg (fmul X, Y)) -> (fmul (fneg X), Y) or (fmul X, (fneg Y))
     if (char V = isNegatibleForFree(Op.getOperand(0), LegalOperations, TLI,
                                     Options, ForCodeSize, Depth + 1))
       return V;
 
+    // Ignore X * 2.0 because that is expected to be canonicalized to X + X.
+    if (auto *C = isConstOrConstSplatFP(Op.getOperand(1)))
+      if (C->isExactlyValue(2.0) && Op.getOpcode() == ISD::FMUL)
+        return 0;
+
     return isNegatibleForFree(Op.getOperand(1), LegalOperations, TLI, Options,
                               ForCodeSize, Depth + 1);
 
   case ISD::FP_EXTEND:
   case ISD::FP_ROUND:
   case ISD::FSIN:
     return isNegatibleForFree(Op.getOperand(0), LegalOperations, TLI, Options,
                               ForCodeSize, Depth + 1);

llvm/trunk/test/CodeGen/AArch64/fadd-combines.ll

; CHECK-NEXT:    ret
   ret float %a3
 }
 
 ; PR32939 - https://bugs.llvm.org/show_bug.cgi?id=32939
 
 define double @fmul2_negated(double %a, double %b, double %c) {
 ; CHECK-LABEL: fmul2_negated:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fmov d3, #-2.00000000
-; CHECK-NEXT:    fmul d1, d1, d3
+; CHECK-NEXT:    fadd d1, d1, d1
 ; CHECK-NEXT:    fmul d1, d1, d2
-; CHECK-NEXT:    fadd d0, d0, d1
+; CHECK-NEXT:    fsub d0, d0, d1
 ; CHECK-NEXT:    ret
   %mul = fmul double %b, 2.0
   %mul1 = fmul double %mul, %c
   %sub = fsub double %a, %mul1
   ret double %sub
 }
 
 define <2 x double> @fmul2_negated_vec(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 ; CHECK-LABEL: fmul2_negated_vec:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    fmov v3.2d, #-2.00000000
-; CHECK-NEXT:    fmul v1.2d, v1.2d, v3.2d
+; CHECK-NEXT:    fadd v1.2d, v1.2d, v1.2d
 ; CHECK-NEXT:    fmul v1.2d, v1.2d, v2.2d
-; CHECK-NEXT:    fadd v0.2d, v0.2d, v1.2d
+; CHECK-NEXT:    fsub v0.2d, v0.2d, v1.2d
 ; CHECK-NEXT:    ret
   %mul = fmul <2 x double> %b, <double 2.0, double 2.0>
   %mul1 = fmul <2 x double> %mul, %c
   %sub = fsub <2 x double> %a, %mul1
   ret <2 x double> %sub
 }
 
 declare void @use(double)
 

llvm/trunk/test/CodeGen/X86/fadd-combines.ll

; CHECK-NEXT:    retq
   ret float %a3
 }
 
 ; PR32939 - https://bugs.llvm.org/show_bug.cgi?id=32939
 
 define double @fmul2_negated(double %a, double %b, double %c) {
 ; CHECK-LABEL: fmul2_negated:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    mulsd {{.*}}(%rip), %xmm1
+; CHECK-NEXT:    addsd %xmm1, %xmm1
 ; CHECK-NEXT:    mulsd %xmm2, %xmm1
-; CHECK-NEXT:    addsd %xmm1, %xmm0
+; CHECK-NEXT:    subsd %xmm1, %xmm0
 ; CHECK-NEXT:    retq
   %mul = fmul double %b, 2.0
   %mul1 = fmul double %mul, %c
   %sub = fsub double %a, %mul1
   ret double %sub
 }
 
 define <2 x double> @fmul2_negated_vec(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 ; CHECK-LABEL: fmul2_negated_vec:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    mulpd {{.*}}(%rip), %xmm1
+; CHECK-NEXT:    addpd %xmm1, %xmm1
 ; CHECK-NEXT:    mulpd %xmm2, %xmm1
-; CHECK-NEXT:    addpd %xmm1, %xmm0
+; CHECK-NEXT:    subpd %xmm1, %xmm0
 ; CHECK-NEXT:    retq
   %mul = fmul <2 x double> %b, <double 2.0, double 2.0>
   %mul1 = fmul <2 x double> %mul, %c
   %sub = fsub <2 x double> %a, %mul1
   ret <2 x double> %sub
 }
 
 attributes #0 = { "less-precise-fpmad"="true" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "unsafe-fp-math"="true" "no-signed-zeros-fp-math"="true" }