[DAGCombine] GetNegatedExpression - add FMA\FMAD support

If the accumulator and either of the multiply operands are negatable then we can we negate the entire expression.

Differential Revision: https://reviews.llvm.org/D63141

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@369746 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Simon Pilgrim
2019-08-23 10:49:46 +00:00
parent 9c7f3a51fd
commit fe649a72d4
2 changed files with 64 additions and 13 deletions
+52 -1
View File
@@ -876,6 +876,27 @@ static char isNegatibleForFree(SDValue Op, bool LegalOperations,
return isNegatibleForFree(Op.getOperand(1), LegalOperations, TLI, Options,
ForCodeSize, Depth + 1);
case ISD::FMA:
case ISD::FMAD: {
if (!Options->NoSignedZerosFPMath && !Flags.hasNoSignedZeros())
return 0;
// fold (fneg (fma X, Y, Z)) -> (fma (fneg X), Y, (fneg Z))
// fold (fneg (fma X, Y, Z)) -> (fma X, (fneg Y), (fneg Z))
char V2 = isNegatibleForFree(Op.getOperand(2), LegalOperations, TLI,
Options, ForCodeSize, Depth + 1);
if (!V2)
return 0;
// One of Op0/Op1 must be cheaply negatible, then select the cheapest.
char V0 = isNegatibleForFree(Op.getOperand(0), LegalOperations, TLI,
Options, ForCodeSize, Depth + 1);
char V1 = isNegatibleForFree(Op.getOperand(1), LegalOperations, TLI,
Options, ForCodeSize, Depth + 1);
char V01 = std::max(V0, V1);
return V01 ? std::max(V01, V2) : 0;
}
case ISD::FP_EXTEND:
case ISD::FP_ROUND:
case ISD::FSIN:
@@ -917,7 +938,8 @@ static SDValue GetNegatedExpression(SDValue Op, SelectionDAG &DAG,
return DAG.getBuildVector(Op.getValueType(), SDLoc(Op), Ops);
}
case ISD::FADD:
assert(Options.NoSignedZerosFPMath || Flags.hasNoSignedZeros());
assert((Options.NoSignedZerosFPMath || Flags.hasNoSignedZeros()) &&
"Expected NSZ fp-flag");
// fold (fneg (fadd A, B)) -> (fsub (fneg A), B)
if (isNegatibleForFree(Op.getOperand(0), LegalOperations,
@@ -964,6 +986,35 @@ static SDValue GetNegatedExpression(SDValue Op, SelectionDAG &DAG,
LegalOperations, ForCodeSize,
Depth + 1), Flags);
case ISD::FMA:
case ISD::FMAD: {
assert((Options.NoSignedZerosFPMath || Flags.hasNoSignedZeros()) &&
"Expected NSZ fp-flag");
SDValue Neg2 = GetNegatedExpression(Op.getOperand(2), DAG, LegalOperations,
ForCodeSize, Depth + 1);
char V0 = isNegatibleForFree(Op.getOperand(0), LegalOperations,
DAG.getTargetLoweringInfo(), &Options,
ForCodeSize, Depth + 1);
char V1 = isNegatibleForFree(Op.getOperand(1), LegalOperations,
DAG.getTargetLoweringInfo(), &Options,
ForCodeSize, Depth + 1);
if (V0 >= V1) {
// fold (fneg (fma X, Y, Z)) -> (fma (fneg X), Y, (fneg Z))
SDValue Neg0 = GetNegatedExpression(
Op.getOperand(0), DAG, LegalOperations, ForCodeSize, Depth + 1);
return DAG.getNode(Op.getOpcode(), SDLoc(Op), Op.getValueType(), Neg0,
Op.getOperand(1), Neg2, Flags);
}
// fold (fneg (fma X, Y, Z)) -> (fma X, (fneg Y), (fneg Z))
SDValue Neg1 = GetNegatedExpression(Op.getOperand(1), DAG, LegalOperations,
ForCodeSize, Depth + 1);
return DAG.getNode(Op.getOpcode(), SDLoc(Op), Op.getValueType(),
Op.getOperand(0), Neg1, Neg2, Flags);
}
case ISD::FP_EXTEND:
case ISD::FSIN:
return DAG.getNode(Op.getOpcode(), SDLoc(Op), Op.getValueType(),
+12 -12
View File
@@ -5,14 +5,14 @@
define float @test_fneg_fma_subx_y_negz_f32(float %w, float %x, float %y, float %z) {
; FMA3-LABEL: test_fneg_fma_subx_y_negz_f32:
; FMA3: # %bb.0: # %entry
; FMA3-NEXT: vsubss %xmm1, %xmm0, %xmm0
; FMA3-NEXT: vfnmadd213ss {{.*#+}} xmm0 = -(xmm2 * xmm0) + xmm3
; FMA3-NEXT: vsubss %xmm0, %xmm1, %xmm0
; FMA3-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm2 * xmm0) + xmm3
; FMA3-NEXT: retq
;
; FMA4-LABEL: test_fneg_fma_subx_y_negz_f32:
; FMA4: # %bb.0: # %entry
; FMA4-NEXT: vsubss %xmm1, %xmm0, %xmm0
; FMA4-NEXT: vfnmaddss %xmm3, %xmm2, %xmm0, %xmm0
; FMA4-NEXT: vsubss %xmm0, %xmm1, %xmm0
; FMA4-NEXT: vfmaddss %xmm3, %xmm2, %xmm0, %xmm0
; FMA4-NEXT: retq
entry:
%subx = fsub nsz float %w, %x
@@ -25,14 +25,14 @@ entry:
define float @test_fneg_fma_x_suby_negz_f32(float %w, float %x, float %y, float %z) {
; FMA3-LABEL: test_fneg_fma_x_suby_negz_f32:
; FMA3: # %bb.0: # %entry
; FMA3-NEXT: vsubss %xmm2, %xmm0, %xmm0
; FMA3-NEXT: vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm3
; FMA3-NEXT: vsubss %xmm0, %xmm2, %xmm0
; FMA3-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm3
; FMA3-NEXT: retq
;
; FMA4-LABEL: test_fneg_fma_x_suby_negz_f32:
; FMA4: # %bb.0: # %entry
; FMA4-NEXT: vsubss %xmm2, %xmm0, %xmm0
; FMA4-NEXT: vfnmaddss %xmm3, %xmm0, %xmm1, %xmm0
; FMA4-NEXT: vsubss %xmm0, %xmm2, %xmm0
; FMA4-NEXT: vfmaddss %xmm3, %xmm0, %xmm1, %xmm0
; FMA4-NEXT: retq
entry:
%suby = fsub nsz float %w, %y
@@ -45,16 +45,16 @@ entry:
define float @test_fneg_fma_subx_suby_negz_f32(float %w, float %x, float %y, float %z) {
; FMA3-LABEL: test_fneg_fma_subx_suby_negz_f32:
; FMA3: # %bb.0: # %entry
; FMA3-NEXT: vsubss %xmm1, %xmm0, %xmm1
; FMA3-NEXT: vsubss %xmm0, %xmm1, %xmm1
; FMA3-NEXT: vsubss %xmm2, %xmm0, %xmm0
; FMA3-NEXT: vfnmadd213ss {{.*#+}} xmm0 = -(xmm1 * xmm0) + xmm3
; FMA3-NEXT: vfmadd213ss {{.*#+}} xmm0 = (xmm1 * xmm0) + xmm3
; FMA3-NEXT: retq
;
; FMA4-LABEL: test_fneg_fma_subx_suby_negz_f32:
; FMA4: # %bb.0: # %entry
; FMA4-NEXT: vsubss %xmm1, %xmm0, %xmm1
; FMA4-NEXT: vsubss %xmm0, %xmm1, %xmm1
; FMA4-NEXT: vsubss %xmm2, %xmm0, %xmm0
; FMA4-NEXT: vfnmaddss %xmm3, %xmm0, %xmm1, %xmm0
; FMA4-NEXT: vfmaddss %xmm3, %xmm0, %xmm1, %xmm0
; FMA4-NEXT: retq
entry:
%subx = fsub nsz float %w, %x