Skip to content

[SDAG] Share signed zero handling for maximum and maximumnum #142762

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 1 commit into from
Closed
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
44 changes: 21 additions & 23 deletions llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -8559,6 +8559,23 @@ SDValue TargetLowering::expandFMINNUM_FMAXNUM(SDNode *Node,
return SDValue();
}

static SDValue emitSignedZeroOrdering(SelectionDAG &DAG, bool IsMax,
SDValue MinMax, SDValue LHS, SDValue RHS,
EVT CCVT, SDNodeFlags Flags,
const SDLoc &DL) {
EVT VT = MinMax.getValueType();
SDValue IsZero = DAG.getSetCC(DL, CCVT, MinMax,
DAG.getConstantFP(0.0, DL, VT), ISD::SETOEQ);
FloatSignAsInt State;
DAG.getSignAsIntValue(State, DL, LHS);
SDValue IsSpecificZero =
DAG.getSetCC(DL, CCVT, State.IntValue,
DAG.getConstant(0, DL, State.IntValue.getValueType()),
IsMax ? ISD::SETEQ : ISD::SETNE);
SDValue Sel = DAG.getSelect(DL, VT, IsSpecificZero, LHS, RHS, Flags);
return DAG.getSelect(DL, VT, IsZero, Sel, MinMax, Flags);
}

SDValue TargetLowering::expandFMINIMUM_FMAXIMUM(SDNode *N,
SelectionDAG &DAG) const {
if (SDValue Expanded = expandVectorNaryOpBySplitting(N, DAG))
Expand Down Expand Up @@ -8609,18 +8626,9 @@ SDValue TargetLowering::expandFMINIMUM_FMAXIMUM(SDNode *N,

// fminimum/fmaximum requires -0.0 less than +0.0
if (!MinMaxMustRespectOrderedZero && !N->getFlags().hasNoSignedZeros() &&
!DAG.isKnownNeverZeroFloat(RHS) && !DAG.isKnownNeverZeroFloat(LHS)) {
SDValue IsZero = DAG.getSetCC(DL, CCVT, MinMax,
DAG.getConstantFP(0.0, DL, VT), ISD::SETOEQ);
FloatSignAsInt State;
DAG.getSignAsIntValue(State, DL, LHS);
SDValue IsSpecificZero =
DAG.getSetCC(DL, CCVT, State.IntValue,
DAG.getConstant(0, DL, State.IntValue.getValueType()),
IsMax ? ISD::SETEQ : ISD::SETNE);
SDValue Sel = DAG.getSelect(DL, VT, IsSpecificZero, LHS, RHS, Flags);
MinMax = DAG.getSelect(DL, VT, IsZero, Sel, MinMax, Flags);
}
!DAG.isKnownNeverZeroFloat(RHS) && !DAG.isKnownNeverZeroFloat(LHS))
return emitSignedZeroOrdering(DAG, IsMax, MinMax, LHS, RHS, CCVT, Flags,
DL);

return MinMax;
}
Expand Down Expand Up @@ -8697,17 +8705,7 @@ SDValue TargetLowering::expandFMINIMUMNUM_FMAXIMUMNUM(SDNode *Node,
DAG.isKnownNeverZeroFloat(LHS) || DAG.isKnownNeverZeroFloat(RHS)) {
return MinMax;
}
SDValue TestZero =
DAG.getTargetConstant(IsMax ? fcPosZero : fcNegZero, DL, MVT::i32);
SDValue IsZero = DAG.getSetCC(DL, CCVT, MinMax,
DAG.getConstantFP(0.0, DL, VT), ISD::SETEQ);
SDValue LCmp = DAG.getSelect(
DL, VT, DAG.getNode(ISD::IS_FPCLASS, DL, CCVT, LHS, TestZero), LHS,
MinMax, Flags);
SDValue RCmp = DAG.getSelect(
DL, VT, DAG.getNode(ISD::IS_FPCLASS, DL, CCVT, RHS, TestZero), RHS, LCmp,
Flags);
return DAG.getSelect(DL, VT, IsZero, RCmp, MinMax, Flags);
return emitSignedZeroOrdering(DAG, IsMax, MinMax, LHS, RHS, CCVT, Flags, DL);
}

/// Returns a true value if if this FPClassTest can be performed with an ordered
Expand Down
745 changes: 312 additions & 433 deletions llvm/test/CodeGen/AMDGPU/fmax3-maximumnum.ll

Large diffs are not rendered by default.

834 changes: 354 additions & 480 deletions llvm/test/CodeGen/AMDGPU/fmin3-minimumnum.ll

Large diffs are not rendered by default.

17,251 changes: 7,782 additions & 9,469 deletions llvm/test/CodeGen/AMDGPU/maximumnum.bf16.ll

Large diffs are not rendered by default.

17,958 changes: 8,103 additions & 9,855 deletions llvm/test/CodeGen/AMDGPU/minimumnum.bf16.ll

Large diffs are not rendered by default.

90 changes: 30 additions & 60 deletions llvm/test/CodeGen/Mips/fp-maximumnum-minimumnum.ll
Original file line number Diff line number Diff line change
Expand Up @@ -17,18 +17,16 @@ define float @maximumnum_float(float %x, float %y) {
;
; MIPS64R2-LABEL: maximumnum_float:
; MIPS64R2: # %bb.0:
; MIPS64R2-NEXT: mov.s $f0, $f13
; MIPS64R2-NEXT: c.un.s $f12, $f12
; MIPS64R2-NEXT: movt.s $f12, $f13, $fcc0
; MIPS64R2-NEXT: c.un.s $f13, $f13
; MIPS64R2-NEXT: movt.s $f13, $f12, $fcc0
; MIPS64R2-NEXT: c.ule.s $f12, $f13
; MIPS64R2-NEXT: mov.s $f0, $f13
; MIPS64R2-NEXT: movf.s $f0, $f12, $fcc0
; MIPS64R2-NEXT: movt.s $f0, $f12, $fcc0
; MIPS64R2-NEXT: mfc1 $1, $f12
; MIPS64R2-NEXT: mov.s $f1, $f0
; MIPS64R2-NEXT: movz.s $f1, $f12, $1
; MIPS64R2-NEXT: mfc1 $1, $f13
; MIPS64R2-NEXT: movz.s $f1, $f13, $1
; MIPS64R2-NEXT: c.ule.s $f12, $f0
; MIPS64R2-NEXT: movf.s $f0, $f12, $fcc0
; MIPS64R2-NEXT: mtc1 $zero, $f2
; MIPS64R2-NEXT: c.eq.s $f0, $f2
; MIPS64R2-NEXT: jr $ra
Expand Down Expand Up @@ -67,14 +65,12 @@ define float @maximumnum_float_nnan(float %x, float %y) {
;
; MIPS64R2-LABEL: maximumnum_float_nnan:
; MIPS64R2: # %bb.0:
; MIPS64R2-NEXT: c.ule.s $f12, $f13
; MIPS64R2-NEXT: mov.s $f0, $f13
; MIPS64R2-NEXT: movf.s $f0, $f12, $fcc0
; MIPS64R2-NEXT: mfc1 $1, $f12
; MIPS64R2-NEXT: mov.s $f1, $f0
; MIPS64R2-NEXT: mov.s $f1, $f13
; MIPS64R2-NEXT: movz.s $f1, $f12, $1
; MIPS64R2-NEXT: mfc1 $1, $f13
; MIPS64R2-NEXT: movz.s $f1, $f13, $1
; MIPS64R2-NEXT: c.ule.s $f12, $f13
; MIPS64R2-NEXT: movf.s $f0, $f12, $fcc0
; MIPS64R2-NEXT: mtc1 $zero, $f2
; MIPS64R2-NEXT: c.eq.s $f0, $f2
; MIPS64R2-NEXT: jr $ra
Expand All @@ -94,18 +90,16 @@ define double @maximumnum_double(double %x, double %y) {
;
; MIPS64R2-LABEL: maximumnum_double:
; MIPS64R2: # %bb.0:
; MIPS64R2-NEXT: mov.d $f0, $f13
; MIPS64R2-NEXT: c.un.d $f12, $f12
; MIPS64R2-NEXT: movt.d $f12, $f13, $fcc0
; MIPS64R2-NEXT: c.un.d $f13, $f13
; MIPS64R2-NEXT: movt.d $f13, $f12, $fcc0
; MIPS64R2-NEXT: c.ule.d $f12, $f13
; MIPS64R2-NEXT: mov.d $f0, $f13
; MIPS64R2-NEXT: movf.d $f0, $f12, $fcc0
; MIPS64R2-NEXT: movt.d $f0, $f12, $fcc0
; MIPS64R2-NEXT: dmfc1 $1, $f12
; MIPS64R2-NEXT: mov.d $f1, $f0
; MIPS64R2-NEXT: movz.d $f1, $f12, $1
; MIPS64R2-NEXT: dmfc1 $1, $f13
; MIPS64R2-NEXT: movz.d $f1, $f13, $1
; MIPS64R2-NEXT: c.ule.d $f12, $f0
; MIPS64R2-NEXT: movf.d $f0, $f12, $fcc0
; MIPS64R2-NEXT: dmtc1 $zero, $f2
; MIPS64R2-NEXT: c.eq.d $f0, $f2
; MIPS64R2-NEXT: jr $ra
Expand Down Expand Up @@ -144,14 +138,12 @@ define double @maximumnum_double_nnan(double %x, double %y) {
;
; MIPS64R2-LABEL: maximumnum_double_nnan:
; MIPS64R2: # %bb.0:
; MIPS64R2-NEXT: c.ule.d $f12, $f13
; MIPS64R2-NEXT: mov.d $f0, $f13
; MIPS64R2-NEXT: movf.d $f0, $f12, $fcc0
; MIPS64R2-NEXT: dmfc1 $1, $f12
; MIPS64R2-NEXT: mov.d $f1, $f0
; MIPS64R2-NEXT: mov.d $f1, $f13
; MIPS64R2-NEXT: movz.d $f1, $f12, $1
; MIPS64R2-NEXT: dmfc1 $1, $f13
; MIPS64R2-NEXT: movz.d $f1, $f13, $1
; MIPS64R2-NEXT: c.ule.d $f12, $f13
; MIPS64R2-NEXT: movf.d $f0, $f12, $fcc0
; MIPS64R2-NEXT: dmtc1 $zero, $f2
; MIPS64R2-NEXT: c.eq.d $f0, $f2
; MIPS64R2-NEXT: jr $ra
Expand All @@ -170,21 +162,16 @@ define float @minimumnum_float(float %x, float %y) {
;
; MIPS64R2-LABEL: minimumnum_float:
; MIPS64R2: # %bb.0:
; MIPS64R2-NEXT: mov.s $f0, $f13
; MIPS64R2-NEXT: c.un.s $f12, $f12
; MIPS64R2-NEXT: movt.s $f12, $f13, $fcc0
; MIPS64R2-NEXT: c.un.s $f13, $f13
; MIPS64R2-NEXT: movt.s $f13, $f12, $fcc0
; MIPS64R2-NEXT: c.olt.s $f12, $f13
; MIPS64R2-NEXT: mov.s $f0, $f13
; MIPS64R2-NEXT: movt.s $f0, $f12, $fcc0
; MIPS64R2-NEXT: mfc1 $1, $f12
; MIPS64R2-NEXT: lui $2, 32768
; MIPS64R2-NEXT: xor $1, $1, $2
; MIPS64R2-NEXT: mov.s $f1, $f0
; MIPS64R2-NEXT: movz.s $f1, $f12, $1
; MIPS64R2-NEXT: mfc1 $1, $f13
; MIPS64R2-NEXT: xor $1, $1, $2
; MIPS64R2-NEXT: movz.s $f1, $f13, $1
; MIPS64R2-NEXT: movn.s $f1, $f12, $1
; MIPS64R2-NEXT: c.olt.s $f12, $f0
; MIPS64R2-NEXT: movt.s $f0, $f12, $fcc0
; MIPS64R2-NEXT: mtc1 $zero, $f2
; MIPS64R2-NEXT: c.eq.s $f0, $f2
; MIPS64R2-NEXT: jr $ra
Expand Down Expand Up @@ -223,17 +210,12 @@ define float @minimumnum_float_nnan(float %x, float %y) {
;
; MIPS64R2-LABEL: minimumnum_float_nnan:
; MIPS64R2: # %bb.0:
; MIPS64R2-NEXT: c.olt.s $f12, $f13
; MIPS64R2-NEXT: mov.s $f0, $f13
; MIPS64R2-NEXT: movt.s $f0, $f12, $fcc0
; MIPS64R2-NEXT: mfc1 $1, $f12
; MIPS64R2-NEXT: lui $2, 32768
; MIPS64R2-NEXT: xor $1, $1, $2
; MIPS64R2-NEXT: mov.s $f1, $f0
; MIPS64R2-NEXT: movz.s $f1, $f12, $1
; MIPS64R2-NEXT: mfc1 $1, $f13
; MIPS64R2-NEXT: xor $1, $1, $2
; MIPS64R2-NEXT: movz.s $f1, $f13, $1
; MIPS64R2-NEXT: mov.s $f1, $f13
; MIPS64R2-NEXT: movn.s $f1, $f12, $1
; MIPS64R2-NEXT: c.olt.s $f12, $f13
; MIPS64R2-NEXT: movt.s $f0, $f12, $fcc0
; MIPS64R2-NEXT: mtc1 $zero, $f2
; MIPS64R2-NEXT: c.eq.s $f0, $f2
; MIPS64R2-NEXT: jr $ra
Expand All @@ -252,22 +234,16 @@ define double @minimumnum_double(double %x, double %y) {
;
; MIPS64R2-LABEL: minimumnum_double:
; MIPS64R2: # %bb.0:
; MIPS64R2-NEXT: mov.d $f0, $f13
; MIPS64R2-NEXT: c.un.d $f12, $f12
; MIPS64R2-NEXT: movt.d $f12, $f13, $fcc0
; MIPS64R2-NEXT: c.un.d $f13, $f13
; MIPS64R2-NEXT: movt.d $f13, $f12, $fcc0
; MIPS64R2-NEXT: c.olt.d $f12, $f13
; MIPS64R2-NEXT: mov.d $f0, $f13
; MIPS64R2-NEXT: movt.d $f0, $f12, $fcc0
; MIPS64R2-NEXT: dmfc1 $1, $f12
; MIPS64R2-NEXT: daddiu $2, $zero, 1
; MIPS64R2-NEXT: dsll $2, $2, 63
; MIPS64R2-NEXT: xor $1, $1, $2
; MIPS64R2-NEXT: mov.d $f1, $f0
; MIPS64R2-NEXT: movz.d $f1, $f12, $1
; MIPS64R2-NEXT: dmfc1 $1, $f13
; MIPS64R2-NEXT: xor $1, $1, $2
; MIPS64R2-NEXT: movz.d $f1, $f13, $1
; MIPS64R2-NEXT: movn.d $f1, $f12, $1
; MIPS64R2-NEXT: c.olt.d $f12, $f0
; MIPS64R2-NEXT: movt.d $f0, $f12, $fcc0
; MIPS64R2-NEXT: dmtc1 $zero, $f2
; MIPS64R2-NEXT: c.eq.d $f0, $f2
; MIPS64R2-NEXT: jr $ra
Expand Down Expand Up @@ -306,18 +282,12 @@ define double @minimumnum_double_nnan(double %x, double %y) {
;
; MIPS64R2-LABEL: minimumnum_double_nnan:
; MIPS64R2: # %bb.0:
; MIPS64R2-NEXT: c.olt.d $f12, $f13
; MIPS64R2-NEXT: mov.d $f0, $f13
; MIPS64R2-NEXT: dmfc1 $1, $f12
; MIPS64R2-NEXT: mov.d $f1, $f13
; MIPS64R2-NEXT: movn.d $f1, $f12, $1
; MIPS64R2-NEXT: c.olt.d $f12, $f13
; MIPS64R2-NEXT: movt.d $f0, $f12, $fcc0
; MIPS64R2-NEXT: daddiu $1, $zero, 1
; MIPS64R2-NEXT: dsll $1, $1, 63
; MIPS64R2-NEXT: dmfc1 $2, $f12
; MIPS64R2-NEXT: xor $2, $2, $1
; MIPS64R2-NEXT: mov.d $f1, $f0
; MIPS64R2-NEXT: movz.d $f1, $f12, $2
; MIPS64R2-NEXT: dmfc1 $2, $f13
; MIPS64R2-NEXT: xor $1, $2, $1
; MIPS64R2-NEXT: movz.d $f1, $f13, $1
; MIPS64R2-NEXT: dmtc1 $zero, $f2
; MIPS64R2-NEXT: c.eq.d $f0, $f2
; MIPS64R2-NEXT: jr $ra
Expand Down
Loading
Loading