Skip to content

[AArch64] Remove copy in SVE/SME predicate spill and fill #81716

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.

Sign up for GitHub

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

Merged
merged 19 commits into from
Apr 9, 2024
Merged

[AArch64] Remove copy in SVE/SME predicate spill and fill #81716

Show file tree
Hide file tree
Changes from 13 commits
Commits
Show all changes
19 commits
Select commit Hold shift + click to select a range
c0c0609
[AArch64] Remove copy in SVE/SME predicate spill
SamTebbs33 Feb 13, 2024
929b111
fixup: constrain reg class instead of convert
SamTebbs33 Feb 21, 2024
514b9b5
fixup: let predicate spill and fill take ppr or pnr
SamTebbs33 Feb 27, 2024
264a844
fixup: remove braces
SamTebbs33 Feb 27, 2024
53a3177
fixup: add decode function
SamTebbs33 Feb 28, 2024
6a4502c
fixup: remove inst aliases
SamTebbs33 Feb 28, 2024
808ddbb
fixup: match reg class in AArch64AsmParser
SamTebbs33 Feb 28, 2024
ed8519e
Add svcount to reg class
SamTebbs33 Feb 28, 2024
6a55451
fixup: format definitions
SamTebbs33 Feb 28, 2024
a229c76
fixup: use reg class numbers in MIR tests
SamTebbs33 Feb 29, 2024
8c5b567
fixup: Add parsing and rendering functions
SamTebbs33 Mar 6, 2024
a64ca22
fixup: remove spaces
SamTebbs33 Mar 7, 2024
9e59c45
fixup: format decode function
SamTebbs33 Mar 7, 2024
1a398e1
fixup: merge pprorpnr and ppraspnr handling
SamTebbs33 Mar 20, 2024
2f6d418
fixup: combine ppr and pnr stack slot handling
SamTebbs33 Apr 2, 2024
f06c0ce
fixup: add <= pn15 to asm parser
SamTebbs33 Apr 2, 2024
23b051b
fixup: simplify load/storeRegToStackSlot
SamTebbs33 Apr 5, 2024
a53aeb4
fixup: remove extra parentheses
SamTebbs33 Apr 5, 2024
03aabd2
fixup: remove IsPPR and move error match case statement
SamTebbs33 Apr 9, 2024
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
12 changes: 4 additions & 8 deletions llvm/lib/Target/AArch64/AArch64InstrInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4821,13 +4821,9 @@ void AArch64InstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
} else if (AArch64::PNRRegClass.hasSubClassEq(RC)) {
assert((Subtarget.hasSVE2p1() || Subtarget.hasSME2()) &&
"Unexpected register store without SVE2p1 or SME2");
if (SrcReg.isVirtual()) {
auto NewSrcReg =
MF.getRegInfo().createVirtualRegister(&AArch64::PPRRegClass);
BuildMI(MBB, MBBI, DebugLoc(), get(TargetOpcode::COPY), NewSrcReg)
.addReg(SrcReg);
SrcReg = NewSrcReg;
} else
if (SrcReg.isVirtual())
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Now that you've made the instructions accept both p0-p15 and pn0-pn15, we can remove all the code that tries to handle PNR registers differently from P registers in storeRegToStackSlot and loadRegFromStackSlot.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good idea, done.

MF.getRegInfo().constrainRegClass(SrcReg, &AArch64::PPRRegClass);
else
SrcReg = (SrcReg - AArch64::PN0) + AArch64::P0;
Opc = AArch64::STR_PXI;
StackID = TargetStackID::ScalableVector;
Expand Down Expand Up @@ -5006,7 +5002,7 @@ void AArch64InstrInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
"Unexpected register load without SVE2p1 or SME2");
PNRReg = DestReg;
if (DestReg.isVirtual())
DestReg = MF.getRegInfo().createVirtualRegister(&AArch64::PPRRegClass);
MF.getRegInfo().constrainRegClass(DestReg, &AArch64::PPRRegClass);
else
DestReg = (DestReg - AArch64::PN0) + AArch64::P0;
Opc = AArch64::LDR_PXI;
Expand Down
21 changes: 21 additions & 0 deletions llvm/lib/Target/AArch64/AArch64RegisterInfo.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1004,6 +1004,27 @@ let Namespace = "AArch64" in {
def psub1 : SubRegIndex<16, -1>;
}

class PPRorPNRClass : RegisterClass<
"AArch64",
[ nxv16i1, nxv8i1, nxv4i1, nxv2i1, nxv1i1, aarch64svcount ], 16,
(add PPR, PNR)> {
let Size = 16;
}

class PPRorPNRAsmOperand<string name, string RegClass, int Width>: AsmOperandClass {
let Name = "SVE" # name # "Reg";
let PredicateMethod = "isSVEPredicateOrPredicateAsCounterRegOfWidth<"
# Width # ", " # "AArch64::"
# RegClass # "RegClassID>";
let DiagnosticType = "InvalidSVE" # name # "Reg";
let RenderMethod = "addPPRorPNRRegOperands";
let ParserMethod = "tryParseSVEPredicateOrPredicateAsCounterVector";
}

def PPRorPNR : PPRorPNRClass;
def PPRorPNRAsmOpAny : PPRorPNRAsmOperand<"PPRorPNRAny", "PPRorPNR", 0>;
def PPRorPNRAny : PPRRegOp<"", PPRorPNRAsmOpAny, ElementSizeNone, PPRorPNR>;

// Pairs of SVE predicate vector registers.
def PSeqPairs : RegisterTuples<[psub0, psub1], [(rotl PPR, 0), (rotl PPR, 1)]>;

Expand Down
10 changes: 0 additions & 10 deletions llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4017,16 +4017,6 @@ let Predicates = [HasSVEorSME] in {

// Aliases for existing SVE instructions for which predicate-as-counter are
// accepted as an operand to the instruction
def : InstAlias<"ldr $Pt, [$Rn, $imm9, mul vl]",
(LDR_PXI PNRasPPRAny:$Pt, GPR64sp:$Rn, simm9:$imm9), 0>;
def : InstAlias<"ldr $Pt, [$Rn]",
(LDR_PXI PNRasPPRAny:$Pt, GPR64sp:$Rn, 0), 0>;

def : InstAlias<"str $Pt, [$Rn, $imm9, mul vl]",
(STR_PXI PNRasPPRAny:$Pt, GPR64sp:$Rn, simm9:$imm9), 0>;
def : InstAlias<"str $Pt, [$Rn]",
(STR_PXI PNRasPPRAny:$Pt, GPR64sp:$Rn, 0), 0>;

def : InstAlias<"mov $Pd, $Pn",
(ORR_PPzPP PNRasPPR8:$Pd, PNRasPPR8:$Pn, PNRasPPR8:$Pn, PNRasPPR8:$Pn), 0>;

Expand Down
38 changes: 38 additions & 0 deletions llvm/lib/Target/AArch64/AsmParser/AArch64AsmParser.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -276,6 +276,8 @@ class AArch64AsmParser : public MCTargetAsmParser {
ParseStatus tryParseSVEDataVector(OperandVector &Operands);
template <RegKind RK>
ParseStatus tryParseSVEPredicateVector(OperandVector &Operands);
ParseStatus
tryParseSVEPredicateOrPredicateAsCounterVector(OperandVector &Operands);
template <RegKind VectorKind>
ParseStatus tryParseVectorList(OperandVector &Operands,
bool ExpectMatch = false);
Expand Down Expand Up @@ -1241,6 +1243,7 @@ class AArch64Operand : public MCParsedAsmOperand {
case AArch64::PPR_p8to15RegClassID:
case AArch64::PNRRegClassID:
case AArch64::PNR_p8to15RegClassID:
case AArch64::PPRorPNRRegClassID:
RK = RegKind::SVEPredicateAsCounter;
break;
default:
Expand All @@ -1264,6 +1267,7 @@ class AArch64Operand : public MCParsedAsmOperand {
case AArch64::PPR_p8to15RegClassID:
case AArch64::PNRRegClassID:
case AArch64::PNR_p8to15RegClassID:
case AArch64::PPRorPNRRegClassID:
RK = RegKind::SVEPredicateVector;
break;
default:
Expand All @@ -1290,6 +1294,20 @@ class AArch64Operand : public MCParsedAsmOperand {
return DiagnosticPredicateTy::NearMatch;
}

template <int ElementWidth, unsigned Class>
DiagnosticPredicate isSVEPredicateOrPredicateAsCounterRegOfWidth() const {
if (Kind != k_Register || (Reg.Kind != RegKind::SVEPredicateAsCounter &&
Reg.Kind != RegKind::SVEPredicateVector))
return DiagnosticPredicateTy::NoMatch;

if ((isSVEPredicateAsCounterReg<Class>() ||
isSVEPredicateVectorRegOfWidth<ElementWidth, Class>()) &&
(Reg.ElementWidth == ElementWidth))
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: no parenthesis needed in:

Suggested change
(Reg.ElementWidth == ElementWidth))
Reg.ElementWidth == ElementWidth)

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done

return DiagnosticPredicateTy::Match;

return DiagnosticPredicateTy::NearMatch;
}

template <int ElementWidth, unsigned Class>
DiagnosticPredicate isSVEPredicateAsCounterRegOfWidth() const {
if (Kind != k_Register || Reg.Kind != RegKind::SVEPredicateAsCounter)
Expand Down Expand Up @@ -1770,6 +1788,15 @@ class AArch64Operand : public MCParsedAsmOperand {
Inst.addOperand(MCOperand::createReg(AArch64::Z0 + getReg() - Base));
}

void addPPRorPNRRegOperands(MCInst &Inst, unsigned N) const {
assert(N == 1 && "Invalid number of operands!");
unsigned Reg = getReg();
// Normalise to PPR
if (Reg >= AArch64::PN0)
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit:

Suggested change
if (Reg >= AArch64::PN0)
if (Reg >= AArch64::PN0 && Reg <= AArch64::PN15)

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done

Reg = Reg - AArch64::PN0 + AArch64::P0;
Inst.addOperand(MCOperand::createReg(Reg));
}

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can addPNRasPPRRegOperands be removed now? Or perhaps I should ask: is PNRasPPR still required, or can those instructions that use it also use PPRorPNRRegOperand ?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

PPRasPNR can be removed. Thanks for the idea.

void addPNRasPPRRegOperands(MCInst &Inst, unsigned N) const {
assert(N == 1 && "Invalid number of operands!");
Inst.addOperand(
Expand Down Expand Up @@ -4167,6 +4194,15 @@ ParseStatus AArch64AsmParser::tryParseVectorRegister(MCRegister &Reg,
return ParseStatus::NoMatch;
}

ParseStatus AArch64AsmParser::tryParseSVEPredicateOrPredicateAsCounterVector(
OperandVector &Operands) {
ParseStatus Status =
tryParseSVEPredicateVector<RegKind::SVEPredicateAsCounter>(Operands);
if (!Status.isSuccess())
Status = tryParseSVEPredicateVector<RegKind::SVEPredicateVector>(Operands);
return Status;
}

/// tryParseSVEPredicateVector - Parse a SVE predicate register operand.
template <RegKind RK>
ParseStatus
Expand Down Expand Up @@ -6019,6 +6055,7 @@ bool AArch64AsmParser::showMatchError(SMLoc Loc, unsigned ErrCode,
return Error(Loc, "Invalid restricted vector register, expected z0.d..z15.d");
case Match_InvalidSVEPattern:
return Error(Loc, "invalid predicate pattern");
case Match_InvalidSVEPPRorPNRAnyReg:
case Match_InvalidSVEPredicateAnyReg:
case Match_InvalidSVEPredicateBReg:
case Match_InvalidSVEPredicateHReg:
Expand Down Expand Up @@ -6653,6 +6690,7 @@ bool AArch64AsmParser::MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode,
case Match_InvalidZPR_4b16:
case Match_InvalidZPR_4b32:
case Match_InvalidZPR_4b64:
case Match_InvalidSVEPPRorPNRAnyReg:
case Match_InvalidSVEPredicateAnyReg:
case Match_InvalidSVEPattern:
case Match_InvalidSVEVecLenSpecifier:
Expand Down
15 changes: 15 additions & 0 deletions llvm/lib/Target/AArch64/Disassembler/AArch64Disassembler.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -143,6 +143,9 @@ DecodeMatrixTileListRegisterClass(MCInst &Inst, unsigned RegMask,
static DecodeStatus DecodePPRRegisterClass(MCInst &Inst, unsigned RegNo,
uint64_t Address,
const MCDisassembler *Decoder);
static DecodeStatus DecodePPRorPNRRegisterClass(MCInst &Inst, unsigned RegNo,
uint64_t Addr,
const MCDisassembler *Decoder);
static DecodeStatus DecodePNRRegisterClass(MCInst &Inst, unsigned RegNo,
uint64_t Address,
const MCDisassembler *Decoder);
Expand Down Expand Up @@ -741,6 +744,18 @@ static DecodeStatus DecodeMatrixTile(MCInst &Inst, unsigned RegNo,
return Success;
}

static DecodeStatus DecodePPRorPNRRegisterClass(MCInst &Inst, unsigned RegNo,
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Cleanup suggestion for a separate patch:

All these decoder classes looks rather identical, it would be nice to clean this up with something like this:

template <unsigned RegClassID, unsigned FirstReg, unsigned NumRegsInClass>
static DecodeStatus DecodeSimpleReg(MCInst &Inst, unsigned RegNo,
                                                uint64_t Addr,
                                                const MCDisassembler *Decoder) {
  if (RegNo > (NumRegsInClass-1))
    return Fail;

  unsigned Register =
      AArch64MCRegisterClasses[RegClassID].getRegister(RegNo);
  Inst.addOperand(MCOperand::createReg(Register + FirstReg));
  return Success;
}

And then remove the other functions and just specify e.g. DecodeSimpleReg<AArch64::PPRorPNRRegClassID, 0, 16> in the .td file.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That sounds good to me. We can do it in a separate patch.

uint64_t Addr,
const MCDisassembler *Decoder) {
if (RegNo > 15)
return Fail;

unsigned Register =
AArch64MCRegisterClasses[AArch64::PPRorPNRRegClassID].getRegister(RegNo);
Inst.addOperand(MCOperand::createReg(Register));
return Success;
}

static DecodeStatus DecodePPRRegisterClass(MCInst &Inst, unsigned RegNo,
uint64_t Addr,
const MCDisassembler *Decoder) {
Expand Down
8 changes: 4 additions & 4 deletions llvm/lib/Target/AArch64/SVEInstrFormats.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6689,7 +6689,7 @@ multiclass sve_mem_z_spill<string asm> {
}

class sve_mem_p_spill<string asm>
: I<(outs), (ins PPRAny:$Pt, GPR64sp:$Rn, simm9:$imm9),
: I<(outs), (ins PPRorPNRAny:$Pt, GPR64sp:$Rn, simm9:$imm9),
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This removes the need for having the following InstAliases in AArch64SVEInstrInfo.td:

// Aliases for existing SVE instructions for which predicate-as-counter are
// accepted as an operand to the instruction
def : InstAlias<"ldr $Pt, [$Rn, $imm9, mul vl]",
               (LDR_PXI PNRasPPRAny:$Pt, GPR64sp:$Rn, simm9:$imm9), 0>;
def : InstAlias<"ldr $Pt, [$Rn]",
               (LDR_PXI PNRasPPRAny:$Pt, GPR64sp:$Rn, 0), 0>;

def : InstAlias<"str $Pt, [$Rn, $imm9, mul vl]",
               (STR_PXI PNRasPPRAny:$Pt, GPR64sp:$Rn, simm9:$imm9), 0>;
def : InstAlias<"str $Pt, [$Rn]",
               (STR_PXI PNRasPPRAny:$Pt, GPR64sp:$Rn, 0), 0>;

Can you remove them?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done.

asm, "\t$Pt, [$Rn, $imm9, mul vl]",
"",
[]>, Sched<[]> {
Expand All @@ -6712,7 +6712,7 @@ multiclass sve_mem_p_spill<string asm> {
def NAME : sve_mem_p_spill<asm>;

def : InstAlias<asm # "\t$Pt, [$Rn]",
(!cast<Instruction>(NAME) PPRAny:$Pt, GPR64sp:$Rn, 0), 1>;
(!cast<Instruction>(NAME) PPRorPNRAny:$Pt, GPR64sp:$Rn, 0), 1>;
}

//===----------------------------------------------------------------------===//
Expand Down Expand Up @@ -7858,7 +7858,7 @@ multiclass sve_mem_z_fill<string asm> {
}

class sve_mem_p_fill<string asm>
: I<(outs PPRAny:$Pt), (ins GPR64sp:$Rn, simm9:$imm9),
: I<(outs PPRorPNRAny:$Pt), (ins GPR64sp:$Rn, simm9:$imm9),
asm, "\t$Pt, [$Rn, $imm9, mul vl]",
"",
[]>, Sched<[]> {
Expand All @@ -7881,7 +7881,7 @@ multiclass sve_mem_p_fill<string asm> {
def NAME : sve_mem_p_fill<asm>;

def : InstAlias<asm # "\t$Pt, [$Rn]",
(!cast<Instruction>(NAME) PPRAny:$Pt, GPR64sp:$Rn, 0), 1>;
(!cast<Instruction>(NAME) PPRorPNRAny:$Pt, GPR64sp:$Rn, 0), 1>;
}

class sve2_mem_gldnt_vs_base<bits<5> opc, dag iops, string asm,
Expand Down
8 changes: 4 additions & 4 deletions llvm/test/CodeGen/AArch64/GlobalISel/regbank-inlineasm.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,11 +57,11 @@ tracksRegLiveness: true
body: |
bb.1:
; CHECK-LABEL: name: inlineasm_virt_reg_output
; CHECK: INLINEASM &"mov ${0:w}, 7", 0 /* attdialect */, 1310730 /* regdef:PPR2_with_psub_in_PNR_3b_and_PPR2_with_psub1_in_PPR_p8to15 */, def %0
; CHECK: INLINEASM &"mov ${0:w}, 7", 0 /* attdialect */, 1769482 /* regdef:GPR32common */, def %0
; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr(s32) = COPY %0
; CHECK-NEXT: $w0 = COPY [[COPY]](s32)
; CHECK-NEXT: RET_ReallyLR implicit $w0
INLINEASM &"mov ${0:w}, 7", 0 /* attdialect */, 1310730 /* regdef:GPR32common */, def %0:gpr32common
INLINEASM &"mov ${0:w}, 7", 0 /* attdialect */, 1769482 /* regdef:GPR32common */, def %0:gpr32common
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Suggested change
INLINEASM &"mov ${0:w}, 7", 0 /* attdialect */, 1769482 /* regdef:GPR32common */, def %0:gpr32common
INLINEASM &"mov ${0:w}, 7", 0 /* attdialect */, {{[0-9]+}} /* regdef:GPR32common */, def %0:gpr32common

(same in other places)

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We can only use the regex in a CHECK statement, not in the IR being tested.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah I didn't spot the leading ;!

In that case, I would suggest not using a regex for these numbers, because they must match the ones from the MIR.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Good point, I've changed the .mir tests to explicitly match the regclass numbers.

%1:_(s32) = COPY %0
$w0 = COPY %1(s32)
RET_ReallyLR implicit $w0
Expand All @@ -75,12 +75,12 @@ tracksRegLiveness: true
body: |
bb.1:
; CHECK-LABEL: name: inlineasm_virt_mixed_types
; CHECK: INLINEASM &"mov $0, #0; mov $1, #0", 0 /* attdialect */, 1310730 /* regdef:PPR2_with_psub_in_PNR_3b_and_PPR2_with_psub1_in_PPR_p8to15 */, def %0, 2162698 /* regdef:WSeqPairsClass */, def %1
; CHECK: INLINEASM &"mov $0, #0; mov $1, #0", 0 /* attdialect */, 1769482 /* regdef:GPR32common */, def %0, {{[0-9]+}} /* regdef:FPR64 */, def %1
; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr(s32) = COPY %0
; CHECK-NEXT: [[COPY1:%[0-9]+]]:fpr(s64) = COPY %1
; CHECK-NEXT: $d0 = COPY [[COPY1]](s64)
; CHECK-NEXT: RET_ReallyLR implicit $d0
INLINEASM &"mov $0, #0; mov $1, #0", 0 /* attdialect */, 1310730 /* regdef:GPR32common */, def %0:gpr32common, 2162698 /* regdef:FPR64 */, def %1:fpr64
INLINEASM &"mov $0, #0; mov $1, #0", 0 /* attdialect */, 1769482 /* regdef:GPR32common */, def %0:gpr32common, 2621450 /* regdef:FPR64 */, def %1:fpr64
%3:_(s32) = COPY %0
%4:_(s64) = COPY %1
$d0 = COPY %4(s64)
Expand Down
8 changes: 4 additions & 4 deletions llvm/test/CodeGen/AArch64/emit_fneg_with_non_register_operand.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -91,10 +91,10 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[LOADgot:%[0-9]+]]:gpr64common = LOADgot target-flags(aarch64-got) @c
; CHECK-NEXT: [[LDRDui:%[0-9]+]]:fpr64 = LDRDui [[LOADgot]], 0 :: (dereferenceable load (s64) from @c)
; CHECK-NEXT: INLINEASM &"", 1 /* sideeffect attdialect */, {{[0-9]+}} /* regdef:WSeqPairsClass_with_sube32_in_MatrixIndexGPR32_12_15 */, def %2, 2147483657 /* reguse tiedto:$0 */, [[LDRDui]](tied-def 3)
; CHECK-NEXT: INLINEASM &"", 1 /* sideeffect attdialect */, 2621450 /* regdef:FPR64 */, def %2, 2147483657 /* reguse tiedto:$0 */, [[LDRDui]](tied-def 3)
; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr64 = COPY %2
; CHECK-NEXT: [[LDRDui1:%[0-9]+]]:fpr64 = LDRDui [[LOADgot]], 0 :: (dereferenceable load (s64) from @c)
; CHECK-NEXT: INLINEASM &"", 1 /* sideeffect attdialect */, {{[0-9]+}} /* regdef:WSeqPairsClass_with_sube32_in_MatrixIndexGPR32_12_15 */, def %4, 2147483657 /* reguse tiedto:$0 */, [[LDRDui1]](tied-def 3)
; CHECK-NEXT: INLINEASM &"", 1 /* sideeffect attdialect */, 2621450 /* regdef:FPR64 */, def %4, 2147483657 /* reguse tiedto:$0 */, [[LDRDui1]](tied-def 3)
; CHECK-NEXT: [[FNEGDr:%[0-9]+]]:fpr64 = FNEGDr %2
; CHECK-NEXT: nofpexcept FCMPDrr %4, killed [[FNEGDr]], implicit-def $nzcv, implicit $fpcr
; CHECK-NEXT: Bcc 1, %bb.2, implicit $nzcv
Expand All @@ -111,10 +111,10 @@ body: |

%6:gpr64common = LOADgot target-flags(aarch64-got) @c
%3:fpr64 = LDRDui %6, 0 :: (dereferenceable load (s64) from @c)
INLINEASM &"", 1 /* sideeffect attdialect */, 2359306 /* regdef:FPR64 */, def %2, 2147483657 /* reguse tiedto:$0 */, %3(tied-def 3)
INLINEASM &"", 1 /* sideeffect attdialect */, 2621450 /* regdef:FPR64 */, def %2, 2147483657 /* reguse tiedto:$0 */, %3(tied-def 3)
%0:fpr64 = COPY %2
%5:fpr64 = LDRDui %6, 0 :: (dereferenceable load (s64) from @c)
INLINEASM &"", 1 /* sideeffect attdialect */, 2359306 /* regdef:FPR64 */, def %4, 2147483657 /* reguse tiedto:$0 */, %5(tied-def 3)
INLINEASM &"", 1 /* sideeffect attdialect */, 2621450 /* regdef:FPR64 */, def %4, 2147483657 /* reguse tiedto:$0 */, %5(tied-def 3)
%7:fpr64 = FNEGDr %2
nofpexcept FCMPDrr %4, killed %7, implicit-def $nzcv, implicit $fpcr
Bcc 1, %bb.2, implicit $nzcv
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/AArch64/peephole-insvigpr.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -487,7 +487,7 @@ body: |
; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
; CHECK-NEXT: [[DEF:%[0-9]+]]:gpr64all = IMPLICIT_DEF
; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64sp = COPY [[DEF]]
; CHECK-NEXT: INLINEASM &"ldr ${0:s}, $1", 8 /* mayload attdialect */, {{[0-9]+}} /* regdef:WSeqPairsClass_with_sube32_in_MatrixIndexGPR32_12_15 */, def %1, 262158 /* mem:m */, killed [[COPY1]]
; CHECK-NEXT: INLINEASM &"ldr ${0:s}, $1", 8 /* mayload attdialect */, 2621450 /* regdef:FPR64 */, def %1, 262158 /* mem:m */, killed [[COPY1]]
; CHECK-NEXT: [[MOVIv2d_ns:%[0-9]+]]:fpr128 = MOVIv2d_ns 0
; CHECK-NEXT: [[COPY2:%[0-9]+]]:fpr64 = COPY [[MOVIv2d_ns]].dsub
; CHECK-NEXT: [[DEF1:%[0-9]+]]:fpr128 = IMPLICIT_DEF
Expand All @@ -505,7 +505,7 @@ body: |
%0:gpr64common = COPY $x0
%2:gpr64all = IMPLICIT_DEF
%3:gpr64sp = COPY %2
INLINEASM &"ldr ${0:s}, $1", 8 /* mayload attdialect */, 2359306 /* regdef:FPR64 */, def %1, 262158 /* mem:m */, killed %3
INLINEASM &"ldr ${0:s}, $1", 8 /* mayload attdialect */, 2621450 /* regdef:FPR64 */, def %1, 262158 /* mem:m */, killed %3
%4:fpr128 = MOVIv2d_ns 0
%5:fpr64 = COPY %4.dsub
%7:fpr128 = IMPLICIT_DEF
Expand Down
8 changes: 3 additions & 5 deletions llvm/test/CodeGen/AArch64/spillfill-sve.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
# RUN: llc -mtriple=aarch64-linux-gnu -run-pass=greedy %s -o - | FileCheck %s
# RUN: llc -mtriple=aarch64-linux-gnu -start-before=greedy -stop-after=aarch64-expand-pseudo %s -o - | FileCheck %s --check-prefix=EXPAND
# RUN: llc -mtriple=aarch64-linux-gnu -start-before=greedy -stop-after=aarch64-expand-pseudo -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=EXPAND
--- |
; ModuleID = '<stdin>'
source_filename = "<stdin>"
Expand Down Expand Up @@ -213,11 +213,9 @@ body: |

; EXPAND-LABEL: name: spills_fills_stack_id_virtreg_pnr
; EXPAND: renamable $pn8 = WHILEGE_CXX_B
; EXPAND: $p0 = ORR_PPzPP $p8, $p8, killed $p8
; EXPAND: STR_PXI killed renamable $p0, $sp, 7
; EXPAND: STR_PXI killed renamable $pn8, $sp, 7
;
; EXPAND: renamable $p0 = LDR_PXI $sp, 7
; EXPAND: $p8 = ORR_PPzPP $p0, $p0, killed $p0, implicit-def $pn8
; EXPAND: renamable $pn8 = LDR_PXI $sp, 7
; EXPAND: $p0 = PEXT_PCI_B killed renamable $pn8, 0


Expand Down