Skip to content

[AIE] Add FP64 support to G_FCMP legalizer (issue #431) #439

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

Open
wants to merge 2 commits into
base: aie-public
Choose a base branch
from
Open

[AIE] Add FP64 support to G_FCMP legalizer (issue #431) #439

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
8af6dec
[AIE2] Add FP64 support to G_FCMP legalizer
jseaton Apr 16, 2025
c2d8451
[AIE2] Add G_FCMP F64 legalizer tests
jseaton Apr 17, 2025
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
2 changes: 1 addition & 1 deletion llvm/lib/Target/AIE/AIE2LegalizerInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -177,7 +177,7 @@ AIE2LegalizerInfo::AIE2LegalizerInfo(const AIE2Subtarget &ST) : AIEHelper(ST) {

getActionDefinitionsBuilder(G_FCMP)
.clampScalar(0, S32, S32)
.customFor({{S32, S16}, {S32, S32}});
.customFor({{S32, S16}, {S32, S32}, {S32, S64}});

getActionDefinitionsBuilder(G_FPTRUNC)
.libcallFor({{S32, S64}})
Expand Down
79 changes: 64 additions & 15 deletions llvm/lib/Target/AIE/AIELegalizerHelper.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -805,8 +805,8 @@ bool AIELegalizerHelper::legalizeG_INSERT_VECTOR_ELT(LegalizerHelper &Helper,
return true;
}

static RTLIB::Libcall getFCmpLibCall(CmpInst::Predicate Predicate,
CmpInst::Predicate &IPredicate) {
static RTLIB::Libcall getFCmp32LibCall(CmpInst::Predicate Predicate,
CmpInst::Predicate &IPredicate) {
switch (Predicate) {
default:
llvm_unreachable("Unsupported FCmp predicate");
Expand Down Expand Up @@ -851,18 +851,66 @@ static RTLIB::Libcall getFCmpLibCall(CmpInst::Predicate Predicate,
}
}

bool AIELegalizerHelper::legalizeG_FCMP_FP32(
static RTLIB::Libcall getFCmp64LibCall(CmpInst::Predicate Predicate,
CmpInst::Predicate &IPredicate) {
switch (Predicate) {
default:
llvm_unreachable("Unsupported FCmp predicate");
case CmpInst::FCMP_OEQ:
IPredicate = CmpInst::ICMP_EQ;
return RTLIB::OEQ_F64;
case CmpInst::FCMP_OGE:
IPredicate = CmpInst::ICMP_SGE;
return RTLIB::OGE_F64;
case CmpInst::FCMP_OGT:
IPredicate = CmpInst::ICMP_SGT;
return RTLIB::OGT_F64;
case CmpInst::FCMP_OLE:
IPredicate = CmpInst::ICMP_SLE;
return RTLIB::OLE_F64;
case CmpInst::FCMP_OLT:
IPredicate = CmpInst::ICMP_SLT;
return RTLIB::OLT_F64;
case CmpInst::FCMP_ORD:
IPredicate = CmpInst::ICMP_EQ;
return RTLIB::UO_F64;
/* Unordered comparisons are built from
* the complement of the ordered ones */
case CmpInst::FCMP_UGE:
IPredicate = CmpInst::ICMP_SGE;
return RTLIB::OLT_F64;
case CmpInst::FCMP_UGT:
IPredicate = CmpInst::ICMP_SGT;
return RTLIB::OLE_F64;
case CmpInst::FCMP_ULE:
IPredicate = CmpInst::ICMP_SLE;
return RTLIB::OGT_F64;
case CmpInst::FCMP_ULT:
IPredicate = CmpInst::ICMP_SLT;
return RTLIB::OGE_F64;
case CmpInst::FCMP_UNE:
IPredicate = CmpInst::ICMP_NE;
return RTLIB::UNE_F64;
case CmpInst::FCMP_UNO:
IPredicate = CmpInst::ICMP_NE;
return RTLIB::UO_F64;
}
}

bool AIELegalizerHelper::legalizeG_FCMP_FP32_FP64(
LegalizerHelper &Helper, MachineInstr &MI,
const CmpInst::Predicate FPredicate,
LostDebugLocObserver &LocObserver) const {
LostDebugLocObserver &LocObserver,
int ArgSize) const {
MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
MachineRegisterInfo &MRI = *MIRBuilder.getMRI();
assert(MRI.getType(MI.getOperand(2).getReg()) == LLT::scalar(32) &&
MRI.getType(MI.getOperand(3).getReg()) == LLT::scalar(32) &&
"Expected single precision floating point operands!");
assert(MRI.getType(MI.getOperand(2).getReg()) == LLT::scalar(ArgSize) &&
MRI.getType(MI.getOperand(3).getReg()) == LLT::scalar(ArgSize) &&
"Expected single or double precision floating point operands!");

LLVMContext &Ctx = MIRBuilder.getMF().getFunction().getContext();
auto DstReg = MI.getOperand(0).getReg();
bool Is32 = ArgSize == 32;

RTLIB::Libcall Libcall = RTLIB::UNKNOWN_LIBCALL;
CmpInst::Predicate IPredicate = CmpInst::BAD_ICMP_PREDICATE;
Expand All @@ -871,21 +919,21 @@ bool AIELegalizerHelper::legalizeG_FCMP_FP32(
/* Compose ordered and unequal operands as follows:
* a != b ==> a > b || a < b */
case CmpInst::FCMP_ONE:
Libcalls.push_back(std::make_pair(RTLIB::OGT_F32, CmpInst::ICMP_SGT));
Libcalls.push_back(std::make_pair(RTLIB::OLT_F32, CmpInst::ICMP_SLT));
Libcalls.push_back(std::make_pair(Is32 ? RTLIB::OGT_F32 : RTLIB::OGT_F64, CmpInst::ICMP_SGT));
Libcalls.push_back(std::make_pair(Is32 ? RTLIB::OLT_F32 : RTLIB::OLT_F64, CmpInst::ICMP_SLT));
break;
/* Compose unordered or equal operands as follows:
* (unord(a, b) or a == b) ==> a == b || a != b */
case CmpInst::FCMP_UEQ:
Libcalls.push_back(std::make_pair(RTLIB::OEQ_F32, CmpInst::ICMP_EQ));
Libcalls.push_back(std::make_pair(RTLIB::UO_F32, CmpInst::ICMP_NE));
Libcalls.push_back(std::make_pair(Is32 ? RTLIB::OEQ_F32 : RTLIB::OEQ_F64, CmpInst::ICMP_EQ));
Libcalls.push_back(std::make_pair(Is32 ? RTLIB::UO_F32 : RTLIB::UO_F64, CmpInst::ICMP_NE));
break;
default:
Libcall = getFCmpLibCall(FPredicate, IPredicate);
Libcall = Is32 ? getFCmp32LibCall(FPredicate, IPredicate) : getFCmp64LibCall(FPredicate, IPredicate);
Libcalls.push_back(std::make_pair(Libcall, IPredicate));
break;
}
auto *ArgTy = Type::getFloatTy(Ctx);
auto *ArgTy = Is32 ? Type::getFloatTy(Ctx) : Type::getDoubleTy(Ctx);
auto *RetTy = Type::getInt32Ty(Ctx);

SmallVector<Register, 2> Results;
Expand All @@ -911,6 +959,7 @@ bool AIELegalizerHelper::legalizeG_FCMP_FP32(
MIRBuilder.buildICmp(IDestPred, NewDstReg, LibcallResult, Zero);
Results.push_back(NewDstReg);
}

// OR the results when we have two libcalls
if (Results.size() != 1) {
assert(Results.size() == 2 && "Unexpected Number of Results");
Expand Down Expand Up @@ -1006,8 +1055,8 @@ bool AIELegalizerHelper::legalizeG_FCMP(
}

const unsigned LHSSize = MRI.getType(LHS).getSizeInBits();
if (LHSSize == 32)
return legalizeG_FCMP_FP32(Helper, MI, FPred, LocObserver);
if (LHSSize == 32 || LHSSize == 64)
return legalizeG_FCMP_FP32_FP64(Helper, MI, FPred, LocObserver, LHSSize);

assert(LHSSize == 16 && "Expected bf16 operands for FCmp");

Expand Down
7 changes: 4 additions & 3 deletions llvm/lib/Target/AIE/AIELegalizerHelper.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,9 +63,10 @@ class AIELegalizerHelper {
MachineInstr &MI) const;
bool legalizeG_FCMP(LegalizerHelper &Helper, MachineInstr &MI,
LostDebugLocObserver &LocObserver) const;
bool legalizeG_FCMP_FP32(LegalizerHelper &Helper, MachineInstr &MI,
const CmpInst::Predicate FPredicate,
LostDebugLocObserver &LocObserver) const;
bool legalizeG_FCMP_FP32_FP64(LegalizerHelper &Helper, MachineInstr &MI,
const CmpInst::Predicate FPredicate,
LostDebugLocObserver &LocObserver,
int ArgSize) const;
bool legalizeG_FPTRUNC(LegalizerHelper &Helper, MachineInstr &MI) const;
bool legalizeG_FPEXT(LegalizerHelper &Helper, MachineInstr &MI) const;
bool legalizeG_FABS(LegalizerHelper &Helper, MachineInstr &MI) const;
Expand Down
2 changes: 1 addition & 1 deletion llvm/lib/Target/AIE/aie2p/AIE2PLegalizerInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -204,7 +204,7 @@ AIE2PLegalizerInfo::AIE2PLegalizerInfo(const AIE2PSubtarget &ST)

getActionDefinitionsBuilder(G_FCMP)
.clampScalar(0, S32, S32)
.customFor({{S32, S16}, {S32, S32}});
.customFor({{S32, S16}, {S32, S32}, {S32, S64}});

getActionDefinitionsBuilder(G_FPTRUNC)
.libcallFor({{S32, S64}})
Expand Down
Loading