#include "R600Defines.h"
#include "R600MachineFunctionInfo.h"
#include "R600RegisterInfo.h"
-#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#define GET_INSTRINFO_CTOR
R600InstrInfo::R600InstrInfo(AMDGPUTargetMachine &tm)
: AMDGPUInstrInfo(tm),
- RI(tm, *this),
+ RI(tm),
ST(tm.getSubtarget<AMDGPUSubtarget>())
{ }
MachineInstr *NewMI = buildDefaultInstruction(MBB, MI, AMDGPU::MOV,
DestReg, SrcReg);
- NewMI->getOperand(getOperandIdx(*NewMI, R600Operands::SRC0))
+ NewMI->getOperand(getOperandIdx(*NewMI, AMDGPU::OpName::src0))
.setIsKill(KillSrc);
}
}
}
bool R600InstrInfo::isReductionOp(unsigned Opcode) const {
- switch(Opcode) {
- default: return false;
- case AMDGPU::DOT4_r600_pseudo:
- case AMDGPU::DOT4_eg_pseudo:
- return true;
- }
+ return false;
}
bool R600InstrInfo::isCubeOp(unsigned Opcode) const {
bool R600InstrInfo::isALUInstr(unsigned Opcode) const {
unsigned TargetFlags = get(Opcode).TSFlags;
+ return (TargetFlags & R600_InstFlag::ALU_INST);
+}
+
+bool R600InstrInfo::hasInstrModifiers(unsigned Opcode) const {
+ unsigned TargetFlags = get(Opcode).TSFlags;
+
return ((TargetFlags & R600_InstFlag::OP1) |
(TargetFlags & R600_InstFlag::OP2) |
(TargetFlags & R600_InstFlag::OP3));
}
+bool R600InstrInfo::isLDSInstr(unsigned Opcode) const {
+ unsigned TargetFlags = get(Opcode).TSFlags;
+
+ return ((TargetFlags & R600_InstFlag::LDS_1A) |
+ (TargetFlags & R600_InstFlag::LDS_1A1D));
+}
+
bool R600InstrInfo::isTransOnly(unsigned Opcode) const {
return (get(Opcode).TSFlags & R600_InstFlag::TRANS_ONLY);
}
}
bool R600InstrInfo::usesVertexCache(unsigned Opcode) const {
- return ST.hasVertexCache() && get(Opcode).TSFlags & R600_InstFlag::VTX_INST;
+ return ST.hasVertexCache() && IS_VTX(get(Opcode));
}
bool R600InstrInfo::usesVertexCache(const MachineInstr *MI) const {
}
bool R600InstrInfo::usesTextureCache(unsigned Opcode) const {
- return (!ST.hasVertexCache() && get(Opcode).TSFlags & R600_InstFlag::VTX_INST) ||
- (get(Opcode).TSFlags & R600_InstFlag::TEX_INST);
+ return (!ST.hasVertexCache() && IS_VTX(get(Opcode))) || IS_TEX(get(Opcode));
}
bool R600InstrInfo::usesTextureCache(const MachineInstr *MI) const {
usesTextureCache(MI->getOpcode());
}
+bool R600InstrInfo::mustBeLastInClause(unsigned Opcode) const {
+ switch (Opcode) {
+ case AMDGPU::KILLGT:
+ case AMDGPU::GROUP_BARRIER:
+ return true;
+ default:
+ return false;
+ }
+}
+
+SmallVector<std::pair<MachineOperand *, int64_t>, 3>
+R600InstrInfo::getSrcs(MachineInstr *MI) const {
+ SmallVector<std::pair<MachineOperand *, int64_t>, 3> Result;
+
+ if (MI->getOpcode() == AMDGPU::DOT_4) {
+ static const unsigned OpTable[8][2] = {
+ {AMDGPU::OpName::src0_X, AMDGPU::OpName::src0_sel_X},
+ {AMDGPU::OpName::src0_Y, AMDGPU::OpName::src0_sel_Y},
+ {AMDGPU::OpName::src0_Z, AMDGPU::OpName::src0_sel_Z},
+ {AMDGPU::OpName::src0_W, AMDGPU::OpName::src0_sel_W},
+ {AMDGPU::OpName::src1_X, AMDGPU::OpName::src1_sel_X},
+ {AMDGPU::OpName::src1_Y, AMDGPU::OpName::src1_sel_Y},
+ {AMDGPU::OpName::src1_Z, AMDGPU::OpName::src1_sel_Z},
+ {AMDGPU::OpName::src1_W, AMDGPU::OpName::src1_sel_W},
+ };
+
+ for (unsigned j = 0; j < 8; j++) {
+ MachineOperand &MO = MI->getOperand(getOperandIdx(MI->getOpcode(),
+ OpTable[j][0]));
+ unsigned Reg = MO.getReg();
+ if (Reg == AMDGPU::ALU_CONST) {
+ unsigned Sel = MI->getOperand(getOperandIdx(MI->getOpcode(),
+ OpTable[j][1])).getImm();
+ Result.push_back(std::pair<MachineOperand *, int64_t>(&MO, Sel));
+ continue;
+ }
+
+ }
+ return Result;
+ }
+
+ static const unsigned OpTable[3][2] = {
+ {AMDGPU::OpName::src0, AMDGPU::OpName::src0_sel},
+ {AMDGPU::OpName::src1, AMDGPU::OpName::src1_sel},
+ {AMDGPU::OpName::src2, AMDGPU::OpName::src2_sel},
+ };
+
+ for (unsigned j = 0; j < 3; j++) {
+ int SrcIdx = getOperandIdx(MI->getOpcode(), OpTable[j][0]);
+ if (SrcIdx < 0)
+ break;
+ MachineOperand &MO = MI->getOperand(SrcIdx);
+ unsigned Reg = MI->getOperand(SrcIdx).getReg();
+ if (Reg == AMDGPU::ALU_CONST) {
+ unsigned Sel = MI->getOperand(
+ getOperandIdx(MI->getOpcode(), OpTable[j][1])).getImm();
+ Result.push_back(std::pair<MachineOperand *, int64_t>(&MO, Sel));
+ continue;
+ }
+ if (Reg == AMDGPU::ALU_LITERAL_X) {
+ unsigned Imm = MI->getOperand(
+ getOperandIdx(MI->getOpcode(), AMDGPU::OpName::literal)).getImm();
+ Result.push_back(std::pair<MachineOperand *, int64_t>(&MO, Imm));
+ continue;
+ }
+ Result.push_back(std::pair<MachineOperand *, int64_t>(&MO, 0));
+ }
+ return Result;
+}
+
+std::vector<std::pair<int, unsigned> >
+R600InstrInfo::ExtractSrcs(MachineInstr *MI,
+ const DenseMap<unsigned, unsigned> &PV,
+ unsigned &ConstCount) const {
+ ConstCount = 0;
+ const SmallVector<std::pair<MachineOperand *, int64_t>, 3> Srcs = getSrcs(MI);
+ const std::pair<int, unsigned> DummyPair(-1, 0);
+ std::vector<std::pair<int, unsigned> > Result;
+ unsigned i = 0;
+ for (unsigned n = Srcs.size(); i < n; ++i) {
+ unsigned Reg = Srcs[i].first->getReg();
+ unsigned Index = RI.getEncodingValue(Reg) & 0xff;
+ if (Reg == AMDGPU::OQAP) {
+ Result.push_back(std::pair<int, unsigned>(Index, 0));
+ }
+ if (PV.find(Reg) != PV.end()) {
+ // 255 is used to tells its a PS/PV reg
+ Result.push_back(std::pair<int, unsigned>(255, 0));
+ continue;
+ }
+ if (Index > 127) {
+ ConstCount++;
+ Result.push_back(DummyPair);
+ continue;
+ }
+ unsigned Chan = RI.getHWRegChan(Reg);
+ Result.push_back(std::pair<int, unsigned>(Index, Chan));
+ }
+ for (; i < 3; ++i)
+ Result.push_back(DummyPair);
+ return Result;
+}
+
+static std::vector<std::pair<int, unsigned> >
+Swizzle(std::vector<std::pair<int, unsigned> > Src,
+ R600InstrInfo::BankSwizzle Swz) {
+ switch (Swz) {
+ case R600InstrInfo::ALU_VEC_012_SCL_210:
+ break;
+ case R600InstrInfo::ALU_VEC_021_SCL_122:
+ std::swap(Src[1], Src[2]);
+ break;
+ case R600InstrInfo::ALU_VEC_102_SCL_221:
+ std::swap(Src[0], Src[1]);
+ break;
+ case R600InstrInfo::ALU_VEC_120_SCL_212:
+ std::swap(Src[0], Src[1]);
+ std::swap(Src[0], Src[2]);
+ break;
+ case R600InstrInfo::ALU_VEC_201:
+ std::swap(Src[0], Src[2]);
+ std::swap(Src[0], Src[1]);
+ break;
+ case R600InstrInfo::ALU_VEC_210:
+ std::swap(Src[0], Src[2]);
+ break;
+ }
+ return Src;
+}
+
+static unsigned
+getTransSwizzle(R600InstrInfo::BankSwizzle Swz, unsigned Op) {
+ switch (Swz) {
+ case R600InstrInfo::ALU_VEC_012_SCL_210: {
+ unsigned Cycles[3] = { 2, 1, 0};
+ return Cycles[Op];
+ }
+ case R600InstrInfo::ALU_VEC_021_SCL_122: {
+ unsigned Cycles[3] = { 1, 2, 2};
+ return Cycles[Op];
+ }
+ case R600InstrInfo::ALU_VEC_120_SCL_212: {
+ unsigned Cycles[3] = { 2, 1, 2};
+ return Cycles[Op];
+ }
+ case R600InstrInfo::ALU_VEC_102_SCL_221: {
+ unsigned Cycles[3] = { 2, 2, 1};
+ return Cycles[Op];
+ }
+ default:
+ llvm_unreachable("Wrong Swizzle for Trans Slot");
+ return 0;
+ }
+}
+
+/// returns how many MIs (whose inputs are represented by IGSrcs) can be packed
+/// in the same Instruction Group while meeting read port limitations given a
+/// Swz swizzle sequence.
+unsigned R600InstrInfo::isLegalUpTo(
+ const std::vector<std::vector<std::pair<int, unsigned> > > &IGSrcs,
+ const std::vector<R600InstrInfo::BankSwizzle> &Swz,
+ const std::vector<std::pair<int, unsigned> > &TransSrcs,
+ R600InstrInfo::BankSwizzle TransSwz) const {
+ int Vector[4][3];
+ memset(Vector, -1, sizeof(Vector));
+ for (unsigned i = 0, e = IGSrcs.size(); i < e; i++) {
+ const std::vector<std::pair<int, unsigned> > &Srcs =
+ Swizzle(IGSrcs[i], Swz[i]);
+ for (unsigned j = 0; j < 3; j++) {
+ const std::pair<int, unsigned> &Src = Srcs[j];
+ if (Src.first < 0 || Src.first == 255)
+ continue;
+ if (Src.first == GET_REG_INDEX(RI.getEncodingValue(AMDGPU::OQAP))) {
+ if (Swz[i] != R600InstrInfo::ALU_VEC_012_SCL_210 &&
+ Swz[i] != R600InstrInfo::ALU_VEC_021_SCL_122) {
+ // The value from output queue A (denoted by register OQAP) can
+ // only be fetched during the first cycle.
+ return false;
+ }
+ // OQAP does not count towards the normal read port restrictions
+ continue;
+ }
+ if (Vector[Src.second][j] < 0)
+ Vector[Src.second][j] = Src.first;
+ if (Vector[Src.second][j] != Src.first)
+ return i;
+ }
+ }
+ // Now check Trans Alu
+ for (unsigned i = 0, e = TransSrcs.size(); i < e; ++i) {
+ const std::pair<int, unsigned> &Src = TransSrcs[i];
+ unsigned Cycle = getTransSwizzle(TransSwz, i);
+ if (Src.first < 0)
+ continue;
+ if (Src.first == 255)
+ continue;
+ if (Vector[Src.second][Cycle] < 0)
+ Vector[Src.second][Cycle] = Src.first;
+ if (Vector[Src.second][Cycle] != Src.first)
+ return IGSrcs.size() - 1;
+ }
+ return IGSrcs.size();
+}
+
+/// Given a swizzle sequence SwzCandidate and an index Idx, returns the next
+/// (in lexicographic term) swizzle sequence assuming that all swizzles after
+/// Idx can be skipped
+static bool
+NextPossibleSolution(
+ std::vector<R600InstrInfo::BankSwizzle> &SwzCandidate,
+ unsigned Idx) {
+ assert(Idx < SwzCandidate.size());
+ int ResetIdx = Idx;
+ while (ResetIdx > -1 && SwzCandidate[ResetIdx] == R600InstrInfo::ALU_VEC_210)
+ ResetIdx --;
+ for (unsigned i = ResetIdx + 1, e = SwzCandidate.size(); i < e; i++) {
+ SwzCandidate[i] = R600InstrInfo::ALU_VEC_012_SCL_210;
+ }
+ if (ResetIdx == -1)
+ return false;
+ int NextSwizzle = SwzCandidate[ResetIdx] + 1;
+ SwzCandidate[ResetIdx] = (R600InstrInfo::BankSwizzle)NextSwizzle;
+ return true;
+}
+
+/// Enumerate all possible Swizzle sequence to find one that can meet all
+/// read port requirements.
+bool R600InstrInfo::FindSwizzleForVectorSlot(
+ const std::vector<std::vector<std::pair<int, unsigned> > > &IGSrcs,
+ std::vector<R600InstrInfo::BankSwizzle> &SwzCandidate,
+ const std::vector<std::pair<int, unsigned> > &TransSrcs,
+ R600InstrInfo::BankSwizzle TransSwz) const {
+ unsigned ValidUpTo = 0;
+ do {
+ ValidUpTo = isLegalUpTo(IGSrcs, SwzCandidate, TransSrcs, TransSwz);
+ if (ValidUpTo == IGSrcs.size())
+ return true;
+ } while (NextPossibleSolution(SwzCandidate, ValidUpTo));
+ return false;
+}
+
+/// Instructions in Trans slot can't read gpr at cycle 0 if they also read
+/// a const, and can't read a gpr at cycle 1 if they read 2 const.
+static bool
+isConstCompatible(R600InstrInfo::BankSwizzle TransSwz,
+ const std::vector<std::pair<int, unsigned> > &TransOps,
+ unsigned ConstCount) {
+ for (unsigned i = 0, e = TransOps.size(); i < e; ++i) {
+ const std::pair<int, unsigned> &Src = TransOps[i];
+ unsigned Cycle = getTransSwizzle(TransSwz, i);
+ if (Src.first < 0)
+ continue;
+ if (ConstCount > 0 && Cycle == 0)
+ return false;
+ if (ConstCount > 1 && Cycle == 1)
+ return false;
+ }
+ return true;
+}
+
+bool
+R600InstrInfo::fitsReadPortLimitations(const std::vector<MachineInstr *> &IG,
+ const DenseMap<unsigned, unsigned> &PV,
+ std::vector<BankSwizzle> &ValidSwizzle,
+ bool isLastAluTrans)
+ const {
+ //Todo : support shared src0 - src1 operand
+
+ std::vector<std::vector<std::pair<int, unsigned> > > IGSrcs;
+ ValidSwizzle.clear();
+ unsigned ConstCount;
+ BankSwizzle TransBS = ALU_VEC_012_SCL_210;
+ for (unsigned i = 0, e = IG.size(); i < e; ++i) {
+ IGSrcs.push_back(ExtractSrcs(IG[i], PV, ConstCount));
+ unsigned Op = getOperandIdx(IG[i]->getOpcode(),
+ AMDGPU::OpName::bank_swizzle);
+ ValidSwizzle.push_back( (R600InstrInfo::BankSwizzle)
+ IG[i]->getOperand(Op).getImm());
+ }
+ std::vector<std::pair<int, unsigned> > TransOps;
+ if (!isLastAluTrans)
+ return FindSwizzleForVectorSlot(IGSrcs, ValidSwizzle, TransOps, TransBS);
+
+ TransOps = IGSrcs.back();
+ IGSrcs.pop_back();
+ ValidSwizzle.pop_back();
+
+ static const R600InstrInfo::BankSwizzle TransSwz[] = {
+ ALU_VEC_012_SCL_210,
+ ALU_VEC_021_SCL_122,
+ ALU_VEC_120_SCL_212,
+ ALU_VEC_102_SCL_221
+ };
+ for (unsigned i = 0; i < 4; i++) {
+ TransBS = TransSwz[i];
+ if (!isConstCompatible(TransBS, TransOps, ConstCount))
+ continue;
+ bool Result = FindSwizzleForVectorSlot(IGSrcs, ValidSwizzle, TransOps,
+ TransBS);
+ if (Result) {
+ ValidSwizzle.push_back(TransBS);
+ return true;
+ }
+ }
+
+ return false;
+}
+
+
bool
R600InstrInfo::fitsConstReadLimitations(const std::vector<unsigned> &Consts)
const {
}
bool
-R600InstrInfo::canBundle(const std::vector<MachineInstr *> &MIs) const {
+R600InstrInfo::fitsConstReadLimitations(const std::vector<MachineInstr *> &MIs)
+ const {
std::vector<unsigned> Consts;
for (unsigned i = 0, n = MIs.size(); i < n; i++) {
- const MachineInstr *MI = MIs[i];
-
- const R600Operands::Ops OpTable[3][2] = {
- {R600Operands::SRC0, R600Operands::SRC0_SEL},
- {R600Operands::SRC1, R600Operands::SRC1_SEL},
- {R600Operands::SRC2, R600Operands::SRC2_SEL},
- };
-
+ MachineInstr *MI = MIs[i];
if (!isALUInstr(MI->getOpcode()))
continue;
- for (unsigned j = 0; j < 3; j++) {
- int SrcIdx = getOperandIdx(MI->getOpcode(), OpTable[j][0]);
- if (SrcIdx < 0)
- break;
- unsigned Reg = MI->getOperand(SrcIdx).getReg();
- if (Reg == AMDGPU::ALU_CONST) {
- unsigned Const = MI->getOperand(
- getOperandIdx(MI->getOpcode(), OpTable[j][1])).getImm();
- Consts.push_back(Const);
- continue;
- }
- if (AMDGPU::R600_KC0RegClass.contains(Reg) ||
- AMDGPU::R600_KC1RegClass.contains(Reg)) {
- unsigned Index = RI.getEncodingValue(Reg) & 0xff;
- unsigned Chan = RI.getHWRegChan(Reg);
+ const SmallVector<std::pair<MachineOperand *, int64_t>, 3> &Srcs =
+ getSrcs(MI);
+
+ for (unsigned j = 0, e = Srcs.size(); j < e; j++) {
+ std::pair<MachineOperand *, unsigned> Src = Srcs[j];
+ if (Src.first->getReg() == AMDGPU::ALU_CONST)
+ Consts.push_back(Src.second);
+ if (AMDGPU::R600_KC0RegClass.contains(Src.first->getReg()) ||
+ AMDGPU::R600_KC1RegClass.contains(Src.first->getReg())) {
+ unsigned Index = RI.getEncodingValue(Src.first->getReg()) & 0xff;
+ unsigned Chan = RI.getHWRegChan(Src.first->getReg());
Consts.push_back((Index << 2) | Chan);
- continue;
}
}
}
};
}
+static
+MachineBasicBlock::iterator FindLastAluClause(MachineBasicBlock &MBB) {
+ for (MachineBasicBlock::reverse_iterator It = MBB.rbegin(), E = MBB.rend();
+ It != E; ++It) {
+ if (It->getOpcode() == AMDGPU::CF_ALU ||
+ It->getOpcode() == AMDGPU::CF_ALU_PUSH_BEFORE)
+ return llvm::prior(It.base());
+ }
+ return MBB.end();
+}
+
unsigned
R600InstrInfo::InsertBranch(MachineBasicBlock &MBB,
MachineBasicBlock *TBB,
BuildMI(&MBB, DL, get(AMDGPU::JUMP_COND))
.addMBB(TBB)
.addReg(AMDGPU::PREDICATE_BIT, RegState::Kill);
+ MachineBasicBlock::iterator CfAlu = FindLastAluClause(MBB);
+ if (CfAlu == MBB.end())
+ return 1;
+ assert (CfAlu->getOpcode() == AMDGPU::CF_ALU);
+ CfAlu->setDesc(get(AMDGPU::CF_ALU_PUSH_BEFORE));
return 1;
}
} else {
.addMBB(TBB)
.addReg(AMDGPU::PREDICATE_BIT, RegState::Kill);
BuildMI(&MBB, DL, get(AMDGPU::JUMP)).addMBB(FBB);
+ MachineBasicBlock::iterator CfAlu = FindLastAluClause(MBB);
+ if (CfAlu == MBB.end())
+ return 2;
+ assert (CfAlu->getOpcode() == AMDGPU::CF_ALU);
+ CfAlu->setDesc(get(AMDGPU::CF_ALU_PUSH_BEFORE));
return 2;
}
}
MachineInstr *predSet = findFirstPredicateSetterFrom(MBB, I);
clearFlag(predSet, 0, MO_FLAG_PUSH);
I->eraseFromParent();
+ MachineBasicBlock::iterator CfAlu = FindLastAluClause(MBB);
+ if (CfAlu == MBB.end())
+ break;
+ assert (CfAlu->getOpcode() == AMDGPU::CF_ALU_PUSH_BEFORE);
+ CfAlu->setDesc(get(AMDGPU::CF_ALU));
break;
}
case AMDGPU::JUMP:
MachineInstr *predSet = findFirstPredicateSetterFrom(MBB, I);
clearFlag(predSet, 0, MO_FLAG_PUSH);
I->eraseFromParent();
+ MachineBasicBlock::iterator CfAlu = FindLastAluClause(MBB);
+ if (CfAlu == MBB.end())
+ break;
+ assert (CfAlu->getOpcode() == AMDGPU::CF_ALU_PUSH_BEFORE);
+ CfAlu->setDesc(get(AMDGPU::CF_ALU));
break;
}
case AMDGPU::JUMP:
if (MI->getOpcode() == AMDGPU::KILLGT) {
return false;
+ } else if (MI->getOpcode() == AMDGPU::CF_ALU) {
+ // If the clause start in the middle of MBB then the MBB has more
+ // than a single clause, unable to predicate several clauses.
+ if (MI->getParent()->begin() != MachineBasicBlock::iterator(MI))
+ return false;
+ // TODO: We don't support KC merging atm
+ if (MI->getOperand(3).getImm() != 0 || MI->getOperand(4).getImm() != 0)
+ return false;
+ return true;
} else if (isVector(*MI)) {
return false;
} else {
const SmallVectorImpl<MachineOperand> &Pred) const {
int PIdx = MI->findFirstPredOperandIdx();
+ if (MI->getOpcode() == AMDGPU::CF_ALU) {
+ MI->getOperand(8).setImm(0);
+ return true;
+ }
+
if (PIdx != -1) {
MachineOperand &PMO = MI->getOperand(PIdx);
PMO.setReg(Pred[2].getReg());
unsigned AddrReg = AMDGPU::R600_AddrRegClass.getRegister(Address);
MachineInstr *MOVA = buildDefaultInstruction(*MBB, I, AMDGPU::MOVA_INT_eg,
AMDGPU::AR_X, OffsetReg);
- setImmOperand(MOVA, R600Operands::WRITE, 0);
+ setImmOperand(MOVA, AMDGPU::OpName::write, 0);
MachineInstrBuilder Mov = buildDefaultInstruction(*MBB, I, AMDGPU::MOV,
AddrReg, ValueReg)
- .addReg(AMDGPU::AR_X, RegState::Implicit);
- setImmOperand(Mov, R600Operands::DST_REL, 1);
+ .addReg(AMDGPU::AR_X,
+ RegState::Implicit | RegState::Kill);
+ setImmOperand(Mov, AMDGPU::OpName::dst_rel, 1);
return Mov;
}
MachineInstr *MOVA = buildDefaultInstruction(*MBB, I, AMDGPU::MOVA_INT_eg,
AMDGPU::AR_X,
OffsetReg);
- setImmOperand(MOVA, R600Operands::WRITE, 0);
+ setImmOperand(MOVA, AMDGPU::OpName::write, 0);
MachineInstrBuilder Mov = buildDefaultInstruction(*MBB, I, AMDGPU::MOV,
ValueReg,
AddrReg)
- .addReg(AMDGPU::AR_X, RegState::Implicit);
- setImmOperand(Mov, R600Operands::SRC0_REL, 1);
+ .addReg(AMDGPU::AR_X,
+ RegState::Implicit | RegState::Kill);
+ setImmOperand(Mov, AMDGPU::OpName::src0_rel, 1);
return Mov;
}
return MIB;
}
+#define OPERAND_CASE(Label) \
+ case Label: { \
+ static const unsigned Ops[] = \
+ { \
+ Label##_X, \
+ Label##_Y, \
+ Label##_Z, \
+ Label##_W \
+ }; \
+ return Ops[Slot]; \
+ }
+
+static unsigned getSlotedOps(unsigned Op, unsigned Slot) {
+ switch (Op) {
+ OPERAND_CASE(AMDGPU::OpName::update_exec_mask)
+ OPERAND_CASE(AMDGPU::OpName::update_pred)
+ OPERAND_CASE(AMDGPU::OpName::write)
+ OPERAND_CASE(AMDGPU::OpName::omod)
+ OPERAND_CASE(AMDGPU::OpName::dst_rel)
+ OPERAND_CASE(AMDGPU::OpName::clamp)
+ OPERAND_CASE(AMDGPU::OpName::src0)
+ OPERAND_CASE(AMDGPU::OpName::src0_neg)
+ OPERAND_CASE(AMDGPU::OpName::src0_rel)
+ OPERAND_CASE(AMDGPU::OpName::src0_abs)
+ OPERAND_CASE(AMDGPU::OpName::src0_sel)
+ OPERAND_CASE(AMDGPU::OpName::src1)
+ OPERAND_CASE(AMDGPU::OpName::src1_neg)
+ OPERAND_CASE(AMDGPU::OpName::src1_rel)
+ OPERAND_CASE(AMDGPU::OpName::src1_abs)
+ OPERAND_CASE(AMDGPU::OpName::src1_sel)
+ OPERAND_CASE(AMDGPU::OpName::pred_sel)
+ default:
+ llvm_unreachable("Wrong Operand");
+ }
+}
+
+#undef OPERAND_CASE
+
+MachineInstr *R600InstrInfo::buildSlotOfVectorInstruction(
+ MachineBasicBlock &MBB, MachineInstr *MI, unsigned Slot, unsigned DstReg)
+ const {
+ assert (MI->getOpcode() == AMDGPU::DOT_4 && "Not Implemented");
+ unsigned Opcode;
+ const AMDGPUSubtarget &ST = TM.getSubtarget<AMDGPUSubtarget>();
+ if (ST.getGeneration() <= AMDGPUSubtarget::R700)
+ Opcode = AMDGPU::DOT4_r600;
+ else
+ Opcode = AMDGPU::DOT4_eg;
+ MachineBasicBlock::iterator I = MI;
+ MachineOperand &Src0 = MI->getOperand(
+ getOperandIdx(MI->getOpcode(), getSlotedOps(AMDGPU::OpName::src0, Slot)));
+ MachineOperand &Src1 = MI->getOperand(
+ getOperandIdx(MI->getOpcode(), getSlotedOps(AMDGPU::OpName::src1, Slot)));
+ MachineInstr *MIB = buildDefaultInstruction(
+ MBB, I, Opcode, DstReg, Src0.getReg(), Src1.getReg());
+ static const unsigned Operands[14] = {
+ AMDGPU::OpName::update_exec_mask,
+ AMDGPU::OpName::update_pred,
+ AMDGPU::OpName::write,
+ AMDGPU::OpName::omod,
+ AMDGPU::OpName::dst_rel,
+ AMDGPU::OpName::clamp,
+ AMDGPU::OpName::src0_neg,
+ AMDGPU::OpName::src0_rel,
+ AMDGPU::OpName::src0_abs,
+ AMDGPU::OpName::src0_sel,
+ AMDGPU::OpName::src1_neg,
+ AMDGPU::OpName::src1_rel,
+ AMDGPU::OpName::src1_abs,
+ AMDGPU::OpName::src1_sel,
+ };
+
+ for (unsigned i = 0; i < 14; i++) {
+ MachineOperand &MO = MI->getOperand(
+ getOperandIdx(MI->getOpcode(), getSlotedOps(Operands[i], Slot)));
+ assert (MO.isImm());
+ setImmOperand(MIB, Operands[i], MO.getImm());
+ }
+ MIB->getOperand(20).setImm(0);
+ return MIB;
+}
+
MachineInstr *R600InstrInfo::buildMovImm(MachineBasicBlock &BB,
MachineBasicBlock::iterator I,
unsigned DstReg,
uint64_t Imm) const {
MachineInstr *MovImm = buildDefaultInstruction(BB, I, AMDGPU::MOV, DstReg,
AMDGPU::ALU_LITERAL_X);
- setImmOperand(MovImm, R600Operands::IMM, Imm);
+ setImmOperand(MovImm, AMDGPU::OpName::literal, Imm);
return MovImm;
}
-int R600InstrInfo::getOperandIdx(const MachineInstr &MI,
- R600Operands::Ops Op) const {
+int R600InstrInfo::getOperandIdx(const MachineInstr &MI, unsigned Op) const {
return getOperandIdx(MI.getOpcode(), Op);
}
-int R600InstrInfo::getOperandIdx(unsigned Opcode,
- R600Operands::Ops Op) const {
- unsigned TargetFlags = get(Opcode).TSFlags;
- unsigned OpTableIdx;
-
- if (!HAS_NATIVE_OPERANDS(TargetFlags)) {
- switch (Op) {
- case R600Operands::DST: return 0;
- case R600Operands::SRC0: return 1;
- case R600Operands::SRC1: return 2;
- case R600Operands::SRC2: return 3;
- default:
- assert(!"Unknown operand type for instruction");
- return -1;
- }
- }
-
- if (TargetFlags & R600_InstFlag::OP1) {
- OpTableIdx = 0;
- } else if (TargetFlags & R600_InstFlag::OP2) {
- OpTableIdx = 1;
- } else {
- assert((TargetFlags & R600_InstFlag::OP3) && "OP1, OP2, or OP3 not defined "
- "for this instruction");
- OpTableIdx = 2;
- }
-
- return R600Operands::ALUOpTable[OpTableIdx][Op];
+int R600InstrInfo::getOperandIdx(unsigned Opcode, unsigned Op) const {
+ return AMDGPU::getNamedOperandIdx(Opcode, Op);
}
-void R600InstrInfo::setImmOperand(MachineInstr *MI, R600Operands::Ops Op,
+void R600InstrInfo::setImmOperand(MachineInstr *MI, unsigned Op,
int64_t Imm) const {
int Idx = getOperandIdx(*MI, Op);
assert(Idx != -1 && "Operand not supported for this instruction.");
bool IsOP3 = (TargetFlags & R600_InstFlag::OP3) == R600_InstFlag::OP3;
switch (Flag) {
case MO_FLAG_CLAMP:
- FlagIndex = getOperandIdx(*MI, R600Operands::CLAMP);
+ FlagIndex = getOperandIdx(*MI, AMDGPU::OpName::clamp);
break;
case MO_FLAG_MASK:
- FlagIndex = getOperandIdx(*MI, R600Operands::WRITE);
+ FlagIndex = getOperandIdx(*MI, AMDGPU::OpName::write);
break;
case MO_FLAG_NOT_LAST:
case MO_FLAG_LAST:
- FlagIndex = getOperandIdx(*MI, R600Operands::LAST);
+ FlagIndex = getOperandIdx(*MI, AMDGPU::OpName::last);
break;
case MO_FLAG_NEG:
switch (SrcIdx) {
- case 0: FlagIndex = getOperandIdx(*MI, R600Operands::SRC0_NEG); break;
- case 1: FlagIndex = getOperandIdx(*MI, R600Operands::SRC1_NEG); break;
- case 2: FlagIndex = getOperandIdx(*MI, R600Operands::SRC2_NEG); break;
+ case 0: FlagIndex = getOperandIdx(*MI, AMDGPU::OpName::src0_neg); break;
+ case 1: FlagIndex = getOperandIdx(*MI, AMDGPU::OpName::src1_neg); break;
+ case 2: FlagIndex = getOperandIdx(*MI, AMDGPU::OpName::src2_neg); break;
}
break;
"instructions.");
(void)IsOP3;
switch (SrcIdx) {
- case 0: FlagIndex = getOperandIdx(*MI, R600Operands::SRC0_ABS); break;
- case 1: FlagIndex = getOperandIdx(*MI, R600Operands::SRC1_ABS); break;
+ case 0: FlagIndex = getOperandIdx(*MI, AMDGPU::OpName::src0_abs); break;
+ case 1: FlagIndex = getOperandIdx(*MI, AMDGPU::OpName::src1_abs); break;
}
break;
projects / oota-llvm.git / blobdiff