#include "AMDGPUInstrInfo.h"
#include "AMDGPURegisterInfo.h"
#include "AMDGPUTargetMachine.h"
-#include "AMDIL.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
-#define GET_INSTRINFO_CTOR
+using namespace llvm;
+
+#define GET_INSTRINFO_CTOR_DTOR
+#define GET_INSTRINFO_NAMED_OPS
#define GET_INSTRMAP_INFO
#include "AMDGPUGenInstrInfo.inc"
-using namespace llvm;
+// Pin the vtable to this file.
+void AMDGPUInstrInfo::anchor() {}
-AMDGPUInstrInfo::AMDGPUInstrInfo(TargetMachine &tm)
- : AMDGPUGenInstrInfo(0,0), RI(tm, *this), TM(tm) { }
+AMDGPUInstrInfo::AMDGPUInstrInfo(const AMDGPUSubtarget &st)
+ : AMDGPUGenInstrInfo(-1,-1), RI(st), ST(st) { }
const AMDGPURegisterInfo &AMDGPUInstrInfo::getRegisterInfo() const {
return RI;
MachineBasicBlock::iterator &MBBI,
LiveVariables *LV) const {
// TODO: Implement this function
- return NULL;
+ return nullptr;
}
bool AMDGPUInstrInfo::getNextBranchInstr(MachineBasicBlock::iterator &iter,
MachineBasicBlock &MBB) const {
return false;
}
-MachineBasicBlock::iterator skipFlowControl(MachineBasicBlock *MBB) {
- MachineBasicBlock::iterator tmp = MBB->end();
- if (!MBB->size()) {
- return MBB->end();
- }
- while (--tmp) {
- if (tmp->getOpcode() == AMDGPU::ENDLOOP
- || tmp->getOpcode() == AMDGPU::ENDIF
- || tmp->getOpcode() == AMDGPU::ELSE) {
- if (tmp == MBB->begin()) {
- return tmp;
- } else {
- continue;
- }
- } else {
- return ++tmp;
- }
- }
- return MBB->end();
-}
-
void
AMDGPUInstrInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
int FrameIndex,
const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI) const {
- assert(!"Not Implemented");
+ llvm_unreachable("Not Implemented");
}
void
unsigned DestReg, int FrameIndex,
const TargetRegisterClass *RC,
const TargetRegisterInfo *TRI) const {
- assert(!"Not Implemented");
+ llvm_unreachable("Not Implemented");
}
+bool AMDGPUInstrInfo::expandPostRAPseudo (MachineBasicBlock::iterator MI) const {
+ MachineBasicBlock *MBB = MI->getParent();
+ int OffsetOpIdx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
+ AMDGPU::OpName::addr);
+ // addr is a custom operand with multiple MI operands, and only the
+ // first MI operand is given a name.
+ int RegOpIdx = OffsetOpIdx + 1;
+ int ChanOpIdx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
+ AMDGPU::OpName::chan);
+ if (isRegisterLoad(*MI)) {
+ int DstOpIdx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
+ AMDGPU::OpName::dst);
+ unsigned RegIndex = MI->getOperand(RegOpIdx).getImm();
+ unsigned Channel = MI->getOperand(ChanOpIdx).getImm();
+ unsigned Address = calculateIndirectAddress(RegIndex, Channel);
+ unsigned OffsetReg = MI->getOperand(OffsetOpIdx).getReg();
+ if (OffsetReg == AMDGPU::INDIRECT_BASE_ADDR) {
+ buildMovInstr(MBB, MI, MI->getOperand(DstOpIdx).getReg(),
+ getIndirectAddrRegClass()->getRegister(Address));
+ } else {
+ buildIndirectRead(MBB, MI, MI->getOperand(DstOpIdx).getReg(),
+ Address, OffsetReg);
+ }
+ } else if (isRegisterStore(*MI)) {
+ int ValOpIdx = AMDGPU::getNamedOperandIdx(MI->getOpcode(),
+ AMDGPU::OpName::val);
+ unsigned RegIndex = MI->getOperand(RegOpIdx).getImm();
+ unsigned Channel = MI->getOperand(ChanOpIdx).getImm();
+ unsigned Address = calculateIndirectAddress(RegIndex, Channel);
+ unsigned OffsetReg = MI->getOperand(OffsetOpIdx).getReg();
+ if (OffsetReg == AMDGPU::INDIRECT_BASE_ADDR) {
+ buildMovInstr(MBB, MI, getIndirectAddrRegClass()->getRegister(Address),
+ MI->getOperand(ValOpIdx).getReg());
+ } else {
+ buildIndirectWrite(MBB, MI, MI->getOperand(ValOpIdx).getReg(),
+ calculateIndirectAddress(RegIndex, Channel),
+ OffsetReg);
+ }
+ } else {
+ return false;
+ }
+
+ MBB->erase(MI);
+ return true;
+}
+
+
MachineInstr *
AMDGPUInstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
MachineInstr *MI,
const SmallVectorImpl<unsigned> &Ops,
int FrameIndex) const {
// TODO: Implement this function
- return 0;
+ return nullptr;
}
MachineInstr*
AMDGPUInstrInfo::foldMemoryOperandImpl(MachineFunction &MF,
const SmallVectorImpl<unsigned> &Ops,
MachineInstr *LoadMI) const {
// TODO: Implement this function
- return 0;
+ return nullptr;
}
bool
AMDGPUInstrInfo::canFoldMemoryOperand(const MachineInstr *MI,
return 0;
}
-bool AMDGPUInstrInfo::shouldScheduleLoadsNear(SDNode *Load1, SDNode *Load2,
- int64_t Offset1, int64_t Offset2,
- unsigned NumLoads) const {
- assert(Offset2 > Offset1
- && "Second offset should be larger than first offset!");
- // If we have less than 16 loads in a row, and the offsets are within 16,
- // then schedule together.
- // TODO: Make the loads schedule near if it fits in a cacheline
- return (NumLoads < 16 && (Offset2 - Offset1) < 16);
+bool AMDGPUInstrInfo::enableClusterLoads() const {
+ return true;
+}
+
+// FIXME: This behaves strangely. If, for example, you have 32 load + stores,
+// the first 16 loads will be interleaved with the stores, and the next 16 will
+// be clustered as expected. It should really split into 2 16 store batches.
+//
+// Loads are clustered until this returns false, rather than trying to schedule
+// groups of stores. This also means we have to deal with saying different
+// address space loads should be clustered, and ones which might cause bank
+// conflicts.
+//
+// This might be deprecated so it might not be worth that much effort to fix.
+bool AMDGPUInstrInfo::shouldScheduleLoadsNear(SDNode *Load0, SDNode *Load1,
+ int64_t Offset0, int64_t Offset1,
+ unsigned NumLoads) const {
+ assert(Offset1 > Offset0 &&
+ "Second offset should be larger than first offset!");
+ // If we have less than 16 loads in a row, and the offsets are within 64
+ // bytes, then schedule together.
+
+ // A cacheline is 64 bytes (for global memory).
+ return (NumLoads <= 16 && (Offset1 - Offset0) < 64);
}
bool
return get(MI.getOpcode()).TSFlags & AMDGPU_FLAG_REGISTER_LOAD;
}
+int AMDGPUInstrInfo::getIndirectIndexBegin(const MachineFunction &MF) const {
+ const MachineRegisterInfo &MRI = MF.getRegInfo();
+ const MachineFrameInfo *MFI = MF.getFrameInfo();
+ int Offset = -1;
-void AMDGPUInstrInfo::convertToISA(MachineInstr & MI, MachineFunction &MF,
- DebugLoc DL) const {
- MachineRegisterInfo &MRI = MF.getRegInfo();
- const AMDGPURegisterInfo & RI = getRegisterInfo();
+ if (MFI->getNumObjects() == 0) {
+ return -1;
+ }
- for (unsigned i = 0; i < MI.getNumOperands(); i++) {
- MachineOperand &MO = MI.getOperand(i);
- // Convert dst regclass to one that is supported by the ISA
- if (MO.isReg() && MO.isDef()) {
- if (TargetRegisterInfo::isVirtualRegister(MO.getReg())) {
- const TargetRegisterClass * oldRegClass = MRI.getRegClass(MO.getReg());
- const TargetRegisterClass * newRegClass = RI.getISARegClass(oldRegClass);
+ if (MRI.livein_empty()) {
+ return 0;
+ }
- assert(newRegClass);
+ const TargetRegisterClass *IndirectRC = getIndirectAddrRegClass();
+ for (MachineRegisterInfo::livein_iterator LI = MRI.livein_begin(),
+ LE = MRI.livein_end();
+ LI != LE; ++LI) {
+ unsigned Reg = LI->first;
+ if (TargetRegisterInfo::isVirtualRegister(Reg) ||
+ !IndirectRC->contains(Reg))
+ continue;
- MRI.setRegClass(MO.getReg(), newRegClass);
- }
+ unsigned RegIndex;
+ unsigned RegEnd;
+ for (RegIndex = 0, RegEnd = IndirectRC->getNumRegs(); RegIndex != RegEnd;
+ ++RegIndex) {
+ if (IndirectRC->getRegister(RegIndex) == Reg)
+ break;
}
+ Offset = std::max(Offset, (int)RegIndex);
+ }
+
+ return Offset + 1;
+}
+
+int AMDGPUInstrInfo::getIndirectIndexEnd(const MachineFunction &MF) const {
+ int Offset = 0;
+ const MachineFrameInfo *MFI = MF.getFrameInfo();
+
+ // Variable sized objects are not supported
+ assert(!MFI->hasVarSizedObjects());
+
+ if (MFI->getNumObjects() == 0) {
+ return -1;
+ }
+
+ Offset = MF.getTarget()
+ .getSubtargetImpl()
+ ->getFrameLowering()
+ ->getFrameIndexOffset(MF, -1);
+
+ return getIndirectIndexBegin(MF) + Offset;
+}
+
+int AMDGPUInstrInfo::getMaskedMIMGOp(uint16_t Opcode, unsigned Channels) const {
+ switch (Channels) {
+ default: return Opcode;
+ case 1: return AMDGPU::getMaskedMIMGOp(Opcode, AMDGPU::Channels_1);
+ case 2: return AMDGPU::getMaskedMIMGOp(Opcode, AMDGPU::Channels_2);
+ case 3: return AMDGPU::getMaskedMIMGOp(Opcode, AMDGPU::Channels_3);
}
}
+
+// Wrapper for Tablegen'd function. enum Subtarget is not defined in any
+// header files, so we need to wrap it in a function that takes unsigned
+// instead.
+namespace llvm {
+namespace AMDGPU {
+int getMCOpcode(uint16_t Opcode, unsigned Gen) {
+ return getMCOpcode(Opcode);
+}
+}
+}
projects / oota-llvm.git / blobdiff