1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 03:33:20 +01:00
llvm-mirror/lib/Target/AMDGPU/AMDGPUMCInstLower.cpp
Stanislav Mekhanoshin 3f2c8aa170 [AMDGPU] Add wave barrier builtin
The wave barrier represents the discardable barrier. Its main purpose is to
carry convergent attribute, thus preventing illegal CFG optimizations. All lanes
in a wave come to convergence point simultaneously with SIMT, thus no special
instruction is needed in the ISA. The barrier is discarded during code generation.

Differential Revision: https://reviews.llvm.org/D26585

llvm-svn: 287007
2016-11-15 19:00:15 +00:00

243 lines
8.2 KiB
C++

//===- AMDGPUMCInstLower.cpp - Lower AMDGPU MachineInstr to an MCInst -----===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
/// \file
/// \brief Code to lower AMDGPU MachineInstrs to their corresponding MCInst.
//
//===----------------------------------------------------------------------===//
//
#include "AMDGPUMCInstLower.h"
#include "AMDGPUAsmPrinter.h"
#include "AMDGPUSubtarget.h"
#include "AMDGPUTargetMachine.h"
#include "InstPrinter/AMDGPUInstPrinter.h"
#include "SIInstrInfo.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalVariable.h"
#include "llvm/MC/MCCodeEmitter.h"
#include "llvm/MC/MCContext.h"
#include "llvm/MC/MCExpr.h"
#include "llvm/MC/MCInst.h"
#include "llvm/MC/MCObjectStreamer.h"
#include "llvm/MC/MCStreamer.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/Format.h"
#include <algorithm>
using namespace llvm;
#include "AMDGPUGenMCPseudoLowering.inc"
AMDGPUMCInstLower::AMDGPUMCInstLower(MCContext &ctx, const AMDGPUSubtarget &st,
const AsmPrinter &ap):
Ctx(ctx), ST(st), AP(ap) { }
static MCSymbolRefExpr::VariantKind getVariantKind(unsigned MOFlags) {
switch (MOFlags) {
default:
return MCSymbolRefExpr::VK_None;
case SIInstrInfo::MO_GOTPCREL:
return MCSymbolRefExpr::VK_GOTPCREL;
case SIInstrInfo::MO_GOTPCREL32_LO:
return MCSymbolRefExpr::VK_AMDGPU_GOTPCREL32_LO;
case SIInstrInfo::MO_GOTPCREL32_HI:
return MCSymbolRefExpr::VK_AMDGPU_GOTPCREL32_HI;
case SIInstrInfo::MO_REL32_LO:
return MCSymbolRefExpr::VK_AMDGPU_REL32_LO;
case SIInstrInfo::MO_REL32_HI:
return MCSymbolRefExpr::VK_AMDGPU_REL32_HI;
}
}
const MCExpr *AMDGPUMCInstLower::getLongBranchBlockExpr(
const MachineBasicBlock &SrcBB,
const MachineOperand &MO) const {
const MCExpr *DestBBSym
= MCSymbolRefExpr::create(MO.getMBB()->getSymbol(), Ctx);
const MCExpr *SrcBBSym = MCSymbolRefExpr::create(SrcBB.getSymbol(), Ctx);
assert(SrcBB.front().getOpcode() == AMDGPU::S_GETPC_B64 &&
ST.getInstrInfo()->get(AMDGPU::S_GETPC_B64).Size == 4);
// s_getpc_b64 returns the address of next instruction.
const MCConstantExpr *One = MCConstantExpr::create(4, Ctx);
SrcBBSym = MCBinaryExpr::createAdd(SrcBBSym, One, Ctx);
if (MO.getTargetFlags() == AMDGPU::TF_LONG_BRANCH_FORWARD)
return MCBinaryExpr::createSub(DestBBSym, SrcBBSym, Ctx);
assert(MO.getTargetFlags() == AMDGPU::TF_LONG_BRANCH_BACKWARD);
return MCBinaryExpr::createSub(SrcBBSym, DestBBSym, Ctx);
}
bool AMDGPUMCInstLower::lowerOperand(const MachineOperand &MO,
MCOperand &MCOp) const {
switch (MO.getType()) {
default:
llvm_unreachable("unknown operand type");
case MachineOperand::MO_Immediate:
MCOp = MCOperand::createImm(MO.getImm());
return true;
case MachineOperand::MO_Register:
MCOp = MCOperand::createReg(AMDGPU::getMCReg(MO.getReg(), ST));
return true;
case MachineOperand::MO_MachineBasicBlock: {
if (MO.getTargetFlags() != 0) {
MCOp = MCOperand::createExpr(
getLongBranchBlockExpr(*MO.getParent()->getParent(), MO));
} else {
MCOp = MCOperand::createExpr(
MCSymbolRefExpr::create(MO.getMBB()->getSymbol(), Ctx));
}
return true;
}
case MachineOperand::MO_GlobalAddress: {
const GlobalValue *GV = MO.getGlobal();
SmallString<128> SymbolName;
AP.getNameWithPrefix(SymbolName, GV);
MCSymbol *Sym = Ctx.getOrCreateSymbol(SymbolName);
const MCExpr *SymExpr =
MCSymbolRefExpr::create(Sym, getVariantKind(MO.getTargetFlags()),Ctx);
const MCExpr *Expr = MCBinaryExpr::createAdd(SymExpr,
MCConstantExpr::create(MO.getOffset(), Ctx), Ctx);
MCOp = MCOperand::createExpr(Expr);
return true;
}
case MachineOperand::MO_ExternalSymbol: {
MCSymbol *Sym = Ctx.getOrCreateSymbol(StringRef(MO.getSymbolName()));
Sym->setExternal(true);
const MCSymbolRefExpr *Expr = MCSymbolRefExpr::create(Sym, Ctx);
MCOp = MCOperand::createExpr(Expr);
return true;
}
}
}
void AMDGPUMCInstLower::lower(const MachineInstr *MI, MCInst &OutMI) const {
int MCOpcode = ST.getInstrInfo()->pseudoToMCOpcode(MI->getOpcode());
if (MCOpcode == -1) {
LLVMContext &C = MI->getParent()->getParent()->getFunction()->getContext();
C.emitError("AMDGPUMCInstLower::lower - Pseudo instruction doesn't have "
"a target-specific version: " + Twine(MI->getOpcode()));
}
OutMI.setOpcode(MCOpcode);
for (const MachineOperand &MO : MI->explicit_operands()) {
MCOperand MCOp;
lowerOperand(MO, MCOp);
OutMI.addOperand(MCOp);
}
}
bool AMDGPUAsmPrinter::lowerOperand(const MachineOperand &MO,
MCOperand &MCOp) const {
const AMDGPUSubtarget &STI = MF->getSubtarget<AMDGPUSubtarget>();
AMDGPUMCInstLower MCInstLowering(OutContext, STI, *this);
return MCInstLowering.lowerOperand(MO, MCOp);
}
void AMDGPUAsmPrinter::EmitInstruction(const MachineInstr *MI) {
if (emitPseudoExpansionLowering(*OutStreamer, MI))
return;
const AMDGPUSubtarget &STI = MF->getSubtarget<AMDGPUSubtarget>();
AMDGPUMCInstLower MCInstLowering(OutContext, STI, *this);
StringRef Err;
if (!STI.getInstrInfo()->verifyInstruction(*MI, Err)) {
LLVMContext &C = MI->getParent()->getParent()->getFunction()->getContext();
C.emitError("Illegal instruction detected: " + Err);
MI->dump();
}
if (MI->isBundle()) {
const MachineBasicBlock *MBB = MI->getParent();
MachineBasicBlock::const_instr_iterator I = ++MI->getIterator();
while (I != MBB->instr_end() && I->isInsideBundle()) {
EmitInstruction(&*I);
++I;
}
} else {
// We don't want SI_MASK_BRANCH/SI_RETURN encoded. They are placeholder
// terminator instructions and should only be printed as comments.
if (MI->getOpcode() == AMDGPU::SI_MASK_BRANCH) {
if (isVerbose()) {
SmallVector<char, 16> BBStr;
raw_svector_ostream Str(BBStr);
const MachineBasicBlock *MBB = MI->getOperand(0).getMBB();
const MCSymbolRefExpr *Expr
= MCSymbolRefExpr::create(MBB->getSymbol(), OutContext);
Expr->print(Str, MAI);
OutStreamer->emitRawComment(" mask branch " + BBStr);
}
return;
}
if (MI->getOpcode() == AMDGPU::SI_RETURN) {
if (isVerbose())
OutStreamer->emitRawComment(" return");
return;
}
if (MI->getOpcode() == AMDGPU::WAVE_BARRIER) {
if (isVerbose())
OutStreamer->emitRawComment(" wave barrier");
return;
}
MCInst TmpInst;
MCInstLowering.lower(MI, TmpInst);
EmitToStreamer(*OutStreamer, TmpInst);
if (STI.dumpCode()) {
// Disassemble instruction/operands to text.
DisasmLines.resize(DisasmLines.size() + 1);
std::string &DisasmLine = DisasmLines.back();
raw_string_ostream DisasmStream(DisasmLine);
AMDGPUInstPrinter InstPrinter(*TM.getMCAsmInfo(),
*STI.getInstrInfo(),
*STI.getRegisterInfo());
InstPrinter.printInst(&TmpInst, DisasmStream, StringRef(), STI);
// Disassemble instruction/operands to hex representation.
SmallVector<MCFixup, 4> Fixups;
SmallVector<char, 16> CodeBytes;
raw_svector_ostream CodeStream(CodeBytes);
auto &ObjStreamer = static_cast<MCObjectStreamer&>(*OutStreamer);
MCCodeEmitter &InstEmitter = ObjStreamer.getAssembler().getEmitter();
InstEmitter.encodeInstruction(TmpInst, CodeStream, Fixups,
MF->getSubtarget<MCSubtargetInfo>());
HexLines.resize(HexLines.size() + 1);
std::string &HexLine = HexLines.back();
raw_string_ostream HexStream(HexLine);
for (size_t i = 0; i < CodeBytes.size(); i += 4) {
unsigned int CodeDWord = *(unsigned int *)&CodeBytes[i];
HexStream << format("%s%08X", (i > 0 ? " " : ""), CodeDWord);
}
DisasmStream.flush();
DisasmLineMaxLen = std::max(DisasmLineMaxLen, DisasmLine.size());
}
}
}