//===----------------------------------------------------------------------===//
#include "NVPTXAsmPrinter.h"
+#include "InstPrinter/NVPTXInstPrinter.h"
#include "MCTargetDesc/NVPTXMCAsmInfo.h"
#include "NVPTX.h"
#include "NVPTXInstrInfo.h"
-#include "NVPTXNumRegisters.h"
+#include "NVPTXMachineFunctionInfo.h"
+#include "NVPTXMCExpr.h"
#include "NVPTXRegisterInfo.h"
#include "NVPTXTargetMachine.h"
#include "NVPTXUtilities.h"
#include "cl_common_defines.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/Analysis/ConstantFolding.h"
-#include "llvm/Assembly/Writer.h"
#include "llvm/CodeGen/Analysis.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
-#include "llvm/DebugInfo.h"
+#include "llvm/IR/DebugInfo.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalVariable.h"
+#include "llvm/IR/Mangler.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Operator.h"
#include "llvm/MC/MCStreamer.h"
#include "llvm/Support/Path.h"
#include "llvm/Support/TargetRegistry.h"
#include "llvm/Support/TimeValue.h"
-#include "llvm/Target/Mangler.h"
#include "llvm/Target/TargetLoweringObjectFile.h"
#include <sstream>
using namespace llvm;
-#include "NVPTXGenAsmWriter.inc"
-
-bool RegAllocNilUsed = true;
-
#define DEPOTNAME "__local_depot"
static cl::opt<bool>
-EmitLineNumbers("nvptx-emit-line-numbers",
+EmitLineNumbers("nvptx-emit-line-numbers", cl::Hidden,
cl::desc("NVPTX Specific: Emit Line numbers even without -G"),
cl::init(true));
-namespace llvm { bool InterleaveSrcInPtx = false; }
-
-static cl::opt<bool, true>
-InterleaveSrc("nvptx-emit-src", cl::ZeroOrMore,
+static cl::opt<bool>
+InterleaveSrc("nvptx-emit-src", cl::ZeroOrMore, cl::Hidden,
cl::desc("NVPTX Specific: Emit source line in ptx file"),
- cl::location(llvm::InterleaveSrcInPtx));
+ cl::init(false));
namespace {
/// DiscoverDependentGlobals - Return a set of GlobalVariables on which \p V
return MCConstantExpr::Create(CI->getZExtValue(), Ctx);
if (const GlobalValue *GV = dyn_cast<GlobalValue>(CV))
- return MCSymbolRefExpr::Create(AP.Mang->getSymbol(GV), Ctx);
+ return MCSymbolRefExpr::Create(AP.getSymbol(GV), Ctx);
if (const BlockAddress *BA = dyn_cast<BlockAddress>(CV))
return MCSymbolRefExpr::Create(AP.GetBlockAddressSymbol(BA), Ctx);
const ConstantExpr *CE = dyn_cast<ConstantExpr>(CV);
- if (CE == 0)
+ if (!CE)
llvm_unreachable("Unknown constant value to lower!");
switch (CE->getOpcode()) {
std::string S;
raw_string_ostream OS(S);
OS << "Unsupported expression in static initializer: ";
- WriteAsOperand(OS, CE, /*PrintType=*/ false,
- !AP.MF ? 0 : AP.MF->getFunction()->getParent());
+ CE->printAsOperand(OS, /*PrintType=*/ false,
+ !AP.MF ? nullptr : AP.MF->getFunction()->getParent());
report_fatal_error(OS.str());
}
+ case Instruction::AddrSpaceCast: {
+ // Strip any addrspace(1)->addrspace(0) addrspace casts. These will be
+ // handled by the generic() logic in the MCExpr printer
+ PointerType *DstTy = cast<PointerType>(CE->getType());
+ PointerType *SrcTy = cast<PointerType>(CE->getOperand(0)->getType());
+ if (SrcTy->getAddressSpace() == 1 && DstTy->getAddressSpace() == 0) {
+ return LowerConstant(cast<const Constant>(CE->getOperand(0)), AP);
+ }
+ std::string S;
+ raw_string_ostream OS(S);
+ OS << "Unsupported expression in static initializer: ";
+ CE->printAsOperand(OS, /*PrintType=*/ false,
+ !AP.MF ? nullptr : AP.MF->getFunction()->getParent());
+ report_fatal_error(OS.str());
+ }
case Instruction::GetElementPtr: {
const DataLayout &TD = *AP.TM.getDataLayout();
// Generate a symbolic expression for the byte address
const LLVMContext &ctx = MF->getFunction()->getContext();
DIScope Scope(curLoc.getScope(ctx));
- if (!Scope.Verify())
- return;
+ assert((!Scope || Scope.isScope()) &&
+ "Scope of a DebugLoc should be null or a DIScope.");
+ if (!Scope)
+ return;
StringRef fileName(Scope.getFilename());
StringRef dirName(Scope.getDirectory());
return;
// Emit the line from the source file.
- if (llvm::InterleaveSrcInPtx)
+ if (InterleaveSrc)
this->emitSrcInText(fileName.str(), curLoc.getLine());
std::stringstream temp;
raw_svector_ostream OS(Str);
if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA)
emitLineNumberAsDotLoc(*MI);
- printInstruction(MI, OS);
- OutStreamer.EmitRawText(OS.str());
+
+ MCInst Inst;
+ lowerToMCInst(MI, Inst);
+ EmitToStreamer(OutStreamer, Inst);
+}
+
+// Handle symbol backtracking for targets that do not support image handles
+bool NVPTXAsmPrinter::lowerImageHandleOperand(const MachineInstr *MI,
+ unsigned OpNo, MCOperand &MCOp) {
+ const MachineOperand &MO = MI->getOperand(OpNo);
+
+ switch (MI->getOpcode()) {
+ default: return false;
+ case NVPTX::TEX_1D_F32_I32:
+ case NVPTX::TEX_1D_F32_F32:
+ case NVPTX::TEX_1D_F32_F32_LEVEL:
+ case NVPTX::TEX_1D_F32_F32_GRAD:
+ case NVPTX::TEX_1D_I32_I32:
+ case NVPTX::TEX_1D_I32_F32:
+ case NVPTX::TEX_1D_I32_F32_LEVEL:
+ case NVPTX::TEX_1D_I32_F32_GRAD:
+ case NVPTX::TEX_1D_ARRAY_F32_I32:
+ case NVPTX::TEX_1D_ARRAY_F32_F32:
+ case NVPTX::TEX_1D_ARRAY_F32_F32_LEVEL:
+ case NVPTX::TEX_1D_ARRAY_F32_F32_GRAD:
+ case NVPTX::TEX_1D_ARRAY_I32_I32:
+ case NVPTX::TEX_1D_ARRAY_I32_F32:
+ case NVPTX::TEX_1D_ARRAY_I32_F32_LEVEL:
+ case NVPTX::TEX_1D_ARRAY_I32_F32_GRAD:
+ case NVPTX::TEX_2D_F32_I32:
+ case NVPTX::TEX_2D_F32_F32:
+ case NVPTX::TEX_2D_F32_F32_LEVEL:
+ case NVPTX::TEX_2D_F32_F32_GRAD:
+ case NVPTX::TEX_2D_I32_I32:
+ case NVPTX::TEX_2D_I32_F32:
+ case NVPTX::TEX_2D_I32_F32_LEVEL:
+ case NVPTX::TEX_2D_I32_F32_GRAD:
+ case NVPTX::TEX_2D_ARRAY_F32_I32:
+ case NVPTX::TEX_2D_ARRAY_F32_F32:
+ case NVPTX::TEX_2D_ARRAY_F32_F32_LEVEL:
+ case NVPTX::TEX_2D_ARRAY_F32_F32_GRAD:
+ case NVPTX::TEX_2D_ARRAY_I32_I32:
+ case NVPTX::TEX_2D_ARRAY_I32_F32:
+ case NVPTX::TEX_2D_ARRAY_I32_F32_LEVEL:
+ case NVPTX::TEX_2D_ARRAY_I32_F32_GRAD:
+ case NVPTX::TEX_3D_F32_I32:
+ case NVPTX::TEX_3D_F32_F32:
+ case NVPTX::TEX_3D_F32_F32_LEVEL:
+ case NVPTX::TEX_3D_F32_F32_GRAD:
+ case NVPTX::TEX_3D_I32_I32:
+ case NVPTX::TEX_3D_I32_F32:
+ case NVPTX::TEX_3D_I32_F32_LEVEL:
+ case NVPTX::TEX_3D_I32_F32_GRAD:
+ {
+ // This is a texture fetch, so operand 4 is a texref and operand 5 is
+ // a samplerref
+ if (OpNo == 4) {
+ lowerImageHandleSymbol(MO.getImm(), MCOp);
+ return true;
+ }
+ if (OpNo == 5) {
+ lowerImageHandleSymbol(MO.getImm(), MCOp);
+ return true;
+ }
+
+ return false;
+ }
+ case NVPTX::SULD_1D_I8_TRAP:
+ case NVPTX::SULD_1D_I16_TRAP:
+ case NVPTX::SULD_1D_I32_TRAP:
+ case NVPTX::SULD_1D_ARRAY_I8_TRAP:
+ case NVPTX::SULD_1D_ARRAY_I16_TRAP:
+ case NVPTX::SULD_1D_ARRAY_I32_TRAP:
+ case NVPTX::SULD_2D_I8_TRAP:
+ case NVPTX::SULD_2D_I16_TRAP:
+ case NVPTX::SULD_2D_I32_TRAP:
+ case NVPTX::SULD_2D_ARRAY_I8_TRAP:
+ case NVPTX::SULD_2D_ARRAY_I16_TRAP:
+ case NVPTX::SULD_2D_ARRAY_I32_TRAP:
+ case NVPTX::SULD_3D_I8_TRAP:
+ case NVPTX::SULD_3D_I16_TRAP:
+ case NVPTX::SULD_3D_I32_TRAP: {
+ // This is a V1 surface load, so operand 1 is a surfref
+ if (OpNo == 1) {
+ lowerImageHandleSymbol(MO.getImm(), MCOp);
+ return true;
+ }
+
+ return false;
+ }
+ case NVPTX::SULD_1D_V2I8_TRAP:
+ case NVPTX::SULD_1D_V2I16_TRAP:
+ case NVPTX::SULD_1D_V2I32_TRAP:
+ case NVPTX::SULD_1D_ARRAY_V2I8_TRAP:
+ case NVPTX::SULD_1D_ARRAY_V2I16_TRAP:
+ case NVPTX::SULD_1D_ARRAY_V2I32_TRAP:
+ case NVPTX::SULD_2D_V2I8_TRAP:
+ case NVPTX::SULD_2D_V2I16_TRAP:
+ case NVPTX::SULD_2D_V2I32_TRAP:
+ case NVPTX::SULD_2D_ARRAY_V2I8_TRAP:
+ case NVPTX::SULD_2D_ARRAY_V2I16_TRAP:
+ case NVPTX::SULD_2D_ARRAY_V2I32_TRAP:
+ case NVPTX::SULD_3D_V2I8_TRAP:
+ case NVPTX::SULD_3D_V2I16_TRAP:
+ case NVPTX::SULD_3D_V2I32_TRAP: {
+ // This is a V2 surface load, so operand 2 is a surfref
+ if (OpNo == 2) {
+ lowerImageHandleSymbol(MO.getImm(), MCOp);
+ return true;
+ }
+
+ return false;
+ }
+ case NVPTX::SULD_1D_V4I8_TRAP:
+ case NVPTX::SULD_1D_V4I16_TRAP:
+ case NVPTX::SULD_1D_V4I32_TRAP:
+ case NVPTX::SULD_1D_ARRAY_V4I8_TRAP:
+ case NVPTX::SULD_1D_ARRAY_V4I16_TRAP:
+ case NVPTX::SULD_1D_ARRAY_V4I32_TRAP:
+ case NVPTX::SULD_2D_V4I8_TRAP:
+ case NVPTX::SULD_2D_V4I16_TRAP:
+ case NVPTX::SULD_2D_V4I32_TRAP:
+ case NVPTX::SULD_2D_ARRAY_V4I8_TRAP:
+ case NVPTX::SULD_2D_ARRAY_V4I16_TRAP:
+ case NVPTX::SULD_2D_ARRAY_V4I32_TRAP:
+ case NVPTX::SULD_3D_V4I8_TRAP:
+ case NVPTX::SULD_3D_V4I16_TRAP:
+ case NVPTX::SULD_3D_V4I32_TRAP: {
+ // This is a V4 surface load, so operand 4 is a surfref
+ if (OpNo == 4) {
+ lowerImageHandleSymbol(MO.getImm(), MCOp);
+ return true;
+ }
+
+ return false;
+ }
+ case NVPTX::SUST_B_1D_B8_TRAP:
+ case NVPTX::SUST_B_1D_B16_TRAP:
+ case NVPTX::SUST_B_1D_B32_TRAP:
+ case NVPTX::SUST_B_1D_V2B8_TRAP:
+ case NVPTX::SUST_B_1D_V2B16_TRAP:
+ case NVPTX::SUST_B_1D_V2B32_TRAP:
+ case NVPTX::SUST_B_1D_V4B8_TRAP:
+ case NVPTX::SUST_B_1D_V4B16_TRAP:
+ case NVPTX::SUST_B_1D_V4B32_TRAP:
+ case NVPTX::SUST_B_1D_ARRAY_B8_TRAP:
+ case NVPTX::SUST_B_1D_ARRAY_B16_TRAP:
+ case NVPTX::SUST_B_1D_ARRAY_B32_TRAP:
+ case NVPTX::SUST_B_1D_ARRAY_V2B8_TRAP:
+ case NVPTX::SUST_B_1D_ARRAY_V2B16_TRAP:
+ case NVPTX::SUST_B_1D_ARRAY_V2B32_TRAP:
+ case NVPTX::SUST_B_1D_ARRAY_V4B8_TRAP:
+ case NVPTX::SUST_B_1D_ARRAY_V4B16_TRAP:
+ case NVPTX::SUST_B_1D_ARRAY_V4B32_TRAP:
+ case NVPTX::SUST_B_2D_B8_TRAP:
+ case NVPTX::SUST_B_2D_B16_TRAP:
+ case NVPTX::SUST_B_2D_B32_TRAP:
+ case NVPTX::SUST_B_2D_V2B8_TRAP:
+ case NVPTX::SUST_B_2D_V2B16_TRAP:
+ case NVPTX::SUST_B_2D_V2B32_TRAP:
+ case NVPTX::SUST_B_2D_V4B8_TRAP:
+ case NVPTX::SUST_B_2D_V4B16_TRAP:
+ case NVPTX::SUST_B_2D_V4B32_TRAP:
+ case NVPTX::SUST_B_2D_ARRAY_B8_TRAP:
+ case NVPTX::SUST_B_2D_ARRAY_B16_TRAP:
+ case NVPTX::SUST_B_2D_ARRAY_B32_TRAP:
+ case NVPTX::SUST_B_2D_ARRAY_V2B8_TRAP:
+ case NVPTX::SUST_B_2D_ARRAY_V2B16_TRAP:
+ case NVPTX::SUST_B_2D_ARRAY_V2B32_TRAP:
+ case NVPTX::SUST_B_2D_ARRAY_V4B8_TRAP:
+ case NVPTX::SUST_B_2D_ARRAY_V4B16_TRAP:
+ case NVPTX::SUST_B_2D_ARRAY_V4B32_TRAP:
+ case NVPTX::SUST_B_3D_B8_TRAP:
+ case NVPTX::SUST_B_3D_B16_TRAP:
+ case NVPTX::SUST_B_3D_B32_TRAP:
+ case NVPTX::SUST_B_3D_V2B8_TRAP:
+ case NVPTX::SUST_B_3D_V2B16_TRAP:
+ case NVPTX::SUST_B_3D_V2B32_TRAP:
+ case NVPTX::SUST_B_3D_V4B8_TRAP:
+ case NVPTX::SUST_B_3D_V4B16_TRAP:
+ case NVPTX::SUST_B_3D_V4B32_TRAP:
+ case NVPTX::SUST_P_1D_B8_TRAP:
+ case NVPTX::SUST_P_1D_B16_TRAP:
+ case NVPTX::SUST_P_1D_B32_TRAP:
+ case NVPTX::SUST_P_1D_V2B8_TRAP:
+ case NVPTX::SUST_P_1D_V2B16_TRAP:
+ case NVPTX::SUST_P_1D_V2B32_TRAP:
+ case NVPTX::SUST_P_1D_V4B8_TRAP:
+ case NVPTX::SUST_P_1D_V4B16_TRAP:
+ case NVPTX::SUST_P_1D_V4B32_TRAP:
+ case NVPTX::SUST_P_1D_ARRAY_B8_TRAP:
+ case NVPTX::SUST_P_1D_ARRAY_B16_TRAP:
+ case NVPTX::SUST_P_1D_ARRAY_B32_TRAP:
+ case NVPTX::SUST_P_1D_ARRAY_V2B8_TRAP:
+ case NVPTX::SUST_P_1D_ARRAY_V2B16_TRAP:
+ case NVPTX::SUST_P_1D_ARRAY_V2B32_TRAP:
+ case NVPTX::SUST_P_1D_ARRAY_V4B8_TRAP:
+ case NVPTX::SUST_P_1D_ARRAY_V4B16_TRAP:
+ case NVPTX::SUST_P_1D_ARRAY_V4B32_TRAP:
+ case NVPTX::SUST_P_2D_B8_TRAP:
+ case NVPTX::SUST_P_2D_B16_TRAP:
+ case NVPTX::SUST_P_2D_B32_TRAP:
+ case NVPTX::SUST_P_2D_V2B8_TRAP:
+ case NVPTX::SUST_P_2D_V2B16_TRAP:
+ case NVPTX::SUST_P_2D_V2B32_TRAP:
+ case NVPTX::SUST_P_2D_V4B8_TRAP:
+ case NVPTX::SUST_P_2D_V4B16_TRAP:
+ case NVPTX::SUST_P_2D_V4B32_TRAP:
+ case NVPTX::SUST_P_2D_ARRAY_B8_TRAP:
+ case NVPTX::SUST_P_2D_ARRAY_B16_TRAP:
+ case NVPTX::SUST_P_2D_ARRAY_B32_TRAP:
+ case NVPTX::SUST_P_2D_ARRAY_V2B8_TRAP:
+ case NVPTX::SUST_P_2D_ARRAY_V2B16_TRAP:
+ case NVPTX::SUST_P_2D_ARRAY_V2B32_TRAP:
+ case NVPTX::SUST_P_2D_ARRAY_V4B8_TRAP:
+ case NVPTX::SUST_P_2D_ARRAY_V4B16_TRAP:
+ case NVPTX::SUST_P_2D_ARRAY_V4B32_TRAP:
+ case NVPTX::SUST_P_3D_B8_TRAP:
+ case NVPTX::SUST_P_3D_B16_TRAP:
+ case NVPTX::SUST_P_3D_B32_TRAP:
+ case NVPTX::SUST_P_3D_V2B8_TRAP:
+ case NVPTX::SUST_P_3D_V2B16_TRAP:
+ case NVPTX::SUST_P_3D_V2B32_TRAP:
+ case NVPTX::SUST_P_3D_V4B8_TRAP:
+ case NVPTX::SUST_P_3D_V4B16_TRAP:
+ case NVPTX::SUST_P_3D_V4B32_TRAP: {
+ // This is a surface store, so operand 0 is a surfref
+ if (OpNo == 0) {
+ lowerImageHandleSymbol(MO.getImm(), MCOp);
+ return true;
+ }
+
+ return false;
+ }
+ case NVPTX::TXQ_CHANNEL_ORDER:
+ case NVPTX::TXQ_CHANNEL_DATA_TYPE:
+ case NVPTX::TXQ_WIDTH:
+ case NVPTX::TXQ_HEIGHT:
+ case NVPTX::TXQ_DEPTH:
+ case NVPTX::TXQ_ARRAY_SIZE:
+ case NVPTX::TXQ_NUM_SAMPLES:
+ case NVPTX::TXQ_NUM_MIPMAP_LEVELS:
+ case NVPTX::SUQ_CHANNEL_ORDER:
+ case NVPTX::SUQ_CHANNEL_DATA_TYPE:
+ case NVPTX::SUQ_WIDTH:
+ case NVPTX::SUQ_HEIGHT:
+ case NVPTX::SUQ_DEPTH:
+ case NVPTX::SUQ_ARRAY_SIZE: {
+ // This is a query, so operand 1 is a surfref/texref
+ if (OpNo == 1) {
+ lowerImageHandleSymbol(MO.getImm(), MCOp);
+ return true;
+ }
+
+ return false;
+ }
+ }
+}
+
+void NVPTXAsmPrinter::lowerImageHandleSymbol(unsigned Index, MCOperand &MCOp) {
+ // Ewwww
+ TargetMachine &TM = const_cast<TargetMachine&>(MF->getTarget());
+ NVPTXTargetMachine &nvTM = static_cast<NVPTXTargetMachine&>(TM);
+ const NVPTXMachineFunctionInfo *MFI = MF->getInfo<NVPTXMachineFunctionInfo>();
+ const char *Sym = MFI->getImageHandleSymbol(Index);
+ std::string *SymNamePtr =
+ nvTM.getManagedStrPool()->getManagedString(Sym);
+ MCOp = GetSymbolRef(OutContext.GetOrCreateSymbol(
+ StringRef(SymNamePtr->c_str())));
+}
+
+void NVPTXAsmPrinter::lowerToMCInst(const MachineInstr *MI, MCInst &OutMI) {
+ OutMI.setOpcode(MI->getOpcode());
+ const NVPTXSubtarget &ST = TM.getSubtarget<NVPTXSubtarget>();
+
+ // Special: Do not mangle symbol operand of CALL_PROTOTYPE
+ if (MI->getOpcode() == NVPTX::CALL_PROTOTYPE) {
+ const MachineOperand &MO = MI->getOperand(0);
+ OutMI.addOperand(GetSymbolRef(
+ OutContext.GetOrCreateSymbol(Twine(MO.getSymbolName()))));
+ return;
+ }
+
+ for (unsigned i = 0, e = MI->getNumOperands(); i != e; ++i) {
+ const MachineOperand &MO = MI->getOperand(i);
+
+ MCOperand MCOp;
+ if (!ST.hasImageHandles()) {
+ if (lowerImageHandleOperand(MI, i, MCOp)) {
+ OutMI.addOperand(MCOp);
+ continue;
+ }
+ }
+
+ if (lowerOperand(MO, MCOp))
+ OutMI.addOperand(MCOp);
+ }
+}
+
+bool NVPTXAsmPrinter::lowerOperand(const MachineOperand &MO,
+ MCOperand &MCOp) {
+ switch (MO.getType()) {
+ default: llvm_unreachable("unknown operand type");
+ case MachineOperand::MO_Register:
+ MCOp = MCOperand::CreateReg(encodeVirtualRegister(MO.getReg()));
+ break;
+ case MachineOperand::MO_Immediate:
+ MCOp = MCOperand::CreateImm(MO.getImm());
+ break;
+ case MachineOperand::MO_MachineBasicBlock:
+ MCOp = MCOperand::CreateExpr(MCSymbolRefExpr::Create(
+ MO.getMBB()->getSymbol(), OutContext));
+ break;
+ case MachineOperand::MO_ExternalSymbol:
+ MCOp = GetSymbolRef(GetExternalSymbolSymbol(MO.getSymbolName()));
+ break;
+ case MachineOperand::MO_GlobalAddress:
+ MCOp = GetSymbolRef(getSymbol(MO.getGlobal()));
+ break;
+ case MachineOperand::MO_FPImmediate: {
+ const ConstantFP *Cnt = MO.getFPImm();
+ APFloat Val = Cnt->getValueAPF();
+
+ switch (Cnt->getType()->getTypeID()) {
+ default: report_fatal_error("Unsupported FP type"); break;
+ case Type::FloatTyID:
+ MCOp = MCOperand::CreateExpr(
+ NVPTXFloatMCExpr::CreateConstantFPSingle(Val, OutContext));
+ break;
+ case Type::DoubleTyID:
+ MCOp = MCOperand::CreateExpr(
+ NVPTXFloatMCExpr::CreateConstantFPDouble(Val, OutContext));
+ break;
+ }
+ break;
+ }
+ }
+ return true;
+}
+
+unsigned NVPTXAsmPrinter::encodeVirtualRegister(unsigned Reg) {
+ if (TargetRegisterInfo::isVirtualRegister(Reg)) {
+ const TargetRegisterClass *RC = MRI->getRegClass(Reg);
+
+ DenseMap<unsigned, unsigned> &RegMap = VRegMapping[RC];
+ unsigned RegNum = RegMap[Reg];
+
+ // Encode the register class in the upper 4 bits
+ // Must be kept in sync with NVPTXInstPrinter::printRegName
+ unsigned Ret = 0;
+ if (RC == &NVPTX::Int1RegsRegClass) {
+ Ret = (1 << 28);
+ } else if (RC == &NVPTX::Int16RegsRegClass) {
+ Ret = (2 << 28);
+ } else if (RC == &NVPTX::Int32RegsRegClass) {
+ Ret = (3 << 28);
+ } else if (RC == &NVPTX::Int64RegsRegClass) {
+ Ret = (4 << 28);
+ } else if (RC == &NVPTX::Float32RegsRegClass) {
+ Ret = (5 << 28);
+ } else if (RC == &NVPTX::Float64RegsRegClass) {
+ Ret = (6 << 28);
+ } else {
+ report_fatal_error("Bad register class");
+ }
+
+ // Insert the vreg number
+ Ret |= (RegNum & 0x0FFFFFFF);
+ return Ret;
+ } else {
+ // Some special-use registers are actually physical registers.
+ // Encode this as the register class ID of 0 and the real register ID.
+ return Reg & 0x0FFFFFFF;
+ }
+}
+
+MCOperand NVPTXAsmPrinter::GetSymbolRef(const MCSymbol *Symbol) {
+ const MCExpr *Expr;
+ Expr = MCSymbolRefExpr::Create(Symbol, MCSymbolRefExpr::VK_None,
+ OutContext);
+ return MCOperand::CreateExpr(Expr);
}
void NVPTXAsmPrinter::printReturnValStr(const Function *F, raw_ostream &O) {
O << " (";
if (isABI) {
- if (Ty->isPrimitiveType() || Ty->isIntegerTy()) {
+ if (Ty->isFloatingPointTy() || Ty->isIntegerTy()) {
unsigned size = 0;
if (const IntegerType *ITy = dyn_cast<IntegerType>(Ty)) {
size = ITy->getBitWidth();
<< " func_retval0";
} else {
if ((Ty->getTypeID() == Type::StructTyID) || isa<VectorType>(Ty)) {
- SmallVector<EVT, 16> vtparts;
- ComputeValueVTs(*TLI, Ty, vtparts);
- unsigned totalsz = 0;
- for (unsigned i = 0, e = vtparts.size(); i != e; ++i) {
- unsigned elems = 1;
- EVT elemtype = vtparts[i];
- if (vtparts[i].isVector()) {
- elems = vtparts[i].getVectorNumElements();
- elemtype = vtparts[i].getVectorElementType();
- }
- for (unsigned j = 0, je = elems; j != je; ++j) {
- unsigned sz = elemtype.getSizeInBits();
- if (elemtype.isInteger() && (sz < 8))
- sz = 8;
- totalsz += sz / 8;
- }
- }
+ unsigned totalsz = TD->getTypeAllocSize(Ty);
unsigned retAlignment = 0;
if (!llvm::getAlign(*F, 0, retAlignment))
retAlignment = TD->getABITypeAlignment(Ty);
}
void NVPTXAsmPrinter::EmitFunctionBodyStart() {
- const TargetRegisterInfo &TRI = *TM.getRegisterInfo();
- unsigned numRegClasses = TRI.getNumRegClasses();
- VRidGlobal2LocalMap = new std::map<unsigned, unsigned>[numRegClasses + 1];
+ VRegMapping.clear();
OutStreamer.EmitRawText(StringRef("{\n"));
setAndEmitFunctionVirtualRegisters(*MF);
void NVPTXAsmPrinter::EmitFunctionBodyEnd() {
OutStreamer.EmitRawText(StringRef("}\n"));
- delete[] VRidGlobal2LocalMap;
+ VRegMapping.clear();
+}
+
+void NVPTXAsmPrinter::emitImplicitDef(const MachineInstr *MI) const {
+ unsigned RegNo = MI->getOperand(0).getReg();
+ const TargetRegisterInfo *TRI = TM.getRegisterInfo();
+ if (TRI->isVirtualRegister(RegNo)) {
+ OutStreamer.AddComment(Twine("implicit-def: ") +
+ getVirtualRegisterName(RegNo));
+ } else {
+ OutStreamer.AddComment(Twine("implicit-def: ") +
+ TM.getRegisterInfo()->getName(RegNo));
+ }
+ OutStreamer.AddBlankLine();
}
void NVPTXAsmPrinter::emitKernelFunctionDirectives(const Function &F,
O << ".minnctapersm " << mincta << "\n";
}
-void NVPTXAsmPrinter::getVirtualRegisterName(unsigned vr, bool isVec,
- raw_ostream &O) {
- const TargetRegisterClass *RC = MRI->getRegClass(vr);
- unsigned id = RC->getID();
+std::string
+NVPTXAsmPrinter::getVirtualRegisterName(unsigned Reg) const {
+ const TargetRegisterClass *RC = MRI->getRegClass(Reg);
- std::map<unsigned, unsigned> ®map = VRidGlobal2LocalMap[id];
- unsigned mapped_vr = regmap[vr];
+ std::string Name;
+ raw_string_ostream NameStr(Name);
- if (!isVec) {
- O << getNVPTXRegClassStr(RC) << mapped_vr;
- return;
- }
- report_fatal_error("Bad register!");
+ VRegRCMap::const_iterator I = VRegMapping.find(RC);
+ assert(I != VRegMapping.end() && "Bad register class");
+ const DenseMap<unsigned, unsigned> &RegMap = I->second;
+
+ VRegMap::const_iterator VI = RegMap.find(Reg);
+ assert(VI != RegMap.end() && "Bad virtual register");
+ unsigned MappedVR = VI->second;
+
+ NameStr << getNVPTXRegClassStr(RC) << MappedVR;
+
+ NameStr.flush();
+ return Name;
}
-void NVPTXAsmPrinter::emitVirtualRegister(unsigned int vr, bool isVec,
+void NVPTXAsmPrinter::emitVirtualRegister(unsigned int vr,
raw_ostream &O) {
- getVirtualRegisterName(vr, isVec, O);
+ O << getVirtualRegisterName(vr);
}
void NVPTXAsmPrinter::printVecModifiedImmediate(
llvm_unreachable("Unknown Modifier on immediate operand");
}
-void NVPTXAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
- raw_ostream &O, const char *Modifier) {
- const MachineOperand &MO = MI->getOperand(opNum);
- switch (MO.getType()) {
- case MachineOperand::MO_Register:
- if (TargetRegisterInfo::isPhysicalRegister(MO.getReg())) {
- if (MO.getReg() == NVPTX::VRDepot)
- O << DEPOTNAME << getFunctionNumber();
- else
- O << getRegisterName(MO.getReg());
- } else {
- if (!Modifier)
- emitVirtualRegister(MO.getReg(), false, O);
- else {
- if (strcmp(Modifier, "vecfull") == 0)
- emitVirtualRegister(MO.getReg(), true, O);
- else
- llvm_unreachable(
- "Don't know how to handle the modifier on virtual register.");
- }
- }
- return;
- case MachineOperand::MO_Immediate:
- if (!Modifier)
- O << MO.getImm();
- else if (strstr(Modifier, "vec") == Modifier)
- printVecModifiedImmediate(MO, Modifier, O);
- else
- llvm_unreachable(
- "Don't know how to handle modifier on immediate operand");
- return;
-
- case MachineOperand::MO_FPImmediate:
- printFPConstant(MO.getFPImm(), O);
- break;
-
- case MachineOperand::MO_GlobalAddress:
- O << *Mang->getSymbol(MO.getGlobal());
- break;
-
- case MachineOperand::MO_ExternalSymbol: {
- const char *symbname = MO.getSymbolName();
- if (strstr(symbname, ".PARAM") == symbname) {
- unsigned index;
- sscanf(symbname + 6, "%u[];", &index);
- printParamName(index, O);
- } else if (strstr(symbname, ".HLPPARAM") == symbname) {
- unsigned index;
- sscanf(symbname + 9, "%u[];", &index);
- O << *CurrentFnSym << "_param_" << index << "_offset";
- } else
- O << symbname;
- break;
- }
-
- case MachineOperand::MO_MachineBasicBlock:
- O << *MO.getMBB()->getSymbol();
- return;
-
- default:
- llvm_unreachable("Operand type not supported.");
- }
-}
-
-void NVPTXAsmPrinter::printImplicitDef(const MachineInstr *MI,
- raw_ostream &O) const {
-#ifndef __OPTIMIZE__
- O << "\t// Implicit def :";
- //printOperand(MI, 0);
- O << "\n";
-#endif
-}
-
-void NVPTXAsmPrinter::printMemOperand(const MachineInstr *MI, int opNum,
- raw_ostream &O, const char *Modifier) {
- printOperand(MI, opNum, O);
-
- if (Modifier && !strcmp(Modifier, "add")) {
- O << ", ";
- printOperand(MI, opNum + 1, O);
- } else {
- if (MI->getOperand(opNum + 1).isImm() &&
- MI->getOperand(opNum + 1).getImm() == 0)
- return; // don't print ',0' or '+0'
- O << "+";
- printOperand(MI, opNum + 1, O);
- }
-}
-
-void NVPTXAsmPrinter::printLdStCode(const MachineInstr *MI, int opNum,
- raw_ostream &O, const char *Modifier) {
- if (Modifier) {
- const MachineOperand &MO = MI->getOperand(opNum);
- int Imm = (int) MO.getImm();
- if (!strcmp(Modifier, "volatile")) {
- if (Imm)
- O << ".volatile";
- } else if (!strcmp(Modifier, "addsp")) {
- switch (Imm) {
- case NVPTX::PTXLdStInstCode::GLOBAL:
- O << ".global";
- break;
- case NVPTX::PTXLdStInstCode::SHARED:
- O << ".shared";
- break;
- case NVPTX::PTXLdStInstCode::LOCAL:
- O << ".local";
- break;
- case NVPTX::PTXLdStInstCode::PARAM:
- O << ".param";
- break;
- case NVPTX::PTXLdStInstCode::CONSTANT:
- O << ".const";
- break;
- case NVPTX::PTXLdStInstCode::GENERIC:
- if (!nvptxSubtarget.hasGenericLdSt())
- O << ".global";
- break;
- default:
- llvm_unreachable("Wrong Address Space");
- }
- } else if (!strcmp(Modifier, "sign")) {
- if (Imm == NVPTX::PTXLdStInstCode::Signed)
- O << "s";
- else if (Imm == NVPTX::PTXLdStInstCode::Unsigned)
- O << "u";
- else
- O << "f";
- } else if (!strcmp(Modifier, "vec")) {
- if (Imm == NVPTX::PTXLdStInstCode::V2)
- O << ".v2";
- else if (Imm == NVPTX::PTXLdStInstCode::V4)
- O << ".v4";
- } else
- llvm_unreachable("Unknown Modifier");
- } else
- llvm_unreachable("Empty Modifier");
-}
void NVPTXAsmPrinter::emitDeclaration(const Function *F, raw_ostream &O) {
else
O << ".func ";
printReturnValStr(F, O);
- O << *Mang->getSymbol(F) << "\n";
+ O << *getSymbol(F) << "\n";
emitFunctionParamList(F, O);
O << ";\n";
}
return true;
}
- for (Value::const_use_iterator ui = C->use_begin(), ue = C->use_end();
- ui != ue; ++ui) {
- const Constant *C = dyn_cast<Constant>(*ui);
- if (usedInGlobalVarDef(C))
- return true;
- }
+ for (const User *U : C->users())
+ if (const Constant *C = dyn_cast<Constant>(U))
+ if (usedInGlobalVarDef(C))
+ return true;
+
return false;
}
(md->getName().str() == "llvm.dbg.sp")))
return true;
- for (User::const_use_iterator ui = U->use_begin(), ue = U->use_end();
- ui != ue; ++ui) {
- if (usedInOneFunc(*ui, oneFunc) == false)
+ for (const User *UU : U->users())
+ if (usedInOneFunc(UU, oneFunc) == false)
return false;
- }
+
return true;
}
if (Pty->getAddressSpace() != llvm::ADDRESS_SPACE_SHARED)
return false;
- const Function *oneFunc = 0;
+ const Function *oneFunc = nullptr;
bool flag = usedInOneFunc(gv, oneFunc);
if (flag == false)
static bool useFuncSeen(const Constant *C,
llvm::DenseMap<const Function *, bool> &seenMap) {
- for (Value::const_use_iterator ui = C->use_begin(), ue = C->use_end();
- ui != ue; ++ui) {
- if (const Constant *cu = dyn_cast<Constant>(*ui)) {
+ for (const User *U : C->users()) {
+ if (const Constant *cu = dyn_cast<Constant>(U)) {
if (useFuncSeen(cu, seenMap))
return true;
- } else if (const Instruction *I = dyn_cast<Instruction>(*ui)) {
+ } else if (const Instruction *I = dyn_cast<Instruction>(U)) {
const BasicBlock *bb = I->getParent();
if (!bb)
continue;
emitDeclaration(F, O);
continue;
}
- for (Value::const_use_iterator iter = F->use_begin(),
- iterEnd = F->use_end();
- iter != iterEnd; ++iter) {
- if (const Constant *C = dyn_cast<Constant>(*iter)) {
+ for (const User *U : F->users()) {
+ if (const Constant *C = dyn_cast<Constant>(U)) {
if (usedInGlobalVarDef(C)) {
// The use is in the initialization of a global variable
// that is a function pointer, so print a declaration
}
}
- if (!isa<Instruction>(*iter))
+ if (!isa<Instruction>(U))
continue;
- const Instruction *instr = cast<Instruction>(*iter);
+ const Instruction *instr = cast<Instruction>(U);
const BasicBlock *bb = instr->getParent();
if (!bb)
continue;
DbgFinder.processModule(M);
unsigned i = 1;
- for (DebugInfoFinder::iterator I = DbgFinder.compile_unit_begin(),
- E = DbgFinder.compile_unit_end();
- I != E; ++I) {
- DICompileUnit DIUnit(*I);
+ for (DICompileUnit DIUnit : DbgFinder.compile_units()) {
StringRef Filename(DIUnit.getFilename());
StringRef Dirname(DIUnit.getDirectory());
SmallString<128> FullPathName = Dirname;
++i;
}
- for (DebugInfoFinder::iterator I = DbgFinder.subprogram_begin(),
- E = DbgFinder.subprogram_end();
- I != E; ++I) {
- DISubprogram SP(*I);
+ for (DISubprogram SP : DbgFinder.subprograms()) {
StringRef Filename(SP.getFilename());
StringRef Dirname(SP.getDirectory());
SmallString<128> FullPathName = Dirname;
const_cast<TargetLoweringObjectFile &>(getObjFileLowering())
.Initialize(OutContext, TM);
- Mang = new Mangler(OutContext, &TM);
+ Mang = new Mangler(TM.getDataLayout());
// Emit header before any dwarf directives are emitted below.
emitHeader(M, OS1);
// Already commented out
//bool Result = AsmPrinter::doInitialization(M);
+ // Emit module-level inline asm if it exists.
+ if (!M.getModuleInlineAsm().empty()) {
+ OutStreamer.AddComment("Start of file scope inline assembly");
+ OutStreamer.AddBlankLine();
+ OutStreamer.EmitRawText(StringRef(M.getModuleInlineAsm()));
+ OutStreamer.AddBlankLine();
+ OutStreamer.AddComment("End of file scope inline assembly");
+ OutStreamer.AddBlankLine();
+ }
+
if (nvptxSubtarget.getDrvInterface() == NVPTX::CUDA)
recordAndEmitFilenames(M);
for (i = 0; i < n; i++)
global_list.insert(global_list.end(), gv_array[i]);
+ clearAnnotationCache(&M);
+
delete[] gv_array;
return ret;
// external global variable with init -> .visible
// external without init -> .extern
// appending -> not allowed, assert.
+// for any linkage other than
+// internal, private, linker_private,
+// linker_private_weak, linker_private_weak_def_auto,
+// we emit -> .weak.
void NVPTXAsmPrinter::emitLinkageDirective(const GlobalValue *V,
raw_ostream &O) {
msg.append(V->getName().str());
msg.append("has unsupported appending linkage type");
llvm_unreachable(msg.c_str());
+ } else if (!V->hasInternalLinkage() &&
+ !V->hasPrivateLinkage()) {
+ O << ".weak ";
}
}
}
// Skip meta data
if (GVar->hasSection()) {
- if (GVar->getSection() == "llvm.metadata")
+ if (GVar->getSection() == StringRef("llvm.metadata"))
return;
}
+ // Skip LLVM intrinsic global variables
+ if (GVar->getName().startswith("llvm.") ||
+ GVar->getName().startswith("nvvm."))
+ return;
+
const DataLayout *TD = TM.getDataLayout();
// GlobalVariables are always constant pointers themselves.
O << ".visible ";
else
O << ".extern ";
+ } else if (GVar->hasLinkOnceLinkage() || GVar->hasWeakLinkage() ||
+ GVar->hasAvailableExternallyLinkage() ||
+ GVar->hasCommonLinkage()) {
+ O << ".weak ";
}
if (llvm::isTexture(*GVar)) {
if (llvm::isSampler(*GVar)) {
O << ".global .samplerref " << llvm::getSamplerName(*GVar);
- const Constant *Initializer = NULL;
+ const Constant *Initializer = nullptr;
if (GVar->hasInitializer())
Initializer = GVar->getInitializer();
- const ConstantInt *CI = NULL;
+ const ConstantInt *CI = nullptr;
if (Initializer)
CI = dyn_cast<ConstantInt>(Initializer);
if (CI) {
O << "linear";
break;
case 2:
- assert(0 && "Anisotropic filtering is not supported");
+ llvm_unreachable("Anisotropic filtering is not supported");
default:
O << "nearest";
break;
return;
}
- const Function *demotedFunc = 0;
+ const Function *demotedFunc = nullptr;
if (!processDemoted && canDemoteGlobalVar(GVar, demotedFunc)) {
O << "// " << GVar->getName().str() << " has been demoted\n";
if (localDecls.find(demotedFunc) != localDecls.end())
O << ".";
emitPTXAddressSpace(PTy->getAddressSpace(), O);
+
+ if (isManaged(*GVar)) {
+ O << " .attribute(.managed)";
+ }
+
if (GVar->getAlignment() == 0)
O << " .align " << (int) TD->getPrefTypeAlignment(ETy);
else
O << " .align " << GVar->getAlignment();
- if (ETy->isPrimitiveType() || ETy->isIntegerTy() || isa<PointerType>(ETy)) {
+ if (ETy->isSingleValueType()) {
O << " .";
// Special case: ABI requires that we use .u8 for predicates
if (ETy->isIntegerTy(1))
else
O << getPTXFundamentalTypeStr(ETy, false);
O << " ";
- O << *Mang->getSymbol(GVar);
+ O << *getSymbol(GVar);
// Ptx allows variable initilization only for constant and global state
// spaces.
- if (((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
- (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST_NOT_GEN) ||
- (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) &&
- GVar->hasInitializer()) {
- const Constant *Initializer = GVar->getInitializer();
- if (!Initializer->isNullValue()) {
- O << " = ";
- printScalarConstant(Initializer, O);
+ if (GVar->hasInitializer()) {
+ if ((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
+ (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) {
+ const Constant *Initializer = GVar->getInitializer();
+ // 'undef' is treated as there is no value spefied.
+ if (!Initializer->isNullValue() && !isa<UndefValue>(Initializer)) {
+ O << " = ";
+ printScalarConstant(Initializer, O);
+ }
+ } else {
+ // The frontend adds zero-initializer to variables that don't have an
+ // initial value, so skip warning for this case.
+ if (!GVar->getInitializer()->isNullValue()) {
+ std::string warnMsg = "initial value of '" + GVar->getName().str() +
+ "' is not allowed in addrspace(" +
+ llvm::utostr_32(PTy->getAddressSpace()) + ")";
+ report_fatal_error(warnMsg.c_str());
+ }
}
}
} else {
// Ptx allows variable initilization only for constant and
// global state spaces.
if (((PTy->getAddressSpace() == llvm::ADDRESS_SPACE_GLOBAL) ||
- (PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST_NOT_GEN) ||
(PTy->getAddressSpace() == llvm::ADDRESS_SPACE_CONST)) &&
GVar->hasInitializer()) {
const Constant *Initializer = GVar->getInitializer();
bufferAggregateConstant(Initializer, &aggBuffer);
if (aggBuffer.numSymbols) {
if (nvptxSubtarget.is64Bit()) {
- O << " .u64 " << *Mang->getSymbol(GVar) << "[";
+ O << " .u64 " << *getSymbol(GVar) << "[";
O << ElementSize / 8;
} else {
- O << " .u32 " << *Mang->getSymbol(GVar) << "[";
+ O << " .u32 " << *getSymbol(GVar) << "[";
O << ElementSize / 4;
}
O << "]";
} else {
- O << " .b8 " << *Mang->getSymbol(GVar) << "[";
+ O << " .b8 " << *getSymbol(GVar) << "[";
O << ElementSize;
O << "]";
}
aggBuffer.print();
O << "}";
} else {
- O << " .b8 " << *Mang->getSymbol(GVar);
+ O << " .b8 " << *getSymbol(GVar);
if (ElementSize) {
O << "[";
O << ElementSize;
}
}
} else {
- O << " .b8 " << *Mang->getSymbol(GVar);
+ O << " .b8 " << *getSymbol(GVar);
if (ElementSize) {
O << "[";
O << ElementSize;
}
break;
default:
- assert(0 && "type not supported yet");
+ llvm_unreachable("type not supported yet");
}
}
O << "global";
break;
case llvm::ADDRESS_SPACE_CONST:
- // This logic should be consistent with that in
- // getCodeAddrSpace() (NVPTXISelDATToDAT.cpp)
- if (nvptxSubtarget.hasGenericLdSt())
- O << "global";
- else
- O << "const";
- break;
- case llvm::ADDRESS_SPACE_CONST_NOT_GEN:
O << "const";
break;
case llvm::ADDRESS_SPACE_SHARED:
return "u32";
}
llvm_unreachable("unexpected type");
- return NULL;
+ return nullptr;
}
void NVPTXAsmPrinter::emitPTXGlobalVariable(const GlobalVariable *GVar,
else
O << " .align " << GVar->getAlignment();
- if (ETy->isPrimitiveType() || ETy->isIntegerTy() || isa<PointerType>(ETy)) {
+ if (ETy->isSingleValueType()) {
O << " .";
O << getPTXFundamentalTypeStr(ETy);
O << " ";
- O << *Mang->getSymbol(GVar);
+ O << *getSymbol(GVar);
return;
}
case Type::ArrayTyID:
case Type::VectorTyID:
ElementSize = TD->getTypeStoreSize(ETy);
- O << " .b8 " << *Mang->getSymbol(GVar) << "[";
+ O << " .b8 " << *getSymbol(GVar) << "[";
if (ElementSize) {
O << itostr(ElementSize);
}
O << "]";
break;
default:
- assert(0 && "type not supported yet");
+ llvm_unreachable("type not supported yet");
}
return;
}
static unsigned int getOpenCLAlignment(const DataLayout *TD, Type *Ty) {
- if (Ty->isPrimitiveType() || Ty->isIntegerTy() || isa<PointerType>(Ty))
+ if (Ty->isSingleValueType())
return TD->getPrefTypeAlignment(Ty);
const ArrayType *ATy = dyn_cast<ArrayType>(Ty);
int paramIndex, raw_ostream &O) {
if ((nvptxSubtarget.getDrvInterface() == NVPTX::NVCL) ||
(nvptxSubtarget.getDrvInterface() == NVPTX::CUDA))
- O << *Mang->getSymbol(I->getParent()) << "_param_" << paramIndex;
+ O << *getSymbol(I->getParent()) << "_param_" << paramIndex;
else {
std::string argName = I->getName();
const char *p = argName.c_str();
first = false;
// Handle image/sampler parameters
- if (llvm::isSampler(*I) || llvm::isImage(*I)) {
- if (llvm::isImage(*I)) {
- std::string sname = I->getName();
- if (llvm::isImageWriteOnly(*I))
- O << "\t.param .surfref " << *Mang->getSymbol(F) << "_param_"
- << paramIndex;
- else // Default image is read_only
- O << "\t.param .texref " << *Mang->getSymbol(F) << "_param_"
- << paramIndex;
- } else // Should be llvm::isSampler(*I)
- O << "\t.param .samplerref " << *Mang->getSymbol(F) << "_param_"
- << paramIndex;
- continue;
+ if (isKernelFunction(*F)) {
+ if (isSampler(*I) || isImage(*I)) {
+ if (isImage(*I)) {
+ std::string sname = I->getName();
+ if (isImageWriteOnly(*I) || isImageReadWrite(*I)) {
+ if (nvptxSubtarget.hasImageHandles())
+ O << "\t.param .u64 .ptr .surfref ";
+ else
+ O << "\t.param .surfref ";
+ O << *CurrentFnSym << "_param_" << paramIndex;
+ }
+ else { // Default image is read_only
+ if (nvptxSubtarget.hasImageHandles())
+ O << "\t.param .u64 .ptr .texref ";
+ else
+ O << "\t.param .texref ";
+ O << *CurrentFnSym << "_param_" << paramIndex;
+ }
+ } else {
+ if (nvptxSubtarget.hasImageHandles())
+ O << "\t.param .u64 .ptr .samplerref ";
+ else
+ O << "\t.param .samplerref ";
+ O << *CurrentFnSym << "_param_" << paramIndex;
+ }
+ continue;
+ }
}
if (PAL.hasAttribute(paramIndex + 1, Attribute::ByVal) == false) {
- if (Ty->isVectorTy()) {
- // Just print .param .b8 .align <a> .param[size];
+ if (Ty->isAggregateType() || Ty->isVectorTy()) {
+ // Just print .param .align <a> .b8 .param[size];
// <a> = PAL.getparamalignment
// size = typeallocsize of element type
unsigned align = PAL.getParamAlignment(paramIndex + 1);
default:
O << ".ptr ";
break;
- case llvm::ADDRESS_SPACE_CONST_NOT_GEN:
+ case llvm::ADDRESS_SPACE_CONST:
O << ".ptr .const ";
break;
case llvm::ADDRESS_SPACE_SHARED:
O << ".ptr .shared ";
break;
case llvm::ADDRESS_SPACE_GLOBAL:
- case llvm::ADDRESS_SPACE_CONST:
O << ".ptr .global ";
break;
}
continue;
}
// Non-kernel function, just print .param .b<size> for ABI
- // and .reg .b<size> for non ABY
+ // and .reg .b<size> for non-ABI
unsigned sz = 0;
if (isa<IntegerType>(Ty)) {
sz = cast<IntegerType>(Ty)->getBitWidth();
Type *ETy = PTy->getElementType();
if (isABI || isKernelFunc) {
- // Just print .param .b8 .align <a> .param[size];
+ // Just print .param .align <a> .b8 .param[size];
// <a> = PAL.getparamalignment
// size = typeallocsize of element type
unsigned align = PAL.getParamAlignment(paramIndex + 1);
for (unsigned i = 0; i < numVRs; i++) {
unsigned int vr = TRI->index2VirtReg(i);
const TargetRegisterClass *RC = MRI->getRegClass(vr);
- std::map<unsigned, unsigned> ®map = VRidGlobal2LocalMap[RC->getID()];
+ DenseMap<unsigned, unsigned> ®map = VRegMapping[RC];
int n = regmap.size();
regmap.insert(std::make_pair(vr, n + 1));
}
// Emit register declarations
// @TODO: Extract out the real register usage
- O << "\t.reg .pred %p<" << NVPTXNumRegisters << ">;\n";
- O << "\t.reg .s16 %rc<" << NVPTXNumRegisters << ">;\n";
- O << "\t.reg .s16 %rs<" << NVPTXNumRegisters << ">;\n";
- O << "\t.reg .s32 %r<" << NVPTXNumRegisters << ">;\n";
- O << "\t.reg .s64 %rl<" << NVPTXNumRegisters << ">;\n";
- O << "\t.reg .f32 %f<" << NVPTXNumRegisters << ">;\n";
- O << "\t.reg .f64 %fl<" << NVPTXNumRegisters << ">;\n";
+ // O << "\t.reg .pred %p<" << NVPTXNumRegisters << ">;\n";
+ // O << "\t.reg .s16 %rc<" << NVPTXNumRegisters << ">;\n";
+ // O << "\t.reg .s16 %rs<" << NVPTXNumRegisters << ">;\n";
+ // O << "\t.reg .s32 %r<" << NVPTXNumRegisters << ">;\n";
+ // O << "\t.reg .s64 %rl<" << NVPTXNumRegisters << ">;\n";
+ // O << "\t.reg .f32 %f<" << NVPTXNumRegisters << ">;\n";
+ // O << "\t.reg .f64 %fl<" << NVPTXNumRegisters << ">;\n";
// Emit declaration of the virtual registers or 'physical' registers for
// each register class
- //for (unsigned i=0; i< numRegClasses; i++) {
- // std::map<unsigned, unsigned> ®map = VRidGlobal2LocalMap[i];
- // const TargetRegisterClass *RC = TRI->getRegClass(i);
- // std::string rcname = getNVPTXRegClassName(RC);
- // std::string rcStr = getNVPTXRegClassStr(RC);
- // //int n = regmap.size();
- // if (!isNVPTXVectorRegClass(RC)) {
- // O << "\t.reg " << rcname << " \t" << rcStr << "<"
- // << NVPTXNumRegisters << ">;\n";
- // }
-
- // Only declare those registers that may be used. And do not emit vector
- // registers as
- // they are all elementized to scalar registers.
- //if (n && !isNVPTXVectorRegClass(RC)) {
- // if (RegAllocNilUsed) {
- // O << "\t.reg " << rcname << " \t" << rcStr << "<" << (n+1)
- // << ">;\n";
- // }
- // else {
- // O << "\t.reg " << rcname << " \t" << StrToUpper(rcStr)
- // << "<" << 32 << ">;\n";
- // }
- //}
- //}
+ for (unsigned i=0; i< TRI->getNumRegClasses(); i++) {
+ const TargetRegisterClass *RC = TRI->getRegClass(i);
+ DenseMap<unsigned, unsigned> ®map = VRegMapping[RC];
+ std::string rcname = getNVPTXRegClassName(RC);
+ std::string rcStr = getNVPTXRegClassStr(RC);
+ int n = regmap.size();
+
+ // Only declare those registers that may be used.
+ if (n) {
+ O << "\t.reg " << rcname << " \t" << rcStr << "<" << (n+1)
+ << ">;\n";
+ }
+ }
OutStreamer.EmitRawText(O.str());
}
return;
}
if (const GlobalValue *GVar = dyn_cast<GlobalValue>(CPV)) {
- O << *Mang->getSymbol(GVar);
+ PointerType *PTy = dyn_cast<PointerType>(GVar->getType());
+ bool IsNonGenericPointer = false;
+ if (PTy && PTy->getAddressSpace() != 0) {
+ IsNonGenericPointer = true;
+ }
+ if (EmitGeneric && !isa<Function>(CPV) && !IsNonGenericPointer) {
+ O << "generic(";
+ O << *getSymbol(GVar);
+ O << ")";
+ } else {
+ O << *getSymbol(GVar);
+ }
return;
}
if (const ConstantExpr *Cexpr = dyn_cast<ConstantExpr>(CPV)) {
const Value *v = Cexpr->stripPointerCasts();
+ PointerType *PTy = dyn_cast<PointerType>(Cexpr->getType());
+ bool IsNonGenericPointer = false;
+ if (PTy && PTy->getAddressSpace() != 0) {
+ IsNonGenericPointer = true;
+ }
if (const GlobalValue *GVar = dyn_cast<GlobalValue>(v)) {
- O << *Mang->getSymbol(GVar);
+ if (EmitGeneric && !isa<Function>(v) && !IsNonGenericPointer) {
+ O << "generic(";
+ O << *getSymbol(GVar);
+ O << ")";
+ } else {
+ O << *getSymbol(GVar);
+ }
return;
} else {
O << *LowerConstant(CPV, *this);
case Type::VectorTyID:
case Type::StructTyID: {
if (isa<ConstantArray>(CPV) || isa<ConstantVector>(CPV) ||
- isa<ConstantStruct>(CPV)) {
+ isa<ConstantStruct>(CPV) || isa<ConstantDataSequential>(CPV)) {
int ElementSize = TD->getTypeAllocSize(CPV->getType());
bufferAggregateConstant(CPV, aggBuffer);
if (Bytes > ElementSize)
return false;
}
-/// PrintAsmOperand - Print out an operand for an inline asm expression.
-///
-bool NVPTXAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
- unsigned AsmVariant,
- const char *ExtraCode, raw_ostream &O) {
- if (ExtraCode && ExtraCode[0]) {
- if (ExtraCode[1] != 0)
- return true; // Unknown modifier.
-
- switch (ExtraCode[0]) {
- default:
- // See if this is a generic print operand
- return AsmPrinter::PrintAsmOperand(MI, OpNo, AsmVariant, ExtraCode, O);
- case 'r':
- break;
- }
- }
-
- printOperand(MI, OpNo, O);
-
- return false;
-}
-
-bool NVPTXAsmPrinter::PrintAsmMemoryOperand(
- const MachineInstr *MI, unsigned OpNo, unsigned AsmVariant,
- const char *ExtraCode, raw_ostream &O) {
- if (ExtraCode && ExtraCode[0])
- return true; // Unknown modifier
-
- O << '[';
- printMemOperand(MI, OpNo, O);
- O << ']';
-
- return false;
-}
bool NVPTXAsmPrinter::ignoreLoc(const MachineInstr &MI) {
switch (MI.getOpcode()) {
case NVPTX::CallArgI32:
case NVPTX::CallArgI32imm:
case NVPTX::CallArgI64:
- case NVPTX::CallArgI8:
case NVPTX::CallArgParam:
case NVPTX::CallVoidInst:
case NVPTX::CallVoidInstReg:
case NVPTX::StoreParamI32:
case NVPTX::StoreParamI64:
case NVPTX::StoreParamI8:
- case NVPTX::StoreParamS32I8:
- case NVPTX::StoreParamU32I8:
- case NVPTX::StoreParamS32I16:
- case NVPTX::StoreParamU32I16:
case NVPTX::StoreRetvalF32:
case NVPTX::StoreRetvalF64:
case NVPTX::StoreRetvalI16:
case NVPTX::LastCallArgI32:
case NVPTX::LastCallArgI32imm:
case NVPTX::LastCallArgI64:
- case NVPTX::LastCallArgI8:
case NVPTX::LastCallArgParam:
case NVPTX::LoadParamMemF32:
case NVPTX::LoadParamMemF64:
case NVPTX::LoadParamMemI32:
case NVPTX::LoadParamMemI64:
case NVPTX::LoadParamMemI8:
- case NVPTX::LoadParamRegF32:
- case NVPTX::LoadParamRegF64:
- case NVPTX::LoadParamRegI16:
- case NVPTX::LoadParamRegI32:
- case NVPTX::LoadParamRegI64:
- case NVPTX::LoadParamRegI8:
case NVPTX::PrototypeInst:
case NVPTX::DBG_VALUE:
return true;
return false;
}
+/// PrintAsmOperand - Print out an operand for an inline asm expression.
+///
+bool NVPTXAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
+ unsigned AsmVariant,
+ const char *ExtraCode, raw_ostream &O) {
+ if (ExtraCode && ExtraCode[0]) {
+ if (ExtraCode[1] != 0)
+ return true; // Unknown modifier.
+
+ switch (ExtraCode[0]) {
+ default:
+ // See if this is a generic print operand
+ return AsmPrinter::PrintAsmOperand(MI, OpNo, AsmVariant, ExtraCode, O);
+ case 'r':
+ break;
+ }
+ }
+
+ printOperand(MI, OpNo, O);
+
+ return false;
+}
+
+bool NVPTXAsmPrinter::PrintAsmMemoryOperand(
+ const MachineInstr *MI, unsigned OpNo, unsigned AsmVariant,
+ const char *ExtraCode, raw_ostream &O) {
+ if (ExtraCode && ExtraCode[0])
+ return true; // Unknown modifier
+
+ O << '[';
+ printMemOperand(MI, OpNo, O);
+ O << ']';
+
+ return false;
+}
+
+void NVPTXAsmPrinter::printOperand(const MachineInstr *MI, int opNum,
+ raw_ostream &O, const char *Modifier) {
+ const MachineOperand &MO = MI->getOperand(opNum);
+ switch (MO.getType()) {
+ case MachineOperand::MO_Register:
+ if (TargetRegisterInfo::isPhysicalRegister(MO.getReg())) {
+ if (MO.getReg() == NVPTX::VRDepot)
+ O << DEPOTNAME << getFunctionNumber();
+ else
+ O << NVPTXInstPrinter::getRegisterName(MO.getReg());
+ } else {
+ emitVirtualRegister(MO.getReg(), O);
+ }
+ return;
+
+ case MachineOperand::MO_Immediate:
+ if (!Modifier)
+ O << MO.getImm();
+ else if (strstr(Modifier, "vec") == Modifier)
+ printVecModifiedImmediate(MO, Modifier, O);
+ else
+ llvm_unreachable(
+ "Don't know how to handle modifier on immediate operand");
+ return;
+
+ case MachineOperand::MO_FPImmediate:
+ printFPConstant(MO.getFPImm(), O);
+ break;
+
+ case MachineOperand::MO_GlobalAddress:
+ O << *getSymbol(MO.getGlobal());
+ break;
+
+ case MachineOperand::MO_MachineBasicBlock:
+ O << *MO.getMBB()->getSymbol();
+ return;
+
+ default:
+ llvm_unreachable("Operand type not supported.");
+ }
+}
+
+void NVPTXAsmPrinter::printMemOperand(const MachineInstr *MI, int opNum,
+ raw_ostream &O, const char *Modifier) {
+ printOperand(MI, opNum, O);
+
+ if (Modifier && !strcmp(Modifier, "add")) {
+ O << ", ";
+ printOperand(MI, opNum + 1, O);
+ } else {
+ if (MI->getOperand(opNum + 1).isImm() &&
+ MI->getOperand(opNum + 1).getImm() == 0)
+ return; // don't print ',0' or '+0'
+ O << "+";
+ printOperand(MI, opNum + 1, O);
+ }
+}
+
+
// Force static initialization.
extern "C" void LLVMInitializeNVPTXBackendAsmPrinter() {
RegisterAsmPrinter<NVPTXAsmPrinter> X(TheNVPTXTarget32);
}
LineReader *NVPTXAsmPrinter::getReader(std::string filename) {
- if (reader == NULL) {
+ if (!reader) {
reader = new LineReader(filename);
}
projects / oota-llvm.git / blobdiff