NVPTXSubtarget.cpp

Go to the documentation of this file.

//===- NVPTXSubtarget.cpp - NVPTX Subtarget Information -------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file implements the NVPTX specific subclass of TargetSubtarget.
//
//===----------------------------------------------------------------------===//
 
#include "NVPTXSubtarget.h"
#include "NVPTXSelectionDAGInfo.h"
#include "NVPTXTargetMachine.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
#include "llvm/Option/LibraryOptions.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/FormatVariadic.h"
 
using namespace llvm;
 
#define DEBUG_TYPE "nvptx-subtarget"
 
#define GET_SUBTARGETINFO_TARGET_DESC
#define GET_SUBTARGETINFO_CTOR
#include "NVPTXGenSubtargetInfo.inc"
 
#define OPTIONS_STRUCT_DEFS
#include "NVPTXOptions.inc"
 
const NVPTXOptions &llvm::getNVPTXOptions(const LLVMContext &C) {
  return C.getLibraryOptions().get<NVPTXOptions>();
}
 
const NVPTXOptions &llvm::getNVPTXOptions(const Module &M) {
  return getNVPTXOptions(M.getContext());
}
 
const NVPTXOptions &llvm::getNVPTXOptions(const Function &F) {
  return getNVPTXOptions(F.getContext());
}
 
// Pin the vtable to this file.
void NVPTXSubtarget::anchor() {}
 
// Returns the minimum PTX version required for a given SM target.
// This must be kept in sync with the "Supported Targets" column of the
// "PTX Release History" table in the PTX ISA documentation:
// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#release-notes-ptx-release-history
//
// Note: LLVM's minimum supported PTX version is 3.2 (see FeaturePTX in
// NVPTX.td), so older SMs that supported earlier PTX versions instead use 3.2
// as their effective minimum.
//
// Note: sm_110* is the PTX ISA 9.0 spelling of sm_101*, which was introduced in
// PTX ISA 8.6 (sm_101f in 8.8) and renamed in 9.0. Since getTargetName()
// rewrites the new spelling back to the old one when emitting older PTX, the
// sm_110* names work with any PTX version that supported sm_101*, so they take
// sm_101*'s minimum PTX version rather than the 9.0 of the rename.
static unsigned minPTXVersion(NVPTX::GPUKind Arch) {
  switch (Arch) {
  case NVPTX::GK_NONE:
    llvm_unreachable("architecture is resolved before this is reached");
  case NVPTX::GK_SM_20:
  case NVPTX::GK_SM_21:
  case NVPTX::GK_SM_30:
  case NVPTX::GK_SM_35:
    return 32;
  case NVPTX::GK_SM_32_:
  case NVPTX::GK_SM_50:
    return 40;
  case NVPTX::GK_SM_37:
  case NVPTX::GK_SM_52:
    return 41;
  case NVPTX::GK_SM_53:
    return 42;
  case NVPTX::GK_SM_60:
  case NVPTX::GK_SM_61:
  case NVPTX::GK_SM_62:
    return 50;
  case NVPTX::GK_SM_70:
    return 60;
  case NVPTX::GK_SM_72:
    return 61;
  case NVPTX::GK_SM_75:
    return 63;
  case NVPTX::GK_SM_80:
    return 70;
  case NVPTX::GK_SM_86:
    return 71;
  case NVPTX::GK_SM_87:
    return 74;
  case NVPTX::GK_SM_89:
  case NVPTX::GK_SM_90:
    return 78;
  case NVPTX::GK_SM_90a:
    return 80;
  case NVPTX::GK_SM_100:
  case NVPTX::GK_SM_100a:
  case NVPTX::GK_SM_101:
  case NVPTX::GK_SM_101a:
  case NVPTX::GK_SM_110:
  case NVPTX::GK_SM_110a:
    return 86;
  case NVPTX::GK_SM_120:
  case NVPTX::GK_SM_120a:
    return 87;
  case NVPTX::GK_SM_100f:
  case NVPTX::GK_SM_101f:
  case NVPTX::GK_SM_103:
  case NVPTX::GK_SM_103f:
  case NVPTX::GK_SM_103a:
  case NVPTX::GK_SM_110f:
  case NVPTX::GK_SM_120f:
  case NVPTX::GK_SM_121:
  case NVPTX::GK_SM_121f:
  case NVPTX::GK_SM_121a:
    return 88;
  case NVPTX::GK_SM_88:
    return 90;
  case NVPTX::GK_SM_107:
  case NVPTX::GK_SM_107f:
  case NVPTX::GK_SM_107a:
    return 94;
  }
  llvm_unreachable("invalid NVPTX GPUKind");
}
 
StringRef NVPTXSubtarget::getTargetName() const {
  // Spell sm_110* as sm_101* when emitting PTX older than 9.0, which does not
  // know the new names. See the note above minPTXVersion().
  if (!hasFeature(NVPTX::PTX90))
    switch (Arch) {
    case NVPTX::GK_SM_110:
      return "sm_101";
    case NVPTX::GK_SM_110f:
      return "sm_101f";
    case NVPTX::GK_SM_110a:
      return "sm_101a";
    default:
      break;
    }
 
  return NVPTX::getArchName(Arch);
}
 
NVPTXSubtarget &NVPTXSubtarget::initializeSubtargetDependencies(StringRef CPU,
                                                                StringRef FS) {
  // If the user did not provide a target we default to the `sm_75` target.
  StringRef RequestedCPU = CPU.empty() ? StringRef("sm_75") : CPU;
  ParseSubtargetFeatures(RequestedCPU, /*TuneCPU=*/RequestedCPU, FS);
 
  Arch = NVPTX::parseArch(RequestedCPU);
 
  // An unrecognized name has already been diagnosed and its features dropped,
  // leaving the subtarget at the oldest architecture, so name it that.
  if (Arch == NVPTX::GK_NONE)
    Arch = NVPTX::GK_SM_20;
 
  unsigned MinPTX = minPTXVersion(Arch);
 
  if (PTXVersion == 0) {
    // User didn't request a specific PTX version; use the minimum for this SM.
    ApplyFeatureFlag(("+ptx" + Twine(MinPTX)).str());
    PTXVersion = MinPTX;
  } else if (PTXVersion < MinPTX) {
    // User explicitly requested an insufficient PTX version.
    reportFatalUsageError(
        formatv("PTX version {0}.{1} does not support target '{2}'. "
                "Minimum required PTX version is {3}.{4}. "
                "Either remove the PTX version to use the default, "
                "or increase it to at least {3}.{4}.",
                PTXVersion / 10, PTXVersion % 10, RequestedCPU, MinPTX / 10,
                MinPTX % 10));
  }
 
  return *this;
}
 
NVPTXSubtarget::NVPTXSubtarget(const Triple &TT, StringRef CPU, StringRef FS,
                               const NVPTXTargetMachine &TM)
    : NVPTXGenSubtargetInfo(TT, CPU, /*TuneCPU=*/CPU, FS),
      CLOpts(TM.getCLOpts()), PTXVersion(0),
      InstrInfo(initializeSubtargetDependencies(CPU, FS)), TLInfo(TM, *this),
      TSInfo(std::make_unique<NVPTXSelectionDAGInfo>()) {}
 
NVPTXSubtarget::~NVPTXSubtarget() = default;
 
bool NVPTXSubtarget::allowFP16Math() const {
  return hasFP16Math() && !CLOpts.no_f16_math;
}
 
bool NVPTXSubtarget::hasF32x2Instructions() const {
  return hasFeature(NVPTX::SM100) && !CLOpts.no_f32x2;
}
 
bool NVPTXSubtarget::hasNativeBF16Support(unsigned Opcode) const {
  if (!hasBF16Math())
    return false;
 
  switch (Opcode) {
  // Several BF16 instructions are available on sm_90 only.
  case ISD::FADD:
  case ISD::FMUL:
  case ISD::FSUB:
  case ISD::SELECT:
  case ISD::SELECT_CC:
  case ISD::SETCC:
  case ISD::FEXP2:
  case ISD::FTANH:
  case ISD::FCEIL:
  case ISD::FFLOOR:
  case ISD::FNEARBYINT:
  case ISD::FRINT:
  case ISD::FROUNDEVEN:
  case ISD::FTRUNC:
    return hasFeature(NVPTX::SM90);
  // Several BF16 instructions are available on sm_80 only.
  case ISD::FMINNUM:
  case ISD::FMAXNUM:
  case ISD::FMAXNUM_IEEE:
  case ISD::FMINNUM_IEEE:
  case ISD::FMAXIMUM:
  case ISD::FMINIMUM:
    return hasFeature(NVPTX::SM80);
  }
  return true;
}