reference, declarationdefinition
definition → references, declarations, derived classes, virtual overrides
reference to multiple definitions → definitions
unreferenced
    1
    2
    3
    4
    5
    6
    7
    8
    9
   10
   11
   12
   13
   14
   15
   16
   17
   18
   19
   20
   21
   22
   23
   24
   25
   26
   27
   28
   29
   30
   31
   32
   33
   34
   35
   36
   37
   38
   39
   40
   41
   42
   43
   44
   45
   46
   47
   48
   49
   50
   51
   52
   53
   54
   55
   56
   57
   58
   59
   60
   61
   62
   63
   64
   65
   66
   67
   68
   69
   70
   71
   72
   73
   74
   75
   76
   77
   78
   79
   80
   81
   82
   83
   84
   85
   86
   87
   88
   89
   90
   91
   92
   93
   94
   95
   96
   97
   98
   99
  100
  101
  102
  103
  104
  105
  106
  107
  108
  109
  110
  111
  112
  113
  114
  115
  116
  117
  118
  119
  120
  121
  122
  123
  124
  125
  126
  127
  128
  129
  130
  131
  132
  133
  134
  135
  136
  137
  138
  139
  140
  141
  142
  143
  144
  145
  146
  147
  148
  149
  150
  151
  152
  153
  154
  155
  156
  157
  158
  159
  160
  161
  162
  163
  164
  165
  166
  167
  168
//===-- AMDGPULowerIntrinsics.cpp -----------------------------------------===//
//
// 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
//
//===----------------------------------------------------------------------===//

#include "AMDGPU.h"
#include "AMDGPUSubtarget.h"
#include "llvm/CodeGen/TargetPassConfig.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Module.h"
#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"

#define DEBUG_TYPE "amdgpu-lower-intrinsics"

using namespace llvm;

namespace {

const unsigned MaxStaticSize = 1024;

class AMDGPULowerIntrinsics : public ModulePass {
private:
  bool makeLIDRangeMetadata(Function &F) const;

public:
  static char ID;

  AMDGPULowerIntrinsics() : ModulePass(ID) {}

  bool runOnModule(Module &M) override;
  bool expandMemIntrinsicUses(Function &F);
  StringRef getPassName() const override {
    return "AMDGPU Lower Intrinsics";
  }

  void getAnalysisUsage(AnalysisUsage &AU) const override {
    AU.addRequired<TargetTransformInfoWrapperPass>();
  }
};

}

char AMDGPULowerIntrinsics::ID = 0;

char &llvm::AMDGPULowerIntrinsicsID = AMDGPULowerIntrinsics::ID;

INITIALIZE_PASS(AMDGPULowerIntrinsics, DEBUG_TYPE, "Lower intrinsics", false,
                false)

// TODO: Should refine based on estimated number of accesses (e.g. does it
// require splitting based on alignment)
static bool shouldExpandOperationWithSize(Value *Size) {
  ConstantInt *CI = dyn_cast<ConstantInt>(Size);
  return !CI || (CI->getZExtValue() > MaxStaticSize);
}

bool AMDGPULowerIntrinsics::expandMemIntrinsicUses(Function &F) {
  Intrinsic::ID ID = F.getIntrinsicID();
  bool Changed = false;

  for (auto I = F.user_begin(), E = F.user_end(); I != E;) {
    Instruction *Inst = cast<Instruction>(*I);
    ++I;

    switch (ID) {
    case Intrinsic::memcpy: {
      auto *Memcpy = cast<MemCpyInst>(Inst);
      if (shouldExpandOperationWithSize(Memcpy->getLength())) {
        Function *ParentFunc = Memcpy->getParent()->getParent();
        const TargetTransformInfo &TTI =
            getAnalysis<TargetTransformInfoWrapperPass>().getTTI(*ParentFunc);
        expandMemCpyAsLoop(Memcpy, TTI);
        Changed = true;
        Memcpy->eraseFromParent();
      }

      break;
    }
    case Intrinsic::memmove: {
      auto *Memmove = cast<MemMoveInst>(Inst);
      if (shouldExpandOperationWithSize(Memmove->getLength())) {
        expandMemMoveAsLoop(Memmove);
        Changed = true;
        Memmove->eraseFromParent();
      }

      break;
    }
    case Intrinsic::memset: {
      auto *Memset = cast<MemSetInst>(Inst);
      if (shouldExpandOperationWithSize(Memset->getLength())) {
        expandMemSetAsLoop(Memset);
        Changed = true;
        Memset->eraseFromParent();
      }

      break;
    }
    default:
      break;
    }
  }

  return Changed;
}

bool AMDGPULowerIntrinsics::makeLIDRangeMetadata(Function &F) const {
  auto *TPC = getAnalysisIfAvailable<TargetPassConfig>();
  if (!TPC)
    return false;

  const TargetMachine &TM = TPC->getTM<TargetMachine>();
  bool Changed = false;

  for (auto *U : F.users()) {
    auto *CI = dyn_cast<CallInst>(U);
    if (!CI)
      continue;

    Changed |= AMDGPUSubtarget::get(TM, F).makeLIDRangeMetadata(CI);
  }
  return Changed;
}

bool AMDGPULowerIntrinsics::runOnModule(Module &M) {
  bool Changed = false;

  for (Function &F : M) {
    if (!F.isDeclaration())
      continue;

    switch (F.getIntrinsicID()) {
    case Intrinsic::memcpy:
    case Intrinsic::memmove:
    case Intrinsic::memset:
      if (expandMemIntrinsicUses(F))
        Changed = true;
      break;

    case Intrinsic::amdgcn_workitem_id_x:
    case Intrinsic::r600_read_tidig_x:
    case Intrinsic::amdgcn_workitem_id_y:
    case Intrinsic::r600_read_tidig_y:
    case Intrinsic::amdgcn_workitem_id_z:
    case Intrinsic::r600_read_tidig_z:
    case Intrinsic::r600_read_local_size_x:
    case Intrinsic::r600_read_local_size_y:
    case Intrinsic::r600_read_local_size_z:
      Changed |= makeLIDRangeMetadata(F);
      break;

    default:
      break;
    }
  }

  return Changed;
}

ModulePass *llvm::createAMDGPULowerIntrinsicsPass() {
  return new AMDGPULowerIntrinsics();
}