File Manager

003 File Manager

Current Path: /usr/src/contrib/llvm-project/llvm/lib/Target/AMDGPU

usr / src / contrib / llvm-project / llvm / lib / Target / AMDGPU /

📁 ..
📄 AMDGPU.h(11.46 KB)
📄 AMDGPU.td(36.97 KB)
📄 AMDGPUAliasAnalysis.cpp(5.58 KB)
📄 AMDGPUAliasAnalysis.h(3.32 KB)
📄 AMDGPUAlwaysInlinePass.cpp(4.83 KB)
📄 AMDGPUAnnotateKernelFeatures.cpp(11.94 KB)
📄 AMDGPUAnnotateUniformValues.cpp(6.13 KB)
📄 AMDGPUArgumentUsageInfo.cpp(7.66 KB)
📄 AMDGPUArgumentUsageInfo.h(4.81 KB)
📄 AMDGPUAsmPrinter.cpp(50.42 KB)
📄 AMDGPUAsmPrinter.h(5.13 KB)
📄 AMDGPUAtomicOptimizer.cpp(23.79 KB)
📄 AMDGPUCallLowering.cpp(28.66 KB)
📄 AMDGPUCallLowering.h(2.37 KB)
📄 AMDGPUCallingConv.td(7.33 KB)
📄 AMDGPUCodeGenPrepare.cpp(46.42 KB)
📄 AMDGPUCombine.td(2.79 KB)
📄 AMDGPUExportClustering.cpp(4.52 KB)
📄 AMDGPUExportClustering.h(533 B)
📄 AMDGPUFeatures.td(1.81 KB)
📄 AMDGPUFixFunctionBitcasts.cpp(1.87 KB)
📄 AMDGPUFrameLowering.cpp(1.98 KB)
📄 AMDGPUFrameLowering.h(1.39 KB)
📄 AMDGPUGISel.td(11.57 KB)
📄 AMDGPUGenRegisterBankInfo.def(5.83 KB)
📄 AMDGPUGlobalISelUtils.cpp(1.77 KB)
📄 AMDGPUGlobalISelUtils.h(2.07 KB)
📄 AMDGPUHSAMetadataStreamer.cpp(31.21 KB)
📄 AMDGPUHSAMetadataStreamer.h(5.46 KB)
📄 AMDGPUISelDAGToDAG.cpp(101.59 KB)
📄 AMDGPUISelLowering.cpp(168.65 KB)
📄 AMDGPUISelLowering.h(19.23 KB)
📄 AMDGPUInline.cpp(7.97 KB)
📄 AMDGPUInstrInfo.cpp(1.71 KB)
📄 AMDGPUInstrInfo.h(1.66 KB)
📄 AMDGPUInstrInfo.td(17.18 KB)
📄 AMDGPUInstructionSelector.cpp(128.53 KB)
📄 AMDGPUInstructionSelector.h(11.04 KB)
📄 AMDGPUInstructions.td(25.36 KB)
📄 AMDGPULegalizerInfo.cpp(149.32 KB)
📄 AMDGPULegalizerInfo.h(8.49 KB)
📄 AMDGPULibCalls.cpp(53.89 KB)
📄 AMDGPULibFunc.cpp(37.85 KB)
📄 AMDGPULibFunc.h(10.99 KB)
📄 AMDGPULowerIntrinsics.cpp(4.55 KB)
📄 AMDGPULowerKernelArguments.cpp(8.89 KB)
📄 AMDGPULowerKernelAttributes.cpp(7.78 KB)
📄 AMDGPUMCInstLower.cpp(14.27 KB)
📄 AMDGPUMachineCFGStructurizer.cpp(101.97 KB)
📄 AMDGPUMachineFunction.cpp(2.24 KB)
📄 AMDGPUMachineFunction.h(2.13 KB)
📄 AMDGPUMachineModuleInfo.cpp(1.34 KB)
📄 AMDGPUMachineModuleInfo.h(5.46 KB)
📄 AMDGPUMacroFusion.cpp(2.28 KB)
📄 AMDGPUMacroFusion.h(679 B)
📄 AMDGPUOpenCLEnqueuedBlockLowering.cpp(5.31 KB)
📄 AMDGPUPTNote.h(1.29 KB)
📄 AMDGPUPerfHintAnalysis.cpp(12.17 KB)
📄 AMDGPUPerfHintAnalysis.h(1.67 KB)
📄 AMDGPUPostLegalizerCombiner.cpp(12.02 KB)
📄 AMDGPUPreLegalizerCombiner.cpp(5.45 KB)
📄 AMDGPUPrintfRuntimeBinding.cpp(21.7 KB)
📄 AMDGPUPromoteAlloca.cpp(35.24 KB)
📄 AMDGPUPropagateAttributes.cpp(11.76 KB)
📄 AMDGPURegBankCombiner.cpp(5.36 KB)
📄 AMDGPURegisterBankInfo.cpp(161.67 KB)
📄 AMDGPURegisterBankInfo.h(7.41 KB)
📄 AMDGPURegisterBanks.td(921 B)
📄 AMDGPURewriteOutArguments.cpp(15.82 KB)
📄 AMDGPUSearchableTables.td(21.04 KB)
📄 AMDGPUSubtarget.cpp(29.62 KB)
📄 AMDGPUSubtarget.h(35.82 KB)
📄 AMDGPUTargetMachine.cpp(42.67 KB)
📄 AMDGPUTargetMachine.h(4.52 KB)
📄 AMDGPUTargetObjectFile.cpp(1.54 KB)
📄 AMDGPUTargetObjectFile.h(1.14 KB)
📄 AMDGPUTargetTransformInfo.cpp(39.07 KB)
📄 AMDGPUTargetTransformInfo.h(11.11 KB)
📄 AMDGPUUnifyDivergentExitNodes.cpp(13.84 KB)
📄 AMDGPUUnifyMetadata.cpp(4.46 KB)
📄 AMDILCFGStructurizer.cpp(56.32 KB)
📄 AMDKernelCodeT.h(32.84 KB)
📁 AsmParser
📄 BUFInstructions.td(110.75 KB)
📄 CaymanInstructions.td(7.93 KB)
📄 DSInstructions.td(52.37 KB)
📁 Disassembler
📄 EvergreenInstructions.td(28.24 KB)
📄 FLATInstructions.td(66.93 KB)
📄 GCNDPPCombine.cpp(19.92 KB)
📄 GCNHazardRecognizer.cpp(45.3 KB)
📄 GCNHazardRecognizer.h(3.96 KB)
📄 GCNILPSched.cpp(11.3 KB)
📄 GCNIterativeScheduler.cpp(20.62 KB)
📄 GCNIterativeScheduler.h(4.16 KB)
📄 GCNMinRegStrategy.cpp(8.47 KB)
📄 GCNNSAReassign.cpp(10.92 KB)
📄 GCNProcessors.td(4.84 KB)
📄 GCNRegBankReassign.cpp(26.68 KB)
📄 GCNRegPressure.cpp(16.27 KB)
📄 GCNRegPressure.h(9.15 KB)
📄 GCNSchedStrategy.cpp(21.67 KB)
📄 GCNSchedStrategy.h(3.77 KB)
📁 MCTargetDesc
📄 MIMGInstructions.td(39.85 KB)
📄 R600.td(1.51 KB)
📄 R600AsmPrinter.cpp(4.46 KB)
📄 R600AsmPrinter.h(1.5 KB)
📄 R600ClauseMergePass.cpp(7.38 KB)
📄 R600ControlFlowFinalizer.cpp(23.4 KB)
📄 R600Defines.h(4.25 KB)
📄 R600EmitClauseMarkers.cpp(12.1 KB)
📄 R600ExpandSpecialInstrs.cpp(10.11 KB)
📄 R600FrameLowering.cpp(1.83 KB)
📄 R600FrameLowering.h(1.25 KB)
📄 R600ISelLowering.cpp(81.88 KB)
📄 R600ISelLowering.h(4.8 KB)
📄 R600InstrFormats.td(11.58 KB)
📄 R600InstrInfo.cpp(49.47 KB)
📄 R600InstrInfo.h(13.7 KB)
📄 R600Instructions.td(55.13 KB)
📄 R600MachineFunctionInfo.cpp(551 B)
📄 R600MachineFunctionInfo.h(824 B)
📄 R600MachineScheduler.cpp(13.57 KB)
📄 R600MachineScheduler.h(2.53 KB)
📄 R600OpenCLImageTypeLoweringPass.cpp(11.75 KB)
📄 R600OptimizeVectorRegisters.cpp(13.4 KB)
📄 R600Packetizer.cpp(13.4 KB)
📄 R600Processors.td(4.42 KB)
📄 R600RegisterInfo.cpp(3.95 KB)
📄 R600RegisterInfo.h(2 KB)
📄 R600RegisterInfo.td(9.75 KB)
📄 R600Schedule.td(1.62 KB)
📄 R700Instructions.td(783 B)
📄 SIAddIMGInit.cpp(6.24 KB)
📄 SIAnnotateControlFlow.cpp(11.18 KB)
📄 SIDefines.h(20.86 KB)
📄 SIFixSGPRCopies.cpp(29.46 KB)
📄 SIFixVGPRCopies.cpp(2 KB)
📄 SIFixupVectorISel.cpp(8.75 KB)
📄 SIFoldOperands.cpp(54.56 KB)
📄 SIFormMemoryClauses.cpp(12.76 KB)
📄 SIFrameLowering.cpp(48.08 KB)
📄 SIFrameLowering.h(2.98 KB)
📄 SIISelLowering.cpp(423.43 KB)
📄 SIISelLowering.h(22.13 KB)
📄 SIInsertHardClauses.cpp(7.01 KB)
📄 SIInsertSkips.cpp(15.29 KB)
📄 SIInsertWaitcnts.cpp(58.33 KB)
📄 SIInstrFormats.td(9.44 KB)
📄 SIInstrInfo.cpp(247.15 KB)
📄 SIInstrInfo.h(41.24 KB)
📄 SIInstrInfo.td(90.7 KB)
📄 SIInstructions.td(77.7 KB)
📄 SILoadStoreOptimizer.cpp(76.21 KB)
📄 SILowerControlFlow.cpp(22.66 KB)
📄 SILowerI1Copies.cpp(27.83 KB)
📄 SILowerSGPRSpills.cpp(12.68 KB)
📄 SIMachineFunctionInfo.cpp(20.01 KB)
📄 SIMachineFunctionInfo.h(26.91 KB)
📄 SIMachineScheduler.cpp(69.44 KB)
📄 SIMachineScheduler.h(15.65 KB)
📄 SIMemoryLegalizer.cpp(45.84 KB)
📄 SIModeRegister.cpp(17.43 KB)
📄 SIOptimizeExecMasking.cpp(12.81 KB)
📄 SIOptimizeExecMaskingPreRA.cpp(11.13 KB)
📄 SIPeepholeSDWA.cpp(42.84 KB)
📄 SIPostRABundler.cpp(3.6 KB)
📄 SIPreAllocateWWMRegs.cpp(6.09 KB)
📄 SIPreEmitPeephole.cpp(10.51 KB)
📄 SIProgramInfo.h(2.04 KB)
📄 SIRegisterInfo.cpp(71.51 KB)
📄 SIRegisterInfo.h(13.04 KB)
📄 SIRegisterInfo.td(37.28 KB)
📄 SIRemoveShortExecBranches.cpp(4.96 KB)
📄 SISchedule.td(7.58 KB)
📄 SIShrinkInstructions.cpp(26.86 KB)
📄 SIWholeQuadMode.cpp(30.22 KB)
📄 SMInstructions.td(48.14 KB)
📄 SOPInstructions.td(60.51 KB)
📁 TargetInfo
📁 Utils
📄 VIInstrFormats.td(645 B)
📄 VOP1Instructions.td(35.53 KB)
📄 VOP2Instructions.td(65.04 KB)
📄 VOP3Instructions.td(53.14 KB)
📄 VOP3PInstructions.td(26.47 KB)
📄 VOPCInstructions.td(63.31 KB)
📄 VOPInstructions.td(23.76 KB)

Editing: AMDGPUInline.cpp

//===- AMDGPUInline.cpp - Code to perform simple function inlining --------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
/// \file
/// This is AMDGPU specific replacement of the standard inliner.
/// The main purpose is to account for the fact that calls not only expensive
/// on the AMDGPU, but much more expensive if a private memory pointer is
/// passed to a function as an argument. In this situation, we are unable to
/// eliminate private memory in the caller unless inlined and end up with slow
/// and expensive scratch access. Thus, we boost the inline threshold for such
/// functions here.
///
//===----------------------------------------------------------------------===//

#include "AMDGPU.h"
#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/CallGraph.h"
#include "llvm/Analysis/InlineCost.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Type.h"
#include "llvm/InitializePasses.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Transforms/IPO.h"
#include "llvm/Transforms/IPO/Inliner.h"

using namespace llvm;

#define DEBUG_TYPE "inline"

static cl::opt<int>
ArgAllocaCost("amdgpu-inline-arg-alloca-cost", cl::Hidden, cl::init(4000),
              cl::desc("Cost of alloca argument"));

// If the amount of scratch memory to eliminate exceeds our ability to allocate
// it into registers we gain nothing by aggressively inlining functions for that
// heuristic.
static cl::opt<unsigned>
ArgAllocaCutoff("amdgpu-inline-arg-alloca-cutoff", cl::Hidden, cl::init(256),
                cl::desc("Maximum alloca size to use for inline cost"));

// Inliner constraint to achieve reasonable compilation time
static cl::opt<size_t>
MaxBB("amdgpu-inline-max-bb", cl::Hidden, cl::init(1100),
      cl::desc("Maximum BB number allowed in a function after inlining"
               " (compile time constraint)"));

namespace {

class AMDGPUInliner : public LegacyInlinerBase {

public:
  AMDGPUInliner() : LegacyInlinerBase(ID) {
    initializeAMDGPUInlinerPass(*PassRegistry::getPassRegistry());
    Params = getInlineParams();
  }

static char ID; // Pass identification, replacement for typeid

unsigned getInlineThreshold(CallBase &CB) const;

InlineCost getInlineCost(CallBase &CB) override;

bool runOnSCC(CallGraphSCC &SCC) override;

void getAnalysisUsage(AnalysisUsage &AU) const override;

private:
  TargetTransformInfoWrapperPass *TTIWP;

InlineParams Params;
};

} // end anonymous namespace

char AMDGPUInliner::ID = 0;
INITIALIZE_PASS_BEGIN(AMDGPUInliner, "amdgpu-inline",
                "AMDGPU Function Integration/Inlining", false, false)
INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
INITIALIZE_PASS_DEPENDENCY(CallGraphWrapperPass)
INITIALIZE_PASS_DEPENDENCY(ProfileSummaryInfoWrapperPass)
INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
INITIALIZE_PASS_END(AMDGPUInliner, "amdgpu-inline",
                "AMDGPU Function Integration/Inlining", false, false)

Pass *llvm::createAMDGPUFunctionInliningPass() { return new AMDGPUInliner(); }

bool AMDGPUInliner::runOnSCC(CallGraphSCC &SCC) {
  TTIWP = &getAnalysis<TargetTransformInfoWrapperPass>();
  return LegacyInlinerBase::runOnSCC(SCC);
}

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

unsigned AMDGPUInliner::getInlineThreshold(CallBase &CB) const {
  int Thres = Params.DefaultThreshold;

Function *Caller = CB.getCaller();
  // Listen to the inlinehint attribute when it would increase the threshold
  // and the caller does not need to minimize its size.
  Function *Callee = CB.getCalledFunction();
  bool InlineHint = Callee && !Callee->isDeclaration() &&
    Callee->hasFnAttribute(Attribute::InlineHint);
  if (InlineHint && Params.HintThreshold && Params.HintThreshold > Thres
      && !Caller->hasFnAttribute(Attribute::MinSize))
    Thres = Params.HintThreshold.getValue() *
            TTIWP->getTTI(*Callee).getInliningThresholdMultiplier();

const DataLayout &DL = Caller->getParent()->getDataLayout();
  if (!Callee)
    return (unsigned)Thres;

// If we have a pointer to private array passed into a function
  // it will not be optimized out, leaving scratch usage.
  // Increase the inline threshold to allow inliniting in this case.
  uint64_t AllocaSize = 0;
  SmallPtrSet<const AllocaInst *, 8> AIVisited;
  for (Value *PtrArg : CB.args()) {
    PointerType *Ty = dyn_cast<PointerType>(PtrArg->getType());
    if (!Ty || (Ty->getAddressSpace() != AMDGPUAS::PRIVATE_ADDRESS &&
                Ty->getAddressSpace() != AMDGPUAS::FLAT_ADDRESS))
      continue;

PtrArg = GetUnderlyingObject(PtrArg, DL);
    if (const AllocaInst *AI = dyn_cast<AllocaInst>(PtrArg)) {
      if (!AI->isStaticAlloca() || !AIVisited.insert(AI).second)
        continue;
      AllocaSize += DL.getTypeAllocSize(AI->getAllocatedType());
      // If the amount of stack memory is excessive we will not be able
      // to get rid of the scratch anyway, bail out.
      if (AllocaSize > ArgAllocaCutoff) {
        AllocaSize = 0;
        break;
      }
    }
  }
  if (AllocaSize)
    Thres += ArgAllocaCost;

return (unsigned)Thres;
}

// Check if call is just a wrapper around another call.
// In this case we only have call and ret instructions.
static bool isWrapperOnlyCall(CallBase &CB) {
  Function *Callee = CB.getCalledFunction();
  if (!Callee || Callee->size() != 1)
    return false;
  const BasicBlock &BB = Callee->getEntryBlock();
  if (const Instruction *I = BB.getFirstNonPHI()) {
    if (!isa<CallInst>(I)) {
      return false;
    }
    if (isa<ReturnInst>(*std::next(I->getIterator()))) {
      LLVM_DEBUG(dbgs() << "    Wrapper only call detected: "
                        << Callee->getName() << '\n');
      return true;
    }
  }
  return false;
}

InlineCost AMDGPUInliner::getInlineCost(CallBase &CB) {
  Function *Callee = CB.getCalledFunction();
  Function *Caller = CB.getCaller();

if (!Callee || Callee->isDeclaration())
    return llvm::InlineCost::getNever("undefined callee");

if (CB.isNoInline())
    return llvm::InlineCost::getNever("noinline");

TargetTransformInfo &TTI = TTIWP->getTTI(*Callee);
  if (!TTI.areInlineCompatible(Caller, Callee))
    return llvm::InlineCost::getNever("incompatible");

if (CB.hasFnAttr(Attribute::AlwaysInline)) {
    auto IsViable = isInlineViable(*Callee);
    if (IsViable.isSuccess())
      return llvm::InlineCost::getAlways("alwaysinline viable");
    return llvm::InlineCost::getNever(IsViable.getFailureReason());
  }

if (isWrapperOnlyCall(CB))
    return llvm::InlineCost::getAlways("wrapper-only call");

InlineParams LocalParams = Params;
  LocalParams.DefaultThreshold = (int)getInlineThreshold(CB);
  bool RemarksEnabled = false;
  const auto &BBs = Caller->getBasicBlockList();
  if (!BBs.empty()) {
    auto DI = OptimizationRemark(DEBUG_TYPE, "", DebugLoc(), &BBs.front());
    if (DI.isEnabled())
      RemarksEnabled = true;
  }

OptimizationRemarkEmitter ORE(Caller);
  auto GetAssumptionCache = [this](Function &F) -> AssumptionCache & {
    return ACT->getAssumptionCache(F);
  };

auto IC = llvm::getInlineCost(CB, Callee, LocalParams, TTI,
                                GetAssumptionCache, GetTLI, nullptr, PSI,
                                RemarksEnabled ? &ORE : nullptr);

if (IC && !IC.isAlways() && !Callee->hasFnAttribute(Attribute::InlineHint)) {
    // Single BB does not increase total BB amount, thus subtract 1
    size_t Size = Caller->size() + Callee->size() - 1;
    if (MaxBB && Size > MaxBB)
      return llvm::InlineCost::getNever("max number of bb exceeded");
  }
  return IC;
}

003 File Manager

Editing: AMDGPUInline.cpp

Upload File

Create Folder