* Copyright (c) 2024 Huawei Technologies Co., Ltd.
* This program is free software, you can redistribute it and/or modify it under the terms and conditions of
* CANN Open Software License Agreement Version 2.0 (the "License").
* Please refer to the License for details. You may not use this file except in compliance with the License.
* THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, EITHER EXPRESS OR IMPLIED,
* INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
* See LICENSE in the root of the software repository for the full text of the License.
*/
#include "atb/runner/runner.h"
#include <sstream>
#include <mki/utils/file_system/file_system.h>
#include <mki/utils/time/timer.h>
#include "atb/utils/config.h"
#include "atb/utils/store_util.h"
#include "atb/utils/tensor_util.h"
#include "atb/operation/operation_base.h"
#include "atb/operation/graph_operation.h"
#include "atb/types.h"
#include "atb/utils/common_utils.h"
#include "atb/utils/log.h"
#include "atb/utils/probe.h"
#include "atb/utils/statistic.h"
#include "atb/utils/singleton.h"
namespace atb {
static const char *TENSOR_FILE_NAME_EXT = ".bin";
constexpr size_t WORKSPACE_ALIGN = 512;
Runner::Runner(const std::string &name) : name_(name)
{
tensorDir_ = name_;
logPrefix_ = name_;
operationName_ = "";
runnerIds_.clear();
}
Runner::~Runner() {}
std::string Runner::GetName() const
{
return name_;
}
std::string Runner::GetOperationName() const
{
return operationName_;
}
Status Runner::Setup(RunnerVariantPack &runnerVariantPack)
{
multiStreamWorkspaceSizes_.clear();
multiStreamWorkspaceSizes_.resize(runnerVariantPack.context->GetExecuteStreams().size());
Status st = SetupImpl(runnerVariantPack);
setupCount_++;
Probe::UpdateConfig();
return st;
}
uint64_t Runner::GetIntermediateBufferSize()
{
return GetIntermediateBufferSizeImpl();
}
uint64_t Runner::GetTilingBufferSize()
{
return GetTilingBufferSizeImpl();
}
Status Runner::FillHostTilingBuffer(uint8_t *hostTilingBuffer, uint64_t tilingBufferSize, ContextBase *context)
{
return FillHostTilingBufferImpl(hostTilingBuffer, tilingBufferSize, context);
}
std::vector<uint64_t> &Runner::GetWorkspaceBufferSize()
{
multiStreamWorkspaceSizes_.at(GetExecuteStreamId(operation_)) =
static_cast<uint64_t>(TensorUtil::AlignInt(GetWorkspaceBufferSizeImpl(), WORKSPACE_ALIGN));
return multiStreamWorkspaceSizes_;
}
Status Runner::PreExecute(RunnerVariantPack &runnerVariantPack)
{
if (!runnerVariantPack.context) {
ATB_LOG(ERROR) << GetLogPrefix() << " runnerVariantPack.context is null";
return ERROR_INVALID_PARAM;
}
if (GetExecuteStream(runnerVariantPack.context) == nullptr) {
ATB_LOG(ERROR) << GetLogPrefix() << " runnerVariantPack.stream is null";
return ERROR_INVALID_PARAM;
}
ChangeWorkspaceBufferByExecuteStream(runnerVariantPack);
Status st = PreExecuteImpl(runnerVariantPack);
if (st != NO_ERROR) {
ATB_LOG(ERROR) << GetLogPrefix() << "PreExecute Failed. st: " << st;
return st;
}
return NO_ERROR;
}
Status Runner::Execute(RunnerVariantPack &runnerVariantPack)
{
OperationBase *opBase = dynamic_cast<OperationBase *>(operation_);
Mki::OperationIr* operationIr = nullptr;
if (opBase) {
operationIr = opBase->GetOperationIr();
}
if (IsSaveTensor() && Probe::IsSaveTensorBefore()) {
std::string tensorDir = tensorDir_ + "/before";
if (operationIr) {
StoreUtil::SaveVariantPack(GetExecuteStream(runnerVariantPack.context), runnerVariantPack, tensorDir, operationIr);
} else {
StoreUtil::SaveVariantPack(GetExecuteStream(runnerVariantPack.context), runnerVariantPack, tensorDir);
}
ATB_LOG(INFO) << GetLogPrefix() << " save variant pack at " << tensorDir;
}
Status st = ExecuteImpl(runnerVariantPack);
if (IsSaveTensor() && Probe::IsSaveTensorAfter()) {
std::string tensorDir = tensorDir_ + "/after";
if (operationIr) {
StoreUtil::SaveVariantPack(GetExecuteStream(runnerVariantPack.context), runnerVariantPack, tensorDir, operationIr);
} else {
StoreUtil::SaveVariantPack(GetExecuteStream(runnerVariantPack.context), runnerVariantPack, tensorDir);
}
ATB_LOG(INFO) << GetLogPrefix() << " save variant pack at " << tensorDir;
}
if (IsSaveTensor() && Probe::IsSaveParam()) {
if (opBase) {
nlohmann::json opParamJson = opBase->GetParamJson();
if (!opParamJson.empty()) {
std::string opParam = opParamJson.dump();
std::string filePath = tensorDir_ + "/op_param.json";
Probe::SaveParam(opParam, filePath);
ATB_LOG(INFO) << GetLogPrefix() << " save param at " << filePath;
}
}
}
GraphOperation *opGraph = dynamic_cast<GraphOperation *>(operation_);
if (Probe::ReportOperationIOTensorEnable() && !opGraph) {
DumpIOTensorInfo(runnerVariantPack);
}
executeCount_++;
if (st != NO_ERROR) {
ATB_LOG(ERROR) << GetLogPrefix() << "Execute Failed. st: " << st;
return st;
}
if (GetSingleton<Config>().IsStreamSyncEveryRunnerEnable()) {
Mki::Timer streamSyncTimer;
int retCode = aclrtSynchronizeStream(GetExecuteStream(runnerVariantPack.context));
GetOpExecuteStatistic().streamSyncTime += streamSyncTimer.ElapsedMicroSecond();
ATB_LOG_IF(retCode != 0, ERROR) << GetLogPrefix() << "stream sync fail, ret:" << retCode;
}
return NO_ERROR;
}
void Runner::SetRunnerInfo(const std::string &operationName, const std::vector<int64_t> &operationIds)
{
operationName_ = operationName;
runnerIds_ = operationIds;
logPrefix_ = GenerateOperationName(name_, runnerIds_);
saveTensorFlag_ = Probe::IsTensorNeedSave(runnerIds_, operationName_);
}
Status Runner::SetupImpl(RunnerVariantPack &runnerVariantPack)
{
ATB_LOG(INFO) << GetLogPrefix() << "variantPack:" << runnerVariantPack.ToString();
return NO_ERROR;
}
uint64_t Runner::GetIntermediateBufferSizeImpl()
{
return 0;
}
uint64_t Runner::GetTilingBufferSizeImpl()
{
return 0;
}
Status Runner::FillHostTilingBufferImpl(uint8_t *hostTilingBuffer, uint64_t tilingBufferSize, ContextBase *context)
{
(void)context;
ATB_LOG(INFO) << GetLogPrefix() << "hostTilingBuffer:" << hostTilingBuffer << ", tilingBufferSize"
<< tilingBufferSize;
return NO_ERROR;
}
uint64_t Runner::GetWorkspaceBufferSizeImpl()
{
return 0;
}
Status Runner::ExecuteImpl(RunnerVariantPack &runnerVariantPack)
{
ATB_LOG(INFO) << GetLogPrefix() << "variantPack:" << runnerVariantPack.ToString();
return NO_ERROR;
}
Status Runner::PreExecuteImpl(RunnerVariantPack &runnerVariantPack)
{
ATB_LOG(INFO) << GetLogPrefix() << "variantPack:" << runnerVariantPack.ToString();
return NO_ERROR;
}
void Runner::SetSaveTensorDir(const std::string &tensorDir)
{
tensorDir_ = tensorDir;
}
bool Runner::IsSaveTensor() const
{
return Probe::IsExecuteCountInRange(executeCount_) && saveTensorFlag_ && Probe::IsSaveTensorInSpecificDir(tensorDir_);
}
std::string Runner::GetLogPrefix() const
{
std::stringstream ss;
ss << logPrefix_ << ":" << std::to_string(executeCount_) << " ";
return ss.str();
}
std::string Runner::GetSaveTensorDir() const
{
return tensorDir_;
}
void Runner::SetRunnerOperation(Operation *operation)
{
operation_ = operation;
}
void Runner::DumpIOTensorInfo(RunnerVariantPack &runnerVariantPack) const
{
OperationBase *opBase = dynamic_cast<OperationBase *>(operation_);
if (!opBase) {
ATB_LOG(ERROR) << GetLogPrefix() << "operation is not inherit from OperationBase";
return;
}
size_t executeCount = executeCount_;
const std::vector<int64_t> &operationBaseIds = opBase->GetOperationBaseIds();
std::string opName = GenerateOperationName(operationName_, operationBaseIds);
nlohmann::json opParamJson = opBase->GetParamJson();
std::string opParam = opParamJson.dump();
std::string tensorDir = tensorDir_ + "/" + (Probe::IsSaveTensorBefore() ? "before" : "after");
std::vector<Probe::Tensor> inTensorsDesc(runnerVariantPack.inTensors.size());
for (size_t i = 0; i < runnerVariantPack.inTensors.size(); i++) {
std::string fileName = "intensor" + std::to_string(i) + TENSOR_FILE_NAME_EXT;
std::string tensorPath = Mki::FileSystem::Join({tensorDir, fileName});
Probe::AtbTensorToProbeTensor(runnerVariantPack.inTensors[i], inTensorsDesc[i], tensorPath);
}
std::vector<Probe::Tensor> outTensorsDesc(runnerVariantPack.outTensors.size());
for (size_t i = 0; i < runnerVariantPack.outTensors.size(); i++) {
std::string fileName = "outtensor" + std::to_string(i) + TENSOR_FILE_NAME_EXT;
std::string tensorPath = Mki::FileSystem::Join({tensorDir, fileName});
Probe::AtbTensorToProbeTensor(runnerVariantPack.outTensors[i], outTensorsDesc[i], tensorPath);
}
if (opParam == "null") {
Probe::ReportOperationIOTensor(executeCount, opName, "{}", inTensorsDesc, outTensorsDesc);
} else {
Probe::ReportOperationIOTensor(executeCount, opName, opParam, inTensorsDesc, outTensorsDesc);
}
}
void Runner::SetParam(const Mki::Any ¶m)
{
(void)param;
}
bool Runner::IsSupportGlbWorkspace()
{
return false;
}
aclrtStream Runner::GetExecuteStream(Context *context) const
{
OperationBase *opBase = dynamic_cast<OperationBase *>(operation_);
if (opBase) {
return opBase->GetExecuteStream(context);
}
ATB_LOG(ERROR) << "this operation is not inherit from OperationBase!";
return nullptr;
}
void Runner::ChangeWorkspaceBufferByExecuteStream(RunnerVariantPack &runnerVariantPack)
{
uint32_t streamId = GetExecuteStreamId(operation_);
runnerVariantPack.workspaceBufferSize = multiStreamWorkspaceSizes_.at(streamId);
uint64_t preWorkspaceSize = 0;
for (size_t i = 0; i < streamId; ++i) {
preWorkspaceSize += multiStreamWorkspaceSizes_.at(i);
}
runnerVariantPack.workspaceBuffer += preWorkspaceSize;
ATB_LOG(INFO) << GetLogPrefix() << "Set the operationName: " << operationName_ << " workspaceBuffer from "
<< preWorkspaceSize << ", and the workspaceBufferSize: " << runnerVariantPack.workspaceBufferSize
<< " at StreamId: " << streamId;
}
uint64_t Runner::GetArgsSize()
{
return 0;
}
Status Runner::BuildArgs()
{
return NO_ERROR;
}
Status Runner::UpdateTensorAddr(RunnerVariantPack &runnerVariantPack)
{
(void)runnerVariantPack;
return NO_ERROR;
}
Status Runner::UpdateWorkspaceBuffer(RunnerVariantPack &runnerVariantPack)
{
(void)runnerVariantPack;
return NO_ERROR;
}
}
