Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into add-code-exp
commit
7b9d9d7089
@ -0,0 +1,62 @@
|
||||
// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
|
||||
//
|
||||
// Licensed under the Apache License, Version 2.0 (the "License");
|
||||
// you may not use this file except in compliance with the License.
|
||||
// You may obtain a copy of the License at
|
||||
//
|
||||
// http://www.apache.org/licenses/LICENSE-2.0
|
||||
//
|
||||
// Unless required by applicable law or agreed to in writing, software
|
||||
// distributed under the License is distributed on an "AS IS" BASIS,
|
||||
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
#include "paddle/fluid/framework/transfer_scope_cache.h"
|
||||
|
||||
namespace paddle {
|
||||
namespace framework {
|
||||
|
||||
// Holds all the transfer scope across the process.
|
||||
std::unordered_map<size_t, Scope*>& global_transfer_data_cache() {
|
||||
typedef std::unordered_map<size_t, Scope*> map_t;
|
||||
thread_local std::unique_ptr<map_t> x(new map_t);
|
||||
return *x;
|
||||
}
|
||||
|
||||
// Holds all the transfer scope for this thread.
|
||||
std::unordered_set<Scope*>& global_transfer_scope_cache() {
|
||||
typedef std::unordered_set<Scope*> set_t;
|
||||
thread_local std::unique_ptr<set_t> x(new set_t);
|
||||
return *x;
|
||||
}
|
||||
|
||||
// Try to create a transfer scope. If one cached scope has match the
|
||||
// requirement, just return that one.
|
||||
// Inputs:
|
||||
// @type0: the source kernel type.
|
||||
// @type1: the target kernel type.
|
||||
// @scope: the execution scope of this op.
|
||||
// Returns: A scope used to hold the transfer data across the different kernel
|
||||
// type.
|
||||
Scope* TryCreateTransferScope(OpKernelType type0, OpKernelType type1,
|
||||
const Scope* scope) {
|
||||
Scope* new_scope{nullptr};
|
||||
size_t infer_cache_key =
|
||||
CombineHash(OpKernelType::Hash()(type0), OpKernelType::Hash()(type1));
|
||||
infer_cache_key =
|
||||
CombineHash(infer_cache_key, std::hash<const Scope*>()(scope));
|
||||
|
||||
auto it = global_transfer_data_cache().find(infer_cache_key);
|
||||
if (it != global_transfer_data_cache().end()) {
|
||||
new_scope = global_transfer_data_cache()[infer_cache_key];
|
||||
} else {
|
||||
new_scope = &scope->NewScope();
|
||||
global_transfer_data_cache()[infer_cache_key] = new_scope;
|
||||
}
|
||||
global_transfer_scope_cache().insert(new_scope);
|
||||
return new_scope;
|
||||
}
|
||||
|
||||
} // namespace framework
|
||||
} // namespace paddle
|
@ -0,0 +1,41 @@
|
||||
// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
|
||||
//
|
||||
// Licensed under the Apache License, Version 2.0 (the "License");
|
||||
// you may not use this file except in compliance with the License.
|
||||
// You may obtain a copy of the License at
|
||||
//
|
||||
// http://www.apache.org/licenses/LICENSE-2.0
|
||||
//
|
||||
// Unless required by applicable law or agreed to in writing, software
|
||||
// distributed under the License is distributed on an "AS IS" BASIS,
|
||||
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
#pragma once
|
||||
|
||||
#include <thread> // NOLINT
|
||||
#include <unordered_map>
|
||||
#include <unordered_set>
|
||||
#include "paddle/fluid/framework/op_kernel_type.h"
|
||||
#include "paddle/fluid/framework/scope.h"
|
||||
|
||||
namespace paddle {
|
||||
namespace framework {
|
||||
|
||||
std::unordered_map<size_t, Scope*>& global_transfer_data_cache();
|
||||
|
||||
std::unordered_set<Scope*>& global_transfer_scope_cache();
|
||||
|
||||
// Combine two hash values to a single hash.
|
||||
static size_t CombineHash(size_t seed, size_t a) {
|
||||
return (seed ^ a) + 0x9e3779b9 + (seed << 6) + (seed >> 2);
|
||||
}
|
||||
|
||||
Scope* TryCreateTransferScope(OpKernelType type0, OpKernelType type1,
|
||||
const Scope* scope);
|
||||
|
||||
void RemoveKidsFromTransferScopeCache(Scope* scope);
|
||||
|
||||
} // namespace framework
|
||||
} // namespace paddle
|
@ -0,0 +1,2 @@
|
||||
cc_library(benchmark SRCS benchmark.cc DEPS enforce)
|
||||
cc_test(test_benchmark SRCS benchmark_tester.cc DEPS benchmark)
|
@ -0,0 +1,49 @@
|
||||
// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
|
||||
//
|
||||
// Licensed under the Apache License, Version 2.0 (the "License");
|
||||
// you may not use this file except in compliance with the License.
|
||||
// You may obtain a copy of the License at
|
||||
//
|
||||
// http://www.apache.org/licenses/LICENSE-2.0
|
||||
//
|
||||
// Unless required by applicable law or agreed to in writing, software
|
||||
// distributed under the License is distributed on an "AS IS" BASIS,
|
||||
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
#include "paddle/fluid/inference/utils/benchmark.h"
|
||||
#include <sstream>
|
||||
#include "paddle/fluid/platform/enforce.h"
|
||||
|
||||
namespace paddle {
|
||||
namespace inference {
|
||||
|
||||
std::string Benchmark::SerializeToString() const {
|
||||
std::stringstream ss;
|
||||
ss << "-----------------------------------------------------\n";
|
||||
ss << "name\t";
|
||||
ss << "batch_size\t";
|
||||
ss << "num_threads\t";
|
||||
ss << "latency\t";
|
||||
ss << "qps";
|
||||
ss << '\n';
|
||||
|
||||
ss << name_ << "\t";
|
||||
ss << batch_size_ << "\t";
|
||||
ss << num_threads_ << "\t";
|
||||
ss << latency_ << "\t";
|
||||
ss << 1000 / latency_;
|
||||
ss << '\n';
|
||||
return ss.str();
|
||||
}
|
||||
void Benchmark::PersistToFile(const std::string &path) const {
|
||||
std::ofstream file(path, std::ios::app);
|
||||
PADDLE_ENFORCE(file.is_open(), "Can not open %s to add benchmark", path);
|
||||
file << SerializeToString();
|
||||
file.flush();
|
||||
file.close();
|
||||
}
|
||||
|
||||
} // namespace inference
|
||||
} // namespace paddle
|
@ -0,0 +1,52 @@
|
||||
// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
|
||||
//
|
||||
// Licensed under the Apache License, Version 2.0 (the "License");
|
||||
// you may not use this file except in compliance with the License.
|
||||
// You may obtain a copy of the License at
|
||||
//
|
||||
// http://www.apache.org/licenses/LICENSE-2.0
|
||||
//
|
||||
// Unless required by applicable law or agreed to in writing, software
|
||||
// distributed under the License is distributed on an "AS IS" BASIS,
|
||||
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
#include <fstream>
|
||||
#include <iostream>
|
||||
|
||||
namespace paddle {
|
||||
namespace inference {
|
||||
|
||||
/*
|
||||
* Helper class to calculate the performance.
|
||||
*/
|
||||
struct Benchmark {
|
||||
int batch_size() const { return batch_size_; }
|
||||
void SetBatchSize(int x) { batch_size_ = x; }
|
||||
|
||||
int num_threads() const { return num_threads_; }
|
||||
void SetNumThreads(int x) { num_threads_ = x; }
|
||||
|
||||
bool use_gpu() const { return use_gpu_; }
|
||||
void SetUseGpu() { use_gpu_ = true; }
|
||||
|
||||
int latency() const { return latency_; }
|
||||
void SetLatency(int x) { latency_ = x; }
|
||||
|
||||
const std::string& name() const { return name_; }
|
||||
void SetName(const std::string& name) { name_ = name; }
|
||||
|
||||
std::string SerializeToString() const;
|
||||
void PersistToFile(const std::string& path) const;
|
||||
|
||||
private:
|
||||
bool use_gpu_{false};
|
||||
int batch_size_{0};
|
||||
int latency_;
|
||||
int num_threads_{1};
|
||||
std::string name_;
|
||||
};
|
||||
|
||||
} // namespace inference
|
||||
} // namespace paddle
|
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in new issue