tensor-tang
7 years ago
commit
557229bd39
@ -0,0 +1,44 @@
|
||||
if (NOT WITH_AMD_GPU)
|
||||
return()
|
||||
endif()
|
||||
|
||||
# rocprim is "ROCm Parallel Primitives" for short.
|
||||
# It is a header-only library providing HIP and HC parallel primitives
|
||||
# for developing performant GPU-accelerated code on AMD ROCm platform.
|
||||
|
||||
if("x${HCC_HOME}" STREQUAL "x")
|
||||
set(HCC_HOME "/opt/rocm/hcc")
|
||||
endif()
|
||||
|
||||
INCLUDE(ExternalProject)
|
||||
|
||||
SET(ROCPRIM_SOURCE_DIR ${THIRD_PARTY_PATH}/rocprim)
|
||||
SET(ROCPRIM_INSTALL_DIR ${THIRD_PARTY_PATH}/install/rocprim)
|
||||
SET(ROCPRIM_INCLUDE_DIR ${ROCPRIM_INSTALL_DIR}/include)
|
||||
|
||||
ExternalProject_Add(
|
||||
extern_rocprim
|
||||
GIT_REPOSITORY "https://github.com/ROCmSoftwarePlatform/rocPRIM.git"
|
||||
GIT_TAG 5bd41b96ab8d8343330fb2c3e1b96775bde3b3fc
|
||||
PREFIX ${ROCPRIM_SOURCE_DIR}
|
||||
UPDATE_COMMAND ""
|
||||
CMAKE_ARGS -DCMAKE_CXX_COMPILER=${HCC_HOME}/bin/hcc
|
||||
CMAKE_ARGS -DONLY_INSTALL=ON
|
||||
CMAKE_ARGS -DBUILD_TEST=OFF
|
||||
CMAKE_ARGS -DCMAKE_INSTALL_PREFIX=${ROCPRIM_INSTALL_DIR}
|
||||
|
||||
INSTALL_DIR ${ROCPRIM_INSTALL_DIR}
|
||||
${EXTERNAL_PROJECT_LOG_ARGS}
|
||||
)
|
||||
|
||||
INCLUDE_DIRECTORIES(${ROCPRIM_INCLUDE_DIR})
|
||||
|
||||
if (${CMAKE_VERSION} VERSION_LESS "3.3.0")
|
||||
set(dummyfile ${CMAKE_CURRENT_BINARY_DIR}/rocprim_dummy.c)
|
||||
file(WRITE ${dummyfile} "const char *dummy_rocprim = \"${dummyfile}\";")
|
||||
add_library(rocprim STATIC ${dummyfile})
|
||||
else()
|
||||
add_library(rocprim INTERFACE)
|
||||
endif()
|
||||
|
||||
add_dependencies(rocprim extern_rocprim)
|
||||
@ -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
|
||||
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in new issue