m53297601
/
Paddle
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

fix unit test cases

Browse Source
local_add_cudnn_lstm
peizhilin 7 years ago
parent bef475c92b
commit 7c8c9dc9bf
43 changed files with 138 additions and 89 deletions
Show Stats Download Patch File Download Diff File

11
cmake/generic.cmake
Unescape View File

@ -349,10 +349,17 @@ function(cc_test TARGET_NAME)
set(oneValueArgs "") set(oneValueArgs "")
set(multiValueArgs SRCS DEPS ARGS) set(multiValueArgs SRCS DEPS ARGS)
cmake_parse_arguments(cc_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN}) cmake_parse_arguments(cc_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
if(WIN32)
list(APPEND win32_deps shlwapi)
if("${cc_test_DEPS};" MATCHES "python;")
list(REMOVE_ITEM cc_test_DEPS python)
list(APPEND win32_deps ${PYTHON_LIBRARIES})
endif()
endif(WIN32)
add_executable(${TARGET_NAME} ${cc_test_SRCS}) add_executable(${TARGET_NAME} ${cc_test_SRCS})
target_link_libraries(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main lod_tensor memory gtest gflags glog) target_link_libraries(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main lod_tensor memory gtest gflags glog)
if(WIN32) if(WIN32)
target_link_libraries(${TARGET_NAME} shlwapi) target_link_libraries(${TARGET_NAME} ${win32_deps})
endif(WIN32) endif(WIN32)
add_dependencies(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main lod_tensor memory gtest gflags glog) add_dependencies(${TARGET_NAME} ${cc_test_DEPS} paddle_gtest_main lod_tensor memory gtest gflags glog)
add_test(NAME ${TARGET_NAME} add_test(NAME ${TARGET_NAME}
@ -679,7 +686,7 @@ function(py_test TARGET_NAME)
set(multiValueArgs SRCS DEPS ARGS ENVS) set(multiValueArgs SRCS DEPS ARGS ENVS)
cmake_parse_arguments(py_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN}) cmake_parse_arguments(py_test "${options}" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
add_test(NAME ${TARGET_NAME} add_test(NAME ${TARGET_NAME}
COMMAND env FLAGS_init_allocated_mem=true FLAGS_cudnn_deterministic=true COMMAND ${CMAKE_COMMAND} -E env FLAGS_init_allocated_mem=true FLAGS_cudnn_deterministic=true
FLAGS_cpu_deterministic=true FLAGS_cpu_deterministic=true
PYTHONPATH=${PADDLE_BINARY_DIR}/python ${py_test_ENVS} PYTHONPATH=${PADDLE_BINARY_DIR}/python ${py_test_ENVS}
${PYTHON_EXECUTABLE} -u ${py_test_SRCS} ${py_test_ARGS} ${PYTHON_EXECUTABLE} -u ${py_test_SRCS} ${py_test_ARGS}

4
paddle/fluid/framework/details/all_reduce_op_handle.cc
Unescape View File

@ -23,7 +23,7 @@ namespace paddle {
namespace framework { namespace framework {
namespace details { namespace details {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
AllReduceOpHandle::AllReduceOpHandle(ir::Node *node, AllReduceOpHandle::AllReduceOpHandle(ir::Node *node,
const std::vector<Scope *> &local_scopes, const std::vector<Scope *> &local_scopes,
const std::vector<platform::Place> &places, const std::vector<platform::Place> &places,
@ -74,7 +74,7 @@ void AllReduceOpHandle::RunImpl() {
} }
if (platform::is_gpu_place(lod_tensors[0]->place())) { if (platform::is_gpu_place(lod_tensors[0]->place())) {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
PADDLE_ENFORCE(nccl_ctxs_, "nccl_ctxs should not be nullptr."); PADDLE_ENFORCE(nccl_ctxs_, "nccl_ctxs should not be nullptr.");
int dtype = -1; int dtype = -1;
size_t numel = 0; size_t numel = 0;

6
paddle/fluid/framework/details/all_reduce_op_handle.h
Unescape View File

@ -20,7 +20,7 @@
#include "paddle/fluid/framework/details/op_handle_base.h" #include "paddle/fluid/framework/details/op_handle_base.h"
#include "paddle/fluid/framework/lod_tensor.h" #include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/framework/scope.h" #include "paddle/fluid/framework/scope.h"
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
#include "paddle/fluid/platform/nccl_helper.h" #include "paddle/fluid/platform/nccl_helper.h"
#endif #endif
@ -29,7 +29,7 @@ namespace framework {
namespace details { namespace details {
struct AllReduceOpHandle : public OpHandleBase { struct AllReduceOpHandle : public OpHandleBase {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
AllReduceOpHandle(ir::Node *node, const std::vector<Scope *> &local_scopes, AllReduceOpHandle(ir::Node *node, const std::vector<Scope *> &local_scopes,
const std::vector<platform::Place> &places, const std::vector<platform::Place> &places,
const platform::NCCLContextMap *ctxs); const platform::NCCLContextMap *ctxs);
@ -49,7 +49,7 @@ struct AllReduceOpHandle : public OpHandleBase {
private: private:
std::vector<Scope *> local_scopes_; std::vector<Scope *> local_scopes_;
std::vector<platform::Place> places_; std::vector<platform::Place> places_;
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
const platform::NCCLContextMap *nccl_ctxs_; const platform::NCCLContextMap *nccl_ctxs_;
#endif #endif
}; };

2
paddle/fluid/framework/details/broadcast_op_handle.cc
Unescape View File

@ -82,7 +82,7 @@ void BroadcastOpHandle::BroadcastOneVar(
}); });
} }
} else { } else {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
VarHandle *out_handle = nullptr; VarHandle *out_handle = nullptr;
int root_id = boost::get<platform::CUDAPlace>(in_tensor.place()).device; int root_id = boost::get<platform::CUDAPlace>(in_tensor.place()).device;
std::vector<std::function<void()>> broadcast_calls; std::vector<std::function<void()>> broadcast_calls;

6
paddle/fluid/framework/details/broadcast_op_handle.h
Unescape View File

@ -24,7 +24,7 @@
#include "paddle/fluid/framework/selected_rows.h" #include "paddle/fluid/framework/selected_rows.h"
#include "paddle/fluid/platform/device_context.h" #include "paddle/fluid/platform/device_context.h"
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
#include "paddle/fluid/platform/nccl_helper.h" #include "paddle/fluid/platform/nccl_helper.h"
#endif #endif
@ -34,7 +34,7 @@ namespace details {
struct BroadcastOpHandle : public OpHandleBase { struct BroadcastOpHandle : public OpHandleBase {
public: public:
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
BroadcastOpHandle(ir::Node *node, const std::vector<Scope *> &local_scopes, BroadcastOpHandle(ir::Node *node, const std::vector<Scope *> &local_scopes,
const std::vector<platform::Place> &places, const std::vector<platform::Place> &places,
const platform::NCCLContextMap *nccl_ctxs) const platform::NCCLContextMap *nccl_ctxs)
@ -68,7 +68,7 @@ struct BroadcastOpHandle : public OpHandleBase {
std::vector<Scope *> local_scopes_; std::vector<Scope *> local_scopes_;
std::vector<platform::Place> places_; std::vector<platform::Place> places_;
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
const platform::NCCLContextMap *nccl_ctxs_; const platform::NCCLContextMap *nccl_ctxs_;
#endif #endif

12
paddle/fluid/framework/details/broadcast_op_handle_test.h
Unescape View File

@ -42,7 +42,7 @@ struct TestBroadcastOpHandle {
std::vector<std::unique_ptr<ir::Node>> nodes_; std::vector<std::unique_ptr<ir::Node>> nodes_;
std::vector<p::Place> place_list_; std::vector<p::Place> place_list_;
bool use_gpu_; bool use_gpu_;
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
std::unique_ptr<platform::NCCLContextMap> nccl_ctxs_; std::unique_ptr<platform::NCCLContextMap> nccl_ctxs_;
#endif #endif
@ -50,7 +50,7 @@ struct TestBroadcastOpHandle {
for (size_t j = 0; j < ctxs_.size(); ++j) { for (size_t j = 0; j < ctxs_.size(); ++j) {
ctxs_[j]->Wait(); ctxs_[j]->Wait();
} }
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
if (nccl_ctxs_) { if (nccl_ctxs_) {
nccl_ctxs_->WaitAll(); nccl_ctxs_->WaitAll();
} }
@ -60,7 +60,7 @@ struct TestBroadcastOpHandle {
void InitCtxOnGpu(bool use_gpu) { void InitCtxOnGpu(bool use_gpu) {
use_gpu_ = use_gpu; use_gpu_ = use_gpu;
if (use_gpu_) { if (use_gpu_) {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
int count = p::GetCUDADeviceCount(); int count = p::GetCUDADeviceCount();
if (count <= 1) { if (count <= 1) {
LOG(WARNING) << "Cannot test multi-gpu Broadcast, because the CUDA " LOG(WARNING) << "Cannot test multi-gpu Broadcast, because the CUDA "
@ -84,7 +84,7 @@ struct TestBroadcastOpHandle {
place_list_.push_back(p); place_list_.push_back(p);
ctxs_.emplace_back(new p::CPUDeviceContext(p)); ctxs_.emplace_back(new p::CPUDeviceContext(p));
} }
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
nccl_ctxs_.reset(nullptr); nccl_ctxs_.reset(nullptr);
#endif #endif
} }
@ -106,14 +106,14 @@ struct TestBroadcastOpHandle {
nodes_.emplace_back( nodes_.emplace_back(
ir::CreateNodeForTest("node0", ir::Node::Type::kOperation)); ir::CreateNodeForTest("node0", ir::Node::Type::kOperation));
if (use_gpu_) { if (use_gpu_) {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
op_handle_ = new BroadcastOpHandle(nodes_.back().get(), local_scopes_, op_handle_ = new BroadcastOpHandle(nodes_.back().get(), local_scopes_,
place_list_, nccl_ctxs_.get()); place_list_, nccl_ctxs_.get());
#else #else
PADDLE_THROW("CUDA is not support."); PADDLE_THROW("CUDA is not support.");
#endif #endif
} else { } else {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
op_handle_ = new BroadcastOpHandle(nodes_.back().get(), local_scopes_, op_handle_ = new BroadcastOpHandle(nodes_.back().get(), local_scopes_,
place_list_, nccl_ctxs_.get()); place_list_, nccl_ctxs_.get());
#else #else

4
paddle/fluid/framework/details/build_strategy.cc
Unescape View File

@ -96,7 +96,7 @@ std::unique_ptr<ir::Graph> BuildStrategy::Apply(
const std::string &loss_var_name, const std::string &loss_var_name,
const std::unordered_set<std::string> &param_names, const std::unordered_set<std::string> &param_names,
const std::vector<Scope *> &local_scopes, const std::vector<Scope *> &local_scopes,
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
const bool use_cuda, platform::NCCLContextMap *nccl_ctxs) const { const bool use_cuda, platform::NCCLContextMap *nccl_ctxs) const {
#else #else
const bool use_cuda) const { const bool use_cuda) const {
@ -118,7 +118,7 @@ std::unique_ptr<ir::Graph> BuildStrategy::Apply(
pass->Erase("local_scopes"); pass->Erase("local_scopes");
pass->SetNotOwned<const std::vector<Scope *>>("local_scopes", pass->SetNotOwned<const std::vector<Scope *>>("local_scopes",
&local_scopes); &local_scopes);
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
platform::NCCLContextMap *nctx = use_cuda ? nccl_ctxs : nullptr; platform::NCCLContextMap *nctx = use_cuda ? nccl_ctxs : nullptr;
pass->Erase("nccl_ctxs"); pass->Erase("nccl_ctxs");
pass->SetNotOwned<platform::NCCLContextMap>("nccl_ctxs", nctx); pass->SetNotOwned<platform::NCCLContextMap>("nccl_ctxs", nctx);

4
paddle/fluid/framework/details/build_strategy.h
Unescape View File

@ -23,7 +23,7 @@
#include "paddle/fluid/platform/device_context.h" #include "paddle/fluid/platform/device_context.h"
#include "paddle/fluid/platform/enforce.h" #include "paddle/fluid/platform/enforce.h"
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
#include "paddle/fluid/platform/nccl_helper.h" #include "paddle/fluid/platform/nccl_helper.h"
#endif #endif
@ -98,7 +98,7 @@ struct BuildStrategy {
const std::string &loss_var_name, const std::string &loss_var_name,
const std::unordered_set<std::string> &param_names, const std::unordered_set<std::string> &param_names,
const std::vector<Scope *> &local_scopes, const std::vector<Scope *> &local_scopes,
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
const bool use_cuda, platform::NCCLContextMap *nccl_ctxs) const; const bool use_cuda, platform::NCCLContextMap *nccl_ctxs) const;
#else #else
const bool use_cuda) const; const bool use_cuda) const;

2
paddle/fluid/framework/details/data_balance_op_handle.cc
Unescape View File

@ -20,7 +20,7 @@ namespace paddle {
namespace framework { namespace framework {
namespace details { namespace details {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
DataBalanceOpHandle::DataBalanceOpHandle( DataBalanceOpHandle::DataBalanceOpHandle(
ir::Node *node, const std::vector<Scope *> &local_scopes, ir::Node *node, const std::vector<Scope *> &local_scopes,
const std::vector<platform::Place> &places, const std::vector<platform::Place> &places,

4
paddle/fluid/framework/details/data_balance_op_handle.h
Unescape View File

@ -19,7 +19,7 @@
#include "paddle/fluid/framework/details/op_handle_base.h" #include "paddle/fluid/framework/details/op_handle_base.h"
#include "paddle/fluid/framework/lod_tensor.h" #include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/framework/scope.h" #include "paddle/fluid/framework/scope.h"
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
#include "paddle/fluid/platform/nccl_helper.h" #include "paddle/fluid/platform/nccl_helper.h"
#endif #endif
@ -29,7 +29,7 @@ namespace details {
struct DataBalanceOpHandle : public OpHandleBase { struct DataBalanceOpHandle : public OpHandleBase {
public: public:
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
DataBalanceOpHandle(ir::Node *node, const std::vector<Scope *> &local_scopes, DataBalanceOpHandle(ir::Node *node, const std::vector<Scope *> &local_scopes,
const std::vector<platform::Place> &places, const std::vector<platform::Place> &places,
const platform::NCCLContextMap *ctxs); const platform::NCCLContextMap *ctxs);

4
paddle/fluid/framework/details/fused_broadcast_op_handle.h
Unescape View File

@ -25,7 +25,7 @@
#include "paddle/fluid/framework/selected_rows.h" #include "paddle/fluid/framework/selected_rows.h"
#include "paddle/fluid/platform/device_context.h" #include "paddle/fluid/platform/device_context.h"
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
#include "paddle/fluid/platform/nccl_helper.h" #include "paddle/fluid/platform/nccl_helper.h"
#endif #endif
@ -35,7 +35,7 @@ namespace details {
struct FusedBroadcastOpHandle : public BroadcastOpHandle { struct FusedBroadcastOpHandle : public BroadcastOpHandle {
public: public:
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
FusedBroadcastOpHandle(ir::Node *node, FusedBroadcastOpHandle(ir::Node *node,
const std::vector<Scope *> local_scopes, const std::vector<Scope *> local_scopes,
const std::vector<platform::Place> &places, const std::vector<platform::Place> &places,

4
paddle/fluid/framework/details/fused_broadcast_op_handle_test.cc
Unescape View File

@ -44,14 +44,14 @@ struct TestFusedBroadcastOpHandle : TestBroadcastOpHandle {
nodes_.emplace_back( nodes_.emplace_back(
ir::CreateNodeForTest("fused_broadcast", ir::Node::Type::kOperation)); ir::CreateNodeForTest("fused_broadcast", ir::Node::Type::kOperation));
if (use_gpu_) { if (use_gpu_) {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
op_handle_ = new FusedBroadcastOpHandle( op_handle_ = new FusedBroadcastOpHandle(
nodes_.back().get(), local_scopes_, place_list_, nccl_ctxs_.get()); nodes_.back().get(), local_scopes_, place_list_, nccl_ctxs_.get());
#else #else
PADDLE_THROW("CUDA is not supported."); PADDLE_THROW("CUDA is not supported.");
#endif #endif
} else { } else {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
op_handle_ = new FusedBroadcastOpHandle( op_handle_ = new FusedBroadcastOpHandle(
nodes_.back().get(), local_scopes_, place_list_, nccl_ctxs_.get()); nodes_.back().get(), local_scopes_, place_list_, nccl_ctxs_.get());
#else #else

16
paddle/fluid/framework/details/multi_devices_graph_pass.cc
Unescape View File

@ -142,7 +142,7 @@ void MultiDevSSAGraphBuilder::Init() const {
places_ = Get<const std::vector<platform::Place>>(kPlaces); places_ = Get<const std::vector<platform::Place>>(kPlaces);
local_scopes_ = Get<const std::vector<Scope *>>(kLocalScopes); local_scopes_ = Get<const std::vector<Scope *>>(kLocalScopes);
strategy_ = Get<const BuildStrategy>(kStrategy); strategy_ = Get<const BuildStrategy>(kStrategy);
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
nccl_ctxs_ = &Get<platform::NCCLContextMap>("nccl_ctxs"); nccl_ctxs_ = &Get<platform::NCCLContextMap>("nccl_ctxs");
#endif #endif
@ -431,7 +431,7 @@ std::unique_ptr<ir::Graph> MultiDevSSAGraphBuilder::ApplyImpl(
} }
} }
bool use_gpu = false; bool use_gpu = false;
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
use_gpu = nccl_ctxs_ != nullptr; use_gpu = nccl_ctxs_ != nullptr;
#endif #endif
@ -478,7 +478,7 @@ bool MultiDevSSAGraphBuilder::IsSparseGradient(const std::string &og) const {
void MultiDevSSAGraphBuilder::SetCommunicationContext( void MultiDevSSAGraphBuilder::SetCommunicationContext(
OpHandleBase *op_handle, const platform::Place &p) const { OpHandleBase *op_handle, const platform::Place &p) const {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
if (nccl_ctxs_ == nullptr) { if (nccl_ctxs_ == nullptr) {
op_handle->SetDeviceContext(p, op_handle->SetDeviceContext(p,
platform::DeviceContextPool::Instance().Get(p)); platform::DeviceContextPool::Instance().Get(p));
@ -492,7 +492,7 @@ void MultiDevSSAGraphBuilder::SetCommunicationContext(
void MultiDevSSAGraphBuilder::CreateBroadcastOp(ir::Graph *result, void MultiDevSSAGraphBuilder::CreateBroadcastOp(ir::Graph *result,
const std::string &p_name, const std::string &p_name,
size_t src_dev_id) const { size_t src_dev_id) const {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
auto *op_handle = new BroadcastOpHandle( auto *op_handle = new BroadcastOpHandle(
result->CreateEmptyNode("broadcast", ir::Node::Type::kOperation), result->CreateEmptyNode("broadcast", ir::Node::Type::kOperation),
local_scopes_, places_, nccl_ctxs_); local_scopes_, places_, nccl_ctxs_);
@ -522,7 +522,7 @@ void MultiDevSSAGraphBuilder::CreateBroadcastOp(ir::Graph *result,
void MultiDevSSAGraphBuilder::CreateFusedBroadcastOp( void MultiDevSSAGraphBuilder::CreateFusedBroadcastOp(
ir::Graph *result, ir::Graph *result,
const std::vector<std::unordered_set<std::string>> &bcast_varnames) const { const std::vector<std::unordered_set<std::string>> &bcast_varnames) const {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
auto *op_handle = new FusedBroadcastOpHandle( auto *op_handle = new FusedBroadcastOpHandle(
result->CreateEmptyNode("fused_broadcast", ir::Node::Type::kOperation), result->CreateEmptyNode("fused_broadcast", ir::Node::Type::kOperation),
local_scopes_, places_, nccl_ctxs_); local_scopes_, places_, nccl_ctxs_);
@ -568,7 +568,7 @@ void MultiDevSSAGraphBuilder::CreateComputationalOp(ir::Graph *result,
void MultiDevSSAGraphBuilder::InsertAllReduceOp(ir::Graph *result, void MultiDevSSAGraphBuilder::InsertAllReduceOp(ir::Graph *result,
const std::string &og) const { const std::string &og) const {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
result->Get<GraphOps>(kGraphOps).emplace_back(new AllReduceOpHandle( result->Get<GraphOps>(kGraphOps).emplace_back(new AllReduceOpHandle(
result->CreateEmptyNode("allreduce", ir::Node::Type::kOperation), result->CreateEmptyNode("allreduce", ir::Node::Type::kOperation),
local_scopes_, places_, nccl_ctxs_)); local_scopes_, places_, nccl_ctxs_));
@ -597,7 +597,7 @@ void MultiDevSSAGraphBuilder::InsertAllReduceOp(ir::Graph *result,
void MultiDevSSAGraphBuilder::InsertDataBalanceOp( void MultiDevSSAGraphBuilder::InsertDataBalanceOp(
ir::Graph *result, const std::vector<std::string> &datas) const { ir::Graph *result, const std::vector<std::string> &datas) const {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
result->Get<GraphOps>(kGraphOps).emplace_back(new DataBalanceOpHandle( result->Get<GraphOps>(kGraphOps).emplace_back(new DataBalanceOpHandle(
result->CreateEmptyNode("data_balance", ir::Node::Type::kOperation), result->CreateEmptyNode("data_balance", ir::Node::Type::kOperation),
local_scopes_, places_, nccl_ctxs_)); local_scopes_, places_, nccl_ctxs_));
@ -694,7 +694,7 @@ void MultiDevSSAGraphBuilder::CreateComputationalOps(ir::Graph *result,
VarHandle *MultiDevSSAGraphBuilder::CreateReduceOp(ir::Graph *result, VarHandle *MultiDevSSAGraphBuilder::CreateReduceOp(ir::Graph *result,
const std::string &og, const std::string &og,
int dst_dev_id) const { int dst_dev_id) const {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
result->Get<GraphOps>(kGraphOps).emplace_back(new ReduceOpHandle( result->Get<GraphOps>(kGraphOps).emplace_back(new ReduceOpHandle(
result->CreateEmptyNode("reduce", ir::Node::Type::kOperation), result->CreateEmptyNode("reduce", ir::Node::Type::kOperation),
local_scopes_, places_, nccl_ctxs_)); local_scopes_, places_, nccl_ctxs_));

2
paddle/fluid/framework/details/multi_devices_graph_pass.h
Unescape View File

@ -40,7 +40,7 @@ class MultiDevSSAGraphBuilder : public ir::Pass {
size_t device_id) const; size_t device_id) const;
void Init() const; void Init() const;
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
mutable platform::NCCLContextMap *nccl_ctxs_; mutable platform::NCCLContextMap *nccl_ctxs_;
#endif #endif

2
paddle/fluid/framework/details/reduce_op_handle.cc
Unescape View File

@ -125,7 +125,7 @@ void ReduceOpHandle::RunImpl() {
} }
}); });
} else if (paddle::platform::is_gpu_place(lod_tensors[0]->place())) { } else if (paddle::platform::is_gpu_place(lod_tensors[0]->place())) {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
auto pre_in = pre_in_var->Get<framework::LoDTensor>(); auto pre_in = pre_in_var->Get<framework::LoDTensor>();
VariableVisitor::ShareDimsAndLoD(*pre_in_var, out_var); VariableVisitor::ShareDimsAndLoD(*pre_in_var, out_var);
VariableVisitor::GetMutableTensor(out_var).mutable_data( VariableVisitor::GetMutableTensor(out_var).mutable_data(

4
paddle/fluid/framework/details/reduce_op_handle.h
Unescape View File

@ -23,7 +23,7 @@
#include "paddle/fluid/framework/scope.h" #include "paddle/fluid/framework/scope.h"
#include "paddle/fluid/framework/selected_rows.h" #include "paddle/fluid/framework/selected_rows.h"
#include "paddle/fluid/platform/device_context.h" #include "paddle/fluid/platform/device_context.h"
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
#include "paddle/fluid/platform/nccl_helper.h" #include "paddle/fluid/platform/nccl_helper.h"
#endif #endif
@ -35,7 +35,7 @@ struct ReduceOpHandle : public OpHandleBase {
std::vector<Scope *> local_scopes_; std::vector<Scope *> local_scopes_;
std::vector<platform::Place> places_; std::vector<platform::Place> places_;
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
const platform::NCCLContextMap *nccl_ctxs_; const platform::NCCLContextMap *nccl_ctxs_;
ReduceOpHandle(ir::Node *node, const std::vector<Scope *> &local_scopes, ReduceOpHandle(ir::Node *node, const std::vector<Scope *> &local_scopes,
const std::vector<platform::Place> &places, const std::vector<platform::Place> &places,

12
paddle/fluid/framework/details/reduce_op_handle_test.cc
Unescape View File

@ -35,7 +35,7 @@ struct TestReduceOpHandle {
std::vector<p::Place> gpu_list_; std::vector<p::Place> gpu_list_;
std::vector<std::unique_ptr<p::DeviceContext>> ctxs_; std::vector<std::unique_ptr<p::DeviceContext>> ctxs_;
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
std::unique_ptr<platform::NCCLContextMap> nccl_ctxs_; std::unique_ptr<platform::NCCLContextMap> nccl_ctxs_;
#endif #endif
@ -43,7 +43,7 @@ struct TestReduceOpHandle {
for (size_t j = 0; j < ctxs_.size(); ++j) { for (size_t j = 0; j < ctxs_.size(); ++j) {
ctxs_[j]->Wait(); ctxs_[j]->Wait();
} }
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
if (nccl_ctxs_) { if (nccl_ctxs_) {
nccl_ctxs_->WaitAll(); nccl_ctxs_->WaitAll();
} }
@ -53,7 +53,7 @@ struct TestReduceOpHandle {
void InitCtxOnGpu(bool use_gpu) { void InitCtxOnGpu(bool use_gpu) {
use_gpu_ = use_gpu; use_gpu_ = use_gpu;
if (use_gpu) { if (use_gpu) {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
int count = p::GetCUDADeviceCount(); int count = p::GetCUDADeviceCount();
if (count <= 1) { if (count <= 1) {
LOG(WARNING) << "Cannot test multi-gpu Broadcast, because the CUDA " LOG(WARNING) << "Cannot test multi-gpu Broadcast, because the CUDA "
@ -77,7 +77,7 @@ struct TestReduceOpHandle {
gpu_list_.push_back(p); gpu_list_.push_back(p);
ctxs_.emplace_back(new p::CPUDeviceContext(p)); ctxs_.emplace_back(new p::CPUDeviceContext(p));
} }
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
nccl_ctxs_.reset(nullptr); nccl_ctxs_.reset(nullptr);
#endif #endif
} }
@ -99,14 +99,14 @@ struct TestReduceOpHandle {
nodes.emplace_back(new ir::Node("node")); nodes.emplace_back(new ir::Node("node"));
if (use_gpu_) { if (use_gpu_) {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
op_handle_.reset(new ReduceOpHandle(nodes.back().get(), local_scopes_, op_handle_.reset(new ReduceOpHandle(nodes.back().get(), local_scopes_,
gpu_list_, nccl_ctxs_.get())); gpu_list_, nccl_ctxs_.get()));
#else #else
PADDLE_THROW("CUDA is not support."); PADDLE_THROW("CUDA is not support.");
#endif #endif
} else { } else {
#ifdef PADDLE_WITH_CUDA #if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
op_handle_.reset(new ReduceOpHandle(nodes.back().get(), local_scopes_, op_handle_.reset(new ReduceOpHandle(nodes.back().get(), local_scopes_,
gpu_list_, nccl_ctxs_.get())); gpu_list_, nccl_ctxs_.get()));
#else #else

5
paddle/fluid/framework/ir/is_test_pass_tester.cc
Unescape View File

@ -15,7 +15,10 @@
#include "paddle/fluid/framework/ir/is_test_pass.h" #include "paddle/fluid/framework/ir/is_test_pass.h"
#include <gtest/gtest.h> #include <gtest/gtest.h>
#ifdef _WIN32
#undef FALSE
#undef TRUE
#endif
namespace paddle { namespace paddle {
namespace framework { namespace framework {
namespace ir { namespace ir {

3
paddle/fluid/inference/analysis/analyzer_tester.cc
Unescape View File

@ -19,6 +19,7 @@
#include "paddle/fluid/inference/analysis/ut_helper.h" #include "paddle/fluid/inference/analysis/ut_helper.h"
#include "paddle/fluid/inference/api/paddle_inference_api.h" #include "paddle/fluid/inference/api/paddle_inference_api.h"
#include "paddle/fluid/inference/api/paddle_inference_pass.h" #include "paddle/fluid/inference/api/paddle_inference_pass.h"
#include "paddle/fluid/platform/port.h"
namespace paddle { namespace paddle {
namespace inference { namespace inference {
@ -75,7 +76,7 @@ void TestWord2vecPrediction(const std::string& model_path) {
0.000932706}; 0.000932706};
const size_t num_elements = outputs.front().data.length() / sizeof(float); const size_t num_elements = outputs.front().data.length() / sizeof(float);
// The outputs' buffers are in CPU memory. // The outputs' buffers are in CPU memory.
for (size_t i = 0; i < std::min(5UL, num_elements); i++) { for (size_t i = 0; i < std::min((size_t)5UL, num_elements); i++) {
LOG(INFO) << "data: " LOG(INFO) << "data: "
<< static_cast<float*>(outputs.front().data.data())[i]; << static_cast<float*>(outputs.front().data.data())[i];
PADDLE_ENFORCE(static_cast<float*>(outputs.front().data.data())[i], PADDLE_ENFORCE(static_cast<float*>(outputs.front().data.data())[i],

5
paddle/fluid/inference/api/helper.h
Unescape View File

@ -15,10 +15,6 @@
#pragma once #pragma once
#include <glog/logging.h> #include <glog/logging.h>
#if !defined(_WIN32)
#include <sys/time.h>
#else
#endif
#include <algorithm> #include <algorithm>
#include <chrono> // NOLINT #include <chrono> // NOLINT
@ -28,6 +24,7 @@
#include <string> #include <string>
#include <vector> #include <vector>
#include "paddle/fluid/inference/api/paddle_inference_api.h" #include "paddle/fluid/inference/api/paddle_inference_api.h"
#include "paddle/fluid/platform/port.h"
#include "paddle/fluid/string/printf.h" #include "paddle/fluid/string/printf.h"
namespace paddle { namespace paddle {

1
paddle/fluid/inference/tests/api/anakin_rnn1_tester.cc
Unescape View File

@ -13,7 +13,6 @@ See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
#include <gflags/gflags.h> #include <gflags/gflags.h>
#include <sys/time.h>
#include <time.h> #include <time.h>
#include <algorithm> #include <algorithm>
#include <fstream> #include <fstream>

1
paddle/fluid/inference/tests/book/test_inference_nlp.cc
Unescape View File

@ -12,7 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
#include <sys/time.h>
#include <time.h> #include <time.h>
#include <fstream> #include <fstream>
#include <thread> // NOLINT #include <thread> // NOLINT

1
paddle/fluid/inference/tests/test_helper.h
Unescape View File

@ -20,6 +20,7 @@ limitations under the License. */
#include "paddle/fluid/framework/lod_tensor.h" #include "paddle/fluid/framework/lod_tensor.h"
#include "paddle/fluid/inference/io.h" #include "paddle/fluid/inference/io.h"
#include "paddle/fluid/platform/port.h"
#include "paddle/fluid/platform/profiler.h" #include "paddle/fluid/platform/profiler.h"
DECLARE_bool(use_mkldnn); DECLARE_bool(use_mkldnn);

16
paddle/fluid/operators/beam_search_op_test.cc
Unescape View File

@ -30,23 +30,23 @@ using std::endl;
void CreateInput(LoDTensor* ids, LoDTensor* scores) { void CreateInput(LoDTensor* ids, LoDTensor* scores) {
LoD lod; LoD lod;
vector<size_t> level0({0, 2, 4}); vector<size_t> level0{0, 2, 4};
vector<size_t> level1({0, 1, 2, 3, 4}); vector<size_t> level1{0, 1, 2, 3, 4};
lod.push_back(level0); lod.push_back(level0);
lod.push_back(level1); lod.push_back(level1);
ids->set_lod(lod); ids->set_lod(lod);
scores->set_lod(lod); scores->set_lod(lod);
auto dims = framework::make_ddim(vector<int64_t>({4, 3})); auto dims = framework::make_ddim(vector<int64_t>{4, 3});
ids->Resize(dims); ids->Resize(dims);
scores->Resize(dims); scores->Resize(dims);
CPUPlace place; CPUPlace place;
auto* ids_data = ids->mutable_data<int64_t>(place); auto* ids_data = ids->mutable_data<int64_t>(place);
auto* scores_data = scores->mutable_data<float>(place); auto* scores_data = scores->mutable_data<float>(place);
vector<int64_t> _ids({4, 2, 5, 2, 1, 3, 3, 5, 2, 8, 2, 1}); vector<int64_t> _ids{4, 2, 5, 2, 1, 3, 3, 5, 2, 8, 2, 1};
vector<float> _scores( vector<float> _scores{0.5f, 0.3f, 0.2f, 0.6f, 0.3f, 0.1f,
{0.5, 0.3, 0.2, 0.6, 0.3, 0.1, 0.9, 0.5, 0.1, 0.7, 0.5, 0.1}); 0.9f, 0.5f, 0.1f, 0.7f, 0.5f, 0.1f};
for (int i = 0; i < 12; i++) { for (int i = 0; i < 12; i++) {
ids_data[i] = _ids[i]; ids_data[i] = _ids[i];
@ -79,8 +79,8 @@ TEST(DISABLED_beam_search_op, run) {
ASSERT_EQ(sids.lod(), sscores.lod()); ASSERT_EQ(sids.lod(), sscores.lod());
vector<int> tids({4, 2, 3, 8}); vector<int> tids{4, 2, 3, 8};
vector<float> tscores({0.5, 0.6, 0.9, 0.7}); vector<float> tscores{0.5f, 0.6f, 0.9f, 0.7f};
for (int i = 0; i < 4; i++) { for (int i = 0; i < 4; i++) {
ASSERT_EQ(tids[i], sids.data<int64_t>()[i]); ASSERT_EQ(tids[i], sids.data<int64_t>()[i]);

2
paddle/fluid/operators/distributed/grpc_client.cc
Unescape View File

@ -12,7 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and See the License for the specific language governing permissions and
limitations under the License. */ limitations under the License. */
#include <sys/time.h>
#include <limits> #include <limits>
#include "glog/logging.h" // For VLOG #include "glog/logging.h" // For VLOG
@ -20,6 +19,7 @@ limitations under the License. */
#include "paddle/fluid/operators/distributed/grpc_client.h" #include "paddle/fluid/operators/distributed/grpc_client.h"
#include "paddle/fluid/operators/distributed/grpc_serde.h" #include "paddle/fluid/operators/distributed/grpc_serde.h"
#include "paddle/fluid/operators/distributed/request_handler.h" #include "paddle/fluid/operators/distributed/request_handler.h"
#include "paddle/fluid/platform/port.h"
#include "paddle/fluid/platform/profiler.h" #include "paddle/fluid/platform/profiler.h"
namespace paddle { namespace paddle {

Some files were not shown because too many files have changed in this diff Show More

Write Preview
Loading…
Cancel
Save