提交 d9400243 编写于 作者: Q qiaolongfei

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into public_to_protected

上级 252d4165 973618b6
develop 2.0.1-rocm-post Ligoml-patch-1 OliverLPH-patch-1 OliverLPH-patch-2 PaddlePM-patch-1 PaddlePM-patch-2 ZHUI-patch-1 add_default_att add_model_benchmark_ci add_some_yaml_config addfile all_new_design_exec ascendrc ascendrelease cherry_undefined_var compile_windows cp_2.4_fix_numpy delete_2.0.1-rocm-post delete_add_default_att delete_all_new_design_exec delete_ascendrc delete_compile_windows delete_delete_addfile delete_disable_iterable_dataset_unittest delete_fix_dataloader_memory_leak delete_fix_imperative_dygraph_error delete_fix_retry_ci delete_fix_undefined_var delete_improve_sccache delete_incubate/lite delete_paddle_tiny_install delete_paralleltest delete_prv-disable-more-cache delete_revert-31068-fix_conv3d_windows delete_revert-31562-mean delete_revert-33630-bug-fix delete_revert-34159-add_npu_bce_logical_dev delete_revert-34910-spinlocks_for_allocator delete_revert-35069-revert-34910-spinlocks_for_allocator delete_revert-36057-dev/read_flags_in_ut dingjiaweiww-patch-1 disable_iterable_dataset_unittest dy2static enable_eager_model_test final_state_gen_python_c final_state_intermediate fix-numpy-issue fix_concat_slice fix_dataloader_memory_leak fix_dlpack_for fix_imperative_dygraph_error fix_npu_ci fix_op_flops fix_retry_ci fix_rnn_docs fix_tensor_type fix_undefined_var fix_var_stop_gradient_error fixiscan fixiscan1 fixiscan2 fixiscan3 github/fork/123malin/netifaces github/fork/123malin/tdm_abacus github/fork/AshburnLee/dev_unique github/fork/ForFishes/fix_memory_matmul github/fork/ForFishes/rm_fluid github/fork/LielinJiang/move-2.0-api github/fork/LielinJiang/visual-dl-cb github/fork/LiuChiachi/add-transformer-generate-square-subsequent-mask-api github/fork/LiuChiachi/fix-example-code-for-hapi-Model github/fork/LiuChiachi/remove-input-requirment-in-dygraph-Model github/fork/MrChengmo/fix_ps_profiler github/fork/MrChengmo/update_ps_heter github/fork/PWhiddy/patch-1 github/fork/Shixiaowei02/dev/save_load_upgrade github/fork/TCChenlong/fix_hapi github/fork/TCChenlong/fix_inden github/fork/Thunderbrook/xpu_slice github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_2 github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_3 github/fork/XieYunshen/timeout_20S_ut github/fork/ZeyuChen/remove-nltk github/fork/arlesniak/arlesniak/selective__mkldnn_flags github/fork/baiyfbupt/code_doc_mig github/fork/chalsliu/set_timeout github/fork/chen-zhiyu/develop github/fork/chenwhql/ci/try_to_find_test_buffer_shared_memory_reuse_pass_error github/fork/chenwhql/dygraph/remove_scale_loss_and_apply_collective_grads github/fork/chenwhql/saveload/add_get_inference_program github/fork/chenwhql/saveload/remove_save_load_config github/fork/cryoco/pass-compatibility-trt github/fork/danleifeng/isempty_api2.0 github/fork/frankwhzhang/api_transfer github/fork/hbwx24/error_msg/cuda_kernel_error_msg github/fork/heavengate/cherry_yolo_box github/fork/heavengate/update_yolo_box github/fork/iclementine/rnn_fix github/fork/iducn/testestse github/fork/jczaja/prv-25537-fix github/fork/jeff41404/release/1.8 github/fork/jiweibo/api_2.0 github/fork/jiweibo/fix_lite_resnet50_test github/fork/juncaipeng/fix_doc_1 github/fork/lfchener/sample_code github/fork/littletomatodonkey/fix_reg_doc github/fork/liym27/dy2stat_update_assign_to_rc20 github/fork/luotao1/profiler_ut github/fork/mapingshuo/add_wait github/fork/mapingshuo/doc_2.0 github/fork/mapingshuo/zero-0.5 github/fork/miraiwk/dev github/fork/pangyoki/add-Categorical-class-branch github/fork/pangyoki/add-multinomial-op-branch github/fork/pangyoki/fix-test_distritbution-CI github/fork/qjing666/doublegrad github/fork/qjing666/fix_hdfs_download github/fork/sandyhouse/add_gather_etc github/fork/sandyhouse/add_send_recv_alltoall_etc github/fork/sandyhouse/pipeline_exe_run github/fork/seiriosPlus/feature/large_scale_kv_save_delta github/fork/seiriosPlus/fix/paddle_errors_fix github/fork/seiriosPlus/fix/paddle_op_errors github/fork/shangzhizhou/fix_test_activation_op_random_bug github/fork/smallv0221/yxp0924 github/fork/smallv0221/yxp0925 github/fork/swtkiwi/del-matplotlib github/fork/tianshuo78520a/kunlun_test github/fork/tianshuo78520a/update_dockerfile github/fork/wanghaoshuang/bert_fuse github/fork/wanghaoshuang/label_smooth github/fork/wanghuancoder/develop_CUDASynchronize github/fork/wanghuancoder/develop_Layer_doc github/fork/wanghuancoder/develop_ParameterList_doc github/fork/wanghuancoder/develop_Sequential_doc github/fork/wanghuancoder/develop_bilinear_tensor_product github/fork/wanghuancoder/develop_coverage_build_sh github/fork/wanghuancoder/develop_in_dynamic_mode_doc github/fork/wanghuancoder/develop_unique_name_doc github/fork/wangxicoding/fleet_meta_combine github/fork/wawltor/error_message_fix_5 github/fork/willthefrog/remove_l2_norm github/fork/windstamp/momentum_op github/fork/windstamp/mv_op_5 github/fork/windstamp/normal_api github/fork/wojtuss/wojtuss/fusion_gru_quantization github/fork/wojtuss/wojtuss/quantization-with-shift github/fork/wzzju/fix_err_info github/fork/wzzju/pure_fp16 github/fork/xiemoyuan/op_error_message github/fork/xiemoyuan/optimize_error_message github/fork/yaoxuefeng6/fix_doc github/fork/yaoxuefeng6/mod_dataset_v2 github/fork/yongqiangma/lod github/fork/ysh329/fix-clip-by-norm-error github/fork/ysh329/fix-error-clip-by-value github/fork/yukavio/error_info github/fork/zhangting2020/conv_filter_grad github/fork/zhangting2020/is_compile_with_cuda github/fork/zhangting2020/place_doc github/fork/zhangting2020/program github/fork/zhhsplendid/fix_any github/fork/zhhsplendid/refine_api2 github/fork/zhhsplendid/refine_api2_test github/fork/zhhsplendid/refine_api_test_ptb_lm github/fork/zhhsplendid/refine_api_test_resnet github/fork/zhhsplendid/refine_api_test_simnet github/fork/zhiqiu/dev/refine_initializer github/fork/zhiqiu/dev/remove_inplace_argument github/fork/zlsh80826/nvinfer_plugin_var_len_cuda11 improve_sccache incubate/frl_train_eval incubate/infrt incubate/lite inplace_addto layer_norm make_flag_adding_easier master matmul_double_grad move_embedding_to_phi move_histogram_to_pten move_sgd_to_phi move_slice_to_pten move_temporal_shift_to_phi move_yolo_box_to_phi npu_fix_alloc numel paddle_tiny_install paralleltest preln_ernie prv-disable-more-cache prv-md-even-more prv-onednn-2.5 prv-reshape-mkldnn-ut2 pten_tensor_refactor release/0.11.0 release/0.12.0 release/0.13.0 release/0.14.0 release/0.15.0 release/1.0.0 release/1.1 release/1.2 release/1.3 release/1.4 release/1.5 release/1.6 release/1.7 release/1.8 release/2.0 release/2.0-alpha release/2.0-beta release/2.0-rc release/2.0-rc1 release/2.1 release/2.2 release/2.3 release/2.3-fc-ernie-fix release/2.4 release/lite-0.1 revert-24981-add_device_attr_for_regulization revert-26856-strategy_example2 revert-27520-disable_pr revert-31068-fix_conv3d_windows revert-31562-mean revert-32290-develop-hardlabel revert-33037-forci revert-33475-fix_cifar_label_dimension revert-33630-bug-fix revert-34159-add_npu_bce_logical_dev revert-34406-add_copy_from_tensor revert-34910-spinlocks_for_allocator revert-35069-revert-34910-spinlocks_for_allocator revert-36057-dev/read_flags_in_ut revert-36201-refine_fast_threaded_ssa_graph_executor revert-36985-add_license revert-37318-refactor_dygraph_to_eager revert-37926-eager_coreops_500 revert-37956-revert-37727-pylayer_support_tuple revert-38100-mingdong revert-38301-allocation_rearrange_pr revert-38703-numpy_bf16_package_reupload revert-38732-remove_useless_header_in_elementwise_mul_grad revert-38959-Reduce_Grad revert-39143-adjust_empty revert-39227-move_trace_op_to_pten revert-39268-dev/remove_concat_fluid_kernel revert-40170-support_partial_grad revert-41056-revert-40727-move_some_activaion_to_phi revert-41065-revert-40993-mv_ele_floordiv_pow revert-41068-revert-40790-phi_new revert-41944-smaller_inference_api_test revert-42149-do-not-reset-default-stream-for-stream-safe-cuda-allocator revert-43155-fix_ut_tempfile revert-43882-revert-41944-smaller_inference_api_test revert-45808-phi/simplify_size_op revert-46827-deform_comment revert-47325-remove_cudnn_hardcode revert-47645-add_npu_storage_dims revert-48815-set_free_when_no_cache_hit_default_value_true revert-49654-prim_api_gen revert-49763-fix_static_composite_gen rocm_dev_0217 support-0D-sort support_weight_transpose test_benchmark_ci test_feature_precision_test_c test_for_Filtetfiles test_model_benchmark test_model_benchmark_ci zhiqiu-patch-1 v2.4.1 v2.4.0 v2.4.0-rc0 v2.3.2 v2.3.1 v2.3.0 v2.3.0-rc0 v2.2.2 v2.2.1 v2.2.0 v2.2.0-rc0 v2.2.0-bak0 v2.1.3 v2.1.2 v2.1.1 v2.1.0 v2.1.0-rc0 v2.0.2 v2.0.1 v2.0.0 v2.0.0-rc1 v2.0.0-rc0 v2.0.0-beta0 v2.0.0-alpha0 v1.8.5 v1.8.4 v1.8.3 v1.8.2 v1.8.1 v1.8.0 v1.7.2 v1.7.1 v1.7.0 v1.6.3 v1.6.2 v1.6.1 v1.6.0 v1.6.0-rc0 v1.5.2 v1.5.1 v1.5.0 v1.4.1 v1.4.0 v1.3.2 v1.3.1 v1.3.0 v1.2.1 v1.2.0 v1.1.0 v1.0.2 v1.0.1 v1.0.0 v1.0.0-rc0 v0.15.0 v0.15.0-rc0 v0.14.0 v0.13.0 v0.12.0 v0.11.1a2 v0.11.1a1 v0.11.0 lite-v0.1
无相关合并请求
......@@ -34,9 +34,6 @@ RUN apt-get update && \
net-tools && \
apt-get clean -y
# paddle is using numpy.flip, which is introduced since 1.12.0
RUN pip --no-cache-dir install 'numpy>=1.12.0'
# Install Go and glide
RUN wget -qO- https://storage.googleapis.com/golang/go1.8.1.linux-amd64.tar.gz | \
tar -xz -C /usr/local && \
......@@ -58,13 +55,16 @@ RUN localedef -i en_US -f UTF-8 en_US.UTF-8
# FIXME: due to temporary ipykernel dependency issue, specify ipykernel jupyter
# version util jupyter fixes this issue.
RUN pip install --upgrade pip && \
pip install -U 'protobuf==3.1.0' && \
pip install -U wheel pillow BeautifulSoup && \
pip install -U wheel && \
pip install -U docopt PyYAML sphinx && \
pip install -U sphinx-rtd-theme==0.1.9 recommonmark && \
pip install pre-commit 'requests==2.9.2' 'ipython==5.3.0' && \
pip install -U sphinx-rtd-theme==0.1.9 recommonmark
RUN pip install pre-commit 'ipython==5.3.0' && \
pip install 'ipykernel==4.6.0' 'jupyter==1.0.0' && \
pip install opencv-python rarfile 'scipy>=0.19.0' 'nltk>=3.2.2'
pip install opencv-python
COPY ./python/requirements.txt /root/
RUN pip install -r /root/requirements.txt
# To fix https://github.com/PaddlePaddle/Paddle/issues/1954, we use
# the solution in https://urllib3.readthedocs.io/en/latest/user-guide.html#ssl-py2
......
......@@ -28,13 +28,6 @@ using OpAttrChecker = framework::OpAttrChecker;
using Scope = framework::Scope;
using DeviceContext = platform::DeviceContext;
class EmptyOp : public OperatorBase {
public:
using OperatorBase::OperatorBase;
void InferShape(const Scope &scope) const override {}
void Run(const Scope &scope, const DeviceContext &dev_ctx) const override {}
};
class RowWiseAddOpMaker : public OpProtoAndCheckerMaker {
public:
RowWiseAddOpMaker(OpProto *proto, OpAttrChecker *op_checker)
......@@ -155,19 +148,16 @@ class AddOpMaker : public OpProtoAndCheckerMaker {
namespace f = paddle::framework;
namespace ops = paddle::operators;
using EnforceNotMet = paddle::platform::EnforceNotMet;
REGISTER_OP(rowwise_add, f::EmptyOp, f::RowWiseAddOpMaker);
REGISTER_GRADIENT_OP(rowwise_add, rowwise_add_grad, f::EmptyOp);
REGISTER_OP(mul, f::EmptyOp, f::MulOpMaker);
REGISTER_GRADIENT_OP(mul, mul_grad, f::EmptyOp);
REGISTER_OP(sigmoid, f::EmptyOp, f::SigmoidOpMaker);
REGISTER_GRADIENT_OP(sigmoid, sigmoid_grad, f::EmptyOp);
REGISTER_OP(nograd, f::EmptyOp, f::NoGradOpMaker);
REGISTER_OP(fill_zeros_like, f::EmptyOp, f::FillZeroOpMaker);
REGISTER_OP(add, f::EmptyOp, f::AddOpMaker);
REGISTER_GRADIENT_OP(add, add_grad, f::EmptyOp);
REGISTER_OP(fc, f::FcOp, f::FcOpMaker);
REGISTER_OP(many_output_op, f::EmptyOp, f::ManyOutputOpMaker);
REGISTER_GRADIENT_OP(many_output_op, many_output_op_grad, f::EmptyOp);
REGISTER_OP(rowwise_add, f::NOP, f::RowWiseAddOpMaker, rowwise_add_grad,
f::NOP);
REGISTER_OP(mul, f::NOP, f::MulOpMaker, mul_grad, f::NOP);
REGISTER_OP(sigmoid, f::NOP, f::SigmoidOpMaker, sigmoid_grad, f::NOP);
REGISTER_OP_WITHOUT_GRADIENT(nograd, f::NOP, f::NoGradOpMaker);
REGISTER_OP_WITHOUT_GRADIENT(fill_zeros_like, f::NOP, f::FillZeroOpMaker);
REGISTER_OP(add, f::NOP, f::AddOpMaker, add_grad, f::NOP);
REGISTER_OP_WITHOUT_GRADIENT(fc, f::FcOp, f::FcOpMaker);
REGISTER_OP(many_output_op, f::NOP, f::ManyOutputOpMaker, many_output_op_grad,
f::NOP);
TEST(Backward, simple_op_grad) {
auto fwd = f::OpRegistry::CreateOp(
......
......@@ -19,16 +19,14 @@ namespace paddle {
namespace framework {
enum class OpArgType { IN, OUT };
static void TransOpArg(const OperatorBase* src_op,
OperatorBase::VarNameMap* vars,
const OpArgType& src_type, bool is_grad) {
static void TransOpArg(const OperatorBase* src_op, const OpArgType& src_type,
bool is_grad, OperatorBase::VarNameMap* vars) {
const auto& src_inout =
src_type == OpArgType::IN ? src_op->Inputs() : src_op->Outputs();
auto& dst_inout = *vars;
const OpProto& proto = OpProtos().at(src_op->Type());
const OpProto* proto = OpRegistry::op_info_map().at(src_op->Type()).proto_;
const auto& src_arg_list =
src_type == OpArgType::IN ? proto.inputs() : proto.outputs();
src_type == OpArgType::IN ? proto->inputs() : proto->outputs();
for (const auto& arg : src_arg_list) {
if (arg.no_gradient() && !is_grad) continue;
const std::string src_name = arg.name();
......@@ -42,22 +40,26 @@ static void TransOpArg(const OperatorBase* src_op,
}
OperatorBase* BuildGradOp(const OperatorBase* op) {
auto gop_type_it = OpRegistry::grad_ops().find(op->Type());
PADDLE_ENFORCE(gop_type_it != OpRegistry::grad_ops().end(),
"Operator %s do not register gradient type", op->Type());
auto& grad_op_type = gop_type_it->second;
auto it = OpRegistry::op_info_map().find(op->Type());
PADDLE_ENFORCE(it != OpRegistry::op_info_map().end(),
"'%s' has not been registered.", op->Type());
PADDLE_ENFORCE(it->second.proto_ != nullptr, "'%s' has no OpProto.",
op->Type());
std::string grad_op_type = it->second.grad_op_type_;
PADDLE_ENFORCE(!grad_op_type.empty(), "'%s' has no gradient operator.",
op->Type());
OperatorBase::VarNameMap inputs;
OperatorBase::VarNameMap outputs;
TransOpArg(op, &inputs, OpArgType::IN, false); // I
TransOpArg(op, &inputs, OpArgType::OUT, false); // O
TransOpArg(op, &inputs, OpArgType::OUT, true); // OG
TransOpArg(op, &outputs, OpArgType::IN, true); // IG
auto gop_it = OpRegistry::op_creators().find(grad_op_type);
PADDLE_ENFORCE(gop_it != OpRegistry::op_creators().end(),
"Operator %s 's Gradient %s's creator cannot be found",
op->Type(), grad_op_type);
TransOpArg(op, OpArgType::IN, false, &inputs); // I
TransOpArg(op, OpArgType::OUT, false, &inputs); // O
TransOpArg(op, OpArgType::OUT, true, &inputs); // OG
TransOpArg(op, OpArgType::IN, true, &outputs); // IG
return gop_it->second(grad_op_type, inputs, outputs, op->Attrs());
it = OpRegistry::op_info_map().find(grad_op_type);
PADDLE_ENFORCE(it != OpRegistry::op_info_map().end(),
"'%s' has not been registered.", grad_op_type);
return it->second.creator_(grad_op_type, inputs, outputs, op->Attrs());
}
} // namespace framework
......
......@@ -8,14 +8,6 @@ USE_OP(add_two);
namespace paddle {
namespace framework {
class NOP : public OperatorBase {
public:
using OperatorBase::OperatorBase;
void InferShape(const Scope &scope) const override {}
void Run(const Scope &scope,
const platform::DeviceContext &dev_ctx) const override {}
};
class MutiInOutOpMaker : public OpProtoAndCheckerMaker {
public:
MutiInOutOpMaker(OpProto *proto, OpAttrChecker *op_checker)
......@@ -62,10 +54,8 @@ TEST(GradOpBuilder, AddTwo) {
EXPECT_EQ(grad_add_op->Output(f::GradVarName("Y")), f::GradVarName("y"));
}
REGISTER_OP(mult_io, f::NOP, f::MutiInOutOpMaker);
REGISTER_GRADIENT_OP(mult_io, mult_io_grad, f::NOP);
REGISTER_OP(io_ignored, f::NOP, f::IOIgnoredOpMaker);
REGISTER_GRADIENT_OP(io_ignored, io_ignored_grad, f::NOP);
REGISTER_OP(mult_io, f::NOP, f::MutiInOutOpMaker, mult_io_grad, f::NOP);
REGISTER_OP(io_ignored, f::NOP, f::IOIgnoredOpMaker, io_ignored_grad, f::NOP);
TEST(GradOpBuilder, MutiInOut) {
std::shared_ptr<f::OperatorBase> test_op(f::OpRegistry::CreateOp(
......
......@@ -17,6 +17,7 @@ limitations under the License. */
#include <algorithm>
#include <atomic>
#include <type_traits>
#include <typeinfo>
#include <unordered_map>
#include <unordered_set>
#include "paddle/framework/attribute.h"
......@@ -119,6 +120,12 @@ class OpProtoAndCheckerMaker {
bool validated_{false};
};
class NOPMaker : public OpProtoAndCheckerMaker {
public:
NOPMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
: OpProtoAndCheckerMaker(proto, op_checker) {}
};
class OpRegistry {
using VarNameMap = OperatorBase::VarNameMap;
using OpCreator = std::function<OperatorBase*(
......@@ -126,45 +133,56 @@ class OpRegistry {
const VarNameMap& /*outputs*/, const AttributeMap& /*attrs*/)>;
public:
template <typename OpType, typename ProtoMakerType>
static void RegisterOp(const std::string& op_type) {
op_creators()[op_type] = [](
const std::string& type, const VarNameMap& inputs,
const VarNameMap& outputs, const AttributeMap& attrs) {
return new OpType(type, inputs, outputs, attrs);
};
OpAttrChecker& op_checker = op_checkers()[op_type];
OpProto& op_proto = OpProtos()[op_type];
auto maker = ProtoMakerType(&op_proto, &op_checker);
maker.Validate();
op_proto.set_type(op_type);
PADDLE_ENFORCE(
op_proto.IsInitialized(),
"Fail to initialize %s's OpProto, because %s is not initialized",
op_type, op_proto.InitializationErrorString());
}
struct OpInfo {
OpCreator creator_;
std::string grad_op_type_;
OpProto* proto_;
OpAttrChecker* checker_;
};
template <typename GradOpType>
static void RegisterGradOp(const std::string& op_type,
const std::string& grad_op_type) {
op_creators()[grad_op_type] = [](
const std::string& type, const VarNameMap& inputs,
const VarNameMap& outputs, const AttributeMap& attrs) {
return new GradOpType(type, inputs, outputs, attrs);
template <typename OpType, typename ProtoMakerType, typename GradOpType>
static void RegisterOp(const std::string& op_type,
const std::string& grad_op_type) {
PADDLE_ENFORCE(op_info_map().count(op_type) == 0,
"'%s' is registered more than once.", op_type);
OpInfo op_info;
op_info.creator_ = [](const std::string& type, const VarNameMap& inputs,
const VarNameMap& outputs,
const AttributeMap& attrs) {
return new OpType(type, inputs, outputs, attrs);
};
grad_ops()[op_type] = grad_op_type;
op_info.grad_op_type_ = grad_op_type;
if (std::type_index(typeid(ProtoMakerType)) !=
std::type_index(typeid(NOPMaker))) {
op_info.proto_ = new OpProto;
op_info.checker_ = new OpAttrChecker;
auto maker = ProtoMakerType(op_info.proto_, op_info.checker_);
maker.Validate();
op_info.proto_->set_type(op_type);
PADDLE_ENFORCE(
op_info.proto_->IsInitialized(),
"Fail to initialize %s's OpProto, because %s is not initialized",
op_type, op_info.proto_->InitializationErrorString());
} else {
op_info.proto_ = nullptr;
op_info.checker_ = nullptr;
}
op_info_map().insert(std::make_pair(op_type, op_info));
// register gradient op
if (!grad_op_type.empty()) {
RegisterOp<GradOpType, NOPMaker, NOP>(grad_op_type, "");
}
}
static std::shared_ptr<OperatorBase> CreateOp(const std::string& type,
const VarNameMap& inputs,
const VarNameMap& outputs,
AttributeMap attrs) {
auto op_create_it = op_creators().find(type);
PADDLE_ENFORCE(op_create_it != op_creators().end(),
"Operator %s cannot be found.", type);
op_checkers().at(type).Check(attrs);
auto op = op_create_it->second(type, inputs, outputs, attrs);
auto it = op_info_map().find(type);
PADDLE_ENFORCE(it != op_info_map().end(),
"Operator '%s' has not been registered.", type);
it->second.checker_->Check(attrs);
auto op = it->second.creator_(type, inputs, outputs, attrs);
return std::shared_ptr<OperatorBase>(op);
}
......@@ -199,49 +217,32 @@ class OpRegistry {
return grad_op;
}
static std::unordered_map<std::string, std::string>& grad_ops() {
static std::unordered_map<std::string, std::string> grad_ops_;
return grad_ops_;
}
static std::unordered_map<std::string, OpCreator>& op_creators() {
static std::unordered_map<std::string, OpCreator> op_creators_;
return op_creators_;
}
private:
static std::unordered_map<std::string, OpAttrChecker>& op_checkers() {
static std::unordered_map<std::string, OpAttrChecker> op_checkers_;
return op_checkers_;
static std::unordered_map<std::string, const OpInfo>& op_info_map() {
static std::unordered_map<std::string, const OpInfo> op_info_map_;
return op_info_map_;
}
};
class Registrar {
public:
// In our design, various kinds of classes, e.g., operators and kernels, have
// their corresponding registry and registrar. The action of registration is
// in the constructor of a global registrar variable, which, however, are not
// used in the code that calls package framework, and would be removed from
// the generated binary file by the linker. To avoid such removal, we add
// Touch to all registrar classes and make USE_OP macros to call this
// method. So, as long as the callee code calls USE_OP, the global
// In our design, various kinds of classes, e.g., operators and kernels,
// have their corresponding registry and registrar. The action of
// registration is in the constructor of a global registrar variable, which,
// however, are not used in the code that calls package framework, and would
// be removed from the generated binary file by the linker. To avoid such
// removal, we add Touch to all registrar classes and make USE_OP macros to
// call this method. So, as long as the callee code calls USE_OP, the global
// registrar variable won't be removed by the linker.
void Touch() {}
};
template <typename OpType, typename ProtoMakerType>
template <typename OpType, typename ProtoMakerType, typename GradOpType>
class OpRegistrar : public Registrar {
public:
explicit OpRegistrar(const char* op_type) {
OpRegistry::RegisterOp<OpType, ProtoMakerType>(op_type);
}
};
template <typename GradOpType>
class GradOpRegistrar : public Registrar {
public:
GradOpRegistrar(const char* op_type, const char* grad_op_type) {
OpRegistry::RegisterGradOp<GradOpType>(op_type, grad_op_type);
explicit OpRegistrar(const char* op_type) { OpRegistrar(op_type, ""); }
OpRegistrar(const char* op_type, const char* grad_op_type) {
OpRegistry::RegisterOp<OpType, ProtoMakerType, GradOpType>(op_type,
grad_op_type);
}
};
......@@ -267,30 +268,20 @@ class OpKernelRegistrar : public Registrar {
/**
* Macro to register Operator.
*/
#define REGISTER_OP(op_type, op_class, op_maker_class) \
#define REGISTER_OP(op_type, op_class, op_maker_class, grad_op_type, \
grad_op_class) \
STATIC_ASSERT_GLOBAL_NAMESPACE( \
__reg_op__##op_type, "REGISTER_OP must be called in global namespace"); \
static ::paddle::framework::OpRegistrar<op_class, op_maker_class> \
__op_registrar_##op_type##__(#op_type); \
static ::paddle::framework::OpRegistrar<op_class, op_maker_class, \
grad_op_class> \
__op_registrar_##op_type##__(#op_type, #grad_op_type); \
int TouchOpRegistrar_##op_type() { \
__op_registrar_##op_type##__.Touch(); \
return 0; \
}
/**
* Macro to register Gradient Operator.
*/
#define REGISTER_GRADIENT_OP(op_type, grad_op_type, grad_op_class) \
STATIC_ASSERT_GLOBAL_NAMESPACE( \
__reg_gradient_op__##op_type##_##grad_op_type, \
"REGISTER_GRADIENT_OP must be called in global namespace"); \
static ::paddle::framework::GradOpRegistrar<grad_op_class> \
__op_gradient_registrar_##op_type##_##grad_op_type##__(#op_type, \
#grad_op_type); \
int TouchOpGradientRegistrar_##op_type() { \
__op_gradient_registrar_##op_type##_##grad_op_type##__.Touch(); \
return 0; \
}
#define REGISTER_OP_WITHOUT_GRADIENT(op_type, op_class, op_maker_class) \
REGISTER_OP(op_type, op_class, op_maker_class, , ::paddle::framework::NOP)
/**
* Macro to register OperatorKernel.
......@@ -306,14 +297,6 @@ class OpKernelRegistrar : public Registrar {
return 0; \
}
/**
* Macro to Forbid user register Gradient Operator.
*/
#define NO_GRADIENT(op_type) \
STATIC_ASSERT_GLOBAL_NAMESPACE( \
__reg_gradient_op__##op_type##_##op_type##_grad, \
"NO_GRADIENT must be called in global namespace")
#define REGISTER_OP_GPU_KERNEL(op_type, ...) \
REGISTER_OP_KERNEL(op_type, GPU, ::paddle::platform::GPUPlace, __VA_ARGS__)
......@@ -332,23 +315,6 @@ class OpKernelRegistrar : public Registrar {
static int use_op_itself_##op_type##_ __attribute__((unused)) = \
TouchOpRegistrar_##op_type()
// TODO(fengjiayi): Most ops' gradient op have not been compeleted. So we use
// `NO_GRAD` to disable micro USE_OP_GRADIENT(op_type). Otherwise the code can't
// be compiled. `NO_GRAD` should be removed after all gradient ops are
// compeleted.
#define NO_GRAD
#ifndef NO_GRAD
#define USE_OP_GRADIENT(op_type) \
STATIC_ASSERT_GLOBAL_NAMESPACE( \
__use_op_gradient_##op_type, \
"USE_OP_GRADIENT must be called in global namespace"); \
extern int TouchOpGradientRegistrar_##op_type(); \
static int use_op_gradient_##op_type##_ __attribute__((unused)) = \
TouchOpGradientRegistrar_##op_type()
#else
#define USE_OP_GRADIENT(op_type)
#endif
#define USE_OP_DEVICE_KERNEL(op_type, DEVICE_TYPE) \
STATIC_ASSERT_GLOBAL_NAMESPACE( \
__use_op_kernel_##op_type##_##DEVICE_TYPE##__, \
......@@ -368,18 +334,13 @@ class OpKernelRegistrar : public Registrar {
USE_OP_DEVICE_KERNEL(op_type, GPU)
#endif
#define USE_NO_GRAD_OP(op_type) \
USE_OP_ITSELF(op_type); \
USE_OP_KERNEL(op_type)
#define USE_CPU_OP(op_type) \
USE_OP_ITSELF(op_type); \
USE_OP_DEVICE_KERNEL(op_type, CPU); \
USE_OP_GRADIENT(op_type)
#define USE_CPU_ONLY_OP(op_type) \
USE_OP_ITSELF(op_type); \
USE_OP_DEVICE_KERNEL(op_type, CPU);
#define USE_OP(op_type) \
USE_NO_GRAD_OP(op_type); \
USE_OP_GRADIENT(op_type)
#define USE_OP(op_type) \
USE_OP_ITSELF(op_type); \
USE_OP_KERNEL(op_type)
} // namespace framework
} // namespace paddle
......@@ -59,11 +59,10 @@ static void BuildVar(const std::string& param_name,
var->add_arguments(arg_name);
}
}
REGISTER_OP(cos_sim, paddle::framework::CosineOp,
paddle::framework::CosineOpProtoAndCheckerMaker);
REGISTER_OP(my_test_op, paddle::framework::MyTestOp,
paddle::framework::MyTestOpProtoAndCheckerMaker);
REGISTER_OP_WITHOUT_GRADIENT(cos_sim, paddle::framework::CosineOp,
paddle::framework::CosineOpProtoAndCheckerMaker);
REGISTER_OP_WITHOUT_GRADIENT(my_test_op, paddle::framework::MyTestOp,
paddle::framework::MyTestOpProtoAndCheckerMaker);
TEST(OpRegistry, CreateOp) {
paddle::framework::OpDesc op_desc;
......
......@@ -33,14 +33,6 @@ ExecutionContext::GetEigenDevice<platform::GPUPlace, Eigen::GpuDevice>() const {
}
#endif
static std::unordered_map<std::string, OpProto>* g_op_protos = nullptr;
std::unordered_map<std::string, OpProto>& OpProtos() {
if (g_op_protos == nullptr) {
g_op_protos = new std::unordered_map<std::string, OpProto>();
}
return *g_op_protos;
}
const std::string& OperatorBase::Input(const std::string& name) const {
auto& ins = Inputs(name);
PADDLE_ENFORCE_EQ(ins.size(), 1UL,
......@@ -149,14 +141,18 @@ std::vector<std::string> OperatorBase::OutputVars(bool has_intermediate) const {
}
return ret_val;
}
auto it = OpProtos().find(type_);
auto it = OpRegistry::op_info_map().find(type_);
PADDLE_ENFORCE(
it != OpProtos().end(),
it != OpRegistry::op_info_map().end(),
"Operator %s not registered, cannot figure out intermediate outputs",
type_);
PADDLE_ENFORCE(
it->second.proto_ != nullptr,
"Operator %s has no OpProto, cannot figure out intermediate outputs",
type_);
// get all OpProto::Var for outputs
for (auto& o : it->second.outputs()) {
for (auto& o : it->second.proto_->outputs()) {
// ignore all intermediate output
if (o.intermediate()) continue;
auto out = outputs_.find(o.name());
......
......@@ -50,8 +50,6 @@ inline std::string GradVarName(const std::string& var_name) {
return var_name + kGradVarSuffix;
}
extern std::unordered_map<std::string, OpProto>& OpProtos();
class OperatorBase;
class InferShapeContext;
class ExecutionContext;
......@@ -132,6 +130,14 @@ class OperatorBase {
AttributeMap attrs_;
};
class NOP : public OperatorBase {
public:
using OperatorBase::OperatorBase;
void InferShape(const Scope& scope) const override {}
void Run(const Scope& scope,
const platform::DeviceContext& dev_ctx) const override {}
};
class InferShapeContext {
public:
InferShapeContext(const OperatorBase& op, const Scope& scope)
......@@ -213,7 +219,7 @@ class InferShapeContext {
[&](const std::string& sub_name) {
auto var = scope_.FindVar(sub_name);
PADDLE_ENFORCE_NOT_NULL(
var, "MultiOutput(%s:%s) should not be nullptr", name,
var, "MultiOutput(%s:%s) should not be nullptr.", name,
sub_name);
return var->GetMutable<T>();
});
......
......@@ -65,8 +65,9 @@ static void BuildVar(const std::string& param_name,
}
}
REGISTER_OP(test_operator, paddle::framework::OpWithoutKernelTest,
paddle::framework::OpeWithoutKernelTestProtoAndCheckerMaker);
REGISTER_OP_WITHOUT_GRADIENT(
test_operator, paddle::framework::OpWithoutKernelTest,
paddle::framework::OpeWithoutKernelTestProtoAndCheckerMaker);
TEST(OperatorBase, all) {
paddle::framework::OpDesc op_desc;
......@@ -184,8 +185,9 @@ class CPUKernalMultiInputsTest : public OpKernel {
} // namespace framework
} // namespace paddle
REGISTER_OP(op_with_kernel, paddle::framework::OpWithKernelTest,
paddle::framework::OpKernelTestProtoAndCheckerMaker);
REGISTER_OP_WITHOUT_GRADIENT(
op_with_kernel, paddle::framework::OpWithKernelTest,
paddle::framework::OpKernelTestProtoAndCheckerMaker);
REGISTER_OP_CPU_KERNEL(op_with_kernel,
paddle::framework::CPUKernelTest<float, float>);
......@@ -210,8 +212,9 @@ TEST(OpKernel, all) {
ASSERT_EQ(paddle::framework::cpu_kernel_run_num, 1);
}
REGISTER_OP(op_multi_inputs_with_kernel, paddle::framework::OpWithKernelTest,
paddle::framework::OpKernelTestMultiInputsProtoAndCheckerMaker);
REGISTER_OP_WITHOUT_GRADIENT(
op_multi_inputs_with_kernel, paddle::framework::OpWithKernelTest,
paddle::framework::OpKernelTestMultiInputsProtoAndCheckerMaker);
REGISTER_OP_CPU_KERNEL(op_multi_inputs_with_kernel,
paddle::framework::CPUKernalMultiInputsTest);
......
......@@ -30,8 +30,8 @@ limitations under the License. */
namespace py = pybind11;
USE_OP(add_two);
USE_CPU_OP(onehot_cross_entropy);
USE_NO_GRAD_OP(sgd);
USE_CPU_ONLY_OP(onehot_cross_entropy);
USE_OP(sgd);
USE_OP(mul);
USE_OP(mean);
USE_OP(sigmoid);
......@@ -160,13 +160,16 @@ All parameter, weight, gradient are variables in Paddle.
//! @note: Be careful! PyBind will return std::string as an unicode, not
//! Python str. If you want a str object, you should cast them in Python.
m.def("get_all_op_protos", []() -> std::vector<py::bytes> {
auto &protos = OpProtos();
auto &op_info_map = OpRegistry::op_info_map();
std::vector<py::bytes> ret_values;
for (auto it = protos.begin(); it != protos.end(); ++it) {
PADDLE_ENFORCE(it->second.IsInitialized(),
"OpProto must all be initialized");
for (auto it = op_info_map.begin(); it != op_info_map.end(); ++it) {
const OpProto *proto = it->second.proto_;
if (proto == nullptr) {
continue;
}
PADDLE_ENFORCE(proto->IsInitialized(), "OpProto must all be initialized");
std::string str;
PADDLE_ENFORCE(it->second.SerializeToString(&str),
PADDLE_ENFORCE(proto->SerializeToString(&str),
"Serialize OpProto Error. This could be a bug of Paddle.");
ret_values.push_back(py::bytes(str));
}
......
......@@ -44,6 +44,8 @@ endfunction()
add_subdirectory(math)
cc_test(gather_test SRCS gather_test.cc DEPS tensor)
cc_test(scatter_test SRCS scatter_test.cc DEPS tensor)
cc_library(net_op SRCS net_op.cc DEPS op_registry)
cc_test(net_op_test SRCS net_op_test.cc DEPS net_op)
......
......@@ -57,8 +57,7 @@ class AddOpGrad : public framework::OperatorWithKernel {
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(add_two, ops::AddOp, ops::AddOpMaker);
REGISTER_GRADIENT_OP(add_two, add_two_grad, ops::AddOpGrad);
REGISTER_OP(add_two, ops::AddOp, ops::AddOpMaker, add_two_grad, ops::AddOpGrad);
REGISTER_OP_CPU_KERNEL(add_two,
ops::AddKernel<paddle::platform::CPUPlace, float>);
......@@ -68,12 +68,11 @@ OnehotCrossEntropy Operator.
namespace ops = paddle::operators;
REGISTER_OP(onehot_cross_entropy, ops::OnehotCrossEntropyOp,
ops::OnehotCrossEntropyOpMaker);
ops::OnehotCrossEntropyOpMaker, onehot_cross_entropy_grad,
ops::OnehotCrossEntropyGradientOp);
REGISTER_OP_CPU_KERNEL(
onehot_cross_entropy,
ops::OnehotCrossEntropyOpKernel<paddle::platform::CPUPlace, float>);
REGISTER_GRADIENT_OP(onehot_cross_entropy, onehot_cross_entropy_grad,
ops::OnehotCrossEntropyGradientOp);
REGISTER_OP_CPU_KERNEL(
onehot_cross_entropy_grad,
ops::OnehotCrossEntropyGradientOpKernel<paddle::platform::CPUPlace, float>);
......@@ -46,7 +46,8 @@ The output will have the same size with input.
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(fill_zeros_like, ops::FillZerosLikeOp, ops::FillZerosLikeOpMaker);
REGISTER_OP_WITHOUT_GRADIENT(fill_zeros_like, ops::FillZerosLikeOp,
ops::FillZerosLikeOpMaker);
REGISTER_OP_CPU_KERNEL(
fill_zeros_like,
ops::FillZerosLikeKernel<paddle::platform::CPUPlace, float>);
......@@ -29,7 +29,7 @@ void CPUGather(const T* params, const int* indices, const int slice_size,
const int index_size, T* output) {
const size_t slice_bytes = slice_size * sizeof(T);
for (size_t i = 0; i < index_size; ++i) {
for (int i = 0; i < index_size; ++i) {
int index_ = indices[i];
memcpy(output + i * slice_size, params + index_ * slice_size, slice_bytes);
}
......@@ -60,7 +60,7 @@ void Gather(const platform::Place& place, const paddle::framework::Tensor* src,
// slice size
int slice_size = 1;
for (size_t i = 1; i < src_dims.size(); ++i) slice_size *= src_dims[i];
for (int i = 1; i < src_dims.size(); ++i) slice_size *= src_dims[i];
// Gathering
if (platform::is_cpu_place(place)) {
......
......@@ -35,7 +35,7 @@ TEST(Gather, GatherData) {
p_src = src->mutable_data<int>(make_ddim({3, 4}), CPUPlace());
p_index = index->mutable_data<int>(make_ddim({2}), CPUPlace());
for (size_t i = 0; i < 12; ++i) p_src[i] = i;
for (int i = 0; i < 12; ++i) p_src[i] = i;
p_index[0] = 1;
p_index[1] = 0;
......@@ -43,6 +43,6 @@ TEST(Gather, GatherData) {
Gather<int>(CPUPlace(), src, index, output);
for (size_t i = 0; i < 4; ++i) EXPECT_EQ(p_output[i], i + 4);
for (size_t i = 4; i < 8; ++i) EXPECT_EQ(p_output[i], i - 4);
for (int i = 0; i < 4; ++i) EXPECT_EQ(p_output[i], i + 4);
for (int i = 4; i < 8; ++i) EXPECT_EQ(p_output[i], i - 4);
}
......@@ -81,5 +81,6 @@ Use to initialize tensor with gaussian random generator.
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(gaussian_random, ops::GaussianRandomOp, ops::GaussianRandomOpMaker);
REGISTER_OP_WITHOUT_GRADIENT(gaussian_random, ops::GaussianRandomOp,
ops::GaussianRandomOpMaker);
REGISTER_OP_CPU_KERNEL(gaussian_random, ops::GaussianRandomKernel<float>);
......@@ -54,9 +54,8 @@ class MeanGradOp : public framework::OperatorWithKernel {
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(mean, ops::MeanOp, ops::MeanOpMaker);
REGISTER_OP(mean, ops::MeanOp, ops::MeanOpMaker, mean_grad, ops::MeanGradOp);
REGISTER_OP_CPU_KERNEL(mean,
ops::MeanKernel<paddle::platform::CPUPlace, float>);
REGISTER_GRADIENT_OP(mean, mean_grad, ops::MeanGradOp);
REGISTER_OP_CPU_KERNEL(mean_grad,
ops::MeanGradKernel<paddle::platform::CPUPlace, float>);
......@@ -70,7 +70,5 @@ class MulOpGrad : public framework::OperatorWithKernel {
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(mul, ops::MulOp, ops::MulOpMaker);
REGISTER_GRADIENT_OP(mul, mul_grad, ops::MulOpGrad);
REGISTER_OP(mul, ops::MulOp, ops::MulOpMaker, mul_grad, ops::MulOpGrad);
REGISTER_OP_CPU_KERNEL(mul, ops::MulKernel<paddle::platform::CPUPlace, float>);
......@@ -20,13 +20,6 @@ class TestOp : public framework::OperatorBase {
}
};
class EmptyOp : public framework::OperatorBase {
public:
using framework::OperatorBase::OperatorBase;
void InferShape(const Scope& scope) const override {}
void Run(const Scope& scope, const DeviceContext& dev_ctx) const override {}
};
template <typename T>
void AssertSameVectorWithoutOrder(const std::vector<T>& expected,
const std::vector<T>& actual) {
......@@ -67,9 +60,9 @@ TEST(OpKernel, all) {
TEST(NetOp, insert_op) {
NetOp net;
auto op1 = std::shared_ptr<EmptyOp>(
new EmptyOp("empty", {{"X", {"x"}}, {"W", {"w1"}}, {"b", {"b1"}}},
{{"Out", {"y"}}}, {}));
auto op1 = std::shared_ptr<framework::NOP>(
new framework::NOP("empty", {{"X", {"x"}}, {"W", {"w1"}}, {"b", {"b1"}}},
{{"Out", {"y"}}}, {}));
net.AddOp(op1);
net.InsertOp(0, op1);
ASSERT_EQ(2UL, net.ops_.size());
......
......@@ -246,5 +246,6 @@ RecurrentGradientOp::RecurrentGradientOp(
} // namespace operators
} // namespace paddle
REGISTER_OP(recurrent_op, paddle::operators::RecurrentOp,
paddle::operators::RecurrentAlgorithmProtoAndCheckerMaker);
REGISTER_OP_WITHOUT_GRADIENT(
recurrent_op, paddle::operators::RecurrentOp,
paddle::operators::RecurrentAlgorithmProtoAndCheckerMaker);
......@@ -54,6 +54,7 @@ for i in xrange(X.shape[0]):
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(rowwise_add, ops::RowWiseAddOp, ops::RowWiseAddOpMaker);
REGISTER_OP_WITHOUT_GRADIENT(rowwise_add, ops::RowWiseAddOp,
ops::RowWiseAddOpMaker);
REGISTER_OP_CPU_KERNEL(
rowwise_add, ops::RowWiseAddKernel<paddle::platform::CPUPlace, float>);
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
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 <cstring>
#include "paddle/framework/ddim.h"
#include "paddle/framework/eigen.h"
#include "paddle/framework/tensor.h"
#include "paddle/platform/place.h"
namespace paddle {
namespace operators {
using Tensor = framework::Tensor;
template <typename T, int MajorType = Eigen::RowMajor,
typename IndexType = Eigen::DenseIndex>
using EigenVector = framework::EigenVector<T, MajorType, IndexType>;
// Implementation of CPU copy
template <typename T>
void CPUScatterUpdate(const paddle::framework::Tensor* src, const int* index,
const size_t index_size,
paddle::framework::Tensor* output) {
paddle::framework::DDim output_dims = output->dims();
for (size_t i = 0; i < index_size; ++i) {
int index_ = index[i];
paddle::framework::Tensor src_ = *src;
paddle::framework::Tensor output_ = *output;
if (index_size > 1) src_ = src->Slice<T>(i, i + 1);
if (output_dims[0] > 1) output_ = output->Slice<T>(index_, index_ + 1);
auto X = EigenVector<T>::Flatten(src_);
auto Y = EigenVector<T>::Flatten(output_);
Y = X + Y;
}
}
// Implementation of GPU scatter:
template <typename T>
void GPUScatterUpdate(const T* src, const int* index, const int slice_size,
const int index_size, T* output);
/**
* Return a updated tensor from source tensor, scattered according to index:
* dst[i] += src[index[i]]
* input[src]: type-T source Tensor
* input[index]: type-int index Tensor (1-D)
* return: output tensor
*/
template <typename T>
void ScatterUpdate(const platform::Place& place,
const paddle::framework::Tensor* src,
const paddle::framework::Tensor* index,
paddle::framework::Tensor* output) {
// check index of shape 1-D
PADDLE_ENFORCE(index->dims().size() == 1);
int index_size = index->dims()[0];
auto src_dims = src->dims();
auto dst_dims = output->dims();
// check src shape and dst shape should match
for (int i = 1; i < src_dims.size(); i++)
PADDLE_ENFORCE(src_dims[i] == dst_dims[i]);
// slice size
size_t slice_size = 1;
for (int i = 0; i < src_dims.size(); ++i) slice_size *= src_dims[i];
if (platform::is_cpu_place(place)) {
CPUScatterUpdate<T>(src, index->data<int>(), index_size, output);
} else {
}
}
} // namespace operators
} // namespace paddle
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
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/operators/scatter.h"
#include "paddle/framework/ddim.h"
#include "paddle/framework/tensor.h"
#include "paddle/platform/place.h"
#include <gtest/gtest.h>
#include <iostream>
#include <string>
TEST(scatter, ScatterUpdate) {
using namespace paddle::framework;
using namespace paddle::platform;
using namespace paddle::operators;
Tensor* src = new Tensor();
Tensor* index = new Tensor();
Tensor* output = new Tensor();
float* p_src = nullptr;
int* p_index = nullptr;
p_src = src->mutable_data<float>(make_ddim({1, 4}), CPUPlace());
p_index = index->mutable_data<int>(make_ddim({1}), CPUPlace());
for (size_t i = 0; i < 4; ++i) p_src[i] = float(i);
p_index[0] = 1;
float* p_output = output->mutable_data<float>(make_ddim({4, 4}), CPUPlace());
ScatterUpdate<float>(CPUPlace(), src, index, output);
for (size_t i = 0; i < 4; ++i) EXPECT_EQ(p_output[i], float(0));
for (size_t i = 0; i < 4; ++i) EXPECT_EQ(output->data<float>()[i], float(0));
for (size_t i = 4; i < 8; ++i) EXPECT_EQ(p_output[i], float(i - 4));
for (size_t i = 4; i < 8; ++i)
EXPECT_EQ(output->data<float>()[i], float(i - 4));
for (size_t i = 8; i < 16; ++i) EXPECT_EQ(p_output[i], float(0));
for (size_t i = 8; i < 16; ++i) EXPECT_EQ(output->data<float>()[i], float(0));
}
......@@ -51,6 +51,6 @@ param_out = param - learning_rate * grad;
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(sgd, ops::SGDOp, ops::SGDOpMaker);
REGISTER_OP_WITHOUT_GRADIENT(sgd, ops::SGDOp, ops::SGDOpMaker);
REGISTER_OP_CPU_KERNEL(sgd,
ops::SGDOpKernel<paddle::platform::CPUPlace, float>);
......@@ -52,9 +52,8 @@ class SigmoidOpGrad : public framework::OperatorWithKernel {
} // namespace paddle
namespace ops = paddle::operators;
REGISTER_OP(sigmoid, ops::SigmoidOp, ops::SigmoidOpMaker);
REGISTER_GRADIENT_OP(sigmoid, sigmoid_grad, ops::SigmoidOpGrad);
REGISTER_OP(sigmoid, ops::SigmoidOp, ops::SigmoidOpMaker, sigmoid_grad,
ops::SigmoidOpGrad);
REGISTER_OP_CPU_KERNEL(sigmoid,
ops::SigmoidKernel<paddle::platform::CPUPlace, float>);
REGISTER_OP_CPU_KERNEL(
......
......@@ -62,9 +62,9 @@ class SoftmaxOpGrad : public framework::OperatorWithKernel {
namespace ops = paddle::operators;
REGISTER_OP(softmax, ops::SoftmaxOp, ops::SoftmaxOpMaker);
REGISTER_OP(softmax, ops::SoftmaxOp, ops::SoftmaxOpMaker, softmax_grad,
ops::SoftmaxOpGrad);
REGISTER_OP_CPU_KERNEL(softmax,
ops::SoftmaxKernel<paddle::platform::CPUPlace, float>);
REGISTER_GRADIENT_OP(softmax, softmax_grad, ops::SoftmaxOpGrad);
REGISTER_OP_CPU_KERNEL(
softmax_grad, ops::SoftmaxGradKernel<paddle::platform::CPUPlace, float>);
......@@ -81,7 +81,7 @@ Used to initialize tensor with uniform random generator.
} // namespace operators
} // namespace paddle
REGISTER_OP(uniform_random, paddle::operators::UniformRandomOp,
paddle::operators::UniformRandomOpMaker);
REGISTER_OP_WITHOUT_GRADIENT(uniform_random, paddle::operators::UniformRandomOp,
paddle::operators::UniformRandomOpMaker);
REGISTER_OP_CPU_KERNEL(uniform_random,
paddle::operators::CPUUniformRandomKernel<float>);
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册
反馈
建议
客服 返回
顶部