提交 5bd14f6e 编写于 作者: D Dong Zhihong

Merge remote-tracking branch 'origin/develop' into fix/scope

上级 d4231b35 d7383c6d
develop 1.8.5 2.0.1-rocm-post 2.4.1 Ligoml-patch-1 OliverLPH-patch-1 OliverLPH-patch-2 PaddlePM-patch-1 PaddlePM-patch-2 ZHUI-patch-1 add_default_att add_kylinv10 add_model_benchmark_ci add_some_yaml_config addfile all_new_design_exec ascendrc ascendrelease bugfix-eval-frame-leakgae cherry-pick-fix-customOP-random-fail 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-run-program-grad-node-mem fix_check fix_concat_slice fix_custom_device_copy_sync fix_dataloader_memory_leak fix_dlpack_for fix_imperative_dygraph_error fix_newexe_gc 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 hack_event improve_sccache incuabte/new_frl incubate/frl_train_eval incubate/infrt incubate/lite incubate/new_frl incubate/new_frl_rc incubate/stride 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 operator_opt paddle_tiny_install paralleltest pass-compile-eval-frame preln_ernie prv-disable-more-cache prv-md-even-more prv-onednn-2.5 prv-reshape-mkldnn-ut2 pten_tensor_refactor release-deleted/2.5 release-rc/2.5 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/2.5 release/lite-0.1 release/llm_2.5 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-49499-test_ninja_on_ci revert-49654-prim_api_gen revert-49673-modify_get_single_cov revert-49763-fix_static_composite_gen revert-50158-fix_found_inf_bug_for_custom_optimizer revert-50188-refine_optimizer_create_accumulators revert-50335-fix_optminizer_set_auxiliary_var_bug revert-51676-flag_delete revert-51850-fix_softmaxce_dev revert-52175-dev_peak_memory revert-52186-deve revert-52523-test_py38 revert-52912-develop revert-53248-set_cmake_policy revert-54029-fix_windows_compile_bug revert-54068-support_translating_op_attribute revert-54214-modify_cmake_dependencies revert-54370-offline_pslib revert-54391-fix_cmake_md5error revert-54411-fix_cpp17_compile revert-54466-offline_pslib revert-54480-cmake-rocksdb revert-55568-fix_BF16_bug1 revert-56328-new_ir_support_vector_type_place_transfer revert-56366-fix_openssl_bug revert-56545-revert-56366-fix_openssl_bug revert-56620-fix_new_ir_ocr_bug revert-56925-check_inputs_grad_semantic revert-57005-refine_stride_flag rocm_dev_0217 sd_conv_linear_autocast semi-auto/rule-base 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 v2.5.1 v2.5.0 v2.5.0-rc1 v2.5.0-rc0 v2.4.2 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
6 合并请求!11636[IMPORTANT] MKLDNN layout: Support for sum operator,!8482Release/0.11.0,!8190Release/0.11.0,!8189Release/0.11.0,!6633给线性回归的get-started代码加上了预测的示例~~,!4725change: NewVar() to Var()
......@@ -44,7 +44,7 @@ cc_test(backward_test SRCS backward_test.cc DEPS backward recurrent_op device_co
cc_library(executor SRCS executor.cc DEPS op_registry device_context scope framework_proto backward)
set(EXECUTOR_TEST_OP elementwise_add_op gaussian_random_op feed_op fetch_op
mul_op sum_op squared_l2_distance_op fill_constant_op sgd_op)
mul_op sum_op squared_l2_distance_op fill_constant_op sgd_op mean_op)
if(WITH_GPU)
nv_test(executor_test SRCS executor_test.cc DEPS executor ${EXECUTOR_TEST_OP})
else()
......
......@@ -273,6 +273,21 @@ static bool AllGradInSet(const std::vector<std::string>& names,
return true;
}
static void CreateGradVarInBlock(BlockDescBind* block_desc,
size_t grad_op_start_index) {
auto ops = block_desc->AllOps();
for (size_t op_index = grad_op_start_index; op_index < ops.size();
++op_index) {
for (const auto& output : ops[op_index]->Outputs()) {
for (const auto& real_output : output.second) {
if (!block_desc->HasVar(real_output)) {
block_desc->Var(real_output);
}
}
}
}
}
std::vector<std::unique_ptr<OpDescBind>> MakeOpGrad(
const std::unique_ptr<OpDescBind>& op_desc,
std::unordered_set<std::string>* no_grad_vars,
......@@ -326,15 +341,16 @@ std::vector<std::unique_ptr<OpDescBind>> MakeBlockBackward(
std::unordered_map<std::string, std::vector<size_t>> dup_out_ops;
size_t grad_desc_idx = 0;
std::vector<std::unique_ptr<OpDescBind>> backward_descs;
for (auto it = op_descs.rbegin(); it != op_descs.rend(); ++it) {
std::vector<std::unique_ptr<OpDescBind>> op_grads =
MakeOpGrad(*it, no_grad_vars, grad_to_var);
if ((*it)->Type() == "recurrent") {
PADDLE_ENFORCE_EQ(
op_grads.size(), size_t(1),
op_grads.size(), static_cast<size_t>(1),
"rnn_op's gradient process should contain only one op.");
int step_block_idx = (*it)->GetBlockAttr("stop_block");
int step_block_idx = (*it)->GetBlockAttr("step_block");
auto backward_block_op_descs = MakeBlockBackward(
program_desc, step_block_idx, no_grad_vars, grad_to_var);
BlockDescBind* backward_block = program_desc.AppendBlock(*cur_block);
......@@ -380,10 +396,11 @@ std::vector<std::unique_ptr<OpDescBind>> MakeBlockBackward(
backward_descs.insert(backward_descs.begin() + p.first + 1,
std::move(p.second));
}
return backward_descs;
}
void AppendBackward(ProgramDescBind& program_desc,
void AppendBackward(ProgramDescBind& program_desc, const VarDescBind& target,
const std::unordered_set<std::string>& no_grad_vars) {
std::unordered_set<std::string> no_grad_var_names;
no_grad_var_names.reserve(no_grad_vars.size() + 1);
......@@ -391,13 +408,34 @@ void AppendBackward(ProgramDescBind& program_desc,
for (auto& name : no_grad_vars) {
no_grad_var_names.insert(GradVarName(name));
}
const int root_block_idx = 0;
auto root_block = program_desc.Block(root_block_idx);
auto& all_ops = root_block->ops_;
// insert fill one op for target
std::string fill_one_op_out = GradVarName(target.Name());
std::unique_ptr<OpDescBind> fill_one_op(
new OpDescBind("fill_constant", {}, {{"Out", {fill_one_op_out}}},
{{"shape", std::vector<int>{1}},
{"value", static_cast<float>(1.0)},
{"dataType", framework::DataType::FP32}}));
all_ops.push_back(std::move(fill_one_op));
size_t forward_op_num = all_ops.size();
size_t forward_block_num = program_desc.Size();
std::unordered_map<std::string, std::string> grad_to_var;
auto backward_op_descs = MakeBlockBackward(program_desc, root_block_idx,
&no_grad_var_names, &grad_to_var);
auto& forw_op_descs = program_desc.Block(root_block_idx)->ops_;
for (auto& ptr : backward_op_descs) {
forw_op_descs.push_back(std::move(ptr));
all_ops.push_back(std::move(ptr));
}
root_block->Var(fill_one_op_out);
// create grad_var for all blocks in this program
CreateGradVarInBlock(root_block, forward_op_num);
for (size_t block_index = forward_block_num;
block_index < program_desc.Size(); ++block_index) {
CreateGradVarInBlock(program_desc.Block(block_index), 0);
}
}
......
......@@ -29,7 +29,7 @@ extern std::unique_ptr<OperatorBase> Backward(
// TODO(jiayi): Add target as parameter and generate backward op
// according to target.
void AppendBackward(ProgramDescBind& program_desc,
void AppendBackward(ProgramDescBind& program_desc, const VarDescBind& target,
const std::unordered_set<std::string>& no_grad_vars);
} // namespace framework
......
......@@ -18,6 +18,7 @@
#include "paddle/framework/block_desc.h"
#include "paddle/framework/op_desc.h"
#include "paddle/framework/op_registry.h"
#include "paddle/framework/var_desc.h"
#include "paddle/operators/net_op.h"
namespace paddle {
......@@ -468,10 +469,14 @@ TEST(Backward, simple_single_op) {
op->SetInput("b", {"b"});
op->SetOutput("Out", {"out"});
AppendBackward(program, {});
auto target = f::VarDescBind("out");
AppendBackward(program, target, {});
ASSERT_EQ(block->AllOps().size(), 2UL);
f::OpDescBind *grad_op = block->AllOps()[1];
ASSERT_EQ(block->AllOps().size(), 3UL);
f::OpDescBind *fill_op = block->AllOps()[1];
EXPECT_EQ(fill_op->Type(), "fill_constant");
f::OpDescBind *grad_op = block->AllOps()[2];
EXPECT_EQ(grad_op->Type(), "rowwise_add_grad");
ASSERT_EQ(grad_op->InputNames().size(), 1UL);
ASSERT_EQ(grad_op->OutputNames().size(), 2UL);
......@@ -494,13 +499,17 @@ TEST(Backward, default_attribute) {
op->SetOutput("Out", {"out"});
op->CheckAttrs();
AppendBackward(program, {});
auto target = f::VarDescBind("out");
AppendBackward(program, target, {});
ASSERT_EQ(block->AllOps().size(), 2UL);
ASSERT_EQ(block->AllOps().size(), 3UL);
EXPECT_EQ(boost::get<int>(op->GetAttr("x_num_col_dims")), 1);
EXPECT_EQ(boost::get<int>(op->GetAttr("y_num_col_dims")), 1);
f::OpDescBind *grad_op = block->AllOps()[1];
f::OpDescBind *fill_op = block->AllOps()[1];
EXPECT_EQ(fill_op->Type(), "fill_constant");
f::OpDescBind *grad_op = block->AllOps()[2];
ASSERT_EQ(grad_op->Type(), "mul_grad");
EXPECT_EQ(boost::get<int>(grad_op->GetAttr("x_num_col_dims")), 1);
EXPECT_EQ(boost::get<int>(grad_op->GetAttr("y_num_col_dims")), 1);
......@@ -528,10 +537,15 @@ TEST(Backward, simple_mult_op) {
op3->SetInput("b", {"b3"});
op3->SetOutput("Out", {"out3"});
AppendBackward(program, {});
auto target = f::VarDescBind("out3");
size_t forward_len = block->AllOps().size();
AppendBackward(program, target, {});
ASSERT_EQ(block->AllOps().size(), 6UL);
f::OpDescBind *grad_op1 = block->AllOps()[5];
ASSERT_EQ(block->AllOps().size(), 6UL + 1);
f::OpDescBind *fill_op = block->AllOps()[forward_len];
EXPECT_EQ(fill_op->Type(), "fill_constant");
f::OpDescBind *grad_op1 = block->AllOps()[6];
EXPECT_EQ(grad_op1->Type(), "rowwise_add_grad");
ASSERT_EQ(grad_op1->InputNames().size(), 1UL);
ASSERT_EQ(grad_op1->OutputNames().size(), 2UL);
......@@ -542,7 +556,7 @@ TEST(Backward, simple_mult_op) {
EXPECT_EQ(grad_op1->Output(f::GradVarName("b")),
std::vector<std::string>({f::GradVarName("b1")}));
f::OpDescBind *grad_op2 = block->AllOps()[4];
f::OpDescBind *grad_op2 = block->AllOps()[5];
EXPECT_EQ(grad_op2->Type(), "mul_grad");
ASSERT_EQ(grad_op2->InputNames().size(), 4UL);
ASSERT_EQ(grad_op2->OutputNames().size(), 2UL);
......@@ -556,7 +570,7 @@ TEST(Backward, simple_mult_op) {
EXPECT_EQ(grad_op2->Output(f::GradVarName("Y")),
std::vector<std::string>({f::GradVarName("y2")}));
f::OpDescBind *grad_op3 = block->AllOps()[3];
f::OpDescBind *grad_op3 = block->AllOps()[4];
EXPECT_EQ(grad_op3->Type(), "rowwise_add_grad");
ASSERT_EQ(grad_op3->InputNames().size(), 1UL);
ASSERT_EQ(grad_op3->OutputNames().size(), 2UL);
......@@ -596,10 +610,15 @@ TEST(Backward, intermedia_var_no_grad) {
op4->SetInput("Y", {"out3"});
op4->SetOutput("Out", {"out4"});
AppendBackward(program, {"out3"});
auto target = f::VarDescBind("out4");
size_t forward_len = block->AllOps().size();
AppendBackward(program, target, {"out3"});
ASSERT_EQ(block->AllOps().size(), 6UL);
f::OpDescBind *grad_op1 = block->AllOps()[5];
ASSERT_EQ(block->AllOps().size(), 7UL);
f::OpDescBind *fill_op = block->AllOps()[forward_len];
EXPECT_EQ(fill_op->Type(), "fill_constant");
f::OpDescBind *grad_op1 = block->AllOps()[6];
EXPECT_EQ(grad_op1->Type(), "rowwise_add_grad");
ASSERT_EQ(grad_op1->InputNames().size(), 1UL);
ASSERT_EQ(grad_op1->OutputNames().size(), 2UL);
......@@ -610,7 +629,7 @@ TEST(Backward, intermedia_var_no_grad) {
EXPECT_EQ(grad_op1->Output(f::GradVarName("b")),
std::vector<std::string>({f::GradVarName("b1")}));
f::OpDescBind *grad_op4 = block->AllOps()[4];
f::OpDescBind *grad_op4 = block->AllOps()[5];
EXPECT_EQ(grad_op4->Type(), "mul_grad");
ASSERT_EQ(grad_op4->InputNames().size(), 4UL);
ASSERT_EQ(grad_op4->OutputNames().size(), 2UL);
......@@ -642,10 +661,15 @@ TEST(Backward, var_no_grad) {
op2->SetOutput("Y", {"y2"});
op2->SetOutput("Z", {"z2"});
AppendBackward(program, {"z1"});
auto target = f::VarDescBind("z2");
size_t forward_len = block->AllOps().size();
AppendBackward(program, target, {"z1"});
ASSERT_EQ(block->AllOps().size(), 5UL);
f::OpDescBind *grad_op2 = block->AllOps()[2];
ASSERT_EQ(block->AllOps().size(), 6UL);
f::OpDescBind *fill_op = block->AllOps()[forward_len];
EXPECT_EQ(fill_op->Type(), "fill_constant");
f::OpDescBind *grad_op2 = block->AllOps()[3];
ASSERT_EQ(grad_op2->Type(), "mult_in_out_grad");
ASSERT_EQ(grad_op2->InputNames().size(), 6UL);
ASSERT_EQ(grad_op2->OutputNames().size(), 2UL);
......@@ -661,7 +685,7 @@ TEST(Backward, var_no_grad) {
std::vector<std::string>({f::GradVarName("y1")}));
EXPECT_EQ(grad_op2->Output(f::GradVarName("H")), std::vector<std::string>());
f::OpDescBind *fill_zero_op = block->AllOps()[3];
f::OpDescBind *fill_zero_op = block->AllOps()[4];
ASSERT_EQ(fill_zero_op->Type(), "fill_zeros_like");
ASSERT_EQ(fill_zero_op->InputNames().size(), 1UL);
ASSERT_EQ(fill_zero_op->OutputNames().size(), 1UL);
......@@ -669,7 +693,7 @@ TEST(Backward, var_no_grad) {
EXPECT_EQ(fill_zero_op->Output("Y"),
std::vector<std::string>({std::string("z1") + f::kZeroVarSuffix}));
f::OpDescBind *grad_op1 = block->AllOps()[4];
f::OpDescBind *grad_op1 = block->AllOps()[5];
ASSERT_EQ(grad_op1->Type(), "mult_in_out_grad");
ASSERT_EQ(grad_op1->InputNames().size(), 6UL);
ASSERT_EQ(grad_op1->OutputNames().size(), 2UL);
......@@ -709,10 +733,15 @@ TEST(Backward, shared_var) {
op3->SetInput("b", {"b3"});
op3->SetOutput("Out", {"out3"});
AppendBackward(program, {});
auto target = f::VarDescBind("out3");
size_t forward_len = block->AllOps().size();
AppendBackward(program, target, {});
ASSERT_EQ(block->AllOps().size(), 7UL);
f::OpDescBind *grad_op3 = block->AllOps()[3];
ASSERT_EQ(block->AllOps().size(), 8UL);
f::OpDescBind *fill_op = block->AllOps()[forward_len];
EXPECT_EQ(fill_op->Type(), "fill_constant");
f::OpDescBind *grad_op3 = block->AllOps()[4];
ASSERT_EQ(grad_op3->Type(), "rowwise_add_grad");
ASSERT_EQ(grad_op3->InputNames().size(), 1UL);
ASSERT_EQ(grad_op3->OutputNames().size(), 2UL);
......@@ -723,7 +752,7 @@ TEST(Backward, shared_var) {
EXPECT_EQ(grad_op3->Output(f::GradVarName("b")),
std::vector<std::string>({f::GradVarName("b3")}));
f::OpDescBind *grad_op4 = block->AllOps()[4];
f::OpDescBind *grad_op4 = block->AllOps()[5];
ASSERT_EQ(grad_op4->Type(), "mul_grad");
ASSERT_EQ(grad_op4->InputNames().size(), 4UL);
ASSERT_EQ(grad_op4->OutputNames().size(), 2UL);
......@@ -737,7 +766,7 @@ TEST(Backward, shared_var) {
EXPECT_EQ(grad_op4->Output(f::GradVarName("Y")),
std::vector<std::string>({f::GradVarName("y2")}));
f::OpDescBind *sum_op = block->AllOps()[5];
f::OpDescBind *sum_op = block->AllOps()[6];
ASSERT_EQ(sum_op->Type(), "sum");
ASSERT_EQ(sum_op->InputNames().size(), 1UL);
ASSERT_EQ(sum_op->OutputNames().size(), 1UL);
......@@ -747,7 +776,7 @@ TEST(Backward, shared_var) {
EXPECT_EQ(sum_op->Output("Out"),
std::vector<std::string>({f::GradVarName("out1")}));
f::OpDescBind *grad_op1 = block->AllOps()[6];
f::OpDescBind *grad_op1 = block->AllOps()[7];
ASSERT_EQ(grad_op1->Type(), "rowwise_add_grad");
ASSERT_EQ(grad_op1->InputNames().size(), 1UL);
ASSERT_EQ(grad_op1->OutputNames().size(), 2UL);
......@@ -769,7 +798,11 @@ TEST(Backward, half_backward) {
op1->SetInput("Y", {"b"});
op1->SetOutput("Out", {"out"});
AppendBackward(program, {"b"});
auto target = f::VarDescBind("out");
size_t forward_len = block->AllOps().size();
AppendBackward(program, target, {"b"});
f::OpDescBind *fill_op = block->AllOps()[forward_len];
EXPECT_EQ(fill_op->Type(), "fill_constant");
auto ops = block->AllOps();
ASSERT_EQ(2UL, ops.size());
}
\ No newline at end of file
ASSERT_EQ(3UL, ops.size());
}
......@@ -39,7 +39,7 @@ class BlockDescBind {
std::unordered_map<std::string, std::string> *grad_to_var);
friend void AppendBackward(
ProgramDescBind &program_desc,
ProgramDescBind &program_desc, const VarDescBind &target,
const std::unordered_set<std::string> &no_grad_vars);
BlockDescBind(ProgramDescBind *prog, BlockDesc *desc)
......
......@@ -34,6 +34,7 @@ USE_OP(mul);
USE_OP(sum);
USE_OP(squared_l2_distance);
USE_OP(fill_constant);
USE_OP(mean);
USE_OP(sgd);
using namespace paddle::platform;
......@@ -45,9 +46,16 @@ void AddOp(const std::string& type, const VariableNameMap& inputs,
// insert output
for (auto kv : outputs) {
for (auto v : kv.second) {
auto var = block->Var(v);
var->SetType(VarDesc::LOD_TENSOR);
var->SetDataType(paddle::framework::DataType::FP32);
// <<<<<<< HEAD
// auto var = block->Var(v);
// var->SetType(VarDesc::LOD_TENSOR);
// var->SetDataType(paddle::framework::DataType::FP32);
// =======
if (!block->HasVar(v)) {
auto var = block->Var(v);
var->SetDataType(paddle::framework::DataType::FP32);
}
// >>>>>>> origin/develop
}
}
......@@ -147,12 +155,12 @@ class ExecutorTesterRandom : public ::testing::Test {
AddOp("squared_l2_distance", {{"X", {"a"}}, {"Y", {"a_out"}}},
{{"Out", {"l2_distance"}}, {"sub_result", {"l2_distance_sub"}}}, {},
root_block);
AddOp("mean", {{"X", {"l2_distance"}}}, {{"Out", {"mean_out"}}}, {},
root_block);
// backward
AddOp("fill_constant", {}, {{"Out", {"l2_distance@GRAD"}}},
{{"shape", std::vector<int>{batch_size, 1}}, {"value", float(1.0)}},
root_block);
AppendBackward(program, {});
auto target = VarDescBind("mean_out");
AppendBackward(program, target, {});
// update
AddOp("fill_constant", {}, {{"Out", {"learning_rate"}}},
......
......@@ -54,7 +54,10 @@ inline void VectorToRepeated(const std::vector<bool> &vec,
class VarDescBind {
public:
explicit VarDescBind(const std::string &name) { desc_.set_name(name); }
explicit VarDescBind(const std::string &name) {
desc_.set_name(name);
desc_.set_type(VarDesc::LOD_TENSOR);
}
VarDesc *Proto() { return &desc_; }
......
......@@ -118,9 +118,9 @@ void BindProgramDesc(py::module &m) {
.def("append_block", &ProgramDescBind::AppendBlock,
py::return_value_policy::reference)
.def("append_backward",
[](ProgramDescBind &program_desc,
[](ProgramDescBind &program_desc, const VarDescBind &target,
const std::unordered_set<std::string> &no_grad_vars) {
AppendBackward(program_desc, no_grad_vars);
AppendBackward(program_desc, target, no_grad_vars);
})
.def("block", &ProgramDescBind::Block, py::return_value_policy::reference)
.def("num_blocks", &ProgramDescBind::Size)
......
......@@ -176,6 +176,18 @@ class Operator(object):
proto = OpProtoHolder.instance().get_op_proto(type)
if inputs is not None:
given = set()
need = set()
for n in inputs:
given.add(n)
for m in proto.inputs:
need.add(m.name)
if not given == need:
raise ValueError(
"Incorrect setting for input(s) of operator \"%s\". Need: [%s] Given: [%s]"
% (type, ", ".join(str(e) for e in need), ", ".join(
str(e) for e in given)))
for in_proto in proto.inputs:
in_argus = inputs[in_proto.name]
if not isinstance(in_argus, list):
......@@ -190,6 +202,18 @@ class Operator(object):
self.desc.set_input(in_proto.name, in_argu_names)
if outputs is not None:
given = set()
need = set()
for n in outputs:
given.add(n)
for m in proto.outputs:
need.add(m.name)
if not given == need:
raise ValueError(
"Incorrect setting for output(s) of operator \"%s\". Need: [%s] Given: [%s]"
% (type, ", ".join(str(e) for e in need), ", ".join(
str(e) for e in given)))
for out_proto in proto.outputs:
out_argus = outputs[out_proto.name]
if not isinstance(out_argus, list):
......
......@@ -51,11 +51,14 @@ class TestProgram(unittest.TestCase):
sum_op_desc.set_input("Y", ["b1"])
sum_op_desc.set_output("Out", ["out2"])
target = block.new_var("out2")
expect_ops = [
"mul", "elementwise_add", "elementwise_add_grad", "mul_grad"
"mul", "elementwise_add", "fill_constant", "elementwise_add_grad",
"mul_grad"
]
actual_ops = []
prog.append_backward(set())
prog.append_backward(target, set())
for op in block.all_ops():
actual_ops.append(op.type())
print(actual_ops)
......
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