Merge remote-tracking branch 'ups/develop' into refine/op/gru

Author: tensor-tang

doc/fluid/howto/optimization/timeline_cn.md

@@ -1,21 +1,27 @@
 # 如何使用timeline工具做性能分析
 
-1. 在训练的主循环外加上`with profiler.profiler(...)`。运行之后，代码会在`/tmp/profile`目录下生成一个profile的记录文件。
+1. 在训练的主循环外加上`profiler.start_profiler(...)`和`profiler.stop_profiler(...)`。运行之后，代码会在`/tmp/profile`目录下生成一个profile的记录文件。
 
 	**提示：**
 	请不要在timeline记录信息时运行太多次迭代，因为timeline中的记录数量和迭代次数是成正比的。
 
 	```python
-	with profiler.profiler('All', 'total', '/tmp/profile') as prof:
-	    for pass_id in range(pass_num):
-	        for batch_id, data in enumerate(train_reader()):
-	            exe.run(fluid.default_main_program(),
-	                    feed=feeder.feed(data),
-	                    fetch_list=[])
+    for pass_id in range(pass_num):
+        for batch_id, data in enumerate(train_reader()):
+            if pass_id == 0 and batch_id == 5:
+                profiler.start_profiler("All")
+            elif pass_id == 0 and batch_id == 10:
+                profiler.stop_profiler("total", "/tmp/profile")
+            exe.run(fluid.default_main_program(),
+                    feed=feeder.feed(data),
+                    fetch_list=[])
 	            ...
 	```
 
 1. 运行`python paddle/tools/timeline.py`来处理`/tmp/profile`，这个程序默认会生成一个`/tmp/timeline`文件，你也可以用命令行参数来修改这个路径，请参考[timeline.py](https://github.com/PaddlePaddle/Paddle/blob/develop/tools/timeline.py)。
+```python
+python Paddle/tools/timeline.py --profile_path=/tmp/profile --timeline_path=timeline
+```
 
 1. 打开chrome浏览器，访问<chrome://tracing/>，用`load`按钮来加载生成的`timeline`文件。
 

doc/fluid/howto/optimization/timeline_en.md

@@ -1,22 +1,28 @@
 # how to use timeline tool to do profile
 
-1. Add `with profiler.profiler(...)` to the main training loop. After run, the code will generate a profile record file `/tmp/profile`. **Warning**: Please do not run too many batches when use profiler to record timeline information, for the profile record will grow with the batch number.
+1. Add `profiler.start_profiler(...)`和`profiler.stop_profiler(...)` to the main training loop. After run, the code will generate a profile record file `/tmp/profile`. **Warning**: Please do not run too many batches when use profiler to record timeline information, for the profile record will grow with the batch number.
 
 	```python
-	with profiler.profiler('All', 'total', '/tmp/profile') as prof:
-	    for pass_id in range(pass_num):
-	        for batch_id, data in enumerate(train_reader()):
-	            exe.run(fluid.default_main_program(),
-	                    feed=feeder.feed(data),
-	                    fetch_list=[],
-	                    use_program_cache=True)
+    for pass_id in range(pass_num):
+        for batch_id, data in enumerate(train_reader()):
+            if pass_id == 0 and batch_id == 5:
+                profiler.start_profiler("All")
+            elif pass_id == 0 and batch_id == 10:
+                profiler.stop_profiler("total", "/tmp/profile")
+            exe.run(fluid.default_main_program(),
+                    feed=feeder.feed(data),
+                    fetch_list=[])
 	            ...
 	```
 
 1. Run `python paddle/tools/timeline.py` to process `/tmp/profile`, it will generate another
 file `/tmp/timeline` by default. You can change the path by cmd parameter, please take a look at
 [timeline.py](https://github.com/PaddlePaddle/Paddle/blob/develop/tools/timeline.py) for details.
 
+```python
+python Paddle/tools/timeline.py --profile_path=/tmp/profile --timeline_path=timeline
+```
+
 1. Open chrome and visit <chrome://tracing/>, use `load` button to load the generated `timeline` file.
 
 	![chrome tracing](./tracing.jpeg)

paddle/fluid/framework/details/all_reduce_op_handle.cc

@@ -17,6 +17,7 @@
 #include "paddle/fluid/framework/details/container_cast.h"
 #include "paddle/fluid/framework/details/reduce_and_gather.h"
 #include "paddle/fluid/framework/details/variable_visitor.h"
+#include "paddle/fluid/platform/profiler.h"
 
 namespace paddle {
 namespace framework {
@@ -45,6 +46,7 @@ AllReduceOpHandle::AllReduceOpHandle(ir::Node *node,
 #endif
 
 void AllReduceOpHandle::RunImpl() {
+  platform::RecordEvent r("all_reduce", nullptr);
   if (NoDummyInputSize() == 1) {
     return;  // No need to all reduce when GPU count = 1;
   } else {

paddle/fluid/framework/details/reduce_op_handle.cc

@@ -16,12 +16,14 @@
 #include "paddle/fluid/framework/details/container_cast.h"
 #include "paddle/fluid/framework/details/reduce_and_gather.h"
 #include "paddle/fluid/framework/details/variable_visitor.h"
+#include "paddle/fluid/platform/profiler.h"
 
 namespace paddle {
 namespace framework {
 namespace details {
 
 void ReduceOpHandle::RunImpl() {
+  platform::RecordEvent r("reduce", nullptr);
   if (places_.size() == 1) return;
   // the input and output may have dummy var.
   auto in_var_handles = DynamicCast<VarHandle>(inputs_);

paddle/fluid/framework/details/scope_buffered_ssa_graph_executor.cc

@@ -17,6 +17,7 @@
 #include <string>
 #include <vector>
 #include "paddle/fluid/framework/executor.h"
+#include "paddle/fluid/platform/profiler.h"
 
 namespace paddle {
 namespace framework {
@@ -62,6 +63,7 @@ FeedFetchList ScopeBufferedSSAGraphExecutor::Run(
     eptr = std::current_exception();
   }
 
+  platform::RecordEvent e("ScopeBufferedSSAGraphExecutorAfterRun", nullptr);
   drop_scope_counter_ += 1;
   if (!fetch_tensors.empty() ||
       drop_scope_counter_ == strategy_.num_iteration_per_drop_scope_) {

paddle/fluid/framework/details/threaded_ssa_graph_executor.cc

@@ -15,6 +15,7 @@
 #include "paddle/fluid/framework/details/threaded_ssa_graph_executor.h"
 
 #include "paddle/fluid/framework/details/ssa_graph_builder.h"
+#include "paddle/fluid/platform/profiler.h"
 
 namespace paddle {
 namespace framework {
@@ -34,6 +35,8 @@ ThreadedSSAGraphExecutor::ThreadedSSAGraphExecutor(
 
 FeedFetchList ThreadedSSAGraphExecutor::Run(
     const std::vector<std::string> &fetch_tensors) {
+  std::unique_ptr<platform::RecordEvent> event(
+      new platform::RecordEvent("ThreadedSSAGraphExecutorPrepare", nullptr));
   std::unordered_map<OpHandleBase *, size_t> pending_ops;
   std::unordered_set<VarHandleBase *> pending_vars;
   BlockingQueue<VarHandleBase *> ready_vars;
@@ -84,6 +87,7 @@ FeedFetchList ThreadedSSAGraphExecutor::Run(
   // Clean run context
   run_op_futures_.clear();
   exception_holder_.Clear();
+  event.reset(nullptr);
 
   // Step 3. Execution
   while (!pending_vars.empty()) {

paddle/fluid/operators/lookup_table_op.cc

@@ -32,11 +32,16 @@ class LookupTableOp : public framework::OperatorWithKernel {
 
     auto table_dims = ctx->GetInputDim("W");
     auto ids_dims = ctx->GetInputDim("Ids");
+    int ids_rank = ids_dims.size();
 
-    PADDLE_ENFORCE_EQ(ids_dims.size(), 2);
-    PADDLE_ENFORCE_EQ(ids_dims[1], 1);
+    PADDLE_ENFORCE_EQ(table_dims.size(), 2);
+    PADDLE_ENFORCE_EQ(ids_dims[ids_rank - 1], 1,
+                      "The last dimension of the 'Ids' tensor must be 1.");
 
-    ctx->SetOutputDim("Out", {ids_dims[0], table_dims[1]});
+    auto output_dims =
+        framework::vectorize(framework::slice_ddim(ids_dims, 0, ids_rank - 1));
+    output_dims.push_back(table_dims[1]);
+    ctx->SetOutputDim("Out", framework::make_ddim(output_dims));
 
     if (ctx->GetOutputsVarType("Out")[0] ==
         framework::proto::VarType::LOD_TENSOR) {
@@ -61,8 +66,7 @@ class LookupTableOpMaker : public framework::OpProtoAndCheckerMaker {
     AddInput("Ids",
              "An input with type int32 or int64 "
              "contains the ids to be looked up in W. "
-             "Ids must be a column vector with rank = 2. "
-             "The 2nd dimension size must be 1.");
+             "The last dimension size must be 1.");
     AddOutput("Out", "The lookup results, which have the same type as W.");
     AddAttr<bool>("is_sparse",
                   "(boolean, default false) "

paddle/fluid/operators/lookup_table_op.cu

@@ -118,28 +118,31 @@ class LookupTableGradCUDAKernel : public framework::OpKernel<T> {
       auto *d_table = context.Output<SelectedRows>(framework::GradVarName("W"));
 
       auto *ids_data = ids->data<int64_t>();
-      auto ids_dim = ids->dims();
+      int64_t ids_num = ids->numel();
 
       auto stream = dev_ctx.stream();
       // copy GPU memory to CPU pinned memory
       framework::Vector<int64_t> new_rows;
-      new_rows.resize(ids_dim[0]);
+      new_rows.resize(ids_num);
       auto gpu_place = boost::get<platform::CUDAPlace>(context.GetPlace());
 
       // TODO(yuyang18): Strange code here.
       memory::Copy(platform::CPUPlace(),
                    new_rows.CUDAMutableData(context.GetPlace()), gpu_place,
-                   ids_data, ids_dim[0] * sizeof(int64_t), stream);
+                   ids_data, ids_num * sizeof(int64_t), stream);
 
       d_table->set_rows(new_rows);
 
       auto *d_table_value = d_table->mutable_value();
-      d_table_value->Resize({ids_dim[0], table->dims()[1]});
+      d_table_value->Resize({ids_num, table->dims()[1]});
       d_table_value->mutable_data<T>(context.GetPlace());
 
       auto *d_table_data = d_table_value->data<T>();
       auto *d_output_data = d_output->data<T>();
-      PADDLE_ENFORCE_EQ(d_table_value->dims(), d_output->dims());
+      auto d_output_dims = d_output->dims();
+      PADDLE_ENFORCE_EQ(
+          d_table_value->dims(),
+          framework::flatten_to_2d(d_output_dims, d_output_dims.size() - 1));
       memory::Copy(gpu_place, d_table_data, gpu_place, d_output_data,
                    d_output->numel() * sizeof(T), stream);
 

paddle/fluid/operators/lookup_table_op.h

@@ -109,36 +109,38 @@ class LookupTableGradKernel : public framework::OpKernel<T> {
       auto *d_table = context.Output<SelectedRows>(framework::GradVarName("W"));
 
       auto *ids_data = ids->data<int64_t>();
-      auto ids_dim = ids->dims();
+      int64_t ids_num = ids->numel();
 
       framework::Vector<int64_t> new_rows;
-      new_rows.reserve(ids_dim[0]);
-      for (int64_t i = 0; i < ids_dim[0]; i++) {
+      new_rows.reserve(ids_num);
+      for (int64_t i = 0; i < ids_num; i++) {
         new_rows.push_back(ids_data[i]);
       }
       d_table->set_rows(new_rows);
 
       auto *d_table_value = d_table->mutable_value();
-      d_table_value->Resize({ids_dim[0], table_dim[1]});
+      d_table_value->Resize({ids_num, table_dim[1]});
       d_table_value->mutable_data<T>(context.GetPlace());
 
       d_table->set_height(table_dim[0]);
 
       auto *d_output_data = d_output->data<T>();
       auto *d_table_data = d_table_value->data<T>();
 
-      PADDLE_ENFORCE_EQ(d_table_value->dims(), d_output->dims());
+      auto d_output_dims = d_output->dims();
+      PADDLE_ENFORCE_EQ(
+          d_table_value->dims(),
+          framework::flatten_to_2d(d_output_dims, d_output_dims.size() - 1));
       memcpy(d_table_data, d_output_data, sizeof(T) * d_output->numel());
     } else {
       auto *ids = context.Input<LoDTensor>("Ids");
       auto *d_output = context.Input<LoDTensor>(framework::GradVarName("Out"));
       auto *d_table = context.Output<LoDTensor>(framework::GradVarName("W"));
 
       auto *ids_data = ids->data<int64_t>();
-      auto ids_dim = ids->dims();
 
       int N = table_dim[0];
-      int D = d_output->dims()[1];
+      int D = table_dim[1];
 
       auto *d_output_data = d_output->data<T>();
       auto *d_table_data = d_table->mutable_data<T>(context.GetPlace());

paddle/fluid/operators/parallel_do_op.cc

@@ -18,7 +18,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/threadpool.h"
 #include "paddle/fluid/operators/detail/safe_ref.h"
-#include "paddle/fluid/platform/profiler.h"
 
 namespace paddle {
 namespace operators {
@@ -166,8 +165,6 @@ class ParallelDoOp : public framework::OperatorBase {
 
       workers.emplace_back(
           framework::Async([program, cur_scope, place, block, place_idx] {
-            // Give the thread an id to distinguish parallel block with same id.
-            platform::RecordThread rt(static_cast<int>(place_idx) + 1);
             framework::Executor executor(place);
             executor.Run(*program, cur_scope, block->ID(),
                          false /*create_local_scope*/);
@@ -244,8 +241,6 @@ class ParallelDoGradOp : public framework::OperatorBase {
       // execute
       workers.emplace_back(
           framework::Async([program, cur_scope, place, block, i] {
-            // Give the thread an id to distinguish parallel block with same id.
-            platform::RecordThread rt(static_cast<int>(i) + 1);
             framework::Executor executor(place);
             executor.Run(*program, cur_scope, block->ID(),
                          false /*create_local_scope*/);