Feature/add transform data dygraph (#19707)

* refactor dygraph,test=develop

* fix failed unittest,test=develop

* polish code,test=develop

* check windows ci error,test=develop
try to fix windows ci error by np.allclose,test=develop

* polish vlog and profiler, test=develop

* try to fix preceding ops order,test=develop

* test transformer in windows ci, test=develop

* use python c-api to speed up tracer.trace,test=develop

* test=develop, fix docker with paddle nccl problem

* test=develop, add ut for debug string and gradient_accumulator

* test=develop, add tests for layer/gradient_accumulator/prepared_op

* test=develop, fix complie error for test_prepared_op

* test=develop, add more ut for dygraph

* test=develop, create API.spec for dygraph api change

* add transform_data to dygraph

* test=develop, refoctor name to make it easier to understand

* test=develop, refoctor name to make it easier to understand

* add test and change input to const ref for safety

* test=develop, fix multi-gpu failed problem , add Tracer tests, change PADDLEENFORCE to PADDLEENFORCE_EQ

* add ut for data transform

* refine ut for data_transform

* test=develop, fix ut failed on parallel se-resnext

* test=develop, change one more PADDLE_ENFORCE

* add test_tracer on multiple devices

* test=develop, change place to mutable for data transform

* test=develop, add transform data on same place test and remove useless log

* test=develop, Add to do for data layout and and ut for conv2d with no bias

paddle/fluid/framework/operator.h

@@ -463,7 +463,8 @@ class OperatorWithKernel : public OperatorBase {
 
   std::vector<KernelConfig>* GetKernelConfig(const OpKernelType& key) const;
 
- protected:
+  // change this to public so that in dygraph mode we can call it to check if we
+  // need transform data
   virtual OpKernelType GetKernelTypeForVar(
       const std::string& var_name, const Tensor& tensor,
       const OpKernelType& expected_kernel_type) const;

paddle/fluid/imperative/CMakeLists.txt

@@ -1,6 +1,6 @@
 cc_library(imperative_flag SRCS flags.cc DEPS gflags) 
 
-cc_library(prepared_operator SRCS prepared_operator.cc DEPS proto_desc operator device_context lod_tensor selected_rows var_type_traits)
+cc_library(prepared_operator SRCS prepared_operator.cc DEPS proto_desc operator device_context lod_tensor selected_rows var_type_traits op_kernel_type data_transform)
 cc_library(layer SRCS layer.cc DEPS prepared_operator math_function imperative_flag variable_helper op_registry)
 cc_library(gradient_accumulator SRCS gradient_accumulator.cc DEPS blas operator lod_tensor selected_rows var_type_traits layer)
 cc_library(tracer SRCS tracer.cc DEPS layer engine)

paddle/fluid/imperative/layer.cc

@@ -264,11 +264,11 @@ void OpBase::Run(const NameVarBaseMap& ins, const NameVarBaseMap& outs) {
   VLOG(3) << "Running Op " << Type();
   VLOG(5) << LayerDebugString(Type(), ins, outs);
   auto runtime_ctx = PrepareRuntimeContext(ins, outs);
-  auto runtime_place = PreparedOp::GetExpectedPlace(place(), ins);
 
-  auto prepared_op =
-      PreparedOp::Prepare(runtime_ctx, *op_kernel, runtime_place);
+  VLOG(6) << "start preparing op: " << Type();
+  auto prepared_op = PreparedOp::Prepare(runtime_ctx, *op_kernel, place(), ins);
 
+  VLOG(6) << "finish preparing op: " << Type();
   prepared_op.Run();
 
   VLOG(4) << LayerDebugString(Type(), ins, outs);

paddle/fluid/imperative/prepared_operator.cc

@@ -28,22 +28,34 @@ const framework::Tensor* GetTensorFromVar(const framework::Variable& var) {
   }
 }
 
-platform::Place PreparedOp::GetExpectedPlace(const platform::Place& place,
-                                             const NameVarBaseMap& ins) {
-  bool found = false;
-  for (auto& name_pair : ins) {
-    for (auto& var_base : name_pair.second) {
+void PreparedOp::PrepareData(
+    const platform::Place& place, const NameVarBaseMap& ins,
+    const framework::OperatorWithKernel& op,
+    const framework::OpKernelType& expected_kernel_key) {
+  for (const auto& name_pair : ins) {
+    for (const auto& var_base : name_pair.second) {
       const auto* tensor = GetTensorFromVar(var_base->Var());
       if (tensor && tensor->IsInitialized()) {
         auto tmp_place = tensor->place();
-        PADDLE_ENFORCE_EQ(!found || tmp_place == place, true,
-                          "Input variable should keep in the same place: %s, "
-                          "but get place: %s of input %s instead",
-                          place, tmp_place, name_pair.first);
+        // TODO(jiabin): Support transform data layout when we Verify it on more
+        // tests
+        if (!(tmp_place == place)) {
+          auto kernel_type_for_var = op.GetKernelTypeForVar(
+              name_pair.first, *tensor, expected_kernel_key);
+          if (!NeedTransform(kernel_type_for_var, expected_kernel_key)) {
+            continue;
+          } else {
+            VLOG(3) << "Transform Variable " << var_base->Name() << " from "
+                    << kernel_type_for_var << " to " << expected_kernel_key;
+            framework::Tensor out;
+            TransformData(expected_kernel_key, kernel_type_for_var, *tensor,
+                          &out);
+            SetTensorToVariable(var_base->Var(), out, var_base->MutableVar());
+          }
+        }
       }
     }
   }
-  return place;
 }
 
 PreparedOp::PreparedOp(const framework::OperatorBase& op,
@@ -59,8 +71,10 @@ PreparedOp::PreparedOp(const framework::OperatorBase& op,
 
 PreparedOp PreparedOp::Prepare(const framework::RuntimeContext& ctx,
                                const framework::OperatorWithKernel& op,
-                               const platform::Place& place) {
-  auto* dev_ctx = platform::DeviceContextPool::Instance().Get(place);
+                               platform::Place place,
+                               const NameVarBaseMap& ins) {
+  platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
+  auto* dev_ctx = pool.Get(place);
 
   // check if op[type] has kernel registered.
   auto& all_op_kernels = op.AllOpKernels();
@@ -86,6 +100,13 @@ PreparedOp PreparedOp::Prepare(const framework::RuntimeContext& ctx,
   }
   std::vector<framework::KernelConfig>* kernel_configs =
       op.GetKernelConfig(expected_kernel_key);
+
+  if (!(expected_kernel_key.place_ == place)) {
+    dev_ctx = pool.Get(expected_kernel_key.place_);
+    place = dev_ctx->GetPlace();
+  }
+
+  PrepareData(place, ins, op, expected_kernel_key);
   return PreparedOp(op, ctx, kernel_iter->second, dev_ctx, kernel_configs);
 }
 
@@ -93,6 +114,7 @@ void PreparedOp::Run() {
   // TODO(zjl): remove scope in dygraph
   framework::Scope scope;
   op_.RuntimeInferShape(scope, dev_ctx_->GetPlace(), ctx_);
+  VLOG(6) << "Finish Runtime infer shape";
   func_(framework::ExecutionContext(op_, scope, *dev_ctx_, ctx_,
                                     kernel_configs_));
 }

paddle/fluid/imperative/prepared_operator.h

@@ -17,6 +17,8 @@
 #include <string>
 #include <utility>
 #include <vector>
+#include "paddle/fluid/framework/data_transform.h"
+#include "paddle/fluid/framework/op_kernel_type.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/imperative/layer.h"
 #include "paddle/fluid/imperative/type_defs.h"
@@ -30,14 +32,16 @@ class PreparedOp {
  public:
   static PreparedOp Prepare(const framework::RuntimeContext& ctx,
                             const framework::OperatorWithKernel& op,
-                            const platform::Place& place);
+                            platform::Place place, const NameVarBaseMap& ins);
 
   inline platform::DeviceContext* GetDeviceContext() const { return dev_ctx_; }
 
   void Run();
 
-  static platform::Place GetExpectedPlace(const platform::Place& place,
-                                          const NameVarBaseMap& ins);
+  static void PrepareData(const platform::Place& place,
+                          const NameVarBaseMap& ins,
+                          const framework::OperatorWithKernel& op,
+                          const framework::OpKernelType& expected_kernel_key);
 
  private:
   PreparedOp(const framework::OperatorBase& op,

paddle/fluid/imperative/tests/CMakeLists.txt

@@ -1,5 +1,5 @@
 cc_test(nccl_context_test SRCS nccl_context_test.cc DEPS nccl_context)
 cc_test(test_gradient_accmulator SRCS test_gradient_accmulator.cc DEPS gradient_accumulator memcpy)
-cc_test(test_layer SRCS test_layer.cc DEPS layer proto_desc operator op_registry variable_helper mul_op)
-cc_test(test_prepare_op SRCS test_prepare_op.cc DEPS prepared_operator op_info split_op layer concat_and_split)
-cc_test(test_tracer SRCS test_tracer.cc DEPS tracer layer proto_desc operator op_registry variable_helper mul_op)
+cc_test(test_layer SRCS test_layer.cc DEPS layer proto_desc operator op_registry variable_helper mul_op memcpy)
+cc_test(test_prepare_op SRCS test_prepare_op.cc DEPS prepared_operator op_info split_op layer concat_and_split assign_op place)
+cc_test(test_tracer SRCS test_tracer.cc DEPS tracer layer proto_desc operator op_registry variable_helper mul_op memcpy)

paddle/fluid/imperative/tests/test_prepare_op.cc

@@ -111,7 +111,7 @@ TEST(test_prepare_op, test_prepare_op) {
                                    split_attr_map);
   framework::RuntimeContext ctx = PrepareRuntimeContext(ins, outs);
   ASSERT_NO_FATAL_FAILURE(PreparedOp preparedOp =
-                              PreparedOp::Prepare(ctx, op, place));
+                              PreparedOp::Prepare(ctx, op, place, ins));
 }
 
 const framework::Tensor* GetTensorFromVar(const framework::Variable& var);
@@ -123,8 +123,94 @@ TEST(test_prepare_op, test_get_tensor_from_var) {
   auto* ts = GetTensorFromVar(*vout_error->MutableVar());
   ASSERT_TRUE(ts != nullptr);
 }
+#if defined(PADDLE_WITH_CUDA)
+TEST(test_prepare_op, test_prepare_data) {
+  std::shared_ptr<imperative::VarBase> vin(
+      new imperative::VarBase(false, "vin"));
+  std::shared_ptr<imperative::VarBase> vout(
+      new imperative::VarBase(false, "vout"));
+
+  framework::OpDesc desc;
+  platform::CPUPlace cpu_place;
+  platform::CUDAPlace gpu_place(0);
+  std::vector<float> src_data(10, 2.0);
+  std::vector<int64_t> dims = {2, 5};
+
+  // prepare an cpu only input
+  auto* vin_tensor = vin->MutableVar()->GetMutable<framework::LoDTensor>();
+  vin_tensor->Resize(framework::make_ddim(dims));
+  auto* vin_mutable_tensor = vin_tensor->mutable_data<float>(cpu_place);
+  paddle::memory::Copy(cpu_place, vin_mutable_tensor, cpu_place,
+                       src_data.data(), sizeof(float) * src_data.size());
 
+  var_pair x_pair = var_pair("X", vb_vector(1, vin));
+  var_pair out_pair = var_pair("Out", vb_vector(1, vout));
+  imperative::NameVarBaseMap ins = {x_pair};
+  imperative::NameVarBaseMap outs = {out_pair};
+  framework::AttributeMap assign_attr_map;
+  const auto& info = framework::OpInfoMap::Instance().Get("assign");
+  framework::VariableNameMap var_in_map =
+      CreateVarNameMap(info, "assign", ins, true);
+  framework::VariableNameMap var_out_map =
+      CreateVarNameMap(info, "assign", outs, false);
+  framework::OperatorWithKernel assign_op("assign", var_in_map, var_out_map,
+                                          assign_attr_map);
+  framework::RuntimeContext ctx = PrepareRuntimeContext(ins, outs);
+
+  // test if it can be transformed to GPU place
+  PreparedOp prepared_op = PreparedOp::Prepare(ctx, assign_op, gpu_place, ins);
+  for (const auto& name_pair : ins) {
+    for (const auto& vb : name_pair.second) {
+      ASSERT_TRUE(platform::is_same_place(
+          vb->Var().Get<framework::LoDTensor>().place(), gpu_place));
+    }
+  }
+}
+#endif
+
+TEST(test_prepare_op, test_prepare_data_same_place) {
+  std::shared_ptr<imperative::VarBase> vin(
+      new imperative::VarBase(false, "vin"));
+  std::shared_ptr<imperative::VarBase> vout(
+      new imperative::VarBase(false, "vout"));
+
+  framework::OpDesc desc;
+  platform::CPUPlace cpu_place;
+  std::vector<float> src_data(10, 2.0);
+  std::vector<int64_t> dims = {2, 5};
+
+  // prepare an cpu only input
+  auto* vin_tensor = vin->MutableVar()->GetMutable<framework::LoDTensor>();
+  vin_tensor->Resize(framework::make_ddim(dims));
+  auto* vin_mutable_tensor = vin_tensor->mutable_data<float>(cpu_place);
+  paddle::memory::Copy(cpu_place, vin_mutable_tensor, cpu_place,
+                       src_data.data(), sizeof(float) * src_data.size());
+
+  var_pair x_pair = var_pair("X", vb_vector(1, vin));
+  var_pair out_pair = var_pair("Out", vb_vector(1, vout));
+  imperative::NameVarBaseMap ins = {x_pair};
+  imperative::NameVarBaseMap outs = {out_pair};
+  framework::AttributeMap assign_attr_map;
+  const auto& info = framework::OpInfoMap::Instance().Get("assign");
+  framework::VariableNameMap var_in_map =
+      CreateVarNameMap(info, "assign", ins, true);
+  framework::VariableNameMap var_out_map =
+      CreateVarNameMap(info, "assign", outs, false);
+  framework::OperatorWithKernel assign_op("assign", var_in_map, var_out_map,
+                                          assign_attr_map);
+  framework::RuntimeContext ctx = PrepareRuntimeContext(ins, outs);
+
+  // test if it never transfered on GPU place
+  PreparedOp prepared_op = PreparedOp::Prepare(ctx, assign_op, cpu_place, ins);
+  for (const auto& name_pair : ins) {
+    for (const auto& vb : name_pair.second) {
+      ASSERT_TRUE(platform::is_same_place(
+          vb->Var().Get<framework::LoDTensor>().place(), cpu_place));
+    }
+  }
+}
 }  // namespace imperative
 }  // namespace paddle
 
 USE_OP(split);
+USE_OP(assign);

paddle/fluid/imperative/tests/test_tracer.cc

@@ -22,6 +22,7 @@
 #include <vector>
 #include "gtest/gtest.h"
 #include "paddle/fluid/imperative/tracer.h"
+#include "paddle/fluid/memory/memcpy.h"
 
 namespace imperative = paddle::imperative;
 namespace platform = paddle::platform;
@@ -142,6 +143,48 @@ TEST(test_tracer, test_track_backward_input) {
   mul_attr_map["use_mkldnn"] = false;
   ASSERT_ANY_THROW(tracer.TraceOp("mul", ins, outs, mul_attr_map, place, true));
 }
+#if defined(PADDLE_WITH_CUDA)
+TEST(test_tracer, test_trace_op_with_multi_device_inputs) {
+  // Doing an mul
+  imperative::Tracer tracer;
+  std::shared_ptr<imperative::VarBase> x_in(
+      new imperative::VarBase(true, "x_in"));
+  std::shared_ptr<imperative::VarBase> y_in(
+      new imperative::VarBase(true, "y_in"));
+  std::shared_ptr<imperative::VarBase> vout(
+      new imperative::VarBase(true, "vout"));
+  platform::CPUPlace place;
+  platform::CUDAPlace gpu_place(0);
+  std::vector<float> src_data(10, 2.0);
+  std::vector<int64_t> dims1 = {2, 5};
+  std::vector<int64_t> dims2 = {5, 2};
+
+  auto* x_in_tensor = x_in->MutableVar()->GetMutable<framework::LoDTensor>();
+  auto* y_in_tensor = y_in->MutableVar()->GetMutable<framework::LoDTensor>();
+  x_in_tensor->Resize(framework::make_ddim(dims1));
+  auto* mutable_x = x_in_tensor->mutable_data<float>(place);
+  paddle::memory::Copy(place, mutable_x, place, src_data.data(),
+                       sizeof(float) * src_data.size());
+  y_in_tensor->Resize(framework::make_ddim(dims2));
+  auto* mutable_y = y_in_tensor->mutable_data<float>(gpu_place);
+  paddle::memory::Copy(gpu_place, mutable_y, place, src_data.data(),
+                       sizeof(float) * src_data.size(), 0);
+  var_pair x_pair = var_pair("X", vb_vector(1, x_in));
+  var_pair y_pair = var_pair("Y", vb_vector(1, y_in));
+  var_pair out_pair = var_pair("Out", vb_vector(1, vout));
+  imperative::NameVarBaseMap ins = {x_pair, y_pair};
+  imperative::NameVarBaseMap outs = {out_pair};
+  framework::AttributeMap mul_attr_map;
+  mul_attr_map["use_mkldnn"] = false;
+  tracer.TraceOp("mul", ins, outs, mul_attr_map, gpu_place, true);
+  framework::LoDTensor rlt;
+  framework::TensorCopySync(vout->Var().Get<framework::LoDTensor>(), place,
+                            &rlt);
+  for (size_t i = 0; i < rlt.numel(); i++) {
+    ASSERT_EQ(rlt.data<float>()[i], 20.0);
+  }
+}
+#endif
 }  // namespace imperative
 }  // namespace paddle
 

paddle/fluid/imperative/tracer.cc

@@ -45,11 +45,13 @@ void Tracer::TraceOp(const std::string& type, const NameVarBaseMap& ins,
     TraceBackward(op, framework::OpDesc(op->Type(), op->InputNameMap(),
                                         op->OutputNameMap(), op->Attrs()),
                   ins, outs);
+    VLOG(6) << "Finish tracking Backward of op: " << type;
   }
+  VLOG(6) << "Finish tracing fwd op: " << type;
 }
 
 bool Tracer::ComputeRequiredGrad(const NameVarBaseMap& ins,
-                                 const NameVarBaseMap outs,
+                                 const NameVarBaseMap& outs,
                                  bool trace_backward) {
   // TODO(jiabin): Implement auto prune here
   return trace_backward;

paddle/fluid/imperative/tracer.h

@@ -40,8 +40,8 @@ class Tracer {
                const NameVarBaseMap& outs, framework::AttributeMap attrs,
                const platform::Place& place, bool trace_bacward);
 
-  bool ComputeRequiredGrad(const NameVarBaseMap& ins, const NameVarBaseMap outs,
-                           bool trace_backward);
+  bool ComputeRequiredGrad(const NameVarBaseMap& ins,
+                           const NameVarBaseMap& outs, bool trace_backward);
 
   void TraceBackward(const std::shared_ptr<OpBase>& fwd_op,
                      const framework::OpDesc& fwd_op_desc,

python/paddle/fluid/tests/unittests/test_layers.py

@@ -223,6 +223,13 @@ class TestLayer(LayerTest):
             conv2d = nn.Conv2D('conv2d', num_filters=3, filter_size=[2, 2])
             dy_ret = conv2d(base.to_variable(images))
 
+        with self.dynamic_graph():
+            images = np.ones([2, 3, 5, 5], dtype='float32')
+            conv2d = nn.Conv2D(
+                'conv2d', num_filters=3, filter_size=[2, 2], bias_attr=False)
+            dy_ret = conv2d(base.to_variable(images))
+            self.assertTrue(conv2d._bias_param is None)
+
         self.assertTrue(np.allclose(static_ret, dy_ret.numpy()))
         self.assertTrue(np.allclose(static_ret, static_ret2))
 
@@ -994,6 +1001,105 @@ class TestLayer(LayerTest):
 
         self.assertTrue(np.allclose(static_ret, dy_ret.numpy()))
 
+    def test_compare(self):
+        value_a = np.arange(3)
+        value_b = np.arange(3)
+        # less than
+        with self.static_graph():
+            a = layers.data(name='a', shape=[1], dtype='int64')
+            b = layers.data(name='b', shape=[1], dtype='int64')
+            cond = layers.less_than(x=a, y=b)
+            static_ret = self.get_static_graph_result(
+                feed={"a": value_a,
+                      "b": value_b}, fetch_list=[cond])[0]
+        with self.dynamic_graph():
+            da = base.to_variable(value_a)
+            db = base.to_variable(value_b)
+            dcond = layers.less_than(x=da, y=db)
+
+        for i in range(len(static_ret)):
+            self.assertTrue(dcond.numpy()[i] == static_ret[i])
+
+        # less equal
+        with self.static_graph():
+            a1 = layers.data(name='a1', shape=[1], dtype='int64')
+            b1 = layers.data(name='b1', shape=[1], dtype='int64')
+            cond1 = layers.less_equal(x=a1, y=b1)
+            static_ret1 = self.get_static_graph_result(
+                feed={"a1": value_a,
+                      "b1": value_b}, fetch_list=[cond1])[0]
+        with self.dynamic_graph():
+            da1 = base.to_variable(value_a)
+            db1 = base.to_variable(value_b)
+            dcond1 = layers.less_equal(x=da1, y=db1)
+
+            for i in range(len(static_ret1)):
+                self.assertTrue(dcond1.numpy()[i] == static_ret1[i])
+
+        #greater than
+        with self.static_graph():
+            a2 = layers.data(name='a2', shape=[1], dtype='int64')
+            b2 = layers.data(name='b2', shape=[1], dtype='int64')
+            cond2 = layers.greater_than(x=a2, y=b2)
+            static_ret2 = self.get_static_graph_result(
+                feed={"a2": value_a,
+                      "b2": value_b}, fetch_list=[cond2])[0]
+        with self.dynamic_graph():
+            da2 = base.to_variable(value_a)
+            db2 = base.to_variable(value_b)
+            dcond2 = layers.greater_than(x=da2, y=db2)
+
+            for i in range(len(static_ret2)):
+                self.assertTrue(dcond2.numpy()[i] == static_ret2[i])
+
+        #greater equal
+        with self.static_graph():
+            a3 = layers.data(name='a3', shape=[1], dtype='int64')
+            b3 = layers.data(name='b3', shape=[1], dtype='int64')
+            cond3 = layers.greater_equal(x=a3, y=b3)
+            static_ret3 = self.get_static_graph_result(
+                feed={"a3": value_a,
+                      "b3": value_b}, fetch_list=[cond3])[0]
+        with self.dynamic_graph():
+            da3 = base.to_variable(value_a)
+            db3 = base.to_variable(value_b)
+            dcond3 = layers.greater_equal(x=da3, y=db3)
+
+            for i in range(len(static_ret3)):
+                self.assertTrue(dcond3.numpy()[i] == static_ret3[i])
+
+        # equal
+        with self.static_graph():
+            a4 = layers.data(name='a4', shape=[1], dtype='int64')
+            b4 = layers.data(name='b4', shape=[1], dtype='int64')
+            cond4 = layers.equal(x=a4, y=b4)
+            static_ret4 = self.get_static_graph_result(
+                feed={"a4": value_a,
+                      "b4": value_b}, fetch_list=[cond4])[0]
+        with self.dynamic_graph():
+            da4 = base.to_variable(value_a)
+            db4 = base.to_variable(value_b)
+            dcond4 = layers.equal(x=da4, y=db4)
+
+            for i in range(len(static_ret4)):
+                self.assertTrue(dcond4.numpy()[i] == static_ret4[i])
+
+        # not equal
+        with self.static_graph():
+            a5 = layers.data(name='a5', shape=[1], dtype='int64')
+            b5 = layers.data(name='b5', shape=[1], dtype='int64')
+            cond5 = layers.equal(x=a5, y=b5)
+            static_ret5 = self.get_static_graph_result(
+                feed={"a5": value_a,
+                      "b5": value_b}, fetch_list=[cond5])[0]
+        with self.dynamic_graph():
+            da5 = base.to_variable(value_a)
+            db5 = base.to_variable(value_b)
+            dcond5 = layers.equal(x=da5, y=db5)
+
+            for i in range(len(static_ret5)):
+                self.assertTrue(dcond5.numpy()[i] == static_ret5[i])
+
 
 class TestBook(LayerTest):
     def test_all_layers(self):