Move dropout to phi (#40148)

* move dropout to phi; test=develop

* fix xpu, npu compile error; test=develop

paddle/fluid/inference/tensorrt/convert/dropout_op.cc

@@ -89,5 +89,5 @@ class DropoutOpConverter : public OpConverter {
 }  // namespace inference
 }  // namespace paddle
 
-USE_OP(dropout);
+USE_OP_ITSELF(dropout);
 REGISTER_TRT_OP_CONVERTER(dropout, DropoutOpConverter);

paddle/fluid/inference/tensorrt/convert/test_dropout_op.cc

@@ -57,4 +57,4 @@ TEST(DropoutOpConverter, main) {
 }  // namespace inference
 }  // namespace paddle
 
-USE_OP(dropout);
+USE_OP_ITSELF(dropout);

paddle/fluid/operators/assign_op_npu_test.cc

@@ -23,7 +23,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/collective/c_allgather_op_npu_test.cc

@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/collective/c_allreduce_max_op_npu_test.cc

@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/collective/c_allreduce_sum_op_npu_test.cc

@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/collective/c_broadcast_op_npu_test.cc

@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/collective/c_reduce_sum_op_npu_test.cc

@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/collective/c_reducescatter_op_npu_test.cc

@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/collective/c_sync_comm_stream_op_npu_test.cc

@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/collective/checknumeric_npu_test.cc

@@ -27,7 +27,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/collective/recv_v2_op_npu_test.cc

@@ -26,7 +26,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/collective/send_v2_op_npu_test.cc

@@ -25,7 +25,6 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/program_desc.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/string/printf.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 

paddle/fluid/operators/dropout_impl.cu.h

@@ -32,10 +32,9 @@ limitations under the License. */
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/fluid/operators/amp/fp16_type_traits.h"
 #include "paddle/fluid/operators/dropout_impl_util.h"
-#include "paddle/fluid/operators/dropout_op.h"
 #include "paddle/fluid/operators/elementwise/elementwise_op_impl.cu.h"
-#include "paddle/fluid/platform/device/gpu/gpu_launch_config.h"
-#include "paddle/phi/kernels/funcs/aligned_vector.h"
+#include "paddle/fluid/platform/aligned_vector.h"
+#include "paddle/phi/backends/gpu/gpu_launch_config.h"
 #include "paddle/phi/kernels/funcs/functors.h"
 
 namespace paddle {
@@ -177,12 +176,13 @@ __global__ void DropoutGradCUDAKernel(
 }
 
 template <typename T>
-void DropoutFwGPUKernelDriver(const platform::CUDADeviceContext& dev_ctx,
-                              bool is_test,
+void DropoutFwGPUKernelDriver(const phi::GPUContext& dev_ctx, bool is_test,
                               const std::string dropout_implementation,
                               float dropout_prob, bool upscale_in_train,
-                              bool is_fix_seed, int seed_val, const Tensor& x,
-                              const Tensor* seed, Tensor* mask, Tensor* y) {
+                              bool is_fix_seed, int seed_val,
+                              const framework::Tensor& x,
+                              const framework::Tensor* seed,
+                              framework::Tensor* mask, framework::Tensor* y) {
   auto& place = *dev_ctx.eigen_device();
   int64_t x_numel = x.numel();
   auto stream = dev_ctx.stream();
@@ -220,7 +220,8 @@ void DropoutFwGPUKernelDriver(const platform::CUDADeviceContext& dev_ctx,
     // VectorizedRandomGenerator use curand_uniform4, so we only support
     // vec_size is 4;
     int vec_size = (phi::GetVectorizedSize<T>(x_data) == 4) ? 4 : 1;
-    auto gpu_config = GetGpuLaunchConfig1D(dev_ctx, x_numel, vec_size);
+    auto gpu_config =
+        phi::backends::gpu::GetGpuLaunchConfig1D(dev_ctx, x_numel, vec_size);
     auto offset =
         ((x_numel - 1) / (gpu_config.GetThreadNum() * vec_size) + 1) * vec_size;
 
@@ -278,11 +279,13 @@ void DropoutFwGPUKernelDriver(const platform::CUDADeviceContext& dev_ctx,
 }
 
 template <typename T>
-void DropoutGradGPUKernelDriver(const platform::CUDADeviceContext& dev_ctx,
+void DropoutGradGPUKernelDriver(const phi::GPUContext& dev_ctx,
                                 const std::string dropout_implementation,
-                                float dropout_prob, const Tensor& grad_y,
-                                const Tensor& mask, int64_t size,
-                                Tensor* grad_x, bool is_test = false) {
+                                float dropout_prob,
+                                const framework::Tensor& grad_y,
+                                const framework::Tensor& mask, int64_t size,
+                                framework::Tensor* grad_x,
+                                bool is_test = false) {
   using MT = typename details::MPTypeTrait<T>::Type;
   auto stream = dev_ctx.stream();
   MT factor;

paddle/fluid/operators/dropout_impl_util.h

@@ -20,7 +20,7 @@ limitations under the License. */
 namespace paddle {
 namespace operators {
 
-inline void GetSeedDataAndIncrement(const platform::CUDADeviceContext& dev_ctx,
+inline void GetSeedDataAndIncrement(const phi::GPUContext& dev_ctx,
                                     const framework::Tensor* seed,
                                     const bool is_fix_seed, const int seed_val,
                                     const int offset, uint64_t* seed_data,

paddle/fluid/operators/dropout_op.cc

@@ -12,9 +12,9 @@ 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/fluid/operators/dropout_op.h"
 #include <memory>
 #include <string>
+#include "paddle/fluid/framework/op_registry.h"
 
 namespace paddle {
 namespace operators {
@@ -177,14 +177,3 @@ REGISTER_OPERATOR(dropout, ops::DropoutOp, ops::DropoutOpMaker,
                   ops::DropoutGradOpMaker<paddle::framework::OpDesc>,
                   ops::DropoutGradOpMaker<paddle::imperative::OpBase>);
 REGISTER_OPERATOR(dropout_grad, ops::DropoutOpGrad);
-REGISTER_OP_CPU_KERNEL(
-    dropout, ops::CPUDropoutKernel<paddle::platform::CPUDeviceContext, float>,
-    ops::CPUDropoutKernel<paddle::platform::CPUDeviceContext, double>,
-    ops::CPUDropoutKernel<paddle::platform::CPUDeviceContext,
-                          paddle::platform::bfloat16>);
-REGISTER_OP_CPU_KERNEL(
-    dropout_grad,
-    ops::DropoutGradKernel<paddle::platform::CPUDeviceContext, float>,
-    ops::DropoutGradKernel<paddle::platform::CPUDeviceContext, double>,
-    ops::DropoutGradKernel<paddle::platform::CPUDeviceContext,
-                           paddle::platform::bfloat16>);