[PHI Decoupling] move sequence_scale from fluid to phi

paddle/fluid/operators/math/CMakeLists.txt

@@ -30,7 +30,6 @@ math_library(sampler DEPS generator)
 
 math_library(sequence_padding)
 math_library(sequence_pooling DEPS math_function jit_kernel_helper)
-math_library(sequence_scale)
 if(WITH_ASCEND_CL)
   math_library(beam_search DEPS math_function beam_search_npu)
 elseif(WITH_XPU)

paddle/phi/kernels/funcs/CMakeLists.txt

@@ -22,6 +22,7 @@ math_library(vol2col)
 math_library(softmax DEPS math_function)
 math_library(maxouting)
 math_library(matrix_bit_code)
+math_library(sequence_scale)
 
 cc_library(
   phi_data_layout_transform

paddle/fluid/operators/math/sequence_scale.cc → paddle/phi/kernels/funcs/sequence_scale.cc

@@ -12,17 +12,15 @@ 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/math/sequence_scale.h"
-
+#include "paddle/phi/kernels/funcs/sequence_scale.h"
 #include "paddle/phi/backends/cpu/cpu_context.h"
 
 namespace phi {
 class DenseTensor;
 }  // namespace phi
 
-namespace paddle {
-namespace operators {
-namespace math {
+namespace phi {
+namespace funcs {
 
 template <typename T>
 class ScaleLoDTensorFunctor<phi::CPUContext, T> {
@@ -34,9 +32,9 @@ class ScaleLoDTensorFunctor<phi::CPUContext, T> {
     auto lod = seq->lod();
     const size_t num_seq = lod[level].size() - 1;
     size_t seq_width = seq->dims()[1];
-    framework::LoD abs_offset_lod = framework::ToAbsOffset(lod);
+    paddle::framework::LoD abs_offset_lod = paddle::framework::ToAbsOffset(lod);
 
-    T* seq_data = seq->mutable_data<T>(context.GetPlace());
+    T* seq_data = context.template Alloc<T>(seq);
     for (size_t i = 0; i < num_seq; ++i) {
       for (size_t j = lod[level][i] * seq_width;
            j < lod[level][i + 1] * seq_width;
@@ -50,6 +48,5 @@ class ScaleLoDTensorFunctor<phi::CPUContext, T> {
 template class ScaleLoDTensorFunctor<phi::CPUContext, float>;
 template class ScaleLoDTensorFunctor<phi::CPUContext, double>;
 
-}  // namespace math
-}  // namespace operators
-}  // namespace paddle
+}  // namespace funcs
+}  // namespace phi

paddle/fluid/operators/math/sequence_scale.cu → paddle/phi/kernels/funcs/sequence_scale.cu

@@ -12,13 +12,12 @@ 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/math/sequence_scale.h"
+#include "paddle/phi/kernels/funcs/sequence_scale.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/backends/gpu/gpu_primitives.h"
 
-namespace paddle {
-namespace operators {
-namespace math {
+namespace phi {
+namespace funcs {
 
 using phi::PADDLE_CUDA_NUM_THREADS;
 
@@ -45,8 +44,8 @@ class ScaleLoDTensorFunctor<phi::GPUContext, T> {
     auto lod = seq->lod();
     const size_t num_seq = lod[level].size() - 1;
     const size_t seq_width = seq->numel() / seq->dims()[0];
-    auto abs_offset_lod = framework::ToAbsOffset(lod);
-    T* seq_data = seq->mutable_data<T>(context.GetPlace());
+    auto abs_offset_lod = paddle::framework::ToAbsOffset(lod);
+    T* seq_data = context.template Alloc<T>(seq);
     paddle::framework::MixVector<size_t> mix_vector(&(abs_offset_lod[level]));
 
 #ifdef PADDLE_WITH_HIP
@@ -75,6 +74,5 @@ class ScaleLoDTensorFunctor<phi::GPUContext, T> {
 template class ScaleLoDTensorFunctor<phi::GPUContext, float>;
 template class ScaleLoDTensorFunctor<phi::GPUContext, double>;
 
-}  // namespace math
-}  // namespace operators
-}  // namespace paddle
+}  // namespace funcs
+}  // namespace phi

paddle/fluid/operators/math/sequence_scale.h → paddle/phi/kernels/funcs/sequence_scale.h

@@ -17,9 +17,8 @@ limitations under the License. */
 #include "paddle/fluid/framework/lod_tensor.h"
 #include "paddle/fluid/platform/device_context.h"
 
-namespace paddle {
-namespace operators {
-namespace math {
+namespace phi {
+namespace funcs {
 
 /*
  * \brief   Scale a sequence.
@@ -45,6 +44,7 @@ namespace math {
  * \param num_seq       Number of sequence
  *
  */
+
 template <typename DeviceContext, typename T>
 class ScaleLoDTensorFunctor {
  public:
@@ -53,6 +53,5 @@ class ScaleLoDTensorFunctor {
                   phi::DenseTensor* seq);
 };
 
-}  // namespace math
-}  // namespace operators
-}  // namespace paddle
+}  // namespace funcs
+}  // namespace phi

paddle/phi/kernels/impl/warpctc_grad_kernel_impl.h

@@ -17,12 +17,12 @@
 #include <vector>
 
 #include "paddle/fluid/operators/math/sequence_padding.h"
-#include "paddle/fluid/operators/math/sequence_scale.h"
 #include "paddle/phi/backends/dynload/warpctc.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
+#include "paddle/phi/kernels/funcs/sequence_scale.h"
 #include "paddle/utils/optional.h"
 
 namespace phi {
@@ -79,7 +79,7 @@ void WarpctcGradKernel(const Context& dev_ctx,
         paddle::operators::math::kLengthBatchWidth);
 
     const T* loss_grad_data = loss_grad.data<T>();
-    paddle::operators::math::ScaleLoDTensorFunctor<Context, T>()(
+    phi::funcs::ScaleLoDTensorFunctor<Context, T>()(
         dev_ctx, loss_grad_data, logits_grad);
   }
 }

paddle/phi/kernels/impl/warpctc_kernel_impl.h

@@ -17,12 +17,12 @@
 #include <vector>
 
 #include "paddle/fluid/operators/math/sequence_padding.h"
-#include "paddle/fluid/operators/math/sequence_scale.h"
 #include "paddle/phi/backends/dynload/warpctc.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/empty_kernel.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
+#include "paddle/phi/kernels/funcs/sequence_scale.h"
 #include "paddle/utils/optional.h"
 
 namespace phi {