【CUDA Kernel No.81、88、89、91、92】算子Kernel修复 (#2192)

* fix

* 修改路径错误

---------

Co-authored-by: Tao Luo <[email protected]>
Co-authored-by: Yuqiang Ge <[email protected]>

Author: 3 people

backends/iluvatar_gpu/kernels/cuda_kernels/moe_unpermute_kernel_register.cu

@@ -15,7 +15,7 @@
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/full_kernel.h"
 #include "paddle/phi/kernels/gpu/moe_permute_utils.h"
-#include "paddle/phi/kernels/gpu/moe_unpermute_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/gpu/moe_unpermute_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(moe_unpermute,
                           iluvatar_gpu,

backends/iluvatar_gpu/kernels/cuda_kernels/partial_allgather_kernel_register.cu

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/partial_allgather_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/gpu/partial_allgather_kernel.h"
 #include "paddle/phi/kernels/partial_concat_kernel.h"
 
 #if (NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000) || \

backends/iluvatar_gpu/kernels/cuda_kernels/partial_concat_grad_kernel_register.cu

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/partial_concat_grad_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/gpu/partial_concat_grad_kernel.h"
 PD_CUSTOM_KERNEL_REGISTER(partial_concat_grad,
                           iluvatar_gpu,
                           ALL_LAYOUT,

backends/iluvatar_gpu/kernels/cuda_kernels/partial_recv_kernel_register.cu

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/partial_recv_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/partial_recv_kernel.h"
 #if (NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000) || \
     defined(PADDLE_WITH_HIP)
 PD_CUSTOM_KERNEL_REGISTER(partial_recv,

backends/iluvatar_gpu/kernels/cuda_kernels/partial_send_kernel_register.cu

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/partial_send_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/gpu/partial_send_kernel.h"
 #if (NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000) || \
     defined(PADDLE_WITH_HIP)
 PD_CUSTOM_KERNEL_REGISTER(partial_send,

backends/metax_gpu/CMakeLists.txt

@@ -124,6 +124,11 @@ file(
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/funcs/math/*.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/funcs/eigen/*.cu
   # kernels/gpu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/partial_send_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/partial_recv_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/partial_concat_grad_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/partial_allgather_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/ctc_align_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/yolo_box_head_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/stft_grad_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/soft_relu_grad_kernel.cu

backends/metax_gpu/kernels/cuda_kernels/moe_unpermute_kernel_register.cu

@@ -15,7 +15,7 @@
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/full_kernel.h"
 #include "paddle/phi/kernels/gpu/moe_permute_utils.h"
-#include "paddle/phi/kernels/gpu/moe_unpermute_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/gpu/moe_unpermute_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(moe_unpermute,
                           metax_gpu,

backends/metax_gpu/kernels/cuda_kernels/partial_allgather_kernel_register.cu

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/partial_allgather_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/gpu/partial_allgather_kernel.h"
 #include "paddle/phi/kernels/partial_concat_kernel.h"
 
 #if (NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000) || \

backends/metax_gpu/kernels/cuda_kernels/partial_concat_grad_kernel_register.cu

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/partial_concat_grad_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/gpu/partial_concat_grad_kernel.h"
 PD_CUSTOM_KERNEL_REGISTER(partial_concat_grad,
                           metax_gpu,
                           ALL_LAYOUT,

backends/metax_gpu/kernels/cuda_kernels/partial_recv_kernel_register.cu

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/partial_recv_kernel.cu"  //NOLINT
+#include "paddle/phi/kernels/partial_recv_kernel.h"
 #if (NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000) || \
     defined(PADDLE_WITH_HIP)
 PD_CUSTOM_KERNEL_REGISTER(partial_recv,