fix include for multiclass_nms3_kernel (PaddlePaddle#2220)

Author: fsylmxx

backends/iluvatar_gpu/CMakeLists.txt

@@ -357,6 +357,7 @@ file(
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/moe_unpermute_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/momentum_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/mp_allreduce_sum_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/multiclass_nms3_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/multiplex_grad_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/nonzero_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/pad3d_kernel.cu

backends/iluvatar_gpu/kernels/cuda_kernels/multiclass_nms3_kernel_register.cu

@@ -13,7 +13,6 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/multiclass_nms3_kernel.cu"  //NOLINT
 #include "paddle/phi/kernels/multiclass_nms3_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(multiclass_nms3,

backends/metax_gpu/CMakeLists.txt

@@ -252,6 +252,7 @@ file(
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/mean_all_grad_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/multiplex_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/multiplex_grad_kernel.cu
+  ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/multiclass_nms3_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/mp_allreduce_sum_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/pow2_decay_with_linear_warmup_kernel.cu
   ${PADDLE_SOURCE_DIR}/paddle/phi/kernels/gpu/sigmoid_cross_entropy_with_logits_grad_kernel.cu

backends/metax_gpu/kernels/cuda_kernels/multiclass_nms3_kernel_register.cu

@@ -13,7 +13,6 @@
 // limitations under the License.
 
 #include "paddle/phi/core/kernel_registry.h"
-#include "paddle/phi/kernels/gpu/multiclass_nms3_kernel.cu"  //NOLINT
 #include "paddle/phi/kernels/multiclass_nms3_kernel.h"
 
 PD_CUSTOM_KERNEL_REGISTER(multiclass_nms3,