modify cmake file for cuda11.8 compile (#49020)

* modify cmake file for cuda11.8 compile

* add op_library(fused_embedding_eltwise_layernorm_op DEPS bert_encoder_functor)

cmake/third_party.cmake

@@ -317,8 +317,9 @@ if(WITH_ONNXRUNTIME)
 endif()
 
 if(WITH_GPU)
-  if(${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0 OR ${CMAKE_CUDA_COMPILER_VERSION}
-                                                 GREATER_EQUAL 11.6)
+  if(${CMAKE_CUDA_COMPILER_VERSION} LESS 11.0
+     OR (${CMAKE_CUDA_COMPILER_VERSION} GREATER_EQUAL 11.6
+         AND ${CMAKE_CUDA_COMPILER_VERSION} LESS 11.8))
     include(external/cub) # download cub
     list(APPEND third_party_deps extern_cub)
   endif()

paddle/fluid/operators/fused/CMakeLists.txt

@@ -69,7 +69,7 @@ if(WITH_GPU OR WITH_ROCM)
   op_library(skip_layernorm_op)
   op_library(yolo_box_head_op)
   op_library(yolo_box_post_op)
-  op_library(fused_embedding_eltwise_layernorm_op)
+  op_library(fused_embedding_eltwise_layernorm_op DEPS bert_encoder_functor)
   op_library(fused_gate_attention_op)
   # fusion_group
   if(NOT APPLE AND NOT WIN32)