remove trt_skip_layernorm_fuse_pass from gpu passes (#45293)

paddle/fluid/framework/ir/CMakeLists.txt

@@ -167,6 +167,7 @@ if(WITH_TENSORRT)
   pass_library(trt_map_matmul_to_mul_pass inference)
   pass_library(trt_embedding_eltwise_layernorm_fuse_pass inference)
   pass_library(trt_multihead_matmul_fuse_pass inference)
+  pass_library(trt_skip_layernorm_fuse_pass inference)
   pass_library(preln_embedding_eltwise_layernorm_fuse_pass inference)
   pass_library(preln_skip_layernorm_fuse_pass inference)
   pass_library(set_transformer_input_convert_pass inference)
@@ -177,7 +178,6 @@ endif()
 if(WITH_GPU OR WITH_ROCM)
   pass_library(cudnn_placement_pass base DEPS placement_pass_base)
   pass_library(embedding_eltwise_layernorm_fuse_pass inference)
-  pass_library(trt_skip_layernorm_fuse_pass inference)
 endif()
 
 if(WITH_MKLDNN)

paddle/fluid/inference/api/paddle_pass_builder.cc

@@ -167,7 +167,6 @@ const std::vector<std::string> kGpuLowerPrecisionPasses{
     "fc_fuse_pass",
     "fc_elementwise_layernorm_fuse_pass",
     "embedding_eltwise_layernorm_fuse_pass",
-    "trt_skip_layernorm_fuse_pass",
     "runtime_context_cache_pass",
 };
 
@@ -192,7 +191,6 @@ GpuPassStrategy::GpuPassStrategy() : PassStrategy({}) {
         "conv_bn_fuse_pass",                      //
         "conv_eltwiseadd_bn_fuse_pass",           //
         "embedding_eltwise_layernorm_fuse_pass",  //
-        "trt_skip_layernorm_fuse_pass",           //
         "multihead_matmul_fuse_pass_v2",          //
         "gpu_cpu_squeeze2_matmul_fuse_pass",      //
         "gpu_cpu_reshape2_matmul_fuse_pass",      //