diff --git a/paddle/fluid/framework/new_executor/new_ir_interpreter.cc b/paddle/fluid/framework/new_executor/new_ir_interpreter.cc index 9587511af77284602211aca6caff35ecfb631b2a..28752ef1f7eb1a643935e76afab0a7fc2e7c6d26 100644 --- a/paddle/fluid/framework/new_executor/new_ir_interpreter.cc +++ b/paddle/fluid/framework/new_executor/new_ir_interpreter.cc @@ -346,9 +346,6 @@ void NewIRInterpreter::UpdateSyncOpNum() { void NewIRInterpreter::UpdateNcclOpNum() { static std::set nccl_op_set = { - "pd.sync_batch_norm_", - "pd.sync_batch_norm", - "pd.sync_batch_norm_grad", "pd.c_softmax_with_cross_entropy", "pd.c_allgather", "pd.c_allreduce_max", @@ -360,6 +357,7 @@ void NewIRInterpreter::UpdateNcclOpNum() { "pd.c_reduce_prod", "pd.c_reducescatter", "pd.c_broadcast", + "pd.c_broadcast_", "pd.c_scatter", "pd.partial_send", "pd.partial_recv", @@ -370,11 +368,12 @@ void NewIRInterpreter::UpdateNcclOpNum() { "pd.barrier", "pd.alltoall", "pd.global_gather", - "pd.distributed_fused_lamb_op", - "pd.margin_cross_entropy_op", - "pd.sync_batch_norm_op", - "pd.data_norm_op", - "pd.class_center_sample_op", + "pd.distributed_fused_lamb", + "pd.margin_cross_entropy", + "pd.sync_batch_norm", + "pd.sync_batch_norm_", + "pd.data_norm", + "pd.class_center_sample", "pd.all_to_all", "pd.dist_concat", "pd.all_gather", @@ -405,11 +404,12 @@ void NewIRInterpreter::UpdateNcclOpNum() { "pd.barrier_grad", "pd.alltoall_grad", "pd.global_gather_grad", - "pd.distributed_fused_lamb_op_grad", - "pd.margin_cross_entropy_op_grad", - "pd.sync_batch_norm_op_grad", - "pd.data_norm_op_grad", - "pd.class_center_sample_op_grad", + "pd.distributed_fused_lamb_grad", + "pd.margin_cross_entropy_grad", + "pd.margin_cross_entropy_grad_" + "pd.sync_batch_norm_grad", + "pd.data_norm_grad", + "pd.class_center_sample_grad", "pd.all_to_all_grad", "pd.dist_concat_grad", "pd.all_gather_grad",