提交 3c9cea59 编写于 作者: Y Yu Yang

Add more log

上级 f8f1a963
...@@ -365,6 +365,7 @@ struct NCCLAllReduceOpHandle : public OpHandle { ...@@ -365,6 +365,7 @@ struct NCCLAllReduceOpHandle : public OpHandle {
auto &p = static_cast<VarHandle *>(in)->place_; auto &p = static_cast<VarHandle *>(in)->place_;
in->generated_op_->Wait(dev_ctx_[p]); in->generated_op_->Wait(dev_ctx_[p]);
} }
VLOG(3) << "Before NCCL";
PADDLE_ENFORCE(cudaDeviceSynchronize()); PADDLE_ENFORCE(cudaDeviceSynchronize());
auto &var_name = static_cast<VarHandle *>(this->inputs_[0])->name_; auto &var_name = static_cast<VarHandle *>(this->inputs_[0])->name_;
...@@ -394,8 +395,9 @@ struct NCCLAllReduceOpHandle : public OpHandle { ...@@ -394,8 +395,9 @@ struct NCCLAllReduceOpHandle : public OpHandle {
nccl_ctx.comm, nccl_ctx.stream()); nccl_ctx.comm, nccl_ctx.stream());
} }
platform::dynload::ncclGroupEnd(); platform::dynload::ncclGroupEnd();
PADDLE_ENFORCE(cudaDeviceSynchronize()); PADDLE_ENFORCE(cudaDeviceSynchronize());
VLOG(3) << "After NCCL";
} }
} }
}; };
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册