comaniac commented on a change in pull request #14: URL: https://github.com/apache/tvm-rfcs/pull/14#discussion_r683728888
########## File path: rfcs/0012-pipeline-executor.md ########## @@ -0,0 +1,367 @@ +<!--- Licensed to the Apache Software Foundation (ASF) under one --> +<!--- or more contributor license agreements. See the NOTICE file --> +<!--- distributed with this work for additional information --> +<!--- regarding copyright ownership. The ASF licenses this file --> +<!--- to you under the Apache License, Version 2.0 (the --> +<!--- "License"); you may not use this file except in compliance --> +<!--- with the License. You may obtain a copy of the License at --> + +<!--- http://www.apache.org/licenses/LICENSE-2.0 --> + +<!--- Unless required by applicable law or agreed to in writing, --> +<!--- software distributed under the License is distributed on an --> +<!--- "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY --> +<!--- KIND, either express or implied. See the License for the --> +<!--- specific language governing permissions and limitations --> +<!--- under the License. --> +- Feature Name: (fill me in with a unique identifier, `my_awesome_feature`) +- Start Date: (fill me in with today's date, YYYY-MM-DD) +- RFC PR: [apache/tvm-rfcs#0014](https://github.com/apache/tvm-rfcs/pull/0014) +- GitHub Issue: [apache/tvm#8596](https://github.com/apache/tvm/issues/8596) + +## 1. Summary + + +This proposal introduces Pipeline Executor: A runtime executor that by scheduling +splitted subgraph of relay graph in pipeline to implement task level parallism to +reduce compute latency. Review comment: I actually don't think it's a typical case only in ML. This concept is also widely used in HPC. The "latency" is based on the time of executing one task; while the "throughput" is the number of finished task in a time unit. Pipelining definitely cannot improve latency (of executing one task) but throughput, because you can finish more tasks in the same time period. General speaking, I think this is just the terminology issue, but it would be good to use consistent terminology for efficient communications. The confusion of "not being processed together" falls into the similar reason. Jorn and probably most people may feel confuse about why two stages of the same task can be processed together, but you actually meant stage 2 of the _first task_ and the stage 1 of the _second task_ can be processed in parallel. This is the point you should make it clear in the RFC. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
