adstraw commented on code in PR #12411: URL: https://github.com/apache/tvm/pull/12411#discussion_r955267553
########## tests/cpp-runtime/hexagon/ring_buffer_tests.cc: ########## @@ -0,0 +1,109 @@ +/* + * 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. + */ + Review Comment: I would change your example slightly to make the descriptor IDs zero based. This is how it works in the code. The first descriptor has ID=0 and uses ring buffer slot 0. The second has ID=1 and uses slot 1. When ID is equivalent to ring buffer size we reuse ring buffer slot 0. And so on. ``` Ringbuffer[size=4] Initial state: desc0@0[finished] , desc1@1[finished] , desc2@2[finished] , desc3@3[finished] New DMA needed: desc4@0[inflight] , desc1@1[finished] , desc2@2[finished] , desc3@3[finished] ``` The case you have described above should work OK. I believe we can cover it with `RingBuffer` unit tests. I believe existing tests already hit this case, to some degree. I will add a test case to explicitly cover this scenario. When we go to add `desc4` we first check if the ring buffer is full by checking that `InFlight` count (0) is less than the size of the ring buffer (4). All good. Then, we `GetAddr` of the "next" ID = 4 which reuses ring buffer slot 0. Then, we increment the "next" ID to 5. Then we return a pointer to ring buffer slot 0. There is one issue with this flow that I thought of ... the DMA descriptor at ring buffer slot 0 is "finished" and we simply return a pointer to it and expect that the user will add an "inflight" DMA descriptor at that location. If the user were to call `InFlight` before adding the "inflight" descriptor then ring buffer slot 0 would be treated as "finished" from the perspective of the `RingBuffer` class where it is still "inflight" from the perspective of the user. I had initially coded an `Add` function which took a function pointer which would add a T to the RingBuffer rather that `Next` to get around this issue. It felt to complicated but it does expose a gap. -- 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: commits-unsubscr...@tvm.apache.org For queries about this service, please contact Infrastructure at: us...@infra.apache.org
