My proposal is now implemented.
I ended up completely replacing the content of graph_plan_memory.cc with a python implementation: - Redirect to Python: https://github.com/tum-ei-eda/tvm/blob/e9184d948edd58635e79c3f21355f2b83b361401/src/relay/backend/graph_plan_memory.cc#L890 - Main implementation: https://github.com/tum-ei-eda/tvm/blob/tumeda_memplan/python/tvm/relay/memplan.py - Support code: https://github.com/tum-ei-eda/tvm/tree/tumeda_memplan/python/tvm/relay/memopt To work with the data-flow graph more conveniently this adds `networkx` as dependency. For the optimal solution, this would drag in the optimizer `OR-Tools` as well. The code is currently also pretty filled with debug functionality that might not be wanted by TVM. How to contribute this back? - I have started a PR to add the possibility to return offsets for buffers: https://github.com/apache/tvm/pull/8134 - Suggestion: Add the TFLM inspired heuristic as default and leave the optimized version as external extension that plugs into an interface? That would drop a lot of code and could be implemented without pretty much anything in the memopt folder. - How should the interface to a memory planning implementation look? How about a registered function that is called only if it exists? Are there any similar _optional extension_ interfaces? - Any other suggestions? First results are shown below. This is evaluated on RISC-V and using the [code generator](https://github.com/tum-ei-eda/utvm_staticrt_codegen/). There still might be some issue, because I would not expect the RAM usage to rise in any example. But looks promising so far with 10% reduction for a cifar10 model and 18% for resnet!  --- [Visit Topic](https://discuss.tvm.apache.org/t/discussion-alignment-memory-planning/9730/6) to respond. You are receiving this because you enabled mailing list mode. To unsubscribe from these emails, [click here](https://discuss.tvm.apache.org/email/unsubscribe/ee3839259e6f0cc652c7af0922fbe09a6e74837bce826f633d12f7d2f97860fc).
