masahi opened a new pull request, #13341: URL: https://github.com/apache/tvm/pull/13341
Fixes the issue reported in https://discuss.tvm.apache.org/t/model-compilation-fails-on-gpu-target/13866 For certain `conv2d_transpose` workloads, we get funny allocations like ``` allocate data_pad[float32 * ((floordiv(((blockIdx.y*2) + 1), 3)*4608) + 576)], storage_scope = shared allocate p1.shared[float32 * ((floordiv(((blockIdx.y*2) + 1), 3)*12288) + 1536)], storage_scope = shared ``` This is apparently due to trying to split a small channel by a large factor. cc @vinx13 -- 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]
