cconvey opened a new pull request, #11163: URL: https://github.com/apache/tvm/pull/11163
Re-enable a benchmark configuration that was failing because of a bug in TVM's new dimension-mapping code. Here's the new output: ``` CONFIGURATION: dtype:int8-schedtype:2-memscope:global.vtcm-numvecs:2048 dtype sched_type mem_scope num_vecs_per_tensor status median(µsec) min(µsec) max(µsec) comment int8 1 global 1 OK 0.7 0.7 0.7 int8 1 global 16 OK 0.5 0.5 0.5 int8 1 global 64 OK 1.2 1.2 1.2 int8 1 global 512 OK 3.0 3.0 3.0 int8 1 global 2048 OK 16.5 16.5 16.5 int8 1 global.vtcm 1 OK 0.7 0.7 0.7 int8 1 global.vtcm 16 OK 0.8 0.8 0.8 int8 1 global.vtcm 64 OK 1.4 1.4 1.4 int8 1 global.vtcm 512 OK 5.4 5.4 5.4 int8 1 global.vtcm 2048 SKIP Expect to exceed VTCM budget. int8 2 global 1 OK 0.6 0.6 0.6 int8 2 global 16 OK 0.6 0.6 0.6 int8 2 global 64 OK 0.8 0.8 0.8 int8 2 global 512 OK 2.5 2.5 2.5 int8 2 global 2048 OK 17.2 17.2 17.2 int8 2 global.vtcm 1 OK 0.8 0.8 0.8 int8 2 global.vtcm 16 OK 0.7 0.7 0.7 int8 2 global.vtcm 64 OK 1.3 1.3 1.3 int8 2 global.vtcm 512 OK 5.3 5.3 5.3 int8 2 global.vtcm 2048 SKIP Expect to exceed VTCM budget. ``` -- 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]
