matteosal commented on issue #20675: URL: https://github.com/apache/incubator-mxnet/issues/20675#issuecomment-1014641982
@leezu it was 2 independent slowdowns one on top of the other: for the case of this toy model, using VC2017 introduced a ~4.5x slowdown (55/12) while INT64 was around 2.5x (140/55) on my machine. As for the cmake changes, I think the combination `(NOT USE_ONEDNN) AND USE_INT64_TENSOR_SIZE` should produce a warning instead of a failure because that's a genuine combination that someone might still want to use for some reason (e.g. in an environment that requires MKL ILP64), even if slower. VC2017 on the other hand might fail explicitly because the compiler version is generally not a constraint. -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
