mozga-intel commented on a change in pull request #20533:
URL: https://github.com/apache/incubator-mxnet/pull/20533#discussion_r694551661
##########
File path: src/operator/nn/mkldnn/mkldnn_base-inl.h
##########
@@ -305,17 +305,28 @@ inline static mkldnn::memory::desc GetMemDesc(const
NDArray& arr, int dtype = -1
return mkldnn::memory::desc{dims, get_mkldnn_type(dtype),
mkldnn::memory::format_tag::any};
}
-inline static mkldnn::memory::desc GetFCWeightDesc(const NDArray& arr, int
dtype = -1) {
+inline static bool ChooseBRGEMMImpl(mkldnn::memory::dims weight_dims, size_t
batch_size) {
+ // Conditions based on measurement results done on CLX8280
+ // https://github.com/apache/incubator-mxnet/pull/20533
+ return weight_dims[0] % 64 == 0 && weight_dims[1] % 64 == 0 &&
weight_dims[0] >= 1024 &&
Review comment:
Do the perf results are repeated? The perf result for all dims is really
close, I guess that perf comes from format settings, not it comes from its
internal implementation [?]
So, what would happen if the cache mechanism is enabled? (how do the
differences look like: results are close, are completely different or are the
same) [?]
--
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]
