You are viewing a plain text version of this content. The canonical link for it is here.

Posted to discuss-archive@mxnet.apache.org by LookingForPerf via MXNet Forum <no...@mxnet.discoursemail.com> on 2022/03/25 07:29:26 UTC

[MXNet Forum] [Performance] SIMD based symbol or imperative operators


Greetings,

First of all I want to praise all the developers involved in this outstanding project!
I am doing my 1st year masters thesis on re-using ML frameworks for columnar databases internal workload.
At this point I benchmark/compare ML frameworks. Here is my graph based on 1st order int64 flat Tensors( shape = (8192)). 
where(
  or (
    equal(src_tensor, first_filter_tensor),
    equal(src_tensor, first_filter_tensor)),
  ),
  src_tensor, empty_magic_tensor)
Did I get it right there is no SIMD-based CPU implementation for: where, or,  equal  ops in mshadow ATM? Does it make any sense to try Symbol ops or they are translated in the same GOMP + mshadow primitives?
I appreciate any suggestion on performance/imperative code. 
[Here](https://www.toptal.com/developers/hastebin/orogafamuw.properties) is the actual code I used to measure the timings for the graph I mentioned.





---
[Visit Topic](https://discuss.mxnet.apache.org/t/simd-based-symbol-or-imperative-operators/7163/1) or reply to this email to respond.

You are receiving this because you enabled mailing list mode.

To unsubscribe from these emails, [click here](https://discuss.mxnet.apache.org/email/unsubscribe/45de41ecbe98a7f0298cf79b9fbba0183d97b0ad0d1cf75695b41bae36422fba).