-
Notifications
You must be signed in to change notification settings - Fork 6.8k
[FEATURE] Add oneDNN support for numpy concatenate operator #20652
Conversation
Hey @agrabows , Thanks for submitting the PR
CI supported jobs: [clang, edge, centos-gpu, unix-gpu, unix-cpu, sanity, miscellaneous, windows-gpu, centos-cpu, website, windows-cpu] Note: |
2102099
to
964df6a
Compare
964df6a
to
2c8d63b
Compare
@mxnet-bot run ci [unix-cpu] |
Jenkins CI successfully triggered : [unix-cpu] |
ab1e7a6
to
f36c352
Compare
@mxnet-bot run ci [unix-cpu] |
Jenkins CI successfully triggered : [unix-cpu] |
@mxnet-bot run ci [unix-gpu] |
Jenkins CI successfully triggered : [unix-gpu] |
@mxnet-bot run ci [unix-gpu] |
Jenkins CI successfully triggered : [unix-gpu] |
@szha Could you please take a look and possibly merge this PR? |
@agrabows thanks! |
Description
Function concatenate() from module mxnet.numpy will now be executed by oneDNN primitives as it is done in module mxnet.ndarray. Additionally concat from mxnet.ndarray will gain numpy fuctionality which is flattening input arrays when parameter dim=None.
Checklist
Essentials
Changes
Comments
It was possible to merge both ndarray and numpy concatenate operators into one NNVM_REGISTER_OP, additional functionality was added to ndarray concatenate by this process.
Performance tests from mxnet profiler with 10 warmup runs and 50 test runs of this operator are given below:
Numpy concatenate was about 9 times faster in those tests