openvino
openvino copied to clipboard
Optimize DFT operation
Details:
Cache twiddles for FFT function and add x86 JIT implementations.
ID: 62580
It depends on https://github.com/openvinotoolkit/oneDNN/pull/142
@liubo-intel Could you please take a review?
@chenhu-wang @xuchen-intel @usstq please help review this PR. Thanks.
Please refer to https://github.com/openvinotoolkit/openvino/blob/master/src/tests/functional/plugin/cpu/single_layer_tests/matmul.cpp and add CPUSpecificParams unit tests to cover all different executor branches. Thanks.
@dmitry-gorokhov anyone from Russia team could have a look at this PR?
@nshchego Please review the PR
ache twiddles for FFT function and add x86 JIT implementations.
Please add the ticket number.
@nshchego Any additional comments from your side?
Is there a PR for oneDNN changes?
Yes, it's located here: https://github.com/openvinotoolkit/oneDNN/pull/142
The code LGTM. We need to run Bench to prove the performance gain. I'll try to do it soon.
@avoskoboinyk-lohika Could you please add a link to oneDNN PR to the description
The code LGTM. We need to run Bench to prove the performance gain. I'll try to do it soon.
Did you have a chance to check the performance?
@ceciliapeng2011, do we have enough data on performance validation to merge the PR?