bgawrych opened a new pull request #18885:
URL: https://github.com/apache/incubator-mxnet/pull/18885


   ## Description ##
   This PR deals with issue 
https://github.com/apache/incubator-mxnet/issues/18854
   
   
   Benchmark:
   ```
   import mxnet as mx
   import time
   
   mx.random.seed(123)
   tic = time.time()
   
   for i in range(1000):
       rand_a = mx.nd.random.randint(1000, 10000).asscalar()
       rand_b = mx.nd.random.randint(1000, 10000).asscalar()
       shape = (rand_a, rand_b)
       a = mx.nd.random.uniform(shape=shape)
       b = mx.nd.softmax(a)
       a.wait_to_read()
       b.wait_to_read()
   
   toc = time.time()
   print(toc - tic)
   ```
   
   Intel(R) Xeon(R) CPU E5-2697 v2 @ 2.70GHz  (without AVX512)
   ```
   KMP_AFFINITY=granularity=fine,noduplicates,compact,1,0 OMP_NUM_THREADS=12 
numactl --physcpubind=0-11 --membind=0 python3 bench.py  
   ```
   | Aligned | Not Aligned |
   | ------ | ------ |
   | 232.6839632987976 | 225.73387098312378 |
   | 231.99822640419006 | 225.51357173919678 | 
   
   It's about 3% performance drop on this particular test case.
   
   Intel(R) Xeon(R) Platinum 8260L CPU @ 2.40GHz
   ```
   KMP_AFFINITY=granularity=fine,noduplicates,compact,1,0 OMP_NUM_THREADS=24 
numactl --physcpubind=0-23 --membind=0 python3 bench.py
   ```
   
   | Aligned | Not Aligned |
   | ------ | ------ |
   | 15.814676523208618 | 15.833234071731567 |
   | 15.925291061401367 | 15.907763004302979 | 
   | 15.941178321838379 | 15.9696786403656| 
   | 16.01563811302185 | 16.00200080871582 | 
   
   For better CPU with AVX512 there is no difference in this case
   
   ## Checklist ##
   ### Essentials ###
   Please feel free to remove inapplicable items for your PR.
   - [x] Changes are complete (i.e. I finished coding on this PR)
   - [x] All changes have test coverage:
   - [x] To the best of my knowledge, examples are either not affected by this 
change, or have been fixed to be compatible with this change
   


----------------------------------------------------------------
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.

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to