anko-intel opened a new pull request #20780:
URL: https://github.com/apache/incubator-mxnet/pull/20780


   ## Description ##
   Add microbenchmark for Fully Connected with add fusion
   and utility scripts to run it with proper OMP parameters set and check 
performance against different number of threads
   
   ## Checklist ##
   ### Essentials ###
   - [ ] PR's title starts with a category (e.g. [BUGFIX], [MODEL], [TUTORIAL], 
[FEATURE], [DOC], etc)
   - [ ] Changes are complete (i.e. I finished coding on this PR)
   
   
   ## Comments ##
   Examples of test output:
   - ./run.sh fc_add.py
   
   elemwise_add, float 
   
   |    Shape    | Hidden | Mean [ms] |
   |------------:|-------:|----------:|
   | (   1, 224) |    512 |     0.236 |
   | (   1, 224) |   4096 |     0.227 |
   | (  16,1024) |   1024 |     0.337 |
   | (  32,4096) |   1024 |     0.530 |
   | (  32,4096) |   4096 |     1.539 |
   | ( 512, 512) |   4096 |     0.956 |
   
   
   npi_add, float 
   
   |    Shape    | Hidden | Mean [ms] |
   |------------:|-------:|----------:|
   | (   1, 224) |    512 |     0.165 |
   | (   1, 224) |   4096 |     0.164 |
   | (  16,1024) |   1024 |     0.286 |
   | (  32,4096) |   1024 |     0.523 |
   | (  32,4096) |   4096 |     1.444 |
   | ( 512, 512) |   4096 |     0.846 |
   
   
   elemwise_add, mode = smart, granularity = tensor-wise 
   
   |    Shape    | Hidden | Mean [ms] |
   |------------:|-------:|----------:|
   | (   1, 224) |    512 |     0.240 |
   | (   1, 224) |   4096 |     0.236 |
   | (  16,1024) |   1024 |     0.330 |
   | (  32,4096) |   1024 |     0.396 |
   | (  32,4096) |   4096 |     0.541 |
   | ( 512, 512) |   4096 |     0.900 |
   
   - ./run_per_thread.sh 48 8 64 fc_add.py
   
   NUM_THREADS = 48 56 64
   
   
   elemwise_add, float   
   
   |    Shape    | Hidden | Mean [ms] | Mean [ms] | Mean [ms] |
   |------------:|-------:|----------:|----------:|----------:|
   | (   1, 224) |    512 |     0.204 |     0.214 |     0.244 |
   | (   1, 224) |   4096 |     0.205 |     0.211 |     0.232 |
   | (  16,1024) |   1024 |     0.307 |     0.321 |     0.346 |
   | (  32,4096) |   1024 |     0.569 |     0.535 |     0.536 |
   | (  32,4096) |   4096 |     2.028 |     1.796 |     1.531 |
   | ( 512, 512) |   4096 |     1.124 |     1.085 |     0.962 |
   
   


-- 
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]


Reply via email to