[GitHub] haojin2 commented on issue #11179: [MXNET-404] elemwise_add/sub between rsp and rsp on GPU

[GitBox <gi...@apache.org>]

haojin2 commented on issue #11179: [MXNET-404] elemwise_add/sub between rsp and rsp on GPU
URL: https://github.com/apache/incubator-mxnet/pull/11179#issuecomment-395225648
 
 
   Benchmark script:
   ```Python
   import mxnet as mx
   import sys
   import os
   import scipy
   import numpy as np
   from mxnet.test_utils import rand_ndarray, assert_almost_equal
   import time
   
   def measure_cost(repeat, a, b, out=None):
       # start bench
       start = time.time()
       results = []
       for i in range(repeat):
           results.append(mx.nd.elemwise_add(a, b, out=out))
       for result in results:
           result.wait_to_read()
       end = time.time()
       diff = end - start
       return diff / repeat
   
   def measure_fallback(repeat, a):
       # start bench
       start = time.time()
       results = []
       for i in range(repeat):
           results.append(a.tostype('default'))
       for result in results:
           result.wait_to_read()
       end = time.time()
       diff = end - start
       return diff / repeat
   
   def main():
       shape = (1000000, 512)
       context = mx.gpu(0)
       # context = mx.cpu()
       for lhs_density in [0.01, 0.005, 0.001, 0.0005, 0.0001, 0.000]:
           mx_lhs = rand_ndarray(shape, stype='row_sparse', density=lhs_density).as_in_context(context)
           mx_lhs_dns = mx_lhs.tostype('default')
           for rhs_density in [0.01, 0.005, 0.001, 0.0005, 0.0001, 0.000]:
               mx_rhs = rand_ndarray(shape=shape, stype='row_sparse', density=rhs_density).as_in_context(context)
               mx_rhs_dns = mx_rhs.tostype('default')
               #warmup
               sparse_cost = 0.0
               dns_cost = 0.0
               np_lhs = mx_lhs_dns.asnumpy()
               check = mx.nd.elemwise_add(mx_lhs, mx_rhs)
               np_lhs = np_lhs + mx_rhs.asnumpy()
               assert_almost_equal(check.asnumpy(), np_lhs, atol=1e-5, rtol=1e-4)
               mx.nd.waitall()
               for i in range(100):
                   sparse_cost += measure_cost(1, mx_lhs, mx_rhs)
                   dns_cost += measure_cost(1, mx_lhs_dns, mx_rhs_dns)
               print("%.2f %% %.2f %%" % (lhs_density*100, rhs_density*100), dns_cost / sparse_cost)
   
       for rhs_density in [1.000, 0.01, 0.005, 0.001, 0.0005, 0.0001, 0.000]:
           mx_lhs_dns = mx.nd.ones(shape, ctx=context)
           mx_lhs = mx_lhs_dns.tostype('row_sparse')
           mx_rhs = rand_ndarray(shape=shape, stype='row_sparse', density=rhs_density).as_in_context(context)
           mx_rhs_dns = mx_rhs.tostype('default')
           #warmup
           sparse_cost = 0.0
           dns_cost = 0.0
           np_lhs = mx_lhs_dns.asnumpy()
           mx.nd.elemwise_add(mx_lhs, mx_rhs, out=mx_lhs)
           np_lhs = np_lhs + mx_rhs.asnumpy()
           assert_almost_equal(mx_lhs.asnumpy(), np_lhs, atol=1e-5, rtol=1e-4)
           mx.nd.waitall()
           for i in range(100):
               sparse_cost += measure_cost(1, mx_lhs, mx_rhs, out=mx_lhs)
               dns_cost += measure_cost(1, mx_lhs_dns, mx_rhs_dns, out=mx_lhs_dns)
           print("%.2f %% %.2f %%" % (1.00000*100, rhs_density*100), dns_cost / sparse_cost)
   
       for lhs_density in [1.000, 0.01, 0.005, 0.001, 0.0005, 0.0001, 0.000]:
           mx_rhs_dns = mx.nd.ones(shape, ctx=context)
           mx_rhs = mx_rhs_dns.tostype('row_sparse')
           mx_lhs = rand_ndarray(shape=shape, stype='row_sparse', density=lhs_density).as_in_context(context)
           mx_lhs_dns = mx_lhs.tostype('default')
           #warmup
           sparse_cost = 0.0
           dns_cost = 0.0
           np_rhs = mx_rhs_dns.asnumpy()
           mx.nd.elemwise_add(mx_lhs, mx_rhs, out=mx_rhs)
           np_rhs = np_rhs + mx_lhs.asnumpy()
           assert_almost_equal(mx_rhs.asnumpy(), np_rhs, atol=1e-5, rtol=1e-4)
           mx.nd.waitall()
           for i in range(100):
               sparse_cost += measure_cost(1, mx_lhs, mx_rhs, out=mx_rhs)
               dns_cost += measure_cost(1, mx_lhs_dns, mx_rhs_dns, out=mx_rhs_dns)
           print("%.2f %% %.2f %%" % (1.00000*100, lhs_density*100), dns_cost / sparse_cost)
   
   
   if __name__ == "__main__":
       main()
   ```

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services