roywei edited a comment on issue #15429: Operator Performance Regression on CPU URL: https://github.com/apache/incubator-mxnet/issues/15429#issuecomment-508831150 Update --- There is no much regression (accounting variance also) between 1.4.1 and 1.5.0 on broadcast ops if using python `time()`. it may due to profiler change between versions. Result on C5.18xlarge. | 1.4.1 | 1.4.1 | 1.4.1 | average | 1.5.0 | 1.5.0 | 1.5.0 | 1.5.0 | 1.5.0 average -- | -- | -- | -- | -- | -- | -- | -- | -- | -- broadcast_add | 0.055 | 0.059 | 0.056624 | 0.0567635 | 0.058 | 0.06 | 0.057 | 0.057 | 0.058183 elemwise_add | 0.104 | 0.104460 | 0.104491 | 0.10444 | 0.121223 | 0.127 | 0.124 | 0.124 | 0.125145 Thanks to @reminisce for helping out and providing the scripts. ``` import time import mxnet as mx mx.random.seed(0) a = mx.nd.random.uniform(shape=(1024, 1024)) b = mx.nd.random.uniform(shape=(1024, 1024)) repeat = 10000 mx.nd.waitall() start = time.time() for _ in range(repeat): c = mx.nd.broadcast_add(a, b) # c = mx.nd.elemwise_add(a, b) c.wait_to_read() elapse = time.time() - start print("elapse time: %fms" % (elapse * 1000 / repeat)) ```
---------------------------------------------------------------- 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] With regards, Apache Git Services
