HyperZealot commented on issue #12997: A better take forward kernel for CPU URL: https://github.com/apache/incubator-mxnet/pull/12997#issuecomment-433747551 @pengzhao-intel script: ```Python import mxnet as mx import random from mxnet.test_utils import rand_ndarray, check_speed num_rows = 1000000 test_rows = 20000 num_cols = 512 shape = (num_rows, num_cols) data = rand_ndarray(shape, stype='default') print(data.dtype) x = mx.sym.Variable('data') idx = mx.sym.Variable('indices') sym = mx.sym.take(a=x, indices=idx, axis=0) trials = int(num_rows / test_rows) total_time = 0.0 all_indices = [i for i in range(num_rows)] random.shuffle(all_indices) for i in range(trials): indices = mx.nd.array(all_indices[i * test_rows: (i + 1) * test_rows]) total_time += check_speed(sym, location={'data': data, 'indices': indices}, ctx=mx.cpu(), N=100, typ="forward") print(test_rows * num_cols * 4 * trials / total_time / 1024 / 1024 / 1024, "GB/s") ``` Before with mxnet-mkl release version: 2.0371343465121163 GB/s After: 12.2486153514499 GB/s ~6x speedup with this new one
---------------------------------------------------------------- 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: [email protected] With regards, Apache Git Services
