zixuanweeei commented on issue #14713: [WIP]MKLDNN RNN Inference Integration(fp32 LSTM and vRNN with tanh and relu) URL: https://github.com/apache/incubator-mxnet/pull/14713#issuecomment-493834966 Here is the test for all combinations covered by this PR. We evaluated the performance with MKLDNN commit 41bee20d7eb4a67feeeeb8d597b3598994eb1959 on local Skylake-6148 with 1 Sockets and 20 cores. ### Layer=1 bidirectional = False |API|MKLDNN = 0 (sample/sec)|MKLDNN = 1(sample/sec)|speedup| |---|-----------------------|----------------------|-------| |FusedLSTM|290 |659 |2.27x | |FusedvRNN with tanh|978 |1367 |1.40x | |FusedvRNN with relu|1297 |1341 |1.03x | ### Layer=1 bidirectional = True |API|MKLDNN = 0 (sample/sec)|MKLDNN = 1(sample/sec)|speedup| |---|-----------------------|----------------------|-------| |FusedLSTM|155 |340 |2.19x | |FusedvRNN with tanh|479 |703 |1.47x | |FusedvRNN with relu|627 |684 |1.09x | ### Layer=5 bidirectional = False |API|MKLDNN = 0 (sample/sec)|MKLDNN = 1(sample/sec)|speedup| |---|-----------------------|----------------------|-------| |FusedLSTM|64 |144 |2.25x | |FusedvRNN with tanh|215 |327 |1.52x | |FusedvRNN with relu|286 |320 |1.12x | ### Layer=5 bidirectional = True |API|MKLDNN = 0 (sample/sec)|MKLDNN = 1(sample/sec)|speedup| |---|-----------------------|----------------------|-------| |FusedLSTM|29 |61 |2.10x | |FusedvRNN with tanh|91 |142 |1.56x | |FusedvRNN with relu|119 |138 |1.16x |
---------------------------------------------------------------- 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
