Benchmarking

[apuignav]

First try at serious benchmarking.

Still, we're quite slow, also because creating multiple graphs takes a long time (total time is 202 seconds!):

(zfit36) [10:34]farm-gpu:~/zfit/tfphasespace/benchmark[benchmarks]$ CUDA_VISIBLE_DEVICES= python3 bench_tfphasespace.py
2019-03-12 10:35:18.778705: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
2019-03-12 10:35:19.994733: E tensorflow/stream_executor/cuda/cuda_driver.cc:300] failed call to cuInit: CUDA_ERROR_NO_DEVICE: no CUDA-capable device is detected
2019-03-12 10:35:19.994797: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:163] retrieving CUDA diagnostic information for host: farm-gpu
2019-03-12 10:35:19.994807: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:170] hostname: farm-gpu
2019-03-12 10:35:19.994851: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:194] libcuda reported version is: 410.78.0
2019-03-12 10:35:19.994891: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:198] kernel reported version is: 410.78.0
2019-03-12 10:35:19.994900: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:305] kernel version seems to match DSO: 410.78.0
Initial run (may takes more time than consequent runs)
Elapsed time: 93251.58056803048 ms
starting benchmark
Total number of generated samples 100000100
Shape of one particle momentum (4, 1000001)
Elapsed time: 70495.63611880876 ms
Time per sample: 7.049556562324313e-07 ms
CUDA_VISIBLE_DEVICES= python3 bench_tfphasespace.py  202.81s user 12.83s system 99% cpu 3:37.76 total

vs

(zfit36) [10:38]farm-gpu:~/zfit/tfphasespace/benchmark[benchmarks]$ root -q bench_tgenphasespace.cxx+

Processing bench_tgenphasespace.cxx+...
(int) 0
root -l -q bench_tgenphasespace.cxx+  26.52s user 0.11s system 97% cpu 27.206 total