gistfile1.txt

~/local/pytorch  $ python benchmarks/profiler_benchmark/profiler_bench.py
Payload: loop; 256 iterations, N = 100

Profiling disabled, tensor size 1x1
        avg. time: 0.663 ms, stddev: 0.036 ms ;  time per iteration: 0.003 ms

Profiling enabled , tensor size 1x1
        avg. time: 20.641 ms, stddev: 0.643 ms ;  time per iteration: 0.081 ms

Profiling disabled, tensor size 32x32
        avg. time: 1.256 ms, stddev: 0.055 ms ;  time per iteration: 0.005 ms

Profiling enabled , tensor size 32x32
        avg. time: 21.430 ms, stddev: 0.626 ms ;  time per iteration: 0.084 ms

Profiling disabled, tensor size 128x128
        avg. time: 9.208 ms, stddev: 0.424 ms ;  time per iteration: 0.036 ms

Profiling enabled , tensor size 128x128
        avg. time: 29.457 ms, stddev: 1.884 ms ;  time per iteration: 0.115 ms

Profiling disabled, tensor size 256x256
        avg. time: 16.352 ms, stddev: 0.346 ms ;  time per iteration: 0.064 ms

Profiling enabled , tensor size 256x256
        avg. time: 40.883 ms, stddev: 3.686 ms ;  time per iteration: 0.160 ms

Profiling disabled, tensor size 512x512
        avg. time: 119.053 ms, stddev: 1.903 ms ;  time per iteration: 0.465 ms

Profiling enabled , tensor size 512x512
        avg. time: 138.641 ms, stddev: 6.631 ms ;  time per iteration: 0.542 ms

Payload: parallel; 256 iterations, N = 100

Profiling disabled, tensor size 1x1
        avg. time: 0.287 ms, stddev: 0.054 ms ;  time per iteration: 0.001 ms

Profiling enabled , tensor size 1x1
        avg. time: 19.083 ms, stddev: 0.748 ms ;  time per iteration: 0.075 ms

Profiling disabled, tensor size 32x32
        avg. time: 0.492 ms, stddev: 0.118 ms ;  time per iteration: 0.002 ms

Profiling enabled , tensor size 32x32
        avg. time: 20.473 ms, stddev: 2.186 ms ;  time per iteration: 0.080 ms

Profiling disabled, tensor size 128x128
        avg. time: 6.842 ms, stddev: 0.951 ms ;  time per iteration: 0.027 ms

Profiling enabled , tensor size 128x128
        avg. time: 27.044 ms, stddev: 3.586 ms ;  time per iteration: 0.106 ms

Profiling disabled, tensor size 256x256
        avg. time: 15.606 ms, stddev: 1.528 ms ;  time per iteration: 0.061 ms

Profiling enabled , tensor size 256x256
        avg. time: 35.512 ms, stddev: 4.111 ms ;  time per iteration: 0.139 ms

Profiling disabled, tensor size 512x512
        avg. time: 82.051 ms, stddev: 6.189 ms ;  time per iteration: 0.321 ms

Profiling enabled , tensor size 512x512
        avg. time: 103.030 ms, stddev: 8.640 ms ;  time per iteration: 0.402 ms




---


without TLS optimization:


~/local/pytorch  $ python benchmarks/profiler_benchmark/profiler_bench.py
Payload: loop; 256 iterations, N = 100

Profiling disabled, tensor size 1x1
        avg. time: 0.508 ms, stddev: 0.028 ms ;  time per iteration: 0.002 ms

Profiling enabled , tensor size 1x1
        avg. time: 19.410 ms, stddev: 0.543 ms ;  time per iteration: 0.076 ms

Profiling disabled, tensor size 32x32
        avg. time: 0.918 ms, stddev: 0.032 ms ;  time per iteration: 0.004 ms

Profiling enabled , tensor size 32x32
        avg. time: 20.344 ms, stddev: 0.825 ms ;  time per iteration: 0.079 ms

Profiling disabled, tensor size 128x128
        avg. time: 9.021 ms, stddev: 0.387 ms ;  time per iteration: 0.035 ms

Profiling enabled , tensor size 128x128
        avg. time: 29.780 ms, stddev: 0.967 ms ;  time per iteration: 0.116 ms

Profiling disabled, tensor size 256x256
        avg. time: 15.854 ms, stddev: 0.608 ms ;  time per iteration: 0.062 ms

Profiling enabled , tensor size 256x256
        avg. time: 39.132 ms, stddev: 5.546 ms ;  time per iteration: 0.153 ms

Profiling disabled, tensor size 512x512
        avg. time: 119.083 ms, stddev: 10.857 ms ;  time per iteration: 0.465 ms

Profiling enabled , tensor size 512x512
        avg. time: 139.171 ms, stddev: 3.472 ms ;  time per iteration: 0.544 ms

Payload: parallel; 256 iterations, N = 100

Profiling disabled, tensor size 1x1
        avg. time: 0.275 ms, stddev: 0.066 ms ;  time per iteration: 0.001 ms

Profiling enabled , tensor size 1x1
        avg. time: 20.333 ms, stddev: 1.327 ms ;  time per iteration: 0.079 ms

Profiling disabled, tensor size 32x32
        avg. time: 0.478 ms, stddev: 0.115 ms ;  time per iteration: 0.002 ms

Profiling enabled , tensor size 32x32
        avg. time: 20.322 ms, stddev: 0.373 ms ;  time per iteration: 0.079 ms

Profiling disabled, tensor size 128x128
        avg. time: 6.615 ms, stddev: 0.931 ms ;  time per iteration: 0.026 ms

Profiling enabled , tensor size 128x128
        avg. time: 26.535 ms, stddev: 2.230 ms ;  time per iteration: 0.104 ms

Profiling disabled, tensor size 256x256
        avg. time: 15.361 ms, stddev: 1.943 ms ;  time per iteration: 0.060 ms

Profiling enabled , tensor size 256x256
        avg. time: 34.873 ms, stddev: 1.484 ms ;  time per iteration: 0.136 ms

Profiling disabled, tensor size 512x512
        avg. time: 84.417 ms, stddev: 6.605 ms ;  time per iteration: 0.330 ms

Profiling enabled , tensor size 512x512
        avg. time: 102.702 ms, stddev: 6.269 ms ;  time per iteration: 0.401 ms

---

trunk

~/local/pytorch  $ python benchmarks/profiler_benchmark/profiler_bench.py
Payload: loop; 256 iterations, N = 100

Profiling disabled, tensor size 1x1
        avg. time: 0.527 ms, stddev: 0.010 ms ;  time per iteration: 0.002 ms

Profiling enabled , tensor size 1x1
        avg. time: 21.159 ms, stddev: 3.329 ms ;  time per iteration: 0.083 ms

Profiling disabled, tensor size 32x32
        avg. time: 0.927 ms, stddev: 0.008 ms ;  time per iteration: 0.004 ms

Profiling enabled , tensor size 32x32
        avg. time: 20.517 ms, stddev: 0.528 ms ;  time per iteration: 0.080 ms

Profiling disabled, tensor size 128x128
        avg. time: 8.595 ms, stddev: 0.744 ms ;  time per iteration: 0.034 ms

Profiling enabled , tensor size 128x128
        avg. time: 30.312 ms, stddev: 1.679 ms ;  time per iteration: 0.118 ms

Profiling disabled, tensor size 256x256
        avg. time: 18.019 ms, stddev: 2.982 ms ;  time per iteration: 0.070 ms

Profiling enabled , tensor size 256x256
        avg. time: 38.106 ms, stddev: 0.649 ms ;  time per iteration: 0.149 ms

Profiling disabled, tensor size 512x512
        avg. time: 117.698 ms, stddev: 2.885 ms ;  time per iteration: 0.460 ms

Profiling enabled , tensor size 512x512
        avg. time: 137.435 ms, stddev: 5.857 ms ;  time per iteration: 0.537 ms