Skip to content

only create as many tasks as threads and more inference barriers

Churavy, Valentin requested to merge vc/cpu_perf into master

Created by: vchuravy

Improves performance on the CPU from:

BenchmarkTools.Trial:
  memory estimate:  12.63 GiB
  allocs estimate:  117440577
  --------------
  minimum time:     55.920 s (27.54% GC)
  median time:      55.920 s (27.54% GC)
  mean time:        55.920 s (27.54% GC)
  maximum time:     55.920 s (27.54% GC)
  --------------

To:

  minimum time:     1.593 s (0.00% GC)
  median time:      1.649 s (0.00% GC)
  mean time:        1.649 s (0.00% GC)
  maximum time:     1.703 s (0.00% GC)

Benchmark simple_transpose

Merge request reports

Loading