search

openxla / xla

A machine learning compiler for GPUs, CPUs, and ML accelerators
Apache License 2.0
2.61k stars 407 forks source link

Add an XLA:CPU fusion benchmark. #14378

Closed copybara-service[bot] closed 3 months ago

copybara-service[bot] commented 3 months ago

Add an XLA:CPU fusion benchmark.

Thunk runtime ("new") is 5-12% slower than the classic runtime ("old")

name                                                old cpu/op   new cpu/op   delta
BM_FusionF32_2/40/process_time                      1.56µs ± 2%  1.75µs ± 1%   +12.13%  (p=0.008 n=5+5)
BM_FusionF32_2/80/process_time                      2.21µs ± 1%  2.41µs ± 1%    +9.04%  (p=0.008 n=5+5)
BM_FusionF32_2/160/process_time                     3.54µs ± 0%  3.77µs ± 1%    +6.56%  (p=0.008 n=5+5)
BM_FusionF32_2/240/process_time                     4.88µs ± 0%  5.12µs ± 1%    +4.82%  (p=0.008 n=5+5)