bytedance / matxscript

A high-performance, extensible Python AOT compiler.
https://bytedance.github.io/matxscript/
Apache License 2.0
412 stars 39 forks source link

[Kernel] update benchmark #250

Closed jc-bytedance closed 1 year ago

jc-bytedance commented 1 year ago
  1. Enabled o3 and other optimization in the benchmark.
  2. In the kernel benchmark, the creation of an ndarray was intentionally incorporated to serve as a minor overhead, aligning the kernel benchmark with the code used for the numpy benchmark.
  3. Removed the use of the unit test framework in the benchmark code.