tilelang.profiler.msprof¶

Ascend msprof timing implementation.

Functions¶

get_msprof_cache_flush_kernel(numel)

bench_with_msprof(fn, cache, n_repeat[, detailed])

Benchmark using torch_npu msprof timing.

Module Contents¶

tilelang.profiler.msprof.get_msprof_cache_flush_kernel(numel)¶
Parameters:

numel (int)

tilelang.profiler.msprof.bench_with_msprof(fn, cache, n_repeat, detailed=False)¶

Benchmark using torch_npu msprof timing.

Parameters:

  • fn (collections.abc.Callable)

  • cache (torch.Tensor)

  • n_repeat (int)

  • detailed (bool)

Return type:

float