Closed
Description
from torch.profiler import profile, record_function, ProfilerActivity
import torch
x = torch.randn(0)
with profile(activities=[ProfilerActivity.CPU], record_shapes=True) as prof:
with record_function("model_inference"):
x + x
print(prof.key_averages().table(sort_by="cpu_time_total", row_limit=10))Gives:
In [7]: print(prof.key_averages().table(sort_by="cpu_time_total", row_limit=10))
----------------------------- ------------ ------------ ------------ ------------ ------------ ------------
Name Self CPU % Self CPU CPU total % CPU total CPU time avg # of Calls
----------------------------- ------------ ------------ ------------ ------------ ------------ ------------
aten::matmul 0.46% 8.994us 62.32% 1.213ms 606.382us 2
aten::dot 61.72% 1.201ms 61.86% 1.204ms 601.884us 2
model_inference 6.61% 128.555us 8.13% 158.251us 158.251us 1
aten::to 1.04% 20.242us 5.30% 103.077us 3.221us 32
aten::_to_copy 2.19% 42.586us 4.26% 82.835us 2.589us 32
aten::ones 2.08% 40.453us 2.87% 55.895us 13.974us 4
aten::add 2.32% 45.200us 2.59% 50.328us 12.582us 4
aten::abs 1.27% 24.757us 2.20% 42.744us 21.372us 2
aten::__lshift__ 0.67% 12.990us 1.76% 34.283us 34.283us 1
aten::pow 1.40% 27.282us 1.58% 30.817us 10.272us 3
----------------------------- ------------ ------------ ------------ ------------ ------------ ------------
which seems really bizarre
cc @robieta @chaekit @guotuofeng @guyang3532 @dzhulgakov @davidberard98 @briancoutinho @sraikund16 @sanrise