[profier] Fix double printing of FLOPs (#56974)
Summary:
Call table() shouldn't modify the events
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56974
Test Plan:
```
import torch
from torch import nn
from torch.profiler import profile, record_function
model = nn.Conv2d(8, 64, 3, padding=1)
input = torch.randn(1, 8, 272, 272)
with profile(record_shapes=True, with_flops=True) as prof:
with record_function("model_inference"):
model(input)
events = prof.key_averages(group_by_input_shape=True)
print(events.table())
print(events.table())
```
```
---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------- ------------
Name Self CPU % Self CPU CPU total % CPU total CPU time avg # of Calls Input Shapes GFLOPS/s
---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------- ------------
aten::zeros 0.78% 68.000us 1.16% 101.000us 101.000us 1 [[], [], [], [], []] --
aten::empty 0.49% 43.000us 0.49% 43.000us 14.333us 3 [[], [], [], [], [], []] --
aten::zero_ 0.23% 20.000us 0.23% 20.000us 20.000us 1 [[1]] --
model_inference 13.67% 1.195ms 98.84% 8.639ms 8.639ms 1 [] --
aten::conv2d 0.42% 37.000us 85.13% 7.440ms 7.440ms 1 [[1, 8, 272, 272], [64, 8, 3, 3], [64], [], [ 91.645
aten::convolution 0.15% 13.000us 84.70% 7.403ms 7.403ms 1 [[1, 8, 272, 272], [64, 8, 3, 3], [64], [], [ --
aten::_convolution 0.48% 42.000us 84.55% 7.390ms 7.390ms 1 [[1, 8, 272, 272], [64, 8, 3, 3], [64], [], [ --
aten::mkldnn_convolution 83.47% 7.295ms 84.07% 7.348ms 7.348ms 1 [[1, 8, 272, 272], [64, 8, 3, 3], [64], [], [ --
aten::as_strided_ 0.31% 27.000us 0.31% 27.000us 27.000us 1 [[1, 64, 272, 272], [], [], []] --
---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------- ------------
Self CPU time total: 8.740ms
---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------- ------------
Name Self CPU % Self CPU CPU total % CPU total CPU time avg # of Calls Input Shapes GFLOPS/s
---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------- ------------
aten::zeros 0.78% 68.000us 1.16% 101.000us 101.000us 1 [[], [], [], [], []] --
aten::empty 0.49% 43.000us 0.49% 43.000us 14.333us 3 [[], [], [], [], [], []] --
aten::zero_ 0.23% 20.000us 0.23% 20.000us 20.000us 1 [[1]] --
model_inference 13.67% 1.195ms 98.84% 8.639ms 8.639ms 1 [] --
aten::conv2d 0.42% 37.000us 85.13% 7.440ms 7.440ms 1 [[1, 8, 272, 272], [64, 8, 3, 3], [64], [], [ 91.645
aten::convolution 0.15% 13.000us 84.70% 7.403ms 7.403ms 1 [[1, 8, 272, 272], [64, 8, 3, 3], [64], [], [ --
aten::_convolution 0.48% 42.000us 84.55% 7.390ms 7.390ms 1 [[1, 8, 272, 272], [64, 8, 3, 3], [64], [], [ --
aten::mkldnn_convolution 83.47% 7.295ms 84.07% 7.348ms 7.348ms 1 [[1, 8, 272, 272], [64, 8, 3, 3], [64], [], [ --
aten::as_strided_ 0.31% 27.000us 0.31% 27.000us 27.000us 1 [[1, 64, 272, 272], [], [], []] --
---------------------------- ------------ ------------ ------------ ------------ ------------ ------------ --------------------------------------------- ------------
Self CPU time total: 8.740ms
```
Fixes https://github.com/pytorch/pytorch/issues/55606
Reviewed By: xuzhao9
Differential Revision: D28019925
Pulled By: ilia-cher
fbshipit-source-id: 7e7d7ed496059caf917a3dd8dea2daaceb5db920
