NaNAGISaSA commented on PR #12303:
URL: https://github.com/apache/tvm/pull/12303#issuecomment-1208867049
Hi, @tkonolige , thank you for your review.
I think we do not need to change the reorder code below, here is the
benchmark result and code analysis:
### Benchmark
- Model: ResNet50 With Quantization
- Backend: Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz
- Threads: 4
- Benchmark Before Change:
```
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
586.5478 586.6723 587.4650 585.2788 0.7180
```
- Benchmark After Change:
```
Execution time summary:
mean (ms) median (ms) max (ms) min (ms) std (ms)
584.4176 585.1300 586.2465 580.8972 1.8741
```
So we can see that the benchmark result for ResNet50 before and after
changing the code is equal.
### Code Analysis
In the context of the changes, for `iw` variable, it is not used in the
following code in its function. And for `ic_block` variable, it is reassigned
immediately with the `ic_block` axis for `kernel_vec` in the following line 170
and 308, so this change will also not influence the following code.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
