Speed up CI runtime #3189
Speed up CI runtime #3189
Conversation
- Cache metadata - Increase number of workers
* Adding parallelism to cmake. * Adding parallelism to msbuild.
* Move definition outside of nested loop to avoid repeated jit scripting. * Fix potentially undefined var.
* Separating unrelated checks to avoid unnecessary repetition. * Add cache on get_fn_args().
* Upgrade scipy on windows. * Minor clean ups.
|
The failing tests are not related to this PR but just to be safe, I'll wait until the CI issues are resolved before merging. |
| @@ -496,6 +497,7 @@ def expected_fn(self, x, weight, offset, mask, bias, stride=1, padding=0, dilati | |||
| out += bias.view(1, n_out_channels, 1, 1) | |||
| return out | |||
|
|
|||
| @lru_cache(maxsize=None) | |||
There was a problem hiding this comment.
Deciding on caching random data is not that simple IMO. There are pros/cons of that...
There was a problem hiding this comment.
I agree there are pros/cons. Note that this is something we do on tests only and seems to help us speed-wise. I don't have a strong opinion about keeping it if it causes issues.
| @@ -1,3 +1,3 @@ | |||
| @echo on | |||
| msbuild "-p:Configuration=Release" torchvision.vcxproj | |||
| msbuild "-p:Configuration=Release" INSTALL.vcxproj | |||
| msbuild "-p:Configuration=Release" "-p:BuildInParallel=true" "-p:MultiProcessorCompilation=true" "-p:CL_MPCount=%1" torchvision.vcxproj -maxcpucount:%1 | |||
There was a problem hiding this comment.
So, here /MP is not the right option ? Funny that on windows we have to set 2 more flags...
There was a problem hiding this comment.
As far as I understand after reading the Microsoft docs, MultiProcessorCompilation == /MP. And there are many more flags, related to whether the parallelism is across projects or across C++ files. Frankly I can't say I'm fully convinced that this helped much. If we have anyone who is seasoned on working with msbuild, I would love to get their input.
Summary: * Speedup test_ucf101 (#2623 * Speedup Cmake builds (#3186) * Speedup test_autoaugment (#3190) * Speedup DeformConvTester (#3191) * Speedup InceptionV3 and GoogleNet on Windows (#3196) Reviewed By: datumbox Differential Revision: D25954568 fbshipit-source-id: bdcea84b112a9343f27619aef7036369598f631e
This PR contains a number of improvements that aim to speed up the building and testing times of the project. None of the changes alter the quality and quantity of tests that we do (aka we don't modify batch sizes, vector sizes, iterations, number of tests etc). Further improvements can be achieved if we make such modifications on the slowest tests.
See sub-PRs for more details:
To properly estimate the CI cost savings and dev time improvements of all the above changes we will require doing multiple runs which is both impractical and costly. Instead the stats provided above come from a couple of runs and thus should be taken with a grain of salt. At any case to get a rough idea of the impact of the improvements: