Reduce number of Window test configurations (#38482)
Summary:
After this diff, on PR following compilation configuration would be running:
- VS2017 14.11, CUDA10.1
- VS2017 no CUDA, CUDA10.1
- VS2019, CUDA10.1
And tested:
- VS2017 14.11, CUDA10.1
- VS2017 14.11 no CUDA (only 1st half of tests)
- VS2017 14.11 force on CPU (only 1st half of test)
And on master, we would be building both VS2017 14.11 and 14.16, but testing only VS2017 14.11 and VS2019 builds.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38482
Differential Revision: D22111743
Pulled By: malfet
fbshipit-source-id: d660e4bc8f4f17a93f1cc18402cd5f2091b7789d
