Exclude intel SVML functions from missing intrinsics report#1253
Exclude intel SVML functions from missing intrinsics report#1253jhorstmann wants to merge 2 commits into
Conversation
|
Thanks for the pull request, and welcome! The Rust team is excited to review your changes, and you should hear from @Amanieu (or someone else) soon. Please see the contribution instructions for more information. |
| // | ||
| // Open up the network console and you'll see an xml file was downloaded | ||
| // (currently called data-3.4.xml). That's the file we downloaded | ||
| // (currently called data-3.6.xml). That's the file we downloaded |
There was a problem hiding this comment.
Ignoring pack files this update will grow the repo size by 6.74 MB. With pack files, it will likely be a bit smaller, though possibly not by much.
There was a problem hiding this comment.
Good point, I did not consider the size. If that is a concern I could revert and rebase adding this file. It could be added once there are actual plans to implement to implement for example the VNNI or FP16 extensions. My goal is to work on avx512f and avx512vl for now, especially masked loads/stores.
|
☔ The latest upstream changes (presumably 35b0b1e) made this pull request unmergeable. Please resolve the merge conflicts. |
|
#1594 contains an updated and much extended version of this idea |
5 participants
Intel's Short Vector Math Library (SVML) defines optimized math functions which do not directly map to single cpu instructions but rather to a sequence of instructions. Exclude these from the missing intrinsics report so that it is much easier to see which real instructions are still missing for example for full AVX512F support.
This also updates the intrinsics data file to version 3.6