Rip out core::arch wrappers#216
Conversation
… previously that feature was only applied to ARM
|
Should this be merged before #214 or after? |
|
I approved the other PR, not sure if you wanted to wait for input from Daniel though. |
|
Thanks! I'll wait a few days to let Daniel and the others chime in. This isn't blocking any follow-up work as far as I know. If it holds anything up, please let me know and I'll go ahead and merge it. |
|
Now that #214 is in, this should also be good to go. Daniel has explicitly stated he supports removing this: #212 (review) I'd appreciate an approval on this PR |
|
I've not checked yet, as I'm on mobile. There aren't two levels of tokens now, right, or are there? |
|
There aren't two levels, no. The tokens were simply in the same file as all the macros to generate wrappers. |
LaurenzV
left a comment
There was a problem hiding this comment.
LGTM, just two things but can/should be follow-ups:
- Is there even any point in keeping the SSE3/FMA/etc. tokens since those aren't supported natively anyway?
- We probably should remove the subfolders and put everything into a single file per architecture or even just everything into one file.
|
(Removing it from the queue just to make sure you saw my comment, feel free to re-add) |
|
I mean these structures: pub struct Avx2 {
pub avx2: crate::core_arch::x86::Avx2,
}I'm on mobile, so it's not easy to browse your branch, but I don't think those are collapsed. It's fine for that to be another follow-up, fwiw |
|
Ah, good catch! I totally missed that the token structs are wrapped and are indeed duplicated. I'll remove them before re-enabling auto-merge. Thanks! |
e972e69 to
0471987
Compare
…emit the wrapper structs directly without core_arch references
0471987 to
d1711d6
Compare
|
Committing file deletions to git is hard 😅 |
| /// This can be used in combination with the `safe_wrappers` feature to gain checked access to | ||
| /// the level-specific SIMD capabilities. |
There was a problem hiding this comment.
Possibly we should point to the kernel macro here. This isn't urgent
Remove the wrappers for intrinsics inherited from
pulpnow that #214 provides a way to access intrinsics without us having to wrap every single one.Part of #166
Supersedes #209