Adding newly shared SVML Float16 HA imlementation - #9
Merged
Merged
Conversation
Functions added: acos, acosh, asin, asinh, atan, atan2, atanh, cbrt, cos, cosh, exp, exp2, exp10, expm1, log, log2, log10, log1p, pow, sin, sinh, tan, tanh
Add .section .note.GNU-stack,"",@progbits to all 23 FP16 HA assembly files.
- Add endbr64 at every function entry point for IBT (Indirect Branch Tracking) support - Add .note.gnu.property section declaring IBT (0xc0000002) and shadow stack (0x00000003) compatibility flags
r-devulap
changed the title
Member
Author
|
@seberg I will go ahead and merge this. I am curious if they provide any kind of perf benefits. Will experiment with it and pull it into NumPy if it makes sense. |
Member
|
Yes, thanks. Sorry it seems all fine to port. (I feel a bit we have to still figure out how much fp16 SIMD we want in NumPy, but if it's an obvious gain. We had prevoiusly issue with fp16 SIMD accuracy, because users expected float32, or close to it, accuracy.) |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Intel open sourced and shared the AVX-512 FP16 high-accuracy (HA) functions. Adding them here to the repo.
Functions added: acos, acosh, asin, asinh, atan, atan2, atanh, cbrt, cos, cosh, exp, exp2, exp10, expm1, log, log2, log10, log1p, pow, sin, sinh, tan, tanh.
It looks to me that these use FP32 internally to meet accuracy requirements. Current strategy in NumPy is to do that exactly: we convert to FP32, use the FP32 SVML functions and recovert back to FP16 which can all be done using the base AVX512 SKX ISA using vcvtph2ps (which doesn't need AVX512_FP16).