Uh oh!
There was an error while loading. Please reload this page.
[WIP] poly1305: AVX2 based on Bhattacharyya/Sarkar paper - #47
Closed
tarcieri wants to merge 1 commit into
Closed
Conversation
Mechanical translation originally based on Corrode. The original source code seems to be incomplete, but still amenable to a mechanical translation: https://github.com/Sreyosi/Improved-SIMD-Implementation-of-Poly1305 Notably its internal buffering in the original source code is commented out, but not part of the `UniversalHash`-trait API. We could potentially throw away all of the optimizations for larger-than-block-size buffers and the incomplete code for buffering non-block-aligned messages to get a baseline AVX2-optimized implementation of Poly1305. That said, this code is failing the basic test vectors, probably owing to its incompleteness, and it probably makes more sense to base an AVX2 implementation on well-tested code which may perform slower but is known to work and be generally robust.
tarcieri
commented
Jan 20, 2020
Comment on lines
+222
to
+266
| if in_len >= 768 { | ||
| let mut r8 = red5x64(mul130(state.r4, state.r4, state.r45)); | ||
| let mut r12 = red5x64(mul130(r8, state.r4, state.r45)); | ||
| let mut r85 = _mm256_permutevar8x32_epi32( | ||
| (_mm256_add_epi32((r8), (_mm256_slli_epi32((r8), (2))))), | ||
| (_mm256_set_epi32(4, 3, 2, 1, 1, 1, 1, 1)), | ||
| ); | ||
| let mut r125 = _mm256_permutevar8x32_epi32( | ||
| (_mm256_add_epi32((r12), (_mm256_slli_epi32((r12), (2))))), | ||
| (_mm256_set_epi32(4, 3, 2, 1, 1, 1, 1, 1)), | ||
| ); | ||
| r8 = _mm256_blend_epi32((r8), (r85), (0xE0)); | ||
| r12 = _mm256_blend_epi32((r12), (r125), (0xE0)); | ||
| r85 = _mm256_permute2x128_si256((r85), (r85), (0)); | ||
| r125 = _mm256_permute2x128_si256((r125), (r125), (0)); | ||
| loop { | ||
| p = add4x130( | ||
| red4x130(muladd4x130P( | ||
| muladd4x130P(mul4x130P(p, r12, r125), align4x128(load4x128(ip)), r8, r85), | ||
| align4x128(load4x128(ip.add(16))), | ||
| state.r4, | ||
| state.r45, | ||
| )), | ||
| align4x128(load4x128(ip.add(32))), | ||
| ); | ||
| in_len -= 192; | ||
| ip = ip.add(48); | ||
| if in_len < 192 { | ||
| break; | ||
| } | ||
| } | ||
| if in_len >= 128 { | ||
| p = add4x130( | ||
| red4x130(muladd4x130P( | ||
| mul4x130P(p, r8, r85), | ||
| align4x128(load4x128(ip)), | ||
| state.r4, | ||
| state.r45, | ||
| )), | ||
| align4x128(load4x128(ip.add(16))), | ||
| ); | ||
| in_len -= 128; | ||
| ip = ip.add(32); |
MemberAuthor
There was a problem hiding this comment.
It looks like an optimization which only wants to implement the current UniversalHash trait (which precludes optimizations for outputs larger than the UHF itself) could omit this entire codeblock and also any functions which are only invoked for this "wide" optimization.
tarcieri
commented
Jun 16, 2020
MemberAuthor
Closed in favor of #49 |
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 freeto join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Mechanical translation originally based on Corrode.
The original source code seems to be incomplete, but still amenable to a mechanical translation:
https://github.com/Sreyosi/Improved-SIMD-Implementation-of-Poly1305
Notably its internal buffering in the original source code is commented out, but not part of the
UniversalHash-trait API.We could potentially throw away all of the optimizations for larger-than-block-size buffers and the incomplete code for buffering non-block-aligned messages to get a baseline AVX2-optimized implementation of Poly1305.
That said, this code is failing the basic test vectors, probably owing to its incompleteness, and it probably makes more sense to base an AVX2 implementation on well-tested code which may perform slower but is known to work and be generally robust.