AVX2 optimizations for Q5_0, Q5_1#1195
Merged
ggerganov merged 1 commit intoggml-org:q5_0from Apr 26, 2023
Merged
Conversation
ggerganov
reviewed
Apr 26, 2023
Comment on lines
+331
to
+342
| #define B1(c,s,n) 0x ## n ## c , 0x ## n ## s | ||
| #define B2(c,s,n) B1(c,s,n ## c), B1(c,s,n ## s) | ||
| #define B3(c,s,n) B2(c,s,n ## c), B2(c,s,n ## s) | ||
| #define B4(c,s,n) B3(c,s,n ## c), B3(c,s,n ## s) | ||
| #define B5(c,s,n) B4(c,s,n ## c), B4(c,s,n ## s) | ||
| #define B6(c,s,n) B5(c,s,n ## c), B5(c,s,n ## s) | ||
| #define B7(c,s,n) B6(c,s,n ## c), B6(c,s,n ## s) | ||
| #define B8(c,s ) B7(c,s, c), B7(c,s, s) | ||
|
|
||
| // precomputed tables for expanding 8bits to 8 bytes (shl 4) | ||
| static const uint64_t table_b2b_u[1 << 8] = { B8(00, 10) }; | ||
| static const uint64_t table_b2b_i[1 << 8] = { B8(F0, 00) }; |
ggerganov
approved these changes
Apr 26, 2023
Member
ggerganov
left a comment
There was a problem hiding this comment.
The ARM NEON results are the same
ggerganov
added a commit
that referenced
this pull request
Apr 26, 2023
* ggml : add Q5_0 quantization (cuBLAS only) * ggml : fix Q5_0 qh -> uint32_t * ggml : fix q5_0 histogram stats * ggml : q5_0 scalar dot product * ggml : q5_0 ARM NEON dot * ggml : q5_0 more efficient ARM NEON using uint64_t masks * ggml : rename Q5_0 -> Q5_1 * ggml : adding Q5_0 mode * quantize : add Q5_0 and Q5_1 to map * ggml : AVX2 optimizations for Q5_0, Q5_1 (#1195) --------- Co-authored-by: Stephan Walter <stephan@walter.name>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Note: This isn't for master, but the branch in PR #1187.
This adds AVX2 optimizations for Q5_0 and Q5_1, with the help of another look-up table.
I didn't test ARM NEON, maybe I broke something there. You should probably check if the u8->u32->u8 roundtrip still makes sense.
I'll look into using
_mm256_shuffle_epi8, but you may want to merge this to the PR branch as it is.