Remove Q4/Q5 bit shuffling without breaking compatibility #1384
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Outdated suggestions cannot be applied.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.
As an alternative to #1305, this keeps Q4 and Q5 the same but changes the order of Q8.
This is pretty rough for now, but I'm putting it here as a draft to invite discussion, and also to get help for SIMD optimizations for
quantize_row_q8_x
.I can't test ARM NEON or CUDA, but AVX2 should work.