SPU LLVM: Use VDBPSADBW in SUMB #10937
Merged
Conversation
This file contains 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
- This instruction can be used to sum bytes horrizontally if the second input vector is all zeroes.
- The first element can be extracted via vmovd rather than vpextrd, which saves 1 uop.
|
Added a commit to optimize some branches following a byteswap. |
|
Hmm, How about new syntax for intrinsics? See fre/fmax. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
I overlooked this instruction earlier when optimizing the VNNI path for SUMB.
This AVX-512 instruction first calculates the absolute difference between 2 vectors before summing bytes horizontally. By using a vector of all zeroes, we can effectively use the instruction to just sum bytes horizontally.
This instruction should be faster than the VNNI alternative since we don't need to zero out the destination register, and we don't need to load a vector full of the constant 0x01. Additionally, when op.ra == op.rb, we only need a single instruction for the correct behavior.