Hardware-accelerated inverse floating point square root #156
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
|
Apologies, this was not intended to be two commits. |
|
There's potentially a bug in the Inf or NaN branch; I'll add a test for it to the other PR and check if the fix suggested by aikku works. |
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Outdated suggestions cannot be applied.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.
https://github.com/Aikku93 wrote a software divide that can keep up with the square root unit and wrote this implementation in assembly.
They asked me to make a pull request for it. Draft for now because I need to update the hw_math test also.
Some general information on the characteristics of this function:
Execution time seems to be about 37 bus cycles on dsi, 61-62 bus cycles on ds.
Accuracy should be within 1 ulp of the result you would get when doing the calculation with doubles and then casting to single precision, which should more accurate than a simple 1.0f/sqrtf(x).
The result is exact (well, identical) in ~89 % of possible mantissas.
Considerations: In this version the LUT is in main memory, but it may be desirable to place it in tcm on ds. I believe on dsi it doesnt matter where you put it because the overall execution time is limited by the hw square root. Alignment to cache lines could also be a consideration.