Make sve_128 portable and support true march=native builds #504
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
solidpixel
force-pushed
the
sve_native
branch
from
7b003a2
to
fe8ab6b
Compare
solidpixel
force-pushed
the
sve_native
branch
from
be65437
to
e5f4247
Compare
solidpixel
force-pushed
the
sve_native
branch
from
e5f4247
to
b3f192b
Compare
bengaineyarm
reviewed
Oct 8, 2024
| #define ASTCENC_SVE 8 | ||
| // Auto-detected SVE can only assume vector width of 4 is available, but | ||
| // must also allow for hardware being longer and so all use of intrinsics | ||
| // must explicitly use predicate masks to limit to 4-wide. |
There was a problem hiding this comment.
I suppose we could (in future) do the same for >8wide? (with ASTENC_SVE 8)
bengaineyarm
approved these changes
Oct 8, 2024
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Outdated suggestions cannot be applied.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.
This PR makes two changes to improve support for SVE.
The first change makes the sve_128 build portable to any system implementing SVE by removing the fixed vector length at compile time. This code is merely augmented NEON, using native gathers, and will use predicates to force the 128-bit data path no matter what the host vector size is. Testing on Graviton 4 shows that there is no measurable performance difference between portable sve_128 and compile-time enforced sve_128, and testing on Graviton 3 shows that the portability part works fine.
The second change makes the "native" build type actually use
-march=nativeand-mcpu=native, allowing native builds to pick up SVE if the host CPU supports it, as currently native builds rely on compiler defaults which are baseline Arm-v8. Note that this needs Clang-18 or newer to pick up SVE automatically, which is newer than the default compiler on current AWS Linux images for Graviton 3 and 4.