Hi, Thanks for taking a look.
The patch itself uses only NEON (A64 ASIMD) instructions -- tbl, smlal, sqshrn etc. on 128-bit vectors -- no SVE instructions at all, so the SVE revision does not affect the patch. The CPU line in the commit message was just environment info for the checkasm --bench numbers. For the record, Kunpeng 950 (TaiShan V150 cores, ARMv9) supports both SVE1 and SVE2 with a 256-bit vector length; the benchmark was measured on that box. As suggested, the patch is now also submitted through the Forgejo instance: https://code.ffmpeg.org/FFmpeg/FFmpeg/pulls/25799 Best regards, zhujing _______________________________________________ ffmpeg-devel mailing list -- [email protected] To unsubscribe send an email to [email protected]
