Skip to content

Add NEON for highbd inv_txfm - #5205

Merged
leolzhao merged 1 commit into
AOMediaCodec:mainfrom
jjustiss-apple:jjustiss/neon-inv-txfm
Aug 6, 2026
Merged

Add NEON for highbd inv_txfm#5205
leolzhao merged 1 commit into
AOMediaCodec:mainfrom
jjustiss-apple:jjustiss/neon-inv-txfm

Conversation

@jjustiss-apple

@jjustiss-apple jjustiss-apple commented Aug 4, 2026

Copy link
Copy Markdown
Contributor

Add NEON intrinsics for the full inv_txfm pipeline covering all 1D kernels
(DCT, ADST, FDST, IDTX) for sizes 4-32, using matrix-multiply for sizes
4/8/16 and loop-based for size 32.

CTC Results (RA, cpu-used=1, 33 frames, A5):

Metric Delta
Encode time -1.18%
BD-rate Y 0.000%
BD-rate Cb 0.000%
BD-rate Cr 0.000%

Bit-exact confirmed (all PSNR values identical). Unit tests included.

@jjustiss-apple
jjustiss-apple force-pushed the jjustiss/neon-inv-txfm branch 2 times, most recently from 4375f93 to 24c9503 Compare August 4, 2026 20:15
@leolzhao
leolzhao force-pushed the jjustiss/neon-inv-txfm branch from d3d90e9 to d53a1fa Compare August 6, 2026 22:34
@leolzhao
leolzhao merged commit 2969f22 into AOMediaCodec:main Aug 6, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants