Hi,

2012/1/16 Janne Grunau <[email protected]>:
> Overall almost 4% faster, idct_add down from 340 to 92 cycles, idct_dc_add
> down from 96 to 33 cycles.

Nice. I'm glad all those changes are now paying off. Although a bit
off-topic, do you have figures comparing before the changes since
dequant modifications (but after neon optimizations for it) and now?

Another optimization would be to detect when only a few AC coeffs are
set (lower frequency or horizontal-only or vertical-only) and have
specific transforms for that. It would pay quite less though.

(please don't hesitate to break this to a new thread)

Best regards,
Christophe
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel

Reply via email to