On 7/29/26 21:14, James Almer wrote:
Could you make a PR on forgejo so CI can test this?
I'll do it right now.
nit: You seem to only use m0 and m1, so make it 3, 3, 2 instead.
It's 3, 3, 6 because the ADD_DC macro uses registers m2 to m5 to add the
dc.>
psrldq or pshufd may be faster by not doing int -> float -> int
instruction switches, but probably only on older CPUs.
I'll fix it.
_______________________________________________
ffmpeg-devel mailing list -- [email protected]
To unsubscribe send an email to [email protected]
