[mythtv] OSD issue?
Andrew Mahone
andrewmahone at eml.cc
Wed Feb 11 03:30:11 EST 2004
On Wednesday 11 February 2004 00:51, Isaac Richards wrote:
> Well, I'd rather not go SSE, as I was trying to get something that'd work
> on my epia-m. Essentially, I got it working without calculating the proper
> transparency value (or touching the uv planes like it really should), tried
> to use the lookup table, saw that was really slow, and couldn't come up
> with a way to calculate the new transparency value easily enough to make it
> worthwhile. That's 'bout it =)
Ah, ok, I'll try the static reciprocal LUT idea first, then. Don't expect
anything stunning - closer to a 20-30% gain than the 300% that filters with
all-MMX math can approach.
If you still have a version around that uses the pow_lut table + MMX, it might
save me the trouble of tracing your MMX code to work which registers the
index values end up in.
Btw, any idea if the VIA chips with 3DNow! instructions support any integer
SSE? Some tests I'm doing for denoise3d show that using pinsrw and pextrw
can squeeze a little more speed out of LUT ops than I can get with movq,
probably because it saves several accesses to memory (even if it is cached).
If it proves to be a useful gain in the filter, I'll have a patch up shortly.
Not sure whether I need to check MM_MMXEXT or MM_SSE for these though - I
know some AMD chips support integer SSE ops but not the entire SSE set.
--
Andrew Mahone
andrewmahone AT eml DOT cc
More information about the mythtv-dev
mailing list