[mythtv] [PATCH] Altivec optimizations

Jeremiah Morris jm at whpress.com
Mon Dec 6 14:52:22 UTC 2004


On 5 Dec 2004, at 9:38 PM, Nigel Pearson wrote:

> P.S. Does anyone understand why there isn't
> a default X86 implementation of kerneldeint?
> I can understand the code preforming badly on
> non SSE CPUs, but crashing is a bit extreme?

If it requires SSE, maybe it should check the mm_support return value 
for MM_SSE rather than MM_MMX on line 436.

Also, why do the filters each have their own, 
almost-but-not-quite-identical copies of the mm_support() function and 
constants?  Can this be moved to libmyth, where anyone can access it?  
(It would be nice to use this in yuv2rgb.cpp, which has yet another set 
of MMX checks.)

I admit I don't quite see the logic of the filter design in general, 
with the struct and function pointers forming a poor man's C++.  Is the 
overhead of this really that much less than C++ to justify the hassle?  
If that were true, then why is everything else in Myth comfortable with 
using C++ then?  Working in the linearblend filter felt like a 
completely different environment from Myth, and not one I'd like to 
revisit.

- Jeremiah





More information about the mythtv-dev mailing list