[XviD-devel] iA64-Patch for dev-api-4

[Stephan Krause]

Hi,

one tiny thing left thas crashes xvid on iA64 (same thing as everytimes: 
stride is uint32_t so the whole expression is uint32_t and brakes src if 
it is negative)

(I hope the patch is not line-wrapped, but it should be only a single 
line of code)
,------------------------------------------

Index: src/image/interpolate8x8.h
===================================================================
RCS file: /xvid/xvidcore/src/image/interpolate8x8.h,v
retrieving revision 1.10.2.3
diff -r1.10.2.3 interpolate8x8.h
216c216
<       const uint8_t * const src = refn + (y + (dy>>1)) * stride + x + (dx>>1);
---
>       const uint8_t * const src = refn + (int) ((y + (dy>>1)) * stride + x + (dx>>1));

`-------------------------------------------

I've done a little bit of testing: Generic-C produces exactly the same 
files as my athlon at home, so that should be correct. With assembly-
optimizations the files differ (of course) but the PSNR ist quite similar
(says xvid_encraw), so that should be ok, too.

To performance: Generic-C encodes the "football"-clip from

http://www.cipr.rpi.edu/resource/sequences/sif.html

with Avg: enctime(ms) =  87.97, fps =  11.37 on a Itanium2 900
(my athlon 900 does Avg: enctime(ms) = 169.77, fps =   5.89, so 
it is nearly twice as fast), but Assemly is slower than on
my athlon 
(Itanium2: Avg: enctime(ms) =  50.91, fps =  19.64,
Athlon: Avg: enctime(ms) =  49.50, fps =  20.20)

So i thinkt there's some work to do for me yet...

Stephan
PS: Sorry for the broken build-system patch i sent you last time...

-- 
Sig fault. (core dumped)