[x264-devel] unaligned accesses in IA64

[Mauricio Alvarez]

Hi,

I'm testing X264 on an Itanium machine . I want to test the parallel
scalability of the threading model implemented in X264.

The machine is like this:
- 64 cpus Dual Core Montecito(IA-64). Each one of the 128 cores works
at 1,6 GHz, with a 8MB L3 cache and 533 MHz Bus.
- 512 GB RAM (8 GB per blade).
- OS: SuSe Linux 2.6.16.27-0.9

I have compiled the code using both gcc (4.1.0) and icc (v 10.0), and
I execute it using the following command line:

./x264 --qp=20 --fps 25 --frames 100 --progress  --threads=${NTHREADS}
-o 1088p25_blue_sky.h264 1088p25_blue_sky.yuv 1920x1088

And, I'm getting these results, with some messages about unaligned accesses:
(these results are for the version compiled with icc)
x264 [info]: using cpu capabilities: none!
x264(8689): unaligned access to 0x607fffffff32ee5c, ip=0x40000000000afb71
x264(8689): unaligned access to 0x607fffffff32ee64, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee6c, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee74, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee7c, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee5c, ip=0x40000000000afb71
x264(8689): unaligned access to 0x607fffffff32ee64, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee6c, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee74, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee7c, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee5c, ip=0x40000000000afb71
x264(8689): unaligned access to 0x607fffffff32ee64, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee6c, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee74, ip=0x40000000000aff31
x264(8689): unaligned access to 0x607fffffff32ee7c, ip=0x40000000000aff31
x264 [info]: slice I:1     Avg QP:17.00  size:293253  PSNR Mean
Y:46.51 U:48.54 V:49.18 Avg:47.16 Global:47.16
x264 [info]: slice P:9     Avg QP:20.00  size: 53870  PSNR Mean
Y:44.51 U:45.50 V:46.23 Avg:44.91 Global:44.91
x264 [info]: mb I  I16..4: 56.6%  0.0% 43.4%
x264 [info]: mb P  I16..4: 25.7%  0.0%  0.4%  P16..4: 24.1% 14.9%
5.1%  0.0%  0.0%    skip:29.7%
x264 [info]: SSIM Mean Y:0.9813082
x264 [info]: PSNR Mean Y:44.709 U:45.806 V:46.523 Avg:45.136
Global:45.090 kb/s:15561.58

encoded 10 frames, 0.81 fps, 15570.36 kb/s

I found a previous thread related to this problem, but seems that
there was no solution to it:
http://www.via.ecp.fr/via/ml/x264-devel/2006-04/msg00069.html

Do you have any suggestion on how to solve this issue? At the end the
video is coded correctly but I suppose that having those unaligned
accesses have a big performance penalty.

Thanks in advance,

Mauricio Alvarez