stream benchmarking on RPi

[Jia-Shiun Li]

On Sun, Sep 8, 2013 at 4:11 AM, Zbigniew Bodek <zbb at semihalf.com> wrote:
> 2013/9/6 Jia-Shiun Li <jiashiun at gmail.com>
> Hello Jia-Shiun.
>
> I was looking for similar benchmark that will not depend on floating point
> and I found this:
> http://alasir.com/software/ramspeed/
>
> The idea seems to be the same as the stream benchmark but you have option to
> not use FP.
> I've actually made some tests on my ARM and here are the results (default
> benchmark settings):

So for now it is still not suitable for fp-intensive work. Fortunately
I guess most people won't do that on RPi.

Still, thank you all for the info and your hard work on ARM.

FYR ramspeed result on RPi. No clang results because it is clever
enough to optimize all works away even with -O. ;)

jsli at raspberry-pi:~/ramspeed-2.6.0 % time ./ramspeed -b 3 -g 1
RAMspeed (GENERIC) v2.6.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

1Gb per pass mode

INTEGER   Copy:      190.51 MB/s
INTEGER   Scale:     185.59 MB/s
INTEGER   Add:       163.58 MB/s
INTEGER   Triad:     162.43 MB/s
---
INTEGER   AVERAGE:   175.53 MB/s

61.889u 4.631s 1:09.55 95.6%    25+167k 0+0io 0pf+0w
jsli at raspberry-pi:~/ramspeed-2.6.0 % time ./ramspeed -b 6 -g 1
RAMspeed (GENERIC) v2.6.0 by Rhett M. Hollander and Paul V. Bolotoff, 2002-09

1Gb per pass mode

FL-POINT  Copy:      222.57 MB/s
FL-POINT  Scale:     28.49 MB/s
FL-POINT  Add:       65.74 MB/s
FL-POINT  Triad:     16.43 MB/s
---
FL-POINT  AVERAGE:   83.31 MB/s

317.004u 4.545s 5:36.72 95.4%   25+166k 0+0io 0pf+0w


Jia-Shiun.