On Mon, 19 Dec 2011, Nadi Findikli wrote:
> I am trying to estimate if my processor can perform image operations in a reasonable amount of time. Are
> there any estimates on how many processor cycles certain operations take (e.g. change image size, draw
> text, etc)? I think this is more useful than MIPS estimates but I will take those too :^)
There is really no substitute for testing with your own image files
with the exact commands you plan to use.
Some compressed file formats like JPEG and PNG can be slow to read and
write so reading and writing the file can take more time than the
image processing operation itself.
As far as image processing algorithms go, you can learn a lot from the
GraphicsMagick OpenMP results page at
http://www.graphicsmagick.org/OpenMP.html, which includes total
iterations/second result values on different CPUs (some of them quite
old) for many different requests.
The results on this page are a bit dated. I have additional results
which have not yet been posted yet for 12 cores (24 threads) of modern
Intel Xeon E5649 CPU and 16-cores of AMD Opteron 6220 CPU. The good
news is that algorithms which scaled well with four cores continue to
scale well with many more cores. For example, one algorithm achieved
20X speedup on the Intel Xeon E5649 CPU.
Something I have discovered recently is that GraphicsMagick
performance may be considerably better with alternative compilers like
'clang' or 'Open64' than with GCC.
GraphicsMagick Maintainer, http://www.GraphicsMagick.org/