Am 12.05.2017 um 08:21 schrieb Alexander Rabtchevich:
Hello
I've installed a new graphics card with Radeon 580 chip (8Gb) and so
examined darktable performance. Two opearations during regular jpg
export do not use OpenCl - raw demosaic (or decompression) and applying
of final gamma. Their common time equals or even prewails a whole bunch
of all other operations. The OpenCl setting in GUI is - powerful
graphics card. Is there a way to increase the performance? Darktable is
from current git.
Absolutely! You could just write the needed OpenCL processing code for
raw denoise.
Concerning gamma: this is the last module in the pixelpipe. Latest in
that module we need all data been transferred back to CPU memory. The
time that darktable reports to spend in this module is an artifact. The
reason is that the GPU and the CPU run idepenently. The CPU code just
registers the needed GPU calls and then waits until the GPU terminates.
This waiting happens (among other places) in module gamma. If you need
to see the real timings you could set opencl_async_pixelpipe=FALSE. This
way darktable will wait for OpenCL processing to finish after each module.
Ulrich
___________________________________________________________________________
darktable developer mailing list
to unsubscribe send a mail to [email protected]
