Am 11.11.2013 16:26, schrieb Michael Born: > Hi Ullrich, > > Benchmarking the file export is easy. I just have to parse the darktable > -d perf output for lines with [dev_process_export] to get useful values. > How would you benchmark the responsiveness of darktable? There are so > many values in the debug output - I don't see which of them would > represent responsiveness. > > Michael
Hi Michael, I only look at the total time spent in the pixelpipe as the single module figures are off when you run with OpenCL. So look for a line like [dev_process_image] pixel pipeline processing took 0,129 secs (0,422 CPU) In this case 0.129 seconds is the important figure. If you want to have more info about the OpenCL specific timing you may call darktable with '-d opencl -d perf'. You will get after each pixelpipe run the time spent in each of the kernels, e.g: [opencl_profiling] spent 0,0123 seconds in [Write Image (from host to device)] [opencl_profiling] spent 0,0002 seconds in denoiseprofile_precondition [opencl_profiling] spent 0,0011 seconds in denoiseprofile_decompose [opencl_profiling] spent 0,0010 seconds in denoiseprofile_reduce_first [opencl_profiling] spent 0,0000 seconds in denoiseprofile_reduce_second [opencl_profiling] spent 0,0055 seconds in [Read Buffer (from device to host)] [opencl_profiling] spent 0,0007 seconds in denoiseprofile_synthesize [opencl_profiling] spent 0,0002 seconds in denoiseprofile_backtransform [opencl_profiling] spent 0,0010 seconds in blendop_mask_rgb [opencl_profiling] spent 0,0003 seconds in blendop_rgb [opencl_profiling] spent 0,0003 seconds in colorin [opencl_profiling] spent 0,0003 seconds in colorout [opencl_profiling] spent 0,0002 seconds in [Copy Image (on device)] [opencl_profiling] spent 0,0056 seconds in [Read Image (from device to host)] [opencl_profiling] spent 0,0288 seconds totally in command queue (with 0 events missing) Ulrich ------------------------------------------------------------------------------ November Webinars for C, C++, Fortran Developers Accelerate application performance with scalable programming models. Explore techniques for threading, error checking, porting, and tuning. Get the most from the latest Intel processors and coprocessors. See abstracts and register http://pubads.g.doubleclick.net/gampad/clk?id=60136231&iu=/4140/ostg.clktrk _______________________________________________ Darktable-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/darktable-users
