Hello, great work on this project - I'm a big fan. :)
I am trying to benchmark Magick.NET performance against GDI+, and have written a test harness that loads, resizes and writes images in parallel. When I run Magick.NET tests with OpenCL disabled, performance is great and I get no errors. However with OpenCL enabled I get AccessViolationException exceptions after a hundred or so images. I have encountered this with 2 different GPUs, on a laptop (Intel HD Graphics 4000) and on an AWS G2 instance (NVIDIA GRID K520).
I'm attaching my code to this issue. I'm using a test set of 11,000 jpegs from http://apod.nasa.gov/
If you need more information I will do my best to oblige! Thanks again.
Comments: Thanks for the detailed information. I suspect that this is a threading issue inside the OpenCL part. Is your problem gone when you change the Parallel.ForEach to a 'normal' for each? And could you add a link to the APOD archive? I cannot download the whole archive from the site and I think I need a large set to reproduce this ussue. p.s. Please contact me through CodePlex after we fixed this and you have the benchmarks. Really curious about the results :)
I am trying to benchmark Magick.NET performance against GDI+, and have written a test harness that loads, resizes and writes images in parallel. When I run Magick.NET tests with OpenCL disabled, performance is great and I get no errors. However with OpenCL enabled I get AccessViolationException exceptions after a hundred or so images. I have encountered this with 2 different GPUs, on a laptop (Intel HD Graphics 4000) and on an AWS G2 instance (NVIDIA GRID K520).
I'm attaching my code to this issue. I'm using a test set of 11,000 jpegs from http://apod.nasa.gov/
If you need more information I will do my best to oblige! Thanks again.
Comments: Thanks for the detailed information. I suspect that this is a threading issue inside the OpenCL part. Is your problem gone when you change the Parallel.ForEach to a 'normal' for each? And could you add a link to the APOD archive? I cannot download the whole archive from the site and I think I need a large set to reproduce this ussue. p.s. Please contact me through CodePlex after we fixed this and you have the benchmarks. Really curious about the results :)