When benchmarking GPU algorithms it is important to measure kernel execution time separated from memory transfer times (CPU->GPU and GPU->CPU). Is it possible to measure these times separately for existing algorithms (e.g., GPUBinaryThresholdImageFilter?). Thanks in advance.
Hi @Rafael_Palomar ,
Welcome to the ITK community!
To separate the memory transfer times, call
UpdateBuffers() on the images.
Here is an example:
2 posts were split to a new topic: GPU support in Python packages
Thank you, @matt.mccormick! That solved the issue.