Reduce cuFFT memory usage by re-using the memory space between the forward and backward FFT operations #3
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This is achieved by sharing the memory space between the cuFFT operations (since they cannot be executed in parallel), and only allocated the memory needed by the largest operation (though in this case they -should- be the same References https://docs.nvidia.com/cuda/cufft/index.html\#function-cufftgetsizemany https://docs.nvidia.com/cuda/cufft/index.html\#function-cufftsetautoallocation + further functions in 3.7
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Hey Cees,
Not the patch I was intending to upstream next, but I've been having endless issues with OpenMP forcing itself to be serial, so here's a quick win I came across.
cuFFT allows you to manage your memory usage yourself, so here we can only allocate the larger memory block needed for the cuFFT operation rather than allocating two separate blocks for each FFT. This saves me around 2GB in VRAM in my normal configuration, and I ran a test to compare it to what you've mentioned in the past (20 subbands I believe?) and it should be ~1GB saved in your case, so you could increase nforward or sample even more DMs
Cheers,
David