When does CUFFT allocate memory? I had always assumed that it was when the plan was created but I’ve just noticed that if I create a plan and then use up all of the video memory with other allocations then when I try to execute the plan it fails with CUFFT_EXEC_FAILED. However, if I execute the plan once before making the other allocations then it works (as do all subsequent executions). This leads me to conclude that the memory is not actually allocated until the first time the plan executed. Is this correct? Does this also mean that if the FFT is out-of-place then it doesn’t allocate any memory at all?
Related topics
Topic | Replies | Views | Activity | |
---|---|---|---|---|
CUFFT and shared memory | 0 | 2039 | May 18, 2011 | |
Reduced memory on GPU when being used by X11 | 1 | 887 | October 19, 2010 | |
CUFFT memory usage | 0 | 3773 | August 17, 2008 | |
FFT + constant memory cudaFFT uses too much? | 0 | 1492 | June 11, 2008 | |
cufft memory footprint How much memory does cufft need? | 1 | 2810 | October 15, 2008 | |
CUFFT memory usage | 1 | 1371 | May 6, 2011 | |
CUFFT with mapped memory in CUDA 2.2 | 0 | 1515 | June 10, 2009 | |
wasteful memory usage off CUFFT routines cufftPlans taking up unnecessary amounts of space | 2 | 1234 | March 15, 2012 | |
Why are CUFFT plans so large, and what can be done about this? | 3 | 887 | February 20, 2018 | |
memory allocation question | 0 | 1327 | February 25, 2010 |