Can cublasAlloc() be forced to pin the memory it allocates? If not, is it legal to use the cudaAlloc/cudaFree functions directly to manage memory that will be used by cublas kernels?
mfatica
2
You can use cudaMallocHost to allocate pinned memory that is going to be used by cublas kernel.
cublasAlloc will return non-pinned memory.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| do i need cublaAlloc after cudaMalloc to use cublas? | 5 | 3001 | September 26, 2010 | |
| can I use pinned memory? | 6 | 2749 | September 21, 2009 | |
| cudaHostAlloc and thread safety problems with pinned, portable memory | 2 | 1914 | April 8, 2011 | |
| malloc() + cuMemHostRegister() faster than cuMemAllocHost() | 0 | 1140 | October 9, 2013 | |
| check for cudaHostAlloc Portable possibility | 13 | 2985 | July 1, 2015 | |
| cudaMallocHost | 3 | 2847 | June 8, 2011 | |
| DriverAPI: How to set the data to an pinned memory. | 1 | 935 | May 7, 2009 | |
| Is pinned memory possible in mixed cpp and cuda | 3 | 2921 | January 29, 2009 | |
| fine control of memory pinning in CUDA | 12 | 16847 | May 1, 2008 | |
| CUBLAS memory management cublasAlloc and cublasSetMatrix/cublasGetMatrix | 0 | 2079 | May 24, 2009 |