Open discussion for MAGMA library (Matrix Algebra on GPU and Multicore Architectures)
For each function there are two versions of the testing code. For example, testing_sgeqrf.cpp and testing_sgeqrf_gpu.cpp. My understanding is they both use the GPU to do the calculation. What is the difference between them?
- Posts: 5
- Joined: Tue Aug 09, 2011 3:42 am
The difference is in the interface (and the implementation). The _gpu files test the _gpu routines. These routines take input matrices on the GPU memory and produce the result on the GPU memory. The routines without _gpu are CPU LAPACK compliant interface - input matrices and output is on the CPU memory. Both use the GPU for calculations. The GPU interface routines are in general slightly faster.
- Posts: 255
- Joined: Fri Aug 21, 2009 10:39 pm
Return to User discussion
Who is online
Users browsing this forum: Bing [Bot], Google [Bot] and 2 guests