MAGMA Forum

It looks like I can call cublas inside the GPU kernel.

Here is the example from CUDA SDK, although it only creates one GPU thread to call the GPU kernel.
/usr/local/cuda/samples/7_CUDALibraries/simpleDevLibCUBLAS

However, I am wondering if I can do the same for MAGMA functions?

You can call kernels from within kernels by using dynamic parallelism: http://developer.download.nvidia.com/as ... n_CUDA.pdf

Should work for MAGMA kernels too