Matrix Support about thrust HOT 8 CLOSED

Marking as request for enhancement.

thrust is very efficient, but our project need to manipulate the matrices, i 
have to drop it, sad, hoping thrust could native support matrices in future 

I've been working on device_matrix following on an interface of Eigen 3.x. 
device_matrix will wrap the culbas for matrix operation and use expression 
template to enhance the performance. The job is just starting. 

Sun, that sounds like a great idea. Please keep us informed about your progress 
and if there's anything we can do to help!

Than you for your interests! I just finished wraping some matrix-matrix and 
matrix-scalar functions of CUBLAS using expression template. I will wrap 
matrix-vector functions in next few days and complete the matrix operation 
expression interface. Array operation expression interfaces will be added 
later. Matrix sub-blocks operation is also an important task. 

However, current wrap is simple and not efficient enough. Optimization is the 
main objective of next stage. Unfolding the expresson template in the kernel 
function may bring better performance than wraping CUBLAS for plus and minus 
operation. 

I've just published GPUMatrix 0.1 at http://code.google.com/p/gpumatrix/. I'm 
working on my research project so the code is somewhat rough and the document 
is absent. But it do help in my research project. I can switch computation 
between GPU and CPU just define the MatrixType to GPUMatrix or Eigen::MatrixXd. 
I wish it can help. Any suggestion is welcome. 

Best wishes!