Is there a plan to implement a linear algebra and numerical package on top of aparapi. So i can do x=A^-1 B and Aparapi can take it from there.
No plans at present.
In this example are A,B and X primitive arrays? Of course Aparapi is constrained to only allow expressions that would be allowed by the java compiler.
Maybe this would have to be explicitly handled via static helper functions.
Can I also suggest that we continue this conversation in the aparapi discussion list over at aparapi.googlecode.com. This way we can track all these ideas in one place.
like how LibJacket is. Implement a library for atleast linear algebra. It makes using CUDA drop dead simple. you never write lowlevel CUDA, you use libjacket functions in your code and everything just works.
I'll post this at google group as well.
I had not seen this library before, thanks for the link.
I would like to see a way to add library extensions to Aparapi so that such functions can be called from Kernel code. Of course to keep with the Aparapi 'write once run anywhere' Java goal we will need pure Java equivalent methods for each API and a way for the code to be dispatched behind the scenes in case OpenCL is not avaiable.
Again thanks for this.