Interop GPU computing graphics apis stuff

we have
cuda
opencl
opengl
d3d

1.cuda to cuda
1.1.1 between context multiple devices
see shared pinned mem (host pinned for two GPUS), shared pinned host mem
(host pinned for two GPUS use as GPU mem)
optimized between contexts on same device: none existant (and memory is virtualized per context so shaders can access mem by diferents contexts in same virtual address)
We can try of CUDA 3D write textures, namely alloc 1 context set magic free alloc another context see it
1.1.2use opengl interop between devices and CUDA/OGL interop
1.2 cuda/ogl existant texture interop cuda 3.0 (no copies)
between devices optimized for Quadro/Tesla board interop (see wglaffinity)
1.3 cuda/d3d existant texture without copies. cuda 3.0
1.4 cuda/opencl mem objects from GL object by cugraphics API in cuda 3.0? of course not PTX interop (1.4 vs 1.5), not general mem see different cuda context 1.1.1, see my opencl reoprt

2.0 opencl.
1.1 ocl ocl I must see as cuda to cuda but yes pinned but not shared mem not pinned shared mem
1.2 ocl gl yes 195 cl_khr_sharing
1.3 ocl dx yes ati extension must see nvidia

3.ogl
3.1 OpenGL interop between devices
see affinity (commands send to one GPU) only windows, mac and linux yo can
see copy_image (efficient between 1context,contexts,devices)
amd similar extensions for transfering images and commands to 1 GPUs..
sync between gpus in QUadro but Radeon..
3.2 ogl dx wgl_dx_interop same device,winxp currently plans win7 vista, d3d9 currently and d3d10 plans (what d3d 11)

4.D3D 9 10 11 interop
4.1
http://code.msdn.microsoft.com/Project/Download/FileDownload.aspx?ProjectName=D3D9ExDXGISharedSurf&DownloadId=7944

TODO:
study AMD CAL,AMD OpenCL, AMD OpenGL?