A Unified GPGPU API In Gallium3D?
Zack Rusin today has covered on his blog GPGPU and ultimately answers the question as to whether there will be a GPGPU API within the new Gallium3D architecture. GPGPU, or General Purpose computing on GPUs, has been a hot topic as of late with both NVIDIA and ATI/AMD having their own SDK/APIs and the latest graphics cards offering dozens of stream processors. NVIDIA's general purpose technology is CUDA (Compute Unified Device Architecture) while ATI's is called CTM (Close to Metal).
Zack explains though in this blog entry that Gallium3D is not the place for a unified Linux GPGPU API as this Tungsten Graphics architecture is too low-level. However, a unified GPGPU API could be written above Gallium3D using a state tracker. Zack adds that as Gallium3D now uses LLVM (Low Level Virtual Machine) for compiling and accelerating shaders, any programming language could be put on the GPU with "minimal effort."
For more on the Tungsten's Gallium3D architecture for graphics drivers, check out their Wiki page.
Zack explains though in this blog entry that Gallium3D is not the place for a unified Linux GPGPU API as this Tungsten Graphics architecture is too low-level. However, a unified GPGPU API could be written above Gallium3D using a state tracker. Zack adds that as Gallium3D now uses LLVM (Low Level Virtual Machine) for compiling and accelerating shaders, any programming language could be put on the GPU with "minimal effort."
For more on the Tungsten's Gallium3D architecture for graphics drivers, check out their Wiki page.
3 Comments