Hi A. P. Rajshekhar: During the TSM we had discussions about the GPU programmability.
I have been trying to find out more and stumbled upon ARB - OpenGL Assembly Language which is a low-level Shader language. Please take a look at http://en.wikipedia.org/wiki/ARB_(GPU_assembly_language) This is described as - "It was created by the OpenGL ARB to standardize GPU instructions controlling the hardware graphics pipeline." While this is true that at OpenGL level, the usage of instructions is for graphics related matrix computations. This is evident to some extent from the instructions mentioned. My queries are - 1. The instruction as shown seem more like an artifact created to hide the "actual" instruction set that GPUs export. 2. How would one write an matrix computation like A1 := T x A0 A2 := T x A1 My intention is to try an run an 'Monte Carlo' or 'Genetic Algorithm' implementation on the GPU, while the CPU is free to do other tasks. 3. There has been some discussion about stream computing being possible on GPUs. 4. The nVidia slides talk about 65.536 instructions. What are these instructions ? Appreciate your suggestions and opinion in this matter. thanks Saifi.
