Special math functions implemented in GPU hardware

Question

Today I found out that in the NVIDIA GPUs in the vertex unit there are special hardware functions for calculating linear interpolation in a three-dimensional regular grid. I wonder if there are more of this kind and more important if people really use them when they make GPGPUs to speed up codes.

Answer 1

There are a number of features implemented in hardware. The term you are looking for is the "internal functions of CUDA." Linear interpolation is handled by textures, something similar.

See here: http://developer.download.nvidia.com/compute/DevZone/docs/html/C/doc/CUDA_C_Programming_Guide.pdf

Internal functions are usually written with leading double underscores, such as __sin, or enabled globally with the -use_fast_math nvcc option.

And yes, they are really used quite often. :) They are a little more inaccurate from a numerical point of view, so transferring the results of one internal to another repeatedly may have an unacceptable error, depending on your use case. Testing is the key.