understanding Nvidia Kepler Assembly instructions

Question

I came across this document cuobjdump.pdf.

It lists the Fermi and Kepler instruction but with no additional explanation.

Apart from the usual add, subtract, multiply, etc .. I do not get what other instruction mean or do. Can anybody help me with that or point me to a document or link that explain them ??

Answer 1

The document you linked is actually distributed with the CUDA toolkit. For example, with a standard install of the linux CUDA 5 toolkit, that document will be located in the /usr/local/cuda/doc/pdf directory. (On windows just use the file search to find that filename.)

Also in that directory (or here) is the PTX ISA document. Although PTX (an intermediate code) and the actual machine code (the output produced by cuobjdump) are not indentical, they are close enough that if you study the PTX document, you should have a pretty good understanding of the machine code instructions. Most instructions in machine code have a similarly named counterpart in PTX. For example, SHF (integer funnel shift) is documented in the PTX ISA document, and is referenced under the Kepler section of the cuobjdump document.