/CUDA

Modules to expose CUDA runtime functions

Primary LanguageCMake

SAC CUDA Module

About

This is a SaC module that wraps around some of the runtime functions of CUDA.

Its main purpose is to provide profiling functions.

Build Instructions

You'll need to have installed sac2c and have a copy of the Stdlib installed as well. sac2c will need to have at the very least a functioning backend for CUDA, e.g. target cuda and friends.

$ cd CUDA
$ git submodule init
$ git submodule update
$ mkdir build
$ cd build
$ cmake ..
$ make -j4
$ make install //you might need to use `sudo'

Variables that can be passed to CMake

When running CMake it is possible to pass the following variables:

  • -DTARGETS=x;y;z --- build stdlib for targets x, y and z. (defaults are cuda)
  • -DSAC2C_EXEC=/path/to/sac2c --- specify sac2c executable directly. Otherwise CMake will try to find sac2c on yout PATH.
  • -DLINKSETSIZE=n --- set -linksetsize n when calling sac2c. This option is responsible for the number of C functions that are put in a single C file when compiling a SaC program. The rule of thumb:

    • value 0 is the fastest time-wise but potentially results in a large memory consumption

    • value 1 reduces the memory consumption to minimum, buy significantly increases compilation time.

      Default value: 500.

  • -DENABLE_CUDA_CHECK=ON activates CUDA runtime checks, default is OFF.