4

Is there a tool to speed up and distribute compilation of GPGPU programs using CUDA C (where nvcc compiler is part of compilation chain), i.e. something like distcc + ccache (or derivatives and wrappers like distcc-pump, ccontrol, Icecream (IceCC) or DMUCS) for C / C++ and GCC compiler?

Or alternatively is it possible to configure distcc for use with nvcc?

Jakub Narębski
  • 309,089
  • 65
  • 217
  • 230

1 Answers1

2

Unfortunately I don't think it's at all straightforward to make distcc work with non-gcc compilers. It's not impossible, just a nuisance because you have to go modify the distcc implementation to understand the new compiler's options, as mentioned on the distcc FAQ.

An alternative is to use something like Electric Make, which solves the same problem (accelerating builds) in roughly the same way (distribution and parallelization), but at the level of the build tool instead of the compiler. That means that it's completely indifferent to the particular compiler that you use. You can get a free version from http://www.electric-cloud.com/eade.

Disclaimer: I am the Chief Architect of Electric Make

Eric Melski
  • 16,432
  • 3
  • 38
  • 52
  • When I try to download ElectricFlow/Make/whatever, it says 1.7 GB. Also, I don't see a Debian repository link anywhere, just a stand-alone download. – einpoklum May 17 '16 at 12:35
  • ElectricFlow is a different product. The download for ElectricAccelerator Huddle, the free version of ElectricAccelerator which includes Electric Make, is only about 100MB on Linux. No, there is no .deb package for it. – Eric Melski May 17 '16 at 17:03