Cuda-x86 details

Hi All,

I am considering to try Cuda-x86 for some project, and I would like to get some more information on it. So could anybody answer these questions?

  • What warp size, and what compute capabiliy is used? Can I set it somehow?
  • Is the warp size dependent on the CPU architecture?
  • What compute capability is shown for CPUs? Can I change it somehow?
  • How is the machine code vectorised? Are you using tricks similar to those described here for Intel OpenCL?

Hi icchan,

You might find this article useful: http://www.pgroup.com/lit/articles/insider/v3n2a1.htm

What warp size, and what compute capabiliy is used? Can I set it somehow?

Warpsize is 1 and the CC is 99. No, you can not change these settings.

Is the warp size dependent on the CPU architecture?

No.

What compute capability is shown for CPUs? Can I change it somehow?

CC is 99, and no you can’t change it.

How is the machine code vectorised?

Classical compiler auto-vectorization using SSE/AVX.

  • Mat