NVIDIA Thrust & Compute Capability

@ Nathan, Jared - Creators of Thrust & Other Trustable Thrust Enthusiasts,

Since Thrust applications statically compile CUDA kernels, I wonder:

  1. What is the compute capability they support?
  2. Is a thrust application portable across compute capabilities.

Please help me with your perspectives,

Always Best,

Can i use Thrust algorithms only on CU file? and not on CPP file?
This way, I can pass “-arch=sm_xx” to optimize for particular compute capability.
Can some1 confirm if this is the case, please?


You are able to use thrust commands within a CU file, I have done it. I believe the restriction is that they cannot be called from a kernel, but for all i know that may have changed, or may be changing with SM 3.5