GPU-accel matrix-free "forward map" routines

Front page Forums General GPU-accel matrix-free "forward map" routines

This topic contains 0 replies, has 1 voice, and was last updated by  eulersdilemma 6 months, 3 weeks ago.

Viewing 1 post (of 1 total)
  • Author
    Posts
  • #1496

    I’m interested using Paralution’s GMRES solver for a course project; the linear system in question has no storage requirements but its action (matvec multiplication) is implicitly computed at every call. That is, I do not want to form the matrix. This appears to be what the LocalStencil class is for, if I am not misunderstanding things. Could someone confirm that this is what I want? That is, I’ll have to implement a new gpu-side LocalStencil for my forward map.

    My stencil routine will call cuFFT. Since cuFFT only allows host-side calls, will I still be fine here? A cursory glance at the code indicates the answer is in the affirmative but I would like some help confirming.

    Or is there a better way to accomplish what I want to do?

    Thanks,

Viewing 1 post (of 1 total)

You must be logged in to reply to this topic.