Project

General

Profile

Task #2531

Feature #2054: PME on GPU

Task #2453: PME OpenCL porting effort

Consider optimizing tabulated data access on GPU

Added by Aleksei Iupinov 5 months ago. Updated 3 months ago.

Status:
New
Priority:
Low
Assignee:
-
Category:
-
Target version:
-
Difficulty:
uncategorized
Close

Description

NB/PME use 1D texture objects for lookup tables of constant data in CUDA; OpenCL implementations use raw device buffers (cl_mem).
In OpenCL, one could consider using something like image1d_t.

History

#1 Updated by Aleksei Iupinov 5 months ago

  • Description updated (diff)

#2 Updated by Aleksei Iupinov 5 months ago

  • Parent task deleted (#2453)

#3 Updated by Aleksei Iupinov 5 months ago

  • Parent task set to #2453

#4 Updated by Szilárd Páll 3 months ago

  • Priority changed from Normal to Low

Dropping priority because it would require significant amount of host code for potentially marginal benefits.

Also available in: Atom PDF