Computer Science – Performance
Scientific paper
Mar 2012
adsabs.harvard.edu/cgi-bin/nph-data_query?bibcode=2012aps..apr.e1032b&link_type=abstract
American Physical Society, APS April Meeting 2012, March 31-Apr 3, 2012, abstract #E1.032
Computer Science
Performance
Scientific paper
In conservative numerical schemes for general relativistic magnetohydrodynamics (GRMHD), a procedure is required to convert between the ``conserved" variables and the ``primitive" variables (density, velocity, internal energy) at each time integration step. However, there is no explicit form for the recovery of the primitive variables from the conserved variables, and a system of non-linear equations must therefore be solved numerically with an iterative approach. Primitive variable recovery lies at the heart of GRMHD schemes, and its cost is of great concern to large-scale simulations. We investigate the use of the OpenCL framework to accelerate primitive variable recovery. We describe our parallelization approach, and present our performance results on both the CPU and GPU.
Boydstun Kristen
Moesta Philipp
Ott Christian David
No associations
LandOfFree
OpenCL-Accelerated Primitive Variable Recovery for General Relativistic Magnetohydrodynamics does not yet have a rating. At this time, there are no reviews or comments for this scientific paper.
If you have personal experience with OpenCL-Accelerated Primitive Variable Recovery for General Relativistic Magnetohydrodynamics, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and OpenCL-Accelerated Primitive Variable Recovery for General Relativistic Magnetohydrodynamics will most certainly appreciate the feedback.
Profile ID: LFWR-SCP-O-1365706