opm-simulators icon indicating copy to clipboard operation
opm-simulators copied to clipboard

Published 20 hours ago • verified publisher icon OPM

Add support for using float Scalars in simulators

Open akva2 opened this issue 10 months ago • 2 comments

Downstream of https://github.com/OPM/opm-common/pull/4016 Downstream of https://github.com/OPM/opm-models/pull/896

These are the changes necessary to support float Scalars.

Some TODOs left:

  • UMFPack cannot handle floats. This again means MSW is disabled with floats. SuperLU supports floats so that is a possible solution for this.

Runtime has been lightly tested with the BDA gpu solvers, opencl, cusparse and amgcl (vexcl and cuda backends) and dune cpu solvers. It seems to work although there is non-uniform results (opencl seems to do better than cusparse for some reason).

akva2 avatar Apr 16 '24 12:04 akva2

okay, I have pushed a bisectable version (ie it builds at each step) incorporating your suggestions. I guess this is ready for review now.

akva2 avatar Jun 03 '24 10:06 akva2

umfpack might support single precision in the future, see DrTimothyAldenDavis/SuiteSparse#446

blattms avatar Jun 19 '24 11:06 blattms