[petsc-dev] ASM for each field solve on GPUs

[Jed Brown]

Barry Smith <bsmith at petsc.dev> writes:

>   I was assuming the triangular systems were all solvable and setup in the correct format so the triangular solver would not be the place to first detect problems with the factor. Though not completely generally that seems to cover most cases.

You could write a solver that way, but that doesn't appear to be what cusparse has done. Perhaps it's because the function takes the raw arrays as inputs (rather than an opaque structure that can only be created by an analysis phase that confirms validity). For all we know, they might do 90% of the work synchronously and just leave the last kernel in the stream.