GPU support

[csp256]

What would it take to add basic GPU support for the sparse solver backend? How much of a benefit would this even be?

I have seen the OpenOF project but a ~2x speedup for switching to a GPU for linear solves did not inspire a lot of confidence in me.

[ghost]

Actually ~2x speedup is close to the state-of-the-art for sparse solvers (this is what CHOLMOD gets on average).

http://on-demand.gputechconf.com/gtc/2016/presentation/s6387-steven-rennich-gpu-acceleration-cholmod.pdf

[csp256]

Sorry for the late response.

To what extent can the problem-specific form of bundle adjustment improve upon the speed up of general-purpose sparse solvers? Could additional constraints (say, maximal number of observations per camera, or maximal number of cameras) improve upon this?

[somedadaism]

Have you seen this?
Appears to be substantially faster.
https://github.com/fixstars/cuda-bundle-adjustment

[github-actions]

This issue is stale because it has been open 60 days with no activity. Remove stale label or comment or this will be closed in 14 days.