How to correctly call the official CUDA module of ceres?

[LYK0520]

options.dense_linear_algebra_library_type = ceres::CUDA;
To call cuda, only the simple code above is needed to implement the three methods, respectively the DENSE_QR, DENSE_NORMAL_CHOLESKY and DENSE_SCHUR.
It is worth noting that without the line of code,the program runs normally and caculates the correct result. But after adding the call code, DENSE_QR,DENSE_NORMAL_CHOLWSKY will report an error, DENSE_SCHUR can run but the gpu is not called. I have confirmed that I have cuda installed.

Detailed examples of cuda calls are nowhere to be found in official documentation, and there doesn't seem to be much third-party explanation online. Is there a complete case explaining the call? I saw the official TEST in cuda_dense_cholesky_test.cc, but it was hard to understand.
If I have the honor to receive your reply, I will be very grateful! Thanks again!

[sandwichmaker]

you should not have to do anything other than set
Solver::Options::dense_linear_algebra_library = ceres::CUDA.

What error are you getting when you enable CUDA?

[LYK0520]

I'm sorry I didn't follow you. The program runs smoothly when we don't use cuda; After cuda is used, only schur does not report errors, but also shows that the gpu usage is 0.
I'm sure I only added the line, but the cpu usage is always high and the gpu usage is 0. The IDE I use is 2022 visual studio. Is there any other configuration required for the program?

[sandwichmaker]

Have you tried using cuda with a smaller number of images?

…

On Thu, Feb 1, 2024, 7:08 PM LYK0520 ***@***.***> wrote: I'm sorry I didn't follow you. The program runs smoothly when we don't use cuda; After cuda is used, only schur does not report errors, but also shows that the gpu usage is 0. I'm sure I only added the line, but the cpu usage is always high and the gpu usage is 0. The IDE I use is 2022 visual studio. Is there any other configuration required for the program? — Reply to this email directly, view it on GitHub <#1045 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AANCABMWIB6MP4J6K2IY3TDYRRKDLAVCNFSM6AAAAABCKDTJBSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSMRSG4ZDOMBYHA> . You are receiving this because you were assigned.Message ID: ***@***.***>

[838619643]

I have a similar problem. Looking forward to the official answer.

[sandwichmaker]

I think since you are trying to use this as part of colmap, I think you should make sure that the colmap build environment is correct. It is hard for me to diagnose this problem from here.

…

On Sat, Feb 3, 2024, 5:34 PM 838619643 ***@***.***> wrote: I have a similar problem. Looking forward to the official answer. — Reply to this email directly, view it on GitHub <#1045 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AANCABJBQJ4RDD3TJWRNQBDYR3QQVAVCNFSM6AAAAABCKDTJBSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSMRVGUZDCMRSGM> . You are receiving this because you were assigned.Message ID: ***@***.***>

[AlexanderAKalinkin]

Hi Guys,
Have you ever thought about adding cuDSS as Direct Sparse Solver for GPU?
Thanks,