Windows10 CUDA11.8 build failed

[lucasjinreal]

[bottler]

Can you copy the full output? Text is best. And there may be a log file. And please say something about your environment.

[lucasjinreal]

@bottler hhh, sorry about env info lack, my sys:

win10
CUDA11.8
pytorch2.0-stable
python3.9

pip install "git+https://github.com/facebookresearch/pytorch3d.git@stable"

I unable to catch error info everytime, seems csrc/pulsar/cuda/enderreder.construct.gpu.cu has some code un compatible with cuda11.8

[bottler]

Please copy the whole output / error message / log file / whatever you have.

[lucasjinreal]

sorry, I said I am unable to get error messages every time. Leave this as an issue if you don't want take any look at the code got error I pasted above.

[bottler]

Sorry - I just realised your error is similar to #1227 . Can you try the solution that worked there, i.e. copying a newer cub into the cuda directory?

Or, if you don't need pulsar, you could comment out the relevant call(s) to cub, which maybe requires seeing more of the output.

[lucasjinreal]

@bottler
I replaced it but still error.

I am using CUDA11.8 and tried cub 1.17 cub 1.17.1 cub 1.17.2

all fails!!

[bottler]

Did the line number change in the error? Was it referring to a location which was changed by you? Can you check if the location in the error makes sense in all the versions? I don't think I can help further, but you might be able to follow this up.

[cmy-hhxx]

@bottler hi, bottler, I've met this error.
I am using CUDA11.8 and tried cub 1.17 cub and 2.1.0 cub, copying a newer cub into the cuda directory,
but get error: namespace "thrust::cub" has no member "DebugSyncStream", can you offer some suggestions?

[cmy-hhxx]

@lucasjinreal have you fixed your problem?

[lucasjinreal]

No, op owner claims it our error, not the lib itself. I think this lib is perfect, should not have any errors