Closed twobombs closed 3 months ago
Sorry for this. What version of CUDA compiler and host compiler are you using?
thank you for the reply : I use the cuda:12.3.2-devel-ubuntu22.04 image with the following nvcc and gcc specifications :
Ok, I understand the issue here. I see from the logs you are building for Maxwell, that's a corner case we hadn't covered. Is Maxwell a platform you care about for QUDA.
I can make a fix for this and will file a PR soon.
Noting that there's an easy work around for the issue you've encountered: pass the option -DQUDA_LARGE_KERNEL_ARG=OFF
to CMake when configuring QUDA. This will fix the issue you are seeing. I will ensure that this is the default on Maxwell and Pascal architectures however going forward.
tnx - it builds with this setting :)
Usually builds without problems: this weekend the docker build barn that puts out a quda docker image variant of thereminq-tensors gave out this: