Bug Description

The following code fails when a version of Warp built with the CUDA 11 toolkit (e.g. 11.8) is used:

import warp as wp
import warp.sim.integrator_vbd

wp.init()

wp.load_module("warp.sim.integrator_vbd")

Error:

wp_warp.sim.integrator_vbd_41ba571.cu(12357): Error: Formal parameter space overflowed (5914 bytes required, max 4096 bytes allowed) in function VBD_solve_trimesh_with_self_contact_penetration_free_5b134ddc_cuda_kernel_backward

The issue is the kernel parameter limit, which was increased in CUDA 12.1: https://developer.nvidia.com/blog/cuda-12-1-supports-large-kernel-parameters/

System Information

Ubuntu 24.04.01 LTS