I’m trying to compile some kernels on Linux (Ubuntu 10.04) that I previously compiled on Windows 7 64-bit. The kernels are fairly large, and compiling sometimes passes and sometimes results in either a segmentation fault or *** glibc detected *** ptxas: free(): invalid pointer.
If compile succeeds, some kernels run but two kernels fail with unspecified launch failure. Commenting out lines can cause them to run, so it could be a size problem.
Does anyone know any differences between CUDA on Windows and Linux that could explain the problem?
I’m using CUDA Toolkit 3.2 on both operating systems.
nvcc: NVIDIA ® Cuda compiler driver
Copyright © 2005-2010 NVIDIA Corporation
Built on Wed_Nov__3_16:16:57_PDT_2010
Cuda compilation tools, release 3.2, V0.2.1221
gcc (Ubuntu 4.4.3-4ubuntu5) 4.4.3
Copyright © 2009 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.