Cuda 2.1 beta nvcc linking issue

Accelerated Computing CUDA CUDA Programming and Performance
1

I haven’t found a 100% reliable way to reproduce this yet, but occasionally nvcc fails to unlink .linkinfo. So say I run nvcc to compile a changed .cu file, the file is never processed.

[pstach@beast latsieve]$ nvcc --version

nvcc: NVIDIA (R) Cuda compiler driver

Copyright (c) 2005-2007 NVIDIA Corporation

Built on Fri_Nov__7_04:27:31_PST_2008

Cuda compilation tools, release 2.1, V0.2.1221

using driver 180.06 on CentOS 5.2 x86_64, GTX 280 & 8800GTX (not that the driver or cards matter on this one but the bug submission says to add it)

-Patrick

2

We’re going to need a reliable means of reproducing this in order to investigate.

3

I’m not doing anything special, it triggers with nvcc -o foo foo.cu or using a makefile that some the same. I don’t know what causes it, but it happens 1% of the time or so. Strace shows that it never calls unlink(2) when it does trigger.

-Patrick

4

Also to be clear, this bug is not file specific, it happens randomly with any nvcc compiled file that contains kernels.