Attached are two sigdumps of some segfault behavior that occurs in my code. The segfault always occurs on the very first CUDA call pertaining to allocation, no matter what that call is. In other words, setDevice() doesn’t cause any problems, but creating an cufftPlan or doing some hostMalloc’ing instantly segfaults.
This was code that used to work fine in whatever version of CUDA was out ~7 months ago (2.1?). Any idea what might have changed that could cause this?
Regrettably, I can’t easily post a small reproduction case; but please tell me if anything jumps out at you from the dumps attached.
sigdump19709.txt (12.3 KB)
sigdump19378.txt (12.3 KB)