I am working on a fairly large project.
I have cuda 4.0
I am trying to use cublas but every time I link with cublas.lib any cuda call makes the program crash.
I get cudaError_enum at memory location 0x1011c360.
The little sdk example works just fine.
My GPU device has 16 Multi-Processors, SM 1.1 compute capabilities.
I have tried everything and run out of ideas.
I just ran into the same problem. Unfortunately also I ran out of ideas =/
But maybe I can clarify that (at least for me) any call to cudaMalloc will cause the cudaError_enum-exception when linking with cublas.lib (and using at least a single function to it in the code such as cublas-context-creation, which doesn’t even need to be executed).
When not linking against cublas.lib everything runs like a charm.
I’m compiling a 32bit binary, have the 32bit CUDA Toolkit 4.1 and use the 64bit driver on a 64bit win7.
I have the same problem, when I try to use any cublas function(in my cod taht I wrote in SDK template)my code crash and no longer works, vs when I try to use CUDA kernels in one of my code which contains cublas function( I wrote it in SDK simplecublas) and works well, again it crash and there is a link error.
Does anyone know what I have to do?