When i am using cuda-memcheck from cuda 4.0 sdk i get always memory access violations, when memory is in the opengl buffer and is mapped to a cuda pointer, when accessing it within a kernel.
I get it in all of my kernels. They work with mapped gl buffers without memcheck or with memcheck and normal allocated cudaMemory where i copy content of gl buffer per cudaMemCpy
Idea why it happens?
Does memcheck do some modifications with cuda memory to allow bounds checking and can not access/modify memory in opengl buffer?