I have one question concerning the cuda-dbg debugger. My kernel
consists of 512*400 blocks with 512 threads each. As the kernel
surprisingly generates incorrect data only in the block range from
blockIdx.y=310 to blockIdx.y=400, I want to debug this range with
The problem now is, when switching to those blocks cuda-dbg tells me
“CUDA thread not known or not active”. This rises a few questions:
- How can I a debug those blocks, or more precisely, how can I debug
warps of those blocks, which are currently not active on the gpu?
- Is the only way to set a conditional breakpoint, e.g. cond 1 blockIdx.y==310 ?
- And a more general question: Is there a limit on the grid dimension?
Thanks in advance.