Local memory usage with NVCC 3.1 Probelm of lmem usage while compiling with nvcc 3.1 SDK compiler

Hi,

I have a strange problem. I have a kernel that I compiled using the nvcc compiler coming from the 2.3 CUDA SDK and this compiler reported a local memory usage of 88+0 bytes. When I compile the same kernel with the nvcc compiler from the 3.1 SDK, the lmem usage increases to 844+0 bytes.

Has anyone already experienced the same behaviour and figured out a solution for that ?

Thanks

Is this happening on the exact same system, with nothing changed except the CUDA toolchain version? If so, would it be possible to provide a self-contained test case that reproduces this issue? I would be interested in looking at it. If you do not want to post code publicly, you could send me a private message via the forum.

Is this happening on the exact same system, with nothing changed except the CUDA toolchain version? If so, would it be possible to provide a self-contained test case that reproduces this issue? I would be interested in looking at it. If you do not want to post code publicly, you could send me a private message via the forum.