CUDA shared memory -
i need know cuda shared memory. let's assign 50 blocks 10 threads per block in g80 card. each sm processor of g80 can handle 8 blocks simultaneously. assume that, after doing calculations, shared memory occupied.
what values in shared memory when next 8 new blocks arrive? previous values reside there? or previous values copied global memory , shared memory refreshed next 8 blocks?
it states type qualifiers:
- variables in registers thread, stays in kernel
- variables in global memory thread, stays in kernel
__device__ __shared__
type variable in shared memory block, stays in kernel__device__
type variable in global memory grid, stays until application exits__device__ __constant__
type variable grid, stays until application exits
thus reference, answer question memory should refreshed next 8 blocks if reside in shared memory of device.
Comments
Post a Comment