When I want to dynamically allocate the shared memory for my CUDA kernel, I notice that I can use half of the total shared memory size.
When I allocate a larger shared memory size (> 0.5*shared_memory) it will trigger the
invalid argument error
is there any reason for this? I am using the Tesla V100 with sm_70, which has the 96KB shared memory, while I can only use 48KB when allocating dynamic shared memory.
