Update flash.cu

[eclouder]

determine Bc, Br dynamically

[xavhl]

Thanks for the helpful update! I tested your update and concluded with the following suggestions:

1. divide by sizeof(float), or any other types, since prop.sharedMemPerBlock is in bytes
2. min(shared/4d, N) to cater small input size

// line 91:
const int Bc = min(ceil(prop.sharedMemPerBlock/sizeof(float)/(4*d)), (float)N);