Basic - CUDA - tile width & block width

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello forum,

Is there any difference between tile and block concept ?

I am following the book “Programming massively parallel processors” and in section 4.2 the block size is assumed to be TILE_WIDTH.

But when i go through the cuda fluid example, they did something different.

Could someone in the forum put more light into this idea ? - block width and tile width.

Thanks
Sajjad

2

This is confusing. What is a tile? There is no mentioned of tiles in the CUDA Programming Guide. The size of a block is chosen based on problems.

3

Let me put down the snippet from the book and please explain it for me - TILE_WIDTH specially.

Does this basically mean width of the block?

__global__ void matrixmult(float *Md, float *Nd, float *Pd,int width) 
{
     int row = blockIdx.y * TILE_WIDTH + threadIdx.y;
     int column = blockIdx.x * TILE_WIDTH + threadIdx.x; 

     ......................
     .....................
}

Regards
Sajjad

4

The TILE_WIDTH is a variable he defined for a specific problem. It is not a CUDA variable. For the blocks you have the intrinsec variables: threadIdx.x, blockIdx.x,BlockDim.x, GridDim.x. No TILE_WIDTH. For each problem you define the the number of threads per block as it suits you.