Global id calculation

[vyrphan]

Hi mates,

i’m using a big grid of various dimensions for the call of a kernel. Depends on problem, i resize that dimensions.

My problem now arises when i calculate a global id for each thread executing kernel. The formula i used is:
((blockDim.x * (blockIdx.x + blockIdx.y * gridDim.x) + threadIdx.x))

and i think is right to get my unique id per thread, but it isn’t.

An example of my execution is:

name_kernel <<< Grid, 256 >>> (params…)

where Grid, in the last execution, was (464, 464, 0)

Thanks in advance

[randal]

Hi mates,

i’m using a big grid of various dimensions for the call of a kernel. Depends on problem, i resize that dimensions.

My problem now arises when i calculate a global id for each thread executing kernel. The formula i used is:

((blockDim.x * (blockIdx.x + blockIdx.y * gridDim.x) + threadIdx.x))

and i think is right to get my unique id per thread, but it isn’t.

An example of my execution is:

name_kernel <<< Grid, 256 >>> (params…)

where Grid, in the last execution, was (464, 464, 0)

Thanks in advance

I think the formula u need to use is

Id = (BlockIdx.ygridDim.x + BlockIdx.x )Blockdim.xBlockdim.y+ threadIdx.yblockdim.x +threadIdx.x