blockIdx.x ranges from 0 to 2, as there are three blocks along the x dimension.
blockDim.x is equal to 4, as this is the size of a block in the x dimension.
threadIdx.x refers to the index within one block along the x dimension. It ranges from 0 to 3.
This comment was marked helpful 6 times.
tchitten
Are there any limitations to the dimensions/size of the block? I know at the end of lecture we said a block couldn't have more than 2048 threads which makes sense since each core is limited to 2048 threads and a block has to run in entirety on a core, but are there other requirements?
This comment was marked helpful 0 times.
yanzhan2
There are limitations based on the GPU you are using, you can run some query for the GPU device in CUDA, there are samples code in CUDA 5.5, you can install them in Mac/Linux/Windows, and just run the device query program, it will show a lot information and limitation about your GPU.
The dim3 thing in CUDA is to facilitate calculation on two dimensional or three dimensional arrays. Then no need to do the index conversion, just use it as it is defined.
blockIdx.xranges from 0 to 2, as there are three blocks along the x dimension.blockDim.xis equal to 4, as this is the size of a block in the x dimension.threadIdx.xrefers to the index within one block along the x dimension. It ranges from 0 to 3.This comment was marked helpful 6 times.
Are there any limitations to the dimensions/size of the block? I know at the end of lecture we said a block couldn't have more than 2048 threads which makes sense since each core is limited to 2048 threads and a block has to run in entirety on a core, but are there other requirements?
This comment was marked helpful 0 times.
There are limitations based on the GPU you are using, you can run some query for the GPU device in CUDA, there are samples code in CUDA 5.5, you can install them in Mac/Linux/Windows, and just run the device query program, it will show a lot information and limitation about your GPU.
This comment was marked helpful 0 times.
@tchitten I think CUDA compute capabilities table tells you the details.
This comment was marked helpful 0 times.
The dim3 thing in CUDA is to facilitate calculation on two dimensional or three dimensional arrays. Then no need to do the index conversion, just use it as it is defined.
This comment was marked helpful 0 times.