liveBook · Manning

Cuda.grid

As explained in Fast Python: High performance techniques for large datasets

cuda.grid is a Numba CUDA intrinsic that computes a thread's global index in a one-, two-, or three-dimensional execution grid.

cuda.grid is a Numba CUDA intrinsic for obtaining the calling thread's global position. A kernel typically handles one array element, uses cuda.grid(1) for a one-dimensional index, and writes directly to that element. section 9.2.2

Doubling an array

A kernel can obtain position = cuda.grid(1) and multiply my_array[position] by two. With 1,000 elements, a 50-by-20 launch covers the array exactly; a 63-by-16 launch requires a bounds check.

How Fast Python: High performance techniques for large datasets covers it

How deep it goes
Explains one- and two-dimensional indexing, launch coverage, bounds checks, and the equivalent block/thread arithmetic.
Examples
Uses array doubling and two-dimensional matrix doubling as concrete indexing kernels.
What it assumes
Assumes basic arrays and CUDA kernel launches while teaching thread-to-element mapping through direct examples.
What you can do after
Apply via Numba CUDA by writing element-wise kernels and launching them across configured grids and blocks.
How it connects
Connects CUDA kernels with thread blocks, grids, blockIdx, blockDim, and threadIdx.
Cover of Fast Python: High performance techniques for large datasets
liveBookmanning.com
Read Fast Python: High performance techniques for large datasets on liveBook Get the book at manning.com

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Fast Python: High performance techniques for large datasets. The section links point to the parts of the book it draws on.