Writing a GPU Kernel Function
Express device code in Mojo.
Writing a GPU Kernel Function is a free Mojo Academy lesson on CoddyKit — lesson 3 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the Mojo Academy learning path, one of 4 lessons in the course, and your progress syncs across the web and the CoddyKit app.
What a Kernel Is
A GPU kernel is the function each thread runs. You write the work for one element, and the hardware repeats it across many.
Think One Thread
The trick is to write a kernel from a single thread's view. Each thread does its own small slice of the work.
A Kernel Is Just an fn
In Mojo a kernel is a normal fn. Its strict, typed nature is exactly what device code needs to compile fast.
fn add_kernel():
passPassing in Buffers
A kernel takes pointers to device memory as parameters. These buffers hold the inputs and the output it will fill.
fn add_kernel(a: UnsafePointer[Float32], out: UnsafePointer[Float32]):
passFinding This Thread's Index
Inside the kernel, each thread computes its own position first. That index selects which element it must process.
var i = block_idx.x * block_dim.x + thread_idx.xGuarding the Bounds
Always check the index before touching memory. A bounds check keeps stray threads from reading past the array.
if i < n:
out[i] = a[i] + b[i]Doing the Element's Work
The body is tiny: read inputs, compute, write one result. The whole kernel often fits in a single line of math.
out[i] = a[i] * b[i]Launching the Kernel
You launch it by choosing a grid and block size. The launch fans your one-thread code out across the whole grid.
ctx.enqueue_function[add_kernel](grid_dim=blocks, block_dim=256)No Return Value
Kernels do not return results to the caller. They write into the output buffer, which you read back afterward.
Keep It Branch-Light
Threads run best in lockstep. Heavy branching makes lanes diverge and wait, so keep kernel logic simple and uniform.
Same Idea, Massive Scale
One short kernel plus a big grid equals millions of results. The scale comes from the launch, not from longer code.
Quick Check
You are writing the body of a GPU kernel for one thread.
Recap
A kernel is an fn for one thread: find your index, guard the bounds, do one element's math, then launch over a grid. ⚡
Frequently asked questions
Is the “Writing a GPU Kernel Function” lesson free?
Yes — the full text of “Writing a GPU Kernel Function” is free to read here on the web, and the Mojo Academy course includes 4 lessons in total. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the Mojo Academy course, upgrade to CoddyKit PRO.
What will I learn in “Writing a GPU Kernel Function”?
Express device code in Mojo. You practise Mojo Academy with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.
Do I need any experience to start Mojo Academy?
No prior experience is required. Mojo Academy on CoddyKit is structured for beginners through advanced learners; this is — lesson 3 of 4, so you can start here or from the beginning and move at your own pace.
How long does the “Writing a GPU Kernel Function” lesson take?
Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.
Can I write and run code in this Mojo Academy lesson?
Yes. Every Mojo Academy lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.
All lessons in this course
- Why GPUs for AI Workloads
- Threads, Blocks, and Grids
- Writing a GPU Kernel Function
- Moving Data to and from Device