Shader Printf in HLSL and DX12

therealmjp.github.io

48 points·davikr··6 comments

A strong GPU programmer can do this instinctively. But thinking about it deeper, I think this is exactly the kind of project that demonstrates what is so different about GPUs vs CPU programming.

In CPU programming, just sequentially outputting printf one-at-a-time works. And you can optimize this to go faster-and-faster.

In GPU programming, its innately a parallel job. Any printf call will innately cause all 32-threads of your warp to call printf at the same time (at a minimum!!), which immediately requires all 32 threads to coordinate. Bonus points for expanding out to 1024-threads of a workgroup or cross-workgroup atomics with a wider CUDA-grid.

This "printf" is simple enough to a typical programmer, and yet hits so many advanced parallelization problems (so "reserving" bufferspace and other such tricks to help make the parallel-back-into-sequential for CPU land)

Setting the output color to red was our printf back in the day.

Neat.

In the meantime web developers keep yelling “you don’t need console.log”, lol

Web developers typically have working debuggers available though.

I write SPAs in VueJS, but even with the vue devtools available, I still find myself using console.log so that I don't have to keep drilling down into a very deep component tree every time the page reloads. Plus, the console lets me scroll back and read the history of changes, something I'd only get in the devtools if I threw every bit of local state into pinia. The devtools come in handy from time to time, but the DX is just not there yet, and I doubt it ever will be.

GPU often has debuggers available. PIX is very good these days for example.

Reply on news.ycombinator.com