Real friends don’t let friends fuck around with GENERIC and GIMPLE

I think we both (sadly) already know the answer.

Spotify won’t say it’s done with ICE Paste Magazine is your source for the best music, movies, TV, comedy, videogames, books, comics, craft beer, politics and more. Discover your favorite albums and films.

Ehhh. They aren’t resisting at all.

“Spotify’s stance has not meaningfully shifted in the slightest. The ads are gone only because the money is, and the company remains careful not to say what it will not do.“

www.pastemagazine.com/music/spotif...

Bonus: also the world’s first standard protocol across all languages

This is a wonderful community, so if you (1) are a part of it or (2) generally have interest in supporting an HPC-related community, please consider donating!

I had the pleasure of being mentored by Ian while in graduate school. Queen’s is incredibly lucky to have him!

Day 12 of #AdventOfCode was done twice (cudf and OpenACC)!

github.com/willkill07/A...

github.com/willkill07/A...

This was a very fun year with the shorter duration and increased challenge of a new GPU programming model per day.

Day 11 of #AdventOfCode was done in C++ with std::execution!

github.com/willkill07/A...

Neither the cleanest nor the shortest, but I had to get this programming model in for the event :(

Day 10 of #AdventOfCode was done in CUDA C++!

github.com/willkill07/A...

Parallel BFS for Part 1.
Wrote my own solver for Part 2.

Runs in less than 1.3ms 🚀

Day 9 of #AdventOfCode was done in C++ with OpenMP Offload!

github.com/willkill07/A...

Part 1 and Part 2 are similar structurally, with Part 2 effectively checking for bounding box intersection between the candidate and each edge.

Runs in less than 130 microseconds.

Day 8 of #AdventOfCode was done in CUDA with CCCL!

github.com/willkill07/A...

Uses CCCL to precompute and sort all pairs of nodes by distance and then uses a disjoint set on the GPU (single threaded) to compute the results for part 1 and part 2 separately.

Day 7 of #AdventOfCode was done in ISO C++23 with C++ Standard Parallelism!

github.com/willkill07/A...

Day 6 of #AdventOfCode was done in numba-CUDA! A single kernel was used for both parts.

github.com/willkill07/A...

Day 5 of #AdventOfCode was done in NVIDIA’s Python CCCL library!

github.com/willkill07/A...

A tiny bit of cupy was used, but almost all of the algorithms were dispatched with cccl

Day 4 of #AdventOfCode was so much fun that I did it twice! OpenACC and cudf

OpenACC: github.com/willkill07/A...

cudf: github.com/willkill07/A...

Each are relatively idiomatic. I almost wrote a custom cupy stencil kernel for cudf, but I felt like that was cheating.

Day 3 of #AdventOfCode done using Warp!

github.com/willkill07/A...

The kernel itself yields an array of integer values in the correct order. numpy then reconstructs the number.

The GPU kernel generated runs on a single block where each thread gets its own line of input. Runs in <100us on device.

Day 2 of #AdventOfCode done using cupy!

github.com/willkill07/A...

I tried to avoid nasty repeated division and modulus on the GPU and aimed to do a lot of array-based programming.

With kernel fusion this would be much faster but it still works! 10 more days means 10 GPU programming models left!

Day 1 done using CuTe!

github.com/willkill07/A...

I realize this is kind-of cheating in the sense that it only runs on one GPU thread, but I do take advantage of CuTe Tensors and compile-time JIT via Constexpr!

I do wish that there was an easy way to return a simple value from a CuTe kernel.

GitHub - willkill07/AdventOfCode2025: My solutions to https://adventofcode.com/2025/ My solutions to https://adventofcode.com/2025/. Contribute to willkill07/AdventOfCode2025 development by creating an account on GitHub.

My #AdventOfCode challenge to myself is to ensure all solutions run on my GPU in 12 different programming models officially supported by NVIDIA. I’ve only programmed in half of them 😅

README is live:

github.com/willkill07/A...

The Advent of Code Day 1 countdown showing 23:46:29 remaining until the first puzzle unlocks.

aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa

Definitely one of those

“Your scientists were so preoccupied with whether or not they could, they didn't stop to think if they should.”

situations

Day 2 of #sc25 evening events (official and unofficial)

HPC Ignites Plenary in America’s Ballroom 5:30PM-6:45PM

Ribbon Cutting 6:45PM (Hall 4 corner)

Opening Gala in Exhibit Hall 7:00PM-9:00PM

Beowulf Bash afterwards at City Museum (conference badge required, ~7 blocks west of convention center)

Day 2 of #SC25

Workshops and Tutorials continue to be in full swing today!

Student Programming Events (Room 263)
- Resume Workshop 9:00-9:45AM
- Portfolio Workshop 10:00-10:45AM
- Navigating Education Systems Internationally Panel 11AM-12PM
- 1:1 Career Coaching (preregistration required) 1PM-4PM

The longer you are in the HPC space, the smaller it becomes.

#HPC #SC24. Current registration is 15,446. Peak Scinet bandwidth is 14.72 TBPS. 560 exhibitors.

I’m unsure due to my obligations with the Student Programming events

Day 1 of #sc25

Tutorials and Workshops kick off this morning!

Some Student Programming events are also running today:
- HPC/AI Crash Course - 8:30AM until 3:30PM in Room 263. Please note this required preregistration
- Career Panel - 3:45PM until 4:45PM in Room 263

Wonderful view of the St Louis skyline coming in on my flight. Eagerly anticipating this year’s #sc25 #hpcignites

Frontiere ran on Frontier. This alone is worth recognition.

The TLDR is that a really big HACC job ran on Frontier with a bunch of new algorithmic optimizations. Neato.

Safe (and uneventful) travels!