Conner Moulaison

Projects

Empedocles, 2026

Grid of 8 cubes showcasing dielectric and metallic materials with varying degrees of reflection fuzziness applied, as well as motion and focal blur. Pre-rendered at a 240x240 pixel resolution and 32 samples per pixel. SIMD optimized CPU path traced renderer in x86 assembly

I wrote this in a little over a month as a learning project oriented around understanding x86 assembly, the CPU, and performance optimization generally. I achieved the following goals:

Supported SSE 4.2 SIMD instructions: 128 bit xmm registers allowed me to implement 3D algorithms with single instructions per vector operation.

Created an automatic profiling process: I wrote a set of shell scripts to automatically build the project with different parameters and graph the results to find the optimal settings for varying performance targets.

Modeled dielectric materials, motion and focal blur: I implemented a refractive glass material, sampled across time to produce motion blur, and sampled a unit disk around the camera origin for focal blur.

Achieved realtime performance: As of now, I can render a grid of 8 cubes at 480x480 pixels and 4 samples per pixel at a steady 60 fps. Tested on a relatively low powered laptop CPU: AMD Ryzen 5 PRO 8540U at 3.2 GHz.

Planned future improvements include reducing intersection tests with a bounding volume hierarchy and implementing an AVX-512 codepath utilizing a wavefront approach used in state of the art renderers.

Duck March, 2026

One of the more complex movement puzzles in the game. ~25 of these sorts of static screens were created utilizing several bespoke mechanics. Puzzle adventure game written in 2 weeks for LowRez Jam 2026

LowRez is a 2 week long game jam with a 64x64 resolution constraint. I was interested to see what I could accomplish in a custom C codebase, and I managed to do the following:

Supported both native Linux and the web: Prior to this project, I built only for Linux using the Xlib, ALSA, and OpenGL. I added Emscripten and SDL 2 support, relying on the magic of Emscripten's system to adapt OpenGL calls to WebGL.

Prototyped a complex control scheme: The initial idea for the design was Frogger meets Snake. Design and implementation proved more complex than I had considered, with all sorts of ambiguities to work out, mostly involving moving platforms. After lots of tinkering, I think I landed on something satisfying.

Stress tested my csm_core library: I've been building up a core of utility code which I use as the baseline for all my new projects, with helpers for memory allocation, asset packing pipelines, file processing, math, and more. This was a good stress test, with lots of notes made for improvement.

I'm proud of what I accomplished given the time frame, writing ~4800 new lines of C code, ~25 small puzzle screens, and a number of bespoke mechanics and minigames which unfold as the game progresses.

Netpong, 2025

Two game clients connected to a standalone server on the same machine. Latency is very low in this configuration, but you can still see some client-side interpolation. Online multiplayer Pong clone written like a shooter

I started writing this with the eventual goal of making a faster paced multiplayer 3D mech game, and my approach was consequently much more ambitious than if I was just making a Pong clone. Though I still haven't started work on that more complex game, I did achieve the goals I set out with for this proof of concept, including:

Implemented client-side prediction and server reconciliation: Prioritizing a robust and smooth experience, I ended up with something heavily inspired by the approaches of both Overwatch and Rocket League, with a fixed timestep, client side prediction of both local and remote players, and server reconciliation.

Wrote a custom UDP transport layer: I worked directly with Unix sockets to support a range of features including bit packing of network packets, user/session management, debug network visualizations, detailed frame logging, and tools for simulating varying network latency and packet loss.

Supported a range of network configurations: The server can run standalone or client hosted, with support for fully local play and bot players. Sessions can support any combination of player types, be they hosts, clients only, or bots.

I am still intending to write a more complex multiplayer game using this general approach, though that project has been placed on the back burner for now.