Philippe Rollin

Loading images is one of the most fundamental things a browser does. It turns out that doing it efficiently from JavaScript is an open problem. There are multiple APIs available and the right choice depends on the browser you are targeting. ludicon.com/castano/blog/2…

Graphics Programming weekly - Issue 438 - April 27h, 2026 jendrikillner.com/post/graphics-…

I spent some time reverse-engineering Apple's Lossy texture format. It wasn't as simple as I originally thought, and some of the details surprised me. Check it out! ludicon.com/castano/blog/2…

I've been reverse engineering Apple's, ARM's, and ImgTec's hardware image compression formats. ARM's AFRC is the clear winner, but does native hardware compression make real-time texture encoding obsolete? ludicon.com/castano/blog/2…

@SebAaltonen Why 4 passes? iirc you used single pass with encoded viewport xform in shadow matrix before to save bw on mobile.

Shadow cascade rendering (four passes):

Our new code base uses my Hyper RHI directly in user land code. It's pretty clean. Struct/span based APIs with good defaults. No heap allocs (initializer lists live in stack for the function call duration). This is how the G-buffer pass looks currently:

KTX ETC1S files are compact, but is the quality tradeoff really worth the jarring artifacts? ludicon.com/castano/blog/2…

It's a common misconception that free healthcare costs more than paying for insurance yourself. Healthcare is roughly 2x more expensive in the USA. Money goes to insurance companies, drug companies, investors (profits) and bureaucracy instead of actually treating patients.

@wookash_podcast hund

Tomorrow airs the episode with the legend Fabian Giesen. Known in certain hardcore circles as ryg. Prepare!

My "No Graphics API" blog post is live! Please repost :) sebastianaaltonen.com/blog/no-graphi…

@SebAaltonen I can still hear the voice over just looking at that screenshot :)

I am considering a brick-map style voxel system for HypeHype. Each 1m^3 brick would point to shared subtrees (= DAG). A bit like the Unlimited Detail demo back in the day. We wouldn't need that much detail in the bricks. Aiming for Nintendo game look instead of realistic look.

@royalicing @wookash_podcast Games are modern apps made by modern orgs which are many teams. Each subsystem usually gets a budget and each team has to work within that budget.

@wookash_podcast Because it relies on knowing memory use of all features upfront. Modern apps are architected around modern orgs, which are many teams that work independently and a compiler or runtime or cloud has the job of stitching it all together into one system.

Honestly, why is this *not* the default approach for memory management?

20years in the US, recently moved back to europe; here is my xp so far: - Ppl with those salaries in europe seem to be happier and more fulfilled than ppl with those 400k salary in the US - healthcare is as good; better in the long run, especially if self employed or in small company - decent schools will set you back 30-50k per year per kid in the US… that alone can quickly even things out when you have 2+ kids - housing, property taxes, insurances are much cheaper overall - fruits and vegetable actually have taste But sure… big number is bigger than the small number

Here's some cute bounding box functions for 2D shapes. A link to a longer list in the next post.

@aliasedFreq @matiasgoldberg Beat me to it, you get a 👍

@matiasgoldberg well certainly better than "every entity is actually a door"

Nothing screams Object-Oriented Programming in Gamedev more than "A ground unit is a ship that walks on land"

New blog post about my experiences using spark.js with @threejs: ludicon.com/castano/blog/2… Feedback appreciated!

@FrameworkPuter Use a brother and don’t look back

One of you must have sabotaged our office printer to force us to make one. There is no way HP is actually shipping printers that are this broken.

Compiling branches out is exactly what i was referring to; many mobile gpu still shipping prefer low instruction count rather than dynamic or even static branching. Some of those branches often come with other tradeoffs due to how they might affect the rest of the pipeline (for ex, can use many bindings that another incompatible feature might need in a constrained environment) All of this are problem we don’t have when the problem space is well defined and hardware limited to 1 or 2 targets.

It has nothing to do with any of that. They’re simply trying to eliminate statically branch and eliminate instructions for user-specific that don’t happen to use a particular branch by completely compiling them out. It’s all basically an effort at prioritizing instruction cache optimization. Of course if they didn’t have so much bloat in their shaders already then that might have otherwise not been a problem… Imagine you did codegen for C++ and created a wide number of variants of it depending upon how users configured scene content, all compiled on the fly at runtime? It’d be ridiculous there yet we’re just supposed to accept it in GPU land?

The engine should be architected in such a way as to eliminate the need for developers to even have to optimize anything. Unfortunately that’s require a radically different design. Dynamic shader permutations especially never should’ve been a thing, especially in a world without proper real-time JIT compilers.

I don’t use UE but my educated guess here is that they have to support a much larger combination of hardware, some of them (mobile) are historically not great when it comes to ubershaders (this is becoming less true, but there is a long tail of underpowered devices still in use) They also have to integrate decades of features for a wide variety of use cases. It is easier to optimize a rendering pipeline when the target hardware and the use case are well defined.

@toncijukic The idea that it’s unavoidable simply doesn’t compute to me. Maybe it’s because the simulation industry does things differently enough from the gaming industry, idk.

@ChetHusk why not winget?

Hey #dotnet - we on the .NET SDK team are all about making the `dotnet` CLI more capable and powerful over time. But how do you usually _get_ the `dotnet` CLI and other tooling? Today that answer is a _bunch_ of different ways. We have a plan to centralize these ways into ...

I will be presenting "Stochastic Tile-Based Lighting in HypeHype" at SIGGRAPH 2025 Advances course on Aug 12th, see you there! Here's the full program with many cool talks: advances.realtimerendering.com/s2025/index.ht…