 I have a draft blog post up about Apple's recent Metal shader converter tool: https://github.com/raphlinus/raphlinus.github.io/blob/ec87e0bd22ea848e3ed8bb3d37930a1ef7750d3c/_posts/2023-06-08-shader-converter.md Basically, I feel it furthers fragmentation of the ecosystem, and fails to make progress toward a solid foundation for GPU compute. I also make the case for #WGSL and #WebGPU by comparison. Comments and suggestions are welcome! The video of my talk on Xilem Vector Graphics is now published: https://youtu.be/XjbVnwBtVEk I think these are very promising ideas, and the main reason I'm putting this talk out now is to try to find open source collaborators to develop them further. But it is early days, and this is not yet a usable stack.  @raph loved this talk! There were a lot of links in your slides; are the slides available? I'll be giving a talk on Xilem Vector Graphics at the Bay Area Rust Meetup tomorrow evening: https://www.meetup.com/rust-bay-area/events/293165024/ It's still rough work in progress, but I'm excited about the potential. And I'm really looking forward to talking with people. It won't be livestreamed, but there will be a recording. I've posted the blog post on Bézier path simplification: https://raphlinus.github.io/curves/2023/04/18/bezpath-simplify.html I'm proud of this work. I think it's the best implementation in either the academic literature or shipping products, and am looking forward to continuing it. I'm also interested to see what applications people will find (the core ideas need tuning depending on the use case), and am also open to collaborations along these lines. I have a draft blog post up on simplifying Bézier paths: https://github.com/raphlinus/raphlinus.github.io/blob/7a0834d6c0eb5d46602634a2d1f2cf7dc8b38f95/_posts/2023-04-18-bezpath-simplify.md This points to the latest kurbo code (still in a PR) and also explains what was going on with the bumps, with a reasonably good solution. I believe this is now appropriate for use with fonts and similar applications. Feedback on the draft is welcome! @raph I'm just the peanut gallery here but I have to ask... what's going on at these two points, on the lower edge? normally I'd expect the simplification to be smoother, but these seem to be surprisingly sharp turns on a smooth section of the input curve? I had to double check, are the source and output curves swapped? (just asking because they jumped out, incredibly impressed with this work) Major congratulations to Chrome for landing #WebGPU in M113. This is a game changer. Looking forward to a whole ecosystem developing around this exciting technology. https://developer.chrome.com/blog/webgpu-release/ After many years working in Rust, I've decided to migrate my projects to C++, to benefit from a more mature, stable language. More details about that on my blog: https://raphlinus.github.io/rust/2023/04/01/rust-to-cpp.html
Show previous comments
@raph For anyone who didn't realise, this is an April Fool's post. (In my timezone, this was posted at "Apr 02, 2023, 00:20".) There is no exact translation of the word "petty" to german, but if I were to explain the meaning of it to someone, I would simply show them this article. Good one Raph ^ ^ I will be showing off Vello at the WebGL and #WebGPU meetup on March 22 in San Francisco. I'm excited, I think it'll be a chance to show it to a wider audience and meet a bunch of people. Now I just need to work on getting a compelling demo together.  @raph The demo went great and it was wonderful to be meet and catch up with folks in person! The recording is up on YouTube: https://www.youtube.com/watch?v=oWwtCDv3Pgg Whew! Getting the last of the edge-crossing logic was hard, but I finally have my multisampled path rendering in a working state. The numerical robustness is not perfect yet, but I suspect that will involve relatively minor tweaks. Performance is not quite as good as the area-based approach, but still very fast, and I have ideas how to address that. Now I have to figure out how to write all this up! I should add, this code is in the multi branch of the Vello repo: https://github.com/linebender/vello/tree/multi/ It's not ready to check into main yet for a variety of reasons, one of which is that it only applies to fills, not strokes. The plan for that is to do stroke to fill conversion in a compute shader, but to say that's difficult is an understatement. I do seem to specialize in such things, though. My research retreat was very productive. I didn't quite finish implementing sparse multisampled path rendering on GPU, but I got close (hopefully soon). I ran into a performance snag but have a good idea how to fix it. But the real fruit is probably ideas for even higher quality path rendering - a hybrid of analytical area AA and multisampling - and also compositing. Expect to hear more before long. @raph how much do you think people care about higher quality antialiasing? Skia is looking like they are going all in with MSAA in graphite and that kind of swayed my confidence that better is really needed I'm excited, I'm about to go on a week-long research retreat in a lakefront cottage. During the week, I'll focus on an idea for multisampled path rendering, and be able to tune out all other distractions. I've been craving doing something like this for a while, and the last time I did it was really great. I hope to have an update about what I do during the week. @raph BTW dropped your name a bit in https://www.tbray.org/ongoing/When/202x/2023/02/09/Monospace Anyone in driving range of Berkeley want a Dell P2415Q 4k monitor which may or may not be flaky? It wasn't turning on earlier today, so I got a replacement, but as Murphy's law would have it, it seemed to light when I plugged it in when I got back. I'll list on craigslist if I don't get a response here. @raph What did you replace it with? I cherish the 3 I have left. The lack of similar options nowadays is really frustrating. I just booked a cottage on a lake for a week-long research retreat Feb 13-18. I've done this kind of thing a couple times before and am hoping to make it a regular practice. The idea is to fully immerse into one deep topic, without the usual distractions and gear-switching. This time it'll *probably* be path rendering without conflation artifacts, but there are a couple other deep topics I'd love to address. Attached is an image from that inquiry: how to parallelize line coverage on a grid. @raph Hi Raph, I watched your great talk about Xilem, very interesting stuff, I am currently building a 2D sprite editor in Rust, so this problem you @raph You might want to take a look over how forma currently does it: it reduces the problem to finding the ith term in the sorted union between two arithmatic progressions ax + c and bx + d, which can be solved in O(1) for some given level of accuracy. Today's work on Vello was of the grungy, infrastructural nature. But that's just as important as the high-flying fancy parallel algorithms. The web port works now on main, which is actually a big deal: it uses #WGSL workgroupUniformLoad to avoid uniformity problems, which has now landed in Chrome Canary. Also, clear_buffer was a no-op TODO in the wasm bindings, causing corrupt atomic buffers. Now fixed in wgpu main. Now that the tree is in good shape, on to the fun stuff again! If you were laid off at Google or otherwise impacted, please feel free to reach out to me to chat. I know several excellent people who lost their jobs on Friday. Now is also a good time to plug the Alphabet Workers Union. I've somewhat quietly been a member for exactly a year (tomorrow is my anniversary), but going forward that will be less quietly. Your job does not love you back, but your union will look out for you. @raph Has there been demonstrated evidence of AWU looking out for people in a high value way commensurate with their funding yet? @raph Raph, what do you think about these layoffs across the tech industry, by corporations that have not suffered a decline in profits? Daniel McNab tracked down a missing barrier in the pathtag scan shader, which was causing artifacts on Vulkan (but, as these things go, not affecting mac). The same day, Max finally got around to installing the RTX 4080 (which was a Christmas present). That means I'm now able to run Vello and profile it in Nsight. The results are jaw-dropping. Not just butter-smooth zoom and pan paris-30k at 120fps at 4k, but Nsight reports <1.4ms compute time to render. |
Gregory's Server