← Feed

Posts

• Post #1835239

  Met with the WESL group and put together a slide deck with all the shader composition I'm doing in Use.GPU. Like printf'ing debug geometry from a compute shader: Slides (60MB): http://acko.net/files/use-gpu-12/use.gpu-wesl-export.pdf

• Post #1835238

  If this is true, it is very disturbing: https://www.arkmedic.info/p/would-you-like-plasmids-with-that

• Post #1835237

  @pluralistic This post calls notable people "hatchet men" and a new agency "absurd" and "dangerous" right off the bat, while praising peer review and continuous critique. It doesn't add up, the entire argument is self-defeating. You can't defend expertise by stating that institutions are infallible and that it's wrong for new entrants with a track record of success to question them. It's just Doctorow word sa...

• Post #1835236

  simple existential software razor: Would the problem you're trying to solve still exist if computers didn't exist? If not, it's not actually solving a real problem.

• Post #1835235

  Use.GPU 0.14 release candidate is up on Gitlab, featuring a refactor render pass infra, to enable a very clean impl of GTAO. It has overscan at the screen edges for edge-to-edge coverage of shadows, and temporal reprojection. To use it you just turn on the SSAO flag. This release also improves the ways render passes are composable without requiring the dev/user to wire it all up... often there is only one correct way to do it anyway. Working on a write up, and will release 0.14 final soon.

• Post #1835234

  New Use.GPU release: - 🔍dramatic inspector viewing upgrades - 🌥️a modern ambient-occlusion (SSAO/GTAO) implementation - 🖼️revised render pass infrastructure - ℹ️expanded shader generation for bind groups - 🌴more generated WGSL struct types Full write up here: https://acko.net/blog/occlusion-with-bells-on/

• Post #1835233

  I know this is mastodon but I seriously gotta wonder if the people sharing the group chat news are upset that officials are using unofficial channels to coordinate, or whether they are upset that the wrong team of officials is doing that.

• Post #1835232

  @dneto Then I suppose I can find these people also upset that e.g. Uschi bought Pfizer shots over SMS?

• Post #1835231

  @dneto I deliberately phrased it so as not be an accusation, if you reread it.

• Post #1835230

  @datn Uh, if people are directing their ire to the wrong group, and making them into a stereotype they're not, that is actually their burden to fix. The question is why they believe men in tech are uniformly "bros" and the answer is that it's the news media that has had a vendetta against tech for a decade or so and refuses to make the distinction clearly.

• Post #1835229

  Use.GPU 0.15 is out! Highlights: - Shaded lines and point clouds - Narrow and wide (>90º) spot lights with hemispherical shadow maps - Improved GLTF `.bin` loader for packed/inline textures Also finally rendered a Sponza atrium.

• Post #1835228

  Fun discovery: it's possible to detect multisampled edges very precisely just by checking if the depth samples are planar. If you weigh the MSAA sample offsets, you can get directional sensitivity even. I vaguely remember similar demoscene-experiments from the late 2000s adding AA to a GBuffer based on a multisampled depth pre-pass, but never saw any decent write up about it. Anyone know of anything similar?

• Post #1835227

  Architectural-style outlines with MSAA... turns out just doing edge detection at the sample level and then doing a resolve + dilate works a treat. Will be in the upcoming 0.16 release of Use.GPU, with the normal/depth pre-pass now multisampled. WGSL: https://gitlab.com/unconed/use.gpu/-/tree/master/packages/wgsl/wgsl/outline?ref_type=heads

• Post #1835226

  Oh yeah, Use.GPU now has this sweet black hole raytracer: https://usegpu.live/demo/rtt/black-hole

• Post #1835225

  @tojiro Have you done any additional work on WebXR + WebGPU? In particular was wondering about: - depth24plus as the depth format ... I use depth32float with reverse Z everywhere. The guidelines suggest the depth buffer should be accurate for the given projection matrix, so is reverse Z off the table? - having to create new texture views, view descriptors, color attachments for every single frame seems terrible from a GC point of view? I usually manage all these statically.

• Post #1835223

  New Use.GPU release (0.18) with: - order-independent transparency (WBOIT) - improved dolly zoom to full orthographic, with support for SSAO and other post-fx - rewritten quad-tree map tile renderer with arc labels Docs and demos at: https://usegpu.live