Shadertoys Ported to Rust GPU
rust-gpu.github.io
rust-gpu.github.io
Let me start with the pros:
1. Don't have to learn 2 different languages
2. Modules, crates, and the easier ability to share code
3. Easier sharing between rust structs and shader code.
Now the cons, in comparison to Slang [1]
1. No autodiff mode 2. Strictly outputs SPIR-V, while Slang can do CPU, CUDA, Pytorch, Optix, and all the major graphics APIs
3. Less support - Slang is supported by the Khronos group, and Slang gets use at Nvidia, EA, and Valve.
4. Safety isn't very valuable, most GPU code does not use pointers (it's so rare it's considered a feature by Slang!)
5. slangc probably runs a lot faster than rustc (although I would like to see a benchmark.)
6. Worse debugging experience, slang has better interop with things like NSight Graphics, and their Shader Debugger. Slang recently got support in NSight graphics for shader profiling, for example.
7. Slang has support for reflection, and has a C++ api to directly output a JSON file that contains all the reflected aspects.This makes handling the movement between rust <-> gpu much easier. Also, the example shown on the website uses `bytemuck`, but `bytemuck` won't take into consideration the struct alignment rules[2] when using WebGPU. Instead, you have to use a crate like `encase`[3] to handle that. I'm not sure given the example on the website how it would work with WebGPU.
8. If you have pre-existing shaders in GLSL or HLSL, you can use slangc directly on them. No need to rewrite.
9. In reality, you may not have to learn 2 languages but you have to learn 2 different compute models (CPU vs GPU). This is actually a much harder issue, and AFAICT it is impossible to overcome with a different language. The problem is the programmer needs to understand how the platforms are different.
[1] https://shader-slang.org/ [2] https://webgpufundamentals.org/webgpu/lessons/resources/wgsl... WGSL struct alignment widget [3] https://github.com/teoxoy/encase
It's interesting that Slang looks more like Rust than WGSL does, despite WGSL kind of being de facto "owned by" the Rust community.
Actually this is a divergence point with HLSL202x evolution, which aims to be more C++ like.
WGSL is the only shading language with Rust influence.
Rust is great for tracking the lifetime of long-lived resources, which everything in a shader isn't.
Apart from that, what makes Rust a good fit for this problem domain?
And as an added benefit it means not adding another language to a project and all that comes with it for build system/CI. I.e. cargo takes care of everything. That alone is worth a lot IMHO.
Apart from that its not related to Rust. Just replace Rust with your fav. language and imagine you could also write shaders/GPU-targeted code in it. Isn't that desirable?
Unit testing of shaders is usually a pain. Traditionally they're a black box without any assert() equivalent, and you can at best try to propagate NaN and generate magenta pixels on failure. Sharing Rust code lets you unit-test parts of it on the CPU.
Sharing of data structures between CPU and GPU is nice too. WGSL is superficially similar to Rust, but using plain WGSL requires maintaining bindings and struct layouts by hand, which is a chore.
For CUDA, the alternative is C++. On the upside that's the official first-class API for CUDA, but the downside is that it's C++. With Rust you don't have the legacy language cruft, nor busywork like header files or makefiles. You get working dependency management, and don't get caught in the unix vs Windows schism.
How they did it? Hard to do because GPU hardware can convert data types on the fly, e.g. you can store bytes in VRAM, and convert them to 32-bit floats in [ 0 .. +1 ] in the shader. GPUs can do that for both inputs (loaded texture texels, loaded buffer elements, vertex attributes) and outputs (rendered pixels, stored UAV elements).
To get the format conversion stuff you talk about, you need to use images, vertex input or texel buffers and configure the format conversion explicitly.
It's a good question how much of these conversions are actually done by GPU hardware and how much of it is just software (which you could write yourself in a shader and get same perf). I have not seen an apples to apples benchmark about these format conversions.
Yeah, that will work fine for byte address buffers, to lesser extent constant buffers (they don’t convert data types but the access sematic and alignment are a bit tricky), but not much else. Vertex buffers, textures, and texel buffers / typed buffers in D3D are all widely used in real-time graphics.
> which you could write yourself in a shader and get same perf
Pretty sure it’s hardware. Emulating anisotropic texture sampler with HLSL codes would need hundreds of instructions, prohibitively expensive. Even simpler trilinear sampling is surprisingly tricky to emulate due to these screen-space partial derivatives on input.
> I have not seen an apples to apples benchmark about these format conversions.
Here’s a benchmark for vertex buffers https://wickedengine.net/2017/06/should-we-get-rid-of-vertex... As you see, on AMD GCN4 he indeed measured pretty much the same perf, however on nVidia Maxwell vertex buffers were 2-4 times faster.
This is where sharing the CPU and GPU side struct declaration is helpful. With scalar block layout (VK_EXT_scalar_block_layout in Vulkan, not sure how about d3d land) you don't even need to worry about alignment rules because they're the same for GPU and CPU (just make sure your binding base address/offset is aligned).
> Vertex buffers, textures, and texel buffers / typed buffers in D3D are all widely used in real-time graphics.
Of course. You don't get to share "structs" here between CPU and GPU here transparently because you need to program the GPU hardware (vertex input, texture samplers) to match.
There are some reflection based trickery that can help here but rust-gpu afaik doesn't do that. I've seen some projects use proc macros to generate vertex input layout config for GL/Vulkan from Rust structs with some custom #[attribute] annotations.
> Pretty sure it’s hardware.
Now this is just guessing.
> Emulating anisotropic texture sampler with HLSL codes would need hundreds of instructions...
Texture sampling / interpolation is certainly hardware.
But the conversion from rgba8_unorm to rgba32f, for example? Or r10g10b10a2?
I've not seen any conclusive benchmark results that suggest whether it's faster to just grab these from a storage buffer in a shader and do the few arithmetic instructions or whether it's faster to use an texel buffer. Images are a different beast entirely due to tiling formats (you can't really memory map them so the point of sharing struct declarations is irrelevant).
> Here’s a benchmark for vertex buffers
I am familiar with this benchmark from 8 years ago, which is highly specific to vertex buffers (and post transform cache etc). It's a nicely done benchmark but it has two small flaws in it: the hw tested is quite old by now and it doesn't take into account the benefit of improved batching / reduced draw calls that can only be done with custom vertex fetch (so you don't need BindVertex/IndexBuffer calls). It would be great if this benchmark could be re-run with some newer hw.
But this benchmark doesn't answer the question whether the typed buffer format conversions are faster than doing it in a shader (outside of vertex input).
> however on nVidia Maxwell vertex buffers were 2-4 times faster.
The relevant hardware got revamped in Turing series to facilitate mesh shaders, so can't extrapolate the results to present day hardware.
Fwiw. I've been using custom vertex fetch with buffer device address in my projects for a few years now and I haven't noticed adverse performance implications on any hw I've used (Intel, NV and AMD). But I haven't done rigorous benchmarking that would compare to using vertex input stage.
I'm not using rust-gpu for shaders at the moment, but if I was, it would be helpful to just use the same struct declarations. All my vertex data, instance data, constant buffers and compute buffers are an 1:1 translation from Rust to GLSL struct declarations which is just redundant work.
Indeed, but sharing code between CPU and shaders is not the only way to solve the problem. I wrote a simple design-time tool which loads compiled shaders with shader reflection API, and generates a source file with C++ header (or for other projects C# structures) with these constant buffers. At least with D3D11, compiled shaders have sufficient type and memory layout info to generate these structures, matching memory layout by generating padding fields when necessary.
> not sure how about d3d land
Pretty sure D3D11 doesn’t have an equivalent of that Vulkan extension. Not sure about D3D12 though, only used the 12 briefly.
> I've been using custom vertex fetch with buffer device address in my projects
In my projects, I sometimes use a lot of non-trivial input layout features. Sometimes I need multiple vertex buffers, e.g. to generate normals on GPU with a compute shader. Sometimes I need instancing. Often I need FP16 or SNORM/UNORM vertex attributes, like RG16_UNORM for octahedron-encoded normals.
Thankfully these days we have printf in shaders that you can use for "asserts". You can detect if the shader printed anything and consider it a failure.
You can even add a conditional print in your pixel shader, run your app in renderdoc and find the pixel(s) that printed something. Once you find one, you can step through it in the shader debugger.
This seemingly simple feature is a huge time saver.
"Without dragging in a language runtime" happens to be a primary requirement for writing bare-metal code or writing GPU shaders.
So you have a language that does have a bunch of nice new features (ADTs, acceptably modern type system, etc.) and doesn't have any of the stuff that prevents you from compiling it to random bare-metal targets like GPUs or MMUless microcontrollers.
I'm so in the weeds with dealing with GLSL these days that "What if you could use the same language to implement the CPU logic as the GPU logic" wasn't even a goal I could see. That's actually quite huge; my concerns around Rust were mostly that it's not a deeply simple language and so much of shader frameworks is code-that-writes-code, but it's simple enough that I think it could still be used in that space, while granting the advantages of not having to context-switch in and out of one's CPU-language while developing a shader.
This has promise!
But that is just vertex shaders for things like Shadertoys.
There is much more to computer graphics or GPGPU than this though, Im still learning about that myself X).
[1] https://www.youtube.com/watch?v=eKtsY7hYTPg [2]https://www.youtube.com/watch?v=kfM-yu0iQBk
Programming the shaders themselves are pretty simple imo, they're just pure functions that return color data or triangle positions. The syntax might be a little different than you're used to depending on the shader language, but it should be easy enough to pick up in a day.
If you want to write compute shaders for computation, then it gets a lot more tricky and you need to spend some time learning about memory accesses, the underlying hardware, and profiling.
You need to visit http://thebookofshaders.com/
My main approach to writing shaders is to look at existing programs (e.g. Blender) and see what techniques are in use. The Google Filament renderer documentation [0] is also really good when it comes to BDSF functions.
Some papers from Unreal Engine might also help, such as "Real Shading in Unreal Engine 4" [1]
[0] https://google.github.io/filament/Filament.md.html
[1] https://cdn2.unrealengine.com/Resources/files/2013SiggraphPr...
Are CUDA and OpenCL really "better alternatives" to HLSL and GLSL?
CUDA and OpenCL are compute languages; HLSL and GLSL and shader languages. And while one can theoretically do compute in a shader (and we used to!) or shaders in a compute language, I think it's dishonest to claim that CUDA is intended as an updated alternative to GLSL. It's simply apples and oranges.
CUDA yes, it is already being used in commercial visualisation products like OctaneRender, one of the most important tools in the VFX industry.
NVidia also has plenty of customers on OptiX.
GLSL is dead end, Khronos is on the record they aren't going to develop it further, even for Vulkan, HLSL and now slang, are the way forward.
HLSL due to its use in the games industry, slang due to being developed by NVIDIA and given to Khronos as GLSL replacement.
I've been seeing more comments like yours saying GLSL is a dead end, and it’s making me question whether I should keep learning it.
But at the same time, it’s still very accessible, well-documented, and used in OpenGL demos and educational content. Would you say there's still a case for GLSL in hobby projects, demoscene work, or rapid prototyping? Or is it really time to move on even in those contexts?