Okay fellas, we have compute support now in native web, and outputting kernels are possible e.g. this tutorial: https://jott.live/markdown/webgpu_safari
Who has started or where do we want to start the discussion of a WebGPU-based spiritual successor for GPUjs? The whole javascript-based kernel generation is genius for learning GPU coding and helped me immensely since the performance is fine too, and it's not really that complicated under the hood since it's just transposing text for you. The WebGPU compute pipeline adds some additional boilerplate for setting up command/storage buffers and data structures though it could definitely be macro'd based on what I've seen.
Another thing to think about is chaining compute and fragment shaders, though compute can handle rasterization and possibly faster, while the fragment shader can simply dump the resulting image matrix to screen. E.g. https://github.com/OmarShehata/webgpu-compute-rasterizer/blob/main/how-to-build-a-compute-rasterizer.md
Okay fellas, we have compute support now in native web, and outputting kernels are possible e.g. this tutorial: https://jott.live/markdown/webgpu_safari
Who has started or where do we want to start the discussion of a WebGPU-based spiritual successor for GPUjs? The whole javascript-based kernel generation is genius for learning GPU coding and helped me immensely since the performance is fine too, and it's not really that complicated under the hood since it's just transposing text for you. The WebGPU compute pipeline adds some additional boilerplate for setting up command/storage buffers and data structures though it could definitely be macro'd based on what I've seen.
Another thing to think about is chaining compute and fragment shaders, though compute can handle rasterization and possibly faster, while the fragment shader can simply dump the resulting image matrix to screen. E.g. https://github.com/OmarShehata/webgpu-compute-rasterizer/blob/main/how-to-build-a-compute-rasterizer.md