Premade GPU shaders?

[slouken]

Does it make sense to have some premade shaders that people playing around with the GPU API can use, without adding the complexity of a shader pipeline? Maybe a simple vertex shader and solid color + alpha blending fragment shaders? The vertex shader might be used indefinitely for certain projects, and the solid and blending shaders will get people a long way in tutorials or just playing around with the API.

[flibitijibibo]

We do actually have an equivalent of shadertoy but it's currently in C#, we were planning to port it but mainly as the example for shadercross.

[slouken]

I'm thinking about just bare bones shaders that come with SDL that allow someone to stand up some GPU code and try it out without having to have their own shader workflow out of the box.

I'm looking at what's involved for @captain0xff's GPU text engine example, and the bulk of the commit is shaders:
libsdl-org/SDL_ttf#412

[flibitijibibo]

Maybe something like BasicEffect, where the idea is to implement the equivalent of the old fixed function toggles?

https://github.com/FNA-XNA/FNA/blob/master/src/Graphics/Effect/StockEffects/HLSL/BasicEffect.fx
https://github.com/FNA-XNA/FNA/blob/master/src/Graphics/Effect/StockEffects/BasicEffect.cs

[JBetz]

How about exposing the shaders used by the GPU backend of SDL's renderer? I imagine that a lot of people will want to reimplement basic functionality of SDL's renderer to get started, then extend as needed.

[flibitijibibo]

How about exposing the shaders used by the GPU backend of SDL's renderer? I imagine that a lot of people will want to reimplement basic functionality of SDL's renderer to get started, then extend as needed.

Probably not since these are likely to change over time - the first big change that comes to mind would be moving sprite buffer generation to a compute shader, which would definitely break the existing setup. A fixed-function helper at least has exactly one way to do it, even if there are a handful of permutations we'd have to put up with.

[thatcosmonaut]

I'm pretty strongly against this, it's just a false simplification. Clients who want to use the programming paradigm that revolves almost entirely around shaders should just learn how to use shaders. There's not really any way to sugarcoat it. We'll only confuse people by providing "default" paths which are only defaults if the render pipeline makes the exact assumptions that we bake into the shaders. If someone wants to provide a library that makes basic stuff easier they are more than welcome to do that and take on the burden of the lack of flexibility that comes with making assumptions.

As far as the process of compiling shaders goes, the shadercross tooling I'm currently developing should eliminate a lot of the current friction points.