FifoCI detected that this change impacts graphical rendering. Here are the behavior differences detected by the system:

• sw3-dt on ogl-lin-mesa: diff

_{^{automated-fifoci-reporter}}

Thanks for doing this pokechu! This seems to work fine (tested with per pixel lighting on in my custom shader branch). I did see issues with cached pipelines not matching and therefore had to blow away my cached shaders. Error was:

D3D12 ERROR: ID3D12Device::Create*PipelineState: A pointer was provided to the D3D12_CACHED_PIPELINE_STATE struct which contains a cached PSO which does not match the rest of the description provided. The shaders and state descriptions must match what was originally used to create the cached PSO exactly. [ STATE_CREATION ERROR #887: CREATEPIPELINESTATE_CACHEDBLOBDESCMISMATCH]

I wonder if we should roll the pipeline version for this?

I don't have much D3D12 experience, but I thought that GX_PIPELINE_UID_VERSION was specifically for Dolphin's shader UID cache, which stores information about the GameCube state for all of the shaders that a game has used so that they can be recompiled on launch. I assume there is also a separate cache for the actual compiled shaders, which would be a per-backend thing (because we don't need to invalidate Vulkan shaders, for instance), but I'm not sure if that's actually the case or where that would be.

Yes, we cache per graphics backend.

Looking at the disk cache code, it seems that when validating the header, the logic is only comparing the headers which has the git information and the magic key, so if users had two different builds the users would be fine. It's only because I tested in the same branch that there was issue.

In the past, I cheated and used GX_PIPELINE_UID_VERSION for any pipeline change to flush the cache. That was due to my fear and misunderstanding of how the cache code works. You are right, it is technically only supposed to be incremented for the game state.

I guess this is good to go then.