gallium: the big draw refactor of 2021
In #4608, a discussion was begun regarding the merging of draws where only
index_bias changes, and I hypothesized that this would benefit certain apps, such as Unigine Superposition.
I spent a little time thinking about the process for affecting these changes, and, as I do, I then started deleting code. The result is this MR, which accomplishes a few things:
pipe_draw_info::index_biasoff into the multidraw struct for tc draw merging
- also simplifies the gallium/mesa interfaces a bit by coincidence
pipe_draw_info::drawidto a separate draw_vbo param
- slightly optimizes tc draw merging by removing unmergeable draws from the merge path
Initial results show a slight improvement in drawoverhead figures (initial results seemed more exciting but turned out to just be Intel power/thermal wackiness), and I'm hoping to get my updated multidraw implementation for RADV to stop hanging my GPU long enough to find out how much less CPU I'm using in Superposition soon.
Unfortunately, this MR is a massive amount of changes to every driver. I've run piglit and CTS on zink across a couple drivers, and it seems like everything in core is working as it should be. The majority of the churn is in the struct rename patch, which has no functional changes whatsoever, ideally enabling review to be easier for the remainder of the series.
With that said, this will naturally have to wait until after the upcoming branch point to land since making major changes like this just before a release isn't ideal.