anv: dynamic state re-re-rework
For a while I thought that we could put instructions in the pipeline batch to minimize the amount of dynamic state we have to emit in gfx[78]_cmd_buffer.c
.
But it just doesn't work. We end up either emitting too much or too little.
This MR reverts the last commit that was supposed to fix this Mesa_CI/repos/mesa@f348103f and goes back to the previous strategy :
- If a piece of state is fully in the pipeline, emit it in genX_pipeline.c
- If a piece of state is ever dynamic, for any pipeline, move it to gfx[78]_cmd_buffer.c
There is also an improvement for sample locations on Gfx8+ where the packet 3DSTATE_SAMPLE_PATTERN
holds the patterns for all MSAA combinaisons and if we hold all the value in anv_dynamic_state
we can avoid some emission.