Newer blobs use these values, and turnip missed the updates to them that were done in freedreno's b820c09f.
I'm not sure what I really want to do for SP_CHICKEN
, without a collection of blob traces to look at.
My motivation here: were there any easy register differences affecting performance for #6327 (closed) ? Answer: no. the remaining register setup diffs on a trivial shader trace were little things like:
FP16_NO_INF
-
CONCURRENT_RESOLVE
(which I already knew didn't affect this test) -
RB_UNKNOWN_8E04
left set to the magic value during the renderpass, while freedreno only sets it duringBLIT_OP_SCALE
. VK blob traces seem to leave it set all the time.