panfrost: Add indirect draw support
This version has been simplified (no multi-draw and no indirect draw count) and we no longer have those
compute -> vertex -> tiler
|___________________|
loops.
All dEQP-GLES31.functional.draw_indirect.* tests are passing except those with a huge number of vertices which timeout, but they pass if we increase the kernel-side job timeout.
Things to improve:
-
Collect the number of emitted primitives -
Support XFB -
Lower those 64bit integer divs to something GPU friendly (algo suggested by amonakov http://sprunge.us/HdJHeK) -
Use 8/16 bit arithmetic in the min/max search path (requires some patches in alyssa:indirect-next branch, hasn't been fully tested)