Aperture space rollback snafu
@ickle
Submitted by Chris Wilson Assigned to Intel 3D Bugs Mailing List
Link to original bug (#106668)
Description
Since we may update context state as we compute the batch, that state is preserved as we rollback the batch for a failed aperture check -- the subsequent emission then skips the atoms as the context state is believed to match requirements.
E.g. gen7_upload_urb().
Affects: genX(blorp_exec) brw_dispatch_compute_common brw_draw_single_prim
The aperture rollback has been in place since
commit 3d851ae4 Author: Eric Anholt eric@anholt.net Date: Sat Oct 22 09:37:57 2011 -0700
i965: Use the batch save/reset code to avoid needing the BO validate step.
7.10-branchpoint-6137-g3d851a!
The issue with believing global state for urb in particular seems to be from
commit d5e217db Author: Jason Ekstrand jason.ekstrand@intel.com Date: Fri Aug 18 16:10:39 2017 -0700
i965: Stop looking at NewDriverState when emitting 3DSTATE_URB
I'm not sure if that's the only instance.