panfrost: Track BO lifetime with jobs and reference counts

This (fairly large) patch continues work surrounding the panfrost_job
abstraction to improve job lifetime management. In particular, we add
infrastructure to track which BOs are used by a particular job
(currently limited to the vertex buffer BOs), to reference count these
BOs, and to automatically manage the BOs memory based on the reference
count. This set of changes serves as a code cleanup, as a way of future
proofing for allowing flushing BOs, and immediately as a bugfix to
workaround the missing reference counting for vertex buffer BOs.
Meanwhile, there are a few cleanups to vertex buffer handling code
itself, so in the short-term, this allows us to remove the costly VBO
staging workaround, since this patch addresses the underlying causes.

v2: Use pipe_reference for BO reference counting, rather than managing
it ourselves. Don't duplicate hash-table key removal. Fix vertex buffer
counting.

Signed-off-by: Alyssa Rosenzweig <[email protected]>

1 files changed, 4 insertions, 1 deletions

diff --git a/src/gallium/drivers/panfrost/pan_screen.h b/src/gallium/drivers/panfrost/pan_screen.h
index cbadf813675..4e5efa91f22 100644
--- a/src/gallium/drivers/panfrost/pan_screen.h
+++ b/src/gallium/drivers/panfrost/pan_screen.h
@@ -92,8 +92,11 @@ struct panfrost_screen {
         /* TODO: Where? */
         struct panfrost_resource *display_target;
 
-	int last_fragment_id;
+        /* While we're busy building up the job for frame N, the GPU is
+         * still busy executing frame N-1. So hold a reference to
+         * yesterjob */
 	int last_fragment_flushed;
+        struct panfrost_job *last_job;
 };
 
 #endif /* PAN_SCREEN_H */