Hi John,
You will get many more eyeballs on this question if you submit it to the intel graphics mailing list. Please visit https://lists.freedesktop.org/mailman/listinfo/intel-gfx for information on how to join.
thanks,
Paul
Apr 17, 2016 - 07:33am
Hi John,
You will get many more eyeballs on this question if you submit it to the intel graphics mailing list. Please visit https://lists.freedesktop.org/mailman/listinfo/intel-gfx for information on how to join.
thanks,
Paul
I am hoping this is the right place to post this question.
When using the execlist submission method (on gen9) we have requests being sent to a request queue where they await being submitted to the ELSP.
Suppose you have an application that creates 4 GL contexts (1, 2, 3, 4) and the requests happen to be submitted in the following manner:
rq - queue
| 2,1,2,3,4 |: <--- 3,4 are the next contexts to be submitted to the ELSP.
Note that we have two sets of requests waiting in the request queue for context 2. Since the Logical Ring Context (LRC) area is per context, it seems that the second request for context 2 could have modified the LRC in a way that breaks the first request for context 2.
I am sure I am probably missing something. Could you please shed some light into that?
Thanks!