[PATCH] dma-buf/fence: fix fence_is_later v2

Christian König christian.koenig at amd.com
Tue Mar 1 08:42:01 UTC 2016


Am 01.03.2016 um 09:36 schrieb Maarten Lankhorst:
> Op 29-02-16 om 23:13 schreef Gustavo Padovan:
>> 2016-02-29 Alex Deucher <alexdeucher at gmail.com>:
>>
>>> From: Christian König <christian.koenig at amd.com>
>>>
>>> A fence is never later than itself. This caused a bunch of overhead for AMDGPU.
>>>
>>> v2: simplify check as suggested by Michel.
>>>
>>> Signed-off-by: Christian König <christian.koenig at amd.com>
>>> Reviewed-by: Michel Dänzer <michel.daenzer at amd.com>
>>> Reviewed-by: Alex Deucher <alexander.deucher at amd.com>
>>> Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
>>> ---
>>>   include/linux/fence.h | 2 +-
>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>> Reviewed-by: Gustavo Padovan <gustavo.padovan at collabora.co.uk>
>>
>> 	Gustavo
> Why is it causing overhead? Not really opposed to the patch though.

We tested if flushing the VM is necessary by remembering the last 
flushed VM update operation and comparing that to what operation we need 
to have flushed for the current command submission to work.

So when we where already at the last operation in the timeline we would 
flush anyway which was superfluous.

Regards,
Christian.


More information about the dri-devel mailing list