[PATCH 0/7] A few drm_syncobj optimisations

Tvrtko Ursulin tvrtko.ursulin at igalia.com
Tue Mar 25 09:57:30 UTC 2025


On 24/03/2025 23:17, Maíra Canal wrote:
> Hi Tvrtko,
> 
> Thanks for this patchset! I applied this patchset to the RPi downstream
> kernel 6.13.7 [1] and saw an FPS improvement of approximately 5.85%
> with "vkgears -present-mailbox" on the RPi 5.
> 
> I did five 100 seconds runs on each kernel and here are my results:
> 
> ### 6.13.7
> 
> |   Run    |   Min FPS   |   Max FPS   |   Avg FPS   |
> |----------|-------------|-------------|-------------|
> | Run #1   | 6646.52     | 6874.77     | 6739.313    |
> | Run #2   | 5387.04     | 6723.274    | 6046.773    |
> | Run #3   | 6230.49     | 6823.47     | 6423.923    |
> | Run #4   | 5269.678    | 5870.59     | 5501.858    |
> | Run #5   | 5504.54     | 6285.91     | 5859.724    |
> 
> * Overall Avg FPS: 6114.318 FPS
> 
> 
> ### 6.13.7 + DRM Syncobj optimisations
> 
> |   Run    |   Min FPS   |   Max FPS   |   Avg FPS   |
> |----------|-------------|-------------|-------------|
> | Run #1   | 6089.05     | 7296.27     | 6859.724    |
> | Run #2   | 6022.48     | 7264        | 6818.518    |
> | Run #3   | 5987.68     | 6188.77     | 6041.365    |
> | Run #4   | 5699.27     | 6448.99     | 6190.374    |
> | Run #5   | 6199.27     | 6791.15     | 6450.900    |
> 
> * Overall Avg FPS: 6472.176 FPS

Neat, thanks for testing! I am not surprised a slower CPU benefits more.

Btw if you have the raw data it would be nice to feed it to ministat too.

Regards,

Tvrtko

> [1] https://github.com/raspberrypi/linux/tree/rpi-6.13.y
> 
> Best Regards,
> - Maíra
> 
> On 18/03/25 12:54, Tvrtko Ursulin wrote:
>> A small set of drm_syncobj optimisations which should make things a 
>> tiny bit
>> more efficient on the CPU side of things.
>>
>> Improvement seems to be around 1.5%* more FPS if observed with "vkgears
>> -present-mailbox" on a Steam Deck Plasma desktop, but I am reluctant 
>> to make a
>> definitive claim on the numbers since there is some run to run 
>> variance. But, as
>> suggested by Michel Dänzer, I did do a five ~100 second runs on the 
>> each kernel
>> to be able to show the ministat analysis.
>>
>> x before
>> + after
>> +------------------------------------------------------------+
>> |                          x         +                       |
>> |                   x      x         +                       |
>> |                   x      xx      ++++                      |
>> |                 x x      xx x    ++++                      |
>> |                 x xx   x xx x+   ++++                      |
>> |                xxxxx   xxxxxx+   ++++ + +                  |
>> |                xxxxxxx xxxxxx+x  ++++ +++                  |
>> |              x xxxxxxxxxxx*xx+* x++++++++   ++             |
>> |        x x   xxxxxxxxxxxx**x*+*+*++++++++ ++++ +           |
>> |       xx x   xxxxxxxxxx*x****+***+**+++++ ++++++           |
>> |x     xxx x   xxxxx*x****x***********+*++**+++++++   +  +  +|
>> |               |_______A______|                             |
>> |                             |______A_______|               |
>> +------------------------------------------------------------+
>>      N           Min           Max        Median           Avg        
>> Stddev
>> x 135      21697.58     22809.467     22321.396     22307.707     
>> 198.75011
>> + 118     22200.746      23277.09       22661.4     22671.442     
>> 192.10609
>> Difference at 95.0% confidence
>>      363.735 +/- 48.3345
>>      1.63054% +/- 0.216672%
>>      (Student's t, pooled s = 195.681)
>>
>> Tvrtko Ursulin (7):
>>    drm/syncobj: Remove unhelpful helper
>>    drm/syncobj: Do not allocate an array to store zeros when waiting
>>    drm/syncobj: Avoid one temporary allocation in drm_syncobj_array_find
>>    drm/syncobj: Use put_user in drm_syncobj_query_ioctl
>>    drm/syncobj: Avoid temporary allocation in
>>      drm_syncobj_timeline_signal_ioctl
>>    drm/syncobj: Add a fast path to drm_syncobj_array_wait_timeout
>>    drm/syncobj: Add a fast path to drm_syncobj_array_find
>>
>>   drivers/gpu/drm/drm_syncobj.c | 281 ++++++++++++++++++----------------
>>   1 file changed, 147 insertions(+), 134 deletions(-)
>>
> 



More information about the dri-devel mailing list