[PATCH i-g-t v4 0/3] Improve engine activity accuracy test with spinner reuse

[Soham Purkait <soham.purkait@intel.com>]

      Instead of creating a new spinner during each pass, it's more efficient
to instantiate the spinner once and reuse it by properly ending and resetting
it between iterations.
    This approach reduces the overhead associated with repeated spinner creation,
saving time across multiple passes. By eliminating the cumulative time spent
on spinner initialization (i.e., creation time * number of passes),
the spinner's active duration more accurately reflects the actual engine busyness.
This leads to more precise measurements of the percentage of engine busyness in
terms of spinner running time.

With this change engine-activity-accuracy test result clearly shows the difference:

Before spinner reuse :

Starting subtest: engine-activity-accuracy-90
Starting dynamic subtest: engine-DRM_XE_ENGINE_CLASS_RENDER0
calibration=500ms, test=1000ms, cycle=100ms; ratio=90.00% (90000us/10000us)
error=-2.90% (81.03% vs 83.93%)
Dynamic subtest engine-DRM_XE_ENGINE_CLASS_RENDER0: SUCCESS (1.598s)
Starting dynamic subtest: engine-DRM_XE_ENGINE_CLASS_COPY0
calibration=500ms, test=1000ms, cycle=100ms; ratio=90.00% (90000us/10000us)
error=-2.81% (83.06% vs 85.87%)
Dynamic subtest engine-DRM_XE_ENGINE_CLASS_COPY0: SUCCESS (1.684s)
Starting dynamic subtest: engine-DRM_XE_ENGINE_CLASS_COMPUTE0
calibration=500ms, test=1000ms, cycle=100ms; ratio=90.00% (90000us/10000us)
error=-2.04% (83.83% vs 85.87%)
Dynamic subtest engine-DRM_XE_ENGINE_CLASS_COMPUTE0: SUCCESS (1.667s)
Starting dynamic subtest: engine-DRM_XE_ENGINE_CLASS_VIDEO_DECODE0
calibration=500ms, test=1000ms, cycle=100ms; ratio=90.00% (90000us/10000us)
error=-2.15% (83.99% vs 86.14%)
Dynamic subtest engine-DRM_XE_ENGINE_CLASS_VIDEO_DECODE0: SUCCESS (1.647s)
Starting dynamic subtest: engine-DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE0
calibration=500ms, test=1000ms, cycle=100ms; ratio=90.00% (90000us/10000us)
error=-2.00% (83.59% vs 85.60%)
Dynamic subtest engine-DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE0: SUCCESS (1.664s)
Subtest engine-activity-accuracy-90: SUCCESS (8.264s)

After spinner reuse:

Starting subtest: engine-activity-accuracy-90
Starting dynamic subtest: engine-DRM_XE_ENGINE_CLASS_RENDER0
calibration=500ms, test=1000ms, cycle=100ms; ratio=90.00% (90000us/10000us)
error=0.03% (89.45% vs 89.42%)
Dynamic subtest engine-DRM_XE_ENGINE_CLASS_RENDER0: SUCCESS (1.560s)
Starting dynamic subtest: engine-DRM_XE_ENGINE_CLASS_COPY0
calibration=500ms, test=1000ms, cycle=100ms; ratio=90.00% (90000us/10000us)
error=0.02% (89.56% vs 89.54%)
Dynamic subtest engine-DRM_XE_ENGINE_CLASS_COPY0: SUCCESS (1.559s)
Starting dynamic subtest: engine-DRM_XE_ENGINE_CLASS_COMPUTE0
calibration=500ms, test=1000ms, cycle=100ms; ratio=90.00% (90000us/10000us)
error=0.02% (89.42% vs 89.40%)
Dynamic subtest engine-DRM_XE_ENGINE_CLASS_COMPUTE0: SUCCESS (1.571s)
Starting dynamic subtest: engine-DRM_XE_ENGINE_CLASS_VIDEO_DECODE0
calibration=500ms, test=1000ms, cycle=100ms; ratio=90.00% (90000us/10000us)
error=0.01% (89.28% vs 89.27%)
Dynamic subtest engine-DRM_XE_ENGINE_CLASS_VIDEO_DECODE0: SUCCESS (1.581s)
Starting dynamic subtest: engine-DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE0
calibration=500ms, test=1000ms, cycle=100ms; ratio=90.00% (90000us/10000us)
error=0.01% (89.37% vs 89.36%)
Dynamic subtest engine-DRM_XE_ENGINE_CLASS_VIDEO_ENHANCE0: SUCCESS (1.583s)
Subtest engine-activity-accuracy-90: SUCCESS (7.859s)

Soham Purkait (3):
  lib/xe/xe_spin: Introduce xe_spin_reset
  lib/xe/xe_spin: Export xe_spin_sync_wait
  tests/intel/xe_pmu: Refine engine activity accuracy test

 lib/xe/xe_spin.c     | 43 ++++++++++++++++++++++++++++++++++++++++++-
 lib/xe/xe_spin.h     |  2 ++
 tests/intel/xe_pmu.c | 17 +++++++++++++----
 3 files changed, 57 insertions(+), 5 deletions(-)

-- 
2.34.1