Skip to contentSkip to conversation

drm/i915/pmu: Stop averaging with the previous sample

PATCH In review Tvrtko Ursulin 4 participants 6 messages Nov 24, 2017

Subsystems (2)
lore
Share & tools
lore

Tvrtko Ursulin

author maintainer
#0 lore raw

PATCH 1/1 drm/i915/pmu: Stop averaging with the previous sample

From: Tvrtko Ursulin <[email protected]>

Averaging with the previous sample brings a small statistical improvement to sampling counters, but can leek a little bit of state from a current client to the next which mulls the border between past and present for observing clients.

This is because on event enable clients record the current counter value and use it as reference, but with rapid off-on event cycles, and due the delayed nature of sampling timer self-disarm, previous sample value does not get cleared under these circumstances.

Solution is to stop averaging with the previous sample. This has a small downside of losing some precision with short and spiky signals, but the alternatives look too complicated for the benefit.

Cc Chris Wilson <[email protected]>
Cc Sagar Arun Kamble <[email protected]>
---
 drivers/gpu/drm/i915/i915_pmu.c | 29 ++---------------------------
 drivers/gpu/drm/i915/i915_pmu.h |  1 -
 2 files changed, 2 insertions(+), 28 deletions(-)
drivers/gpu/drm/i915/i915_pmu.c +2 -27
index 5170a46893d7..685aef1d81b9 100644
@@ -181,12 +181,7 @@ static bool grab_forcewake(struct drm_i915_private *i915, bool fw)
 static void
 update_sample(struct i915_pmu_sample *sample, u32 unit, u32 val)
 {
-	/*
-	 * Since we are doing stochastic sampling for these counters,
-	 * average the delta with the previous value for better accuracy.
-	 */
-	sample->cur += div_u64(mul_u32_u32(sample->prev + val, unit), 2);
-	sample->prev = val;
+	sample->cur += mul_u32_u32(val, unit);
 }
 
 static void engines_sample(struct drm_i915_private *dev_priv)
@@ -262,31 +257,13 @@ static void frequency_sample(struct drm_i915_private *dev_priv)
 	}
 }
 
-static void pmu_init_previous_samples(struct drm_i915_private *i915)
-{
-	struct intel_engine_cs *engine;
-	enum intel_engine_id id;
-	unsigned int i;
-
-	for_each_engine(engine, i915, id) {
-		for (i = 0; i < ARRAY_SIZE(engine->pmu.sample); i++)
-			engine->pmu.sample[i].prev = 0;
-	}
-
-	for (i = 0; i < ARRAY_SIZE(i915->pmu.sample); i++)
-		i915->pmu.sample[i].prev = i915->gt_pm.rps.idle_freq;
-}
-
 static enum hrtimer_restart i915_sample(struct hrtimer *hrtimer)
 {
 	struct drm_i915_private *i915 =
 		container_of(hrtimer, struct drm_i915_private, pmu.timer);
 
-	if (!READ_ONCE(i915->pmu.timer_enabled)) {
-		pmu_init_previous_samples(i915);
-
+	if (!READ_ONCE(i915->pmu.timer_enabled))
 		return HRTIMER_NORESTART;
-	}
 
 	engines_sample(i915);
 	frequency_sample(i915);
@@ -867,8 +844,6 @@ void i915_pmu_register(struct drm_i915_private *i915)
 	hrtimer_init(&i915->pmu.timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
 	i915->pmu.timer.function = i915_sample;
 
-	pmu_init_previous_samples(i915);
-
 	for_each_engine(engine, i915, id)
 		INIT_DELAYED_WORK(&engine->pmu.disable_busy_stats,
 				  __disable_busy_stats);
drivers/gpu/drm/i915/i915_pmu.h +0 -1
index e209783a4c5f..40c154d13565 100644
@@ -41,7 +41,6 @@ enum {
 
 struct i915_pmu_sample {
 	u64 cur;
-	u32 prev;
 };
 
 struct i915_pmu {
Signature · 6 lines
2.14.1

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
#1 lore raw

✓ Fi.CI.BAT: success for drm/i915/pmu: Stop averaging with the previous sample

== Series Details ==

== Series Details ==

Series: drm/i915/pmu: Stop averaging with the previous sample
URL : https://patchwork.freedesktop.org/series/34346/
State : success

== Summary ==

Series 34346v1 drm/i915/pmu: Stop averaging with the previous sample
https://patchwork.freedesktop.org/api/1.0/series/34346/revisions/1/mbox/

Test gem_busy:

        Subgroup basic-hang-default:
                incomplete -> PASS       (fi-cfl-s2)
Test gem_exec_reloc:
        Subgroup basic-gtt-cpu-active:
                fail       -> PASS       (fi-gdg-551) fdo#102582 +2

fdo#102582 https://bugs.freedesktop.org/show_bug.cgi?id=102582

fi-bdw-5557u     total:289  pass:268  dwarn:0   dfail:0   fail:0   skip:21  time:443s
fi-bdw-gvtdvm    total:289  pass:265  dwarn:0   dfail:0   fail:0   skip:24  time:464s
fi-blb-e6850     total:289  pass:223  dwarn:1   dfail:0   fail:0   skip:65  time:381s
fi-bsw-n3050     total:289  pass:243  dwarn:0   dfail:0   fail:0   skip:46  time:537s
fi-bwr-2160      total:289  pass:183  dwarn:0   dfail:0   fail:0   skip:106 time:278s
fi-bxt-dsi       total:289  pass:259  dwarn:0   dfail:0   fail:0   skip:30  time:508s
fi-bxt-j4205     total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:513s
fi-byt-j1900     total:289  pass:254  dwarn:0   dfail:0   fail:0   skip:35  time:497s
fi-byt-n2820     total:289  pass:250  dwarn:0   dfail:0   fail:0   skip:39  time:492s
fi-cfl-s2        total:289  pass:263  dwarn:0   dfail:0   fail:0   skip:26  time:612s
fi-elk-e7500     total:289  pass:229  dwarn:0   dfail:0   fail:0   skip:60  time:432s
fi-gdg-551       total:289  pass:178  dwarn:1   dfail:0   fail:1   skip:109 time:269s
fi-glk-1         total:289  pass:261  dwarn:0   dfail:0   fail:0   skip:28  time:540s
fi-hsw-4770      total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:428s
fi-hsw-4770r     total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:442s
fi-ilk-650       total:289  pass:228  dwarn:0   dfail:0   fail:0   skip:61  time:428s
fi-ivb-3520m     total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:471s
fi-ivb-3770      total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:468s
fi-kbl-7500u     total:289  pass:264  dwarn:1   dfail:0   fail:0   skip:24  time:489s
fi-kbl-7560u     total:289  pass:270  dwarn:0   dfail:0   fail:0   skip:19  time:527s
fi-kbl-7567u     total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:476s
fi-kbl-r         total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:533s
fi-pnv-d510      total:289  pass:222  dwarn:1   dfail:0   fail:0   skip:66  time:578s
fi-skl-6260u     total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:454s
fi-skl-6600u     total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:545s
fi-skl-6700hq    total:289  pass:263  dwarn:0   dfail:0   fail:0   skip:26  time:566s
fi-skl-6700k     total:289  pass:265  dwarn:0   dfail:0   fail:0   skip:24  time:518s
fi-skl-6770hq    total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:498s
fi-skl-gvtdvm    total:289  pass:266  dwarn:0   dfail:0   fail:0   skip:23  time:459s
fi-snb-2520m     total:289  pass:250  dwarn:0   dfail:0   fail:0   skip:39  time:559s
fi-snb-2600      total:289  pass:249  dwarn:0   dfail:0   fail:0   skip:40  time:431s

a5d8a220645ed25df2088568070cfdfb1bbf59c4 drm-tip: 2017y-11m-24d-09h-54m-57s UTC integration manifest bc84de5a37ce drm/i915/pmu: Stop averaging with the previous sample

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_7275/
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Reviewed-by on v1 1/1
#3 lore raw
In reply to Tvrtko Ursulin

Quoting Tvrtko Ursulin (2017-11-24 09:49:59)

Quoted 12 lines

From: Tvrtko Ursulin <[email protected]>

Averaging with the previous sample brings a small statistical improvement to sampling counters, but can leek a little bit of state from a current client to the next which mulls the border between past and present for observing clients.

This is because on event enable clients record the current counter value and use it as reference, but with rapid off-on event cycles, and due the delayed nature of sampling timer self-disarm, previous sample value does not get cleared under these circumstances.

Quoted 7 lines

Solution is to stop averaging with the previous sample. This has a small downside of losing some precision with short and spiky signals, but the alternatives look too complicated for the benefit.

Cc Chris Wilson <[email protected]>
Cc Sagar Arun Kamble <[email protected]>

Nothing wrong with the patch, and I agree that chasing that little bit of smoothness is not worth the complexity in the edge cases. Oh well.

(Until such a time as we have a clean way of being perfect, good enough.)
-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

Tested-by
#4 lore raw

Quoting Chris Wilson (2017-11-24 10:42:15)

Quoted 17 lines

Quoting Tvrtko Ursulin (2017-11-24 09:49:59)

Quoted 16 lines

From: Tvrtko Ursulin <[email protected]>

Averaging with the previous sample brings a small statistical improvement to sampling counters, but can leek a little bit of state from a current client to the next which mulls the border between past and present for observing clients.

This is because on event enable clients record the current counter value and use it as reference, but with rapid off-on event cycles, and due the delayed nature of sampling timer self-disarm, previous sample value does not get cleared under these circumstances.

Solution is to stop averaging with the previous sample. This has a small downside of losing some precision with short and spiky signals, but the alternatives look too complicated for the benefit.

Quoted 7 lines
Quoted 3 lines
Cc Chris Wilson <[email protected]>
Cc Sagar Arun Kamble <[email protected]>

Nothing wrong with the patch, and I agree that chasing that little bit of smoothness is not worth the complexity in the edge cases. Oh well.

-Chris
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#5 lore raw

✗ Fi.CI.IGT: warning for drm/i915/pmu: Stop averaging with the previous sample

In reply to Tvrtko Ursulin
== Series Details ==

== Series Details ==

Series: drm/i915/pmu: Stop averaging with the previous sample
URL : https://patchwork.freedesktop.org/series/34346/
State : warning

== Summary ==

Test drv_module_reload:

        Subgroup basic-no-display:
                dmesg-warn -> PASS       (shard-snb) fdo#102707 +1
Test kms_frontbuffer_tracking:
        Subgroup fbc-1p-offscren-pri-shrfb-draw-render:
                fail       -> PASS       (shard-snb) fdo#101623
        Subgroup fbc-tilingchange:
                pass       -> SKIP       (shard-hsw)
Test kms_flip:
        Subgroup flip-vs-absolute-wf_vblank-interruptible:
                fail       -> PASS       (shard-hsw) fdo#100368
Test gem_busy:
        Subgroup close-race:
                pass       -> FAIL       (shard-snb) fdo#103829
Test kms_setmode:
        Subgroup basic:
                pass       -> FAIL       (shard-hsw) fdo#99912

fdo#102707 https://bugs.freedesktop.org/show_bug.cgi?id=102707
fdo#101623 https://bugs.freedesktop.org/show_bug.cgi?id=101623
fdo#100368 https://bugs.freedesktop.org/show_bug.cgi?id=100368
fdo#103829 https://bugs.freedesktop.org/show_bug.cgi?id=103829
fdo#99912 https://bugs.freedesktop.org/show_bug.cgi?id=99912

shard-hsw        total:2667 pass:1534 dwarn:1   dfail:0   fail:10  skip:1122 time:9512s
shard-snb        total:2667 pass:1312 dwarn:1   dfail:0   fail:13  skip:1341 time:8119s

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_7275/shards.html
_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

#2 lore raw

Re: ✓ Fi.CI.BAT: success for drm/i915/pmu: Stop averaging with the previous sample

In reply to Patchwork
Quoted 52 lines · Patchwork

== Series Details ==

Series: drm/i915/pmu: Stop averaging with the previous sample
URL : https://patchwork.freedesktop.org/series/34346/
State : success

== Summary ==

Series 34346v1 drm/i915/pmu: Stop averaging with the previous sample
https://patchwork.freedesktop.org/api/1.0/series/34346/revisions/1/mbox/

Test gem_busy:

         Subgroup basic-hang-default:
                 incomplete -> PASS       (fi-cfl-s2)
Test gem_exec_reloc:
         Subgroup basic-gtt-cpu-active:
                 fail       -> PASS       (fi-gdg-551) fdo#102582 +2

fdo#102582 https://bugs.freedesktop.org/show_bug.cgi?id=102582

fi-bdw-5557u     total:289  pass:268  dwarn:0   dfail:0   fail:0   skip:21  time:443s
fi-bdw-gvtdvm    total:289  pass:265  dwarn:0   dfail:0   fail:0   skip:24  time:464s
fi-blb-e6850     total:289  pass:223  dwarn:1   dfail:0   fail:0   skip:65  time:381s
fi-bsw-n3050     total:289  pass:243  dwarn:0   dfail:0   fail:0   skip:46  time:537s
fi-bwr-2160      total:289  pass:183  dwarn:0   dfail:0   fail:0   skip:106 time:278s
fi-bxt-dsi       total:289  pass:259  dwarn:0   dfail:0   fail:0   skip:30  time:508s
fi-bxt-j4205     total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:513s
fi-byt-j1900     total:289  pass:254  dwarn:0   dfail:0   fail:0   skip:35  time:497s
fi-byt-n2820     total:289  pass:250  dwarn:0   dfail:0   fail:0   skip:39  time:492s
fi-cfl-s2        total:289  pass:263  dwarn:0   dfail:0   fail:0   skip:26  time:612s
fi-elk-e7500     total:289  pass:229  dwarn:0   dfail:0   fail:0   skip:60  time:432s
fi-gdg-551       total:289  pass:178  dwarn:1   dfail:0   fail:1   skip:109 time:269s
fi-glk-1         total:289  pass:261  dwarn:0   dfail:0   fail:0   skip:28  time:540s
fi-hsw-4770      total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:428s
fi-hsw-4770r     total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:442s
fi-ilk-650       total:289  pass:228  dwarn:0   dfail:0   fail:0   skip:61  time:428s
fi-ivb-3520m     total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:471s
fi-ivb-3770      total:289  pass:260  dwarn:0   dfail:0   fail:0   skip:29  time:468s
fi-kbl-7500u     total:289  pass:264  dwarn:1   dfail:0   fail:0   skip:24  time:489s
fi-kbl-7560u     total:289  pass:270  dwarn:0   dfail:0   fail:0   skip:19  time:527s
fi-kbl-7567u     total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:476s
fi-kbl-r         total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:533s
fi-pnv-d510      total:289  pass:222  dwarn:1   dfail:0   fail:0   skip:66  time:578s
fi-skl-6260u     total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:454s
fi-skl-6600u     total:289  pass:262  dwarn:0   dfail:0   fail:0   skip:27  time:545s
fi-skl-6700hq    total:289  pass:263  dwarn:0   dfail:0   fail:0   skip:26  time:566s
fi-skl-6700k     total:289  pass:265  dwarn:0   dfail:0   fail:0   skip:24  time:518s
fi-skl-6770hq    total:289  pass:269  dwarn:0   dfail:0   fail:0   skip:20  time:498s
fi-skl-gvtdvm    total:289  pass:266  dwarn:0   dfail:0   fail:0   skip:23  time:459s
fi-snb-2520m     total:289  pass:250  dwarn:0   dfail:0   fail:0   skip:39  time:559s
fi-snb-2600      total:289  pass:249  dwarn:0   dfail:0   fail:0   skip:40  time:431s
Quoted 2 lines

a5d8a220645ed25df2088568070cfdfb1bbf59c4 drm-tip: 2017y-11m-24d-09h-54m-57s UTC integration manifest bc84de5a37ce drm/i915/pmu: Stop averaging with the previous sample

Pushed after seeing that shards were also happy.

Regards,

Tvrtko

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx