From: Soham Purkait <soham.purkait@intel.com>
To: igt-dev@lists.freedesktop.org, riana.tauro@intel.com,
badal.nilawar@intel.com, kamil.konieczny@intel.com
Cc: anshuman.gupta@intel.com, soham.purkait@intel.com,
umesh.nerlige.ramappa@intel.com
Subject: [PATCH i-g-t v4 3/3] tests/intel/xe_pmu: Refine engine activity accuracy test
Date: Tue, 6 Jan 2026 22:18:01 +0530 [thread overview]
Message-ID: <20260106164801.46353-4-soham.purkait@intel.com> (raw)
In-Reply-To: <20260106164801.46353-1-soham.purkait@intel.com>
Creating a new spinner during each pass was introducing unnecessary
overhead, which could distort timing-sensitive measurements
and increase error rates. The cumulative cost of repeated spinner
initialization was impacting test accuracy and efficiency.
To address this, a single spinner instance is now reused and properly
ended and reset 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 (creation time × number of passes). As the spinner's active
duration more accurately reflects actual engine busyness, this leads to
more precise measurements of the percentage of engine busyness in terms
of spinner running time.
With this change, the engine-activity-accuracy test results clearly show
an improvement in error reduction ranging from 98.97% to 99.53%.
v1:
- Add how this patch improves the engine activity accuracy test. (Lucas)
- Show the numbers with the said improvement. (Lucas)
v2:
- Add an overview of the issue. (Riana)
- Add the links in Closes. (Riana)
- Add improvement in percentage. (Kamil)
v3:
- Add spin_sync_wait() with excess time adjusted
accordingly. (Riana)
Fixes: 477154cbad2c ("tests/intel/xe_pmu: Add tests to validate engine activity accuracy")
Closes: https://gitlab.freedesktop.org/drm/xe/kernel/-/issues/6251
Signed-off-by: Soham Purkait <soham.purkait@intel.com>
---
tests/intel/xe_pmu.c | 17 +++++++++++++----
1 file changed, 13 insertions(+), 4 deletions(-)
diff --git a/tests/intel/xe_pmu.c b/tests/intel/xe_pmu.c
index c8b3cebf0..33852270a 100644
--- a/tests/intel/xe_pmu.c
+++ b/tests/intel/xe_pmu.c
@@ -534,6 +534,10 @@ static void accuracy(int fd, struct drm_xe_engine_class_instance *eci,
intel_allocator_init();
ahnd = intel_allocator_open(fd, 0, INTEL_ALLOCATOR_RELOC);
+ spin = igt_spin_new(fd, .ahnd = ahnd, .vm = vm, .hwe = eci);
+ xe_spin_end(spin->xe_spin);
+ xe_spin_sync_wait(fd, spin);
+
for (int pass = 0; pass < ARRAY_SIZE(timeout); pass++) {
unsigned int target_idle_us = idle_us;
struct timespec start = { };
@@ -546,21 +550,25 @@ static void accuracy(int fd, struct drm_xe_engine_class_instance *eci,
while (pass_ns < timeout[pass]) {
unsigned long loop_ns, loop_active_ns, loop_idle_ns, now;
+ unsigned long after_sync;
double err, prev_avg, cur_val;
/* idle sleep */
igt_measured_usleep(target_idle_us);
/* start spinner */
- spin = igt_spin_new(fd, .ahnd = ahnd, .vm = vm, .hwe = eci);
+ xe_spin_reset(fd, spin);
loop_idle_ns = igt_nsec_elapsed(&start);
igt_measured_usleep(active_us);
- igt_spin_free(fd, spin);
+ xe_spin_end(spin->xe_spin);
now = igt_nsec_elapsed(&start);
+ xe_spin_sync_wait(fd, spin);
+ after_sync = igt_nsec_elapsed(&start);
+
loop_active_ns = now - loop_idle_ns;
loop_ns = now - pass_ns;
- pass_ns = now;
+ pass_ns = after_sync;
pass_active_ns += loop_active_ns;
total_active_ns += loop_active_ns;
@@ -590,6 +598,7 @@ static void accuracy(int fd, struct drm_xe_engine_class_instance *eci,
sizeof(expected));
}
+ igt_spin_free(fd, spin);
xe_vm_destroy(fd, vm);
put_ahnd(ahnd);
}
@@ -622,7 +631,7 @@ static void accuracy(int fd, struct drm_xe_engine_class_instance *eci,
igt_info("error=%.2f%% (%.2f%% vs %.2f%%)\n",
(engine_activity - expected) * 100, 100 * engine_activity, 100 * expected);
- assert_within(100.0 * engine_activity, 100.0 * expected, 3);
+ assert_within(100.0 * engine_activity, 100.0 * expected, 2);
}
static void engine_activity_all_fn(int fd, struct drm_xe_engine_class_instance *eci, int num_fns)
--
2.34.1
next prev parent reply other threads:[~2026-01-06 16:53 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2026-01-06 16:47 [PATCH i-g-t v4 0/3] Improve engine activity accuracy test with spinner reuse Soham Purkait
2026-01-06 16:47 ` [PATCH i-g-t v4 1/3] lib/xe/xe_spin: Introduce xe_spin_reset Soham Purkait
2026-01-06 16:48 ` [PATCH i-g-t v4 2/3] lib/xe/xe_spin: Export xe_spin_sync_wait Soham Purkait
2026-01-06 16:48 ` Soham Purkait [this message]
2026-01-07 8:45 ` [PATCH i-g-t v4 3/3] tests/intel/xe_pmu: Refine engine activity accuracy test Riana Tauro
2026-01-06 17:44 ` ✓ Xe.CI.BAT: success for Improve engine activity accuracy test with spinner reuse (rev6) Patchwork
2026-01-06 18:02 ` ✓ i915.CI.BAT: " Patchwork
2026-01-06 18:59 ` ✓ Xe.CI.Full: " Patchwork
2026-01-06 22:15 ` ✓ i915.CI.Full: " Patchwork
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20260106164801.46353-4-soham.purkait@intel.com \
--to=soham.purkait@intel.com \
--cc=anshuman.gupta@intel.com \
--cc=badal.nilawar@intel.com \
--cc=igt-dev@lists.freedesktop.org \
--cc=kamil.konieczny@intel.com \
--cc=riana.tauro@intel.com \
--cc=umesh.nerlige.ramappa@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox