From: "Belgaumkar, Vinay" <vinay.belgaumkar@intel.com>
To: Riana Tauro <riana.tauro@intel.com>, <igt-dev@lists.freedesktop.org>
Cc: badal.nilawar@intel.com
Subject: Re: [igt-dev] [PATCH i-g-t 1/2] tests/intel/xe_guc_pc: Add freq_power test
Date: Mon, 18 Sep 2023 09:31:02 -0700 [thread overview]
Message-ID: <7360a545-fe62-ea4e-b2fd-2eda5ab703e4@intel.com> (raw)
In-Reply-To: <20230907060638.2317485-2-riana.tauro@intel.com>
On 9/6/2023 11:06 PM, Riana Tauro wrote:
> An assumption is that at lower frequencies,
> not only do we run slower, but we save power compared to
> higher frequencies.
>
> This test runs a spinner and sets the min and max frequencies
> to rp0 and rpn respectively. It then checks if power measured
> at lower frequencies is lesser.
>
> Signed-off-by: Riana Tauro <riana.tauro@intel.com>
> ---
> tests/intel/xe_guc_pc.c | 88 +++++++++++++++++++++++++++++++++++++++++
> 1 file changed, 88 insertions(+)
>
> diff --git a/tests/intel/xe_guc_pc.c b/tests/intel/xe_guc_pc.c
> index 032816921..cea7b56fd 100644
> --- a/tests/intel/xe_guc_pc.c
> +++ b/tests/intel/xe_guc_pc.c
> @@ -13,6 +13,7 @@
>
> #include "igt.h"
> #include "lib/igt_syncobj.h"
> +#include "igt_power.h"
> #include "igt_sysfs.h"
>
> #include "xe_drm.h"
> @@ -387,6 +388,82 @@ static void test_reset(int fd, int gt_id, int cycles)
> }
> }
>
> +static int cmp_u64(const void *a, const void *b)
> +{
> + return (*(u64 *)a - *(u64 *)b);
> +}
> +
> +static uint64_t measure_power(int fd, struct igt_power *gpu)
> +{
> + struct power_sample sample[2];
> + uint64_t power[5];
> +
> + for (int i = 0; i < 5; i++) {
> + igt_power_get_energy(gpu, &sample[0]);
> + usleep(10000); /* 10 ms */
> + igt_power_get_energy(gpu, &sample[1]);
> +
> + power[i] = igt_power_get_mW(gpu, &sample[0], &sample[1]);
> + }
> + /* Sort in ascending order and use a triangular filter */
> + qsort(power, 5, sizeof(*power), cmp_u64);
> + return DIV_ROUND_UP(power[1] + 2 * power[2] + power[3], 4);
> +}
> +
> +/**
> + * SUBTEST: freq_power
> + * Description: Validates if running at lower frequencies saves power
> + * Run type: FULL
> + */
> +static void test_freq_power(int fd, int gt_id, struct drm_xe_engine_class_instance *hwe)
> +{
> + uint32_t rp0, rpn, vm;
> + uint64_t ahnd;
> + struct igt_power gpu;
> + struct {
> + uint64_t power;
> + uint32_t freq;
> + } min, max;
> + igt_spin_t *spin;
> +
> + /* Run for engines belonging to the gt */
> + if (gt_id != hwe->gt_id)
> + return;
> +
> + igt_power_open(fd, &gpu, "gpu");
> +
> + rpn = get_freq(fd, gt_id, "rpn");
> + rp0 = get_freq(fd, gt_id, "rp0");
> +
> + vm = xe_vm_create(fd, 0, 0);
> + ahnd = intel_allocator_open(fd, vm, INTEL_ALLOCATOR_RELOC);
> + spin = igt_spin_new(fd, .ahnd = ahnd, .vm = vm, .hwe = hwe);
> +
> + igt_assert(set_freq(fd, gt_id, "min", rpn) > 0);
> + igt_assert(set_freq(fd, gt_id, "max", rpn) > 0);
> + min.freq = get_freq(fd, gt_id, "act");
> + min.power = measure_power(fd, &gpu);
> +
> + igt_assert(set_freq(fd, gt_id, "min", rp0) > 0);
> + igt_assert(set_freq(fd, gt_id, "max", rp0) > 0);
> + max.freq = get_freq(fd, gt_id, "act");
> + max.power = measure_power(fd, &gpu);
> +
> + igt_info("Engine %s:%d min:%lumW @ %uMHz, max:%lumW @ %uMHz\n",
> + xe_engine_class_string(hwe->engine_class), hwe->engine_instance,
> + min.power, min.freq, max.power, max.freq);
> +
> + igt_spin_free(fd, spin);
> + put_ahnd(ahnd);
> + xe_vm_destroy(fd, vm);
> + igt_power_close(&gpu);
> +
> + /* Max power should be greater than min power + 10% of min power */
nit: reword: power@max_freq should be at least 10% greater than
power@min_freq?
> + igt_assert_f((11 * min.power < 10 * max.power),
> + "%s:%d did not conserve power when setting lower frequency!\n",
> + xe_engine_class_string(hwe->engine_class), hwe->engine_instance);
> +}
> +
> igt_main
> {
> struct drm_xe_engine_class_instance *hwe;
> @@ -475,6 +552,17 @@ igt_main
> }
> }
>
> + igt_describe("Validates power measured at lower frequencies");
nit: reword? Validate more power is consumed at higher frequencies?
> + igt_subtest("freq_power") {
> + /* FIXME: Remove skip once hwmon is added */
> + igt_skip_on(xe_has_vram(fd));
> + xe_for_each_gt(fd, gt) {
> + xe_for_each_hw_engine(fd, hwe) {
> + test_freq_power(fd, gt, hwe);
> + }
> + }
> + }
> +
LGTM,
Reviewed-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
> igt_fixture {
> xe_for_each_gt(fd, gt) {
> set_freq(fd, gt, "min", stash_min);
next prev parent reply other threads:[~2023-09-18 16:31 UTC|newest]
Thread overview: 9+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-09-07 6:06 [igt-dev] [PATCH i-g-t 0/2] Add freq_power test Riana Tauro
2023-09-07 6:06 ` [igt-dev] [PATCH i-g-t 1/2] tests/intel/xe_guc_pc: " Riana Tauro
2023-09-18 16:31 ` Belgaumkar, Vinay [this message]
2023-09-19 13:16 ` Kamil Konieczny
2023-09-21 7:06 ` Riana Tauro
2023-09-07 6:06 ` [igt-dev] [PATCH i-g-t 2/2] HAX: Add freq_power to xe-fast-feedback.testlist Riana Tauro
2023-09-07 8:20 ` [igt-dev] ✓ Fi.CI.BAT: success for Add freq_power test Patchwork
2023-09-07 9:09 ` [igt-dev] ✓ CI.xeBAT: " Patchwork
2023-09-07 9:58 ` [igt-dev] ✗ Fi.CI.IGT: failure " 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=7360a545-fe62-ea4e-b2fd-2eda5ab703e4@intel.com \
--to=vinay.belgaumkar@intel.com \
--cc=badal.nilawar@intel.com \
--cc=igt-dev@lists.freedesktop.org \
--cc=riana.tauro@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