[Intel-gfx] [PATCH 06/22] drm/i915/selftests: Repeat the rps clock frequency measurement

[Mika Kuoppala]

Chris Wilson <chris at chris-wilson.co.uk> writes:

> Repeat the measurement of the clock frequency a few times and use the
> median to try and reduce the systematic measurement error.
>
> Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
> ---
>  drivers/gpu/drm/i915/gt/selftest_rps.c | 54 +++++++++++++++++++-------
>  1 file changed, 40 insertions(+), 14 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/gt/selftest_rps.c b/drivers/gpu/drm/i915/gt/selftest_rps.c
> index b89a7d7611f6..bfa1a15564f7 100644
> --- a/drivers/gpu/drm/i915/gt/selftest_rps.c
> +++ b/drivers/gpu/drm/i915/gt/selftest_rps.c
> @@ -56,6 +56,18 @@ static int cmp_u64(const void *A, const void *B)
>  		return 0;
>  }
>  
> +static int cmp_u32(const void *A, const void *B)
> +{
> +	const u32 *a = A, *b = B;
> +
> +	if (a < b)
> +		return -1;
> +	else if (a > b)
> +		return 1;
> +	else
> +		return 0;
> +}
> +
>  static struct i915_vma *
>  create_spin_counter(struct intel_engine_cs *engine,
>  		    struct i915_address_space *vm,
> @@ -236,8 +248,8 @@ int live_rps_clock_interval(void *arg)
>  	for_each_engine(engine, gt, id) {
>  		unsigned long saved_heartbeat;
>  		struct i915_request *rq;
> -		ktime_t dt;
>  		u32 cycles;
> +		u64 dt;
>  
>  		if (!intel_engine_can_store_dword(engine))
>  			continue;
> @@ -286,15 +298,29 @@ int live_rps_clock_interval(void *arg)
>  				  engine->name);
>  			err = -ENODEV;
>  		} else {
> -			preempt_disable();
> -			dt = ktime_get();
> -			cycles = -intel_uncore_read_fw(gt->uncore,
> -						       GEN6_RP_CUR_UP_EI);
> -			udelay(1000);
> -			dt = ktime_sub(ktime_get(), dt);
> -			cycles += intel_uncore_read_fw(gt->uncore,
> -						       GEN6_RP_CUR_UP_EI);
> -			preempt_enable();
> +			ktime_t dt_[5];
> +			u32 cycles_[5];
> +			int i;
> +
> +			for (i = 0; i < 5; i++) {

Could be sizeof.

Reviewed-by: Mika Kuoppala <mika.kuoppala at linux.intel.com>

> +				preempt_disable();
> +
> +				dt_[i] = ktime_get();
> +				cycles_[i] = -intel_uncore_read_fw(gt->uncore, GEN6_RP_CUR_UP_EI);
> +
> +				udelay(1000);
> +
> +				dt_[i] = ktime_sub(ktime_get(), dt_[i]);
> +				cycles_[i] += intel_uncore_read_fw(gt->uncore, GEN6_RP_CUR_UP_EI);
> +
> +				preempt_enable();
> +			}
> +
> +			/* Use the median of both cycle/dt; close enough */
> +			sort(cycles_, 5, sizeof(*cycles_), cmp_u32, NULL);
> +			cycles = (cycles_[1] + 2 * cycles_[2] + cycles_[3]) / 4;
> +			sort(dt_, 5, sizeof(*dt_), cmp_u64, NULL);
> +			dt = div_u64(dt_[1] + 2 * dt_[2] + dt_[3], 4);
>  		}
>  
>  		intel_uncore_write_fw(gt->uncore, GEN6_RP_CONTROL, 0);
> @@ -306,14 +332,14 @@ int live_rps_clock_interval(void *arg)
>  		if (err == 0) {
>  			u64 time = intel_gt_pm_interval_to_ns(gt, cycles);
>  			u32 expected =
> -				intel_gt_ns_to_pm_interval(gt, ktime_to_ns(dt));
> +				intel_gt_ns_to_pm_interval(gt, dt);
>  
>  			pr_info("%s: rps counted %d C0 cycles [%lldns] in %lldns [%d cycles], using GT clock frequency of %uKHz\n",
> -				engine->name, cycles, time, ktime_to_ns(dt), expected,
> +				engine->name, cycles, time, dt, expected,
>  				gt->clock_frequency / 1000);
>  
> -			if (10 * time < 8 * ktime_to_ns(dt) ||
> -			    8 * time > 10 * ktime_to_ns(dt)) {
> +			if (10 * time < 8 * dt ||
> +			    8 * time > 10 * dt) {
>  				pr_err("%s: rps clock time does not match walltime!\n",
>  				       engine->name);
>  				err = -EINVAL;
> -- 
> 2.20.1
>
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx