| Message ID | 20190322092155.1656-3-chris@chris-wilson.co.uk (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [i-g-t,01/24] i915/gem_exec_latency: Measure the latency of context switching | expand |
On 22/03/2019 09:21, Chris Wilson wrote: > How much energy does spinning on a semaphore consume relative to plain > old spinning? > > Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> > --- > tests/i915/gem_exec_schedule.c | 72 +++++++++++++++++++++++++++++++++- > 1 file changed, 71 insertions(+), 1 deletion(-) > > diff --git a/tests/i915/gem_exec_schedule.c b/tests/i915/gem_exec_schedule.c > index a9383000a..4f0577b4e 100644 > --- a/tests/i915/gem_exec_schedule.c > +++ b/tests/i915/gem_exec_schedule.c > @@ -29,9 +29,10 @@ > #include <signal.h> > > #include "igt.h" > -#include "igt_vgem.h" > +#include "igt_gpu_power.h" > #include "igt_rand.h" > #include "igt_sysfs.h" > +#include "igt_vgem.h" > #include "i915/gem_ring.h" > > #define LO 0 > @@ -1202,6 +1203,65 @@ static void test_pi_ringfull(int fd, unsigned int engine) > munmap(result, 4096); > } > > +static void measure_semaphore_power(int i915) > +{ > + struct gpu_power power; > + unsigned int engine, signaler; > + > + igt_require(gpu_power_open(&power) == 0); > + > + for_each_physical_engine(i915, signaler) { > + struct gpu_power_sample s_spin[2]; > + struct gpu_power_sample s_sema[2]; > + double baseline, total; > + int64_t jiffie = 1; > + igt_spin_t *spin; > + > + spin = __igt_spin_batch_new(i915, > + .engine = signaler, > + .flags = IGT_SPIN_POLL_RUN); if (!spin) continue; To skip over !can_store_dword, since you are using the low level constructor which doesn't check. > + gem_wait(i915, spin->handle, &jiffie); /* waitboost */ Waitboost why? Will it be deterministic either way? > + igt_assert(spin->running); > + igt_spin_busywait_until_running(spin); > + > + gpu_power_read(&power, &s_spin[0]); > + usleep(100*1000); > + gpu_power_read(&power, &s_spin[1]); > + > + /* Add a waiter to each engine */ > + for_each_physical_engine(i915, engine) { > + igt_spin_t *sema; > + > + if (engine == signaler) > + continue; > + > + sema = __igt_spin_batch_new(i915, > + .engine = engine, > + .dependency = spin->handle); > + > + igt_spin_batch_free(i915, sema); > + } > + usleep(10); /* just give the tasklets a chance to run */ > + > + gpu_power_read(&power, &s_sema[0]); > + usleep(100*1000); > + gpu_power_read(&power, &s_sema[1]); > + > + igt_spin_batch_free(i915, spin); > + > + baseline = gpu_power_W(&power, &s_spin[0], &s_spin[1]); > + total = gpu_power_W(&power, &s_sema[0], &s_sema[1]); > + > + igt_info("%s: %.1fmW + %.1fmW (total %1.fmW)\n", > + e__->name, > + 1e3 * baseline, > + 1e3 * (total - baseline), > + 1e3 * total); > + } > + > + gpu_power_close(&power); > +} > + > igt_main > { > const struct intel_execution_engine *e; > @@ -1362,6 +1422,16 @@ igt_main > } > } > > + igt_subtest_group { > + igt_fixture { > + igt_require(gem_scheduler_enabled(fd)); > + igt_require(gem_scheduler_has_semaphores(fd)); > + } > + > + igt_subtest("semaphore-power") > + measure_semaphore_power(fd); > + } > + > igt_fixture { > igt_stop_hang_detector(); > close(fd); > Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com> Regards, Tvrtko
Quoting Tvrtko Ursulin (2019-03-26 08:46:34) > > On 22/03/2019 09:21, Chris Wilson wrote: > > +static void measure_semaphore_power(int i915) > > +{ > > + struct gpu_power power; > > + unsigned int engine, signaler; > > + > > + igt_require(gpu_power_open(&power) == 0); > > + > > + for_each_physical_engine(i915, signaler) { > > + struct gpu_power_sample s_spin[2]; > > + struct gpu_power_sample s_sema[2]; > > + double baseline, total; > > + int64_t jiffie = 1; > > + igt_spin_t *spin; > > + > > + spin = __igt_spin_batch_new(i915, > > + .engine = signaler, > > + .flags = IGT_SPIN_POLL_RUN); > > if (!spin) > continue; > > To skip over !can_store_dword, since you are using the low level > constructor which doesn't check. True, will need store-dword checking. Too much pain from spin_batch_new() mystery locking up inside loops has caused me to shy away from using it. > > + gem_wait(i915, spin->handle, &jiffie); /* waitboost */ > > Waitboost why? Will it be deterministic either way? There's nothing in here that should trigger a waitboost, the idea was to put the GPU into as pessimistic state as possible to measure peak power consumption. We could set the freq to min/max via sysfs to see if there is a significant difference. There's always another test waiting in the corner. -Chris
diff --git a/tests/i915/gem_exec_schedule.c b/tests/i915/gem_exec_schedule.c index a9383000a..4f0577b4e 100644 --- a/tests/i915/gem_exec_schedule.c +++ b/tests/i915/gem_exec_schedule.c @@ -29,9 +29,10 @@ #include <signal.h> #include "igt.h" -#include "igt_vgem.h" +#include "igt_gpu_power.h" #include "igt_rand.h" #include "igt_sysfs.h" +#include "igt_vgem.h" #include "i915/gem_ring.h" #define LO 0 @@ -1202,6 +1203,65 @@ static void test_pi_ringfull(int fd, unsigned int engine) munmap(result, 4096); } +static void measure_semaphore_power(int i915) +{ + struct gpu_power power; + unsigned int engine, signaler; + + igt_require(gpu_power_open(&power) == 0); + + for_each_physical_engine(i915, signaler) { + struct gpu_power_sample s_spin[2]; + struct gpu_power_sample s_sema[2]; + double baseline, total; + int64_t jiffie = 1; + igt_spin_t *spin; + + spin = __igt_spin_batch_new(i915, + .engine = signaler, + .flags = IGT_SPIN_POLL_RUN); + gem_wait(i915, spin->handle, &jiffie); /* waitboost */ + igt_assert(spin->running); + igt_spin_busywait_until_running(spin); + + gpu_power_read(&power, &s_spin[0]); + usleep(100*1000); + gpu_power_read(&power, &s_spin[1]); + + /* Add a waiter to each engine */ + for_each_physical_engine(i915, engine) { + igt_spin_t *sema; + + if (engine == signaler) + continue; + + sema = __igt_spin_batch_new(i915, + .engine = engine, + .dependency = spin->handle); + + igt_spin_batch_free(i915, sema); + } + usleep(10); /* just give the tasklets a chance to run */ + + gpu_power_read(&power, &s_sema[0]); + usleep(100*1000); + gpu_power_read(&power, &s_sema[1]); + + igt_spin_batch_free(i915, spin); + + baseline = gpu_power_W(&power, &s_spin[0], &s_spin[1]); + total = gpu_power_W(&power, &s_sema[0], &s_sema[1]); + + igt_info("%s: %.1fmW + %.1fmW (total %1.fmW)\n", + e__->name, + 1e3 * baseline, + 1e3 * (total - baseline), + 1e3 * total); + } + + gpu_power_close(&power); +} + igt_main { const struct intel_execution_engine *e; @@ -1362,6 +1422,16 @@ igt_main } } + igt_subtest_group { + igt_fixture { + igt_require(gem_scheduler_enabled(fd)); + igt_require(gem_scheduler_has_semaphores(fd)); + } + + igt_subtest("semaphore-power") + measure_semaphore_power(fd); + } + igt_fixture { igt_stop_hang_detector(); close(fd);
How much energy does spinning on a semaphore consume relative to plain old spinning? Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> --- tests/i915/gem_exec_schedule.c | 72 +++++++++++++++++++++++++++++++++- 1 file changed, 71 insertions(+), 1 deletion(-)