@@ -59,6 +59,7 @@
#define PREEMPT 0x2
static unsigned int ring_size;
+static double rcs_clock;
static void
poll_ring(int fd, unsigned ring, const char *name)
@@ -238,10 +239,11 @@ static void latency_on_ring(int fd,
igt_assert(offset == obj[2].offset);
gem_set_domain(fd, obj[1].handle, I915_GEM_DOMAIN_GTT, 0);
- igt_info("%s: dispatch latency: %.2f, execution latency: %.2f (target %.2f)\n",
+ igt_info("%s: dispatch latency: %.1fns, execution latency: %.1fns (target %.1fns)\n",
name,
- (end - start) / (double)repeats,
- gpu_latency, (results[repeats - 1] - results[0]) / (double)(repeats - 1));
+ (end - start) / (double)repeats * rcs_clock,
+ gpu_latency * rcs_clock,
+ (results[repeats - 1] - results[0]) / (double)(repeats - 1) * rcs_clock);
munmap(map, 64*1024);
munmap(results, 4096);
@@ -620,6 +622,30 @@ rthog_latency_on_ring(int fd, unsigned int engine, const char *name, unsigned in
munmap(results, MMAP_SZ);
}
+static double clockrate(int reg)
+{
+ volatile uint32_t *mmio;
+ uint32_t r_start, r_end;
+ struct timespec tv;
+ uint64_t t_start, t_end;
+ uint64_t elapsed;
+
+ mmio = (volatile uint32_t *)((volatile char *)igt_global_mmio + reg);
+
+ t_start = igt_nsec_elapsed(&tv);
+ r_start = *mmio;
+ elapsed = igt_nsec_elapsed(&tv) - t_start;
+
+ usleep(1000);
+
+ t_end = igt_nsec_elapsed(&tv);
+ r_end = *mmio;
+ elapsed += igt_nsec_elapsed(&tv) - t_end;
+
+ elapsed = (t_end - t_start) + elapsed / 2;
+ return (r_end - r_start) * 1e9 / elapsed;
+}
+
igt_main
{
const struct intel_execution_engine *e;
@@ -640,6 +666,10 @@ igt_main
ring_size = 1024;
intel_register_access_init(intel_get_pci_device(), false, device);
+ rcs_clock = clockrate(RCS_TIMESTAMP);
+ igt_info("RCS timestamp clock: %.3fKHz, %.1fns\n",
+ rcs_clock / 1e3, 1e9 / rcs_clock);
+ rcs_clock = 1e9 / rcs_clock;
}
igt_subtest("all-rtidle-submit")
Present the latency results in nanoseconds not RCS cycles. Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> --- tests/i915/gem_exec_latency.c | 36 ++++++++++++++++++++++++++++++++--- 1 file changed, 33 insertions(+), 3 deletions(-)