Move the statistics calculation out of the periodic threads to avoid
performing those computation at an RT priority thus impacting the other
threads.
This allows for a nicer output (not intermingled) as the stats are
displayed once all the threads have exited.
Signed-off-by: Sebastien Dugue <[EMAIL PROTECTED]>
Cc: Darren Hart <[EMAIL PROTECTED]>
Cc: Tim Chavez <[EMAIL PROTECTED]>
---
.../func/periodic_cpu_load/periodic_cpu_load.c | 74 +++++++++++++-------
1 files changed, 48 insertions(+), 26 deletions(-)
diff --git a/testcases/realtime/func/periodic_cpu_load/periodic_cpu_load.c
b/testcases/realtime/func/periodic_cpu_load/periodic_cpu_load.c
index 6a6cd54..ba65699 100644
--- a/testcases/realtime/func/periodic_cpu_load/periodic_cpu_load.c
+++ b/testcases/realtime/func/periodic_cpu_load/periodic_cpu_load.c
@@ -56,6 +56,7 @@
#define CALC_LOOPS_B (50*100)
#define CALC_LOOPS_C (72*100)
+#define NUM_GROUPS 3
#define THREADS_PER_GROUP 4
//#define ITERATIONS 100 /* short functional test run */
@@ -64,6 +65,11 @@
// FIXME: need some kind of passing criteria calculation
//#define PASS_US 100
+int fail[THREADS_PER_GROUP * NUM_GROUPS];
+stats_container_t dat[THREADS_PER_GROUP * NUM_GROUPS];
+stats_quantiles_t quantiles[THREADS_PER_GROUP * NUM_GROUPS];
+static const char groupname[NUM_GROUPS] = "ABC";
+
static int run_jvmsim = 0;
static int ret = 0;
@@ -120,47 +126,31 @@ void *periodic_thread(void *thread)
nsec_t period = parg->period;
void*(*func)(void*) = parg->func;
- stats_container_t dat;
- stats_quantiles_t quantiles;
-
int i = 0;
- int fail = 0;
nsec_t next;
nsec_t exe_start, exe_end, exe_time;
- stats_container_init(&dat, ITERATIONS);
- stats_quantiles_init(&quantiles, 3);
-
next = rt_gettime();
while (i < parg->iterations) {
next += period;
if (rt_gettime() > next) {
printf("TID %d missed period, aborting\n", t->id);
- fail = 1;
+ fail[t->id] = 1;
break;
}
exe_start = rt_gettime();
func(parg->arg);
exe_end = rt_gettime();
exe_time = exe_end - exe_start;
- dat.records[i].x = i;
- dat.records[i].y = exe_time/NS_PER_US;
+ dat[t->id].records[i].x = i;
+ dat[t->id].records[i].y = exe_time/NS_PER_US;
i++;
rt_nanosleep(next - rt_gettime());
}
- printf("\nTID %d (prio %d) complete\n", t->id, t->priority);
- printf(" Execution Time Statistics:\n");
- printf(" Min: %ld us\n", stats_min(&dat));
- printf(" Max: %ld us\n", stats_max(&dat));
- printf(" Avg: %f us\n", stats_avg(&dat));
- printf(" StdDev: %f us\n\n", stats_stddev(&dat));
- printf(" Quantiles:\n");
- stats_quantiles_calc(&dat, &quantiles);
- stats_quantiles_print(&quantiles);
- printf("Criteria:TID %d missed a period\n",t->id);
- printf("Result: %s\n", fail ? "FAIL":"PASS");
+ printf("TID %d (%c - prio %d) complete\n", t->id, groupname[t->id>>2],
+ t->priority);
return NULL;
}
@@ -188,6 +178,7 @@ int main(int argc, char *argv[])
printf(" threads: %d\n", THREADS_PER_GROUP);
printf(" priority: %d\n", PRIO_C);
printf(" period: %d ms\n", PERIOD_C/NS_PER_MS);
+ printf("\n");
if (run_jvmsim) {
printf("jvmsim enabled\n");
@@ -196,20 +187,51 @@ int main(int argc, char *argv[])
printf("jvmsim disabled\n");
}
- struct periodic_arg parg_a = { PERIOD_A, ITERATIONS, calc, (void
*)CALC_LOOPS_A };
- struct periodic_arg parg_b = { PERIOD_B, ITERATIONS, calc, (void
*)CALC_LOOPS_B };
- struct periodic_arg parg_c = { PERIOD_C, ITERATIONS, calc, (void
*)CALC_LOOPS_C };
- for (i=0; i < THREADS_PER_GROUP; i++) {
+ for (i=0; i<(THREADS_PER_GROUP * NUM_GROUPS); i++) {
+ stats_container_init(&dat[i], ITERATIONS);
+ stats_quantiles_init(&quantiles[i], log10(ITERATIONS));
+ }
+
+ struct periodic_arg parg_a = {PERIOD_A, ITERATIONS, calc, (void
*)CALC_LOOPS_A };
+ struct periodic_arg parg_b = {PERIOD_B, ITERATIONS, calc, (void
*)CALC_LOOPS_B };
+ struct periodic_arg parg_c = {PERIOD_C, ITERATIONS, calc, (void
*)CALC_LOOPS_C };
+
+ for (i=0; i < THREADS_PER_GROUP; i++)
create_fifo_thread(periodic_thread, (void*)&parg_a, PRIO_A);
+ for (i=0; i < THREADS_PER_GROUP; i++)
create_fifo_thread(periodic_thread, (void*)&parg_b, PRIO_B);
+ for (i=0; i < THREADS_PER_GROUP; i++)
create_fifo_thread(periodic_thread, (void*)&parg_c, PRIO_C);
- }
join_threads();
+ printf("\nExecution Time Statistics:\n\n");
+
+ for (i=0; i<(THREADS_PER_GROUP * NUM_GROUPS); i++) {
+ printf("TID %d (%c)\n", i, groupname[i>>2]);
+ printf(" Min: %ld us\n", stats_min(&dat[i]));
+ printf(" Max: %ld us\n", stats_max(&dat[i]));
+ printf(" Avg: %f us\n", stats_avg(&dat[i]));
+ printf(" StdDev: %f us\n\n", stats_stddev(&dat[i]));
+ printf(" Quantiles:\n");
+ stats_quantiles_calc(&dat[i], &quantiles[i]);
+ stats_quantiles_print(&quantiles[i]);
+ printf("Criteria: TID %d did not miss a period\n", i);
+ printf("Result: %s\n", fail[i] ? "FAIL":"PASS");
+ printf("\n");
+
+ if (fail[i])
+ ret = 1;
+ }
+
// FIXME: define pass criteria
// printf("\nCriteria: latencies < %d us\n", PASS_US);
// printf("Result: %s\n", ret ? "FAIL" : "PASS");
+ for (i=0; i<(THREADS_PER_GROUP * NUM_GROUPS); i++) {
+ stats_container_free(&dat[i]);
+ stats_quantiles_free(&quantiles[i]);
+ }
+
return ret;
}
--
1.5.4.rc2.84.gf85fd-dirty
-------------------------------------------------------------------------
Check out the new SourceForge.net Marketplace.
It's the best place to buy or sell services for
just about anything Open Source.
http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace
_______________________________________________
Ltp-list mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/ltp-list
