Skip to content

Increased calibration resolution #86

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
flhofer wants to merge 5 commits into
base: master
Choose a base branch
from
Open

Increased calibration resolution #86

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
6484a30
changed calib to hold ps. verify int size
flhofer Jun 13, 2019
f7298c2
fixed typo printf calib
flhofer Jun 13, 2019
be8bdc5
update from upstream 'master' into fix-calib-res
Sep 17, 2019
03955a4
modified options to parse ps times for calibration
Sep 17, 2019
33f4b68
Merge branch 'scheduler-tools:master' into fix-calib-res
flhofer Apr 20, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
48 changes: 24 additions & 24 deletions src/rt-app.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,7 +55,7 @@ static volatile sig_atomic_t continue_running;
static pthread_data_t *threads;
static int nthreads;
static volatile sig_atomic_t running_threads;
static int p_load;
static long p_load;
rtapp_options_t opts;
static struct timespec t_zero;
static struct timespec t_start;
Expand Down Expand Up @@ -225,12 +225,12 @@ void waste_cpu_cycles(unsigned long long load_loops)
* We alternate idle period and run period in order to not trig some hw
* protection mechanism like thermal mitgation
*/
int calibrate_cpu_cycles_1(int clock)
long calibrate_cpu_cycles_1(int clock)
{
struct timespec start, stop, sleep;
int max_load_loop = 10000;
unsigned int diff;
int nsec_per_loop, avg_per_loop = 0;
ulong diff;
long psec_per_loop, avg_per_loop = 0;
int cal_trial = 1000;

while (cal_trial) {
Expand All @@ -244,12 +244,12 @@ int calibrate_cpu_cycles_1(int clock)
waste_cpu_cycles(max_load_loop);
clock_gettime(clock, &stop);

diff = (int)timespec_sub_to_ns(&stop, &start);
nsec_per_loop = diff / max_load_loop;
avg_per_loop = (avg_per_loop + nsec_per_loop) >> 1;
diff = (ulong)timespec_sub_to_ns(&stop, &start);
psec_per_loop = diff * 1000 / max_load_loop;
avg_per_loop = (avg_per_loop + psec_per_loop) >> 1;

/* collect a critical mass of samples.*/
if ((abs(nsec_per_loop - avg_per_loop) * 50) < avg_per_loop)
if ((abs(psec_per_loop - avg_per_loop) * 50) < avg_per_loop)
return avg_per_loop;

/*
Expand All @@ -270,12 +270,12 @@ int calibrate_cpu_cycles_1(int clock)
* We continously runs something to ensure that CPU is set to max freq by the
* governor
*/
int calibrate_cpu_cycles_2(int clock)
long calibrate_cpu_cycles_2(int clock)
{
struct timespec start, stop;
int max_load_loop = 10000;
unsigned int diff;
int nsec_per_loop, avg_per_loop = 0;
ulong diff;
long psec_per_loop, avg_per_loop = 0;
int cal_trial = 1000;

while (cal_trial) {
Expand All @@ -285,12 +285,12 @@ int calibrate_cpu_cycles_2(int clock)
waste_cpu_cycles(max_load_loop);
clock_gettime(clock, &stop);

diff = (int)timespec_sub_to_ns(&stop, &start);
nsec_per_loop = diff / max_load_loop;
avg_per_loop = (avg_per_loop + nsec_per_loop) >> 1;
diff = (ulong)timespec_sub_to_ns(&stop, &start);
psec_per_loop = diff * 1000 / max_load_loop;
avg_per_loop = (avg_per_loop + psec_per_loop) >> 1;

/* collect a critical mass of samples.*/
if ((abs(nsec_per_loop - avg_per_loop) * 50) < avg_per_loop)
if ((abs(psec_per_loop - avg_per_loop) * 50) < avg_per_loop)
return avg_per_loop;

/*
Expand All @@ -310,9 +310,9 @@ int calibrate_cpu_cycles_2(int clock)
* Use several methods to calibrate the ns per loop and get the min value which
* correspond to the highest achievable compute capacity.
*/
int calibrate_cpu_cycles(int clock)
long calibrate_cpu_cycles(int clock)
{
int calib1, calib2;
long calib1, calib2;

/* Run 1st method */
calib1 = calibrate_cpu_cycles_1(clock);
Expand All @@ -337,7 +337,7 @@ static inline unsigned long loadwait(unsigned long exec)
* phase. We need to compute it here because both load_count and exec
* might be modified below.
*/
perf = exec / p_load;
perf = exec * 1000 / p_load;

/*
* If exec is still too big, let's run it in bursts
Expand All @@ -346,15 +346,15 @@ static inline unsigned long loadwait(unsigned long exec)
secs = exec / 1000000;

for (i = 0; i < secs; i++) {
load_count = 1000000000/p_load;
load_count = 1000000000000/p_load;
waste_cpu_cycles(load_count);
exec -= 1000000;
}

/*
* Run for the remainig exec (if any).
*/
load_count = (exec * 1000)/p_load;
load_count = (exec * 1000000)/p_load;
waste_cpu_cycles(load_count);

return perf;
Expand Down Expand Up @@ -1572,7 +1572,7 @@ int main(int argc, char* argv[])
continue_running = 1;

/* Needs to calibrate 'calib_cpu' core */
if (opts.calib_ns_per_loop == 0) {
if (opts.calib_ps_per_loop == 0) {
log_notice("Calibrate ns per loop");
cpu_set_t calib_set;

Expand All @@ -1582,10 +1582,10 @@ int main(int argc, char* argv[])
sched_setaffinity(0, sizeof(cpu_set_t), &calib_set);
p_load = calibrate_cpu_cycles(CLOCK_MONOTONIC);
sched_setaffinity(0, sizeof(cpu_set_t), &orig_set);
log_notice("pLoad = %dns : calib_cpu %d", p_load, opts.calib_cpu);
log_notice("pLoad = %ld.%03ldns : calib_cpu %d", p_load / 1000, p_load % 1000, opts.calib_cpu);
} else {
p_load = opts.calib_ns_per_loop;
log_notice("pLoad = %dns", p_load);
p_load = opts.calib_ps_per_loop;
log_notice("pLoad = %ld.%03ldns", p_load / 1000, p_load % 1000 );
}

initialize_cgroups();
Expand Down
14 changes: 9 additions & 5 deletions src/rt-app_parse_config.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1202,7 +1202,7 @@ parse_global(struct json_object *global, rtapp_options_t *opts)
opts->gnuplot = 0;
opts->policy = other;
opts->calib_cpu = 0;
opts->calib_ns_per_loop = 0;
opts->calib_ps_per_loop = 0;
opts->logdir = strdup("./");
opts->logbasename = strdup("rt-app");
opts->logsize = 0;
Expand Down Expand Up @@ -1232,13 +1232,17 @@ parse_global(struct json_object *global, rtapp_options_t *opts)
if (tmp_obj == NULL) {
/* no setting ? Calibrate CPU0 */
opts->calib_cpu = 0;
opts->calib_ns_per_loop = 0;
opts->calib_ps_per_loop = 0;
log_error("missing calibration setting force CPU0");
} else {
if (json_object_is_type(tmp_obj, json_type_int)) {
/* integer (no " ") detected. */
opts->calib_ns_per_loop = json_object_get_int(tmp_obj);
log_debug("ns_per_loop %d", opts->calib_ns_per_loop);
/* integer (no " " and ".") detected. */
opts->calib_ps_per_loop = json_object_get_int(tmp_obj) * 1000;
log_debug("ns_per_loop %ld.%03ld", opts->calib_ps_per_loop / 1000, opts->calib_ps_per_loop % 1000);
} else if (json_object_is_type(tmp_obj, json_type_double)) {
/* double (no " ") detected. */
opts->calib_ps_per_loop = (long) (json_object_get_double(tmp_obj) * 1000.0);
log_debug("ns_per_loop %ld.%03ld", opts->calib_ps_per_loop / 1000, opts->calib_ps_per_loop % 1000);
} else {
/* Get CPU number */
tmp_str = get_string_value_from(global, "calibration",
Expand Down
2 changes: 1 addition & 1 deletion src/rt-app_types.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -279,7 +279,7 @@ typedef struct _rtapp_options_t {
int logsize;
int gnuplot;
int calib_cpu;
int calib_ns_per_loop;
long calib_ps_per_loop; // option b

rtapp_resources_t *resources;
int pi_enabled;
Expand Down