def func(opr,
*,
f0=TimeFuncHelper.eval_time_func(prof_type, args.top_end_key,
np.max)):
t = f0(opr)
if t is not None and (t < args.min_time or t > args.max_time):
return None
return t
def summary():
device_end_func = TimeFuncHelper.eval_time_func("device", "end", np.max)
device_kern_func = TimeFuncHelper.eval_time_func("device", "kern", np.max)
host_end_func = TimeFuncHelper.eval_time_func("host", "end", np.max)
def get_tot_time(func):
rec = analyzer_tot.select(func, aggregate=np.sum)
if not rec:
return "N/A"
rec = rec[0]
return rec.time
tab = []
tot_dev_time = get_tot_time(device_end_func)
tot_host_time = get_tot_time(host_end_func)
tab.append(("total device time", tot_dev_time))
tab.append(("total host time", tot_host_time))
if args.copy_time:
def fmt(a, b):
a = a[0]
b = b[0]
return "tot={:.4f} avg={:.4f}".format(a.time, b.time)
tab.append(
(
"copy time",
fmt(
analyzer.select(
device_end_func,
lambda opr: opr.opr_info["type"] == "Copy",
aggregate=np.sum,
),
analyzer.select(
device_end_func,
lambda opr: opr.opr_info["type"] == "Copy",
aggregate=np.mean,
),
),
)
)
tab.append(
(
"copy wait time",
fmt(
analyzer.select(
device_kern_func,
lambda opr: opr.opr_info["type"] == "Copy",
aggregate=np.sum,
),
analyzer.select(
device_kern_func,
lambda opr: opr.opr_info["type"] == "Copy",
aggregate=np.mean,
),
),
)
)
if args.confluence:
tab_str = _tabulate_confluence(tab, headers=["name", "value"])
else:
tab_str = tabulate(tab)
return tab_str, tot_dev_time, tot_host_time