Skip to content

Commit

Permalink
toplev: add --columns option to show cpu values side-by-side
Browse files Browse the repository at this point in the history 
This requires a buffered output class, that reorders the data.

Also minor cleanups/improvements.
  • Loading branch information
Andi Kleen committed May 3, 2015
1 parent 97409ab commit cf48cff
Show file tree
Hide file tree
Showing 3 changed files with 110 additions and 19 deletions.
1 change: 1 addition & 0 deletions README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@ analysis on Intel CPUs on top of [Linux perf](https://perf.wiki.kernel.org/index

# Recent new features:

* toplev has a new --columns option to print results from different CPUs side-by-side.
* toplev now has a (draft) [tutorial and manual](https://github.com/andikleen/pmu-tools/wiki/toplev-manual)
* toplev now detects kernel counter scheduling problems based on the kernel version and automatically
disables nodes with unsupported events. Can be overridden with --force-events.
Expand Down
3 changes: 3 additions & 0 deletions tl-tester
View file
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,7 @@ EVENTMAP=${cpus[$DCPU]} FORCECPU=$DCPU $WRAP ./toplev.py $LOAD
DIRECT_MSR=1 EVENTMAP=${cpus[$DCPU]} FORCECPU=$DCPU $WRAP ./toplev.py $LOAD
EVENTMAP=${cpus[$DCPU]} FORCECPU=$DCPU $WRAP ./toplev.py --no-desc -d -l4 $LOAD
EVENTMAP=${cpus[$DCPU]} FORCECPU=$DCPU $WRAP ./toplev.py --no-desc -d -l4 --single-thread $LOAD
EVENTMAP=${cpus[$DCPU]} FORCECPU=$DCPU $WRAP ./toplev.py --no-desc -d -l4 --single-thread --columns $LOAD
EVENTMAP=${cpus[$DCPU]} FORCECPU=$DCPU $WRAP ./toplev.py --no-desc -d --all --core S0-C0-T0 $LOAD0
EVENTMAP=${cpus[$DCPU]} FORCECPU=$DCPU $WRAP ./toplev.py --no-desc -v -d -l4 $LOAD
EVENTMAP=${cpus[$DCPU]} FORCECPU=$DCPU $WRAP ./toplev.py --no-desc -x, -v -d -l4 $LOAD
Expand Down Expand Up @@ -95,11 +96,13 @@ done

# misc features
EVENTMAP=${cpus[hsw]} FORCECPU=hsw $WRAP ./toplev.py -d --stats --metrics --no-multiplex -l4 $LOAD
EVENTMAP=${cpus[hsw]} FORCECPU=hsw $WRAP ./toplev.py -d --stats --metrics --no-multiplex --columns -l4 $LOAD
EVENTMAP=${cpus[hsw]} FORCECPU=hsw $WRAP ./toplev.py -d --no-desc --power -l4 $LOAD
EVENTMAP=${cpus[hsw]} FORCECPU=hsw $WRAP ./toplev.py -d --no-desc $ALL --no-group $LOAD
EVENTMAP=${cpus[hsw]} FORCECPU=hsw $WRAP ./toplev.py -d --no-desc --sw -l4 $LOAD
EVENTMAP=${cpus[hsw]} FORCECPU=hsw $WRAP ./toplev.py --graph -o x.png -d --metrics -l4 $LOAD
EVENTMAP=${cpus[ivb]} FORCECPU=ivb $WRAP ./toplev.py -d -l4 -I 100 sleep 1
EVENTMAP=${cpus[ivb]} FORCECPU=ivb $WRAP ./toplev.py -d -l4 -I 100 --columns sleep 1
EVENTMAP=${cpus[ivb]} FORCECPU=ivb $WRAP ./toplev.py -d -l4 --valcsv val.csv --user $LOAD | tee log
grep :u log
grep /u log
Expand Down
125 changes: 106 additions & 19 deletions toplev.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -282,6 +282,7 @@ def event_group(evlist):
p.add_argument('--long-desc', help='Print long descriptions instead of abbreviated ones.',
action='store_true')
p.add_argument('--force-events', help='Assume kernel supports all events. May give wrong results.', action='store_true')
p.add_argument('--columns', help='Print CPU output in multiple columns', action='store_true')
args, rest = p.parse_known_args()

if args.version:
Expand Down Expand Up @@ -358,7 +359,6 @@ def format_valstat(valstat):
class Output:
"""Generate human readable output."""
def __init__(self, logfile):
self.sep = " "
self.logf = logfile
self.printed_descs = set()
self.hdrlen = 46
Expand All @@ -369,6 +369,24 @@ def set_hdr(self, hdr, area):
hdr = "%-7s %s" % (area, hdr)
self.hdrlen = min(max(len(hdr) + 1, self.hdrlen), 78)

def set_cpus(self, cpus):
pass

def print_desc(self, desc, sample):
if desc and not args.no_desc:
print >>self.logf, "\t" + desc
if desc and sample and not args.no_desc:
print >>self.logf, "\t" + "Sampling events: ", sample

def print_timestamp(self, timestamp):
if timestamp:
self.logf.write("%6.9f%s" % (timestamp, " "))

def print_header(self, area, hdr):
hdr = "%-7s %s" % (area, hdr)
#hdroff = min(len(hdr), self.hdrlen)
print >>self.logf, "%-*s " % (self.hdrlen, hdr + ":"),

# timestamp Timestamp in interval mode
# title CPU
# area FE/BE ...
Expand All @@ -381,18 +399,14 @@ def set_hdr(self, hdr, area):
# Example:
# C0 BE Backend_Bound: 62.00 %
def show(self, timestamp, title, area, hdr, s, remark, desc, sample, valstat):
if timestamp:
self.logf.write("%6.9f%s" % (timestamp, self.sep))
self.print_timestamp(timestamp)
if title:
self.logf.write("%-6s" % (title))
vs = format_valstat(valstat)
hdr = "%-7s %s" % (area, hdr)
hdroff = min(len(s), self.hdrlen)
print >>self.logf, "%-*s %s %s" % (self.hdrlen - hdroff + 5, hdr + ":", s, remark + vs)
if desc and not args.no_desc:
print >>self.logf, "\t" + desc
if desc and sample and not args.no_desc:
print >>self.logf, "\t" + "Sampling events: ", sample
self.print_header(area, hdr)
print >>self.logf, s,
print >>self.logf, remark + vs
self.print_desc(desc, sample)

def item(self, area, name, l, timestamp, remark, desc, title, sample, valstat):
if desc in self.printed_descs:
Expand All @@ -404,17 +418,83 @@ def item(self, area, name, l, timestamp, remark, desc, title, sample, valstat):
self.show(timestamp, title, area, name, l, remark, desc, sample, valstat)

def ratio(self, area, name, l, timestamp, remark, desc, title, sample, valstat):
self.item(area, name, "%5s" % ("%2.2f" % (100.0 * l)), timestamp, "% " + remark, desc, title,
self.item(area, name, "%5s" % ("%2.2f%%" % (100.0 * l)), timestamp, remark, desc, title,
sample, valstat)

def metric(self, area, name, l, timestamp, desc, title, unit, valstat):
self.item(area, name, "%5s" % ("%3.2f" % (l)), timestamp, unit, desc, title,
None, valstat)

def flush(self):
pass

def csv_writer(f, sep):
return csv.writer(f, delimiter=sep)

class OutputBuffered(Output):
"""Output data in per-cpu columns."""
def __init__(self, logfile):
Output.__init__(self, logfile)
self.nodes = dict()
self.timestamp = None
self.cpunames = set()

def set_cpus(self, cpus):
self.cpunames = cpus

def show(self, timestamp, title, area, hdr, s, remark, desc, sample, valstat):
if args.single_thread:
Output.show(self, timestamp, title, area, hdr, s, remark, desc, sample, valstat)
return
self.timestamp = timestamp
key = (area, hdr)
if key not in self.nodes:
self.nodes[key] = dict()
assert title not in self.nodes[key]
self.nodes[key][title] = (s, remark, desc, sample, valstat)

def flush(self):
VALCOL_LEN = 10
write = self.logf.write

cpunames = self.cpunames

if self.timestamp:
write("%9s" % "")
self.logf.write("%*s" % (self.hdrlen, ""))
for j in sorted(cpunames):
write("%*s " % (VALCOL_LEN, j))

write("\n")
for key in sorted(sorted(self.nodes.keys(), key=lambda x: x[1]), key=lambda x: x[0] == ""):
node = self.nodes[key]
desc = None
sample = None
remark = None
if self.timestamp:
self.print_timestamp(self.timestamp)

self.print_header(key[0], key[1])
vlist = []
for cpuname in cpunames:
if cpuname in node:
cpu = node[cpuname]
desc, sample, remark, valstat = cpu[2], cpu[3], cpu[1], cpu[4]
if remark in ("above", "below"): # XXX
remark = ""
if valstat:
vlist.append(valstat)
write("%-*s " % (VALCOL_LEN, cpu[0]))
else:
write("%*s " % (VALCOL_LEN, ""))
if remark:
write(" " + remark + format_valstat(combine_valstat(vlist)))
write("\n")
self.print_desc(desc, sample)
self.nodes = dict()

class OutputCSV(Output):
"""Output data in CSV format."""
def __init__(self, logfile, sep):
Output.__init__(self, logfile)
self.writer = csv_writer(self.logf, sep)
Expand Down Expand Up @@ -746,8 +826,18 @@ def matching(name, mapping):
def geoadd(l):
return math.sqrt(sum([x**2 for x in l]))

# use geomean of stddevs and minimum of multiplex ratios for combining
# XXX better way to combine multiplex ratios?
def combine_valstat(l):
if not l:
return []
return ValStat(geoadd([x.stddev for x in l]), min([x.multiplex for x in l]))

def print_keys(runner, res, rev, valstats, out, interval, env):
stat = runner.stat
if len(res.keys()) > 1:
cores = [key_to_coreid(x) for x in res.keys() if not args.core or display_core(int(x))]
out.set_cpus(set(map(core_fmt, cores) + map(thread_fmt, cores)))
if smt_mode:
# compute non SMT nodes, but don't print yet
# this is needed for getting the thresholds correct when
Expand All @@ -762,13 +852,7 @@ def print_keys(runner, res, rev, valstats, out, interval, env):
for core, citer in itertools.groupby(core_keys, key_to_coreid):
cpus = list(citer)
r = list(itertools.izip(*[res[j] for j in cpus]))

# use geomean of stddevs and minimum of multiplex ratios for combining
# XXX better way to combine multiplex ratios?
st = [ValStat(
geoadd([x.stddev for x in z]),
min([x.multiplex for x in z]))
for z in itertools.izip(*[valstats[j] for j in cpus])]
st = [combine_valstat(z) for z in itertools.izip(*[valstats[j] for j in cpus])]
runner.compute(r, rev[cpus[0]], st, env, smt_node, stat)
runner.print_res(out, interval, core_fmt(core), smt_node)

Expand All @@ -783,6 +867,7 @@ def print_keys(runner, res, rev, valstats, out, interval, env):
continue
runner.compute(res[j], rev[j], valstats[j], env, lambda obj: True, stat)
runner.print_res(out, interval, j, lambda obj: True)
out.flush()
stat.referenced_check(res)
stat.compute_errors()

Expand Down Expand Up @@ -1082,7 +1167,7 @@ def sample_event(e):

def sample_desc(s):
try:
return ",".join([sample_event(x) for x in s])
return " ".join([sample_event(x) for x in s])
except BadEvent as e:
#return "Unknown sample event %s" % (e.event)
return ""
Expand Down Expand Up @@ -1553,6 +1638,8 @@ def setup_with_metrics(p, runner):
runner.collect()
if csv_mode:
out = OutputCSV(args.output, csv_mode)
elif args.columns:
out = OutputBuffered(args.output)
else:
out = Output(args.output)
runner.schedule()
Expand Down

0 comments on commit cf48cff

Please sign in to comment.