def samples_to_df(h, need_line):
"""Convert a parsed perf event list to a pandas table.
The pandas table contains all events in a easily to process format.
The pandas table has callchain_aux and branch_aux fields pointing
to Aux object defining the callchains/branches."""
ev = perfdata.get_events(h)
index = []
data = defaultdict(list)
callchains = Aux()
branches = Aux()
used = Counter()
mm = mmap.MmapTracker()
for n in range(0, len(ev)):
mm.lookahead_mmap(ev, n)
j = ev[n]
if j.type != "SAMPLE":
continue
mm.update_sample(j)
add = lambda k, i: do_add(data, used, k, i)
filename, mmap_base, foffset = mm.resolve(j.pid, j.ip)
if filename == "[kernel.kallsyms]_text":
filename = None
add('filename', filename)
sym, soffset, line = elf.resolve_ip(filename, foffset, j.ip, need_line)
add('symbol', sym)
add('line', line)
add('soffset', soffset)
if 'callchain' in j and j.callchain:
id = callchains.add(j.callchain.caller,
lambda: resolve_chain(j.callchain, j, mm, need_line))
add('callchain', id)
if 'branch_stack' in j and j.branch_stack:
branch = j.branch_stack.branch
id = branches.add(map(lambda x: (x['from'], x.to), branch),
lambda: resolve_branch(branch, j, mm, need_line))
add('branch', id)
kernel, guest, hv = cpumodes[j['cpumode']]
add('kernel', kernel)
add('guest', guest)
add('hv', hv)
for name in j:
if name not in ignored:
if j[name]:
used[name] += 1
data[name].append(j[name])
index.append(int(j["time"]))
for j in data.keys():
if used[j] == 0:
del data[j]
df = pd.DataFrame(data, index=index, dtype=np.uint64)
for i in bool_fields:
df[i] = df[i].astype('bool')
df.branch_aux = branches
df.callchain_aux = callchains
return df
def samples_to_df(h):
ev = perfdata.get_events(h)
index = []
data = defaultdict(list)
procs = dict()
maps = defaultdict(list)
pnames = defaultdict(str)
used = Counter()
# comm do not necessarily appear in order
# first build queue of comm in order
updates = []
for j in ev:
# no time stamp: assume it's synthesized and kernel
if j.type == 'MMAP' and j.pid == -1 and j.tid == 0:
bisect.insort(maps[j.pid], (j.addr, j.len, j.filename))
elif j.type in ('COMM','MMAP'):
bisect.insort(updates, (j.time2, j))
for j in ev:
add = lambda k, i: do_add(data, used, k, i)
if j.type != "SAMPLE":
continue
# process pending updates
while len(updates) > 0 and j.time >= updates[0][0]:
u = updates[0][1]
del updates[0]
if u.type == 'MMAP':
pid = u.pid
bisect.insort(maps[pid], (u.addr, u.len, u.filename))
elif u.type == 'COMM':
maps[u.pid] = []
pnames[u.pid] = u.comm
filename, offset = resolve(maps, j.pid, j.ip)
add('filename', filename)
add('foffset', offset)
sym, offset, line = None, None, None
if filename and filename.startswith("/"):
sym, offset, line = elf.resolve_addr(filename, j.ip)
add('symbol', sym)
add('soffset', offset)
add('line', line)
#if 'callchain' in j:
# resolve_chain(j['callchain'], j, maps)
for name in j:
if name not in ignored:
if j[name]:
used[name] += 1
data[name].append(j[name])
index.append(pd.Timestamp(j["time"]))
for j in data.keys():
if used[j] == 0:
del data[j]
return pd.DataFrame(data, index=index, dtype=np.uint64)
def samples_to_df(h):
ev = perfdata.get_events(h)
index = []
data = defaultdict(list)
procs = dict()
maps = defaultdict(list)
pnames = defaultdict(str)
used = Counter()
# comm do not necessarily appear in order
# first build queue of comm in order
updates = []
for j in ev:
# no time stamp: assume it's synthesized and kernel
if j.type == 'MMAP' and j.pid == -1 and j.tid == 0:
bisect.insort(maps[j.pid], (j.addr, j.len, j.filename))
elif j.type in ('COMM','MMAP'):
bisect.insort(updates, (j.time2, j))
for j in ev:
if j.type != "SAMPLE":
continue
# process pending updates
while len(updates) > 0 and j.time >= updates[0][0]:
u = updates[0][1]
del updates[0]
if u.type == 'MMAP':
pid = u.pid
bisect.insort(maps[pid], (u.addr, u.len, u.filename))
elif u.type == 'COMM':
maps[u.pid] = []
pnames[u.pid] = u.comm
filename, offset = resolve(maps, j.pid, j.ip)
data['filename'].append(filename)
data['offset'].append(offset)
used['offset'] += 1
used['filename'] += 1
for name in j:
if name not in ignored:
if j[name]:
used[name] += 1
data[name].append(j[name])
# XXX assumes time exists
index.append(pd.Timestamp(j["time"]))
for j in data.keys():
if used[j] == 0:
del data[j]
return pd.DataFrame(data, index=index, dtype=np.uint64)
def samples_to_df(h, need_line):
"""Convert a parsed perf event list to a pandas table.
The pandas table contains all events in a easy to process format.
The pandas table has callchain_aux and branch_aux fields pointing
to Aux object defining the callchains/branches."""
ev = perfdata.get_events(h)
index = []
data = defaultdict(list)
callchains = Aux()
branches = Aux()
used = Counter()
mm = mmap.MmapTracker()
numsample = 0
for n in range(0, len(ev)):
mm.lookahead_mmap(ev, n)
j = ev[n]
print(j)
if j.type != "SAMPLE":
continue
numsample += 1
mm.update_sample(j)
def add(k, i):
data[k].append(i)
used[k] += 1
filename, mmap_base, foffset = mm.resolve(j.pid, j.ip)
if filename == "[kernel.kallsyms]_text":
filename = None
add('filename', filename)
sym, soffset, line = elf.resolve_ip(filename, foffset, j.ip, need_line)
add('symbol', sym)
add('line', line)
add('soffset', soffset)
if 'callchain' in j and j.callchain:
id = callchains.add(
j.callchain.caller,
lambda: resolve_chain(j.callchain, j, mm, need_line))
add('callchain', id)
if 'branch_stack' in j and j.branch_stack:
branch = j.branch_stack.branch
id = branches.add(map(lambda x: (x['from'], x.to), branch),
lambda: resolve_branch(branch, j, mm, need_line))
add('branch', id)
kernel, guest, hv = cpumodes[j['cpumode']]
add('kernel', kernel)
add('guest', guest)
add('hv', hv)
for name in j:
if name not in ignored:
if j[name]:
used[name] += 1
data[name].append(j[name])
index.append(int(j["time"]))
for j in data.keys():
if used[j] == 0:
del data[j]
df = pd.DataFrame(data, index=index, dtype=np.uint64)
if numsample > 0:
for i in bool_fields:
df[i] = df[i].astype('bool')
df.branch_aux = branches
df.callchain_aux = callchains
return df