diff options
author | Martin Nowack <m.nowack@imperial.ac.uk> | 2020-02-20 12:45:32 +0000 |
---|---|---|
committer | Cristian Cadar <c.cadar@imperial.ac.uk> | 2020-03-01 20:29:18 +0000 |
commit | f62f8b954a00ae8538f8d3ef2aad77c11e6ed710 (patch) | |
tree | 10e6bd41e306924c43c1675c83de0e5dde75785d | |
parent | bab3011da22f3569fb1e70ecf8c5b0c2700358a7 (diff) | |
download | klee-f62f8b954a00ae8538f8d3ef2aad77c11e6ed710.tar.gz |
[klee-stats] Refactor preparing and printing the table
The number and order of statistics in klee-stats is hard coded. Moreover, adding new statistics to KLEE lead to crashes. Rewriting that part of the script generalises and streamlines the process. List of changes: * Extend legend: this is used for known columns to provide shorter names * simplify sqlite handling and make it more robust and reading of data * select columns based on internal KLEE names * streamline addition of artificial columns and make it robust * handle the case if different runs should be compared but not all have the same statistics * fix calculation of summary row: - avg of column if column is showing a relative value or avg value - max of column if column is showing a max value - sum of column entries, else
-rwxr-xr-x | tools/klee-stats/klee-stats | 304 |
1 files changed, 205 insertions, 99 deletions
diff --git a/tools/klee-stats/klee-stats b/tools/klee-stats/klee-stats index 544a06d3..78b7b0cb 100755 --- a/tools/klee-stats/klee-stats +++ b/tools/klee-stats/klee-stats @@ -2,12 +2,12 @@ # -*- encoding: utf-8 -*- # ===-- klee-stats --------------------------------------------------------===## -# +# # The KLEE Symbolic Virtual Machine -# +# # This file is distributed under the University of Illinois Open Source # License. See LICENSE.TXT for details. -# +# # ===----------------------------------------------------------------------===## """Output statistics logged by Klee.""" @@ -16,24 +16,35 @@ import os import sys import argparse import sqlite3 +import collections +# Mapping of: (column head, explanation, internal klee name) +# column head must start with a capital letter Legend = [ - ('Instrs', 'number of executed instructions'), - ('Time', 'total wall time (s)'), - ('TUser', 'total user time'), - ('ICov', 'instruction coverage in the LLVM bitcode (%)'), - ('BCov', 'branch coverage in the LLVM bitcode (%)'), - ('ICount', 'total static instructions in the LLVM bitcode'), - ('TSolver', 'time spent in the constraint solver'), - ('States', 'number of currently active states'), - ('Mem', 'megabytes of memory currently used'), - ('Queries', 'number of queries issued to STP'), - ('AvgQC', 'average number of query constructs per query'), - ('Tcex', 'time spent in the counterexample caching code'), - ('Tfork', 'time spent forking'), - ('TResolve', 'time spent in object resolution'), - ('QCexCMisses', 'Counterexample cache misses'), - ('QCexCHits', 'Counterexample cache hits'), + ('Instrs', 'number of executed instructions', "Instructions"), + ('Time(s)', 'total wall time (s)', "WallTime"), + ('TUser(s)', 'total user time', "UserTime"), + ('ICov(%)', 'instruction coverage in the LLVM bitcode (%)', "ICov"), + ('BCov(%)', 'branch coverage in the LLVM bitcode (%)', "BCov"), + ('ICount', 'total static instructions in the LLVM bitcode', "ICount"), + ('TSolver(s)', 'time spent in the constraint solver', "SolverTime"), + ('TSolver(%)', 'time spent in the constraint solver', "RelSolverTime"), + ('States', 'number of currently active states', "NumStates"), + ('MaxStates', 'maximum number of active states', "maxStates"), + ('AvgStates', 'average number of active states', "avgStates"), + ('Mem(MB)', 'megabytes of memory currently used', "MallocUsage"), + ('MaxMem(MB)', 'megabytes of memory currently used', "MaxMem"), + ('AvgMem(MB)', 'megabytes of memory currently used', "AvgMem"), + ('Queries', 'number of queries issued to STP', "NumQueries"), + ('AvgQC', 'average number of query constructs per query', "AvgQC"), + ('Tcex(s)', 'time spent in the counterexample caching code', "CexCacheTime"), + ('Tcex(%)', 'relative time spent in the counterexample caching code wrt wall time', "CexCacheTime"), + ('Tfork(s)', 'time spent forking', "ForkTime"), + ('Tfork(%)', 'relative time spent forking wrt wall time', "ForkTime"), + ('TResolve(s)', 'time spent in object resolution', "ResolveTime"), + ('TResolve(%)', 'time spent in object resolution wrt wall time', "ResolveTime"), + ('QCexCMisses', 'Counterexample cache misses', "QueryCexCacheMisses"), + ('QCexCHits', 'Counterexample cache hits', "QueryCexCacheHits"), ] def getInfoFile(path): @@ -48,23 +59,34 @@ class LazyEvalList: """Store all the lines in run.stats and eval() when needed.""" def __init__(self, fileName): # The first line in the records contains headers. - self.conn = sqlite3.connect(fileName) - self.c = self.conn.cursor() - self.c.execute("SELECT * FROM stats ORDER BY Instructions DESC LIMIT 1") - self.line = self.c.fetchone() + self.filename = fileName + + def conn(self): + return sqlite3.connect(self.filename) def aggregateRecords(self): - memC = self.conn.cursor() - memC.execute("SELECT max(MallocUsage) / 1024 / 1024, avg(MallocUsage) / 1024 / 1024 from stats") - maxMem, avgMem = memC.fetchone() + try: + memC = self.conn().execute("SELECT max(MallocUsage) / 1024 / 1024, avg(MallocUsage) / 1024 / 1024 from stats") + maxMem, avgMem = memC.fetchone() + except sqlite3.OperationalError as e: + maxMem, avgMem = None, None + + try: + stateC = self.conn().execute("SELECT max(NumStates), avg(NumStates) from stats") + maxStates, avgStates = stateC.fetchone() + except sqlite3.OperationalError as e: + maxStates, avgStates = None, None - stateC = self.conn.cursor() - stateC.execute("SELECT max(NumStates), avg(NumStates) from stats") - maxStates, avgStates = stateC.fetchone() - return (maxMem, avgMem, maxStates, avgStates) + return {"maxMem":maxMem, "avgMem": avgMem, "maxState": maxStates, "avgStates": avgStates} def getLastRecord(self): - return self.line + try: + cursor = self.conn().execute("SELECT * FROM stats ORDER BY Instructions DESC LIMIT 1") + column_names = [description[0] for description in cursor.description] + return dict(zip(column_names, cursor.fetchone())) + except sqlite3.OperationalError as e: + return None + def stripCommonPathPrefix(paths): paths = map(os.path.normpath, paths) @@ -76,6 +98,7 @@ def stripCommonPathPrefix(paths): break return ['/'.join(p[i:]) for p in paths] + def getKleeOutDirs(dirs): kleeOutDirs = [] for dir in dirs: @@ -89,63 +112,70 @@ def getKleeOutDirs(dirs): kleeOutDirs.append(path) return kleeOutDirs -def getLabels(pr): + +def select_columns(record, pr): if pr == 'all': - labels = ('Path', 'Instrs', 'Time(s)', 'ICov(%)', 'BCov(%)', 'ICount', - 'TSolver(%)', 'States', 'maxStates', 'avgStates', 'Mem(MB)', - 'maxMem(MB)', 'avgMem(MB)', 'Queries', 'AvgQC', 'Tcex(%)', - 'Tfork(%)', 'TResolve(%)', 'QCexCMisses', 'QCexCHits') - elif pr == 'reltime': - labels = ('Path', 'Time(s)', 'TUser(%)', 'TSolver(%)', - 'Tcex(%)', 'Tfork(%)', 'TResolve(%)') + return record + + if pr == 'reltime': + s_column = ['Path', 'WallTime', 'RelUserTime', 'RelSolverTime', + 'RelCexCacheTime', 'RelForkTime', 'RelResolveTime'] elif pr == 'abstime': - labels = ('Path', 'Time(s)', 'TUser(s)', 'TSolver(s)', - 'Tcex(s)', 'Tfork(s)', 'TResolve(s)') + s_column = ['Path', 'WallTime', 'UserTime', 'SolverTime', + 'CexCacheTime', 'ForkTime', 'ResolveTime'] elif pr == 'more': - labels = ('Path', 'Instrs', 'Time(s)', 'ICov(%)', 'BCov(%)', 'ICount', - 'TSolver(%)', 'States', 'maxStates', 'Mem(MB)', 'maxMem(MB)') + s_column = ['Path', 'Instructions', 'WallTime', 'ICov', 'BCov', 'ICount', + 'RelSolverTime', 'States', 'maxStates', 'MallocUsage', 'maxMem'] else: - labels = ('Path', 'Instrs', 'Time(s)', 'ICov(%)', - 'BCov(%)', 'ICount', 'TSolver(%)') - return labels + s_column = ['Path', 'Instructions', 'WallTime', 'ICov', + 'BCov', 'ICount', 'RelSolverTime'] + filtered_record = dict() + for column in s_column: + if column in record.keys(): + filtered_record[column] = record[column] -def getRow(record, stats, pr): - """Compose data for the current run into a row.""" - I, BFull, BPart, BTot, T, St, Mem, QTot, QCon,\ - _, Treal, SCov, SUnc, _, Ts, Tcex, Tf, Tr, QCexMiss, QCexHits = record - maxMem, avgMem, maxStates, avgStates = stats + return filtered_record + +def add_artificial_columns(record): # special case for straight-line code: report 100% branch coverage - if BTot == 0: - BFull = BTot = 1 + if "NumBranches" in record and record["NumBranches"] == 0: + record["FullBranches"] = 1 + record["NumBranches"] = 1 - Ts, Tcex, Tf, Tr, T, Treal = [e / 1000000 for e in [Ts, Tcex, Tf, Tr, T, Treal]] #convert from microseconds - Mem = Mem / 1024 / 1024 - AvgQC = int(QCon / max(1, QTot)) + # Convert recorded times from microseconds to seconds + for key in ["UserTime", "WallTime", "QueryTime", "SolverTime", "CexCacheTime", "ForkTime", "ResolveTime"]: + if not key in record: + continue + record[key] /= 1000000 - if pr == 'all': - row = (I, Treal, 100 * SCov / (SCov + SUnc), - 100 * (2 * BFull + BPart) / (2 * BTot), SCov + SUnc, - 100 * Ts / Treal, St, maxStates, avgStates, - Mem, maxMem, avgMem, QTot, AvgQC, 100 * Tcex / Treal, - 100 * Tf / Treal, 100 * Tr / Treal, QCexMiss, QCexHits) - elif pr == 'reltime': - row = (Treal, 100 * T / Treal, 100 * Ts / Treal, - 100 * Tcex / Treal, 100 * Tf / Treal, - 100 * Tr / Treal) - elif pr == 'abstime': - row = (Treal, T, Ts, Tcex, Tf, Tr) - elif pr == 'more': - row = (I, Treal, 100 * SCov / (SCov + SUnc), - 100 * (2 * BFull + BPart) / (2 * BTot), - SCov + SUnc, 100 * Ts / Treal, - St, maxStates, Mem, maxMem) - else: - row = (I, Treal, 100 * SCov / (SCov + SUnc), - 100 * (2 * BFull + BPart) / (2 * BTot), - SCov + SUnc, 100 * Ts / Treal) - return row + # Convert memory from byte to MiB + if "MallocUsage" in record: + record["MallocUsage"] /= (1024*1024) + + # Calculate avg. query construct + if "NumQueryConstructs" in record and "NumQueries" in record: + record["AvgQC"] = int(record["NumQueryConstructs"] / max(1, record["NumQueries"])) + + # Calculate total number of instructions + if "CoveredInstructions" in record and "UncoveredInstructions" in record: + record["ICount"] = (record["CoveredInstructions"] + record["UncoveredInstructions"]) + + # Calculate relative instruction coverage + if "CoveredInstructions" in record and "ICount" in record: + record["ICov"] = 100 * record["CoveredInstructions"] / record["ICount"] + + # Calculate branch coverage + if "FullBranches" in record and "PartialBranches" in record and "NumBranches" in record: + record["BCov"] = 100 * ( 2 * record["FullBranches"] + record["PartialBranches"]) / ( 2 * record["NumBranches"]) + + # Add relative times + for key in ["SolverTime", "CexCacheTime", "ForkTime", "ResolveTime", "UserTime"]: + if "WallTime" in record and key in record: + record["Rel"+key] = 100 * record[key] / record["WallTime"] + + return record def grafana(dirs, host_address, port): @@ -228,7 +258,7 @@ def grafana(dirs, host_address, port): def write_csv(data): import csv data = data[0] - c = data.conn.cursor() + c = data.conn().cursor() sql3_cursor = c.execute("SELECT * FROM stats") csv_out = csv.writer(sys.stdout) # write header @@ -238,6 +268,23 @@ def write_csv(data): csv_out.writerow(result) +def rename_columns(row, name_mapping): + """ + Renames the columns in a row based on the mapping. + If a column name is not found in the mapping, keep the old name + :param row: + :param name_mapping: + :return: updated row + """ + keys = list(row.keys()) + for k in keys: + new_key = name_mapping.get(k, k) + if new_key == k: + continue + row[new_key] = row.pop(k) + return row + + def write_table(args, data, dirs, pr): from tabulate import TableFormat, Line, DataRow, tabulate @@ -254,35 +301,94 @@ def write_table(args, data, dirs, pr): dirs = stripCommonPathPrefix(dirs) # attach the stripped path data = list(zip(dirs, data)) - labels = getLabels(pr) + # build the main body of the table - table = [] - totRecords = [] # accumulated records - totStats = [] # accumulated stats - for path, records in data: - row = [path] + table = dict() + for i, (path, records) in enumerate(data): stats = records.aggregateRecords() - totStats.append(stats) - row.extend(getRow(records.getLastRecord(), stats, pr)) - totRecords.append(records.getLastRecord()) - table.append(row) - # calculate the total - totRecords = [sum(e) for e in zip(*totRecords)] - totStats = [sum(e) for e in zip(*totStats)] - totalRow = ['Total ({0})'.format(len(table))] - totalRow.extend(getRow(totRecords, totStats, pr)) - if len(data) > 1: - table.append(totalRow) - table.insert(0, labels) + # Get raw row + single_row = records.getLastRecord() + if single_row is None: + # print("Error reading SQLite file (probably corrupt)") + # continue + single_row = {} + single_row['Path'] = path + single_row.update(stats) + + # Extend row with additional entries + single_row = add_artificial_columns(single_row) + single_row = select_columns(single_row, pr) + + for key in single_row: + # Get an existing column or add an empty one + column = table.setdefault(key, []) + + # Resize the column if needed + missing_entries = i - len(column) - 1 + if missing_entries > 0: + # Append empty entries for this column + column.extend([None]* missing_entries) + + # add the value + column.append(single_row[key]) + + # Add missing entries if needed + max_len = len(data) + for c_name in table: + c_length = len(table[c_name]) + if c_length < max_len: + table[c_name].extend([None] * (max_len - c_length)) + + # Rename columns + name_mapping = dict() + for entry in Legend: + name_mapping[entry[2]] = entry[0] + table = rename_columns(table, name_mapping) + + # Add a summary row + if max_len > 1: + # calculate the total + for k in table: + if k == "Path": # Skip path + continue + # TODO: this is a bit bad but ... . In a nutshell, if the name of a column starts or ends with certain + # pattern change the summary function. + if k.startswith("Avg") or k.endswith("(%)"): + total = sum([e for e in table[k] if e is not None])/max_len + elif k.startswith("Max"): + total = max([e for e in table[k] if e is not None]) + else: + total = sum([e for e in table[k] if e is not None]) + table[k].append(total) + + table['Path'].append('Total ({0})'.format(max_len)) + + # Prepare the order of the header: start to order entries according to the order in legend and add the unknown entries at the end + headers = ["Path"] + available_headers = list(table.keys()) + for entry in Legend: + l_name = entry[0] + if l_name in available_headers: + headers.append(l_name) + available_headers.remove(l_name) + headers += available_headers + + # Make sure we keep the correct order of the column entries + final_table = collections.OrderedDict() + for column in headers: + final_table[column] = table[column] + table = final_table + + # Output table if args.tableFormat != 'klee': print(tabulate( - table, headers='firstrow', + table, headers='keys', tablefmt=args.tableFormat, floatfmt='.{p}f'.format(p=2), numalign='right', stralign='center')) else: stream = tabulate( - table, headers='firstrow', + table, headers='keys', tablefmt=KleeTable, floatfmt='.{p}f'.format(p=2), numalign='right', stralign='center') |