tests: Add test infrastructure as a Python module

Implement gem5's test infrastructure as a Python module and a run
script that can be used without scons. The new implementation has
several features that were lacking from the previous test
infrastructure such as support for multiple output formats, automatic
runtime tracking, and better support for being run in a cluster
environment.

Tests consist of one or more steps (TestUnit). Units are run in two
stages, the first a run stage and then a verify stage. Units in the
verify stage are automatically skipped if any unit run stage wasn't
run. The library currently contains TestUnit implementations that run
gem5, diff stat files, and diff output files.

Existing tests are implemented by the ClassicTest class and "just
work". New tests can that don't rely on the old "run gem5 once and
diff output" strategy can be implemented by subclassing the Test base
class or ClassicTest.

Test results can be output in multiple formats. The module currently
supports JUnit, text (short and verbose), and Python's pickle
format. JUnit output allows CI systems to automatically get more
information about test failures. The pickled output contains all state
necessary to reconstruct a tests results object and is mainly intended
for the build system and CI systems.

Since many JUnit parsers parsers assume that test suite names look
like Java package names. We currently output path-like names with
slashes separating components. Test names are translated according to
these rules:

  * '.' -> '-"
  * '/' -> '.'

The test tool, tests.py, supports the following features:

  * Test listing. Example: ./tests.py list arm/quick

  * Running tests. Example:
    ./tests.py run -o output.pickle --format pickle \
        ../build/ARM/gem5.opt \
        quick/se/00.hello/arm/linux/simple-timing

  * Displaying pickled results. Example:
    ./tests.py show --format summary *.pickle

Change-Id: I527164bd791237aacfc65e7d7c0b67b695c5d17c
Signed-off-by: Andreas Sandberg <andreas.sandberg@arm.com>
Reviewed-by: Curtis Dunham <curtis.dunham@arm.com>
Reviewed-by: Joel Hestness <jthestness@gmail.com>

tests/testing/__init__.py

@@ -0,0 +1,38 @@
+#!/usr/bin/env python
+#
+# Copyright (c) 2016 ARM Limited
+# All rights reserved
+#
+# The license below extends only to copyright in the software and shall
+# not be construed as granting a license to any other intellectual
+# property including but not limited to intellectual property relating
+# to a hardware implementation of the functionality of the software
+# licensed hereunder.  You may use the software subject to the license
+# terms below provided that you ensure that this notice is replicated
+# unmodified and in its entirety in all distributions of the software,
+# modified or unmodified, in source code or in binary form.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met: redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer;
+# redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in the
+# documentation and/or other materials provided with the distribution;
+# neither the name of the copyright holders nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Authors: Andreas Sandberg

tests/testing/helpers.py

@@ -0,0 +1,132 @@
+#!/usr/bin/env python
+#
+# Copyright (c) 2016 ARM Limited
+# All rights reserved
+#
+# The license below extends only to copyright in the software and shall
+# not be construed as granting a license to any other intellectual
+# property including but not limited to intellectual property relating
+# to a hardware implementation of the functionality of the software
+# licensed hereunder.  You may use the software subject to the license
+# terms below provided that you ensure that this notice is replicated
+# unmodified and in its entirety in all distributions of the software,
+# modified or unmodified, in source code or in binary form.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met: redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer;
+# redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in the
+# documentation and/or other materials provided with the distribution;
+# neither the name of the copyright holders nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Authors: Andreas Sandberg
+
+import subprocess
+from threading import Timer
+import time
+
+class CallTimeoutException(Exception):
+    """Exception that indicates that a process call timed out"""
+
+    def __init__(self, status, stdout, stderr):
+        self.status = status
+        self.stdout = stdout
+        self.stderr = stderr
+
+class ProcessHelper(subprocess.Popen):
+    """Helper class to run child processes.
+
+    This class wraps a subprocess.Popen class and adds support for
+    using it in a with block. When the process goes out of scope, it's
+    automatically terminated.
+
+    with ProcessHelper(["/bin/ls"], stdout=subprocess.PIPE) as p:
+        return p.call()
+    """
+    def __init__(self, *args, **kwargs):
+        super(ProcessHelper, self).__init__(*args, **kwargs)
+
+    def _terminate_nicely(self, timeout=5):
+        def on_timeout():
+            self.kill()
+
+        if self.returncode is not None:
+            return self.returncode
+
+        timer = Timer(timeout, on_timeout)
+        self.terminate()
+        status = self.wait()
+        timer.cancel()
+
+        return status
+
+    def __enter__(self):
+        return self
+
+    def __exit__(self, exc_type, exc_value, traceback):
+        if self.returncode is None:
+            self._terminate_nicely()
+
+    def call(self, timeout=0):
+        self._timeout = False
+        def on_timeout():
+            self._timeout = True
+            self._terminate_nicely()
+
+        status, stdout, stderr = None, None, None
+        timer = Timer(timeout, on_timeout)
+        if timeout:
+            timer.start()
+
+        stdout, stderr = self.communicate()
+        status = self.wait()
+
+        timer.cancel()
+
+        if self._timeout:
+            self._terminate_nicely()
+            raise CallTimeoutException(self.returncode, stdout, stderr)
+        else:
+            return status, stdout, stderr
+
+if __name__ == "__main__":
+    # Run internal self tests to ensure that the helpers are working
+    # properly. The expected output when running this script is
+    # "SUCCESS!".
+
+    cmd_foo = [ "/bin/echo", "-n", "foo" ]
+    cmd_sleep = [ "/bin/sleep", "10" ]
+
+    # Test that things don't break if the process hasn't been started
+    with ProcessHelper(cmd_foo) as p:
+        pass
+
+    with ProcessHelper(cmd_foo, stdout=subprocess.PIPE) as p:
+        status, stdout, stderr = p.call()
+    assert stdout == "foo"
+    assert status == 0
+
+    try:
+        with ProcessHelper(cmd_sleep) as p:
+            status, stdout, stderr = p.call(timeout=1)
+        assert False, "Timeout not triggered"
+    except CallTimeoutException:
+        pass
+
+    print "SUCCESS!"

tests/testing/results.py

@@ -0,0 +1,271 @@
+#!/usr/bin/env python
+#
+# Copyright (c) 2016 ARM Limited
+# All rights reserved
+#
+# The license below extends only to copyright in the software and shall
+# not be construed as granting a license to any other intellectual
+# property including but not limited to intellectual property relating
+# to a hardware implementation of the functionality of the software
+# licensed hereunder.  You may use the software subject to the license
+# terms below provided that you ensure that this notice is replicated
+# unmodified and in its entirety in all distributions of the software,
+# modified or unmodified, in source code or in binary form.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met: redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer;
+# redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in the
+# documentation and/or other materials provided with the distribution;
+# neither the name of the copyright holders nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Authors: Andreas Sandberg
+
+from abc import ABCMeta, abstractmethod
+import inspect
+import pickle
+import string
+import sys
+
+import xml.etree.cElementTree as ET
+
+class UnitResult(object):
+    """Results of a single test unit.
+
+    A test result can be one of:
+        - STATE_OK: Test ran successfully.
+        - STATE_SKIPPED: The test was skipped.
+        - STATE_ERROR: The test failed to run.
+        - STATE_FAILED: Test ran, but failed.
+
+    The difference between STATE_ERROR and STATE_FAILED is very
+    subtle. In a gem5 context, STATE_ERROR would mean that gem5 failed
+    to start or crashed, while STATE_FAILED would mean that a test
+    failed (e.g., statistics mismatch).
+
+    """
+
+    STATE_OK = 0
+    STATE_SKIPPED = 1
+    STATE_ERROR = 2
+    STATE_FAILURE = 3
+
+    state_names = {
+        STATE_OK : "OK",
+        STATE_SKIPPED : "SKIPPED",
+        STATE_ERROR : "ERROR",
+        STATE_FAILURE : "FAILURE",
+    }
+
+    def __init__(self, name, state, message="", stderr="", stdout="",
+                 runtime=0.0):
+        self.name = name
+        self.state = state
+        self.message = message
+        self.stdout = stdout
+        self.stderr = stderr
+        self.runtime = runtime
+
+    def skipped(self):
+        return self.state == UnitResult.STATE_SKIPPED
+
+    def success(self):
+        return self.state == UnitResult.STATE_OK
+
+    def state_name(self):
+        return UnitResult.state_names[self.state]
+
+    def __nonzero__(self):
+        return self.success() or self.skipped()
+
+    def __str__(self):
+        state_name = self.state_name()
+
+        status = "%s: %s" % (state_name, self.message) if self.message else \
+                 state_name
+
+        return "%s: %s" % (self.name, status)
+
+class TestResult(object):
+    """Results for from a single test consisting of one or more units."""
+
+    def __init__(self, name, results=[]):
+        self.name = name
+        self.results = results
+
+    def success(self):
+        return all([ r.success() for r in self.results])
+
+    def skipped(self):
+        return all([ r.skipped() for r in self.results])
+
+    def failed(self):
+        return any([ not r for r in self.results])
+
+    def runtime(self):
+        return sum([ r.runtime for r in self.results ])
+
+    def __nonzero__(self):
+        return all([r for r in self.results])
+
+class ResultFormatter(object):
+    __metaclass__ = ABCMeta
+
+    def __init__(self, fout=sys.stdout, verbose=False):
+        self.verbose = verbose
+        self.fout = fout
+
+    @abstractmethod
+    def dump_suites(self, suites):
+        pass
+
+class Pickle(ResultFormatter):
+    """Save test results as a binary using Python's pickle
+    functionality.
+
+    """
+
+    def __init__(self, **kwargs):
+        super(Pickle, self).__init__(**kwargs)
+
+    def dump_suites(self, suites):
+        pickle.dump(suites, self.fout, pickle.HIGHEST_PROTOCOL)
+
+class Text(ResultFormatter):
+    """Output test results as text."""
+
+    def __init__(self, **kwargs):
+        super(Text, self).__init__(**kwargs)
+
+    def dump_suites(self, suites):
+        fout = self.fout
+        for suite in suites:
+            print >> fout, "--- %s ---" % suite.name
+
+            for t in suite.results:
+                print >> fout, "*** %s" % t
+
+                if t and not self.verbose:
+                    continue
+
+                if t.message:
+                    print >> fout, t.message
+
+                if t.stderr:
+                    print >> fout, t.stderr
+                if t.stdout:
+                    print >> fout, t.stdout
+
+class TextSummary(ResultFormatter):
+    """Output test results as a text summary"""
+
+    def __init__(self, **kwargs):
+        super(TextSummary, self).__init__(**kwargs)
+
+    def dump_suites(self, suites):
+        fout = self.fout
+        for suite in suites:
+            status = "SKIPPED" if suite.skipped() else \
+                     ("OK" if suite else "FAILED")
+            print >> fout, "%s: %s" % (suite.name, status)
+
+class JUnit(ResultFormatter):
+    """Output test results as JUnit XML"""
+
+    def __init__(self, translate_names=True, **kwargs):
+        super(JUnit, self).__init__(**kwargs)
+
+        if translate_names:
+            self.name_table = string.maketrans(
+                "/.",
+                ".-",
+            )
+        else:
+            self.name_table = string.maketrans("", "")
+
+    def convert_unit(self, x_suite, test):
+        x_test = ET.SubElement(x_suite, "testcase",
+                               name=test.name,
+                               time="%f" % test.runtime)
+
+        x_state = None
+        if test.state == UnitResult.STATE_OK:
+            pass
+        elif test.state == UnitResult.STATE_SKIPPED:
+            x_state = ET.SubElement(x_test, "skipped")
+        elif test.state == UnitResult.STATE_FAILURE:
+            x_state = ET.SubElement(x_test, "failure")
+        elif test.state == UnitResult.STATE_ERROR:
+            x_state = ET.SubElement(x_test, "error")
+        else:
+            assert False, "Unknown test state"
+
+        if x_state is not None:
+            if test.message:
+                x_state.set("message", test.message)
+
+            msg = []
+            if test.stderr:
+                msg.append("*** Standard Errror: ***")
+                msg.append(test.stderr)
+            if test.stdout:
+                msg.append("*** Standard Out: ***")
+                msg.append(test.stdout)
+
+            x_state.text = "\n".join(msg)
+
+        return x_test
+
+    def convert_suite(self, x_suites, suite):
+        x_suite = ET.SubElement(x_suites, "testsuite",
+                                name=suite.name.translate(self.name_table),
+                                time="%f" % suite.runtime())
+        errors = 0
+        failures = 0
+        skipped = 0
+
+        for test in suite.results:
+            if test.state != UnitResult.STATE_OK:
+                if test.state == UnitResult.STATE_SKIPPED:
+                    skipped += 1
+                elif test.state == UnitResult.STATE_ERROR:
+                    errors += 1
+                elif test.state == UnitResult.STATE_FAILURE:
+                    failures += 1
+
+            x_test = self.convert_unit(x_suite, test)
+
+        x_suite.set("errors", str(errors))
+        x_suite.set("failures", str(failures))
+        x_suite.set("skipped", str(skipped))
+        x_suite.set("tests", str(len(suite.results)))
+
+        return x_suite
+
+    def convert_suites(self, suites):
+        x_root = ET.Element("testsuites")
+
+        for suite in suites:
+            self.convert_suite(x_root, suite)
+
+        return x_root
+
+    def dump_suites(self, suites):
+        et = ET.ElementTree(self.convert_suites(suites))
+        et.write(self.fout, encoding="UTF-8")

tests/testing/tests.py

@@ -0,0 +1,343 @@
+#!/usr/bin/env python
+#
+# Copyright (c) 2016 ARM Limited
+# All rights reserved
+#
+# The license below extends only to copyright in the software and shall
+# not be construed as granting a license to any other intellectual
+# property including but not limited to intellectual property relating
+# to a hardware implementation of the functionality of the software
+# licensed hereunder.  You may use the software subject to the license
+# terms below provided that you ensure that this notice is replicated
+# unmodified and in its entirety in all distributions of the software,
+# modified or unmodified, in source code or in binary form.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met: redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer;
+# redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in the
+# documentation and/or other materials provided with the distribution;
+# neither the name of the copyright holders nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Authors: Andreas Sandberg
+
+from abc import ABCMeta, abstractmethod
+import os
+from collections import namedtuple
+from units import *
+from results import TestResult
+import shutil
+
+_test_base = os.path.join(os.path.dirname(__file__), "..")
+
+ClassicConfig = namedtuple("ClassicConfig", (
+    "category",
+    "mode",
+    "workload",
+    "isa",
+    "os",
+    "config",
+))
+
+# There are currently two "classes" of test
+# configurations. Architecture-specific ones and generic ones
+# (typically SE mode tests). In both cases, the configuration name
+# matches a file in tests/configs/ that will be picked up by the test
+# runner (run.py).
+#
+# Architecture specific configurations are listed in the arch_configs
+# dictionary. This is indexed by a (cpu architecture, gpu
+# architecture) tuple. GPU architecture is optional and may be None.
+#
+# Generic configurations are listed in the generic_configs tuple.
+#
+# When discovering available test cases, this script look uses the
+# test list as a list of /candidate/ configurations. A configuration
+# is only used if a test has a reference output for that
+# configuration. In addition to the base configurations from
+# arch_configs and generic_configs, a Ruby configuration may be
+# appended to the base name (this is probed /in addition/ to the
+# original name. See get_tests() for details.
+#
+arch_configs = {
+    ("alpha", None) : (
+        'tsunami-simple-atomic',
+        'tsunami-simple-timing',
+        'tsunami-simple-atomic-dual',
+        'tsunami-simple-timing-dual',
+        'twosys-tsunami-simple-atomic',
+        'tsunami-o3', 'tsunami-o3-dual',
+        'tsunami-minor', 'tsunami-minor-dual',
+        'tsunami-switcheroo-full',
+    ),
+
+    ("arm", None) : (
+        'simple-atomic-dummychecker',
+        'o3-timing-checker',
+        'realview-simple-atomic',
+        'realview-simple-atomic-dual',
+        'realview-simple-atomic-checkpoint',
+        'realview-simple-timing',
+        'realview-simple-timing-dual',
+        'realview-o3',
+        'realview-o3-checker',
+        'realview-o3-dual',
+        'realview-minor',
+        'realview-minor-dual',
+        'realview-switcheroo-atomic',
+        'realview-switcheroo-timing',
+        'realview-switcheroo-o3',
+        'realview-switcheroo-full',
+        'realview64-simple-atomic',
+        'realview64-simple-atomic-checkpoint',
+        'realview64-simple-atomic-dual',
+        'realview64-simple-timing',
+        'realview64-simple-timing-dual',
+        'realview64-o3',
+        'realview64-o3-checker',
+        'realview64-o3-dual',
+        'realview64-minor',
+        'realview64-minor-dual',
+        'realview64-switcheroo-atomic',
+        'realview64-switcheroo-timing',
+        'realview64-switcheroo-o3',
+        'realview64-switcheroo-full',
+    ),
+
+    ("sparc", None) : (
+        't1000-simple-atomic',
+        't1000-simple-x86',
+    ),
+
+    ("timing", None) : (
+        'pc-simple-atomic',
+        'pc-simple-timing',
+        'pc-o3-timing',
+        'pc-switcheroo-full',
+    ),
+
+    ("x86", "hsail") : (
+        'gpu',
+    ),
+}
+
+generic_configs = (
+    'simple-atomic',
+    'simple-atomic-mp',
+    'simple-timing',
+    'simple-timing-mp',
+
+    'minor-timing',
+    'minor-timing-mp',
+
+    'o3-timing',
+    'o3-timing-mt',
+    'o3-timing-mp',
+
+    'rubytest',
+    'memcheck',
+    'memtest',
+    'memtest-filter',
+    'tgen-simple-mem',
+    'tgen-dram-ctrl',
+
+    'learning-gem5-p1-simple',
+    'learning-gem5-p1-two-level',
+)
+
+all_categories = ("quick", "long")
+all_modes = ("fs", "se")
+
+class Test(object):
+    """Test case base class.
+
+    Test cases consists of one or more test units that are run in two
+    phases. A run phase (units produced by run_units() and a verify
+    phase (units from verify_units()). The verify phase is skipped if
+    the run phase fails.
+
+    """
+
+    __metaclass__ = ABCMeta
+
+    def __init__(self, name):
+        self.test_name = name
+
+    @abstractmethod
+    def ref_files(self):
+        """Get a list of reference files used by this test case"""
+        pass
+
+    @abstractmethod
+    def run_units(self):
+        """Units (typically RunGem5 instances) that describe the run phase of
+        this test.
+
+        """
+        pass
+
+    @abstractmethod
+    def verify_units(self):
+        """Verify the output from the run phase (see run_units())."""
+        pass
+
+    @abstractmethod
+    def update_ref(self):
+        """Update reference files with files from a test run"""
+        pass
+
+    def run(self):
+        """Run this test case and return a list of results"""
+
+        run_results = [ u.run() for u in self.run_units() ]
+        run_ok = all([not r.skipped() and r for r in run_results ])
+
+        verify_results = [
+            u.run() if run_ok else u.skip()
+            for u in self.verify_units()
+        ]
+
+        return TestResult(self.test_name, run_results + verify_results)
+
+    def __str__(self):
+        return self.test_name
+
+class ClassicTest(Test):
+    diff_ignore_files = [
+        # Stat files use a special stat differ, so don't include them
+        # here.
+        "stats.txt",
+    ]
+
+    def __init__(self, gem5, output_dir, config_tuple,
+                 timeout=None,
+                 skip=False, skip_diff_out=False, skip_diff_stat=False):
+
+        super(ClassicTest, self).__init__("/".join(config_tuple))
+
+        ct = config_tuple
+
+        self.gem5 = os.path.abspath(gem5)
+        self.script = os.path.join(_test_base, "run.py")
+        self.config_tuple = ct
+        self.timeout = timeout
+
+        self.output_dir = output_dir
+        self.ref_dir = os.path.join(_test_base,
+                                    ct.category, ct.mode, ct.workload,
+                                    "ref", ct.isa, ct.os, ct.config)
+        self.skip_run = skip
+        self.skip_diff_out = skip or skip_diff_out
+        self.skip_diff_stat = skip or skip_diff_stat
+
+    def ref_files(self):
+        ref_dir = os.path.abspath(self.ref_dir)
+        for root, dirs, files in os.walk(ref_dir, topdown=False):
+            for f in files:
+                fpath = os.path.join(root[len(ref_dir) + 1:], f)
+                if fpath not in ClassicTest.diff_ignore_files:
+                    yield fpath
+
+    def run_units(self):
+        args = [
+            self.script,
+            "/".join(self.config_tuple),
+        ]
+
+        return [
+            RunGem5(self.gem5, args,
+                    ref_dir=self.ref_dir, test_dir=self.output_dir,
+                    skip=self.skip_run),
+        ]
+
+    def verify_units(self):
+        return [
+            DiffStatFile(ref_dir=self.ref_dir, test_dir=self.output_dir,
+                         skip=self.skip_diff_stat)
+        ] + [
+            DiffOutFile(f,
+                        ref_dir=self.ref_dir, test_dir=self.output_dir,
+                        skip=self.skip_diff_out)
+            for f in self.ref_files()
+            if f not in ClassicTest.diff_ignore_files
+        ]
+
+    def update_ref(self):
+        for fname in self.ref_files():
+            shutil.copy(
+                os.path.join(self.output_dir, fname),
+                os.path.join(self.ref_dir, fname))
+
+def parse_test_filter(test_filter):
+    wildcards = ("", "*")
+
+    _filter = list(test_filter.split("/"))
+    if len(_filter) > 3:
+        raise RuntimeError("Illegal test filter string")
+    _filter += [ "", ] * (3 - len(_filter))
+
+    isa, cat, mode = _filter
+
+    if isa in wildcards:
+        raise RuntimeError("No ISA specified")
+
+    cat = all_categories if cat in wildcards else (cat, )
+    mode = all_modes if mode in wildcards else (mode, )
+
+    return isa, cat, mode
+
+def get_tests(isa,
+              categories=all_categories, modes=all_modes,
+              ruby_protocol=None, gpu_isa=None):
+
+    # Generate a list of candidate configs
+    configs = list(arch_configs.get((isa, gpu_isa), []))
+
+    if (isa, gpu_isa) == ("x86", "hsail"):
+        if ruby_protocol == "GPU_RfO":
+            configs += ['gpu-randomtest']
+    else:
+        configs += generic_configs
+
+    if ruby_protocol == 'MI_example':
+        configs += [ "%s-ruby" % (c, ) for c in configs ]
+    elif ruby_protocol is not None:
+        configs += [ "%s-ruby-%s" % (c, ruby_protocol) for c in configs ]
+
+    # /(quick|long)/(fs|se)/workload/ref/arch/guest/config/
+    for conf_script in configs:
+        for cat in categories:
+            for mode in modes:
+                mode_dir = os.path.join(_test_base, cat, mode)
+                if not os.path.exists(mode_dir):
+                    continue
+
+                for workload in os.listdir(mode_dir):
+                    isa_dir = os.path.join(mode_dir, workload, "ref", isa)
+                    if not os.path.isdir(isa_dir):
+                        continue
+
+                    for _os in os.listdir(isa_dir):
+                        test_dir = os.path.join(isa_dir, _os, conf_script)
+                        if not os.path.exists(test_dir) or \
+                           os.path.exists(os.path.join(test_dir, "skip")):
+                            continue
+
+                        yield ClassicConfig(cat, mode, workload, isa, _os,
+                                            conf_script)

tests/testing/units.py

@@ -0,0 +1,290 @@
+#!/usr/bin/env python
+#
+# Copyright (c) 2016 ARM Limited
+# All rights reserved
+#
+# The license below extends only to copyright in the software and shall
+# not be construed as granting a license to any other intellectual
+# property including but not limited to intellectual property relating
+# to a hardware implementation of the functionality of the software
+# licensed hereunder.  You may use the software subject to the license
+# terms below provided that you ensure that this notice is replicated
+# unmodified and in its entirety in all distributions of the software,
+# modified or unmodified, in source code or in binary form.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met: redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer;
+# redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in the
+# documentation and/or other materials provided with the distribution;
+# neither the name of the copyright holders nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Authors: Andreas Sandberg
+
+from abc import ABCMeta, abstractmethod
+from datetime import datetime
+import difflib
+import functools
+import os
+import re
+import subprocess
+import sys
+import traceback
+
+from results import UnitResult
+from helpers import *
+
+_test_base = os.path.join(os.path.dirname(__file__), "..")
+
+class TestUnit(object):
+    """Base class for all test units.
+
+    A test unit is a part of a larger test case. Test cases usually
+    contain two types of units, run units (run gem5) and verify units
+    (diff output files). All unit implementations inherit from this
+    class.
+
+    A unit implementation overrides the _run() method. The test runner
+    calls the run() method, which wraps _run() to protect against
+    exceptions.
+
+    """
+
+    __metaclass__ = ABCMeta
+
+    def __init__(self, name, ref_dir, test_dir, skip=False):
+        self.name = name
+        self.ref_dir = ref_dir
+        self.test_dir = test_dir
+        self.force_skip = skip
+        self.start_time = None
+        self.stop_time = None
+
+    def result(self, state, **kwargs):
+        if self.start_time is not None and "runtime" not in kwargs:
+            self.stop_time = datetime.utcnow()
+            delta = self.stop_time - self.start_time
+            kwargs["runtime"] = delta.total_seconds()
+
+        return UnitResult(self.name, state, **kwargs)
+
+    def ok(self, **kwargs):
+        return self.result(UnitResult.STATE_OK, **kwargs)
+
+    def skip(self, **kwargs):
+        return self.result(UnitResult.STATE_SKIPPED, **kwargs)
+
+    def error(self, message, **kwargs):
+        return self.result(UnitResult.STATE_ERROR, message=message, **kwargs)
+
+    def failure(self, message, **kwargs):
+        return self.result(UnitResult.STATE_FAILURE, message=message, **kwargs)
+
+    def ref_file(self, fname):
+        return os.path.join(self.ref_dir, fname)
+
+    def out_file(self, fname):
+        return os.path.join(self.test_dir, fname)
+
+    def _read_output(self, fname, default=""):
+        try:
+            with open(self.out_file(fname), "r") as f:
+                return f.read()
+        except IOError:
+            return default
+
+    def run(self):
+        self.start_time = datetime.utcnow()
+        try:
+            if self.force_skip:
+                return self.skip()
+            else:
+                return self._run()
+        except:
+            return self.error("Python exception:\n%s" % traceback.format_exc())
+
+    @abstractmethod
+    def _run(self):
+        pass
+
+class RunGem5(TestUnit):
+    """Test unit representing a gem5 run.
+
+    Possible failure modes:
+       - gem5 failed to run -> STATE_ERROR
+       - timeout -> STATE_ERROR
+       - non-zero exit code -> STATE_ERROR
+
+    Possible non-failure results:
+       - exit code == 0 -> STATE_OK
+       - exit code == 2 -> STATE_SKIPPED
+    """
+
+    def __init__(self, gem5, gem5_args, timeout=0, **kwargs):
+        super(RunGem5, self).__init__("gem5", **kwargs)
+        self.gem5 = gem5
+        self.args = gem5_args
+        self.timeout = timeout
+
+    def _run(self):
+        gem5_cmd = [
+            self.gem5,
+            "-d", self.test_dir,
+            "-re",
+        ] + self.args
+
+        try:
+            with ProcessHelper(gem5_cmd, stdout=subprocess.PIPE,
+                               stderr=subprocess.PIPE) as p:
+                status, gem5_stdout, gem5_stderr = p.call(timeout=self.timeout)
+        except CallTimeoutException as te:
+            return self.error("Timeout", stdout=te.stdout, stderr=te.stderr)
+        except OSError as ose:
+            return self.error("Failed to launch gem5: %s" % ose)
+
+        stderr = "\n".join([
+            "*** gem5 stderr ***",
+            gem5_stderr,
+            "",
+            "*** m5out/simerr ***",
+            self._read_output("simerr"),
+        ])
+
+        stdout = "\n".join([
+            "*** gem5 stdout ***",
+            gem5_stdout,
+            "",
+            "*** m5out/simout ***",
+            self._read_output("simout"),
+        ])
+
+        # Signal
+        if status < 0:
+            return self.error("gem5 terminated by signal %i" % (-status, ),
+                              stdout=stdout, stderr=stderr)
+        elif status == 2:
+            return self.skip(stdout=stdout, stderr=stderr)
+        elif status > 0:
+            return self.error("gem5 exited with non-zero status: %i" % status,
+                              stdout=stdout, stderr=stderr)
+        else:
+            return self.ok(stdout=stdout, stderr=stderr)
+
+class DiffOutFile(TestUnit):
+    """Test unit comparing and output file and a reference file."""
+
+    # regular expressions of lines to ignore when diffing outputs
+    diff_ignore_regexes = {
+        "simout" : [
+            re.compile('^Redirecting (stdout|stderr) to'),
+            re.compile('^gem5 compiled '),
+            re.compile('^gem5 started '),
+            re.compile('^gem5 executing on '),
+            re.compile('^command line:'),
+            re.compile("^Couldn't import dot_parser,"),
+            re.compile("^info: kernel located at:"),
+            re.compile("^Couldn't unlink "),
+            re.compile("^Using GPU kernel code file\(s\) "),
+        ],
+        "simerr" : [
+            #re.compile('^Simulation complete at'),
+        ],
+        "config.ini" : [
+            re.compile("^(executable|readfile|kernel|image_file)="),
+            re.compile("^(cwd|input|codefile)="),
+        ],
+        "config.json" : [
+            re.compile(r'''^\s*"(executable|readfile|kernel|image_file)":'''),
+            re.compile(r'''^\s*"(cwd|input|codefile)":'''),
+        ],
+    }
+
+    def __init__(self, fname, **kwargs):
+        super(DiffOutFile, self).__init__("diff[%s]" % fname,
+                                          **kwargs)
+
+        self.fname = fname
+        self.line_filters = DiffOutFile.diff_ignore_regexes.get(fname, tuple())
+
+    def _filter_file(self, fname):
+        def match_line(l):
+            for r in self.line_filters:
+                if r.match(l):
+                    return True
+            return False
+
+        with open(fname, "r") as f:
+            for l in f:
+                if not match_line(l):
+                    yield l
+
+
+    def _run(self):
+        fname = self.fname
+        ref = self.ref_file(fname)
+        out = self.out_file(fname)
+
+        if not os.path.exists(ref):
+            return self.error("%s doesn't exist in reference directory" \
+                              % fname)
+
+        if not os.path.exists(out):
+            return self.error("%s doesn't exist in output directory" % fname)
+
+        diff = difflib.unified_diff(
+            tuple(self._filter_file(ref)),
+            tuple(self._filter_file(out)),
+            fromfile="ref/%s" % fname, tofile="out/%s" % fname)
+
+        diff = list(diff)
+        if diff:
+            return self.error("ref/%s and out/%s differ" % (fname, fname),
+                              stderr="".join(diff))
+        else:
+            return self.ok(stdout="-- ref/%s and out/%s are identical --" \
+                           % (fname, fname))
+
+class DiffStatFile(TestUnit):
+    """Test unit comparing two gem5 stat files."""
+
+    def __init__(self, **kwargs):
+        super(DiffStatFile, self).__init__("stat_diff", **kwargs)
+
+        self.stat_diff = os.path.join(_test_base, "diff-out")
+
+    def _run(self):
+        stats = "stats.txt"
+
+        cmd = [
+            self.stat_diff,
+            self.ref_file(stats), self.out_file(stats),
+        ]
+        with ProcessHelper(cmd,
+                           stdout=subprocess.PIPE,
+                           stderr=subprocess.PIPE) as p:
+            status, stdout, stderr = p.call()
+
+        if status == 0:
+            return self.ok(stdout=stdout, stderr=stderr)
+        if status == 1:
+            return self.failure("Statistics mismatch",
+                                stdout=stdout, stderr=stderr)
+        else:
+            return self.error("diff-out returned an error: %i" % status,
+                              stdout=stdout, stderr=stderr)

tests/tests.py

@@ -0,0 +1,287 @@
+#!/usr/bin/env python
+#
+# Copyright (c) 2016 ARM Limited
+# All rights reserved
+#
+# The license below extends only to copyright in the software and shall
+# not be construed as granting a license to any other intellectual
+# property including but not limited to intellectual property relating
+# to a hardware implementation of the functionality of the software
+# licensed hereunder.  You may use the software subject to the license
+# terms below provided that you ensure that this notice is replicated
+# unmodified and in its entirety in all distributions of the software,
+# modified or unmodified, in source code or in binary form.
+#
+# Redistribution and use in source and binary forms, with or without
+# modification, are permitted provided that the following conditions are
+# met: redistributions of source code must retain the above copyright
+# notice, this list of conditions and the following disclaimer;
+# redistributions in binary form must reproduce the above copyright
+# notice, this list of conditions and the following disclaimer in the
+# documentation and/or other materials provided with the distribution;
+# neither the name of the copyright holders nor the names of its
+# contributors may be used to endorse or promote products derived from
+# this software without specific prior written permission.
+#
+# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+# "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+# OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+# SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+# LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+# DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+# THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+#
+# Authors: Andreas Sandberg
+
+import argparse
+import sys
+import os
+import pickle
+
+from testing.tests import *
+import testing.results
+
+class ParagraphHelpFormatter(argparse.HelpFormatter):
+    def _fill_text(self, text, width, indent):
+        return "\n\n".join([
+            super(ParagraphHelpFormatter, self)._fill_text(p, width, indent) \
+            for p in text.split("\n\n") ])
+
+formatters = {
+    "junit" : testing.results.JUnit,
+    "text" : testing.results.Text,
+    "summary" : testing.results.TextSummary,
+    "pickle" : testing.results.Pickle,
+}
+
+
+def _add_format_args(parser):
+    parser.add_argument("--format", choices=formatters, default="text",
+                        help="Output format")
+
+    parser.add_argument("--no-junit-xlate-names", action="store_true",
+                        help="Don't translate test names to " \
+                        "package-like names")
+
+    parser.add_argument("--output", "-o",
+                        type=argparse.FileType('w'), default=sys.stdout,
+                        help="Test result output file")
+
+
+def _create_formatter(args):
+    formatter = formatters[args.format]
+    kwargs = {
+        "fout" : args.output,
+        "verbose" : args.verbose
+    }
+
+    if issubclass(formatter, testing.results.JUnit):
+        kwargs.update({
+            "translate_names" : not args.no_junit_xlate_names,
+        })
+
+    return formatter(**kwargs)
+
+
+def _list_tests_args(subparsers):
+    parser = subparsers.add_parser(
+        "list",
+        formatter_class=ParagraphHelpFormatter,
+        help="List available tests",
+        description="List available tests",
+        epilog="""
+        Generate a list of available tests using a list filter.
+
+        The filter is a string consisting of the target ISA optionally
+        followed by the test category and mode separated by
+        slashes. The test names emitted by this command can be fed
+        into the run command.
+
+        For example, to list all quick arm tests, run the following:
+        tests.py list arm/quick
+
+        Non-mandatory parts of the filter string (anything other than
+        the ISA) can be left out or replaced with the wildcard
+        character. For example, all full-system tests can be listed
+        with this command: tests.py list arm/*/fs""")
+
+    parser.add_argument("--ruby-protocol", type=str, default=None,
+                        help="Ruby protocol")
+
+    parser.add_argument("--gpu-isa", type=str, default=None,
+                        help="GPU ISA")
+
+    parser.add_argument("list_filter", metavar="ISA[/category/mode]",
+                        action="append", type=str,
+                        help="List available test cases")
+
+def _list_tests(args):
+    for isa, categories, modes in \
+        ( parse_test_filter(f) for f in args.list_filter ):
+
+        for test in get_tests(isa, categories=categories, modes=modes,
+                              ruby_protocol=args.ruby_protocol,
+                              gpu_isa=args.gpu_isa):
+            print "/".join(test)
+    sys.exit(0)
+
+def _run_tests_args(subparsers):
+    parser = subparsers.add_parser(
+        "run",
+        formatter_class=ParagraphHelpFormatter,
+        help='Run one or more tests',
+        description="Run one or more tests.",
+        epilog="""
+        Run one or more tests described by a gem5 test tuple.
+
+        The test tuple consists of a test category (quick or long), a
+        test mode (fs or se), a workload name, an isa, an operating
+        system, and a config name separate by slashes. For example:
+        quick/se/00.hello/arm/linux/simple-timing
+
+        Available tests can be listed using the 'list' sub-command
+        (e.g., "tests.py list arm/quick" or one of the scons test list
+        targets (e.g., "scons build/ARM/tests/opt/quick.list").
+
+        The test results can be stored in multiple different output
+        formats. See the help for the show command for more details
+        about output formatting.""")
+
+    parser.add_argument("gem5", type=str,
+                        help="gem5 binary")
+
+    parser.add_argument("test", type=str, nargs="*",
+                        help="List of tests to execute")
+
+    parser.add_argument("--directory", "-d",
+                        type=str, default="m5tests",
+                        help="Test work directory")
+
+    parser.add_argument("--timeout", "-t",
+                        type=int, default="0", metavar="MINUTES",
+                        help="Timeout, 0 to disable")
+
+    parser.add_argument("--skip-diff-out", action="store_true",
+                        help="Skip output diffing stage")
+
+    parser.add_argument("--skip-diff-stat", action="store_true",
+                        help="Skip stat diffing stage")
+
+    _add_format_args(parser)
+
+def _run_tests(args):
+    formatter = _create_formatter(args)
+
+    out_base = os.path.abspath(args.directory)
+    if not os.path.exists(out_base):
+        os.mkdir(out_base)
+    tests = []
+    for test_name in args.test:
+        config = ClassicConfig(*test_name.split("/"))
+        out_dir = os.path.join(out_base, "/".join(config))
+        tests.append(
+            ClassicTest(args.gem5, out_dir, config,
+                        timeout=args.timeout,
+                        skip_diff_stat=args.skip_diff_stat,
+                        skip_diff_out=args.skip_diff_out))
+
+    all_results = []
+    print "Running %i tests" % len(tests)
+    for testno, test in enumerate(tests):
+        print "%i: Running '%s'..." % (testno, test)
+
+        all_results.append(test.run())
+
+    formatter.dump_suites(all_results)
+
+def _show_args(subparsers):
+    parser = subparsers.add_parser(
+        "show",
+        formatter_class=ParagraphHelpFormatter,
+        help='Display pickled test results',
+        description='Display pickled test results',
+        epilog="""
+        Reformat the pickled output from one or more test runs. This
+        command is typically used with the output from a single test
+        run, but it can also be used to merge the outputs from
+        multiple runs.
+
+        The 'text' format is a verbose output format that provides
+        information about individual test units and the output from
+        failed tests. It's mainly useful for debugging test failures.
+
+        The 'summary' format provides outputs the results of one test
+        per line with the test's overall status (OK, SKIPPED, or
+        FAILED).
+
+        The 'junit' format is primarily intended for use with CI
+        systems. It provides an XML representation of test
+        status. Similar to the text format, it includes detailed
+        information about test failures. Since many JUnit parser make
+        assume that test names look like Java packet strings, the
+        JUnit formatter automatically to something the looks like a
+        Java class path ('.'->'-', '/'->'.').
+
+        The 'pickle' format stores the raw results in a format that
+        can be reformatted using this command. It's typically used
+        with the show command to merge multiple test results into one
+        pickle file.""")
+
+    _add_format_args(parser)
+
+    parser.add_argument("result", type=argparse.FileType("rb"), nargs="*",
+                        help="Pickled test results")
+
+def _show(args):
+    formatter = _create_formatter(args)
+    suites = sum([ pickle.load(f) for f in args.result ], [])
+    formatter.dump_suites(suites)
+
+_commands = {
+    "list" : (_list_tests, _list_tests_args),
+    "run" : (_run_tests, _run_tests_args),
+    "show" : (_show, _show_args),
+}
+
+def main():
+    parser = argparse.ArgumentParser(
+        formatter_class=ParagraphHelpFormatter,
+        description="""gem5 testing multi tool.""",
+        epilog="""
+        This tool provides an interface to gem5's test framework that
+        doesn't depend on gem5's build system. It supports test
+        listing, running, and output formatting.
+
+        The list sub-command (e.g., "test.py list arm/quick") produces
+        a list of tests tuples that can be used by the run command
+        (e.g., "tests.py run gem5.opt
+        quick/se/00.hello/arm/linux/simple-timing").
+
+        The run command supports several output formats. One of them,
+        pickle, contains the raw output from the tests and can be
+        re-formatted using the show command (e.g., "tests.py show
+        --format summary *.pickle"). Such pickle files are also
+        generated by the build system when scons is used to run
+        regressions.
+
+        See the usage strings for the individual sub-commands for
+        details.""")
+
+    parser.add_argument("--verbose", action="store_true",
+                        help="Produce more verbose output")
+
+    subparsers = parser.add_subparsers(dest="command")
+
+    for key, (impl, cmd_parser) in _commands.items():
+        cmd_parser(subparsers)
+
+    args = parser.parse_args()
+    impl, cmd_parser = _commands[args.command]
+    impl(args)
+
+if __name__ == "__main__":
+    main()