verif_utils

Convenience functions and classes for verification scripts.

Verifier

Base verifier class.

Base verifier class which can be inherited to easily develop a custom verification script for any kernel.

Subclasses must override the OUTPUT_UIDS attribute and the get_actual_results(), get_expected_results() and check_results() methods, at a minimum. With respect to the latter, it suffices to invoke the parent method with an explicit value for rtol or atol.

Attributes:

Name	Type	Description
OUTPUT_UIDS		A list of global symbols representing outputs of the simulation. See simulate() method for details on its use.

Source code in util/sim/verif_utils.py

class Verifier:
    """Base verifier class.

    Base verifier class which can be inherited to easily develop a
    custom verification script for any kernel.

    Subclasses must override the `OUTPUT_UIDS` attribute and the
    [`get_actual_results()`][verif_utils.Verifier.get_actual_results],
    [`get_expected_results()`][verif_utils.Verifier.get_expected_results]
    and [`check_results()`][verif_utils.Verifier.check_results] methods,
    at a minimum. With respect to the latter, it suffices to invoke the
    parent method with an explicit value for `rtol` or `atol`.

    Attributes:
        OUTPUT_UIDS: A list of global symbols representing outputs of the
            simulation. See
            [`simulate()`][verif_utils.Verifier.simulate] method
            for details on its use.
    """

    OUTPUT_UIDS = [None]

    def parser(self):
        """Default argument parser for verification scripts.

        It is an instance of the `ArgumentParser` class from the `argparse`
        module. Subclasses can extend this and add custom arguments via
        the parser's `add_argument()` method.
        """
        parser = argparse.ArgumentParser(allow_abbrev=True)
        parser.add_argument(
            'sim_bin',
            help='The simulator binary to be used to start the simulation',
        )
        parser.add_argument(
            'snitch_bin',
            help='The Snitch binary to be executed by the simulated Snitch hardware')
        parser.add_argument(
            '--symbols-bin',
            help='An optional binary containing the I/O symbols. By default, '
                 'these are searched for in snitch_bin. This argument serves as an '
                 'alternative.')
        parser.add_argument(
            '--log',
            help='Redirect simulation output to this log file')
        parser.add_argument(
            '--dump-results',
            action='store_true',
            help='Dump results even if the simulation does not fail')
        return parser

    def parse_args(self):
        """Parse default verification script arguments.

        Returns:
            args: The arguments passed to the verification script, parsed
                using the [`parser()`][verif_utils.Verifier.parser]
                method.
        """
        return self.parser().parse_args()

    def __init__(self):
        """Default constructor.

        Parses command-line arguments using the
        [`parse_args()`][verif_utils.Verifier.parse_args] method and
        stores them in `self.args` for use by other methods.
        """
        self.args = self.parse_args()

    def simulate(self):
        """Launch simulation and retrieve results.

        Spawns a subprocess to simulate the `snitch_bin` binary, using a
        command of the form `sim_bin snitch_bin <ipc_args>`. It
        communicates with the simulation using inter-process communication
        (IPC) facilities, to poll the program for termination and retrieve
        the memory contents where the results of the simulation are
        stored. The results of the simulation must have global symbols
        associated in `snitch_bin` in order to retrieve their address and
        size in memory from `snitch_bin`. Alternatively, if `symbols_bin`
        is given, the global symbols will be looked up in `symbols_bin`.

        It populates a dictionary, mapping UIDs, as listed in
        `OUTPUT_UIDS`, to their memory contents at the end of the
        simulation, formatted as raw bytes. The dictionary is stored in
        the `self.raw_outputs` attribute.
        """
        # Open ELF file for processing
        elf = Elf(self.args.snitch_bin)

        # Start simulation
        sim = SnitchSim(self.args.sim_bin, self.args.snitch_bin, log=self.args.log)
        sim.start()

        # Wait for kernel execution to be over
        tohost = elf.get_symbol_address('tohost')
        sim.poll(tohost, 1, 0)

        # Read out results from memory
        if self.args.symbols_bin:
            elf = Elf(self.args.symbols_bin)
        self.raw_outputs = {}
        for uid in self.OUTPUT_UIDS:
            address = elf.get_symbol_address(uid)
            size = elf.get_symbol_size(uid)
            self.raw_outputs[uid] = sim.read(address, size)

        # Terminate
        sim.finish(wait_for_sim=True)

    def check_results(self, actual, expected, atol=None, rtol=None):
        """Check if the actual results are within the expected range.

        Args:
            expected: The expected results.
            actual: The actual results.
            atol: Absolute tolerance. The maximum absolute difference
                between the expected and actual results. Mutually
                exclusive with `rtol`.
            rtol: Relative tolerance. The maximum relative difference
                between the expected and actual results. Mutually
                exclusive with `atol`.

        Returns:
            retcode: An exit code representing the status of the
                simulation: 1 if the simulation results do not match the
                expected results, 0 otherwise.
        """
        # Compute absolute error
        expected, actual = map(flatten, (expected, actual))
        err = np.abs(expected - actual)
        # Check absolute or relative error
        if atol is not None and rtol is not None:
            raise ValueError('atol and rtol are mutually exclusive.')
        if atol is not None:
            max_err = [atol] * len(flatten(expected))
            # Handle FlexFloat arrays differently
            if expected.dtype == np.dtype(object):
                success = np.all(err <= max_err)
            else:
                success = np.allclose(expected, actual, atol=atol, rtol=0, equal_nan=False)
        elif rtol is not None:
            max_err = rtol * np.abs(expected)
            # Handle FlexFloat arrays differently
            if expected.dtype == np.dtype(object):
                success = np.all(err <= max_err)
            else:
                success = np.allclose(expected, actual, atol=0, rtol=rtol, equal_nan=False)
        else:
            raise ValueError('Either atol or rtol must be specified.')

        # Dump results on failure or if requested
        if not success or self.args.dump_results:
            dump_results_to_csv(expected, actual, err, max_err, Path.cwd() / 'results.csv')

        # Return exit code
        return int(not success)

    def get_actual_results(self):
        """Get actual simulation results.

        Subclasses should override this method to return the simulation
        results in a format comparable to the expected results.
        """
        pass

    def get_expected_results(self):
        """Get expected simulation results.

        Subclasses should override this method to return the expected
        results from the simulation.
        """
        pass

    def get_input_from_symbol(self, *args):
        """Get the value of an input variable from its global symbol.

        Retrieves the contents of a global symbol from `snitch_bin`.
        Alternatively, if `symbols_bin` is provided, it looks up the
        symbol contents from there.

        Args:
            args: Positional arguments accepted by the
                [`from_symbol()`][Elf.Elf.from_symbol] method.
        """
        if self.args.symbols_bin:
            elf = Elf(self.args.symbols_bin)
        else:
            elf = Elf(self.args.snitch_bin)
        return elf.from_symbol(*args)

    def get_output_from_symbol(self, uid, ctype):
        """Get the value of an output variable from its global symbol.

        Retrieves the value of a global symbol from the memory contents
        at the end of the simulation.

        Note:
            Calling this method before a call to the
            [`simulate()`][verif_utils.Verifier.simulate] method results in
            undefined behaviour.

        Args:
            uid: Global symbol to look up in memory.
            ctype: C type specifier used to interpret the data in memory.
        """
        return from_buffer(self.raw_outputs[uid], ctype)

    def main(self):
        """Default main function for data generation scripts."""
        # Run simulation
        self.simulate()

        # Get actual and expected results
        actual_results = self.get_actual_results()
        expected_results = self.get_expected_results()

        # Compare actual and expected results
        retcode = self.check_results(actual_results, expected_results)

        # Verify that overriden check_results() method returns an exit code
        if retcode is not None:
            return retcode
        else:
            raise ValueError('check_results() method must return an exit code')

__init__()

Default constructor.

Parses command-line arguments using the parse_args() method and stores them in self.args for use by other methods.

Source code in util/sim/verif_utils.py

def __init__(self):
    """Default constructor.

    Parses command-line arguments using the
    [`parse_args()`][verif_utils.Verifier.parse_args] method and
    stores them in `self.args` for use by other methods.
    """
    self.args = self.parse_args()

check_results(actual, expected, atol=None, rtol=None)

Check if the actual results are within the expected range.

Parameters:

Name	Type	Description	Default
expected		The expected results.	required
actual		The actual results.	required
atol		Absolute tolerance. The maximum absolute difference between the expected and actual results. Mutually exclusive with rtol.	None
rtol		Relative tolerance. The maximum relative difference between the expected and actual results. Mutually exclusive with atol.	None

Returns:

Name	Type	Description
retcode		An exit code representing the status of the simulation: 1 if the simulation results do not match the expected results, 0 otherwise.

Source code in util/sim/verif_utils.py

def check_results(self, actual, expected, atol=None, rtol=None):
    """Check if the actual results are within the expected range.

    Args:
        expected: The expected results.
        actual: The actual results.
        atol: Absolute tolerance. The maximum absolute difference
            between the expected and actual results. Mutually
            exclusive with `rtol`.
        rtol: Relative tolerance. The maximum relative difference
            between the expected and actual results. Mutually
            exclusive with `atol`.

    Returns:
        retcode: An exit code representing the status of the
            simulation: 1 if the simulation results do not match the
            expected results, 0 otherwise.
    """
    # Compute absolute error
    expected, actual = map(flatten, (expected, actual))
    err = np.abs(expected - actual)
    # Check absolute or relative error
    if atol is not None and rtol is not None:
        raise ValueError('atol and rtol are mutually exclusive.')
    if atol is not None:
        max_err = [atol] * len(flatten(expected))
        # Handle FlexFloat arrays differently
        if expected.dtype == np.dtype(object):
            success = np.all(err <= max_err)
        else:
            success = np.allclose(expected, actual, atol=atol, rtol=0, equal_nan=False)
    elif rtol is not None:
        max_err = rtol * np.abs(expected)
        # Handle FlexFloat arrays differently
        if expected.dtype == np.dtype(object):
            success = np.all(err <= max_err)
        else:
            success = np.allclose(expected, actual, atol=0, rtol=rtol, equal_nan=False)
    else:
        raise ValueError('Either atol or rtol must be specified.')

    # Dump results on failure or if requested
    if not success or self.args.dump_results:
        dump_results_to_csv(expected, actual, err, max_err, Path.cwd() / 'results.csv')

    # Return exit code
    return int(not success)

get_actual_results()

Get actual simulation results.

Subclasses should override this method to return the simulation results in a format comparable to the expected results.

Source code in util/sim/verif_utils.py

def get_actual_results(self):
    """Get actual simulation results.

    Subclasses should override this method to return the simulation
    results in a format comparable to the expected results.
    """
    pass

get_expected_results()

Get expected simulation results.

Subclasses should override this method to return the expected results from the simulation.

Source code in util/sim/verif_utils.py

def get_expected_results(self):
    """Get expected simulation results.

    Subclasses should override this method to return the expected
    results from the simulation.
    """
    pass

get_input_from_symbol(*args)

Get the value of an input variable from its global symbol.

Retrieves the contents of a global symbol from snitch_bin. Alternatively, if symbols_bin is provided, it looks up the symbol contents from there.

Parameters:

Name	Type	Description	Default
args		Positional arguments accepted by the from_symbol() method.	()

Source code in util/sim/verif_utils.py

def get_input_from_symbol(self, *args):
    """Get the value of an input variable from its global symbol.

    Retrieves the contents of a global symbol from `snitch_bin`.
    Alternatively, if `symbols_bin` is provided, it looks up the
    symbol contents from there.

    Args:
        args: Positional arguments accepted by the
            [`from_symbol()`][Elf.Elf.from_symbol] method.
    """
    if self.args.symbols_bin:
        elf = Elf(self.args.symbols_bin)
    else:
        elf = Elf(self.args.snitch_bin)
    return elf.from_symbol(*args)

get_output_from_symbol(uid, ctype)

Get the value of an output variable from its global symbol.

Retrieves the value of a global symbol from the memory contents at the end of the simulation.

Note

Calling this method before a call to the simulate() method results in undefined behaviour.

Parameters:

Name	Type	Description	Default
uid		Global symbol to look up in memory.	required
ctype		C type specifier used to interpret the data in memory.	required

Source code in util/sim/verif_utils.py

def get_output_from_symbol(self, uid, ctype):
    """Get the value of an output variable from its global symbol.

    Retrieves the value of a global symbol from the memory contents
    at the end of the simulation.

    Note:
        Calling this method before a call to the
        [`simulate()`][verif_utils.Verifier.simulate] method results in
        undefined behaviour.

    Args:
        uid: Global symbol to look up in memory.
        ctype: C type specifier used to interpret the data in memory.
    """
    return from_buffer(self.raw_outputs[uid], ctype)

main()

Default main function for data generation scripts.

Source code in util/sim/verif_utils.py

def main(self):
    """Default main function for data generation scripts."""
    # Run simulation
    self.simulate()

    # Get actual and expected results
    actual_results = self.get_actual_results()
    expected_results = self.get_expected_results()

    # Compare actual and expected results
    retcode = self.check_results(actual_results, expected_results)

    # Verify that overriden check_results() method returns an exit code
    if retcode is not None:
        return retcode
    else:
        raise ValueError('check_results() method must return an exit code')

parse_args()

Parse default verification script arguments.

Returns:

Name	Type	Description
args		The arguments passed to the verification script, parsed using the parser() method.

Source code in util/sim/verif_utils.py

def parse_args(self):
    """Parse default verification script arguments.

    Returns:
        args: The arguments passed to the verification script, parsed
            using the [`parser()`][verif_utils.Verifier.parser]
            method.
    """
    return self.parser().parse_args()

parser()

Default argument parser for verification scripts.

It is an instance of the ArgumentParser class from the argparse module. Subclasses can extend this and add custom arguments via the parser's add_argument() method.

Source code in util/sim/verif_utils.py

def parser(self):
    """Default argument parser for verification scripts.

    It is an instance of the `ArgumentParser` class from the `argparse`
    module. Subclasses can extend this and add custom arguments via
    the parser's `add_argument()` method.
    """
    parser = argparse.ArgumentParser(allow_abbrev=True)
    parser.add_argument(
        'sim_bin',
        help='The simulator binary to be used to start the simulation',
    )
    parser.add_argument(
        'snitch_bin',
        help='The Snitch binary to be executed by the simulated Snitch hardware')
    parser.add_argument(
        '--symbols-bin',
        help='An optional binary containing the I/O symbols. By default, '
             'these are searched for in snitch_bin. This argument serves as an '
             'alternative.')
    parser.add_argument(
        '--log',
        help='Redirect simulation output to this log file')
    parser.add_argument(
        '--dump-results',
        action='store_true',
        help='Dump results even if the simulation does not fail')
    return parser

simulate()

Launch simulation and retrieve results.

Spawns a subprocess to simulate the snitch_bin binary, using a command of the form sim_bin snitch_bin <ipc_args>. It communicates with the simulation using inter-process communication (IPC) facilities, to poll the program for termination and retrieve the memory contents where the results of the simulation are stored. The results of the simulation must have global symbols associated in snitch_bin in order to retrieve their address and size in memory from snitch_bin. Alternatively, if symbols_bin is given, the global symbols will be looked up in symbols_bin.

It populates a dictionary, mapping UIDs, as listed in OUTPUT_UIDS, to their memory contents at the end of the simulation, formatted as raw bytes. The dictionary is stored in the self.raw_outputs attribute.

Source code in util/sim/verif_utils.py

def simulate(self):
    """Launch simulation and retrieve results.

    Spawns a subprocess to simulate the `snitch_bin` binary, using a
    command of the form `sim_bin snitch_bin <ipc_args>`. It
    communicates with the simulation using inter-process communication
    (IPC) facilities, to poll the program for termination and retrieve
    the memory contents where the results of the simulation are
    stored. The results of the simulation must have global symbols
    associated in `snitch_bin` in order to retrieve their address and
    size in memory from `snitch_bin`. Alternatively, if `symbols_bin`
    is given, the global symbols will be looked up in `symbols_bin`.

    It populates a dictionary, mapping UIDs, as listed in
    `OUTPUT_UIDS`, to their memory contents at the end of the
    simulation, formatted as raw bytes. The dictionary is stored in
    the `self.raw_outputs` attribute.
    """
    # Open ELF file for processing
    elf = Elf(self.args.snitch_bin)

    # Start simulation
    sim = SnitchSim(self.args.sim_bin, self.args.snitch_bin, log=self.args.log)
    sim.start()

    # Wait for kernel execution to be over
    tohost = elf.get_symbol_address('tohost')
    sim.poll(tohost, 1, 0)

    # Read out results from memory
    if self.args.symbols_bin:
        elf = Elf(self.args.symbols_bin)
    self.raw_outputs = {}
    for uid in self.OUTPUT_UIDS:
        address = elf.get_symbol_address(uid)
        size = elf.get_symbol_size(uid)
        self.raw_outputs[uid] = sim.read(address, size)

    # Terminate
    sim.finish(wait_for_sim=True)

dump_results_to_csv(expected_results, actual_results, error, max_error, path)

Dumps results and errors to a CSV file.

Takes a set of arrays (of the same shape or, at least, same flattened size), flattens them, and dumps them to a CSV file. Each array is mapped to a different column of the CSV, in the same order as they appear as arguments in the function signature. Also dumps an additional column, indicating pass (or failure) status, calculated as error <= max_error.

Parameters:

Name	Type	Description	Default
expected_results		Array of expected results.	required
actual_results		Array of actual results.	required
error		Array with the absolute error abs(expected_results - actual_results).	required
max_error		Array with the maximum allowed error per element. Can also be a function of expected_results e.g. to implement relative error checks.	required
path		Path of the output CSV file.	required

Source code in util/sim/verif_utils.py

def dump_results_to_csv(expected_results, actual_results, error, max_error, path):
    """Dumps results and errors to a CSV file.

    Takes a set of arrays (of the same shape or, at least, same flattened
    size), flattens them, and dumps them to a CSV file. Each array is
    mapped to a different column of the CSV, in the same order as they
    appear as arguments in the function signature.
    Also dumps an additional column, indicating pass (or failure) status,
    calculated as `error <= max_error`.

    Args:
        expected_results: Array of expected results.
        actual_results: Array of actual results.
        error: Array with the absolute error
            `abs(expected_results - actual_results)`.
        max_error: Array with the maximum allowed error per element.
            Can also be a function of `expected_results` e.g. to
            implement relative error checks.
        path: Path of the output CSV file.
    """
    # Flatten and zip arrays
    arrays = (expected_results, actual_results, error, max_error)
    flattened = [flatten(arr) for arr in arrays]
    flattened.append((flattened[2] <= flattened[3]).astype(bool).astype(str))
    zipped = np.column_stack(flattened)
    # Write row-by-row to CSV file
    with open(path, 'w') as csv_file:
        csv_writer = csv.writer(csv_file)
        csv_writer.writerow(['expected', 'actual', 'error', 'max_error', 'pass'])
        for row in zipped:
            csv_writer.writerow(row)
    # Print path where results were written
    print(f"Wrote results to {path}")