scripts/vk_validation_stats.py - master - redgpu/Vulkan-LoaderAndValidationLayers

scripts/vk_validation_stats.py
453 строки25 KB
John Zulauf
layers: Improve doc validation init-list parsing
31 янв 2018, 19:12
31 янв 2018, 19:1243f9ad8
#!/usr/bin/env python3
# Copyright (c) 2015-2017 The Khronos Group Inc.
# Copyright (c) 2015-2017 Valve Corporation
# Copyright (c) 2015-2017 LunarG, Inc.
# Copyright (c) 2015-2017 Google Inc.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# Author: Tobin Ehlis <tobine@google.com>

import argparse
import os
import sys
import platform

# vk_validation_stats.py overview
#
# usage:
#    python vk_validation_stats.py [verbose]
#
#    Arguments:
#        verbose - enables verbose output, including VUID duplicates
#
# This script is intended to generate statistics on the state of validation code
#  based on information parsed from the source files and the database file
# Here's what it currently does:
#  1. Parse vk_validation_error_database.txt to store claimed state of validation checks
#  2. Parse vk_validation_error_messages.h to verify the actual checks in header vs. the
#     claimed state of the checks
#  3. Parse source files to identify which checks are implemented and verify that this
#     exactly matches the list of checks claimed to be implemented in the database
#  4. Parse test file(s) and verify that reported tests exist
#  5. Report out stats on number of checks, implemented checks, and duplicated checks
#
# If a mis-match is found during steps 2, 3, or 4, then the script exits w/ a non-zero error code
#  otherwise, the script will exit(0)
#
# TODO:
#  1. Would also like to report out number of existing checks that don't yet use new, unique enum
#  2. Could use notes to store custom fields (like TODO) and print those out here
#  3. Update test code to check if tests use new, unique enums to check for errors instead of strings

db_file = '../layers/vk_validation_error_database.txt'
generated_layer_source_directories = [
'build',
'dbuild',
'release',
]
generated_layer_source_files = [
'parameter_validation.cpp',
'object_tracker.cpp',
]
layer_source_files = [
'../layers/core_validation.cpp',
'../layers/descriptor_sets.cpp',
'../layers/parameter_validation_utils.cpp',
'../layers/object_tracker_utils.cpp',
'../layers/shader_validation.cpp',
'../layers/buffer_validation.cpp',
]
header_file = '../layers/vk_validation_error_messages.h'
# TODO : Don't hardcode linux path format if we want this to run on windows
test_file = '../tests/layer_validation_tests.cpp'
# List of enums that are allowed to be used more than once so don't warn on their duplicates
duplicate_exceptions = [
'VALIDATION_ERROR_258004ea', # This covers the broad case that all child objects must be destroyed at DestroyInstance time
'VALIDATION_ERROR_24a002f4', # This covers the broad case that all child objects must be destroyed at DestroyDevice time
'VALIDATION_ERROR_0280006e', # Obj tracker check makes sure non-null framebuffer is valid & CV check makes sure it's compatible w/ renderpass framebuffer
'VALIDATION_ERROR_12200682', # This is an aliasing error that we report twice, for each of the two allocations that are aliasing
'VALIDATION_ERROR_1060d201', # Covers valid shader module handle for both Compute & Graphics pipelines
'VALIDATION_ERROR_0c20c601', # This is a case for VkMappedMemoryRange struct that is used by both Flush & Invalidate MappedMemoryRange
'VALIDATION_ERROR_0a400c01', # This is a blanket case for all invalid image aspect bit errors. The spec link has appropriate details for all separate cases.
'VALIDATION_ERROR_0a8007fc', # This case covers two separate checks which are done independently
'VALIDATION_ERROR_0a800800', # This case covers two separate checks which are done independently
'VALIDATION_ERROR_15c0028a', # This is a descriptor set write update error that we use for a couple copy cases as well
'VALIDATION_ERROR_1bc002de', # Single error for mis-matched stageFlags of vkCmdPushConstants() that is flagged for no stage flags & mis-matched flags
'VALIDATION_ERROR_1880000e', # Handles both depth/stencil & compressed image errors for vkCmdClearColorImage()
'VALIDATION_ERROR_0a600152', # Used for the mipLevel check of both dst & src images on vkCmdCopyImage call
'VALIDATION_ERROR_0a600154', # Used for the arraySize check of both dst & src images on vkCmdCopyImage call
'VALIDATION_ERROR_1500099e', # Used for both x & y bounds of viewport
'VALIDATION_ERROR_1d8004a6', # Used for both x & y value of scissors to make sure they're not negative
'VALIDATION_ERROR_1462ec01', # Surface of VkSwapchainCreateInfoKHR must be valid when creating both single or shared swapchains
'VALIDATION_ERROR_1460de01', # oldSwapchain of VkSwapchainCreateInfoKHR must be valid when creating both single or shared swapchains
'VALIDATION_ERROR_146009f2', # Single error for both imageFormat & imageColorSpace requirements when creating swapchain
'VALIDATION_ERROR_15c00294', # Used twice for the same error codepath as both a param & to set a variable, so not really a duplicate
]

class ValidationDatabase:
    def __init__(self, filename=db_file):
        self.db_file = filename
        self.delimiter = '~^~'
        self.db_dict = {} # complete dict of all db values per error enum
        # specialized data structs with slices of complete dict
        self.db_implemented_enums = [] # list of all error enums claiming to be implemented in database file
        self.db_unimplemented_implicit = [] # list of all implicit checks that aren't marked implemented
        self.db_enum_to_tests = {} # dict where enum is key to lookup list of tests implementing the enum
        self.db_invalid_implemented = [] # list of checks with invalid check_implemented flags
        #self.src_implemented_enums
    def read(self):
        """Read a database file into internal data structures, format of each line is <enum><implemented Y|N?><testname><api><errormsg><notes>"""
        #db_dict = {} # This is a simple db of just enum->errormsg, the same as is created from spec
        #max_id = 0
        with open(self.db_file, "r", encoding="utf8") as infile:
            for line in infile:
                line = line.strip()
                if line.startswith('#') or '' == line:
                    continue
                db_line = line.split(self.delimiter)
                if len(db_line) != 8:
                    print("ERROR: Bad database line doesn't have 8 elements: %s" % (line))
                error_enum = db_line[0]
                implemented = db_line[1]
                testname = db_line[2]
                api = db_line[3]
                vuid_string = db_line[4]
                core_ext = db_line[5]
                error_str = db_line[6]
                note = db_line[7]
                # Read complete database contents into our class var for later use
                self.db_dict[error_enum] = {}
                self.db_dict[error_enum]['check_implemented'] = implemented
                self.db_dict[error_enum]['testname'] = testname
                self.db_dict[error_enum]['api'] = api
                self.db_dict[error_enum]['vuid_string'] = vuid_string
                self.db_dict[error_enum]['core_ext'] = core_ext
                self.db_dict[error_enum]['error_string'] = error_str
                self.db_dict[error_enum]['note'] = note
                # Now build custom data structs
                if 'Y' == implemented:
                    self.db_implemented_enums.append(error_enum)
                elif 'implicit' in note: # only make note of non-implemented implicit checks
                    self.db_unimplemented_implicit.append(error_enum)
                if implemented not in ['Y', 'N']:
                    self.db_invalid_implemented.append(error_enum)
                if testname.lower() not in ['unknown', 'none', 'nottestable']:
                    self.db_enum_to_tests[error_enum] = testname.split(',')
                    #if len(self.db_enum_to_tests[error_enum]) > 1:
                    #    print "Found check %s that has multiple tests: %s" % (error_enum, self.db_enum_to_tests[error_enum])
                    #else:
                    #    print "Check %s has single test: %s" % (error_enum, self.db_enum_to_tests[error_enum])
                #unique_id = int(db_line[0].split('_')[-1])
                #if unique_id > max_id:
                #    max_id = unique_id
        #print "Found %d total enums in database" % (len(self.db_dict.keys()))
        #print "Found %d enums claiming to be implemented in source" % (len(self.db_implemented_enums))
        #print "Found %d enums claiming to have tests implemented" % (len(self.db_enum_to_tests.keys()))

class ValidationHeader:
    def __init__(self, filename=header_file):
        self.filename = header_file
        self.enums = []
    def read(self):
        """Read unique error enum header file into internal data structures"""
        grab_enums = False
        with open(self.filename, "r") as infile:
            for line in infile:
                line = line.strip()
                if 'enum UNIQUE_VALIDATION_ERROR_CODE {' in line:
                    grab_enums = True
                    continue
                if grab_enums:
                    if 'VALIDATION_ERROR_MAX_ENUM' in line:
                        grab_enums = False
                        break # done
                    elif 'VALIDATION_ERROR_UNDEFINED' in line:
                        continue
                    elif 'VALIDATION_ERROR_' in line:
                        enum = line.split(' = ')[0]
                        self.enums.append(enum)
        #print "Found %d error enums. First is %s and last is %s." % (len(self.enums), self.enums[0], self.enums[-1])

class ValidationSource:
    def __init__(self, source_file_list, generated_source_file_list, generated_source_directories):
        self.source_files = source_file_list
        self.generated_source_files = generated_source_file_list
        self.generated_source_dirs = generated_source_directories

        if len(self.generated_source_files) > 0:
            qualified_paths = []
            for source in self.generated_source_files:
                for build_dir in self.generated_source_dirs:
                    filepath = '../%s/layers/%s' % (build_dir, source)
                    if os.path.isfile(filepath):
                        qualified_paths.append(filepath)
                        break
            if len(self.generated_source_files) != len(qualified_paths):
                print("Error: Unable to locate one or more of the following source files in the %s directories" % (", ".join(generated_source_directories)))
                print(self.generated_source_files)
                print("Skipping documentation validation test")
                exit(1)
            else:
                self.source_files.extend(qualified_paths)

        self.enum_count_dict = {} # dict of enum values to the count of how much they're used, and location of where they're used
    def parse(self):
        duplicate_checks = 0
        for sf in self.source_files:
            line_num = 0
            with open(sf) as f:
                for line in f:
                    line_num = line_num + 1
                    if True in [line.strip().startswith(comment) for comment in ['//', '/*']]:
                        continue
                    # Find enums
                    #if 'VALIDATION_ERROR_' in line and True not in [ignore in line for ignore in ['[VALIDATION_ERROR_', 'UNIQUE_VALIDATION_ERROR_CODE']]:
                    if 'VALIDATION_ERROR_' in line:
                        # Need to isolate the validation error enum
                        #print("Line has check:%s" % (line))
                        line_list = line.split()
                        enum_list = []
                        for str in line_list:
                            if 'VALIDATION_ERROR_' in str and True not in [ignore_str in str for ignore_str in ['[VALIDATION_ERROR_', 'VALIDATION_ERROR_UNDEFINED', 'UNIQUE_VALIDATION_ERROR_CODE']]:
                                enum_list.append(str.strip(',);{}'))
                                #break
                        for enum in enum_list:
                            if enum != '':
                                if enum not in self.enum_count_dict:
                                    self.enum_count_dict[enum] = {}
                                    self.enum_count_dict[enum]['count'] = 1
                                    self.enum_count_dict[enum]['file_line'] = []
                                    self.enum_count_dict[enum]['file_line'].append('%s,%d' % (sf, line_num))
                                    #print "Found enum %s implemented for first time in file %s" % (enum, sf)
                                else:
                                    self.enum_count_dict[enum]['count'] = self.enum_count_dict[enum]['count'] + 1
                                    self.enum_count_dict[enum]['file_line'].append('%s,%d' % (sf, line_num))
                                    #print "Found enum %s implemented for %d time in file %s" % (enum, self.enum_count_dict[enum], sf)
                                    duplicate_checks = duplicate_checks + 1
                            #else:
                                #print("Didn't find actual check in line:%s" % (line))
        #print "Found %d unique implemented checks and %d are duplicated at least once" % (len(self.enum_count_dict.keys()), duplicate_checks)

# Class to parse the validation layer test source and store testnames
# TODO: Enhance class to detect use of unique error enums in the test
class TestParser:
    def __init__(self, test_file_list, test_group_name=['VkLayerTest', 'VkPositiveLayerTest', 'VkWsiEnabledLayerTest']):
        self.test_files = test_file_list
        self.test_to_errors = {} # Dict where testname maps to list of error enums found in that test
        self.test_trigger_txt_list = []
        for tg in test_group_name:
            self.test_trigger_txt_list.append('TEST_F(%s' % tg)
            #print('Test trigger test list: %s' % (self.test_trigger_txt_list))

    # Parse test files into internal data struct
    def parse(self):
        # For each test file, parse test names into set
        grab_next_line = False # handle testname on separate line than wildcard
        testname = ''
        for test_file in self.test_files:
            with open(test_file) as tf:
                for line in tf:
                    if True in [line.strip().startswith(comment) for comment in ['//', '/*']]:
                        continue

                    if True in [ttt in line for ttt in self.test_trigger_txt_list]:
                        #print('Test wildcard in line: %s' % (line))
                        testname = line.split(',')[-1]
                        testname = testname.strip().strip(' {)')
                        #print('Inserting test: "%s"' % (testname))
                        if ('' == testname):
                            grab_next_line = True
                            continue
                        self.test_to_errors[testname] = []
                    if grab_next_line: # test name on its own line
                        grab_next_line = False
                        testname = testname.strip().strip(' {)')
                        self.test_to_errors[testname] = []
                    if ' VALIDATION_ERROR_' in line:
                        line_list = line.split()
                        for sub_str in line_list:
                            if 'VALIDATION_ERROR_' in sub_str and True not in [ignore_str in sub_str for ignore_str in ['VALIDATION_ERROR_UNDEFINED', 'UNIQUE_VALIDATION_ERROR_CODE', 'VALIDATION_ERROR_MAX_ENUM']]:
                                #print("Trying to add enums for line: %s" % ())
                                #print("Adding enum %s to test %s" % (sub_str.strip(',);'), testname))
                                self.test_to_errors[testname].append(sub_str.strip(',);'))

# Little helper class for coloring cmd line output
class bcolors:

    def __init__(self):
        self.GREEN = '\033[0;32m'
        self.RED = '\033[0;31m'
        self.YELLOW = '\033[1;33m'
        self.ENDC = '\033[0m'
        if 'Linux' != platform.system():
            self.GREEN = ''
            self.RED = ''
            self.YELLOW = ''
            self.ENDC = ''

    def green(self):
        return self.GREEN

    def red(self):
        return self.RED

    def yellow(self):
        return self.YELLOW

    def endc(self):
        return self.ENDC

def main(argv):
    result = 0 # Non-zero result indicates an error case
    verbose_mode = 'verbose' in sys.argv
    # parse db
    val_db = ValidationDatabase()
    val_db.read()
    # parse header
    val_header = ValidationHeader()
    val_header.read()
    # Create parser for layer files
    val_source = ValidationSource(layer_source_files, generated_layer_source_files, generated_layer_source_directories)
    val_source.parse()
    # Parse test files
    test_parser = TestParser([test_file, ])
    test_parser.parse()

    # Process stats - Just doing this inline in main, could make a fancy class to handle
    #   all the processing of data and then get results from that
    txt_color = bcolors()
    if verbose_mode:
        print("Validation Statistics")
    else:
        print("Validation/Documentation Consistency Test")
    # First give number of checks in db & header and report any discrepancies
    db_enums = len(val_db.db_dict.keys())
    hdr_enums = len(val_header.enums)
    if verbose_mode:
        print(" Database file includes %d unique checks" % (db_enums))
        print(" Header file declares %d unique checks" % (hdr_enums))

    # Report any checks that have an invalid check_implemented flag
    if len(val_db.db_invalid_implemented) > 0:
        result = 1
        print(txt_color.red() + "The following checks have an invalid check_implemented flag (must be 'Y' or 'N'):" + txt_color.endc())
        for invalid_imp_enum in val_db.db_invalid_implemented:
            check_implemented = val_db.db_dict[invalid_imp_enum]['check_implemented']
            print(txt_color.red() + "    %s has check_implemented flag '%s'" % (invalid_imp_enum, check_implemented) + txt_color.endc())

    # Report details about how well the Database and Header are synchronized.
    tmp_db_dict = val_db.db_dict
    db_missing = []
    for enum in val_header.enums:
        if not tmp_db_dict.pop(enum, False):
            db_missing.append(enum)
    if db_enums == hdr_enums and len(db_missing) == 0 and len(tmp_db_dict.keys()) == 0:
        if verbose_mode:
            print(txt_color.green() + "  Database and Header match, GREAT!" + txt_color.endc())
    else:
        print(txt_color.red() + "  Uh oh, Database doesn't match Header :(" + txt_color.endc())
        result = 1
        if len(db_missing) != 0:
            print(txt_color.red() + "   The following checks are in header but missing from database:" + txt_color.endc())
            for missing_enum in db_missing:
                print(txt_color.red() + "    %s" % (missing_enum) + txt_color.endc())
        if len(tmp_db_dict.keys()) != 0:
            print(txt_color.red() + "   The following checks are in database but haven't been declared in the header:" + txt_color.endc())
            for extra_enum in tmp_db_dict:
                print(txt_color.red() + "    %s" % (extra_enum) + txt_color.endc())

    # Report out claimed implemented checks vs. found actual implemented checks
    imp_not_found = [] # Checks claimed to implemented in DB file but no source found
    imp_not_claimed = [] # Checks found implemented but not claimed to be in DB
    multiple_uses = False # Flag if any enums are used multiple times
    for db_imp in val_db.db_implemented_enums:
        if db_imp not in val_source.enum_count_dict:
            imp_not_found.append(db_imp)
    for src_enum in val_source.enum_count_dict:
        if val_source.enum_count_dict[src_enum]['count'] > 1 and src_enum not in duplicate_exceptions:
            multiple_uses = True
        if src_enum not in val_db.db_implemented_enums:
            imp_not_claimed.append(src_enum)
    if verbose_mode:
        print(" Database file claims that %d checks (%s) are implemented in source." % (len(val_db.db_implemented_enums), "{0:.0f}%".format(float(len(val_db.db_implemented_enums))/db_enums * 100)))

    if len(val_db.db_unimplemented_implicit) > 0 and verbose_mode:
        print(" Database file claims %d implicit checks (%s) that are not implemented." % (len(val_db.db_unimplemented_implicit), "{0:.0f}%".format(float(len(val_db.db_unimplemented_implicit))/db_enums * 100)))
        total_checks = len(val_db.db_implemented_enums) + len(val_db.db_unimplemented_implicit)
        print(" If all implicit checks are handled by parameter validation this is a total of %d (%s) checks covered." % (total_checks, "{0:.0f}%".format(float(total_checks)/db_enums * 100)))
    if len(imp_not_found) == 0 and len(imp_not_claimed) == 0:
        if verbose_mode:
            print(txt_color.green() + "  All claimed Database implemented checks have been found in source, and no source checks aren't claimed in Database, GREAT!" + txt_color.endc())
    else:
        result = 1
        print(txt_color.red() + "  Uh oh, Database claimed implemented don't match Source :(" + txt_color.endc())
        if len(imp_not_found) != 0:
            print(txt_color.red() + "   The following %d checks are claimed to be implemented in Database, but weren't found in source:" % (len(imp_not_found)) + txt_color.endc())
            for not_imp_enum in imp_not_found:
                print(txt_color.red() + "    %s" % (not_imp_enum) + txt_color.endc())
        if len(imp_not_claimed) != 0:
            print(txt_color.red() + "   The following checks are implemented in source, but not claimed to be in Database:" + txt_color.endc())
            for imp_enum in imp_not_claimed:
                print(txt_color.red() + "    %s" % (imp_enum) + txt_color.endc())

    if multiple_uses and verbose_mode:
        print(txt_color.yellow() + "  Note that some checks are used multiple times. These may be good candidates for new valid usage spec language." + txt_color.endc())
        print(txt_color.yellow() + "  Here is a list of each check used multiple times with its number of uses:" + txt_color.endc())
        for enum in val_source.enum_count_dict:
            if val_source.enum_count_dict[enum]['count'] > 1 and enum not in duplicate_exceptions:
                print(txt_color.yellow() + "   %s: %d uses in file,line:" % (enum, val_source.enum_count_dict[enum]['count']) + txt_color.endc())
                for file_line in val_source.enum_count_dict[enum]['file_line']:
                    print(txt_color.yellow() + "   \t%s" % (file_line) + txt_color.endc())

    # Now check that tests claimed to be implemented are actual test names
    bad_testnames = []
    tests_missing_enum = {} # Report tests that don't use validation error enum to check for error case
    for enum in val_db.db_enum_to_tests:
        for testname in val_db.db_enum_to_tests[enum]:
            if testname not in test_parser.test_to_errors:
                bad_testnames.append(testname)
            else:
                enum_found = False
                for test_enum in test_parser.test_to_errors[testname]:
                    if test_enum == enum:
                        #print("Found test that correctly checks for enum: %s" % (enum))
                        enum_found = True
                if not enum_found:
                    #print("Test %s is not using enum %s to check for error" % (testname, enum))
                    if testname not in tests_missing_enum:
                        tests_missing_enum[testname] = []
                    tests_missing_enum[testname].append(enum)
    if tests_missing_enum and verbose_mode:
        print(txt_color.yellow() + "  \nThe following tests do not use their reported enums to check for the validation error. You may want to update these to pass the expected enum to SetDesiredFailureMsg:" + txt_color.endc())
        for testname in tests_missing_enum:
            print(txt_color.yellow() + "   Testname %s does not explicitly check for these ids:" % (testname) + txt_color.endc())
            for enum in tests_missing_enum[testname]:
                print(txt_color.yellow() + "    %s" % (enum) + txt_color.endc())

    # TODO : Go through all enums found in the test file and make sure they're correctly documented in the database file
    if verbose_mode:
        print(" Database file claims that %d checks have tests written." % len(val_db.db_enum_to_tests))
    if len(bad_testnames) == 0:
        if verbose_mode:
            print(txt_color.green() + "  All claimed tests have valid names. That's good!" + txt_color.endc())
    else:
        print(txt_color.red() + "  The following testnames in Database appear to be invalid:")
        result = 1
        for bt in bad_testnames:
            print(txt_color.red() + "   %s" % (bt) + txt_color.endc())

    return result

if __name__ == "__main__":
    sys.exit(main(sys.argv[1:]))