jyoung/external/simpleperf/report_html.py

#!/usr/bin/python
#
# Copyright (C) 2017 The Android Open Source Project
#
# 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.
#

import argparse
import datetime
import json
import os
import subprocess
import sys
import tempfile

from simpleperf_report_lib import ReportLib
from utils import *


class HtmlWriter(object):

    def __init__(self, output_path):
        self.fh = open(output_path, 'w')
        self.tag_stack = []

    def close(self):
        self.fh.close()

    def open_tag(self, tag, **attrs):
        attr_str = ''
        for key in attrs:
            attr_str += ' %s="%s"' % (key, attrs[key])
        self.fh.write('<%s%s>' % (tag, attr_str))
        self.tag_stack.append(tag)
        return self

    def close_tag(self, tag=None):
        if tag:
            assert tag == self.tag_stack[-1]
        self.fh.write('</%s>\n' % self.tag_stack.pop())

    def add(self, text):
        self.fh.write(text)
        return self

    def add_file(self, file_path):
        file_path = os.path.join(get_script_dir(), file_path)
        with open(file_path, 'r') as f:
            self.add(f.read())
        return self

def modify_text_for_html(text):
    return text.replace('>', '&gt;').replace('<', '&lt;')

class EventScope(object):

    def __init__(self, name):
        self.name = name
        self.processes = {}  # map from pid to ProcessScope
        self.sample_count = 0
        self.event_count = 0

    def get_process(self, pid):
        process = self.processes.get(pid)
        if not process:
            process = self.processes[pid] = ProcessScope(pid)
        return process

    def get_sample_info(self, gen_addr_hit_map):
        result = {}
        result['eventName'] = self.name
        result['eventCount'] = self.event_count
        result['processes'] = [process.get_sample_info(gen_addr_hit_map)
                                  for process in self.processes.values()]
        return result


class ProcessScope(object):

    def __init__(self, pid):
        self.pid = pid
        self.name = ''
        self.event_count = 0
        self.threads = {}  # map from tid to ThreadScope

    def get_thread(self, tid, thread_name):
        thread = self.threads.get(tid)
        if not thread:
            thread = self.threads[tid] = ThreadScope(tid)
        thread.name = thread_name
        if self.pid == tid:
            self.name = thread_name
        return thread

    def get_sample_info(self, gen_addr_hit_map):
        result = {}
        result['pid'] = self.pid
        result['eventCount'] = self.event_count
        result['threads'] = [thread.get_sample_info(gen_addr_hit_map)
                                for thread in self.threads.values()]
        return result


class ThreadScope(object):

    def __init__(self, tid):
        self.tid = tid
        self.name = ''
        self.event_count = 0
        self.libs = {}  # map from lib_id to LibScope

    def add_callstack(self, event_count, callstack, build_addr_hit_map):
        """ callstack is a list of tuple (lib_id, func_id, addr).
            For each i > 0, callstack[i] calls callstack[i-1]."""
        hit_func_ids = set()
        for i in range(len(callstack)):
            lib_id, func_id, addr = callstack[i]
            # When a callstack contains recursive function, only add for each function once.
            if func_id in hit_func_ids:
                continue
            hit_func_ids.add(func_id)

            lib = self.libs.get(lib_id)
            if not lib:
                lib = self.libs[lib_id] = LibScope(lib_id)
            function = lib.get_function(func_id)
            if i == 0:
                lib.event_count += event_count
                function.sample_count += 1
            function.add_reverse_callchain(callstack, i + 1, len(callstack), event_count)

            if build_addr_hit_map:
                function.build_addr_hit_map(addr, event_count if i == 0 else 0, event_count)

        hit_func_ids.clear()
        for i in range(len(callstack) - 1, -1, -1):
            lib_id, func_id, _ = callstack[i]
            # When a callstack contains recursive function, only add for each function once.
            if func_id in hit_func_ids:
                continue
            hit_func_ids.add(func_id)
            lib = self.libs.get(lib_id)
            lib.get_function(func_id).add_callchain(callstack, i - 1, -1, event_count)

    def get_sample_info(self, gen_addr_hit_map):
        result = {}
        result['tid'] = self.tid
        result['eventCount'] = self.event_count
        result['libs'] = [lib.gen_sample_info(gen_addr_hit_map)
                            for lib in self.libs.values()]
        return result


class LibScope(object):

    def __init__(self, lib_id):
        self.lib_id = lib_id
        self.event_count = 0
        self.functions = {}  # map from func_id to FunctionScope.

    def get_function(self, func_id):
        function = self.functions.get(func_id)
        if not function:
            function = self.functions[func_id] = FunctionScope(func_id)
        return function

    def gen_sample_info(self, gen_addr_hit_map):
        result = {}
        result['libId'] = self.lib_id
        result['eventCount'] = self.event_count
        result['functions'] = [func.gen_sample_info(gen_addr_hit_map)
                                  for func in self.functions.values()]
        return result


class FunctionScope(object):

    def __init__(self, func_id):
        self.sample_count = 0
        self.call_graph = CallNode(func_id)
        self.reverse_call_graph = CallNode(func_id)
        self.addr_hit_map = None  # map from addr to [event_count, subtree_event_count].
        # map from (source_file_id, line) to [event_count, subtree_event_count].
        self.line_hit_map = None

    def add_callchain(self, callchain, start, end, event_count):
        node = self.call_graph
        for i in range(start, end, -1):
            node = node.get_child(callchain[i][1])
        node.event_count += event_count

    def add_reverse_callchain(self, callchain, start, end, event_count):
        node = self.reverse_call_graph
        for i in range(start, end):
            node = node.get_child(callchain[i][1])
        node.event_count += event_count

    def build_addr_hit_map(self, addr, event_count, subtree_event_count):
        if self.addr_hit_map is None:
            self.addr_hit_map = {}
        count_info = self.addr_hit_map.get(addr)
        if count_info is None:
            self.addr_hit_map[addr] = [event_count, subtree_event_count]
        else:
            count_info[0] += event_count
            count_info[1] += subtree_event_count

    def build_line_hit_map(self, source_file_id, line, event_count, subtree_event_count):
        if self.line_hit_map is None:
            self.line_hit_map = {}
        key = (source_file_id, line)
        count_info = self.line_hit_map.get(key)
        if count_info is None:
            self.line_hit_map[key] = [event_count, subtree_event_count]
        else:
            count_info[0] += event_count
            count_info[1] += subtree_event_count

    def update_subtree_event_count(self):
        a = self.call_graph.update_subtree_event_count()
        b = self.reverse_call_graph.update_subtree_event_count()
        return max(a, b)

    def limit_callchain_percent(self, min_callchain_percent, hit_func_ids):
        min_limit = min_callchain_percent * 0.01 * self.call_graph.subtree_event_count
        self.call_graph.cut_edge(min_limit, hit_func_ids)
        self.reverse_call_graph.cut_edge(min_limit, hit_func_ids)

    def gen_sample_info(self, gen_addr_hit_map):
        result = {}
        result['c'] = self.sample_count
        result['g'] = self.call_graph.gen_sample_info()
        result['rg'] = self.reverse_call_graph.gen_sample_info()
        if self.line_hit_map:
            items = []
            for key in self.line_hit_map:
                count_info = self.line_hit_map[key]
                item = {'f': key[0], 'l': key[1], 'e': count_info[0], 's': count_info[1]}
                items.append(item)
            result['s'] = items
        if gen_addr_hit_map and self.addr_hit_map:
            items = []
            for addr in sorted(self.addr_hit_map):
                count_info = self.addr_hit_map[addr]
                items.append({'a': addr, 'e': count_info[0], 's': count_info[1]})
            result['a'] = items
        return result


class CallNode(object):

    def __init__(self, func_id):
        self.event_count = 0
        self.subtree_event_count = 0
        self.func_id = func_id
        self.children = {}  # map from func_id to CallNode

    def get_child(self, func_id):
        child = self.children.get(func_id)
        if not child:
            child = self.children[func_id] = CallNode(func_id)
        return child

    def update_subtree_event_count(self):
        self.subtree_event_count = self.event_count
        for child in self.children.values():
            self.subtree_event_count += child.update_subtree_event_count()
        return self.subtree_event_count

    def cut_edge(self, min_limit, hit_func_ids):
        hit_func_ids.add(self.func_id)
        to_del_children = []
        for key in self.children:
            child = self.children[key]
            if child.subtree_event_count < min_limit:
                to_del_children.append(key)
            else:
                child.cut_edge(min_limit, hit_func_ids)
        for key in to_del_children:
            del self.children[key]

    def gen_sample_info(self):
        result = {}
        result['e'] = self.event_count
        result['s'] = self.subtree_event_count
        result['f'] = self.func_id
        result['c'] = [child.gen_sample_info() for child in self.children.values()]
        return result


class LibSet(object):
    """ Collection of shared libraries used in perf.data. """
    def __init__(self):
        self.lib_name_to_id = {}
        self.lib_id_to_name = []

    def get_lib_id(self, lib_name):
        lib_id = self.lib_name_to_id.get(lib_name)
        if lib_id is None:
            lib_id = len(self.lib_id_to_name)
            self.lib_name_to_id[lib_name] = lib_id
            self.lib_id_to_name.append(lib_name)
        return lib_id

    def get_lib_name(self, lib_id):
        return self.lib_id_to_name[lib_id]


class Function(object):
    """ Represent a function in a shared library. """
    def __init__(self, lib_id, func_name, func_id, start_addr, addr_len):
        self.lib_id = lib_id
        self.func_name = func_name
        self.func_id = func_id
        self.start_addr = start_addr
        self.addr_len = addr_len
        self.source_info = None
        self.disassembly = None


class FunctionSet(object):
    """ Collection of functions used in perf.data. """
    def __init__(self):
        self.name_to_func = {}
        self.id_to_func = {}

    def get_func_id(self, lib_id, symbol):
        key = (lib_id, symbol.symbol_name)
        function = self.name_to_func.get(key)
        if function is None:
            func_id = len(self.id_to_func)
            function = Function(lib_id, symbol.symbol_name, func_id, symbol.symbol_addr,
                                symbol.symbol_len)
            self.name_to_func[key] = function
            self.id_to_func[func_id] = function
        return function.func_id

    def trim_functions(self, left_func_ids):
        """ Remove functions excepts those in left_func_ids. """
        for function in self.name_to_func.values():
            if function.func_id not in left_func_ids:
                del self.id_to_func[function.func_id]
        # name_to_func will not be used.
        self.name_to_func = None


class SourceFile(object):
    """ A source file containing source code hit by samples. """
    def __init__(self, file_id, abstract_path):
        self.file_id = file_id
        self.abstract_path = abstract_path  # path reported by addr2line
        self.real_path = None  # file path in the file system
        self.requested_lines = set()
        self.line_to_code = {}  # map from line to code in that line.

    def request_lines(self, start_line, end_line):
        self.requested_lines |= set(range(start_line, end_line + 1))

    def add_source_code(self, real_path):
        self.real_path = real_path
        with open(real_path, 'r') as f:
            source_code = f.readlines()
        max_line = len(source_code)
        for line in self.requested_lines:
            if line > 0 and line <= max_line:
                self.line_to_code[line] = source_code[line - 1]
        # requested_lines is no longer used.
        self.requested_lines = None


class SourceFileSet(object):
    """ Collection of source files. """
    def __init__(self):
        self.path_to_source_files = {}  # map from file path to SourceFile.

    def get_source_file(self, file_path):
        source_file = self.path_to_source_files.get(file_path)
        if source_file is None:
            source_file = SourceFile(len(self.path_to_source_files), file_path)
            self.path_to_source_files[file_path] = source_file
        return source_file

    def load_source_code(self, source_dirs):
        file_searcher = SourceFileSearcher(source_dirs)
        for source_file in self.path_to_source_files.values():
            real_path = file_searcher.get_real_path(source_file.abstract_path)
            if real_path:
                source_file.add_source_code(real_path)


class SourceFileSearcher(object):

    SOURCE_FILE_EXTS = {'.h', '.hh', '.H', '.hxx', '.hpp', '.h++',
                        '.c', '.cc', '.C', '.cxx', '.cpp', '.c++',
                        '.java', '.kt'}

    @classmethod
    def is_source_filename(cls, filename):
        ext = os.path.splitext(filename)[1]
        return ext in cls.SOURCE_FILE_EXTS

    """" Find source file paths in the file system.
        The file paths reported by addr2line are the paths stored in debug sections
        of shared libraries. And we need to convert them to file paths in the file
        system. It is done in below steps:
        1. Collect all file paths under the provided source_dirs. The suffix of a
           source file should contain one of below:
            h: for C/C++ header files.
            c: for C/C++ source files.
            java: for Java source files.
            kt: for Kotlin source files.
        2. Given an abstract_path reported by addr2line, select the best real path
           as below:
           2.1 Find all real paths with the same file name as the abstract path.
           2.2 Select the real path having the longest common suffix with the abstract path.
    """
    def __init__(self, source_dirs):
        # Map from filename to a list of reversed directory path containing filename.
        self.filename_to_rparents = {}
        self._collect_paths(source_dirs)

    def _collect_paths(self, source_dirs):
        for source_dir in source_dirs:
            for parent, _, file_names in os.walk(source_dir):
                rparent = None
                for file_name in file_names:
                    if self.is_source_filename(file_name):
                        rparents = self.filename_to_rparents.get(file_name)
                        if rparents is None:
                            rparents = self.filename_to_rparents[file_name] = []
                        if rparent is None:
                            rparent = parent[::-1]
                        rparents.append(rparent)

    def get_real_path(self, abstract_path):
        abstract_path = abstract_path.replace('/', os.sep)
        abstract_parent, file_name = os.path.split(abstract_path)
        abstract_rparent = abstract_parent[::-1]
        real_rparents = self.filename_to_rparents.get(file_name)
        if real_rparents is None:
            return None
        best_matched_rparent = None
        best_common_length = -1
        for real_rparent in real_rparents:
            length = len(os.path.commonprefix((real_rparent, abstract_rparent)))
            if length > best_common_length:
                best_common_length = length
                best_matched_rparent = real_rparent
        if best_matched_rparent is None:
            return None
        return os.path.join(best_matched_rparent[::-1], file_name)


class RecordData(object):

    """RecordData reads perf.data, and generates data used by report.js in json format.
        All generated items are listed as below:
            1. recordTime: string
            2. machineType: string
            3. androidVersion: string
            4. recordCmdline: string
            5. totalSamples: int
            6. processNames: map from pid to processName.
            7. threadNames: map from tid to threadName.
            8. libList: an array of libNames, indexed by libId.
            9. functionMap: map from functionId to funcData.
                funcData = {
                    l: libId
                    f: functionName
                    s: [sourceFileId, startLine, endLine] [optional]
                    d: [(disassembly, addr)] [optional]
                }

            10.  sampleInfo = [eventInfo]
                eventInfo = {
                    eventName
                    eventCount
                    processes: [processInfo]
                }
                processInfo = {
                    pid
                    eventCount
                    threads: [threadInfo]
                }
                threadInfo = {
                    tid
                    eventCount
                    libs: [libInfo],
                }
                libInfo = {
                    libId,
                    eventCount,
                    functions: [funcInfo]
                }
                funcInfo = {
                    c: sampleCount
                    g: callGraph
                    rg: reverseCallgraph
                    s: [sourceCodeInfo] [optional]
                    a: [addrInfo] (sorted by addrInfo.addr) [optional]
                }
                callGraph and reverseCallGraph are both of type CallNode.
                callGraph shows how a function calls other functions.
                reverseCallGraph shows how a function is called by other functions.
                CallNode {
                    e: selfEventCount
                    s: subTreeEventCount
                    f: functionId
                    c: [CallNode] # children
                }

                sourceCodeInfo {
                    f: sourceFileId
                    l: line
                    e: eventCount
                    s: subtreeEventCount
                }

                addrInfo {
                    a: addr
                    e: eventCount
                    s: subtreeEventCount
                }

            11. sourceFiles: an array of sourceFile, indexed by sourceFileId.
                sourceFile {
                    path
                    code:  # a map from line to code for that line.
                }
    """

    def __init__(self, binary_cache_path, ndk_path, build_addr_hit_map):
        self.binary_cache_path = binary_cache_path
        self.ndk_path = ndk_path
        self.build_addr_hit_map = build_addr_hit_map
        self.meta_info = None
        self.cmdline = None
        self.arch = None
        self.events = {}
        self.libs = LibSet()
        self.functions = FunctionSet()
        self.total_samples = 0
        self.source_files = SourceFileSet()
        self.gen_addr_hit_map_in_record_info = False

    def load_record_file(self, record_file, show_art_frames):
        lib = ReportLib()
        lib.SetRecordFile(record_file)
        # If not showing ip for unknown symbols, the percent of the unknown symbol may be
        # accumulated to very big, and ranks first in the sample table.
        lib.ShowIpForUnknownSymbol()
        if show_art_frames:
            lib.ShowArtFrames()
        if self.binary_cache_path:
            lib.SetSymfs(self.binary_cache_path)
        self.meta_info = lib.MetaInfo()
        self.cmdline = lib.GetRecordCmd()
        self.arch = lib.GetArch()
        while True:
            raw_sample = lib.GetNextSample()
            if not raw_sample:
                lib.Close()
                break
            raw_event = lib.GetEventOfCurrentSample()
            symbol = lib.GetSymbolOfCurrentSample()
            callchain = lib.GetCallChainOfCurrentSample()
            event = self._get_event(raw_event.name)
            self.total_samples += 1
            event.sample_count += 1
            event.event_count += raw_sample.period
            process = event.get_process(raw_sample.pid)
            process.event_count += raw_sample.period
            thread = process.get_thread(raw_sample.tid, raw_sample.thread_comm)
            thread.event_count += raw_sample.period

            lib_id = self.libs.get_lib_id(symbol.dso_name)
            func_id = self.functions.get_func_id(lib_id, symbol)
            callstack = [(lib_id, func_id, symbol.vaddr_in_file)]
            for i in range(callchain.nr):
                symbol = callchain.entries[i].symbol
                lib_id = self.libs.get_lib_id(symbol.dso_name)
                func_id = self.functions.get_func_id(lib_id, symbol)
                callstack.append((lib_id, func_id, symbol.vaddr_in_file))
            thread.add_callstack(raw_sample.period, callstack, self.build_addr_hit_map)

        for event in self.events.values():
            for process in event.processes.values():
                for thread in process.threads.values():
                    for lib in thread.libs.values():
                        for func_id in lib.functions:
                            function = lib.functions[func_id]
                            function.update_subtree_event_count()

    def limit_percents(self, min_func_percent, min_callchain_percent):
        hit_func_ids = set()
        for event in self.events.values():
            min_limit = event.event_count * min_func_percent * 0.01
            for process in event.processes.values():
                for thread in process.threads.values():
                    for lib in thread.libs.values():
                        to_del_func_ids = []
                        for func_id in lib.functions:
                            function = lib.functions[func_id]
                            if function.call_graph.subtree_event_count < min_limit:
                                to_del_func_ids.append(func_id)
                            else:
                                function.limit_callchain_percent(min_callchain_percent,
                                                                 hit_func_ids)
                        for func_id in to_del_func_ids:
                            del lib.functions[func_id]
        self.functions.trim_functions(hit_func_ids)

    def _get_event(self, event_name):
        if event_name not in self.events:
            self.events[event_name] = EventScope(event_name)
        return self.events[event_name]

    def add_source_code(self, source_dirs):
        """ Collect source code information:
            1. Find line ranges for each function in FunctionSet.
            2. Find line for each addr in FunctionScope.addr_hit_map.
            3. Collect needed source code in SourceFileSet.
        """
        addr2line = Addr2Nearestline(self.ndk_path, self.binary_cache_path)
        # Request line range for each function.
        for function in self.functions.id_to_func.values():
            if function.func_name == 'unknown':
                continue
            lib_name = self.libs.get_lib_name(function.lib_id)
            addr2line.add_addr(lib_name, function.start_addr, function.start_addr)
            addr2line.add_addr(lib_name, function.start_addr,
                               function.start_addr + function.addr_len - 1)
        # Request line for each addr in FunctionScope.addr_hit_map.
        for event in self.events.values():
            for process in event.processes.values():
                for thread in process.threads.values():
                    for lib in thread.libs.values():
                        lib_name = self.libs.get_lib_name(lib.lib_id)
                        for function in lib.functions.values():
                            func_addr = self.functions.id_to_func[
                                            function.call_graph.func_id].start_addr
                            for addr in function.addr_hit_map:
                                addr2line.add_addr(lib_name, func_addr, addr)
        addr2line.convert_addrs_to_lines()

        # Set line range for each function.
        for function in self.functions.id_to_func.values():
            if function.func_name == 'unknown':
                continue
            dso = addr2line.get_dso(self.libs.get_lib_name(function.lib_id))
            start_source = addr2line.get_addr_source(dso, function.start_addr)
            end_source = addr2line.get_addr_source(dso,
                            function.start_addr + function.addr_len - 1)
            if not start_source or not end_source:
                continue
            start_file_path, start_line = start_source[-1]
            end_file_path, end_line = end_source[-1]
            if start_file_path != end_file_path or start_line > end_line:
                continue
            source_file = self.source_files.get_source_file(start_file_path)
            source_file.request_lines(start_line, end_line)
            function.source_info = (source_file.file_id, start_line, end_line)

        # Build FunctionScope.line_hit_map.
        for event in self.events.values():
            for process in event.processes.values():
                for thread in process.threads.values():
                    for lib in thread.libs.values():
                        dso = addr2line.get_dso(self.libs.get_lib_name(lib.lib_id))
                        for function in lib.functions.values():
                            for addr in function.addr_hit_map:
                                source = addr2line.get_addr_source(dso, addr)
                                if not source:
                                    continue
                                for file_path, line in source:
                                    source_file = self.source_files.get_source_file(file_path)
                                    # Show [line - 5, line + 5] of the line hit by a sample.
                                    source_file.request_lines(line - 5, line + 5)
                                    count_info = function.addr_hit_map[addr]
                                    function.build_line_hit_map(source_file.file_id, line,
                                                                count_info[0], count_info[1])

        # Collect needed source code in SourceFileSet.
        self.source_files.load_source_code(source_dirs)

    def add_disassembly(self):
        """ Collect disassembly information:
            1. Use objdump to collect disassembly for each function in FunctionSet.
            2. Set flag to dump addr_hit_map when generating record info.
        """
        objdump = Objdump(self.ndk_path, self.binary_cache_path)
        for function in self.functions.id_to_func.values():
            if function.func_name == 'unknown':
                continue
            lib_name = self.libs.get_lib_name(function.lib_id)
            code = objdump.disassemble_code(lib_name, function.start_addr, function.addr_len)
            function.disassembly = code

        self.gen_addr_hit_map_in_record_info = True

    def gen_record_info(self):
        record_info = {}
        timestamp = self.meta_info.get('timestamp')
        if timestamp:
            t = datetime.datetime.fromtimestamp(int(timestamp))
        else:
            t = datetime.datetime.now()
        record_info['recordTime'] = t.strftime('%Y-%m-%d (%A) %H:%M:%S')

        product_props = self.meta_info.get('product_props')
        machine_type = self.arch
        if product_props:
            manufacturer, model, name = product_props.split(':')
            machine_type = '%s (%s) by %s, arch %s' % (model, name, manufacturer, self.arch)
        record_info['machineType'] = machine_type
        record_info['androidVersion'] = self.meta_info.get('android_version', '')
        record_info['recordCmdline'] = self.cmdline
        record_info['totalSamples'] = self.total_samples
        record_info['processNames'] = self._gen_process_names()
        record_info['threadNames'] = self._gen_thread_names()
        record_info['libList'] = self._gen_lib_list()
        record_info['functionMap'] = self._gen_function_map()
        record_info['sampleInfo'] = self._gen_sample_info()
        record_info['sourceFiles'] = self._gen_source_files()
        return record_info

    def _gen_process_names(self):
        process_names = {}
        for event in self.events.values():
            for process in event.processes.values():
                process_names[process.pid] = process.name
        return process_names

    def _gen_thread_names(self):
        thread_names = {}
        for event in self.events.values():
            for process in event.processes.values():
                for thread in process.threads.values():
                    thread_names[thread.tid] = thread.name
        return thread_names

    def _gen_lib_list(self):
        return [modify_text_for_html(x) for x in self.libs.lib_id_to_name]

    def _gen_function_map(self):
        func_map = {}
        for func_id in sorted(self.functions.id_to_func):
            function = self.functions.id_to_func[func_id]
            func_data = {}
            func_data['l'] = function.lib_id
            func_data['f'] = modify_text_for_html(function.func_name)
            if function.source_info:
                func_data['s'] = function.source_info
            if function.disassembly:
                disassembly_list = []
                for code, addr in function.disassembly:
                    disassembly_list.append([modify_text_for_html(code), addr])
                func_data['d'] = disassembly_list
            func_map[func_id] = func_data
        return func_map

    def _gen_sample_info(self):
        return [event.get_sample_info(self.gen_addr_hit_map_in_record_info)
                    for event in self.events.values()]

    def _gen_source_files(self):
        source_files = sorted(self.source_files.path_to_source_files.values(),
                              key=lambda x: x.file_id)
        file_list = []
        for source_file in source_files:
            file_data = {}
            if not source_file.real_path:
                file_data['path'] = ''
                file_data['code'] = {}
            else:
                file_data['path'] = source_file.real_path
                code_map = {}
                for line in source_file.line_to_code:
                    code_map[line] = modify_text_for_html(source_file.line_to_code[line])
                file_data['code'] = code_map
            file_list.append(file_data)
        return file_list


class ReportGenerator(object):

    def __init__(self, html_path):
        self.hw = HtmlWriter(html_path)
        self.hw.open_tag('html')
        self.hw.open_tag('head')
        self.hw.open_tag('link', rel='stylesheet', type='text/css',
            href='https://code.jquery.com/ui/1.12.0/themes/smoothness/jquery-ui.css'
                         ).close_tag()

        self.hw.open_tag('link', rel='stylesheet', type='text/css',
             href='https://cdn.datatables.net/1.10.16/css/jquery.dataTables.min.css'
                         ).close_tag()
        self.hw.open_tag('script', src='https://www.gstatic.com/charts/loader.js').close_tag()
        self.hw.open_tag('script').add(
            "google.charts.load('current', {'packages': ['corechart', 'table']});").close_tag()
        self.hw.open_tag('script', src='https://code.jquery.com/jquery-3.2.1.js').close_tag()
        self.hw.open_tag('script', src='https://code.jquery.com/ui/1.12.1/jquery-ui.js'
                         ).close_tag()
        self.hw.open_tag('script',
            src='https://cdn.datatables.net/1.10.16/js/jquery.dataTables.min.js').close_tag()
        self.hw.open_tag('script',
            src='https://cdn.datatables.net/1.10.16/js/dataTables.jqueryui.min.js').close_tag()
        self.hw.open_tag('style', type='text/css').add("""
            .colForLine { width: 50px; }
            .colForCount { width: 100px; }
            .tableCell { font-size: 17px; }
            .boldTableCell { font-weight: bold; font-size: 17px; }
            """).close_tag()
        self.hw.close_tag('head')
        self.hw.open_tag('body')
        self.record_info = {}

    def write_content_div(self):
        self.hw.open_tag('div', id='report_content').close_tag()

    def write_record_data(self, record_data):
        self.hw.open_tag('script', id='record_data', type='application/json')
        self.hw.add(json.dumps(record_data))
        self.hw.close_tag()

    def write_flamegraph(self, flamegraph):
        self.hw.add(flamegraph)

    def write_script(self):
        self.hw.open_tag('script').add_file('report_html.js').close_tag()

    def finish(self):
        self.hw.close_tag('body')
        self.hw.close_tag('html')
        self.hw.close()


def gen_flamegraph(record_file, show_art_frames):
    fd, flamegraph_path = tempfile.mkstemp()
    os.close(fd)
    inferno_script_path = os.path.join(get_script_dir(), 'inferno', 'inferno.py')
    args = [sys.executable, inferno_script_path, '-sc', '-o', flamegraph_path,
            '--record_file', record_file, '--embedded_flamegraph', '--no_browser']
    if show_art_frames:
        args.append('--show_art_frames')
    subprocess.check_call(args)
    with open(flamegraph_path, 'r') as fh:
        data = fh.read()
    remove(flamegraph_path)
    return data


def main():
    parser = argparse.ArgumentParser(description='report profiling data')
    parser.add_argument('-i', '--record_file', nargs='+', default=['perf.data'], help="""
                        Set profiling data file to report. Default is perf.data.""")
    parser.add_argument('-o', '--report_path', default='report.html', help="""
                        Set output html file. Default is report.html.""")
    parser.add_argument('--min_func_percent', default=0.01, type=float, help="""
                        Set min percentage of functions shown in the report.
                        For example, when set to 0.01, only functions taking >= 0.01%% of total
                        event count are collected in the report. Default is 0.01.""")
    parser.add_argument('--min_callchain_percent', default=0.01, type=float, help="""
                        Set min percentage of callchains shown in the report.
                        It is used to limit nodes shown in the function flamegraph. For example,
                        when set to 0.01, only callchains taking >= 0.01%% of the event count of
                        the starting function are collected in the report. Default is 0.01.""")
    parser.add_argument('--add_source_code', action='store_true', help='Add source code.')
    parser.add_argument('--source_dirs', nargs='+', help='Source code directories.')
    parser.add_argument('--add_disassembly', action='store_true', help='Add disassembled code.')
    parser.add_argument('--ndk_path', nargs=1, help='Find tools in the ndk path.')
    parser.add_argument('--no_browser', action='store_true', help="Don't open report in browser.")
    parser.add_argument('--show_art_frames', action='store_true',
                        help='Show frames of internal methods in the ART Java interpreter.')
    args = parser.parse_args()

    # 1. Process args.
    binary_cache_path = 'binary_cache'
    if not os.path.isdir(binary_cache_path):
        if args.add_source_code or args.add_disassembly:
            log_exit("""binary_cache/ doesn't exist. Can't add source code or disassembled code
                        without collected binaries. Please run binary_cache_builder.py to
                        collect binaries for current profiling data, or run app_profiler.py
                        without -nb option.""")
        binary_cache_path = None

    if args.add_source_code and not args.source_dirs:
        log_exit('--source_dirs is needed to add source code.')
    build_addr_hit_map = args.add_source_code or args.add_disassembly
    ndk_path = None if not args.ndk_path else args.ndk_path[0]

    # 2. Produce record data.
    record_data = RecordData(binary_cache_path, ndk_path, build_addr_hit_map)
    for record_file in args.record_file:
        record_data.load_record_file(record_file, args.show_art_frames)
    record_data.limit_percents(args.min_func_percent, args.min_callchain_percent)
    if args.add_source_code:
        record_data.add_source_code(args.source_dirs)
    if args.add_disassembly:
        record_data.add_disassembly()

    # 3. Generate report html.
    report_generator = ReportGenerator(args.report_path)
    report_generator.write_content_div()
    report_generator.write_record_data(record_data.gen_record_info())
    report_generator.write_script()
    # TODO: support multiple perf.data in flamegraph.
    if len(args.record_file) > 1:
        log_warning('flamegraph will only be shown for %s' % args.record_file[0])
    flamegraph = gen_flamegraph(args.record_file[0], args.show_art_frames)
    report_generator.write_flamegraph(flamegraph)
    report_generator.finish()

    if not args.no_browser:
        open_report_in_browser(args.report_path)
    log_info("Report generated at '%s'." % args.report_path)


if __name__ == '__main__':
    main()