mace_tools.py

# Copyright 2018 Xiaomi, Inc.  All rights reserved.
#
# 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.

# Must run at root dir of libmace project.
# python tools/mace_tools.py \
#     --config=tools/example.yaml \
#     --round=100 \
#     --mode=all

import argparse
import filelock
import hashlib
import os
import sh
import subprocess
import sys
import urllib
import yaml
import re

import sh_commands

from ConfigParser import ConfigParser


def get_target_socs(configs):
    available_socs = sh_commands.adb_get_all_socs()
    target_socs = available_socs
    if hasattr(configs, "target_socs"):
        target_socs = set(configs["target_socs"])
        target_socs = target_socs & available_socs

    if FLAGS.target_socs != "all":
        socs = set(FLAGS.target_socs.split(','))
        target_socs = target_socs & socs
        missing_socs = socs.difference(target_socs)
        if len(missing_socs) > 0:
            print(
                "Error: devices with SoCs are not connected %s" % missing_socs)
            exit(1)

    if not target_socs:
        print("Error: no device to run")
        exit(1)

    return target_socs


def get_data_and_device_type(runtime):
    data_type = ""
    device_type = ""

    if runtime == "dsp":
        data_type = "DT_UINT8"
        device_type = "HEXAGON"
    elif runtime == "gpu":
        data_type = "DT_HALF"
        device_type = "OPENCL"
    elif runtime == "cpu":
        data_type = "DT_FLOAT"
        device_type = "CPU"
    elif runtime == "neon":
        data_type = "DT_FLOAT"
        device_type = "NEON"

    return data_type, device_type


def get_hexagon_mode(configs):
    runtime_list = []
    for model_name in configs["models"]:
        model_runtime = configs["models"][model_name]["runtime"]
        runtime_list.append(model_runtime.lower())

    global_runtime = ""
    if "dsp" in runtime_list:
        return True
    return False


def generate_code(target_soc, target_abi, model_output_dirs, pull_or_not):
    if pull_or_not:
        sh_commands.pull_binaries(
                target_soc, target_abi, model_output_dirs)
    sh_commands.gen_opencl_binary_code(
            target_soc, target_abi, model_output_dirs)
    sh_commands.gen_tuning_param_code(
            target_soc, target_abi, model_output_dirs)


def model_benchmark_stdout_processor(stdout,
                                     target_soc,
                                     abi,
                                     runtime,
                                     running_round,
                                     tuning):
    metrics = {}
    for line in stdout:
        if "Aborted" in line:
            raise Exception("Command failed")
        line = line.strip()
        parts = line.split()
        if len(parts) == 6 and parts[0].startswith("time"):
            metrics["%s.create_net_ms" % model_name] = str(float(parts[1]))
            metrics["%s.mace_engine_ctor_ms" % model_name] = str(
                float(parts[2]))
            metrics["%s.init_ms" % model_name] = str(float(parts[3]))
            metrics["%s.warmup_ms" % model_name] = str(float(parts[4]))
            if float(parts[5]) > 0:
                metrics["%s.avg_latency_ms" % model_name] = str(
                    float(parts[5]))
    tags = {
        "ro.board.platform": target_soc,
        "abi": abi,
        "runtime": runtime,
        "round": running_round,  # TODO(yejianwu) change this to source/binary
        "tuning": tuning
    }
    sh_commands.falcon_push_metrics(
        metrics, endpoint="mace_model_benchmark", tags=tags)


def tuning_run(runtime,
               target_soc,
               target_abi,
               vlog_level,
               embed_model_data,
               model_output_dir,
               input_nodes,
               output_nodes,
               input_shapes,
               output_shapes,
               model_name,
               device_type,
               running_round,
               restart_round,
               out_of_range_check,
               phone_data_dir,
               tuning=False,
               limit_opencl_kernel_time=0,
               option_args=""):
    stdout = sh_commands.tuning_run(
            target_soc,
            target_abi,
            vlog_level,
            embed_model_data,
            model_output_dir,
            input_nodes,
            output_nodes,
            input_shapes,
            output_shapes,
            model_name,
            device_type,
            running_round,
            restart_round,
            limit_opencl_kernel_time,
            tuning,
            out_of_range_check,
            phone_data_dir,
            option_args)
    model_benchmark_stdout_processor(stdout,
                                   target_soc,
                                   target_abi,
                                   runtime,
                                   running_round,
                                   tuning)


def build_mace_run_prod(hexagon_mode, runtime, target_soc, target_abi,
                        vlog_level, embed_model_data, model_output_dir,
                        input_nodes, output_nodes, input_shapes, output_shapes,
                        model_name, device_type, running_round, restart_round,
                        tuning, limit_opencl_kernel_time, phone_data_dir):
    generate_code(target_soc, target_abi, [], False)
    production_or_not = False
    mace_run_target = "//mace/tools/validation:mace_run"
    sh_commands.bazel_build(
            mace_run_target,
            abi=target_abi,
            model_tag=model_name,
            production_mode=False,
            hexagon_mode=hexagon_mode)
    sh_commands.update_mace_run_lib(model_output_dir, target_abi, model_name,
                                    embed_model_data)

    tuning_run(runtime, target_soc, target_abi, vlog_level, embed_model_data,
               model_output_dir, input_nodes, output_nodes, input_shapes,
               output_shapes, model_name, device_type, running_round=0,
               restart_round=1, out_of_range_check=True,
               phone_data_dir=phone_data_dir, tuning=False)

    tuning_run(runtime, target_soc, target_abi, vlog_level, embed_model_data,
               model_output_dir, input_nodes, output_nodes, input_shapes,
               output_shapes, model_name, device_type, running_round=0,
               restart_round=1, out_of_range_check=False,
               phone_data_dir=phone_data_dir, tuning=tuning,
               limit_opencl_kernel_time=limit_opencl_kernel_time)

    generate_code(target_soc, target_abi, [model_output_dir], True)
    production_or_not = True
    sh_commands.bazel_build(
            mace_run_target,
            abi=target_abi,
            model_tag=model_name,
            production_mode=True,
            hexagon_mode=hexagon_mode)
    sh_commands.update_mace_run_lib(model_output_dir, target_abi, model_name,
                                    embed_model_data)


def merge_libs_and_tuning_results(target_soc,
                                  target_abi,
                                  project_name,
                                  output_dir,
                                  model_output_dirs,
                                  hexagon_mode,
                                  embed_model_data):
    generate_code(target_soc, target_abi, model_output_dirs, False)
    sh_commands.build_production_code(target_abi)

    sh_commands.merge_libs(target_soc,
                           target_abi,
                           project_name,
                           output_dir,
                           model_output_dirs,
                           hexagon_mode,
                           embed_model_data)


def download_model_files(model_file_path,
                         model_output_dir,
                         weight_file_path=""):
    model_file = ""
    weight_file = ""
    if model_file_path.startswith("http://") or \
            model_file_path.startswith("https://"):
        model_file = model_output_dir + "/model.pb"
        urllib.urlretrieve(model_file_path, model_file)

    if weight_file_path.startswith("http://") or \
            weight_file_path.startswith("https://"):
        weight_file = model_output_dir + "/model.caffemodel"
        urllib.urlretrieve(weight_file_path, weight_file)

    return model_file, weight_file


def md5sum(str):
    md5 = hashlib.md5()
    md5.update(str)
    return md5.hexdigest()


def parse_model_configs():
    with open(FLAGS.config) as f:
        configs = yaml.load(f)
        return configs


def parse_args():
    """Parses command line arguments."""
    parser = argparse.ArgumentParser()
    parser.register("type", "bool", lambda v: v.lower() == "true")
    parser.add_argument(
        "--config",
        type=str,
        default="./tool/config",
        help="The global config file of models.")
    parser.add_argument(
        "--output_dir", type=str, default="build", help="The output dir.")
    parser.add_argument(
        "--round", type=int, default=1, help="The model running round.")
    parser.add_argument(
        "--run_seconds",
        type=int,
        default=10,
        help="The model throughput test running seconds.")
    parser.add_argument(
        "--restart_round",
        type=int,
        default=1,
        help="The model restart round.")
    parser.add_argument(
        "--tuning", type="bool", default="true", help="Tune opencl params.")
    parser.add_argument(
        "--mode",
        type=str,
        default="all",
        help="[build|run|validate|merge|all|throughput_test].")
    parser.add_argument(
        "--target_socs",
        type=str,
        default="all",
        help="SoCs to build, comma seperated list (getprop ro.board.platform)")
    parser.add_argument(
        "--out_of_range_check",
        type="bool",
        default="false",
        help="Enable out of range check for opencl.")
    return parser.parse_known_args()


def process_models(project_name, configs, embed_model_data, vlog_level,
                   target_soc, target_abi, phone_data_dir, option_args):
    hexagon_mode = get_hexagon_mode(configs)
    model_output_dirs = []
    for model_name in configs["models"]:
        print '===================', model_name, '==================='
        model_config = configs["models"][model_name]
        input_file_list = model_config.get("validation_inputs_data",
                                           [])
        data_type, device_type = get_data_and_device_type(
                model_config["runtime"])

        for key in ["input_nodes", "output_nodes", "input_shapes",
                    "output_shapes"]:
            if not isinstance(model_config[key], list):
                model_config[key] = [model_config[key]]

        # Create model build directory
        model_path_digest = md5sum(model_config["model_file_path"])
        model_output_dir = "%s/%s/%s/%s/%s/%s/%s" % (
            FLAGS.output_dir, project_name, "build",
            model_name, model_path_digest, target_soc, target_abi)
        model_output_dirs.append(model_output_dir)

        if FLAGS.mode == "build" or FLAGS.mode == "all":
            if os.path.exists(model_output_dir):
                sh.rm("-rf", model_output_dir)
            os.makedirs(model_output_dir)
            sh_commands.clear_mace_run_data(
                    target_abi, target_soc, phone_data_dir)

        model_file_path, weight_file_path = download_model_files(
                model_config["model_file_path"],
                model_output_dir,
                model_config.get("weight_file_path", ""))

        if FLAGS.mode == "build" or FLAGS.mode == "run" or \
                FLAGS.mode == "validate" or \
                FLAGS.mode == "benchmark" or FLAGS.mode == "all":
            sh_commands.gen_random_input(model_output_dir,
                                         model_config["input_nodes"],
                                         model_config["input_shapes"],
                                         input_file_list)

        if FLAGS.mode == "build" or FLAGS.mode == "all":
            sh_commands.gen_model_code(
                    "mace/codegen/models/%s" % model_name,
                    model_config["platform"],
                    model_file_path,
                    weight_file_path,
                    model_config["model_sha256_checksum"],
                    ",".join(model_config["input_nodes"]),
                    ",".join(model_config["output_nodes"]),
                    data_type,
                    model_config["runtime"],
                    model_name,
                    ":".join(model_config["input_shapes"]),
                    model_config["dsp_mode"],
                    embed_model_data,
                    model_config["fast_conv"],
                    model_config["obfuscate"])
            build_mace_run_prod(hexagon_mode,
                                model_config["runtime"],
                                target_soc,
                                target_abi,
                                vlog_level,
                                embed_model_data,
                                model_output_dir,
                                model_config["input_nodes"],
                                model_config["output_nodes"],
                                model_config["input_shapes"],
                                model_config["output_shapes"],
                                model_name,
                                device_type,
                                FLAGS.round,
                                FLAGS.restart_round,
                                FLAGS.tuning,
                                model_config["limit_opencl_kernel_time"],
                                phone_data_dir)

        if FLAGS.mode == "run" or FLAGS.mode == "validate" or \
                FLAGS.mode == "all":
            tuning_run(model_config["runtime"],
                       target_soc,
                       target_abi,
                       vlog_level,
                       embed_model_data,
                       model_output_dir,
                       model_config["input_nodes"],
                       model_config["output_nodes"],
                       model_config["input_shapes"],
                       model_config["output_shapes"],
                       model_name,
                       device_type,
                       FLAGS.round,
                       FLAGS.restart_round,
                       FLAGS.out_of_range_check,
                       phone_data_dir)

        if FLAGS.mode == "benchmark":
            sh_commands.benchmark_model(target_soc,
                                        target_abi,
                                        vlog_level,
                                        embed_model_data,
                                        model_output_dir,
                                        model_config["input_nodes"],
                                        model_config["output_nodes"],
                                        model_config["input_shapes"],
                                        model_config["output_shapes"],
                                        model_name,
                                        device_type,
                                        hexagon_mode,
                                        phone_data_dir,
                                        option_args)

        if FLAGS.mode == "validate" or FLAGS.mode == "all":
            sh_commands.validate_model(target_soc,
                                       target_abi,
                                       model_file_path,
                                       weight_file_path,
                                       model_config["platform"],
                                       model_config["runtime"],
                                       model_config["input_nodes"],
                                       model_config["output_nodes"],
                                       model_config["input_shapes"],
                                       model_config["output_shapes"],
                                       model_output_dir,
                                       phone_data_dir)

    if FLAGS.mode == "build" or FLAGS.mode == "merge" or \
            FLAGS.mode == "all":
        merge_libs_and_tuning_results(
            target_soc,
            target_abi,
            project_name,
            FLAGS.output_dir,
            model_output_dirs,
            hexagon_mode,
            embed_model_data)

    if FLAGS.mode == "throughput_test":
        merged_lib_file = FLAGS.output_dir + \
                "/%s/%s/libmace_%s.%s.a" % \
                (project_name, target_abi, project_name, target_soc)
        first_model = configs["models"].values()[0]
        throughput_test_output_dir = "%s/%s/%s/%s" % (
                FLAGS.output_dir, project_name, "build",
                "throughput_test")
        if os.path.exists(throughput_test_output_dir):
            sh.rm("-rf", throughput_test_output_dir)
        os.makedirs(throughput_test_output_dir)
        input_file_list = model_config.get("validation_inputs_data",
                                           [])
        sh_commands.gen_random_input(throughput_test_output_dir,
                                     first_model["input_nodes"],
                                     first_model["input_shapes"],
                                     input_file_list)
        model_tag_dict = {}
        for model_name in configs["models"]:
            runtime = configs["models"][model_name]["runtime"]
            model_tag_dict[runtime] = model_name
        sh_commands.build_run_throughput_test(target_soc,
                                              target_abi,
                                              vlog_level,
                                              FLAGS.run_seconds,
                                              merged_lib_file,
                                              throughput_test_output_dir,
                                              embed_model_data,
                                              model_config["input_nodes"],
                                              model_config["output_nodes"],
                                              model_config["input_shapes"],
                                              model_config["output_shapes"],
                                              model_tag_dict.get("cpu", ""),
                                              model_tag_dict.get("gpu", ""),
                                              model_tag_dict.get("dsp", ""),
                                              phone_data_dir)


def main(unused_args):
    configs = parse_model_configs()

    if FLAGS.mode == "validate":
        FLAGS.round = 1
        FLAGS.restart_round = 1

    project_name = os.path.splitext(os.path.basename(FLAGS.config))[0]
    if FLAGS.mode == "build" or FLAGS.mode == "all":
        # Remove previous output dirs
        if not os.path.exists(FLAGS.output_dir):
            os.makedirs(FLAGS.output_dir)
        elif os.path.exists(os.path.join(FLAGS.output_dir, "libmace")):
            sh.rm("-rf", os.path.join(FLAGS.output_dir, project_name))
            os.makedirs(os.path.join(FLAGS.output_dir, project_name))

        # generate source
        sh_commands.gen_mace_version()
        sh_commands.gen_encrypted_opencl_source()

    option_args = ' '.join(
        [arg for arg in unused_args if arg.startswith('--')])

    target_socs = get_target_socs(configs)

    embed_model_data = configs.get("embed_model_data", 1)
    vlog_level = configs.get("vlog_level", 0)
    phone_data_dir = "/data/local/tmp/mace_run/"
    for target_soc in target_socs:
        for target_abi in configs["target_abis"]:
            serialno = sh_commands.adb_devices([target_soc]).pop()
            props = sh_commands.adb_getprop_by_serialno(serialno)
            print(
                "============================================================="
            )
            print("Trying to lock device", serialno)
            with sh_commands.device_lock(serialno):
                print("Run on device: %s, %s, %s" % (
                      serialno, props["ro.board.platform"],
                      props["ro.product.model"]))
                process_models(project_name, configs, embed_model_data,
                               vlog_level, target_soc, target_abi,
                               phone_data_dir, option_args)

    if FLAGS.mode == "build" or FLAGS.mode == "all":
        sh_commands.packaging_lib(FLAGS.output_dir, project_name)


if __name__ == "__main__":
    FLAGS, unparsed = parse_args()
    main(unused_args=[sys.argv[0]] + unparsed)