import sys 
import os
import yaml
import argparse
"""
{'CPU_UTILIZATION': 0.8, 'MAX_GPU_MEMORY': 0, 'GPU_UTILIZATION': '0 %', 'DAG': {'50': 670.256, '60': 670.256, '70': 670.765, '80': 671.23, '90': 687.546, '95': 687.546, '99': 687.546, 'avg': 670.755625, 'qps': 0.8, 'query_count': 8, 'succ': 1.0}, 'demo': {'midp': 669.484375, 'postp': 0.184875, 'prep': 1.001875}}
"""
class LogHandler(object):
    def __init__(self):
        self.fstr = ""

    def print(self):
        print(self.fstr)

    def dump(self, filename):
        with open(filename,'w') as fout:
            fout.write(self.fstr)

    def append(self, new_str):
        self.fstr += new_str + "\n"

def parse_args():  # pylint: disable=doc-string-missing
    parser = argparse.ArgumentParser("serve")
    parser.add_argument(
        "--benchmark_cfg", type=str, required=True, help="benchmark config yaml. including general info, model info, data info, conf info")
    parser.add_argument(
        "--benchmark_log",
        type=str, required=True,
        help="benchmark log, generated by a web service or pipeline.")
    parser.add_argument(
        "--output",
        type=str,
        default="std_benchmark.log",
        help="the output filename, default std_benchmark.log")
    return parser.parse_args()

if __name__ == "__main__":
    args = parse_args()
    benchmark_cfg_filename = args.benchmark_cfg
    f = open(benchmark_cfg_filename, 'r')
    config = yaml.load(f)
    f.close()
    benchmark_raw_filename = args.benchmark_log
    f = open(benchmark_raw_filename, 'r')
    benchmark_raw = yaml.load(f)
    f.close()
    ## general info
    cuda_version = config["cuda_version"]
    cudnn_version = config["cudnn_version"]
    trt_version = config["cudnn_version"]
    python_version = config["python_version"]
    gcc_version = config["gcc_version"]
    paddle_version = config["paddle_version"]
    cpu = config["cpu"]
    gpu = config["gpu"]
    xpu = config["xpu"]
    api = config["api"]
    owner = config["owner"]
    ## model info
    model_name = config["model_name"]
    model_type = config["model_type"]
    model_source = config["model_source"]
    model_url = config["model_url"]
    ## data info
    batch_size = config["batch_size"]
    num_of_samples = config["num_of_samples"]
    input_shape = config["input_shape"]
    ## conf info
    runtime_device = config["runtime_device"]
    ir_optim = config["ir_optim"]
    enable_memory_optim = config["enable_memory_optim"]
    enable_tensorrt = config["enable_tensorrt"]
    precision = config["precision"]
    enable_mkldnn = config["enable_mkldnn"]
    cpu_math_library_num_threads = config["cpu_math_library_num_threads"]
    ## acc info 
    acc1 = "Nan"
    acc5 = "Nan"
    ## perf info
    average_latency, QPS = benchmark_raw["DAG"]["avg"], benchmark_raw["DAG"]["qps"]
    cost_90, cost_99, succ_rate = benchmark_raw["DAG"]["90"], benchmark_raw["DAG"]["99"], benchmark_raw["DAG"]["succ"]
    process_latency = ""
    cpu_rss, vms, shared, dirty, cpu_usage = "", "", "", "",  benchmark_raw["CPU_MEM"]
    gpu_id, total, free, used, gpu_utilization_rate, gpu_mem_utilization_rate = "","","","", benchmark_raw["GPU_UTIL"], benchmark_raw["GPU_MEM"]

    fh = LogHandler()

    fh.append("cuda_version: {}".format(cuda_version))
    fh.append("cudnn_version: {}".format(cudnn_version))
    fh.append("trt_version: {} ".format(trt_version))
    fh.append("python_version: {}".format(python_version))
    fh.append("gcc_version: {}".format(gcc_version))
    fh.append("paddle_version: {}".format(paddle_version))
    fh.append("cpu: {}".format(cpu))
    fh.append("gpu: {}".format(gpu)) # p4, v100, 1080
    fh.append("xpu: {}".format(xpu)) 
    fh.append("api: {}".format(api))
    fh.append("owner: {}".format(owner))
    fh.append("----------------------- Model info ----------------------")
    fh.append("model_name: {}".format(model_name))
    fh.append("model_type: {}".format(model_type))
    fh.append("model_source: {}".format(model_source))
    fh.append("model_url: {}".format(model_url))
    fh.append("----------------------- Data info -----------------------")
    fh.append("batch_size: {}".format(batch_size))
    fh.append("num_of_samples: {}".format(num_of_samples))
    fh.append("input_shape: {}".format(input_shape))
    fh.append("----------------------- Conf info -----------------------")
    fh.append("runtime_device: {}".format(runtime_device))
    fh.append("ir_optim: {}".format(ir_optim))
    fh.append("enable_memory_optim: {}".format(enable_memory_optim))
    fh.append("enable_tensorrt: {}".format(enable_tensorrt))
    fh.append("precision: {}".format(precision))    # fp32, fp16, int8
    fh.append("enable_mkldnn: {}".format(enable_mkldnn))
    fh.append("cpu_math_library_num_threads: {}".format(cpu_math_library_num_threads))
    fh.append("----------------------- Acc info ------------------------")
    fh.append("acc1:".format(acc1))
    fh.append("acc5:".format(acc5))
    fh.append("----------------------- Perf info -----------------------")
    fh.append("average_latency(ms): {}, QPS: {}".format(average_latency, QPS))
    fh.append("process_latency(ms): {}".format(process_latency))
    fh.append("90%_cost: {}, 99%_cost: {}, succ_rate: {}".format(cost_90, cost_99, succ_rate))
    fh.append("process_name: clas_benchmark, cpu_rss(MB): {}, vms(MB): {}, shared(MB): {}, dirty(MB): {}, cpu_usage(%): {}".format(cpu_rss, vms, shared, dirty, cpu_usage))
    fh.append("gpu_id: {}, total(MB): {}, free(MB): {}, used(MB): {}, gpu_utilization_rate(%): {}, gpu_mem_utilization_rate(%): {}".format(gpu_id, total, free, used, gpu_utilization_rate, gpu_mem_utilization_rate))

    fh.dump(args.output)