Files
markbench-tests/procyon_ai/ulprocai.py
j-lin-lmg b229c4f027 Ffmpeg cpu benchmark (#179)
Added ffmpeg cpu benchmark based on @nharris-lmg commands

3 options for encode (h264, av1, h265)

will report final vmaf score as "score" in report json

logs saved to artifacts
2025-12-23 16:20:51 -08:00

257 lines
8.4 KiB
Python

"""UL Procyon Computer Vision test script"""
# pylint: disable=no-name-in-module
import logging
import subprocess
import sys
import time
from argparse import ArgumentParser
from pathlib import Path
import psutil
from utils import (
find_procyon_version,
find_score_in_xml,
find_test_version,
get_install_path,
is_process_running,
)
PARENT_DIR = str(Path(sys.path[0], ".."))
sys.path.append(PARENT_DIR)
from harness_utils.artifacts import ArtifactManager, ArtifactType
from harness_utils.output import (
DEFAULT_DATE_FORMAT,
DEFAULT_LOGGING_FORMAT,
seconds_to_milliseconds,
setup_log_directory,
write_report_json,
)
from harness_utils.procyoncmd import (
get_cuda_devices,
get_openvino_devices,
get_openvino_gpu,
get_winml_devices,
)
#####
# Globals
#####
SCRIPT_DIR = Path(__file__).resolve().parent
LOG_DIR = SCRIPT_DIR / "run"
DIR_PROCYON = Path(get_install_path())
EXECUTABLE = "ProcyonCmd.exe"
ABS_EXECUTABLE_PATH = DIR_PROCYON / EXECUTABLE
WINML_DEVICES = get_winml_devices(ABS_EXECUTABLE_PATH)
OPENVINO_DEVICES = get_openvino_devices(ABS_EXECUTABLE_PATH)
CUDA_DEVICES = get_cuda_devices(ABS_EXECUTABLE_PATH)
CONFIG_DIR = SCRIPT_DIR / "config"
BENCHMARK_CONFIG = {
"AMD_CPU": {
"config": f'"{CONFIG_DIR}\\ai_computer_vision_winml_cpu.def"',
"process_name": "WinML.exe",
"device_name": "CPU",
# TODO: Find a good way to report the CPU name here.
"device_id": "CPU",
"test_name": "cpu_float32",
"api": "winml",
},
"AMD_GPU0": {
"config": f'"{CONFIG_DIR}\\ai_computer_vision_winml_gpu.def"',
"process_name": "WinML.exe",
"device_name": list(WINML_DEVICES.keys())[0],
"device_id": list(WINML_DEVICES.values())[0],
"test_name": "gpu_float32",
"api": "winml",
},
"AMD_GPU1": {
"config": f'"{CONFIG_DIR}\\ai_computer_vision_winml_gpu.def"',
"process_name": "WinML.exe",
"device_name": list(WINML_DEVICES.keys())[1]
if len(list(WINML_DEVICES.keys())) > 1
else list(WINML_DEVICES.keys())[0],
"device_id": list(WINML_DEVICES.values())[1]
if len(list(WINML_DEVICES.values())) > 1
else list(WINML_DEVICES.values())[0],
"test_name": "gpu_float32",
"api": "winml",
},
"Intel_CPU": {
"config": f'"{CONFIG_DIR}\\ai_computer_vision_openvino_cpu.def"',
"process_name": "OpenVino.exe",
"device_id": "CPU",
"device_name": OPENVINO_DEVICES["CPU"],
"test_name": "cpu_float32",
"api": "openvino",
},
"Intel_GPU0": {
"config": f'"{CONFIG_DIR}\\ai_computer_vision_openvino_gpu.def"',
"process_name": "OpenVino.exe",
"device_id": "GPU.0" if "GPU.0" in list(OPENVINO_DEVICES.keys()) else "GPU",
"device_name": get_openvino_gpu(OPENVINO_DEVICES, "GPU.0"),
"test_name": "gpu_float32",
"api": "openvino",
},
"Intel_GPU1": {
"config": f'"{CONFIG_DIR}\\ai_computer_vision_openvino_gpu.def"',
"process_name": "OpenVino.exe",
"device_id": "GPU.1" if "GPU.1" in list(OPENVINO_DEVICES.keys()) else "GPU",
"device_name": get_openvino_gpu(OPENVINO_DEVICES, "GPU.0"),
"test_name": "gpu_float32",
"api": "openvino",
},
"Intel_NPU": {
"config": f'"{CONFIG_DIR}\\ai_computer_vision_openvino_npu.def"',
"process_name": "OpenVino.exe",
"device_id": "NPU",
"device_name": OPENVINO_DEVICES.get("NPU", "None"),
"test_name": "npu_float32",
"api": "openvino",
},
"NVIDIA_GPU": {
"config": f'"{CONFIG_DIR}\\ai_computer_vision_tensorrt.def"',
"device_id": "cuda:0",
"device_name": CUDA_DEVICES.get("cuda:0"),
"process_name": "TensorRT.exe",
"test_name": "gpu_float32",
"api": "tensorrt",
},
"Qualcomm_HTP": {
"config": f'"{CONFIG_DIR}\\ai_computer_vision_snpe.def"',
"device_id": "CPU",
"device_name": "CPU",
"process_name": "SNPE.exe",
"test_name": "htp_integer",
"api": "snpe",
},
}
RESULTS_FILENAME = "result.xml"
RESULTS_XML_PATH = LOG_DIR / RESULTS_FILENAME
def setup_logging():
"""setup logging"""
setup_log_directory(str(LOG_DIR))
logging.basicConfig(
filename=LOG_DIR / "harness.log",
format=DEFAULT_LOGGING_FORMAT,
datefmt=DEFAULT_DATE_FORMAT,
level=logging.DEBUG,
)
console = logging.StreamHandler()
formatter = logging.Formatter(DEFAULT_LOGGING_FORMAT)
console.setFormatter(formatter)
logging.getLogger("").addHandler(console)
def get_arguments():
"""get arguments"""
parser = ArgumentParser()
parser.add_argument(
"--engine",
dest="engine",
help="Engine test type",
required=True,
choices=BENCHMARK_CONFIG.keys(),
)
argies = parser.parse_args()
return argies
def create_procyon_command(test_option, process_name, device_id):
"""create command string"""
command = str()
if device_id == "CPU":
command = f'"{ABS_EXECUTABLE_PATH}" --definition={test_option} --export="{RESULTS_XML_PATH}"'
else:
match process_name:
case "WinML.exe":
command = f'"{ABS_EXECUTABLE_PATH}" --definition={test_option} --export="{RESULTS_XML_PATH}" --select-winml-device {device_id}'
case "OpenVino.exe":
command = f'"{ABS_EXECUTABLE_PATH}" --definition={test_option} --export="{RESULTS_XML_PATH}" --select-openvino-device {device_id}'
case "TensorRT.exe":
command = f'"{ABS_EXECUTABLE_PATH}" --definition={test_option} --export="{RESULTS_XML_PATH}" --select-cuda-device {device_id}'
command = command.rstrip()
return command
def run_benchmark(process_name, command_to_run):
"""run the benchmark"""
with subprocess.Popen(
command_to_run,
stdout=subprocess.PIPE,
stderr=subprocess.STDOUT,
universal_newlines=True,
) as proc:
logging.info("Procyon AI Computer Vision benchmark has started.")
while True:
now = time.time()
elapsed = now - start_time
if elapsed >= 60: # seconds
raise ValueError("BenchMark subprocess did not start in time")
process = is_process_running(process_name)
if process is not None:
process.nice(psutil.HIGH_PRIORITY_CLASS)
break
time.sleep(0.2)
_, _ = proc.communicate() # blocks until 3dmark exits
return proc
try:
setup_logging()
logging.info("Detected Windows ML Devices: %s", str(WINML_DEVICES))
logging.info("Detected OpenVino Devices: %s", str(OPENVINO_DEVICES))
logging.info("Detected CUDA Devices: %s", (CUDA_DEVICES))
args = get_arguments()
option = BENCHMARK_CONFIG[args.engine]["config"]
proc_name = BENCHMARK_CONFIG[args.engine]["process_name"]
dev_id = BENCHMARK_CONFIG[args.engine]["device_id"]
cmd = create_procyon_command(option, proc_name, dev_id)
logging.info("Starting benchmark!")
logging.info(cmd)
start_time = time.time()
pr = run_benchmark(BENCHMARK_CONFIG[args.engine]["process_name"], cmd)
if pr.returncode > 0:
logging.error("Procyon exited with return code %d", pr.returncode)
sys.exit(pr.returncode)
score = find_score_in_xml()
if score is None:
logging.error("Could not find overall score!")
sys.exit(1)
am = ArtifactManager(LOG_DIR)
am.copy_file(RESULTS_XML_PATH, ArtifactType.RESULTS_TEXT, "results xml file")
am.create_manifest()
end_time = time.time()
elapsed_test_time = round(end_time - start_time, 2)
logging.info("Benchmark took %.2f seconds", elapsed_test_time)
logging.info("Score was %s", score)
report = {
"start_time": seconds_to_milliseconds(start_time),
"end_time": seconds_to_milliseconds(end_time),
"test": "Procyon AI CV",
"test_parameter": BENCHMARK_CONFIG[args.engine]["test_name"],
"api": BENCHMARK_CONFIG[args.engine]["api"],
"test_version": find_test_version(),
"device_name": BENCHMARK_CONFIG[args.engine]["device_name"],
"procyon_version": find_procyon_version(),
"unit": "score",
"score": score,
}
write_report_json(str(LOG_DIR), "report.json", report)
except Exception as e:
logging.error("Something went wrong running the benchmark!")
logging.exception(e)
sys.exit(1)