import glob
import json
import os
from itertools import product
from SCons.Script import Value
from openpilot.common.file_chunker import chunk_file, get_chunk_paths
from openpilot.common.transformations.camera import _ar_ox_fisheye, _os_fisheye
from openpilot.common.transformations.model import MEDMODEL_INPUT_SIZE, DM_INPUT_SIZE
from openpilot.selfdrive.modeld.constants import ModelConstants
from openpilot.selfdrive.modeld.helpers import CompileConfig
from tinygrad import Device

CAMERA_CONFIGS = [
  (_ar_ox_fisheye.width, _ar_ox_fisheye.height),  # tici: 1928x1208
  (_os_fisheye.width, _os_fisheye.height),        # mici: 1344x760
]
MODELD_CONFIGS = [CompileConfig(cam_w, cam_h, prepare_only, 'driving_')
                  for (cam_w, cam_h), prepare_only in product(CAMERA_CONFIGS, [True, False])]
DM_WARP_CONFIGS = [CompileConfig(cam_w, cam_h, True, 'dm_') for cam_w, cam_h in CAMERA_CONFIGS]

Import('env', 'arch')
chunker_file = File("#common/file_chunker.py")
lenv = env.Clone()

tinygrad_root = env.Dir("#").abspath
tinygrad_files = ["#"+x for x in glob.glob(env.Dir("#tinygrad_repo").relpath + "/**", recursive=True, root_dir=tinygrad_root)
                  if 'pycache' not in x and os.path.isfile(os.path.join(tinygrad_root, x))]

def estimate_pickle_max_size(onnx_size):
  return 1.2 * onnx_size  + 10 * 1024 * 1024  # 20% + 10MB is plenty

# get fastest TG config
available = set(Device.get_available_devices())
if 'CUDA' in available:
  tg_backend = 'CUDA'
  tg_flags = f'DEV={tg_backend}'
elif 'QCOM' in available:
  tg_backend = 'QCOM'
  tg_flags = f'DEV={tg_backend} FLOAT16=1 NOLOCALS=1 JIT_BATCH_SIZE=0 OPENPILOT_HACKS=1'
else:
  tg_backend = 'CPU' if arch == 'Darwin' else 'CPU:LLVM'
  # THREADS=0 is need to prevent bug: https://github.com/tinygrad/tinygrad/issues/14689
  tg_flags = f'DEV={tg_backend} THREADS=0'

def write_tg_compiled_flags(target, source, env):
  with open(str(target[0]), "w") as f:
    json.dump({"DEV": tg_backend}, f)
    f.write("\n")

compiled_flags_node = lenv.Command(
  File("models/tg_compiled_flags.json").abspath,
  tinygrad_files + [Value(tg_backend)],
  write_tg_compiled_flags,
)

# tinygrad calls brew which needs a $HOME in the env
mac_brew_string = f'HOME={os.path.expanduser("~")}' if arch == 'Darwin' else ''

# Get model metadata
for model_name in ['driving_vision', 'driving_policy', 'dmonitoring_model']:
  fn = File(f"models/{model_name}").abspath
  script_files = [File(Dir("#selfdrive/modeld").File("get_model_metadata.py").abspath)]
  cmd = f'{tg_flags} {mac_brew_string} python3 {Dir("#selfdrive/modeld").abspath}/get_model_metadata.py {fn}.onnx'
  lenv.Command(fn + "_metadata.pkl", [fn + ".onnx"] + tinygrad_files + script_files + [compiled_flags_node], cmd)

image_flag = {
     'larch64': 'IMAGE=2',
}.get(arch, 'IMAGE=0')
modeld_dir = Dir("#selfdrive/modeld").abspath
compile_modeld_script = [File(f"{modeld_dir}/compile_modeld.py")]
compile_dm_warp_script = [File(f"{modeld_dir}/compile_dm_warp.py")]
driving_onnx_deps = [File(f"models/{m}.onnx").abspath for m in ['driving_vision', 'driving_policy']]
driving_metadata_deps = [File(f"models/{m}_metadata.pkl").abspath for m in ['driving_vision', 'driving_policy']]

model_w, model_h = MEDMODEL_INPUT_SIZE
frame_skip = ModelConstants.MODEL_RUN_FREQ // ModelConstants.MODEL_CONTEXT_FREQ
for cfg in MODELD_CONFIGS:
  cmd = (f'{tg_flags} {mac_brew_string} {image_flag} python3 {modeld_dir}/compile_modeld.py '
         f'--model-size {model_w}x{model_h} '
         f'--nv12 {",".join(str(x) for x in cfg.nv12)} '
         f'--vision-onnx {File("models/driving_vision.onnx").abspath} '
         f'--policy-onnx {File("models/driving_policy.onnx").abspath} '
         f'--output {cfg.pkl_path} --frame-skip {frame_skip}'
         + (' --prepare-only' if cfg.prepare_only else ''))
  node = lenv.Command(cfg.pkl_path, tinygrad_files + compile_modeld_script + driving_onnx_deps + driving_metadata_deps + [chunker_file, compiled_flags_node], cmd)
  onnx_sizes_sum = sum(os.path.getsize(f) for f in driving_onnx_deps)
  chunk_targets = get_chunk_paths(cfg.pkl_path, estimate_pickle_max_size(onnx_sizes_sum))
  def do_chunk(target, source, env, pkl=cfg.pkl_path, chunks=chunk_targets):
    chunk_file(pkl, chunks)
  lenv.Command(chunk_targets, node, do_chunk)

dm_w, dm_h = DM_INPUT_SIZE
for cfg in DM_WARP_CONFIGS:
  cmd = (f'{tg_flags} {mac_brew_string} {image_flag} python3 {modeld_dir}/compile_dm_warp.py '
         f'--nv12 {",".join(str(x) for x in cfg.nv12)} --warp-to {dm_w}x{dm_h} '
         f'--output {cfg.pkl_path}')
  lenv.Command(cfg.pkl_path, tinygrad_files + compile_dm_warp_script + compile_modeld_script + [compiled_flags_node], cmd)

def tg_compile(flags, model_name):
  pythonpath_string = 'PYTHONPATH="${PYTHONPATH}:' + env.Dir("#tinygrad_repo").abspath + '"'
  fn = File(f"models/{model_name}").abspath
  pkl = fn + "_tinygrad.pkl"
  onnx_path = fn + ".onnx"
  chunk_targets = get_chunk_paths(pkl, estimate_pickle_max_size(os.path.getsize(onnx_path)))
  compile_node = lenv.Command(
    pkl,
    [onnx_path] + tinygrad_files + [chunker_file, compiled_flags_node],
    f'{pythonpath_string} {flags} {image_flag} python3 {Dir("#tinygrad_repo").abspath}/examples/openpilot/compile3.py {fn}.onnx {pkl}',
  )
  def do_chunk(target, source, env):
    chunk_file(pkl, chunk_targets)
  return lenv.Command(
    chunk_targets,
    compile_node,
    do_chunk,
  )

tg_compile(tg_flags, 'dmonitoring_model')
