modeld: split warp (#38079)

* compiles * runs * dedupe compiling model * always build for both res * fix does not bind loop variable * rm size multiplier
2026-06-08 11:25:51 +08:00 · 2026-05-22 17:50:26 -07:00
parent aff9f9ffae
commit a3cc9c7ac3
4 changed files with 59 additions and 72 deletions
--- a/2
+++ b/2
@@ -190,7 +190,7 @@ else:
 np_version = SCons.Script.Value(np.__version__)
 Export('envCython', 'np_version')

-Export('env', 'arch', 'acados', 'release')
+Export('env', 'arch', 'acados')

 # Setup cache dir
 cache_dir = '/data/scons_cache' if arch == "larch64" else '/tmp/scons_cache'
--- a/selfdrive/modeld/SConscript
+++ b/selfdrive/modeld/SConscript
@@ -7,24 +7,15 @@ from openpilot.common.file_chunker import chunk_file, get_chunk_targets, get_exi
 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.system.hardware import HARDWARE, PC
 from openpilot.selfdrive.modeld.helpers import TG_INPUT_DEVICES_PATH, usbgpu_present, modeld_pkl_path

-Import('env', 'arch', 'release')

+CAMERA_CONFIGS = [
+  (_ar_ox_fisheye.width, _ar_ox_fisheye.height),  # tici: 1928x1208
+  (_os_fisheye.width, _os_fisheye.height),        # mici: 1344x760
+]

-def get_camera_configs():
-  DEVICE_RESOLUTIONS = {
-    "tici": (_ar_ox_fisheye.width, _ar_ox_fisheye.height),
-    "tizi": (_ar_ox_fisheye.width, _ar_ox_fisheye.height),
-    "mici": (_os_fisheye.width, _os_fisheye.height),
-  }
-  if release or PC or 'CI' in os.environ:
-    return set(DEVICE_RESOLUTIONS.values())
-  return [DEVICE_RESOLUTIONS[HARDWARE.get_device_type()]]
-
-CAMERA_CONFIGS = get_camera_configs()
-
+Import('env', 'arch')
 chunker_file = File("#common/file_chunker.py")
 lenv = env.Clone()

@@ -101,8 +92,7 @@ for usbgpu in [False, True] if USBGPU else [False]:
        f'--policy-onnx {File(f"models/{file_prefix}driving_policy.onnx").abspath} '
        f'--output {target_pkl_path} --frame-skip {frame_skip}')
  onnx_sizes_sum = sum(os.path.getsize(f) for f in driving_onnx_deps)
-  size_multiplier = 1 if usbgpu else 2 # TODO make weight dedupe work on QCOM
-  chunk_targets = get_chunk_targets(target_pkl_path, estimate_pickle_max_size(onnx_sizes_sum)*size_multiplier)
+  chunk_targets = get_chunk_targets(target_pkl_path, estimate_pickle_max_size(onnx_sizes_sum))
  def do_chunk(target, source, env, pkl=target_pkl_path, chunks=chunk_targets):
    chunk_file(pkl, chunks)
  node = lenv.Command(
--- a/selfdrive/modeld/compile_modeld.py
+++ b/selfdrive/modeld/compile_modeld.py
@@ -30,11 +30,10 @@ from tinygrad.helpers import Context
 from tinygrad.device import Device
 from tinygrad.engine.jit import TinyJit

-from openpilot.common.file_chunker import read_file_chunked
-from openpilot.system.hardware.hw import Paths
-

 NV12Frame = namedtuple("NV12Frame", ['width', 'height', 'stride', 'y_height', 'uv_height', 'size'])
+WARP_INPUTS = ['img_q', 'big_img_q', 'tfm', 'big_tfm']
+POLICY_INPUTS = ['feat_q', 'desire_q', 'desire', 'traffic_convention']

 UV_SCALE_MATRIX = np.array([[0.5, 0, 0], [0, 0.5, 0], [0, 0, 1]], dtype=np.float32)
 UV_SCALE_MATRIX_INV = np.linalg.inv(UV_SCALE_MATRIX)
@@ -42,6 +41,10 @@ UV_SCALE_MATRIX_INV = np.linalg.inv(UV_SCALE_MATRIX)
 WARP_DEV = os.getenv('WARP_DEV')


+def make_random_images(keys, shape, device=None):
+  return {k: Tensor.randint(shape, low=0, high=256, dtype='uint8', device=device).realize() for k in keys}
+
+
 def warp_perspective_tinygrad(src_flat, M_inv, dst_shape, src_shape, stride_pad, border_fill_val=None):
  w_dst, h_dst = dst_shape
  h_src, w_src = src_shape
@@ -148,55 +151,49 @@ def sample_desire(buf, frame_skip):
  return buf.reshape(-1, frame_skip, *buf.shape[1:]).max(1).flatten(0, 1).unsqueeze(0)


-def make_run_policy(vision_runner, policy_runner, nv12: NV12Frame, model_w, model_h,
-                    vision_features_slice, frame_skip, prepare_only=False):
+def make_warp(nv12, model_w, model_h, frame_skip):
  frame_prepare = make_frame_prepare(nv12, model_w, model_h)
  sample_skip_fn = partial(sample_skip, frame_skip=frame_skip)
-  sample_desire_fn = partial(sample_desire, frame_skip=frame_skip)

-  def run_policy(img_q, big_img_q, feat_q, desire_q, desire, traffic_convention, tfm, big_tfm, frame, big_frame):
+  def warp_enqueue(img_q, big_img_q, tfm, big_tfm, frame, big_frame):
    tfm = tfm.to(WARP_DEV)
    big_tfm = big_tfm.to(WARP_DEV)
-    desire = desire.to(Device.DEFAULT)
-    traffic_convention = traffic_convention.to(Device.DEFAULT)
-    Tensor.realize(tfm, big_tfm, desire, traffic_convention)
+    Tensor.realize(tfm, big_tfm)

    warped_frame = frame_prepare(frame, tfm).unsqueeze(0).to(Device.DEFAULT)
    warped_big_frame = frame_prepare(big_frame, big_tfm).unsqueeze(0).to(Device.DEFAULT)
    img = shift_and_sample(img_q, warped_frame, sample_skip_fn)
    big_img = shift_and_sample(big_img_q, warped_big_frame, sample_skip_fn)
+    return img, big_img
+  return warp_enqueue

-    if prepare_only:
-      return img, big_img

+def make_run_policy(vision_runner, policy_runner, vision_features_slice, frame_skip):
+  sample_desire_fn = partial(sample_desire, frame_skip=frame_skip)
+  sample_skip_fn = partial(sample_skip, frame_skip=frame_skip)
+
+  def run_policy(img, big_img, feat_q, desire_q, desire, traffic_convention):
+    desire = desire.to(Device.DEFAULT)
+    traffic_convention = traffic_convention.to(Device.DEFAULT)
+    Tensor.realize(desire, traffic_convention)
+    desire_buf = shift_and_sample(desire_q, desire.reshape(1, 1, -1), sample_desire_fn)
    vision_out = next(iter(vision_runner({'img': img, 'big_img': big_img}).values())).cast('float32')

    new_feat = vision_out[:, vision_features_slice].reshape(1, -1).unsqueeze(0)
    feat_buf = shift_and_sample(feat_q, new_feat, sample_skip_fn)
-    desire_buf = shift_and_sample(desire_q, desire.reshape(1, 1, -1), sample_desire_fn)

    inputs = {'features_buffer': feat_buf, 'desire_pulse': desire_buf, 'traffic_convention': traffic_convention}
    policy_out = next(iter(policy_runner(inputs).values())).cast('float32')
-
    return vision_out, policy_out
  return run_policy


-def compile_modeld(nv12: NV12Frame, model_w, model_h, prepare_only, frame_skip,
-                   vision_runner, policy_runner, vision_metadata, policy_metadata):
-  print(f"Compiling combined policy JIT for {nv12.width}x{nv12.height} (prepare_only={prepare_only})...")
-
-  vision_features_slice = vision_metadata['output_slices']['hidden_state']
+def compile_jit(jit, make_random_inputs, input_keys, frame_skip, vision_metadata, policy_metadata):
  vision_input_shapes = vision_metadata['input_shapes']
  policy_input_shapes = policy_metadata['input_shapes']

-  _run = make_run_policy(vision_runner, policy_runner, nv12, model_w, model_h,
-                         vision_features_slice, frame_skip, prepare_only)
-  run_policy_jit = TinyJit(_run, prune=True)
-
  SEED = 42
-
-  def random_inputs_run_fn(fn, seed, test_val=None, test_buffers=None, expect_match=True):
+  def random_inputs_run(fn, seed, test_val=None, test_buffers=None, expect_match=True):
    input_queues, npy = make_input_queues(vision_input_shapes, policy_input_shapes, frame_skip, Device.DEFAULT)
    np.random.seed(seed)
    Tensor.manual_seed(seed)
@@ -205,13 +202,11 @@ def compile_modeld(nv12: NV12Frame, model_w, model_h, prepare_only, frame_skip,
    n_runs = 1 if testing else 3

    for i in range(n_runs):
-      frame = Tensor.randint(nv12.size, low=0, high=256, dtype='uint8', device=WARP_DEV).realize()
-      big_frame = Tensor.randint(nv12.size, low=0, high=256, dtype='uint8', device=WARP_DEV).realize()
      for v in npy.values():
        v[:] = np.random.randn(*v.shape).astype(v.dtype)
      Device.default.synchronize()
      st = time.perf_counter()
-      outs = fn(**input_queues, frame=frame, big_frame=big_frame)
+      outs = fn(**{k: input_queues[k] for k in input_keys}, **make_random_inputs())
      mt = time.perf_counter()
      Device.default.synchronize()
      et = time.perf_counter()
@@ -227,16 +222,15 @@ def compile_modeld(nv12: NV12Frame, model_w, model_h, prepare_only, frame_skip,
    if test_buffers is not None:
      match = all(np.array_equal(a, b) for a, b in zip(buffers, test_buffers, strict=True))
      assert match == expect_match, f"buffers {'differ from' if expect_match else 'match'} baseline (seed={seed})"
-    return fn, val, buffers
+    return val, buffers

  print('capture + replay')
-  run_policy_jit, test_val, test_buffers = random_inputs_run_fn(run_policy_jit, SEED)
-
+  test_val, test_buffers = random_inputs_run(jit, SEED)
  print('pickle round trip')
-  run_policy_jit = pickle.loads(pickle.dumps(run_policy_jit))
-  random_inputs_run_fn(run_policy_jit, SEED, test_val, test_buffers, expect_match=True)
-  random_inputs_run_fn(run_policy_jit, SEED+1, test_val, test_buffers, expect_match=False)
-  return run_policy_jit
+  jit = pickle.loads(pickle.dumps(jit))
+  random_inputs_run(jit, SEED, test_val, test_buffers, expect_match=True)
+  random_inputs_run(jit, SEED+1, test_val, test_buffers, expect_match=False)
+  return jit


 def _parse_size(s):
@@ -245,6 +239,8 @@ def _parse_size(s):


 def read_file_chunked_to_shm(path):
+  from openpilot.common.file_chunker import read_file_chunked
+  from openpilot.system.hardware.hw import Paths
  shm_path = os.path.join(Paths.shm_path(), os.path.basename(path))
  atexit.register(lambda: os.path.exists(shm_path) and os.remove(shm_path))
  with open(shm_path, 'wb') as f:
@@ -255,6 +251,7 @@ def read_file_chunked_to_shm(path):
 if __name__ == "__main__":
  from tinygrad.nn.onnx import OnnxRunner
  from openpilot.system.camerad.cameras.nv12_info import get_nv12_info
+  from openpilot.selfdrive.modeld.get_model_metadata import make_metadata_dict
  p = argparse.ArgumentParser()
  p.add_argument('--model-size', type=_parse_size, required=True, help='model input WxH')
  p.add_argument('--camera-resolutions', type=_parse_size, nargs='+', required=True,
@@ -266,23 +263,26 @@ if __name__ == "__main__":
  args = p.parse_args()

  out = defaultdict(dict)
-  # init runners once so weights are shared
-  from get_model_metadata import make_metadata_dict
  vision_path, policy_path = read_file_chunked_to_shm(args.vision_onnx), read_file_chunked_to_shm(args.policy_onnx)
+  model_w, model_h = args.model_size
+
  vision_runner = OnnxRunner(vision_path)
  policy_runner = OnnxRunner(policy_path)
-  out['metadata']['vision'] = make_metadata_dict(vision_path)
-  out['metadata']['policy'] = make_metadata_dict(policy_path)
+  vision_metadata, policy_metadata = make_metadata_dict(vision_path), make_metadata_dict(policy_path)
+
+  run_policy_jit = TinyJit(make_run_policy(vision_runner, policy_runner, vision_metadata['output_slices']['hidden_state'], args.frame_skip), prune=True)
+
+  out['metadata']['vision'], out['metadata']['policy'] = vision_metadata, policy_metadata
+
+  make_random_model_inputs = partial(make_random_images, keys=['img', 'big_img'], shape=vision_metadata['input_shapes']['img'])
+  out['run_policy'] = compile_jit(run_policy_jit, make_random_model_inputs, POLICY_INPUTS, args.frame_skip, vision_metadata, policy_metadata)

  for cam_w, cam_h in args.camera_resolutions:
    nv12 = NV12Frame(cam_w, cam_h, *get_nv12_info(cam_w, cam_h))
-    model_w, model_h = args.model_size
-    out[(cam_w,cam_h)] = {
-      name: compile_modeld(nv12, model_w, model_h, prepare_only, args.frame_skip,
-                           vision_runner, policy_runner, out['metadata']['vision'], out['metadata']['policy'])
-      for name, prepare_only in [('warp_enqueue', True), ('run_policy', False)]
-    }
+    make_random_warp_inputs = partial(make_random_images, keys=['frame', 'big_frame'], shape=nv12.size, device=WARP_DEV)
+    warp_enqueue = TinyJit(make_warp(nv12, model_w, model_h, args.frame_skip), prune=True)
+    out[(cam_w,cam_h)] = compile_jit(warp_enqueue, make_random_warp_inputs, WARP_INPUTS, args.frame_skip, vision_metadata, policy_metadata)

  with open(args.output, "wb") as f:
    pickle.dump(out, f)
-  print(f"Saved combined JIT to {args.output} ({os.path.getsize(args.output) / 1e6:.2f} MB)")
+  print(f"Saved JITs to {args.output} ({os.path.getsize(args.output) / 1e6:.2f} MB)")
--- a/selfdrive/modeld/modeld.py
+++ b/selfdrive/modeld/modeld.py
@@ -20,7 +20,7 @@ from openpilot.common.transformations.model import get_warp_matrix
 from openpilot.selfdrive.controls.lib.desire_helper import DesireHelper
 from openpilot.selfdrive.controls.lib.drive_helpers import get_accel_from_plan, smooth_value, get_curvature_from_plan
 from openpilot.selfdrive.modeld.parse_model_outputs import Parser
-from openpilot.selfdrive.modeld.compile_modeld import make_input_queues
+from openpilot.selfdrive.modeld.compile_modeld import make_input_queues, WARP_INPUTS, POLICY_INPUTS
 from openpilot.selfdrive.modeld.fill_model_msg import fill_model_msg, fill_pose_msg, PublishState
 from openpilot.common.file_chunker import read_file_chunked, get_manifest_path
 from openpilot.selfdrive.modeld.constants import ModelConstants, Plan
@@ -93,12 +93,8 @@ class ModelState:
    self._blob_cache : dict[int, Tensor] = {}
    self.parser = Parser()
    self.frame_buf_params = {k: get_nv12_info(cam_w, cam_h) for k in ('img', 'big_img')}
-    self.run_policy = jits[(cam_w,cam_h)]['run_policy']
-    self.warp_enqueue = jits[(cam_w,cam_h)]['warp_enqueue']
-    self.warp_enqueue(
-      **self.input_queues,
-      frame=Tensor(np.zeros(self.frame_buf_params['img'][3], dtype=np.uint8), device=self.WARP_DEV).contiguous().realize(),
-      big_frame=Tensor(np.zeros(self.frame_buf_params['big_img'][3], dtype=np.uint8), device=self.WARP_DEV).contiguous().realize())
+    self.run_policy = jits['run_policy']
+    self.warp_enqueue = jits[(cam_w,cam_h)]

  def slice_outputs(self, model_outputs: np.ndarray, output_slices: dict[str, slice]) -> dict[str, np.ndarray]:
    parsed_model_outputs = {k: model_outputs[np.newaxis, v] for k,v in output_slices.items()}
@@ -123,12 +119,13 @@ class ModelState:
    self.npy['tfm'][:,:] = transforms['img'][:,:]
    self.npy['big_tfm'][:,:] = transforms['big_img'][:,:]

+    img, big_img = self.warp_enqueue(**{k: self.input_queues[k] for k in WARP_INPUTS}, frame=self.full_frames['img'], big_frame=self.full_frames['big_img'])
+
    if prepare_only:
-      self.warp_enqueue(**self.input_queues, frame=self.full_frames['img'], big_frame=self.full_frames['big_img'])
      return None

    vision_output, policy_output = self.run_policy(
-      **self.input_queues, frame=self.full_frames['img'], big_frame=self.full_frames['big_img']
+      **{k: self.input_queues[k] for k in POLICY_INPUTS}, img=img, big_img=big_img
    )

    vision_output = vision_output.numpy().flatten()