profile usb gpu link stability (#38368)

* profile usb gpu link before using

* ci
This commit is contained in:
Daniel Koepping
2026-07-20 10:36:09 -07:00
committed by GitHub
parent d9596fa998
commit 24a9b6dae5
3 changed files with 41 additions and 0 deletions
@@ -13,6 +13,7 @@ from collections import namedtuple
import numpy as np
from openpilot.selfdrive.modeld.helpers import dump_oob, load_oob
from openpilot.selfdrive.modeld.usbgpu_link import wait_usbgpu_link
def _patch_tinygrad_fetch_fw():
import hashlib
@@ -311,6 +312,9 @@ if __name__ == "__main__":
p.add_argument('--frame-skip', type=int, required=True)
args = p.parse_args()
if 'USB+AMD' in os.environ.get('DEV', ''):
wait_usbgpu_link()
model_path = read_file_chunked_to_disk(args.onnx)
model_w, model_h = args.model_size
+3
View File
@@ -25,6 +25,7 @@ from openpilot.selfdrive.modeld.fill_model_msg import fill_model_msg, fill_drivi
from openpilot.common.file_chunker import open_file_chunked, get_manifest_path
from openpilot.selfdrive.modeld.constants import ModelConstants, Plan
from openpilot.selfdrive.modeld.helpers import usbgpu_present, modeld_pkl_path, get_tg_input_devices, load_oob
from openpilot.selfdrive.modeld.usbgpu_link import wait_usbgpu_link
PROCESS_NAME = "openpilot.selfdrive.modeld.modeld"
SEND_RAW_PRED = os.getenv('SEND_RAW_PRED')
@@ -168,6 +169,8 @@ def main(demo=False):
if use_extra_client:
cloudlog.warning(f"connected extra cam with buffer size: {vipc_client_extra.buffer_len} ({vipc_client_extra.width} x {vipc_client_extra.height})")
if USBGPU:
wait_usbgpu_link()
st = time.monotonic()
cloudlog.warning("loading model")
model = ModelState(vipc_client_main.width, vipc_client_main.height, USBGPU)
+34
View File
@@ -0,0 +1,34 @@
import time
from pathlib import Path
from openpilot.common.swaglog import cloudlog
from openpilot.common.hardware.usb import CHESTNUT_VENDOR_ID, CHESTNUT_PRODUCT_ID, usb_devices, controller, read_int
STABLE_SECONDS = 2.0
STABLE_THRESHOLD = 5.0 # link errors per second
def _chestnut_portli() -> Path | None:
for device in usb_devices():
if read_int(device / "idVendor", 16) == CHESTNUT_VENDOR_ID and \
read_int(device / "idProduct", 16) == CHESTNUT_PRODUCT_ID:
ctrl = controller(device)
if ctrl is not None and (ctrl / "portli").exists():
return ctrl / "portli"
return None
def wait_usbgpu_link(timeout: float = 30.0) -> None:
portli = _chestnut_portli()
if portli is None:
return
t0 = time.monotonic()
while time.monotonic() - t0 < timeout:
start = read_int(portli, 0)
time.sleep(STABLE_SECONDS)
rate = (read_int(portli, 0) - start) / STABLE_SECONDS
if rate <= STABLE_THRESHOLD:
return
cloudlog.warning(f"usbgpu link not stable: {rate:.0f} errors/s")
cloudlog.error("usbgpu link never stabilized")