IQ.Pilot Release Commit @ 0798119

tinygrad_repo/extra/amdpci/hive_reset.py

@@ -0,0 +1,20 @@
+#!/usr/bin/env python3
+
+import os
+from tinygrad.helpers import Context
+from tinygrad.runtime.support.system import System, PCIDevice
+from tinygrad.runtime.support.hcq import FileIOInterface
+from tinygrad.runtime.support.am.amdev import AMDev
+
+if __name__ == "__main__":
+  gpus = System.pci_scan_bus(0x1002, [(0xffff, [0x74a1, 0x75a0])])
+  for gpu in gpus:
+    drv_path = f"/sys/bus/pci/devices/{gpu}/driver"
+    if FileIOInterface.exists(drv_path) and os.path.basename(os.readlink(drv_path)) == "amdgpu":
+      raise RuntimeError(f"amdgpu is bound to {gpu}. Stopping...")
+  pcidevs = [PCIDevice("AM", gpu) for gpu in gpus]
+  amdevs = []
+  with Context(DEBUG=2):
+    for pcidev in pcidevs:
+      amdevs.append(AMDev(pcidev, reset_mode=True))
+    for amdev in amdevs: amdev.smu.mode1_reset()