github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-04-29 03:00:14 -04:00
Code Issues Packages Projects Releases Wiki Activity

Fix double SDMA_OP_FENCE (#9705)

Browse Source 
Introduced in #9585, probably when i incorrectly resolved merge conflict
while rebasing an old, mi300x-only branch. Seems to be the source of
multi gpu beam llama hangs
This commit is contained in:
uuuvn
2025-04-03 06:43:37 +05:00
committed by GitHub
parent a6fec2f5ae
commit 5bd485c027
1 changed files with 0 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
tinygrad/runtime/ops_amd.py
View File

@@ -339,7 +339,6 @@ class AMDCopyQueue(HWQueue):
def signal(self, signal:AMDSignal, value:sint=0):
fence_flags = self.sdma.SDMA_PKT_FENCE_HEADER_MTYPE(3) if self.dev.gfxver >= 10 else 0
self.q(self.sdma.SDMA_OP_FENCE | fence_flags, *data64_le(signal.value_addr), value)
self.q(self.sdma.SDMA_OP_FENCE, *data64_le(signal.value_addr), value)
if not AMDDevice.driverless and (dev:=signal.timeline_for_device) is not None:
self.q(self.sdma.SDMA_OP_FENCE | fence_flags, *data64_le(dev.queue_event_mailbox_ptr), dev.queue_event.event_id)