sirhcm/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests 2 Actions 8 Activity

amd: fix sigs on xcc path (#10137)

Browse Source
This commit is contained in:
nimlgen
2025-05-02 13:50:56 +03:00
committed by GitHub
parent bdd4dd9238
commit 6a845c2de2
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tinygrad/runtime/ops_amd.py
+2 -2
View File
@@ -902,8 +902,8 @@ class AMDDevice(HCQCompiled):
# XCC setup
self.xcc_sync: tuple[AMDSignal, AMDSignal]|None = None
if self.xccs > 1:
self.xcc_sync_area = self.allocator.alloc(0x1000, BufferSpec(nolru=True))
self.xcc_sync = (AMDSignal(base_addr=self.xcc_sync_area.va_addr), AMDSignal(base_addr=self.xcc_sync_area.va_addr + 256))
self.xcc_sync_area = self.allocator.alloc(0x1000, BufferSpec(nolru=True, cpu_access=True))
self.xcc_sync = (AMDSignal(base_buf=self.xcc_sync_area), AMDSignal(base_buf=self.xcc_sync_area.offset(256)))
AMDComputeQueue(self).xcc_config().submit(self)
# SQTT is disabled by default because of runtime overhead and big file sizes (~200mb to Tensor.full() two 4096x4096 tensors and matmul them)