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

am: setup_rinf returns only doorbell (#15112)

Browse Source
This commit is contained in:
nimlgen
2026-03-03 19:27:41 +03:00
committed by GitHub
parent 4b42bb54aa
commit abd830b260
2 changed files with 9 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
tinygrad/runtime/ops_amd.py
View File

@@ -858,12 +858,12 @@ class PCIIface(PCIIfaceBase):
rcvr_params: tuple
if queue_type == kfd.KFD_IOC_QUEUE_TYPE_SDMA:
pv, doorbell_index = self.dev_impl.sdma.setup_ring(*(rcvr_params:=(ring.va_addr, ring.size, gart.va_addr+rptr, gart.va_addr+wptr, idx)))
doorbell_index = self.dev_impl.sdma.setup_ring(*(rcvr_params:=(ring.va_addr, ring.size, gart.va_addr+rptr, gart.va_addr+wptr, idx)))
else:
pv, doorbell_index = self.dev_impl.gfx.setup_ring(*(rcvr_params:=(ring.va_addr, ring.size, gart.va_addr+rptr, gart.va_addr+wptr,
doorbell_index = self.dev_impl.gfx.setup_ring(*(rcvr_params:=(ring.va_addr, ring.size, gart.va_addr+rptr, gart.va_addr+wptr,
eop_buffer.va_addr, eop_buffer.size, is_aql:=(queue_type==kfd.KFD_IOC_QUEUE_TYPE_COMPUTE_AQL), is_aql)))
return AMDQueueDesc(ring=ring.cpu_view().view(fmt='I'), doorbell=self.dev_impl.doorbell64.view(doorbell_index * 8, 8, fmt='Q'), put_value=pv,
return AMDQueueDesc(ring=ring.cpu_view().view(fmt='I'), doorbell=self.dev_impl.doorbell64.view(doorbell_index * 8, 8, fmt='Q'), put_value=0,
read_ptr=gart.cpu_view().view(offset=rptr, size=8, fmt='Q'), write_ptr=gart.cpu_view().view(offset=wptr, size=8, fmt='Q'), params=rcvr_params)
def _collect_faults(self, reset=False):
@@ -871,8 +871,8 @@ class PCIIface(PCIIfaceBase):
for d in devs:
d.iface.dev_impl.ih.interrupt_handler()
if reset and d.iface.dev_impl.recover(force=d.error_state is not None):
d.compute_queue.put_value, _ = d.iface.dev_impl.gfx.setup_ring(*d.compute_queue.params)
d.compute_queue.read_ptr[0] = d.compute_queue.write_ptr[0] = d.compute_queue.put_value
d.compute_queue.put_value = d.compute_queue.read_ptr[0] = d.compute_queue.write_ptr[0] = 0
d.iface.dev_impl.gfx.setup_ring(*d.compute_queue.params)
d.timeline_signal.value = d.timeline_value - 1
d.error_state = None

8
tinygrad/runtime/support/am/ip.py
View File

@@ -300,7 +300,7 @@ class AM_GFX(AM_IP):
self._config_mec()
self._enable_mec()
def setup_ring(self, ring_addr:int, ring_size:int, rptr_addr:int, wptr_addr:int, eop_addr:int, eop_size:int, idx:int, aql:bool) -> tuple[int, int]:
def setup_ring(self, ring_addr:int, ring_size:int, rptr_addr:int, wptr_addr:int, eop_addr:int, eop_size:int, idx:int, aql:bool) -> int:
pipe, queue, doorbell = idx // 4, idx % 4, am.AMDGPU_NAVI10_DOORBELL_MEC_RING0
for xcc in range(self.xccs if aql else 1):
@@ -332,7 +332,7 @@ class AM_GFX(AM_IP):
self.adev.gmc.flush_hdp()
self._grbm_select(inst=xcc)
return 0, doorbell
return doorbell
def set_clockgating_state(self):
if hasattr(self.adev, 'regMM_ATC_L2_MISC_CG'): self.adev.regMM_ATC_L2_MISC_CG.write(enable=1, mem_ls_enable=1)
@@ -515,7 +515,7 @@ class AM_SDMA(AM_IP):
time.sleep(0.01)
self.adev.regGRBM_SOFT_RESET.write(0x0)
def setup_ring(self, ring_addr:int, ring_size:int, rptr_addr:int, wptr_addr:int, idx:int) -> tuple[int, int]:
def setup_ring(self, ring_addr:int, ring_size:int, rptr_addr:int, wptr_addr:int, idx:int) -> int:
pipe, queue = idx // 4, idx % 4
reg, inst = ("regSDMA_GFX", pipe+queue*4) if self.adev.ip_ver[am.SDMA0_HWIP][:2] == (4,4) else (f"regSDMA{pipe}_QUEUE{queue}", 0)
doorbell = am.AMDGPU_NAVI10_DOORBELL_sDMA_ENGINE0 + (pipe+queue*4) * 0xA
@@ -533,7 +533,7 @@ class AM_SDMA(AM_IP):
self.adev.reg(f"{reg}_RB_CNTL").write(**({f'{self.sdma_name.lower()}_wptr_poll_enable':1} if self.adev.ip_ver[am.SDMA0_HWIP][:2]!=(4,4) else {}),
rb_vmid=0, rptr_writeback_enable=1, rptr_writeback_timer=4, rb_enable=1, rb_priv=1, rb_size=(ring_size//4).bit_length()-1, inst=inst)
self.adev.reg(f"{reg}_IB_CNTL").update(ib_enable=1, inst=inst)
return 0, doorbell
return doorbell
class AM_PSP(AM_IP):
def init_sw(self):