mirror of
https://github.com/tinygrad/tinygrad.git
synced 2026-09-05 21:26:15 +00:00
am: remove doorbells from setup_ring (#13946)
This commit is contained in:
@@ -834,11 +834,11 @@ class PCIIface(PCIIfaceBase):
|
||||
|
||||
if queue_type == kfd.KFD_IOC_QUEUE_TYPE_SDMA:
|
||||
assert idx <= 3, "only 4 SDMA queues supported in am"
|
||||
pv = self.dev_impl.sdma.setup_ring(ring_addr=ring.va_addr, ring_size=ring.size, rptr_addr=gart.va_addr+rptr, wptr_addr=gart.va_addr+wptr,
|
||||
doorbell=(doorbell_index:=am.AMDGPU_NAVI10_DOORBELL_sDMA_ENGINE0 + idx * 0xA * 4), pipe=0, queue=idx)
|
||||
pv, doorbell_index = self.dev_impl.sdma.setup_ring(ring_addr=ring.va_addr, ring_size=ring.size, rptr_addr=gart.va_addr+rptr,
|
||||
wptr_addr=gart.va_addr+wptr, pipe=0, queue=idx)
|
||||
else:
|
||||
pv = self.dev_impl.gfx.setup_ring(ring_addr=ring.va_addr, ring_size=ring.size, rptr_addr=gart.va_addr+rptr, wptr_addr=gart.va_addr+wptr,
|
||||
eop_addr=eop_buffer.va_addr, eop_size=eop_buffer.size, doorbell=(doorbell_index:=am.AMDGPU_NAVI10_DOORBELL_MEC_RING0), pipe=0,
|
||||
pv, doorbell_index = self.dev_impl.gfx.setup_ring(ring_addr=ring.va_addr, ring_size=ring.size, rptr_addr=gart.va_addr+rptr,
|
||||
wptr_addr=gart.va_addr+wptr, eop_addr=eop_buffer.va_addr, eop_size=eop_buffer.size, pipe=0,
|
||||
queue=int(is_aql:=(queue_type==kfd.KFD_IOC_QUEUE_TYPE_COMPUTE_AQL)), aql=is_aql)
|
||||
|
||||
return AMDQueueDesc(ring=ring.cpu_view().view(fmt='I'), doorbells=[self.dev_impl.doorbell64.view(doorbell_index * 8, 8, fmt='Q')],
|
||||
|
||||
@@ -281,9 +281,10 @@ class AM_GFX(AM_IP):
|
||||
self._grbm_select(inst=xcc)
|
||||
for xcc in range(self.xccs): self.adev.regGCVM_CONTEXT0_CNTL.write(0, inst=xcc)
|
||||
|
||||
def setup_ring(self, ring_addr:int, ring_size:int, rptr_addr:int, wptr_addr:int, eop_addr:int, eop_size:int, doorbell:int, pipe:int, queue:int,
|
||||
aql:bool) -> int:
|
||||
def setup_ring(self, ring_addr:int, ring_size:int, rptr_addr:int, wptr_addr:int, eop_addr:int, eop_size:int, pipe:int, queue:int,
|
||||
aql:bool) -> tuple[int, int]:
|
||||
self._grbm_select(me=1, pipe=pipe, queue=queue, inst=0)
|
||||
doorbell = am.AMDGPU_NAVI10_DOORBELL_MEC_RING0
|
||||
restore_queue = aql and self.xccs > 1 and self.adev.partial_boot and (self.adev.regCP_HQD_ACTIVE.read(inst=0) & 1)
|
||||
restore_ptr = (self.adev.regCP_HQD_PQ_WPTR_LO.read(inst=0) | (self.adev.regCP_HQD_PQ_WPTR_HI.read(inst=0) << 32)) if restore_queue else 0
|
||||
if DEBUG >= 2 and restore_queue: print(f"am {self.adev.devfmt}: GFX queue already active, continuing from saved state {restore_ptr=:#x}.")
|
||||
@@ -327,7 +328,7 @@ class AM_GFX(AM_IP):
|
||||
self._grbm_select(inst=xcc)
|
||||
|
||||
self.adev.reg(f"regCP_ME1_PIPE{pipe}_INT_CNTL").update(time_stamp_int_enable=1, generic0_int_enable=1, inst=xcc)
|
||||
return restore_ptr // 16
|
||||
return restore_ptr // 16, 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)
|
||||
@@ -447,9 +448,9 @@ 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, doorbell:int, pipe:int, queue:int) -> int:
|
||||
# Setup the ring
|
||||
def setup_ring(self, ring_addr:int, ring_size:int, rptr_addr:int, wptr_addr:int, pipe:int, queue:int) -> tuple[int, int]:
|
||||
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
|
||||
self.sdma_reginst.append((reg, inst))
|
||||
|
||||
self.adev.reg(f"{reg}_MINOR_PTR_UPDATE").write(0x1, inst=inst)
|
||||
@@ -464,7 +465,7 @@ 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 self.adev.reg(f"{reg}_RB_WPTR").read(inst=inst) | (self.adev.reg(f"{reg}_RB_WPTR_HI").read(inst=inst) << 32)
|
||||
return self.adev.reg(f"{reg}_RB_WPTR").read(inst=inst) | (self.adev.reg(f"{reg}_RB_WPTR_HI").read(inst=inst) << 32), doorbell
|
||||
|
||||
class AM_PSP(AM_IP):
|
||||
def init_sw(self):
|
||||
|
||||
Reference in New Issue
Block a user