geohot/tinygrad
1
0
Fork 0
forked from tinygrad/tinygrad
Code Pull Requests Actions Activity

add arch to AMDRenderer and HIPRenderer (#9431)

Browse Source
This commit is contained in:
Ignacio Sica
2025-03-13 13:06:27 -03:00
committed by GitHub
parent 357e364ab8
commit 459d0cd14f
3 changed files with 6 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tinygrad/renderer/cstyle.py
+4
View File
@@ -402,6 +402,10 @@ class AMDRenderer(CStyleLanguage):
opts=("l0","l0","l0","l0","l1","u1","u1","u1"), swizzle=(((4,9,10,11,0),(1,2,3,5,6,7,8)), ((0,1,2,3,4),(9,10,11,5,6,7,8))))
for di,do in [(dtypes.half,dtypes.float),(dtypes.half,dtypes.half)]]
def __init__(self, arch:str): # gfx942 => MI300, gfx1100 => RX 7900
self.tensor_cores, self.arch = AMDRenderer.tensor_cores, arch
def __reduce__(self): return self.__class__, (self.arch,)
# language options
ockl = [(f"__ockl_get_{name}", "unsigned int", "size_t", "const") for name in ["local_id", "group_id", "local_size"]]
ocml = [(f"__ocml_{name}_f{n}", f"{dt}, {dt}" if "fmax" == name else dt, dt, atr)
tinygrad/runtime/ops_amd.py
+1 -1
View File
@@ -704,7 +704,7 @@ class AMDDevice(HCQCompiled):
self.sdma_queue = self.create_queue(kfd.KFD_IOC_QUEUE_TYPE_SDMA, 0x800000)
super().__init__(device, AMDAllocator(self), AMDRenderer(), AMDCompiler(self.arch), functools.partial(AMDProgram, self),
super().__init__(device, AMDAllocator(self), AMDRenderer(self.arch), AMDCompiler(self.arch), functools.partial(AMDProgram, self),
AMDSignal, AMDComputeQueue, AMDCopyQueue)
# Scratch setup
tinygrad/runtime/ops_hip.py
+1 -1
View File
@@ -14,7 +14,7 @@ class HIPDevice(Compiled):
self.device_id = int(device.split(":")[1]) if ":" in device else 0
self.arch = init_c_var(hip.hipDeviceProp_t(), lambda x: check(hip.hipGetDeviceProperties(x, self.device_id))).gcnArchName.decode()
self.time_event_st, self.time_event_en = [init_c_var(hip.hipEvent_t(), lambda x: hip.hipEventCreate(ctypes.byref(x), 0)) for _ in range(2)]
super().__init__(device, HIPAllocator(self), HIPRenderer(), AMDCompiler(self.arch), functools.partial(HIPProgram, self))
super().__init__(device, HIPAllocator(self), HIPRenderer(self.arch), AMDCompiler(self.arch), functools.partial(HIPProgram, self))
def synchronize(self):
check(hip.hipSetDevice(self.device_id))
check(hip.hipDeviceSynchronize())