github/tinygrad
1
1
Fork 0
mirror of https://github.com/tinygrad/tinygrad.git synced 2026-02-12 23:54:58 -05:00
Code Issues Packages Projects Releases Wiki Activity

Update CLProgram to use unsigned long long for event profiling (#2808)

Browse Source 
On Windows, the unsigned long type is 32-bit, which is not compatible
with the required data size for event profiling.
This commit is contained in:
Marcus Asteborg
2023-12-16 23:48:44 -08:00
committed by GitHub
parent c333bfcf69
commit 1fa4f161fe
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tinygrad/runtime/ops_gpu.py
View File

@@ -51,8 +51,8 @@ class CLProgram:
check(cl.clEnqueueNDRangeKernel(self.device.queue, self.kernel, len(global_size), None, (ctypes.c_size_t * len(global_size))(*global_size), (ctypes.c_size_t * len(local_size))(*local_size) if local_size else None, 0, None, event)) # noqa: E501
if wait:
check(cl.clWaitForEvents(1, ctypes.byref(event)))
start = init_c_var(ctypes.c_ulong(), lambda x: check(cl.clGetEventProfilingInfo(event, cl.CL_PROFILING_COMMAND_START, ctypes.sizeof(x), ctypes.byref(x), None))) # noqa: E501
end = init_c_var(ctypes.c_ulong(), lambda x: check(cl.clGetEventProfilingInfo(event, cl.CL_PROFILING_COMMAND_END, ctypes.sizeof(x), ctypes.byref(x), None))) # noqa: E501
start = init_c_var(ctypes.c_uint64(), lambda x: check(cl.clGetEventProfilingInfo(event, cl.CL_PROFILING_COMMAND_START, ctypes.sizeof(x), ctypes.byref(x), None))) # noqa: E501
end = init_c_var(ctypes.c_uint64(), lambda x: check(cl.clGetEventProfilingInfo(event, cl.CL_PROFILING_COMMAND_END, ctypes.sizeof(x), ctypes.byref(x), None))) # noqa: E501
return float(end.value-start.value) * OSX_TIMING_RATIO * 1e-9
return None