jorisvandenbossche commented on code in PR #41685: URL: https://github.com/apache/arrow/pull/41685#discussion_r1618666183
########## python/pyarrow/device.pxi: ########## @@ -0,0 +1,162 @@ +# Licensed to the Apache Software Foundation (ASF) under one +# or more contributor license agreements. See the NOTICE file +# distributed with this work for additional information +# regarding copyright ownership. The ASF licenses this file +# to you under the Apache License, Version 2.0 (the +# "License"); you may not use this file except in compliance +# with the License. You may obtain a copy of the License at +# +# http://www.apache.org/licenses/LICENSE-2.0 +# +# Unless required by applicable law or agreed to in writing, +# software distributed under the License is distributed on an +# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +# KIND, either express or implied. See the License for the +# specific language governing permissions and limitations +# under the License. + +# cython: profile=False +# distutils: language = c++ +# cython: embedsignature = True + + +cpdef enum DeviceAllocationType: + CPU = <char> CDeviceAllocationType_kCPU + CUDA = <char> CDeviceAllocationType_kCUDA + CUDA_HOST = <char> CDeviceAllocationType_kCUDA_HOST + OPENCL = <char> CDeviceAllocationType_kOPENCL + VULKAN = <char> CDeviceAllocationType_kVULKAN + METAL = <char> CDeviceAllocationType_kMETAL + VPI = <char> CDeviceAllocationType_kVPI + ROCM = <char> CDeviceAllocationType_kROCM + ROCM_HOST = <char> CDeviceAllocationType_kROCM_HOST + EXT_DEV = <char> CDeviceAllocationType_kEXT_DEV + CUDA_MANAGED = <char> CDeviceAllocationType_kCUDA_MANAGED + ONEAPI = <char> CDeviceAllocationType_kONEAPI + WEBGPU = <char> CDeviceAllocationType_kWEBGPU + HEXAGON = <char> CDeviceAllocationType_kHEXAGON + + +cdef object _wrap_device_allocation_type(CDeviceAllocationType device_type): + return DeviceAllocationType(<char> device_type) + + +cdef class Device(_Weakrefable): + """ + Abstract interface for hardware devices + + This object represents a device with access to some memory spaces. + When handling a Buffer or raw memory address, it allows deciding in which + context the raw memory address should be interpreted + (e.g. CPU-accessible memory, or embedded memory on some particular GPU). + """ + + def __init__(self): + raise TypeError("Do not call Device's constructor directly, " + "use the device attribute of the MemoryManager instead.") + + cdef void init(self, const shared_ptr[CDevice]& device): + self.device = device + + @staticmethod + cdef wrap(const shared_ptr[CDevice]& device): + cdef Device self = Device.__new__(Device) + self.init(device) + return self + + def __eq__(self, other): + if not isinstance(other, Device): + return False + return self.device.get().Equals(deref((<Device>other).device.get())) + + def __repr__(self): + return "<pyarrow.Device: {}>".format(frombytes(self.device.get().ToString())) + + @property + def type_name(self): + """ + A shorthand for this device's type. + """ + return frombytes(self.device.get().type_name()) + + @property + def device_id(self): + """ + A device ID to identify this device if there are multiple of this type. + + If there is no "device_id" equivalent (such as for the main CPU device on + non-numa systems) returns -1. + """ + return self.device.get().device_id() + + @property + def is_cpu(self): + """ + Whether this device is the main CPU device. + + This shorthand method is very useful when deciding whether a memory address + is CPU-accessible. + """ + return self.device.get().is_cpu() + + @property + def device_type(self): + """ + Return the DeviceAllocationType of this device. + """ + return _wrap_device_allocation_type(self.device.get().device_type()) + + +cdef class MemoryManager(_Weakrefable): + """ + An object that provides memory management primitives. + + A MemoryManager is always tied to a particular Device instance. + It can also have additional parameters (such as a MemoryPool to + allocate CPU memory). + + """ + + def __init__(self): + raise TypeError("Do not call MemoryManager's constructor directly, " + "use pyarrow.default_cpu_memory_manager() instead.") + + cdef void init(self, const shared_ptr[CMemoryManager]& mm): + self.memory_manager = mm + + @staticmethod + cdef wrap(const shared_ptr[CMemoryManager]& mm): + cdef MemoryManager self = MemoryManager.__new__(MemoryManager) + self.init(mm) + return self + + def __repr__(self): + return "<pyarrow.MemoryManager device: {}>".format( + frombytes(self.memory_manager.get().device().get().ToString()) + ) + + @property + def device(self): + """ + The device this MemoryManager is tied to. + """ + return Device.wrap(self.memory_manager.get().device()) + + @property + def is_cpu(self): + """ + Whether this MemoryManager is tied to the main CPU device. + + This shorthand method is very useful when deciding whether a memory + address is CPU-accessible. + """ + return self.memory_manager.get().is_cpu() + + +def default_cpu_memory_manager(): + """ + Return the default CPU MemoryManager instance. + + The returned singleton instance uses the default MemoryPool. + """ + return MemoryManager.wrap(c_default_cpu_memory_manager()) Review Comment: Not necessarily, for now, if not needed? For CPU I am not sure if it is worth it, as looking at device.h, it seems the main thing the subclass provides is allowing to specify a memory pool when creating (which we could also expose through `default_cpu_memory_manager` if we want to expose this? For CUDA there might be more CUDA-specific methods/attributes where it might be more worth providing a subclass. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
