mindspore.hal

Hal encapsulates interfaces for device, stream, event, and memory. MindSpore abstracts the corresponding modules from different backends, allowing users to schedule hardware resources at the Python layer.

Device

API Name	Description	Supported Platforms
mindspore.hal.device_count	Returns device count of specified backend.
mindspore.hal.get_arch_list	Get the architecture list this MindSpore was compiled for.
mindspore.hal.get_device_capability	Get specified device's capability.
mindspore.hal.get_device_name	Get specified device's name.
mindspore.hal.get_device_properties	Get specified device's properties.
mindspore.hal.is_available	Returns whether specified backend is available.
mindspore.hal.is_initialized	Returns whether specified backend is initialized.

Stream

API Name	Description	Supported Platforms
mindspore.hal.communication_stream	Return default stream on this device.
mindspore.hal.current_stream	Return current stream used on this device.
mindspore.hal.default_stream	Return default stream on this device.
mindspore.hal.set_cur_stream	Sets the current stream.This is a wrapper API to set the stream.
mindspore.hal.synchronize	Synchronize all streams on current device.(Each MindSpore process only occupies one device)
mindspore.hal.Stream	Wrapper around a device stream.
mindspore.hal.StreamCtx	Context-manager that selects a given stream.

Event

API Name	Description	Supported Platforms
mindspore.hal.Event	Wrapper around a device event.

Memory

API Name	Description	Supported Platforms
mindspore.hal.contiguous_tensors_handle.combine_tensor_list_contiguous	Return a contiguous memory handle where contiguous memory has been requested and slicing functionality is provided.
mindspore.hal.contiguous_tensors_handle.ContiguousTensorsHandle	ContiguousTensorsHandle is a handle manage continuous memory.
mindspore.hal.max_memory_allocated	Returns the peak memory size of the memory pool actually occupied by Tensor since the process was started.
mindspore.hal.max_memory_reserved	Returns the peak value of the total memory managed by the memory pool since the process was started.
mindspore.hal.memory_allocated	Returns the actual memory size currently occupied by Tensor.
mindspore.hal.memory_reserved	Returns the total amount of memory currently managed by the memory pool.
mindspore.hal.memory_stats	Returns status information queried from the memory pool.
mindspore.hal.memory_summary	Returns readable memory pool status information.
mindspore.hal.reset_max_memory_reserved	Reset the peak memory size managed by the memory pool.
mindspore.hal.reset_max_memory_allocated	Reset the peak memory size of the memory pool actually occupied by Tensor.
mindspore.hal.reset_peak_memory_stats	Reset the "peak" stats tracked by memory manager.
mindspore.hal.empty_cache	Release all memory fragments in the memory pool, so that memory arrangement will be optimized.