o
    im3                     @   s  d dl Z d dlZd dlmZmZ d dlZd dlmZ d dlm	Z	 ddl
mZmZmZmZ e s7edejjd< ee	eedf Zd-d	d
Zd.deddfddZd.deddfddZd.dedeeef fddZd.dedeeef fddZd.dedefddZd.dedefddZd.dedefddZd.dedefddZd.dedeeef fddZd.dede fddZ!d.d e deddfd!d"Z"G d#d$ d$Z#G d%d& d&e#Z$d'e#ddfd(d)Z%de#fd*d+Z&g d,Z'dS )/    N)AnyUnion)_dummy_type)Device   )_get_device_index_is_compiled
_lazy_initis_initialized_xpu_XPUAllocatorreturnc                   C   s   t  r
tj  dS dS )aZ  Release all unoccupied cached memory currently held by the caching
    allocator so that those can be used in other XPU application.

    .. note::
        :func:`~torch.xpu.empty_cache` doesn't increase the amount of XPU
        memory available for PyTorch. However, it may help reduce fragmentation
        of XPU memory in certain cases.
    N)r
   torch_C_xpu_emptyCache r   r   R/sda-disk/www/egybert/egybert_env/lib/python3.10/site-packages/torch/xpu/memory.pyempty_cache   s   	r   devicec                 C      t | dd} tj| S )a  Reset the "peak" stats tracked by the XPU memory allocator.

    See :func:`~torch.xpu.memory_stats` for details. Peak stats correspond to the
    `"peak"` key in each individual stat dict.

    Args:
        device (torch.device or int or str, optional): selected device. Returns
            statistic for the current device, given by :func:`~torch.xpu.current_device`,
            if :attr:`device` is ``None`` (default).
    Toptional)r   r   r   _xpu_resetPeakMemoryStatsr   r   r   r   reset_peak_memory_stats       r   c                 C   r   )a  Reset the "accumulated" (historical) stats tracked by the XPU memory allocator.

    See :func:`~torch.xpu.memory_stats` for details. Accumulated stats correspond to
    the `"allocated"` and `"freed"` keys in each individual stat dict.

    Args:
        device (torch.device or int or str, optional): selected device. Returns
            statistic for the current device, given by :func:`~torch.xpu.current_device`,
            if :attr:`device` is ``None`` (default).
    Tr   )r   r   r    _xpu_resetAccumulatedMemoryStatsr   r   r   r   reset_accumulated_memory_stats/   r   r   c                 C   s"   t  si S t| dd} tj| S )zLReturn the result of :func:`~torch.xpu.memory_stats` as a nested dictionary.Tr   )r
   r   r   r   _xpu_memoryStatsr   r   r   r   memory_stats_as_nested_dict>   s   r   c                    sF   g dt dtddf fdd t| d} d|   tS )	a@  Return a dictionary of XPU memory allocator statistics for a given device.

    The return value of this function is a dictionary of statistics, each of
    which is a non-negative integer.

    Core statistics:

    - ``"allocated_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``:
      amount of allocated memory.
    - ``"reserved_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``:
      amount of reserved memory.
    - ``"active_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``:
      amount of active memory.
    - ``"requested_bytes.{all,large_pool,small_pool}.{current,peak,allocated,freed}"``:
      memory requested by client code, compare this with allocated_bytes to check if
      allocation rounding adds too much overhead.

    For these core statistics, values are broken down as follows.

    Pool type:

    - ``all``: combined statistics across all memory pools.
    - ``large_pool``: statistics for the large allocation pool (for size >= 1MB allocations).
    - ``small_pool``: statistics for the small allocation pool (for size < 1MB allocations).

    Metric type:

    - ``current``: current value of this metric.
    - ``peak``: maximum value of this metric.
    - ``allocated``: historical total increase in this metric.
    - ``freed``: historical total decrease in this metric.

    Args:
        device (torch.device or int or str, optional): selected device. Returns
            statistics for the current device, given by :func:`~torch.xpu.current_device`,
            if :attr:`device` is ``None`` (default).
    prefixobjr   Nc                    sT   t |tr!t| dkr| d7 } | D ]\}} | | | qd S | |f d S )Nr   .)
isinstancedictlenitemsappend)r   r    kv_recurse_add_to_resultresultr   r   r*   n   s   
z,memory_stats.<locals>._recurse_add_to_resultr    )strr   r   sortcollectionsOrderedDict)r   statsr   r)   r   memory_statsF   s   &
	

r2   c                 C      t | dddS )a  Return the current GPU memory occupied by tensors in bytes for a given device.

    Args:
        device (torch.device or int or str, optional): selected device. Returns
            statistic for the current device, given by :func:`~torch.xpu.current_device`,
            if :attr:`device` is ``None`` (default).

    .. note::
        This is likely less than the amount shown in `xpu-smi` since some
        unused memory can be held by the caching allocator and some context
        needs to be created on GPU.
    r   zallocated_bytes.all.currentr   r2   getr   r   r   r   memory_allocated~   s   r6   c                 C   r3   )a  Return the maximum GPU memory occupied by tensors in bytes for a given device.

    By default, this returns the peak allocated memory since the beginning of
    this program. :func:`~torch.xpu.reset_peak_memory_stats` can be used to
    reset the starting point in tracking this metric. For example, these two
    functions can measure the peak allocated memory usage of each iteration in a
    training loop.

    Args:
        device (torch.device or int or str, optional): selected device. Returns
            statistic for the current device, given by :func:`~torch.xpu.current_device`,
            if :attr:`device` is ``None`` (default).
    r   zallocated_bytes.all.peakr   r4   r   r   r   r   max_memory_allocated      r7   c                 C   r3   )aJ  Return the current GPU memory managed by the caching allocator in bytes for a given device.

    Args:
        device (torch.device or int or str, optional): selected device. Returns
            statistic for the current device, given by :func:`~torch.xpu.current_device`,
            if :attr:`device` is ``None`` (default).
    r   zreserved_bytes.all.currentr   r4   r   r   r   r   memory_reserved   s   r9   c                 C   r3   )a  Return the maximum GPU memory managed by the caching allocator in bytes for a given device.

    By default, this returns the peak cached memory since the beginning of this
    program. :func:`~torch.xpu.reset_peak_memory_stats` can be used to reset
    the starting point in tracking this metric. For example, these two functions
    can measure the peak cached memory amount of each iteration in a training
    loop.

    Args:
        device (torch.device or int or str, optional): selected device. Returns
            statistic for the current device, given by :func:`~torch.xpu.current_device`,
            if :attr:`device` is ``None`` (default).
    r   zreserved_bytes.all.peakr   r4   r   r   r   r   max_memory_reserved   r8   r:   c                 C      t   t| dd} tj| S )a  Return the global free and total GPU memory for a given device.

    Args:
        device (torch.device or int or str, optional): selected device. Returns
            statistic for the current device, given by :func:`~torch.xpu.current_device`,
            if :attr:`device` is ``None`` (default).

    Returns:
        int: the memory available on the device in units of bytes.
        int: the total memory on the device in units of bytes
    Tr   )r	   r   r   r   _xpu_getMemoryInfor   r   r   r   mem_get_info   s   r=   c                 C   r;   )ab  
    Retrieve the memory fraction currently set for a process on a given XPU device.
    This fraction represents the portion of the total device memory that
    the caching allocator is allowed to use. The allowed memory is calculated as:

    .. math:: \text{allowed\_memory} = \text{total\_memory} \times \text{fraction}

    Args:
        device (torch.device or int or str, optional): selected device. It uses the current device,
            given by :func:`~torch.xpu.current_device`, if :attr:`device` is ``None`` (default).

    Returns:
        float: The memory fraction in the range 0.0 to 1.0.
    Tr   )r	   r   r   r   _xpu_getMemoryFractionr   r   r   r   get_per_process_memory_fraction   s   r?   fractionc                 C   s6   t   t|dd}t| tstdtj| | dS )a=  
    Set the memory fraction for a single process on XPU device.
    This function limits the amount of memory that the caching allocator can allocate
    on the specified XPU device. The allowed memory is computed as:

    .. math:: \text{allowed\_memory} = \text{total\_memory} \times \text{fraction}

    If the process attempts to allocate more than this allowed memory,
    an out-of-memory error will be raised by the allocator.

    Arguments:
        fraction (float): Range: 0~1. Allowed memory equals total_memory * fraction.
        device (torch.device or int or str, optional): selected device. It uses the current device,
            given by :func:`~torch.xpu.current_device`, if :attr:`device` is ``None`` (default).

    .. note:: In general, the total available free memory is less than the total capacity.
    Tr   z3Invalid type for fraction argument, must be `float`N)r	   r   r"   float	TypeErrorr   r   _xpu_setMemoryFraction)r@   r   r   r   r   set_per_process_memory_fraction   s
   
rD   c                   @   s*   e Zd ZdZdejjfddZdd ZdS )_XPUAllocatorz,Wrapper over internal XPU memory allocators.	allocatorc                 C   s
   || _ d S N
_allocator)selfrF   r   r   r   __init__   s   
z_XPUAllocator.__init__c                 C   s   | j S rG   rH   )rJ   r   r   r   rF      s   z_XPUAllocator.allocatorN)	__name__
__module____qualname____doc__r   r   r   rK   rF   r   r   r   r   rE      s    rE   c                   @   s&   e Zd ZdZdededefddZdS )XPUPluggableAllocatorz2XPU memory allocator loaded from a shared library.path_to_lib_filealloc_fn_namefree_fn_namec           	      C   sj   t |}t||}t||}t |t jj}t |t jj}|du s'|du r+tdtj	||| _
dS )a  XPU memory allocator loaded dynamically from a shared library.

        This lets users provide custom allocation and free functions implemented
        in a separate shared library. The allocator is registered through
        ``torch._C._xpu_customAllocator`` and becomes available for use via
        ``torch.memory.xpu.change_current_allocator``.

        Arguments:
            path_to_lib_file (str):
                Filesystem path to the shared library file containing the allocation
                and free functions.
            alloc_fn_name (str):
                Name of the allocation function exported from the shared library.
                The function must have the signature:

                    ``void* alloc_fn(size_t size, int device, sycl::queue* queue);``

            free_fn_name (str):
                Name of the free function exported from the shared library.
                The function must have the signature:

                    ``void free_fn(void* ptr, size_t size, sycl::queue* queue);``
        Nz9Failed to load allocator symbols from the shared library.)ctypesCDLLgetattrcastc_void_pvalueRuntimeErrorr   r   _xpu_customAllocatorrI   )	rJ   rQ   rR   rS   allocator_liballoc_fn_ptrfree_fn_ptralloc_fn_addrfree_fn_addrr   r   r   rK     s   


zXPUPluggableAllocator.__init__N)rL   rM   rN   rO   r-   rK   r   r   r   r   rP     s    rP   rF   c                 C   s   t j|   dS )a  Change the currently used memory allocator to be the one provided.

    .. note::
        If the current allocator has already been used/initialized, this function will error.

    Arguments:
        allocator (torch.xpu.memory._XPUAllocator): allocator to be set as the active one.
    N)r   r   _xpu_changeCurrentAllocatorrF   )rF   r   r   r   change_current_allocator/  s   	rb   c                   C   s   t tj S )zxReturn the allocator being currently used.

    Returns:
        _XPUAllocator: the allocator being currently used.
    )rE   r   r   _xpu_getAllocatorr   r   r   r   _get_current_allocator;  s   rd   )rP   rb   r   r?   r7   r:   r=   r6   r9   r2   r   r   r   rD   )r   NrG   )(r/   rT   typingr   r   r   torch._utilsr   torch.typesr   r,   r   r   r	   r
   r   __dict__r-   int	_device_tr   r   r   r#   r   r2   r6   r7   r9   r:   tupler=   rA   r?   rD   rE   rP   rb   rd   __all__r   r   r   r   <module>   s6    
8
+	