AVh@fdZddlZddlZddlZddlmZddlmZddlmZddl m Z ddl m Z ddl m Z dd l m Z dd lmZdd lmZdd lmZd gZdZdZGddej,ddZdZdZdZej6ZdZej<dZed Gdd Z y)z,Critical Section object and execution logic.N)context)dtypes)ops) array_ops)control_flow_ops)gen_resource_variable_ops)tensor_array_ops)nest)object_identity) tf_exportCriticalSectioncritical_sectionscritical_section_executionsceZdZdZy)_ExecutionSignaturezGA class storing an `ExecuteInCriticalResource` op and associated attrs.N)__name__ __module__ __qualname____doc__Z/home/dcms/DCMS/lib/python3.12/site-packages/tensorflow/python/ops/critical_section_ops.pyrr)s Prrophandle resourcesexclusive_resource_accessc t|tjr|jSt|tj rt j|Stjr|ytj|S)zEIdentity op that recognizes `TensorArray`, `Operation`, and `Tensor`.N) isinstancer TensorArrayidentityr Operationrgrouprexecuting_eagerlyr)xs r _identityr&1sb#//0 ::<!S]]#  ! !! $$  "qy    a  rc4|jxs t|SN)device_get_colocationrs r_get_device_or_colocationr,=s  )ob))rcP |jdS#ttf$rYywxYw)z&Get colocation symbol from op, if any._classN)get_attr ValueErrorAttributeErrorr+s rr*r*As- ;;x  n % s %%cx tjS#t$rgt_tjcYSwxYwr()_CRITICAL_SECTION_STACKvaluer1rrr_get_critical_section_stackr5Ls7) " ( (( )$&! " ( (()s $99c#Kt}||vrtd|d|j| d|j}||k7rt d|d|y#|j}||k7rt d|d|wxYww)aPush a CriticalSection._signature to the thread-local stack. If the signature is already on the stack, raise an error because it means we're trying to execute inside the same locked CriticalSection, which will create a deadlock. Args: signature: Tuple of the type `CriticalSection._signature`. Uniquely identifies a CriticalSection by its `shared_name`, `container`, and device. Yields: An empty value. The context is guaranteed to run without deadlock. Raises: ValueError: If the signature is already on the stack. RuntimeError: If another thread or function modifies the current stack entry during the yield. z0Attempting to lock a CriticalSection (signature=zK) in which we are already running. This is illegal and may cause deadlocks.Nz8CriticalSection stack inconsistency: expected signature z but received )r5r0appendpop RuntimeError) signaturestackreceived_signatures r_push_critical_section_stackr=Ts* & '%%  :9+FL L MM,,y< Y&  D K~&8%9 ; <<'Y&  D K~&8%9 ; <<'s/BA'B(BBcLeZdZdZ d dZdZedZd dZdZ dZ d Z y) r a# Critical section. A `CriticalSection` object is a resource in the graph which executes subgraphs in **serial** order. A common example of a subgraph one may wish to run exclusively is the one given by the following function: ```python v = resource_variable_ops.ResourceVariable(0.0, name="v") def count(): value = v.read_value() with tf.control_dependencies([value]): with tf.control_dependencies([v.assign_add(1)]): return tf.identity(value) ``` Here, a snapshot of `v` is captured in `value`; and then `v` is updated. The snapshot value is returned. If multiple workers or threads all execute `count` in parallel, there is no guarantee that access to the variable `v` is atomic at any point within any thread's calculation of `count`. In fact, even implementing an atomic counter that guarantees that the user will see each value `0, 1, ...,` is currently impossible. The solution is to ensure any access to the underlying resource `v` is only processed through a critical section: ```python cs = CriticalSection() f1 = cs.execute(count) f2 = cs.execute(count) output = f1 + f2 session.run(output) ``` The functions `f1` and `f2` will be executed serially, and updates to `v` will be atomic. **NOTES** All resource objects, including the critical section and any captured variables of functions executed on that critical section, will be colocated to the same device (host and cpu/gpu). When using multiple critical sections on the same resources, there is no guarantee of exclusive access to those resources. This behavior is disallowed by default (but see the kwarg `exclusive_resource_access`). For example, running the same function in two separate critical sections will not ensure serial execution: ```python v = tf.compat.v1.get_variable("v", initializer=0.0, use_resource=True) def accumulate(up): x = v.read_value() with tf.control_dependencies([x]): with tf.control_dependencies([v.assign_add(up)]): return tf.identity(x) ex1 = CriticalSection().execute( accumulate, 1.0, exclusive_resource_access=False) ex2 = CriticalSection().execute( accumulate, 1.0, exclusive_resource_access=False) bad_sum = ex1 + ex2 sess.run(v.initializer) sess.run(bad_sum) # May return 0.0 ``` Nctj|r|td|d|d|r td|j||y)zCreates a critical section.NzArguments critical_section_def=z and shared_name=z9 are mutually exclusive. Please only specify one of them.z1Argument `critical_section_def` is not supported.)rensure_initializedr0_init_from_args)selfname shared_namecritical_section_def import_scopes r__init__zCriticalSection.__init__sd   0 89M8NO**57:: ;; J KK 4-rctj|dg5}tj5tjj}||}|d}t j ||||_||xst|jt|jf|_ ddddddtjstjt|yy#1swYAxYw#1swYExYw)z:Initialize the CriticalSection from constructor arguments.r N)rD containerrC)r name_scope init_scopeget_default_graph _containerrmutex_v2_handleidr, _signaturerr$add_to_collectionsCRITICAL_SECTIONS)rBrCrDrJs rrAzCriticalSection._init_from_argss / 45 >> 5))+66  +  )099#ytE    +2dll+ %dll 3 555$  $ $ & .5 '#5555s#C1A8C%%C1%C. *C11C:cB|jjjSr()rPrrC)rBs rrCzCriticalSection.names <|tAjBt5j6|}dddtj g5t5jDt8|}dddtjsFtGj j tI |}tjJtL|cdddS#1swYxYw#1swYAxYw#1swYNxYw#1swYSxYwcc} }wcc} w#1swY+xYw#1swYxYw#1swYxYw#1swYyxYw)aExecute function `fn()` inside the critical section. `fn` should not accept any arguments. To add extra arguments to when calling `fn` in the critical section, create a lambda: ```python critical_section.execute(lambda: fn(*my_args, **my_kwargs)) ``` Args: fn: The function to execute. Must return at least one tensor. exclusive_resource_access: Whether the resources required by `fn` should be exclusive to this `CriticalSection`. Default: `True`. You may want to set this to `False` if you will be accessing a resource in read-only mode in two different CriticalSections. name: The name to use when creating the execute operation. Returns: The tensors returned from `fn()`. Raises: ValueError: If `fn` attempts to lock this `CriticalSection` in any nested or lazy way that may cause a deadlock. ValueError: If `exclusive_resource_access == True` and another `CriticalSection` has an execution requesting the same resources as `fn``. Note, even if `exclusive_resource_access` is `True`, if another execution in another `CriticalSection` was created without `exclusive_resource_access=True`, a `ValueError` will be raised. critical_section_executeNc3@K|]}j|ywr()_is_self_handle).0r%rBs r z*CriticalSection.execute..+sC1t##A&CszPAttempting to lock a CriticalSection in which we are already running (signature=z+). This is illegal and may cause deadlocks.r)'rrKr=rRr mutex_lockrPrr$rM_lockget_operationscontrol_dependenciesset difference!_add_control_dependencies_to_lockrr ObjectIdentitySetinputsdtyperresourceanyr0#_check_multiple_access_to_resourcesr flattenr& colocate_withconsume_mutex_lockpack_sequence_asrtuple map_structurerlistrSCRITICAL_SECTION_EXECUTIONS)rBfnrrClock existing_opsr created_opsrinput_captured_resourcesr%r_flatensure_lock_existsoutputsr:s` rexecutezCriticalSection.executesF< 8" =J ( 8(33DLLA((*$$&,, 6002AACL))4&1 $a  s446EEGH&J|4  6 6''/ A !& & & ( ..{DGGD ->>(@ ))@ ||v. @ @   C0BC C,,0OO+<=))* * 00  9 ;'+ll1o6 ! 6f6  # #F + N   t|| ,  9KK      ! !!%5%;%;DLLO%L M N  # #%7$8 93$$Y23 & & ('ww<<-.&? A  ' 4UJJ   6 6  !2@ &7   N N33}JJsOAN9M*:M AM*N#M7+N3A O?7N 6AON&O= N)N3AN)7ON50A#OM' "M**M4 /N7N <NN ON& !N))N2 .O5N> :OO ct|Dcgc]}|jD]}|j!c}}}|jd|Dt d|D}|D]}|j |j d |jD]}|j |j d |jD](}|j |jj d*|j |j d|j}|sytj|}|j|ycc}}w)z=To avoid deadlocks, all args must be executed before lock_op.c3BK|]}|jD]}|ywr()control_inputs)rZrinput_ops rr[zDCriticalSection._add_control_dependencies_to_lock..Us1Jr7H7HJ+3JJsc38K|]}|j|fywr()_id)rZrs rr[zDCriticalSection._add_control_dependencies_to_lock..\s9""&&"9sN) r`rdrupdatedictr8rr~valuesrr#_add_control_input)rBrulock_oprrvall_args all_args_dicts rrbz1CriticalSection._add_control_dependencies_to_lockQs# L""))LFIILILMH OOJ&JJ999M&%&$$&%&..- t,-gkk4(##%H   %%x0H x(AMs$D; c$t|tjr||juS|jj dk(xr|jj dxr|jj d|jjj dk(xrn|jj|jjjk(xs5t|jt|jjk(S)z!!# ,@ @A AAr)NNNN)TN) rrrrrGrApropertyrCr{rbrYrhrrrr r ysHBH-17; .6.    hT#)L P Ar)!r collections contextlib threadingtensorflow.python.eagerrtensorflow.python.frameworkrrtensorflow.python.opsrrrr tensorflow.python.utilr r tensorflow.python.util.tf_exportr __all__rTrp namedtuplerr&r,r*localr3r5contextmanagerr=r rrrrs3+.++2;2'26  (;K0FG !**)//+) !<!