j UddlZddlZddlmZmZddlmZddlZddlm Z m Z GddZ de de d efd Z d Zeaeed <ejd ZGddZddZdS)N)CallableOptional) deprecated)KernelRegistrationHandlec2eZdZdZdefdZdededefdZdS) FakeImplHolderz0A holder where one can register an fake impl to.qualnamec0||_d|_d|_dSN)r kernellib)selfr s IC:\Projects\Historical\venv\Lib\site-packages\torch/_library/fake_impl.py__init__zFakeImplHolder.__init__s% (, 48funcsourcereturncj%tdjdjjdtjjdrtdjdtjjdrtdjdt||_jEj d d }tj |d _tj}j j|dfd }t|S) z}Register an fake impl. Returns a RegistrationHandle that one can use to de-register this fake impl. Nz!register_fake(...): the operator z( already has an fake impl registered at .Metaz already has an DispatchKey::Meta implementation via a pre-existing torch.library or TORCH_LIBRARY registration. Please either remove that registration or don't call register_fake.CompositeImplicitAutograda% already has an implementation for this device type via a pre-existing registration to DispatchKey::CompositeImplicitAutograd.CompositeImplicitAutograd operators do not need an fake impl; instead, the operator will decompose into its constituents and those can have fake impls defined on them.z::rFRAGMENTcdjr jd_d_dSr )r_destroyr )rsrderegister_fake_classz6FakeImplHolder.register..deregister_fake_classAs3x !!###DKKKr)r RuntimeErrorr rtorch_C%_dispatch_has_kernel_for_dispatch_keyrrsplitlibraryLibraryconstruct_meta_kernelimplr)rrrns meta_kernelrs` rregisterzFakeImplHolder.registers ; ")DM));%)))  8 9 9$- P P "DM"""  8 9 9 M6  8DM888   T6**  8 $$T**1-B},,R<.meta_kernel..error_on_ctxRs8OOvOOO   r)r rset_ctx_getter)argskwargsr4rr0r s @rr(z*construct_meta_kernel..meta_kernelMs&222!(/      L ) ) < <*#*D;F;; < < < < < < < < < < < < < < < < < >> # An operator with data-dependent output shape >>> lib = torch.library.Library("mymodule", "FRAGMENT") >>> lib.define("mymodule::custom_nonzero(Tensor x) -> Tensor") >>> >>> @torch.library.register_fake("mymodule::custom_nonzero") >>> def _(x): >>> # Number of nonzero-elements is data-dependent. >>> # Since we cannot peek at the data in an fake impl, >>> # we use the ctx object to construct a new symint that >>> # represents the data-dependent size. >>> ctx = torch.library.get_ctx() >>> nnz = ctx.new_dynamic_size() >>> shape = [nnz, x.dim()] >>> result = x.new_empty(shape, dtype=torch.int64) >>> return result >>> >>> @torch.library.impl(lib, "custom_nonzero", "CPU") >>> def _(x): >>> x_np = x.numpy() >>> res = np.stack(np.nonzero(x_np), axis=1) >>> return torch.tensor(res, device=x.device) Nzctx.new_dynamic_size(min=z, max=zZ): expected min and max to be statically known ints but got SymInt. This is not supported.rzc, ...): expected min to be greater than or equal to 0: this API can only create non-negative sizes.) rEallow_dynamic_output_shape_opsr _subclasses fake_tensorDynamicOutputShapeExceptionrF isinstanceSymInt ValueError allocate_sizerOs rrNzFakeImplCtx.new_dynamic_sizesf O #?A $#/KKDHUU U c5< ( ( JsEL,I,I *C**s***  77'C'''  T_c3777r) r*r+r,r-rr FutureWarningrrWrPrNr/rrrArAvs ZW-.4777EL777 7'(TF8F8F8elF8F8F8F8F8F8rrAc|}tjjj||||SrM)rPrfx experimentalsymbolic_shapes_constrain_range_for_size)rDmin_valmax_valresults rrYrYsG  - - / /F H)CCGD Mr)rN) contextlibr8typingrrtyping_extensionsrrtorch._library.utilsrrr r.r%r;r<__annotations__contextmanagerr5rArYr/rrrisM%%%%%%%%(((((( ;;;;;;;;<9<9<9<9<9<9<9<9~C>h4'8&&& !!!W8W8W8W8W8W8W8W8tr