Kgj SSKrSSKrSSKJr SSKJrJr SSKJ r \R"SS05\R"SSVs/sHn\R0US9PM snS/S .\ D6\RS\RS \RS\R4S j555rSS \S-S \R"S-4Sjjrgs snf)N)explog)autotune_cache_kwargs HAS_SCALEcUSSL$)Nscale)argss ڥC:\Users\johnn\AppData\Local\Packages\PythonSoftwareFoundation.Python.3.13_qbz5n2kfra8p0\LocalCache\local-packages\Python313\site-packages\fla/ops/utils/logsumexp.pyr s d7m47)  num_warpsD)configskeyBcd[R"S5R[R5[R"S5R[R5pvXt-[R"SU5-nX:n [R "XU--U-U [ S5*S9n U(aX-n [R"U S5n [[R"[X- 5S55U -n [R"X[R"X45--U-U 5 g)Nrrinf)maskother) tl program_idtoint64arangeloadfloatmaxrsumrstorecdiv) xzrrrri_ni_do_dm_db_xb_mb_zs r logsumexp_fwd_kernelr2s(}}Q""288,bmmA.>.A.A"((.K 'BIIaO #C 'C ''!Ag+##eEl] CCk &&a.C bffS^Q' (3 .CHHQrwwq}$ $s *C0r rdtypecURnURSUS5nURupE[[R"U5S5n[R "XV5nUR XG[RS9n[XG4"UUUUUS9 URS5R"USS6nUb%U[R:waURU5nU$)aw Compute the logsumexp of the input tensor over the last dimension. Args: x (Tensor): The input tensor of any shape. scale (Optional[float]): The scale applied to the input tensor. Default: `None`. dtype (Optional[torch.dtype]): The data type of the output tensor. Default: `None`. Returns: Tensor: The logsumexp of the input tensor. i)r3)r)r*rrrN) shapeviewmintritonnext_power_of_2r( new_emptytorchr$r2 logsumexpr ) r)rr3r6NrrNDr*s r logsumexp_fwdr@0s& GGE r59A 77DA F " "1 %y1A Q B A -A!!   BeCRj)A Uekk1 DDK Hr r )NN)r<r9triton.languagelanguagerfla.ops.utils.oprr fla.utilsr heuristicsautotuneConfigjit constexprr2r$r3r@rs0r rJs %+7. -I  bI .-     1 || 1  || 1 || 1 1, $$ 4<$  ;; $ 9 sC