~i>SSKJr SSKJrJr SSKJr SSKJr \S5r\S5r \"\R5r \R\S 55r \R\\R"S 5S 555r\R\\R"S 5SDSj555r\R\SESj55r\S5r\S5r\S5r\S5r\S5r\S5r\S5r\R\\R2"SSSS9SFSj555r\R\\R2"SSS9SGSj555r\S 5r\S!5r\S"5r\S#5r\R\\R2"S$SSS9SFS%j555r \R\\R2"S&SS9SGS'j555r!\S(5r"\S)5r#\R\\R2"S*S+S,9SHSIS-jj555r$\S.5r%\R\\R2"S/5SJS0j555r&\S15r'\R\\R2"S25SES3j555r(\R\\RR"S4S+S,9SKSIS5jj555r*\S65r+\R\\RR"S75SLS8j555r,\SMS9j5r-\SNS:j5r.\SOS;j5r/\SPS<j5r0\S S \Rb4SQS=jj5r2\S \Rb4SRS>jj5r3\SSSTS?jj5r4\S \Rb4SRS@jj5r5\SA5r6\R\SSSBj55r7\SC5r8g )U) annotations)jitconstexpr_function)core)mathc>SnUnUS:aUS-nUS- nUS:aMU$)Nrr)ilog2ns U/home/mande/repo/quber/.venv/lib/python3.13/site-packages/triton/language/standard.py_log2r s5 D A a% a   a% Kc.XS- -S:H=(a US:g$Nrrr )r s r_is_power_of_twors QKA  (!q&(rcXS- -U-$)z Computes the ceiling division of :code:`x` by :code:`div` :param x: the input number :type x: Block :param div: the divisor :type div: Block rr )xdivs rcdivr s qMc !!rsigmoidc<SS[R"U*5-- $)Nr)r exp)rs rrr.s DHHaRL !!rsoftmaxNcUcSnOUnU[XUS9- n[R"U5n[XdUS9n[R"XgU5$)Nr keep_dims)maxr rsumfdiv)rdimr ieee_rounding_dimznumdens rrr5sN { " C9 --A ((1+C c9 -C 99S} --rcB[R"XR/US9$)z^ Returns a contiguous flattened view of :code:`x`. :param x: the input tensor :type x: Block ) can_reorder)rreshapenumel)rr*s rravelr-Cs <<GG9+ >>rc~X-U-nXC-nXV-nXt-n[R"X(- U5nXV-nXU--n XT-n X4$)a Transforms the indices of a row-major `size_i * size_j` matrix into the indices of a column-major matrix for each group of `size_g` rows. For example, for :code:`size_i = size_j = 4` and :code:`size_g = 2`, it will transform :: [[0 , 1 , 2 , 3 ], [4 , 5 , 6 , 7 ], [8 , 9 , 10, 11], [12, 13, 14, 15]] into :: [[0, 2, 4 , 6 ], [1, 3, 5 , 7 ], [8, 10, 12, 14], [9, 11, 13, 15]] rminimum) r jsize_isize_jsize_gijsize_gjgroup_idoff_inew_inew_js r swizzle2dr;Os[, aBoG}H  E \\&.& 1F B K E LE <rc2[R"USU5$)a Returns a tensor filled with the scalar value 0 for the given :code:`shape` and :code:`dtype`. :param shape: Shape of the new array, e.g., (8, 16) or (8, ) :type shape: tuple of ints :param dtype: Data-type of the new array, e.g., :code:`tl.float16` :type dtype: DType r)rfull)shapedtypes rzerosr@ws 99UAu %%rcB[URUR5$)z{ Returns a tensor of zeros with the same shape and type as a given tensor. :param input: input tensor :type input: Tensor )r@r>r?)inputs r zeros_likerCs ekk **rcU(aX:H=(a X:nOSnX:=(d Un[R"X`U5n[R"XaU5nXx4$NFrwhere) value1index1value2index2tie_break_lefttiegtv_reti_rets r_argmax_combinerQsM26?  CB JJr6 *E JJr6 *E <rc[XX#S5$NTrQrHrIrJrKs r_argmax_combine_tie_break_leftrV 664 @@rc[XX#S5$rErTrUs r_argmax_combine_tie_break_fastrY 665 AArc.[R"X5$N)rmaximumabs r_elementwise_maxra << rr]return_indicesreturn_indices_tie_break_left)return_indices_arg tie_break_argc[R"U5nU(a;U(a[R"X[US9$[R"X[US9$[R "UR R5[R "S5:a[R "UR R55(a UR[R5nOEUR R5(dS5eUR[R5n[R"X[US9$Nr z"Expecting input to be integer type)r_promote_bfloat16_to_float32_reduce_with_indicesrVrY constexprr?primitive_bitwidth is_floatingtofloat32is_intint32reducerarBaxisrcrdrs rr r s  - -e 4E (,,U:Xdmn n,,U:Xdmn n >>%++88 9DNN2>%++88 9B >~~ekk55788.{{))++Q-QQ+,{{5(8INNrz minimum indexc [XSX#S9upEU$rw)rrxs rargminrsTQ_ uFA Jrc X-$r\r r^s r _sum_combiner 5LrcUbU$SnUR5(a&URS:a[RnU$SnU$UR 5(a"URS:a[R OSnU$)Nri) is_int_signed int_bitwidthrrris_int_unsigneduint32)in_dtyper? out_dtypes r_pick_sum_dtyper s  I"*"7"7"")rrrrhs r_bitonic_mergersD QeAGGn,-A 51A Q A Hrc&Uc[UR5S- OUn[R"U[UR5S- :HS5 [ URU5nUcUO [ U5n[ UR 5n[R "X(aS/U-OS/5n[R"SUS-5Hn [XX:aSOU5nM [R"US-US-5Hcn U(a#[U[ UR 5S- U- S9O"[U[ UR 5S- U- S9n[XX:aSOU5nMe [R "XRSSSU-/-5nU$)aA Sorts a tensor along a specified dimension. :param x: The input tensor to be sorted. :type x: Tensor :param dim: The dimension along which to sort the tensor. If None, the tensor is sorted along the last dimension. Currently, only sorting along the last dimension is supported. :type dim: int, optional :param k: the number of top elements to select. If none, assume k = x.shape[dim] :type k: int, optional :param descending: If set to True, the tensor is sorted in descending order. If set to False, the tensor is sorted in ascending order. :type descending: bool, optional Nr+only minor dimension is currently supportedr)ru) lenr>rrrr,r+rrr r) rkr# descendingr%log_nlog_krrr s r sort_implrsY03{3qwwrrrr)rr#rr%rs r bitonic_mergers^03{3qwws r _get_flip_dimrs. {%j1n Qw s5z Jrc[R"[UR5*U:*=(a U[UR5:5 [ XR5n[R"[ URU55 [ URU5n[URRSS9n[R"URUSS9URSUS/U--URUS-S-5n[R"U5HnU[XRU-S5- nM [R"XPR5RURSS9nU$)z Flips a tensor `x` along the dimension `dim`. :param x: the first input tensor :type x: Block :param dim: the dimension to flip along :type dim: int TrrNrr)rrrr>rrrrr?rmr+rorr)rr#r%stepsrrr s rrrs) AGG }+Bc!''l0BC(gg6D' 67!!''$-0EQWW%?%? MF QTT&$T/$1#+1MPQPWPWX\_`X`XaPb1bcA   u % !8T* *& Q ##AGGT#:A Hrc[R"X5n[UR5S:XaU$[R"X"RSSSURS-/-5$)a Interleaves the values of two tensors along their last dimension. The two tensors must have the same shape. Equivalent to `tl.join(a, b).reshape(a.shape[:-1] + [2 * a.shape[-1]])` :param a: The first input tensor. :type a: Tensor :param b: The second input tensor. :type b: Tensor rNr)rjoinrr>r+)r_r`cs r interleaversU !A 177|q ||Awws|q1772;.??@@r)NFF)F)NFTF)TF)NFN)r?core.constexprrE)rFN)rF)rrr1r)r r)rrrr)rrrrrr)rrr#rrr)r#rrrr\)rrr#r)9 __future__r runtime.jitrrrr rr get_int_dtyper_tensor_member_fnr_add_math_1arg_docstrrrr-r;r@rCrQrVrYra_add_reduction_docstrr r{rrrrrrrrr!rrrr_add_scan_docstrrrrrrrr CONSTEXPR_0rrrrrrrr rrrs^"1 ))$D$6$67 " "I&"'"I&.'.??$$N & &++AABBI:J*IKOKO"O;KLM  AABBI:J*IKOKO"O;KLM   EW5G6GI&G'GK(F)Fx73 E4 Ey!F"F $  *  %)dhdtdt% % P"&TEUEU8877+/dN^N^99  .AAr