`ir2 NddlZddlZddlmZmZddlZddlmZddlmZddlmZe e e e fZ e e e fZ e e e e e ffZee e e e fZe e dfZejGddZd ed ed efd Zd ed ed ed efdZde edfde e dfd e e dffdZde e dfd e edffdZde e dfde e ee edffd e eeffdZdddee e dfge e dffdee edfge edffd dfdZd'dZd'dZd'dZd'dZ d(ddd dd!edd dfd"Zd#ee d$ee d%eee e d e e ee edfffd&Z!dS))N)CallableOptional)_array)_chunk)_modes.cReZdZUeeed<eeed<eeed<dS) _Blocking i_partitions j_partitions k_partitionsN)__name__ __module__ __qualname__listint__annotations__s/home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/cupyx/distributed/array/_linalg.pyr r sDs)s)s)rr location_map_alocation_map_breturncg}g}g}d}|D]\}}|||||||D]\}}||||||d} | |}| |}| |}d} |D]\}}| ||| |||D]\}}| ||| ||t|||S)Nc|\}}}|dkrtd||||dS)Nz"Step other than 1 is not supported) RuntimeErrorappend)indices partitionsstartstopsteps radd_to_partitionsz)_find_blocking..add_to_partitions6sT#tT 199CDD D%   $rct|dkrtd||dg}t||ddD] \}}||kr||!|S)NrzArray has no chunkr)lenrsortzipr)rresxys rto_unique_sortedz(_find_blocking..to_unique_sortedGs z??a  344 4!}o JqrrN33  DAqAvv 1  rc|\}}}||dz||krtddS)NrzInconsistent index mapping)indexr)rrr r!_s r check_indicesz%_find_blocking..check_indicesXsQ tQ   E " "Q &**:*:4*@*@ @ @;<< < A @r)keysr ) rrr r r r# i_indices k_indices j_indicesr+r/s r_find_blockingr4-s !L L L   !/ 3 3 5 533 9)\222)\2222 . 3 3 5 533 9)\222)\2222   $#L11L##L11L##L11L=== !/ 3 3 5 5// 9 i... i.... . 3 3 5 5// 9 i... i.... \< > >>rblockingc *|j}|j}|j}g}t||ddD]}t||ddD]}t||ddD]} |dz| dzf} | dz|dzf} t || } t || } | | z}|r-|}|| | |ftd| d| Ȍ|S)NrrzAThere is no device that can perform multiplication between block z and ) r r r r'setr0poprr)r5rrr r r plani_rangej_rangek_rangeblock_ablock_b devices_a devices_b intersectiondevs r_make_execution_planrDhs{ (L(L(LD|\!""%566CC<abb)9:: C CG|\!""-=>> C C"T>7T>:"T>7T>:w 7 < < > >?? w 7 < < > >?? (94 C'**,,CKK'3 78888&B*1BB8?BBCCC C C& Krslicesshapect|t|ksJtdt||DS)Nc3FK|]\}}||VdSN)r).0slengths r z%_convert_to_tuples..s2GGyq&6""GGGGGGr)r%tupler')rErFs r_convert_to_tuplesrOsH v;;#e** $ $ $ $ GGC4F4FGGG G GGrtuplesc4td|DS)Nc3(K|] }t|VdSrIslice)rJts rrMz%_convert_to_slices..s&++q++++++r)rN)rPs r_convert_to_slicesrVs! ++F+++ + ++r index_mapcFi}|D]\}}t|D]t\}}t||}|dd|dd} }tjt | } ||i} | | i} || |<u|S)N)items enumeraterOtypingcast _BlockIdx setdefault) rFrW location_mapsrCidxschunk_iidx idx_tuples batch_idx block_idx location_maplocations r_group_by_batchris9;M__&&$$ T%dOO $ $LGS+C77J#-crc?JrssOyI Iy99I(33IrBBL#..y"==H#HSMM $ rarr_array.DistributedArrayf_shapef_idxcdtjdtjffd i}|jD]\}}fd|D||<|j}t j||j||j|j S)Nchunkrc|j|jj}|j}fd|jD}t j||j|||jS)Nc0g|]\}}||fSrr)rJdatarcrms r z>_reshape_array_with..reshape_chunk..s*EEE)$D%%**%EEEr) arrayreshaperFr-updatesr_Chunkready prevent_gc)rorrr-rvrmrls r reshape_chunkz*_reshape_array_with..reshape_chunksz{""775;+<#=#=>>ek""EEEEu}EEE} %+ugu/?AA Arc&g|] }|Srr)rJrorzs rrsz'_reshape_array_with..s#DDDE==//DDDr) rrw _chunks_maprZrFrDistributedArraydtype_mode_comms)rjrlrm chunks_maprCchunksrFrzs `` @r_reshape_array_withrs AV]Av}AAAAAAAJ,,..EE VDDDDVDDD 3 GCI  E  " sy*ci = ==rc(t|ddS)Nc d|zSNr7rrFs rz'_prepend_one_to_shape..s dUlrc(tdf|zSrIrSrcs rrz'_prepend_one_to_shape..sU4[[NS(rrrjs r_prepend_one_to_shaper#  ""(( * **rc(t|ddS)Nc |dzSrrrs rrz&_append_one_to_shape..s edlrc(|tdfzSrIrSrs rrz&_append_one_to_shape..sC5;;.(rrrs r_append_one_to_shaperrrcN|jddksJt|ddS)Nrc|ddSNrrrs rrz!_pop_from_shape..seCRCjrc|ddSrrrs rrz!_pop_from_shape..sCHrrFrrs r_pop_from_shapers: 9R=A          rcN|jddksJt|ddS)Nrrc|ddSNrrrs rrz'_pop_front_from_shape..seABBirc|ddSrrrs rrz'_pop_front_from_shape..sCGrrrs r_pop_front_from_shapers: 9Q<1        raboutc |tddD]}||vrtd|dt|tjrt|tjstd|t j}|t j}dx}}|jdkrd }t|}|jdkrd }t|}|j d d\}}|j d d\} } || ks |j dd |j dd krtd t|j |j } t|j |j } | | krtd d |jD} d}| D]}| |}| |}t!||}t#|||}t%|}|D]c\}}}||}||}|j|||}|j|||}|t j|t j|t+|dt+|dfz}|5}||jt3jjj|j|jfi|}t=j|| |||f}| |!||j"}dddn #1swxYwYe|j dd || fz} tj| || t j#|j$}!|rtK|!}!|rtM|!}!|!S)akMatrix multiplication between distributed arrays. The arguments must have compatible :attr:`~DistributedArray.shape` and :attr:`~DistributedArray.index_map`. This operation converts its operands into the replica mode, and compute their product in the sum mode. Args: a, b: Input distributed arrays. out (optional): A location into which the result is stored. This option is currently not supported. Returns: The matrix product of the inputs. Example: >>> A = distributed_array( ... cupy.arange(6).reshape(2, 3), ... make_2d_index_map([0, 2], [0, 1, 3], ... [[{0}, {1, 2}]])) >>> B = distributed_array( ... cupy.arange(12).reshape(3, 4), ... make_2d_index_map([0, 1, 3], [0, 2, 4], ... [[{0}, {0}], ... [{1}, {2}]])) >>> C = A @ B >>> C.mode 'sum' >>> C.all_chunks() {0: [array([[0, 0], [0, 3]]), array([[0, 0], [6, 9]])], 1: [array([[20, 23], [56, 65]])], 2: [array([[26, 29], [74, 83]])]} >>> C array([[20, 23, 26, 29], [56, 68, 80, 92]]) .. seealso:: :obj:`numpy.matmul` NzArgument `out` is not supported)subokaxesaxisz Argument `z` is not supportedzHMixing a distributed array with a non-distributed array is not supportedFrTrYzShapes are incompatiblezMismatched batch shapesci|]}|gSrr)rJrCs r zmatmul..,s1O1O1Oc#r1O1O1Orr)ry)'r isinstancerr} _to_op_moderREPLICAndimrrrF ValueErrorrirWr0devicesr4rDrVr|flushrTon_ready wait_eventrxcupylinalg_productmatmulrtrrwrecordrr~SUMrrr)"rrrkwargsparam one_prepended one_appendednmm2plocation_maps_alocation_maps_brr~rerrr5r: index_prefixr>r?rCloc_aloc_bchunk_achunk_br-streamchunk_ab_arraychunk_abrFr(s" rrrs0^ <===*GG F??EEEEEFF F  q&1 2 2a!899   fn%%A fn%%A#((MLv{{ !! $ $v{{  # # 7233>> index_map = make_2d_index_map( ... [0, 2, 4], [0, 3, 5], ... [[{0}, {1}], ... [{2}, {0, 1}]]) >>> pprint(index_map) {0: [(slice(0, 2, None), slice(0, 3, None)), (slice(2, 4, None), slice(3, 5, None))], 1: [(slice(0, 2, None), slice(3, 5, None)), (slice(2, 4, None), slice(3, 5, None))], 2: [(slice(2, 4, None), slice(0, 3, None))]} rr)sortedr8r%rangerTr_r) r r rrWiji_starti_stopj_startj_stoprcrCs rmake_2d_index_maprXs: ?a     #l## $ $ 4 4 4 4 ?a     #l## $ $ 4 4 4 446I w<<3|,,q0 0 0 0 0 3w<< : :71:#l"3"3a"77777s71:'' : :A"1oG!!A#&F"1oG!!A#&F&))5&+A+ABCqz!} : :$$S"--44S9999 : : r)rrkrI)" dataclassesr\rrrcupyx.distributed.arrayrrrrNr _SliceIndicesr^dict_BlockLocationMapr_ExecutionPlan _BatchIdx dataclassr r4rDrTrOrVrirrrrrrr8rrrrrs %%%%%%%% ******************c3m$  -. / DcN23eIy#567 -$ %          8?%8?%8?8?8?8?8?v  % &     FH %* H&+CHoH =# HHHH, -$ %, 5#:,,,,  c?'+CeE3J6G1H,H'I )& &'$= "= uSCZ()5+<< ==uUCZ()5+<< == ====*********04xx x%>x + ,xxxxxv1s)1s)1$s3x. !1 #tE%*%& &' 111111r