o i@sddlZddlZddlZddlZddlm Z ddlm Z ddl mZddlmZddlmZddlmZddlmZddlZGd d d eZd d Zd dZddZidddddddddddddddddddddd dfd!d"dd dfd#dd$dd%d&d'd&d(d&d)d&d&ddd*d*d&d&d&d+d&d,d,d-d.d/Zidd0dej dfdd0dd0dd0dd0dd0dd0dd0dd1d!d1d#d0d$d0d%d0d'd0d(d0d)d0d0d0d0ej dfej dfd0d0d0d0d0d1d1ej dfd2d/Z!d3d4Z"e#d5d6Z$d7e%fd8d9Z&ddZ(dd@dZ)ddAdZ*ddDdZ+ddEdZ,ddFdZ-ddGdZ.ddHdZ/ddId#Z0ddJd$Z1dKd%Z2dLd'Z3ddNd(Z4dOd)Z5dPdQZ6dRdSZ7dTdUZ8dVdWZ9dXdYZ:dZd[Z;d\d]ZdbdcZ?dddeZ@GdfdgdgeZAGdhdidieAZBGdjdkdkeAZCGdldmdmeAZDGdndodoeAZEddpdZFddqd!ZG ; ;ddrdsZHddtduZIddwdxZJddydzZKdd{d|ZLdd}d~ZMdS)N)driver)runtime)cusparse)_dtype)device)stream)_utilc@sVeZdZddZeddZddZejfddZ d d Z d d Z d dZ ddZ dS) MatDescriptorcCs ||_dSN) descriptor)selfr r K/home/ubuntu/veenaModal/venv/lib/python3.10/site-packages/cupyx/cusparse.py__init__ zMatDescriptor.__init__cCst}t|Sr ) _cusparsecreateMatDescrr )clsdescrr r rcreateszMatDescriptor.createcCs |jdfS)Nr )r)r r r r __reduce__rzMatDescriptor.__reduce__cCs*|rdS|jrt|jd|_dSdSr )r rdestroyMatDescrr is_shutting_downr r r__del__s   zMatDescriptor.__del__cCt|j|dSr )r setMatTyper )r typr r r set_mat_type%zMatDescriptor.set_mat_typecCrr )rsetMatIndexBaser )r baser r rset_mat_index_base(rz MatDescriptor.set_mat_index_basecCrr )rsetMatFillModer )r fill_moder r rset_mat_fill_mode+rzMatDescriptor.set_mat_fill_modecCrr )rsetMatDiagTyper )r diag_typer r rset_mat_diag_type.rzMatDescriptor.set_mat_diag_typeN)__name__ __module__ __qualname__r classmethodrrrrrrr"r%r(r r r rr s  r cs.dd|D}ttj|fdd|DS)NcSsg|] }|dur|jqSr dtype.0xr r r 3sz%_cast_common_type..cs,g|]}|dur|jkr|n|qSr )r.astyper/r-r rr25s$) _functoolsreduce_numpy promote_types)xsdtypesr r-r_cast_common_type2s  r:cCs|rtjStjSr )rCUSPARSE_OPERATION_TRANSPOSE CUSPARSE_OPERATION_NON_TRANSPOSE)transr r r_transpose_flag9sr>cGsR|dkrd}n|dkrd}n|dkrd}n |dkrd}nttt||}||S)NfsdFcDz) TypeErrorgetattrr)namer.argsprefixr?r r r_call_cusparse@srKcsrmv)@*csrmvExcsrmmcsrmm2csrgeamcsrgeam2)<#Ncsrgemmcsrgemm2)rM.gthrspmv'rN)LinuxWindowsspmmi=( csr2dense csc2densecsrsort)rMNcscsortcoosortcoo2csr)rZN)rTrW)i$,N)i\+N)iP-N)csr2coocsr2csccsc2csr csr2cscEx2 csc2csrEx2 dense2csc dense2csrcsr2csr_compresscsrsm2csrilu02 denseToSparse sparseToDensespgemmspsm)i1N)N)iNcCs8t|trt}||vrd|}t|||S|S)Nz/No version information specified for the OS: {}) isinstancedict _platformsystemformat ValueError)r1os_namemsgr r r_get_avail_version_from_specs r{cCstjs t}t}nt}t}||vrd|}t|||\}}t |}t |}|dur5||kr5dS|dur?||kr?dSdS)Nz1No available version information specified for {}FT) _runtimeis_hip_available_cusparse_versionrget_build_version_available_hipsparse_version_driverrwrxr{)rHavailable_versionversionrz version_addedversion_removedr r rcheck_availabilitys    rreturncCsttSr )r getVersion_deviceget_cusparse_handler r r rrsrFc Cs tdstd|dus|jjsJ|s|jn|jddd}|dt|kr+tdt}|\}} t |||\}}}|j } |durIt || }t || j}t || j}td| |t||jd|jd|j|j|jj|jjj|jjj|jjj|jj|j|jj|S)a\Matrix-vector product for a CSR-matrix and a dense vector. .. math:: y = \alpha * o_a(A) x + \beta y, where :math:`o_a` is a transpose function when ``transa`` is ``True`` and is an identity function otherwise. Args: a (cupyx.cusparse.csr_matrix): Matrix A. x (cupy.ndarray): Vector x. y (cupy.ndarray or None): Vector y. It must be F-contiguous. alpha (float): Coefficient for x. beta (float): Coefficient for y. transa (bool): If ``True``, transpose of ``A`` is used. Returns: cupy.ndarray: Calculated ``y``. rLzcsrmv is not available.Nrdimension mismatchr)r RuntimeErrorflags f_contiguousshapelenrxrrr:r._cupyzerosr6arrayctypesrKr>nnzdata_descrr ptrindptrindices) ar1yalphabetatransaa_shapehandlemnr.r r rrLs, cCsv|jjjddkr dS|jjjddkrdS|jjjddkr!dS|jjddkr+dS|dur9|jjddkr9dSdS)aCheck if the pointers of arguments for csrmvEx are aligned or not Args: a (cupyx.cusparse.csr_matrix): Matrix A. x (cupy.ndarray): Vector x. y (cupy.ndarray or None): Vector y. Check if a, x, y pointers are aligned by 128 bytes as required by csrmvEx. Returns: bool: ``True`` if all pointers are aligned. ``False`` if otherwise. rFNT)rrrr)rr1rr r rcsrmvExIsAlignedsrTcCstdstd|dus|jjsJ|jdt|krtdt}|j\}}t |||\}}}|j } |dur>t || }t | } |rHtjntj} tj} t|| j}t|| j}t|||sdJt|| | |jd|jd|j|j| |jj|jjj| |jjj|jjj|jj| |j| |jj| | } t | d}|jjddksJt || | |jd|jd|j|j| |jj|jjj| |jjj|jjj|jj| |j| |jj| | |jj|S) aMatrix-vector product for a CSR-matrix and a dense vector. .. math:: y = \alpha * A x + \beta y, Args: a (cupyx.cusparse.csr_matrix): Matrix A. x (cupy.ndarray): Vector x. y (cupy.ndarray or None): Vector y. It must be F-contiguous. alpha (float): Coefficient for x. beta (float): Coefficient for y. merge_path (bool): If ``True``, merge path algorithm is used. All pointers must be aligned with 128 bytes. Returns: cupy.ndarray: Calculated ``y``. rOzcsrmvEx is not available.Nrrrbr)!rrrrrrrxrrr:r.rrr to_cuda_dtyperCUSPARSE_ALG_MERGE_PATHCUSPARSE_ALG_NAIVEr<r6rrrcsrmvEx_bufferSizerrrr rrremptyrO)rr1rrr merge_pathrrrr.datatypealgmode transa_flag bufferSizebufr r rrOsN     c CsXtdstd|j|jkrdksJJ|jjsJ|dus(|jjs(J|s-|jn|jddd}|d|jdkrBtdt}|\}} |jd} t |||\}}}|durft || f|j d }| } |} t ||j j}t ||j j}td|j |t||jd| |jd|j|j|jj|jjj|jjj|jjj|jj| |j|jj| |S) a|Matrix-matrix product for a CSR-matrix and a dense matrix. .. math:: C = \alpha o_a(A) B + \beta C, where :math:`o_a` is a transpose function when ``transa`` is ``True`` and is an identity function otherwise. Args: a (cupyx.scipy.sparse.csr): Sparse matrix A. b (cupy.ndarray): Dense matrix B. It must be F-contiguous. c (cupy.ndarray or None): Dense matrix C. It must be F-contiguous. alpha (float): Coefficient for AB. beta (float): Coefficient for C. transa (bool): If ``True``, transpose of A is used. Returns: cupy.ndarray: Calculated C. rPzcsrmm is not available.NrrrrrB)rrndimrrrrxrrr:rrr.r6rrrKr>rrrr rrr) rrrCrrrrrrkrldbldcr r rrPTs6   ?cCstdstd|j|jkrdksJJ|jsJ|jjs#J|dus-|jjs-J|r3|r3J|s8|jn|jddd}|sE|jn|jddd}|d|dkrYtdt } |\} } |d} t |||\}}}|dur|t | | f|j d }|jd} |jd}t|}t|}t||j j}t||j j}td|j | |||jd| |jd|j|j|jj|jjj|jjj|jjj|jj| |j|jj||S) aVMatrix-matrix product for a CSR-matrix and a dense matrix. .. math:: C = \alpha o_a(A) o_b(B) + \beta C, where :math:`o_a` and :math:`o_b` are transpose functions when ``transa`` and ``tranb`` are ``True`` respectively. And they are identity functions otherwise. It is forbidden that both ``transa`` and ``transb`` are ``True`` in cuSPARSE specification. Args: a (cupyx.scipy.sparse.csr): Sparse matrix A. b (cupy.ndarray): Dense matrix B. It must be F-contiguous. c (cupy.ndarray or None): Dense matrix C. It must be F-contiguous. alpha (float): Coefficient for AB. beta (float): Coefficient for C. transa (bool): If ``True``, transpose of A is used. transb (bool): If ``True``, transpose of B is used. Returns: cupy.ndarray: Calculated C. rQzcsrmm2 is not available.rNrrrrrB)rrrhas_canonical_formatrrrrxrrr:rrr.r>r6rrrKrrrr rrr)rrrCrrrtransbrb_shaperrrrrrop_aop_br r rrQs>      c Cstdstdt|tjjjstdt |t|tjjjs*tdt ||j s/J|j s4J|j |j kr>t dt }|j \}}t||\}}tdd}t|tjt}t|dd} t||||jj|j|jjj|jjj|jj|j|jjj|jjj|j| jj|j jtt!|d} tt!||j"} t#||j"j }t#||j"j }t$d|j"||||j|jj|j|jjj|jjj|jjj|j|jj|j|jjj|jjj|jjj|j| jj| jj| jjtjjj| | | f|j d} d | _%| S) eMatrix-matrix addition. .. math:: C = \alpha A + \beta B Args: a (cupyx.scipy.sparse.csr_matrix): Sparse matrix A. b (cupyx.scipy.sparse.csr_matrix): Sparse matrix B. alpha (float): Coefficient for A. beta (float): Coefficient for B. Returns: cupyx.scipy.sparse.csr_matrix: Result matrix. rRzcsrgeam is not available.unsupported type (actual: {})inconsistent shapesr irrT)&rrrscupyxscipysparse csr_matrixrFrwtyperrrxrrr:r6rrsetPointerModeCUSPARSE_POINTER_MODE_HOSTr rr xcsrgeamNnzrr rrrrrrintr.rrK_has_canonical_format) rrrrrrrrc_descrc_indptr c_indicesc_datarCr r rrRsX       cCstdstdt|tjjjstdt |t|tjjjs*tdt ||j s/J|j s4J|j |j kr>t dt }|j \}}t||\}}tdd}t|tjt||jj}t||jj}t}t|dd} d} td |j||||j|jj|j|jjj |j!jj |j"jj |j|jj|j|jjj |j!jj |j"jj |j| | jj | } t| tj#} t$||||jj|j|j!jj |j"jj |jj|j|j!jj |j"jj |j| jj |jj| jj tt%|d} tt%||j}td|j||||j|jj|j|jjj |j!jj |j"jj |j|jj|j|jjj |j!jj |j"jj |j|jj | jj | jj | jj tjjj|| | f|j d }d |_&|S) rrSzcsrgeam2 is not available.rrr rrrcsrgeam2_bufferSizeExtrT)'rrrsrrrrrFrwrrrrxrrr:r6rrrrrr.rr rrrKrrr rrrrint8 xcsrgeam2Nnzrr)rrrrrrrrrrnull_ptr buff_sizebuffrrrCr r rrS sf        cCs6tdstd|j|jkrdksJJ|jsJ|js"J|s'|jn|jddd}|s4|jn|jddd}|d|dkrHtdt}|\}}|d} t||\}}|j dkse|j dkrqt j j j || f|jd St|} t|} td d } t|tjt} t|dd }t|| | || ||jj|j |jjj|jjj|jj|j |jjj|jjj| j|jj| jjtt | d }tt | |j}t!d|j|| | || ||jj|j |jjj|jjj|jjj|jj|j |jjj|jjj|jjj| j|jj|jj|jjt j j j |||f|| fd }d |_"|S)aMatrix-matrix product for CSR-matrix. math:: C = op(A) op(B), Args: a (cupyx.scipy.sparse.csr_matrix): Sparse matrix A. b (cupyx.scipy.sparse.csr_matrix): Sparse matrix B. transa (bool): If ``True``, transpose of A is used. transb (bool): If ``True``, transpose of B is used. Returns: cupyx.scipy.sparse.csr_matrix: Calculated C. rUzcsrgemm is not available.rNrrrrr-r rrT)#rrrrrrxrrr:rrrrrr.r>r6rrrrr rr xcsrgemmNnzrr rrrrrrrKr)rrrrrrrrrrrrrrrrrrCr r rrUPsZ      cCstdstd|j|jkrdksJJt|tjjjs)td t |t|tjjjs:td t ||j s?J|j sDJ|j d|j dkrRt d|dur|jdks]Jt|tjjjsntd t ||j ssJ|j d|j dks|j d|j dkrt dtjrtd krtd t}|j \}}|j \}} |durt||\}}n t|||\}}}t} t||jj}d} |dur| } t} d}| }| }| }n t||jj}|j} |j} |j}|jjj }|j!jj }|j"jj }t#d |j||| ||j|jj$|j|j!jj |j"jj |jj$|j|j!jj |j"jj | | j$|||| }t%&|tj'}t&d d }t(|tj)t}t%&|dd }t*||| ||jj$|j|j!jj |j"jj |jj$|j|j!jj |j"jj | j$||||j$|jj |jj| |jj t%&t+|d }t%&t+||j}t#d|j||| ||j|jj$|j|jjj |j!jj |j"jj |jj$|j|jjj |j!jj |j"jj | | j$|||||j$|jj |jj |jj | |jj tjjj|||f|| fd}d|_,t-| |S)aMatrix-matrix product for CSR-matrix. math:: C = alpha * A * B + beta * D Args: a (cupyx.scipy.sparse.csr_matrix): Sparse matrix A. b (cupyx.scipy.sparse.csr_matrix): Sparse matrix B. d (cupyx.scipy.sparse.csr_matrix or None): Sparse matrix D. alpha (scalar): Coefficient beta (scalar): Coefficient Returns: cupyx.scipy.sparse.csr_matrix rVzcsrgemm2 is not available.rrrrmismatched shapeNrrz'd != None is supported since ROCm 4.2.0csrgemm2_bufferSizeExtr rrT).rrrrsrrrrrFrwrrrrxr|r}rrrrr:rcreateCsrgemm2Infor6rr.rr rrrrrrrrKr rrrrr xcsrgemm2NnzrrdestroyCsrgemm2Info)rrrArrrrr_rinfor beta_datad_descrd_nnzd_datad_indptr d_indicesrrc_nnzrrrrrCr r rrVs    (        c Ctdstd|j}|jdvsJ|dur tj|j|dd}n|jjs&Jt }t d|j||jd|jd|j j |jjj|jjj|jjj|jj|jd |S) a!Converts CSR-matrix to a dense matrix. Args: x (cupyx.scipy.sparse.csr_matrix): A sparse matrix to convert. out (cupy.ndarray or None): A dense metrix to store the result. It must be F-contiguous. Returns: cupy.ndarray: Converted result. r^zcsr2dense is not available.fdFDNrBr.orderrr)rrr.charrrrrrrrrKrr rrrrr1outr.rr r rr^  c Cr) a!Converts CSC-matrix to a dense matrix. Args: x (cupyx.scipy.sparse.csc_matrix): A sparse matrix to convert. out (cupy.ndarray or None): A dense metrix to store the result. It must be F-contiguous. Returns: cupy.ndarray: Converted result. r_zcsc2dense is not available.rNrBrrr)rrr.rrrrrrrrrKrr rrrrrr r rr_rc Ctdstd|j}|dkrdSt}|j\}}t|||||jj j |j j j }t |d}t |d}|j }t|||j j t|||||jj|jj j |j j j |j j |j j tdrwtd|j|||j j |j j j |j j tjdSt||j } t|} t|| j| jdS)zzSorts indices of CSR-matrix in place. Args: x (cupyx.scipy.sparse.csr_matrix): A sparse matrix to sort. r`zcsrsort is not available.rNrrrX)rrrrrrrxcsrsort_bufferSizeExtrrrrrrcopycreateIdentityPermutationxcsrsortrr rKr.CUSPARSE_INDEX_BASE_ZEROSpVecDescriptorrDnVecDescriptorgatherdesc r1rrrr buffer_sizerP data_origdesc_xdesc_yr r rr`>8      c Cr)zzSorts indices of CSC-matrix in place. Args: x (cupyx.scipy.sparse.csc_matrix): A sparse matrix to sort. razcscsort is not available.rNrrrX)rrrrrrrxcscsort_bufferSizeExtrrrrrrrrxcscsortrr rKr.rrrrrrrr r rradrrc Csztdstd|j}|dkrdSt}|j\}}t|||||jj j |j j j }t |d}t |d}|j } t|||j j |dkr_t|||||jj j |j j j |j j |j j n|dkrzt|||||jj j |j j j |j j |j j ntd |jjd krtd rtd |j||| j j |j j j |j j tjnt||j } t| } t|| j| j|dkrd |_dSdS) zSorts indices of COO-matrix in place. Args: x (cupyx.scipy.sparse.coo_matrix): A sparse matrix to sort. sort_by (str): Sort the indices by row ('r', default) or column ('c'). rbzcoosort is not available.rNrrrrCz!sort_by must be either 'r' or 'c'?rXF)rrrrrrrxcoosort_bufferSizeExtrowrrcolrrrr xcoosortByRowxcoosortByColumnrxr.rrKrrrrrrr) r1sort_byrrrrrrrrrrr r rrbsL          cCst}|jd}|j}|dkrt|dd}nt|dd}t||j j j |||j j tj t jjj|j |j|f|jdS)Nrrrr)rrrrrrrrxcoo2csrrrrrrrrrr)r1rrrrr r rrc  cCst}|jd}|j}|dkrt|dd}nt|dd}t||j j j |||j j tj t jjj|j |j|f|jdS)Nrrrr)rrrrrrrrr rrrrrrr csc_matrixr)r1rrrrr r rcoo2cscr rcCsptdstdt}|jd}|j}t|d}t ||j j j |||j j tj tjjj|||ff|jdS)aDConverts a CSR-matrix to COO format. Args: x (cupyx.scipy.sparse.csr_matrix): A matrix to be converted. data (cupy.ndarray): A data array for converted data. indices (cupy.ndarray): An index array for converted data. Returns: cupyx.scipy.sparse.coo_matrix: A converted matrix. rdzcsr2coo is not available.rrr)rrrrrrrrrxcsr2coorrrrrrr coo_matrix)r1rrrrrrr r rrds   rdcCstdstdt}|j\}}|j}t||j}t|d}|dkr.t |dd}n+t|dd}t d|j|||||j j j |j j j |jj j |j j |j j |j j tjtjtjjj|||f|jdS)Nrecsr2csc is not available.rrrr)rrrrrrrrr.rrKrrrrrCUSPARSE_ACTION_NUMERICrrrrr r1rrrrrrrr r rre*   recCs<tdstdt}|j\}}|j}t||j}t|d}|dkr.t |dd}nct|dd}t |j}t j } t j} t j} t |||||jjj|jjj|jjj|jj|jj|jj|| | | } t| tj} t |||||jjj|jjj|jjj|jj|jj|jj|| | | | jjtjjj|||f|jdS)Nrgzcsr2cscEx2 is not available.rrrr)rrrrrrrrr.rrrrrrCUSPARSE_CSR2CSC_ALG1csr2cscEx2_bufferSizerrrrr6rrgrrrr r1rrrrrrrx_dtypeactionibasealgorbufferr r rrg 8    rgcCs`t}|jd}|j}t|d}t||jj j |||j j tj t j jj|||ff|jdS)aDConverts a CSC-matrix to COO format. Args: x (cupyx.scipy.sparse.csc_matrix): A matrix to be converted. data (cupy.ndarray): A data array for converted data. indices (cupy.ndarray): An index array for converted data. Returns: cupyx.scipy.sparse.coo_matrix: A converted matrix. rrr)rrrrrrrrrrrrrrrr)r1rrrrrrr r rcsc2coo*s  rcCstdstdt}|j\}}|j}t||j}t|d}|dkr.t |dd}n+t|dd}t d|j|||||j j j |j j j |jj j |j j |j j |j j tjtjtjjj|||f|jdS)Nrfrrrrrer)rrrrrrrrr.rrKrrrrrrrrrrrrr r rrfBrrfcCs<tdstdt}|j\}}|j}t||j}t|d}|dkr.t |dd}nct|dd}t |j}t j } t j} t j} t |||||jjj|jjj|jjj|jj|jj|jj|| | | } t| tj} t |||||jjj|jjj|jjj|jj|jj|jj|| | | | jjtjjj|||f|jdS)Nrhzcsc2csrEx2 is not available.rrrr)rrrrrrrrr.rrrrrrrrrrrrr6rrgrrrrrr r rrhZrrhc Cstdstd|jdksJt|}tjddd}t}|j \}}t }t|d}t d|j |tj|||j|jj||jj|jj t|}t||j }t|dd}t|d} t d|j ||||j|jj||jj|jj| jj|jj tjjj|| |f|j d } d | _| S) zConverts a dense matrix in CSC format. Args: x (cupy.ndarray): A matrix to be converted. Returns: cupyx.scipy.sparse.csc_matrix: A converted matrix. rizdense2csc is not available.rr rr-rrrT)rrrrasfortranarrayr6rrrrr rrKr.rCUSPARSE_DIRECTION_COLUMNr rrrrrrrr r) r1rrrrr nnz_per_colrrrcscr r rrixs6      ric Cs,tdstd|jdksJt|}tjddd}t}|j \}}t }t|d}t d|j |tj|||j|jj||jj|jj t|}tjrT|dkrTtd t||j }t|d d}t|d} t d|j ||||j|jj||jj|jj|jj| jj tjjj|| |f|j d } d | _| S) zConverts a dense matrix in CSR format. Args: x (cupy.ndarray): A matrix to be converted. Returns: cupyx.scipy.sparse.csr_matrix: A converted matrix. rjzdense2csr is not available.rr rr-rr@hipSPARSE currently cannot handle sparse matrices with null ptrsrrT)rrrrrr6rrrrr rrKr.rCUSPARSE_DIRECTION_ROWr rrrrr|r}rxrrrrr) r1rrrrr nnz_per_rowrrrcsrr r rrjs<      rjc Cstdstd|jjdvsJt}|j\}}t|d}t d|j|||j j |j j j |jj j |j j | }t||j}t|dd}t|d} t d|j||||j j |j j j |jj j |jj j |j|j j |j j | j j |j j |tjjj|| |f|jdS)Nrkz"csr2csr_compress is not available.rr nnz_compressrr)rrr.rrrrrrrKrr rrrrrrrrrr) r1tolrrrr%rrrrr r rrks0      rkcCs.|dkrtjS|dkrtjS|dkrtjSt)Nuint16int32int64)rCUSPARSE_INDEX_16UCUSPARSE_INDEX_32ICUSPARSE_INDEX_64IrFr-r r r_dtype_to_IndexTypesr/c@s,eZdZdddZejfddZddZdS) BaseDescriptorNcCs||_||_||_dSr )rgetdestroy)r r r1 destroyerr r rrs zBaseDescriptor.__init__cCsB|rdS|jdurd|_dS|jdur||jd|_dSdSr )r2rrr r rrs     zBaseDescriptor.__del__cCs |jdurt||j|Str )r1rGrAttributeError)r rHr r r __getattr__s zBaseDescriptor.__getattr__)NN)r)r*r+rrrrr5r r r rr0s  r0c@s eZdZeddZddZdS)SpMatDescriptorc Cs:tjj|s J|j\}}tj}t|j }|j dkr@t |||j |j jj|jjj|jjjt|j j t|jj || }tj}nT|j dkrdt|||j |jjj|jjj|jjjt|jj || }tj}n0|j dkrt|||j |j jj|jjj|jjjt|j j t|jj || }d}ntd |j tj}t|||S)Nr&coor"z3csr, csc and coo format are supported (actual: {}).)rrrissparserrrrrr.rw createCsrrrrrrr/csrGet createCoorrcooGet createCscrx destroySpMatr6) rrrowscolsidx_base cuda_dtyperr1r2r r rrs<      zSpMatDescriptor.createcCst|j||dSr )rspMatSetAttributer)r attributerr r r set_attribute-szSpMatDescriptor.set_attributeN)r)r*r+r,rrEr r r rr6 s  r6c@eZdZeddZdS)rc CsP|j}t|j}t|||jj|jjt|jtj |}tj }tj }t |||Sr ) sizerrr.r createSpVecrrr/rspVecGet destroySpVecr)ridxr1rrBrr1r2r r rr3s  zSpVecDescriptor.createNr)r*r+r,rr r r rr1rc@rF)rcCs8t|j}t|j|jj|}tj}tj }t |||Sr ) rrr.r createDnVecrGrrdnVecGet destroyDnVecr)rr1rBrr1r2r r rrBs  zDnVecDescriptor.createNrLr r r rr@rMrc@rF)DnMatDescriptorc Csf|jdksJ|jjs J|j\}}|}t|j}t||||j j |tj }tj }tj }t|||S)Nr)rrrrrrr.r createDnMatrrCUSPARSE_ORDER_COLdnMatGet destroyDnMatrQ) rrr?r@ldrBrr1r2r r rrMs    zDnMatDescriptor.createNrLr r r rrQKrMrQc Cstdstdt|tjjjr+|j}t|tjjjs&d t |}t ||}| }t|tjjjtjjj fsAt d t ||sF|j n|j ddd}|dt|krZtd|js_J|\} } t|||\}}}|durxt| |j}n t|| krtd|jd kr|d |St|} t|} t|} t}t|}t||jj}t||jj}t !|j}t"j#}t"$|||j%| j&| j&|j%| j&|| }t'|tj(}t")|||j%| j&| j&|j%| j&|||j%j* |S) aMultiplication of sparse matrix and dense vector. .. math:: y = \alpha * op(A) x + \beta * y Args: a (cupyx.scipy.sparse.csr_matrix, csc_matrix or coo_matrix): Sparse matrix A x (cupy.ndarray): Dense vector x y (cupy.ndarray or None): Dense vector y alpha (scalar): Coefficient beta (scalar): Coefficient transa (bool): If ``True``, op(A) = transpose of A. Returns: cupy.ndarray rYzspmv is not available."aT must be csr_matrix (actual: {})rNrrrr)+rrrsrrrr TrrwrrFrrrrxrr:rrr.rfillr6rrrrr>r6rrrrrCUSPARSE_MV_ALG_DEFAULTspMV_bufferSizerrrrspMVr)rr1rrrraTrzrrrdesc_arrrrrBalgrrr r rrY[sX           c Cs`tdstd|j|jkrdksJJ|jjsJ|dus(|jjs(Jt|tjjj rK|j }t|tjjj sFd t |}t||}| }t|tjjj tjjjfsatd t ||sf|jn|jddd} |ss|jn|jddd} | d| d krtd |jsJ| \} } | \} }t|||\}}}|durt| |f|jd }n|jd | ks|jd|krtd |jd kr|d |St|}t|}t|}t}t|}t|}t ||jj!}t ||jj!}t"#|j}t$j%}t$&||||j'|j(|j(|j'|j(|| }t)|tj*}t$+||||j'|j(|j(|j'|j(|||j'j, }|S) aMultiplication of sparse matrix and dense matrix. .. math:: C = \alpha * op(A) op(B) + \beta * C Args: a (cupyx.scipy.sparse.csr_matrix, csc_matrix or coo_matrix): Sparse matrix A b (cupy.ndarray): Dense matrix B c (cupy.ndarray or None): Dense matrix C alpha (scalar): Coefficient beta (scalar): Coefficient transa (bool): If ``True``, op(A) = transpose of A. transb (bool): If ``True``, op(B) = transpose of B. Returns: cupy.ndarray r]zspmm is not available.rNrWrrrrrrB)-rrrrrrsrrrr rXrrwrrFrrrxrr:rrr.rrYr6rrQrrr>r6rrrrrCUSPARSE_MM_ALG_DEFAULTspMM_bufferSizerrrrspMMr)rrrCrrrrr]rzrrrrrrr^desc_bdesc_crrrrBr_rrr r rr]sf             cCs6tdstdtjj|stjj|stdt|t j s$td|j dvr-td|j d|j dkrC|j dksHtd td |j |j krRtd |d urZtj}n|d urbtj}ntd ||d urqtj} n|d urytj} ntd||d urd} n|d urd} ntd||d urtj} n|d urtj} ntd||j } | jdkrd} n| jdkrd} n| jdkrd} n| jdkrd} ntd| tt| d}tt| d}tt| d}|d us|dkrtj}n(|d us|dkrtj}n|dkr| jdvrtj}n tj}ntd |tjj|rF|tjkr1td!|j}tjj|s>Jd|}d|}|j d}|j dkrSdn|j d}|jrbtj}|}n|jrltj}|}ntd"t }t!j"|| d#}t#$}|%tj&|'tj(|)||*| t+}||| |||||j,|j-j.|j/|j.j.j0|j1j.j0|j2j.j0|j.j0||| }t j3|ft!j4d#}||| |||||j,|j-j.|j/|j.j.j0|j1j.j0|j2j.j0|j.j0||| |j.j0||| |||||j,|j-j.|j/|j.j.j0|j1j.j0|j2j.j0|j.j0||| |j.j0t567t8|d$S)%aVSolves a sparse triangular linear system op(a) * x = alpha * b. Args: a (cupyx.scipy.sparse.csr_matrix or cupyx.scipy.sparse.csc_matrix): Sparse matrix with dimension ``(M, M)``. b (cupy.ndarray): Dense vector or matrix with dimension ``(M)`` or ``(M, K)``. alpha (float or complex): Coefficient. lower (bool): True: ``a`` is lower triangle matrix. False: ``a`` is upper triangle matrix. unit_diag (bool): True: diagonal part of ``a`` has unit elements. False: diagonal part of ``a`` has non-unit elements. transa (bool or str): True, False, 'N', 'T' or 'H'. 'N' or False: op(a) == ``a``. 'T' or True: op(a) == ``a.T``. 'H': op(a) == ``a.conj().T``. blocking (bool): True: blocking algorithm is used. False: non-blocking algorithm is used. level_info (bool): True: solves it with level information. False: solves it without level information. Note: ``b`` will be overwritten. rlzcsrsm2 is not available.z"a must be CSR or CSC sparse matrixzb must be cupy.ndarray)rrb.ndim must be 1 or 2rrz invalid shapedtype mismatchTFUnknown lower (actual: {})Unknown unit_diag (actual: {})zUnknown blocking (actual: {})Unknown level_info (actual: {})r?r@rArBrCrDrEInvalid dtype (actual: {})csrsm2_bufferSizeExtcsrsm2_analysis csrsm2_solveNrXHfdzUnknown transa (actual: {})z@If matrix is CSC format and complex dtype,transa must not be 'H''b must be F-contiguous or C-contiguous.r-N)9rrrrrisspmatrix_csrisspmatrix_cscrxrsrndarrayrrr.rFrCUSPARSE_FILL_MODE_LOWERCUSPARSE_FILL_MODE_UPPERrwCUSPARSE_DIAG_TYPE_NON_UNITCUSPARSE_DIAG_TYPE_UNITCUSPARSE_SOLVE_POLICY_NO_LEVELCUSPARSE_SOLVE_POLICY_USE_LEVELrrGr<r;&CUSPARSE_OPERATION_CONJUGATE_TRANSPOSErX _f_contiguous _c_contiguousrrr6rr rrCUSPARSE_MATRIX_TYPE_GENERALr"rr%r(createCsrsm2Inforrrr rrrrr_streamget_current_stream synchronizedestroyCsrsm2Info)rrrlower unit_diagrblocking level_infor$r'rpolicyr.thelperanalysissolvernrhsrrra_descrws_sizewsr r rrls   $              rlc Cs`tdstdtjj|std|jd|jdkr%td |j|dur-t j }n|dur5t j }ntd ||j }|jd krGd }n|jd krOd }n|jd krWd}n|jdkr_d}ntd |tt |d}tt |d}tt |d}tt d}t} |jd} |j} t} | t j| t jt } || | | | j|jjj|jjj|jjj| }tj |ft!j"d}t!j dt!j#d}|| | | | j|jjj|jjj|jjj| ||jj z || | |j$jWnt%ytd |dw|| | | | j|jjj|jjj|jjj| ||jj z || | |j$jWdSt%y/td |dw)aComputes incomplete LU decomposition for a sparse square matrix. Args: a (cupyx.scipy.sparse.csr_matrix): Sparse matrix with dimension ``(M, M)``. level_info (bool): True: solves it with level information. False: solves it without level information. Note: ``a`` will be overwritten. This function does not support fill-in (only ILU(0) is supported) nor pivoting. rmzcsrilu02 is not available.za must be CSR sparse matrixrrzinvalid shape (a.shape: {})FTrir?r@rArBrCrDrErjcsrilu02_bufferSizecsrilu02_analysisxcsrilu02_zeroPivotr-rza({0},{0}) is missingzu({0},{0}) is zeroN)&rrrrrrrrFrrxrwrryrzr.rrGrrrr rrr~r"rcreateCsrilu02Infor rrrrrrr6rr*r Exception)rrrr.rrrrcheckrrrrrrrpositionr r rrm|sj          rmr&cCsltdstd|jdksJ|jjdvsJt|}t|}|dkr1t j j j |j |jd}n'|dkrAt j j j|j |jd}n|dkrQt j j j|j |jd}ntd |t|}tj}t}t||j|j|}t|tj}t||j|j||jjtjd d d} tjd d d} tjd d d} t |j| j!j| j!j| j!jt"| } t#j$r| d krt%d |dkr|j&} t| d } t| |j}t j j j || | f|j d}nH|dkr|j&} t| d } t| |j}t j j j|| | f|j d}n&|dkrt'| d }t'| d }t| |j}t j j j|||ff|j d}t|}t(||j|j||jjd|_)|S)a0Converts a dense matrix into a CSR, CSC or COO format. Args: x (cupy.ndarray): A matrix to be converted. format (str): Format of converted matrix. It must be either 'csr', 'csc' or 'coo'. Returns: cupyx.scipy.sparse.spmatrix: A converted sparse matrix. rnzdenseToSparse is not available.rrr&r-r"r7zunsupported format (actual: {})rr+r#rrT)*rrrr.rrrrQrrrrrrr rrFrwr6r"CUSPARSE_DENSETOSPARSE_ALG_DEFAULTrrdenseToSparse_bufferSizerrrdenseToSparse_analysisrrr6r spMatGetSizerrr|r}rxrrdenseToSparse_convertr)r1rwrrrrrrr num_rows_tmp num_cols_tmprrrrrrr r rrnst                rnc Cstdstd|j}|jdvsJ|dur tj|j|dd}n |jjs&J|j|ks-Jt |}t |}t j }t}t ||j|j|}t|tj}tjr[|jdkr[tdt ||j|j||jj|S) a*Converts sparse matrix to a dense matrix. Args: x (cupyx.scipy.sparse.spmatrix): A sparse matrix to convert. out (cupy.ndarray or None): A dense metrix to store the result. It must be F-contiguous. Returns: cupy.ndarray: A converted dense matrix. rozsparseToDense is not available.rNrBrrr#)rrr.rrrrrrr6rrQr"CUSPARSE_SPARSETODENSE_ALG_DEFAULTrrsparseToDense_bufferSizerrrr|r}rrxrorr) r1rr.rdesc_outrrrrr r rro s.        roc Cstdstd|durd}n|durd}n |dvr"td|d tjj|r*n6tjj|rT|dkr;|j}d}n|dkrE|j}d}n |d krP| j}d}| }n tjj |r\ntd |j seJ|j d krsd}| d d }n |j dkr{d}ntd|jd|jd kr|jdkstdtd|j}|jdvrtd|||jkrtd|durtj}n|durtj}ntd||durtj} n|durtj} ntd||dkrtj} n|dkrtj} n |jdvrtj} ntj} |jrtj} n|jrtdkrtd|j}tj} ntd|j\} } | tjkr/|j\} }n|j\}} | |f}tj||jdd}t !}t"#|}t$#|}t$#|}t%}t&j'||jdj(}t)*|j}tj+}zb|,tj-||,tj.| t/|| | |j0|j1|j1|j1||| }tj2|tj3d}t4|| | |j0|j1|j1|j1||||j0j5 t6|| | |j0|j1|j1|j1||||j0j5 |r| d }|Wt7|St7|w)aSolves a sparse triangular linear system op(a) * x = alpha * op(b). Args: a (cupyx.scipy.sparse.csr_matrix or cupyx.scipy.sparse.coo_matrix): Sparse matrix with dimension ``(M, M)``. b (cupy.ndarray): Dense matrix with dimension ``(M, K)``. alpha (float or complex): Coefficient. lower (bool): True: ``a`` is lower triangle matrix. False: ``a`` is upper triangle matrix. unit_diag (bool): True: diagonal part of ``a`` has unit elements. False: diagonal part of ``a`` has non-unit elements. transa (bool or str): True, False, 'N', 'T' or 'H'. 'N' or False: op(a) == ``a``. 'T' or True: op(a) == ``a.T``. 'H': op(a) == ``a.conj().T``. rqzspsm is not available.FrnTrXNTHzUnknown transa (actual: )roz'a must be CSR, CSC or COO sparse matrixrrrrerrrrjrfrgrhrpi-zb must be F-contiguous.rqr?rr-)8rrrxrrrrrrsrXconjisspmatrix_coorrreshaperr.rrFrwrrurvrwrxr<r;r{r|r}rrrrrr6rrQspSM_createDescrr6rrrrCUSPARSE_SPSM_ALG_DEFAULTrECUSPARSE_SPMAT_FILL_MODECUSPARSE_SPMAT_DIAG_TYPEspSM_bufferSizerrrr spSM_analysisr spSM_solvespSM_destroyDescr)rrrrrr is_b_vectorr.r$r'rrrrrc_shaperCrmat_amat_bmat_c spsm_descrrBrrrr r rrq3s    $             rqcCs2tdstd|j|jkrdksJJt|tjjjs)td t |t|tjjjs:td t ||j s?J|j sDJ|j d|j dkrRt d|j \}}|j \}}t||\}}||f}tjjj||jd}t} t|} t|} t|} t} tj}tj}tj||jdj}tjd|jdj}t|j}tj}d}t| |||j| j| j|j| j||| d| }t !|t j"}t| |||j| j| j|j| j||| ||jj# t$| |||j| j| j|j| j||| d| }t !|t j"}t$| |||j| j| j|j| j||| ||jj# tjdd d}tjdd d}tjdd d}t%| j|jj|jj|jj|dt&|ksDJ|dt&|ksOJt&|}|j'}t !|d }t !||j}t(| j|jj#|jj#|jj#t)| |||j| j| j|j| j||| tjjj|||f|d }t*| |S) a+Matrix-matrix product for CSR-matrix. math:: C = alpha * A * B Args: a (cupyx.scipy.sparse.csr_matrix): Sparse matrix A. b (cupyx.scipy.sparse.csr_matrix): Sparse matrix B. alpha (scalar): Coefficient Returns: cupyx.scipy.sparse.csr_matrix rpzspgemm is not available.rrrrrr-r+rr)+rrrrsrrrrrFrwrrrrxr:r.rrr6rrspGEMM_createDescrr<r6rrrrCUSPARSE_SPGEMM_DEFAULTspGEMM_workEstimationrrrrrrspGEMM_computerrrcsrSetPointers spGEMM_copyspGEMM_destroyDescr)rrrrrrrrrCrrrr spgemm_descrrrrrBrr buff1_sizebuff1 buff2_sizebuff2 c_num_rows c_num_colsrrrrr r rrps             rp)NrrFr )NrrT)NrrFF)rr)FF)Nrr)r)NrrFF)rTFFTF)F)r&)rTFFr)N functoolsr4numpyr6platformrucupyrcupy_backends.cuda.apirrrr|cupy_backends.cuda.libsrr cupy._corer cupy.cudarrrrrcupyx.scipy.sparserobjectr r:r>rKr~infrr{memoizerrrrLrrOrPrQrRrSrUrVr^r_r`rarbrcrrdrergrrfrhrirjrkr/r0r6rrrQrYr]rlrmrnrorqrpr r r rsF       !  '      %   2  D 8 ? @ E E i  & &1*. $  CL   G H (#