1 #if !defined(__CUSPARSEMATIMPL) 2 #define __CUSPARSEMATIMPL 3 4 #include <../src/vec/vec/impls/seq/seqcusp/cuspvecimpl.h> 5 6 #include <cusparse_v2.h> 7 8 #include <algorithm> 9 #include <vector> 10 #include <thrust/sort.h> 11 #include <thrust/fill.h> 12 #include <cusp/csr_matrix.h> 13 14 /* Single instance of the cusparse handle for the class. */ 15 MatCUSPARSEStorageFormat cusparseMatSolveStorageFormat=MAT_CUSPARSE_CSR; 16 17 #if defined(PETSC_USE_COMPLEX) 18 #if defined(PETSC_USE_REAL_SINGLE) 19 #define cusparse_solve cusparseCcsrsv_solve 20 #define cusparse_analysis cusparseCcsrsv_analysis 21 #define cusparse_csr_spmv cusparseCcsrmv 22 #define cusparse_csr2csc cusparseCcsr2csc 23 #define cusparse_hyb_spmv cusparseChybmv 24 #define cusparse_csr2hyb cusparseCcsr2hyb 25 #define cusparse_hyb2csr cusparseChyb2csr 26 cuFloatComplex ALPHA = {1.0f, 0.0f}; 27 cuFloatComplex BETA = {0.0f, 0.0f}; 28 #elif defined(PETSC_USE_REAL_DOUBLE) 29 #define cusparse_solve cusparseZcsrsv_solve 30 #define cusparse_analysis cusparseZcsrsv_analysis 31 #define cusparse_csr_spmv cusparseZcsrmv 32 #define cusparse_csr2csc cusparseZcsr2csc 33 #define cusparse_hyb_spmv cusparseZhybmv 34 #define cusparse_csr2hyb cusparseZcsr2hyb 35 #define cusparse_hyb2csr cusparseZhyb2csr 36 cuDoubleComplex ALPHA = {1.0, 0.0}; 37 cuDoubleComplex BETA = {0.0, 0.0}; 38 #endif 39 #else 40 PetscScalar ALPHA = 1.0; 41 PetscScalar BETA = 0.0; 42 #if defined(PETSC_USE_REAL_SINGLE) 43 #define cusparse_solve cusparseScsrsv_solve 44 #define cusparse_analysis cusparseScsrsv_analysis 45 #define cusparse_csr_spmv cusparseScsrmv 46 #define cusparse_csr2csc cusparseScsr2csc 47 #define cusparse_hyb_spmv cusparseShybmv 48 #define cusparse_csr2hyb cusparseScsr2hyb 49 #define cusparse_hyb2csr cusparseShyb2csr 50 #elif defined(PETSC_USE_REAL_DOUBLE) 51 #define cusparse_solve cusparseDcsrsv_solve 52 #define cusparse_analysis cusparseDcsrsv_analysis 53 #define cusparse_csr_spmv cusparseDcsrmv 54 #define cusparse_csr2csc cusparseDcsr2csc 55 #define cusparse_hyb_spmv cusparseDhybmv 56 #define cusparse_csr2hyb cusparseDcsr2hyb 57 #define cusparse_hyb2csr cusparseDhyb2csr 58 #endif 59 #endif 60 61 #define THRUSTINTARRAY32 thrust::device_vector<int> 62 #define THRUSTINTARRAY thrust::device_vector<PetscInt> 63 #define THRUSTARRAY thrust::device_vector<PetscScalar> 64 65 /* A CSR matrix structure */ 66 struct CsrMatrix { 67 PetscInt num_rows; 68 PetscInt num_cols; 69 PetscInt num_entries; 70 THRUSTINTARRAY32 *row_offsets; 71 THRUSTINTARRAY32 *column_indices; 72 THRUSTARRAY *values; 73 }; 74 75 //#define CUSPMATRIXCSR32 cusp::csr_matrix<int,PetscScalar,cusp::device_memory> 76 77 /* This is struct holding the relevant data needed to a MatSolve */ 78 struct Mat_SeqAIJCUSPARSETriFactorStruct { 79 /* Data needed for triangular solve */ 80 cusparseMatDescr_t descr; 81 cusparseSolveAnalysisInfo_t solveInfo; 82 cusparseOperation_t solveOp; 83 CsrMatrix *csrMat; 84 }; 85 86 /* This is struct holding the relevant data needed to a MatMult */ 87 struct Mat_SeqAIJCUSPARSEMultStruct { 88 void *mat; /* opaque pointer to a matrix. This could be either a cusparseHybMat_t or a CsrMatrix */ 89 cusparseMatDescr_t descr; /* Data needed to describe the matrix for a multiply */ 90 THRUSTINTARRAY *cprowIndices; /* compressed row indices used in the parallel SpMV */ 91 }; 92 93 /* This is a larger struct holding all the triangular factors for a solve, transpose solve, and 94 any indices used in a reordering */ 95 struct Mat_SeqAIJCUSPARSETriFactors { 96 Mat_SeqAIJCUSPARSETriFactorStruct *loTriFactorPtr; /* pointer for lower triangular (factored matrix) on GPU */ 97 Mat_SeqAIJCUSPARSETriFactorStruct *upTriFactorPtr; /* pointer for upper triangular (factored matrix) on GPU */ 98 Mat_SeqAIJCUSPARSETriFactorStruct *loTriFactorPtrTranspose; /* pointer for lower triangular (factored matrix) on GPU for the transpose (useful for BiCG) */ 99 Mat_SeqAIJCUSPARSETriFactorStruct *upTriFactorPtrTranspose; /* pointer for upper triangular (factored matrix) on GPU for the transpose (useful for BiCG)*/ 100 THRUSTINTARRAY *rpermIndices; /* indices used for any reordering */ 101 THRUSTINTARRAY *cpermIndices; /* indices used for any reordering */ 102 THRUSTARRAY *workVector; 103 MatCUSPARSEStorageFormat format; /* the storage format for the matrix on the device */ 104 cusparseHandle_t handle; /* a handle to the cusparse library */ 105 PetscInt nnz; /* number of nonzeros ... need this for accurate logging between ICC and ILU */ 106 }; 107 108 /* This is a larger struct holding all the matrices for a SpMV, and SpMV Tranpose */ 109 struct Mat_SeqAIJCUSPARSE { 110 Mat_SeqAIJCUSPARSEMultStruct *mat; /* pointer to the matrix on the GPU */ 111 Mat_SeqAIJCUSPARSEMultStruct *matTranspose; /* pointer to the matrix on the GPU (for the transpose ... useful for BiCG) */ 112 THRUSTARRAY *workVector; /*pointer to a workvector to which we can copy the relevant indices of a vector we want to multiply */ 113 PetscInt nonzerorow; /* number of nonzero rows ... used in the flop calculations */ 114 MatCUSPARSEStorageFormat format; /* the storage format for the matrix on the device */ 115 cudaStream_t stream; /* a stream for the parallel SpMV ... this is not owned and should not be deleted */ 116 cusparseHandle_t handle; /* a handle to the cusparse library ... this may not be owned (if we're working in parallel i.e. multiGPUs) */ 117 }; 118 119 PETSC_INTERN PetscErrorCode MatCUSPARSECopyToGPU(Mat); 120 #endif 121