1 #ifndef PETSC_CUSPARSEMATIMPL_H 2 #define PETSC_CUSPARSEMATIMPL_H 3 4 #include <petscpkg_version.h> 5 #include <petsc/private/cudavecimpl.h> 6 #include <petscaijdevice.h> 7 8 #include <cusparse_v2.h> 9 10 #include <algorithm> 11 #include <vector> 12 13 #include <thrust/device_vector.h> 14 #include <thrust/device_ptr.h> 15 #include <thrust/device_malloc_allocator.h> 16 #include <thrust/transform.h> 17 #include <thrust/functional.h> 18 #include <thrust/sequence.h> 19 #include <thrust/system/system_error.h> 20 21 #define PetscCallThrust(body) \ 22 do { \ 23 try { \ 24 body; \ 25 } catch (thrust::system_error & e) { \ 26 SETERRQ(PETSC_COMM_SELF, PETSC_ERR_LIB, "Error in Thrust %s", e.what()); \ 27 } \ 28 } while (0) 29 30 #if defined(PETSC_USE_COMPLEX) 31 #if defined(PETSC_USE_REAL_SINGLE) 32 const cuComplex PETSC_CUSPARSE_ONE = {1.0f, 0.0f}; 33 const cuComplex PETSC_CUSPARSE_ZERO = {0.0f, 0.0f}; 34 #define cusparseXcsrilu02_bufferSize(a, b, c, d, e, f, g, h, i) cusparseCcsrilu02_bufferSize(a, b, c, d, (cuComplex *)e, f, g, h, i) 35 #define cusparseXcsrilu02_analysis(a, b, c, d, e, f, g, h, i, j) cusparseCcsrilu02_analysis(a, b, c, d, (cuComplex *)e, f, g, h, i, j) 36 #define cusparseXcsrilu02(a, b, c, d, e, f, g, h, i, j) cusparseCcsrilu02(a, b, c, d, (cuComplex *)e, f, g, h, i, j) 37 #define cusparseXcsric02_bufferSize(a, b, c, d, e, f, g, h, i) cusparseCcsric02_bufferSize(a, b, c, d, (cuComplex *)e, f, g, h, i) 38 #define cusparseXcsric02_analysis(a, b, c, d, e, f, g, h, i, j) cusparseCcsric02_analysis(a, b, c, d, (cuComplex *)e, f, g, h, i, j) 39 #define cusparseXcsric02(a, b, c, d, e, f, g, h, i, j) cusparseCcsric02(a, b, c, d, (cuComplex *)e, f, g, h, i, j) 40 #elif defined(PETSC_USE_REAL_DOUBLE) 41 const cuDoubleComplex PETSC_CUSPARSE_ONE = {1.0, 0.0}; 42 const cuDoubleComplex PETSC_CUSPARSE_ZERO = {0.0, 0.0}; 43 #define cusparseXcsrilu02_bufferSize(a, b, c, d, e, f, g, h, i) cusparseZcsrilu02_bufferSize(a, b, c, d, (cuDoubleComplex *)e, f, g, h, i) 44 #define cusparseXcsrilu02_analysis(a, b, c, d, e, f, g, h, i, j) cusparseZcsrilu02_analysis(a, b, c, d, (cuDoubleComplex *)e, f, g, h, i, j) 45 #define cusparseXcsrilu02(a, b, c, d, e, f, g, h, i, j) cusparseZcsrilu02(a, b, c, d, (cuDoubleComplex *)e, f, g, h, i, j) 46 #define cusparseXcsric02_bufferSize(a, b, c, d, e, f, g, h, i) cusparseZcsric02_bufferSize(a, b, c, d, (cuDoubleComplex *)e, f, g, h, i) 47 #define cusparseXcsric02_analysis(a, b, c, d, e, f, g, h, i, j) cusparseZcsric02_analysis(a, b, c, d, (cuDoubleComplex *)e, f, g, h, i, j) 48 #define cusparseXcsric02(a, b, c, d, e, f, g, h, i, j) cusparseZcsric02(a, b, c, d, (cuDoubleComplex *)e, f, g, h, i, j) 49 #endif 50 #else 51 const PetscScalar PETSC_CUSPARSE_ONE = 1.0; 52 const PetscScalar PETSC_CUSPARSE_ZERO = 0.0; 53 #if defined(PETSC_USE_REAL_SINGLE) 54 #define cusparseXcsrilu02_bufferSize cusparseScsrilu02_bufferSize 55 #define cusparseXcsrilu02_analysis cusparseScsrilu02_analysis 56 #define cusparseXcsrilu02 cusparseScsrilu02 57 #define cusparseXcsric02_bufferSize cusparseScsric02_bufferSize 58 #define cusparseXcsric02_analysis cusparseScsric02_analysis 59 #define cusparseXcsric02 cusparseScsric02 60 #elif defined(PETSC_USE_REAL_DOUBLE) 61 #define cusparseXcsrilu02_bufferSize cusparseDcsrilu02_bufferSize 62 #define cusparseXcsrilu02_analysis cusparseDcsrilu02_analysis 63 #define cusparseXcsrilu02 cusparseDcsrilu02 64 #define cusparseXcsric02_bufferSize cusparseDcsric02_bufferSize 65 #define cusparseXcsric02_analysis cusparseDcsric02_analysis 66 #define cusparseXcsric02 cusparseDcsric02 67 #endif 68 #endif 69 70 #if PETSC_PKG_CUDA_VERSION_GE(9, 0, 0) 71 #define csrsvInfo_t csrsv2Info_t 72 #define cusparseCreateCsrsvInfo cusparseCreateCsrsv2Info 73 #define cusparseDestroyCsrsvInfo cusparseDestroyCsrsv2Info 74 #if defined(PETSC_USE_COMPLEX) 75 #if defined(PETSC_USE_REAL_SINGLE) 76 #define cusparseXcsrsv_buffsize(a, b, c, d, e, f, g, h, i, j) cusparseCcsrsv2_bufferSize(a, b, c, d, e, (cuComplex *)(f), g, h, i, j) 77 #define cusparseXcsrsv_analysis(a, b, c, d, e, f, g, h, i, j, k) cusparseCcsrsv2_analysis(a, b, c, d, e, (const cuComplex *)(f), g, h, i, j, k) 78 #define cusparseXcsrsv_solve(a, b, c, d, e, f, g, h, i, j, k, l, m, n) cusparseCcsrsv2_solve(a, b, c, d, (const cuComplex *)(e), f, (const cuComplex *)(g), h, i, j, (const cuComplex *)(k), (cuComplex *)(l), m, n) 79 #elif defined(PETSC_USE_REAL_DOUBLE) 80 #define cusparseXcsrsv_buffsize(a, b, c, d, e, f, g, h, i, j) cusparseZcsrsv2_bufferSize(a, b, c, d, e, (cuDoubleComplex *)(f), g, h, i, j) 81 #define cusparseXcsrsv_analysis(a, b, c, d, e, f, g, h, i, j, k) cusparseZcsrsv2_analysis(a, b, c, d, e, (const cuDoubleComplex *)(f), g, h, i, j, k) 82 #define cusparseXcsrsv_solve(a, b, c, d, e, f, g, h, i, j, k, l, m, n) cusparseZcsrsv2_solve(a, b, c, d, (const cuDoubleComplex *)(e), f, (const cuDoubleComplex *)(g), h, i, j, (const cuDoubleComplex *)(k), (cuDoubleComplex *)(l), m, n) 83 #endif 84 #else /* not complex */ 85 #if defined(PETSC_USE_REAL_SINGLE) 86 #define cusparseXcsrsv_buffsize cusparseScsrsv2_bufferSize 87 #define cusparseXcsrsv_analysis cusparseScsrsv2_analysis 88 #define cusparseXcsrsv_solve cusparseScsrsv2_solve 89 #elif defined(PETSC_USE_REAL_DOUBLE) 90 #define cusparseXcsrsv_buffsize cusparseDcsrsv2_bufferSize 91 #define cusparseXcsrsv_analysis cusparseDcsrsv2_analysis 92 #define cusparseXcsrsv_solve cusparseDcsrsv2_solve 93 #endif 94 #endif 95 #else /* PETSC_PKG_CUDA_VERSION_GE(9, 0, 0) */ 96 #define csrsvInfo_t cusparseSolveAnalysisInfo_t 97 #define cusparseCreateCsrsvInfo cusparseCreateSolveAnalysisInfo 98 #define cusparseDestroyCsrsvInfo cusparseDestroySolveAnalysisInfo 99 #if defined(PETSC_USE_COMPLEX) 100 #if defined(PETSC_USE_REAL_SINGLE) 101 #define cusparseXcsrsv_solve(a, b, c, d_IGNORED, e, f, g, h, i, j, k, l, m_IGNORED, n_IGNORED) cusparseCcsrsv_solve((a), (b), (c), (cuComplex *)(e), (f), (cuComplex *)(g), (h), (i), (j), (cuComplex *)(k), (cuComplex *)(l)) 102 #define cusparseXcsrsv_analysis(a, b, c, d, e, f, g, h, i, j_IGNORED, k_IGNORED) cusparseCcsrsv_analysis((a), (b), (c), (d), (e), (cuComplex *)(f), (g), (h), (i)) 103 #elif defined(PETSC_USE_REAL_DOUBLE) 104 #define cusparseXcsrsv_solve(a, b, c, d_IGNORED, e, f, g, h, i, j, k, l, m_IGNORED, n_IGNORED) \ 105 cusparseZcsrsv_solve((a), (b), (c), (cuDoubleComplex *)(e), (f), (cuDoubleComplex *)(g), (h), (i), (j), (cuDoubleComplex *)(k), (cuDoubleComplex *)(l)) 106 #define cusparseXcsrsv_analysis(a, b, c, d, e, f, g, h, i, j_IGNORED, k_IGNORED) cusparseZcsrsv_analysis((a), (b), (c), (d), (e), (cuDoubleComplex *)(f), (g), (h), (i)) 107 #endif 108 #else /* not complex */ 109 #if defined(PETSC_USE_REAL_SINGLE) 110 #define cusparseXcsrsv_solve cusparseScsrsv_solve 111 #define cusparseXcsrsv_analysis(a, b, c, d, e, f, g, h, i, j, k) cusparseScsrsv_analysis(a, b, c, d, e, f, g, h, i) 112 #elif defined(PETSC_USE_REAL_DOUBLE) 113 #define cusparseXcsrsv_solve cusparseDcsrsv_solve 114 #define cusparseXcsrsv_analysis(a, b, c, d, e, f, g, h, i, j, k) cusparseDcsrsv_analysis(a, b, c, d, e, f, g, h, i) 115 #endif 116 #endif 117 #endif /* PETSC_PKG_CUDA_VERSION_GE(9, 0, 0) */ 118 119 #if PETSC_PKG_CUDA_VERSION_GE(11, 0, 0) 120 #define cusparse_csr2csc cusparseCsr2cscEx2 121 #if defined(PETSC_USE_COMPLEX) 122 #if defined(PETSC_USE_REAL_SINGLE) 123 #define cusparse_scalartype CUDA_C_32F 124 #define cusparse_csr_spgeam(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t) cusparseCcsrgeam2(a, b, c, (cuComplex *)d, e, f, (cuComplex *)g, h, i, (cuComplex *)j, k, l, (cuComplex *)m, n, o, p, (cuComplex *)q, r, s, t) 125 #define cusparse_csr_spgeam_bufferSize(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t) \ 126 cusparseCcsrgeam2_bufferSizeExt(a, b, c, (cuComplex *)d, e, f, (cuComplex *)g, h, i, (cuComplex *)j, k, l, (cuComplex *)m, n, o, p, (cuComplex *)q, r, s, t) 127 #elif defined(PETSC_USE_REAL_DOUBLE) 128 #define cusparse_scalartype CUDA_C_64F 129 #define cusparse_csr_spgeam(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t) \ 130 cusparseZcsrgeam2(a, b, c, (cuDoubleComplex *)d, e, f, (cuDoubleComplex *)g, h, i, (cuDoubleComplex *)j, k, l, (cuDoubleComplex *)m, n, o, p, (cuDoubleComplex *)q, r, s, t) 131 #define cusparse_csr_spgeam_bufferSize(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t) \ 132 cusparseZcsrgeam2_bufferSizeExt(a, b, c, (cuDoubleComplex *)d, e, f, (cuDoubleComplex *)g, h, i, (cuDoubleComplex *)j, k, l, (cuDoubleComplex *)m, n, o, p, (cuDoubleComplex *)q, r, s, t) 133 #endif 134 #else /* not complex */ 135 #if defined(PETSC_USE_REAL_SINGLE) 136 #define cusparse_scalartype CUDA_R_32F 137 #define cusparse_csr_spgeam cusparseScsrgeam2 138 #define cusparse_csr_spgeam_bufferSize cusparseScsrgeam2_bufferSizeExt 139 #elif defined(PETSC_USE_REAL_DOUBLE) 140 #define cusparse_scalartype CUDA_R_64F 141 #define cusparse_csr_spgeam cusparseDcsrgeam2 142 #define cusparse_csr_spgeam_bufferSize cusparseDcsrgeam2_bufferSizeExt 143 #endif 144 #endif 145 #else /* PETSC_PKG_CUDA_VERSION_GE(11, 0, 0) */ 146 #if defined(PETSC_USE_COMPLEX) 147 #if defined(PETSC_USE_REAL_SINGLE) 148 #define cusparse_csr_spmv(a, b, c, d, e, f, g, h, i, j, k, l, m) cusparseCcsrmv((a), (b), (c), (d), (e), (cuComplex *)(f), (g), (cuComplex *)(h), (i), (j), (cuComplex *)(k), (cuComplex *)(l), (cuComplex *)(m)) 149 #define cusparse_csr_spmm(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p) cusparseCcsrmm((a), (b), (c), (d), (e), (f), (cuComplex *)(g), (h), (cuComplex *)(i), (j), (k), (cuComplex *)(l), (m), (cuComplex *)(n), (cuComplex *)(o), (p)) 150 #define cusparse_csr2csc(a, b, c, d, e, f, g, h, i, j, k, l) cusparseCcsr2csc((a), (b), (c), (d), (cuComplex *)(e), (f), (g), (cuComplex *)(h), (i), (j), (k), (l)) 151 #define cusparse_hyb_spmv(a, b, c, d, e, f, g, h) cusparseChybmv((a), (b), (cuComplex *)(c), (d), (e), (cuComplex *)(f), (cuComplex *)(g), (cuComplex *)(h)) 152 #define cusparse_csr2hyb(a, b, c, d, e, f, g, h, i, j) cusparseCcsr2hyb((a), (b), (c), (d), (cuComplex *)(e), (f), (g), (h), (i), (j)) 153 #define cusparse_hyb2csr(a, b, c, d, e, f) cusparseChyb2csr((a), (b), (c), (cuComplex *)(d), (e), (f)) 154 #define cusparse_csr_spgemm(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t) cusparseCcsrgemm(a, b, c, d, e, f, g, h, (cuComplex *)i, j, k, l, m, (cuComplex *)n, o, p, q, (cuComplex *)r, s, t) 155 #define cusparse_csr_spgeam(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s) cusparseCcsrgeam(a, b, c, (cuComplex *)d, e, f, (cuComplex *)g, h, i, (cuComplex *)j, k, l, (cuComplex *)m, n, o, p, (cuComplex *)q, r, s) 156 #elif defined(PETSC_USE_REAL_DOUBLE) 157 #define cusparse_csr_spmv(a, b, c, d, e, f, g, h, i, j, k, l, m) cusparseZcsrmv((a), (b), (c), (d), (e), (cuDoubleComplex *)(f), (g), (cuDoubleComplex *)(h), (i), (j), (cuDoubleComplex *)(k), (cuDoubleComplex *)(l), (cuDoubleComplex *)(m)) 158 #define cusparse_csr_spmm(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p) \ 159 cusparseZcsrmm((a), (b), (c), (d), (e), (f), (cuDoubleComplex *)(g), (h), (cuDoubleComplex *)(i), (j), (k), (cuDoubleComplex *)(l), (m), (cuDoubleComplex *)(n), (cuDoubleComplex *)(o), (p)) 160 #define cusparse_csr2csc(a, b, c, d, e, f, g, h, i, j, k, l) cusparseZcsr2csc((a), (b), (c), (d), (cuDoubleComplex *)(e), (f), (g), (cuDoubleComplex *)(h), (i), (j), (k), (l)) 161 #define cusparse_hyb_spmv(a, b, c, d, e, f, g, h) cusparseZhybmv((a), (b), (cuDoubleComplex *)(c), (d), (e), (cuDoubleComplex *)(f), (cuDoubleComplex *)(g), (cuDoubleComplex *)(h)) 162 #define cusparse_csr2hyb(a, b, c, d, e, f, g, h, i, j) cusparseZcsr2hyb((a), (b), (c), (d), (cuDoubleComplex *)(e), (f), (g), (h), (i), (j)) 163 #define cusparse_hyb2csr(a, b, c, d, e, f) cusparseZhyb2csr((a), (b), (c), (cuDoubleComplex *)(d), (e), (f)) 164 #define cusparse_csr_spgemm(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s, t) cusparseZcsrgemm(a, b, c, d, e, f, g, h, (cuDoubleComplex *)i, j, k, l, m, (cuDoubleComplex *)n, o, p, q, (cuDoubleComplex *)r, s, t) 165 #define cusparse_csr_spgeam(a, b, c, d, e, f, g, h, i, j, k, l, m, n, o, p, q, r, s) \ 166 cusparseZcsrgeam(a, b, c, (cuDoubleComplex *)d, e, f, (cuDoubleComplex *)g, h, i, (cuDoubleComplex *)j, k, l, (cuDoubleComplex *)m, n, o, p, (cuDoubleComplex *)q, r, s) 167 #endif 168 #else 169 #if defined(PETSC_USE_REAL_SINGLE) 170 #define cusparse_csr_spmv cusparseScsrmv 171 #define cusparse_csr_spmm cusparseScsrmm 172 #define cusparse_csr2csc cusparseScsr2csc 173 #define cusparse_hyb_spmv cusparseShybmv 174 #define cusparse_csr2hyb cusparseScsr2hyb 175 #define cusparse_hyb2csr cusparseShyb2csr 176 #define cusparse_csr_spgemm cusparseScsrgemm 177 #define cusparse_csr_spgeam cusparseScsrgeam 178 #elif defined(PETSC_USE_REAL_DOUBLE) 179 #define cusparse_csr_spmv cusparseDcsrmv 180 #define cusparse_csr_spmm cusparseDcsrmm 181 #define cusparse_csr2csc cusparseDcsr2csc 182 #define cusparse_hyb_spmv cusparseDhybmv 183 #define cusparse_csr2hyb cusparseDcsr2hyb 184 #define cusparse_hyb2csr cusparseDhyb2csr 185 #define cusparse_csr_spgemm cusparseDcsrgemm 186 #define cusparse_csr_spgeam cusparseDcsrgeam 187 #endif 188 #endif 189 #endif /* PETSC_PKG_CUDA_VERSION_GE(11, 0, 0) */ 190 191 #define THRUSTINTARRAY32 thrust::device_vector<int> 192 #define THRUSTINTARRAY thrust::device_vector<PetscInt> 193 #define THRUSTARRAY thrust::device_vector<PetscScalar> 194 195 /* A CSR matrix structure */ 196 struct CsrMatrix { 197 PetscInt num_rows; 198 PetscInt num_cols; 199 PetscInt num_entries; 200 THRUSTINTARRAY32 *row_offsets; 201 THRUSTINTARRAY32 *column_indices; 202 THRUSTARRAY *values; 203 }; 204 205 /* This is struct holding the relevant data needed to a MatSolve */ 206 struct Mat_SeqAIJCUSPARSETriFactorStruct { 207 /* Data needed for triangular solve */ 208 cusparseMatDescr_t descr; 209 cusparseOperation_t solveOp; 210 CsrMatrix *csrMat; 211 csrsvInfo_t solveInfo; 212 cusparseSolvePolicy_t solvePolicy; /* whether level information is generated and used */ 213 int solveBufferSize; 214 void *solveBuffer; 215 size_t csr2cscBufferSize; /* to transpose the triangular factor (only used for CUDA >= 11.0) */ 216 void *csr2cscBuffer; 217 PetscScalar *AA_h; /* managed host buffer for moving values to the GPU */ 218 }; 219 220 /* This is a larger struct holding all the triangular factors for a solve, transpose solve, and any indices used in a reordering */ 221 struct Mat_SeqAIJCUSPARSETriFactors { 222 Mat_SeqAIJCUSPARSETriFactorStruct *loTriFactorPtr; /* pointer for lower triangular (factored matrix) on GPU */ 223 Mat_SeqAIJCUSPARSETriFactorStruct *upTriFactorPtr; /* pointer for upper triangular (factored matrix) on GPU */ 224 Mat_SeqAIJCUSPARSETriFactorStruct *loTriFactorPtrTranspose; /* pointer for lower triangular (factored matrix) on GPU for the transpose (useful for BiCG) */ 225 Mat_SeqAIJCUSPARSETriFactorStruct *upTriFactorPtrTranspose; /* pointer for upper triangular (factored matrix) on GPU for the transpose (useful for BiCG)*/ 226 THRUSTINTARRAY *rpermIndices; /* indices used for any reordering */ 227 THRUSTINTARRAY *cpermIndices; /* indices used for any reordering */ 228 THRUSTARRAY *workVector; 229 cusparseHandle_t handle; /* a handle to the cusparse library */ 230 PetscInt nnz; /* number of nonzeros ... need this for accurate logging between ICC and ILU */ 231 PetscScalar *a_band_d; /* GPU data for banded CSR LU factorization matrix diag(L)=1 */ 232 int *i_band_d; /* this could be optimized away */ 233 cudaDeviceProp dev_prop; 234 PetscBool init_dev_prop; 235 236 /* csrilu0/csric0 appeared in cusparse-8.0, but we use it along with cusparseSpSV, 237 which first appeared in cusparse-11.5 with cuda-11.3. 238 */ 239 PetscBool factorizeOnDevice; /* Do factorization on device or not */ 240 #if CUSPARSE_VERSION >= 11500 241 PetscScalar *csrVal; 242 int *csrRowPtr, *csrColIdx; /* a,i,j of M. Using int since some cusparse APIs only support 32-bit indices */ 243 244 /* Mixed mat descriptor types? yes, different cusparse APIs use different types */ 245 cusparseMatDescr_t matDescr_M; 246 cusparseSpMatDescr_t spMatDescr_L, spMatDescr_U; 247 cusparseSpSVDescr_t spsvDescr_L, spsvDescr_Lt, spsvDescr_U, spsvDescr_Ut; 248 249 cusparseDnVecDescr_t dnVecDescr_X, dnVecDescr_Y; 250 PetscScalar *X, *Y; /* data array of dnVec X and Y */ 251 252 /* Mixed size types? yes, CUDA-11.7.0 declared cusparseDcsrilu02_bufferSizeExt() that returns size_t but did not implement it! */ 253 int factBufferSize_M; /* M ~= LU or LLt */ 254 size_t spsvBufferSize_L, spsvBufferSize_Lt, spsvBufferSize_U, spsvBufferSize_Ut; 255 /* cusparse needs various buffers for factorization and solve of L, U, Lt, or Ut. 256 So save memory, we share the factorization buffer with one of spsvBuffer_L/U. 257 */ 258 void *factBuffer_M, *spsvBuffer_L, *spsvBuffer_U, *spsvBuffer_Lt, *spsvBuffer_Ut; 259 260 csrilu02Info_t ilu0Info_M; 261 csric02Info_t ic0Info_M; 262 int structural_zero, numerical_zero; 263 cusparseSolvePolicy_t policy_M; 264 265 /* In MatSolveTranspose() for ILU0, we use the two flags to do on-demand solve */ 266 PetscBool createdTransposeSpSVDescr; /* Have we created SpSV descriptors for Lt, Ut? */ 267 PetscBool updatedTransposeSpSVAnalysis; /* Have we updated SpSV analysis with the latest L, U values? */ 268 269 PetscLogDouble numericFactFlops; /* Estimated FLOPs in ILU0/ICC0 numeric factorization */ 270 #endif 271 }; 272 273 struct Mat_CusparseSpMV { 274 PetscBool initialized; /* Don't rely on spmvBuffer != NULL to test if the struct is initialized, */ 275 size_t spmvBufferSize; /* since I'm not sure if smvBuffer can be NULL even after cusparseSpMV_bufferSize() */ 276 void *spmvBuffer; 277 #if PETSC_PKG_CUDA_VERSION_GE(11, 0, 0) /* these are present from CUDA 10.1, but PETSc code makes use of them from CUDA 11 on */ 278 cusparseDnVecDescr_t vecXDescr, vecYDescr; /* descriptor for the dense vectors in y=op(A)x */ 279 #endif 280 }; 281 282 /* This is struct holding the relevant data needed to a MatMult */ 283 struct Mat_SeqAIJCUSPARSEMultStruct { 284 void *mat; /* opaque pointer to a matrix. This could be either a cusparseHybMat_t or a CsrMatrix */ 285 cusparseMatDescr_t descr; /* Data needed to describe the matrix for a multiply */ 286 THRUSTINTARRAY *cprowIndices; /* compressed row indices used in the parallel SpMV */ 287 PetscScalar *alpha_one; /* pointer to a device "scalar" storing the alpha parameter in the SpMV */ 288 PetscScalar *beta_zero; /* pointer to a device "scalar" storing the beta parameter in the SpMV as zero*/ 289 PetscScalar *beta_one; /* pointer to a device "scalar" storing the beta parameter in the SpMV as one */ 290 #if PETSC_PKG_CUDA_VERSION_GE(11, 0, 0) 291 cusparseSpMatDescr_t matDescr; /* descriptor for the matrix, used by SpMV and SpMM */ 292 Mat_CusparseSpMV cuSpMV[3]; /* different Mat_CusparseSpMV structs for non-transpose, transpose, conj-transpose */ 293 Mat_SeqAIJCUSPARSEMultStruct() : matDescr(NULL) 294 { 295 for (int i = 0; i < 3; i++) cuSpMV[i].initialized = PETSC_FALSE; 296 } 297 #endif 298 }; 299 300 /* This is a larger struct holding all the matrices for a SpMV, and SpMV Transpose */ 301 struct Mat_SeqAIJCUSPARSE { 302 Mat_SeqAIJCUSPARSEMultStruct *mat; /* pointer to the matrix on the GPU */ 303 Mat_SeqAIJCUSPARSEMultStruct *matTranspose; /* pointer to the matrix on the GPU (for the transpose ... useful for BiCG) */ 304 THRUSTARRAY *workVector; /* pointer to a workvector to which we can copy the relevant indices of a vector we want to multiply */ 305 THRUSTINTARRAY32 *rowoffsets_gpu; /* rowoffsets on GPU in non-compressed-row format. It is used to convert CSR to CSC */ 306 PetscInt nrows; /* number of rows of the matrix seen by GPU */ 307 MatCUSPARSEStorageFormat format; /* the storage format for the matrix on the device */ 308 PetscBool use_cpu_solve; /* Use AIJ_Seq (I)LU solve */ 309 cudaStream_t stream; /* a stream for the parallel SpMV ... this is not owned and should not be deleted */ 310 cusparseHandle_t handle; /* a handle to the cusparse library ... this may not be owned (if we're working in parallel i.e. multiGPUs) */ 311 PetscObjectState nonzerostate; /* track nonzero state to possibly recreate the GPU matrix */ 312 #if PETSC_PKG_CUDA_VERSION_GE(11, 0, 0) 313 size_t csr2cscBufferSize; /* stuff used to compute the matTranspose above */ 314 void *csr2cscBuffer; /* This is used as a C struct and is calloc'ed by PetscNew() */ 315 cusparseCsr2CscAlg_t csr2cscAlg; /* algorithms can be selected from command line options */ 316 cusparseSpMVAlg_t spmvAlg; 317 cusparseSpMMAlg_t spmmAlg; 318 #endif 319 THRUSTINTARRAY *csr2csc_i; 320 PetscSplitCSRDataStructure deviceMat; /* Matrix on device for, eg, assembly */ 321 322 /* Stuff for basic COO support */ 323 THRUSTINTARRAY *cooPerm; /* permutation array that sorts the input coo entris by row and col */ 324 THRUSTINTARRAY *cooPerm_a; /* ordered array that indicate i-th nonzero (after sorting) is the j-th unique nonzero */ 325 326 /* Stuff for extended COO support */ 327 PetscBool use_extended_coo; /* Use extended COO format */ 328 PetscCount *jmap_d; /* perm[disp+jmap[i]..disp+jmap[i+1]) gives indices of entries in v[] associated with i-th nonzero of the matrix */ 329 PetscCount *perm_d; 330 331 Mat_SeqAIJCUSPARSE() : use_extended_coo(PETSC_FALSE), perm_d(NULL), jmap_d(NULL) { } 332 }; 333 334 typedef struct Mat_SeqAIJCUSPARSETriFactors *Mat_SeqAIJCUSPARSETriFactors_p; 335 336 PETSC_INTERN PetscErrorCode MatSeqAIJCUSPARSECopyToGPU(Mat); 337 PETSC_INTERN PetscErrorCode MatSetPreallocationCOO_SeqAIJCUSPARSE_Basic(Mat, PetscCount, PetscInt[], PetscInt[]); 338 PETSC_INTERN PetscErrorCode MatSetValuesCOO_SeqAIJCUSPARSE_Basic(Mat, const PetscScalar[], InsertMode); 339 PETSC_INTERN PetscErrorCode MatSeqAIJCUSPARSEMergeMats(Mat, Mat, MatReuse, Mat *); 340 PETSC_INTERN PetscErrorCode MatSeqAIJCUSPARSETriFactors_Reset(Mat_SeqAIJCUSPARSETriFactors_p *); 341 342 static inline bool isCudaMem(const void *data) 343 { 344 cudaError_t cerr; 345 struct cudaPointerAttributes attr; 346 enum cudaMemoryType mtype; 347 cerr = cudaPointerGetAttributes(&attr, data); /* Do not check error since before CUDA 11.0, passing a host pointer returns cudaErrorInvalidValue */ 348 cudaGetLastError(); /* Reset the last error */ 349 #if (CUDART_VERSION < 10000) 350 mtype = attr.memoryType; 351 #else 352 mtype = attr.type; 353 #endif 354 if (cerr == cudaSuccess && mtype == cudaMemoryTypeDevice) return true; 355 else return false; 356 } 357 358 #endif // PETSC_CUSPARSEMATIMPL_H 359