1 #if !defined(CUSPARSEMATIMPL) 2 #define CUSPARSEMATIMPL 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 PetscStackCallThrust(body) do { \ 22 try { \ 23 body; \ 24 } catch(thrust::system_error& e) { \ 25 SETERRQ(PETSC_COMM_SELF,PETSC_ERR_LIB,"Error in Thrust %s",e.what());\ 26 } \ 27 } while (0) 28 29 #if defined(PETSC_USE_COMPLEX) 30 #if defined(PETSC_USE_REAL_SINGLE) 31 const cuComplex PETSC_CUSPARSE_ONE = {1.0f, 0.0f}; 32 const cuComplex PETSC_CUSPARSE_ZERO = {0.0f, 0.0f}; 33 #elif defined(PETSC_USE_REAL_DOUBLE) 34 const cuDoubleComplex PETSC_CUSPARSE_ONE = {1.0, 0.0}; 35 const cuDoubleComplex PETSC_CUSPARSE_ZERO = {0.0, 0.0}; 36 #endif 37 #else 38 const PetscScalar PETSC_CUSPARSE_ONE = 1.0; 39 const PetscScalar PETSC_CUSPARSE_ZERO = 0.0; 40 #endif 41 42 #if PETSC_PKG_CUDA_VERSION_GE(9,0,0) 43 #define csrsvInfo_t csrsv2Info_t 44 #define cusparseCreateCsrsvInfo cusparseCreateCsrsv2Info 45 #define cusparseDestroyCsrsvInfo cusparseDestroyCsrsv2Info 46 #if defined(PETSC_USE_COMPLEX) 47 #if defined(PETSC_USE_REAL_SINGLE) 48 #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) 49 #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) 50 #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) 51 #elif defined(PETSC_USE_REAL_DOUBLE) 52 #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) 53 #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) 54 #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) 55 #endif 56 #else /* not complex */ 57 #if defined(PETSC_USE_REAL_SINGLE) 58 #define cusparseXcsrsv_buffsize cusparseScsrsv2_bufferSize 59 #define cusparseXcsrsv_analysis cusparseScsrsv2_analysis 60 #define cusparseXcsrsv_solve cusparseScsrsv2_solve 61 #elif defined(PETSC_USE_REAL_DOUBLE) 62 #define cusparseXcsrsv_buffsize cusparseDcsrsv2_bufferSize 63 #define cusparseXcsrsv_analysis cusparseDcsrsv2_analysis 64 #define cusparseXcsrsv_solve cusparseDcsrsv2_solve 65 #endif 66 #endif 67 #else 68 #define csrsvInfo_t cusparseSolveAnalysisInfo_t 69 #define cusparseCreateCsrsvInfo cusparseCreateSolveAnalysisInfo 70 #define cusparseDestroyCsrsvInfo cusparseDestroySolveAnalysisInfo 71 #if defined(PETSC_USE_COMPLEX) 72 #if defined(PETSC_USE_REAL_SINGLE) 73 #define cusparseXcsrsv_solve(a,b,c,d,e,f,g,h,i,j,k) cusparseCcsrsv_solve((a),(b),(c),(cuComplex*)(d),(e),(cuComplex*)(f),(g),(h),(i),(cuComplex*)(j),(cuComplex*)(k)) 74 #define cusparseXcsrsv_analysis(a,b,c,d,e,f,g,h,i) cusparseCcsrsv_analysis((a),(b),(c),(d),(e),(cuComplex*)(f),(g),(h),(i)) 75 #elif defined(PETSC_USE_REAL_DOUBLE) 76 #define cusparseXcsrsv_solve(a,b,c,d,e,f,g,h,i,j,k) cusparseZcsrsv_solve((a),(b),(c),(cuDoubleComplex*)(d),(e),(cuDoubleComplex*)(f),(g),(h),(i),(cuDoubleComplex*)(j),(cuDoubleComplex*)(k)) 77 #define cusparseXcsrsv_analysis(a,b,c,d,e,f,g,h,i) cusparseZcsrsv_analysis((a),(b),(c),(d),(e),(cuDoubleComplex*)(f),(g),(h),(i)) 78 #endif 79 #else /* not complex */ 80 #if defined(PETSC_USE_REAL_SINGLE) 81 #define cusparseXcsrsv_solve cusparseScsrsv_solve 82 #define cusparseXcsrsv_analysis cusparseScsrsv_analysis 83 #elif defined(PETSC_USE_REAL_DOUBLE) 84 #define cusparseXcsrsv_solve cusparseDcsrsv_solve 85 #define cusparseXcsrsv_analysis cusparseDcsrsv_analysis 86 #endif 87 #endif 88 #endif 89 90 #if PETSC_PKG_CUDA_VERSION_GE(11,0,0) 91 #define cusparse_csr2csc cusparseCsr2cscEx2 92 #if defined(PETSC_USE_COMPLEX) 93 #if defined(PETSC_USE_REAL_SINGLE) 94 #define cusparse_scalartype CUDA_C_32F 95 #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) 96 #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) 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) 97 #elif defined(PETSC_USE_REAL_DOUBLE) 98 #define cusparse_scalartype CUDA_C_64F 99 #define cusparse_csr_spgeam(a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s,t) 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) 100 #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) 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) 101 #endif 102 #else /* not complex */ 103 #if defined(PETSC_USE_REAL_SINGLE) 104 #define cusparse_scalartype CUDA_R_32F 105 #define cusparse_csr_spgeam cusparseScsrgeam2 106 #define cusparse_csr_spgeam_bufferSize cusparseScsrgeam2_bufferSizeExt 107 #elif defined(PETSC_USE_REAL_DOUBLE) 108 #define cusparse_scalartype CUDA_R_64F 109 #define cusparse_csr_spgeam cusparseDcsrgeam2 110 #define cusparse_csr_spgeam_bufferSize cusparseDcsrgeam2_bufferSizeExt 111 #endif 112 #endif 113 #else 114 #if defined(PETSC_USE_COMPLEX) 115 #if defined(PETSC_USE_REAL_SINGLE) 116 #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)) 117 #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)) 118 #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)) 119 #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)) 120 #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)) 121 #define cusparse_hyb2csr(a,b,c,d,e,f) cusparseChyb2csr((a),(b),(c),(cuComplex*)(d),(e),(f)) 122 #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) 123 #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) 124 #elif defined(PETSC_USE_REAL_DOUBLE) 125 #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)) 126 #define cusparse_csr_spmm(a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p) cusparseZcsrmm((a),(b),(c),(d),(e),(f),(cuDoubleComplex*)(g),(h),(cuDoubleComplex*)(i),(j),(k),(cuDoubleComplex*)(l),(m),(cuDoubleComplex*)(n),(cuDoubleComplex*)(o),(p)) 127 #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)) 128 #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)) 129 #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)) 130 #define cusparse_hyb2csr(a,b,c,d,e,f) cusparseZhyb2csr((a),(b),(c),(cuDoubleComplex*)(d),(e),(f)) 131 #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) 132 #define cusparse_csr_spgeam(a,b,c,d,e,f,g,h,i,j,k,l,m,n,o,p,q,r,s) 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) 133 #endif 134 #else 135 #if defined(PETSC_USE_REAL_SINGLE) 136 #define cusparse_csr_spmv cusparseScsrmv 137 #define cusparse_csr_spmm cusparseScsrmm 138 #define cusparse_csr2csc cusparseScsr2csc 139 #define cusparse_hyb_spmv cusparseShybmv 140 #define cusparse_csr2hyb cusparseScsr2hyb 141 #define cusparse_hyb2csr cusparseShyb2csr 142 #define cusparse_csr_spgemm cusparseScsrgemm 143 #define cusparse_csr_spgeam cusparseScsrgeam 144 #elif defined(PETSC_USE_REAL_DOUBLE) 145 #define cusparse_csr_spmv cusparseDcsrmv 146 #define cusparse_csr_spmm cusparseDcsrmm 147 #define cusparse_csr2csc cusparseDcsr2csc 148 #define cusparse_hyb_spmv cusparseDhybmv 149 #define cusparse_csr2hyb cusparseDcsr2hyb 150 #define cusparse_hyb2csr cusparseDhyb2csr 151 #define cusparse_csr_spgemm cusparseDcsrgemm 152 #define cusparse_csr_spgeam cusparseDcsrgeam 153 #endif 154 #endif 155 #endif 156 157 #define THRUSTINTARRAY32 thrust::device_vector<int> 158 #define THRUSTINTARRAY thrust::device_vector<PetscInt> 159 #define THRUSTARRAY thrust::device_vector<PetscScalar> 160 161 /* A CSR matrix structure */ 162 struct CsrMatrix { 163 PetscInt num_rows; 164 PetscInt num_cols; 165 PetscInt num_entries; 166 THRUSTINTARRAY32 *row_offsets; 167 THRUSTINTARRAY32 *column_indices; 168 THRUSTARRAY *values; 169 }; 170 171 /* This is struct holding the relevant data needed to a MatSolve */ 172 struct Mat_SeqAIJCUSPARSETriFactorStruct { 173 /* Data needed for triangular solve */ 174 cusparseMatDescr_t descr; 175 cusparseOperation_t solveOp; 176 CsrMatrix *csrMat; 177 csrsvInfo_t solveInfo; 178 cusparseSolvePolicy_t solvePolicy; /* whether level information is generated and used */ 179 int solveBufferSize; 180 void *solveBuffer; 181 size_t csr2cscBufferSize; /* to transpose the triangular factor (only used for CUDA >= 11.0) */ 182 void *csr2cscBuffer; 183 PetscScalar *AA_h; /* managed host buffer for moving values to the GPU */ 184 }; 185 186 /* This is a larger struct holding all the triangular factors for a solve, transpose solve, and any indices used in a reordering */ 187 struct Mat_SeqAIJCUSPARSETriFactors { 188 Mat_SeqAIJCUSPARSETriFactorStruct *loTriFactorPtr; /* pointer for lower triangular (factored matrix) on GPU */ 189 Mat_SeqAIJCUSPARSETriFactorStruct *upTriFactorPtr; /* pointer for upper triangular (factored matrix) on GPU */ 190 Mat_SeqAIJCUSPARSETriFactorStruct *loTriFactorPtrTranspose; /* pointer for lower triangular (factored matrix) on GPU for the transpose (useful for BiCG) */ 191 Mat_SeqAIJCUSPARSETriFactorStruct *upTriFactorPtrTranspose; /* pointer for upper triangular (factored matrix) on GPU for the transpose (useful for BiCG)*/ 192 THRUSTINTARRAY *rpermIndices; /* indices used for any reordering */ 193 THRUSTINTARRAY *cpermIndices; /* indices used for any reordering */ 194 THRUSTARRAY *workVector; 195 cusparseHandle_t handle; /* a handle to the cusparse library */ 196 PetscInt nnz; /* number of nonzeros ... need this for accurate logging between ICC and ILU */ 197 PetscScalar *a_band_d; /* GPU data for banded CSR LU factorization matrix diag(L)=1 */ 198 int *i_band_d; /* this could be optimized away */ 199 cudaDeviceProp dev_prop; 200 PetscBool init_dev_prop; 201 }; 202 203 struct Mat_CusparseSpMV { 204 PetscBool initialized; /* Don't rely on spmvBuffer != NULL to test if the struct is initialized, */ 205 size_t spmvBufferSize; /* since I'm not sure if smvBuffer can be NULL even after cusparseSpMV_bufferSize() */ 206 void *spmvBuffer; 207 #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 */ 208 cusparseDnVecDescr_t vecXDescr,vecYDescr; /* descriptor for the dense vectors in y=op(A)x */ 209 #endif 210 }; 211 212 /* This is struct holding the relevant data needed to a MatMult */ 213 struct Mat_SeqAIJCUSPARSEMultStruct { 214 void *mat; /* opaque pointer to a matrix. This could be either a cusparseHybMat_t or a CsrMatrix */ 215 cusparseMatDescr_t descr; /* Data needed to describe the matrix for a multiply */ 216 THRUSTINTARRAY *cprowIndices; /* compressed row indices used in the parallel SpMV */ 217 PetscScalar *alpha_one; /* pointer to a device "scalar" storing the alpha parameter in the SpMV */ 218 PetscScalar *beta_zero; /* pointer to a device "scalar" storing the beta parameter in the SpMV as zero*/ 219 PetscScalar *beta_one; /* pointer to a device "scalar" storing the beta parameter in the SpMV as one */ 220 #if PETSC_PKG_CUDA_VERSION_GE(11,0,0) 221 cusparseSpMatDescr_t matDescr; /* descriptor for the matrix, used by SpMV and SpMM */ 222 Mat_CusparseSpMV cuSpMV[3]; /* different Mat_CusparseSpMV structs for non-transpose, transpose, conj-transpose */ 223 Mat_SeqAIJCUSPARSEMultStruct() : matDescr(NULL) { 224 for (int i=0; i<3; i++) cuSpMV[i].initialized = PETSC_FALSE; 225 } 226 #endif 227 }; 228 229 /* This is a larger struct holding all the matrices for a SpMV, and SpMV Transpose */ 230 struct Mat_SeqAIJCUSPARSE { 231 Mat_SeqAIJCUSPARSEMultStruct *mat; /* pointer to the matrix on the GPU */ 232 Mat_SeqAIJCUSPARSEMultStruct *matTranspose; /* pointer to the matrix on the GPU (for the transpose ... useful for BiCG) */ 233 THRUSTARRAY *workVector; /* pointer to a workvector to which we can copy the relevant indices of a vector we want to multiply */ 234 THRUSTINTARRAY32 *rowoffsets_gpu; /* rowoffsets on GPU in non-compressed-row format. It is used to convert CSR to CSC */ 235 PetscInt nrows; /* number of rows of the matrix seen by GPU */ 236 MatCUSPARSEStorageFormat format; /* the storage format for the matrix on the device */ 237 PetscBool use_cpu_solve; /* Use AIJ_Seq (I)LU solve */ 238 cudaStream_t stream; /* a stream for the parallel SpMV ... this is not owned and should not be deleted */ 239 cusparseHandle_t handle; /* a handle to the cusparse library ... this may not be owned (if we're working in parallel i.e. multiGPUs) */ 240 PetscObjectState nonzerostate; /* track nonzero state to possibly recreate the GPU matrix */ 241 #if PETSC_PKG_CUDA_VERSION_GE(11,0,0) 242 size_t csr2cscBufferSize; /* stuff used to compute the matTranspose above */ 243 void *csr2cscBuffer; /* This is used as a C struct and is calloc'ed by PetscNewLog() */ 244 cusparseCsr2CscAlg_t csr2cscAlg; /* algorithms can be selected from command line options */ 245 cusparseSpMVAlg_t spmvAlg; 246 cusparseSpMMAlg_t spmmAlg; 247 #endif 248 THRUSTINTARRAY *csr2csc_i; 249 PetscSplitCSRDataStructure deviceMat; /* Matrix on device for, eg, assembly */ 250 251 /* Stuff for basic COO support */ 252 THRUSTINTARRAY *cooPerm; /* permutation array that sorts the input coo entris by row and col */ 253 THRUSTINTARRAY *cooPerm_a; /* ordered array that indicate i-th nonzero (after sorting) is the j-th unique nonzero */ 254 255 /* Stuff for extended COO support */ 256 PetscBool use_extended_coo; /* Use extended COO format */ 257 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 */ 258 PetscCount *perm_d; 259 260 Mat_SeqAIJCUSPARSE() : use_extended_coo(PETSC_FALSE), perm_d(NULL), jmap_d(NULL) {} 261 }; 262 263 PETSC_INTERN PetscErrorCode MatSeqAIJCUSPARSECopyToGPU(Mat); 264 PETSC_INTERN PetscErrorCode MatSetPreallocationCOO_SeqAIJCUSPARSE_Basic(Mat,PetscCount,const PetscInt[],const PetscInt[]); 265 PETSC_INTERN PetscErrorCode MatSetValuesCOO_SeqAIJCUSPARSE_Basic(Mat,const PetscScalar[],InsertMode); 266 PETSC_INTERN PetscErrorCode MatSeqAIJCUSPARSEMergeMats(Mat,Mat,MatReuse,Mat*); 267 PETSC_INTERN PetscErrorCode MatSeqAIJCUSPARSETriFactors_Reset(Mat_SeqAIJCUSPARSETriFactors_p*); 268 269 static inline bool isCudaMem(const void *data) 270 { 271 cudaError_t cerr; 272 struct cudaPointerAttributes attr; 273 enum cudaMemoryType mtype; 274 cerr = cudaPointerGetAttributes(&attr,data); /* Do not check error since before CUDA 11.0, passing a host pointer returns cudaErrorInvalidValue */ 275 cudaGetLastError(); /* Reset the last error */ 276 #if (CUDART_VERSION < 10000) 277 mtype = attr.memoryType; 278 #else 279 mtype = attr.type; 280 #endif 281 if (cerr == cudaSuccess && mtype == cudaMemoryTypeDevice) return true; 282 else return false; 283 } 284 285 #endif 286