Actual source code: cusparsematimpl.h
1: #if !defined(CUSPARSEMATIMPL)
2: #define CUSPARSEMATIMPL
4: #include <petscpkg_version.h>
5: #include <petsc/private/cudavecimpl.h>
6: #include <petscaijdevice.h>
8: #include <cusparse_v2.h>
10: #include <algorithm>
11: #include <vector>
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>
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)
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
42: #if PETSC_PKG_CUDA_VERSION_GE(9,0,0)
43: #define cusparse_create_analysis_info cusparseCreateCsrsv2Info
44: #define cusparse_destroy_analysis_info cusparseDestroyCsrsv2Info
45: #if defined(PETSC_USE_COMPLEX)
46: #if defined(PETSC_USE_REAL_SINGLE)
47: #define cusparse_get_svbuffsize(a,b,c,d,e,f,g,h,i,j) cusparseCcsrsv2_bufferSize(a,b,c,d,e,(cuComplex*)(f),g,h,i,j)
48: #define cusparse_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)
49: #define cusparse_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)
50: #elif defined(PETSC_USE_REAL_DOUBLE)
51: #define cusparse_get_svbuffsize(a,b,c,d,e,f,g,h,i,j) cusparseZcsrsv2_bufferSize(a,b,c,d,e,(cuDoubleComplex*)(f),g,h,i,j)
52: #define cusparse_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)
53: #define cusparse_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)
54: #endif
55: #else /* not complex */
56: #if defined(PETSC_USE_REAL_SINGLE)
57: #define cusparse_get_svbuffsize cusparseScsrsv2_bufferSize
58: #define cusparse_analysis cusparseScsrsv2_analysis
59: #define cusparse_solve cusparseScsrsv2_solve
60: #elif defined(PETSC_USE_REAL_DOUBLE)
61: #define cusparse_get_svbuffsize cusparseDcsrsv2_bufferSize
62: #define cusparse_analysis cusparseDcsrsv2_analysis
63: #define cusparse_solve cusparseDcsrsv2_solve
64: #endif
65: #endif
66: #else
67: #define cusparse_create_analysis_info cusparseCreateSolveAnalysisInfo
68: #define cusparse_destroy_analysis_info cusparseDestroySolveAnalysisInfo
69: #if defined(PETSC_USE_COMPLEX)
70: #if defined(PETSC_USE_REAL_SINGLE)
71: #define cusparse_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))
72: #define cusparse_analysis(a,b,c,d,e,f,g,h,i) cusparseCcsrsv_analysis((a),(b),(c),(d),(e),(cuComplex*)(f),(g),(h),(i))
73: #elif defined(PETSC_USE_REAL_DOUBLE)
74: #define cusparse_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))
75: #define cusparse_analysis(a,b,c,d,e,f,g,h,i) cusparseZcsrsv_analysis((a),(b),(c),(d),(e),(cuDoubleComplex*)(f),(g),(h),(i))
76: #endif
77: #else /* not complex */
78: #if defined(PETSC_USE_REAL_SINGLE)
79: #define cusparse_solve cusparseScsrsv_solve
80: #define cusparse_analysis cusparseScsrsv_analysis
81: #elif defined(PETSC_USE_REAL_DOUBLE)
82: #define cusparse_solve cusparseDcsrsv_solve
83: #define cusparse_analysis cusparseDcsrsv_analysis
84: #endif
85: #endif
86: #endif
88: #if PETSC_PKG_CUDA_VERSION_GE(11,0,0)
89: #define cusparse_csr2csc cusparseCsr2cscEx2
90: #if defined(PETSC_USE_COMPLEX)
91: #if defined(PETSC_USE_REAL_SINGLE)
92: #define cusparse_scalartype CUDA_C_32F
93: #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)
94: #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)
95: #elif defined(PETSC_USE_REAL_DOUBLE)
96: #define cusparse_scalartype CUDA_C_64F
97: #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)
98: #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)
99: #endif
100: #else /* not complex */
101: #if defined(PETSC_USE_REAL_SINGLE)
102: #define cusparse_scalartype CUDA_R_32F
103: #define cusparse_csr_spgeam cusparseScsrgeam2
104: #define cusparse_csr_spgeam_bufferSize cusparseScsrgeam2_bufferSizeExt
105: #elif defined(PETSC_USE_REAL_DOUBLE)
106: #define cusparse_scalartype CUDA_R_64F
107: #define cusparse_csr_spgeam cusparseDcsrgeam2
108: #define cusparse_csr_spgeam_bufferSize cusparseDcsrgeam2_bufferSizeExt
109: #endif
110: #endif
111: #else
112: #if defined(PETSC_USE_COMPLEX)
113: #if defined(PETSC_USE_REAL_SINGLE)
114: #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))
115: #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))
116: #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))
117: #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))
118: #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))
119: #define cusparse_hyb2csr(a,b,c,d,e,f) cusparseChyb2csr((a),(b),(c),(cuComplex*)(d),(e),(f))
120: #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)
121: #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)
122: #elif defined(PETSC_USE_REAL_DOUBLE)
123: #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))
124: #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))
125: #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))
126: #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))
127: #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))
128: #define cusparse_hyb2csr(a,b,c,d,e,f) cusparseZhyb2csr((a),(b),(c),(cuDoubleComplex*)(d),(e),(f))
129: #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)
130: #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)
131: #endif
132: #else
133: #if defined(PETSC_USE_REAL_SINGLE)
134: #define cusparse_csr_spmv cusparseScsrmv
135: #define cusparse_csr_spmm cusparseScsrmm
136: #define cusparse_csr2csc cusparseScsr2csc
137: #define cusparse_hyb_spmv cusparseShybmv
138: #define cusparse_csr2hyb cusparseScsr2hyb
139: #define cusparse_hyb2csr cusparseShyb2csr
140: #define cusparse_csr_spgemm cusparseScsrgemm
141: #define cusparse_csr_spgeam cusparseScsrgeam
142: #elif defined(PETSC_USE_REAL_DOUBLE)
143: #define cusparse_csr_spmv cusparseDcsrmv
144: #define cusparse_csr_spmm cusparseDcsrmm
145: #define cusparse_csr2csc cusparseDcsr2csc
146: #define cusparse_hyb_spmv cusparseDhybmv
147: #define cusparse_csr2hyb cusparseDcsr2hyb
148: #define cusparse_hyb2csr cusparseDhyb2csr
149: #define cusparse_csr_spgemm cusparseDcsrgemm
150: #define cusparse_csr_spgeam cusparseDcsrgeam
151: #endif
152: #endif
153: #endif
155: #define THRUSTINTARRAY32 thrust::device_vector<int>
156: #define THRUSTINTARRAY thrust::device_vector<PetscInt>
157: #define THRUSTARRAY thrust::device_vector<PetscScalar>
159: /* A CSR matrix structure */
160: struct CsrMatrix {
161: PetscInt num_rows;
162: PetscInt num_cols;
163: PetscInt num_entries;
164: THRUSTINTARRAY32 *row_offsets;
165: THRUSTINTARRAY32 *column_indices;
166: THRUSTARRAY *values;
167: };
169: /* This is struct holding the relevant data needed to a MatSolve */
170: struct Mat_SeqAIJCUSPARSETriFactorStruct {
171: /* Data needed for triangular solve */
172: cusparseMatDescr_t descr;
173: cusparseOperation_t solveOp;
174: CsrMatrix *csrMat;
175: #if PETSC_PKG_CUDA_VERSION_GE(9,0,0)
176: csrsv2Info_t solveInfo;
177: #else
178: cusparseSolveAnalysisInfo_t solveInfo;
179: #endif
180: cusparseSolvePolicy_t solvePolicy; /* whether level information is generated and used */
181: int solveBufferSize;
182: void *solveBuffer;
183: size_t csr2cscBufferSize; /* to transpose the triangular factor (only used for CUDA >= 11.0) */
184: void *csr2cscBuffer;
185: PetscScalar *AA_h; /* managed host buffer for moving values to the GPU */
186: };
188: /* This is a larger struct holding all the triangular factors for a solve, transpose solve, and any indices used in a reordering */
189: struct Mat_SeqAIJCUSPARSETriFactors {
190: Mat_SeqAIJCUSPARSETriFactorStruct *loTriFactorPtr; /* pointer for lower triangular (factored matrix) on GPU */
191: Mat_SeqAIJCUSPARSETriFactorStruct *upTriFactorPtr; /* pointer for upper triangular (factored matrix) on GPU */
192: Mat_SeqAIJCUSPARSETriFactorStruct *loTriFactorPtrTranspose; /* pointer for lower triangular (factored matrix) on GPU for the transpose (useful for BiCG) */
193: Mat_SeqAIJCUSPARSETriFactorStruct *upTriFactorPtrTranspose; /* pointer for upper triangular (factored matrix) on GPU for the transpose (useful for BiCG)*/
194: THRUSTINTARRAY *rpermIndices; /* indices used for any reordering */
195: THRUSTINTARRAY *cpermIndices; /* indices used for any reordering */
196: THRUSTARRAY *workVector;
197: cusparseHandle_t handle; /* a handle to the cusparse library */
198: PetscInt nnz; /* number of nonzeros ... need this for accurate logging between ICC and ILU */
199: PetscScalar *a_band_d; /* GPU data for banded CSR LU factorization matrix diag(L)=1 */
200: int *i_band_d; /* this could be optimized away */
201: cudaDeviceProp dev_prop;
202: PetscBool init_dev_prop;
203: };
205: struct Mat_CusparseSpMV {
206: PetscBool initialized; /* Don't rely on spmvBuffer != NULL to test if the struct is initialized, */
207: size_t spmvBufferSize; /* since I'm not sure if smvBuffer can be NULL even after cusparseSpMV_bufferSize() */
208: void *spmvBuffer;
209: #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 */
210: cusparseDnVecDescr_t vecXDescr,vecYDescr; /* descriptor for the dense vectors in y=op(A)x */
211: #endif
212: };
214: /* This is struct holding the relevant data needed to a MatMult */
215: struct Mat_SeqAIJCUSPARSEMultStruct {
216: void *mat; /* opaque pointer to a matrix. This could be either a cusparseHybMat_t or a CsrMatrix */
217: cusparseMatDescr_t descr; /* Data needed to describe the matrix for a multiply */
218: THRUSTINTARRAY *cprowIndices; /* compressed row indices used in the parallel SpMV */
219: PetscScalar *alpha_one; /* pointer to a device "scalar" storing the alpha parameter in the SpMV */
220: PetscScalar *beta_zero; /* pointer to a device "scalar" storing the beta parameter in the SpMV as zero*/
221: PetscScalar *beta_one; /* pointer to a device "scalar" storing the beta parameter in the SpMV as one */
222: #if PETSC_PKG_CUDA_VERSION_GE(11,0,0)
223: cusparseSpMatDescr_t matDescr; /* descriptor for the matrix, used by SpMV and SpMM */
224: Mat_CusparseSpMV cuSpMV[3]; /* different Mat_CusparseSpMV structs for non-transpose, transpose, conj-transpose */
225: Mat_SeqAIJCUSPARSEMultStruct() : matDescr(NULL) {
226: for (int i=0; i<3; i++) cuSpMV[i].initialized = PETSC_FALSE;
227: }
228: #endif
229: };
231: /* This is a larger struct holding all the matrices for a SpMV, and SpMV Transpose */
232: struct Mat_SeqAIJCUSPARSE {
233: Mat_SeqAIJCUSPARSEMultStruct *mat; /* pointer to the matrix on the GPU */
234: Mat_SeqAIJCUSPARSEMultStruct *matTranspose; /* pointer to the matrix on the GPU (for the transpose ... useful for BiCG) */
235: THRUSTARRAY *workVector; /* pointer to a workvector to which we can copy the relevant indices of a vector we want to multiply */
236: THRUSTINTARRAY32 *rowoffsets_gpu; /* rowoffsets on GPU in non-compressed-row format. It is used to convert CSR to CSC */
237: PetscInt nrows; /* number of rows of the matrix seen by GPU */
238: MatCUSPARSEStorageFormat format; /* the storage format for the matrix on the device */
239: PetscBool use_cpu_solve; /* Use AIJ_Seq (I)LU solve */
240: cudaStream_t stream; /* a stream for the parallel SpMV ... this is not owned and should not be deleted */
241: cusparseHandle_t handle; /* a handle to the cusparse library ... this may not be owned (if we're working in parallel i.e. multiGPUs) */
242: PetscObjectState nonzerostate; /* track nonzero state to possibly recreate the GPU matrix */
243: #if PETSC_PKG_CUDA_VERSION_GE(11,0,0)
244: size_t csr2cscBufferSize; /* stuff used to compute the matTranspose above */
245: void *csr2cscBuffer; /* This is used as a C struct and is calloc'ed by PetscNewLog() */
246: cusparseCsr2CscAlg_t csr2cscAlg; /* algorithms can be selected from command line options */
247: cusparseSpMVAlg_t spmvAlg;
248: cusparseSpMMAlg_t spmmAlg;
249: #endif
250: THRUSTINTARRAY *csr2csc_i;
251: PetscSplitCSRDataStructure deviceMat; /* Matrix on device for, eg, assembly */
253: /* Stuff for basic COO support */
254: THRUSTINTARRAY *cooPerm; /* permutation array that sorts the input coo entris by row and col */
255: THRUSTINTARRAY *cooPerm_a; /* ordered array that indicate i-th nonzero (after sorting) is the j-th unique nonzero */
257: /* Stuff for extended COO support */
258: PetscBool use_extended_coo; /* Use extended COO format */
259: 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 */
260: PetscCount *perm_d;
262: Mat_SeqAIJCUSPARSE() : use_extended_coo(PETSC_FALSE), perm_d(NULL), jmap_d(NULL) {}
263: };
265: PETSC_INTERN PetscErrorCode MatSeqAIJCUSPARSECopyToGPU(Mat);
266: PETSC_INTERN PetscErrorCode MatSetPreallocationCOO_SeqAIJCUSPARSE_Basic(Mat,PetscCount,const PetscInt[],const PetscInt[]);
267: PETSC_INTERN PetscErrorCode MatSetValuesCOO_SeqAIJCUSPARSE_Basic(Mat,const PetscScalar[],InsertMode);
268: PETSC_INTERN PetscErrorCode MatSeqAIJCUSPARSEMergeMats(Mat,Mat,MatReuse,Mat*);
269: PETSC_INTERN PetscErrorCode MatSeqAIJCUSPARSETriFactors_Reset(Mat_SeqAIJCUSPARSETriFactors_p*);
271: static inline bool isCudaMem(const void *data)
272: {
273: cudaError_t cerr;
274: struct cudaPointerAttributes attr;
275: enum cudaMemoryType mtype;
276: cerr = cudaPointerGetAttributes(&attr,data); /* Do not check error since before CUDA 11.0, passing a host pointer returns cudaErrorInvalidValue */
277: cudaGetLastError(); /* Reset the last error */
278: #if (CUDART_VERSION < 10000)
279: mtype = attr.memoryType;
280: #else
281: mtype = attr.type;
282: #endif
283: if (cerr == cudaSuccess && mtype == cudaMemoryTypeDevice) return true;
284: else return false;
285: }
287: #endif