[136] | 1 |
|
---|
| 2 | #define BLAS_FUNC(NAME) CAT(CAT(SCALAR_PREFIX,NAME),_)
|
---|
| 3 |
|
---|
| 4 | template<> class blas_interface<SCALAR> : public c_interface_base<SCALAR>
|
---|
| 5 | {
|
---|
| 6 |
|
---|
| 7 | public :
|
---|
| 8 |
|
---|
| 9 | static SCALAR fone;
|
---|
| 10 | static SCALAR fzero;
|
---|
| 11 |
|
---|
| 12 | static inline std::string name()
|
---|
| 13 | {
|
---|
| 14 | return MAKE_STRING(CBLASNAME);
|
---|
| 15 | }
|
---|
| 16 |
|
---|
| 17 | static inline void matrix_vector_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){
|
---|
| 18 | BLAS_FUNC(gemv)(¬rans,&N,&N,&fone,A,&N,B,&intone,&fzero,X,&intone);
|
---|
| 19 | }
|
---|
| 20 |
|
---|
| 21 | static inline void symv(gene_matrix & A, gene_vector & B, gene_vector & X, int N){
|
---|
| 22 | BLAS_FUNC(symv)(&lower, &N,&fone,A,&N,B,&intone,&fzero,X,&intone);
|
---|
| 23 | }
|
---|
| 24 |
|
---|
| 25 | static inline void syr2(gene_matrix & A, gene_vector & B, gene_vector & X, int N){
|
---|
| 26 | BLAS_FUNC(syr2)(&lower,&N,&fone,B,&intone,X,&intone,A,&N);
|
---|
| 27 | }
|
---|
| 28 |
|
---|
| 29 | static inline void ger(gene_matrix & A, gene_vector & X, gene_vector & Y, int N){
|
---|
| 30 | BLAS_FUNC(ger)(&N,&N,&fone,X,&intone,Y,&intone,A,&N);
|
---|
| 31 | }
|
---|
| 32 |
|
---|
| 33 | static inline void rot(gene_vector & A, gene_vector & B, SCALAR c, SCALAR s, int N){
|
---|
| 34 | BLAS_FUNC(rot)(&N,A,&intone,B,&intone,&c,&s);
|
---|
| 35 | }
|
---|
| 36 |
|
---|
| 37 | static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){
|
---|
| 38 | BLAS_FUNC(gemv)(&trans,&N,&N,&fone,A,&N,B,&intone,&fzero,X,&intone);
|
---|
| 39 | }
|
---|
| 40 |
|
---|
| 41 | static inline void matrix_matrix_product(gene_matrix & A, gene_matrix & B, gene_matrix & X, int N){
|
---|
| 42 | BLAS_FUNC(gemm)(¬rans,¬rans,&N,&N,&N,&fone,A,&N,B,&N,&fzero,X,&N);
|
---|
| 43 | }
|
---|
| 44 |
|
---|
| 45 | static inline void transposed_matrix_matrix_product(gene_matrix & A, gene_matrix & B, gene_matrix & X, int N){
|
---|
| 46 | BLAS_FUNC(gemm)(¬rans,¬rans,&N,&N,&N,&fone,A,&N,B,&N,&fzero,X,&N);
|
---|
| 47 | }
|
---|
| 48 |
|
---|
| 49 | // static inline void ata_product(gene_matrix & A, gene_matrix & X, int N){
|
---|
| 50 | // ssyrk_(&lower,&trans,&N,&N,&fone,A,&N,&fzero,X,&N);
|
---|
| 51 | // }
|
---|
| 52 |
|
---|
| 53 | static inline void aat_product(gene_matrix & A, gene_matrix & X, int N){
|
---|
| 54 | BLAS_FUNC(syrk)(&lower,¬rans,&N,&N,&fone,A,&N,&fzero,X,&N);
|
---|
| 55 | }
|
---|
| 56 |
|
---|
| 57 | static inline void axpy(SCALAR coef, const gene_vector & X, gene_vector & Y, int N){
|
---|
| 58 | BLAS_FUNC(axpy)(&N,&coef,X,&intone,Y,&intone);
|
---|
| 59 | }
|
---|
| 60 |
|
---|
| 61 | static inline void axpby(SCALAR a, const gene_vector & X, SCALAR b, gene_vector & Y, int N){
|
---|
| 62 | BLAS_FUNC(scal)(&N,&b,Y,&intone);
|
---|
| 63 | BLAS_FUNC(axpy)(&N,&a,X,&intone,Y,&intone);
|
---|
| 64 | }
|
---|
| 65 |
|
---|
| 66 | static inline void cholesky(const gene_matrix & X, gene_matrix & C, int N){
|
---|
| 67 | int N2 = N*N;
|
---|
| 68 | BLAS_FUNC(copy)(&N2, X, &intone, C, &intone);
|
---|
| 69 | char uplo = 'L';
|
---|
| 70 | int info = 0;
|
---|
| 71 | BLAS_FUNC(potrf)(&uplo, &N, C, &N, &info);
|
---|
| 72 | if(info!=0) std::cerr << "potrf_ error " << info << "\n";
|
---|
| 73 | }
|
---|
| 74 |
|
---|
| 75 | static inline void partial_lu_decomp(const gene_matrix & X, gene_matrix & C, int N){
|
---|
| 76 | int N2 = N*N;
|
---|
| 77 | BLAS_FUNC(copy)(&N2, X, &intone, C, &intone);
|
---|
| 78 | char uplo = 'L';
|
---|
| 79 | int info = 0;
|
---|
| 80 | int * ipiv = (int*)alloca(sizeof(int)*N);
|
---|
| 81 | BLAS_FUNC(getrf)(&N, &N, C, &N, ipiv, &info);
|
---|
| 82 | if(info!=0) std::cerr << "getrf_ error " << info << "\n";
|
---|
| 83 | }
|
---|
| 84 |
|
---|
| 85 | static inline void trisolve_lower(const gene_matrix & L, const gene_vector& B, gene_vector & X, int N){
|
---|
| 86 | BLAS_FUNC(copy)(&N, B, &intone, X, &intone);
|
---|
| 87 | BLAS_FUNC(trsv)(&lower, ¬rans, &nonunit, &N, L, &N, X, &intone);
|
---|
| 88 | }
|
---|
| 89 |
|
---|
| 90 | static inline void trisolve_lower_matrix(const gene_matrix & L, const gene_matrix& B, gene_matrix & X, int N){
|
---|
| 91 | BLAS_FUNC(copy)(&N, B, &intone, X, &intone);
|
---|
| 92 | BLAS_FUNC(trsm)(&right, &lower, ¬rans, &nonunit, &N, &N, &fone, L, &N, X, &N);
|
---|
| 93 | }
|
---|
| 94 |
|
---|
| 95 | static inline void trmm(gene_matrix & A, gene_matrix & B, gene_matrix & X, int N){
|
---|
| 96 | BLAS_FUNC(trmm)(&left, &lower, ¬rans,&nonunit, &N,&N,&fone,A,&N,B,&N);
|
---|
| 97 | }
|
---|
| 98 |
|
---|
| 99 | #ifdef HAS_LAPACK
|
---|
| 100 |
|
---|
| 101 | static inline void lu_decomp(const gene_matrix & X, gene_matrix & C, int N){
|
---|
| 102 | int N2 = N*N;
|
---|
| 103 | BLAS_FUNC(copy)(&N2, X, &intone, C, &intone);
|
---|
| 104 | char uplo = 'L';
|
---|
| 105 | int info = 0;
|
---|
| 106 | int * ipiv = (int*)alloca(sizeof(int)*N);
|
---|
| 107 | int * jpiv = (int*)alloca(sizeof(int)*N);
|
---|
| 108 | BLAS_FUNC(getc2)(&N, C, &N, ipiv, jpiv, &info);
|
---|
| 109 | }
|
---|
| 110 |
|
---|
| 111 |
|
---|
| 112 |
|
---|
| 113 | static inline void hessenberg(const gene_matrix & X, gene_matrix & C, int N){
|
---|
| 114 | {
|
---|
| 115 | int N2 = N*N;
|
---|
| 116 | int inc = 1;
|
---|
| 117 | BLAS_FUNC(copy)(&N2, X, &inc, C, &inc);
|
---|
| 118 | }
|
---|
| 119 | int info = 0;
|
---|
| 120 | int ilo = 1;
|
---|
| 121 | int ihi = N;
|
---|
| 122 | int bsize = 64;
|
---|
| 123 | int worksize = N*bsize;
|
---|
| 124 | SCALAR* d = new SCALAR[N+worksize];
|
---|
| 125 | BLAS_FUNC(gehrd)(&N, &ilo, &ihi, C, &N, d, d+N, &worksize, &info);
|
---|
| 126 | delete[] d;
|
---|
| 127 | }
|
---|
| 128 |
|
---|
| 129 | static inline void tridiagonalization(const gene_matrix & X, gene_matrix & C, int N){
|
---|
| 130 | {
|
---|
| 131 | int N2 = N*N;
|
---|
| 132 | int inc = 1;
|
---|
| 133 | BLAS_FUNC(copy)(&N2, X, &inc, C, &inc);
|
---|
| 134 | }
|
---|
| 135 | char uplo = 'U';
|
---|
| 136 | int info = 0;
|
---|
| 137 | int ilo = 1;
|
---|
| 138 | int ihi = N;
|
---|
| 139 | int bsize = 64;
|
---|
| 140 | int worksize = N*bsize;
|
---|
| 141 | SCALAR* d = new SCALAR[3*N+worksize];
|
---|
| 142 | BLAS_FUNC(sytrd)(&uplo, &N, C, &N, d, d+N, d+2*N, d+3*N, &worksize, &info);
|
---|
| 143 | delete[] d;
|
---|
| 144 | }
|
---|
| 145 |
|
---|
| 146 | #endif // HAS_LAPACK
|
---|
| 147 |
|
---|
| 148 | };
|
---|
| 149 |
|
---|
| 150 | SCALAR blas_interface<SCALAR>::fone = SCALAR(1);
|
---|
| 151 | SCALAR blas_interface<SCALAR>::fzero = SCALAR(0);
|
---|