1 |
commit: 8fb800d2817e2087b4d2e04e30877c42a8b6cb50 |
2 |
Author: Andrea Arteaga <andyspiros <AT> gmail <DOT> com> |
3 |
AuthorDate: Fri Sep 28 15:52:35 2012 +0000 |
4 |
Commit: Andrea Arteaga <andyspiros <AT> gmail <DOT> com> |
5 |
CommitDate: Fri Sep 28 15:52:35 2012 +0000 |
6 |
URL: http://git.overlays.gentoo.org/gitweb/?p=proj/auto-numerical-bench.git;a=commit;h=8fb800d2 |
7 |
|
8 |
Compelete the Fortran BLAS interface as requested by the current BLAS |
9 |
operations. |
10 |
|
11 |
--- |
12 |
.../NI_internal/FortranDeclarations.hpp | 30 ++++++++- |
13 |
NumericInterface/NI_internal/FortranInterface.hpp | 65 +++++++++++++++++++- |
14 |
2 files changed, 89 insertions(+), 6 deletions(-) |
15 |
|
16 |
diff --git a/NumericInterface/NI_internal/FortranDeclarations.hpp b/NumericInterface/NI_internal/FortranDeclarations.hpp |
17 |
index fce99d2..9fe0b04 100644 |
18 |
--- a/NumericInterface/NI_internal/FortranDeclarations.hpp |
19 |
+++ b/NumericInterface/NI_internal/FortranDeclarations.hpp |
20 |
@@ -45,14 +45,36 @@ extern "C" { |
21 |
* LEVEL 2 BLAS * |
22 |
****************/ |
23 |
|
24 |
- void sgemv_(const char*, const int*, const int*, const float*, const float*, const int*, const float*, const int*, const float*, const float*, const int*); |
25 |
- void dgemv_(const char*, const int*, const int*, const double*, const double*, const int*, const double*, const int*, const double*, const double*, const int*); |
26 |
+ void sgemv_(const char*, const int*, const int*, const float*, const float*, const int*, const float*, const int*, const float*, float*, const int*); |
27 |
+ void dgemv_(const char*, const int*, const int*, const double*, const double*, const int*, const double*, const int*, const double*, double*, const int*); |
28 |
+ |
29 |
+ void ssymv_(const char*, const int*, const float*, const float*, const int*, const float*, const int*, const float*, float*, const int*); |
30 |
+ void dsymv_(const char*, const int*, const double*, const double*, const int*, const double*, const int*, const double*, double*, const int*); |
31 |
|
32 |
void strsv_(const char*, const char*, const char*, const int*, const float*, const int*, float*, const int*); |
33 |
void dtrsv_(const char*, const char*, const char*, const int*, const double*, const int*, double*, const int*); |
34 |
|
35 |
- void sger_(const int*, const int*, const float*, const float*, const int*, const float*, const int*, const float*, const int*); |
36 |
- void dger_(const int*, const int*, const double*, const double*, const int*, const double*, const int*, const double*, const int*); |
37 |
+ void sger_(const int*, const int*, const float*, const float*, const int*, const float*, const int*, float*, const int*); |
38 |
+ void dger_(const int*, const int*, const double*, const double*, const int*, const double*, const int*, double*, const int*); |
39 |
+ |
40 |
+ void ssyr2_(const char*, const int*, const float*, const float*, const int*, const float*, const int*, float*, const int*); |
41 |
+ void dsyr2_(const char*, const int*, const double*, const double*, const int*, const double*, const int*, double*, const int*); |
42 |
+ |
43 |
+ |
44 |
+ |
45 |
+ |
46 |
+ /**************** |
47 |
+ * LEVEL 3 BLAS * |
48 |
+ ****************/ |
49 |
+ |
50 |
+ void sgemm_(const char*, const char*, const int*, const int*, const int*, const float*, const float*, const int*, const float*, const int*, const float*, float*, const int*); |
51 |
+ void dgemm_(const char*, const char*, const int*, const int*, const int*, const double*, const double*, const int*, const double*, const int*, const double*, double*, const int*); |
52 |
+ |
53 |
+ void strmm_(const char*, const char*, const char*, const char*, const int*, const int*, const float*, const float*, const int*, float*, const int*); |
54 |
+ void dtrmm_(const char*, const char*, const char*, const char*, const int*, const int*, const double*, const double*, const int*, double*, const int*); |
55 |
+ |
56 |
+ void strsm_(const char*, const char*, const char*, const char*, const int*, const int*, const float*, const float*, const int*, float*, const int*); |
57 |
+ void dtrsm_(const char*, const char*, const char*, const char*, const int*, const int*, const double*, const double*, const int*, double*, const int*); |
58 |
} |
59 |
|
60 |
|
61 |
|
62 |
diff --git a/NumericInterface/NI_internal/FortranInterface.hpp b/NumericInterface/NI_internal/FortranInterface.hpp |
63 |
index 0bdcf89..39dfa84 100644 |
64 |
--- a/NumericInterface/NI_internal/FortranInterface.hpp |
65 |
+++ b/NumericInterface/NI_internal/FortranInterface.hpp |
66 |
@@ -33,6 +33,9 @@ public: |
67 |
|
68 |
static const int ZERO; |
69 |
static const int ONE; |
70 |
+ static const Scalar fONE; |
71 |
+ static const char NoTrans; |
72 |
+ static const char Trans; |
73 |
|
74 |
public: |
75 |
static std::string name() |
76 |
@@ -78,11 +81,20 @@ public: |
77 |
* LEVEL 2 BLAS * |
78 |
****************/ |
79 |
|
80 |
- static void matrixVector(const char& trans, const int& M, const int& N, |
81 |
+ static void matrixVector(const bool& trans, const int& M, const int& N, |
82 |
const Scalar& alpha, const Scalar* A, const Scalar* x, |
83 |
const Scalar& beta, Scalar* y) |
84 |
{ |
85 |
- FORTFUNC(gemv)(&trans, &M, &N, &alpha, A, &M, x, &ONE, &beta, y, &ONE); |
86 |
+ const int LDA = trans ? N : M; |
87 |
+ const char tA = trans ? Trans : NoTrans; |
88 |
+ FORTFUNC(gemv)(&tA, &M, &N, &alpha, A, &LDA, x, &ONE, &beta, y, &ONE); |
89 |
+ } |
90 |
+ |
91 |
+ static void symmetricMatrixVector(const char& uplo, const int& N, |
92 |
+ const Scalar& alpha, const Scalar* A, const Scalar* x, |
93 |
+ const Scalar& beta, Scalar* y) |
94 |
+ { |
95 |
+ FORTFUNC(symv)(&uplo, &N, &alpha, A, &N, x, &ONE, &beta, y, &ONE); |
96 |
} |
97 |
|
98 |
static void triangularSolveVector(const char& uplo, const char& diag, |
99 |
@@ -96,10 +108,59 @@ public: |
100 |
{ |
101 |
FORTFUNC(ger)(&M, &N, &alpha, x, &ONE, y, &ONE, A, &M); |
102 |
} |
103 |
+ |
104 |
+ static void rank2update(const char& uplo, const int& N, const Scalar& alpha, |
105 |
+ const Scalar* x, const Scalar* y, Scalar* A) |
106 |
+ { |
107 |
+ FORTFUNC(syr2)(&uplo, &N, &alpha, x, &ONE, y, &ONE, A, &N); |
108 |
+ } |
109 |
+ |
110 |
+ |
111 |
+ |
112 |
+ /**************** |
113 |
+ * LEVEL 3 BLAS * |
114 |
+ ****************/ |
115 |
+ |
116 |
+ static void matrixMatrix(const bool& transA, const bool& transB, |
117 |
+ const int& M, const int& N, const int& K, |
118 |
+ const Scalar& alpha, const Scalar* A, const Scalar* B, |
119 |
+ const Scalar& beta, Scalar* C) |
120 |
+ { |
121 |
+ int LDA = M, LDB = K; |
122 |
+ char tA = NoTrans, tB = NoTrans; |
123 |
+ |
124 |
+ if (transA) { |
125 |
+ LDA = K; |
126 |
+ tA = Trans; |
127 |
+ } |
128 |
+ if (transB) { |
129 |
+ LDB = N; |
130 |
+ tB = Trans; |
131 |
+ } |
132 |
+ |
133 |
+ const int LDB = transB ? N : K; |
134 |
+ FORTFUNC(gemm)(&tA, &tB, &M, &N, &K, &alpha, A, &LDA, B, &LDB, |
135 |
+ &beta, C, &M); |
136 |
+ } |
137 |
+ |
138 |
+ static void triangularMatrixMatrix(const char& uplo, |
139 |
+ const int& M, const int& N, const Scalar* A, Scalar* B) |
140 |
+ { |
141 |
+ FORTFUNC(trmm)("L", &uplo, "N", "N", &M, &N, &fONE, A, &M, B, &M); |
142 |
+ } |
143 |
+ |
144 |
+ static void triangularSolveMatrix(const char& uplo, |
145 |
+ const int& M, const int& N, const Scalar* A, Scalar *B) |
146 |
+ { |
147 |
+ FORTFUNC(trsm)("L", &uplo, "N", "N", &M, &N, &fONE, A, &M, B, &M); |
148 |
+ } |
149 |
}; |
150 |
|
151 |
const int NumericInterface<NI_SCALAR>::ZERO = 0; |
152 |
const int NumericInterface<NI_SCALAR>::ONE = 1; |
153 |
+const NI_SCALAR NumericInterface<NI_SCALAR>::fONE = 1.; |
154 |
+const char NumericInterface<NI_SCALAR>::NoTrans = 'N'; |
155 |
+const char NumericInterface<NI_SCALAR>::Trans = 'T'; |