1 /*********************************************************************/
2 /* */
3 /* Optimized BLAS libraries */
4 /* By Kazushige Goto <kgoto@tacc.utexas.edu> */
5 /* */
6 /* Copyright (c) The University of Texas, 2009. All rights reserved. */
7 /* UNIVERSITY EXPRESSLY DISCLAIMS ANY AND ALL WARRANTIES CONCERNING */
8 /* THIS SOFTWARE AND DOCUMENTATION, INCLUDING ANY WARRANTIES OF */
9 /* MERCHANTABILITY, FITNESS FOR ANY PARTICULAR PURPOSE, */
10 /* NON-INFRINGEMENT AND WARRANTIES OF PERFORMANCE, AND ANY WARRANTY */
11 /* THAT MIGHT OTHERWISE ARISE FROM COURSE OF DEALING OR USAGE OF */
12 /* TRADE. NO WARRANTY IS EITHER EXPRESS OR IMPLIED WITH RESPECT TO */
13 /* THE USE OF THE SOFTWARE OR DOCUMENTATION. */
14 /* Under no circumstances shall University be liable for incidental, */
15 /* special, indirect, direct or consequential damages or loss of */
16 /* profits, interruption of business, or related expenses which may */
17 /* arise from use of Software or Documentation, including but not */
18 /* limited to those resulting from defects in Software and/or */
19 /* Documentation, or loss or inaccuracy of data of any kind. */
20 /*********************************************************************/
21
22 #include <stdio.h>
23 #include <ctype.h>
24 #include "common.h"
25 #ifdef FUNCTION_PROFILE
26 #include "functable.h"
27 #endif
28
29 #ifdef XDOUBLE
30 #define ERROR_NAME "QSYR2 "
31 #elif defined(DOUBLE)
32 #define ERROR_NAME "DSYR2 "
33 #else
34 #define ERROR_NAME "SSYR2 "
35 #endif
36
37 static int (*syr2[])(BLASLONG, FLOAT, FLOAT *, BLASLONG, FLOAT *, BLASLONG, FLOAT *, BLASLONG, FLOAT *) = {
38 #ifdef XDOUBLE
39 qsyr2_U, qsyr2_L,
40 #elif defined(DOUBLE)
41 dsyr2_U, dsyr2_L,
42 #else
43 ssyr2_U, ssyr2_L,
44 #endif
45 };
46
47 #ifdef SMP
48 static int (*syr2_thread[])(BLASLONG, FLOAT, FLOAT *, BLASLONG, FLOAT *, BLASLONG, FLOAT *, BLASLONG, FLOAT *, int) = {
49 #ifdef XDOUBLE
50 qsyr2_thread_U, qsyr2_thread_L,
51 #elif defined(DOUBLE)
52 dsyr2_thread_U, dsyr2_thread_L,
53 #else
54 ssyr2_thread_U, ssyr2_thread_L,
55 #endif
56 };
57 #endif
58
59 #ifndef CBLAS
60
NAME(char * UPLO,blasint * N,FLOAT * ALPHA,FLOAT * x,blasint * INCX,FLOAT * y,blasint * INCY,FLOAT * a,blasint * LDA)61 void NAME(char *UPLO, blasint *N, FLOAT *ALPHA,
62 FLOAT *x, blasint *INCX, FLOAT *y, blasint *INCY, FLOAT *a, blasint *LDA){
63
64 char uplo_arg = *UPLO;
65 blasint n = *N;
66 FLOAT alpha = *ALPHA;
67 blasint lda = *LDA;
68 blasint incx = *INCX;
69 blasint incy = *INCY;
70
71 blasint info;
72 int uplo;
73 FLOAT *buffer;
74 #ifdef SMP
75 int nthreads;
76 #endif
77
78 PRINT_DEBUG_NAME;
79
80 TOUPPER(uplo_arg);
81 uplo = -1;
82
83 if (uplo_arg == 'U') uplo = 0;
84 if (uplo_arg == 'L') uplo = 1;
85
86 info = 0;
87
88 if (lda < MAX(1, n)) info = 9;
89 if (incy == 0) info = 7;
90 if (incx == 0) info = 5;
91 if (n < 0) info = 2;
92 if (uplo < 0) info = 1;
93
94 if (info != 0) {
95 BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
96 return;
97 }
98
99 #else
100
101 void CNAME(enum CBLAS_ORDER order, enum CBLAS_UPLO Uplo, blasint n, FLOAT alpha, FLOAT *x, blasint incx, FLOAT *y, blasint incy, FLOAT *a, blasint lda) {
102
103 FLOAT *buffer;
104 int trans, uplo;
105 blasint info;
106 #ifdef SMP
107 int nthreads;
108 #endif
109
110 PRINT_DEBUG_CNAME;
111
112 trans = -1;
113 uplo = -1;
114 info = 0;
115
116 if (order == CblasColMajor) {
117
118 if (Uplo == CblasUpper) uplo = 0;
119 if (Uplo == CblasLower) uplo = 1;
120
121 info = -1;
122
123 if (lda < MAX(1, n)) info = 9;
124 if (incy == 0) info = 7;
125 if (incx == 0) info = 5;
126 if (n < 0) info = 2;
127 if (uplo < 0) info = 1;
128 }
129
130 if (order == CblasRowMajor) {
131
132 if (Uplo == CblasUpper) uplo = 1;
133 if (Uplo == CblasLower) uplo = 0;
134
135 info = -1;
136
137 if (lda < MAX(1, n)) info = 9;
138 if (incy == 0) info = 7;
139 if (incx == 0) info = 5;
140 if (n < 0) info = 2;
141 if (uplo < 0) info = 1;
142 }
143
144 if (info >= 0) {
145 BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
146 return;
147 }
148
149 #endif
150
151 if (n == 0) return;
152
153 if (alpha == ZERO) return;
154
155 IDEBUG_START;
156
157 FUNCTION_PROFILE_START();
158
159 if (incx < 0 ) x -= (n - 1) * incx;
160 if (incy < 0 ) y -= (n - 1) * incy;
161
162 buffer = (FLOAT *)blas_memory_alloc(1);
163
164 #ifdef SMP
165 nthreads = num_cpu_avail(2);
166
167 if (nthreads == 1) {
168 #endif
169
170 (syr2[uplo])(n, alpha, x, incx, y, incy, a, lda, buffer);
171
172 #ifdef SMP
173 } else {
174
175 (syr2_thread[uplo])(n, alpha, x, incx, y, incy, a, lda, buffer, nthreads);
176
177 }
178 #endif
179
180 blas_memory_free(buffer);
181
182 FUNCTION_PROFILE_END(1, n * n / 2 + 2 * n, 2 * n * n);
183
184 IDEBUG_END;
185
186 return;
187 }
188