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 "common.h"
24 #ifdef FUNCTION_PROFILE
25 #include "functable.h"
26 #endif
27 
28 #ifdef XDOUBLE
29 #define ERROR_NAME "XLAUUM"
30 #elif defined(DOUBLE)
31 #define ERROR_NAME "ZLAUUM"
32 #else
33 #define ERROR_NAME "CLAUUM"
34 #endif
35 
36 static blasint (*lauum_single[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = {
37   LAUUM_U_SINGLE, LAUUM_L_SINGLE,
38 };
39 
40 #ifdef SMP
41 static blasint (*lauum_parallel[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = {
42   LAUUM_U_PARALLEL, LAUUM_L_PARALLEL,
43 };
44 #endif
45 
NAME(char * UPLO,blasint * N,FLOAT * a,blasint * ldA,blasint * Info)46 int NAME(char *UPLO, blasint *N, FLOAT *a, blasint *ldA, blasint *Info){
47 
48   blas_arg_t args;
49 
50   blasint uplo_arg = *UPLO;
51   blasint uplo;
52   blasint info;
53   FLOAT *buffer;
54 #ifdef PPC440
55   extern
56 #endif
57   FLOAT *sa, *sb;
58 
59   PRINT_DEBUG_NAME;
60 
61   args.n    = *N;
62   args.a    = (void *)a;
63   args.lda  = *ldA;
64 
65   TOUPPER(uplo_arg);
66 
67   uplo = -1;
68   if (uplo_arg == 'U') uplo = 0;
69   if (uplo_arg == 'L') uplo = 1;
70 
71   info  = 0;
72   if (args.lda < MAX(1,args.n)) info = 4;
73   if (args.n   < 0)             info = 2;
74   if (uplo     < 0)             info = 1;
75   if (info) {
76     BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
77     *Info = - info;
78     return 0;
79   }
80 
81   *Info = 0;
82 
83   if (args.n == 0) return 0;
84 
85   IDEBUG_START;
86 
87   FUNCTION_PROFILE_START();
88 
89 #ifndef PPC440
90   buffer = (FLOAT *)blas_memory_alloc(1);
91 
92   sa = (FLOAT *)((BLASLONG)buffer + GEMM_OFFSET_A);
93   sb = (FLOAT *)(((BLASLONG)sa + ((GEMM_P * GEMM_Q * COMPSIZE * SIZE + GEMM_ALIGN) & ~GEMM_ALIGN)) + GEMM_OFFSET_B);
94 #endif
95 
96 #ifdef SMP
97   args.common = NULL;
98   args.nthreads = num_cpu_avail(4);
99 
100   if (args.nthreads == 1) {
101 #endif
102 
103     *Info = (lauum_single[uplo])(&args, NULL, NULL, sa, sb, 0);
104 
105 #ifdef SMP
106   } else {
107 
108     *Info = (lauum_parallel[uplo])(&args, NULL, NULL, sa, sb, 0);
109 
110   }
111 #endif
112 
113 #ifndef PPC440
114   blas_memory_free(buffer);
115 #endif
116 
117   FUNCTION_PROFILE_END(1, .5 * args.n * args.n,
118 		       2. * args.n * (1./3. + args.n * ( 1./2. + args.n * 1./6.))
119 		       +  args.n * (args.n * args.n - 1));
120 
121   IDEBUG_END;
122 
123   return 0;
124 }
125