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 "XPOTRI"
30 #elif defined(DOUBLE)
31 #define ERROR_NAME "ZPOTRI"
32 #else
33 #define ERROR_NAME "CPOTRI"
34 #endif
35
36 static blasint (*trtri_single[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) ={
37 TRTRI_UN_SINGLE, TRTRI_LN_SINGLE,
38 };
39
40 static blasint (*lauum_single[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) ={
41 LAUUM_U_SINGLE, LAUUM_L_SINGLE,
42 };
43
44 #ifdef SMP
45 static blasint (*trtri_parallel[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) ={
46 TRTRI_UN_PARALLEL, TRTRI_LN_PARALLEL,
47 };
48
49 static blasint (*lauum_parallel[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) ={
50 LAUUM_U_PARALLEL, LAUUM_L_PARALLEL,
51 };
52 #endif
53
NAME(char * UPLO,blasint * N,FLOAT * a,blasint * ldA,blasint * Info)54 int NAME(char *UPLO, blasint *N, FLOAT *a, blasint *ldA, blasint *Info){
55
56 blas_arg_t args;
57
58 blasint uplo_arg = *UPLO;
59 blasint uplo;
60 blasint info;
61 FLOAT *buffer;
62 #ifdef PPC440
63 extern
64 #endif
65 FLOAT *sa, *sb;
66
67 PRINT_DEBUG_NAME;
68
69 args.n = *N;
70 args.a = (void *)a;
71 args.lda = *ldA;
72
73 TOUPPER(uplo_arg);
74
75 uplo = -1;
76 if (uplo_arg == 'U') uplo = 0;
77 if (uplo_arg == 'L') uplo = 1;
78
79 info = 0;
80 if (args.lda < MAX(1,args.n)) info = 4;
81 if (args.n < 0) info = 2;
82 if (uplo < 0) info = 1;
83
84 if (info) {
85 BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
86 *Info = - info;
87 return 0;
88 }
89
90 *Info = 0;
91
92 if (args.n == 0) return 0;
93
94 IDEBUG_START;
95
96 FUNCTION_PROFILE_START();
97
98 #ifndef PPC440
99 buffer = (FLOAT *)blas_memory_alloc(1);
100
101 sa = (FLOAT *)((BLASLONG)buffer + GEMM_OFFSET_A);
102 sb = (FLOAT *)(((BLASLONG)sa + ((GEMM_P * GEMM_Q * COMPSIZE * SIZE + GEMM_ALIGN) & ~GEMM_ALIGN)) + GEMM_OFFSET_B);
103 #endif
104
105 #ifdef SMP
106 args.nthreads = num_cpu_avail(4);
107
108 if (args.nthreads == 1) {
109 #endif
110
111 info = (trtri_single[uplo])(&args, NULL, NULL, sa, sb, 0);
112
113 if (!info) {
114 info = (lauum_single[uplo])(&args, NULL, NULL, sa, sb, 0);
115 }
116
117 *Info = info;
118
119 #ifdef SMP
120 } else {
121 info = (trtri_parallel[uplo])(&args, NULL, NULL, sa, sb, 0);
122
123 if (!info) {
124 info = (lauum_parallel[uplo])(&args, NULL, NULL, sa, sb, 0);
125 }
126
127 *Info = info;
128 }
129 #endif
130
131 #ifndef PPC440
132 blas_memory_free(buffer);
133 #endif
134
135 FUNCTION_PROFILE_END(COMPSIZE * COMPSIZE, args.m * args.n, 2. / 3. * args.m * args.n * args.n);
136
137 IDEBUG_END;
138
139 return 0;
140 }
141