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 "XTRTI2"
30 #elif defined(DOUBLE)
31 #define ERROR_NAME "ZTRTI2"
32 #else
33 #define ERROR_NAME "CTRTI2"
34 #endif
35 
36 static blasint (*trti2[])(blas_arg_t *, BLASLONG *, BLASLONG *, FLOAT *, FLOAT *, BLASLONG) = {
37 #ifdef XDOUBLE
38   xtrti2_UU, xtrti2_UN, xtrti2_LU, xtrti2_LN,
39 #elif defined(DOUBLE)
40   ztrti2_UU, ztrti2_UN, ztrti2_LU, ztrti2_LN,
41 #else
42   ctrti2_UU, ctrti2_UN, ctrti2_LU, ctrti2_LN,
43 #endif
44   };
45 
NAME(char * UPLO,char * DIAG,blasint * N,FLOAT * a,blasint * ldA,blasint * Info)46 int NAME(char *UPLO, char *DIAG, blasint *N, FLOAT *a, blasint *ldA, blasint *Info){
47 
48   blas_arg_t args;
49 
50   blasint uplo_arg = *UPLO;
51   blasint diag_arg = *DIAG;
52   blasint uplo, diag;
53   blasint info;
54   FLOAT *buffer;
55 #ifdef PPC440
56   extern
57 #endif
58   FLOAT *sa, *sb;
59 
60   PRINT_DEBUG_NAME;
61 
62   args.n    = *N;
63   args.a    = (void *)a;
64   args.lda  = *ldA;
65 
66   TOUPPER(uplo_arg);
67   TOUPPER(diag_arg);
68 
69   uplo = -1;
70   if (uplo_arg == 'U') uplo = 0;
71   if (uplo_arg == 'L') uplo = 1;
72   diag = -1;
73   if (diag_arg == 'U') diag = 0;
74   if (diag_arg == 'N') diag = 1;
75 
76   info  = 0;
77   if (args.lda  < MAX(1,args.n)) info = 5;
78   if (args.n    < 0)             info = 3;
79   if (diag < 0)                  info = 2;
80   if (uplo < 0)                  info = 1;
81   if (info) {
82     BLASFUNC(xerbla)(ERROR_NAME, &info, sizeof(ERROR_NAME));
83     *Info = - info;
84     return 0;
85   }
86 
87   *Info = 0;
88 
89   if (args.n <= 0) return 0;
90 
91   IDEBUG_START;
92 
93   FUNCTION_PROFILE_START();
94 
95 #ifndef PPC440
96   buffer = (FLOAT *)blas_memory_alloc(1);
97 
98   sa = (FLOAT *)((BLASLONG)buffer + GEMM_OFFSET_A);
99   sb = (FLOAT *)(((BLASLONG)sa + ((GEMM_P * GEMM_Q * COMPSIZE * SIZE + GEMM_ALIGN) & ~GEMM_ALIGN)) + GEMM_OFFSET_B);
100 #endif
101 
102   info = (trti2[(uplo << 1) | diag])(&args, NULL, NULL, sa, sb, 0);
103 
104   *Info = info;
105 
106 #ifndef PPC440
107   blas_memory_free(buffer);
108 #endif
109 
110   FUNCTION_PROFILE_END(1, .5 * args.n * args.n,
111 		          2. * args.n * (1./3. + args.n * ( 1./2. + args.n * 1./6.))
112 		       +  6. * args.n * (1./3. + args.n * (-1./2. + args.n * 1./6.)));
113 
114   IDEBUG_END;
115 
116   return 0;
117 }
118