1 /*********************************************************************/
2 /* Copyright 2009, 2010 The University of Texas at Austin. */
3 /* All rights reserved. */
4 /* */
5 /* Redistribution and use in source and binary forms, with or */
6 /* without modification, are permitted provided that the following */
7 /* conditions are met: */
8 /* */
9 /* 1. Redistributions of source code must retain the above */
10 /* copyright notice, this list of conditions and the following */
11 /* disclaimer. */
12 /* */
13 /* 2. Redistributions in binary form must reproduce the above */
14 /* copyright notice, this list of conditions and the following */
15 /* disclaimer in the documentation and/or other materials */
16 /* provided with the distribution. */
17 /* */
18 /* THIS SOFTWARE IS PROVIDED BY THE UNIVERSITY OF TEXAS AT */
19 /* AUSTIN ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, */
20 /* INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF */
21 /* MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE */
22 /* DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY OF TEXAS AT */
23 /* AUSTIN OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, */
24 /* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES */
25 /* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE */
26 /* GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR */
27 /* BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF */
28 /* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT */
29 /* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT */
30 /* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE */
31 /* POSSIBILITY OF SUCH DAMAGE. */
32 /* */
33 /* The views and conclusions contained in the software and */
34 /* documentation are those of the authors and should not be */
35 /* interpreted as representing official policies, either expressed */
36 /* or implied, of The University of Texas at Austin. */
37 /*********************************************************************/
38
39 #include <stdio.h>
40 #include <stdlib.h>
41 #include <math.h>
42 #include "common.h"
43
CNAME(int mode,blas_arg_t * arg,BLASLONG * range_m,BLASLONG * range_n,int (* function)(),void * sa,void * sb,BLASLONG nthreads)44 int CNAME(int mode, blas_arg_t *arg, BLASLONG *range_m, BLASLONG *range_n, int (*function)(), void *sa, void *sb, BLASLONG nthreads) {
45
46 blas_queue_t queue[MAX_CPU_NUMBER];
47 BLASLONG range[MAX_CPU_NUMBER + 1];
48
49 BLASLONG width, i;
50 BLASLONG n_from, n_to;
51 double dnum, nf, nt, di;
52
53 int num_cpu;
54 int mask = 0;
55
56 if (!(mode & BLAS_COMPLEX)) {
57
58 switch (mode & BLAS_PREC) {
59 case BLAS_SINGLE:
60 mask = MAX(SGEMM_UNROLL_M, SGEMM_UNROLL_N) - 1;
61 break;
62 case BLAS_DOUBLE:
63 mask = MAX(DGEMM_UNROLL_M, DGEMM_UNROLL_N) - 1;
64 break;
65 #ifdef EXPRECISION
66 case BLAS_XDOUBLE:
67 mask = MAX(QGEMM_UNROLL_M, QGEMM_UNROLL_N) - 1;
68 break;
69 #endif
70 }
71 } else {
72 switch (mode & BLAS_PREC) {
73 case BLAS_SINGLE:
74 mask = MAX(CGEMM_UNROLL_M, CGEMM_UNROLL_N) - 1;
75 break;
76 case BLAS_DOUBLE:
77 mask = MAX(ZGEMM_UNROLL_M, ZGEMM_UNROLL_N) - 1;
78 break;
79 #ifdef EXPRECISION
80 case BLAS_XDOUBLE:
81 mask = MAX(XGEMM_UNROLL_M, XGEMM_UNROLL_N) - 1;
82 break;
83 #endif
84 }
85 }
86
87 n_from = 0;
88 n_to = arg -> n;
89
90 if (range_n) {
91 n_from = *(range_n + 0);
92 n_to = *(range_n + 1);
93 }
94
95 if (!(mode & BLAS_UPLO)) {
96
97 nf = (double)(n_from);
98 nt = (double)(n_to);
99
100 dnum = (nt * nt - nf * nf) / (double)nthreads;
101
102 num_cpu = 0;
103
104 range[0] = n_from;
105 i = n_from;
106
107 while (i < n_to){
108
109 if (nthreads - num_cpu > 1) {
110
111 di = (double)i;
112 width = ((BLASLONG)( sqrt(di * di + dnum) - di) + mask) & ~mask;
113
114 if ((width <= 0) || (width > n_to - i)) width = n_to - i;
115
116 } else {
117 width = n_to - i;
118 }
119
120 range[num_cpu + 1] = range[num_cpu] + width;
121
122 queue[num_cpu].mode = mode;
123 queue[num_cpu].routine = function;
124 queue[num_cpu].args = arg;
125 queue[num_cpu].range_m = range_m;
126 queue[num_cpu].range_n = &range[num_cpu];
127 queue[num_cpu].sa = NULL;
128 queue[num_cpu].sb = NULL;
129 queue[num_cpu].next = &queue[num_cpu + 1];
130
131 num_cpu ++;
132 i += width;
133 }
134
135 } else {
136
137 nf = (double)(arg -> n - n_from);
138 nt = (double)(arg -> n - n_to);
139
140 dnum = (nt * nt - nf * nf) / (double)nthreads;
141
142 num_cpu = 0;
143
144 range[0] = n_from;
145 i = n_from;
146
147 while (i < n_to){
148
149 if (nthreads - num_cpu > 1) {
150
151 di = (double)(arg -> n - i);
152 width = ((BLASLONG)(-sqrt(di * di + dnum) + di) + mask) & ~mask;
153
154 if ((width <= 0) || (width > n_to - i)) width = n_to - i;
155
156 } else {
157 width = n_to - i;
158 }
159
160 range[num_cpu + 1] = range[num_cpu] + width;
161
162 queue[num_cpu].mode = mode;
163 queue[num_cpu].routine = function;
164 queue[num_cpu].args = arg;
165 queue[num_cpu].range_m = range_m;
166 queue[num_cpu].range_n = &range[num_cpu];
167 queue[num_cpu].sa = NULL;
168 queue[num_cpu].sb = NULL;
169 queue[num_cpu].next = &queue[num_cpu + 1];
170
171 num_cpu ++;
172 i += width;
173 }
174
175 }
176
177 if (num_cpu) {
178 queue[0].sa = sa;
179 queue[0].sb = sb;
180 queue[num_cpu - 1].next = NULL;
181
182 exec_blas(num_cpu, queue);
183 }
184
185 return 0;
186 }
187