![[BACK]](/icons/cvsweb/back.gif){kind=icon}
Return to mat.c CVS log ![[TXT]](/icons/cvsweb/text.gif){kind=icon}
![[DIR]](/icons/cvsweb/dir.gif){kind=icon}
Up to [local] / OpenXM_contrib2 / asir2000 / engine|
Return to mat.c CVS log | Up to [local] / OpenXM_contrib2 / asir2000 / engine |
1.2 noro 1: /*
2: * Copyright (c) 1994-2000 FUJITSU LABORATORIES LIMITED
3: * All rights reserved.
4: *
5: * FUJITSU LABORATORIES LIMITED ("FLL") hereby grants you a limited,
6: * non-exclusive and royalty-free license to use, copy, modify and
7: * redistribute, solely for non-commercial and non-profit purposes, the
8: * computer program, "Risa/Asir" ("SOFTWARE"), subject to the terms and
9: * conditions of this Agreement. For the avoidance of doubt, you acquire
10: * only a limited right to use the SOFTWARE hereunder, and FLL or any
11: * third party developer retains all rights, including but not limited to
12: * copyrights, in and to the SOFTWARE.
13: *
14: * (1) FLL does not grant you a license in any way for commercial
15: * purposes. You may use the SOFTWARE only for non-commercial and
16: * non-profit purposes only, such as academic, research and internal
17: * business use.
18: * (2) The SOFTWARE is protected by the Copyright Law of Japan and
19: * international copyright treaties. If you make copies of the SOFTWARE,
20: * with or without modification, as permitted hereunder, you shall affix
21: * to all such copies of the SOFTWARE the above copyright notice.
22: * (3) An explicit reference to this SOFTWARE and its copyright owner
23: * shall be made on your publication or presentation in any form of the
24: * results obtained by use of the SOFTWARE.
25: * (4) In the event that you modify the SOFTWARE, you shall notify FLL by
1.3 noro 26: * e-mail at risa-admin@sec.flab.fujitsu.co.jp of the detailed specification
1.2 noro 27: * for such modification or the source code of the modified part of the
28: * SOFTWARE.
29: *
30: * THE SOFTWARE IS PROVIDED AS IS WITHOUT ANY WARRANTY OF ANY KIND. FLL
31: * MAKES ABSOLUTELY NO WARRANTIES, EXPRESSED, IMPLIED OR STATUTORY, AND
32: * EXPRESSLY DISCLAIMS ANY IMPLIED WARRANTY OF MERCHANTABILITY, FITNESS
33: * FOR A PARTICULAR PURPOSE OR NONINFRINGEMENT OF THIRD PARTIES'
34: * RIGHTS. NO FLL DEALER, AGENT, EMPLOYEES IS AUTHORIZED TO MAKE ANY
35: * MODIFICATIONS, EXTENSIONS, OR ADDITIONS TO THIS WARRANTY.
36: * UNDER NO CIRCUMSTANCES AND UNDER NO LEGAL THEORY, TORT, CONTRACT,
37: * OR OTHERWISE, SHALL FLL BE LIABLE TO YOU OR ANY OTHER PERSON FOR ANY
38: * DIRECT, INDIRECT, SPECIAL, INCIDENTAL, PUNITIVE OR CONSEQUENTIAL
39: * DAMAGES OF ANY CHARACTER, INCLUDING, WITHOUT LIMITATION, DAMAGES
40: * ARISING OUT OF OR RELATING TO THE SOFTWARE OR THIS AGREEMENT, DAMAGES
41: * FOR LOSS OF GOODWILL, WORK STOPPAGE, OR LOSS OF DATA, OR FOR ANY
42: * DAMAGES, EVEN IF FLL SHALL HAVE BEEN INFORMED OF THE POSSIBILITY OF
43: * SUCH DAMAGES, OR FOR ANY CLAIM BY ANY OTHER PARTY. EVEN IF A PART
44: * OF THE SOFTWARE HAS BEEN DEVELOPED BY A THIRD PARTY, THE THIRD PARTY
45: * DEVELOPER SHALL HAVE NO LIABILITY IN CONNECTION WITH THE USE,
46: * PERFORMANCE OR NON-PERFORMANCE OF THE SOFTWARE.
47: *
1.15 ! noro 48: * $OpenXM: OpenXM_contrib2/asir2000/engine/mat.c,v 1.14 2005/06/03 07:16:16 saito Exp $
1.2 noro 49: */
1.1 noro 50: #include "ca.h"
1.4 saito 51: #include "../parse/parse.h"
52:
53: extern int StrassenSize;
1.14 saito 54: /* remove miser type
1.12 saito 55: void mulmatmat_miser();
1.14 saito 56: */
1.1 noro 57:
58: void addmat(vl,a,b,c)
59: VL vl;
60: MAT a,b,*c;
61: {
62: int row,col,i,j;
1.4 saito 63: MAT t;
64: pointer *ab,*bb,*tb;
65:
66: if ( !a )
67: *c = b;
68: else if ( !b )
69: *c = a;
70: else if ( (a->row != b->row) || (a->col != b->col) ) {
71: *c = 0; error("addmat : size mismatch add");
72: } else {
73: row = a->row; col = a->col;
74: MKMAT(t,row,col);
75: for ( i = 0; i < row; i++ )
76: for ( j = 0, ab = BDY(a)[i], bb = BDY(b)[i], tb = BDY(t)[i];
77: j < col; j++ )
1.9 noro 78: arf_add(vl,(Obj)ab[j],(Obj)bb[j],(Obj *)&tb[j]);
1.4 saito 79: *c = t;
80: }
1.1 noro 81: }
82:
83: void submat(vl,a,b,c)
84: VL vl;
85: MAT a,b,*c;
86: {
87: int row,col,i,j;
1.4 saito 88: MAT t;
89: pointer *ab,*bb,*tb;
1.1 noro 90:
1.4 saito 91: if ( !a )
92: chsgnmat(b,c);
93: else if ( !b )
94: *c = a;
95: else if ( (a->row != b->row) || (a->col != b->col) ) {
96: *c = 0; error("submat : size mismatch sub");
97: } else {
98: row = a->row; col = a->col;
99: MKMAT(t,row,col);
100: for ( i = 0; i < row; i++ )
101: for ( j = 0, ab = BDY(a)[i], bb = BDY(b)[i], tb = BDY(t)[i];
102: j < col; j++ )
1.9 noro 103: arf_sub(vl,(Obj)ab[j],(Obj)bb[j],(Obj *)&tb[j]);
1.4 saito 104: *c = t;
105: }
1.1 noro 106: }
107:
1.14 saito 108: /* remove miser type
1.11 saito 109: void addmat_miser(vl,a,b,c,ar0,ac0,ar1,ac1,br0,bc0,br1,bc1)
110: VL vl;
111: MAT a,b,*c;
112: int ar0,ac0,ar1,ac1,br0,bc0,br1,bc1;
113: {
114: int row,col,i,j;
115: MAT t;
116: pointer *ab,*bb,*tb;
117: row = ar1 - ar0 + 1; col = ac1 - ac0 + 1;
118:
119: if ( !a )
120: *c = b;
121: else if ( !b )
122: *c = a;
123: else if ( (row != br1 - br0 + 1) || (col != bc1 - bc0 + 1) ) {
124: *c = 0; error("addmat : size mismatch add");
125: } else {
126: MKMAT(t,row,col);
127: for ( i = 0; i < row; i++ ) {
128: if (i+ar0 > a->row-1) {
129: ab = NULL;
130: } else {
131: ab = BDY(a)[i+ar0];
132: }
133: if (i+br0 > b->row-1) {
134: bb = NULL;
135: } else {
136: bb = BDY(b)[i+br0];
137: }
138: tb = BDY(t)[i];
139: for ( j =0; j < col; j++ ) {
140: if ((ab == NULL || j+ac0 > a->col-1) && (bb == NULL || j+bc0 > b->col-1)) {
141: arf_add(vl,NULL,NULL,(Obj *)&tb[j]);
142: } else if ((ab != NULL && j+ac0 <= a->col-1) && (bb == NULL || j+bc0 > b->col-1)){
143: arf_add(vl,(Obj)ab[j+ac0],NULL,(Obj *)&tb[j]);
144: } else if ((ab == NULL || j+ac0 > a->col-1) && (bb != NULL && j+bc0 <= b->col-1)) {
145: arf_add(vl,NULL, (Obj)bb[j+bc0],(Obj *)&tb[j]);
146: } else {
147: arf_add(vl,(Obj)ab[j+ac0],(Obj)bb[j+bc0],(Obj *)&tb[j]);
148: }
149:
150: }
151: }
152: *c = t;
153: }
154: }
155:
156: void submat_miser(vl,a,b,c,ar0,ac0,ar1,ac1,br0,bc0,br1,bc1)
157: VL vl;
158: MAT a,b,*c;
159: int ar0,ac0,ar1,ac1,br0,bc0,br1,bc1;
160: {
161: int row,col,i,j;
162: MAT t;
163: pointer *ab,*bb,*tb;
164:
165: row = ar1 - ar0 + 1; col = ac1 - ac0 + 1;
166:
167: if ( !a )
168: chsgnmat(b,c);
169: else if ( !b )
170: *c = a;
171: else if ( (row != br1 - br0 + 1) || (col != bc1 - bc0 + 1) ) {
172: *c = 0; error("submat : size mismatch sub");
173: } else {
174: MKMAT(t,row,col);
175: for ( i = 0; i < row; i++ ) {
176: if (i+ar0 > a->row-1) {
177: ab = NULL;
178: } else {
179: ab = BDY(a)[i+ar0];
180: }
181: if (i+br0 > b->row-1) {
182: bb = NULL;
183: } else {
184: bb = BDY(b)[i+br0];
185: }
186: tb = BDY(t)[i];
187: for ( j =0; j < col; j++ ) {
188: if ((ab == NULL || j+ac0 > a->col-1) && (bb == NULL || j+bc0 > b->col-1)) {
189: arf_sub(vl,NULL,NULL,(Obj *)&tb[j]);
190: } else if ((ab != NULL && j+ac0 <= a->col-1) && (bb == NULL || j+bc0 > b->col-1)){
191: arf_sub(vl,(Obj)ab[j+ac0],NULL,(Obj *)&tb[j]);
192: } else if ((ab == NULL || j+ac0 > a->col-1) && (bb != NULL && j+bc0 <= b->col-1)) {
193: arf_sub(vl,NULL, (Obj)bb[j+bc0],(Obj *)&tb[j]);
194: } else {
195: arf_sub(vl,(Obj)ab[j+ac0],(Obj)bb[j+bc0],(Obj *)&tb[j]);
196: }
197:
198: }
199: }
200: *c = t;
201: }
202: }
1.14 saito 203: */
1.11 saito 204:
1.1 noro 205: void mulmat(vl,a,b,c)
206: VL vl;
207: Obj a,b,*c;
208: {
1.8 noro 209: VECT vect;
210: MAT mat;
211:
212: if ( !a && !b )
1.1 noro 213: *c = 0;
1.8 noro 214: else if ( !a || !b ) {
215: if ( !a )
216: a = b;
217: switch ( OID(a) ) {
218: case O_VECT:
219: MKVECT(vect,((VECT)a)->len);
220: *c = (Obj)vect;
221: break;
222: case O_MAT:
223: MKMAT(mat,((MAT)a)->row,((MAT)a)->col);
224: *c = (Obj)mat;
225: break;
226: default:
227: *c = 0;
228: break;
229: }
1.10 noro 230: } else if ( OID(a) <= O_R || OID(a) == O_DP )
1.1 noro 231: mulrmat(vl,(Obj)a,(MAT)b,(MAT *)c);
1.10 noro 232: else if ( OID(b) <= O_R || OID(b) == O_DP )
1.1 noro 233: mulrmat(vl,(Obj)b,(MAT)a,(MAT *)c);
234: else
235: switch ( OID(a) ) {
236: case O_VECT:
237: switch ( OID(b) ) {
238: case O_MAT:
239: mulvectmat(vl,(VECT)a,(MAT)b,(VECT *)c); break;
240: case O_VECT: default:
241: notdef(vl,a,b,c); break;
242: }
243: break;
244: case O_MAT:
245: switch ( OID(b) ) {
246: case O_VECT:
247: mulmatvect(vl,(MAT)a,(VECT)b,(VECT *)c); break;
248: case O_MAT:
1.14 saito 249: mulmatmat(vl, (MAT)a, (MAT)b, (MAT *)c); break;
250: /* remove miser type
1.11 saito 251: mulmatmat_miser(vl,(MAT)a,(MAT)b,(MAT *)c, 0,0, ((MAT)a)->row-1, ((MAT)a)->col-1, 0,0,((MAT)b)->row-1, ((MAT)b)->col-1); break;
1.14 saito 252: */
1.1 noro 253: default:
254: notdef(vl,a,b,c); break;
255: }
256: break;
257: default:
258: notdef(vl,a,b,c); break;
259: }
260: }
261:
262: void divmat(vl,a,b,c)
263: VL vl;
264: Obj a,b,*c;
265: {
266: Obj t;
267:
268: if ( !b )
269: error("divmat : division by 0");
270: else if ( !a )
271: *c = 0;
272: else if ( OID(b) > O_R )
273: notdef(vl,a,b,c);
274: else {
1.9 noro 275: arf_div(vl,(Obj)ONE,b,&t); mulrmat(vl,t,(MAT)a,(MAT *)c);
1.1 noro 276: }
277: }
278:
279: void chsgnmat(a,b)
280: MAT a,*b;
281: {
282: MAT t;
283: int row,col,i,j;
284: pointer *ab,*tb;
285:
286: if ( !a )
287: *b = 0;
288: else {
289: row = a->row; col = a->col;
290: MKMAT(t,row,col);
291: for ( i = 0; i < row; i++ )
292: for ( j = 0, ab = BDY(a)[i], tb = BDY(t)[i];
293: j < col; j++ )
1.9 noro 294: arf_chsgn((Obj)ab[j],(Obj *)&tb[j]);
1.1 noro 295: *b = t;
296: }
297: }
298:
299: void pwrmat(vl,a,r,c)
300: VL vl;
301: MAT a;
302: Obj r;
303: MAT *c;
304: {
1.8 noro 305: int n,i;
306: MAT t;
307:
1.1 noro 308: if ( !a )
309: *c = 0;
1.8 noro 310: else if ( !r ) {
311: if ( a->row != a->col ) {
312: *c = 0; error("pwrmat : non square matrix");
313: } else {
314: n = a->row;
315: MKMAT(t,n,n);
316: for ( i = 0; i < n; i++ )
317: t->body[i][i] = ONE;
318: *c = t;
319: }
320: } else if ( !NUM(r) || !RATN(r) ||
1.1 noro 321: !INT(r) || (SGN((Q)r)<0) || (PL(NM((Q)r))>1) ) {
322: *c = 0; error("pwrmat : invalid exponent");
323: } else if ( a->row != a->col ) {
324: *c = 0; error("pwrmat : non square matrix");
325: } else
326: pwrmatmain(vl,a,QTOS((Q)r),c);
327: }
328:
329: void pwrmatmain(vl,a,e,c)
330: VL vl;
331: MAT a;
332: int e;
333: MAT *c;
334: {
335: MAT t,s;
336:
337: if ( e == 1 ) {
338: *c = a;
339: return;
340: }
341:
342: pwrmatmain(vl,a,e/2,&t);
343: mulmat(vl,(Obj)t,(Obj)t,(Obj *)&s);
344: if ( e % 2 )
345: mulmat(vl,(Obj)s,(Obj)a,(Obj *)c);
346: else
347: *c = s;
348: }
349:
350: void mulrmat(vl,a,b,c)
351: VL vl;
352: Obj a;
353: MAT b,*c;
354: {
355: int row,col,i,j;
356: MAT t;
357: pointer *bb,*tb;
358:
359: if ( !a || !b )
360: *c = 0;
361: else {
362: row = b->row; col = b->col;
363: MKMAT(t,row,col);
364: for ( i = 0; i < row; i++ )
365: for ( j = 0, bb = BDY(b)[i], tb = BDY(t)[i];
366: j < col; j++ )
1.9 noro 367: arf_mul(vl,(Obj)a,(Obj)bb[j],(Obj *)&tb[j]);
1.1 noro 368: *c = t;
369: }
370: }
371:
372: void mulmatmat(vl,a,b,c)
373: VL vl;
374: MAT a,b,*c;
375: {
1.4 saito 376: int arow,bcol,i,j,k,m, h, arowh, bcolh;
377: MAT t, a11, a12, a21, a22;
378: MAT p, b11, b12, b21, b22;
379: MAT ans1, ans2, ans3, c11, c12, c21, c22;
380: MAT s1, s2, t1, t2, u1, v1, w1, aa, bb;
381: pointer s,u,v;
382: pointer *ab,*tb;
383: int a1row,a2row, a3row,a4row, a1col, a2col, a3col, a4col;
384: int b1row,b2row, b3row,b4row, b1col, b2col, b3col, b4col;
385: int pflag1, pflag2;
1.5 saito 386: /* mismach col and row */
1.4 saito 387: if ( a->col != b->row ) {
388: *c = 0; error("mulmat : size mismatch");
389: }
390: else {
391: pflag1 = 0; pflag2 = 0;
392: arow = a->row; m = a->col; bcol = b->col;
393: MKMAT(t,arow,bcol);
394: /* StrassenSize == 0 or matrix size less then StrassenSize,
1.5 saito 395: then calc cannonical algorithm. */
396: if((StrassenSize == 0)||(a->row<=StrassenSize || a->col <= StrassenSize) || (b->row<=StrassenSize || b->col <= StrassenSize)) {
1.4 saito 397: for ( i = 0; i < arow; i++ )
398: for ( j = 0, ab = BDY(a)[i], tb = BDY(t)[i]; j < bcol; j++ ) {
399: for ( k = 0, s = 0; k < m; k++ ) {
1.9 noro 400: arf_mul(vl,(Obj)ab[k],(Obj)BDY(b)[k][j],(Obj *)&u);
401: arf_add(vl,(Obj)s,(Obj)u,(Obj *)&v);
1.4 saito 402: s = v;
403: }
404: tb[j] = s;
405: }
406: *c = t;
407: return;
408: }
1.5 saito 409: /* padding odd col and row to even number for zero */
1.4 saito 410: i = arow/2;
411: j = arow - i;
412: if (i != j) {
413: arow++;
414: pflag1 = 1;
415: }
416: i = m/2;
417: j = m - i;
418: if (i != j) {
419: m++;
420: pflag2 = 1;
421: }
422:
1.5 saito 423: /* split matrix A and B */
1.11 saito 424: a1row = arow/2; a1col = m/2;
1.4 saito 425: MKMAT(a11,a1row,a1col);
1.11 saito 426: MKMAT(a21,a1row,a1col);
427: MKMAT(a12,a1row,a1col);
428: MKMAT(a22,a1row,a1col);
1.4 saito 429:
1.11 saito 430: b1row = m/2; b1col = bcol/2;
1.4 saito 431: MKMAT(b11,b1row,b1col);
1.11 saito 432: MKMAT(b21,b1row,b1col);
433: MKMAT(b12,b1row,b1col);
434: MKMAT(b22,b1row,b1col);
1.4 saito 435:
1.5 saito 436: /* make a11 matrix */
1.4 saito 437: for (i = 0; i < a1row; i++) {
438: for (j = 0; j < a1col; j++) {
1.11 saito 439: a11->body[i][j] = a->body[i][j];
1.4 saito 440: }
441: }
442:
1.5 saito 443: /* make a21 matrix */
1.11 saito 444: for (i = a1row; i < a->row; i++) {
1.4 saito 445: for (j = 0; j < a1col; j++) {
1.11 saito 446: a21->body[i-a1row][j] = a->body[i][j];
1.4 saito 447: }
448: }
449:
1.5 saito 450: /* create a12 matrix */
1.4 saito 451: for (i = 0; i < a1row; i++) {
1.11 saito 452: for (j = a1col; j < a->col; j++) {
453: a12->body[i][j-a1col] = a->body[i][j];
1.4 saito 454: }
455: }
456:
1.5 saito 457: /* create a22 matrix */
1.11 saito 458: for (i = a1row; i < a->row; i++) {
459: for (j = a1col; j < a->col; j++) {
460: a22->body[i-a1row][j-a1col] = a->body[i][j];
1.4 saito 461: }
462: }
463:
464:
1.5 saito 465: /* create b11 submatrix */
1.4 saito 466: for (i = 0; i < b1row; i++) {
467: for (j = 0; j < b1col; j++) {
1.11 saito 468: b11->body[i][j] = b->body[i][j];
1.4 saito 469: }
470: }
471:
1.5 saito 472: /* create b21 submatrix */
1.11 saito 473: for (i = b1row; i < b->row; i++) {
1.4 saito 474: for (j = 0; j < b1col; j++) {
1.11 saito 475: b21->body[i-b1row][j] = b->body[i][j];
1.4 saito 476: }
477: }
478:
1.5 saito 479: /* create b12 submatrix */
1.4 saito 480: for (i = 0; i < b1row; i++) {
1.11 saito 481: for (j = b1col; j < b->col; j++) {
482: b12->body[i][j-b1col] = b->body[i][j];
1.4 saito 483: }
484: }
485:
1.5 saito 486: /* create b22 submatrix */
1.11 saito 487: for (i = b1row; i < b->row; i++) {
488: for (j = b1col; j < b->col; j++) {
489: b22->body[i-b1row][j-b1col] = b->body[i][j];
1.4 saito 490: }
491: }
1.5 saito 492: /* expand matrix by Strassen-Winograd algorithm */
1.4 saito 493: /* s1=A21+A22 */
494: addmat(vl,a21,a22,&s1);
495:
496: /* s2=s1-A11 */
497: submat(vl,s1,a11,&s2);
498:
499: /* t1=B12-B11 */
500: submat(vl, b12, b11, &t1);
501:
502: /* t2=B22-t1 */
503: submat(vl, b22, t1, &t2);
504:
505: /* u=(A11-A21)*(B22-B12) */
506: submat(vl, a11, a21, &ans1);
507: submat(vl, b22, b12, &ans2);
508: mulmatmat(vl, ans1, ans2, &u1);
509:
510: /* v=s1*t1 */
511: mulmatmat(vl, s1, t1, &v1);
512:
513: /* w=A11*B11+s2*t2 */
514: mulmatmat(vl, a11, b11, &ans1);
515: mulmatmat(vl, s2, t2, &ans2);
516: addmat(vl, ans1, ans2, &w1);
517:
518: /* C11 = A11*B11+A12*B21 */
519: mulmatmat(vl, a12, b21, &ans2);
520: addmat(vl, ans1, ans2, &c11);
521:
522: /* C12 = w1+v1+(A12-s2)*B22 */
523: submat(vl, a12, s2, &ans1);
524: mulmatmat(vl, ans1, b22, &ans2);
525: addmat(vl, w1, v1, &ans1);
526: addmat(vl, ans1, ans2, &c12);
527:
528: /* C21 = w1+u1+A22*(B21-t2) */
529: submat(vl, b21, t2, &ans1);
530: mulmatmat(vl, a22, ans1, &ans2);
1.6 saito 531: addmat(vl, w1, u1, &ans1);
532: addmat(vl, ans1, ans2, &c21);
533:
534: /* C22 = w1 + u1 + v1 */
535: addmat(vl, ans1, v1, &c22);
536: }
537:
538: for(i =0; i<c11->row; i++) {
539: for ( j=0; j < c11->col; j++) {
540: t->body[i][j] = c11->body[i][j];
541: }
542: }
543: if (pflag1 == 0) {
544: k = c21->row;
545: } else {
546: k = c21->row - 1;
547: }
548: for(i =0; i<k; i++) {
549: for ( j=0; j < c21->col; j++) {
550: t->body[i+c11->row][j] = c21->body[i][j];
551: }
552: }
553: if (pflag2 == 0) {
554: h = c12->col;
555: } else {
556: h = c12->col -1;
557: }
558: for(i =0; i<c12->row; i++) {
1.4 saito 559: for ( j=0; j < k; j++) {
560: t->body[i][j+c11->col] = c12->body[i][j];
561: }
562: }
563: if (pflag1 == 0) {
564: k = c22->row;
565: } else {
566: k = c22->row -1;
567: }
568: if (pflag2 == 0) {
569: h = c22->col;
570: } else {
571: h = c22->col - 1;
572: }
573: for(i =0; i<k; i++) {
574: for ( j=0; j < h; j++) {
575: t->body[i+c11->row][j+c11->col] = c22->body[i][j];
576: }
577: }
578: *c = t;
579: }
580:
1.14 saito 581: #if 0
582: /* remove miser type */
1.11 saito 583: void mulmatmat_miser(vl,a,b,c,ar0,ac0,ar1,ac1,br0,bc0,br1,bc1)
584: VL vl;
585: MAT a,b,*c;
586: int ar0, ac0, ar1, ac1, br0, bc0, br1, bc1;
587: {
588: int arow,bcol,i,j,k,m, h;
589: MAT t, a11, a12, a21, a22;
590: MAT p, b11, b12, b21, b22;
591: MAT ans1, ans2, c11, c12, c21, c22;
592: MAT s1, s2, t1, t2, u1, v1, w1;
593: pointer s,u,v;
594: pointer *ab,*tb, *bb;
595: int a1row, a1col;
596: int b1row, b1col;
597: int pflag1, pflag2;
1.4 saito 598:
1.11 saito 599: arow = ar1-ar0 + 1; m = ac1-ac0 + 1; bcol = bc1 - bc0 + 1;
600: /* mismach col and row */
601: if ( m != br1-br0 + 1 ) {
602: *c = 0; error("mulmat : size mismatch");
603: }
604: else {
605: pflag1 = 0; pflag2 = 0;
606: MKMAT(t,arow,bcol);
607: /* StrassenSize == 0 or matrix size less then StrassenSize,
608: then calc cannonical algorithm. */
609: if((StrassenSize == 0)||(arow<=StrassenSize || m <= StrassenSize) || (m<=StrassenSize || bcol <= StrassenSize)) {
610: for ( i = 0; i < arow; i++ ) {
611: if (i+ar0 > a->row-1) {
612: ab = NULL;
613: } else {
614: ab = BDY(a)[i+ar0];
615: }
616: tb = BDY(t)[i];
617: for ( j = 0; j < bcol; j++ ) {
618: for ( k = 0, s = 0; k < m; k++ ) {
619: if (k+br0 > b->row-1) {
620: bb = NULL;
621: } else {
622: bb = BDY(b)[k+br0];
623: }
624: if ((ab == NULL || k+ac0 > a->col-1) && (bb == NULL || j+bc0 > b->col-1)) {
625: arf_mul(vl,NULL,NULL,(Obj *)&u);
626: } else if ((ab != NULL && k+ac0 <= a->col-1) && (bb == NULL || j+bc0 > b->col-1)){
627: arf_mul(vl,(Obj)ab[k+ac0],NULL,(Obj *)&u);
628: } else if ((ab == NULL || k+ac0 > a->col-1) && (bb != NULL && j+bc0 <= b->col-1)) {
629: arf_mul(vl,NULL,(Obj)bb[j+bc0],(Obj *)&u);
630: } else {
631: arf_mul(vl,(Obj)ab[k+ac0],(Obj)bb[j+bc0],(Obj *)&u);
632: }
633: arf_add(vl,(Obj)s,(Obj)u,(Obj *)&v);
634: s = v;
635: }
636: tb[j] = s;
637: }
638: }
639: *c = t;
640: return;
641:
642: }
643: /* padding odd col and row to even number for zero */
644: i = arow/2;
645: j = arow - i;
646: if (i != j) {
647: arow++;
648: pflag1 = 1;
649: }
650: i = m/2;
651: j = m - i;
652: if (i != j) {
653: m++;
654: pflag2 = 1;
655: }
656:
657: i = bcol/2;
658: j = bcol - i;
659: if (i != j) {
660: bcol++;
661: }
662:
663: /* split matrix A and B */
664: a1row = arow/2; a1col = m/2;
665: b1row = m/2; b1col = bcol/2;
666:
667: /* expand matrix by Strassen-Winograd algorithm */
668: /* s1=A21+A22 */
669: addmat_miser(vl,a,a,&s1, ar0 + a1row, ac0, ar0 + arow -1, ac0 + a1col-1, ar0 + a1row, ac0 + a1col, ar0 + arow -1, ac0 + m-1);
670:
671: /* s2=s1-A11 */
672: submat_miser(vl,s1,a,&s2, 0,0, s1->row-1, s1->col-1, ar0, ac0, ar0 + a1row-1, ac0 + a1col-1);
673:
674: /* t1=B12-B11 */
675: submat_miser(vl, b, b, &t1, br0, bc0 + b1col, br0 + b1row-1, bc0 + bcol - 1, br0,bc0,br0 + b1row-1, bc0 + b1col-1);
676:
677: /* t2=B22-t1 */
678: submat_miser(vl, b, t1, &t2, br0 + b1row, bc0 + b1col, br0 + m-1, bc0 + bcol-1, 0,0,t1->row-1, t1->col-1);
679:
680: /* u=(A11-A21)*(B22-B12) */
681: submat_miser(vl, a, a, &ans1, ar0, ac0, ar0 + a1row-1,ac0 + a1col-1, ar0 + a1row, ac0, ar0 + arow-1, ac0 + a1col-1);
682: submat_miser(vl, b, b, &ans2, br0 + b1row, bc0 + b1col, br0 + m-1, bc0 + bcol-1, br0, bc0 + b1col, br0 + b1row-1, bc0 + bcol-1);
683: mulmatmat_miser(vl, ans1, ans2, &u1, 0, 0, ans1->row -1, ans1->col-1, 0, 0, ans2->row -1, ans2->col-1);
684:
685: /* v=s1*t1 */
686: mulmatmat_miser(vl, s1, t1, &v1, 0, 0, s1->row -1, s1->col-1, 0, 0, t1->row -1, t1->col-1);
687:
688: /* w=A11*B11+s2*t2 */
689: mulmatmat_miser(vl, a, b, &ans1, ar0, ac0, ar0 + a1row-1,ac0 + a1col-1, br0, bc0, br0 + b1row-1,bc0 + b1col-1);
690: mulmatmat_miser(vl, s2, t2, &ans2, 0, 0, s2->row -1, s2->col-1, 0, 0, t2->row -1, t2->col-1);
691: addmat_miser(vl, ans1, ans2, &w1, 0, 0, ans1->row -1, ans1->col-1, 0, 0, ans2->row -1, ans2->col-1);
692:
693: /* C11 = A11*B11+A12*B21 */
694: mulmatmat_miser(vl, a, b, &ans2, ar0, ac0 + a1col, ar0 + a1row-1, ac0 + m-1, br0 + b1row, bc0 + 0, br0 + m-1, bc0 + b1col-1);
695: addmat_miser(vl, ans1, ans2, &c11, 0, 0, ans1->row -1, ans1->col -1, 0, 0, ans2->row -1, ans2->col-1);
696:
697: /* C12 = w1+v1+(A12-s2)*B22 */
698: submat_miser(vl, a, s2, &ans1, ar0, ac0 + a1col, ar0 + a1row-1, ac0 + m-1, 0, 0, s2->row -1, s2->col -1);
699: mulmatmat_miser(vl, ans1, b, &ans2, 0, 0, ans1->row -1, ans1->col -1, br0 + b1row, bc0 + b1col, br0 + m-1, bc0 + bcol-1);
700: addmat_miser(vl, w1, v1, &ans1, 0, 0, w1->row -1, w1->col -1, 0,0, v1->row-1, v1->col -1);
701: addmat_miser(vl, ans1, ans2, &c12, 0, 0, ans1->row -1, ans1->col -1, 0, 0, ans2->row -1, ans2->col-1);
702:
703: /* C21 = w1+u1+A22*(B21-t2) */
704: submat_miser(vl, b, t2, &ans1, br0 + b1row, bc0 + 0, br0 + m-1, bc0 + b1col-1, 0,0, t2->row-1, t2->col-1);
705: mulmatmat_miser(vl, a, ans1, &ans2, ar0 + a1row, ac0 + a1col, ar0 + arow-1, ac0 + m-1, 0, 0, ans1->row -1, ans1->col -1);
706: addmat_miser(vl, w1, u1, &ans1, 0,0,w1->row -1, w1->col-1, 0,0,u1->row -1, u1->col-1);
707: addmat_miser(vl, ans1, ans2, &c21, 0, 0, ans1->row -1, ans1->col -1, 0, 0, ans2->row -1, ans2->col-1);
708:
709: /* C22 = w1 + u1 + v1 */
710: addmat_miser(vl, ans1, v1, &c22, 0, 0, ans1->row -1, ans1->col -1, 0, 0, v1->row-1, v1->col-1);
711: }
712:
713: for(i =0; i<c11->row; i++) {
714: for ( j=0; j < c11->col; j++) {
715: t->body[i][j] = c11->body[i][j];
716: }
717: }
718: if (pflag1 == 0) {
719: k = c21->row;
720: } else {
721: k = c21->row - 1;
722: }
723: for(i =0; i<k; i++) {
724: for ( j=0; j < c21->col; j++) {
725: t->body[i+c11->row][j] = c21->body[i][j];
726: }
727: }
728: if (pflag2 == 0) {
729: h = c12->col;
730: } else {
731: h = c12->col -1;
732: }
733: for(i =0; i<c12->row; i++) {
734: for ( j=0; j < k; j++) {
735: t->body[i][j+c11->col] = c12->body[i][j];
736: }
737: }
738: if (pflag1 == 0) {
739: k = c22->row;
740: } else {
741: k = c22->row -1;
742: }
743: if (pflag2 == 0) {
744: h = c22->col;
745: } else {
746: h = c22->col - 1;
747: }
748: for(i =0; i<k; i++) {
749: for ( j=0; j < h; j++) {
750: t->body[i+c11->row][j+c11->col] = c22->body[i][j];
751: }
752: }
753: *c = t;
754: }
1.14 saito 755: #endif
1.1 noro 756:
757: void mulmatvect(vl,a,b,c)
758: VL vl;
759: MAT a;
760: VECT b;
761: VECT *c;
762: {
763: int arow,i,j,m;
764: VECT t;
765: pointer s,u,v;
766: pointer *ab;
767:
768: if ( !a || !b )
769: *c = 0;
770: else if ( a->col != b->len ) {
771: *c = 0; error("mulmatvect : size mismatch");
772: } else {
1.15 ! noro 773: #if 0
1.7 noro 774: for ( i = 0; i < b->len; i++ )
775: if ( BDY(b)[i] && OID((Obj)BDY(b)[i]) > O_R )
776: error("mulmatvect : invalid argument");
1.15 ! noro 777: #endif
1.1 noro 778: arow = a->row; m = a->col;
779: MKVECT(t,arow);
780: for ( i = 0; i < arow; i++ ) {
781: for ( j = 0, s = 0, ab = BDY(a)[i]; j < m; j++ ) {
1.9 noro 782: arf_mul(vl,(Obj)ab[j],(Obj)BDY(b)[j],(Obj *)&u); arf_add(vl,(Obj)s,(Obj)u,(Obj *)&v); s = v;
1.1 noro 783: }
784: BDY(t)[i] = s;
785: }
786: *c = t;
787: }
788: }
789:
790: void mulvectmat(vl,a,b,c)
791: VL vl;
792: VECT a;
793: MAT b;
794: VECT *c;
795: {
796: int bcol,i,j,m;
797: VECT t;
798: pointer s,u,v;
799:
800: if ( !a || !b )
801: *c = 0;
802: else if ( a->len != b->row ) {
803: *c = 0; error("mulvectmat : size mismatch");
804: } else {
1.7 noro 805: for ( i = 0; i < a->len; i++ )
806: if ( BDY(a)[i] && OID((Obj)BDY(a)[i]) > O_R )
807: error("mulvectmat : invalid argument");
1.1 noro 808: bcol = b->col; m = a->len;
809: MKVECT(t,bcol);
810: for ( j = 0; j < bcol; j++ ) {
811: for ( i = 0, s = 0; i < m; i++ ) {
1.9 noro 812: arf_mul(vl,(Obj)BDY(a)[i],(Obj)BDY(b)[i][j],(Obj *)&u); arf_add(vl,(Obj)s,(Obj)u,(Obj *)&v); s = v;
1.1 noro 813: }
814: BDY(t)[j] = s;
815: }
816: *c = t;
817: }
818: }
819:
820: int compmat(vl,a,b)
821: VL vl;
822: MAT a,b;
823: {
824: int i,j,t,row,col;
825:
826: if ( !a )
827: return b?-1:0;
828: else if ( !b )
829: return 1;
830: else if ( a->row != b->row )
831: return a->row>b->row ? 1 : -1;
832: else if (a->col != b->col )
833: return a->col > b->col ? 1 : -1;
834: else {
835: row = a->row; col = a->col;
836: for ( i = 0; i < row; i++ )
837: for ( j = 0; j < col; j++ )
1.9 noro 838: if ( t = arf_comp(vl,(Obj)BDY(a)[i][j],(Obj)BDY(b)[i][j]) )
1.1 noro 839: return t;
840: return 0;
841: }
842: }
843:
844: pointer **almat_pointer(n,m)
845: int n,m;
846: {
847: pointer **mat;
848: int i;
849:
850: mat = (pointer **)MALLOC(n*sizeof(pointer *));
851: for ( i = 0; i < n; i++ )
852: mat[i] = (pointer *)CALLOC(m,sizeof(pointer));
853: return mat;
854: }