=================================================================== RCS file: /home/cvs/OpenXM_contrib2/asir2000/engine/nd.c,v retrieving revision 1.94 retrieving revision 1.95 diff -u -p -r1.94 -r1.95 --- OpenXM_contrib2/asir2000/engine/nd.c 2004/03/15 07:30:44 1.94 +++ OpenXM_contrib2/asir2000/engine/nd.c 2004/03/15 08:44:52 1.95 @@ -1,4 +1,4 @@ -/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.93 2004/03/13 07:48:30 noro Exp $ */ +/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.94 2004/03/15 07:30:44 noro Exp $ */ #include "nd.h" @@ -849,6 +849,79 @@ ND nd_add(int mod,ND p1,ND p2) can += 2; FREENM(s); } s = m2; m2 = NEXT(m2); FREENM(s); + break; + case 1: + s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s); + break; + case -1: + s = m2; m2 = NEXT(m2); NEXTNM2(mr0,mr,s); + break; + } + } + if ( !mr0 ) + if ( m1 ) mr0 = m1; + else if ( m2 ) mr0 = m2; + else return 0; + else if ( m1 ) NEXT(mr) = m1; + else if ( m2 ) NEXT(mr) = m2; + else NEXT(mr) = 0; + BDY(p1) = mr0; + SG(p1) = MAX(SG(p1),SG(p2)); + LEN(p1) = LEN(p1)+LEN(p2)-can; + FREEND(p2); + return p1; + } +} + +/* XXX on opteron, the inlined manipulation of destructive additon of + * two NM seems to make gcc optimizer get confused, so the part is + * done in a function. + */ + +int nm_destructive_add_q(NM *m1,NM *m2,NM *mr0,NM *mr) +{ + NM s; + Q t; + int can; + + addq(CQ(*m1),CQ(*m2),&t); + s = *m1; *m1 = NEXT(*m1); + if ( t ) { + can = 1; NEXTNM2(*mr0,*mr,s); CQ(*mr) = (t); + } else { + can = 2; FREENM(s); + } + s = *m2; *m2 = NEXT(*m2); FREENM(s); + return can; +} + +ND nd_add_q(ND p1,ND p2) +{ + int n,c,can; + ND r; + NM m1,m2,mr0,mr,s; + Q t; + + if ( !p1 ) return p2; + else if ( !p2 ) return p1; + else { + can = 0; + for ( n = NV(p1), m1 = BDY(p1), m2 = BDY(p2), mr0 = 0; m1 && m2; ) { + c = DL_COMPARE(DL(m1),DL(m2)); + switch ( c ) { + case 0: +#if defined(__x86_64__) + can += nm_destructive_add_q(&m1,&m2,&mr0,&mr); +#else + addq(CQ(m1),CQ(m2),&t); + s = m1; m1 = NEXT(m1); + if ( t ) { + can++; NEXTNM2(mr0,mr,s); CQ(mr) = (t); + } else { + can += 2; FREENM(s); + } + s = m2; m2 = NEXT(m2); FREENM(s); +#endif break; case 1: s = m1; m1 = NEXT(m1); NEXTNM2(mr0,mr,s);