version 1.103, 2004/09/15 06:06:42 |
version 1.108, 2004/09/21 02:43:11 |
|
|
/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.102 2004/09/15 01:43:33 noro Exp $ */ |
/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.107 2004/09/21 02:34:12 noro Exp $ */ |
|
|
#include "nd.h" |
#include "nd.h" |
|
|
Line 3202 ND ndv_mul_nm(int mod,NM m0,NDV p) |
|
Line 3202 ND ndv_mul_nm(int mod,NM m0,NDV p) |
|
} |
} |
} |
} |
|
|
ND nd_quo(int mod,ND p,NDV d) |
ND nd_quo(int mod,PGeoBucket bucket,NDV d) |
{ |
{ |
NM mq0,mq; |
NM mq0,mq; |
NMV tm; |
NMV tm; |
Q q; |
Q q; |
int i,nv,sg,c,c1,c2; |
int i,nv,sg,c,c1,c2,hindex; |
ND t,r; |
ND p,t,r; |
|
N tnm; |
|
|
if ( !p ) return 0; |
if ( !p ) return 0; |
else { |
else { |
nv = NV(p); |
nv = NV(d); |
sg = SG(p); |
|
mq0 = 0; |
mq0 = 0; |
tm = (NMV)ALLOCA(nmv_adv); |
tm = (NMV)ALLOCA(nmv_adv); |
while ( p ) { |
while ( 1 ) { |
|
hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket); |
|
if ( hindex < 0 ) break; |
|
p = bucket->body[hindex]; |
NEXTNM(mq0,mq); |
NEXTNM(mq0,mq); |
ndl_sub(HDL(p),HDL(d),DL(mq)); |
ndl_sub(HDL(p),HDL(d),DL(mq)); |
ndl_copy(DL(mq),DL(tm)); |
ndl_copy(DL(mq),DL(tm)); |
Line 3225 ND nd_quo(int mod,ND p,NDV d) |
|
Line 3228 ND nd_quo(int mod,ND p,NDV d) |
|
DMAR(c1,c2,0,mod,c); CM(mq) = c; |
DMAR(c1,c2,0,mod,c); CM(mq) = c; |
CM(tm) = mod-c; |
CM(tm) = mod-c; |
} else { |
} else { |
divq(HCQ(p),HCQ(d),&CQ(mq)); |
divsn(NM(HCQ(p)),NM(HCQ(d)),&tnm); |
|
NTOQ(tnm,SGN(HCQ(p))*SGN(HCQ(d)),CQ(mq)); |
chsgnq(CQ(mq),&CQ(tm)); |
chsgnq(CQ(mq),&CQ(tm)); |
} |
} |
t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d)); |
t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d)); |
p = nd_add(mod,p,t); |
bucket->body[hindex] = nd_remove_head(p); |
|
t = nd_remove_head(t); |
|
add_pbucket(mod,bucket,t); |
} |
} |
NEXT(mq) = 0; |
if ( !mq0 ) |
for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ ); |
r = 0; |
MKND(nv,mq0,i,r); |
else { |
/* XXX */ |
NEXT(mq) = 0; |
SG(r) = sg-SG(d); |
for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ ); |
|
MKND(nv,mq0,i,r); |
|
/* XXX */ |
|
SG(r) = HTD(r); |
|
} |
return r; |
return r; |
} |
} |
} |
} |
Line 3759 int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r) |
|
Line 3769 int nd_to_vect(int mod,UINT *s0,int n,ND d,UINT *r) |
|
return i; |
return i; |
} |
} |
|
|
int ndv_to_vect(int mod,UINT *s0,int n,NDV d,UINT *r) |
int nd_to_vect_q(UINT *s0,int n,ND d,Q *r) |
{ |
{ |
NMV m; |
NM m; |
UINT *t,*s; |
UINT *t,*s; |
int i,j,len; |
int i; |
|
|
for ( i = 0; i < n; i++ ) r[i] = 0; |
for ( i = 0; i < n; i++ ) r[i] = 0; |
m = BDY(d); |
for ( i = 0, s = s0, m = BDY(d); m; m = NEXT(m) ) { |
len = LEN(d); |
|
for ( i = j = 0, s = s0; j < len; j++, NMV_ADV(m)) { |
|
t = DL(m); |
t = DL(m); |
for ( ; !ndl_equal(t,s); s += nd_wpd, i++ ); |
for ( ; !ndl_equal(t,s); s += nd_wpd, i++ ); |
r[i] = CM(m); |
r[i] = CQ(m); |
} |
} |
for ( i = 0; !r[i]; i++ ); |
for ( i = 0; !r[i]; i++ ); |
return i; |
return i; |
Line 3849 IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0 |
|
Line 3857 IndArray nm_ind_pair_to_vect_compress(int mod,UINT *s0 |
|
} |
} |
|
|
|
|
|
int ndv_reduce_vect_q(Q *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) |
|
{ |
|
int i,j,k,len,pos,prev; |
|
Q cs,mcs,c1,c2,cr,gcd; |
|
IndArray ivect; |
|
unsigned char *ivc; |
|
unsigned short *ivs; |
|
unsigned int *ivi; |
|
NDV redv; |
|
NMV mr; |
|
NODE rp; |
|
int maxrs; |
|
|
|
maxrs = 0; |
|
for ( i = 0; i < nred; i++ ) { |
|
ivect = imat[i]; |
|
k = ivect->head; |
|
if ( svect[k] ) { |
|
maxrs = MAX(maxrs,rp0[i]->sugar); |
|
redv = nd_ps[rp0[i]->index]; |
|
len = LEN(redv); mr = BDY(redv); |
|
igcd_cofactor(svect[k],CQ(mr),&gcd,&cs,&cr); |
|
chsgnq(cs,&mcs); |
|
svect[k] = 0; prev = k; |
|
switch ( ivect->width ) { |
|
case 1: |
|
ivc = ivect->index.c; |
|
for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { |
|
pos = prev+ivc[j]; prev = pos; |
|
mulq(svect[pos],cr,&c1); mulq(CQ(mr),mcs,&c2); addq(c1,c2,&svect[pos]); |
|
} |
|
break; |
|
case 2: |
|
ivs = ivect->index.s; |
|
for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { |
|
pos = prev+ivs[j]; prev = pos; |
|
mulq(svect[pos],cr,&c1); mulq(CQ(mr),mcs,&c2); addq(c1,c2,&svect[pos]); |
|
} |
|
break; |
|
case 4: |
|
ivi = ivect->index.i; |
|
for ( j = 1, NMV_ADV(mr); j < len; j++, NMV_ADV(mr) ) { |
|
pos = prev+ivi[j]; prev = pos; |
|
mulq(svect[pos],cr,&c1); mulq(CQ(mr),mcs,&c2); addq(c1,c2,&svect[pos]); |
|
} |
|
break; |
|
} |
|
} |
|
} |
|
return maxrs; |
|
} |
|
|
int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) |
int ndv_reduce_vect(int m,UINT *svect,int col,IndArray *imat,NM_ind_pair *rp0,int nred) |
{ |
{ |
int i,j,k,len,pos,prev; |
int i,j,k,len,pos,prev; |
Line 3988 NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhea |
|
Line 4048 NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhea |
|
} |
} |
} |
} |
|
|
|
NDV vect_to_ndv_q(Q *vect,int spcol,int col,int *rhead,UINT *s0vect) |
|
{ |
|
int j,k,len; |
|
UINT *p; |
|
Q c; |
|
NDV r; |
|
NMV mr0,mr; |
|
|
|
for ( j = 0, len = 0; j < spcol; j++ ) if ( vect[j] ) len++; |
|
if ( !len ) return 0; |
|
else { |
|
mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len); |
|
#if 0 |
|
ndv_alloc += nmv_adv*len; |
|
#endif |
|
mr = mr0; |
|
p = s0vect; |
|
for ( j = k = 0; j < col; j++, p += nd_wpd ) |
|
if ( !rhead[j] ) { |
|
if ( c = vect[k++] ) { |
|
ndl_copy(p,DL(mr)); CQ(mr) = c; NMV_ADV(mr); |
|
} |
|
} |
|
MKNDV(nd_nvar,mr0,len,r); |
|
return r; |
|
} |
|
} |
|
|
int nd_sp_f4(int m,ND_pairs l,PGeoBucket bucket) |
int nd_sp_f4(int m,ND_pairs l,PGeoBucket bucket) |
{ |
{ |
ND_pairs t; |
ND_pairs t; |
Line 4124 NODE nd_f4(int m) |
|
Line 4212 NODE nd_f4(int m) |
|
NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0) |
NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col,NODE rp0) |
{ |
{ |
IndArray *imat; |
IndArray *imat; |
int nsp,nred,spcol,sprow,a; |
int nsp,nred,i; |
int *rhead; |
int *rhead; |
int i,j,k,l,rank; |
NODE r0,rp; |
NODE rp,r0,r; |
|
ND_pairs sp; |
ND_pairs sp; |
ND spol; |
|
int **spmat; |
|
UINT *svect,*v; |
|
int *colstat; |
|
struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2; |
|
NM_ind_pair *rvect; |
NM_ind_pair *rvect; |
int maxrs; |
|
int *spsugar; |
|
|
|
get_eg(&eg0); |
|
for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ ); |
for ( sp = sp0, nsp = 0; sp; sp = NEXT(sp), nsp++ ); |
nred = length(rp0); spcol = col-nred; |
nred = length(rp0); |
imat = (IndArray *)ALLOCA(nred*sizeof(IndArray)); |
imat = (IndArray *)ALLOCA(nred*sizeof(IndArray)); |
rhead = (int *)ALLOCA(col*sizeof(int)); |
rhead = (int *)ALLOCA(col*sizeof(int)); |
for ( i = 0; i < col; i++ ) rhead[i] = 0; |
for ( i = 0; i < col; i++ ) rhead[i] = 0; |
Line 4152 NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col |
|
Line 4231 NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col |
|
imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rvect[i]); |
imat[i] = nm_ind_pair_to_vect_compress(m,s0vect,col,rvect[i]); |
rhead[imat[i]->head] = 1; |
rhead[imat[i]->head] = 1; |
} |
} |
|
if ( m ) |
|
r0 = nd_f4_red_main(m,sp0,nsp,s0vect,col,rvect,rhead,imat,nred); |
|
else |
|
r0 = nd_f4_red_q_main(sp0,nsp,s0vect,col,rvect,rhead,imat,nred); |
|
return r0; |
|
} |
|
|
|
NODE nd_f4_red_main(int m,ND_pairs sp0,int nsp,UINT *s0vect,int col, |
|
NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred) |
|
{ |
|
int spcol,sprow,a; |
|
int i,j,k,l,rank; |
|
NODE r0,r; |
|
ND_pairs sp; |
|
ND spol; |
|
int **spmat; |
|
UINT *svect,*v; |
|
int *colstat; |
|
struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2; |
|
int maxrs; |
|
int *spsugar; |
|
|
|
spcol = col-nred; |
|
get_eg(&eg0); |
/* elimination (1st step) */ |
/* elimination (1st step) */ |
spmat = (int **)ALLOCA(nsp*sizeof(UINT *)); |
spmat = (int **)ALLOCA(nsp*sizeof(UINT *)); |
svect = (UINT *)ALLOCA(col*sizeof(UINT)); |
svect = (UINT *)ALLOCA(col*sizeof(UINT)); |
Line 4208 NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col |
|
Line 4310 NODE nd_f4_red(int m,ND_pairs sp0,UINT *s0vect,int col |
|
return r0; |
return r0; |
} |
} |
|
|
|
NODE nd_f4_red_q_main(ND_pairs sp0,int nsp,UINT *s0vect,int col, |
|
NM_ind_pair *rvect,int *rhead,IndArray *imat,int nred) |
|
{ |
|
int spcol,sprow,a; |
|
int i,j,k,l,rank; |
|
NODE r0,r; |
|
ND_pairs sp; |
|
ND spol; |
|
Q **spmat; |
|
Q *svect,*v; |
|
int *colstat; |
|
struct oEGT eg0,eg1,eg2,eg_f4,eg_f4_1,eg_f4_2; |
|
int maxrs; |
|
int *spsugar; |
|
|
|
spcol = col-nred; |
|
get_eg(&eg0); |
|
/* elimination (1st step) */ |
|
spmat = (Q **)ALLOCA(nsp*sizeof(UINT *)); |
|
svect = (Q *)ALLOCA(col*sizeof(UINT)); |
|
spsugar = (int *)ALLOCA(nsp*sizeof(UINT)); |
|
for ( a = sprow = 0, sp = sp0; a < nsp; a++, sp = NEXT(sp) ) { |
|
nd_sp(0,0,sp,&spol); |
|
if ( !spol ) continue; |
|
nd_to_vect_q(s0vect,col,spol,svect); |
|
maxrs = ndv_reduce_vect_q(svect,col,imat,rvect,nred); |
|
for ( i = 0; i < col; i++ ) if ( svect[i] ) break; |
|
if ( i < col ) { |
|
spmat[sprow] = v = (Q *)MALLOC_ATOMIC(spcol*sizeof(Q)); |
|
for ( j = k = 0; j < col; j++ ) |
|
if ( !rhead[j] ) v[k++] = svect[j]; |
|
spsugar[sprow] = MAX(maxrs,SG(spol)); |
|
sprow++; |
|
} |
|
nd_free(spol); |
|
} |
|
get_eg(&eg1); init_eg(&eg_f4_1); add_eg(&eg_f4_1,&eg0,&eg1); |
|
if ( DP_Print ) { |
|
fprintf(asir_out,"elim1=%fsec,",eg_f4_1.exectime+eg_f4_1.gctime); |
|
fflush(asir_out); |
|
} |
|
/* free index arrays */ |
|
for ( i = 0; i < nred; i++ ) GC_free(imat[i]->index.c); |
|
|
|
/* elimination (2nd step) */ |
|
colstat = (int *)ALLOCA(spcol*sizeof(int)); |
|
rank = nd_gauss_elim_q(spmat,spsugar,sprow,spcol,colstat); |
|
r0 = 0; |
|
for ( i = 0; i < rank; i++ ) { |
|
NEXTNODE(r0,r); BDY(r) = |
|
(pointer)vect_to_ndv_q(spmat[i],spcol,col,rhead,s0vect); |
|
SG((NDV)BDY(r)) = spsugar[i]; |
|
GC_free(spmat[i]); |
|
} |
|
for ( ; i < sprow; i++ ) GC_free(spmat[i]); |
|
get_eg(&eg2); init_eg(&eg_f4_2); add_eg(&eg_f4_2,&eg1,&eg2); |
|
init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg2); |
|
if ( DP_Print ) { |
|
fprintf(asir_out,"elim2=%fsec\n",eg_f4_2.exectime+eg_f4_2.gctime); |
|
fprintf(asir_out,"nsp=%d,nred=%d,spmat=(%d,%d),rank=%d ", |
|
nsp,nred,sprow,spcol,rank); |
|
fprintf(asir_out,"%fsec\n",eg_f4.exectime+eg_f4.gctime); |
|
} |
|
return r0; |
|
} |
|
|
FILE *nd_write,*nd_read; |
FILE *nd_write,*nd_read; |
|
|
void nd_send_int(int a) { |
void nd_send_int(int a) { |
Line 4464 void nd_exec_f4_red_dist() |
|
Line 4632 void nd_exec_f4_red_dist() |
|
fflush(nd_write); |
fflush(nd_write); |
} |
} |
|
|
|
int nd_gauss_elim_q(Q **mat0,int *sugar,int row,int col,int *colstat) |
|
{ |
|
int mod,i,j,t; |
|
int **wmat; |
|
|
|
/* XXX */ |
|
mod = 99999989; |
|
wmat = (int **)ALLOCA(row*sizeof(int *)); |
|
for ( i = 0; i < row; i++ ) { |
|
wmat[i] = (int *)ALLOCA(col*sizeof(int)); |
|
for ( j = 0; j < col; j++ ) { |
|
if ( mat0[i][j] ) { |
|
t = rem(NM(mat0[i][j]),mod); |
|
if ( SGN(mat0[i][j]) < 0 ) t = mod-t; |
|
wmat[i][j] = t; |
|
} else |
|
wmat[i][j] = 0; |
|
} |
|
} |
|
nd_gauss_elim_mod(wmat,sugar,row,col,mod,colstat); |
|
} |
|
|
int nd_gauss_elim_mod(int **mat0,int *sugar,int row,int col,int md,int *colstat) |
int nd_gauss_elim_mod(int **mat0,int *sugar,int row,int col,int md,int *colstat) |
{ |
{ |
int i,j,k,l,inv,a,rank,s; |
int i,j,k,l,inv,a,rank,s; |
Line 4711 void nd_det(int mod,MAT f,P *rp) |
|
Line 4901 void nd_det(int mod,MAT f,P *rp) |
|
if ( mod ) ndv_mod(mod,d); |
if ( mod ) ndv_mod(mod,d); |
chsgnq(ONE,&mone); |
chsgnq(ONE,&mone); |
for ( j = 0, sgn = 1; j < n; j++ ) { |
for ( j = 0, sgn = 1; j < n; j++ ) { |
|
if ( DP_Print ) fprintf(stderr,"j=%d\n",j); |
for ( i = j; i < n && !dm[i][j]; i++ ); |
for ( i = j; i < n && !dm[i][j]; i++ ); |
if ( i == n ) { |
if ( i == n ) { |
*rp = 0; |
*rp = 0; |
Line 4733 void nd_det(int mod,MAT f,P *rp) |
|
Line 4924 void nd_det(int mod,MAT f,P *rp) |
|
sgn = -sgn; |
sgn = -sgn; |
} |
} |
for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) { |
for ( i = j+1, mj = dm[j], mjj = mj[j]; i < n; i++ ) { |
|
if ( DP_Print ) fprintf(stderr," i=%d\n ",i); |
mi = dm[i]; mij = mi[j]; |
mi = dm[i]; mij = mi[j]; |
if ( mod ) |
if ( mod ) |
ndv_mul_c(mod,mij,mod-1); |
ndv_mul_c(mod,mij,mod-1); |
else |
else |
ndv_mul_c_q(mij,mone); |
ndv_mul_c_q(mij,mone); |
for ( k = j+1; k < n; k++ ) { |
for ( k = j+1; k < n; k++ ) { |
|
if ( DP_Print ) fprintf(stderr,"k=%d ",k); |
bucket = create_pbucket(); |
bucket = create_pbucket(); |
if ( mi[k] ) |
if ( mi[k] ) { |
nmv = BDY(mjj); len = LEN(mjj); |
nmv = BDY(mjj); len = LEN(mjj); |
for ( a = 0; a < len; a++, NMV_ADV(nmv) ) { |
for ( a = 0; a < len; a++, NMV_ADV(nmv) ) { |
u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d))); |
u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d))); |
add_pbucket(mod,bucket,u); |
add_pbucket(mod,bucket,u); |
} |
} |
|
} |
if ( mj[k] && mij ) { |
if ( mj[k] && mij ) { |
nmv = BDY(mij); len = LEN(mij); |
nmv = BDY(mij); len = LEN(mij); |
for ( a = 0; a < len; a++, NMV_ADV(nmv) ) { |
for ( a = 0; a < len; a++, NMV_ADV(nmv) ) { |
Line 4753 void nd_det(int mod,MAT f,P *rp) |
|
Line 4947 void nd_det(int mod,MAT f,P *rp) |
|
add_pbucket(mod,bucket,u); |
add_pbucket(mod,bucket,u); |
} |
} |
} |
} |
u = normalize_pbucket(mod,bucket); |
u = nd_quo(mod,bucket,d); |
u = nd_quo(mod,u,d); |
|
mi[k] = ndtondv(mod,u); |
mi[k] = ndtondv(mod,u); |
} |
} |
|
if ( DP_Print ) fprintf(stderr,"\n",k); |
} |
} |
d = mjj; |
d = mjj; |
} |
} |
Line 4792 ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d) |
|
Line 4986 ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d) |
|
if ( ndl_reducible(DL(tnm),d) ) { |
if ( ndl_reducible(DL(tnm),d) ) { |
NEXTNM(mr0,mr); |
NEXTNM(mr0,mr); |
c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2; |
c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2; |
ndl_add(DL(m),d0,DL(mr)); |
ndl_copy(DL(tnm),DL(mr)); |
} |
} |
} |
} |
} else { |
} else { |
Line 4802 ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d) |
|
Line 4996 ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d) |
|
if ( ndl_reducible(DL(tnm),d) ) { |
if ( ndl_reducible(DL(tnm),d) ) { |
NEXTNM(mr0,mr); |
NEXTNM(mr0,mr); |
mulq(CQ(m),q,&CQ(mr)); |
mulq(CQ(m),q,&CQ(mr)); |
ndl_add(DL(m),d0,DL(mr)); |
ndl_copy(DL(tnm),DL(mr)); |
} |
} |
} |
} |
} |
} |
Line 4810 ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d) |
|
Line 5004 ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d) |
|
return 0; |
return 0; |
else { |
else { |
NEXT(mr) = 0; |
NEXT(mr) = 0; |
|
for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ ); |
MKND(NV(p),mr0,len,r); |
MKND(NV(p),mr0,len,r); |
SG(r) = SG(p) + TD(d0); |
SG(r) = SG(p) + TD(d0); |
return r; |
return r; |