version 1.97, 2004/03/25 01:31:03 |
version 1.107, 2004/09/21 02:34:12 |
|
|
/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.96 2004/03/17 08:16:24 noro Exp $ */ |
/* $OpenXM: OpenXM_contrib2/asir2000/engine/nd.c,v 1.106 2004/09/21 02:23:49 noro Exp $ */ |
|
|
#include "nd.h" |
#include "nd.h" |
|
|
|
|
ND _nd_free_list; |
ND _nd_free_list; |
ND_pairs _ndp_free_list; |
ND_pairs _ndp_free_list; |
|
|
|
#if 0 |
static int ndv_alloc; |
static int ndv_alloc; |
|
#endif |
#if 1 |
#if 1 |
static int nd_f4_nsp=0x7fffffff; |
static int nd_f4_nsp=0x7fffffff; |
#else |
#else |
|
|
d = ndvtond(0,r); |
d = ndvtond(0,r); |
stat = nd_nf(0,d,nd_ps,0,0,&nf); |
stat = nd_nf(0,d,nd_ps,0,0,&nf); |
if ( !stat ) { |
if ( !stat ) { |
nd_reconstruct(0,0,0); |
nd_reconstruct(0,0); |
goto again; |
goto again; |
} else if ( nf ) return 0; |
} else if ( nf ) return 0; |
if ( DP_Print ) { printf("."); fflush(stdout); } |
if ( DP_Print ) { printf("."); fflush(stdout); } |
|
|
stat = nd_sp(m,0,l,&h); |
stat = nd_sp(m,0,l,&h); |
if ( !stat ) { |
if ( !stat ) { |
NEXT(l) = d; d = l; |
NEXT(l) = d; d = l; |
d = nd_reconstruct(m,0,d); |
d = nd_reconstruct(0,d); |
goto again; |
goto again; |
} |
} |
#if USE_GEOBUCKET |
#if USE_GEOBUCKET |
|
|
#endif |
#endif |
if ( !stat ) { |
if ( !stat ) { |
NEXT(l) = d; d = l; |
NEXT(l) = d; d = l; |
d = nd_reconstruct(m,0,d); |
d = nd_reconstruct(0,d); |
goto again; |
goto again; |
} else if ( nf ) { |
} else if ( nf ) { |
if ( checkonly ) return 0; |
if ( checkonly ) return 0; |
|
|
stat = nd_sp(m,0,l,&h); |
stat = nd_sp(m,0,l,&h); |
if ( !stat ) { |
if ( !stat ) { |
NEXT(l) = d; d = l; |
NEXT(l) = d; d = l; |
d = nd_reconstruct(m,1,d); |
d = nd_reconstruct(1,d); |
goto again; |
goto again; |
} |
} |
#if USE_GEOBUCKET |
#if USE_GEOBUCKET |
|
|
#endif |
#endif |
if ( !stat ) { |
if ( !stat ) { |
NEXT(l) = d; d = l; |
NEXT(l) = d; d = l; |
d = nd_reconstruct(m,1,d); |
d = nd_reconstruct(1,d); |
goto again; |
goto again; |
} else if ( nf ) { |
} else if ( nf ) { |
if ( nd_demand ) { |
if ( nd_demand ) { |
|
|
if ( !nfq ) { |
if ( !nfq ) { |
if ( !nd_sp(0,1,l,&h) || !nd_nf(0,h,nd_ps_trace,!Top,0,&nfq) ) { |
if ( !nd_sp(0,1,l,&h) || !nd_nf(0,h,nd_ps_trace,!Top,0,&nfq) ) { |
NEXT(l) = d; d = l; |
NEXT(l) = d; d = l; |
d = nd_reconstruct(m,1,d); |
d = nd_reconstruct(1,d); |
goto again; |
goto again; |
} |
} |
} |
} |
Line 1790 NODE ndv_reduceall(int m,NODE f) |
|
Line 1792 NODE ndv_reduceall(int m,NODE f) |
|
g = nd_separate_head(g,&head); |
g = nd_separate_head(g,&head); |
stat = nd_nf(m,g,nd_ps,1,&dn,&nf); |
stat = nd_nf(m,g,nd_ps,1,&dn,&nf); |
if ( !stat ) |
if ( !stat ) |
nd_reconstruct(m,0,0); |
nd_reconstruct(0,0); |
else { |
else { |
if ( DP_Print ) { printf("."); fflush(stdout); } |
if ( DP_Print ) { printf("."); fflush(stdout); } |
if ( !m ) { |
if ( !m ) { |
Line 2186 void nd_gr(LIST f,LIST v,int m,int f4,struct order_spe |
|
Line 2188 void nd_gr(LIST f,LIST v,int m,int f4,struct order_spe |
|
if ( !m && Demand ) nd_demand = 1; |
if ( !m && Demand ) nd_demand = 1; |
else nd_demand = 0; |
else nd_demand = 0; |
|
|
|
#if 0 |
ndv_alloc = 0; |
ndv_alloc = 0; |
|
#endif |
get_vars((Obj)f,&fv); pltovl(v,&vv); |
get_vars((Obj)f,&fv); pltovl(v,&vv); |
for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ ); |
for ( nvar = 0, tv = vv; tv; tv = NEXT(tv), nvar++ ); |
switch ( ord->id ) { |
switch ( ord->id ) { |
Line 2224 void nd_gr(LIST f,LIST v,int m,int f4,struct order_spe |
|
Line 2228 void nd_gr(LIST f,LIST v,int m,int f4,struct order_spe |
|
} |
} |
if ( r0 ) NEXT(r) = 0; |
if ( r0 ) NEXT(r) = 0; |
MKLIST(*rp,r0); |
MKLIST(*rp,r0); |
|
#if 0 |
fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc); |
fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc); |
|
#endif |
} |
} |
|
|
void nd_gr_trace(LIST f,LIST v,int trace,int homo,struct order_spec *ord,LIST *rp) |
void nd_gr_trace(LIST f,LIST v,int trace,int homo,struct order_spec *ord,LIST *rp) |
Line 2702 UINT *ndv_compute_bound(NDV p) |
|
Line 2708 UINT *ndv_compute_bound(NDV p) |
|
return t; |
return t; |
} |
} |
|
|
|
UINT *nd_compute_bound(ND p) |
|
{ |
|
UINT *d1,*d2,*t; |
|
UINT u; |
|
int i,j,k,l,len,ind; |
|
NM m; |
|
|
|
if ( !p ) |
|
return 0; |
|
d1 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); |
|
d2 = (UINT *)ALLOCA(nd_wpd*sizeof(UINT)); |
|
len = LEN(p); |
|
m = BDY(p); ndl_copy(DL(m),d1); m = NEXT(m); |
|
for ( m = NEXT(m); m; m = NEXT(m) ) { |
|
ndl_lcm(DL(m),d1,d2); |
|
t = d1; d1 = d2; d2 = t; |
|
} |
|
l = nd_nvar+31; |
|
t = (UINT *)MALLOC_ATOMIC(l*sizeof(UINT)); |
|
for ( i = nd_exporigin, ind = 0; i < nd_wpd; i++ ) { |
|
u = d1[i]; |
|
k = (nd_epw-1)*nd_bpe; |
|
for ( j = 0; j < nd_epw; j++, k -= nd_bpe, ind++ ) |
|
t[ind] = (u>>k)&nd_mask0; |
|
} |
|
for ( ; ind < l; ind++ ) t[ind] = 0; |
|
return t; |
|
} |
|
|
int nd_get_exporigin(struct order_spec *ord) |
int nd_get_exporigin(struct order_spec *ord) |
{ |
{ |
switch ( ord->id ) { |
switch ( ord->id ) { |
Line 2760 void nd_setup_parameters(int nvar,int max) { |
|
Line 2795 void nd_setup_parameters(int nvar,int max) { |
|
nd_work_vector = (int *)REALLOC(nd_work_vector,nd_nvar*sizeof(int)); |
nd_work_vector = (int *)REALLOC(nd_work_vector,nd_nvar*sizeof(int)); |
} |
} |
|
|
ND_pairs nd_reconstruct(int mod,int trace,ND_pairs d) |
ND_pairs nd_reconstruct(int trace,ND_pairs d) |
{ |
{ |
int i,obpe,oadv,h; |
int i,obpe,oadv,h; |
static NM prev_nm_free_list; |
static NM prev_nm_free_list; |
Line 3167 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,PGeoBucket bucket,NDV d) |
|
{ |
|
NM mq0,mq; |
|
NMV tm; |
|
Q q; |
|
int i,nv,sg,c,c1,c2,hindex; |
|
ND p,t,r; |
|
N tnm; |
|
|
|
if ( !p ) return 0; |
|
else { |
|
nv = NV(d); |
|
mq0 = 0; |
|
tm = (NMV)ALLOCA(nmv_adv); |
|
while ( 1 ) { |
|
hindex = mod?head_pbucket(mod,bucket):head_pbucket_q(bucket); |
|
if ( hindex < 0 ) break; |
|
p = bucket->body[hindex]; |
|
NEXTNM(mq0,mq); |
|
ndl_sub(HDL(p),HDL(d),DL(mq)); |
|
ndl_copy(DL(mq),DL(tm)); |
|
if ( mod ) { |
|
c1 = invm(HCM(d),mod); c2 = HCM(p); |
|
DMAR(c1,c2,0,mod,c); CM(mq) = c; |
|
CM(tm) = mod-c; |
|
} else { |
|
divsn(NM(HCQ(p)),NM(HCQ(d)),&tnm); |
|
NTOQ(tnm,SGN(HCQ(p))*SGN(HCQ(d)),CQ(mq)); |
|
chsgnq(CQ(mq),&CQ(tm)); |
|
} |
|
t = ndv_mul_nmv_trunc(mod,tm,d,HDL(d)); |
|
bucket->body[hindex] = nd_remove_head(p); |
|
t = nd_remove_head(t); |
|
add_pbucket(mod,bucket,t); |
|
} |
|
if ( !mq0 ) |
|
r = 0; |
|
else { |
|
NEXT(mq) = 0; |
|
for ( i = 0, mq = mq0; mq; mq = NEXT(mq), i++ ); |
|
MKND(nv,mq0,i,r); |
|
/* XXX */ |
|
SG(r) = HTD(r); |
|
} |
|
return r; |
|
} |
|
} |
|
|
void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos) |
void ndv_realloc(NDV p,int obpe,int oadv,EPOS oepos) |
{ |
{ |
NMV m,mr,mr0,t; |
NMV m,mr,mr0,t; |
Line 3316 ND ptond(VL vl,VL dvl,P p) |
|
Line 3399 ND ptond(VL vl,VL dvl,P p) |
|
w = (DCP *)ALLOCA(k*sizeof(DCP)); |
w = (DCP *)ALLOCA(k*sizeof(DCP)); |
for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc; |
for ( dc = DC(p), j = 0; j < k; dc = NEXT(dc), j++ ) w[j] = dc; |
for ( i = 0, tvl = dvl, v = VR(p); |
for ( i = 0, tvl = dvl, v = VR(p); |
vl && tvl->v != v; tvl = NEXT(tvl), i++ ); |
tvl && tvl->v != v; tvl = NEXT(tvl), i++ ); |
if ( !tvl ) { |
if ( !tvl ) { |
for ( j = k-1, s = 0, MKV(v,x); j >= 0; j-- ) { |
for ( j = k-1, s = 0, MKV(v,x); j >= 0; j-- ) { |
t = ptond(vl,dvl,COEF(w[j])); |
t = ptond(vl,dvl,COEF(w[j])); |
Line 3389 NDV ndtondv(int mod,ND p) |
|
Line 3472 NDV ndtondv(int mod,ND p) |
|
m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(len*nmv_adv); |
m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(len*nmv_adv); |
else |
else |
m0 = m = MALLOC(len*nmv_adv); |
m0 = m = MALLOC(len*nmv_adv); |
|
#if 0 |
ndv_alloc += nmv_adv*len; |
ndv_alloc += nmv_adv*len; |
|
#endif |
for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) { |
for ( t = BDY(p), i = 0; t; t = NEXT(t), i++, NMV_ADV(m) ) { |
ndl_copy(DL(t),DL(m)); |
ndl_copy(DL(t),DL(m)); |
CQ(m) = CQ(t); |
CQ(m) = CQ(t); |
Line 3660 void nd_nf_p(P f,LIST g,LIST v,int m,struct order_spec |
|
Line 3745 void nd_nf_p(P f,LIST g,LIST v,int m,struct order_spec |
|
stat = nd_nf(m,nd,nd_ps,1,0,&nf); |
stat = nd_nf(m,nd,nd_ps,1,0,&nf); |
if ( !stat ) { |
if ( !stat ) { |
nd_psn++; |
nd_psn++; |
nd_reconstruct(m,0,0); |
nd_reconstruct(0,0); |
nd_psn--; |
nd_psn--; |
} else |
} else |
break; |
break; |
Line 3684 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 3774 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 3897 NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhea |
|
Line 4032 NDV vect_to_ndv(UINT *vect,int spcol,int col,int *rhea |
|
if ( !len ) return 0; |
if ( !len ) return 0; |
else { |
else { |
mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len); |
mr0 = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len); |
|
#if 0 |
ndv_alloc += nmv_adv*len; |
ndv_alloc += nmv_adv*len; |
|
#endif |
mr = mr0; |
mr = mr0; |
p = s0vect; |
p = s0vect; |
for ( j = k = 0; j < col; j++, p += nd_wpd ) |
for ( j = k = 0; j < col; j++, p += nd_wpd ) |
Line 3911 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 3992 NODE nd_f4(int m) |
|
Line 4157 NODE nd_f4(int m) |
|
|
|
if ( !m ) |
if ( !m ) |
error("nd_f4 : not implemented"); |
error("nd_f4 : not implemented"); |
|
#if 0 |
ndv_alloc = 0; |
ndv_alloc = 0; |
|
#endif |
g = 0; d = 0; |
g = 0; d = 0; |
for ( i = 0; i < nd_psn; i++ ) { |
for ( i = 0; i < nd_psn; i++ ) { |
d = update_pairs(d,g,i); |
d = update_pairs(d,g,i); |
Line 4007 NODE nd_f4(int m) |
|
Line 4174 NODE nd_f4(int m) |
|
if ( !stat ) { |
if ( !stat ) { |
for ( t = l; NEXT(t); t = NEXT(t) ); |
for ( t = l; NEXT(t); t = NEXT(t) ); |
NEXT(t) = d; d = l; |
NEXT(t) = d; d = l; |
d = nd_reconstruct(m,0,d); |
d = nd_reconstruct(0,d); |
continue; |
continue; |
} |
} |
if ( bucket->m < 0 ) continue; |
if ( bucket->m < 0 ) continue; |
Line 4015 NODE nd_f4(int m) |
|
Line 4182 NODE nd_f4(int m) |
|
if ( !col ) { |
if ( !col ) { |
for ( t = l; NEXT(t); t = NEXT(t) ); |
for ( t = l; NEXT(t); t = NEXT(t) ); |
NEXT(t) = d; d = l; |
NEXT(t) = d; d = l; |
d = nd_reconstruct(m,0,d); |
d = nd_reconstruct(0,d); |
continue; |
continue; |
} |
} |
get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1); |
get_eg(&eg1); init_eg(&eg_f4); add_eg(&eg_f4,&eg0,&eg1); |
Line 4036 NODE nd_f4(int m) |
|
Line 4203 NODE nd_f4(int m) |
|
} |
} |
} |
} |
for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps[(int)BDY(r)]; |
for ( r = g; r; r = NEXT(r) ) BDY(r) = (pointer)nd_ps[(int)BDY(r)]; |
|
#if 0 |
fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc); |
fprintf(asir_out,"ndv_alloc=%d\n",ndv_alloc); |
|
#endif |
return g; |
return g; |
} |
} |
|
|
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 4071 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 4127 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 4190 NDV nd_recv_ndv() |
|
Line 4439 NDV nd_recv_ndv() |
|
if ( !len ) return 0; |
if ( !len ) return 0; |
else { |
else { |
m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len); |
m0 = m = (NMV)GC_malloc_atomic_ignore_off_page(nmv_adv*len); |
|
#if 0 |
ndv_alloc += len*nmv_adv; |
ndv_alloc += len*nmv_adv; |
|
#endif |
for ( i = 0; i < len; i++, NMV_ADV(m) ) { |
for ( i = 0; i < len; i++, NMV_ADV(m) ) { |
CM(m) = nd_recv_int(); |
CM(m) = nd_recv_int(); |
nd_recv_intarray(DL(m),nd_wpd); |
nd_recv_intarray(DL(m),nd_wpd); |
Line 4381 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 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 4586 NDV ndv_load(int index) |
|
Line 4841 NDV ndv_load(int index) |
|
MKNDV(nv,m0,len,d); |
MKNDV(nv,m0,len,d); |
SG(d) = sugar; |
SG(d) = sugar; |
return d; |
return d; |
|
} |
|
|
|
void nd_det(int mod,MAT f,P *rp) |
|
{ |
|
VL fv,tv; |
|
int n,i,j,max,e,nvar,sgn,k0,l0,len0,len,k,l,a; |
|
pointer **m; |
|
Q mone; |
|
NDV **dm; |
|
NDV *t,*mi,*mj; |
|
NDV d,s,mij,mjj; |
|
ND u; |
|
NMV nmv; |
|
PGeoBucket bucket; |
|
struct order_spec *ord; |
|
|
|
create_order_spec(0,0,&ord); |
|
nd_init_ord(ord); |
|
get_vars((Obj)f,&fv); |
|
if ( f->row != f->col ) |
|
error("nd_det : non-square matrix"); |
|
n = f->row; |
|
for ( nvar = 0, tv = fv; tv; tv = NEXT(tv), nvar++ ); |
|
m = f->body; |
|
for ( i = 0, max = 0; i < n; i++ ) |
|
for ( j = 0; j < n; j++ ) |
|
for ( tv = fv; tv; tv = NEXT(tv) ) { |
|
e = getdeg(tv->v,(P)m[i][j]); |
|
max = MAX(e,max); |
|
} |
|
nd_setup_parameters(nvar,1024); |
|
dm = (NDV **)almat_pointer(n,n); |
|
for ( i = 0, max = 0; i < n; i++ ) |
|
for ( j = 0; j < n; j++ ) { |
|
dm[i][j] = ptondv(CO,fv,m[i][j]); |
|
if ( mod ) ndv_mod(mod,dm[i][j]); |
|
if ( dm[i][j] && !LEN(dm[i][j]) ) dm[i][j] = 0; |
|
} |
|
d = ptondv(CO,fv,(P)ONE); |
|
if ( mod ) ndv_mod(mod,d); |
|
chsgnq(ONE,&mone); |
|
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++ ); |
|
if ( i == n ) { |
|
*rp = 0; |
|
return; |
|
} |
|
k0 = i; l0 = j; len0 = LEN(dm[k0][l0]); |
|
for ( k = j; k < n; k++ ) |
|
for ( l = j; l < n; l++ ) |
|
if ( dm[k][l] && LEN(dm[k][l]) < len0 ) { |
|
k0 = k; l0 = l; len0 = LEN(dm[k][l]); |
|
} |
|
if ( k0 != j ) { |
|
t = dm[j]; dm[j] = dm[k0]; dm[k0] = t; |
|
sgn = -sgn; |
|
} |
|
if ( l0 != j ) { |
|
for ( k = j; k < n; k++ ) { |
|
s = dm[k][j]; dm[k][j] = dm[k][l0]; dm[k][l0] = s; |
|
} |
|
sgn = -sgn; |
|
} |
|
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]; |
|
if ( mod ) |
|
ndv_mul_c(mod,mij,mod-1); |
|
else |
|
ndv_mul_c_q(mij,mone); |
|
for ( k = j+1; k < n; k++ ) { |
|
if ( DP_Print ) fprintf(stderr,"k=%d ",k); |
|
bucket = create_pbucket(); |
|
if ( mi[k] ) { |
|
nmv = BDY(mjj); len = LEN(mjj); |
|
for ( a = 0; a < len; a++, NMV_ADV(nmv) ) { |
|
u = ndv_mul_nmv_trunc(mod,nmv,mi[k],DL(BDY(d))); |
|
add_pbucket(mod,bucket,u); |
|
} |
|
} |
|
if ( mj[k] && mij ) { |
|
nmv = BDY(mij); len = LEN(mij); |
|
for ( a = 0; a < len; a++, NMV_ADV(nmv) ) { |
|
u = ndv_mul_nmv_trunc(mod,nmv,mj[k],DL(BDY(d))); |
|
add_pbucket(mod,bucket,u); |
|
} |
|
} |
|
u = nd_quo(mod,bucket,d); |
|
mi[k] = ndtondv(mod,u); |
|
} |
|
if ( DP_Print ) fprintf(stderr,"\n",k); |
|
} |
|
d = mjj; |
|
} |
|
if ( sgn < 0 ) |
|
if ( mod ) |
|
ndv_mul_c(mod,d,mod-1); |
|
else |
|
ndv_mul_c_q(d,mone); |
|
*rp = ndvtop(mod,CO,fv,d); |
|
} |
|
|
|
ND ndv_mul_nmv_trunc(int mod,NMV m0,NDV p,UINT *d) |
|
{ |
|
NM mr,mr0; |
|
NM tnm; |
|
NMV m; |
|
UINT *d0,*dt,*dm; |
|
int c,n,td,i,c1,c2,len; |
|
Q q; |
|
ND r; |
|
|
|
if ( !p ) return 0; |
|
else { |
|
n = NV(p); m = BDY(p); len = LEN(p); |
|
d0 = DL(m0); |
|
td = TD(d); |
|
mr0 = 0; |
|
NEWNM(tnm); |
|
if ( mod ) { |
|
c = CM(m0); |
|
for ( i = 0; i < len; i++, NMV_ADV(m) ) { |
|
ndl_add(DL(m),d0,DL(tnm)); |
|
if ( ndl_reducible(DL(tnm),d) ) { |
|
NEXTNM(mr0,mr); |
|
c1 = CM(m); DMAR(c1,c,0,mod,c2); CM(mr) = c2; |
|
ndl_copy(DL(tnm),DL(mr)); |
|
} |
|
} |
|
} else { |
|
q = CQ(m0); |
|
for ( i = 0; i < len; i++, NMV_ADV(m) ) { |
|
ndl_add(DL(m),d0,DL(tnm)); |
|
if ( ndl_reducible(DL(tnm),d) ) { |
|
NEXTNM(mr0,mr); |
|
mulq(CQ(m),q,&CQ(mr)); |
|
ndl_copy(DL(tnm),DL(mr)); |
|
} |
|
} |
|
} |
|
if ( !mr0 ) |
|
return 0; |
|
else { |
|
NEXT(mr) = 0; |
|
for ( len = 0, mr = mr0; mr; mr = NEXT(mr), len++ ); |
|
MKND(NV(p),mr0,len,r); |
|
SG(r) = SG(p) + TD(d0); |
|
return r; |
|
} |
|
} |
} |
} |