version 1.94, 2004/03/15 07:30:44 |
version 1.96, 2004/03/17 08:16:24 |
|
|
/* $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 1.95 2004/03/15 08:44:52 noro Exp $ */ |
|
|
#include "nd.h" |
#include "nd.h" |
|
|
Line 29 static NDV *nd_ps; |
|
Line 29 static NDV *nd_ps; |
|
static NDV *nd_ps_trace; |
static NDV *nd_ps_trace; |
static RHist *nd_psh; |
static RHist *nd_psh; |
static int nd_psn,nd_pslen; |
static int nd_psn,nd_pslen; |
|
|
static RHist *nd_red; |
static RHist *nd_red; |
|
static int *nd_work_vector; |
|
static int **nd_matrix; |
|
static int nd_matrix_len; |
|
|
static int nd_found,nd_create,nd_notfirst; |
static int nd_found,nd_create,nd_notfirst; |
static int nmv_adv; |
static int nmv_adv; |
Line 422 int ndl_block_compare(UINT *d1,UINT *d2) |
|
Line 424 int ndl_block_compare(UINT *d1,UINT *d2) |
|
return 0; |
return 0; |
} |
} |
|
|
|
int ndl_matrix_compare(UINT *d1,UINT *d2) |
|
{ |
|
int i,j,s; |
|
int *v; |
|
|
|
for ( j = 0; j < nd_nvar; j++ ) |
|
nd_work_vector[j] = GET_EXP(d1,j)-GET_EXP(d2,j); |
|
for ( i = 0; i < nd_matrix_len; i++ ) { |
|
v = nd_matrix[i]; |
|
for ( j = 0, s = 0; j < nd_nvar; j++ ) |
|
s += v[j]*nd_work_vector[j]; |
|
if ( s > 0 ) return 1; |
|
else if ( s < 0 ) return -1; |
|
} |
|
return 0; |
|
} |
|
|
/* TDH -> WW -> TD-> RL */ |
/* TDH -> WW -> TD-> RL */ |
|
|
int ndl_ww_lex_compare(UINT *d1,UINT *d2) |
int ndl_ww_lex_compare(UINT *d1,UINT *d2) |
Line 873 ND nd_add(int mod,ND p1,ND p2) |
|
Line 892 ND nd_add(int mod,ND p1,ND p2) |
|
} |
} |
} |
} |
|
|
|
/* 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); |
|
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; |
|
} |
|
} |
|
|
ND nd_add_sf(ND p1,ND p2) |
ND nd_add_sf(ND p1,ND p2) |
{ |
{ |
int n,c,can; |
int n,c,can; |
Line 2551 UINT *ndv_compute_bound(NDV p) |
|
Line 2643 UINT *ndv_compute_bound(NDV p) |
|
int nd_get_exporigin(struct order_spec *ord) |
int nd_get_exporigin(struct order_spec *ord) |
{ |
{ |
switch ( ord->id ) { |
switch ( ord->id ) { |
case 0: |
case 0: case 2: |
return 1; |
return 1; |
case 1: |
case 1: |
/* block order */ |
/* block order */ |
/* d[0]:weight d[1]:w0,...,d[nd_exporigin-1]:w(n-1) */ |
/* d[0]:weight d[1]:w0,...,d[nd_exporigin-1]:w(n-1) */ |
return ord->ord.block.length+1; |
return ord->ord.block.length+1; |
case 2: |
case 3: |
error("nd_get_exporigin : matrix order is not supported yet."); |
error("nd_get_exporigin : composite order is not supported yet."); |
} |
} |
} |
} |
|
|
Line 2603 void nd_setup_parameters(int nvar,int max) { |
|
Line 2695 void nd_setup_parameters(int nvar,int max) { |
|
nmv_adv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(nd_wpd-1)*sizeof(UINT)); |
nmv_adv = ROUND_FOR_ALIGN(sizeof(struct oNMV)+(nd_wpd-1)*sizeof(UINT)); |
nd_epos = nd_create_epos(nd_ord); |
nd_epos = nd_create_epos(nd_ord); |
nd_blockmask = nd_create_blockmask(nd_ord); |
nd_blockmask = nd_create_blockmask(nd_ord); |
|
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 mod,int trace,ND_pairs d) |
Line 3354 void nd_init_ord(struct order_spec *ord) |
|
Line 3447 void nd_init_ord(struct order_spec *ord) |
|
} |
} |
break; |
break; |
case 1: |
case 1: |
|
/* block order */ |
/* XXX */ |
/* XXX */ |
nd_dcomp = -1; |
nd_dcomp = -1; |
nd_isrlex = 0; |
nd_isrlex = 0; |
ndl_compare_function = ndl_block_compare; |
ndl_compare_function = ndl_block_compare; |
break; |
break; |
case 2: |
case 2: |
error("nd_init_ord : matrix order is not supported yet."); |
/* matrix order */ |
|
/* XXX */ |
|
nd_dcomp = -1; |
|
nd_isrlex = 0; |
|
nd_matrix_len = ord->ord.matrix.row; |
|
nd_matrix = ord->ord.matrix.matrix; |
|
ndl_compare_function = ndl_matrix_compare; |
break; |
break; |
|
case 3: |
|
error("nd_init_ord : composite order is not supported yet."); |
|
break; |
} |
} |
nd_ord = ord; |
nd_ord = ord; |
} |
} |
Line 3372 BlockMask nd_create_blockmask(struct order_spec *ord) |
|
Line 3475 BlockMask nd_create_blockmask(struct order_spec *ord) |
|
UINT *t; |
UINT *t; |
BlockMask bm; |
BlockMask bm; |
|
|
if ( !ord->id ) |
/* we only create mask table for block order */ |
|
if ( ord->id != 1 ) |
return 0; |
return 0; |
n = ord->ord.block.length; |
n = ord->ord.block.length; |
bm = (BlockMask)MALLOC(sizeof(struct oBlockMask)); |
bm = (BlockMask)MALLOC(sizeof(struct oBlockMask)); |
Line 3430 EPOS nd_create_epos(struct order_spec *ord) |
|
Line 3534 EPOS nd_create_epos(struct order_spec *ord) |
|
} |
} |
break; |
break; |
case 2: |
case 2: |
error("nd_create_epos : matrix order is not supported yet."); |
/* matrix order */ |
|
case 3: |
|
/* composite order */ |
|
for ( i = 0; i < nd_nvar; i++ ) { |
|
epos[i].i = nd_exporigin + i/nd_epw; |
|
epos[i].s = (nd_epw-(i%nd_epw)-1)*nd_bpe; |
|
} |
|
break; |
} |
} |
return epos; |
return epos; |
} |
} |