/*
* @-
* The operator below is only working for a very limited
* case. It also re-uses oids, which may become a semantic
* problem quickly.
*/
str DCdeleteUpperSlice(int *ret, int *bid, int *pos)
{
BAT *b;
int *readerT, *writerT;
BUN size, i;
(void) ret;
if ((b = BATdescriptor(*bid)) == NULL)
throw(MAL, "dc.deleteUpperSlice", "Cannot access input BAT");
/* check for a failure */
assert(b != NULL);
/* remove Hashes etc */
if (b->H->hash)
HASHremove(b);
if (b->T->hash)
HASHremove(BATmirror(b));
size = BATcount(b);
writerT = (int *) Tloc(b, BUNfirst(b));
readerT = (int *) Tloc(b, BUNfirst(b)) + *pos;
for (i = *pos; i < size; i++)
*writerT++ = *readerT++;
b->batInserted -= *pos;
BATsetcount(b, (BUN) (writerT - (int *) Tloc(b, BUNfirst(b))));
BBPunfix(*bid);
b->batDirty = TRUE;
return MAL_SUCCEED;
}
static BAT *
MATproject_hge( BAT *map, BAT **bats, int len, int ttpe )
{
BAT *res;
int i;
BUN j, cnt = BATcount(map);
hge *resT, **batsT;
bte *mapT;
res = BATnew(TYPE_void, ttpe, cnt, TRANSIENT);
batsT = (hge**)GDKmalloc(sizeof(hge*) * len);
if (res == NULL || batsT == NULL) {
if (res)
BBPreclaim(res);
if (batsT)
GDKfree(batsT);
return NULL;
}
BATseqbase(res, map->hseqbase);
resT = (hge*)Tloc(res, 0);
mapT = (bte*)Tloc(map, 0);
for (i=0; i<len; i++)
batsT[i] = (hge*)Tloc(bats[i], 0);
for (j=0; j<cnt; j++)
resT[j] = *batsT[mapT[j]]++;
BATsetcount(res, j);
res->hrevsorted = j <= 1;
GDKfree(batsT);
return res;
}
str DCselectInsert(int *ret, int *res, int *bid, lng *low, lng *hgh)
{
BAT *b, *r;
lng *readerH, *writerH;
lng *readerT, *writerT;
BUN size, i;
(void) ret;
if ((b = BATdescriptor(*bid)) == NULL)
throw(MAL, "dc.selectInsert", "Cannot access input BAT");
if ((r = BATdescriptor(*res)) == NULL)
throw(MAL, "dc.selectInsert", "Cannot access result BAT");
size = BATcount(b);
if (size > BATcapacity(r) - BATcount(r)) {
BUN ncap;
BUN grows;
BUN needed = size - (BATcapacity(r) - BATcount(r));
ncap = BATcapacity(r) + needed;
grows = BATgrows(r);
if (ncap > grows)
grows = ncap;
if (BATextend(r, grows) == NULL)
throw(MAL, "dc.selectInsert", "Failed to make room for the new values");
}
/*printf("in dc.selectInsert size is "OIDFMT,size);*/
writerH = (lng *) Hloc(r, BUNfirst(r));
writerT = (lng *) Tloc(r, BUNfirst(r));
readerH = (lng *) Hloc(b, BUNfirst(b));
readerT = (lng *) Tloc(b, BUNfirst(b));
for (i = 0; i < size; i++) {
if (*readerT >= *low && *readerT <= *hgh) {
*writerH = *readerH;
*writerT = *readerT;
writerH++;
writerT++;
}
readerH++;
readerT++;
}
BATsetcount(r, (BUN) (writerT - (lng *) Tloc(r, BUNfirst(r))));
BBPunfix(*bid);
BBPunfix(*res);
return MAL_SUCCEED;
}
/*
* @-
* The operator below is only working for a very limited cases.
*/
str DCreplaceTailBasedOnHead(int *ret, int *res, int *bid)
{
BAT *b, *r;
oid *readerH_b;
int *writerT_r, *readerT_b;
BUN size_b, size_r, i;
(void) ret;
if ((b = BATdescriptor(*bid)) == NULL)
throw(MAL, "dc.replaceTailBasedOnHead", "Cannot access input BAT");
/* check for a failure */
assert(b != NULL);
if ((r = BATdescriptor(*res)) == NULL)
throw(MAL, "dc.replaceTailBasedOnHead", "Cannot access result BAT");
/* check for a failure */
assert(r != NULL);
/* remove Hashes etc */
if (r->H->hash)
HASHremove(r);
if (r->T->hash)
HASHremove(BATmirror(r));
size_r = BATcount(r);
size_b = BATcount(b);
if ((b->htype == TYPE_void) && (size_b == size_r)) {
writerT_r = (int *) Tloc(r, BUNfirst(r));
readerT_b = (int *) Tloc(b, BUNfirst(b));
for (i = 0; i < size_r; i++) {
*writerT_r = *readerT_b;
writerT_r++;
readerT_b++;
}
} else if ((b->htype != TYPE_void) && (size_b < size_r)) {
readerH_b = (oid *) Hloc(b, BUNfirst(b));
readerT_b = (int *) Tloc(b, BUNfirst(b));
for (i = 0; i < size_b; i++) {
writerT_r = (int *) Tloc(r, BUNfirst(r)) + *readerH_b;
*writerT_r = *readerT_b;
readerH_b++;
readerT_b++;
}
}
BBPunfix(*bid);
BBPunfix(*res);
r->batDirty = TRUE;
return MAL_SUCCEED;
}
static BAT *
MATsort_bte( BAT **map, BAT **bats, int len, BUN cnt, int rev )
{
BAT *res;
int i;
bte *resT, **batsT, *in;
bte *mapT;
BUN len1, len2;
bte *map_in = NULL;
res = BATnew(TYPE_void, bats[0]->ttype, cnt, TRANSIENT);
*map = BATnew(TYPE_void, TYPE_bte, cnt, TRANSIENT);
if (res == NULL || *map == NULL) {
BBPreclaim(res);
BBPreclaim(*map);
*map = NULL;
return NULL;
}
BATseqbase(res, 0);
BATseqbase(*map, 0);
resT = (bte*)Tloc(res, 0);
mapT = (bte*)Tloc(*map, 0);
batsT = (bte**)GDKmalloc(sizeof(bte*) * len);
for (i=0; i<len; i++)
batsT[i] = (bte*)Tloc(bats[i], 0);
/* merge */
in = batsT[0];
len1 = BATcount(bats[0]);
map_in = NULL;
/* TODO: change into a tree version */
for (i=1; i<len; i++) {
len2 = BATcount(bats[i]);
if (rev) {
MATsortloop_bte_rev( resT+cnt-len1-len2,
mapT+cnt-len1-len2,
in, map_in, len1,
batsT[i], i, len2);
} else {
MATsortloop_bte_( resT+cnt-len1-len2,
mapT+cnt-len1-len2,
in, map_in, len1,
batsT[i], i, len2);
}
in = resT+cnt-len1-len2;
map_in = mapT+cnt-len1-len2;
len1 += len2;
}
BATsetcount(res, len1);
BATsetcount(*map, len1);
res->hrevsorted = len1 <= 1;
(*map)->hrevsorted = len1 <= 1;
GDKfree(batsT);
return res;
}
static BAT *
MATproject_any( BAT *map, BAT **bats, int len )
{
BAT *res;
int i;
BUN j, cnt = BATcount(map);
BATiter *bats_i;
BUN *batsT;
bte *mapT;
res = BATnew(TYPE_void, bats[0]->ttype, cnt, TRANSIENT);
batsT = (BUN*)GDKmalloc(sizeof(BUN) * len);
bats_i = (BATiter*)GDKmalloc(sizeof(BATiter) * len);
if (res == NULL || batsT == NULL || bats_i == NULL) {
if (res)
BBPreclaim(res);
if (batsT)
GDKfree(batsT);
if (bats_i)
GDKfree(bats_i);
return NULL;
}
BATseqbase(res, map->hseqbase);
mapT = (bte*)Tloc(map, 0);
for (i=0; i<len; i++) {
batsT[i] = 0;
bats_i[i] = bat_iterator(bats[i]);
}
for (j=0; j<cnt; j++)
BUNappend(res, BUNtail(bats_i[mapT[j]], batsT[mapT[j]]++), FALSE);
GDKfree(batsT);
GDKfree(bats_i);
return res;
}
static BAT *
MATsort_any( BAT **map, BAT **bats, int len, BUN cnt, int rev )
{
BAT *res = 0, *in;
int i;
bte *mapT;
BUN len1, len2;
bte *map_in = NULL;
*map = BATnew(TYPE_void, TYPE_bte, cnt, TRANSIENT);
if (*map == NULL)
return NULL;
BATseqbase(*map, 0);
mapT = (bte*)Tloc(*map, 0);
/* merge */
/* TODO: change into a tree version */
in = bats[0];
len1 = BATcount(in);
for (i=1; i<len; i++) {
len2 = BATcount(bats[i]);
if (rev) {
res = MATsortloop_rev(
mapT+cnt-len1-len2,
in, map_in, len1,
bats[i], i, len2);
} else {
res = MATsortloop_(
mapT+cnt-len1-len2,
in, map_in, len1,
bats[i], i, len2);
}
if (i != 1)
BBPunfix(in->batCacheid);
if (res == NULL)
return NULL;
in = res;
map_in = mapT+cnt-len1-len2;
len1 += len2;
}
BATsetcount(*map, len1);
(*map)->hrevsorted = len1 <= 1;
return res;
}
str
MKEYbathash(bat *res, const bat *bid)
{
BAT *b, *dst;
wrd *r;
BUN n;
if ((b = BATdescriptor(*bid)) == NULL)
throw(SQL, "mkey.bathash", RUNTIME_OBJECT_MISSING);
assert(BAThvoid(b) || BAThrestricted(b));
n = BATcount(b);
dst = BATnew(TYPE_void, TYPE_wrd, n, TRANSIENT);
if (dst == NULL) {
BBPunfix(b->batCacheid);
throw(SQL, "mkey.bathash", MAL_MALLOC_FAIL);
}
BATseqbase(dst, b->hseqbase);
BATsetcount(dst, n);
r = (wrd *) Tloc(dst, BUNfirst(dst));
switch (ATOMstorage(b->ttype)) {
case TYPE_void: {
oid o = b->tseqbase;
if (o == oid_nil)
while (n-- > 0)
*r++ = wrd_nil;
else
while (n-- > 0)
*r++ = (wrd) o++;
break;
}
case TYPE_bte: {
bte *v = (bte *) Tloc(b, BUNfirst(b));
while (n-- > 0) {
*r++ = MKEYHASH_bte(v);
v++;
}
break;
}
case TYPE_sht: {
sht *v = (sht *) Tloc(b, BUNfirst(b));
while (n-- > 0) {
*r++ = MKEYHASH_sht(v);
v++;
}
break;
}
case TYPE_int:
case TYPE_flt: {
int *v = (int *) Tloc(b, BUNfirst(b));
while (n-- > 0) {
*r++ = MKEYHASH_int(v);
v++;
}
break;
}
case TYPE_lng:
case TYPE_dbl: {
lng *v = (lng *) Tloc(b, BUNfirst(b));
while (n-- > 0) {
*r++ = MKEYHASH_lng(v);
v++;
}
break;
}
#ifdef HAVE_HGE
case TYPE_hge: {
hge *v = (hge *) Tloc(b, BUNfirst(b));
while (n-- > 0) {
*r++ = MKEYHASH_hge(v);
v++;
}
break;
}
#endif
default: {
BATiter bi = bat_iterator(b);
BUN (*hash)(const void *) = BATatoms[b->ttype].atomHash;
int (*cmp)(const void *, const void *) = ATOMcompare(b->ttype);
void *nil = ATOMnilptr(b->ttype);
BUN i;
const void *v;
BATloop(b, i, n) {
v = BUNtail(bi, i);
if ((*cmp)(v, nil) == 0)
*r++ = wrd_nil;
else
*r++ = (wrd) (*hash)(v);
}
break;
}
}
/* BATsample implements sampling for void headed BATs */
BAT *
BATsample(BAT *b, BUN n)
{
BAT *bn;
BUN cnt, slen;
BUN rescnt;
struct oidtreenode *tree = NULL;
BATcheck(b, "BATsample", NULL);
assert(BAThdense(b));
ERRORcheck(n > BUN_MAX, "BATsample: sample size larger than BUN_MAX\n", NULL);
ALGODEBUG
fprintf(stderr, "#BATsample: sample " BUNFMT " elements.\n", n);
cnt = BATcount(b);
/* empty sample size */
if (n == 0) {
bn = BATnew(TYPE_void, TYPE_void, 0, TRANSIENT);
if (bn == NULL) {
GDKerror("BATsample: memory allocation error");
return NULL;
}
BATsetcount(bn, 0);
BATseqbase(bn, 0);
BATseqbase(BATmirror(bn), 0);
/* sample size is larger than the input BAT, return all oids */
} else if (cnt <= n) {
bn = BATnew(TYPE_void, TYPE_void, cnt, TRANSIENT);
if (bn == NULL) {
GDKerror("BATsample: memory allocation error");
return NULL;
}
BATsetcount(bn, cnt);
BATseqbase(bn, 0);
BATseqbase(BATmirror(bn), b->H->seq);
} else {
oid minoid = b->hseqbase;
oid maxoid = b->hseqbase + cnt;
/* if someone samples more than half of our tree, we
* do the antiset */
bit antiset = n > cnt / 2;
slen = n;
if (antiset)
n = cnt - n;
tree = GDKmalloc(n * sizeof(struct oidtreenode));
if (tree == NULL) {
GDKerror("#BATsample: memory allocation error");
return NULL;
}
bn = BATnew(TYPE_void, TYPE_oid, slen, TRANSIENT);
if (bn == NULL) {
GDKfree(tree);
GDKerror("#BATsample: memory allocation error");
return NULL;
}
/* while we do not have enough sample OIDs yet */
for (rescnt = 0; rescnt < n; rescnt++) {
oid candoid;
do {
/* generate a new random OID */
candoid = (oid) (minoid + DRAND * (maxoid - minoid));
/* if that candidate OID was already
* generated, try again */
} while (!OIDTreeMaybeInsert(tree, candoid, rescnt));
}
if (!antiset) {
OIDTreeToBAT(tree, bn);
} else {
OIDTreeToBATAntiset(tree, bn, minoid, maxoid);
}
GDKfree(tree);
BATsetcount(bn, slen);
bn->trevsorted = bn->batCount <= 1;
bn->tsorted = 1;
bn->tkey = 1;
bn->tdense = bn->batCount <= 1;
if (bn->batCount == 1)
bn->tseqbase = *(oid *) Tloc(bn, BUNfirst(bn));
bn->hdense = 1;
bn->hseqbase = 0;
bn->hkey = 1;
bn->hrevsorted = bn->batCount <= 1;
bn->hsorted = 1;
}
return bn;
}
static str
AGGRsubgroupedExt(bat *retval1, bat *retval2, bat *bid, bat *gid, bat *eid, bat *sid,
int skip_nils, int abort_on_error, int tp,
BAT *(*grpfunc1)(BAT *, BAT *, BAT *, BAT *, int, int, int),
gdk_return (*grpfunc2)(BAT **, BAT **, BAT *, BAT *, BAT *, BAT *, int, int, int),
BAT *(*quantilefunc)(BAT *, BAT *, BAT *, BAT *, int, double, int, int),
bat *quantile,
const char *malfunc)
{
BAT *b, *g, *e, *s, *bn = NULL, *cnts, *q = NULL;
double qvalue;
/* one of grpfunc1, grpfunc2 and quantilefunc is non-NULL and the others are */
assert((grpfunc1 && grpfunc2 == NULL && quantilefunc == NULL) ||
(grpfunc1 == NULL && grpfunc2 && quantilefunc == NULL) ||
(grpfunc1 == NULL && grpfunc2 == NULL && quantilefunc) );
/* if retval2 is non-NULL, we must have grpfunc2 */
assert(retval2 == NULL || grpfunc2 != NULL);
b = BATdescriptor(*bid);
g = gid ? BATdescriptor(*gid) : NULL;
e = eid ? BATdescriptor(*eid) : NULL;
q = quantile ? BATdescriptor(*quantile) : NULL;
if (b == NULL || (gid != NULL && g == NULL) || (eid != NULL && e == NULL)) {
if (b)
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
throw(MAL, malfunc, RUNTIME_OBJECT_MISSING);
}
if (tp == TYPE_any && (grpfunc1 == BATgroupmedian || quantilefunc == BATgroupquantile))
tp = b->ttype;
if (sid) {
s = BATdescriptor(*sid);
if (s == NULL) {
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
throw(MAL, malfunc, RUNTIME_OBJECT_MISSING);
}
} else {
if (!BAThdense(b)) {
/* XXX backward compatibility code: ignore non-dense head, but
* only if no candidate list */
s = BATmirror(BATmark(BATmirror(b), 0));
BBPreleaseref(b->batCacheid);
b = s;
}
s = NULL;
}
if (grpfunc1)
bn = (*grpfunc1)(b, g, e, s, tp, skip_nils, abort_on_error);
if (quantilefunc) {
assert(BATcount(q)>0);
assert(q->ttype == TYPE_dbl);
qvalue = ((const double *)Tloc(q, BUNfirst(q)))[0];
if (qvalue < 0|| qvalue > 1) {
char *s;
s = createException(MAL, malfunc, "quantile value of %f is not in range [0,1]", qvalue);
return s;
}
bn = (*quantilefunc)(b, g, e, s, tp, qvalue, skip_nils, abort_on_error);
}
if (grpfunc2 && (*grpfunc2)(&bn, retval2 ? &cnts : NULL, b, g, e, s, tp, skip_nils, abort_on_error) == GDK_FAIL)
bn = NULL;
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
if (s)
BBPreleaseref(s->batCacheid);
if (bn == NULL) {
char *errbuf = GDKerrbuf;
char *s;
if (errbuf && *errbuf) {
if (strncmp(errbuf, "!ERROR: ", 8) == 0)
errbuf += 8;
if (strchr(errbuf, '!') == errbuf + 5) {
s = createException(MAL, malfunc, "%s", errbuf);
} else if ((s = strchr(errbuf, ':')) != NULL && s[1] == ' ') {
s = createException(MAL, malfunc, "%s", s + 2);
} else {
s = createException(MAL, malfunc, "%s", errbuf);
}
*GDKerrbuf = 0;
return s;
}
throw(MAL, malfunc, OPERATION_FAILED);
}
*retval1 = bn->batCacheid;
BBPkeepref(bn->batCacheid);
if (retval2) {
*retval2 = cnts->batCacheid;
BBPkeepref(cnts->batCacheid);
}
return MAL_SUCCEED;
}
/*
* grouped aggregates
*/
static str
AGGRgrouped(bat *retval1, bat *retval2, BAT *b, BAT *g, BAT *e, int tp,
BAT *(*grpfunc1)(BAT *, BAT *, BAT *, BAT *, int, int, int),
gdk_return (*grpfunc2)(BAT **, BAT **, BAT *, BAT *, BAT *, BAT *, int, int, int),
BAT *(*quantilefunc)(BAT *, BAT *, BAT *, BAT *, int, double, int, int),
BAT *quantile,
int skip_nils,
const char *malfunc)
{
BAT *bn, *cnts = NULL, *t, *map;
double qvalue;
/* one of grpfunc1, grpfunc2 and quantilefunc is non-NULL and the others are */
assert((grpfunc1 != NULL && grpfunc2 == NULL && quantilefunc == NULL) ||
(grpfunc1 == NULL && grpfunc2 != NULL && quantilefunc == NULL) ||
(grpfunc1 == NULL && grpfunc2 == NULL && quantilefunc != NULL) );
/* if retval2 is non-NULL, we must have grpfunc2 */
assert(retval2 == NULL || grpfunc2 != NULL);
assert(quantile == NULL || quantilefunc != NULL);
if (b == NULL || g == NULL || e == NULL) {
if (b)
BBPreleaseref(b->batCacheid);
if (g)
BBPreleaseref(g->batCacheid);
if (e)
BBPreleaseref(e->batCacheid);
throw(MAL, malfunc, RUNTIME_OBJECT_MISSING);
}
if (tp == TYPE_any && (grpfunc1 == BATgroupmedian || quantilefunc == BATgroupquantile))
tp = b->ttype;
if (!BAThdense(b) || !BAThdense(g)) {
/* if b or g don't have a dense head, replace the head with a
* dense sequence */
t = BATjoin(BATmirror(b), g, MIN(BATcount(b), BATcount(g)));
BBPreleaseref(b->batCacheid);
BBPreleaseref(g->batCacheid);
b = BATmirror(BATmark(t, 0));
g = BATmirror(BATmark(BATmirror(t), 0));
BBPreleaseref(t->batCacheid);
}
if (b->hseqbase != g->hseqbase || BATcount(b) != BATcount(g)) {
/* b and g are not aligned: align them by creating a view on
* one or the other */
oid min; /* lowest common oid */
oid max; /* highest common oid */
min = b->hseqbase;
if (min < g->hseqbase)
min = g->hseqbase;
max = b->hseqbase + BATcount(b);
if (g->hseqbase + BATcount(g) < max)
max = g->hseqbase + BATcount(g);
if (b->hseqbase != min || b->hseqbase + BATcount(b) != max) {
if (min >= max)
min = max = b->hseqbase;
t = BATslice(b, BUNfirst(b) + (BUN) (min - b->hseqbase),
BUNfirst(b) + (BUN) (max - b->hseqbase));
BBPreleaseref(b->batCacheid);
b = t;
}
if (g->hseqbase != min || g->hseqbase + BATcount(g) != max) {
if (min >= max)
min = max = g->hseqbase;
t = BATslice(g, BUNfirst(g) + (BUN) (min - g->hseqbase),
BUNfirst(g) + (BUN) (max - g->hseqbase));
BBPreleaseref(g->batCacheid);
g = t;
}
}
if (!BAThdense(e)) {
/* if e doesn't have a dense head, renumber the group ids with
* a dense sequence at the cost of some left joins */
map = BATmark(e, 0); /* [gid,newgid(dense)] */
BBPreleaseref(e->batCacheid);
e = BATmirror(map); /* [newgid(dense),gid] */
t = BATleftjoin(g, map, BATcount(g)); /* [oid,newgid] */
BBPreleaseref(g->batCacheid);
g = t;
} else {
map = NULL;
}
if (grpfunc1)
bn = (*grpfunc1)(b, g, e, NULL, tp, skip_nils, 1);
if (quantilefunc) {
assert(BATcount(quantile)>0);
assert(quantile->ttype == TYPE_dbl);
qvalue = ((const double *)Tloc(quantile, BUNfirst(quantile)))[0];
if (qvalue < 0|| qvalue > 1) {
char *s;
s = createException(MAL, malfunc, "quantile value of %f is not in range [0,1]", qvalue);
return s;
}
bn = (*quantilefunc)(b, g, e, NULL, tp, qvalue, skip_nils, 1);
}
if (grpfunc2 && (*grpfunc2)(&bn, retval2 ? &cnts : NULL, b, g, e, NULL, tp, skip_nils, 1) == GDK_FAIL)
bn = NULL;
if (bn != NULL && (grpfunc1 == BATgroupmin || grpfunc1 == BATgroupmax)) {
t = BATproject(bn, b);
BBPreleaseref(bn->batCacheid);
bn = t;
}
BBPreleaseref(b->batCacheid);
BBPreleaseref(g->batCacheid);
if (map == NULL) /* if map!=NULL, e is mirror of map */
BBPreleaseref(e->batCacheid);
if (bn == NULL) {
char *errbuf = GDKerrbuf;
char *s;
if (map)
BBPreleaseref(map->batCacheid);
if (errbuf && *errbuf) {
if (strncmp(errbuf, "!ERROR: ", 8) == 0)
errbuf += 8;
if (strchr(errbuf, '!') == errbuf + 5) {
s = createException(MAL, malfunc, "%s", errbuf);
} else if ((s = strchr(errbuf, ':')) != NULL && s[1] == ' ') {
s = createException(MAL, malfunc, "%s", s + 2);
} else {
s = createException(MAL, malfunc, "%s", errbuf);
}
*GDKerrbuf = 0;
return s;
}
throw(MAL, malfunc, OPERATION_FAILED);
}
if (map) {
t = BATleftjoin(map, bn, BATcount(bn));
BBPreleaseref(bn->batCacheid);
bn = t;
if (cnts) {
t = BATleftjoin(map, cnts, BATcount(cnts));
BBPreleaseref(cnts->batCacheid);
cnts = t;
}
BBPreleaseref(map->batCacheid);
}
*retval1 = bn->batCacheid;
BBPkeepref(bn->batCacheid);
if (retval2) {
*retval2 = cnts->batCacheid;
BBPkeepref(cnts->batCacheid);
}
return MAL_SUCCEED;
}
// merging multiple OID lists, optimized for empty bats
// Further improvement should come from multi-bat merging.
str
MATmergepack(Client cntxt, MalBlkPtr mb, MalStkPtr stk, InstrPtr p)
{
int i,j= 0, *ret = (int*) getArgReference(stk,p,0);
int top=0;
oid **o_end, **o_src, *o, *oo, onxt;
BAT *b, *bn, *bm, **bats;
BUN cap = 0;
(void)cntxt;
(void)mb;
bats = (BAT**) GDKzalloc(sizeof(BAT*) * p->argc);
o_end = (oid**) GDKzalloc(sizeof(oid*) * p->argc);
o_src = (oid**) GDKzalloc(sizeof(oid*) * p->argc);
if ( bats ==0 || o_end == 0 || o_src == 0){
if (bats) GDKfree(bats);
if (o_src) GDKfree(o_src);
if (o_end) GDKfree(o_end);
throw(MAL,"mat.mergepack",MAL_MALLOC_FAIL);
}
for (i = 1; i < p->argc; i++) {
int bid = stk->stk[getArg(p,i)].val.ival;
b = BATdescriptor(abs(bid));
if (b ){
cap += BATcount(b);
if ( BATcount(b) ){
// pre-sort the arguments
onxt = *(oid*) Tloc(b,BUNfirst(b));
for( j =top; j > 0 && onxt < *o_src[j-1]; j--){
o_src[j] = o_src[j-1];
o_end[j] = o_end[j-1];
bats[j] = bats[j-1];
}
o_src[j] = (oid*) Tloc(b,BUNfirst(b));
o_end[j] = o_src[j] + BATcount(b);
bats[j] = b;
top++;
}
}
}
bn = BATnew(TYPE_void, TYPE_oid, cap, TRANSIENT);
if (bn == NULL){
GDKfree(bats);
GDKfree(o_src);
GDKfree(o_end);
throw(MAL, "mat.pack", MAL_MALLOC_FAIL);
}
if ( cap == 0){
BATseqbase(bn, 0);
BATseqbase(BATmirror(bn), 0);
BBPkeepref(*ret = bn->batCacheid);
GDKfree(bats);
GDKfree(o_src);
GDKfree(o_end);
return MAL_SUCCEED;
}
BATseqbase(bn, bats[0]->hseqbase);
// UNROLL THE MULTI-BAT MERGE
o = (oid*) Tloc(bn,BUNfirst(bn));
while( top){
*o++ = *o_src[0];
o_src[0]++;
if( o_src[0] == o_end[0]){
// remove this one
for(j=0; j< top; j++){
o_src[j]= o_src[j+1];
o_end[j]= o_end[j+1];
bats[j] = bats[j+1];
}
top--;
} else{
// resort priority queue
onxt= *o_src[0];
for( j=1; j< top && onxt > *o_src[j]; j++){
oo = o_src[j]; o_src[j]= o_src[j-1]; o_src[j-1]= oo;
oo = o_end[j]; o_end[j]= o_end[j-1]; o_end[j-1]= oo;
bm = bats[j]; bats[j]=bats[j-1]; bats[j-1] = bm;
}
}
}
for( i=0; i< top; i++)
BBPunfix(bats[i]->batCacheid);
BATsetcount(bn, (BUN) (o - (oid *) Tloc(bn, BUNfirst(bn))));
BATseqbase(bn, 0);
BATsettrivprop(bn);
GDKfree(bats);
GDKfree(o_src);
GDKfree(o_end);
/* properties */
bn->trevsorted = 0;
bn->tsorted = 1;
bn->tkey = 1;
bn->T->nil = 0;
bn->T->nonil = 1;
BBPkeepref(*ret = bn->batCacheid);
return MAL_SUCCEED;
}
str
BKCreuseBATmap(bat *ret, const bat *bid, const bat *did)
{
BAT *b, *d, *bn, *bs;
oid bidx, oidx = 0, *o, *ol;
oid *r;
gdk_return res;
if ((b = BATdescriptor(*bid)) == NULL) {
throw(MAL, "bat.shrinkMap", RUNTIME_OBJECT_MISSING);
}
if ( b->htype != TYPE_void) {
BBPunfix(b->batCacheid);
throw(MAL, "bat.shrinkMap", SEMANTIC_TYPE_MISMATCH);
}
if ((d = BATdescriptor(*did)) == NULL) {
BBPunfix(b->batCacheid);
throw(MAL, "bat.shrinkMap", RUNTIME_OBJECT_MISSING);
}
bn= BATnew(TYPE_void, TYPE_oid, BATcount(b) - BATcount(d), TRANSIENT);
if (bn == NULL) {
BBPunfix(b->batCacheid);
BBPunfix(d->batCacheid);
throw(MAL, "bat.shrinkMap", MAL_MALLOC_FAIL );
}
res = BATsubsort(&bs, NULL, NULL, d, NULL, NULL, 0, 0);
BBPunfix(d->batCacheid);
if (res != GDK_SUCCEED) {
BBPunfix(b->batCacheid);
BBPunfix(bn->batCacheid);
throw(MAL, "bat.shrinkMap", MAL_MALLOC_FAIL );
}
oidx = b->hseqbase;
bidx = oidx + BATcount(b)-1;
o = (oid*)Tloc(bs, BUNfirst(bs));
ol = (oid*)Tloc(bs, BUNlast(bs));
r = (oid*)Tloc(bn, BUNfirst(bn));
for (; oidx <= bidx; oidx++) {
if ( *o == oidx ){
while ( ol > o && ol[-1] == bidx) {
bidx--;
ol--;
}
*r++ = bidx;
o += (o < ol);
bidx--;
} else {
*r++ = oidx;
}
}
BATsetcount(bn, BATcount(b)-BATcount(bs));
BATseqbase(bn, b->hseqbase);
bn->tsorted = 0;
bn->trevsorted = 0;
bn->tdense = 0;
if (!(bn->batDirty&2)) BATsetaccess(bn, BAT_READ);
BBPunfix(b->batCacheid);
BBPunfix(bs->batCacheid);
BBPkeepref(*ret= bn->batCacheid);
return MAL_SUCCEED;
}
str
BKCreuseBAT(bat *ret, const bat *bid, const bat *did)
{
BAT *b, *d, *bn, *bs;
oid oidx = 0, bidx, *o, *ol;
gdk_return res;
if ((b = BATdescriptor(*bid)) == NULL) {
throw(MAL, "bat.reuse", RUNTIME_OBJECT_MISSING);
}
if ( b->htype != TYPE_void) {
BBPunfix(b->batCacheid);
throw(MAL, "bat.reuse", SEMANTIC_TYPE_MISMATCH);
}
if ((d = BATdescriptor(*did)) == NULL) {
BBPunfix(b->batCacheid);
throw(MAL, "bat.reuse", RUNTIME_OBJECT_MISSING);
}
bn= BATnew(b->htype, b->ttype, BATcount(b) - BATcount(d), TRANSIENT);
if (bn == NULL) {
BBPunfix(b->batCacheid);
BBPunfix(d->batCacheid);
throw(MAL, "bat.reuse", MAL_MALLOC_FAIL );
}
res = BATsubsort(&bs, NULL, NULL, d, NULL, NULL, 0, 0);
BBPunfix(d->batCacheid);
if (res != GDK_SUCCEED) {
BBPunfix(b->batCacheid);
BBPunfix(bn->batCacheid);
throw(MAL, "bat.reuse", MAL_MALLOC_FAIL );
}
oidx = b->hseqbase;
bidx = oidx + BATcount(b)-1;
o = (oid*)Tloc(bs, BUNfirst(bs));
ol= (oid*)Tloc(bs, BUNlast(bs));
switch(ATOMstorage(b->ttype) ){
case TYPE_bte: reuseloop(bte); break;
case TYPE_sht: reuseloop(sht); break;
case TYPE_int: reuseloop(int); break;
case TYPE_lng: reuseloop(lng); break;
#ifdef HAVE_HGE
case TYPE_hge: reuseloop(hge); break;
#endif
case TYPE_flt: reuseloop(flt); break;
case TYPE_dbl: reuseloop(dbl); break;
case TYPE_oid: reuseloop(oid); break;
case TYPE_str: /* to be done based on its index width */
default:
if (ATOMvarsized(bn->ttype)) {
BUN p = BUNfirst(b);
BUN q = BUNlast(b);
BATiter bi = bat_iterator(b);
for (;p<q; oidx++, p++) {
if ( *o == oidx ){
while ( ol > o && ol[-1] == bidx) {
bidx--;
q--;
ol--;
}
BUNappend(bn, BUNtail(bi, --q), FALSE);
o += (o < ol);
bidx--;
} else {
BUNappend(bn, BUNtail(bi, p), FALSE);
}
}
} else {
switch( b->T->width){
case 1:reuseloop(bte); break;
case 2:reuseloop(sht); break;
case 4:reuseloop(int); break;
case 8:reuseloop(lng); break;
#ifdef HAVE_HGE
case 16:reuseloop(hge); break;
#endif
default:
throw(MAL, "bat.shrink", "Illegal argument type");
}
}
}
BATsetcount(bn, BATcount(b) - BATcount(bs));
BATseqbase(bn, b->hseqbase);
bn->tsorted = 0;
bn->trevsorted = 0;
bn->tdense = 0;
bn->tkey = b->tkey;
if (!(bn->batDirty&2)) BATsetaccess(bn, BAT_READ);
BBPunfix(b->batCacheid);
BBPunfix(bs->batCacheid);
BBPkeepref(*ret= bn->batCacheid);
return MAL_SUCCEED;
}
