Changeset: f4aa4ce5316d for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=f4aa4ce5316d
Modified Files:
gdk/gdk_arrays.c
gdk/gdk_arrays.h
monetdb5/modules/kernel/arrays.c
monetdb5/modules/kernel/arrays.h
monetdb5/modules/kernel/arrays.mal
sql/backends/monet5/rel_bin.c
sql/backends/monet5/sql_gencode.c
Branch: arrays
Log Message:
joining array dimensions with condition a.dim1=b.dim1
diffs (truncated from 701 to 300 lines):
diff --git a/gdk/gdk_arrays.c b/gdk/gdk_arrays.c
--- a/gdk/gdk_arrays.c
+++ b/gdk/gdk_arrays.c
@@ -63,6 +63,26 @@ createDim(lng);
createDim(dbl);
createDim(flt);
+#define createGCD(TPE) \
+TPE gcd_##TPE(TPE val1, TPE val2) { \
+ TPE res; \
+ if(val1 == val2) \
+ return val1; \
+ res=val1-val2; \
+ if(res < val2) \
+ return gcd_##TPE(val2,res); \
+ return gcd_##TPE(res,val2); \
+} \
+
+createGCD(bte);
+createGCD(sht);
+createGCD(int);
+createGCD(wrd);
+createGCD(oid);
+createGCD(lng);
+createGCD(dbl);
+createGCD(flt);
+
gdk_array* arrayNew(unsigned short dimsNum) {
gdk_array *array = (gdk_array*)GDKmalloc(sizeof(gdk_array));
diff --git a/gdk/gdk_arrays.h b/gdk/gdk_arrays.h
--- a/gdk/gdk_arrays.h
+++ b/gdk/gdk_arrays.h
@@ -59,6 +59,15 @@ gdk_export gdk_return arrayDelete(gdk_ar
gdk_export gdk_return analyticDimensionDelete(gdk_analytic_dimension *dim);
gdk_export gdk_array* arrayCopy(gdk_array* array);
+gdk_export bte gcd_bte(bte val1, bte val2);
+gdk_export sht gcd_sht(sht val1, sht val2);
+gdk_export int gcd_int(int val1, int val2);
+gdk_export wrd gcd_wrd(wrd val1, wrd val2);
+gdk_export oid gcd_oid(oid val1, oid val2);
+gdk_export lng gcd_lng(lng val1, lng val2);
+gdk_export dbl gcd_dbl(dbl val1, dbl val2);
+gdk_export flt gcd_flt(flt val1, flt val2);
+
/*find the position in the dimension indices (no repetitions) of the given
value*/
#define dimensionFndValuePos(value, min, step) fmod((value-min), step)?
BUN_NONE : (BUN)(value-min)/step
diff --git a/monetdb5/modules/kernel/arrays.c b/monetdb5/modules/kernel/arrays.c
--- a/monetdb5/modules/kernel/arrays.c
+++ b/monetdb5/modules/kernel/arrays.c
@@ -507,153 +507,58 @@ str ALGdimensionLeftfetchjoin2(bat *resu
return ALGdimensionLeftfetchjoin1(result, dimsCands, dim, dims);
}
-str ALGdimensionLeftfetchjoin3(bat *result, const ptr* dimsCands, const bat*
oidsCands, const ptr *array_in) {
+str ALGdimensionLeftfetchjoin3(bat *result, const ptr* dimsCands, const ptr
*array_in) {
gdk_array *array = (gdk_array*)*array_in;
gdk_array *dims_in = (gdk_array*)*dimsCands;
- BAT *oidsCandsBAT = BATdescriptor(*oidsCands);
BAT *resBAT;
oid *resOids;
- BUN elsNum = 1;
- if(!oidsCandsBAT) {
+ /*count the cells in the output*/
+ BUN elsNum = arrayCellsNum(dims_in);
+ /*create all oids*/
+ if(!(resBAT = BATnew(TYPE_void, TYPE_oid, elsNum, TRANSIENT))) {
arrayDelete(dims_in);
- throw(MAL, "algebra.dimensionLeftfetchjoin2",
RUNTIME_OBJECT_MISSING);
+ throw(MAL, "algebra.dimensionLeftfetchjoin2","Problem
allocating new BAT");
}
- /*this function is called in case of an update after a subselection
- * on the dimensions. In such a case only the dimsCands should have
- * values and the oidsCands should be empty*/
- /*
- * the above is not true. When having update a set v =1 where x=y
- * then the oidsCandsBAT is not empty
- if(BATcount(oidsCandsBAT)) {
- arrayDelete(dims_in);
- throw(MAL,"algebra.dimensionLeftfetchjoin2", "oidsCands is not
empty");
- }*/
+ /* create the oids that should be updated based on the dimsCands */
+ resOids = (oid*)Tloc(resBAT, BUNfirst(resBAT));
+ /* fill the array with the elements in the first dimension */
+ computeOids(&resOids, 0, array->dimsNum-1 , array, dims_in,
arrayCellsNum(array));
- elsNum = BATcount(oidsCandsBAT);
- if(!elsNum) { //there are no oids update - get oids from dimsCands
- /*count the cells in the output*/
- elsNum = arrayCellsNum(dims_in);
- /*create all oids*/
- if(!(resBAT = BATnew(TYPE_void, TYPE_oid, elsNum, TRANSIENT))) {
- arrayDelete(dims_in);
- throw(MAL, "algebra.dimensionLeftfetchjoin2","Problem
allocating new BAT");
- }
+ BATsetcount(resBAT, elsNum);
+ BATseqbase(resBAT, 0);
+ resBAT->tsorted = 1;
+ resBAT->trevsorted = 0;
+ resBAT->tkey = 1;
+ resBAT->tdense = 0;
- /* create the oids that should be updated based on the
dimsCands */
- resOids = (oid*)Tloc(resBAT, BUNfirst(resBAT));
- /* fill the array with the elements in the first dimension */
- computeOids(&resOids, 0, array->dimsNum-1 , array, dims_in,
arrayCellsNum(array));
-
- BATsetcount(resBAT, elsNum);
- BATseqbase(resBAT, 0);
- resBAT->tsorted = 1;
- resBAT->trevsorted = 0;
- resBAT->tkey = 1;
- resBAT->tdense = 0;
-
- BBPkeepref(*result = resBAT->batCacheid);
- BBPunfix(oidsCandsBAT->batCacheid);
- } else {
- //there are oids, use only them
- BBPkeepref(*result = oidsCandsBAT->batCacheid);
- }
+ BBPkeepref(*result = resBAT->batCacheid);
return MAL_SUCCEED;
}
-//str ALGnonDimensionLeftfetchjoin1(bat* result, const ptr *dimsCands, const
bat *oidsCands, const bat *vals, const ptr *dims) {
str ALGnonDimensionLeftfetchjoin1(bat* result, const bat *mbrOids, const bat
*vals, const ptr *dims) {
/* projecting a non-dimensional column does not differ from projecting
any relational column */
- BAT *mbrCandsBAT = NULL/*, *oidsCandsBAT = NULL*/, *valsBAT = NULL,
*resBAT= NULL;
-// gdk_array *dimCands_in = (gdk_array*)*dimsCands;
-// gdk_array *array = (gdk_array*)*dims;
-
-#if 0
- if(!dimCands_in) { //empty
- if(!(resBAT = newempty("nonDimensionLeftfetchjoin1")))
- throw(MAL, "algebra.leftfetchjoin", "Problem allocating new BAT");
- BBPkeepref(*result = resBAT->batCacheid);
-
- return MAL_SUCCEED;
- }
-#endif
+ BAT *mbrCandsBAT = NULL, *valsBAT = NULL, *resBAT= NULL;
(void)*dims;
-// if ((oidsCandsBAT = BATdescriptor(*oidsCands)) == NULL) {
-// throw(MAL, "algebra.leftfetchjoin", RUNTIME_OBJECT_MISSING);
-// }
-
if ((valsBAT = BATdescriptor(*vals)) == NULL) {
-// BBPunfix(oidsCandsBAT->batCacheid);
throw(MAL, "algebra.leftfetchjoin", RUNTIME_OBJECT_MISSING);
}
if ((mbrCandsBAT = BATdescriptor(*mbrOids)) == NULL) {
-// BBPunfix(oidsCandsBAT->batCacheid);
BBPunfix(valsBAT->batCacheid);
throw(MAL, "algebra.leftfetchjoin", RUNTIME_OBJECT_MISSING);
}
- //create the oids using the candidates
-// mbrCandsBAT = projectCells(dimCands_in, array);
-
-#if 0
- if(BATcount(oidsCandsBAT)) { /*there is mbr and oids. Some values need
to be set to null */
- BAT *r1p, *r2p, *r3p;
-
- //left outer join between the mbrCands and the oidsCands
- if(BATsubouterjoin(&r1p, &r2p, mbrCandsBAT, oidsCandsBAT, NULL,
NULL, 0 /*null never match*/, BATcount(mbrCandsBAT) /*the size of the resuls.
BUN_NONE will cause it to be computed*/) != GDK_SUCCEED) {
- BBPunfix(oidsCandsBAT->batCacheid);
- BBPunfix(valsBAT->batCacheid);
- BBPunfix(mbrCandsBAT->batCacheid);
- }
- r3p = BATproject(r2p, oidsCandsBAT);
- resBAT = BATproject(r3p, valsBAT);
-
- BBPunfix(r1p->batCacheid);
- BBPunfix(r2p->batCacheid);
- BBPunfix(r3p->batCacheid);
- } else
-#endif
- resBAT = BATproject(mbrCandsBAT, valsBAT);
-
-#if 0
-
- resBAT = BATproject(mbrCandsBAT, valsBAT);
-
- if(BATcount(resBAT)) { //if there are no oidsCands then there is no MBR
either and the result will be empty
- /*iterate over mbrCandsBAT and oidsCandsBAT and set NULL to all
- * values in resBAT that are found in mbrCandsBAT but not in
oidsCandsBAT */
- BATiter mbrIter = bat_iterator(mbrCandsBAT);
- BATiter oidsIter = bat_iterator(oidsCandsBAT);
-
- void *nilPtr = ATOMnilptr(BATttype(valsBAT));
-
- BUN mbrCurrBun = BUNfirst(mbrCandsBAT);
- BUN mbrLastBun = BUNlast(mbrCandsBAT);
- BUN oidsCurrBun = BUNfirst(oidsCandsBAT);
- BUN oidsLastBun = BUNlast(oidsCandsBAT);
- for(; mbrCurrBun < mbrLastBun && oidsCurrBun < oidsLastBun ;
mbrCurrBun++, oidsCurrBun++) {
- oid mbrOid = *(oid*)BUNtail(mbrIter, mbrCurrBun);
- oid oidsOid = *(oid*)BUNtail(oidsIter, oidsCurrBun);
-
- while(mbrOid != oidsOid && mbrCurrBun < mbrLastBun) {
- BUNreplace(resBAT, &mbrCurrBun, nilPtr, FALSE);
//replace with NULL
- mbrCurrBun++;
- mbrOid = *(oid*)BUNtail(mbrIter, mbrCurrBun);
- }
- }
- }
-#endif
+ resBAT = BATproject(mbrCandsBAT, valsBAT);
BBPunfix(mbrCandsBAT->batCacheid);
BBPunfix(valsBAT->batCacheid);
-// BBPunfix(oidsCandsBAT->batCacheid);
if (resBAT == NULL)
throw(MAL, "algebra.leftfetchjoin", GDK_EXCEPTION);
@@ -685,7 +590,6 @@ str ALGnonDimensionLeftfetchjoin2(bat* r
throw(MAL, "algebra.leftfetchjoin", "Problem materialising
non-dimensional column");
}
-
/*append the missing values to the BAT */
BATappend(nonDimensionalBAT, materialisedBAT, TRUE);
BATsetcount(nonDimensionalBAT, totalCellsNum);
@@ -1207,6 +1111,103 @@ str ALGouterjoin(bat *res, const bat *l,
return MAL_SUCCEED;
}
+str ALGsubjoin1(ptr *dimsResL, ptr *dimsResR, const ptr *dimensionL, const ptr
*dimsL, const ptr *dimensionR, const ptr *dimsR) {
+ gdk_analytic_dimension *analytic_dimensionL =
(gdk_analytic_dimension*)*dimensionL;
+ gdk_analytic_dimension *analytic_dimensionR =
(gdk_analytic_dimension*)*dimensionR;
+ gdk_dimension *dimL, *dimR;
+
+ *dimsResL = *dimsL;
+ *dimsResR = *dimsR;
+
+ dimL = ((gdk_array*)*dimsResL)->dims[analytic_dimensionL->dimNum];
+ dimR = ((gdk_array*)*dimsResR)->dims[analytic_dimensionR->dimNum];
+
+#define compareRanges(TPE) \
+do { \
+ TPE minL = *(TPE*)analytic_dimensionL->min; \
+ TPE maxL = *(TPE*)analytic_dimensionL->max; \
+ TPE stepL = *(TPE*)analytic_dimensionL->step; \
+ TPE minR = *(TPE*)analytic_dimensionR->min; \
+ TPE maxR = *(TPE*)analytic_dimensionR->max; \
+ TPE stepR = *(TPE*)analytic_dimensionR->step; \
+\
+ TPE d = gcd_##TPE(stepR, stepL); \
+ if(fmod((minL-minR),d)) {\
+ dimR->min = dimL->min = dimR->max = dimL->max = 0; \
+ dimR->step = dimL->step = 1; \
+ } else { \
+ /*find the min, max and step of the dimensions*/ \
+ TPE l = minL, r=minR; \
+ dimR->step = stepL/d; \
+ dimL->step = stepR/d;\
+ while(l != r) { \
+ while(l<r) \
+ l+=stepL; \
+ while(r<l) \
+ r+=stepR; \
+ } \
+ /*l = r*/ \
+ dimL->min = (l-minL)/stepL; \
+ dimR->min = (r-minR)/stepR; \
+\
+ l = maxL, r=maxR; \
+ while(l != r) { \
+ while(l>r) \
+ l-=stepL; \
+ while(r>l) \
+ r-=stepR; \
+ } \
+ /*l = r*/ \
+ dimL->max = (l-minL)/stepL; \
+ dimR->max = (r-minR)/stepR; \
+ } \
+} while(0)
+
+ switch(analytic_dimensionL->type) {
+ case TYPE_bte:
+ compareRanges(bte);
+ break;
+ case TYPE_sht:
+ compareRanges(sht);
+ break;
+ case TYPE_int:
+ compareRanges(int);
+ break;
+ case TYPE_wrd:
_______________________________________________
checkin-list mailing list
checkin-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/checkin-list
