Changeset: 8a73b460fa18 for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=8a73b460fa18
Modified Files:
gdk/gdk_arrays.c
gdk/gdk_arrays.h
monetdb5/modules/kernel/arrays.c
Branch: arrays
Log Message:
changing code to use new structure + leftfecth join takes dimCands and oidCands
and should handle them itself
diffs (truncated from 319 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
@@ -222,6 +222,36 @@ BAT* materialise_nonDimensional_column(i
return b;
}
+BAT *projectCells(gdk_array* dims, BAT* oidsBAT) {
+ BAT *resBAT = NULL;
+ BUN resSize = 1;
+ oid *resOIDs = NULL;
+ dim_node *n;
+
+ /*combine the oidsDimensions in order to get the global oids (the cells)*/
+ for(n=dims->h; n; n=n->next) {
+ BUN sz = n->data->elementsNum;
+ if(sz > 0)
+ resSize *= sz;
+ else
+ resSize *= n->data->initialElementsNum;
+ }
+ resSize += BATcount(oidsBAT); //this is not accurate but I believe it is ok
+ //fprintf(stderr, "estiamted size = %u\n", (unsigned int)resSize);
+ /*the size of the result is the same as the number of cells in the
candidatesDimensions */
+ if(!(resBAT = BATnew(TYPE_void, TYPE_oid, resSize, TRANSIENT)))
+ return NULL;
+ resOIDs = (oid*)Tloc(resBAT, BUNfirst(resBAT));
+ resSize = qualifyingOIDs(0, 1, dims, oidsBAT, &resOIDs);
+ //fprintf(stderr, "real size = %u\n", (unsigned int)resSize);
+ BATsetcount(resBAT, resSize);
+ BATseqbase(resBAT, 0);
+ BATderiveProps(resBAT, FALSE);
+
+ return resBAT;
+}
+
+
#if 0
gdk_cells* cells_new(void) {
gdk_cells *cells = GDKmalloc(sizeof(gdk_cells));
@@ -1930,35 +1960,6 @@ static BUN qualifyingOIDs(int dimNum, in
return sz;
}
-BAT *projectCells(gdk_cells* dims, BAT* oidsBAT) {
- BAT *resBAT = NULL;
- BUN resSize = 1;
- oid *resOIDs = NULL;
- dim_node *n;
-
- /*combine the oidsDimensions in order to get the global oids (the cells)*/
- for(n=dims->h; n; n=n->next) {
- BUN sz = n->data->elementsNum;
- if(sz > 0)
- resSize *= sz;
- else
- resSize *= n->data->initialElementsNum;
- }
- resSize += BATcount(oidsBAT); //this is not accurate but I believe it is ok
- //fprintf(stderr, "estiamted size = %u\n", (unsigned int)resSize);
- /*the size of the result is the same as the number of cells in the
candidatesDimensions */
- if(!(resBAT = BATnew(TYPE_void, TYPE_oid, resSize, TRANSIENT)))
- return NULL;
- resOIDs = (oid*)Tloc(resBAT, BUNfirst(resBAT));
- resSize = qualifyingOIDs(0, 1, dims, oidsBAT, &resOIDs);
- //fprintf(stderr, "real size = %u\n", (unsigned int)resSize);
- BATsetcount(resBAT, resSize);
- BATseqbase(resBAT, 0);
- BATderiveProps(resBAT, FALSE);
-
- return resBAT;
-}
-
/*takes a set of dimensions of any type and returns the corresponding
dimensions in indices format */
gdk_cells* arrayToCells(gdk_array *array) {
diff --git a/gdk/gdk_arrays.h b/gdk/gdk_arrays.h
--- a/gdk/gdk_arrays.h
+++ b/gdk/gdk_arrays.h
@@ -322,8 +322,8 @@ gdk_return dimensionBATsubjoin(BAT **out
#endif
/*NEW*/
+gdk_export BAT *projectCells(gdk_array* dims, BAT* oidsBAT);
#if 0
-gdk_export BAT *projectCells(gdk_cells* dims, BAT* oidsBAT);
gdk_export gdk_cells* arrayToCells(gdk_array *array);
gdk_export gdk_array *cellsToArray(gdk_cells *cells);
#endif
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
@@ -17,25 +17,27 @@ static int arrayCellsNum(gdk_array *arra
return jumpSize(array, array->dimsNum);
}
-static BUN oidToIdx(oid oidVal, int dimNum, int currentDimNum, BUN skipCells,
gdk_array *dims) {
+/*UPDATED*/
+static BUN oidToIdx(oid oidVal, int dimNum, int currentDimNum, BUN skipCells,
gdk_array *array) {
BUN oid = 0;
while(currentDimNum < dimNum) {
- skipCells*=dims->dimSizes[currentDimNum];
+ skipCells*=array->dims[currentDimNum]->elsNum;
currentDimNum++;
}
- if(currentDimNum == dims->dimsNum-1)
+ if(currentDimNum == array->dimsNum-1)
oid = oidVal;
else
- oid = oidToIdx(oidVal, dimNum, currentDimNum+1,
skipCells*dims->dimSizes[currentDimNum], dims);
+ oid = oidToIdx(oidVal, dimNum, currentDimNum+1,
skipCells*array->dims[currentDimNum]->elsNum, array);
if(currentDimNum == dimNum) //in the last one we do not compute module
return oid/skipCells;
return oid%skipCells;
}
-static BUN* oidToIdx_bulk(oid* oidVals, int valsNum, int dimNum, int
currentDimNum, BUN skipCells, gdk_array *dims) {
+/*UPDATED*/
+static BUN* oidToIdx_bulk(oid* oidVals, int valsNum, int dimNum, int
currentDimNum, BUN skipCells, gdk_array *array) {
BUN *oids = GDKmalloc(valsNum*sizeof(BUN));
int i;
@@ -45,11 +47,11 @@ static BUN* oidToIdx_bulk(oid* oidVals,
}
while(currentDimNum < dimNum) {
- skipCells*=dims->dimSizes[currentDimNum];
+ skipCells*=array->dims[currentDimNum]->elsNum;
currentDimNum++;
}
- if(currentDimNum == dims->dimsNum-1) { //last dimension, do not go any
deeper
+ if(currentDimNum == array->dimsNum-1) { //last dimension, do not go any
deeper
if(currentDimNum == dimNum) {//in the dimension of interest we
do not compute the module
for(i=0; i<valsNum; i++)
oids[i] = oidVals[i]/skipCells;
@@ -59,7 +61,7 @@ static BUN* oidToIdx_bulk(oid* oidVals,
}
}
else {
- BUN *oidRes = oidToIdx_bulk(oidVals, valsNum, dimNum,
currentDimNum+1, skipCells*dims->dimSizes[currentDimNum], dims);
+ BUN *oidRes = oidToIdx_bulk(oidVals, valsNum, dimNum,
currentDimNum+1, skipCells*array->dims[currentDimNum]->elsNum, array);
if(currentDimNum == dimNum) {//in the dimension of interest we
do not compute the module
for(i=0; i<valsNum; i++)
@@ -353,7 +355,7 @@ static bool updateCandidateResults(gdk_a
return 1;
}
-str ALGdimensionSubselect2(ptr *dimsRes, bat* oidsRes, const ptr *dim, const
ptr* dims, const ptr *dimsCand, const bat* oidsCand,
+str ALGdimensionSubselect2(ptr *dimsRes, bat* oidsRes, const ptr *dim, const
ptr* dims, const ptr *dimCands, const bat* oidCands,
const void *low, const
void *high, const bit *li, const bit *hi, const bit *anti) {
gdk_array *array = (gdk_array*)*dims;
gdk_analytic_dimension *dimension = (gdk_analytic_dimension*)*dim;
@@ -364,7 +366,7 @@ str ALGdimensionSubselect2(ptr *dimsRes,
int type;
const void* nil;
- readCands(&dimCands_in, &candidatesBAT_in, dimsCand, oidsCand, array);
+ readCands(&dimCands_in, &candidatesBAT_in, dimCands, oidCands, array);
if(!dimCands_in) { //empty results
if(candidatesBAT_in)
@@ -512,7 +514,7 @@ str ALGdimensionSubselect2(ptr *dimsRes,
return emptyCandidateResults(dimsRes, oidsRes);
}
- if(oidsCand && candidatesBAT_in != candidatesBAT_out) //there was a
candidatesBAT in the input that is not sent in the output
+ if(oidCands && candidatesBAT_in != candidatesBAT_out) //there was a
candidatesBAT in the input that is not sent in the output
BBPunfix(candidatesBAT_in->batCacheid);
BBPkeepref(*oidsRes = candidatesBAT_out->batCacheid);
@@ -526,7 +528,7 @@ str ALGdimensionSubselect1(ptr *dimsRes,
return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims, NULL, NULL,
low, high, li, hi, anti);
}
-str ALGdimensionThetasubselect2(ptr *dimsRes, bat* oidsRes, const ptr *dim,
const ptr* dims, const ptr *dimsCand, const bat* oidsCand, const void *val,
const char **opp) {
+str ALGdimensionThetasubselect2(ptr *dimsRes, bat* oidsRes, const ptr *dim,
const ptr* dims, const ptr *dimCands, const bat* oidCands, const void *val,
const char **opp) {
bit li = 0;
bit hi = 0;
bit anti = 0;
@@ -538,42 +540,42 @@ str ALGdimensionThetasubselect2(ptr *dim
/* "=" or "==" */
li = hi = 1;
anti = 0;
- return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims, dimsCand,
oidsCand, val, nil, &li, &hi, &anti);
+ return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims, dimCands,
oidCands, val, nil, &li, &hi, &anti);
}
if (op[0] == '!' && op[1] == '=' && op[2] == 0) {
/* "!=" (equivalent to "<>") */
li = hi = anti = 1;
- return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims, dimsCand,
oidsCand, val, nil, &li, &hi, &anti);
+ return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims, dimCands,
oidCands, val, nil, &li, &hi, &anti);
}
if (op[0] == '<') {
if (op[1] == 0) {
/* "<" */
li = hi = anti = 0;
- return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims,
dimsCand, oidsCand, nil, val, &li, &hi, &anti);
+ return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims,
dimCands, oidCands, nil, val, &li, &hi, &anti);
}
if (op[1] == '=' && op[2] == 0) {
/* "<=" */
li = anti = 0;
hi = 1;
- return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims,
dimsCand, oidsCand, nil, val, &li, &hi, &anti);
+ return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims,
dimCands, oidCands, nil, val, &li, &hi, &anti);
}
if (op[1] == '>' && op[2] == 0) {
/* "<>" (equivalent to "!=") */
li = hi = anti = 1;
- return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims,
dimsCand, oidsCand, val, nil, &li, &hi, &anti);
+ return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims,
dimCands, oidCands, val, nil, &li, &hi, &anti);
}
}
if (op[0] == '>') {
if (op[1] == 0) {
/* ">" */
li = hi = anti = 0;
- return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims,
dimsCand, oidsCand, val, nil, &li, &hi, &anti);
+ return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims,
dimCands, oidCands, val, nil, &li, &hi, &anti);
}
if (op[1] == '=' && op[2] == 0) {
/* ">=" */
li = 1;
hi = anti = 0;
- return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims,
dimsCand, oidsCand, val, nil, &li, &hi, &anti);
+ return ALGdimensionSubselect2(dimsRes, oidsRes, dim, dims,
dimCands, oidCands, val, nil, &li, &hi, &anti);
}
}
@@ -584,11 +586,11 @@ str ALGdimensionThetasubselect1(ptr *dim
return ALGdimensionThetasubselect2(dimsRes, oidsRes, dim, dims, NULL,
NULL, val, op);
}
-str ALGdimensionLeftfetchjoin1(bat *result, const ptr *dimsCands, const bat
*batCands, const ptr *dim, const ptr *dims) {
+str ALGdimensionLeftfetchjoin1(bat *result, const ptr *dimCands, const bat
*oidCands, const ptr *dim, const ptr *dims) {
gdk_array *array = (gdk_array*)*dims;
gdk_analytic_dimension *dimension = (gdk_analytic_dimension*)*dim;
- gdk_cells *candidateDimensions = *dimCands;
- BAT *candsBAT = NULL, *resBAT = NULL;
+ gdk_array *dimCands_in = (gdk_array*)*dimCands;
+ BAT *oidCandsBAT = NULL, *candsBAT = NULL, *resBAT = NULL;
BUN resSize = 0;
#define computeValues(TPE) \
@@ -612,10 +614,15 @@ do { \
} while(0)
- if ((candsBAT = BATdescriptor(*batCands)) == NULL) {
- throw(MAL, "algebra.dimensionLeftfetchjoin", RUNTIME_OBJECT_MISSING);
+ if ((oidCandsBAT = BATdescriptor(*oidCands)) == NULL) {
+ throw(MAL, "algebra.leftfetchjoin", RUNTIME_OBJECT_MISSING);
}
+
+//TODO: Find a more clever way to do this without the need to project the cells
+ //create the oids using the candidates
+ candsBAT = projectCells(dimCands_in, oidCandsBAT);
resSize = BATcount(candsBAT);
+
/*for each oid in the candsBAT find the real value of the dimension */
switch(dimension->type) { \
case TYPE_bte: \
@@ -656,17 +663,19 @@ do { \
/*should I release space or MonetDB does it for the input?*/
analyticDimensionDelete(dimension);
arrayDelete(array);
+ BBPunfix(candsBAT->batCacheid);
+ BBPunfix(oidCandsBAT->batCacheid);
return MAL_SUCCEED;
}
-str ALGnonDimensionLeftfetchjoin1(bat* result, const ptr *dimsCands, const bat
*batCands, const bat *vals, const ptr *dims) {
+str ALGnonDimensionLeftfetchjoin1(bat* result, const ptr *dimCands, const bat
*oidCands, const bat *vals, const ptr *dims) {
/* projecting a non-dimensional column does not differ from projecting
any relational column */
- BAT *candsBAT, *valsBAT, *resBAT= NULL;
-
+ BAT *oidCandsBAT, *candsBAT, *valsBAT, *resBAT= NULL;
+ gdk_array *dimCands_in = (gdk_array*)*dimCands;
(void)*dims;
- if ((candsBAT = BATdescriptor(*cands)) == NULL) {
+ if ((oidCandsBAT = BATdescriptor(*oidCands)) == NULL) {
throw(MAL, "algebra.leftfetchjoin", RUNTIME_OBJECT_MISSING);
}
if ((valsBAT = BATdescriptor(*vals)) == NULL) {
@@ -674,6 +683,9 @@ str ALGnonDimensionLeftfetchjoin1(bat* r
throw(MAL, "algebra.leftfetchjoin", RUNTIME_OBJECT_MISSING);
}
+ //create the oids using the candidates
+ candsBAT = projectCells(dimCands_in, oidCandsBAT);
+
resBAT = BATproject(candsBAT, valsBAT);
_______________________________________________
checkin-list mailing list
checkin-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/checkin-list
