Changeset: e8e7cbe1108a for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=e8e7cbe1108a
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/sql_gencode.c
Branch: arrays
Log Message:
a dimension stores an array with all idxs that qualify in case they canot be
expressed with a range
+ subselect fixed to work with that
diffs (truncated from 812 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
@@ -50,6 +50,7 @@ gdk_dimension* createDimension_##TPE(TPE
dim->max = floor((max - min ) / step); \
dim->step = 1; \
dim->elsNum = dim->max +1; \
+ dim->idxs = NULL; \
return dim; \
}
@@ -88,8 +89,11 @@ gdk_array* arrayCopy(gdk_array *array) {
gdk_return arrayDelete(gdk_array *array) {
unsigned short i=0;
- for(i=0; i<array->dimsNum; i++)
+ for(i=0; i<array->dimsNum; i++) {
+ if(array->dims[i]->idxs)
+ GDKfree(array->dims[i]->idxs);
GDKfree(array->dims[i]);
+ }
GDKfree(array->dims);
GDKfree(array);
diff --git a/gdk/gdk_arrays.h b/gdk/gdk_arrays.h
--- a/gdk/gdk_arrays.h
+++ b/gdk/gdk_arrays.h
@@ -32,6 +32,7 @@ typedef struct dimensionStruct {
unsigned int max;
unsigned int step; //initialy this is set to 1
unsigned int elsNum;
+ unsigned int *idxs; //if it cannot be expressed as a dimension store
the idxs analytically
} gdk_dimension;
typedef struct arrayStruct {
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
@@ -17,6 +17,7 @@ static int arrayCellsNum(gdk_array *arra
return jumpSize(array, array->dimsNum);
}
+#if 0
/*UPDATED*/
static BUN oidToIdx(oid oidVal, int dimNum, int currentDimNum, BUN skipCells,
gdk_array *array) {
BUN oid = 0;
@@ -35,6 +36,7 @@ static BUN oidToIdx(oid oidVal, int dimN
return oid/skipCells;
return oid%skipCells;
}
+#endif
/*UPDATED*/
static BUN* oidToIdx_bulk(oid* oidVals, int valsNum, int dimNum, int
currentDimNum, BUN skipCells, gdk_array *array) {
@@ -77,303 +79,58 @@ static BUN* oidToIdx_bulk(oid* oidVals,
return oids;
}
-/*UPDATED*/
-static str readCands(gdk_array** dimCands_res, BAT** oidCands_res, const ptr
*dimCands, const bat *oidCands, gdk_array* array) {
- gdk_array *dimCands_in = NULL;
- BAT *candidatesBAT_in = NULL;
-
- if(oidCands) { //there are candidates that cannot be expressed as
dimensions and thus are expressed with oids
- if ((candidatesBAT_in = BATdescriptor(*oidCands)) == NULL) {
- throw(MAL, "algebra.subselect", RUNTIME_OBJECT_MISSING);
- }
- }
-
- if(dimCands) //there are candidates exressed as dimensions
- dimCands_in = (gdk_array*)*dimCands;
-
- //if there are no candidates then everything is a candidate
- if(!dimCands && !oidCands) {
- dimCands_in = arrayCopy(array);
- //create an empy candidates BAT
- if((candidatesBAT_in = BATnew(TYPE_void, TYPE_oid, 0,
TRANSIENT)) == NULL)
- throw(MAL, "algebra.subselect", GDK_EXCEPTION);
- BATsetcount(candidatesBAT_in, 0);
- BATseqbase(candidatesBAT_in, 0);
- BATderiveProps(candidatesBAT_in, FALSE);
- }
-
- *dimCands_res = dimCands_in;
- *oidCands_res = candidatesBAT_in;
-
- return MAL_SUCCEED;
+static gdk_dimension* updateDimCandRange(gdk_dimension *dimCand, unsigned int
min, unsigned int max) {
+ dimCand->min = min > dimCand->min ? min : dimCand->min;
+ dimCand->max = max < dimCand->max ? max : dimCand->max;
+ dimCand->elsNum = min>max ? 0 : max-min+1; //if 0 then the dimension is
empty
+ return dimCand;
}
-/*UPDATED*/
-/*Empty candidate results are distingueshed from first time results by setting
the dimCands to NULL and the oidCands to empty */
-static str emptyCandidateResults(ptr *dimCandsRes, bat* oidCandsRes) {
- BAT *candidatesBAT = NULL;
+static gdk_dimension* updateDimCandIdxs(gdk_dimension *dimCand, unsigned int
min, unsigned int max) {
+ unsigned int elsNum = 0;
+ unsigned int *idxs = GDKmalloc(sizeof(unsigned int)*dimCand->elsNum);
//at most (quailfyingIdx might alresy be out of the idxs)
+ unsigned int i;
- if((candidatesBAT = BATnew(TYPE_void, TYPE_oid, 0, TRANSIENT)) == NULL)
- throw(MAL, "algebra.dimensionSubselect", GDK_EXCEPTION);
- BATsetcount(candidatesBAT, 0);
- BATseqbase(candidatesBAT, 0);
- BATderiveProps(candidatesBAT, FALSE);
+ for(i=0 ; i<dimCand->elsNum; i++) {
+ if(dimCand->idxs[i] >= min && dimCand->idxs[i] <= max) {
+ idxs[elsNum] = dimCand->idxs[i];
+ elsNum++;
+ }
+ }
+
+ //release the previous idxs
+ GDKfree(dimCand->idxs);
+ //store the new idxs
+ dimCand->elsNum = elsNum; //if 0 then the dimension is empty
+ dimCand->idxs = idxs;
+ //upadte the min max if needed
+ dimCand->min = min > dimCand->min ? min : dimCand->min;
+ dimCand->max = max < dimCand->max ? max : dimCand->max;
- BBPkeepref(*oidCandsRes = candidatesBAT->batCacheid);
- *dimCandsRes = NULL;
-
- return MAL_SUCCEED;
+ return dimCand;
}
-/*UPDATED*/
-static gdk_dimension* updateCandidateDimensionRange(gdk_dimension *dim,
unsigned int min, unsigned int max) {
- if(dim->max < min || dim->min > max) //non-overlaping ranges
- return NULL;
-
- /*the biggest of the mins and the smallest of the maximums */
- dim->min = dim->min > min ? dim->min : min;
- dim->max = dim->max < max ? dim->max : max;
- dim->elsNum = floor((dim->max - dim->min)/dim->step)+1;
-//TODO: Take care of cases were a dimension has step <>1 as a result of
multiple selections on it
-
- //the dimensions that are merged should have the same order
- //they also have the same number of initial elements because they came
from the same dimension
- return dim;
-}
-
-static BAT* joinBATs(BAT *candsBAT, BAT* dimBAT, gdk_array *array, int dimNum)
{
- oid *candsOIDs, *dimOIDs, *mergedOIDs;
- BAT* mergedBAT;
-
- BUN i=0, j=0, k=0, minPos;
- BUN minIdx =0 ;
- BUN dimSkip = jumpSize(array, dimNum);
-
- oid dimOIDs_min = 0;
- bool set = 0;
- BUN moduloRes;
-
- candsOIDs = (oid*)Tloc(candsBAT, BUNfirst(candsBAT));
- dimOIDs = (oid*)Tloc(dimBAT, BUNfirst(dimBAT));
-
- //the oids in dimBAT have been computed assuming that 0 is allowed in
all other dimensions
- //is this really true? I can verify that using the first oid in candsBAT
- //if a dimension after filtering is expressed again as a dimension the
min might
- //not be 0 but I do not care about it when it comes to the BAT. I will
resolve tha at the end
- //when projecting the cells where dimensions and BAT will be combined
- for(j=0; j< (unsigned long)dimNum; j++) {
- BUN skipCells = 0;
- //find the min oid of this dimension in the the first
qualifying oid
- minIdx = oidToIdx(candsOIDs[0], j+1, 0, 1, array);
- if(minIdx == 0)
- continue;
- //all oids in the dimOIDs should be updated to comply with the
min oid of the dimension
- skipCells = jumpSize(array, j);
- for(i=0; i<BATcount(dimBAT); i++)
- dimOIDs[i] += skipCells*minIdx;
- }
-
- if(!(mergedBAT = BATnew(TYPE_void, TYPE_oid,
BATcount(candsBAT)+BATcount(dimBAT), TRANSIENT)))
- return NULL;
- mergedOIDs = (oid*)Tloc(mergedBAT, BUNfirst(mergedBAT));
-
- moduloRes = dimOIDs[0]%dimSkip;
- /* find the oids in cands that are there to reflect dimNum and keep
only those that dim and cand have in common */
- for(i=0, j=0; i<BATcount(candsBAT) && j<BATcount(dimBAT);) {
- /* oids in this dimension should be multiples of dimSkip */
- if(candsOIDs[i]%dimSkip == moduloRes) {
- if(candsOIDs[i] < dimOIDs[j]) //it exists in one but
not in the other
- i++;
- else if (candsOIDs[i] > dimOIDs[j])
- j++;
- else { //common
- mergedOIDs[k] = candsOIDs[i];
-
- if(!set) {
- dimOIDs_min = candsOIDs[i];
- minPos = k;
- set = 1;
- }
-
- i++;
- j++;
- k++;
- }
- } else {
- /*not related with the dimension. send it to the
output*/
- mergedOIDs[k] = candsOIDs[i];
- i++;
- k++;
- }
- }
-
- BATseqbase(mergedBAT, 0);
- BATsetcount(mergedBAT, k);
- BATderiveProps(mergedBAT, FALSE);
-
- //adapt the candidates BAT to reflect the minimum value for the new
dimension
- //only the ones that are not reflecting the y dimension should be
updates
- minIdx = oidToIdx(dimOIDs_min, dimNum, 0, 1, array);
- if(minIdx > 0) {
- BUN skipCells = jumpSize(array, dimNum);
- /*split it in 2 parts. Firts update all oids that are before
the minPos position
- * all those oids will be increased and become greater than the
oid in minPos */
- if(minPos > 0) {
- for(i=minPos-1; i>0; i--) {
- /*all will change because they are above the
minimum oid regarding the dimension*/
- mergedOIDs[i+1] =
mergedOIDs[i]+skipCells*minIdx;
- }
- /* excluded from the loop because i>=0 always true
(infinite loop)*/
- mergedOIDs[1] = mergedOIDs[0]+skipCells*minIdx;
- }
- mergedOIDs[0] = dimOIDs_min;
- for(i=minPos+1; i<BATcount(mergedBAT) ; i++)
- if(mergedOIDs[i]%dimSkip != moduloRes)
- mergedOIDs[i] += skipCells*minIdx;
- }
-
- return mergedBAT;
-}
-
-static BAT* mergeBATs(BAT *candsBAT, BAT* dimBAT, gdk_array *array, int
dimNum) {
- oid *candsOIDs, *dimOIDs, *mergedOIDs;
- BAT* mergedBAT;
-
- BUN i=0, j=0, k=0;
- BUN minIdx =0 ;
-
- candsOIDs = (oid*)Tloc(candsBAT, BUNfirst(candsBAT));
- dimOIDs = (oid*)Tloc(dimBAT, BUNfirst(dimBAT));
-
- //the oids in dimBAT have been computed assuming that 0 is allowed in
all other dimensions
- //is this really true? I can verify that using the first oid in candsBAT
- //if a dimension after filtering is expressed again as a dimension the
min might
- //not be 0 but I do not care about it when it comes to the BAT. I will
resolve tha at the end
- //when projecting the cells where dimensions and BAT will be combined
- for(j=0; j< (unsigned long)dimNum; j++) {
- BUN skipCells = 0;
- //find the min oid of this dimension in the the first
qualifying oid
- minIdx = oidToIdx(candsOIDs[0], j+1, 0, 1, array);
- if(minIdx == 0)
- continue;
- //all oids in the dimOIDs should be updated to comply with the
min oid of the dimension
- skipCells = jumpSize(array, j+1);
- for(i=0; i<BATcount(dimBAT); i++)
- dimOIDs[i] += skipCells*minIdx;
- }
- //adapt the candidates BAT to reflect the minimum value for the new
dimension
- minIdx = oidToIdx(dimOIDs[0], dimNum, 0, 1, array);
- if(minIdx > 0) {
- BUN skipCells = jumpSize(array, dimNum);
- for(i=0; i<BATcount(candsBAT); i++)
- candsOIDs[i] += skipCells*minIdx;
- }
-
- //finaly merge the two BATs
- if(!(mergedBAT = BATnew(TYPE_void, TYPE_oid,
BATcount(candsBAT)+BATcount(dimBAT), TRANSIENT)))
- return NULL;
- mergedOIDs = (oid*)Tloc(mergedBAT, BUNfirst(mergedBAT));
- for(i=0, j=0; i<BATcount(candsBAT) && j<BATcount(dimBAT);) {
- if(candsOIDs[i] < dimOIDs[j]) {
- mergedOIDs[k] = candsOIDs[i];
- i++;
- } else if(candsOIDs[i] > dimOIDs[j]) {
- mergedOIDs[k] = dimOIDs[j];
- j++;
- } else {
- mergedOIDs[k] = candsOIDs[i];
- i++;
- j++;
- }
- k++;
-
- if(i == BATcount(candsBAT)) {
- for(; j<BATcount(dimBAT); j++, k++)
- mergedOIDs[k] = dimOIDs[j];
- }
-
- if(j == BATcount(dimBAT)) {
- for(; i<BATcount(dimBAT); i++, k++)
_______________________________________________
checkin-list mailing list
checkin-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/checkin-list
