Changeset: 8c5e3968b78a for MonetDB
URL: http://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=8c5e3968b78a
Modified Files:
geom/monetdb5/geom.mal
geom/monetdb5/geomPoints.c
monetdb5/optimizer/opt_geospatial.c
Branch: geo
Log Message:
geospatial optimiser handles conjunctions of spatial relations
diffs (266 lines):
diff --git a/geom/monetdb5/geom.mal b/geom/monetdb5/geom.mal
--- a/geom/monetdb5/geom.mal
+++ b/geom/monetdb5/geom.mal
@@ -463,9 +463,9 @@ command Contains2(g:wkb, x:bat[:oid,:dbl
command Filter(a:wkb, b:bat[:oid,:wkb]) :bat[:oid,:wkb] address
wkbFilter_geom_bat
comment "Filters the points in the bats according to the MBR of the other
bat.";
command Filter(a:bat[:oid,:wkb], b:wkb) :bat[:oid,:wkb] address
wkbFilter_bat_geom;
-command Filter1(geom:wkb, x:bat[:oid,:dbl], y:bat[:oid,:dbl]) :bat[:oid,:oid]
address wkbFilterWithImprints_geom_bat
+command Filter(geom:wkb, x:bat[:oid,:dbl], y:bat[:oid,:dbl]) :bat[:oid,:oid]
address wkbFilterWithImprints_geom_bat
comment "Filters x,y coordinates in the BATs using imprints";
-command Filter(geom:wkb, x:bat[:oid,:dbl], y:bat[:oid,:dbl]) :bat[:oid,:oid]
address wkbFilterWithPBSM_geom_bat
+command Filter1(geom:wkb, x:bat[:oid,:dbl], y:bat[:oid,:dbl]) :bat[:oid,:oid]
address wkbFilterWithPBSM_geom_bat
comment "Filters x,y coordinates in the BATs using PBSM";
#command point(x:bat[:oid,:dbl],y:bat[:oid,:dbl]) :bat[:oid,:wkb]
diff --git a/geom/monetdb5/geomPoints.c b/geom/monetdb5/geomPoints.c
--- a/geom/monetdb5/geomPoints.c
+++ b/geom/monetdb5/geomPoints.c
@@ -243,6 +243,9 @@ str wkbPointsContains1_geom_bat(bat* out
}
BBPkeepref(*outBAT_id = outBAT->batCacheid);
+
+// fprintf(stderr, "Contains1: IN %u - OUT %u\n", (unsigned
int)BATcount(xBAT), (unsigned int)BATcount(outBAT));
+
goto clean;
clean:
@@ -305,7 +308,7 @@ static str pnpoly_(int *out, const GEOSG
BBPreleaseref(bpy->batCacheid);
throw(MAL, "geom.point", "both point bats must have dense and
aligned heads");
}
-
+;
/*Create output BAT*/
if ((bo = BATnew(TYPE_void, ATOMindex("bte"), BATcount(bpx),
TRANSIENT)) == NULL) {
BBPreleaseref(bpx->batCacheid);
@@ -369,6 +372,7 @@ static str pnpoly_(int *out, const GEOSG
GDKfree(xPoints);
GDKfree(yPoints);
+//fprintf(stderr, "Contains2: IN %u - OUT %u\n", (unsigned int)BATcount(bpx),
(unsigned int)BATcount(bo));
return MAL_SUCCEED;
}
diff --git a/monetdb5/optimizer/opt_geospatial.c
b/monetdb5/optimizer/opt_geospatial.c
--- a/monetdb5/optimizer/opt_geospatial.c
+++ b/monetdb5/optimizer/opt_geospatial.c
@@ -1,23 +1,42 @@
#include "monetdb_config.h"
#include "opt_geospatial.h"
+
+typedef struct {
+ int first;
+ int second;
+} arguments;
+
+static arguments subselectInputs[5];
+static int spatials;
+
+static arguments projectInputs[5];
+//static int subselectFirstInput[5];
+//static int subselectSecondInput[5];
+//static int foundItems = 0;
+static int j = 0;
+//static int subselectsReturnArgument[5];
+//static int projectReturnArgument[5];
+static int subselects =0;
+
static void createFilterInstruction(MalBlkPtr mb, InstrPtr *oldInstrPtr, int
instructionNum, int filterFirstArgument) {
- InstrPtr filterInstrPtr, projectXInstrPtr, projectYInstrPtr,
projectInstrPtr;
- int filterReturnId, subselectReturnId, projectXReturnId,
projectYReturnId;
+ InstrPtr filterInstrPtr, projectXInstrPtr, projectYInstrPtr;//,
subselectInstrPtr, projectInstrPtr;
+ int filterReturnId, projectXReturnId, projectYReturnId;//,
subselectReturnId;
//create and put in the MAL plan the new instructions
filterInstrPtr = newStmt(mb, "batgeom", "Filter");
projectXInstrPtr = newStmt(mb, "algebra", "leftfetchjoin");
projectYInstrPtr = newStmt(mb, "algebra", "leftfetchjoin");
pushInstruction(mb, oldInstrPtr[instructionNum]);
- pushInstruction(mb, oldInstrPtr[instructionNum+1]);
- projectInstrPtr = newStmt(mb, "algebra", "leftfetchjoin");
+// //pushInstruction(mb, oldInstrPtr[instructionNum+1]);
+// subselectInstrPtr = newStmt(mb, "algebra", "subselect");
+// projectInstrPtr = newStmt(mb, "algebra", "leftfetchjoin");
//make new return variables
filterReturnId = newTmpVariable(mb, newBatType(TYPE_oid, TYPE_oid));
- projectXReturnId = newVariable(mb, GDKstrdup("xBATfiltered"),
getArgType(mb,oldInstrPtr[instructionNum],2));
- projectYReturnId = newVariable(mb, GDKstrdup("yBATfiltered"),
getArgType(mb,oldInstrPtr[instructionNum],3));
- subselectReturnId = newTmpVariable(mb, newBatType(TYPE_oid, TYPE_oid));
+ projectXReturnId = newTmpVariable(mb,
getArgType(mb,oldInstrPtr[instructionNum],2));
+ projectYReturnId = newTmpVariable(mb,
getArgType(mb,oldInstrPtr[instructionNum],3));
+// subselectReturnId = newTmpVariable(mb, newBatType(TYPE_oid, TYPE_oid));
//set the arguments for filter
setReturnArgument(filterInstrPtr, filterReturnId);
@@ -40,26 +59,101 @@ static void createFilterInstruction(MalB
setArgument(mb, oldInstrPtr[instructionNum], 2, projectXReturnId);
delArgument(oldInstrPtr[instructionNum], 3);
setArgument(mb, oldInstrPtr[instructionNum], 3, projectYReturnId);
+
+ //store the return variable of the spatial function
+ subselectInputs[spatials].first = getArg(oldInstrPtr[instructionNum],0);
+ subselectInputs[spatials].second = filterReturnId;
+//fprintf(stderr, "%d -> SpatialReturnId: %d, FilterReturnId: %d\n",
foundItems, subselectFirstInput[foundItems], subselectSecondInput[foundItems]);
+ spatials++;
- //the new subselect does not use candidates
+/* //the new subselect does not use candidates
setReturnArgument(projectInstrPtr,
getArg(oldInstrPtr[instructionNum+1],0)); //get the variable before changing it
setReturnArgument(oldInstrPtr[instructionNum+1], subselectReturnId);
if(oldInstrPtr[instructionNum+1]->argc == 8)
delArgument(oldInstrPtr[instructionNum+1], 2);
-
+
//add a new function that gets the oids of the original BAT that
qualified the spatial function
projectInstrPtr = pushArgument(mb, projectInstrPtr, subselectReturnId);
projectInstrPtr = pushArgument(mb, projectInstrPtr, filterReturnId);
-
+*/
}
+static void fixSubselect(MalBlkPtr mb, InstrPtr *oldInstrPtr, int
instructionNum, int filterReturnId) {
+ InstrPtr projectInstrPtr;
+ int subselectReturnId, projectReturnId;
+ int k=0;
+ projectInstrPtr = newStmt(mb, "algebra", "leftfetchjoin");
+
+
+ //get the return variable of this subselect
+ projectInputs[subselects].first = getArg(oldInstrPtr[instructionNum],0);
+ //unless the subselect involves results from other subselect the return
variable of
+ //it will be the return variable of the projection
+ projectReturnId = projectInputs[subselects].first;
+
+ //create the new subselect command that does not use any candidates
+ subselectReturnId = newTmpVariable(mb, newBatType(TYPE_oid, TYPE_oid));
+ setReturnArgument(oldInstrPtr[instructionNum], subselectReturnId);
+ if(oldInstrPtr[instructionNum]->argc == 8) {
+ fprintf(stderr, "secondArg %d\n",
getArg(oldInstrPtr[instructionNum], 2));
+
+ //check if the second argument to this subselect is something
coming from another subselect
+ for(k = 0; k<subselects; k++) {
+ if(getArg(oldInstrPtr[instructionNum], 2) ==
projectInputs[k].first) {
+ InstrPtr joinInstrPtr = newStmt(mb, "algebra",
"subjoin");
+ InstrPtr extraProjectInstrPtr = newStmt(mb,
"algebra", "leftfetchjoin"); //to get the oids from the original BAT that
correspond to the oids tha satisfy the join
+ int joinReturnId1 = newTmpVariable(mb,
newBatType(TYPE_oid, TYPE_oid));
+ int joinReturnId2 = newTmpVariable(mb,
newBatType(TYPE_oid, TYPE_oid));
+ projectReturnId = newTmpVariable(mb,
newBatType(TYPE_oid, TYPE_oid));
+
+
+ setReturnArgument(joinInstrPtr, joinReturnId1);
+ joinInstrPtr = pushArgument(mb, joinInstrPtr,
joinReturnId2);
+ joinInstrPtr = pushArgument(mb, joinInstrPtr,
projectReturnId); //the first BAT is the result of the projection of this
subselect
+ joinInstrPtr = pushArgument(mb, joinInstrPtr,
projectInputs[k].first); //the second BAT is the results of the projection of
the other subselect
+ joinInstrPtr = pushNil(mb, joinInstrPtr,
TYPE_bat);
+ joinInstrPtr = pushNil(mb, joinInstrPtr,
TYPE_bat);
+ joinInstrPtr = pushBit(mb, joinInstrPtr, 0);
//do not match null values
+ joinInstrPtr = pushNil(mb, joinInstrPtr,
TYPE_lng); //I do not have an estimation of the size
+ joinInstrPtr->retc=2;
+
+ setReturnArgument(extraProjectInstrPtr,
projectInputs[subselects].first);
+ extraProjectInstrPtr = pushArgument(mb,
extraProjectInstrPtr, joinReturnId1); //the result of the join
+ extraProjectInstrPtr = pushArgument(mb,
extraProjectInstrPtr, projectReturnId); //the BAT used for the join computation
+
+ }
+ }
+ delArgument(oldInstrPtr[instructionNum], 2);
+
+ }
+
+
+ //add a new function that gets the oids of the original BAT that
qualified the spatial function
+ setReturnArgument(projectInstrPtr, projectReturnId);
+ projectInstrPtr = pushArgument(mb, projectInstrPtr, subselectReturnId);
+ projectInstrPtr = pushArgument(mb, projectInstrPtr, filterReturnId);
+
+ subselects++;
+}
+
+static int getSubselectSecondInput(int currentSubselectFirstInput) {
+ int k=0;
+
+ for(k=0; k<spatials ; k++)
+ if(currentSubselectFirstInput == subselectInputs[k].first)
+ return subselectInputs[k].second;
+ return -1;
+}
int OPTgeospatialImplementation(Client cntxt, MalBlkPtr mb, MalStkPtr stk,
InstrPtr pci) {
int i=0, actions = 0;
int nextFreeSlot = mb->stop; //up to this there are instructions
InstrPtr *oldInstrPtr = mb->stmt; //pointer to the first instruction
int slimit = mb->ssize; //what is
-
+
+ spatials = 0;
+ subselects = 0;
+
(void) pci;
(void) stk;
(void) cntxt;
@@ -70,17 +164,22 @@ int OPTgeospatialImplementation(Client c
//iterate over the instructions
- for(i=0; i<nextFreeSlot; i++) {
+ for(i=0; i<nextFreeSlot; i++) {
+ //chech the module and function name
+ if(getModuleId(oldInstrPtr[i]) &&
!idcmp(getModuleId(oldInstrPtr[i]),"algebra") &&
!idcmp(getFunctionId(oldInstrPtr[i]), "subselect")) {
+ int filterReturnId = 0;
+ pushInstruction(mb, oldInstrPtr[i]);
- //chech the module and function name
- if(getModuleId(oldInstrPtr[i]) &&
!strcasecmp(getModuleId(oldInstrPtr[i]),"batgeom")) {
+ if((filterReturnId =
getSubselectSecondInput(getArg(oldInstrPtr[i], 1))) > 0)
+ fixSubselect(mb, oldInstrPtr, i,
filterReturnId);
+ } else if(getModuleId(oldInstrPtr[i]) &&
!strcasecmp(getModuleId(oldInstrPtr[i]),"batgeom")) {
if((strcasecmp(getFunctionId(oldInstrPtr[i]),
"contains1") == 0) || (strcasecmp(getFunctionId(oldInstrPtr[i]), "contains2")
== 0)) {
if(oldInstrPtr[i]->argc == 5) {
//call all necessary intructions for
the filter and the evaluation of the spatial relation
createFilterInstruction(mb,
oldInstrPtr, i, getArg(oldInstrPtr[i],1));
- //skip the algebra.subselect command
- i++;
+ ////skip the algebra.subselect command
+ //i++;
actions += 5;
} else {
@@ -140,8 +239,8 @@ int OPTgeospatialImplementation(Client c
//call all necessary intructions for
the filter and the evaluation of the spatial relation
createFilterInstruction(mb,
oldInstrPtr, i, bufferReturnId);
- //skip the algebra.thetasubselect
command
- i++;
+ ////skip the algebra.thetasubselect
command
+ //i++;
actions += 5;
} else {
@@ -189,10 +288,20 @@ int OPTgeospatialImplementation(Client c
actions+=3;
}
- } else //put back all other instructions from batgeom
+ } else {//put back all other instructions from batgeom
pushInstruction(mb, oldInstrPtr[i]);
- } else //put all other instructions back
+ }
+ } else { //put all other instructions back
pushInstruction(mb, oldInstrPtr[i]);
+
+ //I need to track the first input to the subselect
+ for(j=0; j<spatials ; j++) {
+ if(getArg(oldInstrPtr[i],1) ==
subselectInputs[j].first) {
+ subselectInputs[j].first =
getArg(oldInstrPtr[i], 0);
+ break;
+ }
+ }
+ }
}
GDKfree(oldInstrPtr);
_______________________________________________
checkin-list mailing list
checkin-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/checkin-list
