function cgExtractColumnFragment($att, $chunk, $start, $end, $wpName)
{
?>
// extracting <?php
echo $att;
?>
:
Column <?php
echo attCol($att);
?>
;
if (<?php
echo attQrys($att);
?>
.Overlaps(queriesToRun)){
<?php
echo $chunk;
?>
.SwapColumn(<?php
echo attCol($att);
?>
, <?php
echo attSlot($att);
?>
);
if (! <?php
echo attCol($att);
?>
.IsValid()){
FATAL("Error: Column <?php
echo $att;
?>
not found in <?php
echo $wpName;
?>
\n");
}
}
<?php
echo attIteratorType($att);
?>
<?php
echo attData($att);
?>
(<?php
echo attCol($att);
?>
/*, 8192*/, <?php
echo $start;
?>
, <?php
echo $end;
?>
);
<?php
}
function JoinLHS($wpName, $jDesc)
{
$rhsAttOrder = [];
foreach ($jDesc->hash_RHS_attr as $attr) {
$att = lookupAttribute($attr);
$rhsAttOrder[$att->slot()] = $attr;
}
ksort($rhsAttOrder);
$jDesc->hash_RHS_attr = $rhsAttOrder;
?>
//+{"kind":"WPF", "name":"LHS Lookup", "action":"start"}
extern "C"
int JoinLHSWorkFunc_<?php
echo $wpName;
?>
(WorkDescription &workDescription, ExecEngineData &result) {
double start_time = global_clock.GetTime();
PROFILING2_START;
// this is the area where all of the intermediate, serialized records are stored
SerializedSegmentArray serializedSegments [NUM_SEGS];
// this is the area where all of the records are serialized to;
// 10K bytes are initially used for this
char *serializeHere = (char *) malloc (10000);
// this is the output chunk
Chunk output;
// go to the work description and get the input chunk
JoinLHSWorkDescription myWork;
myWork.swap (workDescription);
Chunk &input = myWork.get_chunkToProcess ();
// get the waypoint ID from the chunk
int wayPointID = myWork.get_wayPointID ();
QueryIDSet queriesToRun = QueryExitsToQueries(myWork.get_whichQueryExits ());
<?php
cgAccessColumns($jDesc->attribute_queries_LHS, 'input', $wpName);
?>
BStringIterator myInBStringIter;
input.SwapBitmap (myInBStringIter);
// start the iterators for the output columns for LHS; used only if stillShallow = 0
<?php
foreach ($jDesc->attribute_queries_LHS_copy as $att => $queries) {
?>
<?php
echo attIteratorType($att);
?>
<?php
echo $att;
?>
_Column_Out;
<?php
}
/*foreach*/
?>
// these manage the output columns that come from the RHS (now stored in the hash table)
<?php
cgConstructColumns(array_keys($jDesc->attribute_queries_RHS_copy));
?>
// this is the ouput bitstring
MMappedStorage myStore;
Column bitmapOut (myStore);
BStringIterator myOutBStringIter (bitmapOut, queriesToRun);
// now we extract all of the hash table segments... after this, myEntries will hold them all
HashTableView myView;
myWork.get_centralHashTable ().EnterReader (myView);
HashTableSegment myEntries[NUM_SEGS];
myView.ExtractAllSegments (myEntries);
// this tells us that we are "still shallow"---not making a deep copy of the LHS atts to the output
int stillShallow = 1;
// the bitstring that will be exracted from the hash table
QueryIDSet *bitstringRHS = 0;
QueryIDSet existsTarget(<?php
echo $jDesc->exists_target;
?>
, true);
QueryIDSet notExistsTarget(<?php
echo $jDesc->not_exists_target;
?>
, true);
// these are all of the attribute values that come from the hash table...
// for each att we need a pointer as well as a dummy value that the pointer will be set to by default
<?php
foreach ($jDesc->attribute_queries_RHS as $att => $queries) {
?>
QueryIDSet <?php
echo attQrys($att);
?>
_RHS(<?php
echo $queries;
?>
, true);
<?php
echo attType($att);
?>
<?php
echo $att;
?>
RHSShadow;
<?php
echo attType($att);
?>
*<?php
echo $att;
?>
RHS = NULL;
<?php
echo attType($att);
?>
<?php
echo $att;
?>
RHSobj;
<?php
}
/*foreach*/
?>
// now actually try to match up all of the tuples!
int totalNum = 0;
while (!myInBStringIter.AtEndOfColumn ()) { // TBD, probably this is not working TBD
// counts how many matches for this query
int numHits = 0;
// extract values of attributes from streams
// now go through the LHS input atts one at a time and extract if it is needed by an active query
// see which queries match up
QueryIDSet curBits = myInBStringIter.GetCurrent ();
curBits.Intersect (queriesToRun);
QueryIDSet exists; // keeps track of the queries for which a match is found
QueryIDSet oldBitstringLHS; // last value of bistringLHS
// if the input query is not empty
if (!curBits.IsEmpty ()) {
totalNum++;
// compute the hash for LHS
HT_INDEX_TYPE hashValue = HASH_INIT;
<?php
foreach ($jDesc->LHS_keys as $att) {
?>
hashValue = CongruentHash(Hash(<?php
echo $att;
?>
_Column.GetCurrent()), hashValue);
<?php
}
/*foreach*/
?>
// figure out which of the hash buckets it goes into
unsigned int index = WHICH_SEGMENT (hashValue);
// now, go to that index and extract matching tuples!
HT_INDEX_TYPE curSlot = WHICH_SLOT (hashValue);
hashValue = curSlot;
// this loops through all of the possible RHS hits
while (1) {
// this is the bitstring that will go in the output
QueryIDSet bitstringLHS;
// for safety (in case we look at a bitstring that spans multiple
// entries that is not done being written by a concurrent writer)
// empty out the inital bitstring
((QueryIDSet *) serializeHere)->Empty ();
// give safe "shadow" values to all of the RHS attributes
<?php
foreach ($jDesc->hash_RHS_attr as $att) {
?>
<?php
echo $att;
?>
RHS = &<?php
echo $att;
?>
RHSShadow;
<?php
}
/*foreach*/
?>
// here we go through and extract the atts one at a time from the hash
// table. Note that the atts must be extracted IN ORDER. That is, the
// bitstring is first, followed by the att mapped to the lowerest column
// position, followed by the att mapped to the next lowest, and so on.
// The Extract function pulls an attribute out of the hash table...
int lenSoFar = 0, dummy, done;
int lastLen = myEntries[index].Extract (serializeHere, curSlot, hashValue, wayPointID, BITMAP, dummy, done);
// if we cannot find a bitstring, there was no tuple here, and we are done
if (lastLen == 0) {
break;
}
// remember the bitstring
bitstringRHS = (QueryIDSet *) serializeHere;
lenSoFar += lastLen;
// next look for other hashed attributes
<?php
foreach ($jDesc->hash_RHS_attr as $att) {
?>
lastLen = myEntries[index].Extract (serializeHere + lenSoFar, curSlot, hashValue, wayPointID, <?php
echo attSlot($att);
?>
, dummy, done);
// see if we got attribute
if (lastLen > 0) {
Deserialize(serializeHere + lenSoFar, <?php
echo $att;
?>
RHSobj);
//<?php
echo attOptimizedDeserialize($att, $att . "RHSobj", "serializeHere", "lenSoFar");
?>
;
<?php
echo $att;
?>
RHS = &<?php
echo $att;
?>
RHSobj;
lenSoFar += lastLen;
} else {
FATALIF(<?php
echo attQrys($att);
?>
_RHS.Overlaps(*bitstringRHS),
"Did not find attribute <?php
echo $att;
?>
in active RHS tuple");
}
<?php
}
/*foreach*/
?>
// see if we have any query matches
bitstringRHS->Intersect (curBits);
QueryIDSet qBits;
//printf("TPLLLLL: cust_acctbal = %f orders_custkey = %d cust_custkey = %d\n", *customer_c_acctbalRHS, orders_o_custkey_Column.GetCurrent(), *customer_c_custkeyRHS);
<?php
foreach ($jDesc->queries_attribute_comparison as $qClass) {
?>
// See if any query in query class is eligible for this comparision
qBits = QueryIDSet(<?php
echo $qClass->qClass;
?>
, true);
qBits.Intersect(*bitstringRHS);
if (
!qBits.IsEmpty () &&
<?php
foreach ($qClass->att_pairs as $pair) {
?>
*<?php
echo $pair->rhs;
?>
RHS == <?php
echo $pair->lhs;
?>
_Column.GetCurrent() &&
<?php
}
/*foreach pair*/
?>
1 )
{
bitstringLHS.Union (qBits);
}
<?php
}
/*foreach query class*/
?>
// if any of them hit...
if (!bitstringLHS.IsEmpty ()) {
exists.Union(bitstringLHS);
numHits++;
// see if we need to move from shallow to deep
if (numHits == 2 && stillShallow) {
<?php
foreach ($jDesc->attribute_queries_LHS_copy as $att => $qrys) {
?>
<?php
echo attData($att);
?>
_Out.CreateDeepCopy (<?php
echo attData($att);
?>
);
<?php
echo attData($att);
?>
_Out.Insert (<?php
echo attData($att);
?>
.GetCurrent());
<?php
echo attData($att);
?>
_Out.Advance();
<?php
}
/*foreach*/
?>
stillShallow = 0;
}
// now, add all of the outputs over... first deal with the LHS input atts
// that get copied into output atts
if (!stillShallow) {
<?php
foreach ($jDesc->attribute_queries_LHS_copy as $att => $qrys) {
?>
<?php
echo attData($att);
?>
_Out.Insert (<?php
echo attData($att);
?>
.GetCurrent());
<?php
echo attData($att);
?>
_Out.Advance();
<?php
}
/*foreach*/
?>
}
// now, deal with the output atts that come from the hash table
<?php
foreach ($jDesc->attribute_queries_RHS_copy as $att => $qrys) {
?>
<?php
echo attData($att);
?>
_Out.Insert (*<?php
echo $att;
?>
RHS);
<?php
echo attData($att);
?>
_Out.Advance();
<?php
}
/*foreach*/
?>
// finally, set the bitmap. We are one element behind
if (!oldBitstringLHS.IsEmpty()){
myOutBStringIter.Insert (oldBitstringLHS);
myOutBStringIter.Advance ();
}
oldBitstringLHS=bitstringLHS;
} // empty bistring
}
}
// compute the true exist queries
QueryIDSet tmp = existsTarget;
tmp.Intersect(exists);
tmp.Intersect(curBits); // not needed but I'm paranoid
// compute the true not exits queries
QueryIDSet tmp2 = curBits;
tmp2.Intersect(notExistsTarget);
tmp2.Difference(exists);
// now put everything in bitstringLHS
oldBitstringLHS.Union(tmp);
oldBitstringLHS.Union(tmp2);
if (!oldBitstringLHS.IsEmpty()){
myOutBStringIter.Insert (oldBitstringLHS);
myOutBStringIter.Advance ();
}
// at this point, we are done trying to join this tuple... any join results have been
// written to the output columns. Note that we don't have to advance in the output data
// columns; if we are shallow, we don't touch the output columns. If we are not shallow,
// if there were no results, we have nothing to write. HOWEVER, if we are shallow and
// we did not get a match, we need to add an empty btstring
if (stillShallow && numHits == 0) {
<?php
foreach ($jDesc->attribute_queries_RHS_copy as $att => $qrys) {
?>
<?php
echo attType($att);
?>
tmp_<?php
echo attData($att);
?>
;
<?php
echo attData($att);
?>
_Out.Insert (tmp_<?php
echo attData($att);
?>
);
<?php
echo attData($att);
?>
_Out.Advance();
<?php
}
/*foreach*/
?>
if (oldBitstringLHS.IsEmpty()){ // no not exist and no join match
myOutBStringIter.Insert (oldBitstringLHS);
myOutBStringIter.Advance ();
}
}
// lastly, we need to advance in the INPUT tuples
<?php
foreach ($jDesc->attribute_queries_LHS as $att => $qrys) {
?>
<?php
echo attData($att);
?>
.Advance();
<?php
}
/*foreach*/
?>
// advance the input bitstring
myInBStringIter.Advance ();
}
// DONE! So construct the output tuple
// if we are still shallow, put the original data into the output
if (stillShallow) {
<?php
foreach ($jDesc->attribute_queries_LHS_copy as $att => $qrys) {
?>
Column col_<?php
echo $att;
?>
;
<?php
echo attData($att);
?>
.Done(col_<?php
echo $att;
?>
);
output.SwapColumn (col_<?php
echo $att;
?>
, <?php
echo attSlot($att);
?>
);
<?php
}
/*foreach*/
?>
} else {
<?php
foreach ($jDesc->attribute_queries_LHS_copy as $att => $qrys) {
?>
Column col_<?php
echo $att;
?>
;
<?php
echo attData($att);
?>
_Out.Done(col_<?php
echo $att;
?>
);
output.SwapColumn (col_<?php
echo $att;
?>
, <?php
echo attSlot($att);
?>
);
<?php
}
/*foreach*/
?>
}
{
<?php
foreach ($jDesc->attribute_queries_RHS_copy as $att => $qrys) {
?>
Column col_<?php
echo $att;
?>
;
<?php
echo attData($att);
?>
_Out.Done(col_<?php
echo $att;
?>
);
output.SwapColumn (col_<?php
echo $att;
?>
, <?php
echo attSlot($att);
?>
);
<?php
}
/*foreach*/
?>
}
// put in the output bitmap
myOutBStringIter.Done ();
output.SwapBitmap (myOutBStringIter);
// and give back the result
ChunkContainer tempResult (output);
tempResult.swap (result);
PROFILING2_END;
PROFILING(start_time, "<?php
echo $wpName;
?>
", "LHS_lookup", "%d", totalNum);
PROFILING(0.0, "HashTable", "fillrate", "%2.4f", HashTableSegment::globalFillRate*100.0);
// Finish performance counters
// Use the Set functionality in case we add additional counters later.
PCounterList counterList;
PCounter totalCnt("tpi lhs", totalNum, "<?php
echo $wpName;
?>
");
counterList.Append(totalCnt);
PROFILING2_SET(counterList, "<?php
echo $wpName;
?>
");
int64_t hFillRate = int64_t(HashTableSegment::globalFillRate * 1000);
PROFILING2_INSTANT("hfr", hFillRate, "global");
free (serializeHere);
return 1;
}
//+{"kind":"WPF", "name":"LHS Lookup", "action":"end"}
<?php
}
function JoinMerge($wpName, $jDesc)
{
?>
#define USE_PQ
#ifdef USE_PQ
//#include <algorithm>
#include <vector>
#else
#include <vector>
#include<set>
#endif
#include "WorkDescription.h"
#include "ExecEngineData.h"
#include "Column.h"
#include "ColumnIterator.cc"
#include "MMappedStorage.h"
#include "BString.h"
#include "BStringIterator.h"
#include "HashTableMacros.h"
#include <string.h>
using namespace std;
#define MAX_HASH 0xffffffffffff
#define MAX_NUM_CHUNKS 128 // maximum number of chunks we can have for linear scan solution
#ifndef USE_PQ
struct HashWrapper {
HashWrapper() : hash(MAX_HASH), chunkNo(0) {}
__uint64_t hash:48;
__uint64_t chunkNo:16;
};
#endif
<?php
// Macro to find minimum value in a list
// this is the loop that finds the best entry using linear scan
// # minIndex$1 is set at the best position
// in put assumed in hashes$1
// Arguments:
// $side is "LHS" or "RHS"
$FIND_MIN = function ($side) {
?>
best<?php
echo $side;
?>
= MAX_HASH;
for (int i=0; i < num<?php
echo $side;
?>
Chunks; i++) {
HT_INDEX_TYPE hash = hashes<?php
echo $side;
?>
[i]; // current value
minIndex<?php
echo $side;
?>
= (best<?php
echo $side;
?>
> hash) ? i : minIndex<?php
echo $side;
?>
;
best<?php
echo $side;
?>
= (best<?php
echo $side;
?>
> hash) ? hash: best<?php
echo $side;
?>
;
}
// remove this after debugging
FATALIF(best<?php
echo $side;
?>
!= MAX_HASH && WHICH_SEGMENT(best<?php
echo $side;
?>
) !=0,
"Hash out of range %llu", (unsigned long long) best<?php
echo $side;
?>
);
<?php
};
?>
/*
void check_correctness(int start, int end, BStringIterator& biter, ColumnIterator<__uint64_t>& hiter) {
assert(start == end);
int s = 0;
int e = 262143;
if (start > 0) {
s = start * 262143 + (start-1);
e = s + 262143 + 1;
}
int tuple = 0;
while (!biter.AtEndOfColumn()) {
if (!(s < hiter.GetCurrent() && hiter.GetCurrent() <= e))
printf("\n ------- (%d,%d)(%d,%d) currVal = %d", start,end,s,e,hiter.GetCurrent());
assert(s <= hiter.GetCurrent() && hiter.GetCurrent() <= e);
hiter.Advance();
biter.Advance();
tuple++;
}
printf("\n num tuples = %d", tuple);
}
*/
#ifndef USE_PQ
// make heap comparator
struct compare_key {
bool operator()( const HashWrapper lhs, const HashWrapper rhs )
{
return (rhs.hash < lhs.hash);
}
};
struct compare_key_rev {
bool operator()( const HashWrapper lhs, const HashWrapper rhs )
{
return (lhs.hash < rhs.hash);
}
};
#endif
<?php
// More macros
// Macro to Advance() columns
// $side is LHS or RHS
// $chunkNum is the chunk number
// $sde is Lhs or Rhs
$ADVANCE_CALL = function ($side, $chunkNum, $sde) use($jDesc) {
$list = "attribute_queries_" . $side;
foreach ($jDesc->{$list} as $att => $queries) {
?>
col<?php
echo $side;
?>
IterVec_<?php
echo $att;
?>
[<?php
echo $chunkNum;
?>
].Advance();
<?php
}
?>
myInBStringIter<?php
echo $sde;
?>
Vec[<?php
echo $chunkNum;
?>
].Advance();
col<?php
echo $side;
?>
IterVecHash[<?php
echo $chunkNum;
?>
].Advance();
<?php
};
$ADVANCE_SEARCH_CALL = function ($col, $index, $side, $sde) use($jDesc, $FIND_MIN) {
$list = "attribute_queries_" . $side;
?>
//while (!myInBStringIter$4Vec[$2].AtEndOfColumn())
if (!myInBStringIter<?php
echo $sde;
?>
Vec[<?php
echo $index;
?>
].AtEndOfColumn()) {
//while (!<?php
echo $col;
?>
[<?php
echo $index;
?>
].AtUnwrittenByte())
// Now find the first tuple for which query is active
//Bitstring curBits(myInBStringIterLhsVec[<?php
echo $index;
?>
].GetCurrent ());
//curBits.Intersect (queriesToRun);
// If tuple has some active query, fill the heap
//if (!curBits.IsEmpty())
totalguys<?php
echo $side;
?>
++;
#ifdef USE_PQ
// set new value in place of old value >
// ALIN: Please fix the line below, I don't know what the $1 is supposed to be '
hashes<?php
echo $side;
?>
[minIndex<?php
echo $side;
?>
] = <?php
echo $col;
?>
[minIndex<?php
echo $side;
?>
].GetCurrent();
#else
HashWrapper w;
w.hash = <?php
echo $col;
?>
[<?php
echo $index;
?>
].GetCurrent(); // assume first column of each chunk is hash
w.chunkNo = <?php
echo $index;
?>
;
minHeap<?php
echo $side;
?>
.insert (w);
#endif
/*
break; // while loop
} else {
// If this tuple don't have active query, advance all the columns of this chunk '
// including Bitstring column
<?php
foreach ($jDesc->{$list} as $att) {
?>
col<?php
echo $side;
?>
IterVec_<?php
echo $att;
?>
[<?php
echo $index;
?>
].Advance();
<?php
}
/* foreach */
?>
myInBStringIter<?php
echo $sde;
?>
Vec[<?php
echo $index;
?>
].Advance();
col<?php
echo $side;
?>
IterVecHash[<?php
echo $index;
?>
].Advance();
}
*/
}
#ifdef USE_PQ
<?php
$FIND_MIN($side);
?>
#endif
<?php
};
/* ADVANCE_SEARCH_CALL */
?>
/*
This takes two sorted list of chunks, lhs list and rhs list, and do the sort merge join.
It maintaines a min heap of hash values on top of each list and virtually now we have
just 2 lists to do sort merge. When some value matches from LHS and RHS heap, we need
to checkpoint all the iterators of all the columns of all the RHS chunks including storing
the heap, so that we can restore them all if another consecutive LHS value matches.
*/
//+{"kind":"WPF", "name":"Merge", "action":"start"}
extern "C"
int JoinMergeWorkFunc_<?php
echo $wpName;
?>
_writer (WorkDescription &workDescription, ExecEngineData &result) {
int totalguysLHS = 0;
int totalguysRHS = 0;
int total = 0;
int totalhash = 0;
double start_time = global_clock.GetTime();
// get the input LHS and RHS chunk container from work descripton
JoinMergeWorkDescription myWork;
myWork.swap (workDescription);
ContainerOfChunks &inputLHSList = myWork.get_chunksLHS ();
ContainerOfChunks &inputRHSList = myWork.get_chunksRHS ();
int start = myWork.get_start();
int end = myWork.get_end();
// get the number of chunks in each list to create vectors of this length
int numLHSChunks = inputLHSList.Length();
int numRHSChunks = inputRHSList.Length();
// get the waypoint identifier
unsigned int wayPointID = myWork.get_wayPointID ().GetID();
vector<ColumnIterator<__uint64_t> > colLHSIterVecHash;
vector<ColumnIterator<__uint64_t> > colRHSIterVecHash;
colLHSIterVecHash.resize(numLHSChunks);
colRHSIterVecHash.resize(numRHSChunks);
// get the input bitmap out of the LHS input chunks
vector<BStringIterator> myInBStringIterLhsVec;
myInBStringIterLhsVec.resize(numLHSChunks); // set the vector size
assert(inputLHSList.Length());
inputLHSList.MoveToStart();
int i = 0;
while (inputLHSList.RightLength()) {
BStringIterator myInBStringIter;
inputLHSList.Current().SwapBitmap (myInBStringIter);
myInBStringIterLhsVec[i].swap (myInBStringIter);
myInBStringIterLhsVec[i].SetFragmentRange(start, end);
// get the hash now
Column col_hash;
inputLHSList.Current().SwapHash(col_hash);
assert(col_hash.IsValid());
ColumnIterator<__uint64_t> iter (col_hash, start, end);
colLHSIterVecHash[i].swap(iter);
//check_correctness(start, end, myInBStringIterLhsVec[i], colLHSIterVecHash[i]);
inputLHSList.Advance ();
i++;
}
// get the input bitmap out of the RHS input chunks
vector<BStringIterator> myInBStringIterRhsVec;
myInBStringIterRhsVec.resize(numRHSChunks); // set the vector size
assert(inputRHSList.Length());
inputRHSList.MoveToStart();
i = 0;
while (inputRHSList.RightLength()) {
BStringIterator myInBStringIter;
inputRHSList.Current().SwapBitmap(myInBStringIter);
myInBStringIterRhsVec[i].swap (myInBStringIter);
myInBStringIterRhsVec[i].SetFragmentRange(start, end);
//fprintf(stderr, "\nRHS %d", myInBStringIterRhsVec[i].GetNumTuples());
// get the hash now
Column col_hash;
inputRHSList.Current().SwapHash(col_hash);
assert(col_hash.IsValid());
ColumnIterator<__uint64_t> iter (col_hash, start, end);
colRHSIterVecHash[i].swap(iter);
//check_correctness(start, end, myInBStringIterRhsVec[i], colRHSIterVecHash[i]);
inputRHSList.Advance ();
i++;
}
// get all of the queries that are active here
QueryIDSet queriesToRun = QueryExitsToQueries(myWork.get_whichQueryExits ());
// this is the output chunk
Chunk output;
// create output iterators
<?php
foreach ($jDesc->attribute_queries_LHS_copy as $att => $queries) {
?>
MMappedStorage store_<?php
echo $att;
?>
;
Column col_<?php
echo $att;
?>
(store_<?php
echo $att;
?>
);
<?php
echo attIteratorType($att);
?>
colLHSOutIter_<?php
echo $att;
?>
(col_<?php
echo $att;
?>
);
<?php
}
/*foreach*/
?>
<?php
foreach ($jDesc->attribute_queries_RHS_copy as $att => $queries) {
?>
MMappedStorage store_<?php
echo $att;
?>
;
Column col_<?php
echo $att;
?>
(store_<?php
echo $att;
?>
);
<?php
echo attIteratorType($att);
?>
colRHSOutIter_<?php
echo $att;
?>
(col_<?php
echo $att;
?>
);
<?php
}
/*foreach*/
?>
// Create output BitString
MMappedStorage myStore;
Column bitmapOut (myStore);
BStringIterator myOutBStringIter (bitmapOut, queriesToRun);
// Build input iterators vectors first
// Define only those attributes iterator which are required by some query
// vector contains same type of column in all chunks, each vector index represents
// a chunk
<?php
foreach ($jDesc->attribute_queries_LHS as $att => $queries) {
?>
vector< <?php
echo attIteratorType($att);
?>
> colLHSIterVec_<?php
echo $att;
?>
;
colLHSIterVec_<?php
echo $att;
?>
.resize(numLHSChunks);
<?php
}
/* foreach */
?>
<?php
foreach ($jDesc->attribute_queries_RHS as $att => $queries) {
?>
vector< <?php
echo attIteratorType($att);
?>
> colRHSIterVec_<?php
echo $att;
?>
;
colRHSIterVec_<?php
echo $att;
?>
.resize(numRHSChunks);
<?php
}
/* foreach */
?>
// Extract columns now.
// This extracts columns if there is any query for it in this WP
i = 0;
<?php
foreach ($jDesc->attribute_queries_LHS as $att => $queries) {
?>
{
i = 0;
inputLHSList.MoveToStart();
while (inputLHSList.RightLength()) {
// WP node queries intersect Translator provided queries
QueryIDSet <?php
echo attQrys($att);
?>
(<?php
echo $queries;
?>
, true);
<?php
cgExtractColumnFragment($att, "inputLHSList.Current()", "start", "end", $wpName);
?>
if (<?php
echo attQrys($att);
?>
.Overlaps(queriesToRun)){
colLHSIterVec_<?php
echo $att;
?>
[i].swap(<?php
echo attData($att);
?>
);
}
inputLHSList.Advance ();
i++;
}
}
<?php
}
/* foreach */
?>
<?php
foreach ($jDesc->attribute_queries_RHS as $att => $queries) {
?>
{
i = 0;
inputRHSList.MoveToStart();
while (inputRHSList.RightLength()) {
// WP node queries intersect Translator provided queries
QueryIDSet <?php
echo attQrys($att);
?>
(<?php
echo $queries;
?>
, true);
<?php
cgExtractColumnFragment($att, "inputRHSList.Current()", "start", "end", $wpName);
?>
if (<?php
echo attQrys($att);
?>
.Overlaps(queriesToRun)){
colRHSIterVec_<?php
echo $att;
?>
[i].swap(<?php
echo attData($att);
?>
);
}
inputRHSList.Advance ();
i++;
}
}
<?php
}
/* foreach */
?>
// Here we start the sort merge join
// Create priority queue for hash column values
#ifdef USE_PQ
HT_INDEX_TYPE hashesLHS[MAX_NUM_CHUNKS];
HT_INDEX_TYPE hashesRHS[MAX_NUM_CHUNKS];
FATALIF(numLHSChunks >= MAX_NUM_CHUNKS, "Too many chunks for LHS in Merge");
FATALIF(numRHSChunks >= MAX_NUM_CHUNKS, "Too many chunks for RHS in Merge");
HT_INDEX_TYPE bestLHS = MAX_HASH;
HT_INDEX_TYPE bestRHS = MAX_HASH;
int minIndexLHS = -1; // this indicates the chunk that is the best at this point
// this invariant is maintained throught the code
int minIndexRHS = -1;
#else
multiset<HashWrapper, compare_key_rev> minHeapLHS;
multiset<HashWrapper, compare_key_rev> minHeapRHS;
#endif
// Fill the first value of each LHS chunk in LHS heap
for (i = 0; i < numLHSChunks; i++) {
#ifdef USE_PQ
minIndexLHS = i;
#endif
<?php
$ADVANCE_SEARCH_CALL("colLHSIterVecHash", "i", "LHS", "Lhs");
?>
}
// Fill the first value of each RHS chunk in RHS heap
for (i = 0; i < numRHSChunks; i++) {
#ifdef USE_PQ
minIndexRHS = i;
#endif
<?php
$ADVANCE_SEARCH_CALL("colRHSIterVecHash", "i", "RHS", "Rhs");
?>
}
// Now pick one of each heap and keep comparing until one of the heap is exhausted
#ifdef USE_PQ
/* while (hashesLHS[minIndexLHS] != MAX_HASH && hashesRHS[minIndexRHS] != MAX_HASH) {
HT_INDEX_TYPE wl = hashesLHS[minIndexLHS];
HT_INDEX_TYPE wr = hashesRHS[minIndexRHS]; -- INEFFICIENT */
while (bestLHS != MAX_HASH && bestRHS != MAX_HASH) {
HT_INDEX_TYPE wl = bestLHS;
HT_INDEX_TYPE wr = bestRHS;
#else
while (!minHeapLHS.empty() && !minHeapRHS.empty()) {
HashWrapper wlT = *(minHeapLHS.begin());
HashWrapper wrT = *(minHeapRHS.begin());
HT_INDEX_TYPE wl = wlT.hash;
HT_INDEX_TYPE wr = wrT.hash;
int minIndexLHS = wlT.chunkNo;
int minIndexRHS = wrT.chunkNo;
#endif
if (wl < wr) {
// printf("\n Hash val not found is = %ld, segment = %ld, slot = %ld", wl, WHICH_SEGMENT (wl), WHICH_SLOT (wl)); fflush(stdout); assert(0);
// erase the minimum element
#ifdef USE_PQ
hashesLHS[minIndexLHS] = MAX_HASH;
#else
assert (!minHeapLHS.empty());
minHeapLHS.erase(minHeapLHS.begin());
#endif
// here advance all columns of LHS of chunk number wl.chunkNo
<?php
$ADVANCE_CALL("LHS", "minIndexLHS", "Lhs");
$ADVANCE_SEARCH_CALL("colLHSIterVecHash", "minIndexLHS", "LHS", "Lhs");
?>
} else if (wl > wr) {
// erase the minimum element
#ifdef USE_PQ
hashesRHS[minIndexRHS] = MAX_HASH;
#else
assert (!minHeapRHS.empty());
minHeapRHS.erase(minHeapRHS.begin());
#endif
// here advance all columns of RHS of chunk number wr.chunkNo
<?php
$ADVANCE_CALL("RHS", "minIndexRHS", "Rhs");
$ADVANCE_SEARCH_CALL("colRHSIterVecHash", "minIndexRHS", "RHS", "Rhs");
?>
} else { // (wl == wr)
HT_INDEX_TYPE matchingHash = wl;
// Save checkpoint for all rhs columns before incrementing both chunks
for (int chk = 0; chk < numRHSChunks; chk++) {
<?php
foreach ($jDesc->attribute_queries_RHS as $att => $queries) {
?>
colRHSIterVec_<?php
echo $att;
?>
[chk].CheckpointSave();
<?php
}
/* foreach */
?>
myInBStringIterRhsVec[chk].CheckpointSave();
colRHSIterVecHash[chk].CheckpointSave();
}
// Also save the state of the heap
#ifdef USE_PQ
HT_INDEX_TYPE hashesRHS_copy[MAX_NUM_CHUNKS];
memcpy(hashesRHS_copy, hashesRHS, numRHSChunks*sizeof(HT_INDEX_TYPE));
int minIndexRHS_copy = minIndexRHS;
#else
multiset<HashWrapper, compare_key_rev> minHeapRHSCheckpoint(minHeapRHS);
#endif
#ifdef USE_PQ
while (bestLHS != MAX_HASH)
#else
while (!minHeapLHS.empty()) // break from this if 2 consecutive LHS mismatches
#endif
{
#ifdef USE_PQ
// HT_INDEX_TYPE wl1 = hashesLHS[minIndexLHS];
HT_INDEX_TYPE wl1 = bestLHS;
#else
HashWrapper wl1T = *(minHeapLHS.begin());
HT_INDEX_TYPE wl1 = wl1T.hash;
int minIndexLHS = wl1T.chunkNo;
#endif
if (wl1 != matchingHash) { // next LHS dont match to previous LHS hash, first time always match
break;
} else { // restore everything
// restore the RHS Column iterators to original value
for (int chk = 0; chk < numRHSChunks; chk++) {
<?php
foreach ($jDesc->attribute_queries_RHS as $att => $queries) {
?>
colRHSIterVec_<?php
echo $att;
?>
[chk].CheckpointRestore();
<?php
}
/* foreach */
?>
myInBStringIterRhsVec[chk].CheckpointRestore();
colRHSIterVecHash[chk].CheckpointRestore();
}
// restore the original heap state
#ifdef USE_PQ
memcpy(hashesRHS, hashesRHS_copy, numRHSChunks*sizeof(HT_INDEX_TYPE));
minIndexRHS = minIndexRHS_copy;
bestRHS = hashesRHS[minIndexRHS_copy];
#else
minHeapRHS.clear();
minHeapRHS = minHeapRHSCheckpoint;
#endif
}
#ifdef USE_PQ
// while (hashesRHS[minIndexRHS] != MAX_HASH) {
while (bestRHS != MAX_HASH) {
#else
while (!minHeapRHS.empty()) {
#endif
#ifdef USE_PQ
// HT_INDEX_TYPE wr1 = hashesRHS[minIndexRHS];
HT_INDEX_TYPE wr1 = bestRHS;
#else
HashWrapper wr1T = *(minHeapRHS.begin());
HT_INDEX_TYPE wr1 = wr1T.hash;
int minIndexRHS = wr1T.chunkNo;
#endif
if (wl1 == wr1) { // first one will obviously match as it matched before
// Merge all columns here for wl1.chunkNo and wr1.chunkNo after matching attributes
// Make sure both of their bitstrings intersect
Bitstring rez = queriesToRun;
rez.Intersect (myInBStringIterRhsVec[minIndexRHS].GetCurrent ());
rez.Intersect (myInBStringIterLhsVec[minIndexLHS].GetCurrent ());
// This contains union of all translater queries
Bitstring uni = 0;
Bitstring qBits;
totalhash++;
// Do the actual comparision
bool anyOneMatch = false;
<?php
foreach ($jDesc->queries_attribute_comparison as $qClass) {
?>
// See if any query in query class is eligible for this comparision
qBits = QueryIDSet(<?php
echo $qClass->qClass;
?>
, true);
qBits.Intersect(rez);
if (
/* !qBits.IsEmpty () && */
<?php
foreach ($qClass->att_pairs as $pair) {
?>
colLHSIterVec_<?php
echo $pair->lhs;
?>
[minIndexLHS].GetCurrent() == colRHSIterVec_<?php
echo $pair->rhs;
?>
[minIndexRHS].GetCurrent() &&
<?php
}
/*foreach pair*/
?>
1 ) {
anyOneMatch = true;
uni.Union (qBits);
}
<?php
}
/* foreach qClass */
?>
if (anyOneMatch) {
// fill the output iterators
<?php
foreach ($jDesc->attribute_queries_LHS_copy as $att => $queries) {
?>
//if (rez.Overlaps(<?php
echo $queries;
?>
)) { Should we check translator queries?
colLHSOutIter_<?php
echo $att;
?>
.Insert(colLHSIterVec_<?php
echo $att;
?>
[minIndexLHS].GetCurrent());
colLHSOutIter_<?php
echo $att;
?>
.Advance();
//}
<?php
}
/* foreach */
?>
<?php
foreach ($jDesc->attribute_queries_RHS_copy as $att => $queries) {
?>
//if (rez.Overlaps(<?php
echo $queries;
?>
)) { Should we check translator queries?
colRHSOutIter_<?php
echo $att;
?>
.Insert(colRHSIterVec_<?php
echo $att;
?>
[minIndexRHS].GetCurrent());
colRHSOutIter_<?php
echo $att;
?>
.Advance();
//}
<?php
}
/* foreach */
?>
myOutBStringIter.Insert (uni);
total++;
myOutBStringIter.Advance ();
} else {
//Bitstring b(0, true);
//myOutBStringIter.Insert (b);
//myOutBStringIter.Advance ();
}
// erase the minimum element
#ifdef USE_PQ
hashesRHS[minIndexRHS] = MAX_HASH;
#else
minHeapRHS.erase(minHeapRHS.begin());
#endif
// here advance all columns of RHS of chunk number wr1.chunkNo
<?php
$ADVANCE_CALL("RHS", "minIndexRHS", "Rhs");
$ADVANCE_SEARCH_CALL("colRHSIterVecHash", "minIndexRHS", "RHS", "Rhs");
?>
} else {
break;
}
}
// erase the minimum element
#ifdef USE_PQ
hashesLHS[minIndexLHS] = MAX_HASH;
#else
minHeapLHS.erase(minHeapLHS.begin());
#endif
// here advance all columns of LHS of chunk number wl1.chunkNo
<?php
$ADVANCE_CALL("LHS", "minIndexLHS", "Lhs");
$ADVANCE_SEARCH_CALL("colLHSIterVecHash", "minIndexLHS", "LHS", "Lhs");
?>
}
}
}
// fill the output iterators
<?php
foreach ($jDesc->attribute_queries_LHS_copy as $att => $queries) {
?>
Column collhs_<?php
echo $att;
?>
;
colLHSOutIter_<?php
echo $att;
?>
.Done(collhs_<?php
echo $att;
?>
);
output.SwapColumn (collhs_<?php
echo $att;
?>
, <?php
echo attSlot($att);
?>
);
<?php
}
/* foreach */
foreach ($jDesc->attribute_queries_RHS_copy as $att => $queries) {
?>
Column colrhs_<?php
echo $att;
?>
;
colRHSOutIter_<?php
echo $att;
?>
.Done(colrhs_<?php
echo $att;
?>
);
output.SwapColumn (colrhs_<?php
echo $att;
?>
, <?php
echo attSlot($att);
?>
);
<?php
}
/* foreach */
?>
//myOutBStringIter.Done (bitmapOut);
myOutBStringIter.Done ();
// printf("\nTuples %d %d %d %d %d", myOutBStringIter.GetNumTuples(), total, totalguysLHS, totalguysRHS, totalhash);
// fflush(stdout);
PROFILING(start_time, "<?php
echo $wpName;
?>
", "Merge", "%d\t%d", totalhash, total);
//output.SwapBitmap (bitmapOut);
output.SwapBitmap (myOutBStringIter);
/*
MMappedStorage st;
Column co (st);
Bitstring patt(0xf, true);
BStringIterator iter(co, patt, 0);
output.SwapBitmap(iter);
*/
// and give back the result
ChunkContainer tempResult (output);
tempResult.swap (result);
return 0; // have to return something
}
//+{"kind":"WPF", "name":"Merge", "action":"end"}
<?php
}
