sortbuild_hash_A_2.patch

commit e73b27c9412ab916a7fb6e111238c9ecd1a72def
Author: mithun <mithun@localhost.localdomain>
Date:   Sat Apr 1 12:55:13 2017 +0530

    commit 2

diff --git a/src/backend/access/hash/hashsort.c b/src/backend/access/hash/hashsort.c
index 0e0f393..41d615d 100644
--- a/src/backend/access/hash/hashsort.c
+++ b/src/backend/access/hash/hashsort.c
@@ -37,7 +37,15 @@ struct HSpool
 {
 	Tuplesortstate *sortstate;	/* state data for tuplesort.c */
 	Relation	index;
-	uint32		hash_mask;		/* bitmask for hash codes */
+
+	/*
+	 * We sort the hash keys based on the buckets they belong to. Below masks
+	 * are used in _hash_hashkey2bucket to determine the bucket of given hash
+	 * key.
+	 */
+	uint32		high_mask;
+	uint32		low_mask;
+	uint32		max_buckets;
 };
 
 
@@ -56,11 +64,12 @@ _h_spoolinit(Relation heap, Relation index, uint32 num_buckets)
 	 * num_buckets buckets in the index, the appropriate mask can be computed
 	 * as follows.
 	 *
-	 * Note: at present, the passed-in num_buckets is always a power of 2, so
-	 * we could just compute num_buckets - 1.  We prefer not to assume that
-	 * here, though.
+	 * NOTE : This hash mask calculation should be in sync with similar
+	 * calculation in _hash_init_metabuffer.
 	 */
-	hspool->hash_mask = (((uint32) 1) << _hash_log2(num_buckets)) - 1;
+	hspool->high_mask = (((uint32) 1) << _hash_log2(num_buckets + 1)) - 1;
+	hspool->low_mask = (hspool->high_mask >> 1);
+	hspool->max_buckets = num_buckets - 1;
 
 	/*
 	 * We size the sort area as maintenance_work_mem rather than work_mem to
@@ -69,7 +78,9 @@ _h_spoolinit(Relation heap, Relation index, uint32 num_buckets)
 	 */
 	hspool->sortstate = tuplesort_begin_index_hash(heap,
 												   index,
-												   hspool->hash_mask,
+												   hspool->high_mask,
+												   hspool->low_mask,
+												   hspool->max_buckets,
 												   maintenance_work_mem,
 												   false);
 
@@ -122,7 +133,9 @@ _h_indexbuild(HSpool *hspool, Relation heapRel)
 #ifdef USE_ASSERT_CHECKING
 		uint32		lasthashkey = hashkey;
 
-		hashkey = _hash_get_indextuple_hashkey(itup) & hspool->hash_mask;
+		hashkey = _hash_hashkey2bucket(_hash_get_indextuple_hashkey(itup),
+									   hspool->max_buckets, hspool->high_mask,
+									   hspool->low_mask);
 		Assert(hashkey >= lasthashkey);
 #endif
 
diff --git a/src/backend/utils/sort/tuplesort.c b/src/backend/utils/sort/tuplesort.c
index e1e692d..65cda52 100644
--- a/src/backend/utils/sort/tuplesort.c
+++ b/src/backend/utils/sort/tuplesort.c
@@ -127,6 +127,7 @@
 
 #include "access/htup_details.h"
 #include "access/nbtree.h"
+#include "access/hash.h"
 #include "catalog/index.h"
 #include "catalog/pg_am.h"
 #include "commands/tablespace.h"
@@ -473,7 +474,9 @@ struct Tuplesortstate
 	bool		enforceUnique;	/* complain if we find duplicate tuples */
 
 	/* These are specific to the index_hash subcase: */
-	uint32		hash_mask;		/* mask for sortable part of hash code */
+	uint32		high_mask;		/* masks for sortable part of hash code */
+	uint32		low_mask;
+	uint32		max_buckets;
 
 	/*
 	 * These variables are specific to the Datum case; they are set by
@@ -991,7 +994,9 @@ tuplesort_begin_index_btree(Relation heapRel,
 Tuplesortstate *
 tuplesort_begin_index_hash(Relation heapRel,
 						   Relation indexRel,
-						   uint32 hash_mask,
+						   uint32 high_mask,
+						   uint32 low_mask,
+						   uint32 max_buckets,
 						   int workMem, bool randomAccess)
 {
 	Tuplesortstate *state = tuplesort_begin_common(workMem, randomAccess);
@@ -1002,8 +1007,11 @@ tuplesort_begin_index_hash(Relation heapRel,
 #ifdef TRACE_SORT
 	if (trace_sort)
 		elog(LOG,
-		"begin index sort: hash_mask = 0x%x, workMem = %d, randomAccess = %c",
-			 hash_mask,
+			 "begin index sort: high_mask = 0x%x, low_mask = 0x%x, "
+			 "max_buckets = 0x%x, workMem = %d, randomAccess = %c",
+			 high_mask,
+			 low_mask,
+			 max_buckets,
 			 workMem, randomAccess ? 't' : 'f');
 #endif
 
@@ -1017,7 +1025,9 @@ tuplesort_begin_index_hash(Relation heapRel,
 	state->heapRel = heapRel;
 	state->indexRel = indexRel;
 
-	state->hash_mask = hash_mask;
+	state->high_mask = high_mask;
+	state->low_mask = low_mask;
+	state->max_buckets = max_buckets;
 
 	MemoryContextSwitchTo(oldcontext);
 
@@ -4157,8 +4167,8 @@ static int
 comparetup_index_hash(const SortTuple *a, const SortTuple *b,
 					  Tuplesortstate *state)
 {
-	uint32		hash1;
-	uint32		hash2;
+	Bucket		bucket1;
+	Bucket		bucket2;
 	IndexTuple	tuple1;
 	IndexTuple	tuple2;
 
@@ -4167,13 +4177,16 @@ comparetup_index_hash(const SortTuple *a, const SortTuple *b,
 	 * that the first column of the index tuple is the hash key.
 	 */
 	Assert(!a->isnull1);
-	hash1 = DatumGetUInt32(a->datum1) & state->hash_mask;
+	bucket1 = _hash_hashkey2bucket(DatumGetUInt32(a->datum1),
+								   state->max_buckets, state->high_mask,
+								   state->low_mask);
 	Assert(!b->isnull1);
-	hash2 = DatumGetUInt32(b->datum1) & state->hash_mask;
-
-	if (hash1 > hash2)
+	bucket2 = _hash_hashkey2bucket(DatumGetUInt32(b->datum1),
+								   state->max_buckets, state->high_mask,
+								   state->low_mask);
+	if (bucket1 > bucket2)
 		return 1;
-	else if (hash1 < hash2)
+	else if (bucket1 < bucket2)
 		return -1;
 
 	/*
diff --git a/src/include/utils/tuplesort.h b/src/include/utils/tuplesort.h
index 5b3f475..9719db4 100644
--- a/src/include/utils/tuplesort.h
+++ b/src/include/utils/tuplesort.h
@@ -72,7 +72,9 @@ extern Tuplesortstate *tuplesort_begin_index_btree(Relation heapRel,
 							int workMem, bool randomAccess);
 extern Tuplesortstate *tuplesort_begin_index_hash(Relation heapRel,
 						   Relation indexRel,
-						   uint32 hash_mask,
+						   uint32 high_mask,
+						   uint32 low_mask,
+						   uint32 max_buckets,
 						   int workMem, bool randomAccess);
 extern Tuplesortstate *tuplesort_begin_datum(Oid datumType,
 					  Oid sortOperator, Oid sortCollation,