Added concatenate operator for vectors [skip ci]

.github/workflows/build.yml

@@ -40,13 +40,21 @@ jobs:
           sudo apt-get install libipc-run-perl
       - run: make prove_installcheck
   mac:
-    runs-on: macos-latest
+    runs-on: ${{ matrix.os }}
     if: ${{ !startsWith(github.ref_name, 'windows') }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - postgres: 16
+            os: macos-14
+          - postgres: 14
+            os: macos-12
     steps:
       - uses: actions/checkout@v4
       - uses: ankane/setup-postgres@v1
         with:
-          postgres-version: 14
+          postgres-version: ${{ matrix.postgres }}
       - run: make
         env:
           PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter
@@ -54,13 +62,19 @@ jobs:
       - run: make installcheck
       - if: ${{ failure() }}
         run: cat regression.diffs
+      # Homebrew Postgres does not enable TAP tests, so need to download
       - run: |
           brew install cpanm
           cpanm --notest IPC::Run
-          wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_10.tar.gz
-          tar xf REL_14_10.tar.gz
-      - run: make prove_installcheck PROVE_FLAGS="-I ./postgres-REL_14_10/src/test/perl" PERL5LIB="/Users/runner/perl5/lib/perl5"
-      - run: make clean && /usr/local/opt/llvm@15/bin/scan-build --status-bugs make
+          wget -q https://github.com/postgres/postgres/archive/refs/tags/$TAG.tar.gz
+          tar xf $TAG.tar.gz
+          mv postgres-$TAG postgres
+        env:
+          TAG: ${{ matrix.postgres == 16 && 'REL_16_2' || 'REL_14_11' }}
+      - run: make prove_installcheck PROVE_FLAGS="-I ./postgres/src/test/perl -I ./test/perl"
+        env:
+          PERL5LIB: /Users/runner/perl5/lib/perl5
+      - run: make clean && $(brew --prefix llvm@15)/bin/scan-build --status-bugs make
         env:
           PG_CFLAGS: -DUSE_ASSERT_CHECKING
   windows:

CHANGELOG.md

@@ -1,3 +1,13 @@
+## 0.7.0 (unreleased)
+
+- Added support for bit vectors to HNSW
+- Added `hamming_distance` function
+- Added `jaccard_distance` function
+- Added `quantize_binary` function
+- Added `subvector` function
+- Added concatenate operator for vectors
+- Updated comparison operators to support vectors with different dimensions
+
 ## 0.6.2 (2024-03-18)
 
 - Reduced lock contention with parallel HNSW index builds

Makefile

@@ -3,7 +3,7 @@ EXTVERSION = 0.6.2
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*.sql)
-OBJS = src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/vector.o
+OBJS = src/bitvector.o src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/vector.o
 HEADERS = src/vector.h
 
 TESTS = $(wildcard test/sql/*.sql)
@@ -12,7 +12,7 @@ REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)
 
 OPTFLAGS = -march=native
 
-# Mac ARM doesn't support -march=native
+# Mac ARM doesn't always support -march=native
 ifeq ($(shell uname -s), Darwin)
 	ifeq ($(shell uname -p), arm)
 		# no difference with -march=armv8.5-a

Makefile.win

@@ -1,10 +1,10 @@
 EXTENSION = vector
 EXTVERSION = 0.6.2
 
-OBJS = src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
+OBJS = src\bitvector.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
 HEADERS = src\vector.h
 
-REGRESS = btree cast copy functions input ivfflat_cosine ivfflat_ip ivfflat_l2 ivfflat_options ivfflat_unlogged
+REGRESS = bit_functions btree cast copy hnsw_cosine hnsw_hamming hnsw_ip hnsw_jaccard hnsw_l2 hnsw_options hnsw_unlogged ivfflat_cosine ivfflat_ip ivfflat_l2 ivfflat_options ivfflat_unlogged vector_functions vector_input
 REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)
 
 # For /arch flags

README.md

@@ -221,7 +221,19 @@ Cosine distance
 CREATE INDEX ON items USING hnsw (embedding vector_cosine_ops);
 ```
 
-Vectors with up to 2,000 dimensions can be indexed.
+Hamming distance - unreleased
+
+```sql
+CREATE INDEX ON items USING hnsw (embedding bit_hamming_ops);
+```
+
+Jaccard distance - unreleased
+
+```sql
+CREATE INDEX ON items USING hnsw (embedding bit_jaccard_ops);
+```
+
+Vectors with up to 2,000 dimensions can be indexed, or bit vectors with up to 64,000 dimensions.
 
 ### Index Options
 
@@ -699,6 +711,9 @@ Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosi
 
 ## Reference
 
+- [Vector](#vector-type)
+- [Bit](#bit-type)
+
 ### Vector Type
 
 Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a single-precision floating-point number (like the `real` type in Postgres), and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 16,000 dimensions.
@@ -710,6 +725,7 @@ Operator | Description | Added
 \+ | element-wise addition |
 \- | element-wise subtraction |
 \* | element-wise multiplication | 0.5.0
+\|\| | concatenate | unreleased
 <-> | Euclidean distance |
 <#> | negative inner product |
 <=> | cosine distance |
@@ -722,6 +738,8 @@ cosine_distance(vector, vector) → double precision | cosine distance |
 inner_product(vector, vector) → double precision | inner product |
 l2_distance(vector, vector) → double precision | Euclidean distance |
 l1_distance(vector, vector) → double precision | taxicab distance | 0.5.0
+quantize_binary(vector) → bit | quantize | unreleased
+subvector(vector, integer, integer) → vector | subvector | unreleased
 vector_dims(vector) → integer | number of dimensions |
 vector_norm(vector) → double precision | Euclidean norm |
 
@@ -732,6 +750,24 @@ Function | Description | Added
 avg(vector) → vector | average |
 sum(vector) → vector | sum | 0.5.0
 
+### Bit Type
+
+Each bit vector takes `dimensions / 8 + (5 or 8)` bytes of storage. See the [Postgres docs](https://www.postgresql.org/docs/current/datatype-bit.html) for more info.
+
+### Bit Operators
+
+Operator | Description | Added
+--- | --- | ---
+<~> | Hamming distance | unreleased
+<%> | Jaccard distance | unreleased
+
+### Bit Functions
+
+Function | Description | Added
+--- | --- | ---
+hamming_distance(bit, bit) → double precision | Hamming distance | unreleased
+jaccard_distance(bit, bit) → double precision | Jaccard distance | unreleased
+
 ## Installation Notes - Linux and Mac
 
 ### Postgres Location

sql/vector--0.6.2--0.7.0.sql

@@ -0,0 +1,41 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.7.0'" to load this file. \quit
+
+CREATE FUNCTION quantize_binary(vector) RETURNS bit
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION subvector(vector, int, int) RETURNS vector
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION vector_concat(vector, vector) RETURNS vector
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE OPERATOR || (
+	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_concat
+);
+
+CREATE FUNCTION hamming_distance(bit, bit) RETURNS float8
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION jaccard_distance(bit, bit) RETURNS float8
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE OPERATOR <~> (
+	LEFTARG = bit, RIGHTARG = bit, PROCEDURE = hamming_distance,
+	COMMUTATOR = '<~>'
+);
+
+CREATE OPERATOR <%> (
+	LEFTARG = bit, RIGHTARG = bit, PROCEDURE = jaccard_distance,
+	COMMUTATOR = '<%>'
+);
+
+CREATE OPERATOR CLASS bit_hamming_ops
+	FOR TYPE bit USING hnsw AS
+	OPERATOR 1 <~> (bit, bit) FOR ORDER BY float_ops,
+	FUNCTION 1 hamming_distance(bit, bit);
+
+CREATE OPERATOR CLASS bit_jaccard_ops
+	FOR TYPE bit USING hnsw AS
+	OPERATOR 1 <%> (bit, bit) FOR ORDER BY float_ops,
+	FUNCTION 1 jaccard_distance(bit, bit);

sql/vector.sql

@@ -58,6 +58,12 @@ CREATE FUNCTION vector_sub(vector, vector) RETURNS vector
 CREATE FUNCTION vector_mul(vector, vector) RETURNS vector
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
+CREATE FUNCTION quantize_binary(vector) RETURNS bit
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION subvector(vector, int, int) RETURNS vector
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
 -- vector private functions
 
 CREATE FUNCTION vector_lt(vector, vector) RETURNS bool
@@ -99,6 +105,9 @@ CREATE FUNCTION vector_avg(double precision[]) RETURNS vector
 CREATE FUNCTION vector_combine(double precision[], double precision[]) RETURNS double precision[]
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
+CREATE FUNCTION vector_concat(vector, vector) RETURNS vector
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
 -- vector aggregates
 
 CREATE AGGREGATE avg(vector) (
@@ -188,6 +197,10 @@ CREATE OPERATOR * (
 	COMMUTATOR = *
 );
 
+CREATE OPERATOR || (
+	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_concat
+);
+
 CREATE OPERATOR < (
 	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_lt,
 	COMMUTATOR = > , NEGATOR = >= ,
@@ -287,3 +300,31 @@ CREATE OPERATOR CLASS vector_cosine_ops
 	OPERATOR 1 <=> (vector, vector) FOR ORDER BY float_ops,
 	FUNCTION 1 vector_negative_inner_product(vector, vector),
 	FUNCTION 2 vector_norm(vector);
+
+-- bit functions
+
+CREATE FUNCTION hamming_distance(bit, bit) RETURNS float8
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION jaccard_distance(bit, bit) RETURNS float8
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE OPERATOR <~> (
+	LEFTARG = bit, RIGHTARG = bit, PROCEDURE = hamming_distance,
+	COMMUTATOR = '<~>'
+);
+
+CREATE OPERATOR <%> (
+	LEFTARG = bit, RIGHTARG = bit, PROCEDURE = jaccard_distance,
+	COMMUTATOR = '<%>'
+);
+
+CREATE OPERATOR CLASS bit_hamming_ops
+	FOR TYPE bit USING hnsw AS
+	OPERATOR 1 <~> (bit, bit) FOR ORDER BY float_ops,
+	FUNCTION 1 hamming_distance(bit, bit);
+
+CREATE OPERATOR CLASS bit_jaccard_ops
+	FOR TYPE bit USING hnsw AS
+	OPERATOR 1 <%> (bit, bit) FOR ORDER BY float_ops,
+	FUNCTION 1 jaccard_distance(bit, bit);

src/bitvector.c

@@ -0,0 +1,90 @@
+#include "postgres.h"
+
+#include "bitvector.h"
+#include "port/pg_bitutils.h"
+#include "utils/varbit.h"
+
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
+
+/*
+ * Allocate and initialize a new bit vector
+ */
+VarBit *
+InitBitVector(int dim)
+{
+	VarBit	   *result;
+	int			size;
+
+	size = VARBITTOTALLEN(dim);
+	result = (VarBit *) palloc0(size);
+	SET_VARSIZE(result, size);
+	VARBITLEN(result) = dim;
+
+	return result;
+}
+
+/*
+ * Ensure same dimensions
+ */
+static inline void
+CheckDims(VarBit *a, VarBit *b)
+{
+	if (VARBITLEN(a) != VARBITLEN(b))
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("different bit lengths %u and %u", VARBITLEN(a), VARBITLEN(b))));
+}
+
+/*
+ * Get the Hamming distance between two bit vectors
+ */
+PGDLLEXPORT PG_FUNCTION_INFO_V1(hamming_distance);
+Datum
+hamming_distance(PG_FUNCTION_ARGS)
+{
+	VarBit	   *a = PG_GETARG_VARBIT_P(0);
+	VarBit	   *b = PG_GETARG_VARBIT_P(1);
+	unsigned char *ax = VARBITS(a);
+	unsigned char *bx = VARBITS(b);
+	uint64		distance = 0;
+
+	CheckDims(a, b);
+
+	/* TODO Improve performance */
+	for (uint32 i = 0; i < VARBITBYTES(a); i++)
+		distance += pg_number_of_ones[ax[i] ^ bx[i]];
+
+	PG_RETURN_FLOAT8((double) distance);
+}
+
+/*
+ * Get the Jaccard distance between two bit vectors
+ */
+PGDLLEXPORT PG_FUNCTION_INFO_V1(jaccard_distance);
+Datum
+jaccard_distance(PG_FUNCTION_ARGS)
+{
+	VarBit	   *a = PG_GETARG_VARBIT_P(0);
+	VarBit	   *b = PG_GETARG_VARBIT_P(1);
+	unsigned char *ax = VARBITS(a);
+	unsigned char *bx = VARBITS(b);
+	uint64		ab = 0;
+	uint64		aa;
+	uint64		bb;
+
+	CheckDims(a, b);
+
+	/* TODO Improve performance */
+	for (uint32 i = 0; i < VARBITBYTES(a); i++)
+		ab += pg_number_of_ones[ax[i] & bx[i]];
+
+	if (ab == 0)
+		PG_RETURN_FLOAT8(1);
+
+	aa = pg_popcount((char *) ax, VARBITBYTES(a));
+	bb = pg_popcount((char *) bx, VARBITBYTES(b));
+
+	PG_RETURN_FLOAT8(1 - (ab / ((double) (aa + bb - ab))));
+}

src/bitvector.h

@@ -0,0 +1,8 @@
+#ifndef BITVECTOR_H
+#define BITVECTOR_H
+
+#include "utils/varbit.h"
+
+VarBit	   *InitBitVector(int dim);
+
+#endif

src/hnsw.h

@@ -57,7 +57,8 @@
 
 typedef enum HnswType
 {
-	HNSW_TYPE_VECTOR
+	HNSW_TYPE_VECTOR,
+	HNSW_TYPE_BIT
 }			HnswType;
 
 /* Build phases */

src/hnswbuild.c

@@ -44,6 +44,7 @@
 #include "access/xact.h"
 #include "access/xloginsert.h"
 #include "catalog/index.h"
+#include "catalog/pg_type_d.h"
 #include "commands/progress.h"
 #include "hnsw.h"
 #include "miscadmin.h"
@@ -665,13 +666,27 @@ HnswSharedMemoryAlloc(Size size, void *state)
 	return chunk;
 }
 
+/*
+ * Get max dimensions
+ */
+static int
+GetMaxDimensions(HnswType type)
+{
+	int			maxDimensions = HNSW_MAX_DIM;
+
+	if (type == HNSW_TYPE_BIT)
+		maxDimensions *= 32;
+
+	return maxDimensions;
+}
+
 /*
  * Initialize the build state
  */
 static void
 InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, IndexInfo *indexInfo, ForkNumber forkNum)
 {
-	int			maxDimensions = HNSW_MAX_DIM;
+	int			maxDimensions;
 
 	buildstate->heap = heap;
 	buildstate->index = index;
@@ -683,6 +698,8 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 	buildstate->efConstruction = HnswGetEfConstruction(index);
 	buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
 
+	maxDimensions = GetMaxDimensions(buildstate->type);
+
 	/* Require column to have dimensions to be indexed */
 	if (buildstate->dimensions < 0)
 		elog(ERROR, "column does not have dimensions");

src/hnswscan.c

@@ -1,6 +1,8 @@
 #include "postgres.h"
 
 #include "access/relscan.h"
+#include "bitvector.h"
+#include "catalog/pg_type_d.h"
 #include "hnsw.h"
 #include "pgstat.h"
 #include "storage/bufmgr.h"

src/hnswutils.c

@@ -3,6 +3,7 @@
 #include <math.h>
 
 #include "access/generic_xlog.h"
+#include "catalog/pg_type_d.h"
 #include "hnsw.h"
 #include "lib/pairingheap.h"
 #include "storage/bufmgr.h"
@@ -150,11 +151,16 @@ HnswOptionalProcInfo(Relation index, uint16 procnum)
 }
 
 /*
- * Get vector type
+ * Get type
  */
 HnswType
 HnswGetType(Relation index)
 {
+	Oid			typid = TupleDescAttr(index->rd_att, 0)->atttypid;
+
+	if (typid == BITOID || typid == VARBITOID)
+		return HNSW_TYPE_BIT;
+
 	return HNSW_TYPE_VECTOR;
 }
 

src/vector.c

@@ -2,6 +2,7 @@
 
 #include <math.h>
 
+#include "bitvector.h"
 #include "catalog/pg_type.h"
 #include "common/shortest_dec.h"
 #include "fmgr.h"
@@ -197,8 +198,6 @@ vector_in(PG_FUNCTION_ARGS)
 
 	while (pt != NULL && *stringEnd != ']')
 	{
-		float		val;
-
 		if (dim == VECTOR_MAX_DIM)
 			ereport(ERROR,
 					(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
@@ -214,23 +213,15 @@ vector_in(PG_FUNCTION_ARGS)
 					 errmsg("invalid input syntax for type vector: \"%s\"", lit)));
 
 		/* Use strtof like float4in to avoid a double-rounding problem */
-		errno = 0;
-		val = strtof(pt, &stringEnd);
+		x[dim] = strtof(pt, &stringEnd);
+		CheckElement(x[dim]);
+		dim++;
 
 		if (stringEnd == pt)
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
 					 errmsg("invalid input syntax for type vector: \"%s\"", lit)));
 
-		/* Check for range error like float4in */
-		if (errno == ERANGE && (val == 0 || isinf(val)))
-			ereport(ERROR,
-					(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-					 errmsg("\"%s\" is out of range for type vector", pt)));
-
-		CheckElement(val);
-		x[dim++] = val;
-
 		while (vector_isspace(*stringEnd))
 			stringEnd++;
 
@@ -285,6 +276,9 @@ vector_in(PG_FUNCTION_ARGS)
 	PG_RETURN_POINTER(result);
 }
 
+#define AppendChar(ptr, c) (*(ptr)++ = (c))
+#define AppendFloat(ptr, f) ((ptr) += float_to_shortest_decimal_bufn((f), (ptr)))
+
 /*
  * Convert internal representation to textual representation
  */
@@ -296,7 +290,6 @@ vector_out(PG_FUNCTION_ARGS)
 	int			dim = vector->dim;
 	char	   *buf;
 	char	   *ptr;
-	int			n;
 
 	/*
 	 * Need:
@@ -311,21 +304,17 @@ vector_out(PG_FUNCTION_ARGS)
 	buf = (char *) palloc(FLOAT_SHORTEST_DECIMAL_LEN * dim + 2);
 	ptr = buf;
 
-	*ptr = '[';
-	ptr++;
+	AppendChar(ptr, '[');
+
 	for (int i = 0; i < dim; i++)
 	{
 		if (i > 0)
-		{
-			*ptr = ',';
-			ptr++;
+			AppendChar(ptr, ',');
+
+		AppendFloat(ptr, vector->x[i]);
 	}
 
-		n = float_to_shortest_decimal_bufn(vector->x[i], ptr);
-		ptr += n;
-	}
-	*ptr = ']';
-	ptr++;
+	AppendChar(ptr, ']');
 	*ptr = '\0';
 
 	PG_FREE_IF_COPY(vector, 0);
@@ -870,6 +859,80 @@ vector_mul(PG_FUNCTION_ARGS)
 	PG_RETURN_POINTER(result);
 }
 
+/*
+ * Quantize a vector
+ */
+PGDLLEXPORT PG_FUNCTION_INFO_V1(quantize_binary);
+Datum
+quantize_binary(PG_FUNCTION_ARGS)
+{
+	Vector	   *a = PG_GETARG_VECTOR_P(0);
+	float	   *ax = a->x;
+	VarBit	   *result = InitBitVector(a->dim);
+	unsigned char *rx = VARBITS(result);
+
+	for (int i = 0; i < a->dim; i++)
+		rx[i / 8] |= (ax[i] > 0) << (7 - (i % 8));
+
+	PG_RETURN_VARBIT_P(result);
+}
+
+/*
+ * Get a subvector
+ */
+PGDLLEXPORT PG_FUNCTION_INFO_V1(subvector);
+Datum
+subvector(PG_FUNCTION_ARGS)
+{
+	Vector	   *a = PG_GETARG_VECTOR_P(0);
+	int32		start = PG_GETARG_INT32(1);
+	int32		count = PG_GETARG_INT32(2);
+	int32		end = start + count;
+	float	   *ax = a->x;
+	Vector	   *result;
+	int			dim;
+
+	/* Indexing starts at 1, like substring */
+	if (start < 1)
+		start = 1;
+
+	if (end > a->dim)
+		end = a->dim + 1;
+
+	dim = end - start;
+	CheckDim(dim);
+	result = InitVector(dim);
+
+	for (int i = 0; i < dim; i++)
+		result->x[i] = ax[start - 1 + i];
+
+	PG_RETURN_POINTER(result);
+}
+
+/*
+ * Concatenate vectors
+ */
+PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_concat);
+Datum
+vector_concat(PG_FUNCTION_ARGS)
+{
+	Vector	   *a = PG_GETARG_VECTOR_P(0);
+	Vector	   *b = PG_GETARG_VECTOR_P(1);
+	Vector	   *result;
+	int			dim = a->dim + b->dim;
+
+	CheckDim(dim);
+	result = InitVector(dim);
+
+	for (int i = 0; i < a->dim; i++)
+		result->x[i] = a->x[i];
+
+	for (int i = 0; i < b->dim; i++)
+		result->x[i + a->dim] = b->x[i];
+
+	PG_RETURN_POINTER(result);
+}
+
 /*
  * Internal helper to compare vectors
  */
@@ -907,9 +970,6 @@ vector_lt(PG_FUNCTION_ARGS)
 	Vector	   *a = PG_GETARG_VECTOR_P(0);
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 
-	/* TODO Remove in 0.7.0 */
-	CheckDims(a, b);
-
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) < 0);
 }
 
@@ -923,9 +983,6 @@ vector_le(PG_FUNCTION_ARGS)
 	Vector	   *a = PG_GETARG_VECTOR_P(0);
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 
-	/* TODO Remove in 0.7.0 */
-	CheckDims(a, b);
-
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) <= 0);
 }
 
@@ -939,9 +996,6 @@ vector_eq(PG_FUNCTION_ARGS)
 	Vector	   *a = PG_GETARG_VECTOR_P(0);
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 
-	/* TODO Remove in 0.7.0 */
-	CheckDims(a, b);
-
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) == 0);
 }
 
@@ -955,9 +1009,6 @@ vector_ne(PG_FUNCTION_ARGS)
 	Vector	   *a = PG_GETARG_VECTOR_P(0);
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 
-	/* TODO Remove in 0.7.0 */
-	CheckDims(a, b);
-
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) != 0);
 }
 
@@ -971,9 +1022,6 @@ vector_ge(PG_FUNCTION_ARGS)
 	Vector	   *a = PG_GETARG_VECTOR_P(0);
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 
-	/* TODO Remove in 0.7.0 */
-	CheckDims(a, b);
-
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) >= 0);
 }
 
@@ -987,9 +1035,6 @@ vector_gt(PG_FUNCTION_ARGS)
 	Vector	   *a = PG_GETARG_VECTOR_P(0);
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 
-	/* TODO Remove in 0.7.0 */
-	CheckDims(a, b);
-
 	PG_RETURN_BOOL(vector_cmp_internal(a, b) > 0);
 }
 

test/expected/bit_functions.out

@@ -0,0 +1,64 @@
+SELECT hamming_distance(B'111', B'111');
+ hamming_distance 
+------------------
+                0
+(1 row)
+
+SELECT hamming_distance(B'111', B'110');
+ hamming_distance 
+------------------
+                1
+(1 row)
+
+SELECT hamming_distance(B'111', B'100');
+ hamming_distance 
+------------------
+                2
+(1 row)
+
+SELECT hamming_distance(B'111', B'000');
+ hamming_distance 
+------------------
+                3
+(1 row)
+
+SELECT hamming_distance(B'111', B'00');
+ERROR:  different bit lengths 3 and 2
+SELECT jaccard_distance(B'1111', B'1111');
+ jaccard_distance 
+------------------
+                0
+(1 row)
+
+SELECT jaccard_distance(B'1111', B'1110');
+ jaccard_distance 
+------------------
+             0.25
+(1 row)
+
+SELECT jaccard_distance(B'1111', B'1100');
+ jaccard_distance 
+------------------
+              0.5
+(1 row)
+
+SELECT jaccard_distance(B'1111', B'1000');
+ jaccard_distance 
+------------------
+             0.75
+(1 row)
+
+SELECT jaccard_distance(B'1111', B'0000');
+ jaccard_distance 
+------------------
+                1
+(1 row)
+
+SELECT jaccard_distance(B'1100', B'1000');
+ jaccard_distance 
+------------------
+              0.5
+(1 row)
+
+SELECT jaccard_distance(B'1111', B'000');
+ERROR:  different bit lengths 4 and 3

test/expected/hnsw_hamming.out

@@ -0,0 +1,21 @@
+SET enable_seqscan = off;
+CREATE TABLE t (val bit(3));
+INSERT INTO t (val) VALUES (B'000'), (B'100'), (B'111'), (NULL);
+CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
+INSERT INTO t (val) VALUES (B'110');
+SELECT * FROM t ORDER BY val <~> B'111';
+ val 
+-----
+ 111
+ 110
+ 100
+ 000
+(4 rows)
+
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
+ count 
+-------
+     4
+(1 row)
+
+DROP TABLE t;

test/expected/hnsw_jaccard.out

@@ -0,0 +1,21 @@
+SET enable_seqscan = off;
+CREATE TABLE t (val bit(4));
+INSERT INTO t (val) VALUES (B'0000'), (B'1100'), (B'1111'), (NULL);
+CREATE INDEX ON t USING hnsw (val bit_jaccard_ops);
+INSERT INTO t (val) VALUES (B'1110');
+SELECT * FROM t ORDER BY val <%> B'1111';
+ val  
+------
+ 1111
+ 1110
+ 1100
+ 0000
+(4 rows)
+
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <%> (SELECT NULL::bit)) t2;
+ count 
+-------
+     4
+(1 row)
+
+DROP TABLE t;

test/expected/vector_functions.out

@@ -24,6 +24,38 @@ SELECT '[1e37]'::vector * '[1e37]';
 ERROR:  value out of range: overflow
 SELECT '[1e-37]'::vector * '[1e-37]';
 ERROR:  value out of range: underflow
+SELECT '[1,2,3]'::vector || '[4,5]'::vector;
+  ?column?   
+-------------
+ [1,2,3,4,5]
+(1 row)
+
+SELECT array_fill(0, ARRAY[16000])::vector || '[1]'::vector;
+ERROR:  vector cannot have more than 16000 dimensions
+SELECT '[1,2,3]'::vector < '[1,2,3]';
+ ?column? 
+----------
+ f
+(1 row)
+
+SELECT '[1,2,3]'::vector < '[1,2]';
+ ?column? 
+----------
+ f
+(1 row)
+
+SELECT '[1,2,3]'::vector <= '[1,2,3]';
+ ?column? 
+----------
+ t
+(1 row)
+
+SELECT '[1,2,3]'::vector <= '[1,2]';
+ ?column? 
+----------
+ f
+(1 row)
+
 SELECT '[1,2,3]'::vector = '[1,2,3]';
  ?column? 
 ----------
@@ -31,7 +63,47 @@ SELECT '[1,2,3]'::vector = '[1,2,3]';
 (1 row)
 
 SELECT '[1,2,3]'::vector = '[1,2]';
-ERROR:  different vector dimensions 3 and 2
+ ?column? 
+----------
+ f
+(1 row)
+
+SELECT '[1,2,3]'::vector != '[1,2,3]';
+ ?column? 
+----------
+ f
+(1 row)
+
+SELECT '[1,2,3]'::vector != '[1,2]';
+ ?column? 
+----------
+ t
+(1 row)
+
+SELECT '[1,2,3]'::vector >= '[1,2,3]';
+ ?column? 
+----------
+ t
+(1 row)
+
+SELECT '[1,2,3]'::vector >= '[1,2]';
+ ?column? 
+----------
+ t
+(1 row)
+
+SELECT '[1,2,3]'::vector > '[1,2,3]';
+ ?column? 
+----------
+ f
+(1 row)
+
+SELECT '[1,2,3]'::vector > '[1,2]';
+ ?column? 
+----------
+ t
+(1 row)
+
 SELECT vector_cmp('[1,2,3]', '[1,2,3]');
  vector_cmp 
 ------------
@@ -104,110 +176,152 @@ SELECT vector_norm('[3e37,4e37]')::real;
        5e+37
 (1 row)
 
-SELECT l2_distance('[0,0]', '[3,4]');
+SELECT l2_distance('[0,0]'::vector, '[3,4]');
  l2_distance 
 -------------
            5
 (1 row)
 
-SELECT l2_distance('[0,0]', '[0,1]');
+SELECT l2_distance('[0,0]'::vector, '[0,1]');
  l2_distance 
 -------------
            1
 (1 row)
 
-SELECT l2_distance('[1,2]', '[3]');
+SELECT l2_distance('[1,2]'::vector, '[3]');
 ERROR:  different vector dimensions 2 and 1
-SELECT l2_distance('[3e38]', '[-3e38]');
+SELECT l2_distance('[3e38]'::vector, '[-3e38]');
  l2_distance 
 -------------
     Infinity
 (1 row)
 
-SELECT inner_product('[1,2]', '[3,4]');
+SELECT inner_product('[1,2]'::vector, '[3,4]');
  inner_product 
 ---------------
             11
 (1 row)
 
-SELECT inner_product('[1,2]', '[3]');
+SELECT inner_product('[1,2]'::vector, '[3]');
 ERROR:  different vector dimensions 2 and 1
-SELECT inner_product('[3e38]', '[3e38]');
+SELECT inner_product('[3e38]'::vector, '[3e38]');
  inner_product 
 ---------------
       Infinity
 (1 row)
 
-SELECT cosine_distance('[1,2]', '[2,4]');
+SELECT cosine_distance('[1,2]'::vector, '[2,4]');
  cosine_distance 
 -----------------
                0
 (1 row)
 
-SELECT cosine_distance('[1,2]', '[0,0]');
+SELECT cosine_distance('[1,2]'::vector, '[0,0]');
  cosine_distance 
 -----------------
              NaN
 (1 row)
 
-SELECT cosine_distance('[1,1]', '[1,1]');
+SELECT cosine_distance('[1,1]'::vector, '[1,1]');
  cosine_distance 
 -----------------
                0
 (1 row)
 
-SELECT cosine_distance('[1,0]', '[0,2]');
+SELECT cosine_distance('[1,0]'::vector, '[0,2]');
  cosine_distance 
 -----------------
                1
 (1 row)
 
-SELECT cosine_distance('[1,1]', '[-1,-1]');
+SELECT cosine_distance('[1,1]'::vector, '[-1,-1]');
  cosine_distance 
 -----------------
                2
 (1 row)
 
-SELECT cosine_distance('[1,2]', '[3]');
+SELECT cosine_distance('[1,2]'::vector, '[3]');
 ERROR:  different vector dimensions 2 and 1
-SELECT cosine_distance('[1,1]', '[1.1,1.1]');
+SELECT cosine_distance('[1,1]'::vector, '[1.1,1.1]');
  cosine_distance 
 -----------------
                0
 (1 row)
 
-SELECT cosine_distance('[1,1]', '[-1.1,-1.1]');
+SELECT cosine_distance('[1,1]'::vector, '[-1.1,-1.1]');
  cosine_distance 
 -----------------
                2
 (1 row)
 
-SELECT cosine_distance('[3e38]', '[3e38]');
+SELECT cosine_distance('[3e38]'::vector, '[3e38]');
  cosine_distance 
 -----------------
              NaN
 (1 row)
 
-SELECT l1_distance('[0,0]', '[3,4]');
+SELECT l1_distance('[0,0]'::vector, '[3,4]');
  l1_distance 
 -------------
            7
 (1 row)
 
-SELECT l1_distance('[0,0]', '[0,1]');
+SELECT l1_distance('[0,0]'::vector, '[0,1]');
  l1_distance 
 -------------
            1
 (1 row)
 
-SELECT l1_distance('[1,2]', '[3]');
+SELECT l1_distance('[1,2]'::vector, '[3]');
 ERROR:  different vector dimensions 2 and 1
-SELECT l1_distance('[3e38]', '[-3e38]');
+SELECT l1_distance('[3e38]'::vector, '[-3e38]');
  l1_distance 
 -------------
     Infinity
 (1 row)
 
+SELECT quantize_binary('[1,0,-1]');
+ quantize_binary 
+-----------------
+ 100
+(1 row)
+
+SELECT quantize_binary('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]');
+ quantize_binary 
+-----------------
+ 01001110101
+(1 row)
+
+SELECT subvector('[1,2,3,4,5]', 1, 3);
+ subvector 
+-----------
+ [1,2,3]
+(1 row)
+
+SELECT subvector('[1,2,3,4,5]', 3, 2);
+ subvector 
+-----------
+ [3,4]
+(1 row)
+
+SELECT subvector('[1,2,3,4,5]', -1, 3);
+ subvector 
+-----------
+ [1]
+(1 row)
+
+SELECT subvector('[1,2,3,4,5]', 3, 9);
+ subvector 
+-----------
+ [3,4,5]
+(1 row)
+
+SELECT subvector('[1,2,3,4,5]', 1, 0);
+ERROR:  vector must have at least 1 dimension
+SELECT subvector('[1,2,3,4,5]', 3, -1);
+ERROR:  vector must have at least 1 dimension
+SELECT subvector('[1,2,3,4,5]', -1, 2);
+ERROR:  vector must have at least 1 dimension
 SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
     avg    
 -----------

test/expected/vector_input.out

@@ -63,21 +63,25 @@ SELECT '[1.5e-38,-1.5e-38]'::vector;
 (1 row)
 
 SELECT '[4e38,1]'::vector;
-ERROR:  "4e38" is out of range for type vector
+ERROR:  infinite value not allowed in vector
 LINE 1: SELECT '[4e38,1]'::vector;
                ^
 SELECT '[-4e38,1]'::vector;
-ERROR:  "-4e38" is out of range for type vector
+ERROR:  infinite value not allowed in vector
 LINE 1: SELECT '[-4e38,1]'::vector;
                ^
 SELECT '[1e-46,1]'::vector;
-ERROR:  "1e-46" is out of range for type vector
-LINE 1: SELECT '[1e-46,1]'::vector;
-               ^
+ vector 
+--------
+ [0,1]
+(1 row)
+
 SELECT '[-1e-46,1]'::vector;
-ERROR:  "-1e-46" is out of range for type vector
-LINE 1: SELECT '[-1e-46,1]'::vector;
-               ^
+ vector 
+--------
+ [-0,1]
+(1 row)
+
 SELECT '[1,2,3'::vector;
 ERROR:  malformed vector literal: "[1,2,3"
 LINE 1: SELECT '[1,2,3'::vector;

test/sql/bit_functions.sql

@@ -0,0 +1,13 @@
+SELECT hamming_distance(B'111', B'111');
+SELECT hamming_distance(B'111', B'110');
+SELECT hamming_distance(B'111', B'100');
+SELECT hamming_distance(B'111', B'000');
+SELECT hamming_distance(B'111', B'00');
+
+SELECT jaccard_distance(B'1111', B'1111');
+SELECT jaccard_distance(B'1111', B'1110');
+SELECT jaccard_distance(B'1111', B'1100');
+SELECT jaccard_distance(B'1111', B'1000');
+SELECT jaccard_distance(B'1111', B'0000');
+SELECT jaccard_distance(B'1100', B'1000');
+SELECT jaccard_distance(B'1111', B'000');

test/sql/functions.sql

@@ -1,62 +0,0 @@
-SELECT '[1,2,3]'::vector + '[4,5,6]';
-SELECT '[3e38]'::vector + '[3e38]';
-SELECT '[1,2,3]'::vector - '[4,5,6]';
-SELECT '[-3e38]'::vector - '[3e38]';
-SELECT '[1,2,3]'::vector * '[4,5,6]';
-SELECT '[1e37]'::vector * '[1e37]';
-SELECT '[1e-37]'::vector * '[1e-37]';
-
-SELECT '[1,2,3]'::vector = '[1,2,3]';
-SELECT '[1,2,3]'::vector = '[1,2]';
-
-SELECT vector_cmp('[1,2,3]', '[1,2,3]');
-SELECT vector_cmp('[1,2,3]', '[0,0,0]');
-SELECT vector_cmp('[0,0,0]', '[1,2,3]');
-SELECT vector_cmp('[1,2]', '[1,2,3]');
-SELECT vector_cmp('[1,2,3]', '[1,2]');
-SELECT vector_cmp('[1,2]', '[2,3,4]');
-SELECT vector_cmp('[2,3]', '[1,2,3]');
-
-SELECT vector_dims('[1,2,3]');
-
-SELECT round(vector_norm('[1,1]')::numeric, 5);
-SELECT vector_norm('[3,4]');
-SELECT vector_norm('[0,1]');
-SELECT vector_norm('[3e37,4e37]')::real;
-
-SELECT l2_distance('[0,0]', '[3,4]');
-SELECT l2_distance('[0,0]', '[0,1]');
-SELECT l2_distance('[1,2]', '[3]');
-SELECT l2_distance('[3e38]', '[-3e38]');
-
-SELECT inner_product('[1,2]', '[3,4]');
-SELECT inner_product('[1,2]', '[3]');
-SELECT inner_product('[3e38]', '[3e38]');
-
-SELECT cosine_distance('[1,2]', '[2,4]');
-SELECT cosine_distance('[1,2]', '[0,0]');
-SELECT cosine_distance('[1,1]', '[1,1]');
-SELECT cosine_distance('[1,0]', '[0,2]');
-SELECT cosine_distance('[1,1]', '[-1,-1]');
-SELECT cosine_distance('[1,2]', '[3]');
-SELECT cosine_distance('[1,1]', '[1.1,1.1]');
-SELECT cosine_distance('[1,1]', '[-1.1,-1.1]');
-SELECT cosine_distance('[3e38]', '[3e38]');
-
-SELECT l1_distance('[0,0]', '[3,4]');
-SELECT l1_distance('[0,0]', '[0,1]');
-SELECT l1_distance('[1,2]', '[3]');
-SELECT l1_distance('[3e38]', '[-3e38]');
-
-SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
-SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
-SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
-SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
-SELECT avg(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
-SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;
-
-SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
-SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
-SELECT sum(v) FROM unnest(ARRAY[]::vector[]) v;
-SELECT sum(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
-SELECT sum(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;

test/sql/hnsw_hamming.sql

@@ -0,0 +1,12 @@
+SET enable_seqscan = off;
+
+CREATE TABLE t (val bit(3));
+INSERT INTO t (val) VALUES (B'000'), (B'100'), (B'111'), (NULL);
+CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
+
+INSERT INTO t (val) VALUES (B'110');
+
+SELECT * FROM t ORDER BY val <~> B'111';
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
+
+DROP TABLE t;

test/sql/hnsw_jaccard.sql

@@ -0,0 +1,12 @@
+SET enable_seqscan = off;
+
+CREATE TABLE t (val bit(4));
+INSERT INTO t (val) VALUES (B'0000'), (B'1100'), (B'1111'), (NULL);
+CREATE INDEX ON t USING hnsw (val bit_jaccard_ops);
+
+INSERT INTO t (val) VALUES (B'1110');
+
+SELECT * FROM t ORDER BY val <%> B'1111';
+SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <%> (SELECT NULL::bit)) t2;
+
+DROP TABLE t;

test/sql/vector_functions.sql

@@ -0,0 +1,86 @@
+SELECT '[1,2,3]'::vector + '[4,5,6]';
+SELECT '[3e38]'::vector + '[3e38]';
+SELECT '[1,2,3]'::vector - '[4,5,6]';
+SELECT '[-3e38]'::vector - '[3e38]';
+SELECT '[1,2,3]'::vector * '[4,5,6]';
+SELECT '[1e37]'::vector * '[1e37]';
+SELECT '[1e-37]'::vector * '[1e-37]';
+
+SELECT '[1,2,3]'::vector || '[4,5]'::vector;
+SELECT array_fill(0, ARRAY[16000])::vector || '[1]'::vector;
+
+SELECT '[1,2,3]'::vector < '[1,2,3]';
+SELECT '[1,2,3]'::vector < '[1,2]';
+SELECT '[1,2,3]'::vector <= '[1,2,3]';
+SELECT '[1,2,3]'::vector <= '[1,2]';
+SELECT '[1,2,3]'::vector = '[1,2,3]';
+SELECT '[1,2,3]'::vector = '[1,2]';
+SELECT '[1,2,3]'::vector != '[1,2,3]';
+SELECT '[1,2,3]'::vector != '[1,2]';
+SELECT '[1,2,3]'::vector >= '[1,2,3]';
+SELECT '[1,2,3]'::vector >= '[1,2]';
+SELECT '[1,2,3]'::vector > '[1,2,3]';
+SELECT '[1,2,3]'::vector > '[1,2]';
+
+SELECT vector_cmp('[1,2,3]', '[1,2,3]');
+SELECT vector_cmp('[1,2,3]', '[0,0,0]');
+SELECT vector_cmp('[0,0,0]', '[1,2,3]');
+SELECT vector_cmp('[1,2]', '[1,2,3]');
+SELECT vector_cmp('[1,2,3]', '[1,2]');
+SELECT vector_cmp('[1,2]', '[2,3,4]');
+SELECT vector_cmp('[2,3]', '[1,2,3]');
+
+SELECT vector_dims('[1,2,3]');
+
+SELECT round(vector_norm('[1,1]')::numeric, 5);
+SELECT vector_norm('[3,4]');
+SELECT vector_norm('[0,1]');
+SELECT vector_norm('[3e37,4e37]')::real;
+
+SELECT l2_distance('[0,0]'::vector, '[3,4]');
+SELECT l2_distance('[0,0]'::vector, '[0,1]');
+SELECT l2_distance('[1,2]'::vector, '[3]');
+SELECT l2_distance('[3e38]'::vector, '[-3e38]');
+
+SELECT inner_product('[1,2]'::vector, '[3,4]');
+SELECT inner_product('[1,2]'::vector, '[3]');
+SELECT inner_product('[3e38]'::vector, '[3e38]');
+
+SELECT cosine_distance('[1,2]'::vector, '[2,4]');
+SELECT cosine_distance('[1,2]'::vector, '[0,0]');
+SELECT cosine_distance('[1,1]'::vector, '[1,1]');
+SELECT cosine_distance('[1,0]'::vector, '[0,2]');
+SELECT cosine_distance('[1,1]'::vector, '[-1,-1]');
+SELECT cosine_distance('[1,2]'::vector, '[3]');
+SELECT cosine_distance('[1,1]'::vector, '[1.1,1.1]');
+SELECT cosine_distance('[1,1]'::vector, '[-1.1,-1.1]');
+SELECT cosine_distance('[3e38]'::vector, '[3e38]');
+
+SELECT l1_distance('[0,0]'::vector, '[3,4]');
+SELECT l1_distance('[0,0]'::vector, '[0,1]');
+SELECT l1_distance('[1,2]'::vector, '[3]');
+SELECT l1_distance('[3e38]'::vector, '[-3e38]');
+
+SELECT quantize_binary('[1,0,-1]');
+SELECT quantize_binary('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]');
+
+SELECT subvector('[1,2,3,4,5]', 1, 3);
+SELECT subvector('[1,2,3,4,5]', 3, 2);
+SELECT subvector('[1,2,3,4,5]', -1, 3);
+SELECT subvector('[1,2,3,4,5]', 3, 9);
+SELECT subvector('[1,2,3,4,5]', 1, 0);
+SELECT subvector('[1,2,3,4,5]', 3, -1);
+SELECT subvector('[1,2,3,4,5]', -1, 2);
+
+SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
+SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
+SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
+SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
+SELECT avg(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
+SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;
+
+SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
+SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
+SELECT sum(v) FROM unnest(ARRAY[]::vector[]) v;
+SELECT sum(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
+SELECT sum(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;

test/t/020_hnsw_bit_build_recall.pl

@@ -0,0 +1,137 @@
+use strict;
+use warnings;
+use PostgresNode;
+use TestLib;
+use Test::More;
+
+my $node;
+my @queries = ();
+my @expected;
+my $limit = 20;
+my $dim = 52;
+my $max = 2**$dim;
+
+sub test_recall
+{
+	my ($min, $operator) = @_;
+	my $correct = 0;
+	my $total = 0;
+
+	my $explain = $node->safe_psql("postgres", qq(
+		SET enable_seqscan = off;
+		SET hnsw.ef_search = 100;
+		EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v $operator $queries[0] LIMIT $limit;
+	));
+	like($explain, qr/Index Scan/);
+
+	for my $i (0 .. $#queries)
+	{
+		my $actual = $node->safe_psql("postgres", qq(
+			SET enable_seqscan = off;
+			SET hnsw.ef_search = 100;
+			SELECT i FROM tst ORDER BY v $operator $queries[$i] LIMIT $limit;
+		));
+		my @actual_ids = split("\n", $actual);
+
+		my @expected_ids = split("\n", $expected[$i]);
+		my %expected_set = map { $_ => 1 } @expected_ids;
+
+		foreach (@actual_ids)
+		{
+			if (exists($expected_set{$_}))
+			{
+				$correct++;
+			}
+		}
+
+		$total += $limit;
+	}
+
+	cmp_ok($correct / $total, ">=", $min, $operator);
+}
+
+# Initialize node
+$node = get_new_node('node');
+$node->init;
+$node->start;
+
+# Create table
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v bit($dim));");
+$node->safe_psql("postgres",
+	"INSERT INTO tst SELECT i, (random() * $max)::bigint::bit($dim) FROM generate_series(1, 10000) i;"
+);
+
+# Generate queries
+for (1 .. 20)
+{
+	my $r = int(rand() * $max);
+	push(@queries, "${r}::bigint::bit($dim)");
+}
+
+# Check each index type
+my @operators = ("<~>", "<\%>");
+my @opclasses = ("bit_hamming_ops", "bit_jaccard_ops");
+
+for my $i (0 .. $#operators)
+{
+	my $operator = $operators[$i];
+	my $opclass = $opclasses[$i];
+
+	# Get exact results
+	@expected = ();
+	foreach (@queries)
+	{
+		# Handle ties
+		my $res = $node->safe_psql("postgres", qq(
+			WITH top AS (
+				SELECT v $operator $_ AS distance FROM tst ORDER BY v $operator $_ LIMIT $limit
+			)
+			SELECT i FROM tst WHERE (v $operator $_) <= (SELECT MAX(distance) FROM top)
+		));
+		push(@expected, $res);
+	}
+
+	# Build index serially
+	$node->safe_psql("postgres", qq(
+		SET max_parallel_maintenance_workers = 0;
+		CREATE INDEX idx ON tst USING hnsw (v $opclass);
+	));
+
+	# Test approximate results
+	my $min = $operator eq "<\%>" ? 0.96 : 0.98;
+	test_recall($min, $operator);
+
+	$node->safe_psql("postgres", "DROP INDEX idx;");
+
+	# Build index in parallel in memory
+	my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
+		SET client_min_messages = DEBUG;
+		SET min_parallel_table_scan_size = 1;
+		CREATE INDEX idx ON tst USING hnsw (v $opclass);
+	));
+	is($ret, 0, $stderr);
+	like($stderr, qr/using \d+ parallel workers/);
+
+	# Test approximate results
+	test_recall($min, $operator);
+
+	$node->safe_psql("postgres", "DROP INDEX idx;");
+
+	# Build index in parallel on disk
+	# Set parallel_workers on table to use workers with low maintenance_work_mem
+	($ret, $stdout, $stderr) = $node->psql("postgres", qq(
+		ALTER TABLE tst SET (parallel_workers = 2);
+		SET client_min_messages = DEBUG;
+		SET maintenance_work_mem = '4MB';
+		CREATE INDEX idx ON tst USING hnsw (v $opclass);
+		ALTER TABLE tst RESET (parallel_workers);
+	));
+	is($ret, 0, $stderr);
+	like($stderr, qr/using \d+ parallel workers/);
+	like($stderr, qr/hnsw graph no longer fits into maintenance_work_mem/);
+
+	$node->safe_psql("postgres", "DROP INDEX idx;");
+}
+
+done_testing();