Added HnswGetElementTupleSize method

Previously, the cost estimation formula for a HNSW index scan utilized
a methodology that only factored in the entry level for an HNSW scan
and the "m" index parameter, which reflects the number of tuples (or
vectors) to scan at each step of a HNSW graph traversal. While this
would bias the PostgreSQL query planner to choose an HNSW index scan
over other available paths, this could lead to potential suboptimal
index selection, for example, choosing to use a HNSW index instead of
an available B-tree index that has better selectivity.

The number of tuples scanned during HNSW graph traversal is principally
influenced by these factors:

 * The number of tuples stored in the index
 * `m` - the number of tuples that are scanned in each step of the graph
   traversal
 * `hnsw.ef_search` - which influences the total number of steps it
   takes for the scan to converge on the approximated nearest neighbors

Through testing different source models for vectors, we also observed
that the correlation of vectors in mdoels would impact this convergence.
For this first iteration, we've opted to hardcode a constant scaling
factor and set it to `0.55`, though a future commit may turn this into
a configurable parameter.

The high-level formula for estimating the cost of a HNSW index scan is
as such:

```
(entryLevel * m) + (layer0TuplesMax * layer0Selectivity)
```

where

- `(entryLevel * m)` is the lower bound of tuples to scan, as it
accounts for the graph traversal to layer 0 (L0). (L1 and above has an ef=1)
- `layer0TuplesMax` is an estimate of the maximum number of tuples to
scan at L0. This accounts for tuples that may end up being discarded due
to them already being visited. Testing shows that the number of steps
until converge is similar to the value of `hnsw.ef_search`, thus we can
estimate tuples max at `hnsw.ef_search * m * 2`
- `layer0Selectivity` - estimates the percentage of tuples that will
actually be scanned during the index traversal, multipled by the scaling
factor

In addition to the `m` build parameter and `hsnw.ef_search`, costs
estimates can be influenced by standard PostgreSQL costing parameters,
though adjusting those (e.g. `random_page_cost`) should be done with
care.

Co-authored-by: @ankane

.github/workflows/build.yml

@@ -8,17 +8,17 @@ jobs:
       fail-fast: false
       matrix:
         include:
+          # - postgres: 18
+          #   os: ubuntu-24.04
           - postgres: 17
-            os: ubuntu-22.04
+            os: ubuntu-24.04
           - postgres: 16
             os: ubuntu-22.04
           - postgres: 15
             os: ubuntu-22.04
           - postgres: 14
-            os: ubuntu-22.04
-          - postgres: 13
             os: ubuntu-20.04
-          - postgres: 12
+          - postgres: 13
             os: ubuntu-20.04
     steps:
       - uses: actions/checkout@v4

CHANGELOG.md

@@ -1,3 +1,23 @@
+## 0.8.0 (unreleased)
+
+- Added support for inline filtering with HNSW
+- Added casts for arrays to `sparsevec`
+- Improved cost estimation
+- Improved performance of HNSW inserts and on-disk index builds
+- Reduced memory usage for HNSW index scans
+- Dropped support for Postgres 12
+
+## 0.7.4 (2024-08-05)
+
+- Fixed locking for parallel HNSW index builds
+- Fixed compilation error with GCC 14 on i386 when SSE2 is not enabled
+
+## 0.7.3 (2024-07-22)
+
+- Fixed `failed to add index item` error with `sparsevec`
+- Fixed compilation error with FreeBSD ARM
+- Fixed compilation warning with MSVC and Postgres 16
+
 ## 0.7.2 (2024-06-11)
 
 - Fixed initialization fork for indexes on unlogged tables

Dockerfile

@@ -1,4 +1,4 @@
-ARG PG_MAJOR=16
+ARG PG_MAJOR=17
 FROM postgres:$PG_MAJOR
 ARG PG_MAJOR
 

META.json

@@ -2,7 +2,7 @@
 	"name": "vector",
 	"abstract": "Open-source vector similarity search for Postgres",
 	"description": "Supports L2 distance, inner product, and cosine distance",
-	"version": "0.7.2",
+	"version": "0.7.4",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.7.2",
+			"version": "0.7.4",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,8 +1,9 @@
 EXTENSION = vector
-EXTVERSION = 0.7.2
+EXTVERSION = 0.7.4
 
 MODULE_big = vector
-DATA = $(wildcard sql/*--*.sql)
+DATA = $(wildcard sql/*--*--*.sql)
+DATA_built = sql/$(EXTENSION)--$(EXTVERSION).sql
 OBJS = src/bitutils.o src/bitvec.o src/halfutils.o src/halfvec.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/sparsevec.o src/vector.o
 HEADERS = src/halfvec.h src/sparsevec.h src/vector.h
 
@@ -42,8 +43,6 @@ all: sql/$(EXTENSION)--$(EXTVERSION).sql
 sql/$(EXTENSION)--$(EXTVERSION).sql: sql/$(EXTENSION).sql
 	cp $< $@
 
-EXTRA_CLEAN = sql/$(EXTENSION)--$(EXTVERSION).sql
-
 PG_CONFIG ?= pg_config
 PGXS := $(shell $(PG_CONFIG) --pgxs)
 include $(PGXS)
@@ -53,7 +52,7 @@ ifeq ($(PROVE),)
 	PROVE = prove
 endif
 
-# for Postgres 15
+# for Postgres < 15
 PROVE_FLAGS += -I ./test/perl
 
 prove_installcheck:
@@ -67,7 +66,7 @@ dist:
 	git archive --format zip --prefix=$(EXTENSION)-$(EXTVERSION)/ --output dist/$(EXTENSION)-$(EXTVERSION).zip master
 
 # for Docker
-PG_MAJOR ?= 16
+PG_MAJOR ?= 17
 
 .PHONY: docker
 

Makefile.win

@@ -1,6 +1,7 @@
 EXTENSION = vector
-EXTVERSION = 0.7.2
+EXTVERSION = 0.7.4
 
+DATA_built = sql\$(EXTENSION)--$(EXTVERSION).sql
 OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.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\sparsevec.obj src\vector.obj
 HEADERS = src\halfvec.h src\sparsevec.h src\vector.h
 
@@ -19,11 +20,6 @@ PG_CFLAGS = $(PG_CFLAGS) $(OPTFLAGS) /O2 /fp:fast
 # https://learn.microsoft.com/en-us/cpp/error-messages/tool-errors/vectorizer-and-parallelizer-messages
 # PG_CFLAGS = $(PG_CFLAGS) /Qvec-report:2
 
-all: sql\$(EXTENSION)--$(EXTVERSION).sql
-
-sql\$(EXTENSION)--$(EXTVERSION).sql: sql\$(EXTENSION).sql
-	copy sql\$(EXTENSION).sql $@
-
 # TODO use pg_config
 !ifndef PGROOT
 !error PGROOT is not set
@@ -43,15 +39,18 @@ SHLIB = $(EXTENSION).dll
 
 LIBS = "$(LIBDIR)\postgres.lib"
 
+all: $(SHLIB) $(DATA_built)
+
 .c.obj:
 	$(CC) $(CFLAGS) /c $< /Fo$@
 
 $(SHLIB): $(OBJS)
 	$(CC) $(CFLAGS) $(OBJS) $(LIBS) /link /DLL /OUT:$(SHLIB)
 
-all: $(SHLIB)
+sql\$(EXTENSION)--$(EXTVERSION).sql: sql\$(EXTENSION).sql
+	copy sql\$(EXTENSION).sql $@
 
-install:
+install: all
 	copy $(SHLIB) "$(PKGLIBDIR)"
 	copy $(EXTENSION).control "$(SHAREDIR)\extension"
 	copy sql\$(EXTENSION)--*.sql "$(SHAREDIR)\extension"
@@ -70,6 +69,6 @@ uninstall:
 
 clean:
 	del /f $(SHLIB) $(EXTENSION).lib $(EXTENSION).exp
+	del /f $(DATA_built)
 	del /f $(OBJS)
-	del /f sql\$(EXTENSION)--$(EXTVERSION).sql
 	del /f /s /q results regression.diffs regression.out tmp_check tmp_check_iso log output_iso

README.md

@@ -21,7 +21,7 @@ Compile and install the extension (supports Postgres 12+)
 
 ```sh
 cd /tmp
-git clone --branch v0.7.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
@@ -46,12 +46,14 @@ Then use `nmake` to build:
 ```cmd
 set "PGROOT=C:\Program Files\PostgreSQL\16"
 cd %TEMP%
-git clone --branch v0.7.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
 ```
 
+Note: Postgres 17 is not supported yet due to an upstream issue
+
 See the [installation notes](#installation-notes---windows) if you run into issues
 
 You can also install it with [Docker](#docker) or [conda-forge](#conda-forge).
@@ -100,13 +102,15 @@ Or add a vector column to an existing table
 ALTER TABLE items ADD COLUMN embedding vector(3);
 ```
 
+Also supports [half-precision](#half-precision-vectors), [binary](#binary-vectors), and [sparse](#sparse-vectors) vectors
+
 Insert vectors
 
 ```sql
 INSERT INTO items (embedding) VALUES ('[1,2,3]'), ('[4,5,6]');
 ```
 
-Or load vectors in bulk using `COPY` ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/bulk_loading.py))
+Or load vectors in bulk using `COPY` ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/loading/example.py))
 
 ```sql
 COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
@@ -145,6 +149,8 @@ Supported distance functions are:
 - `<#>` - (negative) inner product
 - `<=>` - cosine distance
 - `<+>` - L1 distance (added in 0.7.0)
+- `<~>` - Hamming distance (binary vectors, added in 0.7.0)
+- `<%>` - Jaccard distance (binary vectors, added in 0.7.0)
 
 Get the nearest neighbors to a row
 
@@ -202,7 +208,7 @@ You can add an index to use approximate nearest neighbor search, which trades so
 
 Supported index types are:
 
-- [HNSW](#hnsw) - added in 0.5.0
+- [HNSW](#hnsw)
 - [IVFFlat](#ivfflat)
 
 ## HNSW
@@ -433,6 +439,12 @@ Create an index on one [or more](https://www.postgresql.org/docs/current/indexes
 CREATE INDEX ON items (category_id);
 ```
 
+Or a composite HNSW index for approximate search (added in 0.8.0)
+
+```sql
+CREATE INDEX ON items USING hnsw (embedding vector_l2_ops, category_id);
+```
+
 Or a [partial index](https://www.postgresql.org/docs/current/indexes-partial.html) on the vector column for approximate search
 
 ```sql
@@ -473,7 +485,7 @@ SELECT * FROM items ORDER BY embedding::halfvec(3) <-> '[1,2,3]' LIMIT 5;
 
 ## Binary Vectors
 
-Use the `bit` type to store binary vectors ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/hash_image_search.py))
+Use the `bit` type to store binary vectors ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/imagehash/example.py))
 
 ```sql
 CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
@@ -551,7 +563,7 @@ SELECT id, content FROM items, plainto_tsquery('hello search') query
     WHERE textsearch @@ query ORDER BY ts_rank_cd(textsearch, query) DESC LIMIT 5;
 ```
 
-You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search_rrf.py) or a [cross-encoder](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search.py) to combine results.
+You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search/rrf.py) or a [cross-encoder](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search/cross_encoder.py) to combine results.
 
 ## Indexing Subvectors
 
@@ -597,7 +609,7 @@ Be sure to restart Postgres for changes to take effect.
 
 ### Loading
 
-Use `COPY` for bulk loading data ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/bulk_loading.py)).
+Use `COPY` for bulk loading data ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/loading/example.py)).
 
 ```sql
 COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
@@ -687,7 +699,7 @@ Scale pgvector the same way you scale Postgres.
 
 Scale vertically by increasing memory, CPU, and storage on a single instance. Use existing tools to [tune parameters](#tuning) and [monitor performance](#monitoring).
 
-Scale horizontally with [replicas](https://www.postgresql.org/docs/current/hot-standby.html), or use [Citus](https://github.com/citusdata/citus) or another approach for sharding ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/citus.py)).
+Scale horizontally with [replicas](https://www.postgresql.org/docs/current/hot-standby.html), or use [Citus](https://github.com/citusdata/citus) or another approach for sharding ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/citus/example.py)).
 
 ## Languages
 
@@ -983,7 +995,7 @@ l2_normalize(sparsevec) → sparsevec | Normalize with Euclidean norm | 0.7.0
 If your machine has multiple Postgres installations, specify the path to [pg_config](https://www.postgresql.org/docs/current/app-pgconfig.html) with:
 
 ```sh
-export PG_CONFIG=/Library/PostgreSQL/16/bin/pg_config
+export PG_CONFIG=/Library/PostgreSQL/17/bin/pg_config
 ```
 
 Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use:
@@ -994,11 +1006,11 @@ sudo --preserve-env=PG_CONFIG make install
 
 A few common paths on Mac are:
 
-- EDB installer - `/Library/PostgreSQL/16/bin/pg_config`
-- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@16/bin/pg_config`
-- Homebrew (x86-64) - `/usr/local/opt/postgresql@16/bin/pg_config`
+- EDB installer - `/Library/PostgreSQL/17/bin/pg_config`
+- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@17/bin/pg_config`
+- Homebrew (x86-64) - `/usr/local/opt/postgresql@17/bin/pg_config`
 
-Note: Replace `16` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### Missing Header
 
@@ -1007,10 +1019,10 @@ If compilation fails with `fatal error: postgres.h: No such file or directory`,
 For Ubuntu and Debian, use:
 
 ```sh
-sudo apt install postgresql-server-dev-16
+sudo apt install postgresql-server-dev-17
 ```
 
-Note: Replace `16` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### Missing SDK
 
@@ -1043,17 +1055,17 @@ If installation fails with `Access is denied`, re-run the installation instructi
 Get the [Docker image](https://hub.docker.com/r/pgvector/pgvector) with:
 
 ```sh
-docker pull pgvector/pgvector:pg16
+docker pull pgvector/pgvector:pg17
 ```
 
-This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (replace `16` with your Postgres server version, and run it the same way).
+This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (replace `17` with your Postgres server version, and run it the same way).
 
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.7.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
 cd pgvector
-docker build --pull --build-arg PG_MAJOR=16 -t myuser/pgvector .
+docker build --pull --build-arg PG_MAJOR=17 -t myuser/pgvector .
 ```
 
 ### Homebrew
@@ -1064,7 +1076,7 @@ With Homebrew Postgres, you can use:
 brew install pgvector
 ```
 
-Note: This only adds it to the `postgresql@14` formula
+Note: This only adds it to the `postgresql@17` and `postgresql@14` formulas
 
 ### PGXN
 
@@ -1079,22 +1091,22 @@ pgxn install vector
 Debian and Ubuntu packages are available from the [PostgreSQL APT Repository](https://wiki.postgresql.org/wiki/Apt). Follow the [setup instructions](https://wiki.postgresql.org/wiki/Apt#Quickstart) and run:
 
 ```sh
-sudo apt install postgresql-16-pgvector
+sudo apt install postgresql-17-pgvector
 ```
 
-Note: Replace `16` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### Yum
 
 RPM packages are available from the [PostgreSQL Yum Repository](https://yum.postgresql.org/). Follow the [setup instructions](https://www.postgresql.org/download/linux/redhat/) for your distribution and run:
 
 ```sh
-sudo yum install pgvector_16
+sudo yum install pgvector_17
 # or
-sudo dnf install pgvector_16
+sudo dnf install pgvector_17
 ```
 
-Note: Replace `16` with your Postgres server version
+Note: Replace `17` with your Postgres server version
 
 ### pkg
 
@@ -1183,6 +1195,7 @@ Thanks to:
 - [k-means++: The Advantage of Careful Seeding](https://theory.stanford.edu/~sergei/papers/kMeansPP-soda.pdf)
 - [Concept Decompositions for Large Sparse Text Data using Clustering](https://www.cs.utexas.edu/users/inderjit/public_papers/concept_mlj.pdf)
 - [Efficient and Robust Approximate Nearest Neighbor Search using Hierarchical Navigable Small World Graphs](https://arxiv.org/ftp/arxiv/papers/1603/1603.09320.pdf)
+- [HQANN: Efficient and Robust Similarity Search for Hybrid Queries with Structured and Unstructured Constraints](https://arxiv.org/pdf/2207.07940.pdf)
 
 ## History
 

sql/vector--0.7.2--0.7.3.sql

@@ -0,0 +1,2 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.7.3'" to load this file. \quit

sql/vector--0.7.3--0.7.4.sql

@@ -0,0 +1,2 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.7.4'" to load this file. \quit

sql/vector--0.7.4--0.8.0.sql

@@ -0,0 +1,34 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.8.0'" to load this file. \quit
+
+CREATE FUNCTION array_to_sparsevec(integer[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(real[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(double precision[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(numeric[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE CAST (integer[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(integer[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (real[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(real[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (double precision[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(double precision[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (numeric[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(numeric[], integer, boolean) AS ASSIGNMENT;
+
+CREATE FUNCTION hnsw_attribute_distance(integer, integer) RETURNS float8
+	AS 'MODULE_PATHNAME', 'hnsw_int4_attribute_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE OPERATOR CLASS vector_integer_ops
+	DEFAULT FOR TYPE integer USING hnsw AS
+	OPERATOR 2 = (integer, integer),
+	FUNCTION 4 hnsw_attribute_distance(integer, integer);

sql/vector.sql

@@ -782,6 +782,18 @@ CREATE FUNCTION halfvec_to_sparsevec(halfvec, integer, boolean) RETURNS sparseve
 CREATE FUNCTION sparsevec_to_halfvec(sparsevec, integer, boolean) RETURNS halfvec
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
+CREATE FUNCTION array_to_sparsevec(integer[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(real[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(double precision[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE FUNCTION array_to_sparsevec(numeric[], integer, boolean) RETURNS sparsevec
+	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
 -- sparsevec casts
 
 CREATE CAST (sparsevec AS sparsevec)
@@ -799,6 +811,18 @@ CREATE CAST (sparsevec AS halfvec)
 CREATE CAST (halfvec AS sparsevec)
 	WITH FUNCTION halfvec_to_sparsevec(halfvec, integer, boolean) AS IMPLICIT;
 
+CREATE CAST (integer[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(integer[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (real[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(real[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (double precision[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(double precision[], integer, boolean) AS ASSIGNMENT;
+
+CREATE CAST (numeric[] AS sparsevec)
+	WITH FUNCTION array_to_sparsevec(numeric[], integer, boolean) AS ASSIGNMENT;
+
 -- sparsevec operators
 
 CREATE OPERATOR <-> (
@@ -892,3 +916,13 @@ CREATE OPERATOR CLASS sparsevec_l1_ops
 	OPERATOR 1 <+> (sparsevec, sparsevec) FOR ORDER BY float_ops,
 	FUNCTION 1 l1_distance(sparsevec, sparsevec),
 	FUNCTION 3 hnsw_sparsevec_support(internal);
+
+-- hnsw attributes
+
+CREATE FUNCTION hnsw_attribute_distance(integer, integer) RETURNS float8
+	AS 'MODULE_PATHNAME', 'hnsw_int4_attribute_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
+
+CREATE OPERATOR CLASS vector_integer_ops
+	DEFAULT FOR TYPE integer USING hnsw AS
+	OPERATOR 2 = (integer, integer),
+	FUNCTION 4 hnsw_attribute_distance(integer, integer);

src/bitutils.h

@@ -4,8 +4,8 @@
 #include "postgres.h"
 
 /* Check version in first header */
-#if PG_VERSION_NUM < 120000
-#error "Requires PostgreSQL 12+"
+#if PG_VERSION_NUM < 130000
+#error "Requires PostgreSQL 13+"
 #endif
 
 extern uint64 (*BitHammingDistance) (uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 distance);

src/bitvec.c

@@ -3,6 +3,7 @@
 #include "bitutils.h"
 #include "bitvec.h"
 #include "utils/varbit.h"
+#include "vector.h"
 
 #if PG_VERSION_NUM >= 160000
 #include "varatt.h"
@@ -40,7 +41,7 @@ CheckDims(VarBit *a, VarBit *b)
 /*
  * Get the Hamming distance between two bit vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(hamming_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hamming_distance);
 Datum
 hamming_distance(PG_FUNCTION_ARGS)
 {
@@ -55,7 +56,7 @@ hamming_distance(PG_FUNCTION_ARGS)
 /*
  * Get the Jaccard distance between two bit vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(jaccard_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(jaccard_distance);
 Datum
 jaccard_distance(PG_FUNCTION_ARGS)
 {

src/halfvec.c

@@ -19,11 +19,6 @@
 #include "utils/numeric.h"
 #include "vector.h"
 
-#if PG_VERSION_NUM < 130000
-#define TYPALIGN_DOUBLE 'd'
-#define TYPALIGN_INT 'i'
-#endif
-
 #define STATE_DIMS(x) (ARR_DIMS(x)[0] - 1)
 #define CreateStateDatums(dim) palloc(sizeof(Datum) * (dim + 1))
 
@@ -164,28 +159,10 @@ CheckStateArray(ArrayType *statearray, const char *caller)
 	return (float8 *) ARR_DATA_PTR(statearray);
 }
 
-#if PG_VERSION_NUM < 120003
-static pg_noinline void
-float_overflow_error(void)
-{
-	ereport(ERROR,
-			(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-			 errmsg("value out of range: overflow")));
-}
-
-static pg_noinline void
-float_underflow_error(void)
-{
-	ereport(ERROR,
-			(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-			 errmsg("value out of range: underflow")));
-}
-#endif
-
 /*
  * Convert textual representation to internal representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_in);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_in);
 Datum
 halfvec_in(PG_FUNCTION_ARGS)
 {
@@ -299,7 +276,7 @@ halfvec_in(PG_FUNCTION_ARGS)
 /*
  * Convert internal representation to textual representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_out);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_out);
 Datum
 halfvec_out(PG_FUNCTION_ARGS)
 {
@@ -345,7 +322,7 @@ halfvec_out(PG_FUNCTION_ARGS)
 /*
  * Convert type modifier
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_typmod_in);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_typmod_in);
 Datum
 halfvec_typmod_in(PG_FUNCTION_ARGS)
 {
@@ -376,7 +353,7 @@ halfvec_typmod_in(PG_FUNCTION_ARGS)
 /*
  * Convert external binary representation to internal representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_recv);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_recv);
 Datum
 halfvec_recv(PG_FUNCTION_ARGS)
 {
@@ -410,7 +387,7 @@ halfvec_recv(PG_FUNCTION_ARGS)
 /*
  * Convert internal representation to the external binary representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_send);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_send);
 Datum
 halfvec_send(PG_FUNCTION_ARGS)
 {
@@ -430,7 +407,7 @@ halfvec_send(PG_FUNCTION_ARGS)
  * Convert half vector to half vector
  * This is needed to check the type modifier
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec);
 Datum
 halfvec(PG_FUNCTION_ARGS)
 {
@@ -445,7 +422,7 @@ halfvec(PG_FUNCTION_ARGS)
 /*
  * Convert array to half vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(array_to_halfvec);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(array_to_halfvec);
 Datum
 array_to_halfvec(PG_FUNCTION_ARGS)
 {
@@ -519,7 +496,7 @@ array_to_halfvec(PG_FUNCTION_ARGS)
 /*
  * Convert half vector to float4[]
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_to_float4);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_to_float4);
 Datum
 halfvec_to_float4(PG_FUNCTION_ARGS)
 {
@@ -543,7 +520,7 @@ halfvec_to_float4(PG_FUNCTION_ARGS)
 /*
  * Convert vector to half vec
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_to_halfvec);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_to_halfvec);
 Datum
 vector_to_halfvec(PG_FUNCTION_ARGS)
 {
@@ -565,7 +542,7 @@ vector_to_halfvec(PG_FUNCTION_ARGS)
 /*
  * Get the L2 distance between half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_l2_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_l2_distance);
 Datum
 halfvec_l2_distance(PG_FUNCTION_ARGS)
 {
@@ -580,7 +557,7 @@ halfvec_l2_distance(PG_FUNCTION_ARGS)
 /*
  * Get the L2 squared distance between half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_l2_squared_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_l2_squared_distance);
 Datum
 halfvec_l2_squared_distance(PG_FUNCTION_ARGS)
 {
@@ -595,7 +572,7 @@ halfvec_l2_squared_distance(PG_FUNCTION_ARGS)
 /*
  * Get the inner product of two half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_inner_product);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_inner_product);
 Datum
 halfvec_inner_product(PG_FUNCTION_ARGS)
 {
@@ -610,7 +587,7 @@ halfvec_inner_product(PG_FUNCTION_ARGS)
 /*
  * Get the negative inner product of two half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_negative_inner_product);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_negative_inner_product);
 Datum
 halfvec_negative_inner_product(PG_FUNCTION_ARGS)
 {
@@ -625,7 +602,7 @@ halfvec_negative_inner_product(PG_FUNCTION_ARGS)
 /*
  * Get the cosine distance between two half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_cosine_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_cosine_distance);
 Datum
 halfvec_cosine_distance(PG_FUNCTION_ARGS)
 {
@@ -657,7 +634,7 @@ halfvec_cosine_distance(PG_FUNCTION_ARGS)
  * Currently uses angular distance since needs to satisfy triangle inequality
  * Assumes inputs are unit vectors (skips norm)
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_spherical_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_spherical_distance);
 Datum
 halfvec_spherical_distance(PG_FUNCTION_ARGS)
 {
@@ -681,7 +658,7 @@ halfvec_spherical_distance(PG_FUNCTION_ARGS)
 /*
  * Get the L1 distance between two half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_l1_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_l1_distance);
 Datum
 halfvec_l1_distance(PG_FUNCTION_ARGS)
 {
@@ -696,7 +673,7 @@ halfvec_l1_distance(PG_FUNCTION_ARGS)
 /*
  * Get the dimensions of a half vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_vector_dims);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_vector_dims);
 Datum
 halfvec_vector_dims(PG_FUNCTION_ARGS)
 {
@@ -708,7 +685,7 @@ halfvec_vector_dims(PG_FUNCTION_ARGS)
 /*
  * Get the L2 norm of a half vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_l2_norm);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_l2_norm);
 Datum
 halfvec_l2_norm(PG_FUNCTION_ARGS)
 {
@@ -730,7 +707,7 @@ halfvec_l2_norm(PG_FUNCTION_ARGS)
 /*
  * Normalize a half vector with the L2 norm
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_l2_normalize);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_l2_normalize);
 Datum
 halfvec_l2_normalize(PG_FUNCTION_ARGS)
 {
@@ -769,7 +746,7 @@ halfvec_l2_normalize(PG_FUNCTION_ARGS)
 /*
  * Add half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_add);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_add);
 Datum
 halfvec_add(PG_FUNCTION_ARGS)
 {
@@ -808,7 +785,7 @@ halfvec_add(PG_FUNCTION_ARGS)
 /*
  * Subtract half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_sub);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_sub);
 Datum
 halfvec_sub(PG_FUNCTION_ARGS)
 {
@@ -847,7 +824,7 @@ halfvec_sub(PG_FUNCTION_ARGS)
 /*
  * Multiply half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_mul);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_mul);
 Datum
 halfvec_mul(PG_FUNCTION_ARGS)
 {
@@ -889,7 +866,7 @@ halfvec_mul(PG_FUNCTION_ARGS)
 /*
  * Concatenate half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_concat);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_concat);
 Datum
 halfvec_concat(PG_FUNCTION_ARGS)
 {
@@ -913,7 +890,7 @@ halfvec_concat(PG_FUNCTION_ARGS)
 /*
  * Quantize a half vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_binary_quantize);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_binary_quantize);
 Datum
 halfvec_binary_quantize(PG_FUNCTION_ARGS)
 {
@@ -931,7 +908,7 @@ halfvec_binary_quantize(PG_FUNCTION_ARGS)
 /*
  * Get a subvector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_subvector);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_subvector);
 Datum
 halfvec_subvector(PG_FUNCTION_ARGS)
 {
@@ -1005,7 +982,7 @@ halfvec_cmp_internal(HalfVector * a, HalfVector * b)
 /*
  * Less than
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_lt);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_lt);
 Datum
 halfvec_lt(PG_FUNCTION_ARGS)
 {
@@ -1018,7 +995,7 @@ halfvec_lt(PG_FUNCTION_ARGS)
 /*
  * Less than or equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_le);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_le);
 Datum
 halfvec_le(PG_FUNCTION_ARGS)
 {
@@ -1031,7 +1008,7 @@ halfvec_le(PG_FUNCTION_ARGS)
 /*
  * Equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_eq);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_eq);
 Datum
 halfvec_eq(PG_FUNCTION_ARGS)
 {
@@ -1044,7 +1021,7 @@ halfvec_eq(PG_FUNCTION_ARGS)
 /*
  * Not equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_ne);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_ne);
 Datum
 halfvec_ne(PG_FUNCTION_ARGS)
 {
@@ -1057,7 +1034,7 @@ halfvec_ne(PG_FUNCTION_ARGS)
 /*
  * Greater than or equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_ge);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_ge);
 Datum
 halfvec_ge(PG_FUNCTION_ARGS)
 {
@@ -1070,7 +1047,7 @@ halfvec_ge(PG_FUNCTION_ARGS)
 /*
  * Greater than
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_gt);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_gt);
 Datum
 halfvec_gt(PG_FUNCTION_ARGS)
 {
@@ -1083,7 +1060,7 @@ halfvec_gt(PG_FUNCTION_ARGS)
 /*
  * Compare half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_cmp);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_cmp);
 Datum
 halfvec_cmp(PG_FUNCTION_ARGS)
 {
@@ -1096,7 +1073,7 @@ halfvec_cmp(PG_FUNCTION_ARGS)
 /*
  * Accumulate half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_accum);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_accum);
 Datum
 halfvec_accum(PG_FUNCTION_ARGS)
 {
@@ -1157,7 +1134,7 @@ halfvec_accum(PG_FUNCTION_ARGS)
 /*
  * Average half vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_avg);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_avg);
 Datum
 halfvec_avg(PG_FUNCTION_ARGS)
 {
@@ -1191,7 +1168,7 @@ halfvec_avg(PG_FUNCTION_ARGS)
 /*
  * Convert sparse vector to half vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_to_halfvec);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_to_halfvec);
 Datum
 sparsevec_to_halfvec(PG_FUNCTION_ARGS)
 {

src/halfvec.h

@@ -38,7 +38,7 @@
 /* F16C has better performance than _Float16 (on x86-64) */
 #if defined(__F16C__)
 #define F16C_SUPPORT
-#elif defined(__FLT16_MAX__) && !defined(HALFVEC_DISPATCH)
+#elif defined(__FLT16_MAX__) && !defined(HALFVEC_DISPATCH) && !defined(__FreeBSD__) && (!defined(__i386__) || defined(__SSE2__))
 #define FLT16_SUPPORT
 #endif
 

src/hnsw.c

@@ -9,8 +9,10 @@
 #include "commands/vacuum.h"
 #include "hnsw.h"
 #include "miscadmin.h"
+#include "utils/float.h"
 #include "utils/guc.h"
 #include "utils/selfuncs.h"
+#include "utils/spccache.h"
 
 #if PG_VERSION_NUM < 150000
 #define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x)
@@ -59,17 +61,9 @@ HnswInit(void)
 
 	hnsw_relopt_kind = add_reloption_kind();
 	add_int_reloption(hnsw_relopt_kind, "m", "Max number of connections",
-					  HNSW_DEFAULT_M, HNSW_MIN_M, HNSW_MAX_M
-#if PG_VERSION_NUM >= 130000
-					  ,AccessExclusiveLock
-#endif
-		);
+					  HNSW_DEFAULT_M, HNSW_MIN_M, HNSW_MAX_M, AccessExclusiveLock);
 	add_int_reloption(hnsw_relopt_kind, "ef_construction", "Size of the dynamic candidate list for construction",
-					  HNSW_DEFAULT_EF_CONSTRUCTION, HNSW_MIN_EF_CONSTRUCTION, HNSW_MAX_EF_CONSTRUCTION
-#if PG_VERSION_NUM >= 130000
-					  ,AccessExclusiveLock
-#endif
-		);
+					  HNSW_DEFAULT_EF_CONSTRUCTION, HNSW_MIN_EF_CONSTRUCTION, HNSW_MAX_EF_CONSTRUCTION, AccessExclusiveLock);
 
 	DefineCustomIntVariable("hnsw.ef_search", "Sets the size of the dynamic candidate list for search",
 							"Valid range is 1..1000.", &hnsw_ef_search,
@@ -106,14 +100,16 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 {
 	GenericCosts costs;
 	int			m;
-	int			entryLevel;
+	double		ratio;
+	double		startupPages;
+	double		spc_seq_page_cost;
 	Relation	index;
 
 	/* Never use index without order */
 	if (path->indexorderbys == NULL)
 	{
-		*indexStartupCost = DBL_MAX;
-		*indexTotalCost = DBL_MAX;
+		*indexStartupCost = get_float8_infinity();
+		*indexTotalCost = get_float8_infinity();
 		*indexSelectivity = 0;
 		*indexCorrelation = 0;
 		*indexPages = 0;
@@ -122,21 +118,71 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 
 	MemSet(&costs, 0, sizeof(costs));
 
+	genericcostestimate(root, path, loop_count, &costs);
+
 	index = index_open(path->indexinfo->indexoid, NoLock);
 	HnswGetMetaPageInfo(index, &m, NULL);
 	index_close(index, NoLock);
 
-	/* Approximate entry level */
-	entryLevel = (int) -log(1.0 / path->indexinfo->tuples) * HnswGetMl(m);
+	/*
+	 * HNSW cost estimation follows a formula that accounts for the total
+	 * number of tuples indexed combined with the parameters that most
+	 * influence the duration of the index scan, namely: m - the number of
+	 * tuples that are scanned in each step of the HNSW graph traversal
+	 * ef_search - which influences the total number of steps taken at layer 0
+	 *
+	 * The source of the vector data can impact how many steps it takes to
+	 * converge on the set of vectors to return to the executor. Currently, we
+	 * use a hardcoded scaling factor (HNSWScanScalingFactor) to help
+	 * influence that, but this could later become a configurable parameter
+	 * based on the cost estimations.
+	 *
+	 * The tuple estimator formula is below:
+	 *
+	 * numIndexTuples = entryLevel * m + layer0TuplesMax * layer0Selectivity
+	 *
+	 * "entryLevel * m" represents the floor of tuples we need to scan to get
+	 * to layer 0 (L0).
+	 *
+	 * "layer0TuplesMax" is the estimated total number of tuples we'd scan at
+	 * L0 if we weren't discarding already visited tuples as part of the scan.
+	 *
+	 * "layer0Selectivity" estimates the percentage of tuples that are scanned
+	 * at L0, accounting for previously visited tuples, multiplied by the
+	 * "scalingFactor" (currently hardcoded).
+	 */
+	if (path->indexinfo->tuples > 0)
+	{
+		double		scalingFactor = 0.55;
+		int			entryLevel = (int) (log(path->indexinfo->tuples) * HnswGetMl(m));
+		int			layer0TuplesMax = HnswGetLayerM(m, 0) * hnsw_ef_search;
+		double		layer0Selectivity = scalingFactor * log(path->indexinfo->tuples) / (log(m) * (1 + log(hnsw_ef_search)));
 
-	/* TODO Improve estimate of visited tuples (currently underestimates) */
-	/* Account for number of tuples (or entry level), m, and ef_search */
-	costs.numIndexTuples = (entryLevel + 2) * m;
+		ratio = (entryLevel * m + layer0TuplesMax * layer0Selectivity) / path->indexinfo->tuples;
 
-	genericcostestimate(root, path, loop_count, &costs);
+		if (ratio > 1)
+			ratio = 1;
+	}
+	else
+		ratio = 1;
 
-	/* Use total cost since most work happens before first tuple is returned */
-	*indexStartupCost = costs.indexTotalCost;
+	get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
+
+	/* Startup cost is cost before returning the first row */
+	costs.indexStartupCost = costs.indexTotalCost * ratio;
+
+	/* Adjust cost if needed since TOAST not included in seq scan cost */
+	startupPages = costs.numIndexPages * ratio;
+	if (startupPages > path->indexinfo->rel->pages && ratio < 0.5)
+	{
+		/* Change all page cost from random to sequential */
+		costs.indexStartupCost -= startupPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+
+		/* Remove cost of extra pages */
+		costs.indexStartupCost -= (startupPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
+	}
+
+	*indexStartupCost = costs.indexStartupCost;
 	*indexTotalCost = costs.indexTotalCost;
 	*indexSelectivity = costs.indexSelectivity;
 	*indexCorrelation = costs.indexCorrelation;
@@ -154,23 +200,10 @@ hnswoptions(Datum reloptions, bool validate)
 		{"ef_construction", RELOPT_TYPE_INT, offsetof(HnswOptions, efConstruction)},
 	};
 
-#if PG_VERSION_NUM >= 130000
 	return (bytea *) build_reloptions(reloptions, validate,
 									  hnsw_relopt_kind,
 									  sizeof(HnswOptions),
 									  tab, lengthof(tab));
-#else
-	relopt_value *options;
-	int			numoptions;
-	HnswOptions *rdopts;
-
-	options = parseRelOptions(reloptions, validate, hnsw_relopt_kind, &numoptions);
-	rdopts = allocateReloptStruct(sizeof(HnswOptions), options, numoptions);
-	fillRelOptions((void *) rdopts, sizeof(HnswOptions), options, numoptions,
-				   validate, tab, lengthof(tab));
-
-	return (bytea *) rdopts;
-#endif
 }
 
 /*
@@ -187,22 +220,20 @@ hnswvalidate(Oid opclassoid)
  *
  * See https://www.postgresql.org/docs/current/index-api.html
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(hnswhandler);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnswhandler);
 Datum
 hnswhandler(PG_FUNCTION_ARGS)
 {
 	IndexAmRoutine *amroutine = makeNode(IndexAmRoutine);
 
 	amroutine->amstrategies = 0;
-	amroutine->amsupport = 3;
-#if PG_VERSION_NUM >= 130000
+	amroutine->amsupport = 4;
 	amroutine->amoptsprocnum = 0;
-#endif
 	amroutine->amcanorder = false;
 	amroutine->amcanorderbyop = true;
 	amroutine->amcanbackward = false;	/* can change direction mid-scan */
 	amroutine->amcanunique = false;
-	amroutine->amcanmulticol = false;
+	amroutine->amcanmulticol = true;
 	amroutine->amoptionalkey = true;
 	amroutine->amsearcharray = false;
 	amroutine->amsearchnulls = false;
@@ -210,17 +241,24 @@ hnswhandler(PG_FUNCTION_ARGS)
 	amroutine->amclusterable = false;
 	amroutine->ampredlocks = false;
 	amroutine->amcanparallel = false;
-	amroutine->amcaninclude = false;
-#if PG_VERSION_NUM >= 130000
-	amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
-	amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
+#if PG_VERSION_NUM >= 170000
+	amroutine->amcanbuildparallel = true;
 #endif
+	amroutine->amcaninclude = false;
+	amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
+#if PG_VERSION_NUM >= 160000
+	amroutine->amsummarizing = false;
+#endif
+	amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
 	amroutine->amkeytype = InvalidOid;
 
 	/* Interface functions */
 	amroutine->ambuild = hnswbuild;
 	amroutine->ambuildempty = hnswbuildempty;
 	amroutine->aminsert = hnswinsert;
+#if PG_VERSION_NUM >= 170000
+	amroutine->aminsertcleanup = NULL;
+#endif
 	amroutine->ambulkdelete = hnswbulkdelete;
 	amroutine->amvacuumcleanup = hnswvacuumcleanup;
 	amroutine->amcanreturn = NULL;
@@ -247,3 +285,17 @@ hnswhandler(PG_FUNCTION_ARGS)
 
 	PG_RETURN_POINTER(amroutine);
 }
+
+/*
+ * Get the distance between two int4 attributes
+ */
+PGDLLEXPORT PG_FUNCTION_INFO_V1(hnsw_int4_attribute_distance);
+Datum
+hnsw_int4_attribute_distance(PG_FUNCTION_ARGS)
+{
+	int32		a = PG_GETARG_INT32(0);
+	int32		b = PG_GETARG_INT32(1);
+	double		distance = ((double) a) - ((double) b);
+
+	PG_RETURN_FLOAT8(distance);
+}

src/hnsw.h

@@ -19,6 +19,7 @@
 #define HNSW_DISTANCE_PROC 1
 #define HNSW_NORM_PROC 2
 #define HNSW_TYPE_INFO_PROC 3
+#define HNSW_ATTRIBUTE_DISTANCE_PROC 4
 
 #define HNSW_VERSION	1
 #define HNSW_MAGIC_NUMBER 0xA953A953
@@ -76,11 +77,6 @@
 #define SeedRandom(seed) srandom(seed)
 #endif
 
-#if PG_VERSION_NUM < 130000
-#define list_delete_last(list) list_truncate(list, list_length(list) - 1)
-#define list_sort(list, cmp) ((list) = list_qsort(list, cmp))
-#endif
-
 #define HnswIsElementTuple(tup) ((tup)->type == HNSW_ELEMENT_TUPLE_TYPE)
 #define HnswIsNeighborTuple(tup) ((tup)->type == HNSW_NEIGHBOR_TUPLE_TYPE)
 
@@ -109,6 +105,8 @@
 #define HnswPtrPointer(hp) (hp).ptr
 #define HnswPtrOffset(hp) relptr_offset((hp).relptr)
 
+#define HnswUseIndexTuple(index) (IndexRelationGetNumberOfAttributes(index) > 1)
+
 /* Variables */
 extern int	hnsw_ef_search;
 extern int	hnsw_lock_tranche_id;
@@ -126,6 +124,7 @@ HnswPtrDeclare(HnswElementData, HnswElementRelptr, HnswElementPtr);
 HnswPtrDeclare(HnswNeighborArray, HnswNeighborArrayRelptr, HnswNeighborArrayPtr);
 HnswPtrDeclare(HnswNeighborArrayPtr, HnswNeighborsRelptr, HnswNeighborsPtr);
 HnswPtrDeclare(char, DatumRelptr, DatumPtr);
+HnswPtrDeclare(IndexTupleData, IndexTupleRelptr, IndexTuplePtr);
 
 struct HnswElementData
 {
@@ -141,6 +140,7 @@ struct HnswElementData
 	OffsetNumber neighborOffno;
 	BlockNumber neighborPage;
 	DatumPtr	value;
+	IndexTuplePtr itup;
 	LWLock		lock;
 };
 
@@ -160,11 +160,14 @@ struct HnswNeighborArray
 	HnswCandidate items[FLEXIBLE_ARRAY_MEMBER];
 };
 
-typedef struct HnswPairingHeapNode
+typedef struct HnswSearchCandidate
 {
-	pairingheap_node ph_node;
-	HnswCandidate *inner;
-}			HnswPairingHeapNode;
+	pairingheap_node c_node;
+	pairingheap_node w_node;
+	HnswElementPtr element;
+	double		distance;
+	bool		matches;
+}			HnswSearchCandidate;
 
 /* HNSW index options */
 typedef struct HnswOptions
@@ -188,8 +191,8 @@ typedef struct HnswGraph
 
 	/* Allocations state */
 	LWLock		allocatorLock;
-	long		memoryUsed;
-	long		memoryTotal;
+	Size		memoryUsed;
+	Size		memoryTotal;
 
 	/* Flushed state */
 	LWLock		flushLock;
@@ -259,15 +262,17 @@ typedef struct HnswBuildState
 	double		reltuples;
 
 	/* Support functions */
-	FmgrInfo   *procinfo;
+	FmgrInfo   *procinfo[2];
 	FmgrInfo   *normprocinfo;
-	Oid			collation;
+	Oid		   *collation;
 
 	/* Variables */
 	HnswGraph	graphData;
 	HnswGraph  *graph;
 	double		ml;
 	int			maxLevel;
+	bool		useIndexTuple;
+	TupleDesc	tupdesc;
 
 	/* Memory */
 	MemoryContext graphCtx;
@@ -336,9 +341,9 @@ typedef struct HnswScanOpaqueData
 	MemoryContext tmpCtx;
 
 	/* Support functions */
-	FmgrInfo   *procinfo;
+	FmgrInfo   *procinfo[2];
 	FmgrInfo   *normprocinfo;
-	Oid			collation;
+	Oid		   *collation;
 }			HnswScanOpaqueData;
 
 typedef HnswScanOpaqueData * HnswScanOpaque;
@@ -356,8 +361,8 @@ typedef struct HnswVacuumState
 	int			efConstruction;
 
 	/* Support functions */
-	FmgrInfo   *procinfo;
-	Oid			collation;
+	FmgrInfo   *procinfo[2];
+	Oid		   *collation;
 
 	/* Variables */
 	struct tidhash_hash *deleted;
@@ -378,28 +383,32 @@ bool		HnswCheckNorm(FmgrInfo *procinfo, Oid collation, Datum value);
 Buffer		HnswNewBuffer(Relation index, ForkNumber forkNum);
 void		HnswInitPage(Buffer buf, Page page);
 void		HnswInit(void);
-List	   *HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool inserting, HnswElement skipElement);
+List	   *HnswSearchLayer(char *base, Datum q, IndexTuple qtup, ScanKeyData *keyData, List *ep, int ef, int lc, Relation index, FmgrInfo **procinfo, Oid *collation, int m, bool inserting, HnswElement skipElement, bool inMemory);
 HnswElement HnswGetEntryPoint(Relation index);
 void		HnswGetMetaPageInfo(Relation index, int *m, HnswElement * entryPoint);
 void	   *HnswAlloc(HnswAllocator * allocator, Size size);
 HnswElement HnswInitElement(char *base, ItemPointer tid, int m, double ml, int maxLevel, HnswAllocator * alloc);
 HnswElement HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno);
-void		HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, bool existing);
-HnswCandidate *HnswEntryCandidate(char *base, HnswElement em, Datum q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadVec);
+void		HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo **procinfo, Oid *collation, int m, int efConstruction, bool existing, bool inMemory);
+HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement em, Datum q, IndexTuple qtup, ScanKeyData *keyData, Relation rel, FmgrInfo **procinfo, Oid *collation, bool loadVec, bool inMemory);
 void		HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum, bool building);
 void		HnswSetNeighborTuple(char *base, HnswNeighborTuple ntup, HnswElement e, int m);
 void		HnswAddHeapTid(HnswElement element, ItemPointer heaptid);
+HnswNeighborArray *HnswInitNeighborArray(int lm, HnswAllocator * allocator);
 void		HnswInitNeighbors(char *base, HnswElement element, int m, HnswAllocator * alloc);
 bool		HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull, ItemPointer heap_tid, bool building);
-void		HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting, bool building);
-void		HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec);
-void		HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec, float *maxDistance);
-void		HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element);
-void		HnswUpdateConnection(char *base, HnswElement element, HnswCandidate * hc, int lm, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation);
-void		HnswLoadNeighbors(HnswElement element, Relation index, int m);
+void		HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo **procinfo, Oid *collation, HnswElement e, int m, bool checkExisting, bool building);
+void		HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec, Relation index);
+void		HnswLoadElement(HnswElement element, double *distance, bool *matches, Datum *q, IndexTuple qtup, ScanKeyData *keyData, Relation index, FmgrInfo **procinfo, Oid *collation, bool loadVec, double *maxDistance);
+void		HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element, bool useIndexTuple);
+void		HnswUpdateConnection(char *base, HnswNeighborArray * neighbors, HnswElement newElement, float distance, int lm, int *updateIdx, Relation index, FmgrInfo **procinfo, Oid *collation);
+bool		HnswLoadNeighborTids(HnswElement element, ItemPointerData *indextids, Relation index, int m, int lm, int lc);
 void		HnswInitLockTranche(void);
 const		HnswTypeInfo *HnswGetTypeInfo(Relation index);
 PGDLLEXPORT void HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc);
+void		HnswInitProcinfo(FmgrInfo **procinfo, Oid **collation, Relation index);
+Size		HnswGetElementTupleSize(char *base, HnswElement element, bool useIndexTuple);
+bool		HnswIndexTupleIsEqual(IndexTuple a, IndexTuple b, TupleDesc tupdesc);
 
 /* Index access methods */
 IndexBuildResult *hnswbuild(Relation heap, Relation index, IndexInfo *indexInfo);

src/hnswbuild.c

@@ -60,12 +60,6 @@
 #include "pgstat.h"
 #endif
 
-#if PG_VERSION_NUM >= 130000
-#define CALLBACK_ITEM_POINTER ItemPointer tid
-#else
-#define CALLBACK_ITEM_POINTER HeapTuple hup
-#endif
-
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_status.h"
 #include "utils/wait_event.h"
@@ -75,10 +69,6 @@
 #define PARALLEL_KEY_HNSW_AREA			UINT64CONST(0xA000000000000002)
 #define PARALLEL_KEY_QUERY_TEXT			UINT64CONST(0xA000000000000003)
 
-#if PG_VERSION_NUM < 130000
-#define GENERATIONCHUNK_RAWSIZE (SIZEOF_SIZE_T + SIZEOF_VOID_P * 2)
-#endif
-
 /*
  * Create the metapage
  */
@@ -158,6 +148,7 @@ CreateGraphPages(HnswBuildState * buildstate)
 	Page		page;
 	HnswElementPtr iter = buildstate->graph->head;
 	char	   *base = buildstate->hnswarea;
+	bool		useIndexTuple = buildstate->useIndexTuple;
 
 	/* Calculate sizes */
 	maxSize = HNSW_MAX_SIZE;
@@ -177,7 +168,6 @@ CreateGraphPages(HnswBuildState * buildstate)
 		Size		etupSize;
 		Size		ntupSize;
 		Size		combinedSize;
-		Pointer		valuePtr = HnswPtrAccess(base, element->value);
 
 		/* Update iterator */
 		iter = element->next;
@@ -186,15 +176,17 @@ CreateGraphPages(HnswBuildState * buildstate)
 		MemSet(etup, 0, HNSW_TUPLE_ALLOC_SIZE);
 
 		/* Calculate sizes */
-		etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(valuePtr));
+		etupSize = HnswGetElementTupleSize(base, element, useIndexTuple);
 		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, buildstate->m);
 		combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
 
 		/* Initial size check */
 		if (etupSize > HNSW_TUPLE_ALLOC_SIZE)
-			elog(ERROR, "index tuple too large");
+			ereport(ERROR,
+					(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+					 errmsg("index tuple too large")));
 
-		HnswSetElementTuple(base, etup, element);
+		HnswSetElementTuple(base, etup, element, useIndexTuple);
 
 		/* Keep element and neighbors on the same page if possible */
 		if (PageGetFreeSpace(page) < etupSize || (combinedSize <= maxSize && PageGetFreeSpace(page) < combinedSize))
@@ -335,20 +327,29 @@ AddDuplicateInMemory(HnswElement element, HnswElement dup)
  * Find duplicate element
  */
 static bool
-FindDuplicateInMemory(char *base, HnswElement element)
+FindDuplicateInMemory(char *base, HnswElement element, bool useIndexTuple, TupleDesc tupdesc)
 {
 	HnswNeighborArray *neighbors = HnswGetNeighbors(base, element, 0);
 	Datum		value = HnswGetValue(base, element);
+	IndexTuple	itup = HnswPtrAccess(base, element->itup);
 
 	for (int i = 0; i < neighbors->length; i++)
 	{
 		HnswCandidate *neighbor = &neighbors->items[i];
 		HnswElement neighborElement = HnswPtrAccess(base, neighbor->element);
-		Datum		neighborValue = HnswGetValue(base, neighborElement);
 
-		/* Exit early since ordered by distance */
-		if (!datumIsEqual(value, neighborValue, false, -1))
-			return false;
+		if (useIndexTuple)
+		{
+			/* Exit early since ordered by distance */
+			if (!HnswIndexTupleIsEqual(itup, HnswPtrAccess(base, neighborElement->itup), tupdesc))
+				return false;
+		}
+		else
+		{
+			/* Exit early since ordered by distance */
+			if (!datumIsEqual(value, HnswGetValue(base, neighborElement), false, -1))
+				return false;
+		}
 
 		/* Check for space */
 		if (AddDuplicateInMemory(element, neighborElement))
@@ -374,12 +375,18 @@ AddElementInMemory(char *base, HnswGraph * graph, HnswElement element)
  * Update neighbors
  */
 static void
-UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswElement e, int m)
+UpdateNeighborsInMemory(char *base, Relation index, FmgrInfo **procinfo, Oid *collation, HnswElement e, int m)
 {
 	for (int lc = e->level; lc >= 0; lc--)
 	{
 		int			lm = HnswGetLayerM(m, lc);
-		HnswNeighborArray *neighbors = HnswGetNeighbors(base, e, lc);
+		Size		neighborsSize = HNSW_NEIGHBOR_ARRAY_SIZE(lm);
+		HnswNeighborArray *neighbors = palloc(neighborsSize);
+
+		/* Copy neighbors to local memory */
+		LWLockAcquire(&e->lock, LW_SHARED);
+		memcpy(neighbors, HnswGetNeighbors(base, e, lc), neighborsSize);
+		LWLockRelease(&e->lock);
 
 		for (int i = 0; i < neighbors->length; i++)
 		{
@@ -389,9 +396,8 @@ UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswEleme
 			/* Keep scan-build happy on Mac x86-64 */
 			Assert(neighborElement);
 
-			/* Use element for lock instead of hc since hc can be replaced */
 			LWLockAcquire(&neighborElement->lock, LW_EXCLUSIVE);
-			HnswUpdateConnection(base, e, hc, lm, lc, NULL, NULL, procinfo, collation);
+			HnswUpdateConnection(base, HnswGetNeighbors(base, neighborElement, lc), e, hc->distance, lm, NULL, index, procinfo, collation);
 			LWLockRelease(&neighborElement->lock);
 		}
 	}
@@ -401,20 +407,20 @@ UpdateNeighborsInMemory(char *base, FmgrInfo *procinfo, Oid collation, HnswEleme
  * Update graph in memory
  */
 static void
-UpdateGraphInMemory(FmgrInfo *procinfo, Oid collation, HnswElement element, int m, int efConstruction, HnswElement entryPoint, HnswBuildState * buildstate)
+UpdateGraphInMemory(FmgrInfo **procinfo, Oid *collation, HnswElement element, int m, int efConstruction, HnswElement entryPoint, HnswBuildState * buildstate)
 {
 	HnswGraph  *graph = buildstate->graph;
 	char	   *base = buildstate->hnswarea;
 
 	/* Look for duplicate */
-	if (FindDuplicateInMemory(base, element))
+	if (FindDuplicateInMemory(base, element, buildstate->useIndexTuple, buildstate->tupdesc))
 		return;
 
 	/* Add element */
 	AddElementInMemory(base, graph, element);
 
 	/* Update neighbors */
-	UpdateNeighborsInMemory(base, procinfo, collation, element, m);
+	UpdateNeighborsInMemory(base, buildstate->index, procinfo, collation, element, m);
 
 	/* Update entry point if needed (already have lock) */
 	if (entryPoint == NULL || element->level > entryPoint->level)
@@ -427,8 +433,9 @@ UpdateGraphInMemory(FmgrInfo *procinfo, Oid collation, HnswElement element, int
 static void
 InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 {
-	FmgrInfo   *procinfo = buildstate->procinfo;
-	Oid			collation = buildstate->collation;
+	Relation	index = buildstate->index;
+	FmgrInfo  **procinfo = buildstate->procinfo;
+	Oid		   *collation = buildstate->collation;
 	HnswGraph  *graph = buildstate->graph;
 	HnswElement entryPoint;
 	LWLock	   *entryLock = &graph->entryLock;
@@ -461,7 +468,7 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 	}
 
 	/* Find neighbors for element */
-	HnswFindElementNeighbors(base, element, entryPoint, NULL, procinfo, collation, m, efConstruction, false);
+	HnswFindElementNeighbors(base, element, entryPoint, index, procinfo, collation, m, efConstruction, false, true);
 
 	/* Update graph in memory */
 	UpdateGraphInMemory(procinfo, collation, element, m, efConstruction, entryPoint, buildstate);
@@ -484,6 +491,11 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	Pointer		valuePtr;
 	LWLock	   *flushLock = &graph->flushLock;
 	char	   *base = buildstate->hnswarea;
+	bool		useIndexTuple = buildstate->useIndexTuple;
+	TupleDesc	tupdesc = buildstate->tupdesc;
+	IndexTuple	itup;
+	Size		itupSize;
+	IndexTuple	itupPtr;
 
 	/* Detoast once for all calls */
 	Datum		value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
@@ -495,10 +507,10 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	/* Normalize if needed */
 	if (buildstate->normprocinfo != NULL)
 	{
-		if (!HnswCheckNorm(buildstate->normprocinfo, buildstate->collation, value))
+		if (!HnswCheckNorm(buildstate->normprocinfo, buildstate->collation[0], value))
 			return false;
 
-		value = HnswNormValue(typeInfo, buildstate->collation, value);
+		value = HnswNormValue(typeInfo, buildstate->collation[0], value);
 	}
 
 	/* Get datum size */
@@ -549,7 +561,17 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 
 	/* Ok, we can proceed to allocate the element */
 	element = HnswInitElement(base, heaptid, buildstate->m, buildstate->ml, buildstate->maxLevel, allocator);
-	valuePtr = HnswAlloc(allocator, valueSize);
+
+	if (useIndexTuple)
+	{
+		/* TODO fix */
+		values[0] = value;
+		itup = index_form_tuple(tupdesc, values, isnull);
+		itupSize = IndexTupleSize(itup);
+		itupPtr = HnswAlloc(allocator, itupSize);
+	}
+	else
+		valuePtr = HnswAlloc(allocator, valueSize);
 
 	/*
 	 * We have now allocated the space needed for the element, so we don't
@@ -559,8 +581,19 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	LWLockRelease(&graph->allocatorLock);
 
 	/* Copy the datum */
-	memcpy(valuePtr, DatumGetPointer(value), valueSize);
-	HnswPtrStore(base, element->value, valuePtr);
+	if (useIndexTuple)
+	{
+		bool		unused;
+
+		memcpy(itupPtr, itup, itupSize);
+		HnswPtrStore(base, element->itup, itupPtr);
+		HnswPtrStore(base, element->value, DatumGetPointer(index_getattr(itupPtr, 1, tupdesc, &unused)));
+	}
+	else
+	{
+		memcpy(valuePtr, DatumGetPointer(value), valueSize);
+		HnswPtrStore(base, element->value, valuePtr);
+	}
 
 	/* Create a lock for the element */
 	LWLockInitialize(&element->lock, hnsw_lock_tranche_id);
@@ -578,17 +611,13 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
  * Callback for table_index_build_scan
  */
 static void
-BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
+BuildCallback(Relation index, ItemPointer tid, Datum *values,
 			  bool *isnull, bool tupleIsAlive, void *state)
 {
 	HnswBuildState *buildstate = (HnswBuildState *) state;
 	HnswGraph  *graph = buildstate->graph;
 	MemoryContext oldCtx;
 
-#if PG_VERSION_NUM < 130000
-	ItemPointer tid = &hup->t_self;
-#endif
-
 	/* Skip nulls */
 	if (isnull[0])
 		return;
@@ -614,7 +643,7 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
  * Initialize the graph
  */
 static void
-InitGraph(HnswGraph * graph, char *base, long memoryTotal)
+InitGraph(HnswGraph * graph, char *base, Size memoryTotal)
 {
 	/* Initialize the lock tranche if needed */
 	HnswInitLockTranche();
@@ -651,11 +680,7 @@ HnswMemoryContextAlloc(Size size, void *state)
 	HnswBuildState *buildstate = (HnswBuildState *) state;
 	void	   *chunk = MemoryContextAlloc(buildstate->graphCtx, size);
 
-#if PG_VERSION_NUM >= 130000
 	buildstate->graphData.memoryUsed = MemoryContextMemAllocated(buildstate->graphCtx, false);
-#else
-	buildstate->graphData.memoryUsed += MAXALIGN(size);
-#endif
 
 	return chunk;
 }
@@ -691,30 +716,52 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 
 	/* Disallow varbit since require fixed dimensions */
 	if (TupleDescAttr(index->rd_att, 0)->atttypid == VARBITOID)
-		elog(ERROR, "type not supported for hnsw index");
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("type not supported for hnsw index")));
+
+	/* TODO See if needed */
+	if (IndexRelationGetNumberOfKeyAttributes(index) > 2)
+		elog(ERROR, "index cannot have more than two columns");
+
+	if (!OidIsValid(index_getprocid(index, 1, HNSW_DISTANCE_PROC)))
+		elog(ERROR, "first column must be a vector");
+
+	for (int i = 1; i < IndexRelationGetNumberOfKeyAttributes(index); i++)
+	{
+		if (!OidIsValid(index_getprocid(index, i + 1, HNSW_ATTRIBUTE_DISTANCE_PROC)))
+			elog(ERROR, "column %d cannot be a vector", i + 1);
+	}
 
 	/* Require column to have dimensions to be indexed */
 	if (buildstate->dimensions < 0)
-		elog(ERROR, "column does not have dimensions");
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("column does not have dimensions")));
 
 	if (buildstate->dimensions > buildstate->typeInfo->maxDimensions)
-		elog(ERROR, "column cannot have more than %d dimensions for hnsw index", buildstate->typeInfo->maxDimensions);
+		ereport(ERROR,
+				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+				 errmsg("column cannot have more than %d dimensions for hnsw index", buildstate->typeInfo->maxDimensions)));
 
 	if (buildstate->efConstruction < 2 * buildstate->m)
-		elog(ERROR, "ef_construction must be greater than or equal to 2 * m");
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("ef_construction must be greater than or equal to 2 * m")));
 
 	buildstate->reltuples = 0;
 	buildstate->indtuples = 0;
 
 	/* Get support functions */
-	buildstate->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
+	HnswInitProcinfo(buildstate->procinfo, &buildstate->collation, index);
 	buildstate->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
-	buildstate->collation = index->rd_indcollation[0];
 
-	InitGraph(&buildstate->graphData, NULL, maintenance_work_mem * 1024L);
+	InitGraph(&buildstate->graphData, NULL, (Size) maintenance_work_mem * 1024L);
 	buildstate->graph = &buildstate->graphData;
 	buildstate->ml = HnswGetMl(buildstate->m);
 	buildstate->maxLevel = HnswGetMaxLevel(buildstate->m);
+	buildstate->useIndexTuple = HnswUseIndexTuple(index);
+	buildstate->tupdesc = RelationGetDescr(index);
 
 	buildstate->graphCtx = GenerationContextCreate(CurrentMemoryContext,
 												   "Hnsw build graph context",

src/hnswinsert.c

@@ -36,14 +36,15 @@ GetInsertPage(Relation index)
  * Check for a free offset
  */
 static bool
-HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size ntupSize, Buffer *nbuf, Page *npage, OffsetNumber *freeOffno, OffsetNumber *freeNeighborOffno, BlockNumber *newInsertPage)
+HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size etupSize, Size ntupSize, Buffer *nbuf, Page *npage, OffsetNumber *freeOffno, OffsetNumber *freeNeighborOffno, BlockNumber *newInsertPage)
 {
 	OffsetNumber offno;
 	OffsetNumber maxoffno = PageGetMaxOffsetNumber(page);
 
 	for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
 	{
-		HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
+		ItemId		eitemid = PageGetItemId(page, offno);
+		HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, eitemid);
 
 		/* Skip neighbor tuples */
 		if (!HnswIsElementTuple(etup))
@@ -54,7 +55,9 @@ HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size
 			BlockNumber elementPage = BufferGetBlockNumber(buf);
 			BlockNumber neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
 			OffsetNumber neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
-			ItemId		itemid;
+			ItemId		nitemid;
+			Size		pageFree;
+			Size		npageFree;
 
 			if (!BlockNumberIsValid(*newInsertPage))
 				*newInsertPage = elementPage;
@@ -73,10 +76,25 @@ HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size
 				*npage = BufferGetPage(*nbuf);
 			}
 
-			itemid = PageGetItemId(*npage, neighborOffno);
+			nitemid = PageGetItemId(*npage, neighborOffno);
 
-			/* Check for space on neighbor tuple page */
-			if (PageGetFreeSpace(*npage) + ItemIdGetLength(itemid) - sizeof(ItemIdData) >= ntupSize)
+			/* Ensure aligned for space check */
+			Assert(etupSize == MAXALIGN(etupSize));
+			Assert(ntupSize == MAXALIGN(ntupSize));
+
+			/*
+			 * Calculate free space individually since tuples are overwritten
+			 * individually (in separate calls to PageIndexTupleOverwrite)
+			 */
+			pageFree = ItemIdGetLength(eitemid) + PageGetExactFreeSpace(page);
+			npageFree = ItemIdGetLength(nitemid);
+			if (neighborPage != elementPage)
+				npageFree += PageGetExactFreeSpace(*npage);
+			else if (pageFree >= etupSize)
+				npageFree += pageFree - etupSize;
+
+			/* Check for space */
+			if (pageFree >= etupSize && npageFree >= ntupSize)
 			{
 				*freeOffno = offno;
 				*freeNeighborOffno = neighborOffno;
@@ -136,9 +154,10 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 	OffsetNumber freeNeighborOffno = InvalidOffsetNumber;
 	BlockNumber newInsertPage = InvalidBlockNumber;
 	char	   *base = NULL;
+	bool		useIndexTuple = HnswUseIndexTuple(index);
 
 	/* Calculate sizes */
-	etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(HnswPtrAccess(base, e->value)));
+	etupSize = HnswGetElementTupleSize(base, e, useIndexTuple);
 	ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m);
 	combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
 	maxSize = HNSW_MAX_SIZE;
@@ -146,7 +165,7 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 
 	/* Prepare element tuple */
 	etup = palloc0(etupSize);
-	HnswSetElementTuple(base, etup, e);
+	HnswSetElementTuple(base, etup, e, useIndexTuple);
 
 	/* Prepare neighbor tuple */
 	ntup = palloc0(ntupSize);
@@ -184,7 +203,7 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 		}
 
 		/* Next, try space from a deleted element */
-		if (HnswFreeOffset(index, buf, page, e, ntupSize, &nbuf, &npage, &freeOffno, &freeNeighborOffno, &newInsertPage))
+		if (HnswFreeOffset(index, buf, page, e, etupSize, ntupSize, &nbuf, &npage, &freeOffno, &freeNeighborOffno, &newInsertPage))
 		{
 			if (nbuf != buf)
 			{
@@ -316,6 +335,107 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
 		*updatedInsertPage = newInsertPage;
 }
 
+/*
+ * Load neighbors
+ */
+static HnswNeighborArray *
+HnswLoadNeighbors(HnswElement element, Relation index, int m, int lm, int lc)
+{
+	char	   *base = NULL;
+	HnswNeighborArray *neighbors = HnswInitNeighborArray(lm, NULL);
+	ItemPointerData indextids[HNSW_MAX_M * 2];
+
+	if (!HnswLoadNeighborTids(element, indextids, index, m, lm, lc))
+		return neighbors;
+
+	for (int i = 0; i < lm; i++)
+	{
+		ItemPointer indextid = &indextids[i];
+		HnswElement e;
+		HnswCandidate *hc;
+
+		if (!ItemPointerIsValid(indextid))
+			break;
+
+		e = HnswInitElementFromBlock(ItemPointerGetBlockNumber(indextid), ItemPointerGetOffsetNumber(indextid));
+		hc = &neighbors->items[neighbors->length++];
+		HnswPtrStore(base, hc->element, e);
+	}
+
+	return neighbors;
+}
+
+/*
+ * Load elements for insert
+ */
+static void
+LoadElementsForInsert(HnswNeighborArray * neighbors, Datum q, IndexTuple qtup, int *idx, Relation index, FmgrInfo **procinfo, Oid *collation)
+{
+	char	   *base = NULL;
+
+	for (int i = 0; i < neighbors->length; i++)
+	{
+		HnswCandidate *hc = &neighbors->items[i];
+		HnswElement element = HnswPtrAccess(base, hc->element);
+		double		distance;
+		bool		matches;
+
+		HnswLoadElement(element, &distance, &matches, &q, qtup, NULL, index, procinfo, collation, true, NULL);
+		hc->distance = distance;
+
+		/* Prune element if being deleted */
+		if (element->heaptidsLength == 0)
+		{
+			*idx = i;
+			break;
+		}
+	}
+}
+
+/*
+ * Get update index
+ */
+static int
+GetUpdateIndex(HnswElement element, HnswElement newElement, float distance, int m, int lm, int lc, Relation index, FmgrInfo **procinfo, Oid *collation, MemoryContext updateCtx)
+{
+	char	   *base = NULL;
+	int			idx = -1;
+	HnswNeighborArray *neighbors;
+	MemoryContext oldCtx = MemoryContextSwitchTo(updateCtx);
+
+	/*
+	 * Get latest neighbors since they may have changed. Do not lock yet since
+	 * selecting neighbors can take time. Could use optimistic locking to
+	 * retry if another update occurs before getting exclusive lock.
+	 */
+	neighbors = HnswLoadNeighbors(element, index, m, lm, lc);
+
+	/*
+	 * Could improve performance for vacuuming by checking neighbors against
+	 * list of elements being deleted to find index. It's important to exclude
+	 * already deleted elements for this since they can be replaced at any
+	 * time.
+	 */
+
+	if (neighbors->length < lm)
+		idx = -2;
+	else
+	{
+		Datum		q = HnswGetValue(base, element);
+		IndexTuple	qtup = HnswPtrAccess(base, element->itup);;
+
+		LoadElementsForInsert(neighbors, q, qtup, &idx, index, procinfo, collation);
+
+		if (idx == -1)
+			HnswUpdateConnection(base, neighbors, newElement, distance, lm, &idx, index, procinfo, collation);
+	}
+
+	MemoryContextSwitchTo(oldCtx);
+	MemoryContextReset(updateCtx);
+
+	return idx;
+}
+
 /*
  * Check if connection already exists
  */
@@ -336,14 +456,94 @@ ConnectionExists(HnswElement e, HnswNeighborTuple ntup, int startIdx, int lm)
 	return false;
 }
 
+/*
+ * Update neighbor
+ */
+static void
+UpdateNeighborOnDisk(HnswElement element, HnswElement newElement, int idx, int m, int lm, int lc, Relation index, bool checkExisting, bool building)
+{
+	Buffer		buf;
+	Page		page;
+	GenericXLogState *state;
+	HnswNeighborTuple ntup;
+	int			startIdx;
+	OffsetNumber offno = element->neighborOffno;
+
+	/* Register page */
+	buf = ReadBuffer(index, element->neighborPage);
+	LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
+	if (building)
+	{
+		state = NULL;
+		page = BufferGetPage(buf);
+	}
+	else
+	{
+		state = GenericXLogStart(index);
+		page = GenericXLogRegisterBuffer(state, buf, 0);
+	}
+
+	/* Get tuple */
+	ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, offno));
+
+	/* Calculate index for update */
+	startIdx = (element->level - lc) * m;
+
+	/* Check for existing connection */
+	if (checkExisting && ConnectionExists(newElement, ntup, startIdx, lm))
+		idx = -1;
+	else if (idx == -2)
+	{
+		/* Find free offset if still exists */
+		/* TODO Retry updating connections if not */
+		for (int j = 0; j < lm; j++)
+		{
+			if (!ItemPointerIsValid(&ntup->indextids[startIdx + j]))
+			{
+				idx = startIdx + j;
+				break;
+			}
+		}
+	}
+	else
+		idx += startIdx;
+
+	/* Make robust to issues */
+	if (idx >= 0 && idx < ntup->count)
+	{
+		ItemPointer indextid = &ntup->indextids[idx];
+
+		/* Update neighbor on the buffer */
+		ItemPointerSet(indextid, newElement->blkno, newElement->offno);
+
+		/* Commit */
+		if (building)
+			MarkBufferDirty(buf);
+		else
+			GenericXLogFinish(state);
+	}
+	else if (!building)
+		GenericXLogAbort(state);
+
+	UnlockReleaseBuffer(buf);
+}
+
 /*
  * Update neighbors
  */
 void
-HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting, bool building)
+HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo **procinfo, Oid *collation, HnswElement e, int m, bool checkExisting, bool building)
 {
 	char	   *base = NULL;
 
+	/* Use separate memory context to improve performance for larger vectors */
+	MemoryContext updateCtx = GenerationContextCreate(CurrentMemoryContext,
+													  "Hnsw insert update context",
+#if PG_VERSION_NUM >= 150000
+													  128 * 1024, 128 * 1024,
+#endif
+													  128 * 1024);
+
 	for (int lc = e->level; lc >= 0; lc--)
 	{
 		int			lm = HnswGetLayerM(m, lc);
@@ -352,92 +552,20 @@ HnswUpdateNeighborsOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, Hns
 		for (int i = 0; i < neighbors->length; i++)
 		{
 			HnswCandidate *hc = &neighbors->items[i];
-			Buffer		buf;
-			Page		page;
-			GenericXLogState *state;
-			HnswNeighborTuple ntup;
-			int			idx = -1;
-			int			startIdx;
 			HnswElement neighborElement = HnswPtrAccess(base, hc->element);
-			OffsetNumber offno = neighborElement->neighborOffno;
+			int			idx;
 
-			/* Get latest neighbors since they may have changed */
-			/* Do not lock yet since selecting neighbors can take time */
-			HnswLoadNeighbors(neighborElement, index, m);
-
-			/*
-			 * Could improve performance for vacuuming by checking neighbors
-			 * against list of elements being deleted to find index. It's
-			 * important to exclude already deleted elements for this since
-			 * they can be replaced at any time.
-			 */
-
-			/* Select neighbors */
-			HnswUpdateConnection(NULL, e, hc, lm, lc, &idx, index, procinfo, collation);
+			idx = GetUpdateIndex(neighborElement, e, hc->distance, m, lm, lc, index, procinfo, collation, updateCtx);
 
 			/* New element was not selected as a neighbor */
 			if (idx == -1)
 				continue;
 
-			/* Register page */
-			buf = ReadBuffer(index, neighborElement->neighborPage);
-			LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-			if (building)
-			{
-				state = NULL;
-				page = BufferGetPage(buf);
-			}
-			else
-			{
-				state = GenericXLogStart(index);
-				page = GenericXLogRegisterBuffer(state, buf, 0);
-			}
-
-			/* Get tuple */
-			ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, offno));
-
-			/* Calculate index for update */
-			startIdx = (neighborElement->level - lc) * m;
-
-			/* Check for existing connection */
-			if (checkExisting && ConnectionExists(e, ntup, startIdx, lm))
-				idx = -1;
-			else if (idx == -2)
-			{
-				/* Find free offset if still exists */
-				/* TODO Retry updating connections if not */
-				for (int j = 0; j < lm; j++)
-				{
-					if (!ItemPointerIsValid(&ntup->indextids[startIdx + j]))
-					{
-						idx = startIdx + j;
-						break;
-					}
-				}
-			}
-			else
-				idx += startIdx;
-
-			/* Make robust to issues */
-			if (idx >= 0 && idx < ntup->count)
-			{
-				ItemPointer indextid = &ntup->indextids[idx];
-
-				/* Update neighbor on the buffer */
-				ItemPointerSet(indextid, e->blkno, e->offno);
-
-				/* Commit */
-				if (building)
-					MarkBufferDirty(buf);
-				else
-					GenericXLogFinish(state);
-			}
-			else if (!building)
-				GenericXLogAbort(state);
-
-			UnlockReleaseBuffer(buf);
+			UpdateNeighborOnDisk(neighborElement, e, idx, m, lm, lc, index, checkExisting, building);
 		}
 	}
+
+	MemoryContextDelete(updateCtx);
 }
 
 /*
@@ -505,16 +633,26 @@ FindDuplicateOnDisk(Relation index, HnswElement element, bool building)
 	char	   *base = NULL;
 	HnswNeighborArray *neighbors = HnswGetNeighbors(base, element, 0);
 	Datum		value = HnswGetValue(base, element);
+	IndexTuple	itup = HnswPtrAccess(base, element->itup);
+	TupleDesc	tupdesc = RelationGetDescr(index);
 
 	for (int i = 0; i < neighbors->length; i++)
 	{
 		HnswCandidate *neighbor = &neighbors->items[i];
 		HnswElement neighborElement = HnswPtrAccess(base, neighbor->element);
-		Datum		neighborValue = HnswGetValue(base, neighborElement);
 
-		/* Exit early since ordered by distance */
-		if (!datumIsEqual(value, neighborValue, false, -1))
-			return false;
+		if (HnswUseIndexTuple(index))
+		{
+			/* Exit early since ordered by distance */
+			if (!HnswIndexTupleIsEqual(itup, HnswPtrAccess(base, neighborElement->itup), tupdesc))
+				return false;
+		}
+		else
+		{
+			/* Exit early since ordered by distance */
+			if (!datumIsEqual(value, HnswGetValue(base, neighborElement), false, -1))
+				return false;
+		}
 
 		if (AddDuplicateOnDisk(index, element, neighborElement, building))
 			return true;
@@ -527,7 +665,7 @@ FindDuplicateOnDisk(Relation index, HnswElement element, bool building)
  * Update graph on disk
  */
 static void
-UpdateGraphOnDisk(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement element, int m, int efConstruction, HnswElement entryPoint, bool building)
+UpdateGraphOnDisk(Relation index, FmgrInfo **procinfo, Oid *collation, HnswElement element, int m, int efConstruction, HnswElement entryPoint, bool building)
 {
 	BlockNumber newInsertPage = InvalidBlockNumber;
 
@@ -560,11 +698,13 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
 	HnswElement element;
 	int			m;
 	int			efConstruction = HnswGetEfConstruction(index);
-	FmgrInfo   *procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
-	Oid			collation = index->rd_indcollation[0];
+	FmgrInfo   *procinfo[2];
+	Oid		   *collation;
 	LOCKMODE	lockmode = ShareLock;
 	char	   *base = NULL;
 
+	HnswInitProcinfo(procinfo, &collation, index);
+
 	/*
 	 * Get a shared lock. This allows vacuum to ensure no in-flight inserts
 	 * before repairing graph. Use a page lock so it does not interfere with
@@ -577,7 +717,23 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
 
 	/* Create an element */
 	element = HnswInitElement(base, heap_tid, m, HnswGetMl(m), HnswGetMaxLevel(m), NULL);
-	HnswPtrStore(base, element->value, DatumGetPointer(value));
+	if (HnswUseIndexTuple(index))
+	{
+		/* TODO no toast */
+		TupleDesc	tupdesc = RelationGetDescr(index);
+		IndexTuple	itup;
+		bool		unused;
+
+		/* TODO fix */
+		values[0] = value;
+		itup = index_form_tuple(tupdesc, values, isnull);
+
+		HnswPtrStore(base, element->itup, itup);
+		HnswPtrStore(base, element->value, DatumGetPointer(index_getattr(itup, 1, tupdesc, &unused)));
+
+	}
+	else
+		HnswPtrStore(base, element->value, DatumGetPointer(value));
 
 	/* Prevent concurrent inserts when likely updating entry point */
 	if (entryPoint == NULL || element->level > entryPoint->level)
@@ -594,7 +750,7 @@ HnswInsertTupleOnDisk(Relation index, Datum value, Datum *values, bool *isnull,
 	}
 
 	/* Find neighbors for element */
-	HnswFindElementNeighbors(base, element, entryPoint, index, procinfo, collation, m, efConstruction, false);
+	HnswFindElementNeighbors(base, element, entryPoint, index, procinfo, collation, m, efConstruction, false, false);
 
 	/* Update graph on disk */
 	UpdateGraphOnDisk(index, procinfo, collation, element, m, efConstruction, entryPoint, building);
@@ -614,7 +770,7 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
 	Datum		value;
 	const		HnswTypeInfo *typeInfo = HnswGetTypeInfo(index);
 	FmgrInfo   *normprocinfo;
-	Oid			collation = index->rd_indcollation[0];
+	Oid		   *collation = index->rd_indcollation;
 
 	/* Detoast once for all calls */
 	value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
@@ -627,10 +783,10 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
 	normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
 	if (normprocinfo != NULL)
 	{
-		if (!HnswCheckNorm(normprocinfo, collation, value))
+		if (!HnswCheckNorm(normprocinfo, collation[0], value))
 			return;
 
-		value = HnswNormValue(typeInfo, collation, value);
+		value = HnswNormValue(typeInfo, collation[0], value);
 	}
 
 	HnswInsertTupleOnDisk(index, value, values, isnull, heap_tid, false);

src/hnswscan.c

@@ -15,13 +15,15 @@ GetScanItems(IndexScanDesc scan, Datum q)
 {
 	HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
 	Relation	index = scan->indexRelation;
-	FmgrInfo   *procinfo = so->procinfo;
-	Oid			collation = so->collation;
+	FmgrInfo  **procinfo = so->procinfo;
+	Oid		   *collation = so->collation;
 	List	   *ep;
 	List	   *w;
 	int			m;
 	HnswElement entryPoint;
 	char	   *base = NULL;
+	bool		inMemory = false;
+	ScanKeyData *keyData = scan->keyData;
 
 	/* Get m and entry point */
 	HnswGetMetaPageInfo(index, &m, &entryPoint);
@@ -29,15 +31,15 @@ GetScanItems(IndexScanDesc scan, Datum q)
 	if (entryPoint == NULL)
 		return NIL;
 
-	ep = list_make1(HnswEntryCandidate(base, entryPoint, q, index, procinfo, collation, false));
+	ep = list_make1(HnswEntryCandidate(base, entryPoint, q, NULL, keyData, index, procinfo, collation, false, inMemory));
 
 	for (int lc = entryPoint->level; lc >= 1; lc--)
 	{
-		w = HnswSearchLayer(base, q, ep, 1, lc, index, procinfo, collation, m, false, NULL);
+		w = HnswSearchLayer(base, q, NULL, keyData, ep, 1, lc, index, procinfo, collation, m, false, NULL, inMemory);
 		ep = w;
 	}
 
-	return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL);
+	return HnswSearchLayer(base, q, NULL, keyData, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL, inMemory);
 }
 
 /*
@@ -61,7 +63,7 @@ GetScanValue(IndexScanDesc scan)
 
 		/* Normalize if needed */
 		if (so->normprocinfo != NULL)
-			value = HnswNormValue(so->typeInfo, so->collation, value);
+			value = HnswNormValue(so->typeInfo, so->collation[0], value);
 	}
 
 	return value;
@@ -86,9 +88,8 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
 									   ALLOCSET_DEFAULT_SIZES);
 
 	/* Set support functions */
-	so->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
+	HnswInitProcinfo(so->procinfo, &so->collation, index);
 	so->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
-	so->collation = index->rd_indcollation[0];
 
 	scan->opaque = so;
 
@@ -160,20 +161,20 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
 
 		so->first = false;
 
-#if defined(HNSW_MEMORY) && PG_VERSION_NUM >= 130000
-		elog(INFO, "memory: %zu MB", MemoryContextMemAllocated(so->tmpCtx, false) / (1024 * 1024));
+#if defined(HNSW_MEMORY)
+		elog(INFO, "memory: %zu KB", MemoryContextMemAllocated(so->tmpCtx, false) / 1024);
 #endif
 	}
 
 	while (list_length(so->w) > 0)
 	{
 		char	   *base = NULL;
-		HnswCandidate *hc = llast(so->w);
+		HnswSearchCandidate *hc = llast(so->w);
 		HnswElement element = HnswPtrAccess(base, hc->element);
 		ItemPointer heaptid;
 
 		/* Move to next element if no valid heap TIDs */
-		if (element->heaptidsLength == 0)
+		if (!hc->matches || element->heaptidsLength == 0)
 		{
 			so->w = list_delete_last(so->w);
 			continue;

src/hnswvacuum.c

@@ -189,8 +189,8 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
 	GenericXLogState *state;
 	int			m = vacuumstate->m;
 	int			efConstruction = vacuumstate->efConstruction;
-	FmgrInfo   *procinfo = vacuumstate->procinfo;
-	Oid			collation = vacuumstate->collation;
+	FmgrInfo  **procinfo = vacuumstate->procinfo;
+	Oid		   *collation = vacuumstate->collation;
 	BufferAccessStrategy bas = vacuumstate->bas;
 	HnswNeighborTuple ntup = vacuumstate->ntup;
 	Size		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, m);
@@ -205,7 +205,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
 	element->heaptidsLength = 0;
 
 	/* Find neighbors for element, skipping itself */
-	HnswFindElementNeighbors(base, element, entryPoint, index, procinfo, collation, m, efConstruction, true);
+	HnswFindElementNeighbors(base, element, entryPoint, index, procinfo, collation, m, efConstruction, true, false);
 
 	/* Zero memory for each element */
 	MemSet(ntup, 0, HNSW_TUPLE_ALLOC_SIZE);
@@ -256,7 +256,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
 		LockPage(index, HNSW_UPDATE_LOCK, ShareLock);
 
 		/* Load element */
-		HnswLoadElement(highestPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true, NULL);
+		HnswLoadElement(highestPoint, NULL, NULL, NULL, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true, NULL);
 
 		/* Repair if needed */
 		if (NeedsUpdated(vacuumstate, highestPoint))
@@ -294,7 +294,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
 			 * is outdated, this can remove connections at higher levels in
 			 * the graph until they are repaired, but this should be fine.
 			 */
-			HnswLoadElement(entryPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true, NULL);
+			HnswLoadElement(entryPoint, NULL, NULL, NULL, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true, NULL);
 
 			if (NeedsUpdated(vacuumstate, entryPoint))
 			{
@@ -370,7 +370,7 @@ RepairGraph(HnswVacuumState * vacuumstate)
 
 			/* Create an element */
 			element = HnswInitElementFromBlock(blkno, offno);
-			HnswLoadElementFromTuple(element, etup, false, true);
+			HnswLoadElementFromTuple(element, etup, false, true, index);
 
 			elements = lappend(elements, element);
 		}
@@ -440,6 +440,7 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 	BlockNumber insertPage = InvalidBlockNumber;
 	Relation	index = vacuumstate->index;
 	BufferAccessStrategy bas = vacuumstate->bas;
+	bool		useIndexTuple = HnswUseIndexTuple(index);
 
 	/*
 	 * Wait for index scans to complete. Scans before this point may contain
@@ -521,7 +522,14 @@ MarkDeleted(HnswVacuumState * vacuumstate)
 
 			/* Overwrite element */
 			etup->deleted = 1;
-			MemSet(&etup->data, 0, VARSIZE_ANY(&etup->data));
+			if (useIndexTuple)
+			{
+				IndexTuple	itup = (IndexTuple) &etup->data;
+
+				MemSet(itup, 0, IndexTupleSize(itup));
+			}
+			else
+				MemSet(&etup->data, 0, VARSIZE_ANY(&etup->data));
 
 			/* Overwrite neighbors */
 			for (int i = 0; i < ntup->count; i++)
@@ -573,8 +581,7 @@ InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkD
 	vacuumstate->callback_state = callback_state;
 	vacuumstate->efConstruction = HnswGetEfConstruction(index);
 	vacuumstate->bas = GetAccessStrategy(BAS_BULKREAD);
-	vacuumstate->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
-	vacuumstate->collation = index->rd_indcollation[0];
+	HnswInitProcinfo(vacuumstate->procinfo, &vacuumstate->collation, index);
 	vacuumstate->ntup = palloc0(HNSW_TUPLE_ALLOC_SIZE);
 	vacuumstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
 												"Hnsw vacuum temporary context",

src/ivfbuild.c

@@ -26,12 +26,6 @@
 #include "pgstat.h"
 #endif
 
-#if PG_VERSION_NUM >= 130000
-#define CALLBACK_ITEM_POINTER ItemPointer tid
-#else
-#define CALLBACK_ITEM_POINTER HeapTuple hup
-#endif
-
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_status.h"
 #include "utils/wait_event.h"
@@ -96,7 +90,7 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
  * Callback for sampling
  */
 static void
-SampleCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
+SampleCallback(Relation index, ItemPointer tid, Datum *values,
 			   bool *isnull, bool tupleIsAlive, void *state)
 {
 	IvfflatBuildState *buildstate = (IvfflatBuildState *) state;
@@ -207,16 +201,12 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
  * Callback for table_index_build_scan
  */
 static void
-BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
+BuildCallback(Relation index, ItemPointer tid, Datum *values,
 			  bool *isnull, bool tupleIsAlive, void *state)
 {
 	IvfflatBuildState *buildstate = (IvfflatBuildState *) state;
 	MemoryContext oldCtx;
 
-#if PG_VERSION_NUM < 130000
-	ItemPointer tid = &hup->t_self;
-#endif
-
 	/* Skip nulls */
 	if (isnull[0])
 		return;
@@ -335,14 +325,20 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 
 	/* Disallow varbit since require fixed dimensions */
 	if (TupleDescAttr(index->rd_att, 0)->atttypid == VARBITOID)
-		elog(ERROR, "type not supported for ivfflat index");
+		ereport(ERROR,
+				(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+				 errmsg("type not supported for ivfflat index")));
 
 	/* Require column to have dimensions to be indexed */
 	if (buildstate->dimensions < 0)
-		elog(ERROR, "column does not have dimensions");
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("column does not have dimensions")));
 
 	if (buildstate->dimensions > buildstate->typeInfo->maxDimensions)
-		elog(ERROR, "column cannot have more than %d dimensions for ivfflat index", buildstate->typeInfo->maxDimensions);
+		ereport(ERROR,
+				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
+				 errmsg("column cannot have more than %d dimensions for ivfflat index", buildstate->typeInfo->maxDimensions)));
 
 	buildstate->reltuples = 0;
 	buildstate->indtuples = 0;
@@ -355,7 +351,9 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 
 	/* Require more than one dimension for spherical k-means */
 	if (buildstate->kmeansnormprocinfo != NULL && buildstate->dimensions == 1)
-		elog(ERROR, "dimensions must be greater than one for this opclass");
+		ereport(ERROR,
+				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+				 errmsg("dimensions must be greater than one for this opclass")));
 
 	/* Create tuple description for sorting */
 	buildstate->tupdesc = CreateTemplateTupleDesc(3);
@@ -562,6 +560,20 @@ PrintKmeansMetrics(IvfflatBuildState * buildstate)
 }
 #endif
 
+/*
+ * Initialize build sort state
+ */
+static Tuplesortstate *
+InitBuildSortState(TupleDesc tupdesc, int memory, SortCoordinate coordinate)
+{
+	AttrNumber	attNums[] = {1};
+	Oid			sortOperators[] = {Int4LessOperator};
+	Oid			sortCollations[] = {InvalidOid};
+	bool		nullsFirstFlags[] = {false};
+
+	return tuplesort_begin_heap(tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, memory, coordinate, false);
+}
+
 /*
  * Within leader, wait for end of heap scan
  */
@@ -609,12 +621,6 @@ IvfflatParallelScanAndSort(IvfflatSpool * ivfspool, IvfflatShared * ivfshared, S
 	double		reltuples;
 	IndexInfo  *indexInfo;
 
-	/* Sort options, which must match AssignTuples */
-	AttrNumber	attNums[] = {1};
-	Oid			sortOperators[] = {Int4LessOperator};
-	Oid			sortCollations[] = {InvalidOid};
-	bool		nullsFirstFlags[] = {false};
-
 	/* Initialize local tuplesort coordination state */
 	coordinate = palloc0(sizeof(SortCoordinateData));
 	coordinate->isWorker = true;
@@ -627,7 +633,7 @@ IvfflatParallelScanAndSort(IvfflatSpool * ivfspool, IvfflatShared * ivfshared, S
 	InitBuildState(&buildstate, ivfspool->heap, ivfspool->index, indexInfo);
 	memcpy(buildstate.centers->items, ivfcenters, buildstate.centers->itemsize * buildstate.centers->maxlen);
 	buildstate.centers->length = buildstate.centers->maxlen;
-	ivfspool->sortstate = tuplesort_begin_heap(buildstate.tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, sortmem, coordinate, false);
+	ivfspool->sortstate = InitBuildSortState(buildstate.tupdesc, sortmem, coordinate);
 	buildstate.sortstate = ivfspool->sortstate;
 	scan = table_beginscan_parallel(ivfspool->heap,
 									ParallelTableScanFromIvfflatShared(ivfshared));
@@ -924,12 +930,6 @@ AssignTuples(IvfflatBuildState * buildstate)
 	int			parallel_workers = 0;
 	SortCoordinate coordinate = NULL;
 
-	/* Sort options, which must match IvfflatParallelScanAndSort */
-	AttrNumber	attNums[] = {1};
-	Oid			sortOperators[] = {Int4LessOperator};
-	Oid			sortCollations[] = {InvalidOid};
-	bool		nullsFirstFlags[] = {false};
-
 	pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_ASSIGN);
 
 	/* Calculate parallel workers */
@@ -950,7 +950,7 @@ AssignTuples(IvfflatBuildState * buildstate)
 	}
 
 	/* Begin serial/leader tuplesort */
-	buildstate->sortstate = tuplesort_begin_heap(buildstate->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, maintenance_work_mem, coordinate, false);
+	buildstate->sortstate = InitBuildSortState(buildstate->tupdesc, maintenance_work_mem, coordinate);
 
 	/* Add tuples to sort */
 	if (buildstate->heap != NULL)

src/ivfflat.c

@@ -7,6 +7,7 @@
 #include "commands/progress.h"
 #include "commands/vacuum.h"
 #include "ivfflat.h"
+#include "utils/float.h"
 #include "utils/guc.h"
 #include "utils/selfuncs.h"
 #include "utils/spccache.h"
@@ -26,11 +27,7 @@ IvfflatInit(void)
 {
 	ivfflat_relopt_kind = add_reloption_kind();
 	add_int_reloption(ivfflat_relopt_kind, "lists", "Number of inverted lists",
-					  IVFFLAT_DEFAULT_LISTS, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS
-#if PG_VERSION_NUM >= 130000
-					  ,AccessExclusiveLock
-#endif
-		);
+					  IVFFLAT_DEFAULT_LISTS, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS, AccessExclusiveLock);
 
 	DefineCustomIntVariable("ivfflat.probes", "Sets the number of probes",
 							"Valid range is 1..lists.", &ivfflat_probes,
@@ -72,14 +69,16 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	GenericCosts costs;
 	int			lists;
 	double		ratio;
+	double		sequentialRatio = 0.5;
+	double		startupPages;
 	double		spc_seq_page_cost;
 	Relation	index;
 
 	/* Never use index without order */
 	if (path->indexorderbys == NULL)
 	{
-		*indexStartupCost = DBL_MAX;
-		*indexTotalCost = DBL_MAX;
+		*indexStartupCost = get_float8_infinity();
+		*indexTotalCost = get_float8_infinity();
 		*indexSelectivity = 0;
 		*indexCorrelation = 0;
 		*indexPages = 0;
@@ -88,6 +87,8 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 
 	MemSet(&costs, 0, sizeof(costs));
 
+	genericcostestimate(root, path, loop_count, &costs);
+
 	index = index_open(path->indexinfo->indexoid, NoLock);
 	IvfflatGetMetaPageInfo(index, &lists, NULL);
 	index_close(index, NoLock);
@@ -97,41 +98,26 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	if (ratio > 1.0)
 		ratio = 1.0;
 
-	/*
-	 * This gives us the subset of tuples to visit. This value is passed into
-	 * the generic cost estimator to determine the number of pages to visit
-	 * during the index scan.
-	 */
-	costs.numIndexTuples = path->indexinfo->tuples * ratio;
-
-	genericcostestimate(root, path, loop_count, &costs);
-
 	get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
 
+	/* Change some page cost from random to sequential */
+	costs.indexTotalCost -= sequentialRatio * costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+
+	/* Startup cost is cost before returning the first row */
+	costs.indexStartupCost = costs.indexTotalCost * ratio;
+
 	/* Adjust cost if needed since TOAST not included in seq scan cost */
-	if (costs.numIndexPages > path->indexinfo->rel->pages && ratio < 0.5)
+	startupPages = costs.numIndexPages * ratio;
+	if (startupPages > path->indexinfo->rel->pages && ratio < 0.5)
 	{
-		/* Change all page cost from random to sequential */
-		costs.indexTotalCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+		/* Change rest of page cost from random to sequential */
+		costs.indexStartupCost -= (1 - sequentialRatio) * startupPages * (costs.spc_random_page_cost - spc_seq_page_cost);
 
 		/* Remove cost of extra pages */
-		costs.indexTotalCost -= (costs.numIndexPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
-	}
-	else
-	{
-		/* Change some page cost from random to sequential */
-		costs.indexTotalCost -= 0.5 * costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+		costs.indexStartupCost -= (startupPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
 	}
 
-	/*
-	 * If the list selectivity is lower than what is returned from the generic
-	 * cost estimator, use that.
-	 */
-	if (ratio < costs.indexSelectivity)
-		costs.indexSelectivity = ratio;
-
-	/* Use total cost since most work happens before first tuple is returned */
-	*indexStartupCost = costs.indexTotalCost;
+	*indexStartupCost = costs.indexStartupCost;
 	*indexTotalCost = costs.indexTotalCost;
 	*indexSelectivity = costs.indexSelectivity;
 	*indexCorrelation = costs.indexCorrelation;
@@ -148,23 +134,10 @@ ivfflatoptions(Datum reloptions, bool validate)
 		{"lists", RELOPT_TYPE_INT, offsetof(IvfflatOptions, lists)},
 	};
 
-#if PG_VERSION_NUM >= 130000
 	return (bytea *) build_reloptions(reloptions, validate,
 									  ivfflat_relopt_kind,
 									  sizeof(IvfflatOptions),
 									  tab, lengthof(tab));
-#else
-	relopt_value *options;
-	int			numoptions;
-	IvfflatOptions *rdopts;
-
-	options = parseRelOptions(reloptions, validate, ivfflat_relopt_kind, &numoptions);
-	rdopts = allocateReloptStruct(sizeof(IvfflatOptions), options, numoptions);
-	fillRelOptions((void *) rdopts, sizeof(IvfflatOptions), options, numoptions,
-				   validate, tab, lengthof(tab));
-
-	return (bytea *) rdopts;
-#endif
 }
 
 /*
@@ -181,7 +154,7 @@ ivfflatvalidate(Oid opclassoid)
  *
  * See https://www.postgresql.org/docs/current/index-api.html
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(ivfflathandler);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflathandler);
 Datum
 ivfflathandler(PG_FUNCTION_ARGS)
 {
@@ -189,9 +162,7 @@ ivfflathandler(PG_FUNCTION_ARGS)
 
 	amroutine->amstrategies = 0;
 	amroutine->amsupport = 5;
-#if PG_VERSION_NUM >= 130000
 	amroutine->amoptsprocnum = 0;
-#endif
 	amroutine->amcanorder = false;
 	amroutine->amcanorderbyop = true;
 	amroutine->amcanbackward = false;	/* can change direction mid-scan */
@@ -204,17 +175,24 @@ ivfflathandler(PG_FUNCTION_ARGS)
 	amroutine->amclusterable = false;
 	amroutine->ampredlocks = false;
 	amroutine->amcanparallel = false;
-	amroutine->amcaninclude = false;
-#if PG_VERSION_NUM >= 130000
-	amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
-	amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
+#if PG_VERSION_NUM >= 170000
+	amroutine->amcanbuildparallel = true;
 #endif
+	amroutine->amcaninclude = false;
+	amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
+#if PG_VERSION_NUM >= 160000
+	amroutine->amsummarizing = false;
+#endif
+	amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
 	amroutine->amkeytype = InvalidOid;
 
 	/* Interface functions */
 	amroutine->ambuild = ivfflatbuild;
 	amroutine->ambuildempty = ivfflatbuildempty;
 	amroutine->aminsert = ivfflatinsert;
+#if PG_VERSION_NUM >= 170000
+	amroutine->aminsertcleanup = NULL;
+#endif
 	amroutine->ambulkdelete = ivfflatbulkdelete;
 	amroutine->amvacuumcleanup = ivfflatvacuumcleanup;
 	amroutine->amcanreturn = NULL;	/* tuple not included in heapsort */

src/ivfflat.h

@@ -253,8 +253,9 @@ typedef struct IvfflatScanOpaqueData
 	/* Sorting */
 	Tuplesortstate *sortstate;
 	TupleDesc	tupdesc;
-	TupleTableSlot *slot;
-	bool		isnull;
+	TupleTableSlot *vslot;
+	TupleTableSlot *mslot;
+	BufferAccessStrategy bas;
 
 	/* Support functions */
 	FmgrInfo   *procinfo;

src/ivfkmeans.c

@@ -151,12 +151,8 @@ RandomCenters(Relation index, VectorArray centers, const IvfflatTypeInfo * typeI
 static void
 ShowMemoryUsage(MemoryContext context, Size estimatedSize)
 {
-#if PG_VERSION_NUM >= 130000
 	elog(INFO, "total memory: %zu MB",
 		 MemoryContextMemAllocated(context, true) / (1024 * 1024));
-#else
-	MemoryContextStats(context);
-#endif
 	elog(INFO, "estimated memory: %zu MB", estimatedSize / (1024 * 1024));
 }
 #endif
@@ -327,7 +323,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
 	newCenters->length = numCenters;
 
 #ifdef IVFFLAT_MEMORY
-	ShowMemoryUsage(MemoryContextGetParent(CurrentMemoryContext));
+	ShowMemoryUsage(MemoryContextGetParent(CurrentMemoryContext), totalSize);
 #endif
 
 	/* Pick initial centers */

src/ivfscan.c

@@ -11,16 +11,23 @@
 #include "pgstat.h"
 #include "storage/bufmgr.h"
 
+#ifdef IVFFLAT_MEMORY
+#include "utils/memutils.h"
+#endif
+
+#define GetScanList(ptr) pairingheap_container(IvfflatScanList, ph_node, ptr)
+#define GetScanListConst(ptr) pairingheap_const_container(IvfflatScanList, ph_node, ptr)
+
 /*
  * Compare list distances
  */
 static int
 CompareLists(const pairingheap_node *a, const pairingheap_node *b, void *arg)
 {
-	if (((const IvfflatScanList *) a)->distance > ((const IvfflatScanList *) b)->distance)
+	if (GetScanListConst(a)->distance > GetScanListConst(b)->distance)
 		return 1;
 
-	if (((const IvfflatScanList *) a)->distance < ((const IvfflatScanList *) b)->distance)
+	if (GetScanListConst(a)->distance < GetScanListConst(b)->distance)
 		return -1;
 
 	return 0;
@@ -72,14 +79,14 @@ GetScanLists(IndexScanDesc scan, Datum value)
 
 				/* Calculate max distance */
 				if (listCount == so->probes)
-					maxDistance = ((IvfflatScanList *) pairingheap_first(so->listQueue))->distance;
+					maxDistance = GetScanList(pairingheap_first(so->listQueue))->distance;
 			}
 			else if (distance < maxDistance)
 			{
 				IvfflatScanList *scanlist;
 
 				/* Remove */
-				scanlist = (IvfflatScanList *) pairingheap_remove_first(so->listQueue);
+				scanlist = GetScanList(pairingheap_remove_first(so->listQueue));
 
 				/* Reuse */
 				scanlist->startPage = list->startPage;
@@ -87,7 +94,7 @@ GetScanLists(IndexScanDesc scan, Datum value)
 				pairingheap_add(so->listQueue, &scanlist->ph_node);
 
 				/* Update max distance */
-				maxDistance = ((IvfflatScanList *) pairingheap_first(so->listQueue))->distance;
+				maxDistance = GetScanList(pairingheap_first(so->listQueue))->distance;
 			}
 		}
 
@@ -106,19 +113,12 @@ GetScanItems(IndexScanDesc scan, Datum value)
 	IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
 	TupleDesc	tupdesc = RelationGetDescr(scan->indexRelation);
 	double		tuples = 0;
-	TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
-
-	/*
-	 * Reuse same set of shared buffers for scan
-	 *
-	 * See postgres/src/backend/storage/buffer/README for description
-	 */
-	BufferAccessStrategy bas = GetAccessStrategy(BAS_BULKREAD);
+	TupleTableSlot *slot = so->vslot;
 
 	/* Search closest probes lists */
 	while (!pairingheap_is_empty(so->listQueue))
 	{
-		BlockNumber searchPage = ((IvfflatScanList *) pairingheap_remove_first(so->listQueue))->startPage;
+		BlockNumber searchPage = GetScanList(pairingheap_remove_first(so->listQueue))->startPage;
 
 		/* Search all entry pages for list */
 		while (BlockNumberIsValid(searchPage))
@@ -127,7 +127,7 @@ GetScanItems(IndexScanDesc scan, Datum value)
 			Page		page;
 			OffsetNumber maxoffno;
 
-			buf = ReadBufferExtended(scan->indexRelation, MAIN_FORKNUM, searchPage, RBM_NORMAL, bas);
+			buf = ReadBufferExtended(scan->indexRelation, MAIN_FORKNUM, searchPage, RBM_NORMAL, so->bas);
 			LockBuffer(buf, BUFFER_LOCK_SHARE);
 			page = BufferGetPage(buf);
 			maxoffno = PageGetMaxOffsetNumber(page);
@@ -166,8 +166,6 @@ GetScanItems(IndexScanDesc scan, Datum value)
 		}
 	}
 
-	FreeAccessStrategy(bas);
-
 	if (tuples < 100)
 		ereport(DEBUG1,
 				(errmsg("index scan found few tuples"),
@@ -217,6 +215,20 @@ GetScanValue(IndexScanDesc scan)
 	return value;
 }
 
+/*
+ * Initialize scan sort state
+ */
+static Tuplesortstate *
+InitScanSortState(TupleDesc tupdesc)
+{
+	AttrNumber	attNums[] = {1};
+	Oid			sortOperators[] = {Float8LessOperator};
+	Oid			sortCollations[] = {InvalidOid};
+	bool		nullsFirstFlags[] = {false};
+
+	return tuplesort_begin_heap(tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
+}
+
 /*
  * Prepare for an index scan
  */
@@ -227,10 +239,6 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	IvfflatScanOpaque so;
 	int			lists;
 	int			dimensions;
-	AttrNumber	attNums[] = {1};
-	Oid			sortOperators[] = {Float8LessOperator};
-	Oid			sortCollations[] = {InvalidOid};
-	bool		nullsFirstFlags[] = {false};
 	int			probes = ivfflat_probes;
 
 	scan = RelationGetIndexScan(index, nkeys, norderbys);
@@ -258,9 +266,18 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "heaptid", TIDOID, -1, 0);
 
 	/* Prep sort */
-	so->sortstate = tuplesort_begin_heap(so->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
+	so->sortstate = InitScanSortState(so->tupdesc);
 
-	so->slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsMinimalTuple);
+	/* Need separate slots for puttuple and gettuple */
+	so->vslot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
+	so->mslot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsMinimalTuple);
+
+	/*
+	 * Reuse same set of shared buffers for scan
+	 *
+	 * See postgres/src/backend/storage/buffer/README for description
+	 */
+	so->bas = GetAccessStrategy(BAS_BULKREAD);
 
 	so->listQueue = pairingheap_allocate(CompareLists, scan);
 
@@ -277,10 +294,8 @@ ivfflatrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int
 {
 	IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
 
-#if PG_VERSION_NUM >= 130000
 	if (!so->first)
 		tuplesort_reset(so->sortstate);
-#endif
 
 	so->first = true;
 	pairingheap_reset(so->listQueue);
@@ -327,14 +342,19 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 		IvfflatBench("GetScanItems", GetScanItems(scan, value));
 		so->first = false;
 
+#if defined(IVFFLAT_MEMORY)
+		elog(INFO, "memory: %zu MB", MemoryContextMemAllocated(CurrentMemoryContext, true) / (1024 * 1024));
+#endif
+
 		/* Clean up if we allocated a new value */
 		if (value != scan->orderByData->sk_argument)
 			pfree(DatumGetPointer(value));
 	}
 
-	if (tuplesort_gettupleslot(so->sortstate, true, false, so->slot, NULL))
+	if (tuplesort_gettupleslot(so->sortstate, true, false, so->mslot, NULL))
 	{
-		ItemPointer heaptid = (ItemPointer) DatumGetPointer(slot_getattr(so->slot, 2, &so->isnull));
+		bool		isnull;
+		ItemPointer heaptid = (ItemPointer) DatumGetPointer(slot_getattr(so->mslot, 2, &isnull));
 
 		scan->xs_heaptid = *heaptid;
 		scan->xs_recheck = false;
@@ -355,6 +375,10 @@ ivfflatendscan(IndexScanDesc scan)
 
 	pairingheap_free(so->listQueue);
 	tuplesort_end(so->sortstate);
+	FreeAccessStrategy(so->bas);
+	FreeTupleDesc(so->tupdesc);
+
+	/* TODO Free vslot and mslot without freeing TupleDesc */
 
 	pfree(so);
 	scan->opaque = NULL;

src/ivfutils.c

@@ -342,7 +342,7 @@ IvfflatGetTypeInfo(Relation index)
 		return (const IvfflatTypeInfo *) DatumGetPointer(FunctionCall0Coll(procinfo, InvalidOid));
 }
 
-PGDLLEXPORT PG_FUNCTION_INFO_V1(ivfflat_halfvec_support);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflat_halfvec_support);
 Datum
 ivfflat_halfvec_support(PG_FUNCTION_ARGS)
 {
@@ -357,7 +357,7 @@ ivfflat_halfvec_support(PG_FUNCTION_ARGS)
 	PG_RETURN_POINTER(&typeInfo);
 };
 
-PGDLLEXPORT PG_FUNCTION_INFO_V1(ivfflat_bit_support);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflat_bit_support);
 Datum
 ivfflat_bit_support(PG_FUNCTION_ARGS)
 {

src/sparsevec.c

@@ -3,6 +3,7 @@
 #include <limits.h>
 #include <math.h>
 
+#include "catalog/pg_type.h"
 #include "common/string.h"
 #include "fmgr.h"
 #include "halfutils.h"
@@ -11,6 +12,7 @@
 #include "sparsevec.h"
 #include "utils/array.h"
 #include "utils/builtins.h"
+#include "utils/lsyscache.h"
 #include "vector.h"
 
 #if PG_VERSION_NUM >= 120000
@@ -188,7 +190,7 @@ CompareIndices(const void *a, const void *b)
 /*
  * Convert textual representation to internal representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_in);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_in);
 Datum
 sparsevec_in(PG_FUNCTION_ARGS)
 {
@@ -409,7 +411,7 @@ sparsevec_in(PG_FUNCTION_ARGS)
 /*
  * Convert internal representation to textual representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_out);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_out);
 Datum
 sparsevec_out(PG_FUNCTION_ARGS)
 {
@@ -462,7 +464,7 @@ sparsevec_out(PG_FUNCTION_ARGS)
 /*
  * Convert type modifier
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_typmod_in);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_typmod_in);
 Datum
 sparsevec_typmod_in(PG_FUNCTION_ARGS)
 {
@@ -493,7 +495,7 @@ sparsevec_typmod_in(PG_FUNCTION_ARGS)
 /*
  * Convert external binary representation to internal representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_recv);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_recv);
 Datum
 sparsevec_recv(PG_FUNCTION_ARGS)
 {
@@ -545,7 +547,7 @@ sparsevec_recv(PG_FUNCTION_ARGS)
 /*
  * Convert internal representation to the external binary representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_send);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_send);
 Datum
 sparsevec_send(PG_FUNCTION_ARGS)
 {
@@ -572,7 +574,7 @@ sparsevec_send(PG_FUNCTION_ARGS)
  * Convert sparse vector to sparse vector
  * This is needed to check the type modifier
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec);
 Datum
 sparsevec(PG_FUNCTION_ARGS)
 {
@@ -587,7 +589,7 @@ sparsevec(PG_FUNCTION_ARGS)
 /*
  * Convert dense vector to sparse vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_to_sparsevec);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_to_sparsevec);
 Datum
 vector_to_sparsevec(PG_FUNCTION_ARGS)
 {
@@ -630,7 +632,7 @@ vector_to_sparsevec(PG_FUNCTION_ARGS)
 /*
  * Convert half vector to sparse vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_to_sparsevec);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_to_sparsevec);
 Datum
 halfvec_to_sparsevec(PG_FUNCTION_ARGS)
 {
@@ -670,6 +672,137 @@ halfvec_to_sparsevec(PG_FUNCTION_ARGS)
 	PG_RETURN_POINTER(result);
 }
 
+/*
+ * Convert array to sparse vector
+ */
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(array_to_sparsevec);
+Datum
+array_to_sparsevec(PG_FUNCTION_ARGS)
+{
+	ArrayType  *array = PG_GETARG_ARRAYTYPE_P(0);
+	int32		typmod = PG_GETARG_INT32(1);
+	SparseVector *result;
+	int16		typlen;
+	bool		typbyval;
+	char		typalign;
+	Datum	   *elemsp;
+	int			nelemsp;
+	int			nnz = 0;
+	float	   *values;
+	int			j = 0;
+
+	if (ARR_NDIM(array) > 1)
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("array must be 1-D")));
+
+	if (ARR_HASNULL(array) && array_contains_nulls(array))
+		ereport(ERROR,
+				(errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
+				 errmsg("array must not contain nulls")));
+
+	get_typlenbyvalalign(ARR_ELEMTYPE(array), &typlen, &typbyval, &typalign);
+	deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, NULL, &nelemsp);
+
+	CheckDim(nelemsp);
+	CheckExpectedDim(typmod, nelemsp);
+
+#ifdef _MSC_VER
+/* /fp:fast may not propagate +/-Infinity or NaN */
+#define IS_NOT_ZERO(v) (isnan((float) (v)) || isinf((float) (v)) || ((float) (v)) != 0)
+#else
+#define IS_NOT_ZERO(v) (((float) (v)) != 0)
+#endif
+
+	if (ARR_ELEMTYPE(array) == INT4OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			nnz += IS_NOT_ZERO(DatumGetInt32(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == FLOAT8OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			nnz += IS_NOT_ZERO(DatumGetFloat8(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == FLOAT4OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			nnz += IS_NOT_ZERO(DatumGetFloat4(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == NUMERICOID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			nnz += IS_NOT_ZERO(DirectFunctionCall1(numeric_float4, elemsp[i]));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("unsupported array type")));
+	}
+
+	result = InitSparseVector(nelemsp, nnz);
+	values = SPARSEVEC_VALUES(result);
+
+#define PROCESS_ARRAY_ELEM(elem) \
+	do { \
+		float v = (float) (elem); \
+		if (IS_NOT_ZERO(v)) { \
+			/* Safety check */ \
+			if (j >= result->nnz) \
+				elog(ERROR, "safety check failed"); \
+			result->indices[j] = i; \
+			values[j] = v; \
+			j++; \
+		} \
+	} while (0)
+
+	if (ARR_ELEMTYPE(array) == INT4OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			PROCESS_ARRAY_ELEM(DatumGetInt32(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == FLOAT8OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			PROCESS_ARRAY_ELEM(DatumGetFloat8(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == FLOAT4OID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			PROCESS_ARRAY_ELEM(DatumGetFloat4(elemsp[i]));
+	}
+	else if (ARR_ELEMTYPE(array) == NUMERICOID)
+	{
+		for (int i = 0; i < nelemsp; i++)
+			PROCESS_ARRAY_ELEM(DatumGetFloat4(DirectFunctionCall1(numeric_float4, elemsp[i])));
+	}
+	else
+	{
+		ereport(ERROR,
+				(errcode(ERRCODE_DATA_EXCEPTION),
+				 errmsg("unsupported array type")));
+	}
+
+#undef PROCESS_ARRAY_ELEM
+#undef IS_NOT_ZERO
+
+	/*
+	 * Free allocation from deconstruct_array. Do not free individual elements
+	 * when pass-by-reference since they point to original array.
+	 */
+	pfree(elemsp);
+
+	if (j != result->nnz)
+		elog(ERROR, "correctness check failed");
+
+	/* Check elements */
+	for (int i = 0; i < result->nnz; i++)
+		CheckElement(values[i]);
+
+	PG_RETURN_POINTER(result);
+}
+
 /*
  * Get the L2 squared distance between sparse vectors
  */
@@ -721,7 +854,7 @@ SparsevecL2SquaredDistance(SparseVector * a, SparseVector * b)
 /*
  * Get the L2 distance between sparse vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_l2_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_l2_distance);
 Datum
 sparsevec_l2_distance(PG_FUNCTION_ARGS)
 {
@@ -737,7 +870,7 @@ sparsevec_l2_distance(PG_FUNCTION_ARGS)
  * Get the L2 squared distance between sparse vectors
  * This saves a sqrt calculation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_l2_squared_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_l2_squared_distance);
 Datum
 sparsevec_l2_squared_distance(PG_FUNCTION_ARGS)
 {
@@ -788,7 +921,7 @@ SparsevecInnerProduct(SparseVector * a, SparseVector * b)
 /*
  * Get the inner product of two sparse vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_inner_product);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_inner_product);
 Datum
 sparsevec_inner_product(PG_FUNCTION_ARGS)
 {
@@ -803,7 +936,7 @@ sparsevec_inner_product(PG_FUNCTION_ARGS)
 /*
  * Get the negative inner product of two sparse vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_negative_inner_product);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_negative_inner_product);
 Datum
 sparsevec_negative_inner_product(PG_FUNCTION_ARGS)
 {
@@ -818,7 +951,7 @@ sparsevec_negative_inner_product(PG_FUNCTION_ARGS)
 /*
  * Get the cosine distance between two sparse vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_cosine_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_cosine_distance);
 Datum
 sparsevec_cosine_distance(PG_FUNCTION_ARGS)
 {
@@ -863,7 +996,7 @@ sparsevec_cosine_distance(PG_FUNCTION_ARGS)
 /*
  * Get the L1 distance between two sparse vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_l1_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_l1_distance);
 Datum
 sparsevec_l1_distance(PG_FUNCTION_ARGS)
 {
@@ -912,7 +1045,7 @@ sparsevec_l1_distance(PG_FUNCTION_ARGS)
 /*
  * Get the L2 norm of a sparse vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_l2_norm);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_l2_norm);
 Datum
 sparsevec_l2_norm(PG_FUNCTION_ARGS)
 {
@@ -930,7 +1063,7 @@ sparsevec_l2_norm(PG_FUNCTION_ARGS)
 /*
  * Normalize a sparse vector with the L2 norm
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_l2_normalize);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_l2_normalize);
 Datum
 sparsevec_l2_normalize(PG_FUNCTION_ARGS)
 {
@@ -1040,7 +1173,7 @@ sparsevec_cmp_internal(SparseVector * a, SparseVector * b)
 /*
  * Less than
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_lt);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_lt);
 Datum
 sparsevec_lt(PG_FUNCTION_ARGS)
 {
@@ -1053,7 +1186,7 @@ sparsevec_lt(PG_FUNCTION_ARGS)
 /*
  * Less than or equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_le);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_le);
 Datum
 sparsevec_le(PG_FUNCTION_ARGS)
 {
@@ -1066,7 +1199,7 @@ sparsevec_le(PG_FUNCTION_ARGS)
 /*
  * Equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_eq);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_eq);
 Datum
 sparsevec_eq(PG_FUNCTION_ARGS)
 {
@@ -1079,7 +1212,7 @@ sparsevec_eq(PG_FUNCTION_ARGS)
 /*
  * Not equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_ne);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_ne);
 Datum
 sparsevec_ne(PG_FUNCTION_ARGS)
 {
@@ -1092,7 +1225,7 @@ sparsevec_ne(PG_FUNCTION_ARGS)
 /*
  * Greater than or equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_ge);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_ge);
 Datum
 sparsevec_ge(PG_FUNCTION_ARGS)
 {
@@ -1105,7 +1238,7 @@ sparsevec_ge(PG_FUNCTION_ARGS)
 /*
  * Greater than
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_gt);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_gt);
 Datum
 sparsevec_gt(PG_FUNCTION_ARGS)
 {
@@ -1118,7 +1251,7 @@ sparsevec_gt(PG_FUNCTION_ARGS)
 /*
  * Compare sparse vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_cmp);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_cmp);
 Datum
 sparsevec_cmp(PG_FUNCTION_ARGS)
 {

src/vector.c

@@ -26,11 +26,6 @@
 #include "varatt.h"
 #endif
 
-#if PG_VERSION_NUM < 130000
-#define TYPALIGN_DOUBLE 'd'
-#define TYPALIGN_INT 'i'
-#endif
-
 #define STATE_DIMS(x) (ARR_DIMS(x)[0] - 1)
 #define CreateStateDatums(dim) palloc(sizeof(Datum) * (dim + 1))
 
@@ -160,28 +155,10 @@ CheckStateArray(ArrayType *statearray, const char *caller)
 	return (float8 *) ARR_DATA_PTR(statearray);
 }
 
-#if PG_VERSION_NUM < 120003
-static pg_noinline void
-float_overflow_error(void)
-{
-	ereport(ERROR,
-			(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-			 errmsg("value out of range: overflow")));
-}
-
-static pg_noinline void
-float_underflow_error(void)
-{
-	ereport(ERROR,
-			(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-			 errmsg("value out of range: underflow")));
-}
-#endif
-
 /*
  * Convert textual representation to internal representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_in);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_in);
 Datum
 vector_in(PG_FUNCTION_ARGS)
 {
@@ -294,7 +271,7 @@ vector_in(PG_FUNCTION_ARGS)
 /*
  * Convert internal representation to textual representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_out);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_out);
 Datum
 vector_out(PG_FUNCTION_ARGS)
 {
@@ -348,7 +325,7 @@ PrintVector(char *msg, Vector * vector)
 /*
  * Convert type modifier
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_typmod_in);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_typmod_in);
 Datum
 vector_typmod_in(PG_FUNCTION_ARGS)
 {
@@ -379,7 +356,7 @@ vector_typmod_in(PG_FUNCTION_ARGS)
 /*
  * Convert external binary representation to internal representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_recv);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_recv);
 Datum
 vector_recv(PG_FUNCTION_ARGS)
 {
@@ -413,7 +390,7 @@ vector_recv(PG_FUNCTION_ARGS)
 /*
  * Convert internal representation to the external binary representation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_send);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_send);
 Datum
 vector_send(PG_FUNCTION_ARGS)
 {
@@ -433,7 +410,7 @@ vector_send(PG_FUNCTION_ARGS)
  * Convert vector to vector
  * This is needed to check the type modifier
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector);
 Datum
 vector(PG_FUNCTION_ARGS)
 {
@@ -448,7 +425,7 @@ vector(PG_FUNCTION_ARGS)
 /*
  * Convert array to vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(array_to_vector);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(array_to_vector);
 Datum
 array_to_vector(PG_FUNCTION_ARGS)
 {
@@ -522,7 +499,7 @@ array_to_vector(PG_FUNCTION_ARGS)
 /*
  * Convert vector to float4[]
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_to_float4);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_to_float4);
 Datum
 vector_to_float4(PG_FUNCTION_ARGS)
 {
@@ -546,7 +523,7 @@ vector_to_float4(PG_FUNCTION_ARGS)
 /*
  * Convert half vector to vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(halfvec_to_vector);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(halfvec_to_vector);
 Datum
 halfvec_to_vector(PG_FUNCTION_ARGS)
 {
@@ -584,7 +561,7 @@ VectorL2SquaredDistance(int dim, float *ax, float *bx)
 /*
  * Get the L2 distance between vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(l2_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(l2_distance);
 Datum
 l2_distance(PG_FUNCTION_ARGS)
 {
@@ -600,7 +577,7 @@ l2_distance(PG_FUNCTION_ARGS)
  * Get the L2 squared distance between vectors
  * This saves a sqrt calculation
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_l2_squared_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_l2_squared_distance);
 Datum
 vector_l2_squared_distance(PG_FUNCTION_ARGS)
 {
@@ -627,7 +604,7 @@ VectorInnerProduct(int dim, float *ax, float *bx)
 /*
  * Get the inner product of two vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(inner_product);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(inner_product);
 Datum
 inner_product(PG_FUNCTION_ARGS)
 {
@@ -642,7 +619,7 @@ inner_product(PG_FUNCTION_ARGS)
 /*
  * Get the negative inner product of two vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_negative_inner_product);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_negative_inner_product);
 Datum
 vector_negative_inner_product(PG_FUNCTION_ARGS)
 {
@@ -676,7 +653,7 @@ VectorCosineSimilarity(int dim, float *ax, float *bx)
 /*
  * Get the cosine distance between two vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(cosine_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(cosine_distance);
 Datum
 cosine_distance(PG_FUNCTION_ARGS)
 {
@@ -708,7 +685,7 @@ cosine_distance(PG_FUNCTION_ARGS)
  * Currently uses angular distance since needs to satisfy triangle inequality
  * Assumes inputs are unit vectors (skips norm)
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_spherical_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_spherical_distance);
 Datum
 vector_spherical_distance(PG_FUNCTION_ARGS)
 {
@@ -745,7 +722,7 @@ VectorL1Distance(int dim, float *ax, float *bx)
 /*
  * Get the L1 distance between two vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(l1_distance);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(l1_distance);
 Datum
 l1_distance(PG_FUNCTION_ARGS)
 {
@@ -760,7 +737,7 @@ l1_distance(PG_FUNCTION_ARGS)
 /*
  * Get the dimensions of a vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_dims);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_dims);
 Datum
 vector_dims(PG_FUNCTION_ARGS)
 {
@@ -772,7 +749,7 @@ vector_dims(PG_FUNCTION_ARGS)
 /*
  * Get the L2 norm of a vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_norm);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_norm);
 Datum
 vector_norm(PG_FUNCTION_ARGS)
 {
@@ -790,7 +767,7 @@ vector_norm(PG_FUNCTION_ARGS)
 /*
  * Normalize a vector with the L2 norm
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(l2_normalize);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(l2_normalize);
 Datum
 l2_normalize(PG_FUNCTION_ARGS)
 {
@@ -829,7 +806,7 @@ l2_normalize(PG_FUNCTION_ARGS)
 /*
  * Add vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_add);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_add);
 Datum
 vector_add(PG_FUNCTION_ARGS)
 {
@@ -862,7 +839,7 @@ vector_add(PG_FUNCTION_ARGS)
 /*
  * Subtract vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_sub);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_sub);
 Datum
 vector_sub(PG_FUNCTION_ARGS)
 {
@@ -895,7 +872,7 @@ vector_sub(PG_FUNCTION_ARGS)
 /*
  * Multiply vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_mul);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_mul);
 Datum
 vector_mul(PG_FUNCTION_ARGS)
 {
@@ -931,7 +908,7 @@ vector_mul(PG_FUNCTION_ARGS)
 /*
  * Concatenate vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_concat);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_concat);
 Datum
 vector_concat(PG_FUNCTION_ARGS)
 {
@@ -955,7 +932,7 @@ vector_concat(PG_FUNCTION_ARGS)
 /*
  * Quantize a vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(binary_quantize);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(binary_quantize);
 Datum
 binary_quantize(PG_FUNCTION_ARGS)
 {
@@ -973,7 +950,7 @@ binary_quantize(PG_FUNCTION_ARGS)
 /*
  * Get a subvector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(subvector);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(subvector);
 Datum
 subvector(PG_FUNCTION_ARGS)
 {
@@ -1047,7 +1024,7 @@ vector_cmp_internal(Vector * a, Vector * b)
 /*
  * Less than
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_lt);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_lt);
 Datum
 vector_lt(PG_FUNCTION_ARGS)
 {
@@ -1060,7 +1037,7 @@ vector_lt(PG_FUNCTION_ARGS)
 /*
  * Less than or equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_le);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_le);
 Datum
 vector_le(PG_FUNCTION_ARGS)
 {
@@ -1073,7 +1050,7 @@ vector_le(PG_FUNCTION_ARGS)
 /*
  * Equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_eq);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_eq);
 Datum
 vector_eq(PG_FUNCTION_ARGS)
 {
@@ -1086,7 +1063,7 @@ vector_eq(PG_FUNCTION_ARGS)
 /*
  * Not equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_ne);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_ne);
 Datum
 vector_ne(PG_FUNCTION_ARGS)
 {
@@ -1099,7 +1076,7 @@ vector_ne(PG_FUNCTION_ARGS)
 /*
  * Greater than or equal
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_ge);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_ge);
 Datum
 vector_ge(PG_FUNCTION_ARGS)
 {
@@ -1112,7 +1089,7 @@ vector_ge(PG_FUNCTION_ARGS)
 /*
  * Greater than
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_gt);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_gt);
 Datum
 vector_gt(PG_FUNCTION_ARGS)
 {
@@ -1125,7 +1102,7 @@ vector_gt(PG_FUNCTION_ARGS)
 /*
  * Compare vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_cmp);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_cmp);
 Datum
 vector_cmp(PG_FUNCTION_ARGS)
 {
@@ -1138,7 +1115,7 @@ vector_cmp(PG_FUNCTION_ARGS)
 /*
  * Accumulate vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_accum);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_accum);
 Datum
 vector_accum(PG_FUNCTION_ARGS)
 {
@@ -1199,7 +1176,7 @@ vector_accum(PG_FUNCTION_ARGS)
 /*
  * Combine vectors or half vectors (also used for halfvec_combine)
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_combine);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_combine);
 Datum
 vector_combine(PG_FUNCTION_ARGS)
 {
@@ -1270,7 +1247,7 @@ vector_combine(PG_FUNCTION_ARGS)
 /*
  * Average vectors
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_avg);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(vector_avg);
 Datum
 vector_avg(PG_FUNCTION_ARGS)
 {
@@ -1304,7 +1281,7 @@ vector_avg(PG_FUNCTION_ARGS)
 /*
  * Convert sparse vector to dense vector
  */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_to_vector);
+FUNCTION_PREFIX PG_FUNCTION_INFO_V1(sparsevec_to_vector);
 Datum
 sparsevec_to_vector(PG_FUNCTION_ARGS)
 {

src/vector.h

@@ -20,4 +20,11 @@ Vector	   *InitVector(int dim);
 void		PrintVector(char *msg, Vector * vector);
 int			vector_cmp_internal(Vector * a, Vector * b);
 
+/* TODO Move to better place */
+#if PG_VERSION_NUM >= 160000
+#define FUNCTION_PREFIX
+#else
+#define FUNCTION_PREFIX PGDLLEXPORT
+#endif
+
 #endif

test/expected/cast.out

@@ -208,6 +208,62 @@ SELECT '{1:1e-8}/1'::sparsevec::halfvec;
  [0]
 (1 row)
 
+SELECT ARRAY[1,0,2,0,3,0]::sparsevec;
+      array      
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT ARRAY[1.0,0.0,2.0,0.0,3.0,0.0]::sparsevec;
+      array      
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT ARRAY[1,0,2,0,3,0]::float4[]::sparsevec;
+      array      
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT ARRAY[1,0,2,0,3,0]::float8[]::sparsevec;
+      array      
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT ARRAY[1,0,2,0,3,0]::numeric[]::sparsevec;
+      array      
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec;
+    sparsevec    
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(6);
+    sparsevec    
+-----------------
+ {1:1,3:2,5:3}/6
+(1 row)
+
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(5);
+ERROR:  expected 5 dimensions, not 6
+SELECT '{NULL}'::real[]::sparsevec;
+ERROR:  array must not contain nulls
+SELECT '{NaN}'::real[]::sparsevec;
+ERROR:  NaN not allowed in sparsevec
+SELECT '{Infinity}'::real[]::sparsevec;
+ERROR:  infinite value not allowed in sparsevec
+SELECT '{-Infinity}'::real[]::sparsevec;
+ERROR:  infinite value not allowed in sparsevec
+SELECT '{}'::real[]::sparsevec;
+ERROR:  sparsevec must have at least 1 dimension
+SELECT '{{1}}'::real[]::sparsevec;
+ERROR:  array must be 1-D
 SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
 ERROR:  vector cannot have more than 16000 dimensions
 SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;

test/perl/PostgreSQL/Test/Cluster.pm

@@ -0,0 +1,11 @@
+package PostgreSQL::Test::Cluster;
+
+use PostgresNode;
+
+sub new
+{
+	my ($class, $name) = @_;
+	return get_new_node($name);
+}
+
+1;

test/perl/PostgreSQL/Test/Utils.pm

@@ -0,0 +1,5 @@
+package PostgreSQL::Test::Utils;
+
+use TestLib;
+
+1;

test/perl/PostgresNode.pm

@@ -1,8 +0,0 @@
-use PostgreSQL::Test::Cluster;
-
-sub get_new_node
-{
-	return PostgreSQL::Test::Cluster->new(@_);
-}
-
-1;

test/perl/TestLib.pm

@@ -1,3 +0,0 @@
-use PostgreSQL::Test::Utils;
-
-1;

test/sql/cast.sql

@@ -58,6 +58,22 @@ SELECT '{}/16001'::sparsevec::halfvec;
 SELECT '{1:65520}/1'::sparsevec::halfvec;
 SELECT '{1:1e-8}/1'::sparsevec::halfvec;
 
+SELECT ARRAY[1,0,2,0,3,0]::sparsevec;
+SELECT ARRAY[1.0,0.0,2.0,0.0,3.0,0.0]::sparsevec;
+SELECT ARRAY[1,0,2,0,3,0]::float4[]::sparsevec;
+SELECT ARRAY[1,0,2,0,3,0]::float8[]::sparsevec;
+SELECT ARRAY[1,0,2,0,3,0]::numeric[]::sparsevec;
+
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec;
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(6);
+SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(5);
+SELECT '{NULL}'::real[]::sparsevec;
+SELECT '{NaN}'::real[]::sparsevec;
+SELECT '{Infinity}'::real[]::sparsevec;
+SELECT '{-Infinity}'::real[]::sparsevec;
+SELECT '{}'::real[]::sparsevec;
+SELECT '{{1}}'::real[]::sparsevec;
+
 SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
 SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;
 

test/t/001_ivfflat_wal.pl

@@ -2,9 +2,9 @@
 
 # Test generic xlog record work for ivfflat index replication.
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 32;
@@ -49,7 +49,7 @@ sub test_index_replay
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize primary node
-$node_primary = get_new_node('primary');
+$node_primary = PostgreSQL::Test::Cluster->new('primary');
 $node_primary->init(allows_streaming => 1);
 if ($dim > 32)
 {
@@ -67,7 +67,7 @@ my $backup_name = 'my_backup';
 $node_primary->backup($backup_name);
 
 # Create streaming replica linking to primary
-$node_replica = get_new_node('replica');
+$node_replica = PostgreSQL::Test::Cluster->new('replica');
 $node_replica->init_from_backup($node_primary, $backup_name, has_streaming => 1);
 $node_replica->start;
 

test/t/002_ivfflat_vacuum.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 3;
@@ -15,7 +15,7 @@ for (1 .. $dim)
 my $array_sql = join(", ", @r);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/003_ivfflat_vector_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -49,7 +49,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/004_ivfflat_vector_insert_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/005_ivfflat_query_recall.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/006_ivfflat_lists.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/007_ivfflat_inserts.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 768;
@@ -9,7 +9,7 @@ my $dim = 768;
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/008_ivfflat_centers.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/009_ivfflat_filtering.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 3;
@@ -11,7 +11,7 @@ my $limit = 20;
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 
@@ -94,8 +94,7 @@ like($explain, qr/Seq Scan/);
 $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query';
 ));
-# TODO Do not use index
-like($explain, qr/Index Scan using idx/);
+like($explain, qr/Seq Scan/);
 
 # Test attribute index
 $node->safe_psql("postgres", "CREATE INDEX attribute_idx ON tst (c);");
@@ -110,7 +109,6 @@ $node->safe_psql("postgres", "CREATE INDEX partial_idx ON tst USING ivfflat (v v
 $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
 ));
-# TODO Use partial index
-like($explain, qr/Index Scan using idx/);
+like($explain, qr/Index Scan using partial_idx/);
 
 done_testing();

test/t/010_hnsw_wal.pl

@@ -2,9 +2,9 @@
 
 # Test generic xlog record work for hnsw index replication.
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 32;
@@ -49,7 +49,7 @@ sub test_index_replay
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize primary node
-$node_primary = get_new_node('primary');
+$node_primary = PostgreSQL::Test::Cluster->new('primary');
 $node_primary->init(allows_streaming => 1);
 if ($dim > 32)
 {
@@ -67,7 +67,7 @@ my $backup_name = 'my_backup';
 $node_primary->backup($backup_name);
 
 # Create streaming replica linking to primary
-$node_replica = get_new_node('replica');
+$node_replica = PostgreSQL::Test::Cluster->new('replica');
 $node_replica->init_from_backup($node_primary, $backup_name, has_streaming => 1);
 $node_replica->start;
 

test/t/011_hnsw_vacuum.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 3;
@@ -15,7 +15,7 @@ for (1 .. $dim)
 my $array_sql = join(", ", @r);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/012_hnsw_vector_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -47,7 +47,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/013_hnsw_vector_insert_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -47,7 +47,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/014_hnsw_vector_vacuum_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/015_hnsw_vector_duplicates.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/016_hnsw_inserts.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Ensures elements and neighbors on both same and different pages
@@ -10,7 +10,7 @@ my $dim = 1900;
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/017_hnsw_filtering.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 3;
@@ -11,16 +11,20 @@ my $limit = 20;
 my $array_sql = join(",", ('random()') x $dim);
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 
 # Create table and index
 $node->safe_psql("postgres", "CREATE EXTENSION vector;");
 $node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim), c int4, t text);");
+$node->safe_psql("postgres", "CREATE TABLE cat (i int4 PRIMARY KEY, t text, b boolean);");
 $node->safe_psql("postgres",
 	"INSERT INTO tst SELECT i, ARRAY[$array_sql], i % $nc, 'test ' || i FROM generate_series(1, 10000) i;"
 );
+$node->safe_psql("postgres",
+	"INSERT INTO cat SELECT i, 'cat ' || i, i % 5 = 0 FROM generate_series(1, $nc) i;"
+);
 $node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);");
 $node->safe_psql("postgres", "ANALYZE tst;");
 
@@ -37,8 +41,7 @@ my $c = int(rand() * $nc);
 my $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
 ));
-# TODO Do not use index
-like($explain, qr/Index Scan using idx/);
+like($explain, qr/Seq Scan/);
 
 # Test attribute filtering with few rows removed
 $explain = $node->safe_psql("postgres", qq(
@@ -56,8 +59,7 @@ like($explain, qr/Index Scan using idx/);
 $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE c < 1 ORDER BY v <-> '$query' LIMIT $limit;
 ));
-# TODO Do not use index
-like($explain, qr/Index Scan using idx/);
+like($explain, qr/Seq Scan/);
 
 # Test attribute filtering with few rows removed like
 $explain = $node->safe_psql("postgres", qq(
@@ -96,13 +98,25 @@ $explain = $node->safe_psql("postgres", qq(
 ));
 like($explain, qr/Seq Scan/);
 
+# Test join
+$explain = $node->safe_psql("postgres", qq(
+	EXPLAIN ANALYZE SELECT cat.t FROM cat INNER JOIN tst ON cat.i = tst.c ORDER BY v <-> '$query' LIMIT $limit;
+));
+like($explain, qr/Index Scan using idx/);
+
+# Test join with attribute filtering
+$explain = $node->safe_psql("postgres", qq(
+	EXPLAIN ANALYZE SELECT cat.t FROM cat INNER JOIN tst ON cat.i = tst.c WHERE cat.b = 't' ORDER BY v <-> '$query' LIMIT $limit;
+));
+like($explain, qr/Index Scan using idx/);
+
 # Test attribute index
 $node->safe_psql("postgres", "CREATE INDEX attribute_idx ON tst (c);");
 $explain = $node->safe_psql("postgres", qq(
 	EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
 ));
-# TODO Use attribute index
-like($explain, qr/Index Scan using idx/);
+# Use attribute index
+like($explain, qr/Bitmap Index Scan on attribute_idx/);
 
 # Test partial index
 $node->safe_psql("postgres", "CREATE INDEX partial_idx ON tst USING hnsw (v vector_l2_ops) WHERE (c = $c);");

test/t/018_aggregates.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 
@@ -53,7 +53,7 @@ sub test_aggregate
 	else
 	{
 		# Does not raise overflow error in this instance due to loss of precision
-		is($res, "[24576,24576,49152]")
+		is($res, "[24576,24576,49152]");
 	}
 }
 

test/t/019_storage.pl

@@ -1,13 +1,13 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $dim = 1024;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/020_hnsw_bit_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -51,7 +51,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/021_hnsw_bit_insert_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -51,7 +51,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/022_hnsw_bit_vacuum_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -51,7 +51,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/023_hnsw_bit_duplicates.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/024_hnsw_halfvec_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/025_hnsw_halfvec_insert_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/026_hnsw_halfvec_vacuum_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/027_hnsw_halfvec_duplicates.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/028_hnsw_sparsevec_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -47,7 +47,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/029_hnsw_sparsevec_insert_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -47,7 +47,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/030_hnsw_sparsevec_vacuum_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -48,7 +48,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/031_hnsw_sparsevec_duplicates.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/032_ivfflat_halfvec_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -51,7 +51,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/033_comparison.pl

@@ -1,14 +1,14 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
 my $array_sql = join(",", ('floor(random() * 2)::int - 1') x 3);
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/034_distance_functions.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -10,7 +10,7 @@ my $dim = 5;
 my $array_sql = join(",", ('floor(random() * 4)::int - 2') x $dim);
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/035_ivfflat_bit_build_recall.pl

@@ -1,7 +1,7 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 my $node;
@@ -51,7 +51,7 @@ sub test_recall
 }
 
 # Initialize node
-$node = get_new_node('node');
+$node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/036_ivfflat_bit_centers.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/037_inputs.pl

@@ -1,11 +1,11 @@
 use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
 use Test::More;
 
 # Initialize node
-my $node = get_new_node('node');
+my $node = PostgreSQL::Test::Cluster->new('node');
 $node->init;
 $node->start;
 

test/t/038_hnsw_sparsevec_vacuum_insert.pl

@@ -0,0 +1,42 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+# Initialize node
+my $node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+# Create table and index
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (i serial, v sparsevec(100000));");
+$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v sparsevec_l2_ops);");
+
+for (1 .. 3)
+{
+	for (1 .. 100)
+	{
+		my @elements;
+		my %indices;
+		for (1 .. int(rand() * 100))
+		{
+			my $index = int(rand() * (100000 - 1)) + 1;
+			if (!exists($indices{$index}))
+			{
+				my $value = rand();
+				push(@elements, "$index:$value");
+				$indices{$index} = 1;
+			}
+		}
+		my $embedding = "{" . join(",", @elements) . "}/100000";
+		$node->safe_psql("postgres", "INSERT INTO tst (v) VALUES ('$embedding');");
+	}
+
+	$node->safe_psql("postgres", "DELETE FROM tst WHERE i % 2 = 0;");
+	$node->safe_psql("postgres", "VACUUM tst;");
+	is(1, 1);
+}
+
+done_testing();

test/t/039_hnsw_cost.pl

@@ -0,0 +1,60 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my @dims = (384, 1536);
+my $limit = 10;
+
+# Initialize node
+my $node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+
+for my $dim (@dims)
+{
+	my $array_sql = join(",", ('random()') x $dim);
+
+	# Create table and index
+	$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
+	$node->safe_psql("postgres",
+		"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 2000) i;"
+	);
+	$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);");
+	$node->safe_psql("postgres", "ANALYZE tst;");
+
+	# Generate query
+	my @r = ();
+	for (1 .. $dim)
+	{
+		push(@r, rand());
+	}
+	my $query = "[" . join(",", @r) . "]";
+
+	my $explain = $node->safe_psql("postgres", qq(
+		EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v <-> '$query' LIMIT $limit;
+	));
+	like($explain, qr/Index Scan using idx/);
+
+	# 3x the rows are needed for distance filters
+	# since the planner uses DEFAULT_INEQ_SEL for the selectivity (should be 1)
+	# Recreate index for performance
+	$node->safe_psql("postgres", "DROP INDEX idx;");
+	$node->safe_psql("postgres",
+		"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(2001, 6000) i;"
+	);
+	$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);");
+	$node->safe_psql("postgres", "ANALYZE tst;");
+
+	$explain = $node->safe_psql("postgres", qq(
+		EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query' LIMIT $limit;
+	));
+	like($explain, qr/Index Scan using idx/);
+
+	$node->safe_psql("postgres", "DROP TABLE tst;");
+}
+
+done_testing();

test/t/040_ivfflat_cost.pl

@@ -0,0 +1,50 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my @dims = (384, 1536);
+my $limit = 10;
+
+# Initialize node
+my $node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+
+for my $dim (@dims)
+{
+	my $array_sql = join(",", ('random()') x $dim);
+
+	# Create table and index
+	$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
+	$node->safe_psql("postgres",
+		"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 5000) i;"
+	);
+	$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING ivfflat (v vector_l2_ops) WITH (lists = 5);");
+	$node->safe_psql("postgres", "ANALYZE tst;");
+
+	# Generate query
+	my @r = ();
+	for (1 .. $dim)
+	{
+		push(@r, rand());
+	}
+	my $query = "[" . join(",", @r) . "]";
+
+	my $explain = $node->safe_psql("postgres", qq(
+		EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v <-> '$query' LIMIT $limit;
+	));
+	like($explain, qr/Index Scan using idx/);
+
+	$explain = $node->safe_psql("postgres", qq(
+		EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query' LIMIT $limit;
+	));
+	like($explain, qr/Index Scan using idx/);
+
+	$node->safe_psql("postgres", "DROP TABLE tst;");
+}
+
+done_testing();

test/t/041_hnsw_filtering.pl

@@ -0,0 +1,109 @@
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+my $node;
+my @queries = ();
+my @cs = ();
+my @expected;
+my $limit = 20;
+my $dim = 3;
+my $array_sql = join(",", ('random()') x $dim);
+my $nc = 50;
+
+sub test_recall
+{
+	my ($min, $operator) = @_;
+	my $correct = 0;
+	my $total = 0;
+
+	my $explain = $node->safe_psql("postgres", qq(
+		SET enable_seqscan = off;
+		EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $cs[0] ORDER BY v $operator '$queries[0]' LIMIT $limit;
+	));
+	like($explain, qr/Index Cond/);
+
+	for my $i (0 .. $#queries)
+	{
+		my $actual = $node->safe_psql("postgres", qq(
+			SET enable_seqscan = off;
+			SELECT i FROM tst WHERE c = $cs[$i] ORDER BY v $operator '$queries[$i]' LIMIT $limit;
+		));
+		my @actual_ids = split("\n", $actual);
+		my %actual_set = map { $_ => 1 } @actual_ids;
+
+		is(scalar(@actual_ids), $limit);
+
+		my @expected_ids = split("\n", $expected[$i]);
+
+		foreach (@expected_ids)
+		{
+			if (exists($actual_set{$_}))
+			{
+				$correct++;
+			}
+			$total++;
+		}
+	}
+
+	cmp_ok($correct / $total, ">=", $min, $operator);
+}
+
+# Initialize node
+$node = PostgreSQL::Test::Cluster->new('node');
+$node->init;
+$node->start;
+
+# Create table
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim), c int4);");
+$node->safe_psql("postgres",
+	"INSERT INTO tst SELECT i, ARRAY[$array_sql], i % $nc FROM generate_series(1, 20000) i;"
+);
+
+# Generate queries
+for (1 .. 20)
+{
+	my @r = ();
+	for (1 .. $dim)
+	{
+		push(@r, rand());
+	}
+	push(@queries, "[" . join(",", @r) . "]");
+	push(@cs, int(rand() * $nc));
+}
+
+# Get exact results
+@expected = ();
+for my $i (0 .. $#queries)
+{
+	my $res = $node->safe_psql("postgres", "SELECT i FROM tst WHERE c = $cs[$i] ORDER BY v <-> '$queries[$i]' LIMIT $limit;");
+	push(@expected, $res);
+}
+
+# Add index
+$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops, c);");
+
+# Test recall
+test_recall(0.99, '<->');
+
+# Test vacuum
+$node->safe_psql("postgres", "DELETE FROM tst WHERE c > 5;");
+$node->safe_psql("postgres", "VACUUM tst;");
+
+# Test columns
+my ($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (c);");
+like($stderr, qr/first column must be a vector/);
+
+($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (c, v vector_l2_ops);");
+like($stderr, qr/first column must be a vector/);
+
+($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops, c, c);");
+like($stderr, qr/index cannot have more than two columns/);
+
+($ret, $stdout, $stderr) = $node->psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops, v vector_l2_ops);");
+like($stderr, qr/column 2 cannot be a vector/);
+
+done_testing();

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat and hnsw access methods'
-default_version = '0.7.2'
+default_version = '0.7.4'
 module_pathname = '$libdir/vector'
 relocatable = true