Simplified CI

.github/workflows/build.yml

@@ -28,10 +28,8 @@ jobs:
           dev-files: true
       - run: make
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
+          PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
-      - run: |
+      - run: sudo make install
-          export PG_CONFIG=`which pg_config`
-          sudo --preserve-env=PG_CONFIG make install
       - run: make installcheck
       - if: ${{ failure() }}
         run: cat regression.diffs
@@ -39,76 +37,3 @@ jobs:
           sudo apt-get update
           sudo apt-get install libipc-run-perl
       - run: make prove_installcheck
-  mac:
-    runs-on: macos-latest
-    if: ${{ !startsWith(github.ref_name, 'windows') }}
-    steps:
-      - uses: actions/checkout@v4
-      - uses: ankane/setup-postgres@v1
-        with:
-          postgres-version: 14
-      - run: make
-        env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter
-      - run: make install
-      - run: make installcheck
-      - if: ${{ failure() }}
-        run: cat regression.diffs
-      - run: |
-          brew install cpanm
-          cpanm --notest IPC::Run
-          wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_10.tar.gz
-          tar xf REL_14_10.tar.gz
-      - run: make prove_installcheck PROVE_FLAGS="-I ./postgres-REL_14_10/src/test/perl" PERL5LIB="/Users/runner/perl5/lib/perl5"
-      - run: make clean && /usr/local/opt/llvm@15/bin/scan-build --status-bugs make
-        env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING
-  windows:
-    runs-on: windows-latest
-    if: ${{ !startsWith(github.ref_name, 'mac') }}
-    steps:
-      - uses: actions/checkout@v4
-      - uses: ankane/setup-postgres@v1
-        with:
-          postgres-version: 14
-      - run: |
-          call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
-          cd %TEMP% && ^
-          nmake /NOLOGO /F Makefile.win && ^
-          nmake /NOLOGO /F Makefile.win install && ^
-          nmake /NOLOGO /F Makefile.win installcheck && ^
-          nmake /NOLOGO /F Makefile.win clean && ^
-          nmake /NOLOGO /F Makefile.win uninstall
-        shell: cmd
-  i386:
-    if: ${{ !startsWith(github.ref_name, 'mac') && !startsWith(github.ref_name, 'windows') }}
-    runs-on: ubuntu-latest
-    container:
-      image: debian:12
-      options: --platform linux/386
-    steps:
-      - run: apt-get update && apt-get install -y build-essential git libipc-run-perl postgresql-15 postgresql-server-dev-15 sudo
-      - run: service postgresql start
-      - run: |
-          git clone https://github.com/${{ github.repository }}.git pgvector
-          cd pgvector
-          git fetch origin ${{ github.ref }}
-          git reset --hard FETCH_HEAD
-          make
-          make install
-          chown -R postgres .
-          sudo -u postgres make installcheck
-          sudo -u postgres make prove_installcheck
-        env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
-  valgrind:
-    if: ${{ !startsWith(github.ref_name, 'mac') && !startsWith(github.ref_name, 'windows') }}
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-      - uses: ankane/setup-postgres-valgrind@v1
-        with:
-          postgres-version: 16
-      - run: make
-      - run: sudo --preserve-env=PG_CONFIG make install
-      - run: make installcheck

CHANGELOG.md

@@ -1,16 +1,6 @@
-## 0.7.0 (unreleased)
+## 0.6.1 (unreleased)
-
-- Added `sparsevec` type
-
-## 0.6.2 (2024-03-18)
-
-- Reduced lock contention with parallel HNSW index builds
-
-## 0.6.1 (2024-03-04)
 
 - Fixed error with `ANALYZE` and vectors with different dimensions
-- Fixed segmentation fault with `shared_preload_libraries`
-- Fixed vector subtraction being marked as commutative
 
 ## 0.6.0 (2024-01-29)
 

LICENSE

@@ -1,4 +1,4 @@
-Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
+Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
 
 Portions Copyright (c) 1994, The Regents of the University of California
 

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": "0.6.2",
+	"version": "0.6.0",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.6.2",
+			"version": "0.6.0",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,10 +1,10 @@
 EXTENSION = vector
-EXTVERSION = 0.6.2
+EXTVERSION = 0.6.0
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*.sql)
-OBJS = src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/sparsevec.o src/vector.o
+OBJS = src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/vector.o
-HEADERS = src/sparsevec.h src/vector.h
+HEADERS = src/vector.h
 
 TESTS = $(wildcard test/sql/*.sql)
 REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))
@@ -71,9 +71,11 @@ PG_MAJOR ?= 16
 .PHONY: docker
 
 docker:
-	docker build --pull --no-cache --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
+	docker build --pull --no-cache --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) .
+	docker build --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
 
 .PHONY: docker-release
 
 docker-release:
-	docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .
+	docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR) .
+	docker buildx build --push --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) .

Makefile.win

@@ -1,8 +1,8 @@
 EXTENSION = vector
-EXTVERSION = 0.6.2
+EXTVERSION = 0.6.0
 
-OBJS = src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj
+OBJS = src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
-HEADERS = src\sparsevec.h src\vector.h
+HEADERS = src\vector.h
 
 REGRESS = btree cast copy functions input ivfflat_cosine ivfflat_ip ivfflat_l2 ivfflat_options ivfflat_unlogged
 REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)

README.md

@@ -20,13 +20,13 @@ Compile and install the extension (supports Postgres 12+)
 
 ```sh
 cd /tmp
-git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
 ```
 
-See the [installation notes](#installation-notes---linux-and-mac) if you run into issues
+See the [installation notes](#installation-notes) if you run into issues
 
 You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), [pkg](#pkg), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres). There are also instructions for [GitHub Actions](https://github.com/pgvector/setup-pgvector).
 
@@ -44,15 +44,12 @@ Then use `nmake` to build:
 
 ```cmd
 set "PGROOT=C:\Program Files\PostgreSQL\16"
-cd %TEMP%
+git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
-git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
 ```
 
-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).
 
 ## Getting Started
@@ -105,12 +102,6 @@ Insert vectors
 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))
-
-```sql
-COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
-```
-
 Upsert vectors
 
 ```sql
@@ -273,8 +264,6 @@ HINT:  Increase maintenance_work_mem to speed up builds.
 
 Note: Do not set `maintenance_work_mem` so high that it exhausts the memory on the server
 
-Like other index types, it’s faster to create an index after loading your initial data
-
 Starting with 0.6.0, you can also speed up index creation by increasing the number of parallel workers (2 by default)
 
 ```sql
@@ -413,51 +402,13 @@ You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python
 
 ## Performance
 
-### Tuning
-
-Use a tool like [PgTune](https://pgtune.leopard.in.ua/) to set initial values for Postgres server parameters. For instance, `shared_buffers` should typically be 25% of the server’s memory. You can find the config file with:
-
-```sql
-SHOW config_file;
-```
-
-And check individual settings with:
-
-```sql
-SHOW shared_buffers;
-```
-
-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)).
-
-```sql
-COPY items (embedding) FROM STDIN WITH (FORMAT BINARY);
-```
-
-Add any indexes *after* loading the initial data for best performance.
-
-### Indexing
-
-See index build time for [HNSW](#index-build-time) and [IVFFlat](#index-build-time-1).
-
-In production environments, create indexes concurrently to avoid blocking writes.
-
-```sql
-CREATE INDEX CONCURRENTLY ...
-```
-
-### Querying
-
 Use `EXPLAIN ANALYZE` to debug performance.
 
 ```sql
 EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
-#### Exact Search
+### Exact Search
 
 To speed up queries without an index, increase `max_parallel_workers_per_gather`.
 
@@ -471,7 +422,7 @@ If vectors are normalized to length 1 (like [OpenAI embeddings](https://platform
 SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
 ```
 
-#### Approximate Search
+### Approximate Search
 
 To speed up queries with an IVFFlat index, increase the number of inverted lists (at the expense of recall).
 
@@ -479,7 +430,7 @@ To speed up queries with an IVFFlat index, increase the number of inverted lists
 CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 1000);
 ```
 
-### Vacuuming
+## Vacuuming
 
 Vacuuming can take a while for HNSW indexes. Speed it up by reindexing first.
 
@@ -488,61 +439,6 @@ REINDEX INDEX CONCURRENTLY index_name;
 VACUUM table_name;
 ```
 
-## Monitoring
-
-Monitor performance with [pg_stat_statements](https://www.postgresql.org/docs/current/pgstatstatements.html) (be sure to add it to `shared_preload_libraries`).
-
-```sql
-CREATE EXTENSION pg_stat_statements;
-```
-
-Get the most time-consuming queries with:
-
-```sql
-SELECT query, calls, ROUND((total_plan_time + total_exec_time) / calls) AS avg_time_ms,
-    ROUND((total_plan_time + total_exec_time) / 60000) AS total_time_min
-    FROM pg_stat_statements ORDER BY total_plan_time + total_exec_time DESC LIMIT 20;
-```
-
-Note: Replace `total_plan_time + total_exec_time` with `total_time` for Postgres < 13
-
-Monitor recall by comparing results from approximate search with exact search.
-
-```sql
-BEGIN;
-SET LOCAL enable_indexscan = off; -- use exact search
-SELECT ...
-COMMIT;
-```
-
-## Scaling
-
-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)).
-
-## Sparse Vectors
-
-Create a sparse vector column with 10 dimensions
-
-```sql
-CREATE TABLE items (id bigserial PRIMARY KEY, embedding sparsevec(10));
-```
-
-Insert vectors
-
-```sql
-INSERT INTO items (embedding) VALUES ('{0:1,1:2,2:3}/10'), ('{0:4,1:5,2:6}/10');
-```
-
-Get the nearest neighbors by L2 distance
-
-```sql
-SELECT * FROM items ORDER BY embedding <-> '{0:3,1:1,2:2}/10' LIMIT 5;
-```
-
 ## Languages
 
 Use pgvector from any language with a Postgres client. You can even generate and store vectors in one language and query them in another.
@@ -636,18 +532,6 @@ and query with:
 SELECT * FROM items ORDER BY embedding::vector(3) <-> '[3,1,2]' LIMIT 5;
 ```
 
-#### Are binary vectors supported?
-
-You can store binary vectors and perform exact nearest neighbor search by Hamming distance in Postgres without an extension ([example](https://github.com/pgvector/pgvector-python/blob/master/examples/hash_image_search.py)).
-
-```tsql
-CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
-INSERT INTO items (embedding) VALUES (B'000'), (B'111');
-SELECT * FROM items ORDER BY bit_count(embedding # B'101') LIMIT 5;
-```
-
-Indexing is not currently supported.
-
 #### Do indexes need to fit into memory?
 
 No, but like other index types, you’ll likely see better performance if they do. You can get the size of an index with:
@@ -660,17 +544,7 @@ SELECT pg_size_pretty(pg_relation_size('index_name'));
 
 #### Why isn’t a query using an index?
 
-The query needs to have an `ORDER BY` and `LIMIT`, and the `ORDER BY` must be the result of a distance operator, not an expression.
+The cost estimation in pgvector < 0.4.3 does not always work well with the planner. You can encourage the planner to use an index for a query with:
-
-```sql
--- index
-ORDER BY embedding <=> '[3,1,2]' LIMIT 5;
-
--- no index
-ORDER BY 1 - (embedding <=> '[3,1,2]') DESC LIMIT 5;
-```
-
-You can encourage the planner to use an index for a query with:
 
 ```sql
 BEGIN;
@@ -699,12 +573,6 @@ or choose to store vectors inline:
 ALTER TABLE items ALTER COLUMN embedding SET STORAGE PLAIN;
 ```
 
-#### Why are there less results for a query after adding an HNSW index?
-
-Results are limited by the size of the dynamic candidate list (`hnsw.ef_search`). There may be even less results due to dead tuples or filtering conditions in the query. We recommend setting `hnsw.ef_search` to at least twice the `LIMIT` of the query. If you need more than 500 results, use an IVFFlat index instead.
-
-Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
-
 #### Why are there less results for a query after adding an IVFFlat index?
 
 The index was likely created with too little data for the number of lists. Drop the index until the table has more data.
@@ -713,15 +581,11 @@ The index was likely created with too little data for the number of lists. Drop
 DROP INDEX index_name;
 ```
 
-Results can also be limited by the number of probes (`ivfflat.probes`).
-
-Also, note that `NULL` vectors are not indexed (as well as zero vectors for cosine distance).
-
 ## Reference
 
 ### Vector Type
 
-Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a single-precision floating-point number (like the `real` type in Postgres), and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 16,000 dimensions.
+Each vector takes `4 * dimensions + 8` bytes of storage. Each element is a single precision floating-point number (like the `real` type in Postgres), and all elements must be finite (no `NaN`, `Infinity` or `-Infinity`). Vectors can have up to 16,000 dimensions.
 
 ### Vector Operators
 
@@ -745,14 +609,14 @@ l1_distance(vector, vector) → double precision | taxicab distance | 0.5.0
 vector_dims(vector) → integer | number of dimensions |
 vector_norm(vector) → double precision | Euclidean norm |
 
-### Vector Aggregate Functions
+### Aggregate Functions
 
 Function | Description | Added
 --- | --- | ---
 avg(vector) → vector | average |
 sum(vector) → vector | sum | 0.5.0
 
-## Installation Notes - Linux and Mac
+## Installation Notes
 
 ### Postgres Location
 
@@ -792,26 +656,6 @@ Note: Replace `16` with your Postgres server version
 
 If compilation fails and the output includes `warning: no such sysroot directory` on Mac, reinstall Xcode Command Line Tools.
 
-### Portability
-
-By default, pgvector compiles with `-march=native` on some platforms for best performance. However, this can lead to `Illegal instruction` errors if trying to run the compiled extension on a different machine.
-
-To compile for portability, use:
-
-```sh
-make OPTFLAGS=""
-```
-
-## Installation Notes - Windows
-
-### Missing Header
-
-If compilation fails with `Cannot open include file: 'postgres.h': No such file or directory`, make sure `PGROOT` is correct.
-
-### Permissions
-
-If installation fails with `Access is denied`, re-run the installation instructions as an administrator.
-
 ## Additional Installation Methods
 
 ### Docker
@@ -827,7 +671,7 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
+git clone --branch v0.6.0 https://github.com/pgvector/pgvector.git
 cd pgvector
 docker build --build-arg PG_MAJOR=16 -t myuser/pgvector .
 ```
@@ -883,7 +727,7 @@ pkg install postgresql15-pg_vector
 or the port with:
 
 ```sh
-cd /usr/ports/databases/pgvector
+cd /usr/ports/databases/pg_vector
 make install
 ```
 
@@ -996,12 +840,6 @@ make installcheck REGRESS=functions                            # regression test
 make prove_installcheck PROVE_TESTS=test/t/001_ivfflat_wal.pl  # TAP test
 ```
 
-To enable assertions:
-
-```sh
-make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
-```
-
 To enable benchmarking:
 
 ```sh
@@ -1014,6 +852,12 @@ To show memory usage:
 make clean && PG_CFLAGS="-DHNSW_MEMORY -DIVFFLAT_MEMORY" make && make install
 ```
 
+To enable assertions:
+
+```sh
+make clean && PG_CFLAGS="-DUSE_ASSERT_CHECKING" make && make install
+```
+
 To get k-means metrics:
 
 ```sh

sql/vector--0.6.0--0.6.1.sql

@@ -1,16 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.1'" to load this file. \quit
-
-DROP OPERATOR - (vector, vector);
-
-CREATE OPERATOR - (
-	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub
-);
-
-ALTER OPERATOR <= (vector, vector) SET (
-	RESTRICT = scalarlesel, JOIN = scalarlejoinsel
-);
-
-ALTER OPERATOR >= (vector, vector) SET (
-	RESTRICT = scalargesel, JOIN = scalargejoinsel
-);

sql/vector--0.6.1--0.6.2.sql

@@ -1,2 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.2'" to load this file. \quit

sql/vector--0.6.2--0.7.0.sql

@@ -1,95 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.7.0'" to load this file. \quit
-
-CREATE TYPE sparsevec;
-
-CREATE FUNCTION sparsevec_in(cstring, oid, integer) RETURNS sparsevec
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_out(sparsevec) RETURNS cstring
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_typmod_in(cstring[]) RETURNS integer
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_recv(internal, oid, integer) RETURNS sparsevec
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_send(sparsevec) RETURNS bytea
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE TYPE sparsevec (
-	INPUT     = sparsevec_in,
-	OUTPUT    = sparsevec_out,
-	TYPMOD_IN = sparsevec_typmod_in,
-	RECEIVE   = sparsevec_recv,
-	SEND      = sparsevec_send,
-	STORAGE   = external
-);
-
-CREATE FUNCTION l2_distance(sparsevec, sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME', 'sparsevec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION inner_product(sparsevec, sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME', 'sparsevec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION cosine_distance(sparsevec, sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME', 'sparsevec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_norm(sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_l2_squared_distance(sparsevec, sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_negative_inner_product(sparsevec, sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec(sparsevec, integer, boolean) RETURNS sparsevec
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION vector_to_sparsevec(vector, integer, boolean) RETURNS sparsevec
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_to_vector(sparsevec, integer, boolean) RETURNS vector
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE CAST (sparsevec AS sparsevec)
-	WITH FUNCTION sparsevec(sparsevec, integer, boolean) AS IMPLICIT;
-
-CREATE CAST (sparsevec AS vector)
-	WITH FUNCTION sparsevec_to_vector(sparsevec, integer, boolean) AS IMPLICIT;
-
-CREATE CAST (vector AS sparsevec)
-	WITH FUNCTION vector_to_sparsevec(vector, integer, boolean) AS IMPLICIT;
-
-CREATE OPERATOR <-> (
-	LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = l2_distance,
-	COMMUTATOR = '<->'
-);
-
-CREATE OPERATOR <#> (
-	LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_negative_inner_product,
-	COMMUTATOR = '<#>'
-);
-
-CREATE OPERATOR <=> (
-	LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = cosine_distance,
-	COMMUTATOR = '<=>'
-);
-
-CREATE OPERATOR CLASS sparsevec_l2_ops
-	FOR TYPE sparsevec USING hnsw AS
-	OPERATOR 1 <-> (sparsevec, sparsevec) FOR ORDER BY float_ops,
-	FUNCTION 1 sparsevec_l2_squared_distance(sparsevec, sparsevec);
-
-CREATE OPERATOR CLASS sparsevec_ip_ops
-	FOR TYPE sparsevec USING hnsw AS
-	OPERATOR 1 <#> (sparsevec, sparsevec) FOR ORDER BY float_ops,
-	FUNCTION 1 sparsevec_negative_inner_product(sparsevec, sparsevec);
-
-CREATE OPERATOR CLASS sparsevec_cosine_ops
-	FOR TYPE sparsevec USING hnsw AS
-	OPERATOR 1 <=> (sparsevec, sparsevec) FOR ORDER BY float_ops,
-	FUNCTION 1 sparsevec_negative_inner_product(sparsevec, sparsevec),
-	FUNCTION 2 sparsevec_norm(sparsevec);

sql/vector.sql

@@ -1,7 +1,7 @@
 -- complain if script is sourced in psql, rather than via CREATE EXTENSION
 \echo Use "CREATE EXTENSION vector" to load this file. \quit
 
--- vector type
+-- type
 
 CREATE TYPE vector;
 
@@ -29,7 +29,7 @@ CREATE TYPE vector (
 	STORAGE   = external
 );
 
--- vector functions
+-- functions
 
 CREATE FUNCTION l2_distance(vector, vector) RETURNS float8
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -58,7 +58,7 @@ CREATE FUNCTION vector_sub(vector, vector) RETURNS vector
 CREATE FUNCTION vector_mul(vector, vector) RETURNS vector
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
--- vector private functions
+-- private functions
 
 CREATE FUNCTION vector_lt(vector, vector) RETURNS bool
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -99,7 +99,7 @@ CREATE FUNCTION vector_avg(double precision[]) RETURNS vector
 CREATE FUNCTION vector_combine(double precision[], double precision[]) RETURNS double precision[]
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
--- vector aggregates
+-- aggregates
 
 CREATE AGGREGATE avg(vector) (
 	SFUNC = vector_accum,
@@ -117,7 +117,7 @@ CREATE AGGREGATE sum(vector) (
 	PARALLEL = SAFE
 );
 
--- vector cast functions
+-- cast functions
 
 CREATE FUNCTION vector(vector, integer, boolean) RETURNS vector
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -137,7 +137,7 @@ CREATE FUNCTION array_to_vector(numeric[], integer, boolean) RETURNS vector
 CREATE FUNCTION vector_to_float4(vector, integer, boolean) RETURNS real[]
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
--- vector casts
+-- casts
 
 CREATE CAST (vector AS vector)
 	WITH FUNCTION vector(vector, integer, boolean) AS IMPLICIT;
@@ -157,7 +157,7 @@ CREATE CAST (double precision[] AS vector)
 CREATE CAST (numeric[] AS vector)
 	WITH FUNCTION array_to_vector(numeric[], integer, boolean) AS ASSIGNMENT;
 
--- vector operators
+-- operators
 
 CREATE OPERATOR <-> (
 	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = l2_distance,
@@ -180,7 +180,8 @@ CREATE OPERATOR + (
 );
 
 CREATE OPERATOR - (
-	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub
+	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub,
+	COMMUTATOR = -
 );
 
 CREATE OPERATOR * (
@@ -194,10 +195,11 @@ CREATE OPERATOR < (
 	RESTRICT = scalarltsel, JOIN = scalarltjoinsel
 );
 
+-- should use scalarlesel and scalarlejoinsel, but not supported in Postgres < 11
 CREATE OPERATOR <= (
 	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_le,
 	COMMUTATOR = >= , NEGATOR = > ,
-	RESTRICT = scalarlesel, JOIN = scalarlejoinsel
+	RESTRICT = scalarltsel, JOIN = scalarltjoinsel
 );
 
 CREATE OPERATOR = (
@@ -212,10 +214,11 @@ CREATE OPERATOR <> (
 	RESTRICT = eqsel, JOIN = eqjoinsel
 );
 
+-- should use scalargesel and scalargejoinsel, but not supported in Postgres < 11
 CREATE OPERATOR >= (
 	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_ge,
 	COMMUTATOR = <= , NEGATOR = < ,
-	RESTRICT = scalargesel, JOIN = scalargejoinsel
+	RESTRICT = scalargtsel, JOIN = scalargtjoinsel
 );
 
 CREATE OPERATOR > (
@@ -240,7 +243,7 @@ CREATE ACCESS METHOD hnsw TYPE INDEX HANDLER hnswhandler;
 
 COMMENT ON ACCESS METHOD hnsw IS 'hnsw index access method';
 
--- vector opclasses
+-- opclasses
 
 CREATE OPERATOR CLASS vector_ops
 	DEFAULT FOR TYPE vector USING btree AS
@@ -287,110 +290,3 @@ CREATE OPERATOR CLASS vector_cosine_ops
 	OPERATOR 1 <=> (vector, vector) FOR ORDER BY float_ops,
 	FUNCTION 1 vector_negative_inner_product(vector, vector),
 	FUNCTION 2 vector_norm(vector);
-
---- sparsevec type
-
-CREATE TYPE sparsevec;
-
-CREATE FUNCTION sparsevec_in(cstring, oid, integer) RETURNS sparsevec
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_out(sparsevec) RETURNS cstring
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_typmod_in(cstring[]) RETURNS integer
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_recv(internal, oid, integer) RETURNS sparsevec
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_send(sparsevec) RETURNS bytea
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE TYPE sparsevec (
-	INPUT     = sparsevec_in,
-	OUTPUT    = sparsevec_out,
-	TYPMOD_IN = sparsevec_typmod_in,
-	RECEIVE   = sparsevec_recv,
-	SEND      = sparsevec_send,
-	STORAGE   = external
-);
-
--- sparsevec functions
-
-CREATE FUNCTION l2_distance(sparsevec, sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME', 'sparsevec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION inner_product(sparsevec, sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME', 'sparsevec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION cosine_distance(sparsevec, sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME', 'sparsevec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_norm(sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
--- sparsevec private functions
-
-CREATE FUNCTION sparsevec_l2_squared_distance(sparsevec, sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_negative_inner_product(sparsevec, sparsevec) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
--- sparsevec cast functions
-
-CREATE FUNCTION sparsevec(sparsevec, integer, boolean) RETURNS sparsevec
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION vector_to_sparsevec(vector, integer, boolean) RETURNS sparsevec
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION sparsevec_to_vector(sparsevec, integer, boolean) RETURNS vector
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
--- sparsevec casts
-
-CREATE CAST (sparsevec AS sparsevec)
-	WITH FUNCTION sparsevec(sparsevec, integer, boolean) AS IMPLICIT;
-
-CREATE CAST (sparsevec AS vector)
-	WITH FUNCTION sparsevec_to_vector(sparsevec, integer, boolean) AS IMPLICIT;
-
-CREATE CAST (vector AS sparsevec)
-	WITH FUNCTION vector_to_sparsevec(vector, integer, boolean) AS IMPLICIT;
-
--- sparsevec operators
-
-CREATE OPERATOR <-> (
-	LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = l2_distance,
-	COMMUTATOR = '<->'
-);
-
-CREATE OPERATOR <#> (
-	LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_negative_inner_product,
-	COMMUTATOR = '<#>'
-);
-
-CREATE OPERATOR <=> (
-	LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = cosine_distance,
-	COMMUTATOR = '<=>'
-);
-
--- sparsevec opclasses
-
-CREATE OPERATOR CLASS sparsevec_l2_ops
-	FOR TYPE sparsevec USING hnsw AS
-	OPERATOR 1 <-> (sparsevec, sparsevec) FOR ORDER BY float_ops,
-	FUNCTION 1 sparsevec_l2_squared_distance(sparsevec, sparsevec);
-
-CREATE OPERATOR CLASS sparsevec_ip_ops
-	FOR TYPE sparsevec USING hnsw AS
-	OPERATOR 1 <#> (sparsevec, sparsevec) FOR ORDER BY float_ops,
-	FUNCTION 1 sparsevec_negative_inner_product(sparsevec, sparsevec);
-
-CREATE OPERATOR CLASS sparsevec_cosine_ops
-	FOR TYPE sparsevec USING hnsw AS
-	OPERATOR 1 <=> (sparsevec, sparsevec) FOR ORDER BY float_ops,
-	FUNCTION 1 sparsevec_negative_inner_product(sparsevec, sparsevec),
-	FUNCTION 2 sparsevec_norm(sparsevec);

src/hnsw.c

@@ -8,7 +8,6 @@
 #include "commands/progress.h"
 #include "commands/vacuum.h"
 #include "hnsw.h"
-#include "miscadmin.h"
 #include "utils/guc.h"
 #include "utils/selfuncs.h"
 
@@ -29,7 +28,7 @@ static relopt_kind hnsw_relopt_kind;
  * this grows bigger, we should use a shmem_request_hook and
  * RequestAddinShmemSpace() to pre-reserve space for this.
  */
-void
+static void
 HnswInitLockTranche(void)
 {
 	int		   *tranche_ids;
@@ -54,8 +53,7 @@ HnswInitLockTranche(void)
 void
 HnswInit(void)
 {
-	if (!process_shared_preload_libraries_in_progress)
+	HnswInitLockTranche();
-		HnswInitLockTranche();
 
 	hnsw_relopt_kind = add_reloption_kind();
 	add_int_reloption(hnsw_relopt_kind, "m", "Max number of connections",

src/hnsw.h

@@ -17,7 +17,6 @@
 #endif
 
 #define HNSW_MAX_DIM 2000
-#define HNSW_MAX_NNZ 1000
 
 /* Support functions */
 #define HNSW_DISTANCE_PROC 1
@@ -56,12 +55,6 @@
 #define HNSW_UPDATE_ENTRY_GREATER 1
 #define HNSW_UPDATE_ENTRY_ALWAYS 2
 
-typedef enum HnswType
-{
-	HNSW_TYPE_VECTOR,
-	HNSW_TYPE_SPARSEVEC
-}			HnswType;
-
 /* Build phases */
 /* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
 #define PROGRESS_HNSW_PHASE_LOAD		2
@@ -87,7 +80,7 @@ typedef enum HnswType
 
 #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))
+#define list_sort(list, cmp) list_qsort(list, cmp)
 #endif
 
 #define HnswIsElementTuple(tup) ((tup)->type == HNSW_ELEMENT_TUPLE_TYPE)
@@ -136,7 +129,7 @@ HnswPtrDeclare(HnswNeighborArray, HnswNeighborArrayRelptr, HnswNeighborArrayPtr)
 HnswPtrDeclare(HnswNeighborArrayPtr, HnswNeighborsRelptr, HnswNeighborsPtr);
 HnswPtrDeclare(char, DatumRelptr, DatumPtr);
 
-struct HnswElementData
+typedef struct HnswElementData
 {
 	HnswElementPtr next;
 	ItemPointerData heaptids[HNSW_HEAPTIDS];
@@ -151,7 +144,7 @@ struct HnswElementData
 	BlockNumber neighborPage;
 	DatumPtr	value;
 	LWLock		lock;
-};
+}			HnswElementData;
 
 typedef HnswElementData * HnswElement;
 
@@ -162,12 +155,12 @@ typedef struct HnswCandidate
 	bool		closer;
 }			HnswCandidate;
 
-struct HnswNeighborArray
+typedef struct HnswNeighborArray
 {
 	int			length;
 	bool		closerSet;
 	HnswCandidate items[FLEXIBLE_ARRAY_MEMBER];
-};
+}			HnswNeighborArray;
 
 typedef struct HnswPairingHeapNode
 {
@@ -192,7 +185,6 @@ typedef struct HnswGraph
 
 	/* Entry state */
 	LWLock		entryLock;
-	LWLock		entryWaitLock;
 	HnswElementPtr entryPoint;
 
 	/* Allocations state */
@@ -249,7 +241,6 @@ typedef struct HnswBuildState
 	Relation	index;
 	IndexInfo  *indexInfo;
 	ForkNumber	forkNum;
-	HnswType	type;
 
 	/* Settings */
 	int			dimensions;
@@ -270,6 +261,7 @@ typedef struct HnswBuildState
 	HnswGraph  *graph;
 	double		ml;
 	int			maxLevel;
+	Vector	   *normvec;
 
 	/* Memory */
 	MemoryContext graphCtx;
@@ -374,9 +366,7 @@ typedef struct HnswVacuumState
 int			HnswGetM(Relation index);
 int			HnswGetEfConstruction(Relation index);
 FmgrInfo   *HnswOptionalProcInfo(Relation index, uint16 procnum);
-HnswType	HnswGetType(Relation index);
+bool		HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
-bool		HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type);
-void		HnswCheckValue(Datum value, HnswType type);
 Buffer		HnswNewBuffer(Relation index, ForkNumber forkNum);
 void		HnswInitPage(Buffer buf, Page page);
 void		HnswInit(void);
@@ -399,7 +389,6 @@ void		HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation i
 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		HnswInitLockTranche(void);
 PGDLLEXPORT void HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc);
 
 /* Index access methods */

src/hnswbuild.c

@@ -431,15 +431,10 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 	HnswGraph  *graph = buildstate->graph;
 	HnswElement entryPoint;
 	LWLock	   *entryLock = &graph->entryLock;
-	LWLock	   *entryWaitLock = &graph->entryWaitLock;
 	int			efConstruction = buildstate->efConstruction;
 	int			m = buildstate->m;
 	char	   *base = buildstate->hnswarea;
 
-	/* Wait if another process needs exclusive lock on entry lock */
-	LWLockAcquire(entryWaitLock, LW_EXCLUSIVE);
-	LWLockRelease(entryWaitLock);
-
 	/* Get entry point */
 	LWLockAcquire(entryLock, LW_SHARED);
 	entryPoint = HnswPtrAccess(base, graph->entryPoint);
@@ -450,10 +445,8 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 		/* Release shared lock */
 		LWLockRelease(entryLock);
 
-		/* Tell other processes to wait and get exclusive lock */
+		/* Get exclusive lock */
-		LWLockAcquire(entryWaitLock, LW_EXCLUSIVE);
 		LWLockAcquire(entryLock, LW_EXCLUSIVE);
-		LWLockRelease(entryWaitLock);
 
 		/* Get latest entry point after lock is acquired */
 		entryPoint = HnswPtrAccess(base, graph->entryPoint);
@@ -486,13 +479,10 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid, Hn
 	/* Detoast once for all calls */
 	Datum		value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
 
-	/* Check value */
-	HnswCheckValue(value, buildstate->type);
-
 	/* Normalize if needed */
 	if (buildstate->normprocinfo != NULL)
 	{
-		if (!HnswNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->type))
+		if (!HnswNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->normvec))
 			return false;
 	}
 
@@ -611,9 +601,6 @@ BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
 static void
 InitGraph(HnswGraph * graph, char *base, long memoryTotal)
 {
-	/* Initialize the lock tranche if needed */
-	HnswInitLockTranche();
-
 	HnswPtrStore(base, graph->head, (HnswElement) NULL);
 	HnswPtrStore(base, graph->entryPoint, (HnswElement) NULL);
 	graph->memoryUsed = 0;
@@ -622,7 +609,6 @@ InitGraph(HnswGraph * graph, char *base, long memoryTotal)
 	graph->indtuples = 0;
 	SpinLockInit(&graph->lock);
 	LWLockInitialize(&graph->entryLock, hnsw_lock_tranche_id);
-	LWLockInitialize(&graph->entryWaitLock, hnsw_lock_tranche_id);
 	LWLockInitialize(&graph->allocatorLock, hnsw_lock_tranche_id);
 	LWLockInitialize(&graph->flushLock, hnsw_lock_tranche_id);
 }
@@ -674,28 +660,21 @@ HnswSharedMemoryAlloc(Size size, void *state)
 static void
 InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, IndexInfo *indexInfo, ForkNumber forkNum)
 {
-	int			maxDimensions = HNSW_MAX_DIM;
-
 	buildstate->heap = heap;
 	buildstate->index = index;
 	buildstate->indexInfo = indexInfo;
 	buildstate->forkNum = forkNum;
-	buildstate->type = HnswGetType(index);
 
 	buildstate->m = HnswGetM(index);
 	buildstate->efConstruction = HnswGetEfConstruction(index);
 	buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
 
-	/* No limit on sparse vector dimensions */
-	if (buildstate->type == HNSW_TYPE_SPARSEVEC)
-		maxDimensions = INT_MAX;
-
 	/* Require column to have dimensions to be indexed */
 	if (buildstate->dimensions < 0)
 		elog(ERROR, "column does not have dimensions");
 
-	if (buildstate->dimensions > maxDimensions)
+	if (buildstate->dimensions > HNSW_MAX_DIM)
-		elog(ERROR, "column cannot have more than %d dimensions for hnsw index", maxDimensions);
+		elog(ERROR, "column cannot have more than %d dimensions for hnsw index", HNSW_MAX_DIM);
 
 	if (buildstate->efConstruction < 2 * buildstate->m)
 		elog(ERROR, "ef_construction must be greater than or equal to 2 * m");
@@ -713,6 +692,9 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 	buildstate->ml = HnswGetMl(buildstate->m);
 	buildstate->maxLevel = HnswGetMaxLevel(buildstate->m);
 
+	/* Reuse for each tuple */
+	buildstate->normvec = InitVector(buildstate->dimensions);
+
 	buildstate->graphCtx = GenerationContextCreate(CurrentMemoryContext,
 												   "Hnsw build graph context",
 #if PG_VERSION_NUM >= 150000
@@ -736,6 +718,7 @@ InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, Index
 static void
 FreeBuildState(HnswBuildState * buildstate)
 {
+	pfree(buildstate->normvec);
 	MemoryContextDelete(buildstate->graphCtx);
 	MemoryContextDelete(buildstate->tmpCtx);
 }
@@ -991,14 +974,6 @@ HnswBeginParallel(HnswBuildState * buildstate, bool isconcurrent, int request)
 	/* Report less than allocated so never fails */
 	InitGraph(&hnswshared->graphData, hnswarea, esthnswarea - 1024 * 1024);
 
-	/*
-	 * Avoid base address for relptr for Postgres < 14.5
-	 * https://github.com/postgres/postgres/commit/7201cd18627afc64850537806da7f22150d1a83b
-	 */
-#if PG_VERSION_NUM < 140005
-	hnswshared->graphData.memoryUsed += MAXALIGN(1);
-#endif
-
 	shm_toc_insert(pcxt->toc, PARALLEL_KEY_HNSW_SHARED, hnswshared);
 	shm_toc_insert(pcxt->toc, PARALLEL_KEY_HNSW_AREA, hnswarea);
 

src/hnswinsert.c

@@ -614,19 +614,15 @@ HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_ti
 	Datum		value;
 	FmgrInfo   *normprocinfo;
 	Oid			collation = index->rd_indcollation[0];
-	HnswType	type = HnswGetType(index);
 
 	/* Detoast once for all calls */
 	value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
 
-	/* Check value */
-	HnswCheckValue(value, type);
-
 	/* Normalize if needed */
 	normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
 	if (normprocinfo != NULL)
 	{
-		if (!HnswNormValue(normprocinfo, collation, &value, type))
+		if (!HnswNormValue(normprocinfo, collation, &value, NULL))
 			return;
 	}
 

src/hnswscan.c

@@ -40,6 +40,29 @@ GetScanItems(IndexScanDesc scan, Datum q)
 	return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, procinfo, collation, m, false, NULL);
 }
 
+/*
+ * Get dimensions from metapage
+ */
+static int
+GetDimensions(Relation index)
+{
+	Buffer		buf;
+	Page		page;
+	HnswMetaPage metap;
+	int			dimensions;
+
+	buf = ReadBuffer(index, HNSW_METAPAGE_BLKNO);
+	LockBuffer(buf, BUFFER_LOCK_SHARE);
+	page = BufferGetPage(buf);
+	metap = HnswPageGetMeta(page);
+
+	dimensions = metap->dimensions;
+
+	UnlockReleaseBuffer(buf);
+
+	return dimensions;
+}
+
 /*
  * Get scan value
  */
@@ -50,7 +73,7 @@ GetScanValue(IndexScanDesc scan)
 	Datum		value;
 
 	if (scan->orderByData->sk_flags & SK_ISNULL)
-		value = PointerGetDatum(NULL);
+		value = PointerGetDatum(InitVector(GetDimensions(scan->indexRelation)));
 	else
 	{
 		value = scan->orderByData->sk_argument;
@@ -61,7 +84,7 @@ GetScanValue(IndexScanDesc scan)
 
 		/* Fine if normalization fails */
 		if (so->normprocinfo != NULL)
-			HnswNormValue(so->normprocinfo, so->collation, &value, HnswGetType(scan->indexRelation));
+			HnswNormValue(so->normprocinfo, so->collation, &value, NULL);
 	}
 
 	return value;

src/hnswutils.c

@@ -3,15 +3,12 @@
 #include <math.h>
 
 #include "access/generic_xlog.h"
-#include "catalog/pg_type.h"
 #include "hnsw.h"
 #include "lib/pairingheap.h"
-#include "sparsevec.h"
 #include "storage/bufmgr.h"
 #include "utils/datum.h"
 #include "utils/memdebug.h"
 #include "utils/rel.h"
-#include "utils/syscache.h"
 #include "vector.h"
 
 #if PG_VERSION_NUM >= 130000
@@ -152,32 +149,6 @@ HnswOptionalProcInfo(Relation index, uint16 procnum)
 	return index_getprocinfo(index, 1, procnum);
 }
 
-/*
- * Get vector type
- */
-HnswType
-HnswGetType(Relation index)
-{
-	Oid			typeOid = TupleDescAttr(index->rd_att, 0)->atttypid;
-	HeapTuple	tuple;
-	Form_pg_type type;
-	int			result;
-
-	tuple = SearchSysCache1(TYPEOID, ObjectIdGetDatum(typeOid));
-	if (!HeapTupleIsValid(tuple))
-		elog(ERROR, "cache lookup failed for type %u", typeOid);
-
-	type = (Form_pg_type) GETSTRUCT(tuple);
-	if (strcmp(NameStr(type->typname), "sparsevec") == 0)
-		result = HNSW_TYPE_SPARSEVEC;
-	else
-		result = HNSW_TYPE_VECTOR;
-
-	ReleaseSysCache(tuple);
-
-	return result;
-}
-
 /*
  * Divide by the norm
  *
@@ -187,40 +158,21 @@ HnswGetType(Relation index)
  * if it's different than the original value
  */
 bool
-HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type)
+HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
 {
 	double		norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
 
 	if (norm > 0)
 	{
-		/* TODO Remove vector-specific code */
+		Vector	   *v = DatumGetVector(*value);
-		if (type == HNSW_TYPE_VECTOR)
-		{
-			Vector	   *v = DatumGetVector(*value);
-			Vector	   *result = InitVector(v->dim);
 
-			for (int i = 0; i < v->dim; i++)
+		if (result == NULL)
-				result->x[i] = v->x[i] / norm;
+			result = InitVector(v->dim);
 
-			*value = PointerGetDatum(result);
+		for (int i = 0; i < v->dim; i++)
-		}
+			result->x[i] = v->x[i] / norm;
-		else if (type == HNSW_TYPE_SPARSEVEC)
-		{
-			SparseVector *v = DatumGetSparseVector(*value);
-			SparseVector *result = InitSparseVector(v->dim, v->nnz);
-			float	   *vx = SPARSEVEC_VALUES(v);
-			float	   *rx = SPARSEVEC_VALUES(result);
 
-			for (int i = 0; i < v->nnz; i++)
+		*value = PointerGetDatum(result);
-			{
-				result->indices[i] = v->indices[i];
-				rx[i] = vx[i] / norm;
-			}
-
-			*value = PointerGetDatum(result);
-		}
-		else
-			elog(ERROR, "Unsupported type");
 
 		return true;
 	}
@@ -228,21 +180,6 @@ HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, HnswType type)
 	return false;
 }
 
-/*
- * Check if a value can be indexed
- */
-void
-HnswCheckValue(Datum value, HnswType type)
-{
-	if (type == HNSW_TYPE_SPARSEVEC)
-	{
-		SparseVector *vec = DatumGetSparseVector(value);
-
-		if (vec->nnz > HNSW_MAX_NNZ)
-			elog(ERROR, "sparsevec cannot have more than %d non-zero elements for hnsw index", HNSW_MAX_NNZ);
-	}
-}
-
 /*
  * New buffer
  */
@@ -638,12 +575,7 @@ HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index,
 
 	/* Calculate distance */
 	if (distance != NULL)
-	{
+		*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->data)));
-		if (DatumGetPointer(*q) == NULL)
-			*distance = 0;
-		else
-			*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->data)));
-	}
 
 	UnlockReleaseBuffer(buf);
 }
@@ -928,15 +860,12 @@ HnswSearchLayer(char *base, Datum q, List *ep, int ef, int lc, Relation index, F
 static int
 #if PG_VERSION_NUM >= 130000
 CompareCandidateDistances(const ListCell *a, const ListCell *b)
-{
-	HnswCandidate *hca = lfirst(a);
-	HnswCandidate *hcb = lfirst(b);
 #else
 CompareCandidateDistances(const void *a, const void *b)
-{
-	HnswCandidate *hca = lfirst(*(ListCell **) a);
-	HnswCandidate *hcb = lfirst(*(ListCell **) b);
 #endif
+{
+	HnswCandidate *hca = lfirst((ListCell *) a);
+	HnswCandidate *hcb = lfirst((ListCell *) b);
 
 	if (hca->distance < hcb->distance)
 		return 1;
@@ -959,15 +888,12 @@ CompareCandidateDistances(const void *a, const void *b)
 static int
 #if PG_VERSION_NUM >= 130000
 CompareCandidateDistancesOffset(const ListCell *a, const ListCell *b)
-{
-	HnswCandidate *hca = lfirst(a);
-	HnswCandidate *hcb = lfirst(b);
 #else
 CompareCandidateDistancesOffset(const void *a, const void *b)
-{
-	HnswCandidate *hca = lfirst(*(ListCell **) a);
-	HnswCandidate *hcb = lfirst(*(ListCell **) b);
 #endif
+{
+	HnswCandidate *hca = lfirst((ListCell *) a);
+	HnswCandidate *hcb = lfirst((ListCell *) b);
 
 	if (hca->distance < hcb->distance)
 		return 1;
@@ -1026,9 +952,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
 {
 	List	   *r = NIL;
 	List	   *w = list_copy(c);
-	HnswCandidate **wd;
+	pairingheap *wd;
-	int			wdlen = 0;
-	int			wdoff = 0;
 	HnswNeighborArray *neighbors = HnswGetNeighbors(base, e2, lc);
 	bool		mustCalculate = !neighbors->closerSet;
 	List	   *added = NIL;
@@ -1037,7 +961,7 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
 	if (list_length(w) <= lm)
 		return w;
 
-	wd = palloc(sizeof(HnswCandidate *) * list_length(w));
+	wd = pairingheap_allocate(CompareNearestCandidates, NULL);
 
 	/* Ensure order of candidates is deterministic for closer caching */
 	if (sortCandidates)
@@ -1103,21 +1027,21 @@ SelectNeighbors(char *base, List *c, int lm, int lc, FmgrInfo *procinfo, Oid col
 		if (e->closer)
 			r = lappend(r, e);
 		else
-			wd[wdlen++] = e;
+			pairingheap_add(wd, &(CreatePairingHeapNode(e)->ph_node));
 	}
 
 	/* Cached value can only be used in future if sorted deterministically */
 	neighbors->closerSet = sortCandidates;
 
 	/* Keep pruned connections */
-	while (wdoff < wdlen && list_length(r) < lm)
+	while (!pairingheap_is_empty(wd) && list_length(r) < lm)
-		r = lappend(r, wd[wdoff++]);
+		r = lappend(r, ((HnswPairingHeapNode *) pairingheap_remove_first(wd))->inner);
 
 	/* Return pruned for update connections */
 	if (pruned != NULL)
 	{
-		if (wdoff < wdlen)
+		if (!pairingheap_is_empty(wd))
-			*pruned = wd[wdoff];
+			*pruned = ((HnswPairingHeapNode *) pairingheap_first(wd))->inner;
 		else
 			*pruned = linitial(w);
 	}

src/ivfbuild.c

@@ -57,7 +57,7 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
 	 */
 	if (buildstate->kmeansnormprocinfo != NULL)
 	{
-		if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value))
+		if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value, buildstate->normvec))
 			return;
 	}
 
@@ -105,7 +105,7 @@ SampleCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
 	oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
 
 	/* Add sample */
-	AddSample(values, buildstate);
+	AddSample(values, state);
 
 	/* Reset memory context */
 	MemoryContextSwitchTo(oldCtx);
@@ -153,7 +153,7 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
 	/* Normalize if needed */
 	if (buildstate->normprocinfo != NULL)
 	{
-		if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value))
+		if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->normvec))
 			return;
 	}
 
@@ -356,6 +356,9 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 	buildstate->centers = VectorArrayInit(buildstate->lists, buildstate->dimensions);
 	buildstate->listInfo = palloc(sizeof(ListInfo) * buildstate->lists);
 
+	/* Reuse for each tuple */
+	buildstate->normvec = InitVector(buildstate->dimensions);
+
 	buildstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
 											   "Ivfflat build temporary context",
 											   ALLOCSET_DEFAULT_SIZES);
@@ -377,6 +380,7 @@ FreeBuildState(IvfflatBuildState * buildstate)
 {
 	VectorArrayFree(buildstate->centers);
 	pfree(buildstate->listInfo);
+	pfree(buildstate->normvec);
 
 #ifdef IVFFLAT_KMEANS_DEBUG
 	pfree(buildstate->listSums);

src/ivfflat.h

@@ -172,6 +172,7 @@ typedef struct IvfflatBuildState
 	VectorArray samples;
 	VectorArray centers;
 	ListInfo   *listInfo;
+	Vector	   *normvec;
 
 #ifdef IVFFLAT_KMEANS_DEBUG
 	double		inertia;
@@ -266,7 +267,7 @@ void		VectorArrayFree(VectorArray arr);
 void		PrintVectorArray(char *msg, VectorArray arr);
 void		IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers);
 FmgrInfo   *IvfflatOptionalProcInfo(Relation index, uint16 procnum);
-bool		IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value);
+bool		IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
 int			IvfflatGetLists(Relation index);
 void		IvfflatGetMetaPageInfo(Relation index, int *lists, int *dimensions);
 void		IvfflatUpdateList(Relation index, ListInfo listInfo, BlockNumber insertPage, BlockNumber originalInsertPage, BlockNumber startPage, ForkNumber forkNum);

src/ivfinsert.c

@@ -85,7 +85,7 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, R
 	normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
 	if (normprocinfo != NULL)
 	{
-		if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value))
+		if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value, NULL))
 			return;
 	}
 

src/ivfscan.c

@@ -293,7 +293,7 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 
 			/* Fine if normalization fails */
 			if (so->normprocinfo != NULL)
-				IvfflatNormValue(so->normprocinfo, so->collation, &value);
+				IvfflatNormValue(so->normprocinfo, so->collation, &value, NULL);
 		}
 
 		IvfflatBench("GetScanLists", GetScanLists(scan, value));

src/ivfutils.c

@@ -75,14 +75,16 @@ IvfflatOptionalProcInfo(Relation index, uint16 procnum)
  * if it's different than the original value
  */
 bool
-IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value)
+IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
 {
 	double		norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
 
 	if (norm > 0)
 	{
 		Vector	   *v = DatumGetVector(*value);
-		Vector	   *result = InitVector(v->dim);
+
+		if (result == NULL)
+			result = InitVector(v->dim);
 
 		for (int i = 0; i < v->dim; i++)
 			result->x[i] = v->x[i] / norm;

src/sparsevec.c

@@ -1,779 +0,0 @@
-#include "postgres.h"
-
-#include <limits.h>
-#include <math.h>
-
-#include "fmgr.h"
-#include "libpq/pqformat.h"
-#include "sparsevec.h"
-#include "utils/array.h"
-#include "utils/builtins.h"
-#include "vector.h"
-
-#if PG_VERSION_NUM >= 120000
-#include "common/shortest_dec.h"
-#include "utils/float.h"
-#else
-#include <float.h>
-#include "utils/builtins.h"
-#endif
-
-/*
- * Ensure same dimensions
- */
-static inline void
-CheckDims(SparseVector * a, SparseVector * b)
-{
-	if (a->dim != b->dim)
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("different sparsevec dimensions %d and %d", a->dim, b->dim)));
-}
-
-/*
- * Ensure expected dimensions
- */
-static inline void
-CheckExpectedDim(int32 typmod, int dim)
-{
-	if (typmod != -1 && typmod != dim)
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("expected %d dimensions, not %d", typmod, dim)));
-}
-
-/*
- * Ensure valid dimensions
- */
-static inline void
-CheckDim(int dim)
-{
-	if (dim < 1)
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("sparsevec must have at least 1 dimension")));
-
-	if (dim > SPARSEVEC_MAX_DIM)
-		ereport(ERROR,
-				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
-				 errmsg("sparsevec cannot have more than %d dimensions", SPARSEVEC_MAX_DIM)));
-}
-
-/*
- * Ensure valid nnz
- */
-static inline void
-CheckNnz(int nnz, int dim)
-{
-	if (nnz < 0)
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("sparsevec must have at least one element")));
-
-	if (nnz > dim)
-		ereport(ERROR,
-				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
-				 errmsg("sparsevec cannot have more elements than dimensions")));
-}
-
-/*
- * Ensure valid index
- */
-static inline void
-CheckIndex(int32 *indices, int i, int dim)
-{
-	int32		index = indices[i];
-
-	if (index < 0)
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("index must not be negative")));
-
-	if (index >= dim)
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("index must be less than dimensions")));
-
-	if (i > 0)
-	{
-		if (index < indices[i - 1])
-			ereport(ERROR,
-					(errcode(ERRCODE_DATA_EXCEPTION),
-					 errmsg("indexes must be in ascending order")));
-
-		if (index == indices[i - 1])
-			ereport(ERROR,
-					(errcode(ERRCODE_DATA_EXCEPTION),
-					 errmsg("indexes must not contain duplicates")));
-	}
-}
-
-/*
- * Ensure finite element
- */
-static inline void
-CheckElement(float value)
-{
-	if (isnan(value))
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("NaN not allowed in sparsevec")));
-
-	if (isinf(value))
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("infinite value not allowed in sparsevec")));
-}
-
-/*
- * Allocate and initialize a new sparse vector
- */
-SparseVector *
-InitSparseVector(int dim, int nnz)
-{
-	SparseVector *result;
-	int			size;
-
-	size = SPARSEVEC_SIZE(nnz);
-	result = (SparseVector *) palloc0(size);
-	SET_VARSIZE(result, size);
-	result->dim = dim;
-	result->nnz = nnz;
-
-	return result;
-}
-
-/*
- * Check for whitespace, since array_isspace() is static
- */
-static inline bool
-sparsevec_isspace(char ch)
-{
-	if (ch == ' ' ||
-		ch == '\t' ||
-		ch == '\n' ||
-		ch == '\r' ||
-		ch == '\v' ||
-		ch == '\f')
-		return true;
-	return false;
-}
-
-/*
- * Convert textual representation to internal representation
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_in);
-Datum
-sparsevec_in(PG_FUNCTION_ARGS)
-{
-	char	   *lit = PG_GETARG_CSTRING(0);
-	int32		typmod = PG_GETARG_INT32(2);
-	int			dim;
-	char	   *pt;
-	char	   *stringEnd;
-	SparseVector *result;
-	float	   *rvalues;
-	char	   *litcopy = pstrdup(lit);
-	char	   *str = litcopy;
-	int32	   *indices;
-	float	   *values;
-	int			maxNnz;
-	int			nnz = 0;
-
-	maxNnz = 1;
-	pt = str;
-	while (*pt != '\0')
-	{
-		if (*pt == ',')
-			maxNnz++;
-
-		pt++;
-	}
-
-	indices = palloc(maxNnz * sizeof(int32));
-	values = palloc(maxNnz * sizeof(float));
-
-	while (sparsevec_isspace(*str))
-		str++;
-
-	if (*str != '{')
-		ereport(ERROR,
-				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-				 errmsg("malformed sparsevec literal: \"%s\"", lit),
-				 errdetail("Vector contents must start with \"{\".")));
-
-	str++;
-	pt = strtok(str, ",");
-	stringEnd = pt;
-
-	while (pt != NULL && *stringEnd != '}')
-	{
-		long		index;
-		float		value;
-
-		/* TODO Better error */
-		if (nnz == maxNnz)
-			ereport(ERROR,
-					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("ran out of buffer: \"%s\"", lit)));
-
-		while (sparsevec_isspace(*pt))
-			pt++;
-
-		/* Check for empty string like float4in */
-		if (*pt == '\0')
-			ereport(ERROR,
-					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
-
-		/* Use similar logic as int2vectorin */
-		errno = 0;
-		index = strtol(pt, &stringEnd, 10);
-
-		if (stringEnd == pt)
-			ereport(ERROR,
-					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
-
-		if (errno == ERANGE || index < 0 || index > INT_MAX)
-			ereport(ERROR,
-					(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-					 errmsg("index \"%ld\" is out of range for type sparsevec", index)));
-
-		if (stringEnd == pt)
-			ereport(ERROR,
-					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
-
-		while (sparsevec_isspace(*stringEnd))
-			stringEnd++;
-
-		if (*stringEnd != ':')
-			ereport(ERROR,
-					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
-
-		stringEnd++;
-
-		while (sparsevec_isspace(*stringEnd))
-			stringEnd++;
-
-		errno = 0;
-		pt = stringEnd;
-		value = strtof(pt, &stringEnd);
-
-		if (stringEnd == pt)
-			ereport(ERROR,
-					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
-
-		/* Check for range error like float4in */
-		if (errno == ERANGE && (value == 0 || isinf(value)))
-			ereport(ERROR,
-					(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-					 errmsg("\"%s\" is out of range for type sparsevec", pt)));
-
-		/* TODO Decide whether to store zero values */
-		if (value != 0)
-		{
-			indices[nnz] = index;
-			values[nnz] = value;
-			nnz++;
-		}
-
-		if (*stringEnd != '\0' && *stringEnd != '}')
-			ereport(ERROR,
-					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
-
-		pt = strtok(NULL, ",");
-	}
-
-	if (stringEnd == NULL || *stringEnd != '}')
-		ereport(ERROR,
-				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-				 errmsg("malformed sparsevec literal: \"%s\"", lit),
-				 errdetail("Unexpected end of input.")));
-
-	stringEnd++;
-
-	if (*stringEnd != '/')
-		ereport(ERROR,
-				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-				 errmsg("malformed sparsevec literal: \"%s\"", lit),
-				 errdetail("Unexpected end of input.")));
-
-	stringEnd++;
-
-	/* Use similar logic as int2vectorin */
-	errno = 0;
-	pt = stringEnd;
-	dim = strtol(pt, &stringEnd, 10);
-
-	if (stringEnd == pt)
-		ereport(ERROR,
-				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-				 errmsg("invalid input syntax for type sparsevec: \"%s\"", lit)));
-
-	/* Only whitespace is allowed after the closing brace */
-	while (sparsevec_isspace(*stringEnd))
-		stringEnd++;
-
-	if (*stringEnd != '\0')
-		ereport(ERROR,
-				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-				 errmsg("malformed sparsevec literal: \"%s\"", lit),
-				 errdetail("Junk after closing.")));
-
-	pfree(litcopy);
-
-	CheckDim(dim);
-	CheckExpectedDim(typmod, dim);
-
-	result = InitSparseVector(dim, nnz);
-	rvalues = SPARSEVEC_VALUES(result);
-	for (int i = 0; i < nnz; i++)
-	{
-		result->indices[i] = indices[i];
-		rvalues[i] = values[i];
-
-		CheckIndex(result->indices, i, dim);
-		CheckElement(rvalues[i]);
-	}
-
-	PG_RETURN_POINTER(result);
-}
-
-#define AppendChar(ptr, c) (*(ptr)++ = (c))
-#define AppendFloat(ptr, f) ((ptr) += float_to_shortest_decimal_bufn((f), (ptr)))
-
-#if PG_VERSION_NUM >= 140000
-#define AppendInt(ptr, i) ((ptr) += pg_ltoa((i), (ptr)))
-#else
-#define AppendInt(ptr, i) \
-	do { \
-		pg_ltoa(i, ptr); \
-		while (*ptr != '\0') \
-			ptr++; \
-	} while (0)
-#endif
-
-/*
- * Convert internal representation to textual representation
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_out);
-Datum
-sparsevec_out(PG_FUNCTION_ARGS)
-{
-	SparseVector *sparsevec = PG_GETARG_SPARSEVEC_P(0);
-	float	   *values = SPARSEVEC_VALUES(sparsevec);
-	char	   *buf;
-	char	   *ptr;
-
-	/*
-	 * Need:
-	 *
-	 * nnz * 10 bytes for index (positive integer)
-	 *
-	 * nnz bytes for :
-	 *
-	 * nnz * (FLOAT_SHORTEST_DECIMAL_LEN - 1) bytes for
-	 * float_to_shortest_decimal_bufn
-	 *
-	 * nnz - 1 bytes for ,
-	 *
-	 * 10 bytes for dimensions
-	 *
-	 * 4 bytes for {, }, /, and \0
-	 */
-	buf = (char *) palloc((11 + FLOAT_SHORTEST_DECIMAL_LEN) * sparsevec->nnz + 13);
-	ptr = buf;
-
-	AppendChar(ptr, '{');
-
-	for (int i = 0; i < sparsevec->nnz; i++)
-	{
-		if (i > 0)
-			AppendChar(ptr, ',');
-
-		AppendInt(ptr, sparsevec->indices[i]);
-		AppendChar(ptr, ':');
-		AppendFloat(ptr, values[i]);
-	}
-
-	AppendChar(ptr, '}');
-	AppendChar(ptr, '/');
-	AppendInt(ptr, sparsevec->dim);
-
-	*ptr = '\0';
-
-	PG_FREE_IF_COPY(sparsevec, 0);
-	PG_RETURN_CSTRING(buf);
-}
-
-/*
- * Convert type modifier
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_typmod_in);
-Datum
-sparsevec_typmod_in(PG_FUNCTION_ARGS)
-{
-	ArrayType  *ta = PG_GETARG_ARRAYTYPE_P(0);
-	int32	   *tl;
-	int			n;
-
-	tl = ArrayGetIntegerTypmods(ta, &n);
-
-	if (n != 1)
-		ereport(ERROR,
-				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-				 errmsg("invalid type modifier")));
-
-	if (*tl < 1)
-		ereport(ERROR,
-				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-				 errmsg("dimensions for type sparsevec must be at least 1")));
-
-	if (*tl > SPARSEVEC_MAX_DIM)
-		ereport(ERROR,
-				(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
-				 errmsg("dimensions for type sparsevec cannot exceed %d", SPARSEVEC_MAX_DIM)));
-
-	PG_RETURN_INT32(*tl);
-}
-
-/*
- * Convert external binary representation to internal representation
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_recv);
-Datum
-sparsevec_recv(PG_FUNCTION_ARGS)
-{
-	StringInfo	buf = (StringInfo) PG_GETARG_POINTER(0);
-	int32		typmod = PG_GETARG_INT32(2);
-	SparseVector *result;
-	int32		dim;
-	int32		nnz;
-	int32		unused;
-	float	   *values;
-
-	dim = pq_getmsgint(buf, sizeof(int32));
-	nnz = pq_getmsgint(buf, sizeof(int32));
-	unused = pq_getmsgint(buf, sizeof(int32));
-
-	CheckDim(dim);
-	CheckNnz(nnz, dim);
-	CheckExpectedDim(typmod, dim);
-
-	if (unused != 0)
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("expected unused to be 0, not %d", unused)));
-
-	result = InitSparseVector(dim, nnz);
-	values = SPARSEVEC_VALUES(result);
-
-	for (int i = 0; i < nnz; i++)
-	{
-		result->indices[i] = pq_getmsgint(buf, sizeof(int32));
-		CheckIndex(result->indices, i, dim);
-	}
-
-	for (int i = 0; i < nnz; i++)
-	{
-		values[i] = pq_getmsgfloat4(buf);
-		CheckElement(values[i]);
-	}
-
-	PG_RETURN_POINTER(result);
-}
-
-/*
- * Convert internal representation to the external binary representation
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_send);
-Datum
-sparsevec_send(PG_FUNCTION_ARGS)
-{
-	SparseVector *svec = PG_GETARG_SPARSEVEC_P(0);
-	float	   *values = SPARSEVEC_VALUES(svec);
-	StringInfoData buf;
-
-	pq_begintypsend(&buf);
-	pq_sendint(&buf, svec->dim, sizeof(int32));
-	pq_sendint(&buf, svec->nnz, sizeof(int32));
-	pq_sendint(&buf, svec->unused, sizeof(int32));
-	for (int i = 0; i < svec->nnz; i++)
-		pq_sendint(&buf, svec->indices[i], sizeof(int32));
-	for (int i = 0; i < svec->nnz; i++)
-		pq_sendfloat4(&buf, values[i]);
-
-	PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
-}
-
-/*
- * Convert sparse vector to sparse vector
- * This is needed to check the type modifier
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec);
-Datum
-sparsevec(PG_FUNCTION_ARGS)
-{
-	SparseVector *svec = PG_GETARG_SPARSEVEC_P(0);
-	int32		typmod = PG_GETARG_INT32(1);
-
-	CheckExpectedDim(typmod, svec->dim);
-
-	PG_RETURN_POINTER(svec);
-}
-
-/*
- * Convert dense vector to sparse vector
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_to_sparsevec);
-Datum
-vector_to_sparsevec(PG_FUNCTION_ARGS)
-{
-	Vector	   *vec = PG_GETARG_VECTOR_P(0);
-	int32		typmod = PG_GETARG_INT32(1);
-	SparseVector *result;
-	int			dim = vec->dim;
-	int			nnz = 0;
-	float	   *values;
-	int			j = 0;
-
-	CheckDim(dim);
-	CheckExpectedDim(typmod, dim);
-
-	for (int i = 0; i < dim; i++)
-	{
-		if (vec->x[i] != 0)
-			nnz++;
-	}
-
-	result = InitSparseVector(dim, nnz);
-	values = SPARSEVEC_VALUES(result);
-	for (int i = 0; i < dim; i++)
-	{
-		if (vec->x[i] != 0)
-		{
-			/* Safety check */
-			if (j == nnz)
-				elog(ERROR, "safety check failed");
-
-			result->indices[j] = i;
-			values[j] = vec->x[i];
-			j++;
-		}
-	}
-
-	PG_RETURN_POINTER(result);
-}
-
-/*
- *  Get the L2 squared distance between sparse vectors
- */
-static double
-l2_distance_squared_internal(SparseVector * a, SparseVector * b)
-{
-	float	   *ax = SPARSEVEC_VALUES(a);
-	float	   *bx = SPARSEVEC_VALUES(b);
-	double		distance = 0.0;
-	int			bpos = 0;
-
-	for (int i = 0; i < a->nnz; i++)
-	{
-		int			ai = a->indices[i];
-		int			bi = -1;
-
-		for (int j = bpos; j < b->nnz; j++)
-		{
-			bi = b->indices[j];
-
-			if (ai == bi)
-			{
-				double		diff = ax[i] - bx[j];
-
-				distance += diff * diff;
-			}
-			else if (ai > bi)
-				distance += bx[j] * bx[j];
-
-			/* Update start for next iteration */
-			if (ai >= bi)
-				bpos = j + 1;
-
-			/* Found or passed it */
-			if (bi >= ai)
-				break;
-		}
-
-		if (ai != bi)
-			distance += ax[i] * ax[i];
-	}
-
-	for (int j = bpos; j < b->nnz; j++)
-		distance += bx[j] * bx[j];
-
-	return distance;
-}
-
-/*
- * Get the L2 distance between sparse vectors
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_l2_distance);
-Datum
-sparsevec_l2_distance(PG_FUNCTION_ARGS)
-{
-	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
-	SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
-
-	CheckDims(a, b);
-
-	PG_RETURN_FLOAT8(sqrt(l2_distance_squared_internal(a, b)));
-}
-
-/*
- * Get the L2 squared distance between sparse vectors
- * This saves a sqrt calculation
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_l2_squared_distance);
-Datum
-sparsevec_l2_squared_distance(PG_FUNCTION_ARGS)
-{
-	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
-	SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
-
-	CheckDims(a, b);
-
-	PG_RETURN_FLOAT8(l2_distance_squared_internal(a, b));
-}
-
-/*
- * Get the inner product of two sparse vectors
- */
-static double
-inner_product_internal(SparseVector * a, SparseVector * b)
-{
-	float	   *ax = SPARSEVEC_VALUES(a);
-	float	   *bx = SPARSEVEC_VALUES(b);
-	double		distance = 0.0;
-	int			bpos = 0;
-
-	for (int i = 0; i < a->nnz; i++)
-	{
-		int			ai = a->indices[i];
-
-		for (int j = bpos; j < b->nnz; j++)
-		{
-			int			bi = b->indices[j];
-
-			/* Only update when the same index */
-			if (ai == bi)
-				distance += ax[i] * bx[j];
-
-			/* Update start for next iteration */
-			if (ai >= bi)
-				bpos = j + 1;
-
-			/* Found or passed it */
-			if (bi >= ai)
-				break;
-		}
-	}
-
-	return distance;
-}
-
-/*
- * Get the inner product of two sparse vectors
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_inner_product);
-Datum
-sparsevec_inner_product(PG_FUNCTION_ARGS)
-{
-	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
-	SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
-
-	CheckDims(a, b);
-
-	PG_RETURN_FLOAT8(inner_product_internal(a, b));
-}
-
-/*
- * Get the negative inner product of two sparse vectors
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_negative_inner_product);
-Datum
-sparsevec_negative_inner_product(PG_FUNCTION_ARGS)
-{
-	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
-	SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
-
-	CheckDims(a, b);
-
-	PG_RETURN_FLOAT8(-inner_product_internal(a, b));
-}
-
-/*
- * Get the cosine distance between two sparse vectors
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_cosine_distance);
-Datum
-sparsevec_cosine_distance(PG_FUNCTION_ARGS)
-{
-	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
-	SparseVector *b = PG_GETARG_SPARSEVEC_P(1);
-	float	   *ax = SPARSEVEC_VALUES(a);
-	float	   *bx = SPARSEVEC_VALUES(b);
-	float		norma = 0.0;
-	float		normb = 0.0;
-	double		similarity;
-
-	CheckDims(a, b);
-
-	similarity = inner_product_internal(a, b);
-
-	/* Auto-vectorized */
-	for (int i = 0; i < a->nnz; i++)
-		norma += ax[i] * ax[i];
-
-	/* Auto-vectorized */
-	for (int i = 0; i < b->nnz; i++)
-		normb += bx[i] * bx[i];
-
-	/* Use sqrt(a * b) over sqrt(a) * sqrt(b) */
-	similarity /= sqrt((double) norma * (double) normb);
-
-#ifdef _MSC_VER
-	/* /fp:fast may not propagate NaN */
-	if (isnan(similarity))
-		PG_RETURN_FLOAT8(NAN);
-#endif
-
-	/* Keep in range */
-	if (similarity > 1)
-		similarity = 1.0;
-	else if (similarity < -1)
-		similarity = -1.0;
-
-	PG_RETURN_FLOAT8(1.0 - similarity);
-}
-
-/*
- * Get the L2 norm of a sparse vector
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_norm);
-Datum
-sparsevec_norm(PG_FUNCTION_ARGS)
-{
-	SparseVector *a = PG_GETARG_SPARSEVEC_P(0);
-	float	   *ax = SPARSEVEC_VALUES(a);
-	double		norm = 0.0;
-
-	/* Auto-vectorized */
-	for (int i = 0; i < a->nnz; i++)
-		norm += (double) ax[i] * (double) ax[i];
-
-	PG_RETURN_FLOAT8(sqrt(norm));
-}

src/sparsevec.h

@@ -1,23 +0,0 @@
-#ifndef SPARSEVEC_H
-#define SPARSEVEC_H
-
-#define SPARSEVEC_MAX_DIM 100000
-
-#define SPARSEVEC_SIZE(_nnz)		(offsetof(SparseVector, indices) + MAXALIGN((_nnz) * sizeof(int32)) + (_nnz * sizeof(float)))
-#define SPARSEVEC_VALUES(x)		((float *) (((char *) (x)) + offsetof(SparseVector, indices) + MAXALIGN((x)->nnz * sizeof(int32))))
-#define DatumGetSparseVector(x)		((SparseVector *) PG_DETOAST_DATUM(x))
-#define PG_GETARG_SPARSEVEC_P(x)	DatumGetSparseVector(PG_GETARG_DATUM(x))
-#define PG_RETURN_SPARSEVEC_P(x)	PG_RETURN_POINTER(x)
-
-typedef struct SparseVector
-{
-	int32		vl_len_;		/* varlena header (do not touch directly!) */
-	int32		dim;			/* number of dimensions */
-	int32		nnz;
-	int32		unused;
-	int32		indices[FLEXIBLE_ARRAY_MEMBER];
-}			SparseVector;
-
-SparseVector *InitSparseVector(int dim, int nnz);
-
-#endif

src/vector.c

@@ -10,7 +10,6 @@
 #include "lib/stringinfo.h"
 #include "libpq/pqformat.h"
 #include "port.h"				/* for strtof() */
-#include "sparsevec.h"
 #include "utils/array.h"
 #include "utils/builtins.h"
 #include "utils/float.h"
@@ -1161,26 +1160,3 @@ vector_avg(PG_FUNCTION_ARGS)
 
 	PG_RETURN_POINTER(result);
 }
-
-/*
- * Convert sparse vector to dense vector
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(sparsevec_to_vector);
-Datum
-sparsevec_to_vector(PG_FUNCTION_ARGS)
-{
-	SparseVector *svec = PG_GETARG_SPARSEVEC_P(0);
-	int32		typmod = PG_GETARG_INT32(1);
-	Vector	   *result;
-	int			dim = svec->dim;
-	float	   *values = SPARSEVEC_VALUES(svec);
-
-	CheckDim(dim);
-	CheckExpectedDim(typmod, dim);
-
-	result = InitVector(dim);
-	for (int i = 0; i < svec->nnz; i++)
-		result->x[svec->indices[i]] = values[i];
-
-	PG_RETURN_POINTER(result);
-}

test/expected/functions.out

@@ -104,105 +104,105 @@ SELECT vector_norm('[3e37,4e37]')::real;
        5e+37
 (1 row)
 
-SELECT l2_distance('[0,0]'::vector, '[3,4]');
+SELECT l2_distance('[0,0]', '[3,4]');
  l2_distance 
 -------------
            5
 (1 row)
 
-SELECT l2_distance('[0,0]'::vector, '[0,1]');
+SELECT l2_distance('[0,0]', '[0,1]');
  l2_distance 
 -------------
            1
 (1 row)
 
-SELECT l2_distance('[1,2]'::vector, '[3]');
+SELECT l2_distance('[1,2]', '[3]');
 ERROR:  different vector dimensions 2 and 1
-SELECT l2_distance('[3e38]'::vector, '[-3e38]');
+SELECT l2_distance('[3e38]', '[-3e38]');
  l2_distance 
 -------------
     Infinity
 (1 row)
 
-SELECT inner_product('[1,2]'::vector, '[3,4]');
+SELECT inner_product('[1,2]', '[3,4]');
  inner_product 
 ---------------
             11
 (1 row)
 
-SELECT inner_product('[1,2]'::vector, '[3]');
+SELECT inner_product('[1,2]', '[3]');
 ERROR:  different vector dimensions 2 and 1
-SELECT inner_product('[3e38]'::vector, '[3e38]');
+SELECT inner_product('[3e38]', '[3e38]');
  inner_product 
 ---------------
       Infinity
 (1 row)
 
-SELECT cosine_distance('[1,2]'::vector, '[2,4]');
+SELECT cosine_distance('[1,2]', '[2,4]');
  cosine_distance 
 -----------------
                0
 (1 row)
 
-SELECT cosine_distance('[1,2]'::vector, '[0,0]');
+SELECT cosine_distance('[1,2]', '[0,0]');
  cosine_distance 
 -----------------
              NaN
 (1 row)
 
-SELECT cosine_distance('[1,1]'::vector, '[1,1]');
+SELECT cosine_distance('[1,1]', '[1,1]');
  cosine_distance 
 -----------------
                0
 (1 row)
 
-SELECT cosine_distance('[1,0]'::vector, '[0,2]');
+SELECT cosine_distance('[1,0]', '[0,2]');
  cosine_distance 
 -----------------
                1
 (1 row)
 
-SELECT cosine_distance('[1,1]'::vector, '[-1,-1]');
+SELECT cosine_distance('[1,1]', '[-1,-1]');
  cosine_distance 
 -----------------
                2
 (1 row)
 
-SELECT cosine_distance('[1,2]'::vector, '[3]');
+SELECT cosine_distance('[1,2]', '[3]');
 ERROR:  different vector dimensions 2 and 1
-SELECT cosine_distance('[1,1]'::vector, '[1.1,1.1]');
+SELECT cosine_distance('[1,1]', '[1.1,1.1]');
  cosine_distance 
 -----------------
                0
 (1 row)
 
-SELECT cosine_distance('[1,1]'::vector, '[-1.1,-1.1]');
+SELECT cosine_distance('[1,1]', '[-1.1,-1.1]');
  cosine_distance 
 -----------------
                2
 (1 row)
 
-SELECT cosine_distance('[3e38]'::vector, '[3e38]');
+SELECT cosine_distance('[3e38]', '[3e38]');
  cosine_distance 
 -----------------
              NaN
 (1 row)
 
-SELECT l1_distance('[0,0]'::vector, '[3,4]');
+SELECT l1_distance('[0,0]', '[3,4]');
  l1_distance 
 -------------
            7
 (1 row)
 
-SELECT l1_distance('[0,0]'::vector, '[0,1]');
+SELECT l1_distance('[0,0]', '[0,1]');
  l1_distance 
 -------------
            1
 (1 row)
 
-SELECT l1_distance('[1,2]'::vector, '[3]');
+SELECT l1_distance('[1,2]', '[3]');
 ERROR:  different vector dimensions 2 and 1
-SELECT l1_distance('[3e38]'::vector, '[-3e38]');
+SELECT l1_distance('[3e38]', '[-3e38]');
  l1_distance 
 -------------
     Infinity

test/expected/hnsw_l2.out

@@ -12,11 +12,14 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
  [0,0,0]
 (4 rows)
 
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector)) t2;
+SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
- count 
+   val   
--------
+---------
-     4
+ [0,0,0]
-(1 row)
+ [1,1,1]
+ [1,2,3]
+ [1,2,4]
+(4 rows)
 
 SELECT COUNT(*) FROM t;
  count 

test/expected/hnsw_sparsevec_cosine.out

@@ -1,26 +0,0 @@
-SET enable_seqscan = off;
-CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
-CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
-SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}/3';
-       val       
------------------
- {0:1,1:1,2:1}/3
- {0:1,1:2,2:3}/3
- {0:1,1:2,2:4}/3
-(3 rows)
-
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}/3') t2;
- count 
--------
-     3
-(1 row)
-
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::sparsevec)) t2;
- count 
--------
-     3
-(1 row)
-
-DROP TABLE t;

test/expected/hnsw_sparsevec_ip.out

@@ -1,21 +0,0 @@
-SET enable_seqscan = off;
-CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
-CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
-SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}/3';
-       val       
------------------
- {0:1,1:2,2:4}/3
- {0:1,1:2,2:3}/3
- {0:1,1:1,2:1}/3
- {}/3
-(4 rows)
-
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;
- count 
--------
-     4
-(1 row)
-
-DROP TABLE t;

test/expected/hnsw_sparsevec_l2.out

@@ -1,43 +0,0 @@
-SET enable_seqscan = off;
-CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
-CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
-SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
-       val       
------------------
- {0:1,1:2,2:3}/3
- {0:1,1:2,2:4}/3
- {0:1,1:1,2:1}/3
- {}/3
-(4 rows)
-
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::sparsevec)) t2;
- count 
--------
-     4
-(1 row)
-
-SELECT COUNT(*) FROM t;
- count 
--------
-     5
-(1 row)
-
-TRUNCATE t;
-SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
- val 
------
-(0 rows)
-
-DROP TABLE t;
--- TODO move
-CREATE TABLE t (val sparsevec(1001));
-INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
-CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
-ERROR:  sparsevec cannot have more than 1000 non-zero elements for hnsw index
-TRUNCATE t;
-CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
-INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
-ERROR:  sparsevec cannot have more than 1000 non-zero elements for hnsw index
-DROP TABLE t;

test/expected/input.out

@@ -66,22 +66,6 @@ SELECT '[4e38,1]'::vector;
 ERROR:  infinite value not allowed in vector
 LINE 1: SELECT '[4e38,1]'::vector;
                ^
-SELECT '[-4e38,1]'::vector;
-ERROR:  infinite value not allowed in vector
-LINE 1: SELECT '[-4e38,1]'::vector;
-               ^
-SELECT '[1e-46,1]'::vector;
- vector 
---------
- [0,1]
-(1 row)
-
-SELECT '[-1e-46,1]'::vector;
- vector 
---------
- [-0,1]
-(1 row)
-
 SELECT '[1,2,3'::vector;
 ERROR:  malformed vector literal: "[1,2,3"
 LINE 1: SELECT '[1,2,3'::vector;
@@ -132,30 +116,8 @@ SELECT '[1, ,3]'::vector;
 ERROR:  invalid input syntax for type vector: "[1, ,3]"
 LINE 1: SELECT '[1, ,3]'::vector;
                ^
-SELECT '[1,2,3]'::vector(3);
- vector  
----------
- [1,2,3]
-(1 row)
-
 SELECT '[1,2,3]'::vector(2);
 ERROR:  expected 2 dimensions, not 3
-SELECT '[1,2,3]'::vector(3, 2);
-ERROR:  invalid type modifier
-LINE 1: SELECT '[1,2,3]'::vector(3, 2);
-                          ^
-SELECT '[1,2,3]'::vector('a');
-ERROR:  invalid input syntax for type integer: "a"
-LINE 1: SELECT '[1,2,3]'::vector('a');
-                          ^
-SELECT '[1,2,3]'::vector(0);
-ERROR:  dimensions for type vector must be at least 1
-LINE 1: SELECT '[1,2,3]'::vector(0);
-                          ^
-SELECT '[1,2,3]'::vector(16001);
-ERROR:  dimensions for type vector cannot exceed 16000
-LINE 1: SELECT '[1,2,3]'::vector(16001);
-                          ^
 SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
  unnest  
 ---------

test/expected/sparsevec_functions.out

@@ -1,62 +0,0 @@
-SELECT l2_distance('{}/2'::sparsevec, '{0:3,1:4}/2');
- l2_distance 
--------------
-           5
-(1 row)
-
-SELECT l2_distance('{}/2'::sparsevec, '{1:1}/2');
- l2_distance 
--------------
-           1
-(1 row)
-
-SELECT '{}/2'::sparsevec <-> '{0:3,1:4}/2';
- ?column? 
-----------
-        5
-(1 row)
-
-SELECT inner_product('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
- inner_product 
----------------
-            10
-(1 row)
-
-SELECT sparsevec_negative_inner_product('{0:1,1:2}/2', '{0:2,1:4}/2');
- sparsevec_negative_inner_product 
-----------------------------------
-                              -10
-(1 row)
-
-SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
- cosine_distance 
------------------
-               0
-(1 row)
-
-SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{}/2');
- cosine_distance 
------------------
-             NaN
-(1 row)
-
-SELECT cosine_distance('{0:1,1:1}/2'::sparsevec, '{0:-1,1:-1}/2');
- cosine_distance 
------------------
-               2
-(1 row)
-
-SELECT cosine_distance('{0:1}/2'::sparsevec, '{1:2}/2');
- cosine_distance 
------------------
-               1
-(1 row)
-
-SELECT cosine_distance('{}/1'::sparsevec, '{}/1');
- cosine_distance 
------------------
-             NaN
-(1 row)
-
-SELECT cosine_distance('{0:1}/2'::sparsevec, '{0:1}/3');
-ERROR:  different sparsevec dimensions 2 and 3

test/expected/sparsevec_input.out

@@ -1,62 +0,0 @@
-SELECT '{0:1.5,2:3.5}/5'::sparsevec;
-    sparsevec    
------------------
- {0:1.5,2:3.5}/5
-(1 row)
-
-SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector;
-     vector      
------------------
- [1.5,0,3.5,0,0]
-(1 row)
-
-SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(5);
-     vector      
------------------
- [1.5,0,3.5,0,0]
-(1 row)
-
-SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(4);
-ERROR:  expected 4 dimensions, not 5
-SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
-    sparsevec    
------------------
- {1:1.5,3:3.5}/5
-(1 row)
-
-SELECT '{0:0,1:1,2:0}/3'::sparsevec;
- sparsevec 
------------
- {1:1}/3
-(1 row)
-
-SELECT '{1:1,0:1}/2'::sparsevec;
-ERROR:  indexes must be in ascending order
-LINE 1: SELECT '{1:1,0:1}/2'::sparsevec;
-               ^
-SELECT '{}/5'::sparsevec;
- sparsevec 
------------
- {}/5
-(1 row)
-
-SELECT '{}/-1'::sparsevec;
-ERROR:  sparsevec must have at least 1 dimension
-LINE 1: SELECT '{}/-1'::sparsevec;
-               ^
-SELECT '{}/100001'::sparsevec;
-ERROR:  sparsevec cannot have more than 100000 dimensions
-LINE 1: SELECT '{}/100001'::sparsevec;
-               ^
-SELECT '{}/16001'::sparsevec::vector;
-ERROR:  vector cannot have more than 16000 dimensions
-SELECT '{-1:1}/1'::sparsevec;
-ERROR:  index "-1" is out of range for type sparsevec
-LINE 1: SELECT '{-1:1}/1'::sparsevec;
-               ^
-SELECT '{1:1}/1'::sparsevec;
-ERROR:  index must be less than dimensions
-LINE 1: SELECT '{1:1}/1'::sparsevec;
-               ^
-SELECT '{}/1'::sparsevec(2);
-ERROR:  expected 2 dimensions, not 1

test/sql/functions.sql

@@ -24,29 +24,29 @@ SELECT vector_norm('[3,4]');
 SELECT vector_norm('[0,1]');
 SELECT vector_norm('[3e37,4e37]')::real;
 
-SELECT l2_distance('[0,0]'::vector, '[3,4]');
+SELECT l2_distance('[0,0]', '[3,4]');
-SELECT l2_distance('[0,0]'::vector, '[0,1]');
+SELECT l2_distance('[0,0]', '[0,1]');
-SELECT l2_distance('[1,2]'::vector, '[3]');
+SELECT l2_distance('[1,2]', '[3]');
-SELECT l2_distance('[3e38]'::vector, '[-3e38]');
+SELECT l2_distance('[3e38]', '[-3e38]');
 
-SELECT inner_product('[1,2]'::vector, '[3,4]');
+SELECT inner_product('[1,2]', '[3,4]');
-SELECT inner_product('[1,2]'::vector, '[3]');
+SELECT inner_product('[1,2]', '[3]');
-SELECT inner_product('[3e38]'::vector, '[3e38]');
+SELECT inner_product('[3e38]', '[3e38]');
 
-SELECT cosine_distance('[1,2]'::vector, '[2,4]');
+SELECT cosine_distance('[1,2]', '[2,4]');
-SELECT cosine_distance('[1,2]'::vector, '[0,0]');
+SELECT cosine_distance('[1,2]', '[0,0]');
-SELECT cosine_distance('[1,1]'::vector, '[1,1]');
+SELECT cosine_distance('[1,1]', '[1,1]');
-SELECT cosine_distance('[1,0]'::vector, '[0,2]');
+SELECT cosine_distance('[1,0]', '[0,2]');
-SELECT cosine_distance('[1,1]'::vector, '[-1,-1]');
+SELECT cosine_distance('[1,1]', '[-1,-1]');
-SELECT cosine_distance('[1,2]'::vector, '[3]');
+SELECT cosine_distance('[1,2]', '[3]');
-SELECT cosine_distance('[1,1]'::vector, '[1.1,1.1]');
+SELECT cosine_distance('[1,1]', '[1.1,1.1]');
-SELECT cosine_distance('[1,1]'::vector, '[-1.1,-1.1]');
+SELECT cosine_distance('[1,1]', '[-1.1,-1.1]');
-SELECT cosine_distance('[3e38]'::vector, '[3e38]');
+SELECT cosine_distance('[3e38]', '[3e38]');
 
-SELECT l1_distance('[0,0]'::vector, '[3,4]');
+SELECT l1_distance('[0,0]', '[3,4]');
-SELECT l1_distance('[0,0]'::vector, '[0,1]');
+SELECT l1_distance('[0,0]', '[0,1]');
-SELECT l1_distance('[1,2]'::vector, '[3]');
+SELECT l1_distance('[1,2]', '[3]');
-SELECT l1_distance('[3e38]'::vector, '[-3e38]');
+SELECT l1_distance('[3e38]', '[-3e38]');
 
 SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
 SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;

test/sql/hnsw_l2.sql

@@ -7,7 +7,7 @@ CREATE INDEX ON t USING hnsw (val vector_l2_ops);
 INSERT INTO t (val) VALUES ('[1,2,4]');
 
 SELECT * FROM t ORDER BY val <-> '[3,3,3]';
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector)) t2;
+SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
 SELECT COUNT(*) FROM t;
 
 TRUNCATE t;

test/sql/hnsw_sparsevec_cosine.sql

@@ -1,13 +0,0 @@
-SET enable_seqscan = off;
-
-CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
-CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
-
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
-
-SELECT * FROM t ORDER BY val <=> '{0:3,1:3,2:3}/3';
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '{}/3') t2;
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::sparsevec)) t2;
-
-DROP TABLE t;

test/sql/hnsw_sparsevec_ip.sql

@@ -1,12 +0,0 @@
-SET enable_seqscan = off;
-
-CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
-CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
-
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
-
-SELECT * FROM t ORDER BY val <#> '{0:3,1:3,2:3}/3';
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;
-
-DROP TABLE t;

test/sql/hnsw_sparsevec_l2.sql

@@ -1,25 +0,0 @@
-SET enable_seqscan = off;
-
-CREATE TABLE t (val sparsevec(3));
-INSERT INTO t (val) VALUES ('{}/3'), ('{0:1,1:2,2:3}/3'), ('{0:1,1:1,2:1}/3'), (NULL);
-CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
-
-INSERT INTO t (val) VALUES ('{0:1,1:2,2:4}/3');
-
-SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::sparsevec)) t2;
-SELECT COUNT(*) FROM t;
-
-TRUNCATE t;
-SELECT * FROM t ORDER BY val <-> '{0:3,1:3,2:3}/3';
-
-DROP TABLE t;
-
--- TODO move
-CREATE TABLE t (val sparsevec(1001));
-INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
-CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
-TRUNCATE t;
-CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
-INSERT INTO t (val) VALUES (array_fill(1, ARRAY[1001])::vector::sparsevec);
-DROP TABLE t;

test/sql/input.sql

@@ -11,9 +11,6 @@ SELECT '[1.5e38,-1.5e38]'::vector;
 SELECT '[1.5e+38,-1.5e+38]'::vector;
 SELECT '[1.5e-38,-1.5e-38]'::vector;
 SELECT '[4e38,1]'::vector;
-SELECT '[-4e38,1]'::vector;
-SELECT '[1e-46,1]'::vector;
-SELECT '[-1e-46,1]'::vector;
 SELECT '[1,2,3'::vector;
 SELECT '[1,2,3]9'::vector;
 SELECT '1,2,3'::vector;
@@ -25,13 +22,7 @@ SELECT '[1,]'::vector;
 SELECT '[1a]'::vector;
 SELECT '[1,,3]'::vector;
 SELECT '[1, ,3]'::vector;
-
-SELECT '[1,2,3]'::vector(3);
 SELECT '[1,2,3]'::vector(2);
-SELECT '[1,2,3]'::vector(3, 2);
-SELECT '[1,2,3]'::vector('a');
-SELECT '[1,2,3]'::vector(0);
-SELECT '[1,2,3]'::vector(16001);
 
 SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
 SELECT '{"[1,2,3]"}'::vector(2)[];

test/sql/sparsevec_functions.sql

@@ -1,13 +0,0 @@
-SELECT l2_distance('{}/2'::sparsevec, '{0:3,1:4}/2');
-SELECT l2_distance('{}/2'::sparsevec, '{1:1}/2');
-SELECT '{}/2'::sparsevec <-> '{0:3,1:4}/2';
-
-SELECT inner_product('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
-SELECT sparsevec_negative_inner_product('{0:1,1:2}/2', '{0:2,1:4}/2');
-
-SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{0:2,1:4}/2');
-SELECT cosine_distance('{0:1,1:2}/2'::sparsevec, '{}/2');
-SELECT cosine_distance('{0:1,1:1}/2'::sparsevec, '{0:-1,1:-1}/2');
-SELECT cosine_distance('{0:1}/2'::sparsevec, '{1:2}/2');
-SELECT cosine_distance('{}/1'::sparsevec, '{}/1');
-SELECT cosine_distance('{0:1}/2'::sparsevec, '{0:1}/3');

test/sql/sparsevec_input.sql

@@ -1,19 +0,0 @@
-SELECT '{0:1.5,2:3.5}/5'::sparsevec;
-SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector;
-SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(5);
-SELECT '{0:1.5,2:3.5}/5'::sparsevec::vector(4);
-SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
-
-SELECT '{0:0,1:1,2:0}/3'::sparsevec;
-
-SELECT '{1:1,0:1}/2'::sparsevec;
-
-SELECT '{}/5'::sparsevec;
-SELECT '{}/-1'::sparsevec;
-SELECT '{}/100001'::sparsevec;
-SELECT '{}/16001'::sparsevec::vector;
-
-SELECT '{-1:1}/1'::sparsevec;
-SELECT '{1:1}/1'::sparsevec;
-
-SELECT '{}/1'::sparsevec(2);

test/t/016_hnsw_vacuum_recall.pl

@@ -86,7 +86,7 @@ foreach (@queries)
 	push(@expected, $res);
 }
 
-test_recall(0.19, $limit, "before vacuum");
+test_recall(0.20, $limit, "before vacuum");
 test_recall(0.95, 100, "before vacuum");
 
 # TODO Test concurrent inserts with vacuum

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat and hnsw access methods'
-default_version = '0.6.2'
+default_version = '0.6.0'
 module_pathname = '$libdir/vector'
 relocatable = true