Added support for adaptive beam search [skip ci]

Postgres standard way to check for list emptiness is to compare a pointer to
NIL rather than NULL.

.dockerignore

@@ -1,8 +0,0 @@
-/.git/
-/dist/
-/results/
-/tmp_check/
-/sql/vector--?.?.?.sql
-regression.*
-*.o
-*.so

.github/workflows/build.yml

@@ -8,6 +8,8 @@ jobs:
       fail-fast: false
       matrix:
         include:
+          - postgres: 19
+            os: ubuntu-24.04
           - postgres: 18
             os: ubuntu-24.04
           - postgres: 17
@@ -21,14 +23,14 @@ jobs:
           - postgres: 13
             os: ubuntu-22.04
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
       - uses: ankane/setup-postgres@v1
         with:
           postgres-version: ${{ matrix.postgres }}
           dev-files: true
       - run: make
         env:
-          PG_CFLAGS: ${{ matrix.postgres == 18 && '-Wno-missing-field-initializers' || '' }} -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
+          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }}
       - run: |
           export PG_CONFIG=`which pg_config`
           sudo --preserve-env=PG_CONFIG make install
@@ -46,18 +48,18 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - postgres: 17
+          - postgres: 18
-            os: macos-15
+            os: macos-26
           - postgres: 14
-            os: macos-13
+            os: macos-15-intel
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
       - uses: ankane/setup-postgres@v1
         with:
           postgres-version: ${{ matrix.postgres }}
       - run: make
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter
+          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-unknown-warning-option ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }}
       - run: make install
       - run: make installcheck
       - if: ${{ failure() }}
@@ -70,27 +72,35 @@ jobs:
           tar xf $TAG.tar.gz
           mv postgres-$TAG postgres
         env:
-          TAG: ${{ matrix.postgres == 17 && 'REL_17_2' || 'REL_14_15' }}
+          TAG: ${{ matrix.postgres == 18 && 'REL_18_0' || 'REL_14_19' }}
       - run: make prove_installcheck PROVE_FLAGS="-I ./postgres/src/test/perl -I ./test/perl"
         env:
           PERL5LIB: /Users/runner/perl5/lib/perl5
       - run: make clean && $(brew --prefix llvm@$LLVM_VERSION)/bin/scan-build --status-bugs make
         env:
-          LLVM_VERSION: ${{ matrix.os == 'macos-15' && 18 || 15 }}
+          LLVM_VERSION: ${{ matrix.os == 'macos-26' && 20 || 18 }}
           PG_CFLAGS: -DUSE_ASSERT_CHECKING
   windows:
-    runs-on: windows-latest
+    runs-on: ${{ matrix.os }}
     if: ${{ !startsWith(github.ref_name, 'mac') }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - postgres: 17
+            os: windows-2025
+          - postgres: 14
+            os: windows-2022
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
       - uses: ankane/setup-postgres@v1
         with:
-          postgres-version: 14
+          postgres-version: ${{ matrix.postgres }}
       - run: |
           call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
           nmake /NOLOGO /F Makefile.win && ^
           nmake /NOLOGO /F Makefile.win install && ^
-          nmake /NOLOGO /F Makefile.win installcheck && ^
+          nmake /NOLOGO /F Makefile.win installcheck ${{ matrix.postgres != 17 && 'PG_REGRESS=$(PGROOT)\bin\pg_regress' || '' }} && ^
           nmake /NOLOGO /F Makefile.win clean && ^
           nmake /NOLOGO /F Makefile.win uninstall
         shell: cmd
@@ -123,10 +133,10 @@ jobs:
     if: ${{ !startsWith(github.ref_name, 'mac') && !startsWith(github.ref_name, 'windows') }}
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v5
       - uses: ankane/setup-postgres-valgrind@v1
         with:
-          postgres-version: 17
+          postgres-version: 18
           check-ub: yes
       - run: make OPTFLAGS=""
       - run: sudo --preserve-env=PG_CONFIG make install

CHANGELOG.md

@@ -1,3 +1,13 @@
+## 0.8.2 (unreleased)
+
+- Improved `install` target on Windows
+- Fixed `Index Searches` in `EXPLAIN` output for Postgres 18
+
+## 0.8.1 (2025-09-04)
+
+- Added support for Postgres 18 rc1
+- Improved performance of `binary_quantize` function
+
 ## 0.8.0 (2024-10-30)
 
 - Added support for iterative index scans

Dockerfile

@@ -1,8 +1,11 @@
+# syntax=docker/dockerfile:1
+
 ARG PG_MAJOR=17
-FROM postgres:$PG_MAJOR
+ARG DEBIAN_CODENAME=bookworm
+FROM postgres:$PG_MAJOR-$DEBIAN_CODENAME
 ARG PG_MAJOR
 
-COPY . /tmp/pgvector
+ADD https://github.com/pgvector/pgvector.git#v0.8.1 /tmp/pgvector
 
 RUN apt-get update && \
 		apt-mark hold locales && \

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": "0.8.0",
+	"version": "0.8.1",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.8.0",
+			"version": "0.8.1",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.8.0
+EXTVERSION = 0.8.1
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*--*.sql)
@@ -76,4 +76,9 @@ docker:
 .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=$(PG_MAJOR) --build-arg DEBIAN_CODENAME=bookworm -t pgvector/pgvector:pg$(PG_MAJOR) -t pgvector/pgvector:pg$(PG_MAJOR)-bookworm -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR) -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR)-bookworm .
+
+.PHONY: docker-release-trixie
+
+docker-release-trixie:
+	docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 --build-arg PG_MAJOR=$(PG_MAJOR) --build-arg DEBIAN_CODENAME=trixie -t pgvector/pgvector:pg$(PG_MAJOR)-trixie -t pgvector/pgvector:$(EXTVERSION)-pg$(PG_MAJOR)-trixie .

Makefile.win

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.8.0
+EXTVERSION = 0.8.1
 
 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
@@ -31,6 +31,9 @@ LIBDIR = $(PGROOT)\lib
 PKGLIBDIR = $(PGROOT)\lib
 SHAREDIR = $(PGROOT)\share
 
+# Use $(PGROOT)\bin\pg_regress for Postgres < 17
+PG_REGRESS = $(LIBDIR)\pgxs\src\test\regress\pg_regress
+
 CFLAGS = /nologo /I"$(INCLUDEDIR_SERVER)\port\win32_msvc" /I"$(INCLUDEDIR_SERVER)\port\win32" /I"$(INCLUDEDIR_SERVER)" /I"$(INCLUDEDIR)"
 
 CFLAGS = $(CFLAGS) $(PG_CFLAGS)
@@ -54,11 +57,11 @@ install: all
 	copy $(SHLIB) "$(PKGLIBDIR)"
 	copy $(EXTENSION).control "$(SHAREDIR)\extension"
 	copy sql\$(EXTENSION)--*.sql "$(SHAREDIR)\extension"
-	mkdir "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
+	if not exist "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)" mkdir "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
 	for %f in ($(HEADERS)) do copy %f "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
 
 installcheck:
-	"$(BINDIR)\pg_regress" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS)
+	"$(PG_REGRESS)" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS)
 
 uninstall:
 	del /f "$(PKGLIBDIR)\$(SHLIB)"

README.md

@@ -21,7 +21,7 @@ Compile and install the extension (supports Postgres 13+)
 
 ```sh
 cd /tmp
-git clone --branch v0.8.0 https://github.com/pgvector/pgvector.git
+git clone --branch v0.8.1 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
@@ -29,24 +29,16 @@ make install # may need sudo
 
 See the [installation notes](#installation-notes---linux-and-mac) 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).
+You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), [pkg](#pkg), [APK](#apk), 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).
 
 ### Windows
 
-Ensure [C++ support in Visual Studio](https://learn.microsoft.com/en-us/cpp/build/building-on-the-command-line?view=msvc-170#download-and-install-the-tools) is installed, and run:
+Ensure [C++ support in Visual Studio](https://learn.microsoft.com/en-us/cpp/build/building-on-the-command-line?view=msvc-170#download-and-install-the-tools) is installed and run `x64 Native Tools Command Prompt for VS [version]` as administrator. Then use `nmake` to build:
 
 ```cmd
-call "C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvars64.bat"
+set "PGROOT=C:\Program Files\PostgreSQL\18"
-```
-
-Note: The exact path will vary depending on your Visual Studio version and edition
-
-Then use `nmake` to build:
-
-```cmd
-set "PGROOT=C:\Program Files\PostgreSQL\16"
 cd %TEMP%
-git clone --branch v0.8.0 https://github.com/pgvector/pgvector.git
+git clone --branch v0.8.1 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
@@ -500,9 +492,11 @@ With relaxed ordering, you can use a [materialized CTE](https://www.postgresql.o
 ```sql
 WITH relaxed_results AS MATERIALIZED (
     SELECT id, embedding <-> '[1,2,3]' AS distance FROM items WHERE category_id = 123 ORDER BY distance LIMIT 5
-) SELECT * FROM relaxed_results ORDER BY distance;
+) SELECT * FROM relaxed_results ORDER BY distance + 0;
 ```
 
+Note: `+ 0` is needed for Postgres 17+
+
 For queries that filter by distance, use a materialized CTE and place the distance filter outside of it for best performance (due to the [current behavior](https://www.postgresql.org/message-id/flat/CAOdR5yGUoMQ6j7M5hNUXrySzaqZVGf_Ne%2B8fwZMRKTFxU1nbJg%40mail.gmail.com) of the Postgres executor)
 
 ```sql
@@ -701,10 +695,10 @@ CREATE INDEX CONCURRENTLY ...
 
 ### Querying
 
-Use `EXPLAIN ANALYZE` to debug performance.
+Use `EXPLAIN (ANALYZE, BUFFERS)` to debug performance.
 
 ```sql
-EXPLAIN ANALYZE SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
+EXPLAIN (ANALYZE, BUFFERS) SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
 #### Exact Search
@@ -777,6 +771,8 @@ Use pgvector from any language with a Postgres client. You can even generate and
 
 Language | Libraries / Examples
 --- | ---
+Ada | [pgvector-ada](https://github.com/pgvector/pgvector-ada)
+Algol | [pgvector-algol](https://github.com/pgvector/pgvector-algol)
 C | [pgvector-c](https://github.com/pgvector/pgvector-c)
 C++ | [pgvector-cpp](https://github.com/pgvector/pgvector-cpp)
 C#, F#, Visual Basic | [pgvector-dotnet](https://github.com/pgvector/pgvector-dotnet)
@@ -791,15 +787,18 @@ Go | [pgvector-go](https://github.com/pgvector/pgvector-go)
 Haskell | [pgvector-haskell](https://github.com/pgvector/pgvector-haskell)
 Java, Kotlin, Groovy, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java)
 JavaScript, TypeScript | [pgvector-node](https://github.com/pgvector/pgvector-node)
-Julia | [pgvector-julia](https://github.com/pgvector/pgvector-julia)
+Julia | [Pgvector.jl](https://github.com/pgvector/Pgvector.jl)
 Lisp | [pgvector-lisp](https://github.com/pgvector/pgvector-lisp)
 Lua | [pgvector-lua](https://github.com/pgvector/pgvector-lua)
 Nim | [pgvector-nim](https://github.com/pgvector/pgvector-nim)
 OCaml | [pgvector-ocaml](https://github.com/pgvector/pgvector-ocaml)
+Pascal | [pgvector-pascal](https://github.com/pgvector/pgvector-pascal)
 Perl | [pgvector-perl](https://github.com/pgvector/pgvector-perl)
 PHP | [pgvector-php](https://github.com/pgvector/pgvector-php)
+Prolog | [pgvector-prolog](https://github.com/pgvector/pgvector-prolog)
 Python | [pgvector-python](https://github.com/pgvector/pgvector-python)
 R | [pgvector-r](https://github.com/pgvector/pgvector-r)
+Racket | [pgvector-racket](https://github.com/pgvector/pgvector-racket)
 Raku | [pgvector-raku](https://github.com/pgvector/pgvector-raku)
 Ruby | [pgvector-ruby](https://github.com/pgvector/pgvector-ruby), [Neighbor](https://github.com/ankane/neighbor)
 Rust | [pgvector-rust](https://github.com/pgvector/pgvector-rust)
@@ -818,7 +817,7 @@ Yes, pgvector uses the write-ahead log (WAL), which allows for replication and p
 
 #### What if I want to index vectors with more than 2,000 dimensions?
 
-You can use [half-precision indexing](#half-precision-indexing) to index up to 4,000 dimensions or [binary quantization](#binary-quantization) to index up to 64,000 dimensions. Another option is [dimensionality reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction).
+You can use [half-precision vectors](#half-precision-vectors) or [half-precision indexing](#half-precision-indexing) to index up to 4,000 dimensions or [binary quantization](#binary-quantization) to index up to 64,000 dimensions. Other options are [indexing subvectors](#indexing-subvectors) (for models that support it) or [dimensionality reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction).
 
 #### Can I store vectors with different dimensions in the same column?
 
@@ -1070,7 +1069,7 @@ 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/17/bin/pg_config
+export PG_CONFIG=/Library/PostgreSQL/18/bin/pg_config
 ```
 
 Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use:
@@ -1081,11 +1080,11 @@ sudo --preserve-env=PG_CONFIG make install
 
 A few common paths on Mac are:
 
-- EDB installer - `/Library/PostgreSQL/17/bin/pg_config`
+- EDB installer - `/Library/PostgreSQL/18/bin/pg_config`
-- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@17/bin/pg_config`
+- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@18/bin/pg_config`
-- Homebrew (x86-64) - `/usr/local/opt/postgresql@17/bin/pg_config`
+- Homebrew (x86-64) - `/usr/local/opt/postgresql@18/bin/pg_config`
 
-Note: Replace `17` with your Postgres server version
+Note: Replace `18` with your Postgres server version
 
 ### Missing Header
 
@@ -1094,10 +1093,10 @@ 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-17
+sudo apt install postgresql-server-dev-18
 ```
 
-Note: Replace `17` with your Postgres server version
+Note: Replace `18` with your Postgres server version
 
 ### Missing SDK
 
@@ -1127,7 +1126,7 @@ If compilation fails with `Cannot open include file: 'postgres.h': No such file
 
 ### Mismatched Architecture
 
-If compilation fails with `error C2196: case value '4' already used`, make sure `vcvars64.bat` was called. Then run `nmake /F Makefile.win clean` and re-run the installation instructions.
+If compilation fails with `error C2196: case value '4' already used`, make sure you’re using the `x64 Native Tools Command Prompt`. Then run `nmake /F Makefile.win clean` and re-run the installation instructions.
 
 ### Missing Symbol
 
@@ -1144,17 +1143,32 @@ 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:pg17
+docker pull pgvector/pgvector:pg18-trixie
 ```
 
-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).
+This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (replace `18` with your Postgres server version, and run it the same way).
+
+Supported tags are:
+
+- `pg18-trixie`, `0.8.1-pg18-trixie`
+- `pg18-bookworm`, `0.8.1-pg18-bookworm`, `pg18`, `0.8.1-pg18`
+- `pg17-trixie`, `0.8.1-pg17-trixie`
+- `pg17-bookworm`, `0.8.1-pg17-bookworm`, `pg17`, `0.8.1-pg17`
+- `pg16-trixie`, `0.8.1-pg16-trixie`
+- `pg16-bookworm`, `0.8.1-pg16-bookworm`, `pg16`, `0.8.1-pg16`
+- `pg15-trixie`, `0.8.1-pg15-trixie`
+- `pg15-bookworm`, `0.8.1-pg15-bookworm`, `pg15`, `0.8.1-pg15`
+- `pg14-trixie`, `0.8.1-pg14-trixie`
+- `pg14-bookworm`, `0.8.1-pg14-bookworm`, `pg14`, `0.8.1-pg14`
+- `pg13-trixie`, `0.8.1-pg13-trixie`
+- `pg13-bookworm`, `0.8.1-pg13-bookworm`, `pg13`, `0.8.1-pg13`
 
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.8.0 https://github.com/pgvector/pgvector.git
+git clone --branch v0.8.1 https://github.com/pgvector/pgvector.git
 cd pgvector
-docker build --pull --build-arg PG_MAJOR=17 -t myuser/pgvector .
+docker build --pull --build-arg PG_MAJOR=18 -t myuser/pgvector .
 ```
 
 If you increase `maintenance_work_mem`, make sure `--shm-size` is at least that size to avoid an error with parallel HNSW index builds.
@@ -1171,7 +1185,7 @@ With Homebrew Postgres, you can use:
 brew install pgvector
 ```
 
-Note: This only adds it to the `postgresql@17` and `postgresql@14` formulas
+Note: This only adds it to the `postgresql@18` and `postgresql@17` formulas
 
 ### PGXN
 
@@ -1186,29 +1200,29 @@ 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-17-pgvector
+sudo apt install postgresql-18-pgvector
 ```
 
-Note: Replace `17` with your Postgres server version
+Note: Replace `18` 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_17
+sudo yum install pgvector_18
 # or
-sudo dnf install pgvector_17
+sudo dnf install pgvector_18
 ```
 
-Note: Replace `17` with your Postgres server version
+Note: Replace `18` with your Postgres server version
 
 ### pkg
 
 Install the FreeBSD package with:
 
 ```sh
-pkg install postgresql16-pgvector
+pkg install postgresql17-pgvector
 ```
 
 or the port with:
@@ -1218,6 +1232,14 @@ cd /usr/ports/databases/pgvector
 make install
 ```
 
+### APK
+
+Install the Alpine package with:
+
+```sh
+apk add postgresql-pgvector
+```
+
 ### conda-forge
 
 With Conda Postgres, install from [conda-forge](https://anaconda.org/conda-forge/pgvector) with:
@@ -1260,6 +1282,7 @@ 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)
+- [Distance Adaptive Beam Search for Provably Accurate Graph-Based Nearest Neighbor Search](https://arxiv.org/pdf/2505.15636)
 
 ## History
 

sql/vector--0.8.0--0.8.1.sql

@@ -0,0 +1,2 @@
+-- complain if script is sourced in psql, rather than via CREATE EXTENSION
+\echo Use "ALTER EXTENSION vector UPDATE TO '0.8.1'" to load this file. \quit

src/halfvec.c

@@ -898,8 +898,21 @@ halfvec_binary_quantize(PG_FUNCTION_ARGS)
 	half	   *ax = a->x;
 	VarBit	   *result = InitBitVector(a->dim);
 	unsigned char *rx = VARBITS(result);
+	int			i = 0;
+	int			count = (a->dim / 8) * 8;
 
-	for (int i = 0; i < a->dim; i++)
+	/* Auto-vectorized on aarch64 */
+	for (; i < count; i += 8)
+	{
+		unsigned char result_byte = 0;
+
+		for (int j = 0; j < 8; j++)
+			result_byte |= (HalfToFloat4(ax[i + j]) > 0) << (7 - j);
+
+		rx[i / 8] = result_byte;
+	}
+
+	for (; i < a->dim; i++)
 		rx[i / 8] |= (HalfToFloat4(ax[i]) > 0) << (7 - (i % 8));
 
 	PG_RETURN_VARBIT_P(result);

src/hnsw.c

@@ -52,12 +52,20 @@ HnswInitLockTranche(void)
 								  sizeof(int) * 1,
 								  &found);
 	if (!found)
+	{
+#if PG_VERSION_NUM >= 190000
+		tranche_ids[0] = LWLockNewTrancheId("HnswBuild");
+#else
 		tranche_ids[0] = LWLockNewTrancheId();
+#endif
+	}
 	hnsw_lock_tranche_id = tranche_ids[0];
 	LWLockRelease(AddinShmemInitLock);
 
+#if PG_VERSION_NUM < 190000
 	/* Per-backend registration of the tranche ID */
 	LWLockRegisterTranche(hnsw_lock_tranche_id, "HnswBuild");
+#endif
 }
 
 /*
@@ -130,7 +138,7 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	Relation	index;
 
 	/* Never use index without order */
-	if (path->indexorderbys == NULL)
+	if (path->indexorderbys == NIL)
 	{
 		*indexStartupCost = get_float8_infinity();
 		*indexTotalCost = get_float8_infinity();

src/hnswbuild.c

@@ -54,6 +54,10 @@
 #include "utils/datum.h"
 #include "utils/memutils.h"
 
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
+
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_progress.h"
 #else
@@ -398,7 +402,7 @@ UpdateNeighborsInMemory(char *base, HnswSupport * support, HnswElement e, int m)
  * Update graph in memory
  */
 static void
-UpdateGraphInMemory(HnswSupport * support, HnswElement element, int m, int efConstruction, HnswElement entryPoint, HnswBuildState * buildstate)
+UpdateGraphInMemory(HnswSupport * support, HnswElement element, int m, HnswElement entryPoint, HnswBuildState * buildstate)
 {
 	HnswGraph  *graph = buildstate->graph;
 	char	   *base = buildstate->hnswarea;
@@ -460,7 +464,7 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 	HnswFindElementNeighbors(base, element, entryPoint, NULL, support, m, efConstruction, false);
 
 	/* Update graph in memory */
-	UpdateGraphInMemory(support, element, m, efConstruction, entryPoint, buildstate);
+	UpdateGraphInMemory(support, element, m, entryPoint, buildstate);
 
 	/* Release entry lock */
 	LWLockRelease(entryLock);
@@ -1054,7 +1058,7 @@ ComputeParallelWorkers(Relation heap, Relation index)
  * Build graph
  */
 static void
-BuildGraph(HnswBuildState * buildstate, ForkNumber forkNum)
+BuildGraph(HnswBuildState * buildstate)
 {
 	int			parallel_workers = 0;
 
@@ -1102,7 +1106,7 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
 
 	InitBuildState(buildstate, heap, index, indexInfo, forkNum);
 
-	BuildGraph(buildstate, forkNum);
+	BuildGraph(buildstate);
 
 	if (RelationNeedsWAL(index) || forkNum == INIT_FORKNUM)
 		log_newpage_range(index, forkNum, 0, RelationGetNumberOfBlocksInFork(index, forkNum), true);

src/hnswinsert.c

@@ -9,6 +9,10 @@
 #include "utils/datum.h"
 #include "utils/memutils.h"
 
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
+
 /*
  * Get the insert page
  */
@@ -660,7 +664,7 @@ FindDuplicateOnDisk(Relation index, HnswElement element, bool building)
  * Update graph on disk
  */
 static void
-UpdateGraphOnDisk(Relation index, HnswSupport * support, HnswElement element, int m, int efConstruction, HnswElement entryPoint, bool building)
+UpdateGraphOnDisk(Relation index, HnswSupport * support, HnswElement element, int m, HnswElement entryPoint, bool building)
 {
 	BlockNumber newInsertPage = InvalidBlockNumber;
 
@@ -728,7 +732,7 @@ HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPo
 	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false);
 
 	/* Update graph on disk */
-	UpdateGraphOnDisk(index, support, element, m, efConstruction, entryPoint, building);
+	UpdateGraphOnDisk(index, support, element, m, entryPoint, building);
 
 	/* Release lock */
 	UnlockPage(index, HNSW_UPDATE_LOCK, lockmode);

src/hnswscan.c

@@ -193,6 +193,10 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
 
 		/* Count index scan for stats */
 		pgstat_count_index_scan(scan->indexRelation);
+#if PG_VERSION_NUM >= 180000
+		if (scan->instrument)
+			scan->instrument->nsearches++;
+#endif
 
 		/* Safety check */
 		if (scan->orderByData == NULL)

src/hnswutils.c

@@ -15,6 +15,10 @@
 #include "utils/memdebug.h"
 #include "utils/rel.h"
 
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
+
 #if PG_VERSION_NUM < 170000
 static inline uint64
 murmurhash64(uint64 data)
@@ -805,12 +809,6 @@ HnswLoadUnvisitedFromDisk(HnswElement element, HnswUnvisited * unvisited, int *u
 		if (!found)
 			unvisited[(*unvisitedLength)++].indextid = *indextid;
 	}
-
-#ifdef USE_PREFETCH
-	/* TODO limit by get_tablespace_io_concurrency */
-	for (int i = 0; i < *unvisitedLength; i++)
-		PrefetchBuffer(index, MAIN_FORKNUM, ItemPointerGetBlockNumber(&unvisited[i].indextid));
-#endif
 }
 
 /*
@@ -831,6 +829,7 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 	HnswUnvisited *unvisited = palloc(lm * sizeof(HnswUnvisited));
 	int			unvisitedLength;
 	bool		inMemory = index == NULL;
+	double		lambda = 0.12;
 
 	if (v == NULL)
 	{
@@ -886,7 +885,7 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 		HnswSearchCandidate *f = HnswGetSearchCandidate(w_node, pairingheap_first(W));
 		HnswElement cElement;
 
-		if (c->distance > f->distance)
+		if (c->distance > f->distance * (1 + lambda))
 			break;
 
 		cElement = HnswPtrAccess(base, c->element);
@@ -906,8 +905,10 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 			HnswSearchCandidate *e;
 			double		eDistance;
 			bool		alwaysAdd = wlen < ef;
+			double		fDistance;
 
 			f = HnswGetSearchCandidate(w_node, pairingheap_first(W));
+			fDistance = f->distance * (1 + lambda);
 
 			if (inMemory)
 			{
@@ -922,13 +923,13 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 
 				/* Avoid any allocations if not adding */
 				eElement = NULL;
-				HnswLoadElementImpl(blkno, offno, &eDistance, q, index, support, inserting, alwaysAdd || discarded != NULL ? NULL : &f->distance, &eElement);
+				HnswLoadElementImpl(blkno, offno, &eDistance, q, index, support, inserting, alwaysAdd || discarded != NULL ? NULL : &fDistance, &eElement);
 
 				if (eElement == NULL)
 					continue;
 			}
 
-			if (eElement == NULL || !(eDistance < f->distance || alwaysAdd))
+			if (eElement == NULL || !(eDistance < fDistance || alwaysAdd))
 			{
 				if (discarded != NULL)
 				{

src/hnswvacuum.c

@@ -9,6 +9,10 @@
 #include "storage/lmgr.h"
 #include "utils/memutils.h"
 
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
+
 #if PG_VERSION_NUM >= 180000
 #define vacuum_delay_point() vacuum_delay_point(false)
 #endif

src/ivfbuild.c

@@ -20,6 +20,10 @@
 #include "utils/memutils.h"
 #include "vector.h"
 
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
+
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_progress.h"
 #else
@@ -138,7 +142,7 @@ SampleRows(IvfflatBuildState * buildstate)
  * Add tuple to sort
  */
 static void
-AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
+AddTupleToSort(ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
 {
 	double		distance;
 	double		minDistance = DBL_MAX;
@@ -215,7 +219,7 @@ BuildCallback(Relation index, ItemPointer tid, Datum *values,
 	oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
 
 	/* Add tuple to sort */
-	AddTupleToSort(index, tid, values, buildstate);
+	AddTupleToSort(tid, values, buildstate);
 
 	/* Reset memory context */
 	MemoryContextSwitchTo(oldCtx);
@@ -470,8 +474,8 @@ CreateMetaPage(Relation index, int dimensions, int lists, ForkNumber forkNum)
  * Create list pages
  */
 static void
-CreateListPages(Relation index, VectorArray centers, int dimensions,
+CreateListPages(Relation index, VectorArray centers, int lists,
-				int lists, ForkNumber forkNum, ListInfo * *listInfo)
+				ForkNumber forkNum, ListInfo * *listInfo)
 {
 	Buffer		buf;
 	Page		page;
@@ -1004,7 +1008,7 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
 
 	/* Create pages */
 	CreateMetaPage(index, buildstate->dimensions, buildstate->lists, forkNum);
-	CreateListPages(index, buildstate->centers, buildstate->dimensions, buildstate->lists, forkNum, &buildstate->listInfo);
+	CreateListPages(index, buildstate->centers, buildstate->lists, forkNum, &buildstate->listInfo);
 	CreateEntryPages(buildstate, forkNum);
 
 	/* Write WAL for initialization fork since GenericXLog functions do not */
@@ -1023,6 +1027,10 @@ ivfflatbuild(Relation heap, Relation index, IndexInfo *indexInfo)
 	IndexBuildResult *result;
 	IvfflatBuildState buildstate;
 
+#ifdef IVFFLAT_BENCH
+	SeedRandom(42);
+#endif
+
 	BuildIndex(heap, index, indexInfo, &buildstate, MAIN_FORKNUM);
 
 	result = (IndexBuildResult *) palloc(sizeof(IndexBuildResult));

src/ivfflat.c

@@ -92,7 +92,7 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	Relation	index;
 
 	/* Never use index without order */
-	if (path->indexorderbys == NULL)
+	if (path->indexorderbys == NIL)
 	{
 		*indexStartupCost = get_float8_infinity();
 		*indexTotalCost = get_float8_infinity();

src/ivfflat.h

@@ -13,6 +13,10 @@
 #include "utils/tuplesort.h"
 #include "vector.h"
 
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
+
 #if PG_VERSION_NUM >= 150000
 #include "common/pg_prng.h"
 #endif
@@ -73,9 +77,11 @@
 #if PG_VERSION_NUM >= 150000
 #define RandomDouble() pg_prng_double(&pg_global_prng_state)
 #define RandomInt() pg_prng_uint32(&pg_global_prng_state)
+#define SeedRandom(seed) pg_prng_seed(&pg_global_prng_state, seed)
 #else
 #define RandomDouble() (((double) random()) / MAX_RANDOM_VALUE)
 #define RandomInt() random()
+#define SeedRandom(seed) srandom(seed)
 #endif
 
 /* Variables */

src/ivfinsert.c

@@ -65,7 +65,7 @@ FindInsertPage(Relation index, Datum *values, BlockNumber *insertPage, ListInfo
  * Insert a tuple into the index
  */
 static void
-InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel)
+InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid)
 {
 	const		IvfflatTypeInfo *typeInfo = IvfflatGetTypeInfo(index);
 	IndexTuple	itup;
@@ -204,7 +204,7 @@ ivfflatinsert(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid,
 	oldCtx = MemoryContextSwitchTo(insertCtx);
 
 	/* Insert tuple */
-	InsertTuple(index, values, isnull, heap_tid, heap);
+	InsertTuple(index, values, isnull, heap_tid);
 
 	/* Delete memory context */
 	MemoryContextSwitchTo(oldCtx);

src/ivfkmeans.c

@@ -13,6 +13,10 @@
 #include "utils/memutils.h"
 #include "vector.h"
 
+#if PG_VERSION_NUM >= 160000
+#include "varatt.h"
+#endif
+
 /*
  * Initialize with kmeans++
  *

src/ivfscan.c

@@ -355,6 +355,10 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 
 		/* Count index scan for stats */
 		pgstat_count_index_scan(scan->indexRelation);
+#if PG_VERSION_NUM >= 180000
+		if (scan->instrument)
+			scan->instrument->nsearches++;
+#endif
 
 		/* Safety check */
 		if (scan->orderByData == NULL)

src/ivfutils.c

@@ -259,8 +259,8 @@ VectorUpdateCenter(Pointer v, int dimensions, float *x)
 	SET_VARSIZE(vec, VECTOR_SIZE(dimensions));
 	vec->dim = dimensions;
 
-	for (int k = 0; k < dimensions; k++)
+	for (int i = 0; i < dimensions; i++)
-		vec->x[k] = x[k];
+		vec->x[i] = x[i];
 }
 
 static void
@@ -271,8 +271,8 @@ HalfvecUpdateCenter(Pointer v, int dimensions, float *x)
 	SET_VARSIZE(vec, HALFVEC_SIZE(dimensions));
 	vec->dim = dimensions;
 
-	for (int k = 0; k < dimensions; k++)
+	for (int i = 0; i < dimensions; i++)
-		vec->x[k] = Float4ToHalfUnchecked(x[k]);
+		vec->x[i] = Float4ToHalfUnchecked(x[i]);
 }
 
 static void
@@ -284,29 +284,33 @@ BitUpdateCenter(Pointer v, int dimensions, float *x)
 	SET_VARSIZE(vec, VARBITTOTALLEN(dimensions));
 	VARBITLEN(vec) = dimensions;
 
-	for (uint32 k = 0; k < VARBITBYTES(vec); k++)
+	for (uint32 i = 0; i < VARBITBYTES(vec); i++)
-		nx[k] = 0;
+		nx[i] = 0;
 
-	for (int k = 0; k < dimensions; k++)
+	for (int i = 0; i < dimensions; i++)
-		nx[k / 8] |= (x[k] > 0.5 ? 1 : 0) << (7 - (k % 8));
+		nx[i / 8] |= (x[i] > 0.5 ? 1 : 0) << (7 - (i % 8));
 }
 
 static void
 VectorSumCenter(Pointer v, float *x)
 {
 	Vector	   *vec = (Vector *) v;
+	int			dim = vec->dim;
 
-	for (int k = 0; k < vec->dim; k++)
+	/* Auto-vectorized */
-		x[k] += vec->x[k];
+	for (int i = 0; i < dim; i++)
+		x[i] += vec->x[i];
 }
 
 static void
 HalfvecSumCenter(Pointer v, float *x)
 {
 	HalfVector *vec = (HalfVector *) v;
+	int			dim = vec->dim;
 
-	for (int k = 0; k < vec->dim; k++)
+	/* Auto-vectorized on aarch64 */
-		x[k] += HalfToFloat4(vec->x[k]);
+	for (int i = 0; i < dim; i++)
+		x[i] += HalfToFloat4(vec->x[i]);
 }
 
 static void
@@ -314,8 +318,8 @@ BitSumCenter(Pointer v, float *x)
 {
 	VarBit	   *vec = (VarBit *) v;
 
-	for (int k = 0; k < VARBITLEN(vec); k++)
+	for (int i = 0; i < VARBITLEN(vec); i++)
-		x[k] += (float) (((VARBITS(vec)[k / 8]) >> (7 - (k % 8))) & 0x01);
+		x[i] += (float) (((VARBITS(vec)[i / 8]) >> (7 - (i % 8))) & 0x01);
 }
 
 /*

src/vector.c

@@ -35,7 +35,11 @@
 #define VECTOR_TARGET_CLONES
 #endif
 
+#if PG_VERSION_NUM >= 180000
+PG_MODULE_MAGIC_EXT(.name = "vector",.version = "0.8.1");
+#else
 PG_MODULE_MAGIC;
+#endif
 
 /*
  * Initialize index options and variables
@@ -920,11 +924,13 @@ vector_concat(PG_FUNCTION_ARGS)
 	CheckDim(dim);
 	result = InitVector(dim);
 
-	for (int i = 0; i < a->dim; i++)
+	/* Auto-vectorized */
+	for (int i = 0, imax = a->dim; i < imax; i++)
 		result->x[i] = a->x[i];
 
-	for (int i = 0; i < b->dim; i++)
+	/* Auto-vectorized */
-		result->x[i + a->dim] = b->x[i];
+	for (int i = 0, imax = b->dim, start = a->dim; i < imax; i++)
+		result->x[i + start] = b->x[i];
 
 	PG_RETURN_POINTER(result);
 }
@@ -940,8 +946,21 @@ binary_quantize(PG_FUNCTION_ARGS)
 	float	   *ax = a->x;
 	VarBit	   *result = InitBitVector(a->dim);
 	unsigned char *rx = VARBITS(result);
+	int			i = 0;
+	int			count = (a->dim / 8) * 8;
 
-	for (int i = 0; i < a->dim; i++)
+	/* Auto-vectorized */
+	for (; i < count; i += 8)
+	{
+		unsigned char result_byte = 0;
+
+		for (int j = 0; j < 8; j++)
+			result_byte |= (ax[i + j] > 0) << (7 - j);
+
+		rx[i / 8] = result_byte;
+	}
+
+	for (; i < a->dim; i++)
 		rx[i / 8] |= (ax[i] > 0) << (7 - (i % 8));
 
 	PG_RETURN_VARBIT_P(result);

test/expected/halfvec.out

@@ -540,6 +540,12 @@ SELECT binary_quantize('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]'::halfvec)
  01001110101
 (1 row)
 
+SELECT binary_quantize('[1,2,3,-4,5,6,-7,8,1,-2,-3,4,5,-6,7,8,-1,2,3]'::halfvec);
+   binary_quantize   
+---------------------
+ 1110110110011011011
+(1 row)
+
 SELECT subvector('[1,2,3,4,5]'::halfvec, 1, 3);
  subvector 
 -----------

test/expected/vector_type.out

@@ -576,6 +576,12 @@ SELECT binary_quantize('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]'::vector);
  01001110101
 (1 row)
 
+SELECT binary_quantize('[1,2,3,-4,5,6,-7,8,1,-2,-3,4,5,-6,7,8,-1,2,3]'::vector);
+   binary_quantize   
+---------------------
+ 1110110110011011011
+(1 row)
+
 SELECT subvector('[1,2,3,4,5]'::vector, 1, 3);
  subvector 
 -----------

test/sql/halfvec.sql

@@ -121,6 +121,7 @@ SELECT l2_normalize('[65504]'::halfvec);
 
 SELECT binary_quantize('[1,0,-1]'::halfvec);
 SELECT binary_quantize('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]'::halfvec);
+SELECT binary_quantize('[1,2,3,-4,5,6,-7,8,1,-2,-3,4,5,-6,7,8,-1,2,3]'::halfvec);
 
 SELECT subvector('[1,2,3,4,5]'::halfvec, 1, 3);
 SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 2);

test/sql/vector_type.sql

@@ -128,6 +128,7 @@ SELECT l2_normalize('[3e38]'::vector);
 
 SELECT binary_quantize('[1,0,-1]'::vector);
 SELECT binary_quantize('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]'::vector);
+SELECT binary_quantize('[1,2,3,-4,5,6,-7,8,1,-2,-3,4,5,-6,7,8,-1,2,3]'::vector);
 
 SELECT subvector('[1,2,3,4,5]'::vector, 1, 3);
 SELECT subvector('[1,2,3,4,5]'::vector, 3, 2);

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat and hnsw access methods'
-default_version = '0.8.0'
+default_version = '0.8.1'
 module_pathname = '$libdir/vector'
 relocatable = true