Compare commits

..

41 Commits

Author SHA1 Message Date
Andrew Kane
f6058beeda Started support for parallel index scan [skip ci] 2025-09-22 11:21:14 -07:00
Andrew Kane
9e29dff78c Updated CI [skip ci] 2025-09-19 19:16:08 -07:00
Andrew Kane
13c0dbf530 Test with macos-15-intel on CI [skip ci] 2025-09-19 18:34:51 -07:00
Andrew Kane
c205a12107 Updated CI [skip ci] 2025-09-05 17:03:37 -07:00
Andrew Kane
db3755a58f Added windows-2025 to CI [skip ci] 2025-09-05 16:41:22 -07:00
Andrew Kane
778dacf20c Version bump to 0.8.1 [skip ci] 2025-09-04 17:51:09 -07:00
Andrew Kane
3f687687ee Fixed compilation error with Postgres 19 2025-09-04 15:58:23 -07:00
Andrew Kane
815f48e489 Updated changelog [skip ci] 2025-09-04 15:36:18 -07:00
Andrew Kane
67e648b13e Added another test for binary_quantize function [skip ci] 2025-08-29 01:09:18 -07:00
Andrew Kane
c3ff955231 Reordered supported Docker tags [skip ci] 2025-08-23 19:57:14 -07:00
Andrew Kane
6f46a1035d Added new Docker images to readme [skip ci] 2025-08-23 19:46:07 -07:00
Andrew Kane
bbe66e821b Added Docker images for Debian 13 / Trixie [skip ci] 2025-08-23 19:34:11 -07:00
Julien Rouhaud
dd3a1e9137 Use NIL for empty lists (#890)
Postgres standard way to check for list emptiness is to compare a pointer to
NIL rather than NULL.
2025-08-23 03:31:22 -07:00
Andrew Kane
ea4746f6c0 Updated comment [skip ci] 2025-08-19 14:26:36 -07:00
Andrew Kane
6aec80ccdb Improved performance of binary_quantize function for halfvec 2025-08-19 13:47:11 -07:00
Luca Giacchino
0c9070ba82 Add autovectorized implementation of binary quantize 2025-08-19 13:18:50 -07:00
Andrew Kane
30382418da Ran pgindent [skip ci] 2025-08-19 12:44:54 -07:00
Andrew Kane
26b50e536f Updated FreeBSD package name in readme [skip ci] 2025-08-09 12:09:29 -07:00
Andrew Kane
e29fc3aa1a Switched to PG_MODULE_MAGIC_EXT for Postgres 18+ [skip ci] 2025-07-31 19:59:27 -07:00
Andrew Kane
6ef7fccb5c Added Postgres 19 to CI [skip ci] 2025-07-27 19:50:46 -07:00
Andrew Kane
5b8b68ba1d Use consistent style [skip ci] 2025-07-27 18:19:28 -07:00
Andrew Kane
3be8693c13 Added supported Docker tags to readme [skip ci] 2025-07-27 13:44:59 -07:00
Andrew Kane
247c8dc8a5 Updated changelog [skip ci] 2025-07-27 13:25:51 -07:00
Andrew Kane
3600ab93e5 Added Docker images with -bookworm suffix [skip ci] 2025-07-27 13:14:29 -07:00
Andrew Kane
665db75a3c Updated Dockerfile to use release [skip ci] 2025-07-27 12:34:23 -07:00
Andrew Kane
44163d0a97 Revert "Added OS to Dockerfile [skip ci]"
This reverts commit 33ca8a61e2.
2025-07-09 22:01:34 -07:00
Andrew Kane
33ca8a61e2 Added OS to Dockerfile [skip ci] 2025-07-09 16:25:05 -07:00
Andrew Kane
742e2d1d28 Synced .dockerignore with .gitignore [skip ci] 2025-07-09 16:22:43 -07:00
Andrew Kane
a7c49d8283 Updated readme [skip ci] 2025-07-01 11:13:29 -07:00
Andrew Kane
ae9ee81e4d Fixed relaxed results example for Postgres 17 - #862 [skip ci] 2025-07-01 10:11:50 -07:00
Andrew Kane
fa1dee4e3b Updated link [skip ci] 2025-07-01 03:35:26 -07:00
Andrew Kane
e6bad96a03 Ran pgindent [skip ci] 2025-06-18 20:07:46 -07:00
Jon Daniel
3a49d141b3 Vectorizing vector_concat for improved performance (#861)
* Vectorizing vector_concat for improved performance

On an ARM chip this should generate SIMD instructions to copy the two
incoming vectors to the new vector as opposed to doing it all in
software.

* Moving declarations to above CheckDim

* Removing const from dims

* Formatting
2025-06-18 20:06:32 -07:00
Andrew Kane
ce09c9a27a Improved variable names [skip ci] 2025-06-18 17:32:37 -07:00
Andrew Kane
870ca6724d Fixed CI [skip ci] 2025-06-18 16:26:30 -07:00
Andrew Kane
8ad680f009 Ran pgindent [skip ci] 2025-06-18 16:11:15 -07:00
Jon Daniel
fe697e8788 vectorize: optimize VectorSumCenter and HalfvecSumCenter (#860)
* vectorize: optimize VectorSumCenter and HalfvecSumCenter

The functions VectorSumCenter and HalfvecSumCenter were not being
vectorized by the compiler. A few slight changes will allow these
optimizations to take place and get a performance boost by utilizing
SIMD instructions.

This optimization helps improve performance of vector operations in IVF
index building and updating.

* Removing const, commenting that it is only vectoirzed on ARM
2025-06-18 16:09:43 -07:00
Andrew Kane
bf28ed8176 Set random seed for IVFFLAT_BENCH [skip ci] 2025-06-18 10:57:18 -07:00
Andrew Kane
799cfebf70 Updated readme [skip ci] 2025-06-04 14:29:53 -07:00
Andrew Kane
3cd1f09f66 Updated Windows installation instructions - #593 [skip ci] 2025-06-04 14:18:53 -07:00
Andrew Kane
e2efe62fe5 Updated readme [skip ci] 2025-05-12 13:11:43 -07:00
24 changed files with 202 additions and 75 deletions

View File

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

View File

@@ -8,6 +8,8 @@ jobs:
fail-fast: false fail-fast: false
matrix: matrix:
include: include:
- postgres: 19
os: ubuntu-24.04
- postgres: 18 - postgres: 18
os: ubuntu-24.04 os: ubuntu-24.04
- postgres: 17 - postgres: 17
@@ -28,7 +30,7 @@ jobs:
dev-files: true dev-files: true
- run: make - run: make
env: env:
PG_CFLAGS: ${{ matrix.postgres == 18 && '-Wno-missing-field-initializers' || '' }} -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare PG_CFLAGS: ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }} -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
- run: | - run: |
export PG_CONFIG=`which pg_config` export PG_CONFIG=`which pg_config`
sudo --preserve-env=PG_CONFIG make install sudo --preserve-env=PG_CONFIG make install
@@ -47,9 +49,9 @@ jobs:
matrix: matrix:
include: include:
- postgres: 17 - postgres: 17
os: macos-15 os: macos-26
- postgres: 14 - postgres: 14
os: macos-13 os: macos-15-intel
steps: steps:
- uses: actions/checkout@v4 - uses: actions/checkout@v4
- uses: ankane/setup-postgres@v1 - uses: ankane/setup-postgres@v1
@@ -57,7 +59,7 @@ jobs:
postgres-version: ${{ matrix.postgres }} postgres-version: ${{ matrix.postgres }}
- run: make - run: make
env: 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
- run: make install - run: make install
- run: make installcheck - run: make installcheck
- if: ${{ failure() }} - if: ${{ failure() }}
@@ -70,27 +72,35 @@ jobs:
tar xf $TAG.tar.gz tar xf $TAG.tar.gz
mv postgres-$TAG postgres mv postgres-$TAG postgres
env: env:
TAG: ${{ matrix.postgres == 17 && 'REL_17_2' || 'REL_14_15' }} TAG: ${{ matrix.postgres == 17 && 'REL_17_6' || 'REL_14_19' }}
- run: make prove_installcheck PROVE_FLAGS="-I ./postgres/src/test/perl -I ./test/perl" - run: make prove_installcheck PROVE_FLAGS="-I ./postgres/src/test/perl -I ./test/perl"
env: env:
PERL5LIB: /Users/runner/perl5/lib/perl5 PERL5LIB: /Users/runner/perl5/lib/perl5
- run: make clean && $(brew --prefix llvm@$LLVM_VERSION)/bin/scan-build --status-bugs make - run: make clean && $(brew --prefix llvm@$LLVM_VERSION)/bin/scan-build --status-bugs make
env: env:
LLVM_VERSION: ${{ matrix.os == 'macos-15' && 18 || 15 }} LLVM_VERSION: ${{ matrix.os == 'macos-26' && 20 || 18 }}
PG_CFLAGS: -DUSE_ASSERT_CHECKING PG_CFLAGS: -DUSE_ASSERT_CHECKING
windows: windows:
runs-on: windows-latest runs-on: ${{ matrix.os }}
if: ${{ !startsWith(github.ref_name, 'mac') }} if: ${{ !startsWith(github.ref_name, 'mac') }}
strategy:
fail-fast: false
matrix:
include:
- postgres: 17
os: windows-2025
- postgres: 14
os: windows-2022
steps: steps:
- uses: actions/checkout@v4 - uses: actions/checkout@v4
- uses: ankane/setup-postgres@v1 - uses: ankane/setup-postgres@v1
with: with:
postgres-version: 14 postgres-version: ${{ matrix.postgres }}
- run: | - run: |
call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^ call "C:\Program Files\Microsoft Visual Studio\2022\Enterprise\VC\Auxiliary\Build\vcvars64.bat" && ^
nmake /NOLOGO /F Makefile.win && ^ nmake /NOLOGO /F Makefile.win && ^
nmake /NOLOGO /F Makefile.win install && ^ 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 clean && ^
nmake /NOLOGO /F Makefile.win uninstall nmake /NOLOGO /F Makefile.win uninstall
shell: cmd shell: cmd

View File

@@ -1,3 +1,8 @@
## 0.8.1 (2025-09-04)
- Added support for Postgres 18 rc1
- Improved performance of `binary_quantize` function
## 0.8.0 (2024-10-30) ## 0.8.0 (2024-10-30)
- Added support for iterative index scans - Added support for iterative index scans

View File

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

View File

@@ -2,7 +2,7 @@
"name": "vector", "name": "vector",
"abstract": "Open-source vector similarity search for Postgres", "abstract": "Open-source vector similarity search for Postgres",
"description": "Supports L2 distance, inner product, and cosine distance", "description": "Supports L2 distance, inner product, and cosine distance",
"version": "0.8.0", "version": "0.8.1",
"maintainer": [ "maintainer": [
"Andrew Kane <andrew@ankane.org>" "Andrew Kane <andrew@ankane.org>"
], ],
@@ -20,7 +20,7 @@
"vector": { "vector": {
"file": "sql/vector.sql", "file": "sql/vector.sql",
"docfile": "README.md", "docfile": "README.md",
"version": "0.8.0", "version": "0.8.1",
"abstract": "Open-source vector similarity search for Postgres" "abstract": "Open-source vector similarity search for Postgres"
} }
}, },

View File

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

View File

@@ -1,5 +1,5 @@
EXTENSION = vector EXTENSION = vector
EXTVERSION = 0.8.0 EXTVERSION = 0.8.1
DATA_built = sql\$(EXTENSION)--$(EXTVERSION).sql DATA_built = sql\$(EXTENSION)--$(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 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 PKGLIBDIR = $(PGROOT)\lib
SHAREDIR = $(PGROOT)\share 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"$(INCLUDEDIR)" CFLAGS = /nologo /I"$(INCLUDEDIR_SERVER)\port\win32_msvc" /I"$(INCLUDEDIR_SERVER)\port\win32" /I"$(INCLUDEDIR_SERVER)" /I"$(INCLUDEDIR)"
CFLAGS = $(CFLAGS) $(PG_CFLAGS) CFLAGS = $(CFLAGS) $(PG_CFLAGS)
@@ -58,7 +61,7 @@ install: all
for %f in ($(HEADERS)) do copy %f "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)" for %f in ($(HEADERS)) do copy %f "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
installcheck: installcheck:
"$(BINDIR)\pg_regress" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS) "$(PG_REGRESS)" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS)
uninstall: uninstall:
del /f "$(PKGLIBDIR)\$(SHLIB)" del /f "$(PKGLIBDIR)\$(SHLIB)"

View File

@@ -21,7 +21,7 @@ Compile and install the extension (supports Postgres 13+)
```sh ```sh
cd /tmp 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 cd pgvector
make make
make install # may need sudo make install # may need sudo
@@ -33,20 +33,12 @@ You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#p
### Windows ### 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 ```cmd
call "C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvars64.bat" set "PGROOT=C:\Program Files\PostgreSQL\17"
```
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% 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 cd pgvector
nmake /F Makefile.win nmake /F Makefile.win
nmake /F Makefile.win install nmake /F Makefile.win install
@@ -500,9 +492,11 @@ With relaxed ordering, you can use a [materialized CTE](https://www.postgresql.o
```sql ```sql
WITH relaxed_results AS MATERIALIZED ( 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 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) 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 ```sql
@@ -791,7 +785,7 @@ Go | [pgvector-go](https://github.com/pgvector/pgvector-go)
Haskell | [pgvector-haskell](https://github.com/pgvector/pgvector-haskell) Haskell | [pgvector-haskell](https://github.com/pgvector/pgvector-haskell)
Java, Kotlin, Groovy, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java) Java, Kotlin, Groovy, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java)
JavaScript, TypeScript | [pgvector-node](https://github.com/pgvector/pgvector-node) 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) Lisp | [pgvector-lisp](https://github.com/pgvector/pgvector-lisp)
Lua | [pgvector-lua](https://github.com/pgvector/pgvector-lua) Lua | [pgvector-lua](https://github.com/pgvector/pgvector-lua)
Nim | [pgvector-nim](https://github.com/pgvector/pgvector-nim) Nim | [pgvector-nim](https://github.com/pgvector/pgvector-nim)
@@ -1127,7 +1121,7 @@ If compilation fails with `Cannot open include file: 'postgres.h': No such file
### Mismatched Architecture ### 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 youre using the `x64 Native Tools Command Prompt`. Then run `nmake /F Makefile.win clean` and re-run the installation instructions.
### Missing Symbol ### Missing Symbol
@@ -1144,15 +1138,28 @@ If installation fails with `Access is denied`, re-run the installation instructi
Get the [Docker image](https://hub.docker.com/r/pgvector/pgvector) with: Get the [Docker image](https://hub.docker.com/r/pgvector/pgvector) with:
```sh ```sh
docker pull pgvector/pgvector:pg17 docker pull pgvector/pgvector:pg17-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 `17` with your Postgres server version, and run it the same way).
Supported tags are:
- `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: You can also build the image manually:
```sh ```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 cd pgvector
docker build --pull --build-arg PG_MAJOR=17 -t myuser/pgvector . docker build --pull --build-arg PG_MAJOR=17 -t myuser/pgvector .
``` ```
@@ -1208,7 +1215,7 @@ Note: Replace `17` with your Postgres server version
Install the FreeBSD package with: Install the FreeBSD package with:
```sh ```sh
pkg install postgresql16-pgvector pkg install postgresql17-pgvector
``` ```
or the port with: or the port with:

View File

@@ -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

View File

@@ -898,8 +898,21 @@ halfvec_binary_quantize(PG_FUNCTION_ARGS)
half *ax = a->x; half *ax = a->x;
VarBit *result = InitBitVector(a->dim); VarBit *result = InitBitVector(a->dim);
unsigned char *rx = VARBITS(result); 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)); rx[i / 8] |= (HalfToFloat4(ax[i]) > 0) << (7 - (i % 8));
PG_RETURN_VARBIT_P(result); PG_RETURN_VARBIT_P(result);

View File

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

View File

@@ -805,12 +805,6 @@ HnswLoadUnvisitedFromDisk(HnswElement element, HnswUnvisited * unvisited, int *u
if (!found) if (!found)
unvisited[(*unvisitedLength)++].indextid = *indextid; 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
} }
/* /*

View File

@@ -1023,6 +1023,10 @@ ivfflatbuild(Relation heap, Relation index, IndexInfo *indexInfo)
IndexBuildResult *result; IndexBuildResult *result;
IvfflatBuildState buildstate; IvfflatBuildState buildstate;
#ifdef IVFFLAT_BENCH
SeedRandom(42);
#endif
BuildIndex(heap, index, indexInfo, &buildstate, MAIN_FORKNUM); BuildIndex(heap, index, indexInfo, &buildstate, MAIN_FORKNUM);
result = (IndexBuildResult *) palloc(sizeof(IndexBuildResult)); result = (IndexBuildResult *) palloc(sizeof(IndexBuildResult));

View File

@@ -92,7 +92,7 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
Relation index; Relation index;
/* Never use index without order */ /* Never use index without order */
if (path->indexorderbys == NULL) if (path->indexorderbys == NIL)
{ {
*indexStartupCost = get_float8_infinity(); *indexStartupCost = get_float8_infinity();
*indexTotalCost = get_float8_infinity(); *indexTotalCost = get_float8_infinity();
@@ -143,6 +143,10 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
*indexSelectivity = costs.indexSelectivity; *indexSelectivity = costs.indexSelectivity;
*indexCorrelation = costs.indexCorrelation; *indexCorrelation = costs.indexCorrelation;
*indexPages = costs.numIndexPages; *indexPages = costs.numIndexPages;
elog(INFO, "ivfflatcostestimate = %f", costs.indexTotalCost);
/* Cost estimates for parallel workers applied outside of amcostestimate */
elog(INFO, "parallel_workers = %d, parallel aware = %d", path->path.parallel_workers, path->path.parallel_aware);
} }
/* /*
@@ -170,6 +174,25 @@ ivfflatvalidate(Oid opclassoid)
return true; return true;
} }
static Size
ivfflatestimateparallelscan()
{
elog(INFO, "ivfflatestimateparallelscan");
return 0;
}
static void
ivfflatinitparallelscan(void *target)
{
elog(INFO, "ivfflatinitparallelscan");
}
static void
ivfflatparallelrescan(IndexScanDesc scan)
{
elog(INFO, "ivfflatparallelrescan");
}
/* /*
* Define index handler * Define index handler
* *
@@ -200,7 +223,7 @@ ivfflathandler(PG_FUNCTION_ARGS)
amroutine->amstorage = false; amroutine->amstorage = false;
amroutine->amclusterable = false; amroutine->amclusterable = false;
amroutine->ampredlocks = false; amroutine->ampredlocks = false;
amroutine->amcanparallel = false; amroutine->amcanparallel = true;
#if PG_VERSION_NUM >= 170000 #if PG_VERSION_NUM >= 170000
amroutine->amcanbuildparallel = true; amroutine->amcanbuildparallel = true;
#endif #endif
@@ -242,9 +265,9 @@ ivfflathandler(PG_FUNCTION_ARGS)
amroutine->amrestrpos = NULL; amroutine->amrestrpos = NULL;
/* Interface functions to support parallel index scans */ /* Interface functions to support parallel index scans */
amroutine->amestimateparallelscan = NULL; amroutine->amestimateparallelscan = ivfflatestimateparallelscan;
amroutine->aminitparallelscan = NULL; amroutine->aminitparallelscan = ivfflatinitparallelscan;
amroutine->amparallelrescan = NULL; amroutine->amparallelrescan = ivfflatparallelrescan;
#if PG_VERSION_NUM >= 180000 #if PG_VERSION_NUM >= 180000
amroutine->amtranslatestrategy = NULL; amroutine->amtranslatestrategy = NULL;

View File

@@ -73,9 +73,11 @@
#if PG_VERSION_NUM >= 150000 #if PG_VERSION_NUM >= 150000
#define RandomDouble() pg_prng_double(&pg_global_prng_state) #define RandomDouble() pg_prng_double(&pg_global_prng_state)
#define RandomInt() pg_prng_uint32(&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 #else
#define RandomDouble() (((double) random()) / MAX_RANDOM_VALUE) #define RandomDouble() (((double) random()) / MAX_RANDOM_VALUE)
#define RandomInt() random() #define RandomInt() random()
#define SeedRandom(seed) srandom(seed)
#endif #endif
/* Variables */ /* Variables */

View File

@@ -119,6 +119,9 @@ GetScanItems(IndexScanDesc scan, Datum value)
tuplesort_reset(so->sortstate); tuplesort_reset(so->sortstate);
if (scan->parallel_scan != NULL)
elog(INFO, "parallel scan");
/* Search closest probes lists */ /* Search closest probes lists */
while (so->listIndex < so->maxProbes && (++batchProbes) <= so->probes) while (so->listIndex < so->maxProbes && (++batchProbes) <= so->probes)
{ {

View File

@@ -259,8 +259,8 @@ VectorUpdateCenter(Pointer v, int dimensions, float *x)
SET_VARSIZE(vec, VECTOR_SIZE(dimensions)); SET_VARSIZE(vec, VECTOR_SIZE(dimensions));
vec->dim = 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 static void
@@ -271,8 +271,8 @@ HalfvecUpdateCenter(Pointer v, int dimensions, float *x)
SET_VARSIZE(vec, HALFVEC_SIZE(dimensions)); SET_VARSIZE(vec, HALFVEC_SIZE(dimensions));
vec->dim = 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 static void
@@ -284,29 +284,33 @@ BitUpdateCenter(Pointer v, int dimensions, float *x)
SET_VARSIZE(vec, VARBITTOTALLEN(dimensions)); SET_VARSIZE(vec, VARBITTOTALLEN(dimensions));
VARBITLEN(vec) = 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 static void
VectorSumCenter(Pointer v, float *x) VectorSumCenter(Pointer v, float *x)
{ {
Vector *vec = (Vector *) v; 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 static void
HalfvecSumCenter(Pointer v, float *x) HalfvecSumCenter(Pointer v, float *x)
{ {
HalfVector *vec = (HalfVector *) v; 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 static void
@@ -314,8 +318,8 @@ BitSumCenter(Pointer v, float *x)
{ {
VarBit *vec = (VarBit *) v; 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);
} }
/* /*

View File

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

View File

@@ -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 01001110101
(1 row) (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); SELECT subvector('[1,2,3,4,5]'::halfvec, 1, 3);
subvector subvector
----------- -----------

View File

@@ -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 01001110101
(1 row) (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); SELECT subvector('[1,2,3,4,5]'::vector, 1, 3);
subvector subvector
----------- -----------

View File

@@ -121,6 +121,7 @@ SELECT l2_normalize('[65504]'::halfvec);
SELECT binary_quantize('[1,0,-1]'::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('[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, 1, 3);
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 2); SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 2);

View File

@@ -0,0 +1,16 @@
-- SET force_parallel_mode = on;
SET parallel_setup_cost = 10;
SET parallel_tuple_cost = 0.000001;
SET min_parallel_table_scan_size = 1;
SET min_parallel_index_scan_size = 1;
CREATE TABLE t (id integer, val vector(3));
ALTER TABLE t ALTER COLUMN val SET STORAGE PLAIN;
INSERT INTO t (id, val) SELECT n, ARRAY[random(), random(), random()] FROM generate_series(1,1000000) n;
CREATE INDEX ON t USING ivfflat (val) WITH (lists = 10);
SET ivfflat.probes = 4;
EXPLAIN SELECT * FROM t ORDER BY val <-> '[0.5,0.5,0.5]' LIMIT 5;
SELECT * FROM t ORDER BY val <-> '[0.5,0.5,0.5]' LIMIT 5;
DROP TABLE t;

View File

@@ -128,6 +128,7 @@ SELECT l2_normalize('[3e38]'::vector);
SELECT binary_quantize('[1,0,-1]'::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('[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, 1, 3);
SELECT subvector('[1,2,3,4,5]'::vector, 3, 2); SELECT subvector('[1,2,3,4,5]'::vector, 3, 2);

View File

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