Compare commits

..

4 Commits

Author SHA1 Message Date
Andrew Kane
95bcda94da No need to set heap id [skip ci] 2023-10-22 16:15:34 -07:00
Andrew Kane
84c5aa14bf Added comment [skip ci] 2023-10-22 16:11:03 -07:00
Andrew Kane
72e741a5a6 Fixed issue with freeing 2023-10-22 16:07:56 -07:00
Andrew Kane
84ca6625a7 Added support for index-only scans to HNSW 2023-10-22 15:58:40 -07:00
152 changed files with 3017 additions and 16897 deletions

8
.dockerignore Normal file
View File

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

View File

@@ -8,29 +8,29 @@ jobs:
fail-fast: false
matrix:
include:
- postgres: 19
os: ubuntu-24.04
- postgres: 18
os: ubuntu-24.04
- postgres: 17
os: ubuntu-24.04
os: ubuntu-22.04
- postgres: 16
os: ubuntu-24.04-arm
os: ubuntu-22.04
- postgres: 15
os: ubuntu-22.04
- postgres: 14
os: ubuntu-22.04-arm
- postgres: 13
os: ubuntu-22.04
- postgres: 13
os: ubuntu-20.04
- postgres: 12
os: ubuntu-20.04
- postgres: 11
os: ubuntu-20.04
steps:
- uses: actions/checkout@v6
- uses: actions/checkout@v4
- uses: ankane/setup-postgres@v1
with:
postgres-version: ${{ matrix.postgres }}
dev-files: true
- run: make
env:
PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }}
PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare
- run: |
export PG_CONFIG=`which pg_config`
sudo --preserve-env=PG_CONFIG make install
@@ -42,78 +42,51 @@ jobs:
sudo apt-get install libipc-run-perl
- run: make prove_installcheck
mac:
runs-on: ${{ matrix.os }}
runs-on: macos-latest
if: ${{ !startsWith(github.ref_name, 'windows') }}
strategy:
fail-fast: false
matrix:
include:
- postgres: 18
os: macos-26
- postgres: 14
os: macos-15-intel
steps:
- uses: actions/checkout@v6
- uses: actions/checkout@v4
- uses: ankane/setup-postgres@v1
with:
postgres-version: ${{ matrix.postgres }}
postgres-version: 14
- run: make
env:
PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-unknown-warning-option ${{ matrix.postgres >= 18 && '-Wno-missing-field-initializers' || '' }}
PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter
- run: make install
- run: make installcheck
- if: ${{ failure() }}
run: cat regression.diffs
# Homebrew Postgres does not enable TAP tests, so need to download
- run: |
brew install cpanm
cpanm --notest IPC::Run
wget -q https://github.com/postgres/postgres/archive/refs/tags/$TAG.tar.gz
tar xf $TAG.tar.gz
mv postgres-$TAG postgres
env:
TAG: ${{ matrix.postgres == 18 && 'REL_18_2' || 'REL_14_21' }}
- 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-26' && 20 || 18 }}
PG_CFLAGS: -DUSE_ASSERT_CHECKING
wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_14_5.tar.gz
tar xf REL_14_5.tar.gz
- run: make prove_installcheck PROVE_FLAGS="-I ./postgres-REL_14_5/src/test/perl" PERL5LIB="/Users/runner/perl5/lib/perl5"
- run: make clean && /usr/local/opt/llvm@15/bin/scan-build --status-bugs make
windows:
runs-on: ${{ matrix.os }}
runs-on: windows-latest
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@v6
- uses: actions/checkout@v4
- uses: ankane/setup-postgres@v1
with:
postgres-version: ${{ matrix.postgres }}
postgres-version: 14
- run: |
call "C:\Program Files\Microsoft Visual Studio\${{ matrix.os == 'windows-2025' && 18 || 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 install && ^
nmake /NOLOGO /F Makefile.win installcheck ${{ matrix.postgres != 17 && 'PG_REGRESS=$(PGROOT)\bin\pg_regress' || '' }} && ^
nmake /NOLOGO /F Makefile.win installcheck && ^
nmake /NOLOGO /F Makefile.win clean && ^
nmake /NOLOGO /F Makefile.win uninstall
shell: cmd
- if: ${{ failure() }}
run: cat regression.diffs
i386:
if: ${{ !startsWith(github.ref_name, 'mac') && !startsWith(github.ref_name, 'windows') }}
runs-on: ubuntu-latest
container:
image: debian:12
image: debian:11
options: --platform linux/386
steps:
- run: apt-get update && apt-get install -y build-essential git libipc-run-perl postgresql-15 postgresql-server-dev-15 sudo
- run: apt-get update && apt-get install -y build-essential git libipc-run-perl postgresql-13 postgresql-server-dev-13 sudo
- run: service postgresql start
- run: |
git clone https://github.com/${{ github.repository }}.git pgvector
@@ -126,18 +99,4 @@ jobs:
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
- if: ${{ failure() }}
run: cat pgvector/regression.diffs
valgrind:
if: ${{ !startsWith(github.ref_name, 'mac') && !startsWith(github.ref_name, 'windows') }}
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v6
- uses: ankane/setup-postgres-valgrind@v1
with:
postgres-version: 18
check-ub: yes
- run: make OPTFLAGS=""
- run: sudo --preserve-env=PG_CONFIG make install
- run: make installcheck
PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter -Wno-sign-compare

View File

@@ -1,91 +1,3 @@
## 0.8.3 (2026-06-17)
- Fixed possible index corruption with HNSW vacuuming
- Fixed performance regression with Hamming distance and Jaccard distance with Postgres 18
## 0.8.2 (2026-02-25)
- Fixed buffer overflow with parallel HNSW index build - [more info](https://github.com/pgvector/pgvector/issues/959)
- 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
- Added casts for arrays to `sparsevec`
- Improved cost estimation for better index selection when filtering
- Improved performance of HNSW index scans
- Improved performance of HNSW inserts and on-disk index builds
- Dropped support for Postgres 12
## 0.7.4 (2024-08-05)
- Fixed locking for parallel HNSW index builds
- Fixed compilation error with GCC 14 on i386 when SSE2 is not enabled
## 0.7.3 (2024-07-22)
- Fixed `failed to add index item` error with `sparsevec`
- Fixed compilation error with FreeBSD ARM
- Fixed compilation warning with MSVC and Postgres 16
## 0.7.2 (2024-06-11)
- Fixed initialization fork for indexes on unlogged tables
## 0.7.1 (2024-06-03)
- Improved performance of on-disk HNSW index builds
- Fixed `undefined symbol` error with GCC 8
- Fixed compilation error with universal binaries on Mac
- Fixed compilation warning with Clang < 14
## 0.7.0 (2024-04-29)
- Added `halfvec` type
- Added `sparsevec` type
- Added support for indexing `bit` type
- Added support for indexing L1 distance with HNSW
- Added `binary_quantize` function
- Added `hamming_distance` function
- Added `jaccard_distance` function
- Added `l2_normalize` function
- Added `subvector` function
- Added concatenate operator for vectors
- Added CPU dispatching for distance functions on Linux x86-64
- Updated comparison operators to support vectors with different dimensions
## 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)
If upgrading with Postgres 12 or Docker, see [these notes](https://github.com/pgvector/pgvector#060).
- Added support for parallel index builds for HNSW
- Added validation for GUC parameters
- Changed storage for vector from `extended` to `external`
- Improved performance of HNSW
- Reduced memory usage for HNSW index builds
- Reduced WAL generation for HNSW index builds
- Fixed error with logical replication
- Fixed `invalid memory alloc request size` error with HNSW index builds
- Moved Docker image to `pgvector` org
- Added Docker tags for each supported version of Postgres
- Dropped support for Postgres 11
## 0.5.1 (2023-10-10)
- Improved performance of HNSW index builds
@@ -132,7 +44,7 @@ If upgrading with Postgres 12 or Docker, see [these notes](https://github.com/pg
## 0.4.0 (2023-01-11)
If upgrading with Postgres < 13, see [this note](https://github.com/pgvector/pgvector/blob/v0.4.0/README.md#040).
If upgrading with Postgres < 13, see [this note](https://github.com/pgvector/pgvector#040).
- Changed text representation for vector elements to match `real`
- Changed storage for vector from `plain` to `extended`
@@ -149,7 +61,7 @@ If upgrading with Postgres < 13, see [this note](https://github.com/pgvector/pgv
## 0.3.1 (2022-11-02)
If upgrading from 0.2.7 or 0.3.0, [recreate](https://github.com/pgvector/pgvector/blob/v0.3.1/README.md#031) all `ivfflat` indexes after upgrading to ensure all data is indexed.
If upgrading from 0.2.7 or 0.3.0, [recreate](https://github.com/pgvector/pgvector#031) all `ivfflat` indexes after upgrading to ensure all data is indexed.
- Fixed issue with inserts silently corrupting `ivfflat` indexes (introduced in 0.2.7)
- Fixed segmentation fault with index creation when lists > 6500

View File

@@ -1,11 +1,8 @@
# syntax=docker/dockerfile:1
ARG PG_MAJOR=17
ARG DEBIAN_CODENAME=bookworm
FROM postgres:$PG_MAJOR-$DEBIAN_CODENAME
ARG PG_MAJOR=15
FROM postgres:$PG_MAJOR
ARG PG_MAJOR
ADD https://github.com/pgvector/pgvector.git#v0.8.3 /tmp/pgvector
COPY . /tmp/pgvector
RUN apt-get update && \
apt-mark hold locales && \

View File

@@ -1,4 +1,4 @@
Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
Portions Copyright (c) 1994, The Regents of the University of California

View File

@@ -2,7 +2,7 @@
"name": "vector",
"abstract": "Open-source vector similarity search for Postgres",
"description": "Supports L2 distance, inner product, and cosine distance",
"version": "0.8.3",
"version": "0.5.1",
"maintainer": [
"Andrew Kane <andrew@ankane.org>"
],
@@ -12,7 +12,7 @@
"prereqs": {
"runtime": {
"requires": {
"PostgreSQL": "13.0.0"
"PostgreSQL": "11.0.0"
}
}
},
@@ -20,7 +20,7 @@
"vector": {
"file": "sql/vector.sql",
"docfile": "README.md",
"version": "0.8.3",
"version": "0.5.1",
"abstract": "Open-source vector similarity search for Postgres"
}
},

View File

@@ -1,20 +1,18 @@
EXTENSION = vector
EXTVERSION = 0.8.3
EXTVERSION = 0.5.1
MODULE_big = vector
DATA = $(wildcard sql/*--*--*.sql)
DATA_built = sql/$(EXTENSION)--$(EXTVERSION).sql
OBJS = src/bitutils.o src/bitvec.o src/halfutils.o src/halfvec.o src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/sparsevec.o src/vector.o
HEADERS = src/halfvec.h src/sparsevec.h src/vector.h
DATA = $(wildcard sql/*--*.sql)
OBJS = src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/vector.o
HEADERS = src/vector.h
TESTS = $(wildcard test/sql/*.sql)
REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))
REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)
# To compile for portability, run: make OPTFLAGS=""
OPTFLAGS = -march=native
# Mac ARM doesn't always support -march=native
# Mac ARM doesn't support -march=native
ifeq ($(shell uname -s), Darwin)
ifeq ($(shell uname -p), arm)
# no difference with -march=armv8.5-a
@@ -27,11 +25,6 @@ ifneq ($(filter ppc64%, $(shell uname -m)), )
OPTFLAGS =
endif
# RISC-V64 doesn't support -march=native
ifeq ($(shell uname -m), riscv64)
OPTFLAGS =
endif
# For auto-vectorization:
# - GCC (needs -ftree-vectorize OR -O3) - https://gcc.gnu.org/projects/tree-ssa/vectorization.html
# - Clang (could use pragma instead) - https://llvm.org/docs/Vectorizers.html
@@ -48,6 +41,8 @@ all: sql/$(EXTENSION)--$(EXTVERSION).sql
sql/$(EXTENSION)--$(EXTVERSION).sql: sql/$(EXTENSION).sql
cp $< $@
EXTRA_CLEAN = sql/$(EXTENSION)--$(EXTVERSION).sql
PG_CONFIG ?= pg_config
PGXS := $(shell $(PG_CONFIG) --pgxs)
include $(PGXS)
@@ -57,7 +52,7 @@ ifeq ($(PROVE),)
PROVE = prove
endif
# for Postgres < 15
# for Postgres 15
PROVE_FLAGS += -I ./test/perl
prove_installcheck:
@@ -70,20 +65,13 @@ dist:
mkdir -p dist
git archive --format zip --prefix=$(EXTENSION)-$(EXTVERSION)/ --output dist/$(EXTENSION)-$(EXTVERSION).zip master
# for Docker
PG_MAJOR ?= 17
.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 --platform linux/amd64 -t ankane/pgvector:latest .
.PHONY: docker-release
docker-release:
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$(PG_MAJOR)-trixie .
docker buildx build --push --pull --no-cache --platform linux/amd64,linux/arm64 -t ankane/pgvector:latest .
docker buildx build --push --platform linux/amd64,linux/arm64 -t ankane/pgvector:v$(EXTVERSION) .

View File

@@ -1,11 +1,10 @@
EXTENSION = vector
EXTVERSION = 0.8.3
EXTVERSION = 0.5.1
DATA_built = sql\$(EXTENSION)--$(EXTVERSION).sql
OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj
HEADERS = src\halfvec.h src\sparsevec.h src\vector.h
OBJS = src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
HEADERS = src\vector.h
REGRESS = bit btree cast copy halfvec hnsw_bit hnsw_halfvec hnsw_sparsevec hnsw_vector ivfflat_bit ivfflat_halfvec ivfflat_vector sparsevec vector_type
REGRESS = btree cast copy functions input ivfflat_cosine ivfflat_ip ivfflat_l2 ivfflat_options ivfflat_unlogged
REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)
# For /arch flags
@@ -20,6 +19,11 @@ PG_CFLAGS = $(PG_CFLAGS) $(OPTFLAGS) /O2 /fp:fast
# https://learn.microsoft.com/en-us/cpp/error-messages/tool-errors/vectorizer-and-parallelizer-messages
# PG_CFLAGS = $(PG_CFLAGS) /Qvec-report:2
all: sql\$(EXTENSION)--$(EXTVERSION).sql
sql\$(EXTENSION)--$(EXTVERSION).sql: sql\$(EXTENSION).sql
copy sql\$(EXTENSION).sql $@
# TODO use pg_config
!ifndef PGROOT
!error PGROOT is not set
@@ -31,9 +35,6 @@ 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)
@@ -42,26 +43,23 @@ SHLIB = $(EXTENSION).dll
LIBS = "$(LIBDIR)\postgres.lib"
all: $(SHLIB) $(DATA_built)
.c.obj:
$(CC) $(CFLAGS) /c $< /Fo$@
$(SHLIB): $(OBJS)
$(CC) $(CFLAGS) $(OBJS) $(LIBS) /link /DLL /OUT:$(SHLIB)
sql\$(EXTENSION)--$(EXTVERSION).sql: sql\$(EXTENSION).sql
copy sql\$(EXTENSION).sql $@
all: $(SHLIB)
install: all
install:
copy $(SHLIB) "$(PKGLIBDIR)"
copy $(EXTENSION).control "$(SHAREDIR)\extension"
copy sql\$(EXTENSION)--*.sql "$(SHAREDIR)\extension"
if not exist "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)" mkdir "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
for %f in ($(HEADERS)) do copy %f "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
mkdir "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
copy $(HEADERS) "$(INCLUDEDIR_SERVER)\extension\$(EXTENSION)"
installcheck:
"$(PG_REGRESS)" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS)
"$(BINDIR)\pg_regress" --bindir="$(BINDIR)" $(REGRESS_OPTS) $(REGRESS)
uninstall:
del /f "$(PKGLIBDIR)\$(SHLIB)"
@@ -72,6 +70,6 @@ uninstall:
clean:
del /f $(SHLIB) $(EXTENSION).lib $(EXTENSION).exp
del /f $(DATA_built)
del /f $(OBJS)
del /f sql\$(EXTENSION)--$(EXTVERSION).sql
del /f /s /q results regression.diffs regression.out tmp_check tmp_check_iso log output_iso

976
README.md

File diff suppressed because it is too large Load Diff

View File

@@ -1,5 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.6.0'" to load this file. \quit
-- remove this single line for Postgres < 13
ALTER TYPE vector SET (STORAGE = external);

View File

@@ -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
);

View File

@@ -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

View File

@@ -1,569 +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 FUNCTION l2_normalize(vector) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION binary_quantize(vector) RETURNS bit
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION subvector(vector, int, int) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_concat(vector, vector) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE OPERATOR <+> (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = l1_distance,
COMMUTATOR = '<+>'
);
CREATE OPERATOR || (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_concat
);
CREATE FUNCTION ivfflat_halfvec_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION ivfflat_bit_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION hnsw_halfvec_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION hnsw_bit_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION hnsw_sparsevec_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE OPERATOR CLASS vector_l1_ops
FOR TYPE vector USING hnsw AS
OPERATOR 1 <+> (vector, vector) FOR ORDER BY float_ops,
FUNCTION 1 l1_distance(vector, vector);
CREATE TYPE halfvec;
CREATE FUNCTION halfvec_in(cstring, oid, integer) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_out(halfvec) RETURNS cstring
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_typmod_in(cstring[]) RETURNS integer
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_recv(internal, oid, integer) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_send(halfvec) RETURNS bytea
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE TYPE halfvec (
INPUT = halfvec_in,
OUTPUT = halfvec_out,
TYPMOD_IN = halfvec_typmod_in,
RECEIVE = halfvec_recv,
SEND = halfvec_send,
STORAGE = external
);
CREATE FUNCTION l2_distance(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME', 'halfvec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION inner_product(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME', 'halfvec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION cosine_distance(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME', 'halfvec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l1_distance(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME', 'halfvec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_dims(halfvec) RETURNS integer
AS 'MODULE_PATHNAME', 'halfvec_vector_dims' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_norm(halfvec) RETURNS float8
AS 'MODULE_PATHNAME', 'halfvec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_normalize(halfvec) RETURNS halfvec
AS 'MODULE_PATHNAME', 'halfvec_l2_normalize' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION binary_quantize(halfvec) RETURNS bit
AS 'MODULE_PATHNAME', 'halfvec_binary_quantize' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION subvector(halfvec, int, int) RETURNS halfvec
AS 'MODULE_PATHNAME', 'halfvec_subvector' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_add(halfvec, halfvec) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_sub(halfvec, halfvec) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_mul(halfvec, halfvec) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_concat(halfvec, halfvec) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_lt(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_le(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_eq(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_ne(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_ge(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_gt(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_cmp(halfvec, halfvec) RETURNS int4
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_l2_squared_distance(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_negative_inner_product(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_spherical_distance(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_accum(double precision[], halfvec) RETURNS double precision[]
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_avg(double precision[]) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_combine(double precision[], double precision[]) RETURNS double precision[]
AS 'MODULE_PATHNAME', 'vector_combine' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE AGGREGATE avg(halfvec) (
SFUNC = halfvec_accum,
STYPE = double precision[],
FINALFUNC = halfvec_avg,
COMBINEFUNC = halfvec_combine,
INITCOND = '{0}',
PARALLEL = SAFE
);
CREATE AGGREGATE sum(halfvec) (
SFUNC = halfvec_add,
STYPE = halfvec,
COMBINEFUNC = halfvec_add,
PARALLEL = SAFE
);
CREATE FUNCTION halfvec(halfvec, integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_to_vector(halfvec, integer, boolean) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_to_halfvec(vector, integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_halfvec(integer[], integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_halfvec(real[], integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_halfvec(double precision[], integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_halfvec(numeric[], integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_to_float4(halfvec, integer, boolean) RETURNS real[]
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE CAST (halfvec AS halfvec)
WITH FUNCTION halfvec(halfvec, integer, boolean) AS IMPLICIT;
CREATE CAST (halfvec AS vector)
WITH FUNCTION halfvec_to_vector(halfvec, integer, boolean) AS ASSIGNMENT;
CREATE CAST (vector AS halfvec)
WITH FUNCTION vector_to_halfvec(vector, integer, boolean) AS IMPLICIT;
CREATE CAST (halfvec AS real[])
WITH FUNCTION halfvec_to_float4(halfvec, integer, boolean) AS ASSIGNMENT;
CREATE CAST (integer[] AS halfvec)
WITH FUNCTION array_to_halfvec(integer[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (real[] AS halfvec)
WITH FUNCTION array_to_halfvec(real[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (double precision[] AS halfvec)
WITH FUNCTION array_to_halfvec(double precision[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (numeric[] AS halfvec)
WITH FUNCTION array_to_halfvec(numeric[], integer, boolean) AS ASSIGNMENT;
CREATE OPERATOR <-> (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = l2_distance,
COMMUTATOR = '<->'
);
CREATE OPERATOR <#> (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_negative_inner_product,
COMMUTATOR = '<#>'
);
CREATE OPERATOR <=> (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = cosine_distance,
COMMUTATOR = '<=>'
);
CREATE OPERATOR <+> (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = l1_distance,
COMMUTATOR = '<+>'
);
CREATE OPERATOR + (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_add,
COMMUTATOR = +
);
CREATE OPERATOR - (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_sub
);
CREATE OPERATOR * (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_mul,
COMMUTATOR = *
);
CREATE OPERATOR || (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_concat
);
CREATE OPERATOR < (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_lt,
COMMUTATOR = > , NEGATOR = >= ,
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
);
CREATE OPERATOR <= (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_le,
COMMUTATOR = >= , NEGATOR = > ,
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
);
CREATE OPERATOR = (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_eq,
COMMUTATOR = = , NEGATOR = <> ,
RESTRICT = eqsel, JOIN = eqjoinsel
);
CREATE OPERATOR <> (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_ne,
COMMUTATOR = <> , NEGATOR = = ,
RESTRICT = eqsel, JOIN = eqjoinsel
);
CREATE OPERATOR >= (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_ge,
COMMUTATOR = <= , NEGATOR = < ,
RESTRICT = scalargesel, JOIN = scalargejoinsel
);
CREATE OPERATOR > (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_gt,
COMMUTATOR = < , NEGATOR = <= ,
RESTRICT = scalargtsel, JOIN = scalargtjoinsel
);
CREATE OPERATOR CLASS halfvec_ops
DEFAULT FOR TYPE halfvec USING btree AS
OPERATOR 1 < ,
OPERATOR 2 <= ,
OPERATOR 3 = ,
OPERATOR 4 >= ,
OPERATOR 5 > ,
FUNCTION 1 halfvec_cmp(halfvec, halfvec);
CREATE OPERATOR CLASS halfvec_l2_ops
FOR TYPE halfvec USING ivfflat AS
OPERATOR 1 <-> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_l2_squared_distance(halfvec, halfvec),
FUNCTION 3 l2_distance(halfvec, halfvec),
FUNCTION 5 ivfflat_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_ip_ops
FOR TYPE halfvec USING ivfflat AS
OPERATOR 1 <#> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
FUNCTION 3 halfvec_spherical_distance(halfvec, halfvec),
FUNCTION 4 l2_norm(halfvec),
FUNCTION 5 ivfflat_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_cosine_ops
FOR TYPE halfvec USING ivfflat AS
OPERATOR 1 <=> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
FUNCTION 2 l2_norm(halfvec),
FUNCTION 3 halfvec_spherical_distance(halfvec, halfvec),
FUNCTION 4 l2_norm(halfvec),
FUNCTION 5 ivfflat_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_l2_ops
FOR TYPE halfvec USING hnsw AS
OPERATOR 1 <-> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_l2_squared_distance(halfvec, halfvec),
FUNCTION 3 hnsw_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_ip_ops
FOR TYPE halfvec USING hnsw AS
OPERATOR 1 <#> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
FUNCTION 3 hnsw_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_cosine_ops
FOR TYPE halfvec USING hnsw AS
OPERATOR 1 <=> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
FUNCTION 2 l2_norm(halfvec),
FUNCTION 3 hnsw_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_l1_ops
FOR TYPE halfvec USING hnsw AS
OPERATOR 1 <+> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 l1_distance(halfvec, halfvec),
FUNCTION 3 hnsw_halfvec_support(internal);
CREATE FUNCTION hamming_distance(bit, bit) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION jaccard_distance(bit, bit) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE OPERATOR <~> (
LEFTARG = bit, RIGHTARG = bit, PROCEDURE = hamming_distance,
COMMUTATOR = '<~>'
);
CREATE OPERATOR <%> (
LEFTARG = bit, RIGHTARG = bit, PROCEDURE = jaccard_distance,
COMMUTATOR = '<%>'
);
CREATE OPERATOR CLASS bit_hamming_ops
FOR TYPE bit USING ivfflat AS
OPERATOR 1 <~> (bit, bit) FOR ORDER BY float_ops,
FUNCTION 1 hamming_distance(bit, bit),
FUNCTION 3 hamming_distance(bit, bit),
FUNCTION 5 ivfflat_bit_support(internal);
CREATE OPERATOR CLASS bit_hamming_ops
FOR TYPE bit USING hnsw AS
OPERATOR 1 <~> (bit, bit) FOR ORDER BY float_ops,
FUNCTION 1 hamming_distance(bit, bit),
FUNCTION 3 hnsw_bit_support(internal);
CREATE OPERATOR CLASS bit_jaccard_ops
FOR TYPE bit USING hnsw AS
OPERATOR 1 <%> (bit, bit) FOR ORDER BY float_ops,
FUNCTION 1 jaccard_distance(bit, bit),
FUNCTION 3 hnsw_bit_support(internal);
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 l1_distance(sparsevec, sparsevec) RETURNS float8
AS 'MODULE_PATHNAME', 'sparsevec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_norm(sparsevec) RETURNS float8
AS 'MODULE_PATHNAME', 'sparsevec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_normalize(sparsevec) RETURNS sparsevec
AS 'MODULE_PATHNAME', 'sparsevec_l2_normalize' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_lt(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_le(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_eq(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_ne(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_ge(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_gt(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_cmp(sparsevec, sparsevec) RETURNS int4
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 FUNCTION halfvec_to_sparsevec(halfvec, integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_to_halfvec(sparsevec, integer, boolean) RETURNS halfvec
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 ASSIGNMENT;
CREATE CAST (vector AS sparsevec)
WITH FUNCTION vector_to_sparsevec(vector, integer, boolean) AS IMPLICIT;
CREATE CAST (sparsevec AS halfvec)
WITH FUNCTION sparsevec_to_halfvec(sparsevec, integer, boolean) AS ASSIGNMENT;
CREATE CAST (halfvec AS sparsevec)
WITH FUNCTION halfvec_to_sparsevec(halfvec, 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 <+> (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = l1_distance,
COMMUTATOR = '<+>'
);
CREATE OPERATOR < (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_lt,
COMMUTATOR = > , NEGATOR = >= ,
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
);
CREATE OPERATOR <= (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_le,
COMMUTATOR = >= , NEGATOR = > ,
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
);
CREATE OPERATOR = (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_eq,
COMMUTATOR = = , NEGATOR = <> ,
RESTRICT = eqsel, JOIN = eqjoinsel
);
CREATE OPERATOR <> (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_ne,
COMMUTATOR = <> , NEGATOR = = ,
RESTRICT = eqsel, JOIN = eqjoinsel
);
CREATE OPERATOR >= (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_ge,
COMMUTATOR = <= , NEGATOR = < ,
RESTRICT = scalargesel, JOIN = scalargejoinsel
);
CREATE OPERATOR > (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_gt,
COMMUTATOR = < , NEGATOR = <= ,
RESTRICT = scalargtsel, JOIN = scalargtjoinsel
);
CREATE OPERATOR CLASS sparsevec_ops
DEFAULT FOR TYPE sparsevec USING btree AS
OPERATOR 1 < ,
OPERATOR 2 <= ,
OPERATOR 3 = ,
OPERATOR 4 >= ,
OPERATOR 5 > ,
FUNCTION 1 sparsevec_cmp(sparsevec, sparsevec);
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),
FUNCTION 3 hnsw_sparsevec_support(internal);
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),
FUNCTION 3 hnsw_sparsevec_support(internal);
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 l2_norm(sparsevec),
FUNCTION 3 hnsw_sparsevec_support(internal);
CREATE OPERATOR CLASS sparsevec_l1_ops
FOR TYPE sparsevec USING hnsw AS
OPERATOR 1 <+> (sparsevec, sparsevec) FOR ORDER BY float_ops,
FUNCTION 1 l1_distance(sparsevec, sparsevec),
FUNCTION 3 hnsw_sparsevec_support(internal);

View File

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.7.1'" to load this file. \quit

View File

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.7.2'" to load this file. \quit

View File

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

View File

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

View File

@@ -1,26 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.8.0'" to load this file. \quit
CREATE FUNCTION array_to_sparsevec(integer[], integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_sparsevec(real[], integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_sparsevec(double precision[], integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_sparsevec(numeric[], integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE CAST (integer[] AS sparsevec)
WITH FUNCTION array_to_sparsevec(integer[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (real[] AS sparsevec)
WITH FUNCTION array_to_sparsevec(real[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (double precision[] AS sparsevec)
WITH FUNCTION array_to_sparsevec(double precision[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (numeric[] AS sparsevec)
WITH FUNCTION array_to_sparsevec(numeric[], integer, boolean) AS ASSIGNMENT;

View File

@@ -1,2 +0,0 @@
-- 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

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.8.2'" to load this file. \quit

View File

@@ -1,2 +0,0 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION
\echo Use "ALTER EXTENSION vector UPDATE TO '0.8.3'" to load this file. \quit

View File

@@ -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;
@@ -26,10 +26,10 @@ CREATE TYPE vector (
TYPMOD_IN = vector_typmod_in,
RECEIVE = vector_recv,
SEND = vector_send,
STORAGE = external
STORAGE = extended
);
-- vector functions
-- functions
CREATE FUNCTION l2_distance(vector, vector) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -49,17 +49,6 @@ CREATE FUNCTION vector_dims(vector) RETURNS integer
CREATE FUNCTION vector_norm(vector) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_normalize(vector) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION binary_quantize(vector) RETURNS bit
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION subvector(vector, int, int) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- vector private functions
CREATE FUNCTION vector_add(vector, vector) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -69,8 +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;
CREATE FUNCTION vector_concat(vector, vector) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- private functions
CREATE FUNCTION vector_lt(vector, vector) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
@@ -111,7 +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,
@@ -129,7 +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;
@@ -149,7 +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;
@@ -169,7 +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,
@@ -186,18 +174,14 @@ CREATE OPERATOR <=> (
COMMUTATOR = '<=>'
);
CREATE OPERATOR <+> (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = l1_distance,
COMMUTATOR = '<+>'
);
CREATE OPERATOR + (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_add,
COMMUTATOR = +
);
CREATE OPERATOR - (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_sub,
COMMUTATOR = -
);
CREATE OPERATOR * (
@@ -205,20 +189,17 @@ CREATE OPERATOR * (
COMMUTATOR = *
);
CREATE OPERATOR || (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_concat
);
CREATE OPERATOR < (
LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_lt,
COMMUTATOR = > , NEGATOR = >= ,
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 = (
@@ -233,10 +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 > (
@@ -261,24 +243,7 @@ CREATE ACCESS METHOD hnsw TYPE INDEX HANDLER hnswhandler;
COMMENT ON ACCESS METHOD hnsw IS 'hnsw index access method';
-- access method private functions
CREATE FUNCTION ivfflat_halfvec_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION ivfflat_bit_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION hnsw_halfvec_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION hnsw_bit_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION hnsw_sparsevec_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
-- vector opclasses
-- opclasses
CREATE OPERATOR CLASS vector_ops
DEFAULT FOR TYPE vector USING btree AS
@@ -325,594 +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);
CREATE OPERATOR CLASS vector_l1_ops
FOR TYPE vector USING hnsw AS
OPERATOR 1 <+> (vector, vector) FOR ORDER BY float_ops,
FUNCTION 1 l1_distance(vector, vector);
-- halfvec type
CREATE TYPE halfvec;
CREATE FUNCTION halfvec_in(cstring, oid, integer) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_out(halfvec) RETURNS cstring
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_typmod_in(cstring[]) RETURNS integer
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_recv(internal, oid, integer) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_send(halfvec) RETURNS bytea
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE TYPE halfvec (
INPUT = halfvec_in,
OUTPUT = halfvec_out,
TYPMOD_IN = halfvec_typmod_in,
RECEIVE = halfvec_recv,
SEND = halfvec_send,
STORAGE = external
);
-- halfvec functions
CREATE FUNCTION l2_distance(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME', 'halfvec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION inner_product(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME', 'halfvec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION cosine_distance(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME', 'halfvec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l1_distance(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME', 'halfvec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_dims(halfvec) RETURNS integer
AS 'MODULE_PATHNAME', 'halfvec_vector_dims' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_norm(halfvec) RETURNS float8
AS 'MODULE_PATHNAME', 'halfvec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_normalize(halfvec) RETURNS halfvec
AS 'MODULE_PATHNAME', 'halfvec_l2_normalize' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION binary_quantize(halfvec) RETURNS bit
AS 'MODULE_PATHNAME', 'halfvec_binary_quantize' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION subvector(halfvec, int, int) RETURNS halfvec
AS 'MODULE_PATHNAME', 'halfvec_subvector' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- halfvec private functions
CREATE FUNCTION halfvec_add(halfvec, halfvec) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_sub(halfvec, halfvec) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_mul(halfvec, halfvec) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_concat(halfvec, halfvec) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_lt(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_le(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_eq(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_ne(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_ge(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_gt(halfvec, halfvec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_cmp(halfvec, halfvec) RETURNS int4
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_l2_squared_distance(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_negative_inner_product(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_spherical_distance(halfvec, halfvec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_accum(double precision[], halfvec) RETURNS double precision[]
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_avg(double precision[]) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_combine(double precision[], double precision[]) RETURNS double precision[]
AS 'MODULE_PATHNAME', 'vector_combine' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- halfvec aggregates
CREATE AGGREGATE avg(halfvec) (
SFUNC = halfvec_accum,
STYPE = double precision[],
FINALFUNC = halfvec_avg,
COMBINEFUNC = halfvec_combine,
INITCOND = '{0}',
PARALLEL = SAFE
);
CREATE AGGREGATE sum(halfvec) (
SFUNC = halfvec_add,
STYPE = halfvec,
COMBINEFUNC = halfvec_add,
PARALLEL = SAFE
);
-- halfvec cast functions
CREATE FUNCTION halfvec(halfvec, integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_to_vector(halfvec, integer, boolean) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_to_halfvec(vector, integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_halfvec(integer[], integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_halfvec(real[], integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_halfvec(double precision[], integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_halfvec(numeric[], integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION halfvec_to_float4(halfvec, integer, boolean) RETURNS real[]
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- halfvec casts
CREATE CAST (halfvec AS halfvec)
WITH FUNCTION halfvec(halfvec, integer, boolean) AS IMPLICIT;
CREATE CAST (halfvec AS vector)
WITH FUNCTION halfvec_to_vector(halfvec, integer, boolean) AS ASSIGNMENT;
CREATE CAST (vector AS halfvec)
WITH FUNCTION vector_to_halfvec(vector, integer, boolean) AS IMPLICIT;
CREATE CAST (halfvec AS real[])
WITH FUNCTION halfvec_to_float4(halfvec, integer, boolean) AS ASSIGNMENT;
CREATE CAST (integer[] AS halfvec)
WITH FUNCTION array_to_halfvec(integer[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (real[] AS halfvec)
WITH FUNCTION array_to_halfvec(real[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (double precision[] AS halfvec)
WITH FUNCTION array_to_halfvec(double precision[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (numeric[] AS halfvec)
WITH FUNCTION array_to_halfvec(numeric[], integer, boolean) AS ASSIGNMENT;
-- halfvec operators
CREATE OPERATOR <-> (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = l2_distance,
COMMUTATOR = '<->'
);
CREATE OPERATOR <#> (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_negative_inner_product,
COMMUTATOR = '<#>'
);
CREATE OPERATOR <=> (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = cosine_distance,
COMMUTATOR = '<=>'
);
CREATE OPERATOR <+> (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = l1_distance,
COMMUTATOR = '<+>'
);
CREATE OPERATOR + (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_add,
COMMUTATOR = +
);
CREATE OPERATOR - (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_sub
);
CREATE OPERATOR * (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_mul,
COMMUTATOR = *
);
CREATE OPERATOR || (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_concat
);
CREATE OPERATOR < (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_lt,
COMMUTATOR = > , NEGATOR = >= ,
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
);
CREATE OPERATOR <= (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_le,
COMMUTATOR = >= , NEGATOR = > ,
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
);
CREATE OPERATOR = (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_eq,
COMMUTATOR = = , NEGATOR = <> ,
RESTRICT = eqsel, JOIN = eqjoinsel
);
CREATE OPERATOR <> (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_ne,
COMMUTATOR = <> , NEGATOR = = ,
RESTRICT = eqsel, JOIN = eqjoinsel
);
CREATE OPERATOR >= (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_ge,
COMMUTATOR = <= , NEGATOR = < ,
RESTRICT = scalargesel, JOIN = scalargejoinsel
);
CREATE OPERATOR > (
LEFTARG = halfvec, RIGHTARG = halfvec, PROCEDURE = halfvec_gt,
COMMUTATOR = < , NEGATOR = <= ,
RESTRICT = scalargtsel, JOIN = scalargtjoinsel
);
-- halfvec opclasses
CREATE OPERATOR CLASS halfvec_ops
DEFAULT FOR TYPE halfvec USING btree AS
OPERATOR 1 < ,
OPERATOR 2 <= ,
OPERATOR 3 = ,
OPERATOR 4 >= ,
OPERATOR 5 > ,
FUNCTION 1 halfvec_cmp(halfvec, halfvec);
CREATE OPERATOR CLASS halfvec_l2_ops
FOR TYPE halfvec USING ivfflat AS
OPERATOR 1 <-> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_l2_squared_distance(halfvec, halfvec),
FUNCTION 3 l2_distance(halfvec, halfvec),
FUNCTION 5 ivfflat_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_ip_ops
FOR TYPE halfvec USING ivfflat AS
OPERATOR 1 <#> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
FUNCTION 3 halfvec_spherical_distance(halfvec, halfvec),
FUNCTION 4 l2_norm(halfvec),
FUNCTION 5 ivfflat_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_cosine_ops
FOR TYPE halfvec USING ivfflat AS
OPERATOR 1 <=> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
FUNCTION 2 l2_norm(halfvec),
FUNCTION 3 halfvec_spherical_distance(halfvec, halfvec),
FUNCTION 4 l2_norm(halfvec),
FUNCTION 5 ivfflat_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_l2_ops
FOR TYPE halfvec USING hnsw AS
OPERATOR 1 <-> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_l2_squared_distance(halfvec, halfvec),
FUNCTION 3 hnsw_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_ip_ops
FOR TYPE halfvec USING hnsw AS
OPERATOR 1 <#> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
FUNCTION 3 hnsw_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_cosine_ops
FOR TYPE halfvec USING hnsw AS
OPERATOR 1 <=> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 halfvec_negative_inner_product(halfvec, halfvec),
FUNCTION 2 l2_norm(halfvec),
FUNCTION 3 hnsw_halfvec_support(internal);
CREATE OPERATOR CLASS halfvec_l1_ops
FOR TYPE halfvec USING hnsw AS
OPERATOR 1 <+> (halfvec, halfvec) FOR ORDER BY float_ops,
FUNCTION 1 l1_distance(halfvec, halfvec),
FUNCTION 3 hnsw_halfvec_support(internal);
-- bit functions
CREATE FUNCTION hamming_distance(bit, bit) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION jaccard_distance(bit, bit) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- bit operators
CREATE OPERATOR <~> (
LEFTARG = bit, RIGHTARG = bit, PROCEDURE = hamming_distance,
COMMUTATOR = '<~>'
);
CREATE OPERATOR <%> (
LEFTARG = bit, RIGHTARG = bit, PROCEDURE = jaccard_distance,
COMMUTATOR = '<%>'
);
-- bit opclasses
CREATE OPERATOR CLASS bit_hamming_ops
FOR TYPE bit USING ivfflat AS
OPERATOR 1 <~> (bit, bit) FOR ORDER BY float_ops,
FUNCTION 1 hamming_distance(bit, bit),
FUNCTION 3 hamming_distance(bit, bit),
FUNCTION 5 ivfflat_bit_support(internal);
CREATE OPERATOR CLASS bit_hamming_ops
FOR TYPE bit USING hnsw AS
OPERATOR 1 <~> (bit, bit) FOR ORDER BY float_ops,
FUNCTION 1 hamming_distance(bit, bit),
FUNCTION 3 hnsw_bit_support(internal);
CREATE OPERATOR CLASS bit_jaccard_ops
FOR TYPE bit USING hnsw AS
OPERATOR 1 <%> (bit, bit) FOR ORDER BY float_ops,
FUNCTION 1 jaccard_distance(bit, bit),
FUNCTION 3 hnsw_bit_support(internal);
--- 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 l1_distance(sparsevec, sparsevec) RETURNS float8
AS 'MODULE_PATHNAME', 'sparsevec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_norm(sparsevec) RETURNS float8
AS 'MODULE_PATHNAME', 'sparsevec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_normalize(sparsevec) RETURNS sparsevec
AS 'MODULE_PATHNAME', 'sparsevec_l2_normalize' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- sparsevec private functions
CREATE FUNCTION sparsevec_lt(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_le(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_eq(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_ne(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_ge(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_gt(sparsevec, sparsevec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_cmp(sparsevec, sparsevec) RETURNS int4
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;
-- 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;
CREATE FUNCTION halfvec_to_sparsevec(halfvec, integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION sparsevec_to_halfvec(sparsevec, integer, boolean) RETURNS halfvec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_sparsevec(integer[], integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_sparsevec(real[], integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_sparsevec(double precision[], integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_sparsevec(numeric[], integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- sparsevec casts
CREATE CAST (sparsevec AS sparsevec)
WITH FUNCTION sparsevec(sparsevec, integer, boolean) AS IMPLICIT;
CREATE CAST (sparsevec AS vector)
WITH FUNCTION sparsevec_to_vector(sparsevec, integer, boolean) AS ASSIGNMENT;
CREATE CAST (vector AS sparsevec)
WITH FUNCTION vector_to_sparsevec(vector, integer, boolean) AS IMPLICIT;
CREATE CAST (sparsevec AS halfvec)
WITH FUNCTION sparsevec_to_halfvec(sparsevec, integer, boolean) AS ASSIGNMENT;
CREATE CAST (halfvec AS sparsevec)
WITH FUNCTION halfvec_to_sparsevec(halfvec, integer, boolean) AS IMPLICIT;
CREATE CAST (integer[] AS sparsevec)
WITH FUNCTION array_to_sparsevec(integer[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (real[] AS sparsevec)
WITH FUNCTION array_to_sparsevec(real[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (double precision[] AS sparsevec)
WITH FUNCTION array_to_sparsevec(double precision[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (numeric[] AS sparsevec)
WITH FUNCTION array_to_sparsevec(numeric[], integer, boolean) AS ASSIGNMENT;
-- sparsevec operators
CREATE OPERATOR <-> (
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 <+> (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = l1_distance,
COMMUTATOR = '<+>'
);
CREATE OPERATOR < (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_lt,
COMMUTATOR = > , NEGATOR = >= ,
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
);
CREATE OPERATOR <= (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_le,
COMMUTATOR = >= , NEGATOR = > ,
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
);
CREATE OPERATOR = (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_eq,
COMMUTATOR = = , NEGATOR = <> ,
RESTRICT = eqsel, JOIN = eqjoinsel
);
CREATE OPERATOR <> (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_ne,
COMMUTATOR = <> , NEGATOR = = ,
RESTRICT = eqsel, JOIN = eqjoinsel
);
CREATE OPERATOR >= (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_ge,
COMMUTATOR = <= , NEGATOR = < ,
RESTRICT = scalargesel, JOIN = scalargejoinsel
);
CREATE OPERATOR > (
LEFTARG = sparsevec, RIGHTARG = sparsevec, PROCEDURE = sparsevec_gt,
COMMUTATOR = < , NEGATOR = <= ,
RESTRICT = scalargtsel, JOIN = scalargtjoinsel
);
-- sparsevec opclasses
CREATE OPERATOR CLASS sparsevec_ops
DEFAULT FOR TYPE sparsevec USING btree AS
OPERATOR 1 < ,
OPERATOR 2 <= ,
OPERATOR 3 = ,
OPERATOR 4 >= ,
OPERATOR 5 > ,
FUNCTION 1 sparsevec_cmp(sparsevec, sparsevec);
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),
FUNCTION 3 hnsw_sparsevec_support(internal);
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),
FUNCTION 3 hnsw_sparsevec_support(internal);
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 l2_norm(sparsevec),
FUNCTION 3 hnsw_sparsevec_support(internal);
CREATE OPERATOR CLASS sparsevec_l1_ops
FOR TYPE sparsevec USING hnsw AS
OPERATOR 1 <+> (sparsevec, sparsevec) FOR ORDER BY float_ops,
FUNCTION 1 l1_distance(sparsevec, sparsevec),
FUNCTION 3 hnsw_sparsevec_support(internal);

View File

@@ -1,224 +0,0 @@
#include "postgres.h"
#include "bitutils.h"
#include "halfvec.h" /* for USE_DISPATCH and USE_TARGET_CLONES */
#include "port/pg_bitutils.h"
#if defined(USE_DISPATCH)
#define BIT_DISPATCH
#endif
#ifdef BIT_DISPATCH
#include <immintrin.h>
#if defined(USE__GET_CPUID)
#include <cpuid.h>
#else
#include <intrin.h>
#endif
#ifdef _MSC_VER
#define TARGET_AVX512_POPCOUNT
#else
#define TARGET_AVX512_POPCOUNT __attribute__((target("avx512f,avx512vpopcntdq")))
#endif
#endif
/* Disable for LLVM due to crash with bitcode generation */
#if defined(USE_TARGET_CLONES) && !defined(__POPCNT__) && !defined(__llvm__)
#define BIT_TARGET_CLONES __attribute__((target_clones("default", "popcnt")))
#else
#define BIT_TARGET_CLONES
#endif
/* Use built-ins when possible for Postgres < 19 for inlining */
#if PG_VERSION_NUM >= 190000
#define popcount64(x) pg_popcount64(x)
#elif defined(HAVE__BUILTIN_POPCOUNT) && (defined(HAVE_LONG_INT_64) || SIZEOF_LONG == 8)
#define popcount64(x) __builtin_popcountl(x)
#elif defined(HAVE__BUILTIN_POPCOUNT) && (defined(HAVE_LONG_LONG_INT_64) || SIZEOF_LONG_LONG == 8)
#define popcount64(x) __builtin_popcountll(x)
#elif !defined(_MSC_VER)
/* Fails to resolve with MSVC */
#define popcount64(x) pg_popcount64(x)
#endif
uint64 (*BitHammingDistance) (uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 distance);
double (*BitJaccardDistance) (uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 ab, uint64 aa, uint64 bb);
BIT_TARGET_CLONES static uint64
BitHammingDistanceDefault(uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 distance)
{
#ifdef popcount64
for (; bytes >= sizeof(uint64); bytes -= sizeof(uint64))
{
uint64 axs;
uint64 bxs;
/* Ensure aligned */
memcpy(&axs, ax, sizeof(uint64));
memcpy(&bxs, bx, sizeof(uint64));
distance += popcount64(axs ^ bxs);
ax += sizeof(uint64);
bx += sizeof(uint64);
}
#endif
for (uint32 i = 0; i < bytes; i++)
distance += pg_number_of_ones[ax[i] ^ bx[i]];
return distance;
}
#ifdef BIT_DISPATCH
TARGET_AVX512_POPCOUNT static uint64
BitHammingDistanceAvx512Popcount(uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 distance)
{
__m512i dist = _mm512_setzero_si512();
for (; bytes >= sizeof(__m512i); bytes -= sizeof(__m512i))
{
__m512i axs = _mm512_loadu_si512((const __m512i *) ax);
__m512i bxs = _mm512_loadu_si512((const __m512i *) bx);
dist = _mm512_add_epi64(dist, _mm512_popcnt_epi64(_mm512_xor_si512(axs, bxs)));
ax += sizeof(__m512i);
bx += sizeof(__m512i);
}
distance += _mm512_reduce_add_epi64(dist);
return BitHammingDistanceDefault(bytes, ax, bx, distance);
}
#endif
BIT_TARGET_CLONES static double
BitJaccardDistanceDefault(uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 ab, uint64 aa, uint64 bb)
{
#ifdef popcount64
for (; bytes >= sizeof(uint64); bytes -= sizeof(uint64))
{
uint64 axs;
uint64 bxs;
/* Ensure aligned */
memcpy(&axs, ax, sizeof(uint64));
memcpy(&bxs, bx, sizeof(uint64));
ab += popcount64(axs & bxs);
aa += popcount64(axs);
bb += popcount64(bxs);
ax += sizeof(uint64);
bx += sizeof(uint64);
}
#endif
for (uint32 i = 0; i < bytes; i++)
{
ab += pg_number_of_ones[ax[i] & bx[i]];
aa += pg_number_of_ones[ax[i]];
bb += pg_number_of_ones[bx[i]];
}
if (ab == 0)
return 1;
else
return 1 - (ab / ((double) (aa + bb - ab)));
}
#ifdef BIT_DISPATCH
TARGET_AVX512_POPCOUNT static double
BitJaccardDistanceAvx512Popcount(uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 ab, uint64 aa, uint64 bb)
{
__m512i abx = _mm512_setzero_si512();
__m512i aax = _mm512_setzero_si512();
__m512i bbx = _mm512_setzero_si512();
for (; bytes >= sizeof(__m512i); bytes -= sizeof(__m512i))
{
__m512i axs = _mm512_loadu_si512((const __m512i *) ax);
__m512i bxs = _mm512_loadu_si512((const __m512i *) bx);
abx = _mm512_add_epi64(abx, _mm512_popcnt_epi64(_mm512_and_si512(axs, bxs)));
aax = _mm512_add_epi64(aax, _mm512_popcnt_epi64(axs));
bbx = _mm512_add_epi64(bbx, _mm512_popcnt_epi64(bxs));
ax += sizeof(__m512i);
bx += sizeof(__m512i);
}
ab += _mm512_reduce_add_epi64(abx);
aa += _mm512_reduce_add_epi64(aax);
bb += _mm512_reduce_add_epi64(bbx);
return BitJaccardDistanceDefault(bytes, ax, bx, ab, aa, bb);
}
#endif
#ifdef BIT_DISPATCH
#define CPU_FEATURE_OSXSAVE (1 << 27) /* F1 ECX */
#define CPU_FEATURE_AVX512F (1 << 16) /* F7,0 EBX */
#define CPU_FEATURE_AVX512VPOPCNTDQ (1 << 14) /* F7,0 ECX */
#ifdef _MSC_VER
#define TARGET_XSAVE
#else
#define TARGET_XSAVE __attribute__((target("xsave")))
#endif
TARGET_XSAVE static bool
SupportsAvx512Popcount(void)
{
unsigned int exx[4] = {0, 0, 0, 0};
#if defined(USE__GET_CPUID)
__get_cpuid(1, &exx[0], &exx[1], &exx[2], &exx[3]);
#else
__cpuid(exx, 1);
#endif
/* Check OS supports XSAVE */
if ((exx[2] & CPU_FEATURE_OSXSAVE) != CPU_FEATURE_OSXSAVE)
return false;
/* Check XMM, YMM, and ZMM registers are enabled */
if ((_xgetbv(0) & 0xe6) != 0xe6)
return false;
#if defined(USE__GET_CPUID)
__get_cpuid_count(7, 0, &exx[0], &exx[1], &exx[2], &exx[3]);
#else
__cpuidex(exx, 7, 0);
#endif
/* Check AVX512F */
if ((exx[1] & CPU_FEATURE_AVX512F) != CPU_FEATURE_AVX512F)
return false;
/* Check AVX512VPOPCNTDQ */
return (exx[2] & CPU_FEATURE_AVX512VPOPCNTDQ) == CPU_FEATURE_AVX512VPOPCNTDQ;
}
#endif
void
BitvecInit(void)
{
/*
* Could skip pointer when single function, but no difference in
* performance
*/
BitHammingDistance = BitHammingDistanceDefault;
BitJaccardDistance = BitJaccardDistanceDefault;
#ifdef BIT_DISPATCH
if (SupportsAvx512Popcount())
{
BitHammingDistance = BitHammingDistanceAvx512Popcount;
BitJaccardDistance = BitJaccardDistanceAvx512Popcount;
}
#endif
}

View File

@@ -1,16 +0,0 @@
#ifndef BITUTILS_H
#define BITUTILS_H
#include "postgres.h"
/* Check version in first header */
#if PG_VERSION_NUM < 130000
#error "Requires PostgreSQL 13+"
#endif
extern uint64 (*BitHammingDistance) (uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 distance);
extern double (*BitJaccardDistance) (uint32 bytes, unsigned char *ax, unsigned char *bx, uint64 ab, uint64 aa, uint64 bb);
void BitvecInit(void);
#endif

View File

@@ -1,70 +0,0 @@
#include "postgres.h"
#include "bitutils.h"
#include "bitvec.h"
#include "fmgr.h"
#include "utils/varbit.h"
#include "vector.h"
#if PG_VERSION_NUM >= 160000
#include "varatt.h"
#endif
/*
* Allocate and initialize a new bit vector
*/
VarBit *
InitBitVector(int dim)
{
VarBit *result;
int size;
size = VARBITTOTALLEN(dim);
result = (VarBit *) palloc0(size);
SET_VARSIZE(result, size);
VARBITLEN(result) = dim;
return result;
}
/*
* Ensure same dimensions
*/
static inline void
CheckDims(VarBit *a, VarBit *b)
{
if (VARBITLEN(a) != VARBITLEN(b))
ereport(ERROR,
(errcode(ERRCODE_DATA_EXCEPTION),
errmsg("different bit lengths %u and %u", VARBITLEN(a), VARBITLEN(b))));
}
/*
* Get the Hamming distance between two bit vectors
*/
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hamming_distance);
Datum
hamming_distance(PG_FUNCTION_ARGS)
{
VarBit *a = PG_GETARG_VARBIT_P(0);
VarBit *b = PG_GETARG_VARBIT_P(1);
CheckDims(a, b);
PG_RETURN_FLOAT8((double) BitHammingDistance(VARBITBYTES(a), VARBITS(a), VARBITS(b), 0));
}
/*
* Get the Jaccard distance between two bit vectors
*/
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(jaccard_distance);
Datum
jaccard_distance(PG_FUNCTION_ARGS)
{
VarBit *a = PG_GETARG_VARBIT_P(0);
VarBit *b = PG_GETARG_VARBIT_P(1);
CheckDims(a, b);
PG_RETURN_FLOAT8(BitJaccardDistance(VARBITBYTES(a), VARBITS(a), VARBITS(b), 0, 0, 0));
}

View File

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

View File

@@ -1,300 +0,0 @@
#include "postgres.h"
#include <math.h>
#include "halfutils.h"
#include "halfvec.h"
#ifdef HALFVEC_DISPATCH
#include <immintrin.h>
#if defined(USE__GET_CPUID)
#include <cpuid.h>
#else
#include <intrin.h>
#endif
#ifdef _MSC_VER
#define TARGET_F16C
#else
#define TARGET_F16C __attribute__((target("avx,f16c,fma")))
#endif
#endif
float (*HalfvecL2SquaredDistance) (int dim, half * ax, half * bx);
float (*HalfvecInnerProduct) (int dim, half * ax, half * bx);
double (*HalfvecCosineSimilarity) (int dim, half * ax, half * bx);
float (*HalfvecL1Distance) (int dim, half * ax, half * bx);
static float
HalfvecL2SquaredDistanceDefault(int dim, half * ax, half * bx)
{
float distance = 0.0;
/* Auto-vectorized */
for (int i = 0; i < dim; i++)
{
float diff = HalfToFloat4(ax[i]) - HalfToFloat4(bx[i]);
distance += diff * diff;
}
return distance;
}
#ifdef HALFVEC_DISPATCH
TARGET_F16C static float
HalfvecL2SquaredDistanceF16c(int dim, half * ax, half * bx)
{
float distance;
int i;
float s[8];
int count = (dim / 8) * 8;
__m256 dist = _mm256_setzero_ps();
for (i = 0; i < count; i += 8)
{
__m128i axi = _mm_loadu_si128((__m128i *) (ax + i));
__m128i bxi = _mm_loadu_si128((__m128i *) (bx + i));
__m256 axs = _mm256_cvtph_ps(axi);
__m256 bxs = _mm256_cvtph_ps(bxi);
__m256 diff = _mm256_sub_ps(axs, bxs);
dist = _mm256_fmadd_ps(diff, diff, dist);
}
_mm256_storeu_ps(s, dist);
distance = s[0] + s[1] + s[2] + s[3] + s[4] + s[5] + s[6] + s[7];
for (; i < dim; i++)
{
float diff = HalfToFloat4(ax[i]) - HalfToFloat4(bx[i]);
distance += diff * diff;
}
return distance;
}
#endif
static float
HalfvecInnerProductDefault(int dim, half * ax, half * bx)
{
float distance = 0.0;
/* Auto-vectorized */
for (int i = 0; i < dim; i++)
distance += HalfToFloat4(ax[i]) * HalfToFloat4(bx[i]);
return distance;
}
#ifdef HALFVEC_DISPATCH
TARGET_F16C static float
HalfvecInnerProductF16c(int dim, half * ax, half * bx)
{
float distance;
int i;
float s[8];
int count = (dim / 8) * 8;
__m256 dist = _mm256_setzero_ps();
for (i = 0; i < count; i += 8)
{
__m128i axi = _mm_loadu_si128((__m128i *) (ax + i));
__m128i bxi = _mm_loadu_si128((__m128i *) (bx + i));
__m256 axs = _mm256_cvtph_ps(axi);
__m256 bxs = _mm256_cvtph_ps(bxi);
dist = _mm256_fmadd_ps(axs, bxs, dist);
}
_mm256_storeu_ps(s, dist);
distance = s[0] + s[1] + s[2] + s[3] + s[4] + s[5] + s[6] + s[7];
for (; i < dim; i++)
distance += HalfToFloat4(ax[i]) * HalfToFloat4(bx[i]);
return distance;
}
#endif
static double
HalfvecCosineSimilarityDefault(int dim, half * ax, half * bx)
{
float similarity = 0.0;
float norma = 0.0;
float normb = 0.0;
/* Auto-vectorized */
for (int i = 0; i < dim; i++)
{
float axi = HalfToFloat4(ax[i]);
float bxi = HalfToFloat4(bx[i]);
similarity += axi * bxi;
norma += axi * axi;
normb += bxi * bxi;
}
/* Use sqrt(a * b) over sqrt(a) * sqrt(b) */
return (double) similarity / sqrt((double) norma * (double) normb);
}
#ifdef HALFVEC_DISPATCH
TARGET_F16C static double
HalfvecCosineSimilarityF16c(int dim, half * ax, half * bx)
{
float similarity;
float norma;
float normb;
int i;
float s[8];
int count = (dim / 8) * 8;
__m256 sim = _mm256_setzero_ps();
__m256 na = _mm256_setzero_ps();
__m256 nb = _mm256_setzero_ps();
for (i = 0; i < count; i += 8)
{
__m128i axi = _mm_loadu_si128((__m128i *) (ax + i));
__m128i bxi = _mm_loadu_si128((__m128i *) (bx + i));
__m256 axs = _mm256_cvtph_ps(axi);
__m256 bxs = _mm256_cvtph_ps(bxi);
sim = _mm256_fmadd_ps(axs, bxs, sim);
na = _mm256_fmadd_ps(axs, axs, na);
nb = _mm256_fmadd_ps(bxs, bxs, nb);
}
_mm256_storeu_ps(s, sim);
similarity = s[0] + s[1] + s[2] + s[3] + s[4] + s[5] + s[6] + s[7];
_mm256_storeu_ps(s, na);
norma = s[0] + s[1] + s[2] + s[3] + s[4] + s[5] + s[6] + s[7];
_mm256_storeu_ps(s, nb);
normb = s[0] + s[1] + s[2] + s[3] + s[4] + s[5] + s[6] + s[7];
/* Auto-vectorized */
for (; i < dim; i++)
{
float axi = HalfToFloat4(ax[i]);
float bxi = HalfToFloat4(bx[i]);
similarity += axi * bxi;
norma += axi * axi;
normb += bxi * bxi;
}
/* Use sqrt(a * b) over sqrt(a) * sqrt(b) */
return (double) similarity / sqrt((double) norma * (double) normb);
}
#endif
static float
HalfvecL1DistanceDefault(int dim, half * ax, half * bx)
{
float distance = 0.0;
/* Auto-vectorized */
for (int i = 0; i < dim; i++)
distance += fabsf(HalfToFloat4(ax[i]) - HalfToFloat4(bx[i]));
return distance;
}
#ifdef HALFVEC_DISPATCH
/* Does not require FMA, but keep logic simple */
TARGET_F16C static float
HalfvecL1DistanceF16c(int dim, half * ax, half * bx)
{
float distance;
int i;
float s[8];
int count = (dim / 8) * 8;
__m256 dist = _mm256_setzero_ps();
__m256 sign = _mm256_set1_ps(-0.0);
for (i = 0; i < count; i += 8)
{
__m128i axi = _mm_loadu_si128((__m128i *) (ax + i));
__m128i bxi = _mm_loadu_si128((__m128i *) (bx + i));
__m256 axs = _mm256_cvtph_ps(axi);
__m256 bxs = _mm256_cvtph_ps(bxi);
dist = _mm256_add_ps(dist, _mm256_andnot_ps(sign, _mm256_sub_ps(axs, bxs)));
}
_mm256_storeu_ps(s, dist);
distance = s[0] + s[1] + s[2] + s[3] + s[4] + s[5] + s[6] + s[7];
for (; i < dim; i++)
distance += fabsf(HalfToFloat4(ax[i]) - HalfToFloat4(bx[i]));
return distance;
}
#endif
#ifdef HALFVEC_DISPATCH
#define CPU_FEATURE_FMA (1 << 12)
#define CPU_FEATURE_OSXSAVE (1 << 27)
#define CPU_FEATURE_AVX (1 << 28)
#define CPU_FEATURE_F16C (1 << 29)
#ifdef _MSC_VER
#define TARGET_XSAVE
#else
#define TARGET_XSAVE __attribute__((target("xsave")))
#endif
TARGET_XSAVE static bool
SupportsCpuFeature(unsigned int feature)
{
unsigned int exx[4] = {0, 0, 0, 0};
#if defined(USE__GET_CPUID)
__get_cpuid(1, &exx[0], &exx[1], &exx[2], &exx[3]);
#else
__cpuid(exx, 1);
#endif
/* Check OS supports XSAVE */
if ((exx[2] & CPU_FEATURE_OSXSAVE) != CPU_FEATURE_OSXSAVE)
return false;
/* Check XMM and YMM registers are enabled */
if ((_xgetbv(0) & 6) != 6)
return false;
/* Now check features */
return (exx[2] & feature) == feature;
}
#endif
void
HalfvecInit(void)
{
/*
* Could skip pointer when single function, but no difference in
* performance
*/
HalfvecL2SquaredDistance = HalfvecL2SquaredDistanceDefault;
HalfvecInnerProduct = HalfvecInnerProductDefault;
HalfvecCosineSimilarity = HalfvecCosineSimilarityDefault;
HalfvecL1Distance = HalfvecL1DistanceDefault;
#ifdef HALFVEC_DISPATCH
if (SupportsCpuFeature(CPU_FEATURE_AVX | CPU_FEATURE_F16C | CPU_FEATURE_FMA))
{
HalfvecL2SquaredDistance = HalfvecL2SquaredDistanceF16c;
HalfvecInnerProduct = HalfvecInnerProductF16c;
HalfvecCosineSimilarity = HalfvecCosineSimilarityF16c;
/* Does not require FMA, but keep logic simple */
HalfvecL1Distance = HalfvecL1DistanceF16c;
}
#endif
}

View File

@@ -1,263 +0,0 @@
#ifndef HALFUTILS_H
#define HALFUTILS_H
#include <math.h>
#include "common/shortest_dec.h"
#include "halfvec.h"
#ifdef F16C_SUPPORT
#include <immintrin.h>
#endif
extern float (*HalfvecL2SquaredDistance) (int dim, half * ax, half * bx);
extern float (*HalfvecInnerProduct) (int dim, half * ax, half * bx);
extern double (*HalfvecCosineSimilarity) (int dim, half * ax, half * bx);
extern float (*HalfvecL1Distance) (int dim, half * ax, half * bx);
void HalfvecInit(void);
/*
* Check if half is NaN
*/
static inline bool
HalfIsNan(half num)
{
#ifdef FLT16_SUPPORT
return isnan(num);
#else
return (num & 0x7C00) == 0x7C00 && (num & 0x7FFF) != 0x7C00;
#endif
}
/*
* Check if half is infinite
*/
static inline bool
HalfIsInf(half num)
{
#ifdef FLT16_SUPPORT
return isinf(num);
#else
return (num & 0x7FFF) == 0x7C00;
#endif
}
/*
* Check if half is zero
*/
static inline bool
HalfIsZero(half num)
{
#ifdef FLT16_SUPPORT
return num == 0;
#else
return (num & 0x7FFF) == 0x0000;
#endif
}
/*
* Convert a half to a float4
*/
static inline float
HalfToFloat4(half num)
{
#if defined(F16C_SUPPORT)
return _cvtsh_ss(num);
#elif defined(FLT16_SUPPORT)
return (float) num;
#else
union
{
float f;
uint32 i;
} swapfloat;
union
{
half h;
uint16 i;
} swaphalf;
uint16 bin;
uint32 exponent;
uint32 mantissa;
uint32 result;
swaphalf.h = num;
bin = swaphalf.i;
exponent = (bin & 0x7C00) >> 10;
mantissa = bin & 0x03FF;
/* Sign */
result = (bin & 0x8000) << 16;
if (unlikely(exponent == 31))
{
if (mantissa == 0)
{
/* Infinite */
result |= 0x7F800000;
}
else
{
/* NaN */
result |= 0x7FC00000;
}
}
else if (unlikely(exponent == 0))
{
/* Subnormal */
if (mantissa != 0)
{
exponent = -14;
for (int i = 0; i < 10; i++)
{
mantissa <<= 1;
exponent -= 1;
if ((mantissa >> 10) % 2 == 1)
{
mantissa &= 0x03ff;
break;
}
}
result |= (exponent + 127) << 23;
}
}
else
{
/* Normal */
result |= (exponent - 15 + 127) << 23;
}
result |= mantissa << 13;
swapfloat.i = result;
return swapfloat.f;
#endif
}
/*
* Convert a float4 to a half
*/
static inline half
Float4ToHalfUnchecked(float num)
{
#if defined(F16C_SUPPORT)
return _cvtss_sh(num, 0);
#elif defined(FLT16_SUPPORT)
return (_Float16) num;
#else
union
{
float f;
uint32 i;
} swapfloat;
union
{
half h;
uint16 i;
} swaphalf;
uint32 bin;
int exponent;
int mantissa;
uint16 result;
swapfloat.f = num;
bin = swapfloat.i;
exponent = (bin & 0x7F800000) >> 23;
mantissa = bin & 0x007FFFFF;
/* Sign */
result = (bin & 0x80000000) >> 16;
if (isinf(num))
{
/* Infinite */
result |= 0x7C00;
}
else if (isnan(num))
{
/* NaN */
result |= 0x7E00;
result |= mantissa >> 13;
}
else if (exponent > 98)
{
int m;
int gr;
int s;
exponent -= 127;
s = mantissa & 0x00000FFF;
/* Subnormal */
if (exponent < -14)
{
int diff = -exponent - 14;
mantissa >>= diff;
mantissa += 1 << (23 - diff);
s |= mantissa & 0x00000FFF;
}
m = mantissa >> 13;
/* Round */
gr = (mantissa >> 12) % 4;
if (gr == 3 || (gr == 1 && s != 0))
m += 1;
if (m == 1024)
{
m = 0;
exponent += 1;
}
if (exponent > 15)
{
/* Infinite */
result |= 0x7C00;
}
else
{
if (exponent >= -14)
result |= (exponent + 15) << 10;
result |= m;
}
}
swaphalf.i = result;
return swaphalf.h;
#endif
}
/*
* Convert a float4 to a half
*/
static inline half
Float4ToHalf(float num)
{
half result = Float4ToHalfUnchecked(num);
if (unlikely(HalfIsInf(result)) && !isinf(num))
{
char *buf = palloc(FLOAT_SHORTEST_DECIMAL_LEN);
float_to_shortest_decimal_buf(num, buf);
ereport(ERROR,
(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
errmsg("\"%s\" is out of range for type halfvec", buf)));
}
return result;
}
#endif

File diff suppressed because it is too large Load Diff

View File

@@ -1,70 +0,0 @@
#ifndef HALFVEC_H
#define HALFVEC_H
#define __STDC_WANT_IEC_60559_TYPES_EXT__
#include <float.h>
/* We use two types of dispatching: intrinsics and target_clones */
/* TODO Move to better place */
#ifndef DISABLE_DISPATCH
/* Only enable for more recent compilers to keep build process simple */
#if defined(__x86_64__) && defined(__GNUC__) && __GNUC__ >= 9
#define USE_DISPATCH
#elif defined(__x86_64__) && defined(__clang_major__) && __clang_major__ >= 7
#define USE_DISPATCH
#elif defined(_M_AMD64) && defined(_MSC_VER) && _MSC_VER >= 1920
#define USE_DISPATCH
#endif
#endif
/* target_clones requires glibc */
#if defined(USE_DISPATCH) && defined(__gnu_linux__) && defined(__has_attribute)
/* Use separate line for portability */
#if __has_attribute(target_clones)
#define USE_TARGET_CLONES
#endif
#endif
/* Apple clang check needed for universal binaries on Mac */
#if defined(USE_DISPATCH) && (defined(HAVE__GET_CPUID) || defined(__apple_build_version__))
#define USE__GET_CPUID
#endif
#if defined(USE_DISPATCH)
#define HALFVEC_DISPATCH
#endif
/* F16C has better performance than _Float16 (on x86-64) */
#if defined(__F16C__)
#define F16C_SUPPORT
#elif defined(__FLT16_MAX__) && !defined(HALFVEC_DISPATCH) && !defined(__FreeBSD__) && (!defined(__i386__) || defined(__SSE2__))
#define FLT16_SUPPORT
#endif
#ifdef FLT16_SUPPORT
#define half _Float16
#define HALF_MAX FLT16_MAX
#else
#define half uint16
#define HALF_MAX 65504
#endif
#define HALFVEC_MAX_DIM 16000
#define HALFVEC_SIZE(_dim) (offsetof(HalfVector, x) + sizeof(half)*(_dim))
#define DatumGetHalfVector(x) ((HalfVector *) PG_DETOAST_DATUM(x))
#define PG_GETARG_HALFVEC_P(x) DatumGetHalfVector(PG_GETARG_DATUM(x))
#define PG_RETURN_HALFVEC_P(x) PG_RETURN_POINTER(x)
typedef struct HalfVector
{
int32 vl_len_; /* varlena header (do not touch directly!) */
int16 dim; /* number of dimensions */
int16 unused; /* reserved for future use, always zero */
half x[FLEXIBLE_ARRAY_MEMBER];
} HalfVector;
HalfVector *InitHalfVector(int dim);
#endif

View File

@@ -1,119 +1,50 @@
#include "postgres.h"
#include <float.h>
#include <limits.h>
#include <math.h>
#include "access/amapi.h"
#include "access/genam.h"
#include "access/reloptions.h"
#include "commands/progress.h"
#include "commands/vacuum.h"
#include "fmgr.h"
#include "hnsw.h"
#include "miscadmin.h"
#include "nodes/pg_list.h"
#include "storage/lwlock.h"
#include "utils/float.h"
#include "utils/guc.h"
#include "utils/relcache.h"
#include "utils/selfuncs.h"
#include "utils/spccache.h"
#include "vector.h"
#if PG_VERSION_NUM < 150000
#define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x)
#if PG_VERSION_NUM >= 120000
#include "commands/progress.h"
#endif
static const struct config_enum_entry hnsw_iterative_scan_options[] = {
{"off", HNSW_ITERATIVE_SCAN_OFF, false},
{"relaxed_order", HNSW_ITERATIVE_SCAN_RELAXED, false},
{"strict_order", HNSW_ITERATIVE_SCAN_STRICT, false},
{NULL, 0, false}
};
int hnsw_ef_search;
int hnsw_iterative_scan;
int hnsw_max_scan_tuples;
double hnsw_scan_mem_multiplier;
int hnsw_lock_tranche_id;
static relopt_kind hnsw_relopt_kind;
/*
* Assign a tranche ID for our LWLocks. This only needs to be done by one
* backend, as the tranche ID is remembered in shared memory.
*
* This shared memory area is very small, so we just allocate it from the
* "slop" that PostgreSQL reserves for small allocations like this. If
* this grows bigger, we should use a shmem_request_hook and
* RequestAddinShmemSpace() to pre-reserve space for this.
*/
void
HnswInitLockTranche(void)
{
int *tranche_ids;
bool found;
LWLockAcquire(AddinShmemInitLock, LW_EXCLUSIVE);
tranche_ids = ShmemInitStruct("hnsw LWLock ids",
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
}
/*
* Initialize index options and variables
*/
void
HnswInit(void)
{
if (!process_shared_preload_libraries_in_progress)
HnswInitLockTranche();
hnsw_relopt_kind = add_reloption_kind();
add_int_reloption(hnsw_relopt_kind, "m", "Max number of connections",
HNSW_DEFAULT_M, HNSW_MIN_M, HNSW_MAX_M, AccessExclusiveLock);
HNSW_DEFAULT_M, HNSW_MIN_M, HNSW_MAX_M
#if PG_VERSION_NUM >= 130000
,AccessExclusiveLock
#endif
);
add_int_reloption(hnsw_relopt_kind, "ef_construction", "Size of the dynamic candidate list for construction",
HNSW_DEFAULT_EF_CONSTRUCTION, HNSW_MIN_EF_CONSTRUCTION, HNSW_MAX_EF_CONSTRUCTION, AccessExclusiveLock);
HNSW_DEFAULT_EF_CONSTRUCTION, HNSW_MIN_EF_CONSTRUCTION, HNSW_MAX_EF_CONSTRUCTION
#if PG_VERSION_NUM >= 130000
,AccessExclusiveLock
#endif
);
DefineCustomIntVariable("hnsw.ef_search", "Sets the size of the dynamic candidate list for search",
"Valid range is 1..1000.", &hnsw_ef_search,
HNSW_DEFAULT_EF_SEARCH, HNSW_MIN_EF_SEARCH, HNSW_MAX_EF_SEARCH, PGC_USERSET, 0, NULL, NULL, NULL);
DefineCustomEnumVariable("hnsw.iterative_scan", "Sets the mode for iterative scans",
NULL, &hnsw_iterative_scan,
HNSW_ITERATIVE_SCAN_OFF, hnsw_iterative_scan_options, PGC_USERSET, 0, NULL, NULL, NULL);
/* This is approximate and does not affect the initial scan */
DefineCustomIntVariable("hnsw.max_scan_tuples", "Sets the max number of tuples to visit for iterative scans",
NULL, &hnsw_max_scan_tuples,
20000, 1, INT_MAX, PGC_USERSET, 0, NULL, NULL, NULL);
/* Same range as hash_mem_multiplier */
DefineCustomRealVariable("hnsw.scan_mem_multiplier", "Sets the multiple of work_mem to use for iterative scans",
NULL, &hnsw_scan_mem_multiplier,
1, 1, 1000, PGC_USERSET, 0, NULL, NULL, NULL);
MarkGUCPrefixReserved("hnsw");
}
/*
* Get the name of index build phase
*/
#if PG_VERSION_NUM >= 120000
static char *
hnswbuildphasename(int64 phasenum)
{
@@ -127,6 +58,7 @@ hnswbuildphasename(int64 phasenum)
return NULL;
}
}
#endif
/*
* Estimate the cost of an index scan
@@ -139,93 +71,45 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
{
GenericCosts costs;
int m;
double ratio;
double startupPages;
double spc_seq_page_cost;
int entryLevel;
Relation index;
#if PG_VERSION_NUM < 120000
List *qinfos;
#endif
/* Never use index without order */
if (path->indexorderbys == NIL)
if (path->indexorderbys == NULL)
{
*indexStartupCost = get_float8_infinity();
*indexTotalCost = get_float8_infinity();
*indexStartupCost = DBL_MAX;
*indexTotalCost = DBL_MAX;
*indexSelectivity = 0;
*indexCorrelation = 0;
*indexPages = 0;
#if PG_VERSION_NUM >= 180000
/* See "On disable_cost" thread on pgsql-hackers */
path->path.disabled_nodes = 2;
#endif
return;
}
MemSet(&costs, 0, sizeof(costs));
genericcostestimate(root, path, loop_count, &costs);
index = index_open(path->indexinfo->indexoid, NoLock);
HnswGetMetaPageInfo(index, &m, NULL);
index_close(index, NoLock);
/*
* HNSW cost estimation follows a formula that accounts for the total
* number of tuples indexed combined with the parameters that most
* influence the duration of the index scan, namely: m - the number of
* tuples that are scanned in each step of the HNSW graph traversal
* ef_search - which influences the total number of steps taken at layer 0
*
* The source of the vector data can impact how many steps it takes to
* converge on the set of vectors to return to the executor. Currently, we
* use a hardcoded scaling factor (HNSWScanScalingFactor) to help
* influence that, but this could later become a configurable parameter
* based on the cost estimations.
*
* The tuple estimator formula is below:
*
* numIndexTuples = entryLevel * m + layer0TuplesMax * layer0Selectivity
*
* "entryLevel * m" represents the floor of tuples we need to scan to get
* to layer 0 (L0).
*
* "layer0TuplesMax" is the estimated total number of tuples we'd scan at
* L0 if we weren't discarding already visited tuples as part of the scan.
*
* "layer0Selectivity" estimates the percentage of tuples that are scanned
* at L0, accounting for previously visited tuples, multiplied by the
* "scalingFactor" (currently hardcoded).
*/
if (path->indexinfo->tuples > 0)
{
double scalingFactor = 0.55;
int entryLevel = (int) (log(path->indexinfo->tuples) * HnswGetMl(m));
int layer0TuplesMax = HnswGetLayerM(m, 0) * hnsw_ef_search;
double layer0Selectivity = scalingFactor * log(path->indexinfo->tuples) / (log(m) * (1 + log(hnsw_ef_search)));
/* Approximate entry level */
entryLevel = (int) -log(1.0 / path->indexinfo->tuples) * HnswGetMl(m);
ratio = (entryLevel * m + layer0TuplesMax * layer0Selectivity) / path->indexinfo->tuples;
/* TODO Improve estimate of visited tuples (currently underestimates) */
/* Account for number of tuples (or entry level), m, and ef_search */
costs.numIndexTuples = (entryLevel + 2) * m;
if (ratio > 1)
ratio = 1;
}
else
ratio = 1;
#if PG_VERSION_NUM >= 120000
genericcostestimate(root, path, loop_count, &costs);
#else
qinfos = deconstruct_indexquals(path);
genericcostestimate(root, path, loop_count, qinfos, &costs);
#endif
get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
/* Startup cost is cost before returning the first row */
costs.indexStartupCost = costs.indexTotalCost * ratio;
/* Adjust cost if needed since TOAST not included in seq scan cost */
startupPages = costs.numIndexPages * ratio;
if (startupPages > path->indexinfo->rel->pages && ratio < 0.5)
{
/* Change all page cost from random to sequential */
costs.indexStartupCost -= startupPages * (costs.spc_random_page_cost - spc_seq_page_cost);
/* Remove cost of extra pages */
costs.indexStartupCost -= (startupPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
}
*indexStartupCost = costs.indexStartupCost;
/* Use total cost since most work happens before first tuple is returned */
*indexStartupCost = costs.indexTotalCost;
*indexTotalCost = costs.indexTotalCost;
*indexSelectivity = costs.indexSelectivity;
*indexCorrelation = costs.indexCorrelation;
@@ -243,10 +127,23 @@ hnswoptions(Datum reloptions, bool validate)
{"ef_construction", RELOPT_TYPE_INT, offsetof(HnswOptions, efConstruction)},
};
#if PG_VERSION_NUM >= 130000
return (bytea *) build_reloptions(reloptions, validate,
hnsw_relopt_kind,
sizeof(HnswOptions),
tab, lengthof(tab));
#else
relopt_value *options;
int numoptions;
HnswOptions *rdopts;
options = parseRelOptions(reloptions, validate, hnsw_relopt_kind, &numoptions);
rdopts = allocateReloptStruct(sizeof(HnswOptions), options, numoptions);
fillRelOptions((void *) rdopts, sizeof(HnswOptions), options, numoptions,
validate, tab, lengthof(tab));
return (bytea *) rdopts;
#endif
}
/*
@@ -258,85 +155,33 @@ hnswvalidate(Oid opclassoid)
return true;
}
/*
* Checks if index-only scan is supported
*/
static bool
hnswcanreturn(Relation indexRelation, int attno)
{
return attno == 1 && !OidIsValid(index_getprocid(indexRelation, 1, HNSW_NORM_PROC));
}
/*
* Define index handler
*
* See https://www.postgresql.org/docs/current/index-api.html
*/
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnswhandler);
PGDLLEXPORT PG_FUNCTION_INFO_V1(hnswhandler);
Datum
hnswhandler(PG_FUNCTION_ARGS)
{
#if PG_VERSION_NUM >= 190000
static const IndexAmRoutine amroutine = {
.type = T_IndexAmRoutine,
.amstrategies = 0,
.amsupport = 3,
.amoptsprocnum = 0,
.amcanorder = false,
.amcanorderbyop = true,
.amcanhash = false,
.amconsistentequality = false,
.amconsistentordering = false,
.amcanbackward = false,
.amcanunique = false,
.amcanmulticol = false,
.amoptionalkey = true,
.amsearcharray = false,
.amsearchnulls = false,
.amstorage = false,
.amclusterable = false,
.ampredlocks = false,
.amcanparallel = false,
.amcanbuildparallel = true,
.amcaninclude = false,
.amusemaintenanceworkmem = false,
.amsummarizing = false,
.amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL,
.amkeytype = InvalidOid,
.ambuild = hnswbuild,
.ambuildempty = hnswbuildempty,
.aminsert = hnswinsert,
.aminsertcleanup = NULL,
.ambulkdelete = hnswbulkdelete,
.amvacuumcleanup = hnswvacuumcleanup,
.amcanreturn = NULL,
.amcostestimate = hnswcostestimate,
.amgettreeheight = NULL,
.amoptions = hnswoptions,
.amproperty = NULL,
.ambuildphasename = hnswbuildphasename,
.amvalidate = hnswvalidate,
.amadjustmembers = NULL,
.ambeginscan = hnswbeginscan,
.amrescan = hnswrescan,
.amgettuple = hnswgettuple,
.amgetbitmap = NULL,
.amendscan = hnswendscan,
.ammarkpos = NULL,
.amrestrpos = NULL,
.amestimateparallelscan = NULL,
.aminitparallelscan = NULL,
.amparallelrescan = NULL,
.amtranslatestrategy = NULL,
.amtranslatecmptype = NULL,
};
PG_RETURN_POINTER(&amroutine);
#else
IndexAmRoutine *amroutine = makeNode(IndexAmRoutine);
amroutine->amstrategies = 0;
amroutine->amsupport = 3;
amroutine->amsupport = 2;
#if PG_VERSION_NUM >= 130000
amroutine->amoptsprocnum = 0;
#endif
amroutine->amcanorder = false;
amroutine->amcanorderbyop = true;
#if PG_VERSION_NUM >= 180000
amroutine->amcanhash = false;
amroutine->amconsistentequality = false;
amroutine->amconsistentordering = false;
#endif
amroutine->amcanbackward = false; /* can change direction mid-scan */
amroutine->amcanunique = false;
amroutine->amcanmulticol = false;
@@ -347,34 +192,26 @@ hnswhandler(PG_FUNCTION_ARGS)
amroutine->amclusterable = false;
amroutine->ampredlocks = false;
amroutine->amcanparallel = false;
#if PG_VERSION_NUM >= 170000
amroutine->amcanbuildparallel = true;
#endif
amroutine->amcaninclude = false;
#if PG_VERSION_NUM >= 130000
amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
#if PG_VERSION_NUM >= 160000
amroutine->amsummarizing = false;
#endif
amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
#endif
amroutine->amkeytype = InvalidOid;
/* Interface functions */
amroutine->ambuild = hnswbuild;
amroutine->ambuildempty = hnswbuildempty;
amroutine->aminsert = hnswinsert;
#if PG_VERSION_NUM >= 170000
amroutine->aminsertcleanup = NULL;
#endif
amroutine->ambulkdelete = hnswbulkdelete;
amroutine->amvacuumcleanup = hnswvacuumcleanup;
amroutine->amcanreturn = NULL;
amroutine->amcanreturn = hnswcanreturn;
amroutine->amcostestimate = hnswcostestimate;
#if PG_VERSION_NUM >= 180000
amroutine->amgettreeheight = NULL;
#endif
amroutine->amoptions = hnswoptions;
amroutine->amproperty = NULL; /* TODO AMPROP_DISTANCE_ORDERABLE */
#if PG_VERSION_NUM >= 120000
amroutine->ambuildphasename = hnswbuildphasename;
#endif
amroutine->amvalidate = hnswvalidate;
#if PG_VERSION_NUM >= 140000
amroutine->amadjustmembers = NULL;
@@ -392,11 +229,5 @@ hnswhandler(PG_FUNCTION_ARGS)
amroutine->aminitparallelscan = NULL;
amroutine->amparallelrescan = NULL;
#if PG_VERSION_NUM >= 180000
amroutine->amtranslatestrategy = NULL;
amroutine->amtranslatecmptype = NULL;
#endif
PG_RETURN_POINTER(amroutine);
#endif
}

View File

@@ -3,36 +3,22 @@
#include "postgres.h"
#include <math.h>
#include "access/genam.h"
#include "access/parallel.h"
#include "lib/pairingheap.h"
#include "access/generic_xlog.h"
#include "access/reloptions.h"
#include "nodes/execnodes.h"
#include "port.h" /* for random() */
#include "storage/bufpage.h"
#include "storage/condition_variable.h"
#include "storage/lwlock.h"
#include "storage/s_lock.h"
#include "utils/relptr.h"
#include "utils/sampling.h"
#include "vector.h"
#ifdef HNSW_BENCH
#include "portability/instr_time.h"
#endif
#if PG_VERSION_NUM >= 190000
typedef Pointer Item;
#if PG_VERSION_NUM < 110000
#error "Requires PostgreSQL 11+"
#endif
#define HNSW_MAX_DIM 2000
#define HNSW_MAX_NNZ 1000
/* Support functions */
#define HNSW_DISTANCE_PROC 1
#define HNSW_NORM_PROC 2
#define HNSW_TYPE_INFO_PROC 3
#define HNSW_VERSION 1
#define HNSW_MAGIC_NUMBER 0xA953A953
@@ -72,37 +58,22 @@ typedef Pointer Item;
#define PROGRESS_HNSW_PHASE_LOAD 2
#define HNSW_MAX_SIZE (BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - sizeof(ItemIdData))
#define HNSW_TUPLE_ALLOC_SIZE BLCKSZ
#define HNSW_ELEMENT_TUPLE_SIZE(size) MAXALIGN(offsetof(HnswElementTupleData, data) + (size))
#define HNSW_ELEMENT_TUPLE_SIZE(_dim) MAXALIGN(offsetof(HnswElementTupleData, vec) + VECTOR_SIZE(_dim))
#define HNSW_NEIGHBOR_TUPLE_SIZE(level, m) MAXALIGN(offsetof(HnswNeighborTupleData, indextids) + ((level) + 2) * (m) * sizeof(ItemPointerData))
#define HNSW_NEIGHBOR_ARRAY_SIZE(lm) (offsetof(HnswNeighborArray, items) + sizeof(HnswCandidate) * (lm))
#define HnswPageGetOpaque(page) ((HnswPageOpaque) PageGetSpecialPointer(page))
#define HnswPageGetMeta(page) ((HnswMetaPageData *) PageGetContents(page))
#ifdef HNSW_BENCH
#define HnswBench(name, code) \
do { \
instr_time start; \
instr_time duration; \
INSTR_TIME_SET_CURRENT(start); \
(code); \
INSTR_TIME_SET_CURRENT(duration); \
INSTR_TIME_SUBTRACT(duration, start); \
elog(INFO, "%s: %.3f ms", name, INSTR_TIME_GET_MILLISEC(duration)); \
} while (0)
#else
#define HnswBench(name, code) (code)
#endif
#if PG_VERSION_NUM >= 150000
#define RandomDouble() pg_prng_double(&pg_global_prng_state)
#define SeedRandom(seed) pg_prng_seed(&pg_global_prng_state, seed)
#else
#define RandomDouble() (((double) random()) / MAX_RANDOM_VALUE)
#define SeedRandom(seed) srandom(seed)
#endif
#if PG_VERSION_NUM < 130000
#define list_delete_last(list) list_truncate(list, list_length(list) - 1)
#define list_sort(list, cmp) list_qsort(list, cmp)
#endif
#define HnswIsElementTuple(tup) ((tup)->type == HNSW_ELEMENT_TUPLE_TYPE)
@@ -115,96 +86,47 @@ typedef Pointer Item;
#define HnswGetMl(m) (1 / log(m))
/* Ensure fits on page and in uint8 */
#define HnswGetMaxLevel(m) Min(((BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - offsetof(HnswNeighborTupleData, indextids) - sizeof(ItemIdData)) / (sizeof(ItemPointerData)) / (m)) - 2, 255)
#define HnswGetSearchCandidate(membername, ptr) pairingheap_container(HnswSearchCandidate, membername, ptr)
#define HnswGetSearchCandidateConst(membername, ptr) pairingheap_const_container(HnswSearchCandidate, membername, ptr)
#define HnswGetValue(base, element) PointerGetDatum(HnswPtrAccess(base, (element)->value))
#if PG_VERSION_NUM < 140005
#define relptr_offset(rp) ((rp).relptr_off - 1)
#endif
/* Pointer macros */
#define HnswPtrAccess(base, hp) ((base) == NULL ? (hp).ptr : relptr_access(base, (hp).relptr))
#define HnswPtrStore(base, hp, value) ((base) == NULL ? (void) ((hp).ptr = (value)) : (void) relptr_store(base, (hp).relptr, value))
#define HnswPtrIsNull(base, hp) ((base) == NULL ? (hp).ptr == NULL : relptr_is_null((hp).relptr))
#define HnswPtrEqual(base, hp1, hp2) ((base) == NULL ? (hp1).ptr == (hp2).ptr : relptr_offset((hp1).relptr) == relptr_offset((hp2).relptr))
/* For code paths dedicated to each type */
#define HnswPtrPointer(hp) (hp).ptr
#define HnswPtrOffset(hp) relptr_offset((hp).relptr)
#define HnswGetMaxLevel(m) Min(((BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData)) - offsetof(HnswNeighborTupleData, indextids) - sizeof(ItemIdData)) / (sizeof(ItemPointerData)) / m) - 2, 255)
/* Variables */
extern int hnsw_ef_search;
extern int hnsw_iterative_scan;
extern int hnsw_max_scan_tuples;
extern double hnsw_scan_mem_multiplier;
extern int hnsw_lock_tranche_id;
typedef enum HnswIterativeScanMode
{
HNSW_ITERATIVE_SCAN_OFF,
HNSW_ITERATIVE_SCAN_RELAXED,
HNSW_ITERATIVE_SCAN_STRICT
} HnswIterativeScanMode;
typedef struct HnswElementData HnswElementData;
typedef struct HnswNeighborArray HnswNeighborArray;
#define HnswPtrDeclare(type, relptrtype, ptrtype) \
relptr_declare(type, relptrtype); \
typedef union { type *ptr; relptrtype relptr; } ptrtype
/* Pointers that can be absolute or relative */
/* Use char for DatumPtr so works with Pointer */
HnswPtrDeclare(HnswElementData, HnswElementRelptr, HnswElementPtr);
HnswPtrDeclare(HnswNeighborArray, HnswNeighborArrayRelptr, HnswNeighborArrayPtr);
HnswPtrDeclare(HnswNeighborArrayPtr, HnswNeighborsRelptr, HnswNeighborsPtr);
HnswPtrDeclare(char, DatumRelptr, DatumPtr);
struct HnswElementData
typedef struct HnswElementData
{
HnswElementPtr next;
ItemPointerData heaptids[HNSW_HEAPTIDS];
uint8 heaptidsLength;
List *heaptids;
uint8 level;
uint8 deleted;
uint8 version;
uint32 hash;
HnswNeighborsPtr neighbors;
HnswNeighborArray *neighbors;
BlockNumber blkno;
OffsetNumber offno;
OffsetNumber neighborOffno;
BlockNumber neighborPage;
DatumPtr value;
LWLock lock;
};
Vector *vec;
} HnswElementData;
typedef HnswElementData * HnswElement;
typedef struct HnswCandidate
{
HnswElementPtr element;
HnswElement element;
float distance;
bool closer;
} HnswCandidate;
struct HnswNeighborArray
typedef struct HnswNeighborArray
{
int length;
bool closerSet;
HnswCandidate items[FLEXIBLE_ARRAY_MEMBER];
};
HnswCandidate *items;
} HnswNeighborArray;
typedef struct HnswSearchCandidate
typedef struct HnswPairingHeapNode
{
pairingheap_node c_node;
pairingheap_node w_node;
HnswElementPtr element;
double distance;
} HnswSearchCandidate;
pairingheap_node ph_node;
HnswCandidate *inner;
} HnswPairingHeapNode;
/* HNSW index options */
typedef struct HnswOptions
@@ -214,84 +136,6 @@ typedef struct HnswOptions
int efConstruction; /* size of dynamic candidate list */
} HnswOptions;
typedef struct HnswGraph
{
/* Graph state */
slock_t lock;
HnswElementPtr head;
double indtuples;
/* Entry state */
LWLock entryLock;
LWLock entryWaitLock;
HnswElementPtr entryPoint;
/* Allocations state */
LWLock allocatorLock;
Size memoryUsed;
Size memoryTotal;
/* Flushed state */
LWLock flushLock;
bool flushed;
} HnswGraph;
typedef struct HnswShared
{
/* Immutable state */
Oid heaprelid;
Oid indexrelid;
bool isconcurrent;
/* Worker progress */
ConditionVariable workersdonecv;
/* Mutex for mutable state */
slock_t mutex;
/* Mutable state */
int nparticipantsdone;
double reltuples;
HnswGraph graphData;
} HnswShared;
#define ParallelTableScanFromHnswShared(shared) \
(ParallelTableScanDesc) ((char *) (shared) + BUFFERALIGN(sizeof(HnswShared)))
typedef struct HnswLeader
{
ParallelContext *pcxt;
int nparticipanttuplesorts;
HnswShared *hnswshared;
Snapshot snapshot;
char *hnswarea;
} HnswLeader;
typedef struct HnswAllocator
{
void *(*alloc) (Size size, void *state);
void *state;
} HnswAllocator;
typedef struct HnswTypeInfo
{
int maxDimensions;
Datum (*normalize) (PG_FUNCTION_ARGS);
void (*checkValue) (Pointer v);
} HnswTypeInfo;
typedef struct HnswSupport
{
FmgrInfo *procinfo;
FmgrInfo *normprocinfo;
Oid collation;
} HnswSupport;
typedef struct HnswQuery
{
Datum value;
} HnswQuery;
typedef struct HnswBuildState
{
/* Info */
@@ -299,7 +143,6 @@ typedef struct HnswBuildState
Relation index;
IndexInfo *indexInfo;
ForkNumber forkNum;
const HnswTypeInfo *typeInfo;
/* Settings */
int dimensions;
@@ -311,23 +154,21 @@ typedef struct HnswBuildState
double reltuples;
/* Support functions */
HnswSupport support;
FmgrInfo *procinfo;
FmgrInfo *normprocinfo;
Oid collation;
/* Variables */
HnswGraph graphData;
HnswGraph *graph;
List *elements;
HnswElement entryPoint;
double ml;
int maxLevel;
double maxInMemoryElements;
bool flushed;
Vector *normvec;
/* Memory */
MemoryContext graphCtx;
MemoryContext tmpCtx;
HnswAllocator allocator;
/* Parallel builds */
HnswLeader *hnswleader;
HnswShared *hnswshared;
char *hnswarea;
} HnswBuildState;
typedef struct HnswMetaPageData
@@ -359,11 +200,11 @@ typedef struct HnswElementTupleData
uint8 type;
uint8 level;
uint8 deleted;
uint8 version;
uint8 unused;
ItemPointerData heaptids[HNSW_HEAPTIDS];
ItemPointerData neighbortid;
uint16 unused;
Vector data;
uint16 unused2;
Vector vec;
} HnswElementTupleData;
typedef HnswElementTupleData * HnswElementTuple;
@@ -371,42 +212,23 @@ typedef HnswElementTupleData * HnswElementTuple;
typedef struct HnswNeighborTupleData
{
uint8 type;
uint8 version;
uint8 unused;
uint16 count;
ItemPointerData indextids[FLEXIBLE_ARRAY_MEMBER];
} HnswNeighborTupleData;
typedef HnswNeighborTupleData * HnswNeighborTuple;
typedef union
{
struct pointerhash_hash *pointers;
struct offsethash_hash *offsets;
struct tidhash_hash *tids;
} visited_hash;
typedef union
{
HnswElement element;
ItemPointerData indextid;
} HnswUnvisited;
typedef struct HnswScanOpaqueData
{
const HnswTypeInfo *typeInfo;
bool first;
List *w;
visited_hash v;
pairingheap *discarded;
HnswQuery q;
int m;
int64 tuples;
double previousDistance;
Size maxMemory;
MemoryContext tmpCtx;
/* Support functions */
HnswSupport support;
FmgrInfo *procinfo;
FmgrInfo *normprocinfo;
Oid collation;
} HnswScanOpaqueData;
typedef HnswScanOpaqueData * HnswScanOpaque;
@@ -424,14 +246,14 @@ typedef struct HnswVacuumState
int efConstruction;
/* Support functions */
HnswSupport support;
FmgrInfo *procinfo;
Oid collation;
/* Variables */
struct tidhash_hash *deleted;
HTAB *deleted;
BufferAccessStrategy bas;
HnswNeighborTuple ntup;
HnswElementData highestPoint;
HnswElementData fallbackPoint;
/* Memory */
MemoryContext tmpCtx;
@@ -441,36 +263,32 @@ typedef struct HnswVacuumState
int HnswGetM(Relation index);
int HnswGetEfConstruction(Relation index);
FmgrInfo *HnswOptionalProcInfo(Relation index, uint16 procnum);
void HnswInitSupport(HnswSupport * support, Relation index);
Datum HnswNormValue(const HnswTypeInfo * typeInfo, Oid collation, Datum value);
bool HnswCheckNorm(HnswSupport * support, Datum value);
bool HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
void HnswCommitBuffer(Buffer buf, GenericXLogState *state);
Buffer HnswNewBuffer(Relation index, ForkNumber forkNum);
void HnswInitPage(Buffer buf, Page page);
void HnswInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state);
void HnswInit(void);
List *HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples);
List *HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, int m, bool loadVec, HnswElement skipElement);
HnswElement HnswGetEntryPoint(Relation index);
void HnswGetMetaPageInfo(Relation index, int *m, HnswElement * entryPoint);
void *HnswAlloc(HnswAllocator * allocator, Size size);
HnswElement HnswInitElement(char *base, ItemPointer tid, int m, double ml, int maxLevel, HnswAllocator * alloc);
HnswElement HnswInitElement(ItemPointer tid, int m, double ml, int maxLevel);
void HnswFreeElement(HnswElement element);
HnswElement HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno);
void HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing);
HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement entryPoint, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec);
void HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum, bool building);
void HnswSetNeighborTuple(char *base, HnswNeighborTuple ntup, HnswElement e, int m);
void HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, bool existing);
HnswElement HnswFindDuplicate(HnswElement e);
HnswCandidate *HnswEntryCandidate(HnswElement em, Datum q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadVec);
void HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum);
void HnswSetNeighborTuple(HnswNeighborTuple ntup, HnswElement e, int m);
void HnswAddHeapTid(HnswElement element, ItemPointer heaptid);
HnswNeighborArray *HnswInitNeighborArray(int lm, HnswAllocator * allocator);
void HnswInitNeighbors(char *base, HnswElement element, int m, HnswAllocator * alloc);
bool HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPointer heaptid, bool building);
void HnswUpdateNeighborsOnDisk(Relation index, HnswSupport * support, HnswElement e, int m, bool checkExisting, bool building);
void HnswInitNeighbors(HnswElement element, int m);
bool HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel);
void HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting);
void HnswLoadElementFromTuple(HnswElement element, HnswElementTuple etup, bool loadHeaptids, bool loadVec);
void HnswLoadElement(HnswElement element, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance);
bool HnswFormIndexValue(Datum *out, Datum *values, bool *isnull, const HnswTypeInfo * typeInfo, HnswSupport * support);
void HnswSetElementTuple(char *base, HnswElementTuple etup, HnswElement element);
void HnswUpdateConnection(char *base, HnswNeighborArray * neighbors, HnswElement newElement, float distance, int lm, int *updateIdx, Relation index, HnswSupport * support);
bool HnswLoadNeighborTids(HnswElement element, ItemPointerData *indextids, Relation index, int m, int lm, int lc);
void HnswInitLockTranche(void);
const HnswTypeInfo *HnswGetTypeInfo(Relation index);
PGDLLEXPORT void HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc);
void HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec);
void HnswSetElementTuple(HnswElementTuple etup, HnswElement element);
void HnswUpdateConnection(HnswElement element, HnswCandidate * hc, int m, int lc, int *updateIdx, Relation index, FmgrInfo *procinfo, Oid collation);
void HnswLoadNeighbors(HnswElement element, Relation index, int m);
/* Index access methods */
IndexBuildResult *hnswbuild(Relation heap, Relation index, IndexInfo *indexInfo);
@@ -488,54 +306,4 @@ void hnswrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys,
bool hnswgettuple(IndexScanDesc scan, ScanDirection dir);
void hnswendscan(IndexScanDesc scan);
static inline HnswNeighborArray *
HnswGetNeighbors(char *base, HnswElement element, int lc)
{
HnswNeighborArrayPtr *neighborList = HnswPtrAccess(base, element->neighbors);
Assert(element->level >= lc);
return HnswPtrAccess(base, neighborList[lc]);
}
/* Hash tables */
typedef struct TidHashEntry
{
ItemPointerData tid;
char status;
} TidHashEntry;
#define SH_PREFIX tidhash
#define SH_ELEMENT_TYPE TidHashEntry
#define SH_KEY_TYPE ItemPointerData
#define SH_SCOPE extern
#define SH_DECLARE
#include "lib/simplehash.h"
typedef struct PointerHashEntry
{
uintptr_t ptr;
char status;
} PointerHashEntry;
#define SH_PREFIX pointerhash
#define SH_ELEMENT_TYPE PointerHashEntry
#define SH_KEY_TYPE uintptr_t
#define SH_SCOPE extern
#define SH_DECLARE
#include "lib/simplehash.h"
typedef struct OffsetHashEntry
{
Size offset;
char status;
} OffsetHashEntry;
#define SH_PREFIX offsethash
#define SH_ELEMENT_TYPE OffsetHashEntry
#define SH_KEY_TYPE Size
#define SH_SCOPE extern
#define SH_DECLARE
#include "lib/simplehash.h"
#endif

File diff suppressed because it is too large Load Diff

View File

@@ -1,19 +1,11 @@
#include "postgres.h"
#include "access/genam.h"
#include "access/generic_xlog.h"
#include <math.h>
#include "hnsw.h"
#include "nodes/execnodes.h"
#include "storage/bufmgr.h"
#include "storage/lmgr.h"
#include "storage/lwlock.h"
#include "utils/datum.h"
#include "utils/memutils.h"
#include "utils/rel.h"
#if PG_VERSION_NUM >= 160000
#include "varatt.h"
#endif
/*
* Get the insert page
@@ -42,15 +34,14 @@ GetInsertPage(Relation index)
* Check for a free offset
*/
static bool
HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size etupSize, Size ntupSize, Buffer *nbuf, Page *npage, OffsetNumber *freeOffno, OffsetNumber *freeNeighborOffno, BlockNumber *newInsertPage, uint8 *tupleVersion)
HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size ntupSize, Buffer *nbuf, Page *npage, OffsetNumber *freeOffno, OffsetNumber *freeNeighborOffno, BlockNumber *newInsertPage)
{
OffsetNumber offno;
OffsetNumber maxoffno = PageGetMaxOffsetNumber(page);
for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
{
ItemId eitemid = PageGetItemId(page, offno);
HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, eitemid);
HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
/* Skip neighbor tuples */
if (!HnswIsElementTuple(etup))
@@ -61,9 +52,7 @@ HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size
BlockNumber elementPage = BufferGetBlockNumber(buf);
BlockNumber neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
OffsetNumber neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
ItemId nitemid;
Size pageFree;
Size npageFree;
ItemId itemid;
if (!BlockNumberIsValid(*newInsertPage))
*newInsertPage = elementPage;
@@ -82,29 +71,13 @@ HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size
*npage = BufferGetPage(*nbuf);
}
nitemid = PageGetItemId(*npage, neighborOffno);
itemid = PageGetItemId(*npage, neighborOffno);
/* Ensure aligned for space check */
Assert(etupSize == MAXALIGN(etupSize));
Assert(ntupSize == MAXALIGN(ntupSize));
/*
* Calculate free space individually since tuples are overwritten
* individually (in separate calls to PageIndexTupleOverwrite)
*/
pageFree = ItemIdGetLength(eitemid) + PageGetExactFreeSpace(page);
npageFree = ItemIdGetLength(nitemid);
if (neighborPage != elementPage)
npageFree += PageGetExactFreeSpace(*npage);
else if (pageFree >= etupSize)
npageFree += pageFree - etupSize;
/* Check for space */
if (pageFree >= etupSize && npageFree >= ntupSize)
/* Check for space on neighbor tuple page */
if (PageGetFreeSpace(*npage) + ItemIdGetLength(itemid) - sizeof(ItemIdData) >= ntupSize)
{
*freeOffno = offno;
*freeNeighborOffno = neighborOffno;
*tupleVersion = etup->version;
return true;
}
else if (*nbuf != buf)
@@ -119,7 +92,7 @@ HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size
* Add a new page
*/
static void
HnswInsertAppendPage(Relation index, Buffer *nbuf, Page *npage, GenericXLogState *state, Page page, bool building)
HnswInsertAppendPage(Relation index, Buffer *nbuf, Page *npage, GenericXLogState *state, Page page)
{
/* Add a new page */
LockRelationForExtension(index, ExclusiveLock);
@@ -127,11 +100,7 @@ HnswInsertAppendPage(Relation index, Buffer *nbuf, Page *npage, GenericXLogState
UnlockRelationForExtension(index, ExclusiveLock);
/* Init new page */
if (building)
*npage = BufferGetPage(*nbuf);
else
*npage = GenericXLogRegisterBuffer(state, *nbuf, GENERIC_XLOG_FULL_IMAGE);
*npage = GenericXLogRegisterBuffer(state, *nbuf, GENERIC_XLOG_FULL_IMAGE);
HnswInitPage(*nbuf, *npage);
/* Update previous buffer */
@@ -142,7 +111,7 @@ HnswInsertAppendPage(Relation index, Buffer *nbuf, Page *npage, GenericXLogState
* Add to element and neighbor pages
*/
static void
AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, BlockNumber *updatedInsertPage, bool building)
WriteNewElementPages(Relation index, HnswElement e, int m, BlockNumber insertPage, BlockNumber *updatedInsertPage)
{
Buffer buf;
Page page;
@@ -154,17 +123,16 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
Size minCombinedSize;
HnswElementTuple etup;
BlockNumber currentPage = insertPage;
int dimensions = e->vec->dim;
HnswNeighborTuple ntup;
Buffer nbuf;
Page npage;
OffsetNumber freeOffno = InvalidOffsetNumber;
OffsetNumber freeNeighborOffno = InvalidOffsetNumber;
BlockNumber newInsertPage = InvalidBlockNumber;
uint8 tupleVersion;
char *base = NULL;
/* Calculate sizes */
etupSize = HNSW_ELEMENT_TUPLE_SIZE(VARSIZE_ANY(HnswPtrAccess(base, e->value)));
etupSize = HNSW_ELEMENT_TUPLE_SIZE(dimensions);
ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m);
combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
maxSize = HNSW_MAX_SIZE;
@@ -172,11 +140,11 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
/* Prepare element tuple */
etup = palloc0(etupSize);
HnswSetElementTuple(base, etup, e);
HnswSetElementTuple(etup, e);
/* Prepare neighbor tuple */
ntup = palloc0(ntupSize);
HnswSetNeighborTuple(base, ntup, e, m);
HnswSetNeighborTuple(ntup, e, m);
/* Find a page (or two if needed) to insert the tuples */
for (;;)
@@ -184,16 +152,8 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
buf = ReadBuffer(index, currentPage);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
if (building)
{
state = NULL;
page = BufferGetPage(buf);
}
else
{
state = GenericXLogStart(index);
page = GenericXLogRegisterBuffer(state, buf, 0);
}
state = GenericXLogStart(index);
page = GenericXLogRegisterBuffer(state, buf, 0);
/* Keep track of first page where element at level 0 can fit */
if (!BlockNumberIsValid(newInsertPage) && PageGetFreeSpace(page) >= minCombinedSize)
@@ -210,19 +170,10 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
}
/* Next, try space from a deleted element */
if (HnswFreeOffset(index, buf, page, e, etupSize, ntupSize, &nbuf, &npage, &freeOffno, &freeNeighborOffno, &newInsertPage, &tupleVersion))
if (HnswFreeOffset(index, buf, page, e, ntupSize, &nbuf, &npage, &freeOffno, &freeNeighborOffno, &newInsertPage))
{
if (nbuf != buf)
{
if (building)
npage = BufferGetPage(nbuf);
else
npage = GenericXLogRegisterBuffer(state, nbuf, 0);
}
/* Set tuple version */
etup->version = tupleVersion;
ntup->version = tupleVersion;
npage = GenericXLogRegisterBuffer(state, nbuf, 0);
break;
}
@@ -231,7 +182,7 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
/* Skip if both tuples can fit on the same page */
if (combinedSize > maxSize && PageGetFreeSpace(page) >= etupSize && !BlockNumberIsValid(HnswPageGetOpaque(page)->nextblkno))
{
HnswInsertAppendPage(index, &nbuf, &npage, state, page, building);
HnswInsertAppendPage(index, &nbuf, &npage, state, page);
break;
}
@@ -240,8 +191,7 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
if (BlockNumberIsValid(currentPage))
{
/* Move to next page */
if (!building)
GenericXLogAbort(state);
GenericXLogAbort(state);
UnlockReleaseBuffer(buf);
}
else
@@ -249,33 +199,22 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
Buffer newbuf;
Page newpage;
HnswInsertAppendPage(index, &newbuf, &newpage, state, page, building);
HnswInsertAppendPage(index, &newbuf, &newpage, state, page);
/* Commit */
if (building)
MarkBufferDirty(buf);
else
GenericXLogFinish(state);
GenericXLogFinish(state);
/* Unlock previous buffer */
UnlockReleaseBuffer(buf);
/* Prepare new buffer */
state = GenericXLogStart(index);
buf = newbuf;
if (building)
{
state = NULL;
page = BufferGetPage(buf);
}
else
{
state = GenericXLogStart(index);
page = GenericXLogRegisterBuffer(state, buf, 0);
}
page = GenericXLogRegisterBuffer(state, buf, 0);
/* Create new page for neighbors if needed */
if (PageGetFreeSpace(page) < combinedSize)
HnswInsertAppendPage(index, &nbuf, &npage, state, page, building);
HnswInsertAppendPage(index, &nbuf, &npage, state, page);
else
{
nbuf = buf;
@@ -329,14 +268,7 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
}
/* Commit */
if (building)
{
MarkBufferDirty(buf);
if (nbuf != buf)
MarkBufferDirty(nbuf);
}
else
GenericXLogFinish(state);
GenericXLogFinish(state);
UnlockReleaseBuffer(buf);
if (nbuf != buf)
UnlockReleaseBuffer(nbuf);
@@ -346,107 +278,6 @@ AddElementOnDisk(Relation index, HnswElement e, int m, BlockNumber insertPage, B
*updatedInsertPage = newInsertPage;
}
/*
* Load neighbors
*/
static HnswNeighborArray *
HnswLoadNeighbors(HnswElement element, Relation index, int m, int lm, int lc)
{
char *base = NULL;
HnswNeighborArray *neighbors = HnswInitNeighborArray(lm, NULL);
ItemPointerData indextids[HNSW_MAX_M * 2];
if (!HnswLoadNeighborTids(element, indextids, index, m, lm, lc))
return neighbors;
for (int i = 0; i < lm; i++)
{
ItemPointer indextid = &indextids[i];
HnswElement e;
HnswCandidate *hc;
if (!ItemPointerIsValid(indextid))
break;
e = HnswInitElementFromBlock(ItemPointerGetBlockNumber(indextid), ItemPointerGetOffsetNumber(indextid));
hc = &neighbors->items[neighbors->length++];
HnswPtrStore(base, hc->element, e);
}
return neighbors;
}
/*
* Load elements for insert
*/
static void
LoadElementsForInsert(HnswNeighborArray * neighbors, HnswQuery * q, int *idx, Relation index, HnswSupport * support)
{
char *base = NULL;
for (int i = 0; i < neighbors->length; i++)
{
HnswCandidate *hc = &neighbors->items[i];
HnswElement element = HnswPtrAccess(base, hc->element);
double distance;
HnswLoadElement(element, &distance, q, index, support, true, NULL);
hc->distance = distance;
/* Prune element if being deleted */
if (element->heaptidsLength == 0)
{
*idx = i;
break;
}
}
}
/*
* Get update index
*/
static int
GetUpdateIndex(HnswElement element, HnswElement newElement, float distance, int m, int lm, int lc, Relation index, HnswSupport * support, MemoryContext updateCtx)
{
char *base = NULL;
int idx = -1;
HnswNeighborArray *neighbors;
MemoryContext oldCtx = MemoryContextSwitchTo(updateCtx);
/*
* Get latest neighbors since they may have changed. Do not lock yet since
* selecting neighbors can take time. Could use optimistic locking to
* retry if another update occurs before getting exclusive lock.
*/
neighbors = HnswLoadNeighbors(element, index, m, lm, lc);
/*
* Could improve performance for vacuuming by checking neighbors against
* list of elements being deleted to find index. It's important to exclude
* already deleted elements for this since they can be replaced at any
* time.
*/
if (neighbors->length < lm)
idx = -2;
else
{
HnswQuery q;
q.value = HnswGetValue(base, element);
LoadElementsForInsert(neighbors, &q, &idx, index, support);
if (idx == -1)
HnswUpdateConnection(base, neighbors, newElement, distance, lm, &idx, index, support);
}
MemoryContextSwitchTo(oldCtx);
MemoryContextReset(updateCtx);
return idx;
}
/*
* Check if connection already exists
*/
@@ -467,143 +298,122 @@ ConnectionExists(HnswElement e, HnswNeighborTuple ntup, int startIdx, int lm)
return false;
}
/*
* Update neighbor
*/
static void
UpdateNeighborOnDisk(HnswElement element, HnswElement newElement, int idx, int m, int lm, int lc, Relation index, bool checkExisting, bool building)
{
Buffer buf;
Page page;
GenericXLogState *state;
HnswNeighborTuple ntup;
int startIdx;
OffsetNumber offno = element->neighborOffno;
/* Register page */
buf = ReadBuffer(index, element->neighborPage);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
if (building)
{
state = NULL;
page = BufferGetPage(buf);
}
else
{
state = GenericXLogStart(index);
page = GenericXLogRegisterBuffer(state, buf, 0);
}
/* Get tuple */
ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, offno));
/* Calculate index for update */
startIdx = (element->level - lc) * m;
/* Check for existing connection */
if (checkExisting && ConnectionExists(newElement, ntup, startIdx, lm))
idx = -1;
else if (idx == -2)
{
/* Find free offset if still exists */
/* TODO Retry updating connections if not */
for (int j = 0; j < lm; j++)
{
if (!ItemPointerIsValid(&ntup->indextids[startIdx + j]))
{
idx = startIdx + j;
break;
}
}
}
else
idx += startIdx;
/* Make robust to issues */
if (idx >= 0 && idx < ntup->count)
{
ItemPointer indextid = &ntup->indextids[idx];
/* Update neighbor on the buffer */
ItemPointerSet(indextid, newElement->blkno, newElement->offno);
/* Commit */
if (building)
MarkBufferDirty(buf);
else
GenericXLogFinish(state);
}
else if (!building)
GenericXLogAbort(state);
UnlockReleaseBuffer(buf);
}
/*
* Update neighbors
*/
void
HnswUpdateNeighborsOnDisk(Relation index, HnswSupport * support, HnswElement e, int m, bool checkExisting, bool building)
HnswUpdateNeighborPages(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement e, int m, bool checkExisting)
{
char *base = NULL;
/* Use separate memory context to improve performance for larger vectors */
MemoryContext updateCtx = GenerationContextCreate(CurrentMemoryContext,
"Hnsw insert update context",
#if PG_VERSION_NUM >= 150000
128 * 1024, 128 * 1024,
#endif
128 * 1024);
for (int lc = e->level; lc >= 0; lc--)
{
int lm = HnswGetLayerM(m, lc);
HnswNeighborArray *neighbors = HnswGetNeighbors(base, e, lc);
HnswNeighborArray *neighbors = &e->neighbors[lc];
for (int i = 0; i < neighbors->length; i++)
{
HnswCandidate *hc = &neighbors->items[i];
HnswElement neighborElement = HnswPtrAccess(base, hc->element);
int idx;
Buffer buf;
Page page;
GenericXLogState *state;
ItemId itemid;
HnswNeighborTuple ntup;
Size ntupSize;
int idx = -1;
int startIdx;
OffsetNumber offno = hc->element->neighborOffno;
idx = GetUpdateIndex(neighborElement, e, hc->distance, m, lm, lc, index, support, updateCtx);
/* Get latest neighbors since they may have changed */
/* Do not lock yet since selecting neighbors can take time */
HnswLoadNeighbors(hc->element, index, m);
/*
* Could improve performance for vacuuming by checking neighbors
* against list of elements being deleted to find index. It's
* important to exclude already deleted elements for this since
* they can be replaced at any time.
*/
/* Select neighbors */
HnswUpdateConnection(e, hc, lm, lc, &idx, index, procinfo, collation);
/* New element was not selected as a neighbor */
if (idx == -1)
continue;
UpdateNeighborOnDisk(neighborElement, e, idx, m, lm, lc, index, checkExisting, building);
/* Register page */
buf = ReadBuffer(index, hc->element->neighborPage);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
state = GenericXLogStart(index);
page = GenericXLogRegisterBuffer(state, buf, 0);
/* Get tuple */
itemid = PageGetItemId(page, offno);
ntup = (HnswNeighborTuple) PageGetItem(page, itemid);
ntupSize = ItemIdGetLength(itemid);
/* Calculate index for update */
startIdx = (hc->element->level - lc) * m;
/* Check for existing connection */
if (checkExisting && ConnectionExists(e, ntup, startIdx, lm))
idx = -1;
else if (idx == -2)
{
/* Find free offset if still exists */
/* TODO Retry updating connections if not */
for (int j = 0; j < lm; j++)
{
if (!ItemPointerIsValid(&ntup->indextids[startIdx + j]))
{
idx = startIdx + j;
break;
}
}
}
else
idx += startIdx;
/* Make robust to issues */
if (idx >= 0 && idx < ntup->count)
{
ItemPointer indextid = &ntup->indextids[idx];
/* Update neighbor */
ItemPointerSet(indextid, e->blkno, e->offno);
/* Overwrite tuple */
if (!PageIndexTupleOverwrite(page, offno, (Item) ntup, ntupSize))
elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
/* Commit */
GenericXLogFinish(state);
}
else
GenericXLogAbort(state);
UnlockReleaseBuffer(buf);
}
}
MemoryContextDelete(updateCtx);
}
/*
* Add a heap TID to an existing element
*/
static bool
AddDuplicateOnDisk(Relation index, HnswElement element, HnswElement dup, bool building)
HnswAddDuplicate(Relation index, HnswElement element, HnswElement dup)
{
Buffer buf;
Page page;
GenericXLogState *state;
Size etupSize = HNSW_ELEMENT_TUPLE_SIZE(dup->vec->dim);
HnswElementTuple etup;
int i;
/* Read page */
buf = ReadBuffer(index, dup->blkno);
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
if (building)
{
state = NULL;
page = BufferGetPage(buf);
}
else
{
state = GenericXLogStart(index);
page = GenericXLogRegisterBuffer(state, buf, 0);
}
state = GenericXLogStart(index);
page = GenericXLogRegisterBuffer(state, buf, 0);
/* Find space */
etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, dup->offno));
@@ -616,91 +426,82 @@ AddDuplicateOnDisk(Relation index, HnswElement element, HnswElement dup, bool bu
/* Either being deleted or we lost our chance to another backend */
if (i == 0 || i == HNSW_HEAPTIDS)
{
if (!building)
GenericXLogAbort(state);
GenericXLogAbort(state);
UnlockReleaseBuffer(buf);
return false;
}
/* Add heap TID, modifying the tuple on the page directly */
etup->heaptids[i] = element->heaptids[0];
/* Add heap TID */
etup->heaptids[i] = *((ItemPointer) linitial(element->heaptids));
/* Overwrite tuple */
if (!PageIndexTupleOverwrite(page, dup->offno, (Item) etup, etupSize))
elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
/* Commit */
if (building)
MarkBufferDirty(buf);
else
GenericXLogFinish(state);
GenericXLogFinish(state);
UnlockReleaseBuffer(buf);
return true;
}
/*
* Find duplicate element
*/
static bool
FindDuplicateOnDisk(Relation index, HnswElement element, bool building)
{
char *base = NULL;
HnswNeighborArray *neighbors = HnswGetNeighbors(base, element, 0);
Datum value = HnswGetValue(base, element);
for (int i = 0; i < neighbors->length; i++)
{
HnswCandidate *neighbor = &neighbors->items[i];
HnswElement neighborElement = HnswPtrAccess(base, neighbor->element);
Datum neighborValue = HnswGetValue(base, neighborElement);
/* Exit early since ordered by distance */
if (!datumIsEqual(value, neighborValue, false, -1))
return false;
if (AddDuplicateOnDisk(index, element, neighborElement, building))
return true;
}
return false;
}
/*
* Update graph on disk
* Write changes to disk
*/
static void
UpdateGraphOnDisk(Relation index, HnswSupport * support, HnswElement element, int m, HnswElement entryPoint, bool building)
WriteElement(Relation index, FmgrInfo *procinfo, Oid collation, HnswElement element, int m, int efConstruction, HnswElement dup, HnswElement entryPoint)
{
BlockNumber newInsertPage = InvalidBlockNumber;
/* Look for duplicate */
if (FindDuplicateOnDisk(index, element, building))
return;
/* Try to add to existing page */
if (dup != NULL)
{
if (HnswAddDuplicate(index, element, dup))
return;
}
/* Add element */
AddElementOnDisk(index, element, m, GetInsertPage(index), &newInsertPage, building);
/* Write element and neighbor tuples */
WriteNewElementPages(index, element, m, GetInsertPage(index), &newInsertPage);
/* Update insert page if needed */
if (BlockNumberIsValid(newInsertPage))
HnswUpdateMetaPage(index, 0, NULL, newInsertPage, MAIN_FORKNUM, building);
HnswUpdateMetaPage(index, 0, NULL, newInsertPage, MAIN_FORKNUM);
/* Update neighbors */
HnswUpdateNeighborsOnDisk(index, support, element, m, false, building);
HnswUpdateNeighborPages(index, procinfo, collation, element, m, false);
/* Update entry point if needed */
/* Update metapage if needed */
if (entryPoint == NULL || element->level > entryPoint->level)
HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_GREATER, element, InvalidBlockNumber, MAIN_FORKNUM, building);
HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_GREATER, element, InvalidBlockNumber, MAIN_FORKNUM);
}
/*
* Insert a tuple into the index
*/
bool
HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPointer heaptid, bool building)
HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel)
{
Datum value;
FmgrInfo *normprocinfo;
HnswElement entryPoint;
HnswElement element;
int m;
int efConstruction = HnswGetEfConstruction(index);
FmgrInfo *procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
Oid collation = index->rd_indcollation[0];
HnswElement dup;
LOCKMODE lockmode = ShareLock;
char *base = NULL;
/* Detoast once for all calls */
value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
/* Normalize if needed */
normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
if (normprocinfo != NULL)
{
if (!HnswNormValue(normprocinfo, collation, &value, NULL))
return false;
}
/*
* Get a shared lock. This allows vacuum to ensure no in-flight inserts
@@ -713,8 +514,8 @@ HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPo
HnswGetMetaPageInfo(index, &m, &entryPoint);
/* Create an element */
element = HnswInitElement(base, heaptid, m, HnswGetMl(m), HnswGetMaxLevel(m), NULL);
HnswPtrStore(base, element->value, (char *) DatumGetPointer(value));
element = HnswInitElement(heap_tid, m, HnswGetMl(m), HnswGetMaxLevel(m));
element->vec = DatumGetVector(value);
/* Prevent concurrent inserts when likely updating entry point */
if (entryPoint == NULL || element->level > entryPoint->level)
@@ -730,11 +531,14 @@ HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPo
entryPoint = HnswGetEntryPoint(index);
}
/* Find neighbors for element */
HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false);
/* Insert element in graph */
HnswInsertElement(element, entryPoint, index, procinfo, collation, m, efConstruction, false);
/* Update graph on disk */
UpdateGraphOnDisk(index, support, element, m, entryPoint, building);
/* Look for duplicate */
dup = HnswFindDuplicate(element);
/* Write to disk */
WriteElement(index, procinfo, collation, element, m, efConstruction, dup, entryPoint);
/* Release lock */
UnlockPage(index, HNSW_UPDATE_LOCK, lockmode);
@@ -742,25 +546,6 @@ HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPo
return true;
}
/*
* Insert a tuple into the index
*/
static void
HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heaptid)
{
Datum value;
const HnswTypeInfo *typeInfo = HnswGetTypeInfo(index);
HnswSupport support;
HnswInitSupport(&support, index);
/* Form index value */
if (!HnswFormIndexValue(&value, values, isnull, typeInfo, &support))
return;
HnswInsertTupleOnDisk(index, &support, value, heaptid, false);
}
/*
* Insert a tuple into the index
*/
@@ -787,7 +572,7 @@ hnswinsert(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid,
oldCtx = MemoryContextSwitchTo(insertCtx);
/* Insert tuple */
HnswInsertTuple(index, values, isnull, heap_tid);
HnswInsertTuple(index, values, isnull, heap_tid, heap);
/* Delete memory context */
MemoryContextSwitchTo(oldCtx);

View File

@@ -1,89 +1,66 @@
#include "postgres.h"
#include <limits.h>
#include "access/genam.h"
#include "access/relscan.h"
#include "hnsw.h"
#include "lib/pairingheap.h"
#include "miscadmin.h"
#include "nodes/pg_list.h"
#include "pgstat.h"
#include "storage/bufmgr.h"
#include "storage/lmgr.h"
#include "utils/float.h"
#include "utils/memutils.h"
#include "utils/relcache.h"
#include "utils/snapmgr.h"
#if PG_VERSION_NUM >= 160000
#include "varatt.h"
#endif
/*
* Algorithm 5 from paper
*/
static List *
GetScanItems(IndexScanDesc scan, Datum value)
GetScanItems(IndexScanDesc scan, Datum q)
{
HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
Relation index = scan->indexRelation;
HnswSupport *support = &so->support;
FmgrInfo *procinfo = so->procinfo;
Oid collation = so->collation;
bool loadVec = scan->xs_want_itup;
List *ep;
List *w;
int m;
HnswElement entryPoint;
char *base = NULL;
HnswQuery *q = &so->q;
/* Get m and entry point */
HnswGetMetaPageInfo(index, &m, &entryPoint);
q->value = value;
so->m = m;
if (entryPoint == NULL)
return NIL;
ep = list_make1(HnswEntryCandidate(base, entryPoint, q, index, support, false));
ep = list_make1(HnswEntryCandidate(entryPoint, q, index, procinfo, collation, loadVec));
for (int lc = entryPoint->level; lc >= 1; lc--)
{
w = HnswSearchLayer(base, q, ep, 1, lc, index, support, m, false, NULL, NULL, NULL, true, NULL);
w = HnswSearchLayer(q, ep, 1, lc, index, procinfo, collation, m, loadVec, NULL);
ep = w;
}
return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, support, m, false, NULL, &so->v, hnsw_iterative_scan != HNSW_ITERATIVE_SCAN_OFF ? &so->discarded : NULL, true, &so->tuples);
return HnswSearchLayer(q, ep, hnsw_ef_search, 0, index, procinfo, collation, m, loadVec, NULL);
}
/*
* Resume scan at ground level with discarded candidates
* Get dimensions from metapage
*/
static List *
ResumeScanItems(IndexScanDesc scan)
static int
GetDimensions(Relation index)
{
HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
Relation index = scan->indexRelation;
List *ep = NIL;
char *base = NULL;
int batch_size = hnsw_ef_search;
Buffer buf;
Page page;
HnswMetaPage metap;
int dimensions;
if (pairingheap_is_empty(so->discarded))
return NIL;
buf = ReadBuffer(index, HNSW_METAPAGE_BLKNO);
LockBuffer(buf, BUFFER_LOCK_SHARE);
page = BufferGetPage(buf);
metap = HnswPageGetMeta(page);
/* Get next batch of candidates */
for (int i = 0; i < batch_size; i++)
{
HnswSearchCandidate *sc;
dimensions = metap->dimensions;
if (pairingheap_is_empty(so->discarded))
break;
UnlockReleaseBuffer(buf);
sc = HnswGetSearchCandidate(w_node, pairingheap_remove_first(so->discarded));
ep = lappend(ep, sc);
}
return HnswSearchLayer(base, &so->q, ep, batch_size, 0, index, &so->support, so->m, false, NULL, &so->v, &so->discarded, false, &so->tuples);
return dimensions;
}
/*
@@ -96,7 +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;
@@ -105,25 +82,14 @@ GetScanValue(IndexScanDesc scan)
Assert(!VARATT_IS_COMPRESSED(DatumGetPointer(value)));
Assert(!VARATT_IS_EXTENDED(DatumGetPointer(value)));
/* Normalize if needed */
if (so->support.normprocinfo != NULL)
value = HnswNormValue(so->typeInfo, so->support.collation, value);
/* Fine if normalization fails */
if (so->normprocinfo != NULL)
HnswNormValue(so->normprocinfo, so->collation, &value, NULL);
}
return value;
}
#if defined(HNSW_MEMORY)
/*
* Show memory usage
*/
static void
ShowMemoryUsage(HnswScanOpaque so)
{
elog(INFO, "memory: %zu KB, tuples: " INT64_FORMAT, MemoryContextMemAllocated(so->tmpCtx, false) / 1024, so->tuples);
}
#endif
/*
* Prepare for an index scan
*/
@@ -132,31 +98,25 @@ hnswbeginscan(Relation index, int nkeys, int norderbys)
{
IndexScanDesc scan;
HnswScanOpaque so;
double maxMemory;
scan = RelationGetIndexScan(index, nkeys, norderbys);
so = (HnswScanOpaque) palloc(sizeof(HnswScanOpaqueData));
so->typeInfo = HnswGetTypeInfo(index);
/* Set support functions */
HnswInitSupport(&so->support, index);
/*
* Use a lower max allocation size than default to allow scanning more
* tuples for iterative search before exceeding work_mem
*/
so->first = true;
so->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
"Hnsw scan temporary context",
0, 8 * 1024, 256 * 1024);
ALLOCSET_DEFAULT_SIZES);
/* Calculate max memory */
/* Add 256 extra bytes to fill last block when close */
maxMemory = (double) work_mem * hnsw_scan_mem_multiplier * 1024.0 + 256;
so->maxMemory = Min(maxMemory, (double) (SIZE_MAX / 2));
/* Set support functions */
so->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
so->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
so->collation = index->rd_indcollation[0];
scan->opaque = so;
/* OK to always set since cheap */
scan->xs_itupdesc = RelationGetDescr(index);
return scan;
}
@@ -169,11 +129,6 @@ hnswrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int no
HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
so->first = true;
/* v and discarded are allocated in tmpCtx */
so->v.tids = NULL;
so->discarded = NULL;
so->tuples = 0;
so->previousDistance = -get_float8_infinity();
MemoryContextReset(so->tmpCtx);
if (keys && scan->numberOfKeys > 0)
@@ -204,10 +159,6 @@ 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)
@@ -233,95 +184,41 @@ hnswgettuple(IndexScanDesc scan, ScanDirection dir)
UnlockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
so->first = false;
#if defined(HNSW_MEMORY)
ShowMemoryUsage(so);
#endif
}
for (;;)
while (list_length(so->w) > 0)
{
char *base = NULL;
HnswSearchCandidate *sc;
HnswElement element;
HnswCandidate *hc = llast(so->w);
ItemPointer heaptid;
if (list_length(so->w) == 0)
{
if (hnsw_iterative_scan == HNSW_ITERATIVE_SCAN_OFF)
break;
/* Empty index */
if (so->discarded == NULL)
break;
/* Reached max number of tuples or memory limit */
if (so->tuples >= hnsw_max_scan_tuples || MemoryContextMemAllocated(so->tmpCtx, false) > so->maxMemory)
{
if (pairingheap_is_empty(so->discarded))
break;
/* Return remaining tuples */
so->w = lappend(so->w, HnswGetSearchCandidate(w_node, pairingheap_remove_first(so->discarded)));
}
else
{
/*
* Locking ensures when neighbors are read, the elements they
* reference will not be deleted (and replaced) during the
* iteration.
*
* Elements loaded into memory on previous iterations may have
* been deleted (and replaced), so when reading neighbors, the
* element version must be checked.
*/
LockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
so->w = ResumeScanItems(scan);
UnlockPage(scan->indexRelation, HNSW_SCAN_LOCK, ShareLock);
#if defined(HNSW_MEMORY)
ShowMemoryUsage(so);
#endif
}
if (list_length(so->w) == 0)
break;
}
sc = llast(so->w);
element = HnswPtrAccess(base, sc->element);
/* Move to next element if no valid heap TIDs */
if (element->heaptidsLength == 0)
if (list_length(hc->element->heaptids) == 0)
{
so->w = list_delete_last(so->w);
/* Mark memory as free for next iteration */
if (hnsw_iterative_scan != HNSW_ITERATIVE_SCAN_OFF)
{
pfree(element);
pfree(sc);
}
continue;
}
heaptid = &element->heaptids[--element->heaptidsLength];
heaptid = llast(hc->element->heaptids);
if (hnsw_iterative_scan == HNSW_ITERATIVE_SCAN_STRICT)
hc->element->heaptids = list_delete_last(hc->element->heaptids);
if (scan->xs_want_itup)
{
if (sc->distance < so->previousDistance)
continue;
Datum value = PointerGetDatum(hc->element->vec);
bool isnull = false;
so->previousDistance = sc->distance;
/* Allocate in temporary context, so no need to free */
scan->xs_itup = index_form_tuple(scan->xs_itupdesc, &value, &isnull);
}
MemoryContextSwitchTo(oldCtx);
#if PG_VERSION_NUM >= 120000
scan->xs_heaptid = *heaptid;
scan->xs_recheck = false;
#else
scan->xs_ctup.t_self = *heaptid;
#endif
scan->xs_recheckorderby = false;
return true;
}

File diff suppressed because it is too large Load Diff

View File

@@ -1,30 +1,23 @@
#include "postgres.h"
#include "access/genam.h"
#include "access/generic_xlog.h"
#include <math.h>
#include "commands/vacuum.h"
#include "hnsw.h"
#include "nodes/pg_list.h"
#include "storage/bufmgr.h"
#include "storage/lmgr.h"
#include "utils/memutils.h"
#include "utils/rel.h"
#if PG_VERSION_NUM >= 160000
#include "varatt.h"
#endif
#if PG_VERSION_NUM >= 180000
#define vacuum_delay_point() vacuum_delay_point(false)
#endif
/*
* Check if deleted list contains an index TID
*/
static bool
DeletedContains(tidhash_hash * deleted, ItemPointer indextid)
DeletedContains(HTAB *deleted, ItemPointer indextid)
{
return tidhash_lookup(deleted, *indextid) != NULL;
bool found;
hash_search(deleted, indextid, HASH_FIND, &found);
return found;
}
/*
@@ -37,20 +30,17 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
{
BlockNumber blkno = HNSW_HEAD_BLKNO;
HnswElement highestPoint = &vacuumstate->highestPoint;
HnswElement fallbackPoint = &vacuumstate->fallbackPoint;
Relation index = vacuumstate->index;
BufferAccessStrategy bas = vacuumstate->bas;
HnswElement entryPoint = HnswGetEntryPoint(vacuumstate->index);
IndexBulkDeleteResult *stats = vacuumstate->stats;
/* Store separately since HnswElement level is uint8 */
/* Store separately since highestPoint.level is uint8 */
int highestLevel = -1;
int fallbackLevel = -1;
/* Initialize highest point and fallback point */
/* Initialize highest point */
highestPoint->blkno = InvalidBlockNumber;
highestPoint->offno = InvalidOffsetNumber;
fallbackPoint->blkno = InvalidBlockNumber;
fallbackPoint->offno = InvalidOffsetNumber;
while (BlockNumberIsValid(blkno))
{
@@ -103,10 +93,15 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
if (itemUpdated)
{
Size etupSize = HNSW_ELEMENT_TUPLE_SIZE(etup->vec.dim);
/* Mark rest as invalid */
for (int i = idx; i < HNSW_HEAPTIDS; i++)
ItemPointerSetInvalid(&etup->heaptids[i]);
if (!PageIndexTupleOverwrite(page, offno, (Item) etup, etupSize))
elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
updated = true;
}
}
@@ -114,39 +109,20 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
if (!ItemPointerIsValid(&etup->heaptids[0]))
{
ItemPointerData ip;
bool found;
/* Add to deleted list */
ItemPointerSet(&ip, blkno, offno);
tidhash_insert(vacuumstate->deleted, ip, &found);
Assert(!found);
(void) hash_search(vacuumstate->deleted, &ip, HASH_ENTER, NULL);
}
else if (etup->level > highestLevel)
else if (etup->level > highestLevel && !(entryPoint != NULL && blkno == entryPoint->blkno && offno == entryPoint->offno))
{
if (BlockNumberIsValid(highestPoint->blkno))
{
/* Current highest point becomes fallback */
fallbackPoint->blkno = highestPoint->blkno;
fallbackPoint->offno = highestPoint->offno;
fallbackPoint->level = highestPoint->level;
fallbackLevel = highestLevel;
}
/* Keep track of highest point */
/* Keep track of highest non-entry point */
highestPoint->blkno = blkno;
highestPoint->offno = offno;
highestPoint->level = etup->level;
highestLevel = etup->level;
}
else if (etup->level > fallbackLevel)
{
/* Keep track of second highest point */
fallbackPoint->blkno = blkno;
fallbackPoint->offno = offno;
fallbackPoint->level = etup->level;
fallbackLevel = etup->level;
}
}
blkno = HnswPageGetOpaque(page)->nextblkno;
@@ -158,10 +134,6 @@ RemoveHeapTids(HnswVacuumState * vacuumstate)
UnlockReleaseBuffer(buf);
}
#ifdef HNSW_MEMORY
elog(INFO, "memory: %zu KB", MemoryContextMemAllocated(CurrentMemoryContext, true) / 1024);
#endif
}
/*
@@ -203,12 +175,7 @@ NeedsUpdated(HnswVacuumState * vacuumstate, HnswElement element)
/* Also update if layer 0 is not full */
/* This could indicate too many candidates being deleted during insert */
if (!needsUpdated)
{
/* Keep clang-tidy happy */
Assert(ntup->count > 0);
needsUpdated = !ItemPointerIsValid(&ntup->indextids[ntup->count - 1]);
}
UnlockReleaseBuffer(buf);
@@ -222,34 +189,31 @@ static void
RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswElement entryPoint)
{
Relation index = vacuumstate->index;
HnswSupport *support = &vacuumstate->support;
Buffer buf;
Page page;
GenericXLogState *state;
int m = vacuumstate->m;
int efConstruction = vacuumstate->efConstruction;
FmgrInfo *procinfo = vacuumstate->procinfo;
Oid collation = vacuumstate->collation;
BufferAccessStrategy bas = vacuumstate->bas;
HnswNeighborTuple ntup = vacuumstate->ntup;
Size ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, m);
char *base = NULL;
/* Skip if element is entry point */
if (entryPoint != NULL && element->blkno == entryPoint->blkno && element->offno == entryPoint->offno)
return;
/* Init fields */
HnswInitNeighbors(base, element, m, NULL);
element->heaptidsLength = 0;
HnswInitNeighbors(element, m);
element->heaptids = NIL;
/* Find neighbors for element, skipping itself */
HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, true);
/* Zero memory for each element */
MemSet(ntup, 0, HNSW_TUPLE_ALLOC_SIZE);
/* Add element to graph, skipping itself */
HnswInsertElement(element, entryPoint, index, procinfo, collation, m, efConstruction, true);
/* Update neighbor tuple */
/* Do this before getting page to minimize locking */
HnswSetNeighborTuple(base, ntup, element, m);
HnswSetNeighborTuple(ntup, element, m);
/* Get neighbor page */
buf = ReadBufferExtended(index, MAIN_FORKNUM, element->neighborPage, RBM_NORMAL, bas);
@@ -266,7 +230,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
UnlockReleaseBuffer(buf);
/* Update neighbors */
HnswUpdateNeighborsOnDisk(index, support, element, m, true, false);
HnswUpdateNeighborPages(index, procinfo, collation, element, m, true);
}
/*
@@ -276,7 +240,6 @@ static void
RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
{
Relation index = vacuumstate->index;
HnswSupport *support = &vacuumstate->support;
HnswElement highestPoint = &vacuumstate->highestPoint;
HnswElement entryPoint;
MemoryContext oldCtx = MemoryContextSwitchTo(vacuumstate->tmpCtx);
@@ -293,27 +256,12 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
/* Get a shared lock */
LockPage(index, HNSW_UPDATE_LOCK, ShareLock);
/* Get latest entry point */
entryPoint = HnswGetEntryPoint(index);
/* Load element */
HnswLoadElement(highestPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true);
/* Use fallback point if highest point is entry point */
if (entryPoint != NULL && entryPoint->blkno == highestPoint->blkno && entryPoint->offno == highestPoint->offno)
{
highestPoint = &vacuumstate->fallbackPoint;
if (!BlockNumberIsValid(highestPoint->blkno))
highestPoint = NULL;
}
if (highestPoint != NULL)
{
/* Load element */
HnswLoadElement(highestPoint, NULL, NULL, index, support, true, NULL);
/* Repair if needed */
if (NeedsUpdated(vacuumstate, highestPoint))
RepairGraphElement(vacuumstate, highestPoint, entryPoint);
}
/* Repair if needed */
if (NeedsUpdated(vacuumstate, highestPoint))
RepairGraphElement(vacuumstate, highestPoint, HnswGetEntryPoint(index));
/* Release lock */
UnlockPage(index, HNSW_UPDATE_LOCK, ShareLock);
@@ -338,7 +286,7 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
* point is outdated and empty, the entry point will be empty
* until an element is repaired.
*/
HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_ALWAYS, highestPoint, InvalidBlockNumber, MAIN_FORKNUM, false);
HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_ALWAYS, highestPoint, InvalidBlockNumber, MAIN_FORKNUM);
}
else
{
@@ -347,13 +295,13 @@ RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
* is outdated, this can remove connections at higher levels in
* the graph until they are repaired, but this should be fine.
*/
HnswLoadElement(entryPoint, NULL, NULL, index, support, true, NULL);
HnswLoadElement(entryPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true);
if (NeedsUpdated(vacuumstate, entryPoint))
{
/* Reset neighbors from previous update */
if (highestPoint != NULL)
HnswPtrStore((char *) NULL, highestPoint->neighbors, (HnswNeighborArrayPtr *) NULL);
highestPoint->neighbors = NULL;
RepairGraphElement(vacuumstate, entryPoint, highestPoint);
}
@@ -471,7 +419,7 @@ RepairGraph(HnswVacuumState * vacuumstate)
* was replaced and highest point was outdated.
*/
if (entryPoint == NULL || element->level > entryPoint->level)
HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_GREATER, element, InvalidBlockNumber, MAIN_FORKNUM, false);
HnswUpdateMetaPage(index, HNSW_UPDATE_ENTRY_GREATER, element, InvalidBlockNumber, MAIN_FORKNUM);
/* Release lock */
UnlockPage(index, HNSW_UPDATE_LOCK, lockmode);
@@ -480,99 +428,6 @@ RepairGraph(HnswVacuumState * vacuumstate)
/* Reset memory context */
MemoryContextSwitchTo(oldCtx);
MemoryContextReset(vacuumstate->tmpCtx);
#ifdef HNSW_VACUUM_PROGRESS
if (!BlockNumberIsValid(blkno) || (blkno - HNSW_HEAD_BLKNO) % 1000 == 0)
{
BlockNumber totalBlocks = RelationGetNumberOfBlocks(index);
BlockNumber currentBlocks = BlockNumberIsValid(blkno) ? blkno : totalBlocks;
elog(INFO, "hnsw vacuum progress: %.1f%%", 100.0 * currentBlocks / totalBlocks);
}
#endif
}
}
/*
* Confirm graph was repaired
*/
static void
ConfirmRepaired(HnswVacuumState * vacuumstate)
{
BlockNumber blkno = HNSW_HEAD_BLKNO;
Relation index = vacuumstate->index;
BufferAccessStrategy bas = vacuumstate->bas;
while (BlockNumberIsValid(blkno))
{
Buffer buf;
Page page;
OffsetNumber offno;
OffsetNumber maxoffno;
vacuum_delay_point();
buf = ReadBufferExtended(index, MAIN_FORKNUM, blkno, RBM_NORMAL, bas);
LockBuffer(buf, BUFFER_LOCK_SHARE);
page = BufferGetPage(buf);
maxoffno = PageGetMaxOffsetNumber(page);
/* Iterate over nodes */
for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
{
HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
HnswNeighborTuple ntup;
Buffer nbuf;
Page npage;
BlockNumber neighborPage;
OffsetNumber neighborOffno;
/* Skip neighbor tuples */
if (!HnswIsElementTuple(etup))
continue;
/* Skip if being deleted */
if (!ItemPointerIsValid(&etup->heaptids[0]))
continue;
/* Get neighbor page */
neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
if (neighborPage == blkno)
{
nbuf = buf;
npage = page;
}
else
{
nbuf = ReadBufferExtended(index, MAIN_FORKNUM, neighborPage, RBM_NORMAL, bas);
LockBuffer(nbuf, BUFFER_LOCK_SHARE);
npage = BufferGetPage(nbuf);
}
ntup = (HnswNeighborTuple) PageGetItem(npage, PageGetItemId(npage, neighborOffno));
/* Check neighbors */
for (int i = 0; i < ntup->count; i++)
{
ItemPointer indextid = &ntup->indextids[i];
if (!ItemPointerIsValid(indextid))
continue;
/* Check if in deleted list */
if (DeletedContains(vacuumstate->deleted, indextid))
elog(ERROR, "hnsw graph not repaired");
}
if (nbuf != buf)
UnlockReleaseBuffer(nbuf);
}
blkno = HnswPageGetOpaque(page)->nextblkno;
UnlockReleaseBuffer(buf);
}
}
@@ -624,6 +479,8 @@ MarkDeleted(HnswVacuumState * vacuumstate)
{
HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
HnswNeighborTuple ntup;
Size etupSize;
Size ntupSize;
Buffer nbuf;
Page npage;
BlockNumber neighborPage;
@@ -647,6 +504,10 @@ MarkDeleted(HnswVacuumState * vacuumstate)
if (ItemPointerIsValid(&etup->heaptids[0]))
continue;
/* Calculate sizes */
etupSize = HNSW_ELEMENT_TUPLE_SIZE(etup->vec.dim);
ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(etup->level, vacuumstate->m);
/* Get neighbor page */
neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
@@ -666,26 +527,20 @@ MarkDeleted(HnswVacuumState * vacuumstate)
ntup = (HnswNeighborTuple) PageGetItem(npage, PageGetItemId(npage, neighborOffno));
/* Overwrite element */
/* Use memset instead of MemSet to keep clang-tidy happy */
etup->deleted = 1;
memset(&etup->data, 0, VARSIZE_ANY(&etup->data));
MemSet(&etup->vec.x, 0, etup->vec.dim * sizeof(float));
/* Overwrite neighbors */
for (int i = 0; i < ntup->count; i++)
ItemPointerSetInvalid(&ntup->indextids[i]);
/* Increment version */
/* This is used to avoid incorrect reads for iterative scans */
/* Reserve some bits for future use */
etup->version++;
if (etup->version > 15)
etup->version = 1;
ntup->version = etup->version;
/* Overwrite element tuple */
if (!PageIndexTupleOverwrite(page, offno, (Item) etup, etupSize))
elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
/*
* We modified the tuples in place, no need to call
* PageIndexTupleOverwrite
*/
/* Overwrite neighbor tuple */
if (!PageIndexTupleOverwrite(npage, neighborOffno, (Item) ntup, ntupSize))
elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
/* Commit */
GenericXLogFinish(state);
@@ -708,7 +563,7 @@ MarkDeleted(HnswVacuumState * vacuumstate)
}
/* Update insert page last, after everything has been marked as deleted */
HnswUpdateMetaPage(index, 0, NULL, insertPage, MAIN_FORKNUM, false);
HnswUpdateMetaPage(index, 0, NULL, insertPage, MAIN_FORKNUM);
}
/*
@@ -718,6 +573,7 @@ static void
InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkDeleteResult *stats, IndexBulkDeleteCallback callback, void *callback_state)
{
Relation index = info->index;
HASHCTL hash_ctl;
if (stats == NULL)
stats = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
@@ -728,18 +584,21 @@ InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkD
vacuumstate->callback_state = callback_state;
vacuumstate->efConstruction = HnswGetEfConstruction(index);
vacuumstate->bas = GetAccessStrategy(BAS_BULKREAD);
vacuumstate->ntup = palloc0(HNSW_TUPLE_ALLOC_SIZE);
vacuumstate->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
vacuumstate->collation = index->rd_indcollation[0];
vacuumstate->ntup = palloc0(BLCKSZ);
vacuumstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
"Hnsw vacuum temporary context",
ALLOCSET_DEFAULT_SIZES);
HnswInitSupport(&vacuumstate->support, index);
/* Get m from metapage */
HnswGetMetaPageInfo(index, &vacuumstate->m, NULL);
/* Create hash table */
vacuumstate->deleted = tidhash_create(CurrentMemoryContext, 256, NULL);
hash_ctl.keysize = sizeof(ItemPointerData);
hash_ctl.entrysize = sizeof(ItemPointerData);
hash_ctl.hcxt = CurrentMemoryContext;
vacuumstate->deleted = hash_create("hnswbulkdelete indextids", 256, &hash_ctl, HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
}
/*
@@ -748,7 +607,7 @@ InitVacuumState(HnswVacuumState * vacuumstate, IndexVacuumInfo *info, IndexBulkD
static void
FreeVacuumState(HnswVacuumState * vacuumstate)
{
tidhash_destroy(vacuumstate->deleted);
hash_destroy(vacuumstate->deleted);
FreeAccessStrategy(vacuumstate->bas);
pfree(vacuumstate->ntup);
MemoryContextDelete(vacuumstate->tmpCtx);
@@ -766,16 +625,13 @@ hnswbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
InitVacuumState(&vacuumstate, info, stats, callback, callback_state);
/* Pass 1: Remove heap TIDs */
HnswBench("RemoveHeapTids", RemoveHeapTids(&vacuumstate));
RemoveHeapTids(&vacuumstate);
/* Pass 2: Repair graph */
HnswBench("RepairGraph", RepairGraph(&vacuumstate));
RepairGraph(&vacuumstate);
/* Pass 3: Confirm repaired */
HnswBench("ConfirmRepaired", ConfirmRepaired(&vacuumstate));
/* Pass 4: Mark as deleted */
HnswBench("MarkDeleted", MarkDeleted(&vacuumstate));
/* Pass 3: Mark as deleted */
MarkDeleted(&vacuumstate);
FreeVacuumState(&vacuumstate);

View File

@@ -2,49 +2,58 @@
#include <float.h>
#include "access/genam.h"
#include "access/generic_xlog.h"
#include "access/itup.h"
#include "access/relscan.h"
#include "access/table.h"
#include "access/tableam.h"
#include "access/tupdesc.h"
#include "access/parallel.h"
#include "access/xact.h"
#include "access/xloginsert.h"
#include "catalog/index.h"
#include "catalog/pg_operator_d.h"
#include "catalog/pg_type_d.h"
#include "commands/progress.h"
#include "fmgr.h"
#include "ivfflat.h"
#include "miscadmin.h"
#include "nodes/execnodes.h"
#include "optimizer/optimizer.h"
#include "storage/bufmgr.h"
#include "storage/condition_variable.h"
#include "tcop/tcopprot.h"
#include "utils/memutils.h"
#include "utils/rel.h"
#include "utils/sampling.h"
#include "utils/snapmgr.h"
#include "utils/tuplesort.h"
#if PG_VERSION_NUM >= 160000
#include "varatt.h"
#endif
#if PG_VERSION_NUM >= 140000
#include "utils/backend_progress.h"
#else
#elif PG_VERSION_NUM >= 120000
#include "pgstat.h"
#endif
#if PG_VERSION_NUM >= 120000
#include "access/tableam.h"
#include "commands/progress.h"
#else
#define PROGRESS_CREATEIDX_SUBPHASE 0
#define PROGRESS_CREATEIDX_TUPLES_TOTAL 0
#define PROGRESS_CREATEIDX_TUPLES_DONE 0
#endif
#if PG_VERSION_NUM >= 130000
#define CALLBACK_ITEM_POINTER ItemPointer tid
#else
#define CALLBACK_ITEM_POINTER HeapTuple hup
#endif
#if PG_VERSION_NUM >= 120000
#define UpdateProgress(index, val) pgstat_progress_update_param(index, val)
#else
#define UpdateProgress(index, val) ((void)val)
#endif
#if PG_VERSION_NUM >= 140000
#include "utils/backend_status.h"
#include "utils/wait_event.h"
#endif
#if PG_VERSION_NUM >= 120000
#include "access/table.h"
#include "optimizer/optimizer.h"
#else
#include "access/heapam.h"
#include "optimizer/planner.h"
#include "pgstat.h"
#endif
#define PARALLEL_KEY_IVFFLAT_SHARED UINT64CONST(0xA000000000000001)
#define PARALLEL_KEY_TUPLESORT UINT64CONST(0xA000000000000002)
#define PARALLEL_KEY_IVFFLAT_CENTERS UINT64CONST(0xA000000000000003)
@@ -63,24 +72,24 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
Datum value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
/*
* Check with KMEANS_NORM_PROC that the value can be normalized since
* spherical distance function expects unit vectors
* Normalize with KMEANS_NORM_PROC since spherical distance function
* expects unit vectors
*/
if (buildstate->kmeansnormprocinfo != NULL)
{
if (!IvfflatCheckNorm(buildstate->kmeansnormprocinfo, buildstate->collation, value))
if (!IvfflatNormValue(buildstate->kmeansnormprocinfo, buildstate->collation, &value, buildstate->normvec))
return;
}
if (samples->length < targsamples)
{
VectorArraySet(samples, samples->length, DatumGetPointer(value));
VectorArraySet(samples, samples->length, DatumGetVector(value));
samples->length++;
}
else
{
if (buildstate->rowstoskip < 0)
buildstate->rowstoskip = reservoir_get_next_S(&buildstate->rstate, buildstate->samplerows, targsamples);
buildstate->rowstoskip = reservoir_get_next_S(&buildstate->rstate, samples->length, targsamples);
if (buildstate->rowstoskip <= 0)
{
@@ -91,21 +100,18 @@ AddSample(Datum *values, IvfflatBuildState * buildstate)
#endif
Assert(k >= 0 && k < targsamples);
VectorArraySet(samples, k, DatumGetPointer(value));
VectorArraySet(samples, k, DatumGetVector(value));
}
buildstate->rowstoskip -= 1;
}
/* Increment after reservoir_get_next_S */
buildstate->samplerows += 1;
}
/*
* Callback for sampling
*/
static void
SampleCallback(Relation index, ItemPointer tid, Datum *values,
SampleCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
bool *isnull, bool tupleIsAlive, void *state)
{
IvfflatBuildState *buildstate = (IvfflatBuildState *) state;
@@ -119,7 +125,7 @@ SampleCallback(Relation index, ItemPointer tid, Datum *values,
oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
/* Add sample */
AddSample(values, buildstate);
AddSample(values, state);
/* Reset memory context */
MemoryContextSwitchTo(oldCtx);
@@ -135,7 +141,6 @@ SampleRows(IvfflatBuildState * buildstate)
int targsamples = buildstate->samples->maxlen;
BlockNumber totalblocks = RelationGetNumberOfBlocks(buildstate->heap);
buildstate->samplerows = 0;
buildstate->rowstoskip = -1;
BlockSampler_Init(&buildstate->bs, totalblocks, targsamples, RandomInt());
@@ -145,24 +150,13 @@ SampleRows(IvfflatBuildState * buildstate)
{
BlockNumber targblock = BlockSampler_Next(&buildstate->bs);
/* Set anyvisible to false like table_index_build_scan */
#if PG_VERSION_NUM >= 120000
table_index_build_range_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
false, false, false, targblock, 1, SampleCallback, (void *) buildstate, NULL);
}
/* Normalize if needed */
if (buildstate->kmeansnormprocinfo != NULL)
{
VectorArray samples = buildstate->samples;
for (int i = 0; i < samples->length; i++)
{
Datum value = PointerGetDatum(VectorArrayGet(samples, i));
Datum normValue = IvfflatNormValue(buildstate->typeInfo, buildstate->collation, value);
VectorArraySet(samples, i, DatumGetPointer(normValue));
pfree(DatumGetPointer(normValue));
}
false, true, false, targblock, 1, SampleCallback, (void *) buildstate, NULL);
#else
IndexBuildHeapRangeScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
false, true, targblock, 1, SampleCallback, (void *) buildstate, NULL);
#endif
}
}
@@ -170,7 +164,7 @@ SampleRows(IvfflatBuildState * buildstate)
* Add tuple to sort
*/
static void
AddTupleToSort(ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
{
double distance;
double minDistance = DBL_MAX;
@@ -184,10 +178,8 @@ AddTupleToSort(ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
/* Normalize if needed */
if (buildstate->normprocinfo != NULL)
{
if (!IvfflatCheckNorm(buildstate->normprocinfo, buildstate->collation, value))
if (!IvfflatNormValue(buildstate->normprocinfo, buildstate->collation, &value, buildstate->normvec))
return;
value = IvfflatNormValue(buildstate->typeInfo, buildstate->collation, value);
}
/* Find the list that minimizes the distance */
@@ -233,12 +225,16 @@ AddTupleToSort(ItemPointer tid, Datum *values, IvfflatBuildState * buildstate)
* Callback for table_index_build_scan
*/
static void
BuildCallback(Relation index, ItemPointer tid, Datum *values,
BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
bool *isnull, bool tupleIsAlive, void *state)
{
IvfflatBuildState *buildstate = (IvfflatBuildState *) state;
MemoryContext oldCtx;
#if PG_VERSION_NUM < 130000
ItemPointer tid = &hup->t_self;
#endif
/* Skip nulls */
if (isnull[0])
return;
@@ -247,7 +243,7 @@ BuildCallback(Relation index, ItemPointer tid, Datum *values,
oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
/* Add tuple to sort */
AddTupleToSort(tid, values, buildstate);
AddTupleToSort(index, tid, values, buildstate);
/* Reset memory context */
MemoryContextSwitchTo(oldCtx);
@@ -260,11 +256,11 @@ BuildCallback(Relation index, ItemPointer tid, Datum *values,
static inline void
GetNextTuple(Tuplesortstate *sortstate, TupleDesc tupdesc, TupleTableSlot *slot, IndexTuple *itup, int *list)
{
Datum value;
bool isnull;
if (tuplesort_gettupleslot(sortstate, true, false, slot, NULL))
{
Datum value;
bool isnull;
*list = DatumGetInt32(slot_getattr(slot, 1, &isnull));
value = slot_getattr(slot, 3, &isnull);
@@ -286,12 +282,16 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
IndexTuple itup = NULL; /* silence compiler warning */
int64 inserted = 0;
TupleTableSlot *slot = MakeSingleTupleTableSlot(buildstate->sortdesc, &TTSOpsMinimalTuple);
TupleDesc tupdesc = buildstate->tupdesc;
#if PG_VERSION_NUM >= 120000
TupleTableSlot *slot = MakeSingleTupleTableSlot(buildstate->tupdesc, &TTSOpsMinimalTuple);
#else
TupleTableSlot *slot = MakeSingleTupleTableSlot(buildstate->tupdesc);
#endif
TupleDesc tupdesc = RelationGetDescr(index);
pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_LOAD);
UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_LOAD);
pgstat_progress_update_param(PROGRESS_CREATEIDX_TUPLES_TOTAL, buildstate->indtuples);
UpdateProgress(PROGRESS_CREATEIDX_TUPLES_TOTAL, buildstate->indtuples);
GetNextTuple(buildstate->sortstate, tupdesc, slot, &itup, &list);
@@ -327,7 +327,7 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
pfree(itup);
pgstat_progress_update_param(PROGRESS_CREATEIDX_TUPLES_DONE, ++inserted);
UpdateProgress(PROGRESS_CREATEIDX_TUPLES_DONE, ++inserted);
GetNextTuple(buildstate->sortstate, tupdesc, slot, &itup, &list);
}
@@ -350,28 +350,16 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
buildstate->heap = heap;
buildstate->index = index;
buildstate->indexInfo = indexInfo;
buildstate->typeInfo = IvfflatGetTypeInfo(index);
buildstate->tupdesc = RelationGetDescr(index);
buildstate->lists = IvfflatGetLists(index);
buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
/* Disallow varbit since require fixed dimensions */
if (TupleDescAttr(index->rd_att, 0)->atttypid == VARBITOID)
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("type not supported for ivfflat index")));
/* Require column to have dimensions to be indexed */
if (buildstate->dimensions < 0)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("column does not have dimensions")));
elog(ERROR, "column does not have dimensions");
if (buildstate->dimensions > buildstate->typeInfo->maxDimensions)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("column cannot have more than %d dimensions for ivfflat index", buildstate->typeInfo->maxDimensions)));
if (buildstate->dimensions > IVFFLAT_MAX_DIM)
elog(ERROR, "column cannot have more than %d dimensions for ivfflat index", IVFFLAT_MAX_DIM);
buildstate->reltuples = 0;
buildstate->indtuples = 0;
@@ -384,25 +372,30 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
/* Require more than one dimension for spherical k-means */
if (buildstate->kmeansnormprocinfo != NULL && buildstate->dimensions == 1)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("dimensions must be greater than one for this opclass")));
elog(ERROR, "dimensions must be greater than one for this opclass");
/* Create tuple description for sorting */
buildstate->sortdesc = CreateTemplateTupleDesc(3);
TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 1, "list", INT4OID, -1, 0);
TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
TupleDescInitEntry(buildstate->sortdesc, (AttrNumber) 3, "vector", TupleDescAttr(buildstate->tupdesc, 0)->atttypid, -1, 0);
#if PG_VERSION_NUM >= 190000
TupleDescFinalize(buildstate->sortdesc);
#if PG_VERSION_NUM >= 120000
buildstate->tupdesc = CreateTemplateTupleDesc(3);
#else
buildstate->tupdesc = CreateTemplateTupleDesc(3, false);
#endif
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 1, "list", INT4OID, -1, 0);
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
TupleDescInitEntry(buildstate->tupdesc, (AttrNumber) 3, "vector", RelationGetDescr(index)->attrs[0].atttypid, -1, 0);
#if PG_VERSION_NUM >= 120000
buildstate->slot = MakeSingleTupleTableSlot(buildstate->tupdesc, &TTSOpsVirtual);
#else
buildstate->slot = MakeSingleTupleTableSlot(buildstate->tupdesc);
#endif
buildstate->slot = MakeSingleTupleTableSlot(buildstate->sortdesc, &TTSOpsVirtual);
/* TODO Ensure within maintenance_work_mem */
buildstate->centers = VectorArrayInit(buildstate->lists, buildstate->dimensions, buildstate->typeInfo->itemSize(buildstate->dimensions));
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);
@@ -424,6 +417,7 @@ FreeBuildState(IvfflatBuildState * buildstate)
{
VectorArrayFree(buildstate->centers);
pfree(buildstate->listInfo);
pfree(buildstate->normvec);
#ifdef IVFFLAT_KMEANS_DEBUG
pfree(buildstate->listSums);
@@ -441,7 +435,7 @@ ComputeCenters(IvfflatBuildState * buildstate)
{
int numSamples;
pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_KMEANS);
UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_KMEANS);
/* Target 50 samples per list, with at least 10000 samples */
/* The number of samples has a large effect on index build time */
@@ -455,10 +449,10 @@ ComputeCenters(IvfflatBuildState * buildstate)
/* Sample rows */
/* TODO Ensure within maintenance_work_mem */
buildstate->samples = VectorArrayInit(numSamples, buildstate->dimensions, buildstate->centers->itemsize);
buildstate->samples = VectorArrayInit(numSamples, buildstate->dimensions);
if (buildstate->heap != NULL)
{
IvfflatBench("sample rows", SampleRows(buildstate));
SampleRows(buildstate);
if (buildstate->samples->length < buildstate->lists)
{
@@ -470,7 +464,7 @@ ComputeCenters(IvfflatBuildState * buildstate)
}
/* Calculate centers */
IvfflatBench("k-means", IvfflatKmeans(buildstate->index, buildstate->samples, buildstate->centers, buildstate->typeInfo));
IvfflatBench("k-means", IvfflatKmeans(buildstate->index, buildstate->samples, buildstate->centers));
/* Free samples before we allocate more memory */
VectorArrayFree(buildstate->samples);
@@ -506,8 +500,8 @@ CreateMetaPage(Relation index, int dimensions, int lists, ForkNumber forkNum)
* Create list pages
*/
static void
CreateListPages(Relation index, VectorArray centers, int lists,
ForkNumber forkNum, ListInfo * *listInfo)
CreateListPages(Relation index, VectorArray centers, int dimensions,
int lists, ForkNumber forkNum, ListInfo * *listInfo)
{
Buffer buf;
Page page;
@@ -515,8 +509,8 @@ CreateListPages(Relation index, VectorArray centers, int lists,
Size listSize;
IvfflatList list;
listSize = MAXALIGN(IVFFLAT_LIST_SIZE(centers->itemsize));
list = palloc0(listSize);
listSize = MAXALIGN(IVFFLAT_LIST_SIZE(dimensions));
list = palloc(listSize);
buf = IvfflatNewBuffer(index, forkNum);
IvfflatInitRegisterPage(index, &buf, &page, &state);
@@ -525,13 +519,10 @@ CreateListPages(Relation index, VectorArray centers, int lists,
{
OffsetNumber offno;
/* Zero memory for each list */
MemSet(list, 0, listSize);
/* Load list */
list->startPage = InvalidBlockNumber;
list->insertPage = InvalidBlockNumber;
memcpy(&list->center, VectorArrayGet(centers, i), VARSIZE_ANY(VectorArrayGet(centers, i)));
memcpy(&list->center, VectorArrayGet(centers, i), VECTOR_SIZE(dimensions));
/* Ensure free space */
if (PageGetFreeSpace(page) < listSize)
@@ -552,10 +543,10 @@ CreateListPages(Relation index, VectorArray centers, int lists,
pfree(list);
}
#ifdef IVFFLAT_KMEANS_DEBUG
/*
* Print k-means metrics
*/
#ifdef IVFFLAT_KMEANS_DEBUG
static void
PrintKmeansMetrics(IvfflatBuildState * buildstate)
{
@@ -597,20 +588,6 @@ PrintKmeansMetrics(IvfflatBuildState * buildstate)
}
#endif
/*
* Initialize build sort state
*/
static Tuplesortstate *
InitBuildSortState(TupleDesc tupdesc, int memory, SortCoordinate coordinate)
{
AttrNumber attNums[] = {1};
Oid sortOperators[] = {Int4LessOperator};
Oid sortCollations[] = {InvalidOid};
bool nullsFirstFlags[] = {false};
return tuplesort_begin_heap(tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, memory, coordinate, false);
}
/*
* Within leader, wait for end of heap scan
*/
@@ -650,14 +627,24 @@ ParallelHeapScan(IvfflatBuildState * buildstate)
* Perform a worker's portion of a parallel sort
*/
static void
IvfflatParallelScanAndSort(IvfflatSpool * ivfspool, IvfflatShared * ivfshared, Sharedsort *sharedsort, char *ivfcenters, int sortmem, bool progress)
IvfflatParallelScanAndSort(IvfflatSpool * ivfspool, IvfflatShared * ivfshared, Sharedsort *sharedsort, Vector * ivfcenters, int sortmem, bool progress)
{
SortCoordinate coordinate;
IvfflatBuildState buildstate;
#if PG_VERSION_NUM >= 120000
TableScanDesc scan;
#else
HeapScanDesc scan;
#endif
double reltuples;
IndexInfo *indexInfo;
/* Sort options, which must match AssignTuples */
AttrNumber attNums[] = {1};
Oid sortOperators[] = {Int4LessOperator};
Oid sortCollations[] = {InvalidOid};
bool nullsFirstFlags[] = {false};
/* Initialize local tuplesort coordination state */
coordinate = palloc0(sizeof(SortCoordinateData));
coordinate->isWorker = true;
@@ -668,19 +655,22 @@ IvfflatParallelScanAndSort(IvfflatSpool * ivfspool, IvfflatShared * ivfshared, S
indexInfo = BuildIndexInfo(ivfspool->index);
indexInfo->ii_Concurrent = ivfshared->isconcurrent;
InitBuildState(&buildstate, ivfspool->heap, ivfspool->index, indexInfo);
memcpy(buildstate.centers->items, ivfcenters, buildstate.centers->itemsize * buildstate.centers->maxlen);
memcpy(buildstate.centers->items, ivfcenters, VECTOR_SIZE(buildstate.centers->dim) * buildstate.centers->maxlen);
buildstate.centers->length = buildstate.centers->maxlen;
ivfspool->sortstate = InitBuildSortState(buildstate.sortdesc, sortmem, coordinate);
ivfspool->sortstate = tuplesort_begin_heap(buildstate.tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, sortmem, coordinate, false);
buildstate.sortstate = ivfspool->sortstate;
#if PG_VERSION_NUM >= 120000
scan = table_beginscan_parallel(ivfspool->heap,
ParallelTableScanFromIvfflatShared(ivfshared)
#if PG_VERSION_NUM >= 190000
,SO_NONE
#endif
);
ParallelTableScanFromIvfflatShared(ivfshared));
reltuples = table_index_build_scan(ivfspool->heap, ivfspool->index, indexInfo,
true, progress, BuildCallback,
(void *) &buildstate, scan);
#else
scan = heap_beginscan_parallel(ivfspool->heap, &ivfshared->heapdesc);
reltuples = IndexBuildHeapScan(ivfspool->heap, ivfspool->index, indexInfo,
true, BuildCallback,
(void *) &buildstate, scan);
#endif
/* Execute this worker's part of the sort */
tuplesort_performsort(ivfspool->sortstate);
@@ -720,7 +710,7 @@ IvfflatParallelBuildMain(dsm_segment *seg, shm_toc *toc)
IvfflatSpool *ivfspool;
IvfflatShared *ivfshared;
Sharedsort *sharedsort;
char *ivfcenters;
Vector *ivfcenters;
Relation heapRel;
Relation indexRel;
LOCKMODE heapLockmode;
@@ -750,7 +740,11 @@ IvfflatParallelBuildMain(dsm_segment *seg, shm_toc *toc)
}
/* Open relations within worker */
#if PG_VERSION_NUM >= 120000
heapRel = table_open(ivfshared->heaprelid, heapLockmode);
#else
heapRel = heap_open(ivfshared->heaprelid, heapLockmode);
#endif
indexRel = index_open(ivfshared->indexrelid, indexLockmode);
/* Initialize worker's own spool */
@@ -770,7 +764,11 @@ IvfflatParallelBuildMain(dsm_segment *seg, shm_toc *toc)
/* Close relations within worker */
index_close(indexRel, indexLockmode);
#if PG_VERSION_NUM >= 120000
table_close(heapRel, heapLockmode);
#else
heap_close(heapRel, heapLockmode);
#endif
}
/*
@@ -795,7 +793,19 @@ IvfflatEndParallel(IvfflatLeader * ivfleader)
static Size
ParallelEstimateShared(Relation heap, Snapshot snapshot)
{
#if PG_VERSION_NUM >= 120000
return add_size(BUFFERALIGN(sizeof(IvfflatShared)), table_parallelscan_estimate(heap, snapshot));
#else
if (!IsMVCCSnapshot(snapshot))
{
Assert(snapshot == SnapshotAny);
return sizeof(IvfflatShared);
}
return add_size(offsetof(IvfflatShared, heapdesc) +
offsetof(ParallelHeapScanDescData, phs_snapshot_data),
EstimateSnapshotSpace(snapshot));
#endif
}
/*
@@ -834,7 +844,7 @@ IvfflatBeginParallel(IvfflatBuildState * buildstate, bool isconcurrent, int requ
Size estcenters;
IvfflatShared *ivfshared;
Sharedsort *sharedsort;
char *ivfcenters;
Vector *ivfcenters;
IvfflatLeader *ivfleader = (IvfflatLeader *) palloc0(sizeof(IvfflatLeader));
bool leaderparticipates = true;
int querylen;
@@ -846,7 +856,11 @@ IvfflatBeginParallel(IvfflatBuildState * buildstate, bool isconcurrent, int requ
/* Enter parallel mode and create context */
EnterParallelMode();
Assert(request > 0);
#if PG_VERSION_NUM >= 120000
pcxt = CreateParallelContext("vector", "IvfflatParallelBuildMain", request);
#else
pcxt = CreateParallelContext("vector", "IvfflatParallelBuildMain", request, true);
#endif
scantuplesortstates = leaderparticipates ? request + 1 : request;
@@ -861,7 +875,7 @@ IvfflatBeginParallel(IvfflatBuildState * buildstate, bool isconcurrent, int requ
shm_toc_estimate_chunk(&pcxt->estimator, estivfshared);
estsort = tuplesort_estimate_shared(scantuplesortstates);
shm_toc_estimate_chunk(&pcxt->estimator, estsort);
estcenters = buildstate->centers->itemsize * buildstate->centers->maxlen;
estcenters = VECTOR_SIZE(buildstate->dimensions) * buildstate->lists;
shm_toc_estimate_chunk(&pcxt->estimator, estcenters);
shm_toc_estimate_keys(&pcxt->estimator, 3);
@@ -904,16 +918,20 @@ IvfflatBeginParallel(IvfflatBuildState * buildstate, bool isconcurrent, int requ
#ifdef IVFFLAT_KMEANS_DEBUG
ivfshared->inertia = 0;
#endif
#if PG_VERSION_NUM >= 120000
table_parallelscan_initialize(buildstate->heap,
ParallelTableScanFromIvfflatShared(ivfshared),
snapshot);
#else
heap_parallelscan_initialize(&ivfshared->heapdesc, buildstate->heap, snapshot);
#endif
/* Store shared tuplesort-private state, for which we reserved space */
sharedsort = (Sharedsort *) shm_toc_allocate(pcxt->toc, estsort);
tuplesort_initialize_shared(sharedsort, scantuplesortstates,
pcxt->seg);
ivfcenters = shm_toc_allocate(pcxt->toc, estcenters);
ivfcenters = (Vector *) shm_toc_allocate(pcxt->toc, estcenters);
memcpy(ivfcenters, buildstate->centers->items, estcenters);
shm_toc_insert(pcxt->toc, PARALLEL_KEY_IVFFLAT_SHARED, ivfshared);
@@ -971,7 +989,13 @@ AssignTuples(IvfflatBuildState * buildstate)
int parallel_workers = 0;
SortCoordinate coordinate = NULL;
pgstat_progress_update_param(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_ASSIGN);
/* Sort options, which must match IvfflatParallelScanAndSort */
AttrNumber attNums[] = {1};
Oid sortOperators[] = {Int4LessOperator};
Oid sortCollations[] = {InvalidOid};
bool nullsFirstFlags[] = {false};
UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_ASSIGN);
/* Calculate parallel workers */
if (buildstate->heap != NULL)
@@ -991,7 +1015,7 @@ AssignTuples(IvfflatBuildState * buildstate)
}
/* Begin serial/leader tuplesort */
buildstate->sortstate = InitBuildSortState(buildstate->sortdesc, maintenance_work_mem, coordinate);
buildstate->sortstate = tuplesort_begin_heap(buildstate->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, maintenance_work_mem, coordinate, false);
/* Add tuples to sort */
if (buildstate->heap != NULL)
@@ -999,8 +1023,15 @@ AssignTuples(IvfflatBuildState * buildstate)
if (buildstate->ivfleader)
buildstate->reltuples = ParallelHeapScan(buildstate);
else
{
#if PG_VERSION_NUM >= 120000
buildstate->reltuples = table_index_build_scan(buildstate->heap, buildstate->index, buildstate->indexInfo,
true, true, BuildCallback, (void *) buildstate, NULL);
#else
buildstate->reltuples = IndexBuildHeapScan(buildstate->heap, buildstate->index, buildstate->indexInfo,
true, BuildCallback, (void *) buildstate, NULL);
#endif
}
#ifdef IVFFLAT_KMEANS_DEBUG
PrintKmeansMetrics(buildstate);
@@ -1044,13 +1075,9 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
/* Create pages */
CreateMetaPage(index, buildstate->dimensions, buildstate->lists, forkNum);
CreateListPages(index, buildstate->centers, buildstate->lists, forkNum, &buildstate->listInfo);
CreateListPages(index, buildstate->centers, buildstate->dimensions, buildstate->lists, forkNum, &buildstate->listInfo);
CreateEntryPages(buildstate, forkNum);
/* Write WAL for initialization fork since GenericXLog functions do not */
if (forkNum == INIT_FORKNUM)
log_newpage_range(index, forkNum, 0, RelationGetNumberOfBlocksInFork(index, forkNum), true);
FreeBuildState(buildstate);
}
@@ -1063,10 +1090,6 @@ 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));

View File

@@ -3,35 +3,19 @@
#include <float.h>
#include "access/amapi.h"
#include "access/genam.h"
#include "access/reloptions.h"
#include "commands/progress.h"
#include "commands/vacuum.h"
#include "fmgr.h"
#include "ivfflat.h"
#include "nodes/pg_list.h"
#include "utils/float.h"
#include "utils/guc.h"
#include "utils/relcache.h"
#include "utils/selfuncs.h"
#include "utils/spccache.h"
#include "vector.h"
#if PG_VERSION_NUM < 150000
#define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x)
#if PG_VERSION_NUM >= 120000
#include "commands/progress.h"
#endif
int ivfflat_probes;
int ivfflat_iterative_scan;
int ivfflat_max_probes;
static relopt_kind ivfflat_relopt_kind;
static const struct config_enum_entry ivfflat_iterative_scan_options[] = {
{"off", IVFFLAT_ITERATIVE_SCAN_OFF, false},
{"relaxed_order", IVFFLAT_ITERATIVE_SCAN_RELAXED, false},
{NULL, 0, false}
};
/*
* Initialize index options and variables
*/
@@ -40,27 +24,21 @@ IvfflatInit(void)
{
ivfflat_relopt_kind = add_reloption_kind();
add_int_reloption(ivfflat_relopt_kind, "lists", "Number of inverted lists",
IVFFLAT_DEFAULT_LISTS, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS, AccessExclusiveLock);
IVFFLAT_DEFAULT_LISTS, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS
#if PG_VERSION_NUM >= 130000
,AccessExclusiveLock
#endif
);
DefineCustomIntVariable("ivfflat.probes", "Sets the number of probes",
"Valid range is 1..lists.", &ivfflat_probes,
IVFFLAT_DEFAULT_PROBES, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS, PGC_USERSET, 0, NULL, NULL, NULL);
DefineCustomEnumVariable("ivfflat.iterative_scan", "Sets the mode for iterative scans",
NULL, &ivfflat_iterative_scan,
IVFFLAT_ITERATIVE_SCAN_OFF, ivfflat_iterative_scan_options, PGC_USERSET, 0, NULL, NULL, NULL);
/* If this is less than probes, probes is used */
DefineCustomIntVariable("ivfflat.max_probes", "Sets the max number of probes for iterative scans",
NULL, &ivfflat_max_probes,
IVFFLAT_MAX_LISTS, IVFFLAT_MIN_LISTS, IVFFLAT_MAX_LISTS, PGC_USERSET, 0, NULL, NULL, NULL);
MarkGUCPrefixReserved("ivfflat");
}
/*
* Get the name of index build phase
*/
#if PG_VERSION_NUM >= 120000
static char *
ivfflatbuildphasename(int64 phasenum)
{
@@ -78,6 +56,7 @@ ivfflatbuildphasename(int64 phasenum)
return NULL;
}
}
#endif
/*
* Estimate the cost of an index scan
@@ -91,30 +70,25 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
GenericCosts costs;
int lists;
double ratio;
double sequentialRatio = 0.5;
double startupPages;
double spc_seq_page_cost;
Relation index;
#if PG_VERSION_NUM < 120000
List *qinfos;
#endif
/* Never use index without order */
if (path->indexorderbys == NIL)
if (path->indexorderbys == NULL)
{
*indexStartupCost = get_float8_infinity();
*indexTotalCost = get_float8_infinity();
*indexStartupCost = DBL_MAX;
*indexTotalCost = DBL_MAX;
*indexSelectivity = 0;
*indexCorrelation = 0;
*indexPages = 0;
#if PG_VERSION_NUM >= 180000
/* See "On disable_cost" thread on pgsql-hackers */
path->path.disabled_nodes = 2;
#endif
return;
}
MemSet(&costs, 0, sizeof(costs));
genericcostestimate(root, path, loop_count, &costs);
index = index_open(path->indexinfo->indexoid, NoLock);
IvfflatGetMetaPageInfo(index, &lists, NULL);
index_close(index, NoLock);
@@ -124,26 +98,46 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
if (ratio > 1.0)
ratio = 1.0;
/*
* This gives us the subset of tuples to visit. This value is passed into
* the generic cost estimator to determine the number of pages to visit
* during the index scan.
*/
costs.numIndexTuples = path->indexinfo->tuples * ratio;
#if PG_VERSION_NUM >= 120000
genericcostestimate(root, path, loop_count, &costs);
#else
qinfos = deconstruct_indexquals(path);
genericcostestimate(root, path, loop_count, qinfos, &costs);
#endif
get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
/* Change some page cost from random to sequential */
costs.indexTotalCost -= sequentialRatio * costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
/* Startup cost is cost before returning the first row */
costs.indexStartupCost = costs.indexTotalCost * ratio;
/* Adjust cost if needed since TOAST not included in seq scan cost */
startupPages = costs.numIndexPages * ratio;
if (startupPages > path->indexinfo->rel->pages && ratio < 0.5)
if (costs.numIndexPages > path->indexinfo->rel->pages && ratio < 0.5)
{
/* Change rest of page cost from random to sequential */
costs.indexStartupCost -= (1 - sequentialRatio) * startupPages * (costs.spc_random_page_cost - spc_seq_page_cost);
/* Change all page cost from random to sequential */
costs.indexTotalCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
/* Remove cost of extra pages */
costs.indexStartupCost -= (startupPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
costs.indexTotalCost -= (costs.numIndexPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
}
else
{
/* Change some page cost from random to sequential */
costs.indexTotalCost -= 0.5 * costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
}
*indexStartupCost = costs.indexStartupCost;
/*
* If the list selectivity is lower than what is returned from the generic
* cost estimator, use that.
*/
if (ratio < costs.indexSelectivity)
costs.indexSelectivity = ratio;
/* Use total cost since most work happens before first tuple is returned */
*indexStartupCost = costs.indexTotalCost;
*indexTotalCost = costs.indexTotalCost;
*indexSelectivity = costs.indexSelectivity;
*indexCorrelation = costs.indexCorrelation;
@@ -160,10 +154,23 @@ ivfflatoptions(Datum reloptions, bool validate)
{"lists", RELOPT_TYPE_INT, offsetof(IvfflatOptions, lists)},
};
#if PG_VERSION_NUM >= 130000
return (bytea *) build_reloptions(reloptions, validate,
ivfflat_relopt_kind,
sizeof(IvfflatOptions),
tab, lengthof(tab));
#else
relopt_value *options;
int numoptions;
IvfflatOptions *rdopts;
options = parseRelOptions(reloptions, validate, ivfflat_relopt_kind, &numoptions);
rdopts = allocateReloptStruct(sizeof(IvfflatOptions), options, numoptions);
fillRelOptions((void *) rdopts, sizeof(IvfflatOptions), options, numoptions,
validate, tab, lengthof(tab));
return (bytea *) rdopts;
#endif
}
/*
@@ -180,80 +187,19 @@ ivfflatvalidate(Oid opclassoid)
*
* See https://www.postgresql.org/docs/current/index-api.html
*/
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflathandler);
PGDLLEXPORT PG_FUNCTION_INFO_V1(ivfflathandler);
Datum
ivfflathandler(PG_FUNCTION_ARGS)
{
#if PG_VERSION_NUM >= 190000
static const IndexAmRoutine amroutine = {
.type = T_IndexAmRoutine,
.amstrategies = 0,
.amsupport = 5,
.amoptsprocnum = 0,
.amcanorder = false,
.amcanorderbyop = true,
.amcanhash = false,
.amconsistentequality = false,
.amconsistentordering = false,
.amcanbackward = false,
.amcanunique = false,
.amcanmulticol = false,
.amoptionalkey = true,
.amsearcharray = false,
.amsearchnulls = false,
.amstorage = false,
.amclusterable = false,
.ampredlocks = false,
.amcanparallel = false,
.amcanbuildparallel = true,
.amcaninclude = false,
.amusemaintenanceworkmem = false,
.amsummarizing = false,
.amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL,
.amkeytype = InvalidOid,
.ambuild = ivfflatbuild,
.ambuildempty = ivfflatbuildempty,
.aminsert = ivfflatinsert,
.aminsertcleanup = NULL,
.ambulkdelete = ivfflatbulkdelete,
.amvacuumcleanup = ivfflatvacuumcleanup,
.amcanreturn = NULL,
.amcostestimate = ivfflatcostestimate,
.amgettreeheight = NULL,
.amoptions = ivfflatoptions,
.amproperty = NULL,
.ambuildphasename = ivfflatbuildphasename,
.amvalidate = ivfflatvalidate,
.amadjustmembers = NULL,
.ambeginscan = ivfflatbeginscan,
.amrescan = ivfflatrescan,
.amgettuple = ivfflatgettuple,
.amgetbitmap = NULL,
.amendscan = ivfflatendscan,
.ammarkpos = NULL,
.amrestrpos = NULL,
.amestimateparallelscan = NULL,
.aminitparallelscan = NULL,
.amparallelrescan = NULL,
.amtranslatestrategy = NULL,
.amtranslatecmptype = NULL,
};
PG_RETURN_POINTER(&amroutine);
#else
IndexAmRoutine *amroutine = makeNode(IndexAmRoutine);
amroutine->amstrategies = 0;
amroutine->amsupport = 5;
amroutine->amsupport = 4;
#if PG_VERSION_NUM >= 130000
amroutine->amoptsprocnum = 0;
#endif
amroutine->amcanorder = false;
amroutine->amcanorderbyop = true;
#if PG_VERSION_NUM >= 180000
amroutine->amcanhash = false;
amroutine->amconsistentequality = false;
amroutine->amconsistentordering = false;
#endif
amroutine->amcanbackward = false; /* can change direction mid-scan */
amroutine->amcanunique = false;
amroutine->amcanmulticol = false;
@@ -264,34 +210,26 @@ ivfflathandler(PG_FUNCTION_ARGS)
amroutine->amclusterable = false;
amroutine->ampredlocks = false;
amroutine->amcanparallel = false;
#if PG_VERSION_NUM >= 170000
amroutine->amcanbuildparallel = true;
#endif
amroutine->amcaninclude = false;
#if PG_VERSION_NUM >= 130000
amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
#if PG_VERSION_NUM >= 160000
amroutine->amsummarizing = false;
#endif
amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
#endif
amroutine->amkeytype = InvalidOid;
/* Interface functions */
amroutine->ambuild = ivfflatbuild;
amroutine->ambuildempty = ivfflatbuildempty;
amroutine->aminsert = ivfflatinsert;
#if PG_VERSION_NUM >= 170000
amroutine->aminsertcleanup = NULL;
#endif
amroutine->ambulkdelete = ivfflatbulkdelete;
amroutine->amvacuumcleanup = ivfflatvacuumcleanup;
amroutine->amcanreturn = NULL; /* tuple not included in heapsort */
amroutine->amcostestimate = ivfflatcostestimate;
#if PG_VERSION_NUM >= 180000
amroutine->amgettreeheight = NULL;
#endif
amroutine->amoptions = ivfflatoptions;
amroutine->amproperty = NULL; /* TODO AMPROP_DISTANCE_ORDERABLE */
#if PG_VERSION_NUM >= 120000
amroutine->ambuildphasename = ivfflatbuildphasename;
#endif
amroutine->amvalidate = ivfflatvalidate;
#if PG_VERSION_NUM >= 140000
amroutine->amadjustmembers = NULL;
@@ -309,11 +247,5 @@ ivfflathandler(PG_FUNCTION_ARGS)
amroutine->aminitparallelscan = NULL;
amroutine->amparallelrescan = NULL;
#if PG_VERSION_NUM >= 180000
amroutine->amtranslatestrategy = NULL;
amroutine->amtranslatecmptype = NULL;
#endif
PG_RETURN_POINTER(amroutine);
#endif
}

View File

@@ -3,31 +3,25 @@
#include "postgres.h"
#include "access/genam.h"
#include "access/generic_xlog.h"
#include "access/parallel.h"
#include "lib/pairingheap.h"
#include "access/reloptions.h"
#include "nodes/execnodes.h"
#include "port.h" /* for random() */
#include "storage/condition_variable.h"
#include "utils/sampling.h"
#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
#ifdef IVFFLAT_BENCH
#include "portability/instr_time.h"
#if PG_VERSION_NUM < 120000
#include "access/relscan.h"
#endif
#if PG_VERSION_NUM >= 190000
typedef Pointer Item;
#ifdef IVFFLAT_BENCH
#include "portability/instr_time.h"
#endif
#define IVFFLAT_MAX_DIM 2000
@@ -37,7 +31,6 @@ typedef Pointer Item;
#define IVFFLAT_NORM_PROC 2
#define IVFFLAT_KMEANS_DISTANCE_PROC 3
#define IVFFLAT_KMEANS_NORM_PROC 4
#define IVFFLAT_TYPE_INFO_PROC 5
#define IVFFLAT_VERSION 1
#define IVFFLAT_MAGIC_NUMBER 0x14FF1A7
@@ -59,7 +52,7 @@ typedef Pointer Item;
#define PROGRESS_IVFFLAT_PHASE_ASSIGN 3
#define PROGRESS_IVFFLAT_PHASE_LOAD 4
#define IVFFLAT_LIST_SIZE(size) (offsetof(IvfflatListData, center) + size)
#define IVFFLAT_LIST_SIZE(_dim) (offsetof(IvfflatListData, center) + VECTOR_SIZE(_dim))
#define IvfflatPageGetOpaque(page) ((IvfflatPageOpaque) PageGetSpecialPointer(page))
#define IvfflatPageGetMeta(page) ((IvfflatMetaPageData *) PageGetContents(page))
@@ -82,31 +75,20 @@ typedef Pointer Item;
#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 */
extern int ivfflat_probes;
extern int ivfflat_iterative_scan;
extern int ivfflat_max_probes;
typedef enum IvfflatIterativeScanMode
{
IVFFLAT_ITERATIVE_SCAN_OFF,
IVFFLAT_ITERATIVE_SCAN_RELAXED
} IvfflatIterativeScanMode;
typedef struct VectorArrayData
{
int length;
int maxlen;
int dim;
Size itemsize;
char *items;
Vector *items;
} VectorArrayData;
typedef VectorArrayData * VectorArray;
@@ -153,10 +135,16 @@ typedef struct IvfflatShared
#ifdef IVFFLAT_KMEANS_DEBUG
double inertia;
#endif
#if PG_VERSION_NUM < 120000
ParallelHeapScanDescData heapdesc; /* must come last */
#endif
} IvfflatShared;
#if PG_VERSION_NUM >= 120000
#define ParallelTableScanFromIvfflatShared(shared) \
(ParallelTableScanDesc) ((char *) (shared) + BUFFERALIGN(sizeof(IvfflatShared)))
#endif
typedef struct IvfflatLeader
{
@@ -165,26 +153,15 @@ typedef struct IvfflatLeader
IvfflatShared *ivfshared;
Sharedsort *sharedsort;
Snapshot snapshot;
char *ivfcenters;
Vector *ivfcenters;
} IvfflatLeader;
typedef struct IvfflatTypeInfo
{
int maxDimensions;
Datum (*normalize) (PG_FUNCTION_ARGS);
Size (*itemSize) (int dimensions);
void (*updateCenter) (Pointer v, int dimensions, float *x);
void (*sumCenter) (Pointer v, float *x);
} IvfflatTypeInfo;
typedef struct IvfflatBuildState
{
/* Info */
Relation heap;
Relation index;
IndexInfo *indexInfo;
const IvfflatTypeInfo *typeInfo;
TupleDesc tupdesc;
/* Settings */
int dimensions;
@@ -204,6 +181,7 @@ typedef struct IvfflatBuildState
VectorArray samples;
VectorArray centers;
ListInfo *listInfo;
Vector *normvec;
#ifdef IVFFLAT_KMEANS_DEBUG
double inertia;
@@ -214,12 +192,11 @@ typedef struct IvfflatBuildState
/* Sampling */
BlockSamplerData bs;
ReservoirStateData rstate;
double samplerows;
double rowstoskip;
int rowstoskip;
/* Sorting */
Tuplesortstate *sortstate;
TupleDesc sortdesc;
TupleDesc tupdesc;
TupleTableSlot *slot;
/* Memory */
@@ -266,59 +243,40 @@ typedef struct IvfflatScanList
typedef struct IvfflatScanOpaqueData
{
const IvfflatTypeInfo *typeInfo;
int probes;
int maxProbes;
int dimensions;
bool first;
Datum value;
MemoryContext tmpCtx;
/* Sorting */
Tuplesortstate *sortstate;
TupleDesc tupdesc;
TupleTableSlot *vslot;
TupleTableSlot *mslot;
BufferAccessStrategy bas;
TupleTableSlot *slot;
bool isnull;
/* Support functions */
FmgrInfo *procinfo;
FmgrInfo *normprocinfo;
Oid collation;
Datum (*distfunc) (FmgrInfo *flinfo, Oid collation, Datum arg1, Datum arg2);
/* Lists */
pairingheap *listQueue;
BlockNumber *listPages;
int listIndex;
IvfflatScanList *lists;
IvfflatScanList lists[FLEXIBLE_ARRAY_MEMBER]; /* must come last */
} IvfflatScanOpaqueData;
typedef IvfflatScanOpaqueData * IvfflatScanOpaque;
#define VECTOR_ARRAY_SIZE(_length, _size) (sizeof(VectorArrayData) + (_length) * MAXALIGN(_size))
/* Use functions instead of macros to avoid double evaluation */
static inline Pointer
VectorArrayGet(VectorArray arr, int offset)
{
return ((char *) arr->items) + (offset * arr->itemsize);
}
static inline void
VectorArraySet(VectorArray arr, int offset, Pointer val)
{
memcpy(VectorArrayGet(arr, offset), val, VARSIZE_ANY(val));
}
#define VECTOR_ARRAY_SIZE(_length, _dim) (sizeof(VectorArrayData) + (_length) * VECTOR_SIZE(_dim))
#define VECTOR_ARRAY_OFFSET(_arr, _offset) ((char*) (_arr)->items + (_offset) * VECTOR_SIZE((_arr)->dim))
#define VectorArrayGet(_arr, _offset) ((Vector *) VECTOR_ARRAY_OFFSET(_arr, _offset))
#define VectorArraySet(_arr, _offset, _val) memcpy(VECTOR_ARRAY_OFFSET(_arr, _offset), _val, VECTOR_SIZE((_arr)->dim))
/* Methods */
VectorArray VectorArrayInit(int maxlen, int dimensions, Size itemsize);
VectorArray VectorArrayInit(int maxlen, int dimensions);
void VectorArrayFree(VectorArray arr);
void IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers, const IvfflatTypeInfo * typeInfo);
void PrintVectorArray(char *msg, VectorArray arr);
void IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers);
FmgrInfo *IvfflatOptionalProcInfo(Relation index, uint16 procnum);
Datum IvfflatNormValue(const IvfflatTypeInfo * typeInfo, Oid collation, Datum value);
bool IvfflatCheckNorm(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);
@@ -328,7 +286,6 @@ Buffer IvfflatNewBuffer(Relation index, ForkNumber forkNum);
void IvfflatInitPage(Buffer buf, Page page);
void IvfflatInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state);
void IvfflatInit(void);
const IvfflatTypeInfo *IvfflatGetTypeInfo(Relation index);
PGDLLEXPORT void IvfflatParallelBuildMain(dsm_segment *seg, shm_toc *toc);
/* Index access methods */

View File

@@ -2,16 +2,10 @@
#include <float.h>
#include "access/genam.h"
#include "access/generic_xlog.h"
#include "access/itup.h"
#include "fmgr.h"
#include "ivfflat.h"
#include "nodes/execnodes.h"
#include "storage/bufmgr.h"
#include "storage/lmgr.h"
#include "utils/memutils.h"
#include "utils/rel.h"
/*
* Find the list that minimizes the distance function
@@ -70,9 +64,8 @@ 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)
InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel)
{
const IvfflatTypeInfo *typeInfo = IvfflatGetTypeInfo(index);
IndexTuple itup;
Datum value;
FmgrInfo *normprocinfo;
@@ -91,19 +84,12 @@ InsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid)
normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
if (normprocinfo != NULL)
{
Oid collation = index->rd_indcollation[0];
if (!IvfflatCheckNorm(normprocinfo, collation, value))
if (!IvfflatNormValue(normprocinfo, index->rd_indcollation[0], &value, NULL))
return;
value = IvfflatNormValue(typeInfo, collation, value);
}
/* Ensure index is valid */
IvfflatGetMetaPageInfo(index, NULL, NULL);
/* Find the insert page - sets the page and list info */
FindInsertPage(index, &value, &insertPage, &listInfo);
FindInsertPage(index, values, &insertPage, &listInfo);
Assert(BlockNumberIsValid(insertPage));
originalInsertPage = insertPage;
@@ -209,7 +195,7 @@ ivfflatinsert(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid,
oldCtx = MemoryContextSwitchTo(insertCtx);
/* Insert tuple */
InsertTuple(index, values, isnull, heap_tid);
InsertTuple(index, values, isnull, heap_tid, heap);
/* Delete memory context */
MemoryContextSwitchTo(oldCtx);

View File

@@ -1,19 +1,10 @@
#include "postgres.h"
#include <float.h>
#include <limits.h>
#include <math.h>
#include "access/genam.h"
#include "fmgr.h"
#include "ivfflat.h"
#include "miscadmin.h"
#include "utils/memutils.h"
#include "utils/relcache.h"
#if PG_VERSION_NUM >= 160000
#include "varatt.h"
#endif
/*
* Initialize with kmeans++
@@ -51,12 +42,12 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, float *low
for (j = 0; j < numSamples; j++)
{
Datum vec = PointerGetDatum(VectorArrayGet(samples, j));
Vector *vec = VectorArrayGet(samples, j);
double distance;
/* Only need to compute distance for new center */
/* TODO Use triangle inequality to reduce distance calculations */
distance = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, vec, PointerGetDatum(VectorArrayGet(centers, i))));
distance = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(vec), PointerGetDatum(VectorArrayGet(centers, i))));
/* Set lower bound */
lowerBound[j * numCenters + i] = distance;
@@ -91,162 +82,73 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, float *low
}
/*
* Norm centers
* Apply norm to vector
*/
static void
NormCenters(const IvfflatTypeInfo * typeInfo, Oid collation, VectorArray centers)
static inline void
ApplyNorm(FmgrInfo *normprocinfo, Oid collation, Vector * vec)
{
MemoryContext normCtx = AllocSetContextCreate(CurrentMemoryContext,
"Ivfflat norm temporary context",
ALLOCSET_DEFAULT_SIZES);
MemoryContext oldCtx = MemoryContextSwitchTo(normCtx);
double norm = DatumGetFloat8(FunctionCall1Coll(normprocinfo, collation, PointerGetDatum(vec)));
for (int j = 0; j < centers->length; j++)
/* TODO Handle zero norm */
if (norm > 0)
{
Datum center = PointerGetDatum(VectorArrayGet(centers, j));
Datum newCenter = IvfflatNormValue(typeInfo, collation, center);
Size size = VARSIZE_ANY(DatumGetPointer(newCenter));
if (size > centers->itemsize)
elog(ERROR, "safety check failed");
memcpy(DatumGetPointer(center), DatumGetPointer(newCenter), size);
MemoryContextReset(normCtx);
for (int i = 0; i < vec->dim; i++)
vec->x[i] /= norm;
}
MemoryContextSwitchTo(oldCtx);
MemoryContextDelete(normCtx);
}
/*
* Quick approach if we have no data
* Compare vectors
*/
static int
CompareVectors(const void *a, const void *b)
{
return vector_cmp_internal((Vector *) a, (Vector *) b);
}
/*
* Quick approach if we have little data
*/
static void
RandomCenters(Relation index, VectorArray centers, const IvfflatTypeInfo * typeInfo)
QuickCenters(Relation index, VectorArray samples, VectorArray centers)
{
int dimensions = centers->dim;
FmgrInfo *normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_KMEANS_NORM_PROC);
Oid collation = index->rd_indcollation[0];
float *x = (float *) palloc(sizeof(float) * dimensions);
FmgrInfo *normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_KMEANS_NORM_PROC);
/* Fill with random data */
/* Copy existing vectors while avoiding duplicates */
if (samples->length > 0)
{
qsort(samples->items, samples->length, VECTOR_SIZE(samples->dim), CompareVectors);
for (int i = 0; i < samples->length; i++)
{
Vector *vec = VectorArrayGet(samples, i);
if (i == 0 || CompareVectors(vec, VectorArrayGet(samples, i - 1)) != 0)
{
VectorArraySet(centers, centers->length, vec);
centers->length++;
}
}
}
/* Fill remaining with random data */
while (centers->length < centers->maxlen)
{
Pointer center = VectorArrayGet(centers, centers->length);
Vector *vec = VectorArrayGet(centers, centers->length);
for (int i = 0; i < dimensions; i++)
x[i] = (float) RandomDouble();
SET_VARSIZE(vec, VECTOR_SIZE(dimensions));
vec->dim = dimensions;
typeInfo->updateCenter(center, dimensions, x);
for (int j = 0; j < dimensions; j++)
vec->x[j] = RandomDouble();
/* Normalize if needed (only needed for random centers) */
if (normprocinfo != NULL)
ApplyNorm(normprocinfo, collation, vec);
centers->length++;
}
if (normprocinfo != NULL)
NormCenters(typeInfo, collation, centers);
}
#ifdef IVFFLAT_MEMORY
/*
* Show memory usage
*/
static void
ShowMemoryUsage(MemoryContext context, Size estimatedSize)
{
elog(INFO, "total memory: %zu MB",
MemoryContextMemAllocated(context, true) / (1024 * 1024));
elog(INFO, "estimated memory: %zu MB", estimatedSize / (1024 * 1024));
}
#endif
/*
* Sum centers
*/
static void
SumCenters(VectorArray samples, float *agg, int *closestCenters, const IvfflatTypeInfo * typeInfo)
{
for (int j = 0; j < samples->length; j++)
{
float *x = agg + ((int64) closestCenters[j] * samples->dim);
typeInfo->sumCenter(VectorArrayGet(samples, j), x);
}
}
/*
* Update centers
*/
static void
UpdateCenters(float *agg, VectorArray centers, const IvfflatTypeInfo * typeInfo)
{
for (int j = 0; j < centers->length; j++)
{
float *x = agg + ((int64) j * centers->dim);
typeInfo->updateCenter(VectorArrayGet(centers, j), centers->dim, x);
}
}
/*
* Compute new centers
*/
static void
ComputeNewCenters(VectorArray samples, float *agg, VectorArray newCenters, int *centerCounts, int *closestCenters, FmgrInfo *normprocinfo, Oid collation, const IvfflatTypeInfo * typeInfo)
{
int dimensions = newCenters->dim;
int numCenters = newCenters->length;
int numSamples = samples->length;
/* Reset sum and count */
for (int j = 0; j < numCenters; j++)
{
float *x = agg + ((int64) j * dimensions);
for (int k = 0; k < dimensions; k++)
x[k] = 0.0;
centerCounts[j] = 0;
}
/* Increment sum of closest center */
SumCenters(samples, agg, closestCenters, typeInfo);
/* Increment count of closest center */
for (int j = 0; j < numSamples; j++)
centerCounts[closestCenters[j]] += 1;
/* Divide sum by count */
for (int j = 0; j < numCenters; j++)
{
float *x = agg + ((int64) j * dimensions);
if (centerCounts[j] > 0)
{
/* Double avoids overflow, but requires more memory */
/* TODO Update bounds */
for (int k = 0; k < dimensions; k++)
{
if (isinf(x[k]))
x[k] = x[k] > 0 ? FLT_MAX : -FLT_MAX;
}
for (int k = 0; k < dimensions; k++)
x[k] /= centerCounts[j];
}
else
{
/* TODO Handle empty centers properly */
for (int k = 0; k < dimensions; k++)
x[k] = RandomDouble();
}
}
/* Set new centers */
UpdateCenters(agg, newCenters, typeInfo);
/* Normalize if needed */
if (normprocinfo != NULL)
NormCenters(typeInfo, collation, newCenters);
}
/*
@@ -258,16 +160,19 @@ ComputeNewCenters(VectorArray samples, float *agg, VectorArray newCenters, int *
* https://www.aaai.org/Papers/ICML/2003/ICML03-022.pdf
*/
static void
ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const IvfflatTypeInfo * typeInfo)
ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
{
FmgrInfo *procinfo;
FmgrInfo *normprocinfo;
Oid collation;
Vector *vec;
Vector *newCenter;
int64 j;
int64 k;
int dimensions = centers->dim;
int numCenters = centers->maxlen;
int numSamples = samples->length;
VectorArray newCenters;
float *agg;
int *centerCounts;
int *closestCenters;
float *lowerBound;
@@ -277,10 +182,9 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
float *newcdist;
/* Calculate allocation sizes */
Size samplesSize = VECTOR_ARRAY_SIZE(samples->maxlen, samples->itemsize);
Size centersSize = VECTOR_ARRAY_SIZE(centers->maxlen, centers->itemsize);
Size newCentersSize = VECTOR_ARRAY_SIZE(numCenters, centers->itemsize);
Size aggSize = sizeof(float) * (int64) numCenters * dimensions;
Size samplesSize = VECTOR_ARRAY_SIZE(samples->maxlen, samples->dim);
Size centersSize = VECTOR_ARRAY_SIZE(centers->maxlen, centers->dim);
Size newCentersSize = VECTOR_ARRAY_SIZE(numCenters, dimensions);
Size centerCountsSize = sizeof(int) * numCenters;
Size closestCentersSize = sizeof(int) * numSamples;
Size lowerBoundSize = sizeof(float) * numSamples * numCenters;
@@ -290,7 +194,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
Size newcdistSize = sizeof(float) * numCenters;
/* Calculate total size */
Size totalSize = samplesSize + centersSize + newCentersSize + aggSize + centerCountsSize + closestCentersSize + lowerBoundSize + upperBoundSize + sSize + halfcdistSize + newcdistSize;
Size totalSize = samplesSize + centersSize + newCentersSize + centerCountsSize + closestCentersSize + lowerBoundSize + upperBoundSize + sSize + halfcdistSize + newcdistSize;
/* Check memory requirements */
/* Add one to error message to ceil */
@@ -311,7 +215,6 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
/* Allocate space */
/* Use float instead of double to save memory */
agg = palloc(aggSize);
centerCounts = palloc(centerCountsSize);
closestCenters = palloc(closestCentersSize);
lowerBound = palloc_extended(lowerBoundSize, MCXT_ALLOC_HUGE);
@@ -320,25 +223,25 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
halfcdist = palloc_extended(halfcdistSize, MCXT_ALLOC_HUGE);
newcdist = palloc(newcdistSize);
/* Initialize new centers */
newCenters = VectorArrayInit(numCenters, dimensions, centers->itemsize);
newCenters->length = numCenters;
#ifdef IVFFLAT_MEMORY
ShowMemoryUsage(MemoryContextGetParent(CurrentMemoryContext), totalSize);
#endif
newCenters = VectorArrayInit(numCenters, dimensions);
for (j = 0; j < numCenters; j++)
{
vec = VectorArrayGet(newCenters, j);
SET_VARSIZE(vec, VECTOR_SIZE(dimensions));
vec->dim = dimensions;
}
/* Pick initial centers */
InitCenters(index, samples, centers, lowerBound);
/* Assign each x to its closest initial center c(x) = argmin d(x,c) */
for (int64 j = 0; j < numSamples; j++)
for (j = 0; j < numSamples; j++)
{
float minDistance = FLT_MAX;
int closestCenter = 0;
/* Find closest center */
for (int64 k = 0; k < numCenters; k++)
for (k = 0; k < numCenters; k++)
{
/* TODO Use Lemma 1 in k-means++ initialization */
float distance = lowerBound[j * numCenters + k];
@@ -364,13 +267,13 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
CHECK_FOR_INTERRUPTS();
/* Step 1: For all centers, compute distance */
for (int64 j = 0; j < numCenters; j++)
for (j = 0; j < numCenters; j++)
{
Datum vec = PointerGetDatum(VectorArrayGet(centers, j));
vec = VectorArrayGet(centers, j);
for (int64 k = j + 1; k < numCenters; k++)
for (k = j + 1; k < numCenters; k++)
{
float distance = 0.5 * DatumGetFloat8(FunctionCall2Coll(procinfo, collation, vec, PointerGetDatum(VectorArrayGet(centers, k))));
float distance = 0.5 * DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(vec), PointerGetDatum(VectorArrayGet(centers, k))));
halfcdist[j * numCenters + k] = distance;
halfcdist[k * numCenters + j] = distance;
@@ -378,11 +281,11 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
}
/* For all centers c, compute s(c) */
for (int64 j = 0; j < numCenters; j++)
for (j = 0; j < numCenters; j++)
{
float minDistance = FLT_MAX;
for (int64 k = 0; k < numCenters; k++)
for (k = 0; k < numCenters; k++)
{
float distance;
@@ -399,7 +302,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
rjreset = iteration != 0;
for (int64 j = 0; j < numSamples; j++)
for (j = 0; j < numSamples; j++)
{
bool rj;
@@ -409,9 +312,8 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
rj = rjreset;
for (int64 k = 0; k < numCenters; k++)
for (k = 0; k < numCenters; k++)
{
Datum vec;
float dxcx;
/* Step 3: For all remaining points x and centers c */
@@ -424,12 +326,12 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
if (upperBound[j] <= halfcdist[closestCenters[j] * numCenters + k])
continue;
vec = PointerGetDatum(VectorArrayGet(samples, j));
vec = VectorArrayGet(samples, j);
/* Step 3a */
if (rj)
{
dxcx = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, vec, PointerGetDatum(VectorArrayGet(centers, closestCenters[j]))));
dxcx = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(vec), PointerGetDatum(VectorArrayGet(centers, closestCenters[j]))));
/* d(x,c(x)) computed, which is a form of d(x,c) */
lowerBound[j * numCenters + closestCenters[j]] = dxcx;
@@ -443,7 +345,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
/* Step 3b */
if (dxcx > lowerBound[j * numCenters + k] || dxcx > halfcdist[closestCenters[j] * numCenters + k])
{
float dxc = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, vec, PointerGetDatum(VectorArrayGet(centers, k))));
float dxc = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(vec), PointerGetDatum(VectorArrayGet(centers, k))));
/* d(x,c) calculated */
lowerBound[j * numCenters + k] = dxc;
@@ -462,15 +364,66 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
}
/* Step 4: For each center c, let m(c) be mean of all points assigned */
ComputeNewCenters(samples, agg, newCenters, centerCounts, closestCenters, normprocinfo, collation, typeInfo);
for (j = 0; j < numCenters; j++)
{
vec = VectorArrayGet(newCenters, j);
for (k = 0; k < dimensions; k++)
vec->x[k] = 0.0;
centerCounts[j] = 0;
}
for (j = 0; j < numSamples; j++)
{
int closestCenter;
vec = VectorArrayGet(samples, j);
closestCenter = closestCenters[j];
/* Increment sum and count of closest center */
newCenter = VectorArrayGet(newCenters, closestCenter);
for (k = 0; k < dimensions; k++)
newCenter->x[k] += vec->x[k];
centerCounts[closestCenter] += 1;
}
for (j = 0; j < numCenters; j++)
{
vec = VectorArrayGet(newCenters, j);
if (centerCounts[j] > 0)
{
/* Double avoids overflow, but requires more memory */
/* TODO Update bounds */
for (k = 0; k < dimensions; k++)
{
if (isinf(vec->x[k]))
vec->x[k] = vec->x[k] > 0 ? FLT_MAX : -FLT_MAX;
}
for (k = 0; k < dimensions; k++)
vec->x[k] /= centerCounts[j];
}
else
{
/* TODO Handle empty centers properly */
for (k = 0; k < dimensions; k++)
vec->x[k] = RandomDouble();
}
/* Normalize if needed */
if (normprocinfo != NULL)
ApplyNorm(normprocinfo, collation, vec);
}
/* Step 5 */
for (int j = 0; j < numCenters; j++)
for (j = 0; j < numCenters; j++)
newcdist[j] = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(VectorArrayGet(centers, j)), PointerGetDatum(VectorArrayGet(newCenters, j))));
for (int64 j = 0; j < numSamples; j++)
for (j = 0; j < numSamples; j++)
{
for (int64 k = 0; k < numCenters; k++)
for (k = 0; k < numCenters; k++)
{
float distance = lowerBound[j * numCenters + k] - newcdist[k];
@@ -483,78 +436,77 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers, const Ivff
/* Step 6 */
/* We reset r(x) before Step 3 in the next iteration */
for (int j = 0; j < numSamples; j++)
for (j = 0; j < numSamples; j++)
upperBound[j] += newcdist[closestCenters[j]];
/* Step 7 */
for (int j = 0; j < numCenters; j++)
for (j = 0; j < numCenters; j++)
VectorArraySet(centers, j, VectorArrayGet(newCenters, j));
if (changes == 0 && iteration != 0)
break;
}
}
/*
* Ensure no NaN or infinite values
*/
static void
CheckElements(VectorArray centers, const IvfflatTypeInfo * typeInfo)
{
float *scratch = palloc(sizeof(float) * centers->dim);
for (int i = 0; i < centers->length; i++)
{
for (int j = 0; j < centers->dim; j++)
scratch[j] = 0;
/* /fp:fast may not propagate NaN with MSVC, but that's alright */
typeInfo->sumCenter(VectorArrayGet(centers, i), scratch);
for (int j = 0; j < centers->dim; j++)
{
if (isnan(scratch[j]))
elog(ERROR, "NaN detected. Please report a bug.");
if (isinf(scratch[j]))
elog(ERROR, "Infinite value detected. Please report a bug.");
}
}
}
/*
* Ensure no zero vectors for cosine distance
*/
static void
CheckNorms(VectorArray centers, Relation index)
{
/* Check NORM_PROC instead of KMEANS_NORM_PROC */
FmgrInfo *normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
Oid collation = index->rd_indcollation[0];
if (normprocinfo == NULL)
return;
for (int i = 0; i < centers->length; i++)
{
double norm = DatumGetFloat8(FunctionCall1Coll(normprocinfo, collation, PointerGetDatum(VectorArrayGet(centers, i))));
if (norm == 0)
elog(ERROR, "Zero norm detected. Please report a bug.");
}
VectorArrayFree(newCenters);
pfree(centerCounts);
pfree(closestCenters);
pfree(lowerBound);
pfree(upperBound);
pfree(s);
pfree(halfcdist);
pfree(newcdist);
}
/*
* Detect issues with centers
*/
static void
CheckCenters(Relation index, VectorArray centers, const IvfflatTypeInfo * typeInfo)
CheckCenters(Relation index, VectorArray centers)
{
FmgrInfo *normprocinfo;
if (centers->length != centers->maxlen)
elog(ERROR, "Not enough centers. Please report a bug.");
CheckElements(centers, typeInfo);
CheckNorms(centers, index);
/* Ensure no NaN or infinite values */
for (int i = 0; i < centers->length; i++)
{
Vector *vec = VectorArrayGet(centers, i);
for (int j = 0; j < vec->dim; j++)
{
if (isnan(vec->x[j]))
elog(ERROR, "NaN detected. Please report a bug.");
if (isinf(vec->x[j]))
elog(ERROR, "Infinite value detected. Please report a bug.");
}
}
/* Ensure no duplicate centers */
/* Fine to sort in-place */
qsort(centers->items, centers->length, VECTOR_SIZE(centers->dim), CompareVectors);
for (int i = 1; i < centers->length; i++)
{
if (CompareVectors(VectorArrayGet(centers, i), VectorArrayGet(centers, i - 1)) == 0)
elog(ERROR, "Duplicate centers detected. Please report a bug.");
}
/* Ensure no zero vectors for cosine distance */
/* Check NORM_PROC instead of KMEANS_NORM_PROC */
normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
if (normprocinfo != NULL)
{
Oid collation = index->rd_indcollation[0];
for (int i = 0; i < centers->length; i++)
{
double norm = DatumGetFloat8(FunctionCall1Coll(normprocinfo, collation, PointerGetDatum(VectorArrayGet(centers, i))));
if (norm == 0)
elog(ERROR, "Zero norm detected. Please report a bug.");
}
}
}
/*
@@ -562,20 +514,12 @@ CheckCenters(Relation index, VectorArray centers, const IvfflatTypeInfo * typeIn
* We use spherical k-means for inner product and cosine
*/
void
IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers, const IvfflatTypeInfo * typeInfo)
IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers)
{
MemoryContext kmeansCtx = AllocSetContextCreate(CurrentMemoryContext,
"Ivfflat kmeans temporary context",
ALLOCSET_DEFAULT_SIZES);
MemoryContext oldCtx = MemoryContextSwitchTo(kmeansCtx);
if (samples->length == 0)
RandomCenters(index, centers, typeInfo);
if (samples->length <= centers->maxlen)
QuickCenters(index, samples, centers);
else
ElkanKmeans(index, samples, centers, typeInfo);
ElkanKmeans(index, samples, centers);
CheckCenters(index, centers, typeInfo);
MemoryContextSwitchTo(oldCtx);
MemoryContextDelete(kmeansCtx);
CheckCenters(index, centers);
}

View File

@@ -2,29 +2,13 @@
#include <float.h>
#include "access/genam.h"
#include "access/itup.h"
#include "access/relscan.h"
#include "access/tupdesc.h"
#include "catalog/pg_operator_d.h"
#include "catalog/pg_type_d.h"
#include "fmgr.h"
#include "lib/pairingheap.h"
#include "ivfflat.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "storage/bufmgr.h"
#include "utils/memutils.h"
#include "utils/rel.h"
#include "utils/snapmgr.h"
#include "utils/tuplesort.h"
#if PG_VERSION_NUM >= 160000
#include "varatt.h"
#endif
#define GetScanList(ptr) pairingheap_container(IvfflatScanList, ph_node, ptr)
#define GetScanListConst(ptr) pairingheap_const_container(IvfflatScanList, ph_node, ptr)
/*
* Compare list distances
@@ -32,10 +16,10 @@
static int
CompareLists(const pairingheap_node *a, const pairingheap_node *b, void *arg)
{
if (GetScanListConst(a)->distance > GetScanListConst(b)->distance)
if (((const IvfflatScanList *) a)->distance > ((const IvfflatScanList *) b)->distance)
return 1;
if (GetScanListConst(a)->distance < GetScanListConst(b)->distance)
if (((const IvfflatScanList *) a)->distance < ((const IvfflatScanList *) b)->distance)
return -1;
return 0;
@@ -71,9 +55,9 @@ GetScanLists(IndexScanDesc scan, Datum value)
double distance;
/* Use procinfo from the index instead of scan key for performance */
distance = DatumGetFloat8(so->distfunc(so->procinfo, so->collation, PointerGetDatum(&list->center), value));
distance = DatumGetFloat8(FunctionCall2Coll(so->procinfo, so->collation, PointerGetDatum(&list->center), value));
if (listCount < so->maxProbes)
if (listCount < so->probes)
{
IvfflatScanList *scanlist;
@@ -86,15 +70,15 @@ GetScanLists(IndexScanDesc scan, Datum value)
pairingheap_add(so->listQueue, &scanlist->ph_node);
/* Calculate max distance */
if (listCount == so->maxProbes)
maxDistance = GetScanList(pairingheap_first(so->listQueue))->distance;
if (listCount == so->probes)
maxDistance = ((IvfflatScanList *) pairingheap_first(so->listQueue))->distance;
}
else if (distance < maxDistance)
{
IvfflatScanList *scanlist;
/* Remove */
scanlist = GetScanList(pairingheap_remove_first(so->listQueue));
scanlist = (IvfflatScanList *) pairingheap_remove_first(so->listQueue);
/* Reuse */
scanlist->startPage = list->startPage;
@@ -102,7 +86,7 @@ GetScanLists(IndexScanDesc scan, Datum value)
pairingheap_add(so->listQueue, &scanlist->ph_node);
/* Update max distance */
maxDistance = GetScanList(pairingheap_first(so->listQueue))->distance;
maxDistance = ((IvfflatScanList *) pairingheap_first(so->listQueue))->distance;
}
}
@@ -110,11 +94,6 @@ GetScanLists(IndexScanDesc scan, Datum value)
UnlockReleaseBuffer(cbuf);
}
for (int i = listCount - 1; i >= 0; i--)
so->listPages[i] = GetScanList(pairingheap_remove_first(so->listQueue))->startPage;
Assert(pairingheap_is_empty(so->listQueue));
}
/*
@@ -125,15 +104,25 @@ GetScanItems(IndexScanDesc scan, Datum value)
{
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
TupleTableSlot *slot = so->vslot;
int batchProbes = 0;
double tuples = 0;
tuplesort_reset(so->sortstate);
#if PG_VERSION_NUM >= 120000
TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
#else
TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc);
#endif
/*
* Reuse same set of shared buffers for scan
*
* See postgres/src/backend/storage/buffer/README for description
*/
BufferAccessStrategy bas = GetAccessStrategy(BAS_BULKREAD);
/* Search closest probes lists */
while (so->listIndex < so->maxProbes && (++batchProbes) <= so->probes)
while (!pairingheap_is_empty(so->listQueue))
{
BlockNumber searchPage = so->listPages[so->listIndex++];
BlockNumber searchPage = ((IvfflatScanList *) pairingheap_remove_first(so->listQueue))->startPage;
/* Search all entry pages for list */
while (BlockNumberIsValid(searchPage))
@@ -142,7 +131,7 @@ GetScanItems(IndexScanDesc scan, Datum value)
Page page;
OffsetNumber maxoffno;
buf = ReadBufferExtended(scan->indexRelation, MAIN_FORKNUM, searchPage, RBM_NORMAL, so->bas);
buf = ReadBufferExtended(scan->indexRelation, MAIN_FORKNUM, searchPage, RBM_NORMAL, bas);
LockBuffer(buf, BUFFER_LOCK_SHARE);
page = BufferGetPage(buf);
maxoffno = PageGetMaxOffsetNumber(page);
@@ -164,13 +153,15 @@ GetScanItems(IndexScanDesc scan, Datum value)
* performance
*/
ExecClearTuple(slot);
slot->tts_values[0] = so->distfunc(so->procinfo, so->collation, datum, value);
slot->tts_values[0] = FunctionCall2Coll(so->procinfo, so->collation, datum, value);
slot->tts_isnull[0] = false;
slot->tts_values[1] = PointerGetDatum(&itup->t_tid);
slot->tts_isnull[1] = false;
ExecStoreVirtualTuple(slot);
tuplesort_puttupleslot(so->sortstate, slot);
tuples++;
}
searchPage = IvfflatPageGetOpaque(page)->nextblkno;
@@ -179,71 +170,15 @@ GetScanItems(IndexScanDesc scan, Datum value)
}
}
FreeAccessStrategy(bas);
if (tuples < 100)
ereport(DEBUG1,
(errmsg("index scan found few tuples"),
errdetail("Index may have been created with little data."),
errhint("Recreate the index and possibly decrease lists.")));
tuplesort_performsort(so->sortstate);
#if defined(IVFFLAT_MEMORY)
elog(INFO, "memory: %zu MB", MemoryContextMemAllocated(CurrentMemoryContext, true) / (1024 * 1024));
#endif
}
/*
* Zero distance
*/
static Datum
ZeroDistance(FmgrInfo *flinfo, Oid collation, Datum arg1, Datum arg2)
{
return Float8GetDatum(0.0);
}
/*
* Get scan value
*/
static Datum
GetScanValue(IndexScanDesc scan)
{
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
Datum value;
if (scan->orderByData->sk_flags & SK_ISNULL)
{
value = PointerGetDatum(NULL);
so->distfunc = ZeroDistance;
}
else
{
value = scan->orderByData->sk_argument;
so->distfunc = FunctionCall2Coll;
/* Value should not be compressed or toasted */
Assert(!VARATT_IS_COMPRESSED(DatumGetPointer(value)));
Assert(!VARATT_IS_EXTENDED(DatumGetPointer(value)));
/* Normalize if needed */
if (so->normprocinfo != NULL)
{
MemoryContext oldCtx = MemoryContextSwitchTo(so->tmpCtx);
value = IvfflatNormValue(so->typeInfo, so->collation, value);
MemoryContextSwitchTo(oldCtx);
}
}
return value;
}
/*
* Initialize scan sort state
*/
static Tuplesortstate *
InitScanSortState(TupleDesc tupdesc)
{
AttrNumber attNums[] = {1};
Oid sortOperators[] = {Float8LessOperator};
Oid sortCollations[] = {InvalidOid};
bool nullsFirstFlags[] = {false};
return tuplesort_begin_heap(tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
}
/*
@@ -256,31 +191,23 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
IvfflatScanOpaque so;
int lists;
int dimensions;
AttrNumber attNums[] = {1};
Oid sortOperators[] = {Float8LessOperator};
Oid sortCollations[] = {InvalidOid};
bool nullsFirstFlags[] = {false};
int probes = ivfflat_probes;
int maxProbes;
MemoryContext oldCtx;
scan = RelationGetIndexScan(index, nkeys, norderbys);
/* Get lists and dimensions from metapage */
IvfflatGetMetaPageInfo(index, &lists, &dimensions);
if (ivfflat_iterative_scan != IVFFLAT_ITERATIVE_SCAN_OFF)
maxProbes = Max(ivfflat_max_probes, probes);
else
maxProbes = probes;
if (probes > lists)
probes = lists;
if (maxProbes > lists)
maxProbes = lists;
so = (IvfflatScanOpaque) palloc(sizeof(IvfflatScanOpaqueData));
so->typeInfo = IvfflatGetTypeInfo(index);
so = (IvfflatScanOpaque) palloc(offsetof(IvfflatScanOpaqueData, lists) + probes * sizeof(IvfflatScanList));
so->first = true;
so->probes = probes;
so->maxProbes = maxProbes;
so->dimensions = dimensions;
/* Set support functions */
@@ -288,40 +215,25 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
so->normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
so->collation = index->rd_indcollation[0];
so->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
"Ivfflat scan temporary context",
ALLOCSET_DEFAULT_SIZES);
oldCtx = MemoryContextSwitchTo(so->tmpCtx);
/* Create tuple description for sorting */
#if PG_VERSION_NUM >= 120000
so->tupdesc = CreateTemplateTupleDesc(2);
#else
so->tupdesc = CreateTemplateTupleDesc(2, false);
#endif
TupleDescInitEntry(so->tupdesc, (AttrNumber) 1, "distance", FLOAT8OID, -1, 0);
TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "heaptid", TIDOID, -1, 0);
#if PG_VERSION_NUM >= 190000
TupleDescFinalize(so->tupdesc);
#endif
/* Prep sort */
so->sortstate = InitScanSortState(so->tupdesc);
so->sortstate = tuplesort_begin_heap(so->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
/* Need separate slots for puttuple and gettuple */
so->vslot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
so->mslot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsMinimalTuple);
/*
* Reuse same set of shared buffers for scan
*
* See postgres/src/backend/storage/buffer/README for description
*/
so->bas = GetAccessStrategy(BAS_BULKREAD);
#if PG_VERSION_NUM >= 120000
so->slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsMinimalTuple);
#else
so->slot = MakeSingleTupleTableSlot(so->tupdesc);
#endif
so->listQueue = pairingheap_allocate(CompareLists, scan);
so->listPages = palloc(maxProbes * sizeof(BlockNumber));
so->listIndex = 0;
so->lists = palloc(maxProbes * sizeof(IvfflatScanList));
MemoryContextSwitchTo(oldCtx);
scan->opaque = so;
@@ -336,9 +248,13 @@ ivfflatrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int
{
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
#if PG_VERSION_NUM >= 130000
if (!so->first)
tuplesort_reset(so->sortstate);
#endif
so->first = true;
pairingheap_reset(so->listQueue);
so->listIndex = 0;
if (keys && scan->numberOfKeys > 0)
memmove(scan->keyData, keys, scan->numberOfKeys * sizeof(ScanKeyData));
@@ -354,8 +270,6 @@ bool
ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
{
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
ItemPointer heaptid;
bool isnull;
/*
* Index can be used to scan backward, but Postgres doesn't support
@@ -369,10 +283,6 @@ 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)
@@ -383,27 +293,45 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
if (!IsMVCCSnapshot(scan->xs_snapshot))
elog(ERROR, "non-MVCC snapshots are not supported with ivfflat");
value = GetScanValue(scan);
if (scan->orderByData->sk_flags & SK_ISNULL)
value = PointerGetDatum(InitVector(so->dimensions));
else
{
value = scan->orderByData->sk_argument;
/* Value should not be compressed or toasted */
Assert(!VARATT_IS_COMPRESSED(DatumGetPointer(value)));
Assert(!VARATT_IS_EXTENDED(DatumGetPointer(value)));
/* Fine if normalization fails */
if (so->normprocinfo != NULL)
IvfflatNormValue(so->normprocinfo, so->collation, &value, NULL);
}
IvfflatBench("GetScanLists", GetScanLists(scan, value));
IvfflatBench("GetScanItems", GetScanItems(scan, value));
so->first = false;
so->value = value;
/* Clean up if we allocated a new value */
if (value != scan->orderByData->sk_argument)
pfree(DatumGetPointer(value));
}
while (!tuplesort_gettupleslot(so->sortstate, true, false, so->mslot, NULL))
if (tuplesort_gettupleslot(so->sortstate, true, false, so->slot, NULL))
{
if (so->listIndex == so->maxProbes)
return false;
ItemPointer heaptid = (ItemPointer) DatumGetPointer(slot_getattr(so->slot, 2, &so->isnull));
IvfflatBench("GetScanItems", GetScanItems(scan, so->value));
#if PG_VERSION_NUM >= 120000
scan->xs_heaptid = *heaptid;
#else
scan->xs_ctup.t_self = *heaptid;
#endif
scan->xs_recheckorderby = false;
return true;
}
heaptid = (ItemPointer) DatumGetPointer(slot_getattr(so->mslot, 2, &isnull));
scan->xs_heaptid = *heaptid;
scan->xs_recheck = false;
scan->xs_recheckorderby = false;
return true;
return false;
}
/*
@@ -414,11 +342,9 @@ ivfflatendscan(IndexScanDesc scan)
{
IvfflatScanOpaque so = (IvfflatScanOpaque) scan->opaque;
/* Free any temporary files */
pairingheap_free(so->listQueue);
tuplesort_end(so->sortstate);
MemoryContextDelete(so->tmpCtx);
pfree(so);
scan->opaque = NULL;
}

View File

@@ -1,36 +1,21 @@
#include "postgres.h"
#include "access/genam.h"
#include "access/generic_xlog.h"
#include "fmgr.h"
#include "halfutils.h"
#include "halfvec.h"
#include "ivfflat.h"
#include "storage/bufmgr.h"
#include "utils/relcache.h"
#include "utils/varbit.h"
#include "vector.h"
#if PG_VERSION_NUM >= 160000
#include "varatt.h"
#endif
/*
* Allocate a vector array
*/
VectorArray
VectorArrayInit(int maxlen, int dimensions, Size itemsize)
VectorArrayInit(int maxlen, int dimensions)
{
VectorArray res = palloc(sizeof(VectorArrayData));
/* Ensure items are aligned to prevent UB */
itemsize = MAXALIGN(itemsize);
res->length = 0;
res->maxlen = maxlen;
res->dim = dimensions;
res->itemsize = itemsize;
res->items = palloc_extended(maxlen * itemsize, MCXT_ALLOC_ZERO | MCXT_ALLOC_HUGE);
res->items = palloc_extended(maxlen * VECTOR_SIZE(dimensions), MCXT_ALLOC_ZERO | MCXT_ALLOC_HUGE);
return res;
}
@@ -44,6 +29,16 @@ VectorArrayFree(VectorArray arr)
pfree(arr);
}
/*
* Print vector array - useful for debugging
*/
void
PrintVectorArray(char *msg, VectorArray arr)
{
for (int i = 0; i < arr->length; i++)
PrintVector(msg, VectorArrayGet(arr, i));
}
/*
* Get the number of lists in the index
*/
@@ -71,21 +66,34 @@ IvfflatOptionalProcInfo(Relation index, uint16 procnum)
}
/*
* Normalize value
*/
Datum
IvfflatNormValue(const IvfflatTypeInfo * typeInfo, Oid collation, Datum value)
{
return DirectFunctionCall1Coll(typeInfo->normalize, collation, value);
}
/*
* Check if non-zero norm
* Divide by the norm
*
* Returns false if value should not be indexed
*
* The caller needs to free the pointer stored in value
* if it's different than the original value
*/
bool
IvfflatCheckNorm(FmgrInfo *procinfo, Oid collation, Datum value)
IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
{
return DatumGetFloat8(FunctionCall1Coll(procinfo, collation, value)) > 0;
double norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
if (norm > 0)
{
Vector *v = DatumGetVector(*value);
if (result == NULL)
result = InitVector(v->dim);
for (int i = 0; i < v->dim; i++)
result->x[i] = v->x[i] / norm;
*value = PointerGetDatum(result);
return true;
}
return false;
}
/*
@@ -176,11 +184,7 @@ IvfflatGetMetaPageInfo(Relation index, int *lists, int *dimensions)
page = BufferGetPage(buf);
metap = IvfflatPageGetMeta(page);
if (unlikely(metap->magicNumber != IVFFLAT_MAGIC_NUMBER))
elog(ERROR, "ivfflat index is not valid");
if (lists != NULL)
*lists = metap->lists;
*lists = metap->lists;
if (dimensions != NULL)
*dimensions = metap->dimensions;
@@ -234,150 +238,3 @@ IvfflatUpdateList(Relation index, ListInfo listInfo,
UnlockReleaseBuffer(buf);
}
}
PGDLLEXPORT Datum l2_normalize(PG_FUNCTION_ARGS);
PGDLLEXPORT Datum halfvec_l2_normalize(PG_FUNCTION_ARGS);
PGDLLEXPORT Datum sparsevec_l2_normalize(PG_FUNCTION_ARGS);
static Size
VectorItemSize(int dimensions)
{
return VECTOR_SIZE(dimensions);
}
static Size
HalfvecItemSize(int dimensions)
{
return HALFVEC_SIZE(dimensions);
}
static Size
BitItemSize(int dimensions)
{
return VARBITTOTALLEN(dimensions);
}
static void
VectorUpdateCenter(Pointer v, int dimensions, float *x)
{
Vector *vec = (Vector *) v;
SET_VARSIZE(vec, VECTOR_SIZE(dimensions));
vec->dim = dimensions;
for (int i = 0; i < dimensions; i++)
vec->x[i] = x[i];
}
static void
HalfvecUpdateCenter(Pointer v, int dimensions, float *x)
{
HalfVector *vec = (HalfVector *) v;
SET_VARSIZE(vec, HALFVEC_SIZE(dimensions));
vec->dim = dimensions;
for (int i = 0; i < dimensions; i++)
vec->x[i] = Float4ToHalfUnchecked(x[i]);
}
static void
BitUpdateCenter(Pointer v, int dimensions, float *x)
{
VarBit *vec = (VarBit *) v;
unsigned char *nx = VARBITS(vec);
SET_VARSIZE(vec, VARBITTOTALLEN(dimensions));
VARBITLEN(vec) = dimensions;
for (uint32 i = 0; i < VARBITBYTES(vec); i++)
nx[i] = 0;
for (int i = 0; i < dimensions; i++)
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;
/* Auto-vectorized */
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;
/* Auto-vectorized on aarch64 */
for (int i = 0; i < dim; i++)
x[i] += HalfToFloat4(vec->x[i]);
}
static void
BitSumCenter(Pointer v, float *x)
{
VarBit *vec = (VarBit *) v;
for (int i = 0; i < VARBITLEN(vec); i++)
x[i] += (float) (((VARBITS(vec)[i / 8]) >> (7 - (i % 8))) & 0x01);
}
/*
* Get type info
*/
const IvfflatTypeInfo *
IvfflatGetTypeInfo(Relation index)
{
FmgrInfo *procinfo = IvfflatOptionalProcInfo(index, IVFFLAT_TYPE_INFO_PROC);
if (procinfo == NULL)
{
static const IvfflatTypeInfo typeInfo = {
.maxDimensions = IVFFLAT_MAX_DIM,
.normalize = l2_normalize,
.itemSize = VectorItemSize,
.updateCenter = VectorUpdateCenter,
.sumCenter = VectorSumCenter
};
return (&typeInfo);
}
else
return (const IvfflatTypeInfo *) DatumGetPointer(FunctionCall0Coll(procinfo, InvalidOid));
}
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflat_halfvec_support);
Datum
ivfflat_halfvec_support(PG_FUNCTION_ARGS)
{
static const IvfflatTypeInfo typeInfo = {
.maxDimensions = IVFFLAT_MAX_DIM * 2,
.normalize = halfvec_l2_normalize,
.itemSize = HalfvecItemSize,
.updateCenter = HalfvecUpdateCenter,
.sumCenter = HalfvecSumCenter
};
PG_RETURN_POINTER(&typeInfo);
}
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflat_bit_support);
Datum
ivfflat_bit_support(PG_FUNCTION_ARGS)
{
static const IvfflatTypeInfo typeInfo = {
.maxDimensions = IVFFLAT_MAX_DIM * 32,
.normalize = NULL,
.itemSize = BitItemSize,
.updateCenter = BitUpdateCenter,
.sumCenter = BitSumCenter
};
PG_RETURN_POINTER(&typeInfo);
}

View File

@@ -1,16 +1,8 @@
#include "postgres.h"
#include "access/genam.h"
#include "access/generic_xlog.h"
#include "access/itup.h"
#include "commands/vacuum.h"
#include "ivfflat.h"
#include "storage/bufmgr.h"
#include "utils/relcache.h"
#if PG_VERSION_NUM >= 180000
#define vacuum_delay_point() vacuum_delay_point(false)
#endif
/*
* Bulk delete tuples from the index
@@ -33,7 +25,7 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
Page cpage;
OffsetNumber coffno;
OffsetNumber cmaxoffno;
BlockNumber listPages[MaxOffsetNumber];
BlockNumber startPages[MaxOffsetNumber];
ListInfo listInfo;
cbuf = ReadBuffer(index, blkno);
@@ -47,7 +39,7 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
{
IvfflatList list = (IvfflatList) PageGetItem(cpage, PageGetItemId(cpage, coffno));
listPages[coffno - FirstOffsetNumber] = list->startPage;
startPages[coffno - FirstOffsetNumber] = list->startPage;
}
listInfo.blkno = blkno;
@@ -57,7 +49,7 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
for (coffno = FirstOffsetNumber; coffno <= cmaxoffno; coffno = OffsetNumberNext(coffno))
{
BlockNumber searchPage = listPages[coffno - FirstOffsetNumber];
BlockNumber searchPage = startPages[coffno - FirstOffsetNumber];
BlockNumber insertPage = InvalidBlockNumber;
/* Iterate over entry pages */

File diff suppressed because it is too large Load Diff

View File

@@ -1,40 +0,0 @@
#ifndef SPARSEVEC_H
#define SPARSEVEC_H
#define SPARSEVEC_MAX_DIM 1000000000
#define SPARSEVEC_MAX_NNZ 16000
#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)
/*
* Indices use 0-based numbering for the on-disk (and binary) format (consistent with C)
* and are always sorted. Values come after indices.
*/
typedef struct SparseVector
{
int32 vl_len_; /* varlena header (do not touch directly!) */
int32 dim; /* number of dimensions */
int32 nnz; /* number of non-zero elements */
int32 unused; /* reserved for future use, always zero */
int32 indices[FLEXIBLE_ARRAY_MEMBER];
} SparseVector;
/* Use functions instead of macros to avoid double evaluation */
static inline Size
SPARSEVEC_SIZE(int nnz)
{
return offsetof(SparseVector, indices) + (nnz * sizeof(int32)) + (nnz * sizeof(float));
}
static inline float *
SPARSEVEC_VALUES(SparseVector * x)
{
return (float *) (((char *) x) + offsetof(SparseVector, indices) + (x->nnz * sizeof(int32)));
}
SparseVector *InitSparseVector(int dim, int nnz);
#endif

File diff suppressed because it is too large Load Diff

View File

@@ -12,7 +12,7 @@ typedef struct Vector
{
int32 vl_len_; /* varlena header (do not touch directly!) */
int16 dim; /* number of dimensions */
int16 unused; /* reserved for future use, always zero */
int16 unused;
float x[FLEXIBLE_ARRAY_MEMBER];
} Vector;
@@ -20,11 +20,4 @@ Vector *InitVector(int dim);
void PrintVector(char *msg, Vector * vector);
int vector_cmp_internal(Vector * a, Vector * b);
/* TODO Move to better place */
#if PG_VERSION_NUM >= 160000
#define FUNCTION_PREFIX
#else
#define FUNCTION_PREFIX PGDLLEXPORT
#endif
#endif

View File

@@ -1,140 +0,0 @@
SELECT hamming_distance('111', '111');
hamming_distance
------------------
0
(1 row)
SELECT hamming_distance('111', '110');
hamming_distance
------------------
1
(1 row)
SELECT hamming_distance('111', '100');
hamming_distance
------------------
2
(1 row)
SELECT hamming_distance('111', '000');
hamming_distance
------------------
3
(1 row)
SELECT hamming_distance('10101010101010101010', '01010101010101010101');
hamming_distance
------------------
20
(1 row)
SELECT hamming_distance('101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101', '101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101');
hamming_distance
------------------
0
(1 row)
SELECT hamming_distance('101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101', '010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010');
hamming_distance
------------------
513
(1 row)
SELECT hamming_distance('110000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000011', '100000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000001');
hamming_distance
------------------
2
(1 row)
SELECT hamming_distance('', '');
hamming_distance
------------------
0
(1 row)
SELECT hamming_distance('111', '00');
ERROR: different bit lengths 3 and 2
SELECT hamming_distance('111', '000'::varbit(4));
hamming_distance
------------------
3
(1 row)
SELECT hamming_distance('111', '0000'::varbit(4));
ERROR: different bit lengths 3 and 4
SELECT jaccard_distance('1111', '1111');
jaccard_distance
------------------
0
(1 row)
SELECT jaccard_distance('1111', '1110');
jaccard_distance
------------------
0.25
(1 row)
SELECT jaccard_distance('1111', '1100');
jaccard_distance
------------------
0.5
(1 row)
SELECT jaccard_distance('1111', '1000');
jaccard_distance
------------------
0.75
(1 row)
SELECT jaccard_distance('1111', '0000');
jaccard_distance
------------------
1
(1 row)
SELECT jaccard_distance('1100', '1000');
jaccard_distance
------------------
0.5
(1 row)
SELECT jaccard_distance('10101010101010101010', '01010101010101010101');
jaccard_distance
------------------
1
(1 row)
SELECT jaccard_distance('101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101', '101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101');
jaccard_distance
------------------
0
(1 row)
SELECT jaccard_distance('101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101', '010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010');
jaccard_distance
------------------
1
(1 row)
SELECT jaccard_distance('110000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000011', '100000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000001');
jaccard_distance
------------------
0.5
(1 row)
SELECT jaccard_distance('', '');
jaccard_distance
------------------
1
(1 row)
SELECT jaccard_distance('1111', '000');
ERROR: different bit lengths 4 and 3
SELECT jaccard_distance('1111', '0000'::varbit(5));
jaccard_distance
------------------
1
(1 row)
SELECT jaccard_distance('1111', '00000'::varbit(5));
ERROR: different bit lengths 4 and 5

View File

@@ -1,5 +1,4 @@
SET enable_seqscan = off;
-- vector
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t (val);
@@ -9,53 +8,10 @@ SELECT * FROM t WHERE val = '[1,2,3]';
[1,2,3]
(1 row)
SELECT * FROM t ORDER BY val;
SELECT * FROM t ORDER BY val LIMIT 1;
val
---------
[0,0,0]
[1,1,1]
[1,2,3]
(4 rows)
DROP TABLE t;
-- halfvec
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t (val);
SELECT * FROM t WHERE val = '[1,2,3]';
val
---------
[1,2,3]
(1 row)
SELECT * FROM t ORDER BY val;
val
---------
[0,0,0]
[1,1,1]
[1,2,3]
(4 rows)
DROP TABLE t;
-- sparsevec
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE INDEX ON t (val);
SELECT * FROM t WHERE val = '{1:1,2:2,3:3}/3';
val
-----------------
{1:1,2:2,3:3}/3
(1 row)
SELECT * FROM t ORDER BY val;
val
-----------------
{}/3
{1:1,2:1,3:1}/3
{1:1,2:2,3:3}/3
(4 rows)
DROP TABLE t;

View File

@@ -28,26 +28,6 @@ SELECT ARRAY[1,2,3]::numeric[]::vector;
[1,2,3]
(1 row)
SELECT '[1,2,3]'::vector::real[];
float4
---------
{1,2,3}
(1 row)
SELECT '{1,2,3}'::real[]::vector;
vector
---------
[1,2,3]
(1 row)
SELECT '{1,2,3}'::real[]::vector(3);
vector
---------
[1,2,3]
(1 row)
SELECT '{1,2,3}'::real[]::vector(2);
ERROR: expected 2 dimensions, not 3
SELECT '{NULL}'::real[]::vector;
ERROR: array must not contain nulls
SELECT '{NaN}'::real[]::vector;
@@ -60,210 +40,12 @@ SELECT '{}'::real[]::vector;
ERROR: vector must have at least 1 dimension
SELECT '{{1}}'::real[]::vector;
ERROR: array must be 1-D
SELECT '{1,2,3}'::double precision[]::vector;
vector
SELECT '[1,2,3]'::vector::real[];
float4
---------
[1,2,3]
{1,2,3}
(1 row)
SELECT '{1,2,3}'::double precision[]::vector(3);
vector
---------
[1,2,3]
(1 row)
SELECT '{1,2,3}'::double precision[]::vector(2);
ERROR: expected 2 dimensions, not 3
SELECT '{4e38,-4e38}'::double precision[]::vector;
ERROR: infinite value not allowed in vector
SELECT '{1e-46,-1e-46}'::double precision[]::vector;
vector
--------
[0,-0]
(1 row)
SELECT '[1,2,3]'::vector::halfvec;
halfvec
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::vector::halfvec(3);
halfvec
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::vector::halfvec(2);
ERROR: expected 2 dimensions, not 3
SELECT '[65520]'::vector::halfvec;
ERROR: "65520" is out of range for type halfvec
SELECT '[1e-8]'::vector::halfvec;
halfvec
---------
[0]
(1 row)
SELECT '[1,2,3]'::halfvec::vector;
vector
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::halfvec::vector(3);
vector
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::halfvec::vector(2);
ERROR: expected 2 dimensions, not 3
SELECT '{1,2,3}'::real[]::halfvec;
halfvec
---------
[1,2,3]
(1 row)
SELECT '{1,2,3}'::real[]::halfvec(3);
halfvec
---------
[1,2,3]
(1 row)
SELECT '{1,2,3}'::real[]::halfvec(2);
ERROR: expected 2 dimensions, not 3
SELECT '{65520,-65520}'::real[]::halfvec;
ERROR: "65520" is out of range for type halfvec
SELECT '{1e-8,-1e-8}'::real[]::halfvec;
halfvec
---------
[0,-0]
(1 row)
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
sparsevec
-----------------
{2:1.5,4:3.5}/5
(1 row)
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec(5);
sparsevec
-----------------
{2:1.5,4:3.5}/5
(1 row)
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec(4);
ERROR: expected 4 dimensions, not 5
SELECT '{2:1.5,4:3.5}/5'::sparsevec::vector;
vector
-----------------
[0,1.5,0,3.5,0]
(1 row)
SELECT '{2:1.5,4:3.5}/5'::sparsevec::vector(5);
vector
-----------------
[0,1.5,0,3.5,0]
(1 row)
SELECT '{2:1.5,4:3.5}/5'::sparsevec::vector(4);
ERROR: expected 4 dimensions, not 5
SELECT '{}/16001'::sparsevec::vector;
ERROR: vector cannot have more than 16000 dimensions
SELECT '[0,1.5,0,3.5,0]'::halfvec::sparsevec;
sparsevec
-----------------
{2:1.5,4:3.5}/5
(1 row)
SELECT '[0,1.5,0,3.5,0]'::halfvec::sparsevec(5);
sparsevec
-----------------
{2:1.5,4:3.5}/5
(1 row)
SELECT '[0,1.5,0,3.5,0]'::halfvec::sparsevec(4);
ERROR: expected 4 dimensions, not 5
SELECT '{2:1.5,4:3.5}/5'::sparsevec::halfvec;
halfvec
-----------------
[0,1.5,0,3.5,0]
(1 row)
SELECT '{2:1.5,4:3.5}/5'::sparsevec::halfvec(5);
halfvec
-----------------
[0,1.5,0,3.5,0]
(1 row)
SELECT '{2:1.5,4:3.5}/5'::sparsevec::halfvec(4);
ERROR: expected 4 dimensions, not 5
SELECT '{}/16001'::sparsevec::halfvec;
ERROR: halfvec cannot have more than 16000 dimensions
SELECT '{1:65520}/1'::sparsevec::halfvec;
ERROR: "65520" is out of range for type halfvec
SELECT '{1:1e-8}/1'::sparsevec::halfvec;
halfvec
---------
[0]
(1 row)
SELECT ARRAY[1,0,2,0,3,0]::sparsevec;
array
-----------------
{1:1,3:2,5:3}/6
(1 row)
SELECT ARRAY[1.0,0.0,2.0,0.0,3.0,0.0]::sparsevec;
array
-----------------
{1:1,3:2,5:3}/6
(1 row)
SELECT ARRAY[1,0,2,0,3,0]::float4[]::sparsevec;
array
-----------------
{1:1,3:2,5:3}/6
(1 row)
SELECT ARRAY[1,0,2,0,3,0]::float8[]::sparsevec;
array
-----------------
{1:1,3:2,5:3}/6
(1 row)
SELECT ARRAY[1,0,2,0,3,0]::numeric[]::sparsevec;
array
-----------------
{1:1,3:2,5:3}/6
(1 row)
SELECT '{1,0,2,0,3,0}'::real[]::sparsevec;
sparsevec
-----------------
{1:1,3:2,5:3}/6
(1 row)
SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(6);
sparsevec
-----------------
{1:1,3:2,5:3}/6
(1 row)
SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(5);
ERROR: expected 5 dimensions, not 6
SELECT '{NULL}'::real[]::sparsevec;
ERROR: array must not contain nulls
SELECT '{NaN}'::real[]::sparsevec;
ERROR: NaN not allowed in sparsevec
SELECT '{Infinity}'::real[]::sparsevec;
ERROR: infinite value not allowed in sparsevec
SELECT '{-Infinity}'::real[]::sparsevec;
ERROR: infinite value not allowed in sparsevec
SELECT '{}'::real[]::sparsevec;
ERROR: sparsevec must have at least 1 dimension
SELECT '{{1}}'::real[]::sparsevec;
ERROR: array must be 1-D
SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
ERROR: vector cannot have more than 16000 dimensions
SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;

View File

@@ -1,9 +1,8 @@
-- vector
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE TABLE t2 (val vector(3));
\copy t TO 'results/vector.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/vector.bin' WITH (FORMAT binary)
\copy t TO 'results/data.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/data.bin' WITH (FORMAT binary)
SELECT * FROM t2 ORDER BY val;
val
---------
@@ -15,37 +14,3 @@ SELECT * FROM t2 ORDER BY val;
DROP TABLE t;
DROP TABLE t2;
-- halfvec
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE TABLE t2 (val halfvec(3));
\copy t TO 'results/halfvec.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/halfvec.bin' WITH (FORMAT binary)
SELECT * FROM t2 ORDER BY val;
val
---------
[0,0,0]
[1,1,1]
[1,2,3]
(4 rows)
DROP TABLE t;
DROP TABLE t2;
-- sparsevec
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE TABLE t2 (val sparsevec(3));
\copy t TO 'results/sparsevec.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/sparsevec.bin' WITH (FORMAT binary)
SELECT * FROM t2 ORDER BY val;
val
-----------------
{}/3
{1:1,2:1,3:1}/3
{1:1,2:2,3:3}/3
(4 rows)
DROP TABLE t;
DROP TABLE t2;

210
test/expected/functions.out Normal file
View File

@@ -0,0 +1,210 @@
SELECT '[1,2,3]'::vector + '[4,5,6]';
?column?
----------
[5,7,9]
(1 row)
SELECT '[3e38]'::vector + '[3e38]';
ERROR: value out of range: overflow
SELECT '[1,2,3]'::vector - '[4,5,6]';
?column?
------------
[-3,-3,-3]
(1 row)
SELECT '[-3e38]'::vector - '[3e38]';
ERROR: value out of range: overflow
SELECT '[1,2,3]'::vector * '[4,5,6]';
?column?
-----------
[4,10,18]
(1 row)
SELECT '[1e37]'::vector * '[1e37]';
ERROR: value out of range: overflow
SELECT '[1e-37]'::vector * '[1e-37]';
ERROR: value out of range: underflow
SELECT vector_dims('[1,2,3]');
vector_dims
-------------
3
(1 row)
SELECT round(vector_norm('[1,1]')::numeric, 5);
round
---------
1.41421
(1 row)
SELECT vector_norm('[3,4]');
vector_norm
-------------
5
(1 row)
SELECT vector_norm('[0,1]');
vector_norm
-------------
1
(1 row)
SELECT vector_norm('[3e37,4e37]')::real;
vector_norm
-------------
5e+37
(1 row)
SELECT l2_distance('[0,0]', '[3,4]');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('[0,0]', '[0,1]');
l2_distance
-------------
1
(1 row)
SELECT l2_distance('[1,2]', '[3]');
ERROR: different vector dimensions 2 and 1
SELECT l2_distance('[3e38]', '[-3e38]');
l2_distance
-------------
Infinity
(1 row)
SELECT inner_product('[1,2]', '[3,4]');
inner_product
---------------
11
(1 row)
SELECT inner_product('[1,2]', '[3]');
ERROR: different vector dimensions 2 and 1
SELECT inner_product('[3e38]', '[3e38]');
inner_product
---------------
Infinity
(1 row)
SELECT cosine_distance('[1,2]', '[2,4]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,2]', '[0,0]');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('[1,1]', '[1,1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,0]', '[0,2]');
cosine_distance
-----------------
1
(1 row)
SELECT cosine_distance('[1,1]', '[-1,-1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[1,2]', '[3]');
ERROR: different vector dimensions 2 and 1
SELECT cosine_distance('[1,1]', '[1.1,1.1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,1]', '[-1.1,-1.1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[3e38]', '[3e38]');
cosine_distance
-----------------
NaN
(1 row)
SELECT l1_distance('[0,0]', '[3,4]');
l1_distance
-------------
7
(1 row)
SELECT l1_distance('[0,0]', '[0,1]');
l1_distance
-------------
1
(1 row)
SELECT l1_distance('[1,2]', '[3]');
ERROR: different vector dimensions 2 and 1
SELECT l1_distance('[3e38]', '[-3e38]');
l1_distance
-------------
Infinity
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
avg
-----------
[2,3.5,5]
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
avg
-----------
[2,3.5,5]
(1 row)
SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
avg
-----
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
ERROR: expected 2 dimensions, not 1
SELECT avg(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
avg
---------
[3e+38]
(1 row)
SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;
ERROR: vector cannot have more than 16000 dimensions
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
sum
----------
[4,7,10]
(1 row)
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
sum
----------
[4,7,10]
(1 row)
SELECT sum(v) FROM unnest(ARRAY[]::vector[]) v;
sum
-----
(1 row)
SELECT sum(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
ERROR: different vector dimensions 2 and 1
SELECT sum(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
ERROR: value out of range: overflow

View File

@@ -1,642 +0,0 @@
SELECT '[1,2,3]'::halfvec;
halfvec
---------
[1,2,3]
(1 row)
SELECT '[-1,-2,-3]'::halfvec;
halfvec
------------
[-1,-2,-3]
(1 row)
SELECT '[1.,2.,3.]'::halfvec;
halfvec
---------
[1,2,3]
(1 row)
SELECT ' [ 1, 2 , 3 ] '::halfvec;
halfvec
---------
[1,2,3]
(1 row)
SELECT '[1.23456]'::halfvec;
halfvec
------------
[1.234375]
(1 row)
SELECT '[hello,1]'::halfvec;
ERROR: invalid input syntax for type halfvec: "[hello,1]"
LINE 1: SELECT '[hello,1]'::halfvec;
^
SELECT '[NaN,1]'::halfvec;
ERROR: NaN not allowed in halfvec
LINE 1: SELECT '[NaN,1]'::halfvec;
^
SELECT '[Infinity,1]'::halfvec;
ERROR: infinite value not allowed in halfvec
LINE 1: SELECT '[Infinity,1]'::halfvec;
^
SELECT '[-Infinity,1]'::halfvec;
ERROR: infinite value not allowed in halfvec
LINE 1: SELECT '[-Infinity,1]'::halfvec;
^
SELECT '[65519,-65519]'::halfvec;
halfvec
----------------
[65504,-65504]
(1 row)
SELECT '[65520,-65520]'::halfvec;
ERROR: "65520" is out of range for type halfvec
LINE 1: SELECT '[65520,-65520]'::halfvec;
^
SELECT '[1e-8,-1e-8]'::halfvec;
halfvec
---------
[0,-0]
(1 row)
SELECT '[4e38,1]'::halfvec;
ERROR: "4e38" is out of range for type halfvec
LINE 1: SELECT '[4e38,1]'::halfvec;
^
SELECT '[1e-46,1]'::halfvec;
halfvec
---------
[0,1]
(1 row)
SELECT '[1,2,3'::halfvec;
ERROR: invalid input syntax for type halfvec: "[1,2,3"
LINE 1: SELECT '[1,2,3'::halfvec;
^
SELECT '[1,2,3]9'::halfvec;
ERROR: invalid input syntax for type halfvec: "[1,2,3]9"
LINE 1: SELECT '[1,2,3]9'::halfvec;
^
DETAIL: Junk after closing right brace.
SELECT '1,2,3'::halfvec;
ERROR: invalid input syntax for type halfvec: "1,2,3"
LINE 1: SELECT '1,2,3'::halfvec;
^
DETAIL: Vector contents must start with "[".
SELECT ''::halfvec;
ERROR: invalid input syntax for type halfvec: ""
LINE 1: SELECT ''::halfvec;
^
DETAIL: Vector contents must start with "[".
SELECT '['::halfvec;
ERROR: invalid input syntax for type halfvec: "["
LINE 1: SELECT '['::halfvec;
^
SELECT '[ '::halfvec;
ERROR: invalid input syntax for type halfvec: "[ "
LINE 1: SELECT '[ '::halfvec;
^
SELECT '[,'::halfvec;
ERROR: invalid input syntax for type halfvec: "[,"
LINE 1: SELECT '[,'::halfvec;
^
SELECT '[]'::halfvec;
ERROR: halfvec must have at least 1 dimension
LINE 1: SELECT '[]'::halfvec;
^
SELECT '[ ]'::halfvec;
ERROR: halfvec must have at least 1 dimension
LINE 1: SELECT '[ ]'::halfvec;
^
SELECT '[,]'::halfvec;
ERROR: invalid input syntax for type halfvec: "[,]"
LINE 1: SELECT '[,]'::halfvec;
^
SELECT '[1,]'::halfvec;
ERROR: invalid input syntax for type halfvec: "[1,]"
LINE 1: SELECT '[1,]'::halfvec;
^
SELECT '[1a]'::halfvec;
ERROR: invalid input syntax for type halfvec: "[1a]"
LINE 1: SELECT '[1a]'::halfvec;
^
SELECT '[1,,3]'::halfvec;
ERROR: invalid input syntax for type halfvec: "[1,,3]"
LINE 1: SELECT '[1,,3]'::halfvec;
^
SELECT '[1, ,3]'::halfvec;
ERROR: invalid input syntax for type halfvec: "[1, ,3]"
LINE 1: SELECT '[1, ,3]'::halfvec;
^
SELECT '[1,2,3]'::halfvec(3);
halfvec
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::halfvec(2);
ERROR: expected 2 dimensions, not 3
SELECT '[1,2,3]'::halfvec(3, 2);
ERROR: invalid type modifier
LINE 1: SELECT '[1,2,3]'::halfvec(3, 2);
^
SELECT '[1,2,3]'::halfvec('a');
ERROR: invalid input syntax for type integer: "a"
LINE 1: SELECT '[1,2,3]'::halfvec('a');
^
SELECT '[1,2,3]'::halfvec(0);
ERROR: dimensions for type halfvec must be at least 1
LINE 1: SELECT '[1,2,3]'::halfvec(0);
^
SELECT '[1,2,3]'::halfvec(16001);
ERROR: dimensions for type halfvec cannot exceed 16000
LINE 1: SELECT '[1,2,3]'::halfvec(16001);
^
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::halfvec[]);
unnest
---------
[1,2,3]
[4,5,6]
(2 rows)
SELECT '{"[1,2,3]"}'::halfvec(2)[];
ERROR: expected 2 dimensions, not 3
SELECT '[1,2,3]'::halfvec + '[4,5,6]';
?column?
----------
[5,7,9]
(1 row)
SELECT '[65519]'::halfvec + '[65519]';
ERROR: value out of range: overflow
SELECT '[1,2]'::halfvec + '[3]';
ERROR: different halfvec dimensions 2 and 1
SELECT '[1,2,3]'::halfvec - '[4,5,6]';
?column?
------------
[-3,-3,-3]
(1 row)
SELECT '[-65519]'::halfvec - '[65519]';
ERROR: value out of range: overflow
SELECT '[1,2]'::halfvec - '[3]';
ERROR: different halfvec dimensions 2 and 1
SELECT '[1,2,3]'::halfvec * '[4,5,6]';
?column?
-----------
[4,10,18]
(1 row)
SELECT '[65519]'::halfvec * '[65519]';
ERROR: value out of range: overflow
SELECT '[1e-7]'::halfvec * '[1e-7]';
ERROR: value out of range: underflow
SELECT '[1,2]'::halfvec * '[3]';
ERROR: different halfvec dimensions 2 and 1
SELECT '[1,2,3]'::halfvec || '[4,5]';
?column?
-------------
[1,2,3,4,5]
(1 row)
SELECT array_fill(0, ARRAY[16000])::halfvec || '[1]';
ERROR: halfvec cannot have more than 16000 dimensions
SELECT '[1,2,3]'::halfvec < '[1,2,3]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::halfvec < '[1,2]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::halfvec <= '[1,2,3]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::halfvec <= '[1,2]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::halfvec = '[1,2,3]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::halfvec = '[1,2]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::halfvec != '[1,2,3]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::halfvec != '[1,2]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::halfvec >= '[1,2,3]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::halfvec >= '[1,2]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::halfvec > '[1,2,3]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::halfvec > '[1,2]';
?column?
----------
t
(1 row)
SELECT halfvec_cmp('[1,2,3]', '[1,2,3]');
halfvec_cmp
-------------
0
(1 row)
SELECT halfvec_cmp('[1,2,3]', '[0,0,0]');
halfvec_cmp
-------------
1
(1 row)
SELECT halfvec_cmp('[0,0,0]', '[1,2,3]');
halfvec_cmp
-------------
-1
(1 row)
SELECT halfvec_cmp('[1,2]', '[1,2,3]');
halfvec_cmp
-------------
-1
(1 row)
SELECT halfvec_cmp('[1,2,3]', '[1,2]');
halfvec_cmp
-------------
1
(1 row)
SELECT halfvec_cmp('[1,2]', '[2,3,4]');
halfvec_cmp
-------------
-1
(1 row)
SELECT halfvec_cmp('[2,3]', '[1,2,3]');
halfvec_cmp
-------------
1
(1 row)
SELECT vector_dims('[1,2,3]'::halfvec);
vector_dims
-------------
3
(1 row)
SELECT round(l2_norm('[1,1]'::halfvec)::numeric, 5);
round
---------
1.41421
(1 row)
SELECT l2_norm('[3,4]'::halfvec);
l2_norm
---------
5
(1 row)
SELECT l2_norm('[0,1]'::halfvec);
l2_norm
---------
1
(1 row)
SELECT l2_norm('[0,0]'::halfvec);
l2_norm
---------
0
(1 row)
SELECT l2_norm('[2]'::halfvec);
l2_norm
---------
2
(1 row)
SELECT l2_distance('[0,0]'::halfvec, '[3,4]');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('[0,0]'::halfvec, '[0,1]');
l2_distance
-------------
1
(1 row)
SELECT l2_distance('[1,2]'::halfvec, '[3]');
ERROR: different halfvec dimensions 2 and 1
SELECT l2_distance('[1,1,1,1,1,1,1,1,1]'::halfvec, '[1,1,1,1,1,1,1,4,5]');
l2_distance
-------------
5
(1 row)
SELECT '[0,0]'::halfvec <-> '[3,4]';
?column?
----------
5
(1 row)
SELECT inner_product('[1,2]'::halfvec, '[3,4]');
inner_product
---------------
11
(1 row)
SELECT inner_product('[1,2]'::halfvec, '[3]');
ERROR: different halfvec dimensions 2 and 1
SELECT inner_product('[65504]'::halfvec, '[65504]');
inner_product
---------------
4290774016
(1 row)
SELECT inner_product('[1,1,1,1,1,1,1,1,1]'::halfvec, '[1,2,3,4,5,6,7,8,9]');
inner_product
---------------
45
(1 row)
SELECT '[1,2]'::halfvec <#> '[3,4]';
?column?
----------
-11
(1 row)
SELECT cosine_distance('[1,2]'::halfvec, '[2,4]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,2]'::halfvec, '[0,0]');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('[1,1]'::halfvec, '[1,1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,0]'::halfvec, '[0,2]');
cosine_distance
-----------------
1
(1 row)
SELECT cosine_distance('[1,1]'::halfvec, '[-1,-1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[1,2]'::halfvec, '[3]');
ERROR: different halfvec dimensions 2 and 1
SELECT cosine_distance('[1,1]'::halfvec, '[1.1,1.1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,1]'::halfvec, '[-1.1,-1.1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[1,2,3,4,5,6,7,8,9]'::halfvec, '[1,2,3,4,5,6,7,8,9]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,2,3,4,5,6,7,8,9]'::halfvec, '[-1,-2,-3,-4,-5,-6,-7,-8,-9]');
cosine_distance
-----------------
2
(1 row)
SELECT '[1,2]'::halfvec <=> '[2,4]';
?column?
----------
0
(1 row)
SELECT l1_distance('[0,0]'::halfvec, '[3,4]');
l1_distance
-------------
7
(1 row)
SELECT l1_distance('[0,0]'::halfvec, '[0,1]');
l1_distance
-------------
1
(1 row)
SELECT l1_distance('[1,2]'::halfvec, '[3]');
ERROR: different halfvec dimensions 2 and 1
SELECT l1_distance('[1,2,3,4,5,6,7,8,9]'::halfvec, '[1,2,3,4,5,6,7,8,9]');
l1_distance
-------------
0
(1 row)
SELECT l1_distance('[1,2,3,4,5,6,7,8,9]'::halfvec, '[0,3,2,5,4,7,6,9,8]');
l1_distance
-------------
9
(1 row)
SELECT '[0,0]'::halfvec <+> '[3,4]';
?column?
----------
7
(1 row)
SELECT l2_normalize('[3,4]'::halfvec);
l2_normalize
------------------------
[0.60009766,0.7998047]
(1 row)
SELECT l2_normalize('[3,0]'::halfvec);
l2_normalize
--------------
[1,0]
(1 row)
SELECT l2_normalize('[0,0.1]'::halfvec);
l2_normalize
--------------
[0,1]
(1 row)
SELECT l2_normalize('[0,0]'::halfvec);
l2_normalize
--------------
[0,0]
(1 row)
SELECT l2_normalize('[65504]'::halfvec);
l2_normalize
--------------
[1]
(1 row)
SELECT binary_quantize('[1,0,-1]'::halfvec);
binary_quantize
-----------------
100
(1 row)
SELECT binary_quantize('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]'::halfvec);
binary_quantize
-----------------
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
-----------
[1,2,3]
(1 row)
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 2);
subvector
-----------
[3,4]
(1 row)
SELECT subvector('[1,2,3,4,5]'::halfvec, -1, 3);
subvector
-----------
[1]
(1 row)
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 9);
subvector
-----------
[3,4,5]
(1 row)
SELECT subvector('[1,2,3,4,5]'::halfvec, 1, 0);
ERROR: halfvec must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, -1);
ERROR: halfvec must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::halfvec, -1, 2);
ERROR: halfvec must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::halfvec, 2147483647, 10);
ERROR: halfvec must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 2147483647);
subvector
-----------
[3,4,5]
(1 row)
SELECT subvector('[1,2,3,4,5]'::halfvec, -2147483644, 2147483647);
subvector
-----------
[1,2]
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::halfvec, '[3,5,7]']) v;
avg
-----------
[2,3.5,5]
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::halfvec, '[3,5,7]', NULL]) v;
avg
-----------
[2,3.5,5]
(1 row)
SELECT avg(v) FROM unnest(ARRAY[]::halfvec[]) v;
avg
-----
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2]'::halfvec, '[3]']) v;
ERROR: expected 2 dimensions, not 1
SELECT avg(v) FROM unnest(ARRAY['[65504]'::halfvec, '[65504]']) v;
avg
---------
[65504]
(1 row)
SELECT halfvec_avg(array_agg(n)) FROM generate_series(1, 16002) n;
ERROR: halfvec cannot have more than 16000 dimensions
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::halfvec, '[3,5,7]']) v;
sum
----------
[4,7,10]
(1 row)
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::halfvec, '[3,5,7]', NULL]) v;
sum
----------
[4,7,10]
(1 row)
SELECT sum(v) FROM unnest(ARRAY[]::halfvec[]) v;
sum
-----
(1 row)
SELECT sum(v) FROM unnest(ARRAY['[1,2]'::halfvec, '[3]']) v;
ERROR: different halfvec dimensions 2 and 1
SELECT sum(v) FROM unnest(ARRAY['[65504]'::halfvec, '[65504]']) v;
ERROR: value out of range: overflow

View File

@@ -1,51 +0,0 @@
SET enable_seqscan = off;
-- hamming
CREATE TABLE t (val bit(3));
INSERT INTO t (val) VALUES (B'000'), (B'100'), (B'111'), (NULL);
CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
INSERT INTO t (val) VALUES (B'110');
SELECT * FROM t ORDER BY val <~> B'111';
val
-----
111
110
100
000
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- jaccard
CREATE TABLE t (val bit(4));
INSERT INTO t (val) VALUES (B'0000'), (B'1100'), (B'1111'), (NULL);
CREATE INDEX ON t USING hnsw (val bit_jaccard_ops);
INSERT INTO t (val) VALUES (B'1110');
SELECT * FROM t ORDER BY val <%> B'1111';
val
------
1111
1110
1100
0000
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <%> (SELECT NULL::bit)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- varbit
CREATE TABLE t (val varbit(3));
CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
ERROR: type not supported for hnsw index
CREATE INDEX ON t USING hnsw ((val::bit(3)) bit_hamming_ops);
CREATE INDEX ON t USING hnsw ((val::bit(64001)) bit_hamming_ops);
ERROR: column cannot have more than 64000 dimensions for hnsw index
DROP TABLE t;

View File

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

View File

@@ -1,102 +0,0 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val halfvec_l2_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,2,4]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::halfvec)) t2;
count
-------
4
(1 row)
SELECT COUNT(*) FROM t;
count
-------
5
(1 row)
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
-----
(0 rows)
DROP TABLE t;
-- inner product
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val halfvec_ip_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
val
---------
[1,2,4]
[1,2,3]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::halfvec)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- cosine
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val halfvec_cosine_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
val
---------
[1,1,1]
[1,2,3]
[1,2,4]
(3 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
count
-------
3
(1 row)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::halfvec)) t2;
count
-------
3
(1 row)
DROP TABLE t;
-- L1
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val halfvec_l1_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <+> '[3,3,3]';
val
---------
[1,2,3]
[1,2,4]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <+> (SELECT NULL::halfvec)) t2;
count
-------
4
(1 row)
DROP TABLE t;

21
test/expected/hnsw_ip.out Normal file
View File

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

36
test/expected/hnsw_l2.out Normal file
View File

@@ -0,0 +1,36 @@
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
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]';
val
---------
[1,2,3]
[1,2,4]
[1,1,1]
[0,0,0]
(4 rows)
SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
val
---------
[0,0,0]
[1,1,1]
[1,2,3]
[1,2,4]
(4 rows)
SELECT COUNT(*) FROM t;
count
-------
5
(1 row)
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
-----
(0 rows)
DROP TABLE t;

View File

@@ -0,0 +1,26 @@
CREATE TABLE t (val vector(3));
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 1);
ERROR: value 1 out of bounds for option "m"
DETAIL: Valid values are between "2" and "100".
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 101);
ERROR: value 101 out of bounds for option "m"
DETAIL: Valid values are between "2" and "100".
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (ef_construction = 3);
ERROR: value 3 out of bounds for option "ef_construction"
DETAIL: Valid values are between "4" and "1000".
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (ef_construction = 1001);
ERROR: value 1001 out of bounds for option "ef_construction"
DETAIL: Valid values are between "4" and "1000".
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 16, ef_construction = 31);
ERROR: ef_construction must be greater than or equal to 2 * m
SHOW hnsw.ef_search;
hnsw.ef_search
----------------
40
(1 row)
SET hnsw.ef_search = 0;
ERROR: 0 is outside the valid range for parameter "hnsw.ef_search" (1 .. 1000)
SET hnsw.ef_search = 1001;
ERROR: 1001 is outside the valid range for parameter "hnsw.ef_search" (1 .. 1000)
DROP TABLE t;

View File

@@ -1,112 +0,0 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
SELECT * FROM t ORDER BY val <-> '{1:3,2:3,3:3}/3';
val
-----------------
{1:1,2:2,3:3}/3
{1:1,2:2,3:4}/3
{1:1,2:1,3:1}/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 <-> '{1:3,2:3,3:3}/3';
val
-----
(0 rows)
DROP TABLE t;
-- inner product
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
SELECT * FROM t ORDER BY val <#> '{1:3,2:3,3:3}/3';
val
-----------------
{1:1,2:2,3:4}/3
{1:1,2:2,3:3}/3
{1:1,2:1,3:1}/3
{}/3
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- cosine
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
SELECT * FROM t ORDER BY val <=> '{1:3,2:3,3:3}/3';
val
-----------------
{1:1,2:1,3:1}/3
{1:1,2:2,3:3}/3
{1:1,2:2,3:4}/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;
-- L1
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_l1_ops);
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
SELECT * FROM t ORDER BY val <+> '{1:3,2:3,3:3}/3';
val
-----------------
{1:1,2:2,3:3}/3
{1:1,2:2,3:4}/3
{1:1,2:1,3:1}/3
{}/3
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <+> (SELECT NULL::sparsevec)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- non-zero elements
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;

View File

@@ -0,0 +1,13 @@
SET enable_seqscan = off;
CREATE UNLOGGED TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_l2_ops);
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,1,1]
[0,0,0]
(3 rows)
DROP TABLE t;

View File

@@ -1,201 +0,0 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
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]';
val
---------
[1,2,3]
[1,2,4]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector)) t2;
count
-------
4
(1 row)
SELECT COUNT(*) FROM t;
count
-------
5
(1 row)
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
-----
(0 rows)
DROP TABLE t;
-- inner product
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_ip_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
val
---------
[1,2,4]
[1,2,3]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- cosine
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_cosine_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
val
---------
[1,1,1]
[1,2,3]
[1,2,4]
(3 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
count
-------
3
(1 row)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
count
-------
3
(1 row)
DROP TABLE t;
-- L1
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_l1_ops);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <+> '[3,3,3]';
val
---------
[1,2,3]
[1,2,4]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <+> (SELECT NULL::vector)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- iterative
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_l2_ops);
SET hnsw.iterative_scan = strict_order;
SET hnsw.ef_search = 1;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,1,1]
[0,0,0]
(3 rows)
SET hnsw.iterative_scan = relaxed_order;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,1,1]
[0,0,0]
(3 rows)
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
-----
(0 rows)
RESET hnsw.iterative_scan;
RESET hnsw.ef_search;
DROP TABLE t;
-- unlogged
CREATE UNLOGGED TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_l2_ops);
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,1,1]
[0,0,0]
(3 rows)
DROP TABLE t;
-- options
CREATE TABLE t (val vector(3));
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 1);
ERROR: value 1 out of bounds for option "m"
DETAIL: Valid values are between "2" and "100".
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 101);
ERROR: value 101 out of bounds for option "m"
DETAIL: Valid values are between "2" and "100".
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (ef_construction = 3);
ERROR: value 3 out of bounds for option "ef_construction"
DETAIL: Valid values are between "4" and "1000".
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (ef_construction = 1001);
ERROR: value 1001 out of bounds for option "ef_construction"
DETAIL: Valid values are between "4" and "1000".
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 16, ef_construction = 31);
ERROR: ef_construction must be greater than or equal to 2 * m
SHOW hnsw.ef_search;
hnsw.ef_search
----------------
40
(1 row)
SET hnsw.ef_search = 0;
ERROR: 0 is outside the valid range for parameter "hnsw.ef_search" (1 .. 1000)
SET hnsw.ef_search = 1001;
ERROR: 1001 is outside the valid range for parameter "hnsw.ef_search" (1 .. 1000)
SHOW hnsw.iterative_scan;
hnsw.iterative_scan
---------------------
off
(1 row)
SET hnsw.iterative_scan = on;
ERROR: invalid value for parameter "hnsw.iterative_scan": "on"
HINT: Available values: off, relaxed_order, strict_order.
SHOW hnsw.max_scan_tuples;
hnsw.max_scan_tuples
----------------------
20000
(1 row)
SET hnsw.max_scan_tuples = 0;
ERROR: 0 is outside the valid range for parameter "hnsw.max_scan_tuples" (1 .. 2147483647)
SHOW hnsw.scan_mem_multiplier;
hnsw.scan_mem_multiplier
--------------------------
1
(1 row)
SET hnsw.scan_mem_multiplier = 0;
ERROR: 0 is outside the valid range for parameter "hnsw.scan_mem_multiplier" (1 .. 1000)
SET hnsw.scan_mem_multiplier = 1001;
ERROR: 1001 is outside the valid range for parameter "hnsw.scan_mem_multiplier" (1 .. 1000)
DROP TABLE t;

129
test/expected/input.out Normal file
View File

@@ -0,0 +1,129 @@
SELECT '[1,2,3]'::vector;
vector
---------
[1,2,3]
(1 row)
SELECT '[-1,-2,-3]'::vector;
vector
------------
[-1,-2,-3]
(1 row)
SELECT '[1.,2.,3.]'::vector;
vector
---------
[1,2,3]
(1 row)
SELECT ' [ 1, 2 , 3 ] '::vector;
vector
---------
[1,2,3]
(1 row)
SELECT '[1.23456]'::vector;
vector
-----------
[1.23456]
(1 row)
SELECT '[hello,1]'::vector;
ERROR: invalid input syntax for type vector: "[hello,1]"
LINE 1: SELECT '[hello,1]'::vector;
^
SELECT '[NaN,1]'::vector;
ERROR: NaN not allowed in vector
LINE 1: SELECT '[NaN,1]'::vector;
^
SELECT '[Infinity,1]'::vector;
ERROR: infinite value not allowed in vector
LINE 1: SELECT '[Infinity,1]'::vector;
^
SELECT '[-Infinity,1]'::vector;
ERROR: infinite value not allowed in vector
LINE 1: SELECT '[-Infinity,1]'::vector;
^
SELECT '[1.5e38,-1.5e38]'::vector;
vector
--------------------
[1.5e+38,-1.5e+38]
(1 row)
SELECT '[1.5e+38,-1.5e+38]'::vector;
vector
--------------------
[1.5e+38,-1.5e+38]
(1 row)
SELECT '[1.5e-38,-1.5e-38]'::vector;
vector
--------------------
[1.5e-38,-1.5e-38]
(1 row)
SELECT '[4e38,1]'::vector;
ERROR: infinite value not allowed in vector
LINE 1: SELECT '[4e38,1]'::vector;
^
SELECT '[1,2,3'::vector;
ERROR: malformed vector literal: "[1,2,3"
LINE 1: SELECT '[1,2,3'::vector;
^
DETAIL: Unexpected end of input.
SELECT '[1,2,3]9'::vector;
ERROR: malformed vector literal: "[1,2,3]9"
LINE 1: SELECT '[1,2,3]9'::vector;
^
DETAIL: Junk after closing right brace.
SELECT '1,2,3'::vector;
ERROR: malformed vector literal: "1,2,3"
LINE 1: SELECT '1,2,3'::vector;
^
DETAIL: Vector contents must start with "[".
SELECT ''::vector;
ERROR: malformed vector literal: ""
LINE 1: SELECT ''::vector;
^
DETAIL: Vector contents must start with "[".
SELECT '['::vector;
ERROR: malformed vector literal: "["
LINE 1: SELECT '['::vector;
^
DETAIL: Unexpected end of input.
SELECT '[,'::vector;
ERROR: malformed vector literal: "[,"
LINE 1: SELECT '[,'::vector;
^
DETAIL: Unexpected end of input.
SELECT '[]'::vector;
ERROR: vector must have at least 1 dimension
LINE 1: SELECT '[]'::vector;
^
SELECT '[1,]'::vector;
ERROR: invalid input syntax for type vector: "[1,]"
LINE 1: SELECT '[1,]'::vector;
^
SELECT '[1a]'::vector;
ERROR: invalid input syntax for type vector: "[1a]"
LINE 1: SELECT '[1a]'::vector;
^
SELECT '[1,,3]'::vector;
ERROR: malformed vector literal: "[1,,3]"
LINE 1: SELECT '[1,,3]'::vector;
^
SELECT '[1, ,3]'::vector;
ERROR: invalid input syntax for type vector: "[1, ,3]"
LINE 1: SELECT '[1, ,3]'::vector;
^
SELECT '[1,2,3]'::vector(2);
ERROR: expected 2 dimensions, not 3
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
unnest
---------
[1,2,3]
[4,5,6]
(2 rows)
SELECT '{"[1,2,3]"}'::vector(2)[];
ERROR: expected 2 dimensions, not 3

View File

@@ -1,37 +0,0 @@
SET enable_seqscan = off;
-- hamming
CREATE TABLE t (val bit(3));
INSERT INTO t (val) VALUES (B'000'), (B'100'), (B'111'), (NULL);
CREATE INDEX ON t USING ivfflat (val bit_hamming_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES (B'110');
SELECT * FROM t ORDER BY val <~> B'111';
val
-----
111
110
100
000
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- varbit
CREATE TABLE t (val varbit(3));
CREATE INDEX ON t USING ivfflat (val bit_hamming_ops) WITH (lists = 1);
ERROR: type not supported for ivfflat index
CREATE INDEX ON t USING ivfflat ((val::bit(3)) bit_hamming_ops) WITH (lists = 1);
NOTICE: ivfflat index created with little data
DETAIL: This will cause low recall.
HINT: Drop the index until the table has more data.
CREATE INDEX ON t USING ivfflat ((val::bit(64001)) bit_hamming_ops) WITH (lists = 1);
ERROR: column cannot have more than 64000 dimensions for ivfflat index
CREATE INDEX ON t USING ivfflat ((val::bit(2)) bit_hamming_ops) WITH (lists = 5);
NOTICE: ivfflat index created with little data
DETAIL: This will cause low recall.
HINT: Drop the index until the table has more data.
DROP TABLE t;

View File

@@ -0,0 +1,26 @@
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_cosine_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
val
---------
[1,1,1]
[1,2,3]
[1,2,4]
(3 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
count
-------
3
(1 row)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
count
-------
3
(1 row)
DROP TABLE t;

View File

@@ -1,84 +0,0 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val halfvec_l2_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,2,4]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::halfvec)) t2;
count
-------
4
(1 row)
SELECT COUNT(*) FROM t;
count
-------
5
(1 row)
TRUNCATE t;
NOTICE: ivfflat index created with little data
DETAIL: This will cause low recall.
HINT: Drop the index until the table has more data.
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
-----
(0 rows)
DROP TABLE t;
-- inner product
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val halfvec_ip_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
val
---------
[1,2,4]
[1,2,3]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::halfvec)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- cosine
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val halfvec_cosine_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
val
---------
[1,1,1]
[1,2,3]
[1,2,4]
(3 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
count
-------
3
(1 row)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::halfvec)) t2;
count
-------
3
(1 row)
DROP TABLE t;

View File

@@ -0,0 +1,21 @@
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_ip_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
val
---------
[1,2,4]
[1,2,3]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector)) t2;
count
-------
4
(1 row)
DROP TABLE t;

View File

@@ -0,0 +1,39 @@
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,2,4]
[1,1,1]
[0,0,0]
(4 rows)
SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
val
---------
[0,0,0]
[1,1,1]
[1,2,3]
[1,2,4]
(4 rows)
SELECT COUNT(*) FROM t;
count
-------
5
(1 row)
TRUNCATE t;
NOTICE: ivfflat index created with little data
DETAIL: This will cause low recall.
HINT: Drop the index until the table has more data.
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
-----
(0 rows)
DROP TABLE t;

View File

@@ -0,0 +1,14 @@
CREATE TABLE t (val vector(3));
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 0);
ERROR: value 0 out of bounds for option "lists"
DETAIL: Valid values are between "1" and "32768".
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 32769);
ERROR: value 32769 out of bounds for option "lists"
DETAIL: Valid values are between "1" and "32768".
SHOW ivfflat.probes;
ivfflat.probes
----------------
1
(1 row)
DROP TABLE t;

View File

@@ -0,0 +1,13 @@
SET enable_seqscan = off;
CREATE UNLOGGED TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 1);
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,1,1]
[0,0,0]
(3 rows)
DROP TABLE t;

View File

@@ -1,175 +0,0 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,2,4]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector)) t2;
count
-------
4
(1 row)
SELECT COUNT(*) FROM t;
count
-------
5
(1 row)
TRUNCATE t;
NOTICE: ivfflat index created with little data
DETAIL: This will cause low recall.
HINT: Drop the index until the table has more data.
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
-----
(0 rows)
DROP TABLE t;
-- inner product
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_ip_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <#> '[3,3,3]';
val
---------
[1,2,4]
[1,2,3]
[1,1,1]
[0,0,0]
(4 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- cosine
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_cosine_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <=> '[3,3,3]';
val
---------
[1,1,1]
[1,2,3]
[1,2,4]
(3 rows)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> '[0,0,0]') t2;
count
-------
3
(1 row)
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
count
-------
3
(1 row)
DROP TABLE t;
-- iterative
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 3);
SET ivfflat.iterative_scan = relaxed_order;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,1,1]
[0,0,0]
(3 rows)
SET ivfflat.max_probes = 1;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
(1 row)
SET ivfflat.max_probes = 2;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,1,1]
(2 rows)
TRUNCATE t;
NOTICE: ivfflat index created with little data
DETAIL: This will cause low recall.
HINT: Drop the index until the table has more data.
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
-----
(0 rows)
RESET ivfflat.iterative_scan;
RESET ivfflat.max_probes;
DROP TABLE t;
-- unlogged
CREATE UNLOGGED TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 1);
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
val
---------
[1,2,3]
[1,1,1]
[0,0,0]
(3 rows)
DROP TABLE t;
-- options
CREATE TABLE t (val vector(3));
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 0);
ERROR: value 0 out of bounds for option "lists"
DETAIL: Valid values are between "1" and "32768".
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 32769);
ERROR: value 32769 out of bounds for option "lists"
DETAIL: Valid values are between "1" and "32768".
SHOW ivfflat.probes;
ivfflat.probes
----------------
1
(1 row)
SET ivfflat.probes = 0;
ERROR: 0 is outside the valid range for parameter "ivfflat.probes" (1 .. 32768)
SET ivfflat.probes = 32769;
ERROR: 32769 is outside the valid range for parameter "ivfflat.probes" (1 .. 32768)
SHOW ivfflat.iterative_scan;
ivfflat.iterative_scan
------------------------
off
(1 row)
SET ivfflat.iterative_scan = on;
ERROR: invalid value for parameter "ivfflat.iterative_scan": "on"
HINT: Available values: off, relaxed_order.
SHOW ivfflat.max_probes;
ivfflat.max_probes
--------------------
32768
(1 row)
SET ivfflat.max_probes = 0;
ERROR: 0 is outside the valid range for parameter "ivfflat.max_probes" (1 .. 32768)
SET ivfflat.max_probes = 32769;
ERROR: 32769 is outside the valid range for parameter "ivfflat.max_probes" (1 .. 32768)
DROP TABLE t;

View File

@@ -1,653 +0,0 @@
SELECT '{1:1.5,3:3.5}/5'::sparsevec;
sparsevec
-----------------
{1:1.5,3:3.5}/5
(1 row)
SELECT '{1:-2,3:-4}/5'::sparsevec;
sparsevec
---------------
{1:-2,3:-4}/5
(1 row)
SELECT '{1:2.,3:4.}/5'::sparsevec;
sparsevec
-------------
{1:2,3:4}/5
(1 row)
SELECT ' { 1 : 1.5 , 3 : 3.5 } / 5 '::sparsevec;
sparsevec
-----------------
{1:1.5,3:3.5}/5
(1 row)
SELECT '{1:1.23456}/1'::sparsevec;
sparsevec
---------------
{1:1.23456}/1
(1 row)
SELECT '{1:hello,2:1}/2'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{1:hello,2:1}/2"
LINE 1: SELECT '{1:hello,2:1}/2'::sparsevec;
^
SELECT '{1:NaN,2:1}/2'::sparsevec;
ERROR: NaN not allowed in sparsevec
LINE 1: SELECT '{1:NaN,2:1}/2'::sparsevec;
^
SELECT '{1:Infinity,2:1}/2'::sparsevec;
ERROR: infinite value not allowed in sparsevec
LINE 1: SELECT '{1:Infinity,2:1}/2'::sparsevec;
^
SELECT '{1:-Infinity,2:1}/2'::sparsevec;
ERROR: infinite value not allowed in sparsevec
LINE 1: SELECT '{1:-Infinity,2:1}/2'::sparsevec;
^
SELECT '{1:1.5e38,2:-1.5e38}/2'::sparsevec;
sparsevec
--------------------------
{1:1.5e+38,2:-1.5e+38}/2
(1 row)
SELECT '{1:1.5e+38,2:-1.5e+38}/2'::sparsevec;
sparsevec
--------------------------
{1:1.5e+38,2:-1.5e+38}/2
(1 row)
SELECT '{1:1.5e-38,2:-1.5e-38}/2'::sparsevec;
sparsevec
--------------------------
{1:1.5e-38,2:-1.5e-38}/2
(1 row)
SELECT '{1:4e38,2:1}/2'::sparsevec;
ERROR: "4e38" is out of range for type sparsevec
LINE 1: SELECT '{1:4e38,2:1}/2'::sparsevec;
^
SELECT '{1:-4e38,2:1}/2'::sparsevec;
ERROR: "-4e38" is out of range for type sparsevec
LINE 1: SELECT '{1:-4e38,2:1}/2'::sparsevec;
^
SELECT '{1:1e-46,2:1}/2'::sparsevec;
ERROR: "1e-46" is out of range for type sparsevec
LINE 1: SELECT '{1:1e-46,2:1}/2'::sparsevec;
^
SELECT '{1:-1e-46,2:1}/2'::sparsevec;
ERROR: "-1e-46" is out of range for type sparsevec
LINE 1: SELECT '{1:-1e-46,2:1}/2'::sparsevec;
^
SELECT ''::sparsevec;
ERROR: invalid input syntax for type sparsevec: ""
LINE 1: SELECT ''::sparsevec;
^
DETAIL: Vector contents must start with "{".
SELECT '{'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{"
LINE 1: SELECT '{'::sparsevec;
^
SELECT '{ '::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{ "
LINE 1: SELECT '{ '::sparsevec;
^
SELECT '{:'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{:"
LINE 1: SELECT '{:'::sparsevec;
^
SELECT '{,'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{,"
LINE 1: SELECT '{,'::sparsevec;
^
SELECT '{}'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{}"
LINE 1: SELECT '{}'::sparsevec;
^
DETAIL: Unexpected end of input.
SELECT '{}/'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{}/"
LINE 1: SELECT '{}/'::sparsevec;
^
SELECT '{}/1'::sparsevec;
sparsevec
-----------
{}/1
(1 row)
SELECT '{}/1a'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{}/1a"
LINE 1: SELECT '{}/1a'::sparsevec;
^
DETAIL: Junk after closing.
SELECT '{ }/1'::sparsevec;
sparsevec
-----------
{}/1
(1 row)
SELECT '{:}/1'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{:}/1"
LINE 1: SELECT '{:}/1'::sparsevec;
^
SELECT '{,}/1'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{,}/1"
LINE 1: SELECT '{,}/1'::sparsevec;
^
SELECT '{1,}/1'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{1,}/1"
LINE 1: SELECT '{1,}/1'::sparsevec;
^
SELECT '{:1}/1'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{:1}/1"
LINE 1: SELECT '{:1}/1'::sparsevec;
^
SELECT '{1:}/1'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{1:}/1"
LINE 1: SELECT '{1:}/1'::sparsevec;
^
SELECT '{1a:1}/1'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{1a:1}/1"
LINE 1: SELECT '{1a:1}/1'::sparsevec;
^
SELECT '{1:1a}/1'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{1:1a}/1"
LINE 1: SELECT '{1:1a}/1'::sparsevec;
^
SELECT '{1:1,}/1'::sparsevec;
ERROR: invalid input syntax for type sparsevec: "{1:1,}/1"
LINE 1: SELECT '{1:1,}/1'::sparsevec;
^
SELECT '{1:0,2:1,3:0}/3'::sparsevec;
sparsevec
-----------
{2:1}/3
(1 row)
SELECT '{2:1,1:1}/2'::sparsevec;
sparsevec
-------------
{1:1,2:1}/2
(1 row)
SELECT '{1:1,1:1}/2'::sparsevec;
ERROR: sparsevec indices must not contain duplicates
LINE 1: SELECT '{1:1,1:1}/2'::sparsevec;
^
SELECT '{1:1,2:1,1:1}/2'::sparsevec;
ERROR: sparsevec indices must not contain duplicates
LINE 1: SELECT '{1:1,2:1,1:1}/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 '{}/1000000001'::sparsevec;
ERROR: sparsevec cannot have more than 1000000000 dimensions
LINE 1: SELECT '{}/1000000001'::sparsevec;
^
SELECT '{}/2147483648'::sparsevec;
ERROR: sparsevec cannot have more than 1000000000 dimensions
LINE 1: SELECT '{}/2147483648'::sparsevec;
^
SELECT '{}/-2147483649'::sparsevec;
ERROR: sparsevec must have at least 1 dimension
LINE 1: SELECT '{}/-2147483649'::sparsevec;
^
SELECT '{}/9223372036854775808'::sparsevec;
ERROR: sparsevec cannot have more than 1000000000 dimensions
LINE 1: SELECT '{}/9223372036854775808'::sparsevec;
^
SELECT '{}/-9223372036854775809'::sparsevec;
ERROR: sparsevec must have at least 1 dimension
LINE 1: SELECT '{}/-9223372036854775809'::sparsevec;
^
SELECT '{2147483647:1}/1'::sparsevec;
ERROR: sparsevec index out of bounds
LINE 1: SELECT '{2147483647:1}/1'::sparsevec;
^
SELECT '{2147483648:1}/1'::sparsevec;
ERROR: sparsevec index out of bounds
LINE 1: SELECT '{2147483648:1}/1'::sparsevec;
^
SELECT '{-2147483648:1}/1'::sparsevec;
ERROR: sparsevec index out of bounds
LINE 1: SELECT '{-2147483648:1}/1'::sparsevec;
^
SELECT '{-2147483649:1}/1'::sparsevec;
ERROR: sparsevec index out of bounds
LINE 1: SELECT '{-2147483649:1}/1'::sparsevec;
^
SELECT '{0:1}/1'::sparsevec;
ERROR: sparsevec index out of bounds
LINE 1: SELECT '{0:1}/1'::sparsevec;
^
SELECT '{2:1}/1'::sparsevec;
ERROR: sparsevec index out of bounds
LINE 1: SELECT '{2:1}/1'::sparsevec;
^
SELECT '{}/3'::sparsevec(3);
sparsevec
-----------
{}/3
(1 row)
SELECT '{}/3'::sparsevec(2);
ERROR: expected 2 dimensions, not 3
SELECT '{}/3'::sparsevec(3, 2);
ERROR: invalid type modifier
LINE 1: SELECT '{}/3'::sparsevec(3, 2);
^
SELECT '{}/3'::sparsevec('a');
ERROR: invalid input syntax for type integer: "a"
LINE 1: SELECT '{}/3'::sparsevec('a');
^
SELECT '{}/3'::sparsevec(0);
ERROR: dimensions for type sparsevec must be at least 1
LINE 1: SELECT '{}/3'::sparsevec(0);
^
SELECT '{}/3'::sparsevec(1000000001);
ERROR: dimensions for type sparsevec cannot exceed 1000000000
LINE 1: SELECT '{}/3'::sparsevec(1000000001);
^
SELECT '{1:1,2:2,3:3}/3'::sparsevec < '{1:1,2:2,3:3}/3';
?column?
----------
f
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec < '{1:1,2:2}/2';
?column?
----------
f
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec <= '{1:1,2:2,3:3}/3';
?column?
----------
t
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec <= '{1:1,2:2}/2';
?column?
----------
f
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec = '{1:1,2:2,3:3}/3';
?column?
----------
t
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec = '{1:1,2:2}/2';
?column?
----------
f
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec != '{1:1,2:2,3:3}/3';
?column?
----------
f
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec != '{1:1,2:2}/2';
?column?
----------
t
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec >= '{1:1,2:2,3:3}/3';
?column?
----------
t
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec >= '{1:1,2:2}/2';
?column?
----------
t
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec > '{1:1,2:2,3:3}/3';
?column?
----------
f
(1 row)
SELECT '{1:1,2:2,3:3}/3'::sparsevec > '{1:1,2:2}/2';
?column?
----------
t
(1 row)
SELECT sparsevec_cmp('{1:1,2:2,3:3}/3', '{1:1,2:2,3:3}/3');
sparsevec_cmp
---------------
0
(1 row)
SELECT sparsevec_cmp('{1:1,2:2,3:3}/3', '{}/3');
sparsevec_cmp
---------------
1
(1 row)
SELECT sparsevec_cmp('{}/3', '{1:1,2:2,3:3}/3');
sparsevec_cmp
---------------
-1
(1 row)
SELECT sparsevec_cmp('{1:1,2:2}/2', '{1:1,2:2,3:3}/3');
sparsevec_cmp
---------------
-1
(1 row)
SELECT sparsevec_cmp('{1:1,2:2,3:3}/3', '{1:1,2:2}/2');
sparsevec_cmp
---------------
1
(1 row)
SELECT sparsevec_cmp('{1:1,2:2}/2', '{1:2,2:3,3:4}/3');
sparsevec_cmp
---------------
-1
(1 row)
SELECT sparsevec_cmp('{1:2,2:3}/2', '{1:1,2:2,3:3}/3');
sparsevec_cmp
---------------
1
(1 row)
SELECT round(l2_norm('{1:1,2:1}/2'::sparsevec)::numeric, 5);
round
---------
1.41421
(1 row)
SELECT l2_norm('{1:3,2:4}/2'::sparsevec);
l2_norm
---------
5
(1 row)
SELECT l2_norm('{2:1}/2'::sparsevec);
l2_norm
---------
1
(1 row)
SELECT l2_norm('{1:3e37,2:4e37}/2'::sparsevec)::real;
l2_norm
---------
5e+37
(1 row)
SELECT l2_norm('{}/2'::sparsevec);
l2_norm
---------
0
(1 row)
SELECT l2_norm('{1:2}/1'::sparsevec);
l2_norm
---------
2
(1 row)
SELECT l2_distance('{}/2'::sparsevec, '{1:3,2:4}/2');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('{1:3}/2'::sparsevec, '{2:4}/2');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('{2:4}/2'::sparsevec, '{1:3}/2');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('{1:3,2:4}/2'::sparsevec, '{}/2');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('{}/2'::sparsevec, '{2:1}/2');
l2_distance
-------------
1
(1 row)
SELECT '{}/2'::sparsevec <-> '{1:3,2:4}/2';
?column?
----------
5
(1 row)
SELECT inner_product('{1:1,2:2}/2'::sparsevec, '{1:2,2:4}/2');
inner_product
---------------
10
(1 row)
SELECT inner_product('{1:1,2:2}/2'::sparsevec, '{1:3}/1');
ERROR: different sparsevec dimensions 2 and 1
SELECT inner_product('{1:1,3:3}/4'::sparsevec, '{2:2,4:4}/4');
inner_product
---------------
0
(1 row)
SELECT inner_product('{2:2,4:4}/4'::sparsevec, '{1:1,3:3}/4');
inner_product
---------------
0
(1 row)
SELECT inner_product('{1:1,3:3,5:5}/5'::sparsevec, '{2:4,3:6,4:8}/5');
inner_product
---------------
18
(1 row)
SELECT inner_product('{1:1}/2'::sparsevec, '{}/2');
inner_product
---------------
0
(1 row)
SELECT inner_product('{}/2'::sparsevec, '{1:1}/2');
inner_product
---------------
0
(1 row)
SELECT inner_product('{1:3e38}/1'::sparsevec, '{1:3e38}/1');
inner_product
---------------
Infinity
(1 row)
SELECT inner_product('{1:1,3:3,5:5}/5'::sparsevec, '{2:4,3:6,4:8}/5');
inner_product
---------------
18
(1 row)
SELECT '{1:1,2:2}/2'::sparsevec <#> '{1:3,2:4}/2';
?column?
----------
-11
(1 row)
SELECT cosine_distance('{1:1,2:2}/2'::sparsevec, '{1:2,2:4}/2');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('{1:1,2:2}/2'::sparsevec, '{}/2');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('{1:1,2:1}/2'::sparsevec, '{1:1,2:1}/2');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('{1:1}/2'::sparsevec, '{2:2}/2');
cosine_distance
-----------------
1
(1 row)
SELECT cosine_distance('{1:1,2:1}/2'::sparsevec, '{1:-1,2:-1}/2');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('{1:2}/2'::sparsevec, '{2:2}/2');
cosine_distance
-----------------
1
(1 row)
SELECT cosine_distance('{2:2}/2'::sparsevec, '{1:2}/2');
cosine_distance
-----------------
1
(1 row)
SELECT cosine_distance('{1:1,2:2}/2'::sparsevec, '{1:3}/1');
ERROR: different sparsevec dimensions 2 and 1
SELECT cosine_distance('{1:1,2:1}/2'::sparsevec, '{1:1.1,2:1.1}/2');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('{1:1,2:1}/2'::sparsevec, '{1:-1.1,2:-1.1}/2');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('{1:3e38}/1'::sparsevec, '{1:3e38}/1');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('{}/1'::sparsevec, '{}/1');
cosine_distance
-----------------
NaN
(1 row)
SELECT '{1:1,2:2}/2'::sparsevec <=> '{1:2,2:4}/2';
?column?
----------
0
(1 row)
SELECT l1_distance('{}/2'::sparsevec, '{1:3,2:4}/2');
l1_distance
-------------
7
(1 row)
SELECT l1_distance('{}/2'::sparsevec, '{2:1}/2');
l1_distance
-------------
1
(1 row)
SELECT l1_distance('{1:1,2:2}/2'::sparsevec, '{1:3}/1');
ERROR: different sparsevec dimensions 2 and 1
SELECT l1_distance('{1:3e38}/1'::sparsevec, '{1:-3e38}/1');
l1_distance
-------------
Infinity
(1 row)
SELECT l1_distance('{1:1,3:3,5:5,7:7}/8'::sparsevec, '{2:2,4:4,6:6,8:8}/8');
l1_distance
-------------
36
(1 row)
SELECT l1_distance('{1:1,3:3,5:5,7:7,9:9}/9'::sparsevec, '{2:2,4:4,6:6,8:8}/9');
l1_distance
-------------
45
(1 row)
SELECT '{}/2'::sparsevec <+> '{1:3,2:4}/2';
?column?
----------
7
(1 row)
SELECT l2_normalize('{1:3,2:4}/2'::sparsevec);
l2_normalize
-----------------
{1:0.6,2:0.8}/2
(1 row)
SELECT l2_normalize('{1:3}/2'::sparsevec);
l2_normalize
--------------
{1:1}/2
(1 row)
SELECT l2_normalize('{2:0.1}/2'::sparsevec);
l2_normalize
--------------
{2:1}/2
(1 row)
SELECT l2_normalize('{}/2'::sparsevec);
l2_normalize
--------------
{}/2
(1 row)
SELECT l2_normalize('{1:3e38}/1'::sparsevec);
l2_normalize
--------------
{1:1}/1
(1 row)
SELECT l2_normalize('{1:3e38,2:1e-37}/2'::sparsevec);
l2_normalize
--------------
{1:1}/2
(1 row)
SELECT l2_normalize('{2:3e37,4:3e-37,6:4e37,8:4e-37}/9'::sparsevec);
l2_normalize
-----------------
{2:0.6,6:0.8}/9
(1 row)

View File

@@ -1,678 +0,0 @@
SELECT '[1,2,3]'::vector;
vector
---------
[1,2,3]
(1 row)
SELECT '[-1,-2,-3]'::vector;
vector
------------
[-1,-2,-3]
(1 row)
SELECT '[1.,2.,3.]'::vector;
vector
---------
[1,2,3]
(1 row)
SELECT ' [ 1, 2 , 3 ] '::vector;
vector
---------
[1,2,3]
(1 row)
SELECT '[1.23456]'::vector;
vector
-----------
[1.23456]
(1 row)
SELECT '[hello,1]'::vector;
ERROR: invalid input syntax for type vector: "[hello,1]"
LINE 1: SELECT '[hello,1]'::vector;
^
SELECT '[NaN,1]'::vector;
ERROR: NaN not allowed in vector
LINE 1: SELECT '[NaN,1]'::vector;
^
SELECT '[Infinity,1]'::vector;
ERROR: infinite value not allowed in vector
LINE 1: SELECT '[Infinity,1]'::vector;
^
SELECT '[-Infinity,1]'::vector;
ERROR: infinite value not allowed in vector
LINE 1: SELECT '[-Infinity,1]'::vector;
^
SELECT '[1.5e38,-1.5e38]'::vector;
vector
--------------------
[1.5e+38,-1.5e+38]
(1 row)
SELECT '[1.5e+38,-1.5e+38]'::vector;
vector
--------------------
[1.5e+38,-1.5e+38]
(1 row)
SELECT '[1.5e-38,-1.5e-38]'::vector;
vector
--------------------
[1.5e-38,-1.5e-38]
(1 row)
SELECT '[4e38,1]'::vector;
ERROR: "4e38" is out of range for type vector
LINE 1: SELECT '[4e38,1]'::vector;
^
SELECT '[-4e38,1]'::vector;
ERROR: "-4e38" is out of range for type 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: invalid input syntax for type vector: "[1,2,3"
LINE 1: SELECT '[1,2,3'::vector;
^
SELECT '[1,2,3]9'::vector;
ERROR: invalid input syntax for type vector: "[1,2,3]9"
LINE 1: SELECT '[1,2,3]9'::vector;
^
DETAIL: Junk after closing right brace.
SELECT '1,2,3'::vector;
ERROR: invalid input syntax for type vector: "1,2,3"
LINE 1: SELECT '1,2,3'::vector;
^
DETAIL: Vector contents must start with "[".
SELECT ''::vector;
ERROR: invalid input syntax for type vector: ""
LINE 1: SELECT ''::vector;
^
DETAIL: Vector contents must start with "[".
SELECT '['::vector;
ERROR: invalid input syntax for type vector: "["
LINE 1: SELECT '['::vector;
^
SELECT '[ '::vector;
ERROR: invalid input syntax for type vector: "[ "
LINE 1: SELECT '[ '::vector;
^
SELECT '[,'::vector;
ERROR: invalid input syntax for type vector: "[,"
LINE 1: SELECT '[,'::vector;
^
SELECT '[]'::vector;
ERROR: vector must have at least 1 dimension
LINE 1: SELECT '[]'::vector;
^
SELECT '[ ]'::vector;
ERROR: vector must have at least 1 dimension
LINE 1: SELECT '[ ]'::vector;
^
SELECT '[,]'::vector;
ERROR: invalid input syntax for type vector: "[,]"
LINE 1: SELECT '[,]'::vector;
^
SELECT '[1,]'::vector;
ERROR: invalid input syntax for type vector: "[1,]"
LINE 1: SELECT '[1,]'::vector;
^
SELECT '[1a]'::vector;
ERROR: invalid input syntax for type vector: "[1a]"
LINE 1: SELECT '[1a]'::vector;
^
SELECT '[1,,3]'::vector;
ERROR: invalid input syntax for type vector: "[1,,3]"
LINE 1: SELECT '[1,,3]'::vector;
^
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
---------
[1,2,3]
[4,5,6]
(2 rows)
SELECT '{"[1,2,3]"}'::vector(2)[];
ERROR: expected 2 dimensions, not 3
SELECT '[1,2,3]'::vector + '[4,5,6]';
?column?
----------
[5,7,9]
(1 row)
SELECT '[3e38]'::vector + '[3e38]';
ERROR: value out of range: overflow
SELECT '[1,2]'::vector + '[3]';
ERROR: different vector dimensions 2 and 1
SELECT '[1,2,3]'::vector - '[4,5,6]';
?column?
------------
[-3,-3,-3]
(1 row)
SELECT '[-3e38]'::vector - '[3e38]';
ERROR: value out of range: overflow
SELECT '[1,2]'::vector - '[3]';
ERROR: different vector dimensions 2 and 1
SELECT '[1,2,3]'::vector * '[4,5,6]';
?column?
-----------
[4,10,18]
(1 row)
SELECT '[1e37]'::vector * '[1e37]';
ERROR: value out of range: overflow
SELECT '[1e-37]'::vector * '[1e-37]';
ERROR: value out of range: underflow
SELECT '[1,2]'::vector * '[3]';
ERROR: different vector dimensions 2 and 1
SELECT '[1,2,3]'::vector || '[4,5]';
?column?
-------------
[1,2,3,4,5]
(1 row)
SELECT array_fill(0, ARRAY[16000])::vector || '[1]';
ERROR: vector cannot have more than 16000 dimensions
SELECT '[1,2,3]'::vector < '[1,2,3]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::vector < '[1,2]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::vector <= '[1,2,3]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::vector <= '[1,2]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::vector = '[1,2,3]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::vector = '[1,2]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::vector != '[1,2,3]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::vector != '[1,2]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::vector >= '[1,2,3]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::vector >= '[1,2]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::vector > '[1,2,3]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::vector > '[1,2]';
?column?
----------
t
(1 row)
SELECT vector_cmp('[1,2,3]', '[1,2,3]');
vector_cmp
------------
0
(1 row)
SELECT vector_cmp('[1,2,3]', '[0,0,0]');
vector_cmp
------------
1
(1 row)
SELECT vector_cmp('[0,0,0]', '[1,2,3]');
vector_cmp
------------
-1
(1 row)
SELECT vector_cmp('[1,2]', '[1,2,3]');
vector_cmp
------------
-1
(1 row)
SELECT vector_cmp('[1,2,3]', '[1,2]');
vector_cmp
------------
1
(1 row)
SELECT vector_cmp('[1,2]', '[2,3,4]');
vector_cmp
------------
-1
(1 row)
SELECT vector_cmp('[2,3]', '[1,2,3]');
vector_cmp
------------
1
(1 row)
SELECT vector_dims('[1,2,3]'::vector);
vector_dims
-------------
3
(1 row)
SELECT round(vector_norm('[1,1]')::numeric, 5);
round
---------
1.41421
(1 row)
SELECT vector_norm('[3,4]');
vector_norm
-------------
5
(1 row)
SELECT vector_norm('[0,1]');
vector_norm
-------------
1
(1 row)
SELECT vector_norm('[3e37,4e37]')::real;
vector_norm
-------------
5e+37
(1 row)
SELECT vector_norm('[0,0]');
vector_norm
-------------
0
(1 row)
SELECT vector_norm('[2]');
vector_norm
-------------
2
(1 row)
SELECT l2_distance('[0,0]'::vector, '[3,4]');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('[0,0]'::vector, '[0,1]');
l2_distance
-------------
1
(1 row)
SELECT l2_distance('[1,2]'::vector, '[3]');
ERROR: different vector dimensions 2 and 1
SELECT l2_distance('[3e38]'::vector, '[-3e38]');
l2_distance
-------------
Infinity
(1 row)
SELECT l2_distance('[1,1,1,1,1,1,1,1,1]'::vector, '[1,1,1,1,1,1,1,4,5]');
l2_distance
-------------
5
(1 row)
SELECT '[0,0]'::vector <-> '[3,4]';
?column?
----------
5
(1 row)
SELECT inner_product('[1,2]'::vector, '[3,4]');
inner_product
---------------
11
(1 row)
SELECT inner_product('[1,2]'::vector, '[3]');
ERROR: different vector dimensions 2 and 1
SELECT inner_product('[3e38]'::vector, '[3e38]');
inner_product
---------------
Infinity
(1 row)
SELECT inner_product('[1,1,1,1,1,1,1,1,1]'::vector, '[1,2,3,4,5,6,7,8,9]');
inner_product
---------------
45
(1 row)
SELECT '[1,2]'::vector <#> '[3,4]';
?column?
----------
-11
(1 row)
SELECT cosine_distance('[1,2]'::vector, '[2,4]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,2]'::vector, '[0,0]');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('[1,1]'::vector, '[1,1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,0]'::vector, '[0,2]');
cosine_distance
-----------------
1
(1 row)
SELECT cosine_distance('[1,1]'::vector, '[-1,-1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[1,2]'::vector, '[3]');
ERROR: different vector dimensions 2 and 1
SELECT cosine_distance('[1,1]'::vector, '[1.1,1.1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,1]'::vector, '[-1.1,-1.1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[3e38]'::vector, '[3e38]');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('[1,2,3,4,5,6,7,8,9]'::vector, '[1,2,3,4,5,6,7,8,9]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,2,3,4,5,6,7,8,9]'::vector, '[-1,-2,-3,-4,-5,-6,-7,-8,-9]');
cosine_distance
-----------------
2
(1 row)
SELECT '[1,2]'::vector <=> '[2,4]';
?column?
----------
0
(1 row)
SELECT l1_distance('[0,0]'::vector, '[3,4]');
l1_distance
-------------
7
(1 row)
SELECT l1_distance('[0,0]'::vector, '[0,1]');
l1_distance
-------------
1
(1 row)
SELECT l1_distance('[1,2]'::vector, '[3]');
ERROR: different vector dimensions 2 and 1
SELECT l1_distance('[3e38]'::vector, '[-3e38]');
l1_distance
-------------
Infinity
(1 row)
SELECT l1_distance('[1,2,3,4,5,6,7,8,9]'::vector, '[1,2,3,4,5,6,7,8,9]');
l1_distance
-------------
0
(1 row)
SELECT l1_distance('[1,2,3,4,5,6,7,8,9]'::vector, '[0,3,2,5,4,7,6,9,8]');
l1_distance
-------------
9
(1 row)
SELECT '[0,0]'::vector <+> '[3,4]';
?column?
----------
7
(1 row)
SELECT l2_normalize('[3,4]'::vector);
l2_normalize
--------------
[0.6,0.8]
(1 row)
SELECT l2_normalize('[3,0]'::vector);
l2_normalize
--------------
[1,0]
(1 row)
SELECT l2_normalize('[0,0.1]'::vector);
l2_normalize
--------------
[0,1]
(1 row)
SELECT l2_normalize('[0,0]'::vector);
l2_normalize
--------------
[0,0]
(1 row)
SELECT l2_normalize('[3e38]'::vector);
l2_normalize
--------------
[1]
(1 row)
SELECT binary_quantize('[1,0,-1]'::vector);
binary_quantize
-----------------
100
(1 row)
SELECT binary_quantize('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]'::vector);
binary_quantize
-----------------
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
-----------
[1,2,3]
(1 row)
SELECT subvector('[1,2,3,4,5]'::vector, 3, 2);
subvector
-----------
[3,4]
(1 row)
SELECT subvector('[1,2,3,4,5]'::vector, -1, 3);
subvector
-----------
[1]
(1 row)
SELECT subvector('[1,2,3,4,5]'::vector, 3, 9);
subvector
-----------
[3,4,5]
(1 row)
SELECT subvector('[1,2,3,4,5]'::vector, 1, 0);
ERROR: vector must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::vector, 3, -1);
ERROR: vector must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::vector, -1, 2);
ERROR: vector must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::vector, 2147483647, 10);
ERROR: vector must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::vector, 3, 2147483647);
subvector
-----------
[3,4,5]
(1 row)
SELECT subvector('[1,2,3,4,5]'::vector, -2147483644, 2147483647);
subvector
-----------
[1,2]
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
avg
-----------
[2,3.5,5]
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
avg
-----------
[2,3.5,5]
(1 row)
SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
avg
-----
(1 row)
SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
ERROR: expected 2 dimensions, not 1
SELECT avg(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
avg
---------
[3e+38]
(1 row)
SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;
ERROR: vector cannot have more than 16000 dimensions
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
sum
----------
[4,7,10]
(1 row)
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
sum
----------
[4,7,10]
(1 row)
SELECT sum(v) FROM unnest(ARRAY[]::vector[]) v;
sum
-----
(1 row)
SELECT sum(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
ERROR: different vector dimensions 2 and 1
SELECT sum(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
ERROR: value out of range: overflow

View File

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

View File

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

View File

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

3
test/perl/TestLib.pm Normal file
View File

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

View File

@@ -1,27 +0,0 @@
SELECT hamming_distance('111', '111');
SELECT hamming_distance('111', '110');
SELECT hamming_distance('111', '100');
SELECT hamming_distance('111', '000');
SELECT hamming_distance('10101010101010101010', '01010101010101010101');
SELECT hamming_distance('101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101', '101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101');
SELECT hamming_distance('101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101', '010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010');
SELECT hamming_distance('110000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000011', '100000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000001');
SELECT hamming_distance('', '');
SELECT hamming_distance('111', '00');
SELECT hamming_distance('111', '000'::varbit(4));
SELECT hamming_distance('111', '0000'::varbit(4));
SELECT jaccard_distance('1111', '1111');
SELECT jaccard_distance('1111', '1110');
SELECT jaccard_distance('1111', '1100');
SELECT jaccard_distance('1111', '1000');
SELECT jaccard_distance('1111', '0000');
SELECT jaccard_distance('1100', '1000');
SELECT jaccard_distance('10101010101010101010', '01010101010101010101');
SELECT jaccard_distance('101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101', '101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101');
SELECT jaccard_distance('101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101', '010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010101010');
SELECT jaccard_distance('110000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000011', '100000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000001');
SELECT jaccard_distance('', '');
SELECT jaccard_distance('1111', '000');
SELECT jaccard_distance('1111', '0000'::varbit(5));
SELECT jaccard_distance('1111', '00000'::varbit(5));

View File

@@ -1,34 +1,10 @@
SET enable_seqscan = off;
-- vector
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t (val);
SELECT * FROM t WHERE val = '[1,2,3]';
SELECT * FROM t ORDER BY val;
DROP TABLE t;
-- halfvec
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t (val);
SELECT * FROM t WHERE val = '[1,2,3]';
SELECT * FROM t ORDER BY val;
DROP TABLE t;
-- sparsevec
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE INDEX ON t (val);
SELECT * FROM t WHERE val = '{1:1,2:2,3:3}/3';
SELECT * FROM t ORDER BY val;
SELECT * FROM t ORDER BY val LIMIT 1;
DROP TABLE t;

View File

@@ -3,77 +3,13 @@ SELECT ARRAY[1.0,2.0,3.0]::vector;
SELECT ARRAY[1,2,3]::float4[]::vector;
SELECT ARRAY[1,2,3]::float8[]::vector;
SELECT ARRAY[1,2,3]::numeric[]::vector;
SELECT '[1,2,3]'::vector::real[];
SELECT '{1,2,3}'::real[]::vector;
SELECT '{1,2,3}'::real[]::vector(3);
SELECT '{1,2,3}'::real[]::vector(2);
SELECT '{NULL}'::real[]::vector;
SELECT '{NaN}'::real[]::vector;
SELECT '{Infinity}'::real[]::vector;
SELECT '{-Infinity}'::real[]::vector;
SELECT '{}'::real[]::vector;
SELECT '{{1}}'::real[]::vector;
SELECT '{1,2,3}'::double precision[]::vector;
SELECT '{1,2,3}'::double precision[]::vector(3);
SELECT '{1,2,3}'::double precision[]::vector(2);
SELECT '{4e38,-4e38}'::double precision[]::vector;
SELECT '{1e-46,-1e-46}'::double precision[]::vector;
SELECT '[1,2,3]'::vector::halfvec;
SELECT '[1,2,3]'::vector::halfvec(3);
SELECT '[1,2,3]'::vector::halfvec(2);
SELECT '[65520]'::vector::halfvec;
SELECT '[1e-8]'::vector::halfvec;
SELECT '[1,2,3]'::halfvec::vector;
SELECT '[1,2,3]'::halfvec::vector(3);
SELECT '[1,2,3]'::halfvec::vector(2);
SELECT '{1,2,3}'::real[]::halfvec;
SELECT '{1,2,3}'::real[]::halfvec(3);
SELECT '{1,2,3}'::real[]::halfvec(2);
SELECT '{65520,-65520}'::real[]::halfvec;
SELECT '{1e-8,-1e-8}'::real[]::halfvec;
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec(5);
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec(4);
SELECT '{2:1.5,4:3.5}/5'::sparsevec::vector;
SELECT '{2:1.5,4:3.5}/5'::sparsevec::vector(5);
SELECT '{2:1.5,4:3.5}/5'::sparsevec::vector(4);
SELECT '{}/16001'::sparsevec::vector;
SELECT '[0,1.5,0,3.5,0]'::halfvec::sparsevec;
SELECT '[0,1.5,0,3.5,0]'::halfvec::sparsevec(5);
SELECT '[0,1.5,0,3.5,0]'::halfvec::sparsevec(4);
SELECT '{2:1.5,4:3.5}/5'::sparsevec::halfvec;
SELECT '{2:1.5,4:3.5}/5'::sparsevec::halfvec(5);
SELECT '{2:1.5,4:3.5}/5'::sparsevec::halfvec(4);
SELECT '{}/16001'::sparsevec::halfvec;
SELECT '{1:65520}/1'::sparsevec::halfvec;
SELECT '{1:1e-8}/1'::sparsevec::halfvec;
SELECT ARRAY[1,0,2,0,3,0]::sparsevec;
SELECT ARRAY[1.0,0.0,2.0,0.0,3.0,0.0]::sparsevec;
SELECT ARRAY[1,0,2,0,3,0]::float4[]::sparsevec;
SELECT ARRAY[1,0,2,0,3,0]::float8[]::sparsevec;
SELECT ARRAY[1,0,2,0,3,0]::numeric[]::sparsevec;
SELECT '{1,0,2,0,3,0}'::real[]::sparsevec;
SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(6);
SELECT '{1,0,2,0,3,0}'::real[]::sparsevec(5);
SELECT '{NULL}'::real[]::sparsevec;
SELECT '{NaN}'::real[]::sparsevec;
SELECT '{Infinity}'::real[]::sparsevec;
SELECT '{-Infinity}'::real[]::sparsevec;
SELECT '{}'::real[]::sparsevec;
SELECT '{{1}}'::real[]::sparsevec;
SELECT '[1,2,3]'::vector::real[];
SELECT array_agg(n)::vector FROM generate_series(1, 16001) n;
SELECT array_to_vector(array_agg(n), 16001, false) FROM generate_series(1, 16001) n;

View File

@@ -1,42 +1,10 @@
-- vector
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE TABLE t2 (val vector(3));
\copy t TO 'results/vector.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/vector.bin' WITH (FORMAT binary)
SELECT * FROM t2 ORDER BY val;
DROP TABLE t;
DROP TABLE t2;
-- halfvec
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE TABLE t2 (val halfvec(3));
\copy t TO 'results/halfvec.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/halfvec.bin' WITH (FORMAT binary)
SELECT * FROM t2 ORDER BY val;
DROP TABLE t;
DROP TABLE t2;
-- sparsevec
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE TABLE t2 (val sparsevec(3));
\copy t TO 'results/sparsevec.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/sparsevec.bin' WITH (FORMAT binary)
\copy t TO 'results/data.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/data.bin' WITH (FORMAT binary)
SELECT * FROM t2 ORDER BY val;

51
test/sql/functions.sql Normal file
View File

@@ -0,0 +1,51 @@
SELECT '[1,2,3]'::vector + '[4,5,6]';
SELECT '[3e38]'::vector + '[3e38]';
SELECT '[1,2,3]'::vector - '[4,5,6]';
SELECT '[-3e38]'::vector - '[3e38]';
SELECT '[1,2,3]'::vector * '[4,5,6]';
SELECT '[1e37]'::vector * '[1e37]';
SELECT '[1e-37]'::vector * '[1e-37]';
SELECT vector_dims('[1,2,3]');
SELECT round(vector_norm('[1,1]')::numeric, 5);
SELECT vector_norm('[3,4]');
SELECT vector_norm('[0,1]');
SELECT vector_norm('[3e37,4e37]')::real;
SELECT l2_distance('[0,0]', '[3,4]');
SELECT l2_distance('[0,0]', '[0,1]');
SELECT l2_distance('[1,2]', '[3]');
SELECT l2_distance('[3e38]', '[-3e38]');
SELECT inner_product('[1,2]', '[3,4]');
SELECT inner_product('[1,2]', '[3]');
SELECT inner_product('[3e38]', '[3e38]');
SELECT cosine_distance('[1,2]', '[2,4]');
SELECT cosine_distance('[1,2]', '[0,0]');
SELECT cosine_distance('[1,1]', '[1,1]');
SELECT cosine_distance('[1,0]', '[0,2]');
SELECT cosine_distance('[1,1]', '[-1,-1]');
SELECT cosine_distance('[1,2]', '[3]');
SELECT cosine_distance('[1,1]', '[1.1,1.1]');
SELECT cosine_distance('[1,1]', '[-1.1,-1.1]');
SELECT cosine_distance('[3e38]', '[3e38]');
SELECT l1_distance('[0,0]', '[3,4]');
SELECT l1_distance('[0,0]', '[0,1]');
SELECT l1_distance('[1,2]', '[3]');
SELECT l1_distance('[3e38]', '[-3e38]');
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
SELECT avg(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
SELECT avg(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;
SELECT vector_avg(array_agg(n)) FROM generate_series(1, 16002) n;
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]']) v;
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::vector, '[3,5,7]', NULL]) v;
SELECT sum(v) FROM unnest(ARRAY[]::vector[]) v;
SELECT sum(v) FROM unnest(ARRAY['[1,2]'::vector, '[3]']) v;
SELECT sum(v) FROM unnest(ARRAY['[3e38]'::vector, '[3e38]']) v;

View File

@@ -1,148 +0,0 @@
SELECT '[1,2,3]'::halfvec;
SELECT '[-1,-2,-3]'::halfvec;
SELECT '[1.,2.,3.]'::halfvec;
SELECT ' [ 1, 2 , 3 ] '::halfvec;
SELECT '[1.23456]'::halfvec;
SELECT '[hello,1]'::halfvec;
SELECT '[NaN,1]'::halfvec;
SELECT '[Infinity,1]'::halfvec;
SELECT '[-Infinity,1]'::halfvec;
SELECT '[65519,-65519]'::halfvec;
SELECT '[65520,-65520]'::halfvec;
SELECT '[1e-8,-1e-8]'::halfvec;
SELECT '[4e38,1]'::halfvec;
SELECT '[1e-46,1]'::halfvec;
SELECT '[1,2,3'::halfvec;
SELECT '[1,2,3]9'::halfvec;
SELECT '1,2,3'::halfvec;
SELECT ''::halfvec;
SELECT '['::halfvec;
SELECT '[ '::halfvec;
SELECT '[,'::halfvec;
SELECT '[]'::halfvec;
SELECT '[ ]'::halfvec;
SELECT '[,]'::halfvec;
SELECT '[1,]'::halfvec;
SELECT '[1a]'::halfvec;
SELECT '[1,,3]'::halfvec;
SELECT '[1, ,3]'::halfvec;
SELECT '[1,2,3]'::halfvec(3);
SELECT '[1,2,3]'::halfvec(2);
SELECT '[1,2,3]'::halfvec(3, 2);
SELECT '[1,2,3]'::halfvec('a');
SELECT '[1,2,3]'::halfvec(0);
SELECT '[1,2,3]'::halfvec(16001);
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::halfvec[]);
SELECT '{"[1,2,3]"}'::halfvec(2)[];
SELECT '[1,2,3]'::halfvec + '[4,5,6]';
SELECT '[65519]'::halfvec + '[65519]';
SELECT '[1,2]'::halfvec + '[3]';
SELECT '[1,2,3]'::halfvec - '[4,5,6]';
SELECT '[-65519]'::halfvec - '[65519]';
SELECT '[1,2]'::halfvec - '[3]';
SELECT '[1,2,3]'::halfvec * '[4,5,6]';
SELECT '[65519]'::halfvec * '[65519]';
SELECT '[1e-7]'::halfvec * '[1e-7]';
SELECT '[1,2]'::halfvec * '[3]';
SELECT '[1,2,3]'::halfvec || '[4,5]';
SELECT array_fill(0, ARRAY[16000])::halfvec || '[1]';
SELECT '[1,2,3]'::halfvec < '[1,2,3]';
SELECT '[1,2,3]'::halfvec < '[1,2]';
SELECT '[1,2,3]'::halfvec <= '[1,2,3]';
SELECT '[1,2,3]'::halfvec <= '[1,2]';
SELECT '[1,2,3]'::halfvec = '[1,2,3]';
SELECT '[1,2,3]'::halfvec = '[1,2]';
SELECT '[1,2,3]'::halfvec != '[1,2,3]';
SELECT '[1,2,3]'::halfvec != '[1,2]';
SELECT '[1,2,3]'::halfvec >= '[1,2,3]';
SELECT '[1,2,3]'::halfvec >= '[1,2]';
SELECT '[1,2,3]'::halfvec > '[1,2,3]';
SELECT '[1,2,3]'::halfvec > '[1,2]';
SELECT halfvec_cmp('[1,2,3]', '[1,2,3]');
SELECT halfvec_cmp('[1,2,3]', '[0,0,0]');
SELECT halfvec_cmp('[0,0,0]', '[1,2,3]');
SELECT halfvec_cmp('[1,2]', '[1,2,3]');
SELECT halfvec_cmp('[1,2,3]', '[1,2]');
SELECT halfvec_cmp('[1,2]', '[2,3,4]');
SELECT halfvec_cmp('[2,3]', '[1,2,3]');
SELECT vector_dims('[1,2,3]'::halfvec);
SELECT round(l2_norm('[1,1]'::halfvec)::numeric, 5);
SELECT l2_norm('[3,4]'::halfvec);
SELECT l2_norm('[0,1]'::halfvec);
SELECT l2_norm('[0,0]'::halfvec);
SELECT l2_norm('[2]'::halfvec);
SELECT l2_distance('[0,0]'::halfvec, '[3,4]');
SELECT l2_distance('[0,0]'::halfvec, '[0,1]');
SELECT l2_distance('[1,2]'::halfvec, '[3]');
SELECT l2_distance('[1,1,1,1,1,1,1,1,1]'::halfvec, '[1,1,1,1,1,1,1,4,5]');
SELECT '[0,0]'::halfvec <-> '[3,4]';
SELECT inner_product('[1,2]'::halfvec, '[3,4]');
SELECT inner_product('[1,2]'::halfvec, '[3]');
SELECT inner_product('[65504]'::halfvec, '[65504]');
SELECT inner_product('[1,1,1,1,1,1,1,1,1]'::halfvec, '[1,2,3,4,5,6,7,8,9]');
SELECT '[1,2]'::halfvec <#> '[3,4]';
SELECT cosine_distance('[1,2]'::halfvec, '[2,4]');
SELECT cosine_distance('[1,2]'::halfvec, '[0,0]');
SELECT cosine_distance('[1,1]'::halfvec, '[1,1]');
SELECT cosine_distance('[1,0]'::halfvec, '[0,2]');
SELECT cosine_distance('[1,1]'::halfvec, '[-1,-1]');
SELECT cosine_distance('[1,2]'::halfvec, '[3]');
SELECT cosine_distance('[1,1]'::halfvec, '[1.1,1.1]');
SELECT cosine_distance('[1,1]'::halfvec, '[-1.1,-1.1]');
SELECT cosine_distance('[1,2,3,4,5,6,7,8,9]'::halfvec, '[1,2,3,4,5,6,7,8,9]');
SELECT cosine_distance('[1,2,3,4,5,6,7,8,9]'::halfvec, '[-1,-2,-3,-4,-5,-6,-7,-8,-9]');
SELECT '[1,2]'::halfvec <=> '[2,4]';
SELECT l1_distance('[0,0]'::halfvec, '[3,4]');
SELECT l1_distance('[0,0]'::halfvec, '[0,1]');
SELECT l1_distance('[1,2]'::halfvec, '[3]');
SELECT l1_distance('[1,2,3,4,5,6,7,8,9]'::halfvec, '[1,2,3,4,5,6,7,8,9]');
SELECT l1_distance('[1,2,3,4,5,6,7,8,9]'::halfvec, '[0,3,2,5,4,7,6,9,8]');
SELECT '[0,0]'::halfvec <+> '[3,4]';
SELECT l2_normalize('[3,4]'::halfvec);
SELECT l2_normalize('[3,0]'::halfvec);
SELECT l2_normalize('[0,0.1]'::halfvec);
SELECT l2_normalize('[0,0]'::halfvec);
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);
SELECT subvector('[1,2,3,4,5]'::halfvec, -1, 3);
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 9);
SELECT subvector('[1,2,3,4,5]'::halfvec, 1, 0);
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, -1);
SELECT subvector('[1,2,3,4,5]'::halfvec, -1, 2);
SELECT subvector('[1,2,3,4,5]'::halfvec, 2147483647, 10);
SELECT subvector('[1,2,3,4,5]'::halfvec, 3, 2147483647);
SELECT subvector('[1,2,3,4,5]'::halfvec, -2147483644, 2147483647);
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::halfvec, '[3,5,7]']) v;
SELECT avg(v) FROM unnest(ARRAY['[1,2,3]'::halfvec, '[3,5,7]', NULL]) v;
SELECT avg(v) FROM unnest(ARRAY[]::halfvec[]) v;
SELECT avg(v) FROM unnest(ARRAY['[1,2]'::halfvec, '[3]']) v;
SELECT avg(v) FROM unnest(ARRAY['[65504]'::halfvec, '[65504]']) v;
SELECT halfvec_avg(array_agg(n)) FROM generate_series(1, 16002) n;
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::halfvec, '[3,5,7]']) v;
SELECT sum(v) FROM unnest(ARRAY['[1,2,3]'::halfvec, '[3,5,7]', NULL]) v;
SELECT sum(v) FROM unnest(ARRAY[]::halfvec[]) v;
SELECT sum(v) FROM unnest(ARRAY['[1,2]'::halfvec, '[3]']) v;
SELECT sum(v) FROM unnest(ARRAY['[65504]'::halfvec, '[65504]']) v;

View File

@@ -1,35 +0,0 @@
SET enable_seqscan = off;
-- hamming
CREATE TABLE t (val bit(3));
INSERT INTO t (val) VALUES (B'000'), (B'100'), (B'111'), (NULL);
CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
INSERT INTO t (val) VALUES (B'110');
SELECT * FROM t ORDER BY val <~> B'111';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
DROP TABLE t;
-- jaccard
CREATE TABLE t (val bit(4));
INSERT INTO t (val) VALUES (B'0000'), (B'1100'), (B'1111'), (NULL);
CREATE INDEX ON t USING hnsw (val bit_jaccard_ops);
INSERT INTO t (val) VALUES (B'1110');
SELECT * FROM t ORDER BY val <%> B'1111';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <%> (SELECT NULL::bit)) t2;
DROP TABLE t;
-- varbit
CREATE TABLE t (val varbit(3));
CREATE INDEX ON t USING hnsw (val bit_hamming_ops);
CREATE INDEX ON t USING hnsw ((val::bit(3)) bit_hamming_ops);
CREATE INDEX ON t USING hnsw ((val::bit(64001)) bit_hamming_ops);
DROP TABLE t;

13
test/sql/hnsw_cosine.sql Normal file
View File

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

View File

@@ -1,58 +0,0 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val halfvec_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::halfvec)) t2;
SELECT COUNT(*) FROM t;
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
DROP TABLE t;
-- inner product
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val halfvec_ip_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::halfvec)) t2;
DROP TABLE t;
-- cosine
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val halfvec_cosine_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 <=> '[0,0,0]') t2;
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::halfvec)) t2;
DROP TABLE t;
-- L1
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val halfvec_l1_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::halfvec)) t2;
DROP TABLE t;

12
test/sql/hnsw_ip.sql Normal file
View File

@@ -0,0 +1,12 @@
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_ip_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;
DROP TABLE t;

16
test/sql/hnsw_l2.sql Normal file
View File

@@ -0,0 +1,16 @@
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
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 * FROM t ORDER BY val <-> (SELECT NULL::vector);
SELECT COUNT(*) FROM t;
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
DROP TABLE t;

13
test/sql/hnsw_options.sql Normal file
View File

@@ -0,0 +1,13 @@
CREATE TABLE t (val vector(3));
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 1);
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 101);
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (ef_construction = 3);
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (ef_construction = 1001);
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 16, ef_construction = 31);
SHOW hnsw.ef_search;
SET hnsw.ef_search = 0;
SET hnsw.ef_search = 1001;
DROP TABLE t;

View File

@@ -1,68 +0,0 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_l2_ops);
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
SELECT * FROM t ORDER BY val <-> '{1:3,2:3,3:3}/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 <-> '{1:3,2:3,3:3}/3';
DROP TABLE t;
-- inner product
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_ip_ops);
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
SELECT * FROM t ORDER BY val <#> '{1:3,2:3,3:3}/3';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::sparsevec)) t2;
DROP TABLE t;
-- cosine
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_cosine_ops);
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
SELECT * FROM t ORDER BY val <=> '{1:3,2:3,3:3}/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;
-- L1
CREATE TABLE t (val sparsevec(3));
INSERT INTO t (val) VALUES ('{}/3'), ('{1:1,2:2,3:3}/3'), ('{1:1,2:1,3:1}/3'), (NULL);
CREATE INDEX ON t USING hnsw (val sparsevec_l1_ops);
INSERT INTO t (val) VALUES ('{1:1,2:2,3:4}/3');
SELECT * FROM t ORDER BY val <+> '{1:3,2:3,3:3}/3';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <+> (SELECT NULL::sparsevec)) t2;
DROP TABLE t;
-- non-zero elements
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;

View File

@@ -0,0 +1,9 @@
SET enable_seqscan = off;
CREATE UNLOGGED TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_l2_ops);
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
DROP TABLE t;

View File

@@ -1,117 +0,0 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
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 COUNT(*) FROM t;
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
DROP TABLE t;
-- inner product
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_ip_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;
DROP TABLE t;
-- cosine
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_cosine_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 <=> '[0,0,0]') t2;
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
DROP TABLE t;
-- L1
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_l1_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;
DROP TABLE t;
-- iterative
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_l2_ops);
SET hnsw.iterative_scan = strict_order;
SET hnsw.ef_search = 1;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
SET hnsw.iterative_scan = relaxed_order;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
RESET hnsw.iterative_scan;
RESET hnsw.ef_search;
DROP TABLE t;
-- unlogged
CREATE UNLOGGED TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING hnsw (val vector_l2_ops);
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
DROP TABLE t;
-- options
CREATE TABLE t (val vector(3));
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 1);
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 101);
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (ef_construction = 3);
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (ef_construction = 1001);
CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 16, ef_construction = 31);
SHOW hnsw.ef_search;
SET hnsw.ef_search = 0;
SET hnsw.ef_search = 1001;
SHOW hnsw.iterative_scan;
SET hnsw.iterative_scan = on;
SHOW hnsw.max_scan_tuples;
SET hnsw.max_scan_tuples = 0;
SHOW hnsw.scan_mem_multiplier;
SET hnsw.scan_mem_multiplier = 0;
SET hnsw.scan_mem_multiplier = 1001;
DROP TABLE t;

28
test/sql/input.sql Normal file
View File

@@ -0,0 +1,28 @@
SELECT '[1,2,3]'::vector;
SELECT '[-1,-2,-3]'::vector;
SELECT '[1.,2.,3.]'::vector;
SELECT ' [ 1, 2 , 3 ] '::vector;
SELECT '[1.23456]'::vector;
SELECT '[hello,1]'::vector;
SELECT '[NaN,1]'::vector;
SELECT '[Infinity,1]'::vector;
SELECT '[-Infinity,1]'::vector;
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 '[1,2,3'::vector;
SELECT '[1,2,3]9'::vector;
SELECT '1,2,3'::vector;
SELECT ''::vector;
SELECT '['::vector;
SELECT '[,'::vector;
SELECT '[]'::vector;
SELECT '[1,]'::vector;
SELECT '[1a]'::vector;
SELECT '[1,,3]'::vector;
SELECT '[1, ,3]'::vector;
SELECT '[1,2,3]'::vector(2);
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::vector[]);
SELECT '{"[1,2,3]"}'::vector(2)[];

View File

@@ -1,23 +0,0 @@
SET enable_seqscan = off;
-- hamming
CREATE TABLE t (val bit(3));
INSERT INTO t (val) VALUES (B'000'), (B'100'), (B'111'), (NULL);
CREATE INDEX ON t USING ivfflat (val bit_hamming_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES (B'110');
SELECT * FROM t ORDER BY val <~> B'111';
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <~> (SELECT NULL::bit)) t2;
DROP TABLE t;
-- varbit
CREATE TABLE t (val varbit(3));
CREATE INDEX ON t USING ivfflat (val bit_hamming_ops) WITH (lists = 1);
CREATE INDEX ON t USING ivfflat ((val::bit(3)) bit_hamming_ops) WITH (lists = 1);
CREATE INDEX ON t USING ivfflat ((val::bit(64001)) bit_hamming_ops) WITH (lists = 1);
CREATE INDEX ON t USING ivfflat ((val::bit(2)) bit_hamming_ops) WITH (lists = 5);
DROP TABLE t;

View File

@@ -0,0 +1,13 @@
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_cosine_ops) WITH (lists = 1);
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 <=> '[0,0,0]') t2;
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector)) t2;
DROP TABLE t;

View File

@@ -1,45 +0,0 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val halfvec_l2_ops) WITH (lists = 1);
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::halfvec)) t2;
SELECT COUNT(*) FROM t;
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
DROP TABLE t;
-- inner product
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val halfvec_ip_ops) WITH (lists = 1);
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::halfvec)) t2;
DROP TABLE t;
-- cosine
CREATE TABLE t (val halfvec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val halfvec_cosine_ops) WITH (lists = 1);
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 <=> '[0,0,0]') t2;
SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <=> (SELECT NULL::halfvec)) t2;
DROP TABLE t;

12
test/sql/ivfflat_ip.sql Normal file
View File

@@ -0,0 +1,12 @@
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_ip_ops) WITH (lists = 1);
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;
DROP TABLE t;

16
test/sql/ivfflat_l2.sql Normal file
View File

@@ -0,0 +1,16 @@
SET enable_seqscan = off;
CREATE TABLE t (val vector(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val vector_l2_ops) WITH (lists = 1);
INSERT INTO t (val) VALUES ('[1,2,4]');
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
SELECT * FROM t ORDER BY val <-> (SELECT NULL::vector);
SELECT COUNT(*) FROM t;
TRUNCATE t;
SELECT * FROM t ORDER BY val <-> '[3,3,3]';
DROP TABLE t;

Some files were not shown because too many files have changed in this diff Show More