Compare commits

..

17 Commits

Author SHA1 Message Date
Andrew Kane
e77410a1c7 Improved name [skip ci] 2024-09-24 11:48:34 -07:00
Andrew Kane
33506a645a Added comment [skip ci] 2024-09-24 11:46:44 -07:00
Andrew Kane
8e68625f1c Removed unneeded header [skip ci] 2024-09-24 01:25:48 -07:00
Andrew Kane
d1e2dc8cd4 Fixed lookup table [skip ci] 2024-09-24 01:22:49 -07:00
Andrew Kane
8d75f33eb4 Fixed code [skip ci] 2024-09-24 00:41:34 -07:00
Andrew Kane
d50125c95f Fixed rounding [skip ci] 2024-09-24 00:22:13 -07:00
Andrew Kane
d805378471 Fixed NaN [skip ci] 2024-09-24 00:07:02 -07:00
Andrew Kane
0dd7ace4de Added e2m5 version 2024-09-23 22:04:21 -07:00
Andrew Kane
261ba4a0d9 Added test for IVFFlat [skip ci] 2024-09-23 18:57:13 -07:00
Andrew Kane
4293bc439e Exclude zero vectors for cosine distance to be consistent with other types [skip ci] 2024-09-23 18:55:10 -07:00
Andrew Kane
044f0c6441 Improved precision for cosine distance [skip ci] 2024-09-23 18:53:23 -07:00
Andrew Kane
22efa8ec22 Improved precision for cosine distance [skip ci] 2024-09-23 18:50:10 -07:00
Andrew Kane
0a2803130f Added support for IVFFlat [skip ci] 2024-09-23 18:40:49 -07:00
Andrew Kane
6105a01881 Added more tests [skip ci] 2024-09-23 18:33:08 -07:00
Andrew Kane
9925b92d73 Added more minivec tests [skip ci] 2024-09-23 18:29:17 -07:00
Andrew Kane
e49bae9dc3 Fixed CI 2024-09-23 18:24:21 -07:00
Andrew Kane
eea2c44fae Added minivec type 2024-09-23 18:12:26 -07:00
36 changed files with 3280 additions and 250 deletions

View File

@@ -1,7 +1,6 @@
## 0.8.0 (unreleased) ## 0.8.0 (unreleased)
- Added casts for arrays to `sparsevec` - Added casts for arrays to `sparsevec`
- Improved cost estimation
- Reduced memory usage for HNSW index scans - Reduced memory usage for HNSW index scans
- Dropped support for Postgres 12 - Dropped support for Postgres 12

View File

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

View File

@@ -4,8 +4,8 @@ EXTVERSION = 0.7.4
MODULE_big = vector MODULE_big = vector
DATA = $(wildcard sql/*--*--*.sql) DATA = $(wildcard sql/*--*--*.sql)
DATA_built = sql/$(EXTENSION)--$(EXTVERSION).sql DATA_built = sql/$(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 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/minivec.o src/ivfvacuum.o src/sparsevec.o src/vector.o
HEADERS = src/halfvec.h src/sparsevec.h src/vector.h HEADERS = src/halfvec.h src/minivec.h src/sparsevec.h src/vector.h
TESTS = $(wildcard test/sql/*.sql) TESTS = $(wildcard test/sql/*.sql)
REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS)) REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))
@@ -66,7 +66,7 @@ dist:
git archive --format zip --prefix=$(EXTENSION)-$(EXTVERSION)/ --output dist/$(EXTENSION)-$(EXTVERSION).zip master git archive --format zip --prefix=$(EXTENSION)-$(EXTVERSION)/ --output dist/$(EXTENSION)-$(EXTVERSION).zip master
# for Docker # for Docker
PG_MAJOR ?= 17 PG_MAJOR ?= 16
.PHONY: docker .PHONY: docker

View File

@@ -2,10 +2,10 @@ EXTENSION = vector
EXTVERSION = 0.7.4 EXTVERSION = 0.7.4
DATA_built = sql\$(EXTENSION)--$(EXTVERSION).sql DATA_built = sql\$(EXTENSION)--$(EXTVERSION).sql
OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\minivec.obj src\sparsevec.obj src\vector.obj
HEADERS = src\halfvec.h src\sparsevec.h src\vector.h HEADERS = src\halfvec.h src\minivec.h src\sparsevec.h 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 = bit btree cast copy halfvec hnsw_bit hnsw_halfvec hnsw_sparsevec hnsw_vector ivfflat_bit ivfflat_halfvec ivfflat_vector minivec sparsevec vector_type
REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION) REGRESS_OPTS = --inputdir=test --load-extension=$(EXTENSION)
# For /arch flags # For /arch flags

View File

@@ -52,8 +52,6 @@ nmake /F Makefile.win
nmake /F Makefile.win install nmake /F Makefile.win install
``` ```
Note: Postgres 17 is not supported yet due to an upstream issue
See the [installation notes](#installation-notes---windows) if you run into issues See the [installation notes](#installation-notes---windows) if you run into issues
You can also install it with [Docker](#docker) or [conda-forge](#conda-forge). You can also install it with [Docker](#docker) or [conda-forge](#conda-forge).
@@ -102,8 +100,6 @@ Or add a vector column to an existing table
ALTER TABLE items ADD COLUMN embedding vector(3); ALTER TABLE items ADD COLUMN embedding vector(3);
``` ```
Also supports [half-precision](#half-precision-vectors), [binary](#binary-vectors), and [sparse](#sparse-vectors) vectors
Insert vectors Insert vectors
```sql ```sql
@@ -149,8 +145,6 @@ Supported distance functions are:
- `<#>` - (negative) inner product - `<#>` - (negative) inner product
- `<=>` - cosine distance - `<=>` - cosine distance
- `<+>` - L1 distance (added in 0.7.0) - `<+>` - L1 distance (added in 0.7.0)
- `<~>` - Hamming distance (binary vectors, added in 0.7.0)
- `<%>` - Jaccard distance (binary vectors, added in 0.7.0)
Get the nearest neighbors to a row Get the nearest neighbors to a row
@@ -940,6 +934,37 @@ Function | Description | Added
avg(halfvec) → halfvec | average | 0.7.0 avg(halfvec) → halfvec | average | 0.7.0
sum(halfvec) → halfvec | sum | 0.7.0 sum(halfvec) → halfvec | sum | 0.7.0
### Minivec Type
Each mini vector takes `dimensions + 8` bytes of storage. Each element is a E4M3 8-bit floating-point number, and all elements must be finite (no `NaN`). Mini vectors can have up to 16,000 dimensions.
### Minivec Operators
Operator | Description | Added
--- | --- | ---
\+ | element-wise addition | 0.8.0
\- | element-wise subtraction | 0.8.0
\* | element-wise multiplication | 0.8.0
\|\| | concatenate | 0.8.0
<-> | Euclidean distance | 0.8.0
<#> | negative inner product | 0.8.0
<=> | cosine distance | 0.8.0
<+> | taxicab distance | 0.8.0
### Minivec Functions
Function | Description | Added
--- | --- | ---
binary_quantize(minivec) → bit | binary quantize | 0.8.0
cosine_distance(minivec, minivec) → double precision | cosine distance | 0.8.0
inner_product(minivec, minivec) → double precision | inner product | 0.8.0
l1_distance(minivec, minivec) → double precision | taxicab distance | 0.8.0
l2_distance(minivec, minivec) → double precision | Euclidean distance | 0.8.0
l2_norm(minivec) → double precision | Euclidean norm | 0.8.0
l2_normalize(minivec) → minivec | Normalize with Euclidean norm | 0.8.0
subvector(minivec, integer, integer) → minivec | subvector | 0.8.0
vector_dims(minivec) → integer | number of dimensions | 0.8.0
### Bit Type ### Bit Type
Each bit vector takes `dimensions / 8 + 8` bytes of storage. See the [Postgres docs](https://www.postgresql.org/docs/current/datatype-bit.html) for more info. Each bit vector takes `dimensions / 8 + 8` bytes of storage. See the [Postgres docs](https://www.postgresql.org/docs/current/datatype-bit.html) for more info.
@@ -989,7 +1014,7 @@ l2_normalize(sparsevec) → sparsevec | Normalize with Euclidean norm | 0.7.0
If your machine has multiple Postgres installations, specify the path to [pg_config](https://www.postgresql.org/docs/current/app-pgconfig.html) with: If your machine has multiple Postgres installations, specify the path to [pg_config](https://www.postgresql.org/docs/current/app-pgconfig.html) with:
```sh ```sh
export PG_CONFIG=/Library/PostgreSQL/17/bin/pg_config export PG_CONFIG=/Library/PostgreSQL/16/bin/pg_config
``` ```
Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use: Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use:
@@ -1000,11 +1025,11 @@ sudo --preserve-env=PG_CONFIG make install
A few common paths on Mac are: A few common paths on Mac are:
- EDB installer - `/Library/PostgreSQL/17/bin/pg_config` - EDB installer - `/Library/PostgreSQL/16/bin/pg_config`
- Homebrew (arm64) - `/opt/homebrew/opt/postgresql@17/bin/pg_config` - Homebrew (arm64) - `/opt/homebrew/opt/postgresql@16/bin/pg_config`
- Homebrew (x86-64) - `/usr/local/opt/postgresql@17/bin/pg_config` - Homebrew (x86-64) - `/usr/local/opt/postgresql@16/bin/pg_config`
Note: Replace `17` with your Postgres server version Note: Replace `16` with your Postgres server version
### Missing Header ### Missing Header
@@ -1013,10 +1038,10 @@ If compilation fails with `fatal error: postgres.h: No such file or directory`,
For Ubuntu and Debian, use: For Ubuntu and Debian, use:
```sh ```sh
sudo apt install postgresql-server-dev-17 sudo apt install postgresql-server-dev-16
``` ```
Note: Replace `17` with your Postgres server version Note: Replace `16` with your Postgres server version
### Missing SDK ### Missing SDK
@@ -1049,17 +1074,17 @@ If installation fails with `Access is denied`, re-run the installation instructi
Get the [Docker image](https://hub.docker.com/r/pgvector/pgvector) with: Get the [Docker image](https://hub.docker.com/r/pgvector/pgvector) with:
```sh ```sh
docker pull pgvector/pgvector:pg17 docker pull pgvector/pgvector:pg16
``` ```
This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (replace `17` with your Postgres server version, and run it the same way). This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (replace `16` with your Postgres server version, and run it the same way).
You can also build the image manually: You can also build the image manually:
```sh ```sh
git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git git clone --branch v0.7.4 https://github.com/pgvector/pgvector.git
cd pgvector cd pgvector
docker build --pull --build-arg PG_MAJOR=17 -t myuser/pgvector . docker build --pull --build-arg PG_MAJOR=16 -t myuser/pgvector .
``` ```
### Homebrew ### Homebrew
@@ -1070,7 +1095,7 @@ With Homebrew Postgres, you can use:
brew install pgvector brew install pgvector
``` ```
Note: This only adds it to the `postgresql@17` and `postgresql@14` formulas Note: This only adds it to the `postgresql@14` formula
### PGXN ### PGXN
@@ -1085,22 +1110,22 @@ pgxn install vector
Debian and Ubuntu packages are available from the [PostgreSQL APT Repository](https://wiki.postgresql.org/wiki/Apt). Follow the [setup instructions](https://wiki.postgresql.org/wiki/Apt#Quickstart) and run: Debian and Ubuntu packages are available from the [PostgreSQL APT Repository](https://wiki.postgresql.org/wiki/Apt). Follow the [setup instructions](https://wiki.postgresql.org/wiki/Apt#Quickstart) and run:
```sh ```sh
sudo apt install postgresql-17-pgvector sudo apt install postgresql-16-pgvector
``` ```
Note: Replace `17` with your Postgres server version Note: Replace `16` with your Postgres server version
### Yum ### Yum
RPM packages are available from the [PostgreSQL Yum Repository](https://yum.postgresql.org/). Follow the [setup instructions](https://www.postgresql.org/download/linux/redhat/) for your distribution and run: RPM packages are available from the [PostgreSQL Yum Repository](https://yum.postgresql.org/). Follow the [setup instructions](https://www.postgresql.org/download/linux/redhat/) for your distribution and run:
```sh ```sh
sudo yum install pgvector_17 sudo yum install pgvector_16
# or # or
sudo dnf install pgvector_17 sudo dnf install pgvector_16
``` ```
Note: Replace `17` with your Postgres server version Note: Replace `16` with your Postgres server version
### pkg ### pkg

View File

@@ -1,6 +1,8 @@
-- complain if script is sourced in psql, rather than via CREATE EXTENSION -- 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 \echo Use "ALTER EXTENSION vector UPDATE TO '0.8.0'" to load this file. \quit
-- TODO minivec functions
CREATE FUNCTION array_to_sparsevec(integer[], integer, boolean) RETURNS sparsevec CREATE FUNCTION array_to_sparsevec(integer[], integer, boolean) RETURNS sparsevec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE; AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;

View File

@@ -266,12 +266,18 @@ COMMENT ON ACCESS METHOD hnsw IS 'hnsw index access method';
CREATE FUNCTION ivfflat_halfvec_support(internal) RETURNS internal CREATE FUNCTION ivfflat_halfvec_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C; AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION ivfflat_minivec_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION ivfflat_bit_support(internal) RETURNS internal CREATE FUNCTION ivfflat_bit_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C; AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION hnsw_halfvec_support(internal) RETURNS internal CREATE FUNCTION hnsw_halfvec_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C; AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION hnsw_minivec_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C;
CREATE FUNCTION hnsw_bit_support(internal) RETURNS internal CREATE FUNCTION hnsw_bit_support(internal) RETURNS internal
AS 'MODULE_PATHNAME' LANGUAGE C; AS 'MODULE_PATHNAME' LANGUAGE C;
@@ -647,6 +653,295 @@ CREATE OPERATOR CLASS halfvec_l1_ops
FUNCTION 1 l1_distance(halfvec, halfvec), FUNCTION 1 l1_distance(halfvec, halfvec),
FUNCTION 3 hnsw_halfvec_support(internal); FUNCTION 3 hnsw_halfvec_support(internal);
-- minivec type
CREATE TYPE minivec;
CREATE FUNCTION minivec_in(cstring, oid, integer) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_out(minivec) RETURNS cstring
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_typmod_in(cstring[]) RETURNS integer
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_recv(internal, oid, integer) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_send(minivec) RETURNS bytea
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE TYPE minivec (
INPUT = minivec_in,
OUTPUT = minivec_out,
TYPMOD_IN = minivec_typmod_in,
RECEIVE = minivec_recv,
SEND = minivec_send,
STORAGE = external
);
-- minivec functions
CREATE FUNCTION l2_distance(minivec, minivec) RETURNS float8
AS 'MODULE_PATHNAME', 'minivec_l2_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION inner_product(minivec, minivec) RETURNS float8
AS 'MODULE_PATHNAME', 'minivec_inner_product' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION cosine_distance(minivec, minivec) RETURNS float8
AS 'MODULE_PATHNAME', 'minivec_cosine_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l1_distance(minivec, minivec) RETURNS float8
AS 'MODULE_PATHNAME', 'minivec_l1_distance' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_dims(minivec) RETURNS integer
AS 'MODULE_PATHNAME', 'minivec_vector_dims' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_norm(minivec) RETURNS float8
AS 'MODULE_PATHNAME', 'minivec_l2_norm' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION l2_normalize(minivec) RETURNS minivec
AS 'MODULE_PATHNAME', 'minivec_l2_normalize' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION binary_quantize(minivec) RETURNS bit
AS 'MODULE_PATHNAME', 'minivec_binary_quantize' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION subvector(minivec, int, int) RETURNS minivec
AS 'MODULE_PATHNAME', 'minivec_subvector' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- minivec private functions
CREATE FUNCTION minivec_add(minivec, minivec) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_sub(minivec, minivec) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_mul(minivec, minivec) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_concat(minivec, minivec) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_lt(minivec, minivec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_le(minivec, minivec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_eq(minivec, minivec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_ne(minivec, minivec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_ge(minivec, minivec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_gt(minivec, minivec) RETURNS bool
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_cmp(minivec, minivec) RETURNS int4
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_l2_squared_distance(minivec, minivec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_negative_inner_product(minivec, minivec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_spherical_distance(minivec, minivec) RETURNS float8
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- minivec cast functions
CREATE FUNCTION minivec(minivec, integer, boolean) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_to_vector(minivec, integer, boolean) RETURNS vector
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION vector_to_minivec(vector, integer, boolean) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_minivec(integer[], integer, boolean) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_minivec(real[], integer, boolean) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_minivec(double precision[], integer, boolean) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION array_to_minivec(numeric[], integer, boolean) RETURNS minivec
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
CREATE FUNCTION minivec_to_float4(minivec, integer, boolean) RETURNS real[]
AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-- minivec casts
CREATE CAST (minivec AS minivec)
WITH FUNCTION minivec(minivec, integer, boolean) AS IMPLICIT;
CREATE CAST (minivec AS vector)
WITH FUNCTION minivec_to_vector(minivec, integer, boolean) AS ASSIGNMENT;
CREATE CAST (vector AS minivec)
WITH FUNCTION vector_to_minivec(vector, integer, boolean) AS IMPLICIT;
CREATE CAST (minivec AS real[])
WITH FUNCTION minivec_to_float4(minivec, integer, boolean) AS ASSIGNMENT;
CREATE CAST (integer[] AS minivec)
WITH FUNCTION array_to_minivec(integer[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (real[] AS minivec)
WITH FUNCTION array_to_minivec(real[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (double precision[] AS minivec)
WITH FUNCTION array_to_minivec(double precision[], integer, boolean) AS ASSIGNMENT;
CREATE CAST (numeric[] AS minivec)
WITH FUNCTION array_to_minivec(numeric[], integer, boolean) AS ASSIGNMENT;
-- minivec operators
CREATE OPERATOR <-> (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = l2_distance,
COMMUTATOR = '<->'
);
CREATE OPERATOR <#> (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_negative_inner_product,
COMMUTATOR = '<#>'
);
CREATE OPERATOR <=> (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = cosine_distance,
COMMUTATOR = '<=>'
);
CREATE OPERATOR <+> (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = l1_distance,
COMMUTATOR = '<+>'
);
CREATE OPERATOR + (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_add,
COMMUTATOR = +
);
CREATE OPERATOR - (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_sub
);
CREATE OPERATOR * (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_mul,
COMMUTATOR = *
);
CREATE OPERATOR || (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_concat
);
CREATE OPERATOR < (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_lt,
COMMUTATOR = > , NEGATOR = >= ,
RESTRICT = scalarltsel, JOIN = scalarltjoinsel
);
CREATE OPERATOR <= (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_le,
COMMUTATOR = >= , NEGATOR = > ,
RESTRICT = scalarlesel, JOIN = scalarlejoinsel
);
CREATE OPERATOR = (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_eq,
COMMUTATOR = = , NEGATOR = <> ,
RESTRICT = eqsel, JOIN = eqjoinsel
);
CREATE OPERATOR <> (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_ne,
COMMUTATOR = <> , NEGATOR = = ,
RESTRICT = eqsel, JOIN = eqjoinsel
);
CREATE OPERATOR >= (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_ge,
COMMUTATOR = <= , NEGATOR = < ,
RESTRICT = scalargesel, JOIN = scalargejoinsel
);
CREATE OPERATOR > (
LEFTARG = minivec, RIGHTARG = minivec, PROCEDURE = minivec_gt,
COMMUTATOR = < , NEGATOR = <= ,
RESTRICT = scalargtsel, JOIN = scalargtjoinsel
);
-- minivec op classes
CREATE OPERATOR CLASS minivec_ops
DEFAULT FOR TYPE minivec USING btree AS
OPERATOR 1 < ,
OPERATOR 2 <= ,
OPERATOR 3 = ,
OPERATOR 4 >= ,
OPERATOR 5 > ,
FUNCTION 1 minivec_cmp(minivec, minivec);
CREATE OPERATOR CLASS minivec_l2_ops
FOR TYPE minivec USING ivfflat AS
OPERATOR 1 <-> (minivec, minivec) FOR ORDER BY float_ops,
FUNCTION 1 minivec_l2_squared_distance(minivec, minivec),
FUNCTION 3 l2_distance(minivec, minivec),
FUNCTION 5 ivfflat_minivec_support(internal);
CREATE OPERATOR CLASS minivec_ip_ops
FOR TYPE minivec USING ivfflat AS
OPERATOR 1 <#> (minivec, minivec) FOR ORDER BY float_ops,
FUNCTION 1 minivec_negative_inner_product(minivec, minivec),
FUNCTION 3 minivec_spherical_distance(minivec, minivec),
FUNCTION 4 l2_norm(minivec),
FUNCTION 5 ivfflat_minivec_support(internal);
CREATE OPERATOR CLASS minivec_cosine_ops
FOR TYPE minivec USING ivfflat AS
OPERATOR 1 <=> (minivec, minivec) FOR ORDER BY float_ops,
FUNCTION 1 cosine_distance(minivec, minivec),
FUNCTION 2 l2_norm(minivec),
FUNCTION 3 minivec_spherical_distance(minivec, minivec),
FUNCTION 4 l2_norm(minivec),
FUNCTION 5 ivfflat_minivec_support(internal);
CREATE OPERATOR CLASS minivec_l2_ops
FOR TYPE minivec USING hnsw AS
OPERATOR 1 <-> (minivec, minivec) FOR ORDER BY float_ops,
FUNCTION 1 minivec_l2_squared_distance(minivec, minivec),
FUNCTION 3 hnsw_minivec_support(internal);
CREATE OPERATOR CLASS minivec_ip_ops
FOR TYPE minivec USING hnsw AS
OPERATOR 1 <#> (minivec, minivec) FOR ORDER BY float_ops,
FUNCTION 1 minivec_negative_inner_product(minivec, minivec),
FUNCTION 3 hnsw_minivec_support(internal);
CREATE OPERATOR CLASS minivec_cosine_ops
FOR TYPE minivec USING hnsw AS
OPERATOR 1 <=> (minivec, minivec) FOR ORDER BY float_ops,
FUNCTION 1 cosine_distance(minivec, minivec),
FUNCTION 2 l2_norm(minivec),
FUNCTION 3 hnsw_minivec_support(internal);
CREATE OPERATOR CLASS minivec_l1_ops
FOR TYPE minivec USING hnsw AS
OPERATOR 1 <+> (minivec, minivec) FOR ORDER BY float_ops,
FUNCTION 1 l1_distance(minivec, minivec),
FUNCTION 3 hnsw_minivec_support(internal);
-- bit functions -- bit functions
CREATE FUNCTION hamming_distance(bit, bit) RETURNS float8 CREATE FUNCTION hamming_distance(bit, bit) RETURNS float8

View File

@@ -12,7 +12,6 @@
#include "utils/float.h" #include "utils/float.h"
#include "utils/guc.h" #include "utils/guc.h"
#include "utils/selfuncs.h" #include "utils/selfuncs.h"
#include "utils/spccache.h"
#if PG_VERSION_NUM < 150000 #if PG_VERSION_NUM < 150000
#define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x) #define MarkGUCPrefixReserved(x) EmitWarningsOnPlaceholders(x)
@@ -100,9 +99,7 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
{ {
GenericCosts costs; GenericCosts costs;
int m; int m;
double ratio; int entryLevel;
double startupPages;
double spc_seq_page_cost;
Relation index; Relation index;
/* Never use index without order */ /* Never use index without order */
@@ -118,71 +115,21 @@ hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
MemSet(&costs, 0, sizeof(costs)); MemSet(&costs, 0, sizeof(costs));
genericcostestimate(root, path, loop_count, &costs);
index = index_open(path->indexinfo->indexoid, NoLock); index = index_open(path->indexinfo->indexoid, NoLock);
HnswGetMetaPageInfo(index, &m, NULL); HnswGetMetaPageInfo(index, &m, NULL);
index_close(index, NoLock); index_close(index, NoLock);
/* /* Approximate entry level */
* HNSW cost estimation follows a formula that accounts for the total entryLevel = (int) -log(1.0 / path->indexinfo->tuples) * HnswGetMl(m);
* 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)));
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) genericcostestimate(root, path, loop_count, &costs);
ratio = 1;
}
else
ratio = 1;
get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost); /* Use total cost since most work happens before first tuple is returned */
*indexStartupCost = costs.indexTotalCost;
/* Startup cost is cost before returning the first row */
costs.indexStartupCost = costs.indexTotalCost * ratio;
/* Adjust cost if needed since TOAST not included in seq scan cost */
startupPages = costs.numIndexPages * ratio;
if (startupPages > path->indexinfo->rel->pages && ratio < 0.5)
{
/* Change all page cost from random to sequential */
costs.indexStartupCost -= startupPages * (costs.spc_random_page_cost - spc_seq_page_cost);
/* Remove cost of extra pages */
costs.indexStartupCost -= (startupPages - path->indexinfo->rel->pages) * spc_seq_page_cost;
}
*indexStartupCost = costs.indexStartupCost;
*indexTotalCost = costs.indexTotalCost; *indexTotalCost = costs.indexTotalCost;
*indexSelectivity = costs.indexSelectivity; *indexSelectivity = costs.indexSelectivity;
*indexCorrelation = costs.indexCorrelation; *indexCorrelation = costs.indexCorrelation;

View File

@@ -159,6 +159,9 @@ HnswOptionalProcInfo(Relation index, uint16 procnum)
Datum Datum
HnswNormValue(const HnswTypeInfo * typeInfo, Oid collation, Datum value) HnswNormValue(const HnswTypeInfo * typeInfo, Oid collation, Datum value)
{ {
if (!typeInfo->normalize)
return value;
return DirectFunctionCall1Coll(typeInfo->normalize, collation, value); return DirectFunctionCall1Coll(typeInfo->normalize, collation, value);
} }
@@ -759,16 +762,9 @@ HnswLoadUnvisitedFromDisk(HnswElement element, HnswUnvisited * unvisited, int *u
page = BufferGetPage(buf); page = BufferGetPage(buf);
ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, element->neighborOffno)); ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, element->neighborOffno));
start = (element->level - lc) * m;
/* Ensure expected neighbors */
if (ntup->count != (element->level + 2) * m)
{
UnlockReleaseBuffer(buf);
return;
}
/* Copy to minimize lock time */ /* Copy to minimize lock time */
start = (element->level - lc) * m;
memcpy(&indextids, ntup->indextids + start, lm * sizeof(ItemPointerData)); memcpy(&indextids, ntup->indextids + start, lm * sizeof(ItemPointerData));
UnlockReleaseBuffer(buf); UnlockReleaseBuffer(buf);
@@ -1334,6 +1330,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
PGDLLEXPORT Datum l2_normalize(PG_FUNCTION_ARGS); PGDLLEXPORT Datum l2_normalize(PG_FUNCTION_ARGS);
PGDLLEXPORT Datum halfvec_l2_normalize(PG_FUNCTION_ARGS); PGDLLEXPORT Datum halfvec_l2_normalize(PG_FUNCTION_ARGS);
PGDLLEXPORT Datum minivec_l2_normalize(PG_FUNCTION_ARGS);
PGDLLEXPORT Datum sparsevec_l2_normalize(PG_FUNCTION_ARGS); PGDLLEXPORT Datum sparsevec_l2_normalize(PG_FUNCTION_ARGS);
static void static void
@@ -1382,6 +1379,20 @@ hnsw_halfvec_support(PG_FUNCTION_ARGS)
PG_RETURN_POINTER(&typeInfo); PG_RETURN_POINTER(&typeInfo);
}; };
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnsw_minivec_support);
Datum
hnsw_minivec_support(PG_FUNCTION_ARGS)
{
static const HnswTypeInfo typeInfo = {
.maxDimensions = HNSW_MAX_DIM * 4,
/* Do not normalize to maximize precision */
.normalize = NULL,
.checkValue = NULL
};
PG_RETURN_POINTER(&typeInfo);
};
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnsw_bit_support); FUNCTION_PREFIX PG_FUNCTION_INFO_V1(hnsw_bit_support);
Datum Datum
hnsw_bit_support(PG_FUNCTION_ARGS) hnsw_bit_support(PG_FUNCTION_ARGS)

View File

@@ -69,8 +69,6 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
GenericCosts costs; GenericCosts costs;
int lists; int lists;
double ratio; double ratio;
double sequentialRatio = 0.5;
double startupPages;
double spc_seq_page_cost; double spc_seq_page_cost;
Relation index; Relation index;
@@ -87,8 +85,6 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
MemSet(&costs, 0, sizeof(costs)); MemSet(&costs, 0, sizeof(costs));
genericcostestimate(root, path, loop_count, &costs);
index = index_open(path->indexinfo->indexoid, NoLock); index = index_open(path->indexinfo->indexoid, NoLock);
IvfflatGetMetaPageInfo(index, &lists, NULL); IvfflatGetMetaPageInfo(index, &lists, NULL);
index_close(index, NoLock); index_close(index, NoLock);
@@ -98,26 +94,41 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
if (ratio > 1.0) if (ratio > 1.0)
ratio = 1.0; ratio = 1.0;
/*
* This gives us the subset of tuples to visit. This value is passed into
* the generic cost estimator to determine the number of pages to visit
* during the index scan.
*/
costs.numIndexTuples = path->indexinfo->tuples * ratio;
genericcostestimate(root, path, loop_count, &costs);
get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost); 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 */ /* Adjust cost if needed since TOAST not included in seq scan cost */
startupPages = costs.numIndexPages * ratio; if (costs.numIndexPages > path->indexinfo->rel->pages && ratio < 0.5)
if (startupPages > path->indexinfo->rel->pages && ratio < 0.5)
{ {
/* Change rest of page cost from random to sequential */ /* Change all page cost from random to sequential */
costs.indexStartupCost -= (1 - sequentialRatio) * startupPages * (costs.spc_random_page_cost - spc_seq_page_cost); costs.indexTotalCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
/* Remove cost of extra pages */ /* 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; *indexTotalCost = costs.indexTotalCost;
*indexSelectivity = costs.indexSelectivity; *indexSelectivity = costs.indexSelectivity;
*indexCorrelation = costs.indexCorrelation; *indexCorrelation = costs.indexCorrelation;

View File

@@ -7,6 +7,7 @@
#include "halfutils.h" #include "halfutils.h"
#include "halfvec.h" #include "halfvec.h"
#include "ivfflat.h" #include "ivfflat.h"
#include "minivec.h"
#include "storage/bufmgr.h" #include "storage/bufmgr.h"
/* /*
@@ -70,6 +71,9 @@ IvfflatOptionalProcInfo(Relation index, uint16 procnum)
Datum Datum
IvfflatNormValue(const IvfflatTypeInfo * typeInfo, Oid collation, Datum value) IvfflatNormValue(const IvfflatTypeInfo * typeInfo, Oid collation, Datum value)
{ {
if (!typeInfo->normalize)
return value;
return DirectFunctionCall1Coll(typeInfo->normalize, collation, value); return DirectFunctionCall1Coll(typeInfo->normalize, collation, value);
} }
@@ -231,6 +235,7 @@ IvfflatUpdateList(Relation index, ListInfo listInfo,
PGDLLEXPORT Datum l2_normalize(PG_FUNCTION_ARGS); PGDLLEXPORT Datum l2_normalize(PG_FUNCTION_ARGS);
PGDLLEXPORT Datum halfvec_l2_normalize(PG_FUNCTION_ARGS); PGDLLEXPORT Datum halfvec_l2_normalize(PG_FUNCTION_ARGS);
PGDLLEXPORT Datum minivec_l2_normalize(PG_FUNCTION_ARGS);
PGDLLEXPORT Datum sparsevec_l2_normalize(PG_FUNCTION_ARGS); PGDLLEXPORT Datum sparsevec_l2_normalize(PG_FUNCTION_ARGS);
static Size static Size
@@ -245,6 +250,12 @@ HalfvecItemSize(int dimensions)
return HALFVEC_SIZE(dimensions); return HALFVEC_SIZE(dimensions);
} }
static Size
MinivecItemSize(int dimensions)
{
return MINIVEC_SIZE(dimensions);
}
static Size static Size
BitItemSize(int dimensions) BitItemSize(int dimensions)
{ {
@@ -275,6 +286,18 @@ HalfvecUpdateCenter(Pointer v, int dimensions, float *x)
vec->x[k] = Float4ToHalfUnchecked(x[k]); vec->x[k] = Float4ToHalfUnchecked(x[k]);
} }
static void
MinivecUpdateCenter(Pointer v, int dimensions, float *x)
{
MiniVector *vec = (MiniVector *) v;
SET_VARSIZE(vec, MINIVEC_SIZE(dimensions));
vec->dim = dimensions;
for (int k = 0; k < dimensions; k++)
vec->x[k] = Float4ToFp8Unchecked(x[k]);
}
static void static void
BitUpdateCenter(Pointer v, int dimensions, float *x) BitUpdateCenter(Pointer v, int dimensions, float *x)
{ {
@@ -309,6 +332,15 @@ HalfvecSumCenter(Pointer v, float *x)
x[k] += HalfToFloat4(vec->x[k]); x[k] += HalfToFloat4(vec->x[k]);
} }
static void
MinivecSumCenter(Pointer v, float *x)
{
MiniVector *vec = (MiniVector *) v;
for (int k = 0; k < vec->dim; k++)
x[k] += Fp8ToFloat4(vec->x[k]);
}
static void static void
BitSumCenter(Pointer v, float *x) BitSumCenter(Pointer v, float *x)
{ {
@@ -357,6 +389,22 @@ ivfflat_halfvec_support(PG_FUNCTION_ARGS)
PG_RETURN_POINTER(&typeInfo); PG_RETURN_POINTER(&typeInfo);
}; };
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflat_minivec_support);
Datum
ivfflat_minivec_support(PG_FUNCTION_ARGS)
{
static const IvfflatTypeInfo typeInfo = {
.maxDimensions = IVFFLAT_MAX_DIM * 4,
/* Do not normalize to maximize precision */
.normalize = NULL,
.itemSize = MinivecItemSize,
.updateCenter = MinivecUpdateCenter,
.sumCenter = MinivecSumCenter
};
PG_RETURN_POINTER(&typeInfo);
};
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflat_bit_support); FUNCTION_PREFIX PG_FUNCTION_INFO_V1(ivfflat_bit_support);
Datum Datum
ivfflat_bit_support(PG_FUNCTION_ARGS) ivfflat_bit_support(PG_FUNCTION_ARGS)

1062
src/minivec.c Normal file

File diff suppressed because it is too large Load Diff

177
src/minivec.h Normal file
View File

@@ -0,0 +1,177 @@
#ifndef MINIVEC_H
#define MINIVEC_H
#define MINIVEC_MAX_DIM 16000
#define fp8 uint8
#define MINIVEC_SIZE(_dim) (offsetof(MiniVector, x) + sizeof(fp8)*(_dim))
#define DatumGetMiniVector(x) ((MiniVector *) PG_DETOAST_DATUM(x))
#define PG_GETARG_MINIVEC_P(x) DatumGetMiniVector(PG_GETARG_DATUM(x))
#define PG_RETURN_MINIVEC_P(x) PG_RETURN_POINTER(x)
typedef struct MiniVector
{
int32 vl_len_; /* varlena header (do not touch directly!) */
int16 dim; /* number of dimensions */
int16 unused; /* reserved for future use, always zero */
fp8 x[FLEXIBLE_ARRAY_MEMBER];
} MiniVector;
MiniVector *InitMiniVector(int dim);
/*
* Check if fp8 is NaN
*/
static inline bool
Fp8IsNan(fp8 num)
{
return (num & 0x7C) == 0x7C && (num & 0x7F) != 0x7C;
}
/*
* Check if fp8 is infinite
*/
static inline bool
Fp8IsInf(fp8 num)
{
return (num & 0x7F) == 0x7C;
}
/*
* Check if fp8 is zero
*/
static inline bool
Fp8IsZero(fp8 num)
{
return num == 0;
}
/*
* Convert a fp8 to a float4
*/
static inline float
Fp8ToFloat4(fp8 num)
{
/* Lookup table for non-sign bits */
/* Uses uint32 for correctness */
uint32 lookup[128] = {0, 931135488, 939524096, 943718400, 947912704, 950009856, 952107008, 954204160, 956301312, 958398464, 960495616, 962592768, 964689920, 966787072, 968884224, 970981376, 973078528, 975175680, 977272832, 979369984, 981467136, 983564288, 985661440, 987758592, 989855744, 991952896, 994050048, 996147200, 998244352, 1000341504, 1002438656, 1004535808, 1006632960, 1008730112, 1010827264, 1012924416, 1015021568, 1017118720, 1019215872, 1021313024, 1023410176, 1025507328, 1027604480, 1029701632, 1031798784, 1033895936, 1035993088, 1038090240, 1040187392, 1042284544, 1044381696, 1046478848, 1048576000, 1050673152, 1052770304, 1054867456, 1056964608, 1059061760, 1061158912, 1063256064, 1065353216, 1067450368, 1069547520, 1071644672, 1073741824, 1075838976, 1077936128, 1080033280, 1082130432, 1084227584, 1086324736, 1088421888, 1090519040, 1092616192, 1094713344, 1096810496, 1098907648, 1101004800, 1103101952, 1105199104, 1107296256, 1109393408, 1111490560, 1113587712, 1115684864, 1117782016, 1119879168, 1121976320, 1124073472, 1126170624, 1128267776, 1130364928, 1132462080, 1134559232, 1136656384, 1138753536, 1140850688, 1142947840, 1145044992, 1147142144, 1149239296, 1151336448, 1153433600, 1155530752, 1157627904, 1159725056, 1161822208, 1163919360, 1166016512, 1168113664, 1170210816, 1172307968, 1174405120, 1176502272, 1178599424, 1180696576, 1182793728, 1184890880, 1186988032, 1189085184, 1191182336, 1193279488, 1195376640, 1197473792, 2139095040, 2145386496, 2143289344, 2145386496};
union
{
float f;
uint32 i;
} swap;
swap.i = lookup[num & 0x7F];
return (num & 0x80) == 0x80 ? -swap.f : swap.f;
}
/*
* Convert a float4 to a fp8
*/
static inline fp8
Float4ToFp8Unchecked(float num)
{
union
{
float f;
uint32 i;
} swap;
uint32 bin;
int exponent;
int mantissa;
uint8 result;
swap.f = num;
bin = swap.i;
exponent = (bin & 0x7F800000) >> 23;
mantissa = bin & 0x007FFFFF;
/* Sign */
result = (bin & 0x80000000) >> 24;
if (isinf(num))
{
/* Infinite */
result |= 0x7C;
}
else if (isnan(num))
{
/* NaN */
result |= 0x7F;
}
else if (exponent > 98)
{
int m;
int gr;
int s;
exponent -= 127;
s = mantissa & 0x000FFFFF;
/* Subnormal */
if (exponent < -14)
{
int diff = -exponent - 14;
mantissa >>= diff;
mantissa += 1 << (23 - diff);
s |= mantissa & 0x000FFFFF;
}
m = mantissa >> 21;
/* Round */
gr = (mantissa >> 20) % 4;
if (gr == 3 || (gr == 1 && s != 0))
m += 1;
if (m == 4)
{
m = 0;
exponent += 1;
}
if (exponent > 15)
{
/* Infinite */
result |= 0x7C;
}
else
{
if (exponent >= -14)
result |= (exponent + 15) << 2;
result |= m;
}
}
return result;
}
/*
* Convert a float4 to a fp8
*/
static inline fp8
Float4ToFp8(float num)
{
fp8 result = Float4ToFp8Unchecked(num);
if (unlikely(Fp8IsInf(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 minivec", buf)));
}
return result;
}
#endif

View File

@@ -13,6 +13,7 @@
#include "ivfflat.h" #include "ivfflat.h"
#include "lib/stringinfo.h" #include "lib/stringinfo.h"
#include "libpq/pqformat.h" #include "libpq/pqformat.h"
#include "minivec.h"
#include "port.h" /* for strtof() */ #include "port.h" /* for strtof() */
#include "sparsevec.h" #include "sparsevec.h"
#include "utils/array.h" #include "utils/array.h"
@@ -542,6 +543,28 @@ halfvec_to_vector(PG_FUNCTION_ARGS)
PG_RETURN_POINTER(result); PG_RETURN_POINTER(result);
} }
/*
* Convert fp8 vector to vector
*/
FUNCTION_PREFIX PG_FUNCTION_INFO_V1(minivec_to_vector);
Datum
minivec_to_vector(PG_FUNCTION_ARGS)
{
MiniVector *vec = PG_GETARG_MINIVEC_P(0);
int32 typmod = PG_GETARG_INT32(1);
Vector *result;
CheckDim(vec->dim);
CheckExpectedDim(typmod, vec->dim);
result = InitVector(vec->dim);
for (int i = 0; i < vec->dim; i++)
result->x[i] = Fp8ToFloat4(vec->x[i]);
PG_RETURN_POINTER(result);
}
VECTOR_TARGET_CLONES static float VECTOR_TARGET_CLONES static float
VectorL2SquaredDistance(int dim, float *ax, float *bx) VectorL2SquaredDistance(int dim, float *ax, float *bx)
{ {

View File

@@ -38,6 +38,26 @@ SELECT * FROM t ORDER BY val;
(4 rows) (4 rows)
DROP TABLE t;
-- minivec
CREATE TABLE t (val minivec(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; DROP TABLE t;
-- sparsevec -- sparsevec
CREATE TABLE t (val sparsevec(3)); CREATE TABLE t (val sparsevec(3));

View File

@@ -140,6 +140,64 @@ SELECT '{1e-8,-1e-8}'::real[]::halfvec;
[0,-0] [0,-0]
(1 row) (1 row)
SELECT '[1,2,3]'::vector::minivec;
minivec
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::vector::minivec(3);
minivec
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::vector::minivec(2);
ERROR: expected 2 dimensions, not 3
SELECT '[61440]'::vector::minivec;
ERROR: "61440" is out of range for type minivec
SELECT '[1e-8]'::vector::minivec;
minivec
---------
[0]
(1 row)
SELECT '[1,2,3]'::minivec::vector;
vector
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::minivec::vector(3);
vector
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::minivec::vector(2);
ERROR: expected 2 dimensions, not 3
SELECT '{1,2,3}'::real[]::minivec;
minivec
---------
[1,2,3]
(1 row)
SELECT '{1,2,3}'::real[]::minivec(3);
minivec
---------
[1,2,3]
(1 row)
SELECT '{1,2,3}'::real[]::minivec(2);
ERROR: expected 2 dimensions, not 3
SELECT '{61440,-61440}'::real[]::minivec;
ERROR: "61440" is out of range for type minivec
SELECT '{1e-8,-1e-8}'::real[]::minivec;
minivec
---------
[0,-0]
(1 row)
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec; SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec;
sparsevec sparsevec
----------------- -----------------

View File

@@ -30,6 +30,23 @@ SELECT * FROM t2 ORDER BY val;
(4 rows) (4 rows)
DROP TABLE t;
DROP TABLE t2;
-- minivec
CREATE TABLE t (val minivec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE TABLE t2 (val minivec(3));
\copy t TO 'results/minivec.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/minivec.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 t;
DROP TABLE t2; DROP TABLE t2;
-- sparsevec -- sparsevec

View File

@@ -0,0 +1,84 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val minivec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val minivec_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::minivec)) 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 minivec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val minivec_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::minivec)) t2;
count
-------
4
(1 row)
DROP TABLE t;
-- cosine
CREATE TABLE t (val minivec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val minivec_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::minivec)) t2;
count
-------
3
(1 row)
DROP TABLE t;

590
test/expected/minivec.out Normal file
View File

@@ -0,0 +1,590 @@
SELECT '[1,2,3]'::minivec;
minivec
---------
[1,2,3]
(1 row)
SELECT '[-1,-2,-3]'::minivec;
minivec
------------
[-1,-2,-3]
(1 row)
SELECT '[1.,2.,3.]'::minivec;
minivec
---------
[1,2,3]
(1 row)
SELECT ' [ 1, 2 , 3 ] '::minivec;
minivec
---------
[1,2,3]
(1 row)
SELECT '[1.23456]'::minivec;
minivec
---------
[1.25]
(1 row)
SELECT '[hello,1]'::minivec;
ERROR: invalid input syntax for type minivec: "[hello,1]"
LINE 1: SELECT '[hello,1]'::minivec;
^
SELECT '[NaN,1]'::minivec;
ERROR: NaN not allowed in minivec
LINE 1: SELECT '[NaN,1]'::minivec;
^
SELECT '[Infinity,1]'::minivec;
ERROR: infinite value not allowed in minivec
LINE 1: SELECT '[Infinity,1]'::minivec;
^
SELECT '[-Infinity,1]'::minivec;
ERROR: infinite value not allowed in minivec
LINE 1: SELECT '[-Infinity,1]'::minivec;
^
SELECT '[61439,-61439]'::minivec;
minivec
----------------
[57344,-57344]
(1 row)
SELECT '[61440,-61440]'::minivec;
ERROR: "61440" is out of range for type minivec
LINE 1: SELECT '[61440,-61440]'::minivec;
^
SELECT '[1e-8,-1e-8]'::minivec;
minivec
---------
[0,-0]
(1 row)
SELECT '[4e38,1]'::minivec;
ERROR: "4e38" is out of range for type minivec
LINE 1: SELECT '[4e38,1]'::minivec;
^
SELECT '[1e-46,1]'::minivec;
minivec
---------
[0,1]
(1 row)
SELECT '[1,2,3'::minivec;
ERROR: invalid input syntax for type minivec: "[1,2,3"
LINE 1: SELECT '[1,2,3'::minivec;
^
SELECT '[1,2,3]9'::minivec;
ERROR: invalid input syntax for type minivec: "[1,2,3]9"
LINE 1: SELECT '[1,2,3]9'::minivec;
^
DETAIL: Junk after closing right brace.
SELECT '1,2,3'::minivec;
ERROR: invalid input syntax for type minivec: "1,2,3"
LINE 1: SELECT '1,2,3'::minivec;
^
DETAIL: Vector contents must start with "[".
SELECT ''::minivec;
ERROR: invalid input syntax for type minivec: ""
LINE 1: SELECT ''::minivec;
^
DETAIL: Vector contents must start with "[".
SELECT '['::minivec;
ERROR: invalid input syntax for type minivec: "["
LINE 1: SELECT '['::minivec;
^
SELECT '[ '::minivec;
ERROR: invalid input syntax for type minivec: "[ "
LINE 1: SELECT '[ '::minivec;
^
SELECT '[,'::minivec;
ERROR: invalid input syntax for type minivec: "[,"
LINE 1: SELECT '[,'::minivec;
^
SELECT '[]'::minivec;
ERROR: minivec must have at least 1 dimension
LINE 1: SELECT '[]'::minivec;
^
SELECT '[ ]'::minivec;
ERROR: minivec must have at least 1 dimension
LINE 1: SELECT '[ ]'::minivec;
^
SELECT '[,]'::minivec;
ERROR: invalid input syntax for type minivec: "[,]"
LINE 1: SELECT '[,]'::minivec;
^
SELECT '[1,]'::minivec;
ERROR: invalid input syntax for type minivec: "[1,]"
LINE 1: SELECT '[1,]'::minivec;
^
SELECT '[1a]'::minivec;
ERROR: invalid input syntax for type minivec: "[1a]"
LINE 1: SELECT '[1a]'::minivec;
^
SELECT '[1,,3]'::minivec;
ERROR: invalid input syntax for type minivec: "[1,,3]"
LINE 1: SELECT '[1,,3]'::minivec;
^
SELECT '[1, ,3]'::minivec;
ERROR: invalid input syntax for type minivec: "[1, ,3]"
LINE 1: SELECT '[1, ,3]'::minivec;
^
SELECT '[1,2,3]'::minivec(3);
minivec
---------
[1,2,3]
(1 row)
SELECT '[1,2,3]'::minivec(2);
ERROR: expected 2 dimensions, not 3
SELECT '[1,2,3]'::minivec(3, 2);
ERROR: invalid type modifier
LINE 1: SELECT '[1,2,3]'::minivec(3, 2);
^
SELECT '[1,2,3]'::minivec('a');
ERROR: invalid input syntax for type integer: "a"
LINE 1: SELECT '[1,2,3]'::minivec('a');
^
SELECT '[1,2,3]'::minivec(0);
ERROR: dimensions for type minivec must be at least 1
LINE 1: SELECT '[1,2,3]'::minivec(0);
^
SELECT '[1,2,3]'::minivec(16001);
ERROR: dimensions for type minivec cannot exceed 16000
LINE 1: SELECT '[1,2,3]'::minivec(16001);
^
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::minivec[]);
unnest
---------
[1,2,3]
[4,5,6]
(2 rows)
SELECT '{"[1,2,3]"}'::minivec(2)[];
ERROR: expected 2 dimensions, not 3
SELECT '[1,2,3]'::minivec + '[4,5,6]';
?column?
----------
[5,7,8]
(1 row)
SELECT '[61439]'::minivec + '[61439]';
ERROR: value out of range: overflow
SELECT '[1,2]'::minivec + '[3]';
ERROR: different minivec dimensions 2 and 1
SELECT '[1,2,3]'::minivec - '[4,5,6]';
ERROR: value out of range: overflow
SELECT '[-61439]'::minivec - '[61439]';
?column?
----------
[-0]
(1 row)
SELECT '[1,2]'::minivec - '[3]';
ERROR: different minivec dimensions 2 and 1
SELECT '[1,2,3]'::minivec * '[4,5,6]';
?column?
-----------
[4,10,16]
(1 row)
SELECT '[61439]'::minivec * '[61439]';
ERROR: value out of range: overflow
SELECT '[1e-7]'::minivec * '[1e-7]';
?column?
----------
[0]
(1 row)
SELECT '[1,2]'::minivec * '[3]';
ERROR: different minivec dimensions 2 and 1
SELECT '[1,2,3]'::minivec || '[4,5]';
?column?
-------------
[1,2,3,4,5]
(1 row)
SELECT array_fill(0, ARRAY[16000])::minivec || '[1]';
ERROR: minivec cannot have more than 16000 dimensions
SELECT '[1,2,3]'::minivec < '[1,2,3]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::minivec < '[1,2]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::minivec <= '[1,2,3]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::minivec <= '[1,2]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::minivec = '[1,2,3]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::minivec = '[1,2]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::minivec != '[1,2,3]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::minivec != '[1,2]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::minivec >= '[1,2,3]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::minivec >= '[1,2]';
?column?
----------
t
(1 row)
SELECT '[1,2,3]'::minivec > '[1,2,3]';
?column?
----------
f
(1 row)
SELECT '[1,2,3]'::minivec > '[1,2]';
?column?
----------
t
(1 row)
SELECT minivec_cmp('[1,2,3]', '[1,2,3]');
minivec_cmp
-------------
0
(1 row)
SELECT minivec_cmp('[1,2,3]', '[0,0,0]');
minivec_cmp
-------------
1
(1 row)
SELECT minivec_cmp('[0,0,0]', '[1,2,3]');
minivec_cmp
-------------
-1
(1 row)
SELECT minivec_cmp('[1,2]', '[1,2,3]');
minivec_cmp
-------------
-1
(1 row)
SELECT minivec_cmp('[1,2,3]', '[1,2]');
minivec_cmp
-------------
1
(1 row)
SELECT minivec_cmp('[1,2]', '[2,3,4]');
minivec_cmp
-------------
-1
(1 row)
SELECT minivec_cmp('[2,3]', '[1,2,3]');
minivec_cmp
-------------
1
(1 row)
SELECT vector_dims('[1,2,3]'::minivec);
vector_dims
-------------
3
(1 row)
SELECT round(l2_norm('[1,1]'::minivec)::numeric, 5);
round
---------
1.41421
(1 row)
SELECT l2_norm('[3,4]'::minivec);
l2_norm
---------
5
(1 row)
SELECT l2_norm('[0,1]'::minivec);
l2_norm
---------
1
(1 row)
SELECT l2_norm('[0,0]'::minivec);
l2_norm
---------
0
(1 row)
SELECT l2_norm('[2]'::minivec);
l2_norm
---------
2
(1 row)
SELECT l2_distance('[0,0]'::minivec, '[3,4]');
l2_distance
-------------
5
(1 row)
SELECT l2_distance('[0,0]'::minivec, '[0,1]');
l2_distance
-------------
1
(1 row)
SELECT l2_distance('[1,2]'::minivec, '[3]');
ERROR: different minivec dimensions 2 and 1
SELECT l2_distance('[1,1,1,1,1,1,1,1,1]'::minivec, '[1,1,1,1,1,1,1,4,5]');
l2_distance
-------------
5
(1 row)
SELECT '[0,0]'::minivec <-> '[3,4]';
?column?
----------
5
(1 row)
SELECT inner_product('[1,2]'::minivec, '[3,4]');
inner_product
---------------
11
(1 row)
SELECT inner_product('[1,2]'::minivec, '[3]');
ERROR: different minivec dimensions 2 and 1
SELECT inner_product('[448]'::minivec, '[448]');
inner_product
---------------
200704
(1 row)
SELECT inner_product('[1,1,1,1,1,1,1,1,1]'::minivec, '[1,2,3,4,5,6,7,8,9]');
inner_product
---------------
44
(1 row)
SELECT '[1,2]'::minivec <#> '[3,4]';
?column?
----------
-11
(1 row)
SELECT cosine_distance('[1,2]'::minivec, '[2,4]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,2]'::minivec, '[0,0]');
cosine_distance
-----------------
NaN
(1 row)
SELECT cosine_distance('[1,1]'::minivec, '[1,1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,0]'::minivec, '[0,2]');
cosine_distance
-----------------
1
(1 row)
SELECT cosine_distance('[1,1]'::minivec, '[-1,-1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[1,2]'::minivec, '[3]');
ERROR: different minivec dimensions 2 and 1
SELECT cosine_distance('[1,1]'::minivec, '[1.1,1.1]');
cosine_distance
-----------------
0
(1 row)
SELECT cosine_distance('[1,1]'::minivec, '[-1.1,-1.1]');
cosine_distance
-----------------
2
(1 row)
SELECT cosine_distance('[1,2,3,4,5,6,7,8,9]'::minivec, '[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]'::minivec, '[-1,-2,-3,-4,-5,-6,-7,-8,-9]');
cosine_distance
-----------------
2
(1 row)
SELECT '[1,2]'::minivec <=> '[2,4]';
?column?
----------
0
(1 row)
SELECT l1_distance('[0,0]'::minivec, '[3,4]');
l1_distance
-------------
7
(1 row)
SELECT l1_distance('[0,0]'::minivec, '[0,1]');
l1_distance
-------------
1
(1 row)
SELECT l1_distance('[1,2]'::minivec, '[3]');
ERROR: different minivec dimensions 2 and 1
SELECT l1_distance('[1,2,3,4,5,6,7,8,9]'::minivec, '[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]'::minivec, '[0,3,2,5,4,7,6,9,8]');
l1_distance
-------------
7
(1 row)
SELECT '[0,0]'::minivec <+> '[3,4]';
?column?
----------
7
(1 row)
SELECT l2_normalize('[3,4]'::minivec);
l2_normalize
--------------
[0.625,0.75]
(1 row)
SELECT l2_normalize('[3,0]'::minivec);
l2_normalize
--------------
[1,0]
(1 row)
SELECT l2_normalize('[0,0.1]'::minivec);
l2_normalize
--------------
[0,1]
(1 row)
SELECT l2_normalize('[0,0]'::minivec);
l2_normalize
--------------
[0,0]
(1 row)
SELECT l2_normalize('[448]'::minivec);
l2_normalize
--------------
[1]
(1 row)
SELECT binary_quantize('[1,0,-1]'::minivec);
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]'::minivec);
binary_quantize
-----------------
01001110101
(1 row)
SELECT subvector('[1,2,3,4,5]'::minivec, 1, 3);
subvector
-----------
[1,2,3]
(1 row)
SELECT subvector('[1,2,3,4,5]'::minivec, 3, 2);
subvector
-----------
[3,4]
(1 row)
SELECT subvector('[1,2,3,4,5]'::minivec, -1, 3);
subvector
-----------
[1]
(1 row)
SELECT subvector('[1,2,3,4,5]'::minivec, 3, 9);
subvector
-----------
[3,4,5]
(1 row)
SELECT subvector('[1,2,3,4,5]'::minivec, 1, 0);
ERROR: minivec must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::minivec, 3, -1);
ERROR: minivec must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::minivec, -1, 2);
ERROR: minivec must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::minivec, 2147483647, 10);
ERROR: minivec must have at least 1 dimension
SELECT subvector('[1,2,3,4,5]'::minivec, 3, 2147483647);
subvector
-----------
[3,4,5]
(1 row)
SELECT subvector('[1,2,3,4,5]'::minivec, -2147483644, 2147483647);
subvector
-----------
[1,2]
(1 row)

View File

@@ -22,6 +22,17 @@ SELECT * FROM t ORDER BY val;
DROP TABLE t; DROP TABLE t;
-- minivec
CREATE TABLE t (val minivec(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 -- sparsevec
CREATE TABLE t (val sparsevec(3)); CREATE TABLE t (val sparsevec(3));

View File

@@ -38,6 +38,22 @@ SELECT '{1,2,3}'::real[]::halfvec(2);
SELECT '{65520,-65520}'::real[]::halfvec; SELECT '{65520,-65520}'::real[]::halfvec;
SELECT '{1e-8,-1e-8}'::real[]::halfvec; SELECT '{1e-8,-1e-8}'::real[]::halfvec;
SELECT '[1,2,3]'::vector::minivec;
SELECT '[1,2,3]'::vector::minivec(3);
SELECT '[1,2,3]'::vector::minivec(2);
SELECT '[61440]'::vector::minivec;
SELECT '[1e-8]'::vector::minivec;
SELECT '[1,2,3]'::minivec::vector;
SELECT '[1,2,3]'::minivec::vector(3);
SELECT '[1,2,3]'::minivec::vector(2);
SELECT '{1,2,3}'::real[]::minivec;
SELECT '{1,2,3}'::real[]::minivec(3);
SELECT '{1,2,3}'::real[]::minivec(2);
SELECT '{61440,-61440}'::real[]::minivec;
SELECT '{1e-8,-1e-8}'::real[]::minivec;
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec; 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(5);
SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec(4); SELECT '[0,1.5,0,3.5,0]'::vector::sparsevec(4);

View File

@@ -28,6 +28,21 @@ SELECT * FROM t2 ORDER BY val;
DROP TABLE t; DROP TABLE t;
DROP TABLE t2; DROP TABLE t2;
-- minivec
CREATE TABLE t (val minivec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE TABLE t2 (val minivec(3));
\copy t TO 'results/minivec.bin' WITH (FORMAT binary)
\copy t2 FROM 'results/minivec.bin' WITH (FORMAT binary)
SELECT * FROM t2 ORDER BY val;
DROP TABLE t;
DROP TABLE t2;
-- sparsevec -- sparsevec
CREATE TABLE t (val sparsevec(3)); CREATE TABLE t (val sparsevec(3));

View File

@@ -0,0 +1,45 @@
SET enable_seqscan = off;
-- L2
CREATE TABLE t (val minivec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val minivec_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::minivec)) 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 minivec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val minivec_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::minivec)) t2;
DROP TABLE t;
-- cosine
CREATE TABLE t (val minivec(3));
INSERT INTO t (val) VALUES ('[0,0,0]'), ('[1,2,3]'), ('[1,1,1]'), (NULL);
CREATE INDEX ON t USING ivfflat (val minivec_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::minivec)) t2;
DROP TABLE t;

134
test/sql/minivec.sql Normal file
View File

@@ -0,0 +1,134 @@
SELECT '[1,2,3]'::minivec;
SELECT '[-1,-2,-3]'::minivec;
SELECT '[1.,2.,3.]'::minivec;
SELECT ' [ 1, 2 , 3 ] '::minivec;
SELECT '[1.23456]'::minivec;
SELECT '[hello,1]'::minivec;
SELECT '[NaN,1]'::minivec;
SELECT '[Infinity,1]'::minivec;
SELECT '[-Infinity,1]'::minivec;
SELECT '[61439,-61439]'::minivec;
SELECT '[61440,-61440]'::minivec;
SELECT '[1e-8,-1e-8]'::minivec;
SELECT '[4e38,1]'::minivec;
SELECT '[1e-46,1]'::minivec;
SELECT '[1,2,3'::minivec;
SELECT '[1,2,3]9'::minivec;
SELECT '1,2,3'::minivec;
SELECT ''::minivec;
SELECT '['::minivec;
SELECT '[ '::minivec;
SELECT '[,'::minivec;
SELECT '[]'::minivec;
SELECT '[ ]'::minivec;
SELECT '[,]'::minivec;
SELECT '[1,]'::minivec;
SELECT '[1a]'::minivec;
SELECT '[1,,3]'::minivec;
SELECT '[1, ,3]'::minivec;
SELECT '[1,2,3]'::minivec(3);
SELECT '[1,2,3]'::minivec(2);
SELECT '[1,2,3]'::minivec(3, 2);
SELECT '[1,2,3]'::minivec('a');
SELECT '[1,2,3]'::minivec(0);
SELECT '[1,2,3]'::minivec(16001);
SELECT unnest('{"[1,2,3]", "[4,5,6]"}'::minivec[]);
SELECT '{"[1,2,3]"}'::minivec(2)[];
SELECT '[1,2,3]'::minivec + '[4,5,6]';
SELECT '[61439]'::minivec + '[61439]';
SELECT '[1,2]'::minivec + '[3]';
SELECT '[1,2,3]'::minivec - '[4,5,6]';
SELECT '[-61439]'::minivec - '[61439]';
SELECT '[1,2]'::minivec - '[3]';
SELECT '[1,2,3]'::minivec * '[4,5,6]';
SELECT '[61439]'::minivec * '[61439]';
SELECT '[1e-7]'::minivec * '[1e-7]';
SELECT '[1,2]'::minivec * '[3]';
SELECT '[1,2,3]'::minivec || '[4,5]';
SELECT array_fill(0, ARRAY[16000])::minivec || '[1]';
SELECT '[1,2,3]'::minivec < '[1,2,3]';
SELECT '[1,2,3]'::minivec < '[1,2]';
SELECT '[1,2,3]'::minivec <= '[1,2,3]';
SELECT '[1,2,3]'::minivec <= '[1,2]';
SELECT '[1,2,3]'::minivec = '[1,2,3]';
SELECT '[1,2,3]'::minivec = '[1,2]';
SELECT '[1,2,3]'::minivec != '[1,2,3]';
SELECT '[1,2,3]'::minivec != '[1,2]';
SELECT '[1,2,3]'::minivec >= '[1,2,3]';
SELECT '[1,2,3]'::minivec >= '[1,2]';
SELECT '[1,2,3]'::minivec > '[1,2,3]';
SELECT '[1,2,3]'::minivec > '[1,2]';
SELECT minivec_cmp('[1,2,3]', '[1,2,3]');
SELECT minivec_cmp('[1,2,3]', '[0,0,0]');
SELECT minivec_cmp('[0,0,0]', '[1,2,3]');
SELECT minivec_cmp('[1,2]', '[1,2,3]');
SELECT minivec_cmp('[1,2,3]', '[1,2]');
SELECT minivec_cmp('[1,2]', '[2,3,4]');
SELECT minivec_cmp('[2,3]', '[1,2,3]');
SELECT vector_dims('[1,2,3]'::minivec);
SELECT round(l2_norm('[1,1]'::minivec)::numeric, 5);
SELECT l2_norm('[3,4]'::minivec);
SELECT l2_norm('[0,1]'::minivec);
SELECT l2_norm('[0,0]'::minivec);
SELECT l2_norm('[2]'::minivec);
SELECT l2_distance('[0,0]'::minivec, '[3,4]');
SELECT l2_distance('[0,0]'::minivec, '[0,1]');
SELECT l2_distance('[1,2]'::minivec, '[3]');
SELECT l2_distance('[1,1,1,1,1,1,1,1,1]'::minivec, '[1,1,1,1,1,1,1,4,5]');
SELECT '[0,0]'::minivec <-> '[3,4]';
SELECT inner_product('[1,2]'::minivec, '[3,4]');
SELECT inner_product('[1,2]'::minivec, '[3]');
SELECT inner_product('[448]'::minivec, '[448]');
SELECT inner_product('[1,1,1,1,1,1,1,1,1]'::minivec, '[1,2,3,4,5,6,7,8,9]');
SELECT '[1,2]'::minivec <#> '[3,4]';
SELECT cosine_distance('[1,2]'::minivec, '[2,4]');
SELECT cosine_distance('[1,2]'::minivec, '[0,0]');
SELECT cosine_distance('[1,1]'::minivec, '[1,1]');
SELECT cosine_distance('[1,0]'::minivec, '[0,2]');
SELECT cosine_distance('[1,1]'::minivec, '[-1,-1]');
SELECT cosine_distance('[1,2]'::minivec, '[3]');
SELECT cosine_distance('[1,1]'::minivec, '[1.1,1.1]');
SELECT cosine_distance('[1,1]'::minivec, '[-1.1,-1.1]');
SELECT cosine_distance('[1,2,3,4,5,6,7,8,9]'::minivec, '[1,2,3,4,5,6,7,8,9]');
SELECT cosine_distance('[1,2,3,4,5,6,7,8,9]'::minivec, '[-1,-2,-3,-4,-5,-6,-7,-8,-9]');
SELECT '[1,2]'::minivec <=> '[2,4]';
SELECT l1_distance('[0,0]'::minivec, '[3,4]');
SELECT l1_distance('[0,0]'::minivec, '[0,1]');
SELECT l1_distance('[1,2]'::minivec, '[3]');
SELECT l1_distance('[1,2,3,4,5,6,7,8,9]'::minivec, '[1,2,3,4,5,6,7,8,9]');
SELECT l1_distance('[1,2,3,4,5,6,7,8,9]'::minivec, '[0,3,2,5,4,7,6,9,8]');
SELECT '[0,0]'::minivec <+> '[3,4]';
SELECT l2_normalize('[3,4]'::minivec);
SELECT l2_normalize('[3,0]'::minivec);
SELECT l2_normalize('[0,0.1]'::minivec);
SELECT l2_normalize('[0,0]'::minivec);
SELECT l2_normalize('[448]'::minivec);
SELECT binary_quantize('[1,0,-1]'::minivec);
SELECT binary_quantize('[0,0.1,-0.2,-0.3,0.4,0.5,0.6,-0.7,0.8,-0.9,1]'::minivec);
SELECT subvector('[1,2,3,4,5]'::minivec, 1, 3);
SELECT subvector('[1,2,3,4,5]'::minivec, 3, 2);
SELECT subvector('[1,2,3,4,5]'::minivec, -1, 3);
SELECT subvector('[1,2,3,4,5]'::minivec, 3, 9);
SELECT subvector('[1,2,3,4,5]'::minivec, 1, 0);
SELECT subvector('[1,2,3,4,5]'::minivec, 3, -1);
SELECT subvector('[1,2,3,4,5]'::minivec, -1, 2);
SELECT subvector('[1,2,3,4,5]'::minivec, 2147483647, 10);
SELECT subvector('[1,2,3,4,5]'::minivec, 3, 2147483647);
SELECT subvector('[1,2,3,4,5]'::minivec, -2147483644, 2147483647);

View File

@@ -94,7 +94,8 @@ like($explain, qr/Seq Scan/);
$explain = $node->safe_psql("postgres", qq( $explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query'; EXPLAIN ANALYZE SELECT i FROM tst WHERE v <-> '$query' < 1 ORDER BY v <-> '$query';
)); ));
like($explain, qr/Seq Scan/); # TODO Do not use index
like($explain, qr/Index Scan using idx/);
# Test attribute index # Test attribute index
$node->safe_psql("postgres", "CREATE INDEX attribute_idx ON tst (c);"); $node->safe_psql("postgres", "CREATE INDEX attribute_idx ON tst (c);");
@@ -109,6 +110,7 @@ $node->safe_psql("postgres", "CREATE INDEX partial_idx ON tst USING ivfflat (v v
$explain = $node->safe_psql("postgres", qq( $explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit; EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
)); ));
like($explain, qr/Index Scan using partial_idx/); # TODO Use partial index
like($explain, qr/Index Scan using idx/);
done_testing(); done_testing();

View File

@@ -18,13 +18,9 @@ $node->start;
# Create table and index # Create table and index
$node->safe_psql("postgres", "CREATE EXTENSION vector;"); $node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim), c int4, t text);"); $node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim), c int4, t text);");
$node->safe_psql("postgres", "CREATE TABLE cat (i int4 PRIMARY KEY, t text, b boolean);");
$node->safe_psql("postgres", $node->safe_psql("postgres",
"INSERT INTO tst SELECT i, ARRAY[$array_sql], i % $nc, 'test ' || i FROM generate_series(1, 10000) i;" "INSERT INTO tst SELECT i, ARRAY[$array_sql], i % $nc, 'test ' || i FROM generate_series(1, 10000) i;"
); );
$node->safe_psql("postgres",
"INSERT INTO cat SELECT i, 'cat ' || i, i % 5 = 0 FROM generate_series(1, $nc) i;"
);
$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);"); $node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v vector_l2_ops);");
$node->safe_psql("postgres", "ANALYZE tst;"); $node->safe_psql("postgres", "ANALYZE tst;");
@@ -41,7 +37,8 @@ my $c = int(rand() * $nc);
my $explain = $node->safe_psql("postgres", qq( my $explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit; EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
)); ));
like($explain, qr/Seq Scan/); # TODO Do not use index
like($explain, qr/Index Scan using idx/);
# Test attribute filtering with few rows removed # Test attribute filtering with few rows removed
$explain = $node->safe_psql("postgres", qq( $explain = $node->safe_psql("postgres", qq(
@@ -59,7 +56,8 @@ like($explain, qr/Index Scan using idx/);
$explain = $node->safe_psql("postgres", qq( $explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT i FROM tst WHERE c < 1 ORDER BY v <-> '$query' LIMIT $limit; EXPLAIN ANALYZE SELECT i FROM tst WHERE c < 1 ORDER BY v <-> '$query' LIMIT $limit;
)); ));
like($explain, qr/Seq Scan/); # TODO Do not use index
like($explain, qr/Index Scan using idx/);
# Test attribute filtering with few rows removed like # Test attribute filtering with few rows removed like
$explain = $node->safe_psql("postgres", qq( $explain = $node->safe_psql("postgres", qq(
@@ -98,25 +96,13 @@ $explain = $node->safe_psql("postgres", qq(
)); ));
like($explain, qr/Seq Scan/); like($explain, qr/Seq Scan/);
# Test join
$explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT cat.t FROM cat INNER JOIN tst ON cat.i = tst.c ORDER BY v <-> '$query' LIMIT $limit;
));
like($explain, qr/Index Scan using idx/);
# Test join with attribute filtering
$explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT cat.t FROM cat INNER JOIN tst ON cat.i = tst.c WHERE cat.b = 't' ORDER BY v <-> '$query' LIMIT $limit;
));
like($explain, qr/Index Scan using idx/);
# Test attribute index # Test attribute index
$node->safe_psql("postgres", "CREATE INDEX attribute_idx ON tst (c);"); $node->safe_psql("postgres", "CREATE INDEX attribute_idx ON tst (c);");
$explain = $node->safe_psql("postgres", qq( $explain = $node->safe_psql("postgres", qq(
EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit; EXPLAIN ANALYZE SELECT i FROM tst WHERE c = $c ORDER BY v <-> '$query' LIMIT $limit;
)); ));
# Use attribute index # TODO Use attribute index
like($explain, qr/Bitmap Index Scan on attribute_idx/); like($explain, qr/Index Scan using idx/);
# Test partial index # Test partial index
$node->safe_psql("postgres", "CREATE INDEX partial_idx ON tst USING hnsw (v vector_l2_ops) WHERE (c = $c);"); $node->safe_psql("postgres", "CREATE INDEX partial_idx ON tst USING hnsw (v vector_l2_ops) WHERE (c = $c);");

View File

@@ -40,6 +40,10 @@ for (1 .. 50)
$actual = $node->safe_psql("postgres", "SELECT halfvec_cmp(v::halfvec, '$query'::real[]::halfvec) FROM tst"); $actual = $node->safe_psql("postgres", "SELECT halfvec_cmp(v::halfvec, '$query'::real[]::halfvec) FROM tst");
is($expected, $actual); is($expected, $actual);
# Test minivec
$actual = $node->safe_psql("postgres", "SELECT minivec_cmp(v::minivec, '$query'::real[]::minivec) FROM tst");
is($expected, $actual);
# Test sparsevec # Test sparsevec
$actual = $node->safe_psql("postgres", "SELECT sparsevec_cmp(v::vector::sparsevec, '$query'::real[]::vector::sparsevec) FROM tst"); $actual = $node->safe_psql("postgres", "SELECT sparsevec_cmp(v::vector::sparsevec, '$query'::real[]::vector::sparsevec) FROM tst");
is($expected, $actual); is($expected, $actual);

View File

@@ -45,6 +45,10 @@ for my $function (@functions)
my $actual = $node->safe_psql("postgres", "SELECT $function(v::halfvec, '$query'::vector::halfvec) FROM tst"); my $actual = $node->safe_psql("postgres", "SELECT $function(v::halfvec, '$query'::vector::halfvec) FROM tst");
is($expected, $actual, "halfvec $function"); is($expected, $actual, "halfvec $function");
# Test minivec
$actual = $node->safe_psql("postgres", "SELECT $function(v::minivec, '$query'::vector::minivec) FROM tst");
is($expected, $actual, "minivec $function");
# Test sparsevec # Test sparsevec
$actual = $node->safe_psql("postgres", "SELECT $function(v::sparsevec, '$query'::vector::sparsevec) FROM tst"); $actual = $node->safe_psql("postgres", "SELECT $function(v::sparsevec, '$query'::vector::sparsevec) FROM tst");
is($expected, $actual, "sparsevec $function"); is($expected, $actual, "sparsevec $function");

View File

@@ -12,8 +12,8 @@ $node->start;
# Create extension # Create extension
$node->safe_psql("postgres", "CREATE EXTENSION vector;"); $node->safe_psql("postgres", "CREATE EXTENSION vector;");
my @types = ("vector", "halfvec", "sparsevec"); my @types = ("vector", "halfvec", "minivec", "sparsevec");
my @inputs = ("[1.23,4.56,7.89]", "[1.23,4.56,7.89]", "{1:1.23,2:4.56,3:7.89}/3"); my @inputs = ("[1.23,4.56,7.89]", "[1.23,4.56,7.89]", "[1.23,4.56,7.89]", "{1:1.23,2:4.56,3:7.89}/3");
my @subs = (" ", " ", ",", ":", "-", "1", "9", "\0", "2147483648", "-2147483649"); my @subs = (" ", " ", ",", ":", "-", "1", "9", "\0", "2147483648", "-2147483649");
for my $i (0 .. $#types) for my $i (0 .. $#types)

View File

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

View File

@@ -0,0 +1,132 @@
use strict;
use warnings FATAL => 'all';
use PostgreSQL::Test::Cluster;
use PostgreSQL::Test::Utils;
use Test::More;
my $node;
my @queries = ();
my @expected;
my $limit = 20;
my $dim = 10;
my $array_sql = join(",", ('2 * random() * random()') x $dim);
sub test_recall
{
my ($min, $operator) = @_;
my $correct = 0;
my $total = 0;
my $explain = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v $operator '$queries[0]' LIMIT $limit;
));
like($explain, qr/Index Scan/);
for my $i (0 .. $#queries)
{
my $actual = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SELECT i FROM tst ORDER BY v $operator '$queries[$i]' LIMIT $limit;
));
my @actual_ids = split("\n", $actual);
my %actual_set = map { $_ => 1 } @actual_ids;
my @expected_ids = split("\n", $expected[$i]);
foreach (@expected_ids)
{
if (exists($actual_set{$_}))
{
$correct++;
}
$total++;
}
}
cmp_ok($correct / $total, ">=", $min, $operator);
}
# Initialize node
$node = PostgreSQL::Test::Cluster->new('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v minivec($dim));");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 10000) i;"
);
# Generate queries
for (1 .. 20)
{
my @r = ();
for (1 .. $dim)
{
push(@r, rand());
}
push(@queries, "[" . join(",", @r) . "]");
}
# Check each index type
my @operators = ("<->", "<#>", "<=>", "<+>");
my @opclasses = ("minivec_l2_ops", "minivec_ip_ops", "minivec_cosine_ops", "minivec_l1_ops");
for my $i (0 .. $#operators)
{
my $operator = $operators[$i];
my $opclass = $opclasses[$i];
# Get exact results
@expected = ();
foreach (@queries)
{
my $res = $node->safe_psql("postgres", "SELECT i FROM tst ORDER BY v $operator '$_' LIMIT $limit;");
push(@expected, $res);
}
# Build index serially
$node->safe_psql("postgres", qq(
SET max_parallel_maintenance_workers = 0;
CREATE INDEX idx ON tst USING hnsw (v $opclass);
));
# Test approximate results
my $min = 0.98;
test_recall($min, $operator);
$node->safe_psql("postgres", "DROP INDEX idx;");
# Build index in parallel in memory
my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
SET client_min_messages = DEBUG;
SET min_parallel_table_scan_size = 1;
CREATE INDEX idx ON tst USING hnsw (v $opclass);
));
is($ret, 0, $stderr);
like($stderr, qr/using \d+ parallel workers/);
# Test approximate results
test_recall($min, $operator);
$node->safe_psql("postgres", "DROP INDEX idx;");
# Build index in parallel on disk
# Set parallel_workers on table to use workers with low maintenance_work_mem
($ret, $stdout, $stderr) = $node->psql("postgres", qq(
ALTER TABLE tst SET (parallel_workers = 2);
SET client_min_messages = DEBUG;
SET maintenance_work_mem = '4MB';
CREATE INDEX idx ON tst USING hnsw (v $opclass);
ALTER TABLE tst RESET (parallel_workers);
));
is($ret, 0, $stderr);
like($stderr, qr/using \d+ parallel workers/);
like($stderr, qr/hnsw graph no longer fits into maintenance_work_mem/);
$node->safe_psql("postgres", "DROP INDEX idx;");
}
done_testing();

View File

@@ -0,0 +1,113 @@
use strict;
use warnings FATAL => 'all';
use PostgreSQL::Test::Cluster;
use PostgreSQL::Test::Utils;
use Test::More;
my $node;
my @queries = ();
my @expected;
my $limit = 20;
my $dim = 10;
my $array_sql = join(",", ('2 * random() * random()') x $dim);
sub test_recall
{
my ($min, $operator) = @_;
my $correct = 0;
my $total = 0;
my $explain = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v $operator '$queries[0]' LIMIT $limit;
));
like($explain, qr/Index Scan/);
for my $i (0 .. $#queries)
{
my $actual = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SELECT i FROM tst ORDER BY v $operator '$queries[$i]' LIMIT $limit;
));
my @actual_ids = split("\n", $actual);
my %actual_set = map { $_ => 1 } @actual_ids;
my @expected_ids = split("\n", $expected[$i]);
foreach (@expected_ids)
{
if (exists($actual_set{$_}))
{
$correct++;
}
$total++;
}
}
cmp_ok($correct / $total, ">=", $min, $operator);
}
# Initialize node
$node = PostgreSQL::Test::Cluster->new('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (i serial, v minivec($dim));");
# Generate queries
for (1 .. 20)
{
my @r = ();
for (1 .. $dim)
{
push(@r, rand());
}
push(@queries, "[" . join(",", @r) . "]");
}
# Check each index type
my @operators = ("<->", "<#>", "<=>", "<+>");
my @opclasses = ("minivec_l2_ops", "minivec_ip_ops", "minivec_cosine_ops", "minivec_l1_ops");
for my $i (0 .. $#operators)
{
my $operator = $operators[$i];
my $opclass = $opclasses[$i];
# Add index
$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v $opclass);");
# Use concurrent inserts
$node->pgbench(
"--no-vacuum --client=10 --transactions=1000",
0,
[qr{actually processed}],
[qr{^$}],
"concurrent INSERTs",
{
"040_hnsw_minivec_insert_recall_$opclass" => "INSERT INTO tst (v) VALUES (ARRAY[$array_sql]);"
}
);
# Get exact results
@expected = ();
foreach (@queries)
{
my $res = $node->safe_psql("postgres", qq(
SET enable_indexscan = off;
SELECT i FROM tst ORDER BY v $operator '$_' LIMIT $limit;
));
push(@expected, $res);
}
# Test approximate results
my $min = 0.98;
test_recall($min, $operator);
$node->safe_psql("postgres", "DROP INDEX idx;");
$node->safe_psql("postgres", "TRUNCATE tst;");
}
done_testing();

View File

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

View File

@@ -0,0 +1,97 @@
use strict;
use warnings FATAL => 'all';
use PostgreSQL::Test::Cluster;
use PostgreSQL::Test::Utils;
use Test::More;
my $node;
my @queries = ();
my @expected;
my $limit = 20;
sub test_recall
{
my ($min, $ef_search, $test_name) = @_;
my $correct = 0;
my $total = 0;
my $explain = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SET hnsw.ef_search = $ef_search;
EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v <-> '$queries[0]' LIMIT $limit;
));
like($explain, qr/Index Scan/);
for my $i (0 .. $#queries)
{
my $actual = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SET hnsw.ef_search = $ef_search;
SELECT i FROM tst ORDER BY v <-> '$queries[$i]' LIMIT $limit;
));
my @actual_ids = split("\n", $actual);
my %actual_set = map { $_ => 1 } @actual_ids;
my @expected_ids = split("\n", $expected[$i]);
foreach (@expected_ids)
{
if (exists($actual_set{$_}))
{
$correct++;
}
$total++;
}
}
cmp_ok($correct / $total, ">=", $min, $test_name);
}
# Initialize node
$node = PostgreSQL::Test::Cluster->new('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v minivec(3));");
$node->safe_psql("postgres", "ALTER TABLE tst SET (autovacuum_enabled = false);");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()] FROM generate_series(1, 10000) i;"
);
# Add index
$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v minivec_l2_ops) WITH (m = 4, ef_construction = 8);");
# Delete data
$node->safe_psql("postgres", "DELETE FROM tst WHERE i > 2500;");
# Generate queries
for (1 .. 20)
{
my $r1 = rand();
my $r2 = rand();
my $r3 = rand();
push(@queries, "[$r1,$r2,$r3]");
}
# Get exact results
@expected = ();
foreach (@queries)
{
my $res = $node->safe_psql("postgres", qq(
SET enable_indexscan = off;
SELECT i FROM tst ORDER BY v <-> '$_' LIMIT $limit;
));
push(@expected, $res);
}
test_recall(0.18, $limit, "before vacuum");
test_recall(0.93, 100, "before vacuum");
# TODO Test concurrent inserts with vacuum
$node->safe_psql("postgres", "VACUUM tst;");
test_recall(0.95, $limit, "after vacuum");
done_testing();

View File

@@ -0,0 +1,58 @@
use strict;
use warnings FATAL => 'all';
use PostgreSQL::Test::Cluster;
use PostgreSQL::Test::Utils;
use Test::More;
# Initialize node
my $node = PostgreSQL::Test::Cluster->new('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (v minivec(3));");
sub insert_vectors
{
for my $i (1 .. 20)
{
$node->safe_psql("postgres", "INSERT INTO tst VALUES ('[1,1,1]');");
}
}
sub test_duplicates
{
my $res = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SET hnsw.ef_search = 1;
SELECT COUNT(*) FROM (SELECT * FROM tst ORDER BY v <-> '[1,1,1]') t;
));
is($res, 10);
}
# Test duplicates with build
insert_vectors();
$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v minivec_l2_ops);");
test_duplicates();
# Reset
$node->safe_psql("postgres", "TRUNCATE tst;");
# Test duplicates with inserts
insert_vectors();
test_duplicates();
# Test fallback path for inserts
$node->pgbench(
"--no-vacuum --client=5 --transactions=100",
0,
[qr{actually processed}],
[qr{^$}],
"concurrent INSERTs",
{
"042_hnsw_minivec_duplicates" => "INSERT INTO tst VALUES ('[1,1,1]');"
}
);
done_testing();

View File

@@ -0,0 +1,154 @@
use strict;
use warnings FATAL => 'all';
use PostgreSQL::Test::Cluster;
use PostgreSQL::Test::Utils;
use Test::More;
my $node;
my @queries = ();
my @expected;
my $limit = 20;
my $dim = 10;
my $array_sql = join(",", ('random()') x $dim);
sub test_recall
{
my ($probes, $min, $operator) = @_;
my $correct = 0;
my $total = 0;
my $explain = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SET ivfflat.probes = $probes;
EXPLAIN ANALYZE SELECT i FROM tst ORDER BY v $operator '$queries[0]' LIMIT $limit;
));
like($explain, qr/Index Scan using idx on tst/);
for my $i (0 .. $#queries)
{
my $actual = $node->safe_psql("postgres", qq(
SET enable_seqscan = off;
SET ivfflat.probes = $probes;
SELECT i FROM tst ORDER BY v $operator '$queries[$i]' LIMIT $limit;
));
my @actual_ids = split("\n", $actual);
my @expected_ids = split("\n", $expected[$i]);
my %expected_set = map { $_ => 1 } @expected_ids;
foreach (@actual_ids)
{
if (exists($expected_set{$_}))
{
$correct++;
}
}
$total += $limit;
}
cmp_ok($correct / $total, ">=", $min, $operator);
}
# Initialize node
$node = PostgreSQL::Test::Cluster->new('node');
$node->init;
$node->start;
# Create table
$node->safe_psql("postgres", "CREATE EXTENSION vector;");
$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v minivec($dim));");
$node->safe_psql("postgres",
"INSERT INTO tst SELECT i, ARRAY[$array_sql] FROM generate_series(1, 100000) i;"
);
# Generate queries
for (1 .. 20)
{
my @r = ();
for (1 .. $dim)
{
push(@r, rand());
}
push(@queries, "[" . join(",", @r) . "]");
}
# Check each index type
my @operators = ("<->", "<#>", "<=>");
my @opclasses = ("minivec_l2_ops", "minivec_ip_ops", "minivec_cosine_ops");
for my $i (0 .. $#operators)
{
my $operator = $operators[$i];
my $opclass = $opclasses[$i];
# Get exact results
@expected = ();
foreach (@queries)
{
my $res = $node->safe_psql("postgres", qq(
WITH top AS (
SELECT v $operator '$_' AS distance FROM tst ORDER BY distance LIMIT $limit
)
SELECT i FROM tst WHERE (v $operator '$_') <= (SELECT MAX(distance) FROM top)
));
push(@expected, $res);
}
# Build index serially
$node->safe_psql("postgres", qq(
SET max_parallel_maintenance_workers = 0;
CREATE INDEX idx ON tst USING ivfflat (v $opclass);
));
# Test approximate results
if ($operator ne "<#>")
{
# TODO Fix test (uniform random vectors all have similar inner product)
test_recall(1, 0.33, $operator);
test_recall(10, 0.93, $operator);
}
# Test probes equals lists
if ($operator eq "<=>")
{
test_recall(100, 0.98, $operator);
}
else
{
test_recall(100, 1.00, $operator);
}
$node->safe_psql("postgres", "DROP INDEX idx;");
# Build index in parallel
my ($ret, $stdout, $stderr) = $node->psql("postgres", qq(
SET client_min_messages = DEBUG;
SET min_parallel_table_scan_size = 1;
CREATE INDEX idx ON tst USING ivfflat (v $opclass);
));
is($ret, 0, $stderr);
like($stderr, qr/using \d+ parallel workers/);
# Test approximate results
if ($operator ne "<#>")
{
# TODO Fix test (uniform random vectors all have similar inner product)
test_recall(1, 0.33, $operator);
test_recall(10, 0.93, $operator);
}
# Test probes equals lists
if ($operator eq "<=>")
{
test_recall(100, 0.98, $operator);
}
else
{
test_recall(100, 1.00, $operator);
}
$node->safe_psql("postgres", "DROP INDEX idx;");
}
done_testing();