Use maxItems 10 [skip ci]

.github/workflows/build.yml

@@ -8,8 +8,6 @@ jobs:
       fail-fast: false
       matrix:
         include:
-          - postgres: 16
-            os: ubuntu-22.04
           - postgres: 15
             os: ubuntu-22.04
           - postgres: 14
@@ -27,8 +25,6 @@ jobs:
           postgres-version: ${{ matrix.postgres }}
           dev-files: true
       - run: make
-        env:
-          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
@@ -48,8 +44,6 @@ jobs:
         with:
           postgres-version: 14
       - run: make
-        env:
-          PG_CFLAGS: -Wall -Wextra -Werror -Wno-unused-parameter
       - run: make install
       - run: make installcheck
       - if: ${{ failure() }}
@@ -60,7 +54,6 @@ jobs:
           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: windows-latest
     if: ${{ !startsWith(github.ref_name, 'mac') }}
@@ -77,5 +70,3 @@ jobs:
           nmake /NOLOGO /F Makefile.win clean && ^
           nmake /NOLOGO /F Makefile.win uninstall
         shell: cmd
-      - if: ${{ failure() }}
-        run: cat regression.diffs

CHANGELOG.md

@@ -1,36 +1,6 @@
-## 0.5.0 (unreleased)
-
-- Added HNSW index type
-- Added support for parallel index builds
-- Added `l1_distance` function
-- Added `normalize_l2` function
-- Added element-wise multiplication for vectors
-- Added `sum` aggregate
-- Improved performance of distance functions
-- Fixed out of range results for cosine distance
-- Fixed results for NULL and NaN distances
-
-## 0.4.4 (2023-06-12)
-
-- Improved error message for malformed vector literal
-- Fixed segmentation fault with text input
-- Fixed consecutive delimiters with text input
-
-## 0.4.3 (2023-06-10)
-
-- Improved cost estimation
-- Improved support for spaces with text input
-- Fixed infinite and NaN values with binary input
-- Fixed infinite values with vector addition and subtraction
-- Fixed infinite values with list centers
-- Fixed compilation error when `float8` is pass by reference
-- Fixed compilation error on PowerPC
-- Fixed segmentation fault with index creation on i386
-
-## 0.4.2 (2023-05-13)
+## 0.4.2 (unreleased)
 
 - Added notice when index created with little data
-- Fixed dimensions check for some direct function calls
 - Fixed installation error with Postgres 12.0-12.2
 
 ## 0.4.1 (2023-03-21)

Dockerfile

@@ -5,7 +5,6 @@ ARG PG_MAJOR
 COPY . /tmp/pgvector
 
 RUN apt-get update && \
-		apt-mark hold locales && \
 		apt-get install -y --no-install-recommends build-essential postgresql-server-dev-$PG_MAJOR && \
 		cd /tmp/pgvector && \
 		make clean && \
@@ -16,5 +15,4 @@ RUN apt-get update && \
 		rm -r /tmp/pgvector && \
 		apt-get remove -y build-essential postgresql-server-dev-$PG_MAJOR && \
 		apt-get autoremove -y && \
-		apt-mark unhold locales && \
 		rm -rf /var/lib/apt/lists/*

LICENSE

@@ -1,4 +1,4 @@
-Portions Copyright (c) 1996-2023, PostgreSQL Global Development Group
+Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
 
 Portions Copyright (c) 1994, The Regents of the University of California
 

META.json

@@ -2,7 +2,7 @@
 	"name": "vector",
 	"abstract": "Open-source vector similarity search for Postgres",
 	"description": "Supports L2 distance, inner product, and cosine distance",
-	"version": "0.4.4",
+	"version": "0.4.1",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.4.4",
+			"version": "0.4.1",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,9 +1,9 @@
 EXTENSION = vector
-EXTVERSION = 0.4.4
+EXTVERSION = 0.4.1
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*.sql)
-OBJS = src/hnsw.o src/hnswbuild.o src/hnswinsert.o src/hnswscan.o src/hnswutils.o src/hnswvacuum.o src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/vector.o
+OBJS = src/ivfbuild.o src/ivfflat.o src/ivfinsert.o src/ivfkmeans.o src/ivfscan.o src/ivfutils.o src/ivfvacuum.o src/vector.o
 
 TESTS = $(wildcard test/sql/*.sql)
 REGRESS = $(patsubst test/sql/%.sql,%,$(TESTS))
@@ -19,11 +19,6 @@ ifeq ($(shell uname -s), Darwin)
 	endif
 endif
 
-# PowerPC doesn't support -march=native
-ifneq ($(filter ppc64%, $(shell uname -m)), )
-	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
@@ -68,9 +63,3 @@ dist:
 
 docker:
 	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 -t ankane/pgvector:latest .
-	docker buildx build --push --platform linux/amd64,linux/arm64 -t ankane/pgvector:v$(EXTVERSION) .

Makefile.win

@@ -1,7 +1,7 @@
 EXTENSION = vector
-EXTVERSION = 0.4.4
+EXTVERSION = 0.4.1
 
-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
+OBJS = src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\vector.obj
 
 REGRESS = btree cast copy functions input ivfflat_cosine ivfflat_ip ivfflat_l2 ivfflat_options ivfflat_unlogged
 REGRESS_OPTS = --inputdir=test --load-extension=vector

README.md

@@ -2,14 +2,12 @@
 
 Open-source vector similarity search for Postgres
 
-Store all of your application data in one place. Supports:
+Supports
 
 - exact and approximate nearest neighbor search
 - L2 distance, inner product, and cosine distance
 - any [language](#languages) with a Postgres client
 
-Plus [ACID](https://en.wikipedia.org/wiki/ACID) compliance, point-in-time recovery, JOINs, and all of the other [great features](https://www.postgresql.org/about/) of Postgres
-
 [![Build Status](https://github.com/pgvector/pgvector/workflows/build/badge.svg?branch=master)](https://github.com/pgvector/pgvector/actions)
 
 ## Installation
@@ -18,24 +16,24 @@ Compile and install the extension (supports Postgres 11+)
 
 ```sh
 cd /tmp
-git clone --branch v0.4.4 https://github.com/pgvector/pgvector.git
+git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
 ```
 
-See the [installation notes](#installation-notes) if you run into issues
+Then load it in databases where you want to use it
 
-You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [APT](#apt), [Yum](#yum), or [conda-forge](#conda-forge), and it comes preinstalled with [Postgres.app](#postgresapp) and many [hosted providers](#hosted-postgres)
-
-## Getting Started
-
-Enable the extension (do this once in each database where you want to use it)
-
-```tsql
+```sql
 CREATE EXTENSION vector;
 ```
 
+See the [installation notes](#installation-notes) if you run into issues
+
+You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#pgxn), [Yum](#yum), or [conda-forge](#conda-forge)
+
+## Getting Started
+
 Create a vector column with 3 dimensions
 
 ```sql
@@ -129,7 +127,7 @@ SELECT embedding <-> '[3,1,2]' AS distance FROM items;
 
 For inner product, multiply by -1 (since `<#>` returns the negative inner product)
 
-```tsql
+```sql
 SELECT (embedding <#> '[3,1,2]') * -1 AS inner_product FROM items;
 ```
 
@@ -163,7 +161,7 @@ Three keys to achieving good recall are:
 
 1. Create the index *after* the table has some data
 2. Choose an appropriate number of lists - a good place to start is `rows / 1000` for up to 1M rows and `sqrt(rows)` for over 1M rows
-3. When querying, specify an appropriate number of [probes](#query-options) (higher is better for recall, lower is better for speed) - a good place to start is `sqrt(lists)`
+3. When querying, specify an appropriate number of [probes](#query-options) (higher is better for recall, lower is better for speed) - a good place to start is `lists / 10` for up to 1M rows and `sqrt(lists)` for over 1M rows
 
 Add an index for each distance function you want to use.
 
@@ -223,42 +221,26 @@ The phases are:
 
 Note: `tuples_done` and `tuples_total` are only populated during the `loading tuples` phase
 
-## Filtering
+### Partial Indexes
 
-There are a few ways to index nearest neighbor queries with a `WHERE` clause
+Consider [partial indexes](https://www.postgresql.org/docs/current/indexes-partial.html) for queries with a `WHERE` clause
 
 ```sql
 SELECT * FROM items WHERE category_id = 123 ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
-Create an index on one [or more](https://www.postgresql.org/docs/current/indexes-multicolumn.html) of the `WHERE` columns for exact search
+can be indexed with:
 
 ```sql
-CREATE INDEX ON items (category_id);
+CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100) WHERE (category_id = 123);
 ```
 
-Or a [partial index](https://www.postgresql.org/docs/current/indexes-partial.html) on the vector column for approximate search
-
-```sql
-CREATE INDEX ON items USING ivfflat (embedding vector_l2_ops) WITH (lists = 100)
-    WHERE (category_id = 123);
-```
-
-Use [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html) for approximate search on many different values of the `WHERE` columns
+To index many different values of `category_id`, consider [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html) on `category_id`.
 
 ```sql
 CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(category_id);
 ```
 
-## Hybrid Search
-
-Use together with Postgres [full-text search](https://www.postgresql.org/docs/current/textsearch-intro.html) for hybrid search ([Python example](https://github.com/pgvector/pgvector-python/blob/master/examples/hybrid_search.py)).
-
-```sql
-SELECT id, content FROM items, plainto_tsquery('hello search') query
-    WHERE textsearch @@ query ORDER BY ts_rank_cd(textsearch, query) DESC LIMIT 5;
-```
-
 ## Performance
 
 Use `EXPLAIN ANALYZE` to debug performance.
@@ -277,7 +259,7 @@ SET max_parallel_workers_per_gather = 4;
 
 If vectors are normalized to length 1 (like [OpenAI embeddings](https://platform.openai.com/docs/guides/embeddings/which-distance-function-should-i-use)), use inner product for best performance.
 
-```tsql
+```sql
 SELECT * FROM items ORDER BY embedding <#> '[3,1,2]' LIMIT 5;
 ```
 
@@ -298,7 +280,6 @@ Language | Libraries / Examples
 C++ | [pgvector-cpp](https://github.com/pgvector/pgvector-cpp)
 C# | [pgvector-dotnet](https://github.com/pgvector/pgvector-dotnet)
 Crystal | [pgvector-crystal](https://github.com/pgvector/pgvector-crystal)
-Dart | [pgvector-dart](https://github.com/pgvector/pgvector-dart)
 Elixir | [pgvector-elixir](https://github.com/pgvector/pgvector-elixir)
 Go | [pgvector-go](https://github.com/pgvector/pgvector-go)
 Haskell | [pgvector-haskell](https://github.com/pgvector/pgvector-haskell)
@@ -326,46 +307,10 @@ Yes, pgvector uses the write-ahead log (WAL), which allows for replication and p
 
 #### What if I want to index vectors with more than 2,000 dimensions?
 
-You’ll need to use [dimensionality reduction](https://en.wikipedia.org/wiki/Dimensionality_reduction) at the moment.
+Two things you can try are:
 
-## Troubleshooting
-
-#### Why isn’t a query using an index?
-
-The cost estimation in pgvector < 0.4.3 does not always work well with the planner. You can encourage the planner to use an index for a query with:
-
-```sql
-BEGIN;
-SET LOCAL enable_seqscan = off;
-SELECT ...
-COMMIT;
-```
-
-#### Why isn’t a query using a parallel table scan?
-
-The planner doesn’t consider [out-of-line storage](https://www.postgresql.org/docs/current/storage-toast.html) in cost estimates, which can make a serial scan look cheaper. You can reduce the cost of a parallel scan for a query with:
-
-```sql
-BEGIN;
-SET LOCAL min_parallel_table_scan_size = 1;
-SET LOCAL parallel_setup_cost = 1;
-SELECT ...
-COMMIT;
-```
-
-or choose to store vectors inline:
-
-```sql
-ALTER TABLE items ALTER COLUMN embedding SET STORAGE PLAIN;
-```
-
-#### Why are there less results for a query after adding an index?
-
-The index was likely created with too little data for the number of lists. Drop the index until the table has more data.
-
-```sql
-DROP INDEX index_name;
-```
+1. use dimensionality reduction
+2. compile Postgres with a larger block size (`./configure --with-blocksize=32`) and edit the limit in `src/ivfflat.h`
 
 ## Reference
 
@@ -379,7 +324,6 @@ Operator | Description
 --- | ---
 \+ | element-wise addition
 \- | element-wise subtraction
-\* | element-wise multiplication [unreleased]
 <-> | Euclidean distance
 <#> | negative inner product
 <=> | cosine distance
@@ -391,8 +335,6 @@ Function | Description
 cosine_distance(vector, vector) → double precision | cosine distance
 inner_product(vector, vector) → double precision | inner product
 l2_distance(vector, vector) → double precision | Euclidean distance
-l1_distance(vector, vector) → double precision | taxicab distance [unreleased]
-normalize_l2(vector) → vector | normalize with Euclidean norm [unreleased]
 vector_dims(vector) → integer | number of dimensions
 vector_norm(vector) → double precision | Euclidean norm
 
@@ -401,7 +343,6 @@ vector_norm(vector) → double precision | Euclidean norm
 Function | Description
 --- | ---
 avg(vector) → vector | arithmetic mean
-sum(vector) → vector | sum [unreleased]
 
 ## Installation Notes
 
@@ -413,11 +354,7 @@ If your machine has multiple Postgres installations, specify the path to [pg_con
 export PG_CONFIG=/Applications/Postgres.app/Contents/Versions/latest/bin/pg_config
 ```
 
-Then re-run the installation instructions (run `make clean` before `make` if needed). If `sudo` is needed for `make install`, use:
-
-```sh
-sudo --preserve-env=PG_CONFIG make install
-```
+Then re-run the installation instructions (run `make clean` before `make` if needed)
 
 ### Missing Header
 
@@ -426,26 +363,18 @@ If compilation fails with `fatal error: postgres.h: No such file or directory`,
 For Ubuntu and Debian, use:
 
 ```sh
-sudo apt install postgresql-server-dev-15
+sudo apt-get install postgresql-server-dev-15
 ```
 
 Note: Replace `15` with your Postgres server version
 
 ### Windows
 
-Support for Windows is currently experimental. Ensure [C++ support in Visual Studio](https://learn.microsoft.com/en-us/cpp/build/building-on-the-command-line?view=msvc-170#download-and-install-the-tools) is installed, and run:
-
-```cmd
-call "C:\Program Files\Microsoft Visual Studio\2022\Community\VC\Auxiliary\Build\vcvars64.bat"
-```
-
-Note: The exact path will vary depending on your Visual Studio version and edition
-
-Then use `nmake` to build:
+Support for Windows is currently experimental. Use `nmake` to build:
 
 ```cmd
 set "PGROOT=C:\Program Files\PostgreSQL\15"
-git clone --branch v0.4.4 https://github.com/pgvector/pgvector.git
+git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
@@ -466,10 +395,9 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.4.4 https://github.com/pgvector/pgvector.git
+git clone --branch v0.4.1 https://github.com/pgvector/pgvector.git
 cd pgvector
-git cherry-pick 237a6df
-docker build --build-arg PG_MAJOR=15 -t myuser/pgvector .
+docker build -t pgvector .
 ```
 
 ### Homebrew
@@ -480,8 +408,6 @@ With Homebrew Postgres, you can use:
 brew install pgvector
 ```
 
-Note: This only adds it to the `postgresql@14` formula
-
 ### PGXN
 
 Install from the [PostgreSQL Extension Network](https://pgxn.org/dist/vector) with:
@@ -490,16 +416,6 @@ Install from the [PostgreSQL Extension Network](https://pgxn.org/dist/vector) wi
 pgxn install vector
 ```
 
-### APT
-
-Debian and Ubuntu packages are available from the [PostgreSQL APT Repository](https://wiki.postgresql.org/wiki/Apt). Follow the [setup instructions](https://wiki.postgresql.org/wiki/Apt#Quickstart) and run:
-
-```sh
-sudo apt install postgresql-15-pgvector
-```
-
-Note: Replace `15` with your Postgres server version
-
 ### Yum
 
 RPM packages are available from the [PostgreSQL Yum Repository](https://yum.postgresql.org/). Follow the [setup instructions](https://www.postgresql.org/download/linux/redhat/) for your distribution and run:
@@ -522,14 +438,18 @@ conda install -c conda-forge pgvector
 
 This method is [community-maintained](https://github.com/conda-forge/pgvector-feedstock) by [@mmcauliffe](https://github.com/mmcauliffe)
 
-### Postgres.app
-
-Download the [latest release](https://postgresapp.com/downloads.html) with Postgres 15+.
-
 ## Hosted Postgres
 
 pgvector is available on [these providers](https://github.com/pgvector/pgvector/issues/54).
 
+To request a new extension on other providers:
+
+- Amazon RDS - follow the instructions on [this page](https://aws.amazon.com/rds/postgresql/faqs/)
+- Google Cloud SQL - vote or comment on [this page](https://issuetracker.google.com/issues/265172065)
+- Azure Database - vote or comment on [this page](https://feedback.azure.com/d365community/idea/7b423322-6189-ed11-a81b-000d3ae49307)
+- DigitalOcean Managed Databases - vote or comment on [this page](https://ideas.digitalocean.com/app-framework-services/p/pgvector-extension-for-postgresql)
+- Heroku Postgres - vote or comment on [this page](https://github.com/heroku/roadmap/issues/156)
+
 ## Upgrading
 
 Install the latest version and run:
@@ -570,10 +490,9 @@ Thanks to:
 
 - [PASE: PostgreSQL Ultra-High-Dimensional Approximate Nearest Neighbor Search Extension](https://dl.acm.org/doi/pdf/10.1145/3318464.3386131)
 - [Faiss: A Library for Efficient Similarity Search and Clustering of Dense Vectors](https://github.com/facebookresearch/faiss)
-- [Using the Triangle Inequality to Accelerate k-means](https://cdn.aaai.org/ICML/2003/ICML03-022.pdf)
+- [Using the Triangle Inequality to Accelerate k-means](https://www.aaai.org/Papers/ICML/2003/ICML03-022.pdf)
 - [k-means++: The Advantage of Careful Seeding](https://theory.stanford.edu/~sergei/papers/kMeansPP-soda.pdf)
 - [Concept Decompositions for Large Sparse Text Data using Clustering](https://www.cs.utexas.edu/users/inderjit/public_papers/concept_mlj.pdf)
-- [Efficient and Robust Approximate Nearest Neighbor Search using Hierarchical Navigable Small World Graphs](https://arxiv.org/ftp/arxiv/papers/1603/1603.09320.pdf)
 
 ## History
 
@@ -621,4 +540,4 @@ Resources for contributors
 
 - [Extension Building Infrastructure](https://www.postgresql.org/docs/current/extend-pgxs.html)
 - [Index Access Method Interface Definition](https://www.postgresql.org/docs/current/indexam.html)
-- [Generic WAL Records](https://www.postgresql.org/docs/current/generic-wal.html)
+- [Generic WAL Records](https://www.postgresql.org/docs/13/generic-wal.html)

sql/vector--0.4.1--0.4.2.sql

@@ -1,2 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.4.2'" to load this file. \quit

sql/vector--0.4.2--0.4.3.sql

@@ -1,2 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.4.3'" to load this file. \quit

sql/vector--0.4.3--0.4.4.sql

@@ -1,2 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.4.4'" to load this file. \quit

sql/vector--0.4.4--0.5.0.sql

@@ -1,46 +0,0 @@
--- complain if script is sourced in psql, rather than via CREATE EXTENSION
-\echo Use "ALTER EXTENSION vector UPDATE TO '0.5.0'" to load this file. \quit
-
-CREATE FUNCTION l1_distance(vector, vector) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION normalize_l2(vector) RETURNS vector
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE FUNCTION vector_mul(vector, vector) RETURNS vector
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
-CREATE OPERATOR * (
-	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_mul,
-	COMMUTATOR = *
-);
-
-CREATE AGGREGATE sum(vector) (
-	SFUNC = vector_add,
-	STYPE = vector,
-	COMBINEFUNC = vector_add,
-	PARALLEL = SAFE
-);
-
-CREATE FUNCTION hnswhandler(internal) RETURNS index_am_handler
-	AS 'MODULE_PATHNAME' LANGUAGE C;
-
-CREATE ACCESS METHOD hnsw TYPE INDEX HANDLER hnswhandler;
-
-COMMENT ON ACCESS METHOD hnsw IS 'hnsw index access method';
-
-CREATE OPERATOR CLASS vector_l2_ops
-	FOR TYPE vector USING hnsw AS
-	OPERATOR 1 <-> (vector, vector) FOR ORDER BY float_ops,
-	FUNCTION 1 vector_l2_squared_distance(vector, vector);
-
-CREATE OPERATOR CLASS vector_ip_ops
-	FOR TYPE vector USING hnsw AS
-	OPERATOR 1 <#> (vector, vector) FOR ORDER BY float_ops,
-	FUNCTION 1 vector_negative_inner_product(vector, vector);
-
-CREATE OPERATOR CLASS vector_cosine_ops
-	FOR TYPE vector USING hnsw AS
-	OPERATOR 1 <=> (vector, vector) FOR ORDER BY float_ops,
-	FUNCTION 1 vector_negative_inner_product(vector, vector),
-	FUNCTION 2 vector_norm(vector);

sql/vector.sql

@@ -40,27 +40,18 @@ CREATE FUNCTION inner_product(vector, vector) RETURNS float8
 CREATE FUNCTION cosine_distance(vector, vector) RETURNS float8
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
-CREATE FUNCTION l1_distance(vector, vector) RETURNS float8
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
 CREATE FUNCTION vector_dims(vector) RETURNS integer
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
 CREATE FUNCTION vector_norm(vector) RETURNS float8
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
-CREATE FUNCTION normalize_l2(vector) RETURNS vector
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
 CREATE FUNCTION vector_add(vector, vector) RETURNS vector
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
 CREATE FUNCTION vector_sub(vector, vector) RETURNS vector
 	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
 
-CREATE FUNCTION vector_mul(vector, vector) RETURNS vector
-	AS 'MODULE_PATHNAME' LANGUAGE C IMMUTABLE STRICT PARALLEL SAFE;
-
 -- private functions
 
 CREATE FUNCTION vector_lt(vector, vector) RETURNS bool
@@ -113,13 +104,6 @@ CREATE AGGREGATE avg(vector) (
 	PARALLEL = SAFE
 );
 
-CREATE AGGREGATE sum(vector) (
-	SFUNC = vector_add,
-	STYPE = vector,
-	COMBINEFUNC = vector_add,
-	PARALLEL = SAFE
-);
-
 -- cast functions
 
 CREATE FUNCTION vector(vector, integer, boolean) RETURNS vector
@@ -187,11 +171,6 @@ CREATE OPERATOR - (
 	COMMUTATOR = -
 );
 
-CREATE OPERATOR * (
-	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_mul,
-	COMMUTATOR = *
-);
-
 CREATE OPERATOR < (
 	LEFTARG = vector, RIGHTARG = vector, PROCEDURE = vector_lt,
 	COMMUTATOR = > , NEGATOR = >= ,
@@ -230,7 +209,7 @@ CREATE OPERATOR > (
 	RESTRICT = scalargtsel, JOIN = scalargtjoinsel
 );
 
--- access methods
+-- access method
 
 CREATE FUNCTION ivfflathandler(internal) RETURNS index_am_handler
 	AS 'MODULE_PATHNAME' LANGUAGE C;
@@ -239,13 +218,6 @@ CREATE ACCESS METHOD ivfflat TYPE INDEX HANDLER ivfflathandler;
 
 COMMENT ON ACCESS METHOD ivfflat IS 'ivfflat index access method';
 
-CREATE FUNCTION hnswhandler(internal) RETURNS index_am_handler
-	AS 'MODULE_PATHNAME' LANGUAGE C;
-
-CREATE ACCESS METHOD hnsw TYPE INDEX HANDLER hnswhandler;
-
-COMMENT ON ACCESS METHOD hnsw IS 'hnsw index access method';
-
 -- opclasses
 
 CREATE OPERATOR CLASS vector_ops
@@ -277,19 +249,3 @@ CREATE OPERATOR CLASS vector_cosine_ops
 	FUNCTION 2 vector_norm(vector),
 	FUNCTION 3 vector_spherical_distance(vector, vector),
 	FUNCTION 4 vector_norm(vector);
-
-CREATE OPERATOR CLASS vector_l2_ops
-	FOR TYPE vector USING hnsw AS
-	OPERATOR 1 <-> (vector, vector) FOR ORDER BY float_ops,
-	FUNCTION 1 vector_l2_squared_distance(vector, vector);
-
-CREATE OPERATOR CLASS vector_ip_ops
-	FOR TYPE vector USING hnsw AS
-	OPERATOR 1 <#> (vector, vector) FOR ORDER BY float_ops,
-	FUNCTION 1 vector_negative_inner_product(vector, vector);
-
-CREATE OPERATOR CLASS vector_cosine_ops
-	FOR TYPE vector USING hnsw AS
-	OPERATOR 1 <=> (vector, vector) FOR ORDER BY float_ops,
-	FUNCTION 1 vector_negative_inner_product(vector, vector),
-	FUNCTION 2 vector_norm(vector);

src/hnsw.c

@@ -1,224 +0,0 @@
-#include "postgres.h"
-
-#include <float.h>
-#include <math.h>
-
-#include "access/amapi.h"
-#include "commands/vacuum.h"
-#include "hnsw.h"
-#include "utils/guc.h"
-#include "utils/selfuncs.h"
-
-#if PG_VERSION_NUM >= 120000
-#include "commands/progress.h"
-#endif
-
-int			hnsw_ef_search;
-static relopt_kind hnsw_relopt_kind;
-
-/*
- * Initialize index options and variables
- */
-void
-HnswInit(void)
-{
-	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
-#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
-#if PG_VERSION_NUM >= 130000
-					  ,AccessExclusiveLock
-#endif
-		);
-
-	DefineCustomIntVariable("hnsw.ef_search", "Sets the size of the dynamic candidate list for search",
-							"Valid range is 10..1000.", &hnsw_ef_search,
-							HNSW_DEFAULT_EF_SEARCH, HNSW_MIN_EF_SEARCH, HNSW_MAX_EF_SEARCH, PGC_USERSET, 0, NULL, NULL, NULL);
-}
-
-/*
- * Get the name of index build phase
- */
-#if PG_VERSION_NUM >= 120000
-static char *
-hnswbuildphasename(int64 phasenum)
-{
-	switch (phasenum)
-	{
-		case PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE:
-			return "initializing";
-		case PROGRESS_HNSW_PHASE_LOAD:
-			return "loading tuples";
-		default:
-			return NULL;
-	}
-}
-#endif
-
-/*
- * Estimate the cost of an index scan
- */
-static void
-hnswcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
-				 Cost *indexStartupCost, Cost *indexTotalCost,
-				 Selectivity *indexSelectivity, double *indexCorrelation,
-				 double *indexPages)
-{
-	GenericCosts costs;
-	int			m;
-	int			entryLevel;
-	Relation	index;
-#if PG_VERSION_NUM < 120000
-	List	   *qinfos;
-#endif
-
-	/* Never use index without order */
-	if (path->indexorderbys == NULL)
-	{
-		*indexStartupCost = DBL_MAX;
-		*indexTotalCost = DBL_MAX;
-		*indexSelectivity = 0;
-		*indexCorrelation = 0;
-		*indexPages = 0;
-		return;
-	}
-
-	MemSet(&costs, 0, sizeof(costs));
-
-	index = index_open(path->indexinfo->indexoid, NoLock);
-	m = HnswGetM(index);
-	index_close(index, NoLock);
-
-	/* Approximate entry level */
-	entryLevel = (int) -log(1.0 / path->indexinfo->tuples) * HnswGetMl(m);
-
-	/* 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 PG_VERSION_NUM >= 120000
-	genericcostestimate(root, path, loop_count, &costs);
-#else
-	qinfos = deconstruct_indexquals(path);
-	genericcostestimate(root, path, loop_count, qinfos, &costs);
-#endif
-
-	/* Use total cost since most work happens before first tuple is returned */
-	*indexStartupCost = costs.indexTotalCost;
-	*indexTotalCost = costs.indexTotalCost;
-	*indexSelectivity = costs.indexSelectivity;
-	*indexCorrelation = costs.indexCorrelation;
-	*indexPages = costs.numIndexPages;
-}
-
-/*
- * Parse and validate the reloptions
- */
-static bytea *
-hnswoptions(Datum reloptions, bool validate)
-{
-	static const relopt_parse_elt tab[] = {
-		{"m", RELOPT_TYPE_INT, offsetof(HnswOptions, m)},
-		{"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
-}
-
-/*
- * Validate catalog entries for the specified operator class
- */
-static bool
-hnswvalidate(Oid opclassoid)
-{
-	return true;
-}
-
-/*
- * Define index handler
- *
- * See https://www.postgresql.org/docs/current/index-api.html
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(hnswhandler);
-Datum
-hnswhandler(PG_FUNCTION_ARGS)
-{
-	IndexAmRoutine *amroutine = makeNode(IndexAmRoutine);
-
-	amroutine->amstrategies = 0;
-	amroutine->amsupport = 2;
-#if PG_VERSION_NUM >= 130000
-	amroutine->amoptsprocnum = 0;
-#endif
-	amroutine->amcanorder = false;
-	amroutine->amcanorderbyop = true;
-	amroutine->amcanbackward = false;	/* can change direction mid-scan */
-	amroutine->amcanunique = false;
-	amroutine->amcanmulticol = false;
-	amroutine->amoptionalkey = true;
-	amroutine->amsearcharray = false;
-	amroutine->amsearchnulls = false;
-	amroutine->amstorage = false;
-	amroutine->amclusterable = false;
-	amroutine->ampredlocks = false;
-	amroutine->amcanparallel = false;
-	amroutine->amcaninclude = false;
-#if PG_VERSION_NUM >= 130000
-	amroutine->amusemaintenanceworkmem = false; /* not used during VACUUM */
-	amroutine->amparallelvacuumoptions = VACUUM_OPTION_PARALLEL_BULKDEL;
-#endif
-	amroutine->amkeytype = InvalidOid;
-
-	/* Interface functions */
-	amroutine->ambuild = hnswbuild;
-	amroutine->ambuildempty = hnswbuildempty;
-	amroutine->aminsert = hnswinsert;
-	amroutine->ambulkdelete = hnswbulkdelete;
-	amroutine->amvacuumcleanup = hnswvacuumcleanup;
-	amroutine->amcanreturn = NULL;	/* tuple not included in heapsort */
-	amroutine->amcostestimate = hnswcostestimate;
-	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;
-#endif
-	amroutine->ambeginscan = hnswbeginscan;
-	amroutine->amrescan = hnswrescan;
-	amroutine->amgettuple = hnswgettuple;
-	amroutine->amgetbitmap = NULL;
-	amroutine->amendscan = hnswendscan;
-	amroutine->ammarkpos = NULL;
-	amroutine->amrestrpos = NULL;
-
-	/* Interface functions to support parallel index scans */
-	amroutine->amestimateparallelscan = NULL;
-	amroutine->aminitparallelscan = NULL;
-	amroutine->amparallelrescan = NULL;
-
-	PG_RETURN_POINTER(amroutine);
-}

src/hnsw.h

@@ -1,294 +0,0 @@
-#ifndef HNSW_H
-#define HNSW_H
-
-#include "postgres.h"
-
-#include "access/generic_xlog.h"
-#include "access/reloptions.h"
-#include "nodes/execnodes.h"
-#include "port.h"				/* for random() */
-#include "utils/sampling.h"
-#include "vector.h"
-
-#if PG_VERSION_NUM < 110000
-#error "Requires PostgreSQL 11+"
-#endif
-
-#define HNSW_MAX_DIM 2000
-
-/* Support functions */
-#define HNSW_DISTANCE_PROC 1
-#define HNSW_NORM_PROC 2
-
-#define HNSW_VERSION	1
-#define HNSW_MAGIC_NUMBER 0xA953A953
-#define HNSW_PAGE_ID	0xFF85
-
-/* Preserved page numbers */
-#define HNSW_METAPAGE_BLKNO	0
-#define HNSW_HEAD_BLKNO		1	/* first element page */
-
-#define HNSW_DEFAULT_M	16
-#define HNSW_MIN_M	4
-#define HNSW_MAX_M		100
-#define HNSW_DEFAULT_EF_CONSTRUCTION	40
-#define HNSW_MIN_EF_CONSTRUCTION	10
-#define HNSW_MAX_EF_CONSTRUCTION		1000
-#define HNSW_DEFAULT_EF_SEARCH	40
-#define HNSW_MIN_EF_SEARCH		10
-#define HNSW_MAX_EF_SEARCH		1000
-
-#define HNSW_ELEMENT_TUPLE_TYPE  1
-#define HNSW_NEIGHBOR_TUPLE_TYPE 2
-
-/* Make graph robust against non-HOT updates */
-#define HNSW_HEAPTIDS 10
-
-/* Build phases */
-/* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
-#define PROGRESS_HNSW_PHASE_LOAD		2
-
-#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 HnswPageGetOpaque(page)	((HnswPageOpaque) PageGetSpecialPointer(page))
-#define HnswPageGetMeta(page)	((HnswMetaPageData *) PageGetContents(page))
-
-#if PG_VERSION_NUM >= 150000
-#define RandomDouble() pg_prng_double(&pg_global_prng_state)
-#else
-#define RandomDouble() (((double) random()) / MAX_RANDOM_VALUE)
-#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)
-#define HnswIsNeighborTuple(tup) ((tup)->type == HNSW_NEIGHBOR_TUPLE_TYPE)
-
-#define HnswGetLayerM(m, layer) (layer == 0 ? m * 2 : m)
-#define HnswGetMl(m) (1 / log(m))
-
-/* Ensure fits in uint8 */
-#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;
-
-typedef struct HnswNeighborArray HnswNeighborArray;
-
-typedef struct HnswElementData
-{
-	List	   *heaptids;
-	uint8		level;
-	uint8		deleted;
-	HnswNeighborArray *neighbors;
-	BlockNumber blkno;
-	OffsetNumber offno;
-	OffsetNumber neighborOffno;
-	BlockNumber neighborPage;
-	Vector	   *vec;
-}			HnswElementData;
-
-typedef HnswElementData * HnswElement;
-
-typedef struct HnswCandidate
-{
-	HnswElement element;
-	float		distance;
-}			HnswCandidate;
-
-typedef struct HnswNeighborArray
-{
-	int			length;
-	HnswCandidate *items;
-}			HnswNeighborArray;
-
-typedef struct HnswUpdate
-{
-	HnswCandidate hc;
-	int			level;
-	int			index;
-}			HnswUpdate;
-
-typedef struct HnswPairingHeapNode
-{
-	pairingheap_node ph_node;
-	HnswCandidate *inner;
-}			HnswPairingHeapNode;
-
-/* HNSW index options */
-typedef struct HnswOptions
-{
-	int32		vl_len_;		/* varlena header (do not touch directly!) */
-	int			m;				/* number of connections */
-	int			efConstruction; /* size of dynamic candidate list */
-}			HnswOptions;
-
-typedef struct HnswBuildState
-{
-	/* Info */
-	Relation	heap;
-	Relation	index;
-	IndexInfo  *indexInfo;
-	ForkNumber	forkNum;
-
-	/* Settings */
-	int			dimensions;
-	int			m;
-	int			efConstruction;
-
-	/* Statistics */
-	double		indtuples;
-	double		reltuples;
-
-	/* Support functions */
-	FmgrInfo   *procinfo;
-	FmgrInfo   *normprocinfo;
-	Oid			collation;
-
-	/* Variables */
-	List	   *elements;
-	HnswElement entryPoint;
-	double		ml;
-	int			maxLevel;
-	double		maxInMemoryElements;
-	bool		flushed;
-	Vector	   *normvec;
-
-	/* Memory */
-	MemoryContext tmpCtx;
-}			HnswBuildState;
-
-typedef struct HnswMetaPageData
-{
-	uint32		magicNumber;
-	uint32		version;
-	uint32		dimensions;
-	uint16		m;
-	uint16		efConstruction;
-	BlockNumber entryBlkno;
-	OffsetNumber entryOffno;
-	int16		entryLevel;
-	BlockNumber insertPage;
-}			HnswMetaPageData;
-
-typedef HnswMetaPageData * HnswMetaPage;
-
-typedef struct HnswPageOpaqueData
-{
-	BlockNumber nextblkno;
-	uint16		unused;
-	uint16		page_id;		/* for identification of HNSW indexes */
-}			HnswPageOpaqueData;
-
-typedef HnswPageOpaqueData * HnswPageOpaque;
-
-typedef struct HnswElementTupleData
-{
-	uint8		type;
-	uint8		level;
-	uint8		deleted;
-	uint8		unused;
-	ItemPointerData heaptids[HNSW_HEAPTIDS];
-	ItemPointerData neighbortid;
-	uint16		unused2;
-	Vector		vec;
-}			HnswElementTupleData;
-
-typedef HnswElementTupleData * HnswElementTuple;
-
-typedef struct HnswNeighborTupleData
-{
-	uint8		type;
-	uint8		unused;
-	uint16		count;
-	ItemPointerData indextids[FLEXIBLE_ARRAY_MEMBER];
-}			HnswNeighborTupleData;
-
-typedef HnswNeighborTupleData * HnswNeighborTuple;
-
-typedef struct HnswScanOpaqueData
-{
-	bool		first;
-	Buffer		buf;
-	List	   *w;
-	MemoryContext tmpCtx;
-
-	/* Support functions */
-	FmgrInfo   *procinfo;
-	FmgrInfo   *normprocinfo;
-	Oid			collation;
-}			HnswScanOpaqueData;
-
-typedef HnswScanOpaqueData * HnswScanOpaque;
-
-typedef struct HnswVacuumState
-{
-	/* Info */
-	Relation	index;
-	IndexBulkDeleteResult *stats;
-	IndexBulkDeleteCallback callback;
-	void	   *callback_state;
-
-	/* Settings */
-	int			m;
-	int			efConstruction;
-
-	/* Support functions */
-	FmgrInfo   *procinfo;
-	Oid			collation;
-
-	/* Variables */
-	HTAB	   *deleted;
-	BufferAccessStrategy bas;
-	HnswNeighborTuple ntup;
-	HnswElementData highestPoint;
-
-	/* Memory */
-	MemoryContext tmpCtx;
-}			HnswVacuumState;
-
-/* Methods */
-int			HnswGetM(Relation index);
-int			HnswGetEfConstruction(Relation index);
-FmgrInfo   *HnswOptionalProcInfo(Relation rel, uint16 procnum);
-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(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, bool inserting, BlockNumber *skipPage, OffsetNumber *skipOffno);
-HnswElement HnswGetEntryPoint(Relation index);
-HnswElement HnswInitElement(ItemPointer tid, int m, double ml, int maxLevel);
-void		HnswFreeElement(HnswElement element);
-HnswElement HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, List **updates, bool vacuuming);
-HnswCandidate *HnswEntryCandidate(HnswElement em, Datum q, Relation rel, FmgrInfo *procinfo, Oid collation, bool loadvec);
-void		HnswUpdateMetaPage(Relation index, bool updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum);
-void		HnswSetNeighborTuple(HnswNeighborTuple ntup, HnswElement e, int m);
-void		HnswAddHeapTid(HnswElement element, ItemPointer heaptid);
-void		HnswInitNeighbors(HnswElement element, int m);
-bool		HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel);
-void		HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec);
-void		HnswSetElementTuple(HnswElementTuple etup, HnswElement element);
-
-/* Index access methods */
-IndexBuildResult *hnswbuild(Relation heap, Relation index, IndexInfo *indexInfo);
-void		hnswbuildempty(Relation index);
-bool		hnswinsert(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heap, IndexUniqueCheck checkUnique
-#if PG_VERSION_NUM >= 140000
-					   ,bool indexUnchanged
-#endif
-					   ,IndexInfo *indexInfo
-);
-IndexBulkDeleteResult *hnswbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats, IndexBulkDeleteCallback callback, void *callback_state);
-IndexBulkDeleteResult *hnswvacuumcleanup(IndexVacuumInfo *info, IndexBulkDeleteResult *stats);
-IndexScanDesc hnswbeginscan(Relation index, int nkeys, int norderbys);
-void		hnswrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int norderbys);
-bool		hnswgettuple(IndexScanDesc scan, ScanDirection dir);
-void		hnswendscan(IndexScanDesc scan);
-
-#endif

src/hnswbuild.c

@@ -1,506 +0,0 @@
-#include "postgres.h"
-
-#include <math.h>
-
-#include "catalog/index.h"
-#include "hnsw.h"
-#include "miscadmin.h"
-#include "lib/pairingheap.h"
-#include "nodes/pg_list.h"
-#include "storage/bufmgr.h"
-#include "utils/memutils.h"
-
-#if PG_VERSION_NUM >= 140000
-#include "utils/backend_progress.h"
-#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_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
-
-/*
- * Create the metapage
- */
-static void
-CreateMetaPage(HnswBuildState * buildstate)
-{
-	Relation	index = buildstate->index;
-	ForkNumber	forkNum = buildstate->forkNum;
-	Buffer		buf;
-	Page		page;
-	GenericXLogState *state;
-	HnswMetaPage metap;
-
-	buf = HnswNewBuffer(index, forkNum);
-	HnswInitRegisterPage(index, &buf, &page, &state);
-
-	/* Set metapage data */
-	metap = HnswPageGetMeta(page);
-	metap->magicNumber = HNSW_MAGIC_NUMBER;
-	metap->version = HNSW_VERSION;
-	metap->dimensions = buildstate->dimensions;
-	metap->m = buildstate->m;
-	metap->efConstruction = buildstate->efConstruction;
-	metap->entryBlkno = InvalidBlockNumber;
-	metap->entryOffno = InvalidOffsetNumber;
-	metap->insertPage = InvalidBlockNumber;
-	((PageHeader) page)->pd_lower =
-		((char *) metap + sizeof(HnswMetaPageData)) - (char *) page;
-
-	HnswCommitBuffer(buf, state);
-}
-
-/*
- * Add a new page
- */
-static void
-HnswBuildAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state, ForkNumber forkNum)
-{
-	/* Add a new page */
-	Buffer		newbuf = HnswNewBuffer(index, forkNum);
-
-	/* Update previous page */
-	HnswPageGetOpaque(*page)->nextblkno = BufferGetBlockNumber(newbuf);
-
-	/* Commit */
-	MarkBufferDirty(*buf);
-	GenericXLogFinish(*state);
-	UnlockReleaseBuffer(*buf);
-
-	/* Can take a while, so ensure we can interrupt */
-	/* Needs to be called when no buffer locks are held */
-	LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
-	CHECK_FOR_INTERRUPTS();
-	LockBuffer(newbuf, BUFFER_LOCK_EXCLUSIVE);
-
-	/* Prepare new page */
-	*buf = newbuf;
-	*state = GenericXLogStart(index);
-	*page = GenericXLogRegisterBuffer(*state, *buf, GENERIC_XLOG_FULL_IMAGE);
-	HnswInitPage(*buf, *page);
-}
-
-/*
- * Create element pages
- */
-static void
-CreateElementPages(HnswBuildState * buildstate)
-{
-	Relation	index = buildstate->index;
-	ForkNumber	forkNum = buildstate->forkNum;
-	int			dimensions = buildstate->dimensions;
-	Size		etupSize;
-	Size		maxSize;
-	HnswElementTuple etup;
-	HnswNeighborTuple ntup;
-	BlockNumber insertPage;
-	Buffer		buf;
-	Page		page;
-	GenericXLogState *state;
-	ListCell   *lc;
-
-	/* Calculate sizes */
-	maxSize = BLCKSZ - MAXALIGN(SizeOfPageHeaderData) - MAXALIGN(sizeof(HnswPageOpaqueData));
-	etupSize = HNSW_ELEMENT_TUPLE_SIZE(dimensions);
-
-	/* Allocate once */
-	etup = palloc0(etupSize);
-	ntup = palloc0(maxSize);
-
-	/* Prepare first page */
-	buf = HnswNewBuffer(index, forkNum);
-	state = GenericXLogStart(index);
-	page = GenericXLogRegisterBuffer(state, buf, GENERIC_XLOG_FULL_IMAGE);
-	HnswInitPage(buf, page);
-
-	foreach(lc, buildstate->elements)
-	{
-		HnswElement element = lfirst(lc);
-		Size		ntupSize;
-		Size		combinedSize;
-
-		HnswSetElementTuple(etup, element);
-
-		/* Calculate sizes */
-		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, buildstate->m);
-		combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
-
-		/* Keep element and neighbors on the same page if possible */
-		if (PageGetFreeSpace(page) < etupSize || (combinedSize <= maxSize && PageGetFreeSpace(page) < combinedSize))
-			HnswBuildAppendPage(index, &buf, &page, &state, forkNum);
-
-		/* Calculate offsets */
-		element->blkno = BufferGetBlockNumber(buf);
-		element->offno = OffsetNumberNext(PageGetMaxOffsetNumber(page));
-		if (combinedSize <= maxSize)
-		{
-			element->neighborPage = element->blkno;
-			element->neighborOffno = OffsetNumberNext(element->offno);
-		}
-		else
-		{
-			element->neighborPage = element->blkno + 1;
-			element->neighborOffno = FirstOffsetNumber;
-		}
-
-		ItemPointerSet(&etup->neighbortid, element->neighborPage, element->neighborOffno);
-
-		/* Add element */
-		if (PageAddItem(page, (Item) etup, etupSize, InvalidOffsetNumber, false, false) != element->offno)
-			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-
-		/* Add new page if needed */
-		if (PageGetFreeSpace(page) < ntupSize)
-			HnswBuildAppendPage(index, &buf, &page, &state, forkNum);
-
-		/* Add placeholder for neighbors */
-		if (PageAddItem(page, (Item) ntup, ntupSize, InvalidOffsetNumber, false, false) != element->neighborOffno)
-			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-	}
-
-	insertPage = BufferGetBlockNumber(buf);
-
-	/* Commit */
-	MarkBufferDirty(buf);
-	GenericXLogFinish(state);
-	UnlockReleaseBuffer(buf);
-
-	HnswUpdateMetaPage(index, true, buildstate->entryPoint, insertPage, forkNum);
-
-	pfree(etup);
-	pfree(ntup);
-}
-
-/*
- * Create neighbor pages
- */
-static void
-CreateNeighborPages(HnswBuildState * buildstate)
-{
-	Relation	index = buildstate->index;
-	ForkNumber	forkNum = buildstate->forkNum;
-	int			m = buildstate->m;
-	ListCell   *lc;
-	HnswNeighborTuple ntup;
-
-	/* Allocate once */
-	ntup = palloc0(BLCKSZ);
-
-	foreach(lc, buildstate->elements)
-	{
-		HnswElement e = lfirst(lc);
-		Buffer		buf;
-		Page		page;
-		GenericXLogState *state;
-		Size		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m);
-
-		/* Can take a while, so ensure we can interrupt */
-		/* Needs to be called when no buffer locks are held */
-		CHECK_FOR_INTERRUPTS();
-
-		buf = ReadBufferExtended(index, forkNum, e->neighborPage, RBM_NORMAL, NULL);
-		LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-		state = GenericXLogStart(index);
-		page = GenericXLogRegisterBuffer(state, buf, 0);
-
-		HnswSetNeighborTuple(ntup, e, m);
-
-		if (!PageIndexTupleOverwrite(page, e->neighborOffno, (Item) ntup, ntupSize))
-			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-
-		/* Commit */
-		MarkBufferDirty(buf);
-		GenericXLogFinish(state);
-		UnlockReleaseBuffer(buf);
-	}
-
-	pfree(ntup);
-}
-
-/*
- * Free elements
- */
-static void
-FreeElements(HnswBuildState * buildstate)
-{
-	ListCell   *lc;
-
-	foreach(lc, buildstate->elements)
-		HnswFreeElement(lfirst(lc));
-
-	list_free(buildstate->elements);
-}
-
-/*
- * Flush pages
- */
-static void
-FlushPages(HnswBuildState * buildstate)
-{
-	CreateMetaPage(buildstate);
-	CreateElementPages(buildstate);
-	CreateNeighborPages(buildstate);
-
-	buildstate->flushed = true;
-	FreeElements(buildstate);
-}
-
-/*
- * Insert tuple
- */
-static bool
-InsertTuple(Relation index, Datum *values, HnswElement element, HnswBuildState * buildstate, HnswElement * dup)
-{
-	FmgrInfo   *procinfo = buildstate->procinfo;
-	Oid			collation = buildstate->collation;
-	HnswElement entryPoint = buildstate->entryPoint;
-	int			efConstruction = buildstate->efConstruction;
-	int			m = buildstate->m;
-
-	/* Detoast once for all calls */
-	Datum		value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
-
-	/* Normalize if needed */
-	if (buildstate->normprocinfo != NULL)
-	{
-		if (!HnswNormValue(buildstate->normprocinfo, collation, &value, buildstate->normvec))
-			return false;
-	}
-
-	/* Copy value to element so accessible outside of memory context */
-	memcpy(element->vec, DatumGetVector(value), VECTOR_SIZE(buildstate->dimensions));
-
-	/* Insert element in graph */
-	*dup = HnswInsertElement(element, entryPoint, NULL, procinfo, collation, m, efConstruction, NULL, false);
-
-	/* Update entry point if needed */
-	if (*dup == NULL && (entryPoint == NULL || element->level > entryPoint->level))
-		buildstate->entryPoint = element;
-
-	UpdateProgress(PROGRESS_CREATEIDX_TUPLES_DONE, ++buildstate->indtuples);
-
-	return *dup == NULL;
-}
-
-/*
- * Callback for table_index_build_scan
- */
-static void
-BuildCallback(Relation index, CALLBACK_ITEM_POINTER, Datum *values,
-			  bool *isnull, bool tupleIsAlive, void *state)
-{
-	HnswBuildState *buildstate = (HnswBuildState *) state;
-	MemoryContext oldCtx;
-	HnswElement element;
-	HnswElement dup = NULL;
-	bool		inserted;
-
-#if PG_VERSION_NUM < 130000
-	ItemPointer tid = &hup->t_self;
-#endif
-
-	/* Skip nulls */
-	if (isnull[0])
-		return;
-
-	if (buildstate->indtuples >= buildstate->maxInMemoryElements)
-	{
-		if (!buildstate->flushed)
-		{
-			ereport(NOTICE,
-					(errmsg("hnsw graph no longer fits into maintenance_work_mem after " INT64_FORMAT " tuples", (int64) buildstate->indtuples),
-					 errdetail("Building will take significantly more time."),
-					 errhint("Increase maintenance_work_mem to speed up builds.")));
-
-			FlushPages(buildstate);
-		}
-
-		oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
-
-		if (HnswInsertTuple(buildstate->index, values, isnull, tid, buildstate->heap))
-			UpdateProgress(PROGRESS_CREATEIDX_TUPLES_DONE, ++buildstate->indtuples);
-
-		/* Reset memory context */
-		MemoryContextSwitchTo(oldCtx);
-		MemoryContextReset(buildstate->tmpCtx);
-
-		return;
-	}
-
-	/* Allocate necessary memory outside of memory context */
-	element = HnswInitElement(tid, buildstate->m, buildstate->ml, buildstate->maxLevel);
-	element->vec = palloc(VECTOR_SIZE(buildstate->dimensions));
-
-	/* Use memory context since detoast can allocate */
-	oldCtx = MemoryContextSwitchTo(buildstate->tmpCtx);
-
-	/* Insert tuple */
-	inserted = InsertTuple(index, values, element, buildstate, &dup);
-
-	/* Reset memory context */
-	MemoryContextSwitchTo(oldCtx);
-	MemoryContextReset(buildstate->tmpCtx);
-
-	/* Add outside memory context */
-	if (dup != NULL)
-		HnswAddHeapTid(dup, tid);
-
-	/* Add to buildstate or free */
-	if (inserted)
-		buildstate->elements = lappend(buildstate->elements, element);
-	else
-		HnswFreeElement(element);
-}
-
-/*
- * Get the max number of elements that fit into maintenance_work_mem
- */
-static double
-HnswGetMaxInMemoryElements(int m, double ml, int dimensions)
-{
-	Size		elementSize = sizeof(HnswElementData);
-	double		avgLevel = -log(0.5) * ml;
-
-	elementSize += sizeof(HnswNeighborArray) * (avgLevel + 1);
-	elementSize += sizeof(HnswCandidate) * (m * (avgLevel + 2));
-	elementSize += sizeof(ItemPointerData);
-	elementSize += VECTOR_SIZE(dimensions);
-	return (maintenance_work_mem * 1024L) / elementSize;
-}
-
-/*
- * Initialize the build state
- */
-static void
-InitBuildState(HnswBuildState * buildstate, Relation heap, Relation index, IndexInfo *indexInfo, ForkNumber forkNum)
-{
-	buildstate->heap = heap;
-	buildstate->index = index;
-	buildstate->indexInfo = indexInfo;
-	buildstate->forkNum = forkNum;
-
-	buildstate->m = HnswGetM(index);
-	buildstate->efConstruction = HnswGetEfConstruction(index);
-	buildstate->dimensions = TupleDescAttr(index->rd_att, 0)->atttypmod;
-
-	/* Require column to have dimensions to be indexed */
-	if (buildstate->dimensions < 0)
-		elog(ERROR, "column does not have dimensions");
-
-	if (buildstate->dimensions > HNSW_MAX_DIM)
-		elog(ERROR, "column cannot have more than %d dimensions for hnsw index", HNSW_MAX_DIM);
-
-	buildstate->reltuples = 0;
-	buildstate->indtuples = 0;
-
-	/* Get support functions */
-	buildstate->procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
-	buildstate->normprocinfo = HnswOptionalProcInfo(index, HNSW_NORM_PROC);
-	buildstate->collation = index->rd_indcollation[0];
-
-	buildstate->elements = NIL;
-	buildstate->entryPoint = NULL;
-	buildstate->ml = HnswGetMl(buildstate->m);
-	buildstate->maxLevel = HnswGetMaxLevel(buildstate->m);
-	buildstate->maxInMemoryElements = HnswGetMaxInMemoryElements(buildstate->m, buildstate->ml, buildstate->dimensions);
-	buildstate->flushed = false;
-
-	/* Reuse for each tuple */
-	buildstate->normvec = InitVector(buildstate->dimensions);
-
-	buildstate->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
-											   "Hnsw build temporary context",
-											   ALLOCSET_DEFAULT_SIZES);
-}
-
-/*
- * Free resources
- */
-static void
-FreeBuildState(HnswBuildState * buildstate)
-{
-	pfree(buildstate->normvec);
-	MemoryContextDelete(buildstate->tmpCtx);
-}
-
-/*
- * Build graph
- */
-static void
-BuildGraph(HnswBuildState * buildstate, ForkNumber forkNum)
-{
-	UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_HNSW_PHASE_LOAD);
-
-#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
-}
-
-/*
- * Build the index
- */
-static void
-BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
-		   HnswBuildState * buildstate, ForkNumber forkNum)
-{
-	InitBuildState(buildstate, heap, index, indexInfo, forkNum);
-
-	if (buildstate->heap != NULL)
-		BuildGraph(buildstate, forkNum);
-
-	if (!buildstate->flushed)
-		FlushPages(buildstate);
-
-	FreeBuildState(buildstate);
-}
-
-/*
- * Build the index for a logged table
- */
-IndexBuildResult *
-hnswbuild(Relation heap, Relation index, IndexInfo *indexInfo)
-{
-	IndexBuildResult *result;
-	HnswBuildState buildstate;
-
-	BuildIndex(heap, index, indexInfo, &buildstate, MAIN_FORKNUM);
-
-	result = (IndexBuildResult *) palloc(sizeof(IndexBuildResult));
-	result->heap_tuples = buildstate.reltuples;
-	result->index_tuples = buildstate.indtuples;
-
-	return result;
-}
-
-/*
- * Build the index for an unlogged table
- */
-void
-hnswbuildempty(Relation index)
-{
-	IndexInfo  *indexInfo = BuildIndexInfo(index);
-	HnswBuildState buildstate;
-
-	BuildIndex(NULL, index, indexInfo, &buildstate, INIT_FORKNUM);
-}

src/hnswinsert.c

@@ -1,490 +0,0 @@
-#include "postgres.h"
-
-#include <math.h>
-
-#include "hnsw.h"
-#include "storage/bufmgr.h"
-#include "storage/lmgr.h"
-#include "utils/memutils.h"
-
-/*
- * Get the insert page
- */
-static BlockNumber
-GetInsertPage(Relation index)
-{
-	Buffer		buf;
-	Page		page;
-	HnswMetaPage metap;
-	BlockNumber insertPage;
-
-	buf = ReadBuffer(index, HNSW_METAPAGE_BLKNO);
-	LockBuffer(buf, BUFFER_LOCK_SHARE);
-	page = BufferGetPage(buf);
-	metap = HnswPageGetMeta(page);
-
-	insertPage = metap->insertPage;
-
-	UnlockReleaseBuffer(buf);
-
-	return insertPage;
-}
-
-/*
- * Check for a free offset
- */
-static bool
-HnswFreeOffset(Relation index, Buffer buf, Page page, HnswElement element, Size ntupSize, Buffer *nbuf, Page *npage, OffsetNumber *freeOffno, OffsetNumber *freeNeighborOffno, BlockNumber *firstFreePage)
-{
-	OffsetNumber offno;
-	OffsetNumber maxoffno = PageGetMaxOffsetNumber(page);
-
-	for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
-	{
-		HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
-
-		/* Skip neighbor tuples */
-		if (!HnswIsElementTuple(etup))
-			continue;
-
-		if (etup->deleted)
-		{
-			BlockNumber neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
-			OffsetNumber neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
-			ItemId		itemid;
-
-			if (!BlockNumberIsValid(*firstFreePage))
-				*firstFreePage = neighborPage;
-
-			if (neighborPage == BufferGetBlockNumber(buf))
-			{
-				*nbuf = buf;
-				*npage = page;
-			}
-			else
-			{
-				*nbuf = ReadBuffer(index, neighborPage);
-				LockBuffer(*nbuf, BUFFER_LOCK_EXCLUSIVE);
-
-				/* Skip WAL for now */
-				*npage = BufferGetPage(*nbuf);
-			}
-
-			itemid = PageGetItemId(*npage, neighborOffno);
-
-			/* Check for space on neighbor tuple page */
-			if (PageGetFreeSpace(*npage) + ItemIdGetLength(itemid) - sizeof(ItemIdData) >= ntupSize)
-			{
-				*freeOffno = offno;
-				*freeNeighborOffno = neighborOffno;
-				return true;
-			}
-			else if (*nbuf != buf)
-				UnlockReleaseBuffer(*nbuf);
-		}
-	}
-
-	return false;
-}
-
-/*
- * Add a new page
- */
-static void
-HnswInsertAppendPage(Relation index, Buffer *nbuf, Page *npage, GenericXLogState *state, Page page)
-{
-	/* Add a new page */
-	LockRelationForExtension(index, ExclusiveLock);
-	*nbuf = HnswNewBuffer(index, MAIN_FORKNUM);
-	UnlockRelationForExtension(index, ExclusiveLock);
-
-	/* Init new page */
-	*npage = GenericXLogRegisterBuffer(state, *nbuf, GENERIC_XLOG_FULL_IMAGE);
-	HnswInitPage(*nbuf, *npage);
-
-	/* Update previous buffer */
-	HnswPageGetOpaque(page)->nextblkno = BufferGetBlockNumber(*nbuf);
-}
-
-/*
- * Add to element and neighbor pages
- */
-static void
-WriteNewElementPages(Relation index, HnswElement e, int m)
-{
-	Buffer		buf;
-	Page		page;
-	GenericXLogState *state;
-	Size		etupSize;
-	Size		ntupSize;
-	Size		combinedSize;
-	HnswElementTuple etup;
-	BlockNumber insertPage = GetInsertPage(index);
-	BlockNumber originalInsertPage = insertPage;
-	int			dimensions = e->vec->dim;
-	HnswNeighborTuple ntup;
-	Buffer		nbuf;
-	Page		npage;
-	OffsetNumber freeOffno = InvalidOffsetNumber;
-	OffsetNumber freeNeighborOffno = InvalidOffsetNumber;
-	BlockNumber firstFreePage = InvalidBlockNumber;
-
-	/* Calculate sizes */
-	etupSize = HNSW_ELEMENT_TUPLE_SIZE(dimensions);
-	ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(e->level, m);
-	combinedSize = etupSize + ntupSize + sizeof(ItemIdData);
-
-	/* Prepare element tuple */
-	etup = palloc0(etupSize);
-	HnswSetElementTuple(etup, e);
-
-	/* Prepare neighbor tuple */
-	ntup = palloc0(ntupSize);
-	HnswSetNeighborTuple(ntup, e, m);
-
-	/* Find a page to insert the item */
-	for (;;)
-	{
-		buf = ReadBuffer(index, insertPage);
-		LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-
-		state = GenericXLogStart(index);
-		page = GenericXLogRegisterBuffer(state, buf, 0);
-
-		/* Space for both */
-		if (PageGetFreeSpace(page) >= combinedSize)
-		{
-			nbuf = buf;
-			npage = page;
-			break;
-		}
-
-		/* Space for element but not neighbors and last page */
-		if (PageGetFreeSpace(page) >= etupSize && !BlockNumberIsValid(HnswPageGetOpaque(page)->nextblkno))
-		{
-			HnswInsertAppendPage(index, &nbuf, &npage, state, page);
-			break;
-		}
-
-		/* Space from deleted item */
-		if (HnswFreeOffset(index, buf, page, e, ntupSize, &nbuf, &npage, &freeOffno, &freeNeighborOffno, &firstFreePage))
-		{
-			if (nbuf != buf)
-				npage = GenericXLogRegisterBuffer(state, nbuf, 0);
-
-			break;
-		}
-
-		insertPage = HnswPageGetOpaque(page)->nextblkno;
-
-		if (BlockNumberIsValid(insertPage))
-		{
-			/* Move to next page */
-			GenericXLogAbort(state);
-			UnlockReleaseBuffer(buf);
-		}
-		else
-		{
-			Buffer		newbuf;
-			Page		newpage;
-
-			HnswInsertAppendPage(index, &newbuf, &newpage, state, page);
-
-			/* Commit */
-			MarkBufferDirty(newbuf);
-			MarkBufferDirty(buf);
-			GenericXLogFinish(state);
-
-			/* Unlock previous buffer */
-			UnlockReleaseBuffer(buf);
-
-			/* Prepare new buffer */
-			state = GenericXLogStart(index);
-			buf = newbuf;
-			page = GenericXLogRegisterBuffer(state, buf, 0);
-
-			/* Create new page for neighbors if needed */
-			if (PageGetFreeSpace(page) < combinedSize)
-				HnswInsertAppendPage(index, &nbuf, &npage, state, page);
-			else
-			{
-				nbuf = buf;
-				npage = page;
-			}
-
-			break;
-		}
-	}
-
-	e->blkno = BufferGetBlockNumber(buf);
-	e->neighborPage = BufferGetBlockNumber(nbuf);
-
-	insertPage = e->neighborPage;
-
-	if (OffsetNumberIsValid(freeOffno))
-	{
-		e->offno = freeOffno;
-		e->neighborOffno = freeNeighborOffno;
-	}
-	else
-	{
-		e->offno = OffsetNumberNext(PageGetMaxOffsetNumber(page));
-		if (nbuf == buf)
-			e->neighborOffno = OffsetNumberNext(e->offno);
-		else
-			e->neighborOffno = FirstOffsetNumber;
-	}
-
-	ItemPointerSet(&etup->neighbortid, e->neighborPage, e->neighborOffno);
-
-	/* Add element and neighbors */
-	if (OffsetNumberIsValid(freeOffno))
-	{
-		if (!PageIndexTupleOverwrite(page, e->offno, (Item) etup, etupSize))
-			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-
-		if (!PageIndexTupleOverwrite(npage, e->neighborOffno, (Item) ntup, ntupSize))
-			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-	}
-	else
-	{
-		if (PageAddItem(page, (Item) etup, etupSize, InvalidOffsetNumber, false, false) != e->offno)
-			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-
-		if (PageAddItem(npage, (Item) ntup, ntupSize, InvalidOffsetNumber, false, false) != e->neighborOffno)
-			elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-	}
-
-	/* Commit */
-	MarkBufferDirty(buf);
-	if (nbuf != buf)
-		MarkBufferDirty(nbuf);
-	GenericXLogFinish(state);
-	UnlockReleaseBuffer(buf);
-	if (nbuf != buf)
-		UnlockReleaseBuffer(nbuf);
-
-	/* Update the insert page */
-	if (insertPage != originalInsertPage && (!OffsetNumberIsValid(freeOffno) || firstFreePage == insertPage))
-		HnswUpdateMetaPage(index, false, NULL, insertPage, MAIN_FORKNUM);
-}
-
-/*
- * Calculate index for update
- */
-static int
-HnswGetIndex(HnswUpdate * update, int m)
-{
-	return (update->hc.element->level - update->level) * m + update->index;
-}
-
-/*
- * Update neighbors
- */
-static void
-UpdateNeighborPages(Relation index, HnswElement e, int m, List *updates)
-{
-	ListCell   *lc;
-
-	/* Could update multiple at once for same element */
-	/* but should only happen a low percent of time, so keep simple for now */
-	foreach(lc, updates)
-	{
-		Buffer		buf;
-		Page		page;
-		GenericXLogState *state;
-		HnswUpdate *update = lfirst(lc);
-		ItemId		itemid;
-		HnswNeighborTuple ntup;
-		Size		ntupSize;
-		int			idx;
-		OffsetNumber offno = update->hc.element->neighborOffno;
-
-		/* Register page */
-		buf = ReadBuffer(index, update->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 */
-		idx = HnswGetIndex(update, m);
-
-		/* Make robust to issues */
-		if (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 */
-			MarkBufferDirty(buf);
-			GenericXLogFinish(state);
-		}
-		else
-			GenericXLogAbort(state);
-
-		UnlockReleaseBuffer(buf);
-	}
-}
-
-/*
- * Add a heap TID to an existing element
- */
-static bool
-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);
-	state = GenericXLogStart(index);
-	page = GenericXLogRegisterBuffer(state, buf, 0);
-
-	/* Find space */
-	etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, dup->offno));
-	for (i = 0; i < HNSW_HEAPTIDS; i++)
-	{
-		if (!ItemPointerIsValid(&etup->heaptids[i]))
-			break;
-	}
-
-	/* Either being deleted or we lost our chance to another backend */
-	if (i == 0 || i == HNSW_HEAPTIDS)
-	{
-		GenericXLogAbort(state);
-		UnlockReleaseBuffer(buf);
-		return false;
-	}
-
-	/* 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 */
-	MarkBufferDirty(buf);
-	GenericXLogFinish(state);
-	UnlockReleaseBuffer(buf);
-
-	return true;
-}
-
-/*
- * Write changes to disk
- */
-static void
-WriteElement(Relation index, HnswElement element, int m, List *updates, HnswElement dup, HnswElement entryPoint)
-{
-	/* Try to add to existing page */
-	if (dup != NULL)
-	{
-		if (HnswAddDuplicate(index, element, dup))
-			return;
-	}
-
-	/* If fails, take this path */
-	WriteNewElementPages(index, element, m);
-	UpdateNeighborPages(index, element, m, updates);
-
-	/* Update metapage if needed */
-	if (entryPoint == NULL || element->level > entryPoint->level)
-		HnswUpdateMetaPage(index, true, element, InvalidBlockNumber, MAIN_FORKNUM);
-}
-
-/*
- * Insert a tuple into the index
- */
-bool
-HnswInsertTuple(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid, Relation heapRel)
-{
-	Datum		value;
-	FmgrInfo   *normprocinfo;
-	HnswElement entryPoint;
-	HnswElement element;
-	int			m = HnswGetM(index);
-	int			efConstruction = HnswGetEfConstruction(index);
-	double		ml = HnswGetMl(m);
-	FmgrInfo   *procinfo = index_getprocinfo(index, 1, HNSW_DISTANCE_PROC);
-	Oid			collation = index->rd_indcollation[0];
-	List	   *updates = NIL;
-	HnswElement dup;
-
-	/* 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;
-	}
-
-	/* Create an element */
-	element = HnswInitElement(heap_tid, m, ml, HnswGetMaxLevel(m));
-	element->vec = DatumGetVector(value);
-
-	/* Get entry point */
-	entryPoint = HnswGetEntryPoint(index);
-
-	/* Insert element in graph */
-	dup = HnswInsertElement(element, entryPoint, index, procinfo, collation, m, efConstruction, &updates, false);
-
-	/* Write to disk */
-	WriteElement(index, element, m, updates, dup, entryPoint);
-
-	return true;
-}
-
-/*
- * Insert a tuple into the index
- */
-bool
-hnswinsert(Relation index, Datum *values, bool *isnull, ItemPointer heap_tid,
-		   Relation heap, IndexUniqueCheck checkUnique
-#if PG_VERSION_NUM >= 140000
-		   ,bool indexUnchanged
-#endif
-		   ,IndexInfo *indexInfo
-)
-{
-	MemoryContext oldCtx;
-	MemoryContext insertCtx;
-
-	/* Skip nulls */
-	if (isnull[0])
-		return false;
-
-	/* Create memory context */
-	insertCtx = AllocSetContextCreate(CurrentMemoryContext,
-									  "Hnsw insert temporary context",
-									  ALLOCSET_DEFAULT_SIZES);
-	oldCtx = MemoryContextSwitchTo(insertCtx);
-
-	/* Insert tuple */
-	HnswInsertTuple(index, values, isnull, heap_tid, heap);
-
-	/* Delete memory context */
-	MemoryContextSwitchTo(oldCtx);
-	MemoryContextDelete(insertCtx);
-
-	return false;
-}

src/hnswscan.c

@@ -1,212 +0,0 @@
-#include "postgres.h"
-
-#include "access/relscan.h"
-#include "hnsw.h"
-#include "pgstat.h"
-#include "storage/bufmgr.h"
-#include "utils/memutils.h"
-
-/*
- * Algorithm 5 from paper
- */
-static void
-GetScanItems(IndexScanDesc scan, Datum q)
-{
-	HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
-	Relation	index = scan->indexRelation;
-	FmgrInfo   *procinfo = so->procinfo;
-	Oid			collation = so->collation;
-	List	   *ep = NIL;
-	List	   *w;
-	HnswElement entryPoint = HnswGetEntryPoint(index);
-
-	if (entryPoint == NULL)
-		return;
-
-	ep = lappend(ep, HnswEntryCandidate(entryPoint, q, index, procinfo, collation, false));
-
-	for (int lc = entryPoint->level; lc >= 1; lc--)
-	{
-		w = HnswSearchLayer(q, ep, 1, lc, index, procinfo, collation, false, NULL, NULL);
-		ep = w;
-	}
-
-	so->w = HnswSearchLayer(q, ep, hnsw_ef_search, 0, index, procinfo, collation, false, NULL, NULL);
-}
-
-/*
- * Get dimensions from metapage
- */
-static int
-GetDimensions(Relation index)
-{
-	Buffer		buf;
-	Page		page;
-	HnswMetaPage metap;
-	int			dimensions;
-
-	buf = ReadBuffer(index, HNSW_METAPAGE_BLKNO);
-	LockBuffer(buf, BUFFER_LOCK_SHARE);
-	page = BufferGetPage(buf);
-	metap = HnswPageGetMeta(page);
-
-	dimensions = metap->dimensions;
-
-	UnlockReleaseBuffer(buf);
-
-	return dimensions;
-}
-
-/*
- * Prepare for an index scan
- */
-IndexScanDesc
-hnswbeginscan(Relation index, int nkeys, int norderbys)
-{
-	IndexScanDesc scan;
-	HnswScanOpaque so;
-
-	scan = RelationGetIndexScan(index, nkeys, norderbys);
-
-	so = (HnswScanOpaque) palloc(sizeof(HnswScanOpaqueData));
-	so->buf = InvalidBuffer;
-	so->first = true;
-	so->tmpCtx = AllocSetContextCreate(CurrentMemoryContext,
-									   "Hnsw scan temporary context",
-									   ALLOCSET_DEFAULT_SIZES);
-
-	/* 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;
-
-	return scan;
-}
-
-/*
- * Start or restart an index scan
- */
-void
-hnswrescan(IndexScanDesc scan, ScanKey keys, int nkeys, ScanKey orderbys, int norderbys)
-{
-	HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
-
-	so->first = true;
-	MemoryContextReset(so->tmpCtx);
-
-	if (keys && scan->numberOfKeys > 0)
-		memmove(scan->keyData, keys, scan->numberOfKeys * sizeof(ScanKeyData));
-
-	if (orderbys && scan->numberOfOrderBys > 0)
-		memmove(scan->orderByData, orderbys, scan->numberOfOrderBys * sizeof(ScanKeyData));
-}
-
-/*
- * Fetch the next tuple in the given scan
- */
-bool
-hnswgettuple(IndexScanDesc scan, ScanDirection dir)
-{
-	HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
-	MemoryContext oldCtx = MemoryContextSwitchTo(so->tmpCtx);
-
-	/*
-	 * Index can be used to scan backward, but Postgres doesn't support
-	 * backward scan on operators
-	 */
-	Assert(ScanDirectionIsForward(dir));
-
-	if (so->first)
-	{
-		Datum		value;
-
-		/* Count index scan for stats */
-		pgstat_count_index_scan(scan->indexRelation);
-
-		/* Safety check */
-		if (scan->orderByData == NULL)
-			elog(ERROR, "cannot scan hnsw index without order");
-
-		if (scan->orderByData->sk_flags & SK_ISNULL)
-			value = PointerGetDatum(InitVector(GetDimensions(scan->indexRelation)));
-		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)
-				HnswNormValue(so->normprocinfo, so->collation, &value, NULL);
-		}
-
-		GetScanItems(scan, value);
-		so->first = false;
-	}
-
-	while (list_length(so->w) > 0)
-	{
-		HnswCandidate *hc = llast(so->w);
-		ItemPointer tid;
-		BlockNumber indexblkno;
-
-		/* Move to next element if no valid heap tids */
-		if (list_length(hc->element->heaptids) == 0)
-		{
-			so->w = list_delete_last(so->w);
-			continue;
-		}
-
-		tid = llast(hc->element->heaptids);
-		indexblkno = hc->element->blkno;
-
-		hc->element->heaptids = list_delete_last(hc->element->heaptids);
-
-		MemoryContextSwitchTo(oldCtx);
-
-#if PG_VERSION_NUM >= 120000
-		scan->xs_heaptid = *tid;
-#else
-		scan->xs_ctup.t_self = *tid;
-#endif
-
-		if (BufferIsValid(so->buf))
-			ReleaseBuffer(so->buf);
-
-		/*
-		 * An index scan must maintain a pin on the index page holding the
-		 * item last returned by amgettuple
-		 *
-		 * https://www.postgresql.org/docs/current/index-locking.html
-		 */
-		so->buf = ReadBuffer(scan->indexRelation, indexblkno);
-
-		scan->xs_recheckorderby = false;
-		return true;
-	}
-
-	MemoryContextSwitchTo(oldCtx);
-	return false;
-}
-
-/*
- * End a scan and release resources
- */
-void
-hnswendscan(IndexScanDesc scan)
-{
-	HnswScanOpaque so = (HnswScanOpaque) scan->opaque;
-
-	/* Release pin */
-	if (BufferIsValid(so->buf))
-		ReleaseBuffer(so->buf);
-
-	MemoryContextDelete(so->tmpCtx);
-
-	pfree(so);
-	scan->opaque = NULL;
-}

src/hnswutils.c

@@ -1,981 +0,0 @@
-#include "postgres.h"
-
-#include <math.h>
-
-#include "hnsw.h"
-#include "storage/bufmgr.h"
-#include "vector.h"
-
-/*
- * Get the number of connection in the index
- */
-int
-HnswGetM(Relation index)
-{
-	HnswOptions *opts = (HnswOptions *) index->rd_options;
-
-	if (opts)
-		return opts->m;
-
-	return HNSW_DEFAULT_M;
-}
-
-/*
- * Get the size of the dynamic candidate list in the index
- */
-int
-HnswGetEfConstruction(Relation index)
-{
-	HnswOptions *opts = (HnswOptions *) index->rd_options;
-
-	if (opts)
-		return opts->efConstruction;
-
-	return HNSW_DEFAULT_EF_CONSTRUCTION;
-}
-
-/*
- * Get proc
- */
-FmgrInfo *
-HnswOptionalProcInfo(Relation rel, uint16 procnum)
-{
-	if (!OidIsValid(index_getprocid(rel, 1, procnum)))
-		return NULL;
-
-	return index_getprocinfo(rel, 1, procnum);
-}
-
-/*
- * 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
-HnswNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
-{
-	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;
-}
-
-/*
- * New buffer
- */
-Buffer
-HnswNewBuffer(Relation index, ForkNumber forkNum)
-{
-	Buffer		buf = ReadBufferExtended(index, forkNum, P_NEW, RBM_NORMAL, NULL);
-
-	LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-	return buf;
-}
-
-/*
- * Init page
- */
-void
-HnswInitPage(Buffer buf, Page page)
-{
-	PageInit(page, BufferGetPageSize(buf), sizeof(HnswPageOpaqueData));
-	HnswPageGetOpaque(page)->nextblkno = InvalidBlockNumber;
-	HnswPageGetOpaque(page)->page_id = HNSW_PAGE_ID;
-}
-
-/*
- * Init and register page
- */
-void
-HnswInitRegisterPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state)
-{
-	*state = GenericXLogStart(index);
-	*page = GenericXLogRegisterBuffer(*state, *buf, GENERIC_XLOG_FULL_IMAGE);
-	HnswInitPage(*buf, *page);
-}
-
-/*
- * Commit buffer
- */
-void
-HnswCommitBuffer(Buffer buf, GenericXLogState *state)
-{
-	MarkBufferDirty(buf);
-	GenericXLogFinish(state);
-	UnlockReleaseBuffer(buf);
-}
-
-/*
- * Allocate neighbors
- */
-void
-HnswInitNeighbors(HnswElement element, int m)
-{
-	int			level = element->level;
-
-	element->neighbors = palloc(sizeof(HnswNeighborArray) * (level + 1));
-
-	for (int lc = 0; lc <= level; lc++)
-	{
-		HnswNeighborArray *a;
-		int			lm = HnswGetLayerM(m, lc);
-
-		a = &element->neighbors[lc];
-		a->length = 0;
-		a->items = palloc(sizeof(HnswCandidate) * lm);
-	}
-}
-
-/*
- * Allocate an element
- */
-HnswElement
-HnswInitElement(ItemPointer heaptid, int m, double ml, int maxLevel)
-{
-	HnswElement element = palloc(sizeof(HnswElementData));
-
-	int			level = (int) (-log(RandomDouble()) * ml);
-
-	/* Cap level */
-	if (level > maxLevel)
-		level = maxLevel;
-
-	element->heaptids = NIL;
-	HnswAddHeapTid(element, heaptid);
-
-	element->level = level;
-	element->deleted = 0;
-
-	HnswInitNeighbors(element, m);
-
-	return element;
-}
-
-/*
- * Free an element
- */
-void
-HnswFreeElement(HnswElement element)
-{
-	list_free_deep(element->heaptids);
-	for (int lc = 0; lc <= element->level; lc++)
-		pfree(element->neighbors[lc].items);
-	pfree(element->neighbors);
-	pfree(element->vec);
-	pfree(element);
-}
-
-/*
- * Add a heap TID to an element
- */
-void
-HnswAddHeapTid(HnswElement element, ItemPointer heaptid)
-{
-	ItemPointer copy = palloc(sizeof(ItemPointerData));
-
-	ItemPointerCopy(heaptid, copy);
-	element->heaptids = lappend(element->heaptids, copy);
-}
-
-/*
- * Allocate an element from block and offset numbers
- */
-static HnswElement
-InitElementFromBlock(BlockNumber blkno, OffsetNumber offno)
-{
-	HnswElement element = palloc(sizeof(HnswElementData));
-
-	element->blkno = blkno;
-	element->offno = offno;
-	element->neighbors = NULL;
-	element->vec = NULL;
-	return element;
-}
-
-/*
- * Get the entry point
- */
-HnswElement
-HnswGetEntryPoint(Relation index)
-{
-	Buffer		buf;
-	Page		page;
-	HnswMetaPage metap;
-	HnswElement entryPoint = NULL;
-
-	buf = ReadBuffer(index, HNSW_METAPAGE_BLKNO);
-	LockBuffer(buf, BUFFER_LOCK_SHARE);
-	page = BufferGetPage(buf);
-	metap = HnswPageGetMeta(page);
-
-	if (BlockNumberIsValid(metap->entryBlkno))
-		entryPoint = InitElementFromBlock(metap->entryBlkno, metap->entryOffno);
-
-	UnlockReleaseBuffer(buf);
-
-	return entryPoint;
-}
-
-/*
- * Update the metapage
- */
-void
-HnswUpdateMetaPage(Relation index, bool updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum)
-{
-	Buffer		buf;
-	Page		page;
-	GenericXLogState *state;
-	HnswMetaPage metap;
-
-	buf = ReadBufferExtended(index, forkNum, HNSW_METAPAGE_BLKNO, RBM_NORMAL, NULL);
-	LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-	state = GenericXLogStart(index);
-	page = GenericXLogRegisterBuffer(state, buf, 0);
-
-	metap = HnswPageGetMeta(page);
-
-	if (updateEntry)
-	{
-		if (entryPoint == NULL)
-		{
-			metap->entryBlkno = InvalidBlockNumber;
-			metap->entryOffno = InvalidOffsetNumber;
-			metap->entryLevel = -1;
-		}
-		else
-		{
-			metap->entryBlkno = entryPoint->blkno;
-			metap->entryOffno = entryPoint->offno;
-			metap->entryLevel = entryPoint->level;
-		}
-	}
-
-	if (BlockNumberIsValid(insertPage))
-		metap->insertPage = insertPage;
-
-	HnswCommitBuffer(buf, state);
-}
-
-/*
- * Set element tuple, except for neighbor info
- */
-void
-HnswSetElementTuple(HnswElementTuple etup, HnswElement element)
-{
-	etup->type = HNSW_ELEMENT_TUPLE_TYPE;
-	etup->level = element->level;
-	etup->deleted = 0;
-	for (int i = 0; i < HNSW_HEAPTIDS; i++)
-	{
-		if (i < list_length(element->heaptids))
-			etup->heaptids[i] = *((ItemPointer) list_nth(element->heaptids, i));
-		else
-			ItemPointerSetInvalid(&etup->heaptids[i]);
-	}
-	memcpy(&etup->vec, element->vec, VECTOR_SIZE(element->vec->dim));
-}
-
-/*
- * Set neighbor tuple
- */
-void
-HnswSetNeighborTuple(HnswNeighborTuple ntup, HnswElement e, int m)
-{
-	int			idx = 0;
-
-	ntup->type = HNSW_NEIGHBOR_TUPLE_TYPE;
-
-	for (int lc = e->level; lc >= 0; lc--)
-	{
-		HnswNeighborArray *neighbors = &e->neighbors[lc];
-		int			lm = HnswGetLayerM(m, lc);
-
-		for (int i = 0; i < lm; i++)
-		{
-			ItemPointer indextid = &ntup->indextids[idx++];
-
-			if (i < neighbors->length)
-			{
-				HnswCandidate *hc = &neighbors->items[i];
-
-				ItemPointerSet(indextid, hc->element->blkno, hc->element->offno);
-			}
-			else
-				ItemPointerSetInvalid(indextid);
-		}
-	}
-
-	ntup->count = idx;
-}
-
-/*
- * Load neighbors from page
- */
-static void
-LoadNeighborsFromPage(HnswElement element, Relation index, Page page)
-{
-	HnswNeighborTuple ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, element->neighborOffno));
-	int			m = HnswGetM(index);
-	int			neighborCount = (element->level + 2) * m;
-
-	Assert(HnswIsNeighborTuple(ntup));
-
-	HnswInitNeighbors(element, m);
-
-	/* Ensure expected neighbors */
-	if (ntup->count != neighborCount)
-		return;
-
-	for (int i = 0; i < neighborCount; i++)
-	{
-		HnswElement e;
-		int			level;
-		HnswCandidate *hc;
-		ItemPointer indextid;
-		HnswNeighborArray *neighbors;
-
-		indextid = &ntup->indextids[i];
-
-		if (!ItemPointerIsValid(indextid))
-			continue;
-
-		e = InitElementFromBlock(ItemPointerGetBlockNumber(indextid), ItemPointerGetOffsetNumber(indextid));
-
-		/* Calculate level based on offset */
-		level = element->level - i / m;
-		if (level < 0)
-			level = 0;
-
-		neighbors = &element->neighbors[level];
-		hc = &neighbors->items[neighbors->length++];
-		hc->element = e;
-	}
-}
-
-/*
- * Load neighbors
- */
-static void
-LoadNeighbors(HnswElement element, Relation index)
-{
-	Buffer		buf;
-	Page		page;
-
-	buf = ReadBuffer(index, element->neighborPage);
-	LockBuffer(buf, BUFFER_LOCK_SHARE);
-	page = BufferGetPage(buf);
-
-	LoadNeighborsFromPage(element, index, page);
-
-	UnlockReleaseBuffer(buf);
-}
-
-/*
- * Load an element and optionally get its distance from q
- */
-void
-HnswLoadElement(HnswElement element, float *distance, Datum *q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadVec)
-{
-	Buffer		buf;
-	Page		page;
-	HnswElementTuple etup;
-
-	/* Read vector */
-	buf = ReadBuffer(index, element->blkno);
-	LockBuffer(buf, BUFFER_LOCK_SHARE);
-	page = BufferGetPage(buf);
-
-	etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, element->offno));
-
-	Assert(HnswIsElementTuple(etup));
-
-	/* Load element */
-	element->heaptids = NIL;
-	for (int i = 0; i < HNSW_HEAPTIDS; i++)
-	{
-		/* Can stop at first invalid */
-		if (!ItemPointerIsValid(&etup->heaptids[i]))
-			break;
-
-		HnswAddHeapTid(element, &etup->heaptids[i]);
-	}
-	element->level = etup->level;
-	element->neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
-	element->neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
-	element->deleted = etup->deleted;
-
-	if (loadVec)
-	{
-		element->vec = palloc(VECTOR_SIZE(etup->vec.dim));
-		memcpy(element->vec, &etup->vec, VECTOR_SIZE(etup->vec.dim));
-	}
-
-	/* Calculate distance */
-	if (distance != NULL)
-		*distance = (float) DatumGetFloat8(FunctionCall2Coll(procinfo, collation, *q, PointerGetDatum(&etup->vec)));
-
-	UnlockReleaseBuffer(buf);
-}
-
-/*
- * Get the distance for a candidate
- */
-static float
-GetCandidateDistance(HnswCandidate * hc, Datum q, FmgrInfo *procinfo, Oid collation)
-{
-	return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, q, PointerGetDatum(hc->element->vec)));
-}
-
-/*
- * Create a candidate for the entry point
- */
-HnswCandidate *
-HnswEntryCandidate(HnswElement entryPoint, Datum q, Relation index, FmgrInfo *procinfo, Oid collation, bool loadvec)
-{
-	HnswCandidate *hc = palloc(sizeof(HnswCandidate));
-
-	hc->element = entryPoint;
-	if (index == NULL)
-		hc->distance = GetCandidateDistance(hc, q, procinfo, collation);
-	else
-		HnswLoadElement(hc->element, &hc->distance, &q, index, procinfo, collation, loadvec);
-	return hc;
-}
-
-/*
- * Compare candidate distances
- */
-static int
-CompareNearestCandidates(const pairingheap_node *a, const pairingheap_node *b, void *arg)
-{
-	if (((const HnswPairingHeapNode *) a)->inner->distance < ((const HnswPairingHeapNode *) b)->inner->distance)
-		return 1;
-
-	if (((const HnswPairingHeapNode *) a)->inner->distance > ((const HnswPairingHeapNode *) b)->inner->distance)
-		return -1;
-
-	return 0;
-}
-
-/*
- * Compare candidate distances
- */
-static int
-CompareFurthestCandidates(const pairingheap_node *a, const pairingheap_node *b, void *arg)
-{
-	if (((const HnswPairingHeapNode *) a)->inner->distance < ((const HnswPairingHeapNode *) b)->inner->distance)
-		return -1;
-
-	if (((const HnswPairingHeapNode *) a)->inner->distance > ((const HnswPairingHeapNode *) b)->inner->distance)
-		return 1;
-
-	return 0;
-}
-
-/*
- * Create a pairing heap node for a candidate
- */
-static HnswPairingHeapNode *
-CreatePairingHeapNode(HnswCandidate * c)
-{
-	HnswPairingHeapNode *node = palloc(sizeof(HnswPairingHeapNode));
-
-	node->inner = c;
-	return node;
-}
-
-/*
- * Add to visited
- */
-static inline void
-AddToVisited(HTAB *v, HnswCandidate * hc, Relation index, bool *found)
-{
-	if (index == NULL)
-		hash_search(v, &hc->element, HASH_ENTER, found);
-	else
-	{
-		ItemPointerData indextid;
-
-		ItemPointerSet(&indextid, hc->element->blkno, hc->element->offno);
-		hash_search(v, &indextid, HASH_ENTER, found);
-	}
-}
-
-/*
- * Algorithm 2 from paper
- */
-List *
-HnswSearchLayer(Datum q, List *ep, int ef, int lc, Relation index, FmgrInfo *procinfo, Oid collation, bool inserting, BlockNumber *skipPage, OffsetNumber *skipOffno)
-{
-	ListCell   *lc2;
-
-	List	   *w = NIL;
-	pairingheap *C = pairingheap_allocate(CompareNearestCandidates, NULL);
-	pairingheap *W = pairingheap_allocate(CompareFurthestCandidates, NULL);
-	int			wlen = 0;
-	HASHCTL		hash_ctl;
-	HTAB	   *v;
-
-	/* Create hash table */
-	if (index == NULL)
-	{
-		hash_ctl.keysize = sizeof(HnswElement *);
-		hash_ctl.entrysize = sizeof(HnswElement *);
-	}
-	else
-	{
-		hash_ctl.keysize = sizeof(ItemPointerData);
-		hash_ctl.entrysize = sizeof(ItemPointerData);
-	}
-
-	hash_ctl.hcxt = CurrentMemoryContext;
-	v = hash_create("hnsw visited", 256, &hash_ctl, HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
-
-	/* Add entry points to v, C, and W */
-	foreach(lc2, ep)
-	{
-		HnswCandidate *hc = (HnswCandidate *) lfirst(lc2);
-
-		AddToVisited(v, hc, index, NULL);
-
-		pairingheap_add(C, &(CreatePairingHeapNode(hc)->ph_node));
-		pairingheap_add(W, &(CreatePairingHeapNode(hc)->ph_node));
-
-		wlen++;
-	}
-
-	while (!pairingheap_is_empty(C))
-	{
-		HnswNeighborArray *neighborhood;
-		HnswCandidate *c = ((HnswPairingHeapNode *) pairingheap_remove_first(C))->inner;
-		HnswCandidate *f = ((HnswPairingHeapNode *) pairingheap_first(W))->inner;
-
-		if (c->distance > f->distance)
-			break;
-
-		if (c->element->neighbors == NULL)
-			LoadNeighbors(c->element, index);
-
-		/* Get the neighborhood at layer lc */
-		neighborhood = &c->element->neighbors[lc];
-
-		for (int i = 0; i < neighborhood->length; i++)
-		{
-			HnswCandidate *e = &neighborhood->items[i];
-			bool		visited;
-
-			AddToVisited(v, e, index, &visited);
-
-			if (!visited)
-			{
-				float		eDistance;
-
-				f = ((HnswPairingHeapNode *) pairingheap_first(W))->inner;
-
-				if (index == NULL)
-					eDistance = GetCandidateDistance(e, q, procinfo, collation);
-				else
-					HnswLoadElement(e->element, &eDistance, &q, index, procinfo, collation, inserting);
-
-				/* Skip if fully deleted */
-				if (e->element->deleted)
-					continue;
-
-				/* Skip for inserts if deleting */
-				if (inserting && list_length(e->element->heaptids) == 0)
-					continue;
-
-				/* Skip self for vacuuming update */
-				if (skipPage != NULL && e->element->neighborPage == *skipPage && e->element->neighborOffno == *skipOffno)
-					continue;
-
-				/* Make robust to issues */
-				if (e->element->level < lc)
-					continue;
-
-				if (eDistance < f->distance || wlen < ef)
-				{
-					/* Copy e */
-					HnswCandidate *ec = palloc(sizeof(HnswCandidate));
-
-					ec->element = e->element;
-					ec->distance = eDistance;
-
-					pairingheap_add(C, &(CreatePairingHeapNode(ec)->ph_node));
-					pairingheap_add(W, &(CreatePairingHeapNode(ec)->ph_node));
-					wlen++;
-
-					/* No need to decrement wlen */
-					if (wlen > ef)
-						pairingheap_remove_first(W);
-				}
-			}
-		}
-	}
-
-	/* Add each element of W to w */
-	while (!pairingheap_is_empty(W))
-	{
-		HnswCandidate *hc = ((HnswPairingHeapNode *) pairingheap_remove_first(W))->inner;
-
-		w = lappend(w, hc);
-	}
-
-	return w;
-}
-
-/*
- * Calculate the distance between elements
- */
-static float
-HnswGetDistance(HnswElement a, HnswElement b, int lc, FmgrInfo *procinfo, Oid collation)
-{
-	/* Look for cached distance */
-	if (a->neighbors != NULL)
-	{
-		Assert(a->level >= lc);
-
-		for (int i = 0; i < a->neighbors[lc].length; i++)
-		{
-			if (a->neighbors[lc].items[i].element == b)
-				return a->neighbors[lc].items[i].distance;
-		}
-	}
-
-	if (b->neighbors != NULL)
-	{
-		Assert(b->level >= lc);
-
-		for (int i = 0; i < b->neighbors[lc].length; i++)
-		{
-			if (b->neighbors[lc].items[i].element == a)
-				return b->neighbors[lc].items[i].distance;
-		}
-	}
-
-	return DatumGetFloat8(FunctionCall2Coll(procinfo, collation, PointerGetDatum(a->vec), PointerGetDatum(b->vec)));
-}
-
-/*
- * Check if an element is closer to q than any element from R
- */
-static bool
-CheckElementCloser(HnswCandidate * e, List *r, int lc, FmgrInfo *procinfo, Oid collation)
-{
-	ListCell   *lc2;
-
-	foreach(lc2, r)
-	{
-		HnswCandidate *ri = lfirst(lc2);
-		float		distance = HnswGetDistance(e->element, ri->element, lc, procinfo, collation);
-
-		if (distance <= e->distance)
-			return false;
-	}
-
-	return true;
-}
-
-/*
- * Algorithm 4 from paper
- */
-static List *
-SelectNeighbors(List *c, int m, int lc, FmgrInfo *procinfo, Oid collation, HnswCandidate * *pruned)
-{
-	List	   *r = NIL;
-	List	   *w = list_copy(c);
-	pairingheap *wd;
-
-	if (list_length(w) < m)
-		return w;
-
-	wd = pairingheap_allocate(CompareNearestCandidates, NULL);
-
-	while (list_length(w) > 0 && list_length(r) < m)
-	{
-		/* Assumes w is already ordered desc */
-		HnswCandidate *e = llast(w);
-		bool		closer;
-
-		w = list_delete_last(w);
-
-		closer = CheckElementCloser(e, r, lc, procinfo, collation);
-
-		if (closer)
-			r = lappend(r, e);
-		else
-			pairingheap_add(wd, &(CreatePairingHeapNode(e)->ph_node));
-	}
-
-	/* Keep pruned connections */
-	while (!pairingheap_is_empty(wd) && list_length(r) < m)
-		r = lappend(r, ((HnswPairingHeapNode *) pairingheap_remove_first(wd))->inner);
-
-	/* Return pruned for update connections */
-	if (pruned != NULL)
-	{
-		if (!pairingheap_is_empty(wd))
-			*pruned = ((HnswPairingHeapNode *) pairingheap_first(wd))->inner;
-		else
-			*pruned = linitial(w);
-	}
-
-	return r;
-}
-
-/*
- * Find duplicate element
- */
-static HnswElement
-HnswFindDuplicate(HnswElement e, List *neighbors)
-{
-	ListCell   *lc;
-
-	foreach(lc, neighbors)
-	{
-		HnswCandidate *neighbor = lfirst(lc);
-
-		/* Exit early since ordered by distance */
-		if (vector_cmp_internal(e->vec, neighbor->element->vec) != 0)
-			break;
-
-		/* Check for space */
-		if (list_length(neighbor->element->heaptids) < HNSW_HEAPTIDS)
-			return neighbor->element;
-	}
-
-	return NULL;
-}
-
-/*
- * Add connections
- */
-static void
-AddConnections(HnswElement element, List *neighbors, int m, int lc)
-{
-	ListCell   *lc2;
-	HnswNeighborArray *a = &element->neighbors[lc];
-
-	foreach(lc2, neighbors)
-		a->items[a->length++] = *((HnswCandidate *) lfirst(lc2));
-}
-
-/*
- * Compare candidate distances
- */
-static int
-#if PG_VERSION_NUM >= 130000
-CompareCandidateDistances(const ListCell *a, const ListCell *b)
-#else
-CompareCandidateDistances(const void *a, const void *b)
-#endif
-{
-	HnswCandidate *hca = lfirst((ListCell *) a);
-	HnswCandidate *hcb = lfirst((ListCell *) b);
-
-	if (hca->distance < hcb->distance)
-		return 1;
-
-	if (hca->distance > hcb->distance)
-		return -1;
-
-	return 0;
-}
-
-/*
- * Create update
- */
-static HnswUpdate *
-CreateUpdate(HnswCandidate * hc, int level, int index)
-{
-	HnswUpdate *update = palloc(sizeof(HnswUpdate));
-
-	update->hc = *hc;
-	update->level = level;
-	update->index = index;
-	return update;
-}
-
-/*
- * Update connections
- */
-static void
-UpdateConnections(HnswElement element, List *neighbors, int m, int lc, List **updates, Relation index, FmgrInfo *procinfo, Oid collation)
-{
-	ListCell   *lc2;
-
-	foreach(lc2, neighbors)
-	{
-		HnswCandidate *hc = (HnswCandidate *) lfirst(lc2);
-		HnswNeighborArray *currentNeighbors = &hc->element->neighbors[lc];
-
-		HnswCandidate hc2;
-
-		hc2.element = element;
-		hc2.distance = hc->distance;
-
-		if (currentNeighbors->length < m)
-		{
-			currentNeighbors->items[currentNeighbors->length++] = hc2;
-
-			/* Track updates */
-			if (updates != NULL)
-				*updates = lappend(*updates, CreateUpdate(hc, lc, currentNeighbors->length - 1));
-		}
-		else
-		{
-			/* Shrink connections */
-			HnswCandidate *pruned = NULL;
-			List	   *c = NIL;
-
-			/* Load elements on insert */
-			if (index != NULL)
-			{
-				Datum		q = PointerGetDatum(hc->element->vec);
-
-				for (int i = 0; i < currentNeighbors->length; i++)
-				{
-					HnswCandidate *hc3 = &currentNeighbors->items[i];
-
-					if (hc3->element->vec == NULL)
-						HnswLoadElement(hc3->element, &hc3->distance, &q, index, procinfo, collation, true);
-					else
-						hc3->distance = GetCandidateDistance(hc3, q, procinfo, collation);
-
-					/* Prune element if being deleted */
-					if (list_length(hc3->element->heaptids) == 0)
-					{
-						pruned = &currentNeighbors->items[i];
-						break;
-					}
-				}
-			}
-
-			if (pruned == NULL)
-			{
-				/* Add and sort candidates */
-				for (int i = 0; i < currentNeighbors->length; i++)
-					c = lappend(c, &currentNeighbors->items[i]);
-				c = lappend(c, &hc2);
-				list_sort(c, CompareCandidateDistances);
-
-				SelectNeighbors(c, m, lc, procinfo, collation, &pruned);
-
-				/* Should not happen */
-				if (pruned == NULL)
-					continue;
-			}
-
-			/* Find and replace the pruned element */
-			for (int i = 0; i < currentNeighbors->length; i++)
-			{
-				if (currentNeighbors->items[i].element == pruned->element)
-				{
-					currentNeighbors->items[i] = hc2;
-
-					/* Track updates */
-					if (updates != NULL)
-						*updates = lappend(*updates, CreateUpdate(hc, lc, i));
-
-					break;
-				}
-			}
-		}
-	}
-}
-
-/*
- * Algorithm 1 from paper
- */
-HnswElement
-HnswInsertElement(HnswElement element, HnswElement entryPoint, Relation index, FmgrInfo *procinfo, Oid collation, int m, int efConstruction, List **updates, bool vacuuming)
-{
-	List	   *ep = NIL;
-	List	   *w;
-	int			level = element->level;
-	int			entryLevel;
-	List	  **newNeighbors = palloc(sizeof(List *) * (level + 1));
-	Datum		q = PointerGetDatum(element->vec);
-	HnswElement dup;
-	BlockNumber *skipPage = vacuuming ? &element->neighborPage : NULL;
-	OffsetNumber *skipOffno = vacuuming ? &element->neighborOffno : NULL;
-	bool		removeEntryPoint;
-	HnswCandidate *entryCandidate;
-
-	/* Get entry point and level */
-	if (entryPoint != NULL)
-	{
-		entryCandidate = HnswEntryCandidate(entryPoint, q, index, procinfo, collation, true);
-		ep = lappend(ep, entryCandidate);
-		entryLevel = entryPoint->level;
-		removeEntryPoint = vacuuming && list_length(entryPoint->heaptids) == 0;
-	}
-	else
-	{
-		entryLevel = -1;
-		removeEntryPoint = false;
-	}
-
-	/* 1st phase: greedy search to insert level */
-	for (int lc = entryLevel; lc >= level + 1; lc--)
-	{
-		w = HnswSearchLayer(q, ep, 1, lc, index, procinfo, collation, true, skipPage, skipOffno);
-		ep = w;
-	}
-
-	if (level > entryLevel)
-		level = entryLevel;
-
-	/* 2nd phase */
-	for (int lc = level; lc >= 0; lc--)
-	{
-		int			lm = HnswGetLayerM(m, lc);
-
-		w = HnswSearchLayer(q, ep, efConstruction, lc, index, procinfo, collation, true, skipPage, skipOffno);
-
-		/* Remove entry point if it's being deleted */
-		if (removeEntryPoint)
-			w = list_delete_ptr(w, entryCandidate);
-
-		newNeighbors[lc] = SelectNeighbors(w, lm, lc, procinfo, collation, NULL);
-		ep = w;
-	}
-
-	/* Look for duplicate */
-	if (level >= 0 && !vacuuming)
-	{
-		dup = HnswFindDuplicate(element, newNeighbors[0]);
-		if (dup != NULL)
-			return dup;
-	}
-
-	/* Update connections */
-	for (int lc = level; lc >= 0; lc--)
-	{
-		int			lm = HnswGetLayerM(m, lc);
-
-		AddConnections(element, newNeighbors[lc], lm, lc);
-
-		if (!vacuuming)
-			UpdateConnections(element, newNeighbors[lc], lm, lc, updates, index, procinfo, collation);
-	}
-
-	return NULL;
-}

src/hnswvacuum.c

@@ -1,581 +0,0 @@
-#include "postgres.h"
-
-#include <math.h>
-
-#include "commands/vacuum.h"
-#include "hnsw.h"
-#include "storage/bufmgr.h"
-#include "utils/memutils.h"
-
-/*
- * Check if deleted list contains an index tid
- */
-static bool
-DeletedContains(HTAB *deleted, ItemPointer indextid)
-{
-	bool		found;
-
-	hash_search(deleted, indextid, HASH_FIND, &found);
-	return found;
-}
-
-/*
- * Remove deleted heap TIDs
- *
- * OK to remove for entry point, since always considered for searches and inserts
- */
-static void
-RemoveHeapTids(HnswVacuumState * vacuumstate)
-{
-	BlockNumber blkno = HNSW_HEAD_BLKNO;
-	HnswElement highestPoint = &vacuumstate->highestPoint;
-	Relation	index = vacuumstate->index;
-	BufferAccessStrategy bas = vacuumstate->bas;
-	HnswElement entryPoint = HnswGetEntryPoint(vacuumstate->index);
-
-	/* Store separately since highestPoint.level is uint8 */
-	int			highestLevel = -1;
-
-	/* Initialize highest point */
-	highestPoint->blkno = InvalidBlockNumber;
-	highestPoint->offno = InvalidOffsetNumber;
-
-	while (BlockNumberIsValid(blkno))
-	{
-		Buffer		buf;
-		Page		page;
-		GenericXLogState *state;
-		OffsetNumber offno;
-		OffsetNumber maxoffno;
-		bool		updated = false;
-
-		vacuum_delay_point();
-
-		buf = ReadBufferExtended(index, MAIN_FORKNUM, blkno, RBM_NORMAL, bas);
-		LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-		state = GenericXLogStart(index);
-		page = GenericXLogRegisterBuffer(state, buf, 0);
-		maxoffno = PageGetMaxOffsetNumber(page);
-
-		/* Iterate over nodes */
-		for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
-		{
-			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
-			int			idx = 0;
-			bool		itemUpdated = false;
-
-			/* Skip neighbor tuples */
-			if (!HnswIsElementTuple(etup))
-				continue;
-
-			if (ItemPointerIsValid(&etup->heaptids[0]))
-			{
-				for (int i = 0; i < HNSW_HEAPTIDS; i++)
-				{
-					/* Stop at first unused */
-					if (!ItemPointerIsValid(&etup->heaptids[i]))
-						break;
-
-					if (vacuumstate->callback(&etup->heaptids[i], vacuumstate->callback_state))
-						itemUpdated = true;
-					else
-					{
-						/* Move to front of list */
-						etup->heaptids[idx++] = etup->heaptids[i];
-					}
-				}
-
-				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;
-				}
-			}
-
-			if (!ItemPointerIsValid(&etup->heaptids[0]))
-			{
-				ItemPointerData ip;
-
-				/* Add to deleted list */
-				ItemPointerSet(&ip, blkno, offno);
-
-				(void) hash_search(vacuumstate->deleted, &ip, HASH_ENTER, NULL);
-			}
-			else if (etup->level > highestLevel && !(blkno == entryPoint->blkno && offno == entryPoint->offno))
-			{
-				/* Keep track of highest non-entry point */
-				highestPoint->blkno = blkno;
-				highestPoint->offno = offno;
-				highestPoint->level = etup->level;
-				highestLevel = etup->level;
-			}
-		}
-
-		blkno = HnswPageGetOpaque(page)->nextblkno;
-
-		if (updated)
-		{
-			MarkBufferDirty(buf);
-			GenericXLogFinish(state);
-		}
-		else
-			GenericXLogAbort(state);
-
-		UnlockReleaseBuffer(buf);
-	}
-}
-
-/*
- * Check for deleted neighbors
- */
-static bool
-NeedsUpdated(HnswVacuumState * vacuumstate, HnswElement element)
-{
-	Relation	index = vacuumstate->index;
-	BufferAccessStrategy bas = vacuumstate->bas;
-	Buffer		buf;
-	Page		page;
-	HnswNeighborTuple ntup;
-	bool		needsUpdated = false;
-
-	buf = ReadBufferExtended(index, MAIN_FORKNUM, element->neighborPage, RBM_NORMAL, bas);
-	LockBuffer(buf, BUFFER_LOCK_SHARE);
-	page = BufferGetPage(buf);
-	ntup = (HnswNeighborTuple) PageGetItem(page, PageGetItemId(page, element->neighborOffno));
-
-	Assert(HnswIsNeighborTuple(ntup));
-
-	/* 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))
-		{
-			needsUpdated = true;
-			break;
-		}
-	}
-
-	UnlockReleaseBuffer(buf);
-
-	return needsUpdated;
-}
-
-/*
- * Repair graph for a single element
- */
-static void
-RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element)
-{
-	Relation	index = vacuumstate->index;
-	Buffer		buf;
-	Page		page;
-	GenericXLogState *state;
-	int			m = vacuumstate->m;
-	int			efConstruction = vacuumstate->efConstruction;
-	FmgrInfo   *procinfo = vacuumstate->procinfo;
-	Oid			collation = vacuumstate->collation;
-	HnswElement entryPoint;
-	BufferAccessStrategy bas = vacuumstate->bas;
-	HnswNeighborTuple ntup = vacuumstate->ntup;
-	Size		ntupSize = HNSW_NEIGHBOR_TUPLE_SIZE(element->level, m);
-
-	/* Check if any neighbors point to deleted values */
-	if (!NeedsUpdated(vacuumstate, element))
-		return;
-
-	/* Refresh entry point for each element */
-	entryPoint = HnswGetEntryPoint(index);
-
-	/* Special case for entry point */
-	if (element->blkno == entryPoint->blkno && element->offno == entryPoint->offno)
-	{
-		if (BlockNumberIsValid(vacuumstate->highestPoint.blkno))
-		{
-			/* Already updated */
-			if (vacuumstate->highestPoint.blkno == element->blkno && vacuumstate->highestPoint.offno == element->offno)
-				return;
-
-			entryPoint = &vacuumstate->highestPoint;
-
-			/* Reset neighbors from previous update */
-			entryPoint->neighbors = NULL;
-		}
-		else
-			entryPoint = NULL;
-	}
-
-	/* Init fields */
-	HnswInitNeighbors(element, m);
-	element->heaptids = NIL;
-
-	/* Add element to graph, skipping itself */
-	HnswInsertElement(element, entryPoint, index, procinfo, collation, m, efConstruction, NULL, true);
-
-	/* Update neighbor tuple */
-	/* Do this before getting page to minimize locking */
-	HnswSetNeighborTuple(ntup, element, m);
-
-	/* Get neighbor page */
-	buf = ReadBufferExtended(index, MAIN_FORKNUM, element->neighborPage, RBM_NORMAL, bas);
-	LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
-	state = GenericXLogStart(index);
-	page = GenericXLogRegisterBuffer(state, buf, 0);
-
-	/* Overwrite tuple */
-	if (!PageIndexTupleOverwrite(page, element->neighborOffno, (Item) ntup, ntupSize))
-		elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-
-	/* Commit */
-	MarkBufferDirty(buf);
-	GenericXLogFinish(state);
-	UnlockReleaseBuffer(buf);
-}
-
-/*
- * Repair graph entry point
- */
-static void
-RepairGraphEntryPoint(HnswVacuumState * vacuumstate)
-{
-	Relation	index = vacuumstate->index;
-	HnswElement highestPoint = &vacuumstate->highestPoint;
-	HnswElement entryPoint;
-	MemoryContext oldCtx = MemoryContextSwitchTo(vacuumstate->tmpCtx);
-
-	/* Repair graph for highest non-entry point */
-	/* This may not be the highest with new inserts, but should be fine */
-	if (BlockNumberIsValid(highestPoint->blkno))
-	{
-		HnswLoadElement(highestPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true);
-		RepairGraphElement(vacuumstate, highestPoint);
-	}
-
-	/* See if entry point needs updated */
-	entryPoint = HnswGetEntryPoint(index);
-	if (entryPoint != NULL)
-	{
-		ItemPointerData epData;
-
-		ItemPointerSet(&epData, entryPoint->blkno, entryPoint->offno);
-
-		if (DeletedContains(vacuumstate->deleted, &epData))
-			HnswUpdateMetaPage(index, true, highestPoint, InvalidBlockNumber, MAIN_FORKNUM);
-		else
-		{
-			/* Highest point will be used to repair */
-			HnswLoadElement(entryPoint, NULL, NULL, index, vacuumstate->procinfo, vacuumstate->collation, true);
-			RepairGraphElement(vacuumstate, entryPoint);
-		}
-	}
-
-	/* Reset memory context */
-	MemoryContextSwitchTo(oldCtx);
-	MemoryContextReset(vacuumstate->tmpCtx);
-}
-
-/*
- * Repair graph for all elements
- */
-static void
-RepairGraph(HnswVacuumState * vacuumstate)
-{
-	Relation	index = vacuumstate->index;
-	BufferAccessStrategy bas = vacuumstate->bas;
-	BlockNumber blkno = HNSW_HEAD_BLKNO;
-
-	RepairGraphEntryPoint(vacuumstate);
-
-	while (BlockNumberIsValid(blkno))
-	{
-		Buffer		buf;
-		Page		page;
-		OffsetNumber offno;
-		OffsetNumber maxoffno;
-		List	   *elements = NIL;
-		ListCell   *lc2;
-		MemoryContext oldCtx;
-
-		vacuum_delay_point();
-
-		oldCtx = MemoryContextSwitchTo(vacuumstate->tmpCtx);
-
-		buf = ReadBufferExtended(index, MAIN_FORKNUM, blkno, RBM_NORMAL, bas);
-		LockBuffer(buf, BUFFER_LOCK_SHARE);
-		page = BufferGetPage(buf);
-		maxoffno = PageGetMaxOffsetNumber(page);
-
-		/* Load items into memory to minimize locking */
-		for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
-		{
-			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
-			HnswElement element;
-
-			/* Skip neighbor tuples */
-			if (!HnswIsElementTuple(etup))
-				continue;
-
-			/* Skip updating neighbors if being deleted */
-			if (!ItemPointerIsValid(&etup->heaptids[0]))
-				continue;
-
-			/* Create an element */
-			element = palloc(sizeof(HnswElementData));
-			element->neighborPage = ItemPointerGetBlockNumber(&etup->neighbortid);
-			element->neighborOffno = ItemPointerGetOffsetNumber(&etup->neighbortid);
-			element->level = etup->level;
-			element->blkno = blkno;
-			element->offno = offno;
-			element->vec = palloc(VECTOR_SIZE(etup->vec.dim));
-			memcpy(element->vec, &etup->vec, VECTOR_SIZE(etup->vec.dim));
-
-			elements = lappend(elements, element);
-		}
-
-		blkno = HnswPageGetOpaque(page)->nextblkno;
-
-		UnlockReleaseBuffer(buf);
-
-		/* Update neighbor pages */
-		foreach(lc2, elements)
-			RepairGraphElement(vacuumstate, (HnswElement) lfirst(lc2));
-
-		/* Reset memory context */
-		MemoryContextSwitchTo(oldCtx);
-		MemoryContextReset(vacuumstate->tmpCtx);
-	}
-}
-
-/*
- * Mark items as deleted
- */
-static void
-MarkDeleted(HnswVacuumState * vacuumstate)
-{
-	BlockNumber blkno = HNSW_HEAD_BLKNO;
-	BlockNumber insertPage = InvalidBlockNumber;
-	Relation	index = vacuumstate->index;
-	BufferAccessStrategy bas = vacuumstate->bas;
-	IndexBulkDeleteResult *stats = vacuumstate->stats;
-
-	while (BlockNumberIsValid(blkno))
-	{
-		Buffer		buf;
-		Page		page;
-		GenericXLogState *state;
-		OffsetNumber offno;
-		OffsetNumber maxoffno;
-
-		vacuum_delay_point();
-
-		buf = ReadBufferExtended(index, MAIN_FORKNUM, blkno, RBM_NORMAL, bas);
-
-		/*
-		 * ambulkdelete cannot delete entries from pages that are pinned by
-		 * other backends
-		 *
-		 * https://www.postgresql.org/docs/current/index-locking.html
-		 */
-		LockBufferForCleanup(buf);
-
-		state = GenericXLogStart(index);
-		page = GenericXLogRegisterBuffer(state, buf, 0);
-		maxoffno = PageGetMaxOffsetNumber(page);
-
-		/* Update element and neighbors together */
-		for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
-		{
-			HnswElementTuple etup = (HnswElementTuple) PageGetItem(page, PageGetItemId(page, offno));
-			HnswNeighborTuple ntup;
-			Size		etupSize;
-			Size		ntupSize;
-			Buffer		nbuf;
-			Page		npage;
-			BlockNumber neighborPage;
-			OffsetNumber neighborOffno;
-
-			/* Skip neighbor tuples */
-			if (!HnswIsElementTuple(etup))
-				continue;
-
-			/* Skip deleted tuples */
-			if (etup->deleted)
-				continue;
-
-			/* Skip live tuples */
-			if (ItemPointerIsValid(&etup->heaptids[0]))
-			{
-				stats->num_index_tuples++;
-				continue;
-			}
-
-			/* Update stats */
-			stats->tuples_removed++;
-
-			/* 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);
-
-			if (neighborPage == blkno)
-			{
-				nbuf = buf;
-				npage = page;
-			}
-			else
-			{
-				nbuf = ReadBufferExtended(index, MAIN_FORKNUM, neighborPage, RBM_NORMAL, bas);
-				LockBuffer(nbuf, BUFFER_LOCK_EXCLUSIVE);
-				npage = GenericXLogRegisterBuffer(state, nbuf, 0);
-			}
-
-			ntup = (HnswNeighborTuple) PageGetItem(npage, PageGetItemId(npage, neighborOffno));
-
-			/* Overwrite element */
-			etup->deleted = 1;
-			MemSet(&etup->vec.x, 0, etup->vec.dim * sizeof(float));
-
-			/* Overwrite neighbors */
-			for (int i = 0; i < ntup->count; i++)
-				ItemPointerSetInvalid(&ntup->indextids[i]);
-
-			/* Overwrite element tuple */
-			if (!PageIndexTupleOverwrite(page, offno, (Item) etup, etupSize))
-				elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-
-			/* Overwrite neighbor tuple */
-			if (!PageIndexTupleOverwrite(npage, neighborOffno, (Item) ntup, ntupSize))
-				elog(ERROR, "failed to add index item to \"%s\"", RelationGetRelationName(index));
-
-			/* Commit */
-			MarkBufferDirty(buf);
-			if (nbuf != buf)
-				MarkBufferDirty(nbuf);
-			GenericXLogFinish(state);
-			if (nbuf != buf)
-				UnlockReleaseBuffer(nbuf);
-
-			/* Set to first free page */
-			if (!BlockNumberIsValid(insertPage))
-				insertPage = blkno;
-
-			/* Prepare new xlog */
-			state = GenericXLogStart(index);
-			page = GenericXLogRegisterBuffer(state, buf, 0);
-		}
-
-		blkno = HnswPageGetOpaque(page)->nextblkno;
-
-		GenericXLogAbort(state);
-		UnlockReleaseBuffer(buf);
-	}
-
-	HnswUpdateMetaPage(index, false, NULL, insertPage, MAIN_FORKNUM);
-}
-
-/*
- * Initialize the vacuum state
- */
-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));
-
-	vacuumstate->index = index;
-	vacuumstate->stats = stats;
-	vacuumstate->callback = callback;
-	vacuumstate->callback_state = callback_state;
-	vacuumstate->m = HnswGetM(index);
-	vacuumstate->efConstruction = HnswGetEfConstruction(index);
-	vacuumstate->bas = GetAccessStrategy(BAS_BULKREAD);
-	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);
-
-	/* Create hash table */
-	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);
-}
-
-/*
- * Free resources
- */
-static void
-FreeVacuumState(HnswVacuumState * vacuumstate)
-{
-	hash_destroy(vacuumstate->deleted);
-	FreeAccessStrategy(vacuumstate->bas);
-	pfree(vacuumstate->ntup);
-	MemoryContextDelete(vacuumstate->tmpCtx);
-}
-
-/*
- * Bulk delete tuples from the index
- */
-IndexBulkDeleteResult *
-hnswbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
-			   IndexBulkDeleteCallback callback, void *callback_state)
-{
-	HnswVacuumState vacuumstate;
-
-	InitVacuumState(&vacuumstate, info, stats, callback, callback_state);
-
-	/* Pass 1: Remove heap TIDs */
-	RemoveHeapTids(&vacuumstate);
-
-	/* Pass 2: Repair graph */
-	RepairGraph(&vacuumstate);
-
-	/* Pass 3: Mark as deleted */
-	MarkDeleted(&vacuumstate);
-
-	FreeVacuumState(&vacuumstate);
-
-	return vacuumstate.stats;
-}
-
-/*
- * Clean up after a VACUUM operation
- */
-IndexBulkDeleteResult *
-hnswvacuumcleanup(IndexVacuumInfo *info, IndexBulkDeleteResult *stats)
-{
-	Relation	rel = info->index;
-
-	if (info->analyze_only)
-		return stats;
-
-	/* stats is NULL if ambulkdelete not called */
-	/* OK to return NULL if index not changed */
-	if (stats == NULL)
-		return NULL;
-
-	stats->num_pages = RelationGetNumberOfBlocks(rel);
-
-	return stats;
-}

src/ivfbuild.c

@@ -2,16 +2,11 @@
 
 #include <float.h>
 
-#include "access/parallel.h"
-#include "access/xact.h"
 #include "catalog/index.h"
-#include "catalog/pg_operator_d.h"
-#include "catalog/pg_type_d.h"
 #include "ivfflat.h"
 #include "miscadmin.h"
 #include "storage/bufmgr.h"
 #include "utils/memutils.h"
-#include "tcop/tcopprot.h"
 
 #if PG_VERSION_NUM >= 140000
 #include "utils/backend_progress.h"
@@ -28,6 +23,9 @@
 #define PROGRESS_CREATEIDX_TUPLES_DONE 0
 #endif
 
+#include "catalog/pg_operator_d.h"
+#include "catalog/pg_type_d.h"
+
 #if PG_VERSION_NUM >= 130000
 #define CALLBACK_ITEM_POINTER ItemPointer tid
 #else
@@ -40,25 +38,6 @@
 #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)
-#define PARALLEL_KEY_QUERY_TEXT			UINT64CONST(0xA000000000000004)
-
 /*
  * Add sample
  */
@@ -168,9 +147,10 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
 {
 	double		distance;
 	double		minDistance = DBL_MAX;
-	int			closestCenter = 0;
+	int			closestCenter = -1;
 	VectorArray centers = buildstate->centers;
 	TupleTableSlot *slot = buildstate->slot;
+	int			i;
 
 	/* Detoast once for all calls */
 	Datum		value = PointerGetDatum(PG_DETOAST_DATUM(values[0]));
@@ -183,7 +163,7 @@ AddTupleToSort(Relation index, ItemPointer tid, Datum *values, IvfflatBuildState
 	}
 
 	/* Find the list that minimizes the distance */
-	for (int i = 0; i < centers->length; i++)
+	for (i = 0; i < centers->length; i++)
 	{
 		distance = DatumGetFloat8(FunctionCall2Coll(buildstate->procinfo, buildstate->collation, value, PointerGetDatum(VectorArrayGet(centers, i))));
 
@@ -278,8 +258,15 @@ GetNextTuple(Tuplesortstate *sortstate, TupleDesc tupdesc, TupleTableSlot *slot,
 static void
 InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
 {
+	Buffer		buf;
+	Page		page;
+	GenericXLogState *state;
 	int			list;
 	IndexTuple	itup = NULL;	/* silence compiler warning */
+	BlockNumber startPage;
+	BlockNumber insertPage;
+	Size		itemsz;
+	int			i;
 	int64		inserted = 0;
 
 #if PG_VERSION_NUM >= 120000
@@ -295,14 +282,8 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
 
 	GetNextTuple(buildstate->sortstate, tupdesc, slot, &itup, &list);
 
-	for (int i = 0; i < buildstate->centers->length; i++)
+	for (i = 0; i < buildstate->centers->length; i++)
 	{
-		Buffer		buf;
-		Page		page;
-		GenericXLogState *state;
-		BlockNumber startPage;
-		BlockNumber insertPage;
-
 		/* Can take a while, so ensure we can interrupt */
 		/* Needs to be called when no buffer locks are held */
 		CHECK_FOR_INTERRUPTS();
@@ -316,8 +297,7 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
 		while (list == i)
 		{
 			/* Check for free space */
-			Size		itemsz = MAXALIGN(IndexTupleSize(itup));
-
+			itemsz = MAXALIGN(IndexTupleSize(itup));
 			if (PageGetFreeSpace(page) < itemsz)
 				IvfflatAppendPage(index, &buf, &page, &state, forkNum);
 
@@ -337,7 +317,7 @@ InsertTuples(Relation index, IvfflatBuildState * buildstate, ForkNumber forkNum)
 		IvfflatCommitBuffer(buf, state);
 
 		/* Set the start and insert pages */
-		IvfflatUpdateList(index, buildstate->listInfo[i], insertPage, InvalidBlockNumber, startPage, forkNum);
+		IvfflatUpdateList(index, state, buildstate->listInfo[i], insertPage, InvalidBlockNumber, startPage, forkNum);
 	}
 }
 
@@ -371,7 +351,9 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 	buildstate->collation = index->rd_indcollation[0];
 
 	/* Require more than one dimension for spherical k-means */
-	if (buildstate->kmeansnormprocinfo != NULL && buildstate->dimensions == 1)
+	/* Lists check for backwards compatibility */
+	/* TODO Remove lists check in 0.3.0 */
+	if (buildstate->kmeansnormprocinfo != NULL && buildstate->dimensions == 1 && buildstate->lists > 1)
 		elog(ERROR, "dimensions must be greater than one for this opclass");
 
 	/* Create tuple description for sorting */
@@ -405,8 +387,6 @@ InitBuildState(IvfflatBuildState * buildstate, Relation heap, Relation index, In
 	buildstate->listSums = palloc0(sizeof(double) * buildstate->lists);
 	buildstate->listCounts = palloc0(sizeof(int) * buildstate->lists);
 #endif
-
-	buildstate->ivfleader = NULL;
 }
 
 /*
@@ -458,8 +438,8 @@ ComputeCenters(IvfflatBuildState * buildstate)
 		{
 			ereport(NOTICE,
 					(errmsg("ivfflat index created with little data"),
-					 errdetail("This will cause low recall."),
-					 errhint("Drop the index until the table has more data.")));
+					 errdetail("this will cause poor recall"),
+					 errhint("drop the index until the table has more data")));
 		}
 	}
 
@@ -503,6 +483,7 @@ static void
 CreateListPages(Relation index, VectorArray centers, int dimensions,
 				int lists, ForkNumber forkNum, ListInfo * *listInfo)
 {
+	int			i;
 	Buffer		buf;
 	Page		page;
 	GenericXLogState *state;
@@ -516,7 +497,7 @@ CreateListPages(Relation index, VectorArray centers, int dimensions,
 	buf = IvfflatNewBuffer(index, forkNum);
 	IvfflatInitRegisterPage(index, &buf, &page, &state);
 
-	for (int i = 0; i < lists; i++)
+	for (i = 0; i < lists; i++)
 	{
 		/* Load list */
 		list->startPage = InvalidBlockNumber;
@@ -552,7 +533,7 @@ PrintKmeansMetrics(IvfflatBuildState * buildstate)
 	elog(INFO, "inertia: %.3e", buildstate->inertia);
 
 	/* Calculate Davies-Bouldin index */
-	if (buildstate->lists > 1 && !buildstate->ivfleader)
+	if (buildstate->lists > 1)
 	{
 		double		db = 0.0;
 
@@ -587,478 +568,43 @@ PrintKmeansMetrics(IvfflatBuildState * buildstate)
 }
 #endif
 
-/*
- * Within leader, wait for end of heap scan
- */
-static double
-ParallelHeapScan(IvfflatBuildState * buildstate)
-{
-	IvfflatShared *ivfshared = buildstate->ivfleader->ivfshared;
-	int			nparticipanttuplesorts;
-	double		reltuples;
-
-	nparticipanttuplesorts = buildstate->ivfleader->nparticipanttuplesorts;
-	for (;;)
-	{
-		SpinLockAcquire(&ivfshared->mutex);
-		if (ivfshared->nparticipantsdone == nparticipanttuplesorts)
-		{
-			buildstate->indtuples = ivfshared->indtuples;
-			reltuples = ivfshared->reltuples;
-#ifdef IVFFLAT_KMEANS_DEBUG
-			buildstate->inertia = ivfshared->inertia;
-#endif
-			SpinLockRelease(&ivfshared->mutex);
-			break;
-		}
-		SpinLockRelease(&ivfshared->mutex);
-
-		ConditionVariableSleep(&ivfshared->workersdonecv,
-							   WAIT_EVENT_PARALLEL_CREATE_INDEX_SCAN);
-	}
-
-	ConditionVariableCancelSleep();
-
-	return reltuples;
-}
-
-/*
- * Perform a worker's portion of a parallel sort
- */
-static void
-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;
-	coordinate->nParticipants = -1;
-	coordinate->sharedsort = sharedsort;
-
-	/* Join parallel scan */
-	indexInfo = BuildIndexInfo(ivfspool->index);
-	indexInfo->ii_Concurrent = ivfshared->isconcurrent;
-	InitBuildState(&buildstate, ivfspool->heap, ivfspool->index, indexInfo);
-	memcpy(buildstate.centers->items, ivfcenters, VECTOR_SIZE(buildstate.centers->dim) * buildstate.centers->maxlen);
-	buildstate.centers->length = buildstate.centers->maxlen;
-	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));
-	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);
-
-	/* Record statistics */
-	SpinLockAcquire(&ivfshared->mutex);
-	ivfshared->nparticipantsdone++;
-	ivfshared->reltuples += reltuples;
-	ivfshared->indtuples += buildstate.indtuples;
-#ifdef IVFFLAT_KMEANS_DEBUG
-	ivfshared->inertia += buildstate.inertia;
-#endif
-	SpinLockRelease(&ivfshared->mutex);
-
-	/* Log statistics */
-	if (progress)
-		ereport(DEBUG1, (errmsg("leader processed " INT64_FORMAT " tuples", (int64) reltuples)));
-	else
-		ereport(DEBUG1, (errmsg("worker processed " INT64_FORMAT " tuples", (int64) reltuples)));
-
-	/* Notify leader */
-	ConditionVariableSignal(&ivfshared->workersdonecv);
-
-	/* We can end tuplesorts immediately */
-	tuplesort_end(ivfspool->sortstate);
-
-	FreeBuildState(&buildstate);
-}
-
-/*
- * Perform work within a launched parallel process
- */
-void
-IvfflatParallelBuildMain(dsm_segment *seg, shm_toc *toc)
-{
-	char	   *sharedquery;
-	IvfflatSpool *ivfspool;
-	IvfflatShared *ivfshared;
-	Sharedsort *sharedsort;
-	Vector	   *ivfcenters;
-	Relation	heapRel;
-	Relation	indexRel;
-	LOCKMODE	heapLockmode;
-	LOCKMODE	indexLockmode;
-	int			sortmem;
-
-	/* Set debug_query_string for individual workers first */
-	sharedquery = shm_toc_lookup(toc, PARALLEL_KEY_QUERY_TEXT, true);
-	debug_query_string = sharedquery;
-
-	/* Report the query string from leader */
-	pgstat_report_activity(STATE_RUNNING, debug_query_string);
-
-	/* Look up shared state */
-	ivfshared = shm_toc_lookup(toc, PARALLEL_KEY_IVFFLAT_SHARED, false);
-
-	/* Open relations using lock modes known to be obtained by index.c */
-	if (!ivfshared->isconcurrent)
-	{
-		heapLockmode = ShareLock;
-		indexLockmode = AccessExclusiveLock;
-	}
-	else
-	{
-		heapLockmode = ShareUpdateExclusiveLock;
-		indexLockmode = RowExclusiveLock;
-	}
-
-	/* 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 */
-	ivfspool = (IvfflatSpool *) palloc0(sizeof(IvfflatSpool));
-	ivfspool->heap = heapRel;
-	ivfspool->index = indexRel;
-
-	/* Look up shared state private to tuplesort.c */
-	sharedsort = shm_toc_lookup(toc, PARALLEL_KEY_TUPLESORT, false);
-	tuplesort_attach_shared(sharedsort, seg);
-
-	ivfcenters = shm_toc_lookup(toc, PARALLEL_KEY_IVFFLAT_CENTERS, false);
-
-	/* Perform sorting */
-	sortmem = maintenance_work_mem / ivfshared->scantuplesortstates;
-	IvfflatParallelScanAndSort(ivfspool, ivfshared, sharedsort, ivfcenters, sortmem, false);
-
-	/* Close relations within worker */
-	index_close(indexRel, indexLockmode);
-#if PG_VERSION_NUM >= 120000
-	table_close(heapRel, heapLockmode);
-#else
-	heap_close(heapRel, heapLockmode);
-#endif
-}
-
-/*
- * End parallel build
- */
-static void
-IvfflatEndParallel(IvfflatLeader * ivfleader)
-{
-	/* Shutdown worker processes */
-	WaitForParallelWorkersToFinish(ivfleader->pcxt);
-
-	/* Free last reference to MVCC snapshot, if one was used */
-	if (IsMVCCSnapshot(ivfleader->snapshot))
-		UnregisterSnapshot(ivfleader->snapshot);
-	DestroyParallelContext(ivfleader->pcxt);
-	ExitParallelMode();
-}
-
-/*
- * Return size of shared memory required for parallel index build
- */
-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
-}
-
-/*
- * Within leader, participate as a parallel worker
- */
-static void
-IvfflatLeaderParticipateAsWorker(IvfflatBuildState * buildstate)
-{
-	IvfflatLeader *ivfleader = buildstate->ivfleader;
-	IvfflatSpool *leaderworker;
-	int			sortmem;
-
-	/* Allocate memory and initialize private spool */
-	leaderworker = (IvfflatSpool *) palloc0(sizeof(IvfflatSpool));
-	leaderworker->heap = buildstate->heap;
-	leaderworker->index = buildstate->index;
-
-	/* Perform work common to all participants */
-	sortmem = maintenance_work_mem / ivfleader->nparticipanttuplesorts;
-	IvfflatParallelScanAndSort(leaderworker, ivfleader->ivfshared,
-							   ivfleader->sharedsort, ivfleader->ivfcenters,
-							   sortmem, true);
-}
-
-/*
- * Begin parallel build
- */
-static void
-IvfflatBeginParallel(IvfflatBuildState * buildstate, bool isconcurrent, int request)
-{
-	ParallelContext *pcxt;
-	int			scantuplesortstates;
-	Snapshot	snapshot;
-	Size		estivfshared;
-	Size		estsort;
-	Size		estcenters;
-	IvfflatShared *ivfshared;
-	Sharedsort *sharedsort;
-	Vector	   *ivfcenters;
-	IvfflatLeader *ivfleader = (IvfflatLeader *) palloc0(sizeof(IvfflatLeader));
-	bool		leaderparticipates = true;
-	int			querylen;
-
-#ifdef DISABLE_LEADER_PARTICIPATION
-	leaderparticipates = false;
-#endif
-
-	/* 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;
-
-	/* Get snapshot for table scan */
-	if (!isconcurrent)
-		snapshot = SnapshotAny;
-	else
-		snapshot = RegisterSnapshot(GetTransactionSnapshot());
-
-	/* Estimate size of workspaces */
-	estivfshared = ParallelEstimateShared(buildstate->heap, snapshot);
-	shm_toc_estimate_chunk(&pcxt->estimator, estivfshared);
-	estsort = tuplesort_estimate_shared(scantuplesortstates);
-	shm_toc_estimate_chunk(&pcxt->estimator, estsort);
-	estcenters = VECTOR_SIZE(buildstate->dimensions) * buildstate->lists;
-	shm_toc_estimate_chunk(&pcxt->estimator, estcenters);
-	shm_toc_estimate_keys(&pcxt->estimator, 3);
-
-	/* Finally, estimate PARALLEL_KEY_QUERY_TEXT space */
-	if (debug_query_string)
-	{
-		querylen = strlen(debug_query_string);
-		shm_toc_estimate_chunk(&pcxt->estimator, querylen + 1);
-		shm_toc_estimate_keys(&pcxt->estimator, 1);
-	}
-	else
-		querylen = 0;			/* keep compiler quiet */
-
-	/* Everyone's had a chance to ask for space, so now create the DSM */
-	InitializeParallelDSM(pcxt);
-
-	/* If no DSM segment was available, back out (do serial build) */
-	if (pcxt->seg == NULL)
-	{
-		if (IsMVCCSnapshot(snapshot))
-			UnregisterSnapshot(snapshot);
-		DestroyParallelContext(pcxt);
-		ExitParallelMode();
-		return;
-	}
-
-	/* Store shared build state, for which we reserved space */
-	ivfshared = (IvfflatShared *) shm_toc_allocate(pcxt->toc, estivfshared);
-	/* Initialize immutable state */
-	ivfshared->heaprelid = RelationGetRelid(buildstate->heap);
-	ivfshared->indexrelid = RelationGetRelid(buildstate->index);
-	ivfshared->isconcurrent = isconcurrent;
-	ivfshared->scantuplesortstates = scantuplesortstates;
-	ConditionVariableInit(&ivfshared->workersdonecv);
-	SpinLockInit(&ivfshared->mutex);
-	/* Initialize mutable state */
-	ivfshared->nparticipantsdone = 0;
-	ivfshared->reltuples = 0;
-	ivfshared->indtuples = 0;
-#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 = (Vector *) shm_toc_allocate(pcxt->toc, estcenters);
-	memcpy(ivfcenters, buildstate->centers->items, estcenters);
-
-	shm_toc_insert(pcxt->toc, PARALLEL_KEY_IVFFLAT_SHARED, ivfshared);
-	shm_toc_insert(pcxt->toc, PARALLEL_KEY_TUPLESORT, sharedsort);
-	shm_toc_insert(pcxt->toc, PARALLEL_KEY_IVFFLAT_CENTERS, ivfcenters);
-
-	/* Store query string for workers */
-	if (debug_query_string)
-	{
-		char	   *sharedquery;
-
-		sharedquery = (char *) shm_toc_allocate(pcxt->toc, querylen + 1);
-		memcpy(sharedquery, debug_query_string, querylen + 1);
-		shm_toc_insert(pcxt->toc, PARALLEL_KEY_QUERY_TEXT, sharedquery);
-	}
-
-	/* Launch workers, saving status for leader/caller */
-	LaunchParallelWorkers(pcxt);
-	ivfleader->pcxt = pcxt;
-	ivfleader->nparticipanttuplesorts = pcxt->nworkers_launched;
-	if (leaderparticipates)
-		ivfleader->nparticipanttuplesorts++;
-	ivfleader->ivfshared = ivfshared;
-	ivfleader->sharedsort = sharedsort;
-	ivfleader->snapshot = snapshot;
-	ivfleader->ivfcenters = ivfcenters;
-
-	/* If no workers were successfully launched, back out (do serial build) */
-	if (pcxt->nworkers_launched == 0)
-	{
-		IvfflatEndParallel(ivfleader);
-		return;
-	}
-
-	/* Log participants */
-	ereport(DEBUG1, (errmsg("using %d parallel workers", pcxt->nworkers_launched)));
-
-	/* Save leader state now that it's clear build will be parallel */
-	buildstate->ivfleader = ivfleader;
-
-	/* Join heap scan ourselves */
-	if (leaderparticipates)
-		IvfflatLeaderParticipateAsWorker(buildstate);
-
-	/* Wait for all launched workers */
-	WaitForParallelWorkersToAttach(pcxt);
-}
-
-/*
- * Scan table for tuples to index
- */
-static void
-AssignTuples(IvfflatBuildState * buildstate)
-{
-	int			parallel_workers = 0;
-	SortCoordinate coordinate = NULL;
-
-	/* 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)
-		parallel_workers = plan_create_index_workers(RelationGetRelid(buildstate->heap), RelationGetRelid(buildstate->index));
-
-	/* Attempt to launch parallel worker scan when required */
-	if (parallel_workers > 0)
-		IvfflatBeginParallel(buildstate, buildstate->indexInfo->ii_Concurrent, parallel_workers);
-
-	/* Set up coordination state if at least one worker launched */
-	if (buildstate->ivfleader)
-	{
-		coordinate = (SortCoordinate) palloc0(sizeof(SortCoordinateData));
-		coordinate->isWorker = false;
-		coordinate->nParticipants = buildstate->ivfleader->nparticipanttuplesorts;
-		coordinate->sharedsort = buildstate->ivfleader->sharedsort;
-	}
-
-	/* Begin serial/leader tuplesort */
-	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)
-	{
-		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);
-#endif
-	}
-}
-
 /*
  * Create entry pages
  */
 static void
 CreateEntryPages(IvfflatBuildState * buildstate, ForkNumber forkNum)
 {
-	/* Assign */
-	IvfflatBench("assign tuples", AssignTuples(buildstate));
+	AttrNumber	attNums[] = {1};
+	Oid			sortOperators[] = {Float8LessOperator};
+	Oid			sortCollations[] = {InvalidOid};
+	bool		nullsFirstFlags[] = {false};
+
+	UpdateProgress(PROGRESS_CREATEIDX_SUBPHASE, PROGRESS_IVFFLAT_PHASE_SORT);
+
+	buildstate->sortstate = tuplesort_begin_heap(buildstate->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, maintenance_work_mem, NULL, false);
+
+	/* Add tuples to sort */
+	if (buildstate->heap != NULL)
+	{
+#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
+	}
 
 	/* Sort */
-	IvfflatBench("sort tuples", tuplesort_performsort(buildstate->sortstate));
+	tuplesort_performsort(buildstate->sortstate);
 
-	/* Load */
-	IvfflatBench("load tuples", InsertTuples(buildstate->index, buildstate, forkNum));
+#ifdef IVFFLAT_KMEANS_DEBUG
+	PrintKmeansMetrics(buildstate);
+#endif
 
-	/* End sort */
+	/* Insert */
+	InsertTuples(buildstate->index, buildstate, forkNum);
 	tuplesort_end(buildstate->sortstate);
-
-	/* End parallel build */
-	if (buildstate->ivfleader)
-		IvfflatEndParallel(buildstate->ivfleader);
 }
 
 /*
@@ -1075,7 +621,7 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
 	/* Create pages */
 	CreateMetaPage(index, buildstate->dimensions, buildstate->lists, forkNum);
 	CreateListPages(index, buildstate->centers, buildstate->dimensions, buildstate->lists, forkNum, &buildstate->listInfo);
-	CreateEntryPages(buildstate, forkNum);
+	IvfflatBench("CreateEntryPages", CreateEntryPages(buildstate, forkNum));
 
 	FreeBuildState(buildstate);
 }

src/ivfflat.c

@@ -7,7 +7,6 @@
 #include "ivfflat.h"
 #include "utils/guc.h"
 #include "utils/selfuncs.h"
-#include "utils/spccache.h"
 
 #if PG_VERSION_NUM >= 120000
 #include "commands/progress.h"
@@ -20,7 +19,7 @@ static relopt_kind ivfflat_relopt_kind;
  * Initialize index options and variables
  */
 void
-IvfflatInit(void)
+_PG_init(void)
 {
 	ivfflat_relopt_kind = add_reloption_kind();
 	add_int_reloption(ivfflat_relopt_kind, "lists", "Number of inverted lists",
@@ -48,8 +47,8 @@ ivfflatbuildphasename(int64 phasenum)
 			return "initializing";
 		case PROGRESS_IVFFLAT_PHASE_KMEANS:
 			return "performing k-means";
-		case PROGRESS_IVFFLAT_PHASE_ASSIGN:
-			return "assigning tuples";
+		case PROGRESS_IVFFLAT_PHASE_SORT:
+			return "sorting tuples";
 		case PROGRESS_IVFFLAT_PHASE_LOAD:
 			return "loading tuples";
 		default:
@@ -64,13 +63,13 @@ ivfflatbuildphasename(int64 phasenum)
 static void
 ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 					Cost *indexStartupCost, Cost *indexTotalCost,
-					Selectivity *indexSelectivity, double *indexCorrelation,
-					double *indexPages)
+					Selectivity *indexSelectivity, double *indexCorrelation
+					,double *indexPages
+)
 {
 	GenericCosts costs;
 	int			lists;
 	double		ratio;
-	double		spc_seq_page_cost;
 	Relation	indexRel;
 #if PG_VERSION_NUM < 120000
 	List	   *qinfos;
@@ -89,22 +88,6 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 
 	MemSet(&costs, 0, sizeof(costs));
 
-	indexRel = index_open(path->indexinfo->indexoid, NoLock);
-	lists = IvfflatGetLists(indexRel);
-	index_close(indexRel, NoLock);
-
-	/* Get the ratio of lists that we need to visit */
-	ratio = ((double) ivfflat_probes) / lists;
-	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
@@ -112,31 +95,17 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	genericcostestimate(root, path, loop_count, qinfos, &costs);
 #endif
 
-	get_tablespace_page_costs(path->indexinfo->reltablespace, NULL, &spc_seq_page_cost);
+	indexRel = index_open(path->indexinfo->indexoid, NoLock);
+	lists = IvfflatGetLists(indexRel);
+	index_close(indexRel, NoLock);
 
-	/* Adjust cost if needed since TOAST not included in seq scan cost */
-	if (costs.numIndexPages > path->indexinfo->rel->pages && ratio < 0.5)
-	{
-		/* Change all page cost from random to sequential */
-		costs.indexTotalCost -= costs.numIndexPages * (costs.spc_random_page_cost - spc_seq_page_cost);
+	ratio = ((double) ivfflat_probes) / lists;
+	if (ratio > 1)
+		ratio = 1;
 
-		/* Remove cost of extra pages */
-		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);
-	}
+	costs.indexTotalCost *= ratio;
 
-	/*
-	 * 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 */
+	/* Startup cost and total cost are same */
 	*indexStartupCost = costs.indexTotalCost;
 	*indexTotalCost = costs.indexTotalCost;
 	*indexSelectivity = costs.indexSelectivity;

src/ivfflat.h

@@ -3,11 +3,14 @@
 
 #include "postgres.h"
 
+#if PG_VERSION_NUM < 110000
+#error "Requires PostgreSQL 11+"
+#endif
+
 #include "access/generic_xlog.h"
-#include "access/parallel.h"
 #include "access/reloptions.h"
 #include "nodes/execnodes.h"
-#include "port.h"				/* for random() */
+#include "port.h"				/* for strtof() and random() */
 #include "utils/sampling.h"
 #include "utils/tuplesort.h"
 #include "vector.h"
@@ -16,10 +19,6 @@
 #include "common/pg_prng.h"
 #endif
 
-#if PG_VERSION_NUM < 120000
-#include "access/relscan.h"
-#endif
-
 #ifdef IVFFLAT_BENCH
 #include "portability/instr_time.h"
 #endif
@@ -46,7 +45,7 @@
 /* Build phases */
 /* PROGRESS_CREATEIDX_SUBPHASE_INITIALIZE is 1 */
 #define PROGRESS_IVFFLAT_PHASE_KMEANS	2
-#define PROGRESS_IVFFLAT_PHASE_ASSIGN	3
+#define PROGRESS_IVFFLAT_PHASE_SORT		3
 #define PROGRESS_IVFFLAT_PHASE_LOAD		4
 
 #define IVFFLAT_LIST_SIZE(_dim)	(offsetof(IvfflatListData, center) + VECTOR_SIZE(_dim))
@@ -80,6 +79,9 @@
 /* Variables */
 extern int	ivfflat_probes;
 
+/* Exported functions */
+PGDLLEXPORT void _PG_init(void);
+
 typedef struct VectorArrayData
 {
 	int			length;
@@ -103,56 +105,6 @@ typedef struct IvfflatOptions
 	int			lists;			/* number of lists */
 }			IvfflatOptions;
 
-typedef struct IvfflatSpool
-{
-	Tuplesortstate *sortstate;
-	Relation	heap;
-	Relation	index;
-}			IvfflatSpool;
-
-typedef struct IvfflatShared
-{
-	/* Immutable state */
-	Oid			heaprelid;
-	Oid			indexrelid;
-	bool		isconcurrent;
-	int			scantuplesortstates;
-
-	/* Worker progress */
-	ConditionVariable workersdonecv;
-
-	/* Mutex for mutable state */
-	slock_t		mutex;
-
-	/* Mutable state */
-	int			nparticipantsdone;
-	double		reltuples;
-	double		indtuples;
-
-#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
-{
-	ParallelContext *pcxt;
-	int			nparticipanttuplesorts;
-	IvfflatShared *ivfshared;
-	Sharedsort *sharedsort;
-	Snapshot	snapshot;
-	Vector	   *ivfcenters;
-}			IvfflatLeader;
-
 typedef struct IvfflatBuildState
 {
 	/* Info */
@@ -198,9 +150,6 @@ typedef struct IvfflatBuildState
 
 	/* Memory */
 	MemoryContext tmpCtx;
-
-	/* Parallel builds */
-	IvfflatLeader *ivfleader;
 }			IvfflatBuildState;
 
 typedef struct IvfflatMetaPageData
@@ -238,6 +187,14 @@ typedef struct IvfflatScanList
 	double		distance;
 }			IvfflatScanList;
 
+typedef struct IvfflatScanItem
+{
+	pairingheap_node ph_node;
+	BlockNumber searchPage;
+	double		distance;
+	ItemPointerData tid;
+}			IvfflatScanItem;
+
 typedef struct IvfflatScanOpaqueData
 {
 	int			probes;
@@ -255,6 +212,13 @@ typedef struct IvfflatScanOpaqueData
 	FmgrInfo   *normprocinfo;
 	Oid			collation;
 
+	/* Items */
+	int			maxItems;
+	int			itemCount;
+	pairingheap *itemQueue;
+	IvfflatScanItem *items;
+	IvfflatScanItem **sortedItems;
+
 	/* Lists */
 	pairingheap *listQueue;
 	IvfflatScanList lists[FLEXIBLE_ARRAY_MEMBER];	/* must come last */
@@ -275,14 +239,12 @@ void		IvfflatKmeans(Relation index, VectorArray samples, VectorArray centers);
 FmgrInfo   *IvfflatOptionalProcInfo(Relation rel, uint16 procnum);
 bool		IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result);
 int			IvfflatGetLists(Relation index);
-void		IvfflatUpdateList(Relation index, ListInfo listInfo, BlockNumber insertPage, BlockNumber originalInsertPage, BlockNumber startPage, ForkNumber forkNum);
+void		IvfflatUpdateList(Relation index, GenericXLogState *state, ListInfo listInfo, BlockNumber insertPage, BlockNumber originalInsertPage, BlockNumber startPage, ForkNumber forkNum);
 void		IvfflatCommitBuffer(Buffer buf, GenericXLogState *state);
 void		IvfflatAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **state, ForkNumber forkNum);
 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);
-PGDLLEXPORT void IvfflatParallelBuildMain(dsm_segment *seg, shm_toc *toc);
 
 /* Index access methods */
 IndexBuildResult *ivfflatbuild(Relation heap, Relation index, IndexInfo *indexInfo);

src/ivfinsert.c

@@ -4,7 +4,6 @@
 
 #include "ivfflat.h"
 #include "storage/bufmgr.h"
-#include "storage/lmgr.h"
 #include "utils/memutils.h"
 
 /*
@@ -24,10 +23,6 @@ FindInsertPage(Relation rel, Datum *values, BlockNumber *insertPage, ListInfo *
 	OffsetNumber offno;
 	OffsetNumber maxoffno;
 
-	/* Avoid compiler warning */
-	listInfo->blkno = nextblkno;
-	listInfo->offno = FirstOffsetNumber;
-
 	procinfo = index_getprocinfo(rel, 1, IVFFLAT_DISTANCE_PROC);
 	collation = rel->rd_indcollation[0];
 
@@ -44,7 +39,7 @@ FindInsertPage(Relation rel, Datum *values, BlockNumber *insertPage, ListInfo *
 			list = (IvfflatList) PageGetItem(cpage, PageGetItemId(cpage, offno));
 			distance = DatumGetFloat8(FunctionCall2Coll(procinfo, collation, values[0], PointerGetDatum(&list->center)));
 
-			if (distance < minDistance || !BlockNumberIsValid(*insertPage))
+			if (distance < minDistance)
 			{
 				*insertPage = list->insertPage;
 				listInfo->blkno = nextblkno;
@@ -122,16 +117,23 @@ InsertTuple(Relation rel, Datum *values, bool *isnull, ItemPointer heap_tid, Rel
 		}
 		else
 		{
+			Buffer		metabuf;
 			Buffer		newbuf;
 			Page		newpage;
 
+			/*
+			 * From ReadBufferExtended: Caller is responsible for ensuring
+			 * that only one backend tries to extend a relation at the same
+			 * time!
+			 */
+			metabuf = ReadBuffer(rel, IVFFLAT_METAPAGE_BLKNO);
+			LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
+
 			/* Add a new page */
-			LockRelationForExtension(rel, ExclusiveLock);
 			newbuf = IvfflatNewBuffer(rel, MAIN_FORKNUM);
-			UnlockRelationForExtension(rel, ExclusiveLock);
+			newpage = GenericXLogRegisterBuffer(state, newbuf, GENERIC_XLOG_FULL_IMAGE);
 
 			/* Init new page */
-			newpage = GenericXLogRegisterBuffer(state, newbuf, GENERIC_XLOG_FULL_IMAGE);
 			IvfflatInitPage(newbuf, newpage);
 
 			/* Update insert page */
@@ -145,6 +147,9 @@ InsertTuple(Relation rel, Datum *values, bool *isnull, ItemPointer heap_tid, Rel
 			MarkBufferDirty(buf);
 			GenericXLogFinish(state);
 
+			/* Unlock extend relation lock as early as possible */
+			UnlockReleaseBuffer(metabuf);
+
 			/* Unlock previous buffer */
 			UnlockReleaseBuffer(buf);
 
@@ -164,7 +169,7 @@ InsertTuple(Relation rel, Datum *values, bool *isnull, ItemPointer heap_tid, Rel
 
 	/* Update the insert page */
 	if (insertPage != originalInsertPage)
-		IvfflatUpdateList(rel, listInfo, insertPage, originalInsertPage, InvalidBlockNumber, MAIN_FORKNUM);
+		IvfflatUpdateList(rel, state, listInfo, insertPage, originalInsertPage, InvalidBlockNumber, MAIN_FORKNUM);
 }
 
 /*

src/ivfkmeans.c

@@ -1,7 +1,6 @@
 #include "postgres.h"
 
 #include <float.h>
-#include <math.h>
 
 #include "ivfflat.h"
 #include "miscadmin.h"
@@ -16,6 +15,7 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, float *low
 {
 	FmgrInfo   *procinfo;
 	Oid			collation;
+	int			i;
 	int64		j;
 	double		distance;
 	double		sum;
@@ -35,7 +35,7 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, float *low
 	for (j = 0; j < numSamples; j++)
 		weight[j] = DBL_MAX;
 
-	for (int i = 0; i < numCenters; i++)
+	for (i = 0; i < numCenters; i++)
 	{
 		CHECK_FOR_INTERRUPTS();
 
@@ -87,12 +87,13 @@ InitCenters(Relation index, VectorArray samples, VectorArray centers, float *low
 static inline void
 ApplyNorm(FmgrInfo *normprocinfo, Oid collation, Vector * vec)
 {
+	int			i;
 	double		norm = DatumGetFloat8(FunctionCall1Coll(normprocinfo, collation, PointerGetDatum(vec)));
 
 	/* TODO Handle zero norm */
 	if (norm > 0)
 	{
-		for (int i = 0; i < vec->dim; i++)
+		for (i = 0; i < vec->dim; i++)
 			vec->x[i] /= norm;
 	}
 }
@@ -112,6 +113,8 @@ CompareVectors(const void *a, const void *b)
 static void
 QuickCenters(Relation index, VectorArray samples, VectorArray centers)
 {
+	int			i;
+	int			j;
 	Vector	   *vec;
 	int			dimensions = centers->dim;
 	Oid			collation = index->rd_indcollation[0];
@@ -121,7 +124,7 @@ QuickCenters(Relation index, VectorArray samples, VectorArray centers)
 	if (samples->length > 0)
 	{
 		qsort(samples->items, samples->length, VECTOR_SIZE(samples->dim), CompareVectors);
-		for (int i = 0; i < samples->length; i++)
+		for (i = 0; i < samples->length; i++)
 		{
 			vec = VectorArrayGet(samples, i);
 
@@ -141,7 +144,7 @@ QuickCenters(Relation index, VectorArray samples, VectorArray centers)
 		SET_VARSIZE(vec, VECTOR_SIZE(dimensions));
 		vec->dim = dimensions;
 
-		for (int j = 0; j < dimensions; j++)
+		for (j = 0; j < dimensions; j++)
 			vec->x[j] = RandomDouble();
 
 		/* Normalize if needed (only needed for random centers) */
@@ -208,7 +211,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 
 	/* Check memory requirements */
 	/* Add one to error message to ceil */
-	if (totalSize > (Size) maintenance_work_mem * 1024L)
+	if (totalSize / 1024 > maintenance_work_mem)
 		ereport(ERROR,
 				(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
 				 errmsg("memory required is %zu MB, maintenance_work_mem is %d MB",
@@ -248,7 +251,7 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 	for (j = 0; j < numSamples; j++)
 	{
 		minDistance = DBL_MAX;
-		closestCenter = 0;
+		closestCenter = -1;
 
 		/* Find closest center */
 		for (k = 0; k < numCenters; k++)
@@ -395,14 +398,6 @@ ElkanKmeans(Relation index, VectorArray samples, VectorArray centers)
 
 			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];
 			}
@@ -466,31 +461,16 @@ CheckCenters(Relation index, VectorArray centers)
 {
 	FmgrInfo   *normprocinfo;
 	Oid			collation;
-	Vector	   *vec;
+	int			i;
 	double		norm;
 
 	if (centers->length != centers->maxlen)
 		elog(ERROR, "Not enough centers. Please report a bug.");
 
-	/* Ensure no NaN or infinite values */
-	for (int i = 0; i < centers->length; i++)
-	{
-		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++)
+	for (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.");
@@ -503,7 +483,7 @@ CheckCenters(Relation index, VectorArray centers)
 	{
 		collation = index->rd_indcollation[0];
 
-		for (int i = 0; i < centers->length; i++)
+		for (i = 0; i < centers->length; i++)
 		{
 			norm = DatumGetFloat8(FunctionCall1Coll(normprocinfo, collation, PointerGetDatum(VectorArrayGet(centers, i))));
 			if (norm == 0)

src/ivfscan.c

@@ -3,13 +3,14 @@
 #include <float.h>
 
 #include "access/relscan.h"
-#include "catalog/pg_operator_d.h"
-#include "catalog/pg_type_d.h"
 #include "ivfflat.h"
 #include "miscadmin.h"
 #include "pgstat.h"
 #include "storage/bufmgr.h"
 
+#include "catalog/pg_operator_d.h"
+#include "catalog/pg_type_d.h"
+
 /*
  * Compare list distances
  */
@@ -25,6 +26,21 @@ CompareLists(const pairingheap_node *a, const pairingheap_node *b, void *arg)
 	return 0;
 }
 
+/*
+ * Compare item distances
+ */
+static int
+CompareItems(const pairingheap_node *a, const pairingheap_node *b, void *arg)
+{
+	if (((const IvfflatScanItem *) a)->distance > ((const IvfflatScanItem *) b)->distance)
+		return 1;
+
+	if (((const IvfflatScanItem *) a)->distance < ((const IvfflatScanItem *) b)->distance)
+		return -1;
+
+	return 0;
+}
+
 /*
  * Get lists and sort by distance
  */
@@ -110,13 +126,10 @@ GetScanItems(IndexScanDesc scan, Datum value)
 	Datum		datum;
 	bool		isnull;
 	TupleDesc	tupdesc = RelationGetDescr(scan->indexRelation);
-	double		tuples = 0;
-
-#if PG_VERSION_NUM >= 120000
-	TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc, &TTSOpsVirtual);
-#else
-	TupleTableSlot *slot = MakeSingleTupleTableSlot(so->tupdesc);
-#endif
+	int			i;
+	double		distance;
+	IvfflatScanItem *scanitem;
+	double		maxDistance = DBL_MAX;
 
 	/*
 	 * Reuse same set of shared buffers for scan
@@ -142,25 +155,40 @@ GetScanItems(IndexScanDesc scan, Datum value)
 			{
 				itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offno));
 				datum = index_getattr(itup, 1, tupdesc, &isnull);
+				distance = DatumGetFloat8(FunctionCall2Coll(so->procinfo, so->collation, datum, value));
 
-				/*
-				 * Add virtual tuple
-				 *
-				 * Use procinfo from the index instead of scan key for
-				 * performance
-				 */
-				ExecClearTuple(slot);
-				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;
-				slot->tts_values[2] = Int32GetDatum((int) searchPage);
-				slot->tts_isnull[2] = false;
-				ExecStoreVirtualTuple(slot);
+				if (so->itemCount < so->maxItems)
+				{
+					scanitem = &so->items[so->itemCount];
+					scanitem->searchPage = searchPage;
+					scanitem->tid = itup->t_tid;
+					scanitem->distance = distance;
+					so->itemCount++;
 
-				tuplesort_puttupleslot(so->sortstate, slot);
+					/* Add to heap */
+					pairingheap_add(so->itemQueue, &scanitem->ph_node);
 
-				tuples++;
+					/* Calculate max distance */
+					if (so->itemCount == so->maxItems)
+					{
+						maxDistance = ((IvfflatScanItem *) pairingheap_first(so->itemQueue))->distance;
+						scanitem = &so->items[so->itemCount];
+					}
+				}
+				else if (distance < maxDistance)
+				{
+					/* Reuse */
+					scanitem->searchPage = searchPage;
+					scanitem->tid = itup->t_tid;
+					scanitem->distance = distance;
+					pairingheap_add(so->itemQueue, &scanitem->ph_node);
+
+					/* Remove */
+					scanitem = (IvfflatScanItem *) pairingheap_remove_first(so->itemQueue);
+
+					/* Update max distance */
+					maxDistance = ((IvfflatScanItem *) pairingheap_first(so->itemQueue))->distance;
+				}
 			}
 
 			searchPage = IvfflatPageGetOpaque(page)->nextblkno;
@@ -169,38 +197,10 @@ GetScanItems(IndexScanDesc scan, Datum value)
 		}
 	}
 
-	FreeAccessStrategy(bas);
+	for (i = 0; i < so->itemCount; i++)
+		so->sortedItems[i] = (IvfflatScanItem *) pairingheap_remove_first(so->itemQueue);
 
-	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);
-}
-
-/*
- * Get dimensions from metapage
- */
-static int
-GetDimensions(Relation index)
-{
-	Buffer		buf;
-	Page		page;
-	IvfflatMetaPage metap;
-	int			dimensions;
-
-	buf = ReadBuffer(index, IVFFLAT_METAPAGE_BLKNO);
-	LockBuffer(buf, BUFFER_LOCK_SHARE);
-	page = BufferGetPage(buf);
-	metap = IvfflatPageGetMeta(page);
-
-	dimensions = metap->dimensions;
-
-	UnlockReleaseBuffer(buf);
-
-	return dimensions;
+	Assert(pairingheap_is_empty(so->itemQueue));
 }
 
 /*
@@ -212,10 +212,6 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	IndexScanDesc scan;
 	IvfflatScanOpaque so;
 	int			lists;
-	AttrNumber	attNums[] = {1};
-	Oid			sortOperators[] = {Float8LessOperator};
-	Oid			sortCollations[] = {InvalidOid};
-	bool		nullsFirstFlags[] = {false};
 	int			probes = ivfflat_probes;
 
 	scan = RelationGetIndexScan(index, nkeys, norderbys);
@@ -234,27 +230,14 @@ ivfflatbeginscan(Relation index, int nkeys, int norderbys)
 	so->normprocinfo = IvfflatOptionalProcInfo(index, IVFFLAT_NORM_PROC);
 	so->collation = index->rd_indcollation[0];
 
-	/* Create tuple description for sorting */
-#if PG_VERSION_NUM >= 120000
-	so->tupdesc = CreateTemplateTupleDesc(3);
-#else
-	so->tupdesc = CreateTemplateTupleDesc(3, false);
-#endif
-	TupleDescInitEntry(so->tupdesc, (AttrNumber) 1, "distance", FLOAT8OID, -1, 0);
-	TupleDescInitEntry(so->tupdesc, (AttrNumber) 2, "tid", TIDOID, -1, 0);
-	TupleDescInitEntry(so->tupdesc, (AttrNumber) 3, "indexblkno", INT4OID, -1, 0);
-
-	/* Prep sort */
-	so->sortstate = tuplesort_begin_heap(so->tupdesc, 1, attNums, sortOperators, sortCollations, nullsFirstFlags, work_mem, NULL, false);
-
-#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->maxItems = 10;
+	so->itemCount = 0;
+	so->itemQueue = pairingheap_allocate(CompareItems, scan);
+	so->items = palloc(sizeof(IvfflatScanItem) * (so->maxItems + 1));
+	so->sortedItems = palloc(sizeof(IvfflatScanItem *) * so->maxItems);
+
 	scan->opaque = so;
 
 	return scan;
@@ -268,13 +251,10 @@ 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);
+	pairingheap_reset(so->itemQueue);
+	so->itemCount = 0;
 
 	if (keys && scan->numberOfKeys > 0)
 		memmove(scan->keyData, keys, scan->numberOfKeys * sizeof(ScanKeyData));
@@ -308,19 +288,21 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 		if (scan->orderByData == NULL)
 			elog(ERROR, "cannot scan ivfflat index without order");
 
+		/* No items will match if null */
 		if (scan->orderByData->sk_flags & SK_ISNULL)
-			value = PointerGetDatum(InitVector(GetDimensions(scan->indexRelation)));
-		else
+			return false;
+
+		value = scan->orderByData->sk_argument;
+
+		/* Value should not be compressed or toasted */
+		Assert(!VARATT_IS_COMPRESSED(DatumGetPointer(value)));
+		Assert(!VARATT_IS_EXTENDED(DatumGetPointer(value)));
+
+		if (so->normprocinfo != NULL)
 		{
-			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);
+			/* No items will match if normalization fails */
+			if (!IvfflatNormValue(so->normprocinfo, so->collation, &value, NULL))
+				return false;
 		}
 
 		IvfflatBench("GetScanLists", GetScanLists(scan, value));
@@ -332,15 +314,18 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 			pfree(DatumGetPointer(value));
 	}
 
-	if (tuplesort_gettupleslot(so->sortstate, true, false, so->slot, NULL))
+	if (so->itemCount > 0)
 	{
-		ItemPointer tid = (ItemPointer) DatumGetPointer(slot_getattr(so->slot, 2, &so->isnull));
-		BlockNumber indexblkno = DatumGetInt32(slot_getattr(so->slot, 3, &so->isnull));
+		IvfflatScanItem *scanitem;
+
+		so->itemCount--;
+
+		scanitem = so->sortedItems[so->itemCount];
 
 #if PG_VERSION_NUM >= 120000
-		scan->xs_heaptid = *tid;
+		scan->xs_heaptid = scanitem->tid;
 #else
-		scan->xs_ctup.t_self = *tid;
+		scan->xs_ctup.t_self = scanitem->tid;
 #endif
 
 		if (BufferIsValid(so->buf))
@@ -352,7 +337,7 @@ ivfflatgettuple(IndexScanDesc scan, ScanDirection dir)
 		 *
 		 * https://www.postgresql.org/docs/current/index-locking.html
 		 */
-		so->buf = ReadBuffer(scan->indexRelation, indexblkno);
+		so->buf = ReadBuffer(scan->indexRelation, scanitem->searchPage);
 
 		scan->xs_recheckorderby = false;
 		return true;
@@ -374,7 +359,10 @@ ivfflatendscan(IndexScanDesc scan)
 		ReleaseBuffer(so->buf);
 
 	pairingheap_free(so->listQueue);
-	tuplesort_end(so->sortstate);
+
+	pairingheap_free(so->itemQueue);
+	pfree(so->items);
+	pfree(so->sortedItems);
 
 	pfree(so);
 	scan->opaque = NULL;

src/ivfutils.c

@@ -35,7 +35,9 @@ VectorArrayFree(VectorArray arr)
 void
 PrintVectorArray(char *msg, VectorArray arr)
 {
-	for (int i = 0; i < arr->length; i++)
+	int			i;
+
+	for (i = 0; i < arr->length; i++)
 		PrintVector(msg, VectorArrayGet(arr, i));
 }
 
@@ -76,16 +78,20 @@ IvfflatOptionalProcInfo(Relation rel, uint16 procnum)
 bool
 IvfflatNormValue(FmgrInfo *procinfo, Oid collation, Datum *value, Vector * result)
 {
-	double		norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
+	Vector	   *v;
+	int			i;
+	double		norm;
+
+	norm = DatumGetFloat8(FunctionCall1Coll(procinfo, collation, *value));
 
 	if (norm > 0)
 	{
-		Vector	   *v = DatumGetVector(*value);
+		v = DatumGetVector(*value);
 
 		if (result == NULL)
 			result = InitVector(v->dim);
 
-		for (int i = 0; i < v->dim; i++)
+		for (i = 0; i < v->dim; i++)
 			result->x[i] = v->x[i] / norm;
 
 		*value = PointerGetDatum(result);
@@ -176,13 +182,12 @@ IvfflatAppendPage(Relation index, Buffer *buf, Page *page, GenericXLogState **st
  * Update the start or insert page of a list
  */
 void
-IvfflatUpdateList(Relation index, ListInfo listInfo,
+IvfflatUpdateList(Relation index, GenericXLogState *state, ListInfo listInfo,
 				  BlockNumber insertPage, BlockNumber originalInsertPage,
 				  BlockNumber startPage, ForkNumber forkNum)
 {
 	Buffer		buf;
 	Page		page;
-	GenericXLogState *state;
 	IvfflatList list;
 	bool		changed = false;
 

src/ivfvacuum.c

@@ -12,23 +12,34 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 				  IndexBulkDeleteCallback callback, void *callback_state)
 {
 	Relation	index = info->index;
-	BlockNumber blkno = IVFFLAT_HEAD_BLKNO;
+	Buffer		cbuf;
+	Page		cpage;
+	Buffer		buf;
+	Page		page;
+	IvfflatList list;
+	IndexTuple	itup;
+	ItemPointer htup;
+	OffsetNumber deletable[MaxOffsetNumber];
+	int			ndeletable;
+	BlockNumber startPages[MaxOffsetNumber];
+	BlockNumber nextblkno = IVFFLAT_HEAD_BLKNO;
+	BlockNumber searchPage;
+	BlockNumber insertPage;
+	GenericXLogState *state;
+	OffsetNumber coffno;
+	OffsetNumber cmaxoffno;
+	OffsetNumber offno;
+	OffsetNumber maxoffno;
+	ListInfo	listInfo;
 	BufferAccessStrategy bas = GetAccessStrategy(BAS_BULKREAD);
 
 	if (stats == NULL)
 		stats = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
 
 	/* Iterate over list pages */
-	while (BlockNumberIsValid(blkno))
+	while (BlockNumberIsValid(nextblkno))
 	{
-		Buffer		cbuf;
-		Page		cpage;
-		OffsetNumber coffno;
-		OffsetNumber cmaxoffno;
-		BlockNumber startPages[MaxOffsetNumber];
-		ListInfo	listInfo;
-
-		cbuf = ReadBuffer(index, blkno);
+		cbuf = ReadBuffer(index, nextblkno);
 		LockBuffer(cbuf, BUFFER_LOCK_SHARE);
 		cpage = BufferGetPage(cbuf);
 
@@ -37,32 +48,23 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 		/* Iterate over lists */
 		for (coffno = FirstOffsetNumber; coffno <= cmaxoffno; coffno = OffsetNumberNext(coffno))
 		{
-			IvfflatList list = (IvfflatList) PageGetItem(cpage, PageGetItemId(cpage, coffno));
-
+			list = (IvfflatList) PageGetItem(cpage, PageGetItemId(cpage, coffno));
 			startPages[coffno - FirstOffsetNumber] = list->startPage;
 		}
 
-		listInfo.blkno = blkno;
-		blkno = IvfflatPageGetOpaque(cpage)->nextblkno;
+		listInfo.blkno = nextblkno;
+		nextblkno = IvfflatPageGetOpaque(cpage)->nextblkno;
 
 		UnlockReleaseBuffer(cbuf);
 
 		for (coffno = FirstOffsetNumber; coffno <= cmaxoffno; coffno = OffsetNumberNext(coffno))
 		{
-			BlockNumber searchPage = startPages[coffno - FirstOffsetNumber];
-			BlockNumber insertPage = InvalidBlockNumber;
+			searchPage = startPages[coffno - FirstOffsetNumber];
+			insertPage = InvalidBlockNumber;
 
 			/* Iterate over entry pages */
 			while (BlockNumberIsValid(searchPage))
 			{
-				Buffer		buf;
-				Page		page;
-				GenericXLogState *state;
-				OffsetNumber offno;
-				OffsetNumber maxoffno;
-				OffsetNumber deletable[MaxOffsetNumber];
-				int			ndeletable;
-
 				vacuum_delay_point();
 
 				buf = ReadBufferExtended(index, MAIN_FORKNUM, searchPage, RBM_NORMAL, bas);
@@ -84,8 +86,8 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 				/* Find deleted tuples */
 				for (offno = FirstOffsetNumber; offno <= maxoffno; offno = OffsetNumberNext(offno))
 				{
-					IndexTuple	itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offno));
-					ItemPointer htup = &(itup->t_tid);
+					itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offno));
+					htup = &(itup->t_tid);
 
 					if (callback(htup, callback_state))
 					{
@@ -125,13 +127,11 @@ ivfflatbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
 			if (BlockNumberIsValid(insertPage))
 			{
 				listInfo.offno = coffno;
-				IvfflatUpdateList(index, listInfo, insertPage, InvalidBlockNumber, InvalidBlockNumber, MAIN_FORKNUM);
+				IvfflatUpdateList(index, state, listInfo, insertPage, InvalidBlockNumber, InvalidBlockNumber, MAIN_FORKNUM);
 			}
 		}
 	}
 
-	FreeAccessStrategy(bas);
-
 	return stats;
 }
 

src/vector.c

@@ -2,18 +2,15 @@
 
 #include <math.h>
 
-#include "catalog/pg_type.h"
+#include "vector.h"
 #include "fmgr.h"
-#include "hnsw.h"
-#include "ivfflat.h"
+#include "catalog/pg_type.h"
 #include "lib/stringinfo.h"
 #include "libpq/pqformat.h"
-#include "port.h"				/* for strtof() */
 #include "utils/array.h"
 #include "utils/builtins.h"
 #include "utils/lsyscache.h"
 #include "utils/numeric.h"
-#include "vector.h"
 
 #if PG_VERSION_NUM >= 120000
 #include "common/shortest_dec.h"
@@ -32,16 +29,6 @@
 
 PG_MODULE_MAGIC;
 
-/*
- * Initialize index options and variables
- */
-void
-_PG_init(void)
-{
-	HnswInit();
-	IvfflatInit();
-}
-
 /*
  * Ensure same dimensions
  */
@@ -55,7 +42,7 @@ CheckDims(Vector * a, Vector * b)
 }
 
 /*
- * Ensure expected dimensions
+ * Ensure expected dimension
  */
 static inline void
 CheckExpectedDim(int32 typmod, int dim)
@@ -66,9 +53,7 @@ CheckExpectedDim(int32 typmod, int dim)
 				 errmsg("expected %d dimensions, not %d", typmod, dim)));
 }
 
-/*
- * Ensure valid dimensions
- */
+
 static inline void
 CheckDim(int dim)
 {
@@ -94,28 +79,13 @@ CheckElement(float value)
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("NaN not allowed in vector")));
 
+
 	if (isinf(value))
 		ereport(ERROR,
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("infinite value not allowed in vector")));
 }
 
-/*
- * Check for whitespace, since array_isspace() is static
- */
-static inline bool
-vector_isspace(char ch)
-{
-	if (ch == ' ' ||
-		ch == '\t' ||
-		ch == '\n' ||
-		ch == '\r' ||
-		ch == '\v' ||
-		ch == '\f')
-		return true;
-	return false;
-}
-
 /*
  * Check state array
  */
@@ -138,16 +108,32 @@ float_overflow_error(void)
 			(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
 			 errmsg("value out of range: overflow")));
 }
-
-static pg_noinline void
-float_underflow_error(void)
-{
-	ereport(ERROR,
-			(errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
-			 errmsg("value out of range: underflow")));
-}
 #endif
 
+/*
+ * Print vector - useful for debugging
+ */
+void
+PrintVector(char *msg, Vector * vector)
+{
+	StringInfoData buf;
+	int			dim = vector->dim;
+	int			i;
+
+	initStringInfo(&buf);
+
+	appendStringInfoChar(&buf, '[');
+	for (i = 0; i < dim; i++)
+	{
+		if (i > 0)
+			appendStringInfoString(&buf, ",");
+		appendStringInfoString(&buf, float8out_internal(vector->x[i]));
+	}
+	appendStringInfoChar(&buf, ']');
+
+	elog(INFO, "%s = %s", msg, buf.data);
+}
+
 /*
  * Convert textual representation to internal representation
  */
@@ -157,20 +143,17 @@ vector_in(PG_FUNCTION_ARGS)
 {
 	char	   *str = PG_GETARG_CSTRING(0);
 	int32		typmod = PG_GETARG_INT32(2);
+	int			i;
 	float		x[VECTOR_MAX_DIM];
 	int			dim = 0;
 	char	   *pt;
 	char	   *stringEnd;
 	Vector	   *result;
-	char	   *lit = pstrdup(str);
-
-	while (vector_isspace(*str))
-		str++;
 
 	if (*str != '[')
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-				 errmsg("malformed vector literal: \"%s\"", lit),
+				 errmsg("malformed vector literal: \"%s\"", str),
 				 errdetail("Vector contents must start with \"[\".")));
 
 	str++;
@@ -184,15 +167,6 @@ vector_in(PG_FUNCTION_ARGS)
 					(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
 					 errmsg("vector cannot have more than %d dimensions", VECTOR_MAX_DIM)));
 
-		while (vector_isspace(*pt))
-			pt++;
-
-		/* Check for empty string like float4in */
-		if (*pt == '\0')
-			ereport(ERROR,
-					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type vector: \"%s\"", lit)));
-
 		/* Use strtof like float4in to avoid a double-rounding problem */
 		x[dim] = strtof(pt, &stringEnd);
 		CheckElement(x[dim]);
@@ -201,57 +175,37 @@ vector_in(PG_FUNCTION_ARGS)
 		if (stringEnd == pt)
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type vector: \"%s\"", lit)));
-
-		while (vector_isspace(*stringEnd))
-			stringEnd++;
+					 errmsg("invalid input syntax for type vector: \"%s\"", pt)));
 
 		if (*stringEnd != '\0' && *stringEnd != ']')
 			ereport(ERROR,
 					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("invalid input syntax for type vector: \"%s\"", lit)));
+					 errmsg("invalid input syntax for type vector: \"%s\"", pt)));
 
 		pt = strtok(NULL, ",");
 	}
 
-	if (stringEnd == NULL || *stringEnd != ']')
+	if (*stringEnd != ']')
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-				 errmsg("malformed vector literal: \"%s\"", lit),
+				 errmsg("malformed vector literal"),
 				 errdetail("Unexpected end of input.")));
 
-	stringEnd++;
-
-	/* Only whitespace is allowed after the closing brace */
-	while (vector_isspace(*stringEnd))
-		stringEnd++;
-
-	if (*stringEnd != '\0')
+	if (stringEnd[1] != '\0')
 		ereport(ERROR,
 				(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-				 errmsg("malformed vector literal: \"%s\"", lit),
+				 errmsg("malformed vector literal"),
 				 errdetail("Junk after closing right brace.")));
 
-	/* Ensure no consecutive delimiters since strtok skips */
-	for (pt = lit + 1; *pt != '\0'; pt++)
-	{
-		if (pt[-1] == ',' && *pt == ',')
-			ereport(ERROR,
-					(errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
-					 errmsg("malformed vector literal: \"%s\"", lit)));
-	}
-
 	if (dim < 1)
 		ereport(ERROR,
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("vector must have at least 1 dimension")));
 
-	pfree(lit);
-
 	CheckExpectedDim(typmod, dim);
 
 	result = InitVector(dim);
-	for (int i = 0; i < dim; i++)
+	for (i = 0; i < dim; i++)
 		result->x[i] = x[i];
 
 	PG_RETURN_POINTER(result);
@@ -268,6 +222,7 @@ vector_out(PG_FUNCTION_ARGS)
 	int			dim = vector->dim;
 	char	   *buf;
 	char	   *ptr;
+	int			i;
 	int			n;
 
 #if PG_VERSION_NUM < 120000
@@ -294,7 +249,7 @@ vector_out(PG_FUNCTION_ARGS)
 
 	*ptr = '[';
 	ptr++;
-	for (int i = 0; i < dim; i++)
+	for (i = 0; i < dim; i++)
 	{
 		if (i > 0)
 		{
@@ -317,18 +272,6 @@ vector_out(PG_FUNCTION_ARGS)
 	PG_RETURN_CSTRING(buf);
 }
 
-/*
- * Print vector - useful for debugging
- */
-void
-PrintVector(char *msg, Vector * vector)
-{
-	char	   *out = DatumGetPointer(DirectFunctionCall1(vector_out, PointerGetDatum(vector)));
-
-	elog(INFO, "%s = %s", msg, out);
-	pfree(out);
-}
-
 /*
  * Convert type modifier
  */
@@ -372,6 +315,7 @@ vector_recv(PG_FUNCTION_ARGS)
 	Vector	   *result;
 	int16		dim;
 	int16		unused;
+	int			i;
 
 	dim = pq_getmsgint(buf, sizeof(int16));
 	unused = pq_getmsgint(buf, sizeof(int16));
@@ -385,11 +329,8 @@ vector_recv(PG_FUNCTION_ARGS)
 				 errmsg("expected unused to be 0, not %d", unused)));
 
 	result = InitVector(dim);
-	for (int i = 0; i < dim; i++)
-	{
+	for (i = 0; i < dim; i++)
 		result->x[i] = pq_getmsgfloat4(buf);
-		CheckElement(result->x[i]);
-	}
 
 	PG_RETURN_POINTER(result);
 }
@@ -403,11 +344,12 @@ vector_send(PG_FUNCTION_ARGS)
 {
 	Vector	   *vec = PG_GETARG_VECTOR_P(0);
 	StringInfoData buf;
+	int			i;
 
 	pq_begintypsend(&buf);
 	pq_sendint(&buf, vec->dim, sizeof(int16));
 	pq_sendint(&buf, vec->unused, sizeof(int16));
-	for (int i = 0; i < vec->dim; i++)
+	for (i = 0; i < vec->dim; i++)
 		pq_sendfloat4(&buf, vec->x[i]);
 
 	PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
@@ -437,6 +379,7 @@ array_to_vector(PG_FUNCTION_ARGS)
 {
 	ArrayType  *array = PG_GETARG_ARRAYTYPE_P(0);
 	int32		typmod = PG_GETARG_INT32(1);
+	int			i;
 	Vector	   *result;
 	int16		typlen;
 	bool		typbyval;
@@ -450,49 +393,37 @@ array_to_vector(PG_FUNCTION_ARGS)
 				(errcode(ERRCODE_DATA_EXCEPTION),
 				 errmsg("array must be 1-D")));
 
-	if (ARR_HASNULL(array) && array_contains_nulls(array))
-		ereport(ERROR,
-				(errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
-				 errmsg("array must not contain nulls")));
-
 	get_typlenbyvalalign(ARR_ELEMTYPE(array), &typlen, &typbyval, &typalign);
 	deconstruct_array(array, ARR_ELEMTYPE(array), typlen, typbyval, typalign, &elemsp, &nullsp, &nelemsp);
 
-	CheckDim(nelemsp);
-	CheckExpectedDim(typmod, nelemsp);
+	if (typmod == -1)
+		CheckDim(nelemsp);
+	else
+		CheckExpectedDim(typmod, nelemsp);
 
 	result = InitVector(nelemsp);
-
-	if (ARR_ELEMTYPE(array) == INT4OID)
+	for (i = 0; i < nelemsp; i++)
 	{
-		for (int i = 0; i < nelemsp; i++)
+		if (nullsp[i])
+			ereport(ERROR,
+					(errcode(ERRCODE_NULL_VALUE_NOT_ALLOWED),
+					 errmsg("array must not containing NULLs")));
+
+		if (ARR_ELEMTYPE(array) == INT4OID)
 			result->x[i] = DatumGetInt32(elemsp[i]);
-	}
-	else if (ARR_ELEMTYPE(array) == FLOAT8OID)
-	{
-		for (int i = 0; i < nelemsp; i++)
+		else if (ARR_ELEMTYPE(array) == FLOAT8OID)
 			result->x[i] = DatumGetFloat8(elemsp[i]);
-	}
-	else if (ARR_ELEMTYPE(array) == FLOAT4OID)
-	{
-		for (int i = 0; i < nelemsp; i++)
+		else if (ARR_ELEMTYPE(array) == FLOAT4OID)
 			result->x[i] = DatumGetFloat4(elemsp[i]);
-	}
-	else if (ARR_ELEMTYPE(array) == NUMERICOID)
-	{
-		for (int i = 0; i < nelemsp; i++)
+		else if (ARR_ELEMTYPE(array) == NUMERICOID)
 			result->x[i] = DatumGetFloat4(DirectFunctionCall1(numeric_float4, elemsp[i]));
-	}
-	else
-	{
-		ereport(ERROR,
-				(errcode(ERRCODE_DATA_EXCEPTION),
-				 errmsg("unsupported array type")));
-	}
+		else
+			ereport(ERROR,
+					(errcode(ERRCODE_DATA_EXCEPTION),
+					 errmsg("unsupported array type")));
 
-	/* Check elements */
-	for (int i = 0; i < result->dim; i++)
 		CheckElement(result->x[i]);
+	}
 
 	PG_RETURN_POINTER(result);
 }
@@ -505,18 +436,17 @@ Datum
 vector_to_float4(PG_FUNCTION_ARGS)
 {
 	Vector	   *vec = PG_GETARG_VECTOR_P(0);
-	Datum	   *datums;
+	Datum	   *d;
 	ArrayType  *result;
+	int			i;
 
-	datums = (Datum *) palloc(sizeof(Datum) * vec->dim);
+	d = (Datum *) palloc(sizeof(Datum) * vec->dim);
 
-	for (int i = 0; i < vec->dim; i++)
-		datums[i] = Float4GetDatum(vec->x[i]);
+	for (i = 0; i < vec->dim; i++)
+		d[i] = Float4GetDatum(vec->x[i]);
 
 	/* Use TYPALIGN_INT for float4 */
-	result = construct_array(datums, vec->dim, FLOAT4OID, sizeof(float4), true, TYPALIGN_INT);
-
-	pfree(datums);
+	result = construct_array(d, vec->dim, FLOAT4OID, sizeof(float4), true, TYPALIGN_INT);
 
 	PG_RETURN_POINTER(result);
 }
@@ -532,8 +462,8 @@ l2_distance(PG_FUNCTION_ARGS)
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 	float	   *ax = a->x;
 	float	   *bx = b->x;
-	float		distance = 0.0;
-	float		diff;
+	double		distance = 0.0;
+	double		diff;
 
 	CheckDims(a, b);
 
@@ -544,7 +474,7 @@ l2_distance(PG_FUNCTION_ARGS)
 		distance += diff * diff;
 	}
 
-	PG_RETURN_FLOAT8(sqrt((double) distance));
+	PG_RETURN_FLOAT8(sqrt(distance));
 }
 
 /*
@@ -559,8 +489,8 @@ vector_l2_squared_distance(PG_FUNCTION_ARGS)
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 	float	   *ax = a->x;
 	float	   *bx = b->x;
-	float		distance = 0.0;
-	float		diff;
+	double		distance = 0.0;
+	double		diff;
 
 	CheckDims(a, b);
 
@@ -571,7 +501,7 @@ vector_l2_squared_distance(PG_FUNCTION_ARGS)
 		distance += diff * diff;
 	}
 
-	PG_RETURN_FLOAT8((double) distance);
+	PG_RETURN_FLOAT8(distance);
 }
 
 /*
@@ -585,7 +515,7 @@ inner_product(PG_FUNCTION_ARGS)
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 	float	   *ax = a->x;
 	float	   *bx = b->x;
-	float		distance = 0.0;
+	double		distance = 0.0;
 
 	CheckDims(a, b);
 
@@ -593,7 +523,7 @@ inner_product(PG_FUNCTION_ARGS)
 	for (int i = 0; i < a->dim; i++)
 		distance += ax[i] * bx[i];
 
-	PG_RETURN_FLOAT8((double) distance);
+	PG_RETURN_FLOAT8(distance);
 }
 
 /*
@@ -607,7 +537,7 @@ vector_negative_inner_product(PG_FUNCTION_ARGS)
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 	float	   *ax = a->x;
 	float	   *bx = b->x;
-	float		distance = 0.0;
+	double		distance = 0.0;
 
 	CheckDims(a, b);
 
@@ -615,7 +545,7 @@ vector_negative_inner_product(PG_FUNCTION_ARGS)
 	for (int i = 0; i < a->dim; i++)
 		distance += ax[i] * bx[i];
 
-	PG_RETURN_FLOAT8((double) distance * -1);
+	PG_RETURN_FLOAT8(distance * -1);
 }
 
 /*
@@ -629,10 +559,9 @@ cosine_distance(PG_FUNCTION_ARGS)
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 	float	   *ax = a->x;
 	float	   *bx = b->x;
-	float		distance = 0.0;
-	float		norma = 0.0;
-	float		normb = 0.0;
-	double		similarity;
+	double		distance = 0.0;
+	double		norma = 0.0;
+	double		normb = 0.0;
 
 	CheckDims(a, b);
 
@@ -645,21 +574,7 @@ cosine_distance(PG_FUNCTION_ARGS)
 	}
 
 	/* Use sqrt(a * b) over sqrt(a) * sqrt(b) */
-	similarity = (double) distance / sqrt((double) norma * (double) normb);
-
-#ifdef _MSC_VER
-	/* /fp:fast may not propagate NaN */
-	if (isnan(similarity))
-		PG_RETURN_FLOAT8(NAN);
-#endif
-
-	/* Keep in range */
-	if (similarity > 1)
-		similarity = 1.0;
-	else if (similarity < -1)
-		similarity = -1.0;
-
-	PG_RETURN_FLOAT8(1.0 - similarity);
+	PG_RETURN_FLOAT8(1 - (distance / sqrt(norma * normb)));
 }
 
 /*
@@ -673,16 +588,13 @@ vector_spherical_distance(PG_FUNCTION_ARGS)
 {
 	Vector	   *a = PG_GETARG_VECTOR_P(0);
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
-	float		dp = 0.0;
-	double		distance;
+	double		distance = 0.0;
 
 	CheckDims(a, b);
 
 	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
-		dp += a->x[i] * b->x[i];
-
-	distance = (double) dp;
+		distance += a->x[i] * b->x[i];
 
 	/* Prevent NaN with acos with loss of precision */
 	if (distance > 1)
@@ -693,28 +605,6 @@ vector_spherical_distance(PG_FUNCTION_ARGS)
 	PG_RETURN_FLOAT8(acos(distance) / M_PI);
 }
 
-/*
- * Get the L1 distance between vectors
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(l1_distance);
-Datum
-l1_distance(PG_FUNCTION_ARGS)
-{
-	Vector	   *a = PG_GETARG_VECTOR_P(0);
-	Vector	   *b = PG_GETARG_VECTOR_P(1);
-	float	   *ax = a->x;
-	float	   *bx = b->x;
-	float		distance = 0.0;
-
-	CheckDims(a, b);
-
-	/* Auto-vectorized */
-	for (int i = 0; i < a->dim; i++)
-		distance += fabsf(ax[i] - bx[i]);
-
-	PG_RETURN_FLOAT8((double) distance);
-}
-
 /*
  * Get the dimensions of a vector
  */
@@ -736,52 +626,13 @@ vector_norm(PG_FUNCTION_ARGS)
 {
 	Vector	   *a = PG_GETARG_VECTOR_P(0);
 	float	   *ax = a->x;
-	float		norm = 0.0;
+	double		norm = 0.0;
 
 	/* Auto-vectorized */
 	for (int i = 0; i < a->dim; i++)
 		norm += ax[i] * ax[i];
 
-	PG_RETURN_FLOAT8(sqrt((double) norm));
-}
-
-/*
- * Normalize a vector with the L2 norm
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(normalize_l2);
-Datum
-normalize_l2(PG_FUNCTION_ARGS)
-{
-	Vector	   *a = PG_GETARG_VECTOR_P(0);
-	float	   *ax = a->x;
-	double		norm = 0.0;
-	Vector	   *result;
-	float	   *rx;
-
-	result = InitVector(a->dim);
-	rx = result->x;
-
-	/* Auto-vectorized */
-	for (int i = 0; i < a->dim; i++)
-		norm += (double) ax[i] * (double) ax[i];
-
-	norm = sqrt(norm);
-
-	if (norm > 0)
-	{
-		/* Auto-vectorized */
-		for (int i = 0, imax = a->dim; i < imax; i++)
-			rx[i] = ax[i] / norm;
-
-		/* Check for overflow */
-		for (int i = 0, imax = a->dim; i < imax; i++)
-		{
-			if (isinf(rx[i]))
-				float_overflow_error();
-		}
-	}
-
-	PG_RETURN_POINTER(result);
+	PG_RETURN_FLOAT8(sqrt(norm));
 }
 
 /*
@@ -807,13 +658,6 @@ vector_add(PG_FUNCTION_ARGS)
 	for (int i = 0, imax = a->dim; i < imax; i++)
 		rx[i] = ax[i] + bx[i];
 
-	/* Check for overflow */
-	for (int i = 0, imax = a->dim; i < imax; i++)
-	{
-		if (isinf(rx[i]))
-			float_overflow_error();
-	}
-
 	PG_RETURN_POINTER(result);
 }
 
@@ -840,49 +684,6 @@ vector_sub(PG_FUNCTION_ARGS)
 	for (int i = 0, imax = a->dim; i < imax; i++)
 		rx[i] = ax[i] - bx[i];
 
-	/* Check for overflow */
-	for (int i = 0, imax = a->dim; i < imax; i++)
-	{
-		if (isinf(rx[i]))
-			float_overflow_error();
-	}
-
-	PG_RETURN_POINTER(result);
-}
-
-/*
- * Multiply vectors
- */
-PGDLLEXPORT PG_FUNCTION_INFO_V1(vector_mul);
-Datum
-vector_mul(PG_FUNCTION_ARGS)
-{
-	Vector	   *a = PG_GETARG_VECTOR_P(0);
-	Vector	   *b = PG_GETARG_VECTOR_P(1);
-	float	   *ax = a->x;
-	float	   *bx = b->x;
-	Vector	   *result;
-	float	   *rx;
-
-	CheckDims(a, b);
-
-	result = InitVector(a->dim);
-	rx = result->x;
-
-	/* Auto-vectorized */
-	for (int i = 0, imax = a->dim; i < imax; i++)
-		rx[i] = ax[i] * bx[i];
-
-	/* Check for overflow and underflow */
-	for (int i = 0, imax = a->dim; i < imax; i++)
-	{
-		if (isinf(rx[i]))
-			float_overflow_error();
-
-		if (rx[i] == 0 && !(ax[i] == 0 || bx[i] == 0))
-			float_underflow_error();
-	}
-
 	PG_RETURN_POINTER(result);
 }
 
@@ -892,9 +693,11 @@ vector_mul(PG_FUNCTION_ARGS)
 int
 vector_cmp_internal(Vector * a, Vector * b)
 {
+	int			i;
+
 	CheckDims(a, b);
 
-	for (int i = 0; i < a->dim; i++)
+	for (i = 0; i < a->dim; i++)
 	{
 		if (a->x[i] < b->x[i])
 			return -1;
@@ -1026,12 +829,12 @@ vector_accum(PG_FUNCTION_ARGS)
 	n = statevalues[0] + 1.0;
 
 	statedatums = CreateStateDatums(dim);
-	statedatums[0] = Float8GetDatum(n);
+	statedatums[0] = Float8GetDatumFast(n);
 
 	if (newarr)
 	{
 		for (int i = 0; i < dim; i++)
-			statedatums[i + 1] = Float8GetDatum((double) x[i]);
+			statedatums[i + 1] = Float8GetDatumFast((double) x[i]);
 	}
 	else
 	{
@@ -1039,11 +842,10 @@ vector_accum(PG_FUNCTION_ARGS)
 		{
 			double		v = statevalues[i + 1] + x[i];
 
-			/* Check for overflow */
 			if (isinf(v))
 				float_overflow_error();
 
-			statedatums[i + 1] = Float8GetDatum(v);
+			statedatums[i + 1] = Float8GetDatumFast(v);
 		}
 	}
 
@@ -1088,7 +890,7 @@ vector_combine(PG_FUNCTION_ARGS)
 		dim = STATE_DIMS(statearray2);
 		statedatums = CreateStateDatums(dim);
 		for (int i = 1; i <= dim; i++)
-			statedatums[i] = Float8GetDatum(statevalues2[i]);
+			statedatums[i] = Float8GetDatumFast(statevalues2[i]);
 	}
 	else if (n2 == 0.0)
 	{
@@ -1096,7 +898,7 @@ vector_combine(PG_FUNCTION_ARGS)
 		dim = STATE_DIMS(statearray1);
 		statedatums = CreateStateDatums(dim);
 		for (int i = 1; i <= dim; i++)
-			statedatums[i] = Float8GetDatum(statevalues1[i]);
+			statedatums[i] = Float8GetDatumFast(statevalues1[i]);
 	}
 	else
 	{
@@ -1108,15 +910,14 @@ vector_combine(PG_FUNCTION_ARGS)
 		{
 			double		v = statevalues1[i] + statevalues2[i];
 
-			/* Check for overflow */
 			if (isinf(v))
 				float_overflow_error();
 
-			statedatums[i] = Float8GetDatum(v);
+			statedatums[i] = Float8GetDatumFast(v);
 		}
 	}
 
-	statedatums[0] = Float8GetDatum(n);
+	statedatums[0] = Float8GetDatumFast(n);
 
 	result = construct_array(statedatums, dim + 1,
 							 FLOAT8OID,
@@ -1139,6 +940,7 @@ vector_avg(PG_FUNCTION_ARGS)
 	float8		n;
 	uint16		dim;
 	Vector	   *result;
+	float		v;
 
 	/* Check array before using */
 	statevalues = CheckStateArray(statearray, "vector_avg");
@@ -1150,12 +952,12 @@ vector_avg(PG_FUNCTION_ARGS)
 
 	/* Create vector */
 	dim = STATE_DIMS(statearray);
-	CheckDim(dim);
 	result = InitVector(dim);
 	for (int i = 0; i < dim; i++)
 	{
-		result->x[i] = statevalues[i + 1] / n;
-		CheckElement(result->x[i]);
+		v = statevalues[i + 1] / n;
+		CheckElement(v);
+		result->x[i] = v;
 	}
 
 	PG_RETURN_POINTER(result);

src/vector.h

@@ -14,9 +14,6 @@
 #define PG_GETARG_VECTOR_P(x)	DatumGetVector(PG_GETARG_DATUM(x))
 #define PG_RETURN_VECTOR_P(x)	PG_RETURN_POINTER(x)
 
-/* Exported functions */
-PGDLLEXPORT void _PG_init(void);
-
 typedef struct Vector
 {
 	int32		vl_len_;		/* varlena header (do not touch directly!) */

test/expected/cast.out

@@ -29,7 +29,7 @@ SELECT ARRAY[1,2,3]::numeric[]::vector;
 (1 row)
 
 SELECT '{NULL}'::real[]::vector;
-ERROR:  array must not contain nulls
+ERROR:  array must not containing NULLs
 SELECT '{NaN}'::real[]::vector;
 ERROR:  NaN not allowed in vector
 SELECT '{Infinity}'::real[]::vector;
@@ -38,8 +38,6 @@ SELECT '{-Infinity}'::real[]::vector;
 ERROR:  infinite value not allowed in vector
 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]'::vector::real[];
  float4  
 ---------
@@ -48,8 +46,6 @@ SELECT '[1,2,3]'::vector::real[];
 
 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;
-ERROR:  vector cannot have more than 16000 dimensions
 -- ensure no error
 SELECT ARRAY[1,2,3] = ARRAY[1,2,3];
  ?column? 

test/expected/functions.out

@@ -4,26 +4,12 @@ SELECT '[1,2,3]'::vector + '[4,5,6]';
  [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 
 -------------
@@ -48,36 +34,6 @@ SELECT vector_norm('[0,1]');
            1
 (1 row)
 
-SELECT normalize_l2('[3,4]');
- normalize_l2 
---------------
- [0.6,0.8]
-(1 row)
-
-SELECT normalize_l2('[3,0]');
- normalize_l2 
---------------
- [1,0]
-(1 row)
-
-SELECT normalize_l2('[0,0.1]');
- normalize_l2 
---------------
- [0,1]
-(1 row)
-
-SELECT normalize_l2('[0,0]');
- normalize_l2 
---------------
- [0,0]
-(1 row)
-
-SELECT normalize_l2('[3e38]');
- normalize_l2 
---------------
- [1]
-(1 row)
-
 SELECT l2_distance('[0,0]', '[3,4]');
  l2_distance 
 -------------
@@ -92,12 +48,6 @@ SELECT l2_distance('[0,0]', '[0,1]');
 
 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 
 ---------------
@@ -106,12 +56,6 @@ SELECT inner_product('[1,2]', '[3,4]');
 
 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 
 -----------------
@@ -138,44 +82,6 @@ SELECT cosine_distance('[1,1]', '[-1,-1]');
 
 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    
 -----------
@@ -196,33 +102,3 @@ SELECT avg(v) FROM unnest(ARRAY[]::vector[]) v;
 
 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

test/expected/hnsw_cosine.out

@@ -1,26 +0,0 @@
-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;

test/expected/hnsw_ip.out

@@ -1,21 +0,0 @@
-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;

test/expected/hnsw_l2.out

@@ -1,30 +0,0 @@
-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)
-
-DROP TABLE t;

test/expected/hnsw_options.out

@@ -1,25 +0,0 @@
-SET enable_seqscan = off;
-CREATE TABLE t (val vector(3));
-CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 3);
-ERROR:  value 3 out of bounds for option "m"
-DETAIL:  Valid values are between "4" 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 "4" and "100".
-CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (ef_construction = 9);
-ERROR:  value 9 out of bounds for option "ef_construction"
-DETAIL:  Valid values are between "10" 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 "10" and "1000".
-SHOW hnsw.ef_search;
- hnsw.ef_search 
-----------------
- 40
-(1 row)
-
-SET hnsw.ef_search = 9;
-ERROR:  9 is outside the valid range for parameter "hnsw.ef_search" (10 .. 1000)
-SET hnsw.ef_search = 1001;
-ERROR:  1001 is outside the valid range for parameter "hnsw.ef_search" (10 .. 1000)
-DROP TABLE t;

test/expected/hnsw_unlogged.out

@@ -1,13 +0,0 @@
-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;

test/expected/input.out

@@ -4,22 +4,10 @@ SELECT '[1,2,3]'::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]
+SELECT '[-1,2,3]'::vector;
+  vector  
+----------
+ [-1,2,3]
 (1 row)
 
 SELECT '[1.23456]'::vector;
@@ -29,7 +17,7 @@ SELECT '[1.23456]'::vector;
 (1 row)
 
 SELECT '[hello,1]'::vector;
-ERROR:  invalid input syntax for type vector: "[hello,1]"
+ERROR:  invalid input syntax for type vector: "hello"
 LINE 1: SELECT '[hello,1]'::vector;
                ^
 SELECT '[NaN,1]'::vector;
@@ -44,35 +32,13 @@ 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"
+ERROR:  malformed vector literal
 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"
+ERROR:  malformed vector literal
 LINE 1: SELECT '[1,2,3]9'::vector;
                ^
 DETAIL:  Junk after closing right brace.
@@ -81,41 +47,14 @@ 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,]"
+ERROR:  invalid input syntax for type vector: "]"
 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[]);

test/expected/ivfflat_cosine.out

@@ -11,16 +11,9 @@ SELECT * FROM t ORDER BY val <=> '[3,3,3]';
  [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)
+SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector);
+ val 
+-----
+(0 rows)
 
 DROP TABLE t;

test/expected/ivfflat_ip.out

@@ -12,10 +12,9 @@ SELECT * FROM t ORDER BY val <#> '[3,3,3]';
  [0,0,0]
 (4 rows)
 
-SELECT COUNT(*) FROM (SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector)) t2;
- count 
--------
-     4
-(1 row)
+SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector);
+ val 
+-----
+(0 rows)
 
 DROP TABLE t;

test/expected/ivfflat_l2.out

@@ -13,13 +13,9 @@ SELECT * FROM t ORDER BY val <-> '[3,3,3]';
 (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)
+ val 
+-----
+(0 rows)
 
 SELECT COUNT(*) FROM t;
  count 

test/sql/cast.sql

@@ -8,10 +8,8 @@ SELECT '{NaN}'::real[]::vector;
 SELECT '{Infinity}'::real[]::vector;
 SELECT '{-Infinity}'::real[]::vector;
 SELECT '{}'::real[]::vector;
-SELECT '{{1}}'::real[]::vector;
 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;
 
 -- ensure no error
 SELECT ARRAY[1,2,3] = ARRAY[1,2,3];

test/sql/functions.sql

@@ -1,10 +1,5 @@
 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]');
 
@@ -12,44 +7,20 @@ SELECT round(vector_norm('[1,1]')::numeric, 5);
 SELECT vector_norm('[3,4]');
 SELECT vector_norm('[0,1]');
 
-SELECT normalize_l2('[3,4]');
-SELECT normalize_l2('[3,0]');
-SELECT normalize_l2('[0,0.1]');
-SELECT normalize_l2('[0,0]');
-SELECT normalize_l2('[3e38]');
-
 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('[1,1]', '[-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;

test/sql/hnsw_cosine.sql

@@ -1,13 +0,0 @@
-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;

test/sql/hnsw_ip.sql

@@ -1,12 +0,0 @@
-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;

test/sql/hnsw_l2.sql

@@ -1,13 +0,0 @@
-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;
-
-DROP TABLE t;

test/sql/hnsw_options.sql

@@ -1,14 +0,0 @@
-SET enable_seqscan = off;
-
-CREATE TABLE t (val vector(3));
-CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (m = 3);
-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 = 9);
-CREATE INDEX ON t USING hnsw (val vector_l2_ops) WITH (ef_construction = 1001);
-
-SHOW hnsw.ef_search;
-
-SET hnsw.ef_search = 9;
-SET hnsw.ef_search = 1001;
-
-DROP TABLE t;

test/sql/hnsw_unlogged.sql

@@ -1,9 +0,0 @@
-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;

test/sql/input.sql

@@ -1,27 +1,15 @@
 SELECT '[1,2,3]'::vector;
-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[]);

test/sql/ivfflat_cosine.sql

@@ -7,7 +7,6 @@ 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;
+SELECT * FROM t ORDER BY val <=> (SELECT NULL::vector);
 
 DROP TABLE t;

test/sql/ivfflat_ip.sql

@@ -7,6 +7,6 @@ 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;
+SELECT * FROM t ORDER BY val <#> (SELECT NULL::vector);
 
 DROP TABLE t;

test/t/001_wal.pl

@@ -5,7 +5,7 @@ use strict;
 use warnings;
 use PostgresNode;
 use TestLib;
-use Test::More;
+use Test::More tests => 31;
 
 my $dim = 32;
 
@@ -95,5 +95,3 @@ for my $i (1 .. 10)
 	);
 	test_index_replay("insert $i");
 }
-
-done_testing();

test/t/002_vacuum.pl

@@ -2,7 +2,7 @@ use strict;
 use warnings;
 use PostgresNode;
 use TestLib;
-use Test::More;
+use Test::More tests => 1;
 
 my $dim = 3;
 
@@ -39,5 +39,3 @@ $node->safe_psql("postgres",
 # Check size
 my $new_size = $node->safe_psql("postgres", "SELECT pg_total_relation_size('tst_v_idx');");
 is($size, $new_size, "size does not change");
-
-done_testing();

test/t/003_ivfflat_recall.pl

@@ -1,127 +0,0 @@
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-my $node;
-my @queries = ();
-my @expected;
-my $limit = 20;
-
-sub test_recall
-{
-	my ($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 %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 = get_new_node('node');
-$node->init;
-$node->start;
-
-# Create table
-$node->safe_psql("postgres", "CREATE EXTENSION vector;");
-$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector(3));");
-$node->safe_psql("postgres",
-	"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()] FROM generate_series(1, 100000) i;"
-);
-
-# Generate queries
-for (1..20) {
-	my $r1 = rand();
-	my $r2 = rand();
-	my $r3 = rand();
-	push(@queries, "[$r1,$r2,$r3]");
-}
-
-# Check each index type
-my @operators = ("<->", "<#>", "<=>");
-
-foreach (@operators) {
-	my $operator = $_;
-
-	# 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);
-	}
-
-	my $opclass;
-	if ($operator eq "<->") {
-		$opclass = "vector_l2_ops";
-	} elsif ($operator eq "<#>") {
-		$opclass = "vector_ip_ops";
-	} else {
-		$opclass = "vector_cosine_ops";
-	}
-
-	# 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
-		test_recall(1, 0.75, $operator);
-		test_recall(10, 0.95, $operator);
-	}
-	# Account for equal distances
-	test_recall(100, 0.995, $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
-		test_recall(1, 0.75, $operator);
-		test_recall(10, 0.95, $operator);
-	}
-	# Account for equal distances
-	test_recall(100, 0.995, $operator);
-
-	$node->safe_psql("postgres", "DROP INDEX idx;");
-}
-
-done_testing();

test/t/003_recall.pl

@@ -2,7 +2,7 @@ use strict;
 use warnings;
 use PostgresNode;
 use TestLib;
-use Test::More;
+use Test::More tests => 9;
 
 my $node;
 my @queries = ();
@@ -11,19 +11,14 @@ my $limit = 20;
 
 sub test_recall
 {
-	my ($min, $operator) = @_;
+	my ($probes, $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;
+			SET ivfflat.probes = $probes;
 			SELECT i FROM tst ORDER BY v $operator '$queries[$i]' LIMIT $limit;
 		));
 		my @actual_ids = split("\n", $actual);
@@ -51,7 +46,7 @@ $node->start;
 $node->safe_psql("postgres", "CREATE EXTENSION vector;");
 $node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector(3));");
 $node->safe_psql("postgres",
-	"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()] FROM generate_series(1, 10000) i;"
+	"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()] FROM generate_series(1, 100000) i;"
 );
 
 # Generate queries
@@ -77,20 +72,17 @@ foreach (@operators) {
 
 	# Add index
 	my $opclass;
-	if ($operator eq "<->") {
+	if ($operator == "<->") {
 		$opclass = "vector_l2_ops";
-	} elsif ($operator eq "<#>") {
+	} elsif ($operator == "<#>") {
 		$opclass = "vector_ip_ops";
 	} else {
 		$opclass = "vector_cosine_ops";
 	}
-	$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v $opclass);");
+	$node->safe_psql("postgres", "CREATE INDEX ON tst USING ivfflat (v $opclass);");
 
-	if ($operator eq "<#>") {
-		test_recall(0.80, $operator);
-	} else {
-		test_recall(0.99, $operator);
-	}
+	# Test approximate results
+	test_recall(1, 0.75, $operator);
+	test_recall(10, 0.95, $operator);
+	test_recall(100, 1.0, $operator);
 }
-
-done_testing();

test/t/004_centers.pl

@@ -2,7 +2,7 @@ use strict;
 use warnings;
 use PostgresNode;
 use TestLib;
-use Test::More;
+use Test::More tests => 3;
 
 # Initialize node
 my $node = get_new_node('node');
@@ -34,5 +34,3 @@ $node->safe_psql("postgres",
 
 # Test no error for duplicate centers
 test_centers(10);
-
-done_testing();

test/t/005_query_recall.pl

@@ -2,7 +2,7 @@ use strict;
 use warnings;
 use PostgresNode;
 use TestLib;
-use Test::More;
+use Test::More tests => 60;
 
 # Initialize node
 my $node = get_new_node('node');
@@ -23,9 +23,9 @@ foreach (@operators) {
 
 	# Add index
 	my $opclass;
-	if ($operator eq "<->") {
+	if ($operator == "<->") {
 		$opclass = "vector_l2_ops";
-	} elsif ($operator eq "<#>") {
+	} elsif ($operator == "<#>") {
 		$opclass = "vector_ip_ops";
 	} else {
 		$opclass = "vector_cosine_ops";
@@ -43,5 +43,3 @@ foreach (@operators) {
 		is($res, $query);
 	}
 }
-
-done_testing();

test/t/006_lists.pl

@@ -2,7 +2,7 @@ use strict;
 use warnings;
 use PostgresNode;
 use TestLib;
-use Test::More;
+use Test::More tests => 3;
 
 # Initialize node
 my $node = get_new_node('node');
@@ -29,5 +29,3 @@ my ($ret, $stdout, $stderr) = $node->psql("postgres",
 	"CREATE INDEX lists10000 ON tst USING ivfflat (v) WITH (lists = 10000);"
 );
 like($stderr, qr/memory required is/);
-
-done_testing();

test/t/007_inserts.pl

@@ -2,7 +2,7 @@ use strict;
 use warnings;
 use PostgresNode;
 use TestLib;
-use Test::More;
+use Test::More tests => 7;
 
 my $dim = 768;
 
@@ -53,5 +53,3 @@ $count = $node->safe_psql("postgres", qq(
 ));
 is($count, $expected);
 is(idx_scan(), 1);
-
-done_testing();

test/t/008_aggregates.pl

@@ -1,48 +0,0 @@
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-# Initialize node
-my $node = get_new_node('node');
-$node->init;
-$node->start;
-
-# Create table
-$node->safe_psql("postgres", "CREATE EXTENSION vector;");
-$node->safe_psql("postgres", "CREATE TABLE tst (r1 real, r2 real, r3 real, v vector(3));");
-$node->safe_psql("postgres", qq(
-	INSERT INTO tst SELECT r1, r2, r3, ARRAY[r1, r2, r3] FROM (
-		SELECT random() + 1.01 AS r1, random() + 2.01 AS r2, random() + 3.01 AS r3 FROM generate_series(1, 1000000) t
-	) i;
-));
-
-sub test_aggregate
-{
-	my ($agg) = @_;
-
-	# Test value
-	my $res = $node->safe_psql("postgres", "SELECT $agg(v) FROM tst;");
-	like($res, qr/\[1\.5/);
-	like($res, qr/,2\.5/);
-	like($res, qr/,3\.5/);
-
-	# Test matches real for avg
-	# Cannot test sum since sum(real) varies between calls
-	if ($agg eq 'avg') {
-		my $r1 = $node->safe_psql("postgres", "SELECT $agg(r1)::float4 FROM tst;");
-		my $r2 = $node->safe_psql("postgres", "SELECT $agg(r2)::float4 FROM tst;");
-		my $r3 = $node->safe_psql("postgres", "SELECT $agg(r3)::float4 FROM tst;");
-		is($res, "[$r1,$r2,$r3]");
-	}
-
-	# Test explain
-	my $explain = $node->safe_psql("postgres", "EXPLAIN SELECT $agg(v) FROM tst;");
-	like($explain, qr/Partial Aggregate/);
-}
-
-test_aggregate('avg');
-test_aggregate('sum');
-
-done_testing();

test/t/008_avg.pl

@@ -0,0 +1,35 @@
+use strict;
+use warnings;
+use PostgresNode;
+use TestLib;
+use Test::More tests => 5;
+
+# Initialize node
+my $node = get_new_node('node');
+$node->init;
+$node->start;
+
+# Create table
+$node->safe_psql("postgres", "CREATE EXTENSION vector;");
+$node->safe_psql("postgres", "CREATE TABLE tst (r1 real, r2 real, r3 real, v vector(3));");
+$node->safe_psql("postgres", qq(
+	INSERT INTO tst SELECT r1, r2, r3, ARRAY[r1, r2, r3] FROM (
+		SELECT random() + 1.01 AS r1, random() + 2.01 AS r2, random() + 3.01 AS r3 FROM generate_series(1, 1000000) t
+	) i;
+));
+
+# Test avg
+my $avg = $node->safe_psql("postgres", "SELECT AVG(v) FROM tst;");
+like($avg, qr/\[1\.5/);
+like($avg, qr/,2\.5/);
+like($avg, qr/,3\.5/);
+
+# Test matches real
+my $r1 = $node->safe_psql("postgres", "SELECT AVG(r1)::float4 FROM tst;");
+my $r2 = $node->safe_psql("postgres", "SELECT AVG(r2)::float4 FROM tst;");
+my $r3 = $node->safe_psql("postgres", "SELECT AVG(r3)::float4 FROM tst;");
+is($avg, "[$r1,$r2,$r3]");
+
+# Test explain
+my $explain = $node->safe_psql("postgres", "EXPLAIN SELECT AVG(v) FROM tst;");
+like($explain, qr/Partial Aggregate/);

test/t/009_storage.pl

@@ -2,7 +2,7 @@ use strict;
 use warnings;
 use PostgresNode;
 use TestLib;
-use Test::More;
+use Test::More tests => 1;
 
 my $dim = 1024;
 
@@ -30,5 +30,3 @@ my ($ret, $stdout, $stderr) = $node->psql("postgres",
 	"INSERT INTO tst SELECT array_agg(n), array_agg(n), array_agg(n) FROM generate_series(1, $dim) n"
 );
 like($stderr, qr/row is too big/);
-
-done_testing();

test/t/010_hnsw_wal.pl

@@ -1,99 +0,0 @@
-# Based on postgres/contrib/bloom/t/001_wal.pl
-
-# Test generic xlog record work for hnsw index replication.
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-my $dim = 32;
-
-my $node_primary;
-my $node_replica;
-
-# Run few queries on both primary and replica and check their results match.
-sub test_index_replay
-{
-	my ($test_name) = @_;
-
-	# Wait for replica to catch up
-	my $applname = $node_replica->name;
-
-	my $server_version_num = $node_primary->safe_psql("postgres", "SHOW server_version_num");
-	my $caughtup_query = "SELECT pg_current_wal_lsn() <= replay_lsn FROM pg_stat_replication WHERE application_name = '$applname';";
-	$node_primary->poll_query_until('postgres', $caughtup_query)
-	  or die "Timed out while waiting for replica 1 to catch up";
-
-	my @r = ();
-	for (1 .. $dim) {
-		push(@r, rand());
-	}
-	my $sql = join(",", @r);
-
-	my $queries = qq(
-		SET enable_seqscan = off;
-		SELECT * FROM tst ORDER BY v <-> '[$sql]' LIMIT 10;
-	);
-
-	# Run test queries and compare their result
-	my $primary_result = $node_primary->safe_psql("postgres", $queries);
-	my $replica_result = $node_replica->safe_psql("postgres", $queries);
-
-	is($primary_result, $replica_result, "$test_name: query result matches");
-	return;
-}
-
-# Use ARRAY[random(), random(), random(), ...] over
-# SELECT array_agg(random()) FROM generate_series(1, $dim)
-# to generate different values for each row
-my $array_sql = join(",", ('random()') x $dim);
-
-# Initialize primary node
-$node_primary = get_new_node('primary');
-$node_primary->init(allows_streaming => 1);
-if ($dim > 32) {
-	# TODO use wal_keep_segments for Postgres < 13
-	$node_primary->append_conf('postgresql.conf', qq(wal_keep_size = 1GB));
-}
-if ($dim > 1500) {
-	$node_primary->append_conf('postgresql.conf', qq(maintenance_work_mem = 128MB));
-}
-$node_primary->start;
-my $backup_name = 'my_backup';
-
-# Take backup
-$node_primary->backup($backup_name);
-
-# Create streaming replica linking to primary
-$node_replica = get_new_node('replica');
-$node_replica->init_from_backup($node_primary, $backup_name,
-	has_streaming => 1);
-$node_replica->start;
-
-# Create hnsw index on primary
-$node_primary->safe_psql("postgres", "CREATE EXTENSION vector;");
-$node_primary->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
-$node_primary->safe_psql("postgres",
-	"INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series(1, 1000) i;"
-);
-$node_primary->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops);");
-
-# Test that queries give same result
-test_index_replay('initial');
-
-# Run 10 cycles of table modification. Run test queries after each modification.
-for my $i (1 .. 10)
-{
-	$node_primary->safe_psql("postgres", "DELETE FROM tst WHERE i = $i;");
-	test_index_replay("delete $i");
-	$node_primary->safe_psql("postgres", "VACUUM tst;");
-	test_index_replay("vacuum $i");
-	my ($start, $end) = (1001 + ($i - 1) * 100, 1000 + $i * 100);
-	$node_primary->safe_psql("postgres",
-		"INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series($start, $end) i;"
-	);
-	test_index_replay("insert $i");
-}
-
-done_testing();

test/t/011_hnsw_vacuum.pl

@@ -1,43 +0,0 @@
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-my $dim = 3;
-
-my @r = ();
-for (1 .. $dim) {
-	my $v = int(rand(1000)) + 1;
-	push(@r, "i % $v");
-}
-my $array_sql = join(", ", @r);
-
-# Initialize node
-my $node = get_new_node('node');
-$node->init;
-$node->start;
-
-# Create table and index
-$node->safe_psql("postgres", "CREATE EXTENSION vector;");
-$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector($dim));");
-$node->safe_psql("postgres",
-	"INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series(1, 10000) i;"
-);
-$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops);");
-
-# Get size
-my $size = $node->safe_psql("postgres", "SELECT pg_total_relation_size('tst_v_idx');");
-
-# Delete all, vacuum, and insert same data
-$node->safe_psql("postgres", "DELETE FROM tst;");
-$node->safe_psql("postgres", "VACUUM tst;");
-$node->safe_psql("postgres",
-	"INSERT INTO tst SELECT i % 10, ARRAY[$array_sql] FROM generate_series(1, 10000) i;"
-);
-
-# Check size
-my $new_size = $node->safe_psql("postgres", "SELECT pg_total_relation_size('tst_v_idx');");
-cmp_ok($new_size, "<=", $size * 1.01, "size does not increase too much");
-
-done_testing();

test/t/013_hnsw_insert_recall.pl

@@ -1,103 +0,0 @@
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-my $node;
-my @queries = ();
-my @expected;
-my $limit = 20;
-
-sub test_recall
-{
-	my ($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 = get_new_node('node');
-$node->init;
-$node->start;
-
-# Create table
-$node->safe_psql("postgres", "CREATE EXTENSION vector;");
-$node->safe_psql("postgres", "CREATE TABLE tst (i int4, v vector(3));");
-
-# Generate queries
-for (1..20) {
-	my $r1 = rand();
-	my $r2 = rand();
-	my $r3 = rand();
-	push(@queries, "[$r1,$r2,$r3]");
-}
-
-# Check each index type
-my @operators = ("<->", "<#>", "<=>");
-
-foreach (@operators) {
-	my $operator = $_;
-
-	# Add index
-	my $opclass;
-	if ($operator eq "<->") {
-		$opclass = "vector_l2_ops";
-	} elsif ($operator eq "<#>") {
-		$opclass = "vector_ip_ops";
-	} else {
-		$opclass = "vector_cosine_ops";
-	}
-	$node->safe_psql("postgres", "CREATE INDEX idx ON tst USING hnsw (v $opclass);");
-
-	$node->safe_psql("postgres",
-		"INSERT INTO tst SELECT i, ARRAY[random(), random(), random()] FROM generate_series(1, 10000) i;"
-	);
-
-	# 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, $res);
-	}
-
-	if ($operator eq "<#>") {
-		test_recall(0.80, $operator);
-	} else {
-		test_recall(0.99, $operator);
-	}
-
-	$node->safe_psql("postgres", "DROP INDEX idx;");
-	$node->safe_psql("postgres", "TRUNCATE tst;");
-}
-
-done_testing();

test/t/014_hnsw_inserts.pl

@@ -1,58 +0,0 @@
-use strict;
-use warnings;
-use PostgresNode;
-use TestLib;
-use Test::More;
-
-# Ensures elements and neighbors on both same and different pages
-my $dim = 1900;
-
-my $array_sql = join(",", ('random()') x $dim);
-
-# Initialize node
-my $node = get_new_node('node');
-$node->init;
-$node->start;
-
-# Create table and index
-$node->safe_psql("postgres", "CREATE EXTENSION vector;");
-$node->safe_psql("postgres", "CREATE TABLE tst (v vector($dim));");
-$node->safe_psql("postgres",
-	"INSERT INTO tst SELECT ARRAY[$array_sql] FROM generate_series(1, 100) i;"
-);
-$node->safe_psql("postgres", "CREATE INDEX ON tst USING hnsw (v vector_l2_ops);");
-
-$node->pgbench(
-	"--no-vacuum --client=5 --transactions=100",
-	0,
-	[qr{actually processed}],
-	[qr{^$}],
-	"concurrent INSERTs",
-	{
-		"007_inserts" => "INSERT INTO tst SELECT ARRAY[$array_sql] FROM generate_series(1, 10) i;"
-	}
-);
-
-sub idx_scan
-{
-	# Stats do not update instantaneously
-	# https://www.postgresql.org/docs/current/monitoring-stats.html#MONITORING-STATS-VIEWS
-	sleep(1);
-	$node->safe_psql("postgres", "SELECT idx_scan FROM pg_stat_user_indexes WHERE indexrelid = 'tst_v_idx'::regclass;");
-}
-
-my $expected = 100 + 5 * 100 * 10;
-
-my $count = $node->safe_psql("postgres", "SELECT COUNT(*) FROM tst;");
-is($count, $expected);
-is(idx_scan(), 0);
-
-$count = $node->safe_psql("postgres", qq(
-	SET enable_seqscan = off;
-	SET hnsw.ef_search = 400;
-	SELECT COUNT(*) FROM (SELECT v FROM tst ORDER BY v <-> (SELECT v FROM tst LIMIT 1)) t;
-));
-is($count, 400);
-is(idx_scan(), 1);
-
-done_testing();

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat access method'
-default_version = '0.4.4'
+default_version = '0.4.1'
 module_pathname = '$libdir/vector'
 relocatable = true