Started Neon intrinsics [skip ci]

CHANGELOG.md

@@ -1,4 +1,4 @@
-## 0.7.0 (2024-04-29)
+## 0.7.0 (unreleased)
 
 - Added `halfvec` type
 - Added `sparsevec` type

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": "0.7.0",
+	"version": "0.6.2",
 	"maintainer": [
 		"Andrew Kane <andrew@ankane.org>"
 	],
@@ -20,7 +20,7 @@
 		"vector": {
 			"file": "sql/vector.sql",
 			"docfile": "README.md",
-			"version": "0.7.0",
+			"version": "0.6.2",
 			"abstract": "Open-source vector similarity search for Postgres"
 		}
 	},

Makefile

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.7.0
+EXTVERSION = 0.6.2
 
 MODULE_big = vector
 DATA = $(wildcard sql/*--*.sql)

Makefile.win

@@ -1,5 +1,5 @@
 EXTENSION = vector
-EXTVERSION = 0.7.0
+EXTVERSION = 0.6.2
 
 OBJS = src\bitutils.obj src\bitvec.obj src\halfutils.obj src\halfvec.obj src\hnsw.obj src\hnswbuild.obj src\hnswinsert.obj src\hnswscan.obj src\hnswutils.obj src\hnswvacuum.obj src\ivfbuild.obj src\ivfflat.obj src\ivfinsert.obj src\ivfkmeans.obj src\ivfscan.obj src\ivfutils.obj src\ivfvacuum.obj src\sparsevec.obj src\vector.obj
 HEADERS = src\halfvec.h src\sparsevec.h src\vector.h

README.md

@@ -5,8 +5,7 @@ Open-source vector similarity search for Postgres
 Store your vectors with the rest of your data. Supports:
 
 - exact and approximate nearest neighbor search
-- single-precision, half-precision, binary, and sparse vectors
+- L2 distance, inner product, and cosine distance
-- L2 distance, inner product, cosine distance, L1 distance, Hamming distance, and Jaccard 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
@@ -21,7 +20,7 @@ Compile and install the extension (supports Postgres 12+)
 
 ```sh
 cd /tmp
-git clone --branch v0.7.0 https://github.com/pgvector/pgvector.git
+git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
 cd pgvector
 make
 make install # may need sudo
@@ -46,7 +45,7 @@ Then use `nmake` to build:
 ```cmd
 set "PGROOT=C:\Program Files\PostgreSQL\16"
 cd %TEMP%
-git clone --branch v0.7.0 https://github.com/pgvector/pgvector.git
+git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
 cd pgvector
 nmake /F Makefile.win
 nmake /F Makefile.win install
@@ -82,7 +81,7 @@ Get the nearest neighbors by L2 distance
 SELECT * FROM items ORDER BY embedding <-> '[3,1,2]' LIMIT 5;
 ```
 
-Also supports inner product (`<#>`), cosine distance (`<=>`), and L1 distance (`<+>`, added in 0.7.0)
+Also supports inner product (`<#>`), cosine distance (`<=>`), and L1 distance (`<+>`, unreleased)
 
 Note: `<#>` returns the negative inner product since Postgres only supports `ASC` order index scans on operators
 
@@ -144,7 +143,7 @@ Supported distance functions are:
 - `<->` - L2 distance
 - `<#>` - (negative) inner product
 - `<=>` - cosine distance
-- `<+>` - L1 distance (added in 0.7.0)
+- `<+>` - L1 distance (unreleased)
 
 Get the nearest neighbors to a row
 
@@ -229,19 +228,19 @@ Cosine distance
 CREATE INDEX ON items USING hnsw (embedding vector_cosine_ops);
 ```
 
-L1 distance - added in 0.7.0
+L1 distance - unreleased
 
 ```sql
 CREATE INDEX ON items USING hnsw (embedding vector_l1_ops);
 ```
 
-Hamming distance - added in 0.7.0
+Hamming distance - unreleased
 
 ```sql
 CREATE INDEX ON items USING hnsw (embedding bit_hamming_ops);
 ```
 
-Jaccard distance - added in 0.7.0
+Jaccard distance - unreleased
 
 ```sql
 CREATE INDEX ON items USING hnsw (embedding bit_jaccard_ops);
@@ -250,9 +249,9 @@ CREATE INDEX ON items USING hnsw (embedding bit_jaccard_ops);
 Supported types are:
 
 - `vector` - up to 2,000 dimensions
-- `halfvec` - up to 4,000 dimensions (added in 0.7.0)
+- `halfvec` - up to 4,000 dimensions (unreleased)
-- `bit` - up to 64,000 dimensions (added in 0.7.0)
+- `bit` - up to 64,000 dimensions (unreleased)
-- `sparsevec` - up to 1,000 non-zero elements (added in 0.7.0)
+- `sparsevec` - up to 1,000 non-zero elements (unreleased)
 
 ### Index Options
 
@@ -357,7 +356,7 @@ Cosine distance
 CREATE INDEX ON items USING ivfflat (embedding vector_cosine_ops) WITH (lists = 100);
 ```
 
-Hamming distance - added in 0.7.0
+Hamming distance - unreleased
 
 ```sql
 CREATE INDEX ON items USING ivfflat (embedding bit_hamming_ops) WITH (lists = 100);
@@ -366,8 +365,8 @@ CREATE INDEX ON items USING ivfflat (embedding bit_hamming_ops) WITH (lists = 10
 Supported types are:
 
 - `vector` - up to 2,000 dimensions
-- `halfvec` - up to 4,000 dimensions (added in 0.7.0)
+- `halfvec` - up to 4,000 dimensions (unreleased)
-- `bit` - up to 64,000 dimensions (added in 0.7.0)
+- `bit` - up to 64,000 dimensions (unreleased)
 
 ### Query Options
 
@@ -441,9 +440,9 @@ Use [partitioning](https://www.postgresql.org/docs/current/ddl-partitioning.html
 CREATE TABLE items (embedding vector(3), category_id int) PARTITION BY LIST(category_id);
 ```
 
-## Half-Precision Vectors
+## Half Vectors
 
-*Added in 0.7.0*
+*Unreleased*
 
 Use the `halfvec` type to store half-precision vectors
 
@@ -451,9 +450,9 @@ Use the `halfvec` type to store half-precision vectors
 CREATE TABLE items (id bigserial PRIMARY KEY, embedding halfvec(3));
 ```
 
-## Half-Precision Indexing
+## Half Indexing
 
-*Added in 0.7.0*
+*Unreleased*
 
 Index vectors at half precision for smaller indexes and faster build times
 
@@ -476,23 +475,23 @@ CREATE TABLE items (id bigserial PRIMARY KEY, embedding bit(3));
 INSERT INTO items (embedding) VALUES ('000'), ('111');
 ```
 
-Get the nearest neighbors by Hamming distance (added in 0.7.0)
+Get the nearest neighbors by Hamming distance
-
-```sql
-SELECT * FROM items ORDER BY embedding <~> '101' LIMIT 5;
-```
-
-Or (before 0.7.0)
 
 ```sql
 SELECT * FROM items ORDER BY bit_count(embedding # '101') LIMIT 5;
 ```
 
+Or (unreleased)
+
+```sql
+SELECT * FROM items ORDER BY embedding <~> '101' LIMIT 5;
+```
+
 Also supports Jaccard distance (`<%>`)
 
 ## Binary Quantization
 
-*Added in 0.7.0*
+*Unreleased*
 
 Use expression indexing for binary quantization
 
@@ -516,7 +515,7 @@ SELECT * FROM (
 
 ## Sparse Vectors
 
-*Added in 0.7.0*
+*Unreleased*
 
 Use the `sparsevec` type to store sparse vectors
 
@@ -551,7 +550,7 @@ You can use [Reciprocal Rank Fusion](https://github.com/pgvector/pgvector-python
 
 ## Indexing Subvectors
 
-*Added in 0.7.0*
+*Unreleased*
 
 Use expression indexing to index subvectors
 
@@ -865,23 +864,23 @@ Operator | Description | Added
 \+ | element-wise addition |
 \- | element-wise subtraction |
 \* | element-wise multiplication | 0.5.0
-\|\| | concatenate | 0.7.0
+\|\| | concatenate | unreleased
 <-> | Euclidean distance |
 <#> | negative inner product |
 <=> | cosine distance |
-<+> | taxicab distance | 0.7.0
+<+> | taxicab distance | unreleased
 
 ### Vector Functions
 
 Function | Description | Added
 --- | --- | ---
-binary_quantize(vector) → bit | binary quantize | 0.7.0
+binary_quantize(vector) → bit | binary quantize | unreleased
 cosine_distance(vector, vector) → double precision | cosine distance |
 inner_product(vector, vector) → double precision | inner product |
 l1_distance(vector, vector) → double precision | taxicab distance | 0.5.0
 l2_distance(vector, vector) → double precision | Euclidean distance |
-l2_normalize(vector) → vector | Normalize with Euclidean norm | 0.7.0
+l2_normalize(vector) → vector | Normalize with Euclidean norm | unreleased
-subvector(vector, integer, integer) → vector | subvector | 0.7.0
+subvector(vector, integer, integer) → vector | subvector | unreleased
 vector_dims(vector) → integer | number of dimensions |
 vector_norm(vector) → double precision | Euclidean norm |
 
@@ -900,35 +899,35 @@ Each half vector takes `2 * dimensions + 8` bytes of storage. Each element is a
 
 Operator | Description | Added
 --- | --- | ---
-\+ | element-wise addition | 0.7.0
+\+ | element-wise addition | unreleased
-\- | element-wise subtraction | 0.7.0
+\- | element-wise subtraction | unreleased
-\* | element-wise multiplication | 0.7.0
+\* | element-wise multiplication | unreleased
-\|\| | concatenate | 0.7.0
+\|\| | concatenate | unreleased
-<-> | Euclidean distance | 0.7.0
+<-> | Euclidean distance | unreleased
-<#> | negative inner product | 0.7.0
+<#> | negative inner product | unreleased
-<=> | cosine distance | 0.7.0
+<=> | cosine distance | unreleased
-<+> | taxicab distance | 0.7.0
+<+> | taxicab distance | unreleased
 
 ### Halfvec Functions
 
 Function | Description | Added
 --- | --- | ---
-binary_quantize(halfvec) → bit | binary quantize | 0.7.0
+binary_quantize(halfvec) → bit | binary quantize | unreleased
-cosine_distance(halfvec, halfvec) → double precision | cosine distance | 0.7.0
+cosine_distance(halfvec, halfvec) → double precision | cosine distance | unreleased
-inner_product(halfvec, halfvec) → double precision | inner product | 0.7.0
+inner_product(halfvec, halfvec) → double precision | inner product | unreleased
-l1_distance(halfvec, halfvec) → double precision | taxicab distance | 0.7.0
+l1_distance(halfvec, halfvec) → double precision | taxicab distance | unreleased
-l2_distance(halfvec, halfvec) → double precision | Euclidean distance | 0.7.0
+l2_distance(halfvec, halfvec) → double precision | Euclidean distance | unreleased
-l2_norm(halfvec) → double precision | Euclidean norm | 0.7.0
+l2_norm(halfvec) → double precision | Euclidean norm | unreleased
-l2_normalize(halfvec) → halfvec | Normalize with Euclidean norm | 0.7.0
+l2_normalize(halfvec) → halfvec | Normalize with Euclidean norm | unreleased
-subvector(halfvec, integer, integer) → halfvec | subvector | 0.7.0
+subvector(halfvec, integer, integer) → halfvec | subvector | unreleased
-vector_dims(halfvec) → integer | number of dimensions | 0.7.0
+vector_dims(halfvec) → integer | number of dimensions | unreleased
 
 ### Halfvec Aggregate Functions
 
 Function | Description | Added
 --- | --- | ---
-avg(halfvec) → halfvec | average | 0.7.0
+avg(halfvec) → halfvec | average | unreleased
-sum(halfvec) → halfvec | sum | 0.7.0
+sum(halfvec) → halfvec | sum | unreleased
 
 ### Bit Type
 
@@ -938,15 +937,15 @@ Each bit vector takes `dimensions / 8 + 8` bytes of storage. See the [Postgres d
 
 Operator | Description | Added
 --- | --- | ---
-<~> | Hamming distance | 0.7.0
+<~> | Hamming distance | unreleased
-<%> | Jaccard distance | 0.7.0
+<%> | Jaccard distance | unreleased
 
 ### Bit Functions
 
 Function | Description | Added
 --- | --- | ---
-hamming_distance(bit, bit) → double precision | Hamming distance | 0.7.0
+hamming_distance(bit, bit) → double precision | Hamming distance | unreleased
-jaccard_distance(bit, bit) → double precision | Jaccard distance | 0.7.0
+jaccard_distance(bit, bit) → double precision | Jaccard distance | unreleased
 
 ### Sparsevec Type
 
@@ -956,21 +955,21 @@ Each sparse vector takes `8 * non-zero elements + 16` bytes of storage. Each ele
 
 Operator | Description | Added
 --- | --- | ---
-<-> | Euclidean distance | 0.7.0
+<-> | Euclidean distance | unreleased
-<#> | negative inner product | 0.7.0
+<#> | negative inner product | unreleased
-<=> | cosine distance | 0.7.0
+<=> | cosine distance | unreleased
-<+> | taxicab distance | 0.7.0
+<+> | taxicab distance | unreleased
 
 ### Sparsevec Functions
 
 Function | Description | Added
 --- | --- | ---
-cosine_distance(sparsevec, sparsevec) → double precision | cosine distance | 0.7.0
+cosine_distance(sparsevec, sparsevec) → double precision | cosine distance | unreleased
-inner_product(sparsevec, sparsevec) → double precision | inner product | 0.7.0
+inner_product(sparsevec, sparsevec) → double precision | inner product | unreleased
-l1_distance(sparsevec, sparsevec) → double precision | taxicab distance | 0.7.0
+l1_distance(sparsevec, sparsevec) → double precision | taxicab distance | unreleased
-l2_distance(sparsevec, sparsevec) → double precision | Euclidean distance | 0.7.0
+l2_distance(sparsevec, sparsevec) → double precision | Euclidean distance | unreleased
-l2_norm(sparsevec) → double precision | Euclidean norm | 0.7.0
+l2_norm(sparsevec) → double precision | Euclidean norm | unreleased
-l2_normalize(sparsevec) → sparsevec | Normalize with Euclidean norm | 0.7.0
+l2_normalize(sparsevec) → sparsevec | Normalize with Euclidean norm | unreleased
 
 ## Installation Notes - Linux and Mac
 
@@ -1047,7 +1046,7 @@ This adds pgvector to the [Postgres image](https://hub.docker.com/_/postgres) (r
 You can also build the image manually:
 
 ```sh
-git clone --branch v0.7.0 https://github.com/pgvector/pgvector.git
+git clone --branch v0.6.2 https://github.com/pgvector/pgvector.git
 cd pgvector
 docker build --build-arg PG_MAJOR=16 -t myuser/pgvector .
 ```

src/halfutils.c

@@ -1,5 +1,7 @@
 #include "postgres.h"
 
+#include <arm_neon.h>
+
 #include "halfutils.h"
 #include "halfvec.h"
 
@@ -28,9 +30,28 @@ static float
 HalfvecL2SquaredDistanceDefault(int dim, half * ax, half * bx)
 {
 	float		distance = 0.0;
+	int			i = 0;
+
+/* TODO Improve */
+#ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC
+	int			count = (dim / 8) * 8;
+	float32x4_t dist = vmovq_n_f32(0);
+
+	for (; i < count; i += 8)
+	{
+		float16x8_t axs = vld1q_f16((const __fp16 *) (ax + i));
+		float16x8_t bxs = vld1q_f16((const __fp16 *) (bx + i));
+		float16x8_t diff = vsubq_f16(axs, bxs);
+
+		dist = vfmlalq_low_f16(dist, diff, diff);
+		dist = vfmlalq_high_f16(dist, diff, diff);
+	}
+
+	distance += vaddvq_f32(dist);
+#endif
 
 	/* Auto-vectorized */
-	for (int i = 0; i < dim; i++)
+	for (; i < dim; i++)
 	{
 		float		diff = HalfToFloat4(ax[i]) - HalfToFloat4(bx[i]);
 

vector.control

@@ -1,4 +1,4 @@
 comment = 'vector data type and ivfflat and hnsw access methods'
-default_version = '0.7.0'
+default_version = '0.6.2'
 module_pathname = '$libdir/vector'
 relocatable = true