Improved code [skip ci]

.dockerignore

@@ -1,15 +1,8 @@
 /.git/
 /dist/
-/log/
 /results/
 /tmp_check/
 /sql/vector--?.?.?.sql
 regression.*
 *.o
 *.so
-*.bc
-*.dll
-*.dylib
-*.obj
-*.lib
-*.exp

.github/workflows/build.yml

@@ -57,7 +57,7 @@ jobs:
           postgres-version: ${{ matrix.postgres }}
       - run: make
         env:
-          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter -Wno-unknown-warning-option
+          PG_CFLAGS: -DUSE_ASSERT_CHECKING -Wall -Wextra -Werror -Wno-unused-parameter
       - run: make install
       - run: make installcheck
       - if: ${{ failure() }}

CHANGELOG.md

@@ -1,7 +1,3 @@
-## 0.8.1 (unreleased)
-
-- Improved performance of HNSW index scans for Postgres 17
-
 ## 0.8.0 (2024-10-30)
 
 - Added support for iterative index scans

README.md

@@ -33,10 +33,18 @@ You can also install it with [Docker](#docker), [Homebrew](#homebrew), [PGXN](#p
 
 ### Windows
 
-Ensure [C++ support in Visual Studio](https://learn.microsoft.com/en-us/cpp/build/building-on-the-command-line?view=msvc-170#download-and-install-the-tools) is installed and run `x64 Native Tools Command Prompt for VS [version]` as administrator. Then use `nmake` to build:
+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
-set "PGROOT=C:\Program Files\PostgreSQL\17"
+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:
+
+```cmd
+set "PGROOT=C:\Program Files\PostgreSQL\16"
 cd %TEMP%
 git clone --branch v0.8.0 https://github.com/pgvector/pgvector.git
 cd pgvector
@@ -492,11 +500,9 @@ With relaxed ordering, you can use a [materialized CTE](https://www.postgresql.o
 ```sql
 WITH relaxed_results AS MATERIALIZED (
     SELECT id, embedding <-> '[1,2,3]' AS distance FROM items WHERE category_id = 123 ORDER BY distance LIMIT 5
-) SELECT * FROM relaxed_results ORDER BY distance + 0;
+) SELECT * FROM relaxed_results ORDER BY distance;
 ```
 
-Note: `+ 0` is needed for Postgres 17+
-
 For queries that filter by distance, use a materialized CTE and place the distance filter outside of it for best performance (due to the [current behavior](https://www.postgresql.org/message-id/flat/CAOdR5yGUoMQ6j7M5hNUXrySzaqZVGf_Ne%2B8fwZMRKTFxU1nbJg%40mail.gmail.com) of the Postgres executor)
 
 ```sql
@@ -785,7 +791,7 @@ Go | [pgvector-go](https://github.com/pgvector/pgvector-go)
 Haskell | [pgvector-haskell](https://github.com/pgvector/pgvector-haskell)
 Java, Kotlin, Groovy, Scala | [pgvector-java](https://github.com/pgvector/pgvector-java)
 JavaScript, TypeScript | [pgvector-node](https://github.com/pgvector/pgvector-node)
-Julia | [Pgvector.jl](https://github.com/pgvector/Pgvector.jl)
+Julia | [pgvector-julia](https://github.com/pgvector/pgvector-julia)
 Lisp | [pgvector-lisp](https://github.com/pgvector/pgvector-lisp)
 Lua | [pgvector-lua](https://github.com/pgvector/pgvector-lua)
 Nim | [pgvector-nim](https://github.com/pgvector/pgvector-nim)
@@ -1121,7 +1127,7 @@ If compilation fails with `Cannot open include file: 'postgres.h': No such file
 
 ### Mismatched Architecture
 
-If compilation fails with `error C2196: case value '4' already used`, make sure you’re using the `x64 Native Tools Command Prompt`. Then run `nmake /F Makefile.win clean` and re-run the installation instructions.
+If compilation fails with `error C2196: case value '4' already used`, make sure `vcvars64.bat` was called. Then run `nmake /F Makefile.win clean` and re-run the installation instructions.
 
 ### Missing Symbol
 

src/hnsw.h

@@ -362,13 +362,6 @@ typedef union
 	ItemPointerData indextid;
 }			HnswUnvisited;
 
-typedef struct HnswReadStreamData
-{
-	HnswUnvisited *unvisited;
-	int			unvisitedLength;
-	int			visited;
-}			HnswReadStreamData;
-
 typedef struct HnswScanOpaqueData
 {
 	const		HnswTypeInfo *typeInfo;
@@ -424,13 +417,13 @@ bool		HnswCheckNorm(HnswSupport * support, Datum value);
 Buffer		HnswNewBuffer(Relation index, ForkNumber forkNum);
 void		HnswInitPage(Buffer buf, Page page);
 void		HnswInit(void);
-List	   *HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples, bool maintenance);
+List	   *HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples);
 HnswElement HnswGetEntryPoint(Relation index);
 void		HnswGetMetaPageInfo(Relation index, int *m, HnswElement * entryPoint);
 void	   *HnswAlloc(HnswAllocator * allocator, Size size);
 HnswElement HnswInitElement(char *base, ItemPointer tid, int m, double ml, int maxLevel, HnswAllocator * alloc);
 HnswElement HnswInitElementFromBlock(BlockNumber blkno, OffsetNumber offno);
-void		HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing, bool maintenance);
+void		HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing);
 HnswSearchCandidate *HnswEntryCandidate(char *base, HnswElement em, HnswQuery * q, Relation rel, HnswSupport * support, bool loadVec);
 void		HnswUpdateMetaPage(Relation index, int updateEntry, HnswElement entryPoint, BlockNumber insertPage, ForkNumber forkNum, bool building);
 void		HnswSetNeighborTuple(char *base, HnswNeighborTuple ntup, HnswElement e, int m);

src/hnswbuild.c

@@ -457,7 +457,7 @@ InsertTupleInMemory(HnswBuildState * buildstate, HnswElement element)
 	}
 
 	/* Find neighbors for element */
-	HnswFindElementNeighbors(base, element, entryPoint, NULL, support, m, efConstruction, false, true);
+	HnswFindElementNeighbors(base, element, entryPoint, NULL, support, m, efConstruction, false);
 
 	/* Update graph in memory */
 	UpdateGraphInMemory(support, element, m, efConstruction, entryPoint, buildstate);

src/hnswinsert.c

@@ -725,7 +725,7 @@ HnswInsertTupleOnDisk(Relation index, HnswSupport * support, Datum value, ItemPo
 	}
 
 	/* Find neighbors for element */
-	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false, building);
+	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, false);
 
 	/* Update graph on disk */
 	UpdateGraphOnDisk(index, support, element, m, efConstruction, entryPoint, building);

src/hnswscan.c

@@ -37,11 +37,11 @@ GetScanItems(IndexScanDesc scan, Datum value)
 
 	for (int lc = entryPoint->level; lc >= 1; lc--)
 	{
-		w = HnswSearchLayer(base, q, ep, 1, lc, index, support, m, false, NULL, NULL, NULL, true, NULL, false);
+		w = HnswSearchLayer(base, q, ep, 1, lc, index, support, m, false, NULL, NULL, NULL, true, NULL);
 		ep = w;
 	}
 
-	return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, support, m, false, NULL, &so->v, hnsw_iterative_scan != HNSW_ITERATIVE_SCAN_OFF ? &so->discarded : NULL, true, &so->tuples, false);
+	return HnswSearchLayer(base, q, ep, hnsw_ef_search, 0, index, support, m, false, NULL, &so->v, hnsw_iterative_scan != HNSW_ITERATIVE_SCAN_OFF ? &so->discarded : NULL, true, &so->tuples);
 }
 
 /*
@@ -72,7 +72,7 @@ ResumeScanItems(IndexScanDesc scan)
 		ep = lappend(ep, sc);
 	}
 
-	return HnswSearchLayer(base, &so->q, ep, batch_size, 0, index, &so->support, so->m, false, NULL, &so->v, &so->discarded, false, &so->tuples, false);
+	return HnswSearchLayer(base, &so->q, ep, batch_size, 0, index, &so->support, so->m, false, NULL, &so->v, &so->discarded, false, &so->tuples);
 }
 
 /*

src/hnswutils.c

@@ -15,10 +15,6 @@
 #include "utils/memdebug.h"
 #include "utils/rel.h"
 
-#if PG_VERSION_NUM >= 170000
-#include "storage/read_stream.h"
-#endif
-
 #if PG_VERSION_NUM < 170000
 static inline uint64
 murmurhash64(uint64 data)
@@ -529,12 +525,14 @@ HnswGetDistance(Datum a, Datum b, HnswSupport * support)
  * Load an element and optionally get its distance from q
  */
 static void
-HnswLoadElementImpl(Buffer buf, OffsetNumber offno, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance, HnswElement * element)
+HnswLoadElementImpl(BlockNumber blkno, OffsetNumber offno, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance, HnswElement * element)
 {
+	Buffer		buf;
 	Page		page;
 	HnswElementTuple etup;
 
 	/* Read vector */
+	buf = ReadBuffer(index, blkno);
 	LockBuffer(buf, BUFFER_LOCK_SHARE);
 	page = BufferGetPage(buf);
 
@@ -555,7 +553,7 @@ HnswLoadElementImpl(Buffer buf, OffsetNumber offno, double *distance, HnswQuery
 	if (distance == NULL || maxDistance == NULL || *distance < *maxDistance)
 	{
 		if (*element == NULL)
-			*element = HnswInitElementFromBlock(BufferGetBlockNumber(buf), offno);
+			*element = HnswInitElementFromBlock(blkno, offno);
 
 		HnswLoadElementFromTuple(*element, etup, true, loadVec);
 	}
@@ -569,9 +567,7 @@ HnswLoadElementImpl(Buffer buf, OffsetNumber offno, double *distance, HnswQuery
 void
 HnswLoadElement(HnswElement element, double *distance, HnswQuery * q, Relation index, HnswSupport * support, bool loadVec, double *maxDistance)
 {
-	Buffer		buf = ReadBuffer(index, element->blkno);
+	HnswLoadElementImpl(element->blkno, element->offno, distance, q, index, support, loadVec, maxDistance, &element);
-
-	HnswLoadElementImpl(buf, element->offno, distance, q, index, support, loadVec, maxDistance, &element);
 }
 
 /*
@@ -809,33 +805,19 @@ HnswLoadUnvisitedFromDisk(HnswElement element, HnswUnvisited * unvisited, int *u
 		if (!found)
 			unvisited[(*unvisitedLength)++].indextid = *indextid;
 	}
-}
 
-#if PG_VERSION_NUM >= 170000
+#ifdef USE_PREFETCH
-/*
+	/* TODO limit by get_tablespace_io_concurrency */
- * Get next block number for read stream
+	for (int i = 0; i < *unvisitedLength; i++)
- */
+		PrefetchBuffer(index, MAIN_FORKNUM, ItemPointerGetBlockNumber(&unvisited[i].indextid));
-static BlockNumber
-HnswReadStreamNextBlock(ReadStream *stream, void *callback_private_data, void *per_buffer_data)
-{
-	HnswReadStreamData *streamData = callback_private_data;
-	OffsetNumber *offno = per_buffer_data;
-	HnswUnvisited *uv;
-
-	if (streamData->visited == streamData->unvisitedLength)
-		return InvalidBlockNumber;
-
-	uv = &streamData->unvisited[streamData->visited++];
-	*offno = ItemPointerGetOffsetNumber(&uv->indextid);
-	return ItemPointerGetBlockNumber(&uv->indextid);
-}
 #endif
+}
 
 /*
  * Algorithm 2 from paper
  */
 List *
-HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples, bool maintenance)
+HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation index, HnswSupport * support, int m, bool inserting, HnswElement skipElement, visited_hash * v, pairingheap **discarded, bool initVisited, int64 *tuples)
 {
 	List	   *w = NIL;
 	pairingheap *C = pairingheap_allocate(CompareNearestCandidates, NULL);
@@ -850,18 +832,6 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 	int			unvisitedLength;
 	bool		inMemory = index == NULL;
 
-#if PG_VERSION_NUM >= 170000
-	HnswReadStreamData streamData;
-	ReadStream *stream = NULL;
-
-	if (!inMemory)
-	{
-		int			flags = maintenance ? READ_STREAM_MAINTENANCE : READ_STREAM_DEFAULT;
-
-		stream = read_stream_begin_relation(flags, NULL, index, MAIN_FORKNUM, HnswReadStreamNextBlock, &streamData, sizeof(OffsetNumber));
-	}
-#endif
-
 	if (v == NULL)
 	{
 		v = &vh;
@@ -924,18 +894,8 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 		if (inMemory)
 			HnswLoadUnvisitedFromMemory(base, cElement, unvisited, &unvisitedLength, v, lc, localNeighborhood, neighborhoodSize);
 		else
-		{
 			HnswLoadUnvisitedFromDisk(cElement, unvisited, &unvisitedLength, v, index, m, lm, lc);
 
-#if PG_VERSION_NUM >= 170000
-			read_stream_reset(stream);
-
-			streamData.unvisited = unvisited;
-			streamData.unvisitedLength = unvisitedLength;
-			streamData.visited = 0;
-#endif
-		}
-
 		/* OK to count elements instead of tuples */
 		if (tuples != NULL)
 			(*tuples) += unvisitedLength;
@@ -956,24 +916,13 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 			}
 			else
 			{
-				Buffer		buf;
-				OffsetNumber offno;
-
-#if PG_VERSION_NUM >= 170000
-				void	   *offnoPtr;
-
-				buf = read_stream_next_buffer(stream, &offnoPtr);
-				offno = *((OffsetNumber *) offnoPtr);
-#else
 				ItemPointer indextid = &unvisited[i].indextid;
-
+				BlockNumber blkno = ItemPointerGetBlockNumber(indextid);
-				buf = ReadBuffer(index, ItemPointerGetBlockNumber(indextid));
+				OffsetNumber offno = ItemPointerGetOffsetNumber(indextid);
-				offno = ItemPointerGetOffsetNumber(indextid);
-#endif
 
 				/* Avoid any allocations if not adding */
 				eElement = NULL;
-				HnswLoadElementImpl(buf, offno, &eDistance, q, index, support, inserting, alwaysAdd || discarded != NULL ? NULL : &f->distance, &eElement);
+				HnswLoadElementImpl(blkno, offno, &eDistance, q, index, support, inserting, alwaysAdd || discarded != NULL ? NULL : &f->distance, &eElement);
 
 				if (eElement == NULL)
 					continue;
@@ -1029,11 +978,6 @@ HnswSearchLayer(char *base, HnswQuery * q, List *ep, int ef, int lc, Relation in
 		w = lappend(w, sc);
 	}
 
-#if PG_VERSION_NUM >= 170000
-	if (!inMemory)
-		read_stream_end(stream);
-#endif
-
 	return w;
 }
 
@@ -1329,7 +1273,7 @@ PrecomputeHash(char *base, HnswElement element)
  * Algorithm 1 from paper
  */
 void
-HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing, bool maintenance)
+HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint, Relation index, HnswSupport * support, int m, int efConstruction, bool existing)
 {
 	List	   *ep;
 	List	   *w;
@@ -1356,7 +1300,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
 	/* 1st phase: greedy search to insert level */
 	for (int lc = entryLevel; lc >= level + 1; lc--)
 	{
-		w = HnswSearchLayer(base, &q, ep, 1, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL, maintenance);
+		w = HnswSearchLayer(base, &q, ep, 1, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL);
 		ep = w;
 	}
 
@@ -1375,7 +1319,7 @@ HnswFindElementNeighbors(char *base, HnswElement element, HnswElement entryPoint
 		List	   *lw = NIL;
 		ListCell   *lc2;
 
-		w = HnswSearchLayer(base, &q, ep, efConstruction, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL, maintenance);
+		w = HnswSearchLayer(base, &q, ep, efConstruction, lc, index, support, m, true, skipElement, NULL, NULL, true, NULL);
 
 		/* Convert search candidates to candidates */
 		foreach(lc2, w)

src/hnswvacuum.c

@@ -208,7 +208,7 @@ RepairGraphElement(HnswVacuumState * vacuumstate, HnswElement element, HnswEleme
 	element->heaptidsLength = 0;
 
 	/* Find neighbors for element, skipping itself */
-	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, true, true);
+	HnswFindElementNeighbors(base, element, entryPoint, index, support, m, efConstruction, true);
 
 	/* Zero memory for each element */
 	MemSet(ntup, 0, HNSW_TUPLE_ALLOC_SIZE);

src/ivfbuild.c

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

src/ivfflat.h

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

src/ivfutils.c

@@ -259,8 +259,8 @@ VectorUpdateCenter(Pointer v, int dimensions, float *x)
 	SET_VARSIZE(vec, VECTOR_SIZE(dimensions));
 	vec->dim = dimensions;
 
-	for (int i = 0; i < dimensions; i++)
+	for (int k = 0; k < dimensions; k++)
-		vec->x[i] = x[i];
+		vec->x[k] = x[k];
 }
 
 static void
@@ -271,8 +271,8 @@ HalfvecUpdateCenter(Pointer v, int dimensions, float *x)
 	SET_VARSIZE(vec, HALFVEC_SIZE(dimensions));
 	vec->dim = dimensions;
 
-	for (int i = 0; i < dimensions; i++)
+	for (int k = 0; k < dimensions; k++)
-		vec->x[i] = Float4ToHalfUnchecked(x[i]);
+		vec->x[k] = Float4ToHalfUnchecked(x[k]);
 }
 
 static void
@@ -284,33 +284,29 @@ BitUpdateCenter(Pointer v, int dimensions, float *x)
 	SET_VARSIZE(vec, VARBITTOTALLEN(dimensions));
 	VARBITLEN(vec) = dimensions;
 
-	for (uint32 i = 0; i < VARBITBYTES(vec); i++)
+	for (uint32 k = 0; k < VARBITBYTES(vec); k++)
-		nx[i] = 0;
+		nx[k] = 0;
 
-	for (int i = 0; i < dimensions; i++)
+	for (int k = 0; k < dimensions; k++)
-		nx[i / 8] |= (x[i] > 0.5 ? 1 : 0) << (7 - (i % 8));
+		nx[k / 8] |= (x[k] > 0.5 ? 1 : 0) << (7 - (k % 8));
 }
 
 static void
 VectorSumCenter(Pointer v, float *x)
 {
 	Vector	   *vec = (Vector *) v;
-	int			dim = vec->dim;
 
-	/* Auto-vectorized */
+	for (int k = 0; k < vec->dim; k++)
-	for (int i = 0; i < dim; i++)
+		x[k] += vec->x[k];
-		x[i] += vec->x[i];
 }
 
 static void
 HalfvecSumCenter(Pointer v, float *x)
 {
 	HalfVector *vec = (HalfVector *) v;
-	int			dim = vec->dim;
 
-	/* Auto-vectorized on aarch64 */
+	for (int k = 0; k < vec->dim; k++)
-	for (int i = 0; i < dim; i++)
+		x[k] += HalfToFloat4(vec->x[k]);
-		x[i] += HalfToFloat4(vec->x[i]);
 }
 
 static void
@@ -318,8 +314,8 @@ BitSumCenter(Pointer v, float *x)
 {
 	VarBit	   *vec = (VarBit *) v;
 
-	for (int i = 0; i < VARBITLEN(vec); i++)
+	for (int k = 0; k < VARBITLEN(vec); k++)
-		x[i] += (float) (((VARBITS(vec)[i / 8]) >> (7 - (i % 8))) & 0x01);
+		x[k] += (float) (((VARBITS(vec)[k / 8]) >> (7 - (k % 8))) & 0x01);
 }
 
 /*

src/vector.c

@@ -916,19 +916,15 @@ vector_concat(PG_FUNCTION_ARGS)
 	Vector	   *b = PG_GETARG_VECTOR_P(1);
 	Vector	   *result;
 	int			dim = a->dim + b->dim;
-	int			dim_a = a->dim;
-	int			dim_b = b->dim;
 
 	CheckDim(dim);
 	result = InitVector(dim);
 
-	/* Auto-vectorized */
+	for (int i = 0; i < a->dim; i++)
-	for (int i = 0; i < dim_a; i++)
 		result->x[i] = a->x[i];
 
-	/* Auto-vectorized */
+	for (int i = 0; i < b->dim; i++)
-	for (int i = 0; i < dim_b; i++)
+		result->x[i + a->dim] = b->x[i];
-		result->x[i + dim_a] = b->x[i];
 
 	PG_RETURN_POINTER(result);
 }