Version bump to 0.1.2 [skip ci]

This reverts commit 0aa39a8c0c.

.github/workflows/build.yml

@@ -3,20 +3,37 @@ on: [push, pull_request]
 jobs:
   build:
     if: "!contains(github.event.head_commit.message, '[skip ci]')"
-    runs-on: ubuntu-latest
+    runs-on: ${{ matrix.os }}
     strategy:
       fail-fast: false
       matrix:
+        os: [ubuntu-latest]
         postgres: [13, 12, 11, 10, 9.6]
+        include:
+        - os: macos-latest
+          postgres: 13
     steps:
     - uses: actions/checkout@v2
     - uses: ankane/setup-postgres@v1
       with:
         postgres-version: ${{ matrix.postgres }}
-    - run: sudo apt-get update && sudo apt-get install postgresql-server-dev-${{ matrix.postgres }} libipc-run-perl
+    - if: ${{ startsWith(matrix.os, 'ubuntu') }}
+      run: sudo apt-get update && sudo apt-get install postgresql-server-dev-${{ matrix.postgres }} libipc-run-perl
     - run: make
-    - run: |
+    - if: ${{ startsWith(matrix.os, 'ubuntu') }}
+      run: |
         export PG_CONFIG=`which pg_config`
         sudo --preserve-env=PG_CONFIG make install
+    - if: ${{ startsWith(matrix.os, 'macos') }}
+      run: make install
     - run: make installcheck
-    - run: make prove_installcheck
+    - if: ${{ failure() }}
+      run: cat regression.diffs
+    - if: ${{ startsWith(matrix.os, 'ubuntu') }}
+      run: make prove_installcheck
+    - if: ${{ startsWith(matrix.os, 'macos') }}
+      run: |
+        brew install cpanm && cpanm IPC::Run
+        wget -q https://github.com/postgres/postgres/archive/refs/tags/REL_13_2.tar.gz
+        tar xf REL_13_2.tar.gz
+        make prove_installcheck PROVE=prove PERL5LIB=postgres-REL_13_2/src/test/perl

CHANGELOG.md

@@ -1,3 +1,8 @@
+## 0.1.2 (2021-04-26)
+
+- Vectorized distance calculations
+- Improved cost estimation
+
 ## 0.1.1 (2021-04-25)
 
 - Added binary representation for `COPY`

Makefile

@@ -7,6 +7,17 @@ TESTS = $(wildcard sql/*.sql)
 
 REGRESS = btree cast copy functions ivfflat_cosine ivfflat_ip ivfflat_l2 ivfflat_unlogged vector
 
+# 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
+PG_CFLAGS = -march=native -ftree-vectorize -fassociative-math -fno-signed-zeros -fno-trapping-math
+
+# Debug GCC auto-vectorization
+# PG_CFLAGS += -fopt-info-vec
+
+# Debug Clang auto-vectorization
+# PG_CFLAGS += -Rpass=loop-vectorize -Rpass-analysis=loop-vectorize
+
 PG_CONFIG ?= pg_config
 PGXS := $(shell $(PG_CONFIG) --pgxs)
 include $(PGXS)

README.md

@@ -17,7 +17,7 @@ Supports L2 distance, inner product, and cosine distance
 Compile and install the extension (supports Postgres 9.6+)
 
 ```sh
-git clone --branch v0.1.1 https://github.com/ankane/pgvector.git
+git clone --branch v0.1.2 https://github.com/ankane/pgvector.git
 cd pgvector
 make
 make install # may need sudo
@@ -140,7 +140,7 @@ Libraries that use pgvector:
 
 ### 0.1.1
 
-Follow the [installation instructions](#installation) and run:
+Compile and install the latest version and run:
 
 ```sql
 ALTER EXTENSION vector UPDATE TO '0.1.1';

src/ivfflat.c

@@ -29,8 +29,6 @@ _PG_init(void)
 
 /*
  * Estimate the cost of an index scan
- *
- * TODO Improve cost estimation
  */
 static void
 ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
@@ -42,6 +40,9 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 )
 {
 	GenericCosts costs;
+	int			lists;
+	double		ratio;
+	Relation	indexRel;
 
 #if PG_VERSION_NUM < 120000
 	List	   *qinfos;
@@ -57,7 +58,18 @@ ivfflatcostestimate(PlannerInfo *root, IndexPath *path, double loop_count,
 	genericcostestimate(root, path, loop_count, qinfos, &costs);
 #endif
 
-	*indexStartupCost = costs.indexStartupCost;
+	indexRel = index_open(path->indexinfo->indexoid, NoLock);
+	lists = IvfflatGetLists(indexRel);
+	index_close(indexRel, NoLock);
+
+	ratio = ((double) ivfflat_probes) / lists;
+	if (ratio > 1)
+		ratio = 1;
+
+	costs.indexTotalCost *= ratio;
+
+	/* Startup cost and total cost are same */
+	*indexStartupCost = costs.indexTotalCost;
 	*indexTotalCost = costs.indexTotalCost;
 	*indexSelectivity = costs.indexSelectivity;
 	*indexCorrelation = costs.indexCorrelation;

src/vector.c

@@ -244,11 +244,10 @@ vector_recv(PG_FUNCTION_ARGS)
 	int32		typmod = PG_GETARG_INT32(2);
 	Vector	   *result;
 	int16		dim;
-	int16		unused;
 	int			i;
 
 	dim = pq_getmsgint(buf, sizeof(int16));
-	unused = pq_getmsgint(buf, sizeof(int16));
+	pq_getmsgint(buf, sizeof(int16));	/* unused */
 
 	CheckDim(dim);
 	CheckExpectedDim(typmod, dim);

t/001_wal.pl

@@ -21,8 +21,9 @@ sub test_index_replay
 	my $caughtup_query;
 	my $server_version_num = $node_primary->safe_psql("postgres", "SHOW server_version_num");
 	if ($server_version_num >= 100000) {
-		$caughtup_query = "SELECT pg_current_wal_lsn() <= write_lsn FROM pg_stat_replication WHERE application_name = '$applname';";
+		$caughtup_query = "SELECT pg_current_wal_lsn() <= replay_lsn FROM pg_stat_replication WHERE application_name = '$applname';";
 	} else {
+		# TODO figure out why replay location doesn't work
 		$caughtup_query = "SELECT pg_current_xlog_location() <= write_location FROM pg_stat_replication WHERE application_name = '$applname';";
 	}