Commit Graph

762 Commits

Author SHA1 Message Date
Andrew Kane
958af80e96 Added more functions [skip ci] 2024-09-23 17:39:44 -07:00
Andrew Kane
fd65bcfb10 Added indexing [skip ci] 2024-09-23 17:01:32 -07:00
Andrew Kane
274e6544d4 Added L2 distance [skip ci] 2024-09-23 16:52:07 -07:00
Andrew Kane
035a31ac91 Added minivec type 2024-09-23 16:45:19 -07:00
Andrew Kane
87ac108bf7 Removed code for Postgres 12 [skip ci] 2024-09-23 15:26:31 -07:00
Andrew Kane
97cf990e0f Free TupleDesc [skip ci] 2024-09-21 19:15:34 -07:00
Andrew Kane
55dc735e1a Moved allocations out of GetScanItems [skip ci] 2024-09-21 19:10:25 -07:00
Andrew Kane
be4e9a9df2 Added macros for IvfflatScanList [skip ci] 2024-09-21 18:10:37 -07:00
Andrew Kane
d5e8fc96a5 Changed HnswPairingHeapNode to HnswSearchCandidate to reduce allocations and improve code 2024-09-21 12:07:44 -07:00
Andrew Kane
6d2af6d3f9 Improved code [skip ci] 2024-09-20 15:21:57 -07:00
Andrew Kane
a6ab5d07c0 Fixed CI 2024-09-19 20:50:51 -07:00
Andrew Kane
aa77346103 Improved code [skip ci] 2024-09-19 19:57:16 -07:00
Andrew Kane
b0da2d95d9 Fixed array_to_sparsevec on Windows [skip ci] 2024-09-19 19:52:16 -07:00
Andrew Kane
3fb05eb847 Added casts for arrays to sparsevec - #604
Co-authored-by: Narek Galstyan <narekg@berkeley.edu>
Co-authored-by: Di Qi <di@lantern.dev>
2024-09-19 19:17:05 -07:00
Andrew Kane
b738ffecc1 Dropped support for Postgres 12 2024-09-19 18:13:54 -07:00
Heikki Linnakangas
7117513532 Add error codes to a few errors (#657)
With elog(), you get XX000 "internal_error", which sounds scary.

It's not self-evident what the right error codes for some of these
errors are, but I tried to use my best judgment.
2024-09-19 18:04:23 -07:00
Jonathan S. Katz
05fb382031 Swap max costing values to align with upstream guidance (#658)
A feature targeted for PostgreSQL 18 (postgres/postgres@e2225346)
that makes optimizations around disabled path nodes impacted pgvector
such that PostgreSQL would choose to perform an index scan when it
should have used a different scan (e.g. `SELECT count(*) FROM table`).
Per upstream guidance[1], the recommendation is to switch to using
`get_float8_infinity()`, which achieves the same behavior in backbranches,
and can be adapated to work with the new behavior introduced in PostgreSQL 18.

[1] https://www.postgresql.org/message-id/2281822.1724441531%40sss.pgh.pa.us
2024-09-19 18:01:59 -07:00
Andrew Kane
8e1853fbf3 Improved variable name [skip ci] 2024-09-19 15:09:40 -07:00
Andrew Kane
f9d68a061a Simplified HnswLoadUnvisitedFromMemory [skip ci] 2024-09-19 04:39:46 -07:00
Andrew Kane
4f8ab574c9 Simplified CountElement [skip ci] 2024-09-19 04:32:38 -07:00
Andrew Kane
a15806196e Keep scan-build happy 2024-09-19 04:02:09 -07:00
Andrew Kane
5c9429a0f8 Reduced memory usage for HNSW index scans 2024-09-19 03:27:35 -07:00
Andrew Kane
16ca608f42 Updated AddToVisited to use HnswElementPtr 2024-09-19 02:41:20 -07:00
Andrew Kane
8dde14a736 Reduced memory usage for HNSW index scans
Co-authored-by: Heikki Linnakangas <heikki.linnakangas@iki.fi>
2024-09-19 02:17:51 -07:00
Andrew Kane
d74d3065bc Reduced allocations for pairing heap 2024-09-19 01:59:46 -07:00
Andrew Kane
4af5a127e0 Revert "Improved cleanup for IVFFlat index scans [skip ci]"
This reverts commit da7d3959a3.
2024-09-02 01:52:28 -07:00
Andrew Kane
e47984e616 Reset tuple sort for Postgres 12 [skip ci] 2024-08-24 22:10:26 -07:00
Andrew Kane
da7d3959a3 Improved cleanup for IVFFlat index scans [skip ci] 2024-08-24 21:59:44 -07:00
Andrew Kane
dadbbc3758 Renamed InitSortState to InitScanSortState [skip ci] 2024-08-24 21:53:15 -07:00
Andrew Kane
6af0a43d62 Added InitBuildSortState function [skip ci] 2024-08-24 21:50:31 -07:00
Andrew Kane
ffcb90d094 Added InitSortState function [skip ci] 2024-08-24 21:42:18 -07:00
Andrew Kane
8a312c3c8e Added memory usage for IVFFlat index scans [skip ci] 2024-08-24 21:30:40 -07:00
Andrew Kane
5d86b177ab Fixed -DIVFFLAT_MEMORY [skip ci] 2024-08-24 20:56:33 -07:00
Andrew Kane
ea99957fae Added fields to IndexAmRoutine 2024-08-22 20:39:16 -07:00
Samuel Marks
4cede1a9c9 [src/hnswutils.c] Resolve 1 -Wmaybe-uninitialized (#654) 2024-08-22 19:51:16 -07:00
Andrew Kane
bb855e6cb4 Updated comment [skip ci] 2024-08-06 10:35:26 -07:00
Andrew Kane
57fb706242 Fixed locking for parallel HNSW index builds - fixes #635 2024-08-04 11:27:51 -07:00
Andrew Kane
6c692ef23f Fixed compilation error with GCC 14 on i386 when SSE2 is not enabled 2024-07-27 06:20:27 -07:00
Andrew Kane
06d1fa1402 Added alignment check to ensure consistency with PageIndexTupleOverwrite 2024-07-19 15:50:24 -07:00
Andrew Kane
8c5a4bfb6c Fixed failed to add index item error with sparsevec - fixes #625 2024-07-19 13:54:36 -07:00
Andrew Kane
8772c8de68 Fixed compilation error with FreeBSD ARM 2024-06-30 11:23:39 -07:00
Andrew Kane
61870a0244 Fixed compilation warning with MSVC and Postgres 16 - fixes #598
Co-authored-by: Xing Guo <higuoxing@gmail.com>
2024-06-16 12:09:01 -07:00
Andrew Kane
ad7cad5ecd Improved HnswSearchLayer code 2024-06-11 16:29:14 -07:00
Andrew Kane
2a8b9d689e Moved check 2024-06-11 15:45:03 -07:00
Andrew Kane
18cd8a60c3 Updated comment [skip ci] 2024-06-10 22:02:40 -07:00
Andrew Kane
8c91a9f56a Fixed initialization fork for IVFFlat indexes on unlogged tables - #591 2024-06-10 21:55:17 -07:00
Andrew Kane
9e91af5989 Added checks for invalid indexes - #591 2024-06-10 21:20:54 -07:00
Narek Galstyan
9dcf1bdc80 Fix init_fork WAL-logging on unlogged indexes (#591)
Currently pgvector does not create any WAL records for unlogged tables

Postgres assumes INIT_FORK of unlogged tables is persistent and uses it
to reset the table index to its default empty state after a server
crash.

This patch makes INIT_FORK of unlogged table WAL-tracked, which ensures
an unlogged table is usable after a crash-restart
2024-06-10 21:16:32 -07:00
Andrew Kane
49c1f13095 Improved performance of on-disk HNSW index builds - #570 2024-05-29 12:03:58 -07:00
Andrew Kane
69f49290fb Fixed compilation warning with Clang < 14 - closes #546 2024-05-07 20:53:41 -07:00