summaryrefslogtreecommitdiffstats
path: root/searchlib
Commit message (Collapse)AuthorAgeFilesLines
* Rename AddressSpaceUsage::add() -> set() to better reflect intent.Geir Storli2021-07-155-8/+8
|
* Make static strings const.Geir Storli2021-07-152-4/+4
|
* Generalize how address space usage is reported from attribute vectors.Geir Storli2021-07-1513-68/+134
| | | | | This prepares for reporting address space usage for more attribute vector components, not only enum store and multi-value mapping.
* Remove unneeded local using statement.Tor Egge2021-06-291-2/+0
|
* GC unused members from SearchReplyHenning Baldersheim2021-06-286-55/+17
|
* GC unused 'partitionid' and various other members of MonitorReply that are ↵Henning Baldersheim2021-06-269-53/+9
| | | | long gone.
* Various cleanup after reading indexing threading codeHenning Baldersheim2021-06-243-29/+20
|
* dry run onnx models on setupHåvard Pettersen2021-06-237-7/+125
|
* Remove mlock as option for backend tooHenning Baldersheim2021-06-212-2/+1
|
* Remove unneccesary importLester Solbakken2021-06-211-1/+0
|
* Fix bit function typoLester Solbakken2021-06-211-1/+1
|
* Add bit ranking function in JavaLester Solbakken2021-06-215-3/+28
|
* also test tensorFromLabels with single-value integer attributeArne Juul2021-06-181-0/+18
|
* allow tensorFromLabels with collection type "single"Arne Juul2021-06-181-5/+9
|
* - NULL -> nullptrHenning Baldersheim2021-06-172-36/+20
| | | | | - Remove virtual on override. - minor style changes.
* Various cleanup when reading grouping code.Henning Baldersheim2021-06-162-21/+21
|
* IndexSet -> IndexList and unify on 'using' directives.Henning Baldersheim2021-06-153-58/+54
|
* Use a list instead of a set to make building faster.Henning Baldersheim2021-06-146-38/+36
| | | | Then sort and uniq before applying the list.
* added onnx model cacheHåvard Pettersen2021-06-112-7/+16
|
* Commit before populate.Henning Baldersheim2021-06-111-1/+1
|
* Expose bug reletate to initialization of most costly features K computation.Henning Baldersheim2021-06-116-41/+74
|
* More code unification, no sematic changes.Henning Baldersheim2021-06-102-31/+38
|
* Clean up code layout and GC unused members. No semtic changes.Henning Baldersheim2021-06-0925-204/+239
|
* Unify how property names are generated and decomposed.Henning Baldersheim2021-06-093-3/+31
|
* Use even larger buffers for link arrays (and level arrays).Tor Egge2021-06-071-2/+2
|
* Update _bvs when compacting buffers containing bitvector entries.Tor Egge2021-06-071-1/+4
|
* Merge pull request #18134 from ↵Henning Baldersheim2021-06-041-1/+1
|\ | | | | | | | | vespa-engine/toregge/use-larger-buffers-for-link-arrays Use larger buffers for link arrays (and level arrays).
| * Use larger buffers for link arrays (and level arrays).Tor Egge2021-06-041-1/+1
| |
* | Move roundUp2inN<T> to vespalib.Henning Baldersheim2021-06-042-7/+4
| |
* | If there are hickups in the feed chain preventing normal commit rate, we ↵Henning Baldersheim2021-06-043-16/+63
| | | | | | | | | | | | need to prevent lasting huge temporary buffers. So cap buffers and start from scratch if peaks are detected.
* | Avoid tracking information to keep docids close when that is not necessary.Henning Baldersheim2021-06-0316-135/+120
|/ | | | | Just append on add, and then do an explicit sort the one place that requires operations on a doc to be ordered.
* Use 32M which is the same limit used in some other query related places.Henning Baldersheim2021-06-031-1/+1
|
* Unify on mmap limit at 16M for bitvectors. Should ideally be controlled with ↵Henning Baldersheim2021-06-022-3/+7
| | | | allocator from the outside depending on expected BV lifetime.
* Merge pull request #18093 from ↵Håvard Pettersen2021-06-022-17/+15
|\ | | | | | | | | vespa-engine/havardpe/explicit-rank-expression-name use explicit expression name
| * use explicit expression nameHåvard Pettersen2021-06-022-17/+15
| |
* | Distributing the external rankfiles was a dead end due to th eprocessing ↵Henning Baldersheim2021-06-023-7/+2
|/ | | | | | taking place. Reverting back to get back to the drawing table. It must be done post deriving, based on size of expression.
* Reenable tensor attribute compaction test.Tor Egge2021-06-011-8/+10
|
* Only use tensor store memory usage when considering compaction.Tor Egge2021-06-014-5/+28
|
* Update abi and deprecate unused method.Henning Baldersheim2021-05-314-15/+22
|
* Use inheritance information from the uncompiled rankprofile to sort out whenHenning Baldersheim2021-05-314-6/+18
| | | | to use extrenal files, and when they are overridden.
* Update searchlib/src/vespa/searchlib/tensor/dense_tensor_attribute.cpp Tor Egge2021-05-271-1/+1
| | | | | Apply suggested change. Co-authored-by: Geir Storli <geirst@verizonmedia.com>
* Use modulo instead of bitwise and.Tor Egge2021-05-271-1/+1
|
* Load of tensor attribute with nearest neighbor index enabled but noTor Egge2021-05-271-0/+3
| | | | | saved index will index documents during the load. Add occasional calls to commit() to prevent excessive growth of hold lists.
* external ranking expressionsHåvard Pettersen2021-05-205-7/+53
| | | | loaded from potentially compressed files
* Revert "external ranking expressions"Harald Musum2021-05-205-53/+7
|
* Merge pull request #17885 from ↵Håvard Pettersen2021-05-205-7/+53
|\ | | | | | | | | vespa-engine/havardpe/external-ranking-expressions external ranking expressions
| * external ranking expressionsHåvard Pettersen2021-05-185-7/+53
| | | | | | | | loaded from potentially compressed files
* | Use make_uniqueHenning Baldersheim2021-05-181-1/+1
|/
* support functional constantsHåvard Pettersen2021-05-113-10/+73
| | | | | | | | | | | | be more specific about const number values (GDBT trees): is_const -> is_const_double get_const_value -> get_const_double_value Add more generic 'get_const_value' that can also be used for tensor values. Allow it to be called even in the case of parse error, in which case it does not produce a value (same as for non-const sub-expressions).
* Log as debugBjørn Christian Seime2021-05-071-1/+1
|