summaryrefslogtreecommitdiffstats
path: root/searchcore/src/tests/proton/matching
Commit message (Collapse)AuthorAgeFilesLines
* extend unit testArne H Juul2021-10-191-0/+24
|
* use custom types for true/false (NodeTypes::True, ProtonTrue, etc)Arne H Juul2021-10-191-2/+2
|
* add ITEM_TRUE and ITEM_FALSE query nodesArne H Juul2021-10-181-0/+2
|
* Update 2018 copyright notices.gjoranv2021-10-073-3/+3
|
* Update 2017 copyright notices.gjoranv2021-10-0721-21/+21
|
* external ranking expressionsHåvard Pettersen2021-05-205-2/+34
| | | | loaded from potentially compressed files
* Revert "external ranking expressions"Harald Musum2021-05-205-34/+2
|
* external ranking expressionsHåvard Pettersen2021-05-185-2/+34
| | | | loaded from potentially compressed files
* - Use a LookupKey to further delay data conversion until we know it is ↵Henning Baldersheim2021-03-291-0/+11
| | | | | | necessary. - GC unused code
* - Let DotProduct,Wand and WeightedSet be Term nodes in the query tree as ↵Henning Baldersheim2021-03-293-46/+19
| | | | | | | | they really are. That restricts the nodes to what they can really do and makes them significantly cheaper. - In addition type conversion of numeric terms is delayed to when it is necessary. And as next step they can be avoided completely.
* add 'is_search' flag when creating match tools factoryHåvard Pettersen2021-03-251-2/+2
| | | | drop global filter handling for non-search setups
* all scalars must be doubleHåvard Pettersen2021-03-041-1/+1
| | | | and all operation results must be at least float
* Avoid including singlenumericattribute.hppHenning Baldersheim2021-02-251-1/+0
|
* use size literals in searchcoreArne Juul2021-02-152-2/+4
|
* When making callbacks like notifying about new bucket show the gurad that is ↵Henning Baldersheim2021-02-141-1/+2
| | | | | | held. Then this guard can used instead of possibly making a deadlock if trying to take it yourself.
* Use vespalib::duration for timeoutsHenning Baldersheim2021-01-311-2/+1
|
* use SimpleValue instead of EngineOrFactoryArne Juul2020-12-022-11/+12
|
* cleanup unused includeArne Juul2020-11-272-2/+0
|
* balance second phase ranking workloadHåvard Pettersen2020-11-022-96/+117
| | | | | | ... by first giving each thread the same number of results to re-rank and then exchanging results back to the threads finding them during matching/first phase ranking.
* handle both engine- and factory-based tensorsArne Juul2020-10-152-7/+11
| | | | | | * use EngineOrFactory::get() instead of DefaultTensorEngine::ref() * avoid direct use of DenseTensorView etc where possible * use eval::Value instead of tensor::Tensor where possible
* Add noexcept as indicated by -WnoeexceptHenning Baldersheim2020-10-071-1/+1
|
* Reuse document meta store state from prepare step instead of doingTor Egge2020-10-061-1/+1
| | | | a new lookup in btree mapping from gid to lid during live feed.
* handle onnx model config for inputs and outputsHåvard Pettersen2020-09-221-6/+19
|
* handle onnx models configHåvard Pettersen2020-09-012-2/+15
|
* wire in using new JSON-lite location formatArne Juul2020-08-181-7/+11
|
* allow extracting locations as termsArne Juul2020-08-162-5/+5
|
* finish rewrite to use GeoLocationArne Juul2020-07-151-1/+1
|
* preserve old behaviorArne Juul2020-07-151-1/+2
|
* Consolidate search for GTest.Tor Egge2020-07-073-3/+0
|
* Use find_package to find gtest library.Tor Egge2020-06-293-3/+6
|
* Add a rank feature that will use the distribution key to provide a globally ↵Henning Baldersheim2020-06-221-1/+6
| | | | | | unique number. unique = (docId << 16 | distributionKey)
* Wire in nearest neighbor brute force limit.Tor Egge2020-06-191-1/+3
|
* use vector of UP as MultiSearch::ChildrenArne Juul2020-06-041-14/+17
| | | | | | * add helper class for constructing MultiSearch::Children (mostly for unit tests) * rewrite as needed to adapt
* Prepare to support matched-elements-only for arrays and weighted sets of ↵Geir Storli2020-05-201-10/+10
| | | | | | | primitive types. This renames StructFieldMapper to MatchingElementsFields with added support for basic fields. Also some terminology is aligned.
* - Handle more than 64k hits in the element vector.Henning Baldersheim2020-05-181-6/+6
| | | | | - Avoid computing all vectors in full separately and instead do an incremental inline merge with. - Also avoid requiring the searchiterator aspect on the wrappers.
* Extend semantic lifetime of view resolver variables in searchcoreTor Egge2020-04-221-6/+9
| | | | query test.
* Use enable_shared_from_this to simplify interface and keep the details local.Henning Baldersheim2020-02-171-1/+1
|
* Added ↵Ola Aunrønning2020-02-071-4/+22
| | | | content.proton.documentdb.matching.rank_profile.query_setup_time.average to default public metric set. Replaces query_collateral_time
* c_str -> dataHenning Baldersheim2020-01-212-2/+2
|
* Revert "Revert "Balder/add executeinfo""Henning Baldersheim2020-01-033-10/+12
|
* Revert "Balder/add executeinfo"Arnstein Ressem2020-01-033-12/+10
|
* Merge pull request #11619 from vespa-engine/balder/add-executeinfoHenning Baldersheim2020-01-023-10/+12
|\ | | | | Balder/add executeinfo
| * Add ExecutionInfo to propagate more information than just strictness to ↵Henning Baldersheim2019-12-313-10/+12
| | | | | | | | | | | | | | fetchPostings. The estimated hitrate that each iterator will will enable smarter query evaluation. Especially will it be possible to figure out if static up front cost can be avoided.
* | Remove unused constant (term_index).Tor Egge2020-01-021-1/+0
|/
* Add test for 1% capping.Henning Baldersheim2019-12-201-0/+10
|
* Revert "Revert "Balder/remove steadytimestamp 2""Henning Baldersheim2019-12-165-35/+34
|
* Revert "Balder/remove steadytimestamp 2"Henning Baldersheim2019-12-165-34/+35
|
* Use std::chrono.Henning Baldersheim2019-12-095-35/+34
|
* Make one doom with explicit hard and soft instead of wrapping and nesting.Henning Baldersheim2019-12-041-1/+1
|
* Wrap soft and hard doom in an CombinedDoom.Henning Baldersheim2019-12-031-7/+7
| | | | If softdoom factor has been overridden, we shall not adjust the automatic one.