summaryrefslogtreecommitdiffstats
path: root/eval
Commit message (Expand)AuthorAgeFilesLines
* Add a for_each method for more efficient iteration.Henning Baldersheim2018-01-032-4/+1
* Let implementation be visible too.Henning Baldersheim2017-12-291-3/+1
* Must see implementation too.Henning Baldersheim2017-12-292-0/+2
* Include implementation file too.Henning Baldersheim2017-12-291-4/+3
* Ensure methods are inlined.Henning Baldersheim2017-12-291-1/+5
* Also copy the zero termination.Henning Baldersheim2017-12-291-1/+1
* Avoid all the plt indirections by declaring dll local and using simpler struc...Henning Baldersheim2017-12-294-14/+20
* Use a prime based hash instead.Henning Baldersheim2017-12-192-6/+2
* Add fully parameterized instantiation macros.Henning Baldersheim2017-12-191-1/+4
* Use 2inN modulator instead of prime number modulator to avoid expensive divHenning Baldersheim2017-12-191-1/+2
* Inline DenseTensorAddressCombiner::combineHenning Baldersheim2017-12-192-59/+35
* Avoid type conversion and also use a presized address scratchpadHenning Baldersheim2017-12-191-7/+9
* DenseTensorCellsIterator::nextHenning Baldersheim2017-12-192-16/+12
* Micro optimize DenseTensorCellsIterator::next by expecting most probable bran...Henning Baldersheim2017-12-1914-52/+46
* Avoid division imposed by moduloHenning Baldersheim2017-12-181-2/+4
* Unify on nested namespace.Henning Baldersheim2017-12-1834-200/+116
* Consider overlapping dimensions when estimating resulting tensor space.Henning Baldersheim2017-12-185-24/+40
* Estimate number of cells needed in the hashmap.Henning Baldersheim2017-12-182-2/+3
* Use nested namespace.Henning Baldersheim2017-12-1814-66/+26
* Use nested namespace.Henning Baldersheim2017-12-183-28/+17
* Reserve space up front.Henning Baldersheim2017-12-181-13/+7
* Add method to allocate space up front.Henning Baldersheim2017-12-181-4/+3
* Only use 32 bits for hash and size.Henning Baldersheim2017-12-182-9/+9
* Add test for proving size of sparse tensor address ref.Henning Baldersheim2017-12-182-1/+6
* Merge pull request #4329 from vespa-engine/arnej/better-test-xw-productHåvard Pettersen2017-12-013-310/+65
|\
| * consolidate testsArne Juul2017-12-011-12/+4
| * simplify castsArne Juul2017-12-012-6/+4
| * improve testArne Juul2017-12-011-295/+60
* | use static cast instead of dynamic castHåvard Pettersen2017-12-011-2/+2
|/
* replace appropriate sub-expressions with dense xw productHåvard Pettersen2017-12-015-32/+152
* implement and test XW productArne Juul2017-11-306-0/+500
* optimize vector concatHåvard Pettersen2017-11-292-0/+43
* optimize for NN bias additionHåvard Pettersen2017-11-291-9/+9
* recognize join mul as part of inner productHåvard Pettersen2017-11-292-70/+160
* Merge pull request #4298 from vespa-engine/havardpe/clean-up-old-internal-ten...Håvard Pettersen2017-11-2913-356/+50
|\
| * internalize tensor operation special handlingHåvard Pettersen2017-11-288-102/+22
| * remove tensor print/toString (use TensorSpec instead)Håvard Pettersen2017-11-288-101/+3
| * remove unused tensor functions (add,sub,min,max)Håvard Pettersen2017-11-287-120/+0
| * getType->type, sum->as_doubleHåvard Pettersen2017-11-2812-34/+26
* | use binary format to convert tensorsHåvard Pettersen2017-11-281-4/+7
|/
* added type resolving test for elu operatorHåvard Pettersen2017-11-221-0/+1
* Add new tensor conformance test with elu'sLester Solbakken2017-11-221-1/+19
* Add Elu test to c++ conformance testLester Solbakken2017-11-221-3/+4
* Add Elu as a backend ranking functionLester Solbakken2017-11-2212-2/+20
* use const references and stash instead of UPHåvard Pettersen2017-11-0915-222/+201
* clean up tensor engine APIHåvard Pettersen2017-11-0744-434/+352
* add tests for set membership with tensorsHåvard Pettersen2017-11-064-1/+37
* remove 'equal' concept for tensorsHåvard Pettersen2017-11-0315-161/+27
* handle 'in' operator as custom (tensor) map operationHåvard Pettersen2017-11-0317-305/+224
* unify conformance test format, refactor code and update specHåvard Pettersen2017-11-017-468/+264