summaryrefslogtreecommitdiffstats
path: root/eval
Commit message (Collapse)AuthorAgeFilesLines
* Revert "Copyright header"Jon Bratseth2017-06-13208-208/+208
|
* Copyright headerJon Bratseth2017-06-13208-208/+208
|
* auto-detect appropriate compiled function parameter lazynessHaavard2017-06-076-0/+89
| | | | ... but still allow config override
* Add support for conditionally running available shutdown.Henning Baldersheim2017-06-071-0/+5
|
* Include stringfmt where it was needed.Henning Baldersheim2017-06-061-0/+1
|
* cleanup a global mutexArne H Juul2017-06-021-0/+1
|
* Reduce fastos.h dependency.Henning Baldersheim2017-06-0148-70/+43
|
* call llvm_shutdownHaavard2017-05-291-0/+3
|
* clone for wrapped tensorsHaavard2017-05-244-10/+10
|
* support tensor mapping for wrapped simple tensorsHaavard2017-05-236-265/+367
|
* wrapped simple tensorHaavard2017-05-1814-142/+312
| | | | | | | | | | | | fall back to wrapping SimpleTensors for cases where the default tensor implementation does not support the data model (mixed dimensions). remove special handling of mixed cases in tensor conformance tests. require that the default tensor implementation passes all conformance tests. ironed out a small mismatch in the tensor binary format tag.
* fixes based on review feedbackHaavard2017-05-111-7/+7
|
* test tensor binary formatHaavard2017-05-1012-7/+862
|
* Merge pull request #2418 from yahoo/havardpe/encode-decode-simple-tensorsGeir Storli2017-05-098-134/+342
|\ | | | | binary encode/decode as part of tensor engine API
| * binary encode/decode as part of tensor engine APIHaavard2017-05-088-134/+342
| |
* | Deinline as compiler makes the best choices.Henning Baldersheim2017-05-094-0/+7
| |
* | Fix remaining inline warnings for compilation with -OgTor Brede Vekterli2017-05-052-0/+4
| |
* | gcc 7 does not provide <functional> by default.Henning Baldersheim2017-05-051-3/+4
| |
* | gcc 7 does not provide <functional> by default.Henning Baldersheim2017-05-051-4/+5
|/
* Merge pull request #2340 from yahoo/havardpe/mixed-tensor-binary-format-specGeir Storli2017-05-031-0/+42
|\ | | | | added description of new mixed serialization format
| * minor fixupHaavard2017-05-031-1/+1
| |
| * account for sparse tensors without dimensionsHaavard2017-05-031-6/+11
| |
| * added description of new mixed serialization formatHaavard2017-05-021-0/+37
| | | | | | | | ... and how it relates to old formats (sparse/dense)
* | Fix warnings hidden earlier due to including application headers as system ↵Henning Baldersheim2017-05-0211-17/+40
|/ | | | includes
* Revert "Balder/enforce override 2"Arne H Juul2017-04-241-4/+4
|
* Use overrideHenning Baldersheim2017-04-231-4/+4
|
* add override in eval moduleArne H Juul2017-04-101-2/+3
|
* OverrideHenning Baldersheim2017-04-0411-355/+350
|
* change code to work with LLVM 3.9.1Haavard2017-03-287-73/+96
|
* Move non-inlineable eval test functions into cpp fileTor Brede Vekterli2017-03-132-37/+51
|
* Create explicit, non implicitly inlined function definitionsTor Brede Vekterli2017-03-132-0/+3
|
* Deinline destructors/constructorsHenning Baldersheim2017-03-083-11/+23
|
* deinline large destructorsHenning Baldersheim2017-03-082-7/+13
|
* Specify destructors explicit to control inlining.Henning Baldersheim2017-03-0816-93/+170
| | | | | Conflicts: build_settings.cmake
* more specific initial valueHaavard2017-03-021-1/+1
|
* analyze parameter usageHaavard2017-03-026-1/+210
|
* interpreted functions now always take lazy parametersHaavard2017-02-278-74/+102
|
* also log lazy time/factor in explore benchmarkHaavard2017-02-211-12/+28
|
* support gbdt forest optimizations with lazy parametersHaavard2017-02-217-61/+196
|
* improve and extend micro-benchmarkHaavard2017-02-205-85/+156
|
* added support for lazy parameters to compiled functionsHaavard2017-02-179-22/+54
|
* convert result from map/apply to DoubleValue if type is doubleHaavard2017-02-161-1/+8
|
* let tensor::Tensor::getType return const referenceHaavard2017-02-165-5/+5
| | | | | ... to make it cheaper to check tensor types in implementation-specific code.
* use simple tensor engine as (expensive) fallbackHaavard2017-02-162-21/+39
| | | | ... for new immediate API in default tensor engine
* extend tensor conformance test with new syntaxHaavard2017-02-101-42/+78
|
* add more reduce cases and drop tensor-only eval in double contextHaavard2017-02-101-5/+7
|
* enable converting simple tensors to errorsHaavard2017-02-101-17/+11
|
* allow simple tensors to represent error valuesHaavard2017-02-102-0/+20
|
* wire in immediate evaluation of new syntaxHaavard2017-02-109-73/+109
|
* add support for loading lz4 compressed json tensorsHaavard2017-02-095-10/+39
|