Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | Merge pull request #15168 from vespa-engine/arnej/add-generic-lambda | Arne H Juul | 2020-11-05 | 6 | -9/+211 |
|\ | | | | | Arnej/add generic lambda | ||||
| * | rename method counting typed nodes | Arne Juul | 2020-11-05 | 1 | -3/+3 |
| | | |||||
| * | fix lambda traversing and extend test | Arne Juul | 2020-11-04 | 2 | -4/+18 |
| | | |||||
| * | also add test of type exporting for inner lambdas | Arne Juul | 2020-11-04 | 1 | -0/+12 |
| | | |||||
| * | use stash.create_uninitialized_array | Arne Juul | 2020-11-04 | 1 | -1/+1 |
| | | |||||
| * | use as<TensorLambda> | Arne Juul | 2020-11-03 | 1 | -1/+1 |
| | | |||||
| * | GC unused include | Arne Juul | 2020-11-03 | 1 | -1/+0 |
| | | |||||
| * | test with simple factory also | Arne Juul | 2020-11-03 | 1 | -0/+5 |
| | | |||||
| * | export nodes belonging to inner lambdas | Arne Juul | 2020-11-03 | 1 | -1/+7 |
| | | |||||
| * | add GenericLambda instruction | Arne Juul | 2020-11-03 | 4 | -8/+174 |
| | | |||||
* | | Merge pull request #15171 from vespa-engine/arnej/inline-sparse-no-overlap | Arne H Juul | 2020-11-04 | 4 | -49/+179 |
|\ \ | | | | | | | Arnej/inline sparse no overlap | ||||
| * | | just loop over subspaces | Arne Juul | 2020-11-04 | 1 | -32/+28 |
| | | | |||||
| * | | move detect_type to its own header file | Arne Juul | 2020-11-04 | 3 | -83/+77 |
| | | | |||||
| * | | make a specific inline version of sparse no-overlap join | Arne Juul | 2020-11-04 | 2 | -8/+148 |
| | | | |||||
* | | | Merge pull request #15169 from vespa-engine/arnej/use-output-cell-type-in-reduce | Arne H Juul | 2020-11-04 | 1 | -3/+3 |
|\ \ \ | |/ / |/| | | template aggregator on output cell type | ||||
| * | | template aggregator on output cell type | Arne Juul | 2020-11-03 | 1 | -3/+3 |
| |/ | |||||
* | | Merge pull request #15162 from vespa-engine/havardpe/optimize-tensor-function | Arne H Juul | 2020-11-03 | 10 | -70/+153 |
|\ \ | |/ |/| | untangle factory-based optimization pipeline from DefaultTensorEngine | ||||
| * | untangle factory-based optimization pipeline from DefaultTensorEngine | Håvard Pettersen | 2020-11-03 | 10 | -70/+153 |
| | | |||||
* | | Merge pull request #15133 from vespa-engine/arnej/handle-hash-collision | Arne H Juul | 2020-11-03 | 4 | -25/+111 |
|\ \ | | | | | | | Arnej/handle hash collision | ||||
| * | | split tests | Arne Juul | 2020-11-03 | 1 | -21/+27 |
| | | | |||||
| * | | consistent add_mapping | Arne Juul | 2020-11-03 | 2 | -30/+34 |
| | | | | | | | | | | | | | | | | | | | | | | | | | | | * rename lookup_or_add_mapping to just add_mapping * change the other add_mapping to have same semantics * in FastValueIndex, check the result from add_mapping each place * use __builtin_expect to tell both the reader and the compiler what the expected path is | ||||
| * | | disallow duplicate add_subspace for PackedMixedTensorBuilder | Arne Juul | 2020-11-02 | 1 | -4/+2 |
| | | | |||||
| * | | test add_subspace robustness for FastValue | Arne Juul | 2020-11-02 | 1 | -0/+55 |
| | | | |||||
| * | | handle add_subspace with same address (or hash collision) | Arne Juul | 2020-11-02 | 2 | -2/+25 |
| | | | |||||
* | | | Merge pull request #15132 from vespa-engine/arnej/add-generic-peek-instruction | Arne H Juul | 2020-11-03 | 7 | -1/+646 |
|\ \ \ | |_|/ |/| | | Arnej/add generic peek instruction | ||||
| * | | refactor, simplify, and cleanup | Arne Juul | 2020-11-03 | 1 | -67/+55 |
| | | | |||||
| * | | move add of input_offset to execute startup | Arne Juul | 2020-11-02 | 1 | -2/+2 |
| | | | |||||
| * | | also verify peek wiring in TensorFunction | Arne Juul | 2020-11-02 | 2 | -0/+35 |
| | | | |||||
| * | | add GenericPeek instruction | Arne Juul | 2020-11-02 | 7 | -1/+623 |
| |/ | |||||
* | | Merge pull request #15119 from vespa-engine/havardpe/simple-median-aggr | Håvard Pettersen | 2020-11-02 | 7 | -20/+98 |
|\ \ | |/ |/| | added simple median aggregator | ||||
| * | median gives NaN if there are any NaN values | Håvard Pettersen | 2020-10-31 | 2 | -20/+18 |
| | | |||||
| * | re-write to use nth_element and max_element | Håvard Pettersen | 2020-10-30 | 1 | -5/+7 |
| | | |||||
| * | ignore NaN values | Håvard Pettersen | 2020-10-30 | 2 | -3/+29 |
| | | |||||
| * | added simple median aggregator | Håvard Pettersen | 2020-10-30 | 7 | -20/+72 |
| | | |||||
* | | Merge pull request #15120 from vespa-engine/havardpe/avoid-unrolling-reduce-all | Arne H Juul | 2020-10-30 | 1 | -15/+5 |
|\ \ | | | | | | | avoid unrolling reduce all loop | ||||
| * | | avoid unrolling reduce all loop | Håvard Pettersen | 2020-10-30 | 1 | -15/+5 |
| |/ | | | | | | | | | | | This is to see if it help the number stability of the BERT performance test. Note that unrolling made it go 4 times fatser, so we might want to re-introduce it again later. | ||||
* / | Add explicit instantiation declaration for ScalarValue<double> and | Tor Egge | 2020-10-30 | 1 | -0/+3 |
|/ | | | | ScalarValue<float>. | ||||
* | use generic scalar values for trivial cases | Håvard Pettersen | 2020-10-28 | 3 | -23/+16 |
| | |||||
* | use specific scalar builder for fast value factory | Håvard Pettersen | 2020-10-28 | 3 | -4/+16 |
| | |||||
* | DoubleValue is now ScalarValue<double> | Håvard Pettersen | 2020-10-28 | 4 | -19/+27 |
| | |||||
* | enable talking about float scalars | Håvard Pettersen | 2020-10-28 | 4 | -34/+65 |
| | | | | | | | - error as flag (future: move out of class) - eliminate redundant type enum - preserve no free float behavior - to/from spec now supports 'float' | ||||
* | Merge pull request #15047 from vespa-engine/arnej/add-generic-create | Håvard Pettersen | 2020-10-28 | 7 | -1/+274 |
|\ | | | | | Arnej/add generic create | ||||
| * | avoid potential ambiguity in constructor selection | Arne Juul | 2020-10-28 | 1 | -1/+1 |
| | | |||||
| * | use ArrayArrayMap | Arne Juul | 2020-10-27 | 1 | -26/+27 |
| | | |||||
| * | use SimpleValueBuilderFactory in reference fill | Arne Juul | 2020-10-27 | 1 | -1/+1 |
| | | |||||
| * | just do all the layouts from map test | Arne Juul | 2020-10-27 | 1 | -4/+8 |
| | | |||||
| * | do it like new value codec | Arne Juul | 2020-10-27 | 1 | -12/+17 |
| | | |||||
| * | switch param order | Arne Juul | 2020-10-27 | 4 | -12/+12 |
| | | |||||
| * | unit test GenericCreate | Arne Juul | 2020-10-27 | 3 | -0/+107 |
| | | |||||
| * | take a more generic spec in GenericCreate::make_instruction | Arne Juul | 2020-10-27 | 3 | -16/+23 |
| | |