Commit message (Collapse) | Author | Age | Files | Lines | |
---|---|---|---|---|---|
* | also optimize for float cells | Håvard Pettersen | 2019-07-15 | 21 | -208/+262 |
| | |||||
* | enable hw dot product for float cells | Håvard Pettersen | 2019-07-11 | 3 | -20/+64 |
| | |||||
* | remove concept of dimension 'bound-ness' | Håvard Pettersen | 2019-07-10 | 6 | -80/+13 |
| | | | | clean up tensor attribute code | ||||
* | float cells in attribute | Håvard Pettersen | 2019-07-09 | 1 | -0/+1 |
| | | | | also assert for cell type consistency in dense tensor view | ||||
* | rename dense::apply -> dense::generic_join | Arne Juul | 2019-07-04 | 3 | -13/+13 |
| | |||||
* | add TypedCells and related functionality | Arne Juul | 2019-07-04 | 45 | -821/+692 |
| | | | | | | | | | | | * templated DenseTensor * templated DenseTensorModify * add templated TypedDenseTensorBuilder * remove DirectDenseTensorBuilder * remove unused TensorMapper * add dispatch structs * add unit test for generic dense join * add special handling of reducing all dimensions | ||||
* | ensure input sequences generate data that is exactly representable as float ↵ | Arne Juul | 2019-07-04 | 2 | -41/+56 |
| | | | | values | ||||
* | extend c++ specific conformance test with float cases | Håvard Pettersen | 2019-07-03 | 11 | -48/+95 |
| | |||||
* | Merge pull request #9853 from vespa-engine/havardpe/remove-error-value-constant | Arne H Juul | 2019-07-01 | 5 | -28/+11 |
|\ | | | | | remove error_value constant | ||||
| * | remove error_value constant | Håvard Pettersen | 2019-06-20 | 5 | -28/+11 |
| | | |||||
* | | use common code for left and right step/reset | Arne Juul | 2019-06-25 | 2 | -78/+56 |
| | | |||||
* | | GC unused debug method | Arne Juul | 2019-06-25 | 2 | -23/+0 |
| | | |||||
* | | use common SideDim struct for both left and right dimensions | Arne Juul | 2019-06-25 | 2 | -28/+22 |
| | | |||||
* | | add braces | Arne Juul | 2019-06-25 | 1 | -3/+9 |
| | | |||||
* | | GC unused class | Arne Juul | 2019-06-24 | 6 | -230/+2 |
| | | |||||
* | | use DenseDimensionCombiner class instead | Arne Juul | 2019-06-24 | 3 | -56/+24 |
| | | |||||
* | | add unit test | Arne Juul | 2019-06-24 | 5 | -0/+199 |
| | | |||||
* | | add DenseDimensionCombiner class | Arne Juul | 2019-06-24 | 3 | -0/+246 |
|/ | |||||
* | readjust to a world without ErrorValue | Håvard Pettersen | 2019-06-19 | 3 | -6/+22 |
| | |||||
* | remove ErrorValue from eval library | Håvard Pettersen | 2019-06-18 | 10 | -79/+35 |
| | |||||
* | add float cases to encode/decode conformance test | Håvard Pettersen | 2019-06-12 | 2 | -175/+234 |
| | |||||
* | enable float result type for prod tensor operations | Håvard Pettersen | 2019-06-12 | 4 | -41/+5 |
| | |||||
* | Merge pull request #9729 from vespa-engine/havardpe/float-cell-serialization | Arne H Juul | 2019-06-12 | 9 | -181/+208 |
|\ | | | | | serialize float cells | ||||
| * | serialize float cells | Håvard Pettersen | 2019-06-07 | 9 | -181/+208 |
| | | | | | | | | preserve tensor type across encode/decode | ||||
* | | Add missing includes. | Tor Egge | 2019-06-11 | 1 | -0/+1 |
|/ | |||||
* | remove sparse tensor builder | Håvard Pettersen | 2019-06-06 | 10 | -401/+0 |
| | | | | also remove sparse tensor unsorted address builder | ||||
* | use direct sparse tensor builder | Håvard Pettersen | 2019-06-06 | 15 | -59/+186 |
| | |||||
* | remove dense tensor builder | Håvard Pettersen | 2019-06-06 | 9 | -481/+0 |
| | |||||
* | use direct dense tensor builder | Håvard Pettersen | 2019-06-06 | 7 | -393/+203 |
| | | | | also remove tensor micro-benchmark | ||||
* | Throw instead of silent abort. | Henning Baldersheim | 2019-06-04 | 1 | -1/+1 |
| | |||||
* | remove DefaultTensor type wrapper | Håvard Pettersen | 2019-06-03 | 11 | -94/+49 |
| | | | | | also remove TensorBuilder interface and use SparseTensorBuilder directly | ||||
* | remove tensor factory | Håvard Pettersen | 2019-05-31 | 12 | -660/+170 |
| | |||||
* | preserve cell type during encode decode | Håvard Pettersen | 2019-05-06 | 2 | -93/+113 |
| | |||||
* | disable concat optimization for non-double cells | Håvard Pettersen | 2019-05-03 | 1 | -1/+2 |
| | |||||
* | make sure tensor spec has appropriate type | Håvard Pettersen | 2019-05-03 | 2 | -0/+2 |
| | | | | | adds runtime support for input tensors with float cells in eval fixture used for testing. | ||||
* | disable optimizations for tensors with non-double cells | Håvard Pettersen | 2019-05-03 | 16 | -32/+78 |
| | | | | | | use reference implementation for tensors claiming to have non-double cells to be able to preserve cell type for tensors created from a TensorSpec (specifically lambda tensors with float cells). | ||||
* | add float cases to node type test | Håvard Pettersen | 2019-05-03 | 2 | -4/+21 |
| | | | | also update tensor lambda parsing to support cell type | ||||
* | add concept of float cells in value type | Håvard Pettersen | 2019-05-03 | 6 | -231/+305 |
| | | | | | | | | scalar values are always double preserve cell type when combining tensor with scalar result(float cells, double cells) -> double cells require '()' in tensor type spec clean up and update value type test | ||||
* | avoid abstract value types | Håvard Pettersen | 2019-04-25 | 28 | -756/+247 |
| | | | | | | | | | | | | remove basic value type ANY remove concept of tensors with unknown dimensions disallow value types with unbound indexed dimensions remove predicates talking about abstract types type of unknown values are now ERROR (was ANY) require that overlapping indexed dimensions are of equal size type unification now requires types to be equal ('if' expressions) creating a tensor type without dimensions now gives a double type make rank feature setup fail on invalid types (query/attribute) | ||||
* | support binary formats with cell type in reference implementation | Håvard Pettersen | 2019-04-09 | 3 | -7/+248 |
| | |||||
* | Merge pull request #9029 from ↵ | Henning Baldersheim | 2019-04-08 | 2 | -2/+42 |
|\ | | | | | | | | | vespa-engine/havardpe/parse-and-ignore-tensor-cell-type parse (and ignore for now) tensor cell type in tensor type spec | ||||
| * | Improve handling of malformed cell type specifiers | Håvard Pettersen | 2019-04-08 | 1 | -6/+24 |
| | | | | | | | | | | | | | | | | | | Do not consume any input when encountering a malformed cell type specifier. This avoids parse conflicts between 'less' operators and tensor cell type specifiers in type expressions in the type resolution test which parses partial input into value types. example expression: 'tensor < double' | ||||
| * | parse (and ignore for now) tensor cell type in tensor type spec | Håvard Pettersen | 2019-04-05 | 2 | -2/+24 |
| | | | | | | | | example: tensor<float>(x[10]) | ||||
* | | Use bit number 3 mean 'with cell type' to carry forward the current schema ↵ | Henning Baldersheim | 2019-04-05 | 2 | -5/+7 |
|/ | | | | | | where the bits hav meaning. Update expected result. | ||||
* | Add a method that will extract cells only to a favoured cell type. This is ↵ | Henning Baldersheim | 2019-04-03 | 6 | -11/+72 |
| | | | | temporary until full typed tensor support is in place. | ||||
* | Keep the serialzation of the cells with the serialization for now. | Henning Baldersheim | 2019-04-03 | 7 | -82/+79 |
| | | | | Clean up code for better understanding and reuse. | ||||
* | Instead of specifying type of tensor, specify serialization. | Henning Baldersheim | 2019-04-03 | 13 | -86/+74 |
| | |||||
* | Remove wrappers not neeed. | Henning Baldersheim | 2019-04-02 | 3 | -12/+1 |
| | |||||
* | Remove duplicate typedefs. | Henning Baldersheim | 2019-04-02 | 2 | -12/+8 |
| | |||||
* | Handle dense tensors encoded as float | Henning Baldersheim | 2019-04-02 | 11 | -137/+251 |
| |