Age | Commit message (Collapse) | Author | Files | Lines |
|
* Null-terminating check for backend conf api
- Add null terminaing function
- Add check routine
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Give name to arg max length in backend conf
- Add macro for max length
- 32 for backend name, 64 for op name
* fix format
* add null check
* add null check to nnfw_set_op_backend
* Check null inside nnfw_session
|
|
This will introduce get_node_shape helper in copy_shapedata of FixShape
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
This will update FixShape to use correct copy_shapedata() for binary input nodes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
No compiler module depends on "moco" module.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
|
|
Let's declare the visibility of dependent libraries explicitly.
This change allows us to introduce PRIVATE dependencies.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
|
|
"const" qualifier on function return type has no effect.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
|
|
* [logo] Simplifying `BiasEncode - BiasDecode` subgraph
With this commit, logo will Simplifying `BiasEncode - BiasDecode` subgraph into loco::Forward
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* make code shorter
* remove 'using namespace loco;'
|
|
This commit makes to support ArgMax op for acl neon except int32 type.
- Introduce NEArgMinMax layer
- Apply NEArgMinMax layer for neurun
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
|
|
It introduces `tflite2circle` tool.
Its input is `tflite` and output is `circle`.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
|
|
* [moco-tf] copy_shapedata with two inputs
This will introduce copy_shapedata that can receive two inputs and return shape from broadcast algorithm for binary operation nodes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* add note
|
|
* [exo-tflite] Enhancing TestGraph.h
1) Adds more operations. 2) changed the symantics of append() with no arg.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* typo, adds better comment
|
|
This will introduce IR for Div and Sub
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
- Add infra dir
- Fix tests/framework install position
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
|
|
This commit introduces GraphOutputIndexQueryService interface, and
revises output_nodes helper to use this new service interface.
This commit also implements GraphOutputIndexQueryService for canonical
dialect in order to guarantee backward compatibility.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
|
|
This commit extends CanonicalShapeinferenceRule with v2 API implementation.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
|
|
Several canonicalization implementations shows different behavior
according to Knob configuration.
This design makes it a bit difficult to test each implementation in
isolation.
To address this issue, this commit extracts Knob control code and put
them in top-level Canonicalizer.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
|
|
* [locop] Show data type if possible
LinearV1 Graph Formatter now shows the data type of each node.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Fix a typo
|
|
`get_knob` was renamed to `get` since it always use with knob name and `knob` is duplicate. (e.g., `get<Knob::EnableTFLDialect>`)
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
|
|
- neurun test script for acl_cl and mixed
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
|
|
- Run new neurun test script on tizen
- Remove unnecessary zipped file
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
|
|
Use nnapi in android-10 r2 release
Signed-off-by: Hyeongseok Oh <hseok82.oh@samsung.com>
|
|
nnpackage can have two kinds of models.
`nnpackge_model.fbs` is for `circle`.
Signed-off-by: Sanggyu Lee <sg5.lee@samsung.com>
|
|
* Use string as backend configuration arg
- remove unness. enums for backend/operation
- Use string as backend api arguments
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Fix backend conf api
- Remove unness. functions
- Reflect new backend conf api
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Use const char* in c++ header
|
|
* Update tizen's skiplist
Add arm-linux's skiplist into tizen's
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
* Use arm-ubuntu's skiplist on tizen
- Now, tizen's skiplist is the same as ubuntu's
|
|
Let's make ReLU6 similar to other element-wise unary operations.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
|
|
This commit introduces type inference for TransposedConv2D
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
|
|
This commit registers several testcases as mocotest-tf.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
|
|
This commit adds test case for TransposedConv2D in loco Service
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
|
|
1) "TFL.xxx" -> "tfl.xxx", 2) op name printing in one method (opname())
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
|
|
All the execute method implementations invoke "erase_annot_data"
internally.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
|
|
* [exo-tflite] Comment for TensorBroadcast
This will add comment for TensorBroadcast in CanonicalNodeConverter template list as an ite
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* edit message
* fix typo and sentence
|
|
This will add handler for TFLMul node in TensorBroadcastConverter
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
* [exo-tflite] adding FeatureBiasAdd converter
This adds FeatureBiasAdd converter, which converts loco::FeatureBiasEEdd to locoex::TFLAdd.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
* typo
|
|
This will introduce methods for shape broadcast in FixShapeTransform that will be used in shape inference of binary operation nodes with inputs having different shapes
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
This will update conversion list to convert EltwiseMul and TensorBroadcast node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
This will add Shape and Type inference for TFLMul node
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
This will add a test of broadcast algorithm with TFLAdd
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
This commit deletes deprecated Pad.h.
Signed-off-by: seongwoo <sw4670.chae@samsung.com>
|
|
This will introduce as_node_shape() method in FixShapeTransform that returns loco::NodeShape from ShapeInferenceData
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
* Do not derive backends from `Pass`.
* Move backends into separate directory.
Signed-off-by: Sergei Barannikov <s.barannikov@samsung.com>
|
|
This commit unifies functions for elementwise ops in acl ex.
Signed-off-by: jiseob.jang <jiseob.jang@samsung.com>
|
|
This commit cleans up deprecated code in element-wise binary node
implementations.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
|
|
Add nnfw-devel package on tizen.
Signed-off-by: Chunseok Lee <chunseok.lee@samsung.com>
|
|
* [loco] Introduce TensorAxisSet
This commit will introduce `TensorAxisSet`, which is a set of `TensorAxis`
Signed-off-by: Seok NamKoong <sk.namkoong@samsung.com>
* apply feedbacks
* remove redundant headers
|
|
This will introduce EltwiseMulConverter that converts EltwiseMul to TFLMul
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
* [exo-tflite] Converter for TensorBroadcast IR
This will introduce TensorBroadcastConverter that resolves loco::TensorBroadcast IR that meets some certain condition
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
* make compiler happy
* fix comments
|
|
This will add EltwiseConverter to conversion list from knob status
Signed-off-by: SaeHie Park <saehie.park@samsung.com>
|
|
* [locomotiv] Introduce Eltwise Binary Node Helper
This commit introduces a shared helper method for for element-wise
binary nodes, and rewrites the "execute" method for these nodes using
this helper.
Signed-off-by: Jonghyun Park <jh1302.park@samsung.com>
* Add a blank line
|
|
This commit removes unused includes for TFConv2D and
TFConv2DBackpropInput.
Signed-off-by: Cheongyo Bahk <ch.bahk@samsung.com>
|
|
A knob for loco::FeatureBiasAdd is added.
Signed-off-by: Hyun Sik Yoon <hyunsik.yoon@samsung.com>
|