Age | Commit message (Collapse) | Author |
|
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: Ib733b9fe1a5b6f47dc110c7bb96cbea9d656df0d
|
|
* Add rescales before and/or after ops when using int8
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: Ic9704f91147a1a439411511c7d57d0aa52b93e26
|
|
* Onnx parser.
* Async execution interface.
* Shim and support library.
* Arm NN converter
* GpuFsa backend.
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
Change-Id: Ia9adae4da6d9bd2b92a4f4492a022e8337f57f14
|
|
* Debug layer supported data types updated
* Added deprecation notices in md files to functionalities that will be removed
Signed-off-by: Nikhil Raj <nikhil.raj@arm.com>
Change-Id: I8673bdc2d596bfa4de751fc95dae010b195ad0b1
|
|
* The abort is caused during destruction of flatbuffers::ClassicLocale
* Set FLATBUFFERS_LOCALE_INDEPENDENT=0 in delegates and parser so that
ClassicLocale is not included
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: I34584b05998a62bae2263a2281414fcf8c12d668
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
|
|
* Fix for Neon IsLayerSupported to properly check for multiple Quantization Scales
Signed-off-by: Orlaith Monahan <orlaith.monahan@arm.com>
Change-Id: I9f4558cbd62ce1657adb5025ac16c2b5d69d12b1
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I59fe96b0a272fa6984bfc172bf3e110476f3ce7b
|
|
were generated.
* input_<GUID>
* constant_<GUID>
* intermediate<output slot>_<GUID>
* output<output slot>_<GUID>
Input and constant do not need output slot as input layers as well as
constants only have one output slot, therefore with the GUID is enough
to make them unique.
This was the case for constants, but for inputs we were adding the input
slot, which is not needed.
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: I07393fc60f3135337b59a9780aa3a263a995fc9c
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: I520a9d6cc95faeefaf82947a0547292e869aa523
|
|
* Use ComputeSplitAxis in SplitOperator in tosaCommon mappings
* Fix TosaRef split tests, that were missing outputInfos
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: Ib577eacdc6399242f37d25494e208aa56db6334c
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: I70f81d17b696c30bf7a06491281184c4ddb8afc1
|
|
* TosaConv2d
* TosaQuantization
* TosaRescale
Signed-off-by: John Mcloughlin <john.mcloughlin@arm.com>
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I6c7ceca1f7df62896b41a84e6a6448afd8c32b74
|
|
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
Change-Id: Ie2054393188c6099ecee2e09b7008860b9a7178c
|
|
* Added Comparison and LogicalBinary to AddBroadcastReshapeLayer
optimization.
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
Change-Id: I4f4bafb961daf63a733be9a1f17067fd246607ad
|
|
Currently Arm NN will fail to load models containing quantization
scale value of zero.
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
Change-Id: Ifefcee1279b8667da63d1aa7d42e5d44875f9fbe
|
|
* Keep the order of the operators in TosaRef, so that const ops go first
* Remove IsLayerSupportedTosaReferenceConstantUnsupported and open ticket in MLTOSA
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Ifaa6c26dd8ad7d531f1691320d8c731956b910aa
|
|
Change-Id: I03c26ab763ec306a6efe31f9b4e0b1c058589bf7
Signed-off-by: Nikhil Raj <nikhil.raj@arm.com>
|
|
This reverts commit 518a1e4a438f47267235cff450d03efc8e7599d8.
Reason for revert: <Nightly test failing>
Change-Id: Ie595d1831a8391a1c9624beced4bae37898438a7
|
|
* Skip inference if profiling is set to output details only
Signed-off-by: Declan-ARM <decmce01@arm.com>
Change-Id: I9f35d8b1237bfa66790322a40f7ae6d09bbf6eb1
|
|
* Introduced in IVGCVSW-7853 (causes issues with older compilers)
Signed-off-by: Declan-ARM <decmce01@arm.com>
Change-Id: I945b7be5e0e8fe47c8d4859d1db40b5130392363
|
|
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: If02101d881bb7c937e858e464a043d849ac40ba9
|
|
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: I7148c25679a8a919c6138df6b23d0129e8ddd0a5
|
|
* src/armnn
* src/armnn/layers
Signed-off-by: Declan-ARM <decmce01@arm.com>
Change-Id: Ic78cbbb59e90fbb15f893205a358c45264243721
|
|
* Added parsing functions to the serializer and deserializer
* Added ScatterNd and its Descriptor to the ArmnnSchema.fbs
* Added Unittest for Serializer and Deserializer
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: I1ed674dc32d2e2d0d84dca4c7018984ea367ea50
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Id89233954dd8da600c2f82e718df849b098c8af4
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I5b87304319f5e83e8eba9cb2d934fc4a6aebe85b
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I0774a9580710350bd0a959cf68dfe057f52e4330
|
|
* Failures was: MultiplicationQueueDescriptor: Tensors input_0 & input_1 must have the same number of dimensions in order to be broadcasted
* When trying to apply AddMulAdd, Mul dimensions were being collapsed before checking the second ADD.
* If the second ADD determined that fusion cannot happened MUL should have not been collapse.
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I6dda131c424e1bf7b22fff40c5bef8977f073b24
|
|
(scatter_nd, scatter_nd_add, and scatter_nd_update, scatter_nd_sub, scatter_nd_min, scatter_nd_max, scatter_nd_mul)
* Front end support for ScatterNd added.
* Reference implementation for ScatterNd added.
* Unit tests added.
Signed-off-by: Tianle Cheng <tianle.cheng@arm.com>
Change-Id: I30da9056d9b03ca9b5fb8d09987341128badbcf4
|
|
* Updates build rules for platform
* Failing unit tests excluded (dynamic backends, profiling)
* install-packages.sh generates manual setup information
Signed-off-by: Tracy Narine <tracy.narine@arm.com>
Change-Id: I2d3d434aa615a8796c0cb94cd5b9c35a5acfd148
|
|
* Replace calls to ARMNN_ASSERT with DOCTEST CHECK.
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
Change-Id: I8904d169b2099d57a344e319b2f14cf5d8392ae8
|
|
* Identify usages of ARMNN_ASSERT that should be proper exceptions.
* Change ARMNN_ASSERT in Doctests to CHECK.
* Verify any remaining assertions are reasonable.
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
Change-Id: Ifd1f2a5a4bb60135e8654305035ec70e09c4dc2d
|
|
* Rewrote constexpr check to avoid a compile error
Signed-off-by: Tracy Narine <tracy.narine@arm.com>
Change-Id: I09a61314b1b4a5aa1e2baa52711f470802f04131
|
|
* Skipping the optimization which folds pad and conv2d
together for a specific case: 1x1 filter and
padding size >= filter size
Signed-off-by: Tracy Narine <tracy.narine@arm.com>
Change-Id: I46944e9f736df1ff60469b2d2852e1bba01ab8cd
|
|
* Add Reshape EndToEnd tests to all backends
Signed-off-by: Declan-ARM <decmce01@arm.com>
Change-Id: Ic6d07ba8de0cf3271ed0e4c6d604e070ccb968e3
|
|
* Currently Sigmoid and TanH Functions are implemented.
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: If9483be9201dfe47b86acc41ec7932725ac2e39e
|
|
* Added softmax operator support
* Added test cases
Signed-off-by: John Mcloughlin <john.mcloughlin@arm.com>
Change-Id: I51d530b110c4cb812f5aab31ad1ee4022d81d19e
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Ib95eb0fd71106e684cb7652917b8de9f0ac73f9c
|
|
* Refactor to generalize
* Add MUL
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I2ee273d50d3a8b114b5a41abc8ee7585b15e3308
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I9714c4c57e923ac775dcde2951de07cea35c40ee
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I695ef452d004ed7b606020037cad681ef1fc80c3
|
|
* Added cast operator support
Signed-off-by: Tracy Narine <tracy.narine@arm.com>
Change-Id: Ie12cb1559a7a059ff35e1c395bc77243499243cd
|
|
* Add Pool2d EndToEnd tests to all backends
* Add utility functions for the attributes in a separate file
* Remove some unnecessary includes
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I0f82ebbf7b3301c6368462fb4fb4d4d02b246fc6
|
|
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
Change-Id: I3d089e7f1b75596501130d3ece3a94dd326cc27e
|
|
* Added support for Gpu Sub operator
* Added unit tests
Signed-off-by: John Mcloughlin <john.mcloughlin@arm.com>
Change-Id: I1efaa485772a3716e3781566843bd50bd9bab811
|
|
* Adding support for Gpu Add operator
* Added tests for layer support, end to end and optimization
Signed-off-by: Tracy Narine <tracy.narine@arm.com>
Change-Id: Ie9328d269c5c0ff60a7e10133b728ac9265033af
|
|
* Added DepthwiseConv2d support for GpuFsa backend.
* Updated DepthwiseConv2d End-to-End test
Signed-off-by: Tianle Cheng <tianle.cheng@arm.com>
Change-Id: I646839980d138ae235a00990c97c6e66a4418a5e
|
|
* Updates to the existing GpuFsa backend to incorporate patch:
https://review.mlplatform.org/c/ml/ComputeLibrary/+/10990
* Update the ACL pin to include the patch with the fixes
Change-Id: I08d111265f4617657ee7f20249aeb111f64ba7a9
Signed-off-by: David Monahan <david.monahan@arm.com>
|
|
* Using the tosa defines from the serialization library
to avoid compile errors in other backends
* Fixing a bug in the version compat macro
Signed-off-by: Tracy Narine <tracy.narine@arm.com>
Change-Id: Ie4ee80666c6f8033bb72e0e6cb8ca5ef41933990
|
|
* Add PreCompiledWorkload implementation for GpuFsa
* Add ConstantWorkload implementation for GpuFsa
* Add Input/Output workloads for GpuFsa
* Added CopyMemGeneric workload for GpuFsa
* Separate creation and validation of sketch tensors into seperate functions
Signed-off-by: Kevin May <kevin.may@arm.com>
Signed-off-by: David Monahan <david.monahan@arm.com>
Change-Id: Ie7299a4c61073b5ca03d9f8681458869ef7ce743
|