Age | Commit message (Collapse) | Author |
|
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
|
|
* src/armnn
* src/armnn/layers
Signed-off-by: Declan-ARM <decmce01@arm.com>
Change-Id: Ic78cbbb59e90fbb15f893205a358c45264243721
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I0774a9580710350bd0a959cf68dfe057f52e4330
|
|
(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
|
|
* Adding a one to many tosa mapping for Quantize
* Added tests
* Resolves IVGCVSW-7175
Signed-off-by: John Mcloughlin <john.mcloughlin@arm.com>
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Ia0852fefb618b4a29c2601b9de8b6b2731229801
|
|
Squashed commit of the following:
IVGCVSW-7159 Add GpuFsa backend skeleton
IVGCVSW-7380 Update the GpuFsa Skeleton to build and load ACL
IVGCVSW-7381 Add IsLayerSupported implementation to GpuFsa backend
IVGCVSW-7382 Implementation of Conv2d within GpuFsa
Signed-off-by: James Conroy <james.conroy@arm.com>
Signed-off-by: Matthew Sloyan <matthew.sloyan@arm.com>
Signed-off-by: David Monahan <david.monahan@arm.com>
Change-Id: Id23d9ee598535de7b38a99ca223cdf0ad2102cef
|
|
* Added validation for scale on all Quantized types
* Added Encoder for Per Axis UINT16 Symmetrical Quantized type
* Added error for Per Axis Asymmetrical Quantized type not supported
Signed-off-by: John Mcloughlin <john.mcloughlin@arm.com>
Change-Id: I433519ccacd71219a92bde2b81955d6abf9219c5
|
|
Signed-off-by: Idriss Chaouch <idriss.chaouch@arm.com>
Signed-off-by: Narumol Prangnawarat <narumol.prangnawarat@arm.com>
Change-Id: I94ec5f9120b2d736fdf98d00ec5137a4efd739b8
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Ic778d35b001474b44fb1e433a6fe276e4ec9f565
|
|
Signed-off-by: Francis Murtagh <francis.murtagh@arm.com>
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: I6686ad0a16ef763dbdc771c7ef792ae2890e0ab2
|
|
* Check if preferred backends have FP16 support before enable fp16-turbo-mode
* Unit tests
* Replaced global gpuAccCapabilities with getter method construction
* Replaced deprecated function call in SL shim
Signed-off-by: Narumol Prangnawarat <narumol.prangnawarat@arm.com>
Signed-off-by: Ryan OShea <ryan.oshea3@arm.com>
Change-Id: If29b62b330ca8987de8acf6408db11daf25ca0b5
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Signed-off-by: Cian McGriskin <cian.mcgriskin@arm.com>
Change-Id: I0afb2403fee11c5c1e58ea65e2525e99594d8f2d
|
|
* Follow up review to clean up whitespace and copyright errors mentioned
in https://review.mlplatform.org/c/ml/armnn/+/9885
* Added BinaryElementwiseOperation to .dot files
* Refactored ConnectedToSplitterWithMoreThan4Dims function to more
generally useful ConnectedToLayerType function
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: I0e3d0895888f3a3f0a9758ce30bc031aba50812b
|
|
Layer with 2 inputs
* Changing ReverseV2 to use two inputs
* This is required by the backends
* The ReverseV2Descriptor was removed
* Tests updated
* Added a Run<> templatefor inputs with different data types
Signed-off-by: Tracy Narine <tracy.narine@arm.com>
Change-Id: I22f947de829b4b3da6bda3a74f4ffdef4052cc25
|
|
* Added optimization to remove reshapes for Neon and Ref Backends
by using overridden TensorInfos
* Added ability to delete Subgraphs during Optimization
* Fixed naming error in NeonEndToEndTests and CLEndToEndTests
* Added LayerNameAndTypeCheck for testing.
* Fixed error where layers were not marked as altered when removed in
CLBackend
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: I1ac25cd4ec9821470d961831ae2c8d24882276cc
|
|
* Descriptors added for ReverseV2
* Layer definition added
* Input validation added
* Reference workload implementation for ReverseV2 added
* Reference layer unit tests made for ReverseV2
* CompareTensors method updated to support comparison between empty tensors
* CMake and other build files updated
Signed-off-by: Tianle Cheng <tianle.cheng@arm.com>
Change-Id: I805738454421309fda77c44218a8df171d68dc18
|
|
* Added new ARMNN_STUB_PROFILING and ARMNN_DISABLE_DYNAMIC_BACKEND
defines to replace BUILD_BARE_METAL and BUILD_EXECUTE_NETWORK_STATIC
* Add new CMake variables to disable sample apps and dynamic backends
* Improve BUILD_SHARED_LIBS CMake variable
* Add new archive output location to various libraries so that the
static libraries appear in the same location as the shared libraries
* Fixes for bare metal build
* Add ARMNN_DISABLE_FILE_SYSTEM defines to missing locations
Signed-off-by: Ryan OShea <ryan.oshea3@arm.com>
Change-Id: I1db9a8f483e7134bd261294b35240cf21b150d45
|
|
Signed-off-by: John Mcloughlin <john.mcloughlin@arm.com>
Change-Id: Id4bdc31e3e6f18ccaef232c29a2d2825c915b21c
|
|
* Added Deprecation notices for old ElementwiseBinary layers.
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: I5bd0f186aaed675885d667f47e1e210ee9ec84f8
|
|
This reverts commit 52e90bf59ecbe90d33368d8fc1fd120f07658aaf.
Change-Id: I5a0d244593d8e760ee7ba0c9d38c02377e1bdc24
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
|
|
* Added Deprecation notices for old ElementwiseBinary layers.
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: Iebbbaff38cc9c347b25eb2f9054c914a4f931c68
|
|
!android-nn-driver:9329
* Added ElementwiseBinaryLayer that can represent all ElementwiseBinary
operations including Add, Div, Sub, Maximum, Mul and Minimum.
* Updated Delegate to use ElementwiseBinaryLayer instead of the Add,
Div, Sub, Maximum, Mul and Minimum layers.
* Updated Deserializer to use ElementwiseBinaryLayer instead of the Add,
Div, Sub, Maximum, Mul and Minimum layers.
* Updated OnnxParser to use ElementwiseBinaryLayer instead of the Add
layer.
* Updated TfLiteParser to use ElementwiseBinaryLayer instead of the Add,
Div, Sub, Maximum, Mul and Minimum layers.
* Updated CL and Neon tests to use ElementwiseBinaryLayer.
* Updated CL and Neon Backend Specific Optimizations to accept
ElementBinaryLayers as well as Add, Div, Mul, Sub, Maximum and Minimum
layers.
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: I7cbb96b60eb01f0e2b57b0541016d48a08b86c75
|
|
* Folding of pad into conv2d expected a Constant layer not Dequantisation
* Fusing Dequantisation with Constant to a Constant ensures that.
* Group Constant layer optimizations together where possible.
* Add unit test.
Signed-off-by: Francis Murtagh <francis.murtagh@arm.com>
Change-Id: Id0393313bf097595f2f13738b7513e427116ea4a
|
|
The -F execute network option creates a directory to print intermediate
tensors but minor problems caused serious failures. This attempts
to clean up the error handling.
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
Change-Id: Ia44c008919b1bee299b43a672235b1fcc25bf1bd
|
|
android-nn-driver.
* Not a concern with the delegate/parser as tflite builtin operators have
little float16 support
* Will also fix float16 workloads running on the support_library.
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: Iec2033dbc8ece2140b188de1f193c344a68b9c36
|
|
IsLayerSupported()."
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: I7fa4110a1b545c1fcd79e387d53ae7d6cd381f31
|
|
* Removed weights and bias from Convolution, DepthwiseConv & FullyConnected
layers
* Removed the weight and bias ConstTensorHandles from the QueueDescriptors
* Updated Workloads to take tensors from WorkloadInfo rather than the
QueueDescriptors
* Removed unused RedirectMembersToConstantInputs optimization and tests.
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: I9ffcdc4a1c0dff725539dd69fc435b700bd98a56
|
|
* Done as part of 22.11/23.02 innovation days.
* IsLayerSupported() is called in model prepare (delegate, android-nn-driver and shim/support_library)
and again in ArmNN once model otimization is performed.
* From calling IsLayerSupported() the first time, we should know that the layers are supported
and what backend they are supported on.
* Solution is to set the BackendId of the IConnectableLayer when IsLayerSupported() is called the first time,
* In the Optimize() function we then check if the backend is set. If so, we do not call IsLayerSupported() again.
* In the case a layer that is supported gets optimized, then the BackendId of that layer get set to "Unknown"
for the new optimized layer and IsLayerSupported() will get called on the newly optimized layer.
* Includes bug fix IVGCVSW-7213 for Android Mean FP16 CpuAcc tests. Also related to bug IVGCVSW-7211.
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: I7a7820d0cdb079ffb5a3a2e0c44e252f652df53b
|
|
- Remove Bf16ToFp32 Conversion Layer
- Remove Fp32ToBf16 Conversion Layer
- Remove B16 Conversion tests
* Throw exception if m_ReduceFp32ToBf16 optimzer option is set to true
* Provide comments to enable fast math in order to use bf16
* Update docs to inform users to enable fast math for bf16
Execute Network Changes
* Require bf16_turbo_mode to also have fast_math_enabled set to true
- Remove setting m_ReduceFp32ToBf16 optimizer option
Signed-off-by: Ryan OShea <ryan.oshea3@arm.com>
Change-Id: Ibaa6da9d29c96a1ce32ff5196b0847fde9f04a1c
|
|
* In TF what ArmNN calls quantized data types can be non-quantized as well.
* This patch creates 2 models:
* ArmNN: model where int8 and uint8 will always be quantized, but scale can be 1 and offset 0
* TFLite: model where int8 and uint8 can be quantized and non-quantized
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Id960f2f30988f2bbec88cb4e0c52c189ac957bae
|
|
* Bug fix where files were being overwritten at each debug layer
Signed-off-by: Keith Davis <keith.davis@arm.com>
Change-Id: I609fdc82afcee925824efb02183c7dbc942fced0
|
|
* Add functionality to print output tensors to file in tempdir
* UnitTests
Signed-off-by: Keith Davis <keith.davis@arm.com>
Change-Id: Idfb4c186544187db1fecdfca11c662540f645439
|
|
Signed-off-by: Jim Flynn <jim.flynn@arm.com>
Change-Id: I3573078206272c3a72a2b3acf8781ab458ea6c90
|
|
* Fix for Debug mode in ExNet does not work with ConstTensorsAsInputs
* Remove unnecessary assertion with ambiguous message in LoadedNetwork
Signed-off-by: Keith Davis <keith.davis@arm.com>
Change-Id: I9cd5d1f811dbbc89072d1190c510bf1b22e3069c
|
|
* AddConv and AddDWConv with weights and bias
* ResizeBilinearDescriptor
* b,blacklist option in accuracy tool
!android-nn-driver:8172
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Ibbc04fd18be7f938b11590bf67cd7af103cb4d99
|
|
* Move the Conv2D and DepthwiseConv2D validation to Optimization level
when the weights and tensors are as constant inputs
* Take into account offset and scales values when doing INT8 to FP32 dequantization
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: I1f81f15640395ac041923b10dbe9151159715117
|
|
This reverts commit 1a7f033768acb27da11503bd29abb468d2e77f9e.
List of fixes to be able to add this code again:
* "emplacing_back" the vector inputTensors into the vector m_InputTensorsVec outside the for loop
* GetIOInfo() uses IOptimizedNetwork instead of INetwork, where the infered shapes are not saved
* Add missing data type Signed32 to SetupInputsAndOutputs()
* PrintOutputTensors() prints the actual output without dequantizing
* Add profilingDetailsMethod as input in networkProperties in ArmNNExecutor constructor
* Fix typos
Change-Id: I91de166f87228282db3efa27431fe91458834442
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: Ic6634d48892d11e5f146cdf285e1e333e93e9937
Signed-off-by: Francis Murtagh <francis.murtagh@arm.com>
|
|
* Descriptors added for BatchMatMul
* Layer definition added
* Input validation added (will likely change when opt. param support comes in)
* Ref workload implementation for BatchMatMul added (will also change with opt. param support)
* Ref layer tests made for BatchMatMul
* CMake and other build files updated
Signed-off-by: Samuel Yap <samuel.yap@arm.com>
Change-Id: Ic885301da543ee0fbe7922b85e7f9658c4efc617
|
|
Fp32NetworkToBf16Converter
* Fuse FP32ToBF16Layers with Constant Layer so Conv2d/FullyConnected
can have their weights redirected.
* If BF16 Unsupported in Conv2d || FullyConnected revert fused
Constant Layer to FP32
Change-Id: If523c708a822659d64597d9ae39cca1c2f84b76f
Signed-off-by: Francis Murtagh <francis.murtagh@arm.com>
|
|
Signed-off-by: Nikhil Raj <nikhil.raj@arm.com>
Change-Id: I9ccaefbe28ea572e9e2b4a2168574804667f7460
|
|
This reverts commit a0f8b15d4ddb5075f380003ff31b271d389d3b66.
Reason for revert: <Test ClDmaBufInternalTests review >
Change-Id: Ibc4a77fa008643849da7330391942e4c87b941e2
|
|
This reverts commit 03bf98a8bc51ad20eef4b9ca5fbf6ce15e063721.
Reason for revert: Caused failures in tests located in internal repo.
Change-Id: If35cb0ede349b270e4e7827324382e09455d8cfa
|
|
Only one bool is used to indicate whether inputs should be imported.
However, its possible for the user to want to import inputs but not
export outputs. In addition it's possible for a user to enabled import
during optimize but then pass a memory source that does not require
import.
* Add m_ExportEnabled to INetwork.hpp.
* Modify Network::dNetwork to consider both m_ImportEnabled
and m_ExportEnabled.
* Add ValidateSourcesMatchOptimizedNetwork to LoadedNetwork to validate
import options between optimize and network load.
* Update the TfLite delegate consider exportEnabled flag in the
optimizer.
!armnn-internal-tests:425350
Signed-off-by: Colm Donelan <colm.donelan@arm.com>
Change-Id: I776eab81595898e43f91ab40306962eae61329f4
|
|
* Use new INetwork::AddConvolution2dLayer
instead of deprecated version
* Remove duplicated test in SerlializerTests
* Fix some cosmetics
Signed-off-by: Keith Davis <keith.davis@arm.com>
Change-Id: I3407815bfdc1cdc01ca0a667b8e4d80d8621783f
|
|
* No trailing permute layer after a constant layer
* Unit test for optimization
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: I0d098f5af41d2c55df7cef1ccfb848093320ddc1
|
|
* Support Float16 as input to Dequantize layer
* Add Optimization to substitute Const+Dequantize layers with Const layer
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I58bb7e3871ca480c7b6fca93c4efb2de84e09e64
Signed-off-by: David <david.monahan@arm.com>
|
|
dimensions
* Added allow-expanded-dims to TFLite parser and ArmNN delegate
* If true ArmNN will disregard dimensions with a size of 1 when
validating tensor shapes. Tensor sizes must still match.
* This allows us to support models where tensors have expanded
dimensions (i.e. extra dimensions with a size of 1).
* Fixed bug in Network where it assumed that only the first option
could be ShapeInferenceMethod.
* Fixed bug where m_ShapeInferenceMethod was lost when copying or
moving Graphs.
* Changed Delegate to pass "infer-output-shape", "allow-expanded-dims"
and other BackendOptions through to the Network during construction.
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: Ibe7c5ae6597796fc9164cb07bd372bd7f8f8cacf
|
|
!android-nn-driver:7477
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: Ibf633ccccc385bd980934ff829407d21981323ef
|
|
* Update Front-end and Tools.
* Updated Serializer, Deserializer and unit tests to reflect this.
* Updated TfLiteDelegate, TfLiteParser and OnnxParser.
* Updated Ref.
* Fixed resulting Neon / CL tests
* Unified optimizers for conv2d ops
* Optimizer Fix - Fp32ToBf16
* Partial implementation for ACL backends to fix VTS failures
!android-nn-driver:7477
Signed-off-by: Keith Davis <keith.davis@arm.com>
Change-Id: I5fb18877f7ee32643e15a9818945356274bb401b
|
|
* Added backend specific optimization & test for CpuAcc and GpuAcc: PermuteDepthwiseConv2dWeights
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: I600476b2e9c557a39818a574c1091c9d650b21b1
|