Age | Commit message (Collapse) | Author |
|
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: Ibea856e1420ad3ff14aef2cdb3d8ee918fb1bcf1
|
|
!armnn:9319
* Refactored all functions to convert Add, Div, Maximum, Minimum, Mul
and Sub to use ElementwiseBinary layers instead.
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: Ic05885cd8692e7f1b7032862fb4a395af70e0bcd
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: If9717d6ab236d97c76f6cd39b96bde86c81e4382
|
|
!armnn:8742
* 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 optimization 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.
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: Ie5c6c9cd10d81f90b1ee78dd6e3442f353b6c109
|
|
* Add support for Android-nn-driver
!armnn:7382
Signed-off-by: Keith Davis <keith.davis@arm.com>
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: I7ace53cf1c8954c1f2d5588387616d7179ef3bf7
|
|
!armnn:7417
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: Ic37d6b8677e040c60a90358dd0a4a8eb33fb6ea0
|
|
android-nn-driver do not execute.
* Change to ConvertionUtils.hpp fix ConcatTests_GpuAcc tests.
!armnn:7300
Signed-off-by: Cathal Corbett <cathal.corbett@arm.com>
Change-Id: I1543b57c6d1f9660916ca530b421c87e54b13f72
|
|
* Removed unreachable code
* break after if else where both branches return
* Removed unused operations
* Result of dstPtr++ is unused
* Fixed possible overflow
* Axis dimensionSequence assignment can result in overflow where rank
is not 4
* Removed use of old-style casts
* Fixed spelling mistakes in error messages
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: If2a7ab63fc1d200cb18b494d99a67bbddb42f0f8
|
|
* Changed asserts to check for errors and return appropriate values or
throw exceptions
* Changed unit tests to use Doctest's long macro names as the short
macro names clashed with Android's Logging macros
* Removed unused #includes
* Clarified ambiguous #includes
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: Ice92a37590df727fd581d3be5ff2716665f26a13
|
|
* Some tests on GpuAcc and CpuAcc P & Q were skipping because
IsConstant flag wasn't set on TensorInfo passed to validation function.
Signed-off-by: Matthew Sloyan <matthew.sloyan@arm.com>
Change-Id: Idd977bf621b71bb9625d8fc6620c29915de4b431
|
|
* Constant weights and biases are now stored as Constant layers.
!armnn:5942
Signed-off-by: Matthew Sloyan <matthew.sloyan@arm.com>
Change-Id: I867a696eb684df2d0d5f7db8dd9c38933445f388
|
|
!armnn:5552
Signed-off-by: Jan Eilers <jan.eilers@arm.com>
Change-Id: I5bbc998f1a29030f718d85b646907c5a314ceecf
|
|
Signed-off-by: Francis Murtagh <francis.murtagh@arm.com>
Change-Id: I59359d4ffaf55963442ebb0231e1ee4a77616c28
|
|
* The permutation of the tensor info is now completely handled in
the armnnUtils::Permuted function. That includes quantization
informations too
!armnn:5411
Signed-off-by: Jan Eilers <jan.eilers@arm.com>
Change-Id: I40410141303d950be7888f9e491133251b6f69d8
|
|
* A number of DepthwiseConv2d Cts tests pass a filter operand with a
dimension[0] value greater than 1 in contravention of Android
documentation. Adding a check to handle this gracefully.
* Changing a hard assert in ConversionUtils::ConstTensorPin to a warning.
Signed-off-by: Colm Donelan <Colm.Donelan@arm.com>
Change-Id: I605340d098b717d1d88d259a0ed89a801287764d
|
|
* Enabled weights and bias as inputs in FULLY_CONNECTED operator.
!armnn:5180
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: Id325a8bf5be5a772191d27ae89485e992f0c48fa
|
|
Signed-off-by: Kevin May <kevin.may@arm.com>
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: I1e1db52322092c6b1b7ac6183c3adc90aabcec24
|
|
* Update to use dot operator for IsXXXLayerSupported()
* Call LayerSupportHandle member funciton to check is Backend Registered
Signed-off-by: Francis Murtagh <francis.murtagh@arm.com>
Change-Id: If0b9876068db8957a8bdb690ad323cd8dce544a4
!armnn:4924
|
|
* WIP
!armnn:4231
!armnn:4287
Signed-off-by: James Ward <james.ward@arm.com>
Change-Id: I7844efc84fac018d8aad1f72dd236ede3fd8e7f2
|
|
* Added 3D permutation vector to prevent mismatch between mappings
and numDimensions in dynamic concats
Signed-off-by: David Monahan <david.monahan@arm.com>
Change-Id: Ia580deaa8a4f6245c7e4d319d5d0ed837c7dc02a
|
|
Signed-off-by: Finn Williams <Finn.Williams@arm.com>
Change-Id: I011466158b6736670bafba474e8c9ce12c38e229
|
|
* Replaced with armnn/utility/NumericCast.hpp
Signed-off-by: Matthew Sloyan <matthew.sloyan@arm.com>
Change-Id: I6253261508e49ddeaa5ba3f3893a059486637467
|
|
* Move ProcessActivation from Convert<LayerName>
to SetupAndTrackLayerOutputSlot by passing optional ActivationFn
* Connect Activation after IsTensorInfoSet() is called
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: I472bcb65b12ae6e934bd1e9af8a6f6aceb311c0e
|
|
Signed-off-by: Keith Davis <keith.davis@arm.com>
Change-Id: I13e16d271ba55217b98a439aa82931f809fdeeb8
|
|
* Enabled Dynamic Tensors on LSTM operator
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: I9cae539559570a44088a986870d3d3e41aee9468
|
|
* Enabled dynamic type tensors on Squeeze operator in nn-driver.
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: I2092cf6087da2821a07bd4743c4676d5afb35e66
|
|
unsupported operation with with dynamic inputs
Change-Id: Ia30698d608810dabb419d7d326ad985789d50603
Signed-off-by: Finn Williams <Finn.Williams@Arm.com>
|
|
* Correctly handle HalOperandLifeTime::NO_VALUE in ConvertTranspose
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: I36790a90b5a3eac3da9e4edafc96387bb4375db9
|
|
Signed-off-by: Finn Williams <Finn.Williams@Arm.com>
Change-Id: I4729497e67b711a8f1fbf7c555a26e97fa6c9ab6
|
|
* ConvertToActivation
* ConvertAdd
* ConvertArgMinMax
* ConvertConv2d
* ConvertDepthToSpace
* ConvertDepthwiseConv2d
* ConvertDiv
* ConvertFloor
* ConvertFullyConnected
* ConvertL2Normalization
* ConvertLocalResponseNormalization
* ConvertMean
* ConvertMul
* ConvertPad
* ConvertReshape
* ConvertSub
* ConvertStridedSlice
* ConvertTranspose
* ConvertBatchToSpaceNd
* ConvertSpaceToBatchNd
* ConvertComparison_1_2
* ConvertConv2d_1_2
* ConvertDepthwiseConv2d_1_2
* ConvertElementwiseUnary
* ConvertExpandDims
* ConvertGather
* ConvertGroupedConv2d
* ConvertInstanceNormalization
* ConvertLogSoftmax
* ConvertMaximum
* ConvertMinimum
* ConvertPadV2
* ConvertPrelu
* ConvertQuantize
* ConvertResize
* ConvertSpaceToDepth
* ConvertSoftmax
* ConvertTransposeConv2d
Signed-off-by: Finn Williams <Finn.Williams@Arm.com>
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: Idacf16e5eab56d83fce293570bbc89381ae056dc
|
|
* Change NN Driver m_Network to now have ShapeInferenceMethod::InferAndValidate
* Implement dynamic tensor support for:
- ArgMinMax layer
- Pooling2d layer
- Activation layer
* Skip dynamic tensor tests for any HAL other than 1.3
Change-Id: Icf66c968e49cdd4822b8c79c5f18b3f9e97dc53f
Signed-off-by: Finn Williams <Finn.Williams@Arm.com>
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
|
|
!armnn:3330
Signed-off-by: Finn Williams <Finn.Williams@Arm.com>
Change-Id: Icc429d9fabb570193d12bffef0e00dda7b51032f
|
|
* Added check for negative stride with ShrinkAxisMask
Signed-off-by: Ryan OShea <Ryan.OShea2@arm.com>
Change-Id: I8ba0a5e84cdc5d3367b683efb620f078b6cc4147
|
|
* Add QLSTM support for Android NN Driver
* Add overrideOutputInfo parameter to SetupAndTrackLayerOutputSlot
* Add optional condition to GetInputScalar
* Refactor Quantized 16 Bit LSTM impl
Change-Id: Ie8fa98ad5ee4a62174ef91ca80f1df62b7fde937
Signed-off-by: Keith Davis <keith.davis@arm.com>
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
|
|
* Check if the output is large enough to hold the slice in the driver
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: Iaef16167e891a41186de030324bb45b8bbca4d80
|
|
* Added GetOptionalBool function
* Added checks to Resize for unsupported operands AlignCorners and HalfPixelCenters
!armnn:3062
Signed-off-by: David Monahan <david.monahan@arm.com>
Change-Id: Ic0305014ed2638e3a3d1e7520972e3a2909e504b
|
|
* Optional timeline parameters are not supported yet.
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: Ic5d6cd6af0cd87d4acba102a48fc94ec8b406377
|
|
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
Change-Id: I7962be3a77cacf15dad594f0a907499c5b39bfeb
|
|
Signed-off-by: Narumol Prangnawarat <narumol.prangnawarat@arm.com>
Change-Id: If602024a339df7548333e470545f9400c3daf7b3
|
|
* IVGCVSW-4441 Add Support for ANEURALNETWORKS_ELU
* IVGCVSW-4443 Add Support for ANEURALNETWORKS_HARD_SWISH
* IVGCVSW-4448 Add TENSOR_QUANT8_ASYMM_SIGNED data type support
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: Idb9bb3f463b956221711423c15b6557eeb1af7db
|
|
* Add new 1.3 files HalPolicy, ArmnnDriver, ArmnnDriverImpl
* Add new .rc file for 1.3 service
* Add ArmnnPreparedModel_1_3 and implement new functions
* Update Android.mk with 1.3 driver and service
* Refactor ifdef to include ARMNN_ANDROID_NN_V1_3
* Create Utils getMainModel for new 1.3 Model Main Subgraph
* Use android Utils to convertToV1_X in ArmnnPrepapredModel_1_3
* Refactor HAL 1.2 convert functions into ConversionUtils_1_2.hpp
* Replace ArmnnBurstExecutorWithCache with call to ExecutionBurstServer
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: I514069e9e1b16bcd1c4abfb5d563d25ac22d02e3
|
|
* Added TENSOR_BOOL8 support
* Added Broadcast support to comparision operators
!armnn:2903
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
Change-Id: I844e32b57399eff2dc60af9b2099145316c80cae
|
|
* replaced boost::ignore_unused with armnn::IgnoreUnused
Signed-off-by: Jan Eilers <jan.eilers@arm.com>
Change-Id: I4443a3620ed7d51b151af67c070d4225e410b5fa
|
|
* Update ErrorStatus to V1_0::ErrorStatus
* Update Request to V1_0::Request
* Update OperandType to V1_2::OperandType
* Add namespace android::nn::hal in ArmnnDriverImpl for R only
* Add missing g_RelaxedFloat32toFloat16PerformancePowerUsageName
* Add namespace V1_0 or V1_1 where necessary
* Update Android.mk with R macro and android.hardware.neuralnetworks@1.3
* Remove androidnn.go
* include IAllocator in DriverTestHelpers
* Remove unused LOCAL_CFLAGS
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: I1787f1ed6784b3bbec017536d87d49197405e853
Signed-off-by: Kevin May <kevin.may@arm.com>
|
|
* Changed ConvertTranspose to use Transpose instead of Permute
* Removed unused code related to permutation
!armnn:2787
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: Ie59ca80203109bd6bb2abcdedc5ac53c2b2eb157
|
|
* Fixed driver crashes caused by ignoring the result of GetTensorInt32Values.
* Fixed driver crashes where padding, axis, permutation dimension or block size operands were inputs.
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: I59867e9545a240e6be46b7f68da6adaa120a56d8
|
|
* Changed ConvertTransposeConv2d to use the unsigned values for strideX and strideY
not the signed values from the descriptor.
!ComputeLibrary:2599
!armnn:2612
Signed-off-by: Mike Kelly <mike.kelly@arm.com>
Change-Id: Iee300041f3ed585820b7c7e2edd8ab541f339d59
|
|
!android-nn-driver:2642
Signed-off-by: Derek Lamberti <derek.lamberti@arm.com>
Change-Id: I4db6ea4ed0a192c85f124c4a9ced60b1666a3870
Signed-off-by: Teresa Charlin <teresa.charlinreyes@arm.com>
|
|
!armnn:2643
* Add calls before addition of these layers in ConvertConcatenation
* Add outputInfo parameter wherever needed for IsReshapeSupported
Signed-off-by: Kevin May <kevin.may@arm.com>
Change-Id: Ic5d142ea046161960ff2fc137bd261ebb4e6ac0c
|
|
* Adds missing conversion for permutation vector
from ANN/TF/Numpy format to ArmNN format.
* Corrects legacy tests based on incorrect
permutation vectors.
Signed-off-by: James Conroy <james.conroy@arm.com>
Change-Id: I767142378055e484d22f3ffba34580c528370cfe
|
|
* Account for deprecated Abs and Rsqrt layers in ArmNN
* Update HAL 1.2
* Neg HAL operation support added
!armnn:2542
Signed-off-by: josh minor <josh.minor@arm.com>
Change-Id: I043bf64ac7ed3aea65560d72acf664e079917baf
Signed-off-by: josh minor <josh.minor@arm.com>
|