Age | Commit message (Collapse) | Author |
|
* Refactored Elementwise Workload into the single workload.
* Execute() function will react based on the DataType.
Change-Id: I6d4d6a74cec150ed8cb252e70b629ed968e7093d
Signed-off-by: Sadik Armagan <sadik.armagan@arm.com>
|
|
* fixed typo in Ref Merger Workload comment
* fixed typo in ViewsDescriptor comment
* made the origins descriptor accessable in the ViewsDescriptor
(needed for serialization)
* based the unit test on the use of the splitter in the CaffeParser
Change-Id: I3e716839adb4eee5a695633377b49e7e18ec2aa9
Signed-off-by: Ferran Balaguer <ferran.balaguer@arm.com>
Signed-off-by: Francis Murtagh <francis.murtagh@arm.com>
Signed-off-by: Jim Flynn <jim.flynn@arm.com>
|
|
parameters
Change-Id: Ie67ce4966c5e5fef618876b027292da429de1485
Signed-off-by: Ferran Balaguer <ferran.balaguer@arm.com>
|
|
* Assign output shape of MobileNet SSD to ArmNN network
* Add m_OverridenOutputShapes to TfLiteParser to set shape in GetNetworkOutputBindingInfo
* Use input quantization instead of output quantization params
* Correct data and datatype in Inference test
Change-Id: I01ac2e07ed08e8928ba0df33a4847399e1dd8394
Signed-off-by: Narumol Prangnawarat <narumol.prangnawarat@arm.com>
Signed-off-by: Aron Virginas-Tar <Aron.Virginas-Tar@arm.com>
|
|
* end to end tests for Detection PostProcess float and uint8
* add anchors to AddDetectionPostProcessLayer
* add anchors to VisitDetectionPostProcessLayer
* refactor code
Change-Id: I3c5a9a4a60b74c2246b4a27692bbf3c235163f90
Signed-off-by: Narumol Prangnawarat <narumol.prangnawarat@arm.com>
|
|
* implementation of DetectionPostProcessQueueDescriptor validate
* add Uint8ToFloat32Workload
* add implementation of Detection PostProcess functionalities
* add ref workload implemenentation for float and uint8
* add layer support for Detection PostProcess in ref
* unit tests
Change-Id: I650461f49edbb3c533d68ef8700377af51bc3592
|
|
* Remove Equal and Greater from RefElementwiseWorkload
* Create RefComparisonWorkload and add Equal and Greater
* Update ElementwiseFunction for different input/output types
* Update TfParser to create Equal/Greater with Boolean output
* Update relevant tests to check for Boolean comparison
Change-Id: I299b7f2121769c960ac0c6139764a5f3c89c9c32
|
|
* Add end to end test for Gather operator
* Add Support for int32 to Constant layer for Ref
* Add Int32Workload
* Add RefConstantWorkload as template for float, uint8, int32
* Remove unused RefBaseConstantWorkload
* Remove unused RefConstantFloat32Workload
* Remove unused RefConstantUint8Workload
* Add support check for int32 in LayerSupport functions
Change-Id: Ic970588a49ebe2aafb12be8adef52371feacaa7b
|
|
* add implemenentation for GatherQueueDescriptor validate function
* add FirstInputTypedWorkload to allow type check on the first input tensor only
* add ref workload implemenentation for float and uint8
* add Gather layer support in Ref
* unit tests
Change-Id: I4578a3211f11d24aa29d15bcf7f45b0445bcd1ee
|
|
separate branch) and master
* Brings in all the changes done for the pluggable backends
* Added sub-graph support and tests
* Added precompiled layer support and tests
* Moved BackendSettings to a separate file
* Removed the backend-specific code
* Ported DebugLayer and associated functionality
* Included fixes to make those changes work with master
Change-Id: Id7028fa7917527b844628d5aff5732e3d94c0488
|
|
*Added Unit Tests
Change-Id: I6cceb8e6dcda35ce08415f8e5ca86019a64d26e3
|
|
in favor of a type-indepent implementation that takes void-pointers
* The new implementation requires the size of the type to be passed
to the function
* Updated all the usages accordingly
* Removed the old implementation no longer used
!android-nn-driver:469
Change-Id: I37f4e6d62a38fbb8ec8c39bb559a2c54c83365d4
|
|
* Unified ArmNN's weight format to [ M, I, H, W ] for the depthwise convolution
* Added conversion utilities to permute/reshape the weights as appropriate
when using CL and Neon backends
* Updated the reference implementation of the convolution
* Updated the relevant unit tests accordingly
!android-nn-driver:459
Change-Id: I07d0818efa9d1ca1e5dad82983aac1fe78eadb18
|
|
StringMapping was located in the backendsCommon
Change-Id: I1c4a0bfca986722c721fcb2b460c03d6e1cd4568
|
|
* Added the Greater operation as an element-wise workload
* Added the unit tests
Change-Id: Ie00ee30e47a5f5e17a728032eeb11a085d06c8f2
|
|
* Add reference equal workload
* Add Reference Workload Unit Test
Change-Id: If2848e7dde4248566b99d91726d08143c40ff80d
|
|
Change-Id: Ib2e5de2a057da57ef77a9f5c4367d699d4773294
|
|
* Removed the input swizzling when the data layout is NHWC
* Permuting weights depending on the data layout used
* Added getter methods to ParsedConstTfOperation to get the tensor
info and the storage memory area, needed for swizzling the weights
* Added unit tests for both NHWC and NCHW data layouts
Change-Id: I6543900c594417df630b2663d8551158b93b7836
|
|
* Since DataLayoutIndexed is now required in the TF parser, this
changes move it to the armnnUtils library so that it'll be
accessible by the armnnTfParser
* Modified CMake files and Android.mk files accordingly
Change-Id: Ie2620359ef288aeff64cb9e9bec068a466eee0e9
|
|
Change-Id: I017ca6c23b62a8978982de0ca4ad204cb8cf7c67
|
|
Change-Id: Id7302c6b1df995ebe6eb8eb94bab38bee1b31b0b
|
|
workloads and workload functions
Change-Id: I6f3fce12a55f7d38ceafcdfcd6b5181bf56e2c09
|
|
Change-Id: Ifeacc0adb4547c72537b9ea7a61bf3c4ec3673fa
|
|
Change-Id: If8d8087d9d365e467d3ca9bf9c40d7219cb75cfd
|
|
* Ported Google's LSTM implementation to RefLstmFloat32Workload
* Fixed the code throughout because of an error in the docs around the
scratch buffer size
* Updated IsLstmSupported
* Added the unit tests
!android-nn-driver:127
Change-Id: I5577b7e39ca52df1a7f102a9b437df6aa99520b6
|
|
BatchToSpaceNdLayer.
* added Unit test
Change-Id: I80f55e8c7afb39d96006c8dd027fc9683ea8182e
|
|
* Fixed error in InferOutputShape implementation
* Added better error checking to the BatchToSpace implementation.
* Added defaults to the batchToSpace descriptors.
* Changed crops to be a vector of pairs to align with the SpaceToBatch implementation
Change-Id: Ib1c16d871f0898a1caeb6629c1fee6380a773e14
|
|
Change-Id: I13c6728dbb60643d0e086d171225c5d802987f92
|
|
Change-Id: I27ffebdece6e68460931a44c15b9b029f9fce638
|
|
Change-Id: I663a0a0fccb43ee960ec070121a59df9db0bb04e
|
|
* Fixes bug when calcuating indexes for NHWC in
Pooling2D CpuRef implementation, it now uses
TensorBufferArrayView.
* Adds 2-Channel unit tests for Pooling2d on CpuRef,
Cl and Neon. The single channel tests were not
properly exercising Pooling2d using NHWC data layout.
* Refactors Pooling2D NHWC tests so that the input and
output data are permuted to NHWC when necessary,
instead of hard coding the data in NHWC format.
Change-Id: I5b9d41ed425ff283ea8c8ef6b1266ae0bc80f43b
|
|
* Calculate index for NHWC to compute Convolution
* add more unit test for NHWC
Change-Id: I800d649b9b42be2758c445e3b3e76142888c1377
|
|
* Adds implementation to plumb DataLayout parameter
for Pooling2D on CpuRef.
* Adds unit tests to execute Pooling2D on CpuRef
using NHWC data layout.
* Refactors original tests to use DataLayoutIndexed
and removes duplicate code.
Change-Id: Ife7e0861a886cf58a2042e5be20e5b27af4528c9
|
|
Change-Id: I41f3606198db1fda8d72aaf5169594ba9156eb38
|
|
* Updated the ConvImpl.hpp to use DataLayoutIndex
* Enabled unit test for CpuRef
* Update CreateWorkload Tests for ref with NHWC
Change-Id: Id309b7ef677489d63dcb5e09bd48ab9624b5ebfb
|
|
Normalization workload
* Enabled NHWC support in RefNormalizationFloat32Workload for
Across normalization
* Added unit test for the reference implementation
Change-Id: I0e1f319e76491d43b83b121c273fadb5b259d1a0
|
|
and UInt8
BatchNormalization workloads
* Enabled NHWC support in RefBatchNormalizationFloat32Workload
* Added NHWC unit tests for both FP32 and U8
* Refactored the existing unit tests
Change-Id: I6aa18f1dcc0666b80a17a7ed229cf53607bae147
|
|
Change-Id: Ic7882cb415ccb7a899f5fb353fb96a1f19d03346
|
|
* Enabled the now supported ref layer tests
* Re-enabled the failing test now that the bug has been fixed in
ACL 1903a9976ae24f40cb2203364211ed62fcfbb985
* Added CreateWorkload test for ref L2Normalization NHWC
* Refactoring the ref L2Normalization for clarity
!armnn:153723
Change-Id: Id0067e49072b3e057ffe3ae3b70d928be6091c0f
|
|
* Added NHWC support to TensorBufferArrayView class
Change-Id: I41e1d0acd226a471ec834e380389631d9236cb00
|
|
* Added RefPadWorkload implementation
* Added unit tests and applied them to CL and Ref backends
* Fixed a bug in ClPadWorkload
Change-Id: I8cb76bc9d60ae8a39b08d40f05d628e3b72f6410
|
|
!android-nn-driver:153490
Change-Id: I1c2a5f942e3a1c3626e093c90545ca27c64ba5e8
|
|
* Adds implementation to plumb DataLayout parameter
for ResizeBilinear on CpuRef.
* Adds unit tests to execute ResizeBilinear on CpuRef
using the NHWC data layout.
* Adds DataLayoutIndexed API, allowing easy access to
the Channels, Height and Width of a tensor based on
its data layout. This reduces code duplication.
* Refactors original ResizeBilinear implementation and
tests to use the DataLayoutIndexed API when required.
Change-Id: Ic2b8916cdd2e370d070175547079d774daf6d7bf
|
|
* Added RefPermuteFloat16Workload to serve as a fallback when CL
does not support the required permute configuration for FP16
* Move Half.hpp to armnnUtils as the utils library should not be
including private header files from the armnn library
Change-Id: Ibf0f698451e8406f7ed7cce470dab60b6d16361d
|
|
* moving backends/ClWorkloads to backends/cl
* and moving pure Cl workload related code to
backends/cl/workloads
Change-Id: I019a3c6b4da5e7a23074bf03fb057e63199ad129
|
|
Change-Id: I6fb15c407024e3b91d5abf4513f8090be5821760
|
|
* Reference backend is renamed to backends/reference as per
https://confluence.arm.com/display/MLENG/Pluggable+backends
Change-Id: I27a13c274eb60995dfb459e3c49c0e2f60bcd32c
|