Age | Commit message (Collapse) | Author |
|
code base
* Made the virtual functions in ILayerSupport.hpp pure
* Created a LayerSupportBase class with the default implementation of
the interface
* Made the backend layer support classes inherit from the base
class, instead of directly from the interface
* Refactored the profiler and the profiling event classes to use
the BackendId instead of the Compute
* Implemented a proper MemCopy support method
* Changed Compute to BackendId in the profiling API and objects
* Removed static references to pluggable backends
!android-nn-driver:492
Change-Id: Id6332b5f48c980819e0a09adc818d1effd057296
|
|
Change-Id: I43be451f490db0154021f47a2fd49d1269cf5b95
|
|
Change-Id: I349f663e5e2d404930a5f9289d311e67a5a409ea
|
|
Change-Id: I7c7b65bd77b06925efdaf2c9c98c30994a12de42
|
|
* Added CL unit tests
Change-Id: Ia4af49b133774141384a177f5b042e65764bf682
|
|
Change-Id: I0bc48f118aba419abb659657b83224401fac0715
|
|
Just include the function definition that is specifically needed for each workload.
Also, tighten up the scope where Compute Library functions are available.
Knocks about 30seconds off a 4m30s single-threaded compile of the Neon workloads.
Change-Id: Idac438f3bc77ff978295fbc9505cb42447def145
|
|
*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
|
|
* Added Unit test
* Updated TensorFlowSupport.md file
Change-Id: I5f07de5e91ffb681c0ad17c7c73ee0326e7f1e0a
|
|
* 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
|
|
default cases are handled by DefaultLayerSupport
Change-Id: I31942a5e8515d6fdce9a0adf20906eaabc2b9aa6
|
|
StringMapping was located in the backendsCommon
Change-Id: I1c4a0bfca986722c721fcb2b460c03d6e1cd4568
|
|
Change-Id: Icf7ef88a2eaef80ec32cc718b0ca9d26e830ed07
|
|
Change-Id: Iae6a3bfad06cb669ef0abea465d03163bb8316f5
|
|
Factor out new BuildArmComputeReductionCoordinates function
from CL backend into ArmComputeTensorUtils.
Update NEON LayerSupport and WorkloadFactory objects
Change-Id: Icc975ec699199bffafbdb207323df509d35e1e04
|
|
* Allow control inputs from TensorFlow graph but ignore them in ArmNN graph.
* Add utility function to test ArmNN graph structure.
* Add ArmNN graph structure tests in TensorFlow paresr to ensure that control inputs
are ignored in ArmNN graph as well as their inputs that are not used anywhere else.
Change-Id: Ib0ea0d2df85e3fc79b748fa4c9d20e0649352bc1
|
|
Technical debt work towards adding some new Neon workloads
Change-Id: I08ab6dd14d0e89d4ebc8a878fb69caa5681012bf
|
|
* Unit tests in AddN.cpp
Change-Id: Ifb2fa1051d5d92c5d9a5ca751abee4e81ebe39c9
|
|
Change-Id: I297aac3c57d201fdbd71119f004a64f57ec5657f
|
|
Change-Id: Ibe1b27b268011878c7dce3c96efea01402453027
|
|
Change-Id: I74f0d27881fcc42d962fdd61562811a1f53dd0eb
|
|
* Add documentation to the Descriptors
* Add documentation to the layers
Change-Id: I5e0849753903565227fd47d329a600fd90b2feb9
|
|
* Unit tests in Greater.cpp
Change-Id: Ifb3e4c33be2d6235e33889bb63e6abd78bd7d8b6
|
|
!android-nn-driver:428
Change-Id: I42e59ad96d2c80f46b085182855d34b710a74dfe
|
|
* Add Arithmetic end-to-end test implementation
* Enable tests for float, Uint8 and Broadcast
Change-Id: I81c7096e9b6ad29eaa935b74ad5f30f823be2331
|
|
Change-Id: Iddae3486641d2d195827ae4e0a9dfa3f7ccd65e3
|
|
* Added CLMaximumWorkload implementation
* Added CL unit tests
Change-Id: I922c83fe25d17be21a5d0f4e861038463cb09789
|
|
* Unit tests in Equal.cpp
* Fixed error in Network::AddEqualLayer
* Refactored TfParser::Minimum/Equal to get rid of duplicate code
Change-Id: I0ed6f888eb391c995b88be20dc0c1b916dd14c3c
|
|
Change-Id: I045bc90124ea1303dffb0532e10b36ee107f9ee3
|
|
* 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
|
|
* Added ACl Implementation to ArmNN
* Added helper function for setting CL Strided Slice arguments
Change-Id: Ie10d387fdb054027ea9b8782743d270bb72949c1
|
|
Equal operation
* Add QueueDescriptor in WorkloadData.hpp
* Add CreateEqual function in WorkloadFactory.hpp
* Added stub implementation of the CreateEqual function in RefWorkloadFactory,
NeonWorkloadFactory and ClWorkloadFactory
Change-Id: Iec6dc2f989c67fa6f0c32cfb93508995c8580783
|
|
Change-Id: I56f56398e7c2c2ac57a96952237e7f1bc0ced208
|
|
Change-Id: If2a31a49df3701877ce0287a81c569334a24cd20
|
|
Change-Id: Ib3c7cfccbe8933a0c2423f37c363926571968d9e
|
|
per COMPMID-1818 change to using Blob-based memory manager.
Will change back when COMPMID-1848 is fixed.
Change-Id: I6e22c2edf422202410faa717b330a2c96c3ebbe5
|
|
Greater operation
* Added QueueDescriptor in WorkloadData.hpp
* Added CreateGreater function in WorkloadFactory.hpp
* Added stub implementation of the CreateGreater function in RefWorkloadFactory,
NeonWorkloadFactory and ClWorkloadFactory
* Added GreaterLayer stub implementation
* Renamed ArithmeticBaseLayer to ElementwiseBaseLayer
Change-Id: I7e38c2936de905da921a92ba3f918478169ec7f5
|
|
Change-Id: Ifd28e049192e6f5fe5c0f5d358afb2b530eef882
|
|
Change-Id: I9f9852766e4c51151d6ccf34976c4860a83dfe5c
|
|
* Modified optimizer to support debug mode via DebugLayer
Change-Id: Ic8f313778e55540c182cf99876c44a0823be04c6
|
|
* A recent change in ONNX now forces the data types to be indicated by the
integer corresponding to their original enum value
* The new values are taken from the TensorProto_DataType enum at onnx.pb.h:153
Change-Id: I6054611795a7edb11463396eaefc4dd7e27261a0
|
|
Change-Id: I72d94fff4cdad2c62dff98c8fd52eba78a1908f0
|
|
* Fixed invalid conversion issues to onnx::TensorProto::DataType
Change-Id: If8f1cef1d02a22cfecf24135a960923d3ad0e14c
|
|
Kernel position within the layer is part of the kernel name
Change-Id: I3b9ee870fbd0db025f1515c81245f917189ae884
|
|
Change-Id: I0d69a72df9e348cbeca690926485c050499d1051
|
|
Unit tests not yet added as need Compute Library a84faffd.
Change-Id: Ica16df493e8d6a76da9d1f74bf43b8403f9dff62
|
|
NEFillBorderKernel may not be issued for some activation workloads
depending on the version of the Compute Library.
Change-Id: I01b8960f9eb7a0192cd351eddbb416ac05248b34
|
|
* Removed the input swizzling when the concatenation dimension is 3
in ParseConcat in the TF parser
* No longer using the helper ProcessConcatInputTensorInfo, where
the input was being swizzled if the concatenation dimension was 3
* Added a new convenience constuctor to TensorShape that initializes
a shape to all zeros given only the number of dimensions
Change-Id: I82a207e41bddc5fea21a0b5a38eafa24ad75d1c2
|