ARM-software
Arm NN 22.02 fails to build with flatbuffers 2.0.6
Arm NN 22.02 fails to build on openSUSE Tumbleweed with TensorFlow-Lite 2.7.1 and flatbuffers 2.0.6.
Logs:
[ 116s] [ 53%] Building CXX object src/armnnSerializer/CMakeFiles/armnnSerializer.dir/Serializer.cpp.o
[ 116s] cd /home/abuild/rpmbuild/BUILD/armnn-22.02/build/src/armnnSerializer && /usr/bin/c++ -DARMNNREF_ENABLED -DARMNN_SERIALIZER -DARMNN_SERIALIZER_SCHEMA_PATH=\"/home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/ArmnnSchema.fbs\" -DARMNN_TF_LITE_PARSER -DDYNAMIC_BACKEND_BUILD_DIR=\"/home/abuild/rpmbuild/BUILD/armnn-22.02/build\" -DarmnnSerializer_EXPORTS -I/home/abuild/rpmbuild/BUILD/armnn-22.02/include -I/home/abuild/rpmbuild/BUILD/armnn-22.02/profiling -I/home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/../armnn -I/home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/../armnnUtils -isystem /home/abuild/rpmbuild/BUILD/armnn-22.02/third-party -isystem /home/abuild/rpmbuild/BUILD/armnn-22.02/build/src/armnnSerializer -O2 -Wall -D_FORTIFY_SOURCE=2 -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clash-protection -Werror=return-type -g -pthread -std=c++14 -Wall -Wextra -Werror -Wold-style-cast -Wno-missing-braces -Wconversion -Wsign-conversion -Wno-psabi -O2 -g -fPIC -MD -MT src/armnnSerializer/CMakeFiles/armnnSerializer.dir/Serializer.cpp.o -MF CMakeFiles/armnnSerializer.dir/Serializer.cpp.o.d -o CMakeFiles/armnnSerializer.dir/Serializer.cpp.o -c /home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/Serializer.cpp
[ 116s] In file included from /home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/Serializer.hpp:17,
[ 116s] from /home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/Serializer.cpp:5:
[ 116s] /home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/ArmnnSchema_generated.h: In member function 'bool armnnSerializer::TensorInfo::Verify(flatbuffers::Verifier&) const':
[ 116s] /home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/ArmnnSchema_generated.h:1812:31: error: no matching function for call to 'armnnSerializer::TensorInfo::VerifyField<int8_t>(flatbuffers::Verifier&, armnnSerializer::TensorInfo::FlatBuffersVTableOffset) const'
[ 116s] 1812 | VerifyField<int8_t>(verifier, VT_DATATYPE) &&
[ 116s] | ~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~
[ 116s] In file included from /usr/include/flatbuffers/flatbuffer_builder.h:31,
[ 116s] from /usr/include/flatbuffers/flatbuffers.h:27,
[ 116s] from /home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/ArmnnSchema_generated.h:11:
[ 116s] /usr/include/flatbuffers/table.h:115:8: note: candidate: 'template<class T> bool flatbuffers::Table::VerifyField(const flatbuffers::Verifier&, flatbuffers::voffset_t, size_t) const'
[ 116s] 115 | bool VerifyField(const Verifier &verifier, voffset_t field,
[ 116s] | ^~~~~~~~~~~
[ 116s] /usr/include/flatbuffers/table.h:115:8: note: candidate expects 3 arguments, 2 provided
[ 116s] /home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/ArmnnSchema_generated.h:1813:30: error: no matching function for call to 'armnnSerializer::TensorInfo::VerifyField<float>(flatbuffers::Verifier&, armnnSerializer::TensorInfo::FlatBuffersVTableOffset) const'
[ 116s] 1813 | VerifyField<float>(verifier, VT_QUANTIZATIONSCALE) &&
[ 116s] | ~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[ 116s] /usr/include/flatbuffers/table.h:115:8: note: candidate expects 3 arguments, 2 provided
[ 116s] /home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/ArmnnSchema_generated.h: In member function 'bool armnnSerializer::BindableLayerBase::Verify(flatbuffers::Verifier&) const':
[ 116s] /home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/ArmnnSchema_generated.h:2448:32: error: no matching function for call to 'armnnSerializer::BindableLayerBase::VerifyField<int32_t>(flatbuffers::Verifier&, armnnSerializer::BindableLayerBase::FlatBuffersVTableOffset) const'
[ 116s] 2448 | VerifyField<int32_t>(verifier, VT_LAYERBINDINGID) &&
[ 116s] | ~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[ 116s] /usr/include/flatbuffers/table.h:115:8: note: candidate expects 3 arguments, 2 provided
[ 116s] /home/abuild/rpmbuild/BUILD/armnn-22.02/src/armnnSerializer/ArmnnSchema_generated.h:3220:33: error: no matching function for call to 'armnnSerializer::Convolution2dDescriptor::VerifyField<uint32_t>(flatbuffers::Verifier&, armnnSerializer::Convolution2dDescriptor::FlatBuffersVTableOffset) const'
[ 116s] 3220 | VerifyField<uint32_t>(verifier, VT_STRIDEY) &&
[ 116s] | ~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~
And more errors of this kind.
This seems similar to https://github.com/RPCS3/rpcs3/issues/11713 which was fixed with https://github.com/RPCS3/rpcs3/pull/11870 which adds a 3rd argument.
Hi @ggardet
we currently have an version of ArmnnSchema_generated.h that was created by flatbuffers v1.12.0 checked into which is not overwritten when armnn is built. I'm getting a patch together to hopefully resolve this.
With the patches above, I still get build issues:
[ 188s] [ 51%] Building CXX object src/armnnTfLiteParser/CMakeFiles/armnnTfLiteParser.dir/TfLiteParser.cpp.o
[ 188s] cd /home/abuild/rpmbuild/BUILD/armnn-22.05/build/src/armnnTfLiteParser && /usr/bin/c++ -DARMNNREF_ENABLED -DARMNN_SERIALIZER -DARMNN_SERIALIZER_SCHEMA_PATH=\"/home/abuild/rpmbuild/BUILD/armnn-22.05/src/armnnSerializer/ArmnnSchema.fbs\" -DARMNN_TF_LITE_PARSER -DDYNAMIC_BACKEND_BUILD_DIR=\"/home/abuild/rpmbuild/BUILD/armnn-22.05/build\" -DarmnnTfLiteParser_EXPORTS -I/home/abuild/rpmbuild/BUILD/armnn-22.05/include -I/home/abuild/rpmbuild/BUILD/armnn-22.05/profiling -I/home/abuild/rpmbuild/BUILD/armnn-22.05/src/armnnTfLiteParser/../armnn -I/home/abuild/rpmbuild/BUILD/armnn-22.05/src/armnnTfLiteParser/../armnnUtils -isystem /home/abuild/rpmbuild/BUILD/armnn-22.05/third-party -isystem /usr/include/tensorflow/lite/schema -O2 -Wall -U_FORTIFY_SOURCE -D_FORTIFY_SOURCE=3 -fstack-protector-strong -funwind-tables -fasynchronous-unwind-tables -fstack-clash-protection -Werror=return-type -g -pthread -std=c++14 -Wall -Wextra -Werror -Wold-style-cast -Wno-missing-braces -Wconversion -Wsign-conversion -Wno-psabi -O2 -g -fPIC -Wno-conversion -Wno-sign-conversion -Wno-unused-parameter -Wno-unused-function -MD -MT src/armnnTfLiteParser/CMakeFiles/armnnTfLiteParser.dir/TfLiteParser.cpp.o -MF CMakeFiles/armnnTfLiteParser.dir/TfLiteParser.cpp.o.d -o CMakeFiles/armnnTfLiteParser.dir/TfLiteParser.cpp.o -c /home/abuild/rpmbuild/BUILD/armnn-22.05/src/armnnTfLiteParser/TfLiteParser.cpp
[ 188s] In file included from /home/abuild/rpmbuild/BUILD/armnn-22.05/src/armnnTfLiteParser/TfLiteParser.hpp:11,
[ 188s] from /home/abuild/rpmbuild/BUILD/armnn-22.05/src/armnnTfLiteParser/TfLiteParser.cpp:6:
[ 188s] /usr/include/tensorflow/lite/schema/schema_generated.h: In member function 'bool tflite::QuantizationParameters::Verify(flatbuffers::Verifier&) const':
[ 188s] /usr/include/tensorflow/lite/schema/schema_generated.h:3406:32: error: no matching function for call to 'tflite::QuantizationParameters::VerifyField<uint8_t>(flatbuffers::Verifier&, tflite::QuantizationParameters::FlatBuffersVTableOffset) const'
[ 188s] 3406 | VerifyField<uint8_t>(verifier, VT_DETAILS_TYPE) &&
[ 188s] | ~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~
[ 188s] In file included from /usr/include/flatbuffers/flatbuffer_builder.h:31,
[ 188s] from /usr/include/flatbuffers/flatbuffers.h:27,
[ 188s] from /usr/include/tensorflow/lite/schema/schema_generated.h:21:
[ 188s] /usr/include/flatbuffers/table.h:115:8: note: candidate: 'template<class T> bool flatbuffers::Table::VerifyField(const flatbuffers::Verifier&, flatbuffers::voffset_t, size_t) const'
[ 188s] 115 | bool VerifyField(const Verifier &verifier, voffset_t field,
[ 188s] | ^~~~~~~~~~~
[ 188s] /usr/include/flatbuffers/table.h:115:8: note: candidate expects 3 arguments, 2 provided
[ 188s] /usr/include/tensorflow/lite/schema/schema_generated.h:3409:32: error: no matching function for call to 'tflite::QuantizationParameters::VerifyField<int32_t>(flatbuffers::Verifier&, tflite::QuantizationParameters::FlatBuffersVTableOffset) const'
[ 188s] 3409 | VerifyField<int32_t>(verifier, VT_QUANTIZED_DIMENSION) &&
[ 188s] | ~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[ 188s] /usr/include/flatbuffers/table.h:115:8: note: candidate: 'template<class T> bool flatbuffers::Table::VerifyField(const flatbuffers::Verifier&, flatbuffers::voffset_t, size_t) const'
[ 188s] 115 | bool VerifyField(const Verifier &verifier, voffset_t field,
[ 188s] | ^~~~~~~~~~~
[ 188s] /usr/include/flatbuffers/table.h:115:8: note: candidate expects 3 arguments, 2 provided
[ 188s] /usr/include/tensorflow/lite/schema/schema_generated.h: In member function 'bool tflite::DimensionMetadata::Verify(flatbuffers::Verifier&) const':
[ 188s] /usr/include/tensorflow/lite/schema/schema_generated.h:3756:31: error: no matching function for call to 'tflite::DimensionMetadata::VerifyField<int8_t>(flatbuffers::Verifier&, tflite::DimensionMetadata::FlatBuffersVTableOffset) const'
[ 188s] 3756 | VerifyField<int8_t>(verifier, VT_FORMAT) &&
[ 188s] | ~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~
[ 188s] /usr/include/flatbuffers/table.h:115:8: note: candidate: 'template<class T> bool flatbuffers::Table::VerifyField(const flatbuffers::Verifier&, flatbuffers::voffset_t, size_t) const'
[ 188s] 115 | bool VerifyField(const Verifier &verifier, voffset_t field,
[ 188s] | ^~~~~~~~~~~
[ 188s] /usr/include/flatbuffers/table.h:115:8: note: candidate expects 3 arguments, 2 provided
[ 188s] /usr/include/tensorflow/lite/schema/schema_generated.h:3757:32: error: no matching function for call to 'tflite::DimensionMetadata::VerifyField<int32_t>(flatbuffers::Verifier&, tflite::DimensionMetadata::FlatBuffersVTableOffset) const'
[ 188s] 3757 | VerifyField<int32_t>(verifier, VT_DENSE_SIZE) &&
[ 188s] | ~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
[ 188s] /usr/include/flatbuffers/table.h:115:8: note: candidate: 'template<class T> bool flatbuffers::Table::VerifyField(const flatbuffers::Verifier&, flatbuffers::voffset_t, size_t) const'
[ 188s] 115 | bool VerifyField(const Verifier &verifier, voffset_t field,
[ 188s] | ^~~~~~~~~~~
[ 188s] /usr/include/flatbuffers/table.h:115:8: note: candidate expects 3 arguments, 2 provided
[ 188s] /usr/include/tensorflow/lite/schema/schema_generated.h:3758:32: error: no matching function for call to 'tflite::DimensionMetadata::VerifyField<uint8_t>(flatbuffers::Verifier&, tflite::DimensionMetadata::FlatBuffersVTableOffset) const'
[ 188s] 3758 | VerifyField<uint8_t>(verifier, VT_ARRAY_SEGMENTS_TYPE) &&
[ 188s] | ~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
And more.
It seems to use /usr/include/tensorflow/lite/schema/schema_generated.h now, which is not working either.
Any suggestion?
Hi @ggardet
That's the TFLite schema which is used for reading TFLite models. It looks like an incompatability between the version of TFLite and Flatbuffers. I'll take a look.
Hi @ggardet
That tensorflow lite schema is checked into the tensorflow repository and is not generated by us. However, 3 weeks ago the tensorflow project updated their version of flatbuffers to 2.0.6 with https://github.com/tensorflow/tensorflow/commit/bf0901b663e87f081c7a317fe1ac705ab1a2df8b which updated the schema_generated.h file shipped with tensorflow.
The function calls in the updated schema_generated.h have been changed from (for example):
VerifyField<uint8_t>(verifier, VT_ARRAY_SEGMENTS_TYPE) &&
To:
VerifyField<uint8_t>(verifier, VT_ARRAY_SEGMENTS_TYPE, 1) &&
The last tensorflow release was on May 23, 2022 so that has not included those changes but the next one hopefully should.
Best regards, Mike
Ok, so IIUC, I need to backport https://github.com/tensorflow/tensorflow/commit/bf0901b663e87f081c7a317fe1ac705ab1a2df8b to my TensorFlow lite package and the problem should be gone?
That may be enough looking at the changes made to schema_generated.h
However there was also this commit: https://github.com/tensorflow/tensorflow/commit/625a4045bc0728c0f3d1b63e05749201f8b401dd which was when they updated their version of flatbuffers to 2.0.5 so you may need to backport that too.
Hi @ggardet,
Since 22.11 release we update to TensorFlow 2.10 and Flatbuffers 2.0.6.
As a result I believe this issue can be closed, if not please reopen.
Regards, Francis.
