Hi!
I deployed the edge impulse “Tutorial: Continuous motion recognition” example as a C++ library and tried to import it and compile it in MPLABX with free compiler XC32 v3.01 from microchip for the ATSAML10E16A microcontroller. I followed the guide: https://microchipdeveloper.com/machine-learning:porting-edge-impulse-sdk
But when I finish configuration of the project and without calling any function of the edge impulse SDK in the main.cpp I get the following compilation error:
…/src/edge-impulse-sdk/CMSIS/DSP/Include/arm_math_memory.h:61:33: error: dereferencing type-punned pointer will break strict-aliasing rules [-Werror=strict-aliasing] #define __SIMD32(addr) (*(__SIMD32_TYPE **) & (addr))
I get the same error when I try to compile the “smartbell-SAMD21-BMI160” MPLABX example project.
Has anyone succesfully compiled an edge impulse c++ library for a microchip ARM microcontroller with XC32 compiler? Does anyone know why I get this compilation error?
Hi @jalberolalluch looks like you’re compiling with -Werror=strict-aliasing - is this a flag you see in your compiler settings? If not, can you post all flags that you see?
Hi, if you disable ‘Make warnings into errors’ for xc32-g++ this will work. But I think you might also be able to set -fno-strict-aliasing in the additional options for xc32-g++.
Hi @janjongboom, thanks for your help but I disabled the “Make warnings into errors” checkbox and I got the same error because it is not a warning but an error. Then I tried to set -fno-strict-aliasing` in the additional options for xc32-g++ but it did not work either.
The output of the compiler says:
…/src/edge-impulse-sdk/CMSIS/NN/Source/NNSupportFunctions/arm_q7_to_q15_reordered_no_shift.c: In function ‘arm_q7_to_q15_reordered_no_shift’:
In file included from …/src/edge-impulse-sdk/CMSIS/DSP/Include/dsp/fast_math_functions.h:30,
from …/src/edge-impulse-sdk/CMSIS/DSP/Include/arm_common_tables.h:33,
from …/src/edge-impulse-sdk/CMSIS/NN/Include/arm_nnsupportfunctions.h:33,
from …/src/edge-impulse-sdk/CMSIS/NN/Source/NNSupportFunctions/arm_q7_to_q15_reordered_no_shift.c:31:
…/src/edge-impulse-sdk/CMSIS/DSP/Include/arm_math_memory.h:61:33: error: dereferencing type-punned pointer will break strict-aliasing rules [-Werror=strict-aliasing] #define __SIMD32(addr) (*(__SIMD32_TYPE **) & (addr))
^~~~~~~~~~~~~~~~~~~~~~~~~~~
…/src/edge-impulse-sdk/CMSIS/NN/Source/NNSupportFunctions/arm_q7_to_q15_reordered_no_shift.c:106:10: note: in expansion of macro ‘__SIMD32’ __SIMD32(pDst)++ = in2;
^~~~~~~~
…/src/edge-impulse-sdk/CMSIS/DSP/Include/arm_math_memory.h:61:33: error: dereferencing type-punned pointer will break strict-aliasing rules [-Werror=strict-aliasing] #define __SIMD32(addr) ((__SIMD32_TYPE **) & (addr))
^~~~~~~~~~~~~~~~~~~~~~~~~~~
…/src/edge-impulse-sdk/CMSIS/NN/Source/NNSupportFunctions/arm_q7_to_q15_reordered_no_shift.c:107:10: note: in expansion of macro ‘__SIMD32’
*__SIMD32(pDst)++ = in1;
^~~~~~~~
cc1.exe: all warnings being treated as errors
make[2]: *** [build/default/production/_ext/711107984/arm_q7_to_q15_reordered_no_shift.o] Error 255
make[2]: *** Waiting for unfinished jobs…
make[1]: *** [.build-conf] Error 2
make: *** [.build-impl] Error 2
make[2]: Leaving directory ‘C:/Electronica/AI/firmware/AIcompilation.X’
nbproject/Makefile-default.mk:94: recipe for target ‘.build-conf’ failed
make[1]: Leaving directory ‘C:/Electronica/AI/firmware/AIcompilation.X’
nbproject/Makefile-impl.mk:39: recipe for target ‘.build-impl’ failed
I disabled the “Make warnings into errors” checkbox and I got the same error because it is not a warning but an error
In most compilers (all that I know of at least) this is a warning, and my bet was that it was turned into an error due to that checkbox.
Now that I realize this, a quick fix would just be disabling EIDSP_USE_CMSIS_DSP macro. You’re not on an ARM target anyway, and this will fall back to software kernels that don’t have any of this dereferencing behavior.
Hi @janjongboom
I am sorry but I removed the EIDSP_USE_CMSIS_DSP=1 macro and I still get the same warning/error.
My target by now is an ATSAML10E16A microcontroller from microchip which is an ARM cortex M23 (lowest power consumption family I believe).
Hi @jalberolalluch
I encountered the same error messages that you posted on Nov 2021 as the above.
Have you managed to solve that error? like to get help from you if you manage to solve it.
Do you get the support from Janjongboom, as he asked you to post your project to him?
