Code generation issue with inlining

[iomaganaris]

Installing NEURON with NMODL on Apple Macbook Air with M1 processor generated the following issue when compiling the na3n.mod file from the reduced_dentate repo:

arm64/corenrn/mod2c/na3n.cpp:544:32: error: use of undeclared identifier 'bets_in_0'
                exptrap_in_4 = bets_in_0;

Seems like there is some issue with the code generated in

exptrap_in_4 = bets_in_0;

which should be:

if (x_in_4 >= 700.0) {
    exptrap_in_4 = exp(700.0);
} else {
    exptrap_in_4 = exp(x_in_4);
}

after the inline pass.
Unfortunately I couldn't reproduce the issue when I installed NEURON with CoreNEURON and NMODL using the following commands:

git clone --recursive https://github.com/neuronsimulator/nrn.git
cd nrn
mkdir build_nmodl
python3 -v venv venv
. venv/bin/activate
pip3 install -r nrn_requirements.txt
pip3 install -U pip setuptools scikit-build Jinja2 PyYAML pytest 'sympy>=1.3,<1.9' 'cmake-format==0.6.13'  # NMODL requirements
brew install flex bison
export PATH="/opt/homebrew/opt/bison/bin:/opt/homebrew/opt/flex/bin:$PATH"
export PYTHONPATH="$(pwd)/build_nmodl/install/lib/python:$PYTHONPATH"
export SDKROOT=$(xcrun --sdk macosx --show-sdk-path)
cmake .. -DCMAKE_INSTALL_PREFIX=./install -DNRN_ENABLE_CORENEURON=ON -DCORENRN_ENABLE_NMODL=ON -DNRN_ENABLE_RX3D=OFF -DNRN_ENABLE_INTERVIEWS=OFF -DNRN_ENABLE_TESTS=ON
cmake --build . --target install --parallel 1
ctest -R reduced_dentate

[iomaganaris]

I got a similar issue again when building NEURON with NMODL enabled in M1 but this time in a different mod file:

[NMODL] [warning] :: Python bindings are not available
arm64/corenrn/mod2c/tca.cpp:283:17: error: use of undeclared identifier 'exptrap_in_5'; did you mean 'exptrap_in_4'?
            if (exptrap_in_5 < 1e-4) {
                ^~~~~~~~~~~~
                exptrap_in_4
arm64/corenrn/mod2c/tca.cpp:262:33: note: 'exptrap_in_4' declared here
        double nu, f, KTF_in_0, exptrap_in_4, efun_in_0;
                                ^
1 error generated.
make: *** [arm64/corenrn/build/tca.o] Error 1

[iomaganaris]

Seems like after 2109df6 this issue is somehow fixed

Edit: After this commit the mod files included in NEURON integrated tests pass however I still go the error in another mod file with current NMODL master

[iomaganaris]

@alkino Have you maybe tested whether this fixes the issue in an M1 machine?

[iomaganaris]

It seems like the issue is still not resolved. nmodl still creates wrong output in inlined variables

[iomaganaris]

Some more instructions to reproduce the issue in M1 processors:

python3 -v venv venv
. venv/bin/activate
pip3 install -U pip setuptools scikit-build Jinja2 PyYAML pytest 'sympy>=1.3,<1.9' 'cmake-format==0.6.13'  # NMODL requirements
brew install flex bison
export PATH="/opt/homebrew/opt/bison/bin:/opt/homebrew/opt/flex/bin:$PATH"
export SDKROOT=$(xcrun --sdk macosx --show-sdk-path)
export BASE_DIR=$(pwd)
git clone https://github.com/neuronsimulator/reduced_dentate.git
git clone https://github.com/BlueBrain/nmodl.git
cd nmodl
mkdir build
cd build
cmake .. -DCMAKE_INSTALL_PREFIX=./install
cmake --build . --target install --parallel
cd ../..
git clone https://github.com/BlueBrain/CoreNeuron.git
cd CoreNeuron
mkdir build
cd build
cmake .. -DCMAKE_INSTALL_PREFIX=./install -DCORENRN_NMODL_DIR=${BASE_DIR}/nmodl/build/install -DCORENRN_ENABLE_NMODL=ON
cmake --build . --target install --parallel
cd ..
for i in {0..19}; do rm -rf arm64 && build/install/bin/nrnivmodl-core ${BASE_DIR}/reduced_dentate/mechanisms || break; done

Above script is installing NMODL and CoreNEURON and compiles the mod files from reduced dentate that showcase this issue multiple times until there is a failure.
With the above workflow it's easy to do some change in NMODL, then rebuild and install it and nrnivmodl-core will then use the updated NMODL executable

[pramodk]

I don't think this happens on Apple M1 only! today on my Intel MacBook I saw:

[NMODL] [info] :: Running cnexp visitor
[NMODL] [info] :: Running C backend code generator
x86_64/corenrn/mod2c/Aradi_Ca.cpp:560:36: error: use of undeclared identifier 'exponential_in_2'; did you mean 'exptrap_in_2'?
                    exptrap_in_2 = exponential_in_2;
                                   ^~~~~~~~~~~~~~~~
                                   exptrap_in_2
x86_64/corenrn/mod2c/Aradi_Ca.cpp:550:20: note: 'exptrap_in_2' declared here
            double exptrap_in_2, A_in_7, k_in_7, v_in_20, D_in_7;
                   ^
x86_64/corenrn/mod2c/Aradi_Ca.cpp:1482:48: error: use of undeclared identifier 'exponential_in_2'; did you mean 'exptrap_in_2'?
                                exptrap_in_2 = exponential_in_2;
                                               ^~~~~~~~~~~~~~~~
                                               exptrap_in_2
x86_64/corenrn/mod2c/Aradi_Ca.cpp:1472:32: note: 'exptrap_in_2' declared here
                        double exptrap_in_2, A_in_7, k_in_7, v_in_20, D_in_7;
                               ^
2 errors generated.
make[3]: *** [x86_64/corenrn/build/Aradi_Ca.o] Error 1
make[3]: *** Waiting for unfinished jobs....

The NMODL version:

✗ ./bin/nmodl --help
NMODL : Source-to-Source Code Generation Framework [0.4 14359293 2022-10-28 08:35:35 +0200]

And CMake command I used:

cmake .. -DNRN_ENABLE_INTERVIEWS=OFF -DNRN_ENABLE_RX3D=OFF -DCMAKE_INSTALL_PREFIX=`pwd`/install -DNRN_ENABLE_CORENEURON=ON -DCORENRN_ENABLE_LEGACY_UNITS=ON -DNRN_DYNAMIC_UNITS_USE_LEGACY=ON -DCORENRN_ENABLE_NMODL=ON -DCORENRN_ENABLE_OPENMP=OFF -DNRN_ENABLE_TESTS=ON

Tried to build again and I see:

[NMODL] [info] :: Running C backend code generator
x86_64/corenrn/mod2c/Aradi_Ca.cpp:400:36: error: use of undeclared identifier 'alphaa_in_1'
                    exptrap_in_1 = alphaa_in_1;
                                   ^
x86_64/corenrn/mod2c/Aradi_Ca.cpp:1262:48: error: use of undeclared identifier 'alphaa_in_1'
                                exptrap_in_1 = alphaa_in_1;
                                               ^