Christoph Honal

Using the latest version of emscripten, I managed to compile LAPACK and BLAS for complex numbers, and they seem to work. Basic complex arithmetic also works. However, Fortran intrinsics such...

@tmm1 Thank you for testing! I can reproduce all the errors you are seeing. I suspect the LLVM toolchain was upgraded in Emscripten 2.0.27 in such a way that it...

Which data layout did you end up using? I remember having issues with libpthread and libatomic in the past, these usually only concern multi-threaded systems. However, Emscripten requires an explicit...

Adding `libatomic1` to the packages installed via `apt-get` in the Dockerfile (`tool` stage) got rid of the error for me.

Compiling using 2.0.32 and these fixes now generates a few new warnings: ``` emcc -O3 -m32 -Wall -flto -r -o bin/assembly.bc bin/iolib/iolib.c.bc bin/iolib/iolib.f90.bc bin/test.f90.bc bin/lapack_test/test_eigen.f90.bc bin/lapack_test/lapack_prb.f90.bc /app/lib/libgfortran.a /app/lib/liblapack.a /app/lib/librefblas.a emcc:...

Yea, this definitely looks like a LLVM IR compatibility issue. According to https://llvm.org/docs/OpaquePointers.html, LLVM 15 still defaults to using typed pointers, which LLVM 3.3 generates. I assume Emscripten does not...

Thank you for your effort, the upgrade is much appreciated! Your PR looks good, I'll investigate a bit into https://github.com/StarGate01/Full-Stack-Fortran/issues/10#issuecomment-1121616448 .

Thanks for the research, you are correct. The warning is gone.

I managed to fix the empty object file warnings by linking the object files the proper recommended way, i.e. by first creating an archive. This removes the unnecessary objects and...

Yes, I think there were some bitcode changes. I tried writing a converter which upgrades the textual IR representation to the latest version, but there were quite a lot non-triveal...