Replace make_tables by C++11 constexpr arrays

[derselbst]

Bye bye make_tables. This PR proposes to replace the auto-generated lookup tables (using the hacky ExternalProject CMake approach) with C++11 lookup tables.

References gcem as submodule for C++11 constexpr math, making it necessary to update all CI pipeline to check out submodules.

When this PR is merged, a C++11 compliant compiler will be required to compile fluidsynth!

Currently, this PR is in a PoC state as it's incomplete:

• Port the 2D DSP lookup tables
• Compare the contents of the new array with the old array.

Yet, it compiles fine on all platforms and passes the unit tests. For the latter point, I wrote a unit test to compare with the old lookup tables. It passed with a maximum difference of 1e-4. Since I don't want to add 8000 lines of code, I'll revert it again, as the C++ lookuptable implementation is unlikely to change. For future reference, the commit was b263b24.

[komh]

Hi/2.

I've encountered the following error on OS/2:

[ 10%] Building CXX object src/CMakeFiles/libfluidsynth-OBJ.dir/gentables/fluid_ct2hz.cpp.o
F:/lang/work/fluidsynth/fluidsynth.git/src/gentables/fluid_ct2hz.cpp:5:10: fatal error: gcem.hpp: No such file or directory
    5 | #include "gcem.hpp"
      |          ^~~~~~~~~~
compilation terminated.
make[2]: *** [src/CMakeFiles/libfluidsynth-OBJ.dir/build.make:152: src/CMakeFiles/libfluidsynth-OBJ.dir/gentables/fluid_ct2hz.cpp.o] Error 1
make[2]: *** Waiting for unfinished jobs....

[derselbst]

Looks like you forgot to run git submodule update --init

[komh]

Thanks, it fixed. This step is automated later ?

Anyway, unfortunately, this pr does not produce any sounds at all on my machine.

[derselbst]

Thanks, it fixed. This step is automated later ?

No. I have never seen this automated via e.g. hooks or CMake. It's the developers responsibility to deal with the repository and its content. Triggering git submodule update automatically might unintentionally override / revert changes made to the submodule.

Since the content of the gcem submodule is unlikely to change, using git subtree might be a reasonable alternative. git submodule would still be used for versioning the testdata of fluidsynth's manual testsuite. So for me personally it wouldn't make a difference. If I ever figure out a reliable way to automatically validate the rendering results of the manual testsuite, I would still have to enable submodule checkout in all CI pipelines.

unfortunately, this pr does not produce any sounds at all on my machine

I will investigate next weekend. Yet I'm glad it compiles for OS/2 as well :)

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
30.8% Coverage on New Code
1.5% Duplication on New Code

See analysis details on SonarQube Cloud

[white-axe]

Is there a way to do this without the compile-time long double arithmetic in fluid_cb2amp.cpp, fluid_concave.cpp, fluid_convex.cpp and fluid_pan.cpp? Why not just use regular double? This pull request has caused FluidSynth to fail to build with GCC on PowerPC targets that use the IBM floating-point ABI, such as Debian and Ubuntu, because GCC doesn't have compile-time long double division implemented for this ABI. Clang still works, though.

[derselbst]

Thanks for the hint. Using long double is a leftover from #512. It at least compiles fine for OpenSuse ppc64le. I see no problem to demote it to double now, PR welcome. Since we have no CI build for your particular target platform and ABI, we cannot guarantee that fluidsynth will compile fine. Feel free to contribute a CI build as well.