277 switch between double and float

[ZhenxiZhao]

switch precision TODO: simbody and eigen

[DrChiZhang]

@ZhenxiZhao I have went through the modifications you have made in this pull request, and they are seems OK. Would you please update me of the existing issues of this improvement?

[ZhenxiZhao]

@ZhenxiZhao I have went through the modifications you have made in this pull request, and they are seems OK. Would you please update me of the existing issues of this improvement?

All the constants in the project (not variable, for instance, 1.0) should be rewite as Real(1.0) or ::One(). One related error is shown below,

/home/SPHinXsys/SPHINXsys/src/shared/materials/riemann_solver.cpp:30:77: error: no matching function for call to ‘SMAX(SPH::Real, double)’
   30 |   return rho0c0_geo_ave_ * u_jump * SMIN(3.0 * SMAX(u_jump * inv_c_ave_, 0.0), 1.0);

There are many cast type conversion in the project, for instance,

    template <class DampingAlgorithmType>
    bool DampingWithRandomChoice<DampingAlgorithmType>::RandomChoice()
    {
        return ((double)rand() / (RAND_MAX)) < random_ratio_ ? true : false;
    }

Besides, the words "SimTk::xxx" should be replaced by "SimTKxxx", for instance, on line 43, solid_body_supplementary.cpp

            body_part_volume * solid_body_density_, SimTK::Vec3(0), SimTK::UnitInertia(Ix, Iy, Iz));

[DrChiZhang]

Note that with clang and clang++ compiler, non-constexpr function 'sqrt' cannot be used in a constant expression, I will fix this later.

[Xiangyu-Hu]

@ZhenxiZhao I have went through the modifications you have made in this pull request, and they are seems OK. Would you please update me of the existing issues of this improvement?

All the constants in the project (not variable, for instance, 1.0) should be rewite as Real(1.0) or ::One(). One related error is shown below,

/home/SPHinXsys/SPHINXsys/src/shared/materials/riemann_solver.cpp:30:77: error: no matching function for call to ‘SMAX(SPH::Real, double)’
   30 |   return rho0c0_geo_ave_ * u_jump * SMIN(3.0 * SMAX(u_jump * inv_c_ave_, 0.0), 1.0);

There are many cast type conversion in the project, for instance,

  template <class DampingAlgorithmType>
  bool DampingWithRandomChoice<DampingAlgorithmType>::RandomChoice()
  {
      return ((double)rand() / (RAND_MAX)) < random_ratio_ ? true : false;
  }

Besides, the words "SimTk::xxx" should be replaced by "SimTKxxx", for instance, on line 43, solid_body_supplementary.cpp

          body_part_volume * solid_body_density_, SimTK::Vec3(0), SimTK::UnitInertia(Ix, Iy, Iz));

Yes. Seems that we need do it.

[Xiangyu-Hu]

Actually, using replace function in VS code, such replacement is not hard.

[DrChiZhang]

First Step: test with Real = double by doing the following steps. 1, DoubleVec -> BiVector 2, TripleVec -> TriVector 3, CaculateDoubleDotProduct -> CaculateBiDotProduct (clean the variable name with "double", and change all double to Real in code) 4, double -> Real 5, discards changes in polar_docompostion.

[Xiangyu-Hu]

@ZhenxiZhao could you pull from the main repository to solve the conflicts?

[DrChiZhang]

@ZhenxiZhao @Xiangyu-Hu I have moved the modification of branch "[ZhenxiZhao:277-switch-between-double-and-float]" to branch"[Xiangyu-Hu:277-switch-between-double-and-float]" , which we can continue this implementation.

[DrChiZhang]

@Xiangyu-Hu Do I have the access to push to this branch "[Xiangyu-Hu:277-switch-between-double-and-float]" ?

[DrChiZhang]

@ZhenxiZhao I encounter this error after setting "SimTk::xxx" replaced by "SimTKxxx"

According to Simbody documentation

Does this mean that we should build Simbody with float as the default precision or add another transfer form float vector to double vector there ?

[ZhenxiZhao]

@ZhenxiZhao I encounter this error after setting "SimTk::xxx" replaced by "SimTKxxx"

According to Simbody documentation

Does this mean that we should build Simbody with float as the default precision or add another transfer form float vector to double vector there ?

Yes, I attenmpt to use the fVec or Vec explicitly, since the type is pre-defined before compling simbody in vcpkg. However, the class or struct in simbody use Vec, which has been pre-defined as double. Thus we encounter this error. I think if we can dual compile simbody to simply implement the type switch, one for float, one for double.

[DrChiZhang]

@ZhenxiZhao I encounter this error after setting "SimTk::xxx" replaced by "SimTKxxx" According to Simbody documentation

Does this mean that we should build Simbody with float as the default precision or add another transfer form float vector to double vector there ?

Yes, I attenmpt to use the fVec or Vec explicitly, since the type is pre-defined before compling simbody in vcpkg. However, the class or struct in simbody use Vec, which has been pre-defined as double. Thus we encounter this error. I think if we can dual compile simbody to simply implement the type switch, one for float, one for double.

That would introduce extra efforts for the installation. I am thinking that we can use double precision in SIMBODY, while transfer single precision of Eigen vector to double vector of SIMBODY if needed.

[FabienPean-Virtonomy]

I am thinking that we can use double precision in SIMBODY, while transfer single precision of Eigen vector to double vector of SIMBODY if needed.

I agree with this. Simbody needs far less state than a SPH, so preserving double there wouldn't have a big impact and would preserve FP accuracy. Also doing so does not increase the coupling between the two libraries.

ideally use of Simbody is moved to an optional module (i.e. CMake target based with a user-facing option)

[DrChiZhang]

@FabienPean-Virtonomy Thanks for the suggestion.

[DrChiZhang]

@Xiangyu-Hu @ZhenxiZhao Implementation done, while the following tests fails when using float due to the large error.

10:test_2d_dambreak_python 13:test_2d_elastic_gate 16:test_2d_eulerian_taylor_green 30:test_2d_self_contact 31:test_2d_shell 43:test_2d_oscillating_beam_UL 46:test_2d_stfb 47:test_2d_stlw 53:test_vector_functions_particle_relaxation 58:bernoulli_beam_struct_sim 76:test_3d_passive_cantilever 79:test_3d_roof 80:roof_under_self_weight.dp_4 83:test_3d_roof_parametric_cvt 90:test_3d_twisting_column

[Xiangyu-Hu]

this branch stop to use as another branch on the same function has been merged.