Float 128 testing results

[FX]

> With your full_testing_2.diff patch:
> 
> gfortran appears to build OK. The following code compiles OK with no errors. Run time I get an internal error in write_float.def at line 1073.
> 
> Evidently HAVE_GFC_REAL_16 is not getting defined.  The variable len is being passed in OK with a value of 16
> 
> I noticed when libquad is installed, the libraries are being placed in /home/jerry/gcc/usr/lib.
> I configured with --prefix=/home/jerry/gcc/usr for the library build.
> I configured with --with-quad=/home/jerry/gcc/usr

That looks good. However, the library apparently is picked up by libgfortran configure. A few questions:

  1. is your libquad compiled as 64-bit or 32-bit (in the following, I assume it's 64-bit, which should be the default)

If your libquad is 64-bit, the 32-bit libgfortran will not work for real(kind=16)... giving exactly the kind of error you gave in your email. But the 64-bit libgfortran should work fine. In its build directory ($builddir/x86-64-linux-gnu/libgfortran/), look at config.h:

  2. Does it have HAVE_FLOAT128 defined? If not, check in the config.log why it was not defined

  3. Does it have GFC_WITH_QUAD_LIB defined? If not, check in config.log why.

Tell me what you got. If everything 

> Do I need --disable-multilib in the gcc build configure?


No you don't.

FX