--- old/doc/building.md 2018-12-25 16:06:21.930966326 +0100 +++ new/doc/building.md 2018-12-25 16:06:21.562956937 +0100 @@ -527,6 +527,32 @@ Use `--with-libffi=` if `configure` does not properly locate your libffi files. +### SoftFloat + +[Berkeley SoftFloat-3](http://www.jhauser.us/arithmetic/SoftFloat.html) +can be used on ARM processors without FPU to slightly enhance +the arithmetic precision of some floating point operations. It is not +required, system softfp routines can be used without any problems. +The precision loss is extremely small, but [the JCK detects it]( +http://mail.openjdk.java.net/pipermail/aarch32-port-dev/2016-November/000611.html). + + * To build the library, you will have to download its source and build it + for the target platform. To do so, take a look in its + `build/Linux-ARM-VFPv2-GCC` subdirectory. + +You can enable this library by specifying a library prefix +via `--with-sflt=` or by specifying path to softfloat.a +via `--with-sflt-lib=` and path to directory +containing softfloat.h via `--with-sflt-include=`. You +will also need to specify path to the SoftFloat license file with +`--with-sflt-license=`. If you want to use the prefix option, ensure that +`/lib/softfloat.a`, `/include/softfloat.h` and +`/share/softfloat/softfloat.md` exist. However you can override +the defaults by using the options above. + +If you do not enable this library, standard system libraries +will be used instead. + ## Build Tools Requirements ### Autoconf @@ -694,6 +720,8 @@ * `--with-x=` - Set the path to [X11](#x11) * `--with-alsa=` - Set the path to [ALSA](#alsa) * `--with-libffi=` - Set the path to [libffi](#libffi) + * `--with-sflt=` - Enable and set the path to [SoftFloat](#softfloat) + library. * `--with-jtreg=` - Set the path to JTReg. See [Running Tests]( #running-tests)