Qualify RISC-V fenv against __riscv_flen to support soft float; Qualify testsuite/newlib.fenv against __riscv and __riscv_flen

 - All other ports use `configure.in` instead of `configure.ac`.

I've added an additional linker relaxation that relaxes two instruction
pc-relative sequences to one instruction gp relative sequences when
possible.  This sequence now optimizes the initial gp generation to

  mv gp, gp

which is obviously bogus.  The fix is to disable relaxations when
setting up gp, preventing the linker from relaxing away this setup code.

 - We don't want any padding in struct ldieee, otherwise the offset
   might wrong in most compiler.

This removes the dependence on malloc for minimal programs.

- -march controls which instructions may be emitted, and does not
affect ABI.
- -mno-float is equivalent to passing neither F for D to -march;
-msingle-float is equivalent to passing F but not D to -march;
and -mdouble-float is equivalent to passing F and D to -march.
- -mfloat-abi controls which calling convention is used: "soft" means
no args passed in registers; "single" means only single-precision
values are passed in regisers; "double" means single- and
double-precision values are passed in registers.
- -mfloat-abi defaults to "soft" if the D extension is not present,
or "double" if it is.
- GCC will issue an error if -mfloat-abi requires an extension not
provided by -march, but vice-versa is OK.

Closes #187 (I hope).

@kito-cheng can you test this and let me know what I need to fix?

Soft-float calling convention is default, but an alternate hard-float
calling convention (with doubles passed as in soft-float, but floats
passed as in hard-float) is supported.

There are no guarantees that the break going to be located exactly
at the _end, but brk(0) is guaranteed to return its current value.

QEMU ELF loader moves the initial break up to the next page boundary,
disabling sbrk-based malloc for any allocations smaller than
the adjustment made. And ASLR may do even worse.

https://sourceware.org/ml/newlib/2016/msg01139.html
https://gcc.gnu.org/ml/gcc/2016-12/msg00010.html



There is no change if libtool is used.

Some run-time support libraries provided by GCC (e.g. libgomp) use
configure checks to detect certain features, e.g. availability of
thread-local storage.  The configure script generates a test program and
tries to compile and link it.  It should use target libraries and
startfiles of the build tree if available and not random ones from the
installation prefix for this procedure.  The search directories
specified by -B are a bit special, see for_each_path() in gcc.c of the
GCC sources.  First a search is performed on all search paths with the
multilib directory appended (if desired), then a second search is
performed on demand with the base directory only.  For each multilib
there is a "newlib" subdirectory.  This directory is specified by a -B
option for the support libraries.  In order to find the newlib artifacts
(ctr0.o, libc.a, libg.a and libm.a) they must be located in a proper
multilib subdirectory withing the build directory.

Signed-off-by: Sebastian Huber <sebastian.huber@embedded-brains.de>

RTEMS defined SEM_VALUE_MAX to 32767 unlike other systems like FreeBSD
and glibc.  A common value is INT_MAX.

Signed-off-by: Sebastian Huber <sebastian.huber@embedded-brains.de>

getentropy per OpenBSD

  http://man.openbsd.org/cgi-bin/man.cgi/OpenBSD-current/man2/getentropy.2

getrandom per Linux

  http://man7.org/linux/man-pages/man2/getrandom.2.html

  Note that GRND_NONBLOCK is not handled

Signed-off-by: Corinna Vinschen <corinna@vinschen.de>

     * libc/stdlib/strtod.c (strtof_l): Set errno to ERANGE when double to
     float conversion results in infinity.
     (strtof): Likewise.
     * libc/stdlib/wcstod.c (wcstof_l): Likewise.
     (wcstof): Likewise.