i960 support for soft-float mildly broken

• Date: Wed, 23 Oct 1996 16:56:45 -0700
• From:
  Stephen Williams <steve@icarus.com>
• Reply-To:
  Stephen Williams <steve@icarus.com>

I sent off this bug report to the gcc-bugs list, bug I figure it would
be of interest to others out there as well. So, here you go.

----
GCC 2.7.2 as a cross compiler for i960 targets works perfectly well,
even when the target, such as the i960Jx, does not have any floating
point hardware.

However, the libgcc.a used, even when -msoft-float is passed to gcc,
has been compiled with numeric support included. This comes from the
fact that the default CPU type is the kb, which does have floating
point registers and instructions. GCC without explicit instructions
otherwise, will therefore generate floating point instructions.

The workaround for getting gcc to work with the i960Jx is to recompile
libgcc.a with the -msoft-float flag as follows:

	 % rm libgcc*.a
	 % make libgcc.a CFLAGS=-msoft-float

then install the recompiled libgcc.a in the correct place. The
multilib libgcc.a libraries are probably fine, although I do not have
any i960 targets to test them on.
----

• Prev: Re: GNU for 68K in dos
• Next: i960Jx support for gcc-2.7.2
• Index: Cross gcc Mailing list
• Thread: Cross gcc Mailing list