Daniel Jacobowitz wrote:
On Fri, Nov 07, 2003 at 02:08:55PM -0800, David Daney wrote:
Eric Christopher wrote:
I forget to say that without this patch --with-float= seems to have no
effect on the built compiler.
Well, that's the problem then.
It should already work. Particularly, from mips.h:
#define OPTION_DEFAULT_SPECS \
{"arch", "%{!march=*:%{mips16:-march=%(VALUE)}%{!mips*:-march=%(VALUE)}}"
}, \
{"tune", "%{!mtune=*:-mtune=%(VALUE)}" }, \
{"abi", "%{!mabi=*:-mabi=%(VALUE)}" }, \
{"float", "%{!msoft-float:%{!mhard-float:-m%(VALUE)-float}}" }
And from config.gcc:
mips*-*-*)
supported_defaults="abi arch float tune"
Does the generated configargs.h have an entry for float? If so, what's
wrong with the specs that causes it not to work?
Yes:
[daney@dl gcc]$ cat configargs.h
/* Generated automatically. */
static const char configuration_arguments[] = "../gcccvsw/gcc/configure
--host=i686-pc-linux-gnu --target=mipsisa32el-linux
--enable-languages=c,c++,java
--with-headers=/home/mipsel-linux/mipsel-linux/include
--prefix=/home/testcvs/ --with-float=soft";
static const char thread_model[] = "posix";
static const struct {
const char *name, *value;
} configure_default_options[] = { { "float", "soft" } };
However gcc was still generating floating point instructions.
???