Re: Toolchain for Maverick Crunch FPU

[Kai Ruottu <karuottu at mbnet dot fi>]

Claudio Scordino wrote:
> Kai Ruottu wrote:
> > Claudio Scordino wrote:
> > >   I'm looking for a toolchain capable of compiling floating point 
> > > operations for the Maverick Crunch Math engine of Cirrus EP93xx 
> > > processors.
> > >
> > > Cirrus provides some scripts to build a gcc-3.4 toolchain with such 
> > > support, but unfortunately they don't work.
> > >
> > > I heard that gcc 4 has the support for such FPU as well. Is it true ?
> > A quick look at the current gcc-4.1.2 sources told that there is  
> > support for Maverick as the FPU in ARM...
>
> Good!
>
> > > Does it work ?
> > Don't know how well it works...
>
> That remains my main concern...

Things seem to be that the '-mcpu=ep9312 -mhard-float' combination will 
crash the GCC build in both gcc-4.1.2 and gcc-4.2.0-20070316 prerelease 
like :

/data1/home/src/gcc-4.2.0-20070316/build/./gcc/xgcc 
-B/data1/home/src/gcc-4.2.0-20070316/build/./gcc/ 
-B/usr/local/arm-elf/bin/ -B/usr/local/arm-elf/lib/ -isystem 
/usr/local/arm-elf/include -isystem /usr/local/arm-elf/sys-include -O2  
-O2 -Os  -DIN_GCC -DCROSS_COMPILE   -W -Wall -Wwrite-strings 
-Wstrict-prototypes -Wmissing-prototypes -Wold-style-definition  
-isystem ./include  -fno-inline -g  -DIN_LIBGCC2 
-D__GCC_FLOAT_NOT_NEEDED -Dinhibit_libc -I. -I. -I../../gcc 
-I../../gcc/. -I../../gcc/../include -I../../gcc/../libcpp/include  
-I../../gcc/../libdecnumber -I../libdecnumber -mcpu=ep9312 -mhard-float 
-DL_addsubdf3 -xassembler-with-cpp -c ../../gcc/config/arm/lib1funcs.asm 
-o libgcc/ep9312/fpu/_addsubdf3.o
../../gcc/config/arm/ieee754-df.S: Assembler messages:
../../gcc/config/arm/ieee754-df.S:454: Error: selected processor does 
not support `mvfeqd f0,#0.0'
../../gcc/config/arm/ieee754-df.S:476: Error: selected processor does 
not support `mvfeqd f0,#0.0'
../../gcc/config/arm/ieee754-df.S:530: Error: selected processor does 
not support `ldfd f0,[sp],#8'
make[3]: *** [libgcc/ep9312/fpu/_addsubdf3.o] Error 1
make[3]: Leaving directory `/data1/home/src/gcc-4.2.0-20070316/build/gcc'

when enabling the mhard-float/msoft-float multilibs additionally with 
the 'mcpu=ep9312' ones in the
'gcc/config/arm/t-arm-elf' Makefile-fragment....  Without the 
'-mhard-float' those fp-bit/dp-bit based
soft-float routines will be generated into the 'ep9312/libgcc.a', so the 
bare '-mcpu=ep9312' will mean
"use this CPU with soft-float as default", which would sound being the 
same as generating soft-float
always for 'i586', 'i686' etc. which always have that built-in 'i587', 
'i687' etc.  Assuming that those 'ep93xx'
series  members always that "Maverick FPU"....

Producing GCC with only that '-mcpu=ep9312' addition and then having 
only soft-float routines for it
in the used 'libgcc.a' may or may not work with self-built objects 
produced with both '-mcpu=ep9312'
and '-mhard-float'.  But using  '-mcpu=ep9312 -mfpu=maverick' DOESN'T 
generate  any  "Maverick
specific" FPU instructions :-(

So could someone tell if the '-mcpu=ep9312' should work in gcc-4.1 and 
gcc-4.2 and how it should
work so that using the Maverick FPU could be possible?  Or is the FPU 
itself somehow broken and
people shouldn't try to use it at all?