> (note to robert: did you really mean to comment p, q, a[] and b[] from
> cdct.c? it fails to compile on non x86 systems... naughty.)
I had to uncomment those to get it to build w/o x86 optimizations.
> Does freeamp compile/build/work without the x86 optimizations?
Yes, but since there isn't a ./configure option to disable x86
optimizations, I had to hack the makefiles and config.h to get there. I
suggest that you add a --disable-all-x86-optimizations or something like
that; then it's easier to test it without x86 optimizations. :)
> what does `nm icdct.o | grep coef32` say? It ought to have the coef32
> and wincoef symbols there.
It's there: "00000000 b coef32", according to nm. I haven't the faintest
idea why ld was barfing on the nasm and gas stuff. It works fine now
without the optimizations, which is okay by me since I am using freeamp to
decode from MP3 to PCM data files, not real-time playing.
> What exactly is the generated 'XINGLMCOBJC +=' line for you, before you
> have to edit it? Sounds like configure may not be doing the right things.
> (and for what it's worth, cc/gcc SHOULD know what to do with a .s file,
> and there's already a .asm rule, although that requires perl [although
> it should probably use sed. :)]. feh.)
The XINGLMCOBJC += file is coming out as
"lmc/xingmp3/src/{cdct,cwin,cwin8,mdct,msis}asm.o". I noticed that
configure.in is trying to use the shell to expand that by default, but for
some reason it isn't working.
cc/gcc do know what to do with a .s file; they send it through gcc with
the -S (?) assemble option. The problem is that gcc doesn't seem to be able
to assemble the asm files by itself. This is why I had to put in the
implicit rule.
My solaris-x86-2.5.1 machine has no Sparcworks stuff, just gcc and binutils
downloaded from sunfreeware.com.
--pat
