Re: [Qemu-devel] Undefined reference to 'ffs' when building for win64

[Richard Henderson]

On 2013-02-18 01:52, Kevin Wolf wrote:
> On Fri, Feb 15, 2013 at 10:16:11PM +0100, Stefan Weil wrote:
> > It looks like most (but not all) versions of gcc emit built-in code
> > for ffs() even for compilations without optimisation. I have to provide
> > additional compiler options (e.g. -fno-builtin-ffs) to get a real
> > function call.
> > Only then I get the linker errors which you see.
> >
> > Obviously your compiler behaves different (or did you add extra
> > compiler options?).
> >
> > Try this short test.c code:
> >
> > int ffs(unsigned n);
> > int main(int argc, char *argv[])
> > {
> >    return ffs(argc);
> > }
> >
> > On Debian, I compiled and tested like this:
> >
> > amd64-mingw32msvc-gcc -c -g -O0 test.c
> > amd64-mingw32msvc-nm test.o
> >
> > There is no ffs function call in the resulting binary.
> >
> > If your compiler creates a function call, we have to know
> > why and whether there is a simple rule which versions of
> > gcc behave like that. If there is no simple rule, we need
> > a configure check.
>
> A configure check would require that a compiler either creates always a
> function call, or never. I'm not sure if you can rely on that.

Almost certainly best to stop using a system ffs and use the ctz
for which we have plain old arithmetic fallback support.  It would also 
clean up the subtraction in e.g.

> hw/pci/pci.h:    return (val & mask) >> (ffs(mask) - 1);
> hw/strongarm.c:        bit = ffs(diff) - 1;
> target-ppc/cpu.h:    int tlb_bits = ffs(booke206_tlb_size(env, tlbn)) - 1;

i.e. almost every place that we actually use ffs today.


r~