Re: [PATCH] mips: Use builtins for ffs and ffsll

["Maciej W. Rozycki" <macro@orcam.me.uk>]

On Mon, 5 Feb 2024, Adhemerval Zanella Netto wrote:

> >> __builtin_ffs{,ll} basically on __builtin_ctz{,ll} in MIPS GCC compiler.
> >> The hardware ctz instructions were available after MIPS{32,64} Release1. By
> using builtin ctz. It can also reduce code size of ffs/ffsll.
> >>
> >> Checked on mips o32. mips64.
> >>
> >> Signed-off-by: Junxian Zhu <zhujunxian@oss.cipunited.com>
> > 
> > LGTM, thanks.
> > 
> > Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>
> > 
> >> ---
> >>  sysdeps/mips/math-use-builtins-ffs.h | 2 ++
> >>  1 file changed, 2 insertions(+)
> >>  create mode 100644 sysdeps/mips/math-use-builtins-ffs.h
> >>
> >> diff --git a/sysdeps/mips/math-use-builtins-ffs.h
> b/sysdeps/mips/math-use-builtins-ffs.h
> >> new file mode 100644
> >> index 0000000000..78b3f14fae
> >> --- /dev/null
> >> +++ b/sysdeps/mips/math-use-builtins-ffs.h
> >> @@ -0,0 +1,2 @@
> >> +#define USE_FFS_BUILTIN (__mips_isa_rev >= 1)
> >> +#define USE_FFSLL_BUILTIN (__mips_isa_rev >= 1)
> 
> In fact you need to include <sysdep.h> since gcc only defines __mips_isa_rev
> for -mips32 or higher.

 Also this causes code pessimisation for MIPS32 and `ffsll', as there is 
no hw instruction for this operation in the 32-bit ABI and consequently 
`__builtin_ffsll' resorts to a libcall.  So IMO USE_FFSLL_BUILTIN ought to 
only be set for 64-bit ABIs, making `ffsll' defer to `ffs' (and then use 
32-bit CLZ as appropriate).

 NB contrary to the change description there is no CTZ instruction in the 
MIPS ISA at any level.  There are: CLO, CLZ, and for 64-bit: DCLO, DCLZ, 
and `__builtin_ctz', etc. can be synthesized with the usual calculation.

  Maciej