Message ID | 20170912123435.6592-3-gabriel@inconstante.eti.br |
---|---|
State | New |
Headers | show |
Series | None | expand |
"Gabriel F. T. Gomes" <gabriel@inconstante.eti.br> writes: > On powerpc64le, compiler support for float128 is not enabled by default > on gcc. To enable it, the flag -mfloat128 must be passed as a command > line option to the compiler. This means that only the few files that > actively have -mfloat128 passed as an argument get compiler support for > float128, whereas all other files don't. > > When -mfloat128 becomes enabled by default on powerpc [1], all the files > that do not currently have compiler support for float128 enabled during > their compilation, will start to have it. This will lead to build > errors in s_finite.c, s_isinf.c, and s_isnan.c. > > The errors are due to the unintended macro expansion of __finitef128 to > __redirect_finitef128 in math/bits/mathcalls-helper-functions.h. In AFAIU, the expansion of __finitef128 to __redirect_finitef128 in math/bits/mathcalls-helper-functions.h is expected. The problem happens when the same thing doesn't happen in include/math.h too. In that case, I suggest to rewrite this part as: The errors are due to the macro expansion of __finitef128 to __redirect_finitef128 in math/bits/mathcalls-helper-functions.h without expanding it in include/math.h too. > diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c > index a5ec36b72f..0382bfb95c 100644 > --- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c > +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c > @@ -17,8 +17,15 @@ > <http://www.gnu.org/licenses/>. */ > > #define __finite __redirect___finite > + > +/* The following definitions, although not related to the 'double' > + version of 'finite', are required to compensate for the unintended > + macro expansion of __finitef to __redirect___finitef, etc. in > + math/bits/mathcalls-helper-functions.h. */ Likewise. I suggest the following change: /* The following definitions, although not related to the 'double' version of 'finite', are required to guarantee the macro expansion of __finitef to __redirect___finitef, etc. in math/bits/mathcalls-helper-functions.h and include/math.h. */ > #define __finitef __redirect___finitef > #define __finitel __redirect___finitel > +#define __finitef128 __redirect___finitef128 > + > #include <math.h> > #include <math_ldbl_opt.h> > #include <shlib-compat.h> Don't you have to undefine this macro later in this file?
On 19 Sep 2017, Tulio Magno Quites Machado Filho wrote: >"Gabriel F. T. Gomes" <gabriel@inconstante.eti.br> writes: > >> The errors are due to the unintended macro expansion of __finitef128 to >> __redirect_finitef128 in math/bits/mathcalls-helper-functions.h. In > >AFAIU, the expansion of __finitef128 to __redirect_finitef128 in >math/bits/mathcalls-helper-functions.h is expected. The problem happens when >the same thing doesn't happen in include/math.h too. I mean unintended in the sense that the expansion of __finitef and __finitel, which pertain to float and long double, are happening in s_finite.c, which is the *double* implementation of finite. Compare that with s_finitef.c, which only redefines __finitef, leaving the other macros (__finite and __finitel) untouched. On the other hand, I agree that my comments did not fully explain the problem. They only mention the expansion in mathcalls-helper-functions.h and do not say anything about the hidden prototypes in include/math.h. So, how about expanding this paragraph to: The errors are due to the unintended macro expansion of __finitef128 to __redirect_finitef128 in math/bits/mathcalls-helper-functions.h. In that header, __MATHDECL_1 takes '__finite' and 'f128' as arguments and concatenates them. However, since '__finite' has been redefined in s_finite.c, the function declaration becomes __redirect_finitef128: extern int __redirect___finitef128 (_Float128 __value) __attribute__ ((__nothrow__ )) __attribute__ ((__const__)); This declaration itself is OK. The problem arises when include/math.h creates the hidden prototype ('hidden_proto (__finitef128)'), which expands to: extern __typeof (__finitef128) __finitef128 __attribute__ ((visibility ("hidden"))); Since __finitef128 is not declared, __typeof fails. This effect was already true for the 'float' and 'long double' versions and is now true for float128. Likewise for isinsff128 and isnanf128. >> +/* The following definitions, although not related to the 'double' >> + version of 'finite', are required to compensate for the unintended >> + macro expansion of __finitef to __redirect___finitef, etc. in >> + math/bits/mathcalls-helper-functions.h. */ > >Likewise. >I suggest the following change: > >/* The following definitions, although not related to the 'double' > version of 'finite', are required to guarantee the macro expansion > of __finitef to __redirect___finitef, etc. in > math/bits/mathcalls-helper-functions.h and include/math.h. */ The expansion in mathcalls-helper-functions.h is unintended, but we need to guarantee the same expansion in include/math.h, so what about the following: The following definitions, although not related to the 'double' version of 'finite', are required to guarantee expansions (e.g.: from __finitef to __redirect_finitef) in include/math.h, thus compensating for the unintended macro expansions in math/bits/mathcalls-helper-functions.h. OK with these changes?
"Gabriel F. T. Gomes" <gabriel@inconstante.eti.br> writes: > On 19 Sep 2017, Tulio Magno Quites Machado Filho wrote: > >>"Gabriel F. T. Gomes" <gabriel@inconstante.eti.br> writes: >> >>> The errors are due to the unintended macro expansion of __finitef128 to >>> __redirect_finitef128 in math/bits/mathcalls-helper-functions.h. In >> >>AFAIU, the expansion of __finitef128 to __redirect_finitef128 in >>math/bits/mathcalls-helper-functions.h is expected. The problem happens when >>the same thing doesn't happen in include/math.h too. > > I mean unintended in the sense that the expansion of __finitef and > __finitel, which pertain to float and long double, are happening in > s_finite.c, which is the *double* implementation of finite. Compare that > with s_finitef.c, which only redefines __finitef, leaving the other macros > (__finite and __finitel) untouched. Got it! > On the other hand, I agree that my comments did not fully explain the > problem. They only mention the expansion in mathcalls-helper-functions.h > and do not say anything about the hidden prototypes in include/math.h. > > So, how about expanding this paragraph to: > > The errors are due to the unintended macro expansion of __finitef128 to > __redirect_finitef128 in math/bits/mathcalls-helper-functions.h. In > that header, __MATHDECL_1 takes '__finite' and 'f128' as arguments and > concatenates them. However, since '__finite' has been redefined in > s_finite.c, the function declaration becomes __redirect_finitef128: > > extern int __redirect___finitef128 (_Float128 __value) __attribute__ ((__nothrow__ )) __attribute__ ((__const__)); > > This declaration itself is OK. The problem arises when include/math.h > creates the hidden prototype ('hidden_proto (__finitef128)'), which > expands to: > > extern __typeof (__finitef128) __finitef128 __attribute__ ((visibility ("hidden"))); > > Since __finitef128 is not declared, __typeof fails. This effect was > already true for the 'float' and 'long double' versions and is now true > for float128. Likewise for isinsff128 and isnanf128. Looks good to me. >>> +/* The following definitions, although not related to the 'double' >>> + version of 'finite', are required to compensate for the unintended >>> + macro expansion of __finitef to __redirect___finitef, etc. in >>> + math/bits/mathcalls-helper-functions.h. */ >> >>Likewise. >>I suggest the following change: >> >>/* The following definitions, although not related to the 'double' >> version of 'finite', are required to guarantee the macro expansion >> of __finitef to __redirect___finitef, etc. in >> math/bits/mathcalls-helper-functions.h and include/math.h. */ > > The expansion in mathcalls-helper-functions.h is unintended, but we need > to guarantee the same expansion in include/math.h, so what about the > following: > > The following definitions, although not related to the 'double' version > of 'finite', are required to guarantee expansions (e.g.: from __finitef > to __redirect_finitef) in include/math.h, thus compensating for the > unintended macro expansions in math/bits/mathcalls-helper-functions.h. > > OK with these changes? Yes, assuming you also include the #undef lines I mentioned in my previous email. Thanks!
diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c index a5ec36b72f..0382bfb95c 100644 --- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_finite.c @@ -17,8 +17,15 @@ <http://www.gnu.org/licenses/>. */ #define __finite __redirect___finite + +/* The following definitions, although not related to the 'double' + version of 'finite', are required to compensate for the unintended + macro expansion of __finitef to __redirect___finitef, etc. in + math/bits/mathcalls-helper-functions.h. */ #define __finitef __redirect___finitef #define __finitel __redirect___finitel +#define __finitef128 __redirect___finitef128 + #include <math.h> #include <math_ldbl_opt.h> #include <shlib-compat.h> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf.c index 9c6789c7bd..43827557a3 100644 --- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf.c +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isinf.c @@ -17,8 +17,15 @@ <http://www.gnu.org/licenses/>. */ #define __isinf __redirect___isinf + +/* The following definitions, although not related to the 'double' + version of 'isinf', are required to compensate for the unintended + macro expansion of __isinff to __redirect___isinff, etc. in + math/bits/mathcalls-helper-functions.h. */ #define __isinff __redirect___isinff #define __isinfl __redirect___isinfl +#define __isinff128 __redirect___isinff128 + #include <math.h> #include <math_ldbl_opt.h> #include <shlib-compat.h> diff --git a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c index 3cfe1793da..bbc59808a4 100644 --- a/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c +++ b/sysdeps/powerpc/powerpc64/fpu/multiarch/s_isnan.c @@ -17,8 +17,15 @@ <http://www.gnu.org/licenses/>. */ #define __isnan __redirect___isnan + +/* The following definitions, although not related to the 'double' + version of 'isnan', are required to compensate for the unintended + macro expansion of __isnanf to __redirect___isnanf, etc. in + math/bits/mathcalls-helper-functions.h. */ #define __isnanf __redirect___isnanf #define __isnanl __redirect___isnanl +#define __isnanf128 __redirect___isnanf128 + #include <math.h> #include <math_ldbl_opt.h> #include <shlib-compat.h>