Message ID | 20200403160307.32477-1-murphyp@linux.vnet.ibm.com |
---|---|
State | New |
Headers | show |
Series | powerpc64le/multiarch: don't generate strong aliases for fmaf128-ppc64 | expand |
On 03/04/2020 13:03, Paul E. Murphy via Libc-alpha wrote: > This prevents generating a second alias for __fmaieee128 when > compiling with ldouble == ieee128 redirects. I trying to pinpoint where exactly this strong_alias is being issued, but fix make sense (since the idea is to just provide a unique function to ifunc selection).+ > --- > sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c | 2 ++ > 1 file changed, 2 insertions(+) > > diff --git a/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c b/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c > index 7374a799af..ab0c4d03a8 100644 > --- a/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c > +++ b/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c > @@ -18,6 +18,8 @@ > > #undef weak_alias > #define weak_alias(a, b) > +#undef strong_alias > +#define strong_alias(a, b) > > #define __fmaf128 __fmaf128_ppc64 > >
On 4/3/20 2:04 PM, Adhemerval Zanella via Libc-alpha wrote: > > > On 03/04/2020 13:03, Paul E. Murphy via Libc-alpha wrote: >> This prevents generating a second alias for __fmaieee128 when >> compiling with ldouble == ieee128 redirects. > > I trying to pinpoint where exactly this strong_alias is being > issued, but fix make sense (since the idea is to just provide > a unique function to ifunc selection).+ The aliasing macros for float128 are buried in indirection. Ultimately, the macro libm_alias_float128_other_r_ldbl in ldbl-128ibm-compat/libm-alias-float.h uses strong_alias. > >> --- >> sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c | 2 ++ >> 1 file changed, 2 insertions(+) >> >> diff --git a/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c b/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c >> index 7374a799af..ab0c4d03a8 100644 >> --- a/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c >> +++ b/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c >> @@ -18,6 +18,8 @@ >> >> #undef weak_alias >> #define weak_alias(a, b) >> +#undef strong_alias >> +#define strong_alias(a, b) >> >> #define __fmaf128 __fmaf128_ppc64 >> >>
On 03/04/2020 17:04, Paul E Murphy wrote: > > > On 4/3/20 2:04 PM, Adhemerval Zanella via Libc-alpha wrote: >> >> >> On 03/04/2020 13:03, Paul E. Murphy via Libc-alpha wrote: >>> This prevents generating a second alias for __fmaieee128 when >>> compiling with ldouble == ieee128 redirects. >> >> I trying to pinpoint where exactly this strong_alias is being >> issued, but fix make sense (since the idea is to just provide >> a unique function to ifunc selection).+ > > The aliasing macros for float128 are buried in indirection. Ultimately, the macro libm_alias_float128_other_r_ldbl in ldbl-128ibm-compat/libm-alias-float.h uses strong_alias. Alright, LGTM then. > >> >>> --- >>> sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c | 2 ++ >>> 1 file changed, 2 insertions(+) >>> >>> diff --git a/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c b/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c >>> index 7374a799af..ab0c4d03a8 100644 >>> --- a/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c >>> +++ b/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c >>> @@ -18,6 +18,8 @@ >>> #undef weak_alias >>> #define weak_alias(a, b) >>> +#undef strong_alias >>> +#define strong_alias(a, b) >>> #define __fmaf128 __fmaf128_ppc64 >>>
diff --git a/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c b/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c index 7374a799af..ab0c4d03a8 100644 --- a/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c +++ b/sysdeps/powerpc/powerpc64/le/fpu/multiarch/s_fmaf128-ppc64.c @@ -18,6 +18,8 @@ #undef weak_alias #define weak_alias(a, b) +#undef strong_alias +#define strong_alias(a, b) #define __fmaf128 __fmaf128_ppc64