[10/12] ldbl-128ibm-compat: Redirect long double functions to f128/ieee128 functions

Message ID 20180620020426.20372-11-tuliom@linux.ibm.com
State New
Headers show
Series
  • Introduce ieee128 symbols and redirections
Related show

Commit Message

Tulio Magno Quites Machado Filho June 20, 2018, 2:04 a.m.
Modify the headers to redirect long double functions to global __*f128
symbols or to __*ieee128 otherwise.

Most of the functions in math.h benefit from the infrastructure already
available for __LDBL_COMPAT.  The only exceptions are nexttowardf and
nexttoward that need especial treatment.

Both math/bits/mathcalls-helper-functions.h and math/bits/mathcalls.h
were modified in order to provide alternative redirection destinations
that are essential to support functions that should not be redirected to
the same name pattern of the rest of the functions, i.e.: __fpclassify,
__signbit, __iseqsig, __issignaling, isinf, finite and isnan, which will
be redirected to __*f128 instead of __*ieee128 used for the rest.

2018-06-19  Tulio Magno Quites Machado Filho  <tuliom@linux.ibm.com>

	* include/math.h [__HAVE_DISTINCT_FLOAT128 &&
	!__HAVE_FLOAT128_UNLIKE_LDBL]: Avoid redirecting __finitel,
	__isinfl, __isnanl, __fpclassifyl, __issignalingl, __expl,
	__expml and sqrtl as they conflict with the redirections from
	math/math.h.
	* sysdeps/ieee754/ldbl-128ibm/bits/iscanonical.h
	[__HAVE_DISTINCT_FLOAT128 && !__HAVE_FLOAT128_UNLIKE_LDBL]: Likewise
	for function iscanonical(_Float128).
	* math/math.h [__HAVE_DISTINCT_FLOAT128 &&
	!__HAVE_FLOAT128_UNLIKE_LDBL]: Redirect long double functions to
	__*f128 or _*ieee128.
        (__MATHDECL_ALIAS): New macro.
	* math/bits/mathcalls-helper-functions.h: Replace __MATHDECL_1
	with __MATHDECL_ALIAS.
	* math/bits/mathcalls.h: Likewise for isinf, finite and isnan.

Signed-off-by: Tulio Magno Quites Machado Filho <tuliom@linux.ibm.com>

---
 include/math.h                                 | 12 ++++-
 math/bits/mathcalls-helper-functions.h         | 18 +++----
 math/bits/mathcalls.h                          |  9 ++--
 math/math.h                                    | 66 ++++++++++++++++++++++++--
 sysdeps/ieee754/ldbl-128ibm/bits/iscanonical.h |  6 ++-
 5 files changed, 94 insertions(+), 17 deletions(-)

-- 
2.14.4

Comments

Joseph Myers June 20, 2018, 8:53 p.m. | #1
On Tue, 19 Jun 2018, Tulio Magno Quites Machado Filho wrote:

> diff --git a/math/math.h b/math/math.h

> index c1765e4c8c..8d0880b262 100644

> --- a/math/math.h

> +++ b/math/math.h

> @@ -283,6 +283,10 @@ enum

>    extern type __MATH_PRECNAME(function,suffix) args __THROW

>  #define __MATHDECL_1(type, function, suffix, args) \

>    __MATHDECL_1_IMPL(type, function, suffix, args)

> +/* Ignore the alias by default.  The alias is only useful with

> +   redirections.  */

> +#define __MATHDECL_ALIAS(type, function, suffix, args, alias) \

> +  __MATHDECL_1_IMPL(type, function, suffix, args)


So this is the default definitions of __MATHDECL_1 and __MATHDECL_ALIAS.

>  #define __MATHREDIR(type, function, suffix, args, to) \

>    extern type __REDIRECT_NTH (__MATH_PRECNAME (function, suffix), args, to)

> @@ -364,6 +368,39 @@ extern long double __REDIRECT_NTH (nexttowardl,

>  #   undef __MATHDECL_1

>  #   define __MATHDECL_1(type, function,suffix, args) \

>    __MATHREDIR(type, function, suffix, args, __CONCAT(function,suffix))

> +


And this is the __LDBL_COMPAT redefinition, I think.  You've moved various 
functions to use __MATHDECL_ALIAS, but __MATHDECL_ALIAS isn't being 
redefined for __LDBL_COMPAT.  Are you sure that's correct?  That is, what 
do the macro-generated declarations of functions using __MATHDECL_ALIAS 
look like in a -mlong-double-64 configuration, before and after this 
patch?

> +#define __dremieee128 __remainderieee128

> +#define __gammaieee128 __lgammaieee128


These are missing appropriate preprocessor indendation for their nesting 
level inside #if.

-- 
Joseph S. Myers
joseph@codesourcery.com
Tulio Magno Quites Machado Filho June 20, 2018, 10:48 p.m. | #2
Joseph Myers <joseph@codesourcery.com> writes:

> On Tue, 19 Jun 2018, Tulio Magno Quites Machado Filho wrote:

>

>>  #define __MATHREDIR(type, function, suffix, args, to) \

>>    extern type __REDIRECT_NTH (__MATH_PRECNAME (function, suffix), args, to)

>> @@ -364,6 +368,39 @@ extern long double __REDIRECT_NTH (nexttowardl,

>>  #   undef __MATHDECL_1

>>  #   define __MATHDECL_1(type, function,suffix, args) \

>>    __MATHREDIR(type, function, suffix, args, __CONCAT(function,suffix))

>> +

>

> And this is the __LDBL_COMPAT redefinition, I think.  You've moved various 

> functions to use __MATHDECL_ALIAS, but __MATHDECL_ALIAS isn't being 

> redefined for __LDBL_COMPAT.  Are you sure that's correct?  That is, what 

> do the macro-generated declarations of functions using __MATHDECL_ALIAS 

> look like in a -mlong-double-64 configuration, before and after this 

> patch?


You're right.  The default value for __MATHDECL_ALIAS should have been
__MATHDECL_1.
After changing it I get the following:

extern int __isinfl (long double __value) __asm__ ("" "__isinf") __attribute__ ((__nothrow__ , __leaf__))
  __attribute__ ((__const__));

>> +#define __dremieee128 __remainderieee128

>> +#define __gammaieee128 __lgammaieee128

>

> These are missing appropriate preprocessor indendation for their nesting 

> level inside #if.


Ack.  Fixed.

Thanks!

-- 
Tulio Magno

Patch

diff --git a/include/math.h b/include/math.h
index e21d34b868..57d77cfd13 100644
--- a/include/math.h
+++ b/include/math.h
@@ -18,7 +18,9 @@  hidden_proto (__finitef)
 hidden_proto (__isinff)
 hidden_proto (__isnanf)
 
-#  ifndef __NO_LONG_DOUBLE_MATH
+#  if !defined __NO_LONG_DOUBLE_MATH \
+      && (!__HAVE_DISTINCT_FLOAT128 \
+	  || (__HAVE_DISTINCT_FLOAT128 && __HAVE_FLOAT128_UNLIKE_LDBL))
 hidden_proto (__finitel)
 hidden_proto (__isinfl)
 hidden_proto (__isnanl)
@@ -40,7 +42,9 @@  libm_hidden_proto (__exp)
 libm_hidden_proto (__expf)
 libm_hidden_proto (__roundeven)
 
-# ifndef __NO_LONG_DOUBLE_MATH
+#  if !defined __NO_LONG_DOUBLE_MATH \
+  && (!__HAVE_DISTINCT_FLOAT128 \
+      || (__HAVE_DISTINCT_FLOAT128 && __HAVE_FLOAT128_UNLIKE_LDBL))
 libm_hidden_proto (__fpclassifyl)
 libm_hidden_proto (__issignalingl)
 libm_hidden_proto (__expl)
@@ -61,7 +65,11 @@  libm_hidden_proto (__expm1f128)
 float (sqrtf) (float) asm ("__ieee754_sqrtf");
 double (sqrt) (double) asm ("__ieee754_sqrt");
 #   ifndef __NO_LONG_DOUBLE_MATH
+/* Do not redirect sqrtl twice when the long double already redirects it.  */
+#    if !__HAVE_DISTINCT_FLOAT128 \
+	|| (__HAVE_DISTINCT_FLOAT128 && __HAVE_FLOAT128_UNLIKE_LDBL)
 long double (sqrtl) (long double) asm ("__ieee754_sqrtl");
+#    endif
 #   endif
 #   if __HAVE_DISTINCT_FLOAT128 > 0
 _Float128 (sqrtf128) (_Float128) asm ("__ieee754_sqrtf128");
diff --git a/math/bits/mathcalls-helper-functions.h b/math/bits/mathcalls-helper-functions.h
index ac3115bfdc..a6d83bf4e6 100644
--- a/math/bits/mathcalls-helper-functions.h
+++ b/math/bits/mathcalls-helper-functions.h
@@ -16,28 +16,30 @@ 
    License along with the GNU C Library; if not, see
    <http://www.gnu.org/licenses/>.  */
 
-
 /* Classify given number.  */
-__MATHDECL_1 (int, __fpclassify,, (_Mdouble_ __value))
+__MATHDECL_ALIAS (int, __fpclassify,, (_Mdouble_ __value), fpclassify)
      __attribute__ ((__const__));
 
 /* Test for negative number.  */
-__MATHDECL_1 (int, __signbit,, (_Mdouble_ __value))
+__MATHDECL_ALIAS (int, __signbit,, (_Mdouble_ __value), signbit)
      __attribute__ ((__const__));
 
 /* Return 0 if VALUE is finite or NaN, +1 if it
    is +Infinity, -1 if it is -Infinity.  */
-__MATHDECL_1 (int, __isinf,, (_Mdouble_ __value)) __attribute__ ((__const__));
+__MATHDECL_ALIAS (int, __isinf,, (_Mdouble_ __value), isinf)
+  __attribute__ ((__const__));
 
 /* Return nonzero if VALUE is finite and not NaN.  Used by isfinite macro.  */
-__MATHDECL_1 (int, __finite,, (_Mdouble_ __value)) __attribute__ ((__const__));
+__MATHDECL_ALIAS (int, __finite,, (_Mdouble_ __value), finite)
+  __attribute__ ((__const__));
 
 /* Return nonzero if VALUE is not a number.  */
-__MATHDECL_1 (int, __isnan,, (_Mdouble_ __value)) __attribute__ ((__const__));
+__MATHDECL_ALIAS (int, __isnan,, (_Mdouble_ __value), isnan)
+  __attribute__ ((__const__));
 
 /* Test equality.  */
-__MATHDECL_1 (int, __iseqsig,, (_Mdouble_ __x, _Mdouble_ __y));
+__MATHDECL_ALIAS (int, __iseqsig,, (_Mdouble_ __x, _Mdouble_ __y), iseqsig);
 
 /* Test for signaling NaN.  */
-__MATHDECL_1 (int, __issignaling,, (_Mdouble_ __value))
+__MATHDECL_ALIAS (int, __issignaling,, (_Mdouble_ __value), issignaling)
      __attribute__ ((__const__));
diff --git a/math/bits/mathcalls.h b/math/bits/mathcalls.h
index f4246a7881..9ace71bef6 100644
--- a/math/bits/mathcalls.h
+++ b/math/bits/mathcalls.h
@@ -173,12 +173,14 @@  __MATHCALL_FINITE (fmod,, (_Mdouble_ __x, _Mdouble_ __y));
       && !__MATH_DECLARING_FLOATN
 /* Return 0 if VALUE is finite or NaN, +1 if it
    is +Infinity, -1 if it is -Infinity.  */
-__MATHDECL_1 (int,isinf,, (_Mdouble_ __value)) __attribute__ ((__const__));
+__MATHDECL_ALIAS (int,isinf,, (_Mdouble_ __value), isinf)
+  __attribute__ ((__const__));
 # endif
 
 # if !__MATH_DECLARING_FLOATN
 /* Return nonzero if VALUE is finite and not NaN.  */
-__MATHDECL_1 (int,finite,, (_Mdouble_ __value)) __attribute__ ((__const__));
+__MATHDECL_ALIAS (int,finite,, (_Mdouble_ __value), finite)
+  __attribute__ ((__const__));
 
 /* Return the remainder of X/Y.  */
 __MATHCALL (drem,, (_Mdouble_ __x, _Mdouble_ __y));
@@ -207,7 +209,8 @@  __MATHCALL (nan,, (const char *__tagb));
        || __MATH_DECLARING_DOUBLE == 0)) /* isnanf or isnanl don't.  */ \
       && !__MATH_DECLARING_FLOATN
 /* Return nonzero if VALUE is not a number.  */
-__MATHDECL_1 (int,isnan,, (_Mdouble_ __value)) __attribute__ ((__const__));
+__MATHDECL_ALIAS (int,isnan,, (_Mdouble_ __value), isnan)
+  __attribute__ ((__const__));
 # endif
 #endif
 
diff --git a/math/math.h b/math/math.h
index c1765e4c8c..8d0880b262 100644
--- a/math/math.h
+++ b/math/math.h
@@ -283,6 +283,10 @@  enum
   extern type __MATH_PRECNAME(function,suffix) args __THROW
 #define __MATHDECL_1(type, function, suffix, args) \
   __MATHDECL_1_IMPL(type, function, suffix, args)
+/* Ignore the alias by default.  The alias is only useful with
+   redirections.  */
+#define __MATHDECL_ALIAS(type, function, suffix, args, alias) \
+  __MATHDECL_1_IMPL(type, function, suffix, args)
 
 #define __MATHREDIR(type, function, suffix, args, to) \
   extern type __REDIRECT_NTH (__MATH_PRECNAME (function, suffix), args, to)
@@ -364,6 +368,39 @@  extern long double __REDIRECT_NTH (nexttowardl,
 #   undef __MATHDECL_1
 #   define __MATHDECL_1(type, function,suffix, args) \
   __MATHREDIR(type, function, suffix, args, __CONCAT(function,suffix))
+
+#  elif __HAVE_DISTINCT_FLOAT128 && !__HAVE_FLOAT128_UNLIKE_LDBL
+#   ifdef __REDIRECT_NTH
+#    ifdef __USE_ISOC99
+extern float __REDIRECT_NTH (nexttowardf, (float __x, long double __y),
+			    __nexttowardf_to_ieee128)
+  __attribute__ ((__const__));
+extern double __REDIRECT_NTH (nexttoward, (double __x, long double __y),
+			     __nexttoward_to_ieee128)
+  __attribute__ ((__const__));
+
+#define __dremieee128 __remainderieee128
+#define __gammaieee128 __lgammaieee128
+
+#    endif
+#   endif
+
+#   undef __MATHDECL_1
+#   undef __MATHDECL_ALIAS
+
+#   define __REDIRTO(function, suffix) \
+  __ ## function ## ieee128 ## suffix
+#   define __REDIRTO_ALT(function, suffix) \
+  __ ## function ## f128 ## suffix
+
+#   undef __MATH_FINITE_NAME
+#   define __MATH_FINITE_NAME(function, suffix) \
+  __MATH_FINITE_NAMEX (, __REDIRTO_ALT (function, suffix), _finite)
+
+#   define __MATHDECL_1(type, function, suffix, args) \
+  __MATHREDIR (type, function, suffix, args, __REDIRTO (function, suffix))
+#   define __MATHDECL_ALIAS(type, function, suffix, args, alias) \
+  __MATHREDIR (type, function, suffix, args, __REDIRTO_ALT (alias, suffix))
 #  endif
 
 /* Include the file of declarations again, this time using `long double'
@@ -376,15 +413,25 @@  extern long double __REDIRECT_NTH (nexttowardl,
 #  define __MATH_DECLARE_LDOUBLE   1
 #  include <bits/mathcalls-helper-functions.h>
 #  include <bits/mathcalls.h>
+
 #  undef _Mdouble_
 #  undef __MATH_PRECNAME
 #  undef __MATH_DECLARING_DOUBLE
 #  undef __MATH_DECLARING_FLOATN
 
-#  if defined __LDBL_COMPAT
+#  if defined __LDBL_COMPAT \
+      || (__HAVE_DISTINCT_FLOAT128 && !__HAVE_FLOAT128_UNLIKE_LDBL)
+#   undef __REDIRTO
+#   undef __REDIRTO_ALT
 #   undef __MATHDECL_1
+#   undef __MATHDECL_ALIAS
+#   undef __MATH_FINITE_NAME
+#   define __MATH_FINITE_NAME(function, suffix) \
+  __MATH_FINITE_NAME_IMPL (function, suffix)
 #   define __MATHDECL_1(type, function, suffix, args) \
   __MATHDECL_1_IMPL(type, function, suffix, args)
+#   define __MATHDECL_ALIAS(type, function, suffix, args, alias) \
+  __MATHDECL_1_IMPL(type, function, suffix, args)
 #  endif
 # endif /* !(__NO_LONG_DOUBLE_MATH && _LIBC) || __LDBL_COMPAT */
 
@@ -720,6 +767,7 @@  extern int signgam;
 #undef	__REDIRTO_PUBLIC_X
 #undef	__MATHDECL_1_IMPL
 #undef	__MATHDECL_1
+#undef	__MATHDECL_ALIAS
 #undef	__MATHDECL
 #undef	__MATHCALL
 #undef	__MATHCALL_VEC
@@ -759,6 +807,11 @@  extern int signgam;
 # ifdef __LDBL_COMPAT
 #  define __MATHCALL_REDIR_NAME(name) f ## name
 #  undef __MATHCALL_NARROW
+#  define __MATHCALL_NARROW(func, redir, nargs) \
+  __MATHCALL_NARROW_REDIR (func, redir, nargs)
+# elif __HAVE_DISTINCT_FLOAT128 && !__HAVE_FLOAT128_UNLIKE_LDBL
+#  define __MATHCALL_REDIR_NAME(name) __ ## f32 ## name ## ieee128
+#  undef __MATHCALL_NARROW
 #  define __MATHCALL_NARROW(func, redir, nargs) \
   __MATHCALL_NARROW_REDIR (func, redir, nargs)
 # endif
@@ -766,7 +819,8 @@  extern int signgam;
 # undef _Mret_
 # undef _Marg_
 # undef __MATHCALL_NAME
-# ifdef __LDBL_COMPAT
+# if defined __LDBL_COMPAT \
+     || (__HAVE_DISTINCT_FLOAT128 && !__HAVE_FLOAT128_UNLIKE_LDBL)
 #  undef __MATHCALL_REDIR_NAME
 #  undef __MATHCALL_NARROW
 #  define __MATHCALL_NARROW(func, redir, nargs) \
@@ -779,6 +833,11 @@  extern int signgam;
 # ifdef __LDBL_COMPAT
 #  define __MATHCALL_REDIR_NAME(name) __nldbl_d ## name ## l
 #  undef __MATHCALL_NARROW
+#  define __MATHCALL_NARROW(func, redir, nargs) \
+  __MATHCALL_NARROW_REDIR (func, redir, nargs)
+# elif __HAVE_DISTINCT_FLOAT128 && !__HAVE_FLOAT128_UNLIKE_LDBL
+#  define __MATHCALL_REDIR_NAME(name) __ ## f64 ## name ## ieee128
+#  undef __MATHCALL_NARROW
 #  define __MATHCALL_NARROW(func, redir, nargs) \
   __MATHCALL_NARROW_REDIR (func, redir, nargs)
 # endif
@@ -786,7 +845,8 @@  extern int signgam;
 # undef _Mret_
 # undef _Marg_
 # undef __MATHCALL_NAME
-# ifdef __LDBL_COMPAT
+# if defined __LDBL_COMPAT \
+     || (__HAVE_DISTINCT_FLOAT128 && !__HAVE_FLOAT128_UNLIKE_LDBL)
 #  undef __MATHCALL_REDIR_NAME
 #  undef __MATHCALL_NARROW
 #  define __MATHCALL_NARROW(func, redir, nargs) \
diff --git a/sysdeps/ieee754/ldbl-128ibm/bits/iscanonical.h b/sysdeps/ieee754/ldbl-128ibm/bits/iscanonical.h
index 60c54cca46..f54e5d9ddb 100644
--- a/sysdeps/ieee754/ldbl-128ibm/bits/iscanonical.h
+++ b/sysdeps/ieee754/ldbl-128ibm/bits/iscanonical.h
@@ -50,7 +50,11 @@  extern "C++" {
 inline int iscanonical (float __val) { return __iscanonicalf (__val); }
 inline int iscanonical (double __val) { return __iscanonical (__val); }
 inline int iscanonical (long double __val) { return __iscanonicall (__val); }
-#  if __HAVE_DISTINCT_FLOAT128
+/* When using an IEEE 128-bit long double, _Float128 is defined as long double
+   in C++.  */
+#  if __HAVE_DISTINCT_FLOAT128 \
+      && (!defined __cplusplus \
+	  || defined __cplusplus && __HAVE_FLOAT128_UNLIKE_LDBL)
 inline int iscanonical (_Float128 __val) { return __iscanonicalf128 (__val); }
 #  endif
 }