This RFC exemplifies what I have planned for the *printf and *scanf
family of functions, with respect to the second transition of the long
double format on powerpc64le...
In the light of what's been discussed in a previous thread [1], more
specifically, the explanation about the unnecessary increase in the
number of exported symbols [2] (quotes are indented):
[..] using new *l symbol versions and new exported names for ibm128
functions (which seems to me to increase the number of exported symbols
unnecessarily).
And taking into account that new exported symbols are needed for those
functions that are not part of the TS 18661-3 API:
[..] obsolescent functions with no public *f128 names which shouldn't
get such names, but are part of the *l API so need to be available for
long double in the -mabi=ieeelongdouble case, so you may actually want
to call new __*f128 names instead
I adjusted the implementation that I have been working on (for *printf
functions), so that there are no new exported functions for ibm128.
Instead, there are only new exported functions for binary128 and header
redirections (in stdio.h) from *l functions to these new exports (this
RFC only contains the rework for vfprintf and for the functions it
depends on, i.e.: __printf_fp and __printf_fphex. I haven't reworked
the other functions, that's why I omitted them from this RFC (I hope
that's OK and I believe it even eases the review)).
I tried to add as many comments as possible to the changes, so as to
avoid an overly long commit message. Anyhow, an overview:
* The redirections for user code are provided by the following files:
* libio/stdio.h
* sysdeps/ieee754/ldbl-128ibm-compat/bits/stdio-ieee128.h.
* The following files, which are compiled with -mabi=ieeelongdouble,
redefine function names before including the implementation from
stdio-common:
* sysdeps/ieee754/ldbl-128ibm-compat/ieee128-printf_fp.c
* sysdeps/ieee754/ldbl-128ibm-compat/ieee128-printf_fphex.c
* sysdeps/ieee754/ldbl-128ibm-compat/ieee128-vfprintf.c
* Exported symbols (Versions) and their correct versions
(math_ldbl_opt.h) are defined by these files:
* sysdeps/ieee754/ldbl-128ibm-compat/Versions
* sysdeps/ieee754/ldbl-opt/math_ldbl_opt.h
* sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ldbl-128ibm-compat-abi.h
* For the remaining files, I don't have any special comments.
Any feedback is highly appreciated.
Thanks,
Gabriel
[1] https://sourceware.org/ml/libc-alpha/2018-05/msg00509.html
[2] https://sourceware.org/ml/libc-alpha/2018-05/msg00519.html
---
libio/stdio.h | 9 ++++
sysdeps/ieee754/ldbl-128ibm-compat/Makefile | 20 +++++++++
sysdeps/ieee754/ldbl-128ibm-compat/Versions | 12 +++++
.../ldbl-128ibm-compat/bits/stdio-ieee128.h | 30 +++++++++++++
.../ieee754/ldbl-128ibm-compat/ieee128-printf_fp.c | 16 +++++++
.../ldbl-128ibm-compat/ieee128-printf_fphex.c | 16 +++++++
.../ieee754/ldbl-128ibm-compat/ieee128-vfprintf.c | 13 ++++++
.../ldbl-128ibm-compat/test-printf-ibm128.c | 1 +
.../ldbl-128ibm-compat/test-printf-ieee128.c | 1 +
.../ldbl-128ibm-compat/test-printf-ldbl-compat.c | 51 ++++++++++++++++++++++
sysdeps/ieee754/ldbl-opt/math_ldbl_opt.h | 27 +++++++++++-
.../powerpc/powerpc64/le/ldbl-128ibm-compat-abi.h | 8 ++++
12 files changed, 202 insertions(+), 2 deletions(-)
create mode 100644 sysdeps/ieee754/ldbl-128ibm-compat/Makefile
create mode 100644 sysdeps/ieee754/ldbl-128ibm-compat/Versions
create mode 100644 sysdeps/ieee754/ldbl-128ibm-compat/bits/stdio-ieee128.h
create mode 100644 sysdeps/ieee754/ldbl-128ibm-compat/ieee128-printf_fp.c
create mode 100644 sysdeps/ieee754/ldbl-128ibm-compat/ieee128-printf_fphex.c
create mode 100644 sysdeps/ieee754/ldbl-128ibm-compat/ieee128-vfprintf.c
create mode 100644 sysdeps/ieee754/ldbl-128ibm-compat/test-printf-ibm128.c
create mode 100644 sysdeps/ieee754/ldbl-128ibm-compat/test-printf-ieee128.c
create mode 100644 sysdeps/ieee754/ldbl-128ibm-compat/test-printf-ldbl-compat.c
create mode 100644 sysdeps/unix/sysv/linux/powerpc/powerpc64/le/ldbl-128ibm-compat-abi.h
"Gabriel F. T. Gomes" <gabriel@inconstante.eti.br> writes:
> diff --git a/libio/stdio.h b/libio/stdio.h
> index 731f8e56f4..6d6b3e6062 100644
> --- a/libio/stdio.h
> +++ b/libio/stdio.h
> @@ -864,6 +864,15 @@ extern int __overflow (FILE *, int);
> # include <bits/stdio-ldbl.h>
> #endif
>
> +/* For platforms where long double had double format, then was converted
> + to some non-IEEE format, then finally to IEEE binary128 format, add
> + redirections to the correct implementation of stdio.h functions. */
> +#include <bits/floatn.h>
> +#if __HAVE_DISTINCT_FLOAT128 && __LDBL_MANT_DIG__ == 113 && \
> + ! defined __BUILDING_EXTRA_LDBL_FORMAT
> +# include <bits/stdio-ieee128.h>
> +#endif
I think you have to replace '__LDBL_MANT_DIG__ == 113' with
'!__HAVE_FLOAT128_UNLIKE_LDBL'.
If you make that change would you still need to use
__BUILDING_EXTRA_LDBL_FORMAT here?
> diff --git a/sysdeps/ieee754/ldbl-128ibm-compat/Makefile b/sysdeps/ieee754/ldbl-128ibm-compat/Makefile
> new file mode 100644
> index 0000000000..ef04cbb67f
> --- /dev/null
> +++ b/sysdeps/ieee754/ldbl-128ibm-compat/Makefile
> @@ -0,0 +1,20 @@
> +ifeq ($(subdir),stdio-common)
> +routines += ieee128-printf_fp ieee128-printf_fphex
> +CFLAGS-ieee128-printf_fp.c += -mfloat128 -mabi=ieeelongdouble -Wno-psabi
> +CFLAGS-ieee128-printf_fphex.c += -mfloat128 -mabi=ieeelongdouble -Wno-psabi
> +
> +routines += ieee128-vfprintf
> +CFLAGS-ieee128-vfprintf.c += -mfloat128 -mabi=ieeelongdouble -Wno-psabi
> +
> +tests-internal += test-printf-ieee128
> +CFLAGS-test-printf-ieee128.c := $(filter-out -mlong-double-128, \
> + $(CFLAGS-test-printf-ieee128.c))
I don't think this is mandatory, but helps someone analyzing the compiler
flags. Could you add a source comment clarifying this, please?
> +CFLAGS-test-printf-ieee128.c += -mfloat128 -mabi=ieeelongdouble -Wno-psabi
> +
> +tests-internal += test-printf-ibm128
> +CFLAGS-test-printf-ibm128.c := $(filter-out -mlong-double-128, \
> + $(CFLAGS-test-printf-ibm128.c))
Likewise.
> +__IBM128_REDIR (vfprintf)
Outdated name for the macro? :-D
Notice there is already __LDBL_REDIR_DECL from misc/sys/cdefs.h to help you
with this.
> diff --git a/sysdeps/ieee754/ldbl-128ibm-compat/ieee128-vfprintf.c b/sysdeps/ieee754/ldbl-128ibm-compat/ieee128-vfprintf.c
> new file mode 100644
> index 0000000000..db9cd79b24
> --- /dev/null
> +++ b/sysdeps/ieee754/ldbl-128ibm-compat/ieee128-vfprintf.c
> @@ -0,0 +1,13 @@
> +/* Redefine the names of the functions in vfprintf.c before including it. */
> +#define _IO_vfprintf_internal __ieee128__IO_vfprintf_internal
> +#define __printf_fp __ieee128___printf_fp
> +#define __printf_fphex __ieee128___printf_fphex
> +#define vfprintf __ieee128_vfprintf
> +#define _IO_vfprintf __ieee128__IO_vfprintf
Don't you need to include stdio.h before defining these macros?
> +/* Skip the inclusion of the redirections of *l functions to __ieee128_*
> + functions during the build of glibc (these redirections would
> + conflict with the internal redirections to __GL_* symbols). */
> +#define __BUILDING_EXTRA_LDBL_FORMAT
If you include stdio.h, would you still need this macro?
> diff --git a/sysdeps/ieee754/ldbl-opt/math_ldbl_opt.h b/sysdeps/ieee754/ldbl-opt/math_ldbl_opt.h
> index 61ba784f86..78e5d4b496 100644
> --- a/sysdeps/ieee754/ldbl-opt/math_ldbl_opt.h
> +++ b/sysdeps/ieee754/ldbl-opt/math_ldbl_opt.h
> @@ -5,14 +5,35 @@
> # error "nldbl-abi.h must define LONG_DOUBLE_COMPAT_VERSION"
> #endif
>
> +/* On powerpc64le, a third format for long double is supported since the
> + version described in ldbl-128ibm-compat-abi.h. */
> +#include <bits/floatn.h>
> +#if __HAVE_DISTINCT_FLOAT128 && __LDBL_MANT_DIG__ == 113
Use !__HAVE_FLOAT128_UNLIKE_LDBL again.
@@ -864,6 +864,15 @@ extern int __overflow (FILE *, int);
# include <bits/stdio-ldbl.h>
#endif
+/* For platforms where long double had double format, then was converted
+ to some non-IEEE format, then finally to IEEE binary128 format, add
+ redirections to the correct implementation of stdio.h functions. */
+#include <bits/floatn.h>
+#if __HAVE_DISTINCT_FLOAT128 && __LDBL_MANT_DIG__ == 113 && \
+ ! defined __BUILDING_EXTRA_LDBL_FORMAT
+# include <bits/stdio-ieee128.h>
+#endif
+
__END_DECLS
#endif /* <stdio.h> included. */
new file mode 100644
@@ -0,0 +1,20 @@
+ifeq ($(subdir),stdio-common)
+routines += ieee128-printf_fp ieee128-printf_fphex
+CFLAGS-ieee128-printf_fp.c += -mfloat128 -mabi=ieeelongdouble -Wno-psabi
+CFLAGS-ieee128-printf_fphex.c += -mfloat128 -mabi=ieeelongdouble -Wno-psabi
+
+routines += ieee128-vfprintf
+CFLAGS-ieee128-vfprintf.c += -mfloat128 -mabi=ieeelongdouble -Wno-psabi
+
+tests-internal += test-printf-ieee128
+CFLAGS-test-printf-ieee128.c := $(filter-out -mlong-double-128, \
+ $(CFLAGS-test-printf-ieee128.c))
+CFLAGS-test-printf-ieee128.c += -mfloat128 -mabi=ieeelongdouble -Wno-psabi
+
+tests-internal += test-printf-ibm128
+CFLAGS-test-printf-ibm128.c := $(filter-out -mlong-double-128, \
+ $(CFLAGS-test-printf-ibm128.c))
+CFLAGS-test-printf-ibm128.c += -mabi=ibmlongdouble -Wno-psabi
+
+headers += bits/stdio-ieee128.h
+endif
new file mode 100644
@@ -0,0 +1,12 @@
+%include <ldbl-128ibm-compat-abi.h>
+%ifndef LDBL_IBM128_VERSION
+% error "ldbl-128ibm-compat-abi.h must define LDBL_IBM128_VERSION"
+%endif
+
+libc {
+ LDBL_IBM128_VERSION {
+ __ieee128___printf_fp;
+ __ieee128__IO_vfprintf;
+ __ieee128_vfprintf;
+ }
+}
new file mode 100644
@@ -0,0 +1,30 @@
+/* Redirections for stdio functions for -mabi=ieeelongdouble.
+ Copyright (C) 2018 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#ifndef _STDIO_H
+# error "Never include <bits/stdio-ibm128.h> directly; use <stdio.h> instead."
+#endif
+
+#include <stdio.h>
+
+#define __IBM128_REDIR(name) \
+ extern __typeof (name) __ieee128_##name; \
+ extern __typeof (name) name __asm (__ASMNAME ("__ieee128_" #name));
+
+__IBM128_REDIR (vfprintf)
+/* To be completed with the other functions. */
new file mode 100644
@@ -0,0 +1,16 @@
+/* Redefine the names of the functions in printf_fp.c before including it. */
+#define __printf_fp __ieee128___printf_fp
+#define __printf_fp_l __ieee128___printf_fp_l
+#define ___printf_fp __ieee128____printf_fp
+#define __guess_grouping __ieee128___guess_grouping
+
+/* Use __mpn_extract_float128 since it handles the floating-point
+ parameter as a floating-point value with binary128 format. */
+#define __mpn_extract_long_double __mpn_extract_float128
+
+/* Skip the inclusion of the redirections of *l functions to __ieee128_*
+ functions during the build of glibc (these redirections would
+ conflict with the internal redirections to __GL_* symbols). */
+#define __BUILDING_EXTRA_LDBL_FORMAT
+
+#include <stdio-common/printf_fp.c>
new file mode 100644
@@ -0,0 +1,16 @@
+/* Redefine the name of the function in printf_fphex.c before including it. */
+#define __printf_fphex __ieee128___printf_fphex
+
+/* Skip the inclusion of the redirections of *l functions to __ieee128_*
+ functions during the build of glibc (these redirections would
+ conflict with the internal redirections to __GL_* symbols). */
+#define __BUILDING_EXTRA_LDBL_FORMAT
+
+/* Explicitly include ieee754.h from the ldbl-128 directory, since this
+ file is building __printf_fphex for long double with IEEE binary128
+ format (implicitly, this header would be included from ldbl-128ibm,
+ which has definitions for long double with IBM Extended Precision
+ format for floating-point). */
+#include <sysdeps/ieee754/ldbl-128/ieee754.h>
+
+#include <sysdeps/ieee754/ldbl-128/printf_fphex.c>
new file mode 100644
@@ -0,0 +1,13 @@
+/* Redefine the names of the functions in vfprintf.c before including it. */
+#define _IO_vfprintf_internal __ieee128__IO_vfprintf_internal
+#define __printf_fp __ieee128___printf_fp
+#define __printf_fphex __ieee128___printf_fphex
+#define vfprintf __ieee128_vfprintf
+#define _IO_vfprintf __ieee128__IO_vfprintf
+
+/* Skip the inclusion of the redirections of *l functions to __ieee128_*
+ functions during the build of glibc (these redirections would
+ conflict with the internal redirections to __GL_* symbols). */
+#define __BUILDING_EXTRA_LDBL_FORMAT
+
+#include <stdio-common/vfprintf.c>
new file mode 100644
@@ -0,0 +1 @@
+#include <test-printf-ldbl-compat.c>
new file mode 100644
@@ -0,0 +1 @@
+#include <test-printf-ldbl-compat.c>
new file mode 100644
@@ -0,0 +1,51 @@
+/* Test for the long double variants of *printf functions.
+ Copyright (C) 2018 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include <stdarg.h>
+#include <stdint.h>
+#include <stdio.h>
+
+#include <support/check.h>
+
+static void
+do_test_call_varg (FILE *stream, const char *format, ...)
+{
+ va_list args;
+
+ printf ("%20s", "vfprintf: ");
+ va_start (args, format);
+ vfprintf (stream, format, args);
+ va_end (args);
+ printf ("\n");
+}
+
+static int
+do_test (void)
+{
+ long double ld = -1;
+
+ /* Print in decimal notation. */
+ do_test_call_varg (stdout, "%.60Lf", ld);
+
+ /* Print in hexadecimal notation. */
+ do_test_call_varg (stdout, "%.60La", ld);
+
+ return 0;
+}
+
+#include <support/test-driver.c>
@@ -5,14 +5,35 @@
# error "nldbl-abi.h must define LONG_DOUBLE_COMPAT_VERSION"
#endif
+/* On powerpc64le, a third format for long double is supported since the
+ version described in ldbl-128ibm-compat-abi.h. */
+#include <bits/floatn.h>
+#if __HAVE_DISTINCT_FLOAT128 && __LDBL_MANT_DIG__ == 113
+# include <ldbl-128ibm-compat-abi.h>
+# ifndef LDBL_IBM128_COMPAT_VERSION
+# error "ldbl-128ibm-compat-abi.h must define LDBL_IBM128_COMPAT_VERSION"
+# endif
+#endif
+
#include <shlib-compat.h>
#define LONG_DOUBLE_COMPAT(lib, introduced) \
SHLIB_COMPAT(lib, introduced, LONG_DOUBLE_COMPAT_VERSION)
-#define long_double_symbol(lib, local, symbol) \
+/* When building long double with IEEE long double format on
+ powerpc64le, exposed the symbols with the version defined in
+ ldbl-128ibm-compat-abi.h. */
+#if __HAVE_DISTINCT_FLOAT128 && __LDBL_MANT_DIG__ == 113
+# define long_double_symbol(lib, local, symbol) \
+ long_double_symbol_1 (lib, local, symbol, LDBL_IBM128_COMPAT_VERSION)
+/* Otherwise, expose the symbols with the version from nldbl-abi.h. */
+#else
+# define long_double_symbol(lib, local, symbol) \
long_double_symbol_1 (lib, local, symbol, LONG_DOUBLE_COMPAT_VERSION)
+#endif
#ifdef SHARED
# define ldbl_hidden_def(local, name) libc_hidden_ver (local, name)
# define ldbl_strong_alias(name, aliasname) \
+ ldbl_strong_alias_x (name, aliasname)
+# define ldbl_strong_alias_x(name, aliasname) \
strong_alias (name, __GL_##name##_##aliasname) \
long_double_symbol (libc, __GL_##name##_##aliasname, aliasname);
# define ldbl_weak_alias(name, aliasname) \
@@ -22,7 +43,9 @@
versioned_symbol (lib, local, symbol, version)
#else
# define ldbl_hidden_def(local, name) libc_hidden_def (name)
-# define ldbl_strong_alias(name, aliasname) strong_alias (name, aliasname)
+# define ldbl_strong_alias(name, aliasname) \
+ ldbl_strong_alias_x (name, aliasname)
+# define ldbl_strong_alias_x(name, aliasname) strong_alias (name, aliasname)
# define ldbl_weak_alias(name, aliasname) weak_alias (name, aliasname)
# ifndef __ASSEMBLER__
/* Note that weak_alias cannot be used - it is defined to nothing
new file mode 100644
@@ -0,0 +1,8 @@
+/* ABI version for long double switch to IEEE 128-bit floating point..
+ This is used by the Versions and math_ldbl_opt.h files in
+ sysdeps/ieee754/ldbl-128ibm-compat/. It gives the ABI version where
+ long double == ibm128 was replaced with long double == _Float128
+ for libm *l functions and libc functions using long double. */
+
+#define LDBL_IBM128_VERSION GLIBC_2.28
+#define LDBL_IBM128_COMPAT_VERSION GLIBC_2_28