[src/trunk]: src/lib/libc/gdtoa Honour the floating-point rounding mode in fl...

[riastradh <riastradh%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/788c69d5689e
branches:  trunk
changeset: 458559:788c69d5689e
user:      riastradh <riastradh%NetBSD.org@localhost>
date:      Thu Aug 01 02:06:31 2019 +0000

description:
Honour the floating-point rounding mode in floating-point formatting.

C99, Sec. 7.19.6.1 `The fprintf function', paragraph 13, p. 281:

   (Recommended practice)

   For e, E, f, F, g, and G conversions, if the number of significant
   decimal digits is at most DECIMAL_DIG, then the result should be
   correctly rounded.  If the number of significant decimal digits is
   more than DECIMAL_DIG but the source value is exactly
   representable with DECIMAL_DIG digits, then the result should be
   an exact representation with trailing zeros.  Otherwise, the
   source value is bounded by two adjacent decimal strings L < U,
   both having DECIMAL_DIG significant idgits; the value of the
   resultant decimal string D should satisfy L <= D <= U, _with the
   extra stipulation that the error should have a correct sign for
   the current rounding direction_.  [emphasis added]

The gdtoa code base already supports respecting the floating-point
rounding mode, as long as we compile it with Honor_FLT_ROUNDS
defined.  However, for this to work, fegetround must be available in
libc, which it is not currently -- the fenv logic is in libm.

Fortunately, we don't have to move all of fenv from libm to libc --
programs that do not link against libm don't have fesetround, so the
rounding mode is always the default (barring asm shenanigans that
bypass the API -- tough).  So use a weak reference to fegetround; by
default, assume FE_TONEAREST if it is not defined.

diffstat:

 lib/libc/gdtoa/Makefile.inc    |  6 +++++-
 lib/libc/gdtoa/gdtoa_fltrnds.h |  3 ++-
 lib/libc/gdtoa/gdtoaimp.h      |  5 ++++-
 3 files changed, 11 insertions(+), 3 deletions(-)

diffs (48 lines):

diff -r 18a734a85d63 -r 788c69d5689e lib/libc/gdtoa/Makefile.inc
--- a/lib/libc/gdtoa/Makefile.inc       Thu Aug 01 01:19:21 2019 +0000
+++ b/lib/libc/gdtoa/Makefile.inc       Thu Aug 01 02:06:31 2019 +0000
@@ -1,10 +1,14 @@
-#      $NetBSD: Makefile.inc,v 1.10 2014/01/16 20:31:42 christos Exp $
+#      $NetBSD: Makefile.inc,v 1.11 2019/08/01 02:06:31 riastradh Exp $
 
 # gdtoa sources
 .PATH: ${.CURDIR}/gdtoa
 CPPFLAGS+=-I${.CURDIR}/gdtoa -I${.CURDIR}/locale
 
+.if ${MACHINE_ARCH} == "vax"
 CPPFLAGS+=-DNO_FENV_H
+.else
+CPPFLAGS+=-DHonor_FLT_ROUNDS
+.endif
 
 # machine-dependent directory must provide the following:
 #      arith.h gd_qnan.h
diff -r 18a734a85d63 -r 788c69d5689e lib/libc/gdtoa/gdtoa_fltrnds.h
--- a/lib/libc/gdtoa/gdtoa_fltrnds.h    Thu Aug 01 01:19:21 2019 +0000
+++ b/lib/libc/gdtoa/gdtoa_fltrnds.h    Thu Aug 01 02:06:31 2019 +0000
@@ -1,4 +1,5 @@
-       FPI *fpi, fpi1;
+       CONST FPI *fpi;
+       FPI fpi1;
        int Rounding;
 #ifdef Trust_FLT_ROUNDS /*{{ only define this if FLT_ROUNDS really works! */
        Rounding = Flt_Rounds;
diff -r 18a734a85d63 -r 788c69d5689e lib/libc/gdtoa/gdtoaimp.h
--- a/lib/libc/gdtoa/gdtoaimp.h Thu Aug 01 01:19:21 2019 +0000
+++ b/lib/libc/gdtoa/gdtoaimp.h Thu Aug 01 02:06:31 2019 +0000
@@ -1,4 +1,4 @@
-/* $NetBSD: gdtoaimp.h,v 1.14 2013/04/19 10:41:53 joerg Exp $ */
+/* $NetBSD: gdtoaimp.h,v 1.15 2019/08/01 02:06:31 riastradh Exp $ */
 
 /****************************************************************
 
@@ -200,6 +200,9 @@
 #include "gd_qnan.h"
 #ifdef Honor_FLT_ROUNDS
 #include <fenv.h>
+__weakref_visible int __libc_fegetround_ref(void) __weak_reference(fegetround);
+#define fegetround()                                                   \
+       (__libc_fegetround_ref ? __libc_fegetround_ref() : FE_TONEAREST)
 #endif
 
 #ifdef DEBUG