This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Enforce GMP/MPFR version requires and fix fortran/28276,27021
- From: Steve Kargl <sgk at troutmask dot apl dot washington dot edu>
- To: fortran at gcc dot gnu dot org, gcc-patches at gcc dot gnu dot org
- Cc: mark at codesourcery dot com
- Date: Sun, 24 Sep 2006 21:36:46 -0700
- Subject: Re: [PATCH] Enforce GMP/MPFR version requires and fix fortran/28276,27021
- References: <20060912212027.GA63335@troutmask.apl.washington.edu>
PING. PING.
If no one reviews the toplevel changes within the next few
days, I will commit the changes under the implicitly approved
rule.
steve
On Tue, Sep 12, 2006 at 02:20:27PM -0700, Steve Kargl wrote:
> The attached patch has been tested on amd64-*-freebsd, and it
> should only impact gfortran. However, the patch does touch
> the toplevel configure.in, and hence, configure.
>
> markm, I cc'd you because of the pending branching of 4.2 and
> the fact I touch configure.in. If you prefer that I hold the
> patch until after the branching, I can do so.
>
> The patch causes configure to check that the version of
> GMP is 4.1 or newer and that MPFR is version 2.2.0 or
> newer. For older versions of GMP, configure will die.
> For older versions of MPFR, configure simply checks that
> any version of MPFR is available. If no version of
> MPFR is available, configure will die. If an older version
> of MPFR is found, it will be used; however, several known
> problems with older MPFRs prevent the bug fixes for PR
> fortran/28276 and fortran/27021 from functioning (ie.,
> the old buggy gfortran code is still used).
>
> This patch also removes the local hacks for handling subnormal
> numbers and the folding of atan2(). Again, if an older version
> of MPFR is found, the old buggy gfortran code is used.
>
> 2006-09-12 Steven G. Kargl <kargl@gcc.gnu.org>
>
> * configure.in: Check for GMP 4.1 or newer. Check for MPFR 2.2.0
> or newer.
> * configure: Regenerated.
> * doc/install.texi: Document required versions of GMP and MPFR.
> * fortran/arith.c (arctangent2): Conditionally include code.
> (gfc_check_real_range): Use mpfr_subnormalize in preference to local
> hack.
> * fortran/simplify.c: Wrap Copyright to new line.
> (gfc_simplify_atan2): Use mpfr_atan2 in preference to arctangent2().
> (gfc_simplify_log): Ditto.
>
>
> PR fortran/28276
> * fortran/simplify.c (gfc_simplify_exponent): Use mpfr_get_exp in
> preference to broken local hack.
> * fortran/trans-intrinsic.c (gfc_get_intrinsic_lib_fndecl): Append
> l for long double functions.
>
> PR fortran/27021
> * fortran/simplify.c (gfc_simplify_nearest): Use mpfr_nexttoward and
> mpfr_subnormalize to handle numbers near zero in preference to broken
> local hack.
>
> PR fortran/28276
> * testsuite/gfortran.dg/exponent_1.f90: New test.
>
> PR fortran/27021
> * testsuite/gfortran.dg/nearest_1.f90: New test.
>
> --
> Steve
> Index: gcc/doc/install.texi
> ===================================================================
> --- gcc/doc/install.texi (revision 116905)
> +++ gcc/doc/install.texi (working copy)
> @@ -290,17 +290,20 @@ Necessary (only on some platforms) to un
> systems' @command{tar} programs will also work, only try GNU
> @command{tar} if you have problems.
>
> -@item GNU Multiple Precision Library (GMP) version 4.0 (or later)
> +@item GNU Multiple Precision Library (GMP) version 4.1 (or later)
>
> -Necessary to build the Fortran frontend. If you don't have it
> +Necessary to build the Fortran frontend. If you do not have it
> installed in your library search path, you will have to configure with
> the @option{--with-gmp} or @option{--with-gmp-dir} configure option.
>
> -@item MPFR Library
> +@item MPFR Library version 2.2 (or later)
>
> Necessary to build the Fortran frontend. It can be downloaded from
> -@uref{http://www.mpfr.org/}. It is also included in the current GMP
> -release (4.1.3) when configured with @option{--enable-mpfr}.
> +@uref{http://www.mpfr.org/}. The version of MPFR that is bundled with
> +GMP 4.1.x contains numerous bugs. Although gfortran will appear
> +to function with the buggy verions of MPFR, there are few gfortran bugs
> +that will not be fixed to use this version. It is strongly
> +recommended to upgrade to at least MPFR version 2.2.
>
> The @option{--with-mpfr} or @option{--with-mpfr-dir} configure option should
> be used if your MPFR Library is not installed in your library search path.
> Index: gcc/fortran/arith.c
> ===================================================================
> --- gcc/fortran/arith.c (revision 116905)
> +++ gcc/fortran/arith.c (working copy)
> @@ -75,6 +75,7 @@ gfc_set_model (mpfr_t x)
> mpfr_set_default_prec (mpfr_get_prec (x));
> }
>
> +#if !defined(MPFR_VERSION_MAJOR)
> /* Calculate atan2 (y, x)
>
> atan2(y, x) = atan(y/x) if x > 0,
> @@ -124,7 +125,7 @@ arctangent2 (mpfr_t y, mpfr_t x, mpfr_t
>
> mpfr_clear (t);
> }
> -
> +#endif
>
> /* Given an arithmetic error code, return a pointer to a string that
> explains the error. */
> @@ -407,6 +408,7 @@ gfc_check_real_range (mpfr_t p, int kind
> }
> else if (mpfr_cmp (q, gfc_real_kinds[i].tiny) < 0)
> {
> +#if !defined(MPFR_VERSION_MAJOR)
> /* MPFR operates on a number with a given precision and enormous
> exponential range. To represent subnormal numbers, the exponent is
> allowed to become smaller than emin, but always retains the full
> @@ -428,13 +430,30 @@ gfc_check_real_range (mpfr_t p, int kind
> sprintf (s, "0.%sE%d", bin, (int) e);
> mpfr_set_str (q, s, gfc_real_kinds[i].radix, GMP_RNDN);
>
> + gfc_free (s);
> + gfc_free (bin);
> +#else
> + mp_exp_t emin, emax;
> +
> + /* Save current values of emin and emax. */
> + emin = mpfr_get_emin ();
> + emax = mpfr_get_emax ();
> +
> + /* Set emin and emax for the current model number. */
> + mpfr_set_emin ((mp_exp_t) gfc_real_kinds[i].min_exponent - 1);
> + mpfr_set_emax ((mp_exp_t) gfc_real_kinds[i].max_exponent - 1);
> + mpfr_subnormalize (q, 0, GFC_RND_MODE);
> +
> + /* Reset emin and emax. */
> + mpfr_set_emin (emin);
> + mpfr_set_emax (emax);
> +#endif
> +
> + /* Copy sign if needed. */
> if (mpfr_sgn (p) < 0)
> mpfr_neg (p, q, GMP_RNDN);
> else
> mpfr_set (p, q, GMP_RNDN);
> -
> - gfc_free (s);
> - gfc_free (bin);
>
> retval = ARITH_OK;
> }
> Index: gcc/fortran/trans-intrinsic.c
> ===================================================================
> --- gcc/fortran/trans-intrinsic.c (revision 116905)
> +++ gcc/fortran/trans-intrinsic.c (working copy)
> @@ -590,12 +590,18 @@ gfc_get_intrinsic_lib_fndecl (gfc_intrin
>
> if (m->libm_name)
> {
> - gcc_assert (ts->kind == 4 || ts->kind == 8 || ts->kind == 10
> - || ts->kind == 16);
> - snprintf (name, sizeof (name), "%s%s%s",
> - ts->type == BT_COMPLEX ? "c" : "",
> - m->name,
> - ts->kind == 4 ? "f" : "");
> + if (ts->kind == 4)
> + snprintf (name, sizeof (name), "%s%s%s",
> + ts->type == BT_COMPLEX ? "c" : "", m->name, "f");
> + else if (ts->kind == 8)
> + snprintf (name, sizeof (name), "%s%s",
> + ts->type == BT_COMPLEX ? "c" : "", m->name);
> + else
> + {
> + gcc_assert (ts->kind == 10 || ts->kind == 16);
> + snprintf (name, sizeof (name), "%s%s%s",
> + ts->type == BT_COMPLEX ? "c" : "", m->name, "l");
> + }
> }
> else
> {
> Index: gcc/fortran/simplify.c
> ===================================================================
> --- gcc/fortran/simplify.c (revision 116905)
> +++ gcc/fortran/simplify.c (working copy)
> @@ -1,6 +1,6 @@
> /* Simplify intrinsic functions at compile-time.
> - Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006 Free Software
> - Foundation, Inc.
> + Copyright (C) 2000, 2001, 2002, 2003, 2004, 2005, 2006
> + Free Software Foundation, Inc.
> Contributed by Andy Vaught & Katherine Holcomb
>
> This file is part of GCC.
> @@ -607,7 +607,11 @@ gfc_simplify_atan2 (gfc_expr * y, gfc_ex
> return &gfc_bad_expr;
> }
>
> +#if !defined(MPFR_VERSION_MAJOR)
> arctangent2 (y->value.real, x->value.real, result->value.real);
> +#else
> + mpfr_atan2 (result->value.real, y->value.real, x->value.real, GFC_RND_MODE);
> +#endif
>
> return range_check (result, "ATAN2");
> }
> @@ -1054,9 +1058,12 @@ gfc_expr *
> gfc_simplify_exponent (gfc_expr * x)
> {
> int i;
> - mpfr_t tmp;
> gfc_expr *result;
>
> +#if !defined(MPFR_VERSION_MAJOR)
> + mpfr_t tmp;
> +#endif
> +
> if (x->expr_type != EXPR_CONSTANT)
> return NULL;
>
> @@ -1071,6 +1078,9 @@ gfc_simplify_exponent (gfc_expr * x)
> return result;
> }
>
> +#if !defined(MPFR_VERSION_MAJOR)
> + /* PR fortran/28276 suffers from a buggy MPFR, and this block of code
> + does not function correctly. */
> mpfr_init (tmp);
>
> mpfr_abs (tmp, x->value.real, GFC_RND_MODE);
> @@ -1085,6 +1095,11 @@ gfc_simplify_exponent (gfc_expr * x)
> mpz_add_ui (result->value.integer,result->value.integer, 1);
>
> mpfr_clear (tmp);
> +#else
> + /* Requires MPFR 2.2.0 or newer. */
> + i = (int) mpfr_get_exp (x->value.real);
> + mpz_set_si (result->value.integer, i);
> +#endif
>
> return range_check (result, "EXPONENT");
> }
> @@ -2146,8 +2161,13 @@ gfc_simplify_log (gfc_expr * x)
> mpfr_init (xr);
> mpfr_init (xi);
>
> - arctangent2 (x->value.complex.i, x->value.complex.r,
> - result->value.complex.i);
> +#if !defined(MPFR_VERSION_MAJOR)
> + arctangent2 (x->value.complex.i, x->value.complex.r, result->value.complex.i);
> +#else
> + mpfr_atan2 (result->value.complex.i, x->value.complex.i, x->value.complex.r,
> + GFC_RND_MODE);
> +#endif
> +
>
> mpfr_mul (xr, x->value.complex.r, x->value.complex.r, GFC_RND_MODE);
> mpfr_mul (xi, x->value.complex.i, x->value.complex.i, GFC_RND_MODE);
> @@ -2474,27 +2494,28 @@ gfc_simplify_nearest (gfc_expr * x, gfc_
> {
> gfc_expr *result;
> mpfr_t tmp;
> - int direction, sgn;
> + int sgn;
> +#if !defined(MPFR_VERSION_MAJOR)
> + int direction;
> +#else
> + mp_exp_t emin, emax;
> +#endif
>
> if (x->expr_type != EXPR_CONSTANT || s->expr_type != EXPR_CONSTANT)
> return NULL;
>
> - gfc_set_model_kind (x->ts.kind);
> - result = gfc_copy_expr (x);
> -
> - direction = mpfr_sgn (s->value.real);
> -
> - if (direction == 0)
> + if (mpfr_sgn (s->value.real) == 0)
> {
> - gfc_error ("Second argument of NEAREST at %L may not be zero",
> - &s->where);
> - gfc_free (result);
> + gfc_error ("Second argument of NEAREST at %L shall not be zero", &s->where);
> return &gfc_bad_expr;
> }
>
> - /* TODO: Use mpfr_nextabove and mpfr_nextbelow once we move to a
> - newer version of mpfr. */
> + gfc_set_model_kind (x->ts.kind);
> + result = gfc_copy_expr (x);
> +
> +#if !defined(MPFR_VERSION_MAJOR)
>
> + direction = mpfr_sgn (s->value.real);
> sgn = mpfr_sgn (x->value.real);
>
> if (sgn == 0)
> @@ -2539,6 +2560,28 @@ gfc_simplify_nearest (gfc_expr * x, gfc_
> if (sgn < 0)
> mpfr_neg (result->value.real, result->value.real, GFC_RND_MODE);
> }
> +#else
> +
> + /* Save current values of emin and emax. */
> + emin = mpfr_get_emin ();
> + emax = mpfr_get_emax ();
> +
> + /* Set emin and emax for the current model number. */
> + sgn = gfc_validate_kind (BT_REAL, x->ts.kind, 0);
> + mpfr_set_emin ((mp_exp_t) gfc_real_kinds[sgn].min_exponent - 1);
> + mpfr_set_emax ((mp_exp_t) gfc_real_kinds[sgn].max_exponent - 1);
> +
> + sgn = mpfr_sgn (s->value.real);
> + mpfr_init (tmp);
> + mpfr_set_inf (tmp, sgn);
> + mpfr_nexttoward (result->value.real, tmp);
> + mpfr_subnormalize (result->value.real, 0, GFC_RND_MODE);
> +
> + mpfr_set_emin (emin);
> + mpfr_set_emax (emax);
> +
> + mpfr_clear(tmp);
> +#endif
>
> return range_check (result, "NEAREST");
> }
> Index: configure.in
> ===================================================================
> --- configure.in (revision 116905)
> +++ configure.in (working copy)
> @@ -1103,17 +1103,23 @@ CFLAGS="$CFLAGS $gmpinc"
> # Check GMP actually works
> AC_MSG_CHECKING([for correct version of gmp.h])
> AC_TRY_COMPILE([#include "gmp.h"],[
> -#if __GNU_MP_VERSION < 3
> +#if __GNU_MP_VERSION < 4 || (__GNU_MP_VERSION == 4 && __GNU_MP_VERSION_MINOR < 1)
> choke me
> #endif
> -], [AC_MSG_RESULT([yes])],
> - [AC_MSG_RESULT([no]); have_gmp=no])
> +], [AC_MSG_RESULT([yes])], [AC_MSG_RESULT([no]); have_gmp=no])
>
> if test x"$have_gmp" = xyes; then
> - AC_MSG_CHECKING([for MPFR])
> + AC_MSG_CHECKING([for correct version of mpfr.h])
> + AC_TRY_COMPILE([#include "gmp.h"
> +#include <mpfr.h>],[
> +#if MPFR_VERSION_MAJOR < 2 || (MPFR_VERSION_MAJOR == 2 && MPFR_VERSION_MINOR < 2)
> + choke me
> +#endif
> +], [AC_MSG_RESULT([yes])], [AC_MSG_RESULT([buggy version of MPFR detected])])
>
> saved_LIBS="$LIBS"
> LIBS="$LIBS $gmplibs"
> + AC_MSG_CHECKING([for any version of mpfr.h])
> AC_TRY_LINK([#include <gmp.h>
> #include <mpfr.h>], [mpfr_t n; mpfr_init(n);],
> [AC_MSG_RESULT([yes])], [AC_MSG_RESULT([no]); have_gmp=no])
> @@ -1258,7 +1264,7 @@ if test -d ${srcdir}/gcc; then
> case ,${enable_languages},:${have_gmp}:${need_gmp} in
> *,${language},*:no:yes)
> # Specifically requested language; tell them.
> - AC_MSG_ERROR([GMP with MPFR support is required to build $language])
> + AC_MSG_ERROR([GMP 4.1 and MPFR 2.2 or newer versions required by $language])
> ;;
> *:no:yes)
> # Silently disable.
--
Steve