This is the mail archive of the
fortran@gcc.gnu.org
mailing list for the GNU Fortran project.
[PATCH]: Use mpc_pow in the fortran frontend
- From: "Kaveh R. GHAZI" <ghazi at caip dot rutgers dot edu>
- To: fortran at gcc dot gnu dot org
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Sat, 27 Jun 2009 09:13:00 -0400 (EDT)
- Subject: [PATCH]: Use mpc_pow in the fortran frontend
This patch makes use of mpc_pow in the fortran frontend. Note mpc_pow
is only available in the svn repo for MPC. I've guarded the uses of
mpc_pow with the appropriate version checks.
The call to mpc_pow replaces two separate areas, a general pow, and
the integer exponent specialization. The existing fortran integer
exponent code had a problem immortalized in a testcase, and as
discussed here:
http://gcc.gnu.org/ml/fortran/2009-06/msg00267.html
I'm commenting out some lines in one testcase that were incorrect.
I'll reinsert the corrected lines once we've made MPC a
hard-requirement, and assuming this patch is accepted I'll put a note
in PR40302 so I don't forget.
Bootstrapped on x86_64-unknown-linux-gnu with MPC svn (which has
mpc_pow), with mpc-0.6 (no pow), and without MPC. All three scenarios
had no regressions.
Okay for mainline?
Thanks,
--Kaveh
2009-06-26 Kaveh R. Ghazi <ghazi@caip.rutgers.edu>
gcc/fortran:
* gfortran.h: Define HAVE_mpc_pow.
* arith.c (complex_reciprocal, complex_pow): If HAVE_mpc_pow,
don't define these functions.
(arith_power): If HAVE_mpc_pow, use mpc_pow.
gcc/testsuite:
* gfortran.dg/integer_exponentiation_4.f90: Temporarily
comment out some values and add some cases.
diff -rup orig/egcc-SVN20090625/gcc/fortran/gfortran.h egcc-SVN20090625/gcc/fortran/gfortran.h
--- orig/egcc-SVN20090625/gcc/fortran/gfortran.h 2009-06-25 02:01:27.000000000 +0200
+++ egcc-SVN20090625/gcc/fortran/gfortran.h 2009-06-25 22:54:40.000000000 +0200
@@ -1558,6 +1558,9 @@ gfc_intrinsic_sym;
#include <mpfr.h>
#ifdef HAVE_mpc
#include <mpc.h>
+# if MPC_VERSION >= MPC_VERSION_NUM(0,6,1)
+# define HAVE_mpc_pow
+# endif
#else
#define mpc_realref(X) ((X).r)
#define mpc_imagref(X) ((X).i)
diff -rup orig/egcc-SVN20090625/gcc/fortran/arith.c egcc-SVN20090625/gcc/fortran/arith.c
--- orig/egcc-SVN20090625/gcc/fortran/arith.c 2009-06-19 17:35:32.000000000 +0200
+++ egcc-SVN20090625/gcc/fortran/arith.c 2009-06-25 22:54:40.000000000 +0200
@@ -896,6 +896,7 @@ gfc_arith_divide (gfc_expr *op1, gfc_exp
/* Compute the reciprocal of a complex number (guaranteed nonzero). */
+#if ! defined(HAVE_mpc_pow)
static void
complex_reciprocal (gfc_expr *op)
{
@@ -922,6 +923,7 @@ complex_reciprocal (gfc_expr *op)
}
#endif
}
+#endif /* ! HAVE_mpc_pow */
/* Raise a complex number to positive power (power > 0).
@@ -932,6 +934,7 @@ complex_reciprocal (gfc_expr *op)
"Seminumerical Algorithms", Vol. 2, "The Art of Computer Programming",
3rd Edition, 1998. */
+#if ! defined(HAVE_mpc_pow)
static void
complex_pow (gfc_expr *result, gfc_expr *base, mpz_t power)
{
@@ -988,6 +991,7 @@ complex_pow (gfc_expr *result, gfc_expr
mpfr_clears (x_r, x_i, tmp, re, im, NULL);
}
+#endif /* ! HAVE_mpc_pow */
/* Raise a number to a power. */
@@ -1107,6 +1111,15 @@ arith_power (gfc_expr *op1, gfc_expr *op
case BT_COMPLEX:
{
+#ifdef HAVE_mpc_pow
+ mpc_t apower;
+ gfc_set_model (mpc_realref (op1->value.complex));
+ mpc_init2 (apower, mpfr_get_default_prec());
+ mpc_set_z (apower, op2->value.integer, GFC_MPC_RND_MODE);
+ mpc_pow(result->value.complex, op1->value.complex, apower,
+ GFC_MPC_RND_MODE);
+ mpc_clear (apower);
+#else
mpz_t apower;
/* Compute op1**abs(op2) */
@@ -1118,6 +1131,7 @@ arith_power (gfc_expr *op1, gfc_expr *op
/* If (op2 < 0), compute the inverse. */
if (power_sign < 0)
complex_reciprocal (result);
+#endif /* HAVE_mpc_pow */
}
break;
@@ -1159,6 +1173,10 @@ arith_power (gfc_expr *op1, gfc_expr *op
return ARITH_PROHIBIT;
}
+#ifdef HAVE_mpc_pow
+ mpc_pow (result->value.complex, op1->value.complex,
+ op2->value.complex, GFC_MPC_RND_MODE);
+#else
{
mpfr_t x, y, r, t;
@@ -1211,6 +1229,7 @@ arith_power (gfc_expr *op1, gfc_expr *op
mpfr_mul (mpc_imagref (result->value.complex), x, y, GFC_RND_MODE);
mpfr_clears (r, t, x, y, NULL);
}
+#endif /* HAVE_mpc_pow */
}
break;
default:
diff -rup orig/egcc-SVN20090625/gcc/testsuite/gfortran.dg/integer_exponentiation_4.f90 egcc-SVN20090625/gcc/testsuite/gfortran.dg/integer_exponentiation_4.f90
--- orig/egcc-SVN20090625/gcc/testsuite/gfortran.dg/integer_exponentiation_4.f90 2008-03-14 00:37:50.000000000 +0100
+++ egcc-SVN20090625/gcc/testsuite/gfortran.dg/integer_exponentiation_4.f90 2009-06-27 02:16:40.000000000 +0200
@@ -38,7 +38,10 @@ program test
print *, nearest(1.0,-1.0)**(-huge(0)) ! { dg-error "Arithmetic overflow" }
!!!!!! COMPLEX BASE !!!!!!
- print *, (2.0,-4.3)**huge(0) ! { dg-error "Arithmetic NaN" }
- print *, (2.0,-4.3)**(-huge(0)) ! { dg-error "Arithmetic NaN" }
+! Put these lines back in (and "no-" -> "dg-") prior to gcc-4.5.
+! print *, (2.0,-4.3)**huge(0) ! { no-error "Arithmetic overflow" }
+! print *, (2.0,-4.3)**huge(0_8) ! { no-error "Arithmetic overflow" }
+! print *, (2.0,-4.3)**(-huge(0))
+! print *, (2.0,-4.3)**(-huge(0_8))
end program test