[PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Kaveh R. GHAZI]

This patch partly addresses PR middle-end/30250.  It uses MPFR to
evaluate the reentrant lgamma_r/gamma_r builtins at compile-time when
supplied with constant args.  The reentrant versions supply signgam as
a pointer parameter, and they were slightly easier to implement than
doing the regular lgamma functions, so I did them first.  (The plain
lgamma/gamma patch will come hopefully soon.)

The patch requires several others to apply cleanly:
http://gcc.gnu.org/ml/gcc-patches/2007-04/msg01624.html
http://gcc.gnu.org/ml/gcc-patches/2007-04/msg01663.html
http://gcc.gnu.org/ml/gcc-patches/2007-05/msg00297.html
http://gcc.gnu.org/ml/gcc-patches/2007-05/msg00320.html

Patch below tested on sparc-sun-solaris2.10 in conjuction with the
above four patches, no regressions and the new testcases all pass.  I
also ran a preliminary version of my plain lgamma/gamma patch against
the FP accuracy testsuite to check whether MPFR's lgamma function was
producing correct results back to GCC.  It was, all inputs generated
"perfect" results.

Okay for mainline?

		Thanks,
		--Kaveh

:ADDPATCH middle-end:


2007-05-07  Kaveh R. Ghazi  <ghazi@caip.rutgers.edu>

	PR middle-end/30250
	* builtins.c (do_mpfr_lgamma_r): New.
	(fold_builtin_2): Handle builtin gamma_r/lgamma_r.
	* tree.h (CASE_FLT_FN_REENT): New.

testsuite:
	* gcc.dg/torture/builtin-math-2.c: Add gamma_r/lgamma_r tests.
	* gcc.dg/torture/builtin-math-4.c: Likewise.

diff -rup orig/egcc-SVN20070505/gcc/builtins.c egcc-SVN20070505/gcc/builtins.c
--- orig/egcc-SVN20070505/gcc/builtins.c	2007-05-06 23:43:01.621873788 -0400
+++ egcc-SVN20070505/gcc/builtins.c	2007-05-07 00:55:32.966335476 -0400
@@ -234,6 +234,7 @@ static tree do_mpfr_bessel_n (tree, tree
 			      int (*)(mpfr_ptr, long, mpfr_srcptr, mp_rnd_t),
 			      const REAL_VALUE_TYPE *, bool);
 static tree do_mpfr_remquo (tree, tree, tree);
+static tree do_mpfr_lgamma_r (tree, tree, tree);
 #endif

 /* Return true if NODE should be considered for inline expansion regardless
@@ -9862,6 +9863,13 @@ fold_builtin_2 (tree fndecl, tree arg0,
           && validate_arg(arg1, REAL_TYPE))
         return do_mpfr_arg2 (arg0, arg1, type, mpfr_remainder);
     break;
+
+    CASE_FLT_FN_REENT (BUILT_IN_GAMMA): /* GAMMA_R */
+    CASE_FLT_FN_REENT (BUILT_IN_LGAMMA): /* LGAMMA_R */
+      if (validate_arg (arg0, REAL_TYPE)
+	  && validate_arg(arg1, POINTER_TYPE))
+	return do_mpfr_lgamma_r (arg0, arg1, type);
+    break;
 #endif

     CASE_FLT_FN (BUILT_IN_ATAN2):
@@ -12616,4 +12624,66 @@ do_mpfr_remquo (tree arg0, tree arg1, tr
     }
   return result;
 }
+
+/* If ARG is a REAL_CST, call mpfr_lgamma() on it and return the
+   resulting value as a tree with type TYPE.  The mpfr precision is
+   set to the precision of TYPE.  We assume that this mpfr function
+   returns zero if the result could be calculated exactly within the
+   requested precision.  In addition, the integer pointer represented
+   by ARG_SG will be dereferenced and set to the appropriate signgam
+   (-1,1) value.  */
+
+static tree
+do_mpfr_lgamma_r (tree arg, tree arg_sg, tree type)
+{
+  tree result = NULL_TREE;
+
+  STRIP_NOPS (arg);
+
+  /* To proceed, MPFR must exactly represent the target floating point
+     format, which only happens when the target base equals two.  */
+  if (REAL_MODE_FORMAT (TYPE_MODE (type))->b == 2
+      && TREE_CODE (arg) == REAL_CST && !TREE_OVERFLOW (arg))
+    {
+      const REAL_VALUE_TYPE *const ra = TREE_REAL_CST_PTR (arg);
+
+      /* In addition to NaN and Inf, the argument cannot be zero or a
+	 negative integer.  */
+      if (!real_isnan (ra) && !real_isinf (ra)
+	  && ra->cl != rvc_zero
+	  && !(real_isneg(ra) && real_isinteger(ra, TYPE_MODE (type))))
+        {
+	  const int prec = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
+	  int inexact, sg;
+	  mpfr_t m;
+	  tree result_lg;
+
+	  mpfr_init2 (m, prec);
+	  mpfr_from_real (m, ra, GMP_RNDN);
+	  mpfr_clear_flags ();
+	  inexact = mpfr_lgamma (m, &sg, m, GMP_RNDN);
+	  result_lg = do_mpfr_ckconv (m, type, inexact);
+	  mpfr_clear (m);
+	  if (result_lg)
+	    {
+	      /* Dereference the arg_sg pointer argument.  */
+	      arg_sg = build_fold_indirect_ref (arg_sg);
+	      /* Proceed iff a valid pointer type was passed in.  */
+	      if (TYPE_MAIN_VARIANT (TREE_TYPE (arg_sg)) == integer_type_node)
+	        {
+		  /* Set the value. */
+		  tree result_sg = fold_build2 (MODIFY_EXPR,
+						TREE_TYPE (arg_sg), arg_sg,
+						 build_int_cst (NULL, sg));
+		  TREE_SIDE_EFFECTS (result_sg) = 1;
+		  /* Combine the signgam assignment with the lgamma result.  */
+		  result = non_lvalue (fold_build2 (COMPOUND_EXPR, type,
+						    result_sg, result_lg));
+		}
+	    }
+	}
+    }
+
+  return result;
+}
 #endif
diff -rup orig/egcc-SVN20070505/gcc/tree.h egcc-SVN20070505/gcc/tree.h
--- orig/egcc-SVN20070505/gcc/tree.h	2007-05-03 03:14:35.000000000 -0400
+++ egcc-SVN20070505/gcc/tree.h	2007-05-06 23:46:56.209230787 -0400
@@ -273,6 +273,7 @@ extern const char * built_in_names[(int)
 #define BUILTIN_ROOT_P(FN) (BUILTIN_SQRT_P (FN) || BUILTIN_CBRT_P (FN))

 #define CASE_FLT_FN(FN) case FN: case FN##F: case FN##L
+#define CASE_FLT_FN_REENT(FN) case FN##_R: case FN##F_R: case FN##L_R
 #define CASE_INT_FN(FN) case FN: case FN##L: case FN##LL

 /* An array of _DECL trees for the above.  */
diff -rup orig/egcc-SVN20070505/gcc/testsuite/gcc.dg/torture/builtin-math-2.c egcc-SVN20070505/gcc/testsuite/gcc.dg/torture/builtin-math-2.c
--- orig/egcc-SVN20070505/gcc/testsuite/gcc.dg/torture/builtin-math-2.c	2007-05-06 23:42:44.833833597 -0400
+++ egcc-SVN20070505/gcc/testsuite/gcc.dg/torture/builtin-math-2.c	2007-05-06 23:46:56.202474683 -0400
@@ -49,6 +49,13 @@ extern void fool (long double);
   fool (__builtin_remquol (ARG1##L, ARG2##L, &quo)); \
 } while (0)

+#define TESTIT_REENT(FUNC,ARG1) do { \
+  int sg; \
+  foof (__builtin_##FUNC##f_r (ARG1##F, &sg)); \
+  foo (__builtin_##FUNC##_r (ARG1, &sg)); \
+  fool (__builtin_##FUNC##l_r (ARG1##L, &sg)); \
+} while (0)
+
 void bar()
 {
   /* An argument of NaN is not evaluated at compile-time.  */
@@ -266,6 +273,21 @@ void bar()
   TESTIT2 (remainder, 1.0, -0.0);
   TESTIT2 (drem, 1.0, 0.0);
   TESTIT2 (drem, 1.0, -0.0);
+
+  /* The argument to lgamma* cannot be zero or a negative integer.  */
+  TESTIT_REENT (lgamma, -4.0); /* lgamma_r */
+  TESTIT_REENT (lgamma, -3.0); /* lgamma_r */
+  TESTIT_REENT (lgamma, -2.0); /* lgamma_r */
+  TESTIT_REENT (lgamma, -1.0); /* lgamma_r */
+  TESTIT_REENT (lgamma, -0.0); /* lgamma_r */
+  TESTIT_REENT (lgamma, 0.0); /* lgamma_r */
+
+  TESTIT_REENT (gamma, -4.0); /* gamma_r */
+  TESTIT_REENT (gamma, -3.0); /* gamma_r */
+  TESTIT_REENT (gamma, -2.0); /* gamma_r */
+  TESTIT_REENT (gamma, -1.0); /* gamma_r */
+  TESTIT_REENT (gamma, -0.0); /* gamma_r */
+  TESTIT_REENT (gamma, 0.0); /* gamma_r */
 }

 /* { dg-final { scan-tree-dump-times "exp2 " 9 "original" } } */
@@ -340,4 +362,10 @@ void bar()
 /* { dg-final { scan-tree-dump-times "drem " 2 "original" } } */
 /* { dg-final { scan-tree-dump-times "dremf" 2 "original" } } */
 /* { dg-final { scan-tree-dump-times "dreml" 2 "original" } } */
+/* { dg-final { scan-tree-dump-times "lgamma_r " 6 "original" } } */
+/* { dg-final { scan-tree-dump-times "lgammaf_r" 6 "original" } } */
+/* { dg-final { scan-tree-dump-times "lgammal_r" 6 "original" } } */
+/* { dg-final { scan-tree-dump-times "_gamma_r " 6 "original" } } */
+/* { dg-final { scan-tree-dump-times "_gammaf_r" 6 "original" } } */
+/* { dg-final { scan-tree-dump-times "_gammal_r" 6 "original" } } */
 /* { dg-final { cleanup-tree-dump "original" } } */
diff -rup orig/egcc-SVN20070505/gcc/testsuite/gcc.dg/torture/builtin-math-4.c egcc-SVN20070505/gcc/testsuite/gcc.dg/torture/builtin-math-4.c
--- orig/egcc-SVN20070505/gcc/testsuite/gcc.dg/torture/builtin-math-4.c	2007-05-06 23:42:44.835144977 -0400
+++ egcc-SVN20070505/gcc/testsuite/gcc.dg/torture/builtin-math-4.c	2007-05-07 00:52:46.941083027 -0400
@@ -101,6 +101,44 @@ extern void link_error(int);
     link_error(__LINE__); \
   } while (0)

+/* Test that FUNC(ARG,&SG) == (RES) && SG == RES_SG.  */
+#define TESTIT_LGAMMA_REENT(FUNC,ARG,RES,RES_SG) do { \
+  int sg; \
+  sg = 123; \
+  if (__builtin_##FUNC##f_r(ARG##F,&sg) != RES##F \
+      || sg != RES_SG \
+      || CKSGN_F(__builtin_##FUNC##f_r(ARG##F,&sg),RES##F)) \
+    link_error(__LINE__); \
+  sg = 123; \
+  if (__builtin_##FUNC##_r(ARG,&sg) != RES \
+      || sg != RES_SG \
+      || CKSGN(__builtin_##FUNC##_r(ARG,&sg),RES)) \
+    link_error(__LINE__); \
+  sg = 123; \
+  if (__builtin_##FUNC##l_r(ARG##L,&sg) != RES##L \
+      || sg != RES_SG \
+      || CKSGN_L(__builtin_##FUNC##l_r(ARG##L,&sg),RES##L)) \
+    link_error(__LINE__); \
+  } while (0)
+
+/* Range test, check that (LOW) < FUNC(ARG,&SG) < (HI), and also test
+   that SG == RES_SG.  */
+#define TESTIT_LGAMMA_REENT_R(FUNC,ARG,LOW,HI,RES_SG) do { \
+  int sg; \
+  sg = 123; \
+  if (__builtin_##FUNC##f_r(ARG,&sg) <= (LOW) || __builtin_##FUNC##f_r(ARG,&sg) >= (HI) \
+      || sg != RES_SG) \
+    link_error(__LINE__); \
+  sg = 123; \
+  if (__builtin_##FUNC##_r(ARG,&sg) <= (LOW) || __builtin_##FUNC##_r(ARG,&sg) >= (HI) \
+      || sg != RES_SG) \
+    link_error(__LINE__); \
+  sg = 123; \
+  if (__builtin_##FUNC##l_r(ARG,&sg) <= (LOW) || __builtin_##FUNC##l_r(ARG,&sg) >= (HI) \
+      || sg != RES_SG) \
+    link_error(__LINE__); \
+  } while (0)
+
 int main (void)
 {
 #ifdef __OPTIMIZE__
@@ -247,7 +285,27 @@ int main (void)
     MAXIT(remquol, -(__INT_MAX__+1.0L), 0);
     MAXIT(remquol, -(__INT_MAX__+2.0L), -1);
   }
-#endif

+  /* These tests rely on propagating the variable sg which contains
+     signgam.  This happens only when optimization is turned on.  */
+  TESTIT_LGAMMA_REENT_R (lgamma, -2.5, -0.06, -0.05, -1); /* lgamma_r(-2.5) == -0.056... */
+  TESTIT_LGAMMA_REENT_R (lgamma, -1.5, 0.86, 0.87, 1); /* lgamma_r(-1.5) == 0.860... */
+  TESTIT_LGAMMA_REENT_R (lgamma, -0.5, 1.26, 1.27, -1); /* lgamma_r(-0.5) == 1.265... */
+  TESTIT_LGAMMA_REENT_R (lgamma, 0.5, 0.57, 0.58, 1); /* lgamma_r(0.5) == 0.572... */
+  TESTIT_LGAMMA_REENT (lgamma, 1.0, 0.0, 1); /* lgamma_r(1) == 0 */
+  TESTIT_LGAMMA_REENT_R (lgamma, 1.5, -0.13, -0.12, 1); /* lgamma_r(1.5) == -0.120... */
+  TESTIT_LGAMMA_REENT (lgamma, 2.0, 0.0, 1); /* lgamma_r(2) == 0 */
+  TESTIT_LGAMMA_REENT_R (lgamma, 2.5, 0.28, 0.29, 1); /* lgamma_r(2.5) == 0.284... */
+
+  TESTIT_LGAMMA_REENT_R (gamma, -2.5, -0.06, -0.05, -1); /* gamma_r(-2.5) == -0.056... */
+  TESTIT_LGAMMA_REENT_R (gamma, -1.5, 0.86, 0.87, 1); /* gamma_r(-1.5) == 0.860... */
+  TESTIT_LGAMMA_REENT_R (gamma, -0.5, 1.26, 1.27, -1); /* gamma_r(-0.5) == 1.265... */
+  TESTIT_LGAMMA_REENT_R (gamma, 0.5, 0.57, 0.58, 1); /* gamma_r(0.5) == 0.572... */
+  TESTIT_LGAMMA_REENT (gamma, 1.0, 0.0, 1); /* gamma_r(1) == 0 */
+  TESTIT_LGAMMA_REENT_R (gamma, 1.5, -0.13, -0.12, 1); /* gamma_r(1.5) == -0.120... */
+  TESTIT_LGAMMA_REENT (gamma, 2.0, 0.0, 1); /* gamma_r(2) == 0 */
+  TESTIT_LGAMMA_REENT_R (gamma, 2.5, 0.28, 0.29, 1); /* gamma_r(2.5) == 0.284... */
+#endif
+
   return 0;
 }

Re: [PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Richard Guenther]

On 5/7/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> This patch partly addresses PR middle-end/30250.  It uses MPFR to
> evaluate the reentrant lgamma_r/gamma_r builtins at compile-time when
> supplied with constant args.  The reentrant versions supply signgam as
> a pointer parameter, and they were slightly easier to implement than
> doing the regular lgamma functions, so I did them first.  (The plain
> lgamma/gamma patch will come hopefully soon.)
>
> The patch requires several others to apply cleanly:
> http://gcc.gnu.org/ml/gcc-patches/2007-04/msg01624.html
> http://gcc.gnu.org/ml/gcc-patches/2007-04/msg01663.html
> http://gcc.gnu.org/ml/gcc-patches/2007-05/msg00297.html
> http://gcc.gnu.org/ml/gcc-patches/2007-05/msg00320.html
>
> Patch below tested on sparc-sun-solaris2.10 in conjuction with the
> above four patches, no regressions and the new testcases all pass.  I
> also ran a preliminary version of my plain lgamma/gamma patch against
> the FP accuracy testsuite to check whether MPFR's lgamma function was
> producing correct results back to GCC.  It was, all inputs generated
> "perfect" results.
>
> Okay for mainline?
>
>                 Thanks,
>                 --Kaveh
>
> :ADDPATCH middle-end:
>
>
> 2007-05-07  Kaveh R. Ghazi  <ghazi@caip.rutgers.edu>
>
>         PR middle-end/30250
>         * builtins.c (do_mpfr_lgamma_r): New.
>         (fold_builtin_2): Handle builtin gamma_r/lgamma_r.
>         * tree.h (CASE_FLT_FN_REENT): New.
>
> testsuite:
>         * gcc.dg/torture/builtin-math-2.c: Add gamma_r/lgamma_r tests.
>         * gcc.dg/torture/builtin-math-4.c: Likewise.
>
> diff -rup orig/egcc-SVN20070505/gcc/builtins.c egcc-SVN20070505/gcc/builtins.c
> --- orig/egcc-SVN20070505/gcc/builtins.c        2007-05-06 23:43:01.621873788 -0400
> +++ egcc-SVN20070505/gcc/builtins.c     2007-05-07 00:55:32.966335476 -0400
> @@ -234,6 +234,7 @@ static tree do_mpfr_bessel_n (tree, tree
>                               int (*)(mpfr_ptr, long, mpfr_srcptr, mp_rnd_t),
>                               const REAL_VALUE_TYPE *, bool);
>  static tree do_mpfr_remquo (tree, tree, tree);
> +static tree do_mpfr_lgamma_r (tree, tree, tree);
>  #endif
>
>  /* Return true if NODE should be considered for inline expansion regardless
> @@ -9862,6 +9863,13 @@ fold_builtin_2 (tree fndecl, tree arg0,
>            && validate_arg(arg1, REAL_TYPE))
>          return do_mpfr_arg2 (arg0, arg1, type, mpfr_remainder);
>      break;
> +
> +    CASE_FLT_FN_REENT (BUILT_IN_GAMMA): /* GAMMA_R */
> +    CASE_FLT_FN_REENT (BUILT_IN_LGAMMA): /* LGAMMA_R */
> +      if (validate_arg (arg0, REAL_TYPE)
> +         && validate_arg(arg1, POINTER_TYPE))
> +       return do_mpfr_lgamma_r (arg0, arg1, type);
> +    break;
>  #endif
>
>      CASE_FLT_FN (BUILT_IN_ATAN2):
> @@ -12616,4 +12624,66 @@ do_mpfr_remquo (tree arg0, tree arg1, tr
>      }
>    return result;
>  }
> +
> +/* If ARG is a REAL_CST, call mpfr_lgamma() on it and return the
> +   resulting value as a tree with type TYPE.  The mpfr precision is
> +   set to the precision of TYPE.  We assume that this mpfr function
> +   returns zero if the result could be calculated exactly within the
> +   requested precision.  In addition, the integer pointer represented
> +   by ARG_SG will be dereferenced and set to the appropriate signgam
> +   (-1,1) value.  */
> +
> +static tree
> +do_mpfr_lgamma_r (tree arg, tree arg_sg, tree type)
> +{
> +  tree result = NULL_TREE;
> +
> +  STRIP_NOPS (arg);
> +
> +  /* To proceed, MPFR must exactly represent the target floating point
> +     format, which only happens when the target base equals two.  */
> +  if (REAL_MODE_FORMAT (TYPE_MODE (type))->b == 2
> +      && TREE_CODE (arg) == REAL_CST && !TREE_OVERFLOW (arg))
> +    {
> +      const REAL_VALUE_TYPE *const ra = TREE_REAL_CST_PTR (arg);
> +
> +      /* In addition to NaN and Inf, the argument cannot be zero or a
> +        negative integer.  */
> +      if (!real_isnan (ra) && !real_isinf (ra)
> +         && ra->cl != rvc_zero
> +         && !(real_isneg(ra) && real_isinteger(ra, TYPE_MODE (type))))
> +        {
> +         const int prec = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
> +         int inexact, sg;
> +         mpfr_t m;
> +         tree result_lg;
> +
> +         mpfr_init2 (m, prec);
> +         mpfr_from_real (m, ra, GMP_RNDN);
> +         mpfr_clear_flags ();
> +         inexact = mpfr_lgamma (m, &sg, m, GMP_RNDN);
> +         result_lg = do_mpfr_ckconv (m, type, inexact);
> +         mpfr_clear (m);
> +         if (result_lg)
> +           {
> +             /* Dereference the arg_sg pointer argument.  */
> +             arg_sg = build_fold_indirect_ref (arg_sg);
> +             /* Proceed iff a valid pointer type was passed in.  */
> +             if (TYPE_MAIN_VARIANT (TREE_TYPE (arg_sg)) == integer_type_node)

This check should be done earlier based on the pointed to type of the
argument.

Ok with this change.  Please make sure to XFAIL testcases that rely
on new mpfr.

Thanks,
Richard.

:REVIEWMAIL:

Re: [PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Kaveh R. GHAZI]

On Thu, 17 May 2007, Richard Guenther wrote:

> > +             /* Dereference the arg_sg pointer argument.  */
> > +             arg_sg = build_fold_indirect_ref (arg_sg);
> > +             /* Proceed iff a valid pointer type was passed in.  */
> > +             if (TYPE_MAIN_VARIANT (TREE_TYPE (arg_sg)) == integer_type_node)
>
> This check should be done earlier based on the pointed to type of the
> argument. Ok with this change.

Done.  I installed the revised patch below.  Thanks again for your
reviews.

		Regards,
		--Kaveh

2007-05-07  Kaveh R. Ghazi  <ghazi@caip.rutgers.edu>

	PR middle-end/30250
	* builtins.c (do_mpfr_lgamma_r): New.
	(fold_builtin_2): Handle builtin gamma_r/lgamma_r.
	* tree.h (CASE_FLT_FN_REENT): New.

testsuite:
	* gcc.dg/torture/builtin-math-2.c: Add gamma_r/lgamma_r tests.
	* gcc.dg/torture/builtin-math-4.c: Likewise.

diff -rup orig/egcc-SVN20070517/gcc/builtins.c egcc-SVN20070517/gcc/builtins.c
--- orig/egcc-SVN20070517/gcc/builtins.c	2007-05-17 23:03:48.000000000 -0400
+++ egcc-SVN20070517/gcc/builtins.c	2007-05-18 01:05:54.447296033 -0400
@@ -236,6 +236,7 @@ static tree do_mpfr_bessel_n (tree, tree
 			      int (*)(mpfr_ptr, long, mpfr_srcptr, mp_rnd_t),
 			      const REAL_VALUE_TYPE *, bool);
 static tree do_mpfr_remquo (tree, tree, tree);
+static tree do_mpfr_lgamma_r (tree, tree, tree);
 #endif

 /* Return true if NODE should be considered for inline expansion regardless
@@ -9935,6 +9936,13 @@ fold_builtin_2 (tree fndecl, tree arg0,
           && validate_arg(arg1, REAL_TYPE))
         return do_mpfr_arg2 (arg0, arg1, type, mpfr_remainder);
     break;
+
+    CASE_FLT_FN_REENT (BUILT_IN_GAMMA): /* GAMMA_R */
+    CASE_FLT_FN_REENT (BUILT_IN_LGAMMA): /* LGAMMA_R */
+      if (validate_arg (arg0, REAL_TYPE)
+	  && validate_arg(arg1, POINTER_TYPE))
+	return do_mpfr_lgamma_r (arg0, arg1, type);
+    break;
 #endif

     CASE_FLT_FN (BUILT_IN_ATAN2):
@@ -12692,4 +12700,67 @@ do_mpfr_remquo (tree arg0, tree arg1, tr
     }
   return result;
 }
+
+/* If ARG is a REAL_CST, call mpfr_lgamma() on it and return the
+   resulting value as a tree with type TYPE.  The mpfr precision is
+   set to the precision of TYPE.  We assume that this mpfr function
+   returns zero if the result could be calculated exactly within the
+   requested precision.  In addition, the integer pointer represented
+   by ARG_SG will be dereferenced and set to the appropriate signgam
+   (-1,1) value.  */
+
+static tree
+do_mpfr_lgamma_r (tree arg, tree arg_sg, tree type)
+{
+  tree result = NULL_TREE;
+
+  STRIP_NOPS (arg);
+
+  /* To proceed, MPFR must exactly represent the target floating point
+     format, which only happens when the target base equals two.  Also
+     verify ARG is a constant and that ARG_SG is an int pointer.  */
+  if (REAL_MODE_FORMAT (TYPE_MODE (type))->b == 2
+      && TREE_CODE (arg) == REAL_CST && !TREE_OVERFLOW (arg)
+      && TREE_CODE (TREE_TYPE (arg_sg)) == POINTER_TYPE
+      && TYPE_MAIN_VARIANT (TREE_TYPE (TREE_TYPE (arg_sg))) == integer_type_node)
+    {
+      const REAL_VALUE_TYPE *const ra = TREE_REAL_CST_PTR (arg);
+
+      /* In addition to NaN and Inf, the argument cannot be zero or a
+	 negative integer.  */
+      if (!real_isnan (ra) && !real_isinf (ra)
+	  && ra->cl != rvc_zero
+	  && !(real_isneg(ra) && real_isinteger(ra, TYPE_MODE (type))))
+        {
+	  const int prec = REAL_MODE_FORMAT (TYPE_MODE (type))->p;
+	  int inexact, sg;
+	  mpfr_t m;
+	  tree result_lg;
+
+	  mpfr_init2 (m, prec);
+	  mpfr_from_real (m, ra, GMP_RNDN);
+	  mpfr_clear_flags ();
+	  inexact = mpfr_lgamma (m, &sg, m, GMP_RNDN);
+	  result_lg = do_mpfr_ckconv (m, type, inexact);
+	  mpfr_clear (m);
+	  if (result_lg)
+	    {
+	      tree result_sg;
+
+	      /* Dereference the arg_sg pointer argument.  */
+	      arg_sg = build_fold_indirect_ref (arg_sg);
+	      /* Assign the signgam value into *arg_sg. */
+	      result_sg = fold_build2 (MODIFY_EXPR,
+				       TREE_TYPE (arg_sg), arg_sg,
+				       build_int_cst (NULL, sg));
+	      TREE_SIDE_EFFECTS (result_sg) = 1;
+	      /* Combine the signgam assignment with the lgamma result.  */
+	      result = non_lvalue (fold_build2 (COMPOUND_EXPR, type,
+						result_sg, result_lg));
+	    }
+	}
+    }
+
+  return result;
+}
 #endif
diff -rup orig/egcc-SVN20070517/gcc/tree.h egcc-SVN20070517/gcc/tree.h
--- orig/egcc-SVN20070517/gcc/tree.h	2007-05-03 03:14:35.000000000 -0400
+++ egcc-SVN20070517/gcc/tree.h	2007-05-18 00:27:19.050082272 -0400
@@ -273,6 +273,7 @@ extern const char * built_in_names[(int)
 #define BUILTIN_ROOT_P(FN) (BUILTIN_SQRT_P (FN) || BUILTIN_CBRT_P (FN))

 #define CASE_FLT_FN(FN) case FN: case FN##F: case FN##L
+#define CASE_FLT_FN_REENT(FN) case FN##_R: case FN##F_R: case FN##L_R
 #define CASE_INT_FN(FN) case FN: case FN##L: case FN##LL

 /* An array of _DECL trees for the above.  */
diff -rup orig/egcc-SVN20070517/gcc/testsuite/gcc.dg/torture/builtin-math-2.c egcc-SVN20070517/gcc/testsuite/gcc.dg/torture/builtin-math-2.c
--- orig/egcc-SVN20070517/gcc/testsuite/gcc.dg/torture/builtin-math-2.c	2007-05-17 23:03:22.000000000 -0400
+++ egcc-SVN20070517/gcc/testsuite/gcc.dg/torture/builtin-math-2.c	2007-05-18 00:27:19.051938461 -0400
@@ -49,6 +49,13 @@ extern void fool (long double);
   fool (__builtin_remquol (ARG1##L, ARG2##L, &quo)); \
 } while (0)

+#define TESTIT_REENT(FUNC,ARG1) do { \
+  int sg; \
+  foof (__builtin_##FUNC##f_r (ARG1##F, &sg)); \
+  foo (__builtin_##FUNC##_r (ARG1, &sg)); \
+  fool (__builtin_##FUNC##l_r (ARG1##L, &sg)); \
+} while (0)
+
 void bar()
 {
   /* An argument of NaN is not evaluated at compile-time.  */
@@ -266,6 +273,21 @@ void bar()
   TESTIT2 (remainder, 1.0, -0.0);
   TESTIT2 (drem, 1.0, 0.0);
   TESTIT2 (drem, 1.0, -0.0);
+
+  /* The argument to lgamma* cannot be zero or a negative integer.  */
+  TESTIT_REENT (lgamma, -4.0); /* lgamma_r */
+  TESTIT_REENT (lgamma, -3.0); /* lgamma_r */
+  TESTIT_REENT (lgamma, -2.0); /* lgamma_r */
+  TESTIT_REENT (lgamma, -1.0); /* lgamma_r */
+  TESTIT_REENT (lgamma, -0.0); /* lgamma_r */
+  TESTIT_REENT (lgamma, 0.0); /* lgamma_r */
+
+  TESTIT_REENT (gamma, -4.0); /* gamma_r */
+  TESTIT_REENT (gamma, -3.0); /* gamma_r */
+  TESTIT_REENT (gamma, -2.0); /* gamma_r */
+  TESTIT_REENT (gamma, -1.0); /* gamma_r */
+  TESTIT_REENT (gamma, -0.0); /* gamma_r */
+  TESTIT_REENT (gamma, 0.0); /* gamma_r */
 }

 /* { dg-final { scan-tree-dump-times "exp2 " 9 "original" } } */
@@ -340,4 +362,10 @@ void bar()
 /* { dg-final { scan-tree-dump-times "drem " 2 "original" } } */
 /* { dg-final { scan-tree-dump-times "dremf" 2 "original" } } */
 /* { dg-final { scan-tree-dump-times "dreml" 2 "original" } } */
+/* { dg-final { scan-tree-dump-times "lgamma_r " 6 "original" } } */
+/* { dg-final { scan-tree-dump-times "lgammaf_r" 6 "original" } } */
+/* { dg-final { scan-tree-dump-times "lgammal_r" 6 "original" } } */
+/* { dg-final { scan-tree-dump-times "_gamma_r " 6 "original" } } */
+/* { dg-final { scan-tree-dump-times "_gammaf_r" 6 "original" } } */
+/* { dg-final { scan-tree-dump-times "_gammal_r" 6 "original" } } */
 /* { dg-final { cleanup-tree-dump "original" } } */
diff -rup orig/egcc-SVN20070517/gcc/testsuite/gcc.dg/torture/builtin-math-4.c egcc-SVN20070517/gcc/testsuite/gcc.dg/torture/builtin-math-4.c
--- orig/egcc-SVN20070517/gcc/testsuite/gcc.dg/torture/builtin-math-4.c	2007-05-17 23:03:22.000000000 -0400
+++ egcc-SVN20070517/gcc/testsuite/gcc.dg/torture/builtin-math-4.c	2007-05-18 00:27:19.053318162 -0400
@@ -103,6 +103,44 @@ extern void link_error(int);
     link_error(__LINE__); \
   } while (0)

+/* Test that FUNC(ARG,&SG) == (RES) && SG == RES_SG.  */
+#define TESTIT_LGAMMA_REENT(FUNC,ARG,RES,RES_SG) do { \
+  int sg; \
+  sg = 123; \
+  if (__builtin_##FUNC##f_r(ARG##F,&sg) != RES##F \
+      || sg != RES_SG \
+      || CKSGN_F(__builtin_##FUNC##f_r(ARG##F,&sg),RES##F)) \
+    link_error(__LINE__); \
+  sg = 123; \
+  if (__builtin_##FUNC##_r(ARG,&sg) != RES \
+      || sg != RES_SG \
+      || CKSGN(__builtin_##FUNC##_r(ARG,&sg),RES)) \
+    link_error(__LINE__); \
+  sg = 123; \
+  if (__builtin_##FUNC##l_r(ARG##L,&sg) != RES##L \
+      || sg != RES_SG \
+      || CKSGN_L(__builtin_##FUNC##l_r(ARG##L,&sg),RES##L)) \
+    link_error(__LINE__); \
+  } while (0)
+
+/* Range test, check that (LOW) < FUNC(ARG,&SG) < (HI), and also test
+   that SG == RES_SG.  */
+#define TESTIT_LGAMMA_REENT_R(FUNC,ARG,LOW,HI,RES_SG) do { \
+  int sg; \
+  sg = 123; \
+  if (__builtin_##FUNC##f_r(ARG,&sg) <= (LOW) || __builtin_##FUNC##f_r(ARG,&sg) >= (HI) \
+      || sg != RES_SG) \
+    link_error(__LINE__); \
+  sg = 123; \
+  if (__builtin_##FUNC##_r(ARG,&sg) <= (LOW) || __builtin_##FUNC##_r(ARG,&sg) >= (HI) \
+      || sg != RES_SG) \
+    link_error(__LINE__); \
+  sg = 123; \
+  if (__builtin_##FUNC##l_r(ARG,&sg) <= (LOW) || __builtin_##FUNC##l_r(ARG,&sg) >= (HI) \
+      || sg != RES_SG) \
+    link_error(__LINE__); \
+  } while (0)
+
 int main (void)
 {
 #ifdef __OPTIMIZE__
@@ -249,7 +287,27 @@ int main (void)
     MAXIT(remquol, -(__INT_MAX__+1.0L), 0);
     MAXIT(remquol, -(__INT_MAX__+2.0L), -1);
   }
-#endif

+  /* These tests rely on propagating the variable sg which contains
+     signgam.  This happens only when optimization is turned on.  */
+  TESTIT_LGAMMA_REENT_R (lgamma, -2.5, -0.06, -0.05, -1); /* lgamma_r(-2.5) == -0.056... */
+  TESTIT_LGAMMA_REENT_R (lgamma, -1.5, 0.86, 0.87, 1); /* lgamma_r(-1.5) == 0.860... */
+  TESTIT_LGAMMA_REENT_R (lgamma, -0.5, 1.26, 1.27, -1); /* lgamma_r(-0.5) == 1.265... */
+  TESTIT_LGAMMA_REENT_R (lgamma, 0.5, 0.57, 0.58, 1); /* lgamma_r(0.5) == 0.572... */
+  TESTIT_LGAMMA_REENT (lgamma, 1.0, 0.0, 1); /* lgamma_r(1) == 0 */
+  TESTIT_LGAMMA_REENT_R (lgamma, 1.5, -0.13, -0.12, 1); /* lgamma_r(1.5) == -0.120... */
+  TESTIT_LGAMMA_REENT (lgamma, 2.0, 0.0, 1); /* lgamma_r(2) == 0 */
+  TESTIT_LGAMMA_REENT_R (lgamma, 2.5, 0.28, 0.29, 1); /* lgamma_r(2.5) == 0.284... */
+
+  TESTIT_LGAMMA_REENT_R (gamma, -2.5, -0.06, -0.05, -1); /* gamma_r(-2.5) == -0.056... */
+  TESTIT_LGAMMA_REENT_R (gamma, -1.5, 0.86, 0.87, 1); /* gamma_r(-1.5) == 0.860... */
+  TESTIT_LGAMMA_REENT_R (gamma, -0.5, 1.26, 1.27, -1); /* gamma_r(-0.5) == 1.265... */
+  TESTIT_LGAMMA_REENT_R (gamma, 0.5, 0.57, 0.58, 1); /* gamma_r(0.5) == 0.572... */
+  TESTIT_LGAMMA_REENT (gamma, 1.0, 0.0, 1); /* gamma_r(1) == 0 */
+  TESTIT_LGAMMA_REENT_R (gamma, 1.5, -0.13, -0.12, 1); /* gamma_r(1.5) == -0.120... */
+  TESTIT_LGAMMA_REENT (gamma, 2.0, 0.0, 1); /* gamma_r(2) == 0 */
+  TESTIT_LGAMMA_REENT_R (gamma, 2.5, 0.28, 0.29, 1); /* gamma_r(2.5) == 0.284... */
+#endif
+
   return 0;
 }

Re: [PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Richard Guenther]

On 5/19/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> On Thu, 17 May 2007, Richard Guenther wrote:
>
> > > +             /* Dereference the arg_sg pointer argument.  */
> > > +             arg_sg = build_fold_indirect_ref (arg_sg);
> > > +             /* Proceed iff a valid pointer type was passed in.  */
> > > +             if (TYPE_MAIN_VARIANT (TREE_TYPE (arg_sg)) == integer_type_node)
> >
> > This check should be done earlier based on the pointed to type of the
> > argument. Ok with this change.
>
> Done.  I installed the revised patch below.  Thanks again for your
> reviews.

This:

/* Expect failures at least until mpfr-2.3.0 is released. */
/* { dg-xfail-if "mpfr-2.3.0" { *-*-* } } */

doesn't work.  Or did you forget a testsuite/lib patch?

Richard.

Re: [PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Kaveh R. GHAZI]

On Wed, 23 May 2007, Richard Guenther wrote:

> On 5/19/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> > On Thu, 17 May 2007, Richard Guenther wrote:
> >
> > > > +             /* Dereference the arg_sg pointer argument.  */
> > > > +             arg_sg = build_fold_indirect_ref (arg_sg);
> > > > +             /* Proceed iff a valid pointer type was passed in.  */
> > > > +             if (TYPE_MAIN_VARIANT (TREE_TYPE (arg_sg)) == integer_type_node)
> > >
> > > This check should be done earlier based on the pointed to type of the
> > > argument. Ok with this change.
> >
> > Done.  I installed the revised patch below.  Thanks again for your
> > reviews.
>
> This:
>
> /* Expect failures at least until mpfr-2.3.0 is released. */
> /* { dg-xfail-if "mpfr-2.3.0" { *-*-* } } */
>
> doesn't work.  Or did you forget a testsuite/lib patch?
> Richard.

There was no testsuite patch involved, dg-xfail-if should work just fine.
Can you provide some more info as to what exactly doesn't work?  Here are
some 4.3.0 testresults posted for the trunk dated 20070523 on various
platforms.  None of them show builtin-math-4.c failing:

http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01156.html
http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01155.html
http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01153.html
http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01151.html
http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01143.html
http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01132.html

Ah!  I found one from Joe Buck that doesn't pickup the xfail:
http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01149.html

But I can't explain why.  Maybe he's using an outdated version of dejagnu?

Joe?

		Thanks,
		--Kaveh
--
Kaveh R. Ghazi			ghazi@caip.rutgers.edu

Re: [PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Richard Guenther]

On 5/24/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> On Wed, 23 May 2007, Richard Guenther wrote:
>
> > This:
> >
> > /* Expect failures at least until mpfr-2.3.0 is released. */
> > /* { dg-xfail-if "mpfr-2.3.0" { *-*-* } } */
> >
> > doesn't work.  Or did you forget a testsuite/lib patch?
> > Richard.
>
> There was no testsuite patch involved, dg-xfail-if should work just fine.
> Can you provide some more info as to what exactly doesn't work?  Here are
> some 4.3.0 testresults posted for the trunk dated 20070523 on various
> platforms.  None of them show builtin-math-4.c failing:
>
> http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01156.html
> http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01155.html
> http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01153.html
> http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01151.html
> http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01143.html
> http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01132.html
>
> Ah!  I found one from Joe Buck that doesn't pickup the xfail:
> http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01149.html
>
> But I can't explain why.  Maybe he's using an outdated version of dejagnu?

Hm, I don't know how this dg-xfail-if is supposed to work - where does it
get the mpfr-2.3.0 thing from?  I'm using the (outdated, but the docs
only list 1.4.4 as requirement) dejagnu 1.4.3 and mpfr 2.2.0.

Richard.

Re: [PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Richard Guenther]

On 5/24/07, Richard Guenther <richard.guenther@gmail.com> wrote:
> On 5/24/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> > On Wed, 23 May 2007, Richard Guenther wrote:
> >
> > > This:
> > >
> > > /* Expect failures at least until mpfr-2.3.0 is released. */
> > > /* { dg-xfail-if "mpfr-2.3.0" { *-*-* } } */
> > >
> > > doesn't work.  Or did you forget a testsuite/lib patch?
> > > Richard.
> >
> > There was no testsuite patch involved, dg-xfail-if should work just fine.
> > Can you provide some more info as to what exactly doesn't work?  Here are
> > some 4.3.0 testresults posted for the trunk dated 20070523 on various
> > platforms.  None of them show builtin-math-4.c failing:
> >
> > http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01156.html
> > http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01155.html
> > http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01153.html
> > http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01151.html
> > http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01143.html
> > http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01132.html
> >
> > Ah!  I found one from Joe Buck that doesn't pickup the xfail:
> > http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01149.html
> >
> > But I can't explain why.  Maybe he's using an outdated version of dejagnu?
>
> Hm, I don't know how this dg-xfail-if is supposed to work - where does it
> get the mpfr-2.3.0 thing from?  I'm using the (outdated, but the docs
> only list 1.4.4 as requirement) dejagnu 1.4.3 and mpfr 2.2.0.

Updating to dejagnu 1.4.4 fixed it.  Luckily that still works with old tcl ;)

Richard.

Re: [PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Joe Buck]

On Thu, May 24, 2007 at 10:11:58AM +0200, Richard Guenther wrote:
> On 5/24/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> >On Wed, 23 May 2007, Richard Guenther wrote:
> >
> >> This:
> >>
> >> /* Expect failures at least until mpfr-2.3.0 is released. */
> >> /* { dg-xfail-if "mpfr-2.3.0" { *-*-* } } */
> >>
> >> doesn't work.  Or did you forget a testsuite/lib patch?
> >> Richard.
> >
> >There was no testsuite patch involved, dg-xfail-if should work just fine.
> >Can you provide some more info as to what exactly doesn't work?  Here are
> >some 4.3.0 testresults posted for the trunk dated 20070523 on various
> >platforms.  None of them show builtin-math-4.c failing:
> >
> >http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01156.html
> >http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01155.html
> >http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01153.html
> >http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01151.html
> >http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01143.html
> >http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01132.html
> >
> >Ah!  I found one from Joe Buck that doesn't pickup the xfail:
> >http://gcc.gnu.org/ml/gcc-testresults/2007-05/msg01149.html
> >
> >But I can't explain why.  Maybe he's using an outdated version of dejagnu?

I do have a fairly old version; I'll see about replacing it with
something newer.

Re: [PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Janis Johnson]

On Thu, May 24, 2007 at 10:15:22AM +0200, Richard Guenther wrote:
> On 5/24/07, Richard Guenther <richard.guenther@gmail.com> wrote:
> >On 5/24/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> >> On Wed, 23 May 2007, Richard Guenther wrote:
> >>
> >> > This:
> >> >
> >> > /* Expect failures at least until mpfr-2.3.0 is released. */
> >> > /* { dg-xfail-if "mpfr-2.3.0" { *-*-* } } */
> >> >
> >> > doesn't work.  Or did you forget a testsuite/lib patch?

This should be:

  /* { dg-xfail-if "mpfr-2.3.0" { *-*-* } { "*" } { "" } } */

The "mpfr-2.3.0" is a comment.  The lists of included and excluded
options are not optional; I'm surprised it works.

> Updating to dejagnu 1.4.4 fixed it.  Luckily that still works with old tcl 
> ;)

Magic.

Janis

Re: [PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Kaveh R. GHAZI]

On Thu, 24 May 2007, Janis Johnson wrote:

> On Thu, May 24, 2007 at 10:15:22AM +0200, Richard Guenther wrote:
> > On 5/24/07, Richard Guenther <richard.guenther@gmail.com> wrote:
> > >On 5/24/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> > >> On Wed, 23 May 2007, Richard Guenther wrote:
> > >>
> > >> > This:
> > >> >
> > >> > /* Expect failures at least until mpfr-2.3.0 is released. */
> > >> > /* { dg-xfail-if "mpfr-2.3.0" { *-*-* } } */
> > >> >
> > >> > doesn't work.  Or did you forget a testsuite/lib patch?
>
> This should be:
>   /* { dg-xfail-if "mpfr-2.3.0" { *-*-* } { "*" } { "" } } */
> The "mpfr-2.3.0" is a comment.  The lists of included and excluded
> options are not optional; I'm surprised it works.
> > Updating to dejagnu 1.4.4 fixed it.  Luckily that still works with old tcl
> > ;)
> Magic.
> Janis

Oh okay, I thought those flags were optional because I saw it like that
somewhere else (and of course it passed my own testing).  Search, search,
search... ah, here's one:  gcc.dg/Warray-bounds.c.

Anyway, I'm happy to fix the one I introduced.  I also made the comment
more verbose to clarify that it is in fact a comment and not some magic
pattern search for an mpfr version.

Tested via "make check" using mpfr-2.2.1 and mpfr-2.3.0-svn.  I still get
XFAIL and XPASS respectively as desired.

Okay for mainline?

		Thanks,
		--Kaveh


2007-05-24  Kaveh R. Ghazi  <ghazi@caip.rutgers.edu>

	* gcc.dg/torture/builtin-math-4.c: Fix dg-xfail-if.

diff -rup orig/egcc-SVN20070523/gcc/testsuite/gcc.dg/torture/builtin-math-4.c egcc-SVN20070523/gcc/testsuite/gcc.dg/torture/builtin-math-4.c
--- orig/egcc-SVN20070523/gcc/testsuite/gcc.dg/torture/builtin-math-4.c	2007-05-19 00:21:55.000000000 -0400
+++ egcc-SVN20070523/gcc/testsuite/gcc.dg/torture/builtin-math-4.c	2007-05-24 17:03:41.019767099 -0400
@@ -8,7 +8,7 @@

 /* { dg-do link } */
 /* Expect failures at least until mpfr-2.3.0 is released. */
-/* { dg-xfail-if "mpfr-2.3.0" { *-*-* } } */
+/* { dg-xfail-if "This test requires mpfr-2.3.0" { *-*-* } { "*" } { "" } } */

 /* All references to link_error should go away at compile-time.  */
 extern void link_error(int);

Re: [PATCH]: PR middle-end/30250, use MPFR for lgamma_r/gamma_r

[Janis Johnson]

On Thu, May 24, 2007 at 05:09:11PM -0400, Kaveh R. GHAZI wrote:
> On Thu, 24 May 2007, Janis Johnson wrote:
> 
> > On Thu, May 24, 2007 at 10:15:22AM +0200, Richard Guenther wrote:
> > > On 5/24/07, Richard Guenther <richard.guenther@gmail.com> wrote:
> > > >On 5/24/07, Kaveh R. GHAZI <ghazi@caip.rutgers.edu> wrote:
> > > >> On Wed, 23 May 2007, Richard Guenther wrote:
> > > >>
> > > >> > This:
> > > >> >
> > > >> > /* Expect failures at least until mpfr-2.3.0 is released. */
> > > >> > /* { dg-xfail-if "mpfr-2.3.0" { *-*-* } } */
> > > >> >
> > > >> > doesn't work.  Or did you forget a testsuite/lib patch?
> >
> > This should be:
> >   /* { dg-xfail-if "mpfr-2.3.0" { *-*-* } { "*" } { "" } } */
> > The "mpfr-2.3.0" is a comment.  The lists of included and excluded
> > options are not optional; I'm surprised it works.
> > > Updating to dejagnu 1.4.4 fixed it.  Luckily that still works with old tcl
> > > ;)
> > Magic.
> > Janis
> 
> Oh okay, I thought those flags were optional because I saw it like that
> somewhere else (and of course it passed my own testing).  Search, search,
> search... ah, here's one:  gcc.dg/Warray-bounds.c.
> 
> Anyway, I'm happy to fix the one I introduced.  I also made the comment
> more verbose to clarify that it is in fact a comment and not some magic
> pattern search for an mpfr version.
> 
> Tested via "make check" using mpfr-2.2.1 and mpfr-2.3.0-svn.  I still get
> XFAIL and XPASS respectively as desired.
> 
> Okay for mainline?

Yes, thanks.

Janis