diff mbox series

Only allow (int)trunc(x) to (int)x simplification with -ffp-int-builtin-inexact [PR107723]

Message ID 20231124053241.64194-1-xry111@xry111.site
State Accepted
Headers
Series Only allow (int)trunc(x) to (int)x simplification with -ffp-int-builtin-inexact [PR107723] |

Checks

Context Check Description
snail/gcc-patch-check success Github commit url

Commit Message

Xi Ruoyao Nov. 24, 2023, 5:28 a.m. UTC 
  With -fno-fp-int-builtin-inexact, trunc is not allowed to raise
FE_INEXACT and it should produce an integral result (if the input is not
NaN or Inf).  Thus FE_INEXACT should not be raised.

But (int)x may raise FE_INEXACT when x is a non-integer, non-NaN, and
non-Inf value.  C23 recommends to do so in a footnote.

Thus we should not simplify (int)trunc(x) to (int)x if
-fno-fp-int-builtin-inexact.

gcc/ChangeLog:

	PR middle-end/107723
	* convert.cc (convert_to_integer_1) [case BUILT_IN_TRUNC]: Break
	early if !flag_fp_int_builtin_inexact.

gcc/testsuite/ChangeLog:

	PR middle-end/107723
	* gcc.dg/torture/builtin-fp-int-inexact-trunc.c: New test.
---

Bootstrapped and regtested on x86_64-linux-gnu.  Ok for trunk?

 gcc/convert.cc                                       |  3 ++-
 .../gcc.dg/torture/builtin-fp-int-inexact-trunc.c    | 12 ++++++++++++
 2 files changed, 14 insertions(+), 1 deletion(-)
 create mode 100644 gcc/testsuite/gcc.dg/torture/builtin-fp-int-inexact-trunc.c

Comments

Xi Ruoyao Nov. 24, 2023, 8:31 a.m. UTC | #1 
On Fri, 2023-11-24 at 09:27 +0100, Richard Biener wrote:
> > diff --git a/gcc/convert.cc b/gcc/convert.cc
> > index 46c8bcb31f8..31f63ea783b 100644
> > --- a/gcc/convert.cc
> > +++ b/gcc/convert.cc
> > @@ -591,7 +591,8 @@ convert_to_integer_1 (tree type, tree expr, bool dofold)
> >          CASE_FLT_FN (BUILT_IN_TRUNC):
> >          CASE_FLT_FN_FLOATN_NX (BUILT_IN_TRUNC):
> >            if (call_expr_nargs (s_expr) != 1
> > -             || !SCALAR_FLOAT_TYPE_P (TREE_TYPE (CALL_EXPR_ARG (s_expr, 0))))
> > +             || !SCALAR_FLOAT_TYPE_P (TREE_TYPE (CALL_EXPR_ARG (s_expr, 0)))
> > +             || !flag_fp_int_builtin_inexact)
> 
> shouldn't that be (!flag_fp_int_builtn_inexact && flag_trapping_math)?

Ah, indeed it should be.

But I'm wondering why we don't make -fno-trapping-math automatically set
-ffp-int-builtin-exact (and/or unset -fno-fp-int-builtin-exact)...
diff mbox series

Patch

diff --git a/gcc/convert.cc b/gcc/convert.cc
index 46c8bcb31f8..31f63ea783b 100644
--- a/gcc/convert.cc
+++ b/gcc/convert.cc
@@ -591,7 +591,8 @@  convert_to_integer_1 (tree type, tree expr, bool dofold)
 	CASE_FLT_FN (BUILT_IN_TRUNC):
 	CASE_FLT_FN_FLOATN_NX (BUILT_IN_TRUNC):
 	  if (call_expr_nargs (s_expr) != 1
-	      || !SCALAR_FLOAT_TYPE_P (TREE_TYPE (CALL_EXPR_ARG (s_expr, 0))))
+	      || !SCALAR_FLOAT_TYPE_P (TREE_TYPE (CALL_EXPR_ARG (s_expr, 0)))
+	      || !flag_fp_int_builtin_inexact)
 	    break;
 	  return convert_to_integer_1 (type, CALL_EXPR_ARG (s_expr, 0),
 				       dofold);
diff --git a/gcc/testsuite/gcc.dg/torture/builtin-fp-int-inexact-trunc.c b/gcc/testsuite/gcc.dg/torture/builtin-fp-int-inexact-trunc.c
new file mode 100644
index 00000000000..09731183621
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/torture/builtin-fp-int-inexact-trunc.c
@@ -0,0 +1,12 @@ 
+/* Test -fno-fp-int-builtin-inexact.  */
+/* { dg-do compile } */
+/* { dg-options "-fno-fp-int-builtin-inexact -fdump-tree-original" } */
+
+long
+x (double y)
+{
+  return __builtin_trunc (y);
+}
+
+/* Optimization should not discard the __builtin_trunc call.  */
+/* { dg-final { scan-tree-dump "__builtin_trunc" "original" } } */