diff mbox series

middle-end/105376 - invalid REAL_CST for DFP constant

Message ID 20220428071305.E784413491@imap2.suse-dmz.suse.de
State Committed
Commit e27eef7478f30ea79048dbde3317e89679d75a6e
Headers
Series middle-end/105376 - invalid REAL_CST for DFP constant |

Commit Message

Richard Biener April 28, 2022, 7:13 a.m. UTC 
  We are eventually ICEing in decimal_to_decnumber on non-decimal
REAL_VALUE_TYPE that creep in from uses of build_real (..., dconst*)
for DFP types.  The following extends the decimal_to_decnumber
special-casing of dconst* to build_real, avoiding the bogus REAL_CSTs
from creeping into the IL and modified to ones not handled by
the decimal_to_decnumber special casing.  It also makes sure to
ICE for not handled dconst* values at the point we build the REAL_CST.

Bootstrapped and tested on x86_64-unknown-linux-gnu, OK for stage1
and eventual backporting (I'm considering making the gcc_unreachable
conditional on flag_checking on release branches).

Thanks,
Richard.

2022-04-27  Richard Biener  <rguenther@suse.de>

	PR middle-end/105376
	* tree.cc (build_real): Special case dconst* arguments
	for decimal floating point types.

	* gcc.dg/pr105376.c: New testcase.
---
 gcc/testsuite/gcc.dg/pr105376.c |  9 +++++++++
 gcc/tree.cc                     | 21 +++++++++++++++++++++
 2 files changed, 30 insertions(+)
 create mode 100644 gcc/testsuite/gcc.dg/pr105376.c

Comments

Joseph Myers April 28, 2022, 9:58 p.m. UTC | #1 
On Thu, 28 Apr 2022, Richard Biener via Gcc-patches wrote:

> We are eventually ICEing in decimal_to_decnumber on non-decimal
> REAL_VALUE_TYPE that creep in from uses of build_real (..., dconst*)
> for DFP types.  The following extends the decimal_to_decnumber
> special-casing of dconst* to build_real, avoiding the bogus REAL_CSTs
> from creeping into the IL and modified to ones not handled by
> the decimal_to_decnumber special casing.  It also makes sure to
> ICE for not handled dconst* values at the point we build the REAL_CST.
> 
> Bootstrapped and tested on x86_64-unknown-linux-gnu, OK for stage1
> and eventual backporting (I'm considering making the gcc_unreachable
> conditional on flag_checking on release branches).

OK.
diff mbox series

Patch

diff --git a/gcc/testsuite/gcc.dg/pr105376.c b/gcc/testsuite/gcc.dg/pr105376.c
new file mode 100644
index 00000000000..f19ecf4aab2
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr105376.c
@@ -0,0 +1,9 @@ 
+/* { dg-do compile { target dfp } } */
+/* { dg-options "-O -g" } */
+
+void
+foo (_Decimal64 d, _Decimal64 e)
+{
+  d -= -d;
+  d *= -e;
+}
diff --git a/gcc/tree.cc b/gcc/tree.cc
index 3c39be4f501..4cf3785270b 100644
--- a/gcc/tree.cc
+++ b/gcc/tree.cc
@@ -70,6 +70,7 @@  along with GCC; see the file COPYING3.  If not see
 #include "escaped_string.h"
 #include "gimple-range.h"
 #include "gomp-constants.h"
+#include "dfp.h"
 
 /* Tree code classes.  */
 
@@ -2383,6 +2384,26 @@  build_real (tree type, REAL_VALUE_TYPE d)
   REAL_VALUE_TYPE *dp;
   int overflow = 0;
 
+  /* dconst{1,2,m1,half} are used in various places in
+     the middle-end and optimizers, allow them here
+     even for decimal floating point types as an exception
+     by converting them to decimal.  */
+  if (DECIMAL_FLOAT_MODE_P (TYPE_MODE (type))
+      && d.cl == rvc_normal
+      && !d.decimal)
+    {
+      if (memcmp (&d, &dconst1, sizeof (d)) == 0)
+	decimal_real_from_string (&d, "1");
+      else if (memcmp (&d, &dconst2, sizeof (d)) == 0)
+	decimal_real_from_string (&d, "2");
+      else if (memcmp (&d, &dconstm1, sizeof (d)) == 0)
+	decimal_real_from_string (&d, "-1");
+      else if (memcmp (&d, &dconsthalf, sizeof (d)) == 0)
+	decimal_real_from_string (&d, "0.5");
+      else
+	gcc_unreachable ();
+    }
+
   /* ??? Used to check for overflow here via CHECK_FLOAT_TYPE.
      Consider doing it via real_convert now.  */