This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
PATCH COMMITTED: Fix SPEC failures caused by VRP changes
- From: Ian Lance Taylor <iant at google dot com>
- To: gcc-patches at gcc dot gnu dot org
- Date: 11 Mar 2007 08:54:59 -0700
- Subject: PATCH COMMITTED: Fix SPEC failures caused by VRP changes
When I fixed PR tree-optimization/31034, which corrected the handling
of MINUS_EXPR with overflow infinities, I managed to forget that the
exact same issues applied to PLUS_EXPR. This patch corrects that, and
also corrects the handling of the *_DIV_EXPR and RSHIFT_EXPR codes.
Bootstrapped and tested on i686-pc-linux-gnu. Committed.
Ian
2007-03-11 Ian Lance Taylor <iant@google.com>
* tree-vrp.c (vrp_int_const_binop): Handle PLUS_EXPR and
the *_DIV_EXPR codes correctly with overflow infinities.
Index: tree-vrp.c
===================================================================
--- tree-vrp.c (revision 122819)
+++ tree-vrp.c (working copy)
@@ -1519,15 +1519,26 @@ vrp_int_const_binop (enum tree_code code
&& !supports_overflow_infinity (TREE_TYPE (res)))
return NULL_TREE;
- /* We have to punt on subtracting infinities of the same sign,
- since we can't tell what the sign of the result should
- be. */
- if (code == MINUS_EXPR
- && sgn1 == sgn2
+ /* We have to punt on adding infinities of different signs,
+ since we can't tell what the sign of the result should be.
+ Likewise for subtracting infinities of the same sign. */
+ if (((code == PLUS_EXPR && sgn1 != sgn2)
+ || (code == MINUS_EXPR && sgn1 == sgn2))
&& is_overflow_infinity (val1)
&& is_overflow_infinity (val2))
return NULL_TREE;
+ /* Don't try to handle division or shifting of infinities. */
+ if ((code == TRUNC_DIV_EXPR
+ || code == FLOOR_DIV_EXPR
+ || code == CEIL_DIV_EXPR
+ || code == EXACT_DIV_EXPR
+ || code == ROUND_DIV_EXPR
+ || code == RSHIFT_EXPR)
+ && (is_overflow_infinity (val1)
+ || is_overflow_infinity (val2)))
+ return NULL_TREE;
+
/* Notice that we only need to handle the restricted set of
operations handled by extract_range_from_binary_expr.
Among them, only multiplication, addition and subtraction
@@ -1541,8 +1552,12 @@ vrp_int_const_binop (enum tree_code code
if ((code == MULT_EXPR && sgn1 == sgn2)
/* For addition, the operands must be of the same sign
to yield an overflow. Its sign is therefore that
- of one of the operands, for example the first. */
- || (code == PLUS_EXPR && sgn1 > 0)
+ of one of the operands, for example the first. For
+ infinite operands X + -INF is negative, not positive. */
+ || (code == PLUS_EXPR
+ && (sgn1 >= 0
+ ? !is_negative_overflow_infinity (val2)
+ : is_positive_overflow_infinity (val2)))
/* For subtraction, non-infinite operands must be of
different signs to yield an overflow. Its sign is
therefore that of the first operand or the opposite of