On February 13, 2018 6:51:29 PM GMT+01:00, Jakub Jelinek <ja...@redhat.com>
wrote:
>Hi!
>
>On the following testcase, we recurse infinitely, because
>we have float re-association enabled, but also rounding-math, so
>we try to optimize (cst1 + cst2) + cst3 as (cst2 + cst3) + cst1
>but (cst2 + cst3) doesn't simplify and we try again and optimize
>it as (cst3 + cst1) + cst2 and then (cst1 + cst2) + cst3 and so on
>forever. If @0 is not a CONSTANT_CLASS_P, there is not a problem,
>if it is, the code just checks if we can actually simplify the
>operation between cst2 and cst3 into a constant.
Is there a reason to try simplifying at all for constant @0? I'd rather not
try to avoid all the complex code.
Richard.
>Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?
>
>2018-02-13 Jakub Jelinek <ja...@redhat.com>
>
> PR tree-optimization/84334
> * match.pd ((A +- CST1) +- CST2 -> A + CST3): If A is
> also a CONSTANT_CLASS_P, only optimize if we can fold the
> operation between CST1 and CST2 into a constant.
>
> * gcc.dg/pr84334.c: New test.
>
>--- gcc/match.pd.jj 2018-02-13 09:33:31.000000000 +0100
>+++ gcc/match.pd 2018-02-13 12:14:08.108314686 +0100
>@@ -1733,9 +1733,20 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
> CONSTANT_CLASS_P@2)
> /* If one of the types wraps, use that one. */
> (if (!ANY_INTEGRAL_TYPE_P (type) || TYPE_OVERFLOW_WRAPS (type))
>- (if (outer_op == PLUS_EXPR)
>- (plus (view_convert @0) (inner_op @2 (view_convert @1)))
>- (minus (view_convert @0) (neg_inner_op @2 (view_convert @1))))
>+ /* If all 3 captures are CONSTANT_CLASS_P, only optimize if we
>+ can simplify @2 with @1 into a constant, otherwise we might recurse
>+ forever. */
>+ (if (CONSTANT_CLASS_P (@0))
>+ (with { tree cst = fold_unary (VIEW_CONVERT_EXPR, type, @1);
>+ if (cst && CONSTANT_CLASS_P (cst))
>+ cst = const_binop (outer_op == PLUS_EXPR
>+ ? inner_op : neg_inner_op, type,
>+ @2, cst); }
>+ (if (cst)
>+ (outer_op (view_convert @0) { cst; })))
>+ (if (outer_op == PLUS_EXPR)
>+ (plus (view_convert @0) (inner_op @2 (view_convert @1)))
>+ (minus (view_convert @0) (neg_inner_op @2 (view_convert @1)))))
> (if (!ANY_INTEGRAL_TYPE_P (TREE_TYPE (@0))
> || TYPE_OVERFLOW_WRAPS (TREE_TYPE (@0)))
> (if (outer_op == PLUS_EXPR)
>--- gcc/testsuite/gcc.dg/pr84334.c.jj 2018-02-13 12:18:12.765463667
>+0100
>+++ gcc/testsuite/gcc.dg/pr84334.c 2018-02-13 11:36:56.019632428 +0100
>@@ -0,0 +1,12 @@
>+/* PR tree-optimization/84334 */
>+/* { dg-do compile } */
>+/* { dg-options "-Ofast -frounding-math" } */
>+
>+float
>+foo (void)
>+{
>+ float a = 9.999999974752427078783512115478515625e-7f;
>+ float b = 1.999999994950485415756702423095703125e-6f;
>+ float c = 4.999999873689375817775726318359375e-6f;
>+ return a + b + c;
>+}
>
> Jakub
