On Wed, 23 Feb 2022, Jakub Jelinek wrote:
> On Wed, Feb 23, 2022 at 09:32:41AM +0100, Jakub Jelinek via Gcc-patches wrote:
> > The following patch avoids infinite recursion during generic folding.
> > The (cmp (bswap @0) INTEGER_CST@1) simplification relies on
> > (bswap @1) actually being simplified, if it is not simplified, we just
> > move the bswap from one operand to the other and if @0 is also INTEGER_CST,
> > we apply the same rule next.
> >
> > The reason why bswap @1 isn't folded to INTEGER_CST is that the INTEGER_CST
> > has TREE_OVERFLOW set on it and fold-const-call.cc predicate punts in
> > such cases:
> > static inline bool
> > integer_cst_p (tree t)
> > {
> > return TREE_CODE (t) == INTEGER_CST && !TREE_OVERFLOW (t);
> > }
> > The patch uses ! modifier to ensure the bswap is simplified, but because
> > ! is only supported in gimple-match, guards it also with #if GIMPLE.
>
> Here is another variant, which just breaks the possible ping-pong.
> If @0 is not INTEGER_CST, we still want to canonicalize to bswap on the
> INTEGER_CST (e.g. in the hope that we throw away TREE_OVERFLOW during/after
> gimplification), but if it is INTEGER_CST, we don't want to move bswap
> to the operand with TREE_OVERFLOW on it.
>
> Ok for trunk if this passes bootstrap/regtest (it fixes the testcase too)?
If we go for a match.pd solution I'd go with the other one but as said
in the PR I think we should constant fold bswap (1(OVF)) but simply
make the (OVF) sticky as done in other constant foldings.
Richard.
> 2022-02-23 Jakub Jelinek <ja...@redhat.com>
>
> PR tree-optimization/104644
> * match.pd (cmp (bswap @0) INTEGER_CST@1): Don't simplify
> if TREE_OVERFLOW (@1) and @0 is INTEGER_CST.
>
> * gcc.dg/pr104644.c: New test.
>
> --- gcc/match.pd.jj 2022-02-23 09:17:04.867124392 +0100
> +++ gcc/match.pd 2022-02-23 10:31:05.417304115 +0100
> @@ -3961,8 +3961,9 @@ (define_operator_list SYNC_FETCH_AND_AND
> (cmp (convert:ctype @0) (convert:ctype @1))))
> (simplify
> (cmp (bswap @0) INTEGER_CST@1)
> - (with { tree ctype = TREE_TYPE (@1); }
> - (cmp (convert:ctype @0) (bswap @1)))))
> + (if (TREE_CODE (@0) != INTEGER_CST || !TREE_OVERFLOW (@1))
> + (with { tree ctype = TREE_TYPE (@1); }
> + (cmp (convert:ctype @0) (bswap @1))))))
> /* (bswap(x) >> C1) & C2 can sometimes be simplified to (x >> C3) & C2. */
> (simplify
> (bit_and (convert1? (rshift@0 (convert2? (bswap@4 @1)) INTEGER_CST@2))
> --- gcc/testsuite/gcc.dg/pr104644.c.jj 2022-02-23 10:29:50.704341688
> +0100
> +++ gcc/testsuite/gcc.dg/pr104644.c 2022-02-23 10:29:50.704341688 +0100
> @@ -0,0 +1,9 @@
> +/* PR tree-optimization/104644 */
> +/* { dg-do compile } */
> +/* { dg-options "-Wno-overflow" } */
> +
> +int
> +foo (void)
> +{
> + return __builtin_bswap16 (1.31072e+5f) != (signed char) 1.31072e+5f;
> +}
>
>
> Jakub
>
>
--
Richard Biener <rguent...@suse.de>
SUSE Software Solutions Germany GmbH, Maxfeldstrasse 5, 90409 Nuernberg,
Germany; GF: Ivo Totev; HRB 36809 (AG Nuernberg)
