Re: [PATCH] Improve EQ_EXPR/NE_EXPR folding (PR middle-end/38878)

[Richard Guenther <richard dot guenther at gmail dot com>]

On Thu, Feb 3, 2011 at 7:11 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> Hi!
>
> == and != doesn't care about signedness of the operands, so we
> can IMHO use STRIP_NOPS instead of just STRIP_SIGN_NOPS.
> Additionally, for X +- C == X and C - X == X folding we can also
> strip nops from the +/-/p+ operand for the purpose of operand_equal_p
> checking. ?This together fixes PR38878 regression (P2), though I'm
> not 100% sure we want to do that this late, perhaps we can defer
> it till stage1.
>
> Nevertheless, bootstrapped/regtested on x86_64-linux and i686-linux.

I think we want to defer this for stage1, where it is ok.

Thanks,
Richard.

> 2011-02-03 ?Jakub Jelinek ?<jakub@redhat.com>
>
> ? ? ? ?PR middle-end/38878
> ? ? ? ?* fold-const.c (fold_binary_loc) <case EQ_EXPR, NE_EXPR>: Add
> ? ? ? ?STRIP_NOPS on arg0 and arg1. ?When optimizing X +- C == X
> ? ? ? ?and C - X == X also strip nops from +/-/p+ operand.
> ? ? ? ?When optimizing -X == C, fold C to arg0's type.
>
> ? ? ? ?* gcc.dg/tree-ssa/foldaddr-1.c: Remove xfail.
>
> --- gcc/fold-const.c.jj 2010-12-02 11:51:31.000000000 +0100
> +++ gcc/fold-const.c ? ?2011-02-03 15:45:27.000000000 +0100
> @@ -12111,6 +12111,9 @@ fold_binary_loc (location_t loc,
>
> ? ? case EQ_EXPR:
> ? ? case NE_EXPR:
> + ? ? ?STRIP_NOPS (arg0);
> + ? ? ?STRIP_NOPS (arg1);
> +
> ? ? ? tem = fold_comparison (loc, code, type, op0, op1);
> ? ? ? if (tem != NULL_TREE)
> ? ? ? ?return tem;
> @@ -12193,7 +12196,8 @@ fold_binary_loc (location_t loc,
> ? ? ? /* Similarly for a NEGATE_EXPR. ?*/
> ? ? ? if (TREE_CODE (arg0) == NEGATE_EXPR
> ? ? ? ? ?&& TREE_CODE (arg1) == INTEGER_CST
> - ? ? ? ? && 0 != (tem = negate_expr (arg1))
> + ? ? ? ? && 0 != (tem = negate_expr (fold_convert_loc (loc, TREE_TYPE (arg0),
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? arg1)))
> ? ? ? ? ?&& TREE_CODE (tem) == INTEGER_CST
> ? ? ? ? ?&& !TREE_OVERFLOW (tem))
> ? ? ? ?return fold_build2_loc (loc, code, type, TREE_OPERAND (arg0, 0), tem);
> @@ -12213,7 +12217,9 @@ fold_binary_loc (location_t loc,
> ? ? ? if ((TREE_CODE (arg0) == PLUS_EXPR
> ? ? ? ? ? || TREE_CODE (arg0) == POINTER_PLUS_EXPR
> ? ? ? ? ? || TREE_CODE (arg0) == MINUS_EXPR)
> - ? ? ? ? && operand_equal_p (TREE_OPERAND (arg0, 0), arg1, 0)
> + ? ? ? ? && operand_equal_p (tree_strip_nop_conversions (TREE_OPERAND (arg0,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? 0)),
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? arg1, 0)
> ? ? ? ? ?&& (INTEGRAL_TYPE_P (TREE_TYPE (arg0))
> ? ? ? ? ? ? ?|| POINTER_TYPE_P (TREE_TYPE (arg0))))
> ? ? ? ?{
> @@ -12229,7 +12235,9 @@ fold_binary_loc (location_t loc,
> ? ? ? /* Transform comparisons of the form C - X CMP X if C % 2 == 1. ?*/
> ? ? ? if (TREE_CODE (arg0) == MINUS_EXPR
> ? ? ? ? ?&& TREE_CODE (TREE_OPERAND (arg0, 0)) == INTEGER_CST
> - ? ? ? ? && operand_equal_p (TREE_OPERAND (arg0, 1), arg1, 0)
> + ? ? ? ? && operand_equal_p (tree_strip_nop_conversions (TREE_OPERAND (arg0,
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? 1)),
> + ? ? ? ? ? ? ? ? ? ? ? ? ? ? arg1, 0)
> ? ? ? ? ?&& (TREE_INT_CST_LOW (TREE_OPERAND (arg0, 0)) & 1) == 1)
> ? ? ? ?{
> ? ? ? ? ?return omit_two_operands_loc (loc, type,
> --- gcc/testsuite/gcc.dg/tree-ssa/foldaddr-1.c.jj ? ? ? 2009-01-26 15:24:36.000000000 +0100
> +++ gcc/testsuite/gcc.dg/tree-ssa/foldaddr-1.c ?2011-02-03 15:51:24.000000000 +0100
> @@ -11,6 +11,6 @@ int foo(char *b)
> ?/* Folding should have determined that the two addresses were
> ? ?not identical and thus collapsed the function into a trivial
> ? ?"return 0". ?*/
> -/* { dg-final { scan-tree-dump-times "return 0" 1 "original" { xfail *-*-* } } } */
> +/* { dg-final { scan-tree-dump-times "return 0" 1 "original" } } */
> ?/* { dg-final { cleanup-tree-dump "original" } } */
>
>
> ? ? ? ?Jakub
>