On Wed, Sep 18, 2019 at 8:54 AM Richard Sandiford
<richard.sandif...@arm.com> wrote:
>
> CONSTANT lattice values are symbolic constants rather than
> compile-time constants, so among other things can be POLY_INT_CSTs.
> This patch fixes a case in which we assumed all CONSTANTs were either
> ADDR_EXPRs or INTEGER_CSTs.
>
> This is tested by later SVE patches.
>
> Tested on aarch64-linux-gnu with SVE (with and without follow-on
> patches) and x86_64-linux-gnu. OK to install?
OK.
Richard.
> Richard
>
>
> 2019-09-18 Richard Sandiford <richard.sandif...@arm.com>
>
> gcc/
> * tree-ssa-ccp.c (get_value_for_expr): Check whether CONSTANTs
> are INTEGER_CSTs.
>
> Index: gcc/tree-ssa-ccp.c
> ===================================================================
> --- gcc/tree-ssa-ccp.c 2019-08-21 14:58:05.999057076 +0100
> +++ gcc/tree-ssa-ccp.c 2019-09-18 07:53:36.930481545 +0100
> @@ -615,9 +615,17 @@ get_value_for_expr (tree expr, bool for_
> val.mask = -1;
> }
> if (for_bits_p
> - && val.lattice_val == CONSTANT
> - && TREE_CODE (val.value) == ADDR_EXPR)
> - val = get_value_from_alignment (val.value);
> + && val.lattice_val == CONSTANT)
> + {
> + if (TREE_CODE (val.value) == ADDR_EXPR)
> + val = get_value_from_alignment (val.value);
> + else if (TREE_CODE (val.value) != INTEGER_CST)
> + {
> + val.lattice_val = VARYING;
> + val.value = NULL_TREE;
> + val.mask = -1;
> + }
> + }
> /* Fall back to a copy value. */
> if (!for_bits_p
> && val.lattice_val == VARYING
