On Tue, 22 Nov 2011, Jakub Jelinek wrote:
> Hi!
>
> VEC_INTERLEAVE_*_EXPR trees are unfortunately dependent on BYTES_BIG_ENDIAN,
> what is HIGH vs. LOW is different based on endianity.
Huh, that looks bogus. Both tree codes operate on registers and no
other codes care about "endianess" of vector registers.
(What about VEC_WIDEN_LSHIFT_{HI,LO}_EXPR?)
Can't we simply push the differece to expansion time? Or even later?
Richard.
> The only place that creates these in the IL is:
> if (BYTES_BIG_ENDIAN)
> {
> high_code = VEC_INTERLEAVE_HIGH_EXPR;
> low_code = VEC_INTERLEAVE_LOW_EXPR;
> }
> else
> {
> low_code = VEC_INTERLEAVE_HIGH_EXPR;
> high_code = VEC_INTERLEAVE_LOW_EXPR;
> }
> perm_stmt = gimple_build_assign_with_ops (high_code, perm_dest,
> vect1, vect2);
> ...
> so either folding (and expansion if only vec_perm* is supported) needs to
> be adjusted as done in the patch below, or we'd need to rename them
> to VEC_INTERLEAVE_{FIRST,SECOND}_EXPR or similar and adjust all the patterns
> etc.
>
> Bootstrapped/regtested on x86_64-linux and i686-linux, tested on the
> testcase using powerpc cross. Ok for trunk?
>
> 2011-11-22 Jakub Jelinek <ja...@redhat.com>
>
> PR tree-optimization/51074
> * fold-const.c (fold_binary_loc): Fix up VEC_INTERLEAVE_*_EXPR
> handling for BYTES_BIG_ENDIAN.
> * optabs.c (can_vec_perm_for_code_p): Likewise.
>
> * gcc.dg/vect/pr51074.c: New test.
>
> --- gcc/fold-const.c.jj 2011-11-21 16:22:02.000000000 +0100
> +++ gcc/fold-const.c 2011-11-22 09:59:15.606739333 +0100
> @@ -13483,10 +13483,12 @@ fold_binary_loc (location_t loc,
> sel[i] = i * 2 + 1;
> break;
> case VEC_INTERLEAVE_HIGH_EXPR:
> - sel[i] = (i + nelts) / 2 + ((i & 1) ? nelts : 0);
> + sel[i] = (i + (BYTES_BIG_ENDIAN ? 0 : nelts)) / 2
> + + ((i & 1) ? nelts : 0);
> break;
> case VEC_INTERLEAVE_LOW_EXPR:
> - sel[i] = i / 2 + ((i & 1) ? nelts : 0);
> + sel[i] = (i + (BYTES_BIG_ENDIAN ? nelts : 0)) / 2
> + + ((i & 1) ? nelts : 0);
> break;
> default:
> gcc_unreachable ();
> --- gcc/optabs.c.jj 2011-11-21 16:22:02.000000000 +0100
> +++ gcc/optabs.c 2011-11-22 10:17:04.820399126 +0100
> @@ -6932,9 +6932,9 @@ can_vec_perm_for_code_p (enum tree_code
> break;
>
> case VEC_INTERLEAVE_HIGH_EXPR:
> - alt = nelt / 2;
> - /* FALLTHRU */
> case VEC_INTERLEAVE_LOW_EXPR:
> + if ((BYTES_BIG_ENDIAN != 0) ^ (code == VEC_INTERLEAVE_HIGH_EXPR))
> + alt = nelt / 2;
> for (i = 0; i < nelt / 2; ++i)
> {
> data[i * 2] = i + alt;
> --- gcc/testsuite/gcc.dg/vect/pr51074.c.jj 2011-11-22 10:22:44.247377928
> +0100
> +++ gcc/testsuite/gcc.dg/vect/pr51074.c 2011-11-22 10:22:16.000000000
> +0100
> @@ -0,0 +1,24 @@
> +/* PR tree-optimization/51074 */
> +
> +#include "tree-vect.h"
> +
> +struct S { int a, b; } s[8];
> +
> +int
> +main ()
> +{
> + int i;
> + check_vect ();
> + for (i = 0; i < 8; i++)
> + {
> + s[i].b = 0;
> + s[i].a = i;
> + }
> + asm volatile ("" : : : "memory");
> + for (i = 0; i < 8; i++)
> + if (s[i].b != 0 || s[i].a != i)
> + abort ();
> + return 0;
> +}
> +
> +/* { dg-final { cleanup-tree-dump "vect" } } */
>
> Jakub
>
>
--
Richard Guenther <rguent...@suse.de>
SUSE / SUSE Labs
SUSE LINUX Products GmbH - Nuernberg - AG Nuernberg - HRB 16746
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer
