On Fri, 28 Oct 2011, Richard Henderson wrote:
> VEC_UNPACK_HI_EXPR and VEC_UNPACK_LO_EXPR need even more -- at
> least for signed -- a source of sign-extended data. Either a LT
> comparison that produces -1 (common, but not universal) or an
> ASHIFTRT with a constant. We're definitely in the land
On 10/28/2011 02:54 PM, Hans-Peter Nilsson wrote:
> Also VEC_UNPACK_HI_EXPR, VEC_UNPACK_LO_EXPR, and
> VEC_PACK_TRUNC_EXPR to mention some.
VEC_PACK_TRUNC_EXPR requires extra mode changes. I was intending to get to
this one too, but have gotten side-tracked with other work this week.
VEC_UNPAC
On Mon, 24 Oct 2011, Richard Henderson wrote:
> From: Richard Henderson
> + /* Certain vector operations can be implemented with vector permutation.
> */
> + if (VECTOR_MODE_P (mode))
> +{
> + enum tree_code tcode = ERROR_MARK;
> + rtx sel;
> +
> + if (binoptab == vec_inte
From: Richard Henderson
---
gcc/expr.c| 20 +---
gcc/optabs.c | 116 +
gcc/optabs.h |3 +
gcc/tree-vect-data-refs.c | 80 ---
gcc/tree-vect-generic.c |9
5 fi
