[PATCH 12/14][Vectorizer] Redefine VEC_RSHIFT_EXPR and vec_shr_optab as endianness-neutral
Mon Sep 22 13:27:00 GMT 2014
On Thu, 2014-09-18 at 09:12 -0400, David Edelsohn wrote:
> On Thu, Sep 18, 2014 at 8:42 AM, Alan Lawrence <firstname.lastname@example.org> wrote:
> > The direction of VEC_RSHIFT_EXPR has been endian-dependent, contrary to the
> > general principles of tree. This patch updates fold-const and the vectorizer
> > (the only place where such expressions are created), such that
> > VEC_RSHIFT_EXPR always shifts towards element 0.
> > The tree code still maps directly onto the vec_shr_optab, and so this patch
> > *will break any bigendian platform defining the vec_shr optab*.
> > --> For AArch64_be, patch follows next in series;
> > --> For PowerPC, I think patch/rfc 15 should fix, please inspect;
> > --> For MIPS, I think patch/rfc 16 should fix, please inspect.
> > gcc/ChangeLog:
> > * fold-const.c (const_binop): VEC_RSHIFT_EXPR always shifts towards
> > element 0.
> > * tree-vect-loop.c (vect_create_epilog_for_reduction): always
> > extract
> > the result of a reduction with vector shifts from element 0.
> > * tree.def (VEC_RSHIFT_EXPR, VEC_LSHIFT_EXPR): Comment shift
> > direction.
> > * doc/md.texi (vec_shr_m, vec_shl_m): Document shift direction.
> > Testing Done:
> > Bootstrap and check-gcc on x86_64-none-linux-gnu; check-gcc on
> > aarch64-none-elf.
> Why wasn't this tested on the PowerLinux system in the GCC Compile Farm?
> Also, Bill Schmidt can help check the PPC parts fo the patches.
Sorry for the late response; I just returned from vacation. I think
that patch 15 looks reasonable on the surface, but would be more
comfortable if it had been tested. I would echo David's suggestion that
you please test this on gcc110 in the compile farm to avoid surprises.
Given the similarity between vec_shl_<mode> and vec_shr_<mode> I am ok
with removing the former; it won't be difficult to re-create it later if
Please add some of the language you used above about VEC_RSHIFT_EXPR as
commentary for vec_shr_<mode> in vector.md, as right-shifting towards
element zero is not an obvious concept on a BE machine.
> Thanks, David
More information about the Gcc-patches