This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH, rs6000] (v2) Fold vector shifts in GIMPLE
- From: Will Schmidt <will_schmidt at vnet dot ibm dot com>
- To: Richard Biener <richard dot guenther at gmail dot com>
- Cc: GCC Patches <gcc-patches at gcc dot gnu dot org>, Segher Boessenkool <segher at kernel dot crashing dot org>, Bill Schmidt <wschmidt at linux dot vnet dot ibm dot com>, David Edelsohn <dje dot gcc at gmail dot com>, Jakub Jelinek <jakub at redhat dot com>
- Date: Wed, 14 Jun 2017 09:55:21 -0500
- Subject: Re: [PATCH, rs6000] (v2) Fold vector shifts in GIMPLE
- Authentication-results: sourceware.org; auth=none
- References: <1497304579.24125.144.camel@brimstone.rchland.ibm.com> <CAFiYyc2Wg1puv60d34yYbrVKVALgxMKSr+bhK_8XGJQgW4RxuA@mail.gmail.com>
- Reply-to: will_schmidt at vnet dot ibm dot com
On Tue, 2017-06-13 at 10:03 +0200, Richard Biener wrote:
> On Mon, Jun 12, 2017 at 11:56 PM, Will Schmidt
> <will_schmidt@vnet.ibm.com> wrote:
> > Hi,
> >
> >
> > diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
> > index 63ca2d1..55592fb 100644
> > --- a/gcc/config/rs6000/rs6000.c
> > +++ b/gcc/config/rs6000/rs6000.c
> > @@ -16588,6 +16588,83 @@ rs6000_gimple_fold_builtin (gimple_stmt_iterator *gsi)
> > gsi_replace (gsi, g, true);
> > return true;
> > }
<snip>
> > + /* Flavors of vector shift right. */
> > + case ALTIVEC_BUILTIN_VSRB:
> > + case ALTIVEC_BUILTIN_VSRH:
> > + case ALTIVEC_BUILTIN_VSRW:
> > + case P8V_BUILTIN_VSRD:
> > + {
> > + arg0 = gimple_call_arg (stmt, 0);
> > + arg1 = gimple_call_arg (stmt, 1);
> > + lhs = gimple_call_lhs (stmt);
> > + gimple *g;
> > + /* convert arg0 to unsigned. */
> > + arg0 = convert (unsigned_type_for (TREE_TYPE (arg0)), arg0);
>
> Please do not use 'convert', instead do ...
Hi Richard,
V3 of this patch , using the gimple_build() convenience helper function
has been posted, and is the direction I'm going for with this patch. I
wanted to make sure I fully understood the other options though, so I
have a question/clarification on the other suggestions:
> > + tree arg0_uns = create_tmp_reg_or_ssa_name
> > + (unsigned_type_for (TREE_TYPE (arg0)));
> > + g = gimple_build_assign (arg0_uns, arg0);
>
> g = gimple_build_assign (arg0_uns, VIEW_CONVERT_EXPR, usigned_type, arg0);
I tried a few trivial variations of this:
g = gimple_build_assign (arg0_uns, VIEW_CONVERT_EXPR,
unsigned_type_for (TREE_TYPE(arg0_uns)), arg0);
which lookd good, but it asserts in gimple_build_assign_1(), on the
check
"if (op2)
{
gcc_assert (num_ops > 2);
...
Trolling around the other code for references, i found and tried this,
which uses the build1() helper, and appears to work. Is this the gist
of what you suggested, or would there be another alternative?
g = gimple_build_assign (arg0_uns,
build1(VIEW_CONVERT_EXPR,
unsigned_type_for (TREE_TYPE(arg0_uns)), arg0));
Thanks for the feedback, etc. :-)
-Will
> You also want to avoid spitting out useless copies here if the
> arg/result is already unsigned,
> like via
>
> tree arg0_uns = arg0;
> if (! TYPE_UNSIGNED (TREE_TYPE (arg0_uns)))
> {
> ...
> }
>
> > + gimple_set_location (g, gimple_location (stmt));
> > + gsi_insert_before (gsi, g, GSI_SAME_STMT);
> > + /* convert lhs to unsigned and do the shift. */
>
> Just use lhs if it has the same sign as arg0_uns.
>
> > + tree lhs_uns = create_tmp_reg_or_ssa_name
> > + (unsigned_type_for (TREE_TYPE (lhs)));
>
> You can re-use the type of arg0_uns here.
>
> > + g = gimple_build_assign (lhs_uns, RSHIFT_EXPR, arg0_uns, arg1);
> > + gimple_set_location (g, gimple_location (stmt));
> > + gsi_insert_before (gsi, g, GSI_SAME_STMT);
> > + /* convert lhs back to a signed type for the return. */
> > + lhs_uns = convert (signed_type_for (TREE_TYPE (lhs)),lhs_uns);
> > + g = gimple_build_assign (lhs, lhs_uns);
>
> See above for how to perform the conversion.
>
> Note that you could use the gimple_build convenience to shorten the code
> sequence above to
>
> gimple_seq stmts = NULL;
> tree arg0_unsigned = gimple_build (&stmts, VIEW_CONVERT_EXPR,
>
> unsigned_type_for (...), arg0);
> tree res = gimple_build (&stmts, RSHIFT_EXPR, TREE_TYPE (arg0_uns),
> arg0_uns, arg1);
> res = gimple_build (&stmts, VIEW_CONVERT_EXPR, TREE_TYPE (lhs), res);
> gsi_insert_seq_before (gsi, stmts, GSI_SAME_STMT);
> update_call_from_tree (gsi, res);
>
> The above gimple_build sequence will fold all the stmts thus remove
> useless conversions and apply constant folding, etc.
>
> Richard.
>
> > + gimple_set_location (g, gimple_location (stmt));
> > + gsi_replace (gsi, g, true);
> > + return true;
> > + }
> > default:
> > break;
> > }