[Mesa-stable] [PATCH v3 13/48] intel/fs: Use the original destination region for int MUL lowering

[Jason Ekstrand]

On Wed, Nov 8, 2017 at 9:46 AM, Mark Janes <mark.a.janes at intel.com> wrote:

> This patch is CC'd to stable, but CTS regressions bisected to it:
>
> https://bugs.freedesktop.org/show_bug.cgi?id=103626
>
> Please do not merge to stable until the bug is addressed.
>

Noted.  I can take a look at that bug in a bit.  It should be easy to fix.


> Jason Ekstrand <jason at jlekstrand.net> writes:
>
> > Some hardware (CHV, BXT) have special restrictions on register regions
> > when doing integer multiplication.  We want to respect those when we
> > lower to DxW multiplication.
> >
> > Cc: mesa-stable at lists.freedesktop.org
> > ---
> >  src/intel/compiler/brw_fs.cpp | 16 +++++++++-------
> >  1 file changed, 9 insertions(+), 7 deletions(-)
> >
> > diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.
> cpp
> > index 56455e9..1c4351b 100644
> > --- a/src/intel/compiler/brw_fs.cpp
> > +++ b/src/intel/compiler/brw_fs.cpp
> > @@ -3482,18 +3482,20 @@ fs_visitor::lower_integer_multiplication()
> >
> >              bool needs_mov = false;
> >              fs_reg orig_dst = inst->dst;
> > +            fs_reg low = inst->dst;
> >              if (orig_dst.is_null() || orig_dst.file == MRF ||
> >                  regions_overlap(inst->dst, inst->size_written,
> >                                  inst->src[0], inst->size_read(0)) ||
> >                  regions_overlap(inst->dst, inst->size_written,
> >                                  inst->src[1], inst->size_read(1))) {
> >                 needs_mov = true;
> > -               inst->dst = fs_reg(VGRF, alloc.allocate(dispatch_width /
> 8),
> > -                                  inst->dst.type);
> > +               low.nr = alloc.allocate(regs_written(inst));
> > +               low.offset = low.offset % REG_SIZE;
> >              }
> > -            fs_reg low = inst->dst;
> > -            fs_reg high(VGRF, alloc.allocate(dispatch_width / 8),
> > -                        inst->dst.type);
> > +
> > +            fs_reg high = inst->dst;
> > +            high.nr = alloc.allocate(regs_written(inst));
> > +            high.offset = high.offset % REG_SIZE;
> >
> >              if (devinfo->gen >= 7) {
> >                 if (inst->src[1].file == IMM) {
> > @@ -3514,13 +3516,13 @@ fs_visitor::lower_integer_multiplication()
> >                          inst->src[1]);
> >              }
> >
> > -            ibld.ADD(subscript(inst->dst, BRW_REGISTER_TYPE_UW, 1),
> > +            ibld.ADD(subscript(low, BRW_REGISTER_TYPE_UW, 1),
> >                       subscript(low, BRW_REGISTER_TYPE_UW, 1),
> >                       subscript(high, BRW_REGISTER_TYPE_UW, 0));
> >
> >              if (needs_mov || inst->conditional_mod) {
> >                 set_condmod(inst->conditional_mod,
> > -                           ibld.MOV(orig_dst, inst->dst));
> > +                           ibld.MOV(orig_dst, low));
> >              }
> >           }
> >
> > --
> > 2.5.0.400.gff86faf
> >
> > _______________________________________________
> > mesa-stable mailing list
> > mesa-stable at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/mesa-stable
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-stable/attachments/20171108/dbbe4b86/attachment.html>