On Monday, June 09, 2014 10:02:38 AM Matt Turner wrote:
> On Mon, Jun 9, 2014 at 2:59 AM, Kenneth Graunke <kenn...@whitecape.org>
wrote:
> > We need to invalidate the live intervals when inserting new
> > instructions.
> >
> > Signed-off-by: Kenneth Graunke <kenn...@whitecape.org>
> > Cc: mesa-sta...@lists.freedesktop.org
> > ---
> > src/mesa/drivers/dri/i965/brw_fs.cpp | 6 ++++++
> > 1 file changed, 6 insertions(+)
> >
> > diff --git a/src/mesa/drivers/dri/i965/brw_fs.cpp
b/src/mesa/drivers/dri/i965/brw_fs.cpp
> > index b485e65..be461ac 100644
> > --- a/src/mesa/drivers/dri/i965/brw_fs.cpp
> > +++ b/src/mesa/drivers/dri/i965/brw_fs.cpp
> > @@ -2482,6 +2482,8 @@
fs_visitor::insert_gen4_send_dependency_workarounds()
> > if (brw->gen != 4 || brw->is_g4x)
> > return;
> >
> > + bool progress = false;
> > +
> > /* Note that we're done with register allocation, so GRF fs_regs
always
> > * have a .reg_offset of 0.
> > */
> > @@ -2492,8 +2494,12 @@
fs_visitor::insert_gen4_send_dependency_workarounds()
> > if (inst->mlen != 0 && inst->dst.file == GRF) {
> > insert_gen4_pre_send_dependency_workarounds(inst);
> > insert_gen4_post_send_dependency_workarounds(inst);
> > + progress = true;
> > }
> > }
> > +
> > + if (progress)
> > + invalidate_live_intervals();
> > }
> >
> > /**
> > --
> > 1.9.2
>
> What did this fix? Post-register allocation instruction scheduling?
>
> Both are
>
> Reviewed-by: Matt Turner <matts...@gmail.com>I didn't actually observe this to fix anything. I saw some segfaults in the area, and said "oh, we didn't invalidate the live intervals. that must be it." But it wasn't - it was the list walking snafu. Still, since dump_instructions() depends on live intervals working (for the register pressure code), we ought to fix it. It may also be used for post-RA scheduling. --Ken
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
