[Mesa-dev] [PATCH 2/3] intel/vec4: Fix nir_op_b2[fi] with 64-bit result

[Jason Ekstrand]

On Thu, Oct 11, 2018 at 1:23 PM Ian Romanick <idr at freedesktop.org> wrote:

> On 10/11/2018 11:21 AM, Ian Romanick wrote:
> > On 10/11/2018 08:38 AM, Jason Ekstrand wrote:
> >> ---
> >>  src/intel/compiler/brw_vec4_nir.cpp | 7 ++++++-
> >>  1 file changed, 6 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/src/intel/compiler/brw_vec4_nir.cpp
> b/src/intel/compiler/brw_vec4_nir.cpp
> >> index eaf1754b006..5ccfd1f8940 100644
> >> --- a/src/intel/compiler/brw_vec4_nir.cpp
> >> +++ b/src/intel/compiler/brw_vec4_nir.cpp
> >> @@ -1586,7 +1586,12 @@ vec4_visitor::nir_emit_alu(nir_alu_instr *instr)
> >>
> >>     case nir_op_b2i:
> >>     case nir_op_b2f:
> >> -      emit(MOV(dst, negate(op[0])));
> >> +      if (nir_dest_bit_size(instr->dest.dest) > 32) {
> >> +         assert(dst.type == BRW_REGISTER_TYPE_DF);
> >> +         emit_conversion_to_double(dst, negate(op[0]), false);
> >> +      } else {
> >> +         emit(MOV(dst, negate(op[0])));
> >> +      }
> >
> > I wrote almost the same patch, but I split nir_op_b2i and nir_op_b2f
> > handling and omitted the assert.  I was never able to get the
> > then-clause to trigger.  Were you able to get that to happen?
>
> I guess the next patch causes it?  Assuming that's the case, you should
> add that to the commit message for this patch.
>

Correct.  Sure, I can add that to the commit message.  I was going to CC it
to stable because I don't really know if it's possible to trigger or not.
It definitely does trigger with the previous patch but I can't prove that
it doesn't trigger without it.  Thoughts?


> >>        break;
> >>
> >>     case nir_op_f2b:
> >
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20181011/4fee27eb/attachment.html>