[Mesa-stable] [Mesa-dev] [PATCH 2/2] i965/tes/scalar: fix 64-bit indirect input loads

Iago Toral itoral at igalia.com
Fri Jul 15 11:41:39 UTC 2016


On Fri, 2016-07-15 at 21:20 +1000, Timothy Arceri wrote:
> On Fri, 2016-07-15 at 13:16 +0200, Iago Toral wrote:
> > 
> > On Fri, 2016-07-15 at 20:59 +1000, Timothy Arceri wrote:
> > > 
> > > On Fri, 2016-07-15 at 11:04 +0200, Iago Toral Quiroga wrote:
> > > > 
> > > > 
> > > > We totally ignored this before because there were no piglit
> > > > tests
> > > > for
> > > > indirect loads in tessellation stages with doubles.
> > > > 
> > > > Cc: "12.0" <mesa-stable at lists.freedesktop.org>
> > > > ---
> > > >  src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 86
> > > > ++++++++++++++++++++++++--------
> > > >  1 file changed, 64 insertions(+), 22 deletions(-)
> > > > 
> > > > diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > > > b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > > > index 55383ff..7d5af78 100644
> > > > --- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > > > +++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > > > @@ -2926,31 +2926,73 @@
> > > > fs_visitor::nir_emit_tes_intrinsic(const
> > > > fs_builder &bld,
> > > >           }
> > > >        } else {
> > > >           /* Indirect indexing - use per-slot offsets as well.
> > > > */
> > > > -         const fs_reg srcs[] = {
> > > > -            retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UD),
> > > > -            indirect_offset
> > > > -         };
> > > > -         fs_reg payload = bld.vgrf(BRW_REGISTER_TYPE_UD, 2);
> > > > -         bld.LOAD_PAYLOAD(payload, srcs, ARRAY_SIZE(srcs), 0);
> > > >  
> > > > -         if (first_component != 0) {
> > > > -            unsigned read_components =
> > > > -                instr->num_components + first_component;
> > > > -            fs_reg tmp = bld.vgrf(dest.type, read_components);
> > > > -            inst =
> > > > bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT,
> > > > tmp,
> > > > -                            payload);
> > > > -            inst->regs_written = read_components;
> > > > -            for (unsigned i = 0; i < instr->num_components;
> > > > i++)
> > > > {
> > > > -               bld.MOV(offset(dest, bld, i),
> > > > -                       offset(tmp, bld, i + first_component));
> > > > +         /* We can only read two double components with each
> > > > URB
> > > > read, so
> > > > +          * we send two read messages in that case, each one
> > > > loading
> > > > up to
> > > > +          * two double components.
> > > > +          */
> > > > +         unsigned num_iterations = 1;
> > > > +         unsigned num_components = instr->num_components;
> > > > +         fs_reg orig_dest = dest;
> > > > +         if (type_sz(dest.type) == 8) {
> > > Hi,
> > > 
> > > Not your fault as its not currently enabled but can you add the
> > > following here to allow component packing to work once it finally
> > > gets
> > > enabled:
> > > 
> > >                first_component = first_component / 2;
> > > 
> > > I believe that should be enought to keep it working once the
> > > other
> > > patches land.
> > I think this is already being done in master, right after the
> > assignment to first_component.
> Oh yes you are right. All looks good to me then :)

great!

> > 
> > 
> > > 
> > > Also I don't think this patch is going to apply cleanly to stable
> > > as
> > > the component packing cahnges only landed after branching.

Aha, good point. In that case I think it is probably better to just
drop Cc to stable. Thanks!

Iago

> > > Thanks,
> > > Tim
> > > 
> > > > 
> > > > 
> > > > +            if (instr->num_components > 2) {
> > > > +               num_iterations = 2;
> > > > +               num_components = 2;
> > > > +            }
> > > > +            fs_reg tmp = fs_reg(VGRF, alloc.allocate(4),
> > > > dest.type);
> > > > +            dest = tmp;
> > > > +         }
> > > > +
> > > > +         for (unsigned iter = 0; iter < num_iterations;
> > > > iter++)
> > > > {
> > > > +            const fs_reg srcs[] = {
> > > > +               retype(brw_vec1_grf(0, 0),
> > > > BRW_REGISTER_TYPE_UD),
> > > > +               indirect_offset
> > > > +            };
> > > > +            fs_reg payload = bld.vgrf(BRW_REGISTER_TYPE_UD,
> > > > 2);
> > > > +            bld.LOAD_PAYLOAD(payload, srcs, ARRAY_SIZE(srcs),
> > > > 0);
> > > > +
> > > > +            if (first_component != 0) {
> > > > +               unsigned read_components =
> > > > +                   num_components + first_component;
> > > > +               fs_reg tmp = bld.vgrf(dest.type,
> > > > read_components);
> > > > +               inst =
> > > > bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, tmp,
> > > > +                               payload);
> > > > +               for (unsigned i = 0; i < num_components; i++) {
> > > > +                  bld.MOV(offset(dest, bld, i),
> > > > +                          offset(tmp, bld, i +
> > > > first_component));
> > > > +               }
> > > > +            } else {
> > > > +               inst =
> > > > bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT, dest,
> > > > +                               payload);
> > > > +            }
> > > > +            inst->mlen = 2;
> > > > +            inst->offset = imm_offset;
> > > > +            inst->regs_written =
> > > > +               ((num_components + first_component) *
> > > > type_sz(dest.type) / 4);
> > > > +
> > > > +            /* If we are reading 64-bit data using 32-bit read
> > > > messages we need
> > > > +             * build proper 64-bit data elements by shuffling
> > > > the
> > > > low and high
> > > > +             * 32-bit components around like we do for other
> > > > things
> > > > like UBOs
> > > > +             * or SSBOs.
> > > > +             */
> > > > +            if (type_sz(dest.type) == 8) {
> > > > +               shuffle_32bit_load_result_to_64bit_data(
> > > > +                  bld, dest, retype(dest,
> > > > BRW_REGISTER_TYPE_F),
> > > > num_components);
> > > > +
> > > > +               for (unsigned c = 0; c < num_components; c++) {
> > > > +                  bld.MOV(offset(orig_dest, bld, iter * 2 +
> > > > c),
> > > > +                          offset(dest, bld, c));
> > > > +               }
> > > > +            }
> > > > +
> > > > +            /* If we are loading double data and we need a
> > > > second
> > > > read message
> > > > +             * adjust the offset
> > > > +             */
> > > > +            if (num_iterations > 1) {
> > > > +               num_components = instr->num_components - 2;
> > > > +               imm_offset++;
> > > >              }
> > > > -         } else {
> > > > -            inst =
> > > > bld.emit(SHADER_OPCODE_URB_READ_SIMD8_PER_SLOT,
> > > > dest,
> > > > -                            payload);
> > > > -            inst->regs_written = instr->num_components;
> > > >           }
> > > > -         inst->mlen = 2;
> > > > -         inst->offset = imm_offset;
> > > >        }
> > > >        break;
> > > >     }
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/mesa-dev


More information about the mesa-stable mailing list