On 2015-10-29 02:17:20, Iago Toral wrote:
> On Thu, 2015-10-29 at 00:50 -0700, Jordan Justen wrote:
> > Signed-off-by: Jordan Justen <jordan.l.jus...@intel.com>
> > ---
> > src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 6 ++++--
> > src/mesa/drivers/dri/i965/brw_vec4_nir.cpp | 6 ++++--
> > 2 files changed, 8 insertions(+), 4 deletions(-)
> >
> > diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > index 4950ba4..6d69e96 100644
> > --- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > +++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp
> > @@ -1438,8 +1438,10 @@ fs_visitor::nir_emit_intrinsic(const fs_builder
> > &bld, nir_intrinsic_instr *instr
> > fs_reg surf_index;
> >
> > if (const_index) {
> > - surf_index = fs_reg(stage_prog_data->binding_table.ubo_start +
> > - const_index->u[0]);
> > + unsigned index = stage_prog_data->binding_table.ubo_start +
> > + const_index->u[0];
> > + surf_index = fs_reg(index);
> > + brw_mark_surface_used(prog_data, index);
> > } else {
> > /* The block index is not a constant. Evaluate the index
> > expression
> > * per-channel and add the base UBO index; we have to select a
> > value
> > diff --git a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
> > b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
> > index 0f04f65..efbdaa9 100644
> > --- a/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
> > +++ b/src/mesa/drivers/dri/i965/brw_vec4_nir.cpp
> > @@ -749,8 +749,10 @@ vec4_visitor::nir_emit_intrinsic(nir_intrinsic_instr
> > *instr)
> > /* The block index is a constant, so just emit the binding table
> > entry
> > * as an immediate.
> > */
> > - surf_index = src_reg(prog_data->base.binding_table.ubo_start +
> > - const_block_index->u[0]);
> > + unsigned index = prog_data->base.binding_table.ubo_start +
> > + const_block_index->u[0];
> > + surf_index = src_reg(index);
> > + brw_mark_surface_used(&prog_data->base, index);
> > } else {
> > /* The block index is not a constant. Evaluate the index
> > expression
> > * per-channel and add the base UBO index; we have to select a
> > value
>
> Is this necessary? As far as I can see, the generator opcodes will mark
> surfaces for which we have a constant index as used, so this should not
> be necessary.
Ah. You are correct.
I was debugging a UBO+SSBO test, and I noticed the discrepancy between
SSBO and UBO, but it appears SSBO has to mark it in brw_*_nir.cpp,
whereas the generator code can also mark it for the constant case for
UBOs.
Maybe we should just handle this in brw_*_nir.cpp for consistency?
-Jordan
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
