[Mesa-dev] [PATCH 06/21] etnaviv: nir: hardwire position location

[Rob Clark]

On Tue, Jun 5, 2018 at 10:38 AM, Philipp Zabel <p.zabel at pengutronix.de> wrote:
> The temporary input/output register 0 is reserved for position in the
> fragment shader. Hardwire it to 0 and start other input/output variables
> at 1. The intrinsic input load and output store base corresponds to the
> temporary register number.

I didn't look at this very closely, but at RA time you can pre-color
(pre-assign) specific registers, see ra_set_node_reg().. not sure if
this would be a better approach.  I do this for a few frag shader
registers, mostly because at the time I hadn't found the cmdstream
register/bitfield to configure it to something other than r0.x..

BR,
-R

>
> Signed-off-by: Philipp Zabel <p.zabel at pengutronix.de>
> Signed-off-by: Michael Tretter <m.tretter at pengutronix.de>
> ---
>  src/gallium/drivers/etnaviv/etnaviv_nir.c | 76 +++++++++++++++++++++++
>  1 file changed, 76 insertions(+)
>
> diff --git a/src/gallium/drivers/etnaviv/etnaviv_nir.c b/src/gallium/drivers/etnaviv/etnaviv_nir.c
> index 1ddfbb818922..3b29ea9a0e76 100644
> --- a/src/gallium/drivers/etnaviv/etnaviv_nir.c
> +++ b/src/gallium/drivers/etnaviv/etnaviv_nir.c
> @@ -81,6 +81,81 @@ etna_optimize_loop(nir_shader *s)
>     } while (progress);
>  }
>
> +/* The temporary PS input/output register 0 is reserved for the position */
> +static void
> +etna_hardwire_io_position(nir_shader *shader)
> +{
> +   bool fixup_needed = false;
> +   int pos_in = -1, pos_out = -1;
> +   int max_in = 0, max_out = 0;
> +
> +   if (shader->info.stage != MESA_SHADER_FRAGMENT)
> +      return;
> +
> +   nir_foreach_variable(var, &shader->inputs) {
> +      if (var->data.location == VARYING_SLOT_POS) {
> +         pos_in = var->data.driver_location;
> +      } else {
> +         max_in = MAX2(max_in, var->data.driver_location);
> +         if (var->data.driver_location == 0)
> +            fixup_needed = true;
> +      }
> +   }
> +   nir_foreach_variable(var, &shader->outputs) {
> +      if (var->data.location == FRAG_RESULT_DEPTH) {
> +         pos_out = var->data.driver_location;
> +      } else {
> +         max_out = MAX2(max_out, var->data.driver_location);
> +         if (var->data.driver_location == 0)
> +            fixup_needed = true;
> +      }
> +   }
> +
> +   fixup_needed |= (pos_in > 0) || (pos_out > 0);
> +   if (!fixup_needed)
> +      return;
> +
> +   if (pos_in == -1)
> +      pos_in = max_in + 1;
> +   if (pos_out == -1)
> +      pos_out = max_out + 1;
> +
> +   nir_foreach_variable(var, &shader->inputs) {
> +      if (var->data.location == VARYING_SLOT_POS)
> +         var->data.driver_location = 0;
> +      else if (var->data.driver_location < pos_in)
> +         var->data.driver_location++;
> +   }
> +   nir_foreach_variable(var, &shader->outputs) {
> +      if (var->data.location == FRAG_RESULT_DEPTH)
> +         var->data.driver_location = 0;
> +      else if (var->data.driver_location < pos_out)
> +         var->data.driver_location++;
> +   }
> +
> +   nir_foreach_function(function, shader) {
> +      nir_foreach_block(block, function->impl) {
> +         nir_foreach_instr(instr, block) {
> +            if (instr->type != nir_instr_type_intrinsic)
> +               continue;
> +            nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
> +            unsigned pos;
> +            if (intr->intrinsic == nir_intrinsic_load_input)
> +               pos = pos_in;
> +            else if (intr->intrinsic == nir_intrinsic_store_output)
> +               pos = pos_out;
> +            else
> +               continue;
> +            unsigned base = nir_intrinsic_base(intr);
> +            if (base <= pos)
> +               nir_intrinsic_set_base(intr, (base == pos) ? 0 : (base + 1));
> +         }
> +      }
> +      nir_metadata_preserve(function->impl,
> +                            nir_metadata_block_index | nir_metadata_dominance);
> +   }
> +}
> +
>  /* Move const loads, input load intrinsics, and uniform load intrinsics to the
>   * beginning of the function implementation.
>   *
> @@ -545,6 +620,7 @@ etna_optimize_nir(struct etna_shader *shader,
>     NIR_PASS_V(s, nir_opt_global_to_local);
>     NIR_PASS_V(s, nir_lower_regs_to_ssa);
>     NIR_PASS_V(s, etna_move_load_intrinsics);
> +   NIR_PASS_V(s, etna_hardwire_io_position);
>
>     etna_optimize_loop(s);
>
> --
> 2.17.1
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/mesa-dev