[Mesa-dev] [PATCH 2/3] radeonsi/nir: Set vs_inputs_dual_locations and let NIR do the remap

[Timothy Arceri]

On 01/09/18 13:11, Jason Ekstrand wrote:
> We were going out of our way to disable dual-location re-mapping in NIR
> only to then do the remapping in st_glsl_to_nir.cpp.  Presumably, this
> was so that double_inputs would be correct for the core state tracker.
> However, now that we've it to gl_program::DualSlotInputs which is
> unaffected by NIR lowering, we can let NIR lower things for us.  The one
> tricky bit here is that we have to remap the inputs_read bitfield back
> to the single-slot convention for the gallium state tracker to use.
> 
> Since radeonsi is the only NIR-capable gallium driver that also supports
> GL_ARB_vertex_attrib_64bit, we only have to worry about radeonsi when
> making core gallium state tracker changes.
> ---
>   .../nir/nir_lower_io_arrays_to_elements.c     |  5 +--
>   src/gallium/drivers/radeonsi/si_get.c         |  1 +
>   src/mesa/state_tracker/st_glsl_to_nir.cpp     | 45 +++++++++----------
>   3 files changed, 22 insertions(+), 29 deletions(-)
> 
> diff --git a/src/compiler/nir/nir_lower_io_arrays_to_elements.c b/src/compiler/nir/nir_lower_io_arrays_to_elements.c
> index 16f6233f614..af33d153ea5 100644
> --- a/src/compiler/nir/nir_lower_io_arrays_to_elements.c
> +++ b/src/compiler/nir/nir_lower_io_arrays_to_elements.c
> @@ -36,9 +36,6 @@ static unsigned
>   get_io_offset(nir_builder *b, nir_deref_instr *deref, nir_variable *var,
>                 unsigned *element_index, nir_ssa_def **vertex_index)
>   {
> -   bool vs_in = (b->shader->info.stage == MESA_SHADER_VERTEX) &&
> -                (var->data.mode == nir_var_shader_in);
> -
>      nir_deref_path path;
>      nir_deref_path_init(&path, deref, NULL);
>   
> @@ -60,7 +57,7 @@ get_io_offset(nir_builder *b, nir_deref_instr *deref, nir_variable *var,
>   
>            assert(c);     /* must not be indirect dereference */
>   
> -         unsigned size = glsl_count_attribute_slots((*p)->type, vs_in);
> +         unsigned size = glsl_count_attribute_slots((*p)->type, false);
>            offset += size * c->u32[0];
>   
>            unsigned num_elements = glsl_type_is_array((*p)->type) ?
> diff --git a/src/gallium/drivers/radeonsi/si_get.c b/src/gallium/drivers/radeonsi/si_get.c
> index 90f62edf470..bc3d559861f 100644
> --- a/src/gallium/drivers/radeonsi/si_get.c
> +++ b/src/gallium/drivers/radeonsi/si_get.c
> @@ -497,6 +497,7 @@ static const struct nir_shader_compiler_options nir_options = {
>   	.lower_extract_word = true,
>   	.max_unroll_iterations = 32,
>   	.native_integers = true,
> +	.vs_inputs_dual_locations = true,
>   };
>   
>   static const void *
> diff --git a/src/mesa/state_tracker/st_glsl_to_nir.cpp b/src/mesa/state_tracker/st_glsl_to_nir.cpp
> index 0ee9bd9fef1..d0ec410ec69 100644
> --- a/src/mesa/state_tracker/st_glsl_to_nir.cpp
> +++ b/src/mesa/state_tracker/st_glsl_to_nir.cpp
> @@ -32,6 +32,7 @@
>   #include "program/prog_parameter.h"
>   #include "program/ir_to_mesa.h"
>   #include "main/mtypes.h"
> +#include "main/imports.h"
>   #include "main/errors.h"
>   #include "main/shaderapi.h"
>   #include "main/uniforms.h"
> @@ -83,33 +84,18 @@ st_nir_fixup_varying_slots(struct st_context *st, struct exec_list *var_list)
>   static void
>   st_nir_assign_vs_in_locations(struct gl_program *prog, nir_shader *nir)
>   {
> -   unsigned attr, num_inputs = 0;
> -   unsigned input_to_index[VERT_ATTRIB_MAX] = {0};
> -
> -   /* TODO de-duplicate w/ similar code in st_translate_vertex_program()? */
> -   for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
> -      if ((prog->info.inputs_read & BITFIELD64_BIT(attr)) != 0) {
> -         input_to_index[attr] = num_inputs;
> -         num_inputs++;
> -         if ((prog->DualSlotInputs & BITFIELD64_BIT(attr)) != 0) {
> -            /* add placeholder for second part of a double attribute */
> -            num_inputs++;
> -         }
> -      } else {
> -         input_to_index[attr] = ~0;
> -      }
> -   }
> -
> -   /* bit of a hack, mirroring st_translate_vertex_program */
> -   input_to_index[VERT_ATTRIB_EDGEFLAG] = num_inputs;
> -
>      nir->num_inputs = 0;
>      nir_foreach_variable_safe(var, &nir->inputs) {
> -      attr = var->data.location;
> -      assert(attr < ARRAY_SIZE(input_to_index));
> -
> -      if (input_to_index[attr] != ~0u) {
> -         var->data.driver_location = input_to_index[attr];
> +      /* NIR already assigns dual-slot inputs to two locations so all we have
> +       * to do is compact everything down.
> +       */
> +      if (var->data.location == VERT_ATTRIB_EDGEFLAG) {
> +         /* bit of a hack, mirroring st_translate_vertex_program */
> +         var->data.driver_location = _mesa_bitcount_64(nir->info.inputs_read);
> +      } else if (nir->info.inputs_read & BITFIELD64_BIT(var->data.location)) {
> +         var->data.driver_location =
> +            _mesa_bitcount_64(nir->info.inputs_read &
> +                              BITFIELD64_MASK(var->data.location));

I'm probably missing something obvious but wont this always set 
driver_location to 1?


>            nir->num_inputs++;
>         } else {
>            /* Move unused input variables to the globals list (with no
> @@ -743,6 +729,15 @@ st_link_nir(struct gl_context *ctx,
>   
>         nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));
>         shader->Program->info = nir->info;
> +      if (i == MESA_SHADER_VERTEX) {
> +         /* NIR expands dual-slot inputs out to two locations.  We need to
> +          * compact things back down GL-style single-slot inputs to avoid
> +          * confusing the state tracker.
> +          */
> +         shader->Program->info.inputs_read =
> +            nir_get_single_slot_attribs_mask(nir->info.inputs_read,
> +                                             shader->Program->DualSlotInputs);
> +      }
>   
>         if (prev != -1) {
>            struct gl_program *prev_shader =
>