nir: fix packing of TCS varyings not read by the TES

Unlike other stages TCS outputs not read by the TES cannot always
be demoted to globals e.g. when they are read by other TCS
invocations.

We were not taking these outputs into account when packing which
could result in other outputs being assigned to the same location.

Here we make sure to gather information on these outputs and group
them together when packing.

This fixes rendering issues in QUBE 2 via Proton.

Closes: #2653
Fixes: 26aa460940f6 ("nir: rewrite varying component packing")
Reviewed-by: Samuel Pitoiset <[email protected]>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4328>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4328>

1 files changed, 68 insertions, 4 deletions

diff --git a/src/compiler/nir/nir_linking_helpers.c b/src/compiler/nir/nir_linking_helpers.c
index 6e49c8737c5..7864f042636 100644
--- a/src/compiler/nir/nir_linking_helpers.c
+++ b/src/compiler/nir/nir_linking_helpers.c
@@ -443,6 +443,7 @@ struct varying_component {
    uint8_t interp_loc;
    bool is_32bit;
    bool is_patch;
+   bool is_intra_stage_only;
    bool initialised;
 };
 
@@ -456,6 +457,12 @@ cmp_varying_component(const void *comp1_v, const void *comp2_v)
    if (comp1->is_patch != comp2->is_patch)
       return comp1->is_patch ? 1 : -1;
 
+   /* We want to try to group together TCS outputs that are only read by other
+    * TCS invocations and not consumed by the follow stage.
+    */
+   if (comp1->is_intra_stage_only != comp2->is_intra_stage_only)
+      return comp1->is_intra_stage_only ? 1 : -1;
+
    /* We can only pack varyings with matching interpolation types so group
     * them together.
     */
@@ -471,7 +478,7 @@ cmp_varying_component(const void *comp1_v, const void *comp2_v)
 }
 
 static void
-gather_varying_component_info(nir_shader *consumer,
+gather_varying_component_info(nir_shader *producer, nir_shader *consumer,
                               struct varying_component **varying_comp_info,
                               unsigned *varying_comp_info_size,
                               bool default_to_smooth_interp)
@@ -482,7 +489,7 @@ gather_varying_component_info(nir_shader *consumer,
    /* Count the number of varying that can be packed and create a mapping
     * of those varyings to the array we will pass to qsort.
     */
-   nir_foreach_variable(var, &consumer->inputs) {
+   nir_foreach_variable(var, &producer->outputs) {
 
       /* Only remap things that aren't builtins. */
       if (var->data.location >= VARYING_SLOT_VAR0 &&
@@ -493,7 +500,7 @@ gather_varying_component_info(nir_shader *consumer,
             continue;
 
          const struct glsl_type *type = var->type;
-         if (nir_is_per_vertex_io(var, consumer->info.stage)) {
+         if (nir_is_per_vertex_io(var, producer->info.stage)) {
             assert(glsl_type_is_array(type));
             type = glsl_get_array_element(type);
          }
@@ -561,6 +568,63 @@ gather_varying_component_info(nir_shader *consumer,
             vc_info->interp_loc = get_interp_loc(in_var);
             vc_info->is_32bit = glsl_type_is_32bit(type);
             vc_info->is_patch = in_var->data.patch;
+            vc_info->is_intra_stage_only = false;
+         }
+      }
+   }
+
+   /* Walk over the shader and populate the varying component info array
+    * for varyings which are read by other TCS instances but are not consumed
+    * by the TES.
+    */
+   if (producer->info.stage == MESA_SHADER_TESS_CTRL) {
+      impl = nir_shader_get_entrypoint(producer);
+
+      nir_foreach_block(block, impl) {
+         nir_foreach_instr(instr, block) {
+            if (instr->type != nir_instr_type_intrinsic)
+               continue;
+
+            nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr);
+            if (intr->intrinsic != nir_intrinsic_load_deref)
+               continue;
+
+            nir_deref_instr *deref = nir_src_as_deref(intr->src[0]);
+            if (deref->mode != nir_var_shader_out)
+               continue;
+
+            /* We only remap things that aren't builtins. */
+            nir_variable *out_var = nir_deref_instr_get_variable(deref);
+            if (out_var->data.location < VARYING_SLOT_VAR0)
+               continue;
+
+            unsigned location = out_var->data.location - VARYING_SLOT_VAR0;
+            if (location >= MAX_VARYINGS_INCL_PATCH)
+               continue;
+
+            unsigned var_info_idx =
+               store_varying_info_idx[location][out_var->data.location_frac];
+            if (!var_info_idx)
+               continue;
+
+            struct varying_component *vc_info =
+               &(*varying_comp_info)[var_info_idx-1];
+
+            if (!vc_info->initialised) {
+               const struct glsl_type *type = out_var->type;
+               if (nir_is_per_vertex_io(out_var, producer->info.stage)) {
+                  assert(glsl_type_is_array(type));
+                  type = glsl_get_array_element(type);
+               }
+
+               vc_info->var = out_var;
+               vc_info->interp_type =
+                  get_interp_type(out_var, type, default_to_smooth_interp);
+               vc_info->interp_loc = get_interp_loc(out_var);
+               vc_info->is_32bit = glsl_type_is_32bit(type);
+               vc_info->is_patch = out_var->data.patch;
+               vc_info->is_intra_stage_only = true;
+            }
          }
       }
    }
@@ -648,7 +712,7 @@ compact_components(nir_shader *producer, nir_shader *consumer,
    unsigned varying_comp_info_size;
 
    /* Gather varying component info */
-   gather_varying_component_info(consumer, &varying_comp_info,
+   gather_varying_component_info(producer, consumer, &varying_comp_info,
                                  &varying_comp_info_size,
                                  default_to_smooth_interp);