i965/fs: Lower gl_VertexID and friends to inputs at the NIR level

NIR calls these system values but they come in from the VF unit as
vertex data.  It's terribly convenient to just be able to treat them as
such in the back-end.

Reviewed-by: Alejandro Piñeiro <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>

4 files changed, 74 insertions, 70 deletions

diff --git a/src/intel/compiler/brw_fs.h b/src/intel/compiler/brw_fs.h
index e230b5e0dd9..6c8c0275a76 100644
--- a/src/intel/compiler/brw_fs.h
+++ b/src/intel/compiler/brw_fs.h
@@ -175,7 +175,6 @@ public:
    fs_reg *emit_samplepos_setup();
    fs_reg *emit_sampleid_setup();
    fs_reg *emit_samplemaskin_setup();
-   fs_reg *emit_vs_system_value(int location);
    void emit_interpolation_setup_gen4();
    void emit_interpolation_setup_gen6();
    void compute_sample_position(fs_reg dst, fs_reg int_sample_pos);
diff --git a/src/intel/compiler/brw_fs_nir.cpp b/src/intel/compiler/brw_fs_nir.cpp
index d90eaad32ba..779d74d915d 100644
--- a/src/intel/compiler/brw_fs_nir.cpp
+++ b/src/intel/compiler/brw_fs_nir.cpp
@@ -89,39 +89,11 @@ emit_system_values_block(nir_block *block, fs_visitor *v)
          unreachable("should be lowered by lower_vertex_id().");
 
       case nir_intrinsic_load_vertex_id_zero_base:
-         assert(v->stage == MESA_SHADER_VERTEX);
-         reg = &v->nir_system_values[SYSTEM_VALUE_VERTEX_ID_ZERO_BASE];
-         if (reg->file == BAD_FILE)
-            *reg = *v->emit_vs_system_value(SYSTEM_VALUE_VERTEX_ID_ZERO_BASE);
-         break;
-
       case nir_intrinsic_load_base_vertex:
-         assert(v->stage == MESA_SHADER_VERTEX);
-         reg = &v->nir_system_values[SYSTEM_VALUE_BASE_VERTEX];
-         if (reg->file == BAD_FILE)
-            *reg = *v->emit_vs_system_value(SYSTEM_VALUE_BASE_VERTEX);
-         break;
-
       case nir_intrinsic_load_instance_id:
-         assert(v->stage == MESA_SHADER_VERTEX);
-         reg = &v->nir_system_values[SYSTEM_VALUE_INSTANCE_ID];
-         if (reg->file == BAD_FILE)
-            *reg = *v->emit_vs_system_value(SYSTEM_VALUE_INSTANCE_ID);
-         break;
-
       case nir_intrinsic_load_base_instance:
-         assert(v->stage == MESA_SHADER_VERTEX);
-         reg = &v->nir_system_values[SYSTEM_VALUE_BASE_INSTANCE];
-         if (reg->file == BAD_FILE)
-            *reg = *v->emit_vs_system_value(SYSTEM_VALUE_BASE_INSTANCE);
-         break;
-
       case nir_intrinsic_load_draw_id:
-         assert(v->stage == MESA_SHADER_VERTEX);
-         reg = &v->nir_system_values[SYSTEM_VALUE_DRAW_ID];
-         if (reg->file == BAD_FILE)
-            *reg = *v->emit_vs_system_value(SYSTEM_VALUE_DRAW_ID);
-         break;
+         unreachable("should be lowered by brw_nir_lower_vs_inputs().");
 
       case nir_intrinsic_load_invocation_id:
          if (v->stage == MESA_SHADER_TESS_CTRL)
diff --git a/src/intel/compiler/brw_fs_visitor.cpp b/src/intel/compiler/brw_fs_visitor.cpp
index dd09fbdd64c..758c8bf44a9 100644
--- a/src/intel/compiler/brw_fs_visitor.cpp
+++ b/src/intel/compiler/brw_fs_visitor.cpp
@@ -32,44 +32,6 @@
 
 using namespace brw;
 
-fs_reg *
-fs_visitor::emit_vs_system_value(int location)
-{
-   fs_reg *reg = new(this->mem_ctx)
-      fs_reg(ATTR, 4 * _mesa_bitcount_64(nir->info.inputs_read),
-             BRW_REGISTER_TYPE_D);
-
-   switch (location) {
-   case SYSTEM_VALUE_BASE_VERTEX:
-      reg->offset = 0;
-      break;
-   case SYSTEM_VALUE_BASE_INSTANCE:
-      reg->offset = REG_SIZE;
-      break;
-   case SYSTEM_VALUE_VERTEX_ID:
-      unreachable("should have been lowered");
-   case SYSTEM_VALUE_VERTEX_ID_ZERO_BASE:
-      reg->offset = 2 * REG_SIZE;
-      break;
-   case SYSTEM_VALUE_INSTANCE_ID:
-      reg->offset = 3 * REG_SIZE;
-      break;
-   case SYSTEM_VALUE_DRAW_ID:
-      if (nir->info.system_values_read &
-          (BITFIELD64_BIT(SYSTEM_VALUE_BASE_VERTEX) |
-           BITFIELD64_BIT(SYSTEM_VALUE_BASE_INSTANCE) |
-           BITFIELD64_BIT(SYSTEM_VALUE_VERTEX_ID_ZERO_BASE) |
-           BITFIELD64_BIT(SYSTEM_VALUE_INSTANCE_ID)))
-         reg->nr += 4;
-      reg->offset = 0;
-      break;
-   default:
-      unreachable("not reached");
-   }
-
-   return reg;
-}
-
 /* Sample from the MCS surface attached to this multisample texture. */
 fs_reg
 fs_visitor::emit_mcs_fetch(const fs_reg &coordinate, unsigned components,
diff --git a/src/intel/compiler/brw_nir.c b/src/intel/compiler/brw_nir.c
index 2c44d8f1828..a9d909e8db0 100644
--- a/src/intel/compiler/brw_nir.c
+++ b/src/intel/compiler/brw_nir.c
@@ -259,18 +259,84 @@ brw_nir_lower_vs_inputs(nir_shader *nir,
    if (!is_scalar)
       return;
 
+   /* Whether or not we have any system generated values.  gl_DrawID is not
+    * included here as it lives in its own vec4.
+    */
+   const bool has_sgvs =
+      nir->info.system_values_read &
+      (BITFIELD64_BIT(SYSTEM_VALUE_BASE_VERTEX) |
+       BITFIELD64_BIT(SYSTEM_VALUE_BASE_INSTANCE) |
+       BITFIELD64_BIT(SYSTEM_VALUE_VERTEX_ID_ZERO_BASE) |
+       BITFIELD64_BIT(SYSTEM_VALUE_INSTANCE_ID));
+
+   const unsigned num_inputs = _mesa_bitcount_64(nir->info.inputs_read);
+
    nir_foreach_function(function, nir) {
       if (!function->impl)
          continue;
 
+      nir_builder b;
+      nir_builder_init(&b, function->impl);
+
       nir_foreach_block(block, function->impl) {
-         nir_foreach_instr(instr, block) {
+         nir_foreach_instr_safe(instr, block) {
             if (instr->type != nir_instr_type_intrinsic)
                continue;
 
             nir_intrinsic_instr *intrin = nir_instr_as_intrinsic(instr);
 
-            if (intrin->intrinsic == nir_intrinsic_load_input) {
+            switch (intrin->intrinsic) {
+            case nir_intrinsic_load_base_vertex:
+            case nir_intrinsic_load_base_instance:
+            case nir_intrinsic_load_vertex_id_zero_base:
+            case nir_intrinsic_load_instance_id:
+            case nir_intrinsic_load_draw_id: {
+               b.cursor = nir_after_instr(&intrin->instr);
+
+               /* gl_VertexID and friends are stored by the VF as the last
+                * vertex element.  We convert them to load_input intrinsics at
+                * the right location.
+                */
+               nir_intrinsic_instr *load =
+                  nir_intrinsic_instr_create(nir, nir_intrinsic_load_input);
+               load->src[0] = nir_src_for_ssa(nir_imm_int(&b, 0));
+
+               nir_intrinsic_set_base(load, num_inputs);
+               switch (intrin->intrinsic) {
+               case nir_intrinsic_load_base_vertex:
+                  nir_intrinsic_set_component(load, 0);
+                  break;
+               case nir_intrinsic_load_base_instance:
+                  nir_intrinsic_set_component(load, 1);
+                  break;
+               case nir_intrinsic_load_vertex_id_zero_base:
+                  nir_intrinsic_set_component(load, 2);
+                  break;
+               case nir_intrinsic_load_instance_id:
+                  nir_intrinsic_set_component(load, 3);
+                  break;
+               case nir_intrinsic_load_draw_id:
+                  /* gl_DrawID is stored right after gl_VertexID and friends
+                   * if any of them exist.
+                   */
+                  nir_intrinsic_set_base(load, num_inputs + has_sgvs);
+                  nir_intrinsic_set_component(load, 0);
+                  break;
+               default:
+                  unreachable("Invalid system value intrinsic");
+               }
+
+               load->num_components = 1;
+               nir_ssa_dest_init(&load->instr, &load->dest, 1, 32, NULL);
+               nir_builder_instr_insert(&b, &load->instr);
+
+               nir_ssa_def_rewrite_uses(&intrin->dest.ssa,
+                                        nir_src_for_ssa(&load->dest.ssa));
+               nir_instr_remove(&intrin->instr);
+               break;
+            }
+
+            case nir_intrinsic_load_input: {
                /* Attributes come in a contiguous block, ordered by their
                 * gl_vert_attrib value.  That means we can compute the slot
                 * number for an attribute by masking out the enabled attributes
@@ -280,6 +346,11 @@ brw_nir_lower_vs_inputs(nir_shader *nir,
                int slot = _mesa_bitcount_64(nir->info.inputs_read &
                                             BITFIELD64_MASK(attr));
                nir_intrinsic_set_base(intrin, slot);
+               break;
+            }
+
+            default:
+               break; /* Nothing to do */
             }
          }
       }