panvk: Add FS read support for PrimitiveID

Adds support for reading PrimitiveID in a fragment shader assuming the
previous shader was vertex.

If support for shaders that can modify PrimitiveID is added, additional
changes will be required.

Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35367>

src/panfrost/compiler/bifrost_compile.c

@@ -2231,6 +2231,10 @@ bi_emit_intrinsic(bi_builder *b, nir_intrinsic_instr *instr)
       bi_load_sample_id_to(b, dst);
       break;
 
+   case nir_intrinsic_load_primitive_id:
+      bi_mov_i32_to(b, dst, bi_preload(b, 57));
+      break;
+
    case nir_intrinsic_load_front_face: {
       /* (r58 & 1) == 0 means primitive is front facing */
       bi_index primitive_facing = bi_preload(b, 58);

src/panfrost/lib/pan_shader.c

@@ -143,6 +143,9 @@ pan_shader_compile(nir_shader *s, struct pan_compile_inputs *inputs,
       info->fs.reads_frag_coord =
          (s->info.inputs_read & (1 << VARYING_SLOT_POS)) ||
          BITSET_TEST(s->info.system_values_read, SYSTEM_VALUE_FRAG_COORD);
+      info->fs.reads_primitive_id =
+         (s->info.inputs_read & (1 << VARYING_SLOT_PRIMITIVE_ID)) ||
+         BITSET_TEST(s->info.system_values_read, SYSTEM_VALUE_PRIMITIVE_ID);
       info->fs.reads_point_coord =
          s->info.inputs_read & (1 << VARYING_SLOT_PNTC);
       info->fs.reads_face =

src/panfrost/util/pan_ir.h

@@ -206,6 +206,7 @@ struct pan_shader_info {
       struct {
          bool reads_frag_coord;
          bool reads_point_coord;
+         bool reads_primitive_id;
          bool reads_face;
          bool can_discard;
          bool writes_depth;

src/panfrost/vulkan/csf/panvk_vX_cmd_draw.c

@@ -2085,6 +2085,8 @@ set_tiler_idvs_flags(struct cs_builder *b, struct panvk_cmd_buffer *cmdbuf,
    if (dirty) {
       pan_pack(&tiler_idvs_flags, PRIMITIVE_FLAGS, cfg) {
          cfg.draw_mode = translate_prim_topology(ia->primitive_topology);
+         cfg.primitive_index_enable =
+            fs ? fs->info.fs.reads_primitive_id : false;
 
 #if PAN_ARCH < 13
          cfg.point_size_array_format = writes_point_size

src/panfrost/vulkan/jm/panvk_vX_cmd_draw.c

@@ -835,11 +835,14 @@ panvk_emit_tiler_primitive(struct panvk_cmd_buffer *cmdbuf,
       vs->info.vs.writes_point_size &&
       ia->primitive_topology == VK_PRIMITIVE_TOPOLOGY_POINT_LIST;
    bool secondary_shader = vs->info.vs.secondary_enable && fs != NULL;
+   bool fs_reads_primitive_id = fs ? fs->info.fs.reads_primitive_id : false;
 
    pan_pack(prim, PRIMITIVE, cfg) {
       cfg.draw_mode = translate_prim_topology(ia->primitive_topology);
       if (writes_point_size)
          cfg.point_size_array_format = MALI_POINT_SIZE_ARRAY_FORMAT_FP16;
+      cfg.primitive_index_enable = fs_reads_primitive_id;
+      cfg.primitive_index_writeback = fs_reads_primitive_id;
 
       cfg.first_provoking_vertex =
          cmdbuf->state.gfx.render.first_provoking_vertex != U_TRISTATE_NO;
@@ -900,6 +903,24 @@ panvk_emit_tiler_primitive_size(struct panvk_cmd_buffer *cmdbuf,
    }
 }
 
+static uint32_t
+primitive_vertex_count(enum mali_draw_mode in)
+{
+   switch (in) {
+   case MALI_DRAW_MODE_POINTS:
+      return 1;
+   case MALI_DRAW_MODE_LINES:
+   case MALI_DRAW_MODE_LINE_STRIP:
+      return 2;
+   case MALI_DRAW_MODE_TRIANGLES:
+   case MALI_DRAW_MODE_TRIANGLE_STRIP:
+   case MALI_DRAW_MODE_TRIANGLE_FAN:
+      return 3;
+   default:
+      unreachable("Invalid draw mode");
+   }
+}
+
 static void
 panvk_emit_tiler_dcd(struct panvk_cmd_buffer *cmdbuf,
                      const struct panvk_draw_data *draw,
@@ -936,6 +957,14 @@ panvk_emit_tiler_dcd(struct panvk_cmd_buffer *cmdbuf,
       cfg.offset_start = draw->info.vertex.raw_offset;
       cfg.instance_size =
          draw->info.instance.count > 1 ? draw->padded_vertex_count : 1;
+      uint32_t primitives_per_instance =
+         DIV_ROUND_UP(draw->padded_vertex_count,
+                      primitive_vertex_count(
+                         translate_prim_topology(ia->primitive_topology)));
+      /* instance_primitive_size has the same restrictions as
+       * padded_vertex_count, so we can use pan_padded_vertex_count here. */
+      cfg.instance_primitive_size =
+         pan_padded_vertex_count(primitives_per_instance);
       cfg.uniform_buffers = fs_desc_state->tables[PANVK_BIFROST_DESC_TABLE_UBO];
       cfg.push_uniforms = cmdbuf->state.gfx.fs.push_uniforms;
       cfg.textures = fs_desc_state->tables[PANVK_BIFROST_DESC_TABLE_TEXTURE];

src/panfrost/vulkan/panvk_vX_shader.c

@@ -224,6 +224,26 @@ panvk_lower_load_vs_input(nir_builder *b, nir_intrinsic_instr *intrin,
    return true;
 }
 
+static bool
+panvk_lower_load_fs_input(nir_builder *b, nir_intrinsic_instr *intrin,
+                          UNUSED void *data)
+{
+   if (intrin->intrinsic != nir_intrinsic_load_input)
+      return false;
+
+   /* Lower PrimitiveID varying loads to the equivalent intrinsic. This only
+    * works since v6 and will require additional changes if PrimitiveID is
+    * explicitly written to (for example by a geometry shader). */
+   if (nir_intrinsic_io_semantics(intrin).location ==
+       VARYING_SLOT_PRIMITIVE_ID) {
+      b->cursor = nir_before_instr(&intrin->instr);
+      nir_def_replace(&intrin->def, nir_load_primitive_id(b));
+      return true;
+   }
+
+   return false;
+}
+
 #if PAN_ARCH <= 7
 static bool
 lower_gl_pos_layer_writes(nir_builder *b, nir_instr *instr, void *data)
@@ -867,6 +887,9 @@ panvk_lower_nir(struct panvk_device *dev, nir_shader *nir,
    if (stage == MESA_SHADER_VERTEX)
       NIR_PASS(_, nir, nir_shader_intrinsics_pass, panvk_lower_load_vs_input,
                nir_metadata_control_flow, NULL);
+   else if (stage == MESA_SHADER_FRAGMENT)
+      NIR_PASS(_, nir, nir_shader_intrinsics_pass, panvk_lower_load_fs_input,
+               nir_metadata_control_flow, NULL);
 
    /* since valhall, panvk_per_arch(nir_lower_descriptors) separates the
     * driver set and the user sets, and does not need pan_lower_image_index