source/blender/draw/engines/workbench/shaders/infos/workbench_prepass_info.hh

@@ -126,13 +126,15 @@ GPU_SHADER_CREATE_INFO(workbench_next_prepass)
 
 GPU_SHADER_CREATE_INFO(workbench_color_material)
     .define("WORKBENCH_COLOR_MATERIAL")
-    .storage_buf(WB_MATERIAL_SLOT, Qualifier::READ, "vec4", "materials_data[]");
+    .storage_buf(WB_MATERIAL_SLOT, Qualifier::READ, "vec4", "materials_data[]")
+    .storage_buf(WB_MATERIAL_START_SLOT, Qualifier::READ, "uint", "material_start_indices[]");
 
 GPU_SHADER_CREATE_INFO(workbench_color_texture)
     .define("WORKBENCH_COLOR_TEXTURE")
     .define("WORKBENCH_TEXTURE_IMAGE_ARRAY")
     .define("WORKBENCH_COLOR_MATERIAL")
     .storage_buf(WB_MATERIAL_SLOT, Qualifier::READ, "vec4", "materials_data[]")
+    .storage_buf(WB_MATERIAL_START_SLOT, Qualifier::READ, "uint", "material_start_indices[]")
     .sampler(1, ImageType::FLOAT_2D, "imageTexture", Frequency::BATCH)
     .sampler(2, ImageType::FLOAT_2D_ARRAY, "imageTileArray", Frequency::BATCH)
     .sampler(3, ImageType::FLOAT_1D_ARRAY, "imageTileData", Frequency::BATCH)

source/blender/draw/engines/workbench/shaders/workbench_material_lib.glsl

@@ -17,7 +17,8 @@ void workbench_material_data_get(int handle,
 #else
 
 #  ifdef WORKBENCH_COLOR_MATERIAL
-  vec4 data = materials_data[handle];
+  uint index = material_start_indices[resource_id] + sub_resource_id;
+  vec4 data = materials_data[index];
 #  else
   vec4 data = vec4(0.0);
 #  endif

source/blender/draw/engines/workbench/workbench_defines.hh

@@ -1,10 +1,14 @@
 /* SPDX-License-Identifier: GPL-2.0-or-later */
 
+/* TEXTURE SLOTS */
 #define WB_MATCAP_SLOT 0
 #define WB_TEXTURE_SLOT 1
 #define WB_TILE_ARRAY_SLOT 2
 #define WB_TILE_DATA_SLOT 3
-#define WB_MATERIAL_SLOT 0
+/* UBO SLOTS */
 #define WB_WORLD_SLOT 4
+/* SSBO SLOTS */
+#define WB_MATERIAL_SLOT 0
+#define WB_MATERIAL_START_SLOT 1
 
 #define WB_RESOLVE_GROUP_SIZE 8

source/blender/draw/engines/workbench/workbench_engine.cc

@@ -38,6 +38,8 @@ class Instance {
   DofPass dof_ps;
   AntiAliasingPass anti_aliasing_ps;
 
+  uint material_count = 0;
+
   /* An array of nullptr GPUMaterial pointers so we can call DRW_cache_object_surface_material_get.
    * They never get actually used. */
   Vector<GPUMaterial *> dummy_gpu_materials = {1, nullptr, {}};
@@ -68,6 +70,7 @@ class Instance {
                                  resolution,
                                  GPU_TEXTURE_USAGE_SHADER_READ | GPU_TEXTURE_USAGE_ATTACHMENT |
                                      GPU_TEXTURE_USAGE_MIP_SWIZZLE_VIEW);
+    material_count = 0;
 
     opaque_ps.sync(scene_state, resources);
     transparent_ps.sync(scene_state, resources);
@@ -82,6 +85,7 @@ class Instance {
   void end_sync()
   {
     resources.material_buf.push_update();
+    resources.per_obj_material_start_buf.push_update();
   }
 
   void object_sync(Manager &manager, ObjectRef &ob_ref)
@@ -177,6 +181,7 @@ class Instance {
   void mesh_sync(Manager &manager, ObjectRef &ob_ref, const ObjectState &object_state)
   {
     ResourceHandle handle = manager.resource_handle(ob_ref);
+    resources.per_obj_material_start_buf.get_or_resize(handle.resource_index()) = material_count;
     bool has_transparent_material = false;
 
     if (object_state.sculpt_pbvh) {
@@ -186,7 +191,7 @@ class Instance {
     }
     else {
       if (object_state.use_per_material_batches) {
-        const int material_count = DRW_cache_object_material_count_get(ob_ref.object);
+        const int slot_count = DRW_cache_object_material_count_get(ob_ref.object);
 
         struct GPUBatch **batches;
         if (object_state.color_type == V3D_SHADING_TEXTURE_COLOR) {
@@ -194,20 +199,18 @@ class Instance {
         }
         else {
           batches = DRW_cache_object_surface_material_get(
-              ob_ref.object, get_dummy_gpu_materials(material_count), material_count);
+              ob_ref.object, get_dummy_gpu_materials(slot_count), slot_count);
         }
 
         if (batches) {
-          for (auto i : IndexRange(material_count)) {
+          for (auto i : IndexRange(slot_count)) {
             if (batches[i] == nullptr) {
               continue;
             }
-            /* TODO(fclem): This create a cull-able instance for each sub-object. This is done
-             * for simplicity to reduce complexity. But this increase the overhead per object.
-             * Instead, we should use an indirection buffer to the material buffer. */
-            ResourceHandle _handle = i == 0 ? handle : manager.resource_handle(ob_ref);
+            ResourceHandle _handle = handle;
+            _handle.sub_index = i;
 
-            Material &mat = resources.material_buf.get_or_resize(_handle.resource_index());
+            Material &mat = resources.material_buf.get_or_resize(material_count++);
 
             if (::Material *_mat = BKE_object_material_get_eval(ob_ref.object, i + 1)) {
               mat = Material(*_mat);
@@ -247,7 +250,7 @@ class Instance {
         }
 
         if (batch) {
-          Material &mat = resources.material_buf.get_or_resize(handle.resource_index());
+          Material &mat = resources.material_buf.get_or_resize(material_count++);
 
           if (object_state.color_type == V3D_SHADING_OBJECT_COLOR) {
             mat = Material(*ob_ref.object);

source/blender/draw/engines/workbench/workbench_mesh_passes.cc

@@ -22,8 +22,9 @@ void MeshPass::init_pass(SceneResources &resources, DRWState state, int clip_pla
   PassMain::init();
   state_set(state, clip_planes);
   bind_texture(WB_MATCAP_SLOT, resources.matcap_tx);
-  bind_ssbo(WB_MATERIAL_SLOT, &resources.material_buf);
   bind_ubo(WB_WORLD_SLOT, resources.world_buf);
+  bind_ssbo(WB_MATERIAL_SLOT, &resources.material_buf);
+  bind_ssbo(WB_MATERIAL_START_SLOT, &resources.per_obj_material_start_buf);
   if (clip_planes > 0) {
     bind_ubo(DRW_CLIPPING_UBO_SLOT, resources.clip_planes_buf);
   }

source/blender/draw/engines/workbench/workbench_private.hh

@@ -148,6 +148,7 @@ struct SceneResources {
   TextureFromPool depth_in_front_tx = "wb_depth_in_front_tx";
 
   StorageVectorBuffer<Material> material_buf = {"material_buf"};
+  StorageVectorBuffer<uint> per_obj_material_start_buf = {"per_obj_material_start_buf"};
   UniformBuffer<WorldData> world_buf = {};
   UniformArrayBuffer<float4, 6> clip_planes_buf;