Spawning surfels uses a new shader pipeline (called `capture`) to avoid
the complexity of modifying the deferred or forward pipeline.
The shaders are light to compile and should only be compiled on bake.

This is very WIP: the surfel projection box is hardcoded and the 3
projections are done every frame. The surfels are placed in a regular
grid because of the raster pipeline used to spawn them.

Note: While this is really fast, I am not sure this will scale well for
highly detailed lighting conditions. In the case where the surfel density
is too low, the aliasing might create really poor result.

@@ -595,6 +617,27 @@ class RENDER_PT_eevee_indirect_lighting_display(RenderButtonsPanel, Panel):
         row.prop(props, "gi_irradiance_display_size", text="Irradiance Size")
         row.prop(props, "gi_show_irradiance", text="", toggle=True)
 
+class RENDER_PT_eevee_next_indirect_lighting_display(RenderButtonsPanel, Panel):

@@ -219,3 +219,3 @@
         if (region->regiontype == RGN_TYPE_TOOL_HEADER) {
           if (((sl->spacetype == SPACE_IMAGE) && hide_image_tool_header) ||
-              sl->spacetype == SPACE_SEQ) {
+              sl->spacetype == SPACE_SEQ)

@@ -0,0 +27,4 @@
+ * Create the job description.
+ * This is called for async (modal) bake operator.
+ * The actual work will be done by `EEVEE_NEXT_lightbake_job()`.
+ * IMPORTANT: Must run on the main thread because of potential GPUContext creation.

@@ -0,0 +42,4 @@
+ * Allocate dependency graph and job description (EEVEE_NEXT_LightBake).
+ * Dependency graph evaluation does *not* happen here. It is delayed until
+ * `EEVEE_NEXT_lightbake_job` runs.
+ * IMPORTANT: Must run on the main thread because of potential GPUContext creation.

@@ -0,0 +42,4 @@
+  int grid_index;
+};
+
+struct ReflectionCube : public LightProbe {};

@@ -0,0 +62,4 @@
+  radiance_transfer(receiver, radiance, L);
+}
+
+void radiance_transfer_sky(inout Surfel receiver, vec3 sky_L)

@@ -1037,3 +1053,2 @@
  */
-class Image {
-};
+class Image {};

@@ -1172,3 +1172,3 @@
     case OB_SOLID:
       if (U.experimental.enable_workbench_next &&
-          STREQ(engine_type->idname, "BLENDER_WORKBENCH_NEXT")) {
+          STREQ(engine_type->idname, "BLENDER_WORKBENCH_NEXT"))

@@ -201,3 +202,3 @@
 }
 
-void blo_do_versions_400(FileData * /*fd*/, Library * /*lib*/, Main *bmain)
+void blo_do_versions_400(FileData * fd, Library * /*lib*/, Main *bmain)

@@ -0,0 +301,4 @@
+                              WM_JOB_TYPE_LIGHT_BAKE);
+
+  LightBake *bake = new LightBake(
+      bmain, view_layer, scene, original_probes, true, frame, delay_ms);

@@ -0,0 +316,4 @@
+  return wm_job;
+}
+
+void *EEVEE_NEXT_lightbake_job_data_alloc(struct Main *bmain,

@@ -0,0 +322,4 @@
+                                          blender::Vector<struct Object *> original_probes,
+                                          int frame)
+{
+  LightBake *bake = new LightBake(bmain, view_layer, scene, original_probes, false, frame);

@@ -0,0 +17,4 @@
+struct wmJob;
+
+/** Opaque type hiding eevee::LightBake. */
+typedef struct EEVEE_NEXT_LightBake EEVEE_NEXT_LightBake;

@@ -451,1 +485,4 @@
 
+void Instance::light_bake_irradiance(
+    Object &probe,
+    std::function<void()> context_enable,

@@ -34,1 +90,4 @@
+    return {};
   }
+  Vector<IrradianceBrickPacked> allocated;
+  allocated.resize(brick_len);

@@ -64,0 +389,4 @@
+    display_grids_ps_.bind_texture("irradiance_c_tx", &irradiance_c_tx);
+    display_grids_ps_.bind_texture("irradiance_d_tx", &irradiance_d_tx);
+
+    int sample_count = static_cast<int>(BKE_lightprobe_grid_cache_frame_sample_count(cache));

@@ -0,0 +68,4 @@
+  std::mutex result_mutex_;
+
+ public:
+  LightBake(struct Main *bmain,

@@ -0,0 +330,4 @@
+
+void EEVEE_NEXT_lightbake_job_data_free(void *job_data)
+{
+  delete reinterpret_cast<LightBake *>(job_data);

@@ -0,0 +52,4 @@
+
+void LightProbeModule::sync_probe(const Object *ob, ObjectHandle &handle)
+{
+  const ::LightProbe *light_probe = (const ::LightProbe *)ob->data;

@@ -0,0 +109,4 @@
+    }
+  }
+
+#if 0 /* TODO make this work with new per object light cache. */

@@ -0,0 +3,4 @@
+ * Virtual shadowmapping: Usage tagging
+ *
+ * Shadow pages are only allocated if they are visible.
+ * This pass iterates the surfels buffer and tag all tiles that are needed for light shadowing as

@@ -213,0 +291,4 @@
+/* -------------------------------------------------------------------- */
+/** \name Load/Store
+ *
+ * This section define the compression scheme of spherical harmonic data.

@@ -21,0 +140,4 @@
+    .uniform_buf(IRRADIANCE_GRID_BUF_SLOT,
+                 "IrradianceGridData",
+                 "grids_infos_buf[IRRADIANCE_GRID_MAX]")
+    /* NOTE: Use uint instead of IrradianceBrickPacked because Metal need to know the exact type.*/

@@ -1760,0 +1790,4 @@
+       "ALL",
+       0,
+       "All Light Probes",
+       "Delete all light probes baked lighting data"},

@@ -1760,0 +1800,4 @@
+       "ACTIVE",
+       0,
+       "Active Only",
+       "Only delete the active light probe baked lighting data"},

@@ -158,1 +158,4 @@
 
+  prop = RNA_def_property(srna, "grid_bake_samples", PROP_INT, PROP_NONE);
+  RNA_def_property_ui_text(
+      prop, "Bake Samples", "Number of rays direction to evaluate when baking");