refactor(cnn_v3): simplify sphere SDF in shadow pass, remove per-frame alloc

gbuf_shadow.wgsl — dfWithID():
- Sphere: replace inv_model local-space transform with direct world-space
  formula (length(p - center) - radius). Exact, no matrix multiply, no
  floating-point error from matrix inversion that can corrupt soft-shadow
  penumbra over 64 march steps.
- lp/scale now computed only inside the cases that need them (box/torus/plane)
  instead of eagerly for every object.

gbuffer_effect.cc — upload_scene_data():
- Replace per-frame std::vector<GBufObjectData> heap allocation with a
  file-static staging buffer s_obj_staging[256]: zero alloc per frame.

handoff(Gemini): sphere SDF now exact; shadow march should be cleaner.

1 files changed, 9 insertions, 10 deletions

diff --git a/cnn_v3/src/gbuffer_effect.cc b/cnn_v3/src/gbuffer_effect.cc
index b059915..829c199 100644
--- a/cnn_v3/src/gbuffer_effect.cc
+++ b/cnn_v3/src/gbuffer_effect.cc
@@ -30,6 +30,9 @@ struct GBufObjectData {
 static_assert(sizeof(GBufObjectData) == sizeof(float) * 40,
               "GBufObjectData must be 160 bytes");
 
+// Reusable CPU staging buffer (single-threaded demo — no concurrency needed).
+static GBufObjectData s_obj_staging[kGBufMaxObjects];
+
 // GlobalUniforms struct mirroring renderer.h
 struct GBufGlobalUniforms {
   mat4 view_proj;
@@ -446,23 +449,19 @@ void GBufferEffect::upload_scene_data(const Scene& scene,
   wgpuQueueWriteBuffer(ctx_.queue, global_uniforms_buf_.buffer, 0,
                        &gu, sizeof(GBufGlobalUniforms));
 
-  // Upload object data.
+  // Upload object data (no per-frame heap alloc — reuse s_obj_staging).
   if (num_objects > 0) {
     ensure_objects_buffer(num_objects);
-    std::vector<GBufObjectData> obj_data;
-    obj_data.reserve((size_t)num_objects);
     for (int i = 0; i < num_objects; ++i) {
       const Object3D& obj = scene.objects[(size_t)i];
       const mat4 m = obj.get_model_matrix();
-      GBufObjectData d;
-      d.model     = m;
-      d.inv_model = m.inverse();
-      d.color     = obj.color;
-      d.params    = vec4((float)(int)obj.type, 0.0f, 0.0f, 0.0f);
-      obj_data.push_back(d);
+      s_obj_staging[i].model     = m;
+      s_obj_staging[i].inv_model = m.inverse();
+      s_obj_staging[i].color     = obj.color;
+      s_obj_staging[i].params    = vec4((float)(int)obj.type, 0.0f, 0.0f, 0.0f);
     }
     wgpuQueueWriteBuffer(ctx_.queue, objects_buf_.buffer, 0,
-                         obj_data.data(),
+                         s_obj_staging,
                          (size_t)num_objects * sizeof(GBufObjectData));
   }
 }