Fix T73064: Embree does not like Bevel shader

Embree's local intersection routine was not prepared
for local intersections without per-object BVH.
Now it should be able to handle any kind of local
intersection, such as AO, bevel and SSS.

Differential Revision: https://developer.blender.org/D6602

3 files changed, 56 insertions, 40 deletions

diff --git a/intern/cycles/bvh/bvh_embree.cpp b/intern/cycles/bvh/bvh_embree.cpp
index ac7cf6501ba..3e4978a2c0a 100644
--- a/intern/cycles/bvh/bvh_embree.cpp
+++ b/intern/cycles/bvh/bvh_embree.cpp
@@ -146,7 +146,21 @@ static void rtc_filter_occluded_func(const RTCFilterFunctionNArguments *args)
       }
       break;
     }
+    case CCLIntersectContext::RAY_LOCAL:
     case CCLIntersectContext::RAY_SSS: {
+      /* Check if it's hitting the correct object. */
+      Intersection current_isect;
+      if (ctx->type == CCLIntersectContext::RAY_SSS) {
+        kernel_embree_convert_sss_hit(kg, ray, hit, &current_isect, ctx->local_object_id);
+      }
+      else {
+        kernel_embree_convert_hit(kg, ray, hit, &current_isect);
+        if (ctx->local_object_id != current_isect.object) {
+          /* This tells Embree to continue tracing. */
+          *args->valid = 0;
+        }
+      }
+
       /* No intersection information requested, just return a hit. */
       if (ctx->max_hits == 0) {
         break;
@@ -160,8 +174,8 @@ static void rtc_filter_occluded_func(const RTCFilterFunctionNArguments *args)
       }
 
       /* See triangle_intersect_subsurface() for the native equivalent. */
-      for (int i = min(ctx->max_hits, ctx->ss_isect->num_hits) - 1; i >= 0; --i) {
-        if (ctx->ss_isect->hits[i].t == ray->tfar) {
+      for (int i = min(ctx->max_hits, ctx->local_isect->num_hits) - 1; i >= 0; --i) {
+        if (ctx->local_isect->hits[i].t == ray->tfar) {
           /* This tells Embree to continue tracing. */
           *args->valid = 0;
           break;
@@ -172,14 +186,14 @@ static void rtc_filter_occluded_func(const RTCFilterFunctionNArguments *args)
 
       if (ctx->lcg_state) {
 
-        ++ctx->ss_isect->num_hits;
-        if (ctx->ss_isect->num_hits <= ctx->max_hits) {
-          hit_idx = ctx->ss_isect->num_hits - 1;
+        ++ctx->local_isect->num_hits;
+        if (ctx->local_isect->num_hits <= ctx->max_hits) {
+          hit_idx = ctx->local_isect->num_hits - 1;
         }
         else {
           /* reservoir sampling: if we are at the maximum number of
            * hits, randomly replace element or skip it */
-          hit_idx = lcg_step_uint(ctx->lcg_state) % ctx->ss_isect->num_hits;
+          hit_idx = lcg_step_uint(ctx->lcg_state) % ctx->local_isect->num_hits;
 
           if (hit_idx >= ctx->max_hits) {
             /* This tells Embree to continue tracing. */
@@ -189,15 +203,11 @@ static void rtc_filter_occluded_func(const RTCFilterFunctionNArguments *args)
         }
       }
       else {
-        ctx->ss_isect->num_hits = 1;
+        ctx->local_isect->num_hits = 1;
       }
       /* record intersection */
-      kernel_embree_convert_local_hit(
-          kg, ray, hit, &ctx->ss_isect->hits[hit_idx], ctx->sss_object_id);
-      ctx->ss_isect->Ng[hit_idx].x = hit->Ng_x;
-      ctx->ss_isect->Ng[hit_idx].y = hit->Ng_y;
-      ctx->ss_isect->Ng[hit_idx].z = hit->Ng_z;
-      ctx->ss_isect->Ng[hit_idx] = normalize(ctx->ss_isect->Ng[hit_idx]);
+      ctx->local_isect->hits[hit_idx] = current_isect;
+      ctx->local_isect->Ng[hit_idx] = normalize(make_float3(hit->Ng_x, hit->Ng_y, hit->Ng_z));
       /* This tells Embree to continue tracing .*/
       *args->valid = 0;
       break;
diff --git a/intern/cycles/kernel/bvh/bvh.h b/intern/cycles/kernel/bvh/bvh.h
index 0346f5e09e7..8e17ab9af7a 100644
--- a/intern/cycles/kernel/bvh/bvh.h
+++ b/intern/cycles/kernel/bvh/bvh.h
@@ -329,24 +329,26 @@ ccl_device_intersect bool scene_intersect_local(KernelGlobals *kg,
 
 #    ifdef __EMBREE__
   if (kernel_data.bvh.scene) {
-    CCLIntersectContext ctx(kg, CCLIntersectContext::RAY_SSS);
+    const bool has_bvh = !(kernel_tex_fetch(__object_flag, local_object) &
+                           SD_OBJECT_TRANSFORM_APPLIED);
+    CCLIntersectContext ctx(
+        kg, has_bvh ? CCLIntersectContext::RAY_SSS : CCLIntersectContext::RAY_LOCAL);
     ctx.lcg_state = lcg_state;
     ctx.max_hits = max_hits;
-    ctx.ss_isect = local_isect;
+    ctx.local_isect = local_isect;
     local_isect->num_hits = 0;
-    ctx.sss_object_id = local_object;
+    ctx.local_object_id = local_object;
     IntersectContext rtc_ctx(&ctx);
     RTCRay rtc_ray;
     kernel_embree_setup_ray(*ray, rtc_ray, PATH_RAY_ALL_VISIBILITY);
 
-    /* Get the Embree scene for this intersection. */
-    RTCGeometry geom = rtcGetGeometry(kernel_data.bvh.scene, local_object * 2);
-    if (geom) {
-      float3 P = ray->P;
-      float3 dir = ray->D;
-      float3 idir = ray->D;
-      const int object_flag = kernel_tex_fetch(__object_flag, local_object);
-      if (!(object_flag & SD_OBJECT_TRANSFORM_APPLIED)) {
+    /* If this object has its own BVH, use it. */
+    if (has_bvh) {
+      RTCGeometry geom = rtcGetGeometry(kernel_data.bvh.scene, local_object * 2);
+      if (geom) {
+        float3 P = ray->P;
+        float3 dir = ray->D;
+        float3 idir = ray->D;
         Transform ob_itfm;
         rtc_ray.tfar = bvh_instance_motion_push(
             kg, local_object, ray, &P, &dir, &idir, ray->t, &ob_itfm);
@@ -360,11 +362,15 @@ ccl_device_intersect bool scene_intersect_local(KernelGlobals *kg,
         rtc_ray.dir_x = dir.x;
         rtc_ray.dir_y = dir.y;
         rtc_ray.dir_z = dir.z;
+        RTCScene scene = (RTCScene)rtcGetGeometryUserData(geom);
+        kernel_assert(scene);
+        if (scene) {
+          rtcOccluded1(scene, &rtc_ctx.context, &rtc_ray);
+        }
       }
-      RTCScene scene = (RTCScene)rtcGetGeometryUserData(geom);
-      if (scene) {
-        rtcOccluded1(scene, &rtc_ctx.context, &rtc_ray);
-      }
+    }
+    else {
+      rtcOccluded1(kernel_data.bvh.scene, &rtc_ctx.context, &rtc_ray);
     }
 
     return local_isect->num_hits > 0;
diff --git a/intern/cycles/kernel/bvh/bvh_embree.h b/intern/cycles/kernel/bvh/bvh_embree.h
index 661bba54fd4..ffea7d37440 100644
--- a/intern/cycles/kernel/bvh/bvh_embree.h
+++ b/intern/cycles/kernel/bvh/bvh_embree.h
@@ -28,9 +28,9 @@ struct CCLIntersectContext {
   typedef enum {
     RAY_REGULAR = 0,
     RAY_SHADOW_ALL = 1,
-    RAY_SSS = 2,
-    RAY_VOLUME_ALL = 3,
-
+    RAY_LOCAL = 2,
+    RAY_SSS = 3,
+    RAY_VOLUME_ALL = 4,
   } RayType;
 
   KernelGlobals *kg;
@@ -42,8 +42,8 @@ struct CCLIntersectContext {
   int num_hits;
 
   /* for SSS Rays: */
-  LocalIntersection *ss_isect;
-  int sss_object_id;
+  LocalIntersection *local_isect;
+  int local_object_id;
   uint *lcg_state;
 
   CCLIntersectContext(KernelGlobals *kg_, RayType type_)
@@ -53,8 +53,8 @@ struct CCLIntersectContext {
     max_hits = 1;
     num_hits = 0;
     isect_s = NULL;
-    ss_isect = NULL;
-    sss_object_id = -1;
+    local_isect = NULL;
+    local_object_id = -1;
     lcg_state = NULL;
   }
 };
@@ -121,11 +121,11 @@ ccl_device_inline void kernel_embree_convert_hit(KernelGlobals *kg,
   isect->type = kernel_tex_fetch(__prim_type, isect->prim);
 }
 
-ccl_device_inline void kernel_embree_convert_local_hit(KernelGlobals *kg,
-                                                       const RTCRay *ray,
-                                                       const RTCHit *hit,
-                                                       Intersection *isect,
-                                                       int local_object_id)
+ccl_device_inline void kernel_embree_convert_sss_hit(KernelGlobals *kg,
+                                                     const RTCRay *ray,
+                                                     const RTCHit *hit,
+                                                     Intersection *isect,
+                                                     int local_object_id)
 {
   isect->u = 1.0f - hit->v - hit->u;
   isect->v = hit->u;