Cycles: merge of cycles-x branch, a major update to the renderer

This includes much improved GPU rendering performance, viewport interactivity,
new shadow catcher, revamped sampling settings, subsurface scattering anisotropy,
new GPU volume sampling, improved PMJ sampling pattern, and more.

Some features have also been removed or changed, breaking backwards compatibility.
Including the removal of the OpenCL backend, for which alternatives are under
development.

Release notes and code docs:
https://wiki.blender.org/wiki/Reference/Release_Notes/3.0/Cycles
https://wiki.blender.org/wiki/Source/Render/Cycles

Credits:
* Sergey Sharybin
* Brecht Van Lommel
* Patrick Mours (OptiX backend)
* Christophe Hery (subsurface scattering anisotropy)
* William Leeson (PMJ sampling pattern)
* Alaska (various fixes and tweaks)
* Thomas Dinges (various fixes)

For the full commit history, see the cycles-x branch. This squashes together
all the changes since intermediate changes would often fail building or tests.

Ref T87839, T87837, T87836
Fixes T90734, T89353, T80267, T80267, T77185, T69800

3 files changed, 86 insertions, 42 deletions

diff --git a/intern/cycles/bvh/bvh_build.cpp b/intern/cycles/bvh/bvh_build.cpp
index 048c2b95e40..d3497f3a8d8 100644
--- a/intern/cycles/bvh/bvh_build.cpp
+++ b/intern/cycles/bvh/bvh_build.cpp
@@ -832,18 +832,18 @@ BVHNode *BVHBuild::create_leaf_node(const BVHRange &range, const vector<BVHRefer
   typedef StackAllocator<256, float2> LeafTimeStackAllocator;
   typedef StackAllocator<256, BVHReference> LeafReferenceStackAllocator;
 
-  vector<int, LeafStackAllocator> p_type[PRIMITIVE_NUM_TOTAL];
-  vector<int, LeafStackAllocator> p_index[PRIMITIVE_NUM_TOTAL];
-  vector<int, LeafStackAllocator> p_object[PRIMITIVE_NUM_TOTAL];
-  vector<float2, LeafTimeStackAllocator> p_time[PRIMITIVE_NUM_TOTAL];
-  vector<BVHReference, LeafReferenceStackAllocator> p_ref[PRIMITIVE_NUM_TOTAL];
+  vector<int, LeafStackAllocator> p_type[PRIMITIVE_NUM];
+  vector<int, LeafStackAllocator> p_index[PRIMITIVE_NUM];
+  vector<int, LeafStackAllocator> p_object[PRIMITIVE_NUM];
+  vector<float2, LeafTimeStackAllocator> p_time[PRIMITIVE_NUM];
+  vector<BVHReference, LeafReferenceStackAllocator> p_ref[PRIMITIVE_NUM];
 
   /* TODO(sergey): In theory we should be able to store references. */
   vector<BVHReference, LeafReferenceStackAllocator> object_references;
 
-  uint visibility[PRIMITIVE_NUM_TOTAL] = {0};
+  uint visibility[PRIMITIVE_NUM] = {0};
   /* NOTE: Keep initialization in sync with actual number of primitives. */
-  BoundBox bounds[PRIMITIVE_NUM_TOTAL] = {
+  BoundBox bounds[PRIMITIVE_NUM] = {
       BoundBox::empty, BoundBox::empty, BoundBox::empty, BoundBox::empty};
   int ob_num = 0;
   int num_new_prims = 0;
@@ -877,7 +877,7 @@ BVHNode *BVHBuild::create_leaf_node(const BVHRange &range, const vector<BVHRefer
    * TODO(sergey): With some pointer trickery we can write directly to the
    * destination buffers for the non-spatial split BVH.
    */
-  BVHNode *leaves[PRIMITIVE_NUM_TOTAL + 1] = {NULL};
+  BVHNode *leaves[PRIMITIVE_NUM + 1] = {NULL};
   int num_leaves = 0;
   size_t start_index = 0;
   vector<int, LeafStackAllocator> local_prim_type, local_prim_index, local_prim_object;
@@ -888,7 +888,7 @@ BVHNode *BVHBuild::create_leaf_node(const BVHRange &range, const vector<BVHRefer
   if (need_prim_time) {
     local_prim_time.resize(num_new_prims);
   }
-  for (int i = 0; i < PRIMITIVE_NUM_TOTAL; ++i) {
+  for (int i = 0; i < PRIMITIVE_NUM; ++i) {
     int num = (int)p_type[i].size();
     if (num != 0) {
       assert(p_type[i].size() == p_index[i].size());
diff --git a/intern/cycles/bvh/bvh_embree.cpp b/intern/cycles/bvh/bvh_embree.cpp
index 62f543941a9..96852510b63 100644
--- a/intern/cycles/bvh/bvh_embree.cpp
+++ b/intern/cycles/bvh/bvh_embree.cpp
@@ -37,10 +37,10 @@
 /* Kernel includes are necessary so that the filter function for Embree can access the packed BVH.
  */
 #  include "kernel/bvh/bvh_embree.h"
-#  include "kernel/kernel_compat_cpu.h"
-#  include "kernel/kernel_globals.h"
+#  include "kernel/bvh/bvh_util.h"
+#  include "kernel/device/cpu/compat.h"
+#  include "kernel/device/cpu/globals.h"
 #  include "kernel/kernel_random.h"
-#  include "kernel/split/kernel_split_data_types.h"
 
 #  include "render/hair.h"
 #  include "render/mesh.h"
@@ -73,46 +73,69 @@ static void rtc_filter_occluded_func(const RTCFilterFunctionNArguments *args)
   const RTCRay *ray = (RTCRay *)args->ray;
   RTCHit *hit = (RTCHit *)args->hit;
   CCLIntersectContext *ctx = ((IntersectContext *)args->context)->userRayExt;
-  KernelGlobals *kg = ctx->kg;
+  const KernelGlobals *kg = ctx->kg;
 
   switch (ctx->type) {
     case CCLIntersectContext::RAY_SHADOW_ALL: {
-      /* Append the intersection to the end of the array. */
-      if (ctx->num_hits < ctx->max_hits) {
-        Intersection current_isect;
-        kernel_embree_convert_hit(kg, ray, hit, &current_isect);
-        for (size_t i = 0; i < ctx->max_hits; ++i) {
+      Intersection current_isect;
+      kernel_embree_convert_hit(kg, ray, hit, &current_isect);
+
+      /* If no transparent shadows, all light is blocked. */
+      const int flags = intersection_get_shader_flags(kg, &current_isect);
+      if (!(flags & (SD_HAS_TRANSPARENT_SHADOW)) || ctx->max_hits == 0) {
+        ctx->opaque_hit = true;
+        return;
+      }
+
+      /* Test if we need to record this transparent intersection. */
+      if (ctx->num_hits < ctx->max_hits || ray->tfar < ctx->max_t) {
+        /* Skip already recorded intersections. */
+        int num_recorded_hits = min(ctx->num_hits, ctx->max_hits);
+
+        for (int i = 0; i < num_recorded_hits; ++i) {
           if (current_isect.object == ctx->isect_s[i].object &&
               current_isect.prim == ctx->isect_s[i].prim && current_isect.t == ctx->isect_s[i].t) {
             /* This intersection was already recorded, skip it. */
             *args->valid = 0;
-            break;
+            return;
           }
         }
-        Intersection *isect = &ctx->isect_s[ctx->num_hits];
-        ++ctx->num_hits;
-        *isect = current_isect;
-        int prim = kernel_tex_fetch(__prim_index, isect->prim);
-        int shader = 0;
-        if (kernel_tex_fetch(__prim_type, isect->prim) & PRIMITIVE_ALL_TRIANGLE) {
-          shader = kernel_tex_fetch(__tri_shader, prim);
-        }
-        else {
-          float4 str = kernel_tex_fetch(__curves, prim);
-          shader = __float_as_int(str.z);
-        }
-        int flag = kernel_tex_fetch(__shaders, shader & SHADER_MASK).flags;
-        /* If no transparent shadows, all light is blocked. */
-        if (flag & (SD_HAS_TRANSPARENT_SHADOW)) {
-          /* This tells Embree to continue tracing. */
-          *args->valid = 0;
+
+        /* If maximum number of hits was reached, replace the intersection with the
+         * highest distance. We want to find the N closest intersections. */
+        int isect_index = num_recorded_hits;
+        if (num_recorded_hits + 1 >= ctx->max_hits) {
+          float max_t = ctx->isect_s[0].t;
+          int max_recorded_hit = 0;
+
+          for (int i = 1; i < num_recorded_hits; ++i) {
+            if (ctx->isect_s[i].t > max_t) {
+              max_recorded_hit = i;
+              max_t = ctx->isect_s[i].t;
+            }
+          }
+
+          if (num_recorded_hits >= ctx->max_hits) {
+            isect_index = max_recorded_hit;
+          }
+
+          /* Limit the ray distance and stop counting hits beyond this.
+           * TODO: is there some way we can tell Embree to stop intersecting beyond
+           * this distance when max number of hits is reached?. Or maybe it will
+           * become irrelevant if we make max_hits a very high number on the CPU. */
+          ctx->max_t = max(current_isect.t, max_t);
         }
+
+        ctx->isect_s[isect_index] = current_isect;
       }
-      else {
-        /* Increase the number of hits beyond ray.max_hits
-         * so that the caller can detect this as opaque. */
-        ++ctx->num_hits;
-      }
+
+      /* Always increase the number of hits, even beyond ray.max_hits so that
+       * the caller can detect this as and consider it opaque, or trace another
+       * ray. */
+      ++ctx->num_hits;
+
+      /* This tells Embree to continue tracing. */
+      *args->valid = 0;
       break;
     }
     case CCLIntersectContext::RAY_LOCAL:
@@ -329,7 +352,7 @@ void BVHEmbree::build(Progress &progress, Stats *stats, RTCDevice rtc_device_)
     scene = NULL;
   }
 
-  const bool dynamic = params.bvh_type == SceneParams::BVH_DYNAMIC;
+  const bool dynamic = params.bvh_type == BVH_TYPE_DYNAMIC;
 
   scene = rtcNewScene(rtc_device);
   const RTCSceneFlags scene_flags = (dynamic ? RTC_SCENE_FLAG_DYNAMIC : RTC_SCENE_FLAG_NONE) |
diff --git a/intern/cycles/bvh/bvh_params.h b/intern/cycles/bvh/bvh_params.h
index 2dc10f30363..31b3971c110 100644
--- a/intern/cycles/bvh/bvh_params.h
+++ b/intern/cycles/bvh/bvh_params.h
@@ -31,6 +31,27 @@ CCL_NAMESPACE_BEGIN
  */
 typedef KernelBVHLayout BVHLayout;
 
+/* Type of BVH, in terms whether it is supported dynamic updates of meshes
+ * or whether modifying geometry requires full BVH rebuild.
+ */
+enum BVHType {
+  /* BVH supports dynamic updates of geometry.
+   *
+   * Faster for updating BVH tree when doing modifications in viewport,
+   * but slower for rendering.
+   */
+  BVH_TYPE_DYNAMIC = 0,
+  /* BVH tree is calculated for specific scene, updates in geometry
+   * requires full tree rebuild.
+   *
+   * Slower to update BVH tree when modifying objects in viewport, also
+   * slower to build final BVH tree but gives best possible render speed.
+   */
+  BVH_TYPE_STATIC = 1,
+
+  BVH_NUM_TYPES,
+};
+
 /* Names bitflag type to denote which BVH layouts are supported by
  * particular area.
  *