Cycles: Use SSE-optimized version of triangle intersection for motion triangles

The title says it all actually. Gives up to 10% speedup on test scenes here
on i7-6800K.

Render times on GPU are unreliable here, but there might be some slowdown
caused by watertight nature of intersections.

10 files changed, 10 insertions, 10 deletions

diff --git a/intern/cycles/kernel/bvh/bvh_shadow_all.h b/intern/cycles/kernel/bvh/bvh_shadow_all.h
index a6e23d9f381..8f7c005e961 100644
--- a/intern/cycles/kernel/bvh/bvh_shadow_all.h
+++ b/intern/cycles/kernel/bvh/bvh_shadow_all.h
@@ -209,9 +209,9 @@ bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 #if BVH_FEATURE(BVH_MOTION)
 							case PRIMITIVE_MOTION_TRIANGLE: {
 								hit = motion_triangle_intersect(kg,
+								                                &isect_precalc,
 								                                isect_array,
 								                                P,
-								                                dir,
 								                                ray->time,
 								                                PATH_RAY_SHADOW,
 								                                object,
diff --git a/intern/cycles/kernel/bvh/bvh_subsurface.h b/intern/cycles/kernel/bvh/bvh_subsurface.h
index 11ea1f89c8b..cb7a4e3bc31 100644
--- a/intern/cycles/kernel/bvh/bvh_subsurface.h
+++ b/intern/cycles/kernel/bvh/bvh_subsurface.h
@@ -214,9 +214,9 @@ void BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 						for(; prim_addr < prim_addr2; prim_addr++) {
 							kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
 							motion_triangle_intersect_subsurface(kg,
+							                                     &isect_precalc,
 							                                     ss_isect,
 							                                     P,
-							                                     dir,
 							                                     ray->time,
 							                                     object,
 							                                     prim_addr,
diff --git a/intern/cycles/kernel/bvh/bvh_traversal.h b/intern/cycles/kernel/bvh/bvh_traversal.h
index 7c4671e074f..eac98a3165a 100644
--- a/intern/cycles/kernel/bvh/bvh_traversal.h
+++ b/intern/cycles/kernel/bvh/bvh_traversal.h
@@ -267,9 +267,9 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 								BVH_DEBUG_NEXT_INTERSECTION();
 								kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
 								if(motion_triangle_intersect(kg,
+								                             &isect_precalc,
 								                             isect,
 								                             P,
-								                             dir,
 								                             ray->time,
 								                             visibility,
 								                             object,
diff --git a/intern/cycles/kernel/bvh/bvh_volume.h b/intern/cycles/kernel/bvh/bvh_volume.h
index 45b5a3c687b..da97dae0b99 100644
--- a/intern/cycles/kernel/bvh/bvh_volume.h
+++ b/intern/cycles/kernel/bvh/bvh_volume.h
@@ -215,9 +215,9 @@ bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 									continue;
 								}
 								motion_triangle_intersect(kg,
+								                          &isect_precalc,
 								                          isect,
 								                          P,
-								                          dir,
 								                          ray->time,
 								                          visibility,
 								                          object,
diff --git a/intern/cycles/kernel/bvh/bvh_volume_all.h b/intern/cycles/kernel/bvh/bvh_volume_all.h
index 62018e06e97..6efb7e265d0 100644
--- a/intern/cycles/kernel/bvh/bvh_volume_all.h
+++ b/intern/cycles/kernel/bvh/bvh_volume_all.h
@@ -243,9 +243,9 @@ uint BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 									continue;
 								}
 								hit = motion_triangle_intersect(kg,
+								                                &isect_precalc,
 								                                isect_array,
 								                                P,
-								                                dir,
 								                                ray->time,
 								                                visibility,
 								                                object,
diff --git a/intern/cycles/kernel/bvh/qbvh_shadow_all.h b/intern/cycles/kernel/bvh/qbvh_shadow_all.h
index 5a8ec4768d6..5d960787134 100644
--- a/intern/cycles/kernel/bvh/qbvh_shadow_all.h
+++ b/intern/cycles/kernel/bvh/qbvh_shadow_all.h
@@ -290,9 +290,9 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 #if BVH_FEATURE(BVH_MOTION)
 							case PRIMITIVE_MOTION_TRIANGLE: {
 								hit = motion_triangle_intersect(kg,
+								                                &isect_precalc,
 								                                isect_array,
 								                                P,
-								                                dir,
 								                                ray->time,
 								                                PATH_RAY_SHADOW,
 								                                object,
diff --git a/intern/cycles/kernel/bvh/qbvh_subsurface.h b/intern/cycles/kernel/bvh/qbvh_subsurface.h
index d466fe73d24..d67a7826589 100644
--- a/intern/cycles/kernel/bvh/qbvh_subsurface.h
+++ b/intern/cycles/kernel/bvh/qbvh_subsurface.h
@@ -270,9 +270,9 @@ ccl_device void BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 						for(; prim_addr < prim_addr2; prim_addr++) {
 							kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
 							motion_triangle_intersect_subsurface(kg,
+							                                     &isect_precalc,
 							                                     ss_isect,
 							                                     P,
-							                                     dir,
 							                                     ray->time,
 							                                     object,
 							                                     prim_addr,
diff --git a/intern/cycles/kernel/bvh/qbvh_traversal.h b/intern/cycles/kernel/bvh/qbvh_traversal.h
index bf69bfade1b..d88e0e07203 100644
--- a/intern/cycles/kernel/bvh/qbvh_traversal.h
+++ b/intern/cycles/kernel/bvh/qbvh_traversal.h
@@ -354,9 +354,9 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 								BVH_DEBUG_NEXT_INTERSECTION();
 								kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type);
 								if(motion_triangle_intersect(kg,
+								                             &isect_precalc,
 								                             isect,
 								                             P,
-								                             dir,
 								                             ray->time,
 								                             visibility,
 								                             object,
diff --git a/intern/cycles/kernel/bvh/qbvh_volume.h b/intern/cycles/kernel/bvh/qbvh_volume.h
index d34ddb6ea2b..7cfb2cf8c18 100644
--- a/intern/cycles/kernel/bvh/qbvh_volume.h
+++ b/intern/cycles/kernel/bvh/qbvh_volume.h
@@ -281,7 +281,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 									continue;
 								}
 								/* Intersect ray against primitive. */
-								motion_triangle_intersect(kg, isect, P, dir, ray->time, visibility, object, prim_addr);
+								motion_triangle_intersect(kg, &isect_precalc, isect, P, ray->time, visibility, object, prim_addr);
 							}
 							break;
 						}
diff --git a/intern/cycles/kernel/bvh/qbvh_volume_all.h b/intern/cycles/kernel/bvh/qbvh_volume_all.h
index 4e1ee04916b..3dbac8446d7 100644
--- a/intern/cycles/kernel/bvh/qbvh_volume_all.h
+++ b/intern/cycles/kernel/bvh/qbvh_volume_all.h
@@ -309,7 +309,7 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 									continue;
 								}
 								/* Intersect ray against primitive. */
-								hit = motion_triangle_intersect(kg, isect_array, P, dir, ray->time, visibility, object, prim_addr);
+								hit = motion_triangle_intersect(kg, &isect_precalc, isect_array, P, ray->time, visibility, object, prim_addr);
 								if(hit) {
 									/* Move on to next entry in intersections array. */
 									isect_array++;