Cycles: Move triangle intersection precalc to an util file

This is a preparation work for the followup commit which wil l move
remaining parts of Woop intersection logic to an utility file.

Doing it as a separate commit to keep changes more atomic and easier
to bisect when/if needed.

12 files changed, 116 insertions, 96 deletions

diff --git a/intern/cycles/kernel/bvh/bvh_shadow_all.h b/intern/cycles/kernel/bvh/bvh_shadow_all.h
index b4f65bc8efd..a6e23d9f381 100644
--- a/intern/cycles/kernel/bvh/bvh_shadow_all.h
+++ b/intern/cycles/kernel/bvh/bvh_shadow_all.h
@@ -100,8 +100,8 @@ bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 	gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
 #endif  /* __KERNEL_SSE2__ */
 
-	IsectPrecalc isect_precalc;
-	triangle_intersect_precalc(dir, &isect_precalc);
+	TriangleIsectPrecalc isect_precalc;
+	ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 	/* traversal loop */
 	do {
@@ -314,7 +314,7 @@ bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 					isect_t = bvh_instance_push(kg, object, ray, &P, &dir, &idir, isect_t);
 #  endif
 
-					triangle_intersect_precalc(dir, &isect_precalc);
+					ray_triangle_intersect_precalc(dir, &isect_precalc);
 					num_hits_in_instance = 0;
 					isect_array->t = isect_t;
 
@@ -354,7 +354,7 @@ bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 				bvh_instance_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac);
 #  endif
 
-				triangle_intersect_precalc(dir, &isect_precalc);
+				ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 				/* scale isect->t to adjust for instancing */
 				for(int i = 0; i < num_hits_in_instance; i++) {
@@ -367,7 +367,7 @@ bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 #  else
 				bvh_instance_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX);
 #  endif
-				triangle_intersect_precalc(dir, &isect_precalc);
+				ray_triangle_intersect_precalc(dir, &isect_precalc);
 			}
 
 			isect_t = tmax;
diff --git a/intern/cycles/kernel/bvh/bvh_subsurface.h b/intern/cycles/kernel/bvh/bvh_subsurface.h
index 583f7f7c469..11ea1f89c8b 100644
--- a/intern/cycles/kernel/bvh/bvh_subsurface.h
+++ b/intern/cycles/kernel/bvh/bvh_subsurface.h
@@ -109,8 +109,8 @@ void BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 	gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
 #endif
 
-	IsectPrecalc isect_precalc;
-	triangle_intersect_precalc(dir, &isect_precalc);
+	TriangleIsectPrecalc isect_precalc;
+	ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 	/* traversal loop */
 	do {
diff --git a/intern/cycles/kernel/bvh/bvh_traversal.h b/intern/cycles/kernel/bvh/bvh_traversal.h
index 0eca0c8e38b..7c4671e074f 100644
--- a/intern/cycles/kernel/bvh/bvh_traversal.h
+++ b/intern/cycles/kernel/bvh/bvh_traversal.h
@@ -104,8 +104,8 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 	gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
 #endif
 
-	IsectPrecalc isect_precalc;
-	triangle_intersect_precalc(dir, &isect_precalc);
+	TriangleIsectPrecalc isect_precalc;
+	ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 	/* traversal loop */
 	do {
@@ -358,7 +358,7 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 #  else
 					isect->t = bvh_instance_push(kg, object, ray, &P, &dir, &idir, isect->t);
 #  endif
-					triangle_intersect_precalc(dir, &isect_precalc);
+					ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 #  if defined(__KERNEL_SSE2__)
 					Psplat[0] = ssef(P.x);
@@ -395,7 +395,7 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 #  else
 			isect->t = bvh_instance_pop(kg, object, ray, &P, &dir, &idir, isect->t);
 #  endif
-			triangle_intersect_precalc(dir, &isect_precalc);
+			ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 #  if defined(__KERNEL_SSE2__)
 			Psplat[0] = ssef(P.x);
diff --git a/intern/cycles/kernel/bvh/bvh_volume.h b/intern/cycles/kernel/bvh/bvh_volume.h
index 136034aa484..45b5a3c687b 100644
--- a/intern/cycles/kernel/bvh/bvh_volume.h
+++ b/intern/cycles/kernel/bvh/bvh_volume.h
@@ -97,8 +97,8 @@ bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 	gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
 #endif
 
-	IsectPrecalc isect_precalc;
-	triangle_intersect_precalc(dir, &isect_precalc);
+	TriangleIsectPrecalc isect_precalc;
+	ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 	/* traversal loop */
 	do {
@@ -243,7 +243,7 @@ bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 						isect->t = bvh_instance_push(kg, object, ray, &P, &dir, &idir, isect->t);
 #  endif
 
-						triangle_intersect_precalc(dir, &isect_precalc);
+						ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 #  if defined(__KERNEL_SSE2__)
 						Psplat[0] = ssef(P.x);
@@ -286,7 +286,7 @@ bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 			isect->t = bvh_instance_pop(kg, object, ray, &P, &dir, &idir, isect->t);
 #  endif
 
-			triangle_intersect_precalc(dir, &isect_precalc);
+			ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 #  if defined(__KERNEL_SSE2__)
 			Psplat[0] = ssef(P.x);
diff --git a/intern/cycles/kernel/bvh/bvh_volume_all.h b/intern/cycles/kernel/bvh/bvh_volume_all.h
index 6f3346e7634..62018e06e97 100644
--- a/intern/cycles/kernel/bvh/bvh_volume_all.h
+++ b/intern/cycles/kernel/bvh/bvh_volume_all.h
@@ -101,8 +101,8 @@ uint BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 	gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz);
 #endif  /* __KERNEL_SSE2__ */
 
-	IsectPrecalc isect_precalc;
-	triangle_intersect_precalc(dir, &isect_precalc);
+	TriangleIsectPrecalc isect_precalc;
+	ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 	/* traversal loop */
 	do {
@@ -294,7 +294,7 @@ uint BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 						isect_t = bvh_instance_push(kg, object, ray, &P, &dir, &idir, isect_t);
 #  endif
 
-						triangle_intersect_precalc(dir, &isect_precalc);
+						ray_triangle_intersect_precalc(dir, &isect_precalc);
 						num_hits_in_instance = 0;
 						isect_array->t = isect_t;
 
@@ -340,7 +340,7 @@ uint BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 #  else
 				bvh_instance_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac);
 #  endif
-				triangle_intersect_precalc(dir, &isect_precalc);
+				ray_triangle_intersect_precalc(dir, &isect_precalc);
 				/* Scale isect->t to adjust for instancing. */
 				for(int i = 0; i < num_hits_in_instance; i++) {
 					(isect_array-i-1)->t *= t_fac;
@@ -352,7 +352,7 @@ uint BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg,
 #  else
 				bvh_instance_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX);
 #  endif
-				triangle_intersect_precalc(dir, &isect_precalc);
+				ray_triangle_intersect_precalc(dir, &isect_precalc);
 			}
 
 			isect_t = tmax;
diff --git a/intern/cycles/kernel/bvh/qbvh_shadow_all.h b/intern/cycles/kernel/bvh/qbvh_shadow_all.h
index 2a4da3eea82..5a8ec4768d6 100644
--- a/intern/cycles/kernel/bvh/qbvh_shadow_all.h
+++ b/intern/cycles/kernel/bvh/qbvh_shadow_all.h
@@ -96,8 +96,8 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 	                       &near_x, &near_y, &near_z,
 	                       &far_x, &far_y, &far_z);
 
-	IsectPrecalc isect_precalc;
-	triangle_intersect_precalc(dir, &isect_precalc);
+	TriangleIsectPrecalc isect_precalc;
+	ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 	/* Traversal loop. */
 	do {
@@ -414,7 +414,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 					org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
 #  endif
 
-					triangle_intersect_precalc(dir, &isect_precalc);
+					ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 					++stack_ptr;
 					kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
@@ -471,7 +471,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 			org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
 #  endif
 
-			triangle_intersect_precalc(dir, &isect_precalc);
+			ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 			object = OBJECT_NONE;
 			node_addr = traversal_stack[stack_ptr].addr;
diff --git a/intern/cycles/kernel/bvh/qbvh_subsurface.h b/intern/cycles/kernel/bvh/qbvh_subsurface.h
index a6431a94e6e..d466fe73d24 100644
--- a/intern/cycles/kernel/bvh/qbvh_subsurface.h
+++ b/intern/cycles/kernel/bvh/qbvh_subsurface.h
@@ -105,8 +105,8 @@ ccl_device void BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 	                       &near_x, &near_y, &near_z,
 	                       &far_x, &far_y, &far_z);
 
-	IsectPrecalc isect_precalc;
-	triangle_intersect_precalc(dir, &isect_precalc);
+	TriangleIsectPrecalc isect_precalc;
+	ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 	/* Traversal loop. */
 	do {
diff --git a/intern/cycles/kernel/bvh/qbvh_traversal.h b/intern/cycles/kernel/bvh/qbvh_traversal.h
index c20a8f3703f..bf69bfade1b 100644
--- a/intern/cycles/kernel/bvh/qbvh_traversal.h
+++ b/intern/cycles/kernel/bvh/qbvh_traversal.h
@@ -106,8 +106,8 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 	                       &near_x, &near_y, &near_z,
 	                       &far_x, &far_y, &far_z);
 
-	IsectPrecalc isect_precalc;
-	triangle_intersect_precalc(dir, &isect_precalc);
+	TriangleIsectPrecalc isect_precalc;
+	ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 	/* Traversal loop. */
 	do {
@@ -447,7 +447,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 					org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
 #  endif
 
-					triangle_intersect_precalc(dir, &isect_precalc);
+					ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 					++stack_ptr;
 					kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
@@ -489,7 +489,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 			org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
 #  endif
 
-			triangle_intersect_precalc(dir, &isect_precalc);
+			ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 			object = OBJECT_NONE;
 			node_addr = traversal_stack[stack_ptr].addr;
diff --git a/intern/cycles/kernel/bvh/qbvh_volume.h b/intern/cycles/kernel/bvh/qbvh_volume.h
index 859c5da808b..d34ddb6ea2b 100644
--- a/intern/cycles/kernel/bvh/qbvh_volume.h
+++ b/intern/cycles/kernel/bvh/qbvh_volume.h
@@ -91,8 +91,8 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 	                       &near_x, &near_y, &near_z,
 	                       &far_x, &far_y, &far_z);
 
-	IsectPrecalc isect_precalc;
-	triangle_intersect_precalc(dir, &isect_precalc);
+	TriangleIsectPrecalc isect_precalc;
+	ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 	/* Traversal loop. */
 	do {
@@ -316,7 +316,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 						org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
 #  endif
 
-						triangle_intersect_precalc(dir, &isect_precalc);
+						ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 						++stack_ptr;
 						kernel_assert(stack_ptr < BVH_QSTACK_SIZE);
@@ -362,7 +362,7 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 			org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
 #  endif
 
-			triangle_intersect_precalc(dir, &isect_precalc);
+			ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 			object = OBJECT_NONE;
 			node_addr = traversal_stack[stack_ptr].addr;
diff --git a/intern/cycles/kernel/bvh/qbvh_volume_all.h b/intern/cycles/kernel/bvh/qbvh_volume_all.h
index bbe588c878f..4e1ee04916b 100644
--- a/intern/cycles/kernel/bvh/qbvh_volume_all.h
+++ b/intern/cycles/kernel/bvh/qbvh_volume_all.h
@@ -95,8 +95,8 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 	                       &near_x, &near_y, &near_z,
 	                       &far_x, &far_y, &far_z);
 
-	IsectPrecalc isect_precalc;
-	triangle_intersect_precalc(dir, &isect_precalc);
+	TriangleIsectPrecalc isect_precalc;
+	ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 	/* Traversal loop. */
 	do {
@@ -367,7 +367,7 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 						org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
 #  endif
 
-						triangle_intersect_precalc(dir, &isect_precalc);
+						ray_triangle_intersect_precalc(dir, &isect_precalc);
 						num_hits_in_instance = 0;
 						isect_array->t = isect_t;
 
@@ -432,7 +432,7 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 			org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z));
 #  endif
 
-			triangle_intersect_precalc(dir, &isect_precalc);
+			ray_triangle_intersect_precalc(dir, &isect_precalc);
 
 			object = OBJECT_NONE;
 			node_addr = traversal_stack[stack_ptr].addr;
diff --git a/intern/cycles/kernel/geom/geom_triangle_intersect.h b/intern/cycles/kernel/geom/geom_triangle_intersect.h
index ca6e54c6b33..12700c7a246 100644
--- a/intern/cycles/kernel/geom/geom_triangle_intersect.h
+++ b/intern/cycles/kernel/geom/geom_triangle_intersect.h
@@ -40,64 +40,8 @@ CCL_NAMESPACE_BEGIN
  * http://jcgt.org/published/0002/01/05/paper.pdf
  */
 
-/* Precalculated data for the ray->tri intersection. */
-typedef struct IsectPrecalc {
-	/* Maximal dimension kz, and orthogonal dimensions. */
-	int kx, ky, kz;
-
-	/* Shear constants. */
-	float Sx, Sy, Sz;
-} IsectPrecalc;
-
-#if (defined(__KERNEL_OPENCL_APPLE__)) || \
-    (defined(__KERNEL_CUDA__) && (defined(i386) || defined(_M_IX86)))
-ccl_device_noinline
-#else
-ccl_device_inline
-#endif
-void triangle_intersect_precalc(float3 dir,
-                                IsectPrecalc *isect_precalc)
-{
-	/* Calculate dimension where the ray direction is maximal. */
-#ifndef __KERNEL_SSE__
-	int kz = util_max_axis(make_float3(fabsf(dir.x),
-	                                   fabsf(dir.y),
-	                                   fabsf(dir.z)));
-	int kx = kz + 1; if(kx == 3) kx = 0;
-	int ky = kx + 1; if(ky == 3) ky = 0;
-#else
-	int kx, ky, kz;
-	/* Avoiding mispredicted branch on direction. */
-	kz = util_max_axis(fabs(dir));
-	static const char inc_xaxis[] = {1, 2, 0, 55};
-	static const char inc_yaxis[] = {2, 0, 1, 55};
-	kx = inc_xaxis[kz];
-	ky = inc_yaxis[kz];
-#endif
-
-	float dir_kz = IDX(dir, kz);
-
-	/* Swap kx and ky dimensions to preserve winding direction of triangles. */
-	if(dir_kz < 0.0f) {
-		int tmp = kx;
-		kx = ky;
-		ky = tmp;
-	}
-
-	/* Calculate the shear constants. */
-	float inv_dir_z = 1.0f / dir_kz;
-	isect_precalc->Sx = IDX(dir, kx) * inv_dir_z;
-	isect_precalc->Sy = IDX(dir, ky) * inv_dir_z;
-	isect_precalc->Sz = inv_dir_z;
-
-	/* Store the dimensions. */
-	isect_precalc->kx = kx;
-	isect_precalc->ky = ky;
-	isect_precalc->kz = kz;
-}
-
 ccl_device_inline bool triangle_intersect(KernelGlobals *kg,
-                                          const IsectPrecalc *isect_precalc,
+                                          const TriangleIsectPrecalc *isect_precalc,
                                           Intersection *isect,
                                           float3 P,
                                           uint visibility,
@@ -255,7 +199,7 @@ ccl_device_inline bool triangle_intersect(KernelGlobals *kg,
 #ifdef __SUBSURFACE__
 ccl_device_inline void triangle_intersect_subsurface(
         KernelGlobals *kg,
-        const IsectPrecalc *isect_precalc,
+        const TriangleIsectPrecalc *isect_precalc,
         SubsurfaceIntersection *ss_isect,
         float3 P,
         int object,
diff --git a/intern/cycles/util/util_math_intersect.h b/intern/cycles/util/util_math_intersect.h
index 870f33cee0c..7e02c04d939 100644
--- a/intern/cycles/util/util_math_intersect.h
+++ b/intern/cycles/util/util_math_intersect.h
@@ -79,6 +79,82 @@ ccl_device bool ray_aligned_disk_intersect(
 	return true;
 }
 
+/* Optimized watertight ray-triangle intersection.
+ *
+ * Sven Woop
+ * Watertight Ray/Triangle Intersection
+ *
+ * http://jcgt.org/published/0002/01/05/paper.pdf
+ */
+
+/* Precalculated data for the ray->tri intersection. */
+typedef struct TriangleIsectPrecalc {
+	/* Maximal dimension kz, and orthogonal dimensions. */
+	int kx, ky, kz;
+
+	/* Shear constants. */
+	float Sx, Sy, Sz;
+} TriangleIsectPrecalc;
+
+/* Workaround stupidness of CUDA/OpenCL which doesn't allow to access indexed
+ * component of float3 value.
+ */
+#ifdef __KERNEL_GPU__
+#  define IDX(vec, idx) \
+    ((idx == 0) ? ((vec).x) : ( (idx == 1) ? ((vec).y) : ((vec).z) ))
+#else
+#  define IDX(vec, idx) ((vec)[idx])
+#endif
+
+#if (defined(__KERNEL_OPENCL_APPLE__)) || \
+    (defined(__KERNEL_CUDA__) && (defined(i386) || defined(_M_IX86)))
+ccl_device_noinline
+#else
+ccl_device_inline
+#endif
+void ray_triangle_intersect_precalc(float3 dir,
+                                    TriangleIsectPrecalc *isect_precalc)
+{
+	/* Calculate dimension where the ray direction is maximal. */
+#ifndef __KERNEL_SSE__
+	int kz = util_max_axis(make_float3(fabsf(dir.x),
+	                                   fabsf(dir.y),
+	                                   fabsf(dir.z)));
+	int kx = kz + 1; if(kx == 3) kx = 0;
+	int ky = kx + 1; if(ky == 3) ky = 0;
+#else
+	int kx, ky, kz;
+	/* Avoiding mispredicted branch on direction. */
+	kz = util_max_axis(fabs(dir));
+	static const char inc_xaxis[] = {1, 2, 0, 55};
+	static const char inc_yaxis[] = {2, 0, 1, 55};
+	kx = inc_xaxis[kz];
+	ky = inc_yaxis[kz];
+#endif
+
+	float dir_kz = IDX(dir, kz);
+
+	/* Swap kx and ky dimensions to preserve winding direction of triangles. */
+	if(dir_kz < 0.0f) {
+		int tmp = kx;
+		kx = ky;
+		ky = tmp;
+	}
+
+	/* Calculate the shear constants. */
+	float inv_dir_z = 1.0f / dir_kz;
+	isect_precalc->Sx = IDX(dir, kx) * inv_dir_z;
+	isect_precalc->Sy = IDX(dir, ky) * inv_dir_z;
+	isect_precalc->Sz = inv_dir_z;
+
+	/* Store the dimensions. */
+	isect_precalc->kx = kx;
+	isect_precalc->ky = ky;
+	isect_precalc->kz = kz;
+}
+
+#undef IDX
+
 ccl_device_inline bool ray_triangle_intersect_uv(
         float3 ray_P, float3 ray_D, float ray_t,
         float3 v0, float3 v1, float3 v2,