Cycles: Fix crash with BVH8 on certain scenes

The crash was caused by BVH traversal stack being overflowed.

That overflow was caused by lots of false-positive intersections
for rays originating on a non-finite location.

Not sure why those rays will be existing in the first place,
this is to be investigated separately.

This commit moves pre-SSE4.1 check to a higher level function
and enables it for all miroarchitectures.

11 files changed, 28 insertions, 59 deletions

diff --git a/intern/cycles/kernel/bvh/bvh.h b/intern/cycles/kernel/bvh/bvh.h
index 2ad55d041bf..b3135573878 100644
--- a/intern/cycles/kernel/bvh/bvh.h
+++ b/intern/cycles/kernel/bvh/bvh.h
@@ -160,6 +160,19 @@ CCL_NAMESPACE_BEGIN
 #undef BVH_NAME_EVAL
 #undef BVH_FUNCTION_FULL_NAME
 
+ccl_device_inline bool scene_intersect_valid(const Ray *ray)
+{
+	/* NOTE: Due to some vectorization code  non-finite origin point might
+	 * cause lots of false-positive intersections which will overflow traversal
+	 * stack.
+	 * This code is a quick way to perform early output, to avoid crashes in
+	 * such cases.
+	 * From production scenes so far it seems it's enough to test first element
+	 * only.
+	 */
+	return finite(ray->P.x);
+}
+
 /* Note: ray is passed by value to work around a possible CUDA compiler bug. */
 ccl_device_intersect bool scene_intersect(KernelGlobals *kg,
                                           const Ray ray,
@@ -169,6 +182,9 @@ ccl_device_intersect bool scene_intersect(KernelGlobals *kg,
                                           float difl,
                                           float extmax)
 {
+	if (!scene_intersect_valid(&ray)) {
+		return false;
+	}
 #ifdef __OBJECT_MOTION__
 	if(kernel_data.bvh.have_motion) {
 #  ifdef __HAIR__
@@ -213,6 +229,9 @@ ccl_device_intersect bool scene_intersect_local(KernelGlobals *kg,
                                                 uint *lcg_state,
                                                 int max_hits)
 {
+	if (!scene_intersect_valid(&ray)) {
+		return false;
+	}
 #ifdef __OBJECT_MOTION__
 	if(kernel_data.bvh.have_motion) {
 		return bvh_intersect_local_motion(kg,
@@ -240,6 +259,9 @@ ccl_device_intersect bool scene_intersect_shadow_all(KernelGlobals *kg,
                                                      uint max_hits,
                                                      uint *num_hits)
 {
+	if (!scene_intersect_valid(ray)) {
+		return false;
+	}
 #  ifdef __OBJECT_MOTION__
 	if(kernel_data.bvh.have_motion) {
 #    ifdef __HAIR__
@@ -299,6 +321,9 @@ ccl_device_intersect bool scene_intersect_volume(KernelGlobals *kg,
                                                  Intersection *isect,
                                                  const uint visibility)
 {
+	if (!scene_intersect_valid(ray)) {
+		return false;
+	}
 #  ifdef __OBJECT_MOTION__
 	if(kernel_data.bvh.have_motion) {
 		return bvh_intersect_volume_motion(kg, ray, isect, visibility);
@@ -327,6 +352,9 @@ ccl_device_intersect uint scene_intersect_volume_all(KernelGlobals *kg,
                                                      const uint max_hits,
                                                      const uint visibility)
 {
+	if (!scene_intersect_valid(ray)) {
+		return false;
+	}
 #  ifdef __OBJECT_MOTION__
 	if(kernel_data.bvh.have_motion) {
 		return bvh_intersect_volume_all_motion(kg, ray, isect, max_hits, visibility);
diff --git a/intern/cycles/kernel/bvh/obvh_local.h b/intern/cycles/kernel/bvh/obvh_local.h
index 92143193a6a..eb24a607caa 100644
--- a/intern/cycles/kernel/bvh/obvh_local.h
+++ b/intern/cycles/kernel/bvh/obvh_local.h
@@ -73,12 +73,6 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(OBVH)(KernelGlobals *kg,
 		object = local_object;
 	}
 
-#ifndef __KERNEL_SSE41__
-	if(!isfinite(P.x)) {
-		return false;
-	}
-#endif
-
 	avxf tnear(0.0f), tfar(isect_t);
 #if BVH_FEATURE(BVH_HAIR)
 	avx3f dir4(avxf(dir.x), avxf(dir.y), avxf(dir.z));
diff --git a/intern/cycles/kernel/bvh/obvh_shadow_all.h b/intern/cycles/kernel/bvh/obvh_shadow_all.h
index 3e877065127..6ea93c5e7b2 100644
--- a/intern/cycles/kernel/bvh/obvh_shadow_all.h
+++ b/intern/cycles/kernel/bvh/obvh_shadow_all.h
@@ -66,12 +66,6 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(OBVH)(KernelGlobals *kg,
 	*num_hits = 0;
 	isect_array->t = tmax;
 
-#ifndef __KERNEL_SSE41__
-	if(!isfinite(P.x)) {
-		return false;
-	}
-#endif
-
 #if BVH_FEATURE(BVH_INSTANCING)
 	int num_hits_in_instance = 0;
 #endif
diff --git a/intern/cycles/kernel/bvh/obvh_traversal.h b/intern/cycles/kernel/bvh/obvh_traversal.h
index 98cd8f000ba..5010983057d 100644
--- a/intern/cycles/kernel/bvh/obvh_traversal.h
+++ b/intern/cycles/kernel/bvh/obvh_traversal.h
@@ -64,12 +64,6 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(OBVH)(KernelGlobals *kg,
 	Transform ob_itfm;
 #endif
 
-#ifndef __KERNEL_SSE41__
-	if(!isfinite(P.x)) {
-		return false;
-	}
-#endif
-
 	isect->t = ray->t;
 	isect->u = 0.0f;
 	isect->v = 0.0f;
diff --git a/intern/cycles/kernel/bvh/obvh_volume.h b/intern/cycles/kernel/bvh/obvh_volume.h
index da9ddbd4f24..80d09c59039 100644
--- a/intern/cycles/kernel/bvh/obvh_volume.h
+++ b/intern/cycles/kernel/bvh/obvh_volume.h
@@ -52,12 +52,6 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(OBVH)(KernelGlobals *kg,
 	Transform ob_itfm;
 #endif
 
-#ifndef __KERNEL_SSE41__
-	if(!isfinite(P.x)) {
-		return false;
-	}
-#endif
-
 	isect->t = ray->t;
 	isect->u = 0.0f;
 	isect->v = 0.0f;
diff --git a/intern/cycles/kernel/bvh/obvh_volume_all.h b/intern/cycles/kernel/bvh/obvh_volume_all.h
index a88573e6f86..87216127ddb 100644
--- a/intern/cycles/kernel/bvh/obvh_volume_all.h
+++ b/intern/cycles/kernel/bvh/obvh_volume_all.h
@@ -58,12 +58,6 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(OBVH)(KernelGlobals *kg,
 	uint num_hits = 0;
 	isect_array->t = tmax;
 
-#ifndef __KERNEL_SSE41__
-	if(!isfinite(P.x)) {
-		return 0;
-	}
-#endif
-
 #if BVH_FEATURE(BVH_INSTANCING)
 	int num_hits_in_instance = 0;
 #endif
diff --git a/intern/cycles/kernel/bvh/qbvh_local.h b/intern/cycles/kernel/bvh/qbvh_local.h
index ee3827de309..22d434a8737 100644
--- a/intern/cycles/kernel/bvh/qbvh_local.h
+++ b/intern/cycles/kernel/bvh/qbvh_local.h
@@ -82,12 +82,6 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 		object = local_object;
 	}
 
-#ifndef __KERNEL_SSE41__
-	if(!isfinite(P.x)) {
-		return false;
-	}
-#endif
-
 	ssef tnear(0.0f), tfar(isect_t);
 #if BVH_FEATURE(BVH_HAIR)
 	sse3f dir4(ssef(dir.x), ssef(dir.y), ssef(dir.z));
diff --git a/intern/cycles/kernel/bvh/qbvh_shadow_all.h b/intern/cycles/kernel/bvh/qbvh_shadow_all.h
index 46fd178aed6..3610bdd560b 100644
--- a/intern/cycles/kernel/bvh/qbvh_shadow_all.h
+++ b/intern/cycles/kernel/bvh/qbvh_shadow_all.h
@@ -66,11 +66,6 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 	*num_hits = 0;
 	isect_array->t = tmax;
 
-#ifndef __KERNEL_SSE41__
-	if(!isfinite(P.x)) {
-		return false;
-	}
-#endif
 
 #if BVH_FEATURE(BVH_INSTANCING)
 	int num_hits_in_instance = 0;
diff --git a/intern/cycles/kernel/bvh/qbvh_traversal.h b/intern/cycles/kernel/bvh/qbvh_traversal.h
index b36689b2a2e..ff675cab76b 100644
--- a/intern/cycles/kernel/bvh/qbvh_traversal.h
+++ b/intern/cycles/kernel/bvh/qbvh_traversal.h
@@ -71,12 +71,6 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 	Transform ob_itfm;
 #endif
 
-#ifndef __KERNEL_SSE41__
-	if(!isfinite(P.x)) {
-		return false;
-	}
-#endif
-
 	isect->t = ray->t;
 	isect->u = 0.0f;
 	isect->v = 0.0f;
diff --git a/intern/cycles/kernel/bvh/qbvh_volume.h b/intern/cycles/kernel/bvh/qbvh_volume.h
index 192ce009524..7ec264e5f78 100644
--- a/intern/cycles/kernel/bvh/qbvh_volume.h
+++ b/intern/cycles/kernel/bvh/qbvh_volume.h
@@ -58,12 +58,6 @@ ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 	Transform ob_itfm;
 #endif
 
-#ifndef __KERNEL_SSE41__
-	if(!isfinite(P.x)) {
-		return false;
-	}
-#endif
-
 	isect->t = ray->t;
 	isect->u = 0.0f;
 	isect->v = 0.0f;
diff --git a/intern/cycles/kernel/bvh/qbvh_volume_all.h b/intern/cycles/kernel/bvh/qbvh_volume_all.h
index 1e454e4d36b..dd603d79334 100644
--- a/intern/cycles/kernel/bvh/qbvh_volume_all.h
+++ b/intern/cycles/kernel/bvh/qbvh_volume_all.h
@@ -64,12 +64,6 @@ ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg,
 	uint num_hits = 0;
 	isect_array->t = tmax;
 
-#ifndef __KERNEL_SSE41__
-	if(!isfinite(P.x)) {
-		return 0;
-	}
-#endif
-
 #if BVH_FEATURE(BVH_INSTANCING)
 	int num_hits_in_instance = 0;
 #endif