diff options
Diffstat (limited to 'intern/cycles/kernel/bvh/bvh_volume.h')
-rw-r--r-- | intern/cycles/kernel/bvh/bvh_volume.h | 438 |
1 files changed, 211 insertions, 227 deletions
diff --git a/intern/cycles/kernel/bvh/bvh_volume.h b/intern/cycles/kernel/bvh/bvh_volume.h index b8257e3493e..c83b0d783f4 100644 --- a/intern/cycles/kernel/bvh/bvh_volume.h +++ b/intern/cycles/kernel/bvh/bvh_volume.h @@ -19,9 +19,9 @@ #ifdef __QBVH__ # include "kernel/bvh/qbvh_volume.h" -#ifdef __KERNEL_AVX2__ -# include "kernel/bvh/obvh_volume.h" -#endif +# ifdef __KERNEL_AVX2__ +# include "kernel/bvh/obvh_volume.h" +# endif #endif #if BVH_FEATURE(BVH_HAIR) @@ -43,267 +43,260 @@ ccl_device #else ccl_device_inline #endif -bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg, - const Ray *ray, - Intersection *isect, - const uint visibility) + bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg, + const Ray *ray, + Intersection *isect, + const uint visibility) { - /* todo: - * - test if pushing distance on the stack helps (for non shadow rays) - * - separate version for shadow rays - * - likely and unlikely for if() statements - * - test restrict attribute for pointers - */ - - /* traversal stack in CUDA thread-local memory */ - int traversal_stack[BVH_STACK_SIZE]; - traversal_stack[0] = ENTRYPOINT_SENTINEL; - - /* traversal variables in registers */ - int stack_ptr = 0; - int node_addr = kernel_data.bvh.root; - - /* ray parameters in registers */ - float3 P = ray->P; - float3 dir = bvh_clamp_direction(ray->D); - float3 idir = bvh_inverse_direction(dir); - int object = OBJECT_NONE; + /* todo: + * - test if pushing distance on the stack helps (for non shadow rays) + * - separate version for shadow rays + * - likely and unlikely for if() statements + * - test restrict attribute for pointers + */ + + /* traversal stack in CUDA thread-local memory */ + int traversal_stack[BVH_STACK_SIZE]; + traversal_stack[0] = ENTRYPOINT_SENTINEL; + + /* traversal variables in registers */ + int stack_ptr = 0; + int node_addr = kernel_data.bvh.root; + + /* ray parameters in registers */ + float3 P = ray->P; + float3 dir = bvh_clamp_direction(ray->D); + float3 idir = bvh_inverse_direction(dir); + int object = OBJECT_NONE; #if BVH_FEATURE(BVH_MOTION) - Transform ob_itfm; + Transform ob_itfm; #endif - isect->t = ray->t; - isect->u = 0.0f; - isect->v = 0.0f; - isect->prim = PRIM_NONE; - isect->object = OBJECT_NONE; + isect->t = ray->t; + isect->u = 0.0f; + isect->v = 0.0f; + isect->prim = PRIM_NONE; + isect->object = OBJECT_NONE; #if defined(__KERNEL_SSE2__) - const shuffle_swap_t shuf_identity = shuffle_swap_identity(); - const shuffle_swap_t shuf_swap = shuffle_swap_swap(); + const shuffle_swap_t shuf_identity = shuffle_swap_identity(); + const shuffle_swap_t shuf_swap = shuffle_swap_swap(); - const ssef pn = cast(ssei(0, 0, 0x80000000, 0x80000000)); - ssef Psplat[3], idirsplat[3]; + const ssef pn = cast(ssei(0, 0, 0x80000000, 0x80000000)); + ssef Psplat[3], idirsplat[3]; # if BVH_FEATURE(BVH_HAIR) - ssef tnear(0.0f), tfar(isect->t); + ssef tnear(0.0f), tfar(isect->t); # endif - shuffle_swap_t shufflexyz[3]; + shuffle_swap_t shufflexyz[3]; - Psplat[0] = ssef(P.x); - Psplat[1] = ssef(P.y); - Psplat[2] = ssef(P.z); + Psplat[0] = ssef(P.x); + Psplat[1] = ssef(P.y); + Psplat[2] = ssef(P.z); - ssef tsplat(0.0f, 0.0f, -isect->t, -isect->t); + ssef tsplat(0.0f, 0.0f, -isect->t, -isect->t); - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); + gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); #endif - /* traversal loop */ - do { - do { - /* traverse internal nodes */ - while(node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - int node_addr_child1, traverse_mask; - float dist[2]; - float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr+0); + /* traversal loop */ + do { + do { + /* traverse internal nodes */ + while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { + int node_addr_child1, traverse_mask; + float dist[2]; + float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); #if !defined(__KERNEL_SSE2__) - traverse_mask = NODE_INTERSECT(kg, - P, + traverse_mask = NODE_INTERSECT(kg, + P, # if BVH_FEATURE(BVH_HAIR) - dir, + dir, # endif - idir, - isect->t, - node_addr, - visibility, - dist); + idir, + isect->t, + node_addr, + visibility, + dist); #else // __KERNEL_SSE2__ - traverse_mask = NODE_INTERSECT(kg, - P, - dir, + traverse_mask = NODE_INTERSECT(kg, + P, + dir, # if BVH_FEATURE(BVH_HAIR) - tnear, - tfar, + tnear, + tfar, # endif - tsplat, - Psplat, - idirsplat, - shufflexyz, - node_addr, - visibility, - dist); + tsplat, + Psplat, + idirsplat, + shufflexyz, + node_addr, + visibility, + dist); #endif // __KERNEL_SSE2__ - node_addr = __float_as_int(cnodes.z); - node_addr_child1 = __float_as_int(cnodes.w); - - if(traverse_mask == 3) { - /* Both children were intersected, push the farther one. */ - bool is_closest_child1 = (dist[1] < dist[0]); - if(is_closest_child1) { - int tmp = node_addr; - node_addr = node_addr_child1; - node_addr_child1 = tmp; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_STACK_SIZE); - traversal_stack[stack_ptr] = node_addr_child1; - } - else { - /* One child was intersected. */ - if(traverse_mask == 2) { - node_addr = node_addr_child1; - } - else if(traverse_mask == 0) { - /* Neither child was intersected. */ - node_addr = traversal_stack[stack_ptr]; - --stack_ptr; - } - } - } - - /* if node is leaf, fetch triangle list */ - if(node_addr < 0) { - float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr-1)); - int prim_addr = __float_as_int(leaf.x); + node_addr = __float_as_int(cnodes.z); + node_addr_child1 = __float_as_int(cnodes.w); + + if (traverse_mask == 3) { + /* Both children were intersected, push the farther one. */ + bool is_closest_child1 = (dist[1] < dist[0]); + if (is_closest_child1) { + int tmp = node_addr; + node_addr = node_addr_child1; + node_addr_child1 = tmp; + } + + ++stack_ptr; + kernel_assert(stack_ptr < BVH_STACK_SIZE); + traversal_stack[stack_ptr] = node_addr_child1; + } + else { + /* One child was intersected. */ + if (traverse_mask == 2) { + node_addr = node_addr_child1; + } + else if (traverse_mask == 0) { + /* Neither child was intersected. */ + node_addr = traversal_stack[stack_ptr]; + --stack_ptr; + } + } + } + + /* if node is leaf, fetch triangle list */ + if (node_addr < 0) { + float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr - 1)); + int prim_addr = __float_as_int(leaf.x); #if BVH_FEATURE(BVH_INSTANCING) - if(prim_addr >= 0) { + if (prim_addr >= 0) { #endif - const int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - - /* pop */ - node_addr = traversal_stack[stack_ptr]; - --stack_ptr; - - /* primitive intersection */ - switch(type & PRIMITIVE_ALL) { - case PRIMITIVE_TRIANGLE: { - /* intersect ray against primitive */ - for(; prim_addr < prim_addr2; prim_addr++) { - kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type); - /* only primitives from volume object */ - uint tri_object = (object == OBJECT_NONE)? kernel_tex_fetch(__prim_object, prim_addr): object; - int object_flag = kernel_tex_fetch(__object_flag, tri_object); - if((object_flag & SD_OBJECT_HAS_VOLUME) == 0) { - continue; - } - triangle_intersect(kg, - isect, - P, - dir, - visibility, - object, - prim_addr); - } - break; - } + const int prim_addr2 = __float_as_int(leaf.y); + const uint type = __float_as_int(leaf.w); + + /* pop */ + node_addr = traversal_stack[stack_ptr]; + --stack_ptr; + + /* primitive intersection */ + switch (type & PRIMITIVE_ALL) { + case PRIMITIVE_TRIANGLE: { + /* intersect ray against primitive */ + for (; prim_addr < prim_addr2; prim_addr++) { + kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type); + /* only primitives from volume object */ + uint tri_object = (object == OBJECT_NONE) ? + kernel_tex_fetch(__prim_object, prim_addr) : + object; + int object_flag = kernel_tex_fetch(__object_flag, tri_object); + if ((object_flag & SD_OBJECT_HAS_VOLUME) == 0) { + continue; + } + triangle_intersect(kg, isect, P, dir, visibility, object, prim_addr); + } + break; + } #if BVH_FEATURE(BVH_MOTION) - case PRIMITIVE_MOTION_TRIANGLE: { - /* intersect ray against primitive */ - for(; prim_addr < prim_addr2; prim_addr++) { - kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type); - /* only primitives from volume object */ - uint tri_object = (object == OBJECT_NONE)? kernel_tex_fetch(__prim_object, prim_addr): object; - int object_flag = kernel_tex_fetch(__object_flag, tri_object); - if((object_flag & SD_OBJECT_HAS_VOLUME) == 0) { - continue; - } - motion_triangle_intersect(kg, - isect, - P, - dir, - ray->time, - visibility, - object, - prim_addr); - } - break; - } + case PRIMITIVE_MOTION_TRIANGLE: { + /* intersect ray against primitive */ + for (; prim_addr < prim_addr2; prim_addr++) { + kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type); + /* only primitives from volume object */ + uint tri_object = (object == OBJECT_NONE) ? + kernel_tex_fetch(__prim_object, prim_addr) : + object; + int object_flag = kernel_tex_fetch(__object_flag, tri_object); + if ((object_flag & SD_OBJECT_HAS_VOLUME) == 0) { + continue; + } + motion_triangle_intersect( + kg, isect, P, dir, ray->time, visibility, object, prim_addr); + } + break; + } #endif - default: { - break; - } - } - } + default: { + break; + } + } + } #if BVH_FEATURE(BVH_INSTANCING) - else { - /* instance push */ - object = kernel_tex_fetch(__prim_object, -prim_addr-1); - int object_flag = kernel_tex_fetch(__object_flag, object); - if(object_flag & SD_OBJECT_HAS_VOLUME) { + else { + /* instance push */ + object = kernel_tex_fetch(__prim_object, -prim_addr - 1); + int object_flag = kernel_tex_fetch(__object_flag, object); + if (object_flag & SD_OBJECT_HAS_VOLUME) { # if BVH_FEATURE(BVH_MOTION) - isect->t = bvh_instance_motion_push(kg, object, ray, &P, &dir, &idir, isect->t, &ob_itfm); + isect->t = bvh_instance_motion_push( + kg, object, ray, &P, &dir, &idir, isect->t, &ob_itfm); # else - isect->t = bvh_instance_push(kg, object, ray, &P, &dir, &idir, isect->t); + isect->t = bvh_instance_push(kg, object, ray, &P, &dir, &idir, isect->t); # endif # if defined(__KERNEL_SSE2__) - Psplat[0] = ssef(P.x); - Psplat[1] = ssef(P.y); - Psplat[2] = ssef(P.z); + Psplat[0] = ssef(P.x); + Psplat[1] = ssef(P.y); + Psplat[2] = ssef(P.z); - tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); + tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); # if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect->t); + tfar = ssef(isect->t); # endif - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); + gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); # endif - ++stack_ptr; - kernel_assert(stack_ptr < BVH_STACK_SIZE); - traversal_stack[stack_ptr] = ENTRYPOINT_SENTINEL; - - node_addr = kernel_tex_fetch(__object_node, object); - } - else { - /* pop */ - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr]; - --stack_ptr; - } - } - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while(node_addr != ENTRYPOINT_SENTINEL); + ++stack_ptr; + kernel_assert(stack_ptr < BVH_STACK_SIZE); + traversal_stack[stack_ptr] = ENTRYPOINT_SENTINEL; + + node_addr = kernel_tex_fetch(__object_node, object); + } + else { + /* pop */ + object = OBJECT_NONE; + node_addr = traversal_stack[stack_ptr]; + --stack_ptr; + } + } + } +#endif /* FEATURE(BVH_INSTANCING) */ + } while (node_addr != ENTRYPOINT_SENTINEL); #if BVH_FEATURE(BVH_INSTANCING) - if(stack_ptr >= 0) { - kernel_assert(object != OBJECT_NONE); + if (stack_ptr >= 0) { + kernel_assert(object != OBJECT_NONE); - /* instance pop */ + /* instance pop */ # if BVH_FEATURE(BVH_MOTION) - isect->t = bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, isect->t, &ob_itfm); + isect->t = bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, isect->t, &ob_itfm); # else - isect->t = bvh_instance_pop(kg, object, ray, &P, &dir, &idir, isect->t); + isect->t = bvh_instance_pop(kg, object, ray, &P, &dir, &idir, isect->t); # endif # if defined(__KERNEL_SSE2__) - Psplat[0] = ssef(P.x); - Psplat[1] = ssef(P.y); - Psplat[2] = ssef(P.z); + Psplat[0] = ssef(P.x); + Psplat[1] = ssef(P.y); + Psplat[2] = ssef(P.z); - tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); + tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); # if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect->t); + tfar = ssef(isect->t); # endif - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); + gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); # endif - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr]; - --stack_ptr; - } -#endif /* FEATURE(BVH_MOTION) */ - } while(node_addr != ENTRYPOINT_SENTINEL); + object = OBJECT_NONE; + node_addr = traversal_stack[stack_ptr]; + --stack_ptr; + } +#endif /* FEATURE(BVH_MOTION) */ + } while (node_addr != ENTRYPOINT_SENTINEL); - return (isect->prim != PRIM_NONE); + return (isect->prim != PRIM_NONE); } ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg, @@ -311,29 +304,20 @@ ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg, Intersection *isect, const uint visibility) { - switch(kernel_data.bvh.bvh_layout) { + switch (kernel_data.bvh.bvh_layout) { #ifdef __KERNEL_AVX2__ - case BVH_LAYOUT_BVH8: - return BVH_FUNCTION_FULL_NAME(OBVH)(kg, - ray, - isect, - visibility); + case BVH_LAYOUT_BVH8: + return BVH_FUNCTION_FULL_NAME(OBVH)(kg, ray, isect, visibility); #endif #ifdef __QBVH__ - case BVH_LAYOUT_BVH4: - return BVH_FUNCTION_FULL_NAME(QBVH)(kg, - ray, - isect, - visibility); + case BVH_LAYOUT_BVH4: + return BVH_FUNCTION_FULL_NAME(QBVH)(kg, ray, isect, visibility); #endif - case BVH_LAYOUT_BVH2: - return BVH_FUNCTION_FULL_NAME(BVH)(kg, - ray, - isect, - visibility); - } - kernel_assert(!"Should not happen"); - return false; + case BVH_LAYOUT_BVH2: + return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, isect, visibility); + } + kernel_assert(!"Should not happen"); + return false; } #undef BVH_FUNCTION_NAME |