diff options
author | Brecht Van Lommel <brecht@blender.org> | 2020-06-10 19:55:33 +0300 |
---|---|---|
committer | Brecht Van Lommel <brecht@blender.org> | 2020-06-22 14:28:01 +0300 |
commit | d1ef5146d72d40f97fdcbf28e96da49193c21dea (patch) | |
tree | 7a19a24bd6b809c7de72b4e2499d62b8740e639a /intern/cycles/kernel/bvh | |
parent | 1de0e13af619e405f351bf42924f819dc3a9bc44 (diff) |
Cycles: remove SIMD BVH optimizations, to be replaced by Embree
Ref T73778
Depends on D8011
Maniphest Tasks: T73778
Differential Revision: https://developer.blender.org/D8012
Diffstat (limited to 'intern/cycles/kernel/bvh')
20 files changed, 26 insertions, 5892 deletions
diff --git a/intern/cycles/kernel/bvh/bvh.h b/intern/cycles/kernel/bvh/bvh.h index bf48d3dd826..80b58f46329 100644 --- a/intern/cycles/kernel/bvh/bvh.h +++ b/intern/cycles/kernel/bvh/bvh.h @@ -35,14 +35,6 @@ CCL_NAMESPACE_BEGIN #ifndef __KERNEL_OPTIX__ -/* Common QBVH functions. */ -# ifdef __QBVH__ -# include "kernel/bvh/qbvh_nodes.h" -# ifdef __KERNEL_AVX2__ -# include "kernel/bvh/obvh_nodes.h" -# endif -# endif - /* Regular BVH traversal */ # include "kernel/bvh/bvh_nodes.h" diff --git a/intern/cycles/kernel/bvh/bvh_local.h b/intern/cycles/kernel/bvh/bvh_local.h index 7a069ef1108..4006c9c1632 100644 --- a/intern/cycles/kernel/bvh/bvh_local.h +++ b/intern/cycles/kernel/bvh/bvh_local.h @@ -17,13 +17,6 @@ * limitations under the License. */ -#ifdef __QBVH__ -# include "kernel/bvh/qbvh_local.h" -# ifdef __KERNEL_AVX2__ -# include "kernel/bvh/obvh_local.h" -# endif -#endif - #if BVH_FEATURE(BVH_HAIR) # define NODE_INTERSECT bvh_node_intersect #else @@ -88,26 +81,6 @@ ccl_device_inline object = local_object; } -#if defined(__KERNEL_SSE2__) - 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]; -# if BVH_FEATURE(BVH_HAIR) - ssef tnear(0.0f), tfar(isect_t); -# endif - shuffle_swap_t shufflexyz[3]; - - 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); - - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); -#endif - /* traversal loop */ do { do { @@ -117,33 +90,16 @@ ccl_device_inline float dist[2]; float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); -#if !defined(__KERNEL_SSE2__) traverse_mask = NODE_INTERSECT(kg, P, -# if BVH_FEATURE(BVH_HAIR) +#if BVH_FEATURE(BVH_HAIR) dir, -# endif +#endif idir, isect_t, node_addr, PATH_RAY_ALL_VISIBILITY, dist); -#else // __KERNEL_SSE2__ - traverse_mask = NODE_INTERSECT(kg, - P, - dir, -# if BVH_FEATURE(BVH_HAIR) - tnear, - tfar, -# endif - tsplat, - Psplat, - idirsplat, - shufflexyz, - node_addr, - PATH_RAY_ALL_VISIBILITY, - dist); -#endif // __KERNEL_SSE2__ node_addr = __float_as_int(cnodes.z); node_addr_child1 = __float_as_int(cnodes.w); @@ -247,20 +203,7 @@ ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg, uint *lcg_state, int max_hits) { - switch (kernel_data.bvh.bvh_layout) { -#ifdef __KERNEL_AVX2__ - case BVH_LAYOUT_BVH8: - return BVH_FUNCTION_FULL_NAME(OBVH)(kg, ray, local_isect, local_object, lcg_state, max_hits); -#endif -#ifdef __QBVH__ - case BVH_LAYOUT_BVH4: - return BVH_FUNCTION_FULL_NAME(QBVH)(kg, ray, local_isect, local_object, lcg_state, max_hits); -#endif - case BVH_LAYOUT_BVH2: - return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, local_isect, local_object, lcg_state, max_hits); - } - kernel_assert(!"Should not happen"); - return false; + return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, local_isect, local_object, lcg_state, max_hits); } #undef BVH_FUNCTION_NAME diff --git a/intern/cycles/kernel/bvh/bvh_nodes.h b/intern/cycles/kernel/bvh/bvh_nodes.h index db598d1c7fa..5367bdb633c 100644 --- a/intern/cycles/kernel/bvh/bvh_nodes.h +++ b/intern/cycles/kernel/bvh/bvh_nodes.h @@ -28,7 +28,6 @@ ccl_device_forceinline Transform bvh_unaligned_node_fetch_space(KernelGlobals *k return space; } -#if !defined(__KERNEL_SSE2__) ccl_device_forceinline int bvh_aligned_node_intersect(KernelGlobals *kg, const float3 P, const float3 idir, @@ -39,9 +38,9 @@ ccl_device_forceinline int bvh_aligned_node_intersect(KernelGlobals *kg, { /* fetch node data */ -# ifdef __VISIBILITY_FLAG__ +#ifdef __VISIBILITY_FLAG__ float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); -# endif +#endif float4 node0 = kernel_tex_fetch(__bvh_nodes, node_addr + 1); float4 node1 = kernel_tex_fetch(__bvh_nodes, node_addr + 2); float4 node2 = kernel_tex_fetch(__bvh_nodes, node_addr + 3); @@ -68,13 +67,13 @@ ccl_device_forceinline int bvh_aligned_node_intersect(KernelGlobals *kg, dist[0] = c0min; dist[1] = c1min; -# ifdef __VISIBILITY_FLAG__ +#ifdef __VISIBILITY_FLAG__ /* this visibility test gives a 5% performance hit, how to solve? */ return (((c0max >= c0min) && (__float_as_uint(cnodes.x) & visibility)) ? 1 : 0) | (((c1max >= c1min) && (__float_as_uint(cnodes.y) & visibility)) ? 2 : 0); -# else +#else return ((c0max >= c0min) ? 1 : 0) | ((c1max >= c1min) ? 2 : 0); -# endif +#endif } ccl_device_forceinline bool bvh_unaligned_node_intersect_child(KernelGlobals *kg, @@ -113,21 +112,21 @@ ccl_device_forceinline int bvh_unaligned_node_intersect(KernelGlobals *kg, float dist[2]) { int mask = 0; -# ifdef __VISIBILITY_FLAG__ +#ifdef __VISIBILITY_FLAG__ float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); -# endif +#endif if (bvh_unaligned_node_intersect_child(kg, P, dir, t, node_addr, 0, &dist[0])) { -# ifdef __VISIBILITY_FLAG__ +#ifdef __VISIBILITY_FLAG__ if ((__float_as_uint(cnodes.x) & visibility)) -# endif +#endif { mask |= 1; } } if (bvh_unaligned_node_intersect_child(kg, P, dir, t, node_addr, 1, &dist[1])) { -# ifdef __VISIBILITY_FLAG__ +#ifdef __VISIBILITY_FLAG__ if ((__float_as_uint(cnodes.y) & visibility)) -# endif +#endif { mask |= 2; } @@ -152,125 +151,3 @@ ccl_device_forceinline int bvh_node_intersect(KernelGlobals *kg, return bvh_aligned_node_intersect(kg, P, idir, t, node_addr, visibility, dist); } } - -#else /* !defined(__KERNEL_SSE2__) */ - -int ccl_device_forceinline bvh_aligned_node_intersect(KernelGlobals *kg, - const float3 &P, - const float3 &dir, - const ssef &tsplat, - const ssef Psplat[3], - const ssef idirsplat[3], - const shuffle_swap_t shufflexyz[3], - const int node_addr, - const uint visibility, - float dist[2]) -{ - /* Intersect two child bounding boxes, SSE3 version adapted from Embree */ - const ssef pn = cast(ssei(0, 0, 0x80000000, 0x80000000)); - - /* fetch node data */ - const ssef *bvh_nodes = (ssef *)kg->__bvh_nodes.data + node_addr; - - /* intersect ray against child nodes */ - const ssef tminmaxx = (shuffle_swap(bvh_nodes[1], shufflexyz[0]) - Psplat[0]) * idirsplat[0]; - const ssef tminmaxy = (shuffle_swap(bvh_nodes[2], shufflexyz[1]) - Psplat[1]) * idirsplat[1]; - const ssef tminmaxz = (shuffle_swap(bvh_nodes[3], shufflexyz[2]) - Psplat[2]) * idirsplat[2]; - - /* calculate { c0min, c1min, -c0max, -c1max} */ - ssef minmax = max(max(tminmaxx, tminmaxy), max(tminmaxz, tsplat)); - const ssef tminmax = minmax ^ pn; - const sseb lrhit = tminmax <= shuffle<2, 3, 0, 1>(tminmax); - - dist[0] = tminmax[0]; - dist[1] = tminmax[1]; - - int mask = movemask(lrhit); - -# ifdef __VISIBILITY_FLAG__ - /* this visibility test gives a 5% performance hit, how to solve? */ - float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - int cmask = (((mask & 1) && (__float_as_uint(cnodes.x) & visibility)) ? 1 : 0) | - (((mask & 2) && (__float_as_uint(cnodes.y) & visibility)) ? 2 : 0); - return cmask; -# else - return mask & 3; -# endif -} - -ccl_device_forceinline int bvh_unaligned_node_intersect(KernelGlobals *kg, - const float3 P, - const float3 dir, - const ssef &isect_near, - const ssef &isect_far, - const int node_addr, - const uint visibility, - float dist[2]) -{ - Transform space0 = bvh_unaligned_node_fetch_space(kg, node_addr, 0); - Transform space1 = bvh_unaligned_node_fetch_space(kg, node_addr, 1); - - float3 aligned_dir0 = transform_direction(&space0, dir), - aligned_dir1 = transform_direction(&space1, dir); - float3 aligned_P0 = transform_point(&space0, P), aligned_P1 = transform_point(&space1, P); - float3 nrdir0 = -bvh_inverse_direction(aligned_dir0), - nrdir1 = -bvh_inverse_direction(aligned_dir1); - - ssef lower_x = ssef(aligned_P0.x * nrdir0.x, aligned_P1.x * nrdir1.x, 0.0f, 0.0f), - lower_y = ssef(aligned_P0.y * nrdir0.y, aligned_P1.y * nrdir1.y, 0.0f, 0.0f), - lower_z = ssef(aligned_P0.z * nrdir0.z, aligned_P1.z * nrdir1.z, 0.0f, 0.0f); - - ssef upper_x = lower_x - ssef(nrdir0.x, nrdir1.x, 0.0f, 0.0f), - upper_y = lower_y - ssef(nrdir0.y, nrdir1.y, 0.0f, 0.0f), - upper_z = lower_z - ssef(nrdir0.z, nrdir1.z, 0.0f, 0.0f); - - ssef tnear_x = min(lower_x, upper_x); - ssef tnear_y = min(lower_y, upper_y); - ssef tnear_z = min(lower_z, upper_z); - ssef tfar_x = max(lower_x, upper_x); - ssef tfar_y = max(lower_y, upper_y); - ssef tfar_z = max(lower_z, upper_z); - - const ssef tnear = max4(isect_near, tnear_x, tnear_y, tnear_z); - const ssef tfar = min4(isect_far, tfar_x, tfar_y, tfar_z); - sseb vmask = tnear <= tfar; - dist[0] = tnear.f[0]; - dist[1] = tnear.f[1]; - - int mask = (int)movemask(vmask); - -# ifdef __VISIBILITY_FLAG__ - /* this visibility test gives a 5% performance hit, how to solve? */ - float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - int cmask = (((mask & 1) && (__float_as_uint(cnodes.x) & visibility)) ? 1 : 0) | - (((mask & 2) && (__float_as_uint(cnodes.y) & visibility)) ? 2 : 0); - return cmask; -# else - return mask & 3; -# endif -} - -ccl_device_forceinline int bvh_node_intersect(KernelGlobals *kg, - const float3 &P, - const float3 &dir, - const ssef &isect_near, - const ssef &isect_far, - const ssef &tsplat, - const ssef Psplat[3], - const ssef idirsplat[3], - const shuffle_swap_t shufflexyz[3], - const int node_addr, - const uint visibility, - float dist[2]) -{ - float4 node = kernel_tex_fetch(__bvh_nodes, node_addr); - if (__float_as_uint(node.x) & PATH_RAY_NODE_UNALIGNED) { - return bvh_unaligned_node_intersect( - kg, P, dir, isect_near, isect_far, node_addr, visibility, dist); - } - else { - return bvh_aligned_node_intersect( - kg, P, dir, tsplat, Psplat, idirsplat, shufflexyz, node_addr, visibility, dist); - } -} -#endif /* !defined(__KERNEL_SSE2__) */ diff --git a/intern/cycles/kernel/bvh/bvh_shadow_all.h b/intern/cycles/kernel/bvh/bvh_shadow_all.h index 03916bfdca9..12b88f159e2 100644 --- a/intern/cycles/kernel/bvh/bvh_shadow_all.h +++ b/intern/cycles/kernel/bvh/bvh_shadow_all.h @@ -17,13 +17,6 @@ * limitations under the License. */ -#ifdef __QBVH__ -# include "kernel/bvh/qbvh_shadow_all.h" -# ifdef __KERNEL_AVX2__ -# include "kernel/bvh/obvh_shadow_all.h" -# endif -#endif - #if BVH_FEATURE(BVH_HAIR) # define NODE_INTERSECT bvh_node_intersect #else @@ -80,26 +73,6 @@ ccl_device_inline *num_hits = 0; isect_array->t = tmax; -#if defined(__KERNEL_SSE2__) - 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]; -# if BVH_FEATURE(BVH_HAIR) - ssef tnear(0.0f), tfar(isect_t); -# endif - shuffle_swap_t shufflexyz[3]; - - 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); - - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); -#endif /* __KERNEL_SSE2__ */ - /* traversal loop */ do { do { @@ -109,33 +82,16 @@ ccl_device_inline float dist[2]; float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); -#if !defined(__KERNEL_SSE2__) traverse_mask = NODE_INTERSECT(kg, P, -# if BVH_FEATURE(BVH_HAIR) +#if BVH_FEATURE(BVH_HAIR) dir, -# endif +#endif idir, isect_t, node_addr, visibility, dist); -#else // __KERNEL_SSE2__ - traverse_mask = NODE_INTERSECT(kg, - P, - dir, -# if BVH_FEATURE(BVH_HAIR) - tnear, - tfar, -# endif - 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); @@ -272,18 +228,6 @@ ccl_device_inline num_hits_in_instance = 0; isect_array->t = isect_t; -# if defined(__KERNEL_SSE2__) - 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); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect_t); -# endif - 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; @@ -322,18 +266,6 @@ ccl_device_inline isect_t = tmax; isect_array->t = isect_t; -# if defined(__KERNEL_SSE2__) - 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); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect_t); -# endif - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); -# endif - object = OBJECT_NONE; node_addr = traversal_stack[stack_ptr]; --stack_ptr; @@ -350,20 +282,7 @@ ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg, const uint max_hits, uint *num_hits) { - switch (kernel_data.bvh.bvh_layout) { -#ifdef __KERNEL_AVX2__ - case BVH_LAYOUT_BVH8: - return BVH_FUNCTION_FULL_NAME(OBVH)(kg, ray, isect_array, visibility, max_hits, num_hits); -#endif -#ifdef __QBVH__ - case BVH_LAYOUT_BVH4: - return BVH_FUNCTION_FULL_NAME(QBVH)(kg, ray, isect_array, visibility, max_hits, num_hits); -#endif - case BVH_LAYOUT_BVH2: - return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, isect_array, visibility, max_hits, num_hits); - } - kernel_assert(!"Should not happen"); - return false; + return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, isect_array, visibility, max_hits, num_hits); } #undef BVH_FUNCTION_NAME diff --git a/intern/cycles/kernel/bvh/bvh_traversal.h b/intern/cycles/kernel/bvh/bvh_traversal.h index 2a4677928c5..e6236c93caa 100644 --- a/intern/cycles/kernel/bvh/bvh_traversal.h +++ b/intern/cycles/kernel/bvh/bvh_traversal.h @@ -17,13 +17,6 @@ * limitations under the License. */ -#ifdef __QBVH__ -# include "kernel/bvh/qbvh_traversal.h" -#endif -#ifdef __KERNEL_AVX2__ -# include "kernel/bvh/obvh_traversal.h" -#endif - #if BVH_FEATURE(BVH_HAIR) # define NODE_INTERSECT bvh_node_intersect #else @@ -76,26 +69,6 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg, BVH_DEBUG_INIT(); -#if defined(__KERNEL_SSE2__) - 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]; -# if BVH_FEATURE(BVH_HAIR) - ssef tnear(0.0f), tfar(isect->t); -# endif - shuffle_swap_t shufflexyz[3]; - - 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); - - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); -#endif - /* traversal loop */ do { do { @@ -105,37 +78,18 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg, float dist[2]; float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); -#if !defined(__KERNEL_SSE2__) { traverse_mask = NODE_INTERSECT(kg, P, -# if BVH_FEATURE(BVH_HAIR) +#if BVH_FEATURE(BVH_HAIR) dir, -# endif +#endif idir, isect->t, node_addr, visibility, dist); } -#else // __KERNEL_SSE2__ - { - traverse_mask = NODE_INTERSECT(kg, - P, - dir, -# if BVH_FEATURE(BVH_HAIR) - tnear, - tfar, -# endif - 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); @@ -188,17 +142,8 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg, kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type); if (triangle_intersect(kg, isect, P, dir, visibility, object, prim_addr)) { /* shadow ray early termination */ -#if defined(__KERNEL_SSE2__) if (visibility & PATH_RAY_SHADOW_OPAQUE) return true; - tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect->t); -# endif -#else - if (visibility & PATH_RAY_SHADOW_OPAQUE) - return true; -#endif } } break; @@ -211,17 +156,8 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg, if (motion_triangle_intersect( kg, isect, P, dir, ray->time, visibility, object, prim_addr)) { /* shadow ray early termination */ -# if defined(__KERNEL_SSE2__) - if (visibility & PATH_RAY_SHADOW_OPAQUE) - return true; - tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect->t); -# endif -# else if (visibility & PATH_RAY_SHADOW_OPAQUE) return true; -# endif } } break; @@ -238,17 +174,8 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg, kg, isect, P, dir, visibility, object, prim_addr, ray->time, curve_type); if (hit) { /* shadow ray early termination */ -# if defined(__KERNEL_SSE2__) - if (visibility & PATH_RAY_SHADOW_OPAQUE) - return true; - tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect->t); -# endif -# else if (visibility & PATH_RAY_SHADOW_OPAQUE) return true; -# endif } } break; @@ -267,19 +194,6 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg, 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); - - tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect->t); -# endif - - 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; @@ -301,19 +215,6 @@ ccl_device_noinline bool BVH_FUNCTION_FULL_NAME(BVH)(KernelGlobals *kg, 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); - - tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect->t); -# endif - - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); -# endif - object = OBJECT_NONE; node_addr = traversal_stack[stack_ptr]; --stack_ptr; @@ -328,20 +229,7 @@ ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg, Intersection *isect, const uint visibility) { - switch (kernel_data.bvh.bvh_layout) { -#ifdef __KERNEL_AVX2__ - 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); -#endif /* __QBVH__ */ - case BVH_LAYOUT_BVH2: - return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, isect, visibility); - } - kernel_assert(!"Should not happen"); - return false; + return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, isect, visibility); } #undef BVH_FUNCTION_NAME diff --git a/intern/cycles/kernel/bvh/bvh_types.h b/intern/cycles/kernel/bvh/bvh_types.h index a7bc52d4435..b173568266b 100644 --- a/intern/cycles/kernel/bvh/bvh_types.h +++ b/intern/cycles/kernel/bvh/bvh_types.h @@ -31,8 +31,6 @@ CCL_NAMESPACE_BEGIN /* 64 object BVH + 64 mesh BVH + 64 object node splitting */ #define BVH_STACK_SIZE 192 -#define BVH_QSTACK_SIZE 384 -#define BVH_OSTACK_SIZE 768 /* BVH intersection function variations */ #define BVH_MOTION 1 diff --git a/intern/cycles/kernel/bvh/bvh_volume.h b/intern/cycles/kernel/bvh/bvh_volume.h index d8b0bbccd22..1f2ea47269b 100644 --- a/intern/cycles/kernel/bvh/bvh_volume.h +++ b/intern/cycles/kernel/bvh/bvh_volume.h @@ -17,13 +17,6 @@ * limitations under the License. */ -#ifdef __QBVH__ -# include "kernel/bvh/qbvh_volume.h" -# ifdef __KERNEL_AVX2__ -# include "kernel/bvh/obvh_volume.h" -# endif -#endif - #if BVH_FEATURE(BVH_HAIR) # define NODE_INTERSECT bvh_node_intersect #else @@ -78,26 +71,6 @@ ccl_device_inline 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 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); -# endif - shuffle_swap_t shufflexyz[3]; - - 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); - - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); -#endif - /* traversal loop */ do { do { @@ -107,33 +80,16 @@ ccl_device_inline float dist[2]; float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); -#if !defined(__KERNEL_SSE2__) traverse_mask = NODE_INTERSECT(kg, P, -# if BVH_FEATURE(BVH_HAIR) +#if BVH_FEATURE(BVH_HAIR) dir, -# endif +#endif idir, isect->t, node_addr, visibility, dist); -#else // __KERNEL_SSE2__ - traverse_mask = NODE_INTERSECT(kg, - P, - dir, -# if BVH_FEATURE(BVH_HAIR) - tnear, - tfar, -# endif - 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); @@ -231,19 +187,6 @@ ccl_device_inline 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); - - tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect->t); -# endif - - 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; @@ -270,19 +213,6 @@ ccl_device_inline 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); - - tsplat = ssef(0.0f, 0.0f, -isect->t, -isect->t); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect->t); -# endif - - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); -# endif - object = OBJECT_NONE; node_addr = traversal_stack[stack_ptr]; --stack_ptr; @@ -297,20 +227,7 @@ ccl_device_inline bool BVH_FUNCTION_NAME(KernelGlobals *kg, Intersection *isect, const uint visibility) { - switch (kernel_data.bvh.bvh_layout) { -#ifdef __KERNEL_AVX2__ - 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); -#endif - case BVH_LAYOUT_BVH2: - return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, isect, visibility); - } - kernel_assert(!"Should not happen"); - return false; + return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, isect, visibility); } #undef BVH_FUNCTION_NAME diff --git a/intern/cycles/kernel/bvh/bvh_volume_all.h b/intern/cycles/kernel/bvh/bvh_volume_all.h index 7b1834c7c6f..a8664cc4331 100644 --- a/intern/cycles/kernel/bvh/bvh_volume_all.h +++ b/intern/cycles/kernel/bvh/bvh_volume_all.h @@ -17,13 +17,6 @@ * limitations under the License. */ -#ifdef __QBVH__ -# include "kernel/bvh/qbvh_volume_all.h" -# ifdef __KERNEL_AVX2__ -# include "kernel/bvh/obvh_volume_all.h" -# endif -#endif - #if BVH_FEATURE(BVH_HAIR) # define NODE_INTERSECT bvh_node_intersect #else @@ -80,26 +73,6 @@ ccl_device_inline uint num_hits = 0; isect_array->t = tmax; -#if defined(__KERNEL_SSE2__) - 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]; -# if BVH_FEATURE(BVH_HAIR) - ssef tnear(0.0f), tfar(isect_t); -# endif - shuffle_swap_t shufflexyz[3]; - - 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); - - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); -#endif /* __KERNEL_SSE2__ */ - /* traversal loop */ do { do { @@ -109,33 +82,16 @@ ccl_device_inline float dist[2]; float4 cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); -#if !defined(__KERNEL_SSE2__) traverse_mask = NODE_INTERSECT(kg, P, -# if BVH_FEATURE(BVH_HAIR) +#if BVH_FEATURE(BVH_HAIR) dir, -# endif +#endif idir, isect_t, node_addr, visibility, dist); -#else // __KERNEL_SSE2__ - traverse_mask = NODE_INTERSECT(kg, - P, - dir, -# if BVH_FEATURE(BVH_HAIR) - tnear, - tfar, -# endif - 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); @@ -281,19 +237,6 @@ ccl_device_inline num_hits_in_instance = 0; isect_array->t = isect_t; -# if defined(__KERNEL_SSE2__) - 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); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect_t); -# endif - - 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; @@ -337,19 +280,6 @@ ccl_device_inline isect_t = tmax; isect_array->t = isect_t; -# if defined(__KERNEL_SSE2__) - 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); -# if BVH_FEATURE(BVH_HAIR) - tfar = ssef(isect_t); -# endif - - gen_idirsplat_swap(pn, shuf_identity, shuf_swap, idir, idirsplat, shufflexyz); -# endif - object = OBJECT_NONE; node_addr = traversal_stack[stack_ptr]; --stack_ptr; @@ -365,20 +295,7 @@ ccl_device_inline uint BVH_FUNCTION_NAME(KernelGlobals *kg, const uint max_hits, const uint visibility) { - switch (kernel_data.bvh.bvh_layout) { -#ifdef __KERNEL_AVX2__ - case BVH_LAYOUT_BVH8: - return BVH_FUNCTION_FULL_NAME(OBVH)(kg, ray, isect_array, max_hits, visibility); -#endif -#ifdef __QBVH__ - case BVH_LAYOUT_BVH4: - return BVH_FUNCTION_FULL_NAME(QBVH)(kg, ray, isect_array, max_hits, visibility); -#endif - case BVH_LAYOUT_BVH2: - return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, isect_array, max_hits, visibility); - } - kernel_assert(!"Should not happen"); - return 0; + return BVH_FUNCTION_FULL_NAME(BVH)(kg, ray, isect_array, max_hits, visibility); } #undef BVH_FUNCTION_NAME diff --git a/intern/cycles/kernel/bvh/obvh_local.h b/intern/cycles/kernel/bvh/obvh_local.h deleted file mode 100644 index e6bb548bc5b..00000000000 --- a/intern/cycles/kernel/bvh/obvh_local.h +++ /dev/null @@ -1,398 +0,0 @@ -/* - * Copyright 2011-2013 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -/* This is a template BVH traversal function for subsurface scattering, where - * various features can be enabled/disabled. This way we can compile optimized - * versions for each case without new features slowing things down. - * - * BVH_MOTION: motion blur rendering - */ - -#if BVH_FEATURE(BVH_HAIR) -# define NODE_INTERSECT obvh_node_intersect -#else -# define NODE_INTERSECT obvh_aligned_node_intersect -#endif - -ccl_device bool BVH_FUNCTION_FULL_NAME(OBVH)(KernelGlobals *kg, - const Ray *ray, - LocalIntersection *local_isect, - int local_object, - uint *lcg_state, - int max_hits) -{ - /* Traversal stack in CUDA thread-local memory. */ - OBVHStackItem traversal_stack[BVH_OSTACK_SIZE]; - traversal_stack[0].addr = ENTRYPOINT_SENTINEL; - - /* Traversal variables in registers. */ - int stack_ptr = 0; - int node_addr = kernel_tex_fetch(__object_node, local_object); - - /* 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; - float isect_t = ray->t; - - if (local_isect != NULL) { - local_isect->num_hits = 0; - } - kernel_assert((local_isect == NULL) == (max_hits == 0)); - - const int object_flag = kernel_tex_fetch(__object_flag, local_object); - if (!(object_flag & SD_OBJECT_TRANSFORM_APPLIED)) { -#if BVH_FEATURE(BVH_MOTION) - Transform ob_itfm; - isect_t = bvh_instance_motion_push(kg, local_object, ray, &P, &dir, &idir, isect_t, &ob_itfm); -#else - isect_t = bvh_instance_push(kg, local_object, ray, &P, &dir, &idir, isect_t); -#endif - object = local_object; - } - - avxf tnear(0.0f), tfar(isect_t); -#if BVH_FEATURE(BVH_HAIR) - avx3f dir4(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -#endif - avx3f idir4(avxf(idir.x), avxf(idir.y), avxf(idir.z)); - -#ifdef __KERNEL_AVX2__ - float3 P_idir = P * idir; - avx3f P_idir4(P_idir.x, P_idir.y, P_idir.z); -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - avx3f org4(avxf(P.x), avxf(P.y), avxf(P.z)); -#endif - - /* Offsets to select the side that becomes the lower or upper bound. */ - int near_x, near_y, near_z; - int far_x, far_y, far_z; - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - - /* Traversal loop. */ - do { - do { - /* Traverse internal nodes. */ - while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - avxf dist; - int child_mask = NODE_INTERSECT(kg, - tnear, - tfar, -#ifdef __KERNEL_AVX2__ - P_idir4, -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4, -#endif -#if BVH_FEATURE(BVH_HAIR) - dir4, -#endif - idir4, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - &dist); - - if (child_mask != 0) { - float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - avxf cnodes; -#if BVH_FEATURE(BVH_HAIR) - if (__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) { - cnodes = kernel_tex_fetch_avxf(__bvh_nodes, node_addr + 26); - } - else -#endif - { - cnodes = kernel_tex_fetch_avxf(__bvh_nodes, node_addr + 14); - } - - /* One child is hit, continue with that child. */ - int r = __bscf(child_mask); - if (child_mask == 0) { - node_addr = __float_as_int(cnodes[r]); - continue; - } - - /* Two children are hit, push far child, and continue with - * closer child. - */ - int c0 = __float_as_int(cnodes[r]); - float d0 = ((float *)&dist)[r]; - r = __bscf(child_mask); - int c1 = __float_as_int(cnodes[r]); - float d1 = ((float *)&dist)[r]; - if (child_mask == 0) { - if (d1 < d0) { - node_addr = c1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - continue; - } - else { - node_addr = c0; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - continue; - } - } - - /* Here starts the slow path for 3 or 4 hit children. We push - * all nodes onto the stack to sort them there. - */ - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - - /* Three children are hit, push all onto stack and sort 3 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c2 = __float_as_int(cnodes[r]); - float d2 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Four children are hit, push all onto stack and sort 4 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c3 = __float_as_int(cnodes[r]); - float d3 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - - /* Five children are hit, push all onto stack and sort 5 - * stack items, continue with closest child - */ - r = __bscf(child_mask); - int c4 = __float_as_int(cnodes[r]); - float d4 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - /* Six children are hit, push all onto stack and sort 6 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c5 = __float_as_int(cnodes[r]); - float d5 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c5; - traversal_stack[stack_ptr].dist = d5; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c5; - traversal_stack[stack_ptr].dist = d5; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - - /* Seven children are hit, push all onto stack and sort 7 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c6 = __float_as_int(cnodes[r]); - float d6 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c6; - traversal_stack[stack_ptr].dist = d6; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5], - &traversal_stack[stack_ptr - 6]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - /* Eight children are hit, push all onto stack and sort 8 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c7 = __float_as_int(cnodes[r]); - float d7 = ((float *)&dist)[r]; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c7; - traversal_stack[stack_ptr].dist = d7; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c6; - traversal_stack[stack_ptr].dist = d6; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5], - &traversal_stack[stack_ptr - 6], - &traversal_stack[stack_ptr - 7]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - node_addr = traversal_stack[stack_ptr].addr; - --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); - - int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --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); - if (triangle_intersect_local(kg, - local_isect, - P, - dir, - object, - local_object, - prim_addr, - isect_t, - lcg_state, - max_hits)) { - return true; - } - } - 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); - if (motion_triangle_intersect_local(kg, - local_isect, - P, - dir, - ray->time, - object, - local_object, - prim_addr, - isect_t, - lcg_state, - max_hits)) { - return true; - } - } - break; - } -#endif - default: - break; - } - } - } while (node_addr != ENTRYPOINT_SENTINEL); - } while (node_addr != ENTRYPOINT_SENTINEL); - return false; -} - -#undef NODE_INTERSECT diff --git a/intern/cycles/kernel/bvh/obvh_nodes.h b/intern/cycles/kernel/bvh/obvh_nodes.h deleted file mode 100644 index e5c935b75ed..00000000000 --- a/intern/cycles/kernel/bvh/obvh_nodes.h +++ /dev/null @@ -1,410 +0,0 @@ -/* - * Copyright 2011-2014, Blender Foundation. - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - * - * Aligned nodes intersection AVX code is adopted from Embree, - */ - -struct OBVHStackItem { - int addr; - float dist; -}; - -ccl_device_inline void obvh_near_far_idx_calc(const float3 &idir, - int *ccl_restrict near_x, - int *ccl_restrict near_y, - int *ccl_restrict near_z, - int *ccl_restrict far_x, - int *ccl_restrict far_y, - int *ccl_restrict far_z) - -{ -#ifdef __KERNEL_SSE__ - *near_x = 0; - *far_x = 1; - *near_y = 2; - *far_y = 3; - *near_z = 4; - *far_z = 5; - - const size_t mask = movemask(ssef(idir.m128)); - - const int mask_x = mask & 1; - const int mask_y = (mask & 2) >> 1; - const int mask_z = (mask & 4) >> 2; - - *near_x += mask_x; - *far_x -= mask_x; - *near_y += mask_y; - *far_y -= mask_y; - *near_z += mask_z; - *far_z -= mask_z; -#else - if (idir.x >= 0.0f) { - *near_x = 0; - *far_x = 1; - } - else { - *near_x = 1; - *far_x = 0; - } - if (idir.y >= 0.0f) { - *near_y = 2; - *far_y = 3; - } - else { - *near_y = 3; - *far_y = 2; - } - if (idir.z >= 0.0f) { - *near_z = 4; - *far_z = 5; - } - else { - *near_z = 5; - *far_z = 4; - } -#endif -} - -ccl_device_inline void obvh_item_swap(OBVHStackItem *ccl_restrict a, OBVHStackItem *ccl_restrict b) -{ - OBVHStackItem tmp = *a; - *a = *b; - *b = tmp; -} - -ccl_device_inline void obvh_stack_sort(OBVHStackItem *ccl_restrict s1, - OBVHStackItem *ccl_restrict s2, - OBVHStackItem *ccl_restrict s3) -{ - if (s2->dist < s1->dist) { - obvh_item_swap(s2, s1); - } - if (s3->dist < s2->dist) { - obvh_item_swap(s3, s2); - } - if (s2->dist < s1->dist) { - obvh_item_swap(s2, s1); - } -} - -ccl_device_inline void obvh_stack_sort(OBVHStackItem *ccl_restrict s1, - OBVHStackItem *ccl_restrict s2, - OBVHStackItem *ccl_restrict s3, - OBVHStackItem *ccl_restrict s4) -{ - if (s2->dist < s1->dist) { - obvh_item_swap(s2, s1); - } - if (s4->dist < s3->dist) { - obvh_item_swap(s4, s3); - } - if (s3->dist < s1->dist) { - obvh_item_swap(s3, s1); - } - if (s4->dist < s2->dist) { - obvh_item_swap(s4, s2); - } - if (s3->dist < s2->dist) { - obvh_item_swap(s3, s2); - } -} - -ccl_device_inline void obvh_stack_sort(OBVHStackItem *ccl_restrict s1, - OBVHStackItem *ccl_restrict s2, - OBVHStackItem *ccl_restrict s3, - OBVHStackItem *ccl_restrict s4, - OBVHStackItem *ccl_restrict s5) -{ - obvh_stack_sort(s1, s2, s3, s4); - if (s5->dist < s4->dist) { - obvh_item_swap(s4, s5); - if (s4->dist < s3->dist) { - obvh_item_swap(s3, s4); - if (s3->dist < s2->dist) { - obvh_item_swap(s2, s3); - if (s2->dist < s1->dist) { - obvh_item_swap(s1, s2); - } - } - } - } -} - -ccl_device_inline void obvh_stack_sort(OBVHStackItem *ccl_restrict s1, - OBVHStackItem *ccl_restrict s2, - OBVHStackItem *ccl_restrict s3, - OBVHStackItem *ccl_restrict s4, - OBVHStackItem *ccl_restrict s5, - OBVHStackItem *ccl_restrict s6) -{ - obvh_stack_sort(s1, s2, s3, s4, s5); - if (s6->dist < s5->dist) { - obvh_item_swap(s5, s6); - if (s5->dist < s4->dist) { - obvh_item_swap(s4, s5); - if (s4->dist < s3->dist) { - obvh_item_swap(s3, s4); - if (s3->dist < s2->dist) { - obvh_item_swap(s2, s3); - if (s2->dist < s1->dist) { - obvh_item_swap(s1, s2); - } - } - } - } - } -} - -ccl_device_inline void obvh_stack_sort(OBVHStackItem *ccl_restrict s1, - OBVHStackItem *ccl_restrict s2, - OBVHStackItem *ccl_restrict s3, - OBVHStackItem *ccl_restrict s4, - OBVHStackItem *ccl_restrict s5, - OBVHStackItem *ccl_restrict s6, - OBVHStackItem *ccl_restrict s7) -{ - obvh_stack_sort(s1, s2, s3, s4, s5, s6); - if (s7->dist < s6->dist) { - obvh_item_swap(s6, s7); - if (s6->dist < s5->dist) { - obvh_item_swap(s5, s6); - if (s5->dist < s4->dist) { - obvh_item_swap(s4, s5); - if (s4->dist < s3->dist) { - obvh_item_swap(s3, s4); - if (s3->dist < s2->dist) { - obvh_item_swap(s2, s3); - if (s2->dist < s1->dist) { - obvh_item_swap(s1, s2); - } - } - } - } - } - } -} - -ccl_device_inline void obvh_stack_sort(OBVHStackItem *ccl_restrict s1, - OBVHStackItem *ccl_restrict s2, - OBVHStackItem *ccl_restrict s3, - OBVHStackItem *ccl_restrict s4, - OBVHStackItem *ccl_restrict s5, - OBVHStackItem *ccl_restrict s6, - OBVHStackItem *ccl_restrict s7, - OBVHStackItem *ccl_restrict s8) -{ - obvh_stack_sort(s1, s2, s3, s4, s5, s6, s7); - if (s8->dist < s7->dist) { - obvh_item_swap(s7, s8); - if (s7->dist < s6->dist) { - obvh_item_swap(s6, s7); - if (s6->dist < s5->dist) { - obvh_item_swap(s5, s6); - if (s5->dist < s4->dist) { - obvh_item_swap(s4, s5); - if (s4->dist < s3->dist) { - obvh_item_swap(s3, s4); - if (s3->dist < s2->dist) { - obvh_item_swap(s2, s3); - if (s2->dist < s1->dist) { - obvh_item_swap(s1, s2); - } - } - } - } - } - } - } -} - -/* Axis-aligned nodes intersection */ - -ccl_device_inline int obvh_aligned_node_intersect(KernelGlobals *ccl_restrict kg, - const avxf &isect_near, - const avxf &isect_far, -#ifdef __KERNEL_AVX2__ - const avx3f &org_idir, -#else - const avx3f &org, -#endif - const avx3f &idir, - const int near_x, - const int near_y, - const int near_z, - const int far_x, - const int far_y, - const int far_z, - const int node_addr, - avxf *ccl_restrict dist) -{ - const int offset = node_addr + 2; -#ifdef __KERNEL_AVX2__ - const avxf tnear_x = msub( - kernel_tex_fetch_avxf(__bvh_nodes, offset + near_x * 2), idir.x, org_idir.x); - const avxf tnear_y = msub( - kernel_tex_fetch_avxf(__bvh_nodes, offset + near_y * 2), idir.y, org_idir.y); - const avxf tnear_z = msub( - kernel_tex_fetch_avxf(__bvh_nodes, offset + near_z * 2), idir.z, org_idir.z); - const avxf tfar_x = msub( - kernel_tex_fetch_avxf(__bvh_nodes, offset + far_x * 2), idir.x, org_idir.x); - const avxf tfar_y = msub( - kernel_tex_fetch_avxf(__bvh_nodes, offset + far_y * 2), idir.y, org_idir.y); - const avxf tfar_z = msub( - kernel_tex_fetch_avxf(__bvh_nodes, offset + far_z * 2), idir.z, org_idir.z); - - const avxf tnear = max4(tnear_x, tnear_y, tnear_z, isect_near); - const avxf tfar = min4(tfar_x, tfar_y, tfar_z, isect_far); - const avxb vmask = tnear <= tfar; - int mask = (int)movemask(vmask); - *dist = tnear; - return mask; -#else - return 0; -#endif -} - -/* Unaligned nodes intersection */ - -ccl_device_inline int obvh_unaligned_node_intersect(KernelGlobals *ccl_restrict kg, - const avxf &isect_near, - const avxf &isect_far, -#ifdef __KERNEL_AVX2__ - const avx3f &org_idir, -#endif - const avx3f &org, - const avx3f &dir, - const avx3f &idir, - const int near_x, - const int near_y, - const int near_z, - const int far_x, - const int far_y, - const int far_z, - const int node_addr, - avxf *ccl_restrict dist) -{ - const int offset = node_addr; - const avxf tfm_x_x = kernel_tex_fetch_avxf(__bvh_nodes, offset + 2); - const avxf tfm_x_y = kernel_tex_fetch_avxf(__bvh_nodes, offset + 4); - const avxf tfm_x_z = kernel_tex_fetch_avxf(__bvh_nodes, offset + 6); - - const avxf tfm_y_x = kernel_tex_fetch_avxf(__bvh_nodes, offset + 8); - const avxf tfm_y_y = kernel_tex_fetch_avxf(__bvh_nodes, offset + 10); - const avxf tfm_y_z = kernel_tex_fetch_avxf(__bvh_nodes, offset + 12); - - const avxf tfm_z_x = kernel_tex_fetch_avxf(__bvh_nodes, offset + 14); - const avxf tfm_z_y = kernel_tex_fetch_avxf(__bvh_nodes, offset + 16); - const avxf tfm_z_z = kernel_tex_fetch_avxf(__bvh_nodes, offset + 18); - - const avxf tfm_t_x = kernel_tex_fetch_avxf(__bvh_nodes, offset + 20); - const avxf tfm_t_y = kernel_tex_fetch_avxf(__bvh_nodes, offset + 22); - const avxf tfm_t_z = kernel_tex_fetch_avxf(__bvh_nodes, offset + 24); - - const avxf aligned_dir_x = dir.x * tfm_x_x + dir.y * tfm_x_y + dir.z * tfm_x_z, - aligned_dir_y = dir.x * tfm_y_x + dir.y * tfm_y_y + dir.z * tfm_y_z, - aligned_dir_z = dir.x * tfm_z_x + dir.y * tfm_z_y + dir.z * tfm_z_z; - - const avxf aligned_P_x = org.x * tfm_x_x + org.y * tfm_x_y + org.z * tfm_x_z + tfm_t_x, - aligned_P_y = org.x * tfm_y_x + org.y * tfm_y_y + org.z * tfm_y_z + tfm_t_y, - aligned_P_z = org.x * tfm_z_x + org.y * tfm_z_y + org.z * tfm_z_z + tfm_t_z; - - const avxf neg_one(-1.0f); - const avxf nrdir_x = neg_one / aligned_dir_x, nrdir_y = neg_one / aligned_dir_y, - nrdir_z = neg_one / aligned_dir_z; - - const avxf tlower_x = aligned_P_x * nrdir_x, tlower_y = aligned_P_y * nrdir_y, - tlower_z = aligned_P_z * nrdir_z; - - const avxf tupper_x = tlower_x - nrdir_x, tupper_y = tlower_y - nrdir_y, - tupper_z = tlower_z - nrdir_z; - - const avxf tnear_x = min(tlower_x, tupper_x); - const avxf tnear_y = min(tlower_y, tupper_y); - const avxf tnear_z = min(tlower_z, tupper_z); - const avxf tfar_x = max(tlower_x, tupper_x); - const avxf tfar_y = max(tlower_y, tupper_y); - const avxf tfar_z = max(tlower_z, tupper_z); - const avxf tnear = max4(isect_near, tnear_x, tnear_y, tnear_z); - const avxf tfar = min4(isect_far, tfar_x, tfar_y, tfar_z); - const avxb vmask = tnear <= tfar; - *dist = tnear; - return movemask(vmask); -} - -/* Intersectors wrappers. - * - * They'll check node type and call appropriate intersection code. - */ - -ccl_device_inline int obvh_node_intersect(KernelGlobals *ccl_restrict kg, - const avxf &isect_near, - const avxf &isect_far, -#ifdef __KERNEL_AVX2__ - const avx3f &org_idir, -#endif - const avx3f &org, - const avx3f &dir, - const avx3f &idir, - const int near_x, - const int near_y, - const int near_z, - const int far_x, - const int far_y, - const int far_z, - const int node_addr, - avxf *ccl_restrict dist) -{ - const int offset = node_addr; - const float4 node = kernel_tex_fetch(__bvh_nodes, offset); - if (__float_as_uint(node.x) & PATH_RAY_NODE_UNALIGNED) { - return obvh_unaligned_node_intersect(kg, - isect_near, - isect_far, -#ifdef __KERNEL_AVX2__ - org_idir, -#endif - org, - dir, - idir, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - dist); - } - else { - return obvh_aligned_node_intersect(kg, - isect_near, - isect_far, -#ifdef __KERNEL_AVX2__ - org_idir, -#else - org, -#endif - idir, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - dist); - } -} diff --git a/intern/cycles/kernel/bvh/obvh_shadow_all.h b/intern/cycles/kernel/bvh/obvh_shadow_all.h deleted file mode 100644 index ff82d5105a4..00000000000 --- a/intern/cycles/kernel/bvh/obvh_shadow_all.h +++ /dev/null @@ -1,651 +0,0 @@ -/* - * Copyright 2011-2013 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -/* This is a template BVH traversal function, where various features can be - * enabled/disabled. This way we can compile optimized versions for each case - * without new features slowing things down. - * - * BVH_INSTANCING: object instancing - * BVH_HAIR: hair curve rendering - * BVH_MOTION: motion blur rendering - */ - -#if BVH_FEATURE(BVH_HAIR) -# define NODE_INTERSECT obvh_node_intersect -#else -# define NODE_INTERSECT obvh_aligned_node_intersect -#endif - -ccl_device bool BVH_FUNCTION_FULL_NAME(OBVH)(KernelGlobals *kg, - const Ray *ray, - Intersection *isect_array, - const int skip_object, - const uint max_hits, - uint *num_hits) -{ - /* TODO(sergey): - * - Test if pushing distance on the stack helps. - * - Likely and unlikely for if() statements. - * - Test restrict attribute for pointers. - */ - - /* Traversal stack in CUDA thread-local memory. */ - OBVHStackItem traversal_stack[BVH_OSTACK_SIZE]; - traversal_stack[0].addr = ENTRYPOINT_SENTINEL; - - /* Traversal variables in registers. */ - int stack_ptr = 0; - int node_addr = kernel_data.bvh.root; - - /* Ray parameters in registers. */ - const float tmax = ray->t; - float3 P = ray->P; - float3 dir = bvh_clamp_direction(ray->D); - float3 idir = bvh_inverse_direction(dir); - int object = OBJECT_NONE; - float isect_t = tmax; - -#if BVH_FEATURE(BVH_MOTION) - Transform ob_itfm; -#endif - - *num_hits = 0; - isect_array->t = tmax; - -#if BVH_FEATURE(BVH_INSTANCING) - int num_hits_in_instance = 0; -#endif - - avxf tnear(0.0f), tfar(isect_t); -#if BVH_FEATURE(BVH_HAIR) - avx3f dir4(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -#endif - avx3f idir4(avxf(idir.x), avxf(idir.y), avxf(idir.z)); - -#ifdef __KERNEL_AVX2__ - float3 P_idir = P * idir; - avx3f P_idir4(P_idir.x, P_idir.y, P_idir.z); -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - avx3f org4(avxf(P.x), avxf(P.y), avxf(P.z)); -#endif - - /* Offsets to select the side that becomes the lower or upper bound. */ - int near_x, near_y, near_z; - int far_x, far_y, far_z; - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - - /* Traversal loop. */ - do { - do { - /* Traverse internal nodes. */ - while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - (void)inodes; - - if (false -#ifdef __VISIBILITY_FLAG__ - || ((__float_as_uint(inodes.x) & PATH_RAY_SHADOW) == 0) -#endif -#if BVH_FEATURE(BVH_MOTION) - || UNLIKELY(ray->time < inodes.y) || UNLIKELY(ray->time > inodes.z) -#endif - ) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - avxf dist; - int child_mask = NODE_INTERSECT(kg, - tnear, - tfar, -#ifdef __KERNEL_AVX2__ - P_idir4, -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - //#if !defined(__KERNEL_AVX2__) - org4, -#endif -#if BVH_FEATURE(BVH_HAIR) - dir4, -#endif - idir4, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - &dist); - - if (child_mask != 0) { - avxf cnodes; -#if BVH_FEATURE(BVH_HAIR) - if (__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) { - cnodes = kernel_tex_fetch_avxf(__bvh_nodes, node_addr + 26); - } - else -#endif - { - cnodes = kernel_tex_fetch_avxf(__bvh_nodes, node_addr + 14); - } - - /* One child is hit, continue with that child. */ - int r = __bscf(child_mask); - if (child_mask == 0) { - node_addr = __float_as_int(cnodes[r]); - continue; - } - - /* Two children are hit, push far child, and continue with - * closer child. - */ - int c0 = __float_as_int(cnodes[r]); - float d0 = ((float *)&dist)[r]; - r = __bscf(child_mask); - int c1 = __float_as_int(cnodes[r]); - float d1 = ((float *)&dist)[r]; - if (child_mask == 0) { - if (d1 < d0) { - node_addr = c1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - continue; - } - else { - node_addr = c0; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - continue; - } - } - - /* Here starts the slow path for 3 or 4 hit children. We push - * all nodes onto the stack to sort them there. - */ - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - - /* Three children are hit, push all onto stack and sort 3 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c2 = __float_as_int(cnodes[r]); - float d2 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Four children are hit, push all onto stack and sort 4 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c3 = __float_as_int(cnodes[r]); - float d3 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - - /* Five children are hit, push all onto stack and sort 5 - * stack items, continue with closest child - */ - r = __bscf(child_mask); - int c4 = __float_as_int(cnodes[r]); - float d4 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Six children are hit, push all onto stack and sort 6 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c5 = __float_as_int(cnodes[r]); - float d5 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c5; - traversal_stack[stack_ptr].dist = d5; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c5; - traversal_stack[stack_ptr].dist = d5; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - - /* Seven children are hit, push all onto stack and sort 7 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c6 = __float_as_int(cnodes[r]); - float d6 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c6; - traversal_stack[stack_ptr].dist = d6; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5], - &traversal_stack[stack_ptr - 6]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Eight children are hit, push all onto stack and sort 8 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c7 = __float_as_int(cnodes[r]); - float d7 = ((float *)&dist)[r]; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c7; - traversal_stack[stack_ptr].dist = d7; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c6; - traversal_stack[stack_ptr].dist = d6; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5], - &traversal_stack[stack_ptr - 6], - &traversal_stack[stack_ptr - 7]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } - - /* If node is leaf, fetch triangle list. */ - if (node_addr < 0) { - float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr - 1)); -#ifdef __VISIBILITY_FLAG__ - if ((__float_as_uint(leaf.z) & PATH_RAY_SHADOW) == 0) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } -#endif - - int prim_addr = __float_as_int(leaf.x); - -#if BVH_FEATURE(BVH_INSTANCING) - if (prim_addr >= 0) { -#endif - int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - const uint p_type = type & PRIMITIVE_ALL; - - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - - /* Primitive intersection. */ - if (p_type == PRIMITIVE_TRIANGLE) { - int prim_count = prim_addr2 - prim_addr; - if (prim_count < 3) { - while (prim_addr < prim_addr2) { - kernel_assert((kernel_tex_fetch(__prim_type, prim_addr) & PRIMITIVE_ALL) == - p_type); - int hit = triangle_intersect( - kg, isect_array, P, dir, PATH_RAY_SHADOW, object, prim_addr); - /* Shadow ray early termination. */ - if (hit) { - /* detect if this surface has a shader with transparent shadows */ - - /* todo: optimize so primitive visibility flag indicates if - * the primitive has a transparent shadow shader? */ - int prim = kernel_tex_fetch(__prim_index, isect_array->prim); - int shader = 0; - -#ifdef __HAIR__ - if (kernel_tex_fetch(__prim_type, isect_array->prim) & PRIMITIVE_ALL_TRIANGLE) -#endif - { - shader = kernel_tex_fetch(__tri_shader, prim); - } -#ifdef __HAIR__ - else { - float4 str = kernel_tex_fetch(__curves, prim); - shader = __float_as_int(str.z); - } -#endif - int flag = kernel_tex_fetch(__shaders, (shader & SHADER_MASK)).flags; - - /* if no transparent shadows, all light is blocked */ - if (!(flag & SD_HAS_TRANSPARENT_SHADOW)) { - return true; - } - /* if maximum number of hits reached, block all light */ - else if (*num_hits == max_hits) { - return true; - } - - /* move on to next entry in intersections array */ - isect_array++; - (*num_hits)++; -#if BVH_FEATURE(BVH_INSTANCING) - num_hits_in_instance++; -#endif - - isect_array->t = isect_t; - } - - prim_addr++; - } // while - } - else { - kernel_assert((kernel_tex_fetch(__prim_type, (prim_addr)) & PRIMITIVE_ALL) == - p_type); - -#if BVH_FEATURE(BVH_INSTANCING) - int *nhiptr = &num_hits_in_instance; -#else - int nhi = 0; - int *nhiptr = &nhi; -#endif - - int result = triangle_intersect8(kg, - &isect_array, - P, - dir, - PATH_RAY_SHADOW, - object, - prim_addr, - prim_count, - num_hits, - max_hits, - nhiptr, - isect_t); - if (result == 2) { - return true; - } - } // prim_count - } // PRIMITIVE_TRIANGLE - else { - while (prim_addr < prim_addr2) { - kernel_assert((kernel_tex_fetch(__prim_type, prim_addr) & PRIMITIVE_ALL) == p_type); - -#ifdef __SHADOW_TRICKS__ - uint tri_object = (object == OBJECT_NONE) ? - kernel_tex_fetch(__prim_object, prim_addr) : - object; - if (tri_object == skip_object) { - ++prim_addr; - continue; - } -#endif - - bool hit; - - /* todo: specialized intersect functions which don't fill in - * isect unless needed and check SD_HAS_TRANSPARENT_SHADOW? - * might give a few % performance improvement */ - - switch (p_type) { - -#if BVH_FEATURE(BVH_MOTION) - case PRIMITIVE_MOTION_TRIANGLE: { - hit = motion_triangle_intersect( - kg, isect_array, P, dir, ray->time, PATH_RAY_SHADOW, object, prim_addr); - break; - } -#endif -#if BVH_FEATURE(BVH_HAIR) - case PRIMITIVE_CURVE: - case PRIMITIVE_MOTION_CURVE: { - const uint curve_type = kernel_tex_fetch(__prim_type, prim_addr); - hit = curve_intersect(kg, - isect_array, - P, - dir, - PATH_RAY_SHADOW, - object, - prim_addr, - ray->time, - curve_type); - break; - } -#endif - default: { - hit = false; - break; - } - } - - /* Shadow ray early termination. */ - if (hit) { - /* detect if this surface has a shader with transparent shadows */ - - /* todo: optimize so primitive visibility flag indicates if - * the primitive has a transparent shadow shader? */ - int prim = kernel_tex_fetch(__prim_index, isect_array->prim); - int shader = 0; - -#ifdef __HAIR__ - if (kernel_tex_fetch(__prim_type, isect_array->prim) & PRIMITIVE_ALL_TRIANGLE) -#endif - { - shader = kernel_tex_fetch(__tri_shader, prim); - } -#ifdef __HAIR__ - else { - float4 str = kernel_tex_fetch(__curves, prim); - shader = __float_as_int(str.z); - } -#endif - int flag = kernel_tex_fetch(__shaders, (shader & SHADER_MASK)).flags; - - /* if no transparent shadows, all light is blocked */ - if (!(flag & SD_HAS_TRANSPARENT_SHADOW)) { - return true; - } - /* if maximum number of hits reached, block all light */ - else if (*num_hits == max_hits) { - return true; - } - - /* move on to next entry in intersections array */ - isect_array++; - (*num_hits)++; -#if BVH_FEATURE(BVH_INSTANCING) - num_hits_in_instance++; -#endif - - isect_array->t = isect_t; - } - - prim_addr++; - } // while prim - } - } -#if BVH_FEATURE(BVH_INSTANCING) - else { - /* Instance push. */ - object = kernel_tex_fetch(__prim_object, -prim_addr - 1); - -# if BVH_FEATURE(BVH_MOTION) - 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); -# endif - - num_hits_in_instance = 0; - isect_array->t = isect_t; - - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = avxf(isect_t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = avx3f(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -# endif - idir4 = avx3f(avxf(idir.x), avxf(idir.y), avxf(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = avx3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = avx3f(avxf(P.x), avxf(P.y), avxf(P.z)); -# endif - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = ENTRYPOINT_SENTINEL; - - node_addr = kernel_tex_fetch(__object_node, object); - } - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - -#if BVH_FEATURE(BVH_INSTANCING) - if (stack_ptr >= 0) { - kernel_assert(object != OBJECT_NONE); - - /* Instance pop. */ - if (num_hits_in_instance) { - float t_fac; -# if BVH_FEATURE(BVH_MOTION) - bvh_instance_motion_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac, &ob_itfm); -# else - bvh_instance_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac); -# endif - /* Scale isect->t to adjust for instancing. */ - for (int i = 0; i < num_hits_in_instance; i++) { - (isect_array - i - 1)->t *= t_fac; - } - } - else { -# if BVH_FEATURE(BVH_MOTION) - bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX, &ob_itfm); -# else - bvh_instance_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX); -# endif - } - - isect_t = tmax; - isect_array->t = isect_t; - - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = avxf(isect_t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = avx3f(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -# endif - idir4 = avx3f(avxf(idir.x), avxf(idir.y), avxf(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = avx3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = avx3f(avxf(P.x), avxf(P.y), avxf(P.z)); -# endif - - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - - return false; -} - -#undef NODE_INTERSECT diff --git a/intern/cycles/kernel/bvh/obvh_traversal.h b/intern/cycles/kernel/bvh/obvh_traversal.h deleted file mode 100644 index 3f1e03693c3..00000000000 --- a/intern/cycles/kernel/bvh/obvh_traversal.h +++ /dev/null @@ -1,550 +0,0 @@ -/* - * Copyright 2011-2013 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -/* This is a template BVH traversal function, where various features can be - * enabled/disabled. This way we can compile optimized versions for each case - * without new features slowing things down. - * - * BVH_INSTANCING: object instancing - * BVH_HAIR: hair curve rendering - * BVH_MOTION: motion blur rendering - */ - -#if BVH_FEATURE(BVH_HAIR) -# define NODE_INTERSECT obvh_node_intersect -#else -# define NODE_INTERSECT obvh_aligned_node_intersect -#endif - -ccl_device bool BVH_FUNCTION_FULL_NAME(OBVH)(KernelGlobals *kg, - const Ray *ray, - Intersection *isect, - const uint visibility) -{ - /* Traversal stack in CUDA thread-local memory. */ - OBVHStackItem traversal_stack[BVH_OSTACK_SIZE]; - traversal_stack[0].addr = ENTRYPOINT_SENTINEL; - traversal_stack[0].dist = -FLT_MAX; - - /* Traversal variables in registers. */ - int stack_ptr = 0; - int node_addr = kernel_data.bvh.root; - float node_dist = -FLT_MAX; - - /* 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; -#endif - - isect->t = ray->t; - isect->u = 0.0f; - isect->v = 0.0f; - isect->prim = PRIM_NONE; - isect->object = OBJECT_NONE; - - BVH_DEBUG_INIT(); - avxf tnear(0.0f), tfar(ray->t); -#if BVH_FEATURE(BVH_HAIR) - avx3f dir4(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -#endif - avx3f idir4(avxf(idir.x), avxf(idir.y), avxf(idir.z)); - -#ifdef __KERNEL_AVX2__ - float3 P_idir = P * idir; - avx3f P_idir4 = avx3f(P_idir.x, P_idir.y, P_idir.z); -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - avx3f org4 = avx3f(avxf(P.x), avxf(P.y), avxf(P.z)); -#endif - - /* Offsets to select the side that becomes the lower or upper bound. */ - int near_x, near_y, near_z; - int far_x, far_y, far_z; - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - /* Traversal loop. */ - do { - do { - /* Traverse internal nodes. */ - while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - (void)inodes; - - if (UNLIKELY(node_dist > isect->t) -#if BVH_FEATURE(BVH_MOTION) - || UNLIKELY(ray->time < inodes.y) || UNLIKELY(ray->time > inodes.z) -#endif -#ifdef __VISIBILITY_FLAG__ - || (__float_as_uint(inodes.x) & visibility) == 0 -#endif - ) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - - int child_mask; - avxf dist; - - BVH_DEBUG_NEXT_NODE(); - - { - child_mask = NODE_INTERSECT(kg, - tnear, - tfar, -#ifdef __KERNEL_AVX2__ - P_idir4, -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4, -#endif -#if BVH_FEATURE(BVH_HAIR) - dir4, -#endif - idir4, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - &dist); - } - - if (child_mask != 0) { - avxf cnodes; - /* TODO(sergey): Investigate whether moving cnodes upwards - * gives a speedup (will be different cache pattern but will - * avoid extra check here). - */ -#if BVH_FEATURE(BVH_HAIR) - if (__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) { - cnodes = kernel_tex_fetch_avxf(__bvh_nodes, node_addr + 26); - } - else -#endif - { - cnodes = kernel_tex_fetch_avxf(__bvh_nodes, node_addr + 14); - } - - /* One child is hit, continue with that child. */ - int r = __bscf(child_mask); - float d0 = ((float *)&dist)[r]; - if (child_mask == 0) { - node_addr = __float_as_int(cnodes[r]); - node_dist = d0; - continue; - } - - /* Two children are hit, push far child, and continue with - * closer child. - */ - int c0 = __float_as_int(cnodes[r]); - r = __bscf(child_mask); - int c1 = __float_as_int(cnodes[r]); - float d1 = ((float *)&dist)[r]; - if (child_mask == 0) { - if (d1 < d0) { - node_addr = c1; - node_dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - continue; - } - else { - node_addr = c0; - node_dist = d0; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - continue; - } - } - - /* Here starts the slow path for 3 or 4 hit children. We push - * all nodes onto the stack to sort them there. - */ - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - - /* Three children are hit, push all onto stack and sort 3 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c2 = __float_as_int(cnodes[r]); - float d2 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2]); - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - - /* Four children are hit, push all onto stack and sort 4 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c3 = __float_as_int(cnodes[r]); - float d3 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3]); - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - - /* Five children are hit, push all onto stack and sort 5 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c4 = __float_as_int(cnodes[r]); - float d4 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4]); - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - - /* Six children are hit, push all onto stack and sort 6 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c5 = __float_as_int(cnodes[r]); - float d5 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c5; - traversal_stack[stack_ptr].dist = d5; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5]); - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c5; - traversal_stack[stack_ptr].dist = d5; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - - /* Seven children are hit, push all onto stack and sort 7 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c6 = __float_as_int(cnodes[r]); - float d6 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c6; - traversal_stack[stack_ptr].dist = d6; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5], - &traversal_stack[stack_ptr - 6]); - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - - /* Eight children are hit, push all onto stack and sort 8 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c7 = __float_as_int(cnodes[r]); - float d7 = ((float *)&dist)[r]; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c7; - traversal_stack[stack_ptr].dist = d7; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c6; - traversal_stack[stack_ptr].dist = d6; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5], - &traversal_stack[stack_ptr - 6], - &traversal_stack[stack_ptr - 7]); - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - } - - /* If node is leaf, fetch triangle list. */ - if (node_addr < 0) { - float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr - 1)); - -#ifdef __VISIBILITY_FLAG__ - if (UNLIKELY((node_dist > isect->t) || ((__float_as_uint(leaf.z) & visibility) == 0))) -#else - if (UNLIKELY((node_dist > isect->t))) -#endif - { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - int prim_addr = __float_as_int(leaf.x); - -#if BVH_FEATURE(BVH_INSTANCING) - if (prim_addr >= 0) { -#endif - int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - - /* Primitive intersection. */ - switch (type & PRIMITIVE_ALL) { - case PRIMITIVE_TRIANGLE: { - int prim_count = prim_addr2 - prim_addr; - if (prim_count < 3) { - for (; prim_addr < prim_addr2; prim_addr++) { - BVH_DEBUG_NEXT_INTERSECTION(); - kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type); - if (triangle_intersect(kg, isect, P, dir, visibility, object, prim_addr)) { - tfar = avxf(isect->t); - /* Shadow ray early termination. */ - if (visibility == PATH_RAY_SHADOW_OPAQUE) { - return true; - } - } - } // for - } - else { - kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type); - if (triangle_intersect8(kg, - &isect, - P, - dir, - visibility, - object, - prim_addr, - prim_count, - 0, - 0, - NULL, - 0.0f)) { - tfar = avxf(isect->t); - if (visibility == PATH_RAY_SHADOW_OPAQUE) { - return true; - } - } - } // prim count - break; - } -#if BVH_FEATURE(BVH_MOTION) - case PRIMITIVE_MOTION_TRIANGLE: { - for (; prim_addr < prim_addr2; prim_addr++) { - BVH_DEBUG_NEXT_INTERSECTION(); - kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type); - if (motion_triangle_intersect( - kg, isect, P, dir, ray->time, visibility, object, prim_addr)) { - tfar = avxf(isect->t); - /* Shadow ray early termination. */ - if (visibility == PATH_RAY_SHADOW_OPAQUE) { - return true; - } - } - } - break; - } -#endif /* BVH_FEATURE(BVH_MOTION) */ -#if BVH_FEATURE(BVH_HAIR) - case PRIMITIVE_CURVE: - case PRIMITIVE_MOTION_CURVE: { - for (; prim_addr < prim_addr2; prim_addr++) { - BVH_DEBUG_NEXT_INTERSECTION(); - const uint curve_type = kernel_tex_fetch(__prim_type, prim_addr); - kernel_assert((curve_type & PRIMITIVE_ALL) == (type & PRIMITIVE_ALL)); - bool hit = curve_intersect( - kg, isect, P, dir, visibility, object, prim_addr, ray->time, curve_type); - if (hit) { - tfar = avxf(isect->t); - /* Shadow ray early termination. */ - if (visibility == PATH_RAY_SHADOW_OPAQUE) { - return true; - } - } - } - break; - } -#endif /* BVH_FEATURE(BVH_HAIR) */ - } - } -#if BVH_FEATURE(BVH_INSTANCING) - else { - /* Instance push. */ - object = kernel_tex_fetch(__prim_object, -prim_addr - 1); - -# if BVH_FEATURE(BVH_MOTION) - qbvh_instance_motion_push( - kg, object, ray, &P, &dir, &idir, &isect->t, &node_dist, &ob_itfm); -# else - qbvh_instance_push(kg, object, ray, &P, &dir, &idir, &isect->t, &node_dist); -# endif - - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = avxf(isect->t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = avx3f(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -# endif - idir4 = avx3f(avxf(idir.x), avxf(idir.y), avxf(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = avx3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = avx3f(avxf(P.x), avxf(P.y), avxf(P.z)); -# endif - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = ENTRYPOINT_SENTINEL; - traversal_stack[stack_ptr].dist = -FLT_MAX; - - node_addr = kernel_tex_fetch(__object_node, object); - - BVH_DEBUG_NEXT_INSTANCE(); - } - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - -#if BVH_FEATURE(BVH_INSTANCING) - if (stack_ptr >= 0) { - kernel_assert(object != OBJECT_NONE); - - /* Instance pop. */ -# if BVH_FEATURE(BVH_MOTION) - 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); -# endif - - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = avxf(isect->t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = avx3f(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -# endif - idir4 = avx3f(avxf(idir.x), avxf(idir.y), avxf(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = avx3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = avx3f(avxf(P.x), avxf(P.y), avxf(P.z)); -# endif - - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - - return (isect->prim != PRIM_NONE); -} - -#undef NODE_INTERSECT diff --git a/intern/cycles/kernel/bvh/obvh_volume.h b/intern/cycles/kernel/bvh/obvh_volume.h deleted file mode 100644 index fb41ae783ab..00000000000 --- a/intern/cycles/kernel/bvh/obvh_volume.h +++ /dev/null @@ -1,480 +0,0 @@ -/* - * Copyright 2011-2013 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -/* This is a template BVH traversal function for volumes, where - * various features can be enabled/disabled. This way we can compile optimized - * versions for each case without new features slowing things down. - * - * BVH_INSTANCING: object instancing - * BVH_MOTION: motion blur rendering - */ - -#if BVH_FEATURE(BVH_HAIR) -# define NODE_INTERSECT obvh_node_intersect -#else -# define NODE_INTERSECT obvh_aligned_node_intersect -#endif - -ccl_device bool BVH_FUNCTION_FULL_NAME(OBVH)(KernelGlobals *kg, - const Ray *ray, - Intersection *isect, - const uint visibility) -{ - /* Traversal stack in CUDA thread-local memory. */ - OBVHStackItem traversal_stack[BVH_OSTACK_SIZE]; - traversal_stack[0].addr = 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; -#endif - - isect->t = ray->t; - isect->u = 0.0f; - isect->v = 0.0f; - isect->prim = PRIM_NONE; - isect->object = OBJECT_NONE; - - avxf tnear(0.0f), tfar(ray->t); -#if BVH_FEATURE(BVH_HAIR) - avx3f dir4(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -#endif - avx3f idir4(avxf(idir.x), avxf(idir.y), avxf(idir.z)); - -#ifdef __KERNEL_AVX2__ - float3 P_idir = P * idir; - avx3f P_idir4(P_idir.x, P_idir.y, P_idir.z); -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - avx3f org4(avxf(P.x), avxf(P.y), avxf(P.z)); -#endif - - /* Offsets to select the side that becomes the lower or upper bound. */ - int near_x, near_y, near_z; - int far_x, far_y, far_z; - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - - /* Traversal loop. */ - do { - do { - /* Traverse internal nodes. */ - while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - -#ifdef __VISIBILITY_FLAG__ - if ((__float_as_uint(inodes.x) & visibility) == 0) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } -#endif - - avxf dist; - int child_mask = NODE_INTERSECT(kg, - tnear, - tfar, -#ifdef __KERNEL_AVX2__ - P_idir4, -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4, -#endif -#if BVH_FEATURE(BVH_HAIR) - dir4, -#endif - idir4, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - &dist); - - if (child_mask != 0) { - avxf cnodes; -#if BVH_FEATURE(BVH_HAIR) - if (__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) { - cnodes = kernel_tex_fetch_avxf(__bvh_nodes, node_addr + 26); - } - else -#endif - { - cnodes = kernel_tex_fetch_avxf(__bvh_nodes, node_addr + 14); - } - - /* One child is hit, continue with that child. */ - int r = __bscf(child_mask); - if (child_mask == 0) { - node_addr = __float_as_int(cnodes[r]); - continue; - } - - /* Two children are hit, push far child, and continue with - * closer child. - */ - int c0 = __float_as_int(cnodes[r]); - float d0 = ((float *)&dist)[r]; - r = __bscf(child_mask); - int c1 = __float_as_int(cnodes[r]); - float d1 = ((float *)&dist)[r]; - if (child_mask == 0) { - if (d1 < d0) { - node_addr = c1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - continue; - } - else { - node_addr = c0; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - continue; - } - } - - /* Here starts the slow path for 3 or 4 hit children. We push - * all nodes onto the stack to sort them there. - */ - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - - /* Three children are hit, push all onto stack and sort 3 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c2 = __float_as_int(cnodes[r]); - float d2 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Four children are hit, push all onto stack and sort 4 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c3 = __float_as_int(cnodes[r]); - float d3 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - - /* Five children are hit, push all onto stack and sort 5 - * stack items, continue with closest child - */ - r = __bscf(child_mask); - int c4 = __float_as_int(cnodes[r]); - float d4 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Six children are hit, push all onto stack and sort 6 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c5 = __float_as_int(cnodes[r]); - float d5 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c5; - traversal_stack[stack_ptr].dist = d5; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c5; - traversal_stack[stack_ptr].dist = d5; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - - /* Seven children are hit, push all onto stack and sort 7 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c6 = __float_as_int(cnodes[r]); - float d6 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c6; - traversal_stack[stack_ptr].dist = d6; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5], - &traversal_stack[stack_ptr - 6]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Eight children are hit, push all onto stack and sort 8 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c7 = __float_as_int(cnodes[r]); - float d7 = ((float *)&dist)[r]; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c7; - traversal_stack[stack_ptr].dist = d7; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c6; - traversal_stack[stack_ptr].dist = d6; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5], - &traversal_stack[stack_ptr - 6], - &traversal_stack[stack_ptr - 7]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } - - /* If node is leaf, fetch triangle list. */ - if (node_addr < 0) { - float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr - 1)); - - if ((__float_as_uint(leaf.z) & visibility) == 0) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - int prim_addr = __float_as_int(leaf.x); - -#if BVH_FEATURE(BVH_INSTANCING) - if (prim_addr >= 0) { -#endif - int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - const uint p_type = type & PRIMITIVE_ALL; - - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - - /* Primitive intersection. */ - switch (p_type) { - case PRIMITIVE_TRIANGLE: { - 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; - } - /* Intersect ray against primitive. */ - triangle_intersect(kg, isect, P, dir, visibility, object, prim_addr); - } - break; - } -#if BVH_FEATURE(BVH_MOTION) - case PRIMITIVE_MOTION_TRIANGLE: { - 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; - } - /* Intersect ray against primitive. */ - motion_triangle_intersect( - kg, isect, P, dir, ray->time, visibility, object, prim_addr); - } - break; - } -#endif - } - } -#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) { -# if BVH_FEATURE(BVH_MOTION) - 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); -# endif - - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = avxf(isect->t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = avx3f(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -# endif - idir4 = avx3f(avxf(idir.x), avxf(idir.y), avxf(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = avx3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = avx3f(avxf(P.x), avxf(P.y), avxf(P.z)); -# endif - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = ENTRYPOINT_SENTINEL; - - node_addr = kernel_tex_fetch(__object_node, object); - } - else { - /* Pop. */ - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - --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); - - /* Instance pop. */ -# if BVH_FEATURE(BVH_MOTION) - 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); -# endif - - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = avxf(isect->t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = avx3f(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -# endif - idir4 = avx3f(avxf(idir.x), avxf(idir.y), avxf(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = avx3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = avx3f(avxf(P.x), avxf(P.y), avxf(P.z)); -# endif - - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - - return (isect->prim != PRIM_NONE); -} - -#undef NODE_INTERSECT diff --git a/intern/cycles/kernel/bvh/obvh_volume_all.h b/intern/cycles/kernel/bvh/obvh_volume_all.h deleted file mode 100644 index 56e2afd4a11..00000000000 --- a/intern/cycles/kernel/bvh/obvh_volume_all.h +++ /dev/null @@ -1,551 +0,0 @@ -/* - * Copyright 2011-2013 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -/* This is a template BVH traversal function for volumes, where - * various features can be enabled/disabled. This way we can compile optimized - * versions for each case without new features slowing things down. - * - * BVH_INSTANCING: object instancing - * BVH_MOTION: motion blur rendering - */ - -#if BVH_FEATURE(BVH_HAIR) -# define NODE_INTERSECT obvh_node_intersect -#else -# define NODE_INTERSECT obvh_aligned_node_intersect -#endif - -ccl_device uint BVH_FUNCTION_FULL_NAME(OBVH)(KernelGlobals *kg, - const Ray *ray, - Intersection *isect_array, - const uint max_hits, - const uint visibility) -{ - /* Traversal stack in CUDA thread-local memory. */ - OBVHStackItem traversal_stack[BVH_OSTACK_SIZE]; - traversal_stack[0].addr = ENTRYPOINT_SENTINEL; - - /* Traversal variables in registers. */ - int stack_ptr = 0; - int node_addr = kernel_data.bvh.root; - - /* Ray parameters in registers. */ - const float tmax = ray->t; - float3 P = ray->P; - float3 dir = bvh_clamp_direction(ray->D); - float3 idir = bvh_inverse_direction(dir); - int object = OBJECT_NONE; - float isect_t = tmax; - -#if BVH_FEATURE(BVH_MOTION) - Transform ob_itfm; -#endif - - uint num_hits = 0; - isect_array->t = tmax; - -#if BVH_FEATURE(BVH_INSTANCING) - int num_hits_in_instance = 0; -#endif - - avxf tnear(0.0f), tfar(isect_t); -#if BVH_FEATURE(BVH_HAIR) - avx3f dir4(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -#endif - avx3f idir4(avxf(idir.x), avxf(idir.y), avxf(idir.z)); - -#ifdef __KERNEL_AVX2__ - float3 P_idir = P * idir; - avx3f P_idir4(P_idir.x, P_idir.y, P_idir.z); -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - avx3f org4(avxf(P.x), avxf(P.y), avxf(P.z)); -#endif - - /* Offsets to select the side that becomes the lower or upper bound. */ - int near_x, near_y, near_z; - int far_x, far_y, far_z; - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - - /* Traversal loop. */ - do { - do { - /* Traverse internal nodes. */ - while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - -#ifdef __VISIBILITY_FLAG__ - if ((__float_as_uint(inodes.x) & visibility) == 0) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } -#endif - - avxf dist; - int child_mask = NODE_INTERSECT(kg, - tnear, - tfar, -#ifdef __KERNEL_AVX2__ - P_idir4, -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4, -#endif -#if BVH_FEATURE(BVH_HAIR) - dir4, -#endif - idir4, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - &dist); - - if (child_mask != 0) { - avxf cnodes; -#if BVH_FEATURE(BVH_HAIR) - if (__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) { - cnodes = kernel_tex_fetch_avxf(__bvh_nodes, node_addr + 26); - } - else -#endif - { - cnodes = kernel_tex_fetch_avxf(__bvh_nodes, node_addr + 14); - } - - /* One child is hit, continue with that child. */ - int r = __bscf(child_mask); - if (child_mask == 0) { - node_addr = __float_as_int(cnodes[r]); - continue; - } - - /* Two children are hit, push far child, and continue with - * closer child. - */ - int c0 = __float_as_int(cnodes[r]); - float d0 = ((float *)&dist)[r]; - r = __bscf(child_mask); - int c1 = __float_as_int(cnodes[r]); - float d1 = ((float *)&dist)[r]; - if (child_mask == 0) { - if (d1 < d0) { - node_addr = c1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - continue; - } - else { - node_addr = c0; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - continue; - } - } - - /* Here starts the slow path for 3 or 4 hit children. We push - * all nodes onto the stack to sort them there. - */ - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - - /* Three children are hit, push all onto stack and sort 3 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c2 = __float_as_int(cnodes[r]); - float d2 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Four children are hit, push all onto stack and sort 4 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c3 = __float_as_int(cnodes[r]); - float d3 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - - /* Five children are hit, push all onto stack and sort 5 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c4 = __float_as_int(cnodes[r]); - float d4 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Six children are hit, push all onto stack and sort 6 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c5 = __float_as_int(cnodes[r]); - float d5 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c5; - traversal_stack[stack_ptr].dist = d5; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c5; - traversal_stack[stack_ptr].dist = d5; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c4; - traversal_stack[stack_ptr].dist = d4; - - /* Seven children are hit, push all onto stack and sort 7 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c6 = __float_as_int(cnodes[r]); - float d6 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c6; - traversal_stack[stack_ptr].dist = d6; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5], - &traversal_stack[stack_ptr - 6]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Eight children are hit, push all onto stack and sort 8 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c7 = __float_as_int(cnodes[r]); - float d7 = ((float *)&dist)[r]; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c7; - traversal_stack[stack_ptr].dist = d7; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = c6; - traversal_stack[stack_ptr].dist = d6; - obvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3], - &traversal_stack[stack_ptr - 4], - &traversal_stack[stack_ptr - 5], - &traversal_stack[stack_ptr - 6], - &traversal_stack[stack_ptr - 7]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } - - /* If node is leaf, fetch triangle list. */ - if (node_addr < 0) { - float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr - 1)); - - if ((__float_as_uint(leaf.z) & visibility) == 0) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - int prim_addr = __float_as_int(leaf.x); - -#if BVH_FEATURE(BVH_INSTANCING) - if (prim_addr >= 0) { -#endif - int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - const uint p_type = type & PRIMITIVE_ALL; - bool hit; - - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - - /* Primitive intersection. */ - switch (p_type) { - case PRIMITIVE_TRIANGLE: { - 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; - } - /* Intersect ray against primitive. */ - hit = triangle_intersect(kg, isect_array, P, dir, visibility, object, prim_addr); - if (hit) { - /* Move on to next entry in intersections array. */ - isect_array++; - num_hits++; -#if BVH_FEATURE(BVH_INSTANCING) - num_hits_in_instance++; -#endif - isect_array->t = isect_t; - if (num_hits == max_hits) { -#if BVH_FEATURE(BVH_INSTANCING) -# if BVH_FEATURE(BVH_MOTION) - float t_fac = 1.0f / len(transform_direction(&ob_itfm, dir)); -# else - Transform itfm = object_fetch_transform(kg, object, OBJECT_INVERSE_TRANSFORM); - float t_fac = 1.0f / len(transform_direction(&itfm, dir)); -# endif - for (int i = 0; i < num_hits_in_instance; i++) { - (isect_array - i - 1)->t *= t_fac; - } -#endif /* BVH_FEATURE(BVH_INSTANCING) */ - return num_hits; - } - } - } - break; - } -#if BVH_FEATURE(BVH_MOTION) - case PRIMITIVE_MOTION_TRIANGLE: { - 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; - } - /* Intersect ray against primitive. */ - hit = motion_triangle_intersect( - kg, isect_array, P, dir, ray->time, visibility, object, prim_addr); - if (hit) { - /* Move on to next entry in intersections array. */ - isect_array++; - num_hits++; -# if BVH_FEATURE(BVH_INSTANCING) - num_hits_in_instance++; -# endif - isect_array->t = isect_t; - if (num_hits == max_hits) { -# if BVH_FEATURE(BVH_INSTANCING) -# if BVH_FEATURE(BVH_MOTION) - float t_fac = 1.0f / len(transform_direction(&ob_itfm, dir)); -# else - Transform itfm = object_fetch_transform(kg, object, OBJECT_INVERSE_TRANSFORM); - float t_fac = 1.0f / len(transform_direction(&itfm, dir)); -# endif - for (int i = 0; i < num_hits_in_instance; i++) { - (isect_array - i - 1)->t *= t_fac; - } -# endif /* BVH_FEATURE(BVH_INSTANCING) */ - return num_hits; - } - } - } - break; - } -#endif - } - } -#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) { -# if BVH_FEATURE(BVH_MOTION) - 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); -# endif - - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = avxf(isect_t); - idir4 = avx3f(avxf(idir.x), avxf(idir.y), avxf(idir.z)); -# if BVH_FEATURE(BVH_HAIR) - dir4 = avx3f(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -# endif -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = avx3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = avx3f(avxf(P.x), avxf(P.y), avxf(P.z)); -# endif - - num_hits_in_instance = 0; - isect_array->t = isect_t; - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_OSTACK_SIZE); - traversal_stack[stack_ptr].addr = ENTRYPOINT_SENTINEL; - - node_addr = kernel_tex_fetch(__object_node, object); - } - else { - /* Pop. */ - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - --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); - - /* Instance pop. */ - if (num_hits_in_instance) { - float t_fac; -# if BVH_FEATURE(BVH_MOTION) - bvh_instance_motion_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac, &ob_itfm); -# else - bvh_instance_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac); -# endif - /* Scale isect->t to adjust for instancing. */ - for (int i = 0; i < num_hits_in_instance; i++) { - (isect_array - i - 1)->t *= t_fac; - } - } - else { -# if BVH_FEATURE(BVH_MOTION) - bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX, &ob_itfm); -# else - bvh_instance_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX); -# endif - } - - isect_t = tmax; - isect_array->t = isect_t; - - obvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = avxf(isect_t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = avx3f(avxf(dir.x), avxf(dir.y), avxf(dir.z)); -# endif - idir4 = avx3f(avxf(idir.x), avxf(idir.y), avxf(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = avx3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = avx3f(avxf(P.x), avxf(P.y), avxf(P.z)); -# endif - - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - - return num_hits; -} - -#undef NODE_INTERSECT diff --git a/intern/cycles/kernel/bvh/qbvh_local.h b/intern/cycles/kernel/bvh/qbvh_local.h deleted file mode 100644 index b21f79bd3a0..00000000000 --- a/intern/cycles/kernel/bvh/qbvh_local.h +++ /dev/null @@ -1,291 +0,0 @@ -/* - * Copyright 2011-2013 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -/* This is a template BVH traversal function for finding local intersections - * around the shading point, for subsurface scattering and bevel. We disable - * various features for performance, and for instanced objects avoid traversing - * other parts of the scene. - * - * BVH_MOTION: motion blur rendering - */ - -#if BVH_FEATURE(BVH_HAIR) -# define NODE_INTERSECT qbvh_node_intersect -#else -# define NODE_INTERSECT qbvh_aligned_node_intersect -#endif - -ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg, - const Ray *ray, - LocalIntersection *local_isect, - int local_object, - uint *lcg_state, - int max_hits) -{ - /* TODO(sergey): - * - Test if pushing distance on the stack helps (for non shadow rays). - * - Separate version for shadow rays. - * - Likely and unlikely for if() statements. - * - SSE for hair. - * - Test restrict attribute for pointers. - */ - - /* Traversal stack in CUDA thread-local memory. */ - QBVHStackItem traversal_stack[BVH_QSTACK_SIZE]; - traversal_stack[0].addr = ENTRYPOINT_SENTINEL; - - /* Traversal variables in registers. */ - int stack_ptr = 0; - int node_addr = kernel_tex_fetch(__object_node, local_object); - - /* 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; - float isect_t = ray->t; - - if (local_isect != NULL) { - local_isect->num_hits = 0; - } - kernel_assert((local_isect == NULL) == (max_hits == 0)); - - const int object_flag = kernel_tex_fetch(__object_flag, local_object); - if (!(object_flag & SD_OBJECT_TRANSFORM_APPLIED)) { -#if BVH_FEATURE(BVH_MOTION) - Transform ob_itfm; - isect_t = bvh_instance_motion_push(kg, local_object, ray, &P, &dir, &idir, isect_t, &ob_itfm); -#else - isect_t = bvh_instance_push(kg, local_object, ray, &P, &dir, &idir, isect_t); -#endif - object = local_object; - } - - ssef tnear(0.0f), tfar(isect_t); -#if BVH_FEATURE(BVH_HAIR) - sse3f dir4(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -#endif - sse3f idir4(ssef(idir.x), ssef(idir.y), ssef(idir.z)); - -#ifdef __KERNEL_AVX2__ - float3 P_idir = P * idir; - sse3f P_idir4(P_idir.x, P_idir.y, P_idir.z); -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - sse3f org4(ssef(P.x), ssef(P.y), ssef(P.z)); -#endif - - /* Offsets to select the side that becomes the lower or upper bound. */ - int near_x, near_y, near_z; - int far_x, far_y, far_z; - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - - /* Traversal loop. */ - do { - do { - /* Traverse internal nodes. */ - while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - ssef dist; - int child_mask = NODE_INTERSECT(kg, - tnear, - tfar, -#ifdef __KERNEL_AVX2__ - P_idir4, -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4, -#endif -#if BVH_FEATURE(BVH_HAIR) - dir4, -#endif - idir4, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - &dist); - - if (child_mask != 0) { - float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - float4 cnodes; -#if BVH_FEATURE(BVH_HAIR) - if (__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) { - cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 13); - } - else -#endif - { - cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 7); - } - - /* One child is hit, continue with that child. */ - int r = __bscf(child_mask); - if (child_mask == 0) { - node_addr = __float_as_int(cnodes[r]); - continue; - } - - /* Two children are hit, push far child, and continue with - * closer child. - */ - int c0 = __float_as_int(cnodes[r]); - float d0 = ((float *)&dist)[r]; - r = __bscf(child_mask); - int c1 = __float_as_int(cnodes[r]); - float d1 = ((float *)&dist)[r]; - if (child_mask == 0) { - if (d1 < d0) { - node_addr = c1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - continue; - } - else { - node_addr = c0; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - continue; - } - } - - /* Here starts the slow path for 3 or 4 hit children. We push - * all nodes onto the stack to sort them there. - */ - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - - /* Three children are hit, push all onto stack and sort 3 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c2 = __float_as_int(cnodes[r]); - float d2 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - qbvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Four children are hit, push all onto stack and sort 4 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c3 = __float_as_int(cnodes[r]); - float d3 = ((float *)&dist)[r]; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - qbvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3]); - } - - node_addr = traversal_stack[stack_ptr].addr; - --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); - - int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --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); - if (triangle_intersect_local(kg, - local_isect, - P, - dir, - object, - local_object, - prim_addr, - isect_t, - lcg_state, - max_hits)) { - return true; - } - } - 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); - if (motion_triangle_intersect_local(kg, - local_isect, - P, - dir, - ray->time, - object, - local_object, - prim_addr, - isect_t, - lcg_state, - max_hits)) { - return true; - } - } - break; - } -#endif - default: - break; - } - } - } while (node_addr != ENTRYPOINT_SENTINEL); - } while (node_addr != ENTRYPOINT_SENTINEL); - - return false; -} - -#undef NODE_INTERSECT diff --git a/intern/cycles/kernel/bvh/qbvh_nodes.h b/intern/cycles/kernel/bvh/qbvh_nodes.h deleted file mode 100644 index 070406fb18a..00000000000 --- a/intern/cycles/kernel/bvh/qbvh_nodes.h +++ /dev/null @@ -1,329 +0,0 @@ -/* - * Copyright 2011-2014, Blender Foundation. - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - * - * Aligned nodes intersection SSE code is adopted from Embree, - */ - -struct QBVHStackItem { - int addr; - float dist; -}; - -ccl_device_inline void qbvh_near_far_idx_calc(const float3 &idir, - int *ccl_restrict near_x, - int *ccl_restrict near_y, - int *ccl_restrict near_z, - int *ccl_restrict far_x, - int *ccl_restrict far_y, - int *ccl_restrict far_z) - -{ -#ifdef __KERNEL_SSE__ - *near_x = 0; - *far_x = 1; - *near_y = 2; - *far_y = 3; - *near_z = 4; - *far_z = 5; - - const size_t mask = movemask(ssef(idir.m128)); - - const int mask_x = mask & 1; - const int mask_y = (mask & 2) >> 1; - const int mask_z = (mask & 4) >> 2; - - *near_x += mask_x; - *far_x -= mask_x; - *near_y += mask_y; - *far_y -= mask_y; - *near_z += mask_z; - *far_z -= mask_z; -#else - if (idir.x >= 0.0f) { - *near_x = 0; - *far_x = 1; - } - else { - *near_x = 1; - *far_x = 0; - } - if (idir.y >= 0.0f) { - *near_y = 2; - *far_y = 3; - } - else { - *near_y = 3; - *far_y = 2; - } - if (idir.z >= 0.0f) { - *near_z = 4; - *far_z = 5; - } - else { - *near_z = 5; - *far_z = 4; - } -#endif -} - -/* TOOD(sergey): Investigate if using intrinsics helps for both - * stack item swap and float comparison. - */ -ccl_device_inline void qbvh_item_swap(QBVHStackItem *ccl_restrict a, QBVHStackItem *ccl_restrict b) -{ - QBVHStackItem tmp = *a; - *a = *b; - *b = tmp; -} - -ccl_device_inline void qbvh_stack_sort(QBVHStackItem *ccl_restrict s1, - QBVHStackItem *ccl_restrict s2, - QBVHStackItem *ccl_restrict s3) -{ - if (s2->dist < s1->dist) { - qbvh_item_swap(s2, s1); - } - if (s3->dist < s2->dist) { - qbvh_item_swap(s3, s2); - } - if (s2->dist < s1->dist) { - qbvh_item_swap(s2, s1); - } -} - -ccl_device_inline void qbvh_stack_sort(QBVHStackItem *ccl_restrict s1, - QBVHStackItem *ccl_restrict s2, - QBVHStackItem *ccl_restrict s3, - QBVHStackItem *ccl_restrict s4) -{ - if (s2->dist < s1->dist) { - qbvh_item_swap(s2, s1); - } - if (s4->dist < s3->dist) { - qbvh_item_swap(s4, s3); - } - if (s3->dist < s1->dist) { - qbvh_item_swap(s3, s1); - } - if (s4->dist < s2->dist) { - qbvh_item_swap(s4, s2); - } - if (s3->dist < s2->dist) { - qbvh_item_swap(s3, s2); - } -} - -/* Axis-aligned nodes intersection */ - -// ccl_device_inline int qbvh_aligned_node_intersect(KernelGlobals *ccl_restrict kg, -static int qbvh_aligned_node_intersect(KernelGlobals *ccl_restrict kg, - const ssef &isect_near, - const ssef &isect_far, -#ifdef __KERNEL_AVX2__ - const sse3f &org_idir, -#else - const sse3f &org, -#endif - const sse3f &idir, - const int near_x, - const int near_y, - const int near_z, - const int far_x, - const int far_y, - const int far_z, - const int node_addr, - ssef *ccl_restrict dist) -{ - const int offset = node_addr + 1; -#ifdef __KERNEL_AVX2__ - const ssef tnear_x = msub( - kernel_tex_fetch_ssef(__bvh_nodes, offset + near_x), idir.x, org_idir.x); - const ssef tnear_y = msub( - kernel_tex_fetch_ssef(__bvh_nodes, offset + near_y), idir.y, org_idir.y); - const ssef tnear_z = msub( - kernel_tex_fetch_ssef(__bvh_nodes, offset + near_z), idir.z, org_idir.z); - const ssef tfar_x = msub(kernel_tex_fetch_ssef(__bvh_nodes, offset + far_x), idir.x, org_idir.x); - const ssef tfar_y = msub(kernel_tex_fetch_ssef(__bvh_nodes, offset + far_y), idir.y, org_idir.y); - const ssef tfar_z = msub(kernel_tex_fetch_ssef(__bvh_nodes, offset + far_z), idir.z, org_idir.z); -#else - const ssef tnear_x = (kernel_tex_fetch_ssef(__bvh_nodes, offset + near_x) - org.x) * idir.x; - const ssef tnear_y = (kernel_tex_fetch_ssef(__bvh_nodes, offset + near_y) - org.y) * idir.y; - const ssef tnear_z = (kernel_tex_fetch_ssef(__bvh_nodes, offset + near_z) - org.z) * idir.z; - const ssef tfar_x = (kernel_tex_fetch_ssef(__bvh_nodes, offset + far_x) - org.x) * idir.x; - const ssef tfar_y = (kernel_tex_fetch_ssef(__bvh_nodes, offset + far_y) - org.y) * idir.y; - const ssef tfar_z = (kernel_tex_fetch_ssef(__bvh_nodes, offset + far_z) - org.z) * idir.z; -#endif - -#ifdef __KERNEL_SSE41__ - const ssef tnear = maxi(maxi(tnear_x, tnear_y), maxi(tnear_z, isect_near)); - const ssef tfar = mini(mini(tfar_x, tfar_y), mini(tfar_z, isect_far)); - const sseb vmask = cast(tnear) > cast(tfar); - int mask = (int)movemask(vmask) ^ 0xf; -#else - const ssef tnear = max4(isect_near, tnear_x, tnear_y, tnear_z); - const ssef tfar = min4(isect_far, tfar_x, tfar_y, tfar_z); - const sseb vmask = tnear <= tfar; - int mask = (int)movemask(vmask); -#endif - *dist = tnear; - return mask; -} - -/* Unaligned nodes intersection */ - -ccl_device_inline int qbvh_unaligned_node_intersect(KernelGlobals *ccl_restrict kg, - const ssef &isect_near, - const ssef &isect_far, -#ifdef __KERNEL_AVX2__ - const sse3f &org_idir, -#endif - const sse3f &org, - const sse3f &dir, - const sse3f &idir, - const int near_x, - const int near_y, - const int near_z, - const int far_x, - const int far_y, - const int far_z, - const int node_addr, - ssef *ccl_restrict dist) -{ - const int offset = node_addr; - const ssef tfm_x_x = kernel_tex_fetch_ssef(__bvh_nodes, offset + 1); - const ssef tfm_x_y = kernel_tex_fetch_ssef(__bvh_nodes, offset + 2); - const ssef tfm_x_z = kernel_tex_fetch_ssef(__bvh_nodes, offset + 3); - - const ssef tfm_y_x = kernel_tex_fetch_ssef(__bvh_nodes, offset + 4); - const ssef tfm_y_y = kernel_tex_fetch_ssef(__bvh_nodes, offset + 5); - const ssef tfm_y_z = kernel_tex_fetch_ssef(__bvh_nodes, offset + 6); - - const ssef tfm_z_x = kernel_tex_fetch_ssef(__bvh_nodes, offset + 7); - const ssef tfm_z_y = kernel_tex_fetch_ssef(__bvh_nodes, offset + 8); - const ssef tfm_z_z = kernel_tex_fetch_ssef(__bvh_nodes, offset + 9); - - const ssef tfm_t_x = kernel_tex_fetch_ssef(__bvh_nodes, offset + 10); - const ssef tfm_t_y = kernel_tex_fetch_ssef(__bvh_nodes, offset + 11); - const ssef tfm_t_z = kernel_tex_fetch_ssef(__bvh_nodes, offset + 12); - - const ssef aligned_dir_x = dir.x * tfm_x_x + dir.y * tfm_x_y + dir.z * tfm_x_z, - aligned_dir_y = dir.x * tfm_y_x + dir.y * tfm_y_y + dir.z * tfm_y_z, - aligned_dir_z = dir.x * tfm_z_x + dir.y * tfm_z_y + dir.z * tfm_z_z; - - const ssef aligned_P_x = org.x * tfm_x_x + org.y * tfm_x_y + org.z * tfm_x_z + tfm_t_x, - aligned_P_y = org.x * tfm_y_x + org.y * tfm_y_y + org.z * tfm_y_z + tfm_t_y, - aligned_P_z = org.x * tfm_z_x + org.y * tfm_z_y + org.z * tfm_z_z + tfm_t_z; - - const ssef neg_one(-1.0f, -1.0f, -1.0f, -1.0f); - const ssef nrdir_x = neg_one / aligned_dir_x, nrdir_y = neg_one / aligned_dir_y, - nrdir_z = neg_one / aligned_dir_z; - - const ssef tlower_x = aligned_P_x * nrdir_x, tlower_y = aligned_P_y * nrdir_y, - tlower_z = aligned_P_z * nrdir_z; - - const ssef tupper_x = tlower_x - nrdir_x, tupper_y = tlower_y - nrdir_y, - tupper_z = tlower_z - nrdir_z; - -#ifdef __KERNEL_SSE41__ - const ssef tnear_x = mini(tlower_x, tupper_x); - const ssef tnear_y = mini(tlower_y, tupper_y); - const ssef tnear_z = mini(tlower_z, tupper_z); - const ssef tfar_x = maxi(tlower_x, tupper_x); - const ssef tfar_y = maxi(tlower_y, tupper_y); - const ssef tfar_z = maxi(tlower_z, tupper_z); - const ssef tnear = max4(isect_near, tnear_x, tnear_y, tnear_z); - const ssef tfar = min4(isect_far, tfar_x, tfar_y, tfar_z); - const sseb vmask = tnear <= tfar; - *dist = tnear; - return movemask(vmask); -#else - const ssef tnear_x = min(tlower_x, tupper_x); - const ssef tnear_y = min(tlower_y, tupper_y); - const ssef tnear_z = min(tlower_z, tupper_z); - const ssef tfar_x = max(tlower_x, tupper_x); - const ssef tfar_y = max(tlower_y, tupper_y); - const ssef tfar_z = max(tlower_z, tupper_z); - const ssef tnear = max4(isect_near, tnear_x, tnear_y, tnear_z); - const ssef tfar = min4(isect_far, tfar_x, tfar_y, tfar_z); - const sseb vmask = tnear <= tfar; - *dist = tnear; - return movemask(vmask); -#endif -} - -/* Intersectors wrappers. - * - * They'll check node type and call appropriate intersection code. - */ - -ccl_device_inline int qbvh_node_intersect(KernelGlobals *ccl_restrict kg, - const ssef &isect_near, - const ssef &isect_far, -#ifdef __KERNEL_AVX2__ - const sse3f &org_idir, -#endif - const sse3f &org, - const sse3f &dir, - const sse3f &idir, - const int near_x, - const int near_y, - const int near_z, - const int far_x, - const int far_y, - const int far_z, - const int node_addr, - ssef *ccl_restrict dist) -{ - const int offset = node_addr; - const float4 node = kernel_tex_fetch(__bvh_nodes, offset); - if (__float_as_uint(node.x) & PATH_RAY_NODE_UNALIGNED) { - return qbvh_unaligned_node_intersect(kg, - isect_near, - isect_far, -#ifdef __KERNEL_AVX2__ - org_idir, -#endif - org, - dir, - idir, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - dist); - } - else { - return qbvh_aligned_node_intersect(kg, - isect_near, - isect_far, -#ifdef __KERNEL_AVX2__ - org_idir, -#else - org, -#endif - idir, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - dist); - } -} diff --git a/intern/cycles/kernel/bvh/qbvh_shadow_all.h b/intern/cycles/kernel/bvh/qbvh_shadow_all.h deleted file mode 100644 index 9d428c3e1a7..00000000000 --- a/intern/cycles/kernel/bvh/qbvh_shadow_all.h +++ /dev/null @@ -1,433 +0,0 @@ -/* - * Copyright 2011-2013 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -/* This is a template BVH traversal function, where various features can be - * enabled/disabled. This way we can compile optimized versions for each case - * without new features slowing things down. - * - * BVH_INSTANCING: object instancing - * BVH_HAIR: hair curve rendering - * BVH_MOTION: motion blur rendering - */ - -#if BVH_FEATURE(BVH_HAIR) -# define NODE_INTERSECT qbvh_node_intersect -#else -# define NODE_INTERSECT qbvh_aligned_node_intersect -#endif - -ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg, - const Ray *ray, - Intersection *isect_array, - const uint visibility, - const uint max_hits, - uint *num_hits) -{ - /* TODO(sergey): - * - Test if pushing distance on the stack helps. - * - Likely and unlikely for if() statements. - * - Test restrict attribute for pointers. - */ - - /* Traversal stack in CUDA thread-local memory. */ - QBVHStackItem traversal_stack[BVH_QSTACK_SIZE]; - traversal_stack[0].addr = ENTRYPOINT_SENTINEL; - - /* Traversal variables in registers. */ - int stack_ptr = 0; - int node_addr = kernel_data.bvh.root; - - /* Ray parameters in registers. */ - const float tmax = ray->t; - float3 P = ray->P; - float3 dir = bvh_clamp_direction(ray->D); - float3 idir = bvh_inverse_direction(dir); - int object = OBJECT_NONE; - float isect_t = tmax; - -#if BVH_FEATURE(BVH_MOTION) - Transform ob_itfm; -#endif - - *num_hits = 0; - isect_array->t = tmax; - -#if BVH_FEATURE(BVH_INSTANCING) - int num_hits_in_instance = 0; -#endif - - ssef tnear(0.0f), tfar(isect_t); -#if BVH_FEATURE(BVH_HAIR) - sse3f dir4(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -#endif - sse3f idir4(ssef(idir.x), ssef(idir.y), ssef(idir.z)); - -#ifdef __KERNEL_AVX2__ - float3 P_idir = P * idir; - sse3f P_idir4(P_idir.x, P_idir.y, P_idir.z); -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - sse3f org4(ssef(P.x), ssef(P.y), ssef(P.z)); -#endif - - /* Offsets to select the side that becomes the lower or upper bound. */ - int near_x, near_y, near_z; - int far_x, far_y, far_z; - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - - /* Traversal loop. */ - do { - do { - /* Traverse internal nodes. */ - while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - (void)inodes; - - if (false -#ifdef __VISIBILITY_FLAG__ - || ((__float_as_uint(inodes.x) & visibility) == 0) -#endif -#if BVH_FEATURE(BVH_MOTION) - || UNLIKELY(ray->time < inodes.y) || UNLIKELY(ray->time > inodes.z) -#endif - ) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - ssef dist; - int child_mask = NODE_INTERSECT(kg, - tnear, - tfar, -#ifdef __KERNEL_AVX2__ - P_idir4, -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4, -#endif -#if BVH_FEATURE(BVH_HAIR) - dir4, -#endif - idir4, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - &dist); - - if (child_mask != 0) { - float4 cnodes; -#if BVH_FEATURE(BVH_HAIR) - if (__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) { - cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 13); - } - else -#endif - { - cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 7); - } - - /* One child is hit, continue with that child. */ - int r = __bscf(child_mask); - if (child_mask == 0) { - node_addr = __float_as_int(cnodes[r]); - continue; - } - - /* Two children are hit, push far child, and continue with - * closer child. - */ - int c0 = __float_as_int(cnodes[r]); - float d0 = ((float *)&dist)[r]; - r = __bscf(child_mask); - int c1 = __float_as_int(cnodes[r]); - float d1 = ((float *)&dist)[r]; - if (child_mask == 0) { - if (d1 < d0) { - node_addr = c1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - continue; - } - else { - node_addr = c0; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - continue; - } - } - - /* Here starts the slow path for 3 or 4 hit children. We push - * all nodes onto the stack to sort them there. - */ - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - - /* Three children are hit, push all onto stack and sort 3 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c2 = __float_as_int(cnodes[r]); - float d2 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - qbvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Four children are hit, push all onto stack and sort 4 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c3 = __float_as_int(cnodes[r]); - float d3 = ((float *)&dist)[r]; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - qbvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3]); - } - - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } - - /* If node is leaf, fetch triangle list. */ - if (node_addr < 0) { - float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr - 1)); -#ifdef __VISIBILITY_FLAG__ - if ((__float_as_uint(leaf.z) & visibility) == 0) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } -#endif - - int prim_addr = __float_as_int(leaf.x); - -#if BVH_FEATURE(BVH_INSTANCING) - if (prim_addr >= 0) { -#endif - int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - const uint p_type = type & PRIMITIVE_ALL; - - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - - /* Primitive intersection. */ - while (prim_addr < prim_addr2) { - kernel_assert((kernel_tex_fetch(__prim_type, prim_addr) & PRIMITIVE_ALL) == p_type); - bool hit; - - /* todo: specialized intersect functions which don't fill in - * isect unless needed and check SD_HAS_TRANSPARENT_SHADOW? - * might give a few % performance improvement */ - - switch (p_type) { - case PRIMITIVE_TRIANGLE: { - hit = triangle_intersect(kg, isect_array, P, dir, visibility, object, prim_addr); - break; - } -#if BVH_FEATURE(BVH_MOTION) - case PRIMITIVE_MOTION_TRIANGLE: { - hit = motion_triangle_intersect( - kg, isect_array, P, dir, ray->time, visibility, object, prim_addr); - break; - } -#endif -#if BVH_FEATURE(BVH_HAIR) - case PRIMITIVE_CURVE: - case PRIMITIVE_MOTION_CURVE: { - const uint curve_type = kernel_tex_fetch(__prim_type, prim_addr); - hit = curve_intersect( - kg, isect_array, P, dir, visibility, object, prim_addr, ray->time, curve_type); - break; - } -#endif - default: { - hit = false; - break; - } - } - - /* Shadow ray early termination. */ - if (hit) { - /* detect if this surface has a shader with transparent shadows */ - - /* todo: optimize so primitive visibility flag indicates if - * the primitive has a transparent shadow shader? */ - int prim = kernel_tex_fetch(__prim_index, isect_array->prim); - int shader = 0; - -#ifdef __HAIR__ - if (kernel_tex_fetch(__prim_type, isect_array->prim) & PRIMITIVE_ALL_TRIANGLE) -#endif - { - shader = kernel_tex_fetch(__tri_shader, prim); - } -#ifdef __HAIR__ - else { - float4 str = kernel_tex_fetch(__curves, prim); - shader = __float_as_int(str.z); - } -#endif - int flag = kernel_tex_fetch(__shaders, (shader & SHADER_MASK)).flags; - - /* if no transparent shadows, all light is blocked */ - if (!(flag & SD_HAS_TRANSPARENT_SHADOW)) { - return true; - } - /* if maximum number of hits reached, block all light */ - else if (*num_hits == max_hits) { - return true; - } - - /* move on to next entry in intersections array */ - isect_array++; - (*num_hits)++; -#if BVH_FEATURE(BVH_INSTANCING) - num_hits_in_instance++; -#endif - - isect_array->t = isect_t; - } - - prim_addr++; - } - } -#if BVH_FEATURE(BVH_INSTANCING) - else { - /* Instance push. */ - object = kernel_tex_fetch(__prim_object, -prim_addr - 1); - -# if BVH_FEATURE(BVH_MOTION) - 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); -# endif - - num_hits_in_instance = 0; - isect_array->t = isect_t; - - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = ssef(isect_t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -# endif - idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z)); -# endif - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = ENTRYPOINT_SENTINEL; - - node_addr = kernel_tex_fetch(__object_node, object); - } - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - -#if BVH_FEATURE(BVH_INSTANCING) - if (stack_ptr >= 0) { - kernel_assert(object != OBJECT_NONE); - - /* Instance pop. */ - if (num_hits_in_instance) { - float t_fac; -# if BVH_FEATURE(BVH_MOTION) - bvh_instance_motion_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac, &ob_itfm); -# else - bvh_instance_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac); -# endif - /* Scale isect->t to adjust for instancing. */ - for (int i = 0; i < num_hits_in_instance; i++) { - (isect_array - i - 1)->t *= t_fac; - } - } - else { -# if BVH_FEATURE(BVH_MOTION) - bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX, &ob_itfm); -# else - bvh_instance_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX); -# endif - } - - isect_t = tmax; - isect_array->t = isect_t; - - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = ssef(isect_t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -# endif - idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z)); -# endif - - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - - return false; -} - -#undef NODE_INTERSECT diff --git a/intern/cycles/kernel/bvh/qbvh_traversal.h b/intern/cycles/kernel/bvh/qbvh_traversal.h deleted file mode 100644 index f68579b4d69..00000000000 --- a/intern/cycles/kernel/bvh/qbvh_traversal.h +++ /dev/null @@ -1,413 +0,0 @@ -/* - * Copyright 2011-2013 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -/* This is a template BVH traversal function, where various features can be - * enabled/disabled. This way we can compile optimized versions for each case - * without new features slowing things down. - * - * BVH_INSTANCING: object instancing - * BVH_HAIR: hair curve rendering - * BVH_MOTION: motion blur rendering - */ - -#if BVH_FEATURE(BVH_HAIR) -# define NODE_INTERSECT qbvh_node_intersect -#else -# define NODE_INTERSECT qbvh_aligned_node_intersect -#endif - -ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg, - const Ray *ray, - Intersection *isect, - const uint visibility) -{ - /* TODO(sergey): - * - 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. */ - QBVHStackItem traversal_stack[BVH_QSTACK_SIZE]; - traversal_stack[0].addr = ENTRYPOINT_SENTINEL; - traversal_stack[0].dist = -FLT_MAX; - - /* Traversal variables in registers. */ - int stack_ptr = 0; - int node_addr = kernel_data.bvh.root; - float node_dist = -FLT_MAX; - - /* 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; -#endif - - isect->t = ray->t; - isect->u = 0.0f; - isect->v = 0.0f; - isect->prim = PRIM_NONE; - isect->object = OBJECT_NONE; - - BVH_DEBUG_INIT(); - - ssef tnear(0.0f), tfar(ray->t); -#if BVH_FEATURE(BVH_HAIR) - sse3f dir4(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -#endif - sse3f idir4(ssef(idir.x), ssef(idir.y), ssef(idir.z)); - -#ifdef __KERNEL_AVX2__ - float3 P_idir = P * idir; - sse3f P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z); -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - sse3f org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z)); -#endif - - /* Offsets to select the side that becomes the lower or upper bound. */ - int near_x, near_y, near_z; - int far_x, far_y, far_z; - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - - /* Traversal loop. */ - do { - do { - /* Traverse internal nodes. */ - while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - (void)inodes; - - if (UNLIKELY(node_dist > isect->t) -#if BVH_FEATURE(BVH_MOTION) - || UNLIKELY(ray->time < inodes.y) || UNLIKELY(ray->time > inodes.z) -#endif -#ifdef __VISIBILITY_FLAG__ - || (__float_as_uint(inodes.x) & visibility) == 0 -#endif - ) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - - int child_mask; - ssef dist; - - BVH_DEBUG_NEXT_NODE(); - - { - child_mask = NODE_INTERSECT(kg, - tnear, - tfar, -#ifdef __KERNEL_AVX2__ - P_idir4, -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4, -#endif -#if BVH_FEATURE(BVH_HAIR) - dir4, -#endif - idir4, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - &dist); - } - - if (child_mask != 0) { - float4 cnodes; - /* TODO(sergey): Investigate whether moving cnodes upwards - * gives a speedup (will be different cache pattern but will - * avoid extra check here). - */ -#if BVH_FEATURE(BVH_HAIR) - if (__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) { - cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 13); - } - else -#endif - { - cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 7); - } - - /* One child is hit, continue with that child. */ - int r = __bscf(child_mask); - float d0 = ((float *)&dist)[r]; - if (child_mask == 0) { - node_addr = __float_as_int(cnodes[r]); - node_dist = d0; - continue; - } - - /* Two children are hit, push far child, and continue with - * closer child. - */ - int c0 = __float_as_int(cnodes[r]); - r = __bscf(child_mask); - int c1 = __float_as_int(cnodes[r]); - float d1 = ((float *)&dist)[r]; - if (child_mask == 0) { - if (d1 < d0) { - node_addr = c1; - node_dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - continue; - } - else { - node_addr = c0; - node_dist = d0; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - continue; - } - } - - /* Here starts the slow path for 3 or 4 hit children. We push - * all nodes onto the stack to sort them there. - */ - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - - /* Three children are hit, push all onto stack and sort 3 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c2 = __float_as_int(cnodes[r]); - float d2 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - qbvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2]); - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - - /* Four children are hit, push all onto stack and sort 4 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c3 = __float_as_int(cnodes[r]); - float d3 = ((float *)&dist)[r]; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - qbvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3]); - } - - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - } - - /* If node is leaf, fetch triangle list. */ - if (node_addr < 0) { - float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr - 1)); - -#ifdef __VISIBILITY_FLAG__ - if (UNLIKELY((node_dist > isect->t) || ((__float_as_uint(leaf.z) & visibility) == 0))) -#else - if (UNLIKELY((node_dist > isect->t))) -#endif - { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - continue; - } - - int prim_addr = __float_as_int(leaf.x); - -#if BVH_FEATURE(BVH_INSTANCING) - if (prim_addr >= 0) { -#endif - int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - - /* Primitive intersection. */ - switch (type & PRIMITIVE_ALL) { - case PRIMITIVE_TRIANGLE: { - for (; prim_addr < prim_addr2; prim_addr++) { - BVH_DEBUG_NEXT_INTERSECTION(); - kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type); - if (triangle_intersect(kg, isect, P, dir, visibility, object, prim_addr)) { - tfar = ssef(isect->t); - /* Shadow ray early termination. */ - if (visibility & PATH_RAY_SHADOW_OPAQUE) { - return true; - } - } - } - break; - } -#if BVH_FEATURE(BVH_MOTION) - case PRIMITIVE_MOTION_TRIANGLE: { - for (; prim_addr < prim_addr2; prim_addr++) { - BVH_DEBUG_NEXT_INTERSECTION(); - kernel_assert(kernel_tex_fetch(__prim_type, prim_addr) == type); - if (motion_triangle_intersect( - kg, isect, P, dir, ray->time, visibility, object, prim_addr)) { - tfar = ssef(isect->t); - /* Shadow ray early termination. */ - if (visibility & PATH_RAY_SHADOW_OPAQUE) { - return true; - } - } - } - break; - } -#endif /* BVH_FEATURE(BVH_MOTION) */ -#if BVH_FEATURE(BVH_HAIR) - case PRIMITIVE_CURVE: - case PRIMITIVE_MOTION_CURVE: { - for (; prim_addr < prim_addr2; prim_addr++) { - BVH_DEBUG_NEXT_INTERSECTION(); - const uint curve_type = kernel_tex_fetch(__prim_type, prim_addr); - kernel_assert((curve_type & PRIMITIVE_ALL) == (type & PRIMITIVE_ALL)); - bool hit = curve_intersect( - kg, isect, P, dir, visibility, object, prim_addr, ray->time, curve_type); - if (hit) { - tfar = ssef(isect->t); - /* Shadow ray early termination. */ - if (visibility & PATH_RAY_SHADOW_OPAQUE) { - return true; - } - } - } - break; - } -#endif /* BVH_FEATURE(BVH_HAIR) */ - } - } -#if BVH_FEATURE(BVH_INSTANCING) - else { - /* Instance push. */ - object = kernel_tex_fetch(__prim_object, -prim_addr - 1); - -# if BVH_FEATURE(BVH_MOTION) - qbvh_instance_motion_push( - kg, object, ray, &P, &dir, &idir, &isect->t, &node_dist, &ob_itfm); -# else - qbvh_instance_push(kg, object, ray, &P, &dir, &idir, &isect->t, &node_dist); -# endif - - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = ssef(isect->t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -# endif - idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z)); -# endif - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = ENTRYPOINT_SENTINEL; - traversal_stack[stack_ptr].dist = -FLT_MAX; - - node_addr = kernel_tex_fetch(__object_node, object); - - BVH_DEBUG_NEXT_INSTANCE(); - } - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - -#if BVH_FEATURE(BVH_INSTANCING) - if (stack_ptr >= 0) { - kernel_assert(object != OBJECT_NONE); - - /* Instance pop. */ -# if BVH_FEATURE(BVH_MOTION) - 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); -# endif - - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = ssef(isect->t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -# endif - idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z)); -# endif - - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - node_dist = traversal_stack[stack_ptr].dist; - --stack_ptr; - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - - return (isect->prim != PRIM_NONE); -} - -#undef NODE_INTERSECT diff --git a/intern/cycles/kernel/bvh/qbvh_volume.h b/intern/cycles/kernel/bvh/qbvh_volume.h deleted file mode 100644 index e4eaed04467..00000000000 --- a/intern/cycles/kernel/bvh/qbvh_volume.h +++ /dev/null @@ -1,367 +0,0 @@ -/* - * Copyright 2011-2013 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -/* This is a template BVH traversal function for volumes, where - * various features can be enabled/disabled. This way we can compile optimized - * versions for each case without new features slowing things down. - * - * BVH_INSTANCING: object instancing - * BVH_MOTION: motion blur rendering - */ - -#if BVH_FEATURE(BVH_HAIR) -# define NODE_INTERSECT qbvh_node_intersect -#else -# define NODE_INTERSECT qbvh_aligned_node_intersect -#endif - -ccl_device bool BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg, - const Ray *ray, - Intersection *isect, - const uint visibility) -{ - /* TODO(sergey): - * - Test if pushing distance on the stack helps. - * - Likely and unlikely for if() statements. - * - Test restrict attribute for pointers. - */ - - /* Traversal stack in CUDA thread-local memory. */ - QBVHStackItem traversal_stack[BVH_QSTACK_SIZE]; - traversal_stack[0].addr = 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; -#endif - - isect->t = ray->t; - isect->u = 0.0f; - isect->v = 0.0f; - isect->prim = PRIM_NONE; - isect->object = OBJECT_NONE; - - ssef tnear(0.0f), tfar(ray->t); -#if BVH_FEATURE(BVH_HAIR) - sse3f dir4(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -#endif - sse3f idir4(ssef(idir.x), ssef(idir.y), ssef(idir.z)); - -#ifdef __KERNEL_AVX2__ - float3 P_idir = P * idir; - sse3f P_idir4(P_idir.x, P_idir.y, P_idir.z); -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - sse3f org4(ssef(P.x), ssef(P.y), ssef(P.z)); -#endif - - /* Offsets to select the side that becomes the lower or upper bound. */ - int near_x, near_y, near_z; - int far_x, far_y, far_z; - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - - /* Traversal loop. */ - do { - do { - /* Traverse internal nodes. */ - while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - -#ifdef __VISIBILITY_FLAG__ - if ((__float_as_uint(inodes.x) & visibility) == 0) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } -#endif - - ssef dist; - int child_mask = NODE_INTERSECT(kg, - tnear, - tfar, -#ifdef __KERNEL_AVX2__ - P_idir4, -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4, -#endif -#if BVH_FEATURE(BVH_HAIR) - dir4, -#endif - idir4, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - &dist); - - if (child_mask != 0) { - float4 cnodes; -#if BVH_FEATURE(BVH_HAIR) - if (__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) { - cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 13); - } - else -#endif - { - cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 7); - } - - /* One child is hit, continue with that child. */ - int r = __bscf(child_mask); - if (child_mask == 0) { - node_addr = __float_as_int(cnodes[r]); - continue; - } - - /* Two children are hit, push far child, and continue with - * closer child. - */ - int c0 = __float_as_int(cnodes[r]); - float d0 = ((float *)&dist)[r]; - r = __bscf(child_mask); - int c1 = __float_as_int(cnodes[r]); - float d1 = ((float *)&dist)[r]; - if (child_mask == 0) { - if (d1 < d0) { - node_addr = c1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - continue; - } - else { - node_addr = c0; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - continue; - } - } - - /* Here starts the slow path for 3 or 4 hit children. We push - * all nodes onto the stack to sort them there. - */ - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - - /* Three children are hit, push all onto stack and sort 3 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c2 = __float_as_int(cnodes[r]); - float d2 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - qbvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Four children are hit, push all onto stack and sort 4 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c3 = __float_as_int(cnodes[r]); - float d3 = ((float *)&dist)[r]; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - qbvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3]); - } - - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } - - /* If node is leaf, fetch triangle list. */ - if (node_addr < 0) { - float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr - 1)); - - if ((__float_as_uint(leaf.z) & visibility) == 0) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - int prim_addr = __float_as_int(leaf.x); - -#if BVH_FEATURE(BVH_INSTANCING) - if (prim_addr >= 0) { -#endif - int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - const uint p_type = type & PRIMITIVE_ALL; - - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - - /* Primitive intersection. */ - switch (p_type) { - case PRIMITIVE_TRIANGLE: { - 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; - } - /* Intersect ray against primitive. */ - triangle_intersect(kg, isect, P, dir, visibility, object, prim_addr); - } - break; - } -#if BVH_FEATURE(BVH_MOTION) - case PRIMITIVE_MOTION_TRIANGLE: { - 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; - } - /* Intersect ray against primitive. */ - motion_triangle_intersect( - kg, isect, P, dir, ray->time, visibility, object, prim_addr); - } - break; - } -#endif - } - } -#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) { -# if BVH_FEATURE(BVH_MOTION) - 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); -# endif - - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = ssef(isect->t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -# endif - idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z)); -# endif - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = ENTRYPOINT_SENTINEL; - - node_addr = kernel_tex_fetch(__object_node, object); - } - else { - /* Pop. */ - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - --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); - - /* Instance pop. */ -# if BVH_FEATURE(BVH_MOTION) - 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); -# endif - - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = ssef(isect->t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -# endif - idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z)); -# endif - - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - - return (isect->prim != PRIM_NONE); -} - -#undef NODE_INTERSECT diff --git a/intern/cycles/kernel/bvh/qbvh_volume_all.h b/intern/cycles/kernel/bvh/qbvh_volume_all.h deleted file mode 100644 index eddc48c487e..00000000000 --- a/intern/cycles/kernel/bvh/qbvh_volume_all.h +++ /dev/null @@ -1,444 +0,0 @@ -/* - * Copyright 2011-2013 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -/* This is a template BVH traversal function for volumes, where - * various features can be enabled/disabled. This way we can compile optimized - * versions for each case without new features slowing things down. - * - * BVH_INSTANCING: object instancing - * BVH_MOTION: motion blur rendering - */ - -#if BVH_FEATURE(BVH_HAIR) -# define NODE_INTERSECT qbvh_node_intersect -#else -# define NODE_INTERSECT qbvh_aligned_node_intersect -#endif - -ccl_device uint BVH_FUNCTION_FULL_NAME(QBVH)(KernelGlobals *kg, - const Ray *ray, - Intersection *isect_array, - const uint max_hits, - const uint visibility) -{ - /* TODO(sergey): - * - Test if pushing distance on the stack helps. - * - Likely and unlikely for if() statements. - * - Test restrict attribute for pointers. - */ - - /* Traversal stack in CUDA thread-local memory. */ - QBVHStackItem traversal_stack[BVH_QSTACK_SIZE]; - traversal_stack[0].addr = ENTRYPOINT_SENTINEL; - - /* Traversal variables in registers. */ - int stack_ptr = 0; - int node_addr = kernel_data.bvh.root; - - /* Ray parameters in registers. */ - const float tmax = ray->t; - float3 P = ray->P; - float3 dir = bvh_clamp_direction(ray->D); - float3 idir = bvh_inverse_direction(dir); - int object = OBJECT_NONE; - float isect_t = tmax; - -#if BVH_FEATURE(BVH_MOTION) - Transform ob_itfm; -#endif - - uint num_hits = 0; - isect_array->t = tmax; - -#if BVH_FEATURE(BVH_INSTANCING) - int num_hits_in_instance = 0; -#endif - - ssef tnear(0.0f), tfar(isect_t); -#if BVH_FEATURE(BVH_HAIR) - sse3f dir4(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -#endif - sse3f idir4(ssef(idir.x), ssef(idir.y), ssef(idir.z)); - -#ifdef __KERNEL_AVX2__ - float3 P_idir = P * idir; - sse3f P_idir4(P_idir.x, P_idir.y, P_idir.z); -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - sse3f org4(ssef(P.x), ssef(P.y), ssef(P.z)); -#endif - - /* Offsets to select the side that becomes the lower or upper bound. */ - int near_x, near_y, near_z; - int far_x, far_y, far_z; - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - - /* Traversal loop. */ - do { - do { - /* Traverse internal nodes. */ - while (node_addr >= 0 && node_addr != ENTRYPOINT_SENTINEL) { - float4 inodes = kernel_tex_fetch(__bvh_nodes, node_addr + 0); - -#ifdef __VISIBILITY_FLAG__ - if ((__float_as_uint(inodes.x) & visibility) == 0) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } -#endif - - ssef dist; - int child_mask = NODE_INTERSECT(kg, - tnear, - tfar, -#ifdef __KERNEL_AVX2__ - P_idir4, -#endif -#if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4, -#endif -#if BVH_FEATURE(BVH_HAIR) - dir4, -#endif - idir4, - near_x, - near_y, - near_z, - far_x, - far_y, - far_z, - node_addr, - &dist); - - if (child_mask != 0) { - float4 cnodes; -#if BVH_FEATURE(BVH_HAIR) - if (__float_as_uint(inodes.x) & PATH_RAY_NODE_UNALIGNED) { - cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 13); - } - else -#endif - { - cnodes = kernel_tex_fetch(__bvh_nodes, node_addr + 7); - } - - /* One child is hit, continue with that child. */ - int r = __bscf(child_mask); - if (child_mask == 0) { - node_addr = __float_as_int(cnodes[r]); - continue; - } - - /* Two children are hit, push far child, and continue with - * closer child. - */ - int c0 = __float_as_int(cnodes[r]); - float d0 = ((float *)&dist)[r]; - r = __bscf(child_mask); - int c1 = __float_as_int(cnodes[r]); - float d1 = ((float *)&dist)[r]; - if (child_mask == 0) { - if (d1 < d0) { - node_addr = c1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - continue; - } - else { - node_addr = c0; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - continue; - } - } - - /* Here starts the slow path for 3 or 4 hit children. We push - * all nodes onto the stack to sort them there. - */ - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c1; - traversal_stack[stack_ptr].dist = d1; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c0; - traversal_stack[stack_ptr].dist = d0; - - /* Three children are hit, push all onto stack and sort 3 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c2 = __float_as_int(cnodes[r]); - float d2 = ((float *)&dist)[r]; - if (child_mask == 0) { - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - qbvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2]); - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - /* Four children are hit, push all onto stack and sort 4 - * stack items, continue with closest child. - */ - r = __bscf(child_mask); - int c3 = __float_as_int(cnodes[r]); - float d3 = ((float *)&dist)[r]; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c3; - traversal_stack[stack_ptr].dist = d3; - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = c2; - traversal_stack[stack_ptr].dist = d2; - qbvh_stack_sort(&traversal_stack[stack_ptr], - &traversal_stack[stack_ptr - 1], - &traversal_stack[stack_ptr - 2], - &traversal_stack[stack_ptr - 3]); - } - - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } - - /* If node is leaf, fetch triangle list. */ - if (node_addr < 0) { - float4 leaf = kernel_tex_fetch(__bvh_leaf_nodes, (-node_addr - 1)); - - if ((__float_as_uint(leaf.z) & visibility) == 0) { - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - continue; - } - - int prim_addr = __float_as_int(leaf.x); - -#if BVH_FEATURE(BVH_INSTANCING) - if (prim_addr >= 0) { -#endif - int prim_addr2 = __float_as_int(leaf.y); - const uint type = __float_as_int(leaf.w); - const uint p_type = type & PRIMITIVE_ALL; - bool hit; - - /* Pop. */ - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - - /* Primitive intersection. */ - switch (p_type) { - case PRIMITIVE_TRIANGLE: { - 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; - } - /* Intersect ray against primitive. */ - hit = triangle_intersect(kg, isect_array, P, dir, visibility, object, prim_addr); - if (hit) { - /* Move on to next entry in intersections array. */ - isect_array++; - num_hits++; -#if BVH_FEATURE(BVH_INSTANCING) - num_hits_in_instance++; -#endif - isect_array->t = isect_t; - if (num_hits == max_hits) { -#if BVH_FEATURE(BVH_INSTANCING) - if (object != OBJECT_NONE) { -# if BVH_FEATURE(BVH_MOTION) - float t_fac = 1.0f / len(transform_direction(&ob_itfm, dir)); -# else - Transform itfm = object_fetch_transform( - kg, object, OBJECT_INVERSE_TRANSFORM); - float t_fac = 1.0f / len(transform_direction(&itfm, dir)); -# endif - for (int i = 0; i < num_hits_in_instance; i++) { - (isect_array - i - 1)->t *= t_fac; - } - } -#endif /* BVH_FEATURE(BVH_INSTANCING) */ - return num_hits; - } - } - } - break; - } -#if BVH_FEATURE(BVH_MOTION) - case PRIMITIVE_MOTION_TRIANGLE: { - 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; - } - /* Intersect ray against primitive. */ - hit = motion_triangle_intersect( - kg, isect_array, P, dir, ray->time, visibility, object, prim_addr); - if (hit) { - /* Move on to next entry in intersections array. */ - isect_array++; - num_hits++; -# if BVH_FEATURE(BVH_INSTANCING) - num_hits_in_instance++; -# endif - isect_array->t = isect_t; - if (num_hits == max_hits) { -# if BVH_FEATURE(BVH_INSTANCING) - if (object != OBJECT_NONE) { -# if BVH_FEATURE(BVH_MOTION) - float t_fac = 1.0f / len(transform_direction(&ob_itfm, dir)); -# else - Transform itfm = object_fetch_transform( - kg, object, OBJECT_INVERSE_TRANSFORM); - float t_fac = 1.0f / len(transform_direction(&itfm, dir)); -# endif - for (int i = 0; i < num_hits_in_instance; i++) { - (isect_array - i - 1)->t *= t_fac; - } - } -# endif /* BVH_FEATURE(BVH_INSTANCING) */ - return num_hits; - } - } - } - break; - } -#endif - } - } -#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) { -# if BVH_FEATURE(BVH_MOTION) - 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); -# endif - - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = ssef(isect_t); - idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z)); -# if BVH_FEATURE(BVH_HAIR) - dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -# endif -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z)); -# endif - - num_hits_in_instance = 0; - isect_array->t = isect_t; - - ++stack_ptr; - kernel_assert(stack_ptr < BVH_QSTACK_SIZE); - traversal_stack[stack_ptr].addr = ENTRYPOINT_SENTINEL; - - node_addr = kernel_tex_fetch(__object_node, object); - } - else { - /* Pop. */ - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - --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); - - /* Instance pop. */ - if (num_hits_in_instance) { - float t_fac; -# if BVH_FEATURE(BVH_MOTION) - bvh_instance_motion_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac, &ob_itfm); -# else - bvh_instance_pop_factor(kg, object, ray, &P, &dir, &idir, &t_fac); -# endif - /* Scale isect->t to adjust for instancing. */ - for (int i = 0; i < num_hits_in_instance; i++) { - (isect_array - i - 1)->t *= t_fac; - } - } - else { -# if BVH_FEATURE(BVH_MOTION) - bvh_instance_motion_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX, &ob_itfm); -# else - bvh_instance_pop(kg, object, ray, &P, &dir, &idir, FLT_MAX); -# endif - } - - isect_t = tmax; - isect_array->t = isect_t; - - qbvh_near_far_idx_calc(idir, &near_x, &near_y, &near_z, &far_x, &far_y, &far_z); - tfar = ssef(isect_t); -# if BVH_FEATURE(BVH_HAIR) - dir4 = sse3f(ssef(dir.x), ssef(dir.y), ssef(dir.z)); -# endif - idir4 = sse3f(ssef(idir.x), ssef(idir.y), ssef(idir.z)); -# ifdef __KERNEL_AVX2__ - P_idir = P * idir; - P_idir4 = sse3f(P_idir.x, P_idir.y, P_idir.z); -# endif -# if BVH_FEATURE(BVH_HAIR) || !defined(__KERNEL_AVX2__) - org4 = sse3f(ssef(P.x), ssef(P.y), ssef(P.z)); -# endif - - object = OBJECT_NONE; - node_addr = traversal_stack[stack_ptr].addr; - --stack_ptr; - } -#endif /* FEATURE(BVH_INSTANCING) */ - } while (node_addr != ENTRYPOINT_SENTINEL); - - return num_hits; -} - -#undef NODE_INTERSECT |