Page MenuHome
Differential D4532 Diff 14216 intern/cycles/kernel/bvh/obvh_nodes.h

Changeset View

Standalone View

View Options

intern/cycles/kernel/bvh/obvh_nodes.h

Show First 20 Lines • Show All 222 Lines • ▼ Show 20 Lines#ifdef __KERNEL_AVX2__
int mask = (int)movemask(vmask); int mask = (int)movemask(vmask);
*dist = tnear; *dist = tnear;
return mask; return mask;
#else #else
return 0; return 0;
#endif #endif
} }
ccl_device_inline int obvh_aligned_node_intersect_robust(
KernelGlobals *ccl_restrict kg,
const avxf& isect_near,
const avxf& isect_far,
#ifdef __KERNEL_AVX2__
const avx3f& P_idir,
#else
const avx3f& P,
#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,
const float difl,
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, P_idir.x);
const avxf tfar_x = msub(kernel_tex_fetch_avxf(__bvh_nodes, offset + far_x * 2), idir.x, P_idir.x);
const avxf tnear_y = msub(kernel_tex_fetch_avxf(__bvh_nodes, offset + near_y * 2), idir.y, P_idir.y);
const avxf tfar_y = msub(kernel_tex_fetch_avxf(__bvh_nodes, offset + far_y * 2), idir.y, P_idir.y);
const avxf tnear_z = msub(kernel_tex_fetch_avxf(__bvh_nodes, offset + near_z * 2), idir.z, P_idir.z);
const avxf tfar_z = msub(kernel_tex_fetch_avxf(__bvh_nodes, offset + far_z * 2), idir.z, P_idir.z);
const float round_down = 1.0f - difl;
const float round_up = 1.0f + difl;
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 = round_down*tnear <= round_up*tfar;
int mask = (int)movemask(vmask);
*dist = tnear;
return mask;
#else
return 0;
#endif
}
/* Unaligned nodes intersection */ /* Unaligned nodes intersection */
ccl_device_inline int obvh_unaligned_node_intersect( ccl_device_inline int obvh_unaligned_node_intersect(
KernelGlobals *ccl_restrict kg, KernelGlobals *ccl_restrict kg,
const avxf& isect_near, const avxf& isect_near,
const avxf& isect_far, const avxf& isect_far,
#ifdef __KERNEL_AVX2__ #ifdef __KERNEL_AVX2__
const avx3f& org_idir, const avx3f& org_idir,
▲ Show 20 Lines • Show All 56 Lines • ▼ Show 20 Lines#endif
const avxf tfar_z = max(tlower_z, tupper_z); const avxf tfar_z = max(tlower_z, tupper_z);
const avxf tnear = max4(isect_near, tnear_x, tnear_y, tnear_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 avxf tfar = min4(isect_far, tfar_x, tfar_y, tfar_z);
const avxb vmask = tnear <= tfar; const avxb vmask = tnear <= tfar;
*dist = tnear; *dist = tnear;
return movemask(vmask); return movemask(vmask);
} }
ccl_device_inline int obvh_unaligned_node_intersect_robust(
KernelGlobals *ccl_restrict kg,
const avxf& isect_near,
const avxf& isect_far,
#ifdef __KERNEL_AVX2__
const avx3f& P_idir,
#endif
const avx3f& P,
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,
const float difl,
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 = P.x*tfm_x_x + P.y*tfm_x_y + P.z*tfm_x_z + tfm_t_x,
aligned_P_y = P.x*tfm_y_x + P.y*tfm_y_y + P.z*tfm_y_z + tfm_t_y,
aligned_P_z = P.x*tfm_z_x + P.y*tfm_z_y + P.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 float round_down = 1.0f - difl;
const float round_up = 1.0f + difl;
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 = round_down*tnear <= round_up*tfar;
*dist = tnear;
return movemask(vmask);
}
/* Intersectors wrappers. /* Intersectors wrappers.
* *
* They'll check node type and call appropriate intersection code. * They'll check node type and call appropriate intersection code.
*/ */
ccl_device_inline int obvh_node_intersect( ccl_device_inline int obvh_node_intersect(
KernelGlobals *ccl_restrict kg, KernelGlobals *ccl_restrict kg,
const avxf& isect_near, const avxf& isect_near,
▲ Show 20 Lines • Show All 41 Lines • ▼ Show 20 Lines
#endif #endif
idir, idir,
near_x, near_y, near_z, near_x, near_y, near_z,
far_x, far_y, far_z, far_x, far_y, far_z,
node_addr, node_addr,
dist); dist);
} }
} }
ccl_device_inline int obvh_node_intersect_robust(
KernelGlobals *ccl_restrict kg,
const avxf& isect_near,
const avxf& isect_far,
#ifdef __KERNEL_AVX2__
const avx3f& P_idir,
#endif
const avx3f& P,
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,
const float difl,
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_robust(kg,
isect_near,
isect_far,
#ifdef __KERNEL_AVX2__
P_idir,
#endif
P,
dir,
idir,
near_x, near_y, near_z,
far_x, far_y, far_z,
node_addr,
difl,
dist);
}
else {
return obvh_aligned_node_intersect_robust(kg,
isect_near,
isect_far,
#ifdef __KERNEL_AVX2__
P_idir,
#else
P,
#endif
idir,
near_x, near_y, near_z,
far_x, far_y, far_z,
node_addr,
difl,
dist);
}
}