From 34b3e8f9e2ae076990ecf3b2827eff759ba2abf9 Mon Sep 17 00:00:00 2001 From: jfons Date: Tue, 20 Apr 2021 18:38:09 +0200 Subject: Add Embree-aarch64 thirdparty library --- .../kernels/geometry/disc_intersector.h | 216 +++++++++++++++++++++ 1 file changed, 216 insertions(+) create mode 100644 thirdparty/embree-aarch64/kernels/geometry/disc_intersector.h (limited to 'thirdparty/embree-aarch64/kernels/geometry/disc_intersector.h') diff --git a/thirdparty/embree-aarch64/kernels/geometry/disc_intersector.h b/thirdparty/embree-aarch64/kernels/geometry/disc_intersector.h new file mode 100644 index 0000000000..e8305780e5 --- /dev/null +++ b/thirdparty/embree-aarch64/kernels/geometry/disc_intersector.h @@ -0,0 +1,216 @@ +// Copyright 2009-2020 Intel Corporation +// SPDX-License-Identifier: Apache-2.0 + +#pragma once + +#include "../common/ray.h" +#include "../common/scene_points.h" +#include "curve_intersector_precalculations.h" + +namespace embree +{ + namespace isa + { + template + struct DiscIntersectorHitM + { + __forceinline DiscIntersectorHitM() {} + + __forceinline DiscIntersectorHitM(const vfloat& u, const vfloat& v, const vfloat& t, const Vec3vf& Ng) + : vu(u), vv(v), vt(t), vNg(Ng) + { + } + + __forceinline void finalize() {} + + __forceinline Vec2f uv(const size_t i) const + { + return Vec2f(vu[i], vv[i]); + } + __forceinline float t(const size_t i) const + { + return vt[i]; + } + __forceinline Vec3fa Ng(const size_t i) const + { + return Vec3fa(vNg.x[i], vNg.y[i], vNg.z[i]); + } + + public: + vfloat vu; + vfloat vv; + vfloat vt; + Vec3vf vNg; + }; + + template + struct DiscIntersector1 + { + typedef CurvePrecalculations1 Precalculations; + + template + static __forceinline bool intersect( + const vbool& valid_i, + Ray& ray, + IntersectContext* context, + const Points* geom, + const Precalculations& pre, + const Vec4vf& v0i, + const Epilog& epilog) + { + vbool valid = valid_i; + + const Vec3vf ray_org(ray.org.x, ray.org.y, ray.org.z); + const Vec3vf ray_dir(ray.dir.x, ray.dir.y, ray.dir.z); + const vfloat rd2 = rcp(dot(ray_dir, ray_dir)); + + const Vec4vf v0 = enlargeRadiusToMinWidth(context,geom,ray_org,v0i); + const Vec3vf center = v0.xyz(); + const vfloat radius = v0.w; + + const Vec3vf c0 = center - ray_org; + const vfloat projC0 = dot(c0, ray_dir) * rd2; + + valid &= (vfloat(ray.tnear()) <= projC0) & (projC0 <= vfloat(ray.tfar)); + if (EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR != 0.0f) + valid &= projC0 > float(EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR) * radius * pre.depth_scale; // ignore self intersections + if (unlikely(none(valid))) + return false; + + const Vec3vf perp = c0 - projC0 * ray_dir; + const vfloat l2 = dot(perp, perp); + const vfloat r2 = radius * radius; + valid &= (l2 <= r2); + if (unlikely(none(valid))) + return false; + + DiscIntersectorHitM hit(zero, zero, projC0, -ray_dir); + return epilog(valid, hit); + } + + template + static __forceinline bool intersect(const vbool& valid_i, + Ray& ray, + IntersectContext* context, + const Points* geom, + const Precalculations& pre, + const Vec4vf& v0i, + const Vec3vf& normal, + const Epilog& epilog) + { + vbool valid = valid_i; + const Vec3vf ray_org(ray.org.x, ray.org.y, ray.org.z); + + const Vec4vf v0 = enlargeRadiusToMinWidth(context,geom,ray_org,v0i); + const Vec3vf center = v0.xyz(); + const vfloat radius = v0.w; + + vfloat divisor = dot(Vec3vf((Vec3fa)ray.dir), normal); + const vbool parallel = divisor == vfloat(0.f); + valid &= !parallel; + divisor = select(parallel, 1.f, divisor); // prevent divide by zero + + vfloat t = dot(center - Vec3vf((Vec3fa)ray.org), Vec3vf(normal)) / divisor; + + valid &= (vfloat(ray.tnear()) <= t) & (t <= vfloat(ray.tfar)); + if (unlikely(none(valid))) + return false; + + Vec3vf intersection = Vec3vf((Vec3fa)ray.org) + Vec3vf((Vec3fa)ray.dir) * t; + vfloat dist2 = dot(intersection - center, intersection - center); + valid &= dist2 < radius * radius; + if (unlikely(none(valid))) + return false; + + DiscIntersectorHitM hit(zero, zero, t, normal); + return epilog(valid, hit); + } + }; + + template + struct DiscIntersectorK + { + typedef CurvePrecalculationsK Precalculations; + + template + static __forceinline bool intersect(const vbool& valid_i, + RayK& ray, + size_t k, + IntersectContext* context, + const Points* geom, + const Precalculations& pre, + const Vec4vf& v0i, + const Epilog& epilog) + { + vbool valid = valid_i; + + const Vec3vf ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]); + const Vec3vf ray_dir(ray.dir.x[k], ray.dir.y[k], ray.dir.z[k]); + const vfloat rd2 = rcp(dot(ray_dir, ray_dir)); + + const Vec4vf v0 = enlargeRadiusToMinWidth(context,geom,ray_org,v0i); + const Vec3vf center = v0.xyz(); + const vfloat radius = v0.w; + + const Vec3vf c0 = center - ray_org; + const vfloat projC0 = dot(c0, ray_dir) * rd2; + + valid &= (vfloat(ray.tnear()[k]) <= projC0) & (projC0 <= vfloat(ray.tfar[k])); + if (EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR != 0.0f) + valid &= projC0 > float(EMBREE_CURVE_SELF_INTERSECTION_AVOIDANCE_FACTOR) * radius * pre.depth_scale[k]; // ignore self intersections + if (unlikely(none(valid))) + return false; + + const Vec3vf perp = c0 - projC0 * ray_dir; + const vfloat l2 = dot(perp, perp); + const vfloat r2 = radius * radius; + valid &= (l2 <= r2); + if (unlikely(none(valid))) + return false; + + DiscIntersectorHitM hit(zero, zero, projC0, -ray_dir); + return epilog(valid, hit); + } + + template + static __forceinline bool intersect(const vbool& valid_i, + RayK& ray, + size_t k, + IntersectContext* context, + const Points* geom, + const Precalculations& pre, + const Vec4vf& v0i, + const Vec3vf& normal, + const Epilog& epilog) + { + vbool valid = valid_i; + const Vec3vf ray_org(ray.org.x[k], ray.org.y[k], ray.org.z[k]); + const Vec3vf ray_dir(ray.dir.x[k], ray.dir.y[k], ray.dir.z[k]); + + const Vec4vf v0 = enlargeRadiusToMinWidth(context,geom,ray_org,v0i); + const Vec3vf center = v0.xyz(); + const vfloat radius = v0.w; + + vfloat divisor = dot(Vec3vf(ray_dir), normal); + const vbool parallel = divisor == vfloat(0.f); + valid &= !parallel; + divisor = select(parallel, 1.f, divisor); // prevent divide by zero + + vfloat t = dot(center - Vec3vf(ray_org), Vec3vf(normal)) / divisor; + + valid &= (vfloat(ray.tnear()[k]) <= t) & (t <= vfloat(ray.tfar[k])); + if (unlikely(none(valid))) + return false; + + Vec3vf intersection = Vec3vf(ray_org) + Vec3vf(ray_dir) * t; + vfloat dist2 = dot(intersection - center, intersection - center); + valid &= dist2 < radius * radius; + if (unlikely(none(valid))) + return false; + + DiscIntersectorHitM hit(zero, zero, t, normal); + return epilog(valid, hit); + } + }; + } // namespace isa +} // namespace embree -- cgit v1.2.3