warp-lang 1.0.1__py3-none-manylinux2014_aarch64.whl → 1.1.0__py3-none-manylinux2014_aarch64.whl

warp/native/solid_angle.h

@@ -1,442 +1,442 @@
-// This code is adapted from https://github.com/alecjacobson/WindingNumber/tree/1e6081e52905575d8e98fb8b7c0921274a18752f
-// The original license is below:
-/*
-MIT License
-
-Copyright (c) 2018 Side Effects Software Inc.
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
-*/
-namespace wp
-{
-
-class SolidAngleProps{
-public:
-	vec3 average_p;
-	vec3 normal;
-	
-	vec3 n_ij_diag;
-    vec3 n_ijk_diag; 
-
-    float sum_permute_n_xyz; 
-    float two_n_xxy_n_yxx;
-    float two_n_xxz_n_zxx;
-    float two_n_yyz_n_zyy;
-    float two_n_yyx_n_xyy;
-    float two_n_zzx_n_xzz;
-    float two_n_zzy_n_yzz; 
-
-    float n_xy; float n_yx;
-    float n_yz; float n_zy;
-    float n_zx; float n_xz;	
-	
-	bounds3 box;
-	vec3 area_P;
-	float area;	
-	float max_p_dist_sq;
-};
-
-CUDA_CALLABLE inline void compute_integrals(
-	const vec3 &a,
-	const vec3 &b,
-	const vec3 &c,
-	const vec3 &P,
-	float *integral_ii,
-	float *integral_ij,
-	float *integral_ik,
-	const int i)
-{
-
-	// NOTE: a, b, and c must be in order of the i axis.
-	// We're splitting the triangle at the middle i coordinate.
-	const vec3 oab = b - a;
-	const vec3 oac = c - a;
-	const vec3 ocb = b - c;
-	const float t = oab[i] / oac[i];
-
-	const int j = (i == 2) ? 0 : (i + 1);
-	const int k = (j == 2) ? 0 : (j + 1);
-	const float jdiff = t * oac[j] - oab[j];
-	const float kdiff = t * oac[k] - oab[k];
-	vec3 cross_a;
-	cross_a[0] = (jdiff * oab[k] - kdiff * oab[j]);
-	cross_a[1] = kdiff * oab[i];
-	cross_a[2] = jdiff * oab[i];
-	vec3 cross_c;
-	cross_c[0] = (jdiff * ocb[k] - kdiff * ocb[j]);
-	cross_c[1] = kdiff * ocb[i];
-	cross_c[2] = jdiff * ocb[i];
-	const float area_scale_a = length(cross_a);
-	const float area_scale_c = length(cross_c);
-	const float Pai = a[i] - P[i];
-	const float Pci = c[i] - P[i];
-
-	// Integral over the area of the triangle of (pi^2)dA,
-	// by splitting the triangle into two at b, the a side
-	// and the c side.
-	const float int_ii_a = area_scale_a * (0.5f * Pai * Pai + (2.0f / 3.0f) * Pai * oab[i] + 0.25f * oab[i] * oab[i]);
-	const float int_ii_c = area_scale_c * (0.5f * Pci * Pci + (2.0f / 3.0f) * Pci * ocb[i] + 0.25f * ocb[i] * ocb[i]);
-	*integral_ii = int_ii_a + int_ii_c;
-
-	int jk = j;
-	float *integral = integral_ij;
-	float diff = jdiff;
-	while (true) // This only does 2 iterations, one for j and one for k
-	{
-		if (integral)
-		{
-			float obmidj = b[jk] + 0.5f * diff;
-			float oabmidj = obmidj - a[jk];
-			float ocbmidj = obmidj - c[jk];
-			float Paj = a[jk] - P[jk];
-			float Pcj = c[jk] - P[jk];
-			// Integral over the area of the triangle of (pi*pj)dA
-			const float int_ij_a = area_scale_a * (0.5f * Pai * Paj + (1.0f / 3.0f) * Pai * oabmidj + (1.0f / 3.0f) * Paj * oab[i] + 0.25f * oab[i] * oabmidj);
-			const float int_ij_c = area_scale_c * (0.5f * Pci * Pcj + (1.0f / 3.0f) * Pci * ocbmidj + (1.0f / 3.0f) * Pcj * ocb[i] + 0.25f * ocb[i] * ocbmidj);
-			*integral = int_ij_a + int_ij_c;
-		}
-		if (jk == k)
-			break;
-		jk = k;
-		integral = integral_ik;
-		diff = kdiff;
-	}
-};
-
-CUDA_CALLABLE inline void my_swap(int &a, int &b)
-{
-	int c = a;
-	a = b;
-	b = c;
-}
-
-CUDA_CALLABLE inline void precompute_triangle_solid_angle_props(const vec3 &a, const vec3 &b, const vec3 &c, SolidAngleProps &my_data)
-{
-	const vec3 ab = b - a;
-	const vec3 ac = c - a;
-
-	// Are weighted normal
-	const vec3 N = 0.5f * cross(ab, ac);
-	const float area2 = length_sq(N);
-	const float area = sqrtf(area2);
-	const vec3 P = (a + b + c) / 3.0f;
-	my_data.box.add_point(a);
-	my_data.box.add_point(b);
-	my_data.box.add_point(c);
-	my_data.average_p = P;
-	my_data.area_P = P * area;
-	my_data.normal = N;
-
-	my_data.area = area;
-
-	// NOTE: Due to P being at the centroid, triangles have Nij = 0
-	//       contributions to Nij.
-	my_data.n_ij_diag = 0.0f;
-	my_data.n_xy = 0.0f;
-	my_data.n_yx = 0.0f;
-	my_data.n_yz = 0.0f;
-	my_data.n_zy = 0.0f;
-	my_data.n_zx = 0.0f;
-	my_data.n_xz = 0.0f;
-
-	// If it's zero-length, the results are zero, so we can skip.
-	if (area == 0)
-	{
-		my_data.n_ijk_diag = 0.0f;
-		my_data.sum_permute_n_xyz = 0.0f;
-		my_data.two_n_xxy_n_yxx = 0.0f;
-		my_data.two_n_xxz_n_zxx = 0.0f;
-		my_data.two_n_yyz_n_zyy = 0.0f;
-		my_data.two_n_yyx_n_xyy = 0.0f;
-		my_data.two_n_zzx_n_xzz = 0.0f;
-		my_data.two_n_zzy_n_yzz = 0.0f;
-		return;
-	}
-
-	// We need to use the NORMALIZED normal to multiply the integrals by.
-	vec3 n = N / area;
-
-	// Figure out the order of a, b, and c in x, y, and z
-	// for use in computing the integrals for Nijk.
-	vec3 values[3] = {a, b, c};
-
-	int order_x[3] = {0, 1, 2};
-	if (a[0] > b[0])
-		my_swap(order_x[0], order_x[1]);
-	if (values[order_x[0]][0] > c[0])
-		my_swap(order_x[0], order_x[2]);
-	if (values[order_x[1]][0] > values[order_x[2]][0])
-		my_swap(order_x[1], order_x[2]);
-	float dx = values[order_x[2]][0] - values[order_x[0]][0];
-
-	int order_y[3] = {0, 1, 2};
-	if (a[1] > b[1])
-		my_swap(order_y[0], order_y[1]);
-	if (values[order_y[0]][1] > c[1])
-		my_swap(order_y[0], order_y[2]);
-	if (values[order_y[1]][1] > values[order_y[2]][1])
-		my_swap(order_y[1], order_y[2]);
-	float dy = values[order_y[2]][1] - values[order_y[0]][1];
-
-	int order_z[3] = {0, 1, 2};
-	if (a[2] > b[2])
-		my_swap(order_z[0], order_z[1]);
-	if (values[order_z[0]][2] > c[2])
-		my_swap(order_z[0], order_z[2]);
-	if (values[order_z[1]][2] > values[order_z[2]][2])
-		my_swap(order_z[1], order_z[2]);
-	float dz = values[order_z[2]][2] - values[order_z[0]][2];
-
-	float integral_xx = 0.0f;
-	float integral_xy = 0.0f;
-	float integral_yy = 0.0f;
-	float integral_yz = 0.0f;
-	float integral_zz = 0.0f;
-	float integral_zx = 0.0f;
-	// Note that if the span of any axis is zero, the integral must be zero,
-	// since there's a factor of (p_i-P_i), i.e. value minus average,
-	// and every value must be equal to the average, giving zero.
-	if (dx > 0)
-	{
-		compute_integrals(
-			values[order_x[0]], values[order_x[1]], values[order_x[2]], P,
-			&integral_xx, ((dx >= dy && dy > 0) ? &integral_xy : nullptr), ((dx >= dz && dz > 0) ? &integral_zx : nullptr), 0);
-	}
-	if (dy > 0)
-	{
-		compute_integrals(
-			values[order_y[0]], values[order_y[1]], values[order_y[2]], P,
-			&integral_yy, ((dy >= dz && dz > 0) ? &integral_yz : nullptr), ((dx < dy && dx > 0) ? &integral_xy : nullptr), 1);
-	}
-	if (dz > 0)
-	{
-		compute_integrals(
-			values[order_z[0]], values[order_z[1]], values[order_z[2]], P,
-			&integral_zz, ((dx < dz && dx > 0) ? &integral_zx : nullptr), ((dy < dz && dy > 0) ? &integral_yz : nullptr), 2);
-	}
-
-	vec3 Niii(integral_xx, integral_yy, integral_zz);
-	Niii = cw_mul(Niii, n);
-	my_data.n_ijk_diag = Niii;
-	my_data.sum_permute_n_xyz = 2.0f * (n[0] * integral_yz + n[1] * integral_zx + n[2] * integral_xy);
-	float Nxxy = n[0] * integral_xy;
-	float Nxxz = n[0] * integral_zx;
-	float Nyyz = n[1] * integral_yz;
-	float Nyyx = n[1] * integral_xy;
-	float Nzzx = n[2] * integral_zx;
-	float Nzzy = n[2] * integral_yz;
-	my_data.two_n_xxy_n_yxx = 2.0f * Nxxy + n[1] * integral_xx;
-	my_data.two_n_xxz_n_zxx = 2.0f * Nxxz + n[2] * integral_xx;
-	my_data.two_n_yyz_n_zyy = 2.0f * Nyyz + n[2] * integral_yy;
-	my_data.two_n_yyx_n_xyy = 2.0f * Nyyx + n[0] * integral_yy;
-	my_data.two_n_zzx_n_xzz = 2.0f * Nzzx + n[0] * integral_zz;
-	my_data.two_n_zzy_n_yzz = 2.0f * Nzzy + n[1] * integral_zz;
-}
-
-CUDA_CALLABLE inline void combine_precomputed_solid_angle_props(SolidAngleProps &my_data, const SolidAngleProps *left_child_data, const SolidAngleProps *right_child_data)
-{
-	vec3 N = left_child_data->normal;
-	vec3 areaP = left_child_data->area_P;
-	float area = left_child_data->area;
-	if (right_child_data)
-	{
-		const vec3 local_N = right_child_data->normal;
-		N += local_N;
-		areaP += right_child_data->area_P;
-		area += right_child_data->area;
-	}
-	my_data.normal = N;
-	my_data.area_P = areaP;
-	my_data.area = area;
-	bounds3 box(left_child_data->box);
-	if (right_child_data)
-	{
-		box = bounds_union(box, right_child_data->box);
-	}
-
-	// Normalize P
-	vec3 averageP;
-	if (area > 0)
-	{
-		averageP = areaP / area;
-	}
-	else
-	{
-		averageP = 0.5f * (box.lower + box.upper);
-	}
-	my_data.average_p = averageP;
-
-	my_data.box = box;
-
-	// We now have the current box's P, so we can adjust Nij and Nijk
-	my_data.n_ij_diag = left_child_data->n_ij_diag;
-	my_data.n_xy = 0.0f;
-	my_data.n_yx = 0.0f;
-	my_data.n_yz = 0.0f;
-	my_data.n_zy = 0.0f;
-	my_data.n_zx = 0.0f;
-	my_data.n_xz = 0.0f;
-
-	my_data.n_ijk_diag = left_child_data->n_ijk_diag;
-	my_data.sum_permute_n_xyz = left_child_data->sum_permute_n_xyz;
-	my_data.two_n_xxy_n_yxx = left_child_data->two_n_xxy_n_yxx;
-	my_data.two_n_xxz_n_zxx = left_child_data->two_n_xxz_n_zxx;
-	my_data.two_n_yyz_n_zyy = left_child_data->two_n_yyz_n_zyy;
-	my_data.two_n_yyx_n_xyy = left_child_data->two_n_yyx_n_xyy;
-	my_data.two_n_zzx_n_xzz = left_child_data->two_n_zzx_n_xzz;
-	my_data.two_n_zzy_n_yzz = left_child_data->two_n_zzy_n_yzz;
-
-	if (right_child_data)
-	{
-		my_data.n_ij_diag += right_child_data->n_ij_diag;
-		my_data.n_ijk_diag += right_child_data->n_ijk_diag;
-		my_data.sum_permute_n_xyz += right_child_data->sum_permute_n_xyz;
-		my_data.two_n_xxy_n_yxx += right_child_data->two_n_xxy_n_yxx;
-		my_data.two_n_xxz_n_zxx += right_child_data->two_n_xxz_n_zxx;
-		my_data.two_n_yyz_n_zyy += right_child_data->two_n_yyz_n_zyy;
-		my_data.two_n_yyx_n_xyy += right_child_data->two_n_yyx_n_xyy;
-		my_data.two_n_zzx_n_xzz += right_child_data->two_n_zzx_n_xzz;
-		my_data.two_n_zzy_n_yzz += right_child_data->two_n_zzy_n_yzz;
-	}
-
-	for (int i = 0; i < (right_child_data ? 2 : 1); ++i)
-	{
-		const SolidAngleProps &child_data = (i == 0) ? *left_child_data : *right_child_data;
-		vec3 displacement = child_data.average_p - vec3(my_data.average_p);
-		vec3 N = child_data.normal;
-
-		// Adjust Nij for the change in centre P
-		my_data.n_ij_diag += cw_mul(N, displacement);
-		float Nxy = child_data.n_xy + N[0] * displacement[1];
-		float Nyx = child_data.n_yx + N[1] * displacement[0];
-		float Nyz = child_data.n_yz + N[1] * displacement[2];
-		float Nzy = child_data.n_zy + N[2] * displacement[1];
-		float Nzx = child_data.n_zx + N[2] * displacement[0];
-		float Nxz = child_data.n_xz + N[0] * displacement[2];
-
-		my_data.n_xy += Nxy;
-		my_data.n_yx += Nyx;
-		my_data.n_yz += Nyz;
-		my_data.n_zy += Nzy;
-		my_data.n_zx += Nzx;
-		my_data.n_xz += Nxz;
-
-		// Adjust Nijk for the change in centre P
-		my_data.n_ijk_diag += 2.0f * cw_mul(displacement, child_data.n_ij_diag) + cw_mul(displacement, cw_mul(displacement, child_data.normal));
-		my_data.sum_permute_n_xyz += (displacement[0] * (Nyz + Nzy) + displacement[1] * (Nzx + Nxz) + displacement[2] * (Nxy + Nyx));
-		my_data.two_n_xxy_n_yxx +=
-			2 * (displacement[1] * child_data.n_ij_diag[0] + displacement[0] * child_data.n_xy + N[0] * displacement[0] * displacement[1]) + 2 * child_data.n_yx * displacement[0] + N[1] * displacement[0] * displacement[0];
-		my_data.two_n_xxz_n_zxx +=
-			2 * (displacement[2] * child_data.n_ij_diag[0] + displacement[0] * child_data.n_xz + N[0] * displacement[0] * displacement[2]) + 2 * child_data.n_zx * displacement[0] + N[2] * displacement[0] * displacement[0];
-		my_data.two_n_yyz_n_zyy +=
-			2 * (displacement[2] * child_data.n_ij_diag[1] + displacement[1] * child_data.n_yz + N[1] * displacement[1] * displacement[2]) + 2 * child_data.n_zy * displacement[1] + N[2] * displacement[1] * displacement[1];
-		my_data.two_n_yyx_n_xyy +=
-			2 * (displacement[0] * child_data.n_ij_diag[1] + displacement[1] * child_data.n_yx + N[1] * displacement[1] * displacement[0]) + 2 * child_data.n_xy * displacement[1] + N[0] * displacement[1] * displacement[1];
-		my_data.two_n_zzx_n_xzz +=
-			2 * (displacement[0] * child_data.n_ij_diag[2] + displacement[2] * child_data.n_zx + N[2] * displacement[2] * displacement[0]) + 2 * child_data.n_xz * displacement[2] + N[0] * displacement[2] * displacement[2];
-		my_data.two_n_zzy_n_yzz +=
-			2 * (displacement[1] * child_data.n_ij_diag[2] + displacement[2] * child_data.n_zy + N[2] * displacement[2] * displacement[1]) + 2 * child_data.n_yz * displacement[2] + N[1] * displacement[2] * displacement[2];
-	}
-
-	my_data.max_p_dist_sq = length_sq(max(my_data.average_p - my_data.box.lower, my_data.box.upper - my_data.average_p));
-}
-
-// Return whether need to
-CUDA_CALLABLE inline bool evaluate_node_solid_angle(const vec3 &query_point, SolidAngleProps *current_data, float &solid_angle, const float accuracy_scale_sq)
-{
-	SolidAngleProps &data = *current_data;
-	float max_p_sq = data.max_p_dist_sq;
-	vec3 q = query_point - data.average_p;
-	float qlength2 = length_sq(q);
-
-	if (qlength2 <= max_p_sq * accuracy_scale_sq)
-	{
-		solid_angle = 0.0f;
-		return true;
-	}
-	float omega_approx = 0.0f;
-	// qlength2 must be non-zero, since it's strictly greater than something.
-	// We still need to be careful for NaNs, though, because the 4th power might cause problems.
-	float qlength_m2 = 1.0f / qlength2;
-	float qlength_m1 = sqrtf(qlength_m2);
-
-	// Normalize q to reduce issues with overflow/underflow, since we'd need the 7th power
-	// if we didn't normalize, and (1e-6)^-7 = 1e42, which overflows single-precision.
-	q = q * qlength_m1;
-
-	omega_approx = -qlength_m2 * dot(q, data.normal);
-	vec3 q2 = cw_mul(q, q);
-
-	float qlength_m3 = qlength_m2 * qlength_m1;
-
-	float omega_1 = qlength_m3 * (data.n_ij_diag[0] + data.n_ij_diag[1] + data.n_ij_diag[2] - 3.0f * (dot(q2, data.n_ij_diag) + q[0] * q[1] * (data.n_xy + data.n_yx) + q[0] * q[2] * (data.n_zx + data.n_xz) + q[1] * q[2] * (data.n_yz + data.n_zy)));
-	omega_approx += omega_1;
-
-	vec3 q3 = cw_mul(q2, q);
-	float qlength_m4 = qlength_m2 * qlength_m2;
-	
-	vec3 temp0(data.two_n_yyx_n_xyy + data.two_n_zzx_n_xzz, data.two_n_zzy_n_yzz + data.two_n_xxy_n_yxx, data.two_n_xxz_n_zxx + data.two_n_yyz_n_zyy);
-	vec3 temp1(q[1] * data.two_n_xxy_n_yxx + q[2] * data.two_n_xxz_n_zxx, q[2] * data.two_n_yyz_n_zyy + q[0] * data.two_n_yyx_n_xyy, q[0] * data.two_n_zzx_n_xzz + q[1] * data.two_n_zzy_n_yzz);
-
-	float omega_2 = qlength_m4 * (1.5f * dot(q, 3.0f * data.n_ijk_diag + temp0) - 7.5f * (dot(q3, data.n_ijk_diag) + q[0] * q[1] * q[2] * data.sum_permute_n_xyz + dot(q2, temp1)));
-	omega_approx += omega_2;
-
-	// Safety check if not finite, need to descend instead
-	if (!isfinite(omega_approx))
-	{
-		omega_approx = 0.0f;
-		solid_angle = 0.0;
-		return true;
-	}
-
-	solid_angle = omega_approx;
-	return false;
-}
-
-CUDA_CALLABLE inline float robust_solid_angle(
-	const vec3 &a,
-	const vec3 &b,
-	const vec3 &c,
-	const vec3 &p)
-{
-	vec3 qa = a - p;
-	vec3 qb = b - p;
-	vec3 qc = c - p;
-
-	const float a_length = length(qa);
-	const float b_length = length(qb);
-	const float c_length = length(qc);
-
-	if (a_length == 0.0f || b_length == 0.0f || c_length == 0.0f)
-		return 0.0f;
-
-	qa = qa / a_length;
-	qb = qb / b_length;
-	qc = qc / c_length;
-
-	const float numerator = dot(qa, cross(qb - qa, qc - qa));
-
-	if (numerator == 0.0f)
-		return 0.0f;
-
-	const float denominator = 1.0f + dot(qa, qb) + dot(qa, qc) + dot(qb, qc);
-
-	return 2.0f * atan2(numerator, denominator);
-}
-}
+// This code is adapted from https://github.com/alecjacobson/WindingNumber/tree/1e6081e52905575d8e98fb8b7c0921274a18752f
+// The original license is below:
+/*
+MIT License
+
+Copyright (c) 2018 Side Effects Software Inc.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+*/
+namespace wp
+{
+
+class SolidAngleProps{
+public:
+	vec3 average_p;
+	vec3 normal;
+	
+	vec3 n_ij_diag;
+    vec3 n_ijk_diag; 
+
+    float sum_permute_n_xyz; 
+    float two_n_xxy_n_yxx;
+    float two_n_xxz_n_zxx;
+    float two_n_yyz_n_zyy;
+    float two_n_yyx_n_xyy;
+    float two_n_zzx_n_xzz;
+    float two_n_zzy_n_yzz; 
+
+    float n_xy; float n_yx;
+    float n_yz; float n_zy;
+    float n_zx; float n_xz;	
+	
+	bounds3 box;
+	vec3 area_P;
+	float area;	
+	float max_p_dist_sq;
+};
+
+CUDA_CALLABLE inline void compute_integrals(
+	const vec3 &a,
+	const vec3 &b,
+	const vec3 &c,
+	const vec3 &P,
+	float *integral_ii,
+	float *integral_ij,
+	float *integral_ik,
+	const int i)
+{
+
+	// NOTE: a, b, and c must be in order of the i axis.
+	// We're splitting the triangle at the middle i coordinate.
+	const vec3 oab = b - a;
+	const vec3 oac = c - a;
+	const vec3 ocb = b - c;
+	const float t = oab[i] / oac[i];
+
+	const int j = (i == 2) ? 0 : (i + 1);
+	const int k = (j == 2) ? 0 : (j + 1);
+	const float jdiff = t * oac[j] - oab[j];
+	const float kdiff = t * oac[k] - oab[k];
+	vec3 cross_a;
+	cross_a[0] = (jdiff * oab[k] - kdiff * oab[j]);
+	cross_a[1] = kdiff * oab[i];
+	cross_a[2] = jdiff * oab[i];
+	vec3 cross_c;
+	cross_c[0] = (jdiff * ocb[k] - kdiff * ocb[j]);
+	cross_c[1] = kdiff * ocb[i];
+	cross_c[2] = jdiff * ocb[i];
+	const float area_scale_a = length(cross_a);
+	const float area_scale_c = length(cross_c);
+	const float Pai = a[i] - P[i];
+	const float Pci = c[i] - P[i];
+
+	// Integral over the area of the triangle of (pi^2)dA,
+	// by splitting the triangle into two at b, the a side
+	// and the c side.
+	const float int_ii_a = area_scale_a * (0.5f * Pai * Pai + (2.0f / 3.0f) * Pai * oab[i] + 0.25f * oab[i] * oab[i]);
+	const float int_ii_c = area_scale_c * (0.5f * Pci * Pci + (2.0f / 3.0f) * Pci * ocb[i] + 0.25f * ocb[i] * ocb[i]);
+	*integral_ii = int_ii_a + int_ii_c;
+
+	int jk = j;
+	float *integral = integral_ij;
+	float diff = jdiff;
+	while (true) // This only does 2 iterations, one for j and one for k
+	{
+		if (integral)
+		{
+			float obmidj = b[jk] + 0.5f * diff;
+			float oabmidj = obmidj - a[jk];
+			float ocbmidj = obmidj - c[jk];
+			float Paj = a[jk] - P[jk];
+			float Pcj = c[jk] - P[jk];
+			// Integral over the area of the triangle of (pi*pj)dA
+			const float int_ij_a = area_scale_a * (0.5f * Pai * Paj + (1.0f / 3.0f) * Pai * oabmidj + (1.0f / 3.0f) * Paj * oab[i] + 0.25f * oab[i] * oabmidj);
+			const float int_ij_c = area_scale_c * (0.5f * Pci * Pcj + (1.0f / 3.0f) * Pci * ocbmidj + (1.0f / 3.0f) * Pcj * ocb[i] + 0.25f * ocb[i] * ocbmidj);
+			*integral = int_ij_a + int_ij_c;
+		}
+		if (jk == k)
+			break;
+		jk = k;
+		integral = integral_ik;
+		diff = kdiff;
+	}
+};
+
+CUDA_CALLABLE inline void my_swap(int &a, int &b)
+{
+	int c = a;
+	a = b;
+	b = c;
+}
+
+CUDA_CALLABLE inline void precompute_triangle_solid_angle_props(const vec3 &a, const vec3 &b, const vec3 &c, SolidAngleProps &my_data)
+{
+	const vec3 ab = b - a;
+	const vec3 ac = c - a;
+
+	// Are weighted normal
+	const vec3 N = 0.5f * cross(ab, ac);
+	const float area2 = length_sq(N);
+	const float area = sqrtf(area2);
+	const vec3 P = (a + b + c) / 3.0f;
+	my_data.box.add_point(a);
+	my_data.box.add_point(b);
+	my_data.box.add_point(c);
+	my_data.average_p = P;
+	my_data.area_P = P * area;
+	my_data.normal = N;
+
+	my_data.area = area;
+
+	// NOTE: Due to P being at the centroid, triangles have Nij = 0
+	//       contributions to Nij.
+	my_data.n_ij_diag = 0.0f;
+	my_data.n_xy = 0.0f;
+	my_data.n_yx = 0.0f;
+	my_data.n_yz = 0.0f;
+	my_data.n_zy = 0.0f;
+	my_data.n_zx = 0.0f;
+	my_data.n_xz = 0.0f;
+
+	// If it's zero-length, the results are zero, so we can skip.
+	if (area == 0)
+	{
+		my_data.n_ijk_diag = 0.0f;
+		my_data.sum_permute_n_xyz = 0.0f;
+		my_data.two_n_xxy_n_yxx = 0.0f;
+		my_data.two_n_xxz_n_zxx = 0.0f;
+		my_data.two_n_yyz_n_zyy = 0.0f;
+		my_data.two_n_yyx_n_xyy = 0.0f;
+		my_data.two_n_zzx_n_xzz = 0.0f;
+		my_data.two_n_zzy_n_yzz = 0.0f;
+		return;
+	}
+
+	// We need to use the NORMALIZED normal to multiply the integrals by.
+	vec3 n = N / area;
+
+	// Figure out the order of a, b, and c in x, y, and z
+	// for use in computing the integrals for Nijk.
+	vec3 values[3] = {a, b, c};
+
+	int order_x[3] = {0, 1, 2};
+	if (a[0] > b[0])
+		my_swap(order_x[0], order_x[1]);
+	if (values[order_x[0]][0] > c[0])
+		my_swap(order_x[0], order_x[2]);
+	if (values[order_x[1]][0] > values[order_x[2]][0])
+		my_swap(order_x[1], order_x[2]);
+	float dx = values[order_x[2]][0] - values[order_x[0]][0];
+
+	int order_y[3] = {0, 1, 2};
+	if (a[1] > b[1])
+		my_swap(order_y[0], order_y[1]);
+	if (values[order_y[0]][1] > c[1])
+		my_swap(order_y[0], order_y[2]);
+	if (values[order_y[1]][1] > values[order_y[2]][1])
+		my_swap(order_y[1], order_y[2]);
+	float dy = values[order_y[2]][1] - values[order_y[0]][1];
+
+	int order_z[3] = {0, 1, 2};
+	if (a[2] > b[2])
+		my_swap(order_z[0], order_z[1]);
+	if (values[order_z[0]][2] > c[2])
+		my_swap(order_z[0], order_z[2]);
+	if (values[order_z[1]][2] > values[order_z[2]][2])
+		my_swap(order_z[1], order_z[2]);
+	float dz = values[order_z[2]][2] - values[order_z[0]][2];
+
+	float integral_xx = 0.0f;
+	float integral_xy = 0.0f;
+	float integral_yy = 0.0f;
+	float integral_yz = 0.0f;
+	float integral_zz = 0.0f;
+	float integral_zx = 0.0f;
+	// Note that if the span of any axis is zero, the integral must be zero,
+	// since there's a factor of (p_i-P_i), i.e. value minus average,
+	// and every value must be equal to the average, giving zero.
+	if (dx > 0)
+	{
+		compute_integrals(
+			values[order_x[0]], values[order_x[1]], values[order_x[2]], P,
+			&integral_xx, ((dx >= dy && dy > 0) ? &integral_xy : nullptr), ((dx >= dz && dz > 0) ? &integral_zx : nullptr), 0);
+	}
+	if (dy > 0)
+	{
+		compute_integrals(
+			values[order_y[0]], values[order_y[1]], values[order_y[2]], P,
+			&integral_yy, ((dy >= dz && dz > 0) ? &integral_yz : nullptr), ((dx < dy && dx > 0) ? &integral_xy : nullptr), 1);
+	}
+	if (dz > 0)
+	{
+		compute_integrals(
+			values[order_z[0]], values[order_z[1]], values[order_z[2]], P,
+			&integral_zz, ((dx < dz && dx > 0) ? &integral_zx : nullptr), ((dy < dz && dy > 0) ? &integral_yz : nullptr), 2);
+	}
+
+	vec3 Niii(integral_xx, integral_yy, integral_zz);
+	Niii = cw_mul(Niii, n);
+	my_data.n_ijk_diag = Niii;
+	my_data.sum_permute_n_xyz = 2.0f * (n[0] * integral_yz + n[1] * integral_zx + n[2] * integral_xy);
+	float Nxxy = n[0] * integral_xy;
+	float Nxxz = n[0] * integral_zx;
+	float Nyyz = n[1] * integral_yz;
+	float Nyyx = n[1] * integral_xy;
+	float Nzzx = n[2] * integral_zx;
+	float Nzzy = n[2] * integral_yz;
+	my_data.two_n_xxy_n_yxx = 2.0f * Nxxy + n[1] * integral_xx;
+	my_data.two_n_xxz_n_zxx = 2.0f * Nxxz + n[2] * integral_xx;
+	my_data.two_n_yyz_n_zyy = 2.0f * Nyyz + n[2] * integral_yy;
+	my_data.two_n_yyx_n_xyy = 2.0f * Nyyx + n[0] * integral_yy;
+	my_data.two_n_zzx_n_xzz = 2.0f * Nzzx + n[0] * integral_zz;
+	my_data.two_n_zzy_n_yzz = 2.0f * Nzzy + n[1] * integral_zz;
+}
+
+CUDA_CALLABLE inline void combine_precomputed_solid_angle_props(SolidAngleProps &my_data, const SolidAngleProps *left_child_data, const SolidAngleProps *right_child_data)
+{
+	vec3 N = left_child_data->normal;
+	vec3 areaP = left_child_data->area_P;
+	float area = left_child_data->area;
+	if (right_child_data)
+	{
+		const vec3 local_N = right_child_data->normal;
+		N += local_N;
+		areaP += right_child_data->area_P;
+		area += right_child_data->area;
+	}
+	my_data.normal = N;
+	my_data.area_P = areaP;
+	my_data.area = area;
+	bounds3 box(left_child_data->box);
+	if (right_child_data)
+	{
+		box = bounds_union(box, right_child_data->box);
+	}
+
+	// Normalize P
+	vec3 averageP;
+	if (area > 0)
+	{
+		averageP = areaP / area;
+	}
+	else
+	{
+		averageP = 0.5f * (box.lower + box.upper);
+	}
+	my_data.average_p = averageP;
+
+	my_data.box = box;
+
+	// We now have the current box's P, so we can adjust Nij and Nijk
+	my_data.n_ij_diag = left_child_data->n_ij_diag;
+	my_data.n_xy = 0.0f;
+	my_data.n_yx = 0.0f;
+	my_data.n_yz = 0.0f;
+	my_data.n_zy = 0.0f;
+	my_data.n_zx = 0.0f;
+	my_data.n_xz = 0.0f;
+
+	my_data.n_ijk_diag = left_child_data->n_ijk_diag;
+	my_data.sum_permute_n_xyz = left_child_data->sum_permute_n_xyz;
+	my_data.two_n_xxy_n_yxx = left_child_data->two_n_xxy_n_yxx;
+	my_data.two_n_xxz_n_zxx = left_child_data->two_n_xxz_n_zxx;
+	my_data.two_n_yyz_n_zyy = left_child_data->two_n_yyz_n_zyy;
+	my_data.two_n_yyx_n_xyy = left_child_data->two_n_yyx_n_xyy;
+	my_data.two_n_zzx_n_xzz = left_child_data->two_n_zzx_n_xzz;
+	my_data.two_n_zzy_n_yzz = left_child_data->two_n_zzy_n_yzz;
+
+	if (right_child_data)
+	{
+		my_data.n_ij_diag += right_child_data->n_ij_diag;
+		my_data.n_ijk_diag += right_child_data->n_ijk_diag;
+		my_data.sum_permute_n_xyz += right_child_data->sum_permute_n_xyz;
+		my_data.two_n_xxy_n_yxx += right_child_data->two_n_xxy_n_yxx;
+		my_data.two_n_xxz_n_zxx += right_child_data->two_n_xxz_n_zxx;
+		my_data.two_n_yyz_n_zyy += right_child_data->two_n_yyz_n_zyy;
+		my_data.two_n_yyx_n_xyy += right_child_data->two_n_yyx_n_xyy;
+		my_data.two_n_zzx_n_xzz += right_child_data->two_n_zzx_n_xzz;
+		my_data.two_n_zzy_n_yzz += right_child_data->two_n_zzy_n_yzz;
+	}
+
+	for (int i = 0; i < (right_child_data ? 2 : 1); ++i)
+	{
+		const SolidAngleProps &child_data = (i == 0) ? *left_child_data : *right_child_data;
+		vec3 displacement = child_data.average_p - vec3(my_data.average_p);
+		vec3 N = child_data.normal;
+
+		// Adjust Nij for the change in centre P
+		my_data.n_ij_diag += cw_mul(N, displacement);
+		float Nxy = child_data.n_xy + N[0] * displacement[1];
+		float Nyx = child_data.n_yx + N[1] * displacement[0];
+		float Nyz = child_data.n_yz + N[1] * displacement[2];
+		float Nzy = child_data.n_zy + N[2] * displacement[1];
+		float Nzx = child_data.n_zx + N[2] * displacement[0];
+		float Nxz = child_data.n_xz + N[0] * displacement[2];
+
+		my_data.n_xy += Nxy;
+		my_data.n_yx += Nyx;
+		my_data.n_yz += Nyz;
+		my_data.n_zy += Nzy;
+		my_data.n_zx += Nzx;
+		my_data.n_xz += Nxz;
+
+		// Adjust Nijk for the change in centre P
+		my_data.n_ijk_diag += 2.0f * cw_mul(displacement, child_data.n_ij_diag) + cw_mul(displacement, cw_mul(displacement, child_data.normal));
+		my_data.sum_permute_n_xyz += (displacement[0] * (Nyz + Nzy) + displacement[1] * (Nzx + Nxz) + displacement[2] * (Nxy + Nyx));
+		my_data.two_n_xxy_n_yxx +=
+			2 * (displacement[1] * child_data.n_ij_diag[0] + displacement[0] * child_data.n_xy + N[0] * displacement[0] * displacement[1]) + 2 * child_data.n_yx * displacement[0] + N[1] * displacement[0] * displacement[0];
+		my_data.two_n_xxz_n_zxx +=
+			2 * (displacement[2] * child_data.n_ij_diag[0] + displacement[0] * child_data.n_xz + N[0] * displacement[0] * displacement[2]) + 2 * child_data.n_zx * displacement[0] + N[2] * displacement[0] * displacement[0];
+		my_data.two_n_yyz_n_zyy +=
+			2 * (displacement[2] * child_data.n_ij_diag[1] + displacement[1] * child_data.n_yz + N[1] * displacement[1] * displacement[2]) + 2 * child_data.n_zy * displacement[1] + N[2] * displacement[1] * displacement[1];
+		my_data.two_n_yyx_n_xyy +=
+			2 * (displacement[0] * child_data.n_ij_diag[1] + displacement[1] * child_data.n_yx + N[1] * displacement[1] * displacement[0]) + 2 * child_data.n_xy * displacement[1] + N[0] * displacement[1] * displacement[1];
+		my_data.two_n_zzx_n_xzz +=
+			2 * (displacement[0] * child_data.n_ij_diag[2] + displacement[2] * child_data.n_zx + N[2] * displacement[2] * displacement[0]) + 2 * child_data.n_xz * displacement[2] + N[0] * displacement[2] * displacement[2];
+		my_data.two_n_zzy_n_yzz +=
+			2 * (displacement[1] * child_data.n_ij_diag[2] + displacement[2] * child_data.n_zy + N[2] * displacement[2] * displacement[1]) + 2 * child_data.n_yz * displacement[2] + N[1] * displacement[2] * displacement[2];
+	}
+
+	my_data.max_p_dist_sq = length_sq(max(my_data.average_p - my_data.box.lower, my_data.box.upper - my_data.average_p));
+}
+
+// Return whether need to
+CUDA_CALLABLE inline bool evaluate_node_solid_angle(const vec3 &query_point, SolidAngleProps *current_data, float &solid_angle, const float accuracy_scale_sq)
+{
+	SolidAngleProps &data = *current_data;
+	float max_p_sq = data.max_p_dist_sq;
+	vec3 q = query_point - data.average_p;
+	float qlength2 = length_sq(q);
+
+	if (qlength2 <= max_p_sq * accuracy_scale_sq)
+	{
+		solid_angle = 0.0f;
+		return true;
+	}
+	float omega_approx = 0.0f;
+	// qlength2 must be non-zero, since it's strictly greater than something.
+	// We still need to be careful for NaNs, though, because the 4th power might cause problems.
+	float qlength_m2 = 1.0f / qlength2;
+	float qlength_m1 = sqrtf(qlength_m2);
+
+	// Normalize q to reduce issues with overflow/underflow, since we'd need the 7th power
+	// if we didn't normalize, and (1e-6)^-7 = 1e42, which overflows single-precision.
+	q = q * qlength_m1;
+
+	omega_approx = -qlength_m2 * dot(q, data.normal);
+	vec3 q2 = cw_mul(q, q);
+
+	float qlength_m3 = qlength_m2 * qlength_m1;
+
+	float omega_1 = qlength_m3 * (data.n_ij_diag[0] + data.n_ij_diag[1] + data.n_ij_diag[2] - 3.0f * (dot(q2, data.n_ij_diag) + q[0] * q[1] * (data.n_xy + data.n_yx) + q[0] * q[2] * (data.n_zx + data.n_xz) + q[1] * q[2] * (data.n_yz + data.n_zy)));
+	omega_approx += omega_1;
+
+	vec3 q3 = cw_mul(q2, q);
+	float qlength_m4 = qlength_m2 * qlength_m2;
+	
+	vec3 temp0(data.two_n_yyx_n_xyy + data.two_n_zzx_n_xzz, data.two_n_zzy_n_yzz + data.two_n_xxy_n_yxx, data.two_n_xxz_n_zxx + data.two_n_yyz_n_zyy);
+	vec3 temp1(q[1] * data.two_n_xxy_n_yxx + q[2] * data.two_n_xxz_n_zxx, q[2] * data.two_n_yyz_n_zyy + q[0] * data.two_n_yyx_n_xyy, q[0] * data.two_n_zzx_n_xzz + q[1] * data.two_n_zzy_n_yzz);
+
+	float omega_2 = qlength_m4 * (1.5f * dot(q, 3.0f * data.n_ijk_diag + temp0) - 7.5f * (dot(q3, data.n_ijk_diag) + q[0] * q[1] * q[2] * data.sum_permute_n_xyz + dot(q2, temp1)));
+	omega_approx += omega_2;
+
+	// Safety check if not finite, need to descend instead
+	if (!isfinite(omega_approx))
+	{
+		omega_approx = 0.0f;
+		solid_angle = 0.0;
+		return true;
+	}
+
+	solid_angle = omega_approx;
+	return false;
+}
+
+CUDA_CALLABLE inline float robust_solid_angle(
+	const vec3 &a,
+	const vec3 &b,
+	const vec3 &c,
+	const vec3 &p)
+{
+	vec3 qa = a - p;
+	vec3 qb = b - p;
+	vec3 qc = c - p;
+
+	const float a_length = length(qa);
+	const float b_length = length(qb);
+	const float c_length = length(qc);
+
+	if (a_length == 0.0f || b_length == 0.0f || c_length == 0.0f)
+		return 0.0f;
+
+	qa = qa / a_length;
+	qb = qb / b_length;
+	qc = qc / c_length;
+
+	const float numerator = dot(qa, cross(qb - qa, qc - qa));
+
+	if (numerator == 0.0f)
+		return 0.0f;
+
+	const float denominator = 1.0f + dot(qa, qb) + dot(qa, qc) + dot(qb, qc);
+
+	return 2.0f * atan2(numerator, denominator);
+}
+}