/// <summary>
        /// Attempts to find a slightly better parameterization for u on the given curve.
        /// </summary>
        protected void Reparameterize(int first, int last, CubicBezier curve)
        {
#if OPTIMIZED_REPARAMETERIZE
            using var _ = ReparameterizeMarker.Auto();
            unsafe
            {
                using (_pts.ViewAsNativeArray(out var pts))
                    using (_u.ViewAsNativeArray(out var u))
                        OptimizedHelpers.Reparameterize(first, last, curve, (VECTOR *)pts.GetUnsafePtr(), (FLOAT *)u.GetUnsafePtr());
            }
#else
            using var _ = ReparameterizeMarker.Auto();
            List <VECTOR> pts  = _pts;
            List <FLOAT>  u    = _u;
            int           nPts = last - first;
            for (int i = 1; i < nPts; i++)
            {
                VECTOR p  = pts[first + i];
                FLOAT  t  = u[i];
                FLOAT  ti = 1 - t;

                // Control vertices for Q'
                VECTOR qp0 = (curve.p1 - curve.p0) * 3;
                VECTOR qp1 = (curve.p2 - curve.p1) * 3;
                VECTOR qp2 = (curve.p3 - curve.p2) * 3;

                // Control vertices for Q''
                VECTOR qpp0 = (qp1 - qp0) * 2;
                VECTOR qpp1 = (qp2 - qp1) * 2;

                // Evaluate Q(t), Q'(t), and Q''(t)
                VECTOR p0 = curve.Sample(t);
                VECTOR p1 = ((ti * ti) * qp0) + ((2 * ti * t) * qp1) + ((t * t) * qp2);
                VECTOR p2 = (ti * qpp0) + (t * qpp1);

                // these are the actual fitting calculations using http://en.wikipedia.org/wiki/Newton%27s_method
                // We can't just use .X and .Y because Unity uses lower-case "x" and "y".
                FLOAT num  = ((VectorHelper.GetX(p0) - VectorHelper.GetX(p)) * VectorHelper.GetX(p1)) + ((VectorHelper.GetY(p0) - VectorHelper.GetY(p)) * VectorHelper.GetY(p1));
                FLOAT den  = (VectorHelper.GetX(p1) * VectorHelper.GetX(p1)) + (VectorHelper.GetY(p1) * VectorHelper.GetY(p1)) + ((VectorHelper.GetX(p0) - VectorHelper.GetX(p)) * VectorHelper.GetX(p2)) + ((VectorHelper.GetY(p0) - VectorHelper.GetY(p)) * VectorHelper.GetY(p2));
                FLOAT newU = t - num / den;
                if (Math.Abs(den) > EPSILON && newU >= 0 && newU <= 1)
                {
                    u[i] = newU;
                }
            }
#endif
        }
        /// <summary>
        /// Computes the maximum squared distance from a point to the curve using the current parameterization.
        /// </summary>
        protected FLOAT FindMaxSquaredError(int first, int last, CubicBezier curve, out int split)
        {
#if OPTIMIZED_FINDMAXSQUAREDERROR
            using var _ = FindMaxSquaredErrorMarker.Auto();
            unsafe
            {
                using (_pts.ViewAsNativeArray(out var pts))
                    using (_u.ViewAsNativeArray(out var u))
                    {
                        OptimizedHelpers.FindMaxSquaredError(first, last, curve, out split, (VECTOR *)pts.GetUnsafePtr(), (FLOAT *)u.GetUnsafePtr(), out var max);
                        return(max);
                    }
            }
#else
            using var _ = FindMaxSquaredErrorMarker.Auto();
            List <VECTOR> pts  = _pts;
            List <FLOAT>  u    = _u;
            int           s    = (last - first + 1) / 2;
            int           nPts = last - first + 1;
            FLOAT         max  = 0;
            for (int i = 1; i < nPts; i++)
            {
                VECTOR v0 = pts[first + i];
                VECTOR v1 = curve.Sample(u[i]);
                FLOAT  d  = VectorHelper.DistanceSquared(v0, v1);
                if (d > max)
                {
                    max = d;
                    s   = i;
                }
            }

            // split at point of maximum error
            split = s + first;
            if (split <= first)
            {
                split = first + 1;
            }
            if (split >= last)
            {
                split = last - 1;
            }

            return(max);
#endif
        }
        /// <summary>
        /// Gets the tangent for the start of the cure.
        /// </summary>
        public VECTOR GetLeftTangent(int last)
        {
#if OPTIMIZED_GETLEFTTANGENT
            using var _ = GetLeftTangentMarker.Auto();
            unsafe
            {
                using (_pts.ViewAsNativeArray(out var pts))
                    using (_arclen.ViewAsNativeArray(out var arclen))
                    {
                        OptimizedHelpers.GetLeftTangent(last, (VECTOR *)pts.GetUnsafePtr(), pts.Length, (FLOAT *)arclen.GetUnsafePtr(), arclen.Length, out var tanL);
                        return(tanL);
                    }
            }
#else
            using var _ = GetLeftTangentMarker.Auto();
            List <VECTOR> pts         = _pts;
            List <FLOAT>  arclen      = _arclen;
            FLOAT         totalLen    = arclen[arclen.Count - 1];
            VECTOR        p0          = pts[0];
            VECTOR        tanL        = VectorHelper.Normalize(pts[1] - p0);
            VECTOR        total       = tanL;
            FLOAT         weightTotal = 1;
            last = Math.Min(END_TANGENT_N_PTS, last - 1);
            for (int i = 2; i <= last; i++)
            {
                FLOAT  ti     = 1 - (arclen[i] / totalLen);
                FLOAT  weight = ti * ti * ti;
                VECTOR v      = VectorHelper.Normalize(pts[i] - p0);
                total       += v * weight;
                weightTotal += weight;
            }
            // if the vectors add up to zero (ie going opposite directions), there's no way to normalize them
            if (VectorHelper.Length(total) > EPSILON)
            {
                tanL = VectorHelper.Normalize(total / weightTotal);
            }
            return(tanL);
#endif
        }
        /// <summary>
        /// Gets the tangent for the the end of the curve.
        /// </summary>
        public VECTOR GetRightTangent(int first)
        {
#if OPTIMIZED_GETRIGHTTANGENT
            using var _ = GetRightTangentMarker.Auto();
            unsafe
            {
                using (_pts.ViewAsNativeArray(out var pts))
                    using (_arclen.ViewAsNativeArray(out var arclen))
                    {
                        OptimizedHelpers.GetRightTangent(first, (VECTOR *)pts.GetUnsafePtr(), pts.Length, (FLOAT *)arclen.GetUnsafePtr(), arclen.Length, out var tanR);
                        return(tanR);
                    }
            }
#else
            using var _ = GetRightTangentMarker.Auto();
            List <VECTOR> pts         = _pts;
            List <FLOAT>  arclen      = _arclen;
            FLOAT         totalLen    = arclen[arclen.Count - 1];
            VECTOR        p3          = pts[pts.Count - 1];
            VECTOR        tanR        = VectorHelper.Normalize(pts[pts.Count - 2] - p3);
            VECTOR        total       = tanR;
            FLOAT         weightTotal = 1;
            first = Math.Max(pts.Count - (END_TANGENT_N_PTS + 1), first + 1);
            for (int i = pts.Count - 3; i >= first; i--)
            {
                FLOAT  t      = arclen[i] / totalLen;
                FLOAT  weight = t * t * t;
                VECTOR v      = VectorHelper.Normalize(pts[i] - p3);
                total       += v * weight;
                weightTotal += weight;
            }
            if (VectorHelper.Length(total) > EPSILON)
            {
                tanR = VectorHelper.Normalize(total / weightTotal);
            }
            return(tanR);
#endif
        }
        /// <summary>
        /// Generates a bezier curve for the segment using a least-squares approximation. for the derivation of this and why it works,
        /// see http://read.pudn.com/downloads141/ebook/610086/Graphics_Gems_I.pdf page 626 and beyond. tl;dr: math.
        /// </summary>
        protected CubicBezier GenerateBezier(int first, int last, VECTOR tanL, VECTOR tanR)
        {
#if OPTIMIZED_GENERATEBEZIERSIMD
            using var _ = GenerateBezierMarker.Auto();
            unsafe
            {
                using (_pts.ViewAsNativeArray(out var pts))
                    using (_u.ViewAsNativeArray(out var u))
                    {
                        OptimizedHelpers.GenerateBezier(first, last, tanL, tanR, (VECTOR *)pts.GetUnsafePtr(), (FLOAT *)u.GetUnsafePtr(), out var bezier);
                        return(bezier);
                    }
            }
#elif OPTIMIZED_GENERATEBEZIER
            using var _ = GenerateBezierMarker.Auto();
            unsafe
            {
                using (_pts.ViewAsNativeArray(out var pts))
                    using (_u.ViewAsNativeArray(out var u))
                    {
                        OptimizedHelpers.OptimizedGenerateBezier(first, last, tanL, tanR, (VECTOR *)pts.GetUnsafePtr(), (FLOAT *)u.GetUnsafePtr(), out var bezier);
                        return(bezier);
                    }
            }
#else
            using var _ = GenerateBezierMarker.Auto();
            List <VECTOR> pts = _pts;
            List <FLOAT>  u = _u;
            int           nPts = last - first + 1;
            VECTOR        p0 = pts[first], p3 = pts[last];           // first and last points of curve are actual points on data
            FLOAT         c00 = 0, c01 = 0, c11 = 0, x0 = 0, x1 = 0; // matrix members -- both C[0,1] and C[1,0] are the same, stored in c01
            for (int i = 1; i < nPts; i++)
            {
                // Calculate cubic bezier multipliers
                FLOAT t  = u[i];
                FLOAT ti = 1 - t;
                FLOAT t0 = ti * ti * ti;
                FLOAT t1 = 3 * ti * ti * t;
                FLOAT t2 = 3 * ti * t * t;
                FLOAT t3 = t * t * t;

                // For X matrix; moving this up here since profiling shows it's better up here (maybe a0/a1 not in registers vs only v not in regs)
                VECTOR s = (p0 * t0) + (p0 * t1) + (p3 * t2) + (p3 * t3); // NOTE: this would be Q(t) if p1=p0 and p2=p3
                VECTOR v = pts[first + i] - s;

                // C matrix
                VECTOR a0 = tanL * t1;
                VECTOR a1 = tanR * t2;
                c00 += VectorHelper.Dot(a0, a0);
                c01 += VectorHelper.Dot(a0, a1);
                c11 += VectorHelper.Dot(a1, a1);

                // X matrix
                x0 += VectorHelper.Dot(a0, v);
                x1 += VectorHelper.Dot(a1, v);
            }

            // determinents of X and C matrices
            FLOAT det_C0_C1 = c00 * c11 - c01 * c01;
            FLOAT det_C0_X  = c00 * x1 - c01 * x0;
            FLOAT det_X_C1  = x0 * c11 - x1 * c01;
            FLOAT alphaL    = det_X_C1 / det_C0_C1;
            FLOAT alphaR    = det_C0_X / det_C0_C1;

            // if alpha is negative, zero, or very small (or we can't trust it since C matrix is small), fall back to Wu/Barsky heuristic
            FLOAT linDist  = VectorHelper.Distance(p0, p3);
            FLOAT epsilon2 = EPSILON * linDist;
            if (Math.Abs(det_C0_C1) < EPSILON || alphaL < epsilon2 || alphaR < epsilon2)
            {
                FLOAT  alpha = linDist / 3;
                VECTOR p1    = (tanL * alpha) + p0;
                VECTOR p2    = (tanR * alpha) + p3;
                return(new CubicBezier(p0, p1, p2, p3));
            }
            else
            {
                VECTOR p1 = (tanL * alphaL) + p0;
                VECTOR p2 = (tanR * alphaR) + p3;
                return(new CubicBezier(p0, p1, p2, p3));
            }
#endif
        }