camera.cpp

Go to the documentation of this file.

#include "camera.h"
#include "base/basetypes.hpp"
 
#include <graphics/graphics.h>
 
#include <algorithm>
#include <limits>
 
namespace unravel
{
auto camera::get_zoom_factor() const -> float
{
    if(viewport_size_.height == 0)
    {
        return 0.0f;
    }
 
    return ortho_size_ / (float(viewport_size_.height) / 2.0f);
}
 
auto camera::get_ppu() const -> float
{
    return float(viewport_size_.height) / (2.0f * ortho_size_);
}
 
void camera::set_viewport_size(const usize32_t& viewport_size)
{
    viewport_size_ = viewport_size;
    set_aspect_ratio(float(viewport_size.width) / float(viewport_size.height));
}
 
void camera::set_viewport_pos(const ipoint32_t& viewportPos)
{
    viewport_pos_ = viewportPos;
}
 
auto camera::get_viewport_size() const -> const usize32_t&
{
    return viewport_size_;
}
 
auto camera::get_viewport_pos() const -> const ipoint32_t&
{
    return viewport_pos_;
}
 
void camera::set_orthographic_size(float size)
{
    ortho_size_ = size;
 
    touch();
}
 
auto camera::get_projection_mode() const -> projection_mode
{
    return projection_mode_;
}
 
auto camera::get_fov() const -> float
{
    return fov_;
}
 
auto camera::get_near_clip() const -> float
{
    return near_clip_;
}
 
auto camera::get_far_clip() const -> float
{
    return far_clip_;
}
 
auto camera::get_ortho_size() const -> float
{
    return ortho_size_;
}
 
void camera::set_fov(float fFOVY)
{
    // Skip if no-op
    if(math::equal(fFOVY, fov_, math::epsilon<float>()))
    {
        return;
    }
 
    // Update projection matrix and view frustum
    fov_ = fFOVY;
 
    touch();
}
 
void camera::set_projection_mode(projection_mode mode)
{
    // Bail if this is a no op.
    if(mode == projection_mode_)
    {
        return;
    }
 
    // Alter the projection mode.
    projection_mode_ = mode;
 
    touch();
}
 
void camera::set_near_clip(float distance)
{
    // Skip if this is a no-op
    if(math::equal(distance, near_clip_, math::epsilon<float>()))
    {
        return;
    }
 
    // Store value
    near_clip_ = distance;
 
    touch();
 
    // Make sure near clip is less than the far clip
    if(near_clip_ > far_clip_)
    {
        set_far_clip(near_clip_);
    }
}
 
void camera::set_far_clip(float distance)
{
    // Skip if this is a no-op
    if(math::equal(distance, far_clip_, math::epsilon<float>()))
    {
        return;
    }
 
    // Store value
    far_clip_ = distance;
 
    touch();
 
    // Make sure near clip is less than the far clip
    if(near_clip_ > far_clip_)
    {
        set_near_clip(far_clip_);
    }
}
 
auto camera::get_local_bounding_box() -> math::bbox
{
    if(projection_mode_ == projection_mode::perspective)
    {
        float far_size = math::tan(math::radians<float>(fov_ * 0.5f)) * far_clip_;
        return math::bbox(-far_size * aspect_ratio_,
                          -far_size,
                          near_clip_,
                          far_size * aspect_ratio_,
                          far_size,
                          far_clip_);
    }
 
    float spread = far_clip_ - near_clip_;
    math::vec3 center = {0.0f, 0.0f, (far_clip_ + near_clip_) * 0.5f};
    float orthographicSize = get_ortho_size();
    math::vec3 size = {orthographicSize * 2.0f * aspect_ratio_, orthographicSize * 2.0f, spread};
    return math::bbox(center - (size / 2.0f), center + (size / 2.0f));
}
 
void camera::set_aspect_ratio(float aspect, bool bLocked /* = false */)
{
    // Is this a no-op?
    if(math::equal(aspect, aspect_ratio_, math::epsilon<float>()))
    {
        aspect_locked_ = bLocked;
        return;
 
    } // End if aspect is the same
 
    // Update camera properties
    aspect_ratio_ = aspect;
    aspect_locked_ = bLocked;
    aspect_dirty_ = true;
    frustum_dirty_ = true;
    projection_dirty_ = true;
}
 
auto camera::get_aspect_ratio() const -> float
{
    return aspect_ratio_;
}
 
auto camera::is_aspect_locked() const -> bool
{
    return (get_projection_mode() == projection_mode::orthographic || aspect_locked_);
}
 
auto camera::is_frustum_locked() const -> bool
{
    return frustum_locked_;
}
 
void camera::lock_frustum(bool locked)
{
    frustum_locked_ = locked;
}
 
auto camera::get_projection() const -> const math::transform&
{
    // Only update matrix if something has changed
    if(get_projection_mode() == projection_mode::perspective)
    {
        if(projection_dirty_)
        {
            // Generate the updated perspective projection matrix
            float fov_radians = math::radians<float>(get_fov());
            static const auto perspective_ =
                gfx::is_homogeneous_depth() ? math::perspectiveNO<float> : math::perspectiveZO<float>;
 
            math::mat4 projection = perspective_(fov_radians, aspect_ratio_, near_clip_, far_clip_);
 
            projection[2][0] += aa_data_.z;
            projection[2][1] += aa_data_.w;
            projection_ = projection;
            // Matrix has been updated
            projection_dirty_ = false;
            aspect_dirty_ = false;
 
        } // End if projection matrix needs updating
        else if(aspect_dirty_)
        {
            // Just alter the aspect ratio
            math::mat4 projection = projection_;
            projection[0][0] = projection[1][1] / aspect_ratio_;
            projection_ = projection;
            // Matrix has been updated
            aspect_dirty_ = false;
 
        } // End if only aspect ratio changed
 
    } // End if perspective
    else if(get_projection_mode() == projection_mode::orthographic)
    {
        if(projection_dirty_ || aspect_dirty_)
        {
            // Generate the updated orthographic projection matrix
            float zoom = get_zoom_factor();
            const frect_t rect = {-float(viewport_size_.width) / 2.0f,
                                  float(viewport_size_.height) / 2.0f,
                                  float(viewport_size_.width) / 2.0f,
                                  -float(viewport_size_.height) / 2.0f};
            static const auto ortho_ =
                gfx::is_homogeneous_depth() ? math::orthoNO<float> : math::orthoZO<float>;
 
            math::mat4 projection = ortho_(rect.left * zoom,
                                           rect.right * zoom,
                                           rect.bottom * zoom,
                                           rect.top * zoom,
                                           get_near_clip(),
                                           get_far_clip());
            projection[2][0] += aa_data_.z;
            projection[2][1] += aa_data_.w;
            projection_ = projection;
            // Matrix has been updated
            projection_dirty_ = false;
            aspect_dirty_ = false;
 
        } // End if projection matrix needs updating
 
    } // End if orthographic
 
    // Return the projection matrix.
    return projection_;
}
 
auto camera::get_prev_projection() const -> const math::transform&
{
    return last_projection_;
}
 
auto camera::get_view() const -> const math::transform&
{
    return view_;
}
 
auto camera::get_prev_view() const -> const math::transform&
{
    return last_view_;
}
 
auto camera::get_view_relative() const -> const math::transform&
{
    return view_relative_;
}
 
auto camera::get_prev_view_relative() const -> const math::transform&
{
    return last_view_relative_;
}
 
auto camera::get_view_inverse() const -> const math::transform&
{
    return view_inverse_;
}
 
auto camera::get_view_inverse_relative() const -> const math::transform&
{
    return view_inverse_relative_;
}
 
auto camera::get_view_projection() const -> math::transform
{
    return get_projection() * get_view();
}
 
auto camera::get_prev_view_projection() const -> math::transform
{
    return get_prev_projection() * get_prev_view();
}
 
auto camera::get_taa_prev_view_projection() const -> math::transform
{
    return taa_prev_projection_ * taa_prev_view_;
}
 
auto camera::get_view_projection_relative() const -> math::transform
{
    return get_projection() * get_view_relative();
}
 
auto camera::get_prev_view_projection_relative() const -> math::transform
{
    return get_prev_projection() * get_prev_view_relative();
}
 
void camera::look_at(const math::vec3& vEye, const math::vec3& vAt)
{
    look_at(vEye, vAt, math::vec3(0.0f, 1.0f, 0.0f));
}
 
void camera::look_at(const math::vec3& vEye, const math::vec3& vAt, const math::vec3& vUp)
{
    // First update so the camera can cache the previous matrices
    // record_current_matrices();
 
    view_ = math::lookAt(vEye, vAt, vUp);
    view_inverse_ = math::inverse(view_);
 
 
    view_relative_ = math::lookAt(math::vec3(0.0f, 0.0f, 0.0f), vAt - vEye, vUp);
    view_inverse_relative_ = math::inverse(view_relative_);
 
    touch();
}
 
auto camera::get_position() const -> const math::vec3&
{
    return view_inverse_.get_position();
}
 
auto camera::x_unit_axis() const -> math::vec3
{
    return view_inverse_.x_unit_axis();
}
auto camera::y_unit_axis() const -> math::vec3
{
    return view_inverse_.y_unit_axis();
}
 
auto camera::z_unit_axis() const -> math::vec3
{
    return view_inverse_.z_unit_axis();
}
 
auto camera::get_frustum() const -> const math::frustum&
{
    // Recalculate frustum if necessary
    if(frustum_dirty_ && !frustum_locked_)
    {
        frustum_.update(get_view(), get_projection(), gfx::is_homogeneous_depth());
        frustum_dirty_ = false;
 
        // Also build the frustum / volume that represents the space between the
        // camera position and its near plane. This frustum represents the
        // 'volume' that can end up clipping geometry.
 
        clipping_volume_ = frustum_;
        clipping_volume_.planes[math::volume_plane::far_plane].data.w =
            -clipping_volume_.planes[math::volume_plane::near_plane].data.w; // At near plane
        clipping_volume_.planes[math::volume_plane::near_plane].data.w =
            //-math::dot((math::vec3&)_clipping_volume.planes[math::volume_plane::near_plane],
            //         get_position()); // At camera
            -math::plane::dot_coord(clipping_volume_.planes[math::volume_plane::near_plane], get_position());
        // The corner points also need adjusting in this case such that they sit
        // precisely on the new planes.
        clipping_volume_.points[math::volume_geometry_point::left_bottom_far] =
            clipping_volume_.points[math::volume_geometry_point::left_bottom_near];
        clipping_volume_.points[math::volume_geometry_point::left_top_far] =
            clipping_volume_.points[math::volume_geometry_point::left_top_near];
        clipping_volume_.points[math::volume_geometry_point::right_bottom_far] =
            clipping_volume_.points[math::volume_geometry_point::right_bottom_near];
        clipping_volume_.points[math::volume_geometry_point::right_top_far] =
            clipping_volume_.points[math::volume_geometry_point::right_top_near];
        clipping_volume_.points[math::volume_geometry_point::left_bottom_near] = clipping_volume_.position;
        clipping_volume_.points[math::volume_geometry_point::left_top_near] = clipping_volume_.position;
        clipping_volume_.points[math::volume_geometry_point::right_bottom_near] = clipping_volume_.position;
        clipping_volume_.points[math::volume_geometry_point::right_top_near] = clipping_volume_.position;
 
    } // End if recalc frustum
 
    // Return the frustum
    return frustum_;
}
 
auto camera::get_clipping_volume() const -> const math::frustum&
{
    // Recalculate frustum if necessary
    if(frustum_dirty_ && !frustum_locked_)
    {
        get_frustum();
    }
 
    // Return the clipping volume
    return clipping_volume_;
}
 
auto camera::classify_aabb(const math::bbox& AABB) const -> math::volume_query
{
    // Recompute the frustum as necessary.
    const math::frustum& f = get_frustum();
 
    // Request that frustum classifies
    return f.classify_aabb(AABB);
}
 
auto camera::test_aabb(const math::bbox& AABB) const -> bool
{
    // Recompute the frustum as necessary.
    const math::frustum& f = get_frustum();
 
    // Request that frustum classifies
    return f.test_aabb(AABB);
}
 
auto camera::classify_obb(const math::bbox& AABB, const math::transform& t) const -> math::volume_query
{
    // Recompute the frustum as necessary.
    const math::frustum& f = get_frustum();
 
    // Request that frustum classifies
    return f.classify_obb(AABB, t);
}
 
auto camera::test_obb(const math::bbox& AABB, const math::transform& t) const -> bool
{
    // Recompute the frustum as necessary.
    const math::frustum& f = get_frustum();
 
    // Request that frustum classifies
    return f.test_obb(AABB, t);
}
 
auto camera::test_billboard(float size, const math::transform& t) const -> bool
{
    // Recompute the frustum as necessary.
    const math::frustum& f = get_frustum();
 
    auto center = t.get_position();
    // 2) Build our test‐sphere
    float radius = size * glm::root_two<float>();
    math::bsphere sphere{center, radius};
 
    return f.test_sphere(sphere);
}
 
auto camera::world_to_viewport(const math::vec3& pos) const -> math::vec3
{
    // Ensure we have an up-to-date projection and view matrix
    auto view_proj = get_view_projection();
 
    // Transform the point into clip space
    math::vec4 clip = view_proj * math::vec4{pos.x, pos.y, pos.z, 1.0f};
 
    // Project!
    const float recip_w = 1.0f / clip.w;
    clip.x *= recip_w;
    clip.y *= recip_w;
    clip.z *= recip_w;
 
    // Transform to final screen space position
    math::vec3 point;
    point.x = ((clip.x * 0.5f) + 0.5f) * float(viewport_size_.width) + float(viewport_pos_.x);
    point.y = ((clip.y * -0.5f) + 0.5f) * float(viewport_size_.height) + float(viewport_pos_.y);
    point.z = clip.z;
 
    // Point on screen!
    return point;
}
 
auto camera::viewport_to_ray(const math::vec2& point, math::vec3& vec_ray_start, math::vec3& vec_ray_dir) const -> bool
{
    // Ensure we have an up-to-date projection and view matrix
    math::transform mtx_proj = get_projection();
    math::transform mtx_view = get_view();
    math::transform mtx_inv_view = math::inverse(mtx_view);
 
    // Transform pick position from viewport to normalized device coordinates (NDC)
    float ndc_x = 2.0f * (point.x - viewport_pos_.x) / float(viewport_size_.width) - 1.0f;
    float ndc_y = 2.0f * (point.y - viewport_pos_.y) / float(viewport_size_.height) - 1.0f;
 
           // Ensure projection matrix values are valid
    if (math::abs(mtx_proj[0][0]) < math::epsilon<float>() || math::abs(mtx_proj[1][1]) < math::epsilon<float>())
    {
        return false;
    }
 
    math::vec3 cursor(ndc_x / mtx_proj[0][0], -ndc_y / mtx_proj[1][1], 1.0f);
 
    if (get_projection_mode() == projection_mode::orthographic)
    {
        vec_ray_start = mtx_inv_view.transform_coord(cursor);
        vec_ray_dir = math::normalize(mtx_inv_view[2]); // Z-axis direction
    }
    else
    {
        vec_ray_start = mtx_inv_view.get_position();
        vec_ray_dir = math::normalize(mtx_inv_view.transform_normal(cursor));
    }
 
    return true;
}
 
auto camera::viewport_to_world(const math::vec2& point, const math::plane& pl, math::vec3& world_pos, bool clip) const
    -> bool
{
    if(clip && ((point.x < viewport_pos_.x) || (point.x > (viewport_pos_.x + viewport_size_.width)) ||
                (point.y < viewport_pos_.y) || (point.y > (viewport_pos_.y + viewport_size_.height))))
    {
        return false;
    }
 
    math::vec3 pick_ray_dir;
    math::vec3 pick_ray_origin;
    // Convert the screen coordinates to a ray.
    if(!viewport_to_ray(point, pick_ray_origin, pick_ray_dir))
    {
        return false;
    }
 
    // Get the length of the 'adjacent' side of the virtual triangle formed
    // by the direction and normal.
    float proj_ray_length = math::plane::dot_normal(pl, pick_ray_dir);
    if(math::abs<float>(proj_ray_length) < math::epsilon<float>())
    {
        return false;
    }
 
    // Calculate distance to plane along its normal
    float distance = math::plane::dot_normal(pl, pick_ray_origin) + pl.data.w;
 
    // If both the "direction" and origin are on the same side of the plane
    // then we can't possibly intersect (perspective rule only)
    if(get_projection_mode() == projection_mode::perspective)
    {
        int nSign1 = (distance > 0) ? 1 : (distance < 0) ? -1 : 0;
        int nSign2 = (proj_ray_length > 0) ? 1 : (proj_ray_length < 0) ? -1 : 0;
        if(nSign1 == nSign2)
        {
            return false;
        }
 
    } // End if perspective
 
    // Calculate the actual interval (Distance along the adjacent side / length of
    // adjacent side).
    distance /= -proj_ray_length;
 
    // Store the results
    world_pos = pick_ray_origin + (pick_ray_dir * distance);
 
    // Success!
    return true;
}
 
auto camera::project_to_quad(const math::vec2& viewport_point,
                          const math::transform& quad_transform,
                          uint32_t quad_width,
                          uint32_t quad_height,
                          math::vec2& pixel_out) const -> bool
{
    const math::vec3 quad_center = quad_transform.get_position();
    const math::vec3 quad_normal = quad_transform.z_unit_axis();
    const auto plane = math::plane::from_point_normal(quad_center, quad_normal);
 
    math::vec3 world_pos;
    if(!viewport_to_world(viewport_point, plane, world_pos, false))
    {
        pixel_out = math::vec2(-1.0f, -1.0f);
        return false;
    }
 
    const math::vec3 local = quad_transform.inverse_transform_coord(world_pos);
    if(local.x < -0.5f || local.x > 0.5f || local.y < -0.5f || local.y > 0.5f)
    {
        pixel_out = math::vec2(-1.0f, -1.0f);
        return false;
    }
    pixel_out.x = (local.x + 0.5f) * static_cast<float>(quad_width);
    pixel_out.y = (0.5f - local.y) * static_cast<float>(quad_height);
    return true;
}
 
auto camera::viewport_to_major_axis(const math::vec2& point,
                                    const math::vec3& Origin,
                                    math::vec3& world_pos,
                                    math::vec3& major_axis) const -> bool
{
    return viewport_to_major_axis(point, Origin, z_unit_axis(), world_pos, major_axis);
}
 
auto camera::viewport_to_major_axis(const math::vec2& point,
                                    const math::vec3& Origin,
                                    const math::vec3& Normal,
                                    math::vec3& world_pos,
                                    math::vec3& major_axis) const -> bool
{
    // First select the major axis plane based on the specified normal
    major_axis = math::vec3(1, 0, 0); // YZ
 
    // Get absolute normal vector
    float x = math::abs<float>(Normal.x);
    float y = math::abs<float>(Normal.y);
    float z = math::abs<float>(Normal.z);
 
    // If all the components are effectively equal, select one plane
    if(math::abs<float>(x - y) < math::epsilon<float>() && math::abs<float>(x - z) < math::epsilon<float>())
    {
        major_axis = math::vec3(0, 0, 1); // XY
 
    } // End if components equal
    else
    {
        // Calculate which component of the normal is the major axis
        float norm = x;
        if(norm < y)
        {
            norm = y;
            major_axis = math::vec3(0, 1, 0);
        } // XZ
        if(norm < z)
        {
            norm = z;
            major_axis = math::vec3(0, 0, 1);
        } // XY
 
    } // End if perform compare
 
    // Generate the intersection plane based on this information
    // and pass through to the standard viewportToWorld method
    math::plane p = math::plane::from_point_normal(Origin, major_axis);
    return viewport_to_world(point, p, world_pos, false);
}
 
auto camera::viewport_to_camera(const math::vec3& point, math::vec3& camera_pos) const -> bool
{
    // Ensure that we have an up-to-date projection and view matrix
    auto& mtx_proj = get_projection();
 
    // Transform the pick position from screen space into camera space
    camera_pos.x = (((2.0f * (point.x - viewport_pos_.x)) / float(viewport_size_.width)) - 1) / mtx_proj[0][0];
    camera_pos.y = -(((2.0f * (point.y - viewport_pos_.y)) / float(viewport_size_.height)) - 1) / mtx_proj[1][1];
    camera_pos.z = get_near_clip();
 
    // Success!
    return true;
}
 
auto camera::estimate_zoom_factor(const math::plane& pl) const -> float
{
    // Just return the actual zoom factor if this is orthographic
    if(get_projection_mode() == projection_mode::orthographic)
    {
        return get_zoom_factor();
    }
 
    math::vec3 world;
    // Otherwise, estimate is based on the distance from the grid plane.
    viewport_to_world(math::vec2(float(viewport_size_.width) / 2.0f, float(viewport_size_.height) / 2.0f),
                      pl,
                      world,
                      false);
 
    // Perform full position based estimation
    return estimate_zoom_factor(world);
}
 
//-----------------------------------------------------------------------------
// Name : estimateZoomFactor ()
//-----------------------------------------------------------------------------
auto camera::estimate_zoom_factor(const math::vec3& world_pos) const -> float
{
    return estimate_zoom_factor(world_pos, std::numeric_limits<float>::max());
}
 
auto camera::estimate_zoom_factor(const math::plane& pl, float max_val) const -> float
{
    // Just return the actual zoom factor if this is orthographic
    if(get_projection_mode() == projection_mode::orthographic)
    {
        float factor = get_zoom_factor();
        return math::min(max_val, factor);
 
    } // End if orthographic
    // Otherwise, estimate is based on the distance from the grid plane.
    math::vec3 world;
    viewport_to_world(math::vec2(float(viewport_size_.width) / 2.0f, float(viewport_size_.height) / 2.0f),
                      pl,
                      world,
                      false);
 
    // Perform full position based estimation
    return estimate_zoom_factor(world, max_val);
}
 
auto camera::estimate_zoom_factor(const math::vec3& world_pos, float max_val) const -> float
{
    // Just return the actual zoom factor if this is orthographic
    if(get_projection_mode() == projection_mode::orthographic)
    {
        float factor = get_zoom_factor();
        return math::min(max_val, factor);
 
    } // End if orthographic
 
    // New Zoom factor is based on the distance to this position
    // along the camera's look vector.
    math::vec3 view_pos = get_view().transform_coord(world_pos);
    float distance = view_pos.z / (float(viewport_size_.height) * (45.0f / get_fov()));
    return std::min<float>(max_val, distance);
}
 
auto camera::estimate_pick_tolerance(float wire_tolerance,
                                     const math::vec3& pos,
                                     const math::transform& object_transform) const -> math::vec3
{
    // Scale tolerance based on estimated world space zoom factor.
    math::vec3 v = object_transform.transform_coord(pos);
    wire_tolerance *= estimate_zoom_factor(v);
 
    // Convert into object space tolerance.
    math::vec3 object_wire_tolerance;
    const math::vec3& vAxisScale = object_transform.get_scale();
    return object_wire_tolerance / vAxisScale;
}
 
void camera::record_current_matrices()
{
    last_view_ = get_view();
    last_view_relative_ = get_view_relative();
    last_projection_ = get_projection();
    
}
 
void camera::set_aa_data(const usize32_t& viewport_size,
                         std::uint32_t temporal_frame_index,
                         std::uint32_t temporal_aa_samples,
                         taa_jitter_mode jitter_mode,
                         float jitter_amplitude,
                         float jitter_temporal_phase_scale)
{
    taa_prev_view_ = get_view();
    taa_prev_projection_ = get_projection();
 
    if(temporal_aa_samples > 1)
    {
        float SampleX = 0.0f;
        float SampleY = 0.0f;
        switch(jitter_mode)
        {
        case taa_jitter_mode::halton_2_3:
            math::taa_subpixel_offset_halton(temporal_frame_index,
                                             SampleX,
                                             SampleY,
                                             jitter_temporal_phase_scale);
            break;
        case taa_jitter_mode::r2_low_discrepancy:
            math::taa_subpixel_offset_r2(temporal_frame_index,
                                         SampleX,
                                         SampleY,
                                         jitter_temporal_phase_scale);
            break;
        case taa_jitter_mode::msaa_2_rotating:
        {
            const float SamplesX[] = {-4.0f / 16.0f, 4.0f / 16.0f};
            const float SamplesY[] = {4.0f / 16.0f, -4.0f / 16.0f};
            const std::uint32_t idx = temporal_frame_index % 2u;
            SampleX = SamplesX[idx];
            SampleY = SamplesY[idx];
            break;
        }
        case taa_jitter_mode::msaa_3_rotating:
        {
            // Equilateral triangle inscribed in a circle of radius 0.4 px,
            // centered at the pixel center (vertices at 90°/210°/330°). Keeps
            // all offsets inside the standard [-0.5, +0.5] px convention used
            // by the progressive / Halton / R2 / msaa_4 modes; side length
            // r·√3 ≈ 0.69 px gives a per-frame step comparable to msaa_2.
            const float SamplesX[] = {0.0f, -0.34641016f, 0.34641016f};
            const float SamplesY[] = {0.4f, -0.2f, -0.2f};
            const std::uint32_t idx = temporal_frame_index % 3u;
            SampleX = SamplesX[idx];
            SampleY = SamplesY[idx];
            break;
        }
        case taa_jitter_mode::msaa_4_rotating:
        {
            const float SamplesX[] = {-2.0f / 16.0f, 6.0f / 16.0f, 2.0f / 16.0f, -6.0f / 16.0f};
            const float SamplesY[] = {-6.0f / 16.0f, -2.0f / 16.0f, 6.0f / 16.0f, 2.0f / 16.0f};
            const std::uint32_t idx = temporal_frame_index % 4u;
            SampleX = SamplesX[idx];
            SampleY = SamplesY[idx];
            break;
        }
        case taa_jitter_mode::progressive_golden:
        default:
            math::taa_subpixel_offset_progressive(temporal_frame_index,
                                                  SampleX,
                                                  SampleY,
                                                  jitter_temporal_phase_scale);
            break;
        }
 
        const float amp = std::clamp(jitter_amplitude, 0.0f, 1.5f);
        SampleX *= amp;
        SampleY *= amp;
 
        aa_data_ = math::vec4(float(temporal_frame_index & 0xFFFFu),
                              float(temporal_aa_samples),
                              SampleX,
                              SampleY);
 
        float width = static_cast<float>(viewport_size.width);
        float height = static_cast<float>(viewport_size.height);
        aa_data_.z *= (2.0f / width);
        aa_data_.w *= (2.0f / height);
    }
    else
    {
        aa_data_ = math::vec4(0.0f, 0.0f, 0.0f, 0.0f);
    }
 
    projection_dirty_ = true;
}
 
auto camera::get_aa_data() const -> const math::vec4&
{
    return aa_data_;
}
 
void camera::touch()
{
    // All modifications require projection matrix and
    // frustum to be updated.
    view_dirty_ = true;
    projection_dirty_ = true;
    frustum_dirty_ = true;
}
 
auto camera::get_face_camera(uint32_t face, const math::transform& transform) -> camera
{
    camera cam;
    cam.set_fov(90.0f);
    cam.set_aspect_ratio(1.0f, true);
    cam.set_near_clip(0.01f);
    cam.set_far_clip(256.0f);
 
    // Configurable axis vectors used to construct view matrices. In the
    // case of the omni light, we align all frustums to the world axes.
    math::vec3 X(1, 0, 0);
    math::vec3 Y(0, 1, 0);
    math::vec3 Z(0, 0, 1);
    math::vec3 Zero(0, 0, 0);
    math::transform t;
    // Generate the correct view matrix for the frustum
 
    switch(face)
    {
        case 0: // right
            t.set_rotation(-Z, +Y, +X);
            break;
        case 1: // left
            t.set_rotation(+Z, +Y, -X);
            break;
        case 2: // up
            if(!gfx::is_origin_bottom_left())
            {
                t.set_rotation(+X, -Z, +Y);
            }
            else
            {
                t.set_rotation(+X, +Z, -Y);
            }
            break;
        case 3: // down
            if(!gfx::is_origin_bottom_left())
            {
                t.set_rotation(+X, +Z, -Y);
            }
            else
            {
                t.set_rotation(+X, -Z, +Y);
            }
            break;
        case 4: // front
            t.set_rotation(+X, +Y, +Z);
            break;
        case 5: // back
            t.set_rotation(-X, +Y, -Z);
            break;
    }
 
    t = transform * t;
 
    // Set new transform
    cam.look_at(t.get_position(), t.get_position() + t.z_unit_axis(), t.y_unit_axis());
    cam.record_current_matrices();
    return cam;
}
} // namespace unravel