Index: cc/quads/draw_polygon.cc |
diff --git a/cc/quads/draw_polygon.cc b/cc/quads/draw_polygon.cc |
deleted file mode 100644 |
index 6544b641acb237d07d6525015df5ae3abc93b369..0000000000000000000000000000000000000000 |
--- a/cc/quads/draw_polygon.cc |
+++ /dev/null |
@@ -1,363 +0,0 @@ |
-// Copyright 2014 The Chromium Authors. All rights reserved. |
-// Use of this source code is governed by a BSD-style license that can be |
-// found in the LICENSE file. |
- |
-#include "cc/quads/draw_polygon.h" |
- |
-#include <vector> |
- |
-#include "cc/output/bsp_compare_result.h" |
-#include "cc/quads/draw_quad.h" |
- |
-namespace { |
-// This allows for some imperfection in the normal comparison when checking if |
-// two pieces of geometry are coplanar. |
-static const float coplanar_dot_epsilon = 0.001f; |
-// This threshold controls how "thick" a plane is. If a point's distance is |
-// <= |compare_threshold|, then it is considered on the plane. Only when this |
-// boundary is crossed do we consider doing splitting. |
-static const float compare_threshold = 1.0f; |
-// |split_threshold| is lower in this case because we want the points created |
-// during splitting to be well within the range of |compare_threshold| for |
-// comparison purposes. The splitting operation will produce intersection points |
-// that fit within a tighter distance to the splitting plane as a result of this |
-// value. By using a value >= |compare_threshold| we run the risk of creating |
-// points that SHOULD be intersecting the "thick plane", but actually fail to |
-// test positively for it because |split_threshold| allowed them to be outside |
-// this range. |
-// This is really supposd to be compare_threshold / 2.0f, but that would |
-// create another static initializer. |
-static const float split_threshold = 0.5f; |
- |
-static const float normalized_threshold = 0.001f; |
-} // namespace |
- |
-namespace cc { |
- |
-DrawPolygon::DrawPolygon() { |
-} |
- |
-DrawPolygon::DrawPolygon(const DrawQuad* original, |
- const std::vector<gfx::Point3F>& in_points, |
- const gfx::Vector3dF& normal, |
- int draw_order_index) |
- : order_index_(draw_order_index), original_ref_(original), is_split_(true) { |
- for (size_t i = 0; i < in_points.size(); i++) { |
- points_.push_back(in_points[i]); |
- } |
- normal_ = normal; |
-} |
- |
-// This takes the original DrawQuad that this polygon should be based on, |
-// a visible content rect to make the 4 corner points from, and a transformation |
-// to move it and its normal into screen space. |
-DrawPolygon::DrawPolygon(const DrawQuad* original_ref, |
- const gfx::RectF& visible_content_rect, |
- const gfx::Transform& transform, |
- int draw_order_index) |
- : normal_(0.0f, 0.0f, 1.0f), |
- order_index_(draw_order_index), |
- original_ref_(original_ref), |
- is_split_(false) { |
- gfx::Point3F points[8]; |
- int num_vertices_in_clipped_quad; |
- gfx::QuadF send_quad(visible_content_rect); |
- |
- // Doing this mapping here is very important, since we can't just transform |
- // the points without clipping and not run into strange geometry issues when |
- // crossing w = 0. At this point, in the constructor, we know that we're |
- // working with a quad, so we can reuse the MathUtil::MapClippedQuad3d |
- // function instead of writing a generic polygon version of it. |
- MathUtil::MapClippedQuad3d( |
- transform, send_quad, points, &num_vertices_in_clipped_quad); |
- for (int i = 0; i < num_vertices_in_clipped_quad; i++) { |
- points_.push_back(points[i]); |
- } |
- ApplyTransformToNormal(transform); |
-} |
- |
-DrawPolygon::~DrawPolygon() { |
-} |
- |
-scoped_ptr<DrawPolygon> DrawPolygon::CreateCopy() { |
- scoped_ptr<DrawPolygon> new_polygon(new DrawPolygon()); |
- new_polygon->order_index_ = order_index_; |
- new_polygon->original_ref_ = original_ref_; |
- new_polygon->points_.reserve(points_.size()); |
- new_polygon->points_ = points_; |
- new_polygon->normal_.set_x(normal_.x()); |
- new_polygon->normal_.set_y(normal_.y()); |
- new_polygon->normal_.set_z(normal_.z()); |
- return new_polygon.Pass(); |
-} |
- |
-float DrawPolygon::SignedPointDistance(const gfx::Point3F& point) const { |
- return gfx::DotProduct(point - points_[0], normal_); |
-} |
- |
-// Checks whether or not shape a lies on the front or back side of b, or |
-// whether they should be considered coplanar. If on the back side, we |
-// say A_BEFORE_B because it should be drawn in that order. |
-// Assumes that layers are split and there are no intersecting planes. |
-BspCompareResult DrawPolygon::SideCompare(const DrawPolygon& a, |
- const DrawPolygon& b) { |
- // Let's make sure that both of these are normalized. |
- DCHECK_GE(normalized_threshold, std::abs(a.normal_.LengthSquared() - 1.0f)); |
- DCHECK_GE(normalized_threshold, std::abs(b.normal_.LengthSquared() - 1.0f)); |
- // Right away let's check if they're coplanar |
- double dot = gfx::DotProduct(a.normal_, b.normal_); |
- float sign = 0.0f; |
- bool normal_match = false; |
- // This check assumes that the normals are normalized. |
- if (std::abs(dot) >= 1.0f - coplanar_dot_epsilon) { |
- normal_match = true; |
- // The normals are matching enough that we only have to test one point. |
- sign = b.SignedPointDistance(a.points_[0]); |
- // Is it on either side of the splitter? |
- if (sign < -compare_threshold) { |
- return BSP_BACK; |
- } |
- |
- if (sign > compare_threshold) { |
- return BSP_FRONT; |
- } |
- |
- // No it wasn't, so the sign of the dot product of the normals |
- // along with document order determines which side it goes on. |
- if (dot >= 0.0f) { |
- if (a.order_index_ < b.order_index_) { |
- return BSP_COPLANAR_FRONT; |
- } |
- return BSP_COPLANAR_BACK; |
- } |
- |
- if (a.order_index_ < b.order_index_) { |
- return BSP_COPLANAR_BACK; |
- } |
- return BSP_COPLANAR_FRONT; |
- } |
- |
- int pos_count = 0; |
- int neg_count = 0; |
- for (size_t i = 0; i < a.points_.size(); i++) { |
- if (!normal_match || (normal_match && i > 0)) { |
- sign = gfx::DotProduct(a.points_[i] - b.points_[0], b.normal_); |
- } |
- |
- if (sign < -compare_threshold) { |
- ++neg_count; |
- } else if (sign > compare_threshold) { |
- ++pos_count; |
- } |
- |
- if (pos_count && neg_count) { |
- return BSP_SPLIT; |
- } |
- } |
- |
- if (pos_count) { |
- return BSP_FRONT; |
- } |
- return BSP_BACK; |
-} |
- |
-static bool LineIntersectPlane(const gfx::Point3F& line_start, |
- const gfx::Point3F& line_end, |
- const gfx::Point3F& plane_origin, |
- const gfx::Vector3dF& plane_normal, |
- gfx::Point3F* intersection, |
- float distance_threshold) { |
- gfx::Vector3dF start_to_origin_vector = plane_origin - line_start; |
- gfx::Vector3dF end_to_origin_vector = plane_origin - line_end; |
- |
- double start_distance = gfx::DotProduct(start_to_origin_vector, plane_normal); |
- double end_distance = gfx::DotProduct(end_to_origin_vector, plane_normal); |
- |
- // The case where one vertex lies on the thick-plane and the other |
- // is outside of it. |
- if (std::abs(start_distance) <= distance_threshold && |
- std::abs(end_distance) > distance_threshold) { |
- intersection->SetPoint(line_start.x(), line_start.y(), line_start.z()); |
- return true; |
- } |
- |
- // This is the case where we clearly cross the thick-plane. |
- if ((start_distance > distance_threshold && |
- end_distance < -distance_threshold) || |
- (start_distance < -distance_threshold && |
- end_distance > distance_threshold)) { |
- gfx::Vector3dF v = line_end - line_start; |
- float total_distance = std::abs(start_distance) + std::abs(end_distance); |
- float lerp_factor = std::abs(start_distance) / total_distance; |
- |
- intersection->SetPoint(line_start.x() + (v.x() * lerp_factor), |
- line_start.y() + (v.y() * lerp_factor), |
- line_start.z() + (v.z() * lerp_factor)); |
- |
- return true; |
- } |
- return false; |
-} |
- |
-// This function is separate from ApplyTransform because it is often unnecessary |
-// to transform the normal with the rest of the polygon. |
-// When drawing these polygons, it is necessary to move them back into layer |
-// space before sending them to OpenGL, which requires using ApplyTransform, |
-// but normal information is no longer needed after sorting. |
-void DrawPolygon::ApplyTransformToNormal(const gfx::Transform& transform) { |
- // Now we use the inverse transpose of |transform| to transform the normal. |
- gfx::Transform inverse_transform; |
- bool inverted = transform.GetInverse(&inverse_transform); |
- DCHECK(inverted); |
- if (!inverted) |
- return; |
- inverse_transform.Transpose(); |
- |
- gfx::Point3F new_normal(normal_.x(), normal_.y(), normal_.z()); |
- inverse_transform.TransformPoint(&new_normal); |
- // Make sure our normal is still normalized. |
- normal_ = gfx::Vector3dF(new_normal.x(), new_normal.y(), new_normal.z()); |
- float normal_magnitude = normal_.Length(); |
- if (normal_magnitude != 0 && normal_magnitude != 1) { |
- normal_.Scale(1.0f / normal_magnitude); |
- } |
-} |
- |
-void DrawPolygon::ApplyTransform(const gfx::Transform& transform) { |
- for (size_t i = 0; i < points_.size(); i++) { |
- transform.TransformPoint(&points_[i]); |
- } |
-} |
- |
-// TransformToScreenSpace assumes we're moving a layer from its layer space |
-// into 3D screen space, which for sorting purposes requires the normal to |
-// be transformed along with the vertices. |
-void DrawPolygon::TransformToScreenSpace(const gfx::Transform& transform) { |
- ApplyTransform(transform); |
- ApplyTransformToNormal(transform); |
-} |
- |
-// In the case of TransformToLayerSpace, we assume that we are giving the |
-// inverse transformation back to the polygon to move it back into layer space |
-// but we can ignore the costly process of applying the inverse to the normal |
-// since we know the normal will just reset to its original state. |
-void DrawPolygon::TransformToLayerSpace( |
- const gfx::Transform& inverse_transform) { |
- ApplyTransform(inverse_transform); |
- normal_ = gfx::Vector3dF(0.0f, 0.0f, -1.0f); |
-} |
- |
-bool DrawPolygon::Split(const DrawPolygon& splitter, |
- scoped_ptr<DrawPolygon>* front, |
- scoped_ptr<DrawPolygon>* back) { |
- gfx::Point3F intersections[2]; |
- std::vector<gfx::Point3F> out_points[2]; |
- // vertex_before stores the index of the vertex before its matching |
- // intersection. |
- // i.e. vertex_before[0] stores the vertex we saw before we crossed the plane |
- // which resulted in the line/plane intersection giving us intersections[0]. |
- size_t vertex_before[2]; |
- size_t points_size = points_.size(); |
- size_t current_intersection = 0; |
- |
- size_t current_vertex = 0; |
- // We will only have two intersection points because we assume all polygons |
- // are convex. |
- while (current_intersection < 2) { |
- if (LineIntersectPlane(points_[(current_vertex % points_size)], |
- points_[(current_vertex + 1) % points_size], |
- splitter.points_[0], |
- splitter.normal_, |
- &intersections[current_intersection], |
- split_threshold)) { |
- vertex_before[current_intersection] = current_vertex % points_size; |
- current_intersection++; |
- // We found both intersection points so we're done already. |
- if (current_intersection == 2) { |
- break; |
- } |
- } |
- if (current_vertex++ > (points_size)) { |
- break; |
- } |
- } |
- DCHECK_EQ(current_intersection, static_cast<size_t>(2)); |
- |
- // Since we found both the intersection points, we can begin building the |
- // vertex set for both our new polygons. |
- size_t start1 = (vertex_before[0] + 1) % points_size; |
- size_t start2 = (vertex_before[1] + 1) % points_size; |
- size_t points_remaining = points_size; |
- |
- // First polygon. |
- out_points[0].push_back(intersections[0]); |
- DCHECK_GE(vertex_before[1], start1); |
- for (size_t i = start1; i <= vertex_before[1]; i++) { |
- out_points[0].push_back(points_[i]); |
- --points_remaining; |
- } |
- out_points[0].push_back(intersections[1]); |
- |
- // Second polygon. |
- out_points[1].push_back(intersections[1]); |
- size_t index = start2; |
- for (size_t i = 0; i < points_remaining; i++) { |
- out_points[1].push_back(points_[index % points_size]); |
- ++index; |
- } |
- out_points[1].push_back(intersections[0]); |
- |
- // Give both polygons the original splitting polygon's ID, so that they'll |
- // still be sorted properly in co-planar instances. |
- scoped_ptr<DrawPolygon> poly1( |
- new DrawPolygon(original_ref_, out_points[0], normal_, order_index_)); |
- scoped_ptr<DrawPolygon> poly2( |
- new DrawPolygon(original_ref_, out_points[1], normal_, order_index_)); |
- |
- DCHECK_GE(poly1->points().size(), 3u); |
- DCHECK_GE(poly2->points().size(), 3u); |
- |
- if (SideCompare(*poly1, splitter) == BSP_FRONT) { |
- *front = poly1.Pass(); |
- *back = poly2.Pass(); |
- } else { |
- *front = poly2.Pass(); |
- *back = poly1.Pass(); |
- } |
- return true; |
-} |
- |
-// This algorithm takes the first vertex in the polygon and uses that as a |
-// pivot point to fan out and create quads from the rest of the vertices. |
-// |offset| starts off as the second vertex, and then |op1| and |op2| indicate |
-// offset+1 and offset+2 respectively. |
-// After the first quad is created, the first vertex in the next quad is the |
-// same as all the rest, the pivot point. The second vertex in the next quad is |
-// the old |op2|, the last vertex added to the previous quad. This continues |
-// until all points are exhausted. |
-// The special case here is where there are only 3 points remaining, in which |
-// case we use the same values for vertex 3 and 4 to make a degenerate quad |
-// that represents a triangle. |
-void DrawPolygon::ToQuads2D(std::vector<gfx::QuadF>* quads) const { |
- if (points_.size() <= 2) |
- return; |
- |
- gfx::PointF first(points_[0].x(), points_[0].y()); |
- size_t offset = 1; |
- while (offset < points_.size() - 1) { |
- size_t op1 = offset + 1; |
- size_t op2 = offset + 2; |
- if (op2 >= points_.size()) { |
- // It's going to be a degenerate triangle. |
- op2 = op1; |
- } |
- quads->push_back( |
- gfx::QuadF(first, |
- gfx::PointF(points_[offset].x(), points_[offset].y()), |
- gfx::PointF(points_[op1].x(), points_[op1].y()), |
- gfx::PointF(points_[op2].x(), points_[op2].y()))); |
- offset = op2; |
- } |
-} |
- |
-} // namespace cc |