| // Copyright 2012 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/base/math_util.h" |
| |
| #include <algorithm> |
| #include <cmath> |
| #include <limits> |
| #if defined(ARCH_CPU_X86_FAMILY) |
| #include <xmmintrin.h> |
| #endif |
| |
| #include "base/cxx17_backports.h" |
| #include "base/trace_event/traced_value.h" |
| #include "base/values.h" |
| #include "ui/gfx/geometry/angle_conversions.h" |
| #include "ui/gfx/geometry/linear_gradient.h" |
| #include "ui/gfx/geometry/quad_f.h" |
| #include "ui/gfx/geometry/rect.h" |
| #include "ui/gfx/geometry/rect_conversions.h" |
| #include "ui/gfx/geometry/rect_f.h" |
| #include "ui/gfx/geometry/rrect_f.h" |
| #include "ui/gfx/geometry/transform.h" |
| #include "ui/gfx/geometry/vector2d_f.h" |
| #include "ui/gfx/geometry/vector3d_f.h" |
| |
| namespace cc { |
| |
| static HomogeneousCoordinate ProjectHomogeneousPoint( |
| const gfx::Transform& transform, |
| const gfx::PointF& p) { |
| SkScalar m22 = transform.matrix().rc(2, 2); |
| // In this case, the layer we are trying to project onto is perpendicular to |
| // ray (point p and z-axis direction) that we are trying to project. This |
| // happens when the layer is rotated so that it is infinitesimally thin, or |
| // when it is co-planar with the camera origin -- i.e. when the layer is |
| // invisible anyway. |
| if (!std::isnormal(m22)) |
| return HomogeneousCoordinate(0.0, 0.0, 0.0, 1.0); |
| SkScalar z = |
| -(transform.matrix().rc(2, 0) * p.x() + |
| transform.matrix().rc(2, 1) * p.y() + transform.matrix().rc(2, 3)) / |
| m22; |
| // Same underlying condition as the previous early return. |
| if (!std::isfinite(z)) |
| return HomogeneousCoordinate(0.0, 0.0, 0.0, 1.0); |
| |
| HomogeneousCoordinate result(p.x(), p.y(), z, 1.0); |
| transform.matrix().mapScalars(result.vec, result.vec); |
| return result; |
| } |
| |
| static HomogeneousCoordinate ProjectHomogeneousPoint( |
| const gfx::Transform& transform, |
| const gfx::PointF& p, |
| bool* clipped) { |
| HomogeneousCoordinate h = ProjectHomogeneousPoint(transform, p); |
| *clipped = h.w() <= 0; |
| return h; |
| } |
| |
| static HomogeneousCoordinate MapHomogeneousPoint( |
| const gfx::Transform& transform, |
| const gfx::Point3F& p) { |
| HomogeneousCoordinate result(p.x(), p.y(), p.z(), 1.0); |
| transform.matrix().mapScalars(result.vec, result.vec); |
| return result; |
| } |
| |
| namespace { |
| |
| // This is the tolerance for detecting an eyepoint-aligned edge. |
| const float kStationaryPointEpsilon = 0.00001f; |
| |
| } // namespace |
| |
| static void homogeneousLimitAtZero(SkScalar a1, |
| SkScalar w1, |
| SkScalar a2, |
| SkScalar w2, |
| float t, |
| float* limit) { |
| if (std::abs(a1 * w2 / w1 / a2 - 1.0f) > kStationaryPointEpsilon) { |
| // We are going to explode towards an infinity, but we choose the one that |
| // corresponds to the one on the positive side of w. |
| if (((1.0f - t) * a1 + t * a2) > 0) { |
| *limit = HomogeneousCoordinate::kInfiniteCoordinate; |
| } else { |
| *limit = -HomogeneousCoordinate::kInfiniteCoordinate; |
| } |
| } else { |
| *limit = a1 / w1; // (== a2 / w2) && == (1.0f - t) * a1 / w1 + t * a2 / w2 |
| } |
| } |
| |
| static gfx::PointF ComputeClippedCartesianPoint2dForEdge( |
| const HomogeneousCoordinate& h1, |
| const HomogeneousCoordinate& h2) { |
| // Points h1 and h2 form a line in 4d, and any point on that line can be |
| // represented as an interpolation between h1 and h2: |
| // p = (1-t) h1 + (t) h2 |
| // |
| // We want to compute the limit in 2 space of |
| // x = ((1-t) h1.x + (t) h2.x) / ((1-t) h1.w + (t) h2.w) |
| // y = ((1-t) h1.y + (t) h2.y) / ((1-t) h1.w + (t) h2.w) |
| // as ((1-t) h1.w + (t) h2.w) -> 0+ |
| |
| // The only answers to this are h1.x/h1.w == h2.x/h2.w, +/- infinity |
| // i.e., either the coordinate is not moving, or is trending to one |
| // infinity or the other. |
| |
| // This assertion isn't really as strong as it looks because |
| // std::isfinite(h1.w()) or std::isfinite(h2.w()) might not be true |
| // (and they could be NaN). |
| // TODO(crbug.com/1219622): We should be able to assert something |
| // stronger here, and avoid dependencies on undefined floating point |
| // behavior. |
| DCHECK_NE(h1.w() <= 0, h2.w() <= 0); |
| |
| float t = h1.w() / (h1.w() - h2.w()); |
| float x; |
| float y; |
| |
| homogeneousLimitAtZero(h1.x(), h1.w(), h2.x(), h2.w(), t, &x); |
| homogeneousLimitAtZero(h1.y(), h1.w(), h2.y(), h2.w(), t, &y); |
| |
| return gfx::PointF(x, y); |
| } |
| |
| static void homogeneousLimitNearZero(SkScalar a1, |
| SkScalar w1, |
| SkScalar a2, |
| SkScalar w2, |
| float t, |
| float* limit) { |
| if (std::abs(a1 * w2 / w1 / a2 - 1.0f) > kStationaryPointEpsilon) { |
| // t has been computed so that w is near but not at zero. |
| *limit = ((1.0f - t) * a1 + t * a2) / ((1.0f - t) * w1 + t * w2); |
| // std::abs(*limit) should now be somewhere near |
| // HomogeneousCoordinate::kInfiniteCoordinate, preferably smaller than it, |
| // but there are edge cases where it will be larger (for example, if the |
| // point where a crosses 0 is very close to the point where w crosses 0), |
| // so it's hard to DCHECK() that this is the case. |
| } else { |
| *limit = a1 / w1; // (== a2 / w2) && == (1.0f - t) * a1 / w1 + t * a2 / w2 |
| } |
| } |
| |
| static gfx::Point3F ComputeClippedCartesianPoint3dForEdge( |
| const HomogeneousCoordinate& h1, |
| const HomogeneousCoordinate& h2) { |
| // Points h1 and h2 form a line in 4d, and any point on that line can be |
| // represented as an interpolation between h1 and h2: |
| // p = (1-t) h1 + (t) h2 |
| // |
| // We want to compute the limit in 3 space of |
| // x = ((1-t) h1.x + (t) h2.x) / ((1-t) h1.w + (t) h2.w) |
| // y = ((1-t) h1.y + (t) h2.y) / ((1-t) h1.w + (t) h2.w) |
| // z = ((1-t) h1.z + (t) h2.z) / ((1-t) h1.w + (t) h2.w) |
| // as ((1-t) h1.w + (t) h2.w) -> 0+ |
| |
| // The only answers to this are h1.x/h1.w == h2.x/h2.w, +/- infinity |
| // i.e., either the coordinate is not moving, or is trending to one |
| // infinity or the other. |
| |
| // When we clamp to HomogeneousCoordinate::kInfiniteCoordinate we want |
| // to keep the result in the correct plane, which we do by computing |
| // a t that will result in the largest (in absolute value) of x, y, or |
| // z being HomogeneousCoordinate::kInfiniteCoordinate |
| |
| // This assertion isn't really as strong as it looks because |
| // std::isfinite(h1.w()) or std::isfinite(h2.w()) might not be true |
| // (and they could be NaN). |
| // TODO(crbug.com/1219622): We should be able to assert something |
| // stronger here, and avoid dependencies on undefined floating point |
| // behavior. |
| DCHECK_NE(h1.w() <= 0, h2.w() <= 0); |
| |
| float w_diff = h1.w() - h2.w(); |
| float t = h1.w() / w_diff; |
| float max_numerator = std::max({std::abs((1.0f - t) * h1.x() + t * h2.x()), |
| std::abs((1.0f - t) * h1.y() + t * h2.y()), |
| std::abs((1.0f - t) * h1.z() + t * h2.z())}); |
| |
| // Shift t away from the point where w is zero, far enough so that the |
| // largest of the resulting x, y, and z will be about |
| // kInfiniteCoordinate. Add an extra epsilon() / 2.0 so that there's |
| // always enough movement (in case t_shift is very small, which it |
| // often is). |
| const float t_shift = |
| max_numerator / w_diff / HomogeneousCoordinate::kInfiniteCoordinate; |
| constexpr float half_epsilon = std::numeric_limits<float>::epsilon() / 2.0f; |
| DCHECK_EQ(w_diff > 0, t_shift > 0); |
| if (w_diff > 0) { |
| t = std::max(0.0f, t - (t_shift + half_epsilon)); |
| } else { |
| t = std::min(1.0f, t - (t_shift - half_epsilon)); |
| } |
| |
| float x; |
| float y; |
| float z; |
| |
| homogeneousLimitNearZero(h1.x(), h1.w(), h2.x(), h2.w(), t, &x); |
| homogeneousLimitNearZero(h1.y(), h1.w(), h2.y(), h2.w(), t, &y); |
| homogeneousLimitNearZero(h1.z(), h1.w(), h2.z(), h2.w(), t, &z); |
| |
| return gfx::Point3F(x, y, z); |
| } |
| |
| static inline void ExpandBoundsToIncludePoint(float* xmin, |
| float* xmax, |
| float* ymin, |
| float* ymax, |
| const gfx::PointF& p) { |
| *xmin = std::min(p.x(), *xmin); |
| *xmax = std::max(p.x(), *xmax); |
| *ymin = std::min(p.y(), *ymin); |
| *ymax = std::max(p.y(), *ymax); |
| } |
| |
| static inline bool IsNearlyTheSame(float f, float g) { |
| // The idea behind this is to use this fraction of the larger of the |
| // two numbers as the limit of the difference. This breaks down near |
| // zero, so we reuse this as the minimum absolute size we will use |
| // for the base of the scale too. |
| static const float epsilon_scale = 0.00001f; |
| return std::abs(f - g) < |
| epsilon_scale * std::max({std::abs(f), std::abs(g), epsilon_scale}); |
| } |
| |
| static inline bool IsNearlyTheSame(const gfx::PointF& lhs, |
| const gfx::PointF& rhs) { |
| return IsNearlyTheSame(lhs.x(), rhs.x()) && IsNearlyTheSame(lhs.y(), rhs.y()); |
| } |
| |
| static inline bool IsNearlyTheSame(const gfx::Point3F& lhs, |
| const gfx::Point3F& rhs) { |
| return IsNearlyTheSame(lhs.x(), rhs.x()) && |
| IsNearlyTheSame(lhs.y(), rhs.y()) && IsNearlyTheSame(lhs.z(), rhs.z()); |
| } |
| |
| static inline void AddVertexToClippedQuad3d(const gfx::Point3F& new_vertex, |
| gfx::Point3F clipped_quad[6], |
| int* num_vertices_in_clipped_quad, |
| bool* need_to_clamp) { |
| if (*num_vertices_in_clipped_quad > 0 && |
| IsNearlyTheSame(clipped_quad[*num_vertices_in_clipped_quad - 1], |
| new_vertex)) |
| return; |
| |
| DCHECK_LT(*num_vertices_in_clipped_quad, 6); |
| clipped_quad[*num_vertices_in_clipped_quad] = new_vertex; |
| (*num_vertices_in_clipped_quad)++; |
| if (new_vertex.x() < -HomogeneousCoordinate::kInfiniteCoordinate || |
| new_vertex.x() > HomogeneousCoordinate::kInfiniteCoordinate || |
| new_vertex.y() < -HomogeneousCoordinate::kInfiniteCoordinate || |
| new_vertex.y() > HomogeneousCoordinate::kInfiniteCoordinate || |
| new_vertex.z() < -HomogeneousCoordinate::kInfiniteCoordinate || |
| new_vertex.z() > HomogeneousCoordinate::kInfiniteCoordinate) { |
| *need_to_clamp = true; |
| } |
| } |
| |
| gfx::Rect MathUtil::MapEnclosingClippedRect(const gfx::Transform& transform, |
| const gfx::Rect& src_rect) { |
| return MapEnclosingClippedRectIgnoringError(transform, src_rect, 0.f); |
| } |
| |
| gfx::Rect MathUtil::MapEnclosingClippedRectIgnoringError( |
| const gfx::Transform& transform, |
| const gfx::Rect& src_rect, |
| float ignore_error) { |
| if (transform.IsIdentityOrIntegerTranslation()) { |
| gfx::Vector2d offset(static_cast<int>(transform.matrix().rc(0, 3)), |
| static_cast<int>(transform.matrix().rc(1, 3))); |
| return src_rect + offset; |
| } |
| gfx::RectF mapped_rect = MapClippedRect(transform, gfx::RectF(src_rect)); |
| |
| // gfx::ToEnclosingRect crashes if called on a RectF with any NaN coordinate. |
| if (std::isnan(mapped_rect.x()) || std::isnan(mapped_rect.y()) || |
| std::isnan(mapped_rect.right()) || std::isnan(mapped_rect.bottom())) |
| return gfx::Rect(); |
| |
| return gfx::ToEnclosingRectIgnoringError(mapped_rect, ignore_error); |
| } |
| |
| gfx::RectF MathUtil::MapClippedRect(const gfx::Transform& transform, |
| const gfx::RectF& src_rect) { |
| if (transform.IsIdentityOrTranslation()) { |
| gfx::Vector2dF offset(transform.matrix().rc(0, 3), |
| transform.matrix().rc(1, 3)); |
| return src_rect + offset; |
| } |
| |
| // Apply the transform, but retain the result in homogeneous coordinates. |
| |
| SkScalar quad[4 * 2]; // input: 4 x 2D points |
| quad[0] = src_rect.x(); |
| quad[1] = src_rect.y(); |
| quad[2] = src_rect.right(); |
| quad[3] = src_rect.y(); |
| quad[4] = src_rect.right(); |
| quad[5] = src_rect.bottom(); |
| quad[6] = src_rect.x(); |
| quad[7] = src_rect.bottom(); |
| |
| SkScalar result[4 * 4]; // output: 4 x 4D homogeneous points |
| transform.matrix().map2(quad, 4, result); |
| |
| HomogeneousCoordinate hc0(result[0], result[1], result[2], result[3]); |
| HomogeneousCoordinate hc1(result[4], result[5], result[6], result[7]); |
| HomogeneousCoordinate hc2(result[8], result[9], result[10], result[11]); |
| HomogeneousCoordinate hc3(result[12], result[13], result[14], result[15]); |
| return ComputeEnclosingClippedRect(hc0, hc1, hc2, hc3); |
| } |
| |
| gfx::Rect MathUtil::ProjectEnclosingClippedRect(const gfx::Transform& transform, |
| const gfx::Rect& src_rect) { |
| if (transform.IsIdentityOrIntegerTranslation()) { |
| gfx::Vector2d offset(static_cast<int>(transform.matrix().rc(0, 3)), |
| static_cast<int>(transform.matrix().rc(1, 3))); |
| return src_rect + offset; |
| } |
| gfx::RectF projected_rect = |
| ProjectClippedRect(transform, gfx::RectF(src_rect)); |
| |
| // gfx::ToEnclosingRect crashes if called on a RectF with any NaN coordinate. |
| if (std::isnan(projected_rect.x()) || std::isnan(projected_rect.y()) || |
| std::isnan(projected_rect.right()) || std::isnan(projected_rect.bottom())) |
| return gfx::Rect(); |
| |
| return gfx::ToEnclosingRect(projected_rect); |
| } |
| |
| gfx::RectF MathUtil::ProjectClippedRect(const gfx::Transform& transform, |
| const gfx::RectF& src_rect) { |
| if (transform.IsIdentityOrTranslation()) { |
| gfx::Vector2dF offset(transform.matrix().rc(0, 3), |
| transform.matrix().rc(1, 3)); |
| return src_rect + offset; |
| } |
| |
| // Perform the projection, but retain the result in homogeneous coordinates. |
| gfx::QuadF q = gfx::QuadF(src_rect); |
| HomogeneousCoordinate h1 = ProjectHomogeneousPoint(transform, q.p1()); |
| HomogeneousCoordinate h2 = ProjectHomogeneousPoint(transform, q.p2()); |
| HomogeneousCoordinate h3 = ProjectHomogeneousPoint(transform, q.p3()); |
| HomogeneousCoordinate h4 = ProjectHomogeneousPoint(transform, q.p4()); |
| |
| return ComputeEnclosingClippedRect(h1, h2, h3, h4); |
| } |
| |
| gfx::QuadF MathUtil::InverseMapQuadToLocalSpace( |
| const gfx::Transform& device_transform, |
| const gfx::QuadF& device_quad) { |
| gfx::Transform inverse_device_transform(gfx::Transform::kSkipInitialization); |
| DCHECK(device_transform.IsInvertible()); |
| DCHECK(device_transform.IsFlat()); |
| bool did_invert = device_transform.GetInverse(&inverse_device_transform); |
| DCHECK(did_invert); |
| bool clipped = false; |
| gfx::QuadF local_quad = |
| MathUtil::MapQuad(inverse_device_transform, device_quad, &clipped); |
| // We should not DCHECK(!clipped) here, because anti-aliasing inflation may |
| // cause device_quad to become clipped. To our knowledge this scenario does |
| // not need to be handled differently than the unclipped case. |
| return local_quad; |
| } |
| |
| gfx::Rect MathUtil::MapEnclosedRectWith2dAxisAlignedTransform( |
| const gfx::Transform& transform, |
| const gfx::Rect& rect) { |
| DCHECK(transform.Preserves2dAxisAlignment()); |
| DCHECK_GT(transform.matrix().rc(3, 3), 0); |
| DCHECK(std::isnormal(transform.matrix().rc(3, 3))); |
| |
| if (transform.IsIdentityOrIntegerTranslation()) { |
| gfx::Vector2d offset(static_cast<int>(transform.matrix().rc(0, 3)), |
| static_cast<int>(transform.matrix().rc(1, 3))); |
| return rect + offset; |
| } |
| if (transform.IsIdentityOrTranslation()) { |
| gfx::Vector2dF offset(transform.matrix().rc(0, 3), |
| transform.matrix().rc(1, 3)); |
| return gfx::ToEnclosedRect(gfx::RectF(rect) + offset); |
| } |
| |
| SkScalar quad[2 * 2]; // input: 2 x 2D points |
| quad[0] = rect.x(); |
| quad[1] = rect.y(); |
| quad[2] = rect.right(); |
| quad[3] = rect.bottom(); |
| |
| SkScalar result[4 * 2]; // output: 2 x 4D homogeneous points |
| transform.matrix().map2(quad, 2, result); |
| |
| HomogeneousCoordinate hc0(result[0], result[1], result[2], result[3]); |
| HomogeneousCoordinate hc1(result[4], result[5], result[6], result[7]); |
| DCHECK(!hc0.ShouldBeClipped()); |
| DCHECK(!hc1.ShouldBeClipped()); |
| |
| gfx::PointF top_left(hc0.CartesianPoint2d()); |
| gfx::PointF bottom_right(hc1.CartesianPoint2d()); |
| return gfx::ToEnclosedRect(gfx::BoundingRect(top_left, bottom_right)); |
| } |
| |
| bool MathUtil::MapClippedQuad3d(const gfx::Transform& transform, |
| const gfx::QuadF& src_quad, |
| gfx::Point3F clipped_quad[6], |
| int* num_vertices_in_clipped_quad) { |
| // This is different from the 2D version because, when we clamp |
| // coordinates to [-HomogeneousCoordinate::kInfiniteCoordinate, |
| // HomogeneousCoordinate::kInfiniteCoordinate], we need to do the |
| // clamping while keeping the points coplanar. |
| |
| HomogeneousCoordinate h1 = |
| MapHomogeneousPoint(transform, gfx::Point3F(src_quad.p1())); |
| HomogeneousCoordinate h2 = |
| MapHomogeneousPoint(transform, gfx::Point3F(src_quad.p2())); |
| HomogeneousCoordinate h3 = |
| MapHomogeneousPoint(transform, gfx::Point3F(src_quad.p3())); |
| HomogeneousCoordinate h4 = |
| MapHomogeneousPoint(transform, gfx::Point3F(src_quad.p4())); |
| |
| // The order of adding the vertices to the array is chosen so that |
| // clockwise / counter-clockwise orientation is retained. |
| |
| *num_vertices_in_clipped_quad = 0; |
| bool need_to_clamp = false; |
| |
| if (!h1.ShouldBeClipped()) { |
| AddVertexToClippedQuad3d(h1.CartesianPoint3dUnclamped(), clipped_quad, |
| num_vertices_in_clipped_quad, &need_to_clamp); |
| } |
| |
| if (h1.ShouldBeClipped() ^ h2.ShouldBeClipped()) { |
| AddVertexToClippedQuad3d(ComputeClippedCartesianPoint3dForEdge(h1, h2), |
| clipped_quad, num_vertices_in_clipped_quad, |
| &need_to_clamp); |
| } |
| |
| if (!h2.ShouldBeClipped()) { |
| AddVertexToClippedQuad3d(h2.CartesianPoint3dUnclamped(), clipped_quad, |
| num_vertices_in_clipped_quad, &need_to_clamp); |
| } |
| |
| if (h2.ShouldBeClipped() ^ h3.ShouldBeClipped()) { |
| AddVertexToClippedQuad3d(ComputeClippedCartesianPoint3dForEdge(h2, h3), |
| clipped_quad, num_vertices_in_clipped_quad, |
| &need_to_clamp); |
| } |
| |
| if (!h3.ShouldBeClipped()) { |
| AddVertexToClippedQuad3d(h3.CartesianPoint3dUnclamped(), clipped_quad, |
| num_vertices_in_clipped_quad, &need_to_clamp); |
| } |
| |
| if (h3.ShouldBeClipped() ^ h4.ShouldBeClipped()) { |
| AddVertexToClippedQuad3d(ComputeClippedCartesianPoint3dForEdge(h3, h4), |
| clipped_quad, num_vertices_in_clipped_quad, |
| &need_to_clamp); |
| } |
| |
| if (!h4.ShouldBeClipped()) { |
| AddVertexToClippedQuad3d(h4.CartesianPoint3dUnclamped(), clipped_quad, |
| num_vertices_in_clipped_quad, &need_to_clamp); |
| } |
| |
| if (h4.ShouldBeClipped() ^ h1.ShouldBeClipped()) { |
| AddVertexToClippedQuad3d(ComputeClippedCartesianPoint3dForEdge(h4, h1), |
| clipped_quad, num_vertices_in_clipped_quad, |
| &need_to_clamp); |
| } |
| |
| if (*num_vertices_in_clipped_quad > 2 && |
| IsNearlyTheSame(clipped_quad[0], |
| clipped_quad[*num_vertices_in_clipped_quad - 1])) |
| *num_vertices_in_clipped_quad -= 1; |
| |
| if (need_to_clamp) { |
| // Some of the values need to be clamped, but we need to keep them |
| // coplanar while doing so. |
| |
| // First, build a normal vector to the plane by averaging the |
| // cross products of adjacent edges. |
| gfx::Vector3dF normal(0.0f, 0.0f, 0.0f); |
| if (*num_vertices_in_clipped_quad > 2) { |
| gfx::Vector3dF loop_vector = |
| clipped_quad[0] - clipped_quad[*num_vertices_in_clipped_quad - 1]; |
| gfx::Vector3dF prev_vector(loop_vector); |
| for (int i = 1; i < *num_vertices_in_clipped_quad; ++i) { |
| gfx::Vector3dF cur_vector = clipped_quad[i] - clipped_quad[i - 1]; |
| normal += CrossProduct(prev_vector, cur_vector); |
| prev_vector = cur_vector; |
| } |
| normal += CrossProduct(prev_vector, loop_vector); |
| } |
| |
| bool clamp_by_points = false; |
| float length = normal.Length(); |
| if (std::isnormal(length)) { // exclude 0, denormals, +/- inf, NaN |
| normal.Scale(1.0f / length); |
| |
| // Find the vector to the point in the plane closest to (0,0,0). |
| gfx::Vector3dF shortest_from_zero(normal); |
| shortest_from_zero.Scale( |
| DotProduct(normal, clipped_quad[0] - gfx::Point3F(0.0f, 0.0f, 0.0f))); |
| |
| // Find the the point in the plane that is at x=0 and y=0 |
| float z_at_xy_zero = 0.0f; |
| if (shortest_from_zero.x() == 0.0f && shortest_from_zero.y() == 0.0f) { |
| z_at_xy_zero = shortest_from_zero.z(); |
| } else if (shortest_from_zero.z() != 0) { |
| // Compute the vector v pointing from the shortest_from_zero |
| // point to the point with x=0 and y=0. If both v and normal |
| // are projected into the x/y plane, they should point in |
| // opposite directions. |
| gfx::Vector3dF v = CrossProduct( |
| normal, CrossProduct(gfx::Vector3dF(0.0f, 0.0f, 1.0f), normal)); |
| DCHECK(std::abs(normal.x() * v.y() - normal.y() * v.x()) < 0.00001f); |
| // It doesn't matter whether we use x or y, unless one of them |
| // is zero or very close to it. |
| float r = std::abs(v.x()) > std::abs(v.y()) |
| ? shortest_from_zero.x() / v.x() |
| : shortest_from_zero.y() / v.y(); |
| z_at_xy_zero = shortest_from_zero.z() - v.z() * r; |
| } else { |
| // Plane is parallel to the z axis. This means it's not |
| // visible, so just fall back to clamping by points. |
| clamp_by_points = true; |
| } |
| |
| if (!clamp_by_points) { |
| // If z_at_xy_zero is more than 3/4 of kInfiniteCoordinate |
| // distance from zero, move everything in the z axis so |
| // z_at_xy_zero is that distance from zero, so that we don't end |
| // up clamping away the parts that fit within what's likely to |
| // be the visible area. |
| constexpr float max_distance = |
| 0.75 * HomogeneousCoordinate::kInfiniteCoordinate; |
| if (std::abs(z_at_xy_zero) > max_distance) { |
| float z_delta; |
| if (z_at_xy_zero > 0) { |
| z_delta = max_distance - z_at_xy_zero; |
| } else { |
| z_delta = -max_distance - z_at_xy_zero; |
| } |
| for (int i = 0; i < *num_vertices_in_clipped_quad; ++i) { |
| clipped_quad[i].set_z(clipped_quad[i].z() + z_delta); |
| } |
| z_at_xy_zero += z_delta; |
| } |
| |
| // Move all the points towards (0, 0, z_at_xy_zero) until all |
| // their coordinates are less than kInfiniteCoordinate. |
| for (int i = 0; i < *num_vertices_in_clipped_quad; ++i) { |
| gfx::Point3F& point = clipped_quad[i]; |
| float t = 1.0f; |
| |
| float x_abs = std::abs(point.x()); |
| if (x_abs > HomogeneousCoordinate::kInfiniteCoordinate) { |
| t = std::min(t, HomogeneousCoordinate::kInfiniteCoordinate / x_abs); |
| } |
| |
| float y_abs = std::abs(point.y()); |
| if (y_abs > HomogeneousCoordinate::kInfiniteCoordinate) { |
| t = std::min(t, HomogeneousCoordinate::kInfiniteCoordinate / y_abs); |
| } |
| |
| float z = point.z(); |
| if (std::abs(z) > HomogeneousCoordinate::kInfiniteCoordinate) { |
| // From the clamping to max_distance above, we should have |
| // made std::abs(z_at_xy_zero) < kInfiniteCoordinate. |
| // However, if it started off very large we might not have. |
| float z_at_xy_zero_clamped = |
| std::min(float{HomogeneousCoordinate::kInfiniteCoordinate}, |
| std::max(-HomogeneousCoordinate::kInfiniteCoordinate, |
| z_at_xy_zero)); |
| float z_offset = z - z_at_xy_zero_clamped; |
| float z_space = |
| (z > 0 ? HomogeneousCoordinate::kInfiniteCoordinate |
| : -HomogeneousCoordinate::kInfiniteCoordinate) - |
| z_at_xy_zero_clamped; |
| DCHECK_NE(z_offset, 0.0f); |
| DCHECK_NE(z_space, 0.0f); |
| DCHECK_EQ(z_offset > 0, z_space > 0); |
| t = std::min(t, z_space / z_offset); |
| } |
| |
| if (t != 1.0f) { |
| DCHECK(0.0f <= t && t < 1.0f); |
| point.set_x(t * point.x()); |
| point.set_y(t * point.y()); |
| point.set_z((1.0f - t) * z_at_xy_zero + t * point.z()); |
| } |
| } |
| } |
| } else { |
| // Our points were colinear, so there's no plane to maintain. |
| clamp_by_points = true; |
| } |
| |
| if (clamp_by_points) { |
| // Just clamp each point separately in each axis, just like we do |
| // for 2D. |
| for (int i = 0; i < *num_vertices_in_clipped_quad; ++i) { |
| gfx::Point3F& point = clipped_quad[i]; |
| point.set_x( |
| base::clamp(point.x(), -HomogeneousCoordinate::kInfiniteCoordinate, |
| float{HomogeneousCoordinate::kInfiniteCoordinate})); |
| point.set_y( |
| base::clamp(point.y(), -HomogeneousCoordinate::kInfiniteCoordinate, |
| float{HomogeneousCoordinate::kInfiniteCoordinate})); |
| point.set_z( |
| base::clamp(point.z(), -HomogeneousCoordinate::kInfiniteCoordinate, |
| float{HomogeneousCoordinate::kInfiniteCoordinate})); |
| } |
| } |
| } |
| |
| DCHECK_LE(*num_vertices_in_clipped_quad, 6); |
| return (*num_vertices_in_clipped_quad >= 4); |
| } |
| |
| gfx::RectF MathUtil::ComputeEnclosingRectOfVertices( |
| const gfx::PointF vertices[], |
| int num_vertices) { |
| if (num_vertices < 2) |
| return gfx::RectF(); |
| |
| float xmin = std::numeric_limits<float>::max(); |
| float xmax = -std::numeric_limits<float>::max(); |
| float ymin = std::numeric_limits<float>::max(); |
| float ymax = -std::numeric_limits<float>::max(); |
| |
| for (int i = 0; i < num_vertices; ++i) |
| ExpandBoundsToIncludePoint(&xmin, &xmax, &ymin, &ymax, vertices[i]); |
| |
| return gfx::RectF(gfx::PointF(xmin, ymin), |
| gfx::SizeF(xmax - xmin, ymax - ymin)); |
| } |
| |
| gfx::RectF MathUtil::ComputeEnclosingClippedRect( |
| const HomogeneousCoordinate& h1, |
| const HomogeneousCoordinate& h2, |
| const HomogeneousCoordinate& h3, |
| const HomogeneousCoordinate& h4) { |
| // This function performs clipping as necessary and computes the enclosing 2d |
| // gfx::RectF of the vertices. Doing these two steps simultaneously allows us |
| // to avoid the overhead of storing an unknown number of clipped vertices. |
| |
| // If no vertices on the quad are clipped, then we can simply return the |
| // enclosing rect directly. |
| bool something_clipped = h1.ShouldBeClipped() || h2.ShouldBeClipped() || |
| h3.ShouldBeClipped() || h4.ShouldBeClipped(); |
| if (!something_clipped) { |
| gfx::QuadF mapped_quad = gfx::QuadF(h1.CartesianPoint2d(), |
| h2.CartesianPoint2d(), |
| h3.CartesianPoint2d(), |
| h4.CartesianPoint2d()); |
| return mapped_quad.BoundingBox(); |
| } |
| |
| bool everything_clipped = h1.ShouldBeClipped() && h2.ShouldBeClipped() && |
| h3.ShouldBeClipped() && h4.ShouldBeClipped(); |
| if (everything_clipped) |
| return gfx::RectF(); |
| |
| float xmin = std::numeric_limits<float>::max(); |
| float xmax = -std::numeric_limits<float>::max(); |
| float ymin = std::numeric_limits<float>::max(); |
| float ymax = -std::numeric_limits<float>::max(); |
| |
| if (!h1.ShouldBeClipped()) |
| ExpandBoundsToIncludePoint(&xmin, &xmax, &ymin, &ymax, |
| h1.CartesianPoint2d()); |
| |
| if (h1.ShouldBeClipped() ^ h2.ShouldBeClipped()) |
| ExpandBoundsToIncludePoint(&xmin, &xmax, &ymin, &ymax, |
| ComputeClippedCartesianPoint2dForEdge(h1, h2)); |
| |
| if (!h2.ShouldBeClipped()) |
| ExpandBoundsToIncludePoint(&xmin, &xmax, &ymin, &ymax, |
| h2.CartesianPoint2d()); |
| |
| if (h2.ShouldBeClipped() ^ h3.ShouldBeClipped()) |
| ExpandBoundsToIncludePoint(&xmin, &xmax, &ymin, &ymax, |
| ComputeClippedCartesianPoint2dForEdge(h2, h3)); |
| |
| if (!h3.ShouldBeClipped()) |
| ExpandBoundsToIncludePoint(&xmin, &xmax, &ymin, &ymax, |
| h3.CartesianPoint2d()); |
| |
| if (h3.ShouldBeClipped() ^ h4.ShouldBeClipped()) |
| ExpandBoundsToIncludePoint(&xmin, &xmax, &ymin, &ymax, |
| ComputeClippedCartesianPoint2dForEdge(h3, h4)); |
| |
| if (!h4.ShouldBeClipped()) |
| ExpandBoundsToIncludePoint(&xmin, &xmax, &ymin, &ymax, |
| h4.CartesianPoint2d()); |
| |
| if (h4.ShouldBeClipped() ^ h1.ShouldBeClipped()) |
| ExpandBoundsToIncludePoint(&xmin, &xmax, &ymin, &ymax, |
| ComputeClippedCartesianPoint2dForEdge(h4, h1)); |
| |
| return gfx::RectF(gfx::PointF(xmin, ymin), |
| gfx::SizeF(xmax - xmin, ymax - ymin)); |
| } |
| |
| gfx::QuadF MathUtil::MapQuad(const gfx::Transform& transform, |
| const gfx::QuadF& q, |
| bool* clipped) { |
| if (transform.IsIdentityOrTranslation()) { |
| gfx::QuadF mapped_quad(q); |
| mapped_quad += gfx::Vector2dF(transform.matrix().rc(0, 3), |
| transform.matrix().rc(1, 3)); |
| *clipped = false; |
| return mapped_quad; |
| } |
| |
| HomogeneousCoordinate h1 = |
| MapHomogeneousPoint(transform, gfx::Point3F(q.p1())); |
| HomogeneousCoordinate h2 = |
| MapHomogeneousPoint(transform, gfx::Point3F(q.p2())); |
| HomogeneousCoordinate h3 = |
| MapHomogeneousPoint(transform, gfx::Point3F(q.p3())); |
| HomogeneousCoordinate h4 = |
| MapHomogeneousPoint(transform, gfx::Point3F(q.p4())); |
| |
| *clipped = h1.ShouldBeClipped() || h2.ShouldBeClipped() || |
| h3.ShouldBeClipped() || h4.ShouldBeClipped(); |
| |
| // Result will be invalid if clipped == true. But, compute it anyway just in |
| // case, to emulate existing behavior. |
| return gfx::QuadF(h1.CartesianPoint2d(), |
| h2.CartesianPoint2d(), |
| h3.CartesianPoint2d(), |
| h4.CartesianPoint2d()); |
| } |
| |
| gfx::PointF MathUtil::MapPoint(const gfx::Transform& transform, |
| const gfx::PointF& p, |
| bool* clipped) { |
| HomogeneousCoordinate h = MapHomogeneousPoint(transform, gfx::Point3F(p)); |
| |
| if (h.w() > 0) { |
| *clipped = false; |
| return h.CartesianPoint2d(); |
| } |
| |
| // The cartesian coordinates will be invalid after dividing by w. |
| *clipped = true; |
| |
| // Avoid dividing by w if w == 0. |
| if (!h.w()) |
| return gfx::PointF(); |
| |
| // This return value will be invalid because clipped == true, but (1) users of |
| // this code should be ignoring the return value when clipped == true anyway, |
| // and (2) this behavior is more consistent with existing behavior of WebKit |
| // transforms if the user really does not ignore the return value. |
| return h.CartesianPoint2d(); |
| } |
| |
| gfx::PointF MathUtil::ProjectPoint(const gfx::Transform& transform, |
| const gfx::PointF& p, |
| bool* clipped) { |
| HomogeneousCoordinate h = ProjectHomogeneousPoint(transform, p, clipped); |
| // Avoid dividing by w if w == 0. |
| if (!h.w()) |
| return gfx::PointF(); |
| |
| // This return value will be invalid if clipped == true, but (1) users of |
| // this code should be ignoring the return value when clipped == true anyway, |
| // and (2) this behavior is more consistent with existing behavior of WebKit |
| // transforms if the user really does not ignore the return value. |
| return h.CartesianPoint2d(); |
| } |
| |
| gfx::Point3F MathUtil::ProjectPoint3D(const gfx::Transform& transform, |
| const gfx::PointF& p, |
| bool* clipped) { |
| HomogeneousCoordinate h = ProjectHomogeneousPoint(transform, p, clipped); |
| if (!h.w()) |
| return gfx::Point3F(); |
| return h.CartesianPoint3d(); |
| } |
| |
| gfx::RectF MathUtil::ScaleRectProportional(const gfx::RectF& input_outer_rect, |
| const gfx::RectF& scale_outer_rect, |
| const gfx::RectF& scale_inner_rect) { |
| gfx::RectF output_inner_rect = input_outer_rect; |
| float scale_rect_to_input_scale_x = |
| scale_outer_rect.width() / input_outer_rect.width(); |
| float scale_rect_to_input_scale_y = |
| scale_outer_rect.height() / input_outer_rect.height(); |
| |
| gfx::Vector2dF top_left_diff = |
| scale_inner_rect.origin() - scale_outer_rect.origin(); |
| gfx::Vector2dF bottom_right_diff = |
| scale_inner_rect.bottom_right() - scale_outer_rect.bottom_right(); |
| output_inner_rect.Inset( |
| gfx::InsetsF::TLBR(top_left_diff.y() / scale_rect_to_input_scale_y, |
| top_left_diff.x() / scale_rect_to_input_scale_x, |
| -bottom_right_diff.y() / scale_rect_to_input_scale_y, |
| -bottom_right_diff.x() / scale_rect_to_input_scale_x)); |
| return output_inner_rect; |
| } |
| |
| float MathUtil::SmallestAngleBetweenVectors(const gfx::Vector2dF& v1, |
| const gfx::Vector2dF& v2) { |
| double dot_product = gfx::DotProduct(v1, v2) / v1.Length() / v2.Length(); |
| // Clamp to compensate for rounding errors. |
| dot_product = base::clamp(dot_product, -1.0, 1.0); |
| return static_cast<float>(gfx::RadToDeg(std::acos(dot_product))); |
| } |
| |
| gfx::Vector2dF MathUtil::ProjectVector(const gfx::Vector2dF& source, |
| const gfx::Vector2dF& destination) { |
| float projected_length = |
| gfx::DotProduct(source, destination) / destination.LengthSquared(); |
| return gfx::Vector2dF(projected_length * destination.x(), |
| projected_length * destination.y()); |
| } |
| |
| bool MathUtil::FromValue(const base::Value* raw_value, gfx::Rect* out_rect) { |
| if (!raw_value->is_list()) |
| return false; |
| |
| base::Value::ConstListView list_view = raw_value->GetListDeprecated(); |
| |
| if (list_view.size() != 4) |
| return false; |
| |
| for (const auto& val : list_view) { |
| if (!val.is_int()) { |
| return false; |
| } |
| } |
| |
| int x = list_view[0].GetInt(); |
| int y = list_view[1].GetInt(); |
| int w = list_view[2].GetInt(); |
| int h = list_view[3].GetInt(); |
| |
| *out_rect = gfx::Rect(x, y, w, h); |
| return true; |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::Size& s, |
| base::trace_event::TracedValue* res) { |
| res->BeginDictionary(name); |
| res->SetDouble("width", s.width()); |
| res->SetDouble("height", s.height()); |
| res->EndDictionary(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::SizeF& s, |
| base::trace_event::TracedValue* res) { |
| res->BeginDictionary(name); |
| res->SetDouble("width", s.width()); |
| res->SetDouble("height", s.height()); |
| res->EndDictionary(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::Rect& r, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendInteger(r.x()); |
| res->AppendInteger(r.y()); |
| res->AppendInteger(r.width()); |
| res->AppendInteger(r.height()); |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::Point& pt, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendInteger(pt.x()); |
| res->AppendInteger(pt.y()); |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::PointF& pt, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendDouble(pt.x()); |
| res->AppendDouble(pt.y()); |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::Point3F& pt, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendDouble(pt.x()); |
| res->AppendDouble(pt.y()); |
| res->AppendDouble(pt.z()); |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::Vector2d& v, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendInteger(v.x()); |
| res->AppendInteger(v.y()); |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::Vector2dF& v, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendDouble(v.x()); |
| res->AppendDouble(v.y()); |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::QuadF& q, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendDouble(q.p1().x()); |
| res->AppendDouble(q.p1().y()); |
| res->AppendDouble(q.p2().x()); |
| res->AppendDouble(q.p2().y()); |
| res->AppendDouble(q.p3().x()); |
| res->AppendDouble(q.p3().y()); |
| res->AppendDouble(q.p4().x()); |
| res->AppendDouble(q.p4().y()); |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::RectF& rect, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendDouble(rect.x()); |
| res->AppendDouble(rect.y()); |
| res->AppendDouble(rect.width()); |
| res->AppendDouble(rect.height()); |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::Transform& transform, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| for (int row = 0; row < 4; ++row) { |
| for (int col = 0; col < 4; ++col) |
| res->AppendDouble(transform.matrix().rc(row, col)); |
| } |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::BoxF& box, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendInteger(box.x()); |
| res->AppendInteger(box.y()); |
| res->AppendInteger(box.z()); |
| res->AppendInteger(box.width()); |
| res->AppendInteger(box.height()); |
| res->AppendInteger(box.depth()); |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::RRectF& rect, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendDouble(rect.rect().x()); |
| res->AppendDouble(rect.rect().y()); |
| res->AppendDouble(rect.rect().width()); |
| res->AppendDouble(rect.rect().height()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kUpperLeft).x()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kUpperLeft).y()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kUpperRight).x()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kUpperRight).y()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kLowerRight).x()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kLowerRight).y()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kLowerLeft).x()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kLowerLeft).y()); |
| res->EndArray(); |
| } |
| |
| void MathUtil::AddCornerRadiiToTracedValue( |
| const char* name, |
| const gfx::RRectF& rect, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kUpperLeft).x()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kUpperLeft).y()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kUpperRight).x()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kUpperRight).y()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kLowerRight).x()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kLowerRight).y()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kLowerLeft).x()); |
| res->AppendDouble(rect.GetCornerRadii(gfx::RRectF::Corner::kLowerLeft).y()); |
| } |
| |
| void MathUtil::AddToTracedValue(const char* name, |
| const gfx::LinearGradient& gradient, |
| base::trace_event::TracedValue* res) { |
| res->BeginArray(name); |
| res->AppendInteger(gradient.angle()); |
| res->AppendInteger(gradient.step_count()); |
| for (size_t i = 0; i < gradient.step_count(); i++) { |
| res->AppendDouble(gradient.steps()[i].percent); |
| res->AppendInteger(gradient.steps()[i].alpha); |
| } |
| res->EndArray(); |
| } |
| |
| double MathUtil::AsDoubleSafely(double value) { |
| return std::min(value, std::numeric_limits<double>::max()); |
| } |
| |
| float MathUtil::AsFloatSafely(float value) { |
| return std::min(value, std::numeric_limits<float>::max()); |
| } |
| |
| gfx::Vector3dF MathUtil::GetXAxis(const gfx::Transform& transform) { |
| return gfx::Vector3dF(transform.matrix().rc(0, 0), |
| transform.matrix().rc(1, 0), |
| transform.matrix().rc(2, 0)); |
| } |
| |
| gfx::Vector3dF MathUtil::GetYAxis(const gfx::Transform& transform) { |
| return gfx::Vector3dF(transform.matrix().rc(0, 1), |
| transform.matrix().rc(1, 1), |
| transform.matrix().rc(2, 1)); |
| } |
| |
| ScopedSubnormalFloatDisabler::ScopedSubnormalFloatDisabler() { |
| #if defined(ARCH_CPU_X86_FAMILY) |
| // Turn on "subnormals are zero" and "flush to zero" CSR flags. |
| orig_state_ = _mm_getcsr(); |
| _mm_setcsr(orig_state_ | 0x8040); |
| #endif |
| } |
| |
| ScopedSubnormalFloatDisabler::~ScopedSubnormalFloatDisabler() { |
| #if defined(ARCH_CPU_X86_FAMILY) |
| _mm_setcsr(orig_state_); |
| #endif |
| } |
| |
| bool MathUtil::IsFloatNearlyTheSame(float left, float right) { |
| return IsNearlyTheSame(left, right); |
| } |
| |
| bool MathUtil::IsNearlyTheSameForTesting(const gfx::PointF& left, |
| const gfx::PointF& right) { |
| return IsNearlyTheSame(left, right); |
| } |
| |
| bool MathUtil::IsNearlyTheSameForTesting(const gfx::Point3F& left, |
| const gfx::Point3F& right) { |
| return IsNearlyTheSame(left, right); |
| } |
| |
| // Equivalent to SkMatrix::HasPerspective |
| bool MathUtil::SkM44HasPerspective(const SkM44& m) { |
| return (m.rc(3, 0) != 0 || m.rc(3, 1) != 0 || m.rc(3, 2) != 0 || |
| m.rc(3, 3) != 1); |
| } |
| |
| // Since some operations assume a 2d transformation, check to make sure that |
| // is the case by seeing that the z-axis is identity |
| bool MathUtil::SkM44Is2D(const SkM44& m) { |
| return (m.rc(0, 2) == 0 && m.rc(1, 2) == 0 && m.rc(2, 2) == 1 && |
| m.rc(2, 0) == 0 && m.rc(2, 1) == 0 && m.rc(2, 3) == 0 && |
| m.rc(3, 2) == 0); |
| } |
| |
| // Equivalent to SkMatrix::PreservesAxisAlignment |
| // Checks if the transformation is a 90 degree rotation or scaling |
| // See SkMatrix::computeTypeMask |
| bool MathUtil::SkM44Preserves2DAxisAlignment(const SkM44& m) { |
| // Conservatively assume that perspective transforms would not preserve |
| // axis-alignment |
| if (!SkM44Is2D(m) || SkM44HasPerspective(m)) |
| return false; |
| |
| // Does the matrix have skew components |
| if (m.rc(0, 1) != 0 || m.rc(1, 0) != 0) { |
| // Rects only map to rects if both skews are non-zero and both scale |
| // components are zero (i.e. it's a +/-90-degree rotation) |
| return (m.rc(0, 0) == 0 && m.rc(1, 1) == 0 && m.rc(0, 1) != 0 && |
| m.rc(1, 0) != 0); |
| } |
| // Since the matrix has no skewing, it maps to a rectangle so long as the |
| // scale components are non-zero |
| return (m.rc(0, 0) != 0 && m.rc(1, 1) != 0); |
| } |
| |
| } // namespace cc |