pango 3.1.1-x64-mingw32 → 3.1.2-x64-mingw32

data/vendor/local/include/graphene-1.0/graphene-simd4x4f.h

@@ -0,0 +1,1045 @@
+/* graphene-simd4x4f.h: 4x4 float vector operations
+ *
+ * Copyright 2014  Emmanuele Bassi
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#ifndef __GRAPHENE_SIMD4X4F_H__
+#define __GRAPHENE_SIMD4X4F_H__
+
+#include "graphene-simd4f.h"
+
+#include <math.h>
+#include <float.h>
+
+GRAPHENE_BEGIN_DECLS
+
+/**
+ * graphene_simd4x4f_t:
+ *
+ * A SIMD-based matrix type that uses four #graphene_simd4f_t vectors.
+ *
+ * The matrix is treated as row-major, i.e. the x, y, z, and w vectors
+ * are rows, and elements of each vector are a column:
+ *
+ * |[<!-- language="C" -->
+ *   graphene_simd4x4f_t = {
+ *     x.x, x.y, x.z, x.w,
+ *     y.x, y.y, y.z, y.w,
+ *     z.x, z.y, z.z, z.w,
+ *     w.x, w.y, w.z, w.w
+ *   }
+ * ]|
+ *
+ * The contents of the #graphene_simd4x4f_t type are private and
+ * cannot be accessed directly; use the provided API instead.
+ *
+ * Since: 1.0
+ */
+
+/**
+ * graphene_simd4x4f_init:
+ * @x: a #graphene_simd4f_t for the first row
+ * @y: a #graphene_simd4f_t for the second row
+ * @z: a #graphene_simd4f_t for the third row
+ * @w: a #graphene_simd4f_t for the fourth row
+ *
+ * Creates a new #graphene_simd4x4f_t using the given row vectors
+ * to initialize it.
+ *
+ * Returns: the newly created #graphene_simd4x4f_t
+ *
+ * Since: 1.0
+ */
+static inline graphene_simd4x4f_t GRAPHENE_VECTORCALL
+graphene_simd4x4f_init (graphene_simd4f_t x,
+                        graphene_simd4f_t y,
+                        graphene_simd4f_t z,
+                        graphene_simd4f_t w)
+{
+  graphene_simd4x4f_t s;
+
+  s.x = x;
+  s.y = y;
+  s.z = z;
+  s.w = w;
+
+  return s;
+}
+
+/**
+ * graphene_simd4x4f_init_identity:
+ * @m: a #graphene_simd4x4f_t
+ *
+ * Initializes @m to be the identity matrix.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_init_identity (graphene_simd4x4f_t *m)
+{
+  *m = graphene_simd4x4f_init (graphene_simd4f_init (1.0f, 0.0f, 0.0f, 0.0f),
+                               graphene_simd4f_init (0.0f, 1.0f, 0.0f, 0.0f),
+                               graphene_simd4f_init (0.0f, 0.0f, 1.0f, 0.0f),
+                               graphene_simd4f_init (0.0f, 0.0f, 0.0f, 1.0f));
+}
+
+/**
+ * graphene_simd4x4f_init_from_float:
+ * @m: a #graphene_simd4x4f_t
+ * @f: (array fixed-size=16): an array of 16 floating point values
+ *
+ * Initializes a #graphene_simd4x4f_t with the given array
+ * of floating point values.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_init_from_float (graphene_simd4x4f_t *m,
+                                   const float         *f)
+{
+  m->x = graphene_simd4f_init_4f (f +  0);
+  m->y = graphene_simd4f_init_4f (f +  4);
+  m->z = graphene_simd4f_init_4f (f +  8);
+  m->w = graphene_simd4f_init_4f (f + 12);
+}
+
+/**
+ * graphene_simd4x4f_to_float:
+ * @m: a #graphene_sidm4x4f_t
+ * @v: (out caller-allocates) (array fixed-size=16): a floating
+ *   point values vector capable of holding at least 16 values
+ *
+ * Copies the content of @m in a float array.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_to_float (const graphene_simd4x4f_t *m,
+                            float                     *v)
+{
+  graphene_simd4f_dup_4f (m->x, v +  0);
+  graphene_simd4f_dup_4f (m->y, v +  4);
+  graphene_simd4f_dup_4f (m->z, v +  8);
+  graphene_simd4f_dup_4f (m->w, v + 12);
+}
+
+GRAPHENE_AVAILABLE_IN_1_0
+void    graphene_simd4x4f_transpose_in_place    (graphene_simd4x4f_t *s);
+
+#if defined(GRAPHENE_USE_SSE)
+
+#ifdef __GNUC__
+#define graphene_simd4x4f_transpose_in_place(s) \
+  (__extension__ ({ \
+    _MM_TRANSPOSE4_PS ((s)->x, (s)->y, (s)->z, (s)->w); \
+  }))
+#elif defined (_MSC_VER)
+#define graphene_simd4x4f_transpose_in_place(s) \
+  _MM_TRANSPOSE4_PS ((s)->x, (s)->y, (s)->z, (s)->w)
+#endif
+
+#elif defined(GRAPHENE_USE_GCC)
+
+#define graphene_simd4x4f_transpose_in_place(s) \
+  (__extension__ ({ \
+    const graphene_simd4f_t sx = (s)->x; \
+    const graphene_simd4f_t sy = (s)->y; \
+    const graphene_simd4f_t sz = (s)->z; \
+    const graphene_simd4f_t sw = (s)->w; \
+    (s)->x = graphene_simd4f_init (sx[0], sy[0], sz[0], sw[0]); \
+    (s)->y = graphene_simd4f_init (sx[1], sy[1], sz[1], sw[1]); \
+    (s)->z = graphene_simd4f_init (sx[2], sy[2], sz[2], sw[2]); \
+    (s)->w = graphene_simd4f_init (sx[3], sy[3], sz[3], sw[3]); \
+  }))
+
+#elif defined(GRAPHENE_USE_ARM_NEON)
+
+#define graphene_simd4x4f_transpose_in_place(s) \
+  (__extension__ ({ \
+    const graphene_simd4f_union_t sx = { (s)->x }; \
+    const graphene_simd4f_union_t sy = { (s)->y }; \
+    const graphene_simd4f_union_t sz = { (s)->z }; \
+    const graphene_simd4f_union_t sw = { (s)->w }; \
+    (s)->x = graphene_simd4f_init (sx.f[0], sy.f[0], sz.f[0], sw.f[0]); \
+    (s)->y = graphene_simd4f_init (sx.f[1], sy.f[1], sz.f[1], sw.f[1]); \
+    (s)->z = graphene_simd4f_init (sx.f[2], sy.f[2], sz.f[2], sw.f[2]); \
+    (s)->w = graphene_simd4f_init (sx.f[3], sy.f[3], sz.f[3], sw.f[3]); \
+  }))
+
+#elif defined(GRAPHENE_USE_SCALAR)
+
+#define graphene_simd4x4f_transpose_in_place(s) \
+  (graphene_simd4x4f_transpose_in_place ((graphene_simd4x4f_t *) (s)))
+
+#else
+# error "No implementation for graphene_simd4x4f_t defined."
+#endif
+
+/**
+ * graphene_simd4x4f_sum:
+ * @a: a #graphene_simd4f_t
+ * @res: (out): return location for the sum vector
+ *
+ * Adds all the row vectors of @a.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_sum (const graphene_simd4x4f_t *a,
+                       graphene_simd4f_t         *res)
+{
+  graphene_simd4f_t s = graphene_simd4f_add (a->x, a->y);
+  s = graphene_simd4f_add (s, a->z);
+  s = graphene_simd4f_add (s, a->w);
+  *res = s;
+}
+
+/**
+ * graphene_simd4x4f_vec4_mul:
+ * @a: a #graphene_simd4x4f_t
+ * @b: a #graphene_simd4f_t
+ * @res: (out): return location for a #graphene_simd4f_t
+ *
+ * Multiplies the given #graphene_simd4x4f_t with the given
+ * #graphene_simd4f_t using a dot product.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_vec4_mul (const graphene_simd4x4f_t *a,
+                            const graphene_simd4f_t   *b,
+                            graphene_simd4f_t         *res)
+{
+  const graphene_simd4f_t v = *b;
+  const graphene_simd4f_t v_x = graphene_simd4f_splat_x (v);
+  const graphene_simd4f_t v_y = graphene_simd4f_splat_y (v);
+  const graphene_simd4f_t v_z = graphene_simd4f_splat_z (v);
+  const graphene_simd4f_t v_w = graphene_simd4f_splat_w (v);
+
+  *res = graphene_simd4f_add (graphene_simd4f_add (graphene_simd4f_mul (a->x, v_x),
+                                                   graphene_simd4f_mul (a->y, v_y)),
+                              graphene_simd4f_add (graphene_simd4f_mul (a->z, v_z),
+                                                   graphene_simd4f_mul (a->w, v_w)));
+}
+
+/**
+ * graphene_simd4x4f_vec3_mul:
+ * @m: a #graphene_simd4x4f_t
+ * @v: a #graphene_simd4f_t
+ * @res: (out): return location for a #graphene_simd4f_t
+ *
+ * Multiplies the given #graphene_simd4x4f_t with the given
+ * #graphene_simd4f_t, using only the first three row vectors
+ * of the matrix, and the first three components of the vector.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_vec3_mul (const graphene_simd4x4f_t *m,
+                            const graphene_simd4f_t   *v,
+                            graphene_simd4f_t         *res)
+{
+  const graphene_simd4f_t v_x = graphene_simd4f_splat_x (*v);
+  const graphene_simd4f_t v_y = graphene_simd4f_splat_y (*v);
+  const graphene_simd4f_t v_z = graphene_simd4f_splat_z (*v);
+
+  *res = graphene_simd4f_add (graphene_simd4f_add (graphene_simd4f_mul (m->x, v_x),
+                                                   graphene_simd4f_mul (m->y, v_y)),
+                              graphene_simd4f_mul (m->z, v_z));
+}
+
+/**
+ * graphene_simd4x4f_point3_mul:
+ * @m: a #graphene_simd4x4f_t
+ * @p: a #graphene_simd4f_t
+ * @res: (out): return location for a #graphene_simd4f_t
+ *
+ * Multiplies the given #graphene_simd4x4f_t with the given
+ * #graphene_simd4f_t.
+ *
+ * Unlike graphene_simd4x4f_vec3_mul(), this function will
+ * also use the fourth row vector of the matrix.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_point3_mul (const graphene_simd4x4f_t *m,
+                              const graphene_simd4f_t   *p,
+                              graphene_simd4f_t         *res)
+{
+  const graphene_simd4f_t v = *p;
+  const graphene_simd4f_t v_x = graphene_simd4f_splat_x (v);
+  const graphene_simd4f_t v_y = graphene_simd4f_splat_y (v);
+  const graphene_simd4f_t v_z = graphene_simd4f_splat_z (v);
+
+  *res = graphene_simd4f_add (graphene_simd4f_add (graphene_simd4f_mul (m->x, v_x),
+                                                   graphene_simd4f_mul (m->y, v_y)),
+                              graphene_simd4f_add (graphene_simd4f_mul (m->z, v_z),
+                                                   m->w));
+}
+
+/**
+ * graphene_simd4x4f_transpose:
+ * @s: a #graphene_simd4x4f_t
+ * @res: (out): return location for the transposed matrix
+ *
+ * Transposes the given #graphene_simd4x4f_t.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_transpose (const graphene_simd4x4f_t *s,
+                             graphene_simd4x4f_t       *res)
+{
+  *res = *s;
+  graphene_simd4x4f_transpose_in_place (res);
+}
+
+/**
+ * graphene_simd4x4f_inv_ortho_vec3_mul:
+ * @a: a #graphene_simd4x4f_t
+ * @b: a #graphene_simd4f_t
+ * @res: (out): return location for the transformed vector
+ *
+ * Performs the inverse orthographic transformation of the first
+ * three components in the given vector, using the first three
+ * row vectors of the given SIMD matrix.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_inv_ortho_vec3_mul (const graphene_simd4x4f_t *a,
+                                      const graphene_simd4f_t   *b,
+                                      graphene_simd4f_t         *res)
+{
+  graphene_simd4x4f_t transpose = *a;
+  graphene_simd4f_t translation = *b;
+
+  transpose.w = graphene_simd4f_init (0.f, 0.f, 0.f, 0.f);
+  graphene_simd4x4f_transpose_in_place (&transpose);
+
+  graphene_simd4x4f_vec3_mul (&transpose, &translation, res);
+}
+
+/**
+ * graphene_simd4x4f_inv_ortho_point3_mul:
+ * @a: a #graphene_simd4x4f_t
+ * @b: a #graphene_simd4x4f_t
+ * @res: (out): return location for the result vector
+ *
+ * Performs the inverse orthographic transformation of the first
+ * three components in the given vector, using the given SIMD
+ * matrix.
+ *
+ * Unlike graphene_simd4x4f_inv_ortho_vec3_mul(), this function
+ * will also use the fourth row vector of the SIMD matrix.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_inv_ortho_point3_mul (const graphene_simd4x4f_t *a,
+                                        const graphene_simd4f_t   *b,
+                                        graphene_simd4f_t         *res)
+{
+  graphene_simd4f_t translation = graphene_simd4f_sub (*b, a->w);
+  graphene_simd4x4f_t transpose = *a;
+
+  transpose.w = graphene_simd4f_init (0.f, 0.f, 0.f, 0.f);
+  graphene_simd4x4f_transpose_in_place (&transpose);
+
+  graphene_simd4x4f_point3_mul (&transpose, &translation, res);
+}
+
+/**
+ * graphene_simd4x4f_matrix_mul:
+ * @a: a #graphene_simd4x4f_t
+ * @b: a #graphene_simd4x4f_t
+ * @res: (out): return location for the result
+ *
+ * Multiplies the two matrices.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_matrix_mul (const graphene_simd4x4f_t *a,
+                              const graphene_simd4x4f_t *b,
+                              graphene_simd4x4f_t       *res)
+{
+#if 0
+  /* this is the classic naive A*B implementation of the row * column
+   * matrix product. using a SIMD scalar implementation, it's fairly
+   * slow at 329ns per multiplication; the SSE implementation makes it
+   * about 10x faster, at 32ns; the GCC vector implementation is only
+   * 5x faster, at 66ns. the biggest culprits are the transpose operation
+   * and the multiple, one lane reads to compute the scalar sum.
+   */
+  graphene_simd4x4f_t t;
+
+  graphene_simd4x4f_transpose (b, &t);
+
+  res->x =
+    graphene_simd4f_init (graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->x, t.x)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->x, t.y)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->x, t.z)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->x, t.w)));
+
+  res->y =
+    graphene_simd4f_init (graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->y, t.x)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->y, t.y)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->y, t.z)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->y, t.w)));
+
+  res->z =
+    graphene_simd4f_init (graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->z, t.x)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->z, t.y)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->z, t.z)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->z, t.w)));
+
+  res->w =
+    graphene_simd4f_init (graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->w, t.x)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->w, t.y)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->w, t.z)),
+                          graphene_simd4f_sum_scalar (graphene_simd4f_mul (a->w, t.w)));
+#else
+  /* this is an optimized version of the matrix multiplication, using
+   * four dot products for each row vector. this yields drastically
+   * better numbers while retaining the same correct results as above:
+   * the scalar implementation now clocks at 91ns; the GCC vector
+   * implementation is 19ns; and the SSE implementation is 16ns.
+   *
+   * the order is correct if we want to multiply A with B; remember
+   * that matrix multiplication is non-commutative.
+   */
+  graphene_simd4x4f_vec4_mul (b, &a->x, &res->x);
+  graphene_simd4x4f_vec4_mul (b, &a->y, &res->y);
+  graphene_simd4x4f_vec4_mul (b, &a->z, &res->z);
+  graphene_simd4x4f_vec4_mul (b, &a->w, &res->w);
+#endif
+}
+
+/**
+ * graphene_simd4x4f_init_perspective:
+ * @m: a #graphene_simd4x4f_t
+ * @fovy_rad: the angle of the field of vision, in radians
+ * @aspect: the aspect value
+ * @z_near: the depth of the near clipping plane
+ * @z_far: the depth of the far clipping plane
+ *
+ * Initializes a #graphene_simd4x4f_t with a perspective projection.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_init_perspective (graphene_simd4x4f_t *m,
+                                    float                fovy_rad,
+                                    float                aspect,
+                                    float                z_near,
+                                    float                z_far)
+{
+  float delta_z = z_far - z_near;
+  float cotangent = tanf (GRAPHENE_PI_2 - fovy_rad * 0.5f);
+
+  float a = cotangent / aspect;
+  float b = cotangent;
+  float c = -(z_far + z_near) / delta_z;
+  float d = -2 * z_near * z_far / delta_z;
+
+  m->x = graphene_simd4f_init (   a, 0.0f, 0.0f,  0.0f);
+  m->y = graphene_simd4f_init (0.0f,    b, 0.0f,  0.0f);
+  m->z = graphene_simd4f_init (0.0f, 0.0f,    c, -1.0f);
+  m->w = graphene_simd4f_init (0.0f, 0.0f,    d,  0.0f);
+}
+
+/**
+ * graphene_simd4x4f_init_ortho:
+ * @m: a #graphene_simd4x4f_t
+ * @left: edge of the left clipping plane
+ * @right: edge of the right clipping plane
+ * @bottom: edge of the bottom clipping plane
+ * @top: edge of the top clipping plane
+ * @z_near: depth of the near clipping plane
+ * @z_far: depth of the far clipping plane
+ *
+ * Initializes the given SIMD matrix with an orthographic projection.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_init_ortho (graphene_simd4x4f_t *m,
+                              float                left,
+                              float                right,
+                              float                bottom,
+                              float                top,
+                              float                z_near,
+                              float                z_far)
+{
+  float delta_x = right - left;
+  float delta_y = top - bottom;
+  float delta_z = z_far - z_near;
+
+  float a = 2.0f / delta_x;
+  float b = -(right + left) / delta_x;
+  float c = 2.0f / delta_y;
+  float d = -(top + bottom) / delta_y;
+  float e = -2.0f / delta_z;
+  float f = -(z_far + z_near) / delta_z;
+
+  m->x = graphene_simd4f_init (   a, 0.0f, 0.0f, 0.0f);
+  m->y = graphene_simd4f_init (0.0f,    c, 0.0f, 0.0f);
+  m->z = graphene_simd4f_init (0.0f, 0.0f,    e, 0.0f);
+  m->w = graphene_simd4f_init (   b,    d,    f, 1.0f);
+}
+
+/**
+ * graphene_simd4x4f_init_look_at:
+ * @m: a #graphene_simd4x4f_t
+ * @eye: vector for the camera coordinates
+ * @center: vector for the object coordinates
+ * @up: vector for the upwards direction
+ *
+ * Initializes a SIMD matrix with the projection necessary for
+ * the camera at the @eye coordinates to look at the object at
+ * the @center coordinates. The top of the camera is aligned to
+ * the @up vector.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_init_look_at (graphene_simd4x4f_t *m,
+                                graphene_simd4f_t    eye,
+                                graphene_simd4f_t    center,
+                                graphene_simd4f_t    up)
+{
+  const graphene_simd4f_t z_axis = graphene_simd4f_normalize3 (graphene_simd4f_sub (center, eye));
+  const graphene_simd4f_t x_axis = graphene_simd4f_normalize3 (graphene_simd4f_cross3 (z_axis, up));
+  const graphene_simd4f_t y_axis = graphene_simd4f_cross3 (x_axis, z_axis);
+  float eye_v[4];
+
+  graphene_simd4f_dup_4f (eye, eye_v);
+
+  m->x = x_axis;
+  m->y = y_axis;
+  m->z = graphene_simd4f_neg (z_axis);
+  m->w = graphene_simd4f_init (-eye_v[0], -eye_v[1], -eye_v[2], 1.f);
+}
+
+/**
+ * graphene_simd4x4f_init_frustum:
+ * @m: a #graphene_simd4x4f_t
+ * @left: distance of the left clipping plane
+ * @right: distance of the right clipping plane
+ * @bottom: distance of the bottom clipping plane
+ * @top: distance of the top clipping plane
+ * @z_near: distance of the near clipping plane
+ * @z_far: distance of the far clipping plane
+ *
+ * Initializes a SIMD matrix with a frustum described by the distances
+ * of six clipping planes.
+ *
+ * Since: 1.2
+ */
+static inline void
+graphene_simd4x4f_init_frustum (graphene_simd4x4f_t *m,
+                                float                left,
+                                float                right,
+                                float                bottom,
+                                float                top,
+                                float                z_near,
+                                float                z_far)
+{
+  float x = 2.f * z_near / (right - left);
+  float y = 2.f * z_near / (top - bottom);
+
+  float a = (right + left) / (right - left);
+  float b = (top + bottom) / (top - bottom);
+  float c = -1.f * (z_far + z_near) / (z_far - z_near);
+  float d = -2.f * z_far * z_near / (z_far - z_near);
+
+  m->x = graphene_simd4f_init (  x, 0.f, 0.f,  0.f);
+  m->y = graphene_simd4f_init (0.f,   y, 0.f,  0.f);
+  m->z = graphene_simd4f_init (  a,   b,   c, -1.f);
+  m->w = graphene_simd4f_init (0.f, 0.f,   d,  0.f);
+}
+
+/**
+ * graphene_simd4x4f_perspective:
+ * @m: a #graphene_simd4x4f_t
+ * @depth: depth of the perspective
+ *
+ * Adds a perspective transformation for the given @depth.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_perspective (graphene_simd4x4f_t *m,
+                               float                depth)
+{
+#if 1
+  const float m_xw = graphene_simd4f_get_w (m->x);
+  const float m_yw = graphene_simd4f_get_w (m->y);
+  const float m_zw = graphene_simd4f_get_w (m->z);
+  const float m_ww = graphene_simd4f_get_w (m->w);
+
+  const float p0 = graphene_simd4f_get_z (m->x) + -1.0f / depth * m_xw;
+  const float p1 = graphene_simd4f_get_z (m->y) + -1.0f / depth * m_yw;
+  const float p2 = graphene_simd4f_get_z (m->z) + -1.0f / depth * m_zw;
+  const float p3 = graphene_simd4f_get_z (m->w) + -1.0f / depth * m_ww;
+
+  const graphene_simd4f_t p_x = graphene_simd4f_merge_w (m->x, m_xw + p0);
+  const graphene_simd4f_t p_y = graphene_simd4f_merge_w (m->y, m_yw + p1);
+  const graphene_simd4f_t p_z = graphene_simd4f_merge_w (m->z, m_zw + p2);
+  const graphene_simd4f_t p_w = graphene_simd4f_merge_w (m->w, m_ww + p3);
+#else
+  /* this is equivalent to the operations above, but trying to inline
+   * them into SIMD registers as much as possible by transposing the
+   * original matrix and operating on the resulting column vectors. it
+   * should warrant a micro benchmark, because while the above code is
+   * dominated by single channel reads, the code below has a transpose
+   * operation.
+   */
+  graphene_simd4x4f_t t;
+  const graphene_simd4f_t f, p;
+  const graphene_simd4f_t p_x, p_y, p_z, p_w;
+
+  graphene_simd4x4f_transpose (m, &t);
+
+  f = graphene_simd4f_neg (graphene_simd4f_reciprocal (graphene_simd4f_splat (depth)));
+  p = graphene_simd4f_sum (t.w, graphene_simd4f_sum (t.z, graphene_simd4f_mul (f, t.w)));
+  p_x = graphene_simd4f_merge_w (m->x, graphene_simd4f_get_x (p));
+  p_y = graphene_simd4f_merge_w (m->y, graphene_simd4f_get_y (p));
+  p_z = graphene_simd4f_merge_w (m->z, graphene_simd4f_get_z (p));
+  p_w = graphene_simd4f_merge_w (m->w, graphene_simd4f_get_w (p));
+#endif
+
+  *m = graphene_simd4x4f_init (p_x, p_y, p_z, p_w);
+}
+
+/**
+ * graphene_simd4x4f_translation:
+ * @m: a #graphene_simd4x4f_t
+ * @x: coordinate of the X translation
+ * @y: coordinate of the Y translation
+ * @z: coordinate of the Z translation
+ *
+ * Initializes @m to contain a translation to the given coordinates.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_translation (graphene_simd4x4f_t *m,
+                               float                x,
+                               float                y,
+                               float                z)
+{
+  *m = graphene_simd4x4f_init (graphene_simd4f_init (1.0f, 0.0f, 0.0f, 0.0f),
+                               graphene_simd4f_init (0.0f, 1.0f, 0.0f, 0.0f),
+                               graphene_simd4f_init (0.0f, 0.0f, 1.0f, 0.0f),
+                               graphene_simd4f_init (   x,    y,    z, 1.0f));
+}
+
+/**
+ * graphene_simd4x4f_scale:
+ * @m: a #graphene_simd4x4f_t
+ * @x: scaling factor on the X axis
+ * @y: scaling factor on the Y axis
+ * @z: scaling factor on the Z axis
+ *
+ * Initializes @m to contain a scaling transformation with the
+ * given factors.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_scale (graphene_simd4x4f_t *m,
+                         float                x,
+                         float                y,
+                         float                z)
+{
+  *m = graphene_simd4x4f_init (graphene_simd4f_init (   x, 0.0f, 0.0f, 0.0f),
+                               graphene_simd4f_init (0.0f,    y, 0.0f, 0.0f),
+                               graphene_simd4f_init (0.0f, 0.0f,    z, 0.0f),
+                               graphene_simd4f_init (0.0f, 0.0f, 0.0f, 1.0f));
+
+}
+
+/**
+ * graphene_simd4x4f_rotation:
+ * @m: a #graphene_simd4x4f_t
+ * @rad: the rotation, in radians
+ * @axis: the vector of the axis of rotation
+ *
+ * Initializes @m to contain a rotation of the given angle
+ * along the given axis.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_rotation (graphene_simd4x4f_t *m,
+                            float                rad,
+                            graphene_simd4f_t    axis)
+{
+  float sine, cosine;
+  float x, y, z;
+  float ab, bc, ca;
+  float tx, ty, tz;
+  graphene_simd4f_t i, j, k;
+
+  rad = -rad;
+  axis = graphene_simd4f_normalize3 (axis);
+
+  /* We cannot use graphene_sincos() because it's a private function, whereas
+   * graphene-simd4x4f.h is a public header
+   */
+  sine = sinf (rad);
+  cosine = cosf (rad);
+
+  x = graphene_simd4f_get_x (axis);
+  y = graphene_simd4f_get_y (axis);
+  z = graphene_simd4f_get_z (axis);
+
+  ab = x * y * (1.0f - cosine);
+  bc = y * z * (1.0f - cosine);
+  ca = z * x * (1.0f - cosine);
+
+  tx = x * x;
+  ty = y * y;
+  tz = z * z;
+
+  i = graphene_simd4f_init (tx + cosine * (1.0f - tx), ab - z * sine, ca + y * sine, 0.f);
+  j = graphene_simd4f_init (ab + z * sine, ty + cosine * (1.0f - ty), bc - x * sine, 0.f);
+  k = graphene_simd4f_init (ca - y * sine, bc + x * sine, tz + cosine * (1.0f - tz), 0.f);
+
+  *m = graphene_simd4x4f_init (i, j, k, graphene_simd4f_init (0.0f, 0.0f, 0.0f, 1.0f));
+}
+
+/**
+ * graphene_simd4x4f_add:
+ * @a: a #graphene_simd4x4f_t
+ * @b: a #graphene_simd4x4f_t
+ * @res: (out caller-allocates): return location for a #graphene_simd4x4f_t
+ *
+ * Adds each row vector of @a and @b and places the results in @res.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_add (const graphene_simd4x4f_t *a,
+                       const graphene_simd4x4f_t *b,
+                       graphene_simd4x4f_t *res)
+{
+  res->x = graphene_simd4f_add (a->x, b->x);
+  res->y = graphene_simd4f_add (a->y, b->y);
+  res->z = graphene_simd4f_add (a->z, b->z);
+  res->w = graphene_simd4f_add (a->w, b->w);
+}
+
+/**
+ * graphene_simd4x4f_sub:
+ * @a: a #graphene_simd4x4f_t
+ * @b: a #graphene_simd4x4f_t
+ * @res: (out caller-allocates): return location for a #graphene_simd4x4f_t
+ *
+ * Subtracts each row vector of @a and @b and places the results in @res.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_sub (const graphene_simd4x4f_t *a,
+                       const graphene_simd4x4f_t *b,
+                       graphene_simd4x4f_t *res)
+{
+  res->x = graphene_simd4f_sub (a->x, b->x);
+  res->y = graphene_simd4f_sub (a->y, b->y);
+  res->z = graphene_simd4f_sub (a->z, b->z);
+  res->w = graphene_simd4f_sub (a->w, b->w);
+}
+
+/**
+ * graphene_simd4x4f_mul:
+ * @a: a #graphene_simd4x4f_t
+ * @b: a #graphene_simd4x4f_t
+ * @res: (out caller-allocates): return location for a #graphene_simd4x4f_t
+ *
+ * Multiplies each row vector of @a and @b and places the results in @res.
+ *
+ * You most likely want graphene_simd4x4f_matrix_mul() instead.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_mul (const graphene_simd4x4f_t *a,
+                       const graphene_simd4x4f_t *b,
+                       graphene_simd4x4f_t *res)
+{
+  res->x = graphene_simd4f_mul (a->x, b->x);
+  res->y = graphene_simd4f_mul (a->y, b->y);
+  res->z = graphene_simd4f_mul (a->z, b->z);
+  res->w = graphene_simd4f_mul (a->w, b->w);
+}
+
+/**
+ * graphene_simd4x4f_div:
+ * @a: a #graphene_simd4x4f_t
+ * @b: a #graphene_simd4x4f_t
+ * @res: (out caller-allocates): return location for a #graphene_simd4x4f_t
+ *
+ * Divides each row vector of @a and @b and places the results in @res.
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_div (const graphene_simd4x4f_t *a,
+                       const graphene_simd4x4f_t *b,
+                       graphene_simd4x4f_t *res)
+{
+  res->x = graphene_simd4f_div (a->x, b->x);
+  res->y = graphene_simd4f_div (a->y, b->y);
+  res->z = graphene_simd4f_div (a->z, b->z);
+  res->w = graphene_simd4f_div (a->w, b->w);
+}
+
+/**
+ * graphene_simd4x4f_inverse:
+ * @m: a #graphene_simd4x4f_t
+ * @res: (out): return location for the inverse matrix
+ *
+ * Inverts the given #graphene_simd4x4f_t.
+ *
+ * Returns: `true` if the matrix was invertible
+ *
+ * Since: 1.0
+ */
+static inline bool
+graphene_simd4x4f_inverse (const graphene_simd4x4f_t *m,
+                           graphene_simd4x4f_t       *res)
+{
+  /* split rows */
+  const graphene_simd4f_t r0 = m->x;
+  const graphene_simd4f_t r1 = m->y;
+  const graphene_simd4f_t r2 = m->z;
+  const graphene_simd4f_t r3 = m->w;
+
+  /* cofactors */
+  const graphene_simd4f_t r0_wxyz = graphene_simd4f_shuffle_wxyz (r0);
+  const graphene_simd4f_t r0_zwxy = graphene_simd4f_shuffle_zwxy (r0);
+  const graphene_simd4f_t r0_yzwx = graphene_simd4f_shuffle_yzwx (r0);
+
+  const graphene_simd4f_t r1_wxyz = graphene_simd4f_shuffle_wxyz (r1);
+  const graphene_simd4f_t r1_zwxy = graphene_simd4f_shuffle_zwxy (r1);
+  const graphene_simd4f_t r1_yzwx = graphene_simd4f_shuffle_yzwx (r1);
+
+  const graphene_simd4f_t r2_wxyz = graphene_simd4f_shuffle_wxyz (r2);
+  const graphene_simd4f_t r2_zwxy = graphene_simd4f_shuffle_zwxy (r2);
+  const graphene_simd4f_t r2_yzwx = graphene_simd4f_shuffle_yzwx (r2);
+
+  const graphene_simd4f_t r3_wxyz = graphene_simd4f_shuffle_wxyz (r3);
+  const graphene_simd4f_t r3_zwxy = graphene_simd4f_shuffle_zwxy (r3);
+  const graphene_simd4f_t r3_yzwx = graphene_simd4f_shuffle_yzwx (r3);
+
+  const graphene_simd4f_t r0_wxyz_x_r1 = graphene_simd4f_mul (r0_wxyz, r1);
+  const graphene_simd4f_t r0_wxyz_x_r1_yzwx = graphene_simd4f_mul (r0_wxyz, r1_yzwx);
+  const graphene_simd4f_t r0_wxyz_x_r1_zwxy = graphene_simd4f_mul (r0_wxyz, r1_zwxy);
+
+  const graphene_simd4f_t r2_wxyz_x_r3 = graphene_simd4f_mul (r2_wxyz, r3);
+  const graphene_simd4f_t r2_wxyz_x_r3_yzwx = graphene_simd4f_mul (r2_wxyz, r3_yzwx);
+  const graphene_simd4f_t r2_wxyz_x_r3_zwxy = graphene_simd4f_mul (r2_wxyz, r3_zwxy);
+
+  const graphene_simd4f_t ar1 = graphene_simd4f_sub (graphene_simd4f_shuffle_wxyz (r2_wxyz_x_r3_zwxy),
+                                                     graphene_simd4f_shuffle_zwxy (r2_wxyz_x_r3));
+  const graphene_simd4f_t ar2 = graphene_simd4f_sub (graphene_simd4f_shuffle_zwxy (r2_wxyz_x_r3_yzwx),
+                                                     r2_wxyz_x_r3_yzwx);
+  const graphene_simd4f_t ar3 = graphene_simd4f_sub (r2_wxyz_x_r3_zwxy,
+                                                     graphene_simd4f_shuffle_wxyz (r2_wxyz_x_r3));
+
+  const graphene_simd4f_t br1 = graphene_simd4f_sub (graphene_simd4f_shuffle_wxyz (r0_wxyz_x_r1_zwxy),
+                                                     graphene_simd4f_shuffle_zwxy (r0_wxyz_x_r1));
+  const graphene_simd4f_t br2 = graphene_simd4f_sub (graphene_simd4f_shuffle_zwxy (r0_wxyz_x_r1_yzwx),
+                                                     r0_wxyz_x_r1_yzwx);
+  const graphene_simd4f_t br3 = graphene_simd4f_sub (r0_wxyz_x_r1_zwxy,
+                                                     graphene_simd4f_shuffle_wxyz (r0_wxyz_x_r1));
+
+  const graphene_simd4f_t r0_sum =
+    graphene_simd4f_madd (r0_yzwx, ar3,
+                          graphene_simd4f_madd (r0_zwxy, ar2,
+                                                graphene_simd4f_mul (r0_wxyz, ar1)));
+  const graphene_simd4f_t r1_sum =
+    graphene_simd4f_madd (r1_wxyz, ar1,
+                          graphene_simd4f_madd (r1_zwxy, ar2,
+                                                graphene_simd4f_mul (r1_yzwx, ar3)));
+  const graphene_simd4f_t r2_sum =
+    graphene_simd4f_madd (r2_yzwx, br3,
+                          graphene_simd4f_madd (r2_zwxy, br2,
+                                                graphene_simd4f_mul (r2_wxyz, br1)));
+  const graphene_simd4f_t r3_sum =
+    graphene_simd4f_madd (r3_yzwx, br3,
+                          graphene_simd4f_madd (r3_zwxy, br2,
+                                                graphene_simd4f_mul (r3_wxyz, br1)));
+
+  /* determinant and its inverse */
+  const graphene_simd4f_t d0 = graphene_simd4f_mul (r1_sum, r0);
+  const graphene_simd4f_t d1 = graphene_simd4f_add (d0, graphene_simd4f_merge_high (d0, d0));
+  const graphene_simd4f_t det = graphene_simd4f_sub (d1, graphene_simd4f_splat_y (d1));
+  if (graphene_simd4f_get_x (det) != 0.f)
+    {
+      const graphene_simd4f_t invdet = graphene_simd4f_splat_x (graphene_simd4f_div (graphene_simd4f_splat (1.0f), det));
+
+      const graphene_simd4f_t o0 = graphene_simd4f_mul (graphene_simd4f_flip_sign_0101 (r1_sum), invdet);
+      const graphene_simd4f_t o1 = graphene_simd4f_mul (graphene_simd4f_flip_sign_1010 (r0_sum), invdet);
+      const graphene_simd4f_t o2 = graphene_simd4f_mul (graphene_simd4f_flip_sign_0101 (r3_sum), invdet);
+      const graphene_simd4f_t o3 = graphene_simd4f_mul (graphene_simd4f_flip_sign_1010 (r2_sum), invdet);
+
+      graphene_simd4x4f_t mt = graphene_simd4x4f_init (o0, o1, o2, o3);
+
+      /* transpose the resulting matrix */
+      graphene_simd4x4f_transpose (&mt, res);
+
+      return true;
+    }
+
+  return false;
+}
+
+/**
+ * graphene_simd4x4f_determinant:
+ * @m: a #graphene_simd4x4f_t
+ * @det_r: (out): return location for the matrix determinant
+ * @invdet_r: (out): return location for the inverse of the matrix
+ *   determinant
+ *
+ * Computes the determinant (and its inverse) of the given matrix
+ *
+ * Since: 1.0
+ */
+static inline void
+graphene_simd4x4f_determinant (const graphene_simd4x4f_t *m,
+                               graphene_simd4f_t         *det_r,
+                               graphene_simd4f_t         *invdet_r)
+{
+  /* split rows */
+  const graphene_simd4f_t r0 = m->x;
+  const graphene_simd4f_t r1 = m->y;
+  const graphene_simd4f_t r2 = m->z;
+  const graphene_simd4f_t r3 = m->w;
+
+  /* cofactors */
+  const graphene_simd4f_t r1_wxyz = graphene_simd4f_shuffle_wxyz (r1);
+  const graphene_simd4f_t r1_zwxy = graphene_simd4f_shuffle_zwxy (r1);
+  const graphene_simd4f_t r1_yzwx = graphene_simd4f_shuffle_yzwx (r1);
+
+  const graphene_simd4f_t r2_wxyz = graphene_simd4f_shuffle_wxyz (r2);
+
+  const graphene_simd4f_t r3_zwxy = graphene_simd4f_shuffle_zwxy (r3);
+  const graphene_simd4f_t r3_yzwx = graphene_simd4f_shuffle_yzwx (r3);
+
+  const graphene_simd4f_t r2_wxyz_x_r3 = graphene_simd4f_mul (r2_wxyz, r3);
+  const graphene_simd4f_t r2_wxyz_x_r3_yzwx = graphene_simd4f_mul (r2_wxyz, r3_yzwx);
+  const graphene_simd4f_t r2_wxyz_x_r3_zwxy = graphene_simd4f_mul (r2_wxyz, r3_zwxy);
+
+  const graphene_simd4f_t ar1 = graphene_simd4f_sub (graphene_simd4f_shuffle_wxyz (r2_wxyz_x_r3_zwxy),
+                                                     graphene_simd4f_shuffle_zwxy (r2_wxyz_x_r3));
+  const graphene_simd4f_t ar2 = graphene_simd4f_sub (graphene_simd4f_shuffle_zwxy (r2_wxyz_x_r3_yzwx),
+                                                     r2_wxyz_x_r3_yzwx);
+  const graphene_simd4f_t ar3 = graphene_simd4f_sub (r2_wxyz_x_r3_zwxy,
+                                                     graphene_simd4f_shuffle_wxyz (r2_wxyz_x_r3));
+
+  const graphene_simd4f_t r1_sum =
+    graphene_simd4f_madd (r1_wxyz, ar1,
+                          graphene_simd4f_madd (r1_zwxy, ar2,
+                                                graphene_simd4f_mul (r1_yzwx, ar3)));
+
+  /* determinant and its inverse */
+  const graphene_simd4f_t d0 = graphene_simd4f_mul (r1_sum, r0);
+  const graphene_simd4f_t d1 = graphene_simd4f_add (d0, graphene_simd4f_merge_high (d0, d0));
+
+  const graphene_simd4f_t det = graphene_simd4f_sub (d1, graphene_simd4f_splat_y (d1));
+
+  const graphene_simd4f_t invdet = graphene_simd4f_splat_x (graphene_simd4f_div (graphene_simd4f_splat (1.0f), det));
+
+  if (det_r != NULL)
+    *det_r = det;
+
+  if (invdet_r != NULL)
+    *invdet_r = invdet;
+}
+
+/**
+ * graphene_simd4x4f_is_identity:
+ * @m: a #graphene_simd4x4f_t
+ *
+ * Checks whether the given matrix is the identity matrix.
+ *
+ * Returns: `true` if the matrix is the identity matrix
+ *
+ * Since: 1.0
+ */
+static inline bool
+graphene_simd4x4f_is_identity (const graphene_simd4x4f_t *m)
+{
+  const graphene_simd4f_t r0 = graphene_simd4f_init (1.0f, 0.0f, 0.0f, 0.0f);
+  const graphene_simd4f_t r1 = graphene_simd4f_init (0.0f, 1.0f, 0.0f, 0.0f);
+  const graphene_simd4f_t r2 = graphene_simd4f_init (0.0f, 0.0f, 1.0f, 0.0f);
+  const graphene_simd4f_t r3 = graphene_simd4f_init (0.0f, 0.0f, 0.0f, 1.0f);
+
+  return graphene_simd4f_cmp_eq (m->x, r0) &&
+         graphene_simd4f_cmp_eq (m->y, r1) &&
+         graphene_simd4f_cmp_eq (m->z, r2) &&
+         graphene_simd4f_cmp_eq (m->w, r3);
+}
+
+/**
+ * graphene_simd4x4f_is_2d:
+ * @m: a #graphene_simd4x4f_t
+ *
+ * Checks whether the given matrix is compatible with an affine
+ * transformation matrix.
+ *
+ * Returns: `true` if the matrix is compatible with an affine
+ *   transformation matrix
+ *
+ * Since: 1.0
+ */
+static inline bool
+graphene_simd4x4f_is_2d (const graphene_simd4x4f_t *m)
+{
+  float f[4];
+
+  if (!(fabsf (graphene_simd4f_get_z (m->x)) < FLT_EPSILON && fabsf (graphene_simd4f_get_w (m->x)) < FLT_EPSILON))
+    return false;
+
+  if (!(fabsf (graphene_simd4f_get_z (m->y)) < FLT_EPSILON && fabsf (graphene_simd4f_get_w (m->y)) < FLT_EPSILON))
+    return false;
+
+  graphene_simd4f_dup_4f (m->z, f);
+  if (!(fabsf (f[0]) < FLT_EPSILON &&
+        fabsf (f[1]) < FLT_EPSILON &&
+        1.f - fabsf (f[2]) < FLT_EPSILON &&
+        fabsf (f[3]) < FLT_EPSILON))
+    return false;
+
+  if (!(fabsf (graphene_simd4f_get_z (m->w)) < FLT_EPSILON && 1.f - fabsf (graphene_simd4f_get_w (m->w)) < FLT_EPSILON))
+    return false;
+
+  return true;
+}
+
+GRAPHENE_END_DECLS
+
+#endif /* __GRAPHENE_SIMD4X4F_H__ */