brl/mutter
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Nice looking scaledown with mipmap emulation

Browse Source 
Add MutterTextureTower, an abstraction for getting a image with
the right level of detail for rendering at a particular scale,
by manually scaling down by powers of two.

This results in much better looking scaled window images when
mipmaps can't be used with texture_from_pixmap (which is the
typical case for current GL drivers.)

When framebuffer objects are available, they are used to do
the scaledown using the GPU without having to pull the data
back from video memory. A software codepath is also available
for the case when FBO's are not present, though performance
will suffer

https://bugzilla.gnome.org/show_bug.cgi?id=601032
This commit is contained in:
Owen W. Taylor 2009-10-30 09:06:28 -04:00
parent 988d2ffab6
commit 47af6a0bbf
4 changed files with 988 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/Makefile.am
View File

@ -28,6 +28,8 @@ mutter_SOURCES= \
compositor/mutter-plugin-manager.c \
compositor/mutter-plugin-manager.h \
compositor/mutter-shaped-texture.c \
compositor/mutter-texture-tower.c \
compositor/mutter-texture-tower.h \
compositor/mutter-window.c \
compositor/mutter-window-private.h \
compositor/mutter-window-group.c \

86
src/compositor/mutter-shaped-texture.c
View File

@ -26,19 +26,27 @@
#include <config.h>
#include "mutter-shaped-texture.h"
#include "mutter-texture-tower.h"
#include <clutter/clutter.h>
#include <cogl/cogl.h>
#include <string.h>
static void mutter_shaped_texture_dispose (GObject *object);
static void mutter_shaped_texture_finalize (GObject *object);
static void mutter_shaped_texture_notify (GObject *object,
GParamSpec *pspec);
static void mutter_shaped_texture_paint (ClutterActor *actor);
static void mutter_shaped_texture_pick (ClutterActor *actor,
const ClutterColor *color);
static void mutter_shaped_texture_update_area (ClutterX11TexturePixmap *texture,
int x,
int y,
int width,
int height);
static void mutter_shaped_texture_dirty_mask (MutterShapedTexture *stex);
#ifdef HAVE_GLX_TEXTURE_PIXMAP
@ -55,6 +63,7 @@ G_DEFINE_TYPE (MutterShapedTexture, mutter_shaped_texture,
struct _MutterShapedTexturePrivate
{
MutterTextureTower *paint_tower;
CoglHandle mask_texture;
CoglHandle material;
CoglHandle material_unshaped;
@ -74,13 +83,17 @@ mutter_shaped_texture_class_init (MutterShapedTextureClass *klass)
{
GObjectClass *gobject_class = (GObjectClass *) klass;
ClutterActorClass *actor_class = (ClutterActorClass *) klass;
ClutterX11TexturePixmapClass *x11_texture_class = (ClutterX11TexturePixmapClass *) klass;
gobject_class->dispose = mutter_shaped_texture_dispose;
gobject_class->finalize = mutter_shaped_texture_finalize;
gobject_class->notify = mutter_shaped_texture_notify;
actor_class->paint = mutter_shaped_texture_paint;
actor_class->pick = mutter_shaped_texture_pick;
x11_texture_class->update_area = mutter_shaped_texture_update_area;
g_type_class_add_private (klass, sizeof (MutterShapedTexturePrivate));
}
@ -93,6 +106,7 @@ mutter_shaped_texture_init (MutterShapedTexture *self)
priv->rectangles = g_array_new (FALSE, FALSE, sizeof (XRectangle));
priv->paint_tower = mutter_texture_tower_new ();
priv->mask_texture = COGL_INVALID_HANDLE;
}
@ -102,6 +116,10 @@ mutter_shaped_texture_dispose (GObject *object)
MutterShapedTexture *self = (MutterShapedTexture *) object;
MutterShapedTexturePrivate *priv = self->priv;
if (priv->paint_tower)
mutter_texture_tower_free (priv->paint_tower);
priv->paint_tower = NULL;
mutter_shaped_texture_dirty_mask (self);
if (priv->material != COGL_INVALID_HANDLE)
@ -138,6 +156,28 @@ mutter_shaped_texture_finalize (GObject *object)
G_OBJECT_CLASS (mutter_shaped_texture_parent_class)->finalize (object);
}
static void
mutter_shaped_texture_notify (GObject *object,
GParamSpec *pspec)
{
if (G_OBJECT_CLASS (mutter_shaped_texture_parent_class)->notify)
G_OBJECT_CLASS (mutter_shaped_texture_parent_class)->notify (object, pspec);
/* It seems like we could just do this out of update_area(), but unfortunately,
* clutter_glx_texture_pixmap() doesn't call through the vtable on the
* initial update_area, so we need to look for changes to the texture
* explicitly.
*/
if (strcmp (pspec->name, "cogl-texture") == 0)
{
MutterShapedTexture *stex = (MutterShapedTexture *) object;
MutterShapedTexturePrivate *priv = stex->priv;
mutter_texture_tower_set_base_texture (priv->paint_tower,
clutter_texture_get_cogl_texture (CLUTTER_TEXTURE (stex)));
}
}
static void
mutter_shaped_texture_dirty_mask (MutterShapedTexture *stex)
{
@ -269,7 +309,30 @@ mutter_shaped_texture_paint (ClutterActor *actor)
if (!CLUTTER_ACTOR_IS_REALIZED (CLUTTER_ACTOR (stex)))
clutter_actor_realize (CLUTTER_ACTOR (stex));
paint_tex = clutter_texture_get_cogl_texture (CLUTTER_TEXTURE (stex));
/* If mipmaps are supported, then the texture filter quality will
* still be HIGH here. In that case we just want to use the base
* texture. If mipmaps are not support then
* on_glx_texture_pixmap_pre_paint() will have reset the texture
* filter quality to MEDIUM, and we should use the MutterTextureTower
* mipmap emulation.
*
* http://bugzilla.openedhand.com/show_bug.cgi?id=1877 is an RFE
* for a better way of handling this.
*
* While it would be nice to have direct access to the 'can_mipmap'
* boolean in ClutterGLXTexturePixmap, since since MutterTextureTower
* creates the scaled down images on demand there is no substantial
* overhead from doing the work to create and update the tower and
* not using it, other than the memory allocated for the MutterTextureTower
* structure itself.
*/
if (clutter_texture_get_filter_quality (CLUTTER_TEXTURE (stex)) == CLUTTER_TEXTURE_QUALITY_HIGH)
paint_tex = clutter_texture_get_cogl_texture (CLUTTER_TEXTURE (stex));
else
paint_tex = mutter_texture_tower_get_paint_texture (priv->paint_tower);
if (paint_tex == COGL_INVALID_HANDLE)
return;
tex_width = cogl_texture_get_width (paint_tex);
tex_height = cogl_texture_get_height (paint_tex);
@ -277,9 +340,6 @@ mutter_shaped_texture_paint (ClutterActor *actor)
if (tex_width == 0 || tex_height == 0) /* no contents yet */
return;
if (paint_tex == COGL_INVALID_HANDLE)
return;
if (priv->rectangles->len < 1)
{
/* If there are no rectangles use a single-layer texture */
@ -438,6 +498,22 @@ mutter_shaped_texture_pick (ClutterActor *actor,
}
}
static void
mutter_shaped_texture_update_area (ClutterX11TexturePixmap *texture,
int x,
int y,
int width,
int height)
{
MutterShapedTexture *stex = (MutterShapedTexture *) texture;
MutterShapedTexturePrivate *priv = stex->priv;
CLUTTER_X11_TEXTURE_PIXMAP_CLASS (mutter_shaped_texture_parent_class)->update_area (texture,
x, y, width, height);
mutter_texture_tower_update_area (priv->paint_tower, x, y, width, height);
}
ClutterActor *
mutter_shaped_texture_new (void)
{

836
src/compositor/mutter-texture-tower.c Normal file
View File

@ -0,0 +1,836 @@
/* -*- mode: C; c-file-style: "gnu"; indent-tabs-mode: nil; -*- */
/*
* MutterTextureTower
*
* Mipmap emulation by creation of scaled down images
*
* Copyright (C) 2009 Red Hat, Inc.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation; either version 2 of the
* License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
* 02111-1307, USA.
*/
#include <math.h>
#include <string.h>
#include "mutter-texture-tower.h"
#ifndef M_LOG2E
#define M_LOG2E 1.4426950408889634074
#endif
#if !CLUTTER_CHECK_VERSION(1,1,3)
static PFNGLGENFRAMEBUFFERSPROC genFramebuffers;
static PFNGLDELETEFRAMEBUFFERSPROC deleteFramebuffers;
static PFNGLBINDFRAMEBUFFERPROC bindFramebuffer;
static PFNGLFRAMEBUFFERTEXTURE2DPROC framebufferTexture2D;
#endif
#define MAX_TEXTURE_LEVELS 12
/* If the texture format in memory doesn't match this, then Mesa
* will do the conversion, so things will still work, but it might
* be slow depending on how efficient Mesa is. These should be the
* native formats unless the display is 16bpp. If conversions
* here are a bottleneck, investigate whether we are converting when
* storing window data *into* the texture before adding extra code
* to handle multiple texture formats.
*/
#if G_BYTE_ORDER == G_LITTLE_ENDIAN
#define TEXTURE_FORMAT COGL_PIXEL_FORMAT_BGRA_8888_PRE
#else
#define TEXTURE_FORMAT COGL_PIXEL_FORMAT_ARGB_8888_PRE
#endif
typedef struct
{
guint16 x1;
guint16 y1;
guint16 x2;
guint16 y2;
} Box;
struct _MutterTextureTower
{
int n_levels;
CoglHandle textures[MAX_TEXTURE_LEVELS];
#if CLUTTER_CHECK_VERSION(1,1,3)
CoglHandle fbos[MAX_TEXTURE_LEVELS];
#else
GLuint fbos[MAX_TEXTURE_LEVELS];
#endif
Box invalid[MAX_TEXTURE_LEVELS];
};
/**
* mutter_texture_tower_new:
*
* Creates a new texture tower. The base texture has to be set with
* mutter_texture_tower_set_base_texture() before use.
*
* Return value: the new texture tower. Free with mutter_texture_tower_free()
*/
MutterTextureTower *
mutter_texture_tower_new (void)
{
MutterTextureTower *tower;
tower = g_slice_new0 (MutterTextureTower);
return tower;
}
/**
* mutter_texture_tower_free:
* @tower: a #MutterTextureTower
*
* Frees a texture tower created with mutter_texture_tower_new().
*/
void
mutter_texture_tower_free (MutterTextureTower *tower)
{
g_return_if_fail (tower != NULL);
mutter_texture_tower_set_base_texture (tower, COGL_INVALID_HANDLE);
g_slice_free (MutterTextureTower, tower);
}
static gboolean
texture_is_rectangle (CoglHandle texture)
{
GLuint gl_tex;
GLenum gl_target;
cogl_texture_get_gl_texture (texture, &gl_tex, &gl_target);
return gl_target == CGL_TEXTURE_RECTANGLE_ARB;
}
static void
free_texture (CoglHandle texture)
{
GLuint gl_tex;
GLenum gl_target;
cogl_texture_get_gl_texture (texture, &gl_tex, &gl_target);
if (gl_target == CGL_TEXTURE_RECTANGLE_ARB)
glDeleteTextures (1, &gl_tex);
cogl_texture_unref (texture);
}
/**
* mutter_texture_tower_update_area:
* @tower: a MutterTextureTower
* @texture: the new texture used as a base for scaled down versions
*
* Sets the base texture that is the scaled texture that the
* scaled textures of the tower are derived from. The texture itself
* will be used as level 0 of the tower and will be referenced until
* unset or until the tower is freed.
*/
void
mutter_texture_tower_set_base_texture (MutterTextureTower *tower,
CoglHandle texture)
{
int i;
g_return_if_fail (tower != NULL);
if (texture == tower->textures[0])
return;
if (tower->textures[0] != COGL_INVALID_HANDLE)
{
for (i = 1; i < tower->n_levels; i++)
{
if (tower->textures[i] != COGL_INVALID_HANDLE)
{
free_texture (tower->textures[i]);
tower->textures[i] = COGL_INVALID_HANDLE;
}
#if CLUTTER_CHECK_VERSION(1,1,3)
if (tower->fbos[i] != COGL_INVALID_HANDLE)
{
cogl_offscreen_unref (tower->fbos[i]);
tower->fbos[i] = COGL_INVALID_HANDLE;
}
#else
if (tower->fbos[i] != 0)
{
(*deleteFramebuffers) (1, &tower->fbos[i]);
tower->fbos[i] = 0;
}
#endif
}
cogl_texture_unref (tower->textures[0]);
}
tower->textures[0] = texture;
if (tower->textures[0] != COGL_INVALID_HANDLE)
{
int width, height;
cogl_texture_ref (tower->textures[0]);
width = cogl_texture_get_width (tower->textures[0]);
height = cogl_texture_get_height (tower->textures[0]);
tower->n_levels = 1 + MAX ((int)(M_LOG2E * log (width)), (int)(M_LOG2E * log (height)));
tower->n_levels = MIN(tower->n_levels, MAX_TEXTURE_LEVELS);
mutter_texture_tower_update_area (tower, 0, 0, width, height);
}
else
{
tower->n_levels = 0;
}
}
/**
* mutter_texture_tower_update_area:
* @tower: a MutterTextureTower
* @x: X coordinate of upper left of rectangle that changed
* @y: Y coordinate of upper left of rectangle that changed
* @width: width of rectangle that changed
* @height: height rectangle that changed
*
* Mark a region of the base texture as having changed; the next
* time a scaled down version of the base texture is retrieved,
* the appropriate area of the scaled down texture will be updated.
*/
void
mutter_texture_tower_update_area (MutterTextureTower *tower,
int x,
int y,
int width,
int height)
{
int texture_width, texture_height;
Box invalid;
int i;
g_return_if_fail (tower != NULL);
texture_width = cogl_texture_get_width (tower->textures[0]);
texture_height = cogl_texture_get_height (tower->textures[0]);
invalid.x1 = x;
invalid.y1 = y;
invalid.x2 = x + width;
invalid.y2 = y + height;
for (i = 1; i < tower->n_levels; i++)
{
texture_width = MAX (1, texture_width / 2);
texture_height = MAX (1, texture_height / 2);
invalid.x1 = invalid.x1 / 2;
invalid.y1 = invalid.y1 / 2;
invalid.x2 = MIN (texture_width, (invalid.x2 + 1) / 2);
invalid.y2 = MIN (texture_height, (invalid.y2 + 1) / 2);
if (tower->invalid[i].x1 == tower->invalid[i].x2 ||
tower->invalid[i].y1 == tower->invalid[i].y2)
{
tower->invalid[i] = invalid;
}
else
{
tower->invalid[i].x1 = MIN (tower->invalid[i].x1, invalid.x1);
tower->invalid[i].y1 = MIN (tower->invalid[i].y1, invalid.y1);
tower->invalid[i].x2 = MAX (tower->invalid[i].x2, invalid.x2);
tower->invalid[i].y2 = MAX (tower->invalid[i].y2, invalid.y2);
}
}
}
/* It generally looks worse if we scale up a window texture by even a
* small amount than if we scale it down using bilinear filtering, so
* we always pick the *larger* adjacent level. */
#define LOD_BIAS (-0.49)
/* This determines the appropriate level of detail to use when drawing the
* texture, in a way that corresponds to what the GL specification does
* when mip-mapping. This is probably fancier and slower than what we need,
* but we do the computation only once each time we paint a window, and
* its easier to just use the equations from the specification than to
* come up with something simpler.
*
* If window is being painted at an angle from the viewer, then we have to
* pick a point in the texture; we use the middle of the texture (which is
* why the width/height are passed in.) This is not the normal case for
* Mutter.
*/
static int
get_paint_level (int width, int height)
{
CoglMatrix projection, modelview, pm;
float v[4];
double viewport_width, viewport_height;
double u0, v0;
double xc, yc, wc;
double dxdu_, dxdv_, dydu_, dydv_;
double det_, det_sq;
double rho_sq;
double lambda;
/* See
* http://www.opengl.org/registry/doc/glspec32.core.20090803.pdf
* Section 3.8.9, p. 1.6.2. Here we have
*
* u(x,y) = x_o;
* v(x,y) = y_o;
*
* Since we are mapping 1:1 from object coordinates into pixel
* texture coordinates, the clip coordinates are:
*
* (x_c) (x_o) (u)
* (y_c) = (M_projection)(M_modelview) (y_o) = (PM) (v)
* (z_c) (z_o) (0)
* (w_c) (w_o) (1)
*/
cogl_get_projection_matrix (&projection);
cogl_get_modelview_matrix (&modelview);
cogl_matrix_multiply (&pm, &projection, &modelview);
cogl_get_viewport (v);
viewport_width = v[2];
viewport_height = v[3];
u0 = width / 2.;
v0 = height / 2.;
xc = pm.xx * u0 + pm.xy * v0 + pm.xw;
yc = pm.yx * u0 + pm.yy * v0 + pm.yw;
wc = pm.wx * u0 + pm.wy * v0 + pm.ww;
/* We'll simplify the equations below for a bit of micro-optimization.
* The commented out code is the unsimplified version.
// Partial derivates of window coordinates:
//
// x_w = 0.5 * viewport_width * x_c / w_c + viewport_center_x
// y_w = 0.5 * viewport_height * y_c / w_c + viewport_center_y
//
// with respect to u, v, using
// d(a/b)/dx = da/dx * (1/b) - a * db/dx / (b^2)
dxdu = 0.5 * viewport_width * (pm.xx - pm.wx * (xc/wc)) / wc;
dxdv = 0.5 * viewport_width * (pm.xy - pm.wy * (xc/wc)) / wc;
dydu = 0.5 * viewport_height * (pm.yx - pm.wx * (yc/wc)) / wc;
dydv = 0.5 * viewport_height * (pm.yy - pm.wy * (yc/wc)) / wc;
// Compute the inverse partials as the matrix inverse
det = dxdu * dydv - dxdv * dydu;
dudx = dydv / det;
dudy = - dxdv / det;
dvdx = - dydu / det;
dvdy = dvdu / det;
// Scale factor; maximum of the distance in texels for a change of 1 pixel
// in the X direction or 1 pixel in the Y direction
rho = MAX (sqrt (dudx * dudx + dvdx * dvdx), sqrt(dudy * dudy + dvdy * dvdy));
// Level of detail
lambda = log2 (rho) + LOD_BIAS;
*/
/* dxdu * wc, etc */
dxdu_ = 0.5 * viewport_width * (pm.xx - pm.wx * (xc/wc));
dxdv_ = 0.5 * viewport_width * (pm.xy - pm.wy * (xc/wc));
dydu_ = 0.5 * viewport_height * (pm.yx - pm.wx * (yc/wc));
dydv_ = 0.5 * viewport_height * (pm.yy - pm.wy * (yc/wc));
/* det * wc^2 */
det_ = dxdu_ * dydv_ - dxdv_ * dydu_;
det_sq = det_ * det_;
if (det_sq == 0.0)
return -1;
/* (rho * det * wc)^2 */
rho_sq = MAX (dydv_ * dydv_ + dydu_ * dydu_, dxdv_ * dxdv_ + dxdu_ * dxdu_);
lambda = 0.5 * M_LOG2E * log (rho_sq * wc * wc / det_sq) + LOD_BIAS;
#if 0
g_print ("%g %g %g\n", 0.5 * viewport_width * pm.xx / pm.ww, 0.5 * viewport_height * pm.yy / pm.ww, lambda);
#endif
if (lambda <= 0.)
return 0;
else
return (int)(0.5 + lambda);
}
static gboolean
is_power_of_two (int x)
{
return (x & (x - 1)) == 0;
}
static void
texture_tower_create_texture (MutterTextureTower *tower,
int level,
int width,
int height)
{
if ((!is_power_of_two (width) || !is_power_of_two (height)) &&
texture_is_rectangle (tower->textures[level - 1]))
{
GLuint tex = 0;
glGenTextures (1, &tex);
glBindTexture (CGL_TEXTURE_RECTANGLE_ARB, tex);
glTexImage2D (CGL_TEXTURE_RECTANGLE_ARB, 0,
GL_RGBA, width,height,
#if TEXTURE_FORMAT == COGL_PIXEL_FORMAT_BGRA_8888_PRE
0, GL_BGRA, GL_UNSIGNED_BYTE,
#else /* assume big endian */
0, GL_BGRA, GL_UNSIGNED_INT_8_8_8_8_REV,
#endif
NULL);
tower->textures[level] = cogl_texture_new_from_foreign (tex, CGL_TEXTURE_RECTANGLE_ARB,
width, height,
0, 0,
TEXTURE_FORMAT);
}
else
{
tower->textures[level] = cogl_texture_new_with_size (width, height,
COGL_TEXTURE_NO_AUTO_MIPMAP,
TEXTURE_FORMAT);
}
tower->invalid[level].x1 = 0;
tower->invalid[level].y1 = 0;
tower->invalid[level].x2 = width;
tower->invalid[level].y2 = height;
}
/* The COGL fbo (render-to-texture) support is pretty hard to use in
* Clutter 1.0; there's no way to save and restore the old projection
* matrix and viewport without ugly workarounds that require explicit
* access to the ClutterStage. In Clutter 1.2, the save/restore is
* automatic. For now, until we depend on Clutter 1.2, we use GL
* directly for render-to-texture. The main downside (other than
* a lot of verbosity) is that we have to save the state, reset anything
* that we think COGL might have left in a way we don't want it, then
* restore the old state.
*/
#if CLUTTER_CHECK_VERSION(1,1,3)
static gboolean
texture_tower_revalidate_fbo (MutterTextureTower *tower,
int level)
{
CoglHandle source_texture = tower->textures[level - 1];
int source_texture_width = cogl_texture_get_width (source_texture);
int source_texture_height = cogl_texture_get_height (source_texture);
CoglHandle dest_texture = tower->textures[level];
int dest_texture_width = cogl_texture_get_width (dest_texture);
int dest_texture_height = cogl_texture_get_height (dest_texture);
Box *invalid = &tower->invalid[level];
CoglMatrix modelview;
if (tower->fbos[level] == COGL_INVALID_HANDLE)
tower->fbos[level] = cogl_offscreen_new_to_texture (dest_texture);
if (tower->fbos[level] == COGL_INVALID_HANDLE)
return FALSE;
cogl_push_draw_buffer ();
cogl_set_draw_buffer (COGL_OFFSCREEN_BUFFER, tower->fbos[level]);
cogl_ortho (0, dest_texture_width, dest_texture_height, 0, -1., 1.);
cogl_matrix_init_identity (&modelview);
cogl_set_modelview_matrix (&modelview);
cogl_set_source_texture (tower->textures[level - 1]);
cogl_rectangle_with_texture_coords (invalid->x1, invalid->y1,
invalid->x2, invalid->y2,
(2. * invalid->x1) / source_texture_width,
(2. * invalid->y1) / source_texture_height,
(2. * invalid->x2) / source_texture_width,
(2. * invalid->y2) / source_texture_height);
cogl_pop_draw_buffer ();
return TRUE;
}
#else
static void
initialize_gl_functions (void)
{
static gboolean initialized = FALSE;
if (!initialized)
{
initialized = TRUE;
genFramebuffers = (PFNGLGENFRAMEBUFFERSPROC) cogl_get_proc_address ("glGenFramebuffersEXT");
deleteFramebuffers = (PFNGLDELETEFRAMEBUFFERSPROC) cogl_get_proc_address ("glDeleteFramebuffersEXT");
bindFramebuffer = (PFNGLBINDFRAMEBUFFERPROC) cogl_get_proc_address ("glBindFramebufferEXT");
framebufferTexture2D = (PFNGLFRAMEBUFFERTEXTURE2DPROC) cogl_get_proc_address ("glFramebufferTexture2D");
}
}
static gboolean
texture_tower_revalidate_fbo (MutterTextureTower *tower,
int level)
{
CoglHandle source_texture = tower->textures[level - 1];
int source_texture_width = cogl_texture_get_width (source_texture);
int source_texture_height = cogl_texture_get_height (source_texture);
CoglHandle dest_texture = tower->textures[level];
int dest_texture_width = cogl_texture_get_width (dest_texture);
int dest_texture_height = cogl_texture_get_height (dest_texture);
ClutterActorBox source_box;
Box *dest_box;
GLuint source_gl_tex;
GLenum source_gl_target;
if (!cogl_features_available (COGL_FEATURE_OFFSCREEN))
return FALSE;
initialize_gl_functions ();
/* Create the frame-buffer object that renders to the texture, if
* it doesn't exist; just bind it for rendering if it does */
if (tower->fbos[level] == 0)
{
GLuint dest_gl_tex;
GLenum dest_gl_target;
cogl_texture_get_gl_texture (dest_texture, &dest_gl_tex, &dest_gl_target);
(*genFramebuffers) (1, &tower->fbos[level]);
(*bindFramebuffer) (GL_FRAMEBUFFER_EXT, tower->fbos[level]);
(*framebufferTexture2D) (GL_FRAMEBUFFER_EXT, GL_COLOR_ATTACHMENT0_EXT,
dest_gl_target, dest_gl_tex, 0);
}
else
{
(*bindFramebuffer) (GL_FRAMEBUFFER_EXT, tower->fbos[level]);
}
/* Save the old state (other than the transformation matrices) */
glPushAttrib (GL_ENABLE_BIT | GL_COLOR_BUFFER_BIT | GL_TEXTURE_BIT | GL_VIEWPORT_BIT);
/* And set up the state we need */
glDisable (GL_BLEND);
glDisable (GL_SCISSOR_TEST);
glDisable (GL_STENCIL_TEST);
glDisable (GL_CLIP_PLANE3);
glDisable (GL_CLIP_PLANE2);
glDisable (GL_CLIP_PLANE1);
glDisable (GL_CLIP_PLANE0);
cogl_texture_get_gl_texture (source_texture, &source_gl_tex, &source_gl_target);
glActiveTextureARB (GL_TEXTURE0_ARB);
if (source_gl_target == GL_TEXTURE_2D)
glDisable (GL_TEXTURE_RECTANGLE);
else
glDisable (GL_TEXTURE_2D);
glEnable (source_gl_target);
glBindTexture (source_gl_target, source_gl_tex);
glTexParameteri (source_gl_target, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
glTexParameteri (source_gl_target, GL_TEXTURE_WRAP_S, GL_CLAMP);
glTexParameteri (source_gl_target, GL_TEXTURE_WRAP_T, GL_CLAMP);
glTexEnvi (GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
/* In theory, we should loop over all the texture units supported
* by the GL implementation, but here we just assume that no more
* than three are used by Mutter and all GL implementations we care
* about will support at least 3.
*/
glActiveTextureARB (GL_TEXTURE1_ARB);
glDisable (GL_TEXTURE_2D);
glDisable (GL_TEXTURE_RECTANGLE);
glActiveTextureARB (GL_TEXTURE2_ARB);
glDisable (GL_TEXTURE_2D);
glDisable (GL_TEXTURE_RECTANGLE);
glViewport (0, 0, dest_texture_width, dest_texture_height);
/* Save the transformation matrices and set up new ones that map
* coordinates directly onto the destination texture */
glMatrixMode (GL_MODELVIEW);
glPushMatrix ();
glLoadIdentity ();
glMatrixMode (GL_PROJECTION);
glPushMatrix();
glLoadIdentity ();
glOrtho (0, dest_texture_width, 0, dest_texture_height, -1., 1.);
/* Draw */
dest_box = &tower->invalid[level];
if (texture_is_rectangle (source_texture))
{
source_box.x1 = 2 * dest_box->x1;
source_box.y1 = 2 * dest_box->y1;
source_box.x2 = 2 * dest_box->x2;
source_box.y2 = 2 * dest_box->y2;
}
else
{
source_box.x1 = (2. * dest_box->x1) / source_texture_width;
source_box.y1 = (2. * dest_box->y1) / source_texture_height;
source_box.x2 = (2. * dest_box->x2) / source_texture_width;
source_box.y2 = (2. * dest_box->y2) / source_texture_height;
}
glColor3f (0., 1., 1.);
glBegin (GL_QUADS);
glTexCoord2f (source_box.x1, source_box.y1);
glVertex2f (dest_box->x1, dest_box->y1);
glTexCoord2f (source_box.x2, source_box.y1);
glVertex2f (dest_box->x2, dest_box->y1);
glTexCoord2f (source_box.x2, source_box.y2);
glVertex2f (dest_box->x2, dest_box->y2);
glTexCoord2f (source_box.x1, source_box.y2);
glVertex2f (dest_box->x1, dest_box->y2);
glEnd ();
/* And restore everything back the way we found it */
glMatrixMode (GL_PROJECTION);
glPopMatrix ();
glMatrixMode (GL_MODELVIEW);
glPopMatrix ();
glPopAttrib ();
(*bindFramebuffer) (GL_FRAMEBUFFER_EXT, 0);
return TRUE;
}
#endif
static void
fill_copy (guchar *buf,
const guchar *source,
int width)
{
memcpy (buf, source, width * 4);
}
static void
fill_scale_down (guchar *buf,
const guchar *source,
int width)
{
while (width > 1)
{
buf[0] = (source[0] + source[4]) / 2;
buf[1] = (source[1] + source[5]) / 2;
buf[2] = (source[2] + source[6]) / 2;
buf[3] = (source[3] + source[7]) / 2;
buf += 4;
source += 8;
width -= 2;
}
if (width > 0)
{
buf[0] = source[0] / 2;
buf[1] = source[1] / 2;
buf[2] = source[2] / 2;
buf[3] = source[3] / 2;
}
}
static void
texture_tower_revalidate_client (MutterTextureTower *tower,
int level)
{
CoglHandle source_texture = tower->textures[level - 1];
int source_texture_width = cogl_texture_get_width (source_texture);
int source_texture_height = cogl_texture_get_height (source_texture);
guint source_rowstride;
guchar *source_data;
CoglHandle dest_texture = tower->textures[level];
int dest_texture_width = cogl_texture_get_width (dest_texture);
int dest_texture_height = cogl_texture_get_height (dest_texture);
int dest_x = tower->invalid[level].x1;
int dest_y = tower->invalid[level].y1;
int dest_width = tower->invalid[level].x2 - tower->invalid[level].x1;
int dest_height = tower->invalid[level].y2 - tower->invalid[level].y1;
guchar *dest_data;
guchar *source_tmp1 = NULL, *source_tmp2 = NULL;
int i, j;
source_rowstride = source_texture_width * 4;
source_data = g_malloc (source_texture_height * source_rowstride);
cogl_texture_get_data (source_texture, TEXTURE_FORMAT, source_rowstride,
source_data);
dest_data = g_malloc (dest_height * dest_width * 4);
if (dest_texture_height < source_texture_height)
{
source_tmp1 = g_malloc (dest_width * 4);
source_tmp2 = g_malloc (dest_width * 4);
}
for (i = 0; i < dest_height; i++)
{
guchar *dest_row = dest_data + i * dest_width * 4;
if (dest_texture_height < source_texture_height)
{
guchar *source1, *source2;
guchar *dest;
if (dest_texture_width < source_texture_width)
{
fill_scale_down (source_tmp1,
source_data + ((i + dest_y) * 2) * source_rowstride + dest_x * 2 * 4,
dest_width * 2);
fill_scale_down (source_tmp2,
source_data + ((i + dest_y) * 2 + 1) * source_rowstride + dest_x * 2 * 4,
dest_width * 2);
}
else
{
fill_copy (source_tmp1,
source_data + ((i + dest_y) * 2) * source_rowstride + dest_x * 4,
dest_width);
fill_copy (source_tmp2,
source_data + ((i + dest_y) * 2 + 1) * source_rowstride + dest_x * 4,
dest_width);
}
source1 = source_tmp1;
source2 = source_tmp2;
dest = dest_row;
for (j = 0; j < dest_width * 4; j++)
*(dest++) = (*(source1++) + *(source2++)) / 2;
}
else
{
if (dest_texture_width < source_texture_width)
fill_scale_down (dest_row,
source_data + (i + dest_y) * source_rowstride + dest_x * 2 * 4,
dest_width * 2);
else
fill_copy (dest_row,
source_data + (i + dest_y) * source_rowstride,
dest_width);
}
}
cogl_texture_set_region (dest_texture,
0, 0,
dest_x, dest_y,
dest_width, dest_height,
dest_width, dest_height,
TEXTURE_FORMAT,
4 * dest_width,
dest_data);
if (dest_height < source_texture_height)
{
g_free (source_tmp1);
g_free (source_tmp2);
}
g_free (source_data);
g_free (dest_data);
}
static void
texture_tower_revalidate (MutterTextureTower *tower,
int level)
{
if (!texture_tower_revalidate_fbo (tower, level))
texture_tower_revalidate_client (tower, level);
}
/**
* mutter_texture_tower_get_paint_texture:
* @tower: a MutterTextureTower
*
* Gets the texture from the tower that best matches the current
* rendering scale. (On the assumption here the texture is going to
* be rendered with vertex coordinates that correspond to its
* size in pixels, so a 200x200 texture will be rendered on the
* rectangle (0, 0, 200, 200).
*
* Return value: the COGL texture handle to use for painting, or
* %COGL_INVALID_HANDLE if no base texture has yet been set.
*/
CoglHandle
mutter_texture_tower_get_paint_texture (MutterTextureTower *tower)
{
int texture_width, texture_height;
int level;
g_return_val_if_fail (tower != NULL, COGL_INVALID_HANDLE);
if (tower->textures[0] == COGL_INVALID_HANDLE)
return COGL_INVALID_HANDLE;
texture_width = cogl_texture_get_width (tower->textures[0]);
texture_height = cogl_texture_get_height (tower->textures[0]);
level = get_paint_level(texture_width, texture_height);
if (level < 0) /* singular paint matrix, scaled to nothing */
return COGL_INVALID_HANDLE;
level = MIN (level, tower->n_levels - 1);
if (tower->textures[level] == COGL_INVALID_HANDLE ||
(tower->invalid[level].x2 != tower->invalid[level].x1 &&
tower->invalid[level].y2 != tower->invalid[level].y1))
{
int i;
for (i = 1; i <= level; i++)
{
/* Use "floor" convention here to be consistent with the NPOT texture extension */
texture_width = MAX (1, texture_width / 2);
texture_height = MAX (1, texture_height / 2);
if (tower->textures[i] == COGL_INVALID_HANDLE)
texture_tower_create_texture (tower, i, texture_width, texture_height);
}
for (i = 1; i <= level; i++)
{
if (tower->invalid[level].x2 != tower->invalid[level].x1 &&
tower->invalid[level].y2 != tower->invalid[level].y1)
texture_tower_revalidate (tower, i);
}
}
return tower->textures[level];
}

69
src/compositor/mutter-texture-tower.h Normal file
View File

@ -0,0 +1,69 @@
/* -*- mode: C; c-file-style: "gnu"; indent-tabs-mode: nil; -*- */
/*
* MutterTextureTower
*
* Mipmap emulation by creation of scaled down images
*
* Copyright (C) 2009 Red Hat, Inc.
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License as
* published by the Free Software Foundation; either version 2 of the
* License, or (at your option) any later version.
*
* This program is distributed in the hope that it will be useful, but
* WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA
* 02111-1307, USA.
*/
#ifndef __MUTTER_TEXTURE_TOWER_H__
#define __MUTTER_TEXTURE_TOWER_H__
#include <clutter/clutter.h>
G_BEGIN_DECLS
/**
* SECTION:MutterTextureTower
* @short_description: mipmap emulation by creation of scaled down images
*
* A #MutterTextureTower is used to get good looking scaled down images when
* we can't use the GL drivers mipmap support. There are two separate reasons
*
* - Some cards (including radeon cards <= r5xx) only support
* TEXTURE_RECTANGLE_ARB and not NPOT textures. Rectangular textures
* are defined not to support mipmapping.
* - Even when NPOT textures are available, the combination of NPOT
* textures, texture_from_pixmap, and mipmapping doesn't typically
* work, since the X server doesn't allocate pixmaps in the right
* layout for mipmapping.
*
* So, what we do is create the "mipmap" levels ourselves by successive
* power-of-two scaledowns, and when rendering pick the single texture
* that best matches the scale we are rendering at. (Since we aren't
* typically using perspective transforms, we'll frequently have a single
* scale for the entire texture.)
*/
typedef struct _MutterTextureTower MutterTextureTower;
MutterTextureTower *mutter_texture_tower_new (void);
void mutter_texture_tower_free (MutterTextureTower *tower);
void mutter_texture_tower_set_base_texture (MutterTextureTower *tower,
CoglHandle texture);
void mutter_texture_tower_update_area (MutterTextureTower *tower,
int x,
int y,
int width,
int height);
CoglHandle mutter_texture_tower_get_paint_texture (MutterTextureTower *tower);
G_BEGIN_DECLS
#endif /* __MUTTER_TEXTURE_TOWER_H__ */