mutter/cogl/cogl-framebuffer.c

2754 lines
90 KiB
C
Raw Normal View History

[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
/*
* Cogl
*
* An object oriented GL/GLES Abstraction/Utility Layer
*
* Copyright (C) 2007,2008,2009,2012 Intel Corporation.
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library 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
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library. If not, see
* <http://www.gnu.org/licenses/>.
*
*
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include <string.h>
#include "cogl-debug.h"
#include "cogl-context-private.h"
#include "cogl-display-private.h"
#include "cogl-renderer-private.h"
#include "cogl-object-private.h"
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
#include "cogl-util.h"
#include "cogl-texture-private.h"
#include "cogl-framebuffer-private.h"
#include "cogl-onscreen-template-private.h"
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
#include "cogl-clip-stack.h"
#include "cogl-journal-private.h"
#include "cogl-winsys-private.h"
#include "cogl-pipeline-state-private.h"
#include "cogl-matrix-private.h"
#include "cogl-primitive-private.h"
#include "cogl-offscreen.h"
#include "cogl1-context.h"
#include "cogl-private.h"
#include "cogl-primitives-private.h"
#include "cogl-path-private.h"
Adds CoglError api Although we use GLib internally in Cogl we would rather not leak GLib api through Cogl's own api, except through explicitly namespaced cogl_glib_ / cogl_gtype_ feature apis. One of the benefits we see to not leaking GLib through Cogl's public API is that documentation for Cogl won't need to first introduce the Glib API to newcomers, thus hopefully lowering the barrier to learning Cogl. This patch provides a Cogl specific typedef for reporting runtime errors which by no coincidence matches the typedef for GError exactly. If Cogl is built with --enable-glib (default) then developers can even safely assume that a CoglError is a GError under the hood. This patch also enforces a consistent policy for when NULL is passed as an error argument and an error is thrown. In this case we log the error and abort the application, instead of silently ignoring it. In common cases where nothing has been implemented to handle a particular error and/or where applications are just printing the error and aborting themselves then this saves some typing. This also seems more consistent with language based exceptions which usually cause a program to abort if they are not explicitly caught (which passing a non-NULL error signifies in this case) Since this policy for NULL error pointers is stricter than the standard GError convention, there is a clear note in the documentation to warn developers that are used to using the GError api. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit b068d5ea09ab32c37e8c965fc8582c85d1b2db46) Note: Since we can't change the Cogl 1.x api the patch was changed to not rename _error_quark() functions to be _error_domain() functions and although it's a bit ugly, instead of providing our own CoglError type that's compatible with GError we simply #define CoglError to GError unless Cogl is built with glib disabled. Note: this patch does technically introduce an API break since it drops the cogl_error_get_type() symbol generated by glib-mkenum (Since the CoglError enum was replaced by a CoglSystemError enum) but for now we are assuming that this will not affect anyone currently using the Cogl API. If this does turn out to be a problem in practice then we would be able to fix this my manually copying an implementation of cogl_error_get_type() generated by glib-mkenum into a compatibility source file and we could also define the original COGL_ERROR_ enums for compatibility too. Note: another minor concern with cherry-picking this patch to the 1.14 branch is that an api scanner would be lead to believe that some APIs have changed, and for example the gobject-introspection parser which understands the semantics of GError will not understand the semantics of CoglError. We expect most people that have tried to use gobject-introspection with Cogl already understand though that it is not well suited to generating bindings of the Cogl api anyway and we aren't aware or anyone depending on such bindings for apis involving GErrors. (GnomeShell only makes very-very minimal use of Cogl via the gjs bindings for the cogl_rectangle and cogl_color apis.) The main reason we have cherry-picked this patch to the 1.14 branch even given the above concerns is that without it it would become very awkward for us to cherry-pick other beneficial patches from master.
2012-08-31 14:28:27 -04:00
#include "cogl-error-private.h"
#include "cogl-texture-gl-private.h"
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
typedef struct _CoglFramebufferStackEntry
{
CoglFramebuffer *draw_buffer;
CoglFramebuffer *read_buffer;
} CoglFramebufferStackEntry;
extern CoglObjectClass _cogl_onscreen_class;
#ifdef COGL_ENABLE_DEBUG
static CoglUserDataKey wire_pipeline_key;
#endif
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
static void _cogl_offscreen_free (CoglOffscreen *offscreen);
COGL_OBJECT_DEFINE_WITH_CODE (Offscreen, offscreen,
_cogl_offscreen_class.virt_unref =
_cogl_framebuffer_unref);
COGL_OBJECT_DEFINE_DEPRECATED_REF_COUNTING (offscreen);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
/* XXX:
* The CoglObject macros don't support any form of inheritance, so for
* now we implement the CoglObject support for the CoglFramebuffer
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
* abstract class manually.
*/
Adds CoglError api Although we use GLib internally in Cogl we would rather not leak GLib api through Cogl's own api, except through explicitly namespaced cogl_glib_ / cogl_gtype_ feature apis. One of the benefits we see to not leaking GLib through Cogl's public API is that documentation for Cogl won't need to first introduce the Glib API to newcomers, thus hopefully lowering the barrier to learning Cogl. This patch provides a Cogl specific typedef for reporting runtime errors which by no coincidence matches the typedef for GError exactly. If Cogl is built with --enable-glib (default) then developers can even safely assume that a CoglError is a GError under the hood. This patch also enforces a consistent policy for when NULL is passed as an error argument and an error is thrown. In this case we log the error and abort the application, instead of silently ignoring it. In common cases where nothing has been implemented to handle a particular error and/or where applications are just printing the error and aborting themselves then this saves some typing. This also seems more consistent with language based exceptions which usually cause a program to abort if they are not explicitly caught (which passing a non-NULL error signifies in this case) Since this policy for NULL error pointers is stricter than the standard GError convention, there is a clear note in the documentation to warn developers that are used to using the GError api. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit b068d5ea09ab32c37e8c965fc8582c85d1b2db46) Note: Since we can't change the Cogl 1.x api the patch was changed to not rename _error_quark() functions to be _error_domain() functions and although it's a bit ugly, instead of providing our own CoglError type that's compatible with GError we simply #define CoglError to GError unless Cogl is built with glib disabled. Note: this patch does technically introduce an API break since it drops the cogl_error_get_type() symbol generated by glib-mkenum (Since the CoglError enum was replaced by a CoglSystemError enum) but for now we are assuming that this will not affect anyone currently using the Cogl API. If this does turn out to be a problem in practice then we would be able to fix this my manually copying an implementation of cogl_error_get_type() generated by glib-mkenum into a compatibility source file and we could also define the original COGL_ERROR_ enums for compatibility too. Note: another minor concern with cherry-picking this patch to the 1.14 branch is that an api scanner would be lead to believe that some APIs have changed, and for example the gobject-introspection parser which understands the semantics of GError will not understand the semantics of CoglError. We expect most people that have tried to use gobject-introspection with Cogl already understand though that it is not well suited to generating bindings of the Cogl api anyway and we aren't aware or anyone depending on such bindings for apis involving GErrors. (GnomeShell only makes very-very minimal use of Cogl via the gjs bindings for the cogl_rectangle and cogl_color apis.) The main reason we have cherry-picked this patch to the 1.14 branch even given the above concerns is that without it it would become very awkward for us to cherry-pick other beneficial patches from master.
2012-08-31 14:28:27 -04:00
uint32_t
cogl_framebuffer_error_quark (void)
{
return g_quark_from_static_string ("cogl-framebuffer-error-quark");
}
CoglBool
Add -Wmissing-declarations to maintainer flags and fix problems This option to GCC makes it give a warning whenever a global function is defined without a declaration. This should catch cases were we've defined a function but forgot to put it in a header. In that case it is either only used within one file so we should make it static or we should declare it in a header. The following changes where made to fix problems: • Some functions were made static • cogl-path.h (the one containing the 1.0 API) was split into two files, one defining the functions and one defining the enums so that cogl-path.c can include the enum and function declarations from the 2.0 API as well as the function declarations from the 1.0 API. • cogl2-clip-state has been removed. This only had one experimental function called cogl_clip_push_from_path but as this is unstable we might as well remove it favour of the equivalent cogl_framebuffer_* API. • The GLX, SDL and WGL winsys's now have a private header to define their get_vtable function instead of directly declaring in the C file where it is called. • All places that were calling COGL_OBJECT_DEFINE need to have the cogl_is_whatever function declared so these have been added either as a public function or in a private header. • Some files that were not including the header containing their function declarations have been fixed to do so. • Any unused error quark functions have been removed. If we later want them we should add them back one by one and add a declaration for them in a header. • _cogl_is_framebuffer has been renamed to cogl_is_framebuffer and made a public function with a declaration in cogl-framebuffer.h • Similarly for CoglOnscreen. • cogl_vdraw_indexed_attributes is called cogl_framebuffer_vdraw_indexed_attributes in the header. The definition has been changed to match the header. • cogl_index_buffer_allocate has been removed. This had no declaration and I'm not sure what it's supposed to do. • CoglJournal has been changed to use the internal CoglObject macro so that it won't define an exported cogl_is_journal symbol. • The _cogl_blah_pointer_from_handle functions have been removed. CoglHandle isn't used much anymore anyway and in the few places where it is used I think it's safe to just use the implicit cast from void* to the right type. • The test-utils.h header for the conformance tests explicitly disables the -Wmissing-declaration option using a pragma because all of the tests declare their main function without a header. Any mistakes relating to missing declarations aren't really important for the tests. • cogl_quaternion_init_from_quaternion and init_from_matrix have been given declarations in cogl-quaternion.h Reviewed-by: Robert Bragg <robert@linux.intel.com>
2012-03-06 13:21:28 -05:00
cogl_is_framebuffer (void *object)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
CoglObject *obj = object;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
if (obj == NULL)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
return FALSE;
return (obj->klass == &_cogl_onscreen_class ||
obj->klass == &_cogl_offscreen_class);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
void
_cogl_framebuffer_init (CoglFramebuffer *framebuffer,
CoglContext *ctx,
CoglFramebufferType type,
CoglPixelFormat format,
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
int width,
int height)
{
framebuffer->context = ctx;
framebuffer->type = type;
framebuffer->width = width;
framebuffer->height = height;
framebuffer->format = format;
framebuffer->viewport_x = 0;
framebuffer->viewport_y = 0;
framebuffer->viewport_width = width;
framebuffer->viewport_height = height;
framebuffer->viewport_age = 0;
framebuffer->viewport_age_for_scissor_workaround = -1;
framebuffer->dither_enabled = TRUE;
framebuffer->modelview_stack = cogl_matrix_stack_new (ctx);
framebuffer->projection_stack = cogl_matrix_stack_new (ctx);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
framebuffer->dirty_bitmasks = TRUE;
framebuffer->color_mask = COGL_COLOR_MASK_ALL;
framebuffer->samples_per_pixel = 0;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
/* Initialise the clip stack */
_cogl_clip_state_init (&framebuffer->clip_state);
framebuffer->journal = _cogl_journal_new (framebuffer);
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
/* Ensure we know the framebuffer->clear_color* members can't be
* referenced for our fast-path read-pixel optimization (see
* _cogl_journal_try_read_pixel()) until some region of the
* framebuffer is initialized.
*/
framebuffer->clear_clip_dirty = TRUE;
/* XXX: We have to maintain a central list of all framebuffers
* because at times we need to be able to flush all known journals.
*
* Examples where we need to flush all journals are:
* - because journal entries can reference OpenGL texture
* coordinates that may not survive texture-atlas reorganization
* so we need the ability to flush those entries.
* - because although we generally advise against modifying
* pipelines after construction we have to handle that possibility
* and since pipelines may be referenced in journal entries we
* need to be able to flush them before allowing the pipelines to
* be changed.
*
* Note we don't maintain a list of journals and associate
* framebuffers with journals by e.g. having a journal->framebuffer
* reference since that would introduce a circular reference.
*
* Note: As a future change to try and remove the need to index all
* journals it might be possible to defer resolving of OpenGL
* texture coordinates for rectangle primitives until we come to
* flush a journal. This would mean for instance that a single
* rectangle entry in a journal could later be expanded into
* multiple quad primitives to handle sliced textures but would mean
* we don't have to worry about retaining references to OpenGL
* texture coordinates that may later become invalid.
*/
ctx->framebuffers = g_list_prepend (ctx->framebuffers, framebuffer);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
void
_cogl_framebuffer_free (CoglFramebuffer *framebuffer)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
CoglContext *ctx = framebuffer->context;
_cogl_clip_state_destroy (&framebuffer->clip_state);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
cogl_object_unref (framebuffer->modelview_stack);
framebuffer->modelview_stack = NULL;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
cogl_object_unref (framebuffer->projection_stack);
framebuffer->projection_stack = NULL;
cogl_object_unref (framebuffer->journal);
if (ctx->viewport_scissor_workaround_framebuffer == framebuffer)
ctx->viewport_scissor_workaround_framebuffer = NULL;
ctx->framebuffers = g_list_remove (ctx->framebuffers, framebuffer);
if (ctx->current_draw_buffer == framebuffer)
ctx->current_draw_buffer = NULL;
if (ctx->current_read_buffer == framebuffer)
ctx->current_read_buffer = NULL;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
const CoglWinsysVtable *
_cogl_framebuffer_get_winsys (CoglFramebuffer *framebuffer)
{
return framebuffer->context->display->renderer->winsys_vtable;
}
/* This version of cogl_clear can be used internally as an alternative
* to avoid flushing the journal or the framebuffer state. This is
* needed when doing operations that may be called whiling flushing
* the journal */
void
_cogl_framebuffer_clear_without_flush4f (CoglFramebuffer *framebuffer,
unsigned long buffers,
float red,
float green,
float blue,
float alpha)
{
CoglContext *ctx = framebuffer->context;
if (!buffers)
{
static CoglBool shown = FALSE;
if (!shown)
{
g_warning ("You should specify at least one auxiliary buffer "
"when calling cogl_framebuffer_clear");
}
return;
}
ctx->driver_vtable->framebuffer_clear (framebuffer,
buffers,
red, green, blue, alpha);
}
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
void
_cogl_framebuffer_dirty (CoglFramebuffer *framebuffer)
{
framebuffer->clear_clip_dirty = TRUE;
}
void
cogl_framebuffer_clear4f (CoglFramebuffer *framebuffer,
unsigned long buffers,
float red,
float green,
float blue,
float alpha)
{
CoglContext *ctx = framebuffer->context;
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
CoglClipStack *clip_stack = _cogl_framebuffer_get_clip_stack (framebuffer);
int scissor_x0;
int scissor_y0;
int scissor_x1;
int scissor_y1;
CoglBool saved_viewport_scissor_workaround;
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
_cogl_clip_stack_get_bounds (clip_stack,
&scissor_x0, &scissor_y0,
&scissor_x1, &scissor_y1);
/* NB: the previous clear could have had an arbitrary clip.
* NB: everything for the last frame might still be in the journal
* but we can't assume anything about how each entry was
* clipped.
* NB: Clutter will scissor its pick renders which would mean all
* journal entries have a common ClipStack entry, but without
* a layering violation Cogl has to explicitly walk the journal
* entries to determine if this is the case.
* NB: We have a software only read-pixel optimization in the
* journal that determines the color at a given framebuffer
* coordinate for simple scenes without rendering with the GPU.
* When Clutter is hitting this fast-path we can expect to
* receive calls to clear the framebuffer with an un-flushed
* journal.
* NB: To fully support software based picking for Clutter we
* need to be able to reliably detect when the contents of a
* journal can be discarded and when we can skip the call to
* glClear because it matches the previous clear request.
*/
/* Note: we don't check for the stencil buffer being cleared here
* since there isn't any public cogl api to manipulate the stencil
* buffer.
*
* Note: we check for an exact clip match here because
* 1) a smaller clip could mean existing journal entries may
* need to contribute to regions outside the new clear-clip
* 2) a larger clip would mean we need to issue a real
* glClear and we only care about cases avoiding a
* glClear.
*
* Note: Comparing without an epsilon is considered
* appropriate here.
*/
if (buffers & COGL_BUFFER_BIT_COLOR &&
buffers & COGL_BUFFER_BIT_DEPTH &&
!framebuffer->clear_clip_dirty &&
framebuffer->clear_color_red == red &&
framebuffer->clear_color_green == green &&
framebuffer->clear_color_blue == blue &&
framebuffer->clear_color_alpha == alpha &&
scissor_x0 == framebuffer->clear_clip_x0 &&
scissor_y0 == framebuffer->clear_clip_y0 &&
scissor_x1 == framebuffer->clear_clip_x1 &&
scissor_y1 == framebuffer->clear_clip_y1)
{
/* NB: We only have to consider the clip state of journal
* entries if the current clear is clipped since otherwise we
* know every pixel of the framebuffer is affected by the clear
* and so all journal entries become redundant and can simply be
* discarded.
*/
if (clip_stack)
{
/*
* Note: the function for checking the journal entries is
* quite strict. It avoids detailed checking of all entry
* clip_stacks by only checking the details of the first
* entry and then it only verifies that the remaining
* entries share the same clip_stack ancestry. This means
* it's possible for some false negatives here but that will
* just result in us falling back to a real clear.
*/
if (_cogl_journal_all_entries_within_bounds (framebuffer->journal,
scissor_x0, scissor_y0,
scissor_x1, scissor_y1))
{
_cogl_journal_discard (framebuffer->journal);
goto cleared;
}
}
else
{
_cogl_journal_discard (framebuffer->journal);
goto cleared;
}
}
COGL_NOTE (DRAW, "Clear begin");
_cogl_framebuffer_flush_journal (framebuffer);
/* XXX: ONGOING BUG: Intel viewport scissor
*
* The semantics of cogl_framebuffer_clear() are that it should not
* be affected by the current viewport and so if we are currently
* applying a workaround for viewport scissoring we need to
* temporarily disable the workaround before clearing so any
* special scissoring for the workaround will be removed first.
*
* Note: we only need to disable the workaround if the current
* viewport doesn't match the framebuffer's size since otherwise
* the workaround wont affect clearing anyway.
*/
if (ctx->needs_viewport_scissor_workaround &&
(framebuffer->viewport_x != 0 ||
framebuffer->viewport_y != 0 ||
framebuffer->viewport_width != framebuffer->width ||
framebuffer->viewport_height != framebuffer->height))
{
saved_viewport_scissor_workaround = TRUE;
ctx->needs_viewport_scissor_workaround = FALSE;
ctx->current_draw_buffer_changes |= COGL_FRAMEBUFFER_STATE_CLIP;
}
else
saved_viewport_scissor_workaround = FALSE;
/* NB: _cogl_framebuffer_flush_state may disrupt various state (such
* as the pipeline state) when flushing the clip stack, so should
* always be done first when preparing to draw. */
_cogl_framebuffer_flush_state (framebuffer, framebuffer,
COGL_FRAMEBUFFER_STATE_ALL);
_cogl_framebuffer_clear_without_flush4f (framebuffer, buffers,
red, green, blue, alpha);
/* XXX: ONGOING BUG: Intel viewport scissor
*
* See comment about temporarily disabling this workaround above
*/
if (saved_viewport_scissor_workaround)
{
ctx->needs_viewport_scissor_workaround = TRUE;
ctx->current_draw_buffer_changes |= COGL_FRAMEBUFFER_STATE_CLIP;
}
/* This is a debugging variable used to visually display the quad
* batches from the journal. It is reset here to increase the
* chances of getting the same colours for each frame during an
* animation */
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_RECTANGLES)) &&
buffers & COGL_BUFFER_BIT_COLOR)
{
framebuffer->context->journal_rectangles_color = 1;
}
COGL_NOTE (DRAW, "Clear end");
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
cleared:
if (buffers & COGL_BUFFER_BIT_COLOR && buffers & COGL_BUFFER_BIT_DEPTH)
{
/* For our fast-path for reading back a single pixel of simple
* scenes where the whole frame is in the journal we need to
* track the cleared color of the framebuffer in case the point
* read doesn't intersect any of the journal rectangles. */
framebuffer->clear_clip_dirty = FALSE;
framebuffer->clear_color_red = red;
framebuffer->clear_color_green = green;
framebuffer->clear_color_blue = blue;
framebuffer->clear_color_alpha = alpha;
/* NB: A clear may be scissored so we need to track the extents
* that the clear is applicable too... */
if (clip_stack)
{
_cogl_clip_stack_get_bounds (clip_stack,
&framebuffer->clear_clip_x0,
&framebuffer->clear_clip_y0,
&framebuffer->clear_clip_x1,
&framebuffer->clear_clip_y1);
}
else
{
/* FIXME: set degenerate clip */
}
}
else
_cogl_framebuffer_dirty (framebuffer);
}
/* Note: the 'buffers' and 'color' arguments were switched around on
* purpose compared to the original cogl_clear API since it was odd
* that you would be expected to specify a color before even
* necessarily choosing to clear the color buffer.
*/
void
cogl_framebuffer_clear (CoglFramebuffer *framebuffer,
unsigned long buffers,
const CoglColor *color)
{
cogl_framebuffer_clear4f (framebuffer, buffers,
cogl_color_get_red_float (color),
cogl_color_get_green_float (color),
cogl_color_get_blue_float (color),
cogl_color_get_alpha_float (color));
}
int
cogl_framebuffer_get_width (CoglFramebuffer *framebuffer)
{
return framebuffer->width;
}
int
cogl_framebuffer_get_height (CoglFramebuffer *framebuffer)
{
return framebuffer->height;
}
CoglClipState *
_cogl_framebuffer_get_clip_state (CoglFramebuffer *framebuffer)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
return &framebuffer->clip_state;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
CoglClipStack *
_cogl_framebuffer_get_clip_stack (CoglFramebuffer *framebuffer)
{
CoglClipState *clip_state = _cogl_framebuffer_get_clip_state (framebuffer);
return _cogl_clip_state_get_stack (clip_state);
}
void
_cogl_framebuffer_set_clip_stack (CoglFramebuffer *framebuffer,
CoglClipStack *stack)
{
CoglClipState *clip_state = _cogl_framebuffer_get_clip_state (framebuffer);
_cogl_clip_state_set_stack (clip_state, stack);
}
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
void
cogl_framebuffer_set_viewport (CoglFramebuffer *framebuffer,
float x,
float y,
float width,
float height)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
CoglContext *context = framebuffer->context;
_COGL_RETURN_IF_FAIL (width > 0 && height > 0);
if (framebuffer->viewport_x == x &&
framebuffer->viewport_y == y &&
framebuffer->viewport_width == width &&
framebuffer->viewport_height == height)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
return;
_cogl_framebuffer_flush_journal (framebuffer);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
framebuffer->viewport_x = x;
framebuffer->viewport_y = y;
framebuffer->viewport_width = width;
framebuffer->viewport_height = height;
framebuffer->viewport_age++;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
if (context->current_draw_buffer == framebuffer)
{
context->current_draw_buffer_changes |= COGL_FRAMEBUFFER_STATE_VIEWPORT;
if (context->needs_viewport_scissor_workaround)
context->current_draw_buffer_changes |= COGL_FRAMEBUFFER_STATE_CLIP;
}
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
float
cogl_framebuffer_get_viewport_x (CoglFramebuffer *framebuffer)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
return framebuffer->viewport_x;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
float
cogl_framebuffer_get_viewport_y (CoglFramebuffer *framebuffer)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
return framebuffer->viewport_y;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
float
cogl_framebuffer_get_viewport_width (CoglFramebuffer *framebuffer)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
return framebuffer->viewport_width;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
float
cogl_framebuffer_get_viewport_height (CoglFramebuffer *framebuffer)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
return framebuffer->viewport_height;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
void
cogl_framebuffer_get_viewport4fv (CoglFramebuffer *framebuffer,
float *viewport)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
viewport[0] = framebuffer->viewport_x;
viewport[1] = framebuffer->viewport_y;
viewport[2] = framebuffer->viewport_width;
viewport[3] = framebuffer->viewport_height;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
CoglMatrixStack *
_cogl_framebuffer_get_modelview_stack (CoglFramebuffer *framebuffer)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
return framebuffer->modelview_stack;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
CoglMatrixStack *
_cogl_framebuffer_get_projection_stack (CoglFramebuffer *framebuffer)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
return framebuffer->projection_stack;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
void
_cogl_framebuffer_add_dependency (CoglFramebuffer *framebuffer,
CoglFramebuffer *dependency)
{
GList *l;
for (l = framebuffer->deps; l; l = l->next)
{
CoglFramebuffer *existing_dep = l->data;
if (existing_dep == dependency)
return;
}
/* TODO: generalize the primed-array type structure we e.g. use for
* cogl_object_set_user_data or for pipeline children as a way to
* avoid quite a lot of mid-scene micro allocations here... */
framebuffer->deps =
g_list_prepend (framebuffer->deps, cogl_object_ref (dependency));
}
void
_cogl_framebuffer_remove_all_dependencies (CoglFramebuffer *framebuffer)
{
GList *l;
for (l = framebuffer->deps; l; l = l->next)
cogl_object_unref (l->data);
g_list_free (framebuffer->deps);
framebuffer->deps = NULL;
}
void
_cogl_framebuffer_flush_journal (CoglFramebuffer *framebuffer)
{
_cogl_journal_flush (framebuffer->journal);
}
void
_cogl_framebuffer_flush_dependency_journals (CoglFramebuffer *framebuffer)
{
GList *l;
for (l = framebuffer->deps; l; l = l->next)
_cogl_framebuffer_flush_journal (l->data);
_cogl_framebuffer_remove_all_dependencies (framebuffer);
}
CoglOffscreen *
Add a strong CoglTexture type to replace CoglHandle As part of the on going, incremental effort to purge the non type safe CoglHandle type from the Cogl API this patch tackles most of the CoglHandle uses relating to textures. We'd postponed making this change for quite a while because we wanted to have a clearer understanding of how we wanted to evolve the texture APIs towards Cogl 2.0 before exposing type safety here which would be difficult to change later since it would imply breaking APIs. The basic idea that we are steering towards now is that CoglTexture can be considered to be the most primitive interface we have for any object representing a texture. The texture interface would provide roughly these methods: cogl_texture_get_width cogl_texture_get_height cogl_texture_can_repeat cogl_texture_can_mipmap cogl_texture_generate_mipmap; cogl_texture_get_format cogl_texture_set_region cogl_texture_get_region Besides the texture interface we will then start to expose types corresponding to specific texture types: CoglTexture2D, CoglTexture3D, CoglTexture2DSliced, CoglSubTexture, CoglAtlasTexture and CoglTexturePixmapX11. We will then also expose an interface for the high-level texture types we have (such as CoglTexture2DSlice, CoglSubTexture and CoglAtlasTexture) called CoglMetaTexture. CoglMetaTexture is an additional interface that lets you iterate a virtual region of a meta texture and get mappings of primitive textures to sub-regions of that virtual region. Internally we already have this kind of abstraction for dealing with sliced texture, sub-textures and atlas textures in a consistent way, so this will just make that abstraction public. The aim here is to clarify that there is a difference between primitive textures (CoglTexture2D/3D) and some of the other high-level textures, and also enable developers to implement primitives that can support meta textures since they can only be used with the cogl_rectangle API currently. The thing that's not so clean-cut with this are the texture constructors we have currently; such as cogl_texture_new_from_file which no longer make sense when CoglTexture is considered to be an interface. These will basically just become convenient factory functions and it's just a bit unusual that they are within the cogl_texture namespace. It's worth noting here that all the texture type APIs will also have their own type specific constructors so these functions will only be used for the convenience of being able to create a texture without really wanting to know the details of what type of texture you need. Longer term for 2.0 we may come up with replacement names for these factory functions or the other thing we are considering is designing some asynchronous factory functions instead since it's so often detrimental to application performance to be blocked waiting for a texture to be uploaded to the GPU. Reviewed-by: Neil Roberts <neil@linux.intel.com>
2011-08-24 16:30:34 -04:00
_cogl_offscreen_new_to_texture_full (CoglTexture *texture,
CoglOffscreenFlags create_flags,
int level)
{
CoglContext *ctx = texture->context;
CoglOffscreen *offscreen;
CoglFramebuffer *fb;
int level_width;
int level_height;
CoglOffscreen *ret;
_COGL_RETURN_VAL_IF_FAIL (cogl_is_texture (texture), NULL);
_COGL_RETURN_VAL_IF_FAIL (level < _cogl_texture_get_n_levels (texture),
NULL);
_cogl_texture_get_level_size (texture,
level,
&level_width,
&level_height,
NULL);
offscreen = g_new0 (CoglOffscreen, 1);
offscreen->texture = cogl_object_ref (texture);
offscreen->texture_level = level;
offscreen->texture_level_width = level_width;
offscreen->texture_level_height = level_height;
offscreen->create_flags = create_flags;
fb = COGL_FRAMEBUFFER (offscreen);
_cogl_framebuffer_init (fb,
ctx,
COGL_FRAMEBUFFER_TYPE_OFFSCREEN,
cogl_texture_get_format (texture),
level_width,
level_height);
ret = _cogl_offscreen_object_new (offscreen);
_cogl_texture_associate_framebuffer (texture, fb);
return ret;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
CoglOffscreen *
Add a strong CoglTexture type to replace CoglHandle As part of the on going, incremental effort to purge the non type safe CoglHandle type from the Cogl API this patch tackles most of the CoglHandle uses relating to textures. We'd postponed making this change for quite a while because we wanted to have a clearer understanding of how we wanted to evolve the texture APIs towards Cogl 2.0 before exposing type safety here which would be difficult to change later since it would imply breaking APIs. The basic idea that we are steering towards now is that CoglTexture can be considered to be the most primitive interface we have for any object representing a texture. The texture interface would provide roughly these methods: cogl_texture_get_width cogl_texture_get_height cogl_texture_can_repeat cogl_texture_can_mipmap cogl_texture_generate_mipmap; cogl_texture_get_format cogl_texture_set_region cogl_texture_get_region Besides the texture interface we will then start to expose types corresponding to specific texture types: CoglTexture2D, CoglTexture3D, CoglTexture2DSliced, CoglSubTexture, CoglAtlasTexture and CoglTexturePixmapX11. We will then also expose an interface for the high-level texture types we have (such as CoglTexture2DSlice, CoglSubTexture and CoglAtlasTexture) called CoglMetaTexture. CoglMetaTexture is an additional interface that lets you iterate a virtual region of a meta texture and get mappings of primitive textures to sub-regions of that virtual region. Internally we already have this kind of abstraction for dealing with sliced texture, sub-textures and atlas textures in a consistent way, so this will just make that abstraction public. The aim here is to clarify that there is a difference between primitive textures (CoglTexture2D/3D) and some of the other high-level textures, and also enable developers to implement primitives that can support meta textures since they can only be used with the cogl_rectangle API currently. The thing that's not so clean-cut with this are the texture constructors we have currently; such as cogl_texture_new_from_file which no longer make sense when CoglTexture is considered to be an interface. These will basically just become convenient factory functions and it's just a bit unusual that they are within the cogl_texture namespace. It's worth noting here that all the texture type APIs will also have their own type specific constructors so these functions will only be used for the convenience of being able to create a texture without really wanting to know the details of what type of texture you need. Longer term for 2.0 we may come up with replacement names for these factory functions or the other thing we are considering is designing some asynchronous factory functions instead since it's so often detrimental to application performance to be blocked waiting for a texture to be uploaded to the GPU. Reviewed-by: Neil Roberts <neil@linux.intel.com>
2011-08-24 16:30:34 -04:00
cogl_offscreen_new_to_texture (CoglTexture *texture)
{
Add a strong CoglTexture type to replace CoglHandle As part of the on going, incremental effort to purge the non type safe CoglHandle type from the Cogl API this patch tackles most of the CoglHandle uses relating to textures. We'd postponed making this change for quite a while because we wanted to have a clearer understanding of how we wanted to evolve the texture APIs towards Cogl 2.0 before exposing type safety here which would be difficult to change later since it would imply breaking APIs. The basic idea that we are steering towards now is that CoglTexture can be considered to be the most primitive interface we have for any object representing a texture. The texture interface would provide roughly these methods: cogl_texture_get_width cogl_texture_get_height cogl_texture_can_repeat cogl_texture_can_mipmap cogl_texture_generate_mipmap; cogl_texture_get_format cogl_texture_set_region cogl_texture_get_region Besides the texture interface we will then start to expose types corresponding to specific texture types: CoglTexture2D, CoglTexture3D, CoglTexture2DSliced, CoglSubTexture, CoglAtlasTexture and CoglTexturePixmapX11. We will then also expose an interface for the high-level texture types we have (such as CoglTexture2DSlice, CoglSubTexture and CoglAtlasTexture) called CoglMetaTexture. CoglMetaTexture is an additional interface that lets you iterate a virtual region of a meta texture and get mappings of primitive textures to sub-regions of that virtual region. Internally we already have this kind of abstraction for dealing with sliced texture, sub-textures and atlas textures in a consistent way, so this will just make that abstraction public. The aim here is to clarify that there is a difference between primitive textures (CoglTexture2D/3D) and some of the other high-level textures, and also enable developers to implement primitives that can support meta textures since they can only be used with the cogl_rectangle API currently. The thing that's not so clean-cut with this are the texture constructors we have currently; such as cogl_texture_new_from_file which no longer make sense when CoglTexture is considered to be an interface. These will basically just become convenient factory functions and it's just a bit unusual that they are within the cogl_texture namespace. It's worth noting here that all the texture type APIs will also have their own type specific constructors so these functions will only be used for the convenience of being able to create a texture without really wanting to know the details of what type of texture you need. Longer term for 2.0 we may come up with replacement names for these factory functions or the other thing we are considering is designing some asynchronous factory functions instead since it's so often detrimental to application performance to be blocked waiting for a texture to be uploaded to the GPU. Reviewed-by: Neil Roberts <neil@linux.intel.com>
2011-08-24 16:30:34 -04:00
return _cogl_offscreen_new_to_texture_full (texture, 0, 0);
}
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
static void
_cogl_offscreen_free (CoglOffscreen *offscreen)
{
CoglFramebuffer *framebuffer = COGL_FRAMEBUFFER (offscreen);
CoglContext *ctx = framebuffer->context;
ctx->driver_vtable->offscreen_free (offscreen);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
/* Chain up to parent */
_cogl_framebuffer_free (framebuffer);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
if (offscreen->texture != NULL)
Add a strong CoglTexture type to replace CoglHandle As part of the on going, incremental effort to purge the non type safe CoglHandle type from the Cogl API this patch tackles most of the CoglHandle uses relating to textures. We'd postponed making this change for quite a while because we wanted to have a clearer understanding of how we wanted to evolve the texture APIs towards Cogl 2.0 before exposing type safety here which would be difficult to change later since it would imply breaking APIs. The basic idea that we are steering towards now is that CoglTexture can be considered to be the most primitive interface we have for any object representing a texture. The texture interface would provide roughly these methods: cogl_texture_get_width cogl_texture_get_height cogl_texture_can_repeat cogl_texture_can_mipmap cogl_texture_generate_mipmap; cogl_texture_get_format cogl_texture_set_region cogl_texture_get_region Besides the texture interface we will then start to expose types corresponding to specific texture types: CoglTexture2D, CoglTexture3D, CoglTexture2DSliced, CoglSubTexture, CoglAtlasTexture and CoglTexturePixmapX11. We will then also expose an interface for the high-level texture types we have (such as CoglTexture2DSlice, CoglSubTexture and CoglAtlasTexture) called CoglMetaTexture. CoglMetaTexture is an additional interface that lets you iterate a virtual region of a meta texture and get mappings of primitive textures to sub-regions of that virtual region. Internally we already have this kind of abstraction for dealing with sliced texture, sub-textures and atlas textures in a consistent way, so this will just make that abstraction public. The aim here is to clarify that there is a difference between primitive textures (CoglTexture2D/3D) and some of the other high-level textures, and also enable developers to implement primitives that can support meta textures since they can only be used with the cogl_rectangle API currently. The thing that's not so clean-cut with this are the texture constructors we have currently; such as cogl_texture_new_from_file which no longer make sense when CoglTexture is considered to be an interface. These will basically just become convenient factory functions and it's just a bit unusual that they are within the cogl_texture namespace. It's worth noting here that all the texture type APIs will also have their own type specific constructors so these functions will only be used for the convenience of being able to create a texture without really wanting to know the details of what type of texture you need. Longer term for 2.0 we may come up with replacement names for these factory functions or the other thing we are considering is designing some asynchronous factory functions instead since it's so often detrimental to application performance to be blocked waiting for a texture to be uploaded to the GPU. Reviewed-by: Neil Roberts <neil@linux.intel.com>
2011-08-24 16:30:34 -04:00
cogl_object_unref (offscreen->texture);
if (offscreen->depth_texture != NULL)
cogl_object_unref (offscreen->depth_texture);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
g_free (offscreen);
}
CoglBool
cogl_framebuffer_allocate (CoglFramebuffer *framebuffer,
Adds CoglError api Although we use GLib internally in Cogl we would rather not leak GLib api through Cogl's own api, except through explicitly namespaced cogl_glib_ / cogl_gtype_ feature apis. One of the benefits we see to not leaking GLib through Cogl's public API is that documentation for Cogl won't need to first introduce the Glib API to newcomers, thus hopefully lowering the barrier to learning Cogl. This patch provides a Cogl specific typedef for reporting runtime errors which by no coincidence matches the typedef for GError exactly. If Cogl is built with --enable-glib (default) then developers can even safely assume that a CoglError is a GError under the hood. This patch also enforces a consistent policy for when NULL is passed as an error argument and an error is thrown. In this case we log the error and abort the application, instead of silently ignoring it. In common cases where nothing has been implemented to handle a particular error and/or where applications are just printing the error and aborting themselves then this saves some typing. This also seems more consistent with language based exceptions which usually cause a program to abort if they are not explicitly caught (which passing a non-NULL error signifies in this case) Since this policy for NULL error pointers is stricter than the standard GError convention, there is a clear note in the documentation to warn developers that are used to using the GError api. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit b068d5ea09ab32c37e8c965fc8582c85d1b2db46) Note: Since we can't change the Cogl 1.x api the patch was changed to not rename _error_quark() functions to be _error_domain() functions and although it's a bit ugly, instead of providing our own CoglError type that's compatible with GError we simply #define CoglError to GError unless Cogl is built with glib disabled. Note: this patch does technically introduce an API break since it drops the cogl_error_get_type() symbol generated by glib-mkenum (Since the CoglError enum was replaced by a CoglSystemError enum) but for now we are assuming that this will not affect anyone currently using the Cogl API. If this does turn out to be a problem in practice then we would be able to fix this my manually copying an implementation of cogl_error_get_type() generated by glib-mkenum into a compatibility source file and we could also define the original COGL_ERROR_ enums for compatibility too. Note: another minor concern with cherry-picking this patch to the 1.14 branch is that an api scanner would be lead to believe that some APIs have changed, and for example the gobject-introspection parser which understands the semantics of GError will not understand the semantics of CoglError. We expect most people that have tried to use gobject-introspection with Cogl already understand though that it is not well suited to generating bindings of the Cogl api anyway and we aren't aware or anyone depending on such bindings for apis involving GErrors. (GnomeShell only makes very-very minimal use of Cogl via the gjs bindings for the cogl_rectangle and cogl_color apis.) The main reason we have cherry-picked this patch to the 1.14 branch even given the above concerns is that without it it would become very awkward for us to cherry-pick other beneficial patches from master.
2012-08-31 14:28:27 -04:00
CoglError **error)
{
CoglOnscreen *onscreen = COGL_ONSCREEN (framebuffer);
const CoglWinsysVtable *winsys = _cogl_framebuffer_get_winsys (framebuffer);
if (framebuffer->allocated)
return TRUE;
if (framebuffer->type == COGL_FRAMEBUFFER_TYPE_ONSCREEN)
{
if (framebuffer->config.depth_texture_enabled)
{
Adds CoglError api Although we use GLib internally in Cogl we would rather not leak GLib api through Cogl's own api, except through explicitly namespaced cogl_glib_ / cogl_gtype_ feature apis. One of the benefits we see to not leaking GLib through Cogl's public API is that documentation for Cogl won't need to first introduce the Glib API to newcomers, thus hopefully lowering the barrier to learning Cogl. This patch provides a Cogl specific typedef for reporting runtime errors which by no coincidence matches the typedef for GError exactly. If Cogl is built with --enable-glib (default) then developers can even safely assume that a CoglError is a GError under the hood. This patch also enforces a consistent policy for when NULL is passed as an error argument and an error is thrown. In this case we log the error and abort the application, instead of silently ignoring it. In common cases where nothing has been implemented to handle a particular error and/or where applications are just printing the error and aborting themselves then this saves some typing. This also seems more consistent with language based exceptions which usually cause a program to abort if they are not explicitly caught (which passing a non-NULL error signifies in this case) Since this policy for NULL error pointers is stricter than the standard GError convention, there is a clear note in the documentation to warn developers that are used to using the GError api. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit b068d5ea09ab32c37e8c965fc8582c85d1b2db46) Note: Since we can't change the Cogl 1.x api the patch was changed to not rename _error_quark() functions to be _error_domain() functions and although it's a bit ugly, instead of providing our own CoglError type that's compatible with GError we simply #define CoglError to GError unless Cogl is built with glib disabled. Note: this patch does technically introduce an API break since it drops the cogl_error_get_type() symbol generated by glib-mkenum (Since the CoglError enum was replaced by a CoglSystemError enum) but for now we are assuming that this will not affect anyone currently using the Cogl API. If this does turn out to be a problem in practice then we would be able to fix this my manually copying an implementation of cogl_error_get_type() generated by glib-mkenum into a compatibility source file and we could also define the original COGL_ERROR_ enums for compatibility too. Note: another minor concern with cherry-picking this patch to the 1.14 branch is that an api scanner would be lead to believe that some APIs have changed, and for example the gobject-introspection parser which understands the semantics of GError will not understand the semantics of CoglError. We expect most people that have tried to use gobject-introspection with Cogl already understand though that it is not well suited to generating bindings of the Cogl api anyway and we aren't aware or anyone depending on such bindings for apis involving GErrors. (GnomeShell only makes very-very minimal use of Cogl via the gjs bindings for the cogl_rectangle and cogl_color apis.) The main reason we have cherry-picked this patch to the 1.14 branch even given the above concerns is that without it it would become very awkward for us to cherry-pick other beneficial patches from master.
2012-08-31 14:28:27 -04:00
_cogl_set_error (error, COGL_FRAMEBUFFER_ERROR,
COGL_FRAMEBUFFER_ERROR_ALLOCATE,
"Can't allocate onscreen framebuffer with a "
"texture based depth buffer");
return FALSE;
}
if (!winsys->onscreen_init (onscreen, error))
return FALSE;
}
else
{
CoglContext *ctx = framebuffer->context;
CoglOffscreen *offscreen = COGL_OFFSCREEN (framebuffer);
if (!cogl_has_feature (ctx, COGL_FEATURE_ID_OFFSCREEN))
{
_cogl_set_error (error, COGL_SYSTEM_ERROR,
COGL_SYSTEM_ERROR_UNSUPPORTED,
"Offscreen framebuffers not supported by system");
return FALSE;
}
if (cogl_texture_is_sliced (offscreen->texture))
{
_cogl_set_error (error, COGL_SYSTEM_ERROR,
COGL_SYSTEM_ERROR_UNSUPPORTED,
"Can't create offscreen framebuffer from "
"sliced texture");
return FALSE;
}
if (!cogl_texture_allocate (offscreen->texture, error))
return FALSE;
if (!ctx->driver_vtable->offscreen_allocate (offscreen, error))
return FALSE;
}
framebuffer->allocated = TRUE;
return TRUE;
}
static CoglFramebufferStackEntry *
create_stack_entry (CoglFramebuffer *draw_buffer,
CoglFramebuffer *read_buffer)
{
CoglFramebufferStackEntry *entry = g_slice_new (CoglFramebufferStackEntry);
entry->draw_buffer = draw_buffer;
entry->read_buffer = read_buffer;
return entry;
}
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
GSList *
_cogl_create_framebuffer_stack (void)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
CoglFramebufferStackEntry *entry;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
GSList *stack = NULL;
entry = create_stack_entry (NULL, NULL);
return g_slist_prepend (stack, entry);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
void
_cogl_free_framebuffer_stack (GSList *stack)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
GSList *l;
for (l = stack; l != NULL; l = l->next)
{
CoglFramebufferStackEntry *entry = l->data;
if (entry->draw_buffer)
cogl_object_unref (entry->draw_buffer);
if (entry->read_buffer)
cogl_object_unref (entry->draw_buffer);
g_slice_free (CoglFramebufferStackEntry, entry);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
g_slist_free (stack);
}
static void
notify_buffers_changed (CoglFramebuffer *old_draw_buffer,
CoglFramebuffer *new_draw_buffer,
CoglFramebuffer *old_read_buffer,
CoglFramebuffer *new_read_buffer)
{
/* XXX: To support the deprecated cogl_set_draw_buffer API we keep
* track of the last onscreen framebuffer that was set so that it
* can be restored if the COGL_WINDOW_BUFFER enum is used. A
* reference isn't taken to the framebuffer because otherwise we
* would have a circular reference between the context and the
* framebuffer. Instead the pointer is set to NULL in
* _cogl_onscreen_free as a kind of a cheap weak reference */
if (new_draw_buffer &&
new_draw_buffer->type == COGL_FRAMEBUFFER_TYPE_ONSCREEN)
new_draw_buffer->context->window_buffer = new_draw_buffer;
}
/* Set the current framebuffer without checking if it's already the
* current framebuffer. This is used by cogl_pop_framebuffer while
* the top of the stack is currently not up to date. */
static void
_cogl_set_framebuffers_real (CoglFramebuffer *draw_buffer,
CoglFramebuffer *read_buffer)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
CoglFramebufferStackEntry *entry;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
_COGL_GET_CONTEXT (ctx, NO_RETVAL);
_COGL_RETURN_IF_FAIL (ctx != NULL);
_COGL_RETURN_IF_FAIL (draw_buffer && read_buffer ?
draw_buffer->context == read_buffer->context : TRUE);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
entry = ctx->framebuffer_stack->data;
notify_buffers_changed (entry->draw_buffer,
draw_buffer,
entry->read_buffer,
read_buffer);
if (draw_buffer)
cogl_object_ref (draw_buffer);
if (entry->draw_buffer)
cogl_object_unref (entry->draw_buffer);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
if (read_buffer)
cogl_object_ref (read_buffer);
if (entry->read_buffer)
cogl_object_unref (entry->read_buffer);
entry->draw_buffer = draw_buffer;
entry->read_buffer = read_buffer;
}
static void
_cogl_set_framebuffers (CoglFramebuffer *draw_buffer,
CoglFramebuffer *read_buffer)
{
CoglFramebuffer *current_draw_buffer;
CoglFramebuffer *current_read_buffer;
framebuffer: flush journal when switching framebuffers There is currently a problem with per-framebuffer journals in that it's possible to create a framebuffer from a texture which then gets rendered too but the framebuffer (and corresponding journal) can be freed before the texture gets used to draw with. Conceptually we want to make sure when freeing a framebuffer that - if it is associated with a texture - we flush the journal as the last thing before really freeing the framebuffer's meta data. Technically though this is awkward to implement since the obvious mechanism for us to be notified about the framebuffer's destruction (by setting some user data internally with a callback) notifies when the framebuffer has a ref-count of 0. This means we'd have to be careful what we do with the framebuffer to consider e.g. recursive destruction; anything that would set more user data on the framebuffer while it is being destroyed and ensuring nothing else gets notified of the framebuffer's destruction before the journal has been flushed. For simplicity, for now, this patch provides another solution which is to flush framebuffer journals whenever we switch away from a given framebuffer via cogl_set_framebuffer or cogl_push/pop_framebuffer. The disadvantage of this approach is that we can't batch all the geometry of a scene that involves intermediate renders to offscreen framebufers. Clutter is doing this more and more with applications that use the ClutterEffect APIs so this is a shame. Hopefully this will only be a stop-gap solution while we consider how to reliably support journal logging across framebuffer changes.
2011-01-21 14:05:23 -05:00
Add -Wmissing-declarations to maintainer flags and fix problems This option to GCC makes it give a warning whenever a global function is defined without a declaration. This should catch cases were we've defined a function but forgot to put it in a header. In that case it is either only used within one file so we should make it static or we should declare it in a header. The following changes where made to fix problems: • Some functions were made static • cogl-path.h (the one containing the 1.0 API) was split into two files, one defining the functions and one defining the enums so that cogl-path.c can include the enum and function declarations from the 2.0 API as well as the function declarations from the 1.0 API. • cogl2-clip-state has been removed. This only had one experimental function called cogl_clip_push_from_path but as this is unstable we might as well remove it favour of the equivalent cogl_framebuffer_* API. • The GLX, SDL and WGL winsys's now have a private header to define their get_vtable function instead of directly declaring in the C file where it is called. • All places that were calling COGL_OBJECT_DEFINE need to have the cogl_is_whatever function declared so these have been added either as a public function or in a private header. • Some files that were not including the header containing their function declarations have been fixed to do so. • Any unused error quark functions have been removed. If we later want them we should add them back one by one and add a declaration for them in a header. • _cogl_is_framebuffer has been renamed to cogl_is_framebuffer and made a public function with a declaration in cogl-framebuffer.h • Similarly for CoglOnscreen. • cogl_vdraw_indexed_attributes is called cogl_framebuffer_vdraw_indexed_attributes in the header. The definition has been changed to match the header. • cogl_index_buffer_allocate has been removed. This had no declaration and I'm not sure what it's supposed to do. • CoglJournal has been changed to use the internal CoglObject macro so that it won't define an exported cogl_is_journal symbol. • The _cogl_blah_pointer_from_handle functions have been removed. CoglHandle isn't used much anymore anyway and in the few places where it is used I think it's safe to just use the implicit cast from void* to the right type. • The test-utils.h header for the conformance tests explicitly disables the -Wmissing-declaration option using a pragma because all of the tests declare their main function without a header. Any mistakes relating to missing declarations aren't really important for the tests. • cogl_quaternion_init_from_quaternion and init_from_matrix have been given declarations in cogl-quaternion.h Reviewed-by: Robert Bragg <robert@linux.intel.com>
2012-03-06 13:21:28 -05:00
_COGL_RETURN_IF_FAIL (cogl_is_framebuffer (draw_buffer));
_COGL_RETURN_IF_FAIL (cogl_is_framebuffer (read_buffer));
current_draw_buffer = cogl_get_draw_framebuffer ();
current_read_buffer = _cogl_get_read_framebuffer ();
if (current_draw_buffer != draw_buffer ||
current_read_buffer != read_buffer)
_cogl_set_framebuffers_real (draw_buffer, read_buffer);
}
void
cogl_set_framebuffer (CoglFramebuffer *framebuffer)
{
_cogl_set_framebuffers (framebuffer, framebuffer);
}
/* XXX: deprecated API */
void
cogl_set_draw_buffer (CoglBufferTarget target, CoglHandle handle)
{
_COGL_GET_CONTEXT (ctx, NO_RETVAL);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
if (target == COGL_WINDOW_BUFFER)
handle = ctx->window_buffer;
/* This is deprecated public API. The public API doesn't currently
really expose the concept of separate draw and read buffers so
for the time being this actually just sets both buffers */
cogl_set_framebuffer (handle);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
CoglFramebuffer *
cogl_get_draw_framebuffer (void)
{
CoglFramebufferStackEntry *entry;
_COGL_GET_CONTEXT (ctx, NULL);
g_assert (ctx->framebuffer_stack);
entry = ctx->framebuffer_stack->data;
return entry->draw_buffer;
}
CoglFramebuffer *
_cogl_get_read_framebuffer (void)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
CoglFramebufferStackEntry *entry;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
_COGL_GET_CONTEXT (ctx, NULL);
g_assert (ctx->framebuffer_stack);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
entry = ctx->framebuffer_stack->data;
return entry->read_buffer;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
void
_cogl_push_framebuffers (CoglFramebuffer *draw_buffer,
CoglFramebuffer *read_buffer)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
CoglContext *ctx;
CoglFramebuffer *old_draw_buffer, *old_read_buffer;
Add -Wmissing-declarations to maintainer flags and fix problems This option to GCC makes it give a warning whenever a global function is defined without a declaration. This should catch cases were we've defined a function but forgot to put it in a header. In that case it is either only used within one file so we should make it static or we should declare it in a header. The following changes where made to fix problems: • Some functions were made static • cogl-path.h (the one containing the 1.0 API) was split into two files, one defining the functions and one defining the enums so that cogl-path.c can include the enum and function declarations from the 2.0 API as well as the function declarations from the 1.0 API. • cogl2-clip-state has been removed. This only had one experimental function called cogl_clip_push_from_path but as this is unstable we might as well remove it favour of the equivalent cogl_framebuffer_* API. • The GLX, SDL and WGL winsys's now have a private header to define their get_vtable function instead of directly declaring in the C file where it is called. • All places that were calling COGL_OBJECT_DEFINE need to have the cogl_is_whatever function declared so these have been added either as a public function or in a private header. • Some files that were not including the header containing their function declarations have been fixed to do so. • Any unused error quark functions have been removed. If we later want them we should add them back one by one and add a declaration for them in a header. • _cogl_is_framebuffer has been renamed to cogl_is_framebuffer and made a public function with a declaration in cogl-framebuffer.h • Similarly for CoglOnscreen. • cogl_vdraw_indexed_attributes is called cogl_framebuffer_vdraw_indexed_attributes in the header. The definition has been changed to match the header. • cogl_index_buffer_allocate has been removed. This had no declaration and I'm not sure what it's supposed to do. • CoglJournal has been changed to use the internal CoglObject macro so that it won't define an exported cogl_is_journal symbol. • The _cogl_blah_pointer_from_handle functions have been removed. CoglHandle isn't used much anymore anyway and in the few places where it is used I think it's safe to just use the implicit cast from void* to the right type. • The test-utils.h header for the conformance tests explicitly disables the -Wmissing-declaration option using a pragma because all of the tests declare their main function without a header. Any mistakes relating to missing declarations aren't really important for the tests. • cogl_quaternion_init_from_quaternion and init_from_matrix have been given declarations in cogl-quaternion.h Reviewed-by: Robert Bragg <robert@linux.intel.com>
2012-03-06 13:21:28 -05:00
_COGL_RETURN_IF_FAIL (cogl_is_framebuffer (draw_buffer));
_COGL_RETURN_IF_FAIL (cogl_is_framebuffer (read_buffer));
ctx = draw_buffer->context;
_COGL_RETURN_IF_FAIL (ctx != NULL);
_COGL_RETURN_IF_FAIL (draw_buffer->context == read_buffer->context);
_COGL_RETURN_IF_FAIL (ctx->framebuffer_stack != NULL);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
/* Copy the top of the stack so that when we call cogl_set_framebuffer
it will still know what the old framebuffer was */
old_draw_buffer = cogl_get_draw_framebuffer ();
if (old_draw_buffer)
cogl_object_ref (old_draw_buffer);
old_read_buffer = _cogl_get_read_framebuffer ();
if (old_read_buffer)
cogl_object_ref (old_read_buffer);
ctx->framebuffer_stack =
g_slist_prepend (ctx->framebuffer_stack,
create_stack_entry (old_draw_buffer,
old_read_buffer));
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
_cogl_set_framebuffers (draw_buffer, read_buffer);
}
void
cogl_push_framebuffer (CoglFramebuffer *buffer)
{
_cogl_push_framebuffers (buffer, buffer);
}
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
/* XXX: deprecated API */
void
cogl_push_draw_buffer (void)
{
cogl_push_framebuffer (cogl_get_draw_framebuffer ());
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
void
cogl_pop_framebuffer (void)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
CoglFramebufferStackEntry *to_pop;
CoglFramebufferStackEntry *to_restore;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
_COGL_GET_CONTEXT (ctx, NO_RETVAL);
g_assert (ctx->framebuffer_stack != NULL);
g_assert (ctx->framebuffer_stack->next != NULL);
to_pop = ctx->framebuffer_stack->data;
to_restore = ctx->framebuffer_stack->next->data;
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
if (to_pop->draw_buffer != to_restore->draw_buffer ||
to_pop->read_buffer != to_restore->read_buffer)
notify_buffers_changed (to_pop->draw_buffer,
to_restore->draw_buffer,
to_pop->read_buffer,
to_restore->read_buffer);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
cogl_object_unref (to_pop->draw_buffer);
cogl_object_unref (to_pop->read_buffer);
g_slice_free (CoglFramebufferStackEntry, to_pop);
ctx->framebuffer_stack =
g_slist_delete_link (ctx->framebuffer_stack,
ctx->framebuffer_stack);
}
/* XXX: deprecated API */
void
cogl_pop_draw_buffer (void)
{
cogl_pop_framebuffer ();
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
static unsigned long
_cogl_framebuffer_compare_viewport_state (CoglFramebuffer *a,
CoglFramebuffer *b)
{
if (a->viewport_x != b->viewport_x ||
a->viewport_y != b->viewport_y ||
a->viewport_width != b->viewport_width ||
a->viewport_height != b->viewport_height ||
/* NB: we render upside down to offscreen framebuffers and that
* can affect how we setup the GL viewport... */
a->type != b->type)
{
unsigned long differences = COGL_FRAMEBUFFER_STATE_VIEWPORT;
CoglContext *context = a->context;
/* XXX: ONGOING BUG: Intel viewport scissor
*
* Intel gen6 drivers don't currently correctly handle offset
* viewports, since primitives aren't clipped within the bounds of
* the viewport. To workaround this we push our own clip for the
* viewport that will use scissoring to ensure we clip as expected.
*
* This workaround implies that a change in viewport state is
* effectively also a change in the clipping state.
*
* TODO: file a bug upstream!
*/
if (G_UNLIKELY (context->needs_viewport_scissor_workaround))
differences |= COGL_FRAMEBUFFER_STATE_CLIP;
return differences;
}
else
return 0;
}
static unsigned long
_cogl_framebuffer_compare_clip_state (CoglFramebuffer *a,
CoglFramebuffer *b)
{
if (((a->clip_state.stacks == NULL || b->clip_state.stacks == NULL) &&
a->clip_state.stacks != b->clip_state.stacks)
||
a->clip_state.stacks->data != b->clip_state.stacks->data)
return COGL_FRAMEBUFFER_STATE_CLIP;
else
return 0;
}
static unsigned long
_cogl_framebuffer_compare_dither_state (CoglFramebuffer *a,
CoglFramebuffer *b)
{
return a->dither_enabled != b->dither_enabled ?
COGL_FRAMEBUFFER_STATE_DITHER : 0;
}
static unsigned long
_cogl_framebuffer_compare_modelview_state (CoglFramebuffer *a,
CoglFramebuffer *b)
{
Flush matrices in the progend and flip with a vector Previously flushing the matrices was performed as part of the framebuffer state. When on GLES2 this matrix flushing is actually diverted so that it only keeps a reference to the intended matrix stack. This is necessary because on GLES2 there are no builtin uniforms so it can't actually flush the matrices until the program for the pipeline is generated. When the matrices are flushed it would store the age of modifications on the matrix stack so that it could detect when the matrix hasn't changed and avoid flushing it. This patch changes it so that the pipeline is responsible for flushing the matrices even when we are using the GL builtins. The same mechanism for detecting unmodified matrix stacks is used in all cases. There is a new CoglMatrixStackCache type which is used to store a reference to the intended matrix stack along with its last flushed age. There are now two of these attached to the CoglContext to track the flushed state for the global matrix builtins and also two for each glsl progend program state to track the flushed state for a program. The framebuffer matrix flush now just updates the intended matrix stacks without actually trying to flush. When a vertex snippet is attached to the pipeline, the GLSL vertend will now avoid using the projection matrix to flip the rendering. This is necessary because any vertex snippet may cause the projection matrix not to be used. Instead the flip is done as a forced final step by multiplying cogl_position_out by a vec4 uniform. This uniform is updated as part of the progend pre_paint depending on whether the framebuffer is offscreen or not. Reviewed-by: Robert Bragg <robert@linux.intel.com>
2011-11-29 09:21:07 -05:00
/* We always want to flush the modelview state. All this does is set
the current modelview stack on the context to the framebuffer's
stack. */
return COGL_FRAMEBUFFER_STATE_MODELVIEW;
}
static unsigned long
_cogl_framebuffer_compare_projection_state (CoglFramebuffer *a,
CoglFramebuffer *b)
{
Flush matrices in the progend and flip with a vector Previously flushing the matrices was performed as part of the framebuffer state. When on GLES2 this matrix flushing is actually diverted so that it only keeps a reference to the intended matrix stack. This is necessary because on GLES2 there are no builtin uniforms so it can't actually flush the matrices until the program for the pipeline is generated. When the matrices are flushed it would store the age of modifications on the matrix stack so that it could detect when the matrix hasn't changed and avoid flushing it. This patch changes it so that the pipeline is responsible for flushing the matrices even when we are using the GL builtins. The same mechanism for detecting unmodified matrix stacks is used in all cases. There is a new CoglMatrixStackCache type which is used to store a reference to the intended matrix stack along with its last flushed age. There are now two of these attached to the CoglContext to track the flushed state for the global matrix builtins and also two for each glsl progend program state to track the flushed state for a program. The framebuffer matrix flush now just updates the intended matrix stacks without actually trying to flush. When a vertex snippet is attached to the pipeline, the GLSL vertend will now avoid using the projection matrix to flip the rendering. This is necessary because any vertex snippet may cause the projection matrix not to be used. Instead the flip is done as a forced final step by multiplying cogl_position_out by a vec4 uniform. This uniform is updated as part of the progend pre_paint depending on whether the framebuffer is offscreen or not. Reviewed-by: Robert Bragg <robert@linux.intel.com>
2011-11-29 09:21:07 -05:00
/* We always want to flush the projection state. All this does is
set the current projection stack on the context to the
framebuffer's stack. */
return COGL_FRAMEBUFFER_STATE_PROJECTION;
}
static unsigned long
_cogl_framebuffer_compare_color_mask_state (CoglFramebuffer *a,
CoglFramebuffer *b)
{
if (cogl_framebuffer_get_color_mask (a) !=
cogl_framebuffer_get_color_mask (b))
return COGL_FRAMEBUFFER_STATE_COLOR_MASK;
else
return 0;
}
static unsigned long
_cogl_framebuffer_compare_front_face_winding_state (CoglFramebuffer *a,
CoglFramebuffer *b)
{
if (a->type != b->type)
return COGL_FRAMEBUFFER_STATE_FRONT_FACE_WINDING;
else
return 0;
}
unsigned long
_cogl_framebuffer_compare (CoglFramebuffer *a,
CoglFramebuffer *b,
unsigned long state)
{
unsigned long differences = 0;
int bit;
if (state & COGL_FRAMEBUFFER_STATE_BIND)
{
differences |= COGL_FRAMEBUFFER_STATE_BIND;
state &= ~COGL_FRAMEBUFFER_STATE_BIND;
}
COGL_FLAGS_FOREACH_START (&state, 1, bit)
{
/* XXX: We considered having an array of callbacks for each state index
* that we'd call here but decided that this way the compiler is more
* likely going to be able to in-line the comparison functions and use
* the index to jump straight to the required code. */
switch (bit)
{
case COGL_FRAMEBUFFER_STATE_INDEX_VIEWPORT:
differences |=
_cogl_framebuffer_compare_viewport_state (a, b);
break;
case COGL_FRAMEBUFFER_STATE_INDEX_CLIP:
differences |= _cogl_framebuffer_compare_clip_state (a, b);
break;
case COGL_FRAMEBUFFER_STATE_INDEX_DITHER:
differences |= _cogl_framebuffer_compare_dither_state (a, b);
break;
case COGL_FRAMEBUFFER_STATE_INDEX_MODELVIEW:
differences |=
_cogl_framebuffer_compare_modelview_state (a, b);
break;
case COGL_FRAMEBUFFER_STATE_INDEX_PROJECTION:
differences |=
_cogl_framebuffer_compare_projection_state (a, b);
break;
case COGL_FRAMEBUFFER_STATE_INDEX_COLOR_MASK:
differences |=
_cogl_framebuffer_compare_color_mask_state (a, b);
break;
case COGL_FRAMEBUFFER_STATE_INDEX_FRONT_FACE_WINDING:
differences |=
_cogl_framebuffer_compare_front_face_winding_state (a, b);
break;
default:
g_warn_if_reached ();
}
}
COGL_FLAGS_FOREACH_END;
return differences;
}
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
void
_cogl_framebuffer_flush_state (CoglFramebuffer *draw_buffer,
CoglFramebuffer *read_buffer,
CoglFramebufferState state)
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
{
CoglContext *ctx = draw_buffer->context;
ctx->driver_vtable->framebuffer_flush_state (draw_buffer,
read_buffer,
state);
[draw-buffers] First pass at overhauling Cogl's framebuffer management Cogl's support for offscreen rendering was originally written just to support the clutter_texture_new_from_actor API and due to lack of documentation and several confusing - non orthogonal - side effects of using the API it wasn't really possible to use directly. This commit does a number of things: - It removes {gl,gles}/cogl-fbo.{c,h} and adds shared cogl-draw-buffer.{c,h} files instead which should be easier to maintain. - internally CoglFbo objects are now called CoglDrawBuffers. A CoglDrawBuffer is an abstract base class that is inherited from to implement CoglOnscreen and CoglOffscreen draw buffers. CoglOffscreen draw buffers will initially be used to support the cogl_offscreen_new_to_texture API, and CoglOnscreen draw buffers will start to be used internally to represent windows as we aim to migrate some of Clutter's backend code to Cogl. - It makes draw buffer objects the owners of the following state: - viewport - projection matrix stack - modelview matrix stack - clip state (This means when you switch between draw buffers you will automatically be switching to their associated viewport, matrix and clip state) Aside from hopefully making cogl_offscreen_new_to_texture be more useful short term by having simpler and well defined semantics for cogl_set_draw_buffer, as mentioned above this is the first step for a couple of other things: - Its a step toward moving ownership for windows down from Clutter backends into Cogl, by (internally at least) introducing the CoglOnscreen draw buffer. Note: the plan is that cogl_set_draw_buffer will accept on or offscreen draw buffer handles, and the "target" argument will become redundant since we will instead query the type of the given draw buffer handle. - Because we have a common type for on and offscreen framebuffers we can provide a unified API for framebuffer management. Things like: - blitting between buffers - managing ancillary buffers (e.g. attaching depth and stencil buffers) - size requisition - clearing
2009-09-25 09:34:34 -04:00
}
int
cogl_framebuffer_get_red_bits (CoglFramebuffer *framebuffer)
{
CoglContext *ctx = framebuffer->context;
CoglFramebufferBits bits;
ctx->driver_vtable->framebuffer_query_bits (framebuffer, &bits);
return bits.red;
}
int
cogl_framebuffer_get_green_bits (CoglFramebuffer *framebuffer)
{
CoglContext *ctx = framebuffer->context;
CoglFramebufferBits bits;
ctx->driver_vtable->framebuffer_query_bits (framebuffer, &bits);
return bits.green;
}
int
cogl_framebuffer_get_blue_bits (CoglFramebuffer *framebuffer)
{
CoglContext *ctx = framebuffer->context;
CoglFramebufferBits bits;
ctx->driver_vtable->framebuffer_query_bits (framebuffer, &bits);
return bits.blue;
}
int
cogl_framebuffer_get_alpha_bits (CoglFramebuffer *framebuffer)
{
CoglContext *ctx = framebuffer->context;
CoglFramebufferBits bits;
ctx->driver_vtable->framebuffer_query_bits (framebuffer, &bits);
return bits.alpha;
}
int
cogl_framebuffer_get_depth_bits (CoglFramebuffer *framebuffer)
{
CoglContext *ctx = framebuffer->context;
CoglFramebufferBits bits;
ctx->driver_vtable->framebuffer_query_bits (framebuffer, &bits);
return bits.depth;
}
int
_cogl_framebuffer_get_stencil_bits (CoglFramebuffer *framebuffer)
{
CoglContext *ctx = framebuffer->context;
CoglFramebufferBits bits;
ctx->driver_vtable->framebuffer_query_bits (framebuffer, &bits);
return bits.stencil;
}
CoglColorMask
cogl_framebuffer_get_color_mask (CoglFramebuffer *framebuffer)
{
return framebuffer->color_mask;
}
void
cogl_framebuffer_set_color_mask (CoglFramebuffer *framebuffer,
CoglColorMask color_mask)
{
/* XXX: Currently color mask changes don't go through the journal */
_cogl_framebuffer_flush_journal (framebuffer);
framebuffer->color_mask = color_mask;
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_COLOR_MASK;
}
CoglBool
cogl_framebuffer_get_dither_enabled (CoglFramebuffer *framebuffer)
{
return framebuffer->dither_enabled;
}
void
cogl_framebuffer_set_dither_enabled (CoglFramebuffer *framebuffer,
CoglBool dither_enabled)
{
if (framebuffer->dither_enabled == dither_enabled)
return;
cogl_flush (); /* Currently dithering changes aren't tracked in the journal */
framebuffer->dither_enabled = dither_enabled;
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_DITHER;
}
CoglPixelFormat
cogl_framebuffer_get_color_format (CoglFramebuffer *framebuffer)
{
return framebuffer->format;
}
void
cogl_framebuffer_set_depth_texture_enabled (CoglFramebuffer *framebuffer,
CoglBool enabled)
{
_COGL_RETURN_IF_FAIL (!framebuffer->allocated);
framebuffer->config.depth_texture_enabled = enabled;
}
CoglBool
cogl_framebuffer_get_depth_texture_enabled (CoglFramebuffer *framebuffer)
{
return framebuffer->config.depth_texture_enabled;
}
CoglTexture *
cogl_framebuffer_get_depth_texture (CoglFramebuffer *framebuffer)
{
/* lazily allocate the framebuffer... */
if (!cogl_framebuffer_allocate (framebuffer, NULL))
return NULL;
_COGL_RETURN_VAL_IF_FAIL (cogl_is_offscreen (framebuffer), NULL);
return COGL_OFFSCREEN(framebuffer)->depth_texture;
}
int
cogl_framebuffer_get_samples_per_pixel (CoglFramebuffer *framebuffer)
{
if (framebuffer->allocated)
return framebuffer->samples_per_pixel;
else
return framebuffer->config.samples_per_pixel;
}
void
cogl_framebuffer_set_samples_per_pixel (CoglFramebuffer *framebuffer,
int samples_per_pixel)
{
_COGL_RETURN_IF_FAIL (!framebuffer->allocated);
framebuffer->config.samples_per_pixel = samples_per_pixel;
}
void
cogl_framebuffer_resolve_samples (CoglFramebuffer *framebuffer)
{
cogl_framebuffer_resolve_samples_region (framebuffer,
0, 0,
framebuffer->width,
framebuffer->height);
/* TODO: Make this happen implicitly when the resolve texture next gets used
* as a source, either via cogl_texture_get_data(), via cogl_read_pixels() or
* if used as a source for rendering. We would also implicitly resolve if
* necessary before freeing a CoglFramebuffer.
*
* This API should still be kept but it is optional, only necessary
* if the user wants to explicitly control when the resolve happens e.g.
* to ensure it's done in advance of it being used as a source.
*
* Every texture should have a CoglFramebuffer *needs_resolve member
* internally. When the texture gets validated before being used as a source
* we should first check the needs_resolve pointer and if set we'll
* automatically call cogl_framebuffer_resolve_samples ().
*
* Calling cogl_framebuffer_resolve_samples() or
* cogl_framebuffer_resolve_samples_region() should reset the textures
* needs_resolve pointer to NULL.
*
* Rendering anything to a framebuffer will cause the corresponding
* texture's ->needs_resolve pointer to be set.
*
* XXX: Note: we only need to address this TODO item when adding support for
* EXT_framebuffer_multisample because currently we only support hardware
* that resolves implicitly anyway.
*/
}
void
cogl_framebuffer_resolve_samples_region (CoglFramebuffer *framebuffer,
int x,
int y,
int width,
int height)
{
/* NOP for now since we don't support EXT_framebuffer_multisample yet which
* requires an explicit resolve. */
}
CoglContext *
cogl_framebuffer_get_context (CoglFramebuffer *framebuffer)
{
_COGL_RETURN_VAL_IF_FAIL (framebuffer != NULL, NULL);
return framebuffer->context;
}
static CoglBool
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
_cogl_framebuffer_try_fast_read_pixel (CoglFramebuffer *framebuffer,
int x,
int y,
CoglReadPixelsFlags source,
CoglBitmap *bitmap)
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
{
CoglBool found_intersection;
CoglPixelFormat format;
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_DISABLE_FAST_READ_PIXEL)))
return FALSE;
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
if (source != COGL_READ_PIXELS_COLOR_BUFFER)
return FALSE;
format = cogl_bitmap_get_format (bitmap);
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
if (format != COGL_PIXEL_FORMAT_RGBA_8888_PRE &&
format != COGL_PIXEL_FORMAT_RGBA_8888)
return FALSE;
if (!_cogl_journal_try_read_pixel (framebuffer->journal,
x, y, bitmap,
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
&found_intersection))
return FALSE;
/* If we can't determine the color from the primitives in the
* journal then see if we can use the last recorded clear color
*/
/* If _cogl_journal_try_read_pixel() failed even though there was an
* intersection of the given point with a primitive in the journal
* then we can't fallback to the framebuffer's last clear color...
* */
if (found_intersection)
return TRUE;
/* If the framebuffer has been rendered too since it was last
* cleared then we can't return the last known clear color. */
if (framebuffer->clear_clip_dirty)
return FALSE;
if (x >= framebuffer->clear_clip_x0 &&
x < framebuffer->clear_clip_x1 &&
y >= framebuffer->clear_clip_y0 &&
y < framebuffer->clear_clip_y1)
{
uint8_t *pixel;
Allow propogation of OOM errors to apps This allows apps to catch out-of-memory errors when allocating textures. Textures can be pretty huge at times and so it's quite possible for an application to try and allocate more memory than is available. It's also very possible that the application can take some action in response to reduce memory pressure (such as freeing up texture caches perhaps) so we shouldn't just automatically abort like we do for trivial heap allocations. These public functions now take a CoglError argument so applications can catch out of memory errors: cogl_buffer_map cogl_buffer_map_range cogl_buffer_set_data cogl_framebuffer_read_pixels_into_bitmap cogl_pixel_buffer_new cogl_texture_new_from_data cogl_texture_new_from_bitmap Note: we've been quite conservative with how many apis we let throw OOM CoglErrors since we don't really want to put a burdon on developers to be checking for errors with every cogl api call. So long as there is some lower level api for apps to use that let them catch OOM errors for everything necessary that's enough and we don't have to make more convenient apis more awkward to use. The main focus is on bitmaps and texture allocations since they can be particularly large and prone to failing. A new cogl_attribute_buffer_new_with_size() function has been added in case developers need to catch OOM errors when allocating attribute buffers whereby they can first use _buffer_new_with_size() (which doesn't take a CoglError) followed by cogl_buffer_set_data() which will lazily allocate the buffer storage and report OOM errors. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f7735e141ad537a253b02afa2a8238f96340b978) Note: since we can't break the API for Cogl 1.x then actually the main purpose of cherry picking this patch is to keep in-line with changes on the master branch so that we can easily cherry-pick patches. All the api changes relating stable apis released on the 1.12 branch have been reverted as part of cherry-picking this patch so this most just applies all the internal plumbing changes that enable us to correctly propagate OOM errors.
2012-11-08 12:54:10 -05:00
CoglError *ignore_error = NULL;
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
/* we currently only care about cases where the premultiplied or
* unpremultipled colors are equivalent... */
if (framebuffer->clear_color_alpha != 1.0)
return FALSE;
pixel = _cogl_bitmap_map (bitmap,
COGL_BUFFER_ACCESS_WRITE,
Allow propogation of OOM errors to apps This allows apps to catch out-of-memory errors when allocating textures. Textures can be pretty huge at times and so it's quite possible for an application to try and allocate more memory than is available. It's also very possible that the application can take some action in response to reduce memory pressure (such as freeing up texture caches perhaps) so we shouldn't just automatically abort like we do for trivial heap allocations. These public functions now take a CoglError argument so applications can catch out of memory errors: cogl_buffer_map cogl_buffer_map_range cogl_buffer_set_data cogl_framebuffer_read_pixels_into_bitmap cogl_pixel_buffer_new cogl_texture_new_from_data cogl_texture_new_from_bitmap Note: we've been quite conservative with how many apis we let throw OOM CoglErrors since we don't really want to put a burdon on developers to be checking for errors with every cogl api call. So long as there is some lower level api for apps to use that let them catch OOM errors for everything necessary that's enough and we don't have to make more convenient apis more awkward to use. The main focus is on bitmaps and texture allocations since they can be particularly large and prone to failing. A new cogl_attribute_buffer_new_with_size() function has been added in case developers need to catch OOM errors when allocating attribute buffers whereby they can first use _buffer_new_with_size() (which doesn't take a CoglError) followed by cogl_buffer_set_data() which will lazily allocate the buffer storage and report OOM errors. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f7735e141ad537a253b02afa2a8238f96340b978) Note: since we can't break the API for Cogl 1.x then actually the main purpose of cherry picking this patch is to keep in-line with changes on the master branch so that we can easily cherry-pick patches. All the api changes relating stable apis released on the 1.12 branch have been reverted as part of cherry-picking this patch so this most just applies all the internal plumbing changes that enable us to correctly propagate OOM errors.
2012-11-08 12:54:10 -05:00
COGL_BUFFER_MAP_HINT_DISCARD,
&ignore_error);
if (pixel == NULL)
Allow propogation of OOM errors to apps This allows apps to catch out-of-memory errors when allocating textures. Textures can be pretty huge at times and so it's quite possible for an application to try and allocate more memory than is available. It's also very possible that the application can take some action in response to reduce memory pressure (such as freeing up texture caches perhaps) so we shouldn't just automatically abort like we do for trivial heap allocations. These public functions now take a CoglError argument so applications can catch out of memory errors: cogl_buffer_map cogl_buffer_map_range cogl_buffer_set_data cogl_framebuffer_read_pixels_into_bitmap cogl_pixel_buffer_new cogl_texture_new_from_data cogl_texture_new_from_bitmap Note: we've been quite conservative with how many apis we let throw OOM CoglErrors since we don't really want to put a burdon on developers to be checking for errors with every cogl api call. So long as there is some lower level api for apps to use that let them catch OOM errors for everything necessary that's enough and we don't have to make more convenient apis more awkward to use. The main focus is on bitmaps and texture allocations since they can be particularly large and prone to failing. A new cogl_attribute_buffer_new_with_size() function has been added in case developers need to catch OOM errors when allocating attribute buffers whereby they can first use _buffer_new_with_size() (which doesn't take a CoglError) followed by cogl_buffer_set_data() which will lazily allocate the buffer storage and report OOM errors. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f7735e141ad537a253b02afa2a8238f96340b978) Note: since we can't break the API for Cogl 1.x then actually the main purpose of cherry picking this patch is to keep in-line with changes on the master branch so that we can easily cherry-pick patches. All the api changes relating stable apis released on the 1.12 branch have been reverted as part of cherry-picking this patch so this most just applies all the internal plumbing changes that enable us to correctly propagate OOM errors.
2012-11-08 12:54:10 -05:00
{
cogl_error_free (ignore_error);
return FALSE;
}
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
pixel[0] = framebuffer->clear_color_red * 255.0;
pixel[1] = framebuffer->clear_color_green * 255.0;
pixel[2] = framebuffer->clear_color_blue * 255.0;
pixel[3] = framebuffer->clear_color_alpha * 255.0;
_cogl_bitmap_unmap (bitmap);
cogl: Implements a software only read-pixel fast-path This adds a transparent optimization to cogl_read_pixels for when a single pixel is being read back and it happens that all the geometry of the current frame is still available in the framebuffer's associated journal. The intention is to indirectly optimize Clutter's render based picking mechanism in such a way that the 99% of cases where scenes are comprised of trivial quad primitives that can easily be intersected we can avoid the latency of kicking a GPU render and blocking for the result when we know we can calculate the result manually on the CPU probably faster than we could even kick a render. A nice property of this solution is that it maintains all the flexibility of the render based picking provided by Clutter and it can gracefully fall back to GPU rendering if actors are drawn using anything more complex than a quad for their geometry. It seems worth noting that there is a limitation to the extensibility of this approach in that it can only optimize picking a against geometry that passes through Cogl's journal which isn't something Clutter directly controls. For now though this really doesn't matter since basically all apps should end up hitting this fast-path. The current idea to address this longer term would be a pick2 vfunc for ClutterActor that can support geometry and render based input regions of actors and move this optimization up into Clutter instead. Note: currently we don't have a primitive count threshold to consider that there could be scenes with enough geometry for us to compensate for the cost of kicking a render and determine a result more efficiently by utilizing the GPU. We don't currently expect this to be common though. Note: in the future it could still be interesting to revive something like the wip/async-pbo-picking branch to provide an asynchronous read-pixels based optimization for Clutter picking in cases where more complex input regions that necessitate rendering are in use or if we do add a threshold for rendering as mentioned above.
2011-01-12 17:12:41 -05:00
return TRUE;
}
return FALSE;
}
CoglBool
Allow propogation of OOM errors to apps This allows apps to catch out-of-memory errors when allocating textures. Textures can be pretty huge at times and so it's quite possible for an application to try and allocate more memory than is available. It's also very possible that the application can take some action in response to reduce memory pressure (such as freeing up texture caches perhaps) so we shouldn't just automatically abort like we do for trivial heap allocations. These public functions now take a CoglError argument so applications can catch out of memory errors: cogl_buffer_map cogl_buffer_map_range cogl_buffer_set_data cogl_framebuffer_read_pixels_into_bitmap cogl_pixel_buffer_new cogl_texture_new_from_data cogl_texture_new_from_bitmap Note: we've been quite conservative with how many apis we let throw OOM CoglErrors since we don't really want to put a burdon on developers to be checking for errors with every cogl api call. So long as there is some lower level api for apps to use that let them catch OOM errors for everything necessary that's enough and we don't have to make more convenient apis more awkward to use. The main focus is on bitmaps and texture allocations since they can be particularly large and prone to failing. A new cogl_attribute_buffer_new_with_size() function has been added in case developers need to catch OOM errors when allocating attribute buffers whereby they can first use _buffer_new_with_size() (which doesn't take a CoglError) followed by cogl_buffer_set_data() which will lazily allocate the buffer storage and report OOM errors. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f7735e141ad537a253b02afa2a8238f96340b978) Note: since we can't break the API for Cogl 1.x then actually the main purpose of cherry picking this patch is to keep in-line with changes on the master branch so that we can easily cherry-pick patches. All the api changes relating stable apis released on the 1.12 branch have been reverted as part of cherry-picking this patch so this most just applies all the internal plumbing changes that enable us to correctly propagate OOM errors.
2012-11-08 12:54:10 -05:00
_cogl_framebuffer_read_pixels_into_bitmap (CoglFramebuffer *framebuffer,
int x,
int y,
CoglReadPixelsFlags source,
CoglBitmap *bitmap,
CoglError **error)
{
CoglContext *ctx;
int width;
int height;
_COGL_RETURN_VAL_IF_FAIL (source & COGL_READ_PIXELS_COLOR_BUFFER, FALSE);
Add -Wmissing-declarations to maintainer flags and fix problems This option to GCC makes it give a warning whenever a global function is defined without a declaration. This should catch cases were we've defined a function but forgot to put it in a header. In that case it is either only used within one file so we should make it static or we should declare it in a header. The following changes where made to fix problems: • Some functions were made static • cogl-path.h (the one containing the 1.0 API) was split into two files, one defining the functions and one defining the enums so that cogl-path.c can include the enum and function declarations from the 2.0 API as well as the function declarations from the 1.0 API. • cogl2-clip-state has been removed. This only had one experimental function called cogl_clip_push_from_path but as this is unstable we might as well remove it favour of the equivalent cogl_framebuffer_* API. • The GLX, SDL and WGL winsys's now have a private header to define their get_vtable function instead of directly declaring in the C file where it is called. • All places that were calling COGL_OBJECT_DEFINE need to have the cogl_is_whatever function declared so these have been added either as a public function or in a private header. • Some files that were not including the header containing their function declarations have been fixed to do so. • Any unused error quark functions have been removed. If we later want them we should add them back one by one and add a declaration for them in a header. • _cogl_is_framebuffer has been renamed to cogl_is_framebuffer and made a public function with a declaration in cogl-framebuffer.h • Similarly for CoglOnscreen. • cogl_vdraw_indexed_attributes is called cogl_framebuffer_vdraw_indexed_attributes in the header. The definition has been changed to match the header. • cogl_index_buffer_allocate has been removed. This had no declaration and I'm not sure what it's supposed to do. • CoglJournal has been changed to use the internal CoglObject macro so that it won't define an exported cogl_is_journal symbol. • The _cogl_blah_pointer_from_handle functions have been removed. CoglHandle isn't used much anymore anyway and in the few places where it is used I think it's safe to just use the implicit cast from void* to the right type. • The test-utils.h header for the conformance tests explicitly disables the -Wmissing-declaration option using a pragma because all of the tests declare their main function without a header. Any mistakes relating to missing declarations aren't really important for the tests. • cogl_quaternion_init_from_quaternion and init_from_matrix have been given declarations in cogl-quaternion.h Reviewed-by: Robert Bragg <robert@linux.intel.com>
2012-03-06 13:21:28 -05:00
_COGL_RETURN_VAL_IF_FAIL (cogl_is_framebuffer (framebuffer), FALSE);
Allow propogation of OOM errors to apps This allows apps to catch out-of-memory errors when allocating textures. Textures can be pretty huge at times and so it's quite possible for an application to try and allocate more memory than is available. It's also very possible that the application can take some action in response to reduce memory pressure (such as freeing up texture caches perhaps) so we shouldn't just automatically abort like we do for trivial heap allocations. These public functions now take a CoglError argument so applications can catch out of memory errors: cogl_buffer_map cogl_buffer_map_range cogl_buffer_set_data cogl_framebuffer_read_pixels_into_bitmap cogl_pixel_buffer_new cogl_texture_new_from_data cogl_texture_new_from_bitmap Note: we've been quite conservative with how many apis we let throw OOM CoglErrors since we don't really want to put a burdon on developers to be checking for errors with every cogl api call. So long as there is some lower level api for apps to use that let them catch OOM errors for everything necessary that's enough and we don't have to make more convenient apis more awkward to use. The main focus is on bitmaps and texture allocations since they can be particularly large and prone to failing. A new cogl_attribute_buffer_new_with_size() function has been added in case developers need to catch OOM errors when allocating attribute buffers whereby they can first use _buffer_new_with_size() (which doesn't take a CoglError) followed by cogl_buffer_set_data() which will lazily allocate the buffer storage and report OOM errors. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f7735e141ad537a253b02afa2a8238f96340b978) Note: since we can't break the API for Cogl 1.x then actually the main purpose of cherry picking this patch is to keep in-line with changes on the master branch so that we can easily cherry-pick patches. All the api changes relating stable apis released on the 1.12 branch have been reverted as part of cherry-picking this patch so this most just applies all the internal plumbing changes that enable us to correctly propagate OOM errors.
2012-11-08 12:54:10 -05:00
if (!cogl_framebuffer_allocate (framebuffer, error))
return FALSE;
width = cogl_bitmap_get_width (bitmap);
height = cogl_bitmap_get_height (bitmap);
if (width == 1 && height == 1 && !framebuffer->clear_clip_dirty)
{
/* If everything drawn so far for this frame is still in the
* Journal then if all of the rectangles only have a flat
* opaque color we have a fast-path for reading a single pixel
* that avoids the relatively high cost of flushing primitives
* to be drawn on the GPU (considering how simple the geometry
* is in this case) and then blocking on the long GPU pipelines
* for the result.
*/
if (_cogl_framebuffer_try_fast_read_pixel (framebuffer,
x, y, source, bitmap))
return TRUE;
}
ctx = cogl_framebuffer_get_context (framebuffer);
/* make sure any batched primitives get emitted to the driver
* before issuing our read pixels...
*/
_cogl_framebuffer_flush_journal (framebuffer);
return ctx->driver_vtable->framebuffer_read_pixels_into_bitmap (framebuffer,
x, y,
source,
bitmap,
error);
Allow propogation of OOM errors to apps This allows apps to catch out-of-memory errors when allocating textures. Textures can be pretty huge at times and so it's quite possible for an application to try and allocate more memory than is available. It's also very possible that the application can take some action in response to reduce memory pressure (such as freeing up texture caches perhaps) so we shouldn't just automatically abort like we do for trivial heap allocations. These public functions now take a CoglError argument so applications can catch out of memory errors: cogl_buffer_map cogl_buffer_map_range cogl_buffer_set_data cogl_framebuffer_read_pixels_into_bitmap cogl_pixel_buffer_new cogl_texture_new_from_data cogl_texture_new_from_bitmap Note: we've been quite conservative with how many apis we let throw OOM CoglErrors since we don't really want to put a burdon on developers to be checking for errors with every cogl api call. So long as there is some lower level api for apps to use that let them catch OOM errors for everything necessary that's enough and we don't have to make more convenient apis more awkward to use. The main focus is on bitmaps and texture allocations since they can be particularly large and prone to failing. A new cogl_attribute_buffer_new_with_size() function has been added in case developers need to catch OOM errors when allocating attribute buffers whereby they can first use _buffer_new_with_size() (which doesn't take a CoglError) followed by cogl_buffer_set_data() which will lazily allocate the buffer storage and report OOM errors. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f7735e141ad537a253b02afa2a8238f96340b978) Note: since we can't break the API for Cogl 1.x then actually the main purpose of cherry picking this patch is to keep in-line with changes on the master branch so that we can easily cherry-pick patches. All the api changes relating stable apis released on the 1.12 branch have been reverted as part of cherry-picking this patch so this most just applies all the internal plumbing changes that enable us to correctly propagate OOM errors.
2012-11-08 12:54:10 -05:00
}
CoglBool
cogl_framebuffer_read_pixels_into_bitmap (CoglFramebuffer *framebuffer,
int x,
int y,
CoglReadPixelsFlags source,
CoglBitmap *bitmap)
{
CoglError *ignore_error = NULL;
CoglBool status =
_cogl_framebuffer_read_pixels_into_bitmap (framebuffer,
x, y, source, bitmap,
&ignore_error);
if (!status)
cogl_error_free (ignore_error);
return status;
}
CoglBool
cogl_framebuffer_read_pixels (CoglFramebuffer *framebuffer,
int x,
int y,
int width,
int height,
CoglPixelFormat format,
uint8_t *pixels)
{
int bpp = _cogl_pixel_format_get_bytes_per_pixel (format);
CoglBitmap *bitmap;
CoglBool ret;
bitmap = cogl_bitmap_new_for_data (framebuffer->context,
width, height,
format,
bpp * width, /* rowstride */
pixels);
Allow propogation of OOM errors to apps This allows apps to catch out-of-memory errors when allocating textures. Textures can be pretty huge at times and so it's quite possible for an application to try and allocate more memory than is available. It's also very possible that the application can take some action in response to reduce memory pressure (such as freeing up texture caches perhaps) so we shouldn't just automatically abort like we do for trivial heap allocations. These public functions now take a CoglError argument so applications can catch out of memory errors: cogl_buffer_map cogl_buffer_map_range cogl_buffer_set_data cogl_framebuffer_read_pixels_into_bitmap cogl_pixel_buffer_new cogl_texture_new_from_data cogl_texture_new_from_bitmap Note: we've been quite conservative with how many apis we let throw OOM CoglErrors since we don't really want to put a burdon on developers to be checking for errors with every cogl api call. So long as there is some lower level api for apps to use that let them catch OOM errors for everything necessary that's enough and we don't have to make more convenient apis more awkward to use. The main focus is on bitmaps and texture allocations since they can be particularly large and prone to failing. A new cogl_attribute_buffer_new_with_size() function has been added in case developers need to catch OOM errors when allocating attribute buffers whereby they can first use _buffer_new_with_size() (which doesn't take a CoglError) followed by cogl_buffer_set_data() which will lazily allocate the buffer storage and report OOM errors. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f7735e141ad537a253b02afa2a8238f96340b978) Note: since we can't break the API for Cogl 1.x then actually the main purpose of cherry picking this patch is to keep in-line with changes on the master branch so that we can easily cherry-pick patches. All the api changes relating stable apis released on the 1.12 branch have been reverted as part of cherry-picking this patch so this most just applies all the internal plumbing changes that enable us to correctly propagate OOM errors.
2012-11-08 12:54:10 -05:00
/* Note: we don't try and catch errors here since we created the
* bitmap storage up-front and can assume we wont hit an
* out-of-memory error which should be the only exception
* this api throws.
*/
ret = _cogl_framebuffer_read_pixels_into_bitmap (framebuffer,
x, y,
COGL_READ_PIXELS_COLOR_BUFFER,
bitmap,
NULL);
cogl_object_unref (bitmap);
return ret;
}
void
_cogl_blit_framebuffer (CoglFramebuffer *src,
CoglFramebuffer *dest,
int src_x,
int src_y,
int dst_x,
int dst_y,
int width,
int height)
{
CoglContext *ctx = src->context;
_COGL_RETURN_IF_FAIL (ctx->private_feature_flags &
COGL_PRIVATE_FEATURE_OFFSCREEN_BLIT);
/* We can only support blitting between offscreen buffers because
otherwise we would need to mirror the image and GLES2.0 doesn't
support this */
_COGL_RETURN_IF_FAIL (cogl_is_offscreen (src));
_COGL_RETURN_IF_FAIL (cogl_is_offscreen (dest));
/* The buffers must be the same format */
_COGL_RETURN_IF_FAIL (src->format == dest->format);
/* Make sure the current framebuffers are bound. We explicitly avoid
flushing the clip state so we can bind our own empty state */
_cogl_framebuffer_flush_state (dest,
src,
COGL_FRAMEBUFFER_STATE_ALL &
~COGL_FRAMEBUFFER_STATE_CLIP);
/* Flush any empty clip stack because glBlitFramebuffer is affected
by the scissor and we want to hide this feature for the Cogl API
because it's not obvious to an app how the clip state will affect
the scissor */
_cogl_clip_stack_flush (NULL, dest);
/* XXX: Because we are manually flushing clip state here we need to
* make sure that the clip state gets updated the next time we flush
* framebuffer state by marking the current framebuffer's clip state
* as changed */
ctx->current_draw_buffer_changes |= COGL_FRAMEBUFFER_STATE_CLIP;
ctx->glBlitFramebuffer (src_x, src_y,
src_x + width, src_y + height,
dst_x, dst_y,
dst_x + width, dst_y + height,
GL_COLOR_BUFFER_BIT,
GL_NEAREST);
}
void
cogl_framebuffer_discard_buffers (CoglFramebuffer *framebuffer,
unsigned long buffers)
{
CoglContext *ctx = framebuffer->context;
_COGL_RETURN_IF_FAIL (buffers & COGL_BUFFER_BIT_COLOR);
ctx->driver_vtable->framebuffer_discard_buffers (framebuffer, buffers);
}
void
cogl_framebuffer_finish (CoglFramebuffer *framebuffer)
{
CoglContext *ctx = framebuffer->context;
_cogl_framebuffer_flush_journal (framebuffer);
ctx->driver_vtable->framebuffer_finish (framebuffer);
}
void
cogl_framebuffer_push_matrix (CoglFramebuffer *framebuffer)
{
CoglMatrixStack *modelview_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
cogl_matrix_stack_push (modelview_stack);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_MODELVIEW;
}
void
cogl_framebuffer_pop_matrix (CoglFramebuffer *framebuffer)
{
CoglMatrixStack *modelview_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
cogl_matrix_stack_pop (modelview_stack);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_MODELVIEW;
}
void
cogl_framebuffer_identity_matrix (CoglFramebuffer *framebuffer)
{
CoglMatrixStack *modelview_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
cogl_matrix_stack_load_identity (modelview_stack);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_MODELVIEW;
}
void
cogl_framebuffer_scale (CoglFramebuffer *framebuffer,
float x,
float y,
float z)
{
CoglMatrixStack *modelview_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
cogl_matrix_stack_scale (modelview_stack, x, y, z);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_MODELVIEW;
}
void
cogl_framebuffer_translate (CoglFramebuffer *framebuffer,
float x,
float y,
float z)
{
CoglMatrixStack *modelview_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
cogl_matrix_stack_translate (modelview_stack, x, y, z);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_MODELVIEW;
}
void
cogl_framebuffer_rotate (CoglFramebuffer *framebuffer,
float angle,
float x,
float y,
float z)
{
CoglMatrixStack *modelview_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
cogl_matrix_stack_rotate (modelview_stack, angle, x, y, z);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_MODELVIEW;
}
void
cogl_framebuffer_rotate_quaternion (CoglFramebuffer *framebuffer,
const CoglQuaternion *quaternion)
{
CoglMatrixStack *modelview_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
cogl_matrix_stack_rotate_quaternion (modelview_stack, quaternion);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_MODELVIEW;
}
void
cogl_framebuffer_rotate_euler (CoglFramebuffer *framebuffer,
const CoglEuler *euler)
{
CoglMatrixStack *modelview_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
cogl_matrix_stack_rotate_euler (modelview_stack, euler);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_MODELVIEW;
}
void
cogl_framebuffer_transform (CoglFramebuffer *framebuffer,
const CoglMatrix *matrix)
{
CoglMatrixStack *modelview_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
cogl_matrix_stack_multiply (modelview_stack, matrix);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_MODELVIEW;
}
void
cogl_framebuffer_perspective (CoglFramebuffer *framebuffer,
float fov_y,
float aspect,
float z_near,
float z_far)
{
float ymax = z_near * tanf (fov_y * G_PI / 360.0);
cogl_framebuffer_frustum (framebuffer,
-ymax * aspect, /* left */
ymax * aspect, /* right */
-ymax, /* bottom */
ymax, /* top */
z_near,
z_far);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_PROJECTION;
}
void
cogl_framebuffer_frustum (CoglFramebuffer *framebuffer,
float left,
float right,
float bottom,
float top,
float z_near,
float z_far)
{
CoglMatrixStack *projection_stack =
_cogl_framebuffer_get_projection_stack (framebuffer);
/* XXX: The projection matrix isn't currently tracked in the journal
* so we need to flush all journaled primitives first... */
_cogl_framebuffer_flush_journal (framebuffer);
cogl_matrix_stack_load_identity (projection_stack);
cogl_matrix_stack_frustum (projection_stack,
left,
right,
bottom,
top,
z_near,
z_far);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_PROJECTION;
}
void
cogl_framebuffer_orthographic (CoglFramebuffer *framebuffer,
float x_1,
float y_1,
float x_2,
float y_2,
float near,
float far)
{
CoglMatrix ortho;
CoglMatrixStack *projection_stack =
_cogl_framebuffer_get_projection_stack (framebuffer);
/* XXX: The projection matrix isn't currently tracked in the journal
* so we need to flush all journaled primitives first... */
_cogl_framebuffer_flush_journal (framebuffer);
cogl_matrix_init_identity (&ortho);
cogl_matrix_orthographic (&ortho, x_1, y_1, x_2, y_2, near, far);
cogl_matrix_stack_set (projection_stack, &ortho);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_PROJECTION;
}
void
_cogl_framebuffer_push_projection (CoglFramebuffer *framebuffer)
{
CoglMatrixStack *projection_stack =
_cogl_framebuffer_get_projection_stack (framebuffer);
cogl_matrix_stack_push (projection_stack);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_PROJECTION;
}
void
_cogl_framebuffer_pop_projection (CoglFramebuffer *framebuffer)
{
CoglMatrixStack *projection_stack =
_cogl_framebuffer_get_projection_stack (framebuffer);
cogl_matrix_stack_pop (projection_stack);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_PROJECTION;
}
void
cogl_framebuffer_get_modelview_matrix (CoglFramebuffer *framebuffer,
CoglMatrix *matrix)
{
Re-design the matrix stack using a graph of ops This re-designs the matrix stack so we now keep track of each separate operation such as rotating, scaling, translating and multiplying as immutable, ref-counted nodes in a graph. Being a "graph" here means that different transformations composed of a sequence of linked operation nodes may share nodes. The first node in a matrix-stack is always a LOAD_IDENTITY operation. As an example consider if an application where to draw three rectangles A, B and C something like this: cogl_framebuffer_scale (fb, 2, 2, 2); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_translate (fb, 10, 0, 0); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_rotate (fb, 45, 0, 0, 1); cogl_framebuffer_draw_rectangle (...); /* A */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_draw_rectangle (...); /* B */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_set_modelview_matrix (fb, &mv); cogl_framebuffer_draw_rectangle (...); /* C */ cogl_framebuffer_pop_matrix(fb); That would result in a graph of nodes like this: LOAD_IDENTITY | SCALE / \ SAVE LOAD | | TRANSLATE RECTANGLE(C) | \ SAVE RECTANGLE(B) | ROTATE | RECTANGLE(A) Each push adds a SAVE operation which serves as a marker to rewind too when a corresponding pop is issued and also each SAVE node may also store a cached matrix representing the composition of all its ancestor nodes. This means if we repeatedly need to resolve a real CoglMatrix for a given node then we don't need to repeat the composition. Some advantages of this design are: - A single pointer to any node in the graph can now represent a complete, immutable transformation that can be logged for example into a journal. Previously we were storing a full CoglMatrix in each journal entry which is 16 floats for the matrix itself as well as space for flags and another 16 floats for possibly storing a cache of the inverse. This means that we significantly reduce the size of the journal when drawing lots of primitives and we also avoid copying over 128 bytes per entry. - It becomes much cheaper to check for equality. In cases where some (unlikely) false negatives are allowed simply comparing the pointers of two matrix stack graph entries is enough. Previously we would use memcmp() to compare matrices. - It becomes easier to do comparisons of transformations. By looking for the common ancestry between nodes we can determine the operations that differentiate the transforms and use those to gain a high level understanding of the differences. For example we use this in the journal to be able to efficiently determine when two rectangle transforms only differ by some translation so that we can perform software clipping. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f75aee93f6b293ca7a7babbd8fcc326ee6bf7aef)
2012-02-20 10:59:48 -05:00
CoglMatrixEntry *modelview_entry =
_cogl_framebuffer_get_modelview_entry (framebuffer);
cogl_matrix_entry_get (modelview_entry, matrix);
_COGL_MATRIX_DEBUG_PRINT (matrix);
}
void
cogl_framebuffer_set_modelview_matrix (CoglFramebuffer *framebuffer,
const CoglMatrix *matrix)
{
CoglMatrixStack *modelview_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
cogl_matrix_stack_set (modelview_stack, matrix);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_MODELVIEW;
_COGL_MATRIX_DEBUG_PRINT (matrix);
}
void
cogl_framebuffer_get_projection_matrix (CoglFramebuffer *framebuffer,
CoglMatrix *matrix)
{
Re-design the matrix stack using a graph of ops This re-designs the matrix stack so we now keep track of each separate operation such as rotating, scaling, translating and multiplying as immutable, ref-counted nodes in a graph. Being a "graph" here means that different transformations composed of a sequence of linked operation nodes may share nodes. The first node in a matrix-stack is always a LOAD_IDENTITY operation. As an example consider if an application where to draw three rectangles A, B and C something like this: cogl_framebuffer_scale (fb, 2, 2, 2); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_translate (fb, 10, 0, 0); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_rotate (fb, 45, 0, 0, 1); cogl_framebuffer_draw_rectangle (...); /* A */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_draw_rectangle (...); /* B */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_set_modelview_matrix (fb, &mv); cogl_framebuffer_draw_rectangle (...); /* C */ cogl_framebuffer_pop_matrix(fb); That would result in a graph of nodes like this: LOAD_IDENTITY | SCALE / \ SAVE LOAD | | TRANSLATE RECTANGLE(C) | \ SAVE RECTANGLE(B) | ROTATE | RECTANGLE(A) Each push adds a SAVE operation which serves as a marker to rewind too when a corresponding pop is issued and also each SAVE node may also store a cached matrix representing the composition of all its ancestor nodes. This means if we repeatedly need to resolve a real CoglMatrix for a given node then we don't need to repeat the composition. Some advantages of this design are: - A single pointer to any node in the graph can now represent a complete, immutable transformation that can be logged for example into a journal. Previously we were storing a full CoglMatrix in each journal entry which is 16 floats for the matrix itself as well as space for flags and another 16 floats for possibly storing a cache of the inverse. This means that we significantly reduce the size of the journal when drawing lots of primitives and we also avoid copying over 128 bytes per entry. - It becomes much cheaper to check for equality. In cases where some (unlikely) false negatives are allowed simply comparing the pointers of two matrix stack graph entries is enough. Previously we would use memcmp() to compare matrices. - It becomes easier to do comparisons of transformations. By looking for the common ancestry between nodes we can determine the operations that differentiate the transforms and use those to gain a high level understanding of the differences. For example we use this in the journal to be able to efficiently determine when two rectangle transforms only differ by some translation so that we can perform software clipping. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f75aee93f6b293ca7a7babbd8fcc326ee6bf7aef)
2012-02-20 10:59:48 -05:00
CoglMatrixEntry *projection_entry =
_cogl_framebuffer_get_projection_entry (framebuffer);
cogl_matrix_entry_get (projection_entry, matrix);
_COGL_MATRIX_DEBUG_PRINT (matrix);
}
void
cogl_framebuffer_set_projection_matrix (CoglFramebuffer *framebuffer,
const CoglMatrix *matrix)
{
CoglMatrixStack *projection_stack =
_cogl_framebuffer_get_projection_stack (framebuffer);
/* XXX: The projection matrix isn't currently tracked in the journal
* so we need to flush all journaled primitives first... */
_cogl_framebuffer_flush_journal (framebuffer);
cogl_matrix_stack_set (projection_stack, matrix);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_PROJECTION;
_COGL_MATRIX_DEBUG_PRINT (matrix);
}
void
cogl_framebuffer_push_scissor_clip (CoglFramebuffer *framebuffer,
int x,
int y,
int width,
int height)
{
CoglClipState *clip_state = _cogl_framebuffer_get_clip_state (framebuffer);
clip_state->stacks->data =
_cogl_clip_stack_push_window_rectangle (clip_state->stacks->data,
x, y, width, height);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_CLIP;
}
void
cogl_framebuffer_push_rectangle_clip (CoglFramebuffer *framebuffer,
float x_1,
float y_1,
float x_2,
float y_2)
{
CoglClipState *clip_state = _cogl_framebuffer_get_clip_state (framebuffer);
Re-design the matrix stack using a graph of ops This re-designs the matrix stack so we now keep track of each separate operation such as rotating, scaling, translating and multiplying as immutable, ref-counted nodes in a graph. Being a "graph" here means that different transformations composed of a sequence of linked operation nodes may share nodes. The first node in a matrix-stack is always a LOAD_IDENTITY operation. As an example consider if an application where to draw three rectangles A, B and C something like this: cogl_framebuffer_scale (fb, 2, 2, 2); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_translate (fb, 10, 0, 0); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_rotate (fb, 45, 0, 0, 1); cogl_framebuffer_draw_rectangle (...); /* A */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_draw_rectangle (...); /* B */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_set_modelview_matrix (fb, &mv); cogl_framebuffer_draw_rectangle (...); /* C */ cogl_framebuffer_pop_matrix(fb); That would result in a graph of nodes like this: LOAD_IDENTITY | SCALE / \ SAVE LOAD | | TRANSLATE RECTANGLE(C) | \ SAVE RECTANGLE(B) | ROTATE | RECTANGLE(A) Each push adds a SAVE operation which serves as a marker to rewind too when a corresponding pop is issued and also each SAVE node may also store a cached matrix representing the composition of all its ancestor nodes. This means if we repeatedly need to resolve a real CoglMatrix for a given node then we don't need to repeat the composition. Some advantages of this design are: - A single pointer to any node in the graph can now represent a complete, immutable transformation that can be logged for example into a journal. Previously we were storing a full CoglMatrix in each journal entry which is 16 floats for the matrix itself as well as space for flags and another 16 floats for possibly storing a cache of the inverse. This means that we significantly reduce the size of the journal when drawing lots of primitives and we also avoid copying over 128 bytes per entry. - It becomes much cheaper to check for equality. In cases where some (unlikely) false negatives are allowed simply comparing the pointers of two matrix stack graph entries is enough. Previously we would use memcmp() to compare matrices. - It becomes easier to do comparisons of transformations. By looking for the common ancestry between nodes we can determine the operations that differentiate the transforms and use those to gain a high level understanding of the differences. For example we use this in the journal to be able to efficiently determine when two rectangle transforms only differ by some translation so that we can perform software clipping. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f75aee93f6b293ca7a7babbd8fcc326ee6bf7aef)
2012-02-20 10:59:48 -05:00
CoglMatrixEntry *modelview_entry =
_cogl_framebuffer_get_modelview_entry (framebuffer);
CoglMatrixEntry *projection_entry =
_cogl_framebuffer_get_projection_entry (framebuffer);
/* XXX: It would be nicer if we stored the private viewport as a
* vec4 so we could avoid this redundant copy. */
float viewport[] = {
framebuffer->viewport_x,
framebuffer->viewport_y,
framebuffer->viewport_width,
framebuffer->viewport_height
};
clip_state->stacks->data =
_cogl_clip_stack_push_rectangle (clip_state->stacks->data,
x_1, y_1, x_2, y_2,
Re-design the matrix stack using a graph of ops This re-designs the matrix stack so we now keep track of each separate operation such as rotating, scaling, translating and multiplying as immutable, ref-counted nodes in a graph. Being a "graph" here means that different transformations composed of a sequence of linked operation nodes may share nodes. The first node in a matrix-stack is always a LOAD_IDENTITY operation. As an example consider if an application where to draw three rectangles A, B and C something like this: cogl_framebuffer_scale (fb, 2, 2, 2); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_translate (fb, 10, 0, 0); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_rotate (fb, 45, 0, 0, 1); cogl_framebuffer_draw_rectangle (...); /* A */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_draw_rectangle (...); /* B */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_set_modelview_matrix (fb, &mv); cogl_framebuffer_draw_rectangle (...); /* C */ cogl_framebuffer_pop_matrix(fb); That would result in a graph of nodes like this: LOAD_IDENTITY | SCALE / \ SAVE LOAD | | TRANSLATE RECTANGLE(C) | \ SAVE RECTANGLE(B) | ROTATE | RECTANGLE(A) Each push adds a SAVE operation which serves as a marker to rewind too when a corresponding pop is issued and also each SAVE node may also store a cached matrix representing the composition of all its ancestor nodes. This means if we repeatedly need to resolve a real CoglMatrix for a given node then we don't need to repeat the composition. Some advantages of this design are: - A single pointer to any node in the graph can now represent a complete, immutable transformation that can be logged for example into a journal. Previously we were storing a full CoglMatrix in each journal entry which is 16 floats for the matrix itself as well as space for flags and another 16 floats for possibly storing a cache of the inverse. This means that we significantly reduce the size of the journal when drawing lots of primitives and we also avoid copying over 128 bytes per entry. - It becomes much cheaper to check for equality. In cases where some (unlikely) false negatives are allowed simply comparing the pointers of two matrix stack graph entries is enough. Previously we would use memcmp() to compare matrices. - It becomes easier to do comparisons of transformations. By looking for the common ancestry between nodes we can determine the operations that differentiate the transforms and use those to gain a high level understanding of the differences. For example we use this in the journal to be able to efficiently determine when two rectangle transforms only differ by some translation so that we can perform software clipping. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f75aee93f6b293ca7a7babbd8fcc326ee6bf7aef)
2012-02-20 10:59:48 -05:00
modelview_entry,
projection_entry,
viewport);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_CLIP;
}
void
cogl_framebuffer_push_path_clip (CoglFramebuffer *framebuffer,
CoglPath *path)
{
CoglClipState *clip_state = _cogl_framebuffer_get_clip_state (framebuffer);
Re-design the matrix stack using a graph of ops This re-designs the matrix stack so we now keep track of each separate operation such as rotating, scaling, translating and multiplying as immutable, ref-counted nodes in a graph. Being a "graph" here means that different transformations composed of a sequence of linked operation nodes may share nodes. The first node in a matrix-stack is always a LOAD_IDENTITY operation. As an example consider if an application where to draw three rectangles A, B and C something like this: cogl_framebuffer_scale (fb, 2, 2, 2); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_translate (fb, 10, 0, 0); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_rotate (fb, 45, 0, 0, 1); cogl_framebuffer_draw_rectangle (...); /* A */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_draw_rectangle (...); /* B */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_set_modelview_matrix (fb, &mv); cogl_framebuffer_draw_rectangle (...); /* C */ cogl_framebuffer_pop_matrix(fb); That would result in a graph of nodes like this: LOAD_IDENTITY | SCALE / \ SAVE LOAD | | TRANSLATE RECTANGLE(C) | \ SAVE RECTANGLE(B) | ROTATE | RECTANGLE(A) Each push adds a SAVE operation which serves as a marker to rewind too when a corresponding pop is issued and also each SAVE node may also store a cached matrix representing the composition of all its ancestor nodes. This means if we repeatedly need to resolve a real CoglMatrix for a given node then we don't need to repeat the composition. Some advantages of this design are: - A single pointer to any node in the graph can now represent a complete, immutable transformation that can be logged for example into a journal. Previously we were storing a full CoglMatrix in each journal entry which is 16 floats for the matrix itself as well as space for flags and another 16 floats for possibly storing a cache of the inverse. This means that we significantly reduce the size of the journal when drawing lots of primitives and we also avoid copying over 128 bytes per entry. - It becomes much cheaper to check for equality. In cases where some (unlikely) false negatives are allowed simply comparing the pointers of two matrix stack graph entries is enough. Previously we would use memcmp() to compare matrices. - It becomes easier to do comparisons of transformations. By looking for the common ancestry between nodes we can determine the operations that differentiate the transforms and use those to gain a high level understanding of the differences. For example we use this in the journal to be able to efficiently determine when two rectangle transforms only differ by some translation so that we can perform software clipping. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f75aee93f6b293ca7a7babbd8fcc326ee6bf7aef)
2012-02-20 10:59:48 -05:00
CoglMatrixEntry *modelview_entry =
_cogl_framebuffer_get_modelview_entry (framebuffer);
CoglMatrixEntry *projection_entry =
_cogl_framebuffer_get_projection_entry (framebuffer);
/* XXX: It would be nicer if we stored the private viewport as a
* vec4 so we could avoid this redundant copy. */
float viewport[] = {
framebuffer->viewport_x,
framebuffer->viewport_y,
framebuffer->viewport_width,
framebuffer->viewport_height
};
clip_state->stacks->data =
_cogl_clip_stack_push_from_path (clip_state->stacks->data,
path,
Re-design the matrix stack using a graph of ops This re-designs the matrix stack so we now keep track of each separate operation such as rotating, scaling, translating and multiplying as immutable, ref-counted nodes in a graph. Being a "graph" here means that different transformations composed of a sequence of linked operation nodes may share nodes. The first node in a matrix-stack is always a LOAD_IDENTITY operation. As an example consider if an application where to draw three rectangles A, B and C something like this: cogl_framebuffer_scale (fb, 2, 2, 2); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_translate (fb, 10, 0, 0); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_rotate (fb, 45, 0, 0, 1); cogl_framebuffer_draw_rectangle (...); /* A */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_draw_rectangle (...); /* B */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_set_modelview_matrix (fb, &mv); cogl_framebuffer_draw_rectangle (...); /* C */ cogl_framebuffer_pop_matrix(fb); That would result in a graph of nodes like this: LOAD_IDENTITY | SCALE / \ SAVE LOAD | | TRANSLATE RECTANGLE(C) | \ SAVE RECTANGLE(B) | ROTATE | RECTANGLE(A) Each push adds a SAVE operation which serves as a marker to rewind too when a corresponding pop is issued and also each SAVE node may also store a cached matrix representing the composition of all its ancestor nodes. This means if we repeatedly need to resolve a real CoglMatrix for a given node then we don't need to repeat the composition. Some advantages of this design are: - A single pointer to any node in the graph can now represent a complete, immutable transformation that can be logged for example into a journal. Previously we were storing a full CoglMatrix in each journal entry which is 16 floats for the matrix itself as well as space for flags and another 16 floats for possibly storing a cache of the inverse. This means that we significantly reduce the size of the journal when drawing lots of primitives and we also avoid copying over 128 bytes per entry. - It becomes much cheaper to check for equality. In cases where some (unlikely) false negatives are allowed simply comparing the pointers of two matrix stack graph entries is enough. Previously we would use memcmp() to compare matrices. - It becomes easier to do comparisons of transformations. By looking for the common ancestry between nodes we can determine the operations that differentiate the transforms and use those to gain a high level understanding of the differences. For example we use this in the journal to be able to efficiently determine when two rectangle transforms only differ by some translation so that we can perform software clipping. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f75aee93f6b293ca7a7babbd8fcc326ee6bf7aef)
2012-02-20 10:59:48 -05:00
modelview_entry,
projection_entry,
viewport);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_CLIP;
}
void
cogl_framebuffer_push_primitive_clip (CoglFramebuffer *framebuffer,
CoglPrimitive *primitive,
float bounds_x1,
float bounds_y1,
float bounds_x2,
float bounds_y2)
{
CoglClipState *clip_state = _cogl_framebuffer_get_clip_state (framebuffer);
Re-design the matrix stack using a graph of ops This re-designs the matrix stack so we now keep track of each separate operation such as rotating, scaling, translating and multiplying as immutable, ref-counted nodes in a graph. Being a "graph" here means that different transformations composed of a sequence of linked operation nodes may share nodes. The first node in a matrix-stack is always a LOAD_IDENTITY operation. As an example consider if an application where to draw three rectangles A, B and C something like this: cogl_framebuffer_scale (fb, 2, 2, 2); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_translate (fb, 10, 0, 0); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_rotate (fb, 45, 0, 0, 1); cogl_framebuffer_draw_rectangle (...); /* A */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_draw_rectangle (...); /* B */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_set_modelview_matrix (fb, &mv); cogl_framebuffer_draw_rectangle (...); /* C */ cogl_framebuffer_pop_matrix(fb); That would result in a graph of nodes like this: LOAD_IDENTITY | SCALE / \ SAVE LOAD | | TRANSLATE RECTANGLE(C) | \ SAVE RECTANGLE(B) | ROTATE | RECTANGLE(A) Each push adds a SAVE operation which serves as a marker to rewind too when a corresponding pop is issued and also each SAVE node may also store a cached matrix representing the composition of all its ancestor nodes. This means if we repeatedly need to resolve a real CoglMatrix for a given node then we don't need to repeat the composition. Some advantages of this design are: - A single pointer to any node in the graph can now represent a complete, immutable transformation that can be logged for example into a journal. Previously we were storing a full CoglMatrix in each journal entry which is 16 floats for the matrix itself as well as space for flags and another 16 floats for possibly storing a cache of the inverse. This means that we significantly reduce the size of the journal when drawing lots of primitives and we also avoid copying over 128 bytes per entry. - It becomes much cheaper to check for equality. In cases where some (unlikely) false negatives are allowed simply comparing the pointers of two matrix stack graph entries is enough. Previously we would use memcmp() to compare matrices. - It becomes easier to do comparisons of transformations. By looking for the common ancestry between nodes we can determine the operations that differentiate the transforms and use those to gain a high level understanding of the differences. For example we use this in the journal to be able to efficiently determine when two rectangle transforms only differ by some translation so that we can perform software clipping. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f75aee93f6b293ca7a7babbd8fcc326ee6bf7aef)
2012-02-20 10:59:48 -05:00
CoglMatrixEntry *modelview_entry =
_cogl_framebuffer_get_modelview_entry (framebuffer);
CoglMatrixEntry *projection_entry =
_cogl_framebuffer_get_projection_entry (framebuffer);
/* XXX: It would be nicer if we stored the private viewport as a
* vec4 so we could avoid this redundant copy. */
float viewport[] = {
framebuffer->viewport_x,
framebuffer->viewport_y,
framebuffer->viewport_width,
framebuffer->viewport_height
};
clip_state->stacks->data =
_cogl_clip_stack_push_primitive (clip_state->stacks->data,
primitive,
bounds_x1, bounds_y1,
bounds_x2, bounds_y2,
Re-design the matrix stack using a graph of ops This re-designs the matrix stack so we now keep track of each separate operation such as rotating, scaling, translating and multiplying as immutable, ref-counted nodes in a graph. Being a "graph" here means that different transformations composed of a sequence of linked operation nodes may share nodes. The first node in a matrix-stack is always a LOAD_IDENTITY operation. As an example consider if an application where to draw three rectangles A, B and C something like this: cogl_framebuffer_scale (fb, 2, 2, 2); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_translate (fb, 10, 0, 0); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_rotate (fb, 45, 0, 0, 1); cogl_framebuffer_draw_rectangle (...); /* A */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_draw_rectangle (...); /* B */ cogl_framebuffer_pop_matrix(fb); cogl_framebuffer_push_matrix(fb); cogl_framebuffer_set_modelview_matrix (fb, &mv); cogl_framebuffer_draw_rectangle (...); /* C */ cogl_framebuffer_pop_matrix(fb); That would result in a graph of nodes like this: LOAD_IDENTITY | SCALE / \ SAVE LOAD | | TRANSLATE RECTANGLE(C) | \ SAVE RECTANGLE(B) | ROTATE | RECTANGLE(A) Each push adds a SAVE operation which serves as a marker to rewind too when a corresponding pop is issued and also each SAVE node may also store a cached matrix representing the composition of all its ancestor nodes. This means if we repeatedly need to resolve a real CoglMatrix for a given node then we don't need to repeat the composition. Some advantages of this design are: - A single pointer to any node in the graph can now represent a complete, immutable transformation that can be logged for example into a journal. Previously we were storing a full CoglMatrix in each journal entry which is 16 floats for the matrix itself as well as space for flags and another 16 floats for possibly storing a cache of the inverse. This means that we significantly reduce the size of the journal when drawing lots of primitives and we also avoid copying over 128 bytes per entry. - It becomes much cheaper to check for equality. In cases where some (unlikely) false negatives are allowed simply comparing the pointers of two matrix stack graph entries is enough. Previously we would use memcmp() to compare matrices. - It becomes easier to do comparisons of transformations. By looking for the common ancestry between nodes we can determine the operations that differentiate the transforms and use those to gain a high level understanding of the differences. For example we use this in the journal to be able to efficiently determine when two rectangle transforms only differ by some translation so that we can perform software clipping. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f75aee93f6b293ca7a7babbd8fcc326ee6bf7aef)
2012-02-20 10:59:48 -05:00
modelview_entry,
projection_entry,
viewport);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_CLIP;
}
void
cogl_framebuffer_pop_clip (CoglFramebuffer *framebuffer)
{
CoglClipState *clip_state = _cogl_framebuffer_get_clip_state (framebuffer);
clip_state->stacks->data = _cogl_clip_stack_pop (clip_state->stacks->data);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_CLIP;
}
void
_cogl_framebuffer_save_clip_stack (CoglFramebuffer *framebuffer)
{
CoglClipState *clip_state = _cogl_framebuffer_get_clip_state (framebuffer);
_cogl_clip_state_save_clip_stack (clip_state);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_CLIP;
}
void
_cogl_framebuffer_restore_clip_stack (CoglFramebuffer *framebuffer)
{
CoglClipState *clip_state = _cogl_framebuffer_get_clip_state (framebuffer);
_cogl_clip_state_restore_clip_stack (clip_state);
if (framebuffer->context->current_draw_buffer == framebuffer)
framebuffer->context->current_draw_buffer_changes |=
COGL_FRAMEBUFFER_STATE_CLIP;
}
void
_cogl_framebuffer_unref (CoglFramebuffer *framebuffer)
{
/* The journal holds a reference to the framebuffer whenever it is
non-empty. Therefore if the journal is non-empty and we will have
exactly one reference then we know the journal is the only thing
keeping the framebuffer alive. In that case we want to flush the
journal and let the framebuffer die. It is fine at this point if
flushing the journal causes something else to take a reference to
it and it comes back to life */
if (framebuffer->journal->entries->len > 0)
{
unsigned int ref_count = ((CoglObject *) framebuffer)->ref_count;
/* There should be at least two references - the one we are
about to drop and the one held by the journal */
if (ref_count < 2)
g_warning ("Inconsistent ref count on a framebuffer with journal "
"entries.");
if (ref_count == 2)
_cogl_framebuffer_flush_journal (framebuffer);
}
/* Chain-up */
_cogl_object_default_unref (framebuffer);
}
#ifdef COGL_ENABLE_DEBUG
static int
get_index (void *indices,
CoglIndicesType type,
int _index)
{
if (!indices)
return _index;
switch (type)
{
case COGL_INDICES_TYPE_UNSIGNED_BYTE:
return ((uint8_t *)indices)[_index];
case COGL_INDICES_TYPE_UNSIGNED_SHORT:
return ((uint16_t *)indices)[_index];
case COGL_INDICES_TYPE_UNSIGNED_INT:
return ((uint32_t *)indices)[_index];
}
g_return_val_if_reached (0);
}
static void
add_line (uint32_t *line_indices,
int base,
void *user_indices,
CoglIndicesType user_indices_type,
int index0,
int index1,
int *pos)
{
index0 = get_index (user_indices, user_indices_type, index0);
index1 = get_index (user_indices, user_indices_type, index1);
line_indices[(*pos)++] = base + index0;
line_indices[(*pos)++] = base + index1;
}
static int
get_line_count (CoglVerticesMode mode, int n_vertices)
{
if (mode == COGL_VERTICES_MODE_TRIANGLES &&
(n_vertices % 3) == 0)
{
return n_vertices;
}
else if (mode == COGL_VERTICES_MODE_TRIANGLE_FAN &&
n_vertices >= 3)
{
return 2 * n_vertices - 3;
}
else if (mode == COGL_VERTICES_MODE_TRIANGLE_STRIP &&
n_vertices >= 3)
{
return 2 * n_vertices - 3;
}
/* In the journal we are a bit sneaky and actually use GL_QUADS
* which isn't actually a valid CoglVerticesMode! */
#ifdef HAVE_COGL_GL
else if (mode == GL_QUADS && (n_vertices % 4) == 0)
{
return n_vertices;
}
#endif
g_return_val_if_reached (0);
}
static CoglIndices *
get_wire_line_indices (CoglContext *ctx,
CoglVerticesMode mode,
int first_vertex,
int n_vertices_in,
CoglIndices *user_indices,
int *n_indices)
{
int n_lines;
uint32_t *line_indices;
CoglIndexBuffer *index_buffer;
void *indices;
CoglIndicesType indices_type;
int base = first_vertex;
int pos;
int i;
CoglIndices *ret;
if (user_indices)
{
index_buffer = cogl_indices_get_buffer (user_indices);
Allow propogation of OOM errors to apps This allows apps to catch out-of-memory errors when allocating textures. Textures can be pretty huge at times and so it's quite possible for an application to try and allocate more memory than is available. It's also very possible that the application can take some action in response to reduce memory pressure (such as freeing up texture caches perhaps) so we shouldn't just automatically abort like we do for trivial heap allocations. These public functions now take a CoglError argument so applications can catch out of memory errors: cogl_buffer_map cogl_buffer_map_range cogl_buffer_set_data cogl_framebuffer_read_pixels_into_bitmap cogl_pixel_buffer_new cogl_texture_new_from_data cogl_texture_new_from_bitmap Note: we've been quite conservative with how many apis we let throw OOM CoglErrors since we don't really want to put a burdon on developers to be checking for errors with every cogl api call. So long as there is some lower level api for apps to use that let them catch OOM errors for everything necessary that's enough and we don't have to make more convenient apis more awkward to use. The main focus is on bitmaps and texture allocations since they can be particularly large and prone to failing. A new cogl_attribute_buffer_new_with_size() function has been added in case developers need to catch OOM errors when allocating attribute buffers whereby they can first use _buffer_new_with_size() (which doesn't take a CoglError) followed by cogl_buffer_set_data() which will lazily allocate the buffer storage and report OOM errors. Reviewed-by: Neil Roberts <neil@linux.intel.com> (cherry picked from commit f7735e141ad537a253b02afa2a8238f96340b978) Note: since we can't break the API for Cogl 1.x then actually the main purpose of cherry picking this patch is to keep in-line with changes on the master branch so that we can easily cherry-pick patches. All the api changes relating stable apis released on the 1.12 branch have been reverted as part of cherry-picking this patch so this most just applies all the internal plumbing changes that enable us to correctly propagate OOM errors.
2012-11-08 12:54:10 -05:00
indices = _cogl_buffer_map (COGL_BUFFER (index_buffer),
COGL_BUFFER_ACCESS_READ, 0,
NULL);
indices_type = cogl_indices_get_type (user_indices);
}
else
{
index_buffer = NULL;
indices = NULL;
indices_type = COGL_INDICES_TYPE_UNSIGNED_BYTE;
}
n_lines = get_line_count (mode, n_vertices_in);
/* Note: we are using COGL_INDICES_TYPE_UNSIGNED_INT so 4 bytes per index. */
line_indices = g_malloc (4 * n_lines * 2);
pos = 0;
if (mode == COGL_VERTICES_MODE_TRIANGLES &&
(n_vertices_in % 3) == 0)
{
for (i = 0; i < n_vertices_in; i += 3)
{
add_line (line_indices, base, indices, indices_type, i, i+1, &pos);
add_line (line_indices, base, indices, indices_type, i+1, i+2, &pos);
add_line (line_indices, base, indices, indices_type, i+2, i, &pos);
}
}
else if (mode == COGL_VERTICES_MODE_TRIANGLE_FAN &&
n_vertices_in >= 3)
{
add_line (line_indices, base, indices, indices_type, 0, 1, &pos);
add_line (line_indices, base, indices, indices_type, 1, 2, &pos);
add_line (line_indices, base, indices, indices_type, 0, 2, &pos);
for (i = 3; i < n_vertices_in; i++)
{
add_line (line_indices, base, indices, indices_type, i - 1, i, &pos);
add_line (line_indices, base, indices, indices_type, 0, i, &pos);
}
}
else if (mode == COGL_VERTICES_MODE_TRIANGLE_STRIP &&
n_vertices_in >= 3)
{
add_line (line_indices, base, indices, indices_type, 0, 1, &pos);
add_line (line_indices, base, indices, indices_type, 1, 2, &pos);
add_line (line_indices, base, indices, indices_type, 0, 2, &pos);
for (i = 3; i < n_vertices_in; i++)
{
add_line (line_indices, base, indices, indices_type, i - 1, i, &pos);
add_line (line_indices, base, indices, indices_type, i - 2, i, &pos);
}
}
/* In the journal we are a bit sneaky and actually use GL_QUADS
* which isn't actually a valid CoglVerticesMode! */
#ifdef HAVE_COGL_GL
else if (mode == GL_QUADS && (n_vertices_in % 4) == 0)
{
for (i = 0; i < n_vertices_in; i += 4)
{
add_line (line_indices,
base, indices, indices_type, i, i + 1, &pos);
add_line (line_indices,
base, indices, indices_type, i + 1, i + 2, &pos);
add_line (line_indices,
base, indices, indices_type, i + 2, i + 3, &pos);
add_line (line_indices,
base, indices, indices_type, i + 3, i, &pos);
}
}
#endif
if (user_indices)
cogl_buffer_unmap (COGL_BUFFER (index_buffer));
*n_indices = n_lines * 2;
ret = cogl_indices_new (ctx,
COGL_INDICES_TYPE_UNSIGNED_INT,
line_indices,
*n_indices);
g_free (line_indices);
return ret;
}
static CoglBool
remove_layer_cb (CoglPipeline *pipeline,
int layer_index,
void *user_data)
{
cogl_pipeline_remove_layer (pipeline, layer_index);
return TRUE;
}
static void
pipeline_destroyed_cb (CoglPipeline *weak_pipeline, void *user_data)
{
CoglPipeline *original_pipeline = user_data;
/* XXX: I think we probably need to provide a custom unref function for
* CoglPipeline because it's possible that we will reach this callback
* because original_pipeline is being freed which means cogl_object_unref
* will have already freed any associated user data.
*
* Setting more user data here will *probably* succeed but that may allocate
* a new user-data array which could be leaked.
*
* Potentially we could have a _cogl_object_free_user_data function so
* that a custom unref function could be written that can destroy weak
* pipeline children before removing user data.
*/
cogl_object_set_user_data (COGL_OBJECT (original_pipeline),
&wire_pipeline_key, NULL, NULL);
cogl_object_unref (weak_pipeline);
}
static void
draw_wireframe (CoglContext *ctx,
CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglVerticesMode mode,
int first_vertex,
int n_vertices,
CoglAttribute **attributes,
int n_attributes,
CoglIndices *indices,
CoglDrawFlags flags)
{
CoglIndices *wire_indices;
CoglPipeline *wire_pipeline;
int n_indices;
wire_indices = get_wire_line_indices (ctx,
mode,
first_vertex,
n_vertices,
indices,
&n_indices);
wire_pipeline = cogl_object_get_user_data (COGL_OBJECT (pipeline),
&wire_pipeline_key);
if (!wire_pipeline)
{
wire_pipeline =
_cogl_pipeline_weak_copy (pipeline, pipeline_destroyed_cb, NULL);
cogl_object_set_user_data (COGL_OBJECT (pipeline),
&wire_pipeline_key, wire_pipeline,
NULL);
/* If we have glsl then the pipeline may have an associated
* vertex program and since we'd like to see the results of the
* vertex program in the wireframe we just add a final clobber
* of the wire color leaving the rest of the state untouched. */
if (cogl_has_feature (framebuffer->context, COGL_FEATURE_ID_GLSL))
{
static CoglSnippet *snippet = NULL;
/* The snippet is cached so that it will reuse the program
* from the pipeline cache if possible */
if (snippet == NULL)
{
snippet = cogl_snippet_new (COGL_SNIPPET_HOOK_FRAGMENT,
NULL,
NULL);
cogl_snippet_set_replace (snippet,
"cogl_color_out = "
"vec4 (0.0, 1.0, 0.0, 1.0);\n");
}
cogl_pipeline_add_snippet (wire_pipeline, snippet);
}
else
{
cogl_pipeline_foreach_layer (wire_pipeline, remove_layer_cb, NULL);
cogl_pipeline_set_color4f (wire_pipeline, 0, 1, 0, 1);
}
}
/* temporarily disable the wireframe to avoid recursion! */
flags |= COGL_DRAW_SKIP_DEBUG_WIREFRAME;
_cogl_framebuffer_draw_indexed_attributes (
framebuffer,
wire_pipeline,
COGL_VERTICES_MODE_LINES,
0,
n_indices,
wire_indices,
attributes,
n_attributes,
flags);
COGL_DEBUG_SET_FLAG (COGL_DEBUG_WIREFRAME);
cogl_object_unref (wire_indices);
}
#endif
/* This can be called directly by the CoglJournal to draw attributes
* skipping the implicit journal flush, the framebuffer flush and
* pipeline validation. */
void
_cogl_framebuffer_draw_attributes (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglVerticesMode mode,
int first_vertex,
int n_vertices,
CoglAttribute **attributes,
int n_attributes,
CoglDrawFlags flags)
{
#ifdef COGL_ENABLE_DEBUG
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_WIREFRAME) &&
(flags & COGL_DRAW_SKIP_DEBUG_WIREFRAME) == 0) &&
mode != COGL_VERTICES_MODE_LINES &&
mode != COGL_VERTICES_MODE_LINE_LOOP &&
mode != COGL_VERTICES_MODE_LINE_STRIP)
draw_wireframe (framebuffer->context,
framebuffer, pipeline,
mode, first_vertex, n_vertices,
attributes, n_attributes, NULL,
flags);
else
#endif
{
CoglContext *ctx = framebuffer->context;
ctx->driver_vtable->framebuffer_draw_attributes (framebuffer,
pipeline,
mode,
first_vertex,
n_vertices,
attributes,
n_attributes,
flags);
}
}
void
cogl_framebuffer_draw_attributes (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglVerticesMode mode,
int first_vertex,
int n_vertices,
CoglAttribute **attributes,
int n_attributes)
{
_cogl_framebuffer_draw_attributes (framebuffer,
pipeline,
mode,
first_vertex,
n_vertices,
attributes, n_attributes,
COGL_DRAW_SKIP_LEGACY_STATE);
}
void
cogl_framebuffer_vdraw_attributes (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglVerticesMode mode,
int first_vertex,
int n_vertices,
...)
{
va_list ap;
int n_attributes;
CoglAttribute *attribute;
CoglAttribute **attributes;
int i;
va_start (ap, n_vertices);
for (n_attributes = 0; va_arg (ap, CoglAttribute *); n_attributes++)
;
va_end (ap);
attributes = g_alloca (sizeof (CoglAttribute *) * n_attributes);
va_start (ap, n_vertices);
for (i = 0; (attribute = va_arg (ap, CoglAttribute *)); i++)
attributes[i] = attribute;
va_end (ap);
_cogl_framebuffer_draw_attributes (framebuffer,
pipeline,
mode, first_vertex, n_vertices,
attributes, n_attributes,
COGL_DRAW_SKIP_LEGACY_STATE);
}
void
_cogl_framebuffer_draw_indexed_attributes (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglVerticesMode mode,
int first_vertex,
int n_vertices,
CoglIndices *indices,
CoglAttribute **attributes,
int n_attributes,
CoglDrawFlags flags)
{
#ifdef COGL_ENABLE_DEBUG
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_WIREFRAME) &&
(flags & COGL_DRAW_SKIP_DEBUG_WIREFRAME) == 0) &&
mode != COGL_VERTICES_MODE_LINES &&
mode != COGL_VERTICES_MODE_LINE_LOOP &&
mode != COGL_VERTICES_MODE_LINE_STRIP)
draw_wireframe (framebuffer->context,
framebuffer, pipeline,
mode, first_vertex, n_vertices,
attributes, n_attributes, indices,
flags);
else
#endif
{
CoglContext *ctx = framebuffer->context;
ctx->driver_vtable->framebuffer_draw_indexed_attributes (framebuffer,
pipeline,
mode,
first_vertex,
n_vertices,
indices,
attributes,
n_attributes,
flags);
}
}
void
cogl_framebuffer_draw_indexed_attributes (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglVerticesMode mode,
int first_vertex,
int n_vertices,
CoglIndices *indices,
CoglAttribute **attributes,
int n_attributes)
{
_cogl_framebuffer_draw_indexed_attributes (framebuffer,
pipeline,
mode, first_vertex,
n_vertices, indices,
attributes, n_attributes,
COGL_DRAW_SKIP_LEGACY_STATE);
}
void
Add -Wmissing-declarations to maintainer flags and fix problems This option to GCC makes it give a warning whenever a global function is defined without a declaration. This should catch cases were we've defined a function but forgot to put it in a header. In that case it is either only used within one file so we should make it static or we should declare it in a header. The following changes where made to fix problems: • Some functions were made static • cogl-path.h (the one containing the 1.0 API) was split into two files, one defining the functions and one defining the enums so that cogl-path.c can include the enum and function declarations from the 2.0 API as well as the function declarations from the 1.0 API. • cogl2-clip-state has been removed. This only had one experimental function called cogl_clip_push_from_path but as this is unstable we might as well remove it favour of the equivalent cogl_framebuffer_* API. • The GLX, SDL and WGL winsys's now have a private header to define their get_vtable function instead of directly declaring in the C file where it is called. • All places that were calling COGL_OBJECT_DEFINE need to have the cogl_is_whatever function declared so these have been added either as a public function or in a private header. • Some files that were not including the header containing their function declarations have been fixed to do so. • Any unused error quark functions have been removed. If we later want them we should add them back one by one and add a declaration for them in a header. • _cogl_is_framebuffer has been renamed to cogl_is_framebuffer and made a public function with a declaration in cogl-framebuffer.h • Similarly for CoglOnscreen. • cogl_vdraw_indexed_attributes is called cogl_framebuffer_vdraw_indexed_attributes in the header. The definition has been changed to match the header. • cogl_index_buffer_allocate has been removed. This had no declaration and I'm not sure what it's supposed to do. • CoglJournal has been changed to use the internal CoglObject macro so that it won't define an exported cogl_is_journal symbol. • The _cogl_blah_pointer_from_handle functions have been removed. CoglHandle isn't used much anymore anyway and in the few places where it is used I think it's safe to just use the implicit cast from void* to the right type. • The test-utils.h header for the conformance tests explicitly disables the -Wmissing-declaration option using a pragma because all of the tests declare their main function without a header. Any mistakes relating to missing declarations aren't really important for the tests. • cogl_quaternion_init_from_quaternion and init_from_matrix have been given declarations in cogl-quaternion.h Reviewed-by: Robert Bragg <robert@linux.intel.com>
2012-03-06 13:21:28 -05:00
cogl_framebuffer_vdraw_indexed_attributes (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglVerticesMode mode,
int first_vertex,
int n_vertices,
CoglIndices *indices,
...)
{
va_list ap;
int n_attributes;
CoglAttribute **attributes;
int i;
CoglAttribute *attribute;
va_start (ap, indices);
for (n_attributes = 0; va_arg (ap, CoglAttribute *); n_attributes++)
;
va_end (ap);
attributes = g_alloca (sizeof (CoglAttribute *) * n_attributes);
va_start (ap, indices);
for (i = 0; (attribute = va_arg (ap, CoglAttribute *)); i++)
attributes[i] = attribute;
va_end (ap);
_cogl_framebuffer_draw_indexed_attributes (framebuffer,
pipeline,
mode,
first_vertex,
n_vertices,
indices,
attributes,
n_attributes,
COGL_DRAW_SKIP_LEGACY_STATE);
}
void
_cogl_framebuffer_draw_primitive (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglPrimitive *primitive,
CoglDrawFlags flags)
{
if (primitive->indices)
_cogl_framebuffer_draw_indexed_attributes (framebuffer,
pipeline,
primitive->mode,
primitive->first_vertex,
primitive->n_vertices,
primitive->indices,
primitive->attributes,
primitive->n_attributes,
flags);
else
_cogl_framebuffer_draw_attributes (framebuffer,
pipeline,
primitive->mode,
primitive->first_vertex,
primitive->n_vertices,
primitive->attributes,
primitive->n_attributes,
flags);
}
void
cogl_framebuffer_draw_primitive (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglPrimitive *primitive)
{
_cogl_framebuffer_draw_primitive (framebuffer, pipeline, primitive,
COGL_DRAW_SKIP_LEGACY_STATE);
}
void
cogl_framebuffer_draw_rectangle (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
float x_1,
float y_1,
float x_2,
float y_2)
{
const float position[4] = {x_1, y_1, x_2, y_2};
CoglMultiTexturedRect rect;
/* XXX: All the _*_rectangle* APIs normalize their input into an array of
* _CoglMultiTexturedRect rectangles and pass these on to our work horse;
* _cogl_framebuffer_draw_multitextured_rectangles.
*/
rect.position = position;
rect.tex_coords = NULL;
rect.tex_coords_len = 0;
_cogl_framebuffer_draw_multitextured_rectangles (framebuffer,
pipeline,
&rect,
1,
TRUE);
}
void
cogl_framebuffer_draw_textured_rectangle (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
float x_1,
float y_1,
float x_2,
float y_2,
float s_1,
float t_1,
float s_2,
float t_2)
{
const float position[4] = {x_1, y_1, x_2, y_2};
const float tex_coords[4] = {s_1, t_1, s_2, t_2};
CoglMultiTexturedRect rect;
/* XXX: All the _*_rectangle* APIs normalize their input into an array of
* CoglMultiTexturedRect rectangles and pass these on to our work horse;
* _cogl_framebuffer_draw_multitextured_rectangles.
*/
rect.position = position;
rect.tex_coords = tex_coords;
rect.tex_coords_len = 4;
_cogl_framebuffer_draw_multitextured_rectangles (framebuffer,
pipeline,
&rect,
1,
TRUE);
}
void
cogl_framebuffer_draw_multitextured_rectangle (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
float x_1,
float y_1,
float x_2,
float y_2,
const float *tex_coords,
int tex_coords_len)
{
const float position[4] = {x_1, y_1, x_2, y_2};
CoglMultiTexturedRect rect;
/* XXX: All the _*_rectangle* APIs normalize their input into an array of
* CoglMultiTexturedRect rectangles and pass these on to our work horse;
* _cogl_framebuffer_draw_multitextured_rectangles.
*/
rect.position = position;
rect.tex_coords = tex_coords;
rect.tex_coords_len = tex_coords_len;
_cogl_framebuffer_draw_multitextured_rectangles (framebuffer,
pipeline,
&rect,
1,
TRUE);
}
void
cogl_framebuffer_draw_rectangles (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
const float *coordinates,
unsigned int n_rectangles)
{
CoglMultiTexturedRect *rects;
int i;
/* XXX: All the _*_rectangle* APIs normalize their input into an array of
* CoglMultiTexturedRect rectangles and pass these on to our work horse;
* _cogl_framebuffer_draw_multitextured_rectangles.
*/
rects = g_alloca (n_rectangles * sizeof (CoglMultiTexturedRect));
for (i = 0; i < n_rectangles; i++)
{
rects[i].position = &coordinates[i * 4];
rects[i].tex_coords = NULL;
rects[i].tex_coords_len = 0;
}
_cogl_framebuffer_draw_multitextured_rectangles (framebuffer,
pipeline,
rects,
n_rectangles,
TRUE);
}
void
cogl_framebuffer_draw_textured_rectangles (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
const float *coordinates,
unsigned int n_rectangles)
{
CoglMultiTexturedRect *rects;
int i;
/* XXX: All the _*_rectangle* APIs normalize their input into an array of
* _CoglMultiTexturedRect rectangles and pass these on to our work horse;
* _cogl_framebuffer_draw_multitextured_rectangles.
*/
rects = g_alloca (n_rectangles * sizeof (CoglMultiTexturedRect));
for (i = 0; i < n_rectangles; i++)
{
rects[i].position = &coordinates[i * 8];
rects[i].tex_coords = &coordinates[i * 8 + 4];
rects[i].tex_coords_len = 4;
}
_cogl_framebuffer_draw_multitextured_rectangles (framebuffer,
pipeline,
rects,
n_rectangles,
TRUE);
}
void
cogl_framebuffer_fill_path (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglPath *path)
{
_COGL_RETURN_IF_FAIL (cogl_is_framebuffer (framebuffer));
_COGL_RETURN_IF_FAIL (cogl_is_pipeline (pipeline));
_COGL_RETURN_IF_FAIL (cogl_is_path (path));
_cogl_path_fill_nodes (path, framebuffer, pipeline, 0 /* flags */);
}
void
cogl_framebuffer_stroke_path (CoglFramebuffer *framebuffer,
CoglPipeline *pipeline,
CoglPath *path)
{
_COGL_RETURN_IF_FAIL (cogl_is_framebuffer (framebuffer));
_COGL_RETURN_IF_FAIL (cogl_is_pipeline (pipeline));
_COGL_RETURN_IF_FAIL (cogl_is_path (path));
_cogl_path_stroke_nodes (path, framebuffer, pipeline);
}