mutter/cogl/cogl-journal.c

1818 lines
63 KiB
C
Raw Normal View History

/*
* Cogl
*
* An object oriented GL/GLES Abstraction/Utility Layer
*
* Copyright (C) 2007,2008,2009 Intel Corporation.
*
* 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/>.
*
*
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include "cogl-debug.h"
#include "cogl-internal.h"
#include "cogl-context-private.h"
#include "cogl-journal-private.h"
#include "cogl-texture-private.h"
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
#include "cogl-pipeline-private.h"
#include "cogl-pipeline-opengl-private.h"
#include "cogl-vertex-buffer-private.h"
#include "cogl-framebuffer-private.h"
#include "cogl-profile.h"
#include "cogl-attribute-private.h"
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
#include "cogl-point-in-poly-private.h"
Add internal _cogl_push_source to optionally disable legacy state Some code in Cogl such as when flushing a stencil clip assumes that it can push a temporary simple pipeline to reset to a known state for internal drawing operations. However this breaks down if the application has set any legacy state because that is set globally so it will also get applied to the internal pipeline. _cogl_draw_attributes already had an internal flag to disable applying the legacy state but I think this is quite awkward to use because not all places that push a pipeline draw the attribute buffers directly so it is difficult to pass the flag down through the layers. Conceptually the legacy state is meant to be like a layer on top of the purely pipeline-based state API so I think ideally we should have an internal function to push the source without the applying the legacy state. The legacy state can't be applied as the pipeline is pushed because the global state can be modified even after it is pushed. This patch adds a _cogl_push_source() function which takes an extra boolean flag to mark whether to enable the legacy state. The value of this flag is stored alongside the pipeline in the pipeline stack. Another new internal function called _cogl_get_enable_legacy_state queries whether the top entry in the pipeline stack has legacy state enabled. cogl-primitives and the vertex array drawing code now use this to determine whether to apply the legacy state when drawing. The COGL_DRAW_SKIP_LEGACY_STATE flag is now removed. Reviewed-by: Robert Bragg <robert@linux.intel.com>
2011-09-14 07:17:09 -04:00
#include "cogl-private.h"
#include "cogl1-context.h"
#include <string.h>
#include <gmodule.h>
#include <math.h>
/* XXX NB:
* The data logged in logged_vertices is formatted as follows:
*
* Per entry:
* 4 RGBA GLubytes for the color
* 2 floats for the top left position
* 2 * n_layers floats for the top left texture coordinates
* 2 floats for the bottom right position
* 2 * n_layers floats for the bottom right texture coordinates
*/
#define GET_JOURNAL_ARRAY_STRIDE_FOR_N_LAYERS(N_LAYERS) \
(N_LAYERS * 2 + 2)
/* XXX NB:
* Once in the vertex array, the journal's vertex data is arranged as follows:
* 4 vertices per quad:
* 2 or 3 GLfloats per position (3 when doing software transforms)
* 4 RGBA GLubytes,
* 2 GLfloats per tex coord * n_layers
*
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
* Where n_layers corresponds to the number of pipeline layers enabled
*
* To avoid frequent changes in the stride of our vertex data we always pad
* n_layers to be >= 2
*
* There will be four vertices per quad in the vertex array
*
* When we are transforming quads in software we need to also track the z
* coordinate of transformed vertices.
*
* So for a given number of layers this gets the stride in 32bit words:
*/
#define SW_TRANSFORM (!(COGL_DEBUG_ENABLED \
(COGL_DEBUG_DISABLE_SOFTWARE_TRANSFORM)))
#define POS_STRIDE (SW_TRANSFORM ? 3 : 2) /* number of 32bit words */
#define N_POS_COMPONENTS POS_STRIDE
#define COLOR_STRIDE 1 /* number of 32bit words */
#define TEX_STRIDE 2 /* number of 32bit words */
#define MIN_LAYER_PADING 2
#define GET_JOURNAL_VB_STRIDE_FOR_N_LAYERS(N_LAYERS) \
(POS_STRIDE + COLOR_STRIDE + \
TEX_STRIDE * (N_LAYERS < MIN_LAYER_PADING ? MIN_LAYER_PADING : N_LAYERS))
/* If a batch is longer than this threshold then we'll assume it's not
worth doing software clipping and it's cheaper to program the GPU
to do the clip */
#define COGL_JOURNAL_HARDWARE_CLIP_THRESHOLD 8
typedef struct _CoglJournalFlushState
{
CoglContext *ctx;
CoglJournal *journal;
CoglAttributeBuffer *attribute_buffer;
GArray *attributes;
int current_attribute;
size_t stride;
size_t array_offset;
GLuint current_vertex;
Dynamically load the GL or GLES library The GL or GLES library is now dynamically loaded by the CoglRenderer so that it can choose between GL, GLES1 and GLES2 at runtime. The library is loaded by the renderer because it needs to be done before calling eglInitialize. There is a new environment variable called COGL_DRIVER to choose between gl, gles1 or gles2. The #ifdefs for HAVE_COGL_GL, HAVE_COGL_GLES and HAVE_COGL_GLES2 have been changed so that they don't assume the ifdefs are mutually exclusive. They haven't been removed entirely so that it's possible to compile the GLES backends without the the enums from the GL headers. When using GLX the winsys additionally dynamically loads libGL because that also contains the GLX API. It can't be linked in directly because that would probably conflict with the GLES API if the EGL is selected. When compiling with EGL support the library links directly to libEGL because it doesn't contain any GL API so it shouldn't have any conflicts. When building for WGL or OSX Cogl still directly links against the GL API so there is a #define in config.h so that Cogl won't try to dlopen the library. Cogl-pango previously had a #ifdef to detect when the GL backend is used so that it can sneakily pass GL_QUADS to cogl_vertex_buffer_draw. This is now changed so that it queries the CoglContext for the backend. However to get this to work Cogl now needs to export the _cogl_context_get_default symbol and cogl-pango needs some extra -I flags to so that it can include cogl-context-private.h
2011-07-07 15:44:56 -04:00
CoglIndices *indices;
size_t indices_type_size;
Dynamically load the GL or GLES library The GL or GLES library is now dynamically loaded by the CoglRenderer so that it can choose between GL, GLES1 and GLES2 at runtime. The library is loaded by the renderer because it needs to be done before calling eglInitialize. There is a new environment variable called COGL_DRIVER to choose between gl, gles1 or gles2. The #ifdefs for HAVE_COGL_GL, HAVE_COGL_GLES and HAVE_COGL_GLES2 have been changed so that they don't assume the ifdefs are mutually exclusive. They haven't been removed entirely so that it's possible to compile the GLES backends without the the enums from the GL headers. When using GLX the winsys additionally dynamically loads libGL because that also contains the GLX API. It can't be linked in directly because that would probably conflict with the GLES API if the EGL is selected. When compiling with EGL support the library links directly to libEGL because it doesn't contain any GL API so it shouldn't have any conflicts. When building for WGL or OSX Cogl still directly links against the GL API so there is a #define in config.h so that Cogl won't try to dlopen the library. Cogl-pango previously had a #ifdef to detect when the GL backend is used so that it can sneakily pass GL_QUADS to cogl_vertex_buffer_draw. This is now changed so that it queries the CoglContext for the backend. However to get this to work Cogl now needs to export the _cogl_context_get_default symbol and cogl-pango needs some extra -I flags to so that it can include cogl-context-private.h
2011-07-07 15:44:56 -04:00
CoglPipeline *pipeline;
} CoglJournalFlushState;
typedef void (*CoglJournalBatchCallback) (CoglJournalEntry *start,
int n_entries,
void *data);
typedef CoglBool (*CoglJournalBatchTest) (CoglJournalEntry *entry0,
CoglJournalEntry *entry1);
static void _cogl_journal_free (CoglJournal *journal);
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_OBJECT_INTERNAL_DEFINE (Journal, journal);
static void
_cogl_journal_free (CoglJournal *journal)
{
int i;
if (journal->entries)
g_array_free (journal->entries, TRUE);
if (journal->vertices)
g_array_free (journal->vertices, TRUE);
for (i = 0; i < COGL_JOURNAL_VBO_POOL_SIZE; i++)
if (journal->vbo_pool[i])
cogl_object_unref (journal->vbo_pool[i]);
g_slice_free (CoglJournal, journal);
}
CoglJournal *
_cogl_journal_new (CoglFramebuffer *framebuffer)
{
CoglJournal *journal = g_slice_new0 (CoglJournal);
/* The journal keeps a pointer back to the framebuffer because there
is effectively a 1:1 mapping between journals and framebuffers.
However, to avoid a circular reference the journal doesn't take a
reference unless it is non-empty. The framebuffer has a special
unref implementation to ensure that the journal is flushed when
the journal is the only thing keeping it alive */
journal->framebuffer = framebuffer;
journal->entries = g_array_new (FALSE, FALSE, sizeof (CoglJournalEntry));
journal->vertices = g_array_new (FALSE, FALSE, sizeof (float));
return _cogl_journal_object_new (journal);
}
static void
_cogl_journal_dump_logged_quad (uint8_t *data, int n_layers)
{
size_t stride = GET_JOURNAL_ARRAY_STRIDE_FOR_N_LAYERS (n_layers);
int i;
g_print ("n_layers = %d; rgba=0x%02X%02X%02X%02X\n",
n_layers, data[0], data[1], data[2], data[3]);
data += 4;
for (i = 0; i < 2; i++)
{
float *v = (float *)data + (i * stride);
int j;
g_print ("v%d: x = %f, y = %f", i, v[0], v[1]);
for (j = 0; j < n_layers; j++)
{
float *t = v + 2 + TEX_STRIDE * j;
g_print (", tx%d = %f, ty%d = %f", j, t[0], j, t[1]);
}
g_print ("\n");
}
}
static void
_cogl_journal_dump_quad_vertices (uint8_t *data, int n_layers)
{
size_t stride = GET_JOURNAL_VB_STRIDE_FOR_N_LAYERS (n_layers);
int i;
g_print ("n_layers = %d; stride = %d; pos stride = %d; color stride = %d; "
"tex stride = %d; stride in bytes = %d\n",
n_layers, (int)stride, POS_STRIDE, COLOR_STRIDE,
TEX_STRIDE, (int)stride * 4);
for (i = 0; i < 4; i++)
{
float *v = (float *)data + (i * stride);
uint8_t *c = data + (POS_STRIDE * 4) + (i * stride * 4);
int j;
if (G_UNLIKELY (COGL_DEBUG_ENABLED
(COGL_DEBUG_DISABLE_SOFTWARE_TRANSFORM)))
g_print ("v%d: x = %f, y = %f, rgba=0x%02X%02X%02X%02X",
i, v[0], v[1], c[0], c[1], c[2], c[3]);
else
g_print ("v%d: x = %f, y = %f, z = %f, rgba=0x%02X%02X%02X%02X",
i, v[0], v[1], v[2], c[0], c[1], c[2], c[3]);
for (j = 0; j < n_layers; j++)
{
float *t = v + POS_STRIDE + COLOR_STRIDE + TEX_STRIDE * j;
g_print (", tx%d = %f, ty%d = %f", j, t[0], j, t[1]);
}
g_print ("\n");
}
}
static void
_cogl_journal_dump_quad_batch (uint8_t *data, int n_layers, int n_quads)
{
size_t byte_stride = GET_JOURNAL_VB_STRIDE_FOR_N_LAYERS (n_layers) * 4;
int i;
g_print ("_cogl_journal_dump_quad_batch: n_layers = %d, n_quads = %d\n",
n_layers, n_quads);
for (i = 0; i < n_quads; i++)
_cogl_journal_dump_quad_vertices (data + byte_stride * 2 * i, n_layers);
}
static void
batch_and_call (CoglJournalEntry *entries,
int n_entries,
CoglJournalBatchTest can_batch_callback,
CoglJournalBatchCallback batch_callback,
void *data)
{
int i;
int batch_len = 1;
CoglJournalEntry *batch_start = entries;
if (n_entries < 1)
return;
for (i = 1; i < n_entries; i++)
{
CoglJournalEntry *entry0 = &entries[i - 1];
CoglJournalEntry *entry1 = entry0 + 1;
if (can_batch_callback (entry0, entry1))
{
batch_len++;
continue;
}
batch_callback (batch_start, batch_len, data);
batch_start = entry1;
batch_len = 1;
}
/* The last batch... */
batch_callback (batch_start, batch_len, data);
}
static void
_cogl_journal_flush_modelview_and_entries (CoglJournalEntry *batch_start,
int batch_len,
void *data)
{
CoglJournalFlushState *state = data;
CoglContext *ctx = state->ctx;
CoglFramebuffer *framebuffer = state->journal->framebuffer;
CoglAttribute **attributes;
CoglDrawFlags draw_flags = (COGL_DRAW_SKIP_JOURNAL_FLUSH |
COGL_DRAW_SKIP_PIPELINE_VALIDATION |
COGL_DRAW_SKIP_FRAMEBUFFER_FLUSH |
COGL_DRAW_SKIP_LEGACY_STATE);
COGL_STATIC_TIMER (time_flush_modelview_and_entries,
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
"flush: pipeline+entries", /* parent */
"flush: modelview+entries",
"The time spent flushing modelview + entries",
0 /* no application private data */);
COGL_TIMER_START (_cogl_uprof_context, time_flush_modelview_and_entries);
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_BATCHING)))
g_print ("BATCHING: modelview batch len = %d\n", batch_len);
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_DISABLE_SOFTWARE_TRANSFORM)))
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
_cogl_context_set_current_modelview_entry (ctx,
batch_start->modelview_entry);
attributes = (CoglAttribute **)state->attributes->data;
if (!_cogl_pipeline_get_real_blend_enabled (state->pipeline))
draw_flags |= COGL_DRAW_COLOR_ATTRIBUTE_IS_OPAQUE;
#ifdef HAVE_COGL_GL
Add a GL 3 driver This adds a new CoglDriver for GL 3 called COGL_DRIVER_GL3. When requested, the GLX, EGL and SDL2 winsyss will set the necessary attributes to request a forward-compatible core profile 3.1 context. That means it will have no deprecated features. To simplify the explosion of checks for specific combinations of context->driver, many of these conditionals have now been replaced with private feature flags that are checked instead. The GL and GLES drivers now initialise these private feature flags depending on which driver is used. The fixed function backends now explicitly check whether the fixed function private feature is available which means the GL3 driver will fall back to always using the GLSL progend. Since Rob's latest patches the GLSL progend no longer uses any fixed function API anyway so it should just work. The driver is currently lower priority than COGL_DRIVER_GL so it will not be used unless it is specificly requested. We may want to change this priority at some point because apparently Mesa can make some memory savings if a core profile context is used. In GL 3, getting the combined extensions string with glGetString is deprecated so this patch changes it to use glGetStringi to build up an array of extensions instead. _cogl_context_get_gl_extensions now returns this array instead of trying to return a const string. The caller is expected to free the array. Some issues with this patch: • GL 3 does not support GL_ALPHA format textures. We should probably make this a feature flag or something. Cogl uses this to render text which currently just throws a GL error and breaks so it's pretty important to do something about this before considering the GL3 driver to be stable. • GL 3 doesn't support client side vertex buffers. This probably doesn't matter because CoglBuffer won't normally use malloc'd buffers if VBOs are available, but it might but worth making malloc'd buffers a private feature and forcing it not to use them. • GL 3 doesn't support the default vertex array object. This patch just makes it create and bind a single non-default vertex array object which gets used just like the normal default object. Ideally it would be good to use vertex array objects properly and attach them to a CoglPrimitive to cache the state. Reviewed-by: Robert Bragg <robert@linux.intel.com> (cherry picked from commit 66c9db993595b3a22e63f4c201ea468bc9b88cb6)
2012-09-26 15:32:36 -04:00
if ((ctx->private_feature_flags & COGL_PRIVATE_FEATURE_QUADS))
{
Dynamically load the GL or GLES library The GL or GLES library is now dynamically loaded by the CoglRenderer so that it can choose between GL, GLES1 and GLES2 at runtime. The library is loaded by the renderer because it needs to be done before calling eglInitialize. There is a new environment variable called COGL_DRIVER to choose between gl, gles1 or gles2. The #ifdefs for HAVE_COGL_GL, HAVE_COGL_GLES and HAVE_COGL_GLES2 have been changed so that they don't assume the ifdefs are mutually exclusive. They haven't been removed entirely so that it's possible to compile the GLES backends without the the enums from the GL headers. When using GLX the winsys additionally dynamically loads libGL because that also contains the GLX API. It can't be linked in directly because that would probably conflict with the GLES API if the EGL is selected. When compiling with EGL support the library links directly to libEGL because it doesn't contain any GL API so it shouldn't have any conflicts. When building for WGL or OSX Cogl still directly links against the GL API so there is a #define in config.h so that Cogl won't try to dlopen the library. Cogl-pango previously had a #ifdef to detect when the GL backend is used so that it can sneakily pass GL_QUADS to cogl_vertex_buffer_draw. This is now changed so that it queries the CoglContext for the backend. However to get this to work Cogl now needs to export the _cogl_context_get_default symbol and cogl-pango needs some extra -I flags to so that it can include cogl-context-private.h
2011-07-07 15:44:56 -04:00
/* XXX: it's rather evil that we sneak in the GL_QUADS enum here... */
_cogl_framebuffer_draw_attributes (framebuffer,
state->pipeline,
GL_QUADS,
state->current_vertex, batch_len * 4,
attributes,
state->attributes->len,
draw_flags);
}
Dynamically load the GL or GLES library The GL or GLES library is now dynamically loaded by the CoglRenderer so that it can choose between GL, GLES1 and GLES2 at runtime. The library is loaded by the renderer because it needs to be done before calling eglInitialize. There is a new environment variable called COGL_DRIVER to choose between gl, gles1 or gles2. The #ifdefs for HAVE_COGL_GL, HAVE_COGL_GLES and HAVE_COGL_GLES2 have been changed so that they don't assume the ifdefs are mutually exclusive. They haven't been removed entirely so that it's possible to compile the GLES backends without the the enums from the GL headers. When using GLX the winsys additionally dynamically loads libGL because that also contains the GLX API. It can't be linked in directly because that would probably conflict with the GLES API if the EGL is selected. When compiling with EGL support the library links directly to libEGL because it doesn't contain any GL API so it shouldn't have any conflicts. When building for WGL or OSX Cogl still directly links against the GL API so there is a #define in config.h so that Cogl won't try to dlopen the library. Cogl-pango previously had a #ifdef to detect when the GL backend is used so that it can sneakily pass GL_QUADS to cogl_vertex_buffer_draw. This is now changed so that it queries the CoglContext for the backend. However to get this to work Cogl now needs to export the _cogl_context_get_default symbol and cogl-pango needs some extra -I flags to so that it can include cogl-context-private.h
2011-07-07 15:44:56 -04:00
else
#endif /* HAVE_COGL_GL */
{
if (batch_len > 1)
{
CoglVerticesMode mode = COGL_VERTICES_MODE_TRIANGLES;
int first_vertex = state->current_vertex * 6 / 4;
_cogl_framebuffer_draw_indexed_attributes (framebuffer,
state->pipeline,
mode,
first_vertex,
batch_len * 6,
state->indices,
attributes,
state->attributes->len,
draw_flags);
Dynamically load the GL or GLES library The GL or GLES library is now dynamically loaded by the CoglRenderer so that it can choose between GL, GLES1 and GLES2 at runtime. The library is loaded by the renderer because it needs to be done before calling eglInitialize. There is a new environment variable called COGL_DRIVER to choose between gl, gles1 or gles2. The #ifdefs for HAVE_COGL_GL, HAVE_COGL_GLES and HAVE_COGL_GLES2 have been changed so that they don't assume the ifdefs are mutually exclusive. They haven't been removed entirely so that it's possible to compile the GLES backends without the the enums from the GL headers. When using GLX the winsys additionally dynamically loads libGL because that also contains the GLX API. It can't be linked in directly because that would probably conflict with the GLES API if the EGL is selected. When compiling with EGL support the library links directly to libEGL because it doesn't contain any GL API so it shouldn't have any conflicts. When building for WGL or OSX Cogl still directly links against the GL API so there is a #define in config.h so that Cogl won't try to dlopen the library. Cogl-pango previously had a #ifdef to detect when the GL backend is used so that it can sneakily pass GL_QUADS to cogl_vertex_buffer_draw. This is now changed so that it queries the CoglContext for the backend. However to get this to work Cogl now needs to export the _cogl_context_get_default symbol and cogl-pango needs some extra -I flags to so that it can include cogl-context-private.h
2011-07-07 15:44:56 -04:00
}
else
{
_cogl_framebuffer_draw_attributes (framebuffer,
state->pipeline,
COGL_VERTICES_MODE_TRIANGLE_FAN,
state->current_vertex, 4,
attributes,
state->attributes->len,
draw_flags);
Dynamically load the GL or GLES library The GL or GLES library is now dynamically loaded by the CoglRenderer so that it can choose between GL, GLES1 and GLES2 at runtime. The library is loaded by the renderer because it needs to be done before calling eglInitialize. There is a new environment variable called COGL_DRIVER to choose between gl, gles1 or gles2. The #ifdefs for HAVE_COGL_GL, HAVE_COGL_GLES and HAVE_COGL_GLES2 have been changed so that they don't assume the ifdefs are mutually exclusive. They haven't been removed entirely so that it's possible to compile the GLES backends without the the enums from the GL headers. When using GLX the winsys additionally dynamically loads libGL because that also contains the GLX API. It can't be linked in directly because that would probably conflict with the GLES API if the EGL is selected. When compiling with EGL support the library links directly to libEGL because it doesn't contain any GL API so it shouldn't have any conflicts. When building for WGL or OSX Cogl still directly links against the GL API so there is a #define in config.h so that Cogl won't try to dlopen the library. Cogl-pango previously had a #ifdef to detect when the GL backend is used so that it can sneakily pass GL_QUADS to cogl_vertex_buffer_draw. This is now changed so that it queries the CoglContext for the backend. However to get this to work Cogl now needs to export the _cogl_context_get_default symbol and cogl-pango needs some extra -I flags to so that it can include cogl-context-private.h
2011-07-07 15:44:56 -04:00
}
}
/* DEBUGGING CODE XXX: This path will cause all rectangles to be
* drawn with a coloured outline. Each batch will be rendered with
* the same color. This may e.g. help with debugging texture slicing
* issues, visually seeing what is batched and debugging blending
* issues, plus it looks quite cool.
*/
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_RECTANGLES)))
{
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
static CoglPipeline *outline = NULL;
uint8_t color_intensity;
int i;
CoglAttribute *loop_attributes[1];
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
if (outline == NULL)
outline = cogl_pipeline_new (ctx);
/* The least significant three bits represent the three
components so that the order of colours goes red, green,
yellow, blue, magenta, cyan. Black and white are skipped. The
next two bits give four scales of intensity for those colours
in the order 0xff, 0xcc, 0x99, and 0x66. This gives a total
of 24 colours. If there are more than 24 batches on the stage
then it will wrap around */
color_intensity = 0xff - 0x33 * (ctx->journal_rectangles_color >> 3);
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
cogl_pipeline_set_color4ub (outline,
(ctx->journal_rectangles_color & 1) ?
color_intensity : 0,
(ctx->journal_rectangles_color & 2) ?
color_intensity : 0,
(ctx->journal_rectangles_color & 4) ?
color_intensity : 0,
0xff);
loop_attributes[0] = attributes[0]; /* we just want the position */
for (i = 0; i < batch_len; i++)
_cogl_framebuffer_draw_attributes (framebuffer,
outline,
COGL_VERTICES_MODE_LINE_LOOP,
4 * i + state->current_vertex, 4,
loop_attributes,
1,
draw_flags);
/* Go to the next color */
do
ctx->journal_rectangles_color = ((ctx->journal_rectangles_color + 1) &
((1 << 5) - 1));
/* We don't want to use black or white */
while ((ctx->journal_rectangles_color & 0x07) == 0
|| (ctx->journal_rectangles_color & 0x07) == 0x07);
}
state->current_vertex += (4 * batch_len);
COGL_TIMER_STOP (_cogl_uprof_context, time_flush_modelview_and_entries);
}
static CoglBool
compare_entry_modelviews (CoglJournalEntry *entry0,
CoglJournalEntry *entry1)
{
/* Batch together quads with the same model view 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
return entry0->modelview_entry == entry1->modelview_entry;
}
/* At this point we have a run of quads that we know have compatible
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
* pipelines, but they may not all have the same modelview matrix */
static void
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
_cogl_journal_flush_pipeline_and_entries (CoglJournalEntry *batch_start,
cogl: improves header and coding style consistency We've had complaints that our Cogl code/headers are a bit "special" so this is a first pass at tidying things up by giving them some consistency. These changes are all consistent with how new code in Cogl is being written, but the style isn't consistently applied across all code yet. There are two parts to this patch; but since each one required a large amount of effort to maintain tidy indenting it made sense to combine the changes to reduce the time spent re indenting the same lines. The first change is to use a consistent style for declaring function prototypes in headers. Cogl headers now consistently use this style for prototypes: return_type cogl_function_name (CoglType arg0, CoglType arg1); Not everyone likes this style, but it seems that most of the currently active Cogl developers agree on it. The second change is to constrain the use of redundant glib data types in Cogl. Uses of gint, guint, gfloat, glong, gulong and gchar have all been replaced with int, unsigned int, float, long, unsigned long and char respectively. When talking about pixel data; use of guchar has been replaced with guint8, otherwise unsigned char can be used. The glib types that we continue to use for portability are gboolean, gint{8,16,32,64}, guint{8,16,32,64} and gsize. The general intention is that Cogl should look palatable to the widest range of C programmers including those outside the Gnome community so - especially for the public API - we want to minimize the number of foreign looking typedefs.
2010-02-09 20:57:32 -05:00
int batch_len,
void *data)
{
CoglJournalFlushState *state = data;
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
COGL_STATIC_TIMER (time_flush_pipeline_entries,
"flush: texcoords+pipeline+entries", /* parent */
"flush: pipeline+entries",
"The time spent flushing pipeline + entries",
0 /* no application private data */);
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
COGL_TIMER_START (_cogl_uprof_context, time_flush_pipeline_entries);
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_BATCHING)))
g_print ("BATCHING: pipeline batch len = %d\n", batch_len);
state->pipeline = batch_start->pipeline;
/* If we haven't transformed the quads in software then we need to also break
* up batches according to changes in the modelview matrix... */
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_DISABLE_SOFTWARE_TRANSFORM)))
{
batch_and_call (batch_start,
batch_len,
compare_entry_modelviews,
_cogl_journal_flush_modelview_and_entries,
data);
}
else
_cogl_journal_flush_modelview_and_entries (batch_start, batch_len, data);
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
COGL_TIMER_STOP (_cogl_uprof_context, time_flush_pipeline_entries);
}
static CoglBool
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
compare_entry_pipelines (CoglJournalEntry *entry0, CoglJournalEntry *entry1)
{
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
/* batch rectangles using compatible pipelines */
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
if (_cogl_pipeline_equal (entry0->pipeline,
entry1->pipeline,
(COGL_PIPELINE_STATE_ALL &
~COGL_PIPELINE_STATE_COLOR),
COGL_PIPELINE_LAYER_STATE_ALL,
0))
return TRUE;
else
return FALSE;
}
typedef struct _CreateAttributeState
{
int current;
CoglJournalFlushState *flush_state;
} CreateAttributeState;
static CoglBool
create_attribute_cb (CoglPipeline *pipeline,
int layer_number,
void *user_data)
{
CreateAttributeState *state = user_data;
CoglJournalFlushState *flush_state = state->flush_state;
CoglAttribute **attribute_entry =
&g_array_index (flush_state->attributes,
CoglAttribute *,
state->current + 2);
const char *names[] = {
"cogl_tex_coord0_in",
"cogl_tex_coord1_in",
"cogl_tex_coord2_in",
"cogl_tex_coord3_in",
"cogl_tex_coord4_in",
"cogl_tex_coord5_in",
"cogl_tex_coord6_in",
"cogl_tex_coord7_in"
};
char *name;
/* XXX NB:
* Our journal's vertex data is arranged as follows:
* 4 vertices per quad:
* 2 or 3 floats per position (3 when doing software transforms)
* 4 RGBA bytes,
* 2 floats per tex coord * n_layers
* (though n_layers may be padded; see definition of
* GET_JOURNAL_VB_STRIDE_FOR_N_LAYERS for details)
*/
name = layer_number < 8 ? (char *)names[layer_number] :
g_strdup_printf ("cogl_tex_coord%d_in", layer_number);
/* XXX: it may be worth having some form of static initializer for
* attributes... */
*attribute_entry =
cogl_attribute_new (flush_state->attribute_buffer,
name,
flush_state->stride,
flush_state->array_offset +
(POS_STRIDE + COLOR_STRIDE) * 4 +
TEX_STRIDE * 4 * state->current,
2,
COGL_ATTRIBUTE_TYPE_FLOAT);
if (layer_number >= 8)
g_free (name);
state->current++;
return TRUE;
}
/* Since the stride may not reflect the number of texture layers in use
* (due to padding) we deal with texture coordinate offsets separately
* from vertex and color offsets... */
static void
_cogl_journal_flush_texcoord_vbo_offsets_and_entries (
CoglJournalEntry *batch_start,
cogl: improves header and coding style consistency We've had complaints that our Cogl code/headers are a bit "special" so this is a first pass at tidying things up by giving them some consistency. These changes are all consistent with how new code in Cogl is being written, but the style isn't consistently applied across all code yet. There are two parts to this patch; but since each one required a large amount of effort to maintain tidy indenting it made sense to combine the changes to reduce the time spent re indenting the same lines. The first change is to use a consistent style for declaring function prototypes in headers. Cogl headers now consistently use this style for prototypes: return_type cogl_function_name (CoglType arg0, CoglType arg1); Not everyone likes this style, but it seems that most of the currently active Cogl developers agree on it. The second change is to constrain the use of redundant glib data types in Cogl. Uses of gint, guint, gfloat, glong, gulong and gchar have all been replaced with int, unsigned int, float, long, unsigned long and char respectively. When talking about pixel data; use of guchar has been replaced with guint8, otherwise unsigned char can be used. The glib types that we continue to use for portability are gboolean, gint{8,16,32,64}, guint{8,16,32,64} and gsize. The general intention is that Cogl should look palatable to the widest range of C programmers including those outside the Gnome community so - especially for the public API - we want to minimize the number of foreign looking typedefs.
2010-02-09 20:57:32 -05:00
int batch_len,
void *data)
{
CoglJournalFlushState *state = data;
CreateAttributeState create_attrib_state;
int i;
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
COGL_STATIC_TIMER (time_flush_texcoord_pipeline_entries,
"flush: vbo+texcoords+pipeline+entries", /* parent */
"flush: texcoords+pipeline+entries",
"The time spent flushing texcoord offsets + pipeline "
"+ entries",
0 /* no application private data */);
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
COGL_TIMER_START (_cogl_uprof_context, time_flush_texcoord_pipeline_entries);
/* NB: attributes 0 and 1 are position and color */
for (i = 2; i < state->attributes->len; i++)
cogl_object_unref (g_array_index (state->attributes, CoglAttribute *, i));
g_array_set_size (state->attributes, batch_start->n_layers + 2);
create_attrib_state.current = 0;
create_attrib_state.flush_state = state;
cogl_pipeline_foreach_layer (batch_start->pipeline,
create_attribute_cb,
&create_attrib_state);
batch_and_call (batch_start,
batch_len,
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
compare_entry_pipelines,
_cogl_journal_flush_pipeline_and_entries,
data);
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
COGL_TIMER_STOP (_cogl_uprof_context, time_flush_texcoord_pipeline_entries);
}
static CoglBool
compare_entry_layer_numbers (CoglJournalEntry *entry0, CoglJournalEntry *entry1)
{
if (_cogl_pipeline_layer_numbers_equal (entry0->pipeline, entry1->pipeline))
return TRUE;
else
return FALSE;
}
/* At this point we know the stride has changed from the previous batch
* of journal entries */
static void
_cogl_journal_flush_vbo_offsets_and_entries (CoglJournalEntry *batch_start,
cogl: improves header and coding style consistency We've had complaints that our Cogl code/headers are a bit "special" so this is a first pass at tidying things up by giving them some consistency. These changes are all consistent with how new code in Cogl is being written, but the style isn't consistently applied across all code yet. There are two parts to this patch; but since each one required a large amount of effort to maintain tidy indenting it made sense to combine the changes to reduce the time spent re indenting the same lines. The first change is to use a consistent style for declaring function prototypes in headers. Cogl headers now consistently use this style for prototypes: return_type cogl_function_name (CoglType arg0, CoglType arg1); Not everyone likes this style, but it seems that most of the currently active Cogl developers agree on it. The second change is to constrain the use of redundant glib data types in Cogl. Uses of gint, guint, gfloat, glong, gulong and gchar have all been replaced with int, unsigned int, float, long, unsigned long and char respectively. When talking about pixel data; use of guchar has been replaced with guint8, otherwise unsigned char can be used. The glib types that we continue to use for portability are gboolean, gint{8,16,32,64}, guint{8,16,32,64} and gsize. The general intention is that Cogl should look palatable to the widest range of C programmers including those outside the Gnome community so - especially for the public API - we want to minimize the number of foreign looking typedefs.
2010-02-09 20:57:32 -05:00
int batch_len,
void *data)
{
CoglJournalFlushState *state = data;
CoglContext *ctx = state->journal->framebuffer->context;
size_t stride;
int i;
CoglAttribute **attribute_entry;
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
COGL_STATIC_TIMER (time_flush_vbo_texcoord_pipeline_entries,
"flush: clip+vbo+texcoords+pipeline+entries", /* parent */
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
"flush: vbo+texcoords+pipeline+entries",
"The time spent flushing vbo + texcoord offsets + "
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
"pipeline + entries",
0 /* no application private data */);
COGL_TIMER_START (_cogl_uprof_context,
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
time_flush_vbo_texcoord_pipeline_entries);
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_BATCHING)))
g_print ("BATCHING: vbo offset batch len = %d\n", batch_len);
/* XXX NB:
* Our journal's vertex data is arranged as follows:
* 4 vertices per quad:
* 2 or 3 GLfloats per position (3 when doing software transforms)
* 4 RGBA GLubytes,
* 2 GLfloats per tex coord * n_layers
* (though n_layers may be padded; see definition of
* GET_JOURNAL_VB_STRIDE_FOR_N_LAYERS for details)
*/
stride = GET_JOURNAL_VB_STRIDE_FOR_N_LAYERS (batch_start->n_layers);
stride *= sizeof (float);
state->stride = stride;
for (i = 0; i < state->attributes->len; i++)
cogl_object_unref (g_array_index (state->attributes, CoglAttribute *, i));
g_array_set_size (state->attributes, 2);
attribute_entry = &g_array_index (state->attributes, CoglAttribute *, 0);
*attribute_entry = cogl_attribute_new (state->attribute_buffer,
"cogl_position_in",
stride,
state->array_offset,
N_POS_COMPONENTS,
COGL_ATTRIBUTE_TYPE_FLOAT);
attribute_entry = &g_array_index (state->attributes, CoglAttribute *, 1);
*attribute_entry =
cogl_attribute_new (state->attribute_buffer,
"cogl_color_in",
stride,
state->array_offset + (POS_STRIDE * 4),
4,
COGL_ATTRIBUTE_TYPE_UNSIGNED_BYTE);
Add a GL 3 driver This adds a new CoglDriver for GL 3 called COGL_DRIVER_GL3. When requested, the GLX, EGL and SDL2 winsyss will set the necessary attributes to request a forward-compatible core profile 3.1 context. That means it will have no deprecated features. To simplify the explosion of checks for specific combinations of context->driver, many of these conditionals have now been replaced with private feature flags that are checked instead. The GL and GLES drivers now initialise these private feature flags depending on which driver is used. The fixed function backends now explicitly check whether the fixed function private feature is available which means the GL3 driver will fall back to always using the GLSL progend. Since Rob's latest patches the GLSL progend no longer uses any fixed function API anyway so it should just work. The driver is currently lower priority than COGL_DRIVER_GL so it will not be used unless it is specificly requested. We may want to change this priority at some point because apparently Mesa can make some memory savings if a core profile context is used. In GL 3, getting the combined extensions string with glGetString is deprecated so this patch changes it to use glGetStringi to build up an array of extensions instead. _cogl_context_get_gl_extensions now returns this array instead of trying to return a const string. The caller is expected to free the array. Some issues with this patch: • GL 3 does not support GL_ALPHA format textures. We should probably make this a feature flag or something. Cogl uses this to render text which currently just throws a GL error and breaks so it's pretty important to do something about this before considering the GL3 driver to be stable. • GL 3 doesn't support client side vertex buffers. This probably doesn't matter because CoglBuffer won't normally use malloc'd buffers if VBOs are available, but it might but worth making malloc'd buffers a private feature and forcing it not to use them. • GL 3 doesn't support the default vertex array object. This patch just makes it create and bind a single non-default vertex array object which gets used just like the normal default object. Ideally it would be good to use vertex array objects properly and attach them to a CoglPrimitive to cache the state. Reviewed-by: Robert Bragg <robert@linux.intel.com> (cherry picked from commit 66c9db993595b3a22e63f4c201ea468bc9b88cb6)
2012-09-26 15:32:36 -04:00
if (!(ctx->private_feature_flags & COGL_PRIVATE_FEATURE_QUADS))
state->indices = cogl_get_rectangle_indices (ctx, batch_len);
/* We only create new Attributes when the stride within the
* AttributeBuffer changes. (due to a change in the number of pipeline
* layers) While the stride remains constant we walk forward through
* the above AttributeBuffer using a vertex offset passed to
* cogl_draw_attributes
*/
state->current_vertex = 0;
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_JOURNAL)))
{
uint8_t *verts;
/* Mapping a buffer for read is probably a really bad thing to
do but this will only happen during debugging so it probably
doesn't matter */
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
verts = ((uint8_t *)_cogl_buffer_map (COGL_BUFFER (state->attribute_buffer),
COGL_BUFFER_ACCESS_READ, 0,
NULL) +
state->array_offset);
_cogl_journal_dump_quad_batch (verts,
batch_start->n_layers,
batch_len);
cogl_buffer_unmap (COGL_BUFFER (state->attribute_buffer));
}
batch_and_call (batch_start,
batch_len,
compare_entry_layer_numbers,
_cogl_journal_flush_texcoord_vbo_offsets_and_entries,
data);
/* progress forward through the VBO containing all our vertices */
state->array_offset += (stride * 4 * batch_len);
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_JOURNAL)))
g_print ("new vbo offset = %lu\n", (unsigned long)state->array_offset);
COGL_TIMER_STOP (_cogl_uprof_context,
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
time_flush_vbo_texcoord_pipeline_entries);
}
static CoglBool
compare_entry_strides (CoglJournalEntry *entry0, CoglJournalEntry *entry1)
{
/* Currently the only thing that affects the stride for our vertex arrays
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
* is the number of pipeline layers. We need to update our VBO offsets
* whenever the stride changes. */
/* TODO: We should be padding the n_layers == 1 case as if it were
* n_layers == 2 so we can reduce the need to split batches. */
if (entry0->n_layers == entry1->n_layers ||
(entry0->n_layers <= MIN_LAYER_PADING &&
entry1->n_layers <= MIN_LAYER_PADING))
return TRUE;
else
return FALSE;
}
/* At this point we know the batch has a unique clip stack */
static void
_cogl_journal_flush_clip_stacks_and_entries (CoglJournalEntry *batch_start,
int batch_len,
void *data)
{
CoglJournalFlushState *state = data;
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
CoglFramebuffer *framebuffer = state->journal->framebuffer;
CoglContext *ctx = framebuffer->context;
CoglMatrixStack *projection_stack;
COGL_STATIC_TIMER (time_flush_clip_stack_pipeline_entries,
"Journal Flush", /* parent */
"flush: clip+vbo+texcoords+pipeline+entries",
"The time spent flushing clip + vbo + texcoord offsets + "
"pipeline + entries",
0 /* no application private data */);
COGL_TIMER_START (_cogl_uprof_context,
time_flush_clip_stack_pipeline_entries);
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_BATCHING)))
g_print ("BATCHING: clip stack batch len = %d\n", batch_len);
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
_cogl_clip_stack_flush (batch_start->clip_stack, framebuffer);
/* 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;
/* If we have transformed all our quads at log time then we ensure
* no further model transform is applied by loading the identity
* matrix here. We need to do this after flushing the clip stack
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
* because the clip stack flushing code can modify the current
* modelview matrix entry */
if (G_LIKELY (!(COGL_DEBUG_ENABLED (COGL_DEBUG_DISABLE_SOFTWARE_TRANSFORM))))
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
_cogl_context_set_current_modelview_entry (ctx, &ctx->identity_entry);
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
/* Setting up the clip state can sometimes also update the current
* projection matrix entry so we should update it again. This will have
* no affect if the clip code didn't modify the projection */
projection_stack =
_cogl_framebuffer_get_projection_stack (framebuffer);
_cogl_context_set_current_projection_entry (ctx,
projection_stack->last_entry);
batch_and_call (batch_start,
batch_len,
compare_entry_strides,
_cogl_journal_flush_vbo_offsets_and_entries, /* callback */
data);
COGL_TIMER_STOP (_cogl_uprof_context,
time_flush_clip_stack_pipeline_entries);
}
typedef struct
{
float x_1, y_1;
float x_2, y_2;
} ClipBounds;
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
can_software_clip_entry (CoglJournalEntry *journal_entry,
CoglJournalEntry *prev_journal_entry,
CoglClipStack *clip_stack,
ClipBounds *clip_bounds_out)
{
CoglPipeline *pipeline = journal_entry->pipeline;
CoglClipStack *clip_entry;
int layer_num;
clip_bounds_out->x_1 = -G_MAXFLOAT;
clip_bounds_out->y_1 = -G_MAXFLOAT;
clip_bounds_out->x_2 = G_MAXFLOAT;
clip_bounds_out->y_2 = G_MAXFLOAT;
/* Check the pipeline is usable. We can short-cut here for
entries using the same pipeline as the previous entry */
if (prev_journal_entry == NULL || pipeline != prev_journal_entry->pipeline)
{
/* If the pipeline has a user program then we can't reliably modify
the texture coordinates */
if (cogl_pipeline_get_user_program (pipeline))
return FALSE;
/* If any of the pipeline layers have a texture matrix then we can't
reliably modify the texture coordinates */
for (layer_num = cogl_pipeline_get_n_layers (pipeline) - 1;
layer_num >= 0;
layer_num--)
if (_cogl_pipeline_layer_has_user_matrix (pipeline, layer_num))
return FALSE;
}
/* Now we need to verify that each clip entry's matrix is just a
translation of the journal entry's modelview matrix. We can
also work out the bounds of the clip in modelview space using
this translation */
for (clip_entry = clip_stack; clip_entry; clip_entry = clip_entry->parent)
{
float rect_x1, rect_y1, rect_x2, rect_y2;
CoglClipStackRect *clip_rect;
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
float tx, ty, tz;
CoglMatrixEntry *modelview_entry;
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
clip_rect = (CoglClipStackRect *) clip_entry;
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 = journal_entry->modelview_entry;
if (!_cogl_matrix_entry_calculate_translation (clip_rect->matrix_entry,
modelview_entry,
&tx, &ty, &tz))
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 FALSE;
if (clip_rect->x0 < clip_rect->x1)
{
rect_x1 = clip_rect->x0;
rect_x2 = clip_rect->x1;
}
else
{
rect_x1 = clip_rect->x1;
rect_x2 = clip_rect->x0;
}
if (clip_rect->y0 < clip_rect->y1)
{
rect_y1 = clip_rect->y0;
rect_y2 = clip_rect->y1;
}
else
{
rect_y1 = clip_rect->y1;
rect_y2 = clip_rect->y0;
}
clip_bounds_out->x_1 = MAX (clip_bounds_out->x_1, rect_x1 - tx);
clip_bounds_out->y_1 = MAX (clip_bounds_out->y_1, rect_y1 - ty);
clip_bounds_out->x_2 = MIN (clip_bounds_out->x_2, rect_x2 - tx);
clip_bounds_out->y_2 = MIN (clip_bounds_out->y_2, rect_y2 - ty);
}
if (clip_bounds_out->x_2 <= clip_bounds_out->x_1 ||
clip_bounds_out->y_2 <= clip_bounds_out->y_1)
memset (clip_bounds_out, 0, sizeof (ClipBounds));
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;
}
static void
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
software_clip_entry (CoglJournalEntry *journal_entry,
float *verts,
ClipBounds *clip_bounds)
{
size_t stride =
GET_JOURNAL_ARRAY_STRIDE_FOR_N_LAYERS (journal_entry->n_layers);
float rx1, ry1, rx2, ry2;
float vx1, vy1, vx2, vy2;
int layer_num;
/* Remove the clip on the entry */
_cogl_clip_stack_unref (journal_entry->clip_stack);
journal_entry->clip_stack = NULL;
vx1 = verts[0];
vy1 = verts[1];
vx2 = verts[stride];
vy2 = verts[stride + 1];
if (vx1 < vx2)
{
rx1 = vx1;
rx2 = vx2;
}
else
{
rx1 = vx2;
rx2 = vx1;
}
if (vy1 < vy2)
{
ry1 = vy1;
ry2 = vy2;
}
else
{
ry1 = vy2;
ry2 = vy1;
}
rx1 = CLAMP (rx1, clip_bounds->x_1, clip_bounds->x_2);
ry1 = CLAMP (ry1, clip_bounds->y_1, clip_bounds->y_2);
rx2 = CLAMP (rx2, clip_bounds->x_1, clip_bounds->x_2);
ry2 = CLAMP (ry2, clip_bounds->y_1, clip_bounds->y_2);
/* Check if the rectangle intersects the clip at all */
if (rx1 == rx2 || ry1 == ry2)
/* Will set all of the vertex data to 0 in the hope that this
will create a degenerate rectangle and the GL driver will
be able to clip it quickly */
memset (verts, 0, sizeof (float) * stride * 2);
else
{
if (vx1 > vx2)
{
float t = rx1;
rx1 = rx2;
rx2 = t;
}
if (vy1 > vy2)
{
float t = ry1;
ry1 = ry2;
ry2 = t;
}
verts[0] = rx1;
verts[1] = ry1;
verts[stride] = rx2;
verts[stride + 1] = ry2;
/* Convert the rectangle coordinates to a fraction of the original
rectangle */
rx1 = (rx1 - vx1) / (vx2 - vx1);
ry1 = (ry1 - vy1) / (vy2 - vy1);
rx2 = (rx2 - vx1) / (vx2 - vx1);
ry2 = (ry2 - vy1) / (vy2 - vy1);
for (layer_num = 0; layer_num < journal_entry->n_layers; layer_num++)
{
float *t = verts + 2 + 2 * layer_num;
float tx1 = t[0], ty1 = t[1];
float tx2 = t[stride], ty2 = t[stride + 1];
t[0] = rx1 * (tx2 - tx1) + tx1;
t[1] = ry1 * (ty2 - ty1) + ty1;
t[stride] = rx2 * (tx2 - tx1) + tx1;
t[stride + 1] = ry2 * (ty2 - ty1) + ty1;
}
}
}
static void
maybe_software_clip_entries (CoglJournalEntry *batch_start,
int batch_len,
CoglJournalFlushState *state)
{
CoglContext *ctx;
CoglJournal *journal;
CoglClipStack *clip_stack, *clip_entry;
int entry_num;
/* This tries to find cases where the entry is logged with a clip
but it would be faster to modify the vertex and texture
coordinates rather than flush the clip so that it can batch
better */
/* If the batch is reasonably long then it's worthwhile programming
the GPU to do the clip */
if (batch_len >= COGL_JOURNAL_HARDWARE_CLIP_THRESHOLD)
return;
clip_stack = batch_start->clip_stack;
if (clip_stack == NULL)
return;
/* Verify that all of the clip stack entries are a simple rectangle
clip */
for (clip_entry = clip_stack; clip_entry; clip_entry = clip_entry->parent)
if (clip_entry->type != COGL_CLIP_STACK_RECT)
return;
ctx = state->ctx;
journal = state->journal;
/* This scratch buffer is used to store the translation for each
entry in the journal. We store it in a separate buffer because
it's expensive to calculate but at this point we still don't know
whether we can clip all of the entries so we don't want to do the
rest of the dependant calculations until we're sure we can. */
if (ctx->journal_clip_bounds == NULL)
ctx->journal_clip_bounds = g_array_new (FALSE, FALSE, sizeof (ClipBounds));
g_array_set_size (ctx->journal_clip_bounds, batch_len);
for (entry_num = 0; entry_num < batch_len; entry_num++)
{
CoglJournalEntry *journal_entry = batch_start + entry_num;
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
CoglJournalEntry *prev_journal_entry =
entry_num ? batch_start + (entry_num - 1) : NULL;
ClipBounds *clip_bounds = &g_array_index (ctx->journal_clip_bounds,
ClipBounds, entry_num);
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 (!can_software_clip_entry (journal_entry, prev_journal_entry,
clip_stack,
clip_bounds))
return;
}
/* If we make it here then we know we can software clip the entire batch */
COGL_NOTE (CLIPPING, "Software clipping a batch of length %i", batch_len);
for (entry_num = 0; entry_num < batch_len; entry_num++)
{
CoglJournalEntry *journal_entry = batch_start + entry_num;
float *verts = &g_array_index (journal->vertices, float,
journal_entry->array_offset + 1);
ClipBounds *clip_bounds = &g_array_index (ctx->journal_clip_bounds,
ClipBounds, entry_num);
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
software_clip_entry (journal_entry, verts, clip_bounds);
}
return;
}
static void
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_journal_maybe_software_clip_entries (CoglJournalEntry *batch_start,
int batch_len,
void *data)
{
CoglJournalFlushState *state = data;
COGL_STATIC_TIMER (time_check_software_clip,
"Journal Flush", /* parent */
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
"flush: software clipping",
"Time spent software clipping",
0 /* no application private data */);
COGL_TIMER_START (_cogl_uprof_context,
time_check_software_clip);
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
maybe_software_clip_entries (batch_start, batch_len, state);
COGL_TIMER_STOP (_cogl_uprof_context,
time_check_software_clip);
}
static CoglBool
compare_entry_clip_stacks (CoglJournalEntry *entry0, CoglJournalEntry *entry1)
{
return entry0->clip_stack == entry1->clip_stack;
}
/* Gets a new vertex array from the pool. A reference is taken on the
array so it can be treated as if it was just newly allocated */
static CoglAttributeBuffer *
create_attribute_buffer (CoglJournal *journal,
size_t n_bytes)
{
CoglAttributeBuffer *vbo;
CoglContext *ctx = journal->framebuffer->context;
/* If CoglBuffers are being emulated with malloc then there's not
really any point in using the pool so we'll just allocate the
buffer directly */
if (!(ctx->private_feature_flags & COGL_PRIVATE_FEATURE_VBOS))
return cogl_attribute_buffer_new (ctx, n_bytes, NULL);
vbo = journal->vbo_pool[journal->next_vbo_in_pool];
if (vbo == 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
vbo = cogl_attribute_buffer_new_with_size (ctx, n_bytes);
journal->vbo_pool[journal->next_vbo_in_pool] = vbo;
}
else if (cogl_buffer_get_size (COGL_BUFFER (vbo)) < n_bytes)
{
/* If the buffer is too small then we'll just recreate it */
cogl_object_unref (vbo);
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
vbo = cogl_attribute_buffer_new_with_size (ctx, n_bytes);
journal->vbo_pool[journal->next_vbo_in_pool] = vbo;
}
journal->next_vbo_in_pool = ((journal->next_vbo_in_pool + 1) %
COGL_JOURNAL_VBO_POOL_SIZE);
return cogl_object_ref (vbo);
}
static CoglAttributeBuffer *
upload_vertices (CoglJournal *journal,
const CoglJournalEntry *entries,
int n_entries,
size_t needed_vbo_len,
GArray *vertices)
{
CoglAttributeBuffer *attribute_buffer;
CoglBuffer *buffer;
const float *vin;
float *vout;
int entry_num;
int i;
CoglMatrixEntry *last_modelview_entry = NULL;
CoglMatrix modelview;
g_assert (needed_vbo_len);
attribute_buffer = create_attribute_buffer (journal, needed_vbo_len * 4);
buffer = COGL_BUFFER (attribute_buffer);
cogl_buffer_set_update_hint (buffer, COGL_BUFFER_UPDATE_HINT_STATIC);
vout = _cogl_buffer_map_range_for_fill_or_fallback (buffer,
0, /* offset */
needed_vbo_len * 4);
vin = &g_array_index (vertices, float, 0);
/* Expand the number of vertices from 2 to 4 while uploading */
for (entry_num = 0; entry_num < n_entries; entry_num++)
{
const CoglJournalEntry *entry = entries + entry_num;
size_t vb_stride = GET_JOURNAL_VB_STRIDE_FOR_N_LAYERS (entry->n_layers);
size_t array_stride =
GET_JOURNAL_ARRAY_STRIDE_FOR_N_LAYERS (entry->n_layers);
/* Copy the color to all four of the vertices */
for (i = 0; i < 4; i++)
memcpy (vout + vb_stride * i + POS_STRIDE, vin, 4);
vin++;
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_DISABLE_SOFTWARE_TRANSFORM)))
{
vout[vb_stride * 0] = vin[0];
vout[vb_stride * 0 + 1] = vin[1];
vout[vb_stride * 1] = vin[0];
vout[vb_stride * 1 + 1] = vin[array_stride + 1];
vout[vb_stride * 2] = vin[array_stride];
vout[vb_stride * 2 + 1] = vin[array_stride + 1];
vout[vb_stride * 3] = vin[array_stride];
vout[vb_stride * 3 + 1] = vin[1];
}
else
{
float v[8];
v[0] = vin[0];
v[1] = vin[1];
v[2] = vin[0];
v[3] = vin[array_stride + 1];
v[4] = vin[array_stride];
v[5] = vin[array_stride + 1];
v[6] = vin[array_stride];
v[7] = vin[1];
if (entry->modelview_entry != last_modelview_entry)
_cogl_matrix_entry_get (entry->modelview_entry, &modelview);
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
cogl_matrix_transform_points (&modelview,
2, /* n_components */
sizeof (float) * 2, /* stride_in */
v, /* points_in */
/* strideout */
vb_stride * sizeof (float),
vout, /* points_out */
4 /* n_points */);
}
for (i = 0; i < entry->n_layers; i++)
{
const float *tin = vin + 2;
float *tout = vout + POS_STRIDE + COLOR_STRIDE;
tout[vb_stride * 0 + i * 2] = tin[i * 2];
tout[vb_stride * 0 + 1 + i * 2] = tin[i * 2 + 1];
tout[vb_stride * 1 + i * 2] = tin[i * 2];
tout[vb_stride * 1 + 1 + i * 2] = tin[array_stride + i * 2 + 1];
tout[vb_stride * 2 + i * 2] = tin[array_stride + i * 2];
tout[vb_stride * 2 + 1 + i * 2] = tin[array_stride + i * 2 + 1];
tout[vb_stride * 3 + i * 2] = tin[array_stride + i * 2];
tout[vb_stride * 3 + 1 + i * 2] = tin[i * 2 + 1];
}
vin += array_stride * 2;
vout += vb_stride * 4;
}
_cogl_buffer_unmap_for_fill_or_fallback (buffer);
return attribute_buffer;
}
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_journal_discard (CoglJournal *journal)
{
int i;
if (journal->entries->len <= 0)
return;
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
for (i = 0; i < journal->entries->len; i++)
{
CoglJournalEntry *entry =
&g_array_index (journal->entries, CoglJournalEntry, i);
_cogl_pipeline_journal_unref (entry->pipeline);
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
_cogl_matrix_entry_unref (entry->modelview_entry);
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_unref (entry->clip_stack);
}
g_array_set_size (journal->entries, 0);
g_array_set_size (journal->vertices, 0);
journal->needed_vbo_len = 0;
journal->fast_read_pixel_count = 0;
/* The journal only holds a reference to the framebuffer while the
journal is not empty */
cogl_object_unref (journal->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
}
/* Note: A return value of FALSE doesn't mean 'no' it means
* 'unknown' */
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_journal_all_entries_within_bounds (CoglJournal *journal,
float clip_x0,
float clip_y0,
float clip_x1,
float clip_y1)
{
CoglJournalEntry *entry = (CoglJournalEntry *)journal->entries->data;
CoglClipStack *clip_entry;
CoglClipStack *reference = NULL;
int bounds_x0;
int bounds_y0;
int bounds_x1;
int bounds_y1;
int i;
if (journal->entries->len == 0)
return TRUE;
/* Find the shortest clip_stack ancestry that leaves us in the
* required bounds */
for (clip_entry = entry->clip_stack;
clip_entry;
clip_entry = clip_entry->parent)
{
_cogl_clip_stack_get_bounds (clip_entry,
&bounds_x0, &bounds_y0,
&bounds_x1, &bounds_y1);
if (bounds_x0 >= clip_x0 && bounds_y0 >= clip_y0 &&
bounds_x1 <= clip_x1 && bounds_y1 <= clip_y1)
reference = clip_entry;
else
break;
}
if (!reference)
return FALSE;
/* For the remaining journal entries we will only verify they share
* 'reference' as an ancestor in their clip stack since that's
* enough to know that they would be within the required bounds.
*/
for (i = 1; i < journal->entries->len; i++)
{
CoglBool found_reference = 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
entry = &g_array_index (journal->entries, CoglJournalEntry, i);
for (clip_entry = entry->clip_stack;
clip_entry;
clip_entry = clip_entry->parent)
{
if (clip_entry == reference)
{
found_reference = TRUE;
break;
}
}
if (!found_reference)
return FALSE;
}
return TRUE;
}
/* XXX NB: When _cogl_journal_flush() returns all state relating
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
* to pipelines, all glEnable flags and current matrix state
* is undefined.
*/
void
_cogl_journal_flush (CoglJournal *journal)
{
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
CoglFramebuffer *framebuffer;
CoglContext *ctx;
CoglJournalFlushState state;
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
int i;
COGL_STATIC_TIMER (flush_timer,
"Mainloop", /* parent */
"Journal Flush",
"The time spent flushing the Cogl journal",
0 /* no application private data */);
COGL_STATIC_TIMER (discard_timer,
"Journal Flush", /* parent */
"flush: discard",
"The time spent discarding the Cogl journal after a flush",
0 /* no application private data */);
if (journal->entries->len == 0)
return;
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
framebuffer = journal->framebuffer;
ctx = framebuffer->context;
/* The entries in this journal may depend on images in other
* framebuffers which may require that we flush the journals
* associated with those framebuffers before we can flush
* this journal... */
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
_cogl_framebuffer_flush_dependency_journals (framebuffer);
/* Note: we start the timer after flushing dependency journals so
* that the timer isn't started recursively. */
COGL_TIMER_START (_cogl_uprof_context, flush_timer);
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_BATCHING)))
g_print ("BATCHING: journal len = %d\n", journal->entries->len);
/* NB: the journal deals with flushing the modelview stack and clip
state manually */
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
_cogl_framebuffer_flush_state (framebuffer,
framebuffer,
COGL_FRAMEBUFFER_STATE_ALL &
~(COGL_FRAMEBUFFER_STATE_MODELVIEW |
COGL_FRAMEBUFFER_STATE_CLIP));
state.ctx = ctx;
state.journal = journal;
state.attributes = ctx->journal_flush_attributes_array;
if (G_UNLIKELY ((COGL_DEBUG_ENABLED (COGL_DEBUG_DISABLE_SOFTWARE_CLIP)) == 0))
{
/* We do an initial walk of the journal to analyse the clip stack
batches to see if we can do software clipping. We do this as a
separate walk of the journal because we can modify entries and
this may end up joining together clip stack batches in the next
iteration. */
batch_and_call ((CoglJournalEntry *)journal->entries->data, /* first entry */
journal->entries->len, /* max number of entries to consider */
compare_entry_clip_stacks,
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_journal_maybe_software_clip_entries, /* callback */
&state); /* data */
}
/* We upload the vertices after the clip stack pass in case it
modifies the entries */
state.attribute_buffer =
upload_vertices (journal,
&g_array_index (journal->entries, CoglJournalEntry, 0),
journal->entries->len,
journal->needed_vbo_len,
journal->vertices);
state.array_offset = 0;
/* batch_and_call() batches a list of journal entries according to some
* given criteria and calls a callback once for each determined batch.
*
* The process of flushing the journal is staggered to reduce the amount
* of driver/GPU state changes necessary:
* 1) We split the entries according to the clip state.
* 2) We split the entries according to the stride of the vertices:
* Each time the stride of our vertex data changes we need to call
* gl{Vertex,Color}Pointer to inform GL of new VBO offsets.
* Currently the only thing that affects the stride of our vertex data
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
* is the number of pipeline layers.
* 3) We split the entries explicitly by the number of pipeline layers:
* We pad our vertex data when the number of layers is < 2 so that we
* can minimize changes in stride. Each time the number of layers
* changes we need to call glTexCoordPointer to inform GL of new VBO
* offsets.
* 4) We then split according to compatible Cogl pipelines:
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
* This is where we flush pipeline state
* 5) Finally we split according to modelview matrix changes:
* This is when we finally tell GL to draw something.
* Note: Splitting by modelview changes is skipped when are doing the
* vertex transformation in software at log time.
*/
batch_and_call ((CoglJournalEntry *)journal->entries->data, /* first entry */
journal->entries->len, /* max number of entries to consider */
compare_entry_clip_stacks,
_cogl_journal_flush_clip_stacks_and_entries, /* callback */
&state); /* data */
for (i = 0; i < state.attributes->len; i++)
cogl_object_unref (g_array_index (state.attributes, CoglAttribute *, i));
g_array_set_size (state.attributes, 0);
cogl_object_unref (state.attribute_buffer);
COGL_TIMER_START (_cogl_uprof_context, discard_timer);
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_journal_discard (journal);
COGL_TIMER_STOP (_cogl_uprof_context, discard_timer);
COGL_TIMER_STOP (_cogl_uprof_context, flush_timer);
}
static CoglBool
add_framebuffer_deps_cb (CoglPipelineLayer *layer, void *user_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
{
CoglFramebuffer *framebuffer = user_data;
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
CoglTexture *texture = _cogl_pipeline_layer_get_texture_real (layer);
const GList *l;
if (!texture)
return 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
for (l = _cogl_texture_get_associated_framebuffers (texture); l; l = l->next)
_cogl_framebuffer_add_dependency (framebuffer, l->data);
return 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
}
void
_cogl_journal_log_quad (CoglJournal *journal,
const float *position,
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
CoglPipeline *pipeline,
int n_layers,
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
CoglTexture *layer0_override_texture,
const float *tex_coords,
unsigned int tex_coords_len)
{
CoglFramebuffer *framebuffer = journal->framebuffer;
size_t stride;
int next_vert;
float *v;
int i;
int next_entry;
uint32_t disable_layers;
CoglJournalEntry *entry;
CoglPipeline *final_pipeline;
CoglClipStack *clip_stack;
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
CoglPipelineFlushOptions flush_options;
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
CoglMatrixStack *modelview_stack;
COGL_STATIC_TIMER (log_timer,
"Mainloop", /* parent */
"Journal Log",
"The time spent logging in the Cogl journal",
0 /* no application private data */);
COGL_TIMER_START (_cogl_uprof_context, log_timer);
/* If the framebuffer was previously empty then we'll take a
reference to the current framebuffer. This reference will be
removed when the journal is flushed */
if (journal->vertices->len == 0)
cogl_object_ref (framebuffer);
/* The vertex data is logged into a separate array. The data needs
to be copied into a vertex array before it's given to GL so we
only store two vertices per quad and expand it to four while
uploading. */
/* XXX: See definition of GET_JOURNAL_ARRAY_STRIDE_FOR_N_LAYERS for details
* about how we pack our vertex data */
stride = GET_JOURNAL_ARRAY_STRIDE_FOR_N_LAYERS (n_layers);
next_vert = journal->vertices->len;
g_array_set_size (journal->vertices, next_vert + 2 * stride + 1);
v = &g_array_index (journal->vertices, float, next_vert);
/* We calculate the needed size of the vbo as we go because it
depends on the number of layers in each entry and it's not easy
calculate based on the length of the logged vertices array */
journal->needed_vbo_len += GET_JOURNAL_VB_STRIDE_FOR_N_LAYERS (n_layers) * 4;
/* XXX: All the jumping around to fill in this strided buffer doesn't
* seem ideal. */
/* FIXME: This is a hacky optimization, since it will break if we
* change the definition of CoglColor: */
_cogl_pipeline_get_colorubv (pipeline, (uint8_t *) v);
v++;
memcpy (v, position, sizeof (float) * 2);
memcpy (v + stride, position + 2, sizeof (float) * 2);
for (i = 0; i < n_layers; i++)
{
/* XXX: See definition of GET_JOURNAL_ARRAY_STRIDE_FOR_N_LAYERS
* for details about how we pack our vertex data */
GLfloat *t = v + 2 + i * 2;
memcpy (t, tex_coords + i * 4, sizeof (float) * 2);
memcpy (t + stride, tex_coords + i * 4 + 2, sizeof (float) * 2);
}
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_JOURNAL)))
{
g_print ("Logged new quad:\n");
v = &g_array_index (journal->vertices, float, next_vert);
_cogl_journal_dump_logged_quad ((uint8_t *)v, n_layers);
}
next_entry = journal->entries->len;
g_array_set_size (journal->entries, next_entry + 1);
entry = &g_array_index (journal->entries, CoglJournalEntry, next_entry);
entry->n_layers = n_layers;
entry->array_offset = next_vert;
final_pipeline = pipeline;
flush_options.flags = 0;
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
if (G_UNLIKELY (cogl_pipeline_get_n_layers (pipeline) != n_layers))
{
disable_layers = (1 << n_layers) - 1;
disable_layers = ~disable_layers;
flush_options.disable_layers = disable_layers;
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
flush_options.flags |= COGL_PIPELINE_FLUSH_DISABLE_MASK;
}
if (G_UNLIKELY (layer0_override_texture))
{
cogl: rename CoglMaterial -> CoglPipeline This applies an API naming change that's been deliberated over for a while now which is to rename CoglMaterial to CoglPipeline. For now the new pipeline API is marked as experimental and public headers continue to talk about materials not pipelines. The CoglMaterial API is now maintained in terms of the cogl_pipeline API internally. Currently this API is targeting Cogl 2.0 so we will have time to integrate it properly with other upcoming Cogl 2.0 work. The basic reasons for the rename are: - That the term "material" implies to many people that they are constrained to fragment processing; perhaps as some kind of high-level texture abstraction. - In Clutter they get exposed by ClutterTexture actors which may be re-inforcing this misconception. - When comparing how other frameworks use the term material, a material sometimes describes a multi-pass fragment processing technique which isn't the case in Cogl. - In code, "CoglPipeline" will hopefully be a much more self documenting summary of what these objects represent; a full GPU pipeline configuration including, for example, vertex processing, fragment processing and blending. - When considering the API documentation story, at some point we need a document introducing developers to how the "GPU pipeline" works so it should become intuitive that CoglPipeline maps back to that description of the GPU pipeline. - This is consistent in terminology and concept to OpenGL 4's new pipeline object which is a container for program objects. Note: The cogl-material.[ch] files have been renamed to cogl-material-compat.[ch] because otherwise git doesn't seem to treat the change as a moving the old cogl-material.c->cogl-pipeline.c and so we loose all our git-blame history.
2010-10-27 13:54:57 -04:00
flush_options.flags |= COGL_PIPELINE_FLUSH_LAYER0_OVERRIDE;
flush_options.layer0_override_texture = layer0_override_texture;
}
if (G_UNLIKELY (flush_options.flags))
{
final_pipeline = cogl_pipeline_copy (pipeline);
_cogl_pipeline_apply_overrides (final_pipeline, &flush_options);
}
entry->pipeline = _cogl_pipeline_journal_ref (final_pipeline);
clip_stack = _cogl_framebuffer_get_clip_stack (framebuffer);
entry->clip_stack = _cogl_clip_stack_ref (clip_stack);
if (G_UNLIKELY (final_pipeline != pipeline))
cogl_object_unref (final_pipeline);
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_stack =
_cogl_framebuffer_get_modelview_stack (framebuffer);
entry->modelview_entry = _cogl_matrix_entry_ref (modelview_stack->last_entry);
_cogl_pipeline_foreach_layer_internal (pipeline,
add_framebuffer_deps_cb,
framebuffer);
if (G_UNLIKELY (COGL_DEBUG_ENABLED (COGL_DEBUG_DISABLE_BATCHING)))
_cogl_journal_flush (journal);
COGL_TIMER_STOP (_cogl_uprof_context, log_timer);
}
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
static void
entry_to_screen_polygon (CoglFramebuffer *framebuffer,
const CoglJournalEntry *entry,
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
float *vertices,
float *poly)
{
size_t array_stride =
GET_JOURNAL_ARRAY_STRIDE_FOR_N_LAYERS (entry->n_layers);
CoglMatrixStack *projection_stack;
CoglMatrix projection;
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
CoglMatrix modelview;
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
int i;
float viewport[4];
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
poly[0] = vertices[0];
poly[1] = vertices[1];
poly[2] = 0;
poly[3] = 1;
poly[4] = vertices[0];
poly[5] = vertices[array_stride + 1];
poly[6] = 0;
poly[7] = 1;
poly[8] = vertices[array_stride];
poly[9] = vertices[array_stride + 1];
poly[10] = 0;
poly[11] = 1;
poly[12] = vertices[array_stride];
poly[13] = vertices[1];
poly[14] = 0;
poly[15] = 1;
/* TODO: perhaps split the following out into a more generalized
* _cogl_transform_points utility...
*/
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
_cogl_matrix_entry_get (entry->modelview_entry, &modelview);
cogl_matrix_transform_points (&modelview,
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
2, /* n_components */
sizeof (float) * 4, /* stride_in */
poly, /* points_in */
/* strideout */
sizeof (float) * 4,
poly, /* points_out */
4 /* n_points */);
projection_stack =
_cogl_framebuffer_get_projection_stack (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
_cogl_matrix_stack_get (projection_stack, &projection);
cogl_matrix_project_points (&projection,
3, /* n_components */
sizeof (float) * 4, /* stride_in */
poly, /* points_in */
/* strideout */
sizeof (float) * 4,
poly, /* points_out */
4 /* n_points */);
cogl_framebuffer_get_viewport4fv (framebuffer, viewport);
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
/* Scale from OpenGL normalized device coordinates (ranging from -1 to 1)
* to Cogl window/framebuffer coordinates (ranging from 0 to buffer-size) with
* (0,0) being top left. */
#define VIEWPORT_TRANSFORM_X(x, vp_origin_x, vp_width) \
( ( ((x) + 1.0) * ((vp_width) / 2.0) ) + (vp_origin_x) )
/* Note: for Y we first flip all coordinates around the X axis while in
* normalized device coodinates */
#define VIEWPORT_TRANSFORM_Y(y, vp_origin_y, vp_height) \
( ( ((-(y)) + 1.0) * ((vp_height) / 2.0) ) + (vp_origin_y) )
/* Scale from normalized device coordinates (in range [-1,1]) to
* window coordinates ranging [0,window-size] ... */
for (i = 0; i < 4; i++)
{
float w = poly[4 * i + 3];
/* Perform perspective division */
poly[4 * i] /= w;
poly[4 * i + 1] /= w;
/* Apply viewport transform */
poly[4 * i] = VIEWPORT_TRANSFORM_X (poly[4 * i],
viewport[0], viewport[2]);
poly[4 * i + 1] = VIEWPORT_TRANSFORM_Y (poly[4 * i + 1],
viewport[1], viewport[3]);
}
#undef VIEWPORT_TRANSFORM_X
#undef VIEWPORT_TRANSFORM_Y
}
static CoglBool
try_checking_point_hits_entry_after_clipping (CoglFramebuffer *framebuffer,
CoglJournalEntry *entry,
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
float *vertices,
float x,
float y,
CoglBool *hit)
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 can_software_clip = TRUE;
CoglBool needs_software_clip = 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
CoglClipStack *clip_entry;
*hit = TRUE;
/* Verify that all of the clip stack entries are simple rectangle
* clips */
for (clip_entry = entry->clip_stack;
clip_entry;
clip_entry = clip_entry->parent)
{
if (x < clip_entry->bounds_x0 ||
x >= clip_entry->bounds_x1 ||
y < clip_entry->bounds_y0 ||
y >= clip_entry->bounds_y1)
{
*hit = FALSE;
return TRUE;
}
if (clip_entry->type == COGL_CLIP_STACK_WINDOW_RECT)
{
/* XXX: technically we could still run the software clip in
* this case because for our purposes we know this clip
* can be ignored now, but [can_]sofware_clip_entry() doesn't
* know this and will bail out. */
can_software_clip = FALSE;
}
else if (clip_entry->type == COGL_CLIP_STACK_RECT)
{
CoglClipStackRect *rect_entry = (CoglClipStackRect *)entry;
if (rect_entry->can_be_scissor == FALSE)
needs_software_clip = TRUE;
/* If can_be_scissor is TRUE then we know it's screen
* aligned and the hit test we did above has determined
* that we are inside this clip. */
}
else
return FALSE;
}
if (needs_software_clip)
{
ClipBounds clip_bounds;
float poly[16];
if (!can_software_clip)
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 (!can_software_clip_entry (entry, NULL,
entry->clip_stack, &clip_bounds))
return FALSE;
software_clip_entry (entry, vertices, &clip_bounds);
entry_to_screen_polygon (framebuffer, entry, vertices, poly);
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
util: tune point_in_poly test for polys in screen coords This makes a change to the original point_in_poly algorithm from: http://www.ecse.rpi.edu/Homepages/wrf/Research/Short_Notes/pnpoly.html The aim was to tune the test so that tests against screen aligned rectangles are more resilient to some in-precision in how we transformed that rectangle into screen coordinates. In particular gnome-shell was finding that for some stage sizes then row 0 of the stage would become a dead zone when going through the software picking fast-path and this was because the y position of screen aligned rectangles could end up as something like 0.00024 and the way the algorithm works it doesn't have any epsilon/fuz factor to consider that in-precision. We've avoided introducing an epsilon factor to the comparisons since we feel there's a risk of changing some semantics in ways that might not be desirable. One of those is that if you transform two polygons which share an edge and test a point close to that edge then this algorithm will currently give a positive result for only one polygon. Another concern is the way this algorithm resolves the corner case where the horizontal ray being cast to count edge crossings may cross directly through a vertex. The solution is based on the "idea of Simulation of Simplicity" and "pretends to shift the ray infinitesimally down so that it either clearly intersects, or clearly doesn't touch". I'm not familiar with the idea myself so I expect a misplaced epsilon is likely to break that aspect of the algorithm. The simple solution this patch applies is to pixel align the polygon vertices which should eradicate most noise due to in-precision. https://bugzilla.gnome.org/show_bug.cgi?id=641197
2011-03-03 18:19:30 -05:00
*hit = _cogl_util_point_in_screen_poly (x, y, poly, sizeof (float) * 4, 4);
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 TRUE;
}
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_journal_try_read_pixel (CoglJournal *journal,
int x,
int y,
CoglBitmap *bitmap,
CoglBool *found_intersection)
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
{
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
CoglContext *ctx;
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
int i;
/* XXX: this number has been plucked out of thin air, but the idea
* is that if so many pixels are being read from the same un-changed
* journal than we expect that it will be more efficient to fail
* here so we end up flushing and rendering the journal so that
* further reads can directly read from the framebuffer. There will
* be a bit more lag to flush the render but if there are going to
* continue being lots of arbitrary single pixel reads they will end
* up faster in the end. */
if (journal->fast_read_pixel_count > 50)
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;
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
ctx = _cogl_bitmap_get_context (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 = FALSE;
/* NB: The most recently added journal entry is the last entry, and
* assuming this is a simple scene only comprised of opaque coloured
* rectangles with no special pipelines involved (e.g. enabling
* depth testing) then we can assume painter's algorithm for the
* entries and so our fast read-pixel just needs to walk backwards
* through the journal entries trying to intersect each entry with
* the given point of interest. */
for (i = journal->entries->len - 1; i >= 0; i--)
{
CoglJournalEntry *entry =
&g_array_index (journal->entries, CoglJournalEntry, i);
uint8_t *color = (uint8_t *)&g_array_index (journal->vertices, float,
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
entry->array_offset);
float *vertices = (float *)color + 1;
float poly[16];
CoglFramebuffer *framebuffer = journal->framebuffer;
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;
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
entry_to_screen_polygon (framebuffer, entry, vertices, poly);
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
util: tune point_in_poly test for polys in screen coords This makes a change to the original point_in_poly algorithm from: http://www.ecse.rpi.edu/Homepages/wrf/Research/Short_Notes/pnpoly.html The aim was to tune the test so that tests against screen aligned rectangles are more resilient to some in-precision in how we transformed that rectangle into screen coordinates. In particular gnome-shell was finding that for some stage sizes then row 0 of the stage would become a dead zone when going through the software picking fast-path and this was because the y position of screen aligned rectangles could end up as something like 0.00024 and the way the algorithm works it doesn't have any epsilon/fuz factor to consider that in-precision. We've avoided introducing an epsilon factor to the comparisons since we feel there's a risk of changing some semantics in ways that might not be desirable. One of those is that if you transform two polygons which share an edge and test a point close to that edge then this algorithm will currently give a positive result for only one polygon. Another concern is the way this algorithm resolves the corner case where the horizontal ray being cast to count edge crossings may cross directly through a vertex. The solution is based on the "idea of Simulation of Simplicity" and "pretends to shift the ray infinitesimally down so that it either clearly intersects, or clearly doesn't touch". I'm not familiar with the idea myself so I expect a misplaced epsilon is likely to break that aspect of the algorithm. The simple solution this patch applies is to pixel align the polygon vertices which should eradicate most noise due to in-precision. https://bugzilla.gnome.org/show_bug.cgi?id=641197
2011-03-03 18:19:30 -05:00
if (!_cogl_util_point_in_screen_poly (x, y, poly, sizeof (float) * 4, 4))
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
continue;
if (entry->clip_stack)
{
CoglBool hit;
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 (!try_checking_point_hits_entry_after_clipping (framebuffer,
entry,
vertices,
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
x, y, &hit))
return FALSE; /* hit couldn't be determined */
if (!hit)
continue;
}
*found_intersection = TRUE;
/* If we find that the rectangle the point of interest
* intersects has any state more complex than a constant opaque
* color then we bail out. */
if (!_cogl_pipeline_equal (ctx->opaque_color_pipeline, entry->pipeline,
(COGL_PIPELINE_STATE_ALL &
~COGL_PIPELINE_STATE_COLOR),
COGL_PIPELINE_LAYER_STATE_ALL,
0))
return FALSE;
/* we currently only care about cases where the premultiplied or
* unpremultipled colors are equivalent... */
if (color[3] != 0xff)
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] = color[0];
pixel[1] = color[1];
pixel[2] = color[2];
pixel[3] = color[3];
_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
goto success;
}
success:
journal->fast_read_pixel_count++;
return TRUE;
}