In the arbfp backend there is a seqential approach to finding a suitable
arbfp program to use for a given pipeline; first we see if there's
already a program associated with the pipeline, 2nd we try and find a
program associated with the "arbfp-authority" 3rd we try and lookup a
program in a cache and finally we resort to starting code-generation for
a new program. This patch slightly reworks the code of these steps to
hopefully make them a bit clearer.
_cogl_pipeline_needs_blending_enabled tries to determine whether each
layer is using the default combine state. However it was using
argument 0 for both checks so the if-statement would never be true.
There are a set of "EvalFlags" that get passed to _cogl_pipeline_hash
that can tweak the semantics of what state is evaluated for hashing but
these flags weren't getting passed via the HashState state structure
so it would be undefined if you would get the correct semantics.
According to 9cc9033347 the windows headers #define near as nothing,
and presumable the same is true for 'far' too. Apparently this define is
to improve compatibility with code written for Windows 3.1, so it's good
that people will be able to incorporate such code into their Clutter
applications.
We were trying to declare and initializing an arbfp program cache for
GLES but since the prototypes for the _hash and _equal functions were
only available for GL this broke the GLES builds. By #ifdefing the code
to conditionally declare/initialize for GL only this should hopefully
fix GLES builds.
The constant 'True' is defined by Xlib which isn't used for all clutter
builds so this replaces occurrences of True with TRUE which is defined
by glib. This should hopefully fix the win32 builds.
This adds a cache (A GHashTable) of ARBfp programs and before ever
starting to code-generate a new program we will always first try and
find an existing program in the cache. This uses _cogl_pipeline_hash and
_cogl_pipeline_equal to hash and compare the keys for the cache.
There is a new COGL_DEBUG=disable-program-caches option that can disable
the cache for debugging purposes.
This allows us to get a hash for a set of state groups for a given
pipeline. This can be used for example to get a hash of the fragment
processing state of a pipeline so we can implement a cache for compiled
arbfp/glsl programs.
_cogl_pipeline_equal now accepts a mask of pipeline differences and layer
differences to constrain what state will be compared. In addition a set
of flags are passed that can tweak the comparison semantics for some
state groups. For example when comparing layer textures we sometimes
only need to compare the texture target and can ignore the data itself.
In updating the code this patch also changes it so all required pipeline
authorities are resolved in one step up-front instead of resolving the
authority for each state group in turn and repeatedly having to traverse
the pipeline's ancestry. This adds two new functions
_cogl_pipeline_resolve_authorities and
_cogl_pipeline_layer_resolve_authorities to handle resolving a set of
authorities.
This removes the unused array of per-packend priv data pointers
associated with every CoglPipelineLayer. This reduces the size of all
layer allocations and avoids having to zero an array for each
_cogl_pipeline_layer_copy.
A non-static function named cogl_object_get_type was inadvertently added
during the addition of the CoglObject base type, but there is no public
prototype in the headers and it's only referenced inside cogl-object.c
to implement cogl_handle_get_type() for compatibility. This removes the
function since we don't want to commit to CoglObject always simply being
a boxed type. In the future we may want to register hierarchical
GTypeInstance based types.
To allow us to have gobject properties that accept a CoglMatrix value we
need to register a GType. This adds a cogl_gtype_matrix_get_type function
that will register a static boxed type called "CoglMatrix".
This adds a new section to the reference manual for GType integration
functions.
As a pre-requisite for being able to register a boxed GType for
CoglMatrix (enabling us to define gobject properties that accept a
CoglMatrix) this adds cogl_matrix_copy and _free functions.
In _cogl_pipeline_needs_blending_enabled after first checking whether
the property most recently changed requires blending we would then
resort to checking all other properties too in case some other state
also requires blending. We now avoid checking all other properties in
the case that blending was previously disabled and checking the property
recently changed doesn't require blending.
Note: the plan is to improve this further by explicitly keeping track
of the properties that currently cause blending to be enabled so that we
never have to resort to checking all other properties we can constrain
the checks to those masked properties.
This moves _cogl_pipeline_get_parent and _cogl_pipeline_get_authority
into cogl-pipeline-private.h so they can be inlined since they have been
seen to get quite high in profiles. Given that they both contain such
small amounts of code the function call overhead is significant.
This adds a debug option called disable-software-clipping which causes
the journal to always log the clip stack state rather than trying to
manually clip rectangles.
Before flushing the journal there is now a separate iteration that
will try to determine if the matrix of the clip stack and the matrix
of the rectangle in each entry are on the same plane. If they are it
can completely avoid the clip stack and instead manually modify the
vertex and texture coordinates to implement the clip. The has the
advantage that it won't break up batching if a single clipped
rectangle is used in a scene.
The software clip is only used if there is no user program and no
texture matrices. There is a threshold to the size of the batch where
it is assumed that it is worth the cost to break up a batch and
program the GPU to do the clipping. Currently this is set to 8
although this figure is plucked out of thin air.
To check whether the two matrices are on the same plane it tries to
determine if one of the matrices is just a simple translation of the
other. In the process of this it also works out what the translation
would be. These values can be used to translate the clip rectangle
into the coordinate space of the rectangle to be logged. Then we can
do the clip directly in the rectangle's coordinate space.
Previously in cogl-clip-state.c when it detected that the current
modelview matrix is screen-aligned it would convert the clip entry to
a window clip. Instead of doing this cogl-clip-stack.c now contains
the detection and keeps the entry as a rectangle clip but marks that
it is entirely described by its scissor rect. When flusing the clip
stack it doesn't do anything extra for entries that have this mark
(because the clip will already been setup by the scissor). This is
needed so that we can still track the original rectangle coordinates
and modelview matrix to help detect when it would be faster to modify
the rectangle when adding it to the journal rather than having to
break up the batch to set the clip state.
When logging a quad we now only store the 2 vertices representing the
top left and bottom right of the quad. The color is only stored once
per entry. Once we come to upload the data we expand the 2 vertices
into four and copy the color to each vertex. We do this by mapping the
buffer and directly expanding into it. We have to copy the data before
we can render it anyway so it doesn't make much sense to expand the
vertices before uploading and this way should save some space in the
size of the journal. It also makes it slightly easier if we later want
to do pre-processing on the journal entries before uploading such as
doing software clipping.
The modelview matrix is now always copied to the journal entry whereas
before it would only be copied if we aren't doing software
transform. The journal entry struct always has the space for the
modelview matrix so hopefully it's only a small cost to copy the
matrix.
The transform for the four entries is now done using
cogl_matrix_transform_points which may be slightly faster than
transforming them each individually with a call to
cogl_matrix_transfom.
This reverts commit 4cfe90bde2.
GLSL 1.00 on GLES doesn't support unsized arrays so the whole idea
can't work.
Conflicts:
clutter/cogl/cogl/cogl-pipeline-glsl.c
The check for whether we can reuse a program we've already generated
was only being done if the pipeline already had a
glsl_program_state. When there is no glsl_program_state it then looks
for the nearest ancestor it can share the program with. It then
wasn't checking whether that ancestor already had a GL program so it
would start generating the source again. It wouldn't however compile
that source again because _cogl_pipeline_backend_glsl_end does check
whether there is already a program. This patch moves the check until
after it has found the glsl_program_state, whether or not it was found
from an ancestor or as its own state.
Under GLES2 we were defining the cogl_tex_coord_in varying as an array
with a size determined by the number of texture coordinate arrays
enabled whenever the program is used. This meant that we may have to
regenerate the shader with a different size if the shader is used with
more texture coord arrays later. However in OpenGL the equivalent
builtin varying gl_TexCoord is simply defined as:
varying vec4 gl_TexCoord[]; /* <-- no size */
GLSL is documented that if you declare an array with no size then you
can only access it with a constant index and the size of the array
will be determined by the highest index used. If you want to access it
with a non-constant expression you need to redeclare the array
yourself with a size.
We can replicate the same behaviour in our Cogl shaders by instead
declaring the cogl_tex_coord_in with no size. That way we don't have
to pass around the number of tex coord attributes enabled when we
flush a material. It also means that CoglShader can go back to
directly uploading the source string to GL when cogl_shader_source is
called so that we don't have to keep a copy of it around.
If the user wants to access cogl_tex_coord_in with a non-constant
index then they can simply redeclare the array themself. Hopefully
developers will expect to have to do this if they are accustomed to
the gl_TexCoord array.
When compiling for GLES2, the codegen is affected by state other than
the layers. That means when we find an authority for the codegen state
we can't directly look at authority->n_layers to determine the number
of layers because it isn't necessarily the layer state authority. This
patch changes it to use cogl_pipeline_get_n_layers instead. Once we
have two authorities that differ in codegen state we then compare all
of the layers to decide if they would affect codegen. However it was
ignoring the fact that the authorities might also differ by the other
codegen state. This path also adds an extra check for whether
_cogl_pipeline_compare_differences contains any codegen bits other
than COGL_PIPELINE_STATE_LAYERS.
When determining if a layer would require a different shader to be
generated it needs to check a certain set of state changes and it
needs to check whether the texture target is different. However it was
checking whether texture texture was different only if the other state
was also different which doesn't make any sense. It also only checked
the texture difference if that was the only state change which meant
that effectively the code was impossible to reach. Now it does the
texture target check indepent of the other state changes.
The fixed pipeline backend wasn't correctly flushing the combine
constant because it was using the wrong flag to determine if the
combine constant has changed since the last flushed material.
When enabling a unit that was disabled from a previous flush pipeline
it was forgetting to rebind the right texture unit so it wouldn't
work. This was causing the redhand to disappear when using the fixed
function backend in test-cogl-multitexture if anything else is added
to the scene.
For shader generation backends we don't need to worry about changes to
the texture object and changing the user matrix. The missing user
matrix flag was causing test-cogl-multitexture to regenerate the
shader every frame.
Having ctx here produces a warning on GLES. However it's needed for Big
GL as we have at the top of the file:
#ifdef HAVE_COGL_GL
#define glClientActiveTexture ctx->drv.pf_glClientActiveTexture
#endif
This reverts commit 27a3a2056a.
In 6246c2bd6 I moved the code to add the boilerplate to a shader to a
separate function and also made it so that the common boilerplate is
added as a separate string to glShaderSource. However I didn't notice
that the #define for the vertex and fragment shaders already includes
the common part so it was being added twice. Mesa seems to accept this
but it was causing problems on the IMG driver because COGL_VERSION was
defined twice.
Before commit 49898d43 CoglPipeline would compare whether a pipeline
layer's texture is equal by fetching the underlying GL handle. I
changed that so that it would only compare the CoglHandles because
that commit removes the GL handle texture overrides and sliced
textures instead log the underlying primitive texture. However I
forgot that the primitives don't always use
_cogl_texture_foreach_sub_texture_in_region when the quad fits within
the single texture so it won't use a texture override. This meant that
atlas textures and sub textures get logged with the atlas handle so
the comparison still needs to be done using the GL handles. It might
be nice to add a CoglTexture virtual to get the underlying primitive
texture instead to avoid having the pipeline poke around with GL
handles.
If we have to make override changes to the user's source material to
handle cogl_polygon then we need to make sure we unref the override
material at the end.
Previously we used the layers->backend_priv[] members to determine when
to notify backends about layer changes, but it entirely up to the
backends if they want to associate private state with layers, even
though they may still be interested in layer change notifications (they
may associate layer related state with the owner pipeline).
We now make the observation that in
_cogl_pipeline_backend_layer_change_notify we should be able to assume
there can only be one backend currently associated with the layer
because we wouldn't allow changes to a layer with multiple dependants.
This means we can determine the backend to notify by looking at the
owner pipeline instead.
The features_cached member of CoglContext is intended to mark when
we've calculated the features so that we know if they are ready in
cogl_get_features. However we always intialize the features while
creating the context so features_cached will never be FALSE so it's
not useful. We also had the odd behaviour that the COGL_DEBUG feature
overrides were only applied in the first call to
cogl_get_features. However there are other functions that use the
feature flags such as cogl_features_available that don't use this
function so in some cases the feature flags will be interpreted before
the overrides are applied. This patch makes it always initialize the
features and apply the overrides immediately while creating the
context. This fixes a problem with COGL_DEBUG=disable-arbfp where the
first material flushed is done before any call to cogl_get_features so
it may still use ARBfp.
Now that the GLSL backend can generate code it can effectively handle
any pipeline unless there is an ARBfp program. However with current
open source GL drivers the ARBfp compiler is more stable so it makes
sense to prefer ARBfp when possible. The GLSL backend is also lower
than the fixed function backend on the assumption that any driver that
supports GLSL will also support ARBfp so it's quicker to try the fixed
function backend next.
This adds COGL_DEBUG=disable-fixed to disable the fixed function
pipeline backend. This is needed to test the GLSL shader generation
because otherwise the fixed function backend would always override it.
We don't want to use gl_PointCoord to implement point sprites on big
GL because in that case we already use glTexEnv(GL_COORD_REPLACE) to
replace the texture coords with the point sprite coords. Although GL
also supports the gl_PointCoord variable, it requires GLSL 1.2 which
would mean we would have to declare the GLSL version and check for
it. We continue to use gl_PointCoord for GLES2 because it has no
glTexEnv function.
The GLES2 wrapper no longer needs to generate any fragment shader
state because the GLSL pipeline backend will always give the wrapper a
custom fragment shader. This simplifies a lot of the state comparison
done by the wrapper. The fog generation is also removed even though
it's actually part of the vertex shader because only the fixed
function pipeline backend actually calls the fog functions so it would
be disabled when using any of the other backends anyway. We can fix
this when the two shader backends also start generating vertex
shaders.
GLES2 has no glAlphaFunc function so we need to simulate the behaviour
in the fragment shader. The alpha test function is simulated with an
if-statement and a discard statement. The reference value is stored as
a uniform.
Previously the flag to mark the differences for the alpha test
function and reference value were conflated into one. However this is
awkward when generating shader code to simulate the alpha testing for
GLES 2 because in that case changing the function would need a
different program but changing the reference value just requires
updating a uniform. This patch makes the function and reference have
their own state flags.
The GLSL shader generation supports layer combine constants so there's
no need to disable it for GLES2. It looks like there was also code for
it in the GLES2 wrapper so I'm not sure why it was disabled in the
first place.
The GLSL pipeline backend can now generate code to represent the
pipeline state in a similar way to the ARBfp backend. Most of the code
for this is taken from the GLES 2 wrapper.
_cogl_shader_compile_real had some code to create a set of strings to
combine the boilerplate code with a shader before calling
glShaderSource. This has now been moved to its own internal function
so that it could be used from the GLSL pipeline backend as well.
need_texture_combine_separate is moved to cogl-pipeline.c and renamed
to _cogl_pipeline_need_texture_combine_separate. The function is
needed by both the ARBfp and GLSL codegen backends so it makes sense to
share it.
The code for finding the arbfp authority for a pipeline should be the
same as finding the GLSL authority. So that the code can be shared the
function has been moved to cogl-pipeline.c and renamed to
_cogl_pipeline_find_codegen_authority.
Only one of the material backends can be generating code at the same
time so it seems to make sense to share the same source buffer between
arbfp and glsl. The new name is fragment_source_buffer in case we
later want to create a new buffer for the vertex shader. That probably
couldn't share the same buffer because it will likely need to be
generated at the same time.
* cogl_texture_get_data() is converted to use
_cogl_texture_foreach_sub_texture_in_region() to iterate
through the underlying textures.
* When we need to read only a portion of the underlying
texture, we set up a FBO and use _cogl_read_pixels()
to read the portion we need. This is enormously more
efficient for reading a small portion of a large atlas
texture.
* The CoglAtlasTexture, CoglSubTexture, and CoglTexture2dSliced
implementation of get_texture() are removed.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2414
Previously in cogl_read_pixels we assume the format of the framebuffer
is always premultiplied because that is the most likely format with
the default Cogl blend mode. However when the framebuffer is bound to
a texture we should be able to make a better guess at the format
because we know the texture keeps track of the premult status. This
patch adds an internal format member to CoglFramebuffer. For onscreen
framebuffers we still assume it is RGBA_8888_PRE but for offscreen to
textures we copy the texture format. cogl_read_pixels uses this to
determine whether the data returned by glReadPixels will be
premultiplied.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2414
When converting the data in cogl_read_pixels it was using bmp_format
instead of the format passed in to the function. bmp_format is the
same as the passed in format except that it always has the premult bit
set. Therefore the conversion would not handle premultiply correctly.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2414
This is the same as _cogl_read_pixels except that it takes a rowstride
parameter for the destination buffer. Under OpenGL setting the
rowstride this will end up calling GL_ROW_LENGTH so that the buffer
region can be directly written to. Under GLES GL_ROW_LENGTH is not
supported so it will use an intermediate buffer as it does if the
format is not GL_RGBA.
cogl_read_pixels now just calls the full version of the function with
the rowstride set to width*bpp.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2414
This function is the same as cogl_offscreen_new_to_texture but it
takes a level parameter and a set of flags so that FBOs can be used to
render to higher mipmap levels and to disable the depth and stencil
buffers. cogl_offscreen_new_to_texture now just calls the new function
with the level set to zero. This function could be useful in a few
places in Cogl where we want to use FBOs as an implementation detail
such as when copying between textures.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2414
When uploading a 3D texture with an awkward rowstride, on GLES Cogl
will copy the images to an intermediate buffer to pass to GL. However
it was using the wrong height when copying the data so it would end up
overflowing the buffer and crashing.
Since we're using CoglPipelineWrapModeInternal in the internal API
anyway, and the compiler complains loudly when comparing two enumeration
types without casting, the PipelineLayer struct should store the
wrap modes using the internal enumeration.
When using clip planes and we we have to project some vertices into
screen coordinates we used to transform those by the modelview and then
the projection matrix separately. Now we combine the modelview and
projection matrix and then use that to transform the vertices in one
step instead.
When logging quads in the journal it used to be possible to specify a
mask of fallback layers (layers where a default white texture should be
used in-place of the corresponding texture in the current source
pipeline). Since we now handle fallbacks for cogl_rectangle* primitives
when validating the pipeline up-front before logging in the journal we
no longer need the ability for the journal to apply fallbacks too.
This add two new function that allows us to transform or project an
array of points instead of only transforming one point at a time. Recent
benchmarking has shown cogl_matrix_transform_point to be a bottleneck
sometimes, so this should allow us to reduce the overhead when
transforming lots of vertices at the same time, and also reduce the cost
of 3 component, non-projective transforms.
For now they are marked as experimental (you have to define
COGL_ENABLE_EXPERIMENTAL_API) because there is some concern that it
introduces some inconsistent naming. cogl_matrix_transform_point would
have to be renamed cogl_matrix_project_point to be consistent, but that
would be an API break.
Switch _cogl_rectangles_with_multitexture_coords to using
_cogl_pipeline_foreach_layer to iterate the layers of a pipeline when
validating instead of iterating the pipelines internal list, which is
risky since any modifications to pipelines (even to an override pipeline
derived from the original), could potentially corrupt the list as it is
being iterated.
This removes the possibility to specify wrap mode overrides within a
CoglPipelineFlushOptions struct since the right way to handle these
overrides is by copying the user's material and making the changes to
that copy before flushing. All primitives code has already switched away
from using these wrap mode overrides so this patch just removes unused
code and types. It also remove the wrap_mode_overrides argument for
_cogl_journal_log_quad.
With the refactoring to centralize code into CoglBuffer,
_cogl_buffer_fini() was never actually implemented, so all GL
vertex and index buffer objects were leaked.
The duplicate call to glDeleteBuffers() in CoglPixelArray is
removed (it wasn't paying attention to whether the buffer had been
allocated as a PBO or not.)
http://bugzilla.clutter-project.org/show_bug.cgi?id=2423
This adds a COGL_DEBUG=wireframe option to visualize the underlying
geometry of the primitives being drawn via Cogl. This works for triangle
list, triangle fan, triangle strip and quad (internal only) primitives.
It also works for indexed vertex arrays.
In cogl_vertex_buffer_indices_get_for_quads() we sometimes have to
extend the length of an existing array, but when we came to unref the
previous array we didn't first check that it wasn't simply NULL.
This adds an optional data argument for cogl_vertex_array_new() since it
seems that mostly every case where we use this API we follow up with a
cogl_buffer_set_data() matching the size of the new array. This
simplifies all those cases and whenever we want to delay uploading of
data then NULL can simply be passed.
There's no longer any need to use the GL handle in the callback for
_cogl_texture_foreach_sub_texture_in_region because it can now work in
terms of primitive cogl textures so it has now been removed. This
would be helpful if we ever want to make the foreach function public
so that apps could implement their own primitives using sliced
textures.
Since d5634e37 the sliced texture backend now works in terms of
CoglTexture2Ds so there's no need to have special casing for
overriding the texture of a pipeline layer with a GL handle. Instead
we can just use cogl_pipeline_set_layer_texture with the
CoglHandle. The special _cogl_pipeline_set_layer_gl_texture_slice
function has now been removed and parts of the code for comparing
materials have been simplified.
The cogl_texture_foreach_sub_texture_in_region virtual for the sliced
texture backend was previously passing the CoglHandle of the sliced
texture to the callback. Since d5634e37 the slice texture backend now
works in terms of 2D textures so it's possible to pass the underlying
slice texture as a handle too. This makes all of the foreach callbacks
consistent in that they pass a CoglHandle of the primitive texture
type that matches the GL handle.
When COGL_ENABLE_EXPERIMENTAL_2_0_API is defined cogl.h will now include
cogl2-path.h which changes cogl_path_new() so it can directly return a
CoglPath pointer; it no longer exposes a prototype for
cogl_{get,set}_path and all the remaining cogl_path_ functions now take
an explicit path as their first argument.
The idea is that we want to encourage developers to retain path objects
for as long as possible so they can take advantage of us uploading the
path geometry to the GPU. Currently although it is possible to start a
new path and query the current path, it is not convenient.
The other thing is that we want to get Cogl to the point where nothing
depends on a global, current context variable. This will allow us to one
day define a sensible threading model if/when that is ever desired.
For now this new define is simply an alias for
COGL_ENABLE_EXPERIMENTAL_API but the intention is that we will also use
it to start experimenting with changes that need to break the existing
Cogl API in incompatible ways.
We now prepend a set of defines to any given GLSL shader so that we can
define builtin uniforms/attributes within the "cogl" namespace that we
can use to provide compatibility across a range of the earlier versions
of GLSL.
This updates test-cogl-shader-glsl.c and test-shader.c so they no longer
needs to special case GLES vs GL when splicing together its shaders as
well as the blur, colorize and desaturate effects.
To get a feel for the new, portable uniform/attribute names here are the
defines for OpenGL vertex shaders:
#define cogl_position_in gl_Vertex
#define cogl_color_in gl_Color
#define cogl_tex_coord_in gl_MultiTexCoord0
#define cogl_tex_coord0_in gl_MultiTexCoord0
#define cogl_tex_coord1_in gl_MultiTexCoord1
#define cogl_tex_coord2_in gl_MultiTexCoord2
#define cogl_tex_coord3_in gl_MultiTexCoord3
#define cogl_tex_coord4_in gl_MultiTexCoord4
#define cogl_tex_coord5_in gl_MultiTexCoord5
#define cogl_tex_coord6_in gl_MultiTexCoord6
#define cogl_tex_coord7_in gl_MultiTexCoord7
#define cogl_normal_in gl_Normal
#define cogl_position_out gl_Position
#define cogl_point_size_out gl_PointSize
#define cogl_color_out gl_FrontColor
#define cogl_tex_coord_out gl_TexCoord
#define cogl_modelview_matrix gl_ModelViewMatrix
#define cogl_modelview_projection_matrix gl_ModelViewProjectionMatrix
#define cogl_projection_matrix gl_ProjectionMatrix
#define cogl_texture_matrix gl_TextureMatrix
And for fragment shaders we have:
#define cogl_color_in gl_Color
#define cogl_tex_coord_in gl_TexCoord
#define cogl_color_out gl_FragColor
#define cogl_depth_out gl_FragDepth
#define cogl_front_facing gl_FrontFacing
When converting the virtual coordinates of the underlying texture for
a slice to virtual coordinates for the whole texture it was using the
size and offset of the intersection as the size of the child
texture. This would be incorrect if the texture contains waste or the
texture coordinates are not the default. Instead the sliced foreach
function now passes the CoglSpan to the callback instead of the
intersection.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2398
Previously in the tests/tools directory we build a disable-npots
library which was used as an LD_PRELOAD to trick Cogl in to thinking
there is no NPOT texture extension. This is a little awkward to use so
it seems much simpler to just define a COGL_DEBUG option to disable
npot textures.
In some micro-benchmarks testing journal throughput the list
manipulation jumps pretty high in the profile. This replaces the GSList
usage with a GArray instead which is effectively a grow only allocation
that means we avoid ongoing allocations while manipulating the stack
mid-scene.
During _cogl_pipeline_needs_blending_enabled we were always checking the
current lighting properties (ambient,diffuse,specular,emission) which
had a notable impact during micro-benchmarks that exercise journal
throughput of simple colored rectangles. This #if 0's the offending code
considering that Cogl doesn't actually support lighting currently and
when it actually does then we will be able to optimize this by avoiding
the checks when lighting is disabled.
When using cogl_set_source_color4ub there is a notable difference
between colors that require blending and those that dont. When trying to
modify the color of pipeline referenced by the journal we don't force a
flush of the journal unless the color change will also change the
blending state. By using two separate pipeline objects for handing
opaque or transparent colors we can avoid ever flushing the journal when
repeatedly using cogl_set_source_color and jumping between opaque and
transparent colors.
This reworks _cogl_texture_quad_multiple_primitives so instead of using
the CoglPipelineWrapModeOverrides mechanism to force the clamp to edge
repeat mode we now derive an override pipeline using cogl_pipeline_copy
instead. This avoids a relatively large, unconditional, memset.
This avoids using the wrap mode overrides mechanism to implement
_cogl_multitexture_quad_single_primitive which requires memsetting a
fairly large array. This updates it to use cogl_pipeline_foreach_layer()
and we now derive an override_material to handle changes to the wrap
modes instead of using the CoglPipelineWrapModeOverrides.
Previously there was a check to avoid filling the path if there are
zero nodes. However the tesselator also won't generate any triangles
if there are less than 3 nodes so we might as well bail out in that
case too. If we don't emit any triangles then we would end up trying
to create an empty VBO. Although I don't think this should necessarily
be a problem, this seems to cause Mesa to segfault in version 7.8.1
when calling glBufferSubData (although not in
master). test-cogl-primitives tries to fill a path with only two
points so it's convenient to be able to avoid the crash in this case.
When adding a new entry to the journal a reference is now taken on the
current clip stack. Modifying the current clip state no longer causes
a journal flush. The journal flushing code now has an extra stage to
compare the clip state of each entry. The comparison can simply be
done by comparing the pointers. Although different clip states will
still end up with multiple draw calls this at leasts allows a scene
comprising of multiple different clips to be upload with one vbo. It
also lays the groundwork to do certain tricks when drawing clipped
rectangles such as modifying the geometry instead of setting a clip
state.
This adds a flag to avoid flushing the clip state when flushing the
framebuffer state. This will be used by the journal to manage its own
clip state flushing.
Flushing the clip state no longer does anything that would cause the
journal to flush. The clip state is only flushed when flushing the
framebuffer state and in all cases this ends up flushing the journal
in one way or another anyway. Avoiding flushing the journal will make
it easier to log the clip state in the journal.
Previously when trying to set up a rectangle clip that can't be
scissored or when using a path clip the code would use cogl_rectangle
as part of the process to fill the stencil buffer. This is now changed
to use a new internal _cogl_rectangle_immediate function which
directly uses the vertex array API to draw a triangle strip without
affecting the journal. This should be just as efficient as the
previous journalled code because these places would end up flushing
the journal immediately before and after submitting the single
rectangle anyway and flushing the journal always creates a new vbo so
it would effectively do the same thing.
Similarly there is also a new internal _cogl_clear function that does
not flush the journal.
Previously we tracked whether the clip stack needs flushing as part of
the CoglClipState which is part of the CoglFramebuffer state. This is
a bit odd because most of the clipping state (such as the clip planes
and the scissor) are part of the GL context's state rather than the
framebuffer. We were marking the clip state on the framebuffer dirty
every time we change the framebuffer anyway so it seems to make more
sense to have the dirtiness be part of the global context.
Instead of a just a single boolean to record whether the state needs
flushing, the CoglContext now holds a reference to the clip stack that
was flushed. That way we can flush arbitrary stack states and if it
happens to be the same as the state already flushed then Cogl will do
nothing. This will be useful if we log the clip stack in the journal
because then we will need to flush unrelated clip stack states for
each batch.
Instead of having a separate CoglHandle for CoglClipStack the code is
now expected to directly hold a pointer to the top entry on the
stack. The empty stack is then the NULL pointer. This saves an
allocation when we want to copy the stack because we can just take a
reference on a stack entry. The idea is that this will make it
possible to store the clip stack in the journal without any extra
allocations.
The _cogl_get_clip_stack and set functions now take a CoglClipStack
pointer instead of a handle so it would no longer make sense to make
them public. However I think the only reason we would have wanted that
in the first place would be to save the clip state between switching
FBOs and that is no longer necessary.
CoglVertexAttribute has an internal draw function that is used by the
CoglJournal to avoid the call to cogl_journal_flush which would
otherwise end up recursively flushing the journal forever. The
enable_gl_state function called by this was previously also calling
_cogl_flush_framebuffer_state. However the journal code tries to
handle this function specially by calling it with a flag to disable
flushing the modelview matrix. This is useful because the journal
handles flushing the modelview itself. Without this patch the journal
state ends up getting flushed twice. This isn't a particularly big
problem currently because the matrix stack has caching to recognise
when it would push the same state twice and bails out. However if we
later want to use the framebuffer flush flags to override a particular
state of the framebuffer (such as the clip state) then we need to make
sure the flush isn't called twice.
Unless the CoglBuffer is being used for texture data then it's
relatively unlikely that the data will contain an array of bytes. For
example if it's used as a vertex array then it's more likely to be
floats or some vertex struct. In that case it's much more convenient
if set_data and map use void* pointers so that we can avoid a cast.
The convenience constructors for the builtin vertex structs were
creating the primitive and then immediately destroying it and
returning the pointer. I think the intention was to unref the
attributes instead. This adds an internal wrapper around the
new_with_attributes_array constructor which unrefs the attributes
instead of the primitive. The convenience constructors now use that.
The GLES2 wrapper was referring to COGL_MATERIAL_PROGRAM_TYPE_GLSL but
this has since been renamed to COGL_PIPELINE_PROGRAM_TYPE_GLSL so the
GLES2 backend wouldn't compile.
The gles2 wrapper functions don't understand about the CoglBuffer API so
they don't support attributes stored in a CoglVertexArray. Instead of
teaching the backend about buffers we are going to wait until we have
overhauled the GLES 2 backend. We are currently making progress
consolidating the GLES 2 backend with a new GLSL backend for
CoglMaterial. This will hugely simplify the GLES 2 support and share
code with the OpenGL backend. In the end it's hoped that this problem
will simply go away so it doesn't make much sense to solve it with the
current design.
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.
This updates the implementation of cogl_polygon so it sits on the new
CoglVertexArray and CoglVertexAttribute apis. This lets us minimize the
number of different drawing paths we have to maintain in Cogl.
Since the sliced texture support for cogl_polygon has been broken for a
long time now and no one has complained this patch also greatly
simplifies the code by not doing any special material validation so
cogl_polygon will be restricted in the same way as
cogl_draw_vertex_attributes. (i.e. sliced textures not supported).
Instead of using raw OpenGL in the journal we now use the vertex
attributes API instead. This is part of an ongoing effort to reduce the
number of drawing paths we maintain in Cogl.
The functionality of cogl_vertex_buffer_indices_get_for_quads is now
provided by cogl_get_rectangle_indices so this reworks the former to now
work in terms of the latter so we don't have duplicated logic.
As part of an ongoing effort to reduce the number of draw paths we have
in Cogl this re-works CoglVertexBuffer to use the CoglVertexAttribute
and CoglPrimitive APIs instead of using raw GL.
This adds a way to mark that a primitive is in use so that modifications
will generate a warning. The plan is to use this mechanism when batching
primitives in the journal to warn users that mid-scene modifications of
primitives is not allowed.
This adds convenience primitive constructors named like:
cogl_primitive_new_p3 or
cogl_primitive_new_p3c4 or
cogl_primitive_new_p3t2c4
where the letters correspond to the interleved vertex attributes layouts
such as CoglP3Vertex which is a struct with 3 float x,y,z members for
the [p]osition, or CoglP3T2C4Vertex which is a struct with 3 float x,y,z
members for the [p]osition, 2 float s,t members for the [t]exture
coordinates and 4 unsigned byte r,g,b,a members for the [c]olor.
The hope is that people will find these convenient enough to replace
cogl_polygon.
A CoglPrimitive is a retainable object for drawing a single primitive,
such as a triangle strip, fan or list.
CoglPrimitives build on CoglVertexAttributes and CoglIndices which
themselves build on CoglVertexArrays and CoglIndexArrays respectively.
A CoglPrimitive encapsulates enough information such that it can be
retained in a queue (e.g. the Cogl Journal, or renderlists in the
future) and drawn at some later time.
A CoglVertexAttribute defines a single attribute contained in a
CoglVertexArray. I.e. a CoglVertexArray is simply a buffer of N bytes
intended for containing a collection of attributes (position, color,
normals etc) and a CoglVertexAttribute defines one such attribute by
specifying its start offset in the array, its type, the number of
components and the stride etc.
CoglIndices define a range of indices inside a CoglIndexArray. I.e. a
CoglIndexArray is simply a buffer of N bytes and you can then
instantiate multiple CoglIndices collections that define a sub-region of
a CoglIndexArray by specifying a start offset and an index data type.
This adds a new CoglVertexArray object which is a subclass of CoglBuffer
used to hold vertex attributes. A later commit will add a
CoglVertexAttribute API which will be used to describe the attributes
inside a CoglVertexArray.
A CoglIndexArray is a subclass of CoglBuffer and will be used to hold
vertex indices. A later commit will add a CoglIndices API which will
allow describing a range of indices inside a CoglIndexArray.
This adds an internal mechanism to mark that a buffer is in-use so that
a warning can be generated if the user attempts to modify the buffer.
The plans is for the journal to use this mechanism so that we can warn
users about mid-scene modifications of buffers.
We now make _cogl_buffer_bind return a base pointer for the bound buffer
which can be used with OpenGL. The pointer will be NULL for GPU based
buffers or may point to an malloc'd buffer. Since OpenGL expects an
offset instead of a pointer when dealing with buffer objects this means
we can handle fallback malloc buffers and GPU buffers in a consistent
way.
This allows _cogl_material_flush_gl_state to bail out faster if
repeatedly asked to flush the same material and we can see the material
hasn't changed.
Since we can rely on the material age incrementing when any material
property changes or any associated layer property changes then we can
track the age of the material after flushing so it can be compared with
the age of the material if it is subsequently re-flushed. If the age is
the same we only have to re-assert the texture object state.
MaterialNodes are used for the sparse graph of material state and layer
state. In the case of materials there is the idea of weak materials that
don't take a reference on their parent and in that case we need to be
careful not to unref our parent during
_cogl_material_node_unparent_real. This adds a has_parent_reference
member to the CoglMaterialNode struct so we now know when to skip the
unref.
If there is private data associated with a CoglObject then there may be
a user_data_array that needs to be freed. The code was mistakenly
freeing the array inside the loop that was actually iterating over the
user data array notifying the objects destruction instead of waiting
until all the data entries had been destroyed.
This merges the two implementations of CoglProgram for the GLES2 and
GL backends into one. The implementation is more like the GLES2
version which would track the uniform values and delay sending them to
GL. CoglProgram is now effectively just a GList of CoglShaders along
with an array of stored uniform values. CoglProgram never actually
creates a GL program, instead this is left up to the GLSL material
backend. This is necessary on GLES2 where we may need to relink the
user's program with different generated shaders depending on the other
emulated fixed function state. It will also be necessary in the future
GLSL backends for regular OpenGL. The GLSL and ARBfp material backends
are now the ones that create and link the GL program from the list of
shaders. The linked program is attached to the private material state
so that it can be reused if the CoglProgram is used again with the
same material. This does mean the program will get relinked if the
shader is used with multiple materials. This will be particularly bad
if the legacy cogl_program_use function is used because that
effectively always makes one-shot materials. This problem will
hopefully be alleviated if we make a hash table with a cache of
generated programs. The cogl program would then need to become part of
the hash lookup.
Each CoglProgram now has an age counter which is incremented every
time a shader is added. This is used by the material backends to
detect when we need to create a new GL program for the user program.
The internal _cogl_use_program function now takes a GL program handle
rather than a CoglProgram. It no longer needs any special differences
for GLES2. The GLES2 wrapper function now also uses this function to
bind its generated shaders.
The ARBfp shaders no longer store a copy of the program source but
instead just directly create a program object when cogl_shader_source
is called. This avoids having to reupload the source if the same
shader is used in multiple materials.
There are currently a few gross hacks to get the GLES2 backend to work
with this. The problem is that the GLSL material backend is now
generating a complete GL program but the GLES2 wrapper still needs to
add its fixed function emulation shaders if the program doesn't
provide either a vertex or fragment shader. There is a new function in
the GLES2 wrapper called _cogl_gles2_use_program which replaces the
previous cogl_program_use implementation. It extracts the GL shaders
from the GL program object and creates a new GL program containing all
of the shaders plus its fixed function emulation. This new program is
returned to the GLSL material backend so that it can still flush the
custom uniforms using it. The user_program is attached to the GLES2
settings struct as before but its stored using a GL program handle
rather than a CoglProgram pointer. This hack will go away once the
GLSL material backend replaces the GLES2 wrapper by generating the
code itself.
Under Mesa this currently generates some GL errors when glClear is
called in test-cogl-shader-glsl. I think this is due to a bug in Mesa
however. When the user program on the material is changed the GLSL
backend gets notified and deletes the GL program that it linked from
the user shaders. The program will still be bound in GL
however. Leaving a deleted shader bound exposes a bug in Mesa's
glClear implementation. More details are here:
https://bugs.freedesktop.org/show_bug.cgi?id=31194
Previously cogl_set_source_color and cogl_set_source_texture modified
a single global material. If an application then mixes using
cogl_set_source_color and texture then the material will constantly
need a new ARBfp program because the numbers of layers alternates
between 0 and 1. This patch just adds a second global material that is
only used for cogl_set_source_texture. I think it would still end up
flushing the journal if cogl_set_source_texture is used with multiple
different textures but at least it should avoid a recompile unless the
texture target also changes. It might be nice to somehow attach a
material to the CoglTexture for use with cogl_set_source_texture but
it would be difficult to implement this without creating a circular
reference.
This moves the CoglIndicesType and CoglVerticesMode typedefs from
cogl-vertex-buffer.h to cogl-types.h so they can be shared with the
anticipated cogl vertex attribute API.
This renames the BufferBindTarget + BufferUsageHint enums to match the
anticipated new APIs for "index arrays" and "vertex arrays" as opposed
to using the terms "vertices" or "indices".
previously we would silently bail out if the given offset + data size
would overflow the buffer size. Now we use g_return_val_if_fail so we
get a warning if we hit this case.
This adds a store_created bit field to CoglBuffer so we know if the
underlying buffer has been allocated yet. Previously the code was trying
to do something really wrong by accidentally using the
COGL_PIXEL_ARRAY_FLAG_IS_SET macro (note "PIXEL_ARRAY") and what is more
odd was the declaration of a CoglPixelArray *pixel_array in
cogl-buffer.c which the buffer was being cast too before calling using
the macro. Probably this was the fall-out of some previous code
re-factoring.
All the macros get used for are to |= (a new flag bit), &= ~(a flag bit)
or use the & operator to test if a flag bit is set. I haven't found the
code more readable with these macros, but several times now I've felt
the need to double check if these macros do anything else behind the
hood or I've forgotten what flags are available so I've had to go to the
macro definition to see what the full enum names are for the flags (the
macros use symbol concatenation) so I can search for the definition of
all the flags. It turns out they are defined next to the macro so you
don't have to search far, but without the macro that wouldn't have been
necessary.
The more common use of the _IS_SET macro is actually more concise
expanded and imho since it doesn't hide anything in a separate header
file the code is more readable without the macro.
This is a counter part for _cogl_material_layer_get_texture which takes
a layer index instead of a direct CoglMaterialLayer pointer. The aim is
to phase out code that directly iterates the internal layer pointers of
a material since the layer pointers can change if any property of any
layer is changed making direct layer pointers very fragile.
This adds internal _cogl_material_get_layer_filters and
_cogl_material_get_layer_{min,mag}_filter functions which can be used to
query the filters associated with a layer using a layer_index, as
opposed to a layer pointer. Accessing layer pointers is considered
deprecated so we need to provide layer_index based replacements.
When we come to submitting the users given attributes we sort them into
different types of buffers. Previously we had three types; strided,
unstrided and multi-pack. Really though unstrided was just a limited
form of multi-pack buffer and didn't imply any hind of special
optimization so this patch consolidates some code by reducing to just
two types; strided and multi-pack.
This is a counter part for _cogl_material_layer_pre_paint which takes a
layer index instead of a direct CoglMaterialLayer pointer. The aim is to
phase out code that directly iterates the internal layer pointers of a
material since the layer pointers can change if any property of any
layer is changed making direct layer pointers very fragile.
This exposes the idea of a stack of source materials instead of just
having a single current material. This allows the writing of orthogonal
code that can change the current source material and restore it to its
previous state. It also allows the implementation of new composite
primitives that may want to validate the current source material and
possibly make override changes in a derived material.
When compiling for non-glx platforms the winsys feature data array
ends up empty. Empty arrays cause problems for MSVC so this patch adds
a stub entry so that the array always has at least one entry.
Based on a patch by Ole André Vadla Ravnås
Instead of directly manipulating GL textures itself,
CoglTexture2DSliced now works in terms of CoglHandles. It creates the
texture slices using cogl_texture_new_with_size which should always
end up creating a CoglTexture2D because the size should fit. This
allows us to avoid replicating some code such as the first pixel
mipmap tracking and it better enforces the separation that each
texture backend is the only place that contains code dealing with each
texture target.
This adds two new internal functions to create a foreign texture for
the texture 2d and rectangle backends. cogl_texture_new_from_foreign
will now use one of these backends directly if there is no waste
instead of always using the sliced texture backend.
When picking a size for the last slice in a texture, Cogl would always
pick the biggest power of two size that doesn't create too much
waste and is less than or equal to the previous slice size. However
this can end up creating a texture that is bigger than needed if there
is a smaller power of two.
For example, if the maximum waste is 127 (the current default) and we
try to create a texture that is 257 pixels wide it will decide that
the next power of two (512) is too much waste (255) so it will create
the first slice at 256 pixels wide. Then we only have 1 pixel left to
allocate but Cogl would pick the next smaller size that has a small
enough waste which is 128. But of course 1 is already a power of two
so that's redundantly oversized by 127.
This patch fixes it so that whenever it finds a size that would be big
enough, instead of using exactly that it picks the next power of two
up from the size we need to fill.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2355
There are ordering issues in the pixmap destruction with current and
past X11 server, Mesa and dri2. Under some circumstances, an X pixmap
might be destroyed with the GLX pixmap still referencing it, and thus
the X server will decide to destroy the GLX pixmap as well; then, when
Cogl tries to destroy the GLX pixmap, it gets BadDrawable errors.
Clutter 1.2 used to trap + sync all calls to glXDestroyPixmap(), but
then we assumed that the ordering issue had been solved. So, we're back
to square 1.
I left a Big Fat Comment™ right above the glXDestroyPixmap() call
referencing the bug and the reasoning behind the trap, so that we don't
go and remove it in the future without checking that the issue has been
in fact solved.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2324
When using the debug function _cogl_debug_dump_materials_dot_file to
write a dot file representing the sparse graph of material state we now
only show a link between materials and layers when the material directly
owns that layer reference (i.e. just those referenced in
material->layer_differences) This makes it possible to see when
ancestors of a material are being deferred too for layer state.
For example when looking at the graph if you see that a material has an
n_layers of 3 but there is only a link to 2 layers, then you know you
need to look at it's ancestors to find the last layer.
Both of the cogl_texture_2d_sliced_new functions called the
slices_create function which creates the underlying GL
textures. However this was also called by init_base so the textures
would end up being created twice. This would make it leak the GL
textures and the arrays which point to them.
Clutter has now taken responsibility for managing its viewport,
projection matrix and view transform as part of ClutterStage so
_cogl_setup_viewport is no longer used by anything, and since it's quite
an obscure API anyway it's we've taken the opportunity to remove the
function.
*** WARNING: THIS COMMIT CHANGES THE BUILD ***
Do not recurse into the backend directories to build private, internal
libraries.
We only recurse from clutter/ into the cogl sub-directory; from there,
we don't recurse any further. All the backend-specific code in Cogl and
Clutter is compiled conditionally depending on the macros defined by the
configure script.
We still recurse from the top-level directory into doc, clutter and
tests, because gtk-doc and tests do not deal nicely with non-recursive
layouts.
This change makes Clutter compile slightly faster, and cleans up the
build system, especially when dealing with introspection data.
Ideally, we also want to make Cogl part of the top-level build, so that
we can finally drop the sed trick to change the shared library from the
GIR before compiling it.
Currently disabled:
‣ OSX backend
‣ Fruity backend
Currently enabled but untested:
‣ EGL backend
‣ Windows backend
Each time a material property changes we look to see if any of its
ancestry has become redundant and if so we prune that redundant
ancestry.
There was a problem with the logic that handles this though because we
weren't considering that a material which is a layer state authority may
still defer to ancestors to define the state of individual layers.
For example a material that derives from a parent with 5 layers can
become a STATE_LAYERS authority by simply changing it's ->n_layers count
to 4 and in that case it can still defer to its ancestors to define the
state of those 4 layers.
This patch checks first if a material is a layer state authority and if
so only tries to prune its ancestry if it also *owns* all the individual
layers it depends on. (I.e. if g_list_length
(material->layer_differences) != material->n_layers then it's not safe
to try pruning its ancestry!)
http://bugzilla-attachments.gnome.org/attachment.cgi?id=170907
There is GL_INVALID_ENUM error for GL_DEPTH_STENCIL when call
glRenderbufferStorage() with OpenGL ES backend. So enable this
only for OpenGL backend.
Signed-off-by: Robert Bragg <robert@linux.intel.com>
If COGL_OBJECT_DEBUG is defined then cogl-object-private.h will call
COGL_NOTE in the ref and unref macros. For this to work the debug
header needs to also be included or COGL_NOTE won't necessarily be
defined.
cogl_util_next_p2 is declared in cogl-util.h which is a private header
so it shouldn't be possible for an application to use it. It's
probably not a function we'd like to export from Cogl so it seems
better to keep it private. This patch renames it to _cogl_util_next_p2
so that it won't be exported from the shared library.
The documentation for the function is also slightly wrong because it
stated that the function returned the next power greater than
'a'. However the code would actually return 'a' if it's already a
power of two. I think the actual behaviour is more useful so this
patch changes the documentation rather than the code.
Previously CoglVertexBuffer would always set the flush options flags
to at least contain COGL_MATERIAL_FLUSH_FALLBACK_MASK. The code then
later checks whether any flags are set before deciding whether to copy
the material to implement the overrides. This means that it would
always end up copying the material even if there are no fallback
layers. This patch changes it so that it only sets
COGL_MATERIAL_FLUSH_FALLBACK_MASK if fallback_layers != 0.
If a single arbfp program is being shared between multiple CoglMaterials
then we need to make sure we update all program.local params when
switching between materials. Previously we had a dirty flag to track
when combine_constant params were changed but didn't take in to account
that different materials sharing the same program may have different
combine constants.
Previously the backend private state was used to either link to an
authority material or provide authoritative program state. The mechanism
seemed overly complex and felt very fragile. I made a recent comment
which added a lot of documentation to make it easier to understand but
still it didn't feel very elegant.
This patch takes a slightly different approach; we now have a
ref-counted ArbfpProgramState object which encapsulates a single ARBfp
program and the backend private state now just has a single member which
is a pointer to one of these arbfp_program_state objects. We no longer
need to cache pointers to our arbfp-authority and so we can get rid of
a lot of awkward code that ensured these pointers were
updated/invalidated at the right times. The program state objects are
not tightly bound to a material so it will also allow us to later
implement a cache mechanism that lets us share state outside a materials
ancestry. This may help to optimize code not following the
recommendations of deriving materials from templates, avoiding one-shot
materials and not repeatedly modifying materials because even if a
material's ancestry doesn't naturally lead us to shareable state we can
fallback to searching for shareable state using central hash tables.
This adds a way to iterate the layer indices of the given material since
cogl_material_get_layers has been deprecated. The user provides a
callback to be called once for each layer.
Because modification of layers in the callback may potentially
invalidate any number of the internal CoglMaterialLayer structures and
invalidate the material's layer cache this should be more robust than
cogl_material_get_layers() which used to return a const GList *
pointing directly to internal state.
This fixes the material backends to declare their constant vtable in the
c file with a corresponding extern declaration in the header. This
should fix complaints about duplicate symbols seen on OSX.
Instead of lazily incorporating combine constants as arbfp PARAM
constants in the source directly we now use program.local parameters
instead so we can avoid repeating codegen if a material's combine
constant is updated. This should be a big win for applications animating
a constant used for example in an animated interpolation, such as
gnome-shell.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2280
This makes it so we don't consider LAYER_STATE_TEXTURE changes to affect
the arbfp code. This should avoid a lot of unneeded passes of
code generation for applications modifying the texture for a layer.
This makes it so we only notify backends of either a single material
change or a single layer change. Previously all material STATE_LAYERS
changes would be followed by a more detailed layer change.
For backends that perform code generation for fragment processing they
typically need to understand the details of how layers get changed to
determine if they need to repeat codegen. It doesn't help them to report
a material STATE_LAYERS change for all layer changes since it's so
broad, they really need to wait for the layer change to be notified.
What does help though is to report a STATE_LAYERS change for a change in
material->n_layers because they typically do need to repeat codegen in
that case.
This fixes a number of issues relating to how we track the arbfp private
state associated with CoglMaterials. At the same time it adds much more
extensive code documentation to try and make it a bit more approachable.
When notifying a backend about a layer being modified we now pass the
layers current owner for reference. NB: Although a layer can indirectly
be referenced by multiple layers, a layer is considered immutable once
it has dependants, so there is only ever one material associated with a
layer being modified. Passing the material pointer to the backends
layer_pre_change callback can be useful for backends that associate
their private state with materials and may need to update that state in
response to layer changes.
This renames the get_arbfp_authority function to
get_arbfp_authority_no_check to clarify that the function doesn't
validate that the authority cache is still valid by looking at the age
of the referenced material. The function should only be used when we
*know* the cache has already been checked.
We now pass a boolean to _cogl_material_pre_change_notify to know when
a material change is as a result of a layer change. We plan to use this
information to avoid notifying the backends about material changes if
they are as a result of layer changes. This will simplify the handling
of state changes in the backends because they can assume that layer and
material changes are mutually exclusive.
This adds an internal _cogl_material_get_layer_combine_constant function
so we can query the current layer combine constant back. We should
probably make this a public property getter, but for now we just need
this so we can read the constant in the arbfp backend.
We are going to start tracking more per-texture unit state with arbfp
private state so this adds an internal UnitState type and we allocate an
array of these when setting up a new private state structure. The first
thing that has been moved into this is the sampled boolean to know when
a particular texture unit gets sampled from in the generated arbfp code.
This avoids the use of of gcc constructor and destructor attributes to
initialize the cogl uprof context and optionally print a cogl uprof
report at app exit. We now initialize the uprof context in
cogl_context_create instead.
When building with --enable-profile we now depend on the uprof-0.3
developer release which brings a few improvements:
» It lets us "fix" how we initialize uprof so that instead of using a shared
object constructor/destructor (which was a hack used when first adding
uprof support to Clutter) we can now initialize as part of clutter's
normal initialization code. As a side note though, I found that the way
Clutter initializes has some quite serious problems whenever it
involves GOptionGroups. It is not able to guarantee the initialization
of dependencies like uprof and Cogl. For this reason we still use the
contructor/destructor approach to initialize uprof in Cogl.
» uprof-0.3 provides a better API for adding custom columns when reporting
timer and counter statistics which lets us remove quite a lot of manual
report generation code in clutter-profile.c.
» uprof-0.3 provides a shared context for tracking mainloop timer
statistics. This means any mainloop based library following the same
"Mainloop" timer naming convention can use the shared context and no
matter who ends up owning the final mainloop the statistics will always
be in the same place. This allows profiling of Clutter with an
external mainloop such as with the Mutter compositor.
» uprof-0.3 can export statistics over dbus and comes with an ncurses
based ui to vizualize timer and counter stats live.
The latest version of uprof can be cloned from:
git://github.com/rib/UProf.git
When try_creating_fbo fails it deletes any intermediate render buffers
that were created. However it doesn't clear the list so I think if it
failed a second time it would try to delete the render buffers
again. This could potentially cause problems if a subsequent fbo is
created because the destructor for the original might delete the
renderbuffers of the new fbo.
Let's try to keep Cogl's build as non-recursive as possible, in the hope
that one day we'll be able to make it fully non-recursive along with the
rest of Clutter.
Flushing the framebuffer state can cause some drawing to occur if the
framebuffer has a clip stack which needs the stencil buffer. This was
causing the array pointers set up by enable_state_for_drawing_buffer
to get mangled so it would crash when it hits glDrawArrays. This patch
moves the framebuffer state flush to before it sets up the array
pointers.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2297
When disposing a material layer of type 'texture' we should check that
the texture handle is still valid before calling cogl_handle_unref().
This avoids an assertion failure when disposing a ClutterTexture.
This patch merges in substantial work from
Emmanuele Bassi <ebassi@linux.intel.com>
* Use new introspection --include-uninstalled API since we don't want
to try to find the clutter-1.0.pc file before it's installed.
* Use --pkg-export for Clutter-1.0.gir, since we want the .gir file to
contain the associated pkg-config file.
* Drop the use of --pkg for dependencies; those come from the associated
.gir files. (Actually, --pkg is almost never needed)
* Add --quiet
http://bugzilla.clutter-project.org/show_bug.cgi?id=2292
Intel CE3100 and CE4100 SoCs are designed for TVs. They have separate
framebuffers that are blended together by a piece of hardware to make
the final output. The library that allows you to initialize and
configure those planes is called GDL. A EGL GDL winsys can then be
use with those planes as NativeWindowType to select which plane to use.
This patch adds a new ClutterBackendCex100 backend that can be
selected at compile time with the new --with-flavour=cex100 option.
Some minor fixes here and there: missing include, wrongly placed #endif,
unused variable warning fixes, missing #ifdef.
Make ClutterStageEGL a subclass of either ClutterStageX11 or GObject
depending if you compile with X11 support (EGLX) or not (native).
Weak materials are ones that don't take a reference on their parent and
they are associated with a callback that notifies when the material is
destroyed, because its parent was freed or modified.
More details can be found at:
http://wiki.clutter-project.org/wiki/CoglDesign/CoglMaterial
For now the concept is internal only but the plan is to make this public
at some point once we have tested the design internally.
In the case where there is no error log for arbfp we were returning a
"" string literal. The other paths were using g_strdup to return a
string that could be freed with g_free. This makes the arbfp path return
g_strdup ("") instead.
There are quite a few if {} else {} blocks for dealing with arbfp else
glsl and the first block is guarded with #ifdef HAVE_COGL_GL. In this
case though the #endif was before the else so it wouldn't compile for
gles.
We need to include cogl-shader-private.h to have the
COGL_SHADER_TYPE_GLSL define. When building for opengl this wasn't
noticed probably because some other header indirectly includes this
file. It was a problem when building for gles2 though.
Instead of exposing an API that provides an OpenGL state machine style
where you first have to bind the program to the context using
cogl_program_use() followed by updating uniforms using
cogl_program_uniform_xyz we now have uniform setter methods that take an
explicit CoglHandle for the program.
This deprecates cogl_program_use and all the cogl_program_uniform
variants and provides the following replacements:
cogl_program_set_uniform_1i
cogl_program_set_uniform_1f
cogl_program_set_uniform_int
cogl_program_set_uniform_float
cogl_program_set_uniform_matrix
--quiet has been added to g-ir-scanner in the 0.9.1 cycle. We really
want to be able to compile clutter with 0.6.14 to be able to reuse
gir files that are distributed in current distributions.
Use the INTROSPECTION_SCANNER_ARGS (previously unused) variable to
convey --quiet when necessary.
Fixes: http://bugzilla.clutter-project.org/show_bug.cgi?id=2265
CoglAtlas chooses a fairly large default initial size of either
512x512 or 1024x1024 depending on the texture format. There is a
chance that this size will not be supported on some platforms which
would be catastrophic for the glyph cache because it would mean that
it would always fail to put any glyphs in the cache so text wouldn't
work. To fix this the atlas code now checks whether the chosen initial
size is supported by the texture driver and if not it will get halved
until it is supported.
Previously when creating a new rectangle map it would try increasingly
larger texture sizes until GL_MAX_TEXTURE_SIZE is reached. This is bad
because it queries state which should really be owned by the texture
driver. Also GL_MAX_TEXTURE_SIZE is often a conservative estimate so
larger texture sizes can be used if the proxy texture is queried
instead.
Previously each node in the rectangle map tree would store the total
remaining space in all of its children to use as an optimization when
adding nodes. With this it could skip an entire branch of the tree if
it knew there could never be enough space for the new node in the
branch. This modifies that slightly to instead store the largest
single gap. This allows it to skip a branch earlier because often
there would be a lot of small gaps which would add up to enough a
space for the new rectangle, but the space can't be used unless it is
in a single node.
The rectangle map still needs to keep track of the total remaining
space for the whole map for the debugging output so this has been
added back in to the CoglRectangleMap struct. There is a separate
debugging function to verify this value.
Previously when the atlas needs to be migrated it would start by
trying with the same size as the existing atlas if there is enough
space for the new texture. However even if the atlas is completely
sorted there will always be some amount of waste so when the atlas
needs to grow it would usually end up redundantly trying the same size
when it is very unlikely to fit. This patch changes it so that there
must be at least 6% waste available after the new texture is added
otherwise it will start with the next atlas size.
When iterating over the rectangle map a stack is used to implement a
recursive algorithm. Previously this was slice allocating a linked
list. Now it uses a GArray which is retained with the rectangle map to
avoid frequent allocations which is a little bit faster.
Previously the remaining space was managed as part of the
CoglRectangleMap struct. Now it is stored per node so that at any
point in the hierarchy we can quickly determine how much space is
remaining in all of the node's children. That way when adding a
rectangle we can miss out entire branches more quickly if we know that
there is no way the new rectangle would fit in that branch.
This also adds a function to recursively verify the cached state in
the nodes such as the remaining space and the number of
rectangles. This function is only called when the dump-atlas-image
debug flag is set because it is potentially quite slow.
_cogl_atlas_new now has two extra parameters to specify the format of
the textures it creates as well as a set of flags to modify the
behavious of the atlas. One of the flags causes the new textures to be
cleared and the other causes migration to avoid actually copying the
textures. This is needed to use CoglAtlas from the pango glyph cache
because it needs to use COGL_PIXEL_A_8 and to clear the textures as it
does not fill in the gaps between glyphs. It needs to avoid copying
the textures so that it can work on GL implementations without FBO
support.
Instead of storing a pointer to the CoglRectangleMap and a handle to
the atlas texture in the context, there is a now a separate data
structure called a CoglAtlas to manage these two. The context just
contains a pointer to this. The code to reorganise the atlas has been
moved from cogl-atlas-texture.c to cogl-atlas.c
This adds an internal CoglCallbackList type which is just a GSList of
of function pointers along with a data pointer to form a
closure. There are functions to add and remove items and to invoke the
list of functions. This could be used in a number of places in Cogl.
This simply renames CoglAtlas to CoglRectangleMap without making any
functional changes. The old 'CoglAtlas' is just a data structure for
managing unused areas of a rectangle and it doesn't neccessarily have
to be used for an atlas so it wasn't a very good name.
Textures within a layer were compared for equality by comparing their
texture handle. However this means that sub textures and atlas
textures which may be internally using the same GL handle would not be
batched together. Instead it now tries to determine the underlying GL
handle using either the slice override or _cogl_texture_get_gl_texture
and then compares those.
When filtering on allowed formats for atlas textures, it now masks out
the BGR and AFIRST bits in addition to the premult bit. That way it
will accept RGB and RGBA formats in any component order.
In theory it could also accept luminance and alpha-only textures but I
haven't added this because presumably if the application has requested
these formats then it has some reason not to use a full RGB or RGBA
texture and we should respect that.
The special handling for texture unit 1 caught the case where unit
1 was changed for transient purposes, but didn't properly handle
the case where the actual non-transient texture was different between
two materials with no transient binding in between.
If the actual texture has changed when flushing, mark unit 1 as dirty
and needing a rebind.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2261
This makes CoglProgram/Shader automatically detect when the user has
given an ARBfp program by checking for "!!ARBfp1.0" at the beginning of
the user's source.
ARBfp local parameters can be set with cogl_program_uniform_float
assuming you pass a @size of 4 (all ARBfp program.local parameters
are vectors of 4 floats).
This doesn't expose ARBfp environment parameters or double precision
local parameters.
Previously we had an internal only _cogl_material_set_user_program to
redirect legacy usage of cogl_program_use() through CoglMaterial. This
instead makes the API public because until we implement our planned
"snippet" framework we need a stop-gap solution for using shaders in
Cogl.
The plan is to also support ARBfp with the cogl_program/shader API so
this API will also allow clutter-gst to stop using direct OpenGL calls
that conflict with Cogl's state tracking.
A change to a layer is also going to be a change to its owning material
so we have to chain up in _cogl_material_layer_pre_change_notify and
call _cogl_material_pre_change_notify. Previously we were only
considering if the owning material was referenced in the journal but
that ignores that it might also have dependants. We no longer need to
flush the journal directly in layer_pre_change_notify.
In _cogl_material_layer_pre_change_notify when we see that a layer has
dependants and it can't be modified directly then we allocate a new
layer. In this case we also have to link the new layer to its required
owner. If the immutable layer we copied had the same owner though we
weren't unlinking that old layer.
In _cogl_material_pre_change_notify we need to identify if it's a sparse
property being changed and if so initialize the state group if the given
material isn't currently the authority for it.
Previously we were unconditionally calling
_cogl_material_initialize_state which would e.g. NULL the layer
differences list of a material each time a layer change was notified.
It would also call _cogl_material_initialize_state for non-sparse
properties which should always be valid at this point so the function
has been renamed to _cogl_material_initialize_sparse_state to make this
clearer with a corresponding g_return_if_fail check.
This fixes how we copy layer differences in
_cogl_material_copy_layer_differences.
We were making a redundant g_list_copy of the src differences and then
iterating the src list calling _cogl_material_add_layer_difference for
each entry which would double the list length, but the initial copy
directly referenced the original layers which wasn't correct.
Also we were initializing dest->n_layers before copying the layer
differences but the act of copying the differences will re-initialize
n_layers to 0 when adding the first layer_difference since it will
trigger a layer_pre_change_notify and since the dest material isn't yet
a STATE_LAYERS authority the state group is initialized before allowing
the change.
In _cogl_material_texture_storage_change_notify we were potentially
dereferencing layer->texture without checking first that it is the
authority of texture state. We now use
_cogl_material_layer_get_texture() instead.
This improve the dot file output available when calling
_cogl_debug_dump_materials_dot_file. The material graph now directly
points into the layer graph and the layers now show the texture unit
index.
When the texture is set on a layer so that it is back to the parent's
texture it would clear the texture change flag but it wouldn't unref
the texture. The free function for a material layer does not unref the
texture if the change flag is cleared so the texture would end up
leaking. This happens for ClutterTexture because it disposes the
texture by setting layer 0 of the material to COGL_INVALID_HANDLE
which ends up the same as the default material.
In _cogl_material_layer_pre_paint we were mistakenly dereferencing the
layer->texture member for the passed layer instead of dereferencing the
texture state authority which was causing crashes in some cases.
This makes the gles2 cogl_program_use consistent with the GL version by
not binding the program immediately and instead leaving it to
cogl-material.c to bind the program when actually drawing something.
Previously custom uniforms were tracked in _CoglGles2Wrapper but as part
of a process to consolidate the gl/gles2 shader code it seems to make
sense for this state to be tracked in the CoglProgram object instead.
http://bugzilla.o-hand.com/show_bug.cgi?id=2179
Instead of having to query GL and translate the GL enum into a
CoglShaderType each time cogl_shader_get_type is called we now keep
track of the type in CoglShader.
Nothing was storing the shader type when a shader was created so it
would get confused about whether it was a custom vertex or fragment
shader.
Also the 'type' member of CoglShader was a GLenum but the only place
that read it was treating it as if it was CoglShaderType. This changes
it be CoglShaderType.
When loading an RGB image GdkPixbuf will pad the rowstride so that the
beginning of each row is aligned to 4 bytes. This was causing us to
fallback to the code that copies the buffer. It is probably safe to
avoid copying the buffer if we can detect that the rowstride is simply
an alignment of the packed rowstride.
This also changes the copying fallback code so that it uses the
aligned rowstride. However it is now extremely unlikely that the
fallback code would ever be used.
In commit b780413e5a the GdkPixbuf loading code was changed so that
if it needs to copy the pixbuf then it would tightly pack it. However
it was still using the rowstride from the pixbuf so the image would
end up skewed. This fixes it to use the real rowstride.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2235
In OpenGL the 'shininess' lighting parameter is floating point value
limited to the range 0.0→128.0. This number is used to affect the size
of the specular highlight. Cogl materials used to only accept a number
between 0.0 and 1.0 which then gets multiplied by 128.0 before sending
to GL. I think the assumption was that this is just a weird GL quirk
so we don't expose it. However the value is used as an exponent to
raise the attenuation to a power so there is no conceptual limit to
the value.
This removes the mapping and changes some of the documentation.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2222
When flushing a fixed-function or arbfp material it would always call
disable_glsl to try to get rid of the previous GLSL shader. This is
needed even if current_use_program_type is not GLSL because if an
application calls cogl_program_uniform then Cogl will have to bind the
program to set the uniform. If this happens then it won't update
current_use_program_type presumably because the enabled state of arbfp
is still valid.
The problem was that disable_glsl would only select program zero when
the current_use_program_type is set to GLSL which wouldn't be the case
if cogl_program_uniform was called. This patch changes it to just
directly call _cogl_gl_use_program_wrapper(0) instead of having a
separate disable_glsl function. The current program is cached in the
cogl context anyway so it shouldn't cause any extra unnecessary GL
calls.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2232
g_ascii_dtostr was being used in four separate arguments to
g_string_append_printf but all invocations of it were using the same
buffer. This would end up with all of the arguments having the same
value which would depend on whichever order the compiler evaluates
them in. This patches changes it to use a multi-dimensional array and
a loop to fill in the separate buffers.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2219
The ARBfp programs are created with a printf() wrapper, which usually
fails in non-en locales as soon as you start throwing things like
floating point values in the mix.
We should use the g_ascii_dtostr() function which places a double into a
string buffer in a locale-independent way.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2219
This function creates a CoglBitmap which internally references a
CoglBuffer. The map and unmap functions will divert to mapping the
buffer. There are also now bind and unbind functions which should be
used instead of map and unmap whenever the data doesn't need to be
read from the CPU but will instead be passed to GL for packing or
unpacking. For bitmaps created from buffers this just binds the
bitmap.
cogl_texture_new_from_buffer now just uses this function to wrap the
buffer in a bitmap rather than trying to bind the buffer
immediately. This means that the buffer will be bound only at the
point right before the texture data is uploaded.
This approach means that using a pixel array will take the fastest
upload route if possible, but can still fallback to copying the data
by mapping the buffer if some conversion is needed. Previously it
would just crash in this case because the texture functions were all
passed a NULL pointer.
http://bugzilla.clutter-project.org/show_bug.cgi?id=2112
The docs for GdkPixbuf say that the last row of the image won't
necessarily be allocated to the size of the full rowstride. The rest
of Cogl and possibly GL assumes that we can copy the bitmap with
memcpy(height*rowstride) so we previously would copy the pixbuf data
to ensure this. However if the rowstride is the same as bpp*width then
there is no way for the last row to be under-allocated so in this case
we can just directly upload from the gdk pixbuf. Now that CoglBitmap
can be created with a destroy function we can make it keep a reference
to the pixbuf and unref it during its destroy callback. GdkPixbuf
seems to always pack the image with no padding between rows even if it
is RGB so this should end up always avoiding the memcpy.
The fallback code for when we do have to copy the pixbuf is now
simplified so that it copies all of the rows in a single loop. We only
copy the useful region of each row so this should be safe. The
rowstride of the CoglBitmap is now always allocated to bpp*width
regardless of the rowstride of the pixbuf.
The CoglBitmap struct is now only defined within cogl-bitmap.c so that
all of its members can now only be accessed with accessor
functions. To get to the data pointer for the bitmap image you must
first call _cogl_bitmap_map and later call _cogl_bitmap_unmap. The map
function takes the same arguments as cogl_pixel_array_map so that
eventually we can make a bitmap optionally internally divert to a
pixel array.
There is a _cogl_bitmap_new_from_data function which constructs a new
bitmap object and takes ownership of the data pointer. The function
gets passed a destroy callback which gets called when the bitmap is
freed. This is similar to how gdk_pixbuf_new_from_data
works. Alternatively NULL can be passed for the destroy function which
means that the caller will manage the life of the pointer (but must
guarantee that it stays alive at least until the bitmap is
freed). This mechanism is used instead of the old approach of creating
a CoglBitmap struct on the stack and manually filling in the
members. It could also later be used to create a CoglBitmap that owns
a GdkPixbuf ref so that we don't necessarily have to copy the
GdkPixbuf data when converting to a bitmap.
There is also _cogl_bitmap_new_shared. This creates a bitmap using a
reference to another CoglBitmap for the data. This is a bit of a hack
but it is needed by the atlas texture backend which wants to divert
the set_region virtual to another texture but it needs to override the
format of the bitmap to ignore the premult flag.
The 'format' member of CoglTexture2DSliced is returned by
cogl_texture_get_format. All of the other backends return the internal
format of the GL texture in this case. However the sliced backend was
returning the format of the image data used to create the texture. It
doesn't make any sense to retain this information because it doesn't
necessarily indicate the format of the actual texture. This patch
changes it to store the internal format instead.
