Previously when a pipeline is added to the cache it would never be
removed. If the application is generating a lot of unique pipelines
this can end up effectively leaking a large number of resources
including the GL program objects. Arguably this isn't really a problem
because if the application is generating that many unique pipelines
then it is doing something wrong anyway. It also implies that it will
be recompiling shaders very often so the cache leaking will likely be
the least of the problems.
This patch makes it keep track of which pipelines in the cache are in
use. The cache now returns a struct representing the entry instead of
directly returning the pipeline. This entry contains a usage counter
which the pipeline backends can use to mark when there is a pipeline
alive that is using the cache entry. When the hash table decides that
it's a good time to prune some entries, it will make a list of all of
the pipelines that are not in use and then remove the least recently
used half of the pipelines. That way it is less likely to remove
pipelines that the application is actually regenerating often even if
they aren't in use all of the time.
When the cache is pruned the hash table makes a note of how small the
cache could be if it removed all of the unused pipelines. The hash
table starts pruning when there are more entries than twice this
minimum expected size. The idea is that if that case it hit then the
hash table is more than half full of useless pipelines so the
application is generating lots of redundant pipelines and it is a good
time to remove them.
Reviewed-by: Robert Bragg <robert@linux.intel.com>
(cherry picked from commit c21aac22992bb7fef5a8d0913130b8245e67f2eb)
Conflicts:
cogl/driver/gl/cogl-pipeline-fragend-glsl.c
cogl/driver/gl/cogl-pipeline-progend-glsl.c
cogl/driver/gl/cogl-pipeline-vertend-glsl.c
cogl/driver/gl/gl/cogl-pipeline-fragend-arbfp.c
Previously the private feature flags were stored in an enum and we
already had 31 flags. Adding the 32nd flag would presumably make it
add -2³¹ as one of the values which might cause problems. To avoid
this we'll just use an fixed-size array of longs and use indices for
the enum values like we do for the public features.
A slight complication with this is in the CoglDriverDescription where
we were previously using a static intialised value to describe the set
of features that the driver supports. We can't easily do this with the
flags array so instead the features are stored in a fixed-size array
of indices.
Reviewed-by: Robert Bragg <robert@linux.intel.com>
(cherry picked from commit d94cb984e3c93630f3c2e6e3be9d189672aa20f3)
Conflicts:
cogl/cogl-context-private.h
cogl/cogl-context.c
cogl/cogl-private.h
cogl/cogl-renderer.c
cogl/driver/gl/cogl-pipeline-opengl.c
cogl/driver/gl/gl/cogl-driver-gl.c
cogl/driver/gl/gl/cogl-pipeline-progend-fixed-arbfp.c
cogl/driver/gl/gles/cogl-driver-gles.c
cogl/driver/nop/cogl-driver-nop.c
Full GL treats the position attribute specially and requires that it
must be bound to generic attribute location 0 unlike GLES 2.0 or
GL 3.2 core. We now make sure to unconditionally bind the
cogl_position_in attribute to location 0 before linking any glsl program
in cogl.
For reference the relevant part of the GL 3.0 spec that covers these
semantics is Section 2.7 "Vertex Specification" pg 27
After this change there was one remaining problem in
test-custom-attributes where the test_short_verts() test was using its
own "pos" attribute instead of cogl_position_in and so cogl wasn't able
to ensure it would be bound to location 0.
This updates the test to use cogl_position_in but to work around the
fact that glVertexPointer doesn't support UNSIGNED_SHORT components we
force the test to use the glsl backend by setting a shader snippet on
the pipeline.
https://bugs.freedesktop.org/show_bug.cgi?id=67548
Reviewed-by: Neil Roberts <neil@linux.intel.com>
(cherry picked from commit 992ef7b3b49ebb56adde2133bb36330c04133a3f)
Previously on GLES2 where there is no builtin point size uniform then
we would always add a line to the vertex shader to write to the
builtin point size output because when generating the shader it is not
possible to determine if the pipeline will be used to draw points or
not. This patch changes it so that the default point size is 0.0f
which is documented to have undefined results when drawing points.
That way we can avoid adding the point size code to the shader in that
case. The assumption is that any application that is drawing points
will probably have explicitly set the point size on the pipeline
anyway so it is not a big deal to change the default size from 1.0f.
This adds a new pipeline state flag to track whether the point size is
non-zero. This needs to be its own state because altering it needs to
cause a different shader to be added to the pipeline cache. The state
flags that affect the vertex shader have been changed from a constant
to a runtime function because they will be different depending on
whether there is a builtin point size uniform.
There is also a unit test to ensure that changing the point size does
or doesn't generate a new shader depending on the values.
Reviewed-by: Robert Bragg <robert@linux.intel.com>
(cherry picked from commit b2eba06e16b587acbf5c57944a70ceccecb4f175)
Conflicts:
cogl/cogl-pipeline-private.h
cogl/cogl-pipeline-state-private.h
cogl/cogl-pipeline-state.c
cogl/cogl-pipeline.c
This remove cogl-internal.h in favour of using cogl-private.h. Some
things in cogl-internal.h were moved to driver/gl/cogl-util-gl-private.h
and the _cogl_gl_error_to_string function whose prototype was moved from
cogl-internal.h to cogl-util-gl-private.h has had its implementation
moved from cogl.c to cogl-util-gl.c
Reviewed-by: Neil Roberts <neil@linux.intel.com>
(cherry picked from commit 01cc82ece091aa3bec4c07fdd6bc9e5135fca573)
We have found several times now when writing code using Cogl that it
would really help if Cogl's matrix stack api was public as a utility
api. In Rig for example we want to avoid redundant arithmetic when
deriving the matrices of entities used to render and we aren't able
to simply use the framebuffer's matrix stack to achieve this. Also when
implementing cairo-cogl we found that it would be really useful if we
could have a matrix stack utility api.
(cherry picked from commit d17a01fd935d88fab96fe6cc0b906c84026c0067)
_cogl_pipeline_progend_glsl_pre_change_notify and
_cogl_pipeline_progend_glsl_layer_pre_change_notify were only dirtying
the current program state for changes related to fragment processing.
This make both functions also check for changes that affect vertex
shader codegen.
This also fixes a mistake where
_cogl_pipeline_progend_glsl_layer_pre_change_notify was checking for
non-layer related changes which would never be seen, and instead it
should be checking for layer based changes only.
This adds back compatibility for CoglShaders that reference the
cogl_tex_coord_in[] or cogl_tex_coord_out[] varyings. Unlike the
previous way this was done this patch maintains the use of layer numbers
for attributes and maintains forwards compatibility by letting shaders
alternatively access the per-layer tex_coord varyings via
cogl_tex_coord%i_in/out defines that index into the array.
This removes the need to maintain an array of tex_coord varyings and
instead we now just emit a varying per-layer uniquely named using a
layer_number infix like cogl_tex_coord0_out and cogl_tex_coord0_in.
Notable this patch also had to change the journal flushing code to use
pipeline layer numbers to determine the name of texture coordinate
attributes.
We now also break batches by using a deeper comparison of layers so
such that two pipelines with the same number of layers can now cause a
batch break if they use different layer numbers.
This adds an internal _cogl_pipeline_layer_numbers_equal() function that
takes two pipelines and returns TRUE if they have the same number of
layers and all the layer numbers are the same too, otherwise it returns
FALSE.
Where we used to break batches based on changes to the number of layers
we now break according to the status of
_cogl_pipeline_layer_numbers_equal
Reviewed-by: Neil Roberts <neil@linux.intel.com>
(cherry picked from commit e55b64a9cdc93285049d9b969bef67484c2d9fb3)
Note: this will cause a temporary regression for the Cogl 1.x CoglShader
api since it will break compatibility with existing shaders that
reference the texture varyings from the fragment shader.
The intention is to follow up with another patch to add back
CoglShader compatibility.
The ‘builtin uniforms’ are added to the GLSL code generated for the
GLES2 driver to implement missing fixed functionality such as the
builtin point sprite size and the alpha test reference. Previously the
code that accessed these was #ifdef'd to be compiled only when GLES2
was enabled. However since 2701b93f part of this code is now always
used even for non-GLES2 drivers. The code that accessed the builtin
uniforms array was however no longer #ifdef'd which meant that it
wouldn't compile any more if GLES2 was not enabled. This was further
broken becase the GL3 driver actually should be using the alpha test
uniform because that also does not provide any fixed functionality for
alpha testing.
To fix this the builtin uniform array is now always compiled in and
the code to access it is always used. A new member has been added to
the array to mark which private feature the uniform is used to
replace. That is checked before updating the uniform so that under
GLES2 it will update both uniforms but under GL3 it will only update
the alpha test reference.
Reviewed-by: Robert Bragg <robert@linux.intel.com>
(cherry picked from commit 5469a25413883080df75a80153accf5d9124f716)
Since we used to support hybrid fixed-function + glsl pipelines when
running with OpenGL there were numerous differences in how we handled
codegen and uniform updates between GLES2 and full OpenGL. Now that we
only support end-to-end glsl pipelines this patch can largely unify how
we handle GLES2 and OpenGL.
Most notably we now never use the builtin attribute names. This should
also make it easy for us to support creating strict OpenGL 3.1 contexts
where the builtin names have been removed.
Reviewed-by: Neil Roberts <neil@linux.intel.com>
(cherry picked from commit 2701b93f159bf2d3387cedf2d06fe921ad5641f3)
This adds a new "fixed-arbfp" progend so we now have 3 distinct ways of
setting up the state of a pipeline:
» fixed; where the vertex and fragment processing are implemented
using fixed function opengl apis.
» fixed-arbfp; where vertex processing is implemented using fixed
function opengl apis but fragment processing is implemented
using the ARB Fragment Processing language.
» glsl; there vertex and fragment processing are both implemented
using glsl.
This means we avoid unusual, combinations such as glsl for vertex
processing and arbfp for fragment processing, and also avoid pairing
fixed-function vertex processing with glsl fragment processing which we
happen to know hits some awkward code paths in Mesa that lead to poor
performance.
As part of this change, the progend now implies specific vertend and
fragend choices so instead of associating a vertend and fragend with a
pipeline we now just associate a progend choice.
When flushing a pipeline and choosing what progend to use, we now call a
progend->start() method that is able to determine if the vertend and
fragend together will be able to handle the given pipeline so the
vertend and fragend ->start() methods no longer need to return a boolean
status.
Since we now don't need to support glsl used in conjunction with fixed
function this will allow us to avoid ever using OpenGL builtin attribute
names, though this patch doesn't change that yet.
Reviewed-by: Neil Roberts <neil@linux.intel.com>
(cherry picked from commit cec381f50c7a2f2186bd4a8c5f38fecd5f099075)
As part of an effort towards being able to write non-opengl based
backends for Cogl this moves most of the opengl specific code under
drivers/gl. drivers/gl and drivers/gles have been moved to
drivers/gl/gl and drivers/gl/es respectively.
Reviewed-by: Neil Roberts <neil@linux.intel.com>
(cherry picked from commit 7dc482facb0a265c7f48660079e7e12dd7a2813e)
