Full support for EXT_transform_feedback, transform feedback objects
from ARB_transform_feedback2 and equivalent OpenGL ES 3.0 functionality.
Example usage is in src/Magnum/Test/TransformFeedbackGLTest.cpp, I'll
add some example later.
No backward compatibility issues should exist, as the class is in most
(if not all) cases used with unscoped name:
class MyShader: public AbstractShaderProgram {
public:
typedef Attribute<0, Vector3> Position;
// ...
};
With DSA there is slightly different (but completely understandable)
usage -- it's needed to specify the target upon creation, not deferring
it to begin() call. Timestamp queries (TimeQuery::timestamp()) must now
also be created with new TimeQuery::Target::Timestamp target.
The old way (parameterless constructor and begin(Target)) is still
supported, but is marked as deprecated and will be removed in future
release. Also, using the old way the DSA function is simply not used.
Also fixed SampleQuery test to account for cases where the driver might
not support ARB_occlusion_query2.
According to reports on delphigl.de this extension is far more supported
in comparison to the NV version (and also there's much less FF cruft in
the specification).
The indexed binding is allowed for only some types (atomic counters,
uniforms, shader storage and transform feedback), thus we need separate
enum for that. Because the bind() function will be used far more often
than setTargetHint(), the original Target enum is now renamed to
TargetHint and the new Target enum contains (in non-deprecated build)
only three values.
For backwards compatibility, though, we need to have all original Target
values, thus the new Target enum contains also all other values from
TargetHint, but they are marked as deprecated and (at least) run-time
checked in bind() so they aren't accidentaly used for indexed binding.
Similarly there are also deprecated Target overloads of Buffer() and
setTargetHint(). It's ugly, but hopefully will suffice for now. This mess
will be removed as soon as possible in some upcoming version.
As glGen*() only reserves object name without creating it, it must be
ensured that the object is created before calling functions which expect
already created object (such as glObjectLabel() or glBindTextures()).
Allows me to remove quite a lot hacks in the tests.
The documentation of ARB_invalidate_subdata mentions that all the
functions are really just a hint for the implementation to make some
performance optimizations and they are not affecting behavior at all. So
it's perfectly fine to do nothing if the extension is not supported.
I didn't do this originally as I mistakenly thought that invalidating
depth buffer would somehow behave the same as clearing it, but that's
not the case.
Otherwise my NVidia emits GL_INVALID_VALUE with message "that <object>
is not valid object", which isn't in the specs. I hate this "deferred
everything" approach in GL. Is this even legal optimization?
In OpenGL ES this is implemented in EXT_multi_draw_arrays extension, if
it is not available, the functionality is emulated using sequence of
normal draw() calls.
Functionality provided by GL 3.3 and ARB_instanced_arrays, on ES2 this
is again implemented in three different extensions --
{ANGLE,EXT,NV}_instanced_arrays. They are disabled until Magnum has
proper extension loading on ES.
On desktop GL this is provided by ARB_draw_instanced (GL 3.1). Base
instance is available only on desktop GL (4.2, ARB_base_instance). In
ES2 the instanced functionality is provided by three (!) different
extensions (ANGLE_instanced_arrays, EXT_draw_instanced,
NV_draw_instanced), the proper implementation is chosen on context
creation based on what extension is available. Though we don't have
extension loader for ES yet, thus all these extensions are disabled and
the implementation has assertion in it.
Added blind test which tests only that something has been drawn and no
errors were emitted, but not whether the right command is used. I'll
probably need to check this later, because the Mesh::draw() behemoth is
going slightly out of hand :)
Only one value from these two was used in the end, wasting precious
bytes. Also these two values were used to differentiate between indexed
and non-indexed mesh (instead of relying on actual index buffer being
bound), which was very confusing. This approach looks more clean. The
MeshView class is not yet updated, as the change would expose some
features that aren't possible in current implementation (base vertex
specification).
Merged Mesh::setVertexCount() and Mesh::setIndexCount() into one
Mesh::setCount(), the two original functions are now guarded aliases to
the new one, are marked as deprecated and will be removed in future
release, similarly for the getters.
In particular, if the mesh is indexed, setVertexCount() does nothing and
vertexCount() returns 0. The setIndexCount() and indexCount() do and
return the same regardless of whether the mesh is indexed or not.
Vladimír Vondruš