These changes are necessary for programs whose set of active attributes is not fixed at compile time by a template parameter pack, but rather varies based on the generated shader text at runtime. In such cases, the attribute location of a given named attribute may vary between instances of the same Program.
Previously, attribute bindings were implicitly associated with a location based on template parameter order, and -1 was used to indicate an inactive attribute. This left us unable to disable the appropriate attribute when it went from active to inactive. Now, the state tracker for bindings explicitly associates locations and state, and an empty optional is used to indicate an inactive attribute.
In addition, a gl::VertexArray class is now exposed, allowing more flexibility in the relationship between Programs, Segments, and attribute bindings. In this commit, that relationship does not change, but the subsequent commit adjusts it to match gl-js, reduce rebinds, and work around buggy VAO implementations.
VertexArray uses a pimpl idiom in order to support implementations that lack the VAO extension. In that case, all VertexArrays share global binding state, reflecting the platform reality in the absence of VAOs, while still providing a uniform API.
The current value is not passed to `SymbolSizeBinder::uniformValues`, so we shouldn't check `currentValue.isConstant()` in `SymbolSizeBinder::attributeBindings`. If it were true, then we might end up using attribute bindings that are appropriate only for a constant property, but uniform bindings that are appropriate only for a source or composite function.
Instead, just wait for a new bucket to be generated. This will happen automatically, since icon/text-size are layout properties.
Moves the following to the renderer directory and out of the style namespace:
* CascadeParameters
* PropertyEvaluationParameters
* UpdateParameters
* PropertyEvaluator
* DataDrivenPropertyEvaluator
* CrossFadedPropertyEvaluator
* PaintPropertyBinder
* PaintProperyStatistics
* PossiblyEvaluatedPropertyValue
* TransitioningLight
* EvaluatedLight
* Update gl-js and generate style code
* Factor out packUint8Pair() helper function
* Draft implementation of DDS for {text,icon}-size
Ports https://github.com/mapbox/mapbox-gl-js/pull/4455
* Fix text-size/composite-function-line-placement test
* Refactor to PaintPropertyBinders-like strategy
* Dedupe gl::Program construction
* Use exponential function base for interpolation
* Dedupe coveringZoomStops method
* Fixup tests
* Fix CI errors (hidden within #if block)
PaintPropertyBinder and subclass templates were being instantiated for every unique attribute type (e.g. a_color, a_fill_color, a_halo_color) even though they behave identically for a given property value type (e.g. Color). To fix this, a unique type such as a_color no longer derives from gl::Attribute<...> -- instead it has an inner Type typedef, which should be used wherever neither a unique type nor attribute name is required. This reduces binary size substantially:
VM SIZE FILE SIZE
++++++++++++++ GROWING ++++++++++++++
-------------- SHRINKING --------------
-2.0% -49.3Ki __TEXT,__text -49.3Ki -2.0%
-3.1% -5.21Ki [None] -2.79Ki -1.6%
-2.1% -4.12Ki __TEXT,__const -4.12Ki -2.1%
-1.4% -4.04Ki __TEXT,__gcc_except_tab -4.04Ki -1.4%
-19.3% -3.62Ki __DATA,__data -3.62Ki -19.3%
-2.5% -1.65Ki __TEXT,__unwind_info -1.65Ki -2.5%
-4.2% -8 __DATA,__mod_init_func 0 [ = ]
-1.9% -68.0Ki TOTAL -65.6Ki -1.9%
Some devices supported by Mapbox GL provide only 8 vertex attributes; this change packs existing attributes to get us just under that limit.
For properties using a composite function, pack the min and max values into a single attribute with two logical components instead of using two separate attributes and buffers. Special logic is included for color attributes, whose integer components must be packed into the available bits of floating-point attributes. (We don't have access to ivec types in GL ES 2.0.)
For source functions, continue to bind just a one-component attribute even though the GLSL type is vec2 (or vec4 for colors). The type-checking done by gl::Attribute is relaxed slightly to accommodate this.
* Add symbol dds attributes and adapt style code generation
* Update to mapbox-gl-js/master
* Refactor SymbolFeature as a subclass of GeometryTileFeature
Prepares for enabling DDS on symbol paint properties by allowing the
SymbolFeatures, which we keep around after constructing SymbolLayout,
to be used in evaluating data-driven paint properties later in the
layout process.
* Draft approach for splitting icon/text paint properties
The `Program` types are set up to bind GL attributes to each of the
data-driven paint properties specified in the `PaintProperties` type
provided. Since `SymbolPaintProperties` specifies both `Text*` and
`Icon*` properties, the symbolIcon, symbolIconSDF, and symbolGlyph
programs each attempt to bind roughly double the number of attributes
that they actually need.
This change addresses this by:
- Adding the more specific `IconPaintProperties` and `TextPaintProperties` types, which are subsets of the full `SymbolPaintProperties`.
- The symbol layer continues to use its `SymbolPaintProperties paint` member to track layer property state, but it provides helpers that construct objects of each the specific `{Icon,Text}PaintProperties::Evaluated` type, for use by the painter.
- The three symbol programs instantiate `Program<>` using the appropriate `{Icon,Text}PaintProperties` type.
* check in generated style code
* Populate paint buffers for symbol DDS properties
* Address first round of review comments
* Refactor VectorTile{Layer,Feature} to explicitly share data
* Update submodule
ported from -js: ef5582dd3bc5c15a3112e875ed66494dab8e9d0b
Project the extrusion and compare it's projected pixel length with the
actual pixel length and adjust antialiasing accordingly.
The previous approach calculated the adjustment much more indirectly and
had no intuitive explanation.
ported from -js: eb6c6596c6a7a61363d30356674e0002153b1d19
`altitude` was a terribly-named variable that was used to indirectly
control the fov. This should eliminate some confusion.
`altitude` was equivalent to `cameraToCenterDistance / height`
ported from -js: c52a09639ceeeb809cd837360993df9c45b45420
This replaces a hardcoded shader approximation with a more correct,
FOV-dependent one.
`gl_Position.w - 0.5`
is replaced with
`gl_Position.w / tan(fov)`
The `/ tan(fov)` is handled by multiplying `1 / tan(fov)` into `u_gamma`
outside the shader.
I think `- 0.5` was just chosen as something that looked right for the
default fov.lease enter the commit message for your changes. Lines starting
std::to_string() potentially uses a decimal point character different
than a dot (.), depending on the environment's locale. But since this
generated string will be part of GLSL code, there needs to be a dot and
nothing else.
This converts the style property classes (CirclePaintProperties and so on) to the same tuple-based approach as gl::Attribute and gl::Uniform. The approach is outlined in https://github.com/mapbox/cpp/blob/master/C%2B%2B%20Structural%20Metaprogramming.md.
The main advantage of this approach is it allows writing algorithms that work on sets of style properties, without resorting to code generation or manually repetitive code. This lets us iterate on approaches to data-driven properties more easily.
Another advantage is that the cascading, unevaluated, and evaluated states of a set of properties exist as independent structures, instead of individual properties holding their own state. This is a more functional approach that makes data flow clearer and reduces state.
* [core] update shaders for line property functions
update deps, define device pixel ratio for all shaders
[core] create ShaderParameter struct to store pixel ratio and overdraw param
repair rebase errs
update shaders to include pixel ratio
make sure collision_box never overdraws
update test suite, move shaders to Painter::render so the correct pixel ratio is applied
move shader compiling back to the Painter constructor
rebase from shader --> program refactor
re-factor parameters for collisionBox and debug programs
remove unused vars from line-program, move blur math to shader
update core files
remove unecessary files
update shaders PR, remove comments
bump test suite sha
fix formatting, incorporate feedback
refactor program.hpp
* [core] remove line transformations that were moved to the shaders, bump shader sha
* [core] shorten ProgramParameter instantiation
* [core] bump shader+test suite shas
* Extract `ignore` util to separate header.
* `Segment` now tracks offset and length of indices, rather than primitives. This is more natural.
* Introduce `VertexVector` and `IndexVector` types. These types carry information about the intended draw mode (`Triangles`, `LineStrip`, etc.), and ensure that elements are always appended in a group size appropriate for that draw mode, for both indexed and unindexed rendering.
* `Program`, rather than `Drawable`, is now the unifying object for draw calls. `Program` is the best place to type check the draw call, because it is typed to carry information about the intended primitive, vertex type, attributes, and uniforms.
* Use the debug shaders for debug tile rendering, like gl-js.
* Fix the draw mode for background. It was drawing triangle strips with a triangles array. Surprised this didn’t cause issues. Now it’s type checked.
