Compact blending instructions by distinquish the
shape & images. since we gurantee the shape color is
premultiplied alpha.
As a result, this enhanced performance by 0.002s in examples/Performance
on my local machine.
This also contains code cleanup chores.
The CompositeMethod now includes the newly supported InvLumaMask option:
The source pixels are converted to grayscale (luma values),
and the complement of the target's pixels is alpha blended.
As a result, only the part of the source where the grayscale
is not covered by the target is visible.
@APIs: CompositeMethod::InvLumaMask
@Example: examples/InvLumaMasking.cpp
@Issue: https://github.com/thorvg/thorvg/issues/404
Introduced SceneTask, it implements scene clippath behavior
by merging RLE of the scene children render data on any running thread.
Co-authored-by: Michal Szczecinski <m.szczecinsk@partner.samsung.com>
@Issue: https://github.com/thorvg/thorvg/issues/524
re-design the shape data structure so that render backends
are able to access them directly.
This also let us remove tvgShape member data from the Shape::Impl.
To achieve this, migrate shape/stroke/path
from the canvas interface to the render interface.
Introduced CompositeMethod::LumaMask that converts the source pixels to the
grayscale (luma value) before alpha blending. Thanks to it, mask works more like
typical mask in graphics editor software.
Grayscale is calculated with weighted method:
(0.0721*B + 0.7154*G + 0.2125*R) * A
Introduced surface->blender.lumaValue function
Previously, translucent png images are not displayed properly
due to alpha channels premultiplication.
This patch implements that missing part to support it properly
by introducing the Surface data between canvas engine & rasterizer
@Issue: https://github.com/Samsung/thorvg/issues/655
The x and y of spans cannot be negative
because they are specified as coordinates inside the buffer.
Change the type to fix warnings and potential problems
that occur in conversion between int16_t and uint32_t.
apply fast track to fast up the image rasterization.
only shifted image doesn't need to have the matrix-transform computation,
we can avoid it by just shifting offset xy by simple caculating.
@Issue: https://github.com/Samsung/thorvg/issues/206
revert d520da2db8
After considered the scneario seriously,
this doesn't helpful for the performance at all...
just increased code complexity. earlier bad decision... revert it.
After 362d2df the fastTrack cases were applied even for shapes with clips.
These changes fixed this - the check whether a shape is a rect should be done
only if it has no clips.
This patch introduces _STRAIGHT colorspaces (ABGR8888_STRAIGHT and
ARGB8888_STRAIGHT) whose colors are un-alpha-premultiplied. Unmultiplicated
colors are especially needed for wasm thorvg loader and svg2png / tvg2png.
Only C version now.
@issue: #791
there was a wrong condition introduced the bug that image was not updated,
because transformation is not re-applied after the first created the internal image data.
@Issues: https://github.com/Samsung/thorvg/issues/751
