Fixes a bug due to that f32.Rect.Intersect will not return the
empty rectangle for non intersecting rectangles - but instead
a swapped rectangle. By removing the .Canon() call in gpu.go we
ensure that non overlapping clipping rects and paint rects will
lead to no painting.
The Canon() call is not needed since boundsForTransformedRect()
was previously updated to always return a canonical rectangle.
Test case added.
Signed-off-by: Viktor <viktor.ogeman@gmail.com>
This commit fixes a bug where a shape first drawn off-screen
and later moved into screen would not display properly. Since we
cache CPU operations (vertex transform / construction) we need to
upload the constructed data to the GPU after it was build, or a later
frame will use non-initialized memory for it's draw call.
Note that this fix removes the optimization of not processing clip
paths outside the screen - but this is assumed to be uncommon except
when it is first drawn off screen to later be moved in (e.g. in a scrolling list)
in which case we do want to upload the data and prepare for that later
call.
This commit also does a few minor clean ups and adds a test case.
Signed-off-by: Viktor <viktor.ogeman@gmail.com>
To avoid duplicate work when using macros and non-offset transforms,
cache also the new bounding boxes set up for them. The ops.Reader
already generates Keys for all operations, so use them in the cache.
Signed-off-by: Viktor <viktor.ogeman@gmail.com>
Benchmarking showed that the double map access calls
were a bottleneck. Rework the cache to avoid half of them.
The simplest, naive approach would have been to store a
pointer to a struct with a keep field in the map, allowing cheap
update and frame() operation. Benchmarking showed that the
increased GC pressure of that approach decreased performance
however.
Signed-off-by: Viktor <viktor.ogeman@gmail.com>
Previously the cache was only filled during gpu-buffer creation,
resulting in extra work on the CPU to transform vertices if the same
shape was used multiple times in the same frame. Cases such as font
rendering was cached already before this change as it is drawn in it's
own op.Ops that is never reset - and thus re-used from one frame
to the next.
Since we are now calling put() twice per frame an update should no
longer panic.
Signed-off-by: Viktor <viktor.ogeman@gmail.com>
Reintroduce support for offset in stencil vertex so we can reuse
cached values if the only difference in transform is offset. Split
current transform into a pure-offset part and the rest and use
only the complex part as cache key.
Signed-off-by: Viktor <viktor.ogeman@gmail.com>
Cache also CPU operations by moving pathCache into
drawOps and use it in collectOps to avoid splitting and
transformation of quads if in cache. In order to support
this use a concrete type in opCache instead of interface.
Signed-off-by: Viktor <viktor.ogeman@gmail.com>
Add support for affine transformations. The key changes are outlined
below.
- Painting/clipping with rectangles is handled by, for complex
transforms, creating clipping paths representing the transformed
rectangle and using a larger bounding box. Cover/Blit shaders updated
correspondingly to correctly map texture cordinates from the new
bounding boxes.
- Since path splitting must happen on CPU the transforms must happen CPU
side as well - offsets removed from shaders.
- Complex transforms will lead to different path splitting which means
that GPU arrays can no longer be cached if the transform has changed.
Thus the current transform is added as a key to the cache.
- Add a public API to op for setting Affine transformations.
There are a number of optimizations that could be explored further but
which are left out now:
- Caching also of CPU operations (e.g path splitting & transforms) and
not only caching the GPU arrays.
- Allow for re-use of cached GPU vertices if the transformation change
is a pure offset / scaling since the splitting is then the same.
Signed-off-by: Viktor <viktor.ogeman@gmail.com>
Encode TransformOp as an Affince2D matrix instead and use that in gpu and io transform handling.
There are no changes to user facing API and so far only the offset part of the matrix is used.
This patch is a step towards full affine transformations.
Signed-off-by: Viktor <viktor.ogeman@gmail.com>
This is a first step towards supporting affine drawing transforms.
The rendering algorithm relies on quadratic curves that do not cross
x = 0 more than once, thus curves must be split after any rotation/shear
transforms. Move this logic and the generation of vertices to package gpu.
Also close all curves and draw zero-width edges as preparation for
transform since the will no longer implicitly be vertical with no
effect.
This commit will severely affect performance since vertexes are now
transformed also for cached items, using cpu resources.
Signed-off-by: Viktor <viktor.ogeman@gmail.com>
Use greater-than-or-equal test and saturate the z depth buffer
when more than 65k objects are drawn.
Fixes gio#127
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Safari's WebGL does not advertise support for EXT_color_buffer_half_float,
but does support rendering to float FBOs.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Safari's WebGL1 implementation (rightly) complains that first-class
array types are not supported as function result types. Define and
use a struct type instead.
While we're here, use const variables instead of functions.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
The macOS backend doesn't re-create contexts, holding on to the first
created instead. Make sure the GPU leaves the default framebuffer bound,
in case the context is re-used.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Specifying the clear color and depth at the time of clearing is
less error prone and a better for modern GPU APIs. As a bonus, we
can get rid of the BufferAttachment type.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
We're forced by compatibility to encode an integer state into a
floating point. Make the implicit conversion from floating point to
integer more robust against GPUs with low precision floats.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Direct3D 11 supports Direct3D 9.1 level hardware, but only if the shaders are
compiled for target 4_0_level_9_1.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
In particular, the simplifying "-s" flag to gofmt ensures that the automatic
test for unformatted Go source files won't complain.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
OpenGL use the [-1; 1] range for clip depths, Direct3D [0; 1].
Use toClipSpace to encapsulate the difference.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Add fboTextureTransform shader function for cancelling the
implied transformation from fragments output by the fragment
shader and the (u, v) coordinates used to sample from it in a
later pass.
For OpenGL the transformation is the identity.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
OpenGL supports casting from int to float during vertex array
reading. Direct3D doesn't. Since we're transpiling from GLSL, we can't
directly use the Direct3D builtin "asint". So that leaves using
"ivec2" instead of vec2.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Uniform buffers are byte slice backed by Go structs. However, if a uniform
buffer value is embedded in a larger structure with pointers, the Cgo
pointer checker will complain. Avoid the error by moving the uniform
values into separate structures.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
DefaultFramebuffer was set up at Backend creation time, which is
difficult to predict. Instead, let GPU query and cache the current
FBO when created.
Signed-off-by: Elias Naur <mail@eliasnaur.com>
Elias Naur