31abe4bb52
Initialize resized mask textures to rgba(0, 0, 0, 0), not rgba(0, 0, 0, 1).
...
Partially addresses #405 .
2020-07-27 17:07:13 -07:00
dd0ac124d8
Add ColorMatrix filter
2020-07-23 12:46:59 -07:00
94174b81c9
Fix warnings
2020-07-15 13:12:55 -07:00
26bf6cce4e
Document the remainder of the `pathfinder_renderer` API
2020-07-13 19:24:32 -07:00
3d9b5509ba
Add API docs for the renderer and associated types
2020-07-13 16:18:23 -07:00
a49714e0ba
Flesh out the `web_canvas` crate a bit more
2020-07-06 11:01:38 -07:00
e54e0e4760
Start a new `web_canvas` crate, for usage in the browser as a `<canvas>` replacement
2020-07-04 09:33:49 -07:00
5f8fcbeac2
Fix WebGL, and add a simple example of WebGL usage
2020-07-02 19:49:09 -07:00
13ae83d6c5
Combine the Z-buffer and fill indirect draw params buffers to reduce the number
...
of SSBO bindings.
Apparently Mesa RadeonSI drivers have a limit of 8 SSBOs.
Closes #373 .
2020-07-02 11:57:45 -07:00
42289eec6e
Fix Metal memory management and the `canvas_metal_minimal`, `canvas_nanovg`,
...
and `macos_app` examples.
`winit` does not create an autorelease pool, so the Metal backend had not taken
the presence of one into account. Now the Metal backend creates and flushes
autorelease pools as necessary.
Closes #334 .
Closes #376 .
2020-06-29 12:48:49 -07:00
e16f0a045e
Turn off timer queries if the debug UI is off.
...
Closes #360 .
2020-06-25 11:06:17 -07:00
17356311d6
Rework renderer options
2020-06-23 13:15:59 -07:00
df15c855c4
Significantly rework the GPU renderer for the D3D11 backend
2020-06-23 13:15:59 -07:00
23c171fb6f
Refactor GPU a bit more
2020-06-23 13:15:59 -07:00
661da5e12b
Refactor GPU performance measurement
2020-06-23 13:15:59 -07:00
61833168e5
Update the debug UI
2020-06-23 13:15:59 -07:00
ff7c13c8fb
Add a new compute-based D3D11 backend
2020-06-23 13:10:06 -07:00
754a44ae22
Rename the existing renderer to `d3d9` in preparation for the D3D11 branch
2020-06-23 13:10:05 -07:00
ef0c0e0679
Associate texture parameters in shaders with specific texture units, to work
...
around a macOS Radeon driver bug.
Also, ensure there is always a dummy texture bound to every parameter in the
OpenGL backend.
These together stop the macOS Radeon drivers from recompiling the shader on
every drawcall.
Possibly addresses #300 .
2020-05-12 21:40:38 -07:00
2421de6616
Enable compute shader by default if the OpenGL version is high enough.
2020-05-08 12:16:31 -07:00
6299d2c36b
Clear the canvas to the background color if no drawing command did so.
...
Closes #318 .
2020-05-07 21:36:13 -07:00
478008dcf0
Use all four channels in the mask texture.
...
Each bundle of four pixels on a scanline is packed into the RGBA channels of
the mask texture. The area LUT is also expanded to be RGBA so that four pixels'
worth of areas can be looked up at once.
Nice improvement on `paris-30k` from MPVG.
Closes #262 .
2020-05-05 13:11:37 -07:00
f8405d81c0
Bump the number of fills per batch up
2020-05-04 20:01:08 -07:00
6407bf5871
Reduce the size of buffers
2020-05-04 20:01:08 -07:00
d0842f2f4d
Generalize the buffer allocator to support multiple size classes, and use it
...
for tile compositing too.
2020-05-04 15:48:47 -07:00
ac83f79d94
Add a compute shader path, optimize GPU memory management, and switch from SDL
...
to `surfman`.
This is a large commit; explanations of each change follow.
This adds an optional compute shader path, off by default, for rendering fills
to alpha masks. It usually does not improve performance at present, but it
provides a good baseline for further optimizations. Later improvements will
likely aim to avoid writes to the mask texture entirely. Supporting
infrastructure for compute shader has been added to `pathfinder_gpu` for the
OpenGL and Metal backends.
The Metal backend has been optimized to avoid unneccessary buffer allocations
and reflection. As part of this, argument buffers have been removed, as the
current SPIRV-Cross compiler no longer requires them.
The GPU renderer has been improved to avoid stalls. Now, separate buffers are
allocated for each fill batch and for each frame. This can be extended in the
future to allow for separate buffers for tile draw operations as well.
SDL usage has been removed in favor of the native Rust `surfman` and `winit`.
Because `surfman` allows for selection of the integrated GPU on multi-GPU
system, it is chosen by default. The demo supports a new
`--high-performance-gpu` option to opt into the discrete GPU.
2020-04-30 13:33:29 -07:00
104869a6e9
Move the mask enable flag to the tile to reduce drawcall count
2020-04-22 11:04:16 -07:00
cd09177ead
Implement the infrastructure needed to support multiple clip paths.
...
This also lays the groundwork needed to reduce batch breaks between solid and
alpha tiles.
2020-04-21 18:25:20 -07:00
6c996981a3
Fix some transform issues and implement transformed radial gradients in SVG.
...
In particular, this fixes the usvg-to-Pathfinder transform conversion and the
definition of `Transform2F::row_major()`.
Makes `drops.svg` from the MPVG samples work.
2020-04-20 16:40:02 -07:00
5819b8dc60
Stop reallocating texture pages every frame
2020-04-17 11:50:46 -07:00
07ce2a2536
Add a scene field to render target IDs to reduce the chance of misuse
2020-04-16 16:54:52 -07:00
afe1a64f68
Allow multiple tile pages.
...
Closes #151 .
2020-04-15 20:13:37 -07:00
cdbe2fbb6b
Introduce the concept of a base color in order to handle canvas shadow alpha
...
correctly
2020-04-13 18:31:25 -07:00
47919db8f3
Fix frame time measurement in the GL backend, and add it to the NanoVG demo
2020-04-09 17:49:28 -07:00
895f73096e
Switch back to instanced drawing for tiles.
...
Improves tile build time by something around 2x in the NanoVG demo.
2020-04-08 17:16:54 -07:00
eb2b622615
Remove `draw_render_target()` in favor of the standard tile-based rendering
...
path
2020-04-03 13:05:53 -07:00
32a56eab85
Stop leaking textures and framebuffers in the OpenGL backend.
...
Closes #264 .
Possibly addresses #286 .
2020-03-31 17:14:28 -07:00
93f5fbc6d2
Add convenience constructors: `vec2f`, `vec2i`, `Vector2F::zero()`,
...
`Vector2I::zero()`
2020-03-31 11:29:21 -07:00
55df287fec
Move radial gradients from the CPU to the GPU
2020-03-26 21:24:20 -07:00
0c93045f50
Replace the individual tile shaders with an ubershader
2020-03-25 21:59:51 -07:00
2548ab853a
Unify solid tiles and render target tiles.
...
This commit also removes old shaders from the manifest.
2020-03-05 12:22:01 -08:00
9935c9fdff
Don't distinguish between render targets and other images in the texture atlas
...
allocator
2020-03-04 19:41:24 -08:00
a8f7438cd9
Remove transforms from paths.
...
Paints aren't scale-invariant when rendered into the texture atlas, so they
actually do need individual transforms.
2020-03-04 16:28:21 -08:00
27357a23d3
Separate the `AddPaintData` render command into different subcommands
2020-03-04 15:50:53 -08:00
e50039f4c2
Create framebuffers for every texture page
2020-03-04 12:35:07 -08:00
a3736859ba
Rename "paint texture" to just "texture page"
2020-03-04 12:09:08 -08:00
d0f4579864
Move `RenderTargetId` out of the `pattern` module
2020-03-03 15:50:48 -08:00
468848086a
Use the paint texture uniform belonging to the appropriate program
2020-03-03 10:56:32 -08:00
f607b607b0
Composite render targets using tiles, taking the Z-buffer into account.
...
We can do more tile-based optimization, but this should be enough to ensure a
correct rendering.
This temporarily breaks subpixel AA, but it should be fixable by modifying
`blur.fs.glsl`.
Closes #271 .
2020-03-02 20:10:10 -08:00
0f35d9c817
Add an embedded resource loader that places resources directly in the binary.
...
This is useful for WebGL and for downstream crates.io use.
2020-02-28 17:10:53 -08:00
3d39c5e9b7
trimmed whitespace
2020-02-27 08:13:14 -08:00
3a014d78eb
Implement a blur filter for canvas shadows
2020-02-26 18:43:41 -08:00
d1c7da8bd2
Implement pattern repeating and image smoothing control
2020-02-26 14:56:05 -08:00
77b3555828
Separate opacity out from paint.
...
This allows `globalAlpha` to work on render targets.
2020-02-26 12:43:07 -08:00
1deb53fa9c
Implement the remaining Porter-Duff compositing operators
2020-02-25 18:37:23 -08:00
0a3f64eb44
Implement the difference and exclusion blend modes
2020-02-25 14:55:58 -08:00
a8e33d3d3d
Add the soft light blend mode
2020-02-25 14:39:06 -08:00
9b4217300f
Implement the source-in, destination-in, source-out, and destination-atop blend
...
modes to round out the assortment of Porter-Duff blend modes
2020-02-25 12:52:11 -08:00
02012431ca
Implement color dodge and color burn blend modes
2020-02-25 11:37:42 -08:00
5b228ed825
Implement multiply, screen, hard light, and overlay blend modes.
2020-02-24 21:23:17 -08:00
5421525eaa
Implement the HSL filters (called "non-separable blend modes" in the spec).
...
These cannot be implemented with the standard OpenGL blending functions, so we
have to do them manually in a shader, which requires a good deal of machinery
to create intermediate framebuffers and so forth.
2020-02-24 15:37:44 -08:00
67d12adb6c
Replace layers with render targets, which can be used as patterns.
...
This allows us to efficiently handle a lot of workloads that require multiple
HTML canvases, without CPU readback. For example, you can render paths to a
render target, then turn that render target into a repeating pattern that you
fill other paths with. Or you can render paths to a render target and then
composite that render target as a whole with a blend mode.
This introduces a new `DrawRenderTarget` render command that blits a render
target without any paths involved. This is basically just a hack that works
around the fact that our tiled renderer doesn't yet support effects that widen
the ink region (i.e. blurs). It can be removed once we have that support.
2020-02-21 22:14:18 -08:00
3245796445
Add support for multiple paint texture pages.
...
This avoids arbitrary limits on the number of images, gradients, etc. you can
have.
2020-02-21 14:46:10 -08:00
36538d5748
Implement a few more blend modes and switch to premultiplied alpha for layer
...
compositing
2020-02-20 22:22:15 -08:00
4933efb513
Add Lighten and Darken composite ops
2020-02-20 20:38:43 -08:00
ab55b9509b
Fix handling of solid tiles with the Clear blend mode
2020-02-20 18:15:20 -08:00
6acd2d91f7
Implement `clear_rect()` in the canvas front-end by introducing the concept of
...
per-path blend modes.
These should be useful for some canvas composite operations as well.
2020-02-20 15:52:40 -08:00
0c3dad974f
Expose individual blend factors in `pathfinder_gpu`
2020-02-20 15:28:37 -08:00
16a2de88df
Rename "postprocessing" to "effects" and start initial work on composite ops
2020-02-20 14:22:07 -08:00
d9e994e46d
Transition the existing postprocessing system to a layers system.
...
This is preparatory work for composite ops and blurs.
Closes #261 .
2020-02-20 11:21:45 -08:00
3ad1c25796
Implement the even-odd fill rule
2020-02-17 14:44:48 -08:00
677c607a8c
Get clips working for canvas
2020-02-16 13:45:15 -08:00
631e16ce9b
Fix flipped Y axis in mask texture positioning
2020-02-15 14:08:22 -08:00
0883f54e2d
Add a separate mask blit step, in preparation for supporting clips.
...
We can elide this in the future if there are no clips, but it isn't a huge
performance regression right now, so let's just unconditionally do it.
2020-02-14 21:56:43 -08:00
6a73a63336
Factor out mask texture UV calculation, and rename a couple of render commands
2020-02-14 21:37:20 -08:00
84ffc3151e
Split shaders out into a separate module.
2020-02-14 11:56:38 -08:00
90445bac7e
Implement basic support for non-repeating image patterns.
2020-02-11 17:20:21 -08:00
5a21557a6d
Implement basic linear gradients.
...
This is not a very efficient implementation yet, but it seems to work.
2020-02-10 16:01:05 -08:00
a66ed4c073
Specify tile texture coordinates as an affine transform.
...
Part of the groundwork for gradients.
2020-02-05 13:59:32 -08:00
b269723254
Switch to per-pixel texture lookup for paints.
...
This is a prerequisite for supporting gradients and images.
2020-02-04 21:50:13 -08:00
2db43797c3
Split colors out into their own crate.
...
A lot of downstream consumers of `pathfinder_gpu` want to use colors without
the path stuff.
2020-01-31 09:17:04 +01:00
ce3b40cd5c
Make timer queries and framebuffer readback async
2020-01-07 13:03:15 -08:00
0662f6433f
Allow any texture format to be uploaded; fix Metal compilation on newer rustcs
2019-12-20 11:48:51 -08:00
7c655246ae
Add some more features to blending, and expand texture upload features
2019-12-19 08:58:02 -08:00
96758dfc17
Add `rotate` and `scale` methods to `Transform4F`
2019-07-11 15:57:45 -07:00
3f91206d57
Add a `translate` method to 3D transforms
2019-07-11 15:44:01 -07:00
bcac119324
Rename `Transform2DF` to `Transform2F` and `Transform3DF` to `Transform4F`
2019-07-11 14:59:10 -07:00
55825cbc6a
Switch scale and translation constructors for `Transform3DF` over to use vectors
2019-07-11 14:58:45 -07:00
eb0a61679d
Use operator overloading for 3D matrix multiplication
2019-07-11 13:59:39 -07:00
b886c157c1
Use 2-lane instead of 4-lane SIMD types for 2D vectors.
...
Also, this commit rewrites the `add_fill()` method to stop using shuffle
instructions, which can be slow and make the code overly complicated. The
shuffle instructions have been removed from the various SIMD backends.
2019-06-26 16:38:37 -07:00
222fa89b23
Use a 3D transform in the tile vertex shaders.
...
This opens the door to caching tiles at different zoom levels (issue #191 ).
2019-06-25 11:32:39 -07:00
d86ff9011a
Update the postprocessing shader and vertex array for the Metal changes.
...
Closes #201 .
2019-06-21 18:04:18 -07:00
431dcf3d2e
Rename `pathfinder_geometry::basic` to simply `pathfinder_geometry`, and move
...
the remaining code in that crate to a new crate, `pathfinder_content`
The old hierarchy was confusing.
2019-06-21 10:06:19 -07:00
487577a11b
Add Metal support.
...
This commit substantially reworks the `pathfinder_gpu` API to better support
modern APIs like Metal. It should open the door to `gfx-rs`, `wgpu`, Vulkan,
and D3D12 backends relatively straightforwardly.
A new example, `canvas_metal_minimal`, has been added.
Note that the new Metal shaders require a patched version of `spirv-cross` to
build properly. An upstream patch is forthcoming.
2019-06-20 11:47:22 -07:00
dbf02fb742
Make `get_vertex_attr()` fallible
2019-06-12 10:50:24 -07:00
a1b0df0a42
Rename `Point2DF` to `Vector2F`, `Point3DF` to `Vector4F`, and `LineSegmentF`
...
to `LineSegment2F`.
Straw poll on Twitter suggested that these names were preferred.
2019-06-03 12:39:29 -07:00
e1bcc11ace
Change the `F32` and `I32` suffixes to `F` and `I` to match the C API.
...
They're shorter and less noisy.
2019-05-29 19:17:16 -07:00
acf666b701
Stop using triangle fans for DX12/Vulkan compatibility.
...
Closes #165 .
2019-05-29 16:04:33 -07:00
25a6c33a1a
Remove the object index attribute from the shaders, since it's now unused
2019-05-14 15:22:50 -07:00
7a02b78b3d
Explicitly specify color texture coordinates for alpha and solid tiles.
...
This is groundwork for gradients and images.
This commit also refactors the interface for vertex attributes to use named
parameters (via structs), for clarity.
2019-05-14 15:11:04 -07:00
Patrick Walton