Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Support setting AA method dynamically #400

Merged
merged 5 commits into from
Nov 1, 2023
Merged

Support setting AA method dynamically #400

merged 5 commits into from
Nov 1, 2023

Conversation

armansito
Copy link
Collaborator

This replaces the static anti-aliasing setting with a dynamic option in the form of two new settings (one used during pipeline creation and one used during a render):

  • The FullShaders collection now maintains up to 3 fine stage pipeline variants. This can be driven using a new optional RendererOptions field called preferred_antialiasing_method, which determines which fine stage pipeline variants should get instantiated at start up.

  • RenderParams now requires an AaConfig which selects which fine stage variant to use.

  • Added a new key binding (M) to the with_winit example to dynamically cycle through all 3 AA methods.

@armansito armansito force-pushed the dynamic-aa-setting branch 2 times, most recently from 66bdcb1 to 951cacc Compare October 31, 2023 08:26
Copy link
Member

@DJMcNab DJMcNab left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Mostly some style considerations

examples/with_winit/src/lib.rs Outdated Show resolved Hide resolved
examples/with_winit/src/lib.rs Outdated Show resolved Hide resolved
src/lib.rs Outdated

/// The anti-aliasing specialization that should be used when creating the pipelines. `None`
/// initializes all variants.
pub preferred_antialiasing_method: Option<AaConfig>,
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Would a bitflags be better here?

I guess I'd need to consider what the use cases are for the different modes? Why would you choose one over another. Should it be user configurable?

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A partial answer to this is that we think we'll probably want to end up rendering glyphs with area AA and general paths with MSAA. But that speaks to a setting that's even more dynamic than per-render.

Copy link
Collaborator Author

@armansito armansito Oct 31, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Would a bitflags be better here?

A bitfield is probably a good idea. I changed this to use a struct-of-bools instead of bitflags for now.

I mainly wanted to provide a way for existing clients to avoid compiling the variants they don't need. Currently each AaConfig enum maps to exactly one fine stage permutation but this may evolve over time with more configs, such as a single pipeline that is capable of multiple AA modes. Even in that world it may be desirable to turn off AA logic in that pipeline if the client doesn't need it.

I guess I'd need to consider what the use cases are for the different modes? Why would you choose one over another. Should it be user configurable?

This is a great question. Each AA method has a certain trade-off: analytic area AA provides very high quality anti-aliasing but results in conflation artifacts with certain paths (#49) (see also the labyrinth and conflation_artifacts scenes). As @raphlinus said, this is likely the preferred way to render text where quality really matters and glyphs are unlikely to be crafted in a way that can result in conflation artifacts.

The MSAA methods OTOH do not suffer from conflation artifacts but the AA quality is noticeably poorer compared to analytic area. This may be a reasonable trade-off for an application that doesn't have control over all content (for example an SVG viewer or the Canvas API implementation in web a browser). In fact, Skia has had to disable area AA algorithms in the past following user complaints (see crbug.com/913223 and skbug.com/40038110) if you're interested in some of the history of this in Chromium).

Between MSAAx16 and MSAAx8 the trade-off is between performance and quality but maybe that gap will narrow as we invest more time into optimizations.

For now, I chose to provide the option but we can always revisit this.

src/lib.rs Outdated
Comment on lines 113 to 114
/// TODO: Consider evolving this so that the CPU stages can be configured dynamically via
/// `RenderParams`.
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Should this be a doc comment rather than just a comment?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done. To clarify, you wanted this to be a regular comment instead of a doc comment, correct?

src/render.rs Outdated Show resolved Hide resolved
src/shaders.rs Outdated Show resolved Hide resolved
src/shaders.rs Outdated Show resolved Hide resolved
Copy link
Contributor

@raphlinus raphlinus left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good. I agree with most of Daniel's comments, and just had some golfing to make the logic hopefully a little simpler - it feels just a bit clunky and I'm slightly worried about what will happen as the number of permutations grows. But this is definitely useful to have!

Of course, we might reduce permutations by having multiple aa methods present in the same shader, but one nice thing about this setup is that we can experiment with the effect of shared memory size (with effect on occupancy but also likely a nontrivial effect from just clearing the memory, especially as this isn't always optimized well).

src/lib.rs Outdated

/// The anti-aliasing specialization that should be used when creating the pipelines. `None`
/// initializes all variants.
pub preferred_antialiasing_method: Option<AaConfig>,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

A partial answer to this is that we think we'll probably want to end up rendering glyphs with area AA and general paths with MSAA. But that speaks to a setting that's even more dynamic than per-render.

src/render.rs Outdated Show resolved Hide resolved
src/render.rs Outdated
AaConfig::Msaa16 => crate::mask::make_mask_lut_16(),
AaConfig::Msaa8 => crate::mask::make_mask_lut(),
_ => unreachable!(),
};
let buf = recording.upload("mask lut", mask_lut);
self.mask_buf = Some(buf.into());
}
let shader = match fine.aa_config {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Personally I'd call this fine_shader.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done.

src/shaders.rs Outdated
for (i, aa_mode) in AA_MODES.iter().enumerate() {
if options
.preferred_antialiasing_method
.as_ref()
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Could the as_ref be eliminated by comparing m != *aa_mode? It's Copy for this reason.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done.

@armansito
Copy link
Collaborator Author

Looks good. I agree with most of Daniel's comments, and just had some golfing to make the logic hopefully a little simpler - it feels just a bit clunky and I'm slightly worried about what will happen as the number of permutations grows. But this is definitely useful to have!

Of course, we might reduce permutations by having multiple aa methods present in the same shader, but one nice thing about this setup is that we can experiment with the effect of shared memory size (with effect on occupancy but also likely a nontrivial effect from just clearing the memory, especially as this isn't always optimized well).

I agree that this is a little clunky and it would be good to figure out a better way to specify the desired pipeline permutations. For now, this is scoped to 3 AA settings and I imagine it wouldn't be too hard to extend this to an additional "dynamic" mode if we were to explore that.

Copy link
Member

@DJMcNab DJMcNab left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not approving as I haven't had a chance to run this yet, but the code looks good.

There are a couple of small style points, one of which isn't necessarily an improvement (especially if it makes rustfmt break up the aa_modes across 20 lines, which it might accidentally do)

@@ -356,6 +374,7 @@ fn run(
.unwrap_or(Color::BLACK),
width,
height,
antialiasing_method: aa_config,
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'd probably inline this (so that the comment above about base_colour logic is in the right place)

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

aa_config gets reused below so I wanted to have a single binding for it. I moved things around so that the comment is positioned better.

src/shaders.rs Outdated
}
let (range_end_offset, label, aa_config) = match *msaa_info {
Some((label, config)) => (0, label, Some(config)),
None => (1, "fine_area", None),
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

(I had a big comment written out here, which GitHub mobile helpfully deleted)

But broadly, I was wondering if all of this should be inlined into aa_mode, i.e. make it a (bool, usize, &str, Option<&str>).

If we were doing that, I'd then replace the usize offset with the pipeline resources, so fine_area would be:
(support.area, &fine_resources[..7], "fine_area", None)

(Where 7 might not be the right number, but trying to check it lead to the deletion of the previous version of this comment)

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done. I just folded the offset into the tuple to keep the column width manageable.

This replaces the static anti-aliasing setting with a dynamic option in
the form of two new settings (one used during pipeline creation and one
used during a render):

- The `FullShaders` collection now maintains up to 3 `fine` stage
  pipeline variants. This can be driven using a new optional
  `RendererOptions` field called `preferred_antialiasing_method`, which
  determines which fine stage pipeline variants should get  instantiated
  at start up.

- `RenderParams` now requires an `AaConfig` which selects which `fine`
  stage variant to use.

- Added a new key binding (`M`) to the `with_winit` example to
  dynamically cycle through all 3 AA methods.
Instead of passing in a single optional mode to enable, the caller can
now specify which set of AA methods to compile.
Copy link
Contributor

@raphlinus raphlinus left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This needs some merge conflict fixing, but hopefully that should be easy. I haven't run this version but have run a previous version, so I'm ok as long as things didn't get broken. And if they did, that can be a followup PR :)

@@ -180,7 +182,11 @@ fn run(
stats.clear_min_and_max();
}
Some(VirtualKeyCode::M) => {
aa_config_ix = (aa_config_ix + 1) % AA_CONFIGS.len();
aa_config_ix = if modifiers.shift() {
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Personally I would keep the mod n behavior (as the reverse is less discoverable), but that's a style preference. Note though that we do have +/- and mod n for cycling through the test scenes.

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That's fair, the scene cycling logic is actually structured this way (though bound to arrow keys instead of using a modifier) so I just followed that here for consistency.

@armansito armansito merged commit 1e7f850 into main Nov 1, 2023
4 checks passed
@armansito armansito deleted the dynamic-aa-setting branch November 1, 2023 15:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants