Avoid extra blit for GL->D3D. #133

jdm · 2020-02-27T21:44:39Z

Depends on:

asajeffrey

Some minor comments, mostly LGTM, though I think I'd like to revisit this if/when servo/surfman-chains#8 happens.

asajeffrey · 2020-02-28T21:47:05Z

webxr-api/session.rs

 }

+#[derive(Clone, Copy, Debug, Eq, Hash, PartialEq)]
+#[cfg_attr(feature = "ipc", derive(Deserialize, Serialize))]
+pub struct SessionId(pub(crate) u32);


I'm confused, why do we need a swap chain id and a session id? Aren't they always going to be 1:1?

As far as I know they will, but we create an XR session before we end up creating a swapchain, and the code in create_webxr_swap_chain needs to associated the new swapchain with the session's surface provider if it exists.

asajeffrey · 2020-02-28T21:48:35Z

webxr/Cargo.toml

@@ -29,11 +29,12 @@ headless = []
 ipc = ["webxr-api/ipc", "serde"]
 googlevr = ["gvr-sys", "android_injected_glue", "gles"]
 magicleap = ["egl"]
-openxr-api = ["angle", "openxr", "winapi", "wio"]
+openxr-api = ["angle", "openxr", "winapi", "wio", "surfman/sm-angle-default"]


Add a comment that sm-angle-default only sets the default on windows?

asajeffrey · 2020-02-28T21:51:57Z

webxr/openxr/mod.rs

+ let _ = provider_rx.recv();
+ }));
+ // Get the D3D11 device pointer from the webgl thread.
+ let device = device_rx.recv().unwrap();


This will probably be unnecessary in surfmanup, since the openxr device will own the d3d11 device.

The fundamental point of these changes is to ensure that OpenXR uses the WebGL thread's d3d device, rather than an independent one. Without that sharing, the WebGL thread can't share the OpenXR textures.

Can we do it the other way round, where WebGL uses the device provided by OpenXR?

asajeffrey · 2020-02-28T21:54:00Z

webxr/openxr/mod.rs

- // XXXManishearth what if the recommended widths are different?
- width: left_view_configuration.recommended_image_rect_width,
+ width: left_view_configuration.recommended_image_rect_width
+  + right_view_configuration.recommended_image_rect_width,


Ah, HL lets you do this? And it doesn't produce an extra blit inside openxr? Hooray!

It still blits internally. We weren't on the fast path before this change though, either.

Oh boo for the extra blit. It would be sooooo nice if we could do this and hit HL's fast path.

asajeffrey · 2020-02-28T22:12:36Z

webxr/openxr/mod.rs

- self.gl.bind_framebuffer(gl::FRAMEBUFFER, old_framebuffer);
-
- self.left_swapchain.release_image().unwrap();
- self.right_swapchain.release_image().unwrap();


Goodbye horrible GL code! \o/

jdm · 2020-03-02T22:19:25Z

Pushed the updated changes that fix the d3d threading issues in babylon demos.

jdm · 2020-03-04T17:56:09Z

The latest changes in this branch fix #113 as well, which boosts us from 50fps to 60fps on the three.js paint, ball shooter, and lorenz attractor demos!

asajeffrey · 2020-03-05T00:15:04Z

You can r=me when it's ready to land.

jdm · 2020-03-05T20:38:21Z

@bors-servo r=asajeffrey,Manishearth

bors-servo · 2020-03-05T20:38:23Z

📌 Commit 400fbf5 has been approved by asajeffrey,Manishearth

bors-servo · 2020-03-05T20:38:27Z

⌛ Testing commit 400fbf5 with merge 3ac3e83...

bors-servo · 2020-03-05T21:32:25Z

☀️ Test successful - checks-travis
Approved by: asajeffrey,Manishearth
Pushing 3ac3e83 to master...

Remove GL->d3d blit in HoloLens immersive mode Depends on: * servo/surfman#151 * servo/surfman-chains#7 * servo/webxr#133 These changes add two extra APIs for embedders to use when registering a WebXR device - one to allow running any closure as a task in the webgl thread, and one to register an arbitrary surface provider for a particular webxr session. When an openxr session is started, it can then obtain the webgl thread's d3d device from that thread's surfman device and ensure that openxr uses it. Surface providers are traits that have their methods invoked by the webgl thread as part of the the normal swapchain operations. This allows the openxr surface provider to return surfaces that wrap the underlying openxr textures, which are valid in the webgl thread and can be used as the target of an opaque framebuffer. --- - [x] `./mach build -d` does not report any errors - [x] `./mach test-tidy` does not report any errors - [x] These changes fix #25735 - [x] These changes do not require tests because there are no windows immersive mode tests

jdm mentioned this pull request Feb 27, 2020

Remove GL->d3d blit in HoloLens immersive mode servo/servo#25855

Merged

4 tasks

jdm force-pushed the invert branch from ba540fc to e502a0d Compare February 27, 2020 21:51

asajeffrey reviewed Feb 28, 2020

View reviewed changes

jdm force-pushed the invert branch from e502a0d to 967b345 Compare March 2, 2020 22:19

jdm force-pushed the invert branch 2 times, most recently from ae07c08 to 66eaf7d Compare March 4, 2020 17:42

Avoid extra blit for GL->D3D.

400fbf5

jdm force-pushed the invert branch from 66eaf7d to 400fbf5 Compare March 4, 2020 23:15

Manishearth approved these changes Mar 5, 2020

View reviewed changes

bors-servo merged commit 3ac3e83 into servo:master Mar 5, 2020

This was referenced Mar 5, 2020

Handle STOPPING state correctly #109

Closed

openxr: Use recommended swapchain sample count. #111

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Avoid extra blit for GL->D3D. #133

Avoid extra blit for GL->D3D. #133

jdm commented Feb 27, 2020 •

edited

Loading

asajeffrey left a comment

asajeffrey Feb 28, 2020

jdm Mar 2, 2020

asajeffrey Feb 28, 2020

asajeffrey Feb 28, 2020

jdm Feb 28, 2020

asajeffrey Feb 28, 2020

asajeffrey Feb 28, 2020

jdm Feb 28, 2020 •

edited

Loading

asajeffrey Feb 28, 2020

asajeffrey Feb 28, 2020

jdm commented Mar 2, 2020

jdm commented Mar 4, 2020

asajeffrey commented Mar 5, 2020

jdm commented Mar 5, 2020

bors-servo commented Mar 5, 2020

bors-servo commented Mar 5, 2020

bors-servo commented Mar 5, 2020

Avoid extra blit for GL->D3D. #133

Avoid extra blit for GL->D3D. #133

Conversation

jdm commented Feb 27, 2020 • edited Loading

asajeffrey left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jdm Feb 28, 2020 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jdm commented Mar 2, 2020

jdm commented Mar 4, 2020

asajeffrey commented Mar 5, 2020

jdm commented Mar 5, 2020

bors-servo commented Mar 5, 2020

bors-servo commented Mar 5, 2020

bors-servo commented Mar 5, 2020

jdm commented Feb 27, 2020 •

edited

Loading

jdm Feb 28, 2020 •

edited

Loading