backend/headless: use FBOs instead of pbuffers #2063

emersion · 2020-03-10T16:17:22Z

Allows the headless backend to use another backend's renderer.

ammen99 · 2020-03-11T12:52:27Z

Shouldn't wlroots also offer a way to get the FB to render to? For example, Wayfire also switches between various FBs when rendering. Until now, I have always tried to reset to FB 0 to render the the output, which seems to not be the case anymore.

ammen99 · 2020-03-11T12:54:04Z

Oh and by the way, doesn't this make the headless backend dependent on a GLES renderer? From my understanding, this wasn't the case before, but I am not very familiar with EGL.

emersion · 2020-03-11T13:19:36Z

Shouldn't wlroots also offer a way to get the FB to render to? For example, Wayfire also switches between various FBs when rendering. Until now, I have always tried to reset to FB 0 to render the the output, which seems to not be the case anymore.

If you switch between FBs, you should save the previous value with glGet(GL_FRAMEBUFFER_BINDING) and restore it afterwards.

Oh and by the way, doesn't this make the headless backend dependent on a GLES renderer? From my understanding, this wasn't the case before, but I am not very familiar with EGL.

Yes, that's true. Could be fixed by using eglGetProcAddress for all GL functions.

ammen99 · 2020-03-11T13:58:46Z

Yes, that's true. Could be fixed by using eglGetProcAddress for all GL functions.

My understanding was that we want to make it possible to write a pixman-based renderer or something :P (note I do not care about this case, just pointing it out since somebody does)

emersion · 2020-03-11T14:03:33Z

My understanding was that we want to make it possible to write a pixman-based renderer or something :P (note I do not care about this case, just pointing it out since somebody does)

Yeah, I care about this, but I don't think it's related. Currently the headless backend depends on EGL, so this PR doesn't change a lot. With renderer v6 the compositor will manage swapchains directly (or indirectly via helpers) so the backend won't even care.

(My eglGetProcAddress comment was about desktop OpenGL.)

any1 · 2020-03-11T21:36:39Z

CPU load in sway due to glReadPixels() is doubled when running weston-simple-egl in wayvnc after this change is applied. Maybe a different pixel format would give better performance?

emersion · 2020-03-13T00:41:52Z

Can you try GL_RGBA4 and GL_RGBA8_OES instead of GL_RGB8_OES?

any1 · 2020-03-13T21:30:50Z

Yes, using RGBA8_OES performs much better. GL_RGBA4 actually performs much worse.

Most of the pending output state is not forwarded to the backend prior to an output commit. For instance, wlr_output_set_mode just stashes the mode without calling any wlr_output_impl function. wlr_output_impl.commit is responsible for applying the pending mode. However, there are exceptions to this rule. The first one is wlr_output_attach_render. It won't go away before renderer v6 is complete, because it needs to set the current EGL surface. The second one is wlr_output_attach_buffer. wlr_output_impl.attach_buffer is removed in [1]. When wlr_output_rollback is called, all pending state is supposed to be cleared. This works for all the state except the two exceptions mentionned above. To fix this, introduce wlr_output_impl.rollback. Right now, the backend resets the current EGL surface. This prevents GL commands from affecting the output after wlr_output_rollback. This patch is required for FBO-based outputs to work properly. The compositor might be using FBOs for its own purposes [2], having leftover FBO state can have bad consequences. [1]: swaywm#2097 [2]: swaywm#2063 (comment)

emersion · 2020-04-06T10:26:42Z

This is now ready for review.

~~(Note that this PR depends on another PR, so I'll keep it as a draft to prevent it from being merged.)~~

Most of the pending output state is not forwarded to the backend prior to an output commit. For instance, wlr_output_set_mode just stashes the mode without calling any wlr_output_impl function. wlr_output_impl.commit is responsible for applying the pending mode. However, there are exceptions to this rule. The first one is wlr_output_attach_render. It won't go away before renderer v6 is complete, because it needs to set the current EGL surface. The second one is wlr_output_attach_buffer. wlr_output_impl.attach_buffer is removed in [1]. When wlr_output_rollback is called, all pending state is supposed to be cleared. This works for all the state except the two exceptions mentionned above. To fix this, introduce wlr_output_impl.rollback. Right now, the backend resets the current EGL surface. This prevents GL commands from affecting the output after wlr_output_rollback. This patch is required for FBO-based outputs to work properly. The compositor might be using FBOs for its own purposes [2], having leftover FBO state can have bad consequences. [1]: swaywm#2097 [2]: swaywm#2063 (comment)

Most of the pending output state is not forwarded to the backend prior to an output commit. For instance, wlr_output_set_mode just stashes the mode without calling any wlr_output_impl function. wlr_output_impl.commit is responsible for applying the pending mode. However, there are exceptions to this rule. The first one is wlr_output_attach_render. It won't go away before renderer v6 is complete, because it needs to set the current EGL surface. The second one is wlr_output_attach_buffer. wlr_output_impl.attach_buffer is removed in [1]. When wlr_output_rollback is called, all pending state is supposed to be cleared. This works for all the state except the two exceptions mentionned above. To fix this, introduce wlr_output_impl.rollback. Right now, the backend resets the current EGL surface. This prevents GL commands from affecting the output after wlr_output_rollback. This patch is required for FBO-based outputs to work properly. The compositor might be using FBOs for its own purposes [2], having leftover FBO state can have bad consequences. [1]: #2097 [2]: #2063 (comment)

backend/headless/backend.c

any1 · 2020-04-09T19:34:18Z

I tested this. It works well and there are no obvious regressions.

emersion · 2020-04-14T12:45:37Z

Rebased and added a fallback for the internal format.

any1 · 2020-04-14T13:11:43Z

The fallback is perhaps not terribly important, but might I suggest RGB565 instead?

backend/headless/backend.c

ddevault · 2020-04-14T14:05:44Z

Overall, this LGTM

emersion · 2020-04-21T13:07:29Z

Comments addressed.

This allows one to create a headless backend with an existing renderer.

ddevault · 2020-04-22T20:40:57Z

Thanks!

emersion force-pushed the headless-fbo branch from fac19e3 to fa35e9f Compare March 23, 2020 16:25

emersion mentioned this pull request Apr 6, 2020

Reset EGL current surface after output commit/rollback #2104

Merged

emersion force-pushed the headless-fbo branch from fa35e9f to 8149d89 Compare April 6, 2020 10:22

emersion force-pushed the headless-fbo branch from 4ece54e to 0366ab2 Compare April 8, 2020 15:56

emersion marked this pull request as ready for review April 8, 2020 15:56

any1 reviewed Apr 9, 2020

View reviewed changes

backend/headless/backend.c Outdated Show resolved Hide resolved

emersion force-pushed the headless-fbo branch from 0366ab2 to 057318a Compare April 14, 2020 12:09

emersion requested a review from ddevault April 14, 2020 13:13

ddevault reviewed Apr 14, 2020

View reviewed changes

backend/headless/backend.c Outdated Show resolved Hide resolved

ddevault reviewed Apr 14, 2020

View reviewed changes

backend/headless/backend.c Outdated Show resolved Hide resolved

emersion mentioned this pull request Apr 15, 2020

Add a secondary headless backend swaywm/sway#5216

Merged

render/gles2: add wlr_gles2_renderer_check_ext

b00a71e

emersion force-pushed the headless-fbo branch 2 times, most recently from f61db96 to 6ada106 Compare April 21, 2020 13:01

emersion force-pushed the headless-fbo branch from 6ada106 to b66de00 Compare April 21, 2020 13:03

emersion force-pushed the headless-fbo branch from b66de00 to db18bf5 Compare April 21, 2020 13:08

emersion added 2 commits April 21, 2020 15:09

backend/headless: use FBOs instead of pbuffers

ed01aa7

backend/headless: add wlr_headless_backend_create_with_renderer

75a78c7

This allows one to create a headless backend with an existing renderer.

emersion force-pushed the headless-fbo branch from db18bf5 to 75a78c7 Compare April 21, 2020 13:10

ddevault approved these changes Apr 22, 2020

View reviewed changes

ddevault merged commit 6129a6f into swaywm:master Apr 22, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

backend/headless: use FBOs instead of pbuffers #2063

backend/headless: use FBOs instead of pbuffers #2063

emersion commented Mar 10, 2020 •

edited

Loading

ammen99 commented Mar 11, 2020

ammen99 commented Mar 11, 2020

emersion commented Mar 11, 2020

ammen99 commented Mar 11, 2020 •

edited

Loading

emersion commented Mar 11, 2020

any1 commented Mar 11, 2020 •

edited

Loading

emersion commented Mar 13, 2020

any1 commented Mar 13, 2020

emersion commented Apr 6, 2020 •

edited

Loading

any1 commented Apr 9, 2020

emersion commented Apr 14, 2020

any1 commented Apr 14, 2020

ddevault commented Apr 14, 2020

emersion commented Apr 21, 2020

ddevault commented Apr 22, 2020

backend/headless: use FBOs instead of pbuffers #2063

backend/headless: use FBOs instead of pbuffers #2063

Conversation

emersion commented Mar 10, 2020 • edited Loading

ammen99 commented Mar 11, 2020

ammen99 commented Mar 11, 2020

emersion commented Mar 11, 2020

ammen99 commented Mar 11, 2020 • edited Loading

emersion commented Mar 11, 2020

any1 commented Mar 11, 2020 • edited Loading

emersion commented Mar 13, 2020

any1 commented Mar 13, 2020

emersion commented Apr 6, 2020 • edited Loading

any1 commented Apr 9, 2020

emersion commented Apr 14, 2020

any1 commented Apr 14, 2020

ddevault commented Apr 14, 2020

emersion commented Apr 21, 2020

ddevault commented Apr 22, 2020

emersion commented Mar 10, 2020 •

edited

Loading

ammen99 commented Mar 11, 2020 •

edited

Loading

any1 commented Mar 11, 2020 •

edited

Loading

emersion commented Apr 6, 2020 •

edited

Loading