<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.0 TRANSITIONAL//EN"> <HTML> <HEAD> <META HTTP-EQUIV="Content-Type" CONTENT="text/html; CHARSET=UTF-8"> <META NAME="GENERATOR" CONTENT="GtkHTML/4.6.1"> </HEAD> <BODY> [ In reverse order ] We start off with my patches. I haven't yet written docs for cogl-output.c (trying to get this out tonight) ======== commit 55ea53b5b4e7c8352bea59d25cc0cdf47a968229 Author: Owen W. Taylor <<A HREF="mailto:otaylor@fishsoup.net">otaylor@fishsoup.net</A>> Date: Mon Nov 12 11:31:16 2012 -0500 Add CoglOutput and track for the GLX backend The CoglOutput object represents one output such as a monitor or laptop panel, with information about attributes of the output such as the position of the output within the global coordinate space, and the refresh rate. We don't yet publically export the ability to get output information but we track it for the GLX backend, where we'll use it to track the refresh rate. commit a0de9e74d97ef44c327bd1daef800aa8667e9938 Author: Owen W. Taylor <<A HREF="mailto:otaylor@fishsoup.net">otaylor@fishsoup.net</A>> Date: Mon Jan 7 12:24:13 2013 -0500 Add missing cogl-output.c Should have been part of "Add CoglOutput and track for the GLX backend" commit 394a162738f30f9cbf975f93bb9e15399715c423 Author: Owen W. Taylor <<A HREF="mailto:otaylor@fishsoup.net">otaylor@fishsoup.net</A>> Date: Mon Nov 12 11:52:31 2012 -0500 Prefer OML_sync_control over SGI_video_sync when waiting for swap When we block waiting for the swap, prefer doing that using glXWaitForMsc() from OML_sync_control because that returns a system time value for the precise time of the swap. commit 646c2805342db272a1af53537aacbee1fbb477b2 Author: Owen W. Taylor <<A HREF="mailto:otaylor@fishsoup.net">otaylor@fishsoup.net</A>> Date: Mon Nov 12 11:58:10 2012 -0500 Add CoglFrameTimings Add a CoglFrameTimings object that tracks timing information for frames that are drawn. We track a frame counter and frame timing information for each CoglOnscreen. Frames timing information is deliminated by a new cogl_onscreen_begin_frame() and retrieved using cogl_onscreen_get_frame_timings(). ======== Then a bunch of fixes and minor changes from you that I have no issues with. ======== commit 98d2be959c02ef911fdfe038590c226dc8681a1b Author: Robert Bragg <<A HREF="mailto:robert@linux.intel.com">robert@linux.intel.com</A>> Date: Mon Dec 3 23:23:37 2012 +0000 pedantic style fixes TODO: squash back into owen's patch commit 4a881b2fed030981558e02ff1b28f07606cc0692 Author: Robert Bragg <<A HREF="mailto:robert@linux.intel.com">robert@linux.intel.com</A>> Date: Mon Dec 3 23:15:48 2012 +0000 use c99 types if applicable and s/gboolean/CoglBool/ TODO: squash back into owen's patch commit ac24c9db3ed126f8c4c450b258bd26f73489b4be Author: Robert Bragg <<A HREF="mailto:robert@linux.intel.com">robert@linux.intel.com</A>> Date: Mon Dec 3 23:35:14 2012 +0000 move outputs member up to CoglRenderer TODO: squash back into owen's patch commit 2619778124ba347343fb62bd5c9b22e8bbdcabc2 Author: Robert Bragg <<A HREF="mailto:robert@linux.intel.com">robert@linux.intel.com</A>> Date: Mon Dec 3 23:00:15 2012 +0000 rename CoglFrameTimings to CoglSwapInfo The aim is to pass an opaque structure to the swap buffers notification callbacks and enable access to frame timing information this way. Since there could be numerous non-timing related things to report here too this gives the structure a more general name. TODO: fold this back into owen's patch commit 1b6ccda7b65a1b13f60cdd10fa166d4ac9a4ed79 Author: Robert Bragg <<A HREF="mailto:robert@linux.intel.com">robert@linux.intel.com</A>> Date: Mon Dec 3 23:16:01 2012 +0000 use c99 types if applicable and s/gboolean/CoglBool/ TODO: squash back into owen's patch commit 30ad78f96af0cc4bea8a1aca442500d16ba496e6 Author: Robert Bragg <<A HREF="mailto:robert@linux.intel.com">robert@linux.intel.com</A>> Date: Wed Jan 9 20:46:43 2013 +0000 rename SwapInfo to FrameInfo As discussed on the mailing list we can be slightly more consistent with gtk by using "frame" instead of "swap" here. Also we discussed the potential for being able to query information for a frame that's still in progress and in that case "swap" wouldn't make much sense. TODO: squash this back into owen's patch commit 0dd495d28418db0ad6ffd0d8078a27f01c2eba42 Author: Robert Bragg <<A HREF="mailto:robert@linux.intel.com">robert@linux.intel.com</A>> Date: Tue Jan 15 15:37:49 2013 +0000 frame-info: s/_refresh_interval/_refresh_rate/ For consistency this reworks the cogl_frame_info_get_refresh_interval api to instead be cogl_frame_info_get_refresh_rate() and return a frequency in Hertz instead of an interval in milliseconds. This is consistent with cogl_output_get_refresh_rate(). commit c7bda39491fd63cedec92ef79cf2cb9fe3df0fbd Author: Robert Bragg <<A HREF="mailto:robert@linux.intel.com">robert@linux.intel.com</A>> Date: Mon Jan 21 01:18:25 2013 +0000 renderer: expose CoglOutputs This adds a cogl_renderer_foreach_output() function that can be used to iterate the display outputs for a particular renderer. This also updates cogl-info to use this new api so it can dump out all the output information. ======== I also have no issue with the basic idea of the new frame callback API ======== commit 8a1a1f56135af22c7316529f9e6f400059e305b8 Author: Robert Bragg <<A HREF="mailto:robert@linux.intel.com">robert@linux.intel.com</A>> Date: Thu Jan 10 19:23:48 2013 +0000 onscreen: Provide more general CoglFrameCallback api This reworks the cogl_onscreen_add_swap_complete_callback api to instead be named cogl_onscreen_add_frame_callback and to take a callback that can accept other events besides just swap completion events. This means we can add more events in the future that signal the progression of a frame without needing lots of new callback api for each event. TODO: squash this back into owen's patch ======== At this point, we get to a bunch of things extracted from your patch that I don't quite agree with, but I'm willing to go with. cogl_frame_info_get_output() doesn't seem useful to me: ======== commit 5cc26c82555e01ca7370772b73b93ca00772c37a Author: Owen W. Taylor <<A HREF="mailto:otaylor@fishsoup.net">otaylor@fishsoup.net</A>> Date: Thu Jan 24 17:38:38 2013 -0500 Add cogl_frame_info_get_output() If available from the windowing system, it could potentially be useful to know something other than the refresh rate of the output that the monitor is currently presenting on. (Note however, an onscreen may span multiple outputs. The windowing system has to pick *one* for timing, but most other attibutes of the window will apply to the parts of the onscreen on that output.) ======== See discussion of frame time elsewhere. I think it simplifies the lives of people trying to use CoglFrameInfo, but it's not required, minimal low-level functionality, certainly. ======== commit 38ac56be4a0af4048091cfa3f7fd206a0d4db424 Author: Owen W. Taylor <<A HREF="mailto:otaylor@fishsoup.net">otaylor@fishsoup.net</A>> Date: Thu Jan 24 22:12:47 2013 -0500 Remove cogl_frame_info_get_frame_time() The frame_time isn't closely associated with an onscreen - an update could involve multiple stages (in a stage-per-output scenario, perhaps.) If an application or toolkit needs to know the frame time for a FrameInfo it will have to do that association itself. ======== Switching away from g_get_monotonic_time() is needed if Cogl is to be compiled without GLib. The patch here keeps the translation of gettimeofday() UST values into clock_gettime(CLOCK_MONOTONIC, ...). ======== commit 67bc4064ea0f63fdc3a751c4cf1df3edfeeceb2f Author: Owen W. Taylor <<A HREF="mailto:otaylor@fishsoup.net">otaylor@fishsoup.net</A>> Date: Thu Jan 24 22:32:39 2013 -0500 Don't use g_get_monotonic_time() Avoid a hard dependency on GLib and on however GLib is implementing g_get_monotonic_time() by directly referencing clock_gettime(CLOCK_MONOTONIC). The translation of timestamp values for unidentified types of UST is also removed, since it's going to have a big error, and glXGetSyncValues() doesn't return the *current* UST anyways - it returns the UST for the last frame. ======== Hopefully multiplying by 1000 is safe - in which case, I don't care about usec vs. nanosec. ======== commit 36bbcb083156926f50c0b3c3e0b42e24f42fdbf7 Author: Owen W. Taylor <<A HREF="mailto:otaylor@fishsoup.net">otaylor@fishsoup.net</A>> Date: Thu Jan 24 22:43:10 2013 -0500 Switch presentation time to nanoseconds In the future, we may want to track frame timings with very high precision for profiling purposes, so to be consistent with that, use nanoseconds, not microseconds. 63-bits in nanoseconds is 270+ years, so hopefully clock_gettime(CLOCK_MONOTONIC,) will fit. ======== The next patch is a big hunk of changes, but is mostly about removing the history ring in favor of a queue. I don't know how this will work out when we start implementing more complex scenarios than we have currently implemented - like the compositor scenario with _NET_WM_FRAME_DRAWN and _NET_WM_FRAME_TIMINGS - but if we expose nothing, we can always adjust later. There is a potential problem of having a queue of unlimited length - if someone wrote a all-draw-no-events program, the queue would grow without bound. ======== commit 41de98629a23f594715c703d26671d039085cf93 Author: Owen W. Taylor <<A HREF="mailto:otaylor@fishsoup.net">otaylor@fishsoup.net</A>> Date: Thu Jan 24 23:16:25 2013 -0500 Replace frame history with a queue An arbitrary-length frame history seems hard to size, and can be implemented at the application level, so ditch it for a queue of frame info objects that are not yet complete. ======== cogl_frame_info_get_complete() would be useful - e.g., if an application is keeping an application-side history - knowing what frames can be used now and aren't waiting for COMPLETE events. Though with COMPLETE events it can be emulated. Can be added later. ======== commit 47bf79c31f0d8de8c1165d3fe9818a5021a17f1c Author: Owen W. Taylor <<A HREF="mailto:otaylor@fishsoup.net">otaylor@fishsoup.net</A>> Date: Thu Jan 24 23:29:04 2013 -0500 Remove cogl_frame_info_get_complete() An application gets a callback when a frame info is complete, so there may not be a need for a flag as well. ======== I think the main question for me is whether you are OK with gettimeofday => CLOCK_MONOTONIC conversion so we have a known scale. Otherwise, the diffs between the two branches are detail, and there's nothing that I've pushed to wip/frame-synchronization that I have a hard objection to. - Owen </BODY> </HTML>