jachym/dwl
1
0
Fork 0
mirror of https://codeberg.org/dwl/dwl.git synced 2026-02-04 10:13:07 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

fix: let wlroots determine frame_done timestamp for PRIME offload

Browse Source 
When using PRIME GPU offload (e.g., NVIDIA renders, AMD displays),
capturing the timestamp via clock_gettime in rendermon() causes
frame timing issues. The timestamp is captured from the display
GPU's clock, but rendering occurs on the offload GPU with different
timing. This clock divergence causes stuttering that appears as
"frames going backward in time" when the mouse is moved.

By passing NULL to wlr_scene_output_send_frame_done(), we let
wlroots determine the appropriate timestamp, which handles the
PRIME offload case correctly.
This commit is contained in:
Bailey Stoner 2025-12-21 09:28:04 -08:00
parent 6cd26568d5
commit dd163a8ec4
1 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
dwl.c
View File

@ -2154,7 +2154,6 @@ rendermon(struct wl_listener *listener, void *data)
Monitor *m = wl_container_of(listener, m, frame);
Client *c;
struct wlr_output_state pending = {0};
struct timespec now;
/* Render if no XDG clients have an outstanding resize and are visible on
* this monitor. */
@ -2166,9 +2165,10 @@ rendermon(struct wl_listener *listener, void *data)
wlr_scene_output_commit(m->scene_output, NULL);
skip:
/* Let clients know a frame has been rendered */
clock_gettime(CLOCK_MONOTONIC, &now);
wlr_scene_output_send_frame_done(m->scene_output, &now);
/* Let clients know a frame has been rendered. Pass NULL to let wlroots
* determine the timestamp, which is required for correct timing when
* input and output clocks may diverge (e.g., PRIME GPU offload). */
wlr_scene_output_send_frame_done(m->scene_output, NULL);
wlr_output_state_finish(&pending);
}