Game: grass flickers pale white for about a second in the cutscene after learning Epona's Song (Lon Lon Ranch, Steam Deck) #1760
Labels
No labels
area/accessibility
area/audio
area/input
area/install
area/menu
area/renderer
area/saves
area/tooling
area/upscaling
beanstalk: approved
beanstalk: failed
beanstalk: needs approval
beanstalk: running
beanstalk: tested
ci: approved
good first issue
gpu/amd
gpu/igpu
gpu/intel
gpu/nvidia
help wanted
kind
bug
kind
chore
kind
docs
kind
feature
kind
performance
kind
question
needs measurements
platform/linux
platform/macos
platform/steam-deck
platform/windows
preview
priority
critical
priority
high
priority
low
regression
risk
high
risk
low
risk
medium
status
blocked
status
confirmed
status
duplicate
status
invalid
status
needs-info
status
wontfix
ci: approved
No milestone
No project
No assignees
2 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
MagicBeanDev/Longshot#1760
Loading…
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Longshot version
v3.0.0 (source 87cb36aa), a build of the release tag for the Steam Deck.
What kind of problem
Graphics: wrong picture.
What happened
In the short cutscene after Malon teaches Epona's Song at Lon Lon Ranch, the view cuts to an overhead shot of Link and Malon and then pans up over the corral. During that pan, clusters of grass on the ground turn pale white-lavender, mostly inside and at the edges of the long shadows on the ground (the fence's, Link's, the horse's). The patches jump to different places every frame and last about one second (18 to 26 presents in a row), then the picture is right again. Seen by hand in play on the Deck's screen, and reproduced headless on the same build in 6 runs of 6, at the same point of the cutscene each time.
Expected: the grass keeps its colour through the pan.
Steps: a save standing at Malon in the corral as a child with the Fairy Ocarina, before learning the song. Take out the ocarina, talk to Malon, play the song back when she asks. The glitch is in the pan that follows the last note, about 380 presents after it at 21 fps.
Where in the game
Lon Lon Ranch (child, daytime), the cutscene right after learning Epona's Song from Malon.
How often
Every time: 6 of 6 scripted runs, and the one time it was played by hand.
Graphics card and driver version
Steam Deck OLED, RADV VANGOGH, Mesa 26.2.0-devel.
CPU, RAM and operating system
AMD Custom APU 0932, 16 GB, SteamOS 3.9.5.
Changes from the defaults
Graphics: Custom, 1280x800, Render resolution 100%, the game's own upscaler, sharpness 50, frame generation off, the 60 cap.
longshot_env.txt:RX_VK_READBACK_LATE=newest,HOS_SHORT_SLEEP_TIMER=1,HOS_NATIVE_ZSTD=1. No overlay or injector. Mesh shaders run on the compute path (Automatic on RADV).What does not change it, one change each on the same scripted route (pale presents counted from a screenshot of every present, checked by eye):
RX_VK_READBACK_LATEunsetRX_VK_READBACK_LATEunsetHOS_NATIVE_ZSTDunsetSo it is not the late read-backs (which I suspected first, since a late exposure read was the earlier white flashes and #1725 would turn them on for the Deck), not the render scale and not the zstd core rule. The fps is lower than in play because every present was saved as a screenshot.
The flash detector (
RX_VK_FLASH=1) does not flag it: it looks for a frame that differs from both neighbours, and these patches keep moving for a second. In the same run it did flag one frame while the song plays where the whole picture is visibly darker for a single present; I have not looked into that one.Not tried yet: another upscaler or none, native mesh shaders, another GPU, the frame before the cut.
longshot_log.txt
The play session's log shows no error, device-loss or fallback line around the cutscene. I can attach the log and the scripted runs' logs if useful.
Screenshots or video
I have a contact sheet of every present across the pan (40 presents, the pale patches from the 14th on) and can post it here; the forge's command line does not attach files, so say if it is wanted and I'll add it through the web page.
Before you send it
Possibly related, for context:
The contact sheet mentioned above: every present across the pan, left to right and top to bottom, 40 presents from one scripted run on stock v3.0.0 at the settings in the report (320x200 captures). The pale patches start in the third row and are gone by the last.