MagicBeanDev/Longshot

Watch

Fork 71

Game: #366

New issue

Open

opened 2026-11-14 05:07:27 +00:00 by Takuram · 5 comments

Takuram commented

2026-11-14 05:07:27 +00:00

Longshot version

nightly-127-cf605c85 (source 8672daee)

Channel

Nightly

What kind of problem

Graphics (missing, black or wrong-coloured things, lighting, flicker)

What happened

Weird black flicker on some characters, specially during cutscenes.

Where in the game

SacredForestMeadow_Adult (adult)

How often

Sometimes

Graphics card and driver version

NVIDIA GeForce RTX 4060, NVIDIA 617.14

CPU, RAM and operating system

AMD Ryzen 5 5600 6-Core Processor (12 threads), 16 GB, Windows 11 (build 26100)

Changes from the defaults

Frame generation: 2x (default: Off)
Fullscreen: On (default: Off)
V-Sync: Off (default: Auto)
Anisotropic filtering: 4x (default: 16x)
FPS counter: FPS (default: Off)
Menu size: Small (default: Medium)
Layout preset: Swap Face (default: Nintendo (Def))
Button prompts: Xbox (default: Auto)
Depends on this PC (the values now, not counted as changes):
Quality preset: Custom (default depends on this PC)
Output resolution: 2560x1440 (native) (default depends on this PC)
Profile: Docked (default depends on this PC)
Upscaler: NVIDIA DLSS (default depends on this PC)
Render resolution: Quality 67% (1707x960) (default depends on this PC)
Sharpness: 30% (default depends on this PC)
Frame pacing: Off (default depends on this PC)
Window scaling filter: Not needed (window = output) (default depends on this PC)
VRAM budget: 3.6 GB (default depends on this PC)
Game speed: Real time (default depends on this PC)
Switches set (their values are in the attached report.txt): LONGSHOT_NO_UPDATE_CHECK.

longshot_log.txt

attached

Screenshots or video

No response

Before you send it

The problem is still there on the newest release or nightly.
I searched the open issues for it.
I'm not attaching game files, keys, the game's executable, its .bin caches or saves.

### Longshot version nightly-127-cf605c85 (source 8672daee) ### Channel Nightly ### What kind of problem Graphics (missing, black or wrong-coloured things, lighting, flicker) ### What happened Weird black flicker on some characters, specially during cutscenes. ### Where in the game SacredForestMeadow_Adult (adult) ### How often Sometimes ### Graphics card and driver version NVIDIA GeForce RTX 4060, NVIDIA 617.14 ### CPU, RAM and operating system AMD Ryzen 5 5600 6-Core Processor (12 threads), 16 GB, Windows 11 (build 26100) ### Changes from the defaults Frame generation: 2x (default: Off) Fullscreen: On (default: Off) V-Sync: Off (default: Auto) Anisotropic filtering: 4x (default: 16x) FPS counter: FPS (default: Off) Menu size: Small (default: Medium) Layout preset: Swap Face (default: Nintendo (Def)) Button prompts: Xbox (default: Auto) Depends on this PC (the values now, not counted as changes): Quality preset: Custom (default depends on this PC) Output resolution: 2560x1440 (native) (default depends on this PC) Profile: Docked (default depends on this PC) Upscaler: NVIDIA DLSS (default depends on this PC) Render resolution: Quality 67% (1707x960) (default depends on this PC) Sharpness: 30% (default depends on this PC) Frame pacing: Off (default depends on this PC) Window scaling filter: Not needed (window = output) (default depends on this PC) VRAM budget: 3.6 GB (default depends on this PC) Game speed: Real time (default depends on this PC) Switches set (their values are in the attached report.txt): LONGSHOT_NO_UPDATE_CHECK. ### longshot_log.txt attached ### Screenshots or video _No response_ ### Before you send it - [x] The problem is still there on the newest release or nightly. - [x] I searched the open issues for it. - [x] I'm not attaching game files, keys, the game's executable, its .bin caches or saves.

longshot_log.txt

4.3 MiB

Takuram added the

kind

bug

label

2026-11-14 05:07:27 +00:00

Rogasse commented

2026-11-14 16:34:48 +00:00

Additional NVIDIA/Windows reproduction on Longshot nightly-127-cf605c85 (source 8672daee). This looks similar to the black character flicker in this issue, though my location is Zora's Fountain rather than Sacred Forest Meadow.

System: RTX 4050 Laptop GPU (6 GB), NVIDIA 621.14; Intel Core i5-13420H; 32 GB RAM; Windows 11 Home build 26200.

Observation: In ZoraFountain_Child, Link briefly becomes a much darker, near-black silhouette. The attached three-frame crop shows a single-frame event in the middle image, around 0:08 of a local 20-second recording; another event appears around 0:14. The full picture did not go black in those captured frames. The issue also occurred with frame generation Off, so generated frames are not required to trigger it. HDR was Off and the game's own upscaler was selected. I have not retested this on a newer nightly yet.

Same-scene A–B–A, Nightly 127, frame generation Off:

Default (RX_VK_LATE_FIX on): the player saw the flicker.
One launch with RX_VK_LATE_FIX=0: no flicker observed for 1–2 minutes in the same location.
Default again: the flicker reappeared.

This implicates the GPU late-pages path but does not identify the faulty line. The observation windows were short and no FPS cost of the workaround was measured. RX_VK_MIRROR_SPARSE remained on in both arms, so an interaction with sparse mirrors is still possible. Relevant lines from the original default-run log:

longshot nightly-127-cf605c85 (source 8672daee)
vk: sparse VRAM mirrors (RX_VK_MIRROR_SPARSE): on
vk: late pages' fix on (RX_VK_LATE_FIX=0: the wait, as before)
map: ZoraFountain_Child (child)

Source lead: PR #248 changed late_fix_build/late_fix_emit in gpu/vk/vk_memory.cpp and added shaders/late_fix.comp, which patches captured pages from the pool shadow using per-word write-back stamps. The PR description documents black character flashes in an earlier version due to stale per-character animation words and notes that a wrong capture word would appear as a one-frame artifact. The focused late_fix_build and shadow-sync source tests pass locally, but they do not reproduce this gameplay scene; the SPIR-V regeneration check was unavailable because glslangValidator is not installed here. A capture-word comparison in Zora or a same-scene run with sparse mirrors off while the late fix stays on could narrow the exact cause.

Workaround on this PC: RX_VK_LATE_FIX=0 in longshot_env.txt for future launches. The screenshot is a crop from the local recording; no game files, saves, executables or caches are attached.

Additional NVIDIA/Windows reproduction on **Longshot nightly-127-cf605c85 (source 8672daee)**. This looks similar to the black character flicker in this issue, though my location is Zora's Fountain rather than Sacred Forest Meadow. **System:** RTX 4050 Laptop GPU (6 GB), NVIDIA 621.14; Intel Core i5-13420H; 32 GB RAM; Windows 11 Home build 26200. **Observation:** In `ZoraFountain_Child`, Link briefly becomes a much darker, near-black silhouette. The attached three-frame crop shows a single-frame event in the middle image, around 0:08 of a local 20-second recording; another event appears around 0:14. The full picture did not go black in those captured frames. The issue also occurred with frame generation **Off**, so generated frames are not required to trigger it. HDR was Off and the game's own upscaler was selected. I have not retested this on a newer nightly yet. **Same-scene A–B–A, Nightly 127, frame generation Off:** 1. Default (`RX_VK_LATE_FIX` on): the player saw the flicker. 2. One launch with `RX_VK_LATE_FIX=0`: no flicker observed for 1–2 minutes in the same location. 3. Default again: the flicker reappeared. This implicates the GPU late-pages path but does **not** identify the faulty line. The observation windows were short and no FPS cost of the workaround was measured. `RX_VK_MIRROR_SPARSE` remained on in both arms, so an interaction with sparse mirrors is still possible. Relevant lines from the original default-run log: ```text longshot nightly-127-cf605c85 (source 8672daee) vk: sparse VRAM mirrors (RX_VK_MIRROR_SPARSE): on vk: late pages' fix on (RX_VK_LATE_FIX=0: the wait, as before) map: ZoraFountain_Child (child) ``` **Source lead:** [PR #248](https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/248) changed `late_fix_build`/`late_fix_emit` in `gpu/vk/vk_memory.cpp` and added `shaders/late_fix.comp`, which patches captured pages from the pool shadow using per-word write-back stamps. The PR description documents black character flashes in an earlier version due to stale per-character animation words and notes that a wrong capture word would appear as a one-frame artifact. The focused `late_fix_build` and shadow-sync source tests pass locally, but they do not reproduce this gameplay scene; the SPIR-V regeneration check was unavailable because `glslangValidator` is not installed here. A capture-word comparison in Zora or a same-scene run with sparse mirrors off while the late fix stays on could narrow the exact cause. **Workaround on this PC:** `RX_VK_LATE_FIX=0` in `longshot_env.txt` for future launches. The screenshot is a crop from the local recording; no game files, saves, executables or caches are attached. ![event8-compare.png](https://git.magicbean.dev/attachments/c7407e60-07d4-4573-b6ad-ce1f4fcc4798)

event8-compare.png

1.5 MiB

Rogasse commented

2026-11-14 16:41:32 +00:00

Follow-up in the same ZoraFountain_Child save on Nightly 127, with frame generation Off: I launched one session with RX_VK_LATE_FIX=1 and RX_VK_MIRROR_SPARSE=0 (process environment, overriding the local RX_VK_LATE_FIX=0 file). The black flicker on Link still occurred. The run's log confirms late pages' fix on and the Zora map load. Thus sparse mirrors are not required to reproduce this symptom on this PC; the late-pages fix remains the narrower trigger. This does not yet identify the incorrect word or ordering.

One source-test gap worth checking: PR #248's temporary capture-word verification reports 0 pages written back again since the capture in that test. I cannot establish that such a repeated write-back occurs in this Zora scene, but it would be useful to include that case in an instrumented comparison before attributing the flash to it.

Follow-up in the same `ZoraFountain_Child` save on Nightly 127, with frame generation Off: I launched one session with **`RX_VK_LATE_FIX=1` and `RX_VK_MIRROR_SPARSE=0`** (process environment, overriding the local `RX_VK_LATE_FIX=0` file). The black flicker on Link still occurred. The run's log confirms `late pages' fix on` and the Zora map load. Thus sparse mirrors are not required to reproduce this symptom on this PC; the late-pages fix remains the narrower trigger. This does not yet identify the incorrect word or ordering. One source-test gap worth checking: [PR #248](https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/248)'s temporary capture-word verification reports **0 pages written back again since the capture** in that test. I cannot establish that such a repeated write-back occurs in this Zora scene, but it would be useful to include that case in an instrumented comparison before attributing the flash to it.

Rogasse commented

2026-11-14 16:46:47 +00:00

Another observation from the same running Nightly 127 session (RX_VK_LATE_FIX=1, RX_VK_MIRROR_SPARSE=0, frame generation Off): the player sees the black flicker in ZoraFountain_Child but has not noticed it in InsideJabuJabuBelly.

The log shows eight late pages' fix (600 recordings with one) summaries during the Zora portion, typically about 75–87 pages per fixed capture. After the Jabu-Jabu map load, there have been zero such summaries so far, alongside 160 write-back diff (600 submits) summaries. The late-fix counter logs only after 600 nonempty emits, so this does not prove there were zero fixes inside Jabu-Jabu. It does show that this path is much less active there while ordinary write-backs continue. That makes a workload/timing-dependent capture problem a useful lead; it does not identify a particular bad word, shader or scene asset. The next decisive diagnostic remains comparing captured words against the old wait path at the Zora flicker.

Another observation from the **same running Nightly 127 session** (`RX_VK_LATE_FIX=1`, `RX_VK_MIRROR_SPARSE=0`, frame generation Off): the player sees the black flicker in `ZoraFountain_Child` but has not noticed it in `InsideJabuJabuBelly`. The log shows eight `late pages' fix (600 recordings with one)` summaries during the Zora portion, typically about **75–87 pages per fixed capture**. After the Jabu-Jabu map load, there have been **zero** such summaries so far, alongside 160 `write-back diff (600 submits)` summaries. The late-fix counter logs only after 600 nonempty emits, so this does **not** prove there were zero fixes inside Jabu-Jabu. It does show that this path is much less active there while ordinary write-backs continue. That makes a workload/timing-dependent capture problem a useful lead; it does not identify a particular bad word, shader or scene asset. The next decisive diagnostic remains comparing captured words against the old wait path at the Zora flicker.

Rogasse referenced this issue

2026-11-14 17:02:07 +00:00

fix(hos): report the game's mapped pages as used memory; commit the heap lazily on Windows #324

Rogasse referenced this issue

2026-11-14 17:02:39 +00:00

fix(hos): report the game's mapped pages as used memory; commit the heap lazily on Windows #324

Rogasse commented

2026-11-14 17:13:43 +00:00

Further same-save testing on Nightly 127, Windows 11 / RTX 4050 Laptop (NVIDIA 621.14), in ZoraFountain_Child with frame generation Off:

With RX_VK_LATE_FIX=1 and RX_VK_MIRROR_SPARSE=0, the near-black Link flicker still occurs after setting all three PR #324 old-behaviour switches together (HOS_MEMINFO_OLD=1, HOS_HEAP_COMMIT=1, RX_VK_IMPORT_HIGH=0). Thus those three new behaviours are not required for this reproduction on my PC.
Keeping those settings and late_fix on, one 1–2 minute run with RX_VK_SHADOW_FOLLOWS_SYNC=0 showed no flicker to the player. Restoring shadow-follows-sync to its default made the flicker reappear, particularly when the camera moved. In the longer observation it was intermittent rather than continuous. The movement was not scripted or matched frame for frame, and the shadow-off run was short, so this does not yet prove the shadow copy is the defective operation or that the switch is a safe workaround. The known RX_VK_LATE_FIX=0 workaround remains the one saved for normal launches.
The last shadow-on run's log confirms late pages' fix on and map: ZoraFountain_Child (child). It recorded multiple late pages' fix (600 recordings with one) summaries, typically ~58–73 pages per fixed capture after loading Zora. The complete log is preserved locally for inspection; I am not attaching raw local paths or game data.

This narrows the next source check to the timing/content relationship between shadow_sync_copy and late_fix_emit/the per-word stamps in gpu/vk/vk_memory.cpp. It is a lead, not an identified faulty line. A capture-word comparison in the same scene would distinguish a bad copied word from a timing-only change.

Further same-save testing on Nightly 127, Windows 11 / RTX 4050 Laptop (NVIDIA 621.14), in `ZoraFountain_Child` with frame generation Off: - With `RX_VK_LATE_FIX=1` and `RX_VK_MIRROR_SPARSE=0`, the near-black Link flicker **still occurs** after setting all three [PR #324](https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/324#issuecomment-18717) old-behaviour switches together (`HOS_MEMINFO_OLD=1`, `HOS_HEAP_COMMIT=1`, `RX_VK_IMPORT_HIGH=0`). Thus those three new behaviours are not required for this reproduction on my PC. - Keeping those settings and `late_fix` on, one 1–2 minute run with `RX_VK_SHADOW_FOLLOWS_SYNC=0` showed no flicker to the player. Restoring shadow-follows-sync to its default made the flicker reappear, particularly when the camera moved. In the longer observation it was intermittent rather than continuous. The movement was not scripted or matched frame for frame, and the shadow-off run was short, so this **does not yet prove** the shadow copy is the defective operation or that the switch is a safe workaround. The known `RX_VK_LATE_FIX=0` workaround remains the one saved for normal launches. - The last shadow-on run's log confirms `late pages' fix on` and `map: ZoraFountain_Child (child)`. It recorded multiple `late pages' fix (600 recordings with one)` summaries, typically ~58–73 pages per fixed capture after loading Zora. The complete log is preserved locally for inspection; I am not attaching raw local paths or game data. This narrows the next source check to the timing/content relationship between `shadow_sync_copy` and `late_fix_emit`/the per-word stamps in `gpu/vk/vk_memory.cpp`. It is a lead, **not an identified faulty line**. A capture-word comparison in the same scene would distinguish a bad copied word from a timing-only change.

Rogasse commented

2026-11-14 17:16:33 +00:00

Sanitized full longshot_log.txt from my Nightly 127 ZoraFountain_Child run that visibly flickered, with RX_VK_LATE_FIX=1, frame generation off, sparse mirrors off, and the three PR #324 old-behaviour flags. Local user paths and the save identifier were redacted. This is the longer shadow-follows-sync-on run described in my prior comment; the video and game files remain local.

Sanitized log (1,138,520 bytes).

Sanitized full `longshot_log.txt` from my Nightly 127 `ZoraFountain_Child` run that visibly flickered, with `RX_VK_LATE_FIX=1`, frame generation off, sparse mirrors off, and the three PR #324 old-behaviour flags. Local user paths and the save identifier were redacted. This is the longer shadow-follows-sync-on run described in my prior comment; the video and game files remain local. [Sanitized log (1,138,520 bytes)](https://git.magicbean.dev/attachments/4d9da66e-f148-483c-b515-f1e5c5ac71f3).

zora-flicker-nightly127-redacted.txt

1.1 MiB

No milestone

No project

No assignees

2 participants

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

MagicBeanDev/Longshot#366

No description provided.

Rows
Columns