graphical bugs amd

ghostday05 commented

2026-11-07 16:35:41 +00:00

(Migrated from github.com)

some graphical glitches : Some weird black outlined squares, the sky is pure white and some meshes are missing on trees and hud bug with fog especially

some graphical glitches : Some weird black outlined squares, the sky is pure white and some meshes are missing on trees and hud bug with fog especially <img width="1910" height="1074" alt="Image" src="https://github.com/user-attachments/assets/f29a5d13-3ed3-44d7-945a-08c05c1a4de2" /> <img width="1915" height="1075" alt="Image" src="https://github.com/user-attachments/assets/29d36156-7ba4-4381-8836-fe4233ad94a8" /> <img width="1910" height="1074" alt="Image" src="https://github.com/user-attachments/assets/4b8b6652-90f3-43fe-a3a4-ce2dbd68d9ff" /> <img width="1900" height="1075" alt="Image" src="https://github.com/user-attachments/assets/56a83fcf-e961-432e-84d6-c59222731f5b" />

ghostday05 commented

2026-11-07 16:36:15 +00:00

(Migrated from github.com)

rx 6700, ryzen 7 5700 x3D, 32gb ram, build : less-stutters

adaa1262 commented

2026-11-07 16:49:04 +00:00

(Migrated from github.com)

hookshot_log.txt

Same: Glitched text, Missing textures, weird lighting
Latest build 5800X3D 6700XT 32GB RAM
Also shader freezes and fps drops others also mentioned
EDIT: Game froze when collecting a gem

[hookshot_log.txt](https://github.com/user-attachments/files/34524287/hookshot_log.txt) <img width="2560" height="1440" alt="Image" src="https://github.com/user-attachments/assets/9325d1fe-3fb2-41fc-a51b-42be0e64f2d1" /> <img width="2491" height="1402" alt="Image" src="https://github.com/user-attachments/assets/1f569de0-5568-457d-8eba-da9be50da28b" /> <img width="2560" height="1440" alt="Image" src="https://github.com/user-attachments/assets/457a9884-caec-4c7f-99a7-0f0c52475c39" /> <img width="2560" height="1440" alt="Image" src="https://github.com/user-attachments/assets/dabba850-23bc-4177-bcfd-723a8decef66" /> <img width="2560" height="1440" alt="Image" src="https://github.com/user-attachments/assets/79a09755-30f5-4c88-b012-f20646b5ac69" /> <img width="2560" height="1439" alt="Image" src="https://github.com/user-attachments/assets/c0664ff5-238f-4df0-abcc-a93531732130" /> Same: Glitched text, Missing textures, weird lighting Latest build 5800X3D 6700XT 32GB RAM Also shader freezes and fps drops others also mentioned EDIT: Game froze when collecting a gem

AnthonyYuman commented

2026-11-07 17:41:52 +00:00

(Migrated from github.com)

Which version did you guys use? I have the same graphics card and it freezes on a black screen or stays yellow after the tree speaks.

Which version did you guys use? I have the same graphics card and it freezes on a black screen or stays yellow after the tree speaks. <img width="1394" height="859" alt="Image" src="https://github.com/user-attachments/assets/e0f519d6-cc71-43ea-b31b-34ae43b89a0c" />

hellknight0000 commented

2026-11-07 17:47:49 +00:00

(Migrated from github.com)

https://github.com/user-attachments/assets/eea36140-7905-4942-8985-129f2cefd3f4

Use last version

<img width="2560" height="1440" alt="Image" src="https://github.com/user-attachments/assets/3fa403e6-92b9-4d6f-b4e1-791581a0fa50" /> <img width="2560" height="1392" alt="Image" src="https://github.com/user-attachments/assets/57fe6920-acba-479f-ac23-9794f27987d2" /> <img width="2560" height="1392" alt="Image" src="https://github.com/user-attachments/assets/d58a61ad-9ec7-4f6a-a82c-0efe1642db2e" /> https://github.com/user-attachments/assets/eea36140-7905-4942-8985-129f2cefd3f4 Use last version

AnthonyYuman commented

2026-11-07 18:11:46 +00:00

(Migrated from github.com)

Which version worked for you? I've already tried the three listed here but nothing works, they all give the same error. @hellknight0000

Which version worked for you? I've already tried the three listed here but nothing works, they all give the same error. @hellknight0000 <img width="1194" height="938" alt="Image" src="https://github.com/user-attachments/assets/71805faf-dce1-4d48-a2ef-87b60c80be02" />

hellknight0000 commented

2026-11-07 18:21:22 +00:00

(Migrated from github.com)

Is this https://github.com/NotRexed/Hookshot/releases/tag/less-stutter @AnthonyYuman

NotRexed commented

2026-11-07 19:40:48 +00:00

(Migrated from github.com)

To the question about versions: the newest release is https://github.com/NotRexed/Hookshot/releases/tag/foliage-linux-updater (EvilOOT's foliage and lighting work, AMD text fix PR 228). A yellow or black picture after the Deku Tree speaks on RX 6700 cards is the device-lost crash in #150.

pcornejov commented

2026-11-07 21:31:48 +00:00

(Migrated from github.com)

OoT (0400486024DEC000, v131072) — white screen after the intro cutscene on Radeon 890M (iGPU), build foliage-linux-updater

System

GPU: AMD Radeon 890M (RDNA 3.5 iGPU, Ryzen AI Z2 Extreme, ROG Xbox Ally X), 4 GB dedicated VRAM, 24 GB RAM
Driver: AMD 32.0.32015.5008 (Vulkan 1.4, driver 0x800196); mesh shaders yes, host memory import yes, pipeline libraries yes
Windows 11 Pro 10.0.26300
Hookshot foliage-linux-updater, fresh build from HookshotBuilder (no cached objects), language es-419, no hookshot_env.txt

What happens

The title screen, menu and opening cutscene render fine.
After the cutscene the screen goes black for about 1-2 minutes (slow submits of 10 s, 36 s and 76 s while it builds pipelines), then the whole screen turns white.
The game keeps running underneath: about 1,000 draws per frame (0-3 skipped) at 0.6-3 fps, and the process responds normally. The same black-then-white result also happened on earlier builds.

What the log shows

30 mesh programs fail to translate:
mesh program 066eb7b684ed455f not translated: SPIR-V Instruction is not implemented (LoadGlobalS16) is not implemented
The other 29 ids include 0cec079dbd20db04, 1bd44a91a0079c34, 29a8af226b5afad2, 378d39e41362c852, 5278605289a6deb7, c1c24dcb13cf2e71 and fabe5f270c7e3e37.
13 compute programs report not translated: a network left out on this card and then show as rejected compute. Some fall back to a linear upscale, for example program 6676619641320ba7 does not translate: its dispatch replaced by a linear upscale of ... 1457x820 into ... 2048x1152. The ids are e409733c45530d75, 6676619641320ba7, f10782f9e58b6a7c, 0775215f8da720e6, 14140d4c1dbe754f, 3f6f2b1d671badcd, 4c2c129fe950b041, 5a4d7378db437fed, 8306877b7dc8784b, 8b73eaee02efc5e4, a41c78faa56ccf7d, de3c10dba4ff39d3 and e959b9b1451491c0.
A further rejected compute, 6214a97bec4cc0bf, writes 128x72x96 images, which looks like froxel volumetric fog.
TAAU is stood in for: the TAAU network fd029a317a9a4b94 is stood in for (an upscale of its input).
40 lines of storage buffer NOT bound (reads zeros). These come from stage 7 programs, for example 68642bb810ee3e7c and 6b4dea601798c255, at c5+0x60 addr 0x10001c00a size 0x40000000.

My guess is that the white frame comes from the skipped or zeroed compute passes, such as exposure or tonemap reading zeros, combined with the missing LoadGlobalS16 mesh programs. I can share the full hookshot_log.txt.

**OoT (0400486024DEC000, v131072) — white screen after the intro cutscene on Radeon 890M (iGPU), build `foliage-linux-updater`** **System** - GPU: AMD Radeon 890M (RDNA 3.5 iGPU, Ryzen AI Z2 Extreme, ROG Xbox Ally X), 4 GB dedicated VRAM, 24 GB RAM - Driver: AMD 32.0.32015.5008 (Vulkan 1.4, driver 0x800196); mesh shaders yes, host memory import yes, pipeline libraries yes - Windows 11 Pro 10.0.26300 - Hookshot `foliage-linux-updater`, fresh build from HookshotBuilder (no cached objects), language es-419, no `hookshot_env.txt` **What happens** 1. The title screen, menu and opening cutscene render fine. 2. After the cutscene the screen goes black for about 1-2 minutes (slow submits of 10 s, 36 s and 76 s while it builds pipelines), then the whole screen turns **white**. 3. The game keeps running underneath: about 1,000 draws per frame (0-3 skipped) at **0.6-3 fps**, and the process responds normally. The same black-then-white result also happened on earlier builds. **What the log shows** - **30 mesh programs fail to translate:** `mesh program 066eb7b684ed455f not translated: SPIR-V Instruction is not implemented (LoadGlobalS16) is not implemented` The other 29 ids include 0cec079dbd20db04, 1bd44a91a0079c34, 29a8af226b5afad2, 378d39e41362c852, 5278605289a6deb7, c1c24dcb13cf2e71 and fabe5f270c7e3e37. - **13 compute programs report** `not translated: a network left out on this card` and then show as `rejected compute`. Some fall back to a linear upscale, for example `program 6676619641320ba7 does not translate: its dispatch replaced by a linear upscale of ... 1457x820 into ... 2048x1152`. The ids are e409733c45530d75, 6676619641320ba7, f10782f9e58b6a7c, 0775215f8da720e6, 14140d4c1dbe754f, 3f6f2b1d671badcd, 4c2c129fe950b041, 5a4d7378db437fed, 8306877b7dc8784b, 8b73eaee02efc5e4, a41c78faa56ccf7d, de3c10dba4ff39d3 and e959b9b1451491c0. A further rejected compute, 6214a97bec4cc0bf, writes 128x72x96 images, which looks like froxel volumetric fog. - **TAAU** is stood in for: `the TAAU network fd029a317a9a4b94 is stood in for (an upscale of its input)`. - **40 lines of** `storage buffer NOT bound (reads zeros)`. These come from stage 7 programs, for example 68642bb810ee3e7c and 6b4dea601798c255, at `c5+0x60 addr 0x10001c00a size 0x40000000`. My guess is that the white frame comes from the skipped or zeroed compute passes, such as exposure or tonemap reading zeros, combined with the missing `LoadGlobalS16` mesh programs. I can share the full `hookshot_log.txt`.

SantinoMusi commented

2026-11-07 22:42:25 +00:00

(Migrated from github.com)

Ryzen 7 5700X3D, 9070XT, 32Gb 3200Mhz RAM

Emulator for comparison, graphically almost perfect (few glitches here and there), okay performance as long as no new shader is encountered

Previous version, many textures missing, good, stable 60FPS as long as no new shaders are encountered (https://git.magicbean.dev/MagicBeanDev/Hookshot/src/branch/text-hotfix)

Newest version (https://git.magicbean.dev/MagicBeanDev/Hookshot/src/branch/foliage-linux-updater), very improved graphics, tho the frames look... Intermittent? Like first I see frame 1, then frame 3, then frame 2 and then frame 4, it's weird even tho the FPS counter remains stable. Also flickering textures in the distance and the ground fog

Also it'd be cool to see a resolution selection menu instead of a percentage, like 720p, 1080p, 1440p, 4k

hookshot_log.txt

Ryzen 7 5700X3D, 9070XT, 32Gb 3200Mhz RAM Emulator for comparison, graphically almost perfect (few glitches here and there), okay performance as long as no new shader is encountered <img width="2388" height="1342" alt="Image" src="https://github.com/user-attachments/assets/c117e6af-d87c-4b8f-bc3d-f1a0efc20011" /> Previous version, many textures missing, good, stable 60FPS as long as no new shaders are encountered (https://git.magicbean.dev/MagicBeanDev/Hookshot/src/branch/text-hotfix) <img width="3433" height="1431" alt="Image" src="https://github.com/user-attachments/assets/565c420b-7fb8-4d40-9df6-759bc00e2645" /> Newest version (https://git.magicbean.dev/MagicBeanDev/Hookshot/src/branch/foliage-linux-updater), very improved graphics, tho the frames look... Intermittent? Like first I see frame 1, then frame 3, then frame 2 and then frame 4, it's weird even tho the FPS counter remains stable. Also flickering textures in the distance and the ground fog <img width="3439" height="1435" alt="Image" src="https://github.com/user-attachments/assets/81231094-6670-4992-baa4-5609b48a9c3b" /> Also it'd be cool to see a resolution selection menu instead of a percentage, like 720p, 1080p, 1440p, 4k [hookshot_log.txt](https://github.com/user-attachments/files/34530047/hookshot_log.txt)

pcornejov commented

2026-11-08 02:17:32 +00:00

(Migrated from github.com)

Follow-up on my report above (Radeon 890M iGPU, driver 32.0.32015.5008), now on no-visual-studio, a fresh build with the new builder:

✅ The LoadGlobalS16 fix (#351) works for me. 0 mesh programs fail now; before, 30 failed.

❌ After the intro cutscene the screen still goes black and then white. It runs at ~2-5 fps, and in the gameplay frames about half the draws are skipped. The skip count does not go down after several minutes:

12000 presents, 1.9 fps; a frame: 1180 draws (830 skipped), 123 dispatches, 131 passes
12120 presents, 2.4 fps; a frame: 1254 draws (1264 skipped), 148 dispatches, 120 passes
12480 presents, 2.7 fps; a frame: 991 draws (783 skipped), 116 dispatches, 117 passes
12720 presents, 2.3 fps; a frame: 995 draws (1063 skipped), 116 dispatches, 111 passes

31 rejected compute programs. 14 of them are the movie networks left out on this card, which is expected. These are the other 17:
- grid 1x1x1: 09cc9a93a2d7dd81, 0ac7618f173d27b7, 447ebb51fd89e8df, 49fef5f09ec314cf, 54a5c6d465f016bc, 5b7409961f998124, 743ec93a73bd13cb, 8851426c01be325a, 8e9b09db8924dd98, aba5e6af8d4683eb, aea8f1473a9304cc
- grid 2x1x1: 7b73725ff8212e25, 966d67b028787849, b011a5bafe028cd4
- grid 2x1x64: 49867a3053c1947b, fe2b59bfa3cb5e8f
- grid 16x8x1: 3a75eba955a45316
There are still 40 storage buffer NOT bound (reads zeros) lines. Some compute pipelines show SLOW compile of 1-2.3 s, for example 5b0baadba86c1489 and 72aa5bd592cd0135.

The 1x1x1 rejected computes look like small reduction or exposure passes. If they never run, that could explain the white frame.

Follow-up on my report above (Radeon 890M iGPU, driver 32.0.32015.5008), now on **`no-visual-studio`**, a fresh build with the new builder: - ✅ The `LoadGlobalS16` fix (#351) works for me. **0** mesh programs fail now; before, 30 failed. - ❌ After the intro cutscene the screen still goes **black and then white**. It runs at **~2-5 fps**, and in the gameplay frames about **half the draws are skipped**. The skip count does not go down after several minutes: ``` 12000 presents, 1.9 fps; a frame: 1180 draws (830 skipped), 123 dispatches, 131 passes 12120 presents, 2.4 fps; a frame: 1254 draws (1264 skipped), 148 dispatches, 120 passes 12480 presents, 2.7 fps; a frame: 991 draws (783 skipped), 116 dispatches, 117 passes 12720 presents, 2.3 fps; a frame: 995 draws (1063 skipped), 116 dispatches, 111 passes ``` - **31 `rejected compute` programs.** 14 of them are the movie networks left out on this card, which is expected. These are the other 17: - grid 1x1x1: 09cc9a93a2d7dd81, 0ac7618f173d27b7, 447ebb51fd89e8df, 49fef5f09ec314cf, 54a5c6d465f016bc, 5b7409961f998124, 743ec93a73bd13cb, 8851426c01be325a, 8e9b09db8924dd98, aba5e6af8d4683eb, aea8f1473a9304cc - grid 2x1x1: 7b73725ff8212e25, 966d67b028787849, b011a5bafe028cd4 - grid 2x1x64: 49867a3053c1947b, fe2b59bfa3cb5e8f - grid 16x8x1: 3a75eba955a45316 - There are still 40 `storage buffer NOT bound (reads zeros)` lines. Some compute pipelines show `SLOW compile` of 1-2.3 s, for example 5b0baadba86c1489 and 72aa5bd592cd0135. The 1x1x1 rejected computes look like small reduction or exposure passes. If they never run, that could explain the white frame.

pcornejov commented

2026-11-08 14:32:37 +00:00

(Migrated from github.com)

Update on the same machine (Radeon 890M iGPU, 4 GB dedicated VRAM, driver 32.0.32015.5008, Windows 11 26H2), tested on low-vram and honest-eta:

Big improvement: the black/white screen after the intro is gone. Kokiri Forest and the Deku Tree now render with about 1000-1100 draws per frame and 0 skipped (before, 800-1260 were skipped each frame). It runs at 38-41 fps, and only 10-13 rejected compute remain, all of them full-frame (1920x1080) programs that look like the movie networks. The 1x1x1 ones from my last report are gone.

Still broken: all 4 runs end in a GPU hang, then device lost shortly after gameplay starts in Kokiri:

Run	Build	What happened
1	low-vram	Hung on the first gameplay frames: `1 sync waits (11037 ms)`, then `submit 73076 failed (-4)`
2	low-vram (warm cache, `RX_VK_CHECKPOINTS=1`)	Played ~3 min in Kokiri at 38-41 fps, then the GPU froze for 22 s: `submit 113703 failed (-4)`
3	honest-eta	Hung right on entering Kokiri: `1 sync waits (11979 ms)`, then `submit 69786 failed (-4)`
4	honest-eta, after a reboot	White screen at 22-26 fps (1100 draws, 17 skipped), then `0.0 fps over the last 38.1 s` and `submit 88358 failed (-4)`

The retired-pool fix in honest-eta does not change it. With honest-eta the log shows 29 retired still imported (800 MB), 0 retired given back, so pools are kept, and the device is still lost the same way.
RX_VK_CHECKPOINTS=1 printed no markers on this AMD driver. Only device lost: reading the fault info (RX_VK_CHECKPOINTS=1 for markers) appears, with no fault info after it.
Each time, about 40 storage buffer NOT bound (reads zeros) lines appear before the hang (programs 68642bb810ee3e7c, 6b4dea601798c255, 51183fe750d5d0c8, 739acd678ae08655 and d905f52b7517c681, stage 7, with sizes like 0x3fd55555 and 0x449f8155).
After a device lost, the next start without a reboot crashed at once: rx: the Vulkan back end could not start on the window: WebGPU, followed by CRASH: exception 0xe06d7363 in wgpu_native.dll (WER: BEX64, c0000409 at wgpu_native.dll+0x393c4d). A reboot fixed that.

[hos sead::MainThread] nvn: GPU waits in 5 s: 1 sync waits (11979 ms), 0 polls timed out (0 ms), 0 queue finishes (0 ms)
[hos Presentation Thread] vk: submit 69786 failed (-4)
[hos Presentation Thread] vk: device lost: reading the fault info (RX_VK_CHECKPOINTS=1 for markers)
[hos Presentation Thread] vi: 9969 frames presented (0.7 fps over the last 17.0 s, interval 1)

I'm happy to test a build with a specific workaround (for example, turning off the foliage culling compute or the mesh draws).

Update on the same machine (Radeon 890M iGPU, 4 GB dedicated VRAM, driver 32.0.32015.5008, Windows 11 26H2), tested on **`low-vram`** and **`honest-eta`**: **Big improvement:** the black/white screen after the intro is gone. Kokiri Forest and the Deku Tree now render with about 1000-1100 draws per frame and **0 skipped** (before, 800-1260 were skipped each frame). It runs at **38-41 fps**, and only 10-13 `rejected compute` remain, all of them full-frame (1920x1080) programs that look like the movie networks. The 1x1x1 ones from my last report are gone. **Still broken:** all 4 runs end in a **GPU hang, then device lost** shortly after gameplay starts in Kokiri: | Run | Build | What happened | | --- | --- | --- | | 1 | low-vram | Hung on the first gameplay frames: `1 sync waits (11037 ms)`, then `submit 73076 failed (-4)` | | 2 | low-vram (warm cache, `RX_VK_CHECKPOINTS=1`) | Played ~3 min in Kokiri at 38-41 fps, then the GPU froze for 22 s: `submit 113703 failed (-4)` | | 3 | honest-eta | Hung right on entering Kokiri: `1 sync waits (11979 ms)`, then `submit 69786 failed (-4)` | | 4 | honest-eta, after a reboot | White screen at 22-26 fps (1100 draws, 17 skipped), then `0.0 fps over the last 38.1 s` and `submit 88358 failed (-4)` | - **The retired-pool fix in honest-eta does not change it.** With honest-eta the log shows `29 retired still imported (800 MB), 0 retired given back`, so pools are kept, and the device is still lost the same way. - **`RX_VK_CHECKPOINTS=1` printed no markers on this AMD driver.** Only `device lost: reading the fault info (RX_VK_CHECKPOINTS=1 for markers)` appears, with no fault info after it. - Each time, about 40 `storage buffer NOT bound (reads zeros)` lines appear before the hang (programs 68642bb810ee3e7c, 6b4dea601798c255, 51183fe750d5d0c8, 739acd678ae08655 and d905f52b7517c681, stage 7, with sizes like 0x3fd55555 and 0x449f8155). - After a device lost, the next start without a reboot crashed at once: `rx: the Vulkan back end could not start on the window: WebGPU`, followed by `CRASH: exception 0xe06d7363` in `wgpu_native.dll` (WER: BEX64, c0000409 at wgpu_native.dll+0x393c4d). A reboot fixed that. ``` [hos sead::MainThread] nvn: GPU waits in 5 s: 1 sync waits (11979 ms), 0 polls timed out (0 ms), 0 queue finishes (0 ms) [hos Presentation Thread] vk: submit 69786 failed (-4) [hos Presentation Thread] vk: device lost: reading the fault info (RX_VK_CHECKPOINTS=1 for markers) [hos Presentation Thread] vi: 9969 frames presented (0.7 fps over the last 17.0 s, interval 1) ``` I'm happy to test a build with a specific workaround (for example, turning off the foliage culling compute or the mesh draws).

pcornejov commented

2026-11-08 18:06:26 +00:00

(Migrated from github.com)

v2.0.2 fixes the device lost on my Radeon 890M. Thank you!

Same machine as before (Radeon 890M iGPU, 4 GB dedicated VRAM, driver 32.0.32015.5008, Windows 11). On v2.0.2 I played about 10 minutes in Kokiri Forest at 30-45 fps, with 500-2000 draws per frame (0 skipped) and no device lost. On v2.0.0 and every earlier build, all 5 runs ended in a device lost on entering Kokiri or a few minutes later. The bounded compare-and-swap loops (#548) look like the fix.

Two problems are left:

1. Link is drawn as a flat white silhouette. The rest of the scene looks right (Deku Tree, Navi, foliage, lighting), but Link has no shading at all: he is pure white with a blue-white glow around him. 13 rejected compute remain. Most are the full-frame movie networks, but three run on a small 171x96 image, which looks like exposure or bloom work:

0775215f8da720e6 grid 48x27x1
72ba090c70fe2614 grid 43x24x1
f10782f9e58b6a7c grid 43x24x1

There are also the same 40 storage buffer NOT bound (reads zeros) lines as before (programs 68642bb810ee3e7c, 6b4dea601798c255, 51183fe750d5d0c8, 739acd678ae08655 and d905f52b7517c681, stage 7).

2. Two long freezes, each about 2 minutes, that recover by themselves. No device lost. The time is spent in draw on the host while compute pipelines are still compiling:

vk: dispatch of 685adb41edc19de7 left out: its compute pipeline is still compiling
vk: slow submit: 118039.6 ms: draw 117753.0 descriptors 161.2 texture lookups 247.8 uploads 75.6 pipelines 6.8 programs 699.1 commands 3.3 dispatch 199.6
vi: 15146 frames presented (0.1 fps over the last 136.3 s, interval 1)
...
vk: slow submit: 15536.9 ms: draw 15536.9 descriptors 1.1
vk: slow submit: 12216.1 ms: draw 12212.4 descriptors 1.3 commands 2.3 queue submit 1.1
vi: 21637 frames presented (0.8 fps over the last 136.8 s, interval 1)

Some compute pipelines also compile slowly on AMD: SLOW compile, 3489 ms (cc62764061c44f6a) and 3877 ms (76b54f5bb0d7ce97).

I can test again or send the full log if that helps.

**v2.0.2 fixes the device lost on my Radeon 890M.** Thank you! Same machine as before (Radeon 890M iGPU, 4 GB dedicated VRAM, driver 32.0.32015.5008, Windows 11). On v2.0.2 I played **about 10 minutes in Kokiri Forest** at **30-45 fps**, with 500-2000 draws per frame (0 skipped) and **no device lost**. On v2.0.0 and every earlier build, all 5 runs ended in a device lost on entering Kokiri or a few minutes later. The bounded compare-and-swap loops (#548) look like the fix. Two problems are left: **1. Link is drawn as a flat white silhouette.** The rest of the scene looks right (Deku Tree, Navi, foliage, lighting), but Link has no shading at all: he is pure white with a blue-white glow around him. 13 `rejected compute` remain. Most are the full-frame movie networks, but three run on a small 171x96 image, which looks like exposure or bloom work: ``` 0775215f8da720e6 grid 48x27x1 72ba090c70fe2614 grid 43x24x1 f10782f9e58b6a7c grid 43x24x1 ``` There are also the same 40 `storage buffer NOT bound (reads zeros)` lines as before (programs 68642bb810ee3e7c, 6b4dea601798c255, 51183fe750d5d0c8, 739acd678ae08655 and d905f52b7517c681, stage 7). **2. Two long freezes, each about 2 minutes, that recover by themselves.** No device lost. The time is spent in `draw` on the host while compute pipelines are still compiling: ``` vk: dispatch of 685adb41edc19de7 left out: its compute pipeline is still compiling vk: slow submit: 118039.6 ms: draw 117753.0 descriptors 161.2 texture lookups 247.8 uploads 75.6 pipelines 6.8 programs 699.1 commands 3.3 dispatch 199.6 vi: 15146 frames presented (0.1 fps over the last 136.3 s, interval 1) ... vk: slow submit: 15536.9 ms: draw 15536.9 descriptors 1.1 vk: slow submit: 12216.1 ms: draw 12212.4 descriptors 1.3 commands 2.3 queue submit 1.1 vi: 21637 frames presented (0.8 fps over the last 136.8 s, interval 1) ``` Some compute pipelines also compile slowly on AMD: `SLOW compile, 3489 ms` (cc62764061c44f6a) and `3877 ms` (76b54f5bb0d7ce97). I can test again or send the full log if that helps.

NotRexed commented

2026-11-09 05:45:38 +00:00

(Migrated from github.com)

Tracked in #1006, follow there. That issue collects every report of AMD-specific rendering problems, and this one is linked from it with its GPU, version and log details. If it still happens on v2.0.6 (https://github.com/NotRexed/Hookshot/releases/tag/v2.0.6; Update next to Play in the launcher), please add a comment there with your hookshot_log.txt.

Tracked in #1006, follow there. That issue collects every report of AMD-specific rendering problems, and this one is linked from it with its GPU, version and log details. If it still happens on v2.0.6 (https://github.com/NotRexed/Hookshot/releases/tag/v2.0.6; Update next to Play in the launcher), please add a comment there with your `hookshot_log.txt`.

Rows
Columns

graphical bugs amd #224