Skip to content

Beta 6rc For Windows ARM crashes during compile #779

Open

Description

@talynone

Using LAS.Recomp-test-beta_6rc-windows-arm64 on an Asus Zenbook A16 Snapdragon X2 Extreme running Windows 11 ARM Pro 26H1.

During the game compile step it fails, error log is as a follows:

Note: I do notice .\lasgen\lasgen.exe is compiled as an x64 app and not Windows ARM native.

[session] lasrecomp-backup-01dd73dcf71e597a-21380.log
[info] LAS Recomp test-arm-beta_6rc, Windows 11 (10.0.28000, arm64), Snapdragon(R) X2 Elite Extreme - X2E94100 - Qualcomm Oryon(TM) CPU 18 cores, 47.6 GB RAM, zip
[paths] data: C:\Decomps\LAS Recomp
[settings] loaded larecomp.ini
[vk] gpu 0: Qualcomm(R) Adreno(TM) X2-90 GPU (api 1.4.351)
[vk] gpu 1: Microsoft Direct3D12 (Qualcomm(R) Adreno(TM) X2-90 GPU) (api 1.2.359)
[vk] gpu 2: Microsoft Direct3D12 (Microsoft Basic Render Driver) (api 1.2.359)
[vk] using Qualcomm(R) Adreno(TM) X2-90 GPU
[settings] integrated GPU: upscaler bilinear (default), resolution match (default), anti-aliasing off
[vk] lean usage on, SMAA stencil on, per-frame target clears off (integrated GPU)
[vk] driver Qualcomm Technologies Inc. Adreno Vulkan Driver Driver Build: 3777290fe9, Ifc8f55f84d, 1788759568
Date: 09/07/2026
Compiler Version: E17.52.12.00
Driver Branch: auto-glymur-llvm_1
; graphics pipeline library: yes (fast linking yes, independent interpolation decoration yes); vertex input dynamic state: yes; extended dynamic state 3: no extension
[vk] ASTC as the driver lists it (default for this GPU)
[vk] pipeline binaries: feature yes, internal cache 0 (control 0, prefers it 0, precompiled 0), compressed 0 -> pipeline cache (not an AMD driver)
[vk] vertex input: dynamic (VK_EXT_vertex_input_dynamic_state)
[vk] memory (start) heap 0 device-local: 24379 MB, budget 24379 MB, in use 34 MB, ours 0 MB in 0 allocations
[vk] memory (start) heap 1 device-local: 4095 MB, budget 4095 MB, in use 0 MB, ours 0 MB in 0 allocations
[vk] memory (start) heap 2 device-local: 15 MB, budget 15 MB, in use 0 MB, ours 0 MB in 0 allocations
[vk] pipeline cache: loaded 99700 bytes
[vk] swapchain 1280x720, 3 images, mailbox
[vk] GPU timer on (dynamic resolution goes by GPU time)
[vk] guest heap bound to shaders in 16 MB chunks, imported as the game binds them (storage buffers up to 2047 MB here)
[vk] shader branch stacks as arrays (default for this GPU)
[vk] structured shaders on (default for this GPU)
[vk] constant-buffer push constants on (default for this GPU)
[vk] fragment quad shuffles as subgroup shuffles (the driver has them in fragment shaders; subgroup stages 0xf0, operations 0x6ff)
[vk] recolour variants on (default for this GPU)
[vk] fast min / max off (default here)
[vk] recorder thread on
[pcore] hybrid CPU: 12 performance + 6 efficiency logical processors; game/render/vsync threads prefer the 12 performance ones, power throttling off
[pcore] render thread: power throttling off (performance-core CPU sets)
[vk] transient ring: memory type 6 (device-local host-visible cached)
[launcher] game code out of date: updating it
[gamebuild] generator C:\Decomps\LAS Recomp\lasgen\lasgen.exe, compiler C:\Decomps\LAS Recomp\zig\zig.exe cc, 18 jobs
[gamebuild] generator started with Control Flow Guard off
[gamebuild] generator (sdk) failed: exit status 0xC0000409 (a security check stopped it)
[gamebuild] reused 0 objects, compiled 1 objects
[gamebuild] failed in 5s
[launcher] game code: the game code couldn't be prepared
[menu_sfx] SE_SY_CURSOL_MOVE: ADPCM 48000 Hz 1 ch, 821 ms (program 0, key 60, gain 0.107)
[menu_sfx] SE_SY_MENU_LR_MOVE: ADPCM 48000 Hz 1 ch, 381 ms (program 45, key 60, gain 0.293)
[menu_sfx] SE_SY_DECIDE: ADPCM 48000 Hz 1 ch, 919 ms (program 1, key 60, gain 0.239)
[menu_sfx] SE_SY_BEEP: ADPCM 48000 Hz 1 ch, 1160 ms (program 3, key 60, gain 0.293)
[menu_sfx] SE_SY_CANCEL: ADPCM 48000 Hz 1 ch, 206 ms (program 2, key 60, gain 0.220)
[menu_sfx] SE_SY_MENU_OPEN: ADPCM 48000 Hz 1 ch, 1167 ms (program 71, key 60, gain 0.098)
[menu_sfx] SE_SY_MENU_CLOSE: ADPCM 48000 Hz 1 ch, 990 ms (program 70, key 60, gain 0.049)
[menu_sfx] SE_MAP_MATFACE_RUMBLE: ADPCM 48000 Hz 1 ch, 4320 ms (program 96, key 60, gain 0.586)
[menu_sfx] SE_MAP_BOMB_EXPLO: ADPCM 48000 Hz 1 ch, 2851 ms (program 39, key 60, gain 0.342)
[menu_sfx] SE_ENE_BUZZBLOB_SPARK: ADPCM 48000 Hz 1 ch, 614 ms (program 5, key 60, gain 0.059)
[menu_sfx] menu sounds loaded in 11.4 ms
[launcher] game files: version 1.0.0
[gamebuild] generator C:\Decomps\LAS Recomp\lasgen\lasgen.exe, compiler C:\Decomps\LAS Recomp\zig\zig.exe cc, 18 jobs
[gamebuild] generator (sdk) failed: exit status 0xC0000409 (a security check stopped it)
[gamebuild] reused 0 objects, compiled 1 objects
[gamebuild] failed in 4s
[launcher] game code: the game code couldn't be prepared

Activity

  1. peiikai commented on Nov 14, 2026

    @peiikai
    Owner

    Thanks for the report and the log. The game code generator (lasgen.exe) is stopped by a Windows security check (0xC0000409) right after it starts, even with Control Flow Guard turned off for it (the fix from #337). It's an x64 program on purpose: one of the libraries it uses has no Windows on ARM build, so it runs under Windows' x64 emulation.

    Could you attach the generator's own log? It's gen_sdk.log in the .gamebuild folder inside your extracted game files folder (the one with exefs in it). The folder is hidden, so you may need to turn on hidden items in File Explorer. Please also attach lasrecomp.log from C:\Decomps\LAS Recomp.

  2. talynone commented on Nov 14, 2026

    @talynone
    Author

    I've attached lasrecomp.log. I also found gen_sdk.log, but unfortunately it's a 0 byte file.

    lasrecomp.log

    image
  3. github-actions commented on Nov 14, 2026

    @github-actions

    This is an automatic summary of the LAS Recomp log files attached to this issue. It only reads the logs, so a person will still look at the details.

    For a new report, please attach lasrecomp.log from LAS Recomp's data folder (it always holds the latest run; lasrecomp-backup-*.log there are the earlier runs).

    Newest log: lasrecomp.log in comment from 2026-11-16

    • Build: test-779-softshadow (zip)
    • System: Windows 11 (10.0.28000, arm64), Snapdragon(R) X2 Elite Extreme - X2E94100 - Qualcomm Oryon(TM) CPU 18 cores, 47.6 GB RAM
    • GPU: Qualcomm(R) Adreno(TM) X2-90 GPU (integrated), Vulkan 1.4.351, driver Qualcomm Technologies Inc. Adreno Vulkan Driver Driver Bu... (3 GPUs found)
    • Backend: Vulkan
    • Graphics settings: upscaler bilinear (default), resolution match (default), anti-aliasing off; window 1280x720, mailbox
    • How the run ended: the log stops without an exit line (closed from outside, frozen, or still running when copied)
    • Frame rate: 25 samples over 48 s: median 60.0 fps, lowest 49.5 fps, longest frame 186 ms; slowest areas: Field 60.0 fps
    • Notable lines:
      • developer overrides (LARECOMP_*) in use: [vk] shadow lookups compared in the shader (LARECOMP_VK_SOFT_SHADOW)

    Earlier log: lasrecomp.log in comment from 2026-11-15

    • Build: test-arm-779-probe (zip)
    • System: Windows 11 (10.0.28000, arm64), Snapdragon(R) X2 Elite Extreme - X2E94100 - Qualcomm Oryon(TM) CPU 18 cores, 47.6 GB RAM
    • GPU: Qualcomm(R) Adreno(TM) X2-90 GPU (integrated), Vulkan 1.4.351, driver Qualcomm Technologies Inc. Adreno Vulkan Driver Driver Bu... (3 GPUs found)
    • Backend: Vulkan
    • Graphics settings: upscaler bilinear (default), resolution match (default), anti-aliasing off; window 1280x720, mailbox
    • How the run ended: clean exit (window closed)
    • Frame rate: 47 samples over 92 s: median 60.0 fps, lowest 49.0 fps, longest frame 265 ms; slowest areas: Field 60.0 fps
    • Slow frames: 2 [slow] lines; worst: [slow] frame 980 50 ms in Field: 2 full texture hashes, 2 submissions, gpu wait 0 ms, render cpu 0 ms, game cpu 31 ms, queue wait 0 ms, v...

    Earlier log: lasrecomp.log in comment from 2026-11-15

    • Build: test-arm-779-d24 (zip)
    • System: Windows 11 (10.0.28000, arm64), Snapdragon(R) X2 Elite Extreme - X2E94100 - Qualcomm Oryon(TM) CPU 18 cores, 47.6 GB RAM
    • GPU: Qualcomm(R) Adreno(TM) X2-90 GPU (integrated), Vulkan 1.4.351, driver Qualcomm Technologies Inc. Adreno Vulkan Driver Driver Bu... (3 GPUs found)
    • Backend: Vulkan
    • Graphics settings: upscaler bilinear (default), resolution match (default), anti-aliasing off; window 1280x720, mailbox
    • How the run ended: the log stops without an exit line (closed from outside, frozen, or still running when copied)
    • Frame rate: 21 samples over 40 s: median 60.0 fps, lowest 54.5 fps, longest frame 171 ms; slowest areas: Field 60.0 fps

    (10 earlier logs on this issue not summarised.)

  4. peiikai commented on Nov 14, 2026

    @peiikai
    Owner

    Thanks. An empty gen_sdk.log means the generator was stopped before it printed anything, so while it was still starting up. In your screenshot gen_main.log is 1 KB, though: could you attach that one too? It may say how it ended.

  5. talynone commented on Nov 14, 2026

    @talynone
    Author

    Here you go:

    gen_main.log

    Probably won't help, the entire contents are:

    main: game version 1.0.0 (build AE16F71E002AF8CB), address table v1_0_0

  6. peiikai commented on Nov 14, 2026

    @peiikai
    Owner

    It does help: the generator for the main game code got far enough to read your game version, but the one for the SDK part was stopped before it printed anything. So it's something the SDK part does under Windows' x64 emulation on your setup, not the generator failing to start in general. We'll look into it and let you know when there's something to try.

  7. added
    status: investigatingWe're working on it: the ball is in our court
    and removed on Nov 14, 2026
  8. peiikai commented on Nov 14, 2026

    @peiikai
    Owner

    Here's a test build with a likely fix: https://github.com/peiikai/LASRecomp/releases/tag/test-arm-779-shadow (use LAS.Recomp-test-arm-779-shadow-windows-arm64.zip). The generator now starts with Windows' shadow stack protection turned off as well as Control Flow Guard. It also logs which of those protections were really on, so if it still fails, the log will tell us more.

    Please try the game code build with it and attach lasrecomp.log and .gamebuild\gen_sdk.log, whether it works or not.

  9. added
    status: test build outA test build or branch is out; waiting for a tester's result
    and removed
    status: investigatingWe're working on it: the ball is in our court
    on Nov 14, 2026
  10. talynone commented on Nov 14, 2026

    @talynone
    Author

    Thanks, still didn't work, I've attached the requested logs.

    lasrecomp.log

    gen_sdk.log

  11. peiikai commented on Nov 14, 2026

    @peiikai
    Owner

    Thanks for testing. That narrows it down: the generator is now stopped right when it starts the CPU emulator it uses to read the SDK's jump tables, so turning off shadow stacks wasn't enough. We're looking into what that emulator does under Windows' x64 emulation, and will post another build when there's something to try.

  12. 28 remaining items

  13. peiikai commented on Nov 15, 2026

    @peiikai
    Owner

    Thanks, that rules out the depth format too. So the shadow test fails on your GPU whatever format or filter it uses. Next we need to see what's actually in the shadow map.

    This build can log it: https://github.com/peiikai/LASRecomp/releases/tag/test-arm-779-probe

    Please run it once in a new folder with:

    set LARECOMP_VK_SHADOW_PROBE=3000
    

    Then load your save, walk outside Link's house and stay there for about a minute. Then close the game and attach lasrecomp.log. It will have a few lines starting with [vk] shadow probe. No screenshot needed this time.

  14. talynone commented on Nov 15, 2026

    @talynone
    Author

    Thanks, that rules out the depth format too. So the shadow test fails on your GPU whatever format or filter it uses. Next we need to see what's actually in the shadow map.

    This build can log it: https://github.com/peiikai/LASRecomp/releases/tag/test-arm-779-probe

    Please run it once in a new folder with:

    set LARECOMP_VK_SHADOW_PROBE=3000
    

    Then load your save, walk outside Link's house and stay there for about a minute. Then close the game and attach lasrecomp.log. It will have a few lines starting with [vk] shadow probe. No screenshot needed this time.

    Sure, here ya go:
    lasrecomp.log

  15. peiikai commented on Nov 16, 2026

    @peiikai
    Owner

    Thanks, that log helps a lot. Your shadow map holds the same kind of depth values as on an NVIDIA card (we ran the same probe there), so the shadow map itself is drawn correctly. What goes wrong is the shadow test that reads it: your GPU's driver says "in shadow" for every pixel.

    This build can do that test in our own shader code instead of leaving it to the driver: https://github.com/peiikai/LASRecomp/releases/tag/test-779-softshadow

    Please run it once in a new folder with:

    set LARECOMP_VK_SOFT_SHADOW=1
    

    and send a screenshot outside Link's house plus lasrecomp.log. If the sun and shadows come back, we know it's the driver's shadow test and can work around it.

  16. talynone commented on Nov 16, 2026

    @talynone
    Author

    Thanks, that log helps a lot. Your shadow map holds the same kind of depth values as on an NVIDIA card (we ran the same probe there), so the shadow map itself is drawn correctly. What goes wrong is the shadow test that reads it: your GPU's driver says "in shadow" for every pixel.

    This build can do that test in our own shader code instead of leaving it to the driver: https://github.com/peiikai/LASRecomp/releases/tag/test-779-softshadow

    Please run it once in a new folder with:

    set LARECOMP_VK_SOFT_SHADOW=1
    

    and send a screenshot outside Link's house plus lasrecomp.log. If the sun and shadows come back, we know it's the driver's shadow test and can work around it.

    Here ya go:

    lasrecomp.log

    image
  17. peiikai commented on Nov 16, 2026

    @peiikai
    Owner

    Thanks. So it isn't the driver's shadow test either: with our own test it's just as dim. That means one of the numbers going into the test is off on your GPU. This build has a few check modes that light the scene only where a certain thing is true: https://github.com/peiikai/LASRecomp/releases/tag/test-779-shadowtest

    Four runs please, each in a new Command Prompt with only the line listed, and a screenshot outside Link's house each time:

    1. set LARECOMP_VK_SOFT_SHADOW=2
    2. set LARECOMP_VK_SOFT_SHADOW=3
    3. set LARECOMP_VK_SOFT_SHADOW=4
    4. set LARECOMP_VK_SOFT_SHADOW=5

    Each one will look either brightly lit, dim like now, or a mix, and that's the answer we need. No logs needed this time.

  18. talynone commented on Nov 16, 2026

    @talynone
    Author

    Thanks. So it isn't the driver's shadow test either: with our own test it's just as dim. That means one of the numbers going into the test is off on your GPU. This build has a few check modes that light the scene only where a certain thing is true: https://github.com/peiikai/LASRecomp/releases/tag/test-779-shadowtest

    Four runs please, each in a new Command Prompt with only the line listed, and a screenshot outside Link's house each time:

    1. set LARECOMP_VK_SOFT_SHADOW=2
    image
    1. set LARECOMP_VK_SOFT_SHADOW=3
    image
    1. set LARECOMP_VK_SOFT_SHADOW=4
    image
    1. set LARECOMP_VK_SOFT_SHADOW=5
    image

    Each one will look either brightly lit, dim like now, or a mix, and that's the answer we need. No logs needed this time.

    This one seems to have reverted the missing textures/text fixes, even without any environment variable values the missing textures/text problem is back.

  19. peiikai commented on Nov 16, 2026

    @peiikai
    Owner

    Sorry, that one was my mistake: test-779-shadowtest was built without any of the fixes, which is why the text and textures went missing again. Here's a corrected build: https://github.com/peiikai/LASRecomp/releases/tag/test-779-shadowtest2

    Could you do the same four runs with this one? Each in a new Command Prompt with only the line listed, and a screenshot outside Link's house each time:

    1. set LARECOMP_VK_SOFT_SHADOW=2
    2. set LARECOMP_VK_SOFT_SHADOW=3
    3. set LARECOMP_VK_SOFT_SHADOW=4
    4. set LARECOMP_VK_SOFT_SHADOW=5

    Without any setting it should draw as test-779-softshadow did. Each check run will look brightly lit, dim, or a mix.

  20. talynone commented on Nov 16, 2026

    @talynone
    Author

    Sorry, that one was my mistake: test-779-shadowtest was built without any of the fixes, which is why the text and textures went missing again. Here's a corrected build: https://github.com/peiikai/LASRecomp/releases/tag/test-779-shadowtest2

    Could you do the same four runs with this one? Each in a new Command Prompt with only the line listed, and a screenshot outside Link's house each time:

    1. set LARECOMP_VK_SOFT_SHADOW=2
    image
    1. set LARECOMP_VK_SOFT_SHADOW=3
    image
    1. set LARECOMP_VK_SOFT_SHADOW=4
    image
    1. set LARECOMP_VK_SOFT_SHADOW=5
    image

    Without any setting it should draw as test-779-softshadow did. Each check run will look brightly lit, dim, or a mix.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    log: new shadersAttached log has shader variants the shipped list lacksstatus: test build outA test build or branch is out; waiting for a tester's result

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions