RX 6700 XT: 10-11 shader programs fail to translate from the title screen onward (v2.0.5) #1017

Closed

opened 2026-11-09 06:04:53 +00:00 by kafasam · 1 comment

kafasam commented

2026-11-09 06:04:53 +00:00

(Migrated from github.com)

RX 6700 XT: 10–11 shader programs fail to translate from the title screen onward (v2.0.5)

System: Radeon RX 6700 XT 12 GB (driver 32.0.21045.16010), i7-12700K, 32 GB, Windows 11, Hookshot v2.0.5, 4K fullscreen, Lanczos upscaling, frame generation 2, language English. Built by HookshotBuilder from our own v1.1.1 .nup (23.9 GB, verified by the builder).

What I see: the per-frame stats line reports programs (10 failed) from around the title/attract screen (present #240) and 11 failed once in gameplay (Hyrule Field, present ~163k). Ten failures exist before any gameplay, so they are not scene-specific streaming.

I can't find per-program failure names in hookshot_log.txt — only the counter in the frame-stat lines. v2.0.2 fixed seven compute programs "left out on AMD" (#597, #598), and this count (10 at boot) predates play, so these may be more of the same family. Visually I can't pinpoint what's missing; the game generally looks right.

Relevant frame stat (title, 60 fps, everything healthy otherwise):

vk: 600 presents, 60.0 fps; a frame: 2 draws (12 skipped), 2 dispatches, 3 passes, 14 barriers, 6 submits, ...; 2831 pipelines, 0 programs (10 failed), 9 textures (76 MB)

and in Hyrule Field:

vk: 162960 presents, 26.4 fps; ...; 3903 pipelines, 114 programs (11 failed), 1174 textures (2352 MB)

Happy to attach the full hookshot_log.txt or run any instrumented build — is there a flag that names the failing programs (something like a shader dump for the recompiler, the way the frame stats name pipelines)?

Separately, thank you for #823/v2.0.6 — the v2.0.5 terrain re-encode made Hyrule Field noticeably heavier on this exact GPU.

**RX 6700 XT: 10–11 shader programs fail to translate from the title screen onward (v2.0.5)** **System:** Radeon RX 6700 XT 12 GB (driver 32.0.21045.16010), i7-12700K, 32 GB, Windows 11, Hookshot v2.0.5, 4K fullscreen, Lanczos upscaling, frame generation 2, language English. Built by HookshotBuilder from our own v1.1.1 `.nup` (23.9 GB, verified by the builder). **What I see:** the per-frame stats line reports `programs (10 failed)` from around the title/attract screen (present #240) and `11 failed` once in gameplay (Hyrule Field, present ~163k). Ten failures exist before any gameplay, so they are not scene-specific streaming. I can't find per-program failure names in `hookshot_log.txt` — only the counter in the frame-stat lines. v2.0.2 fixed seven compute programs "left out on AMD" (#597, #598), and this count (10 at boot) predates play, so these may be more of the same family. Visually I can't pinpoint what's missing; the game generally looks right. Relevant frame stat (title, 60 fps, everything healthy otherwise): ``` vk: 600 presents, 60.0 fps; a frame: 2 draws (12 skipped), 2 dispatches, 3 passes, 14 barriers, 6 submits, ...; 2831 pipelines, 0 programs (10 failed), 9 textures (76 MB) ``` and in Hyrule Field: ``` vk: 162960 presents, 26.4 fps; ...; 3903 pipelines, 114 programs (11 failed), 1174 textures (2352 MB) ``` Happy to attach the full `hookshot_log.txt` or run any instrumented build — is there a flag that names the failing programs (something like a shader dump for the recompiler, the way the frame stats name pipelines)? Separately, thank you for #823/v2.0.6 — the v2.0.5 terrain re-encode made Hyrule Field noticeably heavier on this exact GPU.

NotRexed commented

2026-11-09 06:05:52 +00:00

(Migrated from github.com)

Tracked in #1006, follow there. That issue collects AMD-specific rendering problems, including compute programs that AMD rejects (#224 found three on an 890M). Your counts (10 failed at the title, 11 in Hyrule Field) are noted there with a link back here. Please attach the full hookshot_log.txt in #1006, ideally from v2.0.6 (https://github.com/NotRexed/Hookshot/releases/tag/v2.0.6; Update next to Play in the launcher).

Commenting is not possible because the repository is archived.