build(release): a contributed pipeline list from an AMD card with FSR 4, 1273 pipelines #2044

Merged
madebyjordan merged 1 commit from pipelines/amd-rx6700xt-fsr4 into main 2026-11-11 00:56:20 +00:00
madebyjordan commented 2026-11-11 00:54:08 +00:00 (Migrated from github.com)

The shipped pipeline lists come from NVIDIA cards, so a first start on an AMD card with FSR 4 still builds many pipelines in play. This adds a list from an RX 6700 XT for #2034: 1,273 pipelines no other list holds (the merged list goes from 11,527 to 12,800), 432 of them for programs the lists don't have yet.

What changed

Two files, written by nx2/pipeline_states.py add:

  • assets/pipeline_states/contrib/madebyjordan-amd-fsr4-field-castle-town.bin.xz
  • assets/pipeline_states/contrib/madebyjordan-amd-fsr4-field-castle-town.md
GPU and driver AMD Radeon RX 6700 XT (12 GB), Adrenalin 26.10.1 (Vulkan driver 2.0.353), Windows 11 Home 25H2
Output resolution 2560x1440
Upscaler + render resolution AMD FSR 4 (4.1.2), 1280x720
Frame generation / HDR off / off
Mods or texture packs none
Where Hyrule Field by day, the bridge and the way into Castle Town, Castle Town, a village at night. About an hour of play, plus replays of a recorded six-minute route from the Hyrule Field bridge into Castle Town

Two things to know before merging

It is not from a stock main build. The README asks for a build of current main; this list was written by Longshot dev builds of 2026-11-10 (pre-1.1.0 a604927): main at 0828863 (2026-11-09) plus AMD changes. Those changes leave the shader translation and the record as they are (gpu/vk/eden untouched, vk_shader.cpp only where the cache files live, the record code only the list's folder and a read-only switch).

That build is from before #1627, which changed what the translation asks about some textures. What I did about it:

  • Compared with the lists here, by program and variant: 1,678 of the recorded pipelines have a program variant another list holds with the same question list, and 117 have one with a different list (mostly 03 where the current lists have 02). I left those 117 out.
  • Some of the 1,273 may be out of date in the same way, where no list here has the same program variant to compare with. The ratio above suggests a few percent.
  • As I read state_list_prepare, such a record is skipped at the first start: its variant is translated from the recorded questions and only used when it comes out with the recorded hash. So it would cost list size and one translation, and build nothing wrong. Say if you would still rather have a list recorded on current main, and I'll record one.

The list is also two lists from the same PC and settings merged into one: the game folder's, and the one written by the route replays.

Tested

  • python -I nx2/pipeline_states.py check: passes, merged: 12800 pipelines (graphics 12159, mesh 641) of 7104 programs
  • tests/test_pipeline_state_sources.py (8 tests), tests/test_pipeline_states.py, tests/test_ship_state_list.py (1 skipped), tests/test_pipeline_state_counts.py: pass
  • Not tested: a first start of a main build with the merged list.

Refs #2034

🤖 Generated with Claude Code

The shipped pipeline lists come from NVIDIA cards, so a first start on an AMD card with FSR 4 still builds many pipelines in play. This adds a list from an RX 6700 XT for #2034: **1,273 pipelines** no other list holds (the merged list goes from 11,527 to 12,800), 432 of them for programs the lists don't have yet. ## What changed Two files, written by `nx2/pipeline_states.py add`: - `assets/pipeline_states/contrib/madebyjordan-amd-fsr4-field-castle-town.bin.xz` - `assets/pipeline_states/contrib/madebyjordan-amd-fsr4-field-castle-town.md` | | | |---|---| | GPU and driver | AMD Radeon RX 6700 XT (12 GB), Adrenalin 26.10.1 (Vulkan driver 2.0.353), Windows 11 Home 25H2 | | Output resolution | 2560x1440 | | Upscaler + render resolution | AMD FSR 4 (4.1.2), 1280x720 | | Frame generation / HDR | off / off | | Mods or texture packs | none | | Where | Hyrule Field by day, the bridge and the way into Castle Town, Castle Town, a village at night. About an hour of play, plus replays of a recorded six-minute route from the Hyrule Field bridge into Castle Town | ## Two things to know before merging **It is not from a stock main build.** The README asks for a build of current `main`; this list was written by [Longshot](https://github.com/madebyjordan/Longshot) dev builds of 2026-11-10 (`pre-1.1.0 a604927`): `main` at 0828863 (2026-11-09) plus AMD changes. Those changes leave the shader translation and the record as they are (`gpu/vk/eden` untouched, `vk_shader.cpp` only where the cache files live, the record code only the list's folder and a read-only switch). **That build is from before #1627**, which changed what the translation asks about some textures. What I did about it: - Compared with the lists here, by program and variant: 1,678 of the recorded pipelines have a program variant another list holds with the same question list, and 117 have one with a different list (mostly `03` where the current lists have `02`). I left those 117 out. - Some of the 1,273 may be out of date in the same way, where no list here has the same program variant to compare with. The ratio above suggests a few percent. - As I read `state_list_prepare`, such a record is skipped at the first start: its variant is translated from the recorded questions and only used when it comes out with the recorded hash. So it would cost list size and one translation, and build nothing wrong. Say if you would still rather have a list recorded on current `main`, and I'll record one. The list is also two lists from the same PC and settings merged into one: the game folder's, and the one written by the route replays. ## Tested - `python -I nx2/pipeline_states.py check`: passes, `merged: 12800 pipelines (graphics 12159, mesh 641) of 7104 programs` - `tests/test_pipeline_state_sources.py` (8 tests), `tests/test_pipeline_states.py`, `tests/test_ship_state_list.py` (1 skipped), `tests/test_pipeline_state_counts.py`: pass - Not tested: a first start of a `main` build with the merged list. Refs #2034 🤖 Generated with [Claude Code](https://claude.com/claude-code)
Kirkidoo commented 2026-11-11 01:07:03 +00:00 (Migrated from github.com)

My pipeline list is attached: hookshot_pipeline_states.zip (only that one file). It holds 1,654 pipelines. Checked with nx2/pipeline_states.py against the list on main (4,276 pipelines): 797 of mine aren't in it yet.

  • Hookshot version: main-abd5e5f (built from main, commit abd5e5f). The list also carries over sessions on v2.0.13 / v2.0.14 in the same game folder.
  • OS, GPU and driver: Windows 11 Pro, AMD Radeon RX 9070 XT, driver 32.0.32015.7029
  • Output resolution: 2560x1440, fullscreen (monitor is 3440x1440)
  • Upscaler + render resolution: game's (Lanczos), 100%. Some earlier sessions used FSR 3.1.
  • Frame generation / HDR: off / off
  • Mods or texture packs: none
  • Where you played: Kokiri Forest

hookshot_pipeline_states.zip

My pipeline list is attached: hookshot_pipeline_states.zip (only that one file). It holds 1,654 pipelines. Checked with nx2/pipeline_states.py against the list on main (4,276 pipelines): 797 of mine aren't in it yet. - Hookshot version: main-abd5e5f (built from main, commit abd5e5f). The list also carries over sessions on v2.0.13 / v2.0.14 in the same game folder. - OS, GPU and driver: Windows 11 Pro, AMD Radeon RX 9070 XT, driver 32.0.32015.7029 - Output resolution: 2560x1440, fullscreen (monitor is 3440x1440) - Upscaler + render resolution: game's (Lanczos), 100%. Some earlier sessions used FSR 3.1. - Frame generation / HDR: off / off - Mods or texture packs: none - Where you played: Kokiri Forest [hookshot_pipeline_states.zip](https://github.com/user-attachments/files/34661556/hookshot_pipeline_states.zip)
Commenting is not possible because the repository is archived.
No description provided.