WIP: feat(vk): the per-PC decisions get one place and one log block: device_policy.h #1341

Closed
HumanBean wants to merge 5 commits from HumanBean/Longshot:feat/device-policy into main
Owner

Type: feat · Risk: low · Tested on: Linux, RTX 3090 · Needs: Windows; an AMD and an Intel GPU

Why

73 rules decide by the PC (the GPU's vendor, driver, VRAM, the CPU count, the OS) in about 40 functions of 14 files. The same fact is asked several ways, switches are parsed seven ways, only two rules name a driver version, and no log shows what a PC got, so a report from an Intel or AMD card starts with guessing which rules ran. Track F of the renderer rework plan: one place for the per-PC decisions, step 1. Stacked on #1309, #1325 and #1326: review the last commit only.

What changed

  • gpu/vk/device_policy.h (new): DeviceFacts, filled once before the device is made, and a pure device_policy() that works out 67 decisions from it, each field with its reason. No Vulkan calls.
  • At the start five vk: policy: lines say the facts and every decision.
  • Four sites read their unset default from the policy: astc_on_cpu, batch_on, FSR 4's tune::on, the spare captures' cap. Each keeps its own switch and its parsing.
  • Every other rule is computed here by the same rule but still decided at its site, because a check cuts that site by text; each field's comment names the step that moves it. Two rules are therefore written twice for now (the Intel native-add list, the integrated GPUs' RAM table) and could drift until their sites move.
  • No behaviour change is intended for any PC or switch value. Rules the design study doubts (small-card mirrors, Linux AMD taken as RADV, Intel rules that reach Mesa) are pinned as they are today, with a comment.
  • Two read-only driver queries are new at start (memory properties and the atomic-float feature).
  • tests/test_device_policy.py (new): 18 made-up PCs with the expected value of every field.

Measurements

Nothing to measure. One session as a check: Kokiri Forest 67.3 fps standing, Hyrule Field 92.0, main's draw counts and GPU times.

Hardware: Linux (CachyOS, kernel 7.3.3), RTX 3090 24 GB, NVIDIA 620.70, Ryzen 9 7900X; 2560x1440, no upscaler

vk: policy: facts 10de:2204, driver id 4, version 0x9b118180 (620.70.06), type 2, not Windows; VRAM 24576 MB, allowed at the start 22859 MB, RAM 31658 MB, resizable BAR 1; CPUs 24 (24 usable); mesh shaders 1, float add 1, cooperative matrices 1, 16-bit stage inputs 0, lanes 32 to 32 (32 on stages 0x3fff)
vk: policy: can and measured (unset defaults): mirrors=1 import_chunk=4 MB ub_snapshot=0 split_ub_flush=1 native_float_add=1 warp64=0 batch_submits=0 mma_unroll_try=0 global_table_loop=0 occ_copy=0 intel_side_by_side_auto=0 fsr4_tune=0 gpu_clock_peak=0 presenter_queue=1 pcache_small_pc=0

Flags and settings

None added or changed.

Testing

Tested: Linux, RTX 3090: built; one session through Kokiri Forest and Hyrule Field, both drew as main does; the five policy lines were printed. Of the four converted sites, this PC takes the same side as before on each (all off or unlimited).

Not tested:

  • tests/run_all.py and the new check, whose about 1,200 expected values were worked out by hand: not run on this PC.
  • Windows: not compiled. The header pulls several others into every renderer file, and an enum name there could meet a Windows macro.
  • Any PC where a converted site takes its other side: RADV (batched submits), an AMD card with FSR 4 (tune), the two AMD integrated chips on Windows (ASTC on the CPU), a card of 6.5 GB or less (spare captures).

Must be tested:

  • A Steam Deck or another RADV card: vk: submits: batched should read as on main (Test Farm: @Beanstalk steamdeck: main against this PR, Kokiri Forest and Hyrule Field).
  • Windows AMD and Windows Intel: a start and the same two areas, and the policy block read against what the card is.

Risks

What could break: a converted site taking the other side on a PC this one is not, or the Windows build.

Turning it off: there is no switch; revert the commit.

Compatibility: none: no cache is rebuilt, MKWVK_API_VERSION stays, no setting changes.

Issues

See #1300

Checklist

Everyone

  • One topic.
  • The title is a Conventional Commit (type(scope): description) of the status line's type.
  • Why gives the reason plainly and links the issues that fit (the tracker searched).
  • Measurements has a verdict, or one line saying why there is nothing to measure.
  • Testing names what was tested, what wasn't, and what must still be tested, where and how.
  • Rebased on current main.
  • python -I -B tests/run_all.py passes (any failure that also fails on main is named). Not run on this PC; left to the forge's checks.
  • No game files, dumps, build output or generated files.
  • third_party/ untouched (a new folder with a PROVENANCE.md instead).
  • The license is OK: GPL-3.0-or-later, nothing incompatible and nothing decompiled.

When it applies

  • AI assistance recorded with Assisted-by: / Co-Authored-By: trailers.
  • Numbers from the same PC, scene and settings, at least 2 interleaved runs each, the median reported.
  • Switches documented (docs/development.md; the README for a menu row) and shipped at their recommended value.
  • A check under tests/ that reads changed source updated in this PR, not weakened.
  • Host code builds on both Windows (llvm-mingw) and Linux, with UTF-8 paths. Built on Linux only here.
  • Docs updated.

🤖 Generated with Claude Code

https://claude.ai/code/session_017vw8dzCsAL3UdW7ft8ptL9

**Type:** feat · **Risk:** low · **Tested on:** Linux, RTX 3090 · **Needs:** Windows; an AMD and an Intel GPU ## Why 73 rules decide by the PC (the GPU's vendor, driver, VRAM, the CPU count, the OS) in about 40 functions of 14 files. The same fact is asked several ways, switches are parsed seven ways, only two rules name a driver version, and no log shows what a PC got, so a report from an Intel or AMD card starts with guessing which rules ran. Track F of the [renderer rework plan](https://git.magicbean.dev/MagicBeanDev/Longshot/wiki/Renderer-rework-plan): one place for the per-PC decisions, step 1. Stacked on #1309, #1325 and #1326: review the last commit only. ## What changed - `gpu/vk/device_policy.h` (new): `DeviceFacts`, filled once before the device is made, and a pure `device_policy()` that works out 67 decisions from it, each field with its reason. No Vulkan calls. - At the start five `vk: policy:` lines say the facts and every decision. - Four sites read their unset default from the policy: `astc_on_cpu`, `batch_on`, FSR 4's `tune::on`, the spare captures' cap. Each keeps its own switch and its parsing. - **Every other rule is computed here by the same rule but still decided at its site**, because a check cuts that site by text; each field's comment names the step that moves it. Two rules are therefore written twice for now (the Intel native-add list, the integrated GPUs' RAM table) and could drift until their sites move. - No behaviour change is intended for any PC or switch value. Rules the design study doubts (small-card mirrors, Linux AMD taken as RADV, Intel rules that reach Mesa) are pinned as they are today, with a comment. - Two read-only driver queries are new at start (memory properties and the atomic-float feature). - `tests/test_device_policy.py` (new): 18 made-up PCs with the expected value of every field. ## Measurements Nothing to measure. One session as a check: Kokiri Forest 67.3 fps standing, Hyrule Field 92.0, main's draw counts and GPU times. Hardware: Linux (CachyOS, kernel 7.3.3), RTX 3090 24 GB, NVIDIA 620.70, Ryzen 9 7900X; 2560x1440, no upscaler ``` vk: policy: facts 10de:2204, driver id 4, version 0x9b118180 (620.70.06), type 2, not Windows; VRAM 24576 MB, allowed at the start 22859 MB, RAM 31658 MB, resizable BAR 1; CPUs 24 (24 usable); mesh shaders 1, float add 1, cooperative matrices 1, 16-bit stage inputs 0, lanes 32 to 32 (32 on stages 0x3fff) vk: policy: can and measured (unset defaults): mirrors=1 import_chunk=4 MB ub_snapshot=0 split_ub_flush=1 native_float_add=1 warp64=0 batch_submits=0 mma_unroll_try=0 global_table_loop=0 occ_copy=0 intel_side_by_side_auto=0 fsr4_tune=0 gpu_clock_peak=0 presenter_queue=1 pcache_small_pc=0 ``` ### Flags and settings None added or changed. ## Testing **Tested:** Linux, RTX 3090: built; one session through Kokiri Forest and Hyrule Field, both drew as main does; the five policy lines were printed. Of the four converted sites, this PC takes the same side as before on each (all off or unlimited). **Not tested:** - `tests/run_all.py` and the new check, whose about 1,200 expected values were worked out by hand: not run on this PC. - Windows: not compiled. The header pulls several others into every renderer file, and an enum name there could meet a Windows macro. - Any PC where a converted site takes its other side: RADV (batched submits), an AMD card with FSR 4 (tune), the two AMD integrated chips on Windows (ASTC on the CPU), a card of 6.5 GB or less (spare captures). **Must be tested:** - A Steam Deck or another RADV card: `vk: submits: batched` should read as on main (Test Farm: `@Beanstalk steamdeck: main against this PR, Kokiri Forest and Hyrule Field`). - Windows AMD and Windows Intel: a start and the same two areas, and the policy block read against what the card is. ## Risks **What could break:** a converted site taking the other side on a PC this one is not, or the Windows build. **Turning it off:** there is no switch; revert the commit. **Compatibility:** none: no cache is rebuilt, `MKWVK_API_VERSION` stays, no setting changes. ## Issues See #1300 ## Checklist ### Everyone - [x] One topic. - [x] The title is a Conventional Commit (`type(scope): description`) of the status line's type. - [x] Why gives the reason plainly and links the issues that fit (the tracker searched). - [x] Measurements has a verdict, or one line saying why there is nothing to measure. - [x] Testing names what was tested, what wasn't, and what must still be tested, where and how. - [x] Rebased on current `main`. - [ ] `python -I -B tests/run_all.py` passes (any failure that also fails on `main` is named). Not run on this PC; left to the forge's checks. - [x] No game files, dumps, build output or generated files. - [x] `third_party/` untouched (a new folder with a `PROVENANCE.md` instead). - [x] The license is OK: GPL-3.0-or-later, nothing incompatible and nothing decompiled. ### When it applies - [x] AI assistance recorded with `Assisted-by:` / `Co-Authored-By:` trailers. - [x] Numbers from the same PC, scene and settings, at least 2 interleaved runs each, the median reported. - [x] Switches documented (docs/development.md; the README for a menu row) and shipped at their recommended value. - [x] A check under `tests/` that reads changed source updated in this PR, not weakened. - [ ] Host code builds on both Windows (llvm-mingw) and Linux, with UTF-8 paths. Built on Linux only here. - [x] Docs updated. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_017vw8dzCsAL3UdW7ft8ptL9
refactor(vk): 93 kill switches and the older paths they kept are deleted from the four large renderer files
All checks were successful
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 7s
Guide screenshots / Guide images and local UI source review (pull_request) Successful in 11s
Third-party parts / As their notes say (pull_request) Successful in 14s
CI / Workflow lint (actionlint) (pull_request) Successful in 19s
PR description / Description follows the template (pull_request_target) Successful in 5s
Website / Website on Cloudflare (pull_request_target) Has been skipped
b9805ef295
The renderer rework plan's track F. A kill switch turned off, or back to an older way, something that is on for
every PC; nobody ships with one set, and each kept an old path alive. vk_draw.cpp 32, vk_texture.cpp 21,
vk_replay.cpp 20, vk_memory.cpp 20: the read and the path are gone, the default is the only behaviour, and each name
is a GONE row of gpu/vk/switches.def, so a run that still sets one is told. 26 are kept, each for a reason (a value
set by a tool or named in the README, a default that follows the PC, a young fix's way out).

Co-Authored-By: Claude Fable 5.2 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8dzCsAL3UdW7ft8ptL9
refactor(vk): 18 more kill switches and their older paths are deleted from the smaller renderer files
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 8s
PR description / Description follows the template (pull_request_target) Successful in 13s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 14s
CI / Workflow lint (actionlint) (pull_request) Successful in 20s
Website / Website on Cloudflare (pull_request_target) Has been skipped
Third-party parts / As their notes say (pull_request) Successful in 23s
5742f65ca1
The renderer rework plan's track F, after the four large files (#1309). vk_vma.cpp, vk_main.cpp, vk_temporal.cpp,
vk_xfer.cpp, vk_dlss.cpp, vk_fsr4.cpp, vk_jobs.cpp, vk_reflex.cpp, vk_framegen.cpp, vk_programs.cpp and vk_progcache.cpp
lose the read and the path; each name is a GONE row of gpu/vk/switches.def. 18 are kept: the README or a script
names them, the shader cache's stamp holds them, or they are a way out that measurement runs or a driver still need.

Co-Authored-By: Claude Fable 5.2 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8dzCsAL3UdW7ft8ptL9
feat(host): the host's switches join the table: every HOS_ and LONGSHOT_ name the game reads has a row
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 13s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 15s
CI / Workflow lint (actionlint) (pull_request) Successful in 17s
Third-party parts / As their notes say (pull_request) Successful in 17s
Website / Website on Cloudflare (pull_request_target) Has been skipped
PR description / Description follows the template (pull_request_target) Successful in 5s
2e0743237d
The renderer rework plan's track F. gpu/vk/switches.def gains the host's 216 switches (89 diagnostics, 50 that turn
something off, 23 player settings, 24 numbers, 20 that turn on something nobody ships, 8 that follow the PC), read
once each from the code. The start's report names them when set and says a HOS_ name that is none; a LONGSHOT_ name
that is none is passed over (the launcher and the Builder have their own). The host's three removed HOS_ names become
GONE rows.

Co-Authored-By: Claude Fable 5.2 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8dzCsAL3UdW7ft8ptL9
feat(vk): the per-PC decisions get one place and one log block: device_policy.h
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 11s
PR description / Description follows the template (pull_request_target) Successful in 15s
CI / Workflow lint (actionlint) (pull_request) Successful in 15s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 16s
Third-party parts / As their notes say (pull_request) Successful in 15s
Website / Website on Cloudflare (pull_request_target) Has been skipped
beanstalk/approval planning the tests
9765371bb6
The renderer rework plan's track F, the per-GPU rules, step 1. DeviceFacts is filled once before the device is made
and device_policy() works out 67 decisions from it, each with its reason; "vk: policy:" says them at the start. Four
sites read their unset default from it (astc_on_cpu, batch_on, FSR 4's tune, the spare captures' cap); the others
still decide where they did, and no switch's parsing changes.

Co-Authored-By: Claude Fable 5.2 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8dzCsAL3UdW7ft8ptL9
Author
Owner

Tested in game: Linux, RTX 3090, NVIDIA 620.70: one session through Kokiri Forest and Hyrule Field at 2560x1440; both drew as main, and the five vk: policy: lines were printed.

Tested in game: Linux, RTX 3090, NVIDIA 620.70: one session through Kokiri Forest and Hyrule Field at 2560x1440; both drew as main, and the five vk: policy: lines were printed.
Author
Owner

A second commit is on this branch. Unlike the first, it changes behaviour on GPUs this PC is not, so it needs runs there before it leaves WIP.

One reader, one rule. device_policy_env reads 15 switches of the converted sites: unset or empty is the PC's default, an off word is off, anything else is on; numbers are digits only and in range, else ignored and said in the log. Values whose meaning changes: =on/=true on RX_VK_BATCH, RX_VK_MIRROR, RX_VK_OCC_COPY, RX_VK_COMPUTE_SERIAL (were off, now on); an empty value on nine switches (was off, now the default); RX_VK_WORKERS=0, a word, or over 16 (was 1 or 16, now ignored); negative or non-numeric sizes (were huge or 0, now ignored). 0 and 1 mean what they did everywhere. docs/development/device-policy.md (new) has the table.

The log block marks a value its switch set with *, names ignored values, and gives the driver version each driver fix that is on was seen on:

vk: policy: auto settings and threads: ... workers 4*, capture helpers 4, ...
vk: policy: switches (* above: set by its switch; the other values are this PC's defaults): RX_VK_MIRROR=on; RX_VK_EQ_BIAS=abc ignored (not a number from 0 to 65536); RX_VK_WORKERS=4

Four rules now test the driver their own comment names, not the vendor:

  1. Compute for task and mesh draws under Automatic was "AMD, not Windows"; it is RADV. AMDVLK on Linux now uses its mesh shaders. Never run there.
  2. One program's builds at a time was every AMD card; it is AMD's Windows driver. RADV and AMDVLK now build a program's variants side by side.
  3. Serial compute builds were every Intel GPU; they are Intel's Windows driver. ANV now builds compute pipelines on the background workers.
  4. Batched submits tested for GCN cards a second way; they use the one definition. RADV with RX_VK_NO_SUBGROUP_PIN=1 on a 64-lane card keeps batching, and (by reading, not a log) RADV on Vega and Raven/Picasso APUs gains batching.

Each is one commented line in device_policy() that says how to go back. Left for hardware nobody here has: the native add's list for other Intel families, the equal-depth margin and part split on ANV, the small-card sizes.

Run in the game (RTX 3090, NVIDIA 620.70, Linux): two sessions through Kokiri Forest, Hyrule Field and the Market, one with three switches set. Scenes drew as main (66.3 / 89.9 / 69.3 fps and 65.2 / 91.5 / 69.0); the block printed as above. None of the four corrected rules applies to this card. Not tested: any AMD or Intel GPU, Windows (not compiled), and tests/run_all.py: nine existing checks were edited by reading alone.

Must be tested: a Steam Deck and a desktop RADV card (@Beanstalk steamdeck: main against this PR, Kokiri Forest, Hyrule Field, the Market), Intel on Linux, and Windows AMD and Intel for the build.

A second commit is on this branch. Unlike the first, **it changes behaviour on GPUs this PC is not**, so it needs runs there before it leaves WIP. **One reader, one rule.** `device_policy_env` reads 15 switches of the converted sites: unset or empty is the PC's default, an off word is off, anything else is on; numbers are digits only and in range, else ignored and said in the log. Values whose meaning changes: `=on`/`=true` on `RX_VK_BATCH`, `RX_VK_MIRROR`, `RX_VK_OCC_COPY`, `RX_VK_COMPUTE_SERIAL` (were off, now on); an empty value on nine switches (was off, now the default); `RX_VK_WORKERS=0`, a word, or over 16 (was 1 or 16, now ignored); negative or non-numeric sizes (were huge or 0, now ignored). `0` and `1` mean what they did everywhere. `docs/development/device-policy.md` (new) has the table. **The log block** marks a value its switch set with `*`, names ignored values, and gives the driver version each driver fix that is on was seen on: ``` vk: policy: auto settings and threads: ... workers 4*, capture helpers 4, ... vk: policy: switches (* above: set by its switch; the other values are this PC's defaults): RX_VK_MIRROR=on; RX_VK_EQ_BIAS=abc ignored (not a number from 0 to 65536); RX_VK_WORKERS=4 ``` **Four rules now test the driver their own comment names, not the vendor:** 1. Compute for task and mesh draws under Automatic was "AMD, not Windows"; it is RADV. **AMDVLK on Linux now uses its mesh shaders.** Never run there. 2. One program's builds at a time was every AMD card; it is AMD's Windows driver. **RADV and AMDVLK now build a program's variants side by side.** 3. Serial compute builds were every Intel GPU; they are Intel's Windows driver. **ANV now builds compute pipelines on the background workers.** 4. Batched submits tested for GCN cards a second way; they use the one definition. RADV with `RX_VK_NO_SUBGROUP_PIN=1` on a 64-lane card keeps batching, and (by reading, not a log) RADV on Vega and Raven/Picasso APUs gains batching. Each is one commented line in `device_policy()` that says how to go back. Left for hardware nobody here has: the native add's list for other Intel families, the equal-depth margin and part split on ANV, the small-card sizes. **Run in the game** (RTX 3090, NVIDIA 620.70, Linux): two sessions through Kokiri Forest, Hyrule Field and the Market, one with three switches set. Scenes drew as main (66.3 / 89.9 / 69.3 fps and 65.2 / 91.5 / 69.0); the block printed as above. None of the four corrected rules applies to this card. **Not tested:** any AMD or Intel GPU, Windows (not compiled), and `tests/run_all.py`: nine existing checks were edited by reading alone. Must be tested: a Steam Deck and a desktop RADV card (`@Beanstalk steamdeck: main against this PR, Kokiri Forest, Hyrule Field, the Market`), Intel on Linux, and Windows AMD and Intel for the build.
feat(vk): 15 per-PC switches are read in one place by one rule, and four rules test the driver they name
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 7s
PR description / Description follows the template (pull_request_target) Successful in 10s
CI / Workflow lint (actionlint) (pull_request) Successful in 11s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 10s
Third-party parts / As their notes say (pull_request) Successful in 16s
beanstalk/approval not approved yet
99d7880615
The renderer rework plan's track F, the per-GPU rules, steps 2 to 4. device_policy_env reads the switches of the
converted sites by one rule (unset or empty: the PC's default; an off word: off; else on; numbers digits only, in
range, else ignored and said), and the start's block marks what a switch set and the driver version each driver fix
was seen on. Fifteen sites read dev.policy; the native add's list and the integrated GPUs' RAM table are written
once. Four rules that tested the vendor now test the driver their reason names: Compute under Automatic is RADV's
(AMDVLK keeps its mesh shaders), one program's builds at a time is AMD's Windows driver's, serial compute builds are
Intel's Windows driver's, and batch_on uses the one definition of GCN.

Co-Authored-By: Claude Fable 5.2 <[email protected]>
Claude-Session: https://claude.ai/code/session_017vw8dzCsAL3UdW7ft8ptL9
Member

Beanstalk test farm

Commit 99d7880 · tier untrusted · approved by nobody yet · dashboard

A member of BigBeans can review this commit and comment @Beanstalk to test it.

<!-- beanstalk:sticky --> ### Beanstalk test farm Commit `99d7880` · tier **untrusted** · approved by nobody yet · [dashboard](https://beanstalk.magicbean.dev/prs/1341) A member of `BigBeans` can review this commit and comment `@Beanstalk` to test it.
Author
Owner

Folded into #1309, which is now the one pull request for switches and the per-GPU rules. The commits are unchanged apart from being rebased into one stack and carrying the fixes the check suite asked for; nothing was dropped.

Folded into #1309, which is now the one pull request for switches and the per-GPU rules. The commits are unchanged apart from being rebased into one stack and carrying the fixes the check suite asked for; nothing was dropped.
HumanBean closed this pull request 2026-11-17 15:48:16 +00:00
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 7s
PR description / Description follows the template (pull_request_target) Successful in 10s
CI / Workflow lint (actionlint) (pull_request) Successful in 11s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 10s
Third-party parts / As their notes say (pull_request) Successful in 16s
beanstalk/approval not approved yet

Pull request closed

Sign in to join this conversation.
No reviewers
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!1341
No description provided.