fix(pipelines): run All prediction after preparing on Most #713
No reviewers
Labels
No labels
area/accessibility
area/audio
area/input
area/install
area/menu
area/renderer
area/saves
area/tooling
area/upscaling
beanstalk: approved
beanstalk: failed
beanstalk: needs approval
beanstalk: running
beanstalk: tested
ci: approved
good first issue
gpu/amd
gpu/igpu
gpu/intel
gpu/nvidia
help wanted
kind
bug
kind
chore
kind
docs
kind
feature
kind
performance
kind
question
needs measurements
platform/linux
platform/macos
platform/steam-deck
platform/windows
preview
priority
critical
priority
high
priority
low
regression
risk
high
risk
low
risk
medium
status
blocked
status
confirmed
status
duplicate
status
invalid
status
needs-info
status
wontfix
ci: approved
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
MagicBeanDev/Longshot!713
Loading…
Reference in a new issue
No description provided.
Delete branch "klimovich008/Longshot:fix/prepare-all-after-most"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Type: fix · Risk: low · Tested on: Windows, RX 9070 XT (26.10.2 and 26.11.1); Linux clang 21 · Needs: none
Why
Choosing All after preparing on Most can finish without predicting any pipelines. Those missing pipelines must then be built during play.
How to reproduce
Seen on: Windows 11, RX 9070 XT, driver 26.10.2.
Settings and switches: prepare on Most, then request All without changing the input or driver.
Steps: run
longshot.exe --prepare-shaders --compile-shaders allon that folder.Expected: prediction runs.
Actual: preparation returns without prediction.
What changed
Explicit All also runs when its prediction-completion record is missing or invalid. Valid completed work, including zero keys, stays skipped.
vk_main.cpp: check completion before skipping explicit All.vk_draw.cpp: share the stamped reader; reject truncated or trailing data before allocating.Measurements
Verdict: Fails on the old code, passes now. This fixes the skipped prediction; it does not claim universal pipeline coverage or a gameplay speedup.
main:
35465d12(2026-11-15)this PR: codex/prepare-all-after-most @
d252a220Baseline: API22 base
82886e89; both native arms retain identical pipeline-cache/progress changes. The source difference is this fix; the new pack also restores the base renderer profile. No timings are compared.Hardware: Windows 11 Home N 26300; RX 9070 XT, 26.10.2; Ryzen 7 5700X3D, 64 GB.
Scene: preparation before the title; no gameplay frames.
How to reproduce: finish Most preparation, then run
longshot.exe --prepare-shaders --compile-shaders all.Method: preparation CLI before the title; completion files set aside to match the genuine Most base. Both arms use
RX_VK_PREDICT_S=1.Flags and settings
No new switches. The short native regression check used the same limit in both arms; the separate All sanity used defaults.
RX_VK_PREDICT_STesting
Tested:
python -I -B tests/test_prepare_cache_save.py --cxx clang++; missing/corrupt/zero-key records, scheduling, progress, save. Native AMD red/green above.python -I -B tests/run_all.py --jobs 4: 536 pass, 19 fail, 38 skip. All 19 fail on clean35465d12 too:test_appimage_player_build,test_cpm_discovery,test_cpp_preflight,test_crash_dialog,test_env_on,test_fpx_sticky,test_lift_cleanup_error,test_linux_preflight,test_make_game_cache,test_make_game_prepare,test_missing_nup,test_oot_state_tap,test_readback_late,test_release,test_spawn_emit,test_tested_configurations,test_vk_pgo,test_vram_budget_row,test_walk_key.Linux clang 21: the regression fixture passes and both changed C++ files compile with the project's Linux flags. Windows native backport linked and ran. No Linux module link or gameplay run.
Not tested: gameplay performance, other GPU drivers, an isolated full same-driver Most-to-All upgrade.
Default-All sanity on AMD26.11.1 also completed in both API22 integration builds, with zero failed/missing noted graphics or computes. These include other cache fixes; their times are not compared. The new tested records name those builds.
Must be tested: None for the scheduling regression. This does not claim universal pipeline coverage.
Risks
What could break: invalid completion records cause an explicit All request to predict again.
Turning it off: revert the commit.
Compatibility: saves, options, translated shaders, GPU cache format and renderer ABI stay unchanged. Automatic startup retry/give-up rules are unchanged.
Checklist
Everyone
type(scope): description) of the status line's type.main.python -I -B tests/run_all.pypasses (any failure that also fails onmainis named).third_party/untouched (a new folder with aPROVENANCE.mdinstead).When it applies
AI assistance recorded with
Assisted-by:/Co-Authored-By:trailers.A file in
docs/tested/for each new measured run (format indocs/tested-configurations.md).A game-free test added for the fix, failing on
mainand passing here.A check under
tests/that reads changed source updated in this PR, not weakened.Docs updated.
Host code builds on both Windows (llvm-mingw) and Linux, with UTF-8 paths.
WIP: fix(vk): All after a Most preparation predicts the pipelines it skippedto WIP: fix(pipelines): run All prediction after preparing on Most3a30d7a7fcd252a220a1Rebased onto main
35465d12(d252a220, Codex's source; only this PR's five files plus two sanity run records). Both default-All preparations on the new AMD driver 26.11.1 completed (24,626 graphics + 9,687 compute pipelines noted, none missing); full suite 536/19/38 with the same 19 failures as clean main; Linux clang 21 compiles the changed files. Taking it out of WIP.WIP: fix(pipelines): run All prediction after preparing on Mostto fix(pipelines): run All prediction after preparing on MostRefreshed onto main
35465d12(d252a220; the fixc6d896f4plus two tested-run records): the regression fixture fails on the old code and passes here, the full suite fails only the 19 tests that also fail on clean35465d12, and default All preparation on AMD 26.11.1 completed with nothing missing in both integration builds. The tracker search found no matching issue. Out of WIP.