MagicBeanDev/Longshot

Watch

Fork 99

revert(vk): the eviction's waits in presents again at any frame rate, as before #771 #873

Merged

HumanBean merged 1 commit from HumanBean/Longshot:revert/evict-seconds into main

2026-11-15 18:50:08 +00:00

HumanBean commented

2026-11-15 18:34:32 +00:00

Owner

Type: revert · Risk: low · Tested on: none (game-free: the tree is main before #771, fd770e7f's parent, byte for byte) · Needs: none

Why

#771 was merged while its review asked for a change. Its idea is right: below 60 fps, the waits before an unused texture is let go should be counted in seconds, not in frames. But the way it measures the frame rate goes wrong right after a load or a long hitch, even at a steady 60 fps.

It ages every texture by the frame rate averaged over the last ten seconds. A gap with no frames, such as an area load's stall, makes that average read far too low for the next ten seconds. At 60 fps after a 30 s gap it reads about 2 fps one second in and about 8.6 fps five seconds in, so textures look 7 to 30 times older than they are:

a texture unused for 1.4 s counts as past the ten-second line, and one unused for 4.3 s as past the thirty-second line;
the minute kept for textures the GPU drew drops to as little as 2 s.

That happens just after an area loads, when VRAM is tightest. The textures let go are uploaded again, and one the GPU drew can come back empty, which the minute's wait exists to prevent. The review's details: #771 (comment)

This takes it back out so that main and the next release candidate don't ship it. The change comes back in its own pull request, with the age measured per texture in seconds instead of through one average rate.

See #771.

What changed

A plain jj revert of fd770e7f (#771). Main goes back to counting the compaction's hold and reserve, and the GPU-drawn textures' keep, in presents at any frame rate, as before #771.

gpu/vk/vk_texture.cpp: the frame-rate marks and present_rate removed; hold_target and the keep take the present counts again.
gpu/vk/vram_compact.h: age_at_60 removed. gpu/vk/keep_rendered.h: at_rate removed.
docs/development.md: the RX_VK_EVICT_BY_PRESENTS row removed.
tests/test_keep_rendered.*, tests/test_vram_compact.*: back to main's before #771.

Measurements

Nothing to measure: this restores main's code from before #771 exactly (jj diff --from fd770e7f- --to this commit is empty), whose behaviour is what every release so far has shipped.

Flags and settings

RX_VK_EVICT_BY_PRESENTS (added by #771) goes away with it: what it switched back to is the only behaviour again.

Flag	What it is	Before this PR	After this PR	Effect	Helps / doesn't help / untested	Recommendation
`RX_VK_EVICT_BY_PRESENTS`	#771's off switch: the waits in presents	unset: waits by the measured rate below 60 fps	removed: waits in presents always	as before #771	before #771's behaviour, shipped in every release	removed with #771

Testing

Tested: game-free only, by reading: the reverted tree is identical to main before #771 (an empty diff against fd770e7f's parent). No tests were run on this PC; the checks are left to CI.

Not tested: nothing new to test: the code is what main ran before #771 merged, today.

Must be tested: None.

Risks

What could break: nothing that main didn't do before today: below 60 fps, the hold's ten and thirty seconds and the keep's minute stretch again (at 3.5 fps they are three, nine and seventeen minutes), which is what #771 set out to fix and its follow-up will fix again.

Turning it off: revert this commit.

Compatibility: Nothing changes: no save, option, cache or vk ABI is touched.

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). Left to CI: no tests are run on this PC.
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.
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. (The code is main's from before #771, which the nightlies built on both.)
Docs updated.

🤖 Generated with Claude Code

**Type:** revert · **Risk:** low · **Tested on:** none (game-free: the tree is main before #771, fd770e7f's parent, byte for byte) · **Needs:** none ## Why [#771](https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/771) was merged while its review asked for a change. Its idea is right: below 60 fps, the waits before an unused texture is let go should be counted in seconds, not in frames. But the way it measures the frame rate goes wrong right after a load or a long hitch, even at a steady 60 fps. It ages every texture by the frame rate averaged over the last ten seconds. A gap with no frames, such as an area load's stall, makes that average read far too low for the next ten seconds. At 60 fps after a 30 s gap it reads about 2 fps one second in and about 8.6 fps five seconds in, so textures look 7 to 30 times older than they are: - a texture unused for 1.4 s counts as past the ten-second line, and one unused for 4.3 s as past the thirty-second line; - the minute kept for textures the GPU drew drops to as little as 2 s. That happens just after an area loads, when VRAM is tightest. The textures let go are uploaded again, and one the GPU drew can come back empty, which the minute's wait exists to prevent. The review's details: https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/771#issuecomment-26880 This takes it back out so that main and the next release candidate don't ship it. The change comes back in its own pull request, with the age measured per texture in seconds instead of through one average rate. See [#771](https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/771). ## What changed A plain `jj revert` of fd770e7f (#771). Main goes back to counting the compaction's hold and reserve, and the GPU-drawn textures' keep, in presents at any frame rate, as before #771. - `gpu/vk/vk_texture.cpp`: the frame-rate marks and `present_rate` removed; `hold_target` and the keep take the present counts again. - `gpu/vk/vram_compact.h`: `age_at_60` removed. `gpu/vk/keep_rendered.h`: `at_rate` removed. - `docs/development.md`: the `RX_VK_EVICT_BY_PRESENTS` row removed. - `tests/test_keep_rendered.*`, `tests/test_vram_compact.*`: back to main's before #771. ## Measurements Nothing to measure: this restores main's code from before #771 exactly (`jj diff --from fd770e7f- --to` this commit is empty), whose behaviour is what every release so far has shipped. ### Flags and settings `RX_VK_EVICT_BY_PRESENTS` (added by #771) goes away with it: what it switched back to is the only behaviour again. | Flag | What it is | Before this PR | After this PR | Effect | Helps / doesn't help / untested | Recommendation | |---|---|---|---|---|---|---| | `RX_VK_EVICT_BY_PRESENTS` | #771's off switch: the waits in presents | unset: waits by the measured rate below 60 fps | removed: waits in presents always | as before #771 | before #771's behaviour, shipped in every release | removed with #771 | ## Testing **Tested:** game-free only, by reading: the reverted tree is identical to main before #771 (an empty diff against fd770e7f's parent). No tests were run on this PC; the checks are left to CI. **Not tested:** nothing new to test: the code is what main ran before #771 merged, today. **Must be tested:** None. ## Risks **What could break:** nothing that main didn't do before today: below 60 fps, the hold's ten and thirty seconds and the keep's minute stretch again (at 3.5 fps they are three, nine and seventeen minutes), which is what #771 set out to fix and its follow-up will fix again. **Turning it off:** revert this commit. **Compatibility:** Nothing changes: no save, option, cache or vk ABI is touched. ## 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). Left to CI: no tests are run on this PC. - [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] 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. - [x] Host code builds on both Windows (llvm-mingw) and Linux, with UTF-8 paths. (The code is main's from before #771, which the nightlies built on both.) - [x] Docs updated. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

HumanBean added 1 commit

2026-11-15 18:34:32 +00:00

revert(vk): the eviction's waits in presents again at any frame rate, as before #771

CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped

Details

CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped

Details

CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped

Details

Third-party parts / The same as upstream (pull_request) Has been skipped

Details

Website / Website on Cloudflare (pull_request_target) Has been skipped

Details

CI / Conventional Commit title (pull_request) Successful in 4s

Details

CI / Workflow lint (actionlint) (pull_request) Successful in 5s

Details

Guide screenshots / Guide images and local UI source review (pull_request) Failing after 5s

Details

Third-party parts / As their notes say (pull_request) Successful in 8s

Details

PR description / Description follows the template (pull_request_target) Successful in 4s

Details

d8c9a4aa28

This reverts commit fd770e7f5f (#771), merged with a request for changes standing:
its rate, averaged over the last ten seconds, reads far too low for ten seconds after any gap without presents (a
load's stall, a long hitch), so textures are aged several times too fast right after an area loads, even at 60 fps.
The change comes back in a new pull request with that fixed.

Co-Authored-By: Claude Opus 5.5 <[email protected]>

BeanIssues added the

area/renderer

kind

bug

labels

2026-11-15 18:34:35 +00:00

HumanBean referenced this pull request

2026-11-15 18:34:55 +00:00

WIP: perf(vk): below 60 fps the eviction's waits are seconds, not presents: the compaction's hold and the GPU-drawn textures' keep #875

HumanBean referenced this pull request

2026-11-15 18:35:04 +00:00

perf(vk): below 60 fps the eviction's waits are seconds, not presents: the compaction's hold and the GPU-drawn textures' keep #771

HumanBean force-pushed revert/evict-seconds from d8c9a4aa28

CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped

Details

CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped

Details

CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped

Details

Third-party parts / The same as upstream (pull_request) Has been skipped

Details

Website / Website on Cloudflare (pull_request_target) Has been skipped

Details

CI / Conventional Commit title (pull_request) Successful in 4s

Details

CI / Workflow lint (actionlint) (pull_request) Successful in 5s

Details

Guide screenshots / Guide images and local UI source review (pull_request) Failing after 5s

Details

Third-party parts / As their notes say (pull_request) Successful in 8s

Details

PR description / Description follows the template (pull_request_target) Successful in 4s

Details

to e020e46983

CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped

Details

CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped

Details

CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped

Details

Third-party parts / The same as upstream (pull_request) Has been skipped

Details

Website / Website on Cloudflare (pull_request) Has been skipped

Details

CI / Conventional Commit title (pull_request) Successful in 5s

Details

CI / Workflow lint (actionlint) (pull_request) Successful in 12s

Details

Guide screenshots / Guide images and local UI source review (pull_request) Failing after 14s

Details

Third-party parts / As their notes say (pull_request) Successful in 23s

Details

PR description / Description follows the template (pull_request_target) Waiting to run

Details

2026-11-15 18:41:00 +00:00

Compare

HumanBean changed title from ~~WIP: revert(vk): the eviction's waits in presents again at any frame rate, as before #771~~ to revert(vk): the eviction's waits in presents again at any frame rate, as before #771

2026-11-15 18:49:58 +00:00

HumanBean merged commit b9e95173a8 into main

2026-11-15 18:50:08 +00:00

HumanBean deleted branch revert/evict-seconds

2026-11-15 18:50:08 +00:00

HumanBean referenced this pull request from a commit

2026-11-15 18:50:10 +00:00

revert(vk): the eviction's waits in presents again at any frame rate, as before #771 (#873)

HumanBean referenced this pull request

2026-11-15 18:50:48 +00:00

WIP: perf(vk): below 60 fps the eviction's waits are seconds, not presents: the compaction's hold and the GPU-drawn textures' keep #875

HumanBean referenced this pull request