WIP: perf(vk): images the renderer lets go of are destroyed on a thread of their own #792

Draft

dyst wants to merge 2 commits from dyst/Longshot:perf/image-reaper into main

pull from: dyst/Longshot:perf/image-reaper

merge into: MagicBeanDev:main

MagicBeanDev:main

Conversation 5 Commits 2 Files changed 6 +179 -13

dyst commented 2026-11-15 12:10:48 +00:00

Member

• Copy link
• Copy Markdown

Type: perf · Risk: medium · Tested on: Windows 11 (game-free checks) · Needs: Windows, NVIDIA (any RTX), turning the camera in the Market and Hyrule Field; any GPU for the compaction

Why

The game stutters while the camera turns. Part of that is the renderer freeing GPU images on its own thread: when it lets go of textures and render targets, texture_frame destroys each image, its views, buffer and memory right there, and the render-target pool's trim destroys up to 16 more a present. On NVIDIA's Windows driver a vmaDestroyImage is a kernel call of a millisecond and more, and a turn of the camera lets go of many in one present.

Project ReLink, a fork of Longshot, measured 2.4 ms of each frame over 25 ms spent there, against 0.1 ms of a normal frame. Moving the frees to a thread of their own cut its frames over 25 ms and its 99th percentile (31689010, 5224e821).

What changed

The images the renderer lets go of are destroyed on a reaper thread. The textures' compaction rule is kept: nothing is freed while a pass is open.

• vk: destroy_image_later (gpu/vk/vk_vma.cpp) queues an image with its views, buffer and memory objects for a reaper thread, started on first use.

  • Thread safety: VMA is thread-safe as we create it (no externally-synchronized flag, vma_init). The images handed over are ones nothing can use again: their last recorded use has finished.

• vk: the compaction rule.

  • The reaper frees only while no defragmentation pass is open: it waits on compact_in_pass under reap_mu.
  • image_compact_pass takes that lock before vmaBeginDefragmentationPass, and image_compact_pass_end holds it through vmaEndDefragmentationPass.
  • Each free runs outside the lock, marked in flight (reaping) under it, so the render thread's destroy_image_later never waits for the queue; image_compact_pass waits for that one free (reap_done_cv) before vmaBeginDefragmentationPass (5333dd63: the first version freed under the lock and kept it while the queue had more, so a hand-over waited for the whole queue).
  • So a pass waits for the one free in flight (a millisecond or so, only when a pass begins), and no free comes inside a pass. texture_frame still hands nothing over while compact.in_pass, as before.

• vk: texture_frame's dead images and pool_destroy (the render-target pool's trim) use it.

• switch: RX_VK_REAPER=0 destroys them at once on the render thread, as before. The log says vk: images are destroyed on a thread of their own when it's on.

• docs: the switch's row in docs/development.md.

• tests: tests/test_image_reaper.py (new) checks:

  • the hand-over;
  • the lock's order around both VMA pass calls;
  • the reaper's wait;
  • the switch.

  tests/test_target_image_pool.py follows the new calls and still checks pooling before freeing.

Measurements

Not measured here. What to measure, on Windows with an NVIDIA card: frames over 25 ms and the 99th percentile while turning the camera in the Market and on arriving in Hyrule Field, on main and on this PR, at least two interleaved runs each, with tex_frame_ms from the frame trace. ReLink's numbers, for reference:

• frames over 25 ms: 71/81 down to 67/50/66;
• 99th percentile: 36-39 ms down to 34-35 ms.

Flags and settings

One new switch, on by default.

Flag	What it is	Before this PR	After this PR	Effect	Helps / doesn't help / untested	Recommendation
RX_VK_REAPER	Images destroyed on a thread of their own	not there (always on the render thread)	on; 0: on the render thread	Takes the driver's image frees off the render thread	Untested here; helped ReLink on NVIDIA Windows	On for everyone; 0 only to compare or if a driver misbehaves

Testing

Tested:

• Rebased onto main @ 7bacf846 (this PR @ 5333dd63): the 63 checks that read this diff (listed below), not the whole suite, through tests/run_all.py --jobs 6 on Windows 11 with the Builder's llvm-mingw (v2.0.14 tools): 42 passed, 9 failed, 12 skipped for want of the Vulkan headers. Of the 9, test_check_pr_description, test_dev_packs, test_drawn_page_map and test_keep_rendered failed with no output under load and pass when run again; test_crash_dialog, test_oot_state_tap, test_vk_pgo and test_walk_key also fail on main @ 1310ecc6 (the full suite there: 500 passed, 17 failed, 79 skipped), and test_vk_occlusion_queries on main @ 7bacf846. The 12 skipped, again with RX_CPM set to the Builder's CPM cache: 11 pass, and nx2/test_bc_storage_views fails as on main @ 7bacf846. After the fix below (5333dd63), vk_vma.cpp and vk_texture.cpp compile (-fsyntax-only) with the renderer's flags from nx2/build.py and the Builder's CPM headers.

  The 63 checks

  • nx2/test_bc_storage_views.py
  • nx2/test_texture_init_barrier.py
  • tests/test_3d_view_create_flags.py
  • tests/test_amd_subgroup_profiles.py
  • tests/test_astc_bc_policy.py
  • tests/test_bc_raw_targets.py
  • tests/test_bc_upload_retry.py
  • tests/test_blocks_at_start.py
  • tests/test_check_pr_description.py
  • tests/test_commit_limit.py
  • tests/test_commit_room.py
  • tests/test_compressed_over_rt.py
  • tests/test_crash_dialog.py
  • tests/test_dev_packs.py
  • tests/test_direct_buffer_writes.py
  • tests/test_drawn_page_map.py
  • tests/test_drawn_writeback.py
  • tests/test_dump_readback.py
  • tests/test_evict_reason.py
  • tests/test_frame_trace_waits.py
  • tests/test_fsr_uniform_ownership.py
  • tests/test_hold_gpu.py
  • tests/test_image_reaper.py
  • tests/test_image_view_retry.py
  • tests/test_inline_write_order.py
  • tests/test_intel_compiler_gate.py
  • tests/test_intel_compiles_auto.py
  • tests/test_jobs_failure.py
  • tests/test_jobs_wait_runs_queued.py
  • tests/test_keep_rendered.py
  • tests/test_linux_pressure.py
  • tests/test_log_levels.py
  • tests/test_low_vram_texture_mips.py
  • tests/test_mem_trim.py
  • tests/test_mesh_fallback.py
  • tests/test_met_wait_capture.py
  • tests/test_native_astc_texture_count.py
  • tests/test_oot_state_tap.py
  • tests/test_parallel_init_order.py
  • tests/test_parallel_rb_refresh.py
  • tests/test_pipeline_state_portable.py
  • tests/test_prepare_stall.py
  • tests/test_prof_clock.py
  • tests/test_prof_sampling.py
  • tests/test_quick_pipelines.py
  • tests/test_sparse_mirror.py
  • tests/test_split_before_draw.py
  • tests/test_staging_arena_fallback.py
  • tests/test_target_image_pool.py
  • tests/test_target_overlap.py
  • tests/test_texel_buffer_ownership.py
  • tests/test_texture_budget.py
  • tests/test_texture_layout_sizes.py
  • tests/test_texture_squeeze.py
  • tests/test_texture_staging_publication.py
  • tests/test_third_party_objects.py
  • tests/test_threaded_rendering.py
  • tests/test_vk_occlusion_queries.py
  • tests/test_vk_pgo.py
  • tests/test_vram_compact.py
  • tests/test_vulkan_clear_mip.py
  • tests/test_walk_key.py
  • tests/test_wb_shadow_remake.py

• tests/run_all.py image_reaper target_image_pool import_buffer compact texture_budget vram sparse on Windows: 6 passed, 2 skipped (no Vulkan headers for those two).

• vk_vma.cpp and vk_texture.cpp compile with the build's flags (llvm-mingw).

Not tested:

• In the game: frame times, a long session with the compaction running passes, and Linux (Mesa and NVIDIA).

Must be tested:

• Windows, NVIDIA RTX:
  1. Turn the camera in the Market, and arrive in Hyrule Field, on main and on this PR.
  2. Expect fewer frames over 25 ms on this PR.
  3. Play 30 minutes with the compaction on (the default) and expect vk: VRAM compacted lines and no device loss (by hand).
• Linux, any GPU: 15 minutes of play. Expect no change in behaviour and no device loss (by hand).

Risks

What could break:

• A missed rule: a free that VMA doesn't allow while a defragmentation pass is open. The lock is meant to rule that out, and the compaction would misbehave if it didn't.
• VRAM given back later: a backlog of dead images on the reaper gives VRAM back a little later than before.

Turning it off: RX_VK_REAPER=0, or revert the commit.

Compatibility: nothing changes for a player beyond where the frees happen; caches and saves are untouched.

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).
• 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.
• Docs updated.

🤖 Generated with Claude Code

**Type:** perf · **Risk:** medium · **Tested on:** Windows 11 (game-free checks) · **Needs:** Windows, NVIDIA (any RTX), turning the camera in the Market and Hyrule Field; any GPU for the compaction ## Why The game stutters while the camera turns. Part of that is the renderer freeing GPU images on its own thread: when it lets go of textures and render targets, `texture_frame` destroys each image, its views, buffer and memory right there, and the render-target pool's trim destroys up to 16 more a present. On NVIDIA's Windows driver a `vmaDestroyImage` is a kernel call of a millisecond and more, and a turn of the camera lets go of many in one present. Project ReLink, a fork of Longshot, measured 2.4 ms of each frame over 25 ms spent there, against 0.1 ms of a normal frame. Moving the frees to a thread of their own cut its frames over 25 ms and its 99th percentile ([31689010](https://github.com/pixl27/ProjectReLink/commit/31689010), [5224e821](https://github.com/pixl27/ProjectReLink/commit/5224e821)). ## What changed The images the renderer lets go of are destroyed on a reaper thread. The textures' compaction rule is kept: nothing is freed while a pass is open. - vk: `destroy_image_later` (`gpu/vk/vk_vma.cpp`) queues an image with its views, buffer and memory objects for a reaper thread, started on first use. - **Thread safety:** VMA is thread-safe as we create it (no externally-synchronized flag, `vma_init`). The images handed over are ones nothing can use again: their last recorded use has finished. - vk: **the compaction rule.** - The reaper frees only while no defragmentation pass is open: it waits on `compact_in_pass` under `reap_mu`. - `image_compact_pass` takes that lock before `vmaBeginDefragmentationPass`, and `image_compact_pass_end` holds it through `vmaEndDefragmentationPass`. - Each free runs outside the lock, marked in flight (`reaping`) under it, so the render thread's `destroy_image_later` never waits for the queue; `image_compact_pass` waits for that one free (`reap_done_cv`) before `vmaBeginDefragmentationPass` (5333dd63: the first version freed under the lock and kept it while the queue had more, so a hand-over waited for the whole queue). - So a pass waits for the one free in flight (a millisecond or so, only when a pass begins), and no free comes inside a pass. `texture_frame` still hands nothing over while `compact.in_pass`, as before. - vk: `texture_frame`'s dead images and `pool_destroy` (the render-target pool's trim) use it. - switch: `RX_VK_REAPER=0` destroys them at once on the render thread, as before. The log says `vk: images are destroyed on a thread of their own` when it's on. - docs: the switch's row in `docs/development.md`. - tests: `tests/test_image_reaper.py` (new) checks: - the hand-over; - the lock's order around both VMA pass calls; - the reaper's wait; - the switch. `tests/test_target_image_pool.py` follows the new calls and still checks pooling before freeing. ## Measurements Not measured here. What to measure, on Windows with an NVIDIA card: frames over 25 ms and the 99th percentile while turning the camera in the Market and on arriving in Hyrule Field, on main and on this PR, at least two interleaved runs each, with `tex_frame_ms` from the frame trace. ReLink's numbers, for reference: - frames over 25 ms: 71/81 down to 67/50/66; - 99th percentile: 36-39 ms down to 34-35 ms. ### Flags and settings One new switch, on by default. | Flag | What it is | Before this PR | After this PR | Effect | Helps / doesn't help / untested | Recommendation | |---|---|---|---|---|---|---| | `RX_VK_REAPER` | Images destroyed on a thread of their own | not there (always on the render thread) | on; `0`: on the render thread | Takes the driver's image frees off the render thread | Untested here; helped ReLink on NVIDIA Windows | On for everyone; `0` only to compare or if a driver misbehaves | ## Testing **Tested:** - Rebased onto `main` @ 7bacf846 (this PR @ 5333dd63): the 63 checks that read this diff (listed below), not the whole suite, through `tests/run_all.py --jobs 6` on Windows 11 with the Builder's llvm-mingw (v2.0.14 tools): 42 passed, 9 failed, 12 skipped for want of the Vulkan headers. Of the 9, `test_check_pr_description`, `test_dev_packs`, `test_drawn_page_map` and `test_keep_rendered` failed with no output under load and pass when run again; `test_crash_dialog`, `test_oot_state_tap`, `test_vk_pgo` and `test_walk_key` also fail on main @ 1310ecc6 (the full suite there: 500 passed, 17 failed, 79 skipped), and `test_vk_occlusion_queries` on main @ 7bacf846. The 12 skipped, again with `RX_CPM` set to the Builder's CPM cache: 11 pass, and `nx2/test_bc_storage_views` fails as on main @ 7bacf846. After the fix below (5333dd63), `vk_vma.cpp` and `vk_texture.cpp` compile (`-fsyntax-only`) with the renderer's flags from `nx2/build.py` and the Builder's CPM headers. <details><summary>The 63 checks</summary> - `nx2/test_bc_storage_views.py` - `nx2/test_texture_init_barrier.py` - `tests/test_3d_view_create_flags.py` - `tests/test_amd_subgroup_profiles.py` - `tests/test_astc_bc_policy.py` - `tests/test_bc_raw_targets.py` - `tests/test_bc_upload_retry.py` - `tests/test_blocks_at_start.py` - `tests/test_check_pr_description.py` - `tests/test_commit_limit.py` - `tests/test_commit_room.py` - `tests/test_compressed_over_rt.py` - `tests/test_crash_dialog.py` - `tests/test_dev_packs.py` - `tests/test_direct_buffer_writes.py` - `tests/test_drawn_page_map.py` - `tests/test_drawn_writeback.py` - `tests/test_dump_readback.py` - `tests/test_evict_reason.py` - `tests/test_frame_trace_waits.py` - `tests/test_fsr_uniform_ownership.py` - `tests/test_hold_gpu.py` - `tests/test_image_reaper.py` - `tests/test_image_view_retry.py` - `tests/test_inline_write_order.py` - `tests/test_intel_compiler_gate.py` - `tests/test_intel_compiles_auto.py` - `tests/test_jobs_failure.py` - `tests/test_jobs_wait_runs_queued.py` - `tests/test_keep_rendered.py` - `tests/test_linux_pressure.py` - `tests/test_log_levels.py` - `tests/test_low_vram_texture_mips.py` - `tests/test_mem_trim.py` - `tests/test_mesh_fallback.py` - `tests/test_met_wait_capture.py` - `tests/test_native_astc_texture_count.py` - `tests/test_oot_state_tap.py` - `tests/test_parallel_init_order.py` - `tests/test_parallel_rb_refresh.py` - `tests/test_pipeline_state_portable.py` - `tests/test_prepare_stall.py` - `tests/test_prof_clock.py` - `tests/test_prof_sampling.py` - `tests/test_quick_pipelines.py` - `tests/test_sparse_mirror.py` - `tests/test_split_before_draw.py` - `tests/test_staging_arena_fallback.py` - `tests/test_target_image_pool.py` - `tests/test_target_overlap.py` - `tests/test_texel_buffer_ownership.py` - `tests/test_texture_budget.py` - `tests/test_texture_layout_sizes.py` - `tests/test_texture_squeeze.py` - `tests/test_texture_staging_publication.py` - `tests/test_third_party_objects.py` - `tests/test_threaded_rendering.py` - `tests/test_vk_occlusion_queries.py` - `tests/test_vk_pgo.py` - `tests/test_vram_compact.py` - `tests/test_vulkan_clear_mip.py` - `tests/test_walk_key.py` - `tests/test_wb_shadow_remake.py` </details> - `tests/run_all.py image_reaper target_image_pool import_buffer compact texture_budget vram sparse` on Windows: 6 passed, 2 skipped (no Vulkan headers for those two). - `vk_vma.cpp` and `vk_texture.cpp` compile with the build's flags (llvm-mingw). **Not tested:** - **In the game:** frame times, a long session with the compaction running passes, and Linux (Mesa and NVIDIA). **Must be tested:** - **Windows, NVIDIA RTX:** 1. Turn the camera in the Market, and arrive in Hyrule Field, on main and on this PR. 2. Expect fewer frames over 25 ms on this PR. 3. Play 30 minutes with the compaction on (the default) and expect `vk: VRAM compacted` lines and no device loss (by hand). - **Linux, any GPU:** 15 minutes of play. Expect no change in behaviour and no device loss (by hand). ## Risks **What could break:** - **A missed rule:** a free that VMA doesn't allow while a defragmentation pass is open. The lock is meant to rule that out, and the compaction would misbehave if it didn't. - **VRAM given back later:** a backlog of dead images on the reaper gives VRAM back a little later than before. **Turning it off:** `RX_VK_REAPER=0`, or revert the commit. **Compatibility:** nothing changes for a player beyond where the frees happen; caches and saves are untouched. ## 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`. - [x] `python -I -B tests/run_all.py` passes (any failure that also fails on `main` is named). - [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] Docs updated. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

dyst added 1 commit 2026-11-15 12:10:48 +00:00

perf(vk): images the renderer lets go of are destroyed on a thread of their own ... 8d361dc63f

texture_frame destroyed the dead images, their views, buffers and memory on the render thread, and the
render-target pool's trim its images too, 16 a present. On NVIDIA's Windows driver a vmaDestroyImage is a
kernel call of a millisecond and more, and a turn of the camera lets go of many in one present: Project
ReLink measured 2.4 ms of each frame over 25 ms there against 0.1 ms of the others, and moving them to a
thread cut its frames over 25 ms and its 99th percentile (31689010, 5224e821).

destroy_image_later (vk_vma.cpp) hands them to a reaper thread. The textures' compaction frees nothing
while a pass is open: the reaper waits on the pass flag under a lock that image_compact_pass takes before
vmaBeginDefragmentationPass and image_compact_pass_end holds through vmaEndDefragmentationPass, so a pass
waits for the one free in flight and no free comes inside it. RX_VK_REAPER=0 destroys them at once.

Co-Authored-By: Claude Opus 5.5 (1M context) <[email protected]>

mooki added the

area/renderer

kind

performance

platform/linux

platform/windows

priority

high

labels 2026-11-15 14:00:09 +00:00

futondestroyer commented 2026-11-15 16:00:14 +00:00

Member

• Copy link
• Copy Markdown

@Beanstalk test Windows with an NVIDIA RTX GPU at head 8d361dc63f. This PC is Windows with an RTX 4070 and the patched Beanstalk agent dev-9585e4a (isolated_gameplay support).

Compare main and this PR with identical settings, frame generation off and no frame-rate cap, using locally created disposable new-game saves. In the Market, measure 20 seconds standing and 20 seconds while turning the camera at full right-stick speed; also measure Hyrule Field arrivals. Use at least two interleaved runs per side, report median frame time, p99, counts of frames over 25 ms if the farm supports them, and VRAM. Record crashes, device loss and invalid-renderer errors. Include comparison screenshots where useful.

Also exercise the default compaction for 30 minutes on the PR, checking for vk: VRAM compacted lines and no device loss. Explicitly state if the farm cannot represent concurrent camera motion and measurement, the long-run duration, the tex_frame_ms trace, or any requested metric; measurements taken after the camera stops must not be described as camera-turn measurements. A short smoke test or skipped scenario is not full performance/lifetime coverage.

Please keep all game files, generated builds, caches and saves local; only normal farm reports, scrubbed logs and requested game screenshots may be uploaded.

@Beanstalk test Windows with an NVIDIA RTX GPU at head 8d361dc63f2a591fa28bfe43579d3f0f36d9234d. This PC is Windows with an RTX 4070 and the patched Beanstalk agent dev-9585e4a (isolated_gameplay support). Compare main and this PR with identical settings, frame generation off and no frame-rate cap, using locally created disposable new-game saves. In the Market, measure 20 seconds standing and 20 seconds while turning the camera at full right-stick speed; also measure Hyrule Field arrivals. Use at least two interleaved runs per side, report median frame time, p99, counts of frames over 25 ms if the farm supports them, and VRAM. Record crashes, device loss and invalid-renderer errors. Include comparison screenshots where useful. Also exercise the default compaction for 30 minutes on the PR, checking for vk: VRAM compacted lines and no device loss. Explicitly state if the farm cannot represent concurrent camera motion and measurement, the long-run duration, the tex_frame_ms trace, or any requested metric; measurements taken after the camera stops must not be described as camera-turn measurements. A short smoke test or skipped scenario is not full performance/lifetime coverage. Please keep all game files, generated builds, caches and saves local; only normal farm reports, scrubbed logs and requested game screenshots may be uploaded.

👀 1

Beanstalk commented 2026-11-15 16:00:16 +00:00

Member

• Copy link
• Copy Markdown

Beanstalk test farm

Commit 5333dd6 · 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 `5333dd6` · tier **untrusted** · approved by nobody yet · [dashboard](https://beanstalk.magicbean.dev/prs/792) A member of `BigBeans` can review this commit and comment `@Beanstalk` to test it.

Beanstalk added the

beanstalk: approved @ 8d361dc

label 2026-11-15 16:00:29 +00:00

Rogasse referenced this pull request 2026-11-15 17:28:44 +00:00

Game: 13-25 real FPS in Castle Town Market on Nightly 189 with RTX 4050 6 GB #813

futondestroyer commented 2026-11-15 18:08:17 +00:00

Member

• Copy link
• Copy Markdown

@Beanstalk cancel

@Beanstalk cancel

👀 1

Beanstalk commented 2026-11-15 18:08:19 +00:00

Member

• Copy link
• Copy Markdown

@futondestroyer cancelled 1 queued job.

@futondestroyer cancelled 1 queued job.

dyst force-pushed perf/image-reaper from 8d361dc63f to 5333dd636f 2026-11-15 19:04:21 +00:00 Compare

Beanstalk removed the

beanstalk: approved @ 8d361dc

label 2026-11-15 19:04:24 +00:00

dyst commented 2026-11-15 19:20:58 +00:00

Author

Member

• Copy link
• Copy Markdown

Rebased onto main @ 7bacf846. Fixed in 5333dd63: the reaper freed each image under reap_mu and, with more queued, never let go of it between frees, so a destroy_image_later on the render thread waited for the whole queue (and a beginning pass for all of it, not the one in flight). Each free now runs outside the lock, marked in flight; image_compact_pass waits for that one before vmaBeginDefragmentationPass, and compact_in_pass, set under the same lock, still keeps every free out of a pass. tests/test_image_reaper.py checks the order. The 63 checks that read this diff: no failure main doesn't have (4 under load passed when run again). Still WIP: Measurements needs its main/this PR table (the Test Farm can't run the camera turn on Windows).

Rebased onto main @ 7bacf846. Fixed in 5333dd63: the reaper freed each image under reap_mu and, with more queued, never let go of it between frees, so a destroy_image_later on the render thread waited for the whole queue (and a beginning pass for all of it, not the one in flight). Each free now runs outside the lock, marked in flight; image_compact_pass waits for that one before vmaBeginDefragmentationPass, and compact_in_pass, set under the same lock, still keeps every free out of a pass. tests/test_image_reaper.py checks the order. The 63 checks that read this diff: no failure main doesn't have (4 under load passed when run again). Still WIP: Measurements needs its main/this PR table (the Test Farm can't run the camera turn on Windows).

Rogasse referenced this pull request 2026-11-15 19:43:49 +00:00

Game: 13-25 real FPS in Castle Town Market on Nightly 189 with RTX 4050 6 GB #813

BeanIssues added the

risk

medium

label 2026-11-15 20:27:37 +00:00

Some checks failed

Hide checks

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) Has been skipped

Details

CI / Conventional Commit title (pull_request) Successful in 3s

Details

CI / Workflow lint (actionlint) (pull_request) Successful in 6s

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 11s

Details

PR description / Description follows the template (pull_request_target) Successful in 5s

Details

Website / Website on Cloudflare (pull_request_target) Has been skipped

Details

beanstalk/approval not approved yet

Details

This pull request is marked as a work in progress.

This branch is out-of-date with the base branch

View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.

git fetch -u 'perf/image-reaper:dyst-perf/image-reaper'

git switch 'dyst-perf/image-reaper'

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels   

area/accessibility

Barriers for people with disabilities

  

area/audio

Sound and music

  

area/input

Controllers, keyboard, mouse

  

area/install

Installing, updating, building from your game

  

area/menu

In-game menu and settings

  

area/renderer

Vulkan renderer, shaders, effects

  

area/saves

Save files

  

area/tooling

Developer tools: scripts, CI, build and release tooling

  

area/upscaling

DLSS, FSR, frame generation

  

beanstalk: approved @ 1697603

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 2c5a62e

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 4fd82b7

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 64d6d09

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 6f91fae

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 70a78f9

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 8015c9a

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 8d361dc

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ 9d8356e

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ abf2227

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ d82f202

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ e5f1e40

Set by Beanstalk for show; the farm's database is what counts

  

beanstalk: approved @ ee8ab5f

Set by Beanstalk for show; the farm's database is what counts

  

ci: approved

A maintainer approved CircleCI for this fork PR (sticky; removed after 14 quiet days)

  

good first issue

A good start for a new contributor

  

gpu/amd

AMD GPUs

  

gpu/igpu

Integrated GPUs

  

gpu/intel

Intel GPUs

  

gpu/nvidia

NVIDIA GPUs

  

help wanted

Extra hands or knowledge wanted

  

kind

bug

Something doesn't work as it should

  

kind

chore

Build, CI, dependencies, tidying

  

kind

docs

Documentation

  

kind

feature

A new feature or an improvement

  

kind

performance

Frame rate, frame time, loading, memory

  

kind

question

A question rather than a change

  

needs measurements

Waiting for before/after measurements

  

platform/linux

Affects Linux (a change for every system carries both Windows and Linux)

  

platform/macos

macOS (Apple silicon or Intel Macs)

  

platform/steam-deck

Steam Deck

  

platform/windows

Affects Windows (a change for every system carries both Windows and Linux)

  

preview

Maintainers: deploy a preview of this PR's website (read its changes first). Removed after the run.

  

priority

critical

Crashes, data loss, or blocks most players

  

priority

high

Important: next in line

  

priority

low

Nice to have

  

regression

Worked in an earlier build, broken in a newer one: highest priority, above other bugs

  

risk

high

Broad or core change: the renderer, defaults on many systems, saves, installs, security

  

risk

low

Small, contained change: little can break

  

risk

medium

Changes behaviour in one area, or a default on some systems

  

status

blocked

Waits on something else first

  

status

confirmed

Reproduced or agreed on

  

status

duplicate

Already reported or proposed

  

status

invalid

Not a problem in this project

  

status

needs-info

Waiting for more information from the reporter

  

status

wontfix

Will not be worked on

  

testing wanted

Wanted: a test on other hardware (posts a test request in #test-requests)

  

ci: approved

No labels

area/accessibility

area/audio

area/input

area/install

area/menu

area/renderer

area/saves

area/tooling

area/upscaling

beanstalk: approved @ 1697603

beanstalk: approved @ 2c5a62e

beanstalk: approved @ 4fd82b7

beanstalk: approved @ 64d6d09

beanstalk: approved @ 6f91fae

beanstalk: approved @ 70a78f9

beanstalk: approved @ 8015c9a

beanstalk: approved @ 8d361dc

beanstalk: approved @ 9d8356e

beanstalk: approved @ abf2227

beanstalk: approved @ d82f202

beanstalk: approved @ e5f1e40

beanstalk: approved @ ee8ab5f

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

testing wanted

ci: approved

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

3 participants

Notifications

Subscribe

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!792

No description provided.