perf(vk): the recorder takes a whole part and submits it, so the render thread never waits at a part's end #1866
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
2 participants
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
MagicBeanDev/Longshot!1866
Loading…
Reference in a new issue
No description provided.
Delete branch "HumanBean/Longshot:perf/recorder-parts"
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?
Summary
With the recorder thread on (Threaded rendering Auto from 6 cores, merged in #1763), the render thread packed every Vulkan call into a ring and woke the recorder per call; at each part's end it then waited until the recorder had made every call, and submitted the part itself. So the two threads overlapped only inside a part, stopped to sync several times a frame, and the recorder woke ~1,100 times a frame (9 packets each) and spun 4 s in every 10 between them: the churn that once let an i7-9700K's clocks drop and cost a Ryzen 9 7900X 3-11% when spinning.
Now the render thread hands the recorder a whole part, and the recorder ends and submits it: the render thread decides the part's signal values up front and goes straight on to the next part. Anything that needs a part to be on the queue (slot reuse, fence and timeline waits, readbacks, the queue lock's holders, DLSS/FSR/XeSS/menu recording of their own) waits for a "submitted up to part N" counter, bounded; everything else waits for nothing. Threads waiting for the recorder sleep instead of spinning.
RX_VK_RECORDER_PARTS=0is the old hand-off, in the same build.Read first:
docs/dev/reference/threads.md"The recorder / The hand-off by part";post_part,part_submitandqueue_flush_for(gpu/vk/vk_replay.cpp);Recorder::loop,wait_until,settle_partsandQueueMutex(gpu/vk/vk_recorder.h,gpu/vk/vk_internal.h).Two independent code reviews found and this fixes: a permanent hang with batched submits on RADV (a held last part with a waiter), unbounded new waits, a too-wide lock escape, a poll that defeated RADV batching, and spinning waits. Their last pass found no hang, deadlock or out-of-order submit left.
Testing
Tested: Linux, RTX 3090 (NVIDIA 620.x), Ryzen with 24 threads, 2560x1440: Castle Town Market standing, walking and turning, two game starts with the hand-off on and two with
RX_VK_RECORDER_PARTS=0, plus a warm start of each; no Vulkan errors, no hang. The recorder's stats line: 28.5 hand-offs a frame (335 packets each) against ~1,100 (9 each); spun idle 0.16 s against 4.15 s in 10 s; blocked on the queue's lock 0.03 s in 10 s (longest 1.3 ms); the render thread waited for it 0.02 ms a frame. Renderer builds on Linux; Windows syntax check of the changed files with llvm-mingw. New game-free testtests/recorder_parts_test.cppwith 8 controls (compiled, not run here).Test Farm on
f08644e3against main, three areas, standing (first results; Intel, the Steam Deck and the integrated GPUs pending): no hang and no crash on any of them. Market: RTX 3080 Ti (Ryzen 9 5900X) +3.2%, RX 9060 XT (Ryzen 5 5600, Windows) +4%, RX 9070 XT (Ryzen 7 5800X3D, Windows) +0.7%, RTX 4070 (5800X3D) +0.5%, GTX 1070 (i7-8750H laptop) -2.8%, Steam Machine (Navi 33, AMD custom 6-core) -2.7% (Hyrule Field -7.4%). The two slower ones are power-limited 6-core CPUs: if they hold up, the recorder's Auto rule should leave such CPUs on the render thread alone (by measured render-thread load rather than core count), a follow-up.Still to test:
mise run test,test_recorder_parts.py,test_recorder_sync.py,test_recorder_own_idle.pyand the fixtures it touched, expecting them to pass (by hand: CI on this pull request)Measurements
Verdict: a small, consistent gain on an RTX 3090 (+1.8% Market standing, better p99), and the recorder's wake-ups and spinning nearly gone; AMD numbers come from the Test Farm.
Baseline: this PR's own build with
RX_VK_RECORDER_PARTS=0, which is main's hand-off, so the runs differ only by this change.Setup: Linux, RTX 3090, 2560x1440, Threaded rendering Auto (render thread + recorder), Castle Town Market; two interleaved game starts each, after a warm start of each.
f08644e3withRX_VK_RECORDER_PARTS=0f08644e3RX_VK_RECORDER_PARTSships on: never slower here, and it removes the recorder's churn on every CPU that runs it.Risks
A wrong wait would show as a hang or a read of a part not yet on the GPU: every new wait is bounded and the reviews checked the lock order;
RX_VK_RECORDER_PARTS=0restores the old hand-off. Threads that take the queue lock (texture uploads) now first wait for the parts handed over before them. Nothing changes for the player's files, and nothing changes where the recorder is off (below 6 cores, the Deck).Checklist
Ready for review
main; the title is a Conventional Commit.mise run testpasses (any failure that also fails onmainis named).main, or Measurements says why there is nothing to measure.third_party/untouched; nothing GPL-incompatible.When it applies
gpu/vk/switches.def.tests/that reads changed source is updated, not weakened.Assisted-by:/Co-Authored-By:trailers.Ready to merge
main, measured again or a comment says why the numbers still hold.🤖 Generated with Claude Code
https://claude.ai/code/session_017vw8e3aEPBsd7QfUkaatjP
Tested in game: Linux, RTX 3090 (NVIDIA 620.x), 2560x1440: Castle Town Market standing/walking/turning, two starts each with RX_VK_RECORDER_PARTS on and =0 plus warm starts; no Vulkan errors or hangs; recorder 28.5 hand-offs a frame vs ~1,100, spun idle 0.16 s vs 4.15 s per 10 s.
@Beanstalk Please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the
kokiri-startsave: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, two game starts a side; make sure each side has reached the area before measuring it. Measure fps, p99 frame time, stutters, crashes and hangs (a hang is the main risk: this changes which thread submits the GPU's work), and compare pictures with main. This only acts where the recorder thread runs (6+ cores); on the Steam Deck and 4-core PCs it should be even.Beanstalk test farm · an earlier run: commit
f08644e(the newer results are in a comment below)Commit
f08644e· approved by HumanBean (reviewed) · dashboardStarted at 11:50 UTC.
Estimate: done in about 2 h 15 min, from the queue and the testers online now; some jobs can't start yet.
⚠ No tester has a PC for
linux+intel(none seen in 14 days): it waits until one signs in.windows+amdwindows+nvidiawindows+intellinux+amdlinux+nvidialinux+intelsteamdeckamd+igpuWhy it failed
windows+amd:three-areaskokiri-after-turn.fps_mean.min_delta_pct:115.4 → 91.8 (−20.5%); limit 10.0windows+amd:three-areasscreenshot.max_diff:0.7 → 0.7; limit 0.1windows+amd:three-areasfield-after-turn.stutter_count.max_delta_pct:0.0 → 0.5; limit 20.0; the baseline is 0windows+amd:three-areasmarket-standing.stutter_count.max_delta_pct:0.0 → 1.5; limit 20.0; the baseline is 0windows+amd:three-areaskokiri-after-turn.frametime_p99_ms.max_delta_pct:11.1 → 13.7 (+23.4%); limit 15.0windows+amd:three-areasmarket-after-turn.stutter_count.max_delta_pct:0.0 → 0.5; limit 20.0; the baseline is 0windows+nvidia:three-areaskokiri-after-turn.fps_mean.min_delta_pct:109.7 → 98.6 (−10.1%); limit 10.0windows+nvidia:three-areasmarket-after-turn.stutter_count.max_delta_pct:0.0 → 1.0; limit 20.0; the baseline is 0Couldn't be judged: the tester's PC could not finish the test. The farm tries another PC, or the same one after a fix; it says nothing about the PR on its own.
windows+nvidia, report 1:the main base failed (exit 1): build.py stopped (exit status 1): fatal error: too many errors emitted, stopping now [-ferror-limit=]What was tested
What will be tested · AI plan (
gpt-6.1-terra)Two starts per side compare the PR with main across the requested Windows and Linux GPU targets, Steam Deck, and AMD integrated graphics. Each start visits Kokiri Forest, Hyrule Field, and Market with V-Sync off and the frame rate uncapped; standing performance and pictures are compared, followed by camera turns and a second measurement. The farm cannot turn the camera concurrently with a measurement, so the 15-second measurements follow the turns rather than measuring the turns themselves. Crashes, failure to reach an area or present a frame, substantial FPS or p99 regressions, increased stutters, and picture differences fail the plan. Each target takes roughly 25 minutes; targets without a matching online PC will wait for one. No tester online now matches linux+intel: that target waits for one.
kokiri-start(settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure kokiri-standing for 20 s (after 5 s warm-up), take the picture kokiri to compare with main, move the stick, measure kokiri-after-turn for 15 s (after 0 s warm-up), teleport to Hyrule Field, measure field-standing for 20 s (after 5 s warm-up), take the picture field to compare with main, measure field-after-turn for 15 s (after 0 s warm-up), teleport to Market, measure market-standing for 20 s (after 5 s warm-up), take the picture market to compare with main, measure market-after-turn for 15 s (after 0 s warm-up); 2 runs on each side.Measurements on `windows+amd`, for the PR description
Verdict: fail on windows+amd (2/1 report):
three-areas kokiri-after-turn.fps_mean.min_delta_pct (-20.5%);three-areas screenshot.max_diff;three-areas field-after-turn.stutter_count.max_delta_pctfailed.Hardware: 1: windows+amd+amd-proprietary+dgpu
AMD Radeon RX 9070 XT, 2.0.416AMD Ryzen 7 5800X3D 8-Core Processor, 32 GB RAM; 2: windows+amd+amd-proprietary+dgpuAMD Radeon RX 9060 XT, 2.0.416AMD Ryzen 5 5600 6-Core Processor, 32 GB RAMScene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (
@Beanstalkon this PR runs it again)89c3d21f08644eFlags and settings
None: the plan varied no setting, switch or build option.
As Markdown for the PR description:
Measurements on `windows+nvidia`, for the PR description
Verdict: fail on windows+nvidia (2/1 report):
three-areas kokiri-after-turn.fps_mean.min_delta_pct (-10.1%);three-areas market-after-turn.stutter_count.max_delta_pctfailed.Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu
NVIDIA GeForce RTX 4080, 621.1.0AMD Ryzen 7 9800X3D 8-Core Processor, 32 GB RAM; 2: windows+nvidia+nvidia-proprietary+dgpuNVIDIA GeForce RTX 3080 Ti, 621.24.0AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAMScene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (
@Beanstalkon this PR runs it again)89c3d21f08644eFlags and settings
None: the plan varied no setting, switch or build option.
As Markdown for the PR description:
Measurements on `linux+amd`, for the PR description
Verdict: fail on linux+amd (1/1 report):
three-areas kokiri-after-turn.fps_mean.min_delta_pct (-10.2%);three-areas field-after-turn.fps_mean.min_delta_pct (-13.8%);three-areas market-after-turn.fps_mean.min_delta_pct (-36.0%)failed.Hardware: 1: linux+amd+radv+dgpu
AMD Steam Machine (RADV NAVI33), 26.1.99AMD Custom CPU 1772, 16 GB RAMScene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (
@Beanstalkon this PR runs it again)89c3d21f08644eFlags and settings
None: the plan varied no setting, switch or build option.
As Markdown for the PR description:
Measurements on `linux+nvidia`, for the PR description
Verdict: fail on linux+nvidia (2/1 report):
three-areas kokiri-standing.stutter_count.max_delta_pct (+325.0%);three-areas field-standing.stutter_count.max_delta_pct (+400.0%);three-areas field-after-turn.stutter_count.max_delta_pct (+200.0%)failed.Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu
NVIDIA GeForce RTX 4070, 615.71.9AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAM; 2: linux+nvidia+nvidia-proprietary+dgpuNVIDIA GeForce GTX 1070, 580.196.7Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz, 32 GB RAMScene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (
@Beanstalkon this PR runs it again)89c3d21f08644eFlags and settings
None: the plan varied no setting, switch or build option.
As Markdown for the PR description:
Measurements on `steamdeck`, for the PR description
Verdict: fail on steamdeck (1/1 report):
three-areas market-standing.stutter_count.max_delta_pct (+50.0%);three-areas market-after-turn.stutter_count.max_delta_pct (+88.9%);three-areas field-after-turn.frametime_p99_ms.max_delta_pct (+19.4%)failed.Hardware: 1: linux+amd+radv+igpu+handheld
AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99AMD Custom APU 0405, 12 GB RAMScene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (
@Beanstalkon this PR runs it again)89c3d21f08644eFlags and settings
None: the plan varied no setting, switch or build option.
As Markdown for the PR description:
Measurements on `amd+igpu`, for the PR description
Verdict: fail on amd+igpu (1/1 report):
three-areas screenshot.max_diff;three-areas kokiri-standing.stutter_count.max_delta_pctfailed.Hardware: 1: linux+amd+radv+igpu+handheld
AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99AMD Custom APU 0405, 12 GB RAMScene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 2 game starts a side, kokiri-standing 20 s after 5 s warm-up, kokiri-after-turn 15 s after 0 s warm-up, field-standing 20 s after 5 s warm-up, field-after-turn 15 s after 0 s warm-up, market-standing 20 s after 5 s warm-up, market-after-turn 15 s after 0 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (
@Beanstalkon this PR runs it again)89c3d21f08644eFlags and settings
None: the plan varied no setting, switch or build option.
As Markdown for the PR description:
The testers' hardware
windows+amdAMD Radeon RX 9070 XT, 2.0.416AMD Ryzen 7 5800X3D 8-Core Processor, 32 GB RAMwindows+amdAMD Radeon RX 9060 XT, 2.0.416AMD Ryzen 5 5600 6-Core Processor, 32 GB RAMwindows+nvidiaNVIDIA GeForce RTX 4080, 621.1.0AMD Ryzen 7 9800X3D 8-Core Processor, 32 GB RAMwindows+nvidiaNVIDIA GeForce RTX 3080 Ti, 621.24.0AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAMlinux+amdAMD Steam Machine (RADV NAVI33), 26.1.99AMD Custom CPU 1772, 16 GB RAMlinux+nvidiaNVIDIA GeForce RTX 4070, 615.71.9AMD Ryzen 7 5800X3D 8-Core Processor, 64 GB RAMlinux+nvidiaNVIDIA GeForce GTX 1070, 580.196.7Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz, 32 GB RAMsteamdeckAMD Custom GPU 0405 (RADV VANGOGH), 26.1.99AMD Custom APU 0405, 12 GB RAMamd+igpuAMD Custom GPU 0405 (RADV VANGOGH), 26.1.99AMD Custom APU 0405, 12 GB RAM@Beanstalk Please test this PR against main on every platform: windows+amd, windows+nvidia, windows+intel, linux+amd, linux+nvidia, linux+intel, the steamdeck, amd+igpu. Settings vsync=2 (Off) and frame_rate_cap=6 (Uncapped), Threaded rendering Auto. From the
kokiri-startsave: Kokiri Forest, Hyrule Field and Castle Town Market, in each 20 s standing then 15 s turning the camera, three game starts a side; make sure each side has reached the area before measuring it. Measure fps, p99 frame time, stutters, crashes and hangs (a hang is the main risk: this changes which thread submits the GPU's work), and compare pictures with main. This only acts where the recorder thread runs (6+ cores); on the Steam Deck and 4-core PCs it should be even.Beanstalk test farm · 🔄 testing (4 of 8 reports so far, a failure already)
Commit
f08644e· approved by HumanBean (reviewed) · dashboardDone (started 16:21 UTC, last report 17:11 UTC). Run 2: re-planned by HumanBean. Comment
@Beanstalkto run it again.windows+amdwindows+nvidiawindows+intellinux+amdlinux+nvidialinux+intelsteamdeckamd+igpuWhy it failed
windows+amd:three-areasforest-standing.fps_mean.min_delta_pct:126.6 → 114.5 (−9.6%); limit 5.0windows+amd:three-areasscreenshot.max_diff:0.7 → 0.9; limit 0.1windows+amd:three-areasforest-standing.frametime_p99_ms.max_delta_pct:9.9 → 11.3 (+14.9%); limit 10.0linux+nvidia:three-areasforest-standing.frametime_p99_ms.max_delta_pct:20.3 → 22.6 (+11.2%); limit 10.0Couldn't be judged: the tester's PC could not finish the test. The farm tries another PC, or the same one after a fix; it says nothing about the PR on its own.
windows+nvidia, report 1:the main base failed (exit 1): build.py stopped (exit status 1): fatal error: too many errors emitted, stopping now [-ferror-limit=]linux+nvidia, report 2:Insufficient frame-time evidence for 6 required comparison(s). See rule results. a step ran out of time: timeout: the save did not load into play within 900 sWhat was tested
What will be tested · AI plan (
gpt-6.1-terra)Against main, three starts per side will visit Kokiri Forest, Hyrule Field and Market with V-Sync off and the frame-rate cap uncapped; threaded rendering remains at its default Auto. Each area gets a 20-second standing measurement and a camera-turn exercise with comparison pictures; the farm cannot measure while issuing the turn input. The plan measures FPS, p99 frame time and stutters, and fails on crashes, area-arrival timeouts, substantial performance regressions or picture differences. Each target takes roughly 35 minutes; the Windows and Linux Intel targets will wait for matching testers to come online. No tester online now matches windows+intel, linux+intel: those targets wait for one.
kokiri-start(settings vsync=2, frame_rate_cap=6), then teleport to Kokiri Forest, measure forest-standing for 20 s (after 5 s warm-up), take the picture forest-standing to compare with main, move the stick, take the picture forest-turned to compare with main, teleport to Hyrule Field, measure field-standing for 20 s (after 5 s warm-up), take the picture field-standing to compare with main, take the picture field-turned to compare with main, teleport to Market, measure market-standing for 20 s (after 5 s warm-up), take the picture market-standing to compare with main, take the picture market-turned to compare with main; 3 runs on each side.Measurements on `windows+amd`, for the PR description
Verdict: fail on windows+amd (1/1 report):
three-areas forest-standing.fps_mean.min_delta_pct (-9.6%);three-areas screenshot.max_diff;three-areas forest-standing.frametime_p99_ms.max_delta_pct (+14.9%)failed.Hardware: 1: windows+amd+amd-proprietary+dgpu
AMD Radeon RX 9070 XT, 2.0.416AMD Ryzen 7 5800X3D 8-Core Processor, 32 GB RAMScene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 3 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (
@Beanstalkon this PR runs it again)89c3d21f08644eFlags and settings
None: the plan varied no setting, switch or build option.
As Markdown for the PR description:
Measurements on `windows+nvidia`, for the PR description
Verdict: could not be judged on windows+nvidia (1/1 report).
Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu
NVIDIA GeForce RTX 4080, 621.1.0AMD Ryzen 7 9800X3D 8-Core Processor, 32 GB RAMScene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 3 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 1 report
How to reproduce: the farm's plan above (
@Beanstalkon this PR runs it again)Nothing was measured: no metric came back.
Flags and settings
None: the plan varied no setting, switch or build option.
As Markdown for the PR description:
Measurements on `linux+nvidia`, for the PR description
Verdict: fail on linux+nvidia (2/1 report):
three-areas forest-standing.frametime_p99_ms.max_delta_pct (+11.2%)failed.Hardware: 1: linux+nvidia+nvidia-proprietary+dgpu
NVIDIA GeForce GTX 1070, 580.196.7Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz, 32 GB RAM; 2: linux+nvidia+nvidia-proprietary+dgpuNVIDIA GeForce RTX 5070, 620.70.6AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAMScene: three-areas, from the repository's save kokiri-start (vsync=2, frame_rate_cap=6); each cell's settings and switches as the table names them
Method: the Beanstalk test farm: main and this PR built and run on the same PC in one session, game starts alternating (ABBA); three-areas: 3 game starts a side, forest-standing 20 s after 5 s warm-up, field-standing 20 s after 5 s warm-up, market-standing 20 s after 5 s warm-up; each number the median of the runs, then of 2 reports
How to reproduce: the farm's plan above (
@Beanstalkon this PR runs it again)89c3d21f08644eFlags and settings
None: the plan varied no setting, switch or build option.
As Markdown for the PR description:
The testers' hardware
windows+amdAMD Radeon RX 9070 XT, 2.0.416AMD Ryzen 7 5800X3D 8-Core Processor, 32 GB RAMwindows+nvidiaNVIDIA GeForce RTX 4080, 621.1.0AMD Ryzen 7 9800X3D 8-Core Processor, 32 GB RAMlinux+nvidiaNVIDIA GeForce GTX 1070, 580.196.7Intel(R) Core(TM) i7-8750H CPU @ 2.20GHz, 32 GB RAMlinux+nvidiaNVIDIA GeForce RTX 5070, 620.70.6AMD Ryzen 5 9600X 6-Core Processor, 32 GB RAMLab run (Linux, RX 580 RADV 26.2.6, i7-9700K 8 cores, 1080p, GPU-bound; this PR @
f08644e3, two interleaved game starts each,RX_VK_RECORDER_PARTS=0as the baseline in the same build):RX_VK_RECORDER_PARTS=0Even to slightly better in fps on a GPU-bound AMD desktop, with 10-16% less CPU in use across the scenes; no hang, no Vulkan error.
On the Test Farm's two slower results: the Steam Machine's Hyrule Field -7.4% is one run with a burst of 44 hitches (p99 69 ms, 44.0 fps) while its second run matched main (49.7 against 50.3 and 48.6); its render thread is idle 22 of 35 ms a frame there. The GTX 1070's -2.8% is a single run of the same size. A second farm run with three starts a side is requested.
WIP: perf(vk): the recorder takes a whole part and submits it, so the render thread never waits at a part's endto perf(vk): the recorder takes a whole part and submits it, so the render thread never waits at a part's end