docs(tested): native mesh shaders on an RX 9070 XT (RADV): memory tour and Market fps against Compute #1087
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!1087
Loading…
Reference in a new issue
No description provided.
Delete branch "GrimVisage/Longshot:docs/tested-mesh-native-rdna4"
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: docs · Risk: low · Tested on: Linux (Fedora 45), Ryzen 5 5600X, RX 9070 XT (RADV 26.2.7) · Needs: none
Why
On Linux AMD the game draws with mesh shaders run as compute shaders (Compute), on every card, since https://github.com/NotRexed/Hookshot/pull/2362: RADV's native task and mesh path hung Navi 2x and the RDNA 3 APUs. Nobody had recorded how the native path does on RDNA 4, where RADV offers it. This PR records eight runs from an RX 9070 XT (GFX1201) on RADV,
RX_VK_MESH_SHADERS=nativeagainst the default, so the next look at the Linux AMD default has numbers from RDNA 4: memory with the repo's memory tour, and fps in the Castle Town market. It changes no default. See #583 (an RX 6600 losing its device with native mesh shaders, on Windows): the same switch, other hardware. No issue or tracking issue on this tracker covers native mesh shaders on Linux AMD (searched:RX_VK_MESH_SHADERS, "mesh shaders",[Tracking], "AMD Linux", "RADV").What changed
Eight run files under
docs/tested/, nothing else. They compare the Linux AMD default (Compute) againstRX_VK_MESH_SHADERS=nativein one build of main1764fed2, on a Ryzen 5 5600X with an RX 9070 XT on Fedora 45.docs/tested/2026-11-15-mesh-shaders-native-memtour-grimvisage-run01.mdtorun04.md, one memory tour each (run01 and run04 Compute, run02 and run03 Native), in the format ofdocs/tested/README.md.docs/tested/2026-11-15-mesh-shaders-native-market-grimvisage-run01.mdtorun04.md, one fps run each in the Market (run01 and run04 Compute, run02 and run03 Native), each with its five parts' numbers.The runs these files record
One build of main
1764fed2(longshot abc-main1764 (source f6bc6f5f)), built as a player's (the builder AppImage's--cli), copied into two game folders, one a side: the program cache's stamp holds the mesh path, so one folder switched back and forth would prepare its programs again in the measured runs. Both folders were prepared and warmed before the series. The log saidvk: mesh shaders: Automatic, from the default: the compute pathin every Compute run andvk: mesh shaders: Native, from RX_VK_MESH_SHADERS: nativein every Native run.Memory tours:
scripts/memtour.pyfrom mainbbe421f5with its defaults (the test profile: no window, no audio; settle 20 s, then destinations 13,1,2,3,0,7,9,11 with 40 s of walking and camera swings in each, memory sampled every 2 s), one child save copied fresh each run, interleaved Compute, Native, Native, Compute.fps runs: fullscreen 2560x1440 (
HOS_FULLSCREEN=1), SDR, the game's own upscaler at Render resolution 100% (Quality preset Auto (this PC); no FSR 4 library in the test folders), frame generation off, the FPS counter off, Threaded rendering Auto (on), uncapped (HOS_FPS_MAX=240),RX_VK_AUTO_TUNE=0, the frame-time log on (RX_VK_FRAME_LOG), Disable camera centering On, one kept options file (HOS_OPTIONS_FILE). Link was steered by his position through the control bridge in the #688 routine, five parts of 30 s each: standing at the gate, laps between the gate and the fountain, the same laps turning the camera (right stick at half), standing at the fountain, turning the camera there. Before each part the clock was set to 12:00 with the Debug tab's Set time (Debug tools on for the test run only,HOS_OOT_CHEATS=1; the log'scheats: requested 12:00checked each time). A warm-up run of each side, then four measured runs interleaved Compute, Native, Native, Compute. Every run's log shows one swapchain size, 2560x1440, from start to end.Measurements
Verdict: nothing to measure in the game for this PR: docs only. The files pass their format check, and the game-free suite gives the same result as on main (CONTRIBUTING.md, "Measurements":
docs).Hardware: Ryzen 5 5600X, 31 GB, Fedora 45 (kernel 7.3.3)
Scene: none (game-free checks)
Method:
python -I -B tests/test_tested_configurations.py, andpython -I -B tests/run_all.pyon main and on this PRHow to reproduce: the two commands above from the repository root
f7922126aa34b58etests/test_tested_configurations.pytests/run_all.pyWhat the runs show
Baseline: main @
1764fed2with mesh shaders Compute (the Linux AMD default), against the same build withRX_VK_MESH_SHADERS=native: the switch is on main already, so its effect is the switch, not a main-against-PR build.Memory (memtour): with its programs prepared, Native holds about 1 GB less of the game's private memory than Compute, in every area from the start; shared memory is the same. The two sides don't overlap (a side's two runs differ by 20 to 115 MB).
1764fed2(run01, run04)1764fed2(run02, run03)fps (Market): within the spread in every part: the two sides' runs overlap in fps, 1% low and frame times. The scene is CPU-bound here (21 to 26 ms a frame against about 12 ms of GPU time), so the GPU time is the number that moves: Native is 0.4 to 0.6 ms a frame lower in four of the five parts, with the sides' runs apart (gate 11.80, 11.60 against 10.95, 11.30 ms; laps 12.90, 13.30 against 12.50, 12.60; fountain 11.70, 11.85 against 11.60, 11.20; fountain turning the camera 12.20, 12.50 against 11.90, 11.90). The replay's submit time leans the other way, by about half a millisecond at the gate (12.61, 12.14 against 12.91, 13.19 ms).
Medians of the two runs a side; fps, 1% low, median and p99 frame time from
RX_VK_FRAME_LOGover each part's presents; the replay's submit time and GPU time the medians of the log's status lines in each part.1764fed21764fed2The laps-turning-the-camera 1% low of the Native side (18.2) is one run's dip (run02: 12.7; run03: 23.7); its fps and frame times are in line.
Stability: no device lost, no
GPU hungline, no amdgpu kernel line in any Native run (the two tours, the warm-up and the two fps runs), nor in a 20-minute Native soak before the series (the Market, then Kokiri Forest, Hyrule Field, Kakariko Village, Death Mountain Trail and back).A caution for a first start: that soak was the Native folder's first start, and it prepared its programs inside the game: the game's private memory reached 7.9 GB (a working set of 11.5 GB), the log said
vk: memory pressure on Linux93 times, and with the PC's 8 GB of swap full (0 MB swap freein those lines) MemAvailable fell to 1.3 GB and the desktop nearly froze. The prepared runs above stay near 2.8 GB private. A PC with 16 GB on a first Native start is the case to watch.A status line of each side, in the fountain part (run01 and run02):
Flags and settings
None: this PR adds or changes no switch, menu row, ini key or build flag (the runs use
RX_VK_MESH_SHADERS, unchanged).Testing
Tested: the eight files with
tests/test_tested_configurations.py(25 tests OK) and the whole game-free suite withtests/run_all.py(the same 13 failures as on mainf7922126), on Fedora 45. The runs themselves are what the files record.Not tested: native mesh shaders on other RDNA 4 cards, on RDNA 2 and 3 under a newer RADV than the one behind https://github.com/NotRexed/Hookshot/pull/2362, and on Windows: this PC only. A first Native start on a 16 GB PC, after the caution above.
Must be tested: None (docs only; the runs make no claim about other PCs).
Risks
What could break: nothing in the game: eight Markdown files under
docs/tested/.Turning it off: revert the commit.
Compatibility: Nothing changes.
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
Assisted-by:/Co-Authored-By:trailers.docs/tested/for each new measured run (format indocs/tested-configurations.md).🤖 Generated with Claude Code
https://claude.ai/code/session_015WSf5yacFywzof5jcUYpsk
WIP: docs(tested): native mesh shaders on an RX 9070 XT (RADV): memory tour and Market fps against Computeto docs(tested): native mesh shaders on an RX 9070 XT (RADV): memory tour and Market fps against ComputeView command line instructions
Checkout
From your project repository, check out a new branch and test the changes.