WIP: perf(vk): a texture binding's slot version read once and handed to the fingerprint and the view #2055
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!2055
Loading…
Reference in a new issue
No description provided.
Delete branch "siantm/Longshot:perf/vk-slot-version-once"
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
Every texture and storage-image binding of a descriptor build (
build_descriptors,gpu/vk/vk_draw.cpp) askedgpu/vk/vk_bank.cppfor its texture slot's version twice: once for the Descriptor reuse fingerprint (desc_fp_note) and once more at the top oftexture_view_of. This reads it once a binding and hands the value to both, so both also see the same version.texture_view_ofanddesc_fp_notekeep their forms without the version (the upscalers' and the movies' callers), which read it as before. The descriptors and the fingerprint are the same; the result expected is a little less work for the replay thread at every textured draw.No switch: the same descriptors either way; a revert gives back main's code.
From c0b017ac on
exp/x86-asm-sync.Read first: the texture and image loops of
build_descriptors,texture_view_ofanddesc_fp_noteingpu/vk/vk_draw.cpp.Testing
Tested: Windows 11, i7-9700, RTX 4060, the builder's llvm-mingw clang 23; head
f854c8f9on main @75317e1f. Every check ran throughtests/run_all.py --strict --exactin build jobs of the PC's job queue, in a hidden console (logt-dc3-vk-slot-version):tests/test_slot_version_once.py(new: cutsdesc_fp_note's two forms and runs them with a countingtexture_slot_version, and checks the loops read the version once and hand it to both): pass.--changed 75317e1f; its own are among them): 146 passed, 15 failed. Failed:tests/test_sampler_admission.py,tests/test_fsr_uniform_ownership.py,tests/test_draw_stream_visibility.py,tests/test_pipeline_play_cache.py,tests/test_device_lost_exit.py,tests/test_draw_binds.py,tests/test_dev_packs.py,tests/test_frame_overlap.py,tests/test_linux_pressure.py,tests/test_mesh_conv.py,tests/test_prepare_cache_save.py,tests/test_recorder_own_idle.py,tests/test_upscaler_rows.py,tests/test_vk_occlusion_queries.py: each fails on main @75317e1ftoo (the same checks on main in the same queue batch, logt-dc3-base).tests/test_recording_admission.pyfails on main @75317e1ftoo, now and then: its "worker slot ownership" case failed on main in 7 of 10 runs made alternately with this branch's (logst-dc5-base-ra-*), with the same message.It has NOT run in the game on this PC yet.
Still to test:
75317e1fagainst this PR, each warmed a minute in the scene first, then interleaved (main, this PR, this PR, main), two runs each, expecting the same picture (no wrong, missing or flickering texture) and the replay's time a frame the same as or lower than main's (by hand)tests/test_slot_version_once.pywith--cxx g++and--cxx clang++, expecting PASS (by hand)Measurements
Verdict: with the fingerprint on (Descriptor reuse) a binding takes 1.14-1.52x less time (1.7-4.2 ns); with it off, 1.10-1.13x in the baseline build; with -march=native 1.08x for handle tables and no gain for distinct handles (0.98x, 0.2 ns slower). The view's lookup is a stand-in: only the version reads' cost is measured here. A micro-benchmark on authored input; not measured in play.
Micro-benchmark (
vk_slot_version_bench.py: a script on this PC, not in this PR):desc_fp_notecut out ofvk_draw.cppat both commits andtexture_slot_versionout ofvk_bank.cpp, compiled in a file of its own (a call a read, as in the build);texture_view_ofis a stand-in kept out of line (its old form reads the version itself, its new form is given it), then an authored table of hits; the texture binding loop as each side has it; 4,096 bindings a repetition, slots under 8,192; three rounds run alone in this PC's job queue (2026-11-22, 01:21-01:23 UTC; logbench-vk-walks.log), the builder's llvm-mingw clang 23.1.5 at -O2 for baseline x86-64 and at -O2 -march=native (this PC's i7-9700, as the game is built here), 301 repetitions a case, old and new taking turns on the same authored inputs, their results equal at every repetition. Each cell: ns per binding, the median of the three rounds' medians, in parentheses the lowest and highest of them; Δ: old over new (above 1, this PR is quicker).75317e1ff854c8f9Risks
A texture resolved against a slot version read earlier than before (once, before the fingerprint, instead of again inside
texture_view_of): a slot registered again between the two reads now gives a fingerprint and a view of the same, older version, which the next build sees as changed and builds again. What could show is a wrong texture for a draw; a revert is enough (no switch). The renderer is compiled again after updating, as for any renderer change; no cache or setting changes.Checklist
Ready for review
main; the title is a Conventional Commit.mise run testpasses (any failure that also fails onmainis named). Not yet: only the checks in Testing ran.main, or Measurements says why there is nothing to measure. Not yet (Measurements).third_party/untouched; nothing GPL-incompatible.When it applies
Assisted-by:/Co-Authored-By:trailers.@Beanstalk This changes the renderer's descriptor builds for every textured draw: a slot's version is read once (gpu/vk). No switch: a revert gives back main's code. It has not run in the game on these PCs yet. Please play Kokiri Forest and Hyrule Field, two minutes each, the PR against main on the same PC, with a picture in each compared with main's, on Linux with NVIDIA, Linux with AMD (RADV), Windows with Intel and the Steam Deck. The same picture as main, no crash, device lost, validation error or FATAL line; frame time p99 equal to main's or better, within the run-to-run spread.
@siantm only members of
BigBeanscan start the farm. One of them can approve this request by reacting 👍 to it or to this comment.@Beanstalk This replaces my request 65766 on this head, so that it follows the farm's picture and hardware rules. This reads a texture binding's slot version once and hands it to the fingerprint and the view, in every textured draw's descriptor build (gpu/vk). No switch: a revert gives back main's code. Please start from the kokiri-start save and take the picture right after it loads and settles, before any input, with a main-against-itself picture on the same PC at that point; then, for frame times only, play Kokiri Forest and Hyrule Field, two minutes each; at least two starts a side, interleaved, the PR against main on the same PC, on Windows with an RTX 4070, Windows with an Arc B580, Linux with an RX 6600 (RADV) and the Steam Deck. The same picture as main within main's own difference from itself; frame time p99 equal to main's or better, within the run-to-run spread. The log must have no FATAL, device lost or validation error line.
@siantm only members of
BigBeanscan start the farm. One of them can approve this request by reacting 👍 to it or to this comment.Measurements: a micro-benchmark on authored input (old against new, three rounds, x86-64 baseline and -march=native; not measured in play) and the Verdict rewritten to match.
@Beanstalk One correction to my request 69698 above, for its picture point: please take each picture 5 s (300 frames at 60 fps) after the first frame whose picture is not a single uniform colour, with no input in between, and take the main-against-itself picture from the same point. A picture taken during the load's fade-in from black is not comparable: run 70269 on #2156 caught the fade in 3 of 4 starts.
@siantm only members of
BigBeanscan start the farm. One of them can approve this request by reacting 👍 to it or to this comment.View command line instructions
Checkout
From your project repository, check out a new branch and test the changes.