perf(vk): RX_VK_TEX_AHEAD=2 prepares every block-compressed texture the game makes, not only those an earlier load drew #1813

Open
GummiNaomi wants to merge 2 commits from GummiNaomi/Longshot:perf/tex-ahead-every into main
Member

Summary

After a teleport or an area change, the first frames in the new place are slow because many textures are made and uploaded at their first draw. #1122's RX_VK_TEX_AHEAD=1 prepares textures during light frames, but only ones an earlier load drew at the same place in memory. An area change in play mostly draws textures the game made seconds before, at new places. This PR adds RX_VK_TEX_AHEAD=2, which prepares every block-compressed texture the game makes during light frames and keeps no list. It is off by default.

Measured again at this head on a 12 GB card, =2 is not better than =1. It improves the Market and Hyrule Field entries, but the next area's entry (Kakariko after the Market) gets much worse: the budget fills with the Market's prepared textures, most of them never drawn, and evicts them again. =1 has the best tails. See Measurements. The default stays off; whether =2 stays as an option is for the Test Farm's 8 GB and Deck runs to settle.

Changed in review (futondestroyer):

  • =2 no longer reads cache/longshot_textures_drawn.bin. The file is read through texture_ahead::load_list, which opens it only for a mode that uses the list, and only once a texture has been taken to look up in it (with =1 it used to be read before the room check too).
  • The room check now counts the texture about to be made. texture_ahead::room(cache, budget, cost) with ahead_cost (texture_ahead::cost_bound, an upper bound from the descriptor: decoded ASTC at RGBA8 or BC's whole blocks, whichever is more). A texture that would take the 256 MB reserve is made at its first draw instead. Before, the cache was checked as it stood and the image charged afterwards, so the last texture prepared could go past the reserve by its own size. The allocator's rounding of that one image is still outside the bound.

Read first: gpu/vk/texture_ahead.h (mode, uses_list, load_list, room, cost_bound), then texture_ahead_run, ahead_load and ahead_cost in gpu/vk/vk_texture.cpp, and tests/test_texture_ahead.cpp.

See #1122. Related: #1810 changes the same loop (a move not yet registered is skipped); the two merge cleanly, with that skip kept.

Testing

Tested: Windows 11 Pro 26200, RTX 4070 SUPER 12 GB (NVIDIA 617.42), Ryzen 9 9950X3D with SMT off, 32 GB, at 3af1d156. One warm-up tour, then 12 measured teleport tours: =0, =1 and =2 interleaved 0,1,2,2,1,0, at 16 cores and again held to 4 cores (affinity 0xF00). All three areas reached in every run, no device-lost, out-of-memory or fatal lines in the logs. With =2 the log has no textures drawn before are known line (the list isn't read); with =1 it has one.

Checks at 3af1d156:

  • tests/test_texture_ahead.py passes on Windows (clang) and on Linux (WSL Ubuntu 24.04, g++). tests/test_switch_table.py passes on Windows.
  • The new cases fail without the fix. With the list read whatever the mode, the seeded-file case fails (!load_list(2, …)). With the cost ignored, !room(2000 MB, 2400 MB, 145 MB) fails. With ahead_load called before the room check and for =2, the source check fails ("the list is asked for with RX_VK_TEX_AHEAD=2").
  • gpu/vk/vk_texture.cpp compiles syntax-only with the builder's Windows flags.
  • Of the checks that read the changed files, 19 fail on Windows and 15 on Linux, and each fails the same way on this PR's base 65488813 (MSVC headers under -Werror, a missing Tooling checkout, markers other changes on main moved). None is this PR's.

Still to test:

  • Windows, 8 GB card (windows+vram8gb): the teleport walk with RX_VK_TEX_AHEAD 0, 1 and 2, expecting =2 to evict more than here and its second arrival to be slower than with 0 (Test Farm: running, @Beanstalk)
  • Windows, 12 GB card (windows+vram12gb): the same, expecting about what Measurements shows (Test Farm: running, @Beanstalk)
  • Steam Deck (steamdeck): the same, expecting =2 no better than =1 and no out-of-memory messages (Test Farm: running, @Beanstalk)
  • Linux, any GPU: one tour with =2, expecting no errors (by hand)

Measurements

Verdict: on a 12 GB card at this head, =2 doesn't beat =1. Its slow-frame time after arrivals is no better than =0 (825, 646 ms against 849, 683 at 16 cores). Its worst frame is worse (439-696 ms against 308-669 for =0 and 235-385 for =1), all of it at the second arrival. It evicts ~2.5-3 times as many textures for the budget. =1 has the fewest frames of 33 ms or more and the lowest p99 on both core counts. Off stays the default.

Setup: the hardware in Testing; the build of this PR @ 3af1d156 (base main @ 65488813). Main's defaults (the control tool's --profile test, so no longshot_env.txt) plus RX_VK_FRAME_TRACE=1 and RX_VK_TEX_AHEAD=0/1/2. The tour: kokiri-start save, then the debug teleport to the Market, Kakariko Village and Hyrule Field (child), ~35 s of walking in each. Runs interleaved 0,1,2,2,1,0, after one =1 warm-up run that filled the list and the pipeline caches. Measured:

  • each arrival's wait: the frames' summed time from the teleport taken to the map reached
  • the frames of 25 ms or more in the 3 s after each arrival
  • frames of 33 ms or more, p99 and the worst frame, from the first teleport to the end of the last walk
  • peak GPU memory used (nvidia-smi, all processes) and the process's peak private bytes, sampled once a second
  • the log's vk: textures ahead: and textures retired lines

Baseline: this PR's own build with RX_VK_TEX_AHEAD=0 and =1. Those are main's two behaviours, unchanged by this PR except for the room check, and one build lets the three values run interleaved on the same binary. Each cell gives the two runs of that value.

Before, off (=0): this PR @ 3af1d156 Before, #1122's list (=1): this PR @ 3af1d156 After (=2): this PR @ 3af1d156
16 cores: frames ≥25 ms in the 3 s after the 3 arrivals, summed 849, 683 ms 748, 595 ms 825, 646 ms
16 cores: frames ≥33 ms 33, 30 25, 28 33, 32
16 cores: p99 15.9, 15.1 ms 15.4, 15.3 ms 18.0, 17.4 ms
16 cores: worst frame 314, 308 ms 235, 238 ms 528, 439 ms
16 cores: arrival waits, the 3 summed 10.2, 9.6 s 9.5, 9.6 s 10.5, 10.0 s
4 cores: frames ≥25 ms after the arrivals 987, 885 ms 829, 728 ms 838, 897 ms
4 cores: frames ≥33 ms 51, 56 34, 32 38, 36
4 cores: p99 24.4, 21.1 ms 19.6, 19.6 ms 21.7, 19.7 ms
4 cores: worst frame 475, 669 ms 385, 262 ms 696, 680 ms
4 cores: arrival waits, the 3 summed 10.6, 10.7 s 11.4, 11.3 s 12.0, 11.8 s
worst frame at each arrival: Market / Kakariko / Hyrule Field (16 cores) 314, 169 / 248, 308 / 171, 151 ms 222, 234 / 235, 238 / 186, 166 ms 132, 155 / 528, 439 / 101, 142 ms
the same, 4 cores 162, 194 / 171, 217 / 149, 198 ms 256, 262 / 385, 242 / 129, 130 ms 116, 130 / 696, 680 / 182, 154 ms
GPU memory peak, all processes (16 cores; 4 cores) 11,040, 11,197; 11,411, 11,400 MB 11,245, 11,245; 10,829, 10,908 MB 11,287, 11,477; 11,466, 11,457 MB
private bytes peak (16 cores; 4 cores) 22,474, 22,302; 23,140, 22,771 MB 22,448, 22,857; 22,360, 22,436 MB 22,988, 22,888; 23,387, 23,380 MB
textures prepared (images) none 2,044-2,129 (2.4-2.5 GB) 9,956-11,118 (12.2-14.5 GB)
textures evicted for the budget 3,657-5,012 3,643-3,942 10,252-11,444

With =2, the worst frame of each run is the Kakariko entry, the second arrival. That frame spent 202-525 ms on uploads: the cache was full of the Market's prepared textures, which the budget then evicted, and Kakariko's own textures had found no room ahead. GPU memory peaks about the same in all three because the budget holds it, so the cost shows as eviction churn rather than as a higher peak.

ta20-c16-m2-3 (=2): vk: textures ahead: 10069 prepared so far (12184 MB of images, 1773 ms); ... the cache had no room at times
ta20-c16-m1-2 (=1): vk: textures ahead: 2084 prepared so far (2461 MB of images, 393 ms); ... the cache had no room at times

The earlier numbers in this description (arrival slow time halved) came from 54a42715 with uncapped play settings and the rule in a temporary form. They don't reproduce on current main with its defaults, so they have been replaced by the table above.

Risks

  • What could break: with =2, many images are made that are never drawn. The budget evicts them again, ~2.5-3 times main's budget evictions over this tour, and the next area's first frame can be slower than with the switch off (up to ~700 ms here, against ~200-300 ms). A smaller card should see this sooner. The light-frame time it uses is bounded by RX_VK_TEX_AHEAD_MS. The room check now keeps the 256 MB reserve after the image is made, so =1 and =2 prepare slightly less near the budget than before; a texture that doesn't fit is made at its first draw, as with the switch off.
  • Turning it off: leave RX_VK_TEX_AHEAD unset (the default), or use 1.
  • After updating: nothing changes for saves or longshot_options.ini. cache/longshot_textures_drawn.bin is read and grown with =1 as before; with =2 it is neither read nor written.

Checklist

Ready for review

  • One topic, on current main; the title is a Conventional Commit.
  • Tested on my own PC, in the game for a change the game runs, as Testing says.
  • mise run test passes (any failure that also fails on main is named). (The changed files' checks on Windows and Linux; every failure also fails on the base 65488813.)
  • Measured against main, or Measurements says why there is nothing to measure.
  • No game files, build output or hand-edited generated files; third_party/ untouched; nothing GPL-incompatible.

When it applies

  • A switch added or changed: it ships at the default the measurements chose, with its row in gpu/vk/switches.def.
  • A check under tests/ that reads changed source is updated, not weakened.
  • Docs updated.
  • AI assistance recorded with Assisted-by: / Co-Authored-By: trailers.

Ready to merge

  • Everything in Still to test has run, its results in Tested.
  • Tier 1 hardware it can affect is tested (docs/dev/standards.md, "Hardware tiers").
  • Approved in review, every comment answered.
  • CI passes on the head; on current main, measured again or a comment says why the numbers still hold.

🤖 Generated with Claude Code

## Summary After a teleport or an area change, the first frames in the new place are slow because many textures are made and uploaded at their first draw. #1122's `RX_VK_TEX_AHEAD=1` prepares textures during light frames, but only ones an earlier load drew at the same place in memory. An area change in play mostly draws textures the game made seconds before, at new places. This PR adds `RX_VK_TEX_AHEAD=2`, which prepares every block-compressed texture the game makes during light frames and keeps no list. It is off by default. Measured again at this head on a 12 GB card, `=2` is not better than `=1`. It improves the Market and Hyrule Field entries, but the next area's entry (Kakariko after the Market) gets much worse: the budget fills with the Market's prepared textures, most of them never drawn, and evicts them again. `=1` has the best tails. See Measurements. The default stays off; whether `=2` stays as an option is for the Test Farm's 8 GB and Deck runs to settle. Changed in review (futondestroyer): - `=2` no longer reads `cache/longshot_textures_drawn.bin`. The file is read through `texture_ahead::load_list`, which opens it only for a mode that uses the list, and only once a texture has been taken to look up in it (with `=1` it used to be read before the room check too). - The room check now counts the texture about to be made. `texture_ahead::room(cache, budget, cost)` with `ahead_cost` (`texture_ahead::cost_bound`, an upper bound from the descriptor: decoded ASTC at RGBA8 or BC's whole blocks, whichever is more). A texture that would take the 256 MB reserve is made at its first draw instead. Before, the cache was checked as it stood and the image charged afterwards, so the last texture prepared could go past the reserve by its own size. The allocator's rounding of that one image is still outside the bound. Read first: `gpu/vk/texture_ahead.h` (`mode`, `uses_list`, `load_list`, `room`, `cost_bound`), then `texture_ahead_run`, `ahead_load` and `ahead_cost` in `gpu/vk/vk_texture.cpp`, and `tests/test_texture_ahead.cpp`. See #1122. Related: #1810 changes the same loop (a move not yet registered is skipped); the two merge cleanly, with that skip kept. ## Testing **Tested:** Windows 11 Pro 26200, RTX 4070 SUPER 12 GB (NVIDIA 617.42), Ryzen 9 9950X3D with SMT off, 32 GB, at 3af1d156. One warm-up tour, then 12 measured teleport tours: `=0`, `=1` and `=2` interleaved 0,1,2,2,1,0, at 16 cores and again held to 4 cores (affinity 0xF00). All three areas reached in every run, no device-lost, out-of-memory or fatal lines in the logs. With `=2` the log has no `textures drawn before are known` line (the list isn't read); with `=1` it has one. Checks at 3af1d156: - `tests/test_texture_ahead.py` passes on Windows (clang) and on Linux (WSL Ubuntu 24.04, g++). `tests/test_switch_table.py` passes on Windows. - The new cases fail without the fix. With the list read whatever the mode, the seeded-file case fails (`!load_list(2, …)`). With the cost ignored, `!room(2000 MB, 2400 MB, 145 MB)` fails. With `ahead_load` called before the room check and for `=2`, the source check fails ("the list is asked for with RX_VK_TEX_AHEAD=2"). - `gpu/vk/vk_texture.cpp` compiles syntax-only with the builder's Windows flags. - Of the checks that read the changed files, 19 fail on Windows and 15 on Linux, and each fails the same way on this PR's base 65488813 (MSVC headers under `-Werror`, a missing Tooling checkout, markers other changes on main moved). None is this PR's. **Still to test:** - Windows, 8 GB card (`windows+vram8gb`): the teleport walk with `RX_VK_TEX_AHEAD` 0, 1 and 2, expecting `=2` to evict more than here and its second arrival to be slower than with 0 (Test Farm: running, @Beanstalk) - Windows, 12 GB card (`windows+vram12gb`): the same, expecting about what Measurements shows (Test Farm: running, @Beanstalk) - Steam Deck (`steamdeck`): the same, expecting `=2` no better than `=1` and no out-of-memory messages (Test Farm: running, @Beanstalk) - Linux, any GPU: one tour with `=2`, expecting no errors (by hand) ## Measurements **Verdict:** on a 12 GB card at this head, `=2` doesn't beat `=1`. Its slow-frame time after arrivals is no better than `=0` (825, 646 ms against 849, 683 at 16 cores). Its worst frame is worse (439-696 ms against 308-669 for `=0` and 235-385 for `=1`), all of it at the second arrival. It evicts ~2.5-3 times as many textures for the budget. `=1` has the fewest frames of 33 ms or more and the lowest p99 on both core counts. Off stays the default. Setup: the hardware in Testing; the build of this PR @ 3af1d156 (base main @ 65488813). Main's defaults (the control tool's `--profile test`, so no `longshot_env.txt`) plus `RX_VK_FRAME_TRACE=1` and `RX_VK_TEX_AHEAD=0/1/2`. The tour: `kokiri-start` save, then the debug teleport to the Market, Kakariko Village and Hyrule Field (child), ~35 s of walking in each. Runs interleaved 0,1,2,2,1,0, after one `=1` warm-up run that filled the list and the pipeline caches. Measured: - each arrival's wait: the frames' summed time from the teleport taken to the map reached - the frames of 25 ms or more in the 3 s after each arrival - frames of 33 ms or more, p99 and the worst frame, from the first teleport to the end of the last walk - peak GPU memory used (nvidia-smi, all processes) and the process's peak private bytes, sampled once a second - the log's `vk: textures ahead:` and `textures retired` lines Baseline: this PR's own build with `RX_VK_TEX_AHEAD=0` and `=1`. Those are main's two behaviours, unchanged by this PR except for the room check, and one build lets the three values run interleaved on the same binary. Each cell gives the two runs of that value. | | Before, off (`=0`): this PR @ 3af1d156 | Before, #1122's list (`=1`): this PR @ 3af1d156 | After (`=2`): this PR @ 3af1d156 | |---|---|---|---| | 16 cores: frames ≥25 ms in the 3 s after the 3 arrivals, summed | 849, 683 ms | 748, 595 ms | 825, 646 ms | | 16 cores: frames ≥33 ms | 33, 30 | 25, 28 | 33, 32 | | 16 cores: p99 | 15.9, 15.1 ms | 15.4, 15.3 ms | 18.0, 17.4 ms | | 16 cores: worst frame | 314, 308 ms | 235, 238 ms | 528, 439 ms | | 16 cores: arrival waits, the 3 summed | 10.2, 9.6 s | 9.5, 9.6 s | 10.5, 10.0 s | | 4 cores: frames ≥25 ms after the arrivals | 987, 885 ms | 829, 728 ms | 838, 897 ms | | 4 cores: frames ≥33 ms | 51, 56 | 34, 32 | 38, 36 | | 4 cores: p99 | 24.4, 21.1 ms | 19.6, 19.6 ms | 21.7, 19.7 ms | | 4 cores: worst frame | 475, 669 ms | 385, 262 ms | 696, 680 ms | | 4 cores: arrival waits, the 3 summed | 10.6, 10.7 s | 11.4, 11.3 s | 12.0, 11.8 s | | worst frame at each arrival: Market / Kakariko / Hyrule Field (16 cores) | 314, 169 / 248, 308 / 171, 151 ms | 222, 234 / 235, 238 / 186, 166 ms | 132, 155 / 528, 439 / 101, 142 ms | | the same, 4 cores | 162, 194 / 171, 217 / 149, 198 ms | 256, 262 / 385, 242 / 129, 130 ms | 116, 130 / 696, 680 / 182, 154 ms | | GPU memory peak, all processes (16 cores; 4 cores) | 11,040, 11,197; 11,411, 11,400 MB | 11,245, 11,245; 10,829, 10,908 MB | 11,287, 11,477; 11,466, 11,457 MB | | private bytes peak (16 cores; 4 cores) | 22,474, 22,302; 23,140, 22,771 MB | 22,448, 22,857; 22,360, 22,436 MB | 22,988, 22,888; 23,387, 23,380 MB | | textures prepared (images) | none | 2,044-2,129 (2.4-2.5 GB) | 9,956-11,118 (12.2-14.5 GB) | | textures evicted for the budget | 3,657-5,012 | 3,643-3,942 | 10,252-11,444 | With `=2`, the worst frame of each run is the Kakariko entry, the second arrival. That frame spent 202-525 ms on uploads: the cache was full of the Market's prepared textures, which the budget then evicted, and Kakariko's own textures had found no room ahead. GPU memory peaks about the same in all three because the budget holds it, so the cost shows as eviction churn rather than as a higher peak. ``` ta20-c16-m2-3 (=2): vk: textures ahead: 10069 prepared so far (12184 MB of images, 1773 ms); ... the cache had no room at times ta20-c16-m1-2 (=1): vk: textures ahead: 2084 prepared so far (2461 MB of images, 393 ms); ... the cache had no room at times ``` The earlier numbers in this description (arrival slow time halved) came from 54a42715 with uncapped play settings and the rule in a temporary form. They don't reproduce on current main with its defaults, so they have been replaced by the table above. ## Risks - **What could break:** with `=2`, many images are made that are never drawn. The budget evicts them again, ~2.5-3 times main's budget evictions over this tour, and the next area's first frame can be slower than with the switch off (up to ~700 ms here, against ~200-300 ms). A smaller card should see this sooner. The light-frame time it uses is bounded by `RX_VK_TEX_AHEAD_MS`. The room check now keeps the 256 MB reserve after the image is made, so `=1` and `=2` prepare slightly less near the budget than before; a texture that doesn't fit is made at its first draw, as with the switch off. - **Turning it off:** leave `RX_VK_TEX_AHEAD` unset (the default), or use `1`. - **After updating:** nothing changes for saves or `longshot_options.ini`. `cache/longshot_textures_drawn.bin` is read and grown with `=1` as before; with `=2` it is neither read nor written. ## Checklist ### Ready for review - [x] One topic, on current `main`; the title is a Conventional Commit. - [x] Tested on my own PC, in the game for a change the game runs, as Testing says. - [x] `mise run test` passes (any failure that also fails on `main` is named). (The changed files' checks on Windows and Linux; every failure also fails on the base 65488813.) - [x] Measured against `main`, or Measurements says why there is nothing to measure. - [x] No game files, build output or hand-edited generated files; `third_party/` untouched; nothing GPL-incompatible. ### When it applies - [x] A switch added or changed: it ships at the default the measurements chose, with its row in `gpu/vk/switches.def`. - [x] A check under `tests/` that reads changed source is updated, not weakened. - [x] Docs updated. - [x] AI assistance recorded with `Assisted-by:` / `Co-Authored-By:` trailers. ### Ready to merge - [ ] Everything in Still to test has run, its results in Tested. - [ ] Tier 1 hardware it can affect is tested (docs/dev/standards.md, "Hardware tiers"). - [ ] Approved in review, every comment answered. - [ ] CI passes on the head; on current `main`, measured again or a comment says why the numbers still hold. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
perf(vk): RX_VK_TEX_AHEAD=2 prepares every block-compressed texture the game makes, not only those an earlier load drew
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 12s
CI / Workflow lint (actionlint) (pull_request) Successful in 16s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 14s
Third-party parts / As their notes say (pull_request) Successful in 19s
Website / Website on Cloudflare (pull_request_target) Has been skipped
PR description / Description follows the template (pull_request_target) Failing after 5s
78143225a2
With RX_VK_TEX_AHEAD=1 a texture is prepared during a light frame only when an earlier
load drew it at the same address. An area change in play draws textures the game made
seconds before at new addresses, which that list misses. =2 prepares every
block-compressed texture the game makes, keeps no list, and stays inside the same room
under the texture cache's budget and the same time a light frame.

Walking the market, Kakariko and Hyrule Field by debug teleport (Ryzen 9 9950X3D, RTX
4070 SUPER): the time in frames of 25 ms or more in the 3 s after an arrival about
halved. It prepares many textures the game never draws; off by default.

Assisted-by: Claude Code
Co-Authored-By: Claude Opus 5.5 <[email protected]>
futondestroyer left a comment

Recommendation: Request changes. Mode 2 still loads the previous-draw list, and the proposed all-BC policy needs exact-head evidence for memory pressure and frame pacing.

Does it help: Unproven for this head. GummiNaomi's historical notes report combined arrival waits of 486–1086 ms falling to 200–601 ms, with 6.3–6.9 GB additional prepared image bytes and roughly 7,600–8,500 never-drawn preparations. Prepared bytes are not peak resident VRAM. The run was on older 54a42715 with a temporary rule and two non-interleaved runs per arm.
Other systems: Recheck low-VRAM GPUs and Deck, measuring peak residency, eviction churn and light-frame tails, as well as arrival waits.
Switches: RX_VK_TEX_AHEAD=2 is opt-in and widens the existing mode 1 policy.
Reviewed: 78143225 against 0de25c3e, with exact Graphify maps and source verification. Read texture-ahead loading, admission, allocation and source checks. Nothing run.
Fixed in this review: Nothing.
Checked and fine: Mode 2 bypasses the membership filter; mode 1 retains it. The pre-allocation budget gate remains in place.

Blocking

  1. gpu/vk/vk_texture.cpp:3011: the path still loads and populates the previous-draw set before mode and room checks. Mode 2 then ignores it, so even an empty or budget-blocked queue can pay the file read and startup cost. Skip loading in mode 2 and add a seeded-file test that confirms mode 1 still reads it.
  2. gpu/vk/vk_texture.cpp:3012,3023: the check sees current use plus reserve before creating the candidate; the allocation is charged afterward. Treat it as a soft preflight in the description or account for candidate cost if promising a hard reserve. Establish the real cost with matched exact-head A/B evidence before merge.

Testing: The head has no gameplay or measurement receipt and the description and Guide screenshots checks are red. Record current-main comparisons and the actual scope of completed game-free checks after rebase.
Overlaps and merge order: #1810 changes the same texture-ahead loop. Preserve its pending-move skip if both land; measure independently first, then test both switches together.

**Recommendation: Request changes.** Mode 2 still loads the previous-draw list, and the proposed all-BC policy needs exact-head evidence for memory pressure and frame pacing. **Does it help:** Unproven for this head. GummiNaomi's historical notes report combined arrival waits of 486–1086 ms falling to 200–601 ms, with 6.3–6.9 GB additional prepared image bytes and roughly 7,600–8,500 never-drawn preparations. Prepared bytes are not peak resident VRAM. The run was on older 54a42715 with a temporary rule and two non-interleaved runs per arm. **Other systems:** Recheck low-VRAM GPUs and Deck, measuring peak residency, eviction churn and light-frame tails, as well as arrival waits. **Switches:** RX_VK_TEX_AHEAD=2 is opt-in and widens the existing mode 1 policy. **Reviewed:** 78143225 against 0de25c3e, with exact Graphify maps and source verification. Read texture-ahead loading, admission, allocation and source checks. Nothing run. **Fixed in this review:** Nothing. **Checked and fine:** Mode 2 bypasses the membership filter; mode 1 retains it. The pre-allocation budget gate remains in place. **Blocking** 1. `gpu/vk/vk_texture.cpp:3011`: the path still loads and populates the previous-draw set before mode and room checks. Mode 2 then ignores it, so even an empty or budget-blocked queue can pay the file read and startup cost. Skip loading in mode 2 and add a seeded-file test that confirms mode 1 still reads it. 2. `gpu/vk/vk_texture.cpp:3012,3023`: the check sees current use plus reserve before creating the candidate; the allocation is charged afterward. Treat it as a soft preflight in the description or account for candidate cost if promising a hard reserve. Establish the real cost with matched exact-head A/B evidence before merge. **Testing:** The head has no gameplay or measurement receipt and the description and Guide screenshots checks are red. Record current-main comparisons and the actual scope of completed game-free checks after rebase. **Overlaps and merge order:** https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/1810 changes the same texture-ahead loop. Preserve its pending-move skip if both land; measure independently first, then test both switches together.
@ -3000,3 +3015,3 @@
}
if (!ahead_queue.take(now, &t)) break;
known = ahead_drawn.count(ahead_signature(t.d, t.L, t.gpu)) != 0;
known = ahead_all() || ahead_drawn.count(ahead_signature(t.d, t.L, t.gpu)) != 0;
Member

This bypasses the learned-list lookup but the unconditional ahead_load above still opens and populates it. Guard the load and test mode 2 with a seeded list.

This bypasses the learned-list lookup but the unconditional ahead_load above still opens and populates it. Guard the load and test mode 2 with a seeded list.
HumanBean added this to the WIP project 2026-11-20 07:10:29 +00:00
GummiNaomi force-pushed perf/tex-ahead-every from 78143225a2
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 12s
CI / Workflow lint (actionlint) (pull_request) Successful in 16s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 14s
Third-party parts / As their notes say (pull_request) Successful in 19s
Website / Website on Cloudflare (pull_request_target) Has been skipped
PR description / Description follows the template (pull_request_target) Failing after 5s
to e9d6b5ddb4
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Failing after 17s
CI / Conventional Commit title (pull_request) Successful in 13s
CI / Workflow lint (actionlint) (pull_request) Successful in 19s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 17s
Website / Website on Cloudflare (pull_request_target) Has been skipped
beanstalk/steamdeck 0/1 pass, 0 running, 2 queued
beanstalk/approval approved by HumanBean (reviewed)
beanstalk/windows+vram8gb 0/1 pass, 1 running, 1 queued
beanstalk/windows+vram12gb 0/1 pass, 0 running, 2 queued
Third-party parts / As their notes say (pull_request) Successful in 26s
CI / Website build and checks (pull_request) Has been cancelled
2026-11-20 07:46:28 +00:00
Compare
Owner

Let's see if this is better for all and maybe replace the old one (-1)?

@Beanstalk

Let's see if this is better for all and maybe replace the old one (-1)? @Beanstalk
Member
Beanstalk test farm · an earlier run: commit e9d6b5d (the newer results are in a comment below)

Commit e9d6b5d · tier reviewed · approved by HumanBean · dashboard

What will be tested · AI plan (gpt-6.1-terra)

On Windows 8 GB and 12 GB GPUs and Steam Deck, this plan compares texture-ahead modes 0, 1 and 2 on main and the PR during two teleport walks through Market, Kakariko Village and Hyrule Field. The mode axis checks the new policy against both off and the previous policy; each tester takes roughly 33 minutes. It measures whole-run hitches, frame times, peak VRAM and load times, and compares arrival screenshots. Crashes, out-of-memory or device-lost messages, visual differences, excessive regressions against main, or more PR-side hitches in mode 2 than in mode 0 fail the plan.

  • Where: windows+vram8gb (8 GB VRAM, 1-2 reports), windows+vram12gb (12 GB VRAM, 1-2 reports), steamdeck (1-2 reports).
  • Matrix: 3 cells, each on main and the PR: RX_VK_TEX_AHEAD 0 or 1 or 2.
  • teleport-walk: from the repository's save kokiri-start, then 2 times: teleport to Market, move the stick, teleport to Kakariko Village, teleport to Hyrule Field, take the picture field to compare with main, measure field for 20 s (after 0 s warm-up); 1 run on each side for each of 3 cells (3 game starts a run).
  • Measured: crashes, frame times, stutters and hitches, VRAM use, load times, the game's log.
  • The PR fails if: it crashes; or the PR's hitches over 50 ms in the whole run with RX_VK_TEX_AHEAD=2 is more than 0% above RX_VK_TEX_AHEAD=0; or a picture differs from main by more than 5%; or peak VRAM is more than 15% above main; or longest load is more than 25% above main; or frame time (p99) is more than 15% above main; or forbidden log lines is above 0.
  • Time: about 38 min of game runs per tester (6 game starts), plus the two builds.

Started at 07:49 UTC.

Estimate: done in about 4 h 30 min, from the queue and the testers online now; some jobs can't start yet.

Target Reports Crash Frame p99 (ms) vs main VRAM peak (MB) vs main Result
windows+vram8gb 0/1 — — — running
windows+vram12gb 0/1 — — — queued
steamdeck 0/1 — — — queued
Scenario · matrix cell Target Frame p99 (ms) vs main Result
teleport-walk · RX_VK_TEX_AHEAD=0 windows+vram8gb — —
teleport-walk · RX_VK_TEX_AHEAD=0 windows+vram12gb — —
teleport-walk · RX_VK_TEX_AHEAD=0 steamdeck — —
teleport-walk · RX_VK_TEX_AHEAD=1 windows+vram8gb — —
teleport-walk · RX_VK_TEX_AHEAD=1 windows+vram12gb — —
teleport-walk · RX_VK_TEX_AHEAD=1 steamdeck — —
teleport-walk · RX_VK_TEX_AHEAD=2 windows+vram8gb — —
teleport-walk · RX_VK_TEX_AHEAD=2 windows+vram12gb — —
teleport-walk · RX_VK_TEX_AHEAD=2 steamdeck — —
<!-- beanstalk:previous --> <details><summary>Beanstalk test farm · an earlier run: commit <code>e9d6b5d</code> (the newer results are in a comment below)</summary> Commit `e9d6b5d` · tier **reviewed** · approved by HumanBean · [dashboard](https://beanstalk.magicbean.dev/prs/1813) **What will be tested** · AI plan (`gpt-6.1-terra`) On Windows 8 GB and 12 GB GPUs and Steam Deck, this plan compares texture-ahead modes 0, 1 and 2 on main and the PR during two teleport walks through Market, Kakariko Village and Hyrule Field. The mode axis checks the new policy against both off and the previous policy; each tester takes roughly 33 minutes. It measures whole-run hitches, frame times, peak VRAM and load times, and compares arrival screenshots. Crashes, out-of-memory or device-lost messages, visual differences, excessive regressions against main, or more PR-side hitches in mode 2 than in mode 0 fail the plan. - **Where:** windows+vram8gb (8 GB VRAM, 1-2 reports), windows+vram12gb (12 GB VRAM, 1-2 reports), steamdeck (1-2 reports). - **Matrix:** 3 cells, each on main and the PR: RX_VK_TEX_AHEAD `0` or `1` or `2`. - **teleport-walk:** from the repository's save `kokiri-start`, then 2 times: teleport to Market, move the stick, teleport to Kakariko Village, teleport to Hyrule Field, take the picture field to compare with main, measure field for 20 s (after 0 s warm-up); 1 run on each side for each of 3 cells (3 game starts a run). - **Measured:** crashes, frame times, stutters and hitches, VRAM use, load times, the game's log. - **The PR fails if:** it crashes; or the PR's hitches over 50 ms in the whole run with RX_VK_TEX_AHEAD=2 is more than 0% above RX_VK_TEX_AHEAD=0; or a picture differs from main by more than 5%; or peak VRAM is more than 15% above main; or longest load is more than 25% above main; or frame time (p99) is more than 15% above main; or forbidden log lines is above 0. - **Time:** about 38 min of game runs per tester (6 game starts), plus the two builds. Started at 07:49 UTC. **Estimate:** done in about 4 h 30 min, from the queue and the testers online now; some jobs can't start yet. | Target | Reports | Crash | Frame p99 (ms) vs main | VRAM peak (MB) vs main | Result | | --- | --- | --- | --- | --- | --- | | windows+vram8gb | 0/1 | — | — | — | running | | windows+vram12gb | 0/1 | — | — | — | queued | | steamdeck | 0/1 | — | — | — | queued | | Scenario · matrix cell | Target | Frame p99 (ms) vs main | Result | | --- | --- | --- | --- | | teleport-walk · `RX_VK_TEX_AHEAD=0` | windows+vram8gb | — | — | | teleport-walk · `RX_VK_TEX_AHEAD=0` | windows+vram12gb | — | — | | teleport-walk · `RX_VK_TEX_AHEAD=0` | steamdeck | — | — | | teleport-walk · `RX_VK_TEX_AHEAD=1` | windows+vram8gb | — | — | | teleport-walk · `RX_VK_TEX_AHEAD=1` | windows+vram12gb | — | — | | teleport-walk · `RX_VK_TEX_AHEAD=1` | steamdeck | — | — | | teleport-walk · `RX_VK_TEX_AHEAD=2` | windows+vram8gb | — | — | | teleport-walk · `RX_VK_TEX_AHEAD=2` | windows+vram12gb | — | — | | teleport-walk · `RX_VK_TEX_AHEAD=2` | steamdeck | — | — | </details>
GummiNaomi force-pushed perf/tex-ahead-every from e9d6b5ddb4
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
PR description / Description follows the template (pull_request_target) Failing after 17s
CI / Conventional Commit title (pull_request) Successful in 13s
CI / Workflow lint (actionlint) (pull_request) Successful in 19s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 17s
Website / Website on Cloudflare (pull_request_target) Has been skipped
beanstalk/steamdeck 0/1 pass, 0 running, 2 queued
beanstalk/approval approved by HumanBean (reviewed)
beanstalk/windows+vram8gb 0/1 pass, 1 running, 1 queued
beanstalk/windows+vram12gb 0/1 pass, 0 running, 2 queued
Third-party parts / As their notes say (pull_request) Successful in 26s
CI / Website build and checks (pull_request) Has been cancelled
to 415777aa95
Some checks failed
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
Website / Website on Cloudflare (pull_request) Has been skipped
CI / Conventional Commit title (pull_request) Successful in 12s
CI / Workflow lint (actionlint) (pull_request) Successful in 19s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 15s
Third-party parts / As their notes say (pull_request) Successful in 18s
beanstalk/approval not approved yet
PR description / Description follows the template (pull_request_target) Failing after 8s
CI / Website build and checks (pull_request) Has been cancelled
2026-11-20 07:51:10 +00:00
Compare
Member

Beanstalk test farm · ❌ failed on windows+vram8gb, windows+vram12gb, steamdeck

Commit 3af1d15 · approved by GummiNaomi (reviewed) · dashboard

Started at 2026-11-20 13:14 UTC.

Target Result Reports VRAM peak (MB) Crashes
windows+vram8gb (8 GB VRAM) ❌ fail (an outlier) 2 1 2 6,214 → 6,150 (−1%) 2 PR runs
windows+vram12gb (12 GB VRAM) ❌ fail 2 1 2 8,601 → 8,551 (−1%) none
steamdeck ❌ fail 1 1 3,283 → 3,144 (−4%) none

teleport-tour · frame time p99 (ms), main → PR at each level

Target RX_VK_TEX_AHEAD=0 (reference) RX_VK_TEX_AHEAD=1 RX_VK_TEX_AHEAD=2
windows+vram8gb (8 GB VRAM) 46.9 → 40.1 (−14%) 50.8 → 65.5 (+29%) 50.5 → 26.6 (−47%)
windows+vram12gb (12 GB VRAM) 28.8 → 26.0 (−10%) 41.1 → 52.8 (+29%) 44.3 → 41.7 (−6%)
steamdeck 464.3 → 238.0 (−49%) 122.2 → 113.7 (−7%) 157.4 → 112.9 (−28%)

Why it failed

  • windows+vram8gb: teleport-tour · RX_VK_TEX_AHEAD=0 screenshot.max_diff: limit 0.1
  • windows+vram8gb: teleport-tour · RX_VK_TEX_AHEAD=2 crash: 1 PR run crashed
  • windows+vram8gb: teleport-tour · RX_VK_TEX_AHEAD=2 load_time_max_s.max_delta_pct: 0.0 → 0.5 (+1075.0%); limit 30.0
  • windows+vram8gb: teleport-tour · RX_VK_TEX_AHEAD=2 log_forbidden_hits.max: 1.0 → 1.0; limit 0.0
  • windows+vram12gb: teleport-tour · RX_VK_TEX_AHEAD=0 screenshot.max_diff: limit 0.1
  • windows+vram12gb: teleport-tour · RX_VK_TEX_AHEAD=0 kokiri-first.vram_peak_mb.max_delta_pct: 5,553 → 9,133 (+64.5%); limit 15.0
  • windows+vram12gb: teleport-tour · RX_VK_TEX_AHEAD=1 kokiri-first.vram_peak_mb.max_delta_pct: 6,097 → 7,269 (+19.2%); limit 15.0
  • windows+vram12gb: teleport-tour · RX_VK_TEX_AHEAD=2 kokiri-first.vram_peak_mb.max_delta_pct: 7,587 → 8,749 (+15.3%); limit 15.0
  • and 2 more, in the reports

Couldn'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+vram12gb, report 1: the pull request's pack failed (exit 1): Failed to find real location of Z:\tools\python\python.exe
What was tested

What will be tested · AI plan (gpt-6.1-terra)

The plan runs the requested teleport tour from Kokiri Forest through Market, Kakariko and Hyrule Field and back on Windows 8 GB, Windows 12 GB and Steam Deck targets. It compares texture-ahead modes 0, 1 and 2 on main and the PR, measuring post-arrival frame tails, memory, load times and eviction counts if the available logs match the supplied pattern. The plan cannot set the requested frame-trace switch, so individual loading-frame maxima may require follow-up. Each target takes roughly 35 minutes; crashes, device loss, out-of-memory errors, visual mismatches or specified regressions fail the PR.

  • Where: windows+vram8gb (8 GB VRAM, 1-2 reports), windows+vram12gb (12 GB VRAM, 1-2 reports), steamdeck (1-2 reports).
  • Matrix: 3 cells, each on main and the PR: RX_VK_TEX_AHEAD 0 or 1 or 2.
  • teleport-tour: from the repository's save kokiri-start, then teleport to Kokiri Forest, measure kokiri-first for 5 s (after 0 s warm-up), move the stick, teleport to Market, measure market for 5 s (after 0 s warm-up), take the picture market to compare with main, teleport to Kakariko Village, measure kakariko for 5 s (after 0 s warm-up), take the picture kakariko to compare with main, teleport to Hyrule Field, measure field for 5 s (after 0 s warm-up), take the picture field to compare with main, measure kokiri-return for 5 s (after 0 s warm-up), take the picture kokiri-return to compare with main; 1 run on each side for each of 3 cells (3 game starts a run).
  • Measured: crashes, frame times, stutters and hitches, load times, VRAM use, memory and CPU time, the game's log, numbers from the game's log.
  • The PR fails if: it crashes; or a picture differs from main by more than 5%; or peak VRAM is more than 15% above main; or longest load is more than 30% above main; or frame time (p99) is more than 30% above main; or forbidden log lines is above 0.
  • Time: about 33 min of game runs per tester (6 game starts), plus the two builds.
Measurements on `windows+vram8gb`, for the PR description

Verdict: fail on windows+vram8gb (2 reports): teleport-tour screenshot.max_diff; teleport-tour crash; teleport-tour load_time_max_s.max_delta_pct (+1075.0%) failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3070, 621.24.0 AMD Ryzen 5 5600X3D 6-Core Processor, 32 GB RAM; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 2080 SUPER, 621.24.0 Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz, 16 GB RAM
Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (@Beanstalk on this PR runs it again)

Before: main @ 6548881 After: this PR @ 3af1d15
teleport-tour · RX_VK_TEX_AHEAD=0 · field · peak VRAM (MB) 6,214 6,150 (−1%)
teleport-tour · RX_VK_TEX_AHEAD=0 · market · peak VRAM (MB) 6,313 6,231 (−1%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kakariko · peak VRAM (MB) 6,313 6,157 (−2%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kokiri-first · peak VRAM (MB) 6,313 6,231 (−1%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kokiri-return · peak VRAM (MB) 6,313 6,231 (−1%)
teleport-tour · RX_VK_TEX_AHEAD=0 · longest load (s) 6.1 6.0 (−1%)
teleport-tour · RX_VK_TEX_AHEAD=0 · field · frame time (p99) (ms) 46.9 40.1 (−14%)
teleport-tour · RX_VK_TEX_AHEAD=0 · market · frame time (p99) (ms) 124.8 68.9 (−45%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kakariko · frame time (p99) (ms) 55.5 48.7 (−12%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kokiri-first · frame time (p99) (ms) 55.7 71.5 (+28%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kokiri-return · frame time (p99) (ms) 30.6 28.6 (−7%)
teleport-tour · RX_VK_TEX_AHEAD=0 · forbidden log lines 0.0 0.0
teleport-tour · RX_VK_TEX_AHEAD=1 · kokiri-first · peak VRAM (MB) 5,516 5,506 (−0%)
teleport-tour · RX_VK_TEX_AHEAD=1 · longest load (s) 5.6 0.0 (−99%)
teleport-tour · RX_VK_TEX_AHEAD=1 · kokiri-first · frame time (p99) (ms) 50.8 65.5 (+29%)
teleport-tour · RX_VK_TEX_AHEAD=1 · forbidden log lines 0.0 0.0
teleport-tour · RX_VK_TEX_AHEAD=2 · kokiri-first · peak VRAM (MB) 3,384 5,296 (+57%)
teleport-tour · RX_VK_TEX_AHEAD=2 · longest load (s) 1.9 0.5 (−76%)
teleport-tour · RX_VK_TEX_AHEAD=2 · kokiri-first · frame time (p99) (ms) 293.1 26.6 (−91%)
teleport-tour · RX_VK_TEX_AHEAD=2 · forbidden log lines 0.5 0.5 (+0%)

Flags and settings

Flag What it is Values tried Before this PR After this PR Effect (peak VRAM) Helps / doesn't help / untested
RX_VK_TEX_AHEAD switch 0, 1, 2 main's default the PR's default teleport-tour, the PR at 0 → at 2: 6,150 → 5,296 (−14%) helps

As Markdown for the PR description:

## Measurements

**Verdict:** fail on windows+vram8gb (2 reports): `teleport-tour screenshot.max_diff`; `teleport-tour crash`; `teleport-tour load_time_max_s.max_delta_pct (+1075.0%)` failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3070, 621.24.0` `AMD Ryzen 5 5600X3D 6-Core Processor, 32 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 2080 SUPER, 621.24.0` `Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz, 16 GB RAM`  
Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (`@Beanstalk` on this PR runs it again)

| | Before: main @ 6548881 | After: this PR @ 3af1d15 |
|---|---|---|
| teleport-tour · `RX_VK_TEX_AHEAD=0` · field · peak VRAM (MB) | 6,214 | 6,150 (−1%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · market · peak VRAM (MB) | 6,313 | 6,231 (−1%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · peak VRAM (MB) | 6,313 | 6,157 (−2%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · peak VRAM (MB) | 6,313 | 6,231 (−1%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · peak VRAM (MB) | 6,313 | 6,231 (−1%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · longest load (s) | 6.1 | 6.0 (−1%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · field · frame time (p99) (ms) | 46.9 | 40.1 (−14%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · market · frame time (p99) (ms) | 124.8 | 68.9 (−45%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · frame time (p99) (ms) | 55.5 | 48.7 (−12%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · frame time (p99) (ms) | 55.7 | 71.5 (+28%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · frame time (p99) (ms) | 30.6 | 28.6 (−7%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · forbidden log lines | 0.0 | 0.0 |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · peak VRAM (MB) | 5,516 | 5,506 (−0%) |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · longest load (s) | 5.6 | 0.0 (−99%) |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · frame time (p99) (ms) | 50.8 | 65.5 (+29%) |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · forbidden log lines | 0.0 | 0.0 |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · peak VRAM (MB) | 3,384 | 5,296 (+57%) |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · longest load (s) | 1.9 | 0.5 (−76%) |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · frame time (p99) (ms) | 293.1 | 26.6 (−91%) |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · forbidden log lines | 0.5 | 0.5 (+0%) |

**Flags and settings**

| Flag | What it is | Values tried | Before this PR | After this PR | Effect (peak VRAM) | Helps / doesn't help / untested |
|---|---|---|---|---|---|---|
| `RX_VK_TEX_AHEAD` | switch | `0`, `1`, `2` | main's default | the PR's default | teleport-tour, the PR at 0 → at 2: 6,150 → 5,296 (−14%) | helps |
Measurements on `windows+vram12gb`, for the PR description

Verdict: fail on windows+vram12gb (2 reports): teleport-tour screenshot.max_diff; teleport-tour kokiri-first.vram_peak_mb.max_delta_pct (+64.5%); teleport-tour kokiri-first.vram_peak_mb.max_delta_pct (+19.2%) failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 4070 Ti, 621.14.0 AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM; 2: windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3080 Ti, 621.24.0 AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM
Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (@Beanstalk on this PR runs it again)

Before: main @ 6548881 After: this PR @ 3af1d15
teleport-tour · RX_VK_TEX_AHEAD=0 · field · peak VRAM (MB) 8,601 8,551 (−1%)
teleport-tour · RX_VK_TEX_AHEAD=0 · market · peak VRAM (MB) 8,601 9,133 (+6%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kakariko · peak VRAM (MB) 8,601 9,133 (+6%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kokiri-first · peak VRAM (MB) 5,553 9,133 (+64%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kokiri-return · peak VRAM (MB) 8,601 8,222 (−4%)
teleport-tour · RX_VK_TEX_AHEAD=0 · longest load (s) 5.4 5.3 (−2%)
teleport-tour · RX_VK_TEX_AHEAD=0 · field · frame time (p99) (ms) 28.8 26.0 (−10%)
teleport-tour · RX_VK_TEX_AHEAD=0 · market · frame time (p99) (ms) 49.8 50.7 (+2%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kakariko · frame time (p99) (ms) 36.3 26.0 (−28%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kokiri-first · frame time (p99) (ms) 60.7 59.8 (−1%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kokiri-return · frame time (p99) (ms) 22.9 22.8 (−0%)
teleport-tour · RX_VK_TEX_AHEAD=0 · forbidden log lines 0.0 0.0
teleport-tour · RX_VK_TEX_AHEAD=1 · field · peak VRAM (MB) 9,286 —
teleport-tour · RX_VK_TEX_AHEAD=1 · market · peak VRAM (MB) 8,304 —
teleport-tour · RX_VK_TEX_AHEAD=1 · kakariko · peak VRAM (MB) 8,527 —
teleport-tour · RX_VK_TEX_AHEAD=1 · kokiri-first · peak VRAM (MB) 6,097 7,269 (+19%)
teleport-tour · RX_VK_TEX_AHEAD=1 · kokiri-return · peak VRAM (MB) 9,286 —
teleport-tour · RX_VK_TEX_AHEAD=1 · longest load (s) 5.2 0.0 (−99%)
teleport-tour · RX_VK_TEX_AHEAD=1 · field · frame time (p99) (ms) 30.4 —
teleport-tour · RX_VK_TEX_AHEAD=1 · market · frame time (p99) (ms) 47.8 —
teleport-tour · RX_VK_TEX_AHEAD=1 · kakariko · frame time (p99) (ms) 21.6 —
teleport-tour · RX_VK_TEX_AHEAD=1 · kokiri-first · frame time (p99) (ms) 41.1 52.8 (+29%)
teleport-tour · RX_VK_TEX_AHEAD=1 · kokiri-return · frame time (p99) (ms) 21.1 —
teleport-tour · RX_VK_TEX_AHEAD=1 · forbidden log lines 0.0 0.0
teleport-tour · RX_VK_TEX_AHEAD=2 · kokiri-first · peak VRAM (MB) 7,587 8,749 (+15%)
teleport-tour · RX_VK_TEX_AHEAD=2 · longest load (s) 0.0 0.0 (−25%)
teleport-tour · RX_VK_TEX_AHEAD=2 · kokiri-first · frame time (p99) (ms) 44.3 41.7 (−6%)
teleport-tour · RX_VK_TEX_AHEAD=2 · forbidden log lines 0.0 0.0

Flags and settings

Flag What it is Values tried Before this PR After this PR Effect (peak VRAM) Helps / doesn't help / untested
RX_VK_TEX_AHEAD switch 0, 1, 2 main's default the PR's default teleport-tour, the PR at 0 → at 2: 8,551 → 8,749 (+2%) doesn't help

As Markdown for the PR description:

## Measurements

**Verdict:** fail on windows+vram12gb (2 reports): `teleport-tour screenshot.max_diff`; `teleport-tour kokiri-first.vram_peak_mb.max_delta_pct (+64.5%)`; `teleport-tour kokiri-first.vram_peak_mb.max_delta_pct (+19.2%)` failed.

Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070 Ti, 621.14.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3080 Ti, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM`  
Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (`@Beanstalk` on this PR runs it again)

| | Before: main @ 6548881 | After: this PR @ 3af1d15 |
|---|---|---|
| teleport-tour · `RX_VK_TEX_AHEAD=0` · field · peak VRAM (MB) | 8,601 | 8,551 (−1%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · market · peak VRAM (MB) | 8,601 | 9,133 (+6%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · peak VRAM (MB) | 8,601 | 9,133 (+6%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · peak VRAM (MB) | 5,553 | 9,133 (+64%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · peak VRAM (MB) | 8,601 | 8,222 (−4%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · longest load (s) | 5.4 | 5.3 (−2%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · field · frame time (p99) (ms) | 28.8 | 26.0 (−10%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · market · frame time (p99) (ms) | 49.8 | 50.7 (+2%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · frame time (p99) (ms) | 36.3 | 26.0 (−28%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · frame time (p99) (ms) | 60.7 | 59.8 (−1%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · frame time (p99) (ms) | 22.9 | 22.8 (−0%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · forbidden log lines | 0.0 | 0.0 |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · field · peak VRAM (MB) | 9,286 | — |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · market · peak VRAM (MB) | 8,304 | — |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · kakariko · peak VRAM (MB) | 8,527 | — |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · peak VRAM (MB) | 6,097 | 7,269 (+19%) |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-return · peak VRAM (MB) | 9,286 | — |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · longest load (s) | 5.2 | 0.0 (−99%) |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · field · frame time (p99) (ms) | 30.4 | — |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · market · frame time (p99) (ms) | 47.8 | — |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · kakariko · frame time (p99) (ms) | 21.6 | — |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · frame time (p99) (ms) | 41.1 | 52.8 (+29%) |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-return · frame time (p99) (ms) | 21.1 | — |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · forbidden log lines | 0.0 | 0.0 |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · peak VRAM (MB) | 7,587 | 8,749 (+15%) |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · longest load (s) | 0.0 | 0.0 (−25%) |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · frame time (p99) (ms) | 44.3 | 41.7 (−6%) |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · forbidden log lines | 0.0 | 0.0 |

**Flags and settings**

| Flag | What it is | Values tried | Before this PR | After this PR | Effect (peak VRAM) | Helps / doesn't help / untested |
|---|---|---|---|---|---|---|
| `RX_VK_TEX_AHEAD` | switch | `0`, `1`, `2` | main's default | the PR's default | teleport-tour, the PR at 0 → at 2: 8,551 → 8,749 (+2%) | doesn't help |
Measurements on `steamdeck`, for the PR description

Verdict: fail on steamdeck (1 report): teleport-tour screenshot.max_diff; teleport-tour load_time_max_s.max_delta_pct (+24200.0%) failed.

Hardware: 1: linux+amd+radv+igpu+handheld AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99 AMD Custom APU 0405, 12 GB RAM
Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (@Beanstalk on this PR runs it again)

Before: main @ 6548881 After: this PR @ 3af1d15
teleport-tour · RX_VK_TEX_AHEAD=0 · market · peak VRAM (MB) 3,283 3,144 (−4%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kokiri-first · peak VRAM (MB) 3,283 3,065 (−7%)
teleport-tour · RX_VK_TEX_AHEAD=0 · longest load (s) 10.1 10.1 (−0%)
teleport-tour · RX_VK_TEX_AHEAD=0 · market · frame time (p99) (ms) 464.3 238.0 (−49%)
teleport-tour · RX_VK_TEX_AHEAD=0 · kokiri-first · frame time (p99) (ms) 130.2 107.7 (−17%)
teleport-tour · RX_VK_TEX_AHEAD=0 · forbidden log lines 0.0 0.0
teleport-tour · RX_VK_TEX_AHEAD=1 · kokiri-first · peak VRAM (MB) 3,182 2,942 (−8%)
teleport-tour · RX_VK_TEX_AHEAD=1 · market · peak VRAM (MB) — 2,942
teleport-tour · RX_VK_TEX_AHEAD=1 · longest load (s) 0.0 9.7 (+24200%)
teleport-tour · RX_VK_TEX_AHEAD=1 · kokiri-first · frame time (p99) (ms) 122.2 113.7 (−7%)
teleport-tour · RX_VK_TEX_AHEAD=1 · market · frame time (p99) (ms) — 312.2
teleport-tour · RX_VK_TEX_AHEAD=1 · forbidden log lines 0.0 0.0
teleport-tour · RX_VK_TEX_AHEAD=2 · kokiri-first · peak VRAM (MB) 2,854 3,118 (+9%)
teleport-tour · RX_VK_TEX_AHEAD=2 · longest load (s) 0.1 0.1 (−44%)
teleport-tour · RX_VK_TEX_AHEAD=2 · kokiri-first · frame time (p99) (ms) 157.4 112.9 (−28%)
teleport-tour · RX_VK_TEX_AHEAD=2 · forbidden log lines 0.0 0.0

Flags and settings

Flag What it is Values tried Before this PR After this PR Effect (peak VRAM) Helps / doesn't help / untested
RX_VK_TEX_AHEAD switch 0, 1, 2 main's default the PR's default teleport-tour, the PR at 0 → at 2: 3,144 → 3,118 (−1%) no clear effect

As Markdown for the PR description:

## Measurements

**Verdict:** fail on steamdeck (1 report): `teleport-tour screenshot.max_diff`; `teleport-tour load_time_max_s.max_delta_pct (+24200.0%)` failed.

Hardware: 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM`  
Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (`@Beanstalk` on this PR runs it again)

| | Before: main @ 6548881 | After: this PR @ 3af1d15 |
|---|---|---|
| teleport-tour · `RX_VK_TEX_AHEAD=0` · market · peak VRAM (MB) | 3,283 | 3,144 (−4%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · peak VRAM (MB) | 3,283 | 3,065 (−7%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · longest load (s) | 10.1 | 10.1 (−0%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · market · frame time (p99) (ms) | 464.3 | 238.0 (−49%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · frame time (p99) (ms) | 130.2 | 107.7 (−17%) |
| teleport-tour · `RX_VK_TEX_AHEAD=0` · forbidden log lines | 0.0 | 0.0 |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · peak VRAM (MB) | 3,182 | 2,942 (−8%) |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · market · peak VRAM (MB) | — | 2,942 |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · longest load (s) | 0.0 | 9.7 (+24200%) |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · frame time (p99) (ms) | 122.2 | 113.7 (−7%) |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · market · frame time (p99) (ms) | — | 312.2 |
| teleport-tour · `RX_VK_TEX_AHEAD=1` · forbidden log lines | 0.0 | 0.0 |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · peak VRAM (MB) | 2,854 | 3,118 (+9%) |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · longest load (s) | 0.1 | 0.1 (−44%) |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · frame time (p99) (ms) | 157.4 | 112.9 (−28%) |
| teleport-tour · `RX_VK_TEX_AHEAD=2` · forbidden log lines | 0.0 | 0.0 |

**Flags and settings**

| Flag | What it is | Values tried | Before this PR | After this PR | Effect (peak VRAM) | Helps / doesn't help / untested |
|---|---|---|---|---|---|---|
| `RX_VK_TEX_AHEAD` | switch | `0`, `1`, `2` | main's default | the PR's default | teleport-tour, the PR at 0 → at 2: 3,144 → 3,118 (−1%) | no clear effect |
The testers' hardware
Target Report Ran on
windows+vram8gb 1 windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3070, 621.24.0 AMD Ryzen 5 5600X3D 6-Core Processor, 32 GB RAM
windows+vram8gb 2 windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 2080 SUPER, 621.24.0 Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz, 16 GB RAM
windows+vram12gb 1 windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 4070 Ti, 621.14.0 AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM
windows+vram12gb 2 windows+nvidia+nvidia-proprietary+dgpu NVIDIA GeForce RTX 3080 Ti, 621.24.0 AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM
steamdeck 1 linux+amd+radv+igpu+handheld AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99 AMD Custom APU 0405, 12 GB RAM
<!-- beanstalk:sticky --> ### Beanstalk test farm · ❌ failed on windows+vram8gb, windows+vram12gb, steamdeck Commit `3af1d15` · approved by GummiNaomi (reviewed) · [dashboard](https://beanstalk.magicbean.dev/prs/1813) Started at 2026-11-20 13:14 UTC. | Target | Result | Reports | VRAM peak (MB) | Crashes | | --- | --- | --- | --- | --- | | `windows+vram8gb` (8 GB VRAM) | ❌ fail (an outlier) | 2 [1](https://beanstalk.magicbean.dev/reports/rp_521666ea63125b4b) [2](https://beanstalk.magicbean.dev/reports/rp_bface65d371268f2) | 6,214 → 6,150 (−1%) | 2 PR runs | | `windows+vram12gb` (12 GB VRAM) | ❌ fail | 2 [1](https://beanstalk.magicbean.dev/reports/rp_746e0328efc1233e) [2](https://beanstalk.magicbean.dev/reports/rp_b85980f0865152bd) | 8,601 → 8,551 (−1%) | none | | `steamdeck` | ❌ fail | 1 [1](https://beanstalk.magicbean.dev/reports/rp_0d035cd77068c69c) | 3,283 → 3,144 (−4%) | none | **`teleport-tour`** · frame time p99 (ms), main → PR at each level | Target | `RX_VK_TEX_AHEAD=0` (reference) | `RX_VK_TEX_AHEAD=1` | `RX_VK_TEX_AHEAD=2` | | --- | --- | --- | --- | | `windows+vram8gb` (8 GB VRAM) | 46.9 → 40.1 (−14%) | 50.8 → 65.5 (+29%) | 50.5 → 26.6 (−47%) | | `windows+vram12gb` (12 GB VRAM) | 28.8 → 26.0 (−10%) | 41.1 → 52.8 (+29%) | 44.3 → 41.7 (−6%) | | `steamdeck` | 464.3 → 238.0 (−49%) | 122.2 → 113.7 (−7%) | 157.4 → 112.9 (−28%) | **Why it failed** - `windows+vram8gb`: `teleport-tour · RX_VK_TEX_AHEAD=0` `screenshot.max_diff`: `limit 0.1` - `windows+vram8gb`: `teleport-tour · RX_VK_TEX_AHEAD=2` `crash`: `1 PR run crashed` - `windows+vram8gb`: `teleport-tour · RX_VK_TEX_AHEAD=2` `load_time_max_s.max_delta_pct`: `0.0 → 0.5 (+1075.0%); limit 30.0` - `windows+vram8gb`: `teleport-tour · RX_VK_TEX_AHEAD=2` `log_forbidden_hits.max`: `1.0 → 1.0; limit 0.0` - `windows+vram12gb`: `teleport-tour · RX_VK_TEX_AHEAD=0` `screenshot.max_diff`: `limit 0.1` - `windows+vram12gb`: `teleport-tour · RX_VK_TEX_AHEAD=0` `kokiri-first.vram_peak_mb.max_delta_pct`: `5,553 → 9,133 (+64.5%); limit 15.0` - `windows+vram12gb`: `teleport-tour · RX_VK_TEX_AHEAD=1` `kokiri-first.vram_peak_mb.max_delta_pct`: `6,097 → 7,269 (+19.2%); limit 15.0` - `windows+vram12gb`: `teleport-tour · RX_VK_TEX_AHEAD=2` `kokiri-first.vram_peak_mb.max_delta_pct`: `7,587 → 8,749 (+15.3%); limit 15.0` - and 2 more, in the reports **Couldn'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+vram12gb`, [report 1](https://beanstalk.magicbean.dev/reports/rp_746e0328efc1233e): `the pull request's pack failed (exit 1): Failed to find real location of Z:\tools\python\python.exe` <details><summary>What was tested</summary> **What will be tested** · AI plan (`gpt-6.1-terra`) The plan runs the requested teleport tour from Kokiri Forest through Market, Kakariko and Hyrule Field and back on Windows 8 GB, Windows 12 GB and Steam Deck targets. It compares texture-ahead modes 0, 1 and 2 on main and the PR, measuring post-arrival frame tails, memory, load times and eviction counts if the available logs match the supplied pattern. The plan cannot set the requested frame-trace switch, so individual loading-frame maxima may require follow-up. Each target takes roughly 35 minutes; crashes, device loss, out-of-memory errors, visual mismatches or specified regressions fail the PR. - **Where:** windows+vram8gb (8 GB VRAM, 1-2 reports), windows+vram12gb (12 GB VRAM, 1-2 reports), steamdeck (1-2 reports). - **Matrix:** 3 cells, each on main and the PR: RX_VK_TEX_AHEAD `0` or `1` or `2`. - **teleport-tour:** from the repository's save `kokiri-start`, then teleport to Kokiri Forest, measure kokiri-first for 5 s (after 0 s warm-up), move the stick, teleport to Market, measure market for 5 s (after 0 s warm-up), take the picture market to compare with main, teleport to Kakariko Village, measure kakariko for 5 s (after 0 s warm-up), take the picture kakariko to compare with main, teleport to Hyrule Field, measure field for 5 s (after 0 s warm-up), take the picture field to compare with main, measure kokiri-return for 5 s (after 0 s warm-up), take the picture kokiri-return to compare with main; 1 run on each side for each of 3 cells (3 game starts a run). - **Measured:** crashes, frame times, stutters and hitches, load times, VRAM use, memory and CPU time, the game's log, numbers from the game's log. - **The PR fails if:** it crashes; or a picture differs from main by more than 5%; or peak VRAM is more than 15% above main; or longest load is more than 30% above main; or frame time (p99) is more than 30% above main; or forbidden log lines is above 0. - **Time:** about 33 min of game runs per tester (6 game starts), plus the two builds. </details> <details><summary>Measurements on `windows+vram8gb`, for the PR description</summary> **Verdict:** fail on windows+vram8gb (2 reports): `teleport-tour screenshot.max_diff`; `teleport-tour crash`; `teleport-tour load_time_max_s.max_delta_pct (+1075.0%)` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3070, 621.24.0` `AMD Ryzen 5 5600X3D 6-Core Processor, 32 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 2080 SUPER, 621.24.0` `Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz, 16 GB RAM` Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (`@Beanstalk` on this PR runs it again) | | Before: main @ 6548881 | After: this PR @ 3af1d15 | |---|---|---| | teleport-tour · `RX_VK_TEX_AHEAD=0` · field · peak VRAM (MB) | 6,214 | 6,150 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · peak VRAM (MB) | 6,313 | 6,231 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · peak VRAM (MB) | 6,313 | 6,157 (−2%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · peak VRAM (MB) | 6,313 | 6,231 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · peak VRAM (MB) | 6,313 | 6,231 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · longest load (s) | 6.1 | 6.0 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · field · frame time (p99) (ms) | 46.9 | 40.1 (−14%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · frame time (p99) (ms) | 124.8 | 68.9 (−45%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · frame time (p99) (ms) | 55.5 | 48.7 (−12%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · frame time (p99) (ms) | 55.7 | 71.5 (+28%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · frame time (p99) (ms) | 30.6 | 28.6 (−7%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · peak VRAM (MB) | 5,516 | 5,506 (−0%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · longest load (s) | 5.6 | 0.0 (−99%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · frame time (p99) (ms) | 50.8 | 65.5 (+29%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · peak VRAM (MB) | 3,384 | 5,296 (+57%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · longest load (s) | 1.9 | 0.5 (−76%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · frame time (p99) (ms) | 293.1 | 26.6 (−91%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · forbidden log lines | 0.5 | 0.5 (+0%) | **Flags and settings** | Flag | What it is | Values tried | Before this PR | After this PR | Effect (peak VRAM) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_TEX_AHEAD` | switch | `0`, `1`, `2` | main's default | the PR's default | teleport-tour, the PR at 0 → at 2: 6,150 → 5,296 (−14%) | helps | As Markdown for the PR description: ```markdown ## Measurements **Verdict:** fail on windows+vram8gb (2 reports): `teleport-tour screenshot.max_diff`; `teleport-tour crash`; `teleport-tour load_time_max_s.max_delta_pct (+1075.0%)` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3070, 621.24.0` `AMD Ryzen 5 5600X3D 6-Core Processor, 32 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 2080 SUPER, 621.24.0` `Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz, 16 GB RAM` Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (`@Beanstalk` on this PR runs it again) | | Before: main @ 6548881 | After: this PR @ 3af1d15 | |---|---|---| | teleport-tour · `RX_VK_TEX_AHEAD=0` · field · peak VRAM (MB) | 6,214 | 6,150 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · peak VRAM (MB) | 6,313 | 6,231 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · peak VRAM (MB) | 6,313 | 6,157 (−2%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · peak VRAM (MB) | 6,313 | 6,231 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · peak VRAM (MB) | 6,313 | 6,231 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · longest load (s) | 6.1 | 6.0 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · field · frame time (p99) (ms) | 46.9 | 40.1 (−14%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · frame time (p99) (ms) | 124.8 | 68.9 (−45%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · frame time (p99) (ms) | 55.5 | 48.7 (−12%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · frame time (p99) (ms) | 55.7 | 71.5 (+28%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · frame time (p99) (ms) | 30.6 | 28.6 (−7%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · peak VRAM (MB) | 5,516 | 5,506 (−0%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · longest load (s) | 5.6 | 0.0 (−99%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · frame time (p99) (ms) | 50.8 | 65.5 (+29%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · peak VRAM (MB) | 3,384 | 5,296 (+57%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · longest load (s) | 1.9 | 0.5 (−76%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · frame time (p99) (ms) | 293.1 | 26.6 (−91%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · forbidden log lines | 0.5 | 0.5 (+0%) | **Flags and settings** | Flag | What it is | Values tried | Before this PR | After this PR | Effect (peak VRAM) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_TEX_AHEAD` | switch | `0`, `1`, `2` | main's default | the PR's default | teleport-tour, the PR at 0 → at 2: 6,150 → 5,296 (−14%) | helps | ``` </details> <details><summary>Measurements on `windows+vram12gb`, for the PR description</summary> **Verdict:** fail on windows+vram12gb (2 reports): `teleport-tour screenshot.max_diff`; `teleport-tour kokiri-first.vram_peak_mb.max_delta_pct (+64.5%)`; `teleport-tour kokiri-first.vram_peak_mb.max_delta_pct (+19.2%)` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070 Ti, 621.14.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3080 Ti, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM` Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (`@Beanstalk` on this PR runs it again) | | Before: main @ 6548881 | After: this PR @ 3af1d15 | |---|---|---| | teleport-tour · `RX_VK_TEX_AHEAD=0` · field · peak VRAM (MB) | 8,601 | 8,551 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · peak VRAM (MB) | 8,601 | 9,133 (+6%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · peak VRAM (MB) | 8,601 | 9,133 (+6%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · peak VRAM (MB) | 5,553 | 9,133 (+64%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · peak VRAM (MB) | 8,601 | 8,222 (−4%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · longest load (s) | 5.4 | 5.3 (−2%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · field · frame time (p99) (ms) | 28.8 | 26.0 (−10%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · frame time (p99) (ms) | 49.8 | 50.7 (+2%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · frame time (p99) (ms) | 36.3 | 26.0 (−28%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · frame time (p99) (ms) | 60.7 | 59.8 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · frame time (p99) (ms) | 22.9 | 22.8 (−0%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=1` · field · peak VRAM (MB) | 9,286 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · market · peak VRAM (MB) | 8,304 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kakariko · peak VRAM (MB) | 8,527 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · peak VRAM (MB) | 6,097 | 7,269 (+19%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-return · peak VRAM (MB) | 9,286 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · longest load (s) | 5.2 | 0.0 (−99%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · field · frame time (p99) (ms) | 30.4 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · market · frame time (p99) (ms) | 47.8 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kakariko · frame time (p99) (ms) | 21.6 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · frame time (p99) (ms) | 41.1 | 52.8 (+29%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-return · frame time (p99) (ms) | 21.1 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · peak VRAM (MB) | 7,587 | 8,749 (+15%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · longest load (s) | 0.0 | 0.0 (−25%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · frame time (p99) (ms) | 44.3 | 41.7 (−6%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · forbidden log lines | 0.0 | 0.0 | **Flags and settings** | Flag | What it is | Values tried | Before this PR | After this PR | Effect (peak VRAM) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_TEX_AHEAD` | switch | `0`, `1`, `2` | main's default | the PR's default | teleport-tour, the PR at 0 → at 2: 8,551 → 8,749 (+2%) | doesn't help | As Markdown for the PR description: ```markdown ## Measurements **Verdict:** fail on windows+vram12gb (2 reports): `teleport-tour screenshot.max_diff`; `teleport-tour kokiri-first.vram_peak_mb.max_delta_pct (+64.5%)`; `teleport-tour kokiri-first.vram_peak_mb.max_delta_pct (+19.2%)` failed. Hardware: 1: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070 Ti, 621.14.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM`; 2: windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3080 Ti, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM` Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (`@Beanstalk` on this PR runs it again) | | Before: main @ 6548881 | After: this PR @ 3af1d15 | |---|---|---| | teleport-tour · `RX_VK_TEX_AHEAD=0` · field · peak VRAM (MB) | 8,601 | 8,551 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · peak VRAM (MB) | 8,601 | 9,133 (+6%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · peak VRAM (MB) | 8,601 | 9,133 (+6%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · peak VRAM (MB) | 5,553 | 9,133 (+64%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · peak VRAM (MB) | 8,601 | 8,222 (−4%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · longest load (s) | 5.4 | 5.3 (−2%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · field · frame time (p99) (ms) | 28.8 | 26.0 (−10%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · frame time (p99) (ms) | 49.8 | 50.7 (+2%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kakariko · frame time (p99) (ms) | 36.3 | 26.0 (−28%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · frame time (p99) (ms) | 60.7 | 59.8 (−1%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-return · frame time (p99) (ms) | 22.9 | 22.8 (−0%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=1` · field · peak VRAM (MB) | 9,286 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · market · peak VRAM (MB) | 8,304 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kakariko · peak VRAM (MB) | 8,527 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · peak VRAM (MB) | 6,097 | 7,269 (+19%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-return · peak VRAM (MB) | 9,286 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · longest load (s) | 5.2 | 0.0 (−99%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · field · frame time (p99) (ms) | 30.4 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · market · frame time (p99) (ms) | 47.8 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kakariko · frame time (p99) (ms) | 21.6 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · frame time (p99) (ms) | 41.1 | 52.8 (+29%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-return · frame time (p99) (ms) | 21.1 | — | | teleport-tour · `RX_VK_TEX_AHEAD=1` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · peak VRAM (MB) | 7,587 | 8,749 (+15%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · longest load (s) | 0.0 | 0.0 (−25%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · frame time (p99) (ms) | 44.3 | 41.7 (−6%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · forbidden log lines | 0.0 | 0.0 | **Flags and settings** | Flag | What it is | Values tried | Before this PR | After this PR | Effect (peak VRAM) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_TEX_AHEAD` | switch | `0`, `1`, `2` | main's default | the PR's default | teleport-tour, the PR at 0 → at 2: 8,551 → 8,749 (+2%) | doesn't help | ``` </details> <details><summary>Measurements on `steamdeck`, for the PR description</summary> **Verdict:** fail on steamdeck (1 report): `teleport-tour screenshot.max_diff`; `teleport-tour load_time_max_s.max_delta_pct (+24200.0%)` failed. Hardware: 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM` Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (`@Beanstalk` on this PR runs it again) | | Before: main @ 6548881 | After: this PR @ 3af1d15 | |---|---|---| | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · peak VRAM (MB) | 3,283 | 3,144 (−4%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · peak VRAM (MB) | 3,283 | 3,065 (−7%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · longest load (s) | 10.1 | 10.1 (−0%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · frame time (p99) (ms) | 464.3 | 238.0 (−49%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · frame time (p99) (ms) | 130.2 | 107.7 (−17%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · peak VRAM (MB) | 3,182 | 2,942 (−8%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · market · peak VRAM (MB) | — | 2,942 | | teleport-tour · `RX_VK_TEX_AHEAD=1` · longest load (s) | 0.0 | 9.7 (+24200%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · frame time (p99) (ms) | 122.2 | 113.7 (−7%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · market · frame time (p99) (ms) | — | 312.2 | | teleport-tour · `RX_VK_TEX_AHEAD=1` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · peak VRAM (MB) | 2,854 | 3,118 (+9%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · longest load (s) | 0.1 | 0.1 (−44%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · frame time (p99) (ms) | 157.4 | 112.9 (−28%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · forbidden log lines | 0.0 | 0.0 | **Flags and settings** | Flag | What it is | Values tried | Before this PR | After this PR | Effect (peak VRAM) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_TEX_AHEAD` | switch | `0`, `1`, `2` | main's default | the PR's default | teleport-tour, the PR at 0 → at 2: 3,144 → 3,118 (−1%) | no clear effect | As Markdown for the PR description: ```markdown ## Measurements **Verdict:** fail on steamdeck (1 report): `teleport-tour screenshot.max_diff`; `teleport-tour load_time_max_s.max_delta_pct (+24200.0%)` failed. Hardware: 1: linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM` Scene: teleport-tour, from the repository's save kokiri-start; 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); teleport-tour: 1 game start a side, kokiri-first 5 s after 0 s warm-up, market 5 s after 0 s warm-up, kakariko 5 s after 0 s warm-up, field 5 s after 0 s warm-up, kokiri-return 5 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 (`@Beanstalk` on this PR runs it again) | | Before: main @ 6548881 | After: this PR @ 3af1d15 | |---|---|---| | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · peak VRAM (MB) | 3,283 | 3,144 (−4%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · peak VRAM (MB) | 3,283 | 3,065 (−7%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · longest load (s) | 10.1 | 10.1 (−0%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · market · frame time (p99) (ms) | 464.3 | 238.0 (−49%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · kokiri-first · frame time (p99) (ms) | 130.2 | 107.7 (−17%) | | teleport-tour · `RX_VK_TEX_AHEAD=0` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · peak VRAM (MB) | 3,182 | 2,942 (−8%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · market · peak VRAM (MB) | — | 2,942 | | teleport-tour · `RX_VK_TEX_AHEAD=1` · longest load (s) | 0.0 | 9.7 (+24200%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · kokiri-first · frame time (p99) (ms) | 122.2 | 113.7 (−7%) | | teleport-tour · `RX_VK_TEX_AHEAD=1` · market · frame time (p99) (ms) | — | 312.2 | | teleport-tour · `RX_VK_TEX_AHEAD=1` · forbidden log lines | 0.0 | 0.0 | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · peak VRAM (MB) | 2,854 | 3,118 (+9%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · longest load (s) | 0.1 | 0.1 (−44%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · kokiri-first · frame time (p99) (ms) | 157.4 | 112.9 (−28%) | | teleport-tour · `RX_VK_TEX_AHEAD=2` · forbidden log lines | 0.0 | 0.0 | **Flags and settings** | Flag | What it is | Values tried | Before this PR | After this PR | Effect (peak VRAM) | Helps / doesn't help / untested | |---|---|---|---|---|---|---| | `RX_VK_TEX_AHEAD` | switch | `0`, `1`, `2` | main's default | the PR's default | teleport-tour, the PR at 0 → at 2: 3,144 → 3,118 (−1%) | no clear effect | ``` </details> <details><summary>The testers' hardware</summary> | Target | Report | Ran on | | --- | --- | --- | | `windows+vram8gb` | [1](https://beanstalk.magicbean.dev/reports/rp_521666ea63125b4b) | windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3070, 621.24.0` `AMD Ryzen 5 5600X3D 6-Core Processor, 32 GB RAM` | | `windows+vram8gb` | [2](https://beanstalk.magicbean.dev/reports/rp_bface65d371268f2) | windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 2080 SUPER, 621.24.0` `Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz, 16 GB RAM` | | `windows+vram12gb` | [1](https://beanstalk.magicbean.dev/reports/rp_746e0328efc1233e) | windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 4070 Ti, 621.14.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM` | | `windows+vram12gb` | [2](https://beanstalk.magicbean.dev/reports/rp_b85980f0865152bd) | windows+nvidia+nvidia-proprietary+dgpu `NVIDIA GeForce RTX 3080 Ti, 621.24.0` `AMD Ryzen 9 5900X 12-Core Processor, 64 GB RAM` | | `steamdeck` | [1](https://beanstalk.magicbean.dev/reports/rp_0d035cd77068c69c) | linux+amd+radv+igpu+handheld `AMD Custom GPU 0405 (RADV VANGOGH), 26.1.99` `AMD Custom APU 0405, 12 GB RAM` | </details>
perf(vk): RX_VK_TEX_AHEAD=2 leaves the list of textures drawn unread; the room counts the texture about to be made
Some checks failed
CI / Conventional Commit title (pull_request) Successful in 11s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 14s
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
CI / Workflow lint (actionlint) (pull_request) Successful in 15s
Third-party parts / As their notes say (pull_request) Successful in 16s
PR description / Description follows the template (pull_request_target) Successful in 4s
Website / Website on Cloudflare (pull_request_target) Has been skipped
CI / Website build and checks (pull_request) Failing after 2m4s
beanstalk/windows+vram12gb 1/1 fail
beanstalk/approval approved by GummiNaomi (reviewed)
beanstalk/windows+vram8gb 2/1 fail
beanstalk/steamdeck 1/1 fail
3af1d156e5
Review of #1813 (futondestroyer):

- The list's file (cache/longshot_textures_drawn.bin) was read, and its set filled, at the first
  look at the queue, before the mode and the room were known: =2 never asks it and paid for it
  all the same. It is now read through texture_ahead::load_list, which opens the file only for a
  mode that uses the list, and only once a texture is taken to be looked up in it. =1 reads it as
  before (a seeded file in tests/test_texture_ahead.cpp: =1 reads it, =2 and off do not).
- The room checked was the cache's as it stood, the texture's image charged only once made: the
  last one prepared could take the 256 MB reserve by its own size. The texture taken is now
  checked with what its image can take at most (texture_ahead::cost_bound: decoded ASTC at RGBA8
  or BC's whole blocks, whichever is more); one that would take the reserve is made at its first
  draw, as without the switch. The check of the cache alone before a texture is taken stays.

RX_VK_TEX_AHEAD is read once through texture_ahead::mode (0 off, 2 every texture, any other the list).

Assisted-by: Claude Code
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Author
Member

Thanks @futondestroyer. Both blocking points are fixed in 3af1d156. That commit is also what I measured below.

1. Mode 2 still read the list (vk_texture.cpp ~3011). Fixed at the root. The file is now read through a new function, texture_ahead::load_list(mode, path, set) in texture_ahead.h. It opens the file only for a mode that uses the list (uses_list: not 0, not 2). texture_ahead_run asks for the list only after it has taken a texture to look up, and never with =2, so a run blocked by the budget or with an empty queue doesn't read it either. RX_VK_TEX_AHEAD is now parsed once, by texture_ahead::mode.

  • Test: tests/test_texture_ahead.cpp writes a list to disk. load_list(1, …) must read it, and load_list(2, …) and load_list(0, …) must leave the set empty. With the old behaviour (the file read whatever the mode) the test fails at !load_list(2, …) && back.empty(). tests/test_texture_ahead.py also checks the renderer: ahead_load goes through load_list, and texture_ahead_run calls it once, after take and behind if (!ahead_all()). With the old call put back (before the room check, for =2 too), it fails with "the list is asked for with RX_VK_TEX_AHEAD=2".
  • In the game: the =2 runs below log no textures drawn before are known line. The =1 runs log it once.

2. The reserve check ran before the candidate was made (~3012/3023). I went for the hard reserve. The texture taken is now checked with its own cost: texture_ahead::room(total_bytes, budget, ahead_cost(t)). ahead_cost uses texture_ahead::cost_bound, an upper bound worked out from the descriptor. Decoded ASTC counts as RGBA8 or as BC's whole 4x4 blocks, whichever is more at each level; other block formats count as whole blocks × bytes per block. Layers, faces and depth are included. A texture that would take the 256 MB reserve is not prepared and gets made at its first draw, as with the switch off. The cache-only check before take is still there, so a full cache still stops the loop without taking anything from the queue.

  • One limit, also written in the comment: the bound leaves out the allocator's rounding for that one image (the pad that create adds). So "never within 256 MB" holds up to one image's rounding.
  • Test: room(2000 MB, 2400 MB, 144 MB) passes and 145 MB fails, plus cost_bound cases for BC1 mips, a BC3 array, and ASTC 8x8 decoded (RGBA8, and BC's 16-byte floor at 1x1). With the cost ignored, the test fails at !room(2000 * MB, 2400 * MB, 145 * MB).
  • Merge with #1810: git merge-tree of this head with #1810 (e841323c) is clean, and #1810's texture_move_wanted skip stays right after if (!known) continue;.

Checks at 3af1d156: test_texture_ahead.py and test_switch_table.py pass on Windows (clang). test_texture_ahead.py passes on Linux (WSL Ubuntu 24.04, g++). vk_texture.cpp compiles syntax-only with the builder's Windows flags. Of the checks that read the changed files, 19 fail on Windows and 15 on Linux, and every one of those fails the same way on this PR's base 65488813. The Guide screenshots check passes locally. The description check was red because of the old template; the new description follows the current one.

What my own runs say about 0 vs 1 vs 2 at this head (for @HumanBean's question; the default is unchanged in this push).

Setup: Windows 11, RTX 4070 SUPER 12 GB, 9950X3D (SMT off), main's defaults (the control tool's test profile) plus RX_VK_FRAME_TRACE=1. The tour: kokiri-start, then teleport to Market, Kakariko and Hyrule Field, ~35 s walking in each. Runs were interleaved 0,1,2,2,1,0 at 16 cores, then again held to 4 cores (0xF00). Each value is one run.

=0 =1 =2
16 cores: frames ≥25 ms in the 3 s after the 3 arrivals (ms) 849, 683 748, 595 825, 646
16 cores: frames ≥33 ms / p99 (first teleport to end) 33, 30 / 15.9, 15.1 ms 25, 28 / 15.4, 15.3 ms 33, 32 / 18.0, 17.4 ms
16 cores: worst frame 314, 308 235, 238 528, 439
16 cores: arrival waits, the 3 summed (ms) 10,176, 9,620 9,525, 9,645 10,462, 10,018
4 cores: frames ≥25 ms after arrivals (ms) 987, 885 829, 728 838, 897
4 cores: frames ≥33 ms / p99 51, 56 / 24.4, 21.1 ms 34, 32 / 19.6, 19.6 ms 38, 36 / 21.7, 19.7 ms
4 cores: worst frame 475, 669 385, 262 696, 680
GPU memory peak, all processes (nvidia-smi, MB; 16 / 4 cores) 11,040, 11,197 / 11,411, 11,400 11,245, 11,245 / 10,829, 10,908 11,287, 11,477 / 11,466, 11,457
private bytes peak (MB; 16 / 4 cores) 22,474, 22,302 / 23,140, 22,771 22,448, 22,857 / 22,360, 22,436 22,988, 22,888 / 23,387, 23,380
textures prepared (images) 0 ~2,050-2,130 (2.4-2.5 GB) ~10,000-11,100 (12.2-14.5 GB)
textures evicted for the budget 3,657-5,012 3,643-3,942 10,252-11,444

My reading: on this 12 GB card, =2 doesn't beat =1 at this head, and on tails it is worse than both.

  • It makes the Market and Hyrule Field entries better: the worst frame in each arrival window is 101-182 ms against 129-314 ms.
  • It makes the Kakariko entry, which comes after the Market, much worse: 439-696 ms against 171-308 ms with =0 and 235-385 ms with =1. The likely cause: the Market's prepared textures (most never drawn) fill the budget, the budget evicts them, and the runs have ~2.5-3 times the budget evictions, and the Kakariko entry frame spends 202-525 ms on uploads.
  • Peak GPU memory is about the same in all three, because the budget holds it.
  • =1 has the best tails on both core counts.

The halving in the old description, measured at 54a42715 with uncapped play settings, doesn't reproduce on current main with its defaults. I've replaced those numbers in the description with these.

So I wouldn't replace =1 with =2 on this evidence. The farm's 8 GB and Deck cells are the ones to watch, since a smaller budget should make the crowding worse. If =2 is kept as an option, a follow-up could stop preparing ahead once the cache passes some share of its budget, rather than only near the reserve. That isn't in this PR. Also not in this push: the ahead_all() comment in vk_texture.cpp still quotes the old 54a42715 numbers. I left it so the pushed head is exactly the build I measured; it gets rewritten or removed once the farm settles what =2 becomes.

Not measured: Linux, AMD, 8 GB cards, the Deck, adult areas, and the Market→Kakariko order reversed.

Thanks @futondestroyer. Both blocking points are fixed in 3af1d156. That commit is also what I measured below. **1. Mode 2 still read the list (`vk_texture.cpp` ~3011).** Fixed at the root. The file is now read through a new function, `texture_ahead::load_list(mode, path, set)` in `texture_ahead.h`. It opens the file only for a mode that uses the list (`uses_list`: not 0, not 2). `texture_ahead_run` asks for the list only after it has taken a texture to look up, and never with `=2`, so a run blocked by the budget or with an empty queue doesn't read it either. `RX_VK_TEX_AHEAD` is now parsed once, by `texture_ahead::mode`. - Test: `tests/test_texture_ahead.cpp` writes a list to disk. `load_list(1, …)` must read it, and `load_list(2, …)` and `load_list(0, …)` must leave the set empty. With the old behaviour (the file read whatever the mode) the test fails at `!load_list(2, …) && back.empty()`. `tests/test_texture_ahead.py` also checks the renderer: `ahead_load` goes through `load_list`, and `texture_ahead_run` calls it once, after `take` and behind `if (!ahead_all())`. With the old call put back (before the room check, for `=2` too), it fails with "the list is asked for with RX_VK_TEX_AHEAD=2". - In the game: the `=2` runs below log no `textures drawn before are known` line. The `=1` runs log it once. **2. The reserve check ran before the candidate was made (~3012/3023).** I went for the hard reserve. The texture taken is now checked with its own cost: `texture_ahead::room(total_bytes, budget, ahead_cost(t))`. `ahead_cost` uses `texture_ahead::cost_bound`, an upper bound worked out from the descriptor. Decoded ASTC counts as RGBA8 or as BC's whole 4x4 blocks, whichever is more at each level; other block formats count as whole blocks × bytes per block. Layers, faces and depth are included. A texture that would take the 256 MB reserve is not prepared and gets made at its first draw, as with the switch off. The cache-only check before `take` is still there, so a full cache still stops the loop without taking anything from the queue. - One limit, also written in the comment: the bound leaves out the allocator's rounding for that one image (the `pad` that `create` adds). So "never within 256 MB" holds up to one image's rounding. - Test: `room(2000 MB, 2400 MB, 144 MB)` passes and `145 MB` fails, plus `cost_bound` cases for BC1 mips, a BC3 array, and ASTC 8x8 decoded (RGBA8, and BC's 16-byte floor at 1x1). With the cost ignored, the test fails at `!room(2000 * MB, 2400 * MB, 145 * MB)`. - Merge with #1810: `git merge-tree` of this head with #1810 (e841323c) is clean, and #1810's `texture_move_wanted` skip stays right after `if (!known) continue;`. **Checks at 3af1d156:** `test_texture_ahead.py` and `test_switch_table.py` pass on Windows (clang). `test_texture_ahead.py` passes on Linux (WSL Ubuntu 24.04, g++). `vk_texture.cpp` compiles syntax-only with the builder's Windows flags. Of the checks that read the changed files, 19 fail on Windows and 15 on Linux, and every one of those fails the same way on this PR's base 65488813. The Guide screenshots check passes locally. The description check was red because of the old template; the new description follows the current one. **What my own runs say about 0 vs 1 vs 2 at this head** (for @HumanBean's question; the default is unchanged in this push). Setup: Windows 11, RTX 4070 SUPER 12 GB, 9950X3D (SMT off), main's defaults (the control tool's test profile) plus `RX_VK_FRAME_TRACE=1`. The tour: `kokiri-start`, then teleport to Market, Kakariko and Hyrule Field, ~35 s walking in each. Runs were interleaved 0,1,2,2,1,0 at 16 cores, then again held to 4 cores (0xF00). Each value is one run. | | `=0` | `=1` | `=2` | |---|---|---|---| | 16 cores: frames ≥25 ms in the 3 s after the 3 arrivals (ms) | 849, 683 | 748, 595 | 825, 646 | | 16 cores: frames ≥33 ms / p99 (first teleport to end) | 33, 30 / 15.9, 15.1 ms | 25, 28 / 15.4, 15.3 ms | 33, 32 / 18.0, 17.4 ms | | 16 cores: worst frame | 314, 308 | 235, 238 | 528, 439 | | 16 cores: arrival waits, the 3 summed (ms) | 10,176, 9,620 | 9,525, 9,645 | 10,462, 10,018 | | 4 cores: frames ≥25 ms after arrivals (ms) | 987, 885 | 829, 728 | 838, 897 | | 4 cores: frames ≥33 ms / p99 | 51, 56 / 24.4, 21.1 ms | 34, 32 / 19.6, 19.6 ms | 38, 36 / 21.7, 19.7 ms | | 4 cores: worst frame | 475, 669 | 385, 262 | 696, 680 | | GPU memory peak, all processes (nvidia-smi, MB; 16 / 4 cores) | 11,040, 11,197 / 11,411, 11,400 | 11,245, 11,245 / 10,829, 10,908 | 11,287, 11,477 / 11,466, 11,457 | | private bytes peak (MB; 16 / 4 cores) | 22,474, 22,302 / 23,140, 22,771 | 22,448, 22,857 / 22,360, 22,436 | 22,988, 22,888 / 23,387, 23,380 | | textures prepared (images) | 0 | ~2,050-2,130 (2.4-2.5 GB) | ~10,000-11,100 (12.2-14.5 GB) | | textures evicted for the budget | 3,657-5,012 | 3,643-3,942 | 10,252-11,444 | My reading: on this 12 GB card, `=2` doesn't beat `=1` at this head, and on tails it is worse than both. - It makes the Market and Hyrule Field entries better: the worst frame in each arrival window is 101-182 ms against 129-314 ms. - It makes the Kakariko entry, which comes after the Market, much worse: 439-696 ms against 171-308 ms with `=0` and 235-385 ms with `=1`. The likely cause: the Market's prepared textures (most never drawn) fill the budget, the budget evicts them, and the runs have ~2.5-3 times the budget evictions, and the Kakariko entry frame spends 202-525 ms on uploads. - Peak GPU memory is about the same in all three, because the budget holds it. - `=1` has the best tails on both core counts. The halving in the old description, measured at 54a42715 with uncapped play settings, doesn't reproduce on current main with its defaults. I've replaced those numbers in the description with these. So I wouldn't replace `=1` with `=2` on this evidence. The farm's 8 GB and Deck cells are the ones to watch, since a smaller budget should make the crowding worse. If `=2` is kept as an option, a follow-up could stop preparing ahead once the cache passes some share of its budget, rather than only near the reserve. That isn't in this PR. Also not in this push: the `ahead_all()` comment in `vk_texture.cpp` still quotes the old 54a42715 numbers. I left it so the pushed head is exactly the build I measured; it gets rewritten or removed once the farm settles what `=2` becomes. Not measured: Linux, AMD, 8 GB cards, the Deck, adult areas, and the Market→Kakariko order reversed.
Author
Member

@Beanstalk windows+vram8gb, windows+vram12gb and steamdeck: the teleport walk from kokiri-start (Kokiri Forest, Castle Town market, Kakariko, Hyrule Field, back to Kokiri) with RX_VK_TEX_AHEAD 0, 1 and 2, RX_VK_FRAME_TRACE=1. Measure each arrival's longest frame and the eviction counts. Expecting on 8 GB and the Steam Deck =2 to evict more and its second arrivals no better than =1 (that is the question for the default); on 12 GB about what the description's Measurements shows.

(Earlier runs on #1846 and #1847 today ended in error before building: the farm's checkout of main failed git rev-parse --verify HEAD^{commit} in packs.py's cmd_base, so nothing was measured there.)

@Beanstalk windows+vram8gb, windows+vram12gb and steamdeck: the teleport walk from kokiri-start (Kokiri Forest, Castle Town market, Kakariko, Hyrule Field, back to Kokiri) with `RX_VK_TEX_AHEAD` 0, 1 and 2, `RX_VK_FRAME_TRACE=1`. Measure each arrival's longest frame and the eviction counts. Expecting on 8 GB and the Steam Deck `=2` to evict more and its second arrivals no better than `=1` (that is the question for the default); on 12 GB about what the description's Measurements shows. (Earlier runs on #1846 and #1847 today ended in `error` before building: the farm's checkout of main failed `git rev-parse --verify HEAD^{commit}` in packs.py's cmd_base, so nothing was measured there.)
Author
Member

The farm run on 3af1d156 is complete (8 GB and 12 GB Windows, and the Steam Deck). A summary, since the sticky marks every target failed:

Slowest frames (p99) on the teleport tour, main → this PR:

Target =1 (the current list) =2 (every texture)
8 GB 50.8 → 65.5 ms (+29 %) 50.5 → 26.6 ms (−47 %)
12 GB 41.1 → 52.8 ms (+29 %) 44.3 → 41.7 ms (−6 %)
Steam Deck 122.2 → 113.7 ms (−7 %) 157.4 → 112.9 ms (−28 %)

So on all three, =2 gives the lower tails and =1 is the weaker of the two. @HumanBean, on replacing =1: these numbers support it. My own 12 GB runs (in the description) had =1 slightly ahead of =2 on the arrivals' slow-frame time, so the two measures disagree there; on the 8 GB card and the Deck the farm's gap is large. I'm happy to make =1 take this PR's policy (every texture) and drop the old list, if you'd like that in this PR.

What the failures are:

  • Crashes on the 8 GB tester: the second report (RTX 2080 SUPER, i7-9700) has both the main and the PR run hanging and then crashing with the same code (0xC0000409), and the farm flags it as an outlier against its peers: that PC, not this PR. The first report's crash (one =2 run, inside NVIDIA's shader compiler nvgpucomp64.dll, with ~3.6 GB of textures made) is the open one: a main run on that same PC also crashed (stack overflow) in another cell, so it isn't attributable yet. Another 8 GB run would settle it.
  • Noise, not the PR: the screenshot limits fail with main against itself (0.24-0.31 on the same tours); the 12 GB kokiri-first VRAM peak is +64 % at =0, where this PR changes nothing; and the Deck's =0 row is −49 %, also a cell this PR doesn't touch. These cells show the run-to-run spread, which is why I'd read only the =1/=2 columns above.
The farm run on 3af1d156 is complete (8 GB and 12 GB Windows, and the Steam Deck). A summary, since the sticky marks every target failed: **Slowest frames (p99) on the teleport tour, main → this PR:** | Target | `=1` (the current list) | `=2` (every texture) | |---|---|---| | 8 GB | 50.8 → 65.5 ms (+29 %) | 50.5 → 26.6 ms (−47 %) | | 12 GB | 41.1 → 52.8 ms (+29 %) | 44.3 → 41.7 ms (−6 %) | | Steam Deck | 122.2 → 113.7 ms (−7 %) | 157.4 → 112.9 ms (−28 %) | So on all three, `=2` gives the lower tails and `=1` is the weaker of the two. @HumanBean, on replacing `=1`: these numbers support it. My own 12 GB runs (in the description) had `=1` slightly ahead of `=2` on the arrivals' slow-frame time, so the two measures disagree there; on the 8 GB card and the Deck the farm's gap is large. I'm happy to make `=1` take this PR's policy (every texture) and drop the old list, if you'd like that in this PR. **What the failures are:** - **Crashes on the 8 GB tester:** the second report (RTX 2080 SUPER, i7-9700) has **both** the main and the PR run hanging and then crashing with the same code (0xC0000409), and the farm flags it as an outlier against its peers: that PC, not this PR. The first report's crash (one `=2` run, inside NVIDIA's shader compiler `nvgpucomp64.dll`, with ~3.6 GB of textures made) is the open one: a main run on that same PC also crashed (stack overflow) in another cell, so it isn't attributable yet. Another 8 GB run would settle it. - **Noise, not the PR:** the screenshot limits fail with main against itself (0.24-0.31 on the same tours); the 12 GB `kokiri-first` VRAM peak is +64 % at `=0`, where this PR changes nothing; and the Deck's `=0` row is −49 %, also a cell this PR doesn't touch. These cells show the run-to-run spread, which is why I'd read only the `=1`/`=2` columns above.
Some checks failed
CI / Conventional Commit title (pull_request) Successful in 11s
Guide screenshots / Guide images and local UI source review (pull_request) Failing after 14s
CI / SPIR-V fixtures (global narrow load/store, direct surface atomics, F2I immediate) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 1/2) (pull_request) Has been skipped
CI / Python source checks (ubuntu-24.04, part 2/2) (pull_request) Has been skipped
Third-party parts / The same as upstream (pull_request) Has been skipped
CI / Workflow lint (actionlint) (pull_request) Successful in 15s
Third-party parts / As their notes say (pull_request) Successful in 16s
PR description / Description follows the template (pull_request_target) Successful in 4s
Website / Website on Cloudflare (pull_request_target) Has been skipped
CI / Website build and checks (pull_request) Failing after 2m4s
beanstalk/windows+vram12gb 1/1 fail
beanstalk/approval approved by GummiNaomi (reviewed)
beanstalk/windows+vram8gb 2/1 fail
beanstalk/steamdeck 1/1 fail
This pull request doesn't have enough approvals yet. 0 of 1 approvals granted.
This branch is out-of-date with the base branch
You are not authorized to merge this pull request.
View command line instructions

Checkout

From your project repository, check out a new branch and test the changes.
git fetch -u 'perf/tex-ahead-every:GummiNaomi-perf/tex-ahead-every'
git switch 'GummiNaomi-perf/tex-ahead-every'
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
4 participants
Notifications
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!1813
No description provided.