Repository navigation
feat(video): support 48:9 triple-monitor picture geometry - #2052
Conversation
Assisted-by: GPT-6 (Codex)
9ed2da5 to
557542c
Compare
ghost
left a comment
ghost
left a comment
There was a problem hiding this comment.
Looks good to me. The 32:9 and 48:9 limits now live in one place (hos_output_size_valid / hos_output_is_surround in hos_wide_size.h), so the host parser and the menu can't drift apart any more. The host and the menu both force the surround shape for a >32:9 custom output, so the menu's Render resolution rows match what boot picks. The 24-row alignment checks out: 7680x1440 goes over the 3840x2160 budget and comes out at exactly 6528x1224 (48:9), and 5760x1080 fits in the budget at full size. Handheld is unaffected because hos_output_size reports 2560x1440 there. The pool goes to 515 MB and 900 MB as the test expects. Nothing on main since the base touches these files, and CI is green.
A few nits, none blocking:
hos_main.c: a surround output now overrides an explicitHOS_ASPECT(see inline).hos_wide_clamp: the 1e-6 snap doesn't catch the 4-decimal form (see inline).vk_menu.cpppicture_for: the doc comment above it still says "with Ultrawide on". A surround output now takes this path with Ultrawide off too, so it's worth a few words there.
The opening-transition black centre is also on the pre-change 32:9 build, so I agree it's not from this PR. A real three-monitor span and Linux are still untested, as the description says.
| uint32_t hos_output_size(int height); | ||
| const uint32_t output_w = hos_output_size(0), output_h = hos_output_size(1); | ||
| /* A custom surround output explicitly requests that shape, even if the OS reports one monitor. */ | ||
| if (hos_output_is_surround(output_w, output_h)) want = (float)output_w / (float)output_h; |
There was a problem hiding this comment.
Nit: this also runs when HOS_ASPECT is set, so HOS_ASPECT=1.7778 (or 0) can't opt out of the surround camera with a 5760x1080/7680x1440 output. The block comment above still says HOS_ASPECT wins. If the override is intended, a word in that comment would do. If not, if (!e && hos_output_is_surround(...)) keeps the env var as the escape hatch. (The menu ignores HOS_ASPECT anyway, so either choice is consistent with it.)
| return want > 32.0 / 9.0 ? 32.0 / 9.0 : want < 5.0 / 4.0 ? 5.0 / 4.0 : want; | ||
| const double surround = 48.0 / 9.0; | ||
| /* Accept the rounded value commonly used for HOS_ASPECT. */ | ||
| return want >= surround - 0.000001 ? surround : want < 5.0 / 4.0 ? 5.0 / 4.0 : want; |
There was a problem hiding this comment.
Nit: the 1e-6 tolerance catches 5.333333 but not 5.3333, which is the 4-decimal form the docs use for HOS_ASPECT (2.3889). With HOS_ASPECT=5.3333 at the 1440 output, r isn't exactly 48/9, so hos_wide_size takes the 8-row path and gives 6608x1240 (5.329) instead of 6528x1224. A tolerance like 1e-4 would snap both. Only matters for the env var: the output-derived ratio is exact.
Document custom surround output precedence and match the menu description. Assisted-by: GPT-6 (Codex)
|
Addressed the three nits at |
Brings in NotRexed#2052 (48:9), NotRexed#2078 (the off value's label) and NotRexed#2041 (the V-Sync row). Conflicts resolved with upstream's text kept and the row's new name on top: - vk_menu.cpp picture_for and hos_main.c's widescreen comment: NotRexed#2052's wording, "Ultrawide" read as the Aspect ratio. - tx_ultrawide: 16:9 again instead of NotRexed#2078's "Off (16:9)". NotRexed#2078 added "Off" because "Ultrawide: 16:9" read as already on; under the name Aspect ratio the value is a shape, and "Aspect ratio: Off (16:9)" would not read well. - readme: the settings table and NotRexed#2052's Output resolution notes, with the row's new name. - test_audio_output_menu.py, test_stick_dead_zone.py: the option stubs take both NotRexed#2041's vsync and the ultrawide key. Also follows the rename where the first commit missed it: recomplauncher.json (label, help, and its default now 1, so the launcher does not write 16:9 over the new default) and tests/test_menu_restart.cpp's sample row names. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Brings in NotRexed#2052 (48:9 surround), NotRexed#2041 and the rest of main. NotRexed#2052's code is kept as merged (output limits, hos_wide_clamp up to 48:9, the 24-row alignment at 48:9); the cap and HOS_WIDE_MAX_PX sit on top. With the 5120x2160-pixel cap, 48:9 comes out as: - the 7680x1440 output: full size (it has exactly 5120x2160's pixels) instead of 6528x1224. Its pool stays 900 MB, set by the output's own size as before; the 5120x2880 output already plays in that pool with more pixels. - 48:9 at the 4K output (11520x2160 wanted): kept to 7680x1440, exactly 48:9, pool 915 MB, the same pixels and pool as 21:9's 5120x2160. Tests follow: test_widescreen_shapes.py takes the full-size 7680x1440 and a 48:9-at-2160 case, and keeps NotRexed#2052's 6528x1224 cases (the 24-row alignment and the four-decimal 5.3333) under HOS_WIDE_MAX_PX=8294400, the old cap; test_output_resolution.py's 7680x1440 output is full size (its scene special case is no longer needed); test_rt_pool_minimum.py has 48:9 at the 4K output (915 MB). The readme's 7680x1440 sentence and the cap follow. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Summary
On a 48:9 desktop the camera stopped widening at 32:9, and the game rejected custom outputs such as 5760×1080 and 7680×1440. Both settings now give a wider view and keep the HUD in the center 16:9 area.
Addresses the 48:9 request in #1007, originally reported in #2039.
Technical changes
Validation
At
0d5e2981, geometry and output fixtures pass on Windows LLVM-MinGW and Linux GCC. The new four-decimal case fails before the fix (6608x1240) and passes after it (6528x1224). Full local suite: Windows 325 passed / 3 failed / 25 skipped; Linux 330 / 1 / 22. All three Windows failures reproduce on mainf915af93; Linux lacks the bake fixture dependency covered by fix(build): run offline bake with portable zstd and Clang 18 #1954.Current-head upstream source checks fail in finish-output fixtures after main added FSR runtime provisioning; existing test: the finish checks give the build its arguments, read since #1866 #2102 owns the fixture correction. These differ from the local failures. Windows build/package and AppImage passed.
At
557542cdond47206ed, the geometry and output fixtures passed on Windows and Ubuntu. All 11 Windows startup cases also passed. The previous revision also passed the geometry, parser and presentation rectangle checks on Windows and Ubuntu.Native Windows smoke at
9ed2da57on RTX 5070 Ti used both output settings in a 1920×360 window. Vulkan reported source sizes of 5760×1080 and 6528×1224, with a full 1920×360 destination in both cases. Captures decoded at the two configured output sizes. The HUD stayed in the middle third and the house-to-Kokiri Forest transition completed in both runs.With 7680×1440 configured, maximizing to a 3440×1369 swapchain kept a 3440×645 centered destination. Both runs exited normally and preserved the original checkpoint hashes.
The brief black center with visible sides during the opening transition was also reproduced at 32:9 on the pre-change build
600960cd. Linux gameplay, campaign-wide cutscenes and a physical three-monitor span remain untested.Dev-work claim.
0d5e2981)