On an 8K panel Match display asks 18x (27x with supersample 1.5), past
the full-VRAM surface: the high-resolution window engages, dynamic
resolution is ineligible there, and the render thread saturates (R4 0.8
on an RTX 4080 at 7680x4320: fixed 27x, render_busy 100 %, Low 51.9
guest Hz). A game.toml key caps the display height that Match display,
its dynamic-resolution floor and preset detection resolve against.
Default 0 keeps today's behaviour for every title.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Keep selector zero stock and seal all choices until the next session restart. Refuse nonzero online activation, scope non-stock save states by frozen choices, and preserve original instruction guards when combined with effective-address adapters. Use overlay ABI34 on current master without the old menu dependency stack.
Extract the audited adapters and card transport onto master without the held selector/menu stack. Keep adapters and codecs tied to mod activation, use overlay ABI 33, retain rejected-load protection on codec registration, and refuse extended-card codecs in netplay before frame zero. Preserve default atomic save publication and guest-visible write failures.
Bit 29 is Beetle's InCommand. The hardware decodes progressively and stalls
while its 32-word OutFIFO is full, so a decode command stays in progress until
all but the last FIFO load of its output has been read. mdec.c decodes a whole
command at once and cleared the bit at execute time.
libpress DecDCTin spins on bit 29 before issuing the next frame. Reporting
ready early let Breath of Fire IV's FMV player issue the next DecDCTin and
DecDCTout while a column DMA was mid-transfer; DecDCTout's CHCR=0 restart
dropped the 205 words already moved, so every later column read started 820
bytes late (vertical stripes at macroblock edges, a one-macroblock vertical
wrap and shifted colour channels in the intro FMV).
Derive the bit from the unread output instead. No new state, so snapshots
are unchanged. Adds mdec_busy_drain_test (fails on the old behaviour).
Renderers that test a 16-bit screen X with an unsigned compare against a width kept in a register (often a loaded clip bound) reject every negative X, so widening the bound only reveals the right band. A screen_x site on sltu now routes through psx_ws_cull_sltu_band: -4m <= (int16)SX < W while widened, identity at 4:3. Four margins keep large near polygons whose corners all lie past one margin; an unbounded left side overflowed a renderer's fixed primitive lists. Covered by ws_cull_screen_x_hi_codegen_test; documented in WIDESCREEN.md.
P1: quality_launcher_apply is now a pure preset -> rows mapping
(psxq::rows_from_runtime plus startup constants); it never writes the live
globals. The in-game menu publishes only thread-safe callbacks: that
mapper, and a re-detect that returns this session's detection (a fresh
probe needs the GL context and mutates the run's state, so it stays in the
pre-game launcher). Summary/reason strings and the detection are fixed on
the main thread at menu setup; settings_get reads only the main-thread
snapshot.
P2: psxq::MenuSelection keeps what the player selected versus what is live.
A pick's next-launch rows and netplay-deferred rows stay selected for the
next seed until the live value matches; other rows follow live. The host
classification moves to psxq::classify_launcher/row_differs (same rules).
Tests: quality_presets_test test_ingame_menu_host_path runs the review's
fixture through the host get/pick/apply/save path (Ultra -> High, then
Antialiasing = NONE; a governed edit is still Custom from High; a netplay-
deferred row stays selected). menu_quality_callbacks_test guards the
draw-thread callbacks against runtime writes and stateful calls.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A hidden run keeps the saved preset without detecting; a menu pick of High
is saved and owns its keys; Smooth motion off while High is in force is a
governed live edit that saves Custom from High with its value.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- [quality.*] accepts only the supported keys (quality_key_supported):
twelve player-editable ones (masked, copied, classified, applied) and
nine preset-only ones (no settings.toml form; applied on select). Any
other key, e.g. vsync or renderer, is a load error.
- Picking a preset now also applies bloom, texture LOD, anisotropy, FMV
chroma smoothing and dithering (quality_apply_video_globals).
- The frontends flag Custom on any tracked row; the host counts only rows
of keys the title governs (psxq::launcher_custom_edit), so Antialiasing
no longer turns Ultra into Custom when no preset sets it.
Tests cover the rejection, every player key's masking/copy, and the
classification cases.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The boot launcher, soft-return launcher and the in-game menu's live
settings now save settings.toml through psxq::settle_save:
- a one-run PSX_QUALITY override never persists: governed keys and the
quality_* keys come from disk, so the previous preset or Custom values
survive even when the boot seed was built empty;
- a governed live edit (psxq::live_edit) or launcher Custom becomes
Custom from the current base, with the edited values, and survives
restart instead of being masked away;
- a launcher pick records the preset; unrelated edits keep it.
Tests reproduce both review cases through a settings.toml round trip.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Nothing is measured about an unknown GPU, so the fallback no longer assumes
Ultra's extra work (supersampling above the display, title Ultra extras);
Low would drop Smooth motion and the PGXP extras on what is usually capable
new hardware. Medium keeps the display's resolution with dynamic
resolution under it. Covers unnamed renderers, Linux with only an AMD PCI
id, and unnamed nouveau chips. docs/QUALITY_PRESETS.md states plainly that
autodetect is a static heuristic over name, threads and memory with no
benchmark, and that the dynamic systems keep the frame rate.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With [timing] guest_cycle_scale active, every one-cycle psx_cyc_charge went
straight to psx_cpu_charge and psx_advance_cycles, bypassing the batch the
scale-1 path uses. [timing] guest_cycle_scale_batch (PSX_GCS_BATCH=0/1)
scales the charge in place, with the same fractional carry and order as
psx_cpu_charge, then batches it like the scale-1 path. It flushes at the same
observation points. Default off.
R4 exactness, frame_fingerprint off vs on, 9000 frames each: headless boot
through intro, title and attract; a 1P Grand Prix race; and a 2P VS race.
cyc, mmio, mc and qc are identical on every frame. sp, wr, pc and ws are
identical at the end. The only differences are single-frame VBlank straddles
of wc/sc (never two frames in a row). 0 dispatch misses.
Speed, release builds, races at 60 Hz:
- M4 (macOS sample): psx_cpu_charge 4.3 % -> 0 % of emulation-thread samples;
busy 46.0 -> 44.4 % (2 pairs).
- 9800X3D: no measurable change (30.0 vs 30.5 %; one core 44.6 vs 44.2 %).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
1. Detection never runs under --headless / --hidden-window (CI, scripted
runs): a saved preset applies, else the title's [video]; nothing is
written. A detection is saved only when the GL probe succeeded; a result
from the OS name alone is used for that run and the next launch probes
again (a failed probe could otherwise pin "unknown GPU" -> Ultra).
2. Linux without a GL probe: the classifier now reads the sysfs PCI id
(Steam Deck parts, Intel integrated vs Arc, NVIDIA).
3. AMD/Intel rules rewritten against real renderer strings: "Radeon RX
Vega 11 Graphics" and other Vega/"Radeon Graphics"/R5-R7/680M-890M APUs
are integrated (Low); RX/Pro/W cards and Strix Halo are Ultra; the Intel
Arc iGPU is Low, Arc A/B cards Ultra. quality_presets_test holds a
44-row table of Windows, Mesa (incl. Steam Deck LCD/OLED, zink) and macOS
GL_RENDERER strings, plus probe-failure and headless cases.
5. generated/OpenBIOS.emitter.sha re-stamped.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Owner correction: low-end machines (software GL, Steam Deck, mobile GPUs,
integrated GPUs up to the 890M/Iris Xe class, Apple M1/M2 base, GeForce
MX/GT, fewer than 5 threads, under 6 GB) start on Low; every other machine
starts on Ultra with the dynamic systems as its safety net. Medium and High
are manual choices.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Owner decision: Ultra on first launch, with dynamic resolution, Smooth
motion from surplus and the adaptive aspect as the live safety net. Apple
silicon, discrete GPUs, unknown GPUs and 8 GB machines now detect as Ultra;
software GL, Steam Deck, mobile, Intel HD/UHD and small AMD iGPUs stay Low,
mid iGPUs Medium; fewer than 4 threads or under 6 GB still caps.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A title may declare [quality.low|medium|high|ultra] tables of [video] keys.
On first launch, and when the hardware fingerprint changes while a preset is
in force, psxrecomp detects the CPU, threads, RAM and GPU (OS facts; a hidden
1x1 GL window reads GL_RENDERER only when a detection is due and the OS did
not name the GPU), picks the matching preset and saves quality_preset,
quality_base, quality_hardware and quality_detected to settings.toml.
- Each preset is the whole game.toml re-parsed with its keys over [video], so
every [video] key works in a preset with its usual validation.
- A preset in force owns its keys: saved settings.toml values for them are
ignored. Custom (the player changed a governed setting in the launcher)
applies the player's values over quality_base and is never overridden;
new hardware under Custom is only reported.
- Launcher: GameInfo.quality_* (offered mask, detected preset, hardware
summary, quality_apply / quality_redetect callbacks) and
Settings.quality_preset/quality_base, behind RECOMP_LAUNCHER_HAS_QUALITY_PRESETS.
- PSX_QUALITY=<preset> for one run, PSX_QUALITY_REDETECT=1.
- Titles without [quality.*] are unchanged.
Rules (quality_tier.c) and the decision are unit-tested in
quality_presets_test; docs/QUALITY_PRESETS.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The detector cannot prove a loop has no later counter-dependent exit, so
the extension now skips only at block-edge PCs the title lists after an
audit. Unlisted sites are never skipped; an empty list disables it. Docs
record R4's v_wait audit as the reference; tests cover unlisted sites.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
idle_skip only skipped loops with zero stores and a single interrupt-check
edge, so it never fired on PsyQ libetc's v_wait -- the VSync(0)/VSync(n)
spin most PsyQ titles run each frame -- which decrements a stack timeout
word every pass and crosses two block edges.
New opt-in on top of idle_skip, [runtime] idle_skip_store_counters
(PSX_IDLE_SKIP_EXT=0/1 for one run):
- an iteration may span up to 8 block edges; the first edge is the anchor
and the edge count must be stable;
- one word store per iteration to the same RAM/scratchpad address whose
value drops by exactly 1 (a register holding it may drop with it).
Skips lower the counter by k (never to <= 1) and, for multi-edge loops, land
on the anchor strictly before the next device event so an IRQ is taken at
the same edge as without the skip. memory.c records the last word store's
address and value for the detector.
R4 race, 3600 frames from one savestate, held input: cyc, mmio, mc, sp, sc
and qc identical every frame; wc differs only by the elided counter stores.
Release build on an Apple M4: emulation-thread busy 47.5% -> 34.9% at 60 Hz.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review of #584: the default must not make guest-visible VRAM depend on the
internal scale. Mode 1 ("on", default; "scaled" kept as an alias) now
puts the PS1 pattern on the native VRAM grid; mode 2 ("fine") is the
elective per-internal-pixel pattern. gl_dithering_test checks the default
mode's native VRAM is identical at 4x and 1x (equal-colour Gouraud now
averages 15.50 in the default at every scale). Docs keep the
semi-transparent approximation explicit. OpenBIOS stamp regenerated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Maintainer review of #584: PS1 dithering is hardware behaviour, so it is
now on by default whenever the guest sets GP0(E1h) bit 9 ([video]
dithering = "on"; "scaled" stays opt-in, "off" is the old true-colour
output).
Eligibility is decided per primitive from its command and carried in the
vertex instead of being inferred from the colour:
- Gouraud polygons and shaded lines: geometry vertex a_col.a + 2*131072
(decoded in GEO_VS, v_dith);
- texture-modulated polygons: a_semi + 8 (v_dith in TEX_VS);
- flat polygons/lines, raw-texture polygons and every rectangle: never.
This fixes the two reproduced bugs (Gouraud with 5-bit-exact colours was
skipped; textured rectangles were dithered). Bit changes no longer flush:
eligibility is fixed when the primitive is queued; mode changes still drain.
gl_dithering_test covers each case (Gouraud incl. equal colours, shaded
line, modulated and neutral-modulated textured polys dithered; flat tri,
flat rect, flat line, raw poly, modulated and raw textured rects never;
bit clear and off unchanged), plus scaled 4x == 1x, render thread, PGXP
depth, native-wide and the accurate-blending basis. OpenBIOS stamp
regenerated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The OpenGL renderer has always drawn shaded and texture-modulated pixels
at 8 bits per channel with no dither pattern (true colour). That stays the
default. 'on' applies the PS1 4x4 offset matrix and 5-bit truncation per
internal-resolution pixel; 'scaled' puts the pattern on the native grid,
so it reads like the console at any scale and the native VRAM is the same
at every internal resolution.
GP0(E1h) bit 9 reaches the backend through a new optional facade hook
(gr_set_dither) recorded for the render thread and Smooth motion redraws.
With the mode off nothing flushes and no uniform changes. Flat, raw and
other 5-bit-exact pixels are never touched.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- config: [video] post_shader is a std::filesystem::path (like memcard_dir);
the frontend turns it into UTF-8 bytes for the GL API. Assigning the path
to a std::string did not compile on MinGW.
- .pfx `generated reuse`: a Smooth motion generated frame keeps that pass's
target from the last real frame. The bloom preset marks its bright and
blur passes so; generated frames run only the masked composite (over their
own picture and mask). bloom debug reports real/generated pass times and
reused passes.
- Mask marking no longer queries GL state per batch: each call site puts
back the blend and mask-bit stencil it set (no driver round trips on the
render thread).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Post effects now run as chains of full-screen fragment passes on one
executor (gl_postfx.inc), as asked in review. A chain is a small .pfx preset
(postfx_chain.c, GL-free): passes with a shader (a file next to the preset or
builtin:<name>), inputs (source, the optional primitive-derived mask,
previous or a named earlier pass), a scale, format, filter and uniforms.
The executor owns target allocation and lifetime, program build/cache,
uniform binding (u_rcp, u_size, u_org) and GL state save/restore.
Stages, in order, each from a fresh copy of the image: the built-in "bloom"
chain ([video] bloom), a user chain ([video] post_shader / PSX_POST_SHADER,
elective, off by default), and the built-in "fxaa"/"fxaa_hq" chain
([video] antialiasing_mode). FXAA folds in as a one-pass chain and its
output is byte-identical to the former fixed pass; it still runs on real
frames only. Generated (Smooth motion) frames run bloom and the user chain.
The "may glow" mask is an optional pipeline input: draws keep stencil bit 1
only while an active chain reads `mask`, and the window-sized mask texture is
built once per present for the chains that use it.
Tests: postfx_chain_test (format, built-ins, errors, file import, no GL);
gl_bloom_test gains an imported two-pass chain (byte-exact invert), the mask
input (HUD unlit, 3D lit), a bad preset, a pass that does not compile, and GL
state restored with bloom + a mask chain + FXAA on, all on the caller thread
and the render thread. docs/POST_EFFECTS.md documents the format.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
[video] bloom = 0.0 (off, default) .. 2.0 adds a glow around lights on the
OpenGL backend. PSX_BLOOM overrides it for one run; {"cmd":"bloom"} reads or
changes strength, threshold and a debug view live. It is a plain [video] key,
so [quality.*] presets can switch it off on low-end machines.
The HUD never glows: with bloom on every textured and untextured draw is
followed by two stencil-only passes of the same vertices that keep a
"may glow" mark in stencil bit 1 (bit 0 stays the mask-bit mirror). Opaque
primitives clear it under their coverage, then PGXP-proven 3D primitives and
additive semi-transparent texels/primitives (modes 1 and 3) set it, painter
ordered like the colour; 3D and non-3D primitives never share a batch while
bloom is on. Semi-transparent primitives leave the mark alone and the PGXP
depth-clear pass (colour writes off) marks nothing. Without PGXP geometry only
additive primitives glow.
The pass runs on the composed game image before AA and the OSD: the presented
surface's marks are blitted into a quarter-resolution stencil, a soft-knee
bright pass runs under it, three separable Gaussian iterations spread it and a
screen blend adds it back. The marks follow the picture: native-wide centre
copies and Smooth motion surfaces (redraw and reprojection) copy stencil with
colour, and generated frames get the same pass. Off: nothing runs, no state
changes, and the composed frame is byte-identical to master.
test_gl_bloom (gl_bloom_test): off is byte-identical before/after bloom and
with --dump against master; on, a bright 2D panel and a panel over a lit 3D
polygon are byte-identical to off, an additive and a 3D polygon glow; guest
VRAM is unchanged; on this thread and on the render thread.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
find_project_root walked up from a release's game.toml to the first
ancestor with a root marker. With a global ~/.gitignore (common) and the
game under ~/Games/<title>, that was the home folder, so [runtime]
memcard_dir = "saves" put memory cards and savestates in ~/saves instead
of beside the game. A folder whose licenses/psxrecomp-LICENSE was staged
by package_game_release.sh now stops the walk.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Owner decision: wheel force feedback is an optional player setting, off by
default. settings.toml [controller] jogcon_force_feedback (UserSettings,
load + save) overrides the title's game.toml
experimental_jogcon_force_feedback default. psx_jogcon_ffb_set_player_enabled
applies it live (off -> every wheel neutral now); the launcher's boot and
lobby soft-return exits seed and read RecompLauncherCSettings
.jogcon_force_feedback (RECOMP_LAUNCHER_HAS_JOGCON_FORCE_FEEDBACK,
RetroPortingToolKit/recomp-ui#95) and persist it. Tests:
jogcon_ffb_settings_test (default off, on, explicit off, round-trip);
jogcon_ffb_test (setting off: no haptic calls; toggled on: the hold drives;
off live: neutral).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review (mstan): the title key is now experimental_jogcon_force_feedback(
_invert) and the runtime logs it as EXPERIMENTAL. Force feedback goes through
a pure driver (psx_jogcon_ffb.h) and an SDL haptic backend
(psx_jogcon_ffb_sdl.cpp): every force has a 120 ms lifetime renewed each
VBlank, so a pause, stall or crash goes slack by itself, and the Rewind and
save-state pause loops, disconnect (close_player) and session end
(close_controller) stop it explicitly. Tests: jogcon_ffb_test drives the
driver against a recording backend (lifetime, renew, stop once, neutral,
idempotence, disable mid-force); jogcon_ffb_sdl_test runs the SDL backend
against an SDL3 virtual wheel. SDL has no virtual haptic device, so it checks
that such a device is not driven, never falls back to rumble and closes
cleanly. Also fixes the SDL3 haptic subsystem init check (psx_sdl.h returns
0 on success).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The emulated JogCon kept its motor command as device state only. With
game.toml [controller] jogcon_force_feedback = true, a presented JogCon's
motor byte (command nibble + strength) now drives a host wheel's SDL haptics:
commands 1 / 2 a constant force signed by direction and scaled by strength,
any other non-stop command a centring spring. Only pads whose haptic device
supports constant or spring effects are driven, so gamepads are untouched.
jogcon_force_feedback_invert flips the turn direction (not checked on JogCon
hardware). SIO keeps the strength nibble beside the command
(sio_get_pad_jogcon_motor). PSX_RUMBLE_TRACE logs the commands, headless too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Shokunin