* vulkan: start without fragmentShaderBarycentric on GPUs that lack it
Pascal (GTX 10xx), RDNA 1 and older drivers have no
VK_KHR_fragment_shader_barycentric, so device selection rejected them
with 'Could not find suitable device' (#1432, #1416, #762). The SPIR-V
backend only emits the capability for pixel shaders with
per-vertex/barycentric inputs, so make the extension and feature
optional: warn at selection and creation, enable only when supported.
Shaders that need custom barycentrics fail at pipeline creation
instead of blocking startup for every game.
* vulkan: fix macOS build and warn about unsupported GPUs
---------
Co-authored-by: Matthew Hurley <171001439+theantipopau@users.noreply.github.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
A float inline constant (240-248) read by a 64-bit operand is the f64 bit
pattern for integer B64 opcodes too: AMD's compilers emit
v_lshrrev_b64 v[4:5], v0, 1.0 for 0x3ff0000000000000 >> lane. ReadU32Pair
returned {f32 bits, 0}, so only F64-typed reads had the right value.
Based on TheCruZ's 284deab28 (closed#1219).
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
V_PK_MIN_F16 and V_PK_MAX_F16 return the quieted NaN for a signaling NaN
operand and the other operand for a quiet one (per half, as V_MIN_F16 and
V_MAX_F16 do). The signaling test masked the half with 0x7c00 and checked
for a nonzero payload, so a quiet NaN with payload bits such as 0x7e01 was
taken as signaling and returned. Mask with 0x7e00 so the quiet bit must be
clear.
Two more vectors in PackedMinMaxF16NanAndSignedZeroEdges cover it.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The 16-bit encodings of the float inline constants are fixed: the exact
halves of 0.5, 1.0, 2.0 and 4.0 and their negations, and 0x3118 for
1/(2*pi). Read16LaneBits converted the 32-bit value at run time instead,
through PackHalf2x16, which the AMD driver rounds toward zero, so
1/(2*pi) became 0x3117. Use the encodings directly.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* shader: support DS_CMPST_RTN_B32
Assassin's Creed Valhalla (PPSA01490, #281) stops on DS opcode 0x30 in a
compute shader. DS compares DATA and stores DATA2, the opposite of
BUFFER_ATOMIC_CMPSWAP, so the 32-bit atomic path swaps the two for this
opcode, as the 64-bit form does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* shader: reject unsupported GDS compare-store variant
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
A patched TLS load calls TlsRegStub and SafeCall, which moved RSP past the
red zone with sub/add rsp,0x80. ADD and SUB write the arithmetic flags, so
the sub ran before SafeCall's pushfq saved them and the add ran after its
popfq restored them: guest code that compares, loads fs:[0] and then
branches on the flags could branch the wrong way. lea moves RSP without
touching the flags. The entry lea plus a two-byte nop keeps every patched
offset in both stubs; the exit lea is one byte longer and moves only ret.
TlsStubsPreserveFlags calls the stubs with two flag patterns and checks
the status flags, the output register and RAX; it fails with the old
bytes.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Performance changes from #1495:
- Sync changed BDA ranges instead of scanning all mappings.
- Let VMA suballocate BDA buffers.
- Remove the renderer lock from presentation.
- Keep compatible pipeline caches across builds.
- Fold masked shifts that always produce zero.
* loader: keep the output of a process LoadExec relaunches
LoadExec on Windows starts a new emulator process and exits the current
one. The new process was created without a console and without the
standard handles, so everything it printed was lost: a redirected log of
a game that relaunches itself ended at the relaunch.
Hand the new process the standard input, output and error handles,
inheriting only those, and let it share the console when there is one.
* loader: flush output before LoadExec starts the new program
On Windows, LoadExec flushed the C streams only after the new process
had been created, so output the old process still held could appear
after the new one's first lines. On Linux they were not flushed at all
before exec, which discards them.
Flush the log and every stream before starting the new program.
* loader: continue the log file across LoadExec
With --printf-direction File, the program LoadExec starts is given the
same --printf-output-file and opened it anew, emptying it. Everything
the first program had logged, its LoadExec line included, was lost.
LoadExec now sets KYTY_RELAUNCHED for the program it starts, which then
appends to the log file instead of replacing it. On Linux the variable
goes into the environment passed to exec rather than this process's
own, which other threads may be reading.
* loader: show LoadExec switches on the console
LoadExec logged the program it switched to only when guest output was
logged, and then only into the log: on a console the new program's
startup output followed the old one's without anything marking where
one program ended and the next began.
Write the switch to the console as well as the log, with the program's
arguments.
* shader: pack v_cvt_pkrtz_f16_f32 with the native half conversion
The round-toward-zero pack was emulated with integer operations: about 20 to 30 operations
per half. GTA V uses it in 613 of its 656 pixel shaders, about 16 times in each G-buffer
pixel shader.
The pair is now packed with PackHalf2x16, unpacked again, and each half whose magnitude
exceeds its input's takes one unit off. Half signs and magnitudes are separate, so that is
the next value toward zero, whatever rounding the native conversion uses as long as it
returns one of the two neighbours. Same bits as before for every finite input and infinity;
a NaN stays a NaN. Checked against the old code on 2 million values on the CPU, and on an
Apple M5 GPU through Metal (400 000 values, subnormals included, safe and fast math).
GTA V on an Apple M5 (MoltenVK), 1080p, fixed script camera, three places, A B B A between
launches: -1.7 to -2.8 ms per frame (+4 to +6 %).
Scope: generic
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* shader: remove half-pack commentary and unused shift helper
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* agc: Cleanup names so they match NIDs
Simple commit tidying up libSceAgc function names so they match their
NIDs. This just a tidy-up commit that makes it more obvious what each
function is.
* agc: Preserve unknown fused shader size name
windows.h pulls in rpcndr.h, which defines `small` as `char`. The R64 BC1
array case added in b7b335ac declared `const auto small`, so the Windows
build of shader_recompiler_compute_tests failed to compile:
ShaderRecompilerComputeTests.cpp(17928,24): error: expected unqualified-id
Rename the local to `small_array`.
* agc: implement sceAgcSetNop
Star Wars Outlaws calls sceAgcSetNop (K2mciNVxUCE) and crashes right after
the unresolved stub returns, without the packet being disabled.
The call turns a packet into a NOP of the same length. Clear the emulator's
custom NOP code as well, so packets that the HLE encodes as NOPs, such as
end-of-pipe actions, stop executing too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* agc: keep packets disabled by sceAgcSetNop from reading as markers
A disabled packet whose first payload word carries the 0x6875 marker
signature still reached the marker handler, which can flip or stop on an
unknown marker ID. Clear the signature so the packet is skipped like any
other NOP.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* agc: tidy SetNop logging and comments
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* controller: use Back as the touchpad click on pads without a touchpad
Gamepads without a touchpad (Switch, Xbox, 8BitDo, ...) cannot click it, so
games that require the touchpad can't be played with them. The keyboard
already emulates it (Backspace/Tab); gamepads had nothing.
For SDL_GAMEPAD_BUTTON_BACK on a pad that reports no touchpads, report
PAD_BUTTON_TOUCH_PAD plus a finger at the centre of the touchpad. Pads with a
real touchpad are unchanged.
Fixes#1298
* controller: establish touch before fallback click
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
tests: avoid empty optional::emplace() on IndirectDescriptor (fixes Linux build)
clang with libstdc++ 14 does not treat std::optional<DescriptorSource::IndirectDescriptor>
as default-constructible (nested struct with default member initializers), so the
argument-less emplace() fails to resolve. Pass an explicit value-initialized
IndirectDescriptor, matching the form already used elsewhere in this file.
WWE 2K26 stops building a shader CFG with "unsupported family=VOP2
opcode=0x11 raw=[0x220404f9 0x00060605] reason=VOP2 SDWA modifier is not
supported for opcode".
V_MIN_I32 takes the same integer path as V_MAX_I32, so give it the same
full-destination SDWA profile. The test runs the game's encoding, which
takes the signed minimum with the zero-extended low byte of a register.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Glyph images were rendered into a fixed 128x128 buffer and their width and
height were clamped to 128. Since glyphs are rasterized at the requested
scale (f96a14a0), large text such as title cards lost every row and column
past 128 pixels, so letters were drawn cut into bands.
The glyph buffers now grow to the glyph's bitmap box and width_byte is the
glyph's width. The VGA fallback font keeps its 128-pixel limit; scalable
glyphs are limited to 4096 pixels only to reject absurd scales.
* cmake: accept glslang's current tool name and document Python 3
Official glslang release archives for Windows ship the compiler only as
glslang.exe, so configuring with one of them on PATH could not find
glslangValidator. Fall back to glslang; glslangValidator is still
preferred where both exist.
The bundled SPIRV-Tools needs Python 3 at configure time, which the
Windows build requirements did not mention.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* docs: list Python 3 in the Linux and macOS prerequisites
The bundled SPIRV-Tools calls find_host_package(Python3 REQUIRED) on every platform, so the apt and Homebrew package lists need it too, not only the Windows requirements.
* cmake: remove glslang naming comment
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* ampr: keep resolved host paths in native form
ResolveOnePath stored each resolved file as a UTF-8 std::string
(Common::PathToString) and the read, stat and size paths turned that
string back into a std::filesystem::path. On POSIX this round trip is a
no-op, but on Windows path(std::string) decodes the bytes with the ANSI
code page, not UTF-8 (/utf-8 only sets the source and execution
charsets). Any non-ASCII character in the game folder therefore produced
a different path: the resolve step found the file through the native
path, then every APR read of it failed with ENOENT.
Tokyo Xtreme Racer (#1156) installed under "D:/Инсталляторы/[ps5]/..."
hits this on its first pak read:
APR read missing host file: .../tokyoxtremeracer/content/paks/global.utoc
Apr read failure 80020002 at CB offset 0 (LowLevelFatalError, abort)
even though KernelStat and the directory listing see the same file.
Store std::filesystem::path in ResolvedPathInfo and the file-id table
and pass it straight to Common::File. Convert to UTF-8 only for the log
lines.
* tests: cover APR stat and size under a non-ASCII host folder
Resolve a file inside a Cyrillic directory through APR and check that
the stat and size exports find it by ID, matching the Tokyo Xtreme
Racer install layout from #1156. On POSIX the old string round trip
also passed, so this guards the Windows path.
* tests: read the non-ASCII APR fixture through an APR command
Review on #1175 noted that the Unicode fixture only checked the stat
and size exports. A submitted APR read opens the stored host path in
ExecuteCommand/ReadHostFileToGuest, which the fixture did not reach.
Add AprReadMatches, which submits one APR read command for a file ID
and compares the bytes, and use it on the file under the Cyrillic
folder. As with the stat and size checks, POSIX hosts run this path but
only Windows would show an ANSI code page round trip.
* Remove APR path tests from this change
* Remove unrelated formatting changes
* Remove native path explanatory comments
---------
Co-authored-by: overlordxrz-source <overlordxrz-source@users.noreply.github.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* libNet: answer sceNpGetAccountLanguage2 as a signed-out user
sceNpGetAccountLanguage2 was not registered, so games that call it hit
the unresolved-import stub. The stub returns 0, which tells the game the
request succeeded and that the language code buffer now holds the
account's language, but nothing was written to it.
Saros (#1053) calls it from its analytics thread right after
sceNpGetAccountAge, each on its own request. sceNpGetAccountAge already
reports SCE_NP_ERROR_SIGNED_OUT there, so the game is told the user is
signed out for one query and signed in for the next. Demon's Souls
(#890), Ghost of Yotei (#1095) and NHL 27 (#1101) import it as well.
Register it next to the other NpManager account queries and answer the
way they do: SCE_NP_ERROR_INVALID_ARGUMENT for a request id <= 0 or a
null output pointer, SCE_NP_ERROR_SIGNED_OUT otherwise. The output is
left untouched because nothing was looked up.
* libNet: reject missing account language requests
---------
Co-authored-by: overlordxrz-source <overlordxrz-source@users.noreply.github.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* Pad: report the adaptive trigger state instead of always 0
scePadGetTriggerEffectState returned 0 for both triggers. Games that wait for
the state of the trigger effect they set before acting on L2/R2 never saw it
change; in Astro Bot, L2/R2 actions such as punches did nothing in some levels.
The state now follows from the game's last effect per trigger (already kept
for the DualSense output) and the trigger's travel, 0..255 over the effect's
ten positions: feedback no force / pushing, weapon not / almost / fully
pressed, vibration not firing / firing, with the values 1..7 that duaLib
(WujekFoliarz/duaLib) reads from a DualSense input report. SDL does not expose
that report, so this is an approximation; it also works with other pads and
the keyboard, where a digital L2/R2 is full travel. The trigger intensity
setting changes only the output, not the reported state.
Tested in Astro Bot (PPSA21567 1.018) by a player in the levels where the
triggers did nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* tests: controller_settings covers scePadGetTriggerEffectState
Feedback, weapon and vibration states across the trigger travel, a trigger without an effect, a masked update keeping the other trigger's effect, and the intensity setting leaving the reported state unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Pad: a feedback trigger reports pushing only while it is pressed
A feedback effect from position 0 reported FEEDBACK_IS_PUSHING on an untouched trigger, so a game waiting for the change from no force to pushing never saw it. In Astro's Playroom the gacha capsules did not break on R2. Feedback states (modes 1, 4, 5) now need some travel (>= 8 of 255). Tested in Astro's Playroom (PPSA01325): the capsules break on R2 again.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Pad: a vibration trigger reports firing only while it is pressed
A vibration effect from position 0 reported VIBRATION_IS_FIRING on an untouched trigger; in Astro's Playroom the gun fired by itself. Vibration states (modes 3 and 6) now need some travel, as the feedback states do. Tested in Astro's Playroom.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Pad: reject invalid handles in scePadGetTriggerEffectState; test the remaining effect modes
scePadGetTriggerEffectState ignored its handle and returned pad 1's state for any handle. It now returns
SCE_PAD_ERROR_INVALID_HANDLE (0x80920003) for handles other than 1, like the other pad queries in libPad.cpp
and duaLib.
controller_settings now also covers multiple-position feedback (mode 4: pushes only at positions with a
strength), slope feedback (mode 5: pushes from its start position, not on an untouched trigger) and
multiple-position vibration (mode 6: fires only at positions with an amplitude and a frequency).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Pad: scePadGetTriggerEffectState returns libScePad's invalid-argument code
A null state returned 0x80960005, which is SCE_USER_SERVICE_ERROR_INVALID_ARGUMENT,
not a pad error; libScePad's code is SCE_PAD_ERROR_INVALID_ARG (0x80920001), as
scePadSetTriggerEffect returns. The call now lives with the other pad calls in
controller.cpp so controller_settings tests its handle and argument checks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Pad: correct trigger effect state reporting
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* tests: cover the 16-bit integer VOPC compares
V_CMP_{EQ,LE,GT,NE}_I16 and V_CMP_{EQ,LE,GT,NE,GE}_U16 decode and translate but
no test exercised them. VectorCompareI16U16Ops compares low halves that
differ in sign interpretation (0x8000 vs 0x7fff) with both the signed and the
unsigned opcodes, and uses operands whose high halves differ to check that only
the low 16 bits are compared. It runs in the default list and with
--int16-compare-only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* tests: cover the 16-bit integer VOP3 ALU ops
V_SUB_NC_U16, V_ADD_NC_I16, V_ASHRREV_I16, V_MAX_U16 and V_MIN_U16 decode and
translate but no test exercised them. Vop3Int16Ops checks wraparound without
clamp, the shift amount taken from the low four bits of src0, unsigned
ordering for max/min on 0x8000 vs 0x7fff, that the high halves of the sources
are ignored and that the high half of the destination is preserved. It runs in
the default list and with --int16-alu-only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* tests: cover the packed 16-bit integer VOP3P ops
V_PK_MAD_I16, V_PK_MAD_U16, V_PK_MUL_LO_U16, V_PK_ADD_I16, V_PK_SUB_U16,
V_PK_LSHLREV_B16, V_PK_LSHRREV_B16, V_PK_ASHRREV_I16, V_PK_MIN_I16, V_PK_MAX_U16
and V_PK_MIN_U16 decode and translate but no test exercised them. Vop3pInt16Ops
runs each on both halves with the default half selection: wraparound without
clamp, shift amounts from the low four bits of each src0 half, and min/max on
0x8000 vs 0x0001 so signed and unsigned ordering give different results. It
runs in the default list and with --int16-alu-only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
TestConcurrentBackingReads (7e011fa5) builds its reader pairs with
std::pair {TryReadBacking, TryReadSparseBacking}. Class template argument
deduction then also considers pair's const T1& / const T2& constructors, which
deduce function types, and libstdc++ and libc++ reject a pair member of
function type ("data member instantiated with function type"), so
virtual_memory_allocation_tests no longer compiles on Linux and macOS. MSVC's
library accepts it, which is why Windows still builds.
Take the functions' addresses so the pairs hold function pointers.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* systemservice: add an option to report the notice screen as skipped
sceSystemServiceGetNoticeScreenSkipFlag always returned false. Return the
new skip_notice_screen option instead. Games that check this flag skip
their boot logo movies when it is set; Marvel's Wolverine, for example,
then does not queue movies/BootSequence/BootSequence.movie.
The option defaults to off, so behaviour is unchanged unless it is
enabled with --skip-notice-screen true or the launcher's "Skip boot
logos" checkbox. The launcher stores it with the other per-game settings,
so it is included in per-game config import and export.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Shorten skip notice screen wording
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* feat(shader): emulate DS_ORDERED_COUNT as a GDS atomic
* fix(shader): issue one ordered-count atomic per wave
* shader: correct ordered-count addressing and wave results
Use the M0 base and pixel packer offset for the GDS counter. Broadcast the
previous value to every lane and preserve the counter when EXEC is empty.
Add GPU regressions for add/exchange with full, partial, and empty EXEC in
wave32 and wave64. The targeted DS and buffer atomic suite passes 63 cases.
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* kernel: apply clang-format to pthread.cpp
Output of the repository's pre-commit clang-format hook (v22.1.3) on
this file; no code change.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* kernel: fix the macOS build of host thread naming
199d0a9d (#1106) names guest threads on the host with the two-argument
glibc pthread_setname_np(thread, name). KYTY_PLATFORM is
KYTY_PLATFORM_LINUX on macOS too, so the macOS job now fails with "no
matching function for call to 'pthread_setname_np'": Apple's version
takes only the name and renames the calling thread. This code already
runs on the new thread itself (it passes pthread_self()), so use the
one-argument form on Apple with Apple's 63-character limit and keep the
Linux call unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Remove formatting
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
PthreadCreate carries a guest thread name, but RunThread never handed it to
the host, so every guest thread showed up as the same generic host thread
name in core dumps, perf samples and htop.
Call pthread_setname_np() on Linux so host-side diagnostics can attribute
work to the guest thread that caused it.
avplayer_file_tests compiles libs/avPlayer.cpp into the test, and its
PRINT_NAME() calls Loader::Timer::GetTimeMs(). Add loader/timer.cpp to the
target, like the other tests that compile HLE library sources. CI links the
test with LTO, which drops the unused reference, so the failure only shows
in non-LTO builds such as the README's no-Qt instructions.
Clang with libstdc++ rejects optional<IndirectDescriptor>::emplace() with
no arguments in ShaderRecompilerComputeTests.cpp. Assign the descriptor the
same way the neighbouring test does.
trophy_system is registered with ctest but kyty_tests did not build its
executable, so ctest reported it as Not Run.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* loader: enable AMD CPU instruction patching on macOS
The guest instruction patcher was compiled out on macOS and runtimeLinker forced emulate_amd off,
so under Rosetta every SSE4a EXTRQ/INSERTQ took the SIGILL path (about 18 us each). Build the
patcher on all hosts and let --amd-cpu enable it on macOS like on Windows and Linux. The host
feature probe already reports no SSE4a under Rosetta, so EXTRQ becomes a native trampoline and the
remaining fallbacks keep using the x64 instruction emulator through the host exception handler.
* tests: macos_sse4a covers the guest instruction patcher on macOS
Patch EXTRQ/INSERTQ blobs in an RWX area with a trampoline page behind it, with the host features of
Rosetta (no SSE4a/RDPID/CLWB), and compare against the SIGILL emulator: EXTRQ must become a native
trampoline without SIGILL, INSERTQ stays a trap handled by the emulator, and nothing is touched
without emulate_amd.
* loader: emulate the patcher's VRSQRTPS trap and MWAITX on macOS
The AMD CPU patcher marks VRSQRTPS it cannot relocate by clearing a reserved
VEX.vvvv bit, which raises #UD on every host including Rosetta. The macOS build
left TryEmulateReciprocalSquareRoot and TryEmulateMonitorxMwaitx out, so the
guest died with an unhandled SIGILL at the first such site.
Both are plain SSE2/libc, so enable them on macOS. The Darwin context also
gains ClearUpperYmm so the VEX.128 result zeroes YMM[255:128] when the thread
has the AVX context. macos_sse4a covers the trapped VRSQRTPS and MWAITX.
* tests: macos_sse4a runs a plain VEX.128 VRSQRTPS through the patcher
The existing case started from an instruction that was already marked as a trap. The
new cases feed the ordinary encoding to PatchGuestInstructions and compare the result
with the emulated reference, both through a native trampoline and, when the instruction
cannot be relocated, through the trap mark and the signal emulator.
* loader: compact macOS instruction patcher coverage
Share the emulator dispatch chain while keeping SHA emulation limited to non-Apple hosts and preserving handler order. Remove a duplicate VRSQRTPS signal test already covered by both native and trapped patcher cases, plus blank lines left by removed platform guards.
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* Map UDS events to trophies via dumped metadata and redesign trophy UI
Correlate each UDS event to its stat ID (uds00.ucp) and trophy target (trophy00.ucp) instead of a hard-coded event, store trophy JSON with unlock date/time outside _SaveData, and add a PS5-style toast, viewer and overview.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
* Fix trophy rule evaluation and share progress accounting
* Match trophy and reload toolbar styling
---------
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* libNet: honour the option mask in sceHttpUriBuild
sceHttpUriBuild's `option` argument (SCE_HTTP_URI_BUILD_WITH_*) selects which
components of the SceHttpUriElement end up in the output. HttpUriBuild logged
the value and ignored it, always returning the full URL.
Black Myth: Wukong calls it with SCHEME alone and requires "https://" (or
"http://") back before it starts an HTTP request. The full URL fails that
check, so the request starter returns false. The caller's "run on game thread
and wait" helper then returns early while its task stays queued, still holding
by-reference captures of a stack frame that is already gone. About 70 seconds
into boot the game thread runs that task against the dead stack and faults
(null read at eboot+0x59662f). With the mask honoured the game gets past that
point and reaches its title cinematic and menu.
Bits: SCHEME 0x01, HOSTNAME 0x02, PORT 0x04, PATH 0x08, USERNAME 0x10,
PASSWORD 0x20, QUERY 0x40, FRAGMENT 0x80. A mask with none of the eight bits
set keeps the previous "everything" behaviour for callers that never pass one.
A scheme-only build returns "https://", the authority marker belonging to the
scheme chunk.
Adds TestBuildHonoursOption to http_uri_parse covering scheme-only, hostname,
port, path, and the 0xff / 0 cases.
* libNet: simplify URI build code and tests
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
Comment out the five explicit flush checks and record the substantial 3DMiniGolf menu FPS drop on NVIDIA RTX 5080 Laptop GPU. Preserve separate multiply/add rounding and both NoContraction decorations.
The matched runs used silent logging, no debugger, no Vulkan validation, and the cd9fc6ff revert on both revisions. Native denormal behavior remains GPU-dependent.
On Linux (PipeWire / ALSA), the 4-channel USB audio playback device for
the DualSense controller is named 'Wireless Controller' based on Sony's
USB product descriptor string. Previously, FindDevice() only matched
names containing 'DualSense', causing quad-channel endpoint discovery to
fail on Linux systems.
Add 'Wireless Controller' to the allowed endpoint names when 4 channels
are present, and add a test case verifying discovery under this name.
All three build badges pointed at the status of the whole build.yml
workflow, so a failure in one platform's job turned every badge red.
Point each badge at its own job through the check-runs badge with a
name filter, so each one reports only its platform.
* window: auto-hide the cursor over the game window
Hide the cursor when the game window is created and whenever the mouse
has been idle for two seconds; any mouse motion, button or wheel event
shows it again. Games that draw their own cursor (e.g. Blue Prince) no
longer show the host pointer on top of it.
Only real mouse input counts as activity: synthetic touch/pen mouse
events and zero-delta motion (sent by some platforms on focus or cursor
visibility changes) are ignored. The event loop only waits with a
timeout while the cursor is visible, so it still blocks indefinitely
once the cursor is hidden. The cursor is restored on shutdown.
Uses SDL_HideCursor/SDL_ShowCursor, which SDL implements for X11,
Wayland, Cocoa and Win32 and which only affect the cursor over the
emulator window.
Closes#282
* window: make cursor auto-hide optional in CLI and launcher
Co-authored-by: Anatolii Kucheruk <anatolii.kucheruk01@gmail.com>
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
Co-authored-by: Anatolii Kucheruk <anatolii.kucheruk01@gmail.com>
* pad: send DualSense speaker and haptic audio over Bluetooth
* ci: fix Opus clang-cl build and macOS memory tracker test
* docs: remove Bluetooth notes from project README
* pad: keep Bluetooth audio work off emulation timing path
* pad: smooth DualSense speaker audio on USB and Bluetooth
* tests: find available macOS guest range for page manager
* pad: keep haptic waveform when Bluetooth queue overflows
* audio: reject zero output sample frequency
* Simplify DualSense audio routing and fix Bluetooth batching
* controller: add global DualSense volume and vibration controls
* Simplify global controller settings
---------
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
CMake 4 warns "Compatibility with CMake < 3.10 will be removed" for every
fetched project that declares an older minimum. zydis and its bundled
zycore both declare cmake_minimum_required(VERSION 3.9), so each configure
prints two deprecation warnings. The existing floor of 3.5 does not lift
them past the threshold; 3.10 does.
Existing build directories keep the cached 3.5 until reconfigured with
-U CMAKE_POLICY_VERSION_MINIMUM or a fresh cache.
Tested on Windows 11, clang-cl 19.1.5, CMake 4.4.3: configure prints 0
deprecation warnings (2 with the old floor); virtual_memory_allocation_tests
builds and links zydis/xbyak; ctest guest_red_zone_patcher passes.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* controller: cycle DualSense speaker volume, vibration and trigger intensity in game
Three new input mapping controls cycle the PS5's controller settings in
game (default keys 1, 2 and 3): speaker volume (100% -> off -> 2% -> 12%
-> 42%), vibration intensity and trigger effect intensity (strong -> off
-> weak -> medium). Each starts at full and is not saved.
Changing vibration or trigger intensity re-sends the game's last rumble
and trigger effect, so the change is felt immediately.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Clean up controller settings test captures and formatting
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* pad: send DualSense speaker and haptic audio over Bluetooth
* ci: fix Opus clang-cl build and macOS memory tracker test
* docs: remove Bluetooth notes from project README
* pad: keep Bluetooth audio work off emulation timing path
* pad: smooth DualSense speaker audio on USB and Bluetooth
* tests: find available macOS guest range for page manager
* pad: keep haptic waveform when Bluetooth queue overflows
* audio: reject zero output sample frequency
* Simplify DualSense audio routing and fix Bluetooth batching
---------
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* network: handle SO_NBIO on non-Windows hosts
Setsockopt only translated the guest SOL_SOCKET/SO_NBIO (0x1200) option
under _WIN32; other hosts rejected it with ENOPROTOOPT. FMOD Studio's Live
Update listener sets SO_NBIO, so on Linux its socket setup failed and the
Studio async manager latched FMOD_ERR_NET_CONNECT. Every blocking Studio
call then failed: loadBankFile returned FMOD_ERR_NOTREADY for all banks,
leaving the game silent and stuck on the splash screen waiting for
RuntimeManager.HaveAllBanksLoaded (PPSA28997).
Toggle O_NONBLOCK with fcntl on non-Windows hosts.
* network: map connection errors from POSIX socket calls
ConvertHostSocketError only covered a handful of errno values on
non-Windows hosts, so anything else reached the guest as EIO. With
SO_NBIO now honoured there, a nonblocking connect() that is still in
progress returned EIO instead of EINPROGRESS (and EALREADY on retry).
Map the same connection and network errors the Windows branch already
translates.
* network: share nonblocking socket error handling
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
launcher: configure and preview controller lightbar color
Add an optional lightbar color to launcher settings and pass it to the emulator with --controller-color. Preview changes on RGB gamepads, handle controller hotplug events, and restore the saved color when settings dialogs close. Release launcher preview handles when starting a game.
Co-authored-by: m66401970-eng <m66401970@gmail.com>
* window: explain why each GPU was rejected in device selection
"Could not find suitable device" gave no hint of what was missing; the
reasons only went to the log file, which users often cannot find (#489).
Device selection now records each rejection reason and the fatal error
lists every GPU with its reasons, for example:
Could not find suitable device:
AMD BC-250 (RADV GFX1013): fragmentShaderBarycentric is not supported
The missing-extension check also names each missing extension instead of
stopping at the first one without saying which. Device requirements and
selection are unchanged; the log keeps the same messages.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* window: simplify GPU rejection reporting and capability checks
* window: identify storage-image format failures accurately
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* audio: play pad speaker ports on the DualSense speaker
Route the right channel to the speaker with the audio-control values the Linux hid-playstation
driver uses without headphones, and fall back to the main output while no DualSense audio device
is active. Also stop the previous DualSense's rumble when another DualSense becomes active.
* audio: harden DualSense speaker routing and fallback
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* libs: fix CP932 iconv conversion on Windows and macOS
SDL3's SDL_SYSTEM_ICONV build option defaults off on Windows and macOS,
so SDL falls back to its own bundled iconv there, which only implements
ASCII/Latin-1/UTF/UCS. SDL_iconv_open("UTF-8", "CP932") always failed as
a result, crashing games that convert Shift-JIS strings via
sceCesMbcsStrToUtf8Str (e.g. save-data dialogs).
Add a small CesIconv shim selected at compile time: Windows converts one
character at a time via the Win32 codepage-932 APIs, replicating
SDL_iconv's chunked semantics and error codes; macOS calls real POSIX
iconv directly, since libiconv is always present there and does support
CP932, unlike SDL's fallback. Linux is untouched, since SDL already
links real system iconv by default there.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* libs: fix SaveDataDialog infinite loop on save
SaveDataDialogOpen() jumped straight from SAVE_STATUS_INITIALIZED (1) to
SAVE_STATUS_FINISHED (3), skipping SAVE_STATUS_RUNNING (2) entirely. At
least one title's save-write dialog stage only accepts a FINISHED status
after observing a RUNNING poll first; without it, the title assumed the
Open() request never took effect and re-issued it forever, hanging the
save flow (this only became visible once the CP932 crash above stopped
masking it).
Open() now starts the dialog as RUNNING. UpdateStatus() advances the
state machine, reporting RUNNING once before flipping to FINISHED for
subsequent polls; GetStatus() is untouched, staying a pure peek with no
side effects, matching the real SDK's pump-vs-peek split between the two
calls.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* Add iconv linking to macOS
* Fix missing macOS iconv link for tests
* Fix CP932 edge cases and save dialog status consistency
---------
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* docs: link weekly updates from README
* docs: add Discord development channel
* docs: use Discord server invite link
* docs: refine weekly updates wording
* audio: play DualSense haptics from vibration audio ports
Games drive the DualSense's actuators through a vibration audio port, one channel per actuator.
These ports had no device, so their samples were dropped. Over USB the DualSense is a 4-channel
audio device whose back channels feed the actuators; play the port's samples there through SDL.
The device is looked up while the port plays, so the controller can be connected at any time. The
queue keeps about the latency of the regular audio path, with the playback rate nudged by up to 3%
when the port is fed off real time.
While haptic audio plays, hold a DualSense's rumble at zero: SDL disables audio haptics whenever
rumble is non-zero.
Add pad_haptics tests for the sample conversion, device-name matching and rate correction.
* Simplify DualSense audio haptics and restore idle rumble
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* AvPlayer: advance the shared audio-master clock after final audio frame
* AvPlayer: initialize padded NV12 planes with neutral black
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
Build and run the CPU-only resource_tracking, scalar_provenance, resource_materialization, and shader_vertex_metadata tests on Windows, macOS, and Linux. Each test executable is excluded from the default build, so add it to the explicit CMake targets and the matching CTest selection. Leave the Vulkan-dependent shader_recompiler_compute suite for a separate CI setup.
* launcher: keep primary update feed answer when the fallback request fails
When the primary feed is reachable but its tag differs from the build tag,
the result was discarded and the GitHub API fallback was consulted. If the
fallback request failed (api.github.com rate-limits unauthenticated
requests), the launcher reported an error even though the primary feed had
already returned a usable update answer with a working release URL. Keep
the primary feed answer and show it if the fallback fails.
* emulator: validate numeric CLI arguments
--screen-width, --screen-height, --gpu, and --vblank-frequency were parsed
with Common::ToInt32/ToInt64, which ignores std::from_chars errors and
returns 0, so values like "--screen-width abc" or "--gpu xyz" were
silently accepted: zero screen size, or GPU 0 selected instead of auto
(-1). Reject invalid values with an error message like the other options,
using from_chars-based helpers (ParseUint32/ParseInt32); reject zero
screen dimensions.
* tests: add LeastRecentlyUsedCache regression tests
Cover the LRU cache walk order, tick filtering in ForEachItemBelow,
early-exit boolean callbacks, Touch reordering and same-tick skips,
Free, and freed-id reuse by later inserts. Modeled on BitArrayTests.
* tests: use a tick cutoff that detects a stale newest-item tick
Collect(cache, 35) also visits c when Touch(c, 35) failed to update
its tick (30 <= 35). Use a cutoff of 30 and expect only b.
* Run LRU cache tests in CI
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* capture PCM at set time when ports share a buffer
* Snapshot size limit and open device check
* Fall back to PCM pointer on stale snapshot
* audio: own AudioOut2 PCM at attribute set time
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
SDWA source extraction and destination insertion are independent. Cover the reported word1-to-word0 preserve form and byte-source partial destinations in compute tests. Verify 64 distinct compute buffers survive resource tracking and binding layout, and remove the obsolete CFG rejection.
(cherry picked from commit 9e74df7dc0)
* docs: fix stale README notes, document building without Qt
- Use the current issue template name, Game Emulation Status Report
(renamed in 027433f).
- Raise the documented Windows CMake minimum from 3.12 to 3.22.1, which
the bundled SPIRV-Tools requires.
- Note that -DKYTY_BUILD_LAUNCHER=OFF builds kyty_emulator and the
regression tests without Qt (#193).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
* docs: add Qt-free Linux build commands
---------
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* audio: support standard NGS2 PCM16 sampler playback
Extend the existing fixed-point PCM path to standard samplers, loops, and progress reporting. Apply voice events immediately and honor callback interruption during source advancement.
Reuse upstream rate and pitch handling. Add regression coverage for streaming, routing, callbacks, loops, and custom sampler controls; ngs2_sampler passes.
* audio: apply NGS2 rear low-pass filters
Handle rear low-pass FCQ controls at the output sample rate with per-channel history and channel masks. Preserve filter tails through streaming gaps and use the standard pi constant for coefficients.
Add regression coverage for frequency response, control layout, channel isolation, and starvation/refill continuity. Emulator build and both audio test suites pass.
* Preserve resampling phase for accepting voices
Only reset voice.sample_phase when there are no blocks and the voice does not accept blocks (change in src/libs/ngs2.cpp). Add TestResamplingPhaseAfterStarvation to tests to ensure sampler and custom sampler retain resampling phase and pending source advancement after starvation, and that SetupSampler still resets the phase. This prevents resampling phase loss during starvation/refill across sample rates.
* audio: correct sampler controls, lifecycle, and filter tails
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* audio: keep blocking AudioOut writes on an even cadence
The blocking queue check only enforced a minimum cushion, so the guest
could queue a burst of buffers and then be blocked for the whole cushion.
Measured on PPSA03647, intervals between guest writes alternated between
0 ms and ~44 ms instead of one 5.33 ms buffer period, while the average
rate stayed correct at 48000 frames/s. Mixers that run on a timer cannot
track that cadence and the music turns slow and warbling.
Keep the cushion as-is, but also hold writes to the buffer period, and
skip that wait when the stream is running low so it cannot underrun.
Intervals become ~5.4 ms with microsecond-level deviation and the menu
music plays correctly.
* audio: reuse output clock after SDL queue priming
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* shader: support V_CMPX_EQ_U16
Decode gfx11 VOPC 0xba and lower it as an unsigned 16-bit equality
compare with CMPX EXEC-mask semantics.
🤖 Generated with Codebuff 🤖
Co-Authored-By: Codebuff <noreply@codebuff.com>
* shader: correct CMPX EQ U16 runtime expectations
Remove the added CFG tests and correct two EXEC expectations in the compute regression case.
---------
Co-authored-by: Matthew Hurley <171001439+theantipopau@users.noreply.github.com>
Co-authored-by: Codebuff <noreply@codebuff.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
The per-frame ACB submit, indirect dispatch, reset-queue and patch-address
parameter dumps, and the flip-pending poll, are logged only with the graphics
debug dump, the way the DCB submit path already is. Otherwise they were about
7.5 MB of log a minute in the Spider-Man world.
LINK_FLAGS is a space-delimited string, so when the build path contains
spaces the linker map option is split into bogus arguments and the final
link fails:
clang++: error: no such file or directory:
'pro/KytyPS5/_Build/linux/kyty_emulator_clang_lld.map'
Move -Wl,-Map=<path> out of LINK_FLAGS into target_link_options, which
CMake passes to the linker as a single argument. This completes #26 for
the WIN32 OR LINUX branch (clang/gcc builds); the clang-cl branch was
already converted there.
* graphics: hoist loop-invariant checks out of draw/attrib hot paths
DrawIndirectMulti recomputed the index_type_and_size switch on every
sub-draw of a multi-draw-indirect call, and ShaderApplyAttribSemantics
re-checked the debug-dump config flag on every vertex attribute. Both
values are invariant across their respective loops, so compute them
once before iterating instead.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* graphics: cache vertex buffer descriptor size in AcquireVertexBuffers
VertexBufferDescriptorSize was computed twice per vertex-buffer slot
on every draw call: once while collecting guest ranges, and again
while rebuilding slot bindings. Both calls use identical inputs, so
cache the first pass's result and reuse it in the second.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
Preserve live lane predicates through RDNA2 whole-quad expansion
Keep the source-live predicate when WQM expands a scalar mask, so entry masks remain known live without an instruction pre-scan. Preserve scalar mask values, SCC, half writes, and raw-source behavior.
Cover entry pixel masks, cross-half copies, and scalar/high-word sources with existing shader regression suites.
Co-authored-by: nmzik <Nmzik@mail.ru>
Co-authored-by: Douglas <douglasproducoes21@gmail.com>
Decode GFX10 DS opcode 0xa0 as DS_WRITE_B8_D16_HI (next to
DS_WRITE_B16_D16_HI at 0xa1) and lower it through the existing 8-bit
shared-memory write. The stored byte is bits 23:16 of the data VGPR,
selected with SDWA BYTE_2 on src1.
Add captured-instruction decode coverage (from GTA V, PPSA04264) and a
runtime regression that writes one byte of a distinctive source value
into a pre-filled LDS dword.
* Add `stretch to window` support
* Address comment
* Allow window resizing by default and recreate swapchain on resize
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* launcher: add a game list refresh action
Dropping new games into an already-configured game folder was only
picked up after restarting the launcher or re-saving the global
settings dialog. Add a refresh toolbar button, an F5 shortcut, and a
"Refresh game list" context-menu entry, all triggering a rescan of the
game folders.
* launcher: do not rescan the game list while a game is running
Clearing cfgs_list destroys the running ConfigurationItem: the running
state is lost and the replacement items allow editing or deleting a
game whose process is still active. Skip the rescan while any item is
running (CodeRabbit review finding for the refresh action).
* launcher: preserve game state when refreshing the list
Share one refresh action across the toolbar, keyboard shortcut, and context menu. Retain running items during rescans, update their settings for the next launch, and preserve selection, filtering, and sort order.
Commit active inline editors before rebuilding rows to avoid losing comments or accessing destroyed configuration data. Tie editor callbacks to item lifetime and guard the context-menu item reference.
Validation: launcher build and startup check passed; 32 Qt interaction checks passed.
* launcher: remove refresh from the game context menu
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* remove outdated ninja setup action
* Remove deprecated ninja setup action
The seanmiddleditch/gha-setup-ninja action has been archived by its
maintainer, who explicitly states that Ninja is now pre-installed on
all GitHub-hosted runners. This removes the action from all three
build jobs (Windows, macOS, Linux) and removes ninja from the package
install lists on macOS and Linux.
* Add automatic ccache/sccache detection to CMakeLists.txt
Detects sccache (preferred on Windows/MSVC) or ccache at configure time
and sets CMAKE_C_COMPILER_LAUNCHER / CMAKE_CXX_COMPILER_LAUNCHER
automatically. Works for local builds without manual flags.
Priority: sccache > ccache (sccache handles MSVC PDBs better).
* Delete github-actions-version-research.md
* Respect user-specified CMAKE_C_COMPILER_LAUNCHER / CMAKE_CXX_COMPILER_LAUNCHER
Only auto-detect sccache/ccache if no launcher was explicitly provided
via -DCMAKE_C_COMPILER_LAUNCHER=... or -DCMAKE_CXX_COMPILER_LAUNCHER=...
* Preserve explicit compiler launcher settings per language
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* shader: add S_BITCMP0_B64 and S_BITCMP1_B64
* test: move S_BITCMP_B64 coverage to compute suite
* test: use the existing compute CTest entry for bit comparisons
* test: remove the dedicated S_BITCMP_B64 command-line selector
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
common: don't follow reparse points in the Windows recursive file walk
SysFileFindFiles recurses into every entry carrying
FILE_ATTRIBUTE_DIRECTORY, which on Windows includes directory symlinks
and junctions. A junction pointing at one of its own ancestors makes the
walk recurse until the stack runs out.
The Linux implementation already guards against this: it uses lstat
rather than stat specifically so a symlink is never followed into a
cycle. Skip reparse-point entries on Windows for the same reason.
Co-authored-by: nmzik <Nmzik@mail.ru>
GetCacheAccessType returned the sys_file_cache_type_t enum values as
CreateFileW flags: SYS_FILE_CACHE_SEQUENTIAL_SCAN (2) became
FILE_ATTRIBUTE_HIDDEN and SYS_FILE_CACHE_AUTO (0) lost the normal
attributes, and no FILE_FLAG_SEQUENTIAL_SCAN hint was ever applied.
Return FILE_FLAG_SEQUENTIAL_SCAN / FILE_ATTRIBUTE_NORMAL instead,
mirroring the POSIX fadvise handling on Linux.
Preserve native Windows paths for filesystem operations, command-line options,
and game icons. Complete Unicode handling in recursive file listings and file
logging, restore unmounting by host directory, and extend regression coverage.
Original implementation by Augusto Domingos (@AghastyGD); retain all five of
his PR commits and include the reviewed follow-up fixes.
Closes#714
Co-authored-by: Augusto Domingos <augustodomingosvasco@gmail.com>
* Persist save-data memory across restarts
* Fix SetSaveDataMemory2 data loss, split saves per user, add atomic writes everywhere
- SaveDataSetSaveDataMemory2 treated data_num=0 as "no entries" even when
a valid data pointer was supplied. Astro Bot calls this with data_num=0
meaning "one entry, at data[0]" (an implicit-count convention), so every
real save write from it was silently dropped as a no-op. Default
data_num to 1 when data is non-null, matching that convention.
- SaveDataSetupSaveDataMemory2/Get/SetSaveDataMemory2 rejected any call
that included a non-null icon pointer. Icon persistence isn't
implemented, but titles are allowed to pass one anyway; rejecting the
whole call over it made every title that does so fail outright. Ignore
icon instead of rejecting.
- SaveDataMount3/SaveDataTransferringMount/SaveDataDelete built their
save-data path from title+dir_name only, dropping the user_id PS5
local users are keyed by (unlike the newer SetSaveDataMemory2 layout,
which already nests by user_id). Nest these under user_id too so saves
from different local users can no longer collide. No migration for the
old flat layout - the project is still pre-release/experimental.
- Classic SaveDataMount3-mounted titles (Blasphemous, Dead Cells, ...)
write their save files directly through the guest's normal file I/O
once mounted, bypassing SaveData's own write path entirely, so they
had none of the crash-safety SetSaveDataMemory2 already had. Added the
same protection at the file-system layer: a guest file opened under a
/savedataN mount that recreates the file from scratch (O_CREAT and/or
O_TRUNC) is now written to a .tmp shadow file and atomically replaces
the real one on close, instead of being truncated in place.
- Both write paths now back up the file they're about to replace into a
sibling "backups" folder (same filename, no added suffix) before the
atomic replace, so a manual recovery copy of the previous save always
survives a write - even one that "succeeds" but wasn't wanted.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* Fix save-memory compatibility and preserve existing saves
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
PatchesDialog::Save silently ignored QSaveFile failures, leaving the
status label showing the old "Loaded N cheat(s)" text so the user could
not tell that their selection was not saved. Report the failure in the
status label and the log instead.
uic from Qt 6.12.0 crashes (0xC0000005) on any non-empty <connections>
block in .ui files. Move the cancel_button -> reject connection from
configuration_edit_dialog.ui into configurationEditDialog.cpp, matching
how the other buttons are already connected there.
Also document the glslang (glslangValidator) requirement for Windows
builds; CMake requires it via find_program(... REQUIRED).
* vscode: point IntelliSense at clang instead of MSVC
KYTY_SYSV_ABI expands to __attribute__((sysv_abi)) with no MSVC fallback,
which is correct for a project that rejects cl.exe at configure time. Nothing
told the C/C++ extension that, so it fell back to windows-msvc-x64 and
flagged every guest-facing function in the tree with "__attribute__ is not a
type name".
Set the IntelliSense mode, compiler path and C++ standard to match the
compiler the build actually uses, and export a compile database for it to
read.
* vscode: resolve the clang toolchain through the VS developer environment
The IntelliSense settings named a specific Visual Studio install, so the
compiler and Ninja only resolved on machines with VS 18 Community under
Program Files. Let CMake Tools source the developer environment instead,
which locates the toolchain through vswhere, and drop the explicit
compilerPath so cpptools takes the compiler from compile_commands.json.
Also point C_Cpp.default.compileCommands at the build directory this
workspace actually configures, and drop CMAKE_EXPORT_COMPILE_COMMANDS
because CMake Tools already exports the database by default.
* loader: emulate immediate SSE4a instructions on macOS
* tests: simplify macOS SSE4a test integration
Keep the Darwin target, CTest registration, and aggregate dependency in
one conditional block without a separate target list. Preserve the
standalone target when BUILD_TESTING is disabled.
Lowercase the test filename and diagnostic prefix, and drop unrelated
formatting changes from the instruction emulator.
Credit Brandon Strong's related SSE4a work in
https://github.com/KytyPS5/KytyPS5/pull/467.
Co-authored-by: Brandon Strong <brandostrong@gmail.com>
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
Co-authored-by: Brandon Strong <brandostrong@gmail.com>
* shader: implement S_ASHR_I64 with compute coverage
* tests: use existing compute suite for S_ASHR_I64
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* Fix launcher build on Qt versions below 6.6
QVulkanInstance::NoPortabilityDrivers is only available since Qt 6.6, so
guard its use the same way as the Apple-specific case.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Stop logging every reference-clock copy
CopyData of the clock counter runs on every GPU wait, so the handler wrote
one line per wait: a single SpongeBob boot that never reaches a frame logs
1312 of them, and the count grows with run time. Keep the first 64 lines
for diagnostics; the copy itself is unchanged.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
* Remove reference-clock logging comment
---------
Co-authored-by: YuhiAida <21134775+YuhiAida@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
Normalize Winsock WSAEMSGSIZE on receive to the copied byte count before converting the peer address. Preserve other receive errors and oversized send failures.
shader: support V_FRACT_F16 and V_CMPX_LT_U16
Decode compact, SDWA, and VOP3 forms using the existing arithmetic and comparison paths. Preserve EXEC masking and VCC for the integer compare, and reject unsupported DPP encoding.
Add captured-instruction, operand-selection, floating-point edge, and execution-mask regression coverage.
Co-authored-by: nmzik <Nmzik@mail.ru>
Resolve PS5 5551 textures with alpha in bit 15 and compose host channel order with guest selectors. Share the existing render-target mapping with sampled textures so packed layouts cannot diverge.
Add native GPU readback coverage for 5551 and 4444, arbitrary selectors, constants, and existing 1555 render-target mapping. Verified PPSA24515 through launch.json with Vulkan validation, advancing scenes and over 12,700 guest flips. Independent correctness and cleanliness audits passed.
* launcher: verify with GitHub when website feed differs and compare tag dates
* launcher: restore kyty_emulator dependency for CI build and install
* launcher: simplify GitHub update verification
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
Comparison textures can carry valid HTILE descriptors before any depth-target binding. Keep encoding checks independent of cached image metadata, matching SDK texture classification and shadPS4 depth promotion. Preserve HTILE address and control validation, and replace cache-history test fixtures with captured first-use coverage.
* controller: enable the DualSense HIDAPI driver, light bar and adaptive triggers
Three additions. SDL was opening a DualSense through the generic gamepad path, so SDL_GameControllerHasLED and SDL_GameControllerGetNumTouchpads both reported nothing and no touchpad events ever arrived - enabling the PS5 HIDAPI driver before SDL_InitSubSystem is what makes the light bar and the existing touchpad support reachable. PadSetLightBar validated its arguments and returned OK without touching the pad; it now drives the LED. PadSetTriggerEffect did the same; it now translates the mode and parameters into a DualSense trigger block and sends it.
* controller: fix DualSense effects
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
This reverts commit 6758d67942.
An EXEC pair loses its mask provenance and falls into the raw high-word path. The saved all-ones high word then keeps waterfall branches true and wedges GPU work.
* graphics: do not arm a fast clear from a read-modify-write dispatch
* tests: cover compute metadata clear classification
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* renderer: decode R32_SFLOAT packed colour clears
DecodePackedColorClear handled R32Uint, R32Sint and the 8/10-bit formats but fell through to the default false case for R32Sfloat. metadata_clear_supported then went false, the clear value was discarded, and the target kept its previous contents. Decode the packed word as float bits, matching DecodePackedDepthClear directly below, and reject non-finite patterns rather than materializing them.
* renderer: preserve all R32_SFLOAT clear bits
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
- apply c8f34f9's ResourceTracking portability fix
- complete V_CMP_LT_U64 and S_FLBIT_I32_B32 support
- add the sceAmprCommandBufferWaitOnAddress NID from f625a47
- add decoder and compute tests for both shader operations
Co-authored-by: Mac <me@itsmemac.com>
* cli: add --gpu <index> to select a Vulkan physical device
Select the Vulkan physical device by enumeration index, defaulting to
automatic selection (-1). Invalid or out-of-range indices fall back to
automatic selection with a clear message, and the selected device is
logged.
Co-Authored-By: Claude <noreply@anthropic.com>
* cli: reject invalid --gpu index, translate comments, apply clang-format
An invalid --gpu index now prints an error and exits with code 1 (usage),
matching --user-id / --present-mode / --console-language. The automatic
fallback is kept only for out-of-range indices detected at runtime.
Comments translated to English and the touched files brought in line
with clang-format (pre-commit v22.1.3).
Co-Authored-By: Claude <noreply@anthropic.com>
* cli: simplify Vulkan device selection
---------
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
* graphics: quiet two misleading render diagnostics
The three PM4 register-offset normalization messages logged once per occurrence with no
limit. They are printed in the colour reserved for problems, and normalizing the offset is
routine, so a short session buries the log in them: PAC-MAN WORLD 2 Re-PAC (PPSA18189)
emitted 24969 of them in about a minute, a fifth of a megabyte and 1.9% of every line
written. Rate-limit all three the way the surrounding diagnostics in these files already
are, which keeps the first occurrences without drowning what follows.
The EQAA check warned whenever a control register was set, including on single-sample
targets where those controls do nothing. That title never multisamples -- all 128 of its
render targets are samples=1 -- yet it still produced the warning, pointing at a cause that
could not apply. Return early when multisampling is off, matching how the neighbouring
sample-location check already reads msaa_num_samples.
Log output only; no rendering behaviour changes.
* graphics: silence normalized offset diagnostics
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* macOS: auto-discover bundled MoltenVK on startup
On macOS the Vulkan loader comes from MoltenVK. SDL's default
probe only searches system paths, so a bundled libMoltenVK.dylib
next to the executable is invisible under the hardened runtime.
Probe it explicitly via SDL_GetBasePath() before falling back to
SDL's default search. Respect SDL_VULKAN_LIBRARY overrides and
log which loader is used. Update README to reflect that no
environment variable is required for bundled builds.
Co-Authored-By: internal-model
* window: fix Vulkan portability enumeration on macOS
MoltenVK requires VK_KHR_portability_enumeration and
VK_INSTANCE_CREATE_ENUMERATE_PORTABILITY_BIT_KHR to surface
portability devices via enumeratePhysicalDevices. Without the flag
the adapter is hidden on some driver versions, causing
"Could not find suitable device" failures.
Add the instance extension and flag when available, and add
diagnostic logs for all MoltenVK fallbacks (depthClipEnable,
colorWriteEnable, depthBounds, fragmentShaderBarycentric)
so black-screen diagnostics are actionable.
* macOS: add KytyPS5.app bundle packaging
Add a proper macOS application bundle when the launcher is built.
launcher becomes KytyPS5.app (CFBundleIdentifier org.kytyps5.kytyps5)
with KytyPS5 as the executable, and kyty_emulator is installed as a
helper inside Contents/MacOS so QCoreApplication::applicationDirPath()
finds it. MoltenVK is auto-discovered via SDL_GetBasePath() in that
directory, so no SDL_VULKAN_LIBRARY is required.
- src/launcher/Info.plist.in: minimal bundle plist (version from
PROJECT_VERSION, LSMinimumSystemVersion 13.0).
- src/launcher/CMakeLists.txt: set MACOSX_BUNDLE, BUNDLE DESTINATION,
install helper, re-sign helper with JIT entitlements, and invoke
macdeployqt when available. Keep flat kyty_emulator for CLI.
- .github/workflows/build.yml: bundle MoltenVK into
KytyPS5.app/Contents/MacOS when present, verify bundle artifacts
(file/lipo/codesign), and handle bundle paths in the release zip.
- README.md: document bundle double-click and MoltenVK locations.
* launcher: fix bundle rpath for Qt frameworks
The bundle executable had @rpath references to Qt frameworks but no
LC_RPATH, causing dyld to abort at launch:
Library not loaded: @rpath/QtConcurrent.framework
Reason: no LC_RPATH's found
Set INSTALL_RPATH to @executable_path/../Frameworks and
BUILD_WITH_INSTALL_RPATH so the built and installed bundles
resolve Qt from Contents/Frameworks. Verified via otool -l
(LC_RPATH now present) and macdeployqt now succeeds without
"Cannot resolve rpath" errors.
* launcher: add Import game button
Add a prominent Import button to ConfigurationListWidget that lets
users select a game folder (containing eboot.bin) or a folder
containing multiple games via QFileDialog. Handles:
- Single game folder selected: adds its parent to game_dirs (Scan
searches subdirectories, so parent is needed)
- Folder with multiple games: adds the selected folder
- Deduplication via PathKey/NormalizeGameDirectories, writes
settings, rescans, and selects the imported game if found
- Shows info/warning dialogs when no games are found
UI: QToolButton with SP_DirOpenIcon and text Import beside the
existing toolbar, styled with a blue background for visibility.
Logic in ConfigurationListWidget::import_game().
* macOS: fix bundled loader and artifact packaging
* Revert "launcher: add Import game button"
This reverts commit 2f670ef95c.
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
When CreateFileMapping cannot reserve the guest backing store the log only
carries the raw Win32 error, which gives no clue that the paging file is the
thing to change. Print the size that was asked for and say so directly when the
failure is ERROR_NO_SYSTEM_RESOURCES or ERROR_COMMITMENT_LIMIT.
Decode the high-half D16 LDS read and route it through the existing unsigned-short load lowering. Represent the destination as a high-half partial write so the low 16 bits of the VGPR remain intact. Add captured-instruction decode coverage and a Vulkan execution regression that verifies the preserved half.
Co-authored-by: SueryRD <the.ai.tester2@gmail.com>
Allow byte, word, and full-dword SDWA source selection for V_NOT_B32 and V_FFBL_B32. Add captured instruction tests that verify decoding, IR extraction, SPIR-V generation, and runtime results.
Co-authored-by: SueryRD <the.ai.tester2@gmail.com>
Scale the SDK 8-bit motor levels to SDL’s 16-bit range and use SDL’s maximum effect duration. Keep lookup and rumble under the controller lock so hot unplug cannot invalidate the SDL pointer.
Run() registers KytyClose with atexit(), but Execute() ends the process with
std::quick_exit(), which invokes at_quick_exit handlers only. Nothing registered
one, so the cleanup never ran and neither did the emergency shutdown handlers
that close open guest files and flush the log.
Register a quick-exit handler that runs the emergency shutdown. It deliberately
skips RuntimeLinker::Clear(): guest threads are still live at that point and
tearing the linker down there deadlocks against a thread parked in a condition
wait.
graphics: stop rejecting valid GE_CNTL group sizes
GE_CNTL group sizes control guest geometry-engine scheduling rather than host wave width. Do not reject the supported vertex path for compiler-selected values above 64, and remove the corresponding stale debug warnings.
* graphics: report the VkResult when a submit or fence wait fails
* graphics: add a --gpu-assisted-validation flag
* graphics: imply Vulkan validation for GPU-assisted validation
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* fixed linux build
- added correct uint long translation
- replaced throw with non-constexpr function so that it fails in case of malformed table
* shader: remove redundant opcode table comment
* shader: use EXIT for invalid opcode tables
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
Calling QApplication::quit() after the user closes the first-run game
folder dialog can segfault during nested modal shutdown. Leave the
launcher open so settings can still be configured. Fixes#121.
No idea why these were reimplemented in the first place at all. The loaded guest libc already provides them; Kyty only needs the lower libkernel pthread imports.
* launcher: quote emulator paths that contain spaces
Run the Linux helper script as a bash file argument instead of bash -c,
and build a properly quoted cmd.exe /K command on Windows so install
folders like "Sony Playstation 5" still launch. Fixes#179.
* launcher: fix cmd command quoting
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
* window: toggle borderless fullscreen with F11 and Alt+Enter
Match the existing SDL_WINDOW_FULLSCREEN_DESKTOP startup path so testers
can switch modes at runtime. Reserve F11 in host input mappings. Fixes#217.
* window: consume fullscreen hotkey input
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
fix(gpu): drop colorWriteEnable dynamic state on depth-only pipelines
Pipelines were created with VK_DYNAMIC_STATE_COLOR_WRITE_ENABLE_EXT
unconditionally on non-Apple platforms, but the draw path only calls
vkCmdSetColorWriteEnableEXT when there is at least one color attachment.
A depth-only draw therefore reached vkCmdDrawIndexed with the dynamic
state declared and never set, which violates
VUID-vkCmdDrawIndexed-None-07749.
With validation enabled the error is fatal, so any title that issues a
depth-only pass aborts. The launcher enables validation by default
(vulkan_validation_enabled = true), so this is the default configuration:
PAC-MAN WORLD Re-PAC crashes from the launcher while the same build runs
normally with --vulkan-validation false.
Gate the declaration on the same color_count the draw path uses. Both
sides read state.color_count from the same draw, so the two conditions
are now identical by construction.
* kernel: do not release a waiter on a zero-count syncOnAddress wake
sceKernelSyncOnAddressWake with a count of 0 must release no waiters. On
Linux it released exactly one.
WakeLinux passes the count straight to FUTEX_WAKE, which does not treat 0
as "wake nobody": the kernel's wake loop tests `if (++ret >= nr_wake) break`
after waking a waiter, so any waiter present is released before the loop can
break. A standalone futex program confirms it -- FUTEX_WAKE with val = 0
returns 1, and the waiter observes its wait return 0 with the word unchanged.
WakePortable already short-circuits count == 0, so the two backends of the
same ABI disagreed, with only the Linux one wrong.
TestWakeZeroIsNoOp in SyncOnAddressTests covers exactly this and fails on
Linux without the fix (15 consecutive failures on the assertion "zero-count
wake releases no waiters"); with the fix the suite passes 15 consecutive runs.
* kernel: shorten zero-count futex comment
---------
Co-authored-by: nmzik <Nmzik@mail.ru>
Testing games and submitting detailed bug reports are useful ways to contribute. Search existing
issues first, then use the **Game Emulation Bug Report** template and attach the complete log file.
issues first, then use the **Game Emulation Status Report** template and attach the complete log file.
Code contributions should be focused, build successfully on the platforms they touch, and include
relevant tests where practical. Windows is the primary target, so a change that alters shared code
should not regress it; changes confined to a platform's own code paths only need to build there. Because KytyPS5 is still evolving quickly, consider opening an issue before
starting a large change.
**Pull requests for general fixes, architectural changes, or refactoring that do not address a concrete bug or launch problem in a specific game will be automatically closed.**
**Large, unfocused commits or pull requests that bundle unrelated changes without a clear purpose or game-related benefit will also be automatically closed.**
Keep each PR focused. Name the affected game, describe the exact problem, and provide evidence showing what the change fixes.
Windows and Linux are the primary target platforms, so changes to shared code should not
introduce regressions on either platform. Changes confined to a platform's own code paths
only need to build on that platform. Because KytyPS5 is still evolving quickly, consider
opening an issue before starting a large change.
### Formatting
Set up the clang-format hook after cloning:
Set up the clang-format hook after cloning.
```powershell
python-mpipinstallpre-commit
Install `pre-commit` using the method appropriate for your platform:
- **Arch Linux / CachyOS:** `sudo pacman -S pre-commit`
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
nmzik
