0888 KRGB and 8888 ARGB, still went through the scalar per-pixel writer.
One packed pixel is exactly one VRAM word, so they fit the same 8-word
interleaved store as the 16-bit path, 8 pixels per iteration. The store
is factored out of storeFramebuffer16Pixels() so both use it.
888 (24-bit packed) stays scalar: its pixels straddle VRAM words, so the
store pattern only repeats every 4 pixels / 3 words.
The address walk is the same as writeFramebufferLW(), so the output is
identical to the scalar path, including when the padding isn't a
multiple of 4. A harness writing 7,056 frames through both paths (both
channel orders, modes 4 to 6, 6 frame sizes, 7 destination addresses
including bank and VRAM edges, 7 line strides, 4 clip windows) gives
byte-identical VRAM. The same algorithm measured about 2x faster on a
640x480 8888 frame (0.213 ms to 0.105 ms) on Apple Silicon.
tests/src/rend/FramebufferWriteTest.cpp checks both 32-bit modes against
an independent reference (one pvr_write32p per pixel, the scalar walk)
over 3,600 cases: widths around the 8-pixel vector width, padding
including values that aren't a multiple of 4, destinations at the 32-bit
bank boundary and the end of VRAM, and clip windows, for both channel
orders. All of VRAM is compared, so a store past the frame fails too. On
arm64 it exercises the NEON path; elsewhere the scalar one. It fails when
the padding computation is deliberately broken.
Other architectures are unchanged.
Co-authored-by: Claude Opus 5.5 <[email protected]>
[ci skip]
ENABLE_CTEST=ON never produced a test runner on macOS. The app's main()
(shell/apple/emulator-osx/SDLApplicationDelegate.mm) was still compiled
in, so the binary launched the emulator UI instead of gtest_main, and the
link needed libcurl, which HttpTest uses directly while the core only
links curl where it is the HTTP client (not on Apple).
- Leave SDLApplicationDelegate.mm, osx-main.mm and the Syphon video
routing out of the target when BUILD_TESTING is set, like linux-dist
main.cpp and winmain.cpp already are.
- Stub the os_* / darw_printf / hostfs / i18n glue those files provided,
in tests/src/test_stubs.cpp, for macOS.
- Link CURL::libcurl into the test target on Apple.
With this the suite builds with both the Xcode and the Ninja generators
and runs on Apple Silicon: 144 tests, 141 pass. The three that fail there
(AicaArmTest.LogicOpsTest, SDLControllerMappingTest.split_axis and
.reverse_axis) and an abort in ~maple_sega_vmu at process exit are
pre-existing and are not changed here. The CI workflow still runs the
tests only on Linux.
Co-authored-by: Claude Fable 5.1 <[email protected]>
[ci skip]
Vulkan without VK_EXT_provoking_vertex, DX9 and DX11 take flat-shaded
colors from the first vertex, while on Dreamcast it's the last. For the
depth-sorted triangle list, setFirstProvokingVertex() copied each
triangle's last vertex colors onto its first vertex. But sortTriangles()
builds these triangles from strips, so neighbors share vertices:
tri0=(P0,P1,P2), tri1=(P2,P1,P3). P2 is tri0's source and tri1's
destination, and the list is processed in depth order, so one
triangle's flat color can overwrite another's before it's read. In
strips of three or more triangles, consecutive triangles even share
their first vertex. Flat-shaded translucent polygons then pick up a
neighbor's color, e.g. the player shadow decals in Tony Hawk's Pro
Skater 2 and Mat Hoffman's Pro BMX turn uniformly dark.
Each sorted triangle has its own three slots in idx, so rotate them
(a,b,c) -> (c,a,b) instead. The winding is unchanged and the provoking
vertex comes first, without touching shared vertices. Strips keep the
color copy, which is safe there.
Adds SorterTest, which fails on the old code and passes with this change.
Co-authored-by: Claude Opus 5.5 <[email protected]>
[ci skip]
Since the switch to hostfs::File (#2299), MD5Sum::add(hostfs::File *)
resolved to the generic add(const T&) template, which hashes the pointer
value. The resulting digests depend on the heap address, so they differ
between peers and between runs. This breaks GGPO peer verification when
a .state.net savestate is used (stateMD5), and with CDI, GDI and CUE
images (gameMD5).
Add an MD5Sum::add(hostfs::File *) overload that hashes the whole file
content, like the std::FILE overload, and reject pointer types in the
template with a static_assert so this cannot happen again silently.
Co-authored-by: unjordi <[email protected]>
[ci skip]
Game ids can have spaces, like "INITIAL D Ver.2". The parser removed all
spaces in the section name so it became "INITIALDVer.2". Now only the
spaces before and after the name are removed.
Co-authored-by: nixxou <[email protected]>
[ci skip]
CHDs converted from Redump MIL-CD dumps were rejected with "Unsupported
subtype or pre/postgap": the audio tracks of the first session carry their
pregap inside their own .bin file, which chdman records as PREGAP:n
PGTYPE:V<type> in the CDROM_TRACK_METADATA2 entry.
Handle these tracks the way MAME does:
- when PGTYPE is prefixed with 'V', the pregap sectors are part of the track
data and counted in FRAMES, so INDEX 01 sits `pregap` frames after the
previous track and the track data starts `pregap` frames into the CHD
- otherwise the pregap only takes up space on the disc
- subcode data doesn't change the sector layout, and postgap sectors are
never stored: warn about both instead of bailing out
Also:
- audio tracks of CD-ROM CHDs are stored big-endian, just like the ones of
GD-ROM CHDs using the CHGD metadata tag. Not byte-swapping them was the
cause of the audio distortion reported with CHD MIL-CDs
- the 11400 frame session gap includes the pregap of the second session's
first track. Don't count those sectors twice when they're in the image
- only treat a multi-track disc as a MIL-CD when its last track is a data
track. A mixed mode CD is a single session disc and its data track is the
first one
- report the lead-out as the first FAD after the last sector, like the CUE
and GDI loaders do, instead of one frame earlier
Tested on a dozen homebrew MIL-CDs, from 2 to 17 tracks: the CUE and CHD
loaders now return the same tracks, sessions and sector data for each of
them, and they all boot.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01NfFd8rJc9aTQB7au8yLrqG
* cue: don't count the session pregap twice in MIL-CDs
The 11400 frame gap added at REM SESSION 02 is the lead-out of the first
session (01:30:00), the lead-in of the second one (01:00:00) and the pregap
of its first track (00:02:00). When that track has an INDEX 00 of its own,
its pregap sectors are stored in the image and were counted a second time,
placing the data track 150 frames after the session start reported in the
TOC, i.e. after the FAD the boot file is looked up at.
Redump lays MIL-CDs out with no INDEX 00 on the second session's track, so
none of the images I have hits this. It only shows up with images written by
other tools, and is included here because the CHD loader needs the same
adjustment to agree with this one.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Claude-Session: https://claude.ai/code/session_01NfFd8rJc9aTQB7au8yLrqG
Add http::get function with request and response header params.
Set a 30 s connection timeout for all http requests (except on macOS).
Linux tests with postb.in
Use cycle_counter to increase the precision of timers.
Use 64-bit where needed to handle underflows. Avoid double operations on
32-bit and 64-bit data.
Add timer test.
Fixes lack of randmness in Soul Calibur. Issue #2093
IniFile: use locale-independant float conversions functions
Use std::ostringstream and std::istringstream as appropriate.
Use C locale for GDI and CUE parsing.
Support for localized dates, collation and messages.
Rewrite ini file reader/writer and command line parser.
Standardize all config stuff in the config namespace.
Range for all axes, including triggers, is now [-32768,32767]
Detect trigger direction when mapping and store in mapping file.
sdl: parse controller bindings string to correctly map axes.
Issue #2014
Add sine effect to haptic api and stop using maple rumble for naomi FFB.
Emulate spring in midi FFB. Thanks to njz3 for his help on decoding the
protocol.
Don't use exponential scale for rumble when using haptic sine effect.
serialize Sh4OCache::writeBack/ThroughBufferCycles
Add button combination mappping support
- Implement tracking of pressed buttons in GamepadDevice
- Add ability to detect button combinations- Fix mapping system to preserve individual button mappings
- Ensure buttons can work both individually and as part of combinations
- Ensured trigger buttons can be the first button in a combination
- Maintained detection state when binding multiple buttons
- Added special handling to recognize both original and normalized trigger codes
---------
Co-authored-by: James Smith <[email protected]>
Reduce gdrom dma buffer size to 16 sectors from 32.
Limit serialization of dma and pio buffers to actual content.
reios: schedule transfer of dma data in chunks.
Fixes Soul Calibuf audio drop outs at boot with HLE BIOS (RA).
Issue #1755
SDL force feedback for racing wheels.
Support for jvs ffb (f355, 18wheeler), atomiswave (maxspeed, ftspeed)
and midi ffb (initd, clubk, kingrt66, sgdrvsim).
Simple haptic rumble for f355 and 18wheeler.
Joe Mattiello