WinUAE 2.6.0 Betas

Bigger update for beta 1 this time.

Quote:

---

Big MMU emulation update:

- Previous (Next emulator) 68030 MMU added (+missing bus error exception stuff implemented by me), some 68040 MMU updates merged. A3000 1.4 boot ROM, 1.3 SuperKickstart (which uses special MMU mapping for "bonus" code), Enforcer and Amix confirmed working (Amix can´t be installed because tape drive is not emulated but pre-installed HD image work)
- 68060 MMU emulation added, mostly same as 68040, different stack frames and 68060 restarts all bus faults, 68040 restarts only read faults (excluding MOVEM).
- Executable is larger, MMU compatible 68030 and 68060 CPU emulation cores added.
- MMU double fault stops the CPU (just like real CPU does), no more instant resets, makes debugging easier. Windowed mode status bar shows "HALTx" (Hx in OSD) in this situation (x = reason code). Only reset gets out of halted state. Also non-MMU mode fake double bus faults or other impossible states (for example odd supervisor stack) now enters halted state.

Note that 68030 MMU is very complex compared to 68040+ = slower than 68040 or 68060 MMU emulation.

- Added second A3000 Quickstart compatibility level, 68030 MMU emulation or fake MMU + JIT.
- Always refresh windowed status bar floppy leds when emulation starts or statefile is loaded.
- Executable as a disk image mounting does not generate identical disk creation dates anymore, wrong date calculation fixed.
- Use name of original file when mounting executable as a disk image, will be truncated if too long.
- Disable floppy sounds when exe mounted as a disk is inserted.
- Added wheel mouse Game Ports device type. Old "Mouse" is renamed "Wheel Mouse" and new "Mouse" is mouse without wheel mapped. Added because some older programs can get confused with mouse wheel keyboard events.
- Win+D (show/hide desktop) has now same behavior as minimize/unminimize.
- Named pipe accepts also UTF16LE (Windows native) and UTF8 strings if BOM is included. No BOM = assumes current "ANSI" character set. Old ipctester.exe works finally again.
- Named pipe also supports debugger commands, either using dbg_ or first setting it to debugger mode using "ipc_debug" command. ("ipc_config" = config mode, "ipc_event" = event mode, "ipc_restore" = back to normal mode). Other direct commands are cfg_. (Works exactly same as uae-configuration commands)
- Call DwmEnableMMCSS(TRUE) if running under Vista or later.
- VSync/HSync counting CIA TOD counters run at half rate if genlock mode is active.
- Added hack that fixes unreliable KS ROM genlock detection in fastest possible CPU modes.
- Enabling or disabling windowed mode status bar does not reopen the window anymore.
- Horizontal programmed screen mode registers´ upper byte must be masked. Fixes Debian Sarge blank screen.
- All debugger commands support MMU logical to physical address translations, enable with "mmu ", 0 = disabled.
- Added support for A2000 old style RTC address range (Adv. chipset panel), this also maps new style clock address range (0x00DC0000) as custom chipset.
- FPU NULL state implemented. Fixes Amix FPU detection, it tests FPU by restoring NULL frame then immediately saving FPU frame and only accepting the FPU if stored FPU frame is NULL frame.
- Filesystem added using FileSys hardfile option will have at least 4000 byte stack, filesystem will be only added to FileSystem.resource if it does not already include filesystem with same dostype. (Previously it was always added, replacing previous one)
- addmemx config option(s) caused crash if emulation was restarted.
- Filter position and zoom values uses now floats, weird integers are gone. Old settings are ignored, config entries renamed!
- Custom input events can send commands to debugger, "dbg ", output goes to console window if open.
- Changing filter position/zoom values using custom events or uae-configuration incorrectly forced display subsystem reset.
- WD33C93/DMAC SCSI emulation improved. (Amix, NetBSD and so on..)
- Added A590/A2091/A3000 SCSI DMAC state file support. Only basic state, any SCSI operation must not be active while saving.
- Windowed mode title bar doubleclick now maximizes normally, shift+doubleclick = fullscreen and ctrl+doubleclick = fullwindow.
- Immediate blitter didn´t check if blitter DMA was enabled.
- Added CHD (Mame) CD image support. Full data, CD audio and subchannel support.
- CHD image as a hardfile support added. Read-only.
- LZMA (7zip) support code updated to latest beta (required by CHD LZMA code)
- Host FPU rounding/accuracy mode was not reset back to correct original values when using filters.
- Pause mode updates, configuration changes while in pause mode work better but currently display becomes blank if display mode changes while in pause mode (Old screen was lost and new data can´t be generated in pause mode)
- Filesystem config (cylinders, sectors etc..) parameter passing rewritten, now uses same structure everywhere instead of passing huge and random amount of ugly variables around. Big change, can break things.
- Partition hardfile parameter/geometry support update, if file named .geo is found in same directory as hardfile, it is parsed for mountlist-like parameters (key=value). It also supports parameters that are not available in GUI or in config file (for example lowcyl, buffers, maxtransfer etc..).
- Log partition hardfile full geometry data when mounting.
- Added native resolution selection to Display panel fullscreen resolution options. Uses always monitor´s native mode. (Actually it is monitor´s desktop resolution because biggest listed resolution may be unsupported)
- Replacing or removing HDF files using the GUI now changes the drive on Amiga side too. Inserting new HDFs won´t work.
- Fixed IDE emulation bug introduced in 2.4b17, IDE status register returned bogus data after execute diagnostics command. (KS driver didn´t care, other drivers probably did)
- 1.5M chip RAM size didn´t work (Note that this can´t work if JIT is enabled)
- Very very long (hours..) startup delay if emulation was started in fastest possible + CPU throttle enabled.
- Windowed RTG to native switch (or vice versa) while window had no focus set internal focus state causing keyboard input to appear on Amiga side.

---

Beta 4:

Quote:

---

- Fixed 68030 MMU MOVEM to memory wrong bus error data output buffer value.
- 68030 MMU FMOVEM didn´t work if it caused bus fault.
- 68060 MMU non-locked Read-Modify-Write bus error bits emulated.
- SCSI emulation: reject >0 LUNs also when LUN is included with SCSI Identify Message.
- IDE single drive setup should return zero when accessing non-existing slave drive´s status register.
- A2065 emulation initialization tweak, fixes NetBSD 1.0 hang.
- CDFS didn´t use CD led (FS)
- Blitter logging/debugging command line option added (was compile time previously) -blitterdebug x (1 = logging, 2 = writes to blitter register while already active logging, 4 = disable D channel, 8 = immediate blitter even in cycle-exact mode)
- Support display setups that have two (or more) horizontal DDFSTRT/DDFSTOP regions. Fixes Subtle Shades / Nuance "face" part. (There is still some small glitches on left side of text part, later..)
- Some remaining instant reset conditions replaced with halted state.

---

Beta 5:

Quote:

---

- RDB hardfile dragndrop to hardfiles panel added empty hardfile.
- Windowed mode bottom border "led" state didn´t always match OSD led state.
- SCSI emulation improved, select and transfer starting from 0x45 can be used to receive data.
- SCSI emulation WD33C93 is now correctly detected as revision A, not B. (only B has Queue Tag register)
- 68060 MMU: BFCHG, BFCLR, BFSET and BFINS are read-modify-write instructions.
- GUI IDE/SCSI CD path was not updated in harddrive list when CD was changed using "Select image file" button.
- 68040/060 MMU emulation fix, PFLUSH/PFLUSHN only flushed supervisor space ATC. Hopefully this was the last major 040/060 MMU bug. Weird NetBSD and Linux crashes are gone.
- Subtle Shades / Nuance remaining graphics glitches fixed. (NOTE: ECS only, OCS has corrupted graphics and it is correct behavior)
- GUI "Swap Ports" cleared custom mappings.
- 68040/060 no FPU/disabled FPU exceptions now include correctly calculated EA field. This should allow "compatible" 040/060 FPU emulation with correctly handled missing instructions in the future (currently 040/060 FPU always supports all 68881/2 instructions)
- Added new keyboard custom event type, "kbr ". For example ´kbr "abc " SHIFT_LEFT "abc" shift_left RETURN´ generates "abc ABC". Input name in upper case = press, lower case = release. Quoted string automatically sends both press and release key codes. Any unreleased keys are automatically released at the end. Non-quoted names are same as input event config names without KEY-prefix. (inputevents.def in source archive)
- Adjusted drive step limit value slightly, Extemporized / Digital works again.
- Fixed virtual mouse driver mouse movement problem (b1)

---

Beta 6:

Quote:

---

MMU emulation is now considered mostly done. Some minor bugs probably still exist but Amix, NetBSD, Debian Linux, Enforcer and WHDLoad MMU option seem to work fine.

- SCSI and IDE emulation HD and CD read and write commands are now executed in background thread, slow devices (for example real optical drive) won´t temporarily pause the emulation anymore.
- ATAPI CD emulation didn´t support split data transfers. Fixes NetBSD installer unpack error.
- Show also error code converted to string if HD to HDF image creation fails.
- Added mostly useless CD list to CD drive GUI. (Someday there will be more than 1 option)
- Do not allow multiple identical HD/CD controllers (if not UAE).
- Increased size of ATAPI data buffer and added size checks. 65536 was too low. (Debian installer)
- SCSI emulation didn´t handle WD33C93 Select-and-Transfer command scatter/gather DMA. (Debian installer)
- CD led didn´t flicker during SCSI and ATAPI CD access.
- Fixed D3D shader error/warning string buffer overflow crash. Some shaders can output really long warning messages.
- D3D internal texture size was incorrect in some situations, causing some shaders to have incorrect geometry.
- Added support for "post" shader filters. Post = apply shader to final output (after scaling, D3D scanlines and RGB masks), Pre = apply shader to internal Amiga display (same as in previous versions).
- D3D shaders can be "stacked", max 4 pre and post shaders supported. (-4 to -1 = pre, 1 to 4 = post), Output of previous shader is input to next shader. Can be useful if you want to have "normal" filter and CRT geometry filter active at the same time. Also stacking same filter multiple times makes interesting results D3D only, software filters not supported.

---

Beta 7:

Quote:

---

- In JIT mode format 2 stack frames had wrong address field contents, this also broke 68060 unimplemented integer instruction exceptions (64-bit variants of DIVL and MULL).
- GUI boot priority/autoboot/automount handling fixed and simplified.
- 68030 "cycle-exact" mode used 020 instruction cache emulation code. No guarantees that this fixes all 68030 "CE" problems.
- D3D texture memory leak fixed (b6)
- A2024/Graffiti extra display buffer memory leak fixed.
- 1/50s file modification date also supported on FAT volumes. Returns FAT file creation date (which has 10ms resolution, file modification date only has 2s resolution) if it is inside file modifcation date´s 2s resolution. (Currently Vista or newer only)
- 68040+ "no FPU" exception stack frame still had incorrect data.
- Disable 68040/060 CPU "More compatible" option if JIT is enabled. JIT active and not emulating unimplemented instructions isn´t safe because JIT and normal emulation does not always agree which instructions are supported.
- A2065 emulation tweak, Ethernet chip documentation STOP/STRT/INIT bit documentation is not very clear.

---

Beta 8:

Quote:

---

- CD SCSI emulation: implemented mode sense CD/DVD page code 0x2A. Used to detect drive capabilities, for example if it can read or write CD-R, DVD-ROM or DVD-R and so on.
- SCSI emulation MODE SENSE didn´t limit returned amount of bytes to allocation length field.
- BPLCON0 SuperHires bit disables normal DDFSTRT/DDFSTOP limits. (Why?)
- Resolve shortcut path if mounting shortcut (.lnk) as a harddrive.
- If inserting RDB hardfile using GUI: get device´s blocksize from RDB block size field.
- Attempted write to SCSI or IDE CD (using normal SCSI Write command) crashed.
- Return WP bit set in SCSI Mode Sense result if SCSI HD is write protected.
- Add harddrive option set wrong parameters, drive was added as non-RDB partition hardfile (some earlier beta).
- Hardfiles and harddrives: attempt to mount as read-only before failing if read-write access is not allowed.
- Hide bootpri/bootable/donotmount boxes in GUI when selected RDB mode/controller combination does not support them.
- Added support for write protected disk images even if physical file is write-enabled. Config file only, floppyXwp=true
- 68020/030 no-FPU F-line exceptions didn´t work correctly, broke in previous 68040+ FPU disabled exception updates.

---

- - - Updated - - -

Beta 9:

Quote:

---

- 68000 NBCD didn´t emulate undocumented V-flag behavior. (Probably forgotten long time ago because SBCD had correct implementation)
- It was impossible to get out of halted CPU state in some situations.
- FDBcc and FScc 6888x FPU instructions without installed FPU returned wrong f-line exception PC field.
- Directory filesystem: emulate Windows shortcut (.lnk) files and symbolic links as Amiga side soft links. Implements ACTION_READ_LINK. Works also across drives if shortcut/link target exists in any mounted directory harddrive. Not guaranteed to be 100% correct implementation, Amiga softlinks are a bit strange.

---

Beta 10:

Quote:

---

SCSI CD MODE SENSE emulation fixes.
SCSI CD Write commands should return unknown command error, it is not a CD writer.
SCSI CD and HD: return correct sense code if Read or Write command tries to access out of bounds LBAs.
SCSI/ATAPI CD: return medium changed unit attention sense data after media change.
OCS and ECS Denise/AGA have slightly different behavior when program writes to BPL1DAT to enable sprites before DDFSTRT (sort of AGA bordersprite option), OCS: only works if hpos is large enough (~40+), ECS: smaller limit (or no limit, didn´t test).
Soft link emulation improved.
Directory filesystem ACTION_MAKE_LINK implemented, supports soft links only, creates shortcut (*.lnk) files.
Entering GUI Harddrives panel removed image from internel CD image mounter if mounted as ATAPI or SCSI. (introduced few betas ago)
Restructured blkdev.cpp unit handling.
Sound autoswitching is disabled when AVI recording is active.
Low latency vsync no buffer mode and doubled frames (100Hz+) was unstable.
POTGO right button bit was temporarily and incorrectly active in some situations when loading state.
D3D Low latency vsync in double and triple buffer double rate modes (100Hz+) now blank every other field to minimize "ghosting" effect. Will be optional later. (http://www.blurbusters.com/ for more information)
Added hack that prevents incorrectly triggering CIA alarm interrupts in non-cycle exact modes due to not that good KS timer.device code. (It did not cause any problems, just wasted some emulated CPU time)
Fixed memory corruption in exe to adf conversion code. (b1)

GamePorts panel test/remap option rewritten.
- Test and Remap buttons combined.
- One or more events can be added.
- Events can be removed one by one.
- Non-key/joystick/mouse events (insert floppy, reset emulator etc..) can be added and removed. Not yet implemented.
- Doubleclick on input list will record single input event.
NOTE: Built-in (non-custom mode) events can not be edited, you need to start from scratch. (Click "Delete all")

---

Beta 11 & 12:

Quote:

---

Beta 12:

AGA border sprite feature emulation is now also cycle-exact. Similar update to borderblank was done in 2.5.0. For example it was possible that sprites were hidden when border sprite bit was set and zero bit planes enabled. (Noticed when doing some unrelated real hardware limit tests)
Black frame insertion was incorrectly used (if enabled) in <100Hz fastest possible CPU modes.
Horizontal blanking rewritten, should be more accurate now, programmed display modes only.
Display emulation tweak, Super72 sprite display corruption is gone.
Fixed bug introduced in b5, blitter emulation ignored DMA off state in cycle exact mode.
Emulate DIV divide by zero undocumented flags when CPU >=68020. So far 68020 and 68060 tested and confirmed. Odyssey / Alcatraz and 68020+ CPU does not anymore get stuck in part where spaceship flies above ground.
CPU halted state didn´t disable CPU interrupts.

Beta 11:

Blank left and right border if programmed mode and part of original borders would be outside of programmed hsync start and end period.
Address error exception always started debugger, forgotten debug option disabled.
Added Ts debugger command, list all exec residents.
Restoring state file with harddrives crashed in some situations. (broke with previous filesystem updates)
CD32 TOC entries were transmitted too quickly due to missing check, correct speed would have been one entry per CD frame (75/sec), not one entry per scanline. CD command buffer overflowed and driver detected CD as broken if CD had enough tracks. Old bug, was mostly invisible until 2.5.0 Universe TOC fix. (Super Street Fighter II Turbo and others CDs with lots of tracks)
IDE emulation ignored optional forced physical cylinder number setting in config file.
Both hires and superhires bits in BPLCON0 set: superhires mode active.
100Hz+ black frame insertion is now optional, setting in Misc panel. (Still D3D/low latency vsync only)
GamePorts panel test/remap improved
- Implemented custom event add option. Select event from select box and click "Add", then press any button/key/etc to map it.
- Added event will be automatically removed if it does not have any mapped events (Works differently than normal events)
- Loading config file that has GamePorts panel custom added events using older version is not supported.

---

Beta 13 & 14:

Quote:

---

Beta 14

A3000 SCSI emulation crashed if no SCSI drives configured.
Added previously config-file only master floppy write protection option to GUI misc panel.
If Windows softlink or .lnk can´t be opened (access denied) or has system attribute flag, handle it as a normal file/directory, just like in older versions.
DIVS/DIVL integer overflow crash fix (INT_MIN/-1 = an integer that does not exist, causes cpu overflow exception)
GUI Add Hardfile added IDE or SCSI was mounted as a CDROM. (Worked if config was saved and then loaded again)
Adjusted read-write-only-custom-register-side-effect in AGA mode. (Artificial Paradise / NGC missing loading screen, does stupid bset #9,$dff104 which, depending on previous accesses, either writes $0200 or "random bits" OR $0200)
Manually added custom addmem2 memory region config entry used addmem1´s memory region..
"Load From"/"Save From" buttons didn´t change config name in window title.


Beta 13

Ignore vendor specific USB usage pages also in received USB HID packets, fixes for example Logitech Precision gamepad ghost button events.<
Fixed border sprite bug introduced in b12, border sprites should stay hidden if border blanking is also enabled.
Black frame insertion is now supported in legacy vsync D3D 100Hz+ modes. DirectDraw modes may also work.
68020+ cycle-exact CPU speed changed after entering and exiting GUI in some situations.
Partition hardfile dragndrop didn´t work.
Automatically loaded ROM path FastFileSystem file always replaced FileSystem.resource FFS when mounting FFS partition hardfiles (b1).
Blitter DMA disabled, writing to BLTSIZE, writing to other blitter register(s) and finally enabling blitter DMA appears to start the blitter normally. (Pink Panther graphics corruption, probably worked accidentally in some old versions)

---

Beta 15:

Quote:

---

- IDE emulation drive status bit fix (DRQ and BSY active at the same time), fixes data corruption if IDEFix was active. (b5).
- Added support for previously unemulated and undocumented feature, sprites can appear 1 lores pixel earlier than first bitplane pixel. (First BPL1DAT access enables sprites immediately, only bitplanes need HRM documented delay). This is not normally visible because horizontal DIWSTRT usually equals bitplane left edge. Fixes "original" Rink-a-Dink/Lemon missing left white border in rotozoomer part.
- Do not use "8372 (Fat-hr) (agnushr),rev 5" VPOSR chipset id when config is ECS 2M chip RAM. This ID does not appear to exist in real world.
- Both bordersprite and borderblank active: sprites were incorrectly visible in upper and lower borders (241b13)
- Rewritten b4 "Subtle Shades" update, it was wrong and caused bad side-effects.

---

Beta 16:

Quote:

---

This beta should fix all strange display glitches introduced in previous 2.6 betas.

- New style filter mask config entry didn´t load. (b6 or later)
- Improved mouse movement mapping in GamePorts panel remap window (It creates mouse movement mappings that makes sense..), also implements 1s delay before next mouse movement is accepted in test/remap modes to reduce accidental events.
- Don´t attempt to combine left/right->horizontal and up/down->vertical if remapping single event.
- Add select menu didn´t include any "standard" device-specific events. (Incorrect filtering)
- DirectDraw + none filter was slow. (b7)
- Blitter changes in b12 still had some problems.
- "Subtle Shades" update was still broken, caused strange glitches in some situations.
- Left and right border was always fully blanked if scanline was in HAM mode.
- Low latency vsync + black frame insertion now always uses threaded vsync mode, normally was only used if fastest possible CPU mode is in use (If you have 100Hz+ monitor, especially lightboost, your PC shouldn´t be low end..), improves stability. No buffer also supported but it can be unstable.
- Added "gfx_autoresolution_delay" config entry. Number of frames to wait before autoresolution mode changes, timeout will always reset if mode change is detected during timeout countdown.

---

Beta 17:

Quote:

---

- One more IDE emulation fix, bogus extra write interrupt removed (I think this was accidentally introduced in some recent update). For some reason all tested drivers seemed to ignore it but it also could have caused data corruption in worst case.
- Added support for very large resolutions (NV Surround/AMD Eyefinity), max is now 8192x8192 (from 3072x2048) (I now have 3xVG248QE in portrait surround, 2d lightboost mode, GTX680 SLI. Debezeling still to do..)
- Display panel GUI didn´t accept new selected resolution if previously loaded config had non-existing resolution.
- B16 didn´t fix all glitches, yet another fix attempt. (Mindriot/Andromeda and more)
- Fixed sprite priorities in HAM mode (Thanks to Paradroid for this and right edge overscan test case)
- Added some uncommon aspect ratios to GUI aspect ratio select menus.
- Added new config entry "chipset_hacks" that is used to enable emulation of not fully known chipset features or bugs. Bit 0 = emulate COPJMP/Blitter cycle conflict (Blitter DMA channel gets loaded with Copper address). Bit 1 = emulate strange right edge overscan display shift if lores bitplane and scroll value is greater than 7. DO NOT enable in your default configuration, these are not fully emulated and can and will break other programs!

---

Beta 18: (Estimated release date: mid May)

Quote:

---

- If config (for example default.uae) has keyboard layout A/B/C selected, it will be remembered as default keyboard layout for selected port, then if new config is loaded and it has non-existing device (for example unplugged gamepad), port is configured for remembered default keyboard layout instead of always defaulting to layout A.
- Patch stupid CD32 ROM CD drive detection CPU delay loop if configured for non-approximate CPU speed modes.
- Validate RTG blit parameters more carefully to prevent invalid memory accesses if program(s) do bad things.
- winuaeenforcer disassembly crashed due to recent debugger changes.
- B15 sprite emulation update caused graphics corruption if display window was "too wide". (Himalaya / Avalanche)
- Mention also Administrator privileges in real harddrive/memory card "block zero write failed" messages.
- Modified emulation of ddfstrt/ddfstop value pairs that cause display corruption on real hardware due to bitplane dma being active too long and conflicting with refresh/strobe slots, prints log message.

---

Beta 20 RC1: (final in a few days)

Quote:

---

- AROS ROM updated.
- DirectSound mode crashed if DS sound play cursor got stuck (which normally shouldn´t happen)

---