WinUAE 3.1.0 beta series
 

New year, new beta series. Not just a 3.0 bug fix series, includes new features too.

Main rule: Before reporting, make sure bug is NOT in previous offical version. (If it is -> post in separate thread). No feature requests.

More information: http://eab.abime.net/showpost.php?p=861456&postcount=1.


http://www.winuae.net/files/b/winuae_3010b1.zip

Beta 1:

- Lowered disk motor startup delay, more closer match with real hardware (at least matches some random drive in my A500)
- Memwatch points in KS ROM address space crashed during reset. (Wrong ROM mirror selection)
- A4000T NCR SCSI data bus width become 16-bit in cycle-exact mode. Caused write triggerable registers to start operation before full 32-bit value was written.
- Memwatch points didn't work in ROM addresses.
- CPU HALT state caused infinite loop in some configurations.
- 2900b16 68040 MMU MOVEM stack frame update was incorrect. (NetBSD crash, probably other Unix-likes too)
- Some 68000 MOVE variants seem to sample interrupt lines before writes and some after writes. Added special case for one variant, allows Made in Croatia / Binary to work in cycle-exact mode. Need more testing later, there does not seem to be any pattern.
- Last line sometimes flickered with random colors in "Double, fields/+" non-interlaced mode.
- RTG mode screenshots in 16-bit host color depth mode had wrong colors.
- Non-cycle exact modes with display vertical diw open but bitplane DMA disabled: in some situations display was corrupted or shifted. (Gloom AGA, Worms Directors Cut title screens)
- Directory filesystem exnext packet update didn't always work correctly if file was deleted during directory enumeration. (Which is technically illegal but it accidentally worked with OFS/FFS and become a feature..)
- Autoconfig space was not unmapped after it finished, broke during previous autoconfig rewrite.
- GUI didn't allow selection of custom SCSI ROM file.
- Completely rewritten CPU idle function, max CPU idle setting should now work without any slowdown side-effects, also lowest CPU idle state should be less than 10%. Please test.
- Last line is not special if A1000 Agnus. I was expecting this result when I noticed that A1000 sprite behavior is different in last line vs later Agnus models. A1000 Agnus: all visible lines are normal, later revisions: last line inhibits bitplane and sprite DMA. Real A1000 confirmed. I think this is last A1000 vs later Agnus revision undocumented difference.
- Small sprite update, only writes to SPRxCTL disarms sprite (I had special case for first DMA write to SPRxPOS too, only because it worked around another old bug..). Now OCS Agnus + ECS Denise + KS3.1 combination mouse cursor sprite corruption is 100% correct.
- On the fly media insertions are now serialized and only processed when Windows PnP manager reports idle state (Media mounting works even if it causes "installing drivers" phase) and multiple simultaneous media insertions also work reliably now (for example USB reader connected with multiple cards already inserted)
- First on the fly inserted device name becomes RDH0:, second RDH1: and so on. Previously number was next available directory filesystem unit number.
- Delay enumeration of inserted/removed input devices by 2 seconds. One removal or insertion can cause multiple device events (which is harmless but sometimes annoying)
- Picasso IV autoconfig missed special case, other boards are 16M in Z3 mode but PIV is 32M. Fixes incorrect address alignment if Z3 fast RAM is also enabled and size is less than 32M. (OS4 black screen hang)
- Picasso IV byteswap VRAM space didn't handle byte reads or writes correctly, only words and longs worked. (OS4 1024*768*16 screen mode glitches, possibly also other 16 bit modes)

New hardware emulated:

- GVP SCSI Series I and II hardware emulation (WD33C93), HC+8 board. v3.15 and v4.15 boot ROMs added to ROM scanner (gvpscsi.device). GVP SCSI enabled + Z2 fast RAM: use GVP Z2 fast ram autoconfig ID.
- DKB 1230/1240 accelerator board + SCSI emulation (FAS246 which appears to be compatible with 53C94/FAS216). SCSI is non-DMA (dkbscsi.device). v1.23 ROM added to ROM scanner. RAM not emulated, no idea about address space.
- Fusion Forty accelerator board emulation. It has interesting menu and graphics in boot ROM, press right mouse button when booting. v3.4 ROM images added to ROM scanner. RAM expansion emulated.
- Toccata Z2 sound card emulation.

Toccata hardware features emulated:
- Mono and stereo channel modes.
- 8-bit and 16-bit linear PCM. Companded (ulaw/alaw) modes are not supported.
- Left and right output channel volume control and mute.
- All AD1848 supported output frequencies.
- Output only. Input channels are not supported.

Technical details:
- Has 1024x9bit FIFO chip, one for output and one for input.
- No DMA. Uses level 6 interrupt to fill FIFO, 512 bytes/interrupt. Interrupt triggers when FIFO has 512 bytes or less
data available. (Which makes about 7 interrupts/frame at 44100Hz/16bit/stereo)

Notes:
- Different implementation than uae ahi and CD audio. Audio goes to Paula emulation backend (after Paula 8 to 16-bit conversions and filters). Very good latency but glitches are possible if configured buffer size is too small or not enough CPU power available (just like with Paula audio). Toccata emulation should be usable for (game) sound effects too, not just for music.
- Because it uses Paula audio path, Toccata audio is included with sound recording. Same method may be used with CD audio in future. (Possibly optionally).
- It is best to set same sample rate in both AHI and UAE sound settings to reduce possible resampling artifacts.
- Does any DMA capable simple sound cards exist? (Simple = basic CODEC chip only, no DSP chip)

Damn it man , have a day off ;)

*Happy New Year*

Paula's low-pass filter is applied to Toccata audio. This shouldn't happen IMHO.

Happy new year.

I just want to say that the toccata emulation seems to works fine in Octamed Soundstudio 1.03c.

How ever in DeliTracker there is no sound, while i can see the scopes moving.

I use the 14Bit-NotePlayer wich has direct Toccata support.
Version 9.10 of the toccata.library is in use. The player would argue if if the card were not there so why does it stay muted?

If I select Toccata in AHIprefs and select AHI in 14Bit-NotePlayer it works as intended.

Is that GVP Series II emulation only? I seem to remember that v4.xx ROMs don't work with Series I cards, so there must be some hardware difference between them.

Not possible!

Which audio filter mode? (none, sinc, anti and so on..)

EDIT: Never mind. I found out what I missed..
http://www.winuae.net/files/b/winuae.zip fixes it. Stereo separation is still after Toccata mixing and it is currently considered a feature.

I need _exact_ same programs and library versions.

I don't exactly know whats the difference between I and II, it seems to be mainly different internal clock rates which of course makes no difference in emulation. (Unless specifically restricted in emulation)

I got 2 Versions of the toccata library here: http://amiga.resource.cx/exp/toccata
DeliPlayer,OctamedSS and AHI are all the latest versions from aminet.
Test OS was OS3.9.

Version 2.5 of the toccata.lib
-Octamed
works
-DeliTracker(14bit-NP)
direct toccata - option is grey/unusable
AHI+toccata complains about library and dos not work
-other software that uses AHI not tested because of the above

Version 9.10 of the toccata.lib
-Octamed
does not work - all I can hear is a short crackling sound(~0.1s)
-DeliTracker(14bit-NP)
direct toccata does not really work,no sound/only visuals and somtimes a crackling noise
AHI+toccata works
-other software using AHI and the Toccata drivers(testet frogger and ffplay)
works

The GVP controllers (at least the Series II ones) have an autoboot disable jumper, so it would be nice to have a ROM disabled option (like for A2091).

I modified the A3000/040 config I used for testing NetBSD to have the HDF connected via GVP or Fastlane SCSI controllers, but then the system doesn't boot from it. I get a blank screen for a while then eventually the insert disk animation. Same with A2000/040/GVP config. [HDF link in this post.]

Does that match real hardware? Maybe the GVP & Fastlane ROMs don't support custom boot code?

A3000/040/Fastlane log output:

With A3000/040/GVP, log output saysbut there are no SCSI commands issued before the insert disk animation appears (with -scsiemulog option).


With an A4000/040/GVP config, there is a lot of log output with -scsiemulog, but the system eventually ends up at the insert disk animation. Log output starts with:Edit: Same problem booting my Debian HDF, which should boot from a normal FFS AmigaDOS partition, so this issue probably isn't related to custom boot code.

Correct test: exact same config but normal RDB with FFS bootable partition (with size less than 1G). If it boots: problem is in boot ROM.

EDIT: FastLane only mounts "OTHER" partition and seems to ignore others. (Easiest to see by booting with WB floppy in drive)

A3000 SCSI + another WD based SCSI controller does not seem to work.

WinUAE 3.1b1 crashes for me when ReKick is used with Zorro III RAM and GVP SCSI controller.

I know that's not a possible real-world config :). Previously I had been using a config to mostly match my A2000 setup, using ReKick to load Kickstart 3.1 from disk. That works fine with only chip + Zorro II RAM configured. However it fails with Zorro III RAM in addition to ZII RAM. I guess that failing isn't the bug, WinUAE crashing is.

First try with hardfile connected to UAE controller...
WinUAE 2.8.1: ReKick runs and resets into the new Kickstart. Emulation seems to lock up (I can press F12 to open settings window then quit) after this log output:WinUAE 3.1b1 is similar to the 2.8.1 case: it doesn't crash.

Trying the same config but with hardfile connected to GVP controller (with GVP ROM 4.15) and WinUAE 3.1b1, after ReKick resets:
Crash happens with both JIT and non-JIT. I can create a test HDF if needed. Dump files were created running under Wine so might not be any use.

Please for any tests use the latest Aminet toccata.library v12:

http://aminet.net/util/libs/toclib12.lha

If not use the attached last v14.1

Done

There is something in those HD images that GVP boot rom hates (Probably some bit in RDSK block that does not match, possibly LUN related), for some reason it starts executing INQUIRY commands forever, like it expects some non-zero LUNs.. Other images seem to work fine.


Fixed. WORD ROM reads accessed invalid memory. (Which means things are going horrible wrong already, it probably isn't going to boot anyway..)

No.

Search for filter author's description. It is not UAE specific.

It is.

Question that also needs to be answered: does old (or "too new") toccata libraries work with OS39 or with those exact version of programs you listed?

Thanks.
All problems solved with your attached lib.
Didn't even think about searching the aminet for the latest drivers. Maybe it was yesterdays hangover. :crazy

Best Version for os3.9 so far is v14.1 of the tocata.library.
AHI works and Octamed or other software, bypassing AHI, works fine,too.

So I wouldn't say there is a "too new" version.
About the two older versions I testd yesterday, see the above post.They just don't work.

Missing logs useless rep... ops sorry :laughing

Happy new year Toni!

Yes but someone with real Toccata should test exact same library and software versions to confirm it. Old libraries may access Toccata hardware differently or even use different addresses (It seems to have huge number of mirror addresses) and I am not sure if my address decoding guess is exactly correct.

-Fusion Forty accelerator board emulation

Thanks! I have a pair of them on the shelf here. Fun stuff.

http://www.winuae.net/files/b/winuae_3010b2.zip

Beta 2:

- Paula audio filter still filtered Toccata audio. Note that stereo separation is after Toccata audio mixing.
- Added "ROM disabled" to GVP SCSI ROM selection.
- GVP SCSI ROM space word reads accessed invalid memory.
- Added Paths panel option to store "saveimages" in same directory as original image file, instead of always saving in saveimages-directory, extension changed to .save_adf to prevent it being visible in directory listings with normal disk image filter type active.
- 68030 MMU on the fly off->on switch crashed. (Not that on the fly MMU switching is really supported..)
- WD33C93 Select without ATN command didn't fully skip message out phase. (AMAX GVP SCSI driver)
- Reading WD33C93 SCSI message in with Transfer Info command is a special case, after it finishes, it does not advance to next state automatically, it must wait for Negate ACK command. (AMAX GVP SCSI driver)
- AMAX rom "dongle" emulation improved floppy drives didn't work reliably.
- Added AMAX rom image selection to ROM panel SCSI select menu, renamed to SCSI/Boot ROMs.
- Added very simple next opcode word prefetch emulation to 68030 MMU mode, only done when MMU state changes. Fixes programs that first switch on (or off) MMU which also unmaps memory where currently executing code is located and then executes JMP (An).
- Added Toccata recording support, currently only internal hardware emulation is implemented (FIFO and interrupts), returns only zero samples.
- Show also inserted data CD volume id (label) in status bar, strip path from file name.
- Show inserted floppy image filename and label (if OFS/FFS image) in status bar.
- CD image audio can be now mixed with Paula audio, just like Toccata output. CD audio can be finally recorded with internal wave or avi recording support (Requested feature). FPS.Adj also adjusts CD audio speed (and pitch) and CD audio sound sync (For example Liberation CD32 intro) is kept even if emulation slows down. Enable option added to Sound panel (can be switched on the fly but mode only changes when CD audio thread restarts), old method is still the default. Note: when enabled, Paula audio slider becomes master volume for CD audio.
- CD32 FMV also supports new audio mixing mode. FPS.Adj not supported yet.

Remember that the less m68k Amiga-like (and/or less interesting to me, usually it is "and") some nice to have (To you) emulation feature is the more support is needed: docs, specs, testing help, reasons why it would be good idea, donations (hardware/money/whatever) and so on.

http://www.winuae.net/files/b/winuae_3100b3.zip

Beta 3:

- Bitplane DMA sequencer emulation rewritten yet again to again match better with real hardware behavior. DDFSTRT and DDFSTOP seems to use hidden counter that counts 4 cycles ahead of real hpos counter. When DDFSTRT matches, next 4 cycles appears to be idle (probably used to init and fill some bitplane dma sequencer shift register?) and after this delay DMA starts normally. Same for stop. Now programs that write to DDFSTRT or DDFSTOP just before, exactly in same cycle or just after it should match now just work without any extra hacks. (Some kind of 4 stage shift register would also explain why changing number of bitplanes by writing to BPLCON0 mid-line always take effect after exactly 4 cycle delay) Can break anything as usual! Correctly breaks SCX_CoolSpot1.adf's intro on ECS Agnus configurations :)
- 512k chip + 512k slow ram + ECS Agnus chip mirroring didn't set slow ram bank with "no memory allocated" flag, could have caused crashes when it was attempted to be freed.
- OCS Agnus 512k chip + 512k+ slow and 1M/0.5+0.5 advanced chipset checkbox checked: move slow ram start at $080000.
- 68000 bus error and address error exception stack frame's first word undocumented bits 6 to 15 seem to contain opcode of faulted instruction. (from Hatari)
- IDE/ATAPI emulation decoupled from Gayle emulation and moved to separate source file.
- Added GVP A3001 board with IDE. Apparently Series I and II have different hardware and drivers. (v4 for Series II) Series I IDE emulation seems to be fully working. Non-DMA, gvpat.device, v3.3 added to ROM scanner. Does not byte swap data words but emulation swaps it automatically to keep HDFs compatible with other controllers.
- Added ES_SYSTEM_REQUIRED flag to SetThreadExecutionState(), perhaps it keeps screensaver inactive. (ES_DISPLAY_REQUIRED seems to only make sure monitor does not switch to power saving mode)
- HAM-E emulation added. Because HAM-E "magic cookie" can be nearly anywhere on screen, enabling HAM-E will slow down emulation (It needs to check nearly all pixels every frame) even when HAM-E mode is not active. "Automatic" setting still only does A2024 and Graffiti detection. DCTV emulation: not going happen without image encoding specs.

Little mistake: