WinUAE 3.6.0 beta series

[AC/DC HACKER!]

Originally Posted by Toni Wilen

Hard reset (GUI Reset button or ctrl+win+win+alt) technically should apply current expansion changes but it isn't doing it yet. It will happen someday..

On the fly without reset can get too risky..

I savvy that. Data corruption, hardware malfunction/wreck...etc. But for emulation at least data corruption in some way. As for your latter comment, cool.

[Toni Wilen]

http://www.winuae.net/files/b/winuae_3600b7.7z
http://www.winuae.net/files/b/winuae64_3600b7.7z

Beta 7:

- Added support for npcap without winpcap backwards compatibility option, fall back to winpcap if not found. (https://nmap.org/npcap/, fully Windows 10 1607+ compatible, original winpcap is not and win10pcap has at least one annoying bug.)
- Fixed partially corrupted first avioutput recorded frame.
- When saving state with avioutput active: last frame was not recorded.
- Number of avioutput frames buffered (waiting for compression thread to process them) counter was reset periodically which caused lost frames (and leaked memory) in recorded video if CPU was not fast enough.
- 680x0 STOP #X instruction undocumented feature emulated (documentation does not mention #X S-bit having any special conditions)
- 68000 (010?) X S-bit cleared and Trace not active: Updates SR, increases PC and then causes privilege violation exception. (Weird, isn't it?)
- 68000 (010?) exits after 4 idle cycles if trace was active, without trace it takes 8 cycles (or more if it needs to stop).
- 68020/030: works as documented, S-bit makes no difference.
- 68040/060 and X S-bit cleared: Does not modify SR, does not increase PC, causes immediate privilege violation exception.
- Added Kupke Golem HD3000 emulation. Uses existing OMTI adapter emulation, not tested, working install disk is MIA.
- Added Music Master dongle emulation.
- Added "unmapped zero" advanced chipset option. If set, all unmapped memory regions return zeros (previously always if non-ce 68020+). Added because B2000 (A2000 not tested) does this which is different than all other 68000 based Amigas, probably due to bus controller chip. This can change behavior of some badly coded programs.
- Directory filesystem ACTION_SET_DATE failed to change datestamp if file was read-only.
- Expansion device GUI changes are now activated after hard reset.

Kupke Golem HD3000:
- OMTI controller based. ("SASI CHS" HD mode)
- Byte wide hardware handshake PIO.
- v1.4 ROM added ("golem driver v1.4 (15 SEP 1989)")

Very experimental Direct3D 11 mode (DXGI 1.2+). Work in progress!

DXGI enables much better control of frame timing (not yet in use), allows windowed variable refresh rate support and more.

- Only windowed/full-window/fullscreen option supported.
- Multiple monitors supported.
- Direct3D11 on the fly switching supported.
- Filter settings are ignored.
- Scaling can't be adjusted.
- Enables (if available) windowed variable refresh rate mode (DXGI_FEATURE_PRESENT_ALLOW_TEARING, DXGI 1.5+). G-Sync confirmed working. Which is nice.
- In theory may work with Windows 7 (must have all updates) but may also accidentally require 8 or 10 (at this point). Windows 10 16299 only tested.
- Useless error message appears if things go wrong.
- Still requires d3d9x (=June 2010 DX redistributable) for matrix calculations.
- Calls SetMaximumFrameLatency(1) but I am not sure if it has any effect in windowed mode.
- Quick and dirty hack of Direct3D11 2D example tutorial.
- Shaders must be unpacked in plugins\filtershaders\direct3d11 (http://www.winuae.net/files/b/direct3d11_shaders.7z)

D3D11 discussion -> http://eab.abime.net/showthread.php?t=88777

[mark_k]

I tested with Wine 2.19 on Lubuntu 16.04.

3.6.0b2 works fine, but 3.6.0b3 and later crash when I load this config:

https://www.media!fire.com/file/d1z9quq95wfw45v/my_test_JIT_uaegfx_GVPI_RalphBabelDualROM.uae.zip

The last few log lines (the err one is output by Wine, not WinUAE):
hfd attempting to open: 'Z:\my_amiga_hd_4gb.bin'
HDF 'Z:\my_amiga_hd_4gb.bin' 06C230B8 opened, size=4194304K mode=1 empty=0
Geometry file '.\Configurations\default.geo' section '4294967296' found
err:seh:setup_exception_record stack overflow 1600 bytes in thread 0009 eip 7bc45e96 esp 069c0cf0 stack 0x69c0000-0x69c1000-0x6c40000

[mark_k]

I tested 3.6.0b7 on Windows 7 32-bit.

On switching from D3D9 to D3D11 mode WinUAE crashed (actually stuck using 100% CPU):
D3D11 free start
D3D11 freed3d start
D3D11 freed3d end
D3D11 free end
D3D11 free start
D3D11 freed3d start
D3D11 freed3d end
D3D11 free end
Allocated draw temp buffer (1600*1280*32) = 3B7A0020
D3D11 init start
D3D11 CreateDXGIFactory4 80004002
ExceptionFilter Trap: 00000000 08D608D2 00000000
D0 00000808 D1 00FF6504 D2 00000000 D3 00000000
D4 00000000 D5 00000000 D6 00000000 D7 00000000
A0 00005A04 A1 00003E60 A2 00C049E8 A3 00C04678
A4 00003E20 A5 00000000 A6 00C03A34 A7 00C049CA
USP 00C049CA ISP 00C80000
T=00 S=0 M=0 X=0 N=0 Z=0 V=0 C=0 IMASK=0 STP=0
Prefetch 00fe (ILLEGAL) 257c (MOVE) Chip latch 00000808
00FEACC6 257c 00fe acf0 ffe6 MOVE.L #$00feacf0,(A2, -$001a) == $00c049ce [00000000]
Access violation! (68KPC=00FEACC6 HOSTADDR=00000000)
Exception 2 (257c 0) at 0 -> fc07dc!
ExceptionFilter Trap: 00000000 08D607DC 08D608D2

Dump file:

http://www.media!fire.com/file/ne781427ynd649q/winuae_3.6.0_b7_2017.10.30_21.35.08.dmp.zip

[mark_k]

Testing on Windows 10 x64 1511 now... changing to D3D11 mode shows D3D11 ERROR! dialog and:
D3D11 free start
D3D11 freed3d start
D3D11 freed3d end
D3D11 free end
D3D11 free start
D3D11 freed3d start
D3D11 freed3d end
D3D11 free end
Allocated draw temp buffer (1600*1280*32) = 1784D020
D3D11 init start
CheckFeatureSupport(DXGI_FEATURE_PRESENT_ALLOW_TEARING) = 00000000 1
Refresh rate 50 not supported, using 0
D3D11CreateDevice 887a0004
D3D11 free start
D3D11 freed3d start
D3D11 freed3d end
D3D11 free end
D3D11 free start
D3D11 freed3d start
D3D11 freed3d end
D3D11 free end
D3D11 ERROR!

(Windows desktop is at 50Hz, config uses windowed mode.)

[Toni Wilen]

Originally Posted by mark_k

I tested with Wine 2.19 on Lubuntu 16.04.

3.6.0b2 works fine, but 3.6.0b3 and later crash when I load this config:

https://www.media!fire.com/file/d1z9quq95wfw45v/my_test_JIT_uaegfx_GVPI_RalphBabelDualROM.uae.zip

The last few log lines (the err one is output by Wine, not WinUAE):
hfd attempting to open: 'Z:\my_amiga_hd_4gb.bin'
HDF 'Z:\my_amiga_hd_4gb.bin' 06C230B8 opened, size=4194304K mode=1 empty=0
Geometry file '.\Configurations\default.geo' section '4294967296' found
err:seh:setup_exception_record stack overflow 1600 bytes in thread 0009 eip 7bc45e96 esp 069c0cf0 stack 0x69c0000-0x69c1000-0x6c40000

Does default.geo exists? If it does, you need to include it.

[Toni Wilen]

Originally Posted by mark_k

Testing on Windows 10 x64 1511 now... changing to D3D11 mode shows D3D11 ERROR! dialog and:

Changing = ? (on the fly changing or changing before starting emulation)

Only parameters to D3D11CreateDevice() are feature level (display mode isn't selected yet), it looks like this system is not D3D_FEATURE_LEVEL_11_0 (or _1) capable.

[Toni Wilen]

Originally Posted by mark_k

I tested 3.6.0b7 on Windows 7 32-bit.

Fixed. (CreateDXGIFactory4 to CreateDXGIFactory2 fallback was broken)

[mark_k]

Originally Posted by Toni Wilen

Does default.geo exists? If it does, you need to include it.

Yep...

http://www.media!fire.com/file/q3h9h3x6zucn1cq/default.geo.zip

Problem isn't Wine-specific, happens on Windows 10 too. Probably something to do with my HDF being exactly 4GB in size (0 mod 2^32). There's no matching entry in default.geo but WinUAE thinks there is.

[mark_k]

Originally Posted by Toni Wilen

Changing = ? (on the fly changing or changing before starting emulation)

Only parameters to D3D11CreateDevice() are feature level (display mode isn't selected yet), it looks like this system is not D3D_FEATURE_LEVEL_11_0 (or _1) capable.

Both on-the-fly (from D3D9) and on starting emulation.

Laptop has Nvidia Quadro NVS 135M which is appearly feature level 10_0.

[Toni Wilen]

Originally Posted by mark_k

Yep...

http://www.media!fire.com/file/q3h9h3x6zucn1cq/default.geo.zip

Problem isn't Wine-specific, happens on Windows 10 too. Probably something to do with my HDF being exactly 4GB in size (0 mod 2^32). There's no matching entry in default.geo but WinUAE thinks there is.

It does not explain it. Variable is long long and %llu is format string.

But log lies, redownload winuae.7z and try again. Now it shows correct detected section and also lists all key/value pairs.

[mark_k]

A quick update. I tested winuae.exe 2017-10-31 20:22 (x86) on Windows 10 1709 x64. That doesn't show the problem, neither does 32-bit 3.6.0b7. But 64-bit 3.6.0b7 does crash.

Could you upload a 64-bit winuae.exe so I can test that on Windows 10? I'll test the latest 32-bit winuae.exe in Wine next.

Edit: I tested latest winuae.exe in Wine. It didn't crash, relevant log output:
load config '.\Configurations\my_test_JIT_uaegfx_GVPI_RalphBabelDualROM.uae':0
fixme:volume:GetVolumeInformationByHandleW 0xf0 (nil) 0 (nil) 0x6c224c0 0x6c224b4 0x6c22cc8 1000
hfd attempting to open: 'Z:\my_amiga_hd_4gb.bin'
HDF 'Z:\my_amiga_hd_4gb.bin' 06C13A14 opened, size=4194304K mode=1 empty=0
Geometry file '.\Configurations\default.geo' section 'default' found
Buffers:50
Geometry file '.\Configurations\default.geo' section 'default' found
Buffers:50
Geometry file '.\Configurations\default.geo' section 'default' found
Buffers:50

Maybe there was some accidental undefined C behaviour which went away when you added the extra logging???

[Toni Wilen]

Originally Posted by mark_k

Laptop has Nvidia Quadro NVS 135M which is appearly feature level 10_0.

It requires feature level 11. No workarounds, 11 or 9 only.

Maybe there was some accidental undefined C behaviour which went away when you added the extra logging???

More likely due to different optimization level. Only official versions and betas are fully optimized (much longer build time). Test versions are not that optimized. Later..

[mark_k]

Originally Posted by Toni Wilen

It requires feature level 11. No workarounds, 11 or 9 only.

What features does WinUAE's D3D11 renderer need that aren't supported by level 10_0 hardware? I looked at the feature table on MSDN but couldn't see anything obvious.

[ceztko]

Originally Posted by mark_k

What features does WinUAE's D3D11 renderer need that aren't supported by level 10_0 hardware? I looked at the feature table on MSDN but couldn't see anything obvious.

What if there is none? D3D11 seems to me a better choice than D3D10 for long term support of the code/drivers from Microsoft and card manufacturers.

[headkase]

Originally Posted by ceztko

What if there is none? D3D11 seems to me a better choice than D3D10 for long term support of the code/drivers from Microsoft and card manufacturers.

Openly documented standards should have priority for real future-proofing. An open standard's documentation is likely to be more commonly accessible decades from now when some future programmer is trying to port the core-code into a corresponding future device.

And DirectX is Windows only, the core of WinUAE is also used in FS-UAE for other platforms, if WinUAE became too dependent on DirectX 11 then that could make the FS-UAE port needlessly hampered.

[Toni Wilen]

What does D3D11 has to do with this? I always use native Windows APIs, not some higher level wrappers that always suck. Core emulation is completely different thing and is portable.

D3D11 support will require full D3D11 capable display card. There won't be any workarounds, it won't be yet another implementation that adds more and more hacks due to old hardware. D3D9 or even DirectDraw isn't going anywhere (DD probably goes away when XP support goes away)

[mark_k]

Direct3D 11 works with any D3D 9 (or better)-class graphics card. Just pass pFeatureLevels = NULL to D3D11CreateDevice.

[Toni Wilen]

Argh. This is getting annoying.. How many times I have to repeat this (also in other thread): D3D11 support is based on D3D11 2D example and it REQUIRES feature level 11! Other parts of initialization code fails with lower feature level. I don't want to learn how D3D11 works, I am not going to change it -> older versions won't be supported! Also old max DX9/DX10 only gpus are also guaranteed to have driver bugs and may be also abandoned by now, this is not going to be D3D9 repeat. Either D3D9 or Buy a new PC.

[Amiga68k]

Not sure if this is beta behaviour, because it also occurs in 3.5, but anyway...

I get the best Sound/Toccata result (with both Eagleplayer and OSS) when I have JIT disabled and the CPU Frequency at "8x" (or even more, not tested that yet).

I start emulation with JIT enabled:

cpu_speed max
cachesize 8192
cycle_exact false

When I want to disable JIT the following settings are changed through uae-configuration:

cpu_speed real
cachesize 0
cycle_exact true

Now comes the part with the CPU Frequency (I want 8x), but the setting cpu_multiplier is not 'active' yet. Only when I change the setting in the GUI (F12) to 8x the option becomes known to uae-configuration.

I also tested the proper use afterwards (with uae-configuration). I can set it back to for instance 4, but only after following the 'GUI route':

uae-configuration cpu_multiplier 4