r.335 - Win9X Game issues

[leilei]

I will be updating this post as more things are tested, i'm posting my preliminary results right now

Mobile Pentium MMX 133-200, a lot cache, S3 ViRGE DX, SB AWE32, Win98SE, 63MB RAM, DirectX 7.0

Standard -O3 -march=i686 -fomit-frame-pointer compile.

S3 ViRGE driver is 4.10.01.2122 as shipped with Win98SE
S3 Trio64 driver is 4.10.00.1686 as shipped with Win98SE

Smacker/Bink video garbage is disregarded.

General 16-bit color - In processors with the MMX recompiler, black lines and some shifted pixels by them occur in some games with the Trio64 and some other games with the ViRGE. They don't occur with the Winchip interpreter.

Airfix Dogfighter (Demo) - Selecting 3Dfx Glide causes PCem to crash after the company logo sequence. Selecting Direct3D(3dfx causes the game to freeze at a black screen after the company logo sequence. Occurs with Pentium, PMMX and Winchip Interpreter. Does NOT occur with Direct3D(ViRGE).

pclog.txt end from the Glide crash:

Code: Select all

Voodoo PCI write 00C3 00
onesec
Swap buffer 00000001 1 15680778
onesec
onesec
onesec
voodoo_readl  : bad addr 0000014C
PIC1 : MASK 28 PEND 00 INS 00 VECTOR 50
PIC2 : MASK 1D PEND 00 INS 00 VECTOR 58
Dumping rram.dmp
Dumping rram4.dmp
Dumping done
EAX=00000001 EBX=C8B82000 ECX=C8B82120 EDX=006EF590
EDI=006EF590 ESI=017B0498 EBP=00836EA0 ESP=006EF56C
PC=1778EE7 CS=0167 DS=016F ES=016F SS=016F FLAGS=0246
0167:1778EE1 0000:0000
377449259 ins
In protected mode
CS : base=000000 limit=FFFFFFFF access=FA  limit_low=00000000 limit_high=FFFFFFFF
DS : base=000000 limit=0000FFFF access=F6  limit_low=00010000 limit_high=FFFFFFFF
ES : base=000000 limit=0000FFFF access=F6  limit_low=00010000 limit_high=FFFFFFFF
FS : base=815FC414 limit=00000037 access=F2  limit_low=00000000 limit_high=00000037
GS : base=FFFFFFFF limit=FFFFFFFF access=00  limit_low=00000000 limit_high=FFFFFFFF
SS : base=000000 limit=0000FFFF access=F6  limit_low=00010000 limit_high=FFFFFFFF
GDT : base=C008347C limit=01FF
LDT : base=80004000 limit=3FFF
IDT : base=8009A000 limit=02FF
TR  : base=C000D7A4 limit=2069
386 in 32-bit mode   stack in 32-bit mode
CR0=80000013 CR2=80013FFF CR3=007FC000
Entries in readlookup : 597    writelookup : 51
Entries in readlookup : 254    writelookup : 7
ST(0)=0.000000	ST(1)=-1.000150	ST(2)=-240.000000	ST(3)=-320.000000	
ST(4)=1.000000	ST(5)=1.000000	ST(6)=0.000000	ST(7)=360.000000	
Status = 7804  Control = 023F  Tag = FFFF

Commandos (Demo) - Freezes after 2 seconds of gameplay on Trio64. Shows black box trails behind cursor. Seems OK in ViRGE

Dune 2000 - Using ViRGE DX will cause the screen buffer to wobble upward every frame. Trio64 does not exhibit this bug. Happens on Pentium, PMMX, and Winchip interpreter.

Freespace 2 - Freezes in main menu. Text texels are shifted in Direct3d(3dfx). Does not freeze and shift in Direct3D(ViRGE). Pentium/MMX/WinchipInterp make no difference.

Jazz Jackrabbit 2 - Using ViRGE DX in a 16-bit color mode shows the in-game menu with a white background instead of purple. Also occurs with IDT Winchip interpreter. Does not occur with Trio64


MDK2 - Can't fade from black. Affects gameplay, as the level starts the screen is stuck black. Most cutscenes aren't affected. Benchmark test is also affected. Also occurs with SGI Software OpenGL, so not a 3dfx bug. Initial and final versions of the game tested.

Sacrifice (build 0492, Oct 25 2000) - small font texel issues with Direct3D(3dfx)

pcemsacrifice.jpg (96.81 KiB) Viewed 21836 times

Simcity 3000 - Title freezes and black lines show up as well as some shifted pixels to the right of them, but only on Pentium MMX and Winchip Recompiler. Also occurs when using the 3dfx Voodoo for the display device (lol). Does not occur on Pentium. Does not occur on Winchip interpreter.




Games that work okay:

Evolva (Demo)
Treadmarks (1.0.1 Demo)

[ppgrainbow]

Have you updated to r335 yet? If so, are you still getting these issues?

[SA1988]

probably not related to this, but the cache should be set to a little for some cpu speed boost.

[leilei]

Performance is not a concern during these tests (and that's pretty important since gratuitous optimization flags can skew results and cause more bugs). If I wanted a boost right now i'd profile generate/use as usual

[SarahWalker]

Rev 341 should fix most of the MMX issues.

I've been unable to reproduce the Dune 2000 issue; does the wobble happen in normal gameplay?

[leilei]

It happens everywhere - menu, briefing, ingame, etc. in both 8-bit and 16-bit color modes I think it even resizes the window


Here's a gif of the wobble in r341

[Battler]

Performance is not a concern during these tests (and that's pretty important since gratuitous optimization flags can skew results and cause more bugs). If I wanted a boost right now i'd profile generate/use as usual

1. So you insist at ignoring the fact the optimization flags were added based on *TESTING* and verified results.
2. He was talking about the cache option in PCem settings anyway, so nothing to do with optimization flags. Maybe in ambiguous situations such as this one, it would be more courteous of you to ask what he meant when he said cache, rather than assuming out of the bat he was talking about optimization flags.

[SarahWalker]

Ah, Dune 2000 only shows that issue under that specific driver version. Rev 342 should fix that.

Battler, stop trying to pick a fight. It just makes you look ridiculous.

[SarahWalker]

Rev 344 fixes Airfix Dogfighter.

[SarahWalker]

Rev 345 fixes the Freespace 2 hang.

[leilei]

I believe 512x384x16 with ViRGE got regressed by the Dune2000 fix. Commandos is vertically stretched in that resolution now, as well as Springy Madness (Yay obscure shareware) They are fine in r335.

It also affected randomly in Jazz 2 with different resolutions.

[ppgrainbow]

I believe 512x384x16 with ViRGE got regressed by the Dune2000 fix. Commandos is vertically stretched in that resolution now, as well as Springy Madness (Yay obscure shareware) They are fine in r335.

It also affected 400x300x8 in Jazz2.

Are you getting the same issue when you update to r347 when you run Jazz2 in 400x300 8-bit colour mode?

[leilei]

The latest reivisions that aren't the ViRGE fix only deal with XT and Voodoo which is irrelevant to a ViRGE isolated issue.

and for the record, my last post dealt with revision 346.


Also I believe a hang in Rogue Squadron was also fixed by these recent changes

[leilei]

quick 347 tries:

Interstate 76 - Game quits/crashes after game loads. Software and Glide tested. Pentium, Pentium MMX and Winchip Interpreter tested.

The old demo also crashes after loading.


NOT PCEM BUG:

Giants: Citizen Kabuto - Blackscreens then crashes. errorlog.txt generated by Giants.xxe refers to an Integer Divide by Zero. Pentium, Pentium MMX and Winchip integer tested. (also i'm not sure if this game would work on an actual Voodoo Graphics card, or if it's faulting on the ViRGE being used for its Direct3D device) The issue is the game trying to initialize and use the S3 ViRGE instead of the 3Dfx Voodoo

[ppgrainbow]

quick 347 tries:

Giants: Citizen Kabuto - Blackscreens then crashes. errorlog.txt generated by Giants.xxe refers to an Integer Divide by Zero. Pentium, Pentium MMX and Winchip integer tested. (also i'm not sure if this game would work on an actual Voodoo Graphics card, or if it's faulting on the ViRGE being used for its Direct3D device)

Interstate 76 - Game quits/crashes after game loads. Software and Glide tested. Pentium, Pentium MMX and Winchip Interpreter tested.

Are you getting the same issues when updating to r352?

[leilei]

That implies i've applied patches that don't really matter to these tests (except the ViRGE patch which i'll get to later)...which i have not

though switching video to Trio64 does show that Giants fails on the ViRGE driver without a chance to initialize the Voodoo next, so chalk that as NOTPCEMBUG.

The black textures are likely actual V1 behavior given the latest drivers' age (1999) and the low video memory it has. This game WAS the "Crysis" of 2000 though, and a Voodoo failing at it is not surprising for a game that made Geforce256s cry...

[SarahWalker]

Rev 353 should get Interstate 76 to start. I'm still investigating why rendering is broken; I think this may be an FPU rounding issue.

[leilei]

3dfx Glide looks a lot better, but that has a buffer issue with the pause menu and also it freezes at some point during the first trip mission.

why i didn't think of testing I76 earlier i'm not sure. I'd assume that'd be fine if Battlezone works relatively better for being technologically derived

[SarahWalker]

Rev 354 fixes the software rendering. Hopefully it won't break anything else.

[leilei]

Revision 354

Dark Colony - Crashes before main menu.

Frogger - Software mode sometimes displays garbage on water and cars' windows in level 1. Glide is fine.

Jedi Knight - With Trio64, Weapon model doesn't appear in software rendering unless Backbuffer in System Memory is checked. Does not occur with ViRGE.

Lemmings Paintball - "Yippee" is not heard with Sound Blaster AWE32. Does not occur with Sound Blaster 16.

Warheads - Sometimes encounter strange physics and AI bugs? One time an enemy got stuck in space right in front of a planet's large crater where a gracitational pull wuith a stopping collission is expected (for it to proceed to detonation and advance the game). Might be too random to be reproducable. Pentium MMX tested.

pcemwarheads.png (55 KiB) Viewed 21080 times

Works OK so far:

Desktop Toys Vol 1
Hexen II
Jet Moto
Mageslayer
Necrodome
Take No Prisoners
Total Annihilation
Twisted Metal II
Urban Assault

[SarahWalker]

Jazz Jackrabbit 2 - Using ViRGE DX in a 16-bit color mode shows the in-game menu with a white background instead of purple. Also occurs with IDT Winchip interpreter. Does not occur with Trio64

I don't think this is a PCem bug - the two cards run the game in different modes; Trio64 in 565, ViRGE in 555. The menu effect doesn't appear to compensate for the difference.

[leilei]

you're probably right. it doesn't happen in windowed mode. chalked out as arjan brusee pebkac

[leilei]

While still on the subject of Epic color errors, r355 might've corrupted Unreal's textures in 3dfx.

Doesn't happen in r354.

[SarahWalker]

Rev 357 should fix that.

[leilei]

While not reported in this thread (but elsewhere), the Battlezone near clipping and particle texturing bug appears to be fixed as of revision 367.

Also I haven't ran into any gravity anomalies in Warheads yet but there is a disruptor explosion particle ramp bug that lasts a little too long.

[leilei]

The video in the Total Mayhem demo goes crazy. Game uses Scitech WinDirect/UVBELib. Trio64 and ViRGE/DX tested. rev370