A lot of has happened past week, perhaps thanks to the summer finally starting to be over, and us having more time for the project again. Here’s a breakdown of most important advances made last week:
Fixed the DIVU.W overflow calculation
A long-time bug in unsigned 16-bit division instruction was finally found, and fixed by Pukka! We accidentally calculated overflow using signed 32767 limit, while it should have been unsigned 65535. This immediately fixed highly noticeable glitches in two of the test applications, Vette! and Test Drive II: The Duel (both ironically of same racing simulation genre):
Before the fix, Test Drive II had noticeable warping and glitching in the road near the bottom, and the polygons in Vette! had sometimes incorrect tangents, causing them to get messy depending on the angle of polygon edge.
This also fixed the “Divide overflow” bug occurring in SoftPC, which was weirdly happening on certain times of day, which does indicate that some calculation related to the computer clock time was failing because of the incorrectly signaled overflow. Now SoftPC works quite reliably.
Bitfield instruction fixes
One recently added test application, Prince of Persia 2, used a lot of the bitfield instructions for the SHAP resource drawing on the screen (apparently RLE/LZ decoding), which originally had a really messy artifacts:
But after fixing the instructions, the levels finally looks as they should:
Time Manager fixes
Besides the bitfield instructions, Prince of Persia 2 also uses Time Manager heavily for timing, and that surfaced a couple nasty bugs in the code, which were fixed.
There are still so oddities to investigate, for example the intro sequence does not advance past second screen yet, and the death and level completion delays do not trigger during the gameplay.
Sadly these fixes do not (yet) fix the one special case in Speedometer, where Dhrystone test does not work properly because of some issues that may be related to Time Manager (it starts the same timer twice, and stops it before test, so it never runs).
Keyboard handling threading bug
There was also a nasty bug in keyboard handling, which had rather surprisingly managed to stay hidden for a quite long time; Basically, the SDL2 key events that were handled by the keyboard code on main thread had chance to accidentally modify data structures and especially the A0 register, while 68K was running in VBL interrupt handler, causing in some rare cases the A0 register value to get corrupted; This caused a lot of crashes in Prince of Persia 2, and is most likely the cause of a very rare crash in Wolfenstein 3D. The offending part code was made thread-safe, which fixed this particular crash completely.
Sound in SoftPC
This was a very minor fix, but still worth mentioning; the reason for why there was no sound in SoftPC was caused by the fact that the application was setting the sound driver volume from the SPVolCtl low memory global, which had been set to zero on startup. By improving the PRAM simulation to populate all low memory PRAM values with sensible default values, the sound works now nicely in SoftPC!
Dark Castle “_Random” bug
As Pukka guessed last Christmas, the reason for buggy behaviour of robots, and some other minor glitches, in Dark Castle was indeed related to calling of the “_Random” trap! Of the last weekend was spent on searching the bug in other places at first, such as robot behaviour code, until we noticed something strange.
In Apple documentation, and all references known to us, the QuickDraw random number generator is defined simply as a Pascal type function with the following signature (in IM: Mathematical and Logical Utilities, 3-51):
pascal short Random (void);
Which means, that the function takes no arguments, and returns a 16-bit signed integer on stack. But, however, the way Dark Castle used was rather odd:
This screenshot of ResEdit showing disassembly of the part of Dark Castle code calling _Random shows, that it actually ignores the result value at offset 000B60 by discarding it off stack! And later looking at the code, it was actually using value in D0 instead…so, a quick look at System 7 _Random trap handler gives some insight to this:
Here actually it is visible, that on real Macs, the trap trashes a number of registers, and has the side-effect of leaving result value in the D0 register (being a toolbox trap, the trap dispatcher does not clean up any of the registers, unlike it does for some of registers after calling OS traps).
So, to fix Dark Castle, we added extra code to set up the D0 register as the game expected it, and everything started to work perfectly.
Some other progress from these changes
A couple other test applications seemed to benefit of these changes too, although they still need a bit of work to make sense: Loony Labyrinth now loads correctly after bitfield instructions were fixed (and physics in it also work correctly after the DIVU.W fix), and Marathon now actually runs and is playable with the Time Manager fixes:
Interestingly both of these games still seem to suffer from having colors completely messed up, but besides this glitch, they appear to be playable and run quite nicely.
And to celebrate getting SoftPC finally to work properly, I recorded this short video of M.A.C.E. running SoftPC running David Murray’s (The 8-bit Guy) Planet X-3:
You can find more information on Planet X-3 here: http://www.the8bitguy.com/product/planet-x3-for-ms-dos-computers/
Full list of changes since last post
2020-09-09 23:29:07 +0300 • Merge branch 'master' of 2020-09-09 23:28:59 +0300 • Update CMake scripts to support Apple notarization 2020-09-09 21:14:01 +0300 • add missing addressing fixes 2020-09-09 20:32:11 +0300 • memory addressing fixes 2020-09-09 20:26:50 +0300 • Fix compilation error in QDColorStretchBits.c 2020-09-09 20:00:29 +0300 • Fix some warnings in StretchBits color blitters 2020-09-09 19:59:57 +0300 • Fix some warnings in CodeResourceWDEF1 2020-09-09 19:26:58 +0300 • Fix return value in C_PlotCIconHandle 2020-09-09 19:18:10 +0300 • Merge branch 'master' of 2020-09-09 19:16:59 +0300 • fixes for bitfield and divuw instructions 2020-09-07 04:23:28 +0300 • Tweak more Dark Castle CMake JSON configs 2020-09-07 00:04:11 +0300 • Simulate PRAM (SysParam) to make SoftPC sound work 2020-09-06 23:04:58 +0300 • Update MACE version in CMake config 2020-09-06 13:14:44 +0300 • Color pixeldouble & (disabled) CRT simulation 2020-09-06 03:45:52 +0300 • Update trap generator XLS file 2020-09-06 03:44:13 +0300 • Tweak Dark Castle JSON config settings 2020-09-06 01:52:48 +0300 • Hack Random result value in D0 to fix Dark Castle 2020-09-05 03:54:17 +0300 • Get correct tmCount in RmvTime for non-first tasks 2020-09-03 02:16:54 +0300 • Fix the conversion of positive count in PrimeTime 2020-09-02 19:30:04 +0300 • Fix threading issue with keyboard input module 2020-09-02 19:28:30 +0300 • Don't crash on purged handles in GetHandleSize 2020-09-02 19:27:42 +0300 • Fix a lot of Time Manager bugs & clean up the code 2020-09-02 19:23:45 +0300 • Fix typing of region handle to CMDoDrawControls 2020-08-31 23:01:40 +0300 • A simplified MakeFSSpec implementation, for POP2 2020-08-31 04:17:44 +0300 • Add missing EqualRect change in EmuUtils.h 2020-08-31 04:16:59 +0300 • Fix GetCCursor crash when ResLoad was FALSE 2020-08-31 04:16:08 +0300 • Add dummy SCSIDispatch for Loony Labyrinth 2020-08-30 21:32:44 +0300 • Add Prince of Persia 2 to test app JSON configs 2020-08-03 01:45:10 +0300 • Make NewGWorld's rowBytes to work w/ Color VETTE! 2020-08-01 20:08:42 +0300 • Record PICT2 opHiliteColor, opOpColor &opDefHilite 2020-08-01 19:45:53 +0300 • Record HiliteBit as opHiliteMode in PICT2 format