Thursday, May 9, 2024

Running SWEET16, Steve Wozniak's "The Dream Machine" from WozMon


The story goes that Steve Wozniak was running out of space in Apple Integer BASIC, and out of necessity, determination, and will power, developed SWEET16 to do 16-bit operations in a more compact fashion.  There is already a great article on SWEET16, the interpretive processing 16-bit supplement to 6502 machine language programs, over at 6502.org.  The article tells the history, instructs how to port, and provides a tutorial on how to use it.  (And see article on Wikipedia, and Steve Wozniak's 1977 article in Byte magazine).

My article will follow the steps to produce a binary to run on my minimal 6502 with MC6850 UART, and run through a few examples to use it together with the Apple-1 Hex Monitor, ported from listings included in the Apple-1 Operation Manual (1976).  The Hex Monitor (or WozMon as people generally refer to it) is the original 256-byte ROM included with the Apple-1 computer.  It only needed to include display memory, edit memory, and run from memory operations to initiate control of that computer.

This article and efforts are inspired by videos produced by Ben Eater and his 6502 breadboard computer.  I wanted to do something similar, and followed instructions he provided to do something similar with my 6502 emulator, but simpler.   

I already had a 6502 emulator.  One version is in C++, is very cross-platform (Windows, Linux, Mac, Arduino, STM32, etc.).  But I had focused on Commodore emulation to match the computers I grew up with learning to program.  While that is working well, it is using Commodore ROMs and some I/O emulation. 

Also inspirational are similar videos produced by Michael Cartwright for his Hopper development environment and 6502 breadboard and PCB computer, inspired by Ben Eater's 6502 breadboard computer.  Michael (sillycowvalley on github) uses the MC6850 UART instead of the MC6821 PIA used in Apple-1, or MOS6522 VIA or MOS6551 UART used by Ben Eater.

I surely didn't want to bit bang serial, and wanted it as simple as possible.  Reviewing resources on the web, including datasheets, it was apparent the MC6850 has an advantage of being rather minimalistic.  It has one data register (read/write), one control register (write), and one status register (read).  While originally intended for use with 6800 series microprocessors, it can be used with others.  Only caveat is that the status is not designed for the 6502 BIT instruction.  But the advantage is that it doesn't have a status bug like the 6551 which intended to replace it.

So, I sat down and added a minimal emulation profile to my emulator including configurable RAM/ROM, and I/O address for the MC6850.  My initial implementation supports polling but not interrupts, which is fine for our purposes today.  It does not include a parallel I/O chip, so no hooking up virtual LEDs.  Wait, what?  Oh yeah, we don't need those (yet?), we exclusively have a terminal console.   But that also means we don't have a way to load or save programs.

Also imagine 60K RAM and 4K ROM here (oops!)

I ported WozMon to the 6502/MC6850 minimum profile so it will serve as our launching off point, similar to how it was utilized on the Apple-1.  Porting required changing the assembler syntax to match my 6502 assembler of choice (ACME), especially note that all the hex values in the file should include # for immediate mode (if extracting from Usenet article instead of Byte article), DFB must stand for define byte, and you do need to include some save/restore register routines.

An advantage of having a minimal system unlike Commodore, is that the 32 consecutive zero-page bytes necessary for SWEET16 are guaranteed to be available because it is a minimal system, addresses are not used by IRQ routines, keyboard handlers, timers, KERNAL I/O, etc.

Steps

1. Clone (or download Zip of) the emulator c-simple-6502-cbm.  Note it is not necessary to gather the Commodore ROMs this time around, because we don't need them.

2. Compile using Visual Studio, or make/gcc (Makefile included), or similar.

3. Look for wozmon.bin in the roms/minimal folder (its ported source is in child src folder).

4. Launch the minimal profile

    c-simple-6502-cbm.exe 1 roms/minimal/wozmon.bin

Now WozMon should be running.  You can review the Apple-1 Operation Manual (1976) for more detailed instructions on how to use WozMon.  But if you're quick, you can follow along with the required inputs here.

5. Paste SWEET16 binary into WozMon.   See copyright at start of article, included here for educational purposes, trusting this is fair use and for your own personal use.  Not for commercial use.

# SWEET16 compiled for EE80 start address
EE70: 20 EA EE 68 85 1E 68 85
EE78: 1F 20 7F EE 4C 79 EE E6
EE80: 1E D0 02 E6 1F A9 EF 48
EE88: A0 00 B1 1E 29 0F 0A AA
EE90: 4A 51 1E F0 0B 86 1D 4A
EE98: 4A 4A A8 B9 C8 EE 48 60
EEA0: E6 1E D0 02 E6 1F BD CB
EEA8: EE 48 A5 1D 4A 60 68 68
EEB0: 20 F6 EE 6C 1E 00 B1 1E
EEB8: 95 01 88 B1 1E 95 00 98
EEC0: 38 65 1E 85 1E 90 02 E6
EEC8: 1F 60 00 F7 02 9B 0B 9C
EED0: 23 AD 14 B0 45 B7 4F BE
EED8: 2D C7 59 D0 83 DB 6C 03
EEE0: 31 E6 6E 91 1C E5 63 E5
EEE8: E5 E5 85 20 86 21 84 22
EEF0: 08 68 85 23 D8 60 A5 23
EEF8: 48 A5 20 A6 21 A4 22 28
EF00: 60 10 B3 B5 00 85 00 B5
EF08: 01 85 01 60 A5 00 95 00
EF10: A5 01 95 01 60 A5 00 81
EF18: 00 A0 00 84 1D F6 00 D0
EF20: 02 F6 01 60 A1 00 85 00
EF28: A0 00 84 01 F0 ED A0 00
EF30: F0 06 20 64 EF A1 00 A8
EF38: 20 64 EF A1 00 85 00 84
EF40: 01 A0 00 84 1D 60 20 24
EF48: EF A1 00 85 01 4C 1D EF
EF50: 20 15 EF A5 01 81 00 4C
EF58: 1D EF 20 64 EF A5 00 81
EF60: 00 4C 41 EF B5 00 D0 02
EF68: D6 01 D6 00 60 A0 00 38
EF70: A5 00 F5 00 99 00 00 A5
EF78: 01 F5 01 99 01 00 98 69
EF80: 00 85 1D 60 A5 00 75 00
EF88: 85 00 A5 01 75 01 A0 00
EF90: F0 E9 A5 1E 20 17 EF A5
EF98: 1F 20 17 EF 18 B0 0E B1
EFA0: 1E 10 01 88 65 1E 85 1E
EFA8: 98 65 1F 85 1F 60 B0 EC
EFB0: 60 0A AA B5 01 10 E8 60
EFB8: 0A AA B5 01 30 E1 60 0A
EFC0: AA B5 00 15 01 F0 D8 60
EFC8: 0A AA B5 00 15 01 D0 CF
EFD0: 60 0A AA B5 00 35 01 49
EFD8: FF F0 C4 60 0A AA B5 00
EFE0: 35 01 49 FF D0 B9 60 A2
EFE8: 18 20 64 EF A1 00 85 1F
EFF0: 20 64 EF A1 00 85 1E 60
EFF8: 4C AE EE 00 00 00 00 00

Note: SWEET16 entry point for this port is $EE70, and the operations are in a single page $EF01-$EFFA.   A save registers routine is at $EEEA and restore registers is at $EEF6.   Zero page usage is $00-$1F for the SWEET16 registers, and $20-$23 for saving 8-bit registers.

6. Let's try it out.  Start by pasting the following to enter a program at 

1000: 20 00 FE A9 2A 20 70 EE 
1008: 11 12 34 12 56 78 21 A2
1010: 33 00 8D F8 FF 00

Which is equivalent to

20 00 FE JSR $FE00
A9 2A    LDA #$2A
20 70 EE JSR $EE70
11 12 34 SET R1,$3412
12 56 78 SET R2,$7856
21       LD  R1 
A2       ADD R2
33       ST  R3
00       RTN
8D F8 FF STA $FFF8
00       BRK

JSR $EE70 switches to SWEET16 interpretation, and these are SWEET16 instructions through RTN, before being back in 6502.  The vectors are set such that a BRK will return to WozMon. ($FE00 is my specific UART initialization routine for MC6850, and storing a byte to $FFF8 will send it across the UART).

All this program does is set the R1 and R2 16-bit registers to constant values, the LD instruction loads the 16-bit accumulator from R1, then ADD R2 adds R2 to the accumulator, and the result is stored in R3 by the ST instruction.   RTN returns to normal 6502 operation.

7. To run the program, enter the following into WozMon:

1000R

8. Display the results by entering

0.1FR

From the last two steps the output should look something like the following (blank lines added for separation)

1000R

1000: 20*\


0.1F

0000: 68 AC 12 34 56 78 68 AC
0008: 00 00 00 00 00 00 00 00
0010: 00 00 00 00 00 00 00 00
0018: 00 00 00 00 00 06 12 10

And the accumulator (R0) at addresses 0 and 1 now has the value $AC68 which appears to be the sum of the other two 16-bit values, with a copy also stored at R3 (addresses 6/7).  The asterisk is printed to demonstrate registers being saved/restored and 6502 operation returning to normal.

That's just a quick example.   Hopefully you've got a little taste of WozMon and SWEET16. 

Example run of SWEET16

But that's not all!  

I made a Commodore port of WozMon, so why not also port SWEET16 to Commodore and run it there too?

# Commodore port of SWEET16
1540: 85 22 86 23 84 24 08 68
1548: 85 25 d8 60 a5 25 48 a5
1550: 22 a6 23 a4 24 28 60 00
1558: 00 00 00 00 00 00 00 00
1560: 00 00 00 00 00 00 00 00
1568: 00 00 00 00 00 00 00 00
1570: 00 00 00 00 00 00 00 00
1578: 00 00 00 00 00 00 00 00
1580: 20 40 15 68 85 7e 68 85
1588: 7f 20 8f 15 4c 89 15 e6
1590: 7e d0 02 e6 7f a9 16 48
1598: a0 00 b1 7e 29 0f 0a aa
15a0: 4a 51 7e f0 0b 86 7d 4a
15a8: 4a 4a a8 b9 d8 15 48 60
15b0: e6 7e d0 02 e6 7f bd db
15b8: 15 48 a5 7d 4a 60 68 68
15c0: 20 4c 15 6c 7e 00 b1 7e
15c8: 95 61 88 b1 7e 95 60 98
15d0: 38 65 7e 85 7e 90 02 e6
15d8: 7f 60 00 f7 02 9b 0b 9c
15e0: 23 ad 14 b0 45 b7 4f be
15e8: 2d c7 59 d0 83 db 6c 03
15f0: 31 e6 6e 91 1c e5 63 e5
15f8: e5 e5 00 00 00 00 00 00
1600: 00 10 c3 b5 60 85 60 b5
1608: 61 85 61 60 a5 60 95 60
1610: a5 61 95 61 60 a5 60 81
1618: 60 a0 00 84 7d f6 60 d0
1620: 02 f6 61 60 a1 60 85 60
1628: a0 00 84 61 f0 ed a0 00
1630: f0 06 20 64 16 a1 60 a8
1638: 20 64 16 a1 60 85 60 84
1640: 61 a0 00 84 7d 60 20 24
1648: 16 a1 60 85 61 4c 1d 16
1650: 20 15 16 a5 61 81 60 4c
1658: 1d 16 20 64 16 a5 60 81
1660: 60 4c 41 16 b5 60 d0 02
1668: d6 61 d6 60 60 a0 00 38
1670: a5 60 f5 60 99 60 00 a5
1678: 61 f5 61 99 61 00 98 69
1680: 00 85 7d 60 a5 60 75 60
1688: 85 60 a5 61 75 61 a0 00
1690: f0 e9 a5 7e 20 17 16 a5
1698: 7f 20 17 16 18 b0 0e b1
16a0: 7e 10 01 88 65 7e 85 7e
16a8: 98 65 7f 85 7f 60 b0 ec
16b0: 60 0a aa b5 61 10 e8 60
16b8: 0a aa b5 61 30 e1 60 0a
16c0: aa b5 60 15 61 f0 d8 60
16c8: 0a aa b5 60 15 61 d0 cf
16d0: 60 0a aa b5 60 35 61 49
16d8: ff f0 c4 60 0a aa b5 60
16e0: 35 61 49 ff d0 b9 60 a2
16e8: 18 20 64 16 a1 60 85 7f
16f0: 20 64 16 a1 60 85 7e 60
16f8: 4c be 15

This SWEET16 entry point for Commodore changed to $1580 so the sample program has to change to call this entry point instead, to call $FFD2 for screen output instead of using the UART, and jump to WozMon entry at end instead of BRK.   Note that this Commodore port puts the SWEET16 registers at $60 (instead of $00) in zero page.   The Commodore port was trivial, disassembly will show there's nothing secret going on, mostly moving around stuff in zero page, and moving the save/load registers code.

1000: A9 2A 20 80 15 11 12 34
1008: 12 56 78 21 A2 33 00 20
1010: D2 FF 4C 00 14
 
Vic-20 running WozMon and SWEET16

C64 running WozMon and SWEET16

It appears there's a bug restoring the registers.  That is not an asterisk!!!   But an underscore.   Exercise left for another day...   Update: I had replaced the call to UART_INIT from the other system with starting SWEET16 followed by 6502 code.  Bad, so bad (see screenshot for bad code).  Have since updated the sample program to be correct.   Disassembling here to double check.   The asterisk is now showing up fine!  Would have been doing some crazy stuff interpreting 6502 opcodes as SWEET16.

A9 2A    LDA #$2A
20 80 15 JSR $1580
11 12 34 SET R1,$3412
12 56 78 SET R2,$7856
21       LD  R1 
A2       ADD R2
33       ST  R3
00       RTN
20 D2 FF JSR $FFD2
4C 00 14 JMP $1400


Wednesday, May 1, 2024

C64 RAM locations for programmable characters

 


Taking the information from Jim Butterfield's interactive program we can determine a few optimal choices for configuring the location of video and character set memory.  They both need to be in the same 16K bank of the C64 memory, video memory needs to be accessible to the KERNAL, and some locations are off limits (offsets $1000-$1FFF within the bank) because the VIC-II only sees D000-DFFF ROM characters there.

Option 1: easiest, video $0400, chars $2000-$2FFF

Option 2: least BASIC free, video $8C00, chars $A000-$AFFF

Option 3: largest free, video $CC00, chars $F000-$FFFF


Option 4: single charset, video $C400, chars $C800-$CFFF


Thanks to Jim Butterfield's SCREENMAP 64 program, it prompts for you to choose between valid options, and then calculates the register and screen page values that you need to change(poke).  Shown above are details for the four options that seem best.  The first three allow for two full sets of characters.  The last one allows for a single character set, or one and a half character sets.

In addition to the steps shown, it may be necessary to install an NMI handler to keep the values set instead of reset to ROM defaults, and it will be necessary to load a font into the desired programable character set address range.

Options 3 and 4 are interesting because they both increase BASIC RAM by 1K more than normal by moving video screen memory elsewhere. Option 1 is the easiest to implement, so is perfect for getting started.


Notes:

1000-1FFF not available for VIC because maps to D000 ROM
5000-5FFF not available for VIC because maps to D000 ROM
9000-9FFF not available for VIC because maps to D000 ROM
D000-DFFF not available for VIC because maps to D000 ROM

C000-CFFF not recommended for prog. chars. because other RAM in that bank not usually available to KERNAL needed for video screen characters.   So if screen characters are there, only room for one set (2K) instead of two sets (4K)

Option 0 (Normal)
0400-07FF screen memory (1K) = 1024
0800-9FFF BASIC RAM (38K) 
C000-CFFF free (4K)
D000-DFFF character ROM banked under I/O

Option 1 (First 16K)
0400-07FF screen memory (1K) = 1024
0800-1FFF free (6K)
2000-2FFF prog. chars. (4K) = 8192
3000-9FFF BASIC RAM (28K)

4000-7FFF is smack dab in the middle of BASIC RAM, seems like a bad idea unless used with 6502 machine language or other assets around it, and not strictly BASIC.

Option 2 (Third 16K)
0400-8BFF BASIC RAM (34K)
8C00-8FFF screen memory (1K) = 35840
9000-9FFF free (4K)
A000-AFFF prog. chars in RAM bank under BASIC ROM = 40960

Option 3 (Fourth 16K)
0400-9FFF BASIC RAM (39K)
C000-CBFF free (3K)
CC00-CFFF screen memory (1K) = 52224
E000-EFFF = 57344 prog. chars in RAM bank under KERNAL ROM *or* F000-FFFF = 61440

Option 4 (Fourth 16K)
0400-9FFF BASIC RAM (39K)
C000-C3FF free (1K) or half lowercase character set (no inverse characters)
C400-C7FF screen memory (1K) = 50176
C800-CFFF prog. chars (1 set, 2K only) = 51200