Z88dk: 8080 support

[suborb]

I keep coming back to this in my head.

I think we can probably do something here that’s not too intrusive. Ideally the 8080 abi would be bcde rather than dehl but that’s a large change.

• [x] Cleanup stack by moving hl into bc
• [x] Disable shift optimisations
• ~[ ] Enforce 32 bit maths - mbf32 is 8080 code~ - actually depends on the library. There might be a 48 bit 8080 library out there somewhere
• [x] Replace all 16 bit sbc with 8 bit equivalent
• [x] Replace ld de,() with ex de,hl pair + hl load
• [x] Implement l_long_mul, l_long_div, l_long_div_u (steal from ack?)
• [x] Assess opt rules (mark with cpu flag)
• [x] Disable builtin functions for 8080
• [x] Review c stdlib
• [x] Revive old 8080 stdio
• [x] Disable extended opcodes in ticks and error so we catch stuff.
• [x] Revive old printf
• [ ] Revive old scanf on demand
• [x] Restore malloc() and friends
• [x] Get testsuites running (in progress)
• [x] Add -m8080 to z80asm, mapping jr -> jp, djnz
• [x] z80asm emulation library for ldir/ldi/ldd/lddr (ldir,ldi done)
• [x] Fix/implement sprintf/snprintf
• [x] Compile CP/M library for 8080

The main target would be cpm of course.

[feilipu]

Another 8008 and 8080 32-bit math option could be built too, the Lawrence Livermore (National) Labs Floating Point Library, circa 1973.

Integrating this library into z88dk is on my list (admittedly sidelined until we have 32-bit math support in newlib), because of its commonality with Am9511A. I'm not 100% certain, but I'd guess that the Am9511A was built to replicate the LLL floating point format.

[suborb]

I had a look at that last night - I couldn’t see the trig functions - did I miss them?

[feilipu]

No. You didn't miss them.
There's nothing except the 4 arithmetic functions, and some helper functions to convert fixed to float and back, in the library. Pretty basic. But probably used to design ICBM flight paths, amongst other things.

This is the original 8080 code, before conversion to Zilog mnemonics, from Herb's stuff.

[suborb]

This is a little bit of an ordeal but the basics are almost there.

Long support is currently broken, stdlib has mostly gone missing along with all the algorithm and allocation libraries - hopefully I can get the regular malloc back in.

The old c based printf (mini version) has reappeared - writing printf in assembler once is enough for me - so bits of the test suite are now working.

[zx70]

Quite obvious, but it is too funny to resist !
This one is tested:

._ldir
    push        af
.ldir_loop
    ld      a,(hl)
    ld      (de),a
    inc     hl
    inc     de
    dec     bc
    ld      a,b
    or      c
    jr      nz,ldir_loop
    pop     af

(ret)

It is 11 bytes long, probably it is worth to keep it as a subroutine and save space.
For a better emulation, the flags after the LDI/LDD instructions should be altered as follows:

S - Sign
Z - Zero flag 
F5 - undocumented flag 
H - Half Carry 
F3 - undocumented flag 
P/V - Parity or Overflow 
N - Subtract 
C - Carry 

SZ5H3VNC
--*0**0-

PV set if BC not 0
F5 is bit 1 of (transferred byte + A)
F3 is bit 3 of (transferred byte + A)

EDIT: the only useful flag in my opinion is P/V when checked after LDI or LDD, but I think we can live without it.

[suborb]

@zx70 yes, I did some magic in z80asm to handle mapping:

• jr cc, label -> jp cc,label
• djnz label -> dec b; jp nz,label
• ldir/ldi/ldd/lddr/cpi/cpir/cpd/cpdr -> __z80_asm_XXXX

The jumps work well as long as they don't reference ASMPC, and the emulation is the same as we did for the Rabbit. So there's an ldir implementation here: https://github.com/z88dk/z88dk/blob/feature/8080/src/z80asm/dev/z80asm_lib/ldir.asm

[suborb]

Comment to indicate the things that have been removed so I can document them later:

• adt: All
• alloc: all allocators apart from malloc_classic
• strings: memops, memopi, memopd
• stdlib: _Number conversion routines_, qsort, bsearch, extract_bits, inp, outp,
• sccz80: builtin functions
• stdio: funopen() and device support, scanf,

SDCC support

The only critical feature removed is number conversion routines. I'll see if I can resurrect some versions from the grave/adapt the newlib ones without too much pain.

[suborb]

This also opens up the idea of 8085 support. Some of the undocumented opcodes look interesting:

• DSUB - 16 bit subtract HL - BC
• ARHL - Arithmetic Right shift HL
• RDEL - Rotate (shift) DE Left into carry, fill in zero
• LDHI - Load DE with HL plus Immediat8 (HL+i8->DE)
• LDSI - Load DE with SP plus Immediat8 (SP+i8->DE)
• SHLX - Store HL indeXed by DE [DE]=HL
• LHLX - Load HL indeXed by DE HL=[DE]

SHLX is basically l_pint for example. Information here: http://electronicerror.blogspot.com/2007/08/undocumented-flags-and-instructions.html

[zx70]

That's a cute one. I'd really like to try to add support for the Kyocera family handheld computers (aka Tandy 100/200)

[suborb]

Yeah, I think this and any 8085 enhancements will open an entire new tranche of targets. Getting to 100 z80 targets feels like it's a long way off.

[zx70]

we are totally crazy :D

[suborb]

Let’s call this one done for now.

[suborb]

Ken Yap (the very first version of z88dk was based off his zcc096.zip package), has a wip 8080 port of sdcc in this repo: https://github.com/kenyapcomau/sdcc-8080