veltas / zenv Goto Github PK

View Code? Open in Web Editor NEW

29.0 3.0 1.0 494 KB

Forth for the ZX Spectrum

License: MIT License

Makefile 0.93% Assembly 97.28% Vim Script 1.80%

forth z80 zx-spectrum retro

zenv's Introduction

zenv - Forth for the ZX Spectrum

This project is a Forth environment for the ZX Spectrum.

Community

IRC channel #zenv on OFTC.

Design

zenv implements most of the ANS Forth core words and some extension/alternative word sets.

zenv can be built with direct threading and tokenised direct threading (see tokenised variable in src/zenv.asm).

zenv's People

Contributors

Stargazers

Watchers

Forkers

gutpuncher

zenv's Issues

Using EX (SP), HL in "does_code:", "constant_code:", "execute:", "um_slash_mod:"

does_code:
POP AF
EX (SP), HL
PUSH AF

…
constant_code:
EX (SP), HL
LD E, (HL)
INC HL
LD D, (HL)
EX DE, HL
JP next
…
execute:
EX (SP), HL
RET

…
um_slash_mod:
…
.skip2:
DEC A
JP NZ, .loop
; IY = quotient
; HL = remainder
EX (SP), IY
EX (SP),HL
JP next

`256. 256 256 UM*/` gives `4294967295.`

Small bug in control.asm: wrong label "j:". It must be "_j:".

`<const> ?` just says `ok` if `?` is undefined

Weirdly, <const> ? just says ok if ? is undefined, when it should say ??. In other situations ? rightly results in ??. There is probably something wrong with the interpreter or SFIND.

Optimize "2OVER" with "2@"

HEADER two_over, "2OVER", 0

two_over:
PUSH HL
LD HL, 4
ADD HL, SP
JR two_fetch

Optimize "DU<" with "U<"

Part of the "U<" code can be used in "DU<".
` HEADER du_less_than, "DU<", 0
du_less_than:
LD C, L
LD B, H
POP DE
POP HL
OR A
SBC HL, BC
JR Z, u_less_than.equal
JR common_less_than.not_equal

HEADER u_less_than, "U<", 0

u_less_than:
EX DE, HL
.equal:
POP HL
common_less_than:
OR A
SBC HL, DE
.not_equal:
LD HL, 0
JR NC, .exit
DEC HL
.exit: JP next
`

`; \ -1 if s1<s2, 1 if s1>s2, 0 if s1=s2
; CODE COMPARE ( a1 u1 a2 u2 -- n )
HEADER compare, "COMPARE", 0
compare:
LD C, L
LD B, H
POP DE
POP HL
PUSH HL
XOR A
SBC HL, BC
JR C, .u2_larger
JR Z, .u1_u2_equal
POP BC
INC A; A=1
JR .cont
.u1_u2_equal:
;A=0
INC A
.u2_larger:
DEC A ;A=-1
POP HL
.cont:
POP HL
PUSH AF
; HL = a1, DE = a2, BC = min(u1,u2)
LD A, C
OR B
JR Z, .skip_loop
.loop:
LD A, (DE)
INC DE
CPI
JR C, .s1_larger
JR NZ, .s2_larger
JP PE, .loop
.skip_loop:
POP AF
LD L, A
LD H, 0
JP next

.s1_larger:
LD HL, 1
JR .cont2

.s2_larger:
LD HL, -1
.cont2:
POP DE
JP next
`

Replace "create_code" with "swap"

I noticed that "create_code" and "swap" are equivalent.
`create_code:
POP DE
PUSH HL
EX DE, HL
JP next
...
swap:
EX (SP), HL
JP next

`
So we can replace "CALL create_code" with "CALL swap", and delete the fragment create_code. Also, "DW create_code" should be replaced by "DW swap".

Stack pointer moving: use low byte increment instead increment of word if stack is aligned

PSP and RS are (or can be done) aligned to even-address. Therefore, when fetching the first byte from the stack, we can write "INC R" instead of "INC RR" (R is low register in register pair RR). It does not reduce the length but reduces the time (for undocumented IX\IY too).
Unfortunately it gives a small gain in zenv: PUSH and POP use automatically 2 bytes at once, the threaded code pointer (IY) is unaligned, RS pointer IX uses indexing instead of increment\decrement.
Here are a few places where it can be applied:

colon_code:                       
   ; Push current PC to RS   
   DEC IX                    
   DEC IXL ; from odd-addr to even-addr                       
...
two_over:
   PUSH HL
   LD HL, 4
   ADD HL, SP
   LD E, (HL)
   INC L ; from even-addr to odd-addr
   LD D, (HL)
   INC HL
   LD C, (HL)
   INC L ; from even-addr to odd-addr
...
r_from:              
   PUSH HL      
   LD L, (IX+0) 
   LD H, (IX+1) 
   INC IXL ;  from even-addr to odd-addr       
   INC IX       
   JP next      
...
exit:               
   LD C, (IX+0)
   LD B, (IX+1)
   INC IXL ;  from even-addr to odd-addr       
   INC IX      
...
to_r:             
   DEC IX                    
   DEC IXL ; from odd-addr to even-addr       
...
pick:             
   ADD HL, HL
   ADD HL, SP
   LD E, (HL)
   INC L ; from even-addr to odd-addr

Doubles aren't compiled properly

: T 1. ; T D. gives 65536

Remove "CALL colon_code" and "DX exit" from words that do not return control

Most high-level words have "CALL colon_code" and "DX exit" at the end. But there are words that do not return control. They are "WHAT?", "CWHAT?", "MAIN", "ABORT", "QUIT". They should not have "DX exit" at the bottom. In addition, we need to replace "CALL colon_code" with "CALL pop_pc_next" at the top of this words.

Another register allocation

Why does the system have such a register allocation (HL as top, IY as PC, IX as RS counter)?

Brad Rodriguez. MOVING FORTH recommends another register allocation for Z80. For example, HL is used as a work register:

HL is undoubtedly the most versatile register, and at one time or another it is tempting to use it for each of the Forth virtual registers. However, because of its versatility -- and because it is the only register which can be fetched byte-wise and used in an indirect jump -- HL should be used for W, Forth's all-purpose working register. (CASE STUDY 3: THE Z80)

In the paper the choice of registers is explained. Maybe the another register allocation could be more optimal?

Macro for asm-code inline

I suggest a macro ASM_INLINE ... END_INLINE to include asm-code into a DX-sequence:

	IF tokenised
		MACRO ASM_INLINE
	    		DB high($+2), low($+2)
		ENDM
	ELSE
		MACRO ASM_INLINE
	    		DW $+2	
		ENDM
	ENDIF

	MACRO END_INLINE
	    LD IY,$+7
	    JP next		
	ENDM

This macro can be placed into zenv.asm after MACRO HEADER, for example.

There are small Forth-words (few asm-instructions) but "JP next" is an expensive operation. Including code allows us to improve efficiency (by decreasing clarity, perhaps).
For example, let's use this macro in "DABS" (zenv.asm).

	; ( d -- |d|)    - in original zenv.asm incomplete comment "; ( d -- d)"
	; : DABS
	HEADER dabs, "DABS", 0
dabs:
	CALL colon_code
	; DUP 0< IF
	DX dup
	DX zero_less
	DX if_raw
	DB .then-$-1
         	; 0. 2SWAP D-
	      ASM_INLINE		
         	LD DE,0
         	POP BC
         	PUSH DE
         	PUSH DE
         	PUSH BC
	      END_INLINE
              	;( 0 d ) 		
	      DX d_minus 
	; THEN ;
.then:
	DX exit

pop_pc_next: 
;drop prefix to change POP IY (FD E1) into POP HL (E1) 
pop_tos_next=$+1 
	POP IY  
	; fall through

Add 16K build mode

Add build mode that supports running on a 16K model (or model with broken upper 32K RAM).