X-Git-Url: https://git.rrq.au/?a=blobdiff_plain;f=main.asm;h=d2cd54deb29e91c65f5d366d344aba29911eba39;hb=02f53b20b4f7244a84a442bbae10eb8401abcfc5;hp=6356be8adb005a3798d95a420783fc0cd3009046;hpb=59d23425b8efe3004f6452e1fa670582b469af2a;p=rrq%2Fjonasforth.git diff --git a/main.asm b/main.asm index 6356be8..d2cd54d 100644 --- a/main.asm +++ b/main.asm @@ -23,54 +23,120 @@ macro popr x { add rbp, 8 } +;; The following macro generates the dictionary header. It updates the +;; initial_latest_entry variable, which is used as the initial value of the +;; latest_entry variable that is made available at runtime. +;; +;; The header contains a link to the previous entry, the length of the name of +;; the word and the word itself as a string literal. +;; +;; This macro also defines a label LABEL_entry. +initial_latest_entry = 0 +macro header label, name, immediate { + local .string_end + +label#_entry: + dq initial_latest_entry + if immediate eq + db 0 + else + db 1 + end if + db .string_end - ($ + 1) + db name + .string_end: +label: + +initial_latest_entry = label#_entry +} + +;; Define a Forth word that is implemented in assembly. See 'header' for details. +macro forth_asm label, name, immediate { + header label, name, immediate + dq .start +.start: +} + +;; Define a Forth word that is implemented in Forth. (The body will be a list of +;; 'dq' statements.) +macro forth label, name, immediate { + header label, name, immediate + dq DOCOL +} + segment readable executable +entry main + +include "impl.asm" + main: cld ; Clear direction flag so LODSQ does the right thing. mov rbp, return_stack_top ; Initialize return stack - mov rsi, program - next + mov rax, MAIN + jmp qword [rax] program: dq MAIN ;; The codeword is the code that will be executed at the beginning of a forth ;; word. It needs to save the old RSI and update it to point to the next word to ;; execute. -docol: +header DOCOL, 'DOCOL' pushr rsi ; Save old value of RSI on return stack; we will continue execution there after we are done executing this word lea rsi, [rax + 8] ; RAX currently points to the address of the codeword, so we want to continue at RAX+8 next ; Execute word pointed to by RSI ;; This word is called at the end of a Forth definition. It just needs to -;; restore the old value of RSI (saved by 'docol') and resume execution. -EXIT: - dq .start -.start: +;; restore the old value of RSI (saved by 'DOCOL') and resume execution. +forth_asm EXIT, 'EXIT' popr rsi next ;; LIT is a special word that reads the next "word pointer" and causes it to be ;; placed on the stack rather than executed. -LIT: - dq .start -.start: +forth_asm LIT, 'LIT' lodsq push rax next +;; Given a string (a pointer following by a size), return the location of the +;; dictionary entry for that word. If no such word exists, return 0. +forth_asm FIND, 'FIND' + mov [.rsi], rsi + + pop [find.search_length] + pop [find.search_buffer] + mov rsi, [latest_entry] ; Start with the last added word + call find + push rsi + + mov rsi, [.rsi] + next + push rsi + + mov rsi, [.rsi] + next + +;; Given an entry in the dictionary, return a pointer to the codeword of that +;; entry. +forth_asm TCFA, '>CFA' + pop rax + add rax, 8 + 1 ; [rax] = length of name + movzx rbx, byte [rax] + inc rax + add rax, rbx ; [rax] = codeword + push rax + next + ;; BRANCH is the fundamental mechanism for branching. BRANCH reads the next word ;; as a signed integer literal and jumps by that offset. -BRANCH: - dq .start -.start: +forth_asm BRANCH, 'BRANCH' add rsi, [rsi] ; [RSI], which is the next word, contains the offset; we add this to the instruction pointer. next ; Then, we can just continue execution as normal ;; 0BRANCH is like BRANCH, but it jumps only if the top of the stack is zero. -ZBRANCH: - dq .start -.start: +forth_asm ZBRANCH, '0BRANCH' ;; Compare top of stack to see if we should branch pop rax cmp rax, 0 @@ -81,10 +147,18 @@ ZBRANCH: add rsi, 8 ; We need to skip over the next word, which contains the offset. next +;; Duplicate the top of the stack. +forth_asm DUP_, 'DUP' + push qword [rsp] + next + +;; Execute the codeword at the given address. +forth_asm EXEC, 'EXEC' + pop rax + jmp qword [rax] + ;; Expects a character on the stack and prints it to standard output. -EMIT: - dq .start -.start: +forth_asm EMIT, 'EMIT' pushr rsi pushr rax mov rax, 1 @@ -98,71 +172,50 @@ EMIT: next ;; Prints a newline to standard output. -NEWLINE: - dq docol +forth NEWLINE, 'NEWLINE' dq LIT, $A dq EMIT dq EXIT +;; Prints a space to standard output. +forth SPACE, 'SPACE' + dq LIT, ' ' + dq EMIT + dq EXIT + ;; Read a word from standard input and push it onto the stack as a pointer and a ;; size. The pointer is valid until the next call to READ_WORD. -READ_WORD: ; 400170 - dq .start -.start: +forth_asm READ_WORD, 'READ-WORD' mov [.rsi], rsi - mov [.rax], rax - -.skip_whitespace: - ;; Read characters into .char_buffer until one of them is not whitespace. - mov rax, 0 - mov rdi, 0 - mov rsi, .char_buffer - mov rdx, 1 - syscall - - cmp [.char_buffer], ' ' - je .skip_whitespace - cmp [.char_buffer], $A - je .skip_whitespace - -.alpha: - ;; We got a character that wasn't whitespace. Now read the actual word. - mov [.length], 0 -.read_alpha: - mov al, [.char_buffer] - movzx rbx, [.length] - mov rsi, .buffer - add rsi, rbx - mov [rsi], al - inc [.length] + call read_word + push rdi ; Buffer + push rdx ; Length - mov rax, 0 - mov rdi, 0 - mov rsi, .char_buffer - mov rdx, 1 - syscall + mov rsi, [.rsi] + next - cmp [.char_buffer], ' ' - je .end - cmp [.char_buffer], $A - jne .read_alpha +;; Takes a string on the stack and replaces it with the decimal number that the +;; string represents. +forth_asm PARSE_NUMBER, 'PARSE-NUMBER' + pop [parse_number.length] ; Length + pop [parse_number.buffer] ; String pointer -.end: - push .buffer - movzx rax, [.length] - push rax - - mov rsi, [.rsi] - mov rax, [.rax] + push rsi + call parse_number + pop rsi + push rax ; Result next +forth READ_NUMBER, 'READ-NUMBER' + dq READ_WORD + dq PARSE_NUMBER + dq EXIT + ;; Takes a string (in the form of a pointer and a length on the stack) and ;; prints it to standard output. -TELL: - dq .start -.start: +forth_asm TELL, 'TELL' mov rbx, rsi mov rcx, rax @@ -177,32 +230,12 @@ TELL: next ;; Exit the program cleanly. -TERMINATE: - dq .start -.start: +forth_asm TERMINATE, 'TERMINATE' mov rax, $3C mov rdi, 0 syscall -PUSH_HELLO_CHARS: - dq docol - dq LIT, $A - dq LIT, 'o' - dq LIT, 'l' - dq LIT, 'l' - dq LIT, 'e' - dq LIT, 'H' - dq EXIT - -PUSH_YOU_TYPED: - dq .start -.start: - push you_typed_string - push you_typed_string.length - next - -HELLO: - dq docol +forth HELLO, 'HELLO' dq LIT, 'H', EMIT dq LIT, 'e', EMIT dq LIT, 'l', EMIT @@ -212,11 +245,92 @@ HELLO: dq NEWLINE dq EXIT +;; Duplicate a pair of elements. +forth_asm PAIRDUP, '2DUP' + pop rbx + pop rax + push rax + push rbx + push rax + push rbx + next + +;; Swap the top two elements on the stack. +forth_asm SWAP, 'SWAP' + pop rax + pop rbx + push rax + push rbx + next + +;; Remove the top element from the stack. +forth_asm DROP, 'DROP' + add rsp, 8 + next + +forth_asm NOT_, 'NOT' + pop rax + cmp rax, 0 + jz .false +.true: + push 0 + next +.false: + push 1 + next + +;; The INTERPRET word reads and interprets user input. It's behavior depends on +;; the current STATE. It provides special handling for integers. +forth INTERPRET, 'INTERPRET' + ;; Read word + dq READ_WORD + dq PAIRDUP + ;; Stack is (word length word length). + dq FIND ; Try to find word + dq DUP_ + dq ZBRANCH, 8 * 20 ; Check if word is found + + ;; - Word is found - + + dq STATE, GET, ZBRANCH, 8 * 9 ; Check whether we are in compilation or immediate mode + + ;; (Word found, compilation mode) + dq DUP_, IS_IMMEDIATE, NOT_, ZBRANCH, 8 * 4 ; If the word is immediate, continue as we would in immediate mode + + ;; Otherwise, we want to compile this word + dq TCFA + dq COMMA + dq EXIT + + ;; (Word found, immediate mode) + ;; Execute word + dq TCFA + ;; Stack is (word length addr) + dq SWAP, DROP + dq SWAP, DROP + ;; Stack is (addr) + dq EXEC + dq EXIT + + ;; - No word is found, assume it is an integer literal - + ;; Stack is (word length addr) + dq DROP + dq PARSE_NUMBER + + dq STATE, GET, ZBRANCH, 8 * 5 ; Check whether we are in compilation or immediate mode + + ;; (Number, compilation mode) + dq LIT, LIT, COMMA + dq COMMA + dq EXIT + + ;; (Number, immediate mode) + dq EXIT + ;; .U prints the value on the stack as an unsigned integer in hexadecimal. -DOTU: - dq .start -.start: +forth_asm DOTU, '.U' mov [.length], 0 + mov [.printed_length], 1 pop rax ; RAX = value to print push rsi ; Save value of RSI @@ -274,25 +388,200 @@ DOTU: pop rsi next -MAIN: - dq docol +;; Takes a value and an address, and stores the value at the given address. +forth_asm PUT, '!' + pop rbx ; Address + pop rax ; Value + mov [rbx], rax + next + +;; Takes an address and returns the value at the given address. +forth_asm GET, '@' + pop rax + mov rax, [rax] + push rax + next + +forth_asm PUT_BYTE, 'C!' + pop rbx + pop rax ; Value + mov [rbx], al + next + +forth_asm GET_BYTE, 'C@' + pop rax + movzx rax, byte [rax] + push rax + next + +;; Add two integers on the stack. +forth_asm PLUS, '+' + pop rax + pop rbx + add rax, rbx + push rax + next + +;; Calculate difference between two integers on the stack. The second number is +;; subtracted from the first. +forth_asm MINUS, '-' + pop rax + pop rbx + sub rbx, rax + push rbx + next + +;; Given two integers a and b on the stack, pushes the quotient and remainder of +;; division of a by b. +forth_asm TIMESMOD, '/MOD' + pop rbx ; b + pop rax ; a + mov rdx, 0 + div rbx + push rax ; a / b + push rdx ; a % b + next + +;; Get the location of the STATE variable. It can be set with '!' and read with +;; '@'. +forth STATE, 'STATE' + dq LIT, var_STATE + dq EXIT + +;; Get the location of the LATEST variable. It can be set with '!' and read with +;; '@'. +forth LATEST, 'LATEST' + dq LIT, latest_entry + dq EXIT + +;; Get the location at which compiled words are expected to be added. This +;; pointer is usually modified automatically when calling ',', but we can also +;; read it manually with 'HERE'. +forth HERE, 'HERE' + dq LIT, here + dq EXIT + +forth COMMA, ',' + dq HERE, GET, PUT ; Set the memory at the address pointed to by HERE + dq HERE, GET, LIT, 8, PLUS ; Calculate new address for HERE to point to + dq HERE, PUT ; Update HERE to point to the new address + dq EXIT + +;; Read user input until next " character is found. Push a string containing the +;; input on the stack as (buffer length). Note that the buffer is only valid +;; until the next call to S" and that no more than 255 character can be read. +forth_asm READ_STRING, 'S"' + push rsi + + mov [.length], 0 + +.read_char: + mov rax, 0 + mov rdi, 0 + mov rsi, .char_buffer + mov rdx, 1 + syscall + + mov al, [.char_buffer] + cmp al, '"' + je .done + + mov rdx, .buffer + add rdx, [.length] + mov [rdx], al + inc [.length] + jmp .read_char + +.done: + pop rsi + + push .buffer + push [.length] + + next + +;; CREATE inserts a new header in the dictionary, and updates LATEST so that it +;; points to the header. To compile a word, the user can then call ',' to +;; continue to append data after the header. +;; +;; It takes the name of the word as a string (address length) on the stack. +forth_asm CREATE, 'CREATE' + pop rcx ; Word string length + pop rdx ; Word string pointer + + mov rdi, [here] ; rdi = Address at which to insert this entry + mov rax, [latest_entry] ; rax = Address of the previous entry + mov [rdi], rax ; Insert link to previous entry + mov [latest_entry], rdi ; Update LATEST to point to this word + + add rdi, 8 + mov [rdi], byte 0 ; Insert immediate flag + + add rdi, 1 + mov [rdi], byte cl ; Insert length + + ;; Insert word string + add rdi, 1 + + push rsi + mov rsi, rdx ; rsi = Word string pointer + rep movsb + pop rsi + + ;; Update HERE + mov [here], rdi + + next + +;; Mark the last added word as immediate. +forth IMMEDIATE, 'IMMEDIATE', 1 + dq LIT, 1 + dq LATEST, GET + dq LIT, 8, PLUS + dq PUT_BYTE + dq EXIT + +;; Given the address of a word, return 0 if the given word is not immediate. +forth IS_IMMEDIATE, 'IMMEDIATE?' + dq LIT, 8, PLUS + dq GET_BYTE + dq EXIT + +;; Enter immediate mode, immediately +forth INTO_IMMEDIATE, '[', 1 + dq LIT, 0, STATE, PUT_BYTE + dq EXIT + +;; Enter compilation mode +forth OUTOF_IMMEDIATE, ']' + dq LIT, 1, STATE, PUT_BYTE + dq EXIT + +forth MAIN, 'MAIN' dq HELLO - dq LIT, 1234567890, DOTU, NEWLINE - dq LIT, $ABCD, DOTU, NEWLINE - dq LIT, $1234ABCD5678EFAB, DOTU, NEWLINE + dq INTERPRET + dq BRANCH, -8 * 2 dq TERMINATE segment readable writable -you_typed_string db 'You typed: ' -.length = $ - you_typed_string +;; The LATEST variable holds a pointer to the word that was last added to the +;; dictionary. This pointer is updated as new words are added, and its value is +;; used by FIND to look up words. +latest_entry dq initial_latest_entry + +;; The STATE variable is 0 when the interpreter is executing, and non-zero when +;; it is compiling. +var_STATE dq 0 + +FIND.rsi dq ? READ_WORD.rsi dq ? -READ_WORD.rax dq ? -READ_WORD.max_size = $FF -READ_WORD.buffer rb READ_WORD.max_size -READ_WORD.length db ? -READ_WORD.char_buffer db ? +READ_WORD.rbp dq ? + +READ_STRING.char_buffer db ? +READ_STRING.buffer rb $FF +READ_STRING.length dq ? DOTU.chars db '0123456789ABCDEF' DOTU.buffer rq 16 ; 64-bit number has no more than 16 digits in hex @@ -300,6 +589,10 @@ DOTU.rbuffer rq 16 DOTU.length dq ? DOTU.printed_length dq ? +;; Reserve space for compiled words, accessed through HERE. +here dq here_top +here_top rq $2000 + ;; Return stack rq $2000 return_stack_top: