(load "typecheck.scm")
(load "ast.scm")
+(load "platform.scm")
+
+(define target host-os)
(define (emit . s)
(begin
(codegen-expr a si env)
(emit "movq %rax, ~a(%rbp)" si)
(codegen-expr b (- si wordsize) env)
- (emit "subq ~a(%rbp), %rax" si)
- (emit "not %rax")
- (emit "andq $1, %rax"))
+ (emit "## ~a = ~b" a b)
+ (emit "cmpq ~a(%rbp), %rax" si)
+ (emit "sete %al"))
+
+ ; 'write file handle addr-string num-bytes
(define (codegen-print x si env)
(codegen-expr x si env) ; x should be a static-string, producing a label
(emit "not %rcx") ; -%rcx = strlen + 1
(emit "dec %rcx")
- (emit "mov %rcx, %rdx") ; number of bytes
- (emit "mov %rbx, %rsi") ; addr of string
- (emit "mov $1, %rax") ; file handle 1 (stdout)
- (emit "mov $1, %rdi") ; syscall 1 (write)
+ (emit "movq %rbx, %rsi") ; string addr
+ (emit "movq %rcx, %rdx") ; num bytes
+ (emit "movq $1, %rdi") ; file handle (stdout)
+ (case target
+ ('darwin (emit "mov $0x2000004, %rax")) ; syscall 4 (write)
+ ('linux (emit "mov $1, %rax"))) ; syscall 1 (write)
(emit "syscall"))
(define (range s n)
(define wordsize 8)
(define (codegen-let bindings body si env)
- (let* ((stack-offsets (map (lambda (x) (- si (* x wordsize)))
+
+ ; is this a closure that captures itself?
+ ; e.g. (let ([x 3] [f (closure lambda0 (f x))]) (f))
+ (define (self-captive-closure? name expr)
+ (and (eqv? (ast-type expr) 'closure)
+ (memv name (caddr expr))))
+
+ (let* ((stack-offsets (map (lambda (name x) ; assoc map of binding name to offset
+ (cons name (- si (* x wordsize))))
+ (map car bindings)
(range 0 (length bindings))))
(inner-si (- si (* (length bindings) wordsize)))
- (names (map car bindings))
- (exprs (map cadr bindings))
-
- ; recursive let bindings: build environment as we go
- (inner-env (fold-left
- (lambda (env name expr offset)
- (codegen-expr expr inner-si env)
- (emit "movq %rax, ~a(%rbp)" offset)
- (cons (cons name offset) env))
- env names exprs stack-offsets)))
+
+ (get-offset (lambda (n) (cdr (assoc n stack-offsets))))
+
+ [inner-env
+ (fold-left
+ (lambda (env comps)
+ (let ([scc-env
+ (fold-left
+ (lambda (acc name)
+ (cons (cons name (get-offset name))
+ acc))
+ env
+ comps)])
+ (for-each
+ (lambda (name)
+ (let ([expr (cadr (assoc name bindings))])
+ (emit "## generating ~a with scc-env ~a" name scc-env)
+ (if (self-captive-closure? name expr)
+ ; if self-captive, insert a flag into the environment to let
+ ; codegen-closure realise this!
+ (codegen-expr expr
+ inner-si
+ (cons (cons name 'self-captive)
+ scc-env))
+ (codegen-expr expr inner-si scc-env))
+ (emit "movq %rax, ~a(%rbp)" (get-offset name))))
+ comps)
+ scc-env))
+ env (reverse (sccs (graph bindings))))])
+
(for-each (lambda (form)
(codegen-expr form inner-si inner-env))
body)))
(format "_lambda~a" (- cur-lambda 1)))
; a closure on the heap looks like:
-; 0-x x+0 x+4 x+12 x+20
-; label #vars var1.... var2.... var3....
+ ; 0 8 16 24
+ ; addr var1.... var2.... var3....
+
(define (codegen-closure label captured si env)
- (let* ((heap-offsets (range 4 (length captured))) ; 4, 12, 20, etc.
- (inner-si (- si (* (length captured) wordsize))))
- (emit "movl $~a, (heap_start)")
- (emit "add $4, (heap_start)")
- (for-each (lambda (var-name new-offset)
- (emit "movq ~a(%rbp), ~a(heap_start)" ; todo: do we need to copy this?
- (cdr (assoc var-name env))
- new-offset)
- (emit "add $8, (heap_start)")
+ (let* ((heap-offsets (map (lambda (i) (+ 8 (* 8 i)))
+ (range 0 (length captured))))) ; 4, 12, 20, etc.
+
+ (emit "## creating closure")
+
+ (emit "movq heap_start@GOTPCREL(%rip), %rbx")
+
+ (emit "movq (%rbx), %rax") ; %rax = heap addr of closure
+
+
+ ; point heap_start to next space
+ (emit "addq $~a, (%rbx)" (+ 8 (* 8 (length captured))))
+
+ (emit "## storing address to lambda")
+ ; store the address to the lambda code
+ (emit "movq ~a@GOTPCREL(%rip), %rbx" label)
+ (emit "movq %rbx, 0(%rax)")
+
+ (emit "## storing captives")
+ ; store the captured vars
+ (for-each
+ (lambda (var-name heap-offset)
+ (let ([stack-offset (cdr (assoc var-name env))])
+ (emit "### captive ~a" var-name)
+ (if (eqv? stack-offset 'self-captive)
+ ; captive refers to this closure:
+ ; move heap addr of this closure to stack!
+ (emit "movq %rax, ~a(%rax)" heap-offset)
+ (begin
+ (emit "movq ~a(%rbp), %rbx" stack-offset)
+ (emit "movq %rbx, ~a(%rax)" heap-offset)))))
captured
- stack-offsets)
-)
+ heap-offsets)))
+
; for now we can only call closures
-(define (codegen-call closure args si env)
-; (codegen-expr closure si env)
- (when (not (eq? (ast-type closure) 'closure))
- (error #f (format "~a is not a closure" closure)))
- (let* ((captured (caddr closure))
- (label (cadr closure))
- (argument-start (length captured)))
-
- ; first move the captured variables into param registers
- (for-each
- (lambda (e i)
- (emit "movq ~a(%rbp), ~a"
- (cdr (assoc e env)) ; offset of the var
- (param-register i)))
- captured (range 0 (length captured)))
+(define (codegen-call f args si env)
+ (codegen-expr f si env)
+
+ (emit "## starting call")
+
+ (emit "movq %rax, ~a(%rbp)" si) ; store address of closure first on stack
- ; then codegen the arguments and move them into the next param registers
+ ; codegen the arguments, store them intermediately
(for-each
(lambda (e i)
(begin
- (codegen-expr e si env)
- ; move result to correct param register
- (emit "movq %rax, ~a" (param-register i))))
- args (range argument-start (length args)))
+ (emit "## arg no. ~a" (- i 1))
+ (codegen-expr e (- si (* wordsize i)) env)
+ ; store intermediate result on stack
+ (emit "movq %rax, ~a(%rbp)" (- si (* wordsize i)))))
+
+ args (range 1 (length args)))
+
+ ; now that we have everything we need on the stack,
+ ; move them into the param registers
+
+ (emit "## moving args into place")
+ (for-each
+ (lambda (i) (emit "movq ~a(%rbp), ~a"
+ (- si (* wordsize i))
+ (param-register i)))
+ (range 1 (length args)))
+
+ ; todo: can this be made more efficient
+ (emit "movq ~a(%rbp), %rax" si) ; load back pointer to closure
+
+ (emit "## moving captives into place")
+
+ ; move captives into first argument
+ (emit "movq %rax, %rbx")
+ (emit "addq $8, %rbx")
+ (emit "movq %rbx, ~a" (param-register 0))
+
+ (emit "## performing call")
(emit "addq $~a, %rsp" si) ; adjust the stack pointer to account all the stuff we put in the env
- (emit "callq ~a" label)
- (emit "subq $~a, %rsp" si)))
+ (emit "callq *(%rax)") ; call closure function
+ (emit "subq $~a, %rsp" si))
+
+ ; LAMBDAS:
+ ; 1st param: pointer to captured args
+ ; 2nd param: 1st arg
+ ; 3rd param: 2nd arg, etc.
(define (codegen-lambda l)
(let* ((label (car l))
- (args (cadr l))
- (captured (caddr l))
- (body (cadddr l))
+ (stuff (cdr l))
+ (captives (car stuff))
+ (args (cadr stuff))
+ (body (caddr stuff))
; params = what actually gets passed
- (params (append captured args))
+ (params (append captives args))
- (param-registers (map param-register
- (range 0 (length params))))
(stack-offsets (map (lambda (i)
- (* (- wordsize) i))
- (range 1 (length params))))
-
- (copy-insts (map (lambda (r o)
- (format "movq ~a, ~a(%rbp)" r o))
- param-registers stack-offsets))
+ (* (- wordsize) (+ 1 i)))
+ (range 0 (length params))))
(env (map cons params stack-offsets)))
(emit "~a:" label)
+
+ (display "## lambda captives: ")
+ (display captives)
+ (newline)
+ (display "## lambda args: ")
+ (display args)
+ (newline)
(display "## lambda body: ")
(display body)
(newline)
- (display "## environment: ")
- (display env)
- (newline)
(emit "push %rbp") ; preserve caller's base pointer
+
(emit "movq %rsp, %rbp") ; set up our own base pointer
- (for-each emit copy-insts)
+ ; load the captured vars onto the stack
+ (for-each
+ (lambda (i)
+ (begin
+ (emit "# loading captive ~a" (list-ref captives i))
+ (emit "movq ~a(~a), %rbx" (* wordsize i) (param-register 0))
+ (emit "movq %rbx, ~a(%rbp)" (* (- wordsize) (+ 1 i)))))
+ (range 0 (length captives)))
+
+ ; load the args onto the stack
+ (for-each
+ (lambda (i)
+ (begin
+ (emit "movq ~a, %rbx" (param-register (+ 1 i)))
+ (emit "movq %rbx, ~a(%rbp)"
+ (* (- wordsize)
+ (+ 1 (length captives) i)))))
+ (range 0 (length args)))
+
(codegen-expr body (* (- wordsize) (+ 1 (length params))) env)
(emit "pop %rbp") ; restore caller's base pointer
(emit "~a:" exit-label)))
(define (codegen-expr e si env)
+ (emit "# ~a" e)
(case (ast-type e)
- ('builtin e)
('closure (codegen-closure (cadr e) (caddr e) si env))
('app
- (let ((callee (codegen-expr (car e) si env)))
- (case callee
+ (case (car e)
('+ (codegen-add (cdr e) si env))
('- (codegen-sub (cadr e) (caddr e) si env))
('* (codegen-mul (cadr e) (caddr e) si env))
('= (codegen-eq (cadr e) (caddr e) si env))
('bool->int (codegen-expr (cadr e) si env))
('print (codegen-print (cadr e) si env))
- (else (codegen-call callee (cdr e) si env)))))
+ (else (codegen-call (car e) (cdr e) si env))))
+
+ ; this is a builtin being passed around as a variable
+ ('builtin (emit "movq $~a, %rax" (builtin-id e)))
('let (codegen-let (let-bindings e)
(let-body e)
('bool-literal (emit "movq $~a, %rax" (if e 1 0)))
('int-literal (emit "movq $~a, %rax" e))
- ('static-string (emit "lea ~a, %rax" (cadr e))) ; move label
+ ('static-string (emit "movq ~a@GOTPCREL(%rip), %rax"
+ (cadr e)))
(else (error #f "don't know how to codegen this"))))
-(define (fold-map f x) (fold-left append '() (map f x)))
(define (free-vars prog)
(define bound '())
(define (collect e)
(case (ast-type e)
('builtin '()) ; do nothing
- ('var (if (memq e bound) '() (list e)))
+ ('var (if (memv e bound) '() (list e)))
('lambda
+ (begin
(set! bound (append (lambda-args e) bound))
- (collect (lambda-body e)))
+ (collect (lambda-body e))))
- ('app (fold-map collect e))
+ ('app (flat-map collect e))
+ ('if (flat-map collect (cdr e)))
('let
- (let ((bind-fvs (fold-map (lambda (a)
- ((set! bound (cons (car a) bound))
+ (let ([bind-fvs (flat-map (lambda (a)
+ (begin
+ (set! bound (cons (car a) bound))
(collect (cdr a))))
- (let-bindings cadr)))
- (body-fvs (fold-map collect (let-body e))))
- (append bind-fvs body-fvs)))
+ (let-bindings e))])
+ (append bind-fvs (flat-map collect (let-body e)))))
(else '())))
(collect prog))
- ; ((lambda (x) (+ x 1)) 42) => {lambda0: (x) (+ x 1)}, (@lambda0 42)
+ ; ((lambda (x) (+ x y)) 42) => ((closure lambda1 (y)) 42)
+ ; [(lambda1 . ((y), (x), (+ x y))]
+ ; for builtins, this generates a closure if it is used
+ ; outside of an immediate app
+ ; but only one closure for each builtin
+
(define (extract-lambdas program)
(define lambdas '())
(define (add-lambda e)
(args (lambda-args e))
(captured (free-vars e))
(body (extract (lambda-body e)))
- (new-lambda (list label args captured body)))
+ (new-lambda (cons label (list captured args body))))
(set! lambdas (cons new-lambda lambdas))
`(closure ,label ,captured))) ; todo: should we string->symbol?
+
+ (define (find-builtin-lambda e)
+ (let [(l (assq (builtin-name e) lambdas))]
+ (if l `(closure ,(car l) ,(caadr l)) #f)))
+
+ (define (builtin-name e)
+ (case e
+ ('+ "_add")
+ ('- "_sub")
+ ('* "_mul")
+ ('! "_not")
+ ('= "_eq")
+ ('bool->int "_bool2int")
+ ('print "_print")
+ (else (error #f "don't know this builtin"))))
+ (define (builtin-args e)
+ (case e
+ ('+ '(x y))
+ ('- '(x y))
+ ('* '(x y))
+ ('! '(x))
+ ('= '(x y))
+ ('bool->int '(x))
+ ('print '(x))
+ (else (error #f "don't know this builtin"))))
+
+ (define (add-builtin-lambda e)
+ (let* [(label (builtin-name e))
+ (captured '())
+ (args (builtin-args e))
+ (body `(,e ,@args))
+ (new-lambda (cons label (list captured args body)))]
+ (set! lambdas (cons new-lambda lambdas))
+ `(closure ,label ,captured)))
+
(define (extract e)
(case (ast-type e)
('lambda (add-lambda e))
- ('let `(let ,(map extract (let-bindings e))
+ ('let `(let ,(map (lambda (b) `(,(car b) ,@(extract (cdr b)))) (let-bindings e))
,@(map extract (let-body e))))
- ('app (append (list (extract (car e)))
+ ('app (append
+ ; if a builtin is used as a function, don't generate lambda
+ (if (eqv? 'builtin (ast-type (car e)))
+ (list (car e))
+ (list (extract (car e))))
(map extract (cdr e))))
+
+ ('builtin
+ (if (find-builtin-lambda e)
+ (find-builtin-lambda e)
+ (add-builtin-lambda e)))
+
+
(else (ast-traverse extract e))))
(let ((transformed (extract program)))
(cons lambdas transformed)))
(else (error #f "need to test out the below"))
(else (format "~a(%rsp)" (- n 6)))))
+(define (initialize-heap)
+ (let ((mmap
+ (case target
+ ('darwin "0x20000c5")
+ ('linux "9"))))
+ ; allocate some heap memory
+ (emit "mov $~a, %rax" mmap) ; mmap
+ (emit "xor %rdi, %rdi") ; addr = null
+ (emit "movq $1024, %rsi") ; length = 1kb
+ (emit "movq $0x3, %rdx") ; prot = read | write = 0x2 | 0x1
+ ; flags = anonymous | private
+ (case target
+ ('darwin (emit "movq $0x1002, %r10")) ; anon = 0x1000, priv = 0x02
+ ('linux (emit "movq $0x22, %r10"))) ; anon = 0x20, priv = 0x02
+ (emit "movq $-1, %r8") ; fd = -1
+ (emit "xor %r9, %r9") ; offset = 0
+ (emit "syscall")
+ ; %rax now contains pointer to the start of the heap
+ ; keep track of it
+
+ (emit "movq heap_start@GOTPCREL(%rip), %rsi")
+ (emit "movq %rax, (%rsi)")))
+
(define (codegen program)
- (let* ((extract-res-0 (extract-strings program))
+ (set! cur-label 0)
+ (set! cur-lambda 0)
+ (let* ([body (program-body program)]
+
+ (extract-res-0 (extract-strings body))
(strings (car extract-res-0))
(extract-res-1 (extract-lambdas (cdr extract-res-0)))
(lambdas (car extract-res-1))
(emit "_start:")
- ; allocate some heap memory
- (emit "mov $9, %rax") ; mmap
- (emit "xor %rdi, %rdi") ; addr = null
- (emit "movq $1024, %rsi") ; length = 1kb
- (emit "movq $0x3, %rdx") ; prot = read | write = 0x2 | 0x1
- (emit "movq $0x22, %r10") ; flags = anonymous | private = 0x20 | 0x02
- (emit "movq $-1, %r8") ; fd = -1
- (emit "xor %r9, %r9") ; offset = 0
- (emit "syscall")
-
- ; %rax now contains pointer to the start of the heap
- ; keep track of it
- (emit "movq %rax, (heap_start)")
+ (initialize-heap)
(emit "movq %rsp, %rbp") ; set up the base pointer
- (codegen-expr xform-prog 0 '())
+
+ (codegen-expr xform-prog (- wordsize) '())
; exit syscall
(emit "mov %rax, %rdi")
- (emit "mov $60, %rax")
+ (case target
+ ('darwin (emit "movq $0x2000001, %rax"))
+ ('linux (emit "mov $60, %rax")))
(emit "syscall")
(emit ".data")
(for-each emit-string-data strings)))
-(define (compile-to-binary program output)
- (when (not (eq? (typecheck program) 'int)) (error #f "not an int"))
+(define (compile-to-binary program output t)
+ (set! target t)
+ (when (not (eq? (typecheck program) 'Int)) (error #f "not an Int"))
(let ([tmp-path "/tmp/a.s"])
(when (file-exists? tmp-path) (delete-file tmp-path))
(with-output-to-file tmp-path
(lambda () (codegen program)))
- (system (format "clang -nostdlib /tmp/a.s -o ~a" output))))
+
+ (case target
+ ('darwin
+ (system "as /tmp/a.s -o /tmp/a.o")
+ (system (format "ld /tmp/a.o -e _start -macosx_version_min 10.14 -static -o ~a" output)))
+ ('linux
+ (system "as /tmp/a.s -o /tmp/a.o")
+ (system (format "ld /tmp/a.o -o ~a" output))))))
; NOTES
-; syscalls in linux use the following arguments for syscall instruction:
+; syscalls in linux and darwin use the following arguments for syscall instruction:
; %rax = syscall #
; %rdi = 1st arg
; %rsi = 2nd arg
; %r10 = 4th arg
; %r8 = 5th arg
; %r9 = 6th arg
+
+; on darwin, the syscall is offset by 0x2000000
+; https://opensource.apple.com/source/xnu/xnu-2782.20.48/bsd/kern/syscalls.master
+; documentation for most syscalls: /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/usr/include/sys