(load "typecheck.scm")
(load "ast.scm")
+(load "platform.scm")
+(load "stdlib.scm")
-(define target 'darwin)
+(define target host-os)
(define (emit . s)
(begin
(apply printf s)
(display "\n")))
+(define wordsize 8)
+
+(define (stack-type? dls type)
+ (if (adt? dls type)
+ (> (type-size dls type) wordsize)
+ #f))
+
+(define (adt? data-layouts type) (assoc type data-layouts))
+
+(define (type-size data-layouts type)
+
+ (define (adt-size adt)
+ (let ([sizes
+ (map (lambda (sum)
+ (fold-left (lambda (acc x) (+ acc (type-size data-layouts x)))
+ wordsize ; one word needed to store tag
+ (cdr sum)))
+ (cdr adt))])
+ (apply max sizes)))
+
+ (case type
+ ['Int wordsize]
+ [else
+ (let ([adt (assoc type data-layouts)])
+ (if adt
+ (adt-size adt)
+ (error #f "unknown size" type)))]))
+
+(define (singletons dls)
+ (flat-map
+ (lambda (sums)
+ (fold-left append '()
+ (filter (lambda (x) (= 1 (length x))) sums)))
+ (map cdr dls)))
+
+(define (singleton? dls x)
+ (memv x (singletons dls)))
+
+ ; returns the size of an expression's result in bytes
+(define (expr-size dls e)
+ (if (eqv? (ast-type e) 'stack)
+ (type-size dls (cadr e))
+ wordsize))
+
+(define (on-stack? expr)
+ (case (ast-type expr)
+ ['stack (cadr expr)]
+ [else #f]))
+
+ ; does a movsq for something on the stack
+ ; src points to the start stack index, but not the top of that index
+ ; likewise for dst
+ ; | ... |
+ ; +------+ <-- to here
+ ; | tag0 |
+ ; +------+ <-- src (size = 16)
+ ; | 42 |
+ ; +------+ <-- start copying from here...
+(define (emit-stack-copy src dst size)
+ (let ([size-to-copy (- size wordsize)])
+ (emit "leaq ~a(%rbp), %rsi" (- src size-to-copy))
+ (emit "leaq ~a(%rbp), %rdi" (- dst size-to-copy))
+ (emit "movq $~a, %rcx" (/ size wordsize))
+ (emit "rep movsq")))
+
+
+ ; an environment consists of adt layouts in scope,
+ ; and any bound variables.
+ ; bound variables are an assoc list with their stack offset
+(define make-env list)
+(define env-data-layouts car)
+(define env-bindings cadr)
+(define (env-append-bindings env bindings)
+ (make-env (env-data-layouts env)
+ (append bindings (env-bindings env))))
+
(define (codegen-add xs si env)
(define (go ys)
(if (null? ys)
(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
+ ; make a copy of string address since %rax and %rdi are clobbered
+ (emit "mov %rax, %rbx")
+
; get the length of the null terminated string
(emit "mov %rax, %rdi")
(emit "xor %al, %al") ; set %al to 0
(emit "not %rcx") ; -%rcx = strlen + 1
(emit "dec %rcx")
- (case target
- ('darwin
- (emit "movq %rax, %rsi") ; string addr
+ (emit "movq %rbx, %rsi") ; string addr
(emit "movq %rcx, %rdx") ; num bytes
(emit "movq $1, %rdi") ; file handle (stdout)
- (emit "movq $0x2000004, %rax")) ; syscall 4 (write)
- ('linux
- (emit "mov %rax, %rsi") ; string addr
- (emit "mov %rcx, %rdx") ; num bytes
- (emit "mov $1, %rax") ; file handle (stdout)
- (emit "mov $1, %rdi"))) ; syscall 1 (write)
+ (case target
+ ('darwin (emit "mov $0x2000004, %rax")) ; syscall 4 (write)
+ ('linux (emit "mov $1, %rax"))) ; syscall 1 (write)
(emit "syscall"))
-(define (range s n)
- (if (= 0 n) '()
- (append (range s (- n 1))
- (list (+ s (- n 1))))))
+(define (codegen-let bindings body si env)
-(define wordsize 8)
+ ; 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))))
+
+ ;; (define (emit-scc scc env)
+ ;; ; acc is a pair of the env and list of touchups
+ ;; (define (emit-binding acc binding)
+ ;; (let ([binding-name (car binding)]
+ ;; [binding-body (cadr binding)]
+
+ ;; [other-bindings (filter
+ ;; (lambda (x) (not (eqv? binding-name x)))
+ ;; scc)]
+ ;; [mutually-recursives
+ ;; (filter
+ ;; (lambda (other-binding)
+ ;; (memv other-binding (references binding-body)))
+ ;; other-bindings)]
+
+ ;; [new-touchups (append touchups (cdr acc))])
+
+ ;; ; TODO: assert that the only mutually recursives are closures
+ ;; (for-each
+ ;; (lambda (binding)
+ ;; (when (not (eqv? (ast-type (cadr binding))
+
+ ;; (emit "asdf")
+ ;; (cons new-env new-touchups)
+ ;; ))
+
+ ;; (fold-left emit-binding (cons env '()) scc))))
+ ; assoc map of binding name to size
+
+
+ (define dls (env-data-layouts env))
+
+ (define stack-sizes
+ (map (lambda (binding) (cons (car binding) (expr-size dls (cadr binding))))
+ bindings))
+
+ ; assoc map of binding name to offset
+ (define stack-offsets
+ ; 2 4 2 8 6
+ (let* ([totals ; 2 6 8 16 22
+ (reverse (fold-left (lambda (acc x)
+ (if (null? acc)
+ (list x)
+ (cons (+ x (car acc)) acc)))
+ '()
+ (map cdr stack-sizes)))]
+ ; 0 2 6 8 16
+ [relative-offsets (map - totals (map cdr stack-sizes))]
+ [absolute-offsets (map (lambda (x) (- si x)) relative-offsets)])
+ (map cons (map car stack-sizes) absolute-offsets)))
+
+ (let* (
+ ; the stack index used when codegening binding body and main body
+ ; -> stack ->
+ ; [stack-offsets | inner-si]
+ [inner-si (- si (fold-left + 0 (map cdr stack-sizes)))]
+
+ [get-offset (lambda (n) (cdr (assoc n stack-offsets)))]
+
+ [inner-env
+ (fold-left
+ (lambda (env comps)
+ (let* ([scc-binding-offsets
+ (fold-left
+ (lambda (acc name)
+ (cons (cons name (get-offset name))
+ acc))
+ (env-bindings env)
+ comps)]
+ [scc-env (make-env (env-data-layouts env) scc-binding-offsets)])
+ (for-each
+ (lambda (name)
+ (let* ([expr (cadr (assoc name bindings))]
+ [size (expr-size dls expr)])
+ (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
+ (make-env
+ (env-data-layouts scc-env)
+ (cons (cons name 'self-captive)
+ (env-bindings scc-env))))
+ (codegen-expr expr inner-si scc-env))
+
+ (if (on-stack? expr)
+ ; copy over whatevers on the stack
+ (emit-stack-copy inner-si (get-offset name) size)
+ (emit "movq %rax, ~a(%rbp)" (get-offset name)))))
+ comps)
+ scc-env))
+ env
+ (reverse (sccs (graph bindings))))])
-(define (codegen-let bindings body si env)
- (let* ((stack-offsets (map (lambda (x) (- si (* x wordsize)))
- (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)))
(for-each (lambda (form)
(codegen-expr form inner-si inner-env))
body)))
-(define (codegen-var name si env)
- (when (not (assoc name env))
+(define (codegen-var e si env)
+ (let* ([stack-type (on-stack? e)]
+ [name (if stack-type (caddr e) e)]
+
+ [dls (env-data-layouts env)]
+
+
+ [stack-offset (assoc name (env-bindings env))])
+
+ (when (and (not stack-offset) (not singleton?))
(error #f (format "Variable ~a is not bound" name)))
- (let ((offset (cdr (assoc name env))))
- (emit "movq ~a(%rbp), %rax" offset)))
+
+ (cond
+ [(singleton? dls name)
+ ; singletons don't need to be in the environment
+ ; just store the tag
+ (let ([tag (data-sum-tag (env-data-layouts env) name)])
+ (if (on-stack? e)
+ (emit "movq $~a, ~a(%rbp)" tag si)
+ (emit "movq $~a, %rax" tag)))]
+ [(on-stack? e) (emit-stack-copy (cdr stack-offset) si (type-size dls stack-type))]
+ [else (emit "movq ~a(%rbp), %rax" (cdr stack-offset))])))
(define cur-lambda 0)
(define (fresh-lambda)
(set! cur-lambda (+ 1 cur-lambda))
(format "_lambda~a" (- cur-lambda 1)))
- ; for now we can only call closures
-(define (codegen-call closure args 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
+ ; a closure on the heap looks like:
+ ; 0 8 16 24
+ ; addr var1.... var2.... var3....
+
+(define (codegen-closure label captured si env)
+ (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 (e i)
- (emit "movq ~a(%rbp), ~a"
- (cdr (assoc e env)) ; offset of the var
- (param-register i)))
- captured (range 0 (length captured)))
+ (lambda (var-name heap-offset)
+ (let ([stack-offset (cdr (assoc var-name (env-bindings 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
+ heap-offsets)))
+
+ ; for now we can only call closures
+(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))
+ (dls (car stuff))
+ (captives (cadr stuff))
+ (args (caddr stuff))
+ (body (cadddr 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)))
+ [bindings (map cons params stack-offsets)]
+ [env (make-env dls bindings)])
(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 "ret")))
-(define (codegen-string label)
- (case target
- ('darwin (emit "movq ~a@GOTPCREL(%rip), %rax" label))
- ('linux (emit "movq $~a, %rax" label))))
-
(define cur-label 0)
(define (fresh-label)
(set! cur-label (+ 1 cur-label))
(codegen-expr else si env)
(emit "~a:" exit-label)))
+(define (codegen-case switch cases si env)
+ (define dls (env-data-layouts env))
+ (define exit-label (fresh-label))
+
+
+ ; checks if equal and returns assoc list of bindings
+ (define (check-equal jne-label type inner-offset x)
+
+ ; TODO: tidy this up! comparibles and binds could be merged
+ ; (foo a 2 (bar x)) -> ((2 Int 1) ((bar x) A 2))
+ ; sum: foo
+ (define (comparibles sum)
+ (let ([product-types (cdr (assoc sum (cdr (assoc type dls))))])
+ (if (null? product-types)
+ '()
+ (filter (lambda (x) (not (eqv? 'var (ast-type (car x)))))
+ (map (lambda (x t i) (list x t i))
+ (cdr x)
+ product-types
+ (range 0 (length product-types)))))))
+
+ (define (binds sum)
+ (let ([product-types (cdr (assoc sum (cdr (assoc type dls))))])
+ (if (null? product-types)
+ '()
+ (filter (lambda (x) (eqv? 'var (ast-type (car x))))
+ (map (lambda (x i)
+ (cons x
+ (- inner-offset
+ (data-product-offset dls type sum i))))
+ (cdr x)
+ (range 0 (length (cdr x))))))))
+
+
+ (if (adt? dls type)
+ (let* ([sum (if (list? x) (car x) x)] ; can sometimes be a singleton
+ [tag (data-sum-tag dls sum)])
+ ; the tag is at the top (beginning) of the adt on the stack
+ (emit "cmpq $~a, ~a(%rbp)" tag inner-offset)
+ (emit "jne ~a" jne-label)
+
+ (append (binds sum)
+ (flat-map
+ (lambda (cmp) ; cmp = (x type index)
+ (check-equal jne-label
+ (cadr cmp)
+ (- inner-offset (data-product-offset dls type sum (caddr cmp)))
+ (car cmp)))
+ (comparibles sum))))
+ (if (eqv? 'var (ast-type x))
+ (list (cons x inner-offset))
+ (begin
+ (emit "cmpq $~a, ~a(%rbp)" x inner-offset)
+ (emit "jne ~a" jne-label)
+ '() ))))
+
+ (define (codegen-adt-match type case)
+ (let* ([match (car case)]
+ [expr (cadr case)]
+ [next-section-label (fresh-label)]
+ [inner-si (- si (type-size dls type))]
+ [new-env (env-append-bindings env
+ (check-equal next-section-label type si match))])
+
+
+ (codegen-expr expr inner-si new-env)
+ (emit "jmp ~a" exit-label)
+
+ (emit "~a:" next-section-label)))
+
+ (define (codegen-literal-match case)
+ (let* ([next-section-label (fresh-label)]
+ [match (car case)]
+ [val (if (singleton? dls match)
+ (data-sum-tag dls match)
+ match)])
+ (emit "cmpq $~a, %rax" val)
+ (emit "jne ~a" next-section-label)
+ (codegen-expr (cadr case) si env)
+ (emit "jmp ~a" exit-label)
+ (emit "~a:" next-section-label)))
+
+ ; generate the switch
+ ; (and store it on the stack if not a stack value)
+ (codegen-expr switch si env)
+
+ (if (on-stack? switch)
+ ; adt pattern match
+ (for-each (lambda (x) (codegen-adt-match (cadr switch) x)) cases)
+ (for-each codegen-literal-match cases))
+ (emit "~a:" exit-label))
+
+
+(define (data-tor env e)
+ (if (not (list? e)) #f
+ (assoc (car e) (flat-map data-tors (env-data-layouts env)))))
+
+ ; returns the internal offset in bytes of a product within an ADT
+ ; given the constructor layout
+ ; constructor-layout: (foo (Int Bool))
+(define (data-product-offset data-layouts type sum index)
+ (let* ([products (cdr (assoc sum (cdr (assoc type data-layouts))))]
+ [to-traverse (list-head products index)])
+ (fold-left
+ (lambda (acc t) (+ acc (type-size data-layouts t)))
+ wordsize ; skip the tag in the first word
+ to-traverse)))
+
+(define (data-sum-tag data-layouts sum)
+
+ (define (go acc sums)
+ (when (null? sums) (error #f "data-sum-tag no sum for type" sum type))
+ (if (eqv? sum (car sums))
+ acc
+ (go (+ 1 acc) (cdr sums))))
+ (let* ([all-sum-names (flat-map (lambda (x) (map (lambda (y) (cons (car y) (car x))) (cdr x))) data-layouts)]
+ [type (cdr (assoc sum all-sum-names))]
+ [type-sums (cdr (assoc type data-layouts))])
+ (go 0 (map car type-sums))))
+
+(define (codegen-data-tor e si env)
+
+ (define dls (env-data-layouts env))
+
+ (define (codegen-destructor tor)
+ (let* ([res (codegen-expr (cadr e) si env)]
+ [info (cadr tor)]
+ [type (car info)]
+ [sum (cadr info)]
+ [index (caddr info)]
+ [product-type (cadddr info)]
+ [product-type-size (type-size dls product-type)]
+
+ [safe-space-offset (- si (type-size dls type))]
+
+ [inner-offset (- si (data-product-offset dls type sum index))])
+
+ (when (not (on-stack? (cadr e)))
+ (error #f "trying to destruct something that isn't a stack expression"))
+ (emit "# deconstructing")
+
+ (if (stack-type? (env-data-layouts env) product-type)
+ ; if copying from the stack, need to first copy
+ ; to a safe space above to avoid overwriting
+ ; the original result on the stack
+ ; this is bad. please remove this in the rewrite.
+ (begin
+ (emit-stack-copy inner-offset safe-space-offset product-type-size)
+ (emit-stack-copy safe-space-offset si product-type-size))
+ (emit "movq ~a(%rbp), %rax" inner-offset))))
+
+ (define (codegen-constructor tor)
+ (let* ([info (cadr tor)]
+ [type (car info)]
+ [sum (cadr info)]
+ [constructor (car e)]
+
+ [args (cdr e)]
+
+ [tag (data-sum-tag (env-data-layouts env)
+ sum)]
+
+ [inner-si (- si (type-size dls type))]
+
+ [product-types (cdr (assoc sum (cdr (assoc type dls))))]
+
+ [insert-product
+ (lambda (expr i product-type)
+ (let ([dest-offset
+ (- si (data-product-offset dls type sum i))]
+ [product-size (type-size dls product-type)])
+ (codegen-expr expr inner-si env)
+ (if (on-stack? expr)
+ (emit-stack-copy inner-si dest-offset product-size)
+ (emit "movq %rax, ~a(%rbp)" dest-offset))))])
+
+ ; emit the tag
+ (emit "movq $~a, ~a(%rbp)" tag si)
+ ; generate products
+ (for-each insert-product args (range 0 (length args)) product-types)))
+
+ (let* ([tor (data-tor env e)]
+ [constructor (eqv? 'constructor (caddr (cadr tor)))])
+ (if constructor
+ (codegen-constructor tor)
+ (codegen-destructor tor))))
+
(define (codegen-expr e si env)
+ (emit "# ~a" e)
(case (ast-type e)
- ('builtin e)
- ('closure 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
+ (if (data-tor env e)
+ (codegen-data-tor e si env)
+ (codegen-call (car e) (cdr e) si env)))))
+
+ ; this is a builtin being passed around as a variable
+ ; this should have been converted to a closure!
+ ('builtin (error #f "passing about a builtin!" e))
('let (codegen-let (let-bindings e)
(let-body e)
('var (codegen-var e si env))
('if (codegen-if (cadr e) (caddr e) (cadddr e) si env))
+ ('case (codegen-case (case-switch e) (case-cases e) si env))
('bool-literal (emit "movq $~a, %rax" (if e 1 0)))
('int-literal (emit "movq $~a, %rax" e))
- ('static-string (codegen-string (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 '())
+ ('static-string (emit "movq ~a@GOTPCREL(%rip), %rax"
+ (cadr e)))
+
+ ('stack (case (ast-type (caddr e))
+ ['var (codegen-var e si env)]
+ [else (codegen-expr (caddr e) si env)]))
+
+ (else (error #f "don't know how to codegen this")))
+ (emit "# done ~a" e))
+
+ ; takes in a expr annotated with types and returns a type-less AST
+ ; with stack values wrapped
+(define (annotate-stack-values data-layouts ann-e)
+ (define (strip e)
+ (ast-traverse strip (ann-expr e)))
+ (let* ([e (ann-expr ann-e)]
+ [type (ann-type ann-e)])
+ (if (stack-type? data-layouts type)
+ `(stack ,type
+ ,(ast-traverse (lambda (x) (annotate-stack-values data-layouts x)) e))
+ (ast-traverse (lambda (x)
+ (annotate-stack-values data-layouts x))
+ e))))
+
+(define (free-vars dls prog)
+ (define bound (singletons dls))
(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)
-(define (extract-lambdas program)
+ ; ((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 dls program)
(define lambdas '())
(define (add-lambda e)
(let* ((label (fresh-lambda))
(args (lambda-args e))
- (captured (free-vars e))
+ (captured (free-vars dls e))
(body (extract (lambda-body e)))
- (new-lambda (list label args captured body)))
+ (new-lambda (cons label (list dls 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 dls 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)))
(let ((transformed (extract program)))
(cons strings transformed))))
-(define (codegen-string-data s)
+(define (emit-string-data s)
(emit "~a:" (car s))
(emit "\t.string \"~a\"" (cdr s)))
-;; (define (amd64-abi f)
-;; ; preserve registers
-;; (emit "push %rbp")
-;; ;; (emit "push %rbx")
-;; ;; (for-each (lambda (i)
-;; ;; (emit (string-append
-;; ;; "push %r"
-;; ;; (number->string i))))
-;; ;; '(12 13 14 15))
-
-;; (emit "movq %rsp, %rbp") ; set up the base pointer
-
-;; (f) ; call stuff
-;; ; restore preserved registers
-;; ;; (for-each (lambda (i)
-;; ;; (emit (string-append
-;; ;; "pop %r"
-;; ;; (number->string i))))
-;; ;; '(15 14 13 12))
-;; ;; (emit "pop %rbx")
-;; (emit "pop %rbp")
-;; (emit "ret"))
-
; 24(%rbp) mem arg 1
; 16(%rbp) mem arg 0 prev frame
; -----------------------
(else (error #f "need to test out the below"))
(else (format "~a(%rsp)" (- n 6)))))
-(define (codegen program)
- (let* ((extract-res-0 (extract-strings program))
- (strings (car extract-res-0))
- (extract-res-1 (extract-lambdas (cdr extract-res-0)))
- (lambdas (car extract-res-1))
- (xform-prog (cdr extract-res-1)))
+(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-without-stdlib)
+ (set! cur-label 0)
+ (set! cur-lambda 0)
+ (let* ([program (append stdlib program-without-stdlib)]
+ [data-layouts (program-data-layouts program)]
+
+ [pattern-matched (expand-pattern-matches program)]
+ [type-annotated (annotate-types pattern-matched)]
+ [stack-annotated (annotate-stack-values data-layouts
+ type-annotated)]
+
+ (strings-res (extract-strings stack-annotated))
+ (strings (car strings-res))
+ (lambdas-res (extract-lambdas data-layouts (cdr strings-res)))
+ (lambdas (car lambdas-res))
+ (xform-prog (cdr lambdas-res)))
+
+ ; verify pattern matches are total
+ (verify-cases data-layouts type-annotated)
(emit "\t.global _start")
(emit "\t.text")
- ; (emit ".p2align 4,,15") is this needed?
(for-each codegen-lambda lambdas)
(emit "_start:")
+
+ (initialize-heap)
+
(emit "movq %rsp, %rbp") ; set up the base pointer
- (codegen-expr xform-prog 0 '())
+ (codegen-expr xform-prog (- wordsize) (make-env data-layouts '()))
; exit syscall
(emit "mov %rax, %rdi")
('linux (emit "mov $60, %rax")))
(emit "syscall")
- (emit "\t.data")
+ (emit ".data")
+
+ (emit "heap_start:")
+ (emit "\t.quad 0")
- (for-each codegen-string-data strings)))
+ (for-each emit-string-data strings)))
(define (compile-to-binary program output t)
(set! target t)
- (when (not (eq? (typecheck program) 'int)) (error #f "not an int"))
+ (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
('linux
(system "as /tmp/a.s -o /tmp/a.o")
(system (format "ld /tmp/a.o -o ~a" output))))))
+
+; NOTES
+; syscalls in linux and darwin use the following arguments for syscall instruction:
+; %rax = syscall #
+; %rdi = 1st arg
+; %rsi = 2nd arg
+; %rdx = 3rd arg
+; %r10 = 4th arg
+; %r8 = 5th arg
+; %r9 = 6th arg
+
+; on darwin, unix/posix syscalls are offset by 0x2000000 (syscall classes)
+; 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
projects / scheme.git / blobdiff