(load "typecheck.scm")
(load "ast.scm")
+(define target 'darwin)
+
(define (emit . s)
(begin
(apply printf s)
(emit "not %rax")
(emit "andq $1, %rax"))
+; '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")
- (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)
+ (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 %rbx, %rsi") ; string addr
+ (emit "mov %rcx, %rdx") ; num bytes
+ (emit "mov $1, %rax") ; file handle (stdout)
+ (emit "mov $1, %rdi"))) ; syscall 1 (write)
(emit "syscall"))
(define (range s n)
(set! cur-lambda (+ 1 cur-lambda))
(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....
++
+ (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)")
+ captured
+ stack-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))
(param-register i)))
captured (range 0 (length captured)))
-
; then codegen the arguments and move them into the next param registers
(for-each
(lambda (e i)
(emit "pop %rbp") ; restore caller's base pointer
(emit "ret")))
- ('linux (emit "movq $~a, %rax" label))))
+(define (codegen-string label)
+ (case target
+ ('darwin (emit "movq ~a@GOTPCREL(%rip), %rax" label))
++ ('linux (emit "lea $~a, %rax" label))))
+
(define cur-label 0)
(define (fresh-label)
(set! cur-label (+ 1 cur-label))
(define (codegen-expr e si env)
(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
('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 (codegen-string (cadr e)))
(else (error #f "don't know how to codegen this"))))
(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)))
(for-each codegen-lambda lambdas)
(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)")
+
(emit "movq %rsp, %rbp") ; set up the base pointer
(codegen-expr xform-prog 0 '())
; 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 "\t.data")
+ (emit ".data")
- (for-each codegen-string-data strings)))
+ (emit "heap_start:")
+ (emit "\t.quad 0")
+
+ (for-each emit-string-data strings)))
-(define (compile-to-binary program output)
+(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:
+ ; %rax = syscall #
+ ; %rdi = 1st arg
+ ; %rsi = 2nd arg
+ ; %rdx = 3rd arg
+ ; %r10 = 4th arg
+ ; %r8 = 5th arg
+ ; %r9 = 6th arg