1: /* command line interpretation, image loading etc. for Gforth
2:
3:
4: Copyright (C) 1995,1996,1997,1998,2000,2003,2004,2005,2006,2007,2008,2009 Free Software Foundation, Inc.
5:
6: This file is part of Gforth.
7:
8: Gforth is free software; you can redistribute it and/or
9: modify it under the terms of the GNU General Public License
10: as published by the Free Software Foundation, either version 3
11: of the License, or (at your option) any later version.
12:
13: This program is distributed in the hope that it will be useful,
14: but WITHOUT ANY WARRANTY; without even the implied warranty of
15: MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16: GNU General Public License for more details.
17:
18: You should have received a copy of the GNU General Public License
19: along with this program; if not, see http://www.gnu.org/licenses/.
20: */
21:
22: #include "config.h"
23: #include "forth.h"
24: #include <errno.h>
25: #include <ctype.h>
26: #include <stdio.h>
27: #include <unistd.h>
28: #include <string.h>
29: #include <math.h>
30: #include <sys/types.h>
31: #ifdef HAVE_ALLOCA_H
32: #include <alloca.h>
33: #endif
34: #ifndef STANDALONE
35: #include <sys/stat.h>
36: #endif
37: #include <fcntl.h>
38: #include <assert.h>
39: #include <stdlib.h>
40: #include <signal.h>
41:
42: #ifndef STANDALONE
43: #if HAVE_SYS_MMAN_H
44: #include <sys/mman.h>
45: #endif
46: #endif
47: #include "io.h"
48: #include "getopt.h"
49: #ifndef STANDALONE
50: #include <locale.h>
51: #endif
52:
53: /* output rules etc. for burg with --debug and --print-sequences */
54: /* #define BURG_FORMAT*/
55:
56: typedef enum prim_num {
57: /* definitions of N_execute etc. */
58: #include PRIM_NUM_I
59: N_START_SUPER
60: } PrimNum;
61:
62: /* global variables for engine.c
63: We put them here because engine.c is compiled several times in
64: different ways for the same engine. */
65: Cell *gforth_SP;
66: Float *gforth_FP;
67: Address gforth_UP=NULL;
68: Cell *gforth_RP;
69: Address gforth_LP;
70:
71: #ifndef HAS_LINKBACK
72: void * gforth_pointers[] = {
73: (void*)&gforth_SP,
74: (void*)&gforth_FP,
75: (void*)&gforth_LP,
76: (void*)&gforth_RP,
77: (void*)&gforth_UP,
78: (void*)gforth_engine
79: #ifdef HAS_FILE
80: ,
81: (void*)cstr,
82: (void*)tilde_cstr
83: #endif
84: };
85: #endif
86:
87: #ifdef HAS_FFCALL
88:
89: #include <callback.h>
90:
91: va_alist gforth_clist;
92:
93: void gforth_callback(Xt* fcall, void * alist)
94: {
95: /* save global valiables */
96: Cell *rp = gforth_RP;
97: Cell *sp = gforth_SP;
98: Float *fp = gforth_FP;
99: Address lp = gforth_LP;
100: va_alist clist = gforth_clist;
101:
102: gforth_clist = (va_alist)alist;
103:
104: gforth_engine(fcall, sp, rp, fp, lp sr_call);
105:
106: /* restore global variables */
107: gforth_RP = rp;
108: gforth_SP = sp;
109: gforth_FP = fp;
110: gforth_LP = lp;
111: gforth_clist = clist;
112: }
113: #endif
114:
115: #ifdef GFORTH_DEBUGGING
116: /* define some VM registers as global variables, so they survive exceptions;
117: global register variables are not up to the task (according to the
118: GNU C manual) */
119: #if defined(GLOBALS_NONRELOC)
120: saved_regs saved_regs_v;
121: saved_regs *saved_regs_p = &saved_regs_v;
122: #else /* !defined(GLOBALS_NONRELOC) */
123: Xt *saved_ip;
124: Cell *rp;
125: #endif /* !defined(GLOBALS_NONRELOC) */
126: #endif /* !defined(GFORTH_DEBUGGING) */
127:
128: #ifdef NO_IP
129: Label next_code;
130: #endif
131:
132: #ifdef HAS_FILE
133: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
134: char* pfileattr[6]={"r","r","r+","r+","w","w"};
135:
136: #ifndef O_BINARY
137: #define O_BINARY 0
138: #endif
139: #ifndef O_TEXT
140: #define O_TEXT 0
141: #endif
142:
143: int ufileattr[6]= {
144: O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
145: O_RDWR |O_BINARY, O_RDWR |O_BINARY,
146: O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
147: #endif
148: /* end global vars for engine.c */
149:
150: #define PRIM_VERSION 1
151: /* increment this whenever the primitives change in an incompatible way */
152:
153: #ifndef DEFAULTPATH
154: # define DEFAULTPATH "."
155: #endif
156:
157: #ifdef MSDOS
158: jmp_buf throw_jmp_buf;
159: #endif
160:
161: #if defined(DOUBLY_INDIRECT)
162: # define CFA(n) ({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
163: #else
164: # define CFA(n) ((Cell)(symbols+((n)&~0x4000UL)))
165: #endif
166:
167: #define maxaligned(n) (typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
168:
169: static UCell dictsize=0;
170: static UCell dsize=0;
171: static UCell rsize=0;
172: static UCell fsize=0;
173: static UCell lsize=0;
174: int offset_image=0;
175: int die_on_signal=0;
176: int ignore_async_signals=0;
177: #ifndef INCLUDE_IMAGE
178: static int clear_dictionary=0;
179: UCell pagesize=1;
180: char *progname;
181: #else
182: char *progname = "gforth";
183: int optind = 1;
184: #endif
185: #ifndef MAP_NORESERVE
186: #define MAP_NORESERVE 0
187: #endif
188: /* IF you have an old Cygwin, this may help:
189: #ifdef __CYGWIN__
190: #define MAP_NORESERVE 0
191: #endif
192: */
193: static int map_noreserve=MAP_NORESERVE;
194:
195: #define CODE_BLOCK_SIZE (512*1024) /* !! overflow handling for -native */
196: Address code_area=0;
197: Cell code_area_size = CODE_BLOCK_SIZE;
198: Address code_here; /* does for code-area what HERE does for the dictionary */
199: Address start_flush=NULL; /* start of unflushed code */
200: Cell last_jump=0; /* if the last prim was compiled without jump, this
201: is it's number, otherwise this contains 0 */
202:
203: static int no_super=0; /* true if compile_prim should not fuse prims */
204: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
205: dynamically */
206: static int print_metrics=0; /* if true, print metrics on exit */
207: static int static_super_number = 10000; /* number of ss used if available */
208: #define MAX_STATE 9 /* maximum number of states */
209: static int maxstates = MAX_STATE; /* number of states for stack caching */
210: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
211: static int diag = 0; /* if true: print diagnostic informations */
212: static int tpa_noequiv = 0; /* if true: no state equivalence checking */
213: static int tpa_noautomaton = 0; /* if true: no tree parsing automaton */
214: static int tpa_trace = 0; /* if true: data for line graph of new states etc. */
215: static int print_sequences = 0; /* print primitive sequences for optimization */
216: static int relocs = 0;
217: static int nonrelocs = 0;
218:
219: #ifdef HAS_DEBUG
220: int debug=0;
221: # define debugp(x...) do { if (debug) fprintf(x); } while (0)
222: #else
223: # define perror(x...)
224: # define fprintf(x...)
225: # define debugp(x...)
226: #endif
227:
228: ImageHeader *gforth_header;
229: Label *vm_prims;
230: #ifdef DOUBLY_INDIRECT
231: Label *xts; /* same content as vm_prims, but should only be used for xts */
232: #endif
233:
234: #ifndef NO_DYNAMIC
235: #ifndef CODE_ALIGNMENT
236: #define CODE_ALIGNMENT 0
237: #endif
238:
239: #define MAX_IMMARGS 2
240:
241: typedef struct {
242: Label start; /* NULL if not relocatable */
243: Cell length; /* only includes the jump iff superend is true*/
244: Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
245: char superend; /* true if primitive ends superinstruction, i.e.,
246: unconditional branch, execute, etc. */
247: Cell nimmargs;
248: struct immarg {
249: Cell offset; /* offset of immarg within prim */
250: char rel; /* true if immarg is relative */
251: } immargs[MAX_IMMARGS];
252: } PrimInfo;
253:
254: PrimInfo *priminfos;
255: PrimInfo **decomp_prims;
256:
257: const char const* const prim_names[]={
258: #include PRIM_NAMES_I
259: };
260:
261: void init_ss_cost(void);
262:
263: static int is_relocatable(int p)
264: {
265: return !no_dynamic && priminfos[p].start != NULL;
266: }
267: #else /* defined(NO_DYNAMIC) */
268: static int is_relocatable(int p)
269: {
270: return 0;
271: }
272: #endif /* defined(NO_DYNAMIC) */
273:
274: #ifdef MEMCMP_AS_SUBROUTINE
275: int gforth_memcmp(const char * s1, const char * s2, size_t n)
276: {
277: return memcmp(s1, s2, n);
278: }
279: #endif
280:
281: static Cell max(Cell a, Cell b)
282: {
283: return a>b?a:b;
284: }
285:
286: static Cell min(Cell a, Cell b)
287: {
288: return a<b?a:b;
289: }
290:
291: #ifndef STANDALONE
292: /* image file format:
293: * "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
294: * padding to a multiple of 8
295: * magic: "Gforth3x" means format 0.6,
296: * where x is a byte with
297: * bit 7: reserved = 0
298: * bit 6:5: address unit size 2^n octets
299: * bit 4:3: character size 2^n octets
300: * bit 2:1: cell size 2^n octets
301: * bit 0: endian, big=0, little=1.
302: * The magic are always 8 octets, no matter what the native AU/character size is
303: * padding to max alignment (no padding necessary on current machines)
304: * ImageHeader structure (see forth.h)
305: * data (size in ImageHeader.image_size)
306: * tags ((if relocatable, 1 bit/data cell)
307: *
308: * tag==1 means that the corresponding word is an address;
309: * If the word is >=0, the address is within the image;
310: * addresses within the image are given relative to the start of the image.
311: * If the word =-1 (CF_NIL), the address is NIL,
312: * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
313: * If the word =CF(DODOES), it's a DOES> CFA
314: * If the word =CF(DOESJUMP), it's a DOES JUMP (2 Cells after DOES>,
315: * possibly containing a jump to dodoes)
316: * If the word is <CF(DOER_MAX) and bit 14 is set, it's the xt of a primitive
317: * If the word is <CF(DOER_MAX) and bit 14 is clear,
318: * it's the threaded code of a primitive
319: * bits 13..9 of a primitive token state which group the primitive belongs to,
320: * bits 8..0 of a primitive token index into the group
321: */
322:
323: Cell groups[32] = {
324: 0,
325: 0
326: #undef GROUP
327: #undef GROUPADD
328: #define GROUPADD(n) +n
329: #define GROUP(x, n) , 0
330: #include PRIM_GRP_I
331: #undef GROUP
332: #undef GROUPADD
333: #define GROUP(x, n)
334: #define GROUPADD(n)
335: };
336:
337: static unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
338: int size, Cell base)
339: /* produce a bitmask marking all the branch targets */
340: {
341: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
342: Cell token;
343: unsigned char bits;
344: unsigned char *result=malloc(steps);
345:
346: memset(result, 0, steps);
347: for(k=0; k<steps; k++) {
348: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
349: if(bits & (1U << (RELINFOBITS-1))) {
350: assert(i*sizeof(Cell) < size);
351: token=image[i];
352: if (token>=base) { /* relocatable address */
353: UCell bitnum=(token-base)/sizeof(Cell);
354: if (bitnum/RELINFOBITS < (UCell)steps)
355: result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
356: }
357: }
358: }
359: }
360: return result;
361: }
362:
363: void gforth_relocate(Cell *image, const Char *bitstring,
364: UCell size, Cell base, Label symbols[])
365: {
366: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
367: Cell token;
368: char bits;
369: Cell max_symbols;
370: /*
371: * A virtual start address that's the real start address minus
372: * the one in the image
373: */
374: Cell *start = (Cell * ) (((void *) image) - ((void *) base));
375: unsigned char *targets = branch_targets(image, bitstring, size, base);
376:
377: /* group index into table */
378: if(groups[31]==0) {
379: int groupsum=0;
380: for(i=0; i<32; i++) {
381: groupsum += groups[i];
382: groups[i] = groupsum;
383: /* printf("group[%d]=%d\n",i,groupsum); */
384: }
385: i=0;
386: }
387:
388: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
389:
390: for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
391: ;
392: max_symbols--;
393:
394: for(k=0; k<steps; k++) {
395: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
396: /* fprintf(stderr,"relocate: image[%d]\n", i);*/
397: if(bits & (1U << (RELINFOBITS-1))) {
398: assert(i*sizeof(Cell) < size);
399: /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
400: token=image[i];
401: if(token<0) {
402: int group = (-token & 0x3E00) >> 9;
403: if(group == 0) {
404: switch(token|0x4000) {
405: case CF_NIL : image[i]=0; break;
406: #if !defined(DOUBLY_INDIRECT)
407: case CF(DOCOL) :
408: case CF(DOVAR) :
409: case CF(DOCON) :
410: case CF(DOVAL) :
411: case CF(DOUSER) :
412: case CF(DODEFER) :
413: case CF(DOFIELD) :
414: case CF(DOABICODE) : MAKE_CF(image+i,symbols[CF(token)]); break;
415: case CF(DOESJUMP): assert(0); image[i]=0; break;
416: #endif /* !defined(DOUBLY_INDIRECT) */
417: case CF(DODOES) :
418: MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
419: break;
420: default : /* backward compatibility */
421: /* printf("Code field generation image[%x]:=CFA(%x)\n",
422: i, CF(image[i])); */
423: if (CF((token | 0x4000))<max_symbols) {
424: image[i]=(Cell)CFA(CF(token));
425: #ifdef DIRECT_THREADED
426: if ((token & 0x4000) == 0) { /* threade code, no CFA */
427: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
428: compile_prim1(0);
429: compile_prim1(&image[i]);
430: }
431: #endif
432: } else
433: fprintf(stderr,"Primitive %ld used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n",(long)CF(token),(long)&image[i], i, PACKAGE_VERSION);
434: }
435: } else {
436: int tok = -token & 0x1FF;
437: if (tok < (groups[group+1]-groups[group])) {
438: #if defined(DOUBLY_INDIRECT)
439: image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
440: #else
441: image[i]=(Cell)CFA((groups[group]+tok));
442: #endif
443: #ifdef DIRECT_THREADED
444: if ((token & 0x4000) == 0) { /* threade code, no CFA */
445: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
446: compile_prim1(0);
447: compile_prim1(&image[i]);
448: }
449: #endif
450: } else
451: fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
452: }
453: } else {
454: /* if base is > 0: 0 is a null reference so don't adjust*/
455: if (token>=base) {
456: image[i]+=(Cell)start;
457: }
458: }
459: }
460: }
461: }
462: free(targets);
463: finish_code();
464: ((ImageHeader*)(image))->base = (Address) image;
465: }
466:
467: #ifndef DOUBLY_INDIRECT
468: static UCell checksum(Label symbols[])
469: {
470: UCell r=PRIM_VERSION;
471: Cell i;
472:
473: for (i=DOCOL; i<=DOER_MAX; i++) {
474: r ^= (UCell)(symbols[i]);
475: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
476: }
477: #ifdef DIRECT_THREADED
478: /* we have to consider all the primitives */
479: for (; symbols[i]!=(Label)0; i++) {
480: r ^= (UCell)(symbols[i]);
481: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
482: }
483: #else
484: /* in indirect threaded code all primitives are accessed through the
485: symbols table, so we just have to put the base address of symbols
486: in the checksum */
487: r ^= (UCell)symbols;
488: #endif
489: return r;
490: }
491: #endif
492:
493: static Address verbose_malloc(Cell size)
494: {
495: Address r;
496: /* leave a little room (64B) for stack underflows */
497: if ((r = malloc(size+64))==NULL) {
498: perror(progname);
499: exit(1);
500: }
501: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
502: debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
503: return r;
504: }
505:
506: static void *next_address=0;
507: static void after_alloc(Address r, Cell size)
508: {
509: if (r != (Address)-1) {
510: debugp(stderr, "success, address=$%lx\n", (long) r);
511: #if 0
512: /* not needed now that we protect the stacks with mprotect */
513: if (pagesize != 1)
514: next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
515: #endif
516: } else {
517: debugp(stderr, "failed: %s\n", strerror(errno));
518: }
519: }
520:
521: #ifndef MAP_FAILED
522: #define MAP_FAILED ((Address) -1)
523: #endif
524: #ifndef MAP_FILE
525: # define MAP_FILE 0
526: #endif
527: #ifndef MAP_PRIVATE
528: # define MAP_PRIVATE 0
529: #endif
530: #ifndef PROT_NONE
531: # define PROT_NONE 0
532: #endif
533: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
534: # define MAP_ANON MAP_ANONYMOUS
535: #endif
536:
537: #if defined(HAVE_MMAP)
538: static Address alloc_mmap(Cell size)
539: {
540: void *r;
541:
542: #if defined(MAP_ANON)
543: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
544: r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE|map_noreserve, -1, 0);
545: #else /* !defined(MAP_ANON) */
546: /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
547: apparently defaults) */
548: static int dev_zero=-1;
549:
550: if (dev_zero == -1)
551: dev_zero = open("/dev/zero", O_RDONLY);
552: if (dev_zero == -1) {
553: r = MAP_FAILED;
554: debugp(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ",
555: strerror(errno));
556: } else {
557: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
558: r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE|map_noreserve, dev_zero, 0);
559: }
560: #endif /* !defined(MAP_ANON) */
561: after_alloc(r, size);
562: return r;
563: }
564:
565: static void page_noaccess(void *a)
566: {
567: /* try mprotect first; with munmap the page might be allocated later */
568: debugp(stderr, "try mprotect(%p,$%lx,PROT_NONE); ", a, (long)pagesize);
569: if (mprotect(a, pagesize, PROT_NONE)==0) {
570: debugp(stderr, "ok\n");
571: return;
572: }
573: debugp(stderr, "failed: %s\n", strerror(errno));
574: debugp(stderr, "try munmap(%p,$%lx); ", a, (long)pagesize);
575: if (munmap(a,pagesize)==0) {
576: debugp(stderr, "ok\n");
577: return;
578: }
579: debugp(stderr, "failed: %s\n", strerror(errno));
580: }
581:
582: static size_t wholepage(size_t n)
583: {
584: return (n+pagesize-1)&~(pagesize-1);
585: }
586: #endif
587:
588: Address gforth_alloc(Cell size)
589: {
590: #if HAVE_MMAP
591: Address r;
592:
593: r=alloc_mmap(size);
594: if (r!=(Address)MAP_FAILED)
595: return r;
596: #endif /* HAVE_MMAP */
597: /* use malloc as fallback */
598: return verbose_malloc(size);
599: }
600:
601: static void *dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
602: {
603: void *image = MAP_FAILED;
604:
605: #if defined(HAVE_MMAP)
606: if (offset==0) {
607: image=alloc_mmap(dictsize);
608: if (image != (void *)MAP_FAILED) {
609: void *image1;
610: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
611: image1 = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE|map_noreserve, fileno(file), 0);
612: after_alloc(image1,dictsize);
613: if (image1 == (void *)MAP_FAILED)
614: goto read_image;
615: }
616: }
617: #endif /* defined(HAVE_MMAP) */
618: if (image == (void *)MAP_FAILED) {
619: image = gforth_alloc(dictsize+offset)+offset;
620: read_image:
621: rewind(file); /* fseek(imagefile,0L,SEEK_SET); */
622: fread(image, 1, imagesize, file);
623: }
624: return image;
625: }
626: #endif
627:
628: void set_stack_sizes(ImageHeader * header)
629: {
630: if (dictsize==0)
631: dictsize = header->dict_size;
632: if (dsize==0)
633: dsize = header->data_stack_size;
634: if (rsize==0)
635: rsize = header->return_stack_size;
636: if (fsize==0)
637: fsize = header->fp_stack_size;
638: if (lsize==0)
639: lsize = header->locals_stack_size;
640: dictsize=maxaligned(dictsize);
641: dsize=maxaligned(dsize);
642: rsize=maxaligned(rsize);
643: lsize=maxaligned(lsize);
644: fsize=maxaligned(fsize);
645: }
646:
647: #ifdef STANDALONE
648: void alloc_stacks(ImageHeader * h)
649: {
650: #define SSTACKSIZE 0x200
651: static Cell dstack[SSTACKSIZE+1];
652: static Cell rstack[SSTACKSIZE+1];
653:
654: h->dict_size=dictsize;
655: h->data_stack_size=dsize;
656: h->fp_stack_size=fsize;
657: h->return_stack_size=rsize;
658: h->locals_stack_size=lsize;
659:
660: h->data_stack_base=dstack+SSTACKSIZE;
661: // h->fp_stack_base=gforth_alloc(fsize);
662: h->return_stack_base=rstack+SSTACKSIZE;
663: // h->locals_stack_base=gforth_alloc(lsize);
664: }
665: #else
666: void alloc_stacks(ImageHeader * h)
667: {
668: h->dict_size=dictsize;
669: h->data_stack_size=dsize;
670: h->fp_stack_size=fsize;
671: h->return_stack_size=rsize;
672: h->locals_stack_size=lsize;
673:
674: #if defined(HAVE_MMAP) && !defined(STANDALONE)
675: if (pagesize > 1) {
676: size_t p = pagesize;
677: size_t totalsize =
678: wholepage(dsize)+wholepage(fsize)+wholepage(rsize)+wholepage(lsize)+5*p;
679: void *a = alloc_mmap(totalsize);
680: if (a != (void *)MAP_FAILED) {
681: page_noaccess(a); a+=p; h-> data_stack_base=a; a+=wholepage(dsize);
682: page_noaccess(a); a+=p; h-> fp_stack_base=a; a+=wholepage(fsize);
683: page_noaccess(a); a+=p; h->return_stack_base=a; a+=wholepage(rsize);
684: page_noaccess(a); a+=p; h->locals_stack_base=a; a+=wholepage(lsize);
685: page_noaccess(a);
686: debugp(stderr,"stack addresses: d=%p f=%p r=%p l=%p\n",
687: h->data_stack_base,
688: h->fp_stack_base,
689: h->return_stack_base,
690: h->locals_stack_base);
691: return;
692: }
693: }
694: #endif
695: h->data_stack_base=gforth_alloc(dsize);
696: h->fp_stack_base=gforth_alloc(fsize);
697: h->return_stack_base=gforth_alloc(rsize);
698: h->locals_stack_base=gforth_alloc(lsize);
699: }
700: #endif
701:
702: #warning You can ignore the warnings about clobbered variables in gforth_go
703: int gforth_go(void *image, int stack, Cell *entries)
704: {
705: volatile ImageHeader *image_header = (ImageHeader *)image;
706: Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
707: Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
708: Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
709: #ifdef GFORTH_DEBUGGING
710: volatile Cell *orig_rp0=rp0;
711: #endif
712: Address lp0=image_header->locals_stack_base + lsize;
713: Xt *ip0=(Xt *)(image_header->boot_entry);
714: #ifdef SYSSIGNALS
715: int throw_code;
716: #endif
717:
718: /* ensure that the cached elements (if any) are accessible */
719: #if !(defined(GFORTH_DEBUGGING) || defined(INDIRECT_THREADED) || defined(DOUBLY_INDIRECT) || defined(VM_PROFILING))
720: sp0 -= 8; /* make stuff below bottom accessible for stack caching */
721: fp0--;
722: #endif
723:
724: for(;stack>0;stack--)
725: *--sp0=entries[stack-1];
726:
727: #if defined(SYSSIGNALS) && !defined(STANDALONE)
728: get_winsize();
729:
730: install_signal_handlers(); /* right place? */
731:
732: if ((throw_code=setjmp(throw_jmp_buf))) {
733: static Cell signal_data_stack[24];
734: static Cell signal_return_stack[16];
735: static Float signal_fp_stack[1];
736:
737: signal_data_stack[15]=throw_code;
738:
739: #ifdef GFORTH_DEBUGGING
740: debugp(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
741: throw_code, saved_ip, rp);
742: if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
743: /* no rstack overflow or underflow */
744: rp0 = rp;
745: *--rp0 = (Cell)saved_ip;
746: }
747: else /* I love non-syntactic ifdefs :-) */
748: rp0 = signal_return_stack+16;
749: #else /* !defined(GFORTH_DEBUGGING) */
750: debugp(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
751: rp0 = signal_return_stack+16;
752: #endif /* !defined(GFORTH_DEBUGGING) */
753: /* fprintf(stderr, "rp=$%x\n",rp0);*/
754:
755: return((int)(Cell)gforth_engine(image_header->throw_entry, signal_data_stack+15,
756: rp0, signal_fp_stack, 0 sr_call));
757: }
758: #endif
759:
760: return((int)(Cell)gforth_engine(ip0,sp0,rp0,fp0,lp0 sr_call));
761: }
762:
763: #if !defined(INCLUDE_IMAGE) && !defined(STANDALONE)
764: static void print_sizes(Cell sizebyte)
765: /* print size information */
766: {
767: static char* endianstring[]= { " big","little" };
768:
769: fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
770: endianstring[sizebyte & 1],
771: 1 << ((sizebyte >> 1) & 3),
772: 1 << ((sizebyte >> 3) & 3),
773: 1 << ((sizebyte >> 5) & 3));
774: }
775:
776: /* static superinstruction stuff */
777:
778: struct cost { /* super_info might be a more accurate name */
779: char loads; /* number of stack loads */
780: char stores; /* number of stack stores */
781: char updates; /* number of stack pointer updates */
782: char branch; /* is it a branch (SET_IP) */
783: unsigned char state_in; /* state on entry */
784: unsigned char state_out; /* state on exit */
785: unsigned char imm_ops; /* number of immediate operands */
786: short offset; /* offset into super2 table */
787: unsigned char length; /* number of components */
788: };
789:
790: PrimNum super2[] = {
791: #include SUPER2_I
792: };
793:
794: struct cost super_costs[] = {
795: #include COSTS_I
796: };
797:
798: struct super_state {
799: struct super_state *next;
800: PrimNum super;
801: };
802:
803: #define HASH_SIZE 256
804:
805: struct super_table_entry {
806: struct super_table_entry *next;
807: PrimNum *start;
808: short length;
809: struct super_state *ss_list; /* list of supers */
810: } *super_table[HASH_SIZE];
811: int max_super=2;
812:
813: struct super_state *state_transitions=NULL;
814:
815: static int hash_super(PrimNum *start, int length)
816: {
817: int i, r;
818:
819: for (i=0, r=0; i<length; i++) {
820: r <<= 1;
821: r += start[i];
822: }
823: return r & (HASH_SIZE-1);
824: }
825:
826: static struct super_state **lookup_super(PrimNum *start, int length)
827: {
828: int hash=hash_super(start,length);
829: struct super_table_entry *p = super_table[hash];
830:
831: /* assert(length >= 2); */
832: for (; p!=NULL; p = p->next) {
833: if (length == p->length &&
834: memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
835: return &(p->ss_list);
836: }
837: return NULL;
838: }
839:
840: static void prepare_super_table()
841: {
842: int i;
843: int nsupers = 0;
844:
845: for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
846: struct cost *c = &super_costs[i];
847: if ((c->length < 2 || nsupers < static_super_number) &&
848: c->state_in < maxstates && c->state_out < maxstates) {
849: struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
850: struct super_state *ss = malloc(sizeof(struct super_state));
851: ss->super= i;
852: if (c->offset==N_noop && i != N_noop) {
853: if (is_relocatable(i)) {
854: ss->next = state_transitions;
855: state_transitions = ss;
856: }
857: } else if (ss_listp != NULL) {
858: ss->next = *ss_listp;
859: *ss_listp = ss;
860: } else {
861: int hash = hash_super(super2+c->offset, c->length);
862: struct super_table_entry **p = &super_table[hash];
863: struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
864: ss->next = NULL;
865: e->next = *p;
866: e->start = super2 + c->offset;
867: e->length = c->length;
868: e->ss_list = ss;
869: *p = e;
870: }
871: if (c->length > max_super)
872: max_super = c->length;
873: if (c->length >= 2)
874: nsupers++;
875: }
876: }
877: debugp(stderr, "Using %d static superinsts\n", nsupers);
878: if (nsupers>0 && !tpa_noautomaton && !tpa_noequiv) {
879: /* Currently these two things don't work together; see Section 3.2
880: of <http://www.complang.tuwien.ac.at/papers/ertl+06pldi.ps.gz>,
881: in particular Footnote 6 for the reason; hmm, we should be able
882: to use an automaton without state equivalence, but that costs
883: significant space so we only do it if the user explicitly
884: disables state equivalence. */
885: debugp(stderr, "Disabling tpa-automaton, because nsupers>0 and state equivalence is enabled.\n");
886: tpa_noautomaton = 1;
887: }
888: }
889:
890: /* dynamic replication/superinstruction stuff */
891:
892: #ifndef NO_DYNAMIC
893: static int compare_priminfo_length(const void *_a, const void *_b)
894: {
895: PrimInfo **a = (PrimInfo **)_a;
896: PrimInfo **b = (PrimInfo **)_b;
897: Cell diff = (*a)->length - (*b)->length;
898: if (diff)
899: return diff;
900: else /* break ties by start address; thus the decompiler produces
901: the earliest primitive with the same code (e.g. noop instead
902: of (char) and @ instead of >code-address */
903: return (*b)->start - (*a)->start;
904: }
905: #endif /* !defined(NO_DYNAMIC) */
906:
907: static char MAYBE_UNUSED superend[]={
908: #include PRIM_SUPEREND_I
909: };
910:
911: Cell npriminfos=0;
912:
913: Label goto_start;
914: Cell goto_len;
915:
916: #ifndef NO_DYNAMIC
917: static int compare_labels(const void *pa, const void *pb)
918: {
919: Label a = *(Label *)pa;
920: Label b = *(Label *)pb;
921: return a-b;
922: }
923: #endif
924:
925: static Label bsearch_next(Label key, Label *a, UCell n)
926: /* a is sorted; return the label >=key that is the closest in a;
927: return NULL if there is no label in a >=key */
928: {
929: int mid = (n-1)/2;
930: if (n<1)
931: return NULL;
932: if (n == 1) {
933: if (a[0] < key)
934: return NULL;
935: else
936: return a[0];
937: }
938: if (a[mid] < key)
939: return bsearch_next(key, a+mid+1, n-mid-1);
940: else
941: return bsearch_next(key, a, mid+1);
942: }
943:
944: static void check_prims(Label symbols1[])
945: {
946: int i;
947: #ifndef NO_DYNAMIC
948: Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted, *goto_p;
949: int nends1j;
950: #endif
951:
952: if (debug)
953: #ifdef __VERSION__
954: fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
955: #else
956: #define xstr(s) str(s)
957: #define str(s) #s
958: fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n");
959: #endif
960: for (i=0; symbols1[i]!=0; i++)
961: ;
962: npriminfos = i;
963:
964: #ifndef NO_DYNAMIC
965: if (no_dynamic)
966: return;
967: symbols2=gforth_engine2(0,0,0,0,0 sr_call);
968: #if NO_IP
969: symbols3=gforth_engine3(0,0,0,0,0 sr_call);
970: #else
971: symbols3=symbols1;
972: #endif
973: ends1 = symbols1+i+1;
974: ends1j = ends1+i;
975: goto_p = ends1j+i+1; /* goto_p[0]==before; ...[1]==after;*/
976: nends1j = i+1;
977: ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
978: memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
979: qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
980:
981: /* check whether the "goto *" is relocatable */
982: goto_len = goto_p[1]-goto_p[0];
983: debugp(stderr, "goto * %p %p len=%ld\n",
984: goto_p[0],symbols2[goto_p-symbols1],(long)goto_len);
985: if (memcmp(goto_p[0],symbols2[goto_p-symbols1],goto_len)!=0) { /* unequal */
986: no_dynamic=1;
987: debugp(stderr," not relocatable, disabling dynamic code generation\n");
988: init_ss_cost();
989: return;
990: }
991: goto_start = goto_p[0];
992:
993: priminfos = calloc(i,sizeof(PrimInfo));
994: for (i=0; symbols1[i]!=0; i++) {
995: int prim_len = ends1[i]-symbols1[i];
996: PrimInfo *pi=&priminfos[i];
997: struct cost *sc=&super_costs[i];
998: int j=0;
999: char *s1 = (char *)symbols1[i];
1000: char *s2 = (char *)symbols2[i];
1001: char *s3 = (char *)symbols3[i];
1002: Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
1003:
1004: pi->start = s1;
1005: pi->superend = superend[i]|no_super;
1006: pi->length = prim_len;
1007: pi->restlength = endlabel - symbols1[i] - pi->length;
1008: pi->nimmargs = 0;
1009: relocs++;
1010: #if defined(BURG_FORMAT)
1011: { /* output as burg-style rules */
1012: int p=super_costs[i].offset;
1013: if (p==N_noop)
1014: debugp(stderr, "S%d: S%d = %d (%d);", sc->state_in, sc->state_out, i+1, pi->length);
1015: else
1016: debugp(stderr, "S%d: op%d(S%d) = %d (%d);", sc->state_in, p, sc->state_out, i+1, pi->length);
1017: }
1018: #else
1019: debugp(stderr, "%-15s %d-%d %4d %p %p %p len=%3ld rest=%2ld send=%1d",
1020: prim_names[i], sc->state_in, sc->state_out,
1021: i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength),
1022: pi->superend);
1023: #endif
1024: if (endlabel == NULL) {
1025: pi->start = NULL; /* not relocatable */
1026: if (pi->length<0) pi->length=100;
1027: #ifndef BURG_FORMAT
1028: debugp(stderr,"\n non_reloc: no J label > start found\n");
1029: #endif
1030: relocs--;
1031: nonrelocs++;
1032: continue;
1033: }
1034: if (ends1[i] > endlabel && !pi->superend) {
1035: pi->start = NULL; /* not relocatable */
1036: pi->length = endlabel-symbols1[i];
1037: #ifndef BURG_FORMAT
1038: debugp(stderr,"\n non_reloc: there is a J label before the K label (restlength<0)\n");
1039: #endif
1040: relocs--;
1041: nonrelocs++;
1042: continue;
1043: }
1044: if (ends1[i] < pi->start && !pi->superend) {
1045: pi->start = NULL; /* not relocatable */
1046: pi->length = endlabel-symbols1[i];
1047: #ifndef BURG_FORMAT
1048: debugp(stderr,"\n non_reloc: K label before I label (length<0)\n");
1049: #endif
1050: relocs--;
1051: nonrelocs++;
1052: continue;
1053: }
1054: assert(pi->length>=0);
1055: assert(pi->restlength >=0);
1056: while (j<(pi->length+pi->restlength)) {
1057: if (s1[j]==s3[j]) {
1058: if (s1[j] != s2[j]) {
1059: pi->start = NULL; /* not relocatable */
1060: #ifndef BURG_FORMAT
1061: debugp(stderr,"\n non_reloc: engine1!=engine2 offset %3d",j);
1062: #endif
1063: /* assert(j<prim_len); */
1064: relocs--;
1065: nonrelocs++;
1066: break;
1067: }
1068: j++;
1069: } else {
1070: struct immarg *ia=&pi->immargs[pi->nimmargs];
1071:
1072: pi->nimmargs++;
1073: ia->offset=j;
1074: if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
1075: ia->rel=0;
1076: debugp(stderr,"\n absolute immarg: offset %3d",j);
1077: } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
1078: symbols1[DOER_MAX+1]) {
1079: ia->rel=1;
1080: debugp(stderr,"\n relative immarg: offset %3d",j);
1081: } else {
1082: pi->start = NULL; /* not relocatable */
1083: #ifndef BURG_FORMAT
1084: debugp(stderr,"\n non_reloc: engine1!=engine3 offset %3d",j);
1085: #endif
1086: /* assert(j<prim_len);*/
1087: relocs--;
1088: nonrelocs++;
1089: break;
1090: }
1091: j+=4;
1092: }
1093: }
1094: debugp(stderr,"\n");
1095: }
1096: decomp_prims = calloc(i,sizeof(PrimInfo *));
1097: for (i=DOER_MAX+1; i<npriminfos; i++)
1098: decomp_prims[i] = &(priminfos[i]);
1099: qsort(decomp_prims+DOER_MAX+1, npriminfos-DOER_MAX-1, sizeof(PrimInfo *),
1100: compare_priminfo_length);
1101: #endif
1102: }
1103:
1104: static void flush_to_here(void)
1105: {
1106: #ifndef NO_DYNAMIC
1107: if (start_flush)
1108: FLUSH_ICACHE((caddr_t)start_flush, code_here-start_flush);
1109: start_flush=code_here;
1110: #endif
1111: }
1112:
1113: static void MAYBE_UNUSED align_code(void)
1114: /* align code_here on some platforms */
1115: {
1116: #ifndef NO_DYNAMIC
1117: #if defined(CODE_PADDING)
1118: Cell alignment = CODE_ALIGNMENT;
1119: static char nops[] = CODE_PADDING;
1120: UCell maxpadding=MAX_PADDING;
1121: UCell offset = ((UCell)code_here)&(alignment-1);
1122: UCell length = alignment-offset;
1123: if (length <= maxpadding) {
1124: memcpy(code_here,nops+offset,length);
1125: code_here += length;
1126: }
1127: #endif /* defined(CODE_PADDING) */
1128: #endif /* defined(NO_DYNAMIC */
1129: }
1130:
1131: #ifndef NO_DYNAMIC
1132: static void append_jump(void)
1133: {
1134: if (last_jump) {
1135: PrimInfo *pi = &priminfos[last_jump];
1136:
1137: memcpy(code_here, pi->start+pi->length, pi->restlength);
1138: code_here += pi->restlength;
1139: memcpy(code_here, goto_start, goto_len);
1140: code_here += goto_len;
1141: align_code();
1142: last_jump=0;
1143: }
1144: }
1145:
1146: /* Gforth remembers all code blocks in this list. On forgetting (by
1147: executing a marker) the code blocks are not freed (because Gforth does
1148: not remember how they were allocated; hmm, remembering that might be
1149: easier and cleaner). Instead, code_here etc. are reset to the old
1150: value, and the "forgotten" code blocks are reused when they are
1151: needed. */
1152:
1153: struct code_block_list {
1154: struct code_block_list *next;
1155: Address block;
1156: Cell size;
1157: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
1158:
1159: static void reserve_code_space(UCell size)
1160: {
1161: if (code_area+code_area_size < code_here+size) {
1162: struct code_block_list *p;
1163: append_jump();
1164: debugp(stderr,"Did not use %ld bytes in code block\n",
1165: (long)(code_area+code_area_size-code_here));
1166: flush_to_here();
1167: if (*next_code_blockp == NULL) {
1168: code_here = start_flush = code_area = gforth_alloc(code_area_size);
1169: p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
1170: *next_code_blockp = p;
1171: p->next = NULL;
1172: p->block = code_here;
1173: p->size = code_area_size;
1174: } else {
1175: p = *next_code_blockp;
1176: code_here = start_flush = code_area = p->block;
1177: }
1178: next_code_blockp = &(p->next);
1179: }
1180: }
1181:
1182: static Address append_prim(Cell p)
1183: {
1184: PrimInfo *pi = &priminfos[p];
1185: Address old_code_here;
1186: reserve_code_space(pi->length+pi->restlength+goto_len+CODE_ALIGNMENT-1);
1187: memcpy(code_here, pi->start, pi->length);
1188: old_code_here = code_here;
1189: code_here += pi->length;
1190: return old_code_here;
1191: }
1192:
1193: static void reserve_code_super(PrimNum origs[], int ninsts)
1194: {
1195: int i;
1196: UCell size = CODE_ALIGNMENT-1; /* alignment may happen first */
1197: if (no_dynamic)
1198: return;
1199: /* use size of the original primitives as an upper bound for the
1200: size of the superinstruction. !! This is only safe if we
1201: optimize for code size (the default) */
1202: for (i=0; i<ninsts; i++) {
1203: PrimNum p = origs[i];
1204: PrimInfo *pi = &priminfos[p];
1205: if (is_relocatable(p))
1206: size += pi->length;
1207: else
1208: if (i>0)
1209: size += priminfos[origs[i-1]].restlength+goto_len+CODE_ALIGNMENT-1;
1210: }
1211: size += priminfos[origs[i-1]].restlength+goto_len;
1212: reserve_code_space(size);
1213: }
1214: #endif
1215:
1216: int forget_dyncode(Address code)
1217: {
1218: #ifdef NO_DYNAMIC
1219: return -1;
1220: #else
1221: struct code_block_list *p, **pp;
1222:
1223: for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
1224: if (code >= p->block && code < p->block+p->size) {
1225: next_code_blockp = &(p->next);
1226: code_here = start_flush = code;
1227: code_area = p->block;
1228: last_jump = 0;
1229: return -1;
1230: }
1231: }
1232: return -no_dynamic;
1233: #endif /* !defined(NO_DYNAMIC) */
1234: }
1235:
1236: static long dyncodesize(void)
1237: {
1238: #ifndef NO_DYNAMIC
1239: struct code_block_list *p;
1240: long size=0;
1241: for (p=code_block_list; p!=NULL; p=p->next) {
1242: if (code_here >= p->block && code_here < p->block+p->size)
1243: return size + (code_here - p->block);
1244: else
1245: size += p->size;
1246: }
1247: #endif /* !defined(NO_DYNAMIC) */
1248: return 0;
1249: }
1250:
1251: Label decompile_code(Label _code)
1252: {
1253: #ifdef NO_DYNAMIC
1254: return _code;
1255: #else /* !defined(NO_DYNAMIC) */
1256: Cell i;
1257: struct code_block_list *p;
1258: Address code=_code;
1259:
1260: /* first, check if we are in code at all */
1261: for (p = code_block_list;; p = p->next) {
1262: if (p == NULL)
1263: return code;
1264: if (code >= p->block && code < p->block+p->size)
1265: break;
1266: }
1267: /* reverse order because NOOP might match other prims */
1268: for (i=npriminfos-1; i>DOER_MAX; i--) {
1269: PrimInfo *pi=decomp_prims[i];
1270: if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1271: return vm_prims[super2[super_costs[pi-priminfos].offset]];
1272: /* return pi->start;*/
1273: }
1274: return code;
1275: #endif /* !defined(NO_DYNAMIC) */
1276: }
1277:
1278: #ifdef NO_IP
1279: int nbranchinfos=0;
1280:
1281: struct branchinfo {
1282: Label **targetpp; /* **(bi->targetpp) is the target */
1283: Cell *addressptr; /* store the target here */
1284: } branchinfos[100000];
1285:
1286: int ndoesexecinfos=0;
1287: struct doesexecinfo {
1288: int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1289: Label *targetp; /*target for branch (because this is not in threaded code)*/
1290: Cell *xt; /* cfa of word whose does-code needs calling */
1291: } doesexecinfos[10000];
1292:
1293: static void set_rel_target(Cell *source, Label target)
1294: {
1295: *source = ((Cell)target)-(((Cell)source)+4);
1296: }
1297:
1298: static void register_branchinfo(Label source, Cell *targetpp)
1299: {
1300: struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1301: bi->targetpp = (Label **)targetpp;
1302: bi->addressptr = (Cell *)source;
1303: nbranchinfos++;
1304: }
1305:
1306: static Address compile_prim1arg(PrimNum p, Cell **argp)
1307: {
1308: Address old_code_here=append_prim(p);
1309:
1310: assert(vm_prims[p]==priminfos[p].start);
1311: *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
1312: return old_code_here;
1313: }
1314:
1315: static Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
1316: {
1317: PrimInfo *pi = &priminfos[N_call2];
1318: Address old_code_here = append_prim(N_call2);
1319:
1320: *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
1321: register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
1322: return old_code_here;
1323: }
1324: #endif
1325:
1326: void finish_code(void)
1327: {
1328: #ifdef NO_IP
1329: Cell i;
1330:
1331: compile_prim1(NULL);
1332: for (i=0; i<ndoesexecinfos; i++) {
1333: struct doesexecinfo *dei = &doesexecinfos[i];
1334: dei->targetp = (Label *)DOES_CODE1((dei->xt));
1335: branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
1336: }
1337: ndoesexecinfos = 0;
1338: for (i=0; i<nbranchinfos; i++) {
1339: struct branchinfo *bi=&branchinfos[i];
1340: set_rel_target(bi->addressptr, **(bi->targetpp));
1341: }
1342: nbranchinfos = 0;
1343: #else
1344: compile_prim1(NULL);
1345: #endif
1346: flush_to_here();
1347: }
1348:
1349: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1350: #ifdef NO_IP
1351: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1352: /* compile prim #p dynamically (mod flags etc.) and return start
1353: address of generated code for putting it into the threaded
1354: code. This function is only called if all the associated
1355: inline arguments of p are already in place (at tcp[1] etc.) */
1356: {
1357: PrimInfo *pi=&priminfos[p];
1358: Cell *next_code_target=NULL;
1359: Address codeaddr;
1360: Address primstart;
1361:
1362: assert(p<npriminfos);
1363: if (p==N_execute || p==N_perform || p==N_lit_perform) {
1364: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1365: primstart = append_prim(p);
1366: goto other_prim;
1367: } else if (p==N_call) {
1368: codeaddr = compile_call2(tcp+1, &next_code_target);
1369: } else if (p==N_does_exec) {
1370: struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1371: Cell *arg;
1372: codeaddr = compile_prim1arg(N_lit,&arg);
1373: *arg = (Cell)PFA(tcp[1]);
1374: /* we cannot determine the callee now (last_start[1] may be a
1375: forward reference), so just register an arbitrary target, and
1376: register in dei that we need to fix this before resolving
1377: branches */
1378: dei->branchinfo = nbranchinfos;
1379: dei->xt = (Cell *)(tcp[1]);
1380: compile_call2(0, &next_code_target);
1381: } else if (!is_relocatable(p)) {
1382: Cell *branch_target;
1383: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1384: compile_prim1arg(N_branch,&branch_target);
1385: set_rel_target(branch_target,vm_prims[p]);
1386: } else {
1387: unsigned j;
1388:
1389: codeaddr = primstart = append_prim(p);
1390: other_prim:
1391: for (j=0; j<pi->nimmargs; j++) {
1392: struct immarg *ia = &(pi->immargs[j]);
1393: Cell *argp = tcp + pi->nimmargs - j;
1394: Cell argval = *argp; /* !! specific to prims */
1395: if (ia->rel) { /* !! assumption: relative refs are branches */
1396: register_branchinfo(primstart + ia->offset, argp);
1397: } else /* plain argument */
1398: *(Cell *)(primstart + ia->offset) = argval;
1399: }
1400: }
1401: if (next_code_target!=NULL)
1402: *next_code_target = (Cell)code_here;
1403: return (Cell)codeaddr;
1404: }
1405: #else /* !defined(NO_IP) */
1406: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1407: /* compile prim #p dynamically (mod flags etc.) and return start
1408: address of generated code for putting it into the threaded code */
1409: {
1410: Cell static_prim = (Cell)vm_prims[p];
1411: #if defined(NO_DYNAMIC)
1412: return static_prim;
1413: #else /* !defined(NO_DYNAMIC) */
1414: Address old_code_here;
1415:
1416: if (no_dynamic)
1417: return static_prim;
1418: if (p>=npriminfos || !is_relocatable(p)) {
1419: append_jump();
1420: return static_prim;
1421: }
1422: old_code_here = append_prim(p);
1423: last_jump = p;
1424: if (priminfos[p].superend)
1425: append_jump();
1426: return (Cell)old_code_here;
1427: #endif /* !defined(NO_DYNAMIC) */
1428: }
1429: #endif /* !defined(NO_IP) */
1430: #endif
1431:
1432: #ifndef NO_DYNAMIC
1433: static int cost_codesize(int prim)
1434: {
1435: return priminfos[prim].length;
1436: }
1437: #endif
1438:
1439: static int cost_ls(int prim)
1440: {
1441: struct cost *c = super_costs+prim;
1442:
1443: return c->loads + c->stores;
1444: }
1445:
1446: static int cost_lsu(int prim)
1447: {
1448: struct cost *c = super_costs+prim;
1449:
1450: return c->loads + c->stores + c->updates;
1451: }
1452:
1453: static int cost_nexts(int prim)
1454: {
1455: return 1;
1456: }
1457:
1458: typedef int Costfunc(int);
1459: Costfunc *ss_cost = /* cost function for optimize_bb */
1460: #ifdef NO_DYNAMIC
1461: cost_lsu;
1462: #else
1463: cost_codesize;
1464: #endif
1465:
1466: struct {
1467: Costfunc *costfunc;
1468: char *metricname;
1469: long sum;
1470: } cost_sums[] = {
1471: #ifndef NO_DYNAMIC
1472: { cost_codesize, "codesize", 0 },
1473: #endif
1474: { cost_ls, "ls", 0 },
1475: { cost_lsu, "lsu", 0 },
1476: { cost_nexts, "nexts", 0 }
1477: };
1478:
1479: #ifndef NO_DYNAMIC
1480: void init_ss_cost(void) {
1481: if (no_dynamic && ss_cost == cost_codesize) {
1482: ss_cost = cost_nexts;
1483: cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
1484: debugp(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
1485: }
1486: }
1487: #endif
1488:
1489: #define MAX_BB 128 /* maximum number of instructions in BB */
1490: #define INF_COST 1000000 /* infinite cost */
1491: #define CANONICAL_STATE 0
1492:
1493: struct waypoint {
1494: int cost; /* the cost from here to the end */
1495: PrimNum inst; /* the inst used from here to the next waypoint */
1496: char relocatable; /* the last non-transition was relocatable */
1497: char no_transition; /* don't use the next transition (relocatability)
1498: * or this transition (does not change state) */
1499: };
1500:
1501: struct tpa_state { /* tree parsing automaton (like) state */
1502: /* labeling is back-to-front */
1503: struct waypoint *inst; /* in front of instruction */
1504: struct waypoint *trans; /* in front of instruction and transition */
1505: };
1506:
1507: struct tpa_state *termstate = NULL; /* initialized in loader() */
1508:
1509: /* statistics about tree parsing (lazyburg) stuff */
1510: long lb_basic_blocks = 0;
1511: long lb_labeler_steps = 0;
1512: long lb_labeler_automaton = 0;
1513: long lb_labeler_dynprog = 0;
1514: long lb_newstate_equiv = 0;
1515: long lb_newstate_new = 0;
1516: long lb_applicable_base_rules = 0;
1517: long lb_applicable_chain_rules = 0;
1518:
1519: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1520: static void init_waypoints(struct waypoint ws[])
1521: {
1522: int k;
1523:
1524: for (k=0; k<maxstates; k++)
1525: ws[k].cost=INF_COST;
1526: }
1527:
1528: static struct tpa_state *empty_tpa_state()
1529: {
1530: struct tpa_state *s = malloc(sizeof(struct tpa_state));
1531:
1532: s->inst = calloc(maxstates,sizeof(struct waypoint));
1533: init_waypoints(s->inst);
1534: s->trans = calloc(maxstates,sizeof(struct waypoint));
1535: /* init_waypoints(s->trans);*/
1536: return s;
1537: }
1538:
1539: static void transitions(struct tpa_state *t)
1540: {
1541: int k;
1542: struct super_state *l;
1543:
1544: for (k=0; k<maxstates; k++) {
1545: t->trans[k] = t->inst[k];
1546: t->trans[k].no_transition = 1;
1547: }
1548: for (l = state_transitions; l != NULL; l = l->next) {
1549: PrimNum s = l->super;
1550: int jcost;
1551: struct cost *c=super_costs+s;
1552: struct waypoint *wi=&(t->trans[c->state_in]);
1553: struct waypoint *wo=&(t->inst[c->state_out]);
1554: lb_applicable_chain_rules++;
1555: if (wo->cost == INF_COST)
1556: continue;
1557: jcost = wo->cost + ss_cost(s);
1558: if (jcost <= wi->cost) {
1559: wi->cost = jcost;
1560: wi->inst = s;
1561: wi->relocatable = wo->relocatable;
1562: wi->no_transition = 0;
1563: /* if (ss_greedy) wi->cost = wo->cost ? */
1564: }
1565: }
1566: }
1567:
1568: static struct tpa_state *make_termstate()
1569: {
1570: struct tpa_state *s = empty_tpa_state();
1571:
1572: s->inst[CANONICAL_STATE].cost = 0;
1573: transitions(s);
1574: return s;
1575: }
1576: #endif
1577:
1578: #define TPA_SIZE 16384
1579:
1580: struct tpa_entry {
1581: struct tpa_entry *next;
1582: PrimNum inst;
1583: struct tpa_state *state_behind; /* note: brack-to-front labeling */
1584: struct tpa_state *state_infront; /* note: brack-to-front labeling */
1585: } *tpa_table[TPA_SIZE];
1586:
1587: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1588: static Cell hash_tpa(PrimNum p, struct tpa_state *t)
1589: {
1590: UCell it = (UCell )t;
1591: return (p+it+(it>>14))&(TPA_SIZE-1);
1592: }
1593:
1594: static struct tpa_state **lookup_tpa(PrimNum p, struct tpa_state *t2)
1595: {
1596: int hash=hash_tpa(p, t2);
1597: struct tpa_entry *te = tpa_table[hash];
1598:
1599: if (tpa_noautomaton) {
1600: static struct tpa_state *t;
1601: t = NULL;
1602: return &t;
1603: }
1604: for (; te!=NULL; te = te->next) {
1605: if (p == te->inst && t2 == te->state_behind)
1606: return &(te->state_infront);
1607: }
1608: te = (struct tpa_entry *)malloc(sizeof(struct tpa_entry));
1609: te->next = tpa_table[hash];
1610: te->inst = p;
1611: te->state_behind = t2;
1612: te->state_infront = NULL;
1613: tpa_table[hash] = te;
1614: return &(te->state_infront);
1615: }
1616:
1617: static void tpa_state_normalize(struct tpa_state *t)
1618: {
1619: /* normalize so cost of canonical state=0; this may result in
1620: negative costs for some states */
1621: int d = t->inst[CANONICAL_STATE].cost;
1622: int i;
1623:
1624: for (i=0; i<maxstates; i++) {
1625: if (t->inst[i].cost != INF_COST)
1626: t->inst[i].cost -= d;
1627: if (t->trans[i].cost != INF_COST)
1628: t->trans[i].cost -= d;
1629: }
1630: }
1631:
1632: static int tpa_state_equivalent(struct tpa_state *t1, struct tpa_state *t2)
1633: {
1634: return (memcmp(t1->inst, t2->inst, maxstates*sizeof(struct waypoint)) == 0 &&
1635: memcmp(t1->trans,t2->trans,maxstates*sizeof(struct waypoint)) == 0);
1636: }
1637: #endif
1638:
1639: struct tpa_state_entry {
1640: struct tpa_state_entry *next;
1641: struct tpa_state *state;
1642: } *tpa_state_table[TPA_SIZE];
1643:
1644: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1645: static Cell hash_tpa_state(struct tpa_state *t)
1646: {
1647: int *ti = (int *)(t->inst);
1648: int *tt = (int *)(t->trans);
1649: int r=0;
1650: int i;
1651:
1652: for (i=0; ti+i < (int *)(t->inst+maxstates); i++)
1653: r += ti[i]+tt[i];
1654: return (r+(r>>14)+(r>>22)) & (TPA_SIZE-1);
1655: }
1656:
1657: static struct tpa_state *lookup_tpa_state(struct tpa_state *t)
1658: {
1659: Cell hash = hash_tpa_state(t);
1660: struct tpa_state_entry *te = tpa_state_table[hash];
1661: struct tpa_state_entry *tn;
1662:
1663: if (!tpa_noequiv) {
1664: for (; te!=NULL; te = te->next) {
1665: if (tpa_state_equivalent(t, te->state)) {
1666: lb_newstate_equiv++;
1667: free(t->inst);
1668: free(t->trans);
1669: free(t);
1670: return te->state;
1671: }
1672: }
1673: tn = (struct tpa_state_entry *)malloc(sizeof(struct tpa_state_entry));
1674: tn->next = te;
1675: tn->state = t;
1676: tpa_state_table[hash] = tn;
1677: }
1678: lb_newstate_new++;
1679: if (tpa_trace)
1680: fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
1681: return t;
1682: }
1683:
1684: /* use dynamic programming to find the shortest paths within the basic
1685: block origs[0..ninsts-1] and rewrite the instructions pointed to by
1686: instps to use it */
1687: static void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
1688: {
1689: int i,j;
1690: struct tpa_state *ts[ninsts+1];
1691: int nextdyn, nextstate, no_transition;
1692: Address old_code_area;
1693:
1694: lb_basic_blocks++;
1695: ts[ninsts] = termstate;
1696: #ifndef NO_DYNAMIC
1697: if (print_sequences) {
1698: for (i=0; i<ninsts; i++)
1699: #if defined(BURG_FORMAT)
1700: fprintf(stderr, "op%d ", super_costs[origs[i]].offset);
1701: #else
1702: fprintf(stderr, "%s ", prim_names[origs[i]]);
1703: #endif
1704: fprintf(stderr, "\n");
1705: }
1706: #endif
1707: for (i=ninsts-1; i>=0; i--) {
1708: struct tpa_state **tp = lookup_tpa(origs[i],ts[i+1]);
1709: struct tpa_state *t = *tp;
1710: lb_labeler_steps++;
1711: if (t) {
1712: ts[i] = t;
1713: lb_labeler_automaton++;
1714: }
1715: else {
1716: lb_labeler_dynprog++;
1717: ts[i] = empty_tpa_state();
1718: for (j=1; j<=max_super && i+j<=ninsts; j++) {
1719: struct super_state **superp = lookup_super(origs+i, j);
1720: if (superp!=NULL) {
1721: struct super_state *supers = *superp;
1722: for (; supers!=NULL; supers = supers->next) {
1723: PrimNum s = supers->super;
1724: int jcost;
1725: struct cost *c=super_costs+s;
1726: struct waypoint *wi=&(ts[i]->inst[c->state_in]);
1727: struct waypoint *wo=&(ts[i+j]->trans[c->state_out]);
1728: int no_transition = wo->no_transition;
1729: lb_applicable_base_rules++;
1730: if (!(is_relocatable(s)) && !wo->relocatable) {
1731: wo=&(ts[i+j]->inst[c->state_out]);
1732: no_transition=1;
1733: }
1734: if (wo->cost == INF_COST)
1735: continue;
1736: jcost = wo->cost + ss_cost(s);
1737: if (jcost <= wi->cost) {
1738: wi->cost = jcost;
1739: wi->inst = s;
1740: wi->relocatable = is_relocatable(s);
1741: wi->no_transition = no_transition;
1742: /* if (ss_greedy) wi->cost = wo->cost ? */
1743: }
1744: }
1745: }
1746: }
1747: transitions(ts[i]);
1748: tpa_state_normalize(ts[i]);
1749: *tp = ts[i] = lookup_tpa_state(ts[i]);
1750: if (tpa_trace)
1751: fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
1752: }
1753: }
1754: /* now rewrite the instructions */
1755: reserve_code_super(origs,ninsts);
1756: old_code_area = code_area;
1757: nextdyn=0;
1758: nextstate=CANONICAL_STATE;
1759: no_transition = ((!ts[0]->trans[nextstate].relocatable)
1760: ||ts[0]->trans[nextstate].no_transition);
1761: for (i=0; i<ninsts; i++) {
1762: Cell tc=0, tc2;
1763: if (i==nextdyn) {
1764: if (!no_transition) {
1765: /* process trans */
1766: PrimNum p = ts[i]->trans[nextstate].inst;
1767: struct cost *c = super_costs+p;
1768: assert(ts[i]->trans[nextstate].cost != INF_COST);
1769: assert(c->state_in==nextstate);
1770: tc = compile_prim_dyn(p,NULL);
1771: nextstate = c->state_out;
1772: }
1773: {
1774: /* process inst */
1775: PrimNum p = ts[i]->inst[nextstate].inst;
1776: struct cost *c=super_costs+p;
1777: assert(c->state_in==nextstate);
1778: assert(ts[i]->inst[nextstate].cost != INF_COST);
1779: #if defined(GFORTH_DEBUGGING)
1780: assert(p == origs[i]);
1781: #endif
1782: tc2 = compile_prim_dyn(p,instps[i]);
1783: if (no_transition || !is_relocatable(p))
1784: /* !! actually what we care about is if and where
1785: * compile_prim_dyn() puts NEXTs */
1786: tc=tc2;
1787: no_transition = ts[i]->inst[nextstate].no_transition;
1788: nextstate = c->state_out;
1789: nextdyn += c->length;
1790: }
1791: } else {
1792: #if defined(GFORTH_DEBUGGING)
1793: assert(0);
1794: #endif
1795: tc=0;
1796: /* tc= (Cell)vm_prims[ts[i]->inst[CANONICAL_STATE].inst]; */
1797: }
1798: *(instps[i]) = tc;
1799: }
1800: if (!no_transition) {
1801: PrimNum p = ts[i]->trans[nextstate].inst;
1802: struct cost *c = super_costs+p;
1803: assert(c->state_in==nextstate);
1804: assert(ts[i]->trans[nextstate].cost != INF_COST);
1805: assert(i==nextdyn);
1806: (void)compile_prim_dyn(p,NULL);
1807: nextstate = c->state_out;
1808: }
1809: assert(nextstate==CANONICAL_STATE);
1810: assert(code_area==old_code_area); /* does reserve_code_super() work? */
1811: }
1812: #endif
1813:
1814: /* compile *start, possibly rewriting it into a static and/or dynamic
1815: superinstruction */
1816: void compile_prim1(Cell *start)
1817: {
1818: #if defined(DOUBLY_INDIRECT)
1819: Label prim;
1820:
1821: if (start==NULL)
1822: return;
1823: prim = (Label)*start;
1824: if (prim<((Label)(xts+DOER_MAX)) || prim>((Label)(xts+npriminfos))) {
1825: fprintf(stderr,"compile_prim encountered xt %p\n", prim);
1826: *start=(Cell)prim;
1827: return;
1828: } else {
1829: *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
1830: return;
1831: }
1832: #elif defined(INDIRECT_THREADED)
1833: return;
1834: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1835: /* !! does not work, for unknown reasons; but something like this is
1836: probably needed to ensure that we don't call compile_prim_dyn
1837: before the inline arguments are there */
1838: static Cell *instps[MAX_BB];
1839: static PrimNum origs[MAX_BB];
1840: static int ninsts=0;
1841: PrimNum prim_num;
1842:
1843: if (start==NULL || ninsts >= MAX_BB ||
1844: (ninsts>0 && superend[origs[ninsts-1]])) {
1845: /* after bb, or at the start of the next bb */
1846: optimize_rewrite(instps,origs,ninsts);
1847: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
1848: ninsts=0;
1849: if (start==NULL) {
1850: align_code();
1851: return;
1852: }
1853: }
1854: prim_num = ((Xt)*start)-vm_prims;
1855: if(prim_num >= npriminfos) {
1856: optimize_rewrite(instps,origs,ninsts);
1857: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
1858: ninsts=0;
1859: return;
1860: }
1861: assert(ninsts<MAX_BB);
1862: instps[ninsts] = start;
1863: origs[ninsts] = prim_num;
1864: ninsts++;
1865: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1866: }
1867:
1868: #ifndef STANDALONE
1869: Address gforth_loader(FILE *imagefile, char* filename)
1870: /* returns the address of the image proper (after the preamble) */
1871: {
1872: ImageHeader header;
1873: Address image;
1874: Address imp; /* image+preamble */
1875: Char magic[8];
1876: char magic7; /* size byte of magic number */
1877: Cell preamblesize=0;
1878: Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1879: UCell check_sum;
1880: Cell ausize = ((RELINFOBITS == 8) ? 0 :
1881: (RELINFOBITS == 16) ? 1 :
1882: (RELINFOBITS == 32) ? 2 : 3);
1883: Cell charsize = ((sizeof(Char) == 1) ? 0 :
1884: (sizeof(Char) == 2) ? 1 :
1885: (sizeof(Char) == 4) ? 2 : 3) + ausize;
1886: Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
1887: (sizeof(Cell) == 2) ? 1 :
1888: (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1889: Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
1890: #ifdef WORDS_BIGENDIAN
1891: 0
1892: #else
1893: 1
1894: #endif
1895: ;
1896:
1897: vm_prims = gforth_engine(0,0,0,0,0 sr_call);
1898: check_prims(vm_prims);
1899: prepare_super_table();
1900: #ifndef DOUBLY_INDIRECT
1901: #ifdef PRINT_SUPER_LENGTHS
1902: print_super_lengths();
1903: #endif
1904: check_sum = checksum(vm_prims);
1905: #else /* defined(DOUBLY_INDIRECT) */
1906: check_sum = (UCell)vm_prims;
1907: #endif /* defined(DOUBLY_INDIRECT) */
1908: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1909: termstate = make_termstate();
1910: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1911:
1912: do {
1913: if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1914: fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
1915: progname, filename);
1916: exit(1);
1917: }
1918: preamblesize+=8;
1919: } while(memcmp(magic,"Gforth3",7));
1920: magic7 = magic[7];
1921: if (debug) {
1922: magic[7]='\0';
1923: fprintf(stderr,"Magic found: %s ", magic);
1924: print_sizes(magic7);
1925: }
1926:
1927: if (magic7 != sizebyte)
1928: {
1929: fprintf(stderr,"This image is: ");
1930: print_sizes(magic7);
1931: fprintf(stderr,"whereas the machine is ");
1932: print_sizes(sizebyte);
1933: exit(-2);
1934: };
1935:
1936: fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1937:
1938: set_stack_sizes(&header);
1939:
1940: #if HAVE_GETPAGESIZE
1941: pagesize=getpagesize(); /* Linux/GNU libc offers this */
1942: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
1943: pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
1944: #elif PAGESIZE
1945: pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
1946: #endif
1947: debugp(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1948:
1949: image = dict_alloc_read(imagefile, preamblesize+header.image_size,
1950: dictsize, data_offset);
1951: imp=image+preamblesize;
1952:
1953: alloc_stacks((ImageHeader *)imp);
1954: if (clear_dictionary)
1955: memset(imp+header.image_size, 0, dictsize-header.image_size-preamblesize);
1956: if(header.base==0 || header.base == (Address)0x100) {
1957: Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
1958: Char reloc_bits[reloc_size];
1959: fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1960: fread(reloc_bits, 1, reloc_size, imagefile);
1961: gforth_relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1962: #if 0
1963: { /* let's see what the relocator did */
1964: FILE *snapshot=fopen("snapshot.fi","wb");
1965: fwrite(image,1,imagesize,snapshot);
1966: fclose(snapshot);
1967: }
1968: #endif
1969: }
1970: else if(header.base!=imp) {
1971: fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
1972: progname, (unsigned long)header.base, (unsigned long)imp);
1973: exit(1);
1974: }
1975: if (header.checksum==0)
1976: ((ImageHeader *)imp)->checksum=check_sum;
1977: else if (header.checksum != check_sum) {
1978: fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
1979: progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
1980: exit(1);
1981: }
1982: #ifdef DOUBLY_INDIRECT
1983: ((ImageHeader *)imp)->xt_base = xts;
1984: #endif
1985: fclose(imagefile);
1986:
1987: /* unnecessary, except maybe for CODE words */
1988: /* FLUSH_ICACHE(imp, header.image_size);*/
1989:
1990: return imp;
1991: }
1992: #endif
1993:
1994: /* pointer to last '/' or '\' in file, 0 if there is none. */
1995: static char *onlypath(char *filename)
1996: {
1997: return strrchr(filename, DIRSEP);
1998: }
1999:
2000: static FILE *openimage(char *fullfilename)
2001: {
2002: FILE *image_file;
2003: char * expfilename = tilde_cstr((Char *)fullfilename, strlen(fullfilename), 1);
2004:
2005: image_file=fopen(expfilename,"rb");
2006: if (image_file!=NULL && debug)
2007: fprintf(stderr, "Opened image file: %s\n", expfilename);
2008: return image_file;
2009: }
2010:
2011: /* try to open image file concat(path[0:len],imagename) */
2012: static FILE *checkimage(char *path, int len, char *imagename)
2013: {
2014: int dirlen=len;
2015: char fullfilename[dirlen+strlen((char *)imagename)+2];
2016:
2017: memcpy(fullfilename, path, dirlen);
2018: if (fullfilename[dirlen-1]!=DIRSEP)
2019: fullfilename[dirlen++]=DIRSEP;
2020: strcpy(fullfilename+dirlen,imagename);
2021: return openimage(fullfilename);
2022: }
2023:
2024: static FILE * open_image_file(char * imagename, char * path)
2025: {
2026: FILE * image_file=NULL;
2027: char *origpath=path;
2028:
2029: if(strchr(imagename, DIRSEP)==NULL) {
2030: /* first check the directory where the exe file is in !! 01may97jaw */
2031: if (onlypath(progname))
2032: image_file=checkimage(progname, onlypath(progname)-progname, imagename);
2033: if (!image_file)
2034: do {
2035: char *pend=strchr(path, PATHSEP);
2036: if (pend==NULL)
2037: pend=path+strlen(path);
2038: if (strlen(path)==0) break;
2039: image_file=checkimage(path, pend-path, imagename);
2040: path=pend+(*pend==PATHSEP);
2041: } while (image_file==NULL);
2042: } else {
2043: image_file=openimage(imagename);
2044: }
2045:
2046: if (!image_file) {
2047: fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
2048: progname, imagename, origpath);
2049: exit(1);
2050: }
2051:
2052: return image_file;
2053: }
2054: #endif
2055:
2056: #ifdef STANDALONE_ALLOC
2057: Address gforth_alloc(Cell size)
2058: {
2059: Address r;
2060: /* leave a little room (64B) for stack underflows */
2061: if ((r = malloc(size+64))==NULL) {
2062: perror(progname);
2063: exit(1);
2064: }
2065: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
2066: debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
2067: return r;
2068: }
2069: #endif
2070:
2071: #ifdef HAS_OS
2072: static UCell convsize(char *s, UCell elemsize)
2073: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
2074: of bytes. the letter at the end indicates the unit, where e stands
2075: for the element size. default is e */
2076: {
2077: char *endp;
2078: UCell n,m;
2079:
2080: m = elemsize;
2081: n = strtoul(s,&endp,0);
2082: if (endp!=NULL) {
2083: if (strcmp(endp,"b")==0)
2084: m=1;
2085: else if (strcmp(endp,"k")==0)
2086: m=1024;
2087: else if (strcmp(endp,"M")==0)
2088: m=1024*1024;
2089: else if (strcmp(endp,"G")==0)
2090: m=1024*1024*1024;
2091: else if (strcmp(endp,"T")==0) {
2092: #if (SIZEOF_CHAR_P > 4)
2093: m=1024L*1024*1024*1024;
2094: #else
2095: fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
2096: exit(1);
2097: #endif
2098: } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
2099: fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
2100: exit(1);
2101: }
2102: }
2103: return n*m;
2104: }
2105:
2106: enum {
2107: ss_number = 256,
2108: ss_states,
2109: ss_min_codesize,
2110: ss_min_ls,
2111: ss_min_lsu,
2112: ss_min_nexts,
2113: opt_code_block_size,
2114: };
2115:
2116: #ifndef STANDALONE
2117: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
2118: {
2119: int c;
2120:
2121: opterr=0;
2122: while (1) {
2123: int option_index=0;
2124: static struct option opts[] = {
2125: {"appl-image", required_argument, NULL, 'a'},
2126: {"image-file", required_argument, NULL, 'i'},
2127: {"dictionary-size", required_argument, NULL, 'm'},
2128: {"data-stack-size", required_argument, NULL, 'd'},
2129: {"return-stack-size", required_argument, NULL, 'r'},
2130: {"fp-stack-size", required_argument, NULL, 'f'},
2131: {"locals-stack-size", required_argument, NULL, 'l'},
2132: {"vm-commit", no_argument, &map_noreserve, 0},
2133: {"path", required_argument, NULL, 'p'},
2134: {"version", no_argument, NULL, 'v'},
2135: {"help", no_argument, NULL, 'h'},
2136: /* put something != 0 into offset_image */
2137: {"offset-image", no_argument, &offset_image, 1},
2138: {"no-offset-im", no_argument, &offset_image, 0},
2139: {"clear-dictionary", no_argument, &clear_dictionary, 1},
2140: {"debug", no_argument, &debug, 1},
2141: {"diag", no_argument, &diag, 1},
2142: {"die-on-signal", no_argument, &die_on_signal, 1},
2143: {"ignore-async-signals", no_argument, &ignore_async_signals, 1},
2144: {"no-super", no_argument, &no_super, 1},
2145: {"no-dynamic", no_argument, &no_dynamic, 1},
2146: {"dynamic", no_argument, &no_dynamic, 0},
2147: {"code-block-size", required_argument, NULL, opt_code_block_size},
2148: {"print-metrics", no_argument, &print_metrics, 1},
2149: {"print-sequences", no_argument, &print_sequences, 1},
2150: {"ss-number", required_argument, NULL, ss_number},
2151: {"ss-states", required_argument, NULL, ss_states},
2152: #ifndef NO_DYNAMIC
2153: {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
2154: #endif
2155: {"ss-min-ls", no_argument, NULL, ss_min_ls},
2156: {"ss-min-lsu", no_argument, NULL, ss_min_lsu},
2157: {"ss-min-nexts", no_argument, NULL, ss_min_nexts},
2158: {"ss-greedy", no_argument, &ss_greedy, 1},
2159: {"tpa-noequiv", no_argument, &tpa_noequiv, 1},
2160: {"tpa-noautomaton", no_argument, &tpa_noautomaton, 1},
2161: {"tpa-trace", no_argument, &tpa_trace, 1},
2162: {0,0,0,0}
2163: /* no-init-file, no-rc? */
2164: };
2165:
2166: c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
2167:
2168: switch (c) {
2169: case EOF: return;
2170: case '?': optind--; return;
2171: case 'a': *imagename = optarg; return;
2172: case 'i': *imagename = optarg; break;
2173: case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
2174: case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
2175: case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
2176: case 'f': fsize = convsize(optarg,sizeof(Float)); break;
2177: case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
2178: case 'p': *path = optarg; break;
2179: case 'o': offset_image = 1; break;
2180: case 'n': offset_image = 0; break;
2181: case 'c': clear_dictionary = 1; break;
2182: case 's': die_on_signal = 1; break;
2183: case 'x': debug = 1; break;
2184: case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
2185: case opt_code_block_size: code_area_size = atoi(optarg); break;
2186: case ss_number: static_super_number = atoi(optarg); break;
2187: case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
2188: #ifndef NO_DYNAMIC
2189: case ss_min_codesize: ss_cost = cost_codesize; break;
2190: #endif
2191: case ss_min_ls: ss_cost = cost_ls; break;
2192: case ss_min_lsu: ss_cost = cost_lsu; break;
2193: case ss_min_nexts: ss_cost = cost_nexts; break;
2194: case 'h':
2195: fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
2196: Engine Options:\n\
2197: --appl-image FILE Equivalent to '--image-file=FILE --'\n\
2198: --clear-dictionary Initialize the dictionary with 0 bytes\n\
2199: --code-block-size=SIZE size of native code blocks [512KB]\n\
2200: -d SIZE, --data-stack-size=SIZE Specify data stack size\n\
2201: --debug Print debugging information during startup\n\
2202: --diag Print diagnostic information during startup\n\
2203: --die-on-signal Exit instead of THROWing some signals\n\
2204: --dynamic Use dynamic native code\n\
2205: -f SIZE, --fp-stack-size=SIZE Specify floating point stack size\n\
2206: -h, --help Print this message and exit\n\
2207: --ignore-async-signals Ignore instead of THROWing async. signals\n\
2208: -i FILE, --image-file=FILE Use image FILE instead of `gforth.fi'\n\
2209: -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
2210: -m SIZE, --dictionary-size=SIZE Specify Forth dictionary size\n\
2211: --no-dynamic Use only statically compiled primitives\n\
2212: --no-offset-im Load image at normal position\n\
2213: --no-super No dynamically formed superinstructions\n\
2214: --offset-image Load image at a different position\n\
2215: -p PATH, --path=PATH Search path for finding image and sources\n\
2216: --print-metrics Print some code generation metrics on exit\n\
2217: --print-sequences Print primitive sequences for optimization\n\
2218: -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
2219: --ss-greedy Greedy, not optimal superinst selection\n\
2220: --ss-min-codesize Select superinsts for smallest native code\n\
2221: --ss-min-ls Minimize loads and stores\n\
2222: --ss-min-lsu Minimize loads, stores, and pointer updates\n\
2223: --ss-min-nexts Minimize the number of static superinsts\n\
2224: --ss-number=N Use N static superinsts (default max)\n\
2225: --ss-states=N N states for stack caching (default max)\n\
2226: --tpa-noequiv Automaton without state equivalence\n\
2227: --tpa-noautomaton Dynamic programming only\n\
2228: --tpa-trace Report new states etc.\n\
2229: -v, --version Print engine version and exit\n\
2230: --vm-commit Use OS default for memory overcommit\n\
2231: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
2232: `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
2233: argv[0]);
2234: optind--;
2235: return;
2236: }
2237: }
2238: }
2239: #endif
2240: #endif
2241:
2242: static void print_diag()
2243: {
2244:
2245: #if !defined(HAVE_GETRUSAGE)
2246: fprintf(stderr, "*** missing functionality ***\n"
2247: #ifndef HAVE_GETRUSAGE
2248: " no getrusage -> CPUTIME broken\n"
2249: #endif
2250: );
2251: #endif
2252: if((relocs < nonrelocs) ||
2253: #if defined(BUGGY_LL_CMP) || defined(BUGGY_LL_MUL) || defined(BUGGY_LL_DIV) || defined(BUGGY_LL_ADD) || defined(BUGGY_LL_SHIFT) || defined(BUGGY_LL_D2F) || defined(BUGGY_LL_F2D)
2254: 1
2255: #else
2256: 0
2257: #endif
2258: )
2259: debugp(stderr, "relocs: %d:%d\n", relocs, nonrelocs);
2260: fprintf(stderr, "*** %sperformance problems ***\n%s%s",
2261: #if defined(BUGGY_LL_CMP) || defined(BUGGY_LL_MUL) || defined(BUGGY_LL_DIV) || defined(BUGGY_LL_ADD) || defined(BUGGY_LL_SHIFT) || defined(BUGGY_LL_D2F) || defined(BUGGY_LL_F2D) || !(defined(FORCE_REG) || defined(FORCE_REG_UNNECESSARY)) || defined(BUGGY_LONG_LONG)
2262: "",
2263: #else
2264: "no ",
2265: #endif
2266: #if defined(BUGGY_LL_CMP) || defined(BUGGY_LL_MUL) || defined(BUGGY_LL_DIV) || defined(BUGGY_LL_ADD) || defined(BUGGY_LL_SHIFT) || defined(BUGGY_LL_D2F) || defined(BUGGY_LL_F2D)
2267: " double-cell integer type buggy ->\n "
2268: #ifdef BUGGY_LL_CMP
2269: "double comparisons, "
2270: #endif
2271: #ifdef BUGGY_LL_MUL
2272: "*/MOD */ M* UM* "
2273: #endif
2274: #ifdef BUGGY_LL_DIV
2275: /* currently nothing is affected */
2276: #endif
2277: #ifdef BUGGY_LL_ADD
2278: "M+ D+ D- DNEGATE "
2279: #endif
2280: #ifdef BUGGY_LL_SHIFT
2281: "D2/ "
2282: #endif
2283: #ifdef BUGGY_LL_D2F
2284: "D>F "
2285: #endif
2286: #ifdef BUGGY_LL_F2D
2287: "F>D "
2288: #endif
2289: "\b\b slow\n"
2290: #endif
2291: #if !(defined(FORCE_REG) || defined(FORCE_REG_UNNECESSARY))
2292: " automatic register allocation: performance degradation possible\n"
2293: #endif
2294: "",
2295: (relocs < nonrelocs) ? "no dynamic code generation (--debug for details) -> factor 2 slowdown\n" : "");
2296: }
2297:
2298: #ifdef STANDALONE
2299: Cell data_abort_pc;
2300:
2301: void data_abort_C(void)
2302: {
2303: while(1) {
2304: }
2305: }
2306: #endif
2307:
2308: int main(int argc, char **argv, char **env)
2309: {
2310: #ifdef HAS_OS
2311: char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
2312: #else
2313: char *path = DEFAULTPATH;
2314: #endif
2315: #ifndef INCLUDE_IMAGE
2316: char *imagename="gforth.fi";
2317: FILE *image_file;
2318: Address image;
2319: #endif
2320: int retvalue;
2321: #if 0 && defined(__i386)
2322: /* disabled because the drawbacks may be worse than the benefits */
2323: /* set 387 precision control to use 53-bit mantissae to avoid most
2324: cases of double rounding */
2325: short fpu_control = 0x027f ;
2326: asm("fldcw %0" : : "m"(fpu_control));
2327: #endif /* defined(__i386) */
2328:
2329: #ifdef MACOSX_DEPLOYMENT_TARGET
2330: setenv("MACOSX_DEPLOYMENT_TARGET", MACOSX_DEPLOYMENT_TARGET, 0);
2331: #endif
2332: #ifdef LTDL_LIBRARY_PATH
2333: setenv("LTDL_LIBRARY_PATH", LTDL_LIBRARY_PATH, 0);
2334: #endif
2335: #ifndef STANDALONE
2336: /* buffering of the user output device */
2337: #ifdef _IONBF
2338: if (isatty(fileno(stdout))) {
2339: fflush(stdout);
2340: setvbuf(stdout,NULL,_IONBF,0);
2341: }
2342: #endif
2343: setlocale(LC_ALL, "");
2344: setlocale(LC_NUMERIC, "C");
2345: #else
2346: prep_terminal();
2347: #endif
2348:
2349: progname = argv[0];
2350:
2351: #ifndef STANDALONE
2352: #ifdef HAVE_LIBLTDL
2353: if (lt_dlinit()!=0) {
2354: fprintf(stderr,"%s: lt_dlinit failed", progname);
2355: exit(1);
2356: }
2357: #endif
2358:
2359: #ifdef HAS_OS
2360: gforth_args(argc, argv, &path, &imagename);
2361: #ifndef NO_DYNAMIC
2362: init_ss_cost();
2363: #endif /* !defined(NO_DYNAMIC) */
2364: #endif /* defined(HAS_OS) */
2365: #endif
2366: code_here = ((void *)0)+code_area_size;
2367: #ifdef STANDALONE
2368: image = gforth_engine(0, 0, 0, 0, 0 sr_call);
2369: alloc_stacks((ImageHeader *)image);
2370: #else
2371: image_file = open_image_file(imagename, path);
2372: image = gforth_loader(image_file, imagename);
2373: #endif
2374: gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
2375:
2376: if (diag)
2377: print_diag();
2378: {
2379: char path2[strlen(path)+1];
2380: char *p1, *p2;
2381: Cell environ[]= {
2382: (Cell)argc-(optind-1),
2383: (Cell)(argv+(optind-1)),
2384: (Cell)strlen(path),
2385: (Cell)path2};
2386: argv[optind-1] = progname;
2387: /*
2388: for (i=0; i<environ[0]; i++)
2389: printf("%s\n", ((char **)(environ[1]))[i]);
2390: */
2391: /* make path OS-independent by replacing path separators with NUL */
2392: for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
2393: if (*p1==PATHSEP)
2394: *p2 = '\0';
2395: else
2396: *p2 = *p1;
2397: *p2='\0';
2398: retvalue = gforth_go(image, 4, environ);
2399: #if defined(SIGPIPE) && !defined(STANDALONE)
2400: bsd_signal(SIGPIPE, SIG_IGN);
2401: #endif
2402: #ifdef VM_PROFILING
2403: vm_print_profile(stderr);
2404: #endif
2405: deprep_terminal();
2406: #ifndef STANDALONE
2407: #ifdef HAVE_LIBLTDL
2408: if (lt_dlexit()!=0)
2409: fprintf(stderr,"%s: lt_dlexit failed", progname);
2410: #endif
2411: #endif
2412: }
2413: if (print_metrics) {
2414: int i;
2415: fprintf(stderr, "code size = %8ld\n", dyncodesize());
2416: #ifndef STANDALONE
2417: for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
2418: fprintf(stderr, "metric %8s: %8ld\n",
2419: cost_sums[i].metricname, cost_sums[i].sum);
2420: #endif
2421: fprintf(stderr,"lb_basic_blocks = %ld\n", lb_basic_blocks);
2422: fprintf(stderr,"lb_labeler_steps = %ld\n", lb_labeler_steps);
2423: fprintf(stderr,"lb_labeler_automaton = %ld\n", lb_labeler_automaton);
2424: fprintf(stderr,"lb_labeler_dynprog = %ld\n", lb_labeler_dynprog);
2425: fprintf(stderr,"lb_newstate_equiv = %ld\n", lb_newstate_equiv);
2426: fprintf(stderr,"lb_newstate_new = %ld\n", lb_newstate_new);
2427: fprintf(stderr,"lb_applicable_base_rules = %ld\n", lb_applicable_base_rules);
2428: fprintf(stderr,"lb_applicable_chain_rules = %ld\n", lb_applicable_chain_rules);
2429: }
2430: if (tpa_trace) {
2431: fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
2432: fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
2433: }
2434: return retvalue;
2435: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>