1: /* command line interpretation, image loading etc. for Gforth
2:
3:
4: Copyright (C) 1995,1996,1997,1998,2000,2003,2004,2005,2006,2007,2008,2009,2010,2011 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: __thread Cell *gforth_SP;
66: __thread Float *gforth_FP;
67: __thread Address gforth_UP=NULL;
68: __thread Cell *gforth_RP;
69: __thread Address gforth_LP;
70:
71: #ifdef HAS_FFCALL
72:
73: #include <callback.h>
74:
75: __thread va_alist gforth_clist;
76:
77: void gforth_callback(Xt* fcall, void * alist)
78: {
79: /* save global valiables */
80: Cell *rp = gforth_RP;
81: Cell *sp = gforth_SP;
82: Float *fp = gforth_FP;
83: Address lp = gforth_LP;
84: va_alist clist = gforth_clist;
85:
86: gforth_clist = (va_alist)alist;
87:
88: gforth_engine(fcall, sp, rp, fp, lp sr_call);
89:
90: /* restore global variables */
91: gforth_RP = rp;
92: gforth_SP = sp;
93: gforth_FP = fp;
94: gforth_LP = lp;
95: gforth_clist = clist;
96: }
97: #endif
98:
99: #ifdef GFORTH_DEBUGGING
100: /* define some VM registers as global variables, so they survive exceptions;
101: global register variables are not up to the task (according to the
102: GNU C manual) */
103: #if defined(GLOBALS_NONRELOC)
104: saved_regs saved_regs_v;
105: __thread saved_regs *saved_regs_p = &saved_regs_v;
106: #else /* !defined(GLOBALS_NONRELOC) */
107: __thread Xt *saved_ip;
108: __thread Cell *rp;
109: #endif /* !defined(GLOBALS_NONRELOC) */
110: #endif /* !defined(GFORTH_DEBUGGING) */
111:
112: #ifdef NO_IP
113: Label next_code;
114: #endif
115:
116: #ifdef HAS_FILE
117: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
118: char* pfileattr[6]={"r","r","r+","r+","w","w"};
119:
120: #ifndef O_BINARY
121: #define O_BINARY 0
122: #endif
123: #ifndef O_TEXT
124: #define O_TEXT 0
125: #endif
126:
127: int ufileattr[6]= {
128: O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
129: O_RDWR |O_BINARY, O_RDWR |O_BINARY,
130: O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
131: #endif
132: /* end global vars for engine.c */
133:
134: #define PRIM_VERSION 1
135: /* increment this whenever the primitives change in an incompatible way */
136:
137: #ifndef DEFAULTPATH
138: # define DEFAULTPATH "."
139: #endif
140:
141: #ifdef MSDOS
142: jmp_buf throw_jmp_buf;
143: #endif
144:
145: #if defined(DOUBLY_INDIRECT)
146: # define CFA(n) ({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
147: #else
148: # define CFA(n) ((Cell)(symbols+((n)&~0x4000UL)))
149: #endif
150:
151: #define maxaligned(n) (typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
152:
153: static UCell dictsize=0;
154: static UCell dsize=0;
155: static UCell rsize=0;
156: static UCell fsize=0;
157: static UCell lsize=0;
158: int offset_image=0;
159: int die_on_signal=0;
160: int ignore_async_signals=0;
161: #ifndef INCLUDE_IMAGE
162: static int clear_dictionary=0;
163: UCell pagesize=1;
164: char *progname;
165: #else
166: char *progname = "gforth";
167: int optind = 1;
168: #endif
169: #ifndef MAP_NORESERVE
170: #define MAP_NORESERVE 0
171: #endif
172: /* IF you have an old Cygwin, this may help:
173: #ifdef __CYGWIN__
174: #define MAP_NORESERVE 0
175: #endif
176: */
177: static int map_noreserve=MAP_NORESERVE;
178:
179: #define CODE_BLOCK_SIZE (512*1024) /* !! overflow handling for -native */
180: Address code_area=0;
181: Cell code_area_size = CODE_BLOCK_SIZE;
182: Address code_here; /* does for code-area what HERE does for the dictionary */
183: Address start_flush=NULL; /* start of unflushed code */
184: Cell last_jump=0; /* if the last prim was compiled without jump, this
185: is it's number, otherwise this contains 0 */
186:
187: static int no_super=0; /* true if compile_prim should not fuse prims */
188: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
189: dynamically */
190: static int print_metrics=0; /* if true, print metrics on exit */
191: static int static_super_number = 10000; /* number of ss used if available */
192: #define MAX_STATE 9 /* maximum number of states */
193: static int maxstates = MAX_STATE; /* number of states for stack caching */
194: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
195: static int diag = 0; /* if true: print diagnostic informations */
196: static int tpa_noequiv = 0; /* if true: no state equivalence checking */
197: static int tpa_noautomaton = 0; /* if true: no tree parsing automaton */
198: static int tpa_trace = 0; /* if true: data for line graph of new states etc. */
199: static int print_sequences = 0; /* print primitive sequences for optimization */
200: static int relocs = 0;
201: static int nonrelocs = 0;
202:
203: #ifdef HAS_DEBUG
204: int debug=0;
205: # define debugp(x...) do { if (debug) fprintf(x); } while (0)
206: #else
207: # define perror(x...)
208: # define fprintf(x...)
209: # define debugp(x...)
210: #endif
211:
212: ImageHeader *gforth_header;
213: Label *vm_prims;
214: #ifdef DOUBLY_INDIRECT
215: Label *xts; /* same content as vm_prims, but should only be used for xts */
216: #endif
217:
218: #ifndef NO_DYNAMIC
219: #ifndef CODE_ALIGNMENT
220: #define CODE_ALIGNMENT 0
221: #endif
222:
223: #define MAX_IMMARGS 2
224:
225: typedef struct {
226: Label start; /* NULL if not relocatable */
227: Cell length; /* only includes the jump iff superend is true*/
228: Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
229: char superend; /* true if primitive ends superinstruction, i.e.,
230: unconditional branch, execute, etc. */
231: Cell nimmargs;
232: struct immarg {
233: Cell offset; /* offset of immarg within prim */
234: char rel; /* true if immarg is relative */
235: } immargs[MAX_IMMARGS];
236: } PrimInfo;
237:
238: PrimInfo *priminfos;
239: PrimInfo **decomp_prims;
240:
241: const char const* const prim_names[]={
242: #include PRIM_NAMES_I
243: };
244:
245: void init_ss_cost(void);
246:
247: static int is_relocatable(int p)
248: {
249: return !no_dynamic && priminfos[p].start != NULL;
250: }
251: #else /* defined(NO_DYNAMIC) */
252: static int is_relocatable(int p)
253: {
254: return 0;
255: }
256: #endif /* defined(NO_DYNAMIC) */
257:
258: #ifdef MEMCMP_AS_SUBROUTINE
259: int gforth_memcmp(const char * s1, const char * s2, size_t n)
260: {
261: return memcmp(s1, s2, n);
262: }
263:
264: Char *gforth_memmove(Char * dest, const Char* src, Cell n)
265: {
266: return memmove(dest, src, n);
267: }
268:
269: Char *gforth_memset(Char * s, Cell c, UCell n)
270: {
271: return memset(s, c, n);
272: }
273:
274: Char *gforth_memcpy(Char * dest, const Char* src, Cell n)
275: {
276: return memcpy(dest, src, n);
277: }
278: #endif
279:
280: static Cell max(Cell a, Cell b)
281: {
282: return a>b?a:b;
283: }
284:
285: static Cell min(Cell a, Cell b)
286: {
287: return a<b?a:b;
288: }
289:
290: #ifndef STANDALONE
291: /* image file format:
292: * "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
293: * padding to a multiple of 8
294: * magic: "Gforth4x" means format 0.8,
295: * where x is a byte with
296: * bit 7: reserved = 0
297: * bit 6:5: address unit size 2^n octets
298: * bit 4:3: character size 2^n octets
299: * bit 2:1: cell size 2^n octets
300: * bit 0: endian, big=0, little=1.
301: * The magic are always 8 octets, no matter what the native AU/character size is
302: * padding to max alignment (no padding necessary on current machines)
303: * ImageHeader structure (see forth.h)
304: * data (size in ImageHeader.image_size)
305: * tags ((if relocatable, 1 bit/data cell)
306: *
307: * tag==1 means that the corresponding word is an address;
308: * If the word is >=0, the address is within the image;
309: * addresses within the image are given relative to the start of the image.
310: * If the word =-1 (CF_NIL), the address is NIL,
311: * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
312: * If the word =CF(DODOES), it's a DOES> CFA
313: * !! ABI-CODE and ;ABI-CODE
314: * If the word is <CF(DOER_MAX) and bit 14 is set, it's the xt of a primitive
315: * If the word is <CF(DOER_MAX) and bit 14 is clear,
316: * it's the threaded code of a primitive
317: * bits 13..9 of a primitive token state which group the primitive belongs to,
318: * bits 8..0 of a primitive token index into the group
319: */
320:
321: Cell groups[32] = {
322: 0,
323: 0
324: #undef GROUP
325: #undef GROUPADD
326: #define GROUPADD(n) +n
327: #define GROUP(x, n) , 0
328: #include PRIM_GRP_I
329: #undef GROUP
330: #undef GROUPADD
331: #define GROUP(x, n)
332: #define GROUPADD(n)
333: };
334:
335: static unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
336: int size, Cell base)
337: /* produce a bitmask marking all the branch targets */
338: {
339: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
340: Cell token;
341: unsigned char bits;
342: unsigned char *result=malloc(steps);
343:
344: memset(result, 0, steps);
345: for(k=0; k<steps; k++) {
346: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
347: if(bits & (1U << (RELINFOBITS-1))) {
348: assert(i*sizeof(Cell) < size);
349: token=image[i];
350: if (token>=base) { /* relocatable address */
351: UCell bitnum=(token-base)/sizeof(Cell);
352: if (bitnum/RELINFOBITS < (UCell)steps)
353: result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
354: }
355: }
356: }
357: }
358: return result;
359: }
360:
361: void gforth_relocate(Cell *image, const Char *bitstring,
362: UCell size, Cell base, Label symbols[])
363: {
364: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
365: Cell token;
366: char bits;
367: Cell max_symbols;
368: /*
369: * A virtual start address that's the real start address minus
370: * the one in the image
371: */
372: Cell *start = (Cell * ) (((void *) image) - ((void *) base));
373: unsigned char *targets = branch_targets(image, bitstring, size, base);
374:
375: /* group index into table */
376: if(groups[31]==0) {
377: int groupsum=0;
378: for(i=0; i<32; i++) {
379: groupsum += groups[i];
380: groups[i] = groupsum;
381: /* printf("group[%d]=%d\n",i,groupsum); */
382: }
383: i=0;
384: }
385:
386: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
387:
388: for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
389: ;
390: max_symbols--;
391:
392: for(k=0; k<steps; k++) {
393: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
394: /* fprintf(stderr,"relocate: image[%d]\n", i);*/
395: if(bits & (1U << (RELINFOBITS-1))) {
396: assert(i*sizeof(Cell) < size);
397: /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
398: token=image[i];
399: if(token<0) {
400: int group = (-token & 0x3E00) >> 9;
401: if(group == 0) {
402: switch(token|0x4000) {
403: case CF_NIL : image[i]=0; break;
404: #if !defined(DOUBLY_INDIRECT)
405: case CF(DOCOL) :
406: case CF(DOVAR) :
407: case CF(DOCON) :
408: case CF(DOVAL) :
409: case CF(DOUSER) :
410: case CF(DODEFER) :
411: case CF(DOFIELD) :
412: case CF(DODOES) :
413: case CF(DOABICODE) :
414: case CF(DOSEMIABICODE):
415: MAKE_CF(image+i,symbols[CF(token)]); break;
416: #endif /* !defined(DOUBLY_INDIRECT) */
417: default : /* backward compatibility */
418: /* printf("Code field generation image[%x]:=CFA(%x)\n",
419: i, CF(image[i])); */
420: if (CF((token | 0x4000))<max_symbols) {
421: image[i]=(Cell)CFA(CF(token));
422: #ifdef DIRECT_THREADED
423: if ((token & 0x4000) == 0) { /* threade code, no CFA */
424: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
425: compile_prim1(0);
426: compile_prim1(&image[i]);
427: }
428: #endif
429: } else
430: 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);
431: }
432: } else {
433: int tok = -token & 0x1FF;
434: if (tok < (groups[group+1]-groups[group])) {
435: #if defined(DOUBLY_INDIRECT)
436: image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
437: #else
438: image[i]=(Cell)CFA((groups[group]+tok));
439: #endif
440: #ifdef DIRECT_THREADED
441: if ((token & 0x4000) == 0) { /* threade code, no CFA */
442: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
443: compile_prim1(0);
444: compile_prim1(&image[i]);
445: }
446: #endif
447: } else
448: fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
449: }
450: } else {
451: /* if base is > 0: 0 is a null reference so don't adjust*/
452: if (token>=base) {
453: image[i]+=(Cell)start;
454: }
455: }
456: }
457: }
458: }
459: free(targets);
460: finish_code();
461: ((ImageHeader*)(image))->base = (Address) image;
462: }
463:
464: #ifndef DOUBLY_INDIRECT
465: static UCell checksum(Label symbols[])
466: {
467: UCell r=PRIM_VERSION;
468: Cell i;
469:
470: for (i=DOCOL; i<=DOER_MAX; i++) {
471: r ^= (UCell)(symbols[i]);
472: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
473: }
474: #ifdef DIRECT_THREADED
475: /* we have to consider all the primitives */
476: for (; symbols[i]!=(Label)0; i++) {
477: r ^= (UCell)(symbols[i]);
478: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
479: }
480: #else
481: /* in indirect threaded code all primitives are accessed through the
482: symbols table, so we just have to put the base address of symbols
483: in the checksum */
484: r ^= (UCell)symbols;
485: #endif
486: return r;
487: }
488: #endif
489:
490: static Address verbose_malloc(Cell size)
491: {
492: Address r;
493: /* leave a little room (64B) for stack underflows */
494: if ((r = malloc(size+64))==NULL) {
495: perror(progname);
496: exit(1);
497: }
498: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
499: debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
500: return r;
501: }
502:
503: static void *next_address=0;
504: static void after_alloc(Address r, Cell size)
505: {
506: if (r != (Address)-1) {
507: debugp(stderr, "success, address=$%lx\n", (long) r);
508: #if 0
509: /* not needed now that we protect the stacks with mprotect */
510: if (pagesize != 1)
511: next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
512: #endif
513: } else {
514: debugp(stderr, "failed: %s\n", strerror(errno));
515: }
516: }
517:
518: #ifndef MAP_FAILED
519: #define MAP_FAILED ((Address) -1)
520: #endif
521: #ifndef MAP_FILE
522: # define MAP_FILE 0
523: #endif
524: #ifndef MAP_PRIVATE
525: # define MAP_PRIVATE 0
526: #endif
527: #ifndef PROT_NONE
528: # define PROT_NONE 0
529: #endif
530: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
531: # define MAP_ANON MAP_ANONYMOUS
532: #endif
533:
534: #if defined(HAVE_MMAP)
535: static Address alloc_mmap(Cell size)
536: {
537: void *r;
538:
539: #if defined(MAP_ANON)
540: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
541: r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE|map_noreserve, -1, 0);
542: #else /* !defined(MAP_ANON) */
543: /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
544: apparently defaults) */
545: static int dev_zero=-1;
546:
547: if (dev_zero == -1)
548: dev_zero = open("/dev/zero", O_RDONLY);
549: if (dev_zero == -1) {
550: r = MAP_FAILED;
551: debugp(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ",
552: strerror(errno));
553: } else {
554: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
555: r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE|map_noreserve, dev_zero, 0);
556: }
557: #endif /* !defined(MAP_ANON) */
558: after_alloc(r, size);
559: return r;
560: }
561:
562: static void page_noaccess(void *a)
563: {
564: /* try mprotect first; with munmap the page might be allocated later */
565: debugp(stderr, "try mprotect(%p,$%lx,PROT_NONE); ", a, (long)pagesize);
566: if (mprotect(a, pagesize, PROT_NONE)==0) {
567: debugp(stderr, "ok\n");
568: return;
569: }
570: debugp(stderr, "failed: %s\n", strerror(errno));
571: debugp(stderr, "try munmap(%p,$%lx); ", a, (long)pagesize);
572: if (munmap(a,pagesize)==0) {
573: debugp(stderr, "ok\n");
574: return;
575: }
576: debugp(stderr, "failed: %s\n", strerror(errno));
577: }
578:
579: static size_t wholepage(size_t n)
580: {
581: return (n+pagesize-1)&~(pagesize-1);
582: }
583: #endif
584:
585: Address gforth_alloc(Cell size)
586: {
587: #if HAVE_MMAP
588: Address r;
589:
590: r=alloc_mmap(size);
591: if (r!=(Address)MAP_FAILED)
592: return r;
593: #endif /* HAVE_MMAP */
594: /* use malloc as fallback */
595: return verbose_malloc(size);
596: }
597:
598: static void *dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
599: {
600: void *image = MAP_FAILED;
601:
602: #if defined(HAVE_MMAP)
603: if (offset==0) {
604: image=alloc_mmap(dictsize);
605: if (image != (void *)MAP_FAILED) {
606: void *image1;
607: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
608: image1 = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE|map_noreserve, fileno(file), 0);
609: after_alloc(image1,dictsize);
610: if (image1 == (void *)MAP_FAILED)
611: goto read_image;
612: }
613: }
614: #endif /* defined(HAVE_MMAP) */
615: if (image == (void *)MAP_FAILED) {
616: image = gforth_alloc(dictsize+offset)+offset;
617: read_image:
618: rewind(file); /* fseek(imagefile,0L,SEEK_SET); */
619: fread(image, 1, imagesize, file);
620: }
621: return image;
622: }
623: #endif
624:
625: void set_stack_sizes(ImageHeader * header)
626: {
627: if (dictsize==0)
628: dictsize = header->dict_size;
629: if (dsize==0)
630: dsize = header->data_stack_size;
631: if (rsize==0)
632: rsize = header->return_stack_size;
633: if (fsize==0)
634: fsize = header->fp_stack_size;
635: if (lsize==0)
636: lsize = header->locals_stack_size;
637: dictsize=maxaligned(dictsize);
638: dsize=maxaligned(dsize);
639: rsize=maxaligned(rsize);
640: lsize=maxaligned(lsize);
641: fsize=maxaligned(fsize);
642: }
643:
644: #ifdef STANDALONE
645: void alloc_stacks(ImageHeader * h)
646: {
647: #define SSTACKSIZE 0x200
648: static Cell dstack[SSTACKSIZE+1];
649: static Cell rstack[SSTACKSIZE+1];
650:
651: h->dict_size=dictsize;
652: h->data_stack_size=dsize;
653: h->fp_stack_size=fsize;
654: h->return_stack_size=rsize;
655: h->locals_stack_size=lsize;
656:
657: h->data_stack_base=dstack+SSTACKSIZE;
658: // h->fp_stack_base=gforth_alloc(fsize);
659: h->return_stack_base=rstack+SSTACKSIZE;
660: // h->locals_stack_base=gforth_alloc(lsize);
661: }
662: #else
663: void alloc_stacks(ImageHeader * h)
664: {
665: h->dict_size=dictsize;
666: h->data_stack_size=dsize;
667: h->fp_stack_size=fsize;
668: h->return_stack_size=rsize;
669: h->locals_stack_size=lsize;
670:
671: #if defined(HAVE_MMAP) && !defined(STANDALONE)
672: if (pagesize > 1) {
673: size_t p = pagesize;
674: size_t totalsize =
675: wholepage(dsize)+wholepage(fsize)+wholepage(rsize)+wholepage(lsize)+5*p;
676: void *a = alloc_mmap(totalsize);
677: if (a != (void *)MAP_FAILED) {
678: page_noaccess(a); a+=p; h-> data_stack_base=a; a+=wholepage(dsize);
679: page_noaccess(a); a+=p; h-> fp_stack_base=a; a+=wholepage(fsize);
680: page_noaccess(a); a+=p; h->return_stack_base=a; a+=wholepage(rsize);
681: page_noaccess(a); a+=p; h->locals_stack_base=a; a+=wholepage(lsize);
682: page_noaccess(a);
683: debugp(stderr,"stack addresses: d=%p f=%p r=%p l=%p\n",
684: h->data_stack_base,
685: h->fp_stack_base,
686: h->return_stack_base,
687: h->locals_stack_base);
688: return;
689: }
690: }
691: #endif
692: h->data_stack_base=gforth_alloc(dsize);
693: h->fp_stack_base=gforth_alloc(fsize);
694: h->return_stack_base=gforth_alloc(rsize);
695: h->locals_stack_base=gforth_alloc(lsize);
696: }
697: #endif
698:
699: #warning You can ignore the warnings about clobbered variables in gforth_go
700: int gforth_go(void *image, int stack, Cell *entries)
701: {
702: volatile ImageHeader *image_header = (ImageHeader *)image;
703: Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
704: Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
705: Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
706: #ifdef GFORTH_DEBUGGING
707: volatile Cell *orig_rp0=rp0;
708: #endif
709: Address lp0=image_header->locals_stack_base + lsize;
710: Xt *ip0=(Xt *)(image_header->boot_entry);
711: #ifdef SYSSIGNALS
712: int throw_code;
713: #endif
714:
715: /* ensure that the cached elements (if any) are accessible */
716: #if !(defined(GFORTH_DEBUGGING) || defined(INDIRECT_THREADED) || defined(DOUBLY_INDIRECT) || defined(VM_PROFILING))
717: sp0 -= 8; /* make stuff below bottom accessible for stack caching */
718: fp0--;
719: #endif
720:
721: for(;stack>0;stack--)
722: *--sp0=entries[stack-1];
723:
724: #if defined(SYSSIGNALS) && !defined(STANDALONE)
725: get_winsize();
726:
727: install_signal_handlers(); /* right place? */
728:
729: debugp(stderr, "setjmp(%lx)\n", throw_jmp_buf);
730: if ((throw_code=setjmp(throw_jmp_buf))) {
731: static Cell signal_data_stack[24];
732: static Cell signal_return_stack[16];
733: static Float signal_fp_stack[1];
734:
735: signal_data_stack[15]=throw_code;
736:
737: #ifdef GFORTH_DEBUGGING
738: debugp(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
739: throw_code, saved_ip, rp);
740: if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
741: /* no rstack overflow or underflow */
742: rp0 = rp;
743: *--rp0 = (Cell)saved_ip;
744: }
745: else /* I love non-syntactic ifdefs :-) */
746: rp0 = signal_return_stack+16;
747: #else /* !defined(GFORTH_DEBUGGING) */
748: debugp(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
749: rp0 = signal_return_stack+16;
750: #endif /* !defined(GFORTH_DEBUGGING) */
751: /* fprintf(stderr, "rp=$%x\n",rp0);*/
752:
753: return((int)(Cell)gforth_engine(image_header->throw_entry, signal_data_stack+15,
754: rp0, signal_fp_stack, 0 sr_call));
755: }
756: #endif
757:
758: return((int)(Cell)gforth_engine(ip0,sp0,rp0,fp0,lp0 sr_call));
759: }
760:
761: #if !defined(INCLUDE_IMAGE) && !defined(STANDALONE)
762: static void print_sizes(Cell sizebyte)
763: /* print size information */
764: {
765: static char* endianstring[]= { " big","little" };
766:
767: fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
768: endianstring[sizebyte & 1],
769: 1 << ((sizebyte >> 1) & 3),
770: 1 << ((sizebyte >> 3) & 3),
771: 1 << ((sizebyte >> 5) & 3));
772: }
773:
774: /* static superinstruction stuff */
775:
776: struct cost { /* super_info might be a more accurate name */
777: char loads; /* number of stack loads */
778: char stores; /* number of stack stores */
779: char updates; /* number of stack pointer updates */
780: char branch; /* is it a branch (SET_IP) */
781: unsigned char state_in; /* state on entry */
782: unsigned char state_out; /* state on exit */
783: unsigned char imm_ops; /* number of immediate operands */
784: short offset; /* offset into super2 table */
785: unsigned char length; /* number of components */
786: };
787:
788: PrimNum super2[] = {
789: #include SUPER2_I
790: };
791:
792: struct cost super_costs[] = {
793: #include COSTS_I
794: };
795:
796: struct super_state {
797: struct super_state *next;
798: PrimNum super;
799: };
800:
801: #define HASH_SIZE 256
802:
803: struct super_table_entry {
804: struct super_table_entry *next;
805: PrimNum *start;
806: short length;
807: struct super_state *ss_list; /* list of supers */
808: } *super_table[HASH_SIZE];
809: int max_super=2;
810:
811: struct super_state *state_transitions=NULL;
812:
813: static int hash_super(PrimNum *start, int length)
814: {
815: int i, r;
816:
817: for (i=0, r=0; i<length; i++) {
818: r <<= 1;
819: r += start[i];
820: }
821: return r & (HASH_SIZE-1);
822: }
823:
824: static struct super_state **lookup_super(PrimNum *start, int length)
825: {
826: int hash=hash_super(start,length);
827: struct super_table_entry *p = super_table[hash];
828:
829: /* assert(length >= 2); */
830: for (; p!=NULL; p = p->next) {
831: if (length == p->length &&
832: memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
833: return &(p->ss_list);
834: }
835: return NULL;
836: }
837:
838: static void prepare_super_table()
839: {
840: int i;
841: int nsupers = 0;
842:
843: for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
844: struct cost *c = &super_costs[i];
845: if ((c->length < 2 || nsupers < static_super_number) &&
846: c->state_in < maxstates && c->state_out < maxstates) {
847: struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
848: struct super_state *ss = malloc(sizeof(struct super_state));
849: ss->super= i;
850: if (c->offset==N_noop && i != N_noop) {
851: if (is_relocatable(i)) {
852: ss->next = state_transitions;
853: state_transitions = ss;
854: }
855: } else if (ss_listp != NULL) {
856: ss->next = *ss_listp;
857: *ss_listp = ss;
858: } else {
859: int hash = hash_super(super2+c->offset, c->length);
860: struct super_table_entry **p = &super_table[hash];
861: struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
862: ss->next = NULL;
863: e->next = *p;
864: e->start = super2 + c->offset;
865: e->length = c->length;
866: e->ss_list = ss;
867: *p = e;
868: }
869: if (c->length > max_super)
870: max_super = c->length;
871: if (c->length >= 2)
872: nsupers++;
873: }
874: }
875: debugp(stderr, "Using %d static superinsts\n", nsupers);
876: if (nsupers>0 && !tpa_noautomaton && !tpa_noequiv) {
877: /* Currently these two things don't work together; see Section 3.2
878: of <http://www.complang.tuwien.ac.at/papers/ertl+06pldi.ps.gz>,
879: in particular Footnote 6 for the reason; hmm, we should be able
880: to use an automaton without state equivalence, but that costs
881: significant space so we only do it if the user explicitly
882: disables state equivalence. */
883: debugp(stderr, "Disabling tpa-automaton, because nsupers>0 and state equivalence is enabled.\n");
884: tpa_noautomaton = 1;
885: }
886: }
887:
888: /* dynamic replication/superinstruction stuff */
889:
890: #ifndef NO_DYNAMIC
891: static int compare_priminfo_length(const void *_a, const void *_b)
892: {
893: PrimInfo **a = (PrimInfo **)_a;
894: PrimInfo **b = (PrimInfo **)_b;
895: Cell diff = (*a)->length - (*b)->length;
896: if (diff)
897: return diff;
898: else /* break ties by start address; thus the decompiler produces
899: the earliest primitive with the same code (e.g. noop instead
900: of (char) and @ instead of >code-address */
901: return (*b)->start - (*a)->start;
902: }
903: #endif /* !defined(NO_DYNAMIC) */
904:
905: static char MAYBE_UNUSED superend[]={
906: #include PRIM_SUPEREND_I
907: };
908:
909: Cell npriminfos=0;
910:
911: Label goto_start;
912: Cell goto_len;
913:
914: #ifndef NO_DYNAMIC
915: static int compare_labels(const void *pa, const void *pb)
916: {
917: Label a = *(Label *)pa;
918: Label b = *(Label *)pb;
919: return a-b;
920: }
921: #endif
922:
923: static Label bsearch_next(Label key, Label *a, UCell n)
924: /* a is sorted; return the label >=key that is the closest in a;
925: return NULL if there is no label in a >=key */
926: {
927: int mid = (n-1)/2;
928: if (n<1)
929: return NULL;
930: if (n == 1) {
931: if (a[0] < key)
932: return NULL;
933: else
934: return a[0];
935: }
936: if (a[mid] < key)
937: return bsearch_next(key, a+mid+1, n-mid-1);
938: else
939: return bsearch_next(key, a, mid+1);
940: }
941:
942: static void check_prims(Label symbols1[])
943: {
944: int i;
945: #ifndef NO_DYNAMIC
946: Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted, *goto_p;
947: int nends1j;
948: #endif
949:
950: if (debug)
951: #ifdef __VERSION__
952: fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
953: #else
954: #define xstr(s) str(s)
955: #define str(s) #s
956: fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n");
957: #endif
958: for (i=0; symbols1[i]!=0; i++)
959: ;
960: npriminfos = i;
961:
962: #ifndef NO_DYNAMIC
963: if (no_dynamic)
964: return;
965: symbols2=gforth_engine2(0,0,0,0,0 sr_call);
966: #if NO_IP
967: symbols3=gforth_engine3(0,0,0,0,0 sr_call);
968: #else
969: symbols3=symbols1;
970: #endif
971: ends1 = symbols1+i+1;
972: ends1j = ends1+i;
973: goto_p = ends1j+i+1; /* goto_p[0]==before; ...[1]==after;*/
974: nends1j = i+1;
975: ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
976: memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
977: qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
978:
979: /* check whether the "goto *" is relocatable */
980: goto_len = goto_p[1]-goto_p[0];
981: debugp(stderr, "goto * %p %p len=%ld\n",
982: goto_p[0],symbols2[goto_p-symbols1],(long)goto_len);
983: if (memcmp(goto_p[0],symbols2[goto_p-symbols1],goto_len)!=0) { /* unequal */
984: no_dynamic=1;
985: debugp(stderr," not relocatable, disabling dynamic code generation\n");
986: init_ss_cost();
987: return;
988: }
989: goto_start = goto_p[0];
990:
991: priminfos = calloc(i,sizeof(PrimInfo));
992: for (i=0; symbols1[i]!=0; i++) {
993: int prim_len = ends1[i]-symbols1[i];
994: PrimInfo *pi=&priminfos[i];
995: struct cost *sc=&super_costs[i];
996: int j=0;
997: char *s1 = (char *)symbols1[i];
998: char *s2 = (char *)symbols2[i];
999: char *s3 = (char *)symbols3[i];
1000: Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
1001:
1002: pi->start = s1;
1003: pi->superend = superend[i]|no_super;
1004: pi->length = prim_len;
1005: pi->restlength = endlabel - symbols1[i] - pi->length;
1006: pi->nimmargs = 0;
1007: relocs++;
1008: #if defined(BURG_FORMAT)
1009: { /* output as burg-style rules */
1010: int p=super_costs[i].offset;
1011: if (p==N_noop)
1012: debugp(stderr, "S%d: S%d = %d (%d);", sc->state_in, sc->state_out, i+1, pi->length);
1013: else
1014: debugp(stderr, "S%d: op%d(S%d) = %d (%d);", sc->state_in, p, sc->state_out, i+1, pi->length);
1015: }
1016: #else
1017: debugp(stderr, "%-15s %d-%d %4d %p %p %p len=%3ld rest=%2ld send=%1d",
1018: prim_names[i], sc->state_in, sc->state_out,
1019: i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength),
1020: pi->superend);
1021: #endif
1022: if (endlabel == NULL) {
1023: pi->start = NULL; /* not relocatable */
1024: if (pi->length<0) pi->length=100;
1025: #ifndef BURG_FORMAT
1026: debugp(stderr,"\n non_reloc: no J label > start found\n");
1027: #endif
1028: relocs--;
1029: nonrelocs++;
1030: continue;
1031: }
1032: if (ends1[i] > endlabel && !pi->superend) {
1033: pi->start = NULL; /* not relocatable */
1034: pi->length = endlabel-symbols1[i];
1035: #ifndef BURG_FORMAT
1036: debugp(stderr,"\n non_reloc: there is a J label before the K label (restlength<0)\n");
1037: #endif
1038: relocs--;
1039: nonrelocs++;
1040: continue;
1041: }
1042: if (ends1[i] < pi->start && !pi->superend) {
1043: pi->start = NULL; /* not relocatable */
1044: pi->length = endlabel-symbols1[i];
1045: #ifndef BURG_FORMAT
1046: debugp(stderr,"\n non_reloc: K label before I label (length<0)\n");
1047: #endif
1048: relocs--;
1049: nonrelocs++;
1050: continue;
1051: }
1052: if (CHECK_PRIM(s1, prim_len)) {
1053: #ifndef BURG_FORMAT
1054: debugp(stderr,"\n non_reloc: architecture specific check failed\n");
1055: #endif
1056: pi->start = NULL; /* not relocatable */
1057: relocs--;
1058: nonrelocs++;
1059: continue;
1060: }
1061: assert(pi->length>=0);
1062: assert(pi->restlength >=0);
1063: while (j<(pi->length+pi->restlength)) {
1064: if (s1[j]==s3[j]) {
1065: if (s1[j] != s2[j]) {
1066: pi->start = NULL; /* not relocatable */
1067: #ifndef BURG_FORMAT
1068: debugp(stderr,"\n non_reloc: engine1!=engine2 offset %3d",j);
1069: #endif
1070: /* assert(j<prim_len); */
1071: relocs--;
1072: nonrelocs++;
1073: break;
1074: }
1075: j++;
1076: } else {
1077: struct immarg *ia=&pi->immargs[pi->nimmargs];
1078:
1079: pi->nimmargs++;
1080: ia->offset=j;
1081: if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
1082: ia->rel=0;
1083: debugp(stderr,"\n absolute immarg: offset %3d",j);
1084: } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
1085: symbols1[DOER_MAX+1]) {
1086: ia->rel=1;
1087: debugp(stderr,"\n relative immarg: offset %3d",j);
1088: } else {
1089: pi->start = NULL; /* not relocatable */
1090: #ifndef BURG_FORMAT
1091: debugp(stderr,"\n non_reloc: engine1!=engine3 offset %3d",j);
1092: #endif
1093: /* assert(j<prim_len);*/
1094: relocs--;
1095: nonrelocs++;
1096: break;
1097: }
1098: j+=4;
1099: }
1100: }
1101: debugp(stderr,"\n");
1102: }
1103: decomp_prims = calloc(i,sizeof(PrimInfo *));
1104: for (i=DOER_MAX+1; i<npriminfos; i++)
1105: decomp_prims[i] = &(priminfos[i]);
1106: qsort(decomp_prims+DOER_MAX+1, npriminfos-DOER_MAX-1, sizeof(PrimInfo *),
1107: compare_priminfo_length);
1108: #endif
1109: }
1110:
1111: static void flush_to_here(void)
1112: {
1113: #ifndef NO_DYNAMIC
1114: if (start_flush)
1115: FLUSH_ICACHE((caddr_t)start_flush, code_here-start_flush);
1116: start_flush=code_here;
1117: #endif
1118: }
1119:
1120: static void MAYBE_UNUSED align_code(void)
1121: /* align code_here on some platforms */
1122: {
1123: #ifndef NO_DYNAMIC
1124: #if defined(CODE_PADDING)
1125: Cell alignment = CODE_ALIGNMENT;
1126: static char nops[] = CODE_PADDING;
1127: UCell maxpadding=MAX_PADDING;
1128: UCell offset = ((UCell)code_here)&(alignment-1);
1129: UCell length = alignment-offset;
1130: if (length <= maxpadding) {
1131: memcpy(code_here,nops+offset,length);
1132: code_here += length;
1133: }
1134: #endif /* defined(CODE_PADDING) */
1135: #endif /* defined(NO_DYNAMIC */
1136: }
1137:
1138: #ifndef NO_DYNAMIC
1139: static void append_jump(void)
1140: {
1141: if (last_jump) {
1142: PrimInfo *pi = &priminfos[last_jump];
1143:
1144: memcpy(code_here, pi->start+pi->length, pi->restlength);
1145: code_here += pi->restlength;
1146: memcpy(code_here, goto_start, goto_len);
1147: code_here += goto_len;
1148: align_code();
1149: last_jump=0;
1150: }
1151: }
1152:
1153: /* Gforth remembers all code blocks in this list. On forgetting (by
1154: executing a marker) the code blocks are not freed (because Gforth does
1155: not remember how they were allocated; hmm, remembering that might be
1156: easier and cleaner). Instead, code_here etc. are reset to the old
1157: value, and the "forgotten" code blocks are reused when they are
1158: needed. */
1159:
1160: struct code_block_list {
1161: struct code_block_list *next;
1162: Address block;
1163: Cell size;
1164: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
1165:
1166: static void reserve_code_space(UCell size)
1167: {
1168: if (code_area+code_area_size < code_here+size) {
1169: struct code_block_list *p;
1170: append_jump();
1171: debugp(stderr,"Did not use %ld bytes in code block\n",
1172: (long)(code_area+code_area_size-code_here));
1173: flush_to_here();
1174: if (*next_code_blockp == NULL) {
1175: code_here = start_flush = code_area = gforth_alloc(code_area_size);
1176: p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
1177: *next_code_blockp = p;
1178: p->next = NULL;
1179: p->block = code_here;
1180: p->size = code_area_size;
1181: } else {
1182: p = *next_code_blockp;
1183: code_here = start_flush = code_area = p->block;
1184: }
1185: next_code_blockp = &(p->next);
1186: }
1187: }
1188:
1189: static Address append_prim(Cell p)
1190: {
1191: PrimInfo *pi = &priminfos[p];
1192: Address old_code_here;
1193: reserve_code_space(pi->length+pi->restlength+goto_len+CODE_ALIGNMENT-1);
1194: memcpy(code_here, pi->start, pi->length);
1195: old_code_here = code_here;
1196: code_here += pi->length;
1197: return old_code_here;
1198: }
1199:
1200: static void reserve_code_super(PrimNum origs[], int ninsts)
1201: {
1202: int i;
1203: UCell size = CODE_ALIGNMENT-1; /* alignment may happen first */
1204: if (no_dynamic)
1205: return;
1206: /* use size of the original primitives as an upper bound for the
1207: size of the superinstruction. !! This is only safe if we
1208: optimize for code size (the default) */
1209: for (i=0; i<ninsts; i++) {
1210: PrimNum p = origs[i];
1211: PrimInfo *pi = &priminfos[p];
1212: if (is_relocatable(p))
1213: size += pi->length;
1214: else
1215: if (i>0)
1216: size += priminfos[origs[i-1]].restlength+goto_len+CODE_ALIGNMENT-1;
1217: }
1218: size += priminfos[origs[i-1]].restlength+goto_len;
1219: reserve_code_space(size);
1220: }
1221: #endif
1222:
1223: int forget_dyncode(Address code)
1224: {
1225: #ifdef NO_DYNAMIC
1226: return -1;
1227: #else
1228: struct code_block_list *p, **pp;
1229:
1230: for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
1231: if (code >= p->block && code < p->block+p->size) {
1232: next_code_blockp = &(p->next);
1233: code_here = start_flush = code;
1234: code_area = p->block;
1235: last_jump = 0;
1236: return -1;
1237: }
1238: }
1239: return -no_dynamic;
1240: #endif /* !defined(NO_DYNAMIC) */
1241: }
1242:
1243: static long dyncodesize(void)
1244: {
1245: #ifndef NO_DYNAMIC
1246: struct code_block_list *p;
1247: long size=0;
1248: for (p=code_block_list; p!=NULL; p=p->next) {
1249: if (code_here >= p->block && code_here < p->block+p->size)
1250: return size + (code_here - p->block);
1251: else
1252: size += p->size;
1253: }
1254: #endif /* !defined(NO_DYNAMIC) */
1255: return 0;
1256: }
1257:
1258: Label decompile_code(Label _code)
1259: {
1260: #ifdef NO_DYNAMIC
1261: return _code;
1262: #else /* !defined(NO_DYNAMIC) */
1263: Cell i;
1264: struct code_block_list *p;
1265: Address code=_code;
1266:
1267: /* first, check if we are in code at all */
1268: for (p = code_block_list;; p = p->next) {
1269: if (p == NULL)
1270: return code;
1271: if (code >= p->block && code < p->block+p->size)
1272: break;
1273: }
1274: /* reverse order because NOOP might match other prims */
1275: for (i=npriminfos-1; i>DOER_MAX; i--) {
1276: PrimInfo *pi=decomp_prims[i];
1277: if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1278: return vm_prims[super2[super_costs[pi-priminfos].offset]];
1279: /* return pi->start;*/
1280: }
1281: return code;
1282: #endif /* !defined(NO_DYNAMIC) */
1283: }
1284:
1285: #ifdef NO_IP
1286: int nbranchinfos=0;
1287:
1288: struct branchinfo {
1289: Label **targetpp; /* **(bi->targetpp) is the target */
1290: Cell *addressptr; /* store the target here */
1291: } branchinfos[100000];
1292:
1293: int ndoesexecinfos=0;
1294: struct doesexecinfo {
1295: int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1296: Label *targetp; /*target for branch (because this is not in threaded code)*/
1297: Cell *xt; /* cfa of word whose does-code needs calling */
1298: } doesexecinfos[10000];
1299:
1300: static void set_rel_target(Cell *source, Label target)
1301: {
1302: *source = ((Cell)target)-(((Cell)source)+4);
1303: }
1304:
1305: static void register_branchinfo(Label source, Cell *targetpp)
1306: {
1307: struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1308: bi->targetpp = (Label **)targetpp;
1309: bi->addressptr = (Cell *)source;
1310: nbranchinfos++;
1311: }
1312:
1313: static Address compile_prim1arg(PrimNum p, Cell **argp)
1314: {
1315: Address old_code_here=append_prim(p);
1316:
1317: assert(vm_prims[p]==priminfos[p].start);
1318: *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
1319: return old_code_here;
1320: }
1321:
1322: static Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
1323: {
1324: PrimInfo *pi = &priminfos[N_call2];
1325: Address old_code_here = append_prim(N_call2);
1326:
1327: *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
1328: register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
1329: return old_code_here;
1330: }
1331: #endif
1332:
1333: void finish_code(void)
1334: {
1335: #ifdef NO_IP
1336: Cell i;
1337:
1338: compile_prim1(NULL);
1339: for (i=0; i<ndoesexecinfos; i++) {
1340: struct doesexecinfo *dei = &doesexecinfos[i];
1341: dei->targetp = (Label *)DOES_CODE1((dei->xt));
1342: branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
1343: }
1344: ndoesexecinfos = 0;
1345: for (i=0; i<nbranchinfos; i++) {
1346: struct branchinfo *bi=&branchinfos[i];
1347: set_rel_target(bi->addressptr, **(bi->targetpp));
1348: }
1349: nbranchinfos = 0;
1350: #else
1351: compile_prim1(NULL);
1352: #endif
1353: flush_to_here();
1354: }
1355:
1356: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1357: #ifdef NO_IP
1358: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1359: /* compile prim #p dynamically (mod flags etc.) and return start
1360: address of generated code for putting it into the threaded
1361: code. This function is only called if all the associated
1362: inline arguments of p are already in place (at tcp[1] etc.) */
1363: {
1364: PrimInfo *pi=&priminfos[p];
1365: Cell *next_code_target=NULL;
1366: Address codeaddr;
1367: Address primstart;
1368:
1369: assert(p<npriminfos);
1370: if (p==N_execute || p==N_perform || p==N_lit_perform) {
1371: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1372: primstart = append_prim(p);
1373: goto other_prim;
1374: } else if (p==N_call) {
1375: codeaddr = compile_call2(tcp+1, &next_code_target);
1376: } else if (p==N_does_exec) {
1377: struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1378: Cell *arg;
1379: codeaddr = compile_prim1arg(N_lit,&arg);
1380: *arg = (Cell)PFA(tcp[1]);
1381: /* we cannot determine the callee now (last_start[1] may be a
1382: forward reference), so just register an arbitrary target, and
1383: register in dei that we need to fix this before resolving
1384: branches */
1385: dei->branchinfo = nbranchinfos;
1386: dei->xt = (Cell *)(tcp[1]);
1387: compile_call2(0, &next_code_target);
1388: } else if (!is_relocatable(p)) {
1389: Cell *branch_target;
1390: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1391: compile_prim1arg(N_branch,&branch_target);
1392: set_rel_target(branch_target,vm_prims[p]);
1393: } else {
1394: unsigned j;
1395:
1396: codeaddr = primstart = append_prim(p);
1397: other_prim:
1398: for (j=0; j<pi->nimmargs; j++) {
1399: struct immarg *ia = &(pi->immargs[j]);
1400: Cell *argp = tcp + pi->nimmargs - j;
1401: Cell argval = *argp; /* !! specific to prims */
1402: if (ia->rel) { /* !! assumption: relative refs are branches */
1403: register_branchinfo(primstart + ia->offset, argp);
1404: } else /* plain argument */
1405: *(Cell *)(primstart + ia->offset) = argval;
1406: }
1407: }
1408: if (next_code_target!=NULL)
1409: *next_code_target = (Cell)code_here;
1410: return (Cell)codeaddr;
1411: }
1412: #else /* !defined(NO_IP) */
1413: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1414: /* compile prim #p dynamically (mod flags etc.) and return start
1415: address of generated code for putting it into the threaded code */
1416: {
1417: Cell static_prim = (Cell)vm_prims[p];
1418: #if defined(NO_DYNAMIC)
1419: return static_prim;
1420: #else /* !defined(NO_DYNAMIC) */
1421: Address old_code_here;
1422:
1423: if (no_dynamic)
1424: return static_prim;
1425: if (p>=npriminfos || !is_relocatable(p)) {
1426: append_jump();
1427: return static_prim;
1428: }
1429: old_code_here = append_prim(p);
1430: last_jump = p;
1431: if (priminfos[p].superend)
1432: append_jump();
1433: return (Cell)old_code_here;
1434: #endif /* !defined(NO_DYNAMIC) */
1435: }
1436: #endif /* !defined(NO_IP) */
1437: #endif
1438:
1439: #ifndef NO_DYNAMIC
1440: static int cost_codesize(int prim)
1441: {
1442: return priminfos[prim].length;
1443: }
1444: #endif
1445:
1446: static int cost_ls(int prim)
1447: {
1448: struct cost *c = super_costs+prim;
1449:
1450: return c->loads + c->stores;
1451: }
1452:
1453: static int cost_lsu(int prim)
1454: {
1455: struct cost *c = super_costs+prim;
1456:
1457: return c->loads + c->stores + c->updates;
1458: }
1459:
1460: static int cost_nexts(int prim)
1461: {
1462: return 1;
1463: }
1464:
1465: typedef int Costfunc(int);
1466: Costfunc *ss_cost = /* cost function for optimize_bb */
1467: #ifdef NO_DYNAMIC
1468: cost_lsu;
1469: #else
1470: cost_codesize;
1471: #endif
1472:
1473: struct {
1474: Costfunc *costfunc;
1475: char *metricname;
1476: long sum;
1477: } cost_sums[] = {
1478: #ifndef NO_DYNAMIC
1479: { cost_codesize, "codesize", 0 },
1480: #endif
1481: { cost_ls, "ls", 0 },
1482: { cost_lsu, "lsu", 0 },
1483: { cost_nexts, "nexts", 0 }
1484: };
1485:
1486: #ifndef NO_DYNAMIC
1487: void init_ss_cost(void) {
1488: if (no_dynamic && ss_cost == cost_codesize) {
1489: ss_cost = cost_nexts;
1490: cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
1491: debugp(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
1492: }
1493: }
1494: #endif
1495:
1496: #define MAX_BB 128 /* maximum number of instructions in BB */
1497: #define INF_COST 1000000 /* infinite cost */
1498: #define CANONICAL_STATE 0
1499:
1500: struct waypoint {
1501: int cost; /* the cost from here to the end */
1502: PrimNum inst; /* the inst used from here to the next waypoint */
1503: char relocatable; /* the last non-transition was relocatable */
1504: char no_transition; /* don't use the next transition (relocatability)
1505: * or this transition (does not change state) */
1506: };
1507:
1508: struct tpa_state { /* tree parsing automaton (like) state */
1509: /* labeling is back-to-front */
1510: struct waypoint *inst; /* in front of instruction */
1511: struct waypoint *trans; /* in front of instruction and transition */
1512: };
1513:
1514: struct tpa_state *termstate = NULL; /* initialized in loader() */
1515:
1516: /* statistics about tree parsing (lazyburg) stuff */
1517: long lb_basic_blocks = 0;
1518: long lb_labeler_steps = 0;
1519: long lb_labeler_automaton = 0;
1520: long lb_labeler_dynprog = 0;
1521: long lb_newstate_equiv = 0;
1522: long lb_newstate_new = 0;
1523: long lb_applicable_base_rules = 0;
1524: long lb_applicable_chain_rules = 0;
1525:
1526: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1527: static void init_waypoints(struct waypoint ws[])
1528: {
1529: int k;
1530:
1531: for (k=0; k<maxstates; k++)
1532: ws[k].cost=INF_COST;
1533: }
1534:
1535: static struct tpa_state *empty_tpa_state()
1536: {
1537: struct tpa_state *s = malloc(sizeof(struct tpa_state));
1538:
1539: s->inst = calloc(maxstates,sizeof(struct waypoint));
1540: init_waypoints(s->inst);
1541: s->trans = calloc(maxstates,sizeof(struct waypoint));
1542: /* init_waypoints(s->trans);*/
1543: return s;
1544: }
1545:
1546: static void transitions(struct tpa_state *t)
1547: {
1548: int k;
1549: struct super_state *l;
1550:
1551: for (k=0; k<maxstates; k++) {
1552: t->trans[k] = t->inst[k];
1553: t->trans[k].no_transition = 1;
1554: }
1555: for (l = state_transitions; l != NULL; l = l->next) {
1556: PrimNum s = l->super;
1557: int jcost;
1558: struct cost *c=super_costs+s;
1559: struct waypoint *wi=&(t->trans[c->state_in]);
1560: struct waypoint *wo=&(t->inst[c->state_out]);
1561: lb_applicable_chain_rules++;
1562: if (wo->cost == INF_COST)
1563: continue;
1564: jcost = wo->cost + ss_cost(s);
1565: if (jcost <= wi->cost) {
1566: wi->cost = jcost;
1567: wi->inst = s;
1568: wi->relocatable = wo->relocatable;
1569: wi->no_transition = 0;
1570: /* if (ss_greedy) wi->cost = wo->cost ? */
1571: }
1572: }
1573: }
1574:
1575: static struct tpa_state *make_termstate()
1576: {
1577: struct tpa_state *s = empty_tpa_state();
1578:
1579: s->inst[CANONICAL_STATE].cost = 0;
1580: transitions(s);
1581: return s;
1582: }
1583: #endif
1584:
1585: #define TPA_SIZE 16384
1586:
1587: struct tpa_entry {
1588: struct tpa_entry *next;
1589: PrimNum inst;
1590: struct tpa_state *state_behind; /* note: brack-to-front labeling */
1591: struct tpa_state *state_infront; /* note: brack-to-front labeling */
1592: } *tpa_table[TPA_SIZE];
1593:
1594: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1595: static Cell hash_tpa(PrimNum p, struct tpa_state *t)
1596: {
1597: UCell it = (UCell )t;
1598: return (p+it+(it>>14))&(TPA_SIZE-1);
1599: }
1600:
1601: static struct tpa_state **lookup_tpa(PrimNum p, struct tpa_state *t2)
1602: {
1603: int hash=hash_tpa(p, t2);
1604: struct tpa_entry *te = tpa_table[hash];
1605:
1606: if (tpa_noautomaton) {
1607: static struct tpa_state *t;
1608: t = NULL;
1609: return &t;
1610: }
1611: for (; te!=NULL; te = te->next) {
1612: if (p == te->inst && t2 == te->state_behind)
1613: return &(te->state_infront);
1614: }
1615: te = (struct tpa_entry *)malloc(sizeof(struct tpa_entry));
1616: te->next = tpa_table[hash];
1617: te->inst = p;
1618: te->state_behind = t2;
1619: te->state_infront = NULL;
1620: tpa_table[hash] = te;
1621: return &(te->state_infront);
1622: }
1623:
1624: static void tpa_state_normalize(struct tpa_state *t)
1625: {
1626: /* normalize so cost of canonical state=0; this may result in
1627: negative costs for some states */
1628: int d = t->inst[CANONICAL_STATE].cost;
1629: int i;
1630:
1631: for (i=0; i<maxstates; i++) {
1632: if (t->inst[i].cost != INF_COST)
1633: t->inst[i].cost -= d;
1634: if (t->trans[i].cost != INF_COST)
1635: t->trans[i].cost -= d;
1636: }
1637: }
1638:
1639: static int tpa_state_equivalent(struct tpa_state *t1, struct tpa_state *t2)
1640: {
1641: return (memcmp(t1->inst, t2->inst, maxstates*sizeof(struct waypoint)) == 0 &&
1642: memcmp(t1->trans,t2->trans,maxstates*sizeof(struct waypoint)) == 0);
1643: }
1644: #endif
1645:
1646: struct tpa_state_entry {
1647: struct tpa_state_entry *next;
1648: struct tpa_state *state;
1649: } *tpa_state_table[TPA_SIZE];
1650:
1651: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1652: static Cell hash_tpa_state(struct tpa_state *t)
1653: {
1654: int *ti = (int *)(t->inst);
1655: int *tt = (int *)(t->trans);
1656: int r=0;
1657: int i;
1658:
1659: for (i=0; ti+i < (int *)(t->inst+maxstates); i++)
1660: r += ti[i]+tt[i];
1661: return (r+(r>>14)+(r>>22)) & (TPA_SIZE-1);
1662: }
1663:
1664: static struct tpa_state *lookup_tpa_state(struct tpa_state *t)
1665: {
1666: Cell hash = hash_tpa_state(t);
1667: struct tpa_state_entry *te = tpa_state_table[hash];
1668: struct tpa_state_entry *tn;
1669:
1670: if (!tpa_noequiv) {
1671: for (; te!=NULL; te = te->next) {
1672: if (tpa_state_equivalent(t, te->state)) {
1673: lb_newstate_equiv++;
1674: free(t->inst);
1675: free(t->trans);
1676: free(t);
1677: return te->state;
1678: }
1679: }
1680: tn = (struct tpa_state_entry *)malloc(sizeof(struct tpa_state_entry));
1681: tn->next = te;
1682: tn->state = t;
1683: tpa_state_table[hash] = tn;
1684: }
1685: lb_newstate_new++;
1686: if (tpa_trace)
1687: fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
1688: return t;
1689: }
1690:
1691: /* use dynamic programming to find the shortest paths within the basic
1692: block origs[0..ninsts-1] and rewrite the instructions pointed to by
1693: instps to use it */
1694: static void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
1695: {
1696: int i,j;
1697: struct tpa_state *ts[ninsts+1];
1698: int nextdyn, nextstate, no_transition;
1699: Address old_code_area;
1700:
1701: lb_basic_blocks++;
1702: ts[ninsts] = termstate;
1703: #ifndef NO_DYNAMIC
1704: if (print_sequences) {
1705: for (i=0; i<ninsts; i++)
1706: #if defined(BURG_FORMAT)
1707: fprintf(stderr, "op%d ", super_costs[origs[i]].offset);
1708: #else
1709: fprintf(stderr, "%s ", prim_names[origs[i]]);
1710: #endif
1711: fprintf(stderr, "\n");
1712: }
1713: #endif
1714: for (i=ninsts-1; i>=0; i--) {
1715: struct tpa_state **tp = lookup_tpa(origs[i],ts[i+1]);
1716: struct tpa_state *t = *tp;
1717: lb_labeler_steps++;
1718: if (t) {
1719: ts[i] = t;
1720: lb_labeler_automaton++;
1721: }
1722: else {
1723: lb_labeler_dynprog++;
1724: ts[i] = empty_tpa_state();
1725: for (j=1; j<=max_super && i+j<=ninsts; j++) {
1726: struct super_state **superp = lookup_super(origs+i, j);
1727: if (superp!=NULL) {
1728: struct super_state *supers = *superp;
1729: for (; supers!=NULL; supers = supers->next) {
1730: PrimNum s = supers->super;
1731: int jcost;
1732: struct cost *c=super_costs+s;
1733: struct waypoint *wi=&(ts[i]->inst[c->state_in]);
1734: struct waypoint *wo=&(ts[i+j]->trans[c->state_out]);
1735: int no_transition = wo->no_transition;
1736: lb_applicable_base_rules++;
1737: if (!(is_relocatable(s)) && !wo->relocatable) {
1738: wo=&(ts[i+j]->inst[c->state_out]);
1739: no_transition=1;
1740: }
1741: if (wo->cost == INF_COST)
1742: continue;
1743: jcost = wo->cost + ss_cost(s);
1744: if (jcost <= wi->cost) {
1745: wi->cost = jcost;
1746: wi->inst = s;
1747: wi->relocatable = is_relocatable(s);
1748: wi->no_transition = no_transition;
1749: /* if (ss_greedy) wi->cost = wo->cost ? */
1750: }
1751: }
1752: }
1753: }
1754: transitions(ts[i]);
1755: tpa_state_normalize(ts[i]);
1756: *tp = ts[i] = lookup_tpa_state(ts[i]);
1757: if (tpa_trace)
1758: fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
1759: }
1760: }
1761: /* now rewrite the instructions */
1762: reserve_code_super(origs,ninsts);
1763: old_code_area = code_area;
1764: nextdyn=0;
1765: nextstate=CANONICAL_STATE;
1766: no_transition = ((!ts[0]->trans[nextstate].relocatable)
1767: ||ts[0]->trans[nextstate].no_transition);
1768: for (i=0; i<ninsts; i++) {
1769: Cell tc=0, tc2;
1770: if (i==nextdyn) {
1771: if (!no_transition) {
1772: /* process trans */
1773: PrimNum p = ts[i]->trans[nextstate].inst;
1774: struct cost *c = super_costs+p;
1775: assert(ts[i]->trans[nextstate].cost != INF_COST);
1776: assert(c->state_in==nextstate);
1777: tc = compile_prim_dyn(p,NULL);
1778: nextstate = c->state_out;
1779: }
1780: {
1781: /* process inst */
1782: PrimNum p = ts[i]->inst[nextstate].inst;
1783: struct cost *c=super_costs+p;
1784: assert(c->state_in==nextstate);
1785: assert(ts[i]->inst[nextstate].cost != INF_COST);
1786: #if defined(GFORTH_DEBUGGING)
1787: assert(p == origs[i]);
1788: #endif
1789: tc2 = compile_prim_dyn(p,instps[i]);
1790: if (no_transition || !is_relocatable(p))
1791: /* !! actually what we care about is if and where
1792: * compile_prim_dyn() puts NEXTs */
1793: tc=tc2;
1794: no_transition = ts[i]->inst[nextstate].no_transition;
1795: nextstate = c->state_out;
1796: nextdyn += c->length;
1797: }
1798: } else {
1799: #if defined(GFORTH_DEBUGGING)
1800: assert(0);
1801: #endif
1802: tc=0;
1803: /* tc= (Cell)vm_prims[ts[i]->inst[CANONICAL_STATE].inst]; */
1804: }
1805: *(instps[i]) = tc;
1806: }
1807: if (!no_transition) {
1808: PrimNum p = ts[i]->trans[nextstate].inst;
1809: struct cost *c = super_costs+p;
1810: assert(c->state_in==nextstate);
1811: assert(ts[i]->trans[nextstate].cost != INF_COST);
1812: assert(i==nextdyn);
1813: (void)compile_prim_dyn(p,NULL);
1814: nextstate = c->state_out;
1815: }
1816: assert(nextstate==CANONICAL_STATE);
1817: assert(code_area==old_code_area); /* does reserve_code_super() work? */
1818: }
1819: #endif
1820:
1821: /* compile *start, possibly rewriting it into a static and/or dynamic
1822: superinstruction */
1823: void compile_prim1(Cell *start)
1824: {
1825: #if defined(DOUBLY_INDIRECT)
1826: Label prim;
1827:
1828: if (start==NULL)
1829: return;
1830: prim = (Label)*start;
1831: if (prim<((Label)(xts+DOER_MAX)) || prim>((Label)(xts+npriminfos))) {
1832: fprintf(stderr,"compile_prim encountered xt %p\n", prim);
1833: *start=(Cell)prim;
1834: return;
1835: } else {
1836: *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
1837: return;
1838: }
1839: #elif defined(INDIRECT_THREADED)
1840: return;
1841: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1842: static Cell *instps[MAX_BB];
1843: static PrimNum origs[MAX_BB];
1844: static int ninsts=0;
1845: PrimNum prim_num;
1846:
1847: if (start==NULL || ninsts >= MAX_BB ||
1848: (ninsts>0 && superend[origs[ninsts-1]])) {
1849: /* after bb, or at the start of the next bb */
1850: optimize_rewrite(instps,origs,ninsts);
1851: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
1852: ninsts=0;
1853: if (start==NULL) {
1854: align_code();
1855: return;
1856: }
1857: }
1858: prim_num = ((Xt)*start)-vm_prims;
1859: if(prim_num >= npriminfos) {
1860: /* code word */
1861: optimize_rewrite(instps,origs,ninsts);
1862: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
1863: ninsts=0;
1864: append_jump();
1865: *start = *(Cell *)*start;
1866: return;
1867: }
1868: assert(ninsts<MAX_BB);
1869: instps[ninsts] = start;
1870: origs[ninsts] = prim_num;
1871: ninsts++;
1872: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1873: }
1874:
1875: #ifndef STANDALONE
1876: Address gforth_loader(FILE *imagefile, char* filename)
1877: /* returns the address of the image proper (after the preamble) */
1878: {
1879: ImageHeader header;
1880: Address image;
1881: Address imp; /* image+preamble */
1882: Char magic[8];
1883: char magic7; /* size byte of magic number */
1884: Cell preamblesize=0;
1885: Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1886: UCell check_sum;
1887: Cell ausize = ((RELINFOBITS == 8) ? 0 :
1888: (RELINFOBITS == 16) ? 1 :
1889: (RELINFOBITS == 32) ? 2 : 3);
1890: Cell charsize = ((sizeof(Char) == 1) ? 0 :
1891: (sizeof(Char) == 2) ? 1 :
1892: (sizeof(Char) == 4) ? 2 : 3) + ausize;
1893: Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
1894: (sizeof(Cell) == 2) ? 1 :
1895: (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1896: Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
1897: #ifdef WORDS_BIGENDIAN
1898: 0
1899: #else
1900: 1
1901: #endif
1902: ;
1903:
1904: vm_prims = gforth_engine(0,0,0,0,0 sr_call);
1905: check_prims(vm_prims);
1906: prepare_super_table();
1907: #ifndef DOUBLY_INDIRECT
1908: #ifdef PRINT_SUPER_LENGTHS
1909: print_super_lengths();
1910: #endif
1911: check_sum = checksum(vm_prims);
1912: #else /* defined(DOUBLY_INDIRECT) */
1913: check_sum = (UCell)vm_prims;
1914: #endif /* defined(DOUBLY_INDIRECT) */
1915: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1916: termstate = make_termstate();
1917: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1918:
1919: do {
1920: if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1921: fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.8) image.\n",
1922: progname, filename);
1923: exit(1);
1924: }
1925: preamblesize+=8;
1926: } while(memcmp(magic,"Gforth4",7));
1927: magic7 = magic[7];
1928: if (debug) {
1929: magic[7]='\0';
1930: fprintf(stderr,"Magic found: %s ", magic);
1931: print_sizes(magic7);
1932: }
1933:
1934: if (magic7 != sizebyte)
1935: {
1936: fprintf(stderr,"This image is: ");
1937: print_sizes(magic7);
1938: fprintf(stderr,"whereas the machine is ");
1939: print_sizes(sizebyte);
1940: exit(-2);
1941: };
1942:
1943: fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1944:
1945: set_stack_sizes(&header);
1946:
1947: #if HAVE_GETPAGESIZE
1948: pagesize=getpagesize(); /* Linux/GNU libc offers this */
1949: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
1950: pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
1951: #elif PAGESIZE
1952: pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
1953: #endif
1954: debugp(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1955:
1956: image = dict_alloc_read(imagefile, preamblesize+header.image_size,
1957: dictsize, data_offset);
1958: imp=image+preamblesize;
1959:
1960: alloc_stacks((ImageHeader *)imp);
1961: if (clear_dictionary)
1962: memset(imp+header.image_size, 0, dictsize-header.image_size-preamblesize);
1963: if(header.base==0 || header.base == (Address)0x100) {
1964: Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
1965: Char reloc_bits[reloc_size];
1966: fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1967: fread(reloc_bits, 1, reloc_size, imagefile);
1968: gforth_relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1969: #if 0
1970: { /* let's see what the relocator did */
1971: FILE *snapshot=fopen("snapshot.fi","wb");
1972: fwrite(image,1,imagesize,snapshot);
1973: fclose(snapshot);
1974: }
1975: #endif
1976: }
1977: else if(header.base!=imp) {
1978: fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
1979: progname, (unsigned long)header.base, (unsigned long)imp);
1980: exit(1);
1981: }
1982: if (header.checksum==0)
1983: ((ImageHeader *)imp)->checksum=check_sum;
1984: else if (header.checksum != check_sum) {
1985: fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
1986: progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
1987: exit(1);
1988: }
1989: #ifdef DOUBLY_INDIRECT
1990: ((ImageHeader *)imp)->xt_base = xts;
1991: #endif
1992: fclose(imagefile);
1993:
1994: /* unnecessary, except maybe for CODE words */
1995: /* FLUSH_ICACHE(imp, header.image_size);*/
1996:
1997: return imp;
1998: }
1999: #endif
2000:
2001: /* pointer to last '/' or '\' in file, 0 if there is none. */
2002: static char *onlypath(char *filename)
2003: {
2004: return strrchr(filename, DIRSEP);
2005: }
2006:
2007: static FILE *openimage(char *fullfilename)
2008: {
2009: FILE *image_file;
2010: char * expfilename = tilde_cstr((Char *)fullfilename, strlen(fullfilename));
2011:
2012: image_file=fopen(expfilename,"rb");
2013: if (image_file!=NULL && debug)
2014: fprintf(stderr, "Opened image file: %s\n", expfilename);
2015: free(expfilename);
2016: return image_file;
2017: }
2018:
2019: /* try to open image file concat(path[0:len],imagename) */
2020: static FILE *checkimage(char *path, int len, char *imagename)
2021: {
2022: int dirlen=len;
2023: char fullfilename[dirlen+strlen((char *)imagename)+2];
2024:
2025: memcpy(fullfilename, path, dirlen);
2026: if (fullfilename[dirlen-1]!=DIRSEP)
2027: fullfilename[dirlen++]=DIRSEP;
2028: strcpy(fullfilename+dirlen,imagename);
2029: return openimage(fullfilename);
2030: }
2031:
2032: static FILE * open_image_file(char * imagename, char * path)
2033: {
2034: FILE * image_file=NULL;
2035: char *origpath=path;
2036:
2037: if(strchr(imagename, DIRSEP)==NULL) {
2038: /* first check the directory where the exe file is in !! 01may97jaw */
2039: if (onlypath(progname))
2040: image_file=checkimage(progname, onlypath(progname)-progname, imagename);
2041: if (!image_file)
2042: do {
2043: char *pend=strchr(path, PATHSEP);
2044: if (pend==NULL)
2045: pend=path+strlen(path);
2046: if (strlen(path)==0) break;
2047: image_file=checkimage(path, pend-path, imagename);
2048: path=pend+(*pend==PATHSEP);
2049: } while (image_file==NULL);
2050: } else {
2051: image_file=openimage(imagename);
2052: }
2053:
2054: if (!image_file) {
2055: fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
2056: progname, imagename, origpath);
2057: exit(1);
2058: }
2059:
2060: return image_file;
2061: }
2062: #endif
2063:
2064: #ifdef STANDALONE_ALLOC
2065: Address gforth_alloc(Cell size)
2066: {
2067: Address r;
2068: /* leave a little room (64B) for stack underflows */
2069: if ((r = malloc(size+64))==NULL) {
2070: perror(progname);
2071: exit(1);
2072: }
2073: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
2074: debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
2075: return r;
2076: }
2077: #endif
2078:
2079: #ifdef HAS_OS
2080: static UCell convsize(char *s, UCell elemsize)
2081: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
2082: of bytes. the letter at the end indicates the unit, where e stands
2083: for the element size. default is e */
2084: {
2085: char *endp;
2086: UCell n,m;
2087:
2088: m = elemsize;
2089: n = strtoul(s,&endp,0);
2090: if (endp!=NULL) {
2091: if (strcmp(endp,"b")==0)
2092: m=1;
2093: else if (strcmp(endp,"k")==0)
2094: m=1024;
2095: else if (strcmp(endp,"M")==0)
2096: m=1024*1024;
2097: else if (strcmp(endp,"G")==0)
2098: m=1024*1024*1024;
2099: else if (strcmp(endp,"T")==0) {
2100: #if (SIZEOF_CHAR_P > 4)
2101: m=1024L*1024*1024*1024;
2102: #else
2103: fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
2104: exit(1);
2105: #endif
2106: } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
2107: fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
2108: exit(1);
2109: }
2110: }
2111: return n*m;
2112: }
2113:
2114: enum {
2115: ss_number = 256,
2116: ss_states,
2117: ss_min_codesize,
2118: ss_min_ls,
2119: ss_min_lsu,
2120: ss_min_nexts,
2121: opt_code_block_size,
2122: };
2123:
2124: #ifndef STANDALONE
2125: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
2126: {
2127: int c;
2128:
2129: opterr=0;
2130: while (1) {
2131: int option_index=0;
2132: static struct option opts[] = {
2133: {"appl-image", required_argument, NULL, 'a'},
2134: {"image-file", required_argument, NULL, 'i'},
2135: {"dictionary-size", required_argument, NULL, 'm'},
2136: {"data-stack-size", required_argument, NULL, 'd'},
2137: {"return-stack-size", required_argument, NULL, 'r'},
2138: {"fp-stack-size", required_argument, NULL, 'f'},
2139: {"locals-stack-size", required_argument, NULL, 'l'},
2140: {"vm-commit", no_argument, &map_noreserve, 0},
2141: {"path", required_argument, NULL, 'p'},
2142: {"version", no_argument, NULL, 'v'},
2143: {"help", no_argument, NULL, 'h'},
2144: /* put something != 0 into offset_image */
2145: {"offset-image", no_argument, &offset_image, 1},
2146: {"no-offset-im", no_argument, &offset_image, 0},
2147: {"clear-dictionary", no_argument, &clear_dictionary, 1},
2148: {"debug", no_argument, &debug, 1},
2149: {"diag", no_argument, &diag, 1},
2150: {"die-on-signal", no_argument, &die_on_signal, 1},
2151: {"ignore-async-signals", no_argument, &ignore_async_signals, 1},
2152: {"no-super", no_argument, &no_super, 1},
2153: {"no-dynamic", no_argument, &no_dynamic, 1},
2154: {"dynamic", no_argument, &no_dynamic, 0},
2155: {"code-block-size", required_argument, NULL, opt_code_block_size},
2156: {"print-metrics", no_argument, &print_metrics, 1},
2157: {"print-sequences", no_argument, &print_sequences, 1},
2158: {"ss-number", required_argument, NULL, ss_number},
2159: {"ss-states", required_argument, NULL, ss_states},
2160: #ifndef NO_DYNAMIC
2161: {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
2162: #endif
2163: {"ss-min-ls", no_argument, NULL, ss_min_ls},
2164: {"ss-min-lsu", no_argument, NULL, ss_min_lsu},
2165: {"ss-min-nexts", no_argument, NULL, ss_min_nexts},
2166: {"ss-greedy", no_argument, &ss_greedy, 1},
2167: {"tpa-noequiv", no_argument, &tpa_noequiv, 1},
2168: {"tpa-noautomaton", no_argument, &tpa_noautomaton, 1},
2169: {"tpa-trace", no_argument, &tpa_trace, 1},
2170: {0,0,0,0}
2171: /* no-init-file, no-rc? */
2172: };
2173:
2174: c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
2175:
2176: switch (c) {
2177: case EOF: return;
2178: case '?': optind--; return;
2179: case 'a': *imagename = optarg; return;
2180: case 'i': *imagename = optarg; break;
2181: case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
2182: case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
2183: case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
2184: case 'f': fsize = convsize(optarg,sizeof(Float)); break;
2185: case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
2186: case 'p': *path = optarg; break;
2187: case 'o': offset_image = 1; break;
2188: case 'n': offset_image = 0; break;
2189: case 'c': clear_dictionary = 1; break;
2190: case 's': die_on_signal = 1; break;
2191: case 'x': debug = 1; break;
2192: case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
2193: case opt_code_block_size: code_area_size = atoi(optarg); break;
2194: case ss_number: static_super_number = atoi(optarg); break;
2195: case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
2196: #ifndef NO_DYNAMIC
2197: case ss_min_codesize: ss_cost = cost_codesize; break;
2198: #endif
2199: case ss_min_ls: ss_cost = cost_ls; break;
2200: case ss_min_lsu: ss_cost = cost_lsu; break;
2201: case ss_min_nexts: ss_cost = cost_nexts; break;
2202: case 'h':
2203: fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
2204: Engine Options:\n\
2205: --appl-image FILE Equivalent to '--image-file=FILE --'\n\
2206: --clear-dictionary Initialize the dictionary with 0 bytes\n\
2207: --code-block-size=SIZE size of native code blocks [512KB]\n\
2208: -d SIZE, --data-stack-size=SIZE Specify data stack size\n\
2209: --debug Print debugging information during startup\n\
2210: --diag Print diagnostic information during startup\n\
2211: --die-on-signal Exit instead of THROWing some signals\n\
2212: --dynamic Use dynamic native code\n\
2213: -f SIZE, --fp-stack-size=SIZE Specify floating point stack size\n\
2214: -h, --help Print this message and exit\n\
2215: --ignore-async-signals Ignore instead of THROWing async. signals\n\
2216: -i FILE, --image-file=FILE Use image FILE instead of `gforth.fi'\n\
2217: -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
2218: -m SIZE, --dictionary-size=SIZE Specify Forth dictionary size\n\
2219: --no-dynamic Use only statically compiled primitives\n\
2220: --no-offset-im Load image at normal position\n\
2221: --no-super No dynamically formed superinstructions\n\
2222: --offset-image Load image at a different position\n\
2223: -p PATH, --path=PATH Search path for finding image and sources\n\
2224: --print-metrics Print some code generation metrics on exit\n\
2225: --print-sequences Print primitive sequences for optimization\n\
2226: -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
2227: --ss-greedy Greedy, not optimal superinst selection\n\
2228: --ss-min-codesize Select superinsts for smallest native code\n\
2229: --ss-min-ls Minimize loads and stores\n\
2230: --ss-min-lsu Minimize loads, stores, and pointer updates\n\
2231: --ss-min-nexts Minimize the number of static superinsts\n\
2232: --ss-number=N Use N static superinsts (default max)\n\
2233: --ss-states=N N states for stack caching (default max)\n\
2234: --tpa-noequiv Automaton without state equivalence\n\
2235: --tpa-noautomaton Dynamic programming only\n\
2236: --tpa-trace Report new states etc.\n\
2237: -v, --version Print engine version and exit\n\
2238: --vm-commit Use OS default for memory overcommit\n\
2239: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
2240: `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
2241: argv[0]);
2242: optind--;
2243: return;
2244: }
2245: }
2246: }
2247: #endif
2248: #endif
2249:
2250: static void print_diag()
2251: {
2252:
2253: #if !defined(HAVE_GETRUSAGE)
2254: fprintf(stderr, "*** missing functionality ***\n"
2255: #ifndef HAVE_GETRUSAGE
2256: " no getrusage -> CPUTIME broken\n"
2257: #endif
2258: );
2259: #endif
2260: if((relocs < nonrelocs) ||
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)
2262: 1
2263: #else
2264: 0
2265: #endif
2266: )
2267: debugp(stderr, "relocs: %d:%d\n", relocs, nonrelocs);
2268: fprintf(stderr, "*** %sperformance problems ***\n%s%s",
2269: #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)
2270: "",
2271: #else
2272: "no ",
2273: #endif
2274: #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)
2275: " double-cell integer type buggy ->\n "
2276: #ifdef BUGGY_LL_CMP
2277: "double comparisons, "
2278: #endif
2279: #ifdef BUGGY_LL_MUL
2280: "*/MOD */ M* UM* "
2281: #endif
2282: #ifdef BUGGY_LL_DIV
2283: /* currently nothing is affected */
2284: #endif
2285: #ifdef BUGGY_LL_ADD
2286: "M+ D+ D- DNEGATE "
2287: #endif
2288: #ifdef BUGGY_LL_SHIFT
2289: "D2/ "
2290: #endif
2291: #ifdef BUGGY_LL_D2F
2292: "D>F "
2293: #endif
2294: #ifdef BUGGY_LL_F2D
2295: "F>D "
2296: #endif
2297: "\b\b slow\n"
2298: #endif
2299: #if !(defined(FORCE_REG) || defined(FORCE_REG_UNNECESSARY))
2300: " automatic register allocation: performance degradation possible\n"
2301: #endif
2302: "",
2303: (relocs < nonrelocs) ? "no dynamic code generation (--debug for details) -> factor 2 slowdown\n" : "");
2304: }
2305:
2306: #ifdef STANDALONE
2307: Cell data_abort_pc;
2308:
2309: void data_abort_C(void)
2310: {
2311: while(1) {
2312: }
2313: }
2314: #endif
2315:
2316: void* gforth_pointers(Cell n)
2317: {
2318: switch(n) {
2319: case 0: return (void*)&gforth_SP;
2320: case 1: return (void*)&gforth_FP;
2321: case 2: return (void*)&gforth_LP;
2322: case 3: return (void*)&gforth_RP;
2323: case 4: return (void*)&gforth_UP;
2324: case 5: return (void*)&gforth_engine;
2325: #ifdef HAS_FILE
2326: case 6: return (void*)&cstr;
2327: case 7: return (void*)&tilde_cstr;
2328: #endif
2329: case 8: return (void*)&throw_jmp_buf;
2330: default: return NULL;
2331: }
2332: }
2333:
2334: int main(int argc, char **argv, char **env)
2335: {
2336: #ifdef HAS_OS
2337: char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
2338: #else
2339: char *path = DEFAULTPATH;
2340: #endif
2341: #ifndef INCLUDE_IMAGE
2342: char *imagename="gforth.fi";
2343: FILE *image_file;
2344: Address image;
2345: #endif
2346: int retvalue;
2347: #if 0 && defined(__i386)
2348: /* disabled because the drawbacks may be worse than the benefits */
2349: /* set 387 precision control to use 53-bit mantissae to avoid most
2350: cases of double rounding */
2351: short fpu_control = 0x027f ;
2352: asm("fldcw %0" : : "m"(fpu_control));
2353: #endif /* defined(__i386) */
2354:
2355: #ifdef MACOSX_DEPLOYMENT_TARGET
2356: setenv("MACOSX_DEPLOYMENT_TARGET", MACOSX_DEPLOYMENT_TARGET, 0);
2357: #endif
2358: #ifdef LTDL_LIBRARY_PATH
2359: setenv("LTDL_LIBRARY_PATH", LTDL_LIBRARY_PATH, 0);
2360: #endif
2361: #ifndef STANDALONE
2362: /* buffering of the user output device */
2363: #ifdef _IONBF
2364: if (isatty(fileno(stdout))) {
2365: fflush(stdout);
2366: setvbuf(stdout,NULL,_IONBF,0);
2367: }
2368: #endif
2369: setlocale(LC_ALL, "");
2370: setlocale(LC_NUMERIC, "C");
2371: #else
2372: prep_terminal();
2373: #endif
2374:
2375: progname = argv[0];
2376:
2377: #ifndef STANDALONE
2378: #ifdef HAVE_LIBLTDL
2379: if (lt_dlinit()!=0) {
2380: fprintf(stderr,"%s: lt_dlinit failed", progname);
2381: exit(1);
2382: }
2383: #endif
2384:
2385: #ifdef HAS_OS
2386: gforth_args(argc, argv, &path, &imagename);
2387: #ifndef NO_DYNAMIC
2388: init_ss_cost();
2389: #endif /* !defined(NO_DYNAMIC) */
2390: #endif /* defined(HAS_OS) */
2391: #endif
2392: code_here = ((void *)0)+code_area_size;
2393: #ifdef STANDALONE
2394: image = gforth_engine(0, 0, 0, 0, 0 sr_call);
2395: alloc_stacks((ImageHeader *)image);
2396: #else
2397: image_file = open_image_file(imagename, path);
2398: image = gforth_loader(image_file, imagename);
2399: #endif
2400: gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
2401:
2402: if (diag)
2403: print_diag();
2404: {
2405: char path2[strlen(path)+1];
2406: char *p1, *p2;
2407: Cell environ[]= {
2408: (Cell)argc-(optind-1),
2409: (Cell)(argv+(optind-1)),
2410: (Cell)strlen(path),
2411: (Cell)path2};
2412: argv[optind-1] = progname;
2413: /*
2414: for (i=0; i<environ[0]; i++)
2415: printf("%s\n", ((char **)(environ[1]))[i]);
2416: */
2417: /* make path OS-independent by replacing path separators with NUL */
2418: for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
2419: if (*p1==PATHSEP)
2420: *p2 = '\0';
2421: else
2422: *p2 = *p1;
2423: *p2='\0';
2424: retvalue = gforth_go(image, 4, environ);
2425: #if defined(SIGPIPE) && !defined(STANDALONE)
2426: bsd_signal(SIGPIPE, SIG_IGN);
2427: #endif
2428: #ifdef VM_PROFILING
2429: vm_print_profile(stderr);
2430: #endif
2431: deprep_terminal();
2432: #ifndef STANDALONE
2433: #ifdef HAVE_LIBLTDL
2434: if (lt_dlexit()!=0)
2435: fprintf(stderr,"%s: lt_dlexit failed", progname);
2436: #endif
2437: #endif
2438: }
2439: if (print_metrics) {
2440: int i;
2441: fprintf(stderr, "code size = %8ld\n", dyncodesize());
2442: #ifndef STANDALONE
2443: for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
2444: fprintf(stderr, "metric %8s: %8ld\n",
2445: cost_sums[i].metricname, cost_sums[i].sum);
2446: #endif
2447: fprintf(stderr,"lb_basic_blocks = %ld\n", lb_basic_blocks);
2448: fprintf(stderr,"lb_labeler_steps = %ld\n", lb_labeler_steps);
2449: fprintf(stderr,"lb_labeler_automaton = %ld\n", lb_labeler_automaton);
2450: fprintf(stderr,"lb_labeler_dynprog = %ld\n", lb_labeler_dynprog);
2451: fprintf(stderr,"lb_newstate_equiv = %ld\n", lb_newstate_equiv);
2452: fprintf(stderr,"lb_newstate_new = %ld\n", lb_newstate_new);
2453: fprintf(stderr,"lb_applicable_base_rules = %ld\n", lb_applicable_base_rules);
2454: fprintf(stderr,"lb_applicable_chain_rules = %ld\n", lb_applicable_chain_rules);
2455: }
2456: if (tpa_trace) {
2457: fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
2458: fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
2459: }
2460: return retvalue;
2461: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>