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: #ifndef HAS_LINKBACK
72: __thread void * gforth_pointers[8];
73: #endif
74:
75: #ifdef HAS_FFCALL
76:
77: #include <callback.h>
78:
79: __thread va_alist gforth_clist;
80:
81: void gforth_callback(Xt* fcall, void * alist)
82: {
83: /* save global valiables */
84: Cell *rp = gforth_RP;
85: Cell *sp = gforth_SP;
86: Float *fp = gforth_FP;
87: Address lp = gforth_LP;
88: va_alist clist = gforth_clist;
89:
90: gforth_clist = (va_alist)alist;
91:
92: gforth_engine(fcall, sp, rp, fp, lp sr_call);
93:
94: /* restore global variables */
95: gforth_RP = rp;
96: gforth_SP = sp;
97: gforth_FP = fp;
98: gforth_LP = lp;
99: gforth_clist = clist;
100: }
101: #endif
102:
103: #ifdef GFORTH_DEBUGGING
104: /* define some VM registers as global variables, so they survive exceptions;
105: global register variables are not up to the task (according to the
106: GNU C manual) */
107: #if defined(GLOBALS_NONRELOC)
108: saved_regs saved_regs_v;
109: __thread saved_regs *saved_regs_p = &saved_regs_v;
110: #else /* !defined(GLOBALS_NONRELOC) */
111: __thread Xt *saved_ip;
112: __thread Cell *rp;
113: #endif /* !defined(GLOBALS_NONRELOC) */
114: #endif /* !defined(GFORTH_DEBUGGING) */
115:
116: #ifdef NO_IP
117: Label next_code;
118: #endif
119:
120: #ifdef HAS_FILE
121: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
122: char* pfileattr[6]={"r","r","r+","r+","w","w"};
123:
124: #ifndef O_BINARY
125: #define O_BINARY 0
126: #endif
127: #ifndef O_TEXT
128: #define O_TEXT 0
129: #endif
130:
131: int ufileattr[6]= {
132: O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
133: O_RDWR |O_BINARY, O_RDWR |O_BINARY,
134: O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
135: #endif
136: /* end global vars for engine.c */
137:
138: #define PRIM_VERSION 1
139: /* increment this whenever the primitives change in an incompatible way */
140:
141: #ifndef DEFAULTPATH
142: # define DEFAULTPATH "."
143: #endif
144:
145: #ifdef MSDOS
146: jmp_buf throw_jmp_buf;
147: #endif
148:
149: #if defined(DOUBLY_INDIRECT)
150: # define CFA(n) ({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
151: #else
152: # define CFA(n) ((Cell)(symbols+((n)&~0x4000UL)))
153: #endif
154:
155: #define maxaligned(n) (typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
156:
157: static UCell dictsize=0;
158: static UCell dsize=0;
159: static UCell rsize=0;
160: static UCell fsize=0;
161: static UCell lsize=0;
162: int offset_image=0;
163: int die_on_signal=0;
164: int ignore_async_signals=0;
165: #ifndef INCLUDE_IMAGE
166: static int clear_dictionary=0;
167: UCell pagesize=1;
168: char *progname;
169: #else
170: char *progname = "gforth";
171: int optind = 1;
172: #endif
173: #ifndef MAP_NORESERVE
174: #define MAP_NORESERVE 0
175: #endif
176: /* IF you have an old Cygwin, this may help:
177: #ifdef __CYGWIN__
178: #define MAP_NORESERVE 0
179: #endif
180: */
181: static int map_noreserve=MAP_NORESERVE;
182:
183: #define CODE_BLOCK_SIZE (512*1024) /* !! overflow handling for -native */
184: Address code_area=0;
185: Cell code_area_size = CODE_BLOCK_SIZE;
186: Address code_here; /* does for code-area what HERE does for the dictionary */
187: Address start_flush=NULL; /* start of unflushed code */
188: Cell last_jump=0; /* if the last prim was compiled without jump, this
189: is it's number, otherwise this contains 0 */
190:
191: static int no_super=0; /* true if compile_prim should not fuse prims */
192: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
193: dynamically */
194: static int print_metrics=0; /* if true, print metrics on exit */
195: static int static_super_number = 10000; /* number of ss used if available */
196: #define MAX_STATE 9 /* maximum number of states */
197: static int maxstates = MAX_STATE; /* number of states for stack caching */
198: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
199: static int diag = 0; /* if true: print diagnostic informations */
200: static int tpa_noequiv = 0; /* if true: no state equivalence checking */
201: static int tpa_noautomaton = 0; /* if true: no tree parsing automaton */
202: static int tpa_trace = 0; /* if true: data for line graph of new states etc. */
203: static int print_sequences = 0; /* print primitive sequences for optimization */
204: static int relocs = 0;
205: static int nonrelocs = 0;
206:
207: #ifdef HAS_DEBUG
208: int debug=0;
209: # define debugp(x...) do { if (debug) fprintf(x); } while (0)
210: #else
211: # define perror(x...)
212: # define fprintf(x...)
213: # define debugp(x...)
214: #endif
215:
216: ImageHeader *gforth_header;
217: Label *vm_prims;
218: #ifdef DOUBLY_INDIRECT
219: Label *xts; /* same content as vm_prims, but should only be used for xts */
220: #endif
221:
222: #ifndef NO_DYNAMIC
223: #ifndef CODE_ALIGNMENT
224: #define CODE_ALIGNMENT 0
225: #endif
226:
227: #define MAX_IMMARGS 2
228:
229: typedef struct {
230: Label start; /* NULL if not relocatable */
231: Cell length; /* only includes the jump iff superend is true*/
232: Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
233: char superend; /* true if primitive ends superinstruction, i.e.,
234: unconditional branch, execute, etc. */
235: Cell nimmargs;
236: struct immarg {
237: Cell offset; /* offset of immarg within prim */
238: char rel; /* true if immarg is relative */
239: } immargs[MAX_IMMARGS];
240: } PrimInfo;
241:
242: PrimInfo *priminfos;
243: PrimInfo **decomp_prims;
244:
245: const char const* const prim_names[]={
246: #include PRIM_NAMES_I
247: };
248:
249: void init_ss_cost(void);
250:
251: static int is_relocatable(int p)
252: {
253: return !no_dynamic && priminfos[p].start != NULL;
254: }
255: #else /* defined(NO_DYNAMIC) */
256: static int is_relocatable(int p)
257: {
258: return 0;
259: }
260: #endif /* defined(NO_DYNAMIC) */
261:
262: #ifdef MEMCMP_AS_SUBROUTINE
263: int gforth_memcmp(const char * s1, const char * s2, size_t n)
264: {
265: return memcmp(s1, s2, n);
266: }
267:
268: Char *gforth_memmove(Char * dest, const Char* src, Cell n)
269: {
270: return memmove(dest, src, n);
271: }
272:
273: Char *gforth_memset(Char * s, Cell c, UCell n)
274: {
275: return memset(s, c, n);
276: }
277:
278: Char *gforth_memcpy(Char * dest, const Char* src, Cell n)
279: {
280: return memcpy(dest, src, n);
281: }
282: #endif
283:
284: static Cell max(Cell a, Cell b)
285: {
286: return a>b?a:b;
287: }
288:
289: static Cell min(Cell a, Cell b)
290: {
291: return a<b?a:b;
292: }
293:
294: #ifndef STANDALONE
295: /* image file format:
296: * "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
297: * padding to a multiple of 8
298: * magic: "Gforth4x" means format 0.8,
299: * where x is a byte with
300: * bit 7: reserved = 0
301: * bit 6:5: address unit size 2^n octets
302: * bit 4:3: character size 2^n octets
303: * bit 2:1: cell size 2^n octets
304: * bit 0: endian, big=0, little=1.
305: * The magic are always 8 octets, no matter what the native AU/character size is
306: * padding to max alignment (no padding necessary on current machines)
307: * ImageHeader structure (see forth.h)
308: * data (size in ImageHeader.image_size)
309: * tags ((if relocatable, 1 bit/data cell)
310: *
311: * tag==1 means that the corresponding word is an address;
312: * If the word is >=0, the address is within the image;
313: * addresses within the image are given relative to the start of the image.
314: * If the word =-1 (CF_NIL), the address is NIL,
315: * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
316: * If the word =CF(DODOES), it's a DOES> CFA
317: * !! ABI-CODE and ;ABI-CODE
318: * If the word is <CF(DOER_MAX) and bit 14 is set, it's the xt of a primitive
319: * If the word is <CF(DOER_MAX) and bit 14 is clear,
320: * it's the threaded code of a primitive
321: * bits 13..9 of a primitive token state which group the primitive belongs to,
322: * bits 8..0 of a primitive token index into the group
323: */
324:
325: Cell groups[32] = {
326: 0,
327: 0
328: #undef GROUP
329: #undef GROUPADD
330: #define GROUPADD(n) +n
331: #define GROUP(x, n) , 0
332: #include PRIM_GRP_I
333: #undef GROUP
334: #undef GROUPADD
335: #define GROUP(x, n)
336: #define GROUPADD(n)
337: };
338:
339: static unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
340: int size, Cell base)
341: /* produce a bitmask marking all the branch targets */
342: {
343: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
344: Cell token;
345: unsigned char bits;
346: unsigned char *result=malloc(steps);
347:
348: memset(result, 0, steps);
349: for(k=0; k<steps; k++) {
350: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
351: if(bits & (1U << (RELINFOBITS-1))) {
352: assert(i*sizeof(Cell) < size);
353: token=image[i];
354: if (token>=base) { /* relocatable address */
355: UCell bitnum=(token-base)/sizeof(Cell);
356: if (bitnum/RELINFOBITS < (UCell)steps)
357: result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
358: }
359: }
360: }
361: }
362: return result;
363: }
364:
365: void gforth_relocate(Cell *image, const Char *bitstring,
366: UCell size, Cell base, Label symbols[])
367: {
368: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
369: Cell token;
370: char bits;
371: Cell max_symbols;
372: /*
373: * A virtual start address that's the real start address minus
374: * the one in the image
375: */
376: Cell *start = (Cell * ) (((void *) image) - ((void *) base));
377: unsigned char *targets = branch_targets(image, bitstring, size, base);
378:
379: /* group index into table */
380: if(groups[31]==0) {
381: int groupsum=0;
382: for(i=0; i<32; i++) {
383: groupsum += groups[i];
384: groups[i] = groupsum;
385: /* printf("group[%d]=%d\n",i,groupsum); */
386: }
387: i=0;
388: }
389:
390: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
391:
392: for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
393: ;
394: max_symbols--;
395:
396: for(k=0; k<steps; k++) {
397: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
398: /* fprintf(stderr,"relocate: image[%d]\n", i);*/
399: if(bits & (1U << (RELINFOBITS-1))) {
400: assert(i*sizeof(Cell) < size);
401: /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
402: token=image[i];
403: if(token<0) {
404: int group = (-token & 0x3E00) >> 9;
405: if(group == 0) {
406: switch(token|0x4000) {
407: case CF_NIL : image[i]=0; break;
408: #if !defined(DOUBLY_INDIRECT)
409: case CF(DOCOL) :
410: case CF(DOVAR) :
411: case CF(DOCON) :
412: case CF(DOVAL) :
413: case CF(DOUSER) :
414: case CF(DODEFER) :
415: case CF(DOFIELD) :
416: case CF(DODOES) :
417: case CF(DOABICODE) :
418: case CF(DOSEMIABICODE):
419: MAKE_CF(image+i,symbols[CF(token)]); break;
420: #endif /* !defined(DOUBLY_INDIRECT) */
421: default : /* backward compatibility */
422: /* printf("Code field generation image[%x]:=CFA(%x)\n",
423: i, CF(image[i])); */
424: if (CF((token | 0x4000))<max_symbols) {
425: image[i]=(Cell)CFA(CF(token));
426: #ifdef DIRECT_THREADED
427: if ((token & 0x4000) == 0) { /* threade code, no CFA */
428: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
429: compile_prim1(0);
430: compile_prim1(&image[i]);
431: }
432: #endif
433: } else
434: 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);
435: }
436: } else {
437: int tok = -token & 0x1FF;
438: if (tok < (groups[group+1]-groups[group])) {
439: #if defined(DOUBLY_INDIRECT)
440: image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
441: #else
442: image[i]=(Cell)CFA((groups[group]+tok));
443: #endif
444: #ifdef DIRECT_THREADED
445: if ((token & 0x4000) == 0) { /* threade code, no CFA */
446: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
447: compile_prim1(0);
448: compile_prim1(&image[i]);
449: }
450: #endif
451: } else
452: fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
453: }
454: } else {
455: /* if base is > 0: 0 is a null reference so don't adjust*/
456: if (token>=base) {
457: image[i]+=(Cell)start;
458: }
459: }
460: }
461: }
462: }
463: free(targets);
464: finish_code();
465: ((ImageHeader*)(image))->base = (Address) image;
466: }
467:
468: #ifndef DOUBLY_INDIRECT
469: static UCell checksum(Label symbols[])
470: {
471: UCell r=PRIM_VERSION;
472: Cell i;
473:
474: for (i=DOCOL; i<=DOER_MAX; i++) {
475: r ^= (UCell)(symbols[i]);
476: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
477: }
478: #ifdef DIRECT_THREADED
479: /* we have to consider all the primitives */
480: for (; symbols[i]!=(Label)0; i++) {
481: r ^= (UCell)(symbols[i]);
482: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
483: }
484: #else
485: /* in indirect threaded code all primitives are accessed through the
486: symbols table, so we just have to put the base address of symbols
487: in the checksum */
488: r ^= (UCell)symbols;
489: #endif
490: return r;
491: }
492: #endif
493:
494: static Address verbose_malloc(Cell size)
495: {
496: Address r;
497: /* leave a little room (64B) for stack underflows */
498: if ((r = malloc(size+64))==NULL) {
499: perror(progname);
500: exit(1);
501: }
502: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
503: debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
504: return r;
505: }
506:
507: static void *next_address=0;
508: static void after_alloc(Address r, Cell size)
509: {
510: if (r != (Address)-1) {
511: debugp(stderr, "success, address=$%lx\n", (long) r);
512: #if 0
513: /* not needed now that we protect the stacks with mprotect */
514: if (pagesize != 1)
515: next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
516: #endif
517: } else {
518: debugp(stderr, "failed: %s\n", strerror(errno));
519: }
520: }
521:
522: #ifndef MAP_FAILED
523: #define MAP_FAILED ((Address) -1)
524: #endif
525: #ifndef MAP_FILE
526: # define MAP_FILE 0
527: #endif
528: #ifndef MAP_PRIVATE
529: # define MAP_PRIVATE 0
530: #endif
531: #ifndef PROT_NONE
532: # define PROT_NONE 0
533: #endif
534: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
535: # define MAP_ANON MAP_ANONYMOUS
536: #endif
537:
538: #if defined(HAVE_MMAP)
539: static Address alloc_mmap(Cell size)
540: {
541: void *r;
542:
543: #if defined(MAP_ANON)
544: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
545: r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE|map_noreserve, -1, 0);
546: #else /* !defined(MAP_ANON) */
547: /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
548: apparently defaults) */
549: static int dev_zero=-1;
550:
551: if (dev_zero == -1)
552: dev_zero = open("/dev/zero", O_RDONLY);
553: if (dev_zero == -1) {
554: r = MAP_FAILED;
555: debugp(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ",
556: strerror(errno));
557: } else {
558: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
559: r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE|map_noreserve, dev_zero, 0);
560: }
561: #endif /* !defined(MAP_ANON) */
562: after_alloc(r, size);
563: return r;
564: }
565:
566: static void page_noaccess(void *a)
567: {
568: /* try mprotect first; with munmap the page might be allocated later */
569: debugp(stderr, "try mprotect(%p,$%lx,PROT_NONE); ", a, (long)pagesize);
570: if (mprotect(a, pagesize, PROT_NONE)==0) {
571: debugp(stderr, "ok\n");
572: return;
573: }
574: debugp(stderr, "failed: %s\n", strerror(errno));
575: debugp(stderr, "try munmap(%p,$%lx); ", a, (long)pagesize);
576: if (munmap(a,pagesize)==0) {
577: debugp(stderr, "ok\n");
578: return;
579: }
580: debugp(stderr, "failed: %s\n", strerror(errno));
581: }
582:
583: static size_t wholepage(size_t n)
584: {
585: return (n+pagesize-1)&~(pagesize-1);
586: }
587: #endif
588:
589: Address gforth_alloc(Cell size)
590: {
591: #if HAVE_MMAP
592: Address r;
593:
594: r=alloc_mmap(size);
595: if (r!=(Address)MAP_FAILED)
596: return r;
597: #endif /* HAVE_MMAP */
598: /* use malloc as fallback */
599: return verbose_malloc(size);
600: }
601:
602: static void *dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
603: {
604: void *image = MAP_FAILED;
605:
606: #if defined(HAVE_MMAP)
607: if (offset==0) {
608: image=alloc_mmap(dictsize);
609: if (image != (void *)MAP_FAILED) {
610: void *image1;
611: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
612: image1 = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE|map_noreserve, fileno(file), 0);
613: after_alloc(image1,dictsize);
614: if (image1 == (void *)MAP_FAILED)
615: goto read_image;
616: }
617: }
618: #endif /* defined(HAVE_MMAP) */
619: if (image == (void *)MAP_FAILED) {
620: image = gforth_alloc(dictsize+offset)+offset;
621: read_image:
622: rewind(file); /* fseek(imagefile,0L,SEEK_SET); */
623: fread(image, 1, imagesize, file);
624: }
625: return image;
626: }
627: #endif
628:
629: void set_stack_sizes(ImageHeader * header)
630: {
631: if (dictsize==0)
632: dictsize = header->dict_size;
633: if (dsize==0)
634: dsize = header->data_stack_size;
635: if (rsize==0)
636: rsize = header->return_stack_size;
637: if (fsize==0)
638: fsize = header->fp_stack_size;
639: if (lsize==0)
640: lsize = header->locals_stack_size;
641: dictsize=maxaligned(dictsize);
642: dsize=maxaligned(dsize);
643: rsize=maxaligned(rsize);
644: lsize=maxaligned(lsize);
645: fsize=maxaligned(fsize);
646: }
647:
648: #ifdef STANDALONE
649: void alloc_stacks(ImageHeader * h)
650: {
651: #define SSTACKSIZE 0x200
652: static Cell dstack[SSTACKSIZE+1];
653: static Cell rstack[SSTACKSIZE+1];
654:
655: h->dict_size=dictsize;
656: h->data_stack_size=dsize;
657: h->fp_stack_size=fsize;
658: h->return_stack_size=rsize;
659: h->locals_stack_size=lsize;
660:
661: h->data_stack_base=dstack+SSTACKSIZE;
662: // h->fp_stack_base=gforth_alloc(fsize);
663: h->return_stack_base=rstack+SSTACKSIZE;
664: // h->locals_stack_base=gforth_alloc(lsize);
665: }
666: #else
667: void alloc_stacks(ImageHeader * h)
668: {
669: h->dict_size=dictsize;
670: h->data_stack_size=dsize;
671: h->fp_stack_size=fsize;
672: h->return_stack_size=rsize;
673: h->locals_stack_size=lsize;
674:
675: #if defined(HAVE_MMAP) && !defined(STANDALONE)
676: if (pagesize > 1) {
677: size_t p = pagesize;
678: size_t totalsize =
679: wholepage(dsize)+wholepage(fsize)+wholepage(rsize)+wholepage(lsize)+5*p;
680: void *a = alloc_mmap(totalsize);
681: if (a != (void *)MAP_FAILED) {
682: page_noaccess(a); a+=p; h-> data_stack_base=a; a+=wholepage(dsize);
683: page_noaccess(a); a+=p; h-> fp_stack_base=a; a+=wholepage(fsize);
684: page_noaccess(a); a+=p; h->return_stack_base=a; a+=wholepage(rsize);
685: page_noaccess(a); a+=p; h->locals_stack_base=a; a+=wholepage(lsize);
686: page_noaccess(a);
687: debugp(stderr,"stack addresses: d=%p f=%p r=%p l=%p\n",
688: h->data_stack_base,
689: h->fp_stack_base,
690: h->return_stack_base,
691: h->locals_stack_base);
692: return;
693: }
694: }
695: #endif
696: h->data_stack_base=gforth_alloc(dsize);
697: h->fp_stack_base=gforth_alloc(fsize);
698: h->return_stack_base=gforth_alloc(rsize);
699: h->locals_stack_base=gforth_alloc(lsize);
700: }
701: #endif
702:
703: #warning You can ignore the warnings about clobbered variables in gforth_go
704: int gforth_go(void *image, int stack, Cell *entries)
705: {
706: volatile ImageHeader *image_header = (ImageHeader *)image;
707: Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
708: Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
709: Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
710: #ifdef GFORTH_DEBUGGING
711: volatile Cell *orig_rp0=rp0;
712: #endif
713: Address lp0=image_header->locals_stack_base + lsize;
714: Xt *ip0=(Xt *)(image_header->boot_entry);
715: #ifdef SYSSIGNALS
716: int throw_code;
717: #endif
718:
719: /* ensure that the cached elements (if any) are accessible */
720: #if !(defined(GFORTH_DEBUGGING) || defined(INDIRECT_THREADED) || defined(DOUBLY_INDIRECT) || defined(VM_PROFILING))
721: sp0 -= 8; /* make stuff below bottom accessible for stack caching */
722: fp0--;
723: #endif
724:
725: for(;stack>0;stack--)
726: *--sp0=entries[stack-1];
727:
728: #if defined(SYSSIGNALS) && !defined(STANDALONE)
729: get_winsize();
730:
731: install_signal_handlers(); /* right place? */
732:
733: if ((throw_code=setjmp(throw_jmp_buf))) {
734: static Cell signal_data_stack[24];
735: static Cell signal_return_stack[16];
736: static Float signal_fp_stack[1];
737:
738: signal_data_stack[15]=throw_code;
739:
740: #ifdef GFORTH_DEBUGGING
741: debugp(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
742: throw_code, saved_ip, rp);
743: if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
744: /* no rstack overflow or underflow */
745: rp0 = rp;
746: *--rp0 = (Cell)saved_ip;
747: }
748: else /* I love non-syntactic ifdefs :-) */
749: rp0 = signal_return_stack+16;
750: #else /* !defined(GFORTH_DEBUGGING) */
751: debugp(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
752: rp0 = signal_return_stack+16;
753: #endif /* !defined(GFORTH_DEBUGGING) */
754: /* fprintf(stderr, "rp=$%x\n",rp0);*/
755:
756: return((int)(Cell)gforth_engine(image_header->throw_entry, signal_data_stack+15,
757: rp0, signal_fp_stack, 0 sr_call));
758: }
759: #endif
760:
761: return((int)(Cell)gforth_engine(ip0,sp0,rp0,fp0,lp0 sr_call));
762: }
763:
764: #if !defined(INCLUDE_IMAGE) && !defined(STANDALONE)
765: static void print_sizes(Cell sizebyte)
766: /* print size information */
767: {
768: static char* endianstring[]= { " big","little" };
769:
770: fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
771: endianstring[sizebyte & 1],
772: 1 << ((sizebyte >> 1) & 3),
773: 1 << ((sizebyte >> 3) & 3),
774: 1 << ((sizebyte >> 5) & 3));
775: }
776:
777: /* static superinstruction stuff */
778:
779: struct cost { /* super_info might be a more accurate name */
780: char loads; /* number of stack loads */
781: char stores; /* number of stack stores */
782: char updates; /* number of stack pointer updates */
783: char branch; /* is it a branch (SET_IP) */
784: unsigned char state_in; /* state on entry */
785: unsigned char state_out; /* state on exit */
786: unsigned char imm_ops; /* number of immediate operands */
787: short offset; /* offset into super2 table */
788: unsigned char length; /* number of components */
789: };
790:
791: PrimNum super2[] = {
792: #include SUPER2_I
793: };
794:
795: struct cost super_costs[] = {
796: #include COSTS_I
797: };
798:
799: struct super_state {
800: struct super_state *next;
801: PrimNum super;
802: };
803:
804: #define HASH_SIZE 256
805:
806: struct super_table_entry {
807: struct super_table_entry *next;
808: PrimNum *start;
809: short length;
810: struct super_state *ss_list; /* list of supers */
811: } *super_table[HASH_SIZE];
812: int max_super=2;
813:
814: struct super_state *state_transitions=NULL;
815:
816: static int hash_super(PrimNum *start, int length)
817: {
818: int i, r;
819:
820: for (i=0, r=0; i<length; i++) {
821: r <<= 1;
822: r += start[i];
823: }
824: return r & (HASH_SIZE-1);
825: }
826:
827: static struct super_state **lookup_super(PrimNum *start, int length)
828: {
829: int hash=hash_super(start,length);
830: struct super_table_entry *p = super_table[hash];
831:
832: /* assert(length >= 2); */
833: for (; p!=NULL; p = p->next) {
834: if (length == p->length &&
835: memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
836: return &(p->ss_list);
837: }
838: return NULL;
839: }
840:
841: static void prepare_super_table()
842: {
843: int i;
844: int nsupers = 0;
845:
846: for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
847: struct cost *c = &super_costs[i];
848: if ((c->length < 2 || nsupers < static_super_number) &&
849: c->state_in < maxstates && c->state_out < maxstates) {
850: struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
851: struct super_state *ss = malloc(sizeof(struct super_state));
852: ss->super= i;
853: if (c->offset==N_noop && i != N_noop) {
854: if (is_relocatable(i)) {
855: ss->next = state_transitions;
856: state_transitions = ss;
857: }
858: } else if (ss_listp != NULL) {
859: ss->next = *ss_listp;
860: *ss_listp = ss;
861: } else {
862: int hash = hash_super(super2+c->offset, c->length);
863: struct super_table_entry **p = &super_table[hash];
864: struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
865: ss->next = NULL;
866: e->next = *p;
867: e->start = super2 + c->offset;
868: e->length = c->length;
869: e->ss_list = ss;
870: *p = e;
871: }
872: if (c->length > max_super)
873: max_super = c->length;
874: if (c->length >= 2)
875: nsupers++;
876: }
877: }
878: debugp(stderr, "Using %d static superinsts\n", nsupers);
879: if (nsupers>0 && !tpa_noautomaton && !tpa_noequiv) {
880: /* Currently these two things don't work together; see Section 3.2
881: of <http://www.complang.tuwien.ac.at/papers/ertl+06pldi.ps.gz>,
882: in particular Footnote 6 for the reason; hmm, we should be able
883: to use an automaton without state equivalence, but that costs
884: significant space so we only do it if the user explicitly
885: disables state equivalence. */
886: debugp(stderr, "Disabling tpa-automaton, because nsupers>0 and state equivalence is enabled.\n");
887: tpa_noautomaton = 1;
888: }
889: }
890:
891: /* dynamic replication/superinstruction stuff */
892:
893: #ifndef NO_DYNAMIC
894: static int compare_priminfo_length(const void *_a, const void *_b)
895: {
896: PrimInfo **a = (PrimInfo **)_a;
897: PrimInfo **b = (PrimInfo **)_b;
898: Cell diff = (*a)->length - (*b)->length;
899: if (diff)
900: return diff;
901: else /* break ties by start address; thus the decompiler produces
902: the earliest primitive with the same code (e.g. noop instead
903: of (char) and @ instead of >code-address */
904: return (*b)->start - (*a)->start;
905: }
906: #endif /* !defined(NO_DYNAMIC) */
907:
908: static char MAYBE_UNUSED superend[]={
909: #include PRIM_SUPEREND_I
910: };
911:
912: Cell npriminfos=0;
913:
914: Label goto_start;
915: Cell goto_len;
916:
917: #ifndef NO_DYNAMIC
918: static int compare_labels(const void *pa, const void *pb)
919: {
920: Label a = *(Label *)pa;
921: Label b = *(Label *)pb;
922: return a-b;
923: }
924: #endif
925:
926: static Label bsearch_next(Label key, Label *a, UCell n)
927: /* a is sorted; return the label >=key that is the closest in a;
928: return NULL if there is no label in a >=key */
929: {
930: int mid = (n-1)/2;
931: if (n<1)
932: return NULL;
933: if (n == 1) {
934: if (a[0] < key)
935: return NULL;
936: else
937: return a[0];
938: }
939: if (a[mid] < key)
940: return bsearch_next(key, a+mid+1, n-mid-1);
941: else
942: return bsearch_next(key, a, mid+1);
943: }
944:
945: static void check_prims(Label symbols1[])
946: {
947: int i;
948: #ifndef NO_DYNAMIC
949: Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted, *goto_p;
950: int nends1j;
951: #endif
952:
953: if (debug)
954: #ifdef __VERSION__
955: fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
956: #else
957: #define xstr(s) str(s)
958: #define str(s) #s
959: fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n");
960: #endif
961: for (i=0; symbols1[i]!=0; i++)
962: ;
963: npriminfos = i;
964:
965: #ifndef NO_DYNAMIC
966: if (no_dynamic)
967: return;
968: symbols2=gforth_engine2(0,0,0,0,0 sr_call);
969: #if NO_IP
970: symbols3=gforth_engine3(0,0,0,0,0 sr_call);
971: #else
972: symbols3=symbols1;
973: #endif
974: ends1 = symbols1+i+1;
975: ends1j = ends1+i;
976: goto_p = ends1j+i+1; /* goto_p[0]==before; ...[1]==after;*/
977: nends1j = i+1;
978: ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
979: memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
980: qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
981:
982: /* check whether the "goto *" is relocatable */
983: goto_len = goto_p[1]-goto_p[0];
984: debugp(stderr, "goto * %p %p len=%ld\n",
985: goto_p[0],symbols2[goto_p-symbols1],(long)goto_len);
986: if (memcmp(goto_p[0],symbols2[goto_p-symbols1],goto_len)!=0) { /* unequal */
987: no_dynamic=1;
988: debugp(stderr," not relocatable, disabling dynamic code generation\n");
989: init_ss_cost();
990: return;
991: }
992: goto_start = goto_p[0];
993:
994: priminfos = calloc(i,sizeof(PrimInfo));
995: for (i=0; symbols1[i]!=0; i++) {
996: int prim_len = ends1[i]-symbols1[i];
997: PrimInfo *pi=&priminfos[i];
998: struct cost *sc=&super_costs[i];
999: int j=0;
1000: char *s1 = (char *)symbols1[i];
1001: char *s2 = (char *)symbols2[i];
1002: char *s3 = (char *)symbols3[i];
1003: Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
1004:
1005: pi->start = s1;
1006: pi->superend = superend[i]|no_super;
1007: pi->length = prim_len;
1008: pi->restlength = endlabel - symbols1[i] - pi->length;
1009: pi->nimmargs = 0;
1010: relocs++;
1011: #if defined(BURG_FORMAT)
1012: { /* output as burg-style rules */
1013: int p=super_costs[i].offset;
1014: if (p==N_noop)
1015: debugp(stderr, "S%d: S%d = %d (%d);", sc->state_in, sc->state_out, i+1, pi->length);
1016: else
1017: debugp(stderr, "S%d: op%d(S%d) = %d (%d);", sc->state_in, p, sc->state_out, i+1, pi->length);
1018: }
1019: #else
1020: debugp(stderr, "%-15s %d-%d %4d %p %p %p len=%3ld rest=%2ld send=%1d",
1021: prim_names[i], sc->state_in, sc->state_out,
1022: i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength),
1023: pi->superend);
1024: #endif
1025: if (endlabel == NULL) {
1026: pi->start = NULL; /* not relocatable */
1027: if (pi->length<0) pi->length=100;
1028: #ifndef BURG_FORMAT
1029: debugp(stderr,"\n non_reloc: no J label > start found\n");
1030: #endif
1031: relocs--;
1032: nonrelocs++;
1033: continue;
1034: }
1035: if (ends1[i] > endlabel && !pi->superend) {
1036: pi->start = NULL; /* not relocatable */
1037: pi->length = endlabel-symbols1[i];
1038: #ifndef BURG_FORMAT
1039: debugp(stderr,"\n non_reloc: there is a J label before the K label (restlength<0)\n");
1040: #endif
1041: relocs--;
1042: nonrelocs++;
1043: continue;
1044: }
1045: if (ends1[i] < pi->start && !pi->superend) {
1046: pi->start = NULL; /* not relocatable */
1047: pi->length = endlabel-symbols1[i];
1048: #ifndef BURG_FORMAT
1049: debugp(stderr,"\n non_reloc: K label before I label (length<0)\n");
1050: #endif
1051: relocs--;
1052: nonrelocs++;
1053: continue;
1054: }
1055: if (CHECK_PRIM(s1, prim_len)) {
1056: #ifndef BURG_FORMAT
1057: debugp(stderr,"\n non_reloc: architecture specific check failed\n");
1058: #endif
1059: pi->start = NULL; /* not relocatable */
1060: relocs--;
1061: nonrelocs++;
1062: continue;
1063: }
1064: assert(pi->length>=0);
1065: assert(pi->restlength >=0);
1066: while (j<(pi->length+pi->restlength)) {
1067: if (s1[j]==s3[j]) {
1068: if (s1[j] != s2[j]) {
1069: pi->start = NULL; /* not relocatable */
1070: #ifndef BURG_FORMAT
1071: debugp(stderr,"\n non_reloc: engine1!=engine2 offset %3d",j);
1072: #endif
1073: /* assert(j<prim_len); */
1074: relocs--;
1075: nonrelocs++;
1076: break;
1077: }
1078: j++;
1079: } else {
1080: struct immarg *ia=&pi->immargs[pi->nimmargs];
1081:
1082: pi->nimmargs++;
1083: ia->offset=j;
1084: if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
1085: ia->rel=0;
1086: debugp(stderr,"\n absolute immarg: offset %3d",j);
1087: } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
1088: symbols1[DOER_MAX+1]) {
1089: ia->rel=1;
1090: debugp(stderr,"\n relative immarg: offset %3d",j);
1091: } else {
1092: pi->start = NULL; /* not relocatable */
1093: #ifndef BURG_FORMAT
1094: debugp(stderr,"\n non_reloc: engine1!=engine3 offset %3d",j);
1095: #endif
1096: /* assert(j<prim_len);*/
1097: relocs--;
1098: nonrelocs++;
1099: break;
1100: }
1101: j+=4;
1102: }
1103: }
1104: debugp(stderr,"\n");
1105: }
1106: decomp_prims = calloc(i,sizeof(PrimInfo *));
1107: for (i=DOER_MAX+1; i<npriminfos; i++)
1108: decomp_prims[i] = &(priminfos[i]);
1109: qsort(decomp_prims+DOER_MAX+1, npriminfos-DOER_MAX-1, sizeof(PrimInfo *),
1110: compare_priminfo_length);
1111: #endif
1112: }
1113:
1114: static void flush_to_here(void)
1115: {
1116: #ifndef NO_DYNAMIC
1117: if (start_flush)
1118: FLUSH_ICACHE((caddr_t)start_flush, code_here-start_flush);
1119: start_flush=code_here;
1120: #endif
1121: }
1122:
1123: static void MAYBE_UNUSED align_code(void)
1124: /* align code_here on some platforms */
1125: {
1126: #ifndef NO_DYNAMIC
1127: #if defined(CODE_PADDING)
1128: Cell alignment = CODE_ALIGNMENT;
1129: static char nops[] = CODE_PADDING;
1130: UCell maxpadding=MAX_PADDING;
1131: UCell offset = ((UCell)code_here)&(alignment-1);
1132: UCell length = alignment-offset;
1133: if (length <= maxpadding) {
1134: memcpy(code_here,nops+offset,length);
1135: code_here += length;
1136: }
1137: #endif /* defined(CODE_PADDING) */
1138: #endif /* defined(NO_DYNAMIC */
1139: }
1140:
1141: #ifndef NO_DYNAMIC
1142: static void append_jump(void)
1143: {
1144: if (last_jump) {
1145: PrimInfo *pi = &priminfos[last_jump];
1146:
1147: memcpy(code_here, pi->start+pi->length, pi->restlength);
1148: code_here += pi->restlength;
1149: memcpy(code_here, goto_start, goto_len);
1150: code_here += goto_len;
1151: align_code();
1152: last_jump=0;
1153: }
1154: }
1155:
1156: /* Gforth remembers all code blocks in this list. On forgetting (by
1157: executing a marker) the code blocks are not freed (because Gforth does
1158: not remember how they were allocated; hmm, remembering that might be
1159: easier and cleaner). Instead, code_here etc. are reset to the old
1160: value, and the "forgotten" code blocks are reused when they are
1161: needed. */
1162:
1163: struct code_block_list {
1164: struct code_block_list *next;
1165: Address block;
1166: Cell size;
1167: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
1168:
1169: static void reserve_code_space(UCell size)
1170: {
1171: if (code_area+code_area_size < code_here+size) {
1172: struct code_block_list *p;
1173: append_jump();
1174: debugp(stderr,"Did not use %ld bytes in code block\n",
1175: (long)(code_area+code_area_size-code_here));
1176: flush_to_here();
1177: if (*next_code_blockp == NULL) {
1178: code_here = start_flush = code_area = gforth_alloc(code_area_size);
1179: p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
1180: *next_code_blockp = p;
1181: p->next = NULL;
1182: p->block = code_here;
1183: p->size = code_area_size;
1184: } else {
1185: p = *next_code_blockp;
1186: code_here = start_flush = code_area = p->block;
1187: }
1188: next_code_blockp = &(p->next);
1189: }
1190: }
1191:
1192: static Address append_prim(Cell p)
1193: {
1194: PrimInfo *pi = &priminfos[p];
1195: Address old_code_here;
1196: reserve_code_space(pi->length+pi->restlength+goto_len+CODE_ALIGNMENT-1);
1197: memcpy(code_here, pi->start, pi->length);
1198: old_code_here = code_here;
1199: code_here += pi->length;
1200: return old_code_here;
1201: }
1202:
1203: static void reserve_code_super(PrimNum origs[], int ninsts)
1204: {
1205: int i;
1206: UCell size = CODE_ALIGNMENT-1; /* alignment may happen first */
1207: if (no_dynamic)
1208: return;
1209: /* use size of the original primitives as an upper bound for the
1210: size of the superinstruction. !! This is only safe if we
1211: optimize for code size (the default) */
1212: for (i=0; i<ninsts; i++) {
1213: PrimNum p = origs[i];
1214: PrimInfo *pi = &priminfos[p];
1215: if (is_relocatable(p))
1216: size += pi->length;
1217: else
1218: if (i>0)
1219: size += priminfos[origs[i-1]].restlength+goto_len+CODE_ALIGNMENT-1;
1220: }
1221: size += priminfos[origs[i-1]].restlength+goto_len;
1222: reserve_code_space(size);
1223: }
1224: #endif
1225:
1226: int forget_dyncode(Address code)
1227: {
1228: #ifdef NO_DYNAMIC
1229: return -1;
1230: #else
1231: struct code_block_list *p, **pp;
1232:
1233: for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
1234: if (code >= p->block && code < p->block+p->size) {
1235: next_code_blockp = &(p->next);
1236: code_here = start_flush = code;
1237: code_area = p->block;
1238: last_jump = 0;
1239: return -1;
1240: }
1241: }
1242: return -no_dynamic;
1243: #endif /* !defined(NO_DYNAMIC) */
1244: }
1245:
1246: static long dyncodesize(void)
1247: {
1248: #ifndef NO_DYNAMIC
1249: struct code_block_list *p;
1250: long size=0;
1251: for (p=code_block_list; p!=NULL; p=p->next) {
1252: if (code_here >= p->block && code_here < p->block+p->size)
1253: return size + (code_here - p->block);
1254: else
1255: size += p->size;
1256: }
1257: #endif /* !defined(NO_DYNAMIC) */
1258: return 0;
1259: }
1260:
1261: Label decompile_code(Label _code)
1262: {
1263: #ifdef NO_DYNAMIC
1264: return _code;
1265: #else /* !defined(NO_DYNAMIC) */
1266: Cell i;
1267: struct code_block_list *p;
1268: Address code=_code;
1269:
1270: /* first, check if we are in code at all */
1271: for (p = code_block_list;; p = p->next) {
1272: if (p == NULL)
1273: return code;
1274: if (code >= p->block && code < p->block+p->size)
1275: break;
1276: }
1277: /* reverse order because NOOP might match other prims */
1278: for (i=npriminfos-1; i>DOER_MAX; i--) {
1279: PrimInfo *pi=decomp_prims[i];
1280: if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1281: return vm_prims[super2[super_costs[pi-priminfos].offset]];
1282: /* return pi->start;*/
1283: }
1284: return code;
1285: #endif /* !defined(NO_DYNAMIC) */
1286: }
1287:
1288: #ifdef NO_IP
1289: int nbranchinfos=0;
1290:
1291: struct branchinfo {
1292: Label **targetpp; /* **(bi->targetpp) is the target */
1293: Cell *addressptr; /* store the target here */
1294: } branchinfos[100000];
1295:
1296: int ndoesexecinfos=0;
1297: struct doesexecinfo {
1298: int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1299: Label *targetp; /*target for branch (because this is not in threaded code)*/
1300: Cell *xt; /* cfa of word whose does-code needs calling */
1301: } doesexecinfos[10000];
1302:
1303: static void set_rel_target(Cell *source, Label target)
1304: {
1305: *source = ((Cell)target)-(((Cell)source)+4);
1306: }
1307:
1308: static void register_branchinfo(Label source, Cell *targetpp)
1309: {
1310: struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1311: bi->targetpp = (Label **)targetpp;
1312: bi->addressptr = (Cell *)source;
1313: nbranchinfos++;
1314: }
1315:
1316: static Address compile_prim1arg(PrimNum p, Cell **argp)
1317: {
1318: Address old_code_here=append_prim(p);
1319:
1320: assert(vm_prims[p]==priminfos[p].start);
1321: *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
1322: return old_code_here;
1323: }
1324:
1325: static Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
1326: {
1327: PrimInfo *pi = &priminfos[N_call2];
1328: Address old_code_here = append_prim(N_call2);
1329:
1330: *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
1331: register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
1332: return old_code_here;
1333: }
1334: #endif
1335:
1336: void finish_code(void)
1337: {
1338: #ifdef NO_IP
1339: Cell i;
1340:
1341: compile_prim1(NULL);
1342: for (i=0; i<ndoesexecinfos; i++) {
1343: struct doesexecinfo *dei = &doesexecinfos[i];
1344: dei->targetp = (Label *)DOES_CODE1((dei->xt));
1345: branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
1346: }
1347: ndoesexecinfos = 0;
1348: for (i=0; i<nbranchinfos; i++) {
1349: struct branchinfo *bi=&branchinfos[i];
1350: set_rel_target(bi->addressptr, **(bi->targetpp));
1351: }
1352: nbranchinfos = 0;
1353: #else
1354: compile_prim1(NULL);
1355: #endif
1356: flush_to_here();
1357: }
1358:
1359: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1360: #ifdef NO_IP
1361: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1362: /* compile prim #p dynamically (mod flags etc.) and return start
1363: address of generated code for putting it into the threaded
1364: code. This function is only called if all the associated
1365: inline arguments of p are already in place (at tcp[1] etc.) */
1366: {
1367: PrimInfo *pi=&priminfos[p];
1368: Cell *next_code_target=NULL;
1369: Address codeaddr;
1370: Address primstart;
1371:
1372: assert(p<npriminfos);
1373: if (p==N_execute || p==N_perform || p==N_lit_perform) {
1374: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1375: primstart = append_prim(p);
1376: goto other_prim;
1377: } else if (p==N_call) {
1378: codeaddr = compile_call2(tcp+1, &next_code_target);
1379: } else if (p==N_does_exec) {
1380: struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1381: Cell *arg;
1382: codeaddr = compile_prim1arg(N_lit,&arg);
1383: *arg = (Cell)PFA(tcp[1]);
1384: /* we cannot determine the callee now (last_start[1] may be a
1385: forward reference), so just register an arbitrary target, and
1386: register in dei that we need to fix this before resolving
1387: branches */
1388: dei->branchinfo = nbranchinfos;
1389: dei->xt = (Cell *)(tcp[1]);
1390: compile_call2(0, &next_code_target);
1391: } else if (!is_relocatable(p)) {
1392: Cell *branch_target;
1393: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1394: compile_prim1arg(N_branch,&branch_target);
1395: set_rel_target(branch_target,vm_prims[p]);
1396: } else {
1397: unsigned j;
1398:
1399: codeaddr = primstart = append_prim(p);
1400: other_prim:
1401: for (j=0; j<pi->nimmargs; j++) {
1402: struct immarg *ia = &(pi->immargs[j]);
1403: Cell *argp = tcp + pi->nimmargs - j;
1404: Cell argval = *argp; /* !! specific to prims */
1405: if (ia->rel) { /* !! assumption: relative refs are branches */
1406: register_branchinfo(primstart + ia->offset, argp);
1407: } else /* plain argument */
1408: *(Cell *)(primstart + ia->offset) = argval;
1409: }
1410: }
1411: if (next_code_target!=NULL)
1412: *next_code_target = (Cell)code_here;
1413: return (Cell)codeaddr;
1414: }
1415: #else /* !defined(NO_IP) */
1416: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1417: /* compile prim #p dynamically (mod flags etc.) and return start
1418: address of generated code for putting it into the threaded code */
1419: {
1420: Cell static_prim = (Cell)vm_prims[p];
1421: #if defined(NO_DYNAMIC)
1422: return static_prim;
1423: #else /* !defined(NO_DYNAMIC) */
1424: Address old_code_here;
1425:
1426: if (no_dynamic)
1427: return static_prim;
1428: if (p>=npriminfos || !is_relocatable(p)) {
1429: append_jump();
1430: return static_prim;
1431: }
1432: old_code_here = append_prim(p);
1433: last_jump = p;
1434: if (priminfos[p].superend)
1435: append_jump();
1436: return (Cell)old_code_here;
1437: #endif /* !defined(NO_DYNAMIC) */
1438: }
1439: #endif /* !defined(NO_IP) */
1440: #endif
1441:
1442: #ifndef NO_DYNAMIC
1443: static int cost_codesize(int prim)
1444: {
1445: return priminfos[prim].length;
1446: }
1447: #endif
1448:
1449: static int cost_ls(int prim)
1450: {
1451: struct cost *c = super_costs+prim;
1452:
1453: return c->loads + c->stores;
1454: }
1455:
1456: static int cost_lsu(int prim)
1457: {
1458: struct cost *c = super_costs+prim;
1459:
1460: return c->loads + c->stores + c->updates;
1461: }
1462:
1463: static int cost_nexts(int prim)
1464: {
1465: return 1;
1466: }
1467:
1468: typedef int Costfunc(int);
1469: Costfunc *ss_cost = /* cost function for optimize_bb */
1470: #ifdef NO_DYNAMIC
1471: cost_lsu;
1472: #else
1473: cost_codesize;
1474: #endif
1475:
1476: struct {
1477: Costfunc *costfunc;
1478: char *metricname;
1479: long sum;
1480: } cost_sums[] = {
1481: #ifndef NO_DYNAMIC
1482: { cost_codesize, "codesize", 0 },
1483: #endif
1484: { cost_ls, "ls", 0 },
1485: { cost_lsu, "lsu", 0 },
1486: { cost_nexts, "nexts", 0 }
1487: };
1488:
1489: #ifndef NO_DYNAMIC
1490: void init_ss_cost(void) {
1491: if (no_dynamic && ss_cost == cost_codesize) {
1492: ss_cost = cost_nexts;
1493: cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
1494: debugp(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
1495: }
1496: }
1497: #endif
1498:
1499: #define MAX_BB 128 /* maximum number of instructions in BB */
1500: #define INF_COST 1000000 /* infinite cost */
1501: #define CANONICAL_STATE 0
1502:
1503: struct waypoint {
1504: int cost; /* the cost from here to the end */
1505: PrimNum inst; /* the inst used from here to the next waypoint */
1506: char relocatable; /* the last non-transition was relocatable */
1507: char no_transition; /* don't use the next transition (relocatability)
1508: * or this transition (does not change state) */
1509: };
1510:
1511: struct tpa_state { /* tree parsing automaton (like) state */
1512: /* labeling is back-to-front */
1513: struct waypoint *inst; /* in front of instruction */
1514: struct waypoint *trans; /* in front of instruction and transition */
1515: };
1516:
1517: struct tpa_state *termstate = NULL; /* initialized in loader() */
1518:
1519: /* statistics about tree parsing (lazyburg) stuff */
1520: long lb_basic_blocks = 0;
1521: long lb_labeler_steps = 0;
1522: long lb_labeler_automaton = 0;
1523: long lb_labeler_dynprog = 0;
1524: long lb_newstate_equiv = 0;
1525: long lb_newstate_new = 0;
1526: long lb_applicable_base_rules = 0;
1527: long lb_applicable_chain_rules = 0;
1528:
1529: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1530: static void init_waypoints(struct waypoint ws[])
1531: {
1532: int k;
1533:
1534: for (k=0; k<maxstates; k++)
1535: ws[k].cost=INF_COST;
1536: }
1537:
1538: static struct tpa_state *empty_tpa_state()
1539: {
1540: struct tpa_state *s = malloc(sizeof(struct tpa_state));
1541:
1542: s->inst = calloc(maxstates,sizeof(struct waypoint));
1543: init_waypoints(s->inst);
1544: s->trans = calloc(maxstates,sizeof(struct waypoint));
1545: /* init_waypoints(s->trans);*/
1546: return s;
1547: }
1548:
1549: static void transitions(struct tpa_state *t)
1550: {
1551: int k;
1552: struct super_state *l;
1553:
1554: for (k=0; k<maxstates; k++) {
1555: t->trans[k] = t->inst[k];
1556: t->trans[k].no_transition = 1;
1557: }
1558: for (l = state_transitions; l != NULL; l = l->next) {
1559: PrimNum s = l->super;
1560: int jcost;
1561: struct cost *c=super_costs+s;
1562: struct waypoint *wi=&(t->trans[c->state_in]);
1563: struct waypoint *wo=&(t->inst[c->state_out]);
1564: lb_applicable_chain_rules++;
1565: if (wo->cost == INF_COST)
1566: continue;
1567: jcost = wo->cost + ss_cost(s);
1568: if (jcost <= wi->cost) {
1569: wi->cost = jcost;
1570: wi->inst = s;
1571: wi->relocatable = wo->relocatable;
1572: wi->no_transition = 0;
1573: /* if (ss_greedy) wi->cost = wo->cost ? */
1574: }
1575: }
1576: }
1577:
1578: static struct tpa_state *make_termstate()
1579: {
1580: struct tpa_state *s = empty_tpa_state();
1581:
1582: s->inst[CANONICAL_STATE].cost = 0;
1583: transitions(s);
1584: return s;
1585: }
1586: #endif
1587:
1588: #define TPA_SIZE 16384
1589:
1590: struct tpa_entry {
1591: struct tpa_entry *next;
1592: PrimNum inst;
1593: struct tpa_state *state_behind; /* note: brack-to-front labeling */
1594: struct tpa_state *state_infront; /* note: brack-to-front labeling */
1595: } *tpa_table[TPA_SIZE];
1596:
1597: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1598: static Cell hash_tpa(PrimNum p, struct tpa_state *t)
1599: {
1600: UCell it = (UCell )t;
1601: return (p+it+(it>>14))&(TPA_SIZE-1);
1602: }
1603:
1604: static struct tpa_state **lookup_tpa(PrimNum p, struct tpa_state *t2)
1605: {
1606: int hash=hash_tpa(p, t2);
1607: struct tpa_entry *te = tpa_table[hash];
1608:
1609: if (tpa_noautomaton) {
1610: static struct tpa_state *t;
1611: t = NULL;
1612: return &t;
1613: }
1614: for (; te!=NULL; te = te->next) {
1615: if (p == te->inst && t2 == te->state_behind)
1616: return &(te->state_infront);
1617: }
1618: te = (struct tpa_entry *)malloc(sizeof(struct tpa_entry));
1619: te->next = tpa_table[hash];
1620: te->inst = p;
1621: te->state_behind = t2;
1622: te->state_infront = NULL;
1623: tpa_table[hash] = te;
1624: return &(te->state_infront);
1625: }
1626:
1627: static void tpa_state_normalize(struct tpa_state *t)
1628: {
1629: /* normalize so cost of canonical state=0; this may result in
1630: negative costs for some states */
1631: int d = t->inst[CANONICAL_STATE].cost;
1632: int i;
1633:
1634: for (i=0; i<maxstates; i++) {
1635: if (t->inst[i].cost != INF_COST)
1636: t->inst[i].cost -= d;
1637: if (t->trans[i].cost != INF_COST)
1638: t->trans[i].cost -= d;
1639: }
1640: }
1641:
1642: static int tpa_state_equivalent(struct tpa_state *t1, struct tpa_state *t2)
1643: {
1644: return (memcmp(t1->inst, t2->inst, maxstates*sizeof(struct waypoint)) == 0 &&
1645: memcmp(t1->trans,t2->trans,maxstates*sizeof(struct waypoint)) == 0);
1646: }
1647: #endif
1648:
1649: struct tpa_state_entry {
1650: struct tpa_state_entry *next;
1651: struct tpa_state *state;
1652: } *tpa_state_table[TPA_SIZE];
1653:
1654: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1655: static Cell hash_tpa_state(struct tpa_state *t)
1656: {
1657: int *ti = (int *)(t->inst);
1658: int *tt = (int *)(t->trans);
1659: int r=0;
1660: int i;
1661:
1662: for (i=0; ti+i < (int *)(t->inst+maxstates); i++)
1663: r += ti[i]+tt[i];
1664: return (r+(r>>14)+(r>>22)) & (TPA_SIZE-1);
1665: }
1666:
1667: static struct tpa_state *lookup_tpa_state(struct tpa_state *t)
1668: {
1669: Cell hash = hash_tpa_state(t);
1670: struct tpa_state_entry *te = tpa_state_table[hash];
1671: struct tpa_state_entry *tn;
1672:
1673: if (!tpa_noequiv) {
1674: for (; te!=NULL; te = te->next) {
1675: if (tpa_state_equivalent(t, te->state)) {
1676: lb_newstate_equiv++;
1677: free(t->inst);
1678: free(t->trans);
1679: free(t);
1680: return te->state;
1681: }
1682: }
1683: tn = (struct tpa_state_entry *)malloc(sizeof(struct tpa_state_entry));
1684: tn->next = te;
1685: tn->state = t;
1686: tpa_state_table[hash] = tn;
1687: }
1688: lb_newstate_new++;
1689: if (tpa_trace)
1690: fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
1691: return t;
1692: }
1693:
1694: /* use dynamic programming to find the shortest paths within the basic
1695: block origs[0..ninsts-1] and rewrite the instructions pointed to by
1696: instps to use it */
1697: static void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
1698: {
1699: int i,j;
1700: struct tpa_state *ts[ninsts+1];
1701: int nextdyn, nextstate, no_transition;
1702: Address old_code_area;
1703:
1704: lb_basic_blocks++;
1705: ts[ninsts] = termstate;
1706: #ifndef NO_DYNAMIC
1707: if (print_sequences) {
1708: for (i=0; i<ninsts; i++)
1709: #if defined(BURG_FORMAT)
1710: fprintf(stderr, "op%d ", super_costs[origs[i]].offset);
1711: #else
1712: fprintf(stderr, "%s ", prim_names[origs[i]]);
1713: #endif
1714: fprintf(stderr, "\n");
1715: }
1716: #endif
1717: for (i=ninsts-1; i>=0; i--) {
1718: struct tpa_state **tp = lookup_tpa(origs[i],ts[i+1]);
1719: struct tpa_state *t = *tp;
1720: lb_labeler_steps++;
1721: if (t) {
1722: ts[i] = t;
1723: lb_labeler_automaton++;
1724: }
1725: else {
1726: lb_labeler_dynprog++;
1727: ts[i] = empty_tpa_state();
1728: for (j=1; j<=max_super && i+j<=ninsts; j++) {
1729: struct super_state **superp = lookup_super(origs+i, j);
1730: if (superp!=NULL) {
1731: struct super_state *supers = *superp;
1732: for (; supers!=NULL; supers = supers->next) {
1733: PrimNum s = supers->super;
1734: int jcost;
1735: struct cost *c=super_costs+s;
1736: struct waypoint *wi=&(ts[i]->inst[c->state_in]);
1737: struct waypoint *wo=&(ts[i+j]->trans[c->state_out]);
1738: int no_transition = wo->no_transition;
1739: lb_applicable_base_rules++;
1740: if (!(is_relocatable(s)) && !wo->relocatable) {
1741: wo=&(ts[i+j]->inst[c->state_out]);
1742: no_transition=1;
1743: }
1744: if (wo->cost == INF_COST)
1745: continue;
1746: jcost = wo->cost + ss_cost(s);
1747: if (jcost <= wi->cost) {
1748: wi->cost = jcost;
1749: wi->inst = s;
1750: wi->relocatable = is_relocatable(s);
1751: wi->no_transition = no_transition;
1752: /* if (ss_greedy) wi->cost = wo->cost ? */
1753: }
1754: }
1755: }
1756: }
1757: transitions(ts[i]);
1758: tpa_state_normalize(ts[i]);
1759: *tp = ts[i] = lookup_tpa_state(ts[i]);
1760: if (tpa_trace)
1761: fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
1762: }
1763: }
1764: /* now rewrite the instructions */
1765: reserve_code_super(origs,ninsts);
1766: old_code_area = code_area;
1767: nextdyn=0;
1768: nextstate=CANONICAL_STATE;
1769: no_transition = ((!ts[0]->trans[nextstate].relocatable)
1770: ||ts[0]->trans[nextstate].no_transition);
1771: for (i=0; i<ninsts; i++) {
1772: Cell tc=0, tc2;
1773: if (i==nextdyn) {
1774: if (!no_transition) {
1775: /* process trans */
1776: PrimNum p = ts[i]->trans[nextstate].inst;
1777: struct cost *c = super_costs+p;
1778: assert(ts[i]->trans[nextstate].cost != INF_COST);
1779: assert(c->state_in==nextstate);
1780: tc = compile_prim_dyn(p,NULL);
1781: nextstate = c->state_out;
1782: }
1783: {
1784: /* process inst */
1785: PrimNum p = ts[i]->inst[nextstate].inst;
1786: struct cost *c=super_costs+p;
1787: assert(c->state_in==nextstate);
1788: assert(ts[i]->inst[nextstate].cost != INF_COST);
1789: #if defined(GFORTH_DEBUGGING)
1790: assert(p == origs[i]);
1791: #endif
1792: tc2 = compile_prim_dyn(p,instps[i]);
1793: if (no_transition || !is_relocatable(p))
1794: /* !! actually what we care about is if and where
1795: * compile_prim_dyn() puts NEXTs */
1796: tc=tc2;
1797: no_transition = ts[i]->inst[nextstate].no_transition;
1798: nextstate = c->state_out;
1799: nextdyn += c->length;
1800: }
1801: } else {
1802: #if defined(GFORTH_DEBUGGING)
1803: assert(0);
1804: #endif
1805: tc=0;
1806: /* tc= (Cell)vm_prims[ts[i]->inst[CANONICAL_STATE].inst]; */
1807: }
1808: *(instps[i]) = tc;
1809: }
1810: if (!no_transition) {
1811: PrimNum p = ts[i]->trans[nextstate].inst;
1812: struct cost *c = super_costs+p;
1813: assert(c->state_in==nextstate);
1814: assert(ts[i]->trans[nextstate].cost != INF_COST);
1815: assert(i==nextdyn);
1816: (void)compile_prim_dyn(p,NULL);
1817: nextstate = c->state_out;
1818: }
1819: assert(nextstate==CANONICAL_STATE);
1820: assert(code_area==old_code_area); /* does reserve_code_super() work? */
1821: }
1822: #endif
1823:
1824: /* compile *start, possibly rewriting it into a static and/or dynamic
1825: superinstruction */
1826: void compile_prim1(Cell *start)
1827: {
1828: #if defined(DOUBLY_INDIRECT)
1829: Label prim;
1830:
1831: if (start==NULL)
1832: return;
1833: prim = (Label)*start;
1834: if (prim<((Label)(xts+DOER_MAX)) || prim>((Label)(xts+npriminfos))) {
1835: fprintf(stderr,"compile_prim encountered xt %p\n", prim);
1836: *start=(Cell)prim;
1837: return;
1838: } else {
1839: *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
1840: return;
1841: }
1842: #elif defined(INDIRECT_THREADED)
1843: return;
1844: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1845: static Cell *instps[MAX_BB];
1846: static PrimNum origs[MAX_BB];
1847: static int ninsts=0;
1848: PrimNum prim_num;
1849:
1850: if (start==NULL || ninsts >= MAX_BB ||
1851: (ninsts>0 && superend[origs[ninsts-1]])) {
1852: /* after bb, or at the start of the next bb */
1853: optimize_rewrite(instps,origs,ninsts);
1854: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
1855: ninsts=0;
1856: if (start==NULL) {
1857: align_code();
1858: return;
1859: }
1860: }
1861: prim_num = ((Xt)*start)-vm_prims;
1862: if(prim_num >= npriminfos) {
1863: /* code word */
1864: optimize_rewrite(instps,origs,ninsts);
1865: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
1866: ninsts=0;
1867: append_jump();
1868: *start = *(Cell *)*start;
1869: return;
1870: }
1871: assert(ninsts<MAX_BB);
1872: instps[ninsts] = start;
1873: origs[ninsts] = prim_num;
1874: ninsts++;
1875: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1876: }
1877:
1878: #ifndef STANDALONE
1879: Address gforth_loader(FILE *imagefile, char* filename)
1880: /* returns the address of the image proper (after the preamble) */
1881: {
1882: ImageHeader header;
1883: Address image;
1884: Address imp; /* image+preamble */
1885: Char magic[8];
1886: char magic7; /* size byte of magic number */
1887: Cell preamblesize=0;
1888: Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1889: UCell check_sum;
1890: Cell ausize = ((RELINFOBITS == 8) ? 0 :
1891: (RELINFOBITS == 16) ? 1 :
1892: (RELINFOBITS == 32) ? 2 : 3);
1893: Cell charsize = ((sizeof(Char) == 1) ? 0 :
1894: (sizeof(Char) == 2) ? 1 :
1895: (sizeof(Char) == 4) ? 2 : 3) + ausize;
1896: Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
1897: (sizeof(Cell) == 2) ? 1 :
1898: (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1899: Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
1900: #ifdef WORDS_BIGENDIAN
1901: 0
1902: #else
1903: 1
1904: #endif
1905: ;
1906:
1907: vm_prims = gforth_engine(0,0,0,0,0 sr_call);
1908: check_prims(vm_prims);
1909: prepare_super_table();
1910: #ifndef DOUBLY_INDIRECT
1911: #ifdef PRINT_SUPER_LENGTHS
1912: print_super_lengths();
1913: #endif
1914: check_sum = checksum(vm_prims);
1915: #else /* defined(DOUBLY_INDIRECT) */
1916: check_sum = (UCell)vm_prims;
1917: #endif /* defined(DOUBLY_INDIRECT) */
1918: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1919: termstate = make_termstate();
1920: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1921:
1922: do {
1923: if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1924: fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.8) image.\n",
1925: progname, filename);
1926: exit(1);
1927: }
1928: preamblesize+=8;
1929: } while(memcmp(magic,"Gforth4",7));
1930: magic7 = magic[7];
1931: if (debug) {
1932: magic[7]='\0';
1933: fprintf(stderr,"Magic found: %s ", magic);
1934: print_sizes(magic7);
1935: }
1936:
1937: if (magic7 != sizebyte)
1938: {
1939: fprintf(stderr,"This image is: ");
1940: print_sizes(magic7);
1941: fprintf(stderr,"whereas the machine is ");
1942: print_sizes(sizebyte);
1943: exit(-2);
1944: };
1945:
1946: fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1947:
1948: set_stack_sizes(&header);
1949:
1950: #if HAVE_GETPAGESIZE
1951: pagesize=getpagesize(); /* Linux/GNU libc offers this */
1952: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
1953: pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
1954: #elif PAGESIZE
1955: pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
1956: #endif
1957: debugp(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1958:
1959: image = dict_alloc_read(imagefile, preamblesize+header.image_size,
1960: dictsize, data_offset);
1961: imp=image+preamblesize;
1962:
1963: alloc_stacks((ImageHeader *)imp);
1964: if (clear_dictionary)
1965: memset(imp+header.image_size, 0, dictsize-header.image_size-preamblesize);
1966: if(header.base==0 || header.base == (Address)0x100) {
1967: Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
1968: Char reloc_bits[reloc_size];
1969: fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1970: fread(reloc_bits, 1, reloc_size, imagefile);
1971: gforth_relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1972: #if 0
1973: { /* let's see what the relocator did */
1974: FILE *snapshot=fopen("snapshot.fi","wb");
1975: fwrite(image,1,imagesize,snapshot);
1976: fclose(snapshot);
1977: }
1978: #endif
1979: }
1980: else if(header.base!=imp) {
1981: fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
1982: progname, (unsigned long)header.base, (unsigned long)imp);
1983: exit(1);
1984: }
1985: if (header.checksum==0)
1986: ((ImageHeader *)imp)->checksum=check_sum;
1987: else if (header.checksum != check_sum) {
1988: fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
1989: progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
1990: exit(1);
1991: }
1992: #ifdef DOUBLY_INDIRECT
1993: ((ImageHeader *)imp)->xt_base = xts;
1994: #endif
1995: fclose(imagefile);
1996:
1997: /* unnecessary, except maybe for CODE words */
1998: /* FLUSH_ICACHE(imp, header.image_size);*/
1999:
2000: return imp;
2001: }
2002: #endif
2003:
2004: /* pointer to last '/' or '\' in file, 0 if there is none. */
2005: static char *onlypath(char *filename)
2006: {
2007: return strrchr(filename, DIRSEP);
2008: }
2009:
2010: static FILE *openimage(char *fullfilename)
2011: {
2012: FILE *image_file;
2013: char * expfilename = tilde_cstr((Char *)fullfilename, strlen(fullfilename));
2014:
2015: image_file=fopen(expfilename,"rb");
2016: if (image_file!=NULL && debug)
2017: fprintf(stderr, "Opened image file: %s\n", expfilename);
2018: free(expfilename);
2019: return image_file;
2020: }
2021:
2022: /* try to open image file concat(path[0:len],imagename) */
2023: static FILE *checkimage(char *path, int len, char *imagename)
2024: {
2025: int dirlen=len;
2026: char fullfilename[dirlen+strlen((char *)imagename)+2];
2027:
2028: memcpy(fullfilename, path, dirlen);
2029: if (fullfilename[dirlen-1]!=DIRSEP)
2030: fullfilename[dirlen++]=DIRSEP;
2031: strcpy(fullfilename+dirlen,imagename);
2032: return openimage(fullfilename);
2033: }
2034:
2035: static FILE * open_image_file(char * imagename, char * path)
2036: {
2037: FILE * image_file=NULL;
2038: char *origpath=path;
2039:
2040: if(strchr(imagename, DIRSEP)==NULL) {
2041: /* first check the directory where the exe file is in !! 01may97jaw */
2042: if (onlypath(progname))
2043: image_file=checkimage(progname, onlypath(progname)-progname, imagename);
2044: if (!image_file)
2045: do {
2046: char *pend=strchr(path, PATHSEP);
2047: if (pend==NULL)
2048: pend=path+strlen(path);
2049: if (strlen(path)==0) break;
2050: image_file=checkimage(path, pend-path, imagename);
2051: path=pend+(*pend==PATHSEP);
2052: } while (image_file==NULL);
2053: } else {
2054: image_file=openimage(imagename);
2055: }
2056:
2057: if (!image_file) {
2058: fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
2059: progname, imagename, origpath);
2060: exit(1);
2061: }
2062:
2063: return image_file;
2064: }
2065: #endif
2066:
2067: #ifdef STANDALONE_ALLOC
2068: Address gforth_alloc(Cell size)
2069: {
2070: Address r;
2071: /* leave a little room (64B) for stack underflows */
2072: if ((r = malloc(size+64))==NULL) {
2073: perror(progname);
2074: exit(1);
2075: }
2076: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
2077: debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
2078: return r;
2079: }
2080: #endif
2081:
2082: #ifdef HAS_OS
2083: static UCell convsize(char *s, UCell elemsize)
2084: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
2085: of bytes. the letter at the end indicates the unit, where e stands
2086: for the element size. default is e */
2087: {
2088: char *endp;
2089: UCell n,m;
2090:
2091: m = elemsize;
2092: n = strtoul(s,&endp,0);
2093: if (endp!=NULL) {
2094: if (strcmp(endp,"b")==0)
2095: m=1;
2096: else if (strcmp(endp,"k")==0)
2097: m=1024;
2098: else if (strcmp(endp,"M")==0)
2099: m=1024*1024;
2100: else if (strcmp(endp,"G")==0)
2101: m=1024*1024*1024;
2102: else if (strcmp(endp,"T")==0) {
2103: #if (SIZEOF_CHAR_P > 4)
2104: m=1024L*1024*1024*1024;
2105: #else
2106: fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
2107: exit(1);
2108: #endif
2109: } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
2110: fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
2111: exit(1);
2112: }
2113: }
2114: return n*m;
2115: }
2116:
2117: enum {
2118: ss_number = 256,
2119: ss_states,
2120: ss_min_codesize,
2121: ss_min_ls,
2122: ss_min_lsu,
2123: ss_min_nexts,
2124: opt_code_block_size,
2125: };
2126:
2127: #ifndef STANDALONE
2128: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
2129: {
2130: int c;
2131:
2132: opterr=0;
2133: while (1) {
2134: int option_index=0;
2135: static struct option opts[] = {
2136: {"appl-image", required_argument, NULL, 'a'},
2137: {"image-file", required_argument, NULL, 'i'},
2138: {"dictionary-size", required_argument, NULL, 'm'},
2139: {"data-stack-size", required_argument, NULL, 'd'},
2140: {"return-stack-size", required_argument, NULL, 'r'},
2141: {"fp-stack-size", required_argument, NULL, 'f'},
2142: {"locals-stack-size", required_argument, NULL, 'l'},
2143: {"vm-commit", no_argument, &map_noreserve, 0},
2144: {"path", required_argument, NULL, 'p'},
2145: {"version", no_argument, NULL, 'v'},
2146: {"help", no_argument, NULL, 'h'},
2147: /* put something != 0 into offset_image */
2148: {"offset-image", no_argument, &offset_image, 1},
2149: {"no-offset-im", no_argument, &offset_image, 0},
2150: {"clear-dictionary", no_argument, &clear_dictionary, 1},
2151: {"debug", no_argument, &debug, 1},
2152: {"diag", no_argument, &diag, 1},
2153: {"die-on-signal", no_argument, &die_on_signal, 1},
2154: {"ignore-async-signals", no_argument, &ignore_async_signals, 1},
2155: {"no-super", no_argument, &no_super, 1},
2156: {"no-dynamic", no_argument, &no_dynamic, 1},
2157: {"dynamic", no_argument, &no_dynamic, 0},
2158: {"code-block-size", required_argument, NULL, opt_code_block_size},
2159: {"print-metrics", no_argument, &print_metrics, 1},
2160: {"print-sequences", no_argument, &print_sequences, 1},
2161: {"ss-number", required_argument, NULL, ss_number},
2162: {"ss-states", required_argument, NULL, ss_states},
2163: #ifndef NO_DYNAMIC
2164: {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
2165: #endif
2166: {"ss-min-ls", no_argument, NULL, ss_min_ls},
2167: {"ss-min-lsu", no_argument, NULL, ss_min_lsu},
2168: {"ss-min-nexts", no_argument, NULL, ss_min_nexts},
2169: {"ss-greedy", no_argument, &ss_greedy, 1},
2170: {"tpa-noequiv", no_argument, &tpa_noequiv, 1},
2171: {"tpa-noautomaton", no_argument, &tpa_noautomaton, 1},
2172: {"tpa-trace", no_argument, &tpa_trace, 1},
2173: {0,0,0,0}
2174: /* no-init-file, no-rc? */
2175: };
2176:
2177: c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
2178:
2179: switch (c) {
2180: case EOF: return;
2181: case '?': optind--; return;
2182: case 'a': *imagename = optarg; return;
2183: case 'i': *imagename = optarg; break;
2184: case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
2185: case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
2186: case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
2187: case 'f': fsize = convsize(optarg,sizeof(Float)); break;
2188: case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
2189: case 'p': *path = optarg; break;
2190: case 'o': offset_image = 1; break;
2191: case 'n': offset_image = 0; break;
2192: case 'c': clear_dictionary = 1; break;
2193: case 's': die_on_signal = 1; break;
2194: case 'x': debug = 1; break;
2195: case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
2196: case opt_code_block_size: code_area_size = atoi(optarg); break;
2197: case ss_number: static_super_number = atoi(optarg); break;
2198: case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
2199: #ifndef NO_DYNAMIC
2200: case ss_min_codesize: ss_cost = cost_codesize; break;
2201: #endif
2202: case ss_min_ls: ss_cost = cost_ls; break;
2203: case ss_min_lsu: ss_cost = cost_lsu; break;
2204: case ss_min_nexts: ss_cost = cost_nexts; break;
2205: case 'h':
2206: fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
2207: Engine Options:\n\
2208: --appl-image FILE Equivalent to '--image-file=FILE --'\n\
2209: --clear-dictionary Initialize the dictionary with 0 bytes\n\
2210: --code-block-size=SIZE size of native code blocks [512KB]\n\
2211: -d SIZE, --data-stack-size=SIZE Specify data stack size\n\
2212: --debug Print debugging information during startup\n\
2213: --diag Print diagnostic information during startup\n\
2214: --die-on-signal Exit instead of THROWing some signals\n\
2215: --dynamic Use dynamic native code\n\
2216: -f SIZE, --fp-stack-size=SIZE Specify floating point stack size\n\
2217: -h, --help Print this message and exit\n\
2218: --ignore-async-signals Ignore instead of THROWing async. signals\n\
2219: -i FILE, --image-file=FILE Use image FILE instead of `gforth.fi'\n\
2220: -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
2221: -m SIZE, --dictionary-size=SIZE Specify Forth dictionary size\n\
2222: --no-dynamic Use only statically compiled primitives\n\
2223: --no-offset-im Load image at normal position\n\
2224: --no-super No dynamically formed superinstructions\n\
2225: --offset-image Load image at a different position\n\
2226: -p PATH, --path=PATH Search path for finding image and sources\n\
2227: --print-metrics Print some code generation metrics on exit\n\
2228: --print-sequences Print primitive sequences for optimization\n\
2229: -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
2230: --ss-greedy Greedy, not optimal superinst selection\n\
2231: --ss-min-codesize Select superinsts for smallest native code\n\
2232: --ss-min-ls Minimize loads and stores\n\
2233: --ss-min-lsu Minimize loads, stores, and pointer updates\n\
2234: --ss-min-nexts Minimize the number of static superinsts\n\
2235: --ss-number=N Use N static superinsts (default max)\n\
2236: --ss-states=N N states for stack caching (default max)\n\
2237: --tpa-noequiv Automaton without state equivalence\n\
2238: --tpa-noautomaton Dynamic programming only\n\
2239: --tpa-trace Report new states etc.\n\
2240: -v, --version Print engine version and exit\n\
2241: --vm-commit Use OS default for memory overcommit\n\
2242: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
2243: `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
2244: argv[0]);
2245: optind--;
2246: return;
2247: }
2248: }
2249: }
2250: #endif
2251: #endif
2252:
2253: static void print_diag()
2254: {
2255:
2256: #if !defined(HAVE_GETRUSAGE)
2257: fprintf(stderr, "*** missing functionality ***\n"
2258: #ifndef HAVE_GETRUSAGE
2259: " no getrusage -> CPUTIME broken\n"
2260: #endif
2261: );
2262: #endif
2263: if((relocs < nonrelocs) ||
2264: #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)
2265: 1
2266: #else
2267: 0
2268: #endif
2269: )
2270: debugp(stderr, "relocs: %d:%d\n", relocs, nonrelocs);
2271: fprintf(stderr, "*** %sperformance problems ***\n%s%s",
2272: #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)
2273: "",
2274: #else
2275: "no ",
2276: #endif
2277: #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)
2278: " double-cell integer type buggy ->\n "
2279: #ifdef BUGGY_LL_CMP
2280: "double comparisons, "
2281: #endif
2282: #ifdef BUGGY_LL_MUL
2283: "*/MOD */ M* UM* "
2284: #endif
2285: #ifdef BUGGY_LL_DIV
2286: /* currently nothing is affected */
2287: #endif
2288: #ifdef BUGGY_LL_ADD
2289: "M+ D+ D- DNEGATE "
2290: #endif
2291: #ifdef BUGGY_LL_SHIFT
2292: "D2/ "
2293: #endif
2294: #ifdef BUGGY_LL_D2F
2295: "D>F "
2296: #endif
2297: #ifdef BUGGY_LL_F2D
2298: "F>D "
2299: #endif
2300: "\b\b slow\n"
2301: #endif
2302: #if !(defined(FORCE_REG) || defined(FORCE_REG_UNNECESSARY))
2303: " automatic register allocation: performance degradation possible\n"
2304: #endif
2305: "",
2306: (relocs < nonrelocs) ? "no dynamic code generation (--debug for details) -> factor 2 slowdown\n" : "");
2307: }
2308:
2309: #ifdef STANDALONE
2310: Cell data_abort_pc;
2311:
2312: void data_abort_C(void)
2313: {
2314: while(1) {
2315: }
2316: }
2317: #endif
2318:
2319: void gforth_linkback()
2320: {
2321: #ifndef HAS_LINKBACK
2322: gforth_pointers[0] = (void*)&gforth_SP;
2323: gforth_pointers[1] = (void*)&gforth_FP;
2324: gforth_pointers[2] = (void*)&gforth_LP;
2325: gforth_pointers[3] = (void*)&gforth_RP;
2326: gforth_pointers[4] = (void*)&gforth_UP;
2327: gforth_pointers[5] = (void*)gforth_engine;
2328: #ifdef HAS_FILE
2329: gforth_pointers[6] = (void*)cstr;
2330: gforth_pointers[7] = (void*)tilde_cstr;
2331: #endif
2332: #endif
2333: }
2334:
2335: int main(int argc, char **argv, char **env)
2336: {
2337: #ifdef HAS_OS
2338: char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
2339: #else
2340: char *path = DEFAULTPATH;
2341: #endif
2342: #ifndef INCLUDE_IMAGE
2343: char *imagename="gforth.fi";
2344: FILE *image_file;
2345: Address image;
2346: #endif
2347: int retvalue;
2348: #if 0 && defined(__i386)
2349: /* disabled because the drawbacks may be worse than the benefits */
2350: /* set 387 precision control to use 53-bit mantissae to avoid most
2351: cases of double rounding */
2352: short fpu_control = 0x027f ;
2353: asm("fldcw %0" : : "m"(fpu_control));
2354: #endif /* defined(__i386) */
2355:
2356: #ifndef HAS_LINKBACK
2357: gforth_linkback();
2358: #endif
2359:
2360: #ifdef MACOSX_DEPLOYMENT_TARGET
2361: setenv("MACOSX_DEPLOYMENT_TARGET", MACOSX_DEPLOYMENT_TARGET, 0);
2362: #endif
2363: #ifdef LTDL_LIBRARY_PATH
2364: setenv("LTDL_LIBRARY_PATH", LTDL_LIBRARY_PATH, 0);
2365: #endif
2366: #ifndef STANDALONE
2367: /* buffering of the user output device */
2368: #ifdef _IONBF
2369: if (isatty(fileno(stdout))) {
2370: fflush(stdout);
2371: setvbuf(stdout,NULL,_IONBF,0);
2372: }
2373: #endif
2374: setlocale(LC_ALL, "");
2375: setlocale(LC_NUMERIC, "C");
2376: #else
2377: prep_terminal();
2378: #endif
2379:
2380: progname = argv[0];
2381:
2382: #ifndef STANDALONE
2383: #ifdef HAVE_LIBLTDL
2384: if (lt_dlinit()!=0) {
2385: fprintf(stderr,"%s: lt_dlinit failed", progname);
2386: exit(1);
2387: }
2388: #endif
2389:
2390: #ifdef HAS_OS
2391: gforth_args(argc, argv, &path, &imagename);
2392: #ifndef NO_DYNAMIC
2393: init_ss_cost();
2394: #endif /* !defined(NO_DYNAMIC) */
2395: #endif /* defined(HAS_OS) */
2396: #endif
2397: code_here = ((void *)0)+code_area_size;
2398: #ifdef STANDALONE
2399: image = gforth_engine(0, 0, 0, 0, 0 sr_call);
2400: alloc_stacks((ImageHeader *)image);
2401: #else
2402: image_file = open_image_file(imagename, path);
2403: image = gforth_loader(image_file, imagename);
2404: #endif
2405: gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
2406:
2407: if (diag)
2408: print_diag();
2409: {
2410: char path2[strlen(path)+1];
2411: char *p1, *p2;
2412: Cell environ[]= {
2413: (Cell)argc-(optind-1),
2414: (Cell)(argv+(optind-1)),
2415: (Cell)strlen(path),
2416: (Cell)path2};
2417: argv[optind-1] = progname;
2418: /*
2419: for (i=0; i<environ[0]; i++)
2420: printf("%s\n", ((char **)(environ[1]))[i]);
2421: */
2422: /* make path OS-independent by replacing path separators with NUL */
2423: for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
2424: if (*p1==PATHSEP)
2425: *p2 = '\0';
2426: else
2427: *p2 = *p1;
2428: *p2='\0';
2429: retvalue = gforth_go(image, 4, environ);
2430: #if defined(SIGPIPE) && !defined(STANDALONE)
2431: bsd_signal(SIGPIPE, SIG_IGN);
2432: #endif
2433: #ifdef VM_PROFILING
2434: vm_print_profile(stderr);
2435: #endif
2436: deprep_terminal();
2437: #ifndef STANDALONE
2438: #ifdef HAVE_LIBLTDL
2439: if (lt_dlexit()!=0)
2440: fprintf(stderr,"%s: lt_dlexit failed", progname);
2441: #endif
2442: #endif
2443: }
2444: if (print_metrics) {
2445: int i;
2446: fprintf(stderr, "code size = %8ld\n", dyncodesize());
2447: #ifndef STANDALONE
2448: for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
2449: fprintf(stderr, "metric %8s: %8ld\n",
2450: cost_sums[i].metricname, cost_sums[i].sum);
2451: #endif
2452: fprintf(stderr,"lb_basic_blocks = %ld\n", lb_basic_blocks);
2453: fprintf(stderr,"lb_labeler_steps = %ld\n", lb_labeler_steps);
2454: fprintf(stderr,"lb_labeler_automaton = %ld\n", lb_labeler_automaton);
2455: fprintf(stderr,"lb_labeler_dynprog = %ld\n", lb_labeler_dynprog);
2456: fprintf(stderr,"lb_newstate_equiv = %ld\n", lb_newstate_equiv);
2457: fprintf(stderr,"lb_newstate_new = %ld\n", lb_newstate_new);
2458: fprintf(stderr,"lb_applicable_base_rules = %ld\n", lb_applicable_base_rules);
2459: fprintf(stderr,"lb_applicable_chain_rules = %ld\n", lb_applicable_chain_rules);
2460: }
2461: if (tpa_trace) {
2462: fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
2463: fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
2464: }
2465: return retvalue;
2466: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>