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