1: /* command line interpretation, image loading etc. for Gforth
2:
3:
4: Copyright (C) 1995,1996,1997,1998,2000,2003,2004,2005,2006 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 2
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, write to the Free Software
20: Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
21: */
22:
23: #include "config.h"
24: #include "forth.h"
25: #include <errno.h>
26: #include <ctype.h>
27: #include <stdio.h>
28: #include <unistd.h>
29: #include <string.h>
30: #include <math.h>
31: #include <sys/types.h>
32: #ifndef STANDALONE
33: #include <sys/stat.h>
34: #endif
35: #include <fcntl.h>
36: #include <assert.h>
37: #include <stdlib.h>
38: #include <signal.h>
39: #ifndef STANDALONE
40: #if HAVE_SYS_MMAN_H
41: #include <sys/mman.h>
42: #endif
43: #endif
44: #include "io.h"
45: #include "getopt.h"
46: #ifdef STANDALONE
47: /* #include <systypes.h> */
48: #endif
49:
50: typedef enum prim_num {
51: /* definitions of N_execute etc. */
52: #include PRIM_NUM_I
53: N_START_SUPER
54: } PrimNum;
55:
56: /* global variables for engine.c
57: We put them here because engine.c is compiled several times in
58: different ways for the same engine. */
59: Cell *gforth_SP;
60: Float *gforth_FP;
61: Address gforth_UP=NULL;
62:
63: #ifdef HAS_FFCALL
64: Cell *gforth_RP;
65: Address gforth_LP;
66:
67: #include <callback.h>
68:
69: va_alist gforth_clist;
70:
71: void gforth_callback(Xt* fcall, void * alist)
72: {
73: /* save global valiables */
74: Cell *rp = gforth_RP;
75: Cell *sp = gforth_SP;
76: Float *fp = gforth_FP;
77: Address lp = gforth_LP;
78: va_alist clist = gforth_clist;
79:
80: gforth_clist = (va_alist)alist;
81:
82: gforth_engine(fcall, sp, rp, fp, lp);
83:
84: /* restore global variables */
85: gforth_RP = rp;
86: gforth_SP = sp;
87: gforth_FP = fp;
88: gforth_LP = lp;
89: gforth_clist = clist;
90: }
91: #endif
92:
93: #ifdef HAS_LIBFFI
94: Cell *gforth_RP;
95: Address gforth_LP;
96:
97: #include <ffi.h>
98:
99: void ** gforth_clist;
100: void * gforth_ritem;
101:
102: void gforth_callback(ffi_cif * cif, void * resp, void ** args, void * ip)
103: {
104: Cell *rp = gforth_RP;
105: Cell *sp = gforth_SP;
106: Float *fp = gforth_FP;
107: Address lp = gforth_LP;
108: void ** clist = gforth_clist;
109: void * ritem = gforth_ritem;
110:
111: gforth_clist = args;
112: gforth_ritem = resp;
113:
114: gforth_engine((Xt *)ip, sp, rp, fp, lp);
115:
116: /* restore global variables */
117: gforth_RP = rp;
118: gforth_SP = sp;
119: gforth_FP = fp;
120: gforth_LP = lp;
121: gforth_clist = clist;
122: gforth_ritem = ritem;
123: }
124: #endif
125:
126: #ifdef GFORTH_DEBUGGING
127: /* define some VM registers as global variables, so they survive exceptions;
128: global register variables are not up to the task (according to the
129: GNU C manual) */
130: Xt *saved_ip;
131: Cell *rp;
132: #endif
133:
134: #ifdef NO_IP
135: Label next_code;
136: #endif
137:
138: #ifdef HAS_FILE
139: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
140: char* pfileattr[6]={"r","r","r+","r+","w","w"};
141:
142: #ifndef O_BINARY
143: #define O_BINARY 0
144: #endif
145: #ifndef O_TEXT
146: #define O_TEXT 0
147: #endif
148:
149: int ufileattr[6]= {
150: O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
151: O_RDWR |O_BINARY, O_RDWR |O_BINARY,
152: O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
153: #endif
154: /* end global vars for engine.c */
155:
156: #define PRIM_VERSION 1
157: /* increment this whenever the primitives change in an incompatible way */
158:
159: #ifndef DEFAULTPATH
160: # define DEFAULTPATH "."
161: #endif
162:
163: #ifdef MSDOS
164: jmp_buf throw_jmp_buf;
165: #endif
166:
167: #if defined(DOUBLY_INDIRECT)
168: # define CFA(n) ({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
169: #else
170: # define CFA(n) ((Cell)(symbols+((n)&~0x4000UL)))
171: #endif
172:
173: #define maxaligned(n) (typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
174:
175: static UCell dictsize=0;
176: static UCell dsize=0;
177: static UCell rsize=0;
178: static UCell fsize=0;
179: static UCell lsize=0;
180: int offset_image=0;
181: int die_on_signal=0;
182: int ignore_async_signals=0;
183: #ifndef INCLUDE_IMAGE
184: static int clear_dictionary=0;
185: UCell pagesize=1;
186: char *progname;
187: #else
188: char *progname = "gforth";
189: int optind = 1;
190: #endif
191: #ifndef MAP_NORESERVE
192: #define MAP_NORESERVE 0
193: #endif
194: /* IF you have an old Cygwin, this may help:
195: #ifdef __CYGWIN__
196: #define MAP_NORESERVE 0
197: #endif
198: */
199: static int map_noreserve=MAP_NORESERVE;
200:
201: #define CODE_BLOCK_SIZE (512*1024) /* !! overflow handling for -native */
202: Address code_area=0;
203: Cell code_area_size = CODE_BLOCK_SIZE;
204: Address code_here=NULL+CODE_BLOCK_SIZE; /* does for code-area what HERE
205: does for the dictionary */
206: Address start_flush=NULL; /* start of unflushed code */
207: Cell last_jump=0; /* if the last prim was compiled without jump, this
208: is it's number, otherwise this contains 0 */
209:
210: static int no_super=0; /* true if compile_prim should not fuse prims */
211: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
212: dynamically */
213: static int print_metrics=0; /* if true, print metrics on exit */
214: static int static_super_number = 0; /* number of ss used if available */
215: /* disabled because of tpa */
216: #define MAX_STATE 9 /* maximum number of states */
217: static int maxstates = MAX_STATE; /* number of states for stack caching */
218: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
219: static int diag = 0; /* if true: print diagnostic informations */
220: static int tpa_noequiv = 0; /* if true: no state equivalence checking */
221: static int tpa_noautomaton = 0; /* if true: no tree parsing automaton */
222: static int tpa_trace = 0; /* if true: data for line graph of new states etc. */
223: static int print_sequences = 0; /* print primitive sequences for optimization */
224: static int relocs = 0;
225: static int nonrelocs = 0;
226:
227: #ifdef HAS_DEBUG
228: int debug=0;
229: # define debugp(x...) if (debug) fprintf(x);
230: #else
231: # define perror(x...)
232: # define fprintf(x...)
233: # define debugp(x...)
234: #endif
235:
236: ImageHeader *gforth_header;
237: Label *vm_prims;
238: #ifdef DOUBLY_INDIRECT
239: Label *xts; /* same content as vm_prims, but should only be used for xts */
240: #endif
241:
242: #ifndef NO_DYNAMIC
243: #ifndef CODE_ALIGNMENT
244: #define CODE_ALIGNMENT 0
245: #endif
246:
247: #define MAX_IMMARGS 2
248:
249: typedef struct {
250: Label start; /* NULL if not relocatable */
251: Cell length; /* only includes the jump iff superend is true*/
252: Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
253: char superend; /* true if primitive ends superinstruction, i.e.,
254: unconditional branch, execute, etc. */
255: Cell nimmargs;
256: struct immarg {
257: Cell offset; /* offset of immarg within prim */
258: char rel; /* true if immarg is relative */
259: } immargs[MAX_IMMARGS];
260: } PrimInfo;
261:
262: PrimInfo *priminfos;
263: PrimInfo **decomp_prims;
264:
265: const char const* const prim_names[]={
266: #include PRIM_NAMES_I
267: };
268:
269: void init_ss_cost(void);
270:
271: static int is_relocatable(int p)
272: {
273: return !no_dynamic && priminfos[p].start != NULL;
274: }
275: #else /* defined(NO_DYNAMIC) */
276: static int is_relocatable(int p)
277: {
278: return 0;
279: }
280: #endif /* defined(NO_DYNAMIC) */
281:
282: #ifdef MEMCMP_AS_SUBROUTINE
283: int gforth_memcmp(const char * s1, const char * s2, size_t n)
284: {
285: return memcmp(s1, s2, n);
286: }
287: #endif
288:
289: static Cell max(Cell a, Cell b)
290: {
291: return a>b?a:b;
292: }
293:
294: static Cell min(Cell a, Cell b)
295: {
296: return a<b?a:b;
297: }
298:
299: #ifndef STANDALONE
300: /* image file format:
301: * "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
302: * padding to a multiple of 8
303: * magic: "Gforth3x" means format 0.6,
304: * where x is a byte with
305: * bit 7: reserved = 0
306: * bit 6:5: address unit size 2^n octets
307: * bit 4:3: character size 2^n octets
308: * bit 2:1: cell size 2^n octets
309: * bit 0: endian, big=0, little=1.
310: * The magic are always 8 octets, no matter what the native AU/character size is
311: * padding to max alignment (no padding necessary on current machines)
312: * ImageHeader structure (see forth.h)
313: * data (size in ImageHeader.image_size)
314: * tags ((if relocatable, 1 bit/data cell)
315: *
316: * tag==1 means that the corresponding word is an address;
317: * If the word is >=0, the address is within the image;
318: * addresses within the image are given relative to the start of the image.
319: * If the word =-1 (CF_NIL), the address is NIL,
320: * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
321: * If the word =CF(DODOES), it's a DOES> CFA
322: * If the word =CF(DOESJUMP), it's a DOES JUMP (2 Cells after DOES>,
323: * possibly containing a jump to dodoes)
324: * If the word is <CF(DOESJUMP) and bit 14 is set, it's the xt of a primitive
325: * If the word is <CF(DOESJUMP) and bit 14 is clear,
326: * it's the threaded code of a primitive
327: * bits 13..9 of a primitive token state which group the primitive belongs to,
328: * bits 8..0 of a primitive token index into the group
329: */
330:
331: Cell groups[32] = {
332: 0,
333: 0
334: #undef GROUP
335: #undef GROUPADD
336: #define GROUPADD(n) +n
337: #define GROUP(x, n) , 0
338: #include PRIM_GRP_I
339: #undef GROUP
340: #undef GROUPADD
341: #define GROUP(x, n)
342: #define GROUPADD(n)
343: };
344:
345: static unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
346: int size, Cell base)
347: /* produce a bitmask marking all the branch targets */
348: {
349: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
350: Cell token;
351: unsigned char bits;
352: unsigned char *result=malloc(steps);
353:
354: memset(result, 0, steps);
355: for(k=0; k<steps; k++) {
356: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
357: if(bits & (1U << (RELINFOBITS-1))) {
358: assert(i*sizeof(Cell) < size);
359: token=image[i];
360: if (token>=base) { /* relocatable address */
361: UCell bitnum=(token-base)/sizeof(Cell);
362: if (bitnum/RELINFOBITS < (UCell)steps)
363: result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
364: }
365: }
366: }
367: }
368: return result;
369: }
370:
371: void gforth_relocate(Cell *image, const Char *bitstring,
372: UCell size, Cell base, Label symbols[])
373: {
374: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
375: Cell token;
376: char bits;
377: Cell max_symbols;
378: /*
379: * A virtual start address that's the real start address minus
380: * the one in the image
381: */
382: Cell *start = (Cell * ) (((void *) image) - ((void *) base));
383: unsigned char *targets = branch_targets(image, bitstring, size, base);
384:
385: /* group index into table */
386: if(groups[31]==0) {
387: int groupsum=0;
388: for(i=0; i<32; i++) {
389: groupsum += groups[i];
390: groups[i] = groupsum;
391: /* printf("group[%d]=%d\n",i,groupsum); */
392: }
393: i=0;
394: }
395:
396: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
397:
398: for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
399: ;
400: max_symbols--;
401:
402: for(k=0; k<steps; k++) {
403: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
404: /* fprintf(stderr,"relocate: image[%d]\n", i);*/
405: if(bits & (1U << (RELINFOBITS-1))) {
406: assert(i*sizeof(Cell) < size);
407: /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
408: token=image[i];
409: if(token<0) {
410: int group = (-token & 0x3E00) >> 9;
411: if(group == 0) {
412: switch(token|0x4000) {
413: case CF_NIL : image[i]=0; break;
414: #if !defined(DOUBLY_INDIRECT)
415: case CF(DOCOL) :
416: case CF(DOVAR) :
417: case CF(DOCON) :
418: case CF(DOVAL) :
419: case CF(DOUSER) :
420: case CF(DODEFER) :
421: case CF(DOFIELD) : MAKE_CF(image+i,symbols[CF(token)]); break;
422: case CF(DOESJUMP): image[i]=0; break;
423: #endif /* !defined(DOUBLY_INDIRECT) */
424: case CF(DODOES) :
425: MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
426: break;
427: default : /* backward compatibility */
428: /* printf("Code field generation image[%x]:=CFA(%x)\n",
429: i, CF(image[i])); */
430: if (CF((token | 0x4000))<max_symbols) {
431: image[i]=(Cell)CFA(CF(token));
432: #ifdef DIRECT_THREADED
433: if ((token & 0x4000) == 0) { /* threade code, no CFA */
434: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
435: compile_prim1(0);
436: compile_prim1(&image[i]);
437: }
438: #endif
439: } else
440: 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);
441: }
442: } else {
443: int tok = -token & 0x1FF;
444: if (tok < (groups[group+1]-groups[group])) {
445: #if defined(DOUBLY_INDIRECT)
446: image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
447: #else
448: image[i]=(Cell)CFA((groups[group]+tok));
449: #endif
450: #ifdef DIRECT_THREADED
451: if ((token & 0x4000) == 0) { /* threade code, no CFA */
452: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
453: compile_prim1(0);
454: compile_prim1(&image[i]);
455: }
456: #endif
457: } else
458: fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
459: }
460: } else {
461: /* if base is > 0: 0 is a null reference so don't adjust*/
462: if (token>=base) {
463: image[i]+=(Cell)start;
464: }
465: }
466: }
467: }
468: }
469: free(targets);
470: finish_code();
471: ((ImageHeader*)(image))->base = (Address) image;
472: }
473:
474: #ifndef DOUBLY_INDIRECT
475: static UCell checksum(Label symbols[])
476: {
477: UCell r=PRIM_VERSION;
478: Cell i;
479:
480: for (i=DOCOL; i<=DOESJUMP; i++) {
481: r ^= (UCell)(symbols[i]);
482: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
483: }
484: #ifdef DIRECT_THREADED
485: /* we have to consider all the primitives */
486: for (; symbols[i]!=(Label)0; i++) {
487: r ^= (UCell)(symbols[i]);
488: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
489: }
490: #else
491: /* in indirect threaded code all primitives are accessed through the
492: symbols table, so we just have to put the base address of symbols
493: in the checksum */
494: r ^= (UCell)symbols;
495: #endif
496: return r;
497: }
498: #endif
499:
500: static Address verbose_malloc(Cell size)
501: {
502: Address r;
503: /* leave a little room (64B) for stack underflows */
504: if ((r = malloc(size+64))==NULL) {
505: perror(progname);
506: exit(1);
507: }
508: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
509: debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
510: return r;
511: }
512:
513: static Address next_address=0;
514: static void after_alloc(Address r, Cell size)
515: {
516: if (r != (Address)-1) {
517: debugp(stderr, "success, address=$%lx\n", (long) r);
518: #if 0
519: /* not needed now that we protect the stacks with mprotect */
520: if (pagesize != 1)
521: next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
522: #endif
523: } else {
524: debugp(stderr, "failed: %s\n", strerror(errno));
525: }
526: }
527:
528: #ifndef MAP_FAILED
529: #define MAP_FAILED ((Address) -1)
530: #endif
531: #ifndef MAP_FILE
532: # define MAP_FILE 0
533: #endif
534: #ifndef MAP_PRIVATE
535: # define MAP_PRIVATE 0
536: #endif
537: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
538: # define MAP_ANON MAP_ANONYMOUS
539: #endif
540:
541: #if defined(HAVE_MMAP)
542: static Address alloc_mmap(Cell size)
543: {
544: Address r;
545:
546: #if defined(MAP_ANON)
547: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
548: r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE|map_noreserve, -1, 0);
549: #else /* !defined(MAP_ANON) */
550: /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
551: apparently defaults) */
552: static int dev_zero=-1;
553:
554: if (dev_zero == -1)
555: dev_zero = open("/dev/zero", O_RDONLY);
556: if (dev_zero == -1) {
557: r = MAP_FAILED;
558: debugp(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ",
559: strerror(errno));
560: } else {
561: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
562: r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE|map_noreserve, dev_zero, 0);
563: }
564: #endif /* !defined(MAP_ANON) */
565: after_alloc(r, size);
566: return r;
567: }
568:
569: static void page_noaccess(Address a)
570: {
571: /* try mprotect first; with munmap the page might be allocated later */
572: debugp(stderr, "try mprotect(%p,%ld,PROT_NONE); ", a, (long)pagesize);
573: if (mprotect(a, pagesize, PROT_NONE)==0) {
574: debugp(stderr, "ok\n");
575: return;
576: }
577: debugp(stderr, "failed: %s\n", strerror(errno));
578: debugp(stderr, "try munmap(%p,%ld); ", a, (long)pagesize);
579: if (munmap(a,pagesize)==0) {
580: debugp(stderr, "ok\n");
581: return;
582: }
583: debugp(stderr, "failed: %s\n", strerror(errno));
584: }
585:
586: static size_t wholepage(size_t n)
587: {
588: return (n+pagesize-1)&~(pagesize-1);
589: }
590: #endif
591:
592: Address gforth_alloc(Cell size)
593: {
594: #if HAVE_MMAP
595: Address r;
596:
597: r=alloc_mmap(size);
598: if (r!=(Address)MAP_FAILED)
599: return r;
600: #endif /* HAVE_MMAP */
601: /* use malloc as fallback */
602: return verbose_malloc(size);
603: }
604:
605: static Address dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
606: {
607: Address image = MAP_FAILED;
608:
609: #if defined(HAVE_MMAP)
610: if (offset==0) {
611: image=alloc_mmap(dictsize);
612: if (image != (Address)MAP_FAILED) {
613: Address image1;
614: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
615: image1 = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE|map_noreserve, fileno(file), 0);
616: after_alloc(image1,dictsize);
617: if (image1 == (Address)MAP_FAILED)
618: goto read_image;
619: }
620: }
621: #endif /* defined(HAVE_MMAP) */
622: if (image == (Address)MAP_FAILED) {
623: image = gforth_alloc(dictsize+offset)+offset;
624: read_image:
625: rewind(file); /* fseek(imagefile,0L,SEEK_SET); */
626: fread(image, 1, imagesize, file);
627: }
628: return image;
629: }
630: #endif
631:
632: void set_stack_sizes(ImageHeader * header)
633: {
634: if (dictsize==0)
635: dictsize = header->dict_size;
636: if (dsize==0)
637: dsize = header->data_stack_size;
638: if (rsize==0)
639: rsize = header->return_stack_size;
640: if (fsize==0)
641: fsize = header->fp_stack_size;
642: if (lsize==0)
643: lsize = header->locals_stack_size;
644: dictsize=maxaligned(dictsize);
645: dsize=maxaligned(dsize);
646: rsize=maxaligned(rsize);
647: lsize=maxaligned(lsize);
648: fsize=maxaligned(fsize);
649: }
650:
651: #ifdef STANDALONE
652: void alloc_stacks(ImageHeader * h)
653: {
654: #define SSTACKSIZE 0x200
655: static Cell dstack[SSTACKSIZE+1];
656: static Cell rstack[SSTACKSIZE+1];
657:
658: h->dict_size=dictsize;
659: h->data_stack_size=dsize;
660: h->fp_stack_size=fsize;
661: h->return_stack_size=rsize;
662: h->locals_stack_size=lsize;
663:
664: h->data_stack_base=dstack+SSTACKSIZE;
665: // h->fp_stack_base=gforth_alloc(fsize);
666: h->return_stack_base=rstack+SSTACKSIZE;
667: // h->locals_stack_base=gforth_alloc(lsize);
668: }
669: #else
670: void alloc_stacks(ImageHeader * h)
671: {
672: h->dict_size=dictsize;
673: h->data_stack_size=dsize;
674: h->fp_stack_size=fsize;
675: h->return_stack_size=rsize;
676: h->locals_stack_size=lsize;
677:
678: #if defined(HAVE_MMAP) && !defined(STANDALONE)
679: if (pagesize > 1) {
680: size_t p = pagesize;
681: size_t totalsize =
682: wholepage(dsize)+wholepage(fsize)+wholepage(rsize)+wholepage(lsize)+5*p;
683: Address a = alloc_mmap(totalsize);
684: if (a != (Address)MAP_FAILED) {
685: page_noaccess(a); a+=p; h-> data_stack_base=a; a+=wholepage(dsize);
686: page_noaccess(a); a+=p; h-> fp_stack_base=a; a+=wholepage(fsize);
687: page_noaccess(a); a+=p; h->return_stack_base=a; a+=wholepage(rsize);
688: page_noaccess(a); a+=p; h->locals_stack_base=a; a+=wholepage(lsize);
689: page_noaccess(a);
690: debugp(stderr,"stack addresses: d=%p f=%p r=%p l=%p\n",
691: h->data_stack_base,
692: h->fp_stack_base,
693: h->return_stack_base,
694: h->locals_stack_base);
695: return;
696: }
697: }
698: #endif
699: h->data_stack_base=gforth_alloc(dsize);
700: h->fp_stack_base=gforth_alloc(fsize);
701: h->return_stack_base=gforth_alloc(rsize);
702: h->locals_stack_base=gforth_alloc(lsize);
703: }
704: #endif
705:
706: #warning You can ignore the warnings about clobbered variables in gforth_go
707: int gforth_go(Address image, int stack, Cell *entries)
708: {
709: volatile ImageHeader *image_header = (ImageHeader *)image;
710: Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
711: Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
712: Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
713: #ifdef GFORTH_DEBUGGING
714: volatile Cell *orig_rp0=rp0;
715: #endif
716: Address lp0=image_header->locals_stack_base + lsize;
717: Xt *ip0=(Xt *)(image_header->boot_entry);
718: #ifdef SYSSIGNALS
719: int throw_code;
720: #endif
721:
722: /* ensure that the cached elements (if any) are accessible */
723: #if !(defined(GFORTH_DEBUGGING) || defined(INDIRECT_THREADED) || defined(DOUBLY_INDIRECT) || defined(VM_PROFILING))
724: sp0 -= 8; /* make stuff below bottom accessible for stack caching */
725: fp0--;
726: #endif
727:
728: for(;stack>0;stack--)
729: *--sp0=entries[stack-1];
730:
731: #if defined(SYSSIGNALS) && !defined(STANDALONE)
732: get_winsize();
733:
734: install_signal_handlers(); /* right place? */
735:
736: if ((throw_code=setjmp(throw_jmp_buf))) {
737: static Cell signal_data_stack[24];
738: static Cell signal_return_stack[16];
739: static Float signal_fp_stack[1];
740:
741: signal_data_stack[15]=throw_code;
742:
743: #ifdef GFORTH_DEBUGGING
744: debugp(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
745: throw_code, saved_ip, rp);
746: if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
747: /* no rstack overflow or underflow */
748: rp0 = rp;
749: *--rp0 = (Cell)saved_ip;
750: }
751: else /* I love non-syntactic ifdefs :-) */
752: rp0 = signal_return_stack+16;
753: #else /* !defined(GFORTH_DEBUGGING) */
754: debugp(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
755: rp0 = signal_return_stack+16;
756: #endif /* !defined(GFORTH_DEBUGGING) */
757: /* fprintf(stderr, "rp=$%x\n",rp0);*/
758:
759: return((int)(Cell)gforth_engine(image_header->throw_entry, signal_data_stack+15,
760: rp0, signal_fp_stack, 0));
761: }
762: #endif
763:
764: return((int)(Cell)gforth_engine(ip0,sp0,rp0,fp0,lp0));
765: }
766:
767: #if !defined(INCLUDE_IMAGE) && !defined(STANDALONE)
768: static void print_sizes(Cell sizebyte)
769: /* print size information */
770: {
771: static char* endianstring[]= { " big","little" };
772:
773: fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
774: endianstring[sizebyte & 1],
775: 1 << ((sizebyte >> 1) & 3),
776: 1 << ((sizebyte >> 3) & 3),
777: 1 << ((sizebyte >> 5) & 3));
778: }
779:
780: /* static superinstruction stuff */
781:
782: struct cost { /* super_info might be a more accurate name */
783: char loads; /* number of stack loads */
784: char stores; /* number of stack stores */
785: char updates; /* number of stack pointer updates */
786: char branch; /* is it a branch (SET_IP) */
787: unsigned char state_in; /* state on entry */
788: unsigned char state_out; /* state on exit */
789: unsigned char imm_ops; /* number of immediate operands */
790: short offset; /* offset into super2 table */
791: unsigned char length; /* number of components */
792: };
793:
794: PrimNum super2[] = {
795: #include SUPER2_I
796: };
797:
798: struct cost super_costs[] = {
799: #include COSTS_I
800: };
801:
802: struct super_state {
803: struct super_state *next;
804: PrimNum super;
805: };
806:
807: #define HASH_SIZE 256
808:
809: struct super_table_entry {
810: struct super_table_entry *next;
811: PrimNum *start;
812: short length;
813: struct super_state *ss_list; /* list of supers */
814: } *super_table[HASH_SIZE];
815: int max_super=2;
816:
817: struct super_state *state_transitions=NULL;
818:
819: static int hash_super(PrimNum *start, int length)
820: {
821: int i, r;
822:
823: for (i=0, r=0; i<length; i++) {
824: r <<= 1;
825: r += start[i];
826: }
827: return r & (HASH_SIZE-1);
828: }
829:
830: static struct super_state **lookup_super(PrimNum *start, int length)
831: {
832: int hash=hash_super(start,length);
833: struct super_table_entry *p = super_table[hash];
834:
835: /* assert(length >= 2); */
836: for (; p!=NULL; p = p->next) {
837: if (length == p->length &&
838: memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
839: return &(p->ss_list);
840: }
841: return NULL;
842: }
843:
844: static void prepare_super_table()
845: {
846: int i;
847: int nsupers = 0;
848:
849: for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
850: struct cost *c = &super_costs[i];
851: if ((c->length < 2 || nsupers < static_super_number) &&
852: c->state_in < maxstates && c->state_out < maxstates) {
853: struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
854: struct super_state *ss = malloc(sizeof(struct super_state));
855: ss->super= i;
856: if (c->offset==N_noop && i != N_noop) {
857: if (is_relocatable(i)) {
858: ss->next = state_transitions;
859: state_transitions = ss;
860: }
861: } else if (ss_listp != NULL) {
862: ss->next = *ss_listp;
863: *ss_listp = ss;
864: } else {
865: int hash = hash_super(super2+c->offset, c->length);
866: struct super_table_entry **p = &super_table[hash];
867: struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
868: ss->next = NULL;
869: e->next = *p;
870: e->start = super2 + c->offset;
871: e->length = c->length;
872: e->ss_list = ss;
873: *p = e;
874: }
875: if (c->length > max_super)
876: max_super = c->length;
877: if (c->length >= 2)
878: nsupers++;
879: }
880: }
881: debugp(stderr, "Using %d static superinsts\n", nsupers);
882: }
883:
884: /* dynamic replication/superinstruction stuff */
885:
886: #ifndef NO_DYNAMIC
887: static int compare_priminfo_length(const void *_a, const void *_b)
888: {
889: PrimInfo **a = (PrimInfo **)_a;
890: PrimInfo **b = (PrimInfo **)_b;
891: Cell diff = (*a)->length - (*b)->length;
892: if (diff)
893: return diff;
894: else /* break ties by start address; thus the decompiler produces
895: the earliest primitive with the same code (e.g. noop instead
896: of (char) and @ instead of >code-address */
897: return (*b)->start - (*a)->start;
898: }
899: #endif /* !defined(NO_DYNAMIC) */
900:
901: static char MAYBE_UNUSED superend[]={
902: #include PRIM_SUPEREND_I
903: };
904:
905: Cell npriminfos=0;
906:
907: Label goto_start;
908: Cell goto_len;
909:
910: #ifndef NO_DYNAMIC
911: static int compare_labels(const void *pa, const void *pb)
912: {
913: Label a = *(Label *)pa;
914: Label b = *(Label *)pb;
915: return a-b;
916: }
917: #endif
918:
919: static Label bsearch_next(Label key, Label *a, UCell n)
920: /* a is sorted; return the label >=key that is the closest in a;
921: return NULL if there is no label in a >=key */
922: {
923: int mid = (n-1)/2;
924: if (n<1)
925: return NULL;
926: if (n == 1) {
927: if (a[0] < key)
928: return NULL;
929: else
930: return a[0];
931: }
932: if (a[mid] < key)
933: return bsearch_next(key, a+mid+1, n-mid-1);
934: else
935: return bsearch_next(key, a, mid+1);
936: }
937:
938: static void check_prims(Label symbols1[])
939: {
940: int i;
941: #ifndef NO_DYNAMIC
942: Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted, *goto_p;
943: int nends1j;
944: #endif
945:
946: if (debug)
947: #ifdef __VERSION__
948: fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
949: #else
950: #define xstr(s) str(s)
951: #define str(s) #s
952: fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n");
953: #endif
954: for (i=0; symbols1[i]!=0; i++)
955: ;
956: npriminfos = i;
957:
958: #ifndef NO_DYNAMIC
959: if (no_dynamic)
960: return;
961: symbols2=gforth_engine2(0,0,0,0,0);
962: #if NO_IP
963: symbols3=gforth_engine3(0,0,0,0,0);
964: #else
965: symbols3=symbols1;
966: #endif
967: ends1 = symbols1+i+1;
968: ends1j = ends1+i;
969: goto_p = ends1j+i+1; /* goto_p[0]==before; ...[1]==after;*/
970: nends1j = i+1;
971: ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
972: memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
973: qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
974:
975: /* check whether the "goto *" is relocatable */
976: goto_len = goto_p[1]-goto_p[0];
977: debugp(stderr, "goto * %p %p len=%ld\n",
978: goto_p[0],symbols2[goto_p-symbols1],goto_len);
979: if (memcmp(goto_p[0],symbols2[goto_p-symbols1],goto_len)!=0) { /* unequal */
980: no_dynamic=1;
981: debugp(stderr," not relocatable, disabling dynamic code generation\n");
982: init_ss_cost();
983: return;
984: }
985: goto_start = goto_p[0];
986:
987: priminfos = calloc(i,sizeof(PrimInfo));
988: for (i=0; symbols1[i]!=0; i++) {
989: int prim_len = ends1[i]-symbols1[i];
990: PrimInfo *pi=&priminfos[i];
991: struct cost *sc=&super_costs[i];
992: int j=0;
993: char *s1 = (char *)symbols1[i];
994: char *s2 = (char *)symbols2[i];
995: char *s3 = (char *)symbols3[i];
996: Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
997:
998: pi->start = s1;
999: pi->superend = superend[i]|no_super;
1000: pi->length = prim_len;
1001: pi->restlength = endlabel - symbols1[i] - pi->length;
1002: pi->nimmargs = 0;
1003: relocs++;
1004: debugp(stderr, "%-15s %d-%d %4d %p %p %p len=%3ld rest=%2ld send=%1d",
1005: prim_names[i], sc->state_in, sc->state_out,
1006: i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength),
1007: pi->superend);
1008: if (endlabel == NULL) {
1009: pi->start = NULL; /* not relocatable */
1010: if (pi->length<0) pi->length=100;
1011: debugp(stderr,"\n non_reloc: no J label > start found\n");
1012: relocs--;
1013: nonrelocs++;
1014: continue;
1015: }
1016: if (ends1[i] > endlabel && !pi->superend) {
1017: pi->start = NULL; /* not relocatable */
1018: pi->length = endlabel-symbols1[i];
1019: debugp(stderr,"\n non_reloc: there is a J label before the K label (restlength<0)\n");
1020: relocs--;
1021: nonrelocs++;
1022: continue;
1023: }
1024: if (ends1[i] < pi->start && !pi->superend) {
1025: pi->start = NULL; /* not relocatable */
1026: pi->length = endlabel-symbols1[i];
1027: debugp(stderr,"\n non_reloc: K label before I label (length<0)\n");
1028: relocs--;
1029: nonrelocs++;
1030: continue;
1031: }
1032: assert(pi->length>=0);
1033: assert(pi->restlength >=0);
1034: while (j<(pi->length+pi->restlength)) {
1035: if (s1[j]==s3[j]) {
1036: if (s1[j] != s2[j]) {
1037: pi->start = NULL; /* not relocatable */
1038: debugp(stderr,"\n non_reloc: engine1!=engine2 offset %3d",j);
1039: /* assert(j<prim_len); */
1040: relocs--;
1041: nonrelocs++;
1042: break;
1043: }
1044: j++;
1045: } else {
1046: struct immarg *ia=&pi->immargs[pi->nimmargs];
1047:
1048: pi->nimmargs++;
1049: ia->offset=j;
1050: if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
1051: ia->rel=0;
1052: debugp(stderr,"\n absolute immarg: offset %3d",j);
1053: } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
1054: symbols1[DOESJUMP+1]) {
1055: ia->rel=1;
1056: debugp(stderr,"\n relative immarg: offset %3d",j);
1057: } else {
1058: pi->start = NULL; /* not relocatable */
1059: debugp(stderr,"\n non_reloc: engine1!=engine3 offset %3d",j);
1060: /* assert(j<prim_len);*/
1061: relocs--;
1062: nonrelocs++;
1063: break;
1064: }
1065: j+=4;
1066: }
1067: }
1068: debugp(stderr,"\n");
1069: }
1070: decomp_prims = calloc(i,sizeof(PrimInfo *));
1071: for (i=DOESJUMP+1; i<npriminfos; i++)
1072: decomp_prims[i] = &(priminfos[i]);
1073: qsort(decomp_prims+DOESJUMP+1, npriminfos-DOESJUMP-1, sizeof(PrimInfo *),
1074: compare_priminfo_length);
1075: #endif
1076: }
1077:
1078: static void flush_to_here(void)
1079: {
1080: #ifndef NO_DYNAMIC
1081: if (start_flush)
1082: FLUSH_ICACHE(start_flush, code_here-start_flush);
1083: start_flush=code_here;
1084: #endif
1085: }
1086:
1087: static void align_code(void)
1088: /* align code_here on some platforms */
1089: {
1090: #ifndef NO_DYNAMIC
1091: #if defined(CODE_PADDING)
1092: Cell alignment = CODE_ALIGNMENT;
1093: static char nops[] = CODE_PADDING;
1094: UCell maxpadding=MAX_PADDING;
1095: UCell offset = ((UCell)code_here)&(alignment-1);
1096: UCell length = alignment-offset;
1097: if (length <= maxpadding) {
1098: memcpy(code_here,nops+offset,length);
1099: code_here += length;
1100: }
1101: #endif /* defined(CODE_PADDING) */
1102: #endif /* defined(NO_DYNAMIC */
1103: }
1104:
1105: #ifndef NO_DYNAMIC
1106: static void append_jump(void)
1107: {
1108: if (last_jump) {
1109: PrimInfo *pi = &priminfos[last_jump];
1110:
1111: memcpy(code_here, pi->start+pi->length, pi->restlength);
1112: code_here += pi->restlength;
1113: memcpy(code_here, goto_start, goto_len);
1114: code_here += goto_len;
1115: align_code();
1116: last_jump=0;
1117: }
1118: }
1119:
1120: /* Gforth remembers all code blocks in this list. On forgetting (by
1121: executing a marker) the code blocks are not freed (because Gforth does
1122: not remember how they were allocated; hmm, remembering that might be
1123: easier and cleaner). Instead, code_here etc. are reset to the old
1124: value, and the "forgotten" code blocks are reused when they are
1125: needed. */
1126:
1127: struct code_block_list {
1128: struct code_block_list *next;
1129: Address block;
1130: Cell size;
1131: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
1132:
1133: static Address append_prim(Cell p)
1134: {
1135: PrimInfo *pi = &priminfos[p];
1136: Address old_code_here = code_here;
1137:
1138: if (code_area+code_area_size < code_here+pi->length+pi->restlength+goto_len+CODE_ALIGNMENT) {
1139: struct code_block_list *p;
1140: append_jump();
1141: flush_to_here();
1142: if (*next_code_blockp == NULL) {
1143: code_here = start_flush = code_area = gforth_alloc(code_area_size);
1144: p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
1145: *next_code_blockp = p;
1146: p->next = NULL;
1147: p->block = code_here;
1148: p->size = code_area_size;
1149: } else {
1150: p = *next_code_blockp;
1151: code_here = start_flush = code_area = p->block;
1152: }
1153: old_code_here = code_here;
1154: next_code_blockp = &(p->next);
1155: }
1156: memcpy(code_here, pi->start, pi->length);
1157: code_here += pi->length;
1158: return old_code_here;
1159: }
1160: #endif
1161:
1162: int forget_dyncode(Address code)
1163: {
1164: #ifdef NO_DYNAMIC
1165: return -1;
1166: #else
1167: struct code_block_list *p, **pp;
1168:
1169: for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
1170: if (code >= p->block && code < p->block+p->size) {
1171: next_code_blockp = &(p->next);
1172: code_here = start_flush = code;
1173: code_area = p->block;
1174: last_jump = 0;
1175: return -1;
1176: }
1177: }
1178: return -no_dynamic;
1179: #endif /* !defined(NO_DYNAMIC) */
1180: }
1181:
1182: static long dyncodesize(void)
1183: {
1184: #ifndef NO_DYNAMIC
1185: struct code_block_list *p;
1186: long size=0;
1187: for (p=code_block_list; p!=NULL; p=p->next) {
1188: if (code_here >= p->block && code_here < p->block+p->size)
1189: return size + (code_here - p->block);
1190: else
1191: size += p->size;
1192: }
1193: #endif /* !defined(NO_DYNAMIC) */
1194: return 0;
1195: }
1196:
1197: Label decompile_code(Label _code)
1198: {
1199: #ifdef NO_DYNAMIC
1200: return _code;
1201: #else /* !defined(NO_DYNAMIC) */
1202: Cell i;
1203: struct code_block_list *p;
1204: Address code=_code;
1205:
1206: /* first, check if we are in code at all */
1207: for (p = code_block_list;; p = p->next) {
1208: if (p == NULL)
1209: return code;
1210: if (code >= p->block && code < p->block+p->size)
1211: break;
1212: }
1213: /* reverse order because NOOP might match other prims */
1214: for (i=npriminfos-1; i>DOESJUMP; i--) {
1215: PrimInfo *pi=decomp_prims[i];
1216: if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1217: return vm_prims[super2[super_costs[pi-priminfos].offset]];
1218: /* return pi->start;*/
1219: }
1220: return code;
1221: #endif /* !defined(NO_DYNAMIC) */
1222: }
1223:
1224: #ifdef NO_IP
1225: int nbranchinfos=0;
1226:
1227: struct branchinfo {
1228: Label **targetpp; /* **(bi->targetpp) is the target */
1229: Cell *addressptr; /* store the target here */
1230: } branchinfos[100000];
1231:
1232: int ndoesexecinfos=0;
1233: struct doesexecinfo {
1234: int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1235: Label *targetp; /*target for branch (because this is not in threaded code)*/
1236: Cell *xt; /* cfa of word whose does-code needs calling */
1237: } doesexecinfos[10000];
1238:
1239: static void set_rel_target(Cell *source, Label target)
1240: {
1241: *source = ((Cell)target)-(((Cell)source)+4);
1242: }
1243:
1244: static void register_branchinfo(Label source, Cell *targetpp)
1245: {
1246: struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1247: bi->targetpp = (Label **)targetpp;
1248: bi->addressptr = (Cell *)source;
1249: nbranchinfos++;
1250: }
1251:
1252: static Address compile_prim1arg(PrimNum p, Cell **argp)
1253: {
1254: Address old_code_here=append_prim(p);
1255:
1256: assert(vm_prims[p]==priminfos[p].start);
1257: *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
1258: return old_code_here;
1259: }
1260:
1261: static Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
1262: {
1263: PrimInfo *pi = &priminfos[N_call2];
1264: Address old_code_here = append_prim(N_call2);
1265:
1266: *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
1267: register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
1268: return old_code_here;
1269: }
1270: #endif
1271:
1272: void finish_code(void)
1273: {
1274: #ifdef NO_IP
1275: Cell i;
1276:
1277: compile_prim1(NULL);
1278: for (i=0; i<ndoesexecinfos; i++) {
1279: struct doesexecinfo *dei = &doesexecinfos[i];
1280: dei->targetp = (Label *)DOES_CODE1((dei->xt));
1281: branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
1282: }
1283: ndoesexecinfos = 0;
1284: for (i=0; i<nbranchinfos; i++) {
1285: struct branchinfo *bi=&branchinfos[i];
1286: set_rel_target(bi->addressptr, **(bi->targetpp));
1287: }
1288: nbranchinfos = 0;
1289: #else
1290: compile_prim1(NULL);
1291: #endif
1292: flush_to_here();
1293: }
1294:
1295: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1296: #ifdef NO_IP
1297: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1298: /* compile prim #p dynamically (mod flags etc.) and return start
1299: address of generated code for putting it into the threaded
1300: code. This function is only called if all the associated
1301: inline arguments of p are already in place (at tcp[1] etc.) */
1302: {
1303: PrimInfo *pi=&priminfos[p];
1304: Cell *next_code_target=NULL;
1305: Address codeaddr;
1306: Address primstart;
1307:
1308: assert(p<npriminfos);
1309: if (p==N_execute || p==N_perform || p==N_lit_perform) {
1310: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1311: primstart = append_prim(p);
1312: goto other_prim;
1313: } else if (p==N_call) {
1314: codeaddr = compile_call2(tcp+1, &next_code_target);
1315: } else if (p==N_does_exec) {
1316: struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1317: Cell *arg;
1318: codeaddr = compile_prim1arg(N_lit,&arg);
1319: *arg = (Cell)PFA(tcp[1]);
1320: /* we cannot determine the callee now (last_start[1] may be a
1321: forward reference), so just register an arbitrary target, and
1322: register in dei that we need to fix this before resolving
1323: branches */
1324: dei->branchinfo = nbranchinfos;
1325: dei->xt = (Cell *)(tcp[1]);
1326: compile_call2(0, &next_code_target);
1327: } else if (!is_relocatable(p)) {
1328: Cell *branch_target;
1329: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1330: compile_prim1arg(N_branch,&branch_target);
1331: set_rel_target(branch_target,vm_prims[p]);
1332: } else {
1333: unsigned j;
1334:
1335: codeaddr = primstart = append_prim(p);
1336: other_prim:
1337: for (j=0; j<pi->nimmargs; j++) {
1338: struct immarg *ia = &(pi->immargs[j]);
1339: Cell *argp = tcp + pi->nimmargs - j;
1340: Cell argval = *argp; /* !! specific to prims */
1341: if (ia->rel) { /* !! assumption: relative refs are branches */
1342: register_branchinfo(primstart + ia->offset, argp);
1343: } else /* plain argument */
1344: *(Cell *)(primstart + ia->offset) = argval;
1345: }
1346: }
1347: if (next_code_target!=NULL)
1348: *next_code_target = (Cell)code_here;
1349: return (Cell)codeaddr;
1350: }
1351: #else /* !defined(NO_IP) */
1352: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1353: /* compile prim #p dynamically (mod flags etc.) and return start
1354: address of generated code for putting it into the threaded code */
1355: {
1356: Cell static_prim = (Cell)vm_prims[p];
1357: #if defined(NO_DYNAMIC)
1358: return static_prim;
1359: #else /* !defined(NO_DYNAMIC) */
1360: Address old_code_here;
1361:
1362: if (no_dynamic)
1363: return static_prim;
1364: if (p>=npriminfos || !is_relocatable(p)) {
1365: append_jump();
1366: return static_prim;
1367: }
1368: old_code_here = append_prim(p);
1369: last_jump = p;
1370: if (priminfos[p].superend)
1371: append_jump();
1372: return (Cell)old_code_here;
1373: #endif /* !defined(NO_DYNAMIC) */
1374: }
1375: #endif /* !defined(NO_IP) */
1376: #endif
1377:
1378: #ifndef NO_DYNAMIC
1379: static int cost_codesize(int prim)
1380: {
1381: return priminfos[prim].length;
1382: }
1383: #endif
1384:
1385: static int cost_ls(int prim)
1386: {
1387: struct cost *c = super_costs+prim;
1388:
1389: return c->loads + c->stores;
1390: }
1391:
1392: static int cost_lsu(int prim)
1393: {
1394: struct cost *c = super_costs+prim;
1395:
1396: return c->loads + c->stores + c->updates;
1397: }
1398:
1399: static int cost_nexts(int prim)
1400: {
1401: return 1;
1402: }
1403:
1404: typedef int Costfunc(int);
1405: Costfunc *ss_cost = /* cost function for optimize_bb */
1406: #ifdef NO_DYNAMIC
1407: cost_lsu;
1408: #else
1409: cost_codesize;
1410: #endif
1411:
1412: struct {
1413: Costfunc *costfunc;
1414: char *metricname;
1415: long sum;
1416: } cost_sums[] = {
1417: #ifndef NO_DYNAMIC
1418: { cost_codesize, "codesize", 0 },
1419: #endif
1420: { cost_ls, "ls", 0 },
1421: { cost_lsu, "lsu", 0 },
1422: { cost_nexts, "nexts", 0 }
1423: };
1424:
1425: #ifndef NO_DYNAMIC
1426: void init_ss_cost(void) {
1427: if (no_dynamic && ss_cost == cost_codesize) {
1428: ss_cost = cost_nexts;
1429: cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
1430: debugp(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
1431: }
1432: }
1433: #endif
1434:
1435: #define MAX_BB 128 /* maximum number of instructions in BB */
1436: #define INF_COST 1000000 /* infinite cost */
1437: #define CANONICAL_STATE 0
1438:
1439: struct waypoint {
1440: int cost; /* the cost from here to the end */
1441: PrimNum inst; /* the inst used from here to the next waypoint */
1442: char relocatable; /* the last non-transition was relocatable */
1443: char no_transition; /* don't use the next transition (relocatability)
1444: * or this transition (does not change state) */
1445: };
1446:
1447: struct tpa_state { /* tree parsing automaton (like) state */
1448: /* labeling is back-to-front */
1449: struct waypoint *inst; /* in front of instruction */
1450: struct waypoint *trans; /* in front of instruction and transition */
1451: };
1452:
1453: struct tpa_state *termstate = NULL; /* initialized in loader() */
1454:
1455: /* statistics about tree parsing (lazyburg) stuff */
1456: long lb_basic_blocks = 0;
1457: long lb_labeler_steps = 0;
1458: long lb_labeler_automaton = 0;
1459: long lb_labeler_dynprog = 0;
1460: long lb_newstate_equiv = 0;
1461: long lb_newstate_new = 0;
1462: long lb_applicable_base_rules = 0;
1463: long lb_applicable_chain_rules = 0;
1464:
1465: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1466: static void init_waypoints(struct waypoint ws[])
1467: {
1468: int k;
1469:
1470: for (k=0; k<maxstates; k++)
1471: ws[k].cost=INF_COST;
1472: }
1473:
1474: static struct tpa_state *empty_tpa_state()
1475: {
1476: struct tpa_state *s = malloc(sizeof(struct tpa_state));
1477:
1478: s->inst = calloc(maxstates,sizeof(struct waypoint));
1479: init_waypoints(s->inst);
1480: s->trans = calloc(maxstates,sizeof(struct waypoint));
1481: /* init_waypoints(s->trans);*/
1482: return s;
1483: }
1484:
1485: static void transitions(struct tpa_state *t)
1486: {
1487: int k;
1488: struct super_state *l;
1489:
1490: for (k=0; k<maxstates; k++) {
1491: t->trans[k] = t->inst[k];
1492: t->trans[k].no_transition = 1;
1493: }
1494: for (l = state_transitions; l != NULL; l = l->next) {
1495: PrimNum s = l->super;
1496: int jcost;
1497: struct cost *c=super_costs+s;
1498: struct waypoint *wi=&(t->trans[c->state_in]);
1499: struct waypoint *wo=&(t->inst[c->state_out]);
1500: lb_applicable_chain_rules++;
1501: if (wo->cost == INF_COST)
1502: continue;
1503: jcost = wo->cost + ss_cost(s);
1504: if (jcost <= wi->cost) {
1505: wi->cost = jcost;
1506: wi->inst = s;
1507: wi->relocatable = wo->relocatable;
1508: wi->no_transition = 0;
1509: /* if (ss_greedy) wi->cost = wo->cost ? */
1510: }
1511: }
1512: }
1513:
1514: static struct tpa_state *make_termstate()
1515: {
1516: struct tpa_state *s = empty_tpa_state();
1517:
1518: s->inst[CANONICAL_STATE].cost = 0;
1519: transitions(s);
1520: return s;
1521: }
1522: #endif
1523:
1524: #define TPA_SIZE 16384
1525:
1526: struct tpa_entry {
1527: struct tpa_entry *next;
1528: PrimNum inst;
1529: struct tpa_state *state_behind; /* note: brack-to-front labeling */
1530: struct tpa_state *state_infront; /* note: brack-to-front labeling */
1531: } *tpa_table[TPA_SIZE];
1532:
1533: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1534: static Cell hash_tpa(PrimNum p, struct tpa_state *t)
1535: {
1536: UCell it = (UCell )t;
1537: return (p+it+(it>>14))&(TPA_SIZE-1);
1538: }
1539:
1540: static struct tpa_state **lookup_tpa(PrimNum p, struct tpa_state *t2)
1541: {
1542: int hash=hash_tpa(p, t2);
1543: struct tpa_entry *te = tpa_table[hash];
1544:
1545: if (tpa_noautomaton) {
1546: static struct tpa_state *t;
1547: t = NULL;
1548: return &t;
1549: }
1550: for (; te!=NULL; te = te->next) {
1551: if (p == te->inst && t2 == te->state_behind)
1552: return &(te->state_infront);
1553: }
1554: te = (struct tpa_entry *)malloc(sizeof(struct tpa_entry));
1555: te->next = tpa_table[hash];
1556: te->inst = p;
1557: te->state_behind = t2;
1558: te->state_infront = NULL;
1559: tpa_table[hash] = te;
1560: return &(te->state_infront);
1561: }
1562:
1563: static void tpa_state_normalize(struct tpa_state *t)
1564: {
1565: /* normalize so cost of canonical state=0; this may result in
1566: negative states for some states */
1567: int d = t->inst[CANONICAL_STATE].cost;
1568: int i;
1569:
1570: for (i=0; i<maxstates; i++) {
1571: if (t->inst[i].cost != INF_COST)
1572: t->inst[i].cost -= d;
1573: if (t->trans[i].cost != INF_COST)
1574: t->trans[i].cost -= d;
1575: }
1576: }
1577:
1578: static int tpa_state_equivalent(struct tpa_state *t1, struct tpa_state *t2)
1579: {
1580: return (memcmp(t1->inst, t2->inst, maxstates*sizeof(struct waypoint)) == 0 &&
1581: memcmp(t1->trans,t2->trans,maxstates*sizeof(struct waypoint)) == 0);
1582: }
1583: #endif
1584:
1585: struct tpa_state_entry {
1586: struct tpa_state_entry *next;
1587: struct tpa_state *state;
1588: } *tpa_state_table[TPA_SIZE];
1589:
1590: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1591: static Cell hash_tpa_state(struct tpa_state *t)
1592: {
1593: int *ti = (int *)(t->inst);
1594: int *tt = (int *)(t->trans);
1595: int r=0;
1596: int i;
1597:
1598: for (i=0; ti+i < (int *)(t->inst+maxstates); i++)
1599: r += ti[i]+tt[i];
1600: return (r+(r>>14)+(r>>22)) & (TPA_SIZE-1);
1601: }
1602:
1603: static struct tpa_state *lookup_tpa_state(struct tpa_state *t)
1604: {
1605: Cell hash = hash_tpa_state(t);
1606: struct tpa_state_entry *te = tpa_state_table[hash];
1607: struct tpa_state_entry *tn;
1608:
1609: if (!tpa_noequiv) {
1610: for (; te!=NULL; te = te->next) {
1611: if (tpa_state_equivalent(t, te->state)) {
1612: lb_newstate_equiv++;
1613: free(t->inst);
1614: free(t->trans);
1615: free(t);
1616: return te->state;
1617: }
1618: }
1619: tn = (struct tpa_state_entry *)malloc(sizeof(struct tpa_state_entry));
1620: tn->next = te;
1621: tn->state = t;
1622: tpa_state_table[hash] = tn;
1623: }
1624: lb_newstate_new++;
1625: if (tpa_trace)
1626: fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
1627: return t;
1628: }
1629:
1630: /* use dynamic programming to find the shortest paths within the basic
1631: block origs[0..ninsts-1] and rewrite the instructions pointed to by
1632: instps to use it */
1633: static void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
1634: {
1635: int i,j;
1636: struct tpa_state *ts[ninsts+1];
1637: int nextdyn, nextstate, no_transition;
1638:
1639: lb_basic_blocks++;
1640: ts[ninsts] = termstate;
1641: #ifndef NO_DYNAMIC
1642: if (print_sequences) {
1643: for (i=0; i<ninsts; i++)
1644: fprintf(stderr, "%s ", prim_names[origs[i]]);
1645: fprintf(stderr, "\n");
1646: }
1647: #endif
1648: for (i=ninsts-1; i>=0; i--) {
1649: struct tpa_state **tp = lookup_tpa(origs[i],ts[i+1]);
1650: struct tpa_state *t = *tp;
1651: lb_labeler_steps++;
1652: if (t) {
1653: ts[i] = t;
1654: lb_labeler_automaton++;
1655: }
1656: else {
1657: lb_labeler_dynprog++;
1658: ts[i] = empty_tpa_state();
1659: for (j=1; j<=max_super && i+j<=ninsts; j++) {
1660: struct super_state **superp = lookup_super(origs+i, j);
1661: if (superp!=NULL) {
1662: struct super_state *supers = *superp;
1663: for (; supers!=NULL; supers = supers->next) {
1664: PrimNum s = supers->super;
1665: int jcost;
1666: struct cost *c=super_costs+s;
1667: struct waypoint *wi=&(ts[i]->inst[c->state_in]);
1668: struct waypoint *wo=&(ts[i+j]->trans[c->state_out]);
1669: int no_transition = wo->no_transition;
1670: lb_applicable_base_rules++;
1671: if (!(is_relocatable(s)) && !wo->relocatable) {
1672: wo=&(ts[i+j]->inst[c->state_out]);
1673: no_transition=1;
1674: }
1675: if (wo->cost == INF_COST)
1676: continue;
1677: jcost = wo->cost + ss_cost(s);
1678: if (jcost <= wi->cost) {
1679: wi->cost = jcost;
1680: wi->inst = s;
1681: wi->relocatable = is_relocatable(s);
1682: wi->no_transition = no_transition;
1683: /* if (ss_greedy) wi->cost = wo->cost ? */
1684: }
1685: }
1686: }
1687: }
1688: transitions(ts[i]);
1689: tpa_state_normalize(ts[i]);
1690: *tp = ts[i] = lookup_tpa_state(ts[i]);
1691: if (tpa_trace)
1692: fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
1693: }
1694: }
1695: /* now rewrite the instructions */
1696: nextdyn=0;
1697: nextstate=CANONICAL_STATE;
1698: no_transition = ((!ts[0]->trans[nextstate].relocatable)
1699: ||ts[0]->trans[nextstate].no_transition);
1700: for (i=0; i<ninsts; i++) {
1701: Cell tc=0, tc2;
1702: if (i==nextdyn) {
1703: if (!no_transition) {
1704: /* process trans */
1705: PrimNum p = ts[i]->trans[nextstate].inst;
1706: struct cost *c = super_costs+p;
1707: assert(ts[i]->trans[nextstate].cost != INF_COST);
1708: assert(c->state_in==nextstate);
1709: tc = compile_prim_dyn(p,NULL);
1710: nextstate = c->state_out;
1711: }
1712: {
1713: /* process inst */
1714: PrimNum p = ts[i]->inst[nextstate].inst;
1715: struct cost *c=super_costs+p;
1716: assert(c->state_in==nextstate);
1717: assert(ts[i]->inst[nextstate].cost != INF_COST);
1718: #if defined(GFORTH_DEBUGGING)
1719: assert(p == origs[i]);
1720: #endif
1721: tc2 = compile_prim_dyn(p,instps[i]);
1722: if (no_transition || !is_relocatable(p))
1723: /* !! actually what we care about is if and where
1724: * compile_prim_dyn() puts NEXTs */
1725: tc=tc2;
1726: no_transition = ts[i]->inst[nextstate].no_transition;
1727: nextstate = c->state_out;
1728: nextdyn += c->length;
1729: }
1730: } else {
1731: #if defined(GFORTH_DEBUGGING)
1732: assert(0);
1733: #endif
1734: tc=0;
1735: /* tc= (Cell)vm_prims[ts[i]->inst[CANONICAL_STATE].inst]; */
1736: }
1737: *(instps[i]) = tc;
1738: }
1739: if (!no_transition) {
1740: PrimNum p = ts[i]->trans[nextstate].inst;
1741: struct cost *c = super_costs+p;
1742: assert(c->state_in==nextstate);
1743: assert(ts[i]->trans[nextstate].cost != INF_COST);
1744: assert(i==nextdyn);
1745: (void)compile_prim_dyn(p,NULL);
1746: nextstate = c->state_out;
1747: }
1748: assert(nextstate==CANONICAL_STATE);
1749: }
1750: #endif
1751:
1752: /* compile *start, possibly rewriting it into a static and/or dynamic
1753: superinstruction */
1754: void compile_prim1(Cell *start)
1755: {
1756: #if defined(DOUBLY_INDIRECT)
1757: Label prim;
1758:
1759: if (start==NULL)
1760: return;
1761: prim = (Label)*start;
1762: if (prim<((Label)(xts+DOESJUMP)) || prim>((Label)(xts+npriminfos))) {
1763: fprintf(stderr,"compile_prim encountered xt %p\n", prim);
1764: *start=(Cell)prim;
1765: return;
1766: } else {
1767: *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
1768: return;
1769: }
1770: #elif defined(INDIRECT_THREADED)
1771: return;
1772: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1773: /* !! does not work, for unknown reasons; but something like this is
1774: probably needed to ensure that we don't call compile_prim_dyn
1775: before the inline arguments are there */
1776: static Cell *instps[MAX_BB];
1777: static PrimNum origs[MAX_BB];
1778: static int ninsts=0;
1779: PrimNum prim_num;
1780:
1781: if (start==NULL || ninsts >= MAX_BB ||
1782: (ninsts>0 && superend[origs[ninsts-1]])) {
1783: /* after bb, or at the start of the next bb */
1784: optimize_rewrite(instps,origs,ninsts);
1785: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
1786: ninsts=0;
1787: if (start==NULL) {
1788: align_code();
1789: return;
1790: }
1791: }
1792: prim_num = ((Xt)*start)-vm_prims;
1793: if(prim_num >= npriminfos) {
1794: optimize_rewrite(instps,origs,ninsts);
1795: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
1796: ninsts=0;
1797: return;
1798: }
1799: assert(ninsts<MAX_BB);
1800: instps[ninsts] = start;
1801: origs[ninsts] = prim_num;
1802: ninsts++;
1803: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1804: }
1805:
1806: #ifndef STANDALONE
1807: Address gforth_loader(FILE *imagefile, char* filename)
1808: /* returns the address of the image proper (after the preamble) */
1809: {
1810: ImageHeader header;
1811: Address image;
1812: Address imp; /* image+preamble */
1813: Char magic[8];
1814: char magic7; /* size byte of magic number */
1815: Cell preamblesize=0;
1816: Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1817: UCell check_sum;
1818: Cell ausize = ((RELINFOBITS == 8) ? 0 :
1819: (RELINFOBITS == 16) ? 1 :
1820: (RELINFOBITS == 32) ? 2 : 3);
1821: Cell charsize = ((sizeof(Char) == 1) ? 0 :
1822: (sizeof(Char) == 2) ? 1 :
1823: (sizeof(Char) == 4) ? 2 : 3) + ausize;
1824: Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
1825: (sizeof(Cell) == 2) ? 1 :
1826: (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1827: Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
1828: #ifdef WORDS_BIGENDIAN
1829: 0
1830: #else
1831: 1
1832: #endif
1833: ;
1834:
1835: vm_prims = gforth_engine(0,0,0,0,0);
1836: check_prims(vm_prims);
1837: prepare_super_table();
1838: #ifndef DOUBLY_INDIRECT
1839: #ifdef PRINT_SUPER_LENGTHS
1840: print_super_lengths();
1841: #endif
1842: check_sum = checksum(vm_prims);
1843: #else /* defined(DOUBLY_INDIRECT) */
1844: check_sum = (UCell)vm_prims;
1845: #endif /* defined(DOUBLY_INDIRECT) */
1846: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1847: termstate = make_termstate();
1848: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1849:
1850: do {
1851: if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1852: fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
1853: progname, filename);
1854: exit(1);
1855: }
1856: preamblesize+=8;
1857: } while(memcmp(magic,"Gforth3",7));
1858: magic7 = magic[7];
1859: if (debug) {
1860: magic[7]='\0';
1861: fprintf(stderr,"Magic found: %s ", magic);
1862: print_sizes(magic7);
1863: }
1864:
1865: if (magic7 != sizebyte)
1866: {
1867: fprintf(stderr,"This image is: ");
1868: print_sizes(magic7);
1869: fprintf(stderr,"whereas the machine is ");
1870: print_sizes(sizebyte);
1871: exit(-2);
1872: };
1873:
1874: fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1875:
1876: set_stack_sizes(&header);
1877:
1878: #if HAVE_GETPAGESIZE
1879: pagesize=getpagesize(); /* Linux/GNU libc offers this */
1880: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
1881: pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
1882: #elif PAGESIZE
1883: pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
1884: #endif
1885: debugp(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1886:
1887: image = dict_alloc_read(imagefile, preamblesize+header.image_size,
1888: preamblesize+dictsize, data_offset);
1889: imp=image+preamblesize;
1890:
1891: alloc_stacks((ImageHeader *)imp);
1892: if (clear_dictionary)
1893: memset(imp+header.image_size, 0, dictsize-header.image_size);
1894: if(header.base==0 || header.base == (Address)0x100) {
1895: Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
1896: Char reloc_bits[reloc_size];
1897: fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1898: fread(reloc_bits, 1, reloc_size, imagefile);
1899: gforth_relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1900: #if 0
1901: { /* let's see what the relocator did */
1902: FILE *snapshot=fopen("snapshot.fi","wb");
1903: fwrite(image,1,imagesize,snapshot);
1904: fclose(snapshot);
1905: }
1906: #endif
1907: }
1908: else if(header.base!=imp) {
1909: fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
1910: progname, (unsigned long)header.base, (unsigned long)imp);
1911: exit(1);
1912: }
1913: if (header.checksum==0)
1914: ((ImageHeader *)imp)->checksum=check_sum;
1915: else if (header.checksum != check_sum) {
1916: fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
1917: progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
1918: exit(1);
1919: }
1920: #ifdef DOUBLY_INDIRECT
1921: ((ImageHeader *)imp)->xt_base = xts;
1922: #endif
1923: fclose(imagefile);
1924:
1925: /* unnecessary, except maybe for CODE words */
1926: /* FLUSH_ICACHE(imp, header.image_size);*/
1927:
1928: return imp;
1929: }
1930: #endif
1931:
1932: /* pointer to last '/' or '\' in file, 0 if there is none. */
1933: static char *onlypath(char *filename)
1934: {
1935: return strrchr(filename, DIRSEP);
1936: }
1937:
1938: static FILE *openimage(char *fullfilename)
1939: {
1940: FILE *image_file;
1941: char * expfilename = tilde_cstr((Char *)fullfilename, strlen(fullfilename), 1);
1942:
1943: image_file=fopen(expfilename,"rb");
1944: if (image_file!=NULL && debug)
1945: fprintf(stderr, "Opened image file: %s\n", expfilename);
1946: return image_file;
1947: }
1948:
1949: /* try to open image file concat(path[0:len],imagename) */
1950: static FILE *checkimage(char *path, int len, char *imagename)
1951: {
1952: int dirlen=len;
1953: char fullfilename[dirlen+strlen((char *)imagename)+2];
1954:
1955: memcpy(fullfilename, path, dirlen);
1956: if (fullfilename[dirlen-1]!=DIRSEP)
1957: fullfilename[dirlen++]=DIRSEP;
1958: strcpy(fullfilename+dirlen,imagename);
1959: return openimage(fullfilename);
1960: }
1961:
1962: static FILE * open_image_file(char * imagename, char * path)
1963: {
1964: FILE * image_file=NULL;
1965: char *origpath=path;
1966:
1967: if(strchr(imagename, DIRSEP)==NULL) {
1968: /* first check the directory where the exe file is in !! 01may97jaw */
1969: if (onlypath(progname))
1970: image_file=checkimage(progname, onlypath(progname)-progname, imagename);
1971: if (!image_file)
1972: do {
1973: char *pend=strchr(path, PATHSEP);
1974: if (pend==NULL)
1975: pend=path+strlen(path);
1976: if (strlen(path)==0) break;
1977: image_file=checkimage(path, pend-path, imagename);
1978: path=pend+(*pend==PATHSEP);
1979: } while (image_file==NULL);
1980: } else {
1981: image_file=openimage(imagename);
1982: }
1983:
1984: if (!image_file) {
1985: fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
1986: progname, imagename, origpath);
1987: exit(1);
1988: }
1989:
1990: return image_file;
1991: }
1992: #endif
1993:
1994: #ifdef STANDALONE_ALLOC
1995: Address gforth_alloc(Cell size)
1996: {
1997: Address r;
1998: /* leave a little room (64B) for stack underflows */
1999: if ((r = malloc(size+64))==NULL) {
2000: perror(progname);
2001: exit(1);
2002: }
2003: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
2004: debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
2005: return r;
2006: }
2007: #endif
2008:
2009: #ifdef HAS_OS
2010: static UCell convsize(char *s, UCell elemsize)
2011: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
2012: of bytes. the letter at the end indicates the unit, where e stands
2013: for the element size. default is e */
2014: {
2015: char *endp;
2016: UCell n,m;
2017:
2018: m = elemsize;
2019: n = strtoul(s,&endp,0);
2020: if (endp!=NULL) {
2021: if (strcmp(endp,"b")==0)
2022: m=1;
2023: else if (strcmp(endp,"k")==0)
2024: m=1024;
2025: else if (strcmp(endp,"M")==0)
2026: m=1024*1024;
2027: else if (strcmp(endp,"G")==0)
2028: m=1024*1024*1024;
2029: else if (strcmp(endp,"T")==0) {
2030: #if (SIZEOF_CHAR_P > 4)
2031: m=1024L*1024*1024*1024;
2032: #else
2033: fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
2034: exit(1);
2035: #endif
2036: } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
2037: fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
2038: exit(1);
2039: }
2040: }
2041: return n*m;
2042: }
2043:
2044: enum {
2045: ss_number = 256,
2046: ss_states,
2047: ss_min_codesize,
2048: ss_min_ls,
2049: ss_min_lsu,
2050: ss_min_nexts,
2051: };
2052:
2053: #ifndef STANDALONE
2054: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
2055: {
2056: int c;
2057:
2058: opterr=0;
2059: while (1) {
2060: int option_index=0;
2061: static struct option opts[] = {
2062: {"appl-image", required_argument, NULL, 'a'},
2063: {"image-file", required_argument, NULL, 'i'},
2064: {"dictionary-size", required_argument, NULL, 'm'},
2065: {"data-stack-size", required_argument, NULL, 'd'},
2066: {"return-stack-size", required_argument, NULL, 'r'},
2067: {"fp-stack-size", required_argument, NULL, 'f'},
2068: {"locals-stack-size", required_argument, NULL, 'l'},
2069: {"vm-commit", no_argument, &map_noreserve, 0},
2070: {"path", required_argument, NULL, 'p'},
2071: {"version", no_argument, NULL, 'v'},
2072: {"help", no_argument, NULL, 'h'},
2073: /* put something != 0 into offset_image */
2074: {"offset-image", no_argument, &offset_image, 1},
2075: {"no-offset-im", no_argument, &offset_image, 0},
2076: {"clear-dictionary", no_argument, &clear_dictionary, 1},
2077: {"die-on-signal", no_argument, &die_on_signal, 1},
2078: {"ignore-async-signals", no_argument, &ignore_async_signals, 1},
2079: {"debug", no_argument, &debug, 1},
2080: {"diag", no_argument, &diag, 1},
2081: {"no-super", no_argument, &no_super, 1},
2082: {"no-dynamic", no_argument, &no_dynamic, 1},
2083: {"dynamic", no_argument, &no_dynamic, 0},
2084: {"print-metrics", no_argument, &print_metrics, 1},
2085: {"print-sequences", no_argument, &print_sequences, 1},
2086: {"ss-number", required_argument, NULL, ss_number},
2087: {"ss-states", required_argument, NULL, ss_states},
2088: #ifndef NO_DYNAMIC
2089: {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
2090: #endif
2091: {"ss-min-ls", no_argument, NULL, ss_min_ls},
2092: {"ss-min-lsu", no_argument, NULL, ss_min_lsu},
2093: {"ss-min-nexts", no_argument, NULL, ss_min_nexts},
2094: {"ss-greedy", no_argument, &ss_greedy, 1},
2095: {"tpa-noequiv", no_argument, &tpa_noequiv, 1},
2096: {"tpa-noautomaton", no_argument, &tpa_noautomaton, 1},
2097: {"tpa-trace", no_argument, &tpa_trace, 1},
2098: {0,0,0,0}
2099: /* no-init-file, no-rc? */
2100: };
2101:
2102: c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
2103:
2104: switch (c) {
2105: case EOF: return;
2106: case '?': optind--; return;
2107: case 'a': *imagename = optarg; return;
2108: case 'i': *imagename = optarg; break;
2109: case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
2110: case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
2111: case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
2112: case 'f': fsize = convsize(optarg,sizeof(Float)); break;
2113: case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
2114: case 'p': *path = optarg; break;
2115: case 'o': offset_image = 1; break;
2116: case 'n': offset_image = 0; break;
2117: case 'c': clear_dictionary = 1; break;
2118: case 's': die_on_signal = 1; break;
2119: case 'x': debug = 1; break;
2120: case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
2121: case ss_number: static_super_number = atoi(optarg); break;
2122: case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
2123: #ifndef NO_DYNAMIC
2124: case ss_min_codesize: ss_cost = cost_codesize; break;
2125: #endif
2126: case ss_min_ls: ss_cost = cost_ls; break;
2127: case ss_min_lsu: ss_cost = cost_lsu; break;
2128: case ss_min_nexts: ss_cost = cost_nexts; break;
2129: case 'h':
2130: fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
2131: Engine Options:\n\
2132: --appl-image FILE Equivalent to '--image-file=FILE --'\n\
2133: --clear-dictionary Initialize the dictionary with 0 bytes\n\
2134: -d SIZE, --data-stack-size=SIZE Specify data stack size\n\
2135: --debug Print debugging information during startup\n\
2136: --diag Print diagnostic information during startup\n\
2137: --die-on-signal Exit instead of THROWing some signals\n\
2138: --dynamic Use dynamic native code\n\
2139: -f SIZE, --fp-stack-size=SIZE Specify floating point stack size\n\
2140: -h, --help Print this message and exit\n\
2141: --ignore-async-signals Ignore instead of THROWing async. signals\n\
2142: -i FILE, --image-file=FILE Use image FILE instead of `gforth.fi'\n\
2143: -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
2144: -m SIZE, --dictionary-size=SIZE Specify Forth dictionary size\n\
2145: --no-dynamic Use only statically compiled primitives\n\
2146: --no-offset-im Load image at normal position\n\
2147: --no-super No dynamically formed superinstructions\n\
2148: --offset-image Load image at a different position\n\
2149: -p PATH, --path=PATH Search path for finding image and sources\n\
2150: --print-metrics Print some code generation metrics on exit\n\
2151: --print-sequences Print primitive sequences for optimization\n\
2152: -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
2153: --ss-greedy Greedy, not optimal superinst selection\n\
2154: --ss-min-codesize Select superinsts for smallest native code\n\
2155: --ss-min-ls Minimize loads and stores\n\
2156: --ss-min-lsu Minimize loads, stores, and pointer updates\n\
2157: --ss-min-nexts Minimize the number of static superinsts\n\
2158: --ss-number=N Use N static superinsts (default max)\n\
2159: --ss-states=N N states for stack caching (default max)\n\
2160: --tpa-noequiv Automaton without state equivalence\n\
2161: --tpa-noautomaton Dynamic programming only\n\
2162: --tpa-trace Report new states etc.\n\
2163: -v, --version Print engine version and exit\n\
2164: --vm-commit Use OS default for memory overcommit\n\
2165: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
2166: `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
2167: argv[0]);
2168: optind--;
2169: return;
2170: }
2171: }
2172: }
2173: #endif
2174: #endif
2175:
2176: static void print_diag()
2177: {
2178:
2179: #if !defined(HAVE_GETRUSAGE) || (!defined(HAS_FFCALL) && !defined(HAS_LIBFFI))
2180: fprintf(stderr, "*** missing functionality ***\n"
2181: #ifndef HAVE_GETRUSAGE
2182: " no getrusage -> CPUTIME broken\n"
2183: #endif
2184: #if !defined(HAS_FFCALL) && !defined(HAS_LIBFFI)
2185: " no ffcall -> only old-style foreign function calls (no fflib.fs)\n"
2186: #endif
2187: );
2188: #endif
2189: if((relocs < nonrelocs) ||
2190: #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)
2191: 1
2192: #else
2193: 0
2194: #endif
2195: )
2196: debugp(stderr, "relocs: %d:%d\n", relocs, nonrelocs);
2197: fprintf(stderr, "*** %sperformance problems ***\n%s",
2198: #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(BUGGY_LONG_LONG)
2199: "",
2200: #else
2201: "no ",
2202: #endif
2203: #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)
2204: " double-cell integer type buggy ->\n "
2205: #ifdef BUGGY_LL_CMP
2206: "CMP, "
2207: #endif
2208: #ifdef BUGGY_LL_MUL
2209: "MUL, "
2210: #endif
2211: #ifdef BUGGY_LL_DIV
2212: "DIV, "
2213: #endif
2214: #ifdef BUGGY_LL_ADD
2215: "ADD, "
2216: #endif
2217: #ifdef BUGGY_LL_SHIFT
2218: "SHIFT, "
2219: #endif
2220: #ifdef BUGGY_LL_D2F
2221: "D2F, "
2222: #endif
2223: #ifdef BUGGY_LL_F2D
2224: "F2D, "
2225: #endif
2226: "\b\b slow\n"
2227: #endif
2228: #ifndef FORCE_REG
2229: " automatic register allocation: performance degradation possible\n"
2230: #endif
2231: #if !defined(FORCE_REG) || defined(BUGGY_LONG_LONG)
2232: "*** Suggested remedy: try ./configure"
2233: #ifndef FORCE_REG
2234: " --enable-force-reg"
2235: #endif
2236: #ifdef BUGGY_LONG_LONG
2237: " --enable-force-ll"
2238: #endif
2239: "\n"
2240: #else
2241: ""
2242: #endif
2243: ,
2244: (relocs < nonrelocs) ? " gcc PR 15242 -> no dynamic code generation (use gcc-2.95 instead)\n" : "");
2245: }
2246:
2247: #ifdef STANDALONE
2248: Cell data_abort_pc;
2249:
2250: void data_abort_C(void)
2251: {
2252: while(1) {
2253: }
2254: }
2255: #endif
2256:
2257: int main(int argc, char **argv, char **env)
2258: {
2259: #ifdef HAS_OS
2260: char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
2261: #else
2262: char *path = DEFAULTPATH;
2263: #endif
2264: #ifndef INCLUDE_IMAGE
2265: char *imagename="gforth.fi";
2266: FILE *image_file;
2267: Address image;
2268: #endif
2269: int retvalue;
2270:
2271: #if defined(i386) && defined(ALIGNMENT_CHECK)
2272: /* turn on alignment checks on the 486.
2273: * on the 386 this should have no effect. */
2274: __asm__("pushfl; popl %eax; orl $0x40000, %eax; pushl %eax; popfl;");
2275: /* this is unusable with Linux' libc.4.6.27, because this library is
2276: not alignment-clean; we would have to replace some library
2277: functions (e.g., memcpy) to make it work. Also GCC doesn't try to keep
2278: the stack FP-aligned. */
2279: #endif
2280:
2281: #ifndef STANDALONE
2282: /* buffering of the user output device */
2283: #ifdef _IONBF
2284: if (isatty(fileno(stdout))) {
2285: fflush(stdout);
2286: setvbuf(stdout,NULL,_IONBF,0);
2287: }
2288: #endif
2289: #else
2290: prep_terminal();
2291: #endif
2292:
2293: progname = argv[0];
2294:
2295: #ifndef STANDALONE
2296: #ifdef HAS_OS
2297: gforth_args(argc, argv, &path, &imagename);
2298: #ifndef NO_DYNAMIC
2299: init_ss_cost();
2300: #endif /* !defined(NO_DYNAMIC) */
2301: #endif /* defined(HAS_OS) */
2302: #endif
2303:
2304: #ifdef STANDALONE
2305: image = gforth_engine(0, 0, 0, 0, 0);
2306: alloc_stacks((ImageHeader *)image);
2307: #else
2308: image_file = open_image_file(imagename, path);
2309: image = gforth_loader(image_file, imagename);
2310: #endif
2311: gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
2312:
2313: if (diag)
2314: print_diag();
2315: {
2316: char path2[strlen(path)+1];
2317: char *p1, *p2;
2318: Cell environ[]= {
2319: (Cell)argc-(optind-1),
2320: (Cell)(argv+(optind-1)),
2321: (Cell)strlen(path),
2322: (Cell)path2};
2323: argv[optind-1] = progname;
2324: /*
2325: for (i=0; i<environ[0]; i++)
2326: printf("%s\n", ((char **)(environ[1]))[i]);
2327: */
2328: /* make path OS-independent by replacing path separators with NUL */
2329: for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
2330: if (*p1==PATHSEP)
2331: *p2 = '\0';
2332: else
2333: *p2 = *p1;
2334: *p2='\0';
2335: retvalue = gforth_go(image, 4, environ);
2336: #if defined(SIGPIPE) && !defined(STANDALONE)
2337: bsd_signal(SIGPIPE, SIG_IGN);
2338: #endif
2339: #ifdef VM_PROFILING
2340: vm_print_profile(stderr);
2341: #endif
2342: deprep_terminal();
2343: }
2344: if (print_metrics) {
2345: int i;
2346: fprintf(stderr, "code size = %8ld\n", dyncodesize());
2347: #ifndef STANDALONE
2348: for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
2349: fprintf(stderr, "metric %8s: %8ld\n",
2350: cost_sums[i].metricname, cost_sums[i].sum);
2351: #endif
2352: fprintf(stderr,"lb_basic_blocks = %ld\n", lb_basic_blocks);
2353: fprintf(stderr,"lb_labeler_steps = %ld\n", lb_labeler_steps);
2354: fprintf(stderr,"lb_labeler_automaton = %ld\n", lb_labeler_automaton);
2355: fprintf(stderr,"lb_labeler_dynprog = %ld\n", lb_labeler_dynprog);
2356: fprintf(stderr,"lb_newstate_equiv = %ld\n", lb_newstate_equiv);
2357: fprintf(stderr,"lb_newstate_new = %ld\n", lb_newstate_new);
2358: fprintf(stderr,"lb_applicable_base_rules = %ld\n", lb_applicable_base_rules);
2359: fprintf(stderr,"lb_applicable_chain_rules = %ld\n", lb_applicable_chain_rules);
2360: }
2361: if (tpa_trace) {
2362: fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
2363: fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
2364: }
2365: return retvalue;
2366: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>