File:
[gforth] /
gforth /
engine /
main.c
Revision
1.184:
download - view:
text,
annotated -
select for diffs
Sat Jun 30 20:28:55 2007 UTC (16 years, 9 months ago) by
anton
Branches:
MAIN
CVS tags:
HEAD
fixed minor bug in -falign-* configure tests
cleaned up dependencies on .h files in engine/Makefile.in
made Alpha port compile with gcc-2.95 (disabled longlong.h division)
optimized Alpha port:
use old division code
native code after a jump is now aligned. Speed effect:
sieve bubble matrix fib
1.950 1.605 1.259 1.653 b1 gcc-3.3.5
1.453 1.932 1.230 1.458 b1 gcc-3.3.5 with alignment
aligning branch targets would probably also help
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 relocs = 0;
224: static int nonrelocs = 0;
225:
226: #ifdef HAS_DEBUG
227: int debug=0;
228: # define debugp(x...) if (debug) fprintf(x);
229: #else
230: # define perror(x...)
231: # define fprintf(x...)
232: # define debugp(x...)
233: #endif
234:
235: ImageHeader *gforth_header;
236: Label *vm_prims;
237: #ifdef DOUBLY_INDIRECT
238: Label *xts; /* same content as vm_prims, but should only be used for xts */
239: #endif
240:
241: #ifndef NO_DYNAMIC
242: #define MAX_IMMARGS 2
243:
244: typedef struct {
245: Label start; /* NULL if not relocatable */
246: Cell length; /* only includes the jump iff superend is true*/
247: Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
248: char superend; /* true if primitive ends superinstruction, i.e.,
249: unconditional branch, execute, etc. */
250: Cell nimmargs;
251: struct immarg {
252: Cell offset; /* offset of immarg within prim */
253: char rel; /* true if immarg is relative */
254: } immargs[MAX_IMMARGS];
255: } PrimInfo;
256:
257: PrimInfo *priminfos;
258: PrimInfo **decomp_prims;
259:
260: const char const* const prim_names[]={
261: #include PRIM_NAMES_I
262: };
263:
264: void init_ss_cost(void);
265:
266: static int is_relocatable(int p)
267: {
268: return !no_dynamic && priminfos[p].start != NULL;
269: }
270: #else /* defined(NO_DYNAMIC) */
271: static int is_relocatable(int p)
272: {
273: return 0;
274: }
275: #endif /* defined(NO_DYNAMIC) */
276:
277: #ifdef MEMCMP_AS_SUBROUTINE
278: int gforth_memcmp(const char * s1, const char * s2, size_t n)
279: {
280: return memcmp(s1, s2, n);
281: }
282: #endif
283:
284: static Cell max(Cell a, Cell b)
285: {
286: return a>b?a:b;
287: }
288:
289: static Cell min(Cell a, Cell b)
290: {
291: return a<b?a:b;
292: }
293:
294: #ifndef STANDALONE
295: /* image file format:
296: * "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
297: * padding to a multiple of 8
298: * magic: "Gforth3x" means format 0.6,
299: * where x is a byte with
300: * bit 7: reserved = 0
301: * bit 6:5: address unit size 2^n octets
302: * bit 4:3: character size 2^n octets
303: * bit 2:1: cell size 2^n octets
304: * bit 0: endian, big=0, little=1.
305: * The magic are always 8 octets, no matter what the native AU/character size is
306: * padding to max alignment (no padding necessary on current machines)
307: * ImageHeader structure (see forth.h)
308: * data (size in ImageHeader.image_size)
309: * tags ((if relocatable, 1 bit/data cell)
310: *
311: * tag==1 means that the corresponding word is an address;
312: * If the word is >=0, the address is within the image;
313: * addresses within the image are given relative to the start of the image.
314: * If the word =-1 (CF_NIL), the address is NIL,
315: * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
316: * If the word =CF(DODOES), it's a DOES> CFA
317: * If the word =CF(DOESJUMP), it's a DOES JUMP (2 Cells after DOES>,
318: * possibly containing a jump to dodoes)
319: * If the word is <CF(DOESJUMP) and bit 14 is set, it's the xt of a primitive
320: * If the word is <CF(DOESJUMP) and bit 14 is clear,
321: * it's the threaded code of a primitive
322: * bits 13..9 of a primitive token state which group the primitive belongs to,
323: * bits 8..0 of a primitive token index into the group
324: */
325:
326: Cell groups[32] = {
327: 0,
328: 0
329: #undef GROUP
330: #undef GROUPADD
331: #define GROUPADD(n) +n
332: #define GROUP(x, n) , 0
333: #include PRIM_GRP_I
334: #undef GROUP
335: #undef GROUPADD
336: #define GROUP(x, n)
337: #define GROUPADD(n)
338: };
339:
340: static unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
341: int size, Cell base)
342: /* produce a bitmask marking all the branch targets */
343: {
344: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
345: Cell token;
346: unsigned char bits;
347: unsigned char *result=malloc(steps);
348:
349: memset(result, 0, steps);
350: for(k=0; k<steps; k++) {
351: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
352: if(bits & (1U << (RELINFOBITS-1))) {
353: assert(i*sizeof(Cell) < size);
354: token=image[i];
355: if (token>=base) { /* relocatable address */
356: UCell bitnum=(token-base)/sizeof(Cell);
357: if (bitnum/RELINFOBITS < (UCell)steps)
358: result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
359: }
360: }
361: }
362: }
363: return result;
364: }
365:
366: void gforth_relocate(Cell *image, const Char *bitstring,
367: UCell size, Cell base, Label symbols[])
368: {
369: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
370: Cell token;
371: char bits;
372: Cell max_symbols;
373: /*
374: * A virtual start address that's the real start address minus
375: * the one in the image
376: */
377: Cell *start = (Cell * ) (((void *) image) - ((void *) base));
378: unsigned char *targets = branch_targets(image, bitstring, size, base);
379:
380: /* group index into table */
381: if(groups[31]==0) {
382: int groupsum=0;
383: for(i=0; i<32; i++) {
384: groupsum += groups[i];
385: groups[i] = groupsum;
386: /* printf("group[%d]=%d\n",i,groupsum); */
387: }
388: i=0;
389: }
390:
391: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
392:
393: for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
394: ;
395: max_symbols--;
396:
397: for(k=0; k<steps; k++) {
398: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
399: /* fprintf(stderr,"relocate: image[%d]\n", i);*/
400: if(bits & (1U << (RELINFOBITS-1))) {
401: assert(i*sizeof(Cell) < size);
402: /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
403: token=image[i];
404: if(token<0) {
405: int group = (-token & 0x3E00) >> 9;
406: if(group == 0) {
407: switch(token|0x4000) {
408: case CF_NIL : image[i]=0; break;
409: #if !defined(DOUBLY_INDIRECT)
410: case CF(DOCOL) :
411: case CF(DOVAR) :
412: case CF(DOCON) :
413: case CF(DOUSER) :
414: case CF(DODEFER) :
415: case CF(DOFIELD) : MAKE_CF(image+i,symbols[CF(token)]); break;
416: case CF(DOESJUMP): image[i]=0; break;
417: #endif /* !defined(DOUBLY_INDIRECT) */
418: case CF(DODOES) :
419: MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
420: break;
421: default : /* backward compatibility */
422: /* printf("Code field generation image[%x]:=CFA(%x)\n",
423: i, CF(image[i])); */
424: if (CF((token | 0x4000))<max_symbols) {
425: image[i]=(Cell)CFA(CF(token));
426: #ifdef DIRECT_THREADED
427: if ((token & 0x4000) == 0) { /* threade code, no CFA */
428: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
429: compile_prim1(0);
430: compile_prim1(&image[i]);
431: }
432: #endif
433: } else
434: fprintf(stderr,"Primitive %ld used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n",(long)CF(token),(long)&image[i], i, PACKAGE_VERSION);
435: }
436: } else {
437: int tok = -token & 0x1FF;
438: if (tok < (groups[group+1]-groups[group])) {
439: #if defined(DOUBLY_INDIRECT)
440: image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
441: #else
442: image[i]=(Cell)CFA((groups[group]+tok));
443: #endif
444: #ifdef DIRECT_THREADED
445: if ((token & 0x4000) == 0) { /* threade code, no CFA */
446: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
447: compile_prim1(0);
448: compile_prim1(&image[i]);
449: }
450: #endif
451: } else
452: fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
453: }
454: } else {
455: /* if base is > 0: 0 is a null reference so don't adjust*/
456: if (token>=base) {
457: image[i]+=(Cell)start;
458: }
459: }
460: }
461: }
462: }
463: free(targets);
464: finish_code();
465: ((ImageHeader*)(image))->base = (Address) image;
466: }
467:
468: #ifndef DOUBLY_INDIRECT
469: static UCell checksum(Label symbols[])
470: {
471: UCell r=PRIM_VERSION;
472: Cell i;
473:
474: for (i=DOCOL; i<=DOESJUMP; i++) {
475: r ^= (UCell)(symbols[i]);
476: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
477: }
478: #ifdef DIRECT_THREADED
479: /* we have to consider all the primitives */
480: for (; symbols[i]!=(Label)0; i++) {
481: r ^= (UCell)(symbols[i]);
482: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
483: }
484: #else
485: /* in indirect threaded code all primitives are accessed through the
486: symbols table, so we just have to put the base address of symbols
487: in the checksum */
488: r ^= (UCell)symbols;
489: #endif
490: return r;
491: }
492: #endif
493:
494: static Address verbose_malloc(Cell size)
495: {
496: Address r;
497: /* leave a little room (64B) for stack underflows */
498: if ((r = malloc(size+64))==NULL) {
499: perror(progname);
500: exit(1);
501: }
502: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
503: debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
504: return r;
505: }
506:
507: static Address next_address=0;
508: static void after_alloc(Address r, Cell size)
509: {
510: if (r != (Address)-1) {
511: debugp(stderr, "success, address=$%lx\n", (long) r);
512: #if 0
513: /* not needed now that we protect the stacks with mprotect */
514: if (pagesize != 1)
515: next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
516: #endif
517: } else {
518: debugp(stderr, "failed: %s\n", strerror(errno));
519: }
520: }
521:
522: #ifndef MAP_FAILED
523: #define MAP_FAILED ((Address) -1)
524: #endif
525: #ifndef MAP_FILE
526: # define MAP_FILE 0
527: #endif
528: #ifndef MAP_PRIVATE
529: # define MAP_PRIVATE 0
530: #endif
531: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
532: # define MAP_ANON MAP_ANONYMOUS
533: #endif
534:
535: #if defined(HAVE_MMAP)
536: static Address alloc_mmap(Cell size)
537: {
538: Address r;
539:
540: #if defined(MAP_ANON)
541: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
542: r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE|map_noreserve, -1, 0);
543: #else /* !defined(MAP_ANON) */
544: /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
545: apparently defaults) */
546: static int dev_zero=-1;
547:
548: if (dev_zero == -1)
549: dev_zero = open("/dev/zero", O_RDONLY);
550: if (dev_zero == -1) {
551: r = MAP_FAILED;
552: debugp(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ",
553: strerror(errno));
554: } else {
555: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
556: r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE|map_noreserve, dev_zero, 0);
557: }
558: #endif /* !defined(MAP_ANON) */
559: after_alloc(r, size);
560: return r;
561: }
562:
563: static void page_noaccess(Address a)
564: {
565: /* try mprotect first; with munmap the page might be allocated later */
566: debugp(stderr, "try mprotect(%p,%ld,PROT_NONE); ", a, (long)pagesize);
567: if (mprotect(a, pagesize, PROT_NONE)==0) {
568: debugp(stderr, "ok\n");
569: return;
570: }
571: debugp(stderr, "failed: %s\n", strerror(errno));
572: debugp(stderr, "try munmap(%p,%ld); ", a, (long)pagesize);
573: if (munmap(a,pagesize)==0) {
574: debugp(stderr, "ok\n");
575: return;
576: }
577: debugp(stderr, "failed: %s\n", strerror(errno));
578: }
579:
580: static size_t wholepage(size_t n)
581: {
582: return (n+pagesize-1)&~(pagesize-1);
583: }
584: #endif
585:
586: Address gforth_alloc(Cell size)
587: {
588: #if HAVE_MMAP
589: Address r;
590:
591: r=alloc_mmap(size);
592: if (r!=(Address)MAP_FAILED)
593: return r;
594: #endif /* HAVE_MMAP */
595: /* use malloc as fallback */
596: return verbose_malloc(size);
597: }
598:
599: static Address dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
600: {
601: Address image = MAP_FAILED;
602:
603: #if defined(HAVE_MMAP)
604: if (offset==0) {
605: image=alloc_mmap(dictsize);
606: if (image != (Address)MAP_FAILED) {
607: Address image1;
608: debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
609: image1 = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE|map_noreserve, fileno(file), 0);
610: after_alloc(image1,dictsize);
611: if (image1 == (Address)MAP_FAILED)
612: goto read_image;
613: }
614: }
615: #endif /* defined(HAVE_MMAP) */
616: if (image == (Address)MAP_FAILED) {
617: image = gforth_alloc(dictsize+offset)+offset;
618: read_image:
619: rewind(file); /* fseek(imagefile,0L,SEEK_SET); */
620: fread(image, 1, imagesize, file);
621: }
622: return image;
623: }
624: #endif
625:
626: void set_stack_sizes(ImageHeader * header)
627: {
628: if (dictsize==0)
629: dictsize = header->dict_size;
630: if (dsize==0)
631: dsize = header->data_stack_size;
632: if (rsize==0)
633: rsize = header->return_stack_size;
634: if (fsize==0)
635: fsize = header->fp_stack_size;
636: if (lsize==0)
637: lsize = header->locals_stack_size;
638: dictsize=maxaligned(dictsize);
639: dsize=maxaligned(dsize);
640: rsize=maxaligned(rsize);
641: lsize=maxaligned(lsize);
642: fsize=maxaligned(fsize);
643: }
644:
645: #ifdef STANDALONE
646: void alloc_stacks(ImageHeader * h)
647: {
648: #define SSTACKSIZE 0x200
649: static Cell dstack[SSTACKSIZE+1];
650: static Cell rstack[SSTACKSIZE+1];
651:
652: h->dict_size=dictsize;
653: h->data_stack_size=dsize;
654: h->fp_stack_size=fsize;
655: h->return_stack_size=rsize;
656: h->locals_stack_size=lsize;
657:
658: h->data_stack_base=dstack+SSTACKSIZE;
659: // h->fp_stack_base=gforth_alloc(fsize);
660: h->return_stack_base=rstack+SSTACKSIZE;
661: // h->locals_stack_base=gforth_alloc(lsize);
662: }
663: #else
664: void alloc_stacks(ImageHeader * h)
665: {
666: h->dict_size=dictsize;
667: h->data_stack_size=dsize;
668: h->fp_stack_size=fsize;
669: h->return_stack_size=rsize;
670: h->locals_stack_size=lsize;
671:
672: #if defined(HAVE_MMAP) && !defined(STANDALONE)
673: if (pagesize > 1) {
674: size_t p = pagesize;
675: size_t totalsize =
676: wholepage(dsize)+wholepage(fsize)+wholepage(rsize)+wholepage(lsize)+5*p;
677: Address a = alloc_mmap(totalsize);
678: if (a != (Address)MAP_FAILED) {
679: page_noaccess(a); a+=p; h-> data_stack_base=a; a+=wholepage(dsize);
680: page_noaccess(a); a+=p; h-> fp_stack_base=a; a+=wholepage(fsize);
681: page_noaccess(a); a+=p; h->return_stack_base=a; a+=wholepage(rsize);
682: page_noaccess(a); a+=p; h->locals_stack_base=a; a+=wholepage(lsize);
683: page_noaccess(a);
684: debugp(stderr,"stack addresses: d=%p f=%p r=%p l=%p\n",
685: h->data_stack_base,
686: h->fp_stack_base,
687: h->return_stack_base,
688: h->locals_stack_base);
689: return;
690: }
691: }
692: #endif
693: h->data_stack_base=gforth_alloc(dsize);
694: h->fp_stack_base=gforth_alloc(fsize);
695: h->return_stack_base=gforth_alloc(rsize);
696: h->locals_stack_base=gforth_alloc(lsize);
697: }
698: #endif
699:
700: #warning You can ignore the warnings about clobbered variables in gforth_go
701: int gforth_go(Address image, int stack, Cell *entries)
702: {
703: volatile ImageHeader *image_header = (ImageHeader *)image;
704: Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
705: Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
706: Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
707: #ifdef GFORTH_DEBUGGING
708: volatile Cell *orig_rp0=rp0;
709: #endif
710: Address lp0=image_header->locals_stack_base + lsize;
711: Xt *ip0=(Xt *)(image_header->boot_entry);
712: #ifdef SYSSIGNALS
713: int throw_code;
714: #endif
715:
716: /* ensure that the cached elements (if any) are accessible */
717: #if !(defined(GFORTH_DEBUGGING) || defined(INDIRECT_THREADED) || defined(DOUBLY_INDIRECT) || defined(VM_PROFILING))
718: sp0 -= 8; /* make stuff below bottom accessible for stack caching */
719: #endif
720: IF_fpTOS(fp0--);
721:
722: for(;stack>0;stack--)
723: *--sp0=entries[stack-1];
724:
725: #if defined(SYSSIGNALS) && !defined(STANDALONE)
726: get_winsize();
727:
728: install_signal_handlers(); /* right place? */
729:
730: if ((throw_code=setjmp(throw_jmp_buf))) {
731: static Cell signal_data_stack[24];
732: static Cell signal_return_stack[16];
733: static Float signal_fp_stack[1];
734:
735: signal_data_stack[15]=throw_code;
736:
737: #ifdef GFORTH_DEBUGGING
738: debugp(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
739: throw_code, saved_ip, rp);
740: if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
741: /* no rstack overflow or underflow */
742: rp0 = rp;
743: *--rp0 = (Cell)saved_ip;
744: }
745: else /* I love non-syntactic ifdefs :-) */
746: rp0 = signal_return_stack+16;
747: #else /* !defined(GFORTH_DEBUGGING) */
748: debugp(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
749: rp0 = signal_return_stack+16;
750: #endif /* !defined(GFORTH_DEBUGGING) */
751: /* fprintf(stderr, "rp=$%x\n",rp0);*/
752:
753: return((int)(Cell)gforth_engine(image_header->throw_entry, signal_data_stack+15,
754: rp0, signal_fp_stack, 0));
755: }
756: #endif
757:
758: return((int)(Cell)gforth_engine(ip0,sp0,rp0,fp0,lp0));
759: }
760:
761: #if !defined(INCLUDE_IMAGE) && !defined(STANDALONE)
762: static void print_sizes(Cell sizebyte)
763: /* print size information */
764: {
765: static char* endianstring[]= { " big","little" };
766:
767: fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
768: endianstring[sizebyte & 1],
769: 1 << ((sizebyte >> 1) & 3),
770: 1 << ((sizebyte >> 3) & 3),
771: 1 << ((sizebyte >> 5) & 3));
772: }
773:
774: /* static superinstruction stuff */
775:
776: struct cost { /* super_info might be a more accurate name */
777: char loads; /* number of stack loads */
778: char stores; /* number of stack stores */
779: char updates; /* number of stack pointer updates */
780: char branch; /* is it a branch (SET_IP) */
781: unsigned char state_in; /* state on entry */
782: unsigned char state_out; /* state on exit */
783: unsigned char imm_ops; /* number of immediate operands */
784: short offset; /* offset into super2 table */
785: unsigned char length; /* number of components */
786: };
787:
788: PrimNum super2[] = {
789: #include SUPER2_I
790: };
791:
792: struct cost super_costs[] = {
793: #include COSTS_I
794: };
795:
796: struct super_state {
797: struct super_state *next;
798: PrimNum super;
799: };
800:
801: #define HASH_SIZE 256
802:
803: struct super_table_entry {
804: struct super_table_entry *next;
805: PrimNum *start;
806: short length;
807: struct super_state *ss_list; /* list of supers */
808: } *super_table[HASH_SIZE];
809: int max_super=2;
810:
811: struct super_state *state_transitions=NULL;
812:
813: static int hash_super(PrimNum *start, int length)
814: {
815: int i, r;
816:
817: for (i=0, r=0; i<length; i++) {
818: r <<= 1;
819: r += start[i];
820: }
821: return r & (HASH_SIZE-1);
822: }
823:
824: static struct super_state **lookup_super(PrimNum *start, int length)
825: {
826: int hash=hash_super(start,length);
827: struct super_table_entry *p = super_table[hash];
828:
829: /* assert(length >= 2); */
830: for (; p!=NULL; p = p->next) {
831: if (length == p->length &&
832: memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
833: return &(p->ss_list);
834: }
835: return NULL;
836: }
837:
838: static void prepare_super_table()
839: {
840: int i;
841: int nsupers = 0;
842:
843: for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
844: struct cost *c = &super_costs[i];
845: if ((c->length < 2 || nsupers < static_super_number) &&
846: c->state_in < maxstates && c->state_out < maxstates) {
847: struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
848: struct super_state *ss = malloc(sizeof(struct super_state));
849: ss->super= i;
850: if (c->offset==N_noop && i != N_noop) {
851: if (is_relocatable(i)) {
852: ss->next = state_transitions;
853: state_transitions = ss;
854: }
855: } else if (ss_listp != NULL) {
856: ss->next = *ss_listp;
857: *ss_listp = ss;
858: } else {
859: int hash = hash_super(super2+c->offset, c->length);
860: struct super_table_entry **p = &super_table[hash];
861: struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
862: ss->next = NULL;
863: e->next = *p;
864: e->start = super2 + c->offset;
865: e->length = c->length;
866: e->ss_list = ss;
867: *p = e;
868: }
869: if (c->length > max_super)
870: max_super = c->length;
871: if (c->length >= 2)
872: nsupers++;
873: }
874: }
875: debugp(stderr, "Using %d static superinsts\n", nsupers);
876: }
877:
878: /* dynamic replication/superinstruction stuff */
879:
880: #ifndef NO_DYNAMIC
881: static int compare_priminfo_length(const void *_a, const void *_b)
882: {
883: PrimInfo **a = (PrimInfo **)_a;
884: PrimInfo **b = (PrimInfo **)_b;
885: Cell diff = (*a)->length - (*b)->length;
886: if (diff)
887: return diff;
888: else /* break ties by start address; thus the decompiler produces
889: the earliest primitive with the same code (e.g. noop instead
890: of (char) and @ instead of >code-address */
891: return (*b)->start - (*a)->start;
892: }
893: #endif /* !defined(NO_DYNAMIC) */
894:
895: static char MAYBE_UNUSED superend[]={
896: #include PRIM_SUPEREND_I
897: };
898:
899: Cell npriminfos=0;
900:
901: Label goto_start;
902: Cell goto_len;
903:
904: #ifndef NO_DYNAMIC
905: static int compare_labels(const void *pa, const void *pb)
906: {
907: Label a = *(Label *)pa;
908: Label b = *(Label *)pb;
909: return a-b;
910: }
911: #endif
912:
913: static Label bsearch_next(Label key, Label *a, UCell n)
914: /* a is sorted; return the label >=key that is the closest in a;
915: return NULL if there is no label in a >=key */
916: {
917: int mid = (n-1)/2;
918: if (n<1)
919: return NULL;
920: if (n == 1) {
921: if (a[0] < key)
922: return NULL;
923: else
924: return a[0];
925: }
926: if (a[mid] < key)
927: return bsearch_next(key, a+mid+1, n-mid-1);
928: else
929: return bsearch_next(key, a, mid+1);
930: }
931:
932: static void check_prims(Label symbols1[])
933: {
934: int i;
935: #ifndef NO_DYNAMIC
936: Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted, *goto_p;
937: int nends1j;
938: #endif
939:
940: if (debug)
941: #ifdef __VERSION__
942: fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
943: #else
944: #define xstr(s) str(s)
945: #define str(s) #s
946: fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n");
947: #endif
948: for (i=0; symbols1[i]!=0; i++)
949: ;
950: npriminfos = i;
951:
952: #ifndef NO_DYNAMIC
953: if (no_dynamic)
954: return;
955: symbols2=gforth_engine2(0,0,0,0,0);
956: #if NO_IP
957: symbols3=gforth_engine3(0,0,0,0,0);
958: #else
959: symbols3=symbols1;
960: #endif
961: ends1 = symbols1+i+1;
962: ends1j = ends1+i;
963: goto_p = ends1j+i+1; /* goto_p[0]==before; ...[1]==after;*/
964: nends1j = i+1;
965: ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
966: memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
967: qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
968:
969: /* check whether the "goto *" is relocatable */
970: goto_len = goto_p[1]-goto_p[0];
971: debugp(stderr, "goto * %p %p len=%ld\n",
972: goto_p[0],symbols2[goto_p-symbols1],goto_len);
973: if (memcmp(goto_p[0],symbols2[goto_p-symbols1],goto_len)!=0) { /* unequal */
974: no_dynamic=1;
975: debugp(stderr," not relocatable, disabling dynamic code generation\n");
976: init_ss_cost();
977: return;
978: }
979: goto_start = goto_p[0];
980:
981: priminfos = calloc(i,sizeof(PrimInfo));
982: for (i=0; symbols1[i]!=0; i++) {
983: int prim_len = ends1[i]-symbols1[i];
984: PrimInfo *pi=&priminfos[i];
985: struct cost *sc=&super_costs[i];
986: int j=0;
987: char *s1 = (char *)symbols1[i];
988: char *s2 = (char *)symbols2[i];
989: char *s3 = (char *)symbols3[i];
990: Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
991:
992: pi->start = s1;
993: pi->superend = superend[i]|no_super;
994: pi->length = prim_len;
995: pi->restlength = endlabel - symbols1[i] - pi->length;
996: pi->nimmargs = 0;
997: relocs++;
998: debugp(stderr, "%-15s %d-%d %4d %p %p %p len=%3ld rest=%2ld send=%1d",
999: prim_names[i], sc->state_in, sc->state_out,
1000: i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength),
1001: pi->superend);
1002: if (endlabel == NULL) {
1003: pi->start = NULL; /* not relocatable */
1004: if (pi->length<0) pi->length=100;
1005: debugp(stderr,"\n non_reloc: no J label > start found\n");
1006: relocs--;
1007: nonrelocs++;
1008: continue;
1009: }
1010: if (ends1[i] > endlabel && !pi->superend) {
1011: pi->start = NULL; /* not relocatable */
1012: pi->length = endlabel-symbols1[i];
1013: debugp(stderr,"\n non_reloc: there is a J label before the K label (restlength<0)\n");
1014: relocs--;
1015: nonrelocs++;
1016: continue;
1017: }
1018: if (ends1[i] < pi->start && !pi->superend) {
1019: pi->start = NULL; /* not relocatable */
1020: pi->length = endlabel-symbols1[i];
1021: debugp(stderr,"\n non_reloc: K label before I label (length<0)\n");
1022: relocs--;
1023: nonrelocs++;
1024: continue;
1025: }
1026: assert(pi->length>=0);
1027: assert(pi->restlength >=0);
1028: while (j<(pi->length+pi->restlength)) {
1029: if (s1[j]==s3[j]) {
1030: if (s1[j] != s2[j]) {
1031: pi->start = NULL; /* not relocatable */
1032: debugp(stderr,"\n non_reloc: engine1!=engine2 offset %3d",j);
1033: /* assert(j<prim_len); */
1034: relocs--;
1035: nonrelocs++;
1036: break;
1037: }
1038: j++;
1039: } else {
1040: struct immarg *ia=&pi->immargs[pi->nimmargs];
1041:
1042: pi->nimmargs++;
1043: ia->offset=j;
1044: if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
1045: ia->rel=0;
1046: debugp(stderr,"\n absolute immarg: offset %3d",j);
1047: } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
1048: symbols1[DOESJUMP+1]) {
1049: ia->rel=1;
1050: debugp(stderr,"\n relative immarg: offset %3d",j);
1051: } else {
1052: pi->start = NULL; /* not relocatable */
1053: debugp(stderr,"\n non_reloc: engine1!=engine3 offset %3d",j);
1054: /* assert(j<prim_len);*/
1055: relocs--;
1056: nonrelocs++;
1057: break;
1058: }
1059: j+=4;
1060: }
1061: }
1062: debugp(stderr,"\n");
1063: }
1064: decomp_prims = calloc(i,sizeof(PrimInfo *));
1065: for (i=DOESJUMP+1; i<npriminfos; i++)
1066: decomp_prims[i] = &(priminfos[i]);
1067: qsort(decomp_prims+DOESJUMP+1, npriminfos-DOESJUMP-1, sizeof(PrimInfo *),
1068: compare_priminfo_length);
1069: #endif
1070: }
1071:
1072: static void flush_to_here(void)
1073: {
1074: #ifndef NO_DYNAMIC
1075: if (start_flush)
1076: FLUSH_ICACHE(start_flush, code_here-start_flush);
1077: start_flush=code_here;
1078: #endif
1079: }
1080:
1081: #ifndef NO_DYNAMIC
1082: static void append_jump(void)
1083: {
1084: if (last_jump) {
1085: PrimInfo *pi = &priminfos[last_jump];
1086:
1087: memcpy(code_here, pi->start+pi->length, pi->restlength);
1088: code_here += pi->restlength;
1089: memcpy(code_here, goto_start, goto_len);
1090: code_here += goto_len;
1091: #if defined(__alpha)
1092: {
1093: /* align code after jump */
1094: Cell alignment = 16;
1095: code_here = (Address)((((UCell)code_here)+(alignment-1))&~(alignment-1));
1096: }
1097: #endif
1098: last_jump=0;
1099: }
1100: }
1101:
1102: /* Gforth remembers all code blocks in this list. On forgetting (by
1103: executing a marker) the code blocks are not freed (because Gforth does
1104: not remember how they were allocated; hmm, remembering that might be
1105: easier and cleaner). Instead, code_here etc. are reset to the old
1106: value, and the "forgotten" code blocks are reused when they are
1107: needed. */
1108:
1109: struct code_block_list {
1110: struct code_block_list *next;
1111: Address block;
1112: Cell size;
1113: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
1114:
1115: static Address append_prim(Cell p)
1116: {
1117: PrimInfo *pi = &priminfos[p];
1118: Address old_code_here = code_here;
1119:
1120: if (code_area+code_area_size < code_here+pi->length+pi->restlength+goto_len) {
1121: struct code_block_list *p;
1122: append_jump();
1123: flush_to_here();
1124: if (*next_code_blockp == NULL) {
1125: code_here = start_flush = code_area = gforth_alloc(code_area_size);
1126: p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
1127: *next_code_blockp = p;
1128: p->next = NULL;
1129: p->block = code_here;
1130: p->size = code_area_size;
1131: } else {
1132: p = *next_code_blockp;
1133: code_here = start_flush = code_area = p->block;
1134: }
1135: old_code_here = code_here;
1136: next_code_blockp = &(p->next);
1137: }
1138: memcpy(code_here, pi->start, pi->length);
1139: code_here += pi->length;
1140: return old_code_here;
1141: }
1142: #endif
1143:
1144: int forget_dyncode(Address code)
1145: {
1146: #ifdef NO_DYNAMIC
1147: return -1;
1148: #else
1149: struct code_block_list *p, **pp;
1150:
1151: for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
1152: if (code >= p->block && code < p->block+p->size) {
1153: next_code_blockp = &(p->next);
1154: code_here = start_flush = code;
1155: code_area = p->block;
1156: last_jump = 0;
1157: return -1;
1158: }
1159: }
1160: return -no_dynamic;
1161: #endif /* !defined(NO_DYNAMIC) */
1162: }
1163:
1164: static long dyncodesize(void)
1165: {
1166: #ifndef NO_DYNAMIC
1167: struct code_block_list *p;
1168: long size=0;
1169: for (p=code_block_list; p!=NULL; p=p->next) {
1170: if (code_here >= p->block && code_here < p->block+p->size)
1171: return size + (code_here - p->block);
1172: else
1173: size += p->size;
1174: }
1175: #endif /* !defined(NO_DYNAMIC) */
1176: return 0;
1177: }
1178:
1179: Label decompile_code(Label _code)
1180: {
1181: #ifdef NO_DYNAMIC
1182: return _code;
1183: #else /* !defined(NO_DYNAMIC) */
1184: Cell i;
1185: struct code_block_list *p;
1186: Address code=_code;
1187:
1188: /* first, check if we are in code at all */
1189: for (p = code_block_list;; p = p->next) {
1190: if (p == NULL)
1191: return code;
1192: if (code >= p->block && code < p->block+p->size)
1193: break;
1194: }
1195: /* reverse order because NOOP might match other prims */
1196: for (i=npriminfos-1; i>DOESJUMP; i--) {
1197: PrimInfo *pi=decomp_prims[i];
1198: if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1199: return vm_prims[super2[super_costs[pi-priminfos].offset]];
1200: /* return pi->start;*/
1201: }
1202: return code;
1203: #endif /* !defined(NO_DYNAMIC) */
1204: }
1205:
1206: #ifdef NO_IP
1207: int nbranchinfos=0;
1208:
1209: struct branchinfo {
1210: Label **targetpp; /* **(bi->targetpp) is the target */
1211: Cell *addressptr; /* store the target here */
1212: } branchinfos[100000];
1213:
1214: int ndoesexecinfos=0;
1215: struct doesexecinfo {
1216: int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1217: Label *targetp; /*target for branch (because this is not in threaded code)*/
1218: Cell *xt; /* cfa of word whose does-code needs calling */
1219: } doesexecinfos[10000];
1220:
1221: static void set_rel_target(Cell *source, Label target)
1222: {
1223: *source = ((Cell)target)-(((Cell)source)+4);
1224: }
1225:
1226: static void register_branchinfo(Label source, Cell *targetpp)
1227: {
1228: struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1229: bi->targetpp = (Label **)targetpp;
1230: bi->addressptr = (Cell *)source;
1231: nbranchinfos++;
1232: }
1233:
1234: static Address compile_prim1arg(PrimNum p, Cell **argp)
1235: {
1236: Address old_code_here=append_prim(p);
1237:
1238: assert(vm_prims[p]==priminfos[p].start);
1239: *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
1240: return old_code_here;
1241: }
1242:
1243: static Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
1244: {
1245: PrimInfo *pi = &priminfos[N_call2];
1246: Address old_code_here = append_prim(N_call2);
1247:
1248: *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
1249: register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
1250: return old_code_here;
1251: }
1252: #endif
1253:
1254: void finish_code(void)
1255: {
1256: #ifdef NO_IP
1257: Cell i;
1258:
1259: compile_prim1(NULL);
1260: for (i=0; i<ndoesexecinfos; i++) {
1261: struct doesexecinfo *dei = &doesexecinfos[i];
1262: dei->targetp = (Label *)DOES_CODE1((dei->xt));
1263: branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
1264: }
1265: ndoesexecinfos = 0;
1266: for (i=0; i<nbranchinfos; i++) {
1267: struct branchinfo *bi=&branchinfos[i];
1268: set_rel_target(bi->addressptr, **(bi->targetpp));
1269: }
1270: nbranchinfos = 0;
1271: #else
1272: compile_prim1(NULL);
1273: #endif
1274: flush_to_here();
1275: }
1276:
1277: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1278: #ifdef NO_IP
1279: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1280: /* compile prim #p dynamically (mod flags etc.) and return start
1281: address of generated code for putting it into the threaded
1282: code. This function is only called if all the associated
1283: inline arguments of p are already in place (at tcp[1] etc.) */
1284: {
1285: PrimInfo *pi=&priminfos[p];
1286: Cell *next_code_target=NULL;
1287: Address codeaddr;
1288: Address primstart;
1289:
1290: assert(p<npriminfos);
1291: if (p==N_execute || p==N_perform || p==N_lit_perform) {
1292: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1293: primstart = append_prim(p);
1294: goto other_prim;
1295: } else if (p==N_call) {
1296: codeaddr = compile_call2(tcp+1, &next_code_target);
1297: } else if (p==N_does_exec) {
1298: struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1299: Cell *arg;
1300: codeaddr = compile_prim1arg(N_lit,&arg);
1301: *arg = (Cell)PFA(tcp[1]);
1302: /* we cannot determine the callee now (last_start[1] may be a
1303: forward reference), so just register an arbitrary target, and
1304: register in dei that we need to fix this before resolving
1305: branches */
1306: dei->branchinfo = nbranchinfos;
1307: dei->xt = (Cell *)(tcp[1]);
1308: compile_call2(0, &next_code_target);
1309: } else if (!is_relocatable(p)) {
1310: Cell *branch_target;
1311: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1312: compile_prim1arg(N_branch,&branch_target);
1313: set_rel_target(branch_target,vm_prims[p]);
1314: } else {
1315: unsigned j;
1316:
1317: codeaddr = primstart = append_prim(p);
1318: other_prim:
1319: for (j=0; j<pi->nimmargs; j++) {
1320: struct immarg *ia = &(pi->immargs[j]);
1321: Cell *argp = tcp + pi->nimmargs - j;
1322: Cell argval = *argp; /* !! specific to prims */
1323: if (ia->rel) { /* !! assumption: relative refs are branches */
1324: register_branchinfo(primstart + ia->offset, argp);
1325: } else /* plain argument */
1326: *(Cell *)(primstart + ia->offset) = argval;
1327: }
1328: }
1329: if (next_code_target!=NULL)
1330: *next_code_target = (Cell)code_here;
1331: return (Cell)codeaddr;
1332: }
1333: #else /* !defined(NO_IP) */
1334: static Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1335: /* compile prim #p dynamically (mod flags etc.) and return start
1336: address of generated code for putting it into the threaded code */
1337: {
1338: Cell static_prim = (Cell)vm_prims[p];
1339: #if defined(NO_DYNAMIC)
1340: return static_prim;
1341: #else /* !defined(NO_DYNAMIC) */
1342: Address old_code_here;
1343:
1344: if (no_dynamic)
1345: return static_prim;
1346: if (p>=npriminfos || !is_relocatable(p)) {
1347: append_jump();
1348: return static_prim;
1349: }
1350: old_code_here = append_prim(p);
1351: last_jump = p;
1352: if (priminfos[p].superend)
1353: append_jump();
1354: return (Cell)old_code_here;
1355: #endif /* !defined(NO_DYNAMIC) */
1356: }
1357: #endif /* !defined(NO_IP) */
1358: #endif
1359:
1360: #ifndef NO_DYNAMIC
1361: static int cost_codesize(int prim)
1362: {
1363: return priminfos[prim].length;
1364: }
1365: #endif
1366:
1367: static int cost_ls(int prim)
1368: {
1369: struct cost *c = super_costs+prim;
1370:
1371: return c->loads + c->stores;
1372: }
1373:
1374: static int cost_lsu(int prim)
1375: {
1376: struct cost *c = super_costs+prim;
1377:
1378: return c->loads + c->stores + c->updates;
1379: }
1380:
1381: static int cost_nexts(int prim)
1382: {
1383: return 1;
1384: }
1385:
1386: typedef int Costfunc(int);
1387: Costfunc *ss_cost = /* cost function for optimize_bb */
1388: #ifdef NO_DYNAMIC
1389: cost_lsu;
1390: #else
1391: cost_codesize;
1392: #endif
1393:
1394: struct {
1395: Costfunc *costfunc;
1396: char *metricname;
1397: long sum;
1398: } cost_sums[] = {
1399: #ifndef NO_DYNAMIC
1400: { cost_codesize, "codesize", 0 },
1401: #endif
1402: { cost_ls, "ls", 0 },
1403: { cost_lsu, "lsu", 0 },
1404: { cost_nexts, "nexts", 0 }
1405: };
1406:
1407: #ifndef NO_DYNAMIC
1408: void init_ss_cost(void) {
1409: if (no_dynamic && ss_cost == cost_codesize) {
1410: ss_cost = cost_nexts;
1411: cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
1412: debugp(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
1413: }
1414: }
1415: #endif
1416:
1417: #define MAX_BB 128 /* maximum number of instructions in BB */
1418: #define INF_COST 1000000 /* infinite cost */
1419: #define CANONICAL_STATE 0
1420:
1421: struct waypoint {
1422: int cost; /* the cost from here to the end */
1423: PrimNum inst; /* the inst used from here to the next waypoint */
1424: char relocatable; /* the last non-transition was relocatable */
1425: char no_transition; /* don't use the next transition (relocatability)
1426: * or this transition (does not change state) */
1427: };
1428:
1429: struct tpa_state { /* tree parsing automaton (like) state */
1430: /* labeling is back-to-front */
1431: struct waypoint *inst; /* in front of instruction */
1432: struct waypoint *trans; /* in front of instruction and transition */
1433: };
1434:
1435: struct tpa_state *termstate = NULL; /* initialized in loader() */
1436:
1437: /* statistics about tree parsing (lazyburg) stuff */
1438: long lb_basic_blocks = 0;
1439: long lb_labeler_steps = 0;
1440: long lb_labeler_automaton = 0;
1441: long lb_labeler_dynprog = 0;
1442: long lb_newstate_equiv = 0;
1443: long lb_newstate_new = 0;
1444: long lb_applicable_base_rules = 0;
1445: long lb_applicable_chain_rules = 0;
1446:
1447: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1448: static void init_waypoints(struct waypoint ws[])
1449: {
1450: int k;
1451:
1452: for (k=0; k<maxstates; k++)
1453: ws[k].cost=INF_COST;
1454: }
1455:
1456: static struct tpa_state *empty_tpa_state()
1457: {
1458: struct tpa_state *s = malloc(sizeof(struct tpa_state));
1459:
1460: s->inst = calloc(maxstates,sizeof(struct waypoint));
1461: init_waypoints(s->inst);
1462: s->trans = calloc(maxstates,sizeof(struct waypoint));
1463: /* init_waypoints(s->trans);*/
1464: return s;
1465: }
1466:
1467: static void transitions(struct tpa_state *t)
1468: {
1469: int k;
1470: struct super_state *l;
1471:
1472: for (k=0; k<maxstates; k++) {
1473: t->trans[k] = t->inst[k];
1474: t->trans[k].no_transition = 1;
1475: }
1476: for (l = state_transitions; l != NULL; l = l->next) {
1477: PrimNum s = l->super;
1478: int jcost;
1479: struct cost *c=super_costs+s;
1480: struct waypoint *wi=&(t->trans[c->state_in]);
1481: struct waypoint *wo=&(t->inst[c->state_out]);
1482: lb_applicable_chain_rules++;
1483: if (wo->cost == INF_COST)
1484: continue;
1485: jcost = wo->cost + ss_cost(s);
1486: if (jcost <= wi->cost) {
1487: wi->cost = jcost;
1488: wi->inst = s;
1489: wi->relocatable = wo->relocatable;
1490: wi->no_transition = 0;
1491: /* if (ss_greedy) wi->cost = wo->cost ? */
1492: }
1493: }
1494: }
1495:
1496: static struct tpa_state *make_termstate()
1497: {
1498: struct tpa_state *s = empty_tpa_state();
1499:
1500: s->inst[CANONICAL_STATE].cost = 0;
1501: transitions(s);
1502: return s;
1503: }
1504: #endif
1505:
1506: #define TPA_SIZE 16384
1507:
1508: struct tpa_entry {
1509: struct tpa_entry *next;
1510: PrimNum inst;
1511: struct tpa_state *state_behind; /* note: brack-to-front labeling */
1512: struct tpa_state *state_infront; /* note: brack-to-front labeling */
1513: } *tpa_table[TPA_SIZE];
1514:
1515: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1516: static Cell hash_tpa(PrimNum p, struct tpa_state *t)
1517: {
1518: UCell it = (UCell )t;
1519: return (p+it+(it>>14))&(TPA_SIZE-1);
1520: }
1521:
1522: static struct tpa_state **lookup_tpa(PrimNum p, struct tpa_state *t2)
1523: {
1524: int hash=hash_tpa(p, t2);
1525: struct tpa_entry *te = tpa_table[hash];
1526:
1527: if (tpa_noautomaton) {
1528: static struct tpa_state *t;
1529: t = NULL;
1530: return &t;
1531: }
1532: for (; te!=NULL; te = te->next) {
1533: if (p == te->inst && t2 == te->state_behind)
1534: return &(te->state_infront);
1535: }
1536: te = (struct tpa_entry *)malloc(sizeof(struct tpa_entry));
1537: te->next = tpa_table[hash];
1538: te->inst = p;
1539: te->state_behind = t2;
1540: te->state_infront = NULL;
1541: tpa_table[hash] = te;
1542: return &(te->state_infront);
1543: }
1544:
1545: static void tpa_state_normalize(struct tpa_state *t)
1546: {
1547: /* normalize so cost of canonical state=0; this may result in
1548: negative states for some states */
1549: int d = t->inst[CANONICAL_STATE].cost;
1550: int i;
1551:
1552: for (i=0; i<maxstates; i++) {
1553: if (t->inst[i].cost != INF_COST)
1554: t->inst[i].cost -= d;
1555: if (t->trans[i].cost != INF_COST)
1556: t->trans[i].cost -= d;
1557: }
1558: }
1559:
1560: static int tpa_state_equivalent(struct tpa_state *t1, struct tpa_state *t2)
1561: {
1562: return (memcmp(t1->inst, t2->inst, maxstates*sizeof(struct waypoint)) == 0 &&
1563: memcmp(t1->trans,t2->trans,maxstates*sizeof(struct waypoint)) == 0);
1564: }
1565: #endif
1566:
1567: struct tpa_state_entry {
1568: struct tpa_state_entry *next;
1569: struct tpa_state *state;
1570: } *tpa_state_table[TPA_SIZE];
1571:
1572: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1573: static Cell hash_tpa_state(struct tpa_state *t)
1574: {
1575: int *ti = (int *)(t->inst);
1576: int *tt = (int *)(t->trans);
1577: int r=0;
1578: int i;
1579:
1580: for (i=0; ti+i < (int *)(t->inst+maxstates); i++)
1581: r += ti[i]+tt[i];
1582: return (r+(r>>14)+(r>>22)) & (TPA_SIZE-1);
1583: }
1584:
1585: static struct tpa_state *lookup_tpa_state(struct tpa_state *t)
1586: {
1587: Cell hash = hash_tpa_state(t);
1588: struct tpa_state_entry *te = tpa_state_table[hash];
1589: struct tpa_state_entry *tn;
1590:
1591: if (!tpa_noequiv) {
1592: for (; te!=NULL; te = te->next) {
1593: if (tpa_state_equivalent(t, te->state)) {
1594: lb_newstate_equiv++;
1595: free(t->inst);
1596: free(t->trans);
1597: free(t);
1598: return te->state;
1599: }
1600: }
1601: tn = (struct tpa_state_entry *)malloc(sizeof(struct tpa_state_entry));
1602: tn->next = te;
1603: tn->state = t;
1604: tpa_state_table[hash] = tn;
1605: }
1606: lb_newstate_new++;
1607: if (tpa_trace)
1608: fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
1609: return t;
1610: }
1611:
1612: /* use dynamic programming to find the shortest paths within the basic
1613: block origs[0..ninsts-1] and rewrite the instructions pointed to by
1614: instps to use it */
1615: static void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
1616: {
1617: int i,j;
1618: struct tpa_state *ts[ninsts+1];
1619: int nextdyn, nextstate, no_transition;
1620:
1621: lb_basic_blocks++;
1622: ts[ninsts] = termstate;
1623: for (i=ninsts-1; i>=0; i--) {
1624: struct tpa_state **tp = lookup_tpa(origs[i],ts[i+1]);
1625: struct tpa_state *t = *tp;
1626: lb_labeler_steps++;
1627: if (t) {
1628: ts[i] = t;
1629: lb_labeler_automaton++;
1630: }
1631: else {
1632: lb_labeler_dynprog++;
1633: ts[i] = empty_tpa_state();
1634: for (j=1; j<=max_super && i+j<=ninsts; j++) {
1635: struct super_state **superp = lookup_super(origs+i, j);
1636: if (superp!=NULL) {
1637: struct super_state *supers = *superp;
1638: for (; supers!=NULL; supers = supers->next) {
1639: PrimNum s = supers->super;
1640: int jcost;
1641: struct cost *c=super_costs+s;
1642: struct waypoint *wi=&(ts[i]->inst[c->state_in]);
1643: struct waypoint *wo=&(ts[i+j]->trans[c->state_out]);
1644: int no_transition = wo->no_transition;
1645: lb_applicable_base_rules++;
1646: if (!(is_relocatable(s)) && !wo->relocatable) {
1647: wo=&(ts[i+j]->inst[c->state_out]);
1648: no_transition=1;
1649: }
1650: if (wo->cost == INF_COST)
1651: continue;
1652: jcost = wo->cost + ss_cost(s);
1653: if (jcost <= wi->cost) {
1654: wi->cost = jcost;
1655: wi->inst = s;
1656: wi->relocatable = is_relocatable(s);
1657: wi->no_transition = no_transition;
1658: /* if (ss_greedy) wi->cost = wo->cost ? */
1659: }
1660: }
1661: }
1662: }
1663: transitions(ts[i]);
1664: tpa_state_normalize(ts[i]);
1665: *tp = ts[i] = lookup_tpa_state(ts[i]);
1666: if (tpa_trace)
1667: fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
1668: }
1669: }
1670: /* now rewrite the instructions */
1671: nextdyn=0;
1672: nextstate=CANONICAL_STATE;
1673: no_transition = ((!ts[0]->trans[nextstate].relocatable)
1674: ||ts[0]->trans[nextstate].no_transition);
1675: for (i=0; i<ninsts; i++) {
1676: Cell tc=0, tc2;
1677: if (i==nextdyn) {
1678: if (!no_transition) {
1679: /* process trans */
1680: PrimNum p = ts[i]->trans[nextstate].inst;
1681: struct cost *c = super_costs+p;
1682: assert(ts[i]->trans[nextstate].cost != INF_COST);
1683: assert(c->state_in==nextstate);
1684: tc = compile_prim_dyn(p,NULL);
1685: nextstate = c->state_out;
1686: }
1687: {
1688: /* process inst */
1689: PrimNum p = ts[i]->inst[nextstate].inst;
1690: struct cost *c=super_costs+p;
1691: assert(c->state_in==nextstate);
1692: assert(ts[i]->inst[nextstate].cost != INF_COST);
1693: #if defined(GFORTH_DEBUGGING)
1694: assert(p == origs[i]);
1695: #endif
1696: tc2 = compile_prim_dyn(p,instps[i]);
1697: if (no_transition || !is_relocatable(p))
1698: /* !! actually what we care about is if and where
1699: * compile_prim_dyn() puts NEXTs */
1700: tc=tc2;
1701: no_transition = ts[i]->inst[nextstate].no_transition;
1702: nextstate = c->state_out;
1703: nextdyn += c->length;
1704: }
1705: } else {
1706: #if defined(GFORTH_DEBUGGING)
1707: assert(0);
1708: #endif
1709: tc=0;
1710: /* tc= (Cell)vm_prims[ts[i]->inst[CANONICAL_STATE].inst]; */
1711: }
1712: *(instps[i]) = tc;
1713: }
1714: if (!no_transition) {
1715: PrimNum p = ts[i]->trans[nextstate].inst;
1716: struct cost *c = super_costs+p;
1717: assert(c->state_in==nextstate);
1718: assert(ts[i]->trans[nextstate].cost != INF_COST);
1719: assert(i==nextdyn);
1720: (void)compile_prim_dyn(p,NULL);
1721: nextstate = c->state_out;
1722: }
1723: assert(nextstate==CANONICAL_STATE);
1724: }
1725: #endif
1726:
1727: /* compile *start, possibly rewriting it into a static and/or dynamic
1728: superinstruction */
1729: void compile_prim1(Cell *start)
1730: {
1731: #if defined(DOUBLY_INDIRECT)
1732: Label prim;
1733:
1734: if (start==NULL)
1735: return;
1736: prim = (Label)*start;
1737: if (prim<((Label)(xts+DOESJUMP)) || prim>((Label)(xts+npriminfos))) {
1738: fprintf(stderr,"compile_prim encountered xt %p\n", prim);
1739: *start=(Cell)prim;
1740: return;
1741: } else {
1742: *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
1743: return;
1744: }
1745: #elif defined(INDIRECT_THREADED)
1746: return;
1747: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1748: /* !! does not work, for unknown reasons; but something like this is
1749: probably needed to ensure that we don't call compile_prim_dyn
1750: before the inline arguments are there */
1751: static Cell *instps[MAX_BB];
1752: static PrimNum origs[MAX_BB];
1753: static int ninsts=0;
1754: PrimNum prim_num;
1755:
1756: if (start==NULL || ninsts >= MAX_BB ||
1757: (ninsts>0 && superend[origs[ninsts-1]])) {
1758: /* after bb, or at the start of the next bb */
1759: optimize_rewrite(instps,origs,ninsts);
1760: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
1761: ninsts=0;
1762: if (start==NULL)
1763: return;
1764: }
1765: prim_num = ((Xt)*start)-vm_prims;
1766: if(prim_num >= npriminfos) {
1767: optimize_rewrite(instps,origs,ninsts);
1768: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
1769: ninsts=0;
1770: return;
1771: }
1772: assert(ninsts<MAX_BB);
1773: instps[ninsts] = start;
1774: origs[ninsts] = prim_num;
1775: ninsts++;
1776: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1777: }
1778:
1779: #ifndef STANDALONE
1780: Address gforth_loader(FILE *imagefile, char* filename)
1781: /* returns the address of the image proper (after the preamble) */
1782: {
1783: ImageHeader header;
1784: Address image;
1785: Address imp; /* image+preamble */
1786: Char magic[8];
1787: char magic7; /* size byte of magic number */
1788: Cell preamblesize=0;
1789: Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1790: UCell check_sum;
1791: Cell ausize = ((RELINFOBITS == 8) ? 0 :
1792: (RELINFOBITS == 16) ? 1 :
1793: (RELINFOBITS == 32) ? 2 : 3);
1794: Cell charsize = ((sizeof(Char) == 1) ? 0 :
1795: (sizeof(Char) == 2) ? 1 :
1796: (sizeof(Char) == 4) ? 2 : 3) + ausize;
1797: Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
1798: (sizeof(Cell) == 2) ? 1 :
1799: (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1800: Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
1801: #ifdef WORDS_BIGENDIAN
1802: 0
1803: #else
1804: 1
1805: #endif
1806: ;
1807:
1808: vm_prims = gforth_engine(0,0,0,0,0);
1809: check_prims(vm_prims);
1810: prepare_super_table();
1811: #ifndef DOUBLY_INDIRECT
1812: #ifdef PRINT_SUPER_LENGTHS
1813: print_super_lengths();
1814: #endif
1815: check_sum = checksum(vm_prims);
1816: #else /* defined(DOUBLY_INDIRECT) */
1817: check_sum = (UCell)vm_prims;
1818: #endif /* defined(DOUBLY_INDIRECT) */
1819: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
1820: termstate = make_termstate();
1821: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1822:
1823: do {
1824: if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1825: fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
1826: progname, filename);
1827: exit(1);
1828: }
1829: preamblesize+=8;
1830: } while(memcmp(magic,"Gforth3",7));
1831: magic7 = magic[7];
1832: if (debug) {
1833: magic[7]='\0';
1834: fprintf(stderr,"Magic found: %s ", magic);
1835: print_sizes(magic7);
1836: }
1837:
1838: if (magic7 != sizebyte)
1839: {
1840: fprintf(stderr,"This image is: ");
1841: print_sizes(magic7);
1842: fprintf(stderr,"whereas the machine is ");
1843: print_sizes(sizebyte);
1844: exit(-2);
1845: };
1846:
1847: fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1848:
1849: set_stack_sizes(&header);
1850:
1851: #if HAVE_GETPAGESIZE
1852: pagesize=getpagesize(); /* Linux/GNU libc offers this */
1853: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
1854: pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
1855: #elif PAGESIZE
1856: pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
1857: #endif
1858: debugp(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1859:
1860: image = dict_alloc_read(imagefile, preamblesize+header.image_size,
1861: preamblesize+dictsize, data_offset);
1862: imp=image+preamblesize;
1863:
1864: alloc_stacks((ImageHeader *)imp);
1865: if (clear_dictionary)
1866: memset(imp+header.image_size, 0, dictsize-header.image_size);
1867: if(header.base==0 || header.base == (Address)0x100) {
1868: Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
1869: Char reloc_bits[reloc_size];
1870: fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1871: fread(reloc_bits, 1, reloc_size, imagefile);
1872: gforth_relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1873: #if 0
1874: { /* let's see what the relocator did */
1875: FILE *snapshot=fopen("snapshot.fi","wb");
1876: fwrite(image,1,imagesize,snapshot);
1877: fclose(snapshot);
1878: }
1879: #endif
1880: }
1881: else if(header.base!=imp) {
1882: fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
1883: progname, (unsigned long)header.base, (unsigned long)imp);
1884: exit(1);
1885: }
1886: if (header.checksum==0)
1887: ((ImageHeader *)imp)->checksum=check_sum;
1888: else if (header.checksum != check_sum) {
1889: fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
1890: progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
1891: exit(1);
1892: }
1893: #ifdef DOUBLY_INDIRECT
1894: ((ImageHeader *)imp)->xt_base = xts;
1895: #endif
1896: fclose(imagefile);
1897:
1898: /* unnecessary, except maybe for CODE words */
1899: /* FLUSH_ICACHE(imp, header.image_size);*/
1900:
1901: return imp;
1902: }
1903: #endif
1904:
1905: /* pointer to last '/' or '\' in file, 0 if there is none. */
1906: static char *onlypath(char *filename)
1907: {
1908: return strrchr(filename, DIRSEP);
1909: }
1910:
1911: static FILE *openimage(char *fullfilename)
1912: {
1913: FILE *image_file;
1914: char * expfilename = tilde_cstr((Char *)fullfilename, strlen(fullfilename), 1);
1915:
1916: image_file=fopen(expfilename,"rb");
1917: if (image_file!=NULL && debug)
1918: fprintf(stderr, "Opened image file: %s\n", expfilename);
1919: return image_file;
1920: }
1921:
1922: /* try to open image file concat(path[0:len],imagename) */
1923: static FILE *checkimage(char *path, int len, char *imagename)
1924: {
1925: int dirlen=len;
1926: char fullfilename[dirlen+strlen((char *)imagename)+2];
1927:
1928: memcpy(fullfilename, path, dirlen);
1929: if (fullfilename[dirlen-1]!=DIRSEP)
1930: fullfilename[dirlen++]=DIRSEP;
1931: strcpy(fullfilename+dirlen,imagename);
1932: return openimage(fullfilename);
1933: }
1934:
1935: static FILE * open_image_file(char * imagename, char * path)
1936: {
1937: FILE * image_file=NULL;
1938: char *origpath=path;
1939:
1940: if(strchr(imagename, DIRSEP)==NULL) {
1941: /* first check the directory where the exe file is in !! 01may97jaw */
1942: if (onlypath(progname))
1943: image_file=checkimage(progname, onlypath(progname)-progname, imagename);
1944: if (!image_file)
1945: do {
1946: char *pend=strchr(path, PATHSEP);
1947: if (pend==NULL)
1948: pend=path+strlen(path);
1949: if (strlen(path)==0) break;
1950: image_file=checkimage(path, pend-path, imagename);
1951: path=pend+(*pend==PATHSEP);
1952: } while (image_file==NULL);
1953: } else {
1954: image_file=openimage(imagename);
1955: }
1956:
1957: if (!image_file) {
1958: fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
1959: progname, imagename, origpath);
1960: exit(1);
1961: }
1962:
1963: return image_file;
1964: }
1965: #endif
1966:
1967: #ifdef STANDALONE_ALLOC
1968: Address gforth_alloc(Cell size)
1969: {
1970: Address r;
1971: /* leave a little room (64B) for stack underflows */
1972: if ((r = malloc(size+64))==NULL) {
1973: perror(progname);
1974: exit(1);
1975: }
1976: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
1977: debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
1978: return r;
1979: }
1980: #endif
1981:
1982: #ifdef HAS_OS
1983: static UCell convsize(char *s, UCell elemsize)
1984: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
1985: of bytes. the letter at the end indicates the unit, where e stands
1986: for the element size. default is e */
1987: {
1988: char *endp;
1989: UCell n,m;
1990:
1991: m = elemsize;
1992: n = strtoul(s,&endp,0);
1993: if (endp!=NULL) {
1994: if (strcmp(endp,"b")==0)
1995: m=1;
1996: else if (strcmp(endp,"k")==0)
1997: m=1024;
1998: else if (strcmp(endp,"M")==0)
1999: m=1024*1024;
2000: else if (strcmp(endp,"G")==0)
2001: m=1024*1024*1024;
2002: else if (strcmp(endp,"T")==0) {
2003: #if (SIZEOF_CHAR_P > 4)
2004: m=1024L*1024*1024*1024;
2005: #else
2006: fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
2007: exit(1);
2008: #endif
2009: } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
2010: fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
2011: exit(1);
2012: }
2013: }
2014: return n*m;
2015: }
2016:
2017: enum {
2018: ss_number = 256,
2019: ss_states,
2020: ss_min_codesize,
2021: ss_min_ls,
2022: ss_min_lsu,
2023: ss_min_nexts,
2024: };
2025:
2026: #ifndef STANDALONE
2027: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
2028: {
2029: int c;
2030:
2031: opterr=0;
2032: while (1) {
2033: int option_index=0;
2034: static struct option opts[] = {
2035: {"appl-image", required_argument, NULL, 'a'},
2036: {"image-file", required_argument, NULL, 'i'},
2037: {"dictionary-size", required_argument, NULL, 'm'},
2038: {"data-stack-size", required_argument, NULL, 'd'},
2039: {"return-stack-size", required_argument, NULL, 'r'},
2040: {"fp-stack-size", required_argument, NULL, 'f'},
2041: {"locals-stack-size", required_argument, NULL, 'l'},
2042: {"vm-commit", no_argument, &map_noreserve, 0},
2043: {"path", required_argument, NULL, 'p'},
2044: {"version", no_argument, NULL, 'v'},
2045: {"help", no_argument, NULL, 'h'},
2046: /* put something != 0 into offset_image */
2047: {"offset-image", no_argument, &offset_image, 1},
2048: {"no-offset-im", no_argument, &offset_image, 0},
2049: {"clear-dictionary", no_argument, &clear_dictionary, 1},
2050: {"die-on-signal", no_argument, &die_on_signal, 1},
2051: {"ignore-async-signals", no_argument, &ignore_async_signals, 1},
2052: {"debug", no_argument, &debug, 1},
2053: {"diag", no_argument, &diag, 1},
2054: {"no-super", no_argument, &no_super, 1},
2055: {"no-dynamic", no_argument, &no_dynamic, 1},
2056: {"dynamic", no_argument, &no_dynamic, 0},
2057: {"print-metrics", no_argument, &print_metrics, 1},
2058: {"ss-number", required_argument, NULL, ss_number},
2059: {"ss-states", required_argument, NULL, ss_states},
2060: #ifndef NO_DYNAMIC
2061: {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
2062: #endif
2063: {"ss-min-ls", no_argument, NULL, ss_min_ls},
2064: {"ss-min-lsu", no_argument, NULL, ss_min_lsu},
2065: {"ss-min-nexts", no_argument, NULL, ss_min_nexts},
2066: {"ss-greedy", no_argument, &ss_greedy, 1},
2067: {"tpa-noequiv", no_argument, &tpa_noequiv, 1},
2068: {"tpa-noautomaton", no_argument, &tpa_noautomaton, 1},
2069: {"tpa-trace", no_argument, &tpa_trace, 1},
2070: {0,0,0,0}
2071: /* no-init-file, no-rc? */
2072: };
2073:
2074: c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
2075:
2076: switch (c) {
2077: case EOF: return;
2078: case '?': optind--; return;
2079: case 'a': *imagename = optarg; return;
2080: case 'i': *imagename = optarg; break;
2081: case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
2082: case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
2083: case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
2084: case 'f': fsize = convsize(optarg,sizeof(Float)); break;
2085: case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
2086: case 'p': *path = optarg; break;
2087: case 'o': offset_image = 1; break;
2088: case 'n': offset_image = 0; break;
2089: case 'c': clear_dictionary = 1; break;
2090: case 's': die_on_signal = 1; break;
2091: case 'x': debug = 1; break;
2092: case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
2093: case ss_number: static_super_number = atoi(optarg); break;
2094: case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
2095: #ifndef NO_DYNAMIC
2096: case ss_min_codesize: ss_cost = cost_codesize; break;
2097: #endif
2098: case ss_min_ls: ss_cost = cost_ls; break;
2099: case ss_min_lsu: ss_cost = cost_lsu; break;
2100: case ss_min_nexts: ss_cost = cost_nexts; break;
2101: case 'h':
2102: fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
2103: Engine Options:\n\
2104: --appl-image FILE Equivalent to '--image-file=FILE --'\n\
2105: --clear-dictionary Initialize the dictionary with 0 bytes\n\
2106: -d SIZE, --data-stack-size=SIZE Specify data stack size\n\
2107: --debug Print debugging information during startup\n\
2108: --diag Print diagnostic information during startup\n\
2109: --die-on-signal Exit instead of THROWing some signals\n\
2110: --dynamic Use dynamic native code\n\
2111: -f SIZE, --fp-stack-size=SIZE Specify floating point stack size\n\
2112: -h, --help Print this message and exit\n\
2113: --ignore-async-signals Ignore instead of THROWing async. signals\n\
2114: -i FILE, --image-file=FILE Use image FILE instead of `gforth.fi'\n\
2115: -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
2116: -m SIZE, --dictionary-size=SIZE Specify Forth dictionary size\n\
2117: --no-dynamic Use only statically compiled primitives\n\
2118: --no-offset-im Load image at normal position\n\
2119: --no-super No dynamically formed superinstructions\n\
2120: --offset-image Load image at a different position\n\
2121: -p PATH, --path=PATH Search path for finding image and sources\n\
2122: --print-metrics Print some code generation metrics on exit\n\
2123: -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
2124: --ss-greedy Greedy, not optimal superinst selection\n\
2125: --ss-min-codesize Select superinsts for smallest native code\n\
2126: --ss-min-ls Minimize loads and stores\n\
2127: --ss-min-lsu Minimize loads, stores, and pointer updates\n\
2128: --ss-min-nexts Minimize the number of static superinsts\n\
2129: --ss-number=N Use N static superinsts (default max)\n\
2130: --ss-states=N N states for stack caching (default max)\n\
2131: --tpa-noequiv Automaton without state equivalence\n\
2132: --tpa-noautomaton Dynamic programming only\n\
2133: --tpa-trace Report new states etc.\n\
2134: -v, --version Print engine version and exit\n\
2135: --vm-commit Use OS default for memory overcommit\n\
2136: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
2137: `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
2138: argv[0]);
2139: optind--;
2140: return;
2141: }
2142: }
2143: }
2144: #endif
2145: #endif
2146:
2147: static void print_diag()
2148: {
2149:
2150: #if !defined(HAVE_GETRUSAGE) || (!defined(HAS_FFCALL) && !defined(HAS_LIBFFI))
2151: fprintf(stderr, "*** missing functionality ***\n"
2152: #ifndef HAVE_GETRUSAGE
2153: " no getrusage -> CPUTIME broken\n"
2154: #endif
2155: #if !defined(HAS_FFCALL) && !defined(HAS_LIBFFI)
2156: " no ffcall -> only old-style foreign function calls (no fflib.fs)\n"
2157: #endif
2158: );
2159: #endif
2160: if((relocs < nonrelocs) ||
2161: #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)
2162: 1
2163: #else
2164: 0
2165: #endif
2166: )
2167: debugp(stderr, "relocs: %d:%d\n", relocs, nonrelocs);
2168: fprintf(stderr, "*** %sperformance problems ***\n%s",
2169: #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)
2170: "",
2171: #else
2172: "no ",
2173: #endif
2174: #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)
2175: " double-cell integer type buggy ->\n "
2176: #ifdef BUGGY_LL_CMP
2177: "CMP, "
2178: #endif
2179: #ifdef BUGGY_LL_MUL
2180: "MUL, "
2181: #endif
2182: #ifdef BUGGY_LL_DIV
2183: "DIV, "
2184: #endif
2185: #ifdef BUGGY_LL_ADD
2186: "ADD, "
2187: #endif
2188: #ifdef BUGGY_LL_SHIFT
2189: "SHIFT, "
2190: #endif
2191: #ifdef BUGGY_LL_D2F
2192: "D2F, "
2193: #endif
2194: #ifdef BUGGY_LL_F2D
2195: "F2D, "
2196: #endif
2197: "\b\b slow\n"
2198: #endif
2199: #ifndef FORCE_REG
2200: " automatic register allocation: performance degradation possible\n"
2201: #endif
2202: #if !defined(FORCE_REG) || defined(BUGGY_LONG_LONG)
2203: "*** Suggested remedy: try ./configure"
2204: #ifndef FORCE_REG
2205: " --enable-force-reg"
2206: #endif
2207: #ifdef BUGGY_LONG_LONG
2208: " --enable-force-ll"
2209: #endif
2210: "\n"
2211: #else
2212: ""
2213: #endif
2214: ,
2215: (relocs < nonrelocs) ? " gcc PR 15242 -> no dynamic code generation (use gcc-2.95 instead)\n" : "");
2216: }
2217:
2218: #ifdef STANDALONE
2219: Cell data_abort_pc;
2220:
2221: void data_abort_C(void)
2222: {
2223: while(1) {
2224: }
2225: }
2226: #endif
2227:
2228: int main(int argc, char **argv, char **env)
2229: {
2230: #ifdef HAS_OS
2231: char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
2232: #else
2233: char *path = DEFAULTPATH;
2234: #endif
2235: #ifndef INCLUDE_IMAGE
2236: char *imagename="gforth.fi";
2237: FILE *image_file;
2238: Address image;
2239: #endif
2240: int retvalue;
2241:
2242: #if defined(i386) && defined(ALIGNMENT_CHECK)
2243: /* turn on alignment checks on the 486.
2244: * on the 386 this should have no effect. */
2245: __asm__("pushfl; popl %eax; orl $0x40000, %eax; pushl %eax; popfl;");
2246: /* this is unusable with Linux' libc.4.6.27, because this library is
2247: not alignment-clean; we would have to replace some library
2248: functions (e.g., memcpy) to make it work. Also GCC doesn't try to keep
2249: the stack FP-aligned. */
2250: #endif
2251:
2252: #ifndef STANDALONE
2253: /* buffering of the user output device */
2254: #ifdef _IONBF
2255: if (isatty(fileno(stdout))) {
2256: fflush(stdout);
2257: setvbuf(stdout,NULL,_IONBF,0);
2258: }
2259: #endif
2260: #else
2261: prep_terminal();
2262: #endif
2263:
2264: progname = argv[0];
2265:
2266: #ifndef STANDALONE
2267: #ifdef HAS_OS
2268: gforth_args(argc, argv, &path, &imagename);
2269: #ifndef NO_DYNAMIC
2270: init_ss_cost();
2271: #endif /* !defined(NO_DYNAMIC) */
2272: #endif /* defined(HAS_OS) */
2273: #endif
2274:
2275: #ifdef STANDALONE
2276: image = gforth_engine(0, 0, 0, 0, 0);
2277: alloc_stacks((ImageHeader *)image);
2278: #else
2279: image_file = open_image_file(imagename, path);
2280: image = gforth_loader(image_file, imagename);
2281: #endif
2282: gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
2283:
2284: if (diag)
2285: print_diag();
2286: {
2287: char path2[strlen(path)+1];
2288: char *p1, *p2;
2289: Cell environ[]= {
2290: (Cell)argc-(optind-1),
2291: (Cell)(argv+(optind-1)),
2292: (Cell)strlen(path),
2293: (Cell)path2};
2294: argv[optind-1] = progname;
2295: /*
2296: for (i=0; i<environ[0]; i++)
2297: printf("%s\n", ((char **)(environ[1]))[i]);
2298: */
2299: /* make path OS-independent by replacing path separators with NUL */
2300: for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
2301: if (*p1==PATHSEP)
2302: *p2 = '\0';
2303: else
2304: *p2 = *p1;
2305: *p2='\0';
2306: retvalue = gforth_go(image, 4, environ);
2307: #if defined(SIGPIPE) && !defined(STANDALONE)
2308: bsd_signal(SIGPIPE, SIG_IGN);
2309: #endif
2310: #ifdef VM_PROFILING
2311: vm_print_profile(stderr);
2312: #endif
2313: deprep_terminal();
2314: }
2315: if (print_metrics) {
2316: int i;
2317: fprintf(stderr, "code size = %8ld\n", dyncodesize());
2318: #ifndef STANDALONE
2319: for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
2320: fprintf(stderr, "metric %8s: %8ld\n",
2321: cost_sums[i].metricname, cost_sums[i].sum);
2322: #endif
2323: fprintf(stderr,"lb_basic_blocks = %ld\n", lb_basic_blocks);
2324: fprintf(stderr,"lb_labeler_steps = %ld\n", lb_labeler_steps);
2325: fprintf(stderr,"lb_labeler_automaton = %ld\n", lb_labeler_automaton);
2326: fprintf(stderr,"lb_labeler_dynprog = %ld\n", lb_labeler_dynprog);
2327: fprintf(stderr,"lb_newstate_equiv = %ld\n", lb_newstate_equiv);
2328: fprintf(stderr,"lb_newstate_new = %ld\n", lb_newstate_new);
2329: fprintf(stderr,"lb_applicable_base_rules = %ld\n", lb_applicable_base_rules);
2330: fprintf(stderr,"lb_applicable_chain_rules = %ld\n", lb_applicable_chain_rules);
2331: }
2332: if (tpa_trace) {
2333: fprintf(stderr, "%ld %ld lb_states\n", lb_labeler_steps, lb_newstate_new);
2334: fprintf(stderr, "%ld %ld lb_table_entries\n", lb_labeler_steps, lb_labeler_dynprog);
2335: }
2336: return retvalue;
2337: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>