1: /* command line interpretation, image loading etc. for Gforth
2:
3:
4: Copyright (C) 1995,1996,1997,1998,2000,2003 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 *SP;
60: Float *FP;
61: Address UP=NULL;
62:
63: #ifdef HAS_FFCALL
64: Cell *RP;
65: Address LP;
66:
67: #include <callback.h>
68:
69: va_alist clist;
70:
71: void engine_callback(Xt* fcall, void * alist)
72: {
73: clist = (va_alist)alist;
74: engine(fcall, SP, RP, FP, LP);
75: }
76: #endif
77:
78: #ifdef GFORTH_DEBUGGING
79: /* define some VM registers as global variables, so they survive exceptions;
80: global register variables are not up to the task (according to the
81: GNU C manual) */
82: Xt *saved_ip;
83: Cell *rp;
84: #endif
85:
86: #ifdef NO_IP
87: Label next_code;
88: #endif
89:
90: #ifdef HAS_FILE
91: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
92: char* pfileattr[6]={"r","r","r+","r+","w","w"};
93:
94: #ifndef O_BINARY
95: #define O_BINARY 0
96: #endif
97: #ifndef O_TEXT
98: #define O_TEXT 0
99: #endif
100:
101: int ufileattr[6]= {
102: O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
103: O_RDWR |O_BINARY, O_RDWR |O_BINARY,
104: O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
105: #endif
106: /* end global vars for engine.c */
107:
108: #define PRIM_VERSION 1
109: /* increment this whenever the primitives change in an incompatible way */
110:
111: #ifndef DEFAULTPATH
112: # define DEFAULTPATH "."
113: #endif
114:
115: #ifdef MSDOS
116: jmp_buf throw_jmp_buf;
117: #endif
118:
119: #if defined(DOUBLY_INDIRECT)
120: # define CFA(n) ({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
121: #else
122: # define CFA(n) ((Cell)(symbols+((n)&~0x4000UL)))
123: #endif
124:
125: #define maxaligned(n) (typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
126:
127: static UCell dictsize=0;
128: static UCell dsize=0;
129: static UCell rsize=0;
130: static UCell fsize=0;
131: static UCell lsize=0;
132: int offset_image=0;
133: int die_on_signal=0;
134: #ifndef INCLUDE_IMAGE
135: static int clear_dictionary=0;
136: UCell pagesize=1;
137: char *progname;
138: #else
139: char *progname = "gforth";
140: int optind = 1;
141: #endif
142:
143: #define CODE_BLOCK_SIZE (256*1024)
144: Address code_area=0;
145: Cell code_area_size = CODE_BLOCK_SIZE;
146: Address code_here=NULL+CODE_BLOCK_SIZE; /* does for code-area what HERE
147: does for the dictionary */
148: Address start_flush=NULL; /* start of unflushed code */
149: Cell last_jump=0; /* if the last prim was compiled without jump, this
150: is it's number, otherwise this contains 0 */
151:
152: static int no_super=0; /* true if compile_prim should not fuse prims */
153: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
154: dynamically */
155: static int print_metrics=0; /* if true, print metrics on exit */
156: static int static_super_number = 10000000; /* number of ss used if available */
157: #define MAX_STATE 4 /* maximum number of states */
158: static int maxstates = MAX_STATE; /* number of states for stack caching */
159: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
160:
161: #ifdef HAS_DEBUG
162: int debug=0;
163: #else
164: # define perror(x...)
165: # define fprintf(x...)
166: #endif
167:
168: ImageHeader *gforth_header;
169: Label *vm_prims;
170: #ifdef DOUBLY_INDIRECT
171: Label *xts; /* same content as vm_prims, but should only be used for xts */
172: #endif
173:
174: #ifndef NO_DYNAMIC
175: #define MAX_IMMARGS 2
176:
177: typedef struct {
178: Label start; /* NULL if not relocatable */
179: Cell length; /* only includes the jump iff superend is true*/
180: Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
181: char superend; /* true if primitive ends superinstruction, i.e.,
182: unconditional branch, execute, etc. */
183: Cell nimmargs;
184: struct immarg {
185: Cell offset; /* offset of immarg within prim */
186: char rel; /* true if immarg is relative */
187: } immargs[MAX_IMMARGS];
188: } PrimInfo;
189:
190: PrimInfo *priminfos;
191: PrimInfo **decomp_prims;
192:
193: static int is_relocatable(int p)
194: {
195: return !no_dynamic && priminfos[p].start != NULL;
196: }
197: #else /* defined(NO_DYNAMIC) */
198: static int is_relocatable(int p)
199: {
200: return 0;
201: }
202: #endif /* defined(NO_DYNAMIC) */
203:
204: #ifdef MEMCMP_AS_SUBROUTINE
205: int gforth_memcmp(const char * s1, const char * s2, size_t n)
206: {
207: return memcmp(s1, s2, n);
208: }
209: #endif
210:
211: static Cell max(Cell a, Cell b)
212: {
213: return a>b?a:b;
214: }
215:
216: static Cell min(Cell a, Cell b)
217: {
218: return a<b?a:b;
219: }
220:
221: /* image file format:
222: * "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
223: * padding to a multiple of 8
224: * magic: "Gforth3x" means format 0.6,
225: * where x is a byte with
226: * bit 7: reserved = 0
227: * bit 6:5: address unit size 2^n octets
228: * bit 4:3: character size 2^n octets
229: * bit 2:1: cell size 2^n octets
230: * bit 0: endian, big=0, little=1.
231: * The magic are always 8 octets, no matter what the native AU/character size is
232: * padding to max alignment (no padding necessary on current machines)
233: * ImageHeader structure (see forth.h)
234: * data (size in ImageHeader.image_size)
235: * tags ((if relocatable, 1 bit/data cell)
236: *
237: * tag==1 means that the corresponding word is an address;
238: * If the word is >=0, the address is within the image;
239: * addresses within the image are given relative to the start of the image.
240: * If the word =-1 (CF_NIL), the address is NIL,
241: * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
242: * If the word =CF(DODOES), it's a DOES> CFA
243: * If the word =CF(DOESJUMP), it's a DOES JUMP (2 Cells after DOES>,
244: * possibly containing a jump to dodoes)
245: * If the word is <CF(DOESJUMP) and bit 14 is set, it's the xt of a primitive
246: * If the word is <CF(DOESJUMP) and bit 14 is clear,
247: * it's the threaded code of a primitive
248: * bits 13..9 of a primitive token state which group the primitive belongs to,
249: * bits 8..0 of a primitive token index into the group
250: */
251:
252: Cell groups[32] = {
253: 0,
254: 0
255: #undef GROUP
256: #undef GROUPADD
257: #define GROUPADD(n) +n
258: #define GROUP(x, n) , 0
259: #include PRIM_GRP_I
260: #undef GROUP
261: #undef GROUPADD
262: #define GROUP(x, n)
263: #define GROUPADD(n)
264: };
265:
266: unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
267: int size, Cell base)
268: /* produce a bitmask marking all the branch targets */
269: {
270: int i=0, j, k, steps=(size/sizeof(Cell))/RELINFOBITS;
271: Cell token;
272: unsigned char bits;
273: unsigned char *result=malloc(steps+1);
274:
275: memset(result, 0, steps+1);
276: for(k=0; k<=steps; k++) {
277: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
278: if((i < size) && (bits & (1U << (RELINFOBITS-1)))) {
279: token=image[i];
280: if (token>=base) { /* relocatable address */
281: UCell bitnum=(token-base)/sizeof(Cell);
282: result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
283: }
284: }
285: }
286: }
287: return result;
288: }
289:
290: void relocate(Cell *image, const unsigned char *bitstring,
291: int size, Cell base, Label symbols[])
292: {
293: int i=0, j, k, steps=(size/sizeof(Cell))/RELINFOBITS;
294: Cell token;
295: char bits;
296: Cell max_symbols;
297: /*
298: * A virtual start address that's the real start address minus
299: * the one in the image
300: */
301: Cell *start = (Cell * ) (((void *) image) - ((void *) base));
302: unsigned char *targets = branch_targets(image, bitstring, size, base);
303:
304: /* group index into table */
305: if(groups[31]==0) {
306: int groupsum=0;
307: for(i=0; i<32; i++) {
308: groupsum += groups[i];
309: groups[i] = groupsum;
310: /* printf("group[%d]=%d\n",i,groupsum); */
311: }
312: i=0;
313: }
314:
315: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
316:
317: for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
318: ;
319: max_symbols--;
320: size/=sizeof(Cell);
321:
322: for(k=0; k<=steps; k++) {
323: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
324: /* fprintf(stderr,"relocate: image[%d]\n", i);*/
325: if((i < size) && (bits & (1U << (RELINFOBITS-1)))) {
326: /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
327: token=image[i];
328: if(token<0) {
329: int group = (-token & 0x3E00) >> 9;
330: if(group == 0) {
331: switch(token|0x4000) {
332: case CF_NIL : image[i]=0; break;
333: #if !defined(DOUBLY_INDIRECT)
334: case CF(DOCOL) :
335: case CF(DOVAR) :
336: case CF(DOCON) :
337: case CF(DOUSER) :
338: case CF(DODEFER) :
339: case CF(DOFIELD) : MAKE_CF(image+i,symbols[CF(token)]); break;
340: case CF(DOESJUMP): image[i]=0; break;
341: #endif /* !defined(DOUBLY_INDIRECT) */
342: case CF(DODOES) :
343: MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
344: break;
345: default : /* backward compatibility */
346: /* printf("Code field generation image[%x]:=CFA(%x)\n",
347: i, CF(image[i])); */
348: if (CF((token | 0x4000))<max_symbols) {
349: image[i]=(Cell)CFA(CF(token));
350: #ifdef DIRECT_THREADED
351: if ((token & 0x4000) == 0) { /* threade code, no CFA */
352: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
353: compile_prim1(0);
354: compile_prim1(&image[i]);
355: }
356: #endif
357: } else
358: 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);
359: }
360: } else {
361: int tok = -token & 0x1FF;
362: if (tok < (groups[group+1]-groups[group])) {
363: #if defined(DOUBLY_INDIRECT)
364: image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
365: #else
366: image[i]=(Cell)CFA((groups[group]+tok));
367: #endif
368: #ifdef DIRECT_THREADED
369: if ((token & 0x4000) == 0) { /* threade code, no CFA */
370: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
371: compile_prim1(0);
372: compile_prim1(&image[i]);
373: }
374: #endif
375: } else
376: fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
377: }
378: } else {
379: /* if base is > 0: 0 is a null reference so don't adjust*/
380: if (token>=base) {
381: image[i]+=(Cell)start;
382: }
383: }
384: }
385: }
386: }
387: free(targets);
388: finish_code();
389: ((ImageHeader*)(image))->base = (Address) image;
390: }
391:
392: UCell checksum(Label symbols[])
393: {
394: UCell r=PRIM_VERSION;
395: Cell i;
396:
397: for (i=DOCOL; i<=DOESJUMP; i++) {
398: r ^= (UCell)(symbols[i]);
399: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
400: }
401: #ifdef DIRECT_THREADED
402: /* we have to consider all the primitives */
403: for (; symbols[i]!=(Label)0; i++) {
404: r ^= (UCell)(symbols[i]);
405: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
406: }
407: #else
408: /* in indirect threaded code all primitives are accessed through the
409: symbols table, so we just have to put the base address of symbols
410: in the checksum */
411: r ^= (UCell)symbols;
412: #endif
413: return r;
414: }
415:
416: Address verbose_malloc(Cell size)
417: {
418: Address r;
419: /* leave a little room (64B) for stack underflows */
420: if ((r = malloc(size+64))==NULL) {
421: perror(progname);
422: exit(1);
423: }
424: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
425: if (debug)
426: fprintf(stderr, "malloc succeeds, address=$%lx\n", (long)r);
427: return r;
428: }
429:
430: static Address next_address=0;
431: void after_alloc(Address r, Cell size)
432: {
433: if (r != (Address)-1) {
434: if (debug)
435: fprintf(stderr, "success, address=$%lx\n", (long) r);
436: if (pagesize != 1)
437: next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
438: } else {
439: if (debug)
440: fprintf(stderr, "failed: %s\n", strerror(errno));
441: }
442: }
443:
444: #ifndef MAP_FAILED
445: #define MAP_FAILED ((Address) -1)
446: #endif
447: #ifndef MAP_FILE
448: # define MAP_FILE 0
449: #endif
450: #ifndef MAP_PRIVATE
451: # define MAP_PRIVATE 0
452: #endif
453: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
454: # define MAP_ANON MAP_ANONYMOUS
455: #endif
456:
457: #if defined(HAVE_MMAP)
458: static Address alloc_mmap(Cell size)
459: {
460: Address r;
461:
462: #if defined(MAP_ANON)
463: if (debug)
464: fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
465: r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE, -1, 0);
466: #else /* !defined(MAP_ANON) */
467: /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
468: apparently defaults) */
469: static int dev_zero=-1;
470:
471: if (dev_zero == -1)
472: dev_zero = open("/dev/zero", O_RDONLY);
473: if (dev_zero == -1) {
474: r = MAP_FAILED;
475: if (debug)
476: fprintf(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ",
477: strerror(errno));
478: } else {
479: if (debug)
480: fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
481: r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE, dev_zero, 0);
482: }
483: #endif /* !defined(MAP_ANON) */
484: after_alloc(r, size);
485: return r;
486: }
487: #endif
488:
489: Address my_alloc(Cell size)
490: {
491: #if HAVE_MMAP
492: Address r;
493:
494: r=alloc_mmap(size);
495: if (r!=(Address)MAP_FAILED)
496: return r;
497: #endif /* HAVE_MMAP */
498: /* use malloc as fallback */
499: return verbose_malloc(size);
500: }
501:
502: Address dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
503: {
504: Address image = MAP_FAILED;
505:
506: #if defined(HAVE_MMAP)
507: if (offset==0) {
508: image=alloc_mmap(dictsize);
509: if (debug)
510: fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
511: image = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE, fileno(file), 0);
512: after_alloc(image,dictsize);
513: }
514: #endif /* defined(HAVE_MMAP) */
515: if (image == (Address)MAP_FAILED) {
516: image = my_alloc(dictsize+offset)+offset;
517: rewind(file); /* fseek(imagefile,0L,SEEK_SET); */
518: fread(image, 1, imagesize, file);
519: }
520: return image;
521: }
522:
523: void set_stack_sizes(ImageHeader * header)
524: {
525: if (dictsize==0)
526: dictsize = header->dict_size;
527: if (dsize==0)
528: dsize = header->data_stack_size;
529: if (rsize==0)
530: rsize = header->return_stack_size;
531: if (fsize==0)
532: fsize = header->fp_stack_size;
533: if (lsize==0)
534: lsize = header->locals_stack_size;
535: dictsize=maxaligned(dictsize);
536: dsize=maxaligned(dsize);
537: rsize=maxaligned(rsize);
538: lsize=maxaligned(lsize);
539: fsize=maxaligned(fsize);
540: }
541:
542: void alloc_stacks(ImageHeader * header)
543: {
544: header->dict_size=dictsize;
545: header->data_stack_size=dsize;
546: header->fp_stack_size=fsize;
547: header->return_stack_size=rsize;
548: header->locals_stack_size=lsize;
549:
550: header->data_stack_base=my_alloc(dsize);
551: header->fp_stack_base=my_alloc(fsize);
552: header->return_stack_base=my_alloc(rsize);
553: header->locals_stack_base=my_alloc(lsize);
554: }
555:
556: #warning You can ignore the warnings about clobbered variables in go_forth
557: int go_forth(Address image, int stack, Cell *entries)
558: {
559: volatile ImageHeader *image_header = (ImageHeader *)image;
560: Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
561: Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
562: Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
563: #ifdef GFORTH_DEBUGGING
564: volatile Cell *orig_rp0=rp0;
565: #endif
566: Address lp0=image_header->locals_stack_base + lsize;
567: Xt *ip0=(Xt *)(image_header->boot_entry);
568: #ifdef SYSSIGNALS
569: int throw_code;
570: #endif
571:
572: /* ensure that the cached elements (if any) are accessible */
573: IF_spTOS(sp0--);
574: IF_fpTOS(fp0--);
575:
576: for(;stack>0;stack--)
577: *--sp0=entries[stack-1];
578:
579: #ifdef SYSSIGNALS
580: get_winsize();
581:
582: install_signal_handlers(); /* right place? */
583:
584: if ((throw_code=setjmp(throw_jmp_buf))) {
585: static Cell signal_data_stack[8];
586: static Cell signal_return_stack[8];
587: static Float signal_fp_stack[1];
588:
589: signal_data_stack[7]=throw_code;
590:
591: #ifdef GFORTH_DEBUGGING
592: if (debug)
593: fprintf(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
594: throw_code, saved_ip, rp);
595: if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
596: /* no rstack overflow or underflow */
597: rp0 = rp;
598: *--rp0 = (Cell)saved_ip;
599: }
600: else /* I love non-syntactic ifdefs :-) */
601: rp0 = signal_return_stack+8;
602: #else /* !defined(GFORTH_DEBUGGING) */
603: if (debug)
604: fprintf(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
605: rp0 = signal_return_stack+8;
606: #endif /* !defined(GFORTH_DEBUGGING) */
607: /* fprintf(stderr, "rp=$%x\n",rp0);*/
608:
609: return((int)(Cell)engine(image_header->throw_entry, signal_data_stack+7,
610: rp0, signal_fp_stack, 0));
611: }
612: #endif
613:
614: return((int)(Cell)engine(ip0,sp0,rp0,fp0,lp0));
615: }
616:
617: #ifndef INCLUDE_IMAGE
618: void print_sizes(Cell sizebyte)
619: /* print size information */
620: {
621: static char* endianstring[]= { " big","little" };
622:
623: fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
624: endianstring[sizebyte & 1],
625: 1 << ((sizebyte >> 1) & 3),
626: 1 << ((sizebyte >> 3) & 3),
627: 1 << ((sizebyte >> 5) & 3));
628: }
629:
630: /* static superinstruction stuff */
631:
632: struct cost {
633: char loads; /* number of stack loads */
634: char stores; /* number of stack stores */
635: char updates; /* number of stack pointer updates */
636: char branch; /* is it a branch (SET_IP) */
637: unsigned char state_in; /* state on entry */
638: unsigned char state_out; /* state on exit */
639: short offset; /* offset into super2 table */
640: unsigned char length; /* number of components */
641: };
642:
643: PrimNum super2[] = {
644: #include SUPER2_I
645: };
646:
647: struct cost super_costs[] = {
648: #include COSTS_I
649: };
650:
651: struct super_state {
652: struct super_state *next;
653: PrimNum super;
654: };
655:
656: #define HASH_SIZE 256
657:
658: struct super_table_entry {
659: struct super_table_entry *next;
660: PrimNum *start;
661: short length;
662: struct super_state *ss_list; /* list of supers */
663: } *super_table[HASH_SIZE];
664: int max_super=2;
665:
666: struct super_state *state_transitions=NULL;
667:
668: int hash_super(PrimNum *start, int length)
669: {
670: int i, r;
671:
672: for (i=0, r=0; i<length; i++) {
673: r <<= 1;
674: r += start[i];
675: }
676: return r & (HASH_SIZE-1);
677: }
678:
679: struct super_state **lookup_super(PrimNum *start, int length)
680: {
681: int hash=hash_super(start,length);
682: struct super_table_entry *p = super_table[hash];
683:
684: /* assert(length >= 2); */
685: for (; p!=NULL; p = p->next) {
686: if (length == p->length &&
687: memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
688: return &(p->ss_list);
689: }
690: return NULL;
691: }
692:
693: void prepare_super_table()
694: {
695: int i;
696: int nsupers = 0;
697:
698: for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
699: struct cost *c = &super_costs[i];
700: if ((c->length < 2 || nsupers < static_super_number) &&
701: c->state_in < maxstates && c->state_out < maxstates) {
702: struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
703: struct super_state *ss = malloc(sizeof(struct super_state));
704: ss->super= i;
705: if (c->offset==N_noop && i != N_noop) {
706: if (is_relocatable(i)) {
707: ss->next = state_transitions;
708: state_transitions = ss;
709: }
710: } else if (ss_listp != NULL) {
711: ss->next = *ss_listp;
712: *ss_listp = ss;
713: } else {
714: int hash = hash_super(super2+c->offset, c->length);
715: struct super_table_entry **p = &super_table[hash];
716: struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
717: ss->next = NULL;
718: e->next = *p;
719: e->start = super2 + c->offset;
720: e->length = c->length;
721: e->ss_list = ss;
722: *p = e;
723: }
724: if (c->length > max_super)
725: max_super = c->length;
726: if (c->length >= 2)
727: nsupers++;
728: }
729: }
730: if (debug)
731: fprintf(stderr, "Using %d static superinsts\n", nsupers);
732: }
733:
734: /* dynamic replication/superinstruction stuff */
735:
736: #ifndef NO_DYNAMIC
737: int compare_priminfo_length(const void *_a, const void *_b)
738: {
739: PrimInfo **a = (PrimInfo **)_a;
740: PrimInfo **b = (PrimInfo **)_b;
741: Cell diff = (*a)->length - (*b)->length;
742: if (diff)
743: return diff;
744: else /* break ties by start address; thus the decompiler produces
745: the earliest primitive with the same code (e.g. noop instead
746: of (char) and @ instead of >code-address */
747: return (*b)->start - (*a)->start;
748: }
749: #endif /* !defined(NO_DYNAMIC) */
750:
751: static char MAYBE_UNUSED superend[]={
752: #include PRIM_SUPEREND_I
753: };
754:
755: Cell npriminfos=0;
756:
757: int compare_labels(const void *pa, const void *pb)
758: {
759: Label a = *(Label *)pa;
760: Label b = *(Label *)pb;
761: return a-b;
762: }
763:
764: Label bsearch_next(Label key, Label *a, UCell n)
765: /* a is sorted; return the label >=key that is the closest in a;
766: return NULL if there is no label in a >=key */
767: {
768: int mid = (n-1)/2;
769: if (n<1)
770: return NULL;
771: if (n == 1) {
772: if (a[0] < key)
773: return NULL;
774: else
775: return a[0];
776: }
777: if (a[mid] < key)
778: return bsearch_next(key, a+mid+1, n-mid-1);
779: else
780: return bsearch_next(key, a, mid+1);
781: }
782:
783: void check_prims(Label symbols1[])
784: {
785: int i;
786: #ifndef NO_DYNAMIC
787: Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted;
788: int nends1j;
789: #endif
790:
791: if (debug)
792: #ifdef __VERSION__
793: fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
794: #else
795: #define xstr(s) str(s)
796: #define str(s) #s
797: fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n");
798: #endif
799: for (i=0; symbols1[i]!=0; i++)
800: ;
801: npriminfos = i;
802:
803: #ifndef NO_DYNAMIC
804: if (no_dynamic)
805: return;
806: symbols2=engine2(0,0,0,0,0);
807: #if NO_IP
808: symbols3=engine3(0,0,0,0,0);
809: #else
810: symbols3=symbols1;
811: #endif
812: ends1 = symbols1+i+1;
813: ends1j = ends1+i;
814: nends1j = i+1;
815: ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
816: memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
817: qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
818:
819: priminfos = calloc(i,sizeof(PrimInfo));
820: for (i=0; symbols1[i]!=0; i++) {
821: int prim_len = ends1[i]-symbols1[i];
822: PrimInfo *pi=&priminfos[i];
823: int j=0;
824: char *s1 = (char *)symbols1[i];
825: char *s2 = (char *)symbols2[i];
826: char *s3 = (char *)symbols3[i];
827: Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
828:
829: pi->start = s1;
830: pi->superend = superend[i]|no_super;
831: if (pi->superend)
832: pi->length = endlabel-symbols1[i];
833: else
834: pi->length = prim_len;
835: pi->restlength = endlabel - symbols1[i] - pi->length;
836: pi->nimmargs = 0;
837: if (debug)
838: fprintf(stderr, "Prim %3d @ %p %p %p, length=%3ld restlength=%2ld superend=%1d",
839: i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength), pi->superend);
840: if (endlabel == NULL) {
841: pi->start = NULL; /* not relocatable */
842: if (pi->length<0) pi->length=100;
843: if (debug)
844: fprintf(stderr,"\n non_reloc: no J label > start found\n");
845: continue;
846: }
847: if (ends1[i] > endlabel && !pi->superend) {
848: pi->start = NULL; /* not relocatable */
849: pi->length = endlabel-symbols1[i];
850: if (debug)
851: fprintf(stderr,"\n non_reloc: there is a J label before the K label (restlength<0)\n");
852: continue;
853: }
854: if (ends1[i] < pi->start && !pi->superend) {
855: pi->start = NULL; /* not relocatable */
856: pi->length = endlabel-symbols1[i];
857: if (debug)
858: fprintf(stderr,"\n non_reloc: K label before I label (length<0)\n");
859: continue;
860: }
861: assert(prim_len>=0);
862: assert(pi->restlength >=0);
863: while (j<(pi->length+pi->restlength)) {
864: if (s1[j]==s3[j]) {
865: if (s1[j] != s2[j]) {
866: pi->start = NULL; /* not relocatable */
867: if (debug)
868: fprintf(stderr,"\n non_reloc: engine1!=engine2 offset %3d",j);
869: /* assert(j<prim_len); */
870: break;
871: }
872: j++;
873: } else {
874: struct immarg *ia=&pi->immargs[pi->nimmargs];
875:
876: pi->nimmargs++;
877: ia->offset=j;
878: if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
879: ia->rel=0;
880: if (debug)
881: fprintf(stderr,"\n absolute immarg: offset %3d",j);
882: } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
883: symbols1[DOESJUMP+1]) {
884: ia->rel=1;
885: if (debug)
886: fprintf(stderr,"\n relative immarg: offset %3d",j);
887: } else {
888: pi->start = NULL; /* not relocatable */
889: if (debug)
890: fprintf(stderr,"\n non_reloc: engine1!=engine3 offset %3d",j);
891: /* assert(j<prim_len);*/
892: break;
893: }
894: j+=4;
895: }
896: }
897: if (debug)
898: fprintf(stderr,"\n");
899: }
900: decomp_prims = calloc(i,sizeof(PrimInfo *));
901: for (i=DOESJUMP+1; i<npriminfos; i++)
902: decomp_prims[i] = &(priminfos[i]);
903: qsort(decomp_prims+DOESJUMP+1, npriminfos-DOESJUMP-1, sizeof(PrimInfo *),
904: compare_priminfo_length);
905: #endif
906: }
907:
908: void flush_to_here(void)
909: {
910: #ifndef NO_DYNAMIC
911: if (start_flush)
912: FLUSH_ICACHE(start_flush, code_here-start_flush);
913: start_flush=code_here;
914: #endif
915: }
916:
917: #ifndef NO_DYNAMIC
918: void append_jump(void)
919: {
920: if (last_jump) {
921: PrimInfo *pi = &priminfos[last_jump];
922:
923: memcpy(code_here, pi->start+pi->length, pi->restlength);
924: code_here += pi->restlength;
925: last_jump=0;
926: }
927: }
928:
929: /* Gforth remembers all code blocks in this list. On forgetting (by
930: executing a marker) the code blocks are not freed (because Gforth does
931: not remember how they were allocated; hmm, remembering that might be
932: easier and cleaner). Instead, code_here etc. are reset to the old
933: value, and the "forgotten" code blocks are reused when they are
934: needed. */
935:
936: struct code_block_list {
937: struct code_block_list *next;
938: Address block;
939: Cell size;
940: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
941:
942: Address append_prim(Cell p)
943: {
944: PrimInfo *pi = &priminfos[p];
945: Address old_code_here = code_here;
946:
947: if (code_area+code_area_size < code_here+pi->length+pi->restlength) {
948: struct code_block_list *p;
949: append_jump();
950: flush_to_here();
951: if (*next_code_blockp == NULL) {
952: code_here = start_flush = code_area = my_alloc(code_area_size);
953: p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
954: *next_code_blockp = p;
955: p->next = NULL;
956: p->block = code_here;
957: p->size = code_area_size;
958: } else {
959: p = *next_code_blockp;
960: code_here = start_flush = code_area = p->block;
961: }
962: old_code_here = code_here;
963: next_code_blockp = &(p->next);
964: }
965: memcpy(code_here, pi->start, pi->length);
966: code_here += pi->length;
967: return old_code_here;
968: }
969: #endif
970:
971: int forget_dyncode(Address code)
972: {
973: #ifdef NO_DYNAMIC
974: return -1;
975: #else
976: struct code_block_list *p, **pp;
977:
978: for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
979: if (code >= p->block && code < p->block+p->size) {
980: next_code_blockp = &(p->next);
981: code_here = start_flush = code;
982: code_area = p->block;
983: last_jump = 0;
984: return -1;
985: }
986: }
987: return -no_dynamic;
988: #endif /* !defined(NO_DYNAMIC) */
989: }
990:
991: long dyncodesize(void)
992: {
993: #ifndef NO_DYNAMIC
994: struct code_block_list *p;
995: long size=0;
996: for (p=code_block_list; p!=NULL; p=p->next) {
997: if (code_here >= p->block && code_here < p->block+p->size)
998: return size + (code_here - p->block);
999: else
1000: size += p->size;
1001: }
1002: #endif /* !defined(NO_DYNAMIC) */
1003: return 0;
1004: }
1005:
1006: Label decompile_code(Label _code)
1007: {
1008: #ifdef NO_DYNAMIC
1009: return _code;
1010: #else /* !defined(NO_DYNAMIC) */
1011: Cell i;
1012: struct code_block_list *p;
1013: Address code=_code;
1014:
1015: /* first, check if we are in code at all */
1016: for (p = code_block_list;; p = p->next) {
1017: if (p == NULL)
1018: return code;
1019: if (code >= p->block && code < p->block+p->size)
1020: break;
1021: }
1022: /* reverse order because NOOP might match other prims */
1023: for (i=npriminfos-1; i>DOESJUMP; i--) {
1024: PrimInfo *pi=decomp_prims[i];
1025: if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1026: return vm_prims[super2[super_costs[pi-priminfos].offset]];
1027: /* return pi->start;*/
1028: }
1029: return code;
1030: #endif /* !defined(NO_DYNAMIC) */
1031: }
1032:
1033: #ifdef NO_IP
1034: int nbranchinfos=0;
1035:
1036: struct branchinfo {
1037: Label *targetptr; /* *(bi->targetptr) is the target */
1038: Cell *addressptr; /* store the target here */
1039: } branchinfos[100000];
1040:
1041: int ndoesexecinfos=0;
1042: struct doesexecinfo {
1043: int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1044: Cell *xt; /* cfa of word whose does-code needs calling */
1045: } doesexecinfos[10000];
1046:
1047: void set_rel_target(Cell *source, Label target)
1048: {
1049: *source = ((Cell)target)-(((Cell)source)+4);
1050: }
1051:
1052: void register_branchinfo(Label source, Cell targetptr)
1053: {
1054: struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1055: bi->targetptr = (Label *)targetptr;
1056: bi->addressptr = (Cell *)source;
1057: nbranchinfos++;
1058: }
1059:
1060: Cell *compile_prim1arg(Cell p)
1061: {
1062: int l = priminfos[p].length;
1063: Address old_code_here=code_here;
1064:
1065: assert(vm_prims[p]==priminfos[p].start);
1066: append_prim(p);
1067: return (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
1068: }
1069:
1070: Cell *compile_call2(Cell targetptr)
1071: {
1072: Cell *next_code_target;
1073: PrimInfo *pi = &priminfos[N_call2];
1074: Address old_code_here = append_prim(N_call2);
1075:
1076: next_code_target = (Cell *)(old_code_here + pi->immargs[0].offset);
1077: register_branchinfo(old_code_here + pi->immargs[1].offset, targetptr);
1078: return next_code_target;
1079: }
1080: #endif
1081:
1082: void finish_code(void)
1083: {
1084: #ifdef NO_IP
1085: Cell i;
1086:
1087: compile_prim1(NULL);
1088: for (i=0; i<ndoesexecinfos; i++) {
1089: struct doesexecinfo *dei = &doesexecinfos[i];
1090: branchinfos[dei->branchinfo].targetptr = DOES_CODE1((dei->xt));
1091: }
1092: ndoesexecinfos = 0;
1093: for (i=0; i<nbranchinfos; i++) {
1094: struct branchinfo *bi=&branchinfos[i];
1095: set_rel_target(bi->addressptr, *(bi->targetptr));
1096: }
1097: nbranchinfos = 0;
1098: #endif
1099: flush_to_here();
1100: }
1101:
1102: #if 0
1103: /* compile *start into a dynamic superinstruction, updating *start */
1104: void compile_prim_dyn(Cell *start)
1105: {
1106: #if defined(NO_IP)
1107: static Cell *last_start=NULL;
1108: static Xt last_prim=NULL;
1109: /* delay work by one call in order to get relocated immargs */
1110:
1111: if (last_start) {
1112: unsigned i = last_prim-vm_prims;
1113: PrimInfo *pi=&priminfos[i];
1114: Cell *next_code_target=NULL;
1115:
1116: assert(i<npriminfos);
1117: if (i==N_execute||i==N_perform||i==N_lit_perform) {
1118: next_code_target = compile_prim1arg(N_set_next_code);
1119: }
1120: if (i==N_call) {
1121: next_code_target = compile_call2(last_start[1]);
1122: } else if (i==N_does_exec) {
1123: struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1124: *compile_prim1arg(N_lit) = (Cell)PFA(last_start[1]);
1125: /* we cannot determine the callee now (last_start[1] may be a
1126: forward reference), so just register an arbitrary target, and
1127: register in dei that we need to fix this before resolving
1128: branches */
1129: dei->branchinfo = nbranchinfos;
1130: dei->xt = (Cell *)(last_start[1]);
1131: next_code_target = compile_call2(NULL);
1132: } else if (!is_relocatable(i)) {
1133: next_code_target = compile_prim1arg(N_set_next_code);
1134: set_rel_target(compile_prim1arg(N_abranch),*(Xt)last_prim);
1135: } else {
1136: unsigned j;
1137: Address old_code_here = append_prim(i);
1138:
1139: for (j=0; j<pi->nimmargs; j++) {
1140: struct immarg *ia = &(pi->immargs[j]);
1141: Cell argval = last_start[pi->nimmargs - j]; /* !! specific to prims */
1142: if (ia->rel) { /* !! assumption: relative refs are branches */
1143: register_branchinfo(old_code_here + ia->offset, argval);
1144: } else /* plain argument */
1145: *(Cell *)(old_code_here + ia->offset) = argval;
1146: }
1147: }
1148: if (next_code_target!=NULL)
1149: *next_code_target = (Cell)code_here;
1150: }
1151: if (start) {
1152: last_prim = (Xt)*start;
1153: *start = (Cell)code_here;
1154: }
1155: last_start = start;
1156: return;
1157: #elif !defined(NO_DYNAMIC)
1158: Label prim=(Label)*start;
1159: unsigned i;
1160: Address old_code_here;
1161:
1162: i = ((Xt)prim)-vm_prims;
1163: prim = *(Xt)prim;
1164: if (no_dynamic) {
1165: *start = (Cell)prim;
1166: return;
1167: }
1168: if (i>=npriminfos || !is_relocatable(i)) {
1169: append_jump();
1170: *start = (Cell)prim;
1171: return;
1172: }
1173: #ifdef ALIGN_CODE
1174: /* ALIGN_CODE;*/
1175: #endif
1176: assert(prim==priminfos[i].start);
1177: old_code_here = append_prim(i);
1178: last_jump = (priminfos[i].superend) ? 0 : i;
1179: *start = (Cell)old_code_here;
1180: return;
1181: #else /* !defined(DOUBLY_INDIRECT), no code replication */
1182: Label prim=(Label)*start;
1183: #if !defined(INDIRECT_THREADED)
1184: prim = *(Xt)prim;
1185: #endif
1186: *start = (Cell)prim;
1187: return;
1188: #endif /* !defined(DOUBLY_INDIRECT) */
1189: }
1190: #endif /* 0 */
1191:
1192: Cell compile_prim_dyn(unsigned p)
1193: {
1194: Cell static_prim = (Cell)vm_prims[p];
1195: #if defined(NO_DYNAMIC)
1196: return static_prim;
1197: #else /* !defined(NO_DYNAMIC) */
1198: Address old_code_here;
1199:
1200: if (no_dynamic)
1201: return static_prim;
1202: if (p>=npriminfos || !is_relocatable(p)) {
1203: append_jump();
1204: return static_prim;
1205: }
1206: old_code_here = append_prim(p);
1207: last_jump = (priminfos[p].superend) ? 0 : p;
1208: return (Cell)old_code_here;
1209: #endif /* !defined(NO_DYNAMIC) */
1210: }
1211:
1212: #ifndef NO_DYNAMIC
1213: int cost_codesize(int prim)
1214: {
1215: return priminfos[prim].length;
1216: }
1217: #endif
1218:
1219: int cost_ls(int prim)
1220: {
1221: struct cost *c = super_costs+prim;
1222:
1223: return c->loads + c->stores;
1224: }
1225:
1226: int cost_lsu(int prim)
1227: {
1228: struct cost *c = super_costs+prim;
1229:
1230: return c->loads + c->stores + c->updates;
1231: }
1232:
1233: int cost_nexts(int prim)
1234: {
1235: return 1;
1236: }
1237:
1238: typedef int Costfunc(int);
1239: Costfunc *ss_cost = /* cost function for optimize_bb */
1240: #ifdef NO_DYNAMIC
1241: cost_lsu;
1242: #else
1243: cost_codesize;
1244: #endif
1245:
1246: struct {
1247: Costfunc *costfunc;
1248: char *metricname;
1249: long sum;
1250: } cost_sums[] = {
1251: #ifndef NO_DYNAMIC
1252: { cost_codesize, "codesize", 0 },
1253: #endif
1254: { cost_ls, "ls", 0 },
1255: { cost_lsu, "lsu", 0 },
1256: { cost_nexts, "nexts", 0 }
1257: };
1258:
1259: #define MAX_BB 128 /* maximum number of instructions in BB */
1260: #define INF_COST 1000000 /* infinite cost */
1261: #define CANONICAL_STATE 0
1262:
1263: struct waypoint {
1264: int cost; /* the cost from here to the end */
1265: PrimNum inst; /* the inst used from here to the next waypoint */
1266: char relocatable; /* the last non-transition was relocatable */
1267: char no_transition; /* don't use the next transition (relocatability)
1268: * or this transition (does not change state) */
1269: };
1270:
1271: void init_waypoints(struct waypoint ws[])
1272: {
1273: int k;
1274:
1275: for (k=0; k<maxstates; k++)
1276: ws[k].cost=INF_COST;
1277: }
1278:
1279: void transitions(struct waypoint inst[], struct waypoint trans[])
1280: {
1281: int k;
1282: struct super_state *l;
1283:
1284: for (k=0; k<maxstates; k++) {
1285: trans[k] = inst[k];
1286: trans[k].no_transition = 1;
1287: }
1288: for (l = state_transitions; l != NULL; l = l->next) {
1289: PrimNum s = l->super;
1290: int jcost;
1291: struct cost *c=super_costs+s;
1292: struct waypoint *wi=&(trans[c->state_in]);
1293: struct waypoint *wo=&(inst[c->state_out]);
1294: if (wo->cost == INF_COST)
1295: continue;
1296: jcost = wo->cost + ss_cost(s);
1297: if (jcost <= wi->cost) {
1298: wi->cost = jcost;
1299: wi->inst = s;
1300: wi->relocatable = wo->relocatable;
1301: wi->no_transition = 0;
1302: /* if (ss_greedy) wi->cost = wo->cost ? */
1303: }
1304: }
1305: }
1306:
1307: /* use dynamic programming to find the shortest paths within the basic
1308: block origs[0..ninsts-1] and rewrite the instructions pointed to by
1309: instps to use it */
1310: void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
1311: {
1312: int i,j;
1313: static struct waypoint inst[MAX_BB+1][MAX_STATE]; /* before instruction*/
1314: static struct waypoint trans[MAX_BB+1][MAX_STATE]; /* before transition */
1315: int nextdyn, nextstate, no_transition;
1316:
1317: init_waypoints(inst[ninsts]);
1318: inst[ninsts][CANONICAL_STATE].cost=0;
1319: transitions(inst[ninsts],trans[ninsts]);
1320: for (i=ninsts-1; i>=0; i--) {
1321: init_waypoints(inst[i]);
1322: for (j=1; j<=max_super && i+j<=ninsts; j++) {
1323: struct super_state **superp = lookup_super(origs+i, j);
1324: if (superp!=NULL) {
1325: struct super_state *supers = *superp;
1326: for (; supers!=NULL; supers = supers->next) {
1327: PrimNum s = supers->super;
1328: int jcost;
1329: struct cost *c=super_costs+s;
1330: struct waypoint *wi=&(inst[i][c->state_in]);
1331: struct waypoint *wo=&(trans[i+j][c->state_out]);
1332: int no_transition = wo->no_transition;
1333: if (!(is_relocatable(s)) && !wo->relocatable) {
1334: wo=&(inst[i+j][c->state_out]);
1335: no_transition=1;
1336: }
1337: if (wo->cost == INF_COST)
1338: continue;
1339: jcost = wo->cost + ss_cost(s);
1340: if (jcost <= wi->cost) {
1341: wi->cost = jcost;
1342: wi->inst = s;
1343: wi->relocatable = is_relocatable(s);
1344: wi->no_transition = no_transition;
1345: /* if (ss_greedy) wi->cost = wo->cost ? */
1346: }
1347: }
1348: }
1349: }
1350: transitions(inst[i],trans[i]);
1351: }
1352: /* now rewrite the instructions */
1353: nextdyn=0;
1354: nextstate=CANONICAL_STATE;
1355: no_transition = ((!trans[0][nextstate].relocatable)
1356: ||trans[0][nextstate].no_transition);
1357: for (i=0; i<ninsts; i++) {
1358: Cell tc=0, tc2;
1359: if (i==nextdyn) {
1360: if (!no_transition) {
1361: /* process trans */
1362: PrimNum p = trans[i][nextstate].inst;
1363: struct cost *c = super_costs+p;
1364: assert(trans[i][nextstate].cost != INF_COST);
1365: assert(c->state_in==nextstate);
1366: tc = compile_prim_dyn(p);
1367: nextstate = c->state_out;
1368: }
1369: {
1370: /* process inst */
1371: PrimNum p = inst[i][nextstate].inst;
1372: struct cost *c=super_costs+p;
1373: assert(c->state_in==nextstate);
1374: assert(inst[i][nextstate].cost != INF_COST);
1375: #if defined(GFORTH_DEBUGGING)
1376: assert(p == origs[i]);
1377: #endif
1378: tc2 = compile_prim_dyn(p);
1379: if (no_transition || !is_relocatable(p))
1380: /* !! actually what we care about is if and where
1381: * compile_prim_dyn() puts NEXTs */
1382: tc=tc2;
1383: no_transition = inst[i][nextstate].no_transition;
1384: nextstate = c->state_out;
1385: nextdyn += c->length;
1386: }
1387: } else {
1388: #if defined(GFORTH_DEBUGGING)
1389: assert(0);
1390: #endif
1391: tc=0;
1392: /* tc= (Cell)vm_prims[inst[i][CANONICAL_STATE].inst]; */
1393: }
1394: *(instps[i]) = tc;
1395: }
1396: if (!no_transition) {
1397: PrimNum p = trans[i][nextstate].inst;
1398: struct cost *c = super_costs+p;
1399: assert(c->state_in==nextstate);
1400: assert(trans[i][nextstate].cost != INF_COST);
1401: assert(i==nextdyn);
1402: (void)compile_prim_dyn(p);
1403: nextstate = c->state_out;
1404: }
1405: assert(nextstate==CANONICAL_STATE);
1406: }
1407:
1408: /* rewrite the instructions pointed to by instps to use the
1409: superinstructions in optimals */
1410: static void rewrite_bb(Cell *instps[], PrimNum *optimals, int ninsts)
1411: {
1412: int i,j, nextdyn;
1413: Cell inst;
1414:
1415: for (i=0, nextdyn=0; i<ninsts; i++) {
1416: if (i==nextdyn) { /* compile dynamically */
1417: nextdyn += super_costs[optimals[i]].length;
1418: inst = compile_prim_dyn(optimals[i]);
1419: for (j=0; j<sizeof(cost_sums)/sizeof(cost_sums[0]); j++)
1420: cost_sums[j].sum += cost_sums[j].costfunc(optimals[i]);
1421: } else { /* compile statically */
1422: inst = (Cell)vm_prims[optimals[i]];
1423: }
1424: *(instps[i]) = inst;
1425: }
1426: }
1427:
1428: /* compile *start, possibly rewriting it into a static and/or dynamic
1429: superinstruction */
1430: void compile_prim1(Cell *start)
1431: {
1432: #if defined(DOUBLY_INDIRECT)
1433: Label prim;
1434:
1435: if (start==NULL)
1436: return;
1437: prim = (Label)*start;
1438: if (prim<((Label)(xts+DOESJUMP)) || prim>((Label)(xts+npriminfos))) {
1439: fprintf(stderr,"compile_prim encountered xt %p\n", prim);
1440: *start=(Cell)prim;
1441: return;
1442: } else {
1443: *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
1444: return;
1445: }
1446: #elif defined(INDIRECT_THREADED)
1447: return;
1448: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1449: static Cell *instps[MAX_BB];
1450: static PrimNum origs[MAX_BB];
1451: static int ninsts=0;
1452: PrimNum prim_num;
1453:
1454: if (start==NULL)
1455: goto end_bb;
1456: prim_num = ((Xt)*start)-vm_prims;
1457: if (prim_num >= npriminfos)
1458: goto end_bb;
1459: assert(ninsts<MAX_BB);
1460: instps[ninsts] = start;
1461: origs[ninsts] = prim_num;
1462: ninsts++;
1463: if (ninsts >= MAX_BB || superend[prim_num]) {
1464: end_bb:
1465: optimize_rewrite(instps,origs,ninsts);
1466: ninsts=0;
1467: }
1468: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1469: }
1470:
1471: Address loader(FILE *imagefile, char* filename)
1472: /* returns the address of the image proper (after the preamble) */
1473: {
1474: ImageHeader header;
1475: Address image;
1476: Address imp; /* image+preamble */
1477: Char magic[8];
1478: char magic7; /* size byte of magic number */
1479: Cell preamblesize=0;
1480: Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1481: UCell check_sum;
1482: Cell ausize = ((RELINFOBITS == 8) ? 0 :
1483: (RELINFOBITS == 16) ? 1 :
1484: (RELINFOBITS == 32) ? 2 : 3);
1485: Cell charsize = ((sizeof(Char) == 1) ? 0 :
1486: (sizeof(Char) == 2) ? 1 :
1487: (sizeof(Char) == 4) ? 2 : 3) + ausize;
1488: Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
1489: (sizeof(Cell) == 2) ? 1 :
1490: (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1491: Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
1492: #ifdef WORDS_BIGENDIAN
1493: 0
1494: #else
1495: 1
1496: #endif
1497: ;
1498:
1499: vm_prims = engine(0,0,0,0,0);
1500: check_prims(vm_prims);
1501: prepare_super_table();
1502: #ifndef DOUBLY_INDIRECT
1503: #ifdef PRINT_SUPER_LENGTHS
1504: print_super_lengths();
1505: #endif
1506: check_sum = checksum(vm_prims);
1507: #else /* defined(DOUBLY_INDIRECT) */
1508: check_sum = (UCell)vm_prims;
1509: #endif /* defined(DOUBLY_INDIRECT) */
1510:
1511: do {
1512: if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1513: fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
1514: progname, filename);
1515: exit(1);
1516: }
1517: preamblesize+=8;
1518: } while(memcmp(magic,"Gforth3",7));
1519: magic7 = magic[7];
1520: if (debug) {
1521: magic[7]='\0';
1522: fprintf(stderr,"Magic found: %s ", magic);
1523: print_sizes(magic7);
1524: }
1525:
1526: if (magic7 != sizebyte)
1527: {
1528: fprintf(stderr,"This image is: ");
1529: print_sizes(magic7);
1530: fprintf(stderr,"whereas the machine is ");
1531: print_sizes(sizebyte);
1532: exit(-2);
1533: };
1534:
1535: fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1536:
1537: set_stack_sizes(&header);
1538:
1539: #if HAVE_GETPAGESIZE
1540: pagesize=getpagesize(); /* Linux/GNU libc offers this */
1541: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
1542: pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
1543: #elif PAGESIZE
1544: pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
1545: #endif
1546: if (debug)
1547: fprintf(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1548:
1549: image = dict_alloc_read(imagefile, preamblesize+header.image_size,
1550: preamblesize+dictsize, data_offset);
1551: imp=image+preamblesize;
1552: alloc_stacks((ImageHeader *)imp);
1553: if (clear_dictionary)
1554: memset(imp+header.image_size, 0, dictsize-header.image_size);
1555: if(header.base==0 || header.base == (Address)0x100) {
1556: Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
1557: char reloc_bits[reloc_size];
1558: fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1559: fread(reloc_bits, 1, reloc_size, imagefile);
1560: relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1561: #if 0
1562: { /* let's see what the relocator did */
1563: FILE *snapshot=fopen("snapshot.fi","wb");
1564: fwrite(image,1,imagesize,snapshot);
1565: fclose(snapshot);
1566: }
1567: #endif
1568: }
1569: else if(header.base!=imp) {
1570: fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
1571: progname, (unsigned long)header.base, (unsigned long)imp);
1572: exit(1);
1573: }
1574: if (header.checksum==0)
1575: ((ImageHeader *)imp)->checksum=check_sum;
1576: else if (header.checksum != check_sum) {
1577: fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
1578: progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
1579: exit(1);
1580: }
1581: #ifdef DOUBLY_INDIRECT
1582: ((ImageHeader *)imp)->xt_base = xts;
1583: #endif
1584: fclose(imagefile);
1585:
1586: /* unnecessary, except maybe for CODE words */
1587: /* FLUSH_ICACHE(imp, header.image_size);*/
1588:
1589: return imp;
1590: }
1591:
1592: /* pointer to last '/' or '\' in file, 0 if there is none. */
1593: char *onlypath(char *filename)
1594: {
1595: return strrchr(filename, DIRSEP);
1596: }
1597:
1598: FILE *openimage(char *fullfilename)
1599: {
1600: FILE *image_file;
1601: char * expfilename = tilde_cstr(fullfilename, strlen(fullfilename), 1);
1602:
1603: image_file=fopen(expfilename,"rb");
1604: if (image_file!=NULL && debug)
1605: fprintf(stderr, "Opened image file: %s\n", expfilename);
1606: return image_file;
1607: }
1608:
1609: /* try to open image file concat(path[0:len],imagename) */
1610: FILE *checkimage(char *path, int len, char *imagename)
1611: {
1612: int dirlen=len;
1613: char fullfilename[dirlen+strlen(imagename)+2];
1614:
1615: memcpy(fullfilename, path, dirlen);
1616: if (fullfilename[dirlen-1]!=DIRSEP)
1617: fullfilename[dirlen++]=DIRSEP;
1618: strcpy(fullfilename+dirlen,imagename);
1619: return openimage(fullfilename);
1620: }
1621:
1622: FILE * open_image_file(char * imagename, char * path)
1623: {
1624: FILE * image_file=NULL;
1625: char *origpath=path;
1626:
1627: if(strchr(imagename, DIRSEP)==NULL) {
1628: /* first check the directory where the exe file is in !! 01may97jaw */
1629: if (onlypath(progname))
1630: image_file=checkimage(progname, onlypath(progname)-progname, imagename);
1631: if (!image_file)
1632: do {
1633: char *pend=strchr(path, PATHSEP);
1634: if (pend==NULL)
1635: pend=path+strlen(path);
1636: if (strlen(path)==0) break;
1637: image_file=checkimage(path, pend-path, imagename);
1638: path=pend+(*pend==PATHSEP);
1639: } while (image_file==NULL);
1640: } else {
1641: image_file=openimage(imagename);
1642: }
1643:
1644: if (!image_file) {
1645: fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
1646: progname, imagename, origpath);
1647: exit(1);
1648: }
1649:
1650: return image_file;
1651: }
1652: #endif
1653:
1654: #ifdef HAS_OS
1655: UCell convsize(char *s, UCell elemsize)
1656: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
1657: of bytes. the letter at the end indicates the unit, where e stands
1658: for the element size. default is e */
1659: {
1660: char *endp;
1661: UCell n,m;
1662:
1663: m = elemsize;
1664: n = strtoul(s,&endp,0);
1665: if (endp!=NULL) {
1666: if (strcmp(endp,"b")==0)
1667: m=1;
1668: else if (strcmp(endp,"k")==0)
1669: m=1024;
1670: else if (strcmp(endp,"M")==0)
1671: m=1024*1024;
1672: else if (strcmp(endp,"G")==0)
1673: m=1024*1024*1024;
1674: else if (strcmp(endp,"T")==0) {
1675: #if (SIZEOF_CHAR_P > 4)
1676: m=1024L*1024*1024*1024;
1677: #else
1678: fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
1679: exit(1);
1680: #endif
1681: } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
1682: fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
1683: exit(1);
1684: }
1685: }
1686: return n*m;
1687: }
1688:
1689: enum {
1690: ss_number = 256,
1691: ss_states,
1692: ss_min_codesize,
1693: ss_min_ls,
1694: ss_min_lsu,
1695: ss_min_nexts,
1696: };
1697:
1698: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
1699: {
1700: int c;
1701:
1702: opterr=0;
1703: while (1) {
1704: int option_index=0;
1705: static struct option opts[] = {
1706: {"appl-image", required_argument, NULL, 'a'},
1707: {"image-file", required_argument, NULL, 'i'},
1708: {"dictionary-size", required_argument, NULL, 'm'},
1709: {"data-stack-size", required_argument, NULL, 'd'},
1710: {"return-stack-size", required_argument, NULL, 'r'},
1711: {"fp-stack-size", required_argument, NULL, 'f'},
1712: {"locals-stack-size", required_argument, NULL, 'l'},
1713: {"path", required_argument, NULL, 'p'},
1714: {"version", no_argument, NULL, 'v'},
1715: {"help", no_argument, NULL, 'h'},
1716: /* put something != 0 into offset_image */
1717: {"offset-image", no_argument, &offset_image, 1},
1718: {"no-offset-im", no_argument, &offset_image, 0},
1719: {"clear-dictionary", no_argument, &clear_dictionary, 1},
1720: {"die-on-signal", no_argument, &die_on_signal, 1},
1721: {"debug", no_argument, &debug, 1},
1722: {"no-super", no_argument, &no_super, 1},
1723: {"no-dynamic", no_argument, &no_dynamic, 1},
1724: {"dynamic", no_argument, &no_dynamic, 0},
1725: {"print-metrics", no_argument, &print_metrics, 1},
1726: {"ss-number", required_argument, NULL, ss_number},
1727: {"ss-states", required_argument, NULL, ss_states},
1728: #ifndef NO_DYNAMIC
1729: {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
1730: #endif
1731: {"ss-min-ls", no_argument, NULL, ss_min_ls},
1732: {"ss-min-lsu", no_argument, NULL, ss_min_lsu},
1733: {"ss-min-nexts", no_argument, NULL, ss_min_nexts},
1734: {"ss-greedy", no_argument, &ss_greedy, 1},
1735: {0,0,0,0}
1736: /* no-init-file, no-rc? */
1737: };
1738:
1739: c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
1740:
1741: switch (c) {
1742: case EOF: return;
1743: case '?': optind--; return;
1744: case 'a': *imagename = optarg; return;
1745: case 'i': *imagename = optarg; break;
1746: case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
1747: case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
1748: case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
1749: case 'f': fsize = convsize(optarg,sizeof(Float)); break;
1750: case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
1751: case 'p': *path = optarg; break;
1752: case 'o': offset_image = 1; break;
1753: case 'n': offset_image = 0; break;
1754: case 'c': clear_dictionary = 1; break;
1755: case 's': die_on_signal = 1; break;
1756: case 'x': debug = 1; break;
1757: case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
1758: case ss_number: static_super_number = atoi(optarg); break;
1759: case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
1760: #ifndef NO_DYNAMIC
1761: case ss_min_codesize: ss_cost = cost_codesize; break;
1762: #endif
1763: case ss_min_ls: ss_cost = cost_ls; break;
1764: case ss_min_lsu: ss_cost = cost_lsu; break;
1765: case ss_min_nexts: ss_cost = cost_nexts; break;
1766: case 'h':
1767: fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
1768: Engine Options:\n\
1769: --appl-image FILE equivalent to '--image-file=FILE --'\n\
1770: --clear-dictionary Initialize the dictionary with 0 bytes\n\
1771: -d SIZE, --data-stack-size=SIZE Specify data stack size\n\
1772: --debug Print debugging information during startup\n\
1773: --die-on-signal exit instead of CATCHing some signals\n\
1774: --dynamic use dynamic native code\n\
1775: -f SIZE, --fp-stack-size=SIZE Specify floating point stack size\n\
1776: -h, --help Print this message and exit\n\
1777: -i FILE, --image-file=FILE Use image FILE instead of `gforth.fi'\n\
1778: -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
1779: -m SIZE, --dictionary-size=SIZE Specify Forth dictionary size\n\
1780: --no-dynamic Use only statically compiled primitives\n\
1781: --no-offset-im Load image at normal position\n\
1782: --no-super No dynamically formed superinstructions\n\
1783: --offset-image Load image at a different position\n\
1784: -p PATH, --path=PATH Search path for finding image and sources\n\
1785: --print-metrics Print some code generation metrics on exit\n\
1786: -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
1787: --ss-greedy greedy, not optimal superinst selection\n\
1788: --ss-min-codesize select superinsts for smallest native code\n\
1789: --ss-min-ls minimize loads and stores\n\
1790: --ss-min-lsu minimize loads, stores, and pointer updates\n\
1791: --ss-min-nexts minimize the number of static superinsts\n\
1792: --ss-number=N use N static superinsts (default max)\n\
1793: --ss-states=N N states for stack caching (default max)\n\
1794: -v, --version Print engine version and exit\n\
1795: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
1796: `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
1797: argv[0]);
1798: optind--;
1799: return;
1800: }
1801: }
1802: }
1803: #endif
1804:
1805: #ifdef INCLUDE_IMAGE
1806: extern Cell image[];
1807: extern const char reloc_bits[];
1808: #endif
1809:
1810: int main(int argc, char **argv, char **env)
1811: {
1812: #ifdef HAS_OS
1813: char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
1814: #else
1815: char *path = DEFAULTPATH;
1816: #endif
1817: #ifndef INCLUDE_IMAGE
1818: char *imagename="gforth.fi";
1819: FILE *image_file;
1820: Address image;
1821: #endif
1822: int retvalue;
1823:
1824: #if defined(i386) && defined(ALIGNMENT_CHECK)
1825: /* turn on alignment checks on the 486.
1826: * on the 386 this should have no effect. */
1827: __asm__("pushfl; popl %eax; orl $0x40000, %eax; pushl %eax; popfl;");
1828: /* this is unusable with Linux' libc.4.6.27, because this library is
1829: not alignment-clean; we would have to replace some library
1830: functions (e.g., memcpy) to make it work. Also GCC doesn't try to keep
1831: the stack FP-aligned. */
1832: #endif
1833:
1834: /* buffering of the user output device */
1835: #ifdef _IONBF
1836: if (isatty(fileno(stdout))) {
1837: fflush(stdout);
1838: setvbuf(stdout,NULL,_IONBF,0);
1839: }
1840: #endif
1841:
1842: progname = argv[0];
1843:
1844: #ifdef HAS_OS
1845: gforth_args(argc, argv, &path, &imagename);
1846: #ifndef NO_DYNAMIC
1847: if (no_dynamic && ss_cost == cost_codesize) {
1848: ss_cost = cost_nexts;
1849: cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
1850: if (debug)
1851: fprintf(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
1852: }
1853: #endif /* !defined(NO_DYNAMIC) */
1854: #endif /* defined(HAS_OS) */
1855:
1856: #ifdef INCLUDE_IMAGE
1857: set_stack_sizes((ImageHeader *)image);
1858: if(((ImageHeader *)image)->base != image)
1859: relocate(image, reloc_bits, ((ImageHeader *)image)->image_size,
1860: (Label*)engine(0, 0, 0, 0, 0));
1861: alloc_stacks((ImageHeader *)image);
1862: #else
1863: image_file = open_image_file(imagename, path);
1864: image = loader(image_file, imagename);
1865: #endif
1866: gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
1867:
1868: {
1869: char path2[strlen(path)+1];
1870: char *p1, *p2;
1871: Cell environ[]= {
1872: (Cell)argc-(optind-1),
1873: (Cell)(argv+(optind-1)),
1874: (Cell)strlen(path),
1875: (Cell)path2};
1876: argv[optind-1] = progname;
1877: /*
1878: for (i=0; i<environ[0]; i++)
1879: printf("%s\n", ((char **)(environ[1]))[i]);
1880: */
1881: /* make path OS-independent by replacing path separators with NUL */
1882: for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
1883: if (*p1==PATHSEP)
1884: *p2 = '\0';
1885: else
1886: *p2 = *p1;
1887: *p2='\0';
1888: retvalue = go_forth(image, 4, environ);
1889: #ifdef SIGPIPE
1890: bsd_signal(SIGPIPE, SIG_IGN);
1891: #endif
1892: #ifdef VM_PROFILING
1893: vm_print_profile(stderr);
1894: #endif
1895: deprep_terminal();
1896: }
1897: if (print_metrics) {
1898: int i;
1899: fprintf(stderr, "code size = %8ld\n", dyncodesize());
1900: for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
1901: fprintf(stderr, "metric %8s: %8ld\n",
1902: cost_sums[i].metricname, cost_sums[i].sum);
1903: }
1904: return retvalue;
1905: }
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>
![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to main.c CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [gforth] / gforth / engine