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 (4096*1024) /* !! overflow handling for -native */
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-1)/sizeof(Cell))/RELINFOBITS)+1;
271: Cell token;
272: unsigned char bits;
273: unsigned char *result=malloc(steps);
274:
275: memset(result, 0, steps);
276: for(k=0; k<steps; k++) {
277: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
278: if(bits & (1U << (RELINFOBITS-1))) {
279: assert(i*sizeof(Cell) < size);
280: token=image[i];
281: if (token>=base) { /* relocatable address */
282: UCell bitnum=(token-base)/sizeof(Cell);
283: result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
284: }
285: }
286: }
287: }
288: return result;
289: }
290:
291: void relocate(Cell *image, const unsigned char *bitstring,
292: int size, Cell base, Label symbols[])
293: {
294: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
295: Cell token;
296: char bits;
297: Cell max_symbols;
298: /*
299: * A virtual start address that's the real start address minus
300: * the one in the image
301: */
302: Cell *start = (Cell * ) (((void *) image) - ((void *) base));
303: unsigned char *targets = branch_targets(image, bitstring, size, base);
304:
305: /* group index into table */
306: if(groups[31]==0) {
307: int groupsum=0;
308: for(i=0; i<32; i++) {
309: groupsum += groups[i];
310: groups[i] = groupsum;
311: /* printf("group[%d]=%d\n",i,groupsum); */
312: }
313: i=0;
314: }
315:
316: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
317:
318: for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
319: ;
320: max_symbols--;
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(bits & (1U << (RELINFOBITS-1))) {
326: assert(i*sizeof(Cell) < size);
327: /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
328: token=image[i];
329: if(token<0) {
330: int group = (-token & 0x3E00) >> 9;
331: if(group == 0) {
332: switch(token|0x4000) {
333: case CF_NIL : image[i]=0; break;
334: #if !defined(DOUBLY_INDIRECT)
335: case CF(DOCOL) :
336: case CF(DOVAR) :
337: case CF(DOCON) :
338: case CF(DOUSER) :
339: case CF(DODEFER) :
340: case CF(DOFIELD) : MAKE_CF(image+i,symbols[CF(token)]); break;
341: case CF(DOESJUMP): image[i]=0; break;
342: #endif /* !defined(DOUBLY_INDIRECT) */
343: case CF(DODOES) :
344: MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
345: break;
346: default : /* backward compatibility */
347: /* printf("Code field generation image[%x]:=CFA(%x)\n",
348: i, CF(image[i])); */
349: if (CF((token | 0x4000))<max_symbols) {
350: image[i]=(Cell)CFA(CF(token));
351: #ifdef DIRECT_THREADED
352: if ((token & 0x4000) == 0) { /* threade code, no CFA */
353: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
354: compile_prim1(0);
355: compile_prim1(&image[i]);
356: }
357: #endif
358: } else
359: 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);
360: }
361: } else {
362: int tok = -token & 0x1FF;
363: if (tok < (groups[group+1]-groups[group])) {
364: #if defined(DOUBLY_INDIRECT)
365: image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
366: #else
367: image[i]=(Cell)CFA((groups[group]+tok));
368: #endif
369: #ifdef DIRECT_THREADED
370: if ((token & 0x4000) == 0) { /* threade code, no CFA */
371: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
372: compile_prim1(0);
373: compile_prim1(&image[i]);
374: }
375: #endif
376: } else
377: fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
378: }
379: } else {
380: /* if base is > 0: 0 is a null reference so don't adjust*/
381: if (token>=base) {
382: image[i]+=(Cell)start;
383: }
384: }
385: }
386: }
387: }
388: free(targets);
389: finish_code();
390: ((ImageHeader*)(image))->base = (Address) image;
391: }
392:
393: UCell checksum(Label symbols[])
394: {
395: UCell r=PRIM_VERSION;
396: Cell i;
397:
398: for (i=DOCOL; i<=DOESJUMP; i++) {
399: r ^= (UCell)(symbols[i]);
400: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
401: }
402: #ifdef DIRECT_THREADED
403: /* we have to consider all the primitives */
404: for (; symbols[i]!=(Label)0; i++) {
405: r ^= (UCell)(symbols[i]);
406: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
407: }
408: #else
409: /* in indirect threaded code all primitives are accessed through the
410: symbols table, so we just have to put the base address of symbols
411: in the checksum */
412: r ^= (UCell)symbols;
413: #endif
414: return r;
415: }
416:
417: Address verbose_malloc(Cell size)
418: {
419: Address r;
420: /* leave a little room (64B) for stack underflows */
421: if ((r = malloc(size+64))==NULL) {
422: perror(progname);
423: exit(1);
424: }
425: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
426: if (debug)
427: fprintf(stderr, "malloc succeeds, address=$%lx\n", (long)r);
428: return r;
429: }
430:
431: static Address next_address=0;
432: void after_alloc(Address r, Cell size)
433: {
434: if (r != (Address)-1) {
435: if (debug)
436: fprintf(stderr, "success, address=$%lx\n", (long) r);
437: if (pagesize != 1)
438: next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
439: } else {
440: if (debug)
441: fprintf(stderr, "failed: %s\n", strerror(errno));
442: }
443: }
444:
445: #ifndef MAP_FAILED
446: #define MAP_FAILED ((Address) -1)
447: #endif
448: #ifndef MAP_FILE
449: # define MAP_FILE 0
450: #endif
451: #ifndef MAP_PRIVATE
452: # define MAP_PRIVATE 0
453: #endif
454: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
455: # define MAP_ANON MAP_ANONYMOUS
456: #endif
457:
458: #if defined(HAVE_MMAP)
459: static Address alloc_mmap(Cell size)
460: {
461: Address r;
462:
463: #if defined(MAP_ANON)
464: if (debug)
465: fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
466: r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE, -1, 0);
467: #else /* !defined(MAP_ANON) */
468: /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
469: apparently defaults) */
470: static int dev_zero=-1;
471:
472: if (dev_zero == -1)
473: dev_zero = open("/dev/zero", O_RDONLY);
474: if (dev_zero == -1) {
475: r = MAP_FAILED;
476: if (debug)
477: fprintf(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ",
478: strerror(errno));
479: } else {
480: if (debug)
481: fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
482: r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE, dev_zero, 0);
483: }
484: #endif /* !defined(MAP_ANON) */
485: after_alloc(r, size);
486: return r;
487: }
488: #endif
489:
490: Address my_alloc(Cell size)
491: {
492: #if HAVE_MMAP
493: Address r;
494:
495: r=alloc_mmap(size);
496: if (r!=(Address)MAP_FAILED)
497: return r;
498: #endif /* HAVE_MMAP */
499: /* use malloc as fallback */
500: return verbose_malloc(size);
501: }
502:
503: Address dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
504: {
505: Address image = MAP_FAILED;
506:
507: #if defined(HAVE_MMAP)
508: if (offset==0) {
509: image=alloc_mmap(dictsize);
510: if (debug)
511: fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
512: image = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE, fileno(file), 0);
513: after_alloc(image,dictsize);
514: }
515: #endif /* defined(HAVE_MMAP) */
516: if (image == (Address)MAP_FAILED) {
517: image = my_alloc(dictsize+offset)+offset;
518: rewind(file); /* fseek(imagefile,0L,SEEK_SET); */
519: fread(image, 1, imagesize, file);
520: }
521: return image;
522: }
523:
524: void set_stack_sizes(ImageHeader * header)
525: {
526: if (dictsize==0)
527: dictsize = header->dict_size;
528: if (dsize==0)
529: dsize = header->data_stack_size;
530: if (rsize==0)
531: rsize = header->return_stack_size;
532: if (fsize==0)
533: fsize = header->fp_stack_size;
534: if (lsize==0)
535: lsize = header->locals_stack_size;
536: dictsize=maxaligned(dictsize);
537: dsize=maxaligned(dsize);
538: rsize=maxaligned(rsize);
539: lsize=maxaligned(lsize);
540: fsize=maxaligned(fsize);
541: }
542:
543: void alloc_stacks(ImageHeader * header)
544: {
545: header->dict_size=dictsize;
546: header->data_stack_size=dsize;
547: header->fp_stack_size=fsize;
548: header->return_stack_size=rsize;
549: header->locals_stack_size=lsize;
550:
551: header->data_stack_base=my_alloc(dsize);
552: header->fp_stack_base=my_alloc(fsize);
553: header->return_stack_base=my_alloc(rsize);
554: header->locals_stack_base=my_alloc(lsize);
555: }
556:
557: #warning You can ignore the warnings about clobbered variables in go_forth
558: int go_forth(Address image, int stack, Cell *entries)
559: {
560: volatile ImageHeader *image_header = (ImageHeader *)image;
561: Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
562: Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
563: Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
564: #ifdef GFORTH_DEBUGGING
565: volatile Cell *orig_rp0=rp0;
566: #endif
567: Address lp0=image_header->locals_stack_base + lsize;
568: Xt *ip0=(Xt *)(image_header->boot_entry);
569: #ifdef SYSSIGNALS
570: int throw_code;
571: #endif
572:
573: /* ensure that the cached elements (if any) are accessible */
574: IF_spTOS(sp0--);
575: IF_fpTOS(fp0--);
576:
577: for(;stack>0;stack--)
578: *--sp0=entries[stack-1];
579:
580: #ifdef SYSSIGNALS
581: get_winsize();
582:
583: install_signal_handlers(); /* right place? */
584:
585: if ((throw_code=setjmp(throw_jmp_buf))) {
586: static Cell signal_data_stack[8];
587: static Cell signal_return_stack[8];
588: static Float signal_fp_stack[1];
589:
590: signal_data_stack[7]=throw_code;
591:
592: #ifdef GFORTH_DEBUGGING
593: if (debug)
594: fprintf(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
595: throw_code, saved_ip, rp);
596: if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
597: /* no rstack overflow or underflow */
598: rp0 = rp;
599: *--rp0 = (Cell)saved_ip;
600: }
601: else /* I love non-syntactic ifdefs :-) */
602: rp0 = signal_return_stack+8;
603: #else /* !defined(GFORTH_DEBUGGING) */
604: if (debug)
605: fprintf(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
606: rp0 = signal_return_stack+8;
607: #endif /* !defined(GFORTH_DEBUGGING) */
608: /* fprintf(stderr, "rp=$%x\n",rp0);*/
609:
610: return((int)(Cell)engine(image_header->throw_entry, signal_data_stack+7,
611: rp0, signal_fp_stack, 0));
612: }
613: #endif
614:
615: return((int)(Cell)engine(ip0,sp0,rp0,fp0,lp0));
616: }
617:
618: #ifndef INCLUDE_IMAGE
619: void print_sizes(Cell sizebyte)
620: /* print size information */
621: {
622: static char* endianstring[]= { " big","little" };
623:
624: fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
625: endianstring[sizebyte & 1],
626: 1 << ((sizebyte >> 1) & 3),
627: 1 << ((sizebyte >> 3) & 3),
628: 1 << ((sizebyte >> 5) & 3));
629: }
630:
631: /* static superinstruction stuff */
632:
633: struct cost {
634: char loads; /* number of stack loads */
635: char stores; /* number of stack stores */
636: char updates; /* number of stack pointer updates */
637: char branch; /* is it a branch (SET_IP) */
638: unsigned char state_in; /* state on entry */
639: unsigned char state_out; /* state on exit */
640: short offset; /* offset into super2 table */
641: unsigned char length; /* number of components */
642: };
643:
644: PrimNum super2[] = {
645: #include SUPER2_I
646: };
647:
648: struct cost super_costs[] = {
649: #include COSTS_I
650: };
651:
652: struct super_state {
653: struct super_state *next;
654: PrimNum super;
655: };
656:
657: #define HASH_SIZE 256
658:
659: struct super_table_entry {
660: struct super_table_entry *next;
661: PrimNum *start;
662: short length;
663: struct super_state *ss_list; /* list of supers */
664: } *super_table[HASH_SIZE];
665: int max_super=2;
666:
667: struct super_state *state_transitions=NULL;
668:
669: int hash_super(PrimNum *start, int length)
670: {
671: int i, r;
672:
673: for (i=0, r=0; i<length; i++) {
674: r <<= 1;
675: r += start[i];
676: }
677: return r & (HASH_SIZE-1);
678: }
679:
680: struct super_state **lookup_super(PrimNum *start, int length)
681: {
682: int hash=hash_super(start,length);
683: struct super_table_entry *p = super_table[hash];
684:
685: /* assert(length >= 2); */
686: for (; p!=NULL; p = p->next) {
687: if (length == p->length &&
688: memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
689: return &(p->ss_list);
690: }
691: return NULL;
692: }
693:
694: void prepare_super_table()
695: {
696: int i;
697: int nsupers = 0;
698:
699: for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
700: struct cost *c = &super_costs[i];
701: if ((c->length < 2 || nsupers < static_super_number) &&
702: c->state_in < maxstates && c->state_out < maxstates) {
703: struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
704: struct super_state *ss = malloc(sizeof(struct super_state));
705: ss->super= i;
706: if (c->offset==N_noop && i != N_noop) {
707: if (is_relocatable(i)) {
708: ss->next = state_transitions;
709: state_transitions = ss;
710: }
711: } else if (ss_listp != NULL) {
712: ss->next = *ss_listp;
713: *ss_listp = ss;
714: } else {
715: int hash = hash_super(super2+c->offset, c->length);
716: struct super_table_entry **p = &super_table[hash];
717: struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
718: ss->next = NULL;
719: e->next = *p;
720: e->start = super2 + c->offset;
721: e->length = c->length;
722: e->ss_list = ss;
723: *p = e;
724: }
725: if (c->length > max_super)
726: max_super = c->length;
727: if (c->length >= 2)
728: nsupers++;
729: }
730: }
731: if (debug)
732: fprintf(stderr, "Using %d static superinsts\n", nsupers);
733: }
734:
735: /* dynamic replication/superinstruction stuff */
736:
737: #ifndef NO_DYNAMIC
738: int compare_priminfo_length(const void *_a, const void *_b)
739: {
740: PrimInfo **a = (PrimInfo **)_a;
741: PrimInfo **b = (PrimInfo **)_b;
742: Cell diff = (*a)->length - (*b)->length;
743: if (diff)
744: return diff;
745: else /* break ties by start address; thus the decompiler produces
746: the earliest primitive with the same code (e.g. noop instead
747: of (char) and @ instead of >code-address */
748: return (*b)->start - (*a)->start;
749: }
750: #endif /* !defined(NO_DYNAMIC) */
751:
752: static char MAYBE_UNUSED superend[]={
753: #include PRIM_SUPEREND_I
754: };
755:
756: Cell npriminfos=0;
757:
758: int compare_labels(const void *pa, const void *pb)
759: {
760: Label a = *(Label *)pa;
761: Label b = *(Label *)pb;
762: return a-b;
763: }
764:
765: Label bsearch_next(Label key, Label *a, UCell n)
766: /* a is sorted; return the label >=key that is the closest in a;
767: return NULL if there is no label in a >=key */
768: {
769: int mid = (n-1)/2;
770: if (n<1)
771: return NULL;
772: if (n == 1) {
773: if (a[0] < key)
774: return NULL;
775: else
776: return a[0];
777: }
778: if (a[mid] < key)
779: return bsearch_next(key, a+mid+1, n-mid-1);
780: else
781: return bsearch_next(key, a, mid+1);
782: }
783:
784: void check_prims(Label symbols1[])
785: {
786: int i;
787: #ifndef NO_DYNAMIC
788: Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted;
789: int nends1j;
790: #endif
791:
792: if (debug)
793: #ifdef __VERSION__
794: fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
795: #else
796: #define xstr(s) str(s)
797: #define str(s) #s
798: fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n");
799: #endif
800: for (i=0; symbols1[i]!=0; i++)
801: ;
802: npriminfos = i;
803:
804: #ifndef NO_DYNAMIC
805: if (no_dynamic)
806: return;
807: symbols2=engine2(0,0,0,0,0);
808: #if NO_IP
809: symbols3=engine3(0,0,0,0,0);
810: #else
811: symbols3=symbols1;
812: #endif
813: ends1 = symbols1+i+1;
814: ends1j = ends1+i;
815: nends1j = i+1;
816: ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
817: memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
818: qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
819:
820: priminfos = calloc(i,sizeof(PrimInfo));
821: for (i=0; symbols1[i]!=0; i++) {
822: int prim_len = ends1[i]-symbols1[i];
823: PrimInfo *pi=&priminfos[i];
824: int j=0;
825: char *s1 = (char *)symbols1[i];
826: char *s2 = (char *)symbols2[i];
827: char *s3 = (char *)symbols3[i];
828: Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
829:
830: pi->start = s1;
831: pi->superend = superend[i]|no_super;
832: if (pi->superend)
833: pi->length = endlabel-symbols1[i];
834: else
835: pi->length = prim_len;
836: pi->restlength = endlabel - symbols1[i] - pi->length;
837: pi->nimmargs = 0;
838: if (debug)
839: fprintf(stderr, "Prim %3d @ %p %p %p, length=%3ld restlength=%2ld superend=%1d",
840: i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength), pi->superend);
841: if (endlabel == NULL) {
842: pi->start = NULL; /* not relocatable */
843: if (pi->length<0) pi->length=100;
844: if (debug)
845: fprintf(stderr,"\n non_reloc: no J label > start found\n");
846: continue;
847: }
848: if (ends1[i] > endlabel && !pi->superend) {
849: pi->start = NULL; /* not relocatable */
850: pi->length = endlabel-symbols1[i];
851: if (debug)
852: fprintf(stderr,"\n non_reloc: there is a J label before the K label (restlength<0)\n");
853: continue;
854: }
855: if (ends1[i] < pi->start && !pi->superend) {
856: pi->start = NULL; /* not relocatable */
857: pi->length = endlabel-symbols1[i];
858: if (debug)
859: fprintf(stderr,"\n non_reloc: K label before I label (length<0)\n");
860: continue;
861: }
862: assert(pi->length>=0);
863: assert(pi->restlength >=0);
864: while (j<(pi->length+pi->restlength)) {
865: if (s1[j]==s3[j]) {
866: if (s1[j] != s2[j]) {
867: pi->start = NULL; /* not relocatable */
868: if (debug)
869: fprintf(stderr,"\n non_reloc: engine1!=engine2 offset %3d",j);
870: /* assert(j<prim_len); */
871: break;
872: }
873: j++;
874: } else {
875: struct immarg *ia=&pi->immargs[pi->nimmargs];
876:
877: pi->nimmargs++;
878: ia->offset=j;
879: if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
880: ia->rel=0;
881: if (debug)
882: fprintf(stderr,"\n absolute immarg: offset %3d",j);
883: } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
884: symbols1[DOESJUMP+1]) {
885: ia->rel=1;
886: if (debug)
887: fprintf(stderr,"\n relative immarg: offset %3d",j);
888: } else {
889: pi->start = NULL; /* not relocatable */
890: if (debug)
891: fprintf(stderr,"\n non_reloc: engine1!=engine3 offset %3d",j);
892: /* assert(j<prim_len);*/
893: break;
894: }
895: j+=4;
896: }
897: }
898: if (debug)
899: fprintf(stderr,"\n");
900: }
901: decomp_prims = calloc(i,sizeof(PrimInfo *));
902: for (i=DOESJUMP+1; i<npriminfos; i++)
903: decomp_prims[i] = &(priminfos[i]);
904: qsort(decomp_prims+DOESJUMP+1, npriminfos-DOESJUMP-1, sizeof(PrimInfo *),
905: compare_priminfo_length);
906: #endif
907: }
908:
909: void flush_to_here(void)
910: {
911: #ifndef NO_DYNAMIC
912: if (start_flush)
913: FLUSH_ICACHE(start_flush, code_here-start_flush);
914: start_flush=code_here;
915: #endif
916: }
917:
918: #ifndef NO_DYNAMIC
919: void append_jump(void)
920: {
921: if (last_jump) {
922: PrimInfo *pi = &priminfos[last_jump];
923:
924: memcpy(code_here, pi->start+pi->length, pi->restlength);
925: code_here += pi->restlength;
926: last_jump=0;
927: }
928: }
929:
930: /* Gforth remembers all code blocks in this list. On forgetting (by
931: executing a marker) the code blocks are not freed (because Gforth does
932: not remember how they were allocated; hmm, remembering that might be
933: easier and cleaner). Instead, code_here etc. are reset to the old
934: value, and the "forgotten" code blocks are reused when they are
935: needed. */
936:
937: struct code_block_list {
938: struct code_block_list *next;
939: Address block;
940: Cell size;
941: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
942:
943: Address append_prim(Cell p)
944: {
945: PrimInfo *pi = &priminfos[p];
946: Address old_code_here = code_here;
947:
948: if (code_area+code_area_size < code_here+pi->length+pi->restlength) {
949: struct code_block_list *p;
950: append_jump();
951: flush_to_here();
952: if (*next_code_blockp == NULL) {
953: code_here = start_flush = code_area = my_alloc(code_area_size);
954: p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
955: *next_code_blockp = p;
956: p->next = NULL;
957: p->block = code_here;
958: p->size = code_area_size;
959: } else {
960: p = *next_code_blockp;
961: code_here = start_flush = code_area = p->block;
962: }
963: old_code_here = code_here;
964: next_code_blockp = &(p->next);
965: }
966: memcpy(code_here, pi->start, pi->length);
967: code_here += pi->length;
968: return old_code_here;
969: }
970: #endif
971:
972: int forget_dyncode(Address code)
973: {
974: #ifdef NO_DYNAMIC
975: return -1;
976: #else
977: struct code_block_list *p, **pp;
978:
979: for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
980: if (code >= p->block && code < p->block+p->size) {
981: next_code_blockp = &(p->next);
982: code_here = start_flush = code;
983: code_area = p->block;
984: last_jump = 0;
985: return -1;
986: }
987: }
988: return -no_dynamic;
989: #endif /* !defined(NO_DYNAMIC) */
990: }
991:
992: long dyncodesize(void)
993: {
994: #ifndef NO_DYNAMIC
995: struct code_block_list *p;
996: long size=0;
997: for (p=code_block_list; p!=NULL; p=p->next) {
998: if (code_here >= p->block && code_here < p->block+p->size)
999: return size + (code_here - p->block);
1000: else
1001: size += p->size;
1002: }
1003: #endif /* !defined(NO_DYNAMIC) */
1004: return 0;
1005: }
1006:
1007: Label decompile_code(Label _code)
1008: {
1009: #ifdef NO_DYNAMIC
1010: return _code;
1011: #else /* !defined(NO_DYNAMIC) */
1012: Cell i;
1013: struct code_block_list *p;
1014: Address code=_code;
1015:
1016: /* first, check if we are in code at all */
1017: for (p = code_block_list;; p = p->next) {
1018: if (p == NULL)
1019: return code;
1020: if (code >= p->block && code < p->block+p->size)
1021: break;
1022: }
1023: /* reverse order because NOOP might match other prims */
1024: for (i=npriminfos-1; i>DOESJUMP; i--) {
1025: PrimInfo *pi=decomp_prims[i];
1026: if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1027: return vm_prims[super2[super_costs[pi-priminfos].offset]];
1028: /* return pi->start;*/
1029: }
1030: return code;
1031: #endif /* !defined(NO_DYNAMIC) */
1032: }
1033:
1034: #ifdef NO_IP
1035: int nbranchinfos=0;
1036:
1037: struct branchinfo {
1038: Label **targetpp; /* **(bi->targetpp) is the target */
1039: Cell *addressptr; /* store the target here */
1040: } branchinfos[100000];
1041:
1042: int ndoesexecinfos=0;
1043: struct doesexecinfo {
1044: int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1045: Label *targetp; /*target for branch (because this is not in threaded code)*/
1046: Cell *xt; /* cfa of word whose does-code needs calling */
1047: } doesexecinfos[10000];
1048:
1049: void set_rel_target(Cell *source, Label target)
1050: {
1051: *source = ((Cell)target)-(((Cell)source)+4);
1052: }
1053:
1054: void register_branchinfo(Label source, Cell *targetpp)
1055: {
1056: struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1057: bi->targetpp = (Label **)targetpp;
1058: bi->addressptr = (Cell *)source;
1059: nbranchinfos++;
1060: }
1061:
1062: Address compile_prim1arg(PrimNum p, Cell **argp)
1063: {
1064: Address old_code_here=append_prim(p);
1065:
1066: assert(vm_prims[p]==priminfos[p].start);
1067: *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
1068: return old_code_here;
1069: }
1070:
1071: Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
1072: {
1073: PrimInfo *pi = &priminfos[N_call2];
1074: Address old_code_here = append_prim(N_call2);
1075:
1076: *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
1077: register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
1078: return old_code_here;
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: dei->targetp = (Label *)DOES_CODE1((dei->xt));
1091: branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
1092: }
1093: ndoesexecinfos = 0;
1094: for (i=0; i<nbranchinfos; i++) {
1095: struct branchinfo *bi=&branchinfos[i];
1096: set_rel_target(bi->addressptr, **(bi->targetpp));
1097: }
1098: nbranchinfos = 0;
1099: #else
1100: compile_prim1(NULL);
1101: #endif
1102: flush_to_here();
1103: }
1104:
1105: #ifdef NO_IP
1106: Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1107: /* compile prim #p dynamically (mod flags etc.) and return start
1108: address of generated code for putting it into the threaded
1109: code. This function is only called if all the associated
1110: inline arguments of p are already in place (at tcp[1] etc.) */
1111: {
1112: PrimInfo *pi=&priminfos[p];
1113: Cell *next_code_target=NULL;
1114: Address codeaddr;
1115: Address primstart;
1116:
1117: assert(p<npriminfos);
1118: if (p==N_execute || p==N_perform || p==N_lit_perform) {
1119: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1120: primstart = append_prim(p);
1121: goto other_prim;
1122: } else if (p==N_call) {
1123: codeaddr = compile_call2(tcp+1, &next_code_target);
1124: } else if (p==N_does_exec) {
1125: struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1126: Cell *arg;
1127: codeaddr = compile_prim1arg(N_lit,&arg);
1128: *arg = (Cell)PFA(tcp[1]);
1129: /* we cannot determine the callee now (last_start[1] may be a
1130: forward reference), so just register an arbitrary target, and
1131: register in dei that we need to fix this before resolving
1132: branches */
1133: dei->branchinfo = nbranchinfos;
1134: dei->xt = (Cell *)(tcp[1]);
1135: compile_call2(0, &next_code_target);
1136: } else if (!is_relocatable(p)) {
1137: Cell *branch_target;
1138: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1139: compile_prim1arg(N_branch,&branch_target);
1140: set_rel_target(branch_target,vm_prims[p]);
1141: } else {
1142: unsigned j;
1143:
1144: codeaddr = primstart = append_prim(p);
1145: other_prim:
1146: for (j=0; j<pi->nimmargs; j++) {
1147: struct immarg *ia = &(pi->immargs[j]);
1148: Cell *argp = tcp + pi->nimmargs - j;
1149: Cell argval = *argp; /* !! specific to prims */
1150: if (ia->rel) { /* !! assumption: relative refs are branches */
1151: register_branchinfo(primstart + ia->offset, argp);
1152: } else /* plain argument */
1153: *(Cell *)(primstart + ia->offset) = argval;
1154: }
1155: }
1156: if (next_code_target!=NULL)
1157: *next_code_target = (Cell)code_here;
1158: return (Cell)codeaddr;
1159: }
1160: #else /* !defined(NO_IP) */
1161: Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1162: /* compile prim #p dynamically (mod flags etc.) and return start
1163: address of generated code for putting it into the threaded code */
1164: {
1165: Cell static_prim = (Cell)vm_prims[p];
1166: #if defined(NO_DYNAMIC)
1167: return static_prim;
1168: #else /* !defined(NO_DYNAMIC) */
1169: Address old_code_here;
1170:
1171: if (no_dynamic)
1172: return static_prim;
1173: if (p>=npriminfos || !is_relocatable(p)) {
1174: append_jump();
1175: return static_prim;
1176: }
1177: old_code_here = append_prim(p);
1178: last_jump = (priminfos[p].superend) ? 0 : p;
1179: return (Cell)old_code_here;
1180: #endif /* !defined(NO_DYNAMIC) */
1181: }
1182: #endif /* !defined(NO_IP) */
1183:
1184: #ifndef NO_DYNAMIC
1185: int cost_codesize(int prim)
1186: {
1187: return priminfos[prim].length;
1188: }
1189: #endif
1190:
1191: int cost_ls(int prim)
1192: {
1193: struct cost *c = super_costs+prim;
1194:
1195: return c->loads + c->stores;
1196: }
1197:
1198: int cost_lsu(int prim)
1199: {
1200: struct cost *c = super_costs+prim;
1201:
1202: return c->loads + c->stores + c->updates;
1203: }
1204:
1205: int cost_nexts(int prim)
1206: {
1207: return 1;
1208: }
1209:
1210: typedef int Costfunc(int);
1211: Costfunc *ss_cost = /* cost function for optimize_bb */
1212: #ifdef NO_DYNAMIC
1213: cost_lsu;
1214: #else
1215: cost_codesize;
1216: #endif
1217:
1218: struct {
1219: Costfunc *costfunc;
1220: char *metricname;
1221: long sum;
1222: } cost_sums[] = {
1223: #ifndef NO_DYNAMIC
1224: { cost_codesize, "codesize", 0 },
1225: #endif
1226: { cost_ls, "ls", 0 },
1227: { cost_lsu, "lsu", 0 },
1228: { cost_nexts, "nexts", 0 }
1229: };
1230:
1231: #define MAX_BB 128 /* maximum number of instructions in BB */
1232: #define INF_COST 1000000 /* infinite cost */
1233: #define CANONICAL_STATE 0
1234:
1235: struct waypoint {
1236: int cost; /* the cost from here to the end */
1237: PrimNum inst; /* the inst used from here to the next waypoint */
1238: char relocatable; /* the last non-transition was relocatable */
1239: char no_transition; /* don't use the next transition (relocatability)
1240: * or this transition (does not change state) */
1241: };
1242:
1243: void init_waypoints(struct waypoint ws[])
1244: {
1245: int k;
1246:
1247: for (k=0; k<maxstates; k++)
1248: ws[k].cost=INF_COST;
1249: }
1250:
1251: void transitions(struct waypoint inst[], struct waypoint trans[])
1252: {
1253: int k;
1254: struct super_state *l;
1255:
1256: for (k=0; k<maxstates; k++) {
1257: trans[k] = inst[k];
1258: trans[k].no_transition = 1;
1259: }
1260: for (l = state_transitions; l != NULL; l = l->next) {
1261: PrimNum s = l->super;
1262: int jcost;
1263: struct cost *c=super_costs+s;
1264: struct waypoint *wi=&(trans[c->state_in]);
1265: struct waypoint *wo=&(inst[c->state_out]);
1266: if (wo->cost == INF_COST)
1267: continue;
1268: jcost = wo->cost + ss_cost(s);
1269: if (jcost <= wi->cost) {
1270: wi->cost = jcost;
1271: wi->inst = s;
1272: wi->relocatable = wo->relocatable;
1273: wi->no_transition = 0;
1274: /* if (ss_greedy) wi->cost = wo->cost ? */
1275: }
1276: }
1277: }
1278:
1279: /* use dynamic programming to find the shortest paths within the basic
1280: block origs[0..ninsts-1] and rewrite the instructions pointed to by
1281: instps to use it */
1282: void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
1283: {
1284: int i,j;
1285: static struct waypoint inst[MAX_BB+1][MAX_STATE]; /* before instruction*/
1286: static struct waypoint trans[MAX_BB+1][MAX_STATE]; /* before transition */
1287: int nextdyn, nextstate, no_transition;
1288:
1289: init_waypoints(inst[ninsts]);
1290: inst[ninsts][CANONICAL_STATE].cost=0;
1291: transitions(inst[ninsts],trans[ninsts]);
1292: for (i=ninsts-1; i>=0; i--) {
1293: init_waypoints(inst[i]);
1294: for (j=1; j<=max_super && i+j<=ninsts; j++) {
1295: struct super_state **superp = lookup_super(origs+i, j);
1296: if (superp!=NULL) {
1297: struct super_state *supers = *superp;
1298: for (; supers!=NULL; supers = supers->next) {
1299: PrimNum s = supers->super;
1300: int jcost;
1301: struct cost *c=super_costs+s;
1302: struct waypoint *wi=&(inst[i][c->state_in]);
1303: struct waypoint *wo=&(trans[i+j][c->state_out]);
1304: int no_transition = wo->no_transition;
1305: if (!(is_relocatable(s)) && !wo->relocatable) {
1306: wo=&(inst[i+j][c->state_out]);
1307: no_transition=1;
1308: }
1309: if (wo->cost == INF_COST)
1310: continue;
1311: jcost = wo->cost + ss_cost(s);
1312: if (jcost <= wi->cost) {
1313: wi->cost = jcost;
1314: wi->inst = s;
1315: wi->relocatable = is_relocatable(s);
1316: wi->no_transition = no_transition;
1317: /* if (ss_greedy) wi->cost = wo->cost ? */
1318: }
1319: }
1320: }
1321: }
1322: transitions(inst[i],trans[i]);
1323: }
1324: /* now rewrite the instructions */
1325: nextdyn=0;
1326: nextstate=CANONICAL_STATE;
1327: no_transition = ((!trans[0][nextstate].relocatable)
1328: ||trans[0][nextstate].no_transition);
1329: for (i=0; i<ninsts; i++) {
1330: Cell tc=0, tc2;
1331: if (i==nextdyn) {
1332: if (!no_transition) {
1333: /* process trans */
1334: PrimNum p = trans[i][nextstate].inst;
1335: struct cost *c = super_costs+p;
1336: assert(trans[i][nextstate].cost != INF_COST);
1337: assert(c->state_in==nextstate);
1338: tc = compile_prim_dyn(p,NULL);
1339: nextstate = c->state_out;
1340: }
1341: {
1342: /* process inst */
1343: PrimNum p = inst[i][nextstate].inst;
1344: struct cost *c=super_costs+p;
1345: assert(c->state_in==nextstate);
1346: assert(inst[i][nextstate].cost != INF_COST);
1347: #if defined(GFORTH_DEBUGGING)
1348: assert(p == origs[i]);
1349: #endif
1350: tc2 = compile_prim_dyn(p,instps[i]);
1351: if (no_transition || !is_relocatable(p))
1352: /* !! actually what we care about is if and where
1353: * compile_prim_dyn() puts NEXTs */
1354: tc=tc2;
1355: no_transition = inst[i][nextstate].no_transition;
1356: nextstate = c->state_out;
1357: nextdyn += c->length;
1358: }
1359: } else {
1360: #if defined(GFORTH_DEBUGGING)
1361: assert(0);
1362: #endif
1363: tc=0;
1364: /* tc= (Cell)vm_prims[inst[i][CANONICAL_STATE].inst]; */
1365: }
1366: *(instps[i]) = tc;
1367: }
1368: if (!no_transition) {
1369: PrimNum p = trans[i][nextstate].inst;
1370: struct cost *c = super_costs+p;
1371: assert(c->state_in==nextstate);
1372: assert(trans[i][nextstate].cost != INF_COST);
1373: assert(i==nextdyn);
1374: (void)compile_prim_dyn(p,NULL);
1375: nextstate = c->state_out;
1376: }
1377: assert(nextstate==CANONICAL_STATE);
1378: }
1379:
1380: /* compile *start, possibly rewriting it into a static and/or dynamic
1381: superinstruction */
1382: void compile_prim1(Cell *start)
1383: {
1384: #if defined(DOUBLY_INDIRECT)
1385: Label prim;
1386:
1387: if (start==NULL)
1388: return;
1389: prim = (Label)*start;
1390: if (prim<((Label)(xts+DOESJUMP)) || prim>((Label)(xts+npriminfos))) {
1391: fprintf(stderr,"compile_prim encountered xt %p\n", prim);
1392: *start=(Cell)prim;
1393: return;
1394: } else {
1395: *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
1396: return;
1397: }
1398: #elif defined(INDIRECT_THREADED)
1399: return;
1400: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1401: /* !! does not work, for unknown reasons; but something like this is
1402: probably needed to ensure that we don't call compile_prim_dyn
1403: before the inline arguments are there */
1404: static Cell *instps[MAX_BB];
1405: static PrimNum origs[MAX_BB];
1406: static int ninsts=0;
1407: PrimNum prim_num;
1408:
1409: if (start==NULL || ninsts >= MAX_BB ||
1410: (ninsts>0 && superend[origs[ninsts-1]])) {
1411: /* after bb, or at the start of the next bb */
1412: optimize_rewrite(instps,origs,ninsts);
1413: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
1414: ninsts=0;
1415: if (start==NULL)
1416: return;
1417: }
1418: prim_num = ((Xt)*start)-vm_prims;
1419: if(prim_num >= npriminfos) {
1420: optimize_rewrite(instps,origs,ninsts);
1421: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
1422: ninsts=0;
1423: return;
1424: }
1425: assert(ninsts<MAX_BB);
1426: instps[ninsts] = start;
1427: origs[ninsts] = prim_num;
1428: ninsts++;
1429: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1430: }
1431:
1432: Address loader(FILE *imagefile, char* filename)
1433: /* returns the address of the image proper (after the preamble) */
1434: {
1435: ImageHeader header;
1436: Address image;
1437: Address imp; /* image+preamble */
1438: Char magic[8];
1439: char magic7; /* size byte of magic number */
1440: Cell preamblesize=0;
1441: Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1442: UCell check_sum;
1443: Cell ausize = ((RELINFOBITS == 8) ? 0 :
1444: (RELINFOBITS == 16) ? 1 :
1445: (RELINFOBITS == 32) ? 2 : 3);
1446: Cell charsize = ((sizeof(Char) == 1) ? 0 :
1447: (sizeof(Char) == 2) ? 1 :
1448: (sizeof(Char) == 4) ? 2 : 3) + ausize;
1449: Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
1450: (sizeof(Cell) == 2) ? 1 :
1451: (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1452: Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
1453: #ifdef WORDS_BIGENDIAN
1454: 0
1455: #else
1456: 1
1457: #endif
1458: ;
1459:
1460: vm_prims = engine(0,0,0,0,0);
1461: check_prims(vm_prims);
1462: prepare_super_table();
1463: #ifndef DOUBLY_INDIRECT
1464: #ifdef PRINT_SUPER_LENGTHS
1465: print_super_lengths();
1466: #endif
1467: check_sum = checksum(vm_prims);
1468: #else /* defined(DOUBLY_INDIRECT) */
1469: check_sum = (UCell)vm_prims;
1470: #endif /* defined(DOUBLY_INDIRECT) */
1471:
1472: do {
1473: if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1474: fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
1475: progname, filename);
1476: exit(1);
1477: }
1478: preamblesize+=8;
1479: } while(memcmp(magic,"Gforth3",7));
1480: magic7 = magic[7];
1481: if (debug) {
1482: magic[7]='\0';
1483: fprintf(stderr,"Magic found: %s ", magic);
1484: print_sizes(magic7);
1485: }
1486:
1487: if (magic7 != sizebyte)
1488: {
1489: fprintf(stderr,"This image is: ");
1490: print_sizes(magic7);
1491: fprintf(stderr,"whereas the machine is ");
1492: print_sizes(sizebyte);
1493: exit(-2);
1494: };
1495:
1496: fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1497:
1498: set_stack_sizes(&header);
1499:
1500: #if HAVE_GETPAGESIZE
1501: pagesize=getpagesize(); /* Linux/GNU libc offers this */
1502: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
1503: pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
1504: #elif PAGESIZE
1505: pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
1506: #endif
1507: if (debug)
1508: fprintf(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1509:
1510: image = dict_alloc_read(imagefile, preamblesize+header.image_size,
1511: preamblesize+dictsize, data_offset);
1512: imp=image+preamblesize;
1513: alloc_stacks((ImageHeader *)imp);
1514: if (clear_dictionary)
1515: memset(imp+header.image_size, 0, dictsize-header.image_size);
1516: if(header.base==0 || header.base == (Address)0x100) {
1517: Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
1518: char reloc_bits[reloc_size];
1519: fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1520: fread(reloc_bits, 1, reloc_size, imagefile);
1521: relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1522: #if 0
1523: { /* let's see what the relocator did */
1524: FILE *snapshot=fopen("snapshot.fi","wb");
1525: fwrite(image,1,imagesize,snapshot);
1526: fclose(snapshot);
1527: }
1528: #endif
1529: }
1530: else if(header.base!=imp) {
1531: fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
1532: progname, (unsigned long)header.base, (unsigned long)imp);
1533: exit(1);
1534: }
1535: if (header.checksum==0)
1536: ((ImageHeader *)imp)->checksum=check_sum;
1537: else if (header.checksum != check_sum) {
1538: fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
1539: progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
1540: exit(1);
1541: }
1542: #ifdef DOUBLY_INDIRECT
1543: ((ImageHeader *)imp)->xt_base = xts;
1544: #endif
1545: fclose(imagefile);
1546:
1547: /* unnecessary, except maybe for CODE words */
1548: /* FLUSH_ICACHE(imp, header.image_size);*/
1549:
1550: return imp;
1551: }
1552:
1553: /* pointer to last '/' or '\' in file, 0 if there is none. */
1554: char *onlypath(char *filename)
1555: {
1556: return strrchr(filename, DIRSEP);
1557: }
1558:
1559: FILE *openimage(char *fullfilename)
1560: {
1561: FILE *image_file;
1562: char * expfilename = tilde_cstr(fullfilename, strlen(fullfilename), 1);
1563:
1564: image_file=fopen(expfilename,"rb");
1565: if (image_file!=NULL && debug)
1566: fprintf(stderr, "Opened image file: %s\n", expfilename);
1567: return image_file;
1568: }
1569:
1570: /* try to open image file concat(path[0:len],imagename) */
1571: FILE *checkimage(char *path, int len, char *imagename)
1572: {
1573: int dirlen=len;
1574: char fullfilename[dirlen+strlen(imagename)+2];
1575:
1576: memcpy(fullfilename, path, dirlen);
1577: if (fullfilename[dirlen-1]!=DIRSEP)
1578: fullfilename[dirlen++]=DIRSEP;
1579: strcpy(fullfilename+dirlen,imagename);
1580: return openimage(fullfilename);
1581: }
1582:
1583: FILE * open_image_file(char * imagename, char * path)
1584: {
1585: FILE * image_file=NULL;
1586: char *origpath=path;
1587:
1588: if(strchr(imagename, DIRSEP)==NULL) {
1589: /* first check the directory where the exe file is in !! 01may97jaw */
1590: if (onlypath(progname))
1591: image_file=checkimage(progname, onlypath(progname)-progname, imagename);
1592: if (!image_file)
1593: do {
1594: char *pend=strchr(path, PATHSEP);
1595: if (pend==NULL)
1596: pend=path+strlen(path);
1597: if (strlen(path)==0) break;
1598: image_file=checkimage(path, pend-path, imagename);
1599: path=pend+(*pend==PATHSEP);
1600: } while (image_file==NULL);
1601: } else {
1602: image_file=openimage(imagename);
1603: }
1604:
1605: if (!image_file) {
1606: fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
1607: progname, imagename, origpath);
1608: exit(1);
1609: }
1610:
1611: return image_file;
1612: }
1613: #endif
1614:
1615: #ifdef HAS_OS
1616: UCell convsize(char *s, UCell elemsize)
1617: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
1618: of bytes. the letter at the end indicates the unit, where e stands
1619: for the element size. default is e */
1620: {
1621: char *endp;
1622: UCell n,m;
1623:
1624: m = elemsize;
1625: n = strtoul(s,&endp,0);
1626: if (endp!=NULL) {
1627: if (strcmp(endp,"b")==0)
1628: m=1;
1629: else if (strcmp(endp,"k")==0)
1630: m=1024;
1631: else if (strcmp(endp,"M")==0)
1632: m=1024*1024;
1633: else if (strcmp(endp,"G")==0)
1634: m=1024*1024*1024;
1635: else if (strcmp(endp,"T")==0) {
1636: #if (SIZEOF_CHAR_P > 4)
1637: m=1024L*1024*1024*1024;
1638: #else
1639: fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
1640: exit(1);
1641: #endif
1642: } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
1643: fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
1644: exit(1);
1645: }
1646: }
1647: return n*m;
1648: }
1649:
1650: enum {
1651: ss_number = 256,
1652: ss_states,
1653: ss_min_codesize,
1654: ss_min_ls,
1655: ss_min_lsu,
1656: ss_min_nexts,
1657: };
1658:
1659: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
1660: {
1661: int c;
1662:
1663: opterr=0;
1664: while (1) {
1665: int option_index=0;
1666: static struct option opts[] = {
1667: {"appl-image", required_argument, NULL, 'a'},
1668: {"image-file", required_argument, NULL, 'i'},
1669: {"dictionary-size", required_argument, NULL, 'm'},
1670: {"data-stack-size", required_argument, NULL, 'd'},
1671: {"return-stack-size", required_argument, NULL, 'r'},
1672: {"fp-stack-size", required_argument, NULL, 'f'},
1673: {"locals-stack-size", required_argument, NULL, 'l'},
1674: {"path", required_argument, NULL, 'p'},
1675: {"version", no_argument, NULL, 'v'},
1676: {"help", no_argument, NULL, 'h'},
1677: /* put something != 0 into offset_image */
1678: {"offset-image", no_argument, &offset_image, 1},
1679: {"no-offset-im", no_argument, &offset_image, 0},
1680: {"clear-dictionary", no_argument, &clear_dictionary, 1},
1681: {"die-on-signal", no_argument, &die_on_signal, 1},
1682: {"debug", no_argument, &debug, 1},
1683: {"no-super", no_argument, &no_super, 1},
1684: {"no-dynamic", no_argument, &no_dynamic, 1},
1685: {"dynamic", no_argument, &no_dynamic, 0},
1686: {"print-metrics", no_argument, &print_metrics, 1},
1687: {"ss-number", required_argument, NULL, ss_number},
1688: {"ss-states", required_argument, NULL, ss_states},
1689: #ifndef NO_DYNAMIC
1690: {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
1691: #endif
1692: {"ss-min-ls", no_argument, NULL, ss_min_ls},
1693: {"ss-min-lsu", no_argument, NULL, ss_min_lsu},
1694: {"ss-min-nexts", no_argument, NULL, ss_min_nexts},
1695: {"ss-greedy", no_argument, &ss_greedy, 1},
1696: {0,0,0,0}
1697: /* no-init-file, no-rc? */
1698: };
1699:
1700: c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
1701:
1702: switch (c) {
1703: case EOF: return;
1704: case '?': optind--; return;
1705: case 'a': *imagename = optarg; return;
1706: case 'i': *imagename = optarg; break;
1707: case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
1708: case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
1709: case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
1710: case 'f': fsize = convsize(optarg,sizeof(Float)); break;
1711: case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
1712: case 'p': *path = optarg; break;
1713: case 'o': offset_image = 1; break;
1714: case 'n': offset_image = 0; break;
1715: case 'c': clear_dictionary = 1; break;
1716: case 's': die_on_signal = 1; break;
1717: case 'x': debug = 1; break;
1718: case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
1719: case ss_number: static_super_number = atoi(optarg); break;
1720: case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
1721: #ifndef NO_DYNAMIC
1722: case ss_min_codesize: ss_cost = cost_codesize; break;
1723: #endif
1724: case ss_min_ls: ss_cost = cost_ls; break;
1725: case ss_min_lsu: ss_cost = cost_lsu; break;
1726: case ss_min_nexts: ss_cost = cost_nexts; break;
1727: case 'h':
1728: fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
1729: Engine Options:\n\
1730: --appl-image FILE equivalent to '--image-file=FILE --'\n\
1731: --clear-dictionary Initialize the dictionary with 0 bytes\n\
1732: -d SIZE, --data-stack-size=SIZE Specify data stack size\n\
1733: --debug Print debugging information during startup\n\
1734: --die-on-signal exit instead of CATCHing some signals\n\
1735: --dynamic use dynamic native code\n\
1736: -f SIZE, --fp-stack-size=SIZE Specify floating point stack size\n\
1737: -h, --help Print this message and exit\n\
1738: -i FILE, --image-file=FILE Use image FILE instead of `gforth.fi'\n\
1739: -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
1740: -m SIZE, --dictionary-size=SIZE Specify Forth dictionary size\n\
1741: --no-dynamic Use only statically compiled primitives\n\
1742: --no-offset-im Load image at normal position\n\
1743: --no-super No dynamically formed superinstructions\n\
1744: --offset-image Load image at a different position\n\
1745: -p PATH, --path=PATH Search path for finding image and sources\n\
1746: --print-metrics Print some code generation metrics on exit\n\
1747: -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
1748: --ss-greedy greedy, not optimal superinst selection\n\
1749: --ss-min-codesize select superinsts for smallest native code\n\
1750: --ss-min-ls minimize loads and stores\n\
1751: --ss-min-lsu minimize loads, stores, and pointer updates\n\
1752: --ss-min-nexts minimize the number of static superinsts\n\
1753: --ss-number=N use N static superinsts (default max)\n\
1754: --ss-states=N N states for stack caching (default max)\n\
1755: -v, --version Print engine version and exit\n\
1756: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
1757: `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
1758: argv[0]);
1759: optind--;
1760: return;
1761: }
1762: }
1763: }
1764: #endif
1765:
1766: #ifdef INCLUDE_IMAGE
1767: extern Cell image[];
1768: extern const char reloc_bits[];
1769: #endif
1770:
1771: int main(int argc, char **argv, char **env)
1772: {
1773: #ifdef HAS_OS
1774: char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
1775: #else
1776: char *path = DEFAULTPATH;
1777: #endif
1778: #ifndef INCLUDE_IMAGE
1779: char *imagename="gforth.fi";
1780: FILE *image_file;
1781: Address image;
1782: #endif
1783: int retvalue;
1784:
1785: #if defined(i386) && defined(ALIGNMENT_CHECK)
1786: /* turn on alignment checks on the 486.
1787: * on the 386 this should have no effect. */
1788: __asm__("pushfl; popl %eax; orl $0x40000, %eax; pushl %eax; popfl;");
1789: /* this is unusable with Linux' libc.4.6.27, because this library is
1790: not alignment-clean; we would have to replace some library
1791: functions (e.g., memcpy) to make it work. Also GCC doesn't try to keep
1792: the stack FP-aligned. */
1793: #endif
1794:
1795: /* buffering of the user output device */
1796: #ifdef _IONBF
1797: if (isatty(fileno(stdout))) {
1798: fflush(stdout);
1799: setvbuf(stdout,NULL,_IONBF,0);
1800: }
1801: #endif
1802:
1803: progname = argv[0];
1804:
1805: #ifdef HAS_OS
1806: gforth_args(argc, argv, &path, &imagename);
1807: #ifndef NO_DYNAMIC
1808: if (no_dynamic && ss_cost == cost_codesize) {
1809: ss_cost = cost_nexts;
1810: cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
1811: if (debug)
1812: fprintf(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
1813: }
1814: #endif /* !defined(NO_DYNAMIC) */
1815: #endif /* defined(HAS_OS) */
1816:
1817: #ifdef INCLUDE_IMAGE
1818: set_stack_sizes((ImageHeader *)image);
1819: if(((ImageHeader *)image)->base != image)
1820: relocate(image, reloc_bits, ((ImageHeader *)image)->image_size,
1821: (Label*)engine(0, 0, 0, 0, 0));
1822: alloc_stacks((ImageHeader *)image);
1823: #else
1824: image_file = open_image_file(imagename, path);
1825: image = loader(image_file, imagename);
1826: #endif
1827: gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
1828:
1829: {
1830: char path2[strlen(path)+1];
1831: char *p1, *p2;
1832: Cell environ[]= {
1833: (Cell)argc-(optind-1),
1834: (Cell)(argv+(optind-1)),
1835: (Cell)strlen(path),
1836: (Cell)path2};
1837: argv[optind-1] = progname;
1838: /*
1839: for (i=0; i<environ[0]; i++)
1840: printf("%s\n", ((char **)(environ[1]))[i]);
1841: */
1842: /* make path OS-independent by replacing path separators with NUL */
1843: for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
1844: if (*p1==PATHSEP)
1845: *p2 = '\0';
1846: else
1847: *p2 = *p1;
1848: *p2='\0';
1849: retvalue = go_forth(image, 4, environ);
1850: #ifdef SIGPIPE
1851: bsd_signal(SIGPIPE, SIG_IGN);
1852: #endif
1853: #ifdef VM_PROFILING
1854: vm_print_profile(stderr);
1855: #endif
1856: deprep_terminal();
1857: }
1858: if (print_metrics) {
1859: int i;
1860: fprintf(stderr, "code size = %8ld\n", dyncodesize());
1861: for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
1862: fprintf(stderr, "metric %8s: %8ld\n",
1863: cost_sums[i].metricname, cost_sums[i].sum);
1864: }
1865: return retvalue;
1866: }
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