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: /* save global valiables */
74: Cell *rp = RP;
75: Cell *sp = SP;
76: Float *fp = FP;
77: Address lp = LP;
78:
79: clist = (va_alist)alist;
80:
81: engine(fcall, sp, rp, fp, lp);
82:
83: /* restore global variables */
84: RP = rp;
85: SP = sp;
86: FP = fp;
87: LP = lp;
88: }
89: #endif
90:
91: #ifdef GFORTH_DEBUGGING
92: /* define some VM registers as global variables, so they survive exceptions;
93: global register variables are not up to the task (according to the
94: GNU C manual) */
95: Xt *saved_ip;
96: Cell *rp;
97: #endif
98:
99: #ifdef NO_IP
100: Label next_code;
101: #endif
102:
103: #ifdef HAS_FILE
104: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
105: char* pfileattr[6]={"r","r","r+","r+","w","w"};
106:
107: #ifndef O_BINARY
108: #define O_BINARY 0
109: #endif
110: #ifndef O_TEXT
111: #define O_TEXT 0
112: #endif
113:
114: int ufileattr[6]= {
115: O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
116: O_RDWR |O_BINARY, O_RDWR |O_BINARY,
117: O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
118: #endif
119: /* end global vars for engine.c */
120:
121: #define PRIM_VERSION 1
122: /* increment this whenever the primitives change in an incompatible way */
123:
124: #ifndef DEFAULTPATH
125: # define DEFAULTPATH "."
126: #endif
127:
128: #ifdef MSDOS
129: jmp_buf throw_jmp_buf;
130: #endif
131:
132: #if defined(DOUBLY_INDIRECT)
133: # define CFA(n) ({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
134: #else
135: # define CFA(n) ((Cell)(symbols+((n)&~0x4000UL)))
136: #endif
137:
138: #define maxaligned(n) (typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
139:
140: static UCell dictsize=0;
141: static UCell dsize=0;
142: static UCell rsize=0;
143: static UCell fsize=0;
144: static UCell lsize=0;
145: int offset_image=0;
146: int die_on_signal=0;
147: #ifndef INCLUDE_IMAGE
148: static int clear_dictionary=0;
149: UCell pagesize=1;
150: char *progname;
151: #else
152: char *progname = "gforth";
153: int optind = 1;
154: #endif
155:
156: #define CODE_BLOCK_SIZE (4096*1024) /* !! overflow handling for -native */
157: Address code_area=0;
158: Cell code_area_size = CODE_BLOCK_SIZE;
159: Address code_here=NULL+CODE_BLOCK_SIZE; /* does for code-area what HERE
160: does for the dictionary */
161: Address start_flush=NULL; /* start of unflushed code */
162: Cell last_jump=0; /* if the last prim was compiled without jump, this
163: is it's number, otherwise this contains 0 */
164:
165: static int no_super=0; /* true if compile_prim should not fuse prims */
166: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
167: dynamically */
168: static int print_metrics=0; /* if true, print metrics on exit */
169: static int static_super_number = 10000000; /* number of ss used if available */
170: #define MAX_STATE 4 /* maximum number of states */
171: static int maxstates = MAX_STATE; /* number of states for stack caching */
172: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
173:
174: #ifdef HAS_DEBUG
175: int debug=0;
176: #else
177: # define perror(x...)
178: # define fprintf(x...)
179: #endif
180:
181: ImageHeader *gforth_header;
182: Label *vm_prims;
183: #ifdef DOUBLY_INDIRECT
184: Label *xts; /* same content as vm_prims, but should only be used for xts */
185: #endif
186:
187: #ifndef NO_DYNAMIC
188: #define MAX_IMMARGS 2
189:
190: typedef struct {
191: Label start; /* NULL if not relocatable */
192: Cell length; /* only includes the jump iff superend is true*/
193: Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
194: char superend; /* true if primitive ends superinstruction, i.e.,
195: unconditional branch, execute, etc. */
196: Cell nimmargs;
197: struct immarg {
198: Cell offset; /* offset of immarg within prim */
199: char rel; /* true if immarg is relative */
200: } immargs[MAX_IMMARGS];
201: } PrimInfo;
202:
203: PrimInfo *priminfos;
204: PrimInfo **decomp_prims;
205:
206: const char const* const prim_names[]={
207: #include PRIM_NAMES_I
208: };
209:
210: static int is_relocatable(int p)
211: {
212: return !no_dynamic && priminfos[p].start != NULL;
213: }
214: #else /* defined(NO_DYNAMIC) */
215: static int is_relocatable(int p)
216: {
217: return 0;
218: }
219: #endif /* defined(NO_DYNAMIC) */
220:
221: #ifdef MEMCMP_AS_SUBROUTINE
222: int gforth_memcmp(const char * s1, const char * s2, size_t n)
223: {
224: return memcmp(s1, s2, n);
225: }
226: #endif
227:
228: static Cell max(Cell a, Cell b)
229: {
230: return a>b?a:b;
231: }
232:
233: static Cell min(Cell a, Cell b)
234: {
235: return a<b?a:b;
236: }
237:
238: /* image file format:
239: * "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
240: * padding to a multiple of 8
241: * magic: "Gforth3x" means format 0.6,
242: * where x is a byte with
243: * bit 7: reserved = 0
244: * bit 6:5: address unit size 2^n octets
245: * bit 4:3: character size 2^n octets
246: * bit 2:1: cell size 2^n octets
247: * bit 0: endian, big=0, little=1.
248: * The magic are always 8 octets, no matter what the native AU/character size is
249: * padding to max alignment (no padding necessary on current machines)
250: * ImageHeader structure (see forth.h)
251: * data (size in ImageHeader.image_size)
252: * tags ((if relocatable, 1 bit/data cell)
253: *
254: * tag==1 means that the corresponding word is an address;
255: * If the word is >=0, the address is within the image;
256: * addresses within the image are given relative to the start of the image.
257: * If the word =-1 (CF_NIL), the address is NIL,
258: * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
259: * If the word =CF(DODOES), it's a DOES> CFA
260: * If the word =CF(DOESJUMP), it's a DOES JUMP (2 Cells after DOES>,
261: * possibly containing a jump to dodoes)
262: * If the word is <CF(DOESJUMP) and bit 14 is set, it's the xt of a primitive
263: * If the word is <CF(DOESJUMP) and bit 14 is clear,
264: * it's the threaded code of a primitive
265: * bits 13..9 of a primitive token state which group the primitive belongs to,
266: * bits 8..0 of a primitive token index into the group
267: */
268:
269: Cell groups[32] = {
270: 0,
271: 0
272: #undef GROUP
273: #undef GROUPADD
274: #define GROUPADD(n) +n
275: #define GROUP(x, n) , 0
276: #include PRIM_GRP_I
277: #undef GROUP
278: #undef GROUPADD
279: #define GROUP(x, n)
280: #define GROUPADD(n)
281: };
282:
283: unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
284: int size, Cell base)
285: /* produce a bitmask marking all the branch targets */
286: {
287: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
288: Cell token;
289: unsigned char bits;
290: unsigned char *result=malloc(steps);
291:
292: memset(result, 0, steps);
293: for(k=0; k<steps; k++) {
294: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
295: if(bits & (1U << (RELINFOBITS-1))) {
296: assert(i*sizeof(Cell) < size);
297: token=image[i];
298: if (token>=base) { /* relocatable address */
299: UCell bitnum=(token-base)/sizeof(Cell);
300: result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
301: }
302: }
303: }
304: }
305: return result;
306: }
307:
308: void relocate(Cell *image, const unsigned char *bitstring,
309: int size, Cell base, Label symbols[])
310: {
311: int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
312: Cell token;
313: char bits;
314: Cell max_symbols;
315: /*
316: * A virtual start address that's the real start address minus
317: * the one in the image
318: */
319: Cell *start = (Cell * ) (((void *) image) - ((void *) base));
320: unsigned char *targets = branch_targets(image, bitstring, size, base);
321:
322: /* group index into table */
323: if(groups[31]==0) {
324: int groupsum=0;
325: for(i=0; i<32; i++) {
326: groupsum += groups[i];
327: groups[i] = groupsum;
328: /* printf("group[%d]=%d\n",i,groupsum); */
329: }
330: i=0;
331: }
332:
333: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
334:
335: for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
336: ;
337: max_symbols--;
338:
339: for(k=0; k<steps; k++) {
340: for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
341: /* fprintf(stderr,"relocate: image[%d]\n", i);*/
342: if(bits & (1U << (RELINFOBITS-1))) {
343: assert(i*sizeof(Cell) < size);
344: /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
345: token=image[i];
346: if(token<0) {
347: int group = (-token & 0x3E00) >> 9;
348: if(group == 0) {
349: switch(token|0x4000) {
350: case CF_NIL : image[i]=0; break;
351: #if !defined(DOUBLY_INDIRECT)
352: case CF(DOCOL) :
353: case CF(DOVAR) :
354: case CF(DOCON) :
355: case CF(DOUSER) :
356: case CF(DODEFER) :
357: case CF(DOFIELD) : MAKE_CF(image+i,symbols[CF(token)]); break;
358: case CF(DOESJUMP): image[i]=0; break;
359: #endif /* !defined(DOUBLY_INDIRECT) */
360: case CF(DODOES) :
361: MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
362: break;
363: default : /* backward compatibility */
364: /* printf("Code field generation image[%x]:=CFA(%x)\n",
365: i, CF(image[i])); */
366: if (CF((token | 0x4000))<max_symbols) {
367: image[i]=(Cell)CFA(CF(token));
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 %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);
377: }
378: } else {
379: int tok = -token & 0x1FF;
380: if (tok < (groups[group+1]-groups[group])) {
381: #if defined(DOUBLY_INDIRECT)
382: image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
383: #else
384: image[i]=(Cell)CFA((groups[group]+tok));
385: #endif
386: #ifdef DIRECT_THREADED
387: if ((token & 0x4000) == 0) { /* threade code, no CFA */
388: if (targets[k] & (1U<<(RELINFOBITS-1-j)))
389: compile_prim1(0);
390: compile_prim1(&image[i]);
391: }
392: #endif
393: } else
394: fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
395: }
396: } else {
397: /* if base is > 0: 0 is a null reference so don't adjust*/
398: if (token>=base) {
399: image[i]+=(Cell)start;
400: }
401: }
402: }
403: }
404: }
405: free(targets);
406: finish_code();
407: ((ImageHeader*)(image))->base = (Address) image;
408: }
409:
410: UCell checksum(Label symbols[])
411: {
412: UCell r=PRIM_VERSION;
413: Cell i;
414:
415: for (i=DOCOL; i<=DOESJUMP; i++) {
416: r ^= (UCell)(symbols[i]);
417: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
418: }
419: #ifdef DIRECT_THREADED
420: /* we have to consider all the primitives */
421: for (; symbols[i]!=(Label)0; i++) {
422: r ^= (UCell)(symbols[i]);
423: r = (r << 5) | (r >> (8*sizeof(Cell)-5));
424: }
425: #else
426: /* in indirect threaded code all primitives are accessed through the
427: symbols table, so we just have to put the base address of symbols
428: in the checksum */
429: r ^= (UCell)symbols;
430: #endif
431: return r;
432: }
433:
434: Address verbose_malloc(Cell size)
435: {
436: Address r;
437: /* leave a little room (64B) for stack underflows */
438: if ((r = malloc(size+64))==NULL) {
439: perror(progname);
440: exit(1);
441: }
442: r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
443: if (debug)
444: fprintf(stderr, "malloc succeeds, address=$%lx\n", (long)r);
445: return r;
446: }
447:
448: static Address next_address=0;
449: void after_alloc(Address r, Cell size)
450: {
451: if (r != (Address)-1) {
452: if (debug)
453: fprintf(stderr, "success, address=$%lx\n", (long) r);
454: if (pagesize != 1)
455: next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
456: } else {
457: if (debug)
458: fprintf(stderr, "failed: %s\n", strerror(errno));
459: }
460: }
461:
462: #ifndef MAP_FAILED
463: #define MAP_FAILED ((Address) -1)
464: #endif
465: #ifndef MAP_FILE
466: # define MAP_FILE 0
467: #endif
468: #ifndef MAP_PRIVATE
469: # define MAP_PRIVATE 0
470: #endif
471: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
472: # define MAP_ANON MAP_ANONYMOUS
473: #endif
474:
475: #if defined(HAVE_MMAP)
476: static Address alloc_mmap(Cell size)
477: {
478: Address r;
479:
480: #if defined(MAP_ANON)
481: if (debug)
482: fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
483: r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE, -1, 0);
484: #else /* !defined(MAP_ANON) */
485: /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
486: apparently defaults) */
487: static int dev_zero=-1;
488:
489: if (dev_zero == -1)
490: dev_zero = open("/dev/zero", O_RDONLY);
491: if (dev_zero == -1) {
492: r = MAP_FAILED;
493: if (debug)
494: fprintf(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ",
495: strerror(errno));
496: } else {
497: if (debug)
498: fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
499: r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE, dev_zero, 0);
500: }
501: #endif /* !defined(MAP_ANON) */
502: after_alloc(r, size);
503: return r;
504: }
505: #endif
506:
507: Address my_alloc(Cell size)
508: {
509: #if HAVE_MMAP
510: Address r;
511:
512: r=alloc_mmap(size);
513: if (r!=(Address)MAP_FAILED)
514: return r;
515: #endif /* HAVE_MMAP */
516: /* use malloc as fallback */
517: return verbose_malloc(size);
518: }
519:
520: Address dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
521: {
522: Address image = MAP_FAILED;
523:
524: #if defined(HAVE_MMAP)
525: if (offset==0) {
526: image=alloc_mmap(dictsize);
527: if (debug)
528: fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
529: image = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE, fileno(file), 0);
530: after_alloc(image,dictsize);
531: }
532: #endif /* defined(HAVE_MMAP) */
533: if (image == (Address)MAP_FAILED) {
534: image = my_alloc(dictsize+offset)+offset;
535: rewind(file); /* fseek(imagefile,0L,SEEK_SET); */
536: fread(image, 1, imagesize, file);
537: }
538: return image;
539: }
540:
541: void set_stack_sizes(ImageHeader * header)
542: {
543: if (dictsize==0)
544: dictsize = header->dict_size;
545: if (dsize==0)
546: dsize = header->data_stack_size;
547: if (rsize==0)
548: rsize = header->return_stack_size;
549: if (fsize==0)
550: fsize = header->fp_stack_size;
551: if (lsize==0)
552: lsize = header->locals_stack_size;
553: dictsize=maxaligned(dictsize);
554: dsize=maxaligned(dsize);
555: rsize=maxaligned(rsize);
556: lsize=maxaligned(lsize);
557: fsize=maxaligned(fsize);
558: }
559:
560: void alloc_stacks(ImageHeader * header)
561: {
562: header->dict_size=dictsize;
563: header->data_stack_size=dsize;
564: header->fp_stack_size=fsize;
565: header->return_stack_size=rsize;
566: header->locals_stack_size=lsize;
567:
568: header->data_stack_base=my_alloc(dsize);
569: header->fp_stack_base=my_alloc(fsize);
570: header->return_stack_base=my_alloc(rsize);
571: header->locals_stack_base=my_alloc(lsize);
572: }
573:
574: #warning You can ignore the warnings about clobbered variables in go_forth
575: int go_forth(Address image, int stack, Cell *entries)
576: {
577: volatile ImageHeader *image_header = (ImageHeader *)image;
578: Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
579: Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
580: Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
581: #ifdef GFORTH_DEBUGGING
582: volatile Cell *orig_rp0=rp0;
583: #endif
584: Address lp0=image_header->locals_stack_base + lsize;
585: Xt *ip0=(Xt *)(image_header->boot_entry);
586: #ifdef SYSSIGNALS
587: int throw_code;
588: #endif
589:
590: /* ensure that the cached elements (if any) are accessible */
591: IF_spTOS(sp0--);
592: IF_fpTOS(fp0--);
593:
594: for(;stack>0;stack--)
595: *--sp0=entries[stack-1];
596:
597: #ifdef SYSSIGNALS
598: get_winsize();
599:
600: install_signal_handlers(); /* right place? */
601:
602: if ((throw_code=setjmp(throw_jmp_buf))) {
603: static Cell signal_data_stack[8];
604: static Cell signal_return_stack[8];
605: static Float signal_fp_stack[1];
606:
607: signal_data_stack[7]=throw_code;
608:
609: #ifdef GFORTH_DEBUGGING
610: if (debug)
611: fprintf(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
612: throw_code, saved_ip, rp);
613: if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
614: /* no rstack overflow or underflow */
615: rp0 = rp;
616: *--rp0 = (Cell)saved_ip;
617: }
618: else /* I love non-syntactic ifdefs :-) */
619: rp0 = signal_return_stack+8;
620: #else /* !defined(GFORTH_DEBUGGING) */
621: if (debug)
622: fprintf(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
623: rp0 = signal_return_stack+8;
624: #endif /* !defined(GFORTH_DEBUGGING) */
625: /* fprintf(stderr, "rp=$%x\n",rp0);*/
626:
627: return((int)(Cell)engine(image_header->throw_entry, signal_data_stack+7,
628: rp0, signal_fp_stack, 0));
629: }
630: #endif
631:
632: return((int)(Cell)engine(ip0,sp0,rp0,fp0,lp0));
633: }
634:
635: #ifndef INCLUDE_IMAGE
636: void print_sizes(Cell sizebyte)
637: /* print size information */
638: {
639: static char* endianstring[]= { " big","little" };
640:
641: fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
642: endianstring[sizebyte & 1],
643: 1 << ((sizebyte >> 1) & 3),
644: 1 << ((sizebyte >> 3) & 3),
645: 1 << ((sizebyte >> 5) & 3));
646: }
647:
648: /* static superinstruction stuff */
649:
650: struct cost {
651: char loads; /* number of stack loads */
652: char stores; /* number of stack stores */
653: char updates; /* number of stack pointer updates */
654: char branch; /* is it a branch (SET_IP) */
655: unsigned char state_in; /* state on entry */
656: unsigned char state_out; /* state on exit */
657: short offset; /* offset into super2 table */
658: unsigned char length; /* number of components */
659: };
660:
661: PrimNum super2[] = {
662: #include SUPER2_I
663: };
664:
665: struct cost super_costs[] = {
666: #include COSTS_I
667: };
668:
669: struct super_state {
670: struct super_state *next;
671: PrimNum super;
672: };
673:
674: #define HASH_SIZE 256
675:
676: struct super_table_entry {
677: struct super_table_entry *next;
678: PrimNum *start;
679: short length;
680: struct super_state *ss_list; /* list of supers */
681: } *super_table[HASH_SIZE];
682: int max_super=2;
683:
684: struct super_state *state_transitions=NULL;
685:
686: int hash_super(PrimNum *start, int length)
687: {
688: int i, r;
689:
690: for (i=0, r=0; i<length; i++) {
691: r <<= 1;
692: r += start[i];
693: }
694: return r & (HASH_SIZE-1);
695: }
696:
697: struct super_state **lookup_super(PrimNum *start, int length)
698: {
699: int hash=hash_super(start,length);
700: struct super_table_entry *p = super_table[hash];
701:
702: /* assert(length >= 2); */
703: for (; p!=NULL; p = p->next) {
704: if (length == p->length &&
705: memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
706: return &(p->ss_list);
707: }
708: return NULL;
709: }
710:
711: void prepare_super_table()
712: {
713: int i;
714: int nsupers = 0;
715:
716: for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
717: struct cost *c = &super_costs[i];
718: if ((c->length < 2 || nsupers < static_super_number) &&
719: c->state_in < maxstates && c->state_out < maxstates) {
720: struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
721: struct super_state *ss = malloc(sizeof(struct super_state));
722: ss->super= i;
723: if (c->offset==N_noop && i != N_noop) {
724: if (is_relocatable(i)) {
725: ss->next = state_transitions;
726: state_transitions = ss;
727: }
728: } else if (ss_listp != NULL) {
729: ss->next = *ss_listp;
730: *ss_listp = ss;
731: } else {
732: int hash = hash_super(super2+c->offset, c->length);
733: struct super_table_entry **p = &super_table[hash];
734: struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
735: ss->next = NULL;
736: e->next = *p;
737: e->start = super2 + c->offset;
738: e->length = c->length;
739: e->ss_list = ss;
740: *p = e;
741: }
742: if (c->length > max_super)
743: max_super = c->length;
744: if (c->length >= 2)
745: nsupers++;
746: }
747: }
748: if (debug)
749: fprintf(stderr, "Using %d static superinsts\n", nsupers);
750: }
751:
752: /* dynamic replication/superinstruction stuff */
753:
754: #ifndef NO_DYNAMIC
755: int compare_priminfo_length(const void *_a, const void *_b)
756: {
757: PrimInfo **a = (PrimInfo **)_a;
758: PrimInfo **b = (PrimInfo **)_b;
759: Cell diff = (*a)->length - (*b)->length;
760: if (diff)
761: return diff;
762: else /* break ties by start address; thus the decompiler produces
763: the earliest primitive with the same code (e.g. noop instead
764: of (char) and @ instead of >code-address */
765: return (*b)->start - (*a)->start;
766: }
767: #endif /* !defined(NO_DYNAMIC) */
768:
769: static char MAYBE_UNUSED superend[]={
770: #include PRIM_SUPEREND_I
771: };
772:
773: Cell npriminfos=0;
774:
775: int compare_labels(const void *pa, const void *pb)
776: {
777: Label a = *(Label *)pa;
778: Label b = *(Label *)pb;
779: return a-b;
780: }
781:
782: Label bsearch_next(Label key, Label *a, UCell n)
783: /* a is sorted; return the label >=key that is the closest in a;
784: return NULL if there is no label in a >=key */
785: {
786: int mid = (n-1)/2;
787: if (n<1)
788: return NULL;
789: if (n == 1) {
790: if (a[0] < key)
791: return NULL;
792: else
793: return a[0];
794: }
795: if (a[mid] < key)
796: return bsearch_next(key, a+mid+1, n-mid-1);
797: else
798: return bsearch_next(key, a, mid+1);
799: }
800:
801: void check_prims(Label symbols1[])
802: {
803: int i;
804: #ifndef NO_DYNAMIC
805: Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted;
806: int nends1j;
807: #endif
808:
809: if (debug)
810: #ifdef __VERSION__
811: fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
812: #else
813: #define xstr(s) str(s)
814: #define str(s) #s
815: fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n");
816: #endif
817: for (i=0; symbols1[i]!=0; i++)
818: ;
819: npriminfos = i;
820:
821: #ifndef NO_DYNAMIC
822: if (no_dynamic)
823: return;
824: symbols2=engine2(0,0,0,0,0);
825: #if NO_IP
826: symbols3=engine3(0,0,0,0,0);
827: #else
828: symbols3=symbols1;
829: #endif
830: ends1 = symbols1+i+1;
831: ends1j = ends1+i;
832: nends1j = i+1;
833: ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
834: memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
835: qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
836:
837: priminfos = calloc(i,sizeof(PrimInfo));
838: for (i=0; symbols1[i]!=0; i++) {
839: int prim_len = ends1[i]-symbols1[i];
840: PrimInfo *pi=&priminfos[i];
841: int j=0;
842: char *s1 = (char *)symbols1[i];
843: char *s2 = (char *)symbols2[i];
844: char *s3 = (char *)symbols3[i];
845: Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
846:
847: pi->start = s1;
848: pi->superend = superend[i]|no_super;
849: if (pi->superend)
850: pi->length = endlabel-symbols1[i];
851: else
852: pi->length = prim_len;
853: pi->restlength = endlabel - symbols1[i] - pi->length;
854: pi->nimmargs = 0;
855: if (debug)
856: fprintf(stderr, "%-15s %3d %p %p %p len=%3ld restlen=%2ld s-end=%1d",
857: prim_names[i], i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength), pi->superend);
858: if (endlabel == NULL) {
859: pi->start = NULL; /* not relocatable */
860: if (pi->length<0) pi->length=100;
861: if (debug)
862: fprintf(stderr,"\n non_reloc: no J label > start found\n");
863: continue;
864: }
865: if (ends1[i] > endlabel && !pi->superend) {
866: pi->start = NULL; /* not relocatable */
867: pi->length = endlabel-symbols1[i];
868: if (debug)
869: fprintf(stderr,"\n non_reloc: there is a J label before the K label (restlength<0)\n");
870: continue;
871: }
872: if (ends1[i] < pi->start && !pi->superend) {
873: pi->start = NULL; /* not relocatable */
874: pi->length = endlabel-symbols1[i];
875: if (debug)
876: fprintf(stderr,"\n non_reloc: K label before I label (length<0)\n");
877: continue;
878: }
879: assert(pi->length>=0);
880: assert(pi->restlength >=0);
881: while (j<(pi->length+pi->restlength)) {
882: if (s1[j]==s3[j]) {
883: if (s1[j] != s2[j]) {
884: pi->start = NULL; /* not relocatable */
885: if (debug)
886: fprintf(stderr,"\n non_reloc: engine1!=engine2 offset %3d",j);
887: /* assert(j<prim_len); */
888: break;
889: }
890: j++;
891: } else {
892: struct immarg *ia=&pi->immargs[pi->nimmargs];
893:
894: pi->nimmargs++;
895: ia->offset=j;
896: if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
897: ia->rel=0;
898: if (debug)
899: fprintf(stderr,"\n absolute immarg: offset %3d",j);
900: } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
901: symbols1[DOESJUMP+1]) {
902: ia->rel=1;
903: if (debug)
904: fprintf(stderr,"\n relative immarg: offset %3d",j);
905: } else {
906: pi->start = NULL; /* not relocatable */
907: if (debug)
908: fprintf(stderr,"\n non_reloc: engine1!=engine3 offset %3d",j);
909: /* assert(j<prim_len);*/
910: break;
911: }
912: j+=4;
913: }
914: }
915: if (debug)
916: fprintf(stderr,"\n");
917: }
918: decomp_prims = calloc(i,sizeof(PrimInfo *));
919: for (i=DOESJUMP+1; i<npriminfos; i++)
920: decomp_prims[i] = &(priminfos[i]);
921: qsort(decomp_prims+DOESJUMP+1, npriminfos-DOESJUMP-1, sizeof(PrimInfo *),
922: compare_priminfo_length);
923: #endif
924: }
925:
926: void flush_to_here(void)
927: {
928: #ifndef NO_DYNAMIC
929: if (start_flush)
930: FLUSH_ICACHE(start_flush, code_here-start_flush);
931: start_flush=code_here;
932: #endif
933: }
934:
935: #ifndef NO_DYNAMIC
936: void append_jump(void)
937: {
938: if (last_jump) {
939: PrimInfo *pi = &priminfos[last_jump];
940:
941: memcpy(code_here, pi->start+pi->length, pi->restlength);
942: code_here += pi->restlength;
943: last_jump=0;
944: }
945: }
946:
947: /* Gforth remembers all code blocks in this list. On forgetting (by
948: executing a marker) the code blocks are not freed (because Gforth does
949: not remember how they were allocated; hmm, remembering that might be
950: easier and cleaner). Instead, code_here etc. are reset to the old
951: value, and the "forgotten" code blocks are reused when they are
952: needed. */
953:
954: struct code_block_list {
955: struct code_block_list *next;
956: Address block;
957: Cell size;
958: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
959:
960: Address append_prim(Cell p)
961: {
962: PrimInfo *pi = &priminfos[p];
963: Address old_code_here = code_here;
964:
965: if (code_area+code_area_size < code_here+pi->length+pi->restlength) {
966: struct code_block_list *p;
967: append_jump();
968: flush_to_here();
969: if (*next_code_blockp == NULL) {
970: code_here = start_flush = code_area = my_alloc(code_area_size);
971: p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
972: *next_code_blockp = p;
973: p->next = NULL;
974: p->block = code_here;
975: p->size = code_area_size;
976: } else {
977: p = *next_code_blockp;
978: code_here = start_flush = code_area = p->block;
979: }
980: old_code_here = code_here;
981: next_code_blockp = &(p->next);
982: }
983: memcpy(code_here, pi->start, pi->length);
984: code_here += pi->length;
985: return old_code_here;
986: }
987: #endif
988:
989: int forget_dyncode(Address code)
990: {
991: #ifdef NO_DYNAMIC
992: return -1;
993: #else
994: struct code_block_list *p, **pp;
995:
996: for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
997: if (code >= p->block && code < p->block+p->size) {
998: next_code_blockp = &(p->next);
999: code_here = start_flush = code;
1000: code_area = p->block;
1001: last_jump = 0;
1002: return -1;
1003: }
1004: }
1005: return -no_dynamic;
1006: #endif /* !defined(NO_DYNAMIC) */
1007: }
1008:
1009: long dyncodesize(void)
1010: {
1011: #ifndef NO_DYNAMIC
1012: struct code_block_list *p;
1013: long size=0;
1014: for (p=code_block_list; p!=NULL; p=p->next) {
1015: if (code_here >= p->block && code_here < p->block+p->size)
1016: return size + (code_here - p->block);
1017: else
1018: size += p->size;
1019: }
1020: #endif /* !defined(NO_DYNAMIC) */
1021: return 0;
1022: }
1023:
1024: Label decompile_code(Label _code)
1025: {
1026: #ifdef NO_DYNAMIC
1027: return _code;
1028: #else /* !defined(NO_DYNAMIC) */
1029: Cell i;
1030: struct code_block_list *p;
1031: Address code=_code;
1032:
1033: /* first, check if we are in code at all */
1034: for (p = code_block_list;; p = p->next) {
1035: if (p == NULL)
1036: return code;
1037: if (code >= p->block && code < p->block+p->size)
1038: break;
1039: }
1040: /* reverse order because NOOP might match other prims */
1041: for (i=npriminfos-1; i>DOESJUMP; i--) {
1042: PrimInfo *pi=decomp_prims[i];
1043: if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1044: return vm_prims[super2[super_costs[pi-priminfos].offset]];
1045: /* return pi->start;*/
1046: }
1047: return code;
1048: #endif /* !defined(NO_DYNAMIC) */
1049: }
1050:
1051: #ifdef NO_IP
1052: int nbranchinfos=0;
1053:
1054: struct branchinfo {
1055: Label **targetpp; /* **(bi->targetpp) is the target */
1056: Cell *addressptr; /* store the target here */
1057: } branchinfos[100000];
1058:
1059: int ndoesexecinfos=0;
1060: struct doesexecinfo {
1061: int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1062: Label *targetp; /*target for branch (because this is not in threaded code)*/
1063: Cell *xt; /* cfa of word whose does-code needs calling */
1064: } doesexecinfos[10000];
1065:
1066: void set_rel_target(Cell *source, Label target)
1067: {
1068: *source = ((Cell)target)-(((Cell)source)+4);
1069: }
1070:
1071: void register_branchinfo(Label source, Cell *targetpp)
1072: {
1073: struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1074: bi->targetpp = (Label **)targetpp;
1075: bi->addressptr = (Cell *)source;
1076: nbranchinfos++;
1077: }
1078:
1079: Address compile_prim1arg(PrimNum p, Cell **argp)
1080: {
1081: Address old_code_here=append_prim(p);
1082:
1083: assert(vm_prims[p]==priminfos[p].start);
1084: *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
1085: return old_code_here;
1086: }
1087:
1088: Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
1089: {
1090: PrimInfo *pi = &priminfos[N_call2];
1091: Address old_code_here = append_prim(N_call2);
1092:
1093: *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
1094: register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
1095: return old_code_here;
1096: }
1097: #endif
1098:
1099: void finish_code(void)
1100: {
1101: #ifdef NO_IP
1102: Cell i;
1103:
1104: compile_prim1(NULL);
1105: for (i=0; i<ndoesexecinfos; i++) {
1106: struct doesexecinfo *dei = &doesexecinfos[i];
1107: dei->targetp = (Label *)DOES_CODE1((dei->xt));
1108: branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
1109: }
1110: ndoesexecinfos = 0;
1111: for (i=0; i<nbranchinfos; i++) {
1112: struct branchinfo *bi=&branchinfos[i];
1113: set_rel_target(bi->addressptr, **(bi->targetpp));
1114: }
1115: nbranchinfos = 0;
1116: #else
1117: compile_prim1(NULL);
1118: #endif
1119: flush_to_here();
1120: }
1121:
1122: #ifdef NO_IP
1123: Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1124: /* compile prim #p dynamically (mod flags etc.) and return start
1125: address of generated code for putting it into the threaded
1126: code. This function is only called if all the associated
1127: inline arguments of p are already in place (at tcp[1] etc.) */
1128: {
1129: PrimInfo *pi=&priminfos[p];
1130: Cell *next_code_target=NULL;
1131: Address codeaddr;
1132: Address primstart;
1133:
1134: assert(p<npriminfos);
1135: if (p==N_execute || p==N_perform || p==N_lit_perform) {
1136: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1137: primstart = append_prim(p);
1138: goto other_prim;
1139: } else if (p==N_call) {
1140: codeaddr = compile_call2(tcp+1, &next_code_target);
1141: } else if (p==N_does_exec) {
1142: struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1143: Cell *arg;
1144: codeaddr = compile_prim1arg(N_lit,&arg);
1145: *arg = (Cell)PFA(tcp[1]);
1146: /* we cannot determine the callee now (last_start[1] may be a
1147: forward reference), so just register an arbitrary target, and
1148: register in dei that we need to fix this before resolving
1149: branches */
1150: dei->branchinfo = nbranchinfos;
1151: dei->xt = (Cell *)(tcp[1]);
1152: compile_call2(0, &next_code_target);
1153: } else if (!is_relocatable(p)) {
1154: Cell *branch_target;
1155: codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1156: compile_prim1arg(N_branch,&branch_target);
1157: set_rel_target(branch_target,vm_prims[p]);
1158: } else {
1159: unsigned j;
1160:
1161: codeaddr = primstart = append_prim(p);
1162: other_prim:
1163: for (j=0; j<pi->nimmargs; j++) {
1164: struct immarg *ia = &(pi->immargs[j]);
1165: Cell *argp = tcp + pi->nimmargs - j;
1166: Cell argval = *argp; /* !! specific to prims */
1167: if (ia->rel) { /* !! assumption: relative refs are branches */
1168: register_branchinfo(primstart + ia->offset, argp);
1169: } else /* plain argument */
1170: *(Cell *)(primstart + ia->offset) = argval;
1171: }
1172: }
1173: if (next_code_target!=NULL)
1174: *next_code_target = (Cell)code_here;
1175: return (Cell)codeaddr;
1176: }
1177: #else /* !defined(NO_IP) */
1178: Cell compile_prim_dyn(PrimNum p, Cell *tcp)
1179: /* compile prim #p dynamically (mod flags etc.) and return start
1180: address of generated code for putting it into the threaded code */
1181: {
1182: Cell static_prim = (Cell)vm_prims[p];
1183: #if defined(NO_DYNAMIC)
1184: return static_prim;
1185: #else /* !defined(NO_DYNAMIC) */
1186: Address old_code_here;
1187:
1188: if (no_dynamic)
1189: return static_prim;
1190: if (p>=npriminfos || !is_relocatable(p)) {
1191: append_jump();
1192: return static_prim;
1193: }
1194: old_code_here = append_prim(p);
1195: last_jump = (priminfos[p].superend) ? 0 : p;
1196: return (Cell)old_code_here;
1197: #endif /* !defined(NO_DYNAMIC) */
1198: }
1199: #endif /* !defined(NO_IP) */
1200:
1201: #ifndef NO_DYNAMIC
1202: int cost_codesize(int prim)
1203: {
1204: return priminfos[prim].length;
1205: }
1206: #endif
1207:
1208: int cost_ls(int prim)
1209: {
1210: struct cost *c = super_costs+prim;
1211:
1212: return c->loads + c->stores;
1213: }
1214:
1215: int cost_lsu(int prim)
1216: {
1217: struct cost *c = super_costs+prim;
1218:
1219: return c->loads + c->stores + c->updates;
1220: }
1221:
1222: int cost_nexts(int prim)
1223: {
1224: return 1;
1225: }
1226:
1227: typedef int Costfunc(int);
1228: Costfunc *ss_cost = /* cost function for optimize_bb */
1229: #ifdef NO_DYNAMIC
1230: cost_lsu;
1231: #else
1232: cost_codesize;
1233: #endif
1234:
1235: struct {
1236: Costfunc *costfunc;
1237: char *metricname;
1238: long sum;
1239: } cost_sums[] = {
1240: #ifndef NO_DYNAMIC
1241: { cost_codesize, "codesize", 0 },
1242: #endif
1243: { cost_ls, "ls", 0 },
1244: { cost_lsu, "lsu", 0 },
1245: { cost_nexts, "nexts", 0 }
1246: };
1247:
1248: #define MAX_BB 128 /* maximum number of instructions in BB */
1249: #define INF_COST 1000000 /* infinite cost */
1250: #define CANONICAL_STATE 0
1251:
1252: struct waypoint {
1253: int cost; /* the cost from here to the end */
1254: PrimNum inst; /* the inst used from here to the next waypoint */
1255: char relocatable; /* the last non-transition was relocatable */
1256: char no_transition; /* don't use the next transition (relocatability)
1257: * or this transition (does not change state) */
1258: };
1259:
1260: void init_waypoints(struct waypoint ws[])
1261: {
1262: int k;
1263:
1264: for (k=0; k<maxstates; k++)
1265: ws[k].cost=INF_COST;
1266: }
1267:
1268: void transitions(struct waypoint inst[], struct waypoint trans[])
1269: {
1270: int k;
1271: struct super_state *l;
1272:
1273: for (k=0; k<maxstates; k++) {
1274: trans[k] = inst[k];
1275: trans[k].no_transition = 1;
1276: }
1277: for (l = state_transitions; l != NULL; l = l->next) {
1278: PrimNum s = l->super;
1279: int jcost;
1280: struct cost *c=super_costs+s;
1281: struct waypoint *wi=&(trans[c->state_in]);
1282: struct waypoint *wo=&(inst[c->state_out]);
1283: if (wo->cost == INF_COST)
1284: continue;
1285: jcost = wo->cost + ss_cost(s);
1286: if (jcost <= wi->cost) {
1287: wi->cost = jcost;
1288: wi->inst = s;
1289: wi->relocatable = wo->relocatable;
1290: wi->no_transition = 0;
1291: /* if (ss_greedy) wi->cost = wo->cost ? */
1292: }
1293: }
1294: }
1295:
1296: /* use dynamic programming to find the shortest paths within the basic
1297: block origs[0..ninsts-1] and rewrite the instructions pointed to by
1298: instps to use it */
1299: void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
1300: {
1301: int i,j;
1302: static struct waypoint inst[MAX_BB+1][MAX_STATE]; /* before instruction*/
1303: static struct waypoint trans[MAX_BB+1][MAX_STATE]; /* before transition */
1304: int nextdyn, nextstate, no_transition;
1305:
1306: init_waypoints(inst[ninsts]);
1307: inst[ninsts][CANONICAL_STATE].cost=0;
1308: transitions(inst[ninsts],trans[ninsts]);
1309: for (i=ninsts-1; i>=0; i--) {
1310: init_waypoints(inst[i]);
1311: for (j=1; j<=max_super && i+j<=ninsts; j++) {
1312: struct super_state **superp = lookup_super(origs+i, j);
1313: if (superp!=NULL) {
1314: struct super_state *supers = *superp;
1315: for (; supers!=NULL; supers = supers->next) {
1316: PrimNum s = supers->super;
1317: int jcost;
1318: struct cost *c=super_costs+s;
1319: struct waypoint *wi=&(inst[i][c->state_in]);
1320: struct waypoint *wo=&(trans[i+j][c->state_out]);
1321: int no_transition = wo->no_transition;
1322: if (!(is_relocatable(s)) && !wo->relocatable) {
1323: wo=&(inst[i+j][c->state_out]);
1324: no_transition=1;
1325: }
1326: if (wo->cost == INF_COST)
1327: continue;
1328: jcost = wo->cost + ss_cost(s);
1329: if (jcost <= wi->cost) {
1330: wi->cost = jcost;
1331: wi->inst = s;
1332: wi->relocatable = is_relocatable(s);
1333: wi->no_transition = no_transition;
1334: /* if (ss_greedy) wi->cost = wo->cost ? */
1335: }
1336: }
1337: }
1338: }
1339: transitions(inst[i],trans[i]);
1340: }
1341: /* now rewrite the instructions */
1342: nextdyn=0;
1343: nextstate=CANONICAL_STATE;
1344: no_transition = ((!trans[0][nextstate].relocatable)
1345: ||trans[0][nextstate].no_transition);
1346: for (i=0; i<ninsts; i++) {
1347: Cell tc=0, tc2;
1348: if (i==nextdyn) {
1349: if (!no_transition) {
1350: /* process trans */
1351: PrimNum p = trans[i][nextstate].inst;
1352: struct cost *c = super_costs+p;
1353: assert(trans[i][nextstate].cost != INF_COST);
1354: assert(c->state_in==nextstate);
1355: tc = compile_prim_dyn(p,NULL);
1356: nextstate = c->state_out;
1357: }
1358: {
1359: /* process inst */
1360: PrimNum p = inst[i][nextstate].inst;
1361: struct cost *c=super_costs+p;
1362: assert(c->state_in==nextstate);
1363: assert(inst[i][nextstate].cost != INF_COST);
1364: #if defined(GFORTH_DEBUGGING)
1365: assert(p == origs[i]);
1366: #endif
1367: tc2 = compile_prim_dyn(p,instps[i]);
1368: if (no_transition || !is_relocatable(p))
1369: /* !! actually what we care about is if and where
1370: * compile_prim_dyn() puts NEXTs */
1371: tc=tc2;
1372: no_transition = inst[i][nextstate].no_transition;
1373: nextstate = c->state_out;
1374: nextdyn += c->length;
1375: }
1376: } else {
1377: #if defined(GFORTH_DEBUGGING)
1378: assert(0);
1379: #endif
1380: tc=0;
1381: /* tc= (Cell)vm_prims[inst[i][CANONICAL_STATE].inst]; */
1382: }
1383: *(instps[i]) = tc;
1384: }
1385: if (!no_transition) {
1386: PrimNum p = trans[i][nextstate].inst;
1387: struct cost *c = super_costs+p;
1388: assert(c->state_in==nextstate);
1389: assert(trans[i][nextstate].cost != INF_COST);
1390: assert(i==nextdyn);
1391: (void)compile_prim_dyn(p,NULL);
1392: nextstate = c->state_out;
1393: }
1394: assert(nextstate==CANONICAL_STATE);
1395: }
1396:
1397: /* compile *start, possibly rewriting it into a static and/or dynamic
1398: superinstruction */
1399: void compile_prim1(Cell *start)
1400: {
1401: #if defined(DOUBLY_INDIRECT)
1402: Label prim;
1403:
1404: if (start==NULL)
1405: return;
1406: prim = (Label)*start;
1407: if (prim<((Label)(xts+DOESJUMP)) || prim>((Label)(xts+npriminfos))) {
1408: fprintf(stderr,"compile_prim encountered xt %p\n", prim);
1409: *start=(Cell)prim;
1410: return;
1411: } else {
1412: *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
1413: return;
1414: }
1415: #elif defined(INDIRECT_THREADED)
1416: return;
1417: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1418: /* !! does not work, for unknown reasons; but something like this is
1419: probably needed to ensure that we don't call compile_prim_dyn
1420: before the inline arguments are there */
1421: static Cell *instps[MAX_BB];
1422: static PrimNum origs[MAX_BB];
1423: static int ninsts=0;
1424: PrimNum prim_num;
1425:
1426: if (start==NULL || ninsts >= MAX_BB ||
1427: (ninsts>0 && superend[origs[ninsts-1]])) {
1428: /* after bb, or at the start of the next bb */
1429: optimize_rewrite(instps,origs,ninsts);
1430: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
1431: ninsts=0;
1432: if (start==NULL)
1433: return;
1434: }
1435: prim_num = ((Xt)*start)-vm_prims;
1436: if(prim_num >= npriminfos) {
1437: optimize_rewrite(instps,origs,ninsts);
1438: /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
1439: ninsts=0;
1440: return;
1441: }
1442: assert(ninsts<MAX_BB);
1443: instps[ninsts] = start;
1444: origs[ninsts] = prim_num;
1445: ninsts++;
1446: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1447: }
1448:
1449: Address loader(FILE *imagefile, char* filename)
1450: /* returns the address of the image proper (after the preamble) */
1451: {
1452: ImageHeader header;
1453: Address image;
1454: Address imp; /* image+preamble */
1455: Char magic[8];
1456: char magic7; /* size byte of magic number */
1457: Cell preamblesize=0;
1458: Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1459: UCell check_sum;
1460: Cell ausize = ((RELINFOBITS == 8) ? 0 :
1461: (RELINFOBITS == 16) ? 1 :
1462: (RELINFOBITS == 32) ? 2 : 3);
1463: Cell charsize = ((sizeof(Char) == 1) ? 0 :
1464: (sizeof(Char) == 2) ? 1 :
1465: (sizeof(Char) == 4) ? 2 : 3) + ausize;
1466: Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
1467: (sizeof(Cell) == 2) ? 1 :
1468: (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1469: Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
1470: #ifdef WORDS_BIGENDIAN
1471: 0
1472: #else
1473: 1
1474: #endif
1475: ;
1476:
1477: vm_prims = engine(0,0,0,0,0);
1478: check_prims(vm_prims);
1479: prepare_super_table();
1480: #ifndef DOUBLY_INDIRECT
1481: #ifdef PRINT_SUPER_LENGTHS
1482: print_super_lengths();
1483: #endif
1484: check_sum = checksum(vm_prims);
1485: #else /* defined(DOUBLY_INDIRECT) */
1486: check_sum = (UCell)vm_prims;
1487: #endif /* defined(DOUBLY_INDIRECT) */
1488:
1489: do {
1490: if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1491: fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
1492: progname, filename);
1493: exit(1);
1494: }
1495: preamblesize+=8;
1496: } while(memcmp(magic,"Gforth3",7));
1497: magic7 = magic[7];
1498: if (debug) {
1499: magic[7]='\0';
1500: fprintf(stderr,"Magic found: %s ", magic);
1501: print_sizes(magic7);
1502: }
1503:
1504: if (magic7 != sizebyte)
1505: {
1506: fprintf(stderr,"This image is: ");
1507: print_sizes(magic7);
1508: fprintf(stderr,"whereas the machine is ");
1509: print_sizes(sizebyte);
1510: exit(-2);
1511: };
1512:
1513: fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1514:
1515: set_stack_sizes(&header);
1516:
1517: #if HAVE_GETPAGESIZE
1518: pagesize=getpagesize(); /* Linux/GNU libc offers this */
1519: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
1520: pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
1521: #elif PAGESIZE
1522: pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
1523: #endif
1524: if (debug)
1525: fprintf(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1526:
1527: image = dict_alloc_read(imagefile, preamblesize+header.image_size,
1528: preamblesize+dictsize, data_offset);
1529: imp=image+preamblesize;
1530: alloc_stacks((ImageHeader *)imp);
1531: if (clear_dictionary)
1532: memset(imp+header.image_size, 0, dictsize-header.image_size);
1533: if(header.base==0 || header.base == (Address)0x100) {
1534: Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
1535: char reloc_bits[reloc_size];
1536: fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1537: fread(reloc_bits, 1, reloc_size, imagefile);
1538: relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1539: #if 0
1540: { /* let's see what the relocator did */
1541: FILE *snapshot=fopen("snapshot.fi","wb");
1542: fwrite(image,1,imagesize,snapshot);
1543: fclose(snapshot);
1544: }
1545: #endif
1546: }
1547: else if(header.base!=imp) {
1548: fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
1549: progname, (unsigned long)header.base, (unsigned long)imp);
1550: exit(1);
1551: }
1552: if (header.checksum==0)
1553: ((ImageHeader *)imp)->checksum=check_sum;
1554: else if (header.checksum != check_sum) {
1555: fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
1556: progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
1557: exit(1);
1558: }
1559: #ifdef DOUBLY_INDIRECT
1560: ((ImageHeader *)imp)->xt_base = xts;
1561: #endif
1562: fclose(imagefile);
1563:
1564: /* unnecessary, except maybe for CODE words */
1565: /* FLUSH_ICACHE(imp, header.image_size);*/
1566:
1567: return imp;
1568: }
1569:
1570: /* pointer to last '/' or '\' in file, 0 if there is none. */
1571: char *onlypath(char *filename)
1572: {
1573: return strrchr(filename, DIRSEP);
1574: }
1575:
1576: FILE *openimage(char *fullfilename)
1577: {
1578: FILE *image_file;
1579: char * expfilename = tilde_cstr(fullfilename, strlen(fullfilename), 1);
1580:
1581: image_file=fopen(expfilename,"rb");
1582: if (image_file!=NULL && debug)
1583: fprintf(stderr, "Opened image file: %s\n", expfilename);
1584: return image_file;
1585: }
1586:
1587: /* try to open image file concat(path[0:len],imagename) */
1588: FILE *checkimage(char *path, int len, char *imagename)
1589: {
1590: int dirlen=len;
1591: char fullfilename[dirlen+strlen(imagename)+2];
1592:
1593: memcpy(fullfilename, path, dirlen);
1594: if (fullfilename[dirlen-1]!=DIRSEP)
1595: fullfilename[dirlen++]=DIRSEP;
1596: strcpy(fullfilename+dirlen,imagename);
1597: return openimage(fullfilename);
1598: }
1599:
1600: FILE * open_image_file(char * imagename, char * path)
1601: {
1602: FILE * image_file=NULL;
1603: char *origpath=path;
1604:
1605: if(strchr(imagename, DIRSEP)==NULL) {
1606: /* first check the directory where the exe file is in !! 01may97jaw */
1607: if (onlypath(progname))
1608: image_file=checkimage(progname, onlypath(progname)-progname, imagename);
1609: if (!image_file)
1610: do {
1611: char *pend=strchr(path, PATHSEP);
1612: if (pend==NULL)
1613: pend=path+strlen(path);
1614: if (strlen(path)==0) break;
1615: image_file=checkimage(path, pend-path, imagename);
1616: path=pend+(*pend==PATHSEP);
1617: } while (image_file==NULL);
1618: } else {
1619: image_file=openimage(imagename);
1620: }
1621:
1622: if (!image_file) {
1623: fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
1624: progname, imagename, origpath);
1625: exit(1);
1626: }
1627:
1628: return image_file;
1629: }
1630: #endif
1631:
1632: #ifdef HAS_OS
1633: UCell convsize(char *s, UCell elemsize)
1634: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
1635: of bytes. the letter at the end indicates the unit, where e stands
1636: for the element size. default is e */
1637: {
1638: char *endp;
1639: UCell n,m;
1640:
1641: m = elemsize;
1642: n = strtoul(s,&endp,0);
1643: if (endp!=NULL) {
1644: if (strcmp(endp,"b")==0)
1645: m=1;
1646: else if (strcmp(endp,"k")==0)
1647: m=1024;
1648: else if (strcmp(endp,"M")==0)
1649: m=1024*1024;
1650: else if (strcmp(endp,"G")==0)
1651: m=1024*1024*1024;
1652: else if (strcmp(endp,"T")==0) {
1653: #if (SIZEOF_CHAR_P > 4)
1654: m=1024L*1024*1024*1024;
1655: #else
1656: fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
1657: exit(1);
1658: #endif
1659: } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
1660: fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
1661: exit(1);
1662: }
1663: }
1664: return n*m;
1665: }
1666:
1667: enum {
1668: ss_number = 256,
1669: ss_states,
1670: ss_min_codesize,
1671: ss_min_ls,
1672: ss_min_lsu,
1673: ss_min_nexts,
1674: };
1675:
1676: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
1677: {
1678: int c;
1679:
1680: opterr=0;
1681: while (1) {
1682: int option_index=0;
1683: static struct option opts[] = {
1684: {"appl-image", required_argument, NULL, 'a'},
1685: {"image-file", required_argument, NULL, 'i'},
1686: {"dictionary-size", required_argument, NULL, 'm'},
1687: {"data-stack-size", required_argument, NULL, 'd'},
1688: {"return-stack-size", required_argument, NULL, 'r'},
1689: {"fp-stack-size", required_argument, NULL, 'f'},
1690: {"locals-stack-size", required_argument, NULL, 'l'},
1691: {"path", required_argument, NULL, 'p'},
1692: {"version", no_argument, NULL, 'v'},
1693: {"help", no_argument, NULL, 'h'},
1694: /* put something != 0 into offset_image */
1695: {"offset-image", no_argument, &offset_image, 1},
1696: {"no-offset-im", no_argument, &offset_image, 0},
1697: {"clear-dictionary", no_argument, &clear_dictionary, 1},
1698: {"die-on-signal", no_argument, &die_on_signal, 1},
1699: {"debug", no_argument, &debug, 1},
1700: {"no-super", no_argument, &no_super, 1},
1701: {"no-dynamic", no_argument, &no_dynamic, 1},
1702: {"dynamic", no_argument, &no_dynamic, 0},
1703: {"print-metrics", no_argument, &print_metrics, 1},
1704: {"ss-number", required_argument, NULL, ss_number},
1705: {"ss-states", required_argument, NULL, ss_states},
1706: #ifndef NO_DYNAMIC
1707: {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
1708: #endif
1709: {"ss-min-ls", no_argument, NULL, ss_min_ls},
1710: {"ss-min-lsu", no_argument, NULL, ss_min_lsu},
1711: {"ss-min-nexts", no_argument, NULL, ss_min_nexts},
1712: {"ss-greedy", no_argument, &ss_greedy, 1},
1713: {0,0,0,0}
1714: /* no-init-file, no-rc? */
1715: };
1716:
1717: c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
1718:
1719: switch (c) {
1720: case EOF: return;
1721: case '?': optind--; return;
1722: case 'a': *imagename = optarg; return;
1723: case 'i': *imagename = optarg; break;
1724: case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
1725: case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
1726: case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
1727: case 'f': fsize = convsize(optarg,sizeof(Float)); break;
1728: case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
1729: case 'p': *path = optarg; break;
1730: case 'o': offset_image = 1; break;
1731: case 'n': offset_image = 0; break;
1732: case 'c': clear_dictionary = 1; break;
1733: case 's': die_on_signal = 1; break;
1734: case 'x': debug = 1; break;
1735: case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
1736: case ss_number: static_super_number = atoi(optarg); break;
1737: case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
1738: #ifndef NO_DYNAMIC
1739: case ss_min_codesize: ss_cost = cost_codesize; break;
1740: #endif
1741: case ss_min_ls: ss_cost = cost_ls; break;
1742: case ss_min_lsu: ss_cost = cost_lsu; break;
1743: case ss_min_nexts: ss_cost = cost_nexts; break;
1744: case 'h':
1745: fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
1746: Engine Options:\n\
1747: --appl-image FILE equivalent to '--image-file=FILE --'\n\
1748: --clear-dictionary Initialize the dictionary with 0 bytes\n\
1749: -d SIZE, --data-stack-size=SIZE Specify data stack size\n\
1750: --debug Print debugging information during startup\n\
1751: --die-on-signal exit instead of CATCHing some signals\n\
1752: --dynamic use dynamic native code\n\
1753: -f SIZE, --fp-stack-size=SIZE Specify floating point stack size\n\
1754: -h, --help Print this message and exit\n\
1755: -i FILE, --image-file=FILE Use image FILE instead of `gforth.fi'\n\
1756: -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
1757: -m SIZE, --dictionary-size=SIZE Specify Forth dictionary size\n\
1758: --no-dynamic Use only statically compiled primitives\n\
1759: --no-offset-im Load image at normal position\n\
1760: --no-super No dynamically formed superinstructions\n\
1761: --offset-image Load image at a different position\n\
1762: -p PATH, --path=PATH Search path for finding image and sources\n\
1763: --print-metrics Print some code generation metrics on exit\n\
1764: -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
1765: --ss-greedy greedy, not optimal superinst selection\n\
1766: --ss-min-codesize select superinsts for smallest native code\n\
1767: --ss-min-ls minimize loads and stores\n\
1768: --ss-min-lsu minimize loads, stores, and pointer updates\n\
1769: --ss-min-nexts minimize the number of static superinsts\n\
1770: --ss-number=N use N static superinsts (default max)\n\
1771: --ss-states=N N states for stack caching (default max)\n\
1772: -v, --version Print engine version and exit\n\
1773: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
1774: `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
1775: argv[0]);
1776: optind--;
1777: return;
1778: }
1779: }
1780: }
1781: #endif
1782:
1783: #ifdef INCLUDE_IMAGE
1784: extern Cell image[];
1785: extern const char reloc_bits[];
1786: #endif
1787:
1788: int main(int argc, char **argv, char **env)
1789: {
1790: #ifdef HAS_OS
1791: char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
1792: #else
1793: char *path = DEFAULTPATH;
1794: #endif
1795: #ifndef INCLUDE_IMAGE
1796: char *imagename="gforth.fi";
1797: FILE *image_file;
1798: Address image;
1799: #endif
1800: int retvalue;
1801:
1802: #if defined(i386) && defined(ALIGNMENT_CHECK)
1803: /* turn on alignment checks on the 486.
1804: * on the 386 this should have no effect. */
1805: __asm__("pushfl; popl %eax; orl $0x40000, %eax; pushl %eax; popfl;");
1806: /* this is unusable with Linux' libc.4.6.27, because this library is
1807: not alignment-clean; we would have to replace some library
1808: functions (e.g., memcpy) to make it work. Also GCC doesn't try to keep
1809: the stack FP-aligned. */
1810: #endif
1811:
1812: /* buffering of the user output device */
1813: #ifdef _IONBF
1814: if (isatty(fileno(stdout))) {
1815: fflush(stdout);
1816: setvbuf(stdout,NULL,_IONBF,0);
1817: }
1818: #endif
1819:
1820: progname = argv[0];
1821:
1822: #ifdef HAS_OS
1823: gforth_args(argc, argv, &path, &imagename);
1824: #ifndef NO_DYNAMIC
1825: if (no_dynamic && ss_cost == cost_codesize) {
1826: ss_cost = cost_nexts;
1827: cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
1828: if (debug)
1829: fprintf(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
1830: }
1831: #endif /* !defined(NO_DYNAMIC) */
1832: #endif /* defined(HAS_OS) */
1833:
1834: #ifdef INCLUDE_IMAGE
1835: set_stack_sizes((ImageHeader *)image);
1836: if(((ImageHeader *)image)->base != image)
1837: relocate(image, reloc_bits, ((ImageHeader *)image)->image_size,
1838: (Label*)engine(0, 0, 0, 0, 0));
1839: alloc_stacks((ImageHeader *)image);
1840: #else
1841: image_file = open_image_file(imagename, path);
1842: image = loader(image_file, imagename);
1843: #endif
1844: gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
1845:
1846: {
1847: char path2[strlen(path)+1];
1848: char *p1, *p2;
1849: Cell environ[]= {
1850: (Cell)argc-(optind-1),
1851: (Cell)(argv+(optind-1)),
1852: (Cell)strlen(path),
1853: (Cell)path2};
1854: argv[optind-1] = progname;
1855: /*
1856: for (i=0; i<environ[0]; i++)
1857: printf("%s\n", ((char **)(environ[1]))[i]);
1858: */
1859: /* make path OS-independent by replacing path separators with NUL */
1860: for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
1861: if (*p1==PATHSEP)
1862: *p2 = '\0';
1863: else
1864: *p2 = *p1;
1865: *p2='\0';
1866: retvalue = go_forth(image, 4, environ);
1867: #ifdef SIGPIPE
1868: bsd_signal(SIGPIPE, SIG_IGN);
1869: #endif
1870: #ifdef VM_PROFILING
1871: vm_print_profile(stderr);
1872: #endif
1873: deprep_terminal();
1874: }
1875: if (print_metrics) {
1876: int i;
1877: fprintf(stderr, "code size = %8ld\n", dyncodesize());
1878: for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
1879: fprintf(stderr, "metric %8s: %8ld\n",
1880: cost_sums[i].metricname, cost_sums[i].sum);
1881: }
1882: return retvalue;
1883: }
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