File:  [gforth] / gforth / engine / main.c
Revision 1.155: download - view: text, annotated - select for diffs
Wed Nov 9 15:46:28 2005 UTC (18 years, 5 months ago) by anton
Branches: MAIN
CVS tags: HEAD
initial work for tree-parsing automaton for stack caching

    1: /* command line interpretation, image loading etc. for Gforth
    2: 
    3: 
    4:   Copyright (C) 1995,1996,1997,1998,2000,2003,2004 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 HAS_LIBFFI
   92: Cell *RP;
   93: Address LP;
   94: 
   95: #include <ffi.h>
   96: 
   97: void ** clist;
   98: void * ritem;
   99: 
  100: void ffi_callback(ffi_cif * cif, void * resp, void ** args, Xt * ip)
  101: {
  102:   Cell *rp = RP;
  103:   Cell *sp = SP;
  104:   Float *fp = FP;
  105:   Address lp = LP;
  106: 
  107:   clist = args;
  108:   ritem = resp;
  109: 
  110:   engine(ip, sp, rp, fp, lp);
  111: 
  112:   /* restore global variables */
  113:   RP = rp;
  114:   SP = sp;
  115:   FP = fp;
  116:   LP = lp;
  117: }
  118: #endif
  119: 
  120: #ifdef GFORTH_DEBUGGING
  121: /* define some VM registers as global variables, so they survive exceptions;
  122:    global register variables are not up to the task (according to the 
  123:    GNU C manual) */
  124: Xt *saved_ip;
  125: Cell *rp;
  126: #endif
  127: 
  128: #ifdef NO_IP
  129: Label next_code;
  130: #endif
  131: 
  132: #ifdef HAS_FILE
  133: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
  134: char* pfileattr[6]={"r","r","r+","r+","w","w"};
  135: 
  136: #ifndef O_BINARY
  137: #define O_BINARY 0
  138: #endif
  139: #ifndef O_TEXT
  140: #define O_TEXT 0
  141: #endif
  142: 
  143: int ufileattr[6]= {
  144:   O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
  145:   O_RDWR  |O_BINARY, O_RDWR  |O_BINARY,
  146:   O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
  147: #endif
  148: /* end global vars for engine.c */
  149: 
  150: #define PRIM_VERSION 1
  151: /* increment this whenever the primitives change in an incompatible way */
  152: 
  153: #ifndef DEFAULTPATH
  154: #  define DEFAULTPATH "."
  155: #endif
  156: 
  157: #ifdef MSDOS
  158: jmp_buf throw_jmp_buf;
  159: #endif
  160: 
  161: #if defined(DOUBLY_INDIRECT)
  162: #  define CFA(n)	({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
  163: #else
  164: #  define CFA(n)	((Cell)(symbols+((n)&~0x4000UL)))
  165: #endif
  166: 
  167: #define maxaligned(n)	(typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
  168: 
  169: static UCell dictsize=0;
  170: static UCell dsize=0;
  171: static UCell rsize=0;
  172: static UCell fsize=0;
  173: static UCell lsize=0;
  174: int offset_image=0;
  175: int die_on_signal=0;
  176: #ifndef INCLUDE_IMAGE
  177: static int clear_dictionary=0;
  178: UCell pagesize=1;
  179: char *progname;
  180: #else
  181: char *progname = "gforth";
  182: int optind = 1;
  183: #endif
  184: 
  185: #define CODE_BLOCK_SIZE (4096*1024) /* !! overflow handling for -native */
  186: Address code_area=0;
  187: Cell code_area_size = CODE_BLOCK_SIZE;
  188: Address code_here=NULL+CODE_BLOCK_SIZE; /* does for code-area what HERE
  189: 					   does for the dictionary */
  190: Address start_flush=NULL; /* start of unflushed code */
  191: Cell last_jump=0; /* if the last prim was compiled without jump, this
  192:                      is it's number, otherwise this contains 0 */
  193: 
  194: static int no_super=0;   /* true if compile_prim should not fuse prims */
  195: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
  196: 					     dynamically */
  197: static int print_metrics=0; /* if true, print metrics on exit */
  198: static int static_super_number = 10000000; /* number of ss used if available */
  199: #define MAX_STATE 9 /* maximum number of states */
  200: static int maxstates = MAX_STATE; /* number of states for stack caching */
  201: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
  202: static int diag = 0; /* if true: print diagnostic informations */
  203: static int relocs = 0;
  204: static int nonrelocs = 0;
  205: 
  206: #ifdef HAS_DEBUG
  207: int debug=0;
  208: # define debugp(x...) if (debug) fprintf(x);
  209: #else
  210: # define perror(x...)
  211: # define fprintf(x...)
  212: # define debugp(x...)
  213: #endif
  214: 
  215: ImageHeader *gforth_header;
  216: Label *vm_prims;
  217: #ifdef DOUBLY_INDIRECT
  218: Label *xts; /* same content as vm_prims, but should only be used for xts */
  219: #endif
  220: 
  221: #ifndef NO_DYNAMIC
  222: #define MAX_IMMARGS 2
  223: 
  224: typedef struct {
  225:   Label start; /* NULL if not relocatable */
  226:   Cell length; /* only includes the jump iff superend is true*/
  227:   Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
  228:   char superend; /* true if primitive ends superinstruction, i.e.,
  229:                      unconditional branch, execute, etc. */
  230:   Cell nimmargs;
  231:   struct immarg {
  232:     Cell offset; /* offset of immarg within prim */
  233:     char rel;    /* true if immarg is relative */
  234:   } immargs[MAX_IMMARGS];
  235: } PrimInfo;
  236: 
  237: PrimInfo *priminfos;
  238: PrimInfo **decomp_prims;
  239: 
  240: const char const* const prim_names[]={
  241: #include PRIM_NAMES_I
  242: };
  243: 
  244: void init_ss_cost(void);
  245: 
  246: static int is_relocatable(int p)
  247: {
  248:   return !no_dynamic && priminfos[p].start != NULL;
  249: }
  250: #else /* defined(NO_DYNAMIC) */
  251: static int is_relocatable(int p)
  252: {
  253:   return 0;
  254: }
  255: #endif /* defined(NO_DYNAMIC) */
  256: 
  257: #ifdef MEMCMP_AS_SUBROUTINE
  258: int gforth_memcmp(const char * s1, const char * s2, size_t n)
  259: {
  260:   return memcmp(s1, s2, n);
  261: }
  262: #endif
  263: 
  264: static Cell max(Cell a, Cell b)
  265: {
  266:   return a>b?a:b;
  267: }
  268: 
  269: static Cell min(Cell a, Cell b)
  270: {
  271:   return a<b?a:b;
  272: }
  273: 
  274: /* image file format:
  275:  *  "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
  276:  *   padding to a multiple of 8
  277:  *   magic: "Gforth3x" means format 0.6,
  278:  *              where x is a byte with
  279:  *              bit 7:   reserved = 0
  280:  *              bit 6:5: address unit size 2^n octets
  281:  *              bit 4:3: character size 2^n octets
  282:  *              bit 2:1: cell size 2^n octets
  283:  *              bit 0:   endian, big=0, little=1.
  284:  *  The magic are always 8 octets, no matter what the native AU/character size is
  285:  *  padding to max alignment (no padding necessary on current machines)
  286:  *  ImageHeader structure (see forth.h)
  287:  *  data (size in ImageHeader.image_size)
  288:  *  tags ((if relocatable, 1 bit/data cell)
  289:  *
  290:  * tag==1 means that the corresponding word is an address;
  291:  * If the word is >=0, the address is within the image;
  292:  * addresses within the image are given relative to the start of the image.
  293:  * If the word =-1 (CF_NIL), the address is NIL,
  294:  * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
  295:  * If the word =CF(DODOES), it's a DOES> CFA
  296:  * If the word =CF(DOESJUMP), it's a DOES JUMP (2 Cells after DOES>,
  297:  *					possibly containing a jump to dodoes)
  298:  * If the word is <CF(DOESJUMP) and bit 14 is set, it's the xt of a primitive
  299:  * If the word is <CF(DOESJUMP) and bit 14 is clear, 
  300:  *                                        it's the threaded code of a primitive
  301:  * bits 13..9 of a primitive token state which group the primitive belongs to,
  302:  * bits 8..0 of a primitive token index into the group
  303:  */
  304: 
  305: Cell groups[32] = {
  306:   0,
  307:   0
  308: #undef GROUP
  309: #undef GROUPADD
  310: #define GROUPADD(n) +n
  311: #define GROUP(x, n) , 0
  312: #include PRIM_GRP_I
  313: #undef GROUP
  314: #undef GROUPADD
  315: #define GROUP(x, n)
  316: #define GROUPADD(n)
  317: };
  318: 
  319: unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
  320: 			      int size, Cell base)
  321:      /* produce a bitmask marking all the branch targets */
  322: {
  323:   int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
  324:   Cell token;
  325:   unsigned char bits;
  326:   unsigned char *result=malloc(steps);
  327: 
  328:   memset(result, 0, steps);
  329:   for(k=0; k<steps; k++) {
  330:     for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
  331:       if(bits & (1U << (RELINFOBITS-1))) {
  332: 	assert(i*sizeof(Cell) < size);
  333:         token=image[i];
  334: 	if (token>=base) { /* relocatable address */
  335: 	  UCell bitnum=(token-base)/sizeof(Cell);
  336: 	  if (bitnum/RELINFOBITS < (UCell)steps)
  337: 	    result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
  338: 	}
  339:       }
  340:     }
  341:   }
  342:   return result;
  343: }
  344: 
  345: void relocate(Cell *image, const unsigned char *bitstring, 
  346:               int size, Cell base, Label symbols[])
  347: {
  348:   int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
  349:   Cell token;
  350:   char bits;
  351:   Cell max_symbols;
  352:   /* 
  353:    * A virtual start address that's the real start address minus 
  354:    * the one in the image 
  355:    */
  356:   Cell *start = (Cell * ) (((void *) image) - ((void *) base));
  357:   unsigned char *targets = branch_targets(image, bitstring, size, base);
  358: 
  359:   /* group index into table */
  360:   if(groups[31]==0) {
  361:     int groupsum=0;
  362:     for(i=0; i<32; i++) {
  363:       groupsum += groups[i];
  364:       groups[i] = groupsum;
  365:       /* printf("group[%d]=%d\n",i,groupsum); */
  366:     }
  367:     i=0;
  368:   }
  369:   
  370: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
  371:   
  372:   for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
  373:     ;
  374:   max_symbols--;
  375: 
  376:   for(k=0; k<steps; k++) {
  377:     for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
  378:       /*      fprintf(stderr,"relocate: image[%d]\n", i);*/
  379:       if(bits & (1U << (RELINFOBITS-1))) {
  380: 	assert(i*sizeof(Cell) < size);
  381: 	/* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
  382:         token=image[i];
  383: 	if(token<0) {
  384: 	  int group = (-token & 0x3E00) >> 9;
  385: 	  if(group == 0) {
  386: 	    switch(token|0x4000) {
  387: 	    case CF_NIL      : image[i]=0; break;
  388: #if !defined(DOUBLY_INDIRECT)
  389: 	    case CF(DOCOL)   :
  390: 	    case CF(DOVAR)   :
  391: 	    case CF(DOCON)   :
  392: 	    case CF(DOUSER)  : 
  393: 	    case CF(DODEFER) : 
  394: 	    case CF(DOFIELD) : MAKE_CF(image+i,symbols[CF(token)]); break;
  395: 	    case CF(DOESJUMP): image[i]=0; break;
  396: #endif /* !defined(DOUBLY_INDIRECT) */
  397: 	    case CF(DODOES)  :
  398: 	      MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
  399: 	      break;
  400: 	    default          : /* backward compatibility */
  401: /*	      printf("Code field generation image[%x]:=CFA(%x)\n",
  402: 		     i, CF(image[i])); */
  403: 	      if (CF((token | 0x4000))<max_symbols) {
  404: 		image[i]=(Cell)CFA(CF(token));
  405: #ifdef DIRECT_THREADED
  406: 		if ((token & 0x4000) == 0) { /* threade code, no CFA */
  407: 		  if (targets[k] & (1U<<(RELINFOBITS-1-j)))
  408: 		    compile_prim1(0);
  409: 		  compile_prim1(&image[i]);
  410: 		}
  411: #endif
  412: 	      } else
  413: 		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);
  414: 	    }
  415: 	  } else {
  416: 	    int tok = -token & 0x1FF;
  417: 	    if (tok < (groups[group+1]-groups[group])) {
  418: #if defined(DOUBLY_INDIRECT)
  419: 	      image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
  420: #else
  421: 	      image[i]=(Cell)CFA((groups[group]+tok));
  422: #endif
  423: #ifdef DIRECT_THREADED
  424: 	      if ((token & 0x4000) == 0) { /* threade code, no CFA */
  425: 		if (targets[k] & (1U<<(RELINFOBITS-1-j)))
  426: 		  compile_prim1(0);
  427: 		compile_prim1(&image[i]);
  428: 	      }
  429: #endif
  430: 	    } else
  431: 	      fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
  432: 	  }
  433: 	} else {
  434:           /* if base is > 0: 0 is a null reference so don't adjust*/
  435:           if (token>=base) {
  436:             image[i]+=(Cell)start;
  437:           }
  438:         }
  439:       }
  440:     }
  441:   }
  442:   free(targets);
  443:   finish_code();
  444:   ((ImageHeader*)(image))->base = (Address) image;
  445: }
  446: 
  447: UCell checksum(Label symbols[])
  448: {
  449:   UCell r=PRIM_VERSION;
  450:   Cell i;
  451: 
  452:   for (i=DOCOL; i<=DOESJUMP; i++) {
  453:     r ^= (UCell)(symbols[i]);
  454:     r = (r << 5) | (r >> (8*sizeof(Cell)-5));
  455:   }
  456: #ifdef DIRECT_THREADED
  457:   /* we have to consider all the primitives */
  458:   for (; symbols[i]!=(Label)0; i++) {
  459:     r ^= (UCell)(symbols[i]);
  460:     r = (r << 5) | (r >> (8*sizeof(Cell)-5));
  461:   }
  462: #else
  463:   /* in indirect threaded code all primitives are accessed through the
  464:      symbols table, so we just have to put the base address of symbols
  465:      in the checksum */
  466:   r ^= (UCell)symbols;
  467: #endif
  468:   return r;
  469: }
  470: 
  471: Address verbose_malloc(Cell size)
  472: {
  473:   Address r;
  474:   /* leave a little room (64B) for stack underflows */
  475:   if ((r = malloc(size+64))==NULL) {
  476:     perror(progname);
  477:     exit(1);
  478:   }
  479:   r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
  480:   debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
  481:   return r;
  482: }
  483: 
  484: static Address next_address=0;
  485: void after_alloc(Address r, Cell size)
  486: {
  487:   if (r != (Address)-1) {
  488:     debugp(stderr, "success, address=$%lx\n", (long) r);
  489:     if (pagesize != 1)
  490:       next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
  491:   } else {
  492:     debugp(stderr, "failed: %s\n", strerror(errno));
  493:   }
  494: }
  495: 
  496: #ifndef MAP_FAILED
  497: #define MAP_FAILED ((Address) -1)
  498: #endif
  499: #ifndef MAP_FILE
  500: # define MAP_FILE 0
  501: #endif
  502: #ifndef MAP_PRIVATE
  503: # define MAP_PRIVATE 0
  504: #endif
  505: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
  506: # define MAP_ANON MAP_ANONYMOUS
  507: #endif
  508: 
  509: #if defined(HAVE_MMAP)
  510: static Address alloc_mmap(Cell size)
  511: {
  512:   Address r;
  513: 
  514: #if defined(MAP_ANON)
  515:   debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
  516:   r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE, -1, 0);
  517: #else /* !defined(MAP_ANON) */
  518:   /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
  519:      apparently defaults) */
  520:   static int dev_zero=-1;
  521: 
  522:   if (dev_zero == -1)
  523:     dev_zero = open("/dev/zero", O_RDONLY);
  524:   if (dev_zero == -1) {
  525:     r = MAP_FAILED;
  526:     debugp(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ", 
  527: 	      strerror(errno));
  528:   } else {
  529:     debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
  530:     r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE, dev_zero, 0);
  531:   }
  532: #endif /* !defined(MAP_ANON) */
  533:   after_alloc(r, size);
  534:   return r;  
  535: }
  536: #endif
  537: 
  538: Address my_alloc(Cell size)
  539: {
  540: #if HAVE_MMAP
  541:   Address r;
  542: 
  543:   r=alloc_mmap(size);
  544:   if (r!=(Address)MAP_FAILED)
  545:     return r;
  546: #endif /* HAVE_MMAP */
  547:   /* use malloc as fallback */
  548:   return verbose_malloc(size);
  549: }
  550: 
  551: Address dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
  552: {
  553:   Address image = MAP_FAILED;
  554: 
  555: #if defined(HAVE_MMAP)
  556:   if (offset==0) {
  557:     image=alloc_mmap(dictsize);
  558:     if (image != (Address)MAP_FAILED) {
  559:       Address image1;
  560:       debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
  561:       image1 = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE, fileno(file), 0);
  562:       after_alloc(image1,dictsize);
  563:       if (image1 == (Address)MAP_FAILED)
  564: 	goto read_image;
  565:     }
  566:   }
  567: #endif /* defined(HAVE_MMAP) */
  568:   if (image == (Address)MAP_FAILED) {
  569:     image = my_alloc(dictsize+offset)+offset;
  570:   read_image:
  571:     rewind(file);  /* fseek(imagefile,0L,SEEK_SET); */
  572:     fread(image, 1, imagesize, file);
  573:   }
  574:   return image;
  575: }
  576: 
  577: void set_stack_sizes(ImageHeader * header)
  578: {
  579:   if (dictsize==0)
  580:     dictsize = header->dict_size;
  581:   if (dsize==0)
  582:     dsize = header->data_stack_size;
  583:   if (rsize==0)
  584:     rsize = header->return_stack_size;
  585:   if (fsize==0)
  586:     fsize = header->fp_stack_size;
  587:   if (lsize==0)
  588:     lsize = header->locals_stack_size;
  589:   dictsize=maxaligned(dictsize);
  590:   dsize=maxaligned(dsize);
  591:   rsize=maxaligned(rsize);
  592:   lsize=maxaligned(lsize);
  593:   fsize=maxaligned(fsize);
  594: }
  595: 
  596: void alloc_stacks(ImageHeader * header)
  597: {
  598:   header->dict_size=dictsize;
  599:   header->data_stack_size=dsize;
  600:   header->fp_stack_size=fsize;
  601:   header->return_stack_size=rsize;
  602:   header->locals_stack_size=lsize;
  603: 
  604:   header->data_stack_base=my_alloc(dsize);
  605:   header->fp_stack_base=my_alloc(fsize);
  606:   header->return_stack_base=my_alloc(rsize);
  607:   header->locals_stack_base=my_alloc(lsize);
  608: }
  609: 
  610: #warning You can ignore the warnings about clobbered variables in go_forth
  611: int go_forth(Address image, int stack, Cell *entries)
  612: {
  613:   volatile ImageHeader *image_header = (ImageHeader *)image;
  614:   Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
  615:   Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
  616:   Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
  617: #ifdef GFORTH_DEBUGGING
  618:   volatile Cell *orig_rp0=rp0;
  619: #endif
  620:   Address lp0=image_header->locals_stack_base + lsize;
  621:   Xt *ip0=(Xt *)(image_header->boot_entry);
  622: #ifdef SYSSIGNALS
  623:   int throw_code;
  624: #endif
  625: 
  626:   /* ensure that the cached elements (if any) are accessible */
  627: #if !(defined(GFORTH_DEBUGGING) || defined(INDIRECT_THREADED) || defined(DOUBLY_INDIRECT) || defined(VM_PROFILING))
  628:   sp0 -= 8; /* make stuff below bottom accessible for stack caching */
  629: #endif
  630:   IF_fpTOS(fp0--);
  631:   
  632:   for(;stack>0;stack--)
  633:     *--sp0=entries[stack-1];
  634: 
  635: #ifdef SYSSIGNALS
  636:   get_winsize();
  637:    
  638:   install_signal_handlers(); /* right place? */
  639:   
  640:   if ((throw_code=setjmp(throw_jmp_buf))) {
  641:     static Cell signal_data_stack[24];
  642:     static Cell signal_return_stack[16];
  643:     static Float signal_fp_stack[1];
  644: 
  645:     signal_data_stack[15]=throw_code;
  646: 
  647: #ifdef GFORTH_DEBUGGING
  648:     debugp(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
  649: 	      throw_code, saved_ip, rp);
  650:     if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
  651:       /* no rstack overflow or underflow */
  652:       rp0 = rp;
  653:       *--rp0 = (Cell)saved_ip;
  654:     }
  655:     else /* I love non-syntactic ifdefs :-) */
  656:       rp0 = signal_return_stack+16;
  657: #else  /* !defined(GFORTH_DEBUGGING) */
  658:     debugp(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
  659:       rp0 = signal_return_stack+16;
  660: #endif /* !defined(GFORTH_DEBUGGING) */
  661:     /* fprintf(stderr, "rp=$%x\n",rp0);*/
  662:     
  663:     return((int)(Cell)engine(image_header->throw_entry, signal_data_stack+15,
  664: 		       rp0, signal_fp_stack, 0));
  665:   }
  666: #endif
  667: 
  668:   return((int)(Cell)engine(ip0,sp0,rp0,fp0,lp0));
  669: }
  670: 
  671: #ifndef INCLUDE_IMAGE
  672: void print_sizes(Cell sizebyte)
  673:      /* print size information */
  674: {
  675:   static char* endianstring[]= { "   big","little" };
  676:   
  677:   fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
  678: 	  endianstring[sizebyte & 1],
  679: 	  1 << ((sizebyte >> 1) & 3),
  680: 	  1 << ((sizebyte >> 3) & 3),
  681: 	  1 << ((sizebyte >> 5) & 3));
  682: }
  683: 
  684: /* static superinstruction stuff */
  685: 
  686: struct cost { /* super_info might be a more accurate name */
  687:   char loads;       /* number of stack loads */
  688:   char stores;      /* number of stack stores */
  689:   char updates;     /* number of stack pointer updates */
  690:   char branch;	    /* is it a branch (SET_IP) */
  691:   unsigned char state_in;    /* state on entry */
  692:   unsigned char state_out;   /* state on exit */
  693:   unsigned char imm_ops;     /* number of immediate operands */
  694:   short offset;     /* offset into super2 table */
  695:   unsigned char length;      /* number of components */
  696: };
  697: 
  698: PrimNum super2[] = {
  699: #include SUPER2_I
  700: };
  701: 
  702: struct cost super_costs[] = {
  703: #include COSTS_I
  704: };
  705: 
  706: struct super_state {
  707:   struct super_state *next;
  708:   PrimNum super;
  709: };
  710: 
  711: #define HASH_SIZE 256
  712: 
  713: struct super_table_entry {
  714:   struct super_table_entry *next;
  715:   PrimNum *start;
  716:   short length;
  717:   struct super_state *ss_list; /* list of supers */
  718: } *super_table[HASH_SIZE];
  719: int max_super=2;
  720: 
  721: struct super_state *state_transitions=NULL;
  722: 
  723: int hash_super(PrimNum *start, int length)
  724: {
  725:   int i, r;
  726:   
  727:   for (i=0, r=0; i<length; i++) {
  728:     r <<= 1;
  729:     r += start[i];
  730:   }
  731:   return r & (HASH_SIZE-1);
  732: }
  733: 
  734: struct super_state **lookup_super(PrimNum *start, int length)
  735: {
  736:   int hash=hash_super(start,length);
  737:   struct super_table_entry *p = super_table[hash];
  738: 
  739:   /* assert(length >= 2); */
  740:   for (; p!=NULL; p = p->next) {
  741:     if (length == p->length &&
  742: 	memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
  743:       return &(p->ss_list);
  744:   }
  745:   return NULL;
  746: }
  747: 
  748: void prepare_super_table()
  749: {
  750:   int i;
  751:   int nsupers = 0;
  752: 
  753:   for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
  754:     struct cost *c = &super_costs[i];
  755:     if ((c->length < 2 || nsupers < static_super_number) &&
  756: 	c->state_in < maxstates && c->state_out < maxstates) {
  757:       struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
  758:       struct super_state *ss = malloc(sizeof(struct super_state));
  759:       ss->super= i;
  760:       if (c->offset==N_noop && i != N_noop) {
  761: 	if (is_relocatable(i)) {
  762: 	  ss->next = state_transitions;
  763: 	  state_transitions = ss;
  764: 	}
  765:       } else if (ss_listp != NULL) {
  766: 	ss->next = *ss_listp;
  767: 	*ss_listp = ss;
  768:       } else {
  769: 	int hash = hash_super(super2+c->offset, c->length);
  770: 	struct super_table_entry **p = &super_table[hash];
  771: 	struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
  772: 	ss->next = NULL;
  773: 	e->next = *p;
  774: 	e->start = super2 + c->offset;
  775: 	e->length = c->length;
  776: 	e->ss_list = ss;
  777: 	*p = e;
  778:       }
  779:       if (c->length > max_super)
  780: 	max_super = c->length;
  781:       if (c->length >= 2)
  782: 	nsupers++;
  783:     }
  784:   }
  785:   debugp(stderr, "Using %d static superinsts\n", nsupers);
  786: }
  787: 
  788: /* dynamic replication/superinstruction stuff */
  789: 
  790: #ifndef NO_DYNAMIC
  791: int compare_priminfo_length(const void *_a, const void *_b)
  792: {
  793:   PrimInfo **a = (PrimInfo **)_a;
  794:   PrimInfo **b = (PrimInfo **)_b;
  795:   Cell diff = (*a)->length - (*b)->length;
  796:   if (diff)
  797:     return diff;
  798:   else /* break ties by start address; thus the decompiler produces
  799:           the earliest primitive with the same code (e.g. noop instead
  800:           of (char) and @ instead of >code-address */
  801:     return (*b)->start - (*a)->start;
  802: }
  803: #endif /* !defined(NO_DYNAMIC) */
  804: 
  805: static char MAYBE_UNUSED superend[]={
  806: #include PRIM_SUPEREND_I
  807: };
  808: 
  809: Cell npriminfos=0;
  810: 
  811: Label goto_start;
  812: Cell goto_len;
  813: 
  814: int compare_labels(const void *pa, const void *pb)
  815: {
  816:   Label a = *(Label *)pa;
  817:   Label b = *(Label *)pb;
  818:   return a-b;
  819: }
  820: 
  821: Label bsearch_next(Label key, Label *a, UCell n)
  822:      /* a is sorted; return the label >=key that is the closest in a;
  823:         return NULL if there is no label in a >=key */
  824: {
  825:   int mid = (n-1)/2;
  826:   if (n<1)
  827:     return NULL;
  828:   if (n == 1) {
  829:     if (a[0] < key)
  830:       return NULL;
  831:     else
  832:       return a[0];
  833:   }
  834:   if (a[mid] < key)
  835:     return bsearch_next(key, a+mid+1, n-mid-1);
  836:   else
  837:     return bsearch_next(key, a, mid+1);
  838: }
  839: 
  840: void check_prims(Label symbols1[])
  841: {
  842:   int i;
  843: #ifndef NO_DYNAMIC
  844:   Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted, *goto_p;
  845:   int nends1j;
  846: #endif
  847: 
  848:   if (debug)
  849: #ifdef __VERSION__
  850:     fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
  851: #else
  852: #define xstr(s) str(s)
  853: #define str(s) #s
  854:   fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n"); 
  855: #endif
  856:   for (i=0; symbols1[i]!=0; i++)
  857:     ;
  858:   npriminfos = i;
  859:   
  860: #ifndef NO_DYNAMIC
  861:   if (no_dynamic)
  862:     return;
  863:   symbols2=engine2(0,0,0,0,0);
  864: #if NO_IP
  865:   symbols3=engine3(0,0,0,0,0);
  866: #else
  867:   symbols3=symbols1;
  868: #endif
  869:   ends1 = symbols1+i+1;
  870:   ends1j =   ends1+i;
  871:   goto_p = ends1j+i+1; /* goto_p[0]==before; ...[1]==after;*/
  872:   nends1j = i+1;
  873:   ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
  874:   memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
  875:   qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
  876: 
  877:   /* check whether the "goto *" is relocatable */
  878:   goto_len = goto_p[1]-goto_p[0];
  879:   debugp(stderr, "goto * %p %p len=%ld\n",
  880: 	 goto_p[0],symbols2[goto_p-symbols1],goto_len);
  881:   if (memcmp(goto_p[0],symbols2[goto_p-symbols1],goto_len)!=0) { /* unequal */
  882:     no_dynamic=1;
  883:     debugp(stderr,"  not relocatable, disabling dynamic code generation\n");
  884:     init_ss_cost();
  885:     return;
  886:   }
  887:   goto_start = goto_p[0];
  888:   
  889:   priminfos = calloc(i,sizeof(PrimInfo));
  890:   for (i=0; symbols1[i]!=0; i++) {
  891:     int prim_len = ends1[i]-symbols1[i];
  892:     PrimInfo *pi=&priminfos[i];
  893:     struct cost *sc=&super_costs[i];
  894:     int j=0;
  895:     char *s1 = (char *)symbols1[i];
  896:     char *s2 = (char *)symbols2[i];
  897:     char *s3 = (char *)symbols3[i];
  898:     Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
  899: 
  900:     pi->start = s1;
  901:     pi->superend = superend[i]|no_super;
  902:     pi->length = prim_len;
  903:     pi->restlength = endlabel - symbols1[i] - pi->length;
  904:     pi->nimmargs = 0;
  905:     relocs++;
  906:     debugp(stderr, "%-15s %d-%d %4d %p %p %p len=%3ld rest=%2ld send=%1d",
  907: 	   prim_names[i], sc->state_in, sc->state_out,
  908: 	   i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength),
  909: 	   pi->superend);
  910:     if (endlabel == NULL) {
  911:       pi->start = NULL; /* not relocatable */
  912:       if (pi->length<0) pi->length=100;
  913:       debugp(stderr,"\n   non_reloc: no J label > start found\n");
  914:       relocs--;
  915:       nonrelocs++;
  916:       continue;
  917:     }
  918:     if (ends1[i] > endlabel && !pi->superend) {
  919:       pi->start = NULL; /* not relocatable */
  920:       pi->length = endlabel-symbols1[i];
  921:       debugp(stderr,"\n   non_reloc: there is a J label before the K label (restlength<0)\n");
  922:       relocs--;
  923:       nonrelocs++;
  924:       continue;
  925:     }
  926:     if (ends1[i] < pi->start && !pi->superend) {
  927:       pi->start = NULL; /* not relocatable */
  928:       pi->length = endlabel-symbols1[i];
  929:       debugp(stderr,"\n   non_reloc: K label before I label (length<0)\n");
  930:       relocs--;
  931:       nonrelocs++;
  932:       continue;
  933:     }
  934:     assert(pi->length>=0);
  935:     assert(pi->restlength >=0);
  936:     while (j<(pi->length+pi->restlength)) {
  937:       if (s1[j]==s3[j]) {
  938: 	if (s1[j] != s2[j]) {
  939: 	  pi->start = NULL; /* not relocatable */
  940: 	  debugp(stderr,"\n   non_reloc: engine1!=engine2 offset %3d",j);
  941: 	  /* assert(j<prim_len); */
  942: 	  relocs--;
  943: 	  nonrelocs++;
  944: 	  break;
  945: 	}
  946: 	j++;
  947:       } else {
  948: 	struct immarg *ia=&pi->immargs[pi->nimmargs];
  949: 
  950: 	pi->nimmargs++;
  951: 	ia->offset=j;
  952: 	if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
  953: 	  ia->rel=0;
  954: 	  debugp(stderr,"\n   absolute immarg: offset %3d",j);
  955: 	} else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
  956: 		   symbols1[DOESJUMP+1]) {
  957: 	  ia->rel=1;
  958: 	  debugp(stderr,"\n   relative immarg: offset %3d",j);
  959: 	} else {
  960: 	  pi->start = NULL; /* not relocatable */
  961: 	  debugp(stderr,"\n   non_reloc: engine1!=engine3 offset %3d",j);
  962: 	  /* assert(j<prim_len);*/
  963: 	  relocs--;
  964: 	  nonrelocs++;
  965: 	  break;
  966: 	}
  967: 	j+=4;
  968:       }
  969:     }
  970:     debugp(stderr,"\n");
  971:   }
  972:   decomp_prims = calloc(i,sizeof(PrimInfo *));
  973:   for (i=DOESJUMP+1; i<npriminfos; i++)
  974:     decomp_prims[i] = &(priminfos[i]);
  975:   qsort(decomp_prims+DOESJUMP+1, npriminfos-DOESJUMP-1, sizeof(PrimInfo *),
  976: 	compare_priminfo_length);
  977: #endif
  978: }
  979: 
  980: void flush_to_here(void)
  981: {
  982: #ifndef NO_DYNAMIC
  983:   if (start_flush)
  984:     FLUSH_ICACHE(start_flush, code_here-start_flush);
  985:   start_flush=code_here;
  986: #endif
  987: }
  988: 
  989: #ifndef NO_DYNAMIC
  990: void append_jump(void)
  991: {
  992:   if (last_jump) {
  993:     PrimInfo *pi = &priminfos[last_jump];
  994:     
  995:     memcpy(code_here, pi->start+pi->length, pi->restlength);
  996:     code_here += pi->restlength;
  997:     memcpy(code_here, goto_start, goto_len);
  998:     code_here += goto_len;
  999:     last_jump=0;
 1000:   }
 1001: }
 1002: 
 1003: /* Gforth remembers all code blocks in this list.  On forgetting (by
 1004: executing a marker) the code blocks are not freed (because Gforth does
 1005: not remember how they were allocated; hmm, remembering that might be
 1006: easier and cleaner).  Instead, code_here etc. are reset to the old
 1007: value, and the "forgotten" code blocks are reused when they are
 1008: needed. */
 1009: 
 1010: struct code_block_list {
 1011:   struct code_block_list *next;
 1012:   Address block;
 1013:   Cell size;
 1014: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
 1015: 
 1016: Address append_prim(Cell p)
 1017: {
 1018:   PrimInfo *pi = &priminfos[p];
 1019:   Address old_code_here = code_here;
 1020: 
 1021:   if (code_area+code_area_size < code_here+pi->length+pi->restlength) {
 1022:     struct code_block_list *p;
 1023:     append_jump();
 1024:     flush_to_here();
 1025:     if (*next_code_blockp == NULL) {
 1026:       code_here = start_flush = code_area = my_alloc(code_area_size);
 1027:       p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
 1028:       *next_code_blockp = p;
 1029:       p->next = NULL;
 1030:       p->block = code_here;
 1031:       p->size = code_area_size;
 1032:     } else {
 1033:       p = *next_code_blockp;
 1034:       code_here = start_flush = code_area = p->block;
 1035:     }
 1036:     old_code_here = code_here;
 1037:     next_code_blockp = &(p->next);
 1038:   }
 1039:   memcpy(code_here, pi->start, pi->length);
 1040:   code_here += pi->length;
 1041:   return old_code_here;
 1042: }
 1043: #endif
 1044: 
 1045: int forget_dyncode(Address code)
 1046: {
 1047: #ifdef NO_DYNAMIC
 1048:   return -1;
 1049: #else
 1050:   struct code_block_list *p, **pp;
 1051: 
 1052:   for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
 1053:     if (code >= p->block && code < p->block+p->size) {
 1054:       next_code_blockp = &(p->next);
 1055:       code_here = start_flush = code;
 1056:       code_area = p->block;
 1057:       last_jump = 0;
 1058:       return -1;
 1059:     }
 1060:   }
 1061:   return -no_dynamic;
 1062: #endif /* !defined(NO_DYNAMIC) */
 1063: }
 1064: 
 1065: long dyncodesize(void)
 1066: {
 1067: #ifndef NO_DYNAMIC
 1068:   struct code_block_list *p;
 1069:   long size=0;
 1070:   for (p=code_block_list; p!=NULL; p=p->next) {
 1071:     if (code_here >= p->block && code_here < p->block+p->size)
 1072:       return size + (code_here - p->block);
 1073:     else
 1074:       size += p->size;
 1075:   }
 1076: #endif /* !defined(NO_DYNAMIC) */
 1077:   return 0;
 1078: }
 1079: 
 1080: Label decompile_code(Label _code)
 1081: {
 1082: #ifdef NO_DYNAMIC
 1083:   return _code;
 1084: #else /* !defined(NO_DYNAMIC) */
 1085:   Cell i;
 1086:   struct code_block_list *p;
 1087:   Address code=_code;
 1088: 
 1089:   /* first, check if we are in code at all */
 1090:   for (p = code_block_list;; p = p->next) {
 1091:     if (p == NULL)
 1092:       return code;
 1093:     if (code >= p->block && code < p->block+p->size)
 1094:       break;
 1095:   }
 1096:   /* reverse order because NOOP might match other prims */
 1097:   for (i=npriminfos-1; i>DOESJUMP; i--) {
 1098:     PrimInfo *pi=decomp_prims[i];
 1099:     if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
 1100:       return vm_prims[super2[super_costs[pi-priminfos].offset]];
 1101:     /* return pi->start;*/
 1102:   }
 1103:   return code;
 1104: #endif /* !defined(NO_DYNAMIC) */
 1105: }
 1106: 
 1107: #ifdef NO_IP
 1108: int nbranchinfos=0;
 1109: 
 1110: struct branchinfo {
 1111:   Label **targetpp; /* **(bi->targetpp) is the target */
 1112:   Cell *addressptr; /* store the target here */
 1113: } branchinfos[100000];
 1114: 
 1115: int ndoesexecinfos=0;
 1116: struct doesexecinfo {
 1117:   int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
 1118:   Label *targetp; /*target for branch (because this is not in threaded code)*/
 1119:   Cell *xt; /* cfa of word whose does-code needs calling */
 1120: } doesexecinfos[10000];
 1121: 
 1122: void set_rel_target(Cell *source, Label target)
 1123: {
 1124:   *source = ((Cell)target)-(((Cell)source)+4);
 1125: }
 1126: 
 1127: void register_branchinfo(Label source, Cell *targetpp)
 1128: {
 1129:   struct branchinfo *bi = &(branchinfos[nbranchinfos]);
 1130:   bi->targetpp = (Label **)targetpp;
 1131:   bi->addressptr = (Cell *)source;
 1132:   nbranchinfos++;
 1133: }
 1134: 
 1135: Address compile_prim1arg(PrimNum p, Cell **argp)
 1136: {
 1137:   Address old_code_here=append_prim(p);
 1138: 
 1139:   assert(vm_prims[p]==priminfos[p].start);
 1140:   *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
 1141:   return old_code_here;
 1142: }
 1143: 
 1144: Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
 1145: {
 1146:   PrimInfo *pi = &priminfos[N_call2];
 1147:   Address old_code_here = append_prim(N_call2);
 1148: 
 1149:   *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
 1150:   register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
 1151:   return old_code_here;
 1152: }
 1153: #endif
 1154: 
 1155: void finish_code(void)
 1156: {
 1157: #ifdef NO_IP
 1158:   Cell i;
 1159: 
 1160:   compile_prim1(NULL);
 1161:   for (i=0; i<ndoesexecinfos; i++) {
 1162:     struct doesexecinfo *dei = &doesexecinfos[i];
 1163:     dei->targetp = (Label *)DOES_CODE1((dei->xt));
 1164:     branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
 1165:   }
 1166:   ndoesexecinfos = 0;
 1167:   for (i=0; i<nbranchinfos; i++) {
 1168:     struct branchinfo *bi=&branchinfos[i];
 1169:     set_rel_target(bi->addressptr, **(bi->targetpp));
 1170:   }
 1171:   nbranchinfos = 0;
 1172: #else
 1173:   compile_prim1(NULL);
 1174: #endif
 1175:   flush_to_here();
 1176: }
 1177: 
 1178: #ifdef NO_IP
 1179: Cell compile_prim_dyn(PrimNum p, Cell *tcp)
 1180:      /* compile prim #p dynamically (mod flags etc.) and return start
 1181: 	address of generated code for putting it into the threaded
 1182: 	code. This function is only called if all the associated
 1183: 	inline arguments of p are already in place (at tcp[1] etc.) */
 1184: {
 1185:   PrimInfo *pi=&priminfos[p];
 1186:   Cell *next_code_target=NULL;
 1187:   Address codeaddr;
 1188:   Address primstart;
 1189:   
 1190:   assert(p<npriminfos);
 1191:   if (p==N_execute || p==N_perform || p==N_lit_perform) {
 1192:     codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
 1193:     primstart = append_prim(p);
 1194:     goto other_prim;
 1195:   } else if (p==N_call) {
 1196:     codeaddr = compile_call2(tcp+1, &next_code_target);
 1197:   } else if (p==N_does_exec) {
 1198:     struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
 1199:     Cell *arg;
 1200:     codeaddr = compile_prim1arg(N_lit,&arg);
 1201:     *arg = (Cell)PFA(tcp[1]);
 1202:     /* we cannot determine the callee now (last_start[1] may be a
 1203:        forward reference), so just register an arbitrary target, and
 1204:        register in dei that we need to fix this before resolving
 1205:        branches */
 1206:     dei->branchinfo = nbranchinfos;
 1207:     dei->xt = (Cell *)(tcp[1]);
 1208:     compile_call2(0, &next_code_target);
 1209:   } else if (!is_relocatable(p)) {
 1210:     Cell *branch_target;
 1211:     codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
 1212:     compile_prim1arg(N_branch,&branch_target);
 1213:     set_rel_target(branch_target,vm_prims[p]);
 1214:   } else {
 1215:     unsigned j;
 1216: 
 1217:     codeaddr = primstart = append_prim(p);
 1218:   other_prim:
 1219:     for (j=0; j<pi->nimmargs; j++) {
 1220:       struct immarg *ia = &(pi->immargs[j]);
 1221:       Cell *argp = tcp + pi->nimmargs - j;
 1222:       Cell argval = *argp; /* !! specific to prims */
 1223:       if (ia->rel) { /* !! assumption: relative refs are branches */
 1224: 	register_branchinfo(primstart + ia->offset, argp);
 1225:       } else /* plain argument */
 1226: 	*(Cell *)(primstart + ia->offset) = argval;
 1227:     }
 1228:   }
 1229:   if (next_code_target!=NULL)
 1230:     *next_code_target = (Cell)code_here;
 1231:   return (Cell)codeaddr;
 1232: }
 1233: #else /* !defined(NO_IP) */
 1234: Cell compile_prim_dyn(PrimNum p, Cell *tcp)
 1235:      /* compile prim #p dynamically (mod flags etc.) and return start
 1236:         address of generated code for putting it into the threaded code */
 1237: {
 1238:   Cell static_prim = (Cell)vm_prims[p];
 1239: #if defined(NO_DYNAMIC)
 1240:   return static_prim;
 1241: #else /* !defined(NO_DYNAMIC) */
 1242:   Address old_code_here;
 1243: 
 1244:   if (no_dynamic)
 1245:     return static_prim;
 1246:   if (p>=npriminfos || !is_relocatable(p)) {
 1247:     append_jump();
 1248:     return static_prim;
 1249:   }
 1250:   old_code_here = append_prim(p);
 1251:   last_jump = p;
 1252:   if (priminfos[p].superend)
 1253:     append_jump();
 1254:   return (Cell)old_code_here;
 1255: #endif  /* !defined(NO_DYNAMIC) */
 1256: }
 1257: #endif /* !defined(NO_IP) */
 1258: 
 1259: #ifndef NO_DYNAMIC
 1260: int cost_codesize(int prim)
 1261: {
 1262:   return priminfos[prim].length;
 1263: }
 1264: #endif
 1265: 
 1266: int cost_ls(int prim)
 1267: {
 1268:   struct cost *c = super_costs+prim;
 1269: 
 1270:   return c->loads + c->stores;
 1271: }
 1272: 
 1273: int cost_lsu(int prim)
 1274: {
 1275:   struct cost *c = super_costs+prim;
 1276: 
 1277:   return c->loads + c->stores + c->updates;
 1278: }
 1279: 
 1280: int cost_nexts(int prim)
 1281: {
 1282:   return 1;
 1283: }
 1284: 
 1285: typedef int Costfunc(int);
 1286: Costfunc *ss_cost =  /* cost function for optimize_bb */
 1287: #ifdef NO_DYNAMIC
 1288: cost_lsu;
 1289: #else
 1290: cost_codesize;
 1291: #endif
 1292: 
 1293: struct {
 1294:   Costfunc *costfunc;
 1295:   char *metricname;
 1296:   long sum;
 1297: } cost_sums[] = {
 1298: #ifndef NO_DYNAMIC
 1299:   { cost_codesize, "codesize", 0 },
 1300: #endif
 1301:   { cost_ls,       "ls",       0 },
 1302:   { cost_lsu,      "lsu",      0 },
 1303:   { cost_nexts,    "nexts",    0 }
 1304: };
 1305: 
 1306: #ifndef NO_DYNAMIC
 1307: void init_ss_cost(void) {
 1308:   if (no_dynamic && ss_cost == cost_codesize) {
 1309:     ss_cost = cost_nexts;
 1310:     cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
 1311:     debugp(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
 1312:   }
 1313: }
 1314: #endif
 1315: 
 1316: #define MAX_BB 128 /* maximum number of instructions in BB */
 1317: #define INF_COST 1000000 /* infinite cost */
 1318: #define CANONICAL_STATE 0
 1319: 
 1320: struct waypoint {
 1321:   int cost;     /* the cost from here to the end */
 1322:   PrimNum inst; /* the inst used from here to the next waypoint */
 1323:   char relocatable; /* the last non-transition was relocatable */
 1324:   char no_transition; /* don't use the next transition (relocatability)
 1325: 		       * or this transition (does not change state) */
 1326: };
 1327: 
 1328: struct tpastate { /* tree parsing automaton (like) state */
 1329:   /* labeling is back-to-front */
 1330:   struct waypoint *inst;  /* in front of instruction */
 1331:   struct waypoint *trans; /* in front of instruction and transition */
 1332: }; 
 1333: 
 1334: struct tpastate *termstate = NULL; /* initialized in loader() */
 1335: 
 1336: void init_waypoints(struct waypoint ws[])
 1337: {
 1338:   int k;
 1339: 
 1340:   for (k=0; k<maxstates; k++)
 1341:     ws[k].cost=INF_COST;
 1342: }
 1343: 
 1344: struct tpastate *empty_tpastate()
 1345: {
 1346:   struct tpastate *s = malloc(sizeof(struct tpastate));
 1347: 
 1348:   s->inst  = malloc(maxstates*sizeof(struct waypoint));
 1349:   init_waypoints(s->inst);
 1350:   s->trans = malloc(maxstates*sizeof(struct waypoint));
 1351:   /* init_waypoints(s->trans);*/
 1352:   return s;
 1353: }
 1354: 
 1355: void transitions(struct tpastate *t)
 1356: {
 1357:   int k;
 1358:   struct super_state *l;
 1359:   
 1360:   for (k=0; k<maxstates; k++) {
 1361:     t->trans[k] = t->inst[k];
 1362:     t->trans[k].no_transition = 1;
 1363:   }
 1364:   for (l = state_transitions; l != NULL; l = l->next) {
 1365:     PrimNum s = l->super;
 1366:     int jcost;
 1367:     struct cost *c=super_costs+s;
 1368:     struct waypoint *wi=&(t->trans[c->state_in]);
 1369:     struct waypoint *wo=&(t->inst[c->state_out]);
 1370:     if (wo->cost == INF_COST)
 1371:       continue;
 1372:     jcost = wo->cost + ss_cost(s);
 1373:     if (jcost <= wi->cost) {
 1374:       wi->cost = jcost;
 1375:       wi->inst = s;
 1376:       wi->relocatable = wo->relocatable;
 1377:       wi->no_transition = 0;
 1378:       /* if (ss_greedy) wi->cost = wo->cost ? */
 1379:     }
 1380:   }
 1381: }
 1382: 
 1383: struct tpastate *make_termstate()
 1384: {
 1385:   struct tpastate *s=empty_tpastate();
 1386: 
 1387:   s->inst[CANONICAL_STATE].cost = 0;
 1388:   transitions(s);
 1389:   return s;
 1390: }
 1391: 
 1392: /* use dynamic programming to find the shortest paths within the basic
 1393:    block origs[0..ninsts-1] and rewrite the instructions pointed to by
 1394:    instps to use it */
 1395: void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
 1396: {
 1397:   int i,j;
 1398:   struct tpastate *ts[ninsts+1];
 1399:   int nextdyn, nextstate, no_transition;
 1400:   
 1401:   ts[ninsts] = termstate;
 1402:   for (i=ninsts-1; i>=0; i--) {
 1403:     ts[i] = empty_tpastate();
 1404:     for (j=1; j<=max_super && i+j<=ninsts; j++) {
 1405:       struct super_state **superp = lookup_super(origs+i, j);
 1406:       if (superp!=NULL) {
 1407: 	struct super_state *supers = *superp;
 1408: 	for (; supers!=NULL; supers = supers->next) {
 1409: 	  PrimNum s = supers->super;
 1410: 	  int jcost;
 1411: 	  struct cost *c=super_costs+s;
 1412: 	  struct waypoint *wi=&(ts[i]->inst[c->state_in]);
 1413: 	  struct waypoint *wo=&(ts[i+j]->trans[c->state_out]);
 1414: 	  int no_transition = wo->no_transition;
 1415: 	  if (!(is_relocatable(s)) && !wo->relocatable) {
 1416: 	    wo=&(ts[i+j]->inst[c->state_out]);
 1417: 	    no_transition=1;
 1418: 	  }
 1419: 	  if (wo->cost == INF_COST) 
 1420: 	    continue;
 1421: 	  jcost = wo->cost + ss_cost(s);
 1422: 	  if (jcost <= wi->cost) {
 1423: 	    wi->cost = jcost;
 1424: 	    wi->inst = s;
 1425: 	    wi->relocatable = is_relocatable(s);
 1426: 	    wi->no_transition = no_transition;
 1427: 	    /* if (ss_greedy) wi->cost = wo->cost ? */
 1428: 	  }
 1429: 	}
 1430:       }
 1431:     }
 1432:     transitions(ts[i]);
 1433:   }
 1434:   /* now rewrite the instructions */
 1435:   nextdyn=0;
 1436:   nextstate=CANONICAL_STATE;
 1437:   no_transition = ((!ts[0]->trans[nextstate].relocatable) 
 1438: 		   ||ts[0]->trans[nextstate].no_transition);
 1439:   for (i=0; i<ninsts; i++) {
 1440:     Cell tc=0, tc2;
 1441:     if (i==nextdyn) {
 1442:       if (!no_transition) {
 1443: 	/* process trans */
 1444: 	PrimNum p = ts[i]->trans[nextstate].inst;
 1445: 	struct cost *c = super_costs+p;
 1446: 	assert(ts[i]->trans[nextstate].cost != INF_COST);
 1447: 	assert(c->state_in==nextstate);
 1448: 	tc = compile_prim_dyn(p,NULL);
 1449: 	nextstate = c->state_out;
 1450:       }
 1451:       {
 1452: 	/* process inst */
 1453: 	PrimNum p = ts[i]->inst[nextstate].inst;
 1454: 	struct cost *c=super_costs+p;
 1455: 	assert(c->state_in==nextstate);
 1456: 	assert(ts[i]->inst[nextstate].cost != INF_COST);
 1457: #if defined(GFORTH_DEBUGGING)
 1458: 	assert(p == origs[i]);
 1459: #endif
 1460: 	tc2 = compile_prim_dyn(p,instps[i]);
 1461: 	if (no_transition || !is_relocatable(p))
 1462: 	  /* !! actually what we care about is if and where
 1463: 	   * compile_prim_dyn() puts NEXTs */
 1464: 	  tc=tc2;
 1465: 	no_transition = ts[i]->inst[nextstate].no_transition;
 1466: 	nextstate = c->state_out;
 1467: 	nextdyn += c->length;
 1468:       }
 1469:     } else {
 1470: #if defined(GFORTH_DEBUGGING)
 1471:       assert(0);
 1472: #endif
 1473:       tc=0;
 1474:       /* tc= (Cell)vm_prims[ts[i]->inst[CANONICAL_STATE].inst]; */
 1475:     }
 1476:     *(instps[i]) = tc;
 1477:   }      
 1478:   if (!no_transition) {
 1479:     PrimNum p = ts[i]->trans[nextstate].inst;
 1480:     struct cost *c = super_costs+p;
 1481:     assert(c->state_in==nextstate);
 1482:     assert(ts[i]->trans[nextstate].cost != INF_COST);
 1483:     assert(i==nextdyn);
 1484:     (void)compile_prim_dyn(p,NULL);
 1485:     nextstate = c->state_out;
 1486:   }
 1487:   assert(nextstate==CANONICAL_STATE);
 1488: }
 1489: 
 1490: /* compile *start, possibly rewriting it into a static and/or dynamic
 1491:    superinstruction */
 1492: void compile_prim1(Cell *start)
 1493: {
 1494: #if defined(DOUBLY_INDIRECT)
 1495:   Label prim;
 1496: 
 1497:   if (start==NULL)
 1498:     return;
 1499:   prim = (Label)*start;
 1500:   if (prim<((Label)(xts+DOESJUMP)) || prim>((Label)(xts+npriminfos))) {
 1501:     fprintf(stderr,"compile_prim encountered xt %p\n", prim);
 1502:     *start=(Cell)prim;
 1503:     return;
 1504:   } else {
 1505:     *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
 1506:     return;
 1507:   }
 1508: #elif defined(INDIRECT_THREADED)
 1509:   return;
 1510: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
 1511:   /* !! does not work, for unknown reasons; but something like this is
 1512:      probably needed to ensure that we don't call compile_prim_dyn
 1513:      before the inline arguments are there */
 1514:   static Cell *instps[MAX_BB];
 1515:   static PrimNum origs[MAX_BB];
 1516:   static int ninsts=0;
 1517:   PrimNum prim_num;
 1518: 
 1519:   if (start==NULL || ninsts >= MAX_BB ||
 1520:       (ninsts>0 && superend[origs[ninsts-1]])) {
 1521:     /* after bb, or at the start of the next bb */
 1522:     optimize_rewrite(instps,origs,ninsts);
 1523:     /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
 1524:     ninsts=0;
 1525:     if (start==NULL)
 1526:       return;
 1527:   }
 1528:   prim_num = ((Xt)*start)-vm_prims;
 1529:   if(prim_num >= npriminfos) {
 1530:     optimize_rewrite(instps,origs,ninsts);
 1531:     /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
 1532:     ninsts=0;
 1533:     return;
 1534:   }    
 1535:   assert(ninsts<MAX_BB);
 1536:   instps[ninsts] = start;
 1537:   origs[ninsts] = prim_num;
 1538:   ninsts++;
 1539: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
 1540: }
 1541: 
 1542: Address loader(FILE *imagefile, char* filename)
 1543: /* returns the address of the image proper (after the preamble) */
 1544: {
 1545:   ImageHeader header;
 1546:   Address image;
 1547:   Address imp; /* image+preamble */
 1548:   Char magic[8];
 1549:   char magic7; /* size byte of magic number */
 1550:   Cell preamblesize=0;
 1551:   Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
 1552:   UCell check_sum;
 1553:   Cell ausize = ((RELINFOBITS ==  8) ? 0 :
 1554: 		 (RELINFOBITS == 16) ? 1 :
 1555: 		 (RELINFOBITS == 32) ? 2 : 3);
 1556:   Cell charsize = ((sizeof(Char) == 1) ? 0 :
 1557: 		   (sizeof(Char) == 2) ? 1 :
 1558: 		   (sizeof(Char) == 4) ? 2 : 3) + ausize;
 1559:   Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
 1560: 		   (sizeof(Cell) == 2) ? 1 :
 1561: 		   (sizeof(Cell) == 4) ? 2 : 3) + ausize;
 1562:   Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
 1563: #ifdef WORDS_BIGENDIAN
 1564:        0
 1565: #else
 1566:        1
 1567: #endif
 1568:     ;
 1569: 
 1570:   vm_prims = engine(0,0,0,0,0);
 1571:   check_prims(vm_prims);
 1572:   prepare_super_table();
 1573: #ifndef DOUBLY_INDIRECT
 1574: #ifdef PRINT_SUPER_LENGTHS
 1575:   print_super_lengths();
 1576: #endif
 1577:   check_sum = checksum(vm_prims);
 1578: #else /* defined(DOUBLY_INDIRECT) */
 1579:   check_sum = (UCell)vm_prims;
 1580: #endif /* defined(DOUBLY_INDIRECT) */
 1581: #if !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED))
 1582:   termstate = make_termstate();
 1583: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
 1584:   
 1585:   do {
 1586:     if(fread(magic,sizeof(Char),8,imagefile) < 8) {
 1587:       fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
 1588: 	      progname, filename);
 1589:       exit(1);
 1590:     }
 1591:     preamblesize+=8;
 1592:   } while(memcmp(magic,"Gforth3",7));
 1593:   magic7 = magic[7];
 1594:   if (debug) {
 1595:     magic[7]='\0';
 1596:     fprintf(stderr,"Magic found: %s ", magic);
 1597:     print_sizes(magic7);
 1598:   }
 1599: 
 1600:   if (magic7 != sizebyte)
 1601:     {
 1602:       fprintf(stderr,"This image is:         ");
 1603:       print_sizes(magic7);
 1604:       fprintf(stderr,"whereas the machine is ");
 1605:       print_sizes(sizebyte);
 1606:       exit(-2);
 1607:     };
 1608: 
 1609:   fread((void *)&header,sizeof(ImageHeader),1,imagefile);
 1610: 
 1611:   set_stack_sizes(&header);
 1612:   
 1613: #if HAVE_GETPAGESIZE
 1614:   pagesize=getpagesize(); /* Linux/GNU libc offers this */
 1615: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
 1616:   pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
 1617: #elif PAGESIZE
 1618:   pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
 1619: #endif
 1620:   debugp(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
 1621: 
 1622:   image = dict_alloc_read(imagefile, preamblesize+header.image_size,
 1623: 			  preamblesize+dictsize, data_offset);
 1624:   imp=image+preamblesize;
 1625:   alloc_stacks((ImageHeader *)imp);
 1626:   if (clear_dictionary)
 1627:     memset(imp+header.image_size, 0, dictsize-header.image_size);
 1628:   if(header.base==0 || header.base  == (Address)0x100) {
 1629:     Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
 1630:     char reloc_bits[reloc_size];
 1631:     fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
 1632:     fread(reloc_bits, 1, reloc_size, imagefile);
 1633:     relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
 1634: #if 0
 1635:     { /* let's see what the relocator did */
 1636:       FILE *snapshot=fopen("snapshot.fi","wb");
 1637:       fwrite(image,1,imagesize,snapshot);
 1638:       fclose(snapshot);
 1639:     }
 1640: #endif
 1641:   }
 1642:   else if(header.base!=imp) {
 1643:     fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
 1644: 	    progname, (unsigned long)header.base, (unsigned long)imp);
 1645:     exit(1);
 1646:   }
 1647:   if (header.checksum==0)
 1648:     ((ImageHeader *)imp)->checksum=check_sum;
 1649:   else if (header.checksum != check_sum) {
 1650:     fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
 1651: 	    progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
 1652:     exit(1);
 1653:   }
 1654: #ifdef DOUBLY_INDIRECT
 1655:   ((ImageHeader *)imp)->xt_base = xts;
 1656: #endif
 1657:   fclose(imagefile);
 1658: 
 1659:   /* unnecessary, except maybe for CODE words */
 1660:   /* FLUSH_ICACHE(imp, header.image_size);*/
 1661: 
 1662:   return imp;
 1663: }
 1664: 
 1665: /* pointer to last '/' or '\' in file, 0 if there is none. */
 1666: char *onlypath(char *filename)
 1667: {
 1668:   return strrchr(filename, DIRSEP);
 1669: }
 1670: 
 1671: FILE *openimage(char *fullfilename)
 1672: {
 1673:   FILE *image_file;
 1674:   char * expfilename = tilde_cstr(fullfilename, strlen(fullfilename), 1);
 1675: 
 1676:   image_file=fopen(expfilename,"rb");
 1677:   if (image_file!=NULL && debug)
 1678:     fprintf(stderr, "Opened image file: %s\n", expfilename);
 1679:   return image_file;
 1680: }
 1681: 
 1682: /* try to open image file concat(path[0:len],imagename) */
 1683: FILE *checkimage(char *path, int len, char *imagename)
 1684: {
 1685:   int dirlen=len;
 1686:   char fullfilename[dirlen+strlen(imagename)+2];
 1687: 
 1688:   memcpy(fullfilename, path, dirlen);
 1689:   if (fullfilename[dirlen-1]!=DIRSEP)
 1690:     fullfilename[dirlen++]=DIRSEP;
 1691:   strcpy(fullfilename+dirlen,imagename);
 1692:   return openimage(fullfilename);
 1693: }
 1694: 
 1695: FILE * open_image_file(char * imagename, char * path)
 1696: {
 1697:   FILE * image_file=NULL;
 1698:   char *origpath=path;
 1699:   
 1700:   if(strchr(imagename, DIRSEP)==NULL) {
 1701:     /* first check the directory where the exe file is in !! 01may97jaw */
 1702:     if (onlypath(progname))
 1703:       image_file=checkimage(progname, onlypath(progname)-progname, imagename);
 1704:     if (!image_file)
 1705:       do {
 1706: 	char *pend=strchr(path, PATHSEP);
 1707: 	if (pend==NULL)
 1708: 	  pend=path+strlen(path);
 1709: 	if (strlen(path)==0) break;
 1710: 	image_file=checkimage(path, pend-path, imagename);
 1711: 	path=pend+(*pend==PATHSEP);
 1712:       } while (image_file==NULL);
 1713:   } else {
 1714:     image_file=openimage(imagename);
 1715:   }
 1716: 
 1717:   if (!image_file) {
 1718:     fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
 1719: 	    progname, imagename, origpath);
 1720:     exit(1);
 1721:   }
 1722: 
 1723:   return image_file;
 1724: }
 1725: #endif
 1726: 
 1727: #ifdef HAS_OS
 1728: UCell convsize(char *s, UCell elemsize)
 1729: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
 1730:    of bytes.  the letter at the end indicates the unit, where e stands
 1731:    for the element size. default is e */
 1732: {
 1733:   char *endp;
 1734:   UCell n,m;
 1735: 
 1736:   m = elemsize;
 1737:   n = strtoul(s,&endp,0);
 1738:   if (endp!=NULL) {
 1739:     if (strcmp(endp,"b")==0)
 1740:       m=1;
 1741:     else if (strcmp(endp,"k")==0)
 1742:       m=1024;
 1743:     else if (strcmp(endp,"M")==0)
 1744:       m=1024*1024;
 1745:     else if (strcmp(endp,"G")==0)
 1746:       m=1024*1024*1024;
 1747:     else if (strcmp(endp,"T")==0) {
 1748: #if (SIZEOF_CHAR_P > 4)
 1749:       m=1024L*1024*1024*1024;
 1750: #else
 1751:       fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
 1752:       exit(1);
 1753: #endif
 1754:     } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
 1755:       fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
 1756:       exit(1);
 1757:     }
 1758:   }
 1759:   return n*m;
 1760: }
 1761: 
 1762: enum {
 1763:   ss_number = 256,
 1764:   ss_states,
 1765:   ss_min_codesize,
 1766:   ss_min_ls,
 1767:   ss_min_lsu,
 1768:   ss_min_nexts,
 1769: };
 1770: 
 1771: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
 1772: {
 1773:   int c;
 1774: 
 1775:   opterr=0;
 1776:   while (1) {
 1777:     int option_index=0;
 1778:     static struct option opts[] = {
 1779:       {"appl-image", required_argument, NULL, 'a'},
 1780:       {"image-file", required_argument, NULL, 'i'},
 1781:       {"dictionary-size", required_argument, NULL, 'm'},
 1782:       {"data-stack-size", required_argument, NULL, 'd'},
 1783:       {"return-stack-size", required_argument, NULL, 'r'},
 1784:       {"fp-stack-size", required_argument, NULL, 'f'},
 1785:       {"locals-stack-size", required_argument, NULL, 'l'},
 1786:       {"path", required_argument, NULL, 'p'},
 1787:       {"version", no_argument, NULL, 'v'},
 1788:       {"help", no_argument, NULL, 'h'},
 1789:       /* put something != 0 into offset_image */
 1790:       {"offset-image", no_argument, &offset_image, 1},
 1791:       {"no-offset-im", no_argument, &offset_image, 0},
 1792:       {"clear-dictionary", no_argument, &clear_dictionary, 1},
 1793:       {"die-on-signal", no_argument, &die_on_signal, 1},
 1794:       {"debug", no_argument, &debug, 1},
 1795:       {"diag", no_argument, &diag, 1},
 1796:       {"no-super", no_argument, &no_super, 1},
 1797:       {"no-dynamic", no_argument, &no_dynamic, 1},
 1798:       {"dynamic", no_argument, &no_dynamic, 0},
 1799:       {"print-metrics", no_argument, &print_metrics, 1},
 1800:       {"ss-number", required_argument, NULL, ss_number},
 1801:       {"ss-states", required_argument, NULL, ss_states},
 1802: #ifndef NO_DYNAMIC
 1803:       {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
 1804: #endif
 1805:       {"ss-min-ls",       no_argument, NULL, ss_min_ls},
 1806:       {"ss-min-lsu",      no_argument, NULL, ss_min_lsu},
 1807:       {"ss-min-nexts",    no_argument, NULL, ss_min_nexts},
 1808:       {"ss-greedy",       no_argument, &ss_greedy, 1},
 1809:       {0,0,0,0}
 1810:       /* no-init-file, no-rc? */
 1811:     };
 1812:     
 1813:     c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
 1814:     
 1815:     switch (c) {
 1816:     case EOF: return;
 1817:     case '?': optind--; return;
 1818:     case 'a': *imagename = optarg; return;
 1819:     case 'i': *imagename = optarg; break;
 1820:     case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
 1821:     case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
 1822:     case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
 1823:     case 'f': fsize = convsize(optarg,sizeof(Float)); break;
 1824:     case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
 1825:     case 'p': *path = optarg; break;
 1826:     case 'o': offset_image = 1; break;
 1827:     case 'n': offset_image = 0; break;
 1828:     case 'c': clear_dictionary = 1; break;
 1829:     case 's': die_on_signal = 1; break;
 1830:     case 'x': debug = 1; break;
 1831:     case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
 1832:     case ss_number: static_super_number = atoi(optarg); break;
 1833:     case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
 1834: #ifndef NO_DYNAMIC
 1835:     case ss_min_codesize: ss_cost = cost_codesize; break;
 1836: #endif
 1837:     case ss_min_ls:       ss_cost = cost_ls;       break;
 1838:     case ss_min_lsu:      ss_cost = cost_lsu;      break;
 1839:     case ss_min_nexts:    ss_cost = cost_nexts;    break;
 1840:     case 'h': 
 1841:       fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
 1842: Engine Options:\n\
 1843:   --appl-image FILE		    equivalent to '--image-file=FILE --'\n\
 1844:   --clear-dictionary		    Initialize the dictionary with 0 bytes\n\
 1845:   -d SIZE, --data-stack-size=SIZE   Specify data stack size\n\
 1846:   --debug			    Print debugging information during startup\n\
 1847:   --diag			    Print diagnostic information during startup\n\
 1848:   --die-on-signal		    exit instead of CATCHing some signals\n\
 1849:   --dynamic			    use dynamic native code\n\
 1850:   -f SIZE, --fp-stack-size=SIZE	    Specify floating point stack size\n\
 1851:   -h, --help			    Print this message and exit\n\
 1852:   -i FILE, --image-file=FILE	    Use image FILE instead of `gforth.fi'\n\
 1853:   -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
 1854:   -m SIZE, --dictionary-size=SIZE   Specify Forth dictionary size\n\
 1855:   --no-dynamic			    Use only statically compiled primitives\n\
 1856:   --no-offset-im		    Load image at normal position\n\
 1857:   --no-super                        No dynamically formed superinstructions\n\
 1858:   --offset-image		    Load image at a different position\n\
 1859:   -p PATH, --path=PATH		    Search path for finding image and sources\n\
 1860:   --print-metrics		    Print some code generation metrics on exit\n\
 1861:   -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
 1862:   --ss-greedy                       greedy, not optimal superinst selection\n\
 1863:   --ss-min-codesize                 select superinsts for smallest native code\n\
 1864:   --ss-min-ls                       minimize loads and stores\n\
 1865:   --ss-min-lsu                      minimize loads, stores, and pointer updates\n\
 1866:   --ss-min-nexts                    minimize the number of static superinsts\n\
 1867:   --ss-number=N                     use N static superinsts (default max)\n\
 1868:   --ss-states=N                     N states for stack caching (default max)\n\
 1869:   -v, --version			    Print engine version and exit\n\
 1870: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
 1871:   `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
 1872: 	      argv[0]);
 1873:       optind--;
 1874:       return;
 1875:     }
 1876:   }
 1877: }
 1878: #endif
 1879: 
 1880: void print_diag()
 1881: {
 1882: 
 1883: #if !defined(HAVE_GETRUSAGE) || (!defined(HAS_FFCALL) && !defined(HAS_LIBFFI))
 1884:   fprintf(stderr, "*** missing functionality ***\n"
 1885: #ifndef HAVE_GETRUSAGE
 1886: 	  "    no getrusage -> CPUTIME broken\n"
 1887: #endif
 1888: #if !defined(HAS_FFCALL) && !defined(HAS_LIBFFI)
 1889: 	  "    no ffcall -> only old-style foreign function calls (no fflib.fs)\n"
 1890: #endif
 1891: 	  );
 1892: #endif
 1893:   if((relocs < nonrelocs) ||
 1894: #if defined(BUGGY_LL_CMP) || defined(BUGGY_LL_MUL) || defined(BUGGY_LL_DIV) || defined(BUGGY_LL_ADD) || defined(BUGGY_LL_SHIFT) || defined(BUGGY_LL_D2F) || defined(BUGGY_LL_F2D)
 1895:      1
 1896: #else
 1897:      0
 1898: #endif
 1899:      )
 1900:     debugp(stderr, "relocs: %d:%d\n", relocs, nonrelocs);
 1901:     fprintf(stderr, "*** performance problems ***\n%s"
 1902: #if defined(BUGGY_LL_CMP) || defined(BUGGY_LL_MUL) || defined(BUGGY_LL_DIV) || defined(BUGGY_LL_ADD) || defined(BUGGY_LL_SHIFT) || defined(BUGGY_LL_D2F) || defined(BUGGY_LL_F2D)
 1903: 	    "    double-cell integer type buggy ->\n        "
 1904: #ifdef BUGGY_LL_CMP
 1905: 	    "CMP, "
 1906: #endif
 1907: #ifdef BUGGY_LL_MUL
 1908: 	    "MUL, "
 1909: #endif
 1910: #ifdef BUGGY_LL_DIV
 1911: 	    "DIV, "
 1912: #endif
 1913: #ifdef BUGGY_LL_ADD
 1914: 	    "ADD, "
 1915: #endif
 1916: #ifdef BUGGY_LL_SHIFT
 1917: 	    "SHIFT, "
 1918: #endif
 1919: #ifdef BUGGY_LL_D2F
 1920: 	    "D2F, "
 1921: #endif
 1922: #ifdef BUGGY_LL_F2D
 1923: 	    "F2D, "
 1924: #endif
 1925: 	    "\b\b slow\n"
 1926: #endif
 1927: #ifndef FORCE_REG
 1928: 	    "    automatic register allocation: performance degradation possible\n"
 1929: #endif
 1930: #if !defined(FORCE_REG) || defined(BUGGY_LONG_LONG)
 1931: 	    "*** Suggested remedy: try ./configure"
 1932: #ifndef FORCE_REG
 1933: 	    " --enable-force-reg"
 1934: #endif
 1935: #ifdef BUGGY_LONG_LONG
 1936: 	    " --enable-force-ll"
 1937: #endif
 1938: 	    "\n"
 1939: #endif
 1940: 	    ,
 1941: 	    (relocs < nonrelocs) ? "    gcc PR 15242 -> no dynamic code generation (use gcc-2.95 instead)\n" : "");
 1942: }
 1943: 
 1944: #ifdef INCLUDE_IMAGE
 1945: extern Cell image[];
 1946: extern const char reloc_bits[];
 1947: #endif
 1948: 
 1949: int main(int argc, char **argv, char **env)
 1950: {
 1951: #ifdef HAS_OS
 1952:   char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
 1953: #else
 1954:   char *path = DEFAULTPATH;
 1955: #endif
 1956: #ifndef INCLUDE_IMAGE
 1957:   char *imagename="gforth.fi";
 1958:   FILE *image_file;
 1959:   Address image;
 1960: #endif
 1961:   int retvalue;
 1962: 	  
 1963: #if defined(i386) && defined(ALIGNMENT_CHECK)
 1964:   /* turn on alignment checks on the 486.
 1965:    * on the 386 this should have no effect. */
 1966:   __asm__("pushfl; popl %eax; orl $0x40000, %eax; pushl %eax; popfl;");
 1967:   /* this is unusable with Linux' libc.4.6.27, because this library is
 1968:      not alignment-clean; we would have to replace some library
 1969:      functions (e.g., memcpy) to make it work. Also GCC doesn't try to keep
 1970:      the stack FP-aligned. */
 1971: #endif
 1972: 
 1973:   /* buffering of the user output device */
 1974: #ifdef _IONBF
 1975:   if (isatty(fileno(stdout))) {
 1976:     fflush(stdout);
 1977:     setvbuf(stdout,NULL,_IONBF,0);
 1978:   }
 1979: #endif
 1980: 
 1981:   progname = argv[0];
 1982: 
 1983: #ifdef HAS_OS
 1984:   gforth_args(argc, argv, &path, &imagename);
 1985: #ifndef NO_DYNAMIC
 1986:   init_ss_cost();
 1987: #endif /* !defined(NO_DYNAMIC) */
 1988: #endif /* defined(HAS_OS) */
 1989: 
 1990: #ifdef INCLUDE_IMAGE
 1991:   set_stack_sizes((ImageHeader *)image);
 1992:   if(((ImageHeader *)image)->base != image)
 1993:     relocate(image, reloc_bits, ((ImageHeader *)image)->image_size,
 1994: 	     (Label*)engine(0, 0, 0, 0, 0));
 1995:   alloc_stacks((ImageHeader *)image);
 1996: #else
 1997:   image_file = open_image_file(imagename, path);
 1998:   image = loader(image_file, imagename);
 1999: #endif
 2000:   gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
 2001: 
 2002:   if (diag)
 2003:     print_diag();
 2004:   {
 2005:     char path2[strlen(path)+1];
 2006:     char *p1, *p2;
 2007:     Cell environ[]= {
 2008:       (Cell)argc-(optind-1),
 2009:       (Cell)(argv+(optind-1)),
 2010:       (Cell)strlen(path),
 2011:       (Cell)path2};
 2012:     argv[optind-1] = progname;
 2013:     /*
 2014:        for (i=0; i<environ[0]; i++)
 2015:        printf("%s\n", ((char **)(environ[1]))[i]);
 2016:        */
 2017:     /* make path OS-independent by replacing path separators with NUL */
 2018:     for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
 2019:       if (*p1==PATHSEP)
 2020: 	*p2 = '\0';
 2021:       else
 2022: 	*p2 = *p1;
 2023:     *p2='\0';
 2024:     retvalue = go_forth(image, 4, environ);
 2025: #ifdef SIGPIPE
 2026:     bsd_signal(SIGPIPE, SIG_IGN);
 2027: #endif
 2028: #ifdef VM_PROFILING
 2029:     vm_print_profile(stderr);
 2030: #endif
 2031:     deprep_terminal();
 2032:   }
 2033:   if (print_metrics) {
 2034:     int i;
 2035:     fprintf(stderr, "code size = %8ld\n", dyncodesize());
 2036:     for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
 2037:       fprintf(stderr, "metric %8s: %8ld\n",
 2038: 	      cost_sums[i].metricname, cost_sums[i].sum);
 2039:   }
 2040:   return retvalue;
 2041: }

FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>