File:  [gforth] / gforth / engine / main.c
Revision 1.122: download - view: text, annotated - select for diffs
Sun Aug 24 09:56:10 2003 UTC (18 years, 3 months ago) by anton
Branches: MAIN
CVS tags: v0-6-2, HEAD
fixed various minor issues:
  workaround for HPUX make and our dependence cycles
  suppress warnings for unused variables in superinstructions
  added some superinstructions
  useful code_size metrics for primitives with funny K and J labels
  default metric for --no-dynamic is now cost_nexts

    1: /* command line interpretation, image loading etc. for Gforth
    2: 
    3: 
    4:   Copyright (C) 1995,1996,1997,1998,2000,2003 Free Software Foundation, Inc.
    5: 
    6:   This file is part of Gforth.
    7: 
    8:   Gforth is free software; you can redistribute it and/or
    9:   modify it under the terms of the GNU General Public License
   10:   as published by the Free Software Foundation; either version 2
   11:   of the License, or (at your option) any later version.
   12: 
   13:   This program is distributed in the hope that it will be useful,
   14:   but WITHOUT ANY WARRANTY; without even the implied warranty of
   15:   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   16:   GNU General Public License for more details.
   17: 
   18:   You should have received a copy of the GNU General Public License
   19:   along with this program; if not, write to the Free Software
   20:   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
   21: */
   22: 
   23: #include "config.h"
   24: #include "forth.h"
   25: #include <errno.h>
   26: #include <ctype.h>
   27: #include <stdio.h>
   28: #include <unistd.h>
   29: #include <string.h>
   30: #include <math.h>
   31: #include <sys/types.h>
   32: #ifndef STANDALONE
   33: #include <sys/stat.h>
   34: #endif
   35: #include <fcntl.h>
   36: #include <assert.h>
   37: #include <stdlib.h>
   38: #include <signal.h>
   39: #ifndef STANDALONE
   40: #if HAVE_SYS_MMAN_H
   41: #include <sys/mman.h>
   42: #endif
   43: #endif
   44: #include "io.h"
   45: #include "getopt.h"
   46: #ifdef STANDALONE
   47: #include <systypes.h>
   48: #endif
   49: 
   50: typedef enum prim_num {
   51: /* definitions of N_execute etc. */
   52: #include "prim_num.i"
   53:   N_START_SUPER
   54: } PrimNum;
   55: 
   56: /* global variables for engine.c 
   57:    We put them here because engine.c is compiled several times in
   58:    different ways for the same engine. */
   59: Cell *SP;
   60: Float *FP;
   61: Address UP=NULL;
   62: 
   63: #ifdef HAS_FFCALL
   64: Cell *RP;
   65: Address LP;
   66: 
   67: #include <callback.h>
   68: 
   69: va_alist clist;
   70: 
   71: void engine_callback(Xt* fcall, void * alist)
   72: {
   73:   clist = (va_alist)alist;
   74:   engine(fcall, SP, RP, FP, LP);
   75: }
   76: #endif
   77: 
   78: #ifdef GFORTH_DEBUGGING
   79: /* define some VM registers as global variables, so they survive exceptions;
   80:    global register variables are not up to the task (according to the 
   81:    GNU C manual) */
   82: Xt *saved_ip;
   83: Cell *rp;
   84: #endif
   85: 
   86: #ifdef NO_IP
   87: Label next_code;
   88: #endif
   89: 
   90: #ifdef HAS_FILE
   91: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
   92: char* pfileattr[6]={"r","r","r+","r+","w","w"};
   93: 
   94: #ifndef O_BINARY
   95: #define O_BINARY 0
   96: #endif
   97: #ifndef O_TEXT
   98: #define O_TEXT 0
   99: #endif
  100: 
  101: int ufileattr[6]= {
  102:   O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
  103:   O_RDWR  |O_BINARY, O_RDWR  |O_BINARY,
  104:   O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
  105: #endif
  106: /* end global vars for engine.c */
  107: 
  108: #define PRIM_VERSION 1
  109: /* increment this whenever the primitives change in an incompatible way */
  110: 
  111: #ifndef DEFAULTPATH
  112: #  define DEFAULTPATH "."
  113: #endif
  114: 
  115: #ifdef MSDOS
  116: jmp_buf throw_jmp_buf;
  117: #endif
  118: 
  119: #if defined(DOUBLY_INDIRECT)
  120: #  define CFA(n)	({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
  121: #else
  122: #  define CFA(n)	((Cell)(symbols+((n)&~0x4000UL)))
  123: #endif
  124: 
  125: #define maxaligned(n)	(typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
  126: 
  127: static UCell dictsize=0;
  128: static UCell dsize=0;
  129: static UCell rsize=0;
  130: static UCell fsize=0;
  131: static UCell lsize=0;
  132: int offset_image=0;
  133: int die_on_signal=0;
  134: #ifndef INCLUDE_IMAGE
  135: static int clear_dictionary=0;
  136: UCell pagesize=1;
  137: char *progname;
  138: #else
  139: char *progname = "gforth";
  140: int optind = 1;
  141: #endif
  142: 
  143: #define CODE_BLOCK_SIZE (256*1024)
  144: Address code_area=0;
  145: Cell code_area_size = CODE_BLOCK_SIZE;
  146: Address code_here=NULL+CODE_BLOCK_SIZE; /* does for code-area what HERE
  147: 					   does for the dictionary */
  148: Address start_flush=NULL; /* start of unflushed code */
  149: Cell last_jump=0; /* if the last prim was compiled without jump, this
  150:                      is it's number, otherwise this contains 0 */
  151: 
  152: static int no_super=0;   /* true if compile_prim should not fuse prims */
  153: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
  154: 					     dynamically */
  155: static int print_metrics=0; /* if true, print metrics on exit */
  156: static int static_super_number = 10000000; /* number of ss used if available */
  157: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
  158: 
  159: #ifdef HAS_DEBUG
  160: int debug=0;
  161: #else
  162: # define perror(x...)
  163: # define fprintf(x...)
  164: #endif
  165: 
  166: ImageHeader *gforth_header;
  167: Label *vm_prims;
  168: #ifdef DOUBLY_INDIRECT
  169: Label *xts; /* same content as vm_prims, but should only be used for xts */
  170: #endif
  171: 
  172: #ifdef MEMCMP_AS_SUBROUTINE
  173: int gforth_memcmp(const char * s1, const char * s2, size_t n)
  174: {
  175:   return memcmp(s1, s2, n);
  176: }
  177: #endif
  178: 
  179: /* image file format:
  180:  *  "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
  181:  *   padding to a multiple of 8
  182:  *   magic: "Gforth3x" means format 0.6,
  183:  *              where x is a byte with
  184:  *              bit 7:   reserved = 0
  185:  *              bit 6:5: address unit size 2^n octets
  186:  *              bit 4:3: character size 2^n octets
  187:  *              bit 2:1: cell size 2^n octets
  188:  *              bit 0:   endian, big=0, little=1.
  189:  *  The magic are always 8 octets, no matter what the native AU/character size is
  190:  *  padding to max alignment (no padding necessary on current machines)
  191:  *  ImageHeader structure (see forth.h)
  192:  *  data (size in ImageHeader.image_size)
  193:  *  tags ((if relocatable, 1 bit/data cell)
  194:  *
  195:  * tag==1 means that the corresponding word is an address;
  196:  * If the word is >=0, the address is within the image;
  197:  * addresses within the image are given relative to the start of the image.
  198:  * If the word =-1 (CF_NIL), the address is NIL,
  199:  * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
  200:  * If the word =CF(DODOES), it's a DOES> CFA
  201:  * If the word =CF(DOESJUMP), it's a DOES JUMP (2 Cells after DOES>,
  202:  *					possibly containing a jump to dodoes)
  203:  * If the word is <CF(DOESJUMP) and bit 14 is set, it's the xt of a primitive
  204:  * If the word is <CF(DOESJUMP) and bit 14 is clear, 
  205:  *                                        it's the threaded code of a primitive
  206:  * bits 13..9 of a primitive token state which group the primitive belongs to,
  207:  * bits 8..0 of a primitive token index into the group
  208:  */
  209: 
  210: Cell groups[32] = {
  211:   0,
  212:   0
  213: #undef GROUP
  214: #undef GROUPADD
  215: #define GROUPADD(n) +n
  216: #define GROUP(x, n) , 0
  217: #include "prim_grp.i"
  218: #undef GROUP
  219: #undef GROUPADD
  220: #define GROUP(x, n)
  221: #define GROUPADD(n)
  222: };
  223: 
  224: void relocate(Cell *image, const unsigned char *bitstring, 
  225:               int size, Cell base, Label symbols[])
  226: {
  227:   int i=0, j, k, steps=(size/sizeof(Cell))/RELINFOBITS;
  228:   Cell token;
  229:   char bits;
  230:   Cell max_symbols;
  231:   /* 
  232:    * A virtual start address that's the real start address minus 
  233:    * the one in the image 
  234:    */
  235:   Cell *start = (Cell * ) (((void *) image) - ((void *) base));
  236: 
  237:   /* group index into table */
  238:   if(groups[31]==0) {
  239:     int groupsum=0;
  240:     for(i=0; i<32; i++) {
  241:       groupsum += groups[i];
  242:       groups[i] = groupsum;
  243:       /* printf("group[%d]=%d\n",i,groupsum); */
  244:     }
  245:     i=0;
  246:   }
  247:   
  248: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
  249:   
  250:   for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
  251:     ;
  252:   max_symbols--;
  253:   size/=sizeof(Cell);
  254: 
  255:   for(k=0; k<=steps; k++) {
  256:     for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
  257:       /*      fprintf(stderr,"relocate: image[%d]\n", i);*/
  258:       if((i < size) && (bits & (1U << (RELINFOBITS-1)))) {
  259: 	/* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
  260:         token=image[i];
  261: 	if(token<0) {
  262: 	  int group = (-token & 0x3E00) >> 9;
  263: 	  if(group == 0) {
  264: 	    switch(token|0x4000) {
  265: 	    case CF_NIL      : image[i]=0; break;
  266: #if !defined(DOUBLY_INDIRECT)
  267: 	    case CF(DOCOL)   :
  268: 	    case CF(DOVAR)   :
  269: 	    case CF(DOCON)   :
  270: 	    case CF(DOUSER)  : 
  271: 	    case CF(DODEFER) : 
  272: 	    case CF(DOFIELD) : MAKE_CF(image+i,symbols[CF(token)]); break;
  273: 	    case CF(DOESJUMP): image[i]=0; break;
  274: #endif /* !defined(DOUBLY_INDIRECT) */
  275: 	    case CF(DODOES)  :
  276: 	      MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
  277: 	      break;
  278: 	    default          : /* backward compatibility */
  279: /*	      printf("Code field generation image[%x]:=CFA(%x)\n",
  280: 		     i, CF(image[i])); */
  281: 	      if (CF((token | 0x4000))<max_symbols) {
  282: 		image[i]=(Cell)CFA(CF(token));
  283: #ifdef DIRECT_THREADED
  284: 		if ((token & 0x4000) == 0) /* threade code, no CFA */
  285: 		  compile_prim1(&image[i]);
  286: #endif
  287: 	      } else
  288: 		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);
  289: 	    }
  290: 	  } else {
  291: 	    int tok = -token & 0x1FF;
  292: 	    if (tok < (groups[group+1]-groups[group])) {
  293: #if defined(DOUBLY_INDIRECT)
  294: 	      image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
  295: #else
  296: 	      image[i]=(Cell)CFA((groups[group]+tok));
  297: #endif
  298: #ifdef DIRECT_THREADED
  299: 	      if ((token & 0x4000) == 0) /* threade code, no CFA */
  300: 		compile_prim1(&image[i]);
  301: #endif
  302: 	    } else
  303: 	      fprintf(stderr,"Primitive %lx, %d of group %d used in this image at $%lx (offset $%x) is not implemented by this\n engine (%s); executing this code will crash.\n", (long)-token, tok, group, (long)&image[i],i,PACKAGE_VERSION);
  304: 	  }
  305: 	} else {
  306:           /* if base is > 0: 0 is a null reference so don't adjust*/
  307:           if (token>=base) {
  308:             image[i]+=(Cell)start;
  309:           }
  310:         }
  311:       }
  312:     }
  313:   }
  314:   finish_code();
  315:   ((ImageHeader*)(image))->base = (Address) image;
  316: }
  317: 
  318: UCell checksum(Label symbols[])
  319: {
  320:   UCell r=PRIM_VERSION;
  321:   Cell i;
  322: 
  323:   for (i=DOCOL; i<=DOESJUMP; i++) {
  324:     r ^= (UCell)(symbols[i]);
  325:     r = (r << 5) | (r >> (8*sizeof(Cell)-5));
  326:   }
  327: #ifdef DIRECT_THREADED
  328:   /* we have to consider all the primitives */
  329:   for (; symbols[i]!=(Label)0; i++) {
  330:     r ^= (UCell)(symbols[i]);
  331:     r = (r << 5) | (r >> (8*sizeof(Cell)-5));
  332:   }
  333: #else
  334:   /* in indirect threaded code all primitives are accessed through the
  335:      symbols table, so we just have to put the base address of symbols
  336:      in the checksum */
  337:   r ^= (UCell)symbols;
  338: #endif
  339:   return r;
  340: }
  341: 
  342: Address verbose_malloc(Cell size)
  343: {
  344:   Address r;
  345:   /* leave a little room (64B) for stack underflows */
  346:   if ((r = malloc(size+64))==NULL) {
  347:     perror(progname);
  348:     exit(1);
  349:   }
  350:   r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
  351:   if (debug)
  352:     fprintf(stderr, "malloc succeeds, address=$%lx\n", (long)r);
  353:   return r;
  354: }
  355: 
  356: static Address next_address=0;
  357: void after_alloc(Address r, Cell size)
  358: {
  359:   if (r != (Address)-1) {
  360:     if (debug)
  361:       fprintf(stderr, "success, address=$%lx\n", (long) r);
  362:     if (pagesize != 1)
  363:       next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
  364:   } else {
  365:     if (debug)
  366:       fprintf(stderr, "failed: %s\n", strerror(errno));
  367:   }
  368: }
  369: 
  370: #ifndef MAP_FAILED
  371: #define MAP_FAILED ((Address) -1)
  372: #endif
  373: #ifndef MAP_FILE
  374: # define MAP_FILE 0
  375: #endif
  376: #ifndef MAP_PRIVATE
  377: # define MAP_PRIVATE 0
  378: #endif
  379: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
  380: # define MAP_ANON MAP_ANONYMOUS
  381: #endif
  382: 
  383: #if defined(HAVE_MMAP)
  384: static Address alloc_mmap(Cell size)
  385: {
  386:   Address r;
  387: 
  388: #if defined(MAP_ANON)
  389:   if (debug)
  390:     fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
  391:   r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE, -1, 0);
  392: #else /* !defined(MAP_ANON) */
  393:   /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
  394:      apparently defaults) */
  395:   static int dev_zero=-1;
  396: 
  397:   if (dev_zero == -1)
  398:     dev_zero = open("/dev/zero", O_RDONLY);
  399:   if (dev_zero == -1) {
  400:     r = MAP_FAILED;
  401:     if (debug)
  402:       fprintf(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ", 
  403: 	      strerror(errno));
  404:   } else {
  405:     if (debug)
  406:       fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
  407:     r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE, dev_zero, 0);
  408:   }
  409: #endif /* !defined(MAP_ANON) */
  410:   after_alloc(r, size);
  411:   return r;  
  412: }
  413: #endif
  414: 
  415: Address my_alloc(Cell size)
  416: {
  417: #if HAVE_MMAP
  418:   Address r;
  419: 
  420:   r=alloc_mmap(size);
  421:   if (r!=(Address)MAP_FAILED)
  422:     return r;
  423: #endif /* HAVE_MMAP */
  424:   /* use malloc as fallback */
  425:   return verbose_malloc(size);
  426: }
  427: 
  428: Address dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
  429: {
  430:   Address image = MAP_FAILED;
  431: 
  432: #if defined(HAVE_MMAP)
  433:   if (offset==0) {
  434:     image=alloc_mmap(dictsize);
  435:     if (debug)
  436:       fprintf(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
  437:     image = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE, fileno(file), 0);
  438:     after_alloc(image,dictsize);
  439:   }
  440: #endif /* defined(HAVE_MMAP) */
  441:   if (image == (Address)MAP_FAILED) {
  442:     image = my_alloc(dictsize+offset)+offset;
  443:     rewind(file);  /* fseek(imagefile,0L,SEEK_SET); */
  444:     fread(image, 1, imagesize, file);
  445:   }
  446:   return image;
  447: }
  448: 
  449: void set_stack_sizes(ImageHeader * header)
  450: {
  451:   if (dictsize==0)
  452:     dictsize = header->dict_size;
  453:   if (dsize==0)
  454:     dsize = header->data_stack_size;
  455:   if (rsize==0)
  456:     rsize = header->return_stack_size;
  457:   if (fsize==0)
  458:     fsize = header->fp_stack_size;
  459:   if (lsize==0)
  460:     lsize = header->locals_stack_size;
  461:   dictsize=maxaligned(dictsize);
  462:   dsize=maxaligned(dsize);
  463:   rsize=maxaligned(rsize);
  464:   lsize=maxaligned(lsize);
  465:   fsize=maxaligned(fsize);
  466: }
  467: 
  468: void alloc_stacks(ImageHeader * header)
  469: {
  470:   header->dict_size=dictsize;
  471:   header->data_stack_size=dsize;
  472:   header->fp_stack_size=fsize;
  473:   header->return_stack_size=rsize;
  474:   header->locals_stack_size=lsize;
  475: 
  476:   header->data_stack_base=my_alloc(dsize);
  477:   header->fp_stack_base=my_alloc(fsize);
  478:   header->return_stack_base=my_alloc(rsize);
  479:   header->locals_stack_base=my_alloc(lsize);
  480: }
  481: 
  482: #warning You can ignore the warnings about clobbered variables in go_forth
  483: int go_forth(Address image, int stack, Cell *entries)
  484: {
  485:   volatile ImageHeader *image_header = (ImageHeader *)image;
  486:   Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
  487:   Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
  488:   Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
  489: #ifdef GFORTH_DEBUGGING
  490:   volatile Cell *orig_rp0=rp0;
  491: #endif
  492:   Address lp0=image_header->locals_stack_base + lsize;
  493:   Xt *ip0=(Xt *)(image_header->boot_entry);
  494: #ifdef SYSSIGNALS
  495:   int throw_code;
  496: #endif
  497: 
  498:   /* ensure that the cached elements (if any) are accessible */
  499:   IF_spTOS(sp0--);
  500:   IF_fpTOS(fp0--);
  501:   
  502:   for(;stack>0;stack--)
  503:     *--sp0=entries[stack-1];
  504: 
  505: #ifdef SYSSIGNALS
  506:   get_winsize();
  507:    
  508:   install_signal_handlers(); /* right place? */
  509:   
  510:   if ((throw_code=setjmp(throw_jmp_buf))) {
  511:     static Cell signal_data_stack[8];
  512:     static Cell signal_return_stack[8];
  513:     static Float signal_fp_stack[1];
  514: 
  515:     signal_data_stack[7]=throw_code;
  516: 
  517: #ifdef GFORTH_DEBUGGING
  518:     if (debug)
  519:       fprintf(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
  520: 	      throw_code, saved_ip, rp);
  521:     if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
  522:       /* no rstack overflow or underflow */
  523:       rp0 = rp;
  524:       *--rp0 = (Cell)saved_ip;
  525:     }
  526:     else /* I love non-syntactic ifdefs :-) */
  527:       rp0 = signal_return_stack+8;
  528: #else  /* !defined(GFORTH_DEBUGGING) */
  529:     if (debug)
  530:       fprintf(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
  531:       rp0 = signal_return_stack+8;
  532: #endif /* !defined(GFORTH_DEBUGGING) */
  533:     /* fprintf(stderr, "rp=$%x\n",rp0);*/
  534:     
  535:     return((int)(Cell)engine(image_header->throw_entry, signal_data_stack+7,
  536: 		       rp0, signal_fp_stack, 0));
  537:   }
  538: #endif
  539: 
  540:   return((int)(Cell)engine(ip0,sp0,rp0,fp0,lp0));
  541: }
  542: 
  543: #ifndef INCLUDE_IMAGE
  544: void print_sizes(Cell sizebyte)
  545:      /* print size information */
  546: {
  547:   static char* endianstring[]= { "   big","little" };
  548:   
  549:   fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
  550: 	  endianstring[sizebyte & 1],
  551: 	  1 << ((sizebyte >> 1) & 3),
  552: 	  1 << ((sizebyte >> 3) & 3),
  553: 	  1 << ((sizebyte >> 5) & 3));
  554: }
  555: 
  556: /* static superinstruction stuff */
  557: 
  558: struct cost {
  559:   char loads;       /* number of stack loads */
  560:   char stores;      /* number of stack stores */
  561:   char updates;     /* number of stack pointer updates */
  562:   short offset;      /* offset into super2 table */
  563:   char length;      /* number of components */
  564: };
  565: 
  566: PrimNum super2[] = {
  567: #include "super2.i"
  568: };
  569: 
  570: struct cost super_costs[] = {
  571: #include "costs.i"
  572: };
  573: 
  574: #define HASH_SIZE 256
  575: 
  576: struct super_table_entry {
  577:   struct super_table_entry *next;
  578:   PrimNum *start;
  579:   short length;
  580:   PrimNum super;
  581: } *super_table[HASH_SIZE];
  582: int max_super=2;
  583: 
  584: int hash_super(PrimNum *start, int length)
  585: {
  586:   int i, r;
  587:   
  588:   for (i=0, r=0; i<length; i++) {
  589:     r <<= 1;
  590:     r += start[i];
  591:   }
  592:   return r & (HASH_SIZE-1);
  593: }
  594: 
  595: int lookup_super(PrimNum *start, int length)
  596: {
  597:   int hash=hash_super(start,length);
  598:   struct super_table_entry *p = super_table[hash];
  599: 
  600:   assert(length >= 2);
  601:   for (; p!=NULL; p = p->next) {
  602:     if (length == p->length &&
  603: 	memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
  604:       return p->super;
  605:   }
  606:   return -1;
  607: }
  608: 
  609: void prepare_super_table()
  610: {
  611:   int i;
  612:   int nsupers = 0;
  613: 
  614:   for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
  615:     struct cost *c = &super_costs[i];
  616:     if (c->length > 1 && nsupers < static_super_number) {
  617:       int hash = hash_super(super2+c->offset, c->length);
  618:       struct super_table_entry **p = &super_table[hash];
  619:       struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
  620:       e->next = *p;
  621:       e->start = super2 + c->offset;
  622:       e->length = c->length;
  623:       e->super = i;
  624:       *p = e;
  625:       if (c->length > max_super)
  626: 	max_super = c->length;
  627:       nsupers++;
  628:     }
  629:   }
  630:   if (debug)
  631:     fprintf(stderr, "Using %d static superinsts\n", nsupers);
  632: }
  633: 
  634: /* dynamic replication/superinstruction stuff */
  635: 
  636: #define MAX_IMMARGS 2
  637: 
  638: #ifndef NO_DYNAMIC
  639: typedef struct {
  640:   Label start;
  641:   Cell length; /* only includes the jump iff superend is true*/
  642:   Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
  643:   char superend; /* true if primitive ends superinstruction, i.e.,
  644:                      unconditional branch, execute, etc. */
  645:   Cell nimmargs;
  646:   struct immarg {
  647:     Cell offset; /* offset of immarg within prim */
  648:     char rel;    /* true if immarg is relative */
  649:   } immargs[MAX_IMMARGS];
  650: } PrimInfo;
  651: 
  652: PrimInfo *priminfos;
  653: PrimInfo **decomp_prims;
  654: 
  655: int compare_priminfo_length(const void *_a, const void *_b)
  656: {
  657:   PrimInfo **a = (PrimInfo **)_a;
  658:   PrimInfo **b = (PrimInfo **)_b;
  659:   Cell diff = (*a)->length - (*b)->length;
  660:   if (diff)
  661:     return diff;
  662:   else /* break ties by start address; thus the decompiler produces
  663:           the earliest primitive with the same code (e.g. noop instead
  664:           of (char) and @ instead of >code-address */
  665:     return (*b)->start - (*a)->start;
  666: }
  667: #endif /* !defined(NO_DYNAMIC) */
  668: 
  669: static char superend[]={
  670: #include "prim_superend.i"
  671: };
  672: 
  673: Cell npriminfos=0;
  674: 
  675: int compare_labels(const void *pa, const void *pb)
  676: {
  677:   Label a = *(Label *)pa;
  678:   Label b = *(Label *)pb;
  679:   return a-b;
  680: }
  681: 
  682: Label bsearch_next(Label key, Label *a, UCell n)
  683:      /* a is sorted; return the label >=key that is the closest in a;
  684:         return NULL if there is no label in a >=key */
  685: {
  686:   int mid = (n-1)/2;
  687:   if (n<1)
  688:     return NULL;
  689:   if (n == 1) {
  690:     if (a[0] < key)
  691:       return NULL;
  692:     else
  693:       return a[0];
  694:   }
  695:   if (a[mid] < key)
  696:     return bsearch_next(key, a+mid+1, n-mid-1);
  697:   else
  698:     return bsearch_next(key, a, mid+1);
  699: }
  700: 
  701: void check_prims(Label symbols1[])
  702: {
  703:   int i;
  704: #ifndef NO_DYNAMIC
  705:   Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted;
  706:   int nends1j;
  707: #endif
  708: 
  709:   if (debug)
  710: #ifdef __VERSION__
  711:     fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
  712: #else
  713: #define xstr(s) str(s)
  714: #define str(s) #s
  715:   fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n"); 
  716: #endif
  717:   for (i=0; symbols1[i]!=0; i++)
  718:     ;
  719:   npriminfos = i;
  720:   
  721: #ifndef NO_DYNAMIC
  722:   if (no_dynamic)
  723:     return;
  724:   symbols2=engine2(0,0,0,0,0);
  725: #if NO_IP
  726:   symbols3=engine3(0,0,0,0,0);
  727: #else
  728:   symbols3=symbols1;
  729: #endif
  730:   ends1 = symbols1+i+1;
  731:   ends1j =   ends1+i;
  732:   nends1j = i+1;
  733:   ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
  734:   memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
  735:   qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
  736:   
  737:   priminfos = calloc(i,sizeof(PrimInfo));
  738:   for (i=0; symbols1[i]!=0; i++) {
  739:     int prim_len = ends1[i]-symbols1[i];
  740:     PrimInfo *pi=&priminfos[i];
  741:     int j=0;
  742:     char *s1 = (char *)symbols1[i];
  743:     char *s2 = (char *)symbols2[i];
  744:     char *s3 = (char *)symbols3[i];
  745:     Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
  746: 
  747:     pi->start = s1;
  748:     pi->superend = superend[i]|no_super;
  749:     if (pi->superend)
  750:       pi->length = endlabel-symbols1[i];
  751:     else
  752:       pi->length = prim_len;
  753:     pi->restlength = endlabel - symbols1[i] - pi->length;
  754:     pi->nimmargs = 0;
  755:     if (debug)
  756:       fprintf(stderr, "Prim %3d @ %p %p %p, length=%3ld restlength=%2ld superend=%1d",
  757: 	      i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength), pi->superend);
  758:     if (endlabel == NULL) {
  759:       pi->start = NULL; /* not relocatable */
  760:       if (pi->length<0) pi->length=100;
  761:       if (debug)
  762: 	fprintf(stderr,"\n   non_reloc: no J label > start found\n");
  763:       continue;
  764:     }
  765:     if (ends1[i] > endlabel && !pi->superend) {
  766:       pi->start = NULL; /* not relocatable */
  767:       pi->length = endlabel-symbols1[i];
  768:       if (debug)
  769: 	fprintf(stderr,"\n   non_reloc: there is a J label before the K label (restlength<0)\n");
  770:       continue;
  771:     }
  772:     if (ends1[i] < pi->start && !pi->superend) {
  773:       pi->start = NULL; /* not relocatable */
  774:       pi->length = endlabel-symbols1[i];
  775:       if (debug)
  776: 	fprintf(stderr,"\n   non_reloc: K label before I label (length<0)\n");
  777:       continue;
  778:     }
  779:     assert(prim_len>=0);
  780:     assert(pi->restlength >=0);
  781:     while (j<(pi->length+pi->restlength)) {
  782:       if (s1[j]==s3[j]) {
  783: 	if (s1[j] != s2[j]) {
  784: 	  pi->start = NULL; /* not relocatable */
  785: 	  if (debug)
  786: 	    fprintf(stderr,"\n   non_reloc: engine1!=engine2 offset %3d",j);
  787: 	  /* assert(j<prim_len); */
  788: 	  break;
  789: 	}
  790: 	j++;
  791:       } else {
  792: 	struct immarg *ia=&pi->immargs[pi->nimmargs];
  793: 
  794: 	pi->nimmargs++;
  795: 	ia->offset=j;
  796: 	if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
  797: 	  ia->rel=0;
  798: 	  if (debug)
  799: 	    fprintf(stderr,"\n   absolute immarg: offset %3d",j);
  800: 	} else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
  801: 		   symbols1[DOESJUMP+1]) {
  802: 	  ia->rel=1;
  803: 	  if (debug)
  804: 	    fprintf(stderr,"\n   relative immarg: offset %3d",j);
  805: 	} else {
  806: 	  pi->start = NULL; /* not relocatable */
  807: 	  if (debug)
  808: 	    fprintf(stderr,"\n   non_reloc: engine1!=engine3 offset %3d",j);
  809: 	  /* assert(j<prim_len);*/
  810: 	  break;
  811: 	}
  812: 	j+=4;
  813:       }
  814:     }
  815:     if (debug)
  816:       fprintf(stderr,"\n");
  817:   }
  818:   decomp_prims = calloc(i,sizeof(PrimInfo *));
  819:   for (i=DOESJUMP+1; i<npriminfos; i++)
  820:     decomp_prims[i] = &(priminfos[i]);
  821:   qsort(decomp_prims+DOESJUMP+1, npriminfos-DOESJUMP-1, sizeof(PrimInfo *),
  822: 	compare_priminfo_length);
  823: #endif
  824: }
  825: 
  826: void flush_to_here(void)
  827: {
  828: #ifndef NO_DYNAMIC
  829:   if (start_flush)
  830:     FLUSH_ICACHE(start_flush, code_here-start_flush);
  831:   start_flush=code_here;
  832: #endif
  833: }
  834: 
  835: #ifndef NO_DYNAMIC
  836: void append_jump(void)
  837: {
  838:   if (last_jump) {
  839:     PrimInfo *pi = &priminfos[last_jump];
  840:     
  841:     memcpy(code_here, pi->start+pi->length, pi->restlength);
  842:     code_here += pi->restlength;
  843:     last_jump=0;
  844:   }
  845: }
  846: 
  847: /* Gforth remembers all code blocks in this list.  On forgetting (by
  848: executing a marker) the code blocks are not freed (because Gforth does
  849: not remember how they were allocated; hmm, remembering that might be
  850: easier and cleaner).  Instead, code_here etc. are reset to the old
  851: value, and the "forgotten" code blocks are reused when they are
  852: needed. */
  853: 
  854: struct code_block_list {
  855:   struct code_block_list *next;
  856:   Address block;
  857:   Cell size;
  858: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
  859: 
  860: Address append_prim(Cell p)
  861: {
  862:   PrimInfo *pi = &priminfos[p];
  863:   Address old_code_here = code_here;
  864: 
  865:   if (code_area+code_area_size < code_here+pi->length+pi->restlength) {
  866:     struct code_block_list *p;
  867:     append_jump();
  868:     flush_to_here();
  869:     if (*next_code_blockp == NULL) {
  870:       code_here = start_flush = code_area = my_alloc(code_area_size);
  871:       p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
  872:       *next_code_blockp = p;
  873:       p->next = NULL;
  874:       p->block = code_here;
  875:       p->size = code_area_size;
  876:     } else {
  877:       p = *next_code_blockp;
  878:       code_here = start_flush = code_area = p->block;
  879:     }
  880:     old_code_here = code_here;
  881:     next_code_blockp = &(p->next);
  882:   }
  883:   memcpy(code_here, pi->start, pi->length);
  884:   code_here += pi->length;
  885:   return old_code_here;
  886: }
  887: #endif
  888: 
  889: int forget_dyncode(Address code)
  890: {
  891: #ifdef NO_DYNAMIC
  892:   return -1;
  893: #else
  894:   struct code_block_list *p, **pp;
  895: 
  896:   for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
  897:     if (code >= p->block && code < p->block+p->size) {
  898:       next_code_blockp = &(p->next);
  899:       code_here = start_flush = code;
  900:       code_area = p->block;
  901:       last_jump = 0;
  902:       return -1;
  903:     }
  904:   }
  905:   return -no_dynamic;
  906: #endif /* !defined(NO_DYNAMIC) */
  907: }
  908: 
  909: long dyncodesize(void)
  910: {
  911: #ifndef NO_DYNAMIC
  912:   struct code_block_list *p;
  913:   long size=0;
  914:   for (p=code_block_list; p!=NULL; p=p->next) {
  915:     if (code_here >= p->block && code_here < p->block+p->size)
  916:       return size + (code_here - p->block);
  917:     else
  918:       size += p->size;
  919:   }
  920: #endif /* !defined(NO_DYNAMIC) */
  921:   return 0;
  922: }
  923: 
  924: Label decompile_code(Label _code)
  925: {
  926: #ifdef NO_DYNAMIC
  927:   return _code;
  928: #else /* !defined(NO_DYNAMIC) */
  929:   Cell i;
  930:   struct code_block_list *p;
  931:   Address code=_code;
  932: 
  933:   /* first, check if we are in code at all */
  934:   for (p = code_block_list;; p = p->next) {
  935:     if (p == NULL)
  936:       return code;
  937:     if (code >= p->block && code < p->block+p->size)
  938:       break;
  939:   }
  940:   /* reverse order because NOOP might match other prims */
  941:   for (i=npriminfos-1; i>DOESJUMP; i--) {
  942:     PrimInfo *pi=decomp_prims[i];
  943:     if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
  944:       return vm_prims[super2[super_costs[pi-priminfos].offset]];
  945:     /* return pi->start;*/
  946:   }
  947:   return code;
  948: #endif /* !defined(NO_DYNAMIC) */
  949: }
  950: 
  951: #ifdef NO_IP
  952: int nbranchinfos=0;
  953: 
  954: struct branchinfo {
  955:   Label *targetptr; /* *(bi->targetptr) is the target */
  956:   Cell *addressptr; /* store the target here */
  957: } branchinfos[100000];
  958: 
  959: int ndoesexecinfos=0;
  960: struct doesexecinfo {
  961:   int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
  962:   Cell *xt; /* cfa of word whose does-code needs calling */
  963: } doesexecinfos[10000];
  964: 
  965: void set_rel_target(Cell *source, Label target)
  966: {
  967:   *source = ((Cell)target)-(((Cell)source)+4);
  968: }
  969: 
  970: void register_branchinfo(Label source, Cell targetptr)
  971: {
  972:   struct branchinfo *bi = &(branchinfos[nbranchinfos]);
  973:   bi->targetptr = (Label *)targetptr;
  974:   bi->addressptr = (Cell *)source;
  975:   nbranchinfos++;
  976: }
  977: 
  978: Cell *compile_prim1arg(Cell p)
  979: {
  980:   int l = priminfos[p].length;
  981:   Address old_code_here=code_here;
  982: 
  983:   assert(vm_prims[p]==priminfos[p].start);
  984:   append_prim(p);
  985:   return (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
  986: }
  987: 
  988: Cell *compile_call2(Cell targetptr)
  989: {
  990:   Cell *next_code_target;
  991:   PrimInfo *pi = &priminfos[N_call2];
  992:   Address old_code_here = append_prim(N_call2);
  993: 
  994:   next_code_target = (Cell *)(old_code_here + pi->immargs[0].offset);
  995:   register_branchinfo(old_code_here + pi->immargs[1].offset, targetptr);
  996:   return next_code_target;
  997: }
  998: #endif
  999: 
 1000: void finish_code(void)
 1001: {
 1002: #ifdef NO_IP
 1003:   Cell i;
 1004: 
 1005:   compile_prim1(NULL);
 1006:   for (i=0; i<ndoesexecinfos; i++) {
 1007:     struct doesexecinfo *dei = &doesexecinfos[i];
 1008:     branchinfos[dei->branchinfo].targetptr = DOES_CODE1((dei->xt));
 1009:   }
 1010:   ndoesexecinfos = 0;
 1011:   for (i=0; i<nbranchinfos; i++) {
 1012:     struct branchinfo *bi=&branchinfos[i];
 1013:     set_rel_target(bi->addressptr, *(bi->targetptr));
 1014:   }
 1015:   nbranchinfos = 0;
 1016: #endif
 1017:   flush_to_here();
 1018: }
 1019: 
 1020: #if 0
 1021: /* compile *start into a dynamic superinstruction, updating *start */
 1022: void compile_prim_dyn(Cell *start)
 1023: {
 1024: #if defined(NO_IP)
 1025:   static Cell *last_start=NULL;
 1026:   static Xt last_prim=NULL;
 1027:   /* delay work by one call in order to get relocated immargs */
 1028: 
 1029:   if (last_start) {
 1030:     unsigned i = last_prim-vm_prims;
 1031:     PrimInfo *pi=&priminfos[i];
 1032:     Cell *next_code_target=NULL;
 1033: 
 1034:     assert(i<npriminfos);
 1035:     if (i==N_execute||i==N_perform||i==N_lit_perform) {
 1036:       next_code_target = compile_prim1arg(N_set_next_code);
 1037:     }
 1038:     if (i==N_call) {
 1039:       next_code_target = compile_call2(last_start[1]);
 1040:     } else if (i==N_does_exec) {
 1041:       struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
 1042:       *compile_prim1arg(N_lit) = (Cell)PFA(last_start[1]);
 1043:       /* we cannot determine the callee now (last_start[1] may be a
 1044:          forward reference), so just register an arbitrary target, and
 1045:          register in dei that we need to fix this before resolving
 1046:          branches */
 1047:       dei->branchinfo = nbranchinfos;
 1048:       dei->xt = (Cell *)(last_start[1]);
 1049:       next_code_target = compile_call2(NULL);
 1050:     } else if (pi->start == NULL) { /* non-reloc */
 1051:       next_code_target = compile_prim1arg(N_set_next_code);
 1052:       set_rel_target(compile_prim1arg(N_abranch),*(Xt)last_prim);
 1053:     } else {
 1054:       unsigned j;
 1055:       Address old_code_here = append_prim(i);
 1056: 
 1057:       for (j=0; j<pi->nimmargs; j++) {
 1058: 	struct immarg *ia = &(pi->immargs[j]);
 1059: 	Cell argval = last_start[pi->nimmargs - j]; /* !! specific to prims */
 1060: 	if (ia->rel) { /* !! assumption: relative refs are branches */
 1061: 	  register_branchinfo(old_code_here + ia->offset, argval);
 1062: 	} else /* plain argument */
 1063: 	  *(Cell *)(old_code_here + ia->offset) = argval;
 1064:       }
 1065:     }
 1066:     if (next_code_target!=NULL)
 1067:       *next_code_target = (Cell)code_here;
 1068:   }
 1069:   if (start) {
 1070:     last_prim = (Xt)*start;
 1071:     *start = (Cell)code_here;
 1072:   }
 1073:   last_start = start;
 1074:   return;
 1075: #elif !defined(NO_DYNAMIC)
 1076:   Label prim=(Label)*start;
 1077:   unsigned i;
 1078:   Address old_code_here;
 1079: 
 1080:   i = ((Xt)prim)-vm_prims;
 1081:   prim = *(Xt)prim;
 1082:   if (no_dynamic) {
 1083:     *start = (Cell)prim;
 1084:     return;
 1085:   }
 1086:   if (i>=npriminfos || priminfos[i].start == 0) { /* not a relocatable prim */
 1087:     append_jump();
 1088:     *start = (Cell)prim;
 1089:     return;
 1090:   }
 1091:   assert(priminfos[i].start = prim); 
 1092: #ifdef ALIGN_CODE
 1093:   /*  ALIGN_CODE;*/
 1094: #endif
 1095:   assert(prim==priminfos[i].start);
 1096:   old_code_here = append_prim(i);
 1097:   last_jump = (priminfos[i].superend) ? 0 : i;
 1098:   *start = (Cell)old_code_here;
 1099:   return;
 1100: #else /* !defined(DOUBLY_INDIRECT), no code replication */
 1101:   Label prim=(Label)*start;
 1102: #if !defined(INDIRECT_THREADED)
 1103:   prim = *(Xt)prim;
 1104: #endif
 1105:   *start = (Cell)prim;
 1106:   return;
 1107: #endif /* !defined(DOUBLY_INDIRECT) */
 1108: }
 1109: #endif /* 0 */
 1110: 
 1111: Cell compile_prim_dyn(unsigned p)
 1112: {
 1113:   Cell static_prim = (Cell)vm_prims[p];
 1114: #if defined(NO_DYNAMIC)
 1115:   return static_prim;
 1116: #else /* !defined(NO_DYNAMIC) */
 1117:   Address old_code_here;
 1118: 
 1119:   if (no_dynamic)
 1120:     return static_prim;
 1121:   if (p>=npriminfos || priminfos[p].start == 0) { /* not a relocatable prim */
 1122:     append_jump();
 1123:     return static_prim;
 1124:   }
 1125:   old_code_here = append_prim(p);
 1126:   last_jump = (priminfos[p].superend) ? 0 : p;
 1127:   return (Cell)old_code_here;
 1128: #endif  /* !defined(NO_DYNAMIC) */
 1129: }
 1130: 
 1131: #ifndef NO_DYNAMIC
 1132: int cost_codesize(int prim)
 1133: {
 1134:   return priminfos[prim].length;
 1135: }
 1136: #endif
 1137: 
 1138: int cost_ls(int prim)
 1139: {
 1140:   struct cost *c = super_costs+prim;
 1141: 
 1142:   return c->loads + c->stores;
 1143: }
 1144: 
 1145: int cost_lsu(int prim)
 1146: {
 1147:   struct cost *c = super_costs+prim;
 1148: 
 1149:   return c->loads + c->stores + c->updates;
 1150: }
 1151: 
 1152: int cost_nexts(int prim)
 1153: {
 1154:   return 1;
 1155: }
 1156: 
 1157: typedef int Costfunc(int);
 1158: Costfunc *ss_cost =  /* cost function for optimize_bb */
 1159: #ifdef NO_DYNAMIC
 1160: cost_lsu;
 1161: #else
 1162: cost_codesize;
 1163: #endif
 1164: 
 1165: struct {
 1166:   Costfunc *costfunc;
 1167:   char *metricname;
 1168:   long sum;
 1169: } cost_sums[] = {
 1170: #ifndef NO_DYNAMIC
 1171:   { cost_codesize, "codesize", 0 },
 1172: #endif
 1173:   { cost_ls,       "ls",       0 },
 1174:   { cost_lsu,      "lsu",      0 },
 1175:   { cost_nexts,    "nexts",    0 }
 1176: };
 1177: 
 1178: #define MAX_BB 128 /* maximum number of instructions in BB */
 1179: 
 1180: /* use dynamic programming to find the shortest paths within the basic
 1181:    block origs[0..ninsts-1]; optimals[i] contains the superinstruction
 1182:    on the shortest path to the end of the BB */
 1183: void optimize_bb(PrimNum origs[], PrimNum optimals[], int ninsts)
 1184: {
 1185:   int i,j, mincost;
 1186:   static int costs[MAX_BB+1];
 1187: 
 1188:   assert(ninsts<MAX_BB);
 1189:   costs[ninsts]=0;
 1190:   for (i=ninsts-1; i>=0; i--) {
 1191:     optimals[i] = origs[i];
 1192:     costs[i] = mincost = costs[i+1] + ss_cost(optimals[i]);
 1193:     for (j=2; j<=max_super && i+j<=ninsts ; j++) {
 1194:       int super, jcost;
 1195: 
 1196:       super = lookup_super(origs+i,j);
 1197:       if (super >= 0) {
 1198: 	jcost = costs[i+j] + ss_cost(super);
 1199: 	if (jcost <= mincost) {
 1200: 	  optimals[i] = super;
 1201: 	  mincost = jcost;
 1202: 	  if (!ss_greedy)
 1203: 	    costs[i] = jcost;
 1204: 	}
 1205:       }
 1206:     }
 1207:   }
 1208: }
 1209: 
 1210: /* rewrite the instructions pointed to by instps to use the
 1211:    superinstructions in optimals */
 1212: void rewrite_bb(Cell *instps[], PrimNum *optimals, int ninsts)
 1213: {
 1214:   int i,j, nextdyn;
 1215:   Cell inst;
 1216: 
 1217:   for (i=0, nextdyn=0; i<ninsts; i++) {
 1218:     if (i==nextdyn) { /* compile dynamically */
 1219:       nextdyn += super_costs[optimals[i]].length;
 1220:       inst = compile_prim_dyn(optimals[i]);
 1221:       for (j=0; j<sizeof(cost_sums)/sizeof(cost_sums[0]); j++)
 1222: 	cost_sums[j].sum += cost_sums[j].costfunc(optimals[i]);
 1223:     } else { /* compile statically */
 1224:       inst = (Cell)vm_prims[optimals[i]];
 1225:     }
 1226:     *(instps[i]) = inst;
 1227:   }
 1228: }
 1229: 
 1230: /* compile *start, possibly rewriting it into a static and/or dynamic
 1231:    superinstruction */
 1232: void compile_prim1(Cell *start)
 1233: {
 1234: #if defined(DOUBLY_INDIRECT)
 1235:   Label prim=(Label)*start;
 1236:   if (prim<((Label)(xts+DOESJUMP)) || prim>((Label)(xts+npriminfos))) {
 1237:     fprintf(stderr,"compile_prim encountered xt %p\n", prim);
 1238:     *start=(Cell)prim;
 1239:     return;
 1240:   } else {
 1241:     *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
 1242:     return;
 1243:   }
 1244: #elif defined(INDIRECT_THREADED)
 1245:   return;
 1246: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
 1247:   static Cell *instps[MAX_BB];
 1248:   static PrimNum origs[MAX_BB];
 1249:   static PrimNum optimals[MAX_BB];
 1250:   static int ninsts=0;
 1251:   PrimNum prim_num;
 1252: 
 1253:   if (start==NULL)
 1254:     goto end_bb;
 1255:   prim_num = ((Xt)*start)-vm_prims;
 1256:   if (prim_num >= npriminfos)
 1257:     goto end_bb;
 1258:   assert(ninsts<MAX_BB);
 1259:   instps[ninsts] = start;
 1260:   origs[ninsts] = prim_num;
 1261:   ninsts++;
 1262:   if (ninsts >= MAX_BB || superend[prim_num]) {
 1263:   end_bb:
 1264:     optimize_bb(origs,optimals,ninsts);
 1265:     rewrite_bb(instps,optimals,ninsts);
 1266:     ninsts=0;
 1267:   }
 1268: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
 1269: }
 1270: 
 1271: Address loader(FILE *imagefile, char* filename)
 1272: /* returns the address of the image proper (after the preamble) */
 1273: {
 1274:   ImageHeader header;
 1275:   Address image;
 1276:   Address imp; /* image+preamble */
 1277:   Char magic[8];
 1278:   char magic7; /* size byte of magic number */
 1279:   Cell preamblesize=0;
 1280:   Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
 1281:   UCell check_sum;
 1282:   Cell ausize = ((RELINFOBITS ==  8) ? 0 :
 1283: 		 (RELINFOBITS == 16) ? 1 :
 1284: 		 (RELINFOBITS == 32) ? 2 : 3);
 1285:   Cell charsize = ((sizeof(Char) == 1) ? 0 :
 1286: 		   (sizeof(Char) == 2) ? 1 :
 1287: 		   (sizeof(Char) == 4) ? 2 : 3) + ausize;
 1288:   Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
 1289: 		   (sizeof(Cell) == 2) ? 1 :
 1290: 		   (sizeof(Cell) == 4) ? 2 : 3) + ausize;
 1291:   Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
 1292: #ifdef WORDS_BIGENDIAN
 1293:        0
 1294: #else
 1295:        1
 1296: #endif
 1297:     ;
 1298: 
 1299:   vm_prims = engine(0,0,0,0,0);
 1300:   check_prims(vm_prims);
 1301:   prepare_super_table();
 1302: #ifndef DOUBLY_INDIRECT
 1303: #ifdef PRINT_SUPER_LENGTHS
 1304:   print_super_lengths();
 1305: #endif
 1306:   check_sum = checksum(vm_prims);
 1307: #else /* defined(DOUBLY_INDIRECT) */
 1308:   check_sum = (UCell)vm_prims;
 1309: #endif /* defined(DOUBLY_INDIRECT) */
 1310:   
 1311:   do {
 1312:     if(fread(magic,sizeof(Char),8,imagefile) < 8) {
 1313:       fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
 1314: 	      progname, filename);
 1315:       exit(1);
 1316:     }
 1317:     preamblesize+=8;
 1318:   } while(memcmp(magic,"Gforth3",7));
 1319:   magic7 = magic[7];
 1320:   if (debug) {
 1321:     magic[7]='\0';
 1322:     fprintf(stderr,"Magic found: %s ", magic);
 1323:     print_sizes(magic7);
 1324:   }
 1325: 
 1326:   if (magic7 != sizebyte)
 1327:     {
 1328:       fprintf(stderr,"This image is:         ");
 1329:       print_sizes(magic7);
 1330:       fprintf(stderr,"whereas the machine is ");
 1331:       print_sizes(sizebyte);
 1332:       exit(-2);
 1333:     };
 1334: 
 1335:   fread((void *)&header,sizeof(ImageHeader),1,imagefile);
 1336: 
 1337:   set_stack_sizes(&header);
 1338:   
 1339: #if HAVE_GETPAGESIZE
 1340:   pagesize=getpagesize(); /* Linux/GNU libc offers this */
 1341: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
 1342:   pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
 1343: #elif PAGESIZE
 1344:   pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
 1345: #endif
 1346:   if (debug)
 1347:     fprintf(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
 1348: 
 1349:   image = dict_alloc_read(imagefile, preamblesize+header.image_size,
 1350: 			  preamblesize+dictsize, data_offset);
 1351:   imp=image+preamblesize;
 1352:   alloc_stacks((ImageHeader *)imp);
 1353:   if (clear_dictionary)
 1354:     memset(imp+header.image_size, 0, dictsize-header.image_size);
 1355:   if(header.base==0 || header.base  == (Address)0x100) {
 1356:     Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
 1357:     char reloc_bits[reloc_size];
 1358:     fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
 1359:     fread(reloc_bits, 1, reloc_size, imagefile);
 1360:     relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
 1361: #if 0
 1362:     { /* let's see what the relocator did */
 1363:       FILE *snapshot=fopen("snapshot.fi","wb");
 1364:       fwrite(image,1,imagesize,snapshot);
 1365:       fclose(snapshot);
 1366:     }
 1367: #endif
 1368:   }
 1369:   else if(header.base!=imp) {
 1370:     fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
 1371: 	    progname, (unsigned long)header.base, (unsigned long)imp);
 1372:     exit(1);
 1373:   }
 1374:   if (header.checksum==0)
 1375:     ((ImageHeader *)imp)->checksum=check_sum;
 1376:   else if (header.checksum != check_sum) {
 1377:     fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
 1378: 	    progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
 1379:     exit(1);
 1380:   }
 1381: #ifdef DOUBLY_INDIRECT
 1382:   ((ImageHeader *)imp)->xt_base = xts;
 1383: #endif
 1384:   fclose(imagefile);
 1385: 
 1386:   /* unnecessary, except maybe for CODE words */
 1387:   /* FLUSH_ICACHE(imp, header.image_size);*/
 1388: 
 1389:   return imp;
 1390: }
 1391: 
 1392: /* pointer to last '/' or '\' in file, 0 if there is none. */
 1393: char *onlypath(char *filename)
 1394: {
 1395:   return strrchr(filename, DIRSEP);
 1396: }
 1397: 
 1398: FILE *openimage(char *fullfilename)
 1399: {
 1400:   FILE *image_file;
 1401:   char * expfilename = tilde_cstr(fullfilename, strlen(fullfilename), 1);
 1402: 
 1403:   image_file=fopen(expfilename,"rb");
 1404:   if (image_file!=NULL && debug)
 1405:     fprintf(stderr, "Opened image file: %s\n", expfilename);
 1406:   return image_file;
 1407: }
 1408: 
 1409: /* try to open image file concat(path[0:len],imagename) */
 1410: FILE *checkimage(char *path, int len, char *imagename)
 1411: {
 1412:   int dirlen=len;
 1413:   char fullfilename[dirlen+strlen(imagename)+2];
 1414: 
 1415:   memcpy(fullfilename, path, dirlen);
 1416:   if (fullfilename[dirlen-1]!=DIRSEP)
 1417:     fullfilename[dirlen++]=DIRSEP;
 1418:   strcpy(fullfilename+dirlen,imagename);
 1419:   return openimage(fullfilename);
 1420: }
 1421: 
 1422: FILE * open_image_file(char * imagename, char * path)
 1423: {
 1424:   FILE * image_file=NULL;
 1425:   char *origpath=path;
 1426:   
 1427:   if(strchr(imagename, DIRSEP)==NULL) {
 1428:     /* first check the directory where the exe file is in !! 01may97jaw */
 1429:     if (onlypath(progname))
 1430:       image_file=checkimage(progname, onlypath(progname)-progname, imagename);
 1431:     if (!image_file)
 1432:       do {
 1433: 	char *pend=strchr(path, PATHSEP);
 1434: 	if (pend==NULL)
 1435: 	  pend=path+strlen(path);
 1436: 	if (strlen(path)==0) break;
 1437: 	image_file=checkimage(path, pend-path, imagename);
 1438: 	path=pend+(*pend==PATHSEP);
 1439:       } while (image_file==NULL);
 1440:   } else {
 1441:     image_file=openimage(imagename);
 1442:   }
 1443: 
 1444:   if (!image_file) {
 1445:     fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
 1446: 	    progname, imagename, origpath);
 1447:     exit(1);
 1448:   }
 1449: 
 1450:   return image_file;
 1451: }
 1452: #endif
 1453: 
 1454: #ifdef HAS_OS
 1455: UCell convsize(char *s, UCell elemsize)
 1456: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
 1457:    of bytes.  the letter at the end indicates the unit, where e stands
 1458:    for the element size. default is e */
 1459: {
 1460:   char *endp;
 1461:   UCell n,m;
 1462: 
 1463:   m = elemsize;
 1464:   n = strtoul(s,&endp,0);
 1465:   if (endp!=NULL) {
 1466:     if (strcmp(endp,"b")==0)
 1467:       m=1;
 1468:     else if (strcmp(endp,"k")==0)
 1469:       m=1024;
 1470:     else if (strcmp(endp,"M")==0)
 1471:       m=1024*1024;
 1472:     else if (strcmp(endp,"G")==0)
 1473:       m=1024*1024*1024;
 1474:     else if (strcmp(endp,"T")==0) {
 1475: #if (SIZEOF_CHAR_P > 4)
 1476:       m=1024L*1024*1024*1024;
 1477: #else
 1478:       fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
 1479:       exit(1);
 1480: #endif
 1481:     } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
 1482:       fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
 1483:       exit(1);
 1484:     }
 1485:   }
 1486:   return n*m;
 1487: }
 1488: 
 1489: enum {
 1490:   ss_number = 256,
 1491:   ss_min_codesize,
 1492:   ss_min_ls,
 1493:   ss_min_lsu,
 1494:   ss_min_nexts,
 1495: };
 1496: 
 1497: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
 1498: {
 1499:   int c;
 1500: 
 1501:   opterr=0;
 1502:   while (1) {
 1503:     int option_index=0;
 1504:     static struct option opts[] = {
 1505:       {"appl-image", required_argument, NULL, 'a'},
 1506:       {"image-file", required_argument, NULL, 'i'},
 1507:       {"dictionary-size", required_argument, NULL, 'm'},
 1508:       {"data-stack-size", required_argument, NULL, 'd'},
 1509:       {"return-stack-size", required_argument, NULL, 'r'},
 1510:       {"fp-stack-size", required_argument, NULL, 'f'},
 1511:       {"locals-stack-size", required_argument, NULL, 'l'},
 1512:       {"path", required_argument, NULL, 'p'},
 1513:       {"version", no_argument, NULL, 'v'},
 1514:       {"help", no_argument, NULL, 'h'},
 1515:       /* put something != 0 into offset_image */
 1516:       {"offset-image", no_argument, &offset_image, 1},
 1517:       {"no-offset-im", no_argument, &offset_image, 0},
 1518:       {"clear-dictionary", no_argument, &clear_dictionary, 1},
 1519:       {"die-on-signal", no_argument, &die_on_signal, 1},
 1520:       {"debug", no_argument, &debug, 1},
 1521:       {"no-super", no_argument, &no_super, 1},
 1522:       {"no-dynamic", no_argument, &no_dynamic, 1},
 1523:       {"dynamic", no_argument, &no_dynamic, 0},
 1524:       {"print-metrics", no_argument, &print_metrics, 1},
 1525:       {"ss-number", required_argument, NULL, ss_number},
 1526: #ifndef NO_DYNAMIC
 1527:       {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
 1528: #endif
 1529:       {"ss-min-ls",       no_argument, NULL, ss_min_ls},
 1530:       {"ss-min-lsu",      no_argument, NULL, ss_min_lsu},
 1531:       {"ss-min-nexts",    no_argument, NULL, ss_min_nexts},
 1532:       {"ss-greedy",       no_argument, &ss_greedy, 1},
 1533:       {0,0,0,0}
 1534:       /* no-init-file, no-rc? */
 1535:     };
 1536:     
 1537:     c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
 1538:     
 1539:     switch (c) {
 1540:     case EOF: return;
 1541:     case '?': optind--; return;
 1542:     case 'a': *imagename = optarg; return;
 1543:     case 'i': *imagename = optarg; break;
 1544:     case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
 1545:     case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
 1546:     case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
 1547:     case 'f': fsize = convsize(optarg,sizeof(Float)); break;
 1548:     case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
 1549:     case 'p': *path = optarg; break;
 1550:     case 'o': offset_image = 1; break;
 1551:     case 'n': offset_image = 0; break;
 1552:     case 'c': clear_dictionary = 1; break;
 1553:     case 's': die_on_signal = 1; break;
 1554:     case 'x': debug = 1; break;
 1555:     case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
 1556:     case ss_number: static_super_number = atoi(optarg); break;
 1557: #ifndef NO_DYNAMIC
 1558:     case ss_min_codesize: ss_cost = cost_codesize; break;
 1559: #endif
 1560:     case ss_min_ls:       ss_cost = cost_ls;       break;
 1561:     case ss_min_lsu:      ss_cost = cost_lsu;      break;
 1562:     case ss_min_nexts:    ss_cost = cost_nexts;    break;
 1563:     case 'h': 
 1564:       fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
 1565: Engine Options:\n\
 1566:   --appl-image FILE		    equivalent to '--image-file=FILE --'\n\
 1567:   --clear-dictionary		    Initialize the dictionary with 0 bytes\n\
 1568:   -d SIZE, --data-stack-size=SIZE   Specify data stack size\n\
 1569:   --debug			    Print debugging information during startup\n\
 1570:   --die-on-signal		    exit instead of CATCHing some signals\n\
 1571:   --dynamic			    use dynamic native code\n\
 1572:   -f SIZE, --fp-stack-size=SIZE	    Specify floating point stack size\n\
 1573:   -h, --help			    Print this message and exit\n\
 1574:   -i FILE, --image-file=FILE	    Use image FILE instead of `gforth.fi'\n\
 1575:   -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
 1576:   -m SIZE, --dictionary-size=SIZE   Specify Forth dictionary size\n\
 1577:   --no-dynamic			    Use only statically compiled primitives\n\
 1578:   --no-offset-im		    Load image at normal position\n\
 1579:   --no-super                        No dynamically formed superinstructions\n\
 1580:   --offset-image		    Load image at a different position\n\
 1581:   -p PATH, --path=PATH		    Search path for finding image and sources\n\
 1582:   --print-metrics		    Print some code generation metrics on exit\n\
 1583:   -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
 1584:   --ss-greedy                       greedy, not optimal superinst selection\n\
 1585:   --ss-min-codesize                 select superinsts for smallest native code\n\
 1586:   --ss-min-ls                       minimize loads and stores\n\
 1587:   --ss-min-lsu                      minimize loads, stores, and pointer updates\n\
 1588:   --ss-min-nexts                    minimize the number of static superinsts\n\
 1589:   --ss-number=N                     use N static superinsts (default max)\n\
 1590:   -v, --version			    Print engine version and exit\n\
 1591: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
 1592:   `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
 1593: 	      argv[0]);
 1594:       optind--;
 1595:       return;
 1596:     }
 1597:   }
 1598: }
 1599: #endif
 1600: 
 1601: #ifdef INCLUDE_IMAGE
 1602: extern Cell image[];
 1603: extern const char reloc_bits[];
 1604: #endif
 1605: 
 1606: int main(int argc, char **argv, char **env)
 1607: {
 1608: #ifdef HAS_OS
 1609:   char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
 1610: #else
 1611:   char *path = DEFAULTPATH;
 1612: #endif
 1613: #ifndef INCLUDE_IMAGE
 1614:   char *imagename="gforth.fi";
 1615:   FILE *image_file;
 1616:   Address image;
 1617: #endif
 1618:   int retvalue;
 1619: 	  
 1620: #if defined(i386) && defined(ALIGNMENT_CHECK)
 1621:   /* turn on alignment checks on the 486.
 1622:    * on the 386 this should have no effect. */
 1623:   __asm__("pushfl; popl %eax; orl $0x40000, %eax; pushl %eax; popfl;");
 1624:   /* this is unusable with Linux' libc.4.6.27, because this library is
 1625:      not alignment-clean; we would have to replace some library
 1626:      functions (e.g., memcpy) to make it work. Also GCC doesn't try to keep
 1627:      the stack FP-aligned. */
 1628: #endif
 1629: 
 1630:   /* buffering of the user output device */
 1631: #ifdef _IONBF
 1632:   if (isatty(fileno(stdout))) {
 1633:     fflush(stdout);
 1634:     setvbuf(stdout,NULL,_IONBF,0);
 1635:   }
 1636: #endif
 1637: 
 1638:   progname = argv[0];
 1639: 
 1640: #ifdef HAS_OS
 1641:   gforth_args(argc, argv, &path, &imagename);
 1642: #ifndef NO_DYNAMIC
 1643:   if (no_dynamic && ss_cost == cost_codesize) {
 1644:     ss_cost = cost_nexts;
 1645:     cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
 1646:     if (debug)
 1647:       fprintf(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
 1648:   }
 1649: #endif /* !defined(NO_DYNAMIC) */
 1650: #endif /* defined(HAS_OS) */
 1651: 
 1652: #ifdef INCLUDE_IMAGE
 1653:   set_stack_sizes((ImageHeader *)image);
 1654:   if(((ImageHeader *)image)->base != image)
 1655:     relocate(image, reloc_bits, ((ImageHeader *)image)->image_size,
 1656: 	     (Label*)engine(0, 0, 0, 0, 0));
 1657:   alloc_stacks((ImageHeader *)image);
 1658: #else
 1659:   image_file = open_image_file(imagename, path);
 1660:   image = loader(image_file, imagename);
 1661: #endif
 1662:   gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
 1663: 
 1664:   {
 1665:     char path2[strlen(path)+1];
 1666:     char *p1, *p2;
 1667:     Cell environ[]= {
 1668:       (Cell)argc-(optind-1),
 1669:       (Cell)(argv+(optind-1)),
 1670:       (Cell)strlen(path),
 1671:       (Cell)path2};
 1672:     argv[optind-1] = progname;
 1673:     /*
 1674:        for (i=0; i<environ[0]; i++)
 1675:        printf("%s\n", ((char **)(environ[1]))[i]);
 1676:        */
 1677:     /* make path OS-independent by replacing path separators with NUL */
 1678:     for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
 1679:       if (*p1==PATHSEP)
 1680: 	*p2 = '\0';
 1681:       else
 1682: 	*p2 = *p1;
 1683:     *p2='\0';
 1684:     retvalue = go_forth(image, 4, environ);
 1685: #ifdef SIGPIPE
 1686:     bsd_signal(SIGPIPE, SIG_IGN);
 1687: #endif
 1688: #ifdef VM_PROFILING
 1689:     vm_print_profile(stderr);
 1690: #endif
 1691:     deprep_terminal();
 1692:   }
 1693:   if (print_metrics) {
 1694:     int i;
 1695:     fprintf(stderr, "code size = %8ld\n", dyncodesize());
 1696:     for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
 1697:       fprintf(stderr, "metric %8s: %8ld\n",
 1698: 	      cost_sums[i].metricname, cost_sums[i].sum);
 1699:   }
 1700:   return retvalue;
 1701: }

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