Annotation of gforth/engine/main.c, revision 1.144

1.1       anton       1: /* command line interpretation, image loading etc. for Gforth
                      2: 
                      3: 
1.143     anton       4:   Copyright (C) 1995,1996,1997,1998,2000,2003,2004 Free Software Foundation, Inc.
1.1       anton       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
1.40      anton      20:   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
1.1       anton      21: */
                     22: 
                     23: #include "config.h"
1.82      anton      24: #include "forth.h"
1.1       anton      25: #include <errno.h>
                     26: #include <ctype.h>
                     27: #include <stdio.h>
1.2       pazsan     28: #include <unistd.h>
1.1       anton      29: #include <string.h>
                     30: #include <math.h>
                     31: #include <sys/types.h>
1.32      pazsan     32: #ifndef STANDALONE
1.1       anton      33: #include <sys/stat.h>
1.32      pazsan     34: #endif
1.1       anton      35: #include <fcntl.h>
                     36: #include <assert.h>
                     37: #include <stdlib.h>
1.102     anton      38: #include <signal.h>
1.11      pazsan     39: #ifndef STANDALONE
1.1       anton      40: #if HAVE_SYS_MMAN_H
                     41: #include <sys/mman.h>
                     42: #endif
1.11      pazsan     43: #endif
1.1       anton      44: #include "io.h"
                     45: #include "getopt.h"
1.11      pazsan     46: #ifdef STANDALONE
                     47: #include <systypes.h>
                     48: #endif
1.1       anton      49: 
1.121     anton      50: typedef enum prim_num {
1.119     anton      51: /* definitions of N_execute etc. */
1.126     anton      52: #include PRIM_NUM_I
1.119     anton      53:   N_START_SUPER
1.121     anton      54: } PrimNum;
1.119     anton      55: 
1.79      anton      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: 
1.115     pazsan     63: #ifdef HAS_FFCALL
                     64: Cell *RP;
                     65: Address LP;
                     66: 
                     67: #include <callback.h>
                     68: 
                     69: va_alist clist;
                     70: 
1.116     pazsan     71: void engine_callback(Xt* fcall, void * alist)
1.115     pazsan     72: {
1.140     pazsan     73:   /* save global valiables */
                     74:   Cell *rp = RP;
                     75:   Cell *sp = SP;
                     76:   Float *fp = FP;
                     77:   Address lp = LP;
                     78: 
1.120     pazsan     79:   clist = (va_alist)alist;
1.140     pazsan     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;
1.115     pazsan     88: }
                     89: #endif
                     90: 
1.79      anton      91: #ifdef GFORTH_DEBUGGING
                     92: /* define some VM registers as global variables, so they survive exceptions;
                     93:    global register variables are not up to the task (according to the 
                     94:    GNU C manual) */
                     95: Xt *saved_ip;
                     96: Cell *rp;
                     97: #endif
                     98: 
                     99: #ifdef NO_IP
                    100: Label next_code;
                    101: #endif
                    102: 
                    103: #ifdef HAS_FILE
                    104: char* fileattr[6]={"rb","rb","r+b","r+b","wb","wb"};
                    105: char* pfileattr[6]={"r","r","r+","r+","w","w"};
                    106: 
                    107: #ifndef O_BINARY
                    108: #define O_BINARY 0
                    109: #endif
                    110: #ifndef O_TEXT
                    111: #define O_TEXT 0
                    112: #endif
                    113: 
                    114: int ufileattr[6]= {
                    115:   O_RDONLY|O_BINARY, O_RDONLY|O_BINARY,
                    116:   O_RDWR  |O_BINARY, O_RDWR  |O_BINARY,
                    117:   O_WRONLY|O_BINARY, O_WRONLY|O_BINARY };
                    118: #endif
                    119: /* end global vars for engine.c */
                    120: 
1.1       anton     121: #define PRIM_VERSION 1
                    122: /* increment this whenever the primitives change in an incompatible way */
                    123: 
1.14      pazsan    124: #ifndef DEFAULTPATH
1.39      anton     125: #  define DEFAULTPATH "."
1.14      pazsan    126: #endif
                    127: 
1.1       anton     128: #ifdef MSDOS
                    129: jmp_buf throw_jmp_buf;
                    130: #endif
                    131: 
1.56      anton     132: #if defined(DOUBLY_INDIRECT)
                    133: #  define CFA(n)       ({Cell _n = (n); ((Cell)(((_n & 0x4000) ? symbols : xts)+(_n&~0x4000UL)));})
1.1       anton     134: #else
1.56      anton     135: #  define CFA(n)       ((Cell)(symbols+((n)&~0x4000UL)))
1.1       anton     136: #endif
                    137: 
                    138: #define maxaligned(n)  (typeof(n))((((Cell)n)+sizeof(Float)-1)&-sizeof(Float))
                    139: 
                    140: static UCell dictsize=0;
                    141: static UCell dsize=0;
                    142: static UCell rsize=0;
                    143: static UCell fsize=0;
                    144: static UCell lsize=0;
                    145: int offset_image=0;
1.4       anton     146: int die_on_signal=0;
1.13      pazsan    147: #ifndef INCLUDE_IMAGE
1.1       anton     148: static int clear_dictionary=0;
1.24      anton     149: UCell pagesize=1;
1.22      pazsan    150: char *progname;
                    151: #else
                    152: char *progname = "gforth";
                    153: int optind = 1;
1.13      pazsan    154: #endif
1.31      pazsan    155: 
1.131     anton     156: #define CODE_BLOCK_SIZE (4096*1024) /* !! overflow handling for -native */
1.48      anton     157: Address code_area=0;
1.73      anton     158: Cell code_area_size = CODE_BLOCK_SIZE;
1.75      anton     159: Address code_here=NULL+CODE_BLOCK_SIZE; /* does for code-area what HERE
                    160:                                           does for the dictionary */
1.100     anton     161: Address start_flush=NULL; /* start of unflushed code */
1.74      anton     162: Cell last_jump=0; /* if the last prim was compiled without jump, this
                    163:                      is it's number, otherwise this contains 0 */
1.48      anton     164: 
1.60      anton     165: static int no_super=0;   /* true if compile_prim should not fuse prims */
1.81      anton     166: static int no_dynamic=NO_DYNAMIC_DEFAULT; /* if true, no code is generated
                    167:                                             dynamically */
1.110     anton     168: static int print_metrics=0; /* if true, print metrics on exit */
1.109     anton     169: static int static_super_number = 10000000; /* number of ss used if available */
1.125     anton     170: #define MAX_STATE 4 /* maximum number of states */
                    171: static int maxstates = MAX_STATE; /* number of states for stack caching */
1.110     anton     172: static int ss_greedy = 0; /* if true: use greedy, not optimal ss selection */
1.144   ! pazsan    173: static int diag = 0; /* if true: print diagnostic informations */
        !           174: static int relocs = 0;
        !           175: static int nonrelocs = 0;
1.60      anton     176: 
1.30      pazsan    177: #ifdef HAS_DEBUG
1.68      anton     178: int debug=0;
1.144   ! pazsan    179: # define debugp(x...) if (debug) fprintf(x);
1.31      pazsan    180: #else
                    181: # define perror(x...)
                    182: # define fprintf(x...)
1.144   ! pazsan    183: # define debugp(x...)
1.30      pazsan    184: #endif
1.31      pazsan    185: 
1.24      anton     186: ImageHeader *gforth_header;
1.43      anton     187: Label *vm_prims;
1.53      anton     188: #ifdef DOUBLY_INDIRECT
                    189: Label *xts; /* same content as vm_prims, but should only be used for xts */
                    190: #endif
1.1       anton     191: 
1.125     anton     192: #ifndef NO_DYNAMIC
                    193: #define MAX_IMMARGS 2
                    194: 
                    195: typedef struct {
                    196:   Label start; /* NULL if not relocatable */
                    197:   Cell length; /* only includes the jump iff superend is true*/
                    198:   Cell restlength; /* length of the rest (i.e., the jump or (on superend) 0) */
                    199:   char superend; /* true if primitive ends superinstruction, i.e.,
                    200:                      unconditional branch, execute, etc. */
                    201:   Cell nimmargs;
                    202:   struct immarg {
                    203:     Cell offset; /* offset of immarg within prim */
                    204:     char rel;    /* true if immarg is relative */
                    205:   } immargs[MAX_IMMARGS];
                    206: } PrimInfo;
                    207: 
                    208: PrimInfo *priminfos;
                    209: PrimInfo **decomp_prims;
                    210: 
1.139     anton     211: const char const* const prim_names[]={
                    212: #include PRIM_NAMES_I
                    213: };
                    214: 
1.125     anton     215: static int is_relocatable(int p)
                    216: {
                    217:   return !no_dynamic && priminfos[p].start != NULL;
                    218: }
                    219: #else /* defined(NO_DYNAMIC) */
                    220: static int is_relocatable(int p)
                    221: {
                    222:   return 0;
                    223: }
                    224: #endif /* defined(NO_DYNAMIC) */
                    225: 
1.30      pazsan    226: #ifdef MEMCMP_AS_SUBROUTINE
                    227: int gforth_memcmp(const char * s1, const char * s2, size_t n)
                    228: {
                    229:   return memcmp(s1, s2, n);
                    230: }
                    231: #endif
                    232: 
1.125     anton     233: static Cell max(Cell a, Cell b)
                    234: {
                    235:   return a>b?a:b;
                    236: }
                    237: 
                    238: static Cell min(Cell a, Cell b)
                    239: {
                    240:   return a<b?a:b;
                    241: }
                    242: 
1.1       anton     243: /* image file format:
1.15      pazsan    244:  *  "#! binary-path -i\n" (e.g., "#! /usr/local/bin/gforth-0.4.0 -i\n")
1.1       anton     245:  *   padding to a multiple of 8
1.84      anton     246:  *   magic: "Gforth3x" means format 0.6,
1.15      pazsan    247:  *              where x is a byte with
                    248:  *              bit 7:   reserved = 0
                    249:  *              bit 6:5: address unit size 2^n octets
                    250:  *              bit 4:3: character size 2^n octets
                    251:  *              bit 2:1: cell size 2^n octets
                    252:  *              bit 0:   endian, big=0, little=1.
                    253:  *  The magic are always 8 octets, no matter what the native AU/character size is
1.1       anton     254:  *  padding to max alignment (no padding necessary on current machines)
1.24      anton     255:  *  ImageHeader structure (see forth.h)
1.1       anton     256:  *  data (size in ImageHeader.image_size)
                    257:  *  tags ((if relocatable, 1 bit/data cell)
                    258:  *
                    259:  * tag==1 means that the corresponding word is an address;
                    260:  * If the word is >=0, the address is within the image;
                    261:  * addresses within the image are given relative to the start of the image.
                    262:  * If the word =-1 (CF_NIL), the address is NIL,
                    263:  * If the word is <CF_NIL and >CF(DODOES), it's a CFA (:, Create, ...)
                    264:  * If the word =CF(DODOES), it's a DOES> CFA
                    265:  * If the word =CF(DOESJUMP), it's a DOES JUMP (2 Cells after DOES>,
                    266:  *                                     possibly containing a jump to dodoes)
1.51      anton     267:  * If the word is <CF(DOESJUMP) and bit 14 is set, it's the xt of a primitive
                    268:  * If the word is <CF(DOESJUMP) and bit 14 is clear, 
                    269:  *                                        it's the threaded code of a primitive
1.85      pazsan    270:  * bits 13..9 of a primitive token state which group the primitive belongs to,
                    271:  * bits 8..0 of a primitive token index into the group
1.1       anton     272:  */
                    273: 
1.115     pazsan    274: Cell groups[32] = {
1.85      pazsan    275:   0,
1.121     anton     276:   0
1.90      anton     277: #undef GROUP
1.115     pazsan    278: #undef GROUPADD
                    279: #define GROUPADD(n) +n
                    280: #define GROUP(x, n) , 0
1.126     anton     281: #include PRIM_GRP_I
1.90      anton     282: #undef GROUP
1.115     pazsan    283: #undef GROUPADD
1.85      pazsan    284: #define GROUP(x, n)
1.115     pazsan    285: #define GROUPADD(n)
1.85      pazsan    286: };
                    287: 
1.125     anton     288: unsigned char *branch_targets(Cell *image, const unsigned char *bitstring,
                    289:                              int size, Cell base)
                    290:      /* produce a bitmask marking all the branch targets */
                    291: {
1.130     anton     292:   int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
1.125     anton     293:   Cell token;
                    294:   unsigned char bits;
1.130     anton     295:   unsigned char *result=malloc(steps);
                    296: 
                    297:   memset(result, 0, steps);
                    298:   for(k=0; k<steps; k++) {
1.125     anton     299:     for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
1.130     anton     300:       if(bits & (1U << (RELINFOBITS-1))) {
                    301:        assert(i*sizeof(Cell) < size);
1.125     anton     302:         token=image[i];
                    303:        if (token>=base) { /* relocatable address */
                    304:          UCell bitnum=(token-base)/sizeof(Cell);
                    305:          result[bitnum/RELINFOBITS] |= 1U << ((~bitnum)&(RELINFOBITS-1));
                    306:        }
                    307:       }
                    308:     }
                    309:   }
                    310:   return result;
                    311: }
                    312: 
1.115     pazsan    313: void relocate(Cell *image, const unsigned char *bitstring, 
1.90      anton     314:               int size, Cell base, Label symbols[])
1.1       anton     315: {
1.130     anton     316:   int i=0, j, k, steps=(((size-1)/sizeof(Cell))/RELINFOBITS)+1;
1.11      pazsan    317:   Cell token;
1.1       anton     318:   char bits;
1.37      anton     319:   Cell max_symbols;
1.46      jwilke    320:   /* 
1.85      pazsan    321:    * A virtual start address that's the real start address minus 
1.46      jwilke    322:    * the one in the image 
                    323:    */
1.45      jwilke    324:   Cell *start = (Cell * ) (((void *) image) - ((void *) base));
1.125     anton     325:   unsigned char *targets = branch_targets(image, bitstring, size, base);
1.1       anton     326: 
1.85      pazsan    327:   /* group index into table */
1.115     pazsan    328:   if(groups[31]==0) {
                    329:     int groupsum=0;
                    330:     for(i=0; i<32; i++) {
                    331:       groupsum += groups[i];
                    332:       groups[i] = groupsum;
                    333:       /* printf("group[%d]=%d\n",i,groupsum); */
                    334:     }
                    335:     i=0;
                    336:   }
1.46      jwilke    337:   
                    338: /* printf("relocating to %x[%x] start=%x base=%x\n", image, size, start, base); */
1.37      anton     339:   
1.121     anton     340:   for (max_symbols=0; symbols[max_symbols]!=0; max_symbols++)
1.37      anton     341:     ;
1.47      anton     342:   max_symbols--;
1.35      pazsan    343: 
1.130     anton     344:   for(k=0; k<steps; k++) {
1.13      pazsan    345:     for(j=0, bits=bitstring[k]; j<RELINFOBITS; j++, i++, bits<<=1) {
1.1       anton     346:       /*      fprintf(stderr,"relocate: image[%d]\n", i);*/
1.130     anton     347:       if(bits & (1U << (RELINFOBITS-1))) {
                    348:        assert(i*sizeof(Cell) < size);
1.35      pazsan    349:        /* fprintf(stderr,"relocate: image[%d]=%d of %d\n", i, image[i], size/sizeof(Cell)); */
1.45      jwilke    350:         token=image[i];
1.85      pazsan    351:        if(token<0) {
                    352:          int group = (-token & 0x3E00) >> 9;
                    353:          if(group == 0) {
                    354:            switch(token|0x4000) {
1.1       anton     355:            case CF_NIL      : image[i]=0; break;
                    356: #if !defined(DOUBLY_INDIRECT)
                    357:            case CF(DOCOL)   :
                    358:            case CF(DOVAR)   :
                    359:            case CF(DOCON)   :
                    360:            case CF(DOUSER)  : 
                    361:            case CF(DODEFER) : 
1.11      pazsan    362:            case CF(DOFIELD) : MAKE_CF(image+i,symbols[CF(token)]); break;
1.92      anton     363:            case CF(DOESJUMP): image[i]=0; break;
1.1       anton     364: #endif /* !defined(DOUBLY_INDIRECT) */
                    365:            case CF(DODOES)  :
1.45      jwilke    366:              MAKE_DOES_CF(image+i,(Xt *)(image[i+1]+((Cell)start)));
1.1       anton     367:              break;
1.85      pazsan    368:            default          : /* backward compatibility */
1.56      anton     369: /*           printf("Code field generation image[%x]:=CFA(%x)\n",
1.1       anton     370:                     i, CF(image[i])); */
1.55      anton     371:              if (CF((token | 0x4000))<max_symbols) {
1.56      anton     372:                image[i]=(Cell)CFA(CF(token));
                    373: #ifdef DIRECT_THREADED
1.125     anton     374:                if ((token & 0x4000) == 0) { /* threade code, no CFA */
                    375:                  if (targets[k] & (1U<<(RELINFOBITS-1-j)))
                    376:                    compile_prim1(0);
1.70      anton     377:                  compile_prim1(&image[i]);
1.125     anton     378:                }
1.56      anton     379: #endif
1.55      anton     380:              } else
1.115     pazsan    381:                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);
1.1       anton     382:            }
1.85      pazsan    383:          } else {
                    384:            int tok = -token & 0x1FF;
                    385:            if (tok < (groups[group+1]-groups[group])) {
                    386: #if defined(DOUBLY_INDIRECT)
                    387:              image[i]=(Cell)CFA(((groups[group]+tok) | (CF(token) & 0x4000)));
                    388: #else
                    389:              image[i]=(Cell)CFA((groups[group]+tok));
                    390: #endif
                    391: #ifdef DIRECT_THREADED
1.125     anton     392:              if ((token & 0x4000) == 0) { /* threade code, no CFA */
                    393:                if (targets[k] & (1U<<(RELINFOBITS-1-j)))
                    394:                  compile_prim1(0);
1.85      pazsan    395:                compile_prim1(&image[i]);
1.125     anton     396:              }
1.85      pazsan    397: #endif
                    398:            } else
1.115     pazsan    399:              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);
1.85      pazsan    400:          }
                    401:        } else {
1.101     anton     402:           /* if base is > 0: 0 is a null reference so don't adjust*/
1.45      jwilke    403:           if (token>=base) {
                    404:             image[i]+=(Cell)start;
                    405:           }
1.46      jwilke    406:         }
1.1       anton     407:       }
                    408:     }
1.31      pazsan    409:   }
1.125     anton     410:   free(targets);
1.70      anton     411:   finish_code();
1.26      jwilke    412:   ((ImageHeader*)(image))->base = (Address) image;
1.1       anton     413: }
                    414: 
                    415: UCell checksum(Label symbols[])
                    416: {
                    417:   UCell r=PRIM_VERSION;
                    418:   Cell i;
                    419: 
                    420:   for (i=DOCOL; i<=DOESJUMP; i++) {
                    421:     r ^= (UCell)(symbols[i]);
                    422:     r = (r << 5) | (r >> (8*sizeof(Cell)-5));
                    423:   }
                    424: #ifdef DIRECT_THREADED
                    425:   /* we have to consider all the primitives */
                    426:   for (; symbols[i]!=(Label)0; i++) {
                    427:     r ^= (UCell)(symbols[i]);
                    428:     r = (r << 5) | (r >> (8*sizeof(Cell)-5));
                    429:   }
                    430: #else
                    431:   /* in indirect threaded code all primitives are accessed through the
                    432:      symbols table, so we just have to put the base address of symbols
                    433:      in the checksum */
                    434:   r ^= (UCell)symbols;
                    435: #endif
                    436:   return r;
                    437: }
                    438: 
1.3       anton     439: Address verbose_malloc(Cell size)
                    440: {
                    441:   Address r;
                    442:   /* leave a little room (64B) for stack underflows */
                    443:   if ((r = malloc(size+64))==NULL) {
                    444:     perror(progname);
                    445:     exit(1);
                    446:   }
                    447:   r = (Address)((((Cell)r)+(sizeof(Float)-1))&(-sizeof(Float)));
1.144   ! pazsan    448:   debugp(stderr, "malloc succeeds, address=$%lx\n", (long)r);
1.3       anton     449:   return r;
                    450: }
                    451: 
1.33      anton     452: static Address next_address=0;
                    453: void after_alloc(Address r, Cell size)
                    454: {
                    455:   if (r != (Address)-1) {
1.144   ! pazsan    456:     debugp(stderr, "success, address=$%lx\n", (long) r);
1.33      anton     457:     if (pagesize != 1)
                    458:       next_address = (Address)(((((Cell)r)+size-1)&-pagesize)+2*pagesize); /* leave one page unmapped */
                    459:   } else {
1.144   ! pazsan    460:     debugp(stderr, "failed: %s\n", strerror(errno));
1.33      anton     461:   }
                    462: }
                    463: 
1.34      anton     464: #ifndef MAP_FAILED
                    465: #define MAP_FAILED ((Address) -1)
                    466: #endif
                    467: #ifndef MAP_FILE
                    468: # define MAP_FILE 0
                    469: #endif
                    470: #ifndef MAP_PRIVATE
                    471: # define MAP_PRIVATE 0
                    472: #endif
1.91      anton     473: #if !defined(MAP_ANON) && defined(MAP_ANONYMOUS)
                    474: # define MAP_ANON MAP_ANONYMOUS
                    475: #endif
1.34      anton     476: 
                    477: #if defined(HAVE_MMAP)
                    478: static Address alloc_mmap(Cell size)
1.1       anton     479: {
                    480:   Address r;
                    481: 
                    482: #if defined(MAP_ANON)
1.144   ! pazsan    483:   debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_ANON, ...); ", (long)next_address, (long)size);
1.34      anton     484:   r = mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_ANON|MAP_PRIVATE, -1, 0);
1.1       anton     485: #else /* !defined(MAP_ANON) */
1.17      anton     486:   /* Ultrix (at least) does not define MAP_FILE and MAP_PRIVATE (both are
                    487:      apparently defaults) */
1.1       anton     488:   static int dev_zero=-1;
                    489: 
                    490:   if (dev_zero == -1)
                    491:     dev_zero = open("/dev/zero", O_RDONLY);
                    492:   if (dev_zero == -1) {
1.34      anton     493:     r = MAP_FAILED;
1.144   ! pazsan    494:     debugp(stderr, "open(\"/dev/zero\"...) failed (%s), no mmap; ", 
1.1       anton     495:              strerror(errno));
                    496:   } else {
1.144   ! pazsan    497:     debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FILE, dev_zero, ...); ", (long)next_address, (long)size);
1.1       anton     498:     r=mmap(next_address, size, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FILE|MAP_PRIVATE, dev_zero, 0);
                    499:   }
                    500: #endif /* !defined(MAP_ANON) */
1.34      anton     501:   after_alloc(r, size);
                    502:   return r;  
                    503: }
                    504: #endif
                    505: 
                    506: Address my_alloc(Cell size)
                    507: {
                    508: #if HAVE_MMAP
                    509:   Address r;
                    510: 
                    511:   r=alloc_mmap(size);
1.117     anton     512:   if (r!=(Address)MAP_FAILED)
1.1       anton     513:     return r;
                    514: #endif /* HAVE_MMAP */
1.3       anton     515:   /* use malloc as fallback */
                    516:   return verbose_malloc(size);
1.1       anton     517: }
                    518: 
1.34      anton     519: Address dict_alloc_read(FILE *file, Cell imagesize, Cell dictsize, Cell offset)
1.33      anton     520: {
1.34      anton     521:   Address image = MAP_FAILED;
1.33      anton     522: 
1.56      anton     523: #if defined(HAVE_MMAP)
1.33      anton     524:   if (offset==0) {
1.34      anton     525:     image=alloc_mmap(dictsize);
1.144   ! pazsan    526:     debugp(stderr,"try mmap($%lx, $%lx, ..., MAP_FIXED|MAP_FILE, imagefile, 0); ", (long)image, (long)imagesize);
1.34      anton     527:     image = mmap(image, imagesize, PROT_EXEC|PROT_READ|PROT_WRITE, MAP_FIXED|MAP_FILE|MAP_PRIVATE, fileno(file), 0);
                    528:     after_alloc(image,dictsize);
1.33      anton     529:   }
1.56      anton     530: #endif /* defined(HAVE_MMAP) */
1.117     anton     531:   if (image == (Address)MAP_FAILED) {
1.56      anton     532:     image = my_alloc(dictsize+offset)+offset;
1.33      anton     533:     rewind(file);  /* fseek(imagefile,0L,SEEK_SET); */
1.34      anton     534:     fread(image, 1, imagesize, file);
1.33      anton     535:   }
                    536:   return image;
                    537: }
                    538: 
1.10      pazsan    539: void set_stack_sizes(ImageHeader * header)
                    540: {
                    541:   if (dictsize==0)
                    542:     dictsize = header->dict_size;
                    543:   if (dsize==0)
                    544:     dsize = header->data_stack_size;
                    545:   if (rsize==0)
                    546:     rsize = header->return_stack_size;
                    547:   if (fsize==0)
                    548:     fsize = header->fp_stack_size;
                    549:   if (lsize==0)
                    550:     lsize = header->locals_stack_size;
                    551:   dictsize=maxaligned(dictsize);
                    552:   dsize=maxaligned(dsize);
                    553:   rsize=maxaligned(rsize);
                    554:   lsize=maxaligned(lsize);
                    555:   fsize=maxaligned(fsize);
                    556: }
                    557: 
                    558: void alloc_stacks(ImageHeader * header)
                    559: {
                    560:   header->dict_size=dictsize;
                    561:   header->data_stack_size=dsize;
                    562:   header->fp_stack_size=fsize;
                    563:   header->return_stack_size=rsize;
                    564:   header->locals_stack_size=lsize;
                    565: 
                    566:   header->data_stack_base=my_alloc(dsize);
                    567:   header->fp_stack_base=my_alloc(fsize);
                    568:   header->return_stack_base=my_alloc(rsize);
                    569:   header->locals_stack_base=my_alloc(lsize);
                    570: }
                    571: 
1.44      pazsan    572: #warning You can ignore the warnings about clobbered variables in go_forth
1.11      pazsan    573: int go_forth(Address image, int stack, Cell *entries)
                    574: {
1.38      anton     575:   volatile ImageHeader *image_header = (ImageHeader *)image;
1.18      anton     576:   Cell *sp0=(Cell*)(image_header->data_stack_base + dsize);
1.44      pazsan    577:   Cell *rp0=(Cell *)(image_header->return_stack_base + rsize);
1.18      anton     578:   Float *fp0=(Float *)(image_header->fp_stack_base + fsize);
1.44      pazsan    579: #ifdef GFORTH_DEBUGGING
1.38      anton     580:   volatile Cell *orig_rp0=rp0;
1.44      pazsan    581: #endif
1.18      anton     582:   Address lp0=image_header->locals_stack_base + lsize;
                    583:   Xt *ip0=(Xt *)(image_header->boot_entry);
1.13      pazsan    584: #ifdef SYSSIGNALS
1.11      pazsan    585:   int throw_code;
1.13      pazsan    586: #endif
1.11      pazsan    587: 
                    588:   /* ensure that the cached elements (if any) are accessible */
1.41      anton     589:   IF_spTOS(sp0--);
                    590:   IF_fpTOS(fp0--);
1.11      pazsan    591:   
                    592:   for(;stack>0;stack--)
1.18      anton     593:     *--sp0=entries[stack-1];
1.11      pazsan    594: 
1.30      pazsan    595: #ifdef SYSSIGNALS
1.11      pazsan    596:   get_winsize();
                    597:    
                    598:   install_signal_handlers(); /* right place? */
                    599:   
                    600:   if ((throw_code=setjmp(throw_jmp_buf))) {
                    601:     static Cell signal_data_stack[8];
                    602:     static Cell signal_return_stack[8];
                    603:     static Float signal_fp_stack[1];
1.13      pazsan    604: 
1.11      pazsan    605:     signal_data_stack[7]=throw_code;
1.18      anton     606: 
                    607: #ifdef GFORTH_DEBUGGING
1.144   ! pazsan    608:     debugp(stderr,"\ncaught signal, throwing exception %d, ip=%p rp=%p\n",
1.97      anton     609:              throw_code, saved_ip, rp);
1.38      anton     610:     if (rp <= orig_rp0 && rp > (Cell *)(image_header->return_stack_base+5)) {
1.18      anton     611:       /* no rstack overflow or underflow */
                    612:       rp0 = rp;
1.63      anton     613:       *--rp0 = (Cell)saved_ip;
1.18      anton     614:     }
                    615:     else /* I love non-syntactic ifdefs :-) */
1.97      anton     616:       rp0 = signal_return_stack+8;
                    617: #else  /* !defined(GFORTH_DEBUGGING) */
1.144   ! pazsan    618:     debugp(stderr,"\ncaught signal, throwing exception %d\n", throw_code);
1.97      anton     619:       rp0 = signal_return_stack+8;
                    620: #endif /* !defined(GFORTH_DEBUGGING) */
1.25      anton     621:     /* fprintf(stderr, "rp=$%x\n",rp0);*/
1.11      pazsan    622:     
1.33      anton     623:     return((int)(Cell)engine(image_header->throw_entry, signal_data_stack+7,
1.18      anton     624:                       rp0, signal_fp_stack, 0));
1.11      pazsan    625:   }
1.13      pazsan    626: #endif
1.11      pazsan    627: 
1.33      anton     628:   return((int)(Cell)engine(ip0,sp0,rp0,fp0,lp0));
1.11      pazsan    629: }
                    630: 
1.30      pazsan    631: #ifndef INCLUDE_IMAGE
1.21      anton     632: void print_sizes(Cell sizebyte)
                    633:      /* print size information */
                    634: {
                    635:   static char* endianstring[]= { "   big","little" };
                    636:   
                    637:   fprintf(stderr,"%s endian, cell=%d bytes, char=%d bytes, au=%d bytes\n",
                    638:          endianstring[sizebyte & 1],
                    639:          1 << ((sizebyte >> 1) & 3),
                    640:          1 << ((sizebyte >> 3) & 3),
                    641:          1 << ((sizebyte >> 5) & 3));
                    642: }
                    643: 
1.106     anton     644: /* static superinstruction stuff */
                    645: 
1.141     anton     646: struct cost { /* super_info might be a more accurate name */
1.106     anton     647:   char loads;       /* number of stack loads */
                    648:   char stores;      /* number of stack stores */
                    649:   char updates;     /* number of stack pointer updates */
1.123     anton     650:   char branch;     /* is it a branch (SET_IP) */
1.125     anton     651:   unsigned char state_in;    /* state on entry */
                    652:   unsigned char state_out;   /* state on exit */
1.142     anton     653:   unsigned char imm_ops;     /* number of immediate operands */
1.123     anton     654:   short offset;     /* offset into super2 table */
1.125     anton     655:   unsigned char length;      /* number of components */
1.106     anton     656: };
                    657: 
1.121     anton     658: PrimNum super2[] = {
1.126     anton     659: #include SUPER2_I
1.106     anton     660: };
                    661: 
                    662: struct cost super_costs[] = {
1.126     anton     663: #include COSTS_I
1.106     anton     664: };
                    665: 
1.125     anton     666: struct super_state {
                    667:   struct super_state *next;
                    668:   PrimNum super;
                    669: };
                    670: 
1.106     anton     671: #define HASH_SIZE 256
                    672: 
                    673: struct super_table_entry {
                    674:   struct super_table_entry *next;
1.121     anton     675:   PrimNum *start;
1.106     anton     676:   short length;
1.125     anton     677:   struct super_state *ss_list; /* list of supers */
1.106     anton     678: } *super_table[HASH_SIZE];
                    679: int max_super=2;
                    680: 
1.125     anton     681: struct super_state *state_transitions=NULL;
                    682: 
1.121     anton     683: int hash_super(PrimNum *start, int length)
1.106     anton     684: {
                    685:   int i, r;
                    686:   
                    687:   for (i=0, r=0; i<length; i++) {
                    688:     r <<= 1;
                    689:     r += start[i];
                    690:   }
                    691:   return r & (HASH_SIZE-1);
                    692: }
                    693: 
1.125     anton     694: struct super_state **lookup_super(PrimNum *start, int length)
1.106     anton     695: {
                    696:   int hash=hash_super(start,length);
                    697:   struct super_table_entry *p = super_table[hash];
                    698: 
1.125     anton     699:   /* assert(length >= 2); */
1.106     anton     700:   for (; p!=NULL; p = p->next) {
                    701:     if (length == p->length &&
1.121     anton     702:        memcmp((char *)p->start, (char *)start, length*sizeof(PrimNum))==0)
1.125     anton     703:       return &(p->ss_list);
1.106     anton     704:   }
1.125     anton     705:   return NULL;
1.106     anton     706: }
                    707: 
                    708: void prepare_super_table()
                    709: {
                    710:   int i;
1.109     anton     711:   int nsupers = 0;
1.106     anton     712: 
                    713:   for (i=0; i<sizeof(super_costs)/sizeof(super_costs[0]); i++) {
                    714:     struct cost *c = &super_costs[i];
1.125     anton     715:     if ((c->length < 2 || nsupers < static_super_number) &&
                    716:        c->state_in < maxstates && c->state_out < maxstates) {
                    717:       struct super_state **ss_listp= lookup_super(super2+c->offset, c->length);
                    718:       struct super_state *ss = malloc(sizeof(struct super_state));
                    719:       ss->super= i;
                    720:       if (c->offset==N_noop && i != N_noop) {
                    721:        if (is_relocatable(i)) {
                    722:          ss->next = state_transitions;
                    723:          state_transitions = ss;
                    724:        }
                    725:       } else if (ss_listp != NULL) {
                    726:        ss->next = *ss_listp;
                    727:        *ss_listp = ss;
                    728:       } else {
                    729:        int hash = hash_super(super2+c->offset, c->length);
                    730:        struct super_table_entry **p = &super_table[hash];
                    731:        struct super_table_entry *e = malloc(sizeof(struct super_table_entry));
                    732:        ss->next = NULL;
                    733:        e->next = *p;
                    734:        e->start = super2 + c->offset;
                    735:        e->length = c->length;
                    736:        e->ss_list = ss;
                    737:        *p = e;
                    738:       }
1.106     anton     739:       if (c->length > max_super)
                    740:        max_super = c->length;
1.125     anton     741:       if (c->length >= 2)
                    742:        nsupers++;
1.106     anton     743:     }
                    744:   }
1.144   ! pazsan    745:   debugp(stderr, "Using %d static superinsts\n", nsupers);
1.106     anton     746: }
                    747: 
                    748: /* dynamic replication/superinstruction stuff */
                    749: 
1.69      anton     750: #ifndef NO_DYNAMIC
1.90      anton     751: int compare_priminfo_length(const void *_a, const void *_b)
1.76      anton     752: {
1.90      anton     753:   PrimInfo **a = (PrimInfo **)_a;
                    754:   PrimInfo **b = (PrimInfo **)_b;
1.77      anton     755:   Cell diff = (*a)->length - (*b)->length;
                    756:   if (diff)
                    757:     return diff;
                    758:   else /* break ties by start address; thus the decompiler produces
                    759:           the earliest primitive with the same code (e.g. noop instead
                    760:           of (char) and @ instead of >code-address */
                    761:     return (*b)->start - (*a)->start;
1.76      anton     762: }
1.112     anton     763: #endif /* !defined(NO_DYNAMIC) */
1.76      anton     764: 
1.125     anton     765: static char MAYBE_UNUSED superend[]={
1.126     anton     766: #include PRIM_SUPEREND_I
1.106     anton     767: };
1.107     anton     768: 
                    769: Cell npriminfos=0;
1.76      anton     770: 
1.114     anton     771: int compare_labels(const void *pa, const void *pb)
1.113     anton     772: {
1.114     anton     773:   Label a = *(Label *)pa;
                    774:   Label b = *(Label *)pb;
                    775:   return a-b;
                    776: }
1.113     anton     777: 
1.114     anton     778: Label bsearch_next(Label key, Label *a, UCell n)
                    779:      /* a is sorted; return the label >=key that is the closest in a;
                    780:         return NULL if there is no label in a >=key */
                    781: {
                    782:   int mid = (n-1)/2;
                    783:   if (n<1)
                    784:     return NULL;
                    785:   if (n == 1) {
                    786:     if (a[0] < key)
                    787:       return NULL;
                    788:     else
                    789:       return a[0];
                    790:   }
                    791:   if (a[mid] < key)
                    792:     return bsearch_next(key, a+mid+1, n-mid-1);
                    793:   else
                    794:     return bsearch_next(key, a, mid+1);
1.113     anton     795: }
                    796: 
1.47      anton     797: void check_prims(Label symbols1[])
                    798: {
                    799:   int i;
1.90      anton     800: #ifndef NO_DYNAMIC
1.119     anton     801:   Label *symbols2, *symbols3, *ends1, *ends1j, *ends1jsorted;
                    802:   int nends1j;
1.90      anton     803: #endif
1.47      anton     804: 
1.66      anton     805:   if (debug)
                    806: #ifdef __VERSION__
                    807:     fprintf(stderr, "Compiled with gcc-" __VERSION__ "\n");
                    808: #else
                    809: #define xstr(s) str(s)
                    810: #define str(s) #s
                    811:   fprintf(stderr, "Compiled with gcc-" xstr(__GNUC__) "." xstr(__GNUC_MINOR__) "\n"); 
                    812: #endif
1.121     anton     813:   for (i=0; symbols1[i]!=0; i++)
1.47      anton     814:     ;
1.55      anton     815:   npriminfos = i;
1.70      anton     816:   
                    817: #ifndef NO_DYNAMIC
1.66      anton     818:   if (no_dynamic)
                    819:     return;
1.55      anton     820:   symbols2=engine2(0,0,0,0,0);
1.70      anton     821: #if NO_IP
                    822:   symbols3=engine3(0,0,0,0,0);
                    823: #else
                    824:   symbols3=symbols1;
                    825: #endif
1.121     anton     826:   ends1 = symbols1+i+1;
1.119     anton     827:   ends1j =   ends1+i;
1.121     anton     828:   nends1j = i+1;
1.119     anton     829:   ends1jsorted = (Label *)alloca(nends1j*sizeof(Label));
                    830:   memcpy(ends1jsorted,ends1j,nends1j*sizeof(Label));
                    831:   qsort(ends1jsorted, nends1j, sizeof(Label), compare_labels);
1.113     anton     832:   
1.47      anton     833:   priminfos = calloc(i,sizeof(PrimInfo));
1.121     anton     834:   for (i=0; symbols1[i]!=0; i++) {
1.70      anton     835:     int prim_len = ends1[i]-symbols1[i];
1.47      anton     836:     PrimInfo *pi=&priminfos[i];
1.70      anton     837:     int j=0;
                    838:     char *s1 = (char *)symbols1[i];
                    839:     char *s2 = (char *)symbols2[i];
                    840:     char *s3 = (char *)symbols3[i];
1.119     anton     841:     Label endlabel = bsearch_next(symbols1[i]+1,ends1jsorted,nends1j);
1.70      anton     842: 
                    843:     pi->start = s1;
1.121     anton     844:     pi->superend = superend[i]|no_super;
1.70      anton     845:     if (pi->superend)
1.113     anton     846:       pi->length = endlabel-symbols1[i];
1.70      anton     847:     else
                    848:       pi->length = prim_len;
1.113     anton     849:     pi->restlength = endlabel - symbols1[i] - pi->length;
1.70      anton     850:     pi->nimmargs = 0;
1.144   ! pazsan    851:     relocs++;
        !           852:     debugp(stderr, "%-15s %3d %p %p %p len=%3ld restlen=%2ld s-end=%1d",
1.139     anton     853:              prim_names[i], i, s1, s2, s3, (long)(pi->length), (long)(pi->restlength), pi->superend);
1.114     anton     854:     if (endlabel == NULL) {
                    855:       pi->start = NULL; /* not relocatable */
1.122     anton     856:       if (pi->length<0) pi->length=100;
1.144   ! pazsan    857:       debugp(stderr,"\n   non_reloc: no J label > start found\n");
        !           858:       relocs--;
        !           859:       nonrelocs++;
1.114     anton     860:       continue;
                    861:     }
                    862:     if (ends1[i] > endlabel && !pi->superend) {
1.113     anton     863:       pi->start = NULL; /* not relocatable */
1.122     anton     864:       pi->length = endlabel-symbols1[i];
1.144   ! pazsan    865:       debugp(stderr,"\n   non_reloc: there is a J label before the K label (restlength<0)\n");
        !           866:       relocs--;
        !           867:       nonrelocs++;
1.113     anton     868:       continue;
                    869:     }
1.114     anton     870:     if (ends1[i] < pi->start && !pi->superend) {
1.113     anton     871:       pi->start = NULL; /* not relocatable */
1.122     anton     872:       pi->length = endlabel-symbols1[i];
1.144   ! pazsan    873:       debugp(stderr,"\n   non_reloc: K label before I label (length<0)\n");
        !           874:       relocs--;
        !           875:       nonrelocs++;
1.113     anton     876:       continue;
                    877:     }
1.138     anton     878:     assert(pi->length>=0);
1.113     anton     879:     assert(pi->restlength >=0);
1.74      anton     880:     while (j<(pi->length+pi->restlength)) {
1.70      anton     881:       if (s1[j]==s3[j]) {
                    882:        if (s1[j] != s2[j]) {
                    883:          pi->start = NULL; /* not relocatable */
1.144   ! pazsan    884:          debugp(stderr,"\n   non_reloc: engine1!=engine2 offset %3d",j);
1.74      anton     885:          /* assert(j<prim_len); */
1.144   ! pazsan    886:          relocs--;
        !           887:          nonrelocs++;
1.70      anton     888:          break;
                    889:        }
                    890:        j++;
                    891:       } else {
                    892:        struct immarg *ia=&pi->immargs[pi->nimmargs];
                    893: 
                    894:        pi->nimmargs++;
                    895:        ia->offset=j;
                    896:        if ((~*(Cell *)&(s1[j]))==*(Cell *)&(s3[j])) {
                    897:          ia->rel=0;
1.144   ! pazsan    898:          debugp(stderr,"\n   absolute immarg: offset %3d",j);
1.70      anton     899:        } else if ((&(s1[j]))+(*(Cell *)&(s1[j]))+4 ==
                    900:                   symbols1[DOESJUMP+1]) {
                    901:          ia->rel=1;
1.144   ! pazsan    902:          debugp(stderr,"\n   relative immarg: offset %3d",j);
1.70      anton     903:        } else {
                    904:          pi->start = NULL; /* not relocatable */
1.144   ! pazsan    905:          debugp(stderr,"\n   non_reloc: engine1!=engine3 offset %3d",j);
1.74      anton     906:          /* assert(j<prim_len);*/
1.144   ! pazsan    907:          relocs--;
        !           908:          nonrelocs++;
1.70      anton     909:          break;
                    910:        }
                    911:        j+=4;
1.47      anton     912:       }
                    913:     }
1.144   ! pazsan    914:     debugp(stderr,"\n");
1.70      anton     915:   }
1.76      anton     916:   decomp_prims = calloc(i,sizeof(PrimInfo *));
                    917:   for (i=DOESJUMP+1; i<npriminfos; i++)
                    918:     decomp_prims[i] = &(priminfos[i]);
                    919:   qsort(decomp_prims+DOESJUMP+1, npriminfos-DOESJUMP-1, sizeof(PrimInfo *),
                    920:        compare_priminfo_length);
1.70      anton     921: #endif
                    922: }
                    923: 
1.74      anton     924: void flush_to_here(void)
                    925: {
1.93      anton     926: #ifndef NO_DYNAMIC
1.100     anton     927:   if (start_flush)
                    928:     FLUSH_ICACHE(start_flush, code_here-start_flush);
1.74      anton     929:   start_flush=code_here;
1.93      anton     930: #endif
1.74      anton     931: }
                    932: 
1.93      anton     933: #ifndef NO_DYNAMIC
1.74      anton     934: void append_jump(void)
                    935: {
                    936:   if (last_jump) {
                    937:     PrimInfo *pi = &priminfos[last_jump];
                    938:     
                    939:     memcpy(code_here, pi->start+pi->length, pi->restlength);
                    940:     code_here += pi->restlength;
                    941:     last_jump=0;
                    942:   }
                    943: }
                    944: 
1.75      anton     945: /* Gforth remembers all code blocks in this list.  On forgetting (by
                    946: executing a marker) the code blocks are not freed (because Gforth does
                    947: not remember how they were allocated; hmm, remembering that might be
                    948: easier and cleaner).  Instead, code_here etc. are reset to the old
                    949: value, and the "forgotten" code blocks are reused when they are
                    950: needed. */
                    951: 
                    952: struct code_block_list {
                    953:   struct code_block_list *next;
                    954:   Address block;
                    955:   Cell size;
                    956: } *code_block_list=NULL, **next_code_blockp=&code_block_list;
                    957: 
1.74      anton     958: Address append_prim(Cell p)
                    959: {
                    960:   PrimInfo *pi = &priminfos[p];
                    961:   Address old_code_here = code_here;
                    962: 
                    963:   if (code_area+code_area_size < code_here+pi->length+pi->restlength) {
1.75      anton     964:     struct code_block_list *p;
1.74      anton     965:     append_jump();
1.93      anton     966:     flush_to_here();
1.75      anton     967:     if (*next_code_blockp == NULL) {
                    968:       code_here = start_flush = code_area = my_alloc(code_area_size);
                    969:       p = (struct code_block_list *)malloc(sizeof(struct code_block_list));
                    970:       *next_code_blockp = p;
                    971:       p->next = NULL;
                    972:       p->block = code_here;
                    973:       p->size = code_area_size;
                    974:     } else {
                    975:       p = *next_code_blockp;
                    976:       code_here = start_flush = code_area = p->block;
                    977:     }
1.74      anton     978:     old_code_here = code_here;
1.75      anton     979:     next_code_blockp = &(p->next);
1.74      anton     980:   }
                    981:   memcpy(code_here, pi->start, pi->length);
                    982:   code_here += pi->length;
                    983:   return old_code_here;
                    984: }
                    985: #endif
1.75      anton     986: 
                    987: int forget_dyncode(Address code)
                    988: {
                    989: #ifdef NO_DYNAMIC
                    990:   return -1;
                    991: #else
                    992:   struct code_block_list *p, **pp;
                    993: 
                    994:   for (pp=&code_block_list, p=*pp; p!=NULL; pp=&(p->next), p=*pp) {
                    995:     if (code >= p->block && code < p->block+p->size) {
                    996:       next_code_blockp = &(p->next);
                    997:       code_here = start_flush = code;
                    998:       code_area = p->block;
                    999:       last_jump = 0;
                   1000:       return -1;
                   1001:     }
                   1002:   }
1.78      anton    1003:   return -no_dynamic;
1.75      anton    1004: #endif /* !defined(NO_DYNAMIC) */
                   1005: }
                   1006: 
1.104     anton    1007: long dyncodesize(void)
                   1008: {
                   1009: #ifndef NO_DYNAMIC
1.106     anton    1010:   struct code_block_list *p;
1.104     anton    1011:   long size=0;
                   1012:   for (p=code_block_list; p!=NULL; p=p->next) {
                   1013:     if (code_here >= p->block && code_here < p->block+p->size)
                   1014:       return size + (code_here - p->block);
                   1015:     else
                   1016:       size += p->size;
                   1017:   }
                   1018: #endif /* !defined(NO_DYNAMIC) */
                   1019:   return 0;
                   1020: }
                   1021: 
1.90      anton    1022: Label decompile_code(Label _code)
1.75      anton    1023: {
1.76      anton    1024: #ifdef NO_DYNAMIC
1.90      anton    1025:   return _code;
1.76      anton    1026: #else /* !defined(NO_DYNAMIC) */
                   1027:   Cell i;
1.77      anton    1028:   struct code_block_list *p;
1.90      anton    1029:   Address code=_code;
1.76      anton    1030: 
1.77      anton    1031:   /* first, check if we are in code at all */
                   1032:   for (p = code_block_list;; p = p->next) {
                   1033:     if (p == NULL)
                   1034:       return code;
                   1035:     if (code >= p->block && code < p->block+p->size)
                   1036:       break;
                   1037:   }
1.76      anton    1038:   /* reverse order because NOOP might match other prims */
                   1039:   for (i=npriminfos-1; i>DOESJUMP; i--) {
                   1040:     PrimInfo *pi=decomp_prims[i];
                   1041:     if (pi->start==code || (pi->start && memcmp(code,pi->start,pi->length)==0))
1.121     anton    1042:       return vm_prims[super2[super_costs[pi-priminfos].offset]];
1.118     anton    1043:     /* return pi->start;*/
1.76      anton    1044:   }
                   1045:   return code;
                   1046: #endif /* !defined(NO_DYNAMIC) */
1.75      anton    1047: }
1.74      anton    1048: 
1.70      anton    1049: #ifdef NO_IP
                   1050: int nbranchinfos=0;
                   1051: 
                   1052: struct branchinfo {
1.136     anton    1053:   Label **targetpp; /* **(bi->targetpp) is the target */
1.70      anton    1054:   Cell *addressptr; /* store the target here */
                   1055: } branchinfos[100000];
                   1056: 
                   1057: int ndoesexecinfos=0;
                   1058: struct doesexecinfo {
                   1059:   int branchinfo; /* fix the targetptr of branchinfos[...->branchinfo] */
1.136     anton    1060:   Label *targetp; /*target for branch (because this is not in threaded code)*/
1.70      anton    1061:   Cell *xt; /* cfa of word whose does-code needs calling */
                   1062: } doesexecinfos[10000];
                   1063: 
                   1064: void set_rel_target(Cell *source, Label target)
                   1065: {
                   1066:   *source = ((Cell)target)-(((Cell)source)+4);
                   1067: }
                   1068: 
1.136     anton    1069: void register_branchinfo(Label source, Cell *targetpp)
1.70      anton    1070: {
                   1071:   struct branchinfo *bi = &(branchinfos[nbranchinfos]);
1.136     anton    1072:   bi->targetpp = (Label **)targetpp;
1.70      anton    1073:   bi->addressptr = (Cell *)source;
                   1074:   nbranchinfos++;
                   1075: }
                   1076: 
1.133     anton    1077: Address compile_prim1arg(PrimNum p, Cell **argp)
1.70      anton    1078: {
1.133     anton    1079:   Address old_code_here=append_prim(p);
1.70      anton    1080: 
1.74      anton    1081:   assert(vm_prims[p]==priminfos[p].start);
1.133     anton    1082:   *argp = (Cell*)(old_code_here+priminfos[p].immargs[0].offset);
                   1083:   return old_code_here;
1.70      anton    1084: }
                   1085: 
1.136     anton    1086: Address compile_call2(Cell *targetpp, Cell **next_code_targetp)
1.70      anton    1087: {
1.73      anton    1088:   PrimInfo *pi = &priminfos[N_call2];
1.74      anton    1089:   Address old_code_here = append_prim(N_call2);
1.70      anton    1090: 
1.134     anton    1091:   *next_code_targetp = (Cell *)(old_code_here + pi->immargs[0].offset);
1.136     anton    1092:   register_branchinfo(old_code_here + pi->immargs[1].offset, targetpp);
1.134     anton    1093:   return old_code_here;
1.70      anton    1094: }
                   1095: #endif
                   1096: 
                   1097: void finish_code(void)
                   1098: {
                   1099: #ifdef NO_IP
                   1100:   Cell i;
                   1101: 
                   1102:   compile_prim1(NULL);
                   1103:   for (i=0; i<ndoesexecinfos; i++) {
                   1104:     struct doesexecinfo *dei = &doesexecinfos[i];
1.136     anton    1105:     dei->targetp = (Label *)DOES_CODE1((dei->xt));
                   1106:     branchinfos[dei->branchinfo].targetpp = &(dei->targetp);
1.70      anton    1107:   }
                   1108:   ndoesexecinfos = 0;
                   1109:   for (i=0; i<nbranchinfos; i++) {
                   1110:     struct branchinfo *bi=&branchinfos[i];
1.136     anton    1111:     set_rel_target(bi->addressptr, **(bi->targetpp));
1.70      anton    1112:   }
                   1113:   nbranchinfos = 0;
1.128     anton    1114: #else
                   1115:   compile_prim1(NULL);
1.48      anton    1116: #endif
1.93      anton    1117:   flush_to_here();
1.48      anton    1118: }
                   1119: 
1.128     anton    1120: #ifdef NO_IP
                   1121: Cell compile_prim_dyn(PrimNum p, Cell *tcp)
                   1122:      /* compile prim #p dynamically (mod flags etc.) and return start
                   1123:        address of generated code for putting it into the threaded
                   1124:        code. This function is only called if all the associated
                   1125:        inline arguments of p are already in place (at tcp[1] etc.) */
                   1126: {
                   1127:   PrimInfo *pi=&priminfos[p];
                   1128:   Cell *next_code_target=NULL;
1.135     anton    1129:   Address codeaddr;
                   1130:   Address primstart;
1.128     anton    1131:   
                   1132:   assert(p<npriminfos);
                   1133:   if (p==N_execute || p==N_perform || p==N_lit_perform) {
1.134     anton    1134:     codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
1.135     anton    1135:     primstart = append_prim(p);
                   1136:     goto other_prim;
                   1137:   } else if (p==N_call) {
1.136     anton    1138:     codeaddr = compile_call2(tcp+1, &next_code_target);
1.128     anton    1139:   } else if (p==N_does_exec) {
                   1140:     struct doesexecinfo *dei = &doesexecinfos[ndoesexecinfos++];
1.133     anton    1141:     Cell *arg;
                   1142:     codeaddr = compile_prim1arg(N_lit,&arg);
                   1143:     *arg = (Cell)PFA(tcp[1]);
1.128     anton    1144:     /* we cannot determine the callee now (last_start[1] may be a
                   1145:        forward reference), so just register an arbitrary target, and
                   1146:        register in dei that we need to fix this before resolving
                   1147:        branches */
                   1148:     dei->branchinfo = nbranchinfos;
                   1149:     dei->xt = (Cell *)(tcp[1]);
1.134     anton    1150:     compile_call2(0, &next_code_target);
1.128     anton    1151:   } else if (!is_relocatable(p)) {
1.133     anton    1152:     Cell *branch_target;
                   1153:     codeaddr = compile_prim1arg(N_set_next_code, &next_code_target);
                   1154:     compile_prim1arg(N_branch,&branch_target);
                   1155:     set_rel_target(branch_target,vm_prims[p]);
1.128     anton    1156:   } else {
                   1157:     unsigned j;
1.135     anton    1158: 
                   1159:     codeaddr = primstart = append_prim(p);
                   1160:   other_prim:
1.128     anton    1161:     for (j=0; j<pi->nimmargs; j++) {
                   1162:       struct immarg *ia = &(pi->immargs[j]);
1.136     anton    1163:       Cell *argp = tcp + pi->nimmargs - j;
                   1164:       Cell argval = *argp; /* !! specific to prims */
1.128     anton    1165:       if (ia->rel) { /* !! assumption: relative refs are branches */
1.136     anton    1166:        register_branchinfo(primstart + ia->offset, argp);
1.128     anton    1167:       } else /* plain argument */
1.135     anton    1168:        *(Cell *)(primstart + ia->offset) = argval;
1.128     anton    1169:     }
                   1170:   }
                   1171:   if (next_code_target!=NULL)
                   1172:     *next_code_target = (Cell)code_here;
1.135     anton    1173:   return (Cell)codeaddr;
1.128     anton    1174: }
                   1175: #else /* !defined(NO_IP) */
                   1176: Cell compile_prim_dyn(PrimNum p, Cell *tcp)
                   1177:      /* compile prim #p dynamically (mod flags etc.) and return start
                   1178:         address of generated code for putting it into the threaded code */
1.108     anton    1179: {
1.121     anton    1180:   Cell static_prim = (Cell)vm_prims[p];
1.108     anton    1181: #if defined(NO_DYNAMIC)
                   1182:   return static_prim;
                   1183: #else /* !defined(NO_DYNAMIC) */
                   1184:   Address old_code_here;
                   1185: 
                   1186:   if (no_dynamic)
                   1187:     return static_prim;
1.125     anton    1188:   if (p>=npriminfos || !is_relocatable(p)) {
1.108     anton    1189:     append_jump();
                   1190:     return static_prim;
                   1191:   }
                   1192:   old_code_here = append_prim(p);
                   1193:   last_jump = (priminfos[p].superend) ? 0 : p;
                   1194:   return (Cell)old_code_here;
                   1195: #endif  /* !defined(NO_DYNAMIC) */
                   1196: }
1.128     anton    1197: #endif /* !defined(NO_IP) */
1.70      anton    1198: 
1.109     anton    1199: #ifndef NO_DYNAMIC
                   1200: int cost_codesize(int prim)
                   1201: {
1.121     anton    1202:   return priminfos[prim].length;
1.109     anton    1203: }
                   1204: #endif
                   1205: 
                   1206: int cost_ls(int prim)
                   1207: {
                   1208:   struct cost *c = super_costs+prim;
                   1209: 
                   1210:   return c->loads + c->stores;
                   1211: }
                   1212: 
                   1213: int cost_lsu(int prim)
                   1214: {
                   1215:   struct cost *c = super_costs+prim;
                   1216: 
                   1217:   return c->loads + c->stores + c->updates;
                   1218: }
                   1219: 
                   1220: int cost_nexts(int prim)
                   1221: {
                   1222:   return 1;
                   1223: }
                   1224: 
                   1225: typedef int Costfunc(int);
                   1226: Costfunc *ss_cost =  /* cost function for optimize_bb */
                   1227: #ifdef NO_DYNAMIC
                   1228: cost_lsu;
                   1229: #else
                   1230: cost_codesize;
                   1231: #endif
                   1232: 
1.110     anton    1233: struct {
                   1234:   Costfunc *costfunc;
                   1235:   char *metricname;
                   1236:   long sum;
                   1237: } cost_sums[] = {
                   1238: #ifndef NO_DYNAMIC
                   1239:   { cost_codesize, "codesize", 0 },
                   1240: #endif
                   1241:   { cost_ls,       "ls",       0 },
                   1242:   { cost_lsu,      "lsu",      0 },
                   1243:   { cost_nexts,    "nexts",    0 }
                   1244: };
                   1245: 
1.106     anton    1246: #define MAX_BB 128 /* maximum number of instructions in BB */
1.125     anton    1247: #define INF_COST 1000000 /* infinite cost */
                   1248: #define CANONICAL_STATE 0
                   1249: 
                   1250: struct waypoint {
                   1251:   int cost;     /* the cost from here to the end */
                   1252:   PrimNum inst; /* the inst used from here to the next waypoint */
                   1253:   char relocatable; /* the last non-transition was relocatable */
                   1254:   char no_transition; /* don't use the next transition (relocatability)
                   1255:                       * or this transition (does not change state) */
                   1256: };
                   1257: 
                   1258: void init_waypoints(struct waypoint ws[])
                   1259: {
                   1260:   int k;
                   1261: 
                   1262:   for (k=0; k<maxstates; k++)
                   1263:     ws[k].cost=INF_COST;
                   1264: }
1.106     anton    1265: 
1.125     anton    1266: void transitions(struct waypoint inst[], struct waypoint trans[])
1.107     anton    1267: {
1.125     anton    1268:   int k;
                   1269:   struct super_state *l;
                   1270:   
                   1271:   for (k=0; k<maxstates; k++) {
                   1272:     trans[k] = inst[k];
                   1273:     trans[k].no_transition = 1;
                   1274:   }
                   1275:   for (l = state_transitions; l != NULL; l = l->next) {
                   1276:     PrimNum s = l->super;
                   1277:     int jcost;
                   1278:     struct cost *c=super_costs+s;
                   1279:     struct waypoint *wi=&(trans[c->state_in]);
                   1280:     struct waypoint *wo=&(inst[c->state_out]);
                   1281:     if (wo->cost == INF_COST)
                   1282:       continue;
                   1283:     jcost = wo->cost + ss_cost(s);
                   1284:     if (jcost <= wi->cost) {
                   1285:       wi->cost = jcost;
                   1286:       wi->inst = s;
                   1287:       wi->relocatable = wo->relocatable;
                   1288:       wi->no_transition = 0;
                   1289:       /* if (ss_greedy) wi->cost = wo->cost ? */
                   1290:     }
                   1291:   }
                   1292: }
1.107     anton    1293: 
1.125     anton    1294: /* use dynamic programming to find the shortest paths within the basic
                   1295:    block origs[0..ninsts-1] and rewrite the instructions pointed to by
                   1296:    instps to use it */
                   1297: void optimize_rewrite(Cell *instps[], PrimNum origs[], int ninsts)
                   1298: {
                   1299:   int i,j;
                   1300:   static struct waypoint inst[MAX_BB+1][MAX_STATE];  /* before instruction*/
                   1301:   static struct waypoint trans[MAX_BB+1][MAX_STATE]; /* before transition */
                   1302:   int nextdyn, nextstate, no_transition;
                   1303:   
                   1304:   init_waypoints(inst[ninsts]);
                   1305:   inst[ninsts][CANONICAL_STATE].cost=0;
                   1306:   transitions(inst[ninsts],trans[ninsts]);
1.107     anton    1307:   for (i=ninsts-1; i>=0; i--) {
1.125     anton    1308:     init_waypoints(inst[i]);
                   1309:     for (j=1; j<=max_super && i+j<=ninsts; j++) {
                   1310:       struct super_state **superp = lookup_super(origs+i, j);
                   1311:       if (superp!=NULL) {
                   1312:        struct super_state *supers = *superp;
                   1313:        for (; supers!=NULL; supers = supers->next) {
                   1314:          PrimNum s = supers->super;
                   1315:          int jcost;
                   1316:          struct cost *c=super_costs+s;
                   1317:          struct waypoint *wi=&(inst[i][c->state_in]);
                   1318:          struct waypoint *wo=&(trans[i+j][c->state_out]);
                   1319:          int no_transition = wo->no_transition;
                   1320:          if (!(is_relocatable(s)) && !wo->relocatable) {
                   1321:            wo=&(inst[i+j][c->state_out]);
                   1322:            no_transition=1;
                   1323:          }
                   1324:          if (wo->cost == INF_COST) 
                   1325:            continue;
                   1326:          jcost = wo->cost + ss_cost(s);
                   1327:          if (jcost <= wi->cost) {
                   1328:            wi->cost = jcost;
                   1329:            wi->inst = s;
                   1330:            wi->relocatable = is_relocatable(s);
                   1331:            wi->no_transition = no_transition;
                   1332:            /* if (ss_greedy) wi->cost = wo->cost ? */
                   1333:          }
1.107     anton    1334:        }
                   1335:       }
                   1336:     }
1.125     anton    1337:     transitions(inst[i],trans[i]);
                   1338:   }
                   1339:   /* now rewrite the instructions */
                   1340:   nextdyn=0;
                   1341:   nextstate=CANONICAL_STATE;
                   1342:   no_transition = ((!trans[0][nextstate].relocatable) 
                   1343:                   ||trans[0][nextstate].no_transition);
                   1344:   for (i=0; i<ninsts; i++) {
                   1345:     Cell tc=0, tc2;
                   1346:     if (i==nextdyn) {
                   1347:       if (!no_transition) {
                   1348:        /* process trans */
                   1349:        PrimNum p = trans[i][nextstate].inst;
                   1350:        struct cost *c = super_costs+p;
                   1351:        assert(trans[i][nextstate].cost != INF_COST);
                   1352:        assert(c->state_in==nextstate);
1.128     anton    1353:        tc = compile_prim_dyn(p,NULL);
1.125     anton    1354:        nextstate = c->state_out;
                   1355:       }
                   1356:       {
                   1357:        /* process inst */
                   1358:        PrimNum p = inst[i][nextstate].inst;
                   1359:        struct cost *c=super_costs+p;
                   1360:        assert(c->state_in==nextstate);
                   1361:        assert(inst[i][nextstate].cost != INF_COST);
                   1362: #if defined(GFORTH_DEBUGGING)
                   1363:        assert(p == origs[i]);
                   1364: #endif
1.128     anton    1365:        tc2 = compile_prim_dyn(p,instps[i]);
1.125     anton    1366:        if (no_transition || !is_relocatable(p))
                   1367:          /* !! actually what we care about is if and where
                   1368:           * compile_prim_dyn() puts NEXTs */
                   1369:          tc=tc2;
                   1370:        no_transition = inst[i][nextstate].no_transition;
                   1371:        nextstate = c->state_out;
                   1372:        nextdyn += c->length;
                   1373:       }
                   1374:     } else {
                   1375: #if defined(GFORTH_DEBUGGING)
                   1376:       assert(0);
                   1377: #endif
                   1378:       tc=0;
                   1379:       /* tc= (Cell)vm_prims[inst[i][CANONICAL_STATE].inst]; */
                   1380:     }
                   1381:     *(instps[i]) = tc;
                   1382:   }      
                   1383:   if (!no_transition) {
                   1384:     PrimNum p = trans[i][nextstate].inst;
                   1385:     struct cost *c = super_costs+p;
                   1386:     assert(c->state_in==nextstate);
                   1387:     assert(trans[i][nextstate].cost != INF_COST);
                   1388:     assert(i==nextdyn);
1.128     anton    1389:     (void)compile_prim_dyn(p,NULL);
1.125     anton    1390:     nextstate = c->state_out;
1.107     anton    1391:   }
1.125     anton    1392:   assert(nextstate==CANONICAL_STATE);
1.107     anton    1393: }
                   1394: 
1.105     anton    1395: /* compile *start, possibly rewriting it into a static and/or dynamic
                   1396:    superinstruction */
                   1397: void compile_prim1(Cell *start)
1.70      anton    1398: {
1.108     anton    1399: #if defined(DOUBLY_INDIRECT)
1.125     anton    1400:   Label prim;
                   1401: 
                   1402:   if (start==NULL)
                   1403:     return;
                   1404:   prim = (Label)*start;
1.108     anton    1405:   if (prim<((Label)(xts+DOESJUMP)) || prim>((Label)(xts+npriminfos))) {
                   1406:     fprintf(stderr,"compile_prim encountered xt %p\n", prim);
                   1407:     *start=(Cell)prim;
                   1408:     return;
                   1409:   } else {
                   1410:     *start = (Cell)(prim-((Label)xts)+((Label)vm_prims));
                   1411:     return;
                   1412:   }
                   1413: #elif defined(INDIRECT_THREADED)
                   1414:   return;
1.112     anton    1415: #else /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1.128     anton    1416:   /* !! does not work, for unknown reasons; but something like this is
                   1417:      probably needed to ensure that we don't call compile_prim_dyn
                   1418:      before the inline arguments are there */
                   1419:   static Cell *instps[MAX_BB];
                   1420:   static PrimNum origs[MAX_BB];
                   1421:   static int ninsts=0;
                   1422:   PrimNum prim_num;
                   1423: 
                   1424:   if (start==NULL || ninsts >= MAX_BB ||
                   1425:       (ninsts>0 && superend[origs[ninsts-1]])) {
                   1426:     /* after bb, or at the start of the next bb */
                   1427:     optimize_rewrite(instps,origs,ninsts);
                   1428:     /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts); */
                   1429:     ninsts=0;
                   1430:     if (start==NULL)
                   1431:       return;
                   1432:   }
                   1433:   prim_num = ((Xt)*start)-vm_prims;
                   1434:   if(prim_num >= npriminfos) {
                   1435:     optimize_rewrite(instps,origs,ninsts);
1.129     anton    1436:     /* fprintf(stderr,"optimize_rewrite(...,%d)\n",ninsts);*/
1.128     anton    1437:     ninsts=0;
                   1438:     return;
                   1439:   }    
                   1440:   assert(ninsts<MAX_BB);
                   1441:   instps[ninsts] = start;
                   1442:   origs[ninsts] = prim_num;
                   1443:   ninsts++;
1.112     anton    1444: #endif /* !(defined(DOUBLY_INDIRECT) || defined(INDIRECT_THREADED)) */
1.47      anton    1445: }
                   1446: 
1.1       anton    1447: Address loader(FILE *imagefile, char* filename)
                   1448: /* returns the address of the image proper (after the preamble) */
                   1449: {
                   1450:   ImageHeader header;
                   1451:   Address image;
                   1452:   Address imp; /* image+preamble */
1.17      anton    1453:   Char magic[8];
                   1454:   char magic7; /* size byte of magic number */
1.1       anton    1455:   Cell preamblesize=0;
1.6       pazsan   1456:   Cell data_offset = offset_image ? 56*sizeof(Cell) : 0;
1.1       anton    1457:   UCell check_sum;
1.15      pazsan   1458:   Cell ausize = ((RELINFOBITS ==  8) ? 0 :
                   1459:                 (RELINFOBITS == 16) ? 1 :
                   1460:                 (RELINFOBITS == 32) ? 2 : 3);
                   1461:   Cell charsize = ((sizeof(Char) == 1) ? 0 :
                   1462:                   (sizeof(Char) == 2) ? 1 :
                   1463:                   (sizeof(Char) == 4) ? 2 : 3) + ausize;
                   1464:   Cell cellsize = ((sizeof(Cell) == 1) ? 0 :
                   1465:                   (sizeof(Cell) == 2) ? 1 :
                   1466:                   (sizeof(Cell) == 4) ? 2 : 3) + ausize;
1.21      anton    1467:   Cell sizebyte = (ausize << 5) + (charsize << 3) + (cellsize << 1) +
                   1468: #ifdef WORDS_BIGENDIAN
                   1469:        0
                   1470: #else
                   1471:        1
                   1472: #endif
                   1473:     ;
1.1       anton    1474: 
1.43      anton    1475:   vm_prims = engine(0,0,0,0,0);
1.47      anton    1476:   check_prims(vm_prims);
1.106     anton    1477:   prepare_super_table();
1.1       anton    1478: #ifndef DOUBLY_INDIRECT
1.59      anton    1479: #ifdef PRINT_SUPER_LENGTHS
                   1480:   print_super_lengths();
                   1481: #endif
1.43      anton    1482:   check_sum = checksum(vm_prims);
1.1       anton    1483: #else /* defined(DOUBLY_INDIRECT) */
1.43      anton    1484:   check_sum = (UCell)vm_prims;
1.1       anton    1485: #endif /* defined(DOUBLY_INDIRECT) */
1.10      pazsan   1486:   
                   1487:   do {
                   1488:     if(fread(magic,sizeof(Char),8,imagefile) < 8) {
1.84      anton    1489:       fprintf(stderr,"%s: image %s doesn't seem to be a Gforth (>=0.6) image.\n",
1.10      pazsan   1490:              progname, filename);
                   1491:       exit(1);
1.1       anton    1492:     }
1.10      pazsan   1493:     preamblesize+=8;
1.84      anton    1494:   } while(memcmp(magic,"Gforth3",7));
1.17      anton    1495:   magic7 = magic[7];
1.1       anton    1496:   if (debug) {
1.17      anton    1497:     magic[7]='\0';
1.21      anton    1498:     fprintf(stderr,"Magic found: %s ", magic);
                   1499:     print_sizes(magic7);
1.1       anton    1500:   }
                   1501: 
1.21      anton    1502:   if (magic7 != sizebyte)
                   1503:     {
                   1504:       fprintf(stderr,"This image is:         ");
                   1505:       print_sizes(magic7);
                   1506:       fprintf(stderr,"whereas the machine is ");
                   1507:       print_sizes(sizebyte);
1.1       anton    1508:       exit(-2);
                   1509:     };
                   1510: 
                   1511:   fread((void *)&header,sizeof(ImageHeader),1,imagefile);
1.10      pazsan   1512: 
                   1513:   set_stack_sizes(&header);
1.1       anton    1514:   
                   1515: #if HAVE_GETPAGESIZE
                   1516:   pagesize=getpagesize(); /* Linux/GNU libc offers this */
                   1517: #elif HAVE_SYSCONF && defined(_SC_PAGESIZE)
                   1518:   pagesize=sysconf(_SC_PAGESIZE); /* POSIX.4 */
                   1519: #elif PAGESIZE
                   1520:   pagesize=PAGESIZE; /* in limits.h according to Gallmeister's POSIX.4 book */
                   1521: #endif
1.144   ! pazsan   1522:   debugp(stderr,"pagesize=%ld\n",(unsigned long) pagesize);
1.1       anton    1523: 
1.34      anton    1524:   image = dict_alloc_read(imagefile, preamblesize+header.image_size,
                   1525:                          preamblesize+dictsize, data_offset);
1.33      anton    1526:   imp=image+preamblesize;
1.57      anton    1527:   alloc_stacks((ImageHeader *)imp);
1.1       anton    1528:   if (clear_dictionary)
1.33      anton    1529:     memset(imp+header.image_size, 0, dictsize-header.image_size);
1.90      anton    1530:   if(header.base==0 || header.base  == (Address)0x100) {
1.1       anton    1531:     Cell reloc_size=((header.image_size-1)/sizeof(Cell))/8+1;
                   1532:     char reloc_bits[reloc_size];
1.33      anton    1533:     fseek(imagefile, preamblesize+header.image_size, SEEK_SET);
1.10      pazsan   1534:     fread(reloc_bits, 1, reloc_size, imagefile);
1.90      anton    1535:     relocate((Cell *)imp, reloc_bits, header.image_size, (Cell)header.base, vm_prims);
1.1       anton    1536: #if 0
                   1537:     { /* let's see what the relocator did */
                   1538:       FILE *snapshot=fopen("snapshot.fi","wb");
                   1539:       fwrite(image,1,imagesize,snapshot);
                   1540:       fclose(snapshot);
                   1541:     }
                   1542: #endif
1.46      jwilke   1543:   }
                   1544:   else if(header.base!=imp) {
                   1545:     fprintf(stderr,"%s: Cannot load nonrelocatable image (compiled for address $%lx) at address $%lx\n",
                   1546:            progname, (unsigned long)header.base, (unsigned long)imp);
                   1547:     exit(1);
1.1       anton    1548:   }
                   1549:   if (header.checksum==0)
                   1550:     ((ImageHeader *)imp)->checksum=check_sum;
                   1551:   else if (header.checksum != check_sum) {
                   1552:     fprintf(stderr,"%s: Checksum of image ($%lx) does not match the executable ($%lx)\n",
                   1553:            progname, (unsigned long)(header.checksum),(unsigned long)check_sum);
                   1554:     exit(1);
                   1555:   }
1.53      anton    1556: #ifdef DOUBLY_INDIRECT
                   1557:   ((ImageHeader *)imp)->xt_base = xts;
                   1558: #endif
1.1       anton    1559:   fclose(imagefile);
                   1560: 
1.56      anton    1561:   /* unnecessary, except maybe for CODE words */
                   1562:   /* FLUSH_ICACHE(imp, header.image_size);*/
1.1       anton    1563: 
                   1564:   return imp;
                   1565: }
                   1566: 
1.72      anton    1567: /* pointer to last '/' or '\' in file, 0 if there is none. */
                   1568: char *onlypath(char *filename)
1.10      pazsan   1569: {
1.72      anton    1570:   return strrchr(filename, DIRSEP);
1.1       anton    1571: }
                   1572: 
                   1573: FILE *openimage(char *fullfilename)
1.10      pazsan   1574: {
                   1575:   FILE *image_file;
1.28      anton    1576:   char * expfilename = tilde_cstr(fullfilename, strlen(fullfilename), 1);
1.10      pazsan   1577: 
1.28      anton    1578:   image_file=fopen(expfilename,"rb");
1.1       anton    1579:   if (image_file!=NULL && debug)
1.28      anton    1580:     fprintf(stderr, "Opened image file: %s\n", expfilename);
1.10      pazsan   1581:   return image_file;
1.1       anton    1582: }
                   1583: 
1.28      anton    1584: /* try to open image file concat(path[0:len],imagename) */
1.1       anton    1585: FILE *checkimage(char *path, int len, char *imagename)
1.10      pazsan   1586: {
                   1587:   int dirlen=len;
1.1       anton    1588:   char fullfilename[dirlen+strlen(imagename)+2];
1.10      pazsan   1589: 
1.1       anton    1590:   memcpy(fullfilename, path, dirlen);
1.71      pazsan   1591:   if (fullfilename[dirlen-1]!=DIRSEP)
                   1592:     fullfilename[dirlen++]=DIRSEP;
1.1       anton    1593:   strcpy(fullfilename+dirlen,imagename);
1.10      pazsan   1594:   return openimage(fullfilename);
1.1       anton    1595: }
                   1596: 
1.10      pazsan   1597: FILE * open_image_file(char * imagename, char * path)
1.1       anton    1598: {
1.10      pazsan   1599:   FILE * image_file=NULL;
1.28      anton    1600:   char *origpath=path;
1.10      pazsan   1601:   
1.71      pazsan   1602:   if(strchr(imagename, DIRSEP)==NULL) {
1.10      pazsan   1603:     /* first check the directory where the exe file is in !! 01may97jaw */
                   1604:     if (onlypath(progname))
1.72      anton    1605:       image_file=checkimage(progname, onlypath(progname)-progname, imagename);
1.10      pazsan   1606:     if (!image_file)
                   1607:       do {
                   1608:        char *pend=strchr(path, PATHSEP);
                   1609:        if (pend==NULL)
                   1610:          pend=path+strlen(path);
                   1611:        if (strlen(path)==0) break;
                   1612:        image_file=checkimage(path, pend-path, imagename);
                   1613:        path=pend+(*pend==PATHSEP);
                   1614:       } while (image_file==NULL);
                   1615:   } else {
                   1616:     image_file=openimage(imagename);
                   1617:   }
1.1       anton    1618: 
1.10      pazsan   1619:   if (!image_file) {
                   1620:     fprintf(stderr,"%s: cannot open image file %s in path %s for reading\n",
1.28      anton    1621:            progname, imagename, origpath);
1.10      pazsan   1622:     exit(1);
1.7       anton    1623:   }
                   1624: 
1.10      pazsan   1625:   return image_file;
                   1626: }
1.11      pazsan   1627: #endif
                   1628: 
                   1629: #ifdef HAS_OS
                   1630: UCell convsize(char *s, UCell elemsize)
                   1631: /* converts s of the format [0-9]+[bekMGT]? (e.g. 25k) into the number
                   1632:    of bytes.  the letter at the end indicates the unit, where e stands
                   1633:    for the element size. default is e */
                   1634: {
                   1635:   char *endp;
                   1636:   UCell n,m;
                   1637: 
                   1638:   m = elemsize;
                   1639:   n = strtoul(s,&endp,0);
                   1640:   if (endp!=NULL) {
                   1641:     if (strcmp(endp,"b")==0)
                   1642:       m=1;
                   1643:     else if (strcmp(endp,"k")==0)
                   1644:       m=1024;
                   1645:     else if (strcmp(endp,"M")==0)
                   1646:       m=1024*1024;
                   1647:     else if (strcmp(endp,"G")==0)
                   1648:       m=1024*1024*1024;
                   1649:     else if (strcmp(endp,"T")==0) {
                   1650: #if (SIZEOF_CHAR_P > 4)
1.24      anton    1651:       m=1024L*1024*1024*1024;
1.11      pazsan   1652: #else
                   1653:       fprintf(stderr,"%s: size specification \"%s\" too large for this machine\n", progname, endp);
                   1654:       exit(1);
                   1655: #endif
                   1656:     } else if (strcmp(endp,"e")!=0 && strcmp(endp,"")!=0) {
                   1657:       fprintf(stderr,"%s: cannot grok size specification %s: invalid unit \"%s\"\n", progname, s, endp);
                   1658:       exit(1);
                   1659:     }
                   1660:   }
                   1661:   return n*m;
                   1662: }
1.10      pazsan   1663: 
1.109     anton    1664: enum {
                   1665:   ss_number = 256,
1.125     anton    1666:   ss_states,
1.109     anton    1667:   ss_min_codesize,
                   1668:   ss_min_ls,
                   1669:   ss_min_lsu,
                   1670:   ss_min_nexts,
                   1671: };
                   1672: 
1.10      pazsan   1673: void gforth_args(int argc, char ** argv, char ** path, char ** imagename)
                   1674: {
                   1675:   int c;
                   1676: 
1.1       anton    1677:   opterr=0;
                   1678:   while (1) {
                   1679:     int option_index=0;
                   1680:     static struct option opts[] = {
1.29      anton    1681:       {"appl-image", required_argument, NULL, 'a'},
1.1       anton    1682:       {"image-file", required_argument, NULL, 'i'},
                   1683:       {"dictionary-size", required_argument, NULL, 'm'},
                   1684:       {"data-stack-size", required_argument, NULL, 'd'},
                   1685:       {"return-stack-size", required_argument, NULL, 'r'},
                   1686:       {"fp-stack-size", required_argument, NULL, 'f'},
                   1687:       {"locals-stack-size", required_argument, NULL, 'l'},
                   1688:       {"path", required_argument, NULL, 'p'},
                   1689:       {"version", no_argument, NULL, 'v'},
                   1690:       {"help", no_argument, NULL, 'h'},
                   1691:       /* put something != 0 into offset_image */
                   1692:       {"offset-image", no_argument, &offset_image, 1},
                   1693:       {"no-offset-im", no_argument, &offset_image, 0},
                   1694:       {"clear-dictionary", no_argument, &clear_dictionary, 1},
1.4       anton    1695:       {"die-on-signal", no_argument, &die_on_signal, 1},
1.1       anton    1696:       {"debug", no_argument, &debug, 1},
1.144   ! pazsan   1697:       {"diag", no_argument, &diag, 1},
1.60      anton    1698:       {"no-super", no_argument, &no_super, 1},
                   1699:       {"no-dynamic", no_argument, &no_dynamic, 1},
1.66      anton    1700:       {"dynamic", no_argument, &no_dynamic, 0},
1.110     anton    1701:       {"print-metrics", no_argument, &print_metrics, 1},
1.109     anton    1702:       {"ss-number", required_argument, NULL, ss_number},
1.125     anton    1703:       {"ss-states", required_argument, NULL, ss_states},
1.109     anton    1704: #ifndef NO_DYNAMIC
                   1705:       {"ss-min-codesize", no_argument, NULL, ss_min_codesize},
                   1706: #endif
                   1707:       {"ss-min-ls",       no_argument, NULL, ss_min_ls},
                   1708:       {"ss-min-lsu",      no_argument, NULL, ss_min_lsu},
                   1709:       {"ss-min-nexts",    no_argument, NULL, ss_min_nexts},
1.110     anton    1710:       {"ss-greedy",       no_argument, &ss_greedy, 1},
1.1       anton    1711:       {0,0,0,0}
                   1712:       /* no-init-file, no-rc? */
                   1713:     };
                   1714:     
1.36      pazsan   1715:     c = getopt_long(argc, argv, "+i:m:d:r:f:l:p:vhoncsx", opts, &option_index);
1.1       anton    1716:     
                   1717:     switch (c) {
1.29      anton    1718:     case EOF: return;
                   1719:     case '?': optind--; return;
                   1720:     case 'a': *imagename = optarg; return;
1.10      pazsan   1721:     case 'i': *imagename = optarg; break;
1.1       anton    1722:     case 'm': dictsize = convsize(optarg,sizeof(Cell)); break;
                   1723:     case 'd': dsize = convsize(optarg,sizeof(Cell)); break;
                   1724:     case 'r': rsize = convsize(optarg,sizeof(Cell)); break;
                   1725:     case 'f': fsize = convsize(optarg,sizeof(Float)); break;
                   1726:     case 'l': lsize = convsize(optarg,sizeof(Cell)); break;
1.10      pazsan   1727:     case 'p': *path = optarg; break;
1.36      pazsan   1728:     case 'o': offset_image = 1; break;
                   1729:     case 'n': offset_image = 0; break;
                   1730:     case 'c': clear_dictionary = 1; break;
                   1731:     case 's': die_on_signal = 1; break;
                   1732:     case 'x': debug = 1; break;
1.83      anton    1733:     case 'v': fputs(PACKAGE_STRING"\n", stderr); exit(0);
1.109     anton    1734:     case ss_number: static_super_number = atoi(optarg); break;
1.125     anton    1735:     case ss_states: maxstates = max(min(atoi(optarg),MAX_STATE),1); break;
1.109     anton    1736: #ifndef NO_DYNAMIC
                   1737:     case ss_min_codesize: ss_cost = cost_codesize; break;
                   1738: #endif
                   1739:     case ss_min_ls:       ss_cost = cost_ls;       break;
                   1740:     case ss_min_lsu:      ss_cost = cost_lsu;      break;
                   1741:     case ss_min_nexts:    ss_cost = cost_nexts;    break;
1.1       anton    1742:     case 'h': 
1.29      anton    1743:       fprintf(stderr, "Usage: %s [engine options] ['--'] [image arguments]\n\
1.1       anton    1744: Engine Options:\n\
1.29      anton    1745:   --appl-image FILE                equivalent to '--image-file=FILE --'\n\
1.10      pazsan   1746:   --clear-dictionary               Initialize the dictionary with 0 bytes\n\
                   1747:   -d SIZE, --data-stack-size=SIZE   Specify data stack size\n\
                   1748:   --debug                          Print debugging information during startup\n\
1.144   ! pazsan   1749:   --diag                           Print diagnostic information during startup\n\
1.10      pazsan   1750:   --die-on-signal                  exit instead of CATCHing some signals\n\
1.66      anton    1751:   --dynamic                        use dynamic native code\n\
1.10      pazsan   1752:   -f SIZE, --fp-stack-size=SIZE            Specify floating point stack size\n\
                   1753:   -h, --help                       Print this message and exit\n\
                   1754:   -i FILE, --image-file=FILE       Use image FILE instead of `gforth.fi'\n\
                   1755:   -l SIZE, --locals-stack-size=SIZE Specify locals stack size\n\
                   1756:   -m SIZE, --dictionary-size=SIZE   Specify Forth dictionary size\n\
1.60      anton    1757:   --no-dynamic                     Use only statically compiled primitives\n\
1.10      pazsan   1758:   --no-offset-im                   Load image at normal position\n\
1.60      anton    1759:   --no-super                        No dynamically formed superinstructions\n\
1.10      pazsan   1760:   --offset-image                   Load image at a different position\n\
                   1761:   -p PATH, --path=PATH             Search path for finding image and sources\n\
1.110     anton    1762:   --print-metrics                  Print some code generation metrics on exit\n\
1.10      pazsan   1763:   -r SIZE, --return-stack-size=SIZE Specify return stack size\n\
1.111     anton    1764:   --ss-greedy                       greedy, not optimal superinst selection\n\
                   1765:   --ss-min-codesize                 select superinsts for smallest native code\n\
                   1766:   --ss-min-ls                       minimize loads and stores\n\
                   1767:   --ss-min-lsu                      minimize loads, stores, and pointer updates\n\
                   1768:   --ss-min-nexts                    minimize the number of static superinsts\n\
                   1769:   --ss-number=N                     use N static superinsts (default max)\n\
1.125     anton    1770:   --ss-states=N                     N states for stack caching (default max)\n\
1.66      anton    1771:   -v, --version                            Print engine version and exit\n\
1.1       anton    1772: SIZE arguments consist of an integer followed by a unit. The unit can be\n\
1.10      pazsan   1773:   `b' (byte), `e' (element; default), `k' (KB), `M' (MB), `G' (GB) or `T' (TB).\n",
                   1774:              argv[0]);
                   1775:       optind--;
                   1776:       return;
1.1       anton    1777:     }
                   1778:   }
1.10      pazsan   1779: }
1.11      pazsan   1780: #endif
1.10      pazsan   1781: 
1.144   ! pazsan   1782: void print_diag()
        !          1783: {
        !          1784:   /* prints something like:
        !          1785: missing functionality:
        !          1786:     no getrusage -> CPUTIME broken
        !          1787:     no ffcall -> only old-style foreign function calls (no fflib.fs)
        !          1788:     Alpha, no ffcall -> only integer args for foreign function calls
        !          1789:     Other functionality checked ok
        !          1790:   performance problems:
        !          1791:     gcc PR 15242 -> no dynamic code generation (use gcc-2.95 instead)
        !          1792:     double-cell integer type buggy -> 
        !          1793:        CMP, MUL, DIV, ADD, SHIFT, D2F, F2D, SIZE slow */
        !          1794: 
        !          1795: #if !defined(HAVE_GETRUSAGE) || !defined(HAS_FFCALL)
        !          1796:   fprintf(stderr, "missing functionality:\n"
        !          1797: #ifndef HAVE_GETRUSAGE
        !          1798:          "    no getrusage -> CPUTIME broken\n"
        !          1799: #endif
        !          1800: #ifndef HAS_FFCALL
        !          1801:          "    no ffcall -> only old-style foreign function calls (no fflib.fs)\n"
        !          1802: #endif
        !          1803:          );
        !          1804: #endif
        !          1805:   if((relocs < nonrelocs) ||
        !          1806: #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)
        !          1807:      1
        !          1808: #else
        !          1809:      0
        !          1810: #endif
        !          1811:      )
        !          1812:     debugp(stderr, "relocs: %d:%d\n", relocs, nonrelocs);
        !          1813:     fprintf(stderr, "performance problems:\n%s"
        !          1814: #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)
        !          1815:            "    double-cell integer type buggy ->\n        "
        !          1816: #ifdef BUGGY_LL_CMP
        !          1817:            "CMP, "
        !          1818: #endif
        !          1819: #ifdef BUGGY_LL_MUL
        !          1820:            "MUL, "
        !          1821: #endif
        !          1822: #ifdef BUGGY_LL_DIV
        !          1823:            "DIV, "
        !          1824: #endif
        !          1825: #ifdef BUGGY_LL_ADD
        !          1826:            "ADD, "
        !          1827: #endif
        !          1828: #ifdef BUGGY_LL_SHIFT
        !          1829:            "SHIFT, "
        !          1830: #endif
        !          1831: #ifdef BUGGY_LL_D2F
        !          1832:            "D2F, "
        !          1833: #endif
        !          1834: #ifdef BUGGY_LL_F2D
        !          1835:            "F2D, "
        !          1836: #endif
        !          1837:            "\b\b slow\n"
        !          1838: #endif
        !          1839:            ,
        !          1840:            (relocs < nonrelocs) ? "    gcc PR 15242 -> no dynamic code generation (use gcc-2.95 instead)\n" : "");
        !          1841: }
        !          1842: 
1.10      pazsan   1843: #ifdef INCLUDE_IMAGE
                   1844: extern Cell image[];
                   1845: extern const char reloc_bits[];
                   1846: #endif
1.67      pazsan   1847: 
1.10      pazsan   1848: int main(int argc, char **argv, char **env)
                   1849: {
1.30      pazsan   1850: #ifdef HAS_OS
1.10      pazsan   1851:   char *path = getenv("GFORTHPATH") ? : DEFAULTPATH;
1.30      pazsan   1852: #else
                   1853:   char *path = DEFAULTPATH;
                   1854: #endif
1.13      pazsan   1855: #ifndef INCLUDE_IMAGE
1.10      pazsan   1856:   char *imagename="gforth.fi";
                   1857:   FILE *image_file;
                   1858:   Address image;
                   1859: #endif
                   1860:   int retvalue;
                   1861:          
1.56      anton    1862: #if defined(i386) && defined(ALIGNMENT_CHECK)
1.10      pazsan   1863:   /* turn on alignment checks on the 486.
                   1864:    * on the 386 this should have no effect. */
                   1865:   __asm__("pushfl; popl %eax; orl $0x40000, %eax; pushl %eax; popfl;");
                   1866:   /* this is unusable with Linux' libc.4.6.27, because this library is
                   1867:      not alignment-clean; we would have to replace some library
                   1868:      functions (e.g., memcpy) to make it work. Also GCC doesn't try to keep
                   1869:      the stack FP-aligned. */
                   1870: #endif
                   1871: 
                   1872:   /* buffering of the user output device */
1.11      pazsan   1873: #ifdef _IONBF
1.10      pazsan   1874:   if (isatty(fileno(stdout))) {
                   1875:     fflush(stdout);
                   1876:     setvbuf(stdout,NULL,_IONBF,0);
1.1       anton    1877:   }
1.11      pazsan   1878: #endif
1.1       anton    1879: 
1.10      pazsan   1880:   progname = argv[0];
                   1881: 
1.11      pazsan   1882: #ifdef HAS_OS
1.10      pazsan   1883:   gforth_args(argc, argv, &path, &imagename);
1.109     anton    1884: #ifndef NO_DYNAMIC
                   1885:   if (no_dynamic && ss_cost == cost_codesize) {
1.122     anton    1886:     ss_cost = cost_nexts;
                   1887:     cost_sums[0] = cost_sums[1]; /* don't use cost_codesize for print-metrics */
1.144   ! pazsan   1888:     debugp(stderr, "--no-dynamic conflicts with --ss-min-codesize, reverting to --ss-min-nexts\n");
1.109     anton    1889:   }
                   1890: #endif /* !defined(NO_DYNAMIC) */
                   1891: #endif /* defined(HAS_OS) */
1.10      pazsan   1892: 
                   1893: #ifdef INCLUDE_IMAGE
                   1894:   set_stack_sizes((ImageHeader *)image);
1.22      pazsan   1895:   if(((ImageHeader *)image)->base != image)
                   1896:     relocate(image, reloc_bits, ((ImageHeader *)image)->image_size,
                   1897:             (Label*)engine(0, 0, 0, 0, 0));
1.10      pazsan   1898:   alloc_stacks((ImageHeader *)image);
                   1899: #else
                   1900:   image_file = open_image_file(imagename, path);
                   1901:   image = loader(image_file, imagename);
                   1902: #endif
1.24      anton    1903:   gforth_header=(ImageHeader *)image; /* used in SIGSEGV handler */
1.1       anton    1904: 
1.144   ! pazsan   1905:   if (diag)
        !          1906:     print_diag();
1.1       anton    1907:   {
1.10      pazsan   1908:     char path2[strlen(path)+1];
1.1       anton    1909:     char *p1, *p2;
                   1910:     Cell environ[]= {
                   1911:       (Cell)argc-(optind-1),
                   1912:       (Cell)(argv+(optind-1)),
1.10      pazsan   1913:       (Cell)strlen(path),
1.1       anton    1914:       (Cell)path2};
                   1915:     argv[optind-1] = progname;
                   1916:     /*
                   1917:        for (i=0; i<environ[0]; i++)
                   1918:        printf("%s\n", ((char **)(environ[1]))[i]);
                   1919:        */
                   1920:     /* make path OS-independent by replacing path separators with NUL */
1.10      pazsan   1921:     for (p1=path, p2=path2; *p1!='\0'; p1++, p2++)
1.1       anton    1922:       if (*p1==PATHSEP)
                   1923:        *p2 = '\0';
                   1924:       else
                   1925:        *p2 = *p1;
                   1926:     *p2='\0';
1.10      pazsan   1927:     retvalue = go_forth(image, 4, environ);
1.102     anton    1928: #ifdef SIGPIPE
                   1929:     bsd_signal(SIGPIPE, SIG_IGN);
                   1930: #endif
1.42      anton    1931: #ifdef VM_PROFILING
                   1932:     vm_print_profile(stderr);
                   1933: #endif
1.1       anton    1934:     deprep_terminal();
1.104     anton    1935:   }
1.110     anton    1936:   if (print_metrics) {
                   1937:     int i;
                   1938:     fprintf(stderr, "code size = %8ld\n", dyncodesize());
                   1939:     for (i=0; i<sizeof(cost_sums)/sizeof(cost_sums[0]); i++)
                   1940:       fprintf(stderr, "metric %8s: %8ld\n",
                   1941:              cost_sums[i].metricname, cost_sums[i].sum);
1.1       anton    1942:   }
1.13      pazsan   1943:   return retvalue;
1.1       anton    1944: }

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