Amiga Format CD 24/Amiga Format AFCD24 (Feb 1998, Issue 108).iso/-in_the_mag-/emulation/macos/uae069b2.src.cpt.hqx/UAE069ß2.SRC.CPT/uae069fl2.src/compiler.c

C/C++ Source or Header | 1997-06-13 | 136KB | 4,561 lines

 /*
  * UAE - The Un*x Amiga Emulator
  *
  * m68k emulation
  *
  * Copyright 1996 Bernd Schmidt
  */

#include "sysconfig.h"
#include "sysdeps.h"

#include "config.h"
#include "options.h"
#include "gensound.h"
#include "sounddep/sound.h"
#include "events.h"
#include "gui.h"
#include "my_memory.h"
#include "custom.h"
#include "machdep/m68k.h"
#include "readcpu.h"
#include "newcpu.h"
#include "ersatz.h"
#include "blitter.h"
#include "debug.h"
#include "autoconf.h"
#include "compiler.h"

#ifdef USE_COMPILER

#include <sys/mman.h>

char *address_space, *good_address_map;

code_execfunc exec_me;
uae_u8 nr_bbs_to_run = 1;
int nr_bbs_start = 40;

static int compile_failure;
static int quiet_compile = 1;
int i_want_to_die = 1;
static int n_compiled = 0;
static int n_max_comp = 99999999;
static uaecptr call_only_me = 0;

int patched_syscalls = 0;

static int count_bits(uae_u16 v)
{
    int bits = 0;
    while (v != 0) {
    if (v & 1)
        bits++;
    v >>= 1;
    }
    return bits;
}

static uae_u16 bitswap(uae_u16 v)
{
    uae_u16 newv = 0;
    uae_u16 m1 = 1, m2 = 0x8000;
    int i;

    for (i = 0; i < 16; i++) {
    if (v & m1)
        newv |= m2;
    m2 >>= 1;
    m1 <<= 1;
    }
    return newv;
}

static long long compiled_hits = 0;

/* 16K areas with 512 byte blocks */
#define SUBUNIT_ORDER 9
#define PAGE_SUBUNIT (1 << SUBUNIT_ORDER)
#define PAGE_ALLOC_UNIT (PAGE_SUBUNIT * 32)

static int zerofd;
static int zeroff;
static struct code_page *first_code_page;

static struct code_page *new_code_page(void)
{
    struct code_page *ncp;

    ncp = (struct code_page *)mmap(NULL, PAGE_ALLOC_UNIT,
                   PROT_EXEC|PROT_READ|PROT_WRITE, MAP_PRIVATE,
                   zerofd, zeroff);
    zeroff += PAGE_ALLOC_UNIT;
    if (ncp) {
    ncp->next = first_code_page;
    first_code_page = ncp;
    ncp->allocmask = 1; /* what a waste */
    }
    return ncp;
}

#define NUM_HASH 32768 /* larger values cause some paging on my 16MB machine */
#define HASH_MASK (NUM_HASH-1)
#define MAX_UNUSED_HASH 512

static int SCAN_MARK = 1; /* Number of calls after which to scan a function */
static int COMPILE_MARK = 5; /* Number of calls after which to compile a function */

/* The main address -> function lookup hashtable. We use the lower bits of
 * the address as hash function. */
static struct hash_entry cpu_hash[NUM_HASH];
/* These aren't really LRU lists... They used to be, but keeping them in that
 * order is costly. The hash LRU list is now a two-part list: Functions that have
 * no code allocated for them are placed at the beginning. Such entries can be
 * recycled when we need a new hash entry. */
static struct hash_block lru_first_block;
static struct hash_entry lru_first_hash;
static struct hash_entry *freelist_hash;
static struct hash_block *freelist_block;
static int num_unused_hash;

static int m68k_scan_func(struct hash_entry *);
static int m68k_compile_block(struct hash_block *);

static char *alloc_code(struct hash_block *hb, int ninsns)
{
    struct code_page *cp;
    long int allocsize = (ninsns * 32 + PAGE_SUBUNIT-1) & ~(PAGE_SUBUNIT-1);
    uae_u32 allocmask;
    int allocbits;
    int j;
    int last_bit;

    if (allocsize >= (PAGE_ALLOC_UNIT - (1 << SUBUNIT_ORDER)))
    return NULL;
    allocbits = (allocsize >> SUBUNIT_ORDER);
    allocmask = (1 << allocbits) - 1;

    for (cp = first_code_page; cp != NULL; cp = cp->next) {
    uae_u32 thispage_alloc = cp->allocmask;
    for (j = 1; j < (33 - allocbits); j++) {
        if ((cp->allocmask & (allocmask << j)) == 0) {
        goto found_page;
        }
    }
    }

    /* Nothing large enough free: make a new page */
    cp = new_code_page();
    if (cp == NULL)
    return NULL;
    j = 1;

found_page:
    /* See whether there is in fact more space for us. If so, allocate all of
     * it. compile_block() will free everything it didn't need. */

    allocmask <<= j;
    last_bit = allocbits + j;
    while (last_bit < 32 && (cp->allocmask & (1 << last_bit)) == 0) {
    allocmask |= 1 << last_bit;
    allocsize += PAGE_SUBUNIT;
    last_bit++;
    }

    hb->page_allocmask = allocmask;
    hb->cpage = cp;
    cp->allocmask |= allocmask;
    hb->compile_start = ((char *)cp + (j << SUBUNIT_ORDER));
    hb->alloclen = allocsize;
    return hb->compile_start;
}

static void remove_hash_from_lists(struct hash_entry *h)
{
    h->lru_next->lru_prev = h->lru_prev;
    h->lru_prev->lru_next = h->lru_next;

    h->next->prev = h->prev;
    h->prev->next = h->next;
}

static void lru_touch(struct hash_entry *h)
{
    h->lru_next->lru_prev = h->lru_prev;
    h->lru_prev->lru_next = h->lru_next;

    h->lru_next = &lru_first_hash;
    h->lru_prev = lru_first_hash.lru_prev;
    h->lru_prev->lru_next = h;
    lru_first_hash.lru_prev = h;
}

static void lru_untouch(struct hash_entry *h)
{
    h->lru_next->lru_prev = h->lru_prev;
    h->lru_prev->lru_next = h->lru_next;

    h->lru_prev = &lru_first_hash;
    h->lru_next = lru_first_hash.lru_next;
    h->lru_next->lru_prev = h;
    lru_first_hash.lru_next = h;
}

static void forget_block(struct hash_block *hb)
{
    struct hash_entry *h = hb->he_first;

    hb->lru_next->lru_prev = hb->lru_prev;
    hb->lru_prev->lru_next = hb->lru_next;

    hb->lru_next = freelist_block;
    freelist_block = hb;

    if (hb->cpage != NULL)
    fprintf(stderr, "Discarding block with code. Tsk.\n");

    do {
    struct hash_entry *next = h->next_same_block;
    h->block = NULL;
    h->execute = NULL;
    h->next_same_block = NULL;
    h = next;
    num_unused_hash++;
    lru_untouch(h);
    } while (h != hb->he_first);
    compiler_flush_jsr_stack();
}

static void lru_touch_block(struct hash_block *h)
{
    h->lru_next->lru_prev = h->lru_prev;
    h->lru_prev->lru_next = h->lru_next;

    h->lru_next = &lru_first_block;
    h->lru_prev = lru_first_block.lru_prev;
    h->lru_prev->lru_next = h;
    lru_first_block.lru_prev = h;
}

static __inline__ int check_block(struct hash_block *hb)
{
#ifndef RELY_ON_LOADSEG_DETECTION
    struct hash_entry *h = hb->he_first;

    do {
    struct hash_entry *next = h->next_same_block;
    if (h->matchword != *(uae_u32 *)get_real_address(h->addr))
        return 0;
    h = next;
    } while (h != hb->he_first);
#endif
    return 1;
}

uae_u32 flush_icache(void)
{
    struct hash_block *hb = lru_first_block.lru_next;

    while (hb != &lru_first_block) {
    struct hash_block *next = hb->lru_next;
    if (hb->cpage != NULL) {
        /* Address in chipmem? Then forget about block*/
        if ((hb->he_first->addr & ~0xF80000) != 0xF80000) {
        hb->cpage->allocmask &= ~hb->page_allocmask;
        hb->cpage = NULL;
        forget_block(hb);
        }
    }
    hb = next;
    }
    return m68k_dreg(regs, 0);
}

void possible_loadseg(void)
{
    fprintf(stderr, "Possible LoadSeg() detected\n");
    flush_icache();
}

static struct hash_block *new_block(void)
{
    struct hash_block *b = freelist_block;

    if (b != NULL) {
    freelist_block = b->lru_next;
    } else
    b = (struct hash_block *)malloc(sizeof *b);
    b->nrefs = 0;
    b->cpage = NULL;
    b->he_first = NULL;
    b->translated = b->untranslatable = b->allocfailed = 0;
    return b;
}

static struct hash_entry *get_free_hash(void)
{
    struct hash_entry *h;

    for (;;) {
    h = freelist_hash;
    if (h != NULL) {
        freelist_hash = h->next_same_block;
        break;
    }
    h = lru_first_hash.lru_next;
    if (num_unused_hash >= MAX_UNUSED_HASH && h->block == NULL
        && !h->locked)
    {
        remove_hash_from_lists(h);
        num_unused_hash--;
        break;
    }
    h = (struct hash_entry *)malloc(sizeof(struct hash_entry));
    h->next_same_block = NULL;
    h->addr = -1;
    break;
    }
    num_unused_hash++;
    h->block = NULL;
    h->ncalls = 0;
    h->locked = h->cacheflush = 0;
    h->execute = NULL;
    return h;
}

static struct hash_entry *new_hash(uaecptr addr)
{
    struct hash_entry *h = get_free_hash();

    h->addr = addr;

    /* Chain the new node */
    h->prev = cpu_hash + ((addr >> 1) & HASH_MASK);
    h->next = h->prev->next;
    h->next->prev = h->prev->next = h;

    h->lru_next = &lru_first_hash;
    h->lru_prev = lru_first_hash.lru_prev;
    h->lru_prev->lru_next = h;
    lru_first_hash.lru_prev = h;

    h->next_same_block = NULL;

    return h;
}
static struct hash_entry *find_hash(uaecptr addr)
{
    struct hash_entry *h;
    struct hash_entry *h1 = cpu_hash + ((addr >> 1) & HASH_MASK);

    if (h1->next->addr == addr)
    return h1->next;

    for (h = h1->next; h != h1; h = h->next) {
    if (h->addr == addr) {
        /* Put it at the head of the list so that the above shortcut
         * works the next time we come here */
        h->next->prev = h->prev; h->prev->next = h->next;
        h->prev = h1;
        h->next = h1->next;
        h->next->prev = h->prev->next = h;
        return h;
    }
    }
    return NULL;
}

static struct hash_entry *get_hash_for_func(uaecptr addr, int mark_locked)
{
    struct hash_entry *h = find_hash(addr);
    if (h == NULL)
    h = new_hash (addr);
#if 0 /* Too expensive */
    else
    lru_touch(h);
#endif
    if (mark_locked)
    h->locked = 1;
    return h;
}

static struct hash_entry *get_hash(uaecptr addr)
{
    struct hash_entry *h = get_hash_for_func(addr, 0);

    if (h->block == NULL) {
    if (++h->ncalls == SCAN_MARK) {
        m68k_scan_func(h);
    }
    } else
    if (!h->block->untranslatable && h->block->nrefs++ == COMPILE_MARK) {
        lru_touch_block(h->block);
        if (m68k_compile_block(h->block)) {
        h->block->untranslatable = 1;
        } else {
        h->block->translated = 1;
        }
    }
    return h;
}

void special_flush_hash(uaecptr addr)
{
    struct hash_entry *h = get_hash_for_func(addr, 0);

    h->cacheflush = 1;
}

static __inline__ void m68k_setpc_hash(uaecptr newpc)
{
    struct hash_entry *h = get_hash(newpc);

    if (h->cacheflush)
    flush_icache();

    if (h->execute != NULL) {
    if ((h->addr & 0xF80000) == 0xF80000 || check_block(h->block)) {
        compiled_hits++;
        if (i_want_to_die && (call_only_me == 0 || call_only_me == newpc)) {
        exec_me = h->execute;
        nr_bbs_to_run = nr_bbs_start;
        regs.spcflags |= SPCFLAG_EXEC;
        }
    } else
        flush_icache();
    }
    regs.pc = newpc;
    regs.pc_p = regs.pc_oldp = get_real_address(newpc);
}

static __inline__ void m68k_setpc_nohash(uaecptr newpc)
{
#if 0
    /* This is probably not too good for efficiency... FIXME */
    struct hash_entry *h = find_hash(newpc);

    if (h != NULL && h->cacheflush)
    flush_icache();
#endif
    regs.pc = newpc;
    regs.pc_p = regs.pc_oldp = get_real_address(newpc);
}

void m68k_setpc(uaecptr newpc)
{
    m68k_setpc_hash(newpc);
}

void m68k_setpc_fast(uaecptr newpc)
{
    m68k_setpc_nohash(newpc);
}

void m68k_setpc_rte(uaecptr newpc)
{
    m68k_setpc_nohash(newpc);
}

void m68k_setpc_bcc(uaecptr newpc)
{
    m68k_setpc_hash(newpc);
}

static void hash_init(void)
{
    int i;
    struct hash_entry **hepp;

    freelist_block = NULL;
    freelist_hash = NULL;

    for(i = 0; i < NUM_HASH; i++) {
    cpu_hash[i].next = cpu_hash[i].prev = cpu_hash + i;
    cpu_hash[i].lru_next = cpu_hash[i].lru_prev = NULL;
    cpu_hash[i].block = NULL;
    cpu_hash[i].locked = 0; cpu_hash[i].cacheflush = 0;
    cpu_hash[i].addr = -1;
    }

    lru_first_hash.lru_next = lru_first_hash.lru_prev = &lru_first_hash;
    lru_first_block.lru_next = lru_first_block.lru_prev = &lru_first_block;

    num_unused_hash = 0;
}

static void code_init(void)
{
    first_code_page = NULL;
    zerofd = open("/dev/zero", O_RDWR);
    zeroff = 0;
}

#define CC68K_C 16
#define CC68K_V 8
#define CC68K_Z 4
#define CC68K_N 2
#define CC68K_X 1

static __inline__ int cc_flagmask_68k(const int cc)
{
    switch(cc){
     case 0: return 0;                       /* T */
     case 1: return 0;                       /* F */
     case 2: return CC68K_C|CC68K_Z;         /* HI */
     case 3: return CC68K_C|CC68K_Z;         /* LS */
     case 4: return CC68K_C;                 /* CC */
     case 5: return CC68K_C;                 /* CS */
     case 6: return CC68K_Z;                 /* NE */
     case 7: return CC68K_Z;                 /* EQ */
     case 8: return CC68K_V;                 /* VC */
     case 9: return CC68K_V;                 /* VS */
     case 10:return CC68K_N;                 /* PL */
     case 11:return CC68K_N;                 /* MI */
     case 12:return CC68K_N|CC68K_V;         /* GE */
     case 13:return CC68K_N|CC68K_V;         /* LT */
     case 14:return CC68K_N|CC68K_V|CC68K_Z; /* GT */
     case 15:return CC68K_N|CC68K_V|CC68K_Z; /* LE */
    }
    abort();
    return 0;
}

static __inline__ void translate_step_over_ea(uae_u8 **pcpp, amodes m,
                          wordsizes size)
{
    switch (m) {
     case Areg:
     case Dreg:
     case Aind:
     case Aipi:
     case Apdi:
     case immi:
    break;

     case imm:
    if (size == sz_long)
        goto is_long;
    /* fall through */
     case Ad16:
     case PC16:
     case imm0:
     case imm1:
     case absw:
    (*pcpp)+=2;
    break;
     case Ad8r:
     case PC8r:
    {
        uae_u16 extra = *(*pcpp)++;
        extra <<= 8;
        extra |= *(*pcpp)++;
        /* @@@ handle 68020 stuff here */
    }
    break;
     case absl:
     case imm2:
    is_long:
    (*pcpp) += 4;
    break;
    }
}

static struct instr *translate_getnextinsn(uae_u8 **pcpp)
{
    uae_u16 opcode;
    struct instr *dp;

    opcode = *(*pcpp)++ << 8;
    opcode |= *(*pcpp)++;

    if (cpufunctbl[opcode] == op_illg) {
    opcode = 0x4AFC;
    }
    dp = table68k + opcode;
    if (dp->suse) {
    translate_step_over_ea(pcpp, dp->smode, dp->size);
    }
    if (dp->duse) {
    translate_step_over_ea(pcpp, dp->dmode, dp->size);
    }
    return dp;
}

#define CB_STACKSIZE 200
#define BB_STACKSIZE 200

static uae_u32 condbranch_stack[CB_STACKSIZE];
static int condbranch_src_stack[CB_STACKSIZE];

struct bb_info {
    struct hash_entry *h;
    uaecptr stopaddr;
    int can_compile_last;
    struct bb_info *bb_next1, *bb_next2;
    int flags_live_at_end;
    int flags_live_at_start;
    int first_iip, last_iip;
} bb_stack[BB_STACKSIZE];

static int top_bb;

static uaecptr bcc_target_stack[BB_STACKSIZE];

static int new_bcc_target(uaecptr addr)
{
    int i;

    for (i = 0; i < top_bb; i++)
    if (bcc_target_stack[i] == addr)
        return 1;

    if (top_bb == BB_STACKSIZE)
    return 0;
    bcc_target_stack[top_bb++] = addr;
    return 1;
}

static int bcc_compfn(const void *a, const void *b)
{
    uaecptr *a1 = (uaecptr *)a, *b1 = (uaecptr *)b;

    if (*a1 == *b1)
    printf("BUG!!\n");

    if (*a1 < *b1)
    return 1;
    return -1;
}

static int bb_compfn(const void *a, const void *b)
{
    struct bb_info *a1 = (struct bb_info *)a, *b1 = (struct bb_info *)b;

    if (a1->h->addr == b1->h->addr)
    printf("BUG!!\n");

    if (a1->h->addr < b1->h->addr)
    return -1;
    return 1;
}

static int find_basic_blocks(struct hash_entry *h)
{
    int current_bb = 0;

    top_bb = 0;
    bcc_target_stack[0] = h->addr;
    new_bcc_target(h->addr);

    while (top_bb > current_bb) {
    uaecptr addr = bcc_target_stack[current_bb];
    int ninsns = 0;
    uae_u8 *realpc = get_real_address(addr);
    uae_u8 *rpc_start = realpc;

    for(;;) {
        uaecptr thisinsn_addr = (realpc - rpc_start) + addr;
        uae_u8 *rpc_save = realpc;
        struct instr *dp = translate_getnextinsn(&realpc);
        uaecptr nextinsn_addr = (realpc - rpc_start) + addr;

        if (dp->mnemo == i_RTS || dp->mnemo == i_RTE
        || dp->mnemo == i_RTR || dp->mnemo == i_RTD
        || dp->mnemo == i_JMP || dp->mnemo == i_ILLG)
        {
        break;
        }

        if (dp->mnemo == i_BSR || dp->mnemo == i_JSR) {
        if (!new_bcc_target(nextinsn_addr))
            return 0;
        break;
        }

        if (dp->mnemo == i_DBcc) {
        uaecptr newaddr = thisinsn_addr + 2 + (uae_s16)((*(rpc_save+2) << 8) | *(rpc_save+3));
        if (!new_bcc_target(nextinsn_addr))
            return 0;
        if (!new_bcc_target(newaddr))
            return 0;
        break;
        }

        if (dp->mnemo == i_Bcc) {
        uaecptr newaddr;
        if (dp->smode == imm1)
            newaddr = thisinsn_addr + 2 + (uae_s16)((*(rpc_save+2) << 8) | *(rpc_save+3));
        else
            newaddr = thisinsn_addr + 2 + (uae_s8)dp->sreg;

        if (dp->cc != 0)
            if (!new_bcc_target(nextinsn_addr))
            return 0;
        if (!new_bcc_target(newaddr))
            return 0;
        break;
        }
    }
    current_bb++;
    }

    qsort(bcc_target_stack, top_bb, sizeof (uaecptr), bcc_compfn);

    return 1;
}

static int m68k_scan_func(struct hash_entry *h)
{
    int i;
    struct hash_block *found_block;
    struct hash_entry **hepp;

    if (!find_basic_blocks(h))
    return 0;

    found_block = NULL;

    /* First, lock the hash entries we already have to prevent grief */
    for (i = 0; i < top_bb; i++) {
    struct hash_entry *h = find_hash(bcc_target_stack[i]);
    if (h != NULL)
        h->locked = 1;
    }

    /* Allocate new ones */
    for (i = 0; i < top_bb; i++) {
    struct hash_entry *h = get_hash_for_func(bcc_target_stack[i], 1);
    bb_stack[i].h = h;
#if 0 /* This doesn't work in all cases */
    if (h->block != NULL && h->block != found_block) {
        if (found_block == NULL) {
        if (h->block->cpage != NULL)
            fprintf(stderr, "Found compiled code\n");
        else
            found_block = h->block;
        } else {
        fprintf(stderr, "Multiple blocks found.\n");
        if (h->block->cpage == NULL)
            forget_block(h->block);
        else if (found_block->cpage == NULL) {
            forget_block(found_block);
            found_block = h->block;
        } else
            fprintf(stderr, "Bad case.\n");
        }
    }
#endif
    }
    if (found_block == NULL) {
    found_block = new_block();

    found_block->lru_next = &lru_first_block;
    found_block->lru_prev = lru_first_block.lru_prev;
    found_block->lru_prev->lru_next = found_block;
    lru_first_block.lru_prev = found_block;
    }

    hepp = &found_block->he_first;
    found_block->he_first = NULL;
    for (i = 0; i < top_bb; i++) {
    struct bb_info *bb = bb_stack + i;

    if (bb->h->block == NULL) {
        num_unused_hash--;
        lru_touch(bb->h);
        bb->h->block = found_block;
        *hepp = bb->h;
        hepp = &bb->h->next_same_block;
    }
    }
    *hepp = found_block->he_first;
    return 1;
}

struct ea_reg_info {
    enum { eat_reg, eat_imem, eat_amem, eat_const } ea_type;
    int regs_set:16;
    int regs_used:16;
    int nr_scratch;
    uae_u32 temp1, temp2;
};

#define MAX_TRANSLATE 2048
struct insn_info_struct {
    uaecptr address;
    struct instr *dp;
    int flags_set;
    int flags_used;
    int flags_live_at_end;
    int jump_target;
    int jumps_to;
    char *compiled_jumpaddr; /* Address to use for jumps to this insn */
    char *compiled_fillin;   /* Address where to put offset if this is a Bcc */
    int regs_set:16;
    int regs_used:16;
    int stop_translation:2;
    int sync_cache:1;
    int sync_flags:1;
    int ccuser_follows:1;
} insn_info [MAX_TRANSLATE];

#define EA_NONE 0
#define EA_LOAD 1
#define EA_STORE 2
#define EA_MODIFY 4

#if 0
static void analyze_ea_for_insn(amodes mode, int reg, wordsizes size,
                struct ea_reg_info *eai,
                uae_u8 **pcpp, uaecptr pca,
                int ea_purpose)
{
    uae_u8 *p = *pcpp;

    switch(mode) {
     case Dreg:
    eai->ea_type = eat_reg;
    if (size != sz_long && (ea_purpose & EA_STORE))
        ea_purpose |= EA_LOAD;
    if (ea_purpose & EA_LOAD)
        eai->regs_used |= 1 << reg;
    if (ea_purpose & EA_STORE)
        eai->regs_set |= 1 << reg;
    break;

     case Areg:
    eai->ea_type = eat_reg;
    if (size != sz_long && (ea_purpose & EA_STORE))
        printf("Areg != long\n");
    if (ea_purpose & EA_LOAD)
        eai->regs_used |= 1 << (8+reg);
    if (ea_purpose & EA_STORE)
        eai->regs_set |= 1 << (8+reg);
    break;

     case Ad16:
     case Aind:
     case Apdi:
     case Aipi:
    eai->ea_type = eat_imem;
    eai->regs_used |= 1 << (8+reg);
    break;

     case Ad8r:
    eai->ea_type = eat_imem;
    pii->regs_used |= 1 << (8+reg);

    eai->temp = (uae_u16)((*p << 8) | *(p+1));
    r = (eai->temp & 0x7000) >> 12;
    (*pcpp) += 2; p += 2;

    if (eai->temp1 & 0x8000)
        pii->regs_used |= 1 << (8+r);
    else
        pii->regs_used |= 1 << r;
    break;

     case PC8r:
    eai->ea_type = eat_imem;
    eai->temp1 = (uae_u16)do_get_mem_word((uae_u16 *)p);
    eai->temp2 = pca + (uae_s8)eai->temp1;
    (*pcpp) += 2; p += 2;
    r = (eai->temp1 & 0x7000) >> 12;

    if (eai->temp1 & 0x8000)
        pii->regs_used |= 1 << (8+r);
    else
        pii->regs_used |= 1 << r;
    break;

     case PC16:
    eai->ea_type = eat_amem;
    eai->temp1 = pca + (uae_s16)do_get_mem_word((uae_u16 *)p);
    (*pcpp) += 2;
    break;

     case absw:
    eai->ea_type = eat_amem;
    eai->temp1 = (uae_s16)do_get_mem_word((uae_u16 *)p);
    (*pcpp) += 2;
    break;

     case absl:
    eai->ea_type = eat_amem;
    eai->temp1 = (uae_s32)do_get_mem_long((uae_u32 *)p);
    (*pcpp) += 4;
    break;

     case imm:
    if (size == sz_long)
        goto imm2_const;
    if (size == sz_word)
        goto imm1_const;

    /* fall through */
     case imm0:
    eai->ea_type = eat_imm;
    eai->temp1 = (uae_s8)*(p+1);
    (*pcpp) += 2;
    break;

     case imm1:
    imm1_const:
    eai->ea_type = eat_imm;
    eai->temp1 = (uae_s16)do_get_mem_word((uae_u16 *)p);
    (*pcpp) += 2;
    break;

     case imm2:
    imm2_const:
    eai->ea_type = eat_imm;
    eai->temp1 = (uae_s32)do_get_mem_long((uae_u32 *)p);
    (*pcpp) += 4;
    break;

     case immi:
    eai->ea_type = eat_imm;
    eai->temp1 = (uae_s8)reg;
    break;

     default:
    break;
    }
}
#endif
static struct bb_info *find_bb(struct hash_entry *h)
{
    int i;

    if (h == NULL)
    printf("Bug...\n");

    for (i = 0; i < top_bb; i++)
    if (bb_stack[i].h == h)
        return bb_stack + i;
    if (!quiet_compile)
    fprintf(stderr, "BB not found!\n");
    return NULL;
}

static int m68k_scan_block(struct hash_block *hb, int *movem_count)
{
    struct hash_entry *h = hb->he_first;
    int i, iip, last_iip;
    int changed, round;

    top_bb = 0;

    do {
    struct bb_info *bb = bb_stack + top_bb;
    bb->h = h;
    bb->bb_next1 = NULL;
    bb->bb_next2 = NULL;
    h = h->next_same_block;
    top_bb++;
    } while (h != hb->he_first);

    qsort(bb_stack, top_bb, sizeof (struct bb_info), bb_compfn);

    *movem_count = 0;

    iip = 0;
    for (i = 0; i < top_bb; i++) {
    struct bb_info *bb = bb_stack + i;
    uae_u8 *realpc = get_real_address(bb->h->addr);
    uae_u8 *rpc_start = realpc;
    uaecptr stop_addr = 0;
    int live_at_start = 31, may_clear_las = 31;
    struct insn_info_struct *prev_ii = NULL;

    if (i < top_bb - 1)
        stop_addr = (bb+1)->h->addr;
    bb->first_iip = iip;

    for (;;) {
        struct insn_info_struct *thisii = insn_info + iip;
        uaecptr thisinsn_addr = (realpc - rpc_start) + bb->h->addr;
        uae_u8 *rpc_save = realpc;
        struct instr *dp = translate_getnextinsn(&realpc);
        uaecptr nextinsn_addr = (realpc - rpc_start) + bb->h->addr;

        int fset = dp->flagdead == -1 ? 31 : dp->flagdead;
        int fuse = dp->flaglive == -1 ? 31 : dp->flaglive;

        if (thisinsn_addr == stop_addr) {
        bb->bb_next1 = find_bb (find_hash (thisinsn_addr));
        break;
        }

        if (dp->mnemo == i_Scc || dp->mnemo == i_Bcc || dp->mnemo == i_DBcc) {
        fset = 0, fuse = cc_flagmask_68k(dp->cc);
        if (prev_ii && dp->mnemo != i_Scc) /* Don't use Scc here: ea can cause an exit */
            prev_ii->ccuser_follows = 1;
        }

        may_clear_las &= ~fuse;
        live_at_start &= ~(fset & may_clear_las);

        thisii->dp = dp;
        thisii->address = thisinsn_addr;
        thisii->stop_translation = 0;
        thisii->ccuser_follows = 0;
/*        thisii->have_reginfo = 0;*/
        thisii->jump_target = 0;
        thisii->sync_cache = thisii->sync_flags = 0;
        thisii->flags_set = fset;
        thisii->flags_used = fuse;
        thisii->regs_set = 0;
        thisii->regs_used = 0;
        iip++;
        if (iip == MAX_TRANSLATE)
        return 0;

        if (dp->mnemo == i_RTS || dp->mnemo == i_RTE
        || dp->mnemo == i_RTR || dp->mnemo == i_RTD
        || dp->mnemo == i_JMP || dp->mnemo == i_ILLG)
        {
        thisii->flags_used = 31;
        thisii->regs_used = 65535;
        thisii->stop_translation = dp->mnemo == i_RTS || dp->mnemo == i_JMP ? 2 : 1;
        break;
        }
        if (dp->mnemo == i_BSR || dp->mnemo == i_JSR)
        {
        thisii->flags_used = 31;
        thisii->regs_used = 65535;
        bb->can_compile_last = 1;
        bb->bb_next1 = find_bb (get_hash_for_func (nextinsn_addr, 1));
        if (bb->bb_next1 == NULL)
            thisii->stop_translation = 1;
        break;
        }

        if (dp->mnemo == i_DBcc) {
        uaecptr newaddr = thisinsn_addr + 2 + (uae_s16)((*(rpc_save+2) << 8) | *(rpc_save+3));
        bb->can_compile_last = 1;
        bb->bb_next1 = find_bb (get_hash_for_func (newaddr, 1));
        if (bb->bb_next1 == NULL)
            thisii->stop_translation = 1;
        bb->bb_next2 = find_bb (get_hash_for_func (nextinsn_addr, 1));
        if (bb->bb_next2 == NULL)
            thisii->stop_translation = 1;
        thisii->regs_used = 65535;
        break;
        }

        if (dp->mnemo == i_Bcc) {
        uaecptr newaddr;
        if (dp->smode == imm1)
            newaddr = thisinsn_addr + 2 + (uae_s16)((*(rpc_save+2) << 8) | *(rpc_save+3));
        else
            newaddr = thisinsn_addr + 2 + (uae_s8)dp->sreg;
        bb->can_compile_last = 1;
        bb->bb_next1 = find_bb(get_hash_for_func(newaddr, 1));
        if (bb->bb_next1 == NULL)
            thisii->stop_translation = 1;
        if (dp->cc != 0) {
            bb->bb_next2 = find_bb(get_hash_for_func(nextinsn_addr, 1));
            if (bb->bb_next2 == NULL)
            thisii->stop_translation = 1;
        }
        thisii->regs_used = 65535;
        break;
        }

        if (dp->mnemo == i_MVMLE || dp->mnemo == i_MVMEL) {
        uae_u16 regmask = (*(rpc_save + 2) << 8) | (*(rpc_save + 3));
        *movem_count += count_bits(regmask);
        if (dp->dmode == Apdi)
            regmask = bitswap(regmask);
        if (dp->mnemo == i_MVMLE)
            thisii->regs_used = regmask;
        else
            thisii->regs_set = regmask;
        }

        prev_ii = thisii;
    }
    bb->last_iip = iip - 1;
    bb->flags_live_at_start = live_at_start;
    }
    last_iip = iip;
    round = 0;
    do {
    changed = 0;
    for (i = 0; i < top_bb; i++) {
        struct bb_info *bb = bb_stack + i;
        int mnemo;
        int current_live;
        struct instr *dp;

        iip = bb->last_iip;
        mnemo = insn_info[iip].dp->mnemo;

        /* Fix up branches */
        if (round == 0 && (mnemo == i_DBcc || mnemo == i_Bcc)) {
        if (bb->bb_next1 != NULL) {
            insn_info[bb->last_iip].jumps_to = bb->bb_next1->first_iip;
            insn_info[bb->bb_next1->first_iip].jump_target = 1;
        }
        }

        /* And take care of flag life information */
        dp = insn_info[iip].dp;
        if (insn_info[iip].stop_translation)
        current_live = 31;
        else if (dp->mnemo == i_DBcc || dp->mnemo == i_Bcc) {
        current_live = 0;
        if (bb->bb_next1 != NULL)
            current_live |= bb->bb_next1->flags_live_at_start;
        if (bb->bb_next2 != NULL)
            current_live |= bb->bb_next2->flags_live_at_start;
        } else {
        if (bb->bb_next1 == NULL && bb->bb_next2 == NULL)
            fprintf(stderr, "Can't happen\n");
        current_live = 0;
        if (bb->bb_next1 != NULL)
            current_live |= bb->bb_next1->flags_live_at_start;
        if (bb->bb_next2 != NULL)
            current_live |= bb->bb_next2->flags_live_at_start;
        }

        do {
        insn_info[iip].flags_live_at_end = current_live;
        current_live &= ~insn_info[iip].flags_set;
        current_live |= insn_info[iip].flags_used;
        } while (iip-- != bb->first_iip);

        if (bb->flags_live_at_start != current_live && !quiet_compile)
        fprintf(stderr, "Fascinating %d!\n", round), changed = 1;
        bb->flags_live_at_start = current_live;
    }
    round++;
    } while (changed);
    return last_iip;
}

#define MAX_JSRS 4096 /* must be a power of two */

static uaecptr jsr_rets[MAX_JSRS];
static struct hash_entry *jsr_hash[MAX_JSRS];
static int jsr_num;
static struct hash_entry dummy_hash; /* This is for safety purposes only */


static void jsr_stack_init(void)
{
    jsr_num = 0;
    dummy_hash.execute = NULL;
}

void compiler_flush_jsr_stack(void)
{
    jsr_num = 0;
}

void m68k_do_rts(void)
{
    m68k_setpc(get_long(m68k_areg(regs, 7)));
    m68k_areg(regs, 7) += 4;
    if (jsr_num > 0)
    jsr_num--;
}

__inline__ void m68k_do_jsr(uaecptr dest)
{
    uae_u32 oldpc = m68k_getpc();
    struct hash_entry *h = find_hash(oldpc);

    if (jsr_num == MAX_JSRS)
    compiler_flush_jsr_stack();
    if (h == NULL) {
    jsr_hash[jsr_num] = &dummy_hash;
    jsr_rets[jsr_num++] = 0xC0DEDBAD;
    } else {
    jsr_hash[jsr_num] = h;
    jsr_rets[jsr_num++] = oldpc;
    }
    m68k_areg(regs, 7) -= 4;
    put_long(m68k_areg(regs, 7), oldpc);
    m68k_setpc(dest);
}

void m68k_do_bsr(uae_s32 offset)
{
    m68k_do_jsr(m68k_getpc() + offset);
}

/* Here starts the actual compiling part */

static char *compile_current_addr;
static char *compile_last_addr;

static __inline__ void assemble(uae_u8 a)
{
    if (compile_current_addr < compile_last_addr) {
    *compile_current_addr++ = a;
    } else {
    compile_failure = 1;
    }
}

static __inline__ void assemble_ulong(uae_u32 a)
{
    assemble(a);
    assemble(a >> 8);
    assemble(a >> 16);
    assemble(a >> 24);
}

static __inline__ void assemble_ulong_68k(uae_u32 a)
{
    assemble(a >> 24);
    assemble(a >> 16);
    assemble(a >> 8);
    assemble(a);
}

static __inline__ void assemble_uword(uae_u16 a)
{
    assemble(a);
    assemble(a >> 8);
}

static __inline__ void assemble_long(void *a)
{
    assemble_ulong((uae_u32)a);
}

static __inline__ void compile_org(char *addr)
{
    compile_current_addr = addr;
}

static __inline__ char *compile_here(void)
{
    return compile_current_addr;
}

#define r_EAX 0
#define r_ECX 1
#define r_EDX 2
#define r_EBX 3
#define r_ESP 4
#define r_EBP 5
#define r_ESI 6
#define r_EDI 7

#define r_AH 0x84
#define r_CH 0x85
#define r_DH 0x86
#define r_BH 0x87

#define ALL_X86_REGS 255
#define ADDRESS_X86_REGS ((1 << r_EBP) | (1 << r_ESI) | (1 << r_EDI))
#define DATA_X86_REGS ((1 << r_EAX) | (1 << r_EDX) | (1 << r_EBX) | (1 << r_ECX))

#define BO_NORMAL 0
#define BO_SWAPPED_LONG 1
#define BO_SWAPPED_WORD 2

struct register_mapping {
    int dreg_map[8], areg_map[8]; /* 68000 register cache */
    int x86_const_offset[8];
    int x86_dirty[8];
    int x86_cache_reg[8]; /* Regs used for the 68000 register cache */
    int x86_cr_type[8]; /* Caching data or address register? */
    int x86_locked[8]; /* Regs used for some purpose */
    int x86_users[8];
    int x86_byteorder[8];
    int x86_verified[8];
};

/*
 * First, code to compile some primitive x86 instructions
 */

static void compile_lea_reg_with_offset(int dstreg, int srcreg, uae_u32 srcoffs)
{
    assemble(0x8D);
    if (srcreg == -2) {
    assemble(0x05 + 8*dstreg);
    assemble_ulong(srcoffs);
    } else if ((uae_s32)srcoffs >= -128 && (uae_s32)srcoffs <= 127) {
    assemble(0x40 + 8*dstreg + srcreg);
    assemble(srcoffs);
    } else {
    assemble(0x80 + 8*dstreg + srcreg);
    assemble_ulong(srcoffs);
    }
}

static void compile_move_reg_reg(int dstreg, int srcreg, wordsizes size)
{
    if (size == sz_byte
    && (((1 << dstreg) & DATA_X86_REGS) == 0
        || ((1 << srcreg) & DATA_X86_REGS) == 0))
    {
    fprintf(stderr, "Moving wrong register types!\n");
    }
    if (size == sz_word)
    assemble(0x66);
    if (size == sz_byte)
    assemble(0x88);
    else
    assemble(0x89);
    assemble(0xC0 + dstreg + 8*srcreg);
}

static void compile_move_between_reg_mem_regoffs(int dstreg, int srcreg,
                         uae_u32 srcoffs, wordsizes size,
                         int code)
{
    if (size == sz_byte && (dstreg & 0x80) != 0)
    dstreg &= ~0x80;
    else if ((size == sz_byte
          && ((1 << dstreg) & DATA_X86_REGS) == 0)
         || (size != sz_byte && (dstreg & 0x80) != 0))
    {
    fprintf(stderr, "Moving wrong register types!\n");
    }
    if (size == sz_word)
    assemble(0x66);
    if (size == sz_byte)
    assemble(code);
    else
    assemble(code + 1);

    if (srcreg == -2) {
    assemble(0x05 + 8*dstreg);
    assemble_ulong(srcoffs);
    } else if ((uae_s32)srcoffs >= -128 && (uae_s32)srcoffs <= 127) {
    assemble(0x40 + 8*dstreg + srcreg);
    assemble(srcoffs);
    } else {
    assemble(0x80 + 8*dstreg + srcreg);
    assemble_ulong(srcoffs);
    }
}

static void compile_move_reg_from_mem_regoffs(int dstreg, int srcreg,
                          uae_u32 srcoffs, wordsizes size)
{
    compile_move_between_reg_mem_regoffs(dstreg, srcreg, srcoffs, size, 0x8A);
}

static void compile_move_reg_to_mem_regoffs(int dstreg, uae_u32 dstoffs,
                        int srcreg, wordsizes size)
{
    compile_move_between_reg_mem_regoffs(srcreg, dstreg, dstoffs, size, 0x88);
}

static void compile_byteswap(int x86r, wordsizes size, int save_flags)
{
    switch(size) {
     case sz_word:
    if (save_flags)
        assemble(0x9C);
    assemble(0x66); /* rolw $8,x86r */
    assemble(0xC1);
    assemble(0xC0 + x86r);
    assemble(8);
    if (save_flags)
        assemble(0x9D);
    break;
     case sz_long:
    assemble(0x0F); /* bswapl x86r */
    assemble(0xC8+x86r);
    break;
     default:
    break;
    }
}

static void compile_force_byteorder(struct register_mapping *map, int x86r,
                    int desired_bo, int save_flags)
{
    if (x86r < 0 || map->x86_byteorder[x86r] == desired_bo)
    return;

    if (map->x86_byteorder[x86r] == BO_SWAPPED_LONG)
    compile_byteswap(x86r, sz_long, save_flags);
    else if (map->x86_byteorder[x86r] == BO_SWAPPED_WORD)
    compile_byteswap(x86r, sz_word, save_flags);

    if (desired_bo == BO_SWAPPED_LONG)
    compile_byteswap(x86r, sz_long, save_flags);
    else if (desired_bo == BO_SWAPPED_WORD)
    compile_byteswap(x86r, sz_word, save_flags);
    map->x86_byteorder[x86r] = desired_bo;
}

/* Add a constant offset to a x86 register. If it's in the cache, make sure
 * we update the const_offset value. The flags are unaffected by this */

static void compile_offset_reg(struct register_mapping *map, int x86r,
                   uae_u32 offset)
{
    int cached_68k;

    if (offset == 0 || x86r == -1 || x86r == -2)
    return;

    compile_force_byteorder(map, x86r, BO_NORMAL, 1);
    cached_68k = map->x86_cache_reg[x86r];
    if (cached_68k != -1) {
    map->x86_const_offset[x86r] -= offset;
    map->x86_dirty[x86r] = 1;
    }
    compile_lea_reg_with_offset(x86r, x86r, offset);
}

static int get_unused_x86_register(struct register_mapping *map)
{
    int x86r;
    for (x86r = 0; x86r < 24; x86r++) {
    if (map->x86_cache_reg[x86r] != -1)
        continue;
    if (map->x86_users[x86r] > 0)
        continue;

    map->x86_verified[x86r] = 0;
    map->x86_byteorder[x86r] = BO_NORMAL;
    return x86r;
    }
    return -1;
}

/*
 * sync_reg() may not touch the flags
 * If may_clobber is 1 and the reg had an offset, the reg will be offsetted
 * by this function
 */
static void sync_reg(struct register_mapping *map, int x86r, void *m68kr,
             uae_u32 offset, int dirty, int may_clobber)
{
    if (dirty || offset != 0)
    compile_force_byteorder(map, x86r, BO_NORMAL, 1);
    if (offset != 0) {
    if (may_clobber) {
        compile_lea_reg_with_offset(x86r, x86r, offset);
        dirty = 1;
    } else {
        int tmpr = get_unused_x86_register(map);
        if (tmpr != -1) {
        compile_lea_reg_with_offset(tmpr, x86r, offset);
        x86r = tmpr;
        dirty = 1;
        } else {
        compile_lea_reg_with_offset(x86r, x86r, offset);
        assemble(0x89);          /* movl x86r,m68kr */
        assemble(0x05 + (x86r << 3));
        assemble_long(m68kr);
        compile_lea_reg_with_offset(x86r, x86r, -offset);
        return;
        }
    }
    }
    if (dirty) {
    assemble(0x89);          /* movl x86r,m68kr */
    assemble(0x05 + (x86r << 3));
    assemble_long(m68kr);
    }
}

static void sync_reg_cache(struct register_mapping *map, int flush)
{
    int i;

    for (i = 0; i < 8; i++) {
    int cr68k = map->x86_cache_reg[i];
    if (cr68k != -1) {
        if (map->x86_cr_type[i] == 1) {
        sync_reg(map, i, regs.regs + cr68k, map->x86_const_offset[i], map->x86_dirty[i], 1);
        if (flush)
            map->dreg_map[cr68k] = -1;
        } else {
        sync_reg(map, i, regs.regs + 8 + cr68k, map->x86_const_offset[i], map->x86_dirty[i], 1);
        if (flush)
            map->areg_map[cr68k] = -1;
        }
        if (flush)
        map->x86_cache_reg[i] = -1;
        map->x86_const_offset[i] = 0;
    }
    }
    memset(map->x86_dirty, 0, sizeof map->x86_dirty);
}

static void remove_x86r_from_cache(struct register_mapping *map, int x86r,
                   int may_clobber)
{
    int j;
    int reg_68k;

    if (x86r == -1)
    return;

    reg_68k = map->x86_cache_reg[x86r];

    if (reg_68k == -1)
    return;

    if (map->x86_cr_type[x86r] == 1) {
    map->dreg_map[reg_68k] = -1;
    sync_reg(map, x86r, regs.regs + reg_68k, map->x86_const_offset[x86r],
         map->x86_dirty[x86r], may_clobber);
    } else {
    map->areg_map[reg_68k] = -1;
    sync_reg(map, x86r, regs.regs + 8 + reg_68k,  map->x86_const_offset[x86r],
         map->x86_dirty[x86r], may_clobber);
    }
    map->x86_dirty[x86r] = 0;
    map->x86_cache_reg[x86r] = -1;
    map->x86_const_offset[x86r] = 0;
    map->x86_verified[x86r] = 0;
    map->x86_byteorder[x86r] = BO_NORMAL;
}

static int get_free_x86_register(struct register_mapping *map,
                 int preferred_mask)
{
    int cnt;
    for (cnt = 0; cnt < 24; cnt++) {
    int x86r = cnt & 7;
    /* In the first two passes, try to get one of the preferred regs */
    if (cnt < 16 && ((1 << x86r) & preferred_mask) == 0)
        continue;
    /* In the first pass, don't discard any registers from the cache */
    if (cnt < 8 && map->x86_cache_reg[x86r] != -1)
        continue;
    /* Never use locked registers */
    if (map->x86_users[x86r] > 0)
        continue;

    remove_x86r_from_cache(map, x86r, 1);
    map->x86_dirty[x86r] = 0;
    map->x86_cache_reg[x86r] = -1;
    map->x86_const_offset[x86r] = 0;
    map->x86_verified[x86r] = 0;
    map->x86_byteorder[x86r] = BO_NORMAL;
    return x86r;
    }
    printf("Out of registers!\n");
    return -1;
}

static int get_typed_x86_register(struct register_mapping *map,
                  int preferred_mask)
{
    int cnt;
    for (cnt = 0; cnt < 16; cnt++) {
    int x86r = cnt & 7;
    /* Get one of the preferred regs */
    if (((1 << x86r) & preferred_mask) == 0)
        continue;
    /* In the first pass, don't discard any registers from the cache */
    if (cnt < 8 && map->x86_cache_reg[x86r] != -1)
        continue;
    /* Never use locked registers */
    if (map->x86_users[x86r] > 0)
        continue;

    remove_x86r_from_cache(map, x86r, 1);
    map->x86_dirty[x86r] = 0;
    map->x86_cache_reg[x86r] = -1;
    map->x86_const_offset[x86r] = 0;
    map->x86_verified[x86r] = 0;
    map->x86_byteorder[x86r] = BO_NORMAL;
    return x86r;
    }
    printf("Out of type registers!\n");
    return -1;
}

static void compile_unlock_reg(struct register_mapping *map, int reg)
{
    if (reg >= 0) {
    if (--map->x86_users[reg] == 0)
        map->x86_locked[reg] = 0;

    }
}

static void lock_reg(struct register_mapping *map, int x86r, int lock_type)
{
#if 1
    switch (map->x86_locked[x86r]) {
     case 0:
    if (map->x86_users[x86r] != 0)
        printf("Users for an unlocked reg!\n");
    break;
     case 1:
    if (lock_type == 2)
        printf("Locking shared reg for exclusive use!\n");
    break;
     case 2:
    printf("Locking exclusive reg!\n");
    break;
     default:
    printf("Unknown lock?\n");
    break;
    }
#endif
    map->x86_locked[x86r] = lock_type;
    map->x86_users[x86r]++;
}

static int get_and_lock_68k_reg(struct register_mapping *map, int reg, int is_dreg,
                int preferred, int no_offset, int lock_type)
{
    int x86r;
    int *regmap;
    uae_u32 *reghome;
    uae_u32 const_off = 0;

    if (reg < 0 || reg > 7) {
    printf("Mad compiler disease\n");
    return 0;
    }

    if (is_dreg)
    regmap = map->dreg_map, reghome = regs.regs;
    else
    regmap = map->areg_map, reghome = regs.regs + 8;

    if (preferred == 0)
    preferred = ALL_X86_REGS;

    x86r = regmap[reg];
    if (x86r == -1) {
    x86r = get_free_x86_register(map, preferred);
    assemble(0x8B); assemble(0x05 + (x86r << 3)); /* movl regs.d[reg],x86r */
    assemble_long(reghome + reg);
    map->x86_cache_reg[x86r] = reg;
    map->x86_cr_type[x86r] = is_dreg;
    map->x86_const_offset[x86r] = 0;
    map->x86_dirty[x86r] = 0;
    map->x86_verified[x86r] = 0;
    map->x86_byteorder[x86r] = BO_NORMAL;
    regmap[reg] = x86r;
    } else {
    const_off = map->x86_const_offset[x86r];

    if (map->x86_locked[x86r] == 2
        || (map->x86_locked[x86r] == 1 && (lock_type == 2 || (const_off != 0 && no_offset))))
    {
        int newr;
        int old_dirty = 0;
        int old_verified;
        int old_bo;

        newr = get_free_x86_register(map, preferred);
        if (const_off == 0) {
        compile_move_reg_reg(newr, x86r, sz_long);
        } else {
        compile_force_byteorder(map, x86r, BO_NORMAL, 1);
        compile_lea_reg_with_offset(newr, x86r, const_off);
        old_dirty = 1;
        const_off = 0;
        }
        /* Remove old reg from cache... */
        map->x86_cache_reg[x86r] = -1;
        map->x86_cr_type[x86r] = is_dreg;
        map->x86_const_offset[x86r] = 0;
        old_dirty |= map->x86_dirty[x86r];
        old_verified = map->x86_verified[x86r];
        old_bo = map->x86_byteorder[x86r];
        map->x86_verified[x86r] = 0;
        map->x86_dirty[x86r] = 0;
        x86r = newr;
        /* ... and make the new one the cache register */
        map->x86_cache_reg[x86r] = reg;
        map->x86_cr_type[x86r] = is_dreg;
        map->x86_const_offset[x86r] = 0;
        map->x86_dirty[x86r] = old_dirty;
        map->x86_verified[x86r] = old_verified;
        map->x86_byteorder[x86r] = old_bo;
        regmap[reg] = x86r;
    }
    }
    if (no_offset && const_off != 0) {
    if (map->x86_locked[x86r] != 0)
        printf("modifying locked reg\n");
    compile_force_byteorder(map, x86r, BO_NORMAL, 1);
    compile_lea_reg_with_offset(x86r, x86r, map->x86_const_offset[x86r]);
    map->x86_const_offset[x86r] = 0;
    map->x86_dirty[x86r] = 1;
    }
    lock_reg(map, x86r, lock_type);
    return x86r;
}

/*
 * Move a constant to a register. Don't do anything if we already have a
 * register, even if it is offset by a constant
 */

static int compile_force_const_reg(struct register_mapping *map, int x86r,
                   uae_u32 *offs, int desired)
{
    int newr = x86r;

    if (newr == -2) {
    if (desired == 0)
        newr = get_free_x86_register(map, ALL_X86_REGS);
    else
        newr = get_typed_x86_register(map, desired);

    assemble(0xB8 + newr);
    assemble_ulong(*offs);
    *offs = 0;
    }
    map->x86_users[newr]++;
    return newr;
}

static void compile_extend_long(struct register_mapping *map, int x86r,
                wordsizes size)
{
    if (x86r < 0) {
    printf("Bad reg in extend_long\n");
    return;
    }

    compile_force_byteorder(map, x86r, BO_NORMAL, 1);

    if (size != sz_long) {
    if (x86r == r_EAX && size == sz_word) {
        assemble(0x98); /* cwtl */
    } else {
        assemble(0x0F);
        if (size == sz_byte) {
        assemble(0xBE);
        } else {
        assemble(0xBF);
        }
        assemble(0xC0 + x86r*9);
    }
    }
}

struct ea_info {
    int reg;
    amodes mode;
    wordsizes size;
    int address_reg;    /* The x86 reg holding the address, or -1 if ea doesn't refer to memory
             * -2 if it refers to memory, but only with a constant address */
    uae_u32 addr_const_off; /* Constant offset to the address */
    int data_reg;         /* The x86 reg that holds the data. -1 if data is not present yet.
               * -2 if data is constant */
    uae_u32 data_const_off;
    int flags;            /* Extra info. Contains the dp field of d8r modes */
    int purpose;
};

static void init_eainfo(struct ea_info *eai)
{
    eai->address_reg = -1;
    eai->addr_const_off = 0;
    eai->data_reg = -1;
    eai->data_const_off = 0;
}

struct insn_reg_needs {
    int checkpoint_no;
    int dreg_needed[8], areg_needed[8];
    int dreg_mask[8], areg_mask[8];
};

/*
 * This structure holds information about predec/postinc addressing modes.
 */

struct pid_undo {
    int used;
    int x86r[2];
    int m68kr[2];
    int dirty[2];
    int offs[2];
};

static void add_undo(struct pid_undo *pud, int x86r, int m68kr, int offs,
             int dirty)
{
    int i;
    for (i = 0; i < pud->used; i++)
    if (pud->m68kr[i] == m68kr)
        return;
    pud->m68kr[i] = m68kr;
    pud->x86r[i] = x86r;
    pud->offs[i] = offs;
    pud->dirty[i] = dirty;
    pud->used++;
}

/*
 * Lock previous contents of address registers used in predec/postinc modes
 * for generate_possible_exit().
 */

static void compile_prepare_undo(struct register_mapping *map, amodes mode,
                  int reg, struct pid_undo *pud)
{
    int x86r;

    switch(mode){
     default:
    break;

     case Apdi:
    x86r = get_and_lock_68k_reg(map, reg, 0, ADDRESS_X86_REGS, 0, 1);
    /* This saves recording the byteorder in the pud structure, and we'll
     * need it in normal byteorder anyway */
    compile_force_byteorder(map, x86r, BO_NORMAL, 0);
    /*
     * Add this reg with its current offset to the undo buffer.
     * Since we have locked it, we are certain that it will not be
     * modified.
     */
    add_undo(pud, x86r, reg, map->x86_const_offset[x86r], map->x86_dirty[x86r]);
    break;

     case Aipi:
    x86r = get_and_lock_68k_reg(map, reg, 0, ADDRESS_X86_REGS, 0, 1);
    compile_force_byteorder(map, x86r, BO_NORMAL, 0);
    add_undo(pud, x86r, reg, map->x86_const_offset[x86r], map->x86_dirty[x86r]);
    break;
    }
}

/*
 * Load all the registers absolutely needed to calculate and verify thea
 * address. Load other registers if convenient.
 * This contains a fair amount of magic to get the register cache working right.
 */

static void compile_prepareea(struct register_mapping *map, amodes mode,
                  int reg, wordsizes size, uae_u8 **pcpp, uaecptr pca,
                  struct ea_info *eainf, int eaino, int ea_purpose,
                  int pidmult)
{
    struct ea_info *eai = eainf + eaino;
    int pdival = size == sz_byte && reg != 7 ? 1 : size == sz_long ? 4 : 2;
    uae_u8 *p = *pcpp;
    uae_u16 dp;
    int r;
    int x86r, tmpr;

    pdival *= pidmult;

    init_eainfo(eai);
    eai->mode = mode;
    eai->size = size;
    eai->reg = reg;

    switch(mode){
     case Dreg:
     case Areg:
    break;

     case Ad16:
    eai->addr_const_off = (uae_s16)do_get_mem_word((uae_u16 *)p);
    (*pcpp) += 2; p += 2;
    x86r = eai->address_reg = get_and_lock_68k_reg(map, reg, 0, ADDRESS_X86_REGS, 0, 1);
    compile_force_byteorder(map, x86r, BO_NORMAL, 0);
    eai->addr_const_off += map->x86_const_offset[x86r];
    break;

     case Aind:
    x86r = eai->address_reg = get_and_lock_68k_reg(map, reg, 0, ADDRESS_X86_REGS, 0, 1);
    compile_force_byteorder(map, x86r, BO_NORMAL, 0);
    eai->addr_const_off = map->x86_const_offset[x86r];
    break;

     case Apdi:
    x86r = eai->address_reg = get_and_lock_68k_reg(map, reg, 0, ADDRESS_X86_REGS, 0, 1);
    compile_force_byteorder(map, x86r, BO_NORMAL, 0);
    map->x86_const_offset[x86r] -= pdival;
    eai->addr_const_off = map->x86_const_offset[x86r];
    break;

     case Aipi:
    x86r = eai->address_reg = get_and_lock_68k_reg(map, reg, 0, ADDRESS_X86_REGS, 0, 1);
    compile_force_byteorder(map, x86r, BO_NORMAL, 0);
    eai->addr_const_off = map->x86_const_offset[x86r];
    map->x86_const_offset[x86r] += pdival;
    break;

     case Ad8r:
    dp = (uae_s16)do_get_mem_word((uae_u16 *)p);
    r = (dp & 0x7000) >> 12;
    (*pcpp) += 2; p += 2;

    tmpr = get_and_lock_68k_reg(map, reg, 0, ADDRESS_X86_REGS, 0, 1);
    compile_force_byteorder(map, tmpr, BO_NORMAL, 0);
    eai->addr_const_off = map->x86_const_offset[tmpr] + (uae_s8)dp;

    if (dp & 0x800) {
        x86r = get_and_lock_68k_reg(map, r, dp & 0x8000 ? 0 : 1, ADDRESS_X86_REGS, 0, 2);
        remove_x86r_from_cache(map, x86r, 0);
        compile_force_byteorder(map, x86r, BO_NORMAL, 0);
        eai->addr_const_off += map->x86_const_offset[x86r];
    } else {
        x86r = get_and_lock_68k_reg(map, r, dp & 0x8000 ? 0 : 1, ADDRESS_X86_REGS, 1, 2);
        remove_x86r_from_cache(map, x86r, 0);
        compile_force_byteorder(map, x86r, BO_NORMAL, 0);
    }
    eai->address_reg = x86r;

    r = (dp & 0x7000) >> 12;

    if (dp & 0x800) {
        if (eai->addr_const_off == 0) {
        assemble(0x03); assemble(0xC0 + tmpr + x86r*8); /* addl basereg,addrreg */
        } else if ((uae_s32)eai->addr_const_off >= -128 && (uae_s32)eai->addr_const_off <= 127) {
        assemble(0x8D);
        assemble(0x44 + x86r*8); /* leal disp8(dispreg,basereg),dispreg */
        assemble(x86r*8 + tmpr);
        assemble(eai->addr_const_off);
        } else {
        assemble(0x8D);
        assemble(0x84 + x86r*8); /* leal disp32(dispreg,basereg),dispreg */
        assemble(x86r*8 + tmpr);
        assemble_ulong(eai->addr_const_off);
        }
        eai->addr_const_off = 0;
    } else {
        assemble(0x0F); assemble(0xBF);
        assemble(0xC0 + x86r*9); /* movswl dispreg,addrreg */
        assemble(0x03); assemble(0xC0 + tmpr + x86r*8); /* addl basereg,addrreg */
    }
    compile_unlock_reg(map, tmpr);
    break;

     case PC8r:
    dp = (uae_s16)do_get_mem_word((uae_u16 *)p);
    (*pcpp) += 2; p += 2;
    r = (dp & 0x7000) >> 12;
    eai->addr_const_off = pca + (uae_s8)dp;
    if (dp & 0x800) {
        x86r = get_and_lock_68k_reg(map, r, dp & 0x8000 ? 0 : 1, ADDRESS_X86_REGS, 0, 1);
        remove_x86r_from_cache(map, x86r, 0);
        compile_force_byteorder(map, x86r, BO_NORMAL, 0);
        eai->addr_const_off += map->x86_const_offset[x86r];
    } else {
        x86r = get_and_lock_68k_reg(map, r, dp & 0x8000 ? 0 : 1, ADDRESS_X86_REGS, 1, 2);
        remove_x86r_from_cache(map, x86r, 0);
        compile_force_byteorder(map, x86r, BO_NORMAL, 0);

        assemble(0x0F); assemble(0xBF);
        assemble(0xC0 + x86r*9); /* movswl dispreg,addrreg */
    }
    eai->address_reg = x86r;
    break;

     case PC16:
    eai->addr_const_off = pca + (uae_s16)do_get_mem_word((uae_u16 *)p);
    eai->address_reg = -2;
    (*pcpp) += 2; p += 2;
    break;

     case absw:
    eai->addr_const_off = (uae_s16)do_get_mem_word((uae_u16 *)p);
    eai->address_reg = -2;
    (*pcpp) += 2; p += 2;
    break;

     case absl:
    eai->addr_const_off = (uae_s32)do_get_mem_long((uae_u32 *)p);
    eai->address_reg = -2;
    (*pcpp) += 4; p += 4;
    break;

     case imm:
    if (size == sz_long)
        goto imm2_const;
    if (size == sz_word)
        goto imm1_const;

    /* fall through */
     case imm0:
    eai->data_const_off = (uae_s8)*(p+1);
    eai->data_reg = -2;
    (*pcpp) += 2; p += 2;
    break;

     case imm1:
    imm1_const:
    eai->data_const_off = (uae_s16)do_get_mem_word((uae_u16 *)p);
    eai->data_reg = -2;
    (*pcpp) += 2; p += 2;
    break;

     case imm2:
    imm2_const:
    eai->data_const_off = (uae_s32)do_get_mem_long((uae_u32 *)p);
    eai->data_reg = -2;
    (*pcpp) += 4; p += 4;
    break;

     case immi:
    eai->data_const_off = (uae_s8)reg;
    eai->data_reg = -2;
    break;

     default:
    break;
    }
    eai->purpose = ea_purpose;
}

static void compile_get_excl_lock(struct register_mapping *map, struct ea_info *eai)
{
    int x86r = eai->data_reg;

    if (x86r >= 0 && map->x86_locked[x86r] == 1) {
    int newr;
    if (eai->size == sz_byte)
        newr = get_typed_x86_register(map, DATA_X86_REGS);
    else
        newr = get_free_x86_register(map, ALL_X86_REGS);

    compile_move_reg_reg(newr, x86r, sz_long);
    eai->data_reg = newr;
    lock_reg(map, eai->data_reg, 2);
    }
}

/*
 * Some functions to assemble some 386 opcodes which have a similar
 * structure (ADD, AND, OR, etc.). These take source and destination
 * addressing modes, check their validity and assemble a complete
 * 386 instruction.
 */

static __inline__ int rmop_long(struct ea_info *eai)
{
    if (eai->data_reg == -2)
    printf("rmop for const\n");
    else if (eai->data_reg == -1) {
    if (eai->address_reg == -2)
        return 5;
    if (eai->address_reg == -1) {
        /* This must be a 68k register in its home location */
        return 5;
    }
#if 0 /* We need to add address_space... */
    if (eai->addr_const_off == 0 && eai->address_reg != r_EBP) {
        return eai->address_reg;
    }
    else if ((uae_s32)eai->addr_const_off >= -128 && (uae_s32)eai->addr_const_off <= 127) {
        return eai->address_reg | 0x40;
    }
#endif
    return eai->address_reg | 0x80;
    } else {
    if (eai->size == sz_byte && ((1 << eai->data_reg) & DATA_X86_REGS) == 0)
        printf("wrong type reg in rmop\n");
    if (eai->data_const_off != 0)
        printf("data_const_off in rmop\n");
    return 0xC0 + eai->data_reg;
    }
    return 0;
}

static __inline__ int rmop_short(struct ea_info *eai)
{
    if (eai->data_reg == -2)
    printf("rmop_short for const\n");
    else if (eai->data_reg == -1) {
    printf("rmop_short for mem\n");
    } else {
    if (eai->size == sz_byte && ((1 << eai->data_reg) & DATA_X86_REGS) == 0)
        printf("wrong type reg in rmop_short\n");
    if (eai->data_const_off != 0)
        printf("data_const_off in rmop_short\n");
    return eai->data_reg*8;
    }
    return 0;
}

static __inline__ void rmop_finalize(struct ea_info *eai)
{
    if (eai->data_reg == -2)
    assemble_ulong(eai->data_const_off);
    else if (eai->data_reg == -1) {
    if (eai->address_reg == -2)
        /* Constant address */
        assemble_long(address_space + (uae_s32)eai->addr_const_off);
    else if (eai->address_reg == -1) {
        /* Register in its home location */
        if (eai->mode == Areg)
        assemble_long(regs.regs + 8  + eai->reg);
        else
        assemble_long(regs.regs + eai->reg);
    } else {
#if 0
        /* Indirect address with offset */
        if ((uae_s32)eai->addr_const_off >= -128 && (uae_s32)eai->addr_const_off <= 127) {
        }
#endif
        assemble_long(address_space + (uae_s32)eai->addr_const_off);
    }
    }
}

static void compile_eas(struct register_mapping *map, struct ea_info *eainf, int eaino_s, int eaino_d,
            int optype)
{
    struct ea_info *eais = eainf + eaino_s;
    struct ea_info *eaid = eainf + eaino_d;
    int szflag = eais->size == sz_byte ? 0 : 1;
    int swapflag = 0;
    int opcode;

    if (eais->data_reg == -1) {
    compile_force_byteorder(map, eais->address_reg, BO_NORMAL, 0);
    eais = eainf + eaino_d;
    eaid = eainf + eaino_s;
    swapflag = 1;
    }
    if (eais->data_reg == -1) {
    compile_force_byteorder(map, eais->address_reg, BO_NORMAL, 0);
    }

    if (eais->size == sz_word)
    assemble(0x66);

    if (eais->data_reg == -2) {
    assemble(0x80+szflag);
    assemble(8*optype | rmop_long(eaid));
    rmop_finalize(eaid);
    switch(eais->size) {
     case sz_byte: assemble(eais->data_const_off); break;
     case sz_word: assemble_uword(eais->data_const_off); break;
     case sz_long: assemble_ulong(eais->data_const_off); break;
    }
    } else {
    assemble(8*optype | szflag | 2*swapflag);
    assemble(rmop_long(eaid) | rmop_short(eais));
    rmop_finalize(eaid);
    }
}

static void compile_fetchmem(struct register_mapping *map, struct ea_info *eai)
{
    int x86r;
    if (eai->size == sz_byte)
    x86r = get_typed_x86_register(map, DATA_X86_REGS);
    else
    x86r = get_free_x86_register(map, ALL_X86_REGS);

    lock_reg(map, x86r, 2);
    compile_force_byteorder(map, eai->address_reg, BO_NORMAL, 0);
    compile_move_reg_from_mem_regoffs(x86r, eai->address_reg,
                      (uae_u32)(eai->addr_const_off + address_space),
                      eai->size);
    map->x86_verified[x86r] = 0;
    switch (eai->size) {
     case sz_byte: map->x86_byteorder[x86r] = BO_NORMAL; break;
     case sz_word: map->x86_byteorder[x86r] = BO_SWAPPED_WORD; break;
     case sz_long: map->x86_byteorder[x86r] = BO_SWAPPED_LONG; break;
    }
    eai->data_reg = x86r;
    eai->data_const_off = 0;
}

static void compile_fetchimm(struct register_mapping *map, struct ea_info *eai, int byteorder)
{
    int x86r;
    if (eai->size == sz_byte)
    x86r = get_typed_x86_register(map, DATA_X86_REGS);
    else
    x86r = get_free_x86_register(map, ALL_X86_REGS);

    switch (byteorder) {
     case BO_SWAPPED_LONG:
    eai->data_const_off = (((eai->data_const_off & 0xFF000000) >> 24)
                   | ((eai->data_const_off & 0xFF0000) >> 8)
                   | ((eai->data_const_off & 0xFF00) << 8)
                   | ((eai->data_const_off & 0xFF) << 24));
    break;
     case BO_SWAPPED_WORD:
    eai->data_const_off = (((eai->data_const_off & 0xFF00) >> 8)
                   | ((eai->data_const_off & 0xFF) << 8)
                   | (eai->data_const_off & 0xFFFF0000));
    break;
     case BO_NORMAL:
    break;
    }
    lock_reg(map, x86r, 2);
    map->x86_byteorder[x86r] = byteorder; map->x86_verified[x86r] = 0;

    switch (eai->size) {
     case sz_byte: assemble(0xC6); assemble(0xC0 + x86r); assemble(eai->data_const_off); break;
     case sz_word: assemble(0x66); assemble(0xC7); assemble(0xC0 + x86r); assemble_uword(eai->data_const_off); break;
     case sz_long: assemble(0xC7); assemble(0xC0 + x86r); assemble_ulong(eai->data_const_off); break;
    }
    eai->data_reg = x86r;
    eai->data_const_off = 0;
}

/*
 * 1: reg
 * 2: mem
 * 4: imm
 */

static int binop_alternatives[] = {
    7, 1,
    5, 3,
    0, 0
};

static int binop_worda_alternatives[] = {
    1, 3,
    0, 0
};

static int regonly_alternatives[] = {
    1, 1,
    0, 0
};

static void compile_loadeas(struct register_mapping *map, struct ea_info *eainf,
               int eaino_s, int eaino_d, int *alternatives,
               int scramble_poss, int load_dest)
{
    struct ea_info *eais = eainf + eaino_s;
    struct ea_info *eaid = eainf + eaino_d;
    int scrambled_bo = eaid->size == sz_long ? BO_SWAPPED_LONG : eaid->size == sz_word ? BO_SWAPPED_WORD : BO_NORMAL;
    int i, scrambled = 0;
    int best = 0;
    int bestcost = -1;
    int *ap;
    uae_u32 *sregp = NULL, *dregp = NULL;
    int screg = -1, dcreg = -1;
    int stype = -1, dtype = -1;
    int asrc, adst;
    int regprefs = eais->size == sz_byte ? DATA_X86_REGS : 0;

    if (eais->mode == Dreg) {
    stype = 0;
    screg = map->dreg_map[eais->reg];
    if (screg == -1)
        sregp = regs.regs + eais->reg;
    } else if (eais->mode == Areg) {
    stype = 0;
    screg = map->areg_map[eais->reg];
    if (screg == -1)
        sregp = regs.regs + 8 + eais->reg;
    } else if (eais->data_reg == -2) {
    stype = -2;
    }

    if (eaid->mode == Dreg) {
    dtype = 0;
    dcreg = map->dreg_map[eaid->reg];
    if (dcreg == -1)
        dregp = regs.regs + eaid->reg;
    } else if (eaid->mode == Areg) {
    dtype = 0;
    dcreg = map->areg_map[eaid->reg];
    if (dcreg == -1)
        dregp = regs.regs + 8 + eaid->reg;
    } else if (eaid->data_reg == -2) {
    dtype = -2;
    }

    ap = alternatives;

    for (i = 0;; i++) {
    int cost = 0;

    asrc = *ap++;
    if (asrc == 0)
        break;
    adst = *ap++;

    if (stype == -2 && (asrc & 4) == 0)
        cost++;
    else if (stype == -1 && ((asrc & 2) == 0 || (eais->size != sz_byte && !scramble_poss)))
        cost++;
    else if (stype == 0 && screg == -1 && (asrc & 2) == 0)
        cost++;

    if (dtype == -1 && ((adst & 2) == 0 || (eaid->size != sz_byte && !scramble_poss)))
        /* The !load_dest case isn't handled by the current code,
         * and it isn't desirable anyway. Use a different alternative
         */
        cost += load_dest ? 1 : 100;
    else if (dtype == 0 && dcreg == -1 && (adst & 2) == 0)
        cost++;

    if (bestcost == -1 || cost < bestcost) {
        bestcost = cost;
        best = i;
    }
    }

    asrc = alternatives[2*best];
    adst = alternatives[2*best+1];

    if (dtype == -1) {
    if (load_dest) {
        if ((adst & 2) == 0 || (eaid->size != sz_byte && !scramble_poss))
        compile_fetchmem(map, eaid);
    } else {
        if ((adst & 2) == 0) {
        printf("Not loading memory operand. Prepare to die.\n");
        if (eaid->size == sz_byte)
            eaid->data_reg = get_typed_x86_register(map, DATA_X86_REGS);
        else
            eaid->data_reg = get_free_x86_register(map, ALL_X86_REGS);
        }
    }
    /* Scrambled in both mem and reg cases */
    if (eaid->size != sz_byte && scramble_poss)
        scrambled = 1;
    } else {
    if (dcreg == -1 && !load_dest && (adst & 2) == 0 && eaid->size == sz_long) {
        /* We need a register, but we don't need to fetch the old data.
         * See storeea for some more code handling this case. This first
         * if statement could be eliminated, we would generate some
         * superfluous moves. This is an optimization. If it were not
         * done, the mem-mem-move warning could be commented in in
         * storeea. */
        if (eaid->size == sz_byte)
        eaid->data_reg = get_typed_x86_register(map, DATA_X86_REGS);
        else
        eaid->data_reg = get_free_x86_register(map, ALL_X86_REGS);
        eaid->data_const_off = 0;
    } else if ((dcreg == -1 && (adst & 2) == 0) || dcreg != -1) {
        int reg_bo;
        eaid->data_reg = get_and_lock_68k_reg(map, eaid->reg, eaid->mode == Dreg, regprefs, 1, 2);
        eaid->data_const_off = 0;

        reg_bo = map->x86_byteorder[eaid->data_reg];

        if (reg_bo != BO_NORMAL) {
        if (reg_bo != scrambled_bo)
            compile_force_byteorder(map, eaid->data_reg, BO_NORMAL, 0);
        else if (scramble_poss)
            scrambled = 1;
        }
    }
    }

    if (stype == -2) {
    /* @@@ may need to scramble imm, this is a workaround */
    if ((asrc & 4) == 0 || scrambled)
        compile_fetchimm(map, eais, scrambled ? scrambled_bo : BO_NORMAL);
    } else if (stype == -1) {
    if ((asrc & 2) == 0 || (eais->size != sz_byte && !scrambled))
        compile_fetchmem(map, eais);
    } else {
    if ((screg == -1 && (asrc & 2) == 0) || screg != -1) {
        eais->data_reg = get_and_lock_68k_reg(map, eais->reg, eais->mode == Dreg, regprefs, 1, 2);
        eais->data_const_off = 0;
    }
    }

    /* Optimization */
    if (scrambled && eais->data_reg >= 0 && !load_dest
    && map->x86_byteorder[eais->data_reg] == BO_NORMAL
    && eaid->size == sz_long && dtype == 0)
    scrambled = 0;

    if (regprefs != 0 && eais->data_reg >= 0 && ((1 << eais->data_reg) & regprefs) == 0) {
    int tmpr = get_typed_x86_register(map, regprefs);
    compile_move_reg_reg(tmpr, eais->data_reg, sz_long);
    eais->data_reg = tmpr;
    }

    if (regprefs != 0 && eaid->data_reg >= 0 && ((1 << eaid->data_reg) & regprefs) == 0) {
    int tmpr = get_typed_x86_register(map, regprefs);
    compile_move_reg_reg(tmpr, eaid->data_reg, sz_long);
    eaid->data_reg = tmpr;
    }

    /* Now set the byteorder once and for all (should already be correct for
     * most cases) */
    if (scrambled) {
    if (eaid->data_reg >= 0)
        compile_force_byteorder(map, eaid->data_reg, scrambled_bo, 0);
    if (eais->data_reg >= 0)
        compile_force_byteorder(map, eais->data_reg, scrambled_bo, 0);
    } else {
    if (eaid->data_reg >= 0)
        compile_force_byteorder(map, eaid->data_reg, BO_NORMAL, 0);
    if (eais->data_reg >= 0)
        compile_force_byteorder(map, eais->data_reg, BO_NORMAL, 0);
    }
}

static void compile_fetchea(struct register_mapping *map, struct ea_info *eainf,
                int eaino, int asrc)
{
    struct ea_info *eais = eainf + eaino;
    int scrambled_bo = eais->size == sz_long ? BO_SWAPPED_LONG : eais->size == sz_word ? BO_SWAPPED_WORD : BO_NORMAL;
    int i, scrambled = 0;
    int best = 0;
    int bestcost = -1;
    int *ap;
    uae_u32 *sregp = NULL;
    int screg = -1, stype = -1;
    int regprefs = eais->size == sz_byte ? DATA_X86_REGS : 0;

    if (eais->mode == Dreg) {
    stype = 0;
    screg = map->dreg_map[eais->reg];
    if (screg == -1)
        sregp = regs.regs + eais->reg;
    } else if (eais->mode == Areg) {
    stype = 0;
    screg = map->areg_map[eais->reg];
    if (screg == -1)
        sregp = regs.regs + 8 + eais->reg;
    } else if (eais->data_reg == -2) {
    stype = -2;
    }

    if (stype == -2) {
    if ((asrc & 4) == 0)
        compile_fetchimm(map, eais, scrambled ? scrambled_bo : BO_NORMAL);
    } else if (stype == -1) {
    if ((asrc & 2) == 0 || eais->size != sz_byte)
        compile_fetchmem(map, eais);
    } else {
    if ((screg == -1 && (asrc & 2) == 0) || screg != -1) {
        eais->data_reg = get_and_lock_68k_reg(map, eais->reg, eais->mode == Dreg, regprefs, 1, 2);
        eais->data_const_off = 0;
    }
    }

    if (eais->data_reg >= 0)
    compile_force_byteorder(map, eais->data_reg, BO_NORMAL, 0);
}

/*
 * compile_note_modify() should be called on destination EAs obtained from
 * compile_loadeas(), if their value was modified (e.g. by the compile_eas()
 * function)
 */

static void compile_note_modify(struct register_mapping *map, struct ea_info *eainf,
                int eaino)
{
    struct ea_info *eai = eainf + eaino;
    int newr;
    int szflag = eai->size == sz_byte ? 0 : 1;

    if (eai->mode == Dreg) {
    /* We only need to do something if we have the value in a register,
     * otherwise, the home location was modified already */
    if (eai->data_reg >= 0) {
        if (eai->data_reg != map->dreg_map[eai->reg]) {
        remove_x86r_from_cache(map, eai->data_reg, 0);
        if (map->dreg_map[eai->reg] >= 0)
            remove_x86r_from_cache(map, map->dreg_map[eai->reg], 0);
        map->x86_cache_reg[eai->data_reg] = eai->reg;
        map->x86_cr_type[eai->data_reg] = 1;
        map->dreg_map[eai->reg] = eai->data_reg;
        }
        map->x86_verified[eai->data_reg] = 0;
        map->x86_const_offset[eai->data_reg] = eai->data_const_off;
        map->x86_dirty[eai->data_reg] = 1;
    }
    return;
    } else if (eai->mode == Areg) {
    if (eai->size != sz_long)
        printf("Areg put != long\n");

    /* We only need to do something if we have the value in a register,
     * otherwise, the home location was modified already */
    if (eai->data_reg >= 0) {
        if (eai->data_reg != map->areg_map[eai->reg]) {
        remove_x86r_from_cache(map, eai->data_reg, 0);
        if (map->areg_map[eai->reg] >= 0)
            remove_x86r_from_cache(map, map->areg_map[eai->reg], 0);
        map->x86_cache_reg[eai->data_reg] = eai->reg;
        map->x86_cr_type[eai->data_reg] = 0;
        map->areg_map[eai->reg] = eai->data_reg;
        }
        map->x86_verified[eai->data_reg] = 0;
        map->x86_const_offset[eai->data_reg] = eai->data_const_off;
        map->x86_dirty[eai->data_reg] = 1;
    }
    return;
    } else {
    /* Storing to memory from reg? */
    if (eai->data_reg >= 0) {
        compile_offset_reg(map, eai->data_reg, eai->data_const_off);

        switch (eai->size) {
         case sz_byte: compile_force_byteorder(map, eai->data_reg, BO_NORMAL, 1); break;
         case sz_word: compile_force_byteorder(map, eai->data_reg, BO_SWAPPED_WORD, 1); break;
         case sz_long: compile_force_byteorder(map, eai->data_reg, BO_SWAPPED_LONG, 1); break;
        }
        compile_force_byteorder(map, eai->address_reg, BO_NORMAL, 0);
        compile_move_reg_to_mem_regoffs(eai->address_reg,
                        (uae_u32)(eai->addr_const_off + address_space),
                        eai->data_reg, eai->size);
    }
    }
}

static void compile_storeea(struct register_mapping *map, struct ea_info *eainf,
                int eaino_s, int eaino_d)
{
    struct ea_info *eais = eainf + eaino_s;
    struct ea_info *eaid = eainf + eaino_d;
    int newr, cacher;
    int szflag = eaid->size == sz_byte ? 0 : 1;

    if (eaid->mode == Dreg) {
    /* Is the reg to move from already the register cache reg for the
     * destination? */
    if (eais->data_reg >= 0 && eais->data_reg == map->dreg_map[eaid->reg]) {
        map->x86_dirty[eais->data_reg] = 1; map->x86_verified[eais->data_reg] = 0;
        map->x86_const_offset[eais->data_reg] = eais->data_const_off;
        return;
    }
    /* Is the destination register in its home location? */
    if (map->dreg_map[eaid->reg] < 0) {
        if (eais->data_reg == -2) {
        /* Move immediate to regs.regs */
        if (eaid->size == sz_word) assemble(0x66);
        assemble(0xC6 + szflag); assemble(0x05); assemble_long(regs.regs + eaid->reg);
        switch (eaid->size) {
         case sz_byte: assemble(eais->data_const_off); break;
         case sz_word: assemble_uword(eais->data_const_off); break;
         case sz_long: assemble_ulong(eais->data_const_off); break;
        }
        } else if (eais->data_reg == -1) {
#if 0
        printf("Shouldn't happen (mem-mem-move)\n");
#endif
        /* This _can_ happen: move.l $4,d0, if d0 isn't in the
         * cache, will come here. But a reg will be allocated for
         * dest. We use this. This _really_ shouldn't happen if
         * the size isn't long. */
        if (eaid->size != sz_long)
            printf("_Really_ shouldn't happen (Dreg case)\n");
        map->x86_cache_reg[eaid->data_reg] = eaid->reg;
        map->x86_cr_type[eaid->data_reg] = 1;
        map->x86_const_offset[eaid->data_reg] = eaid->data_const_off;
        map->dreg_map[eaid->reg] = eaid->data_reg;
        map->x86_verified[eaid->data_reg] = 0;
        goto have_cache_reg_d;
        } else {
        if (eais->size == sz_long) {
            /* Make this the new register cache reg */
            remove_x86r_from_cache(map, eais->data_reg, 0);
            map->x86_cache_reg[eais->data_reg] = eaid->reg;
            map->x86_cr_type[eais->data_reg] = 1;
            map->x86_const_offset[eais->data_reg] = eais->data_const_off;
            map->dreg_map[eaid->reg] = eais->data_reg;
            map->x86_verified[eais->data_reg] = 0;
        } else {
            /* Move from reg to regs.regs */
            compile_force_byteorder(map, eais->data_reg, BO_NORMAL, 1);
            compile_offset_reg (map, eais->data_reg, eais->data_const_off);
            if (eaid->size == sz_word) assemble(0x66);
            assemble(0x88 + szflag); assemble(0x05 + 8*eais->data_reg);
            assemble_long(regs.regs + eaid->reg);
        }
        }
    } else {
        int destr;

        have_cache_reg_d:

        destr = map->dreg_map[eaid->reg];
        if (eaid->size != sz_long)
        compile_force_byteorder(map, destr, BO_NORMAL, 1);

        if (eais->data_reg == -2) {
        /* Move immediate to reg */
        if (eaid->size == sz_word) assemble(0x66);
        assemble(0xC6 + szflag); assemble(0xC0 + destr);
        switch (eaid->size) {
         case sz_byte: assemble(eais->data_const_off); break;
         case sz_word: assemble_uword(eais->data_const_off); break;
         case sz_long: assemble_ulong(eais->data_const_off); break;
        }
        /* normal byteorder comes either from force above or from long
         * const move */
        map->x86_byteorder[destr] = BO_NORMAL;
        } else if (eais->data_reg == -1) {
        if (eais->mode == Dreg) {
            compile_move_reg_from_mem_regoffs(destr, -2, (uae_u32)(regs.regs + eais->reg),
                              eais->size);
            map->x86_byteorder[destr] = BO_NORMAL;
        } else if (eais->mode == Areg) {
            compile_move_reg_from_mem_regoffs(destr, -2, (uae_u32)(regs.regs + 8 + eais->reg),
                              eais->size);
            map->x86_byteorder[destr] = BO_NORMAL;
        } else {
            /* Move mem to reg */
            compile_force_byteorder(map, eais->address_reg, BO_NORMAL, 0);
            compile_move_reg_from_mem_regoffs(destr, eais->address_reg,
                              (uae_u32)(eais->addr_const_off + address_space),
                              eais->size);

            switch (eais->size) {
             case sz_byte: map->x86_byteorder[destr] = BO_NORMAL; break;
             case sz_word: map->x86_byteorder[destr] = BO_SWAPPED_WORD; break;
             case sz_long: map->x86_byteorder[destr] = BO_SWAPPED_LONG; break;
            }
        }
        } else {
        if (eais->size == sz_long) {
            /* Make this the new register cache reg */
            remove_x86r_from_cache(map, eais->data_reg, 0);
            remove_x86r_from_cache(map, destr, 0);
            map->x86_cache_reg[eais->data_reg] = eaid->reg;
            map->x86_cr_type[eais->data_reg] = 1;
            map->x86_const_offset[eais->data_reg] = eais->data_const_off;
            map->dreg_map[eaid->reg] = eais->data_reg;
            map->x86_verified[eais->data_reg] = 0;
        } else {
            /* Move from reg to reg */
            compile_force_byteorder(map, eais->data_reg, BO_NORMAL, 1);
            compile_offset_reg (map, eais->data_reg, eais->data_const_off);
            if (eaid->size == sz_word) assemble(0x66);
            assemble(0x88 + szflag); assemble(0xC0 + destr + 8*eais->data_reg);
        }
        }
    }

    if (map->dreg_map[eaid->reg] >= 0)
        map->x86_dirty[map->dreg_map[eaid->reg]] = 1;
    return;
    } else if (eaid->mode == Areg) {
    if (eaid->size != sz_long)
        printf("Areg put != long\n");

    /* Is the reg to move from already the register cache reg for the
     * destination? */
    if (eais->data_reg >= 0 && eais->data_reg == map->areg_map[eaid->reg]) {
        map->x86_dirty[eais->data_reg] = 1; map->x86_verified[eais->data_reg] = 0;
        map->x86_const_offset[eais->data_reg] = eais->data_const_off;
        return;
    }
    /* Is the destination register in its home location? */
    if (map->areg_map[eaid->reg] < 0) {
        if (eais->data_reg == -2) {
        /* Move immediate to regs.regs */
        assemble(0xC7); assemble(0x05); assemble_long(regs.regs + 8 + eaid->reg);
        assemble_ulong(eais->data_const_off);
        } else if (eais->data_reg == -1) {
#if 0 /* see above... */
        printf("Shouldn't happen (mem-mem-move)\n");
#endif
        map->x86_cache_reg[eaid->data_reg] = eaid->reg;
        map->x86_cr_type[eaid->data_reg] = 0;
        map->x86_const_offset[eaid->data_reg] = eaid->data_const_off;
        map->areg_map[eaid->reg] = eaid->data_reg;
        map->x86_verified[eaid->data_reg] = 0;
        goto have_cache_reg_a;
        } else {
        /* Make this the new register cache reg */
        remove_x86r_from_cache(map, eais->data_reg, 0);
        map->x86_cache_reg[eais->data_reg] = eaid->reg;
        map->x86_cr_type[eais->data_reg] = 0;
        map->x86_const_offset[eais->data_reg] = eais->data_const_off;
        map->areg_map[eaid->reg] = eais->data_reg;
        map->x86_verified[eais->data_reg] = 0;
        }
    } else {
        int destr;

        have_cache_reg_a:

        destr = map->areg_map[eaid->reg];
        if (eaid->size != sz_long)
        compile_force_byteorder(map, destr, BO_NORMAL, 1);

        if (eais->data_reg == -2) {
        /* Move immediate to reg */
        assemble(0xC7); assemble(0xC0 + destr);
        assemble_ulong(eais->data_const_off);

        /* normal byteorder comes either from force above or from long
         * const move */
        map->x86_byteorder[destr] = BO_NORMAL;
        } else if (eais->data_reg == -1) {
        if (eais->mode == Dreg) {
            compile_move_reg_from_mem_regoffs(destr, -2, (uae_u32)(regs.regs + eais->reg),
                              eais->size);
            map->x86_byteorder[destr] = BO_NORMAL;
        } else if (eais->mode == Areg) {
            compile_move_reg_from_mem_regoffs(destr, -2, (uae_u32)(regs.regs + 8 + eais->reg),
                              eais->size);
            map->x86_byteorder[destr] = BO_NORMAL;
        } else {
            /* Move mem to reg */
            compile_force_byteorder(map, eais->address_reg, BO_NORMAL, 0);
            compile_move_reg_from_mem_regoffs(destr, eais->address_reg,
                              (uae_u32)(eais->addr_const_off + address_space),
                              eais->size);

            map->x86_byteorder[destr] = BO_SWAPPED_LONG;
        }
        } else {
        /* Make this the new register cache reg */
        remove_x86r_from_cache(map, eais->data_reg, 0);
        remove_x86r_from_cache(map, destr, 0);
        map->x86_cache_reg[eais->data_reg] = eaid->reg;
        map->x86_cr_type[eais->data_reg] = 0;
        map->x86_const_offset[eais->data_reg] = eais->data_const_off;
        map->areg_map[eaid->reg] = eais->data_reg;
        map->x86_verified[eais->data_reg] = 0;
        }
    }

    if (map->areg_map[eaid->reg] >= 0)
        map->x86_dirty[map->areg_map[eaid->reg]] = 1;
    return;
    }

    if (eais->data_reg == -1)
    printf("Storing to mem, but not from reg\n");
    /* Correct the byteorder */
    if (eais->data_reg != -2) {
    compile_offset_reg(map, eais->data_reg, eais->data_const_off);

    switch (eaid->size) {
     case sz_byte: compile_force_byteorder(map, eais->data_reg, BO_NORMAL, 1); break;
     case sz_word: compile_force_byteorder(map, eais->data_reg, BO_SWAPPED_WORD, 1); break;
     case sz_long: compile_force_byteorder(map, eais->data_reg, BO_SWAPPED_LONG, 1); break;
    }
    compile_force_byteorder(map, eaid->address_reg, BO_NORMAL, 0);
    compile_move_reg_to_mem_regoffs(eaid->address_reg,
                    (uae_u32)(eaid->addr_const_off + address_space),
                    eais->data_reg, eaid->size);
    } else {
    switch (eaid->size) {
     case sz_long:
        eais->data_const_off = (((eais->data_const_off & 0xFF000000) >> 24)
                 | ((eais->data_const_off & 0xFF0000) >> 8)
                 | ((eais->data_const_off & 0xFF00) << 8)
                 | ((eais->data_const_off & 0xFF) << 24));
        break;
     case sz_word:
        eais->data_const_off = (((eais->data_const_off & 0xFF00) >> 8)
                 | ((eais->data_const_off & 0xFF) << 8));
        break;
    }
    compile_force_byteorder(map, eaid->address_reg, BO_NORMAL, 0);
    /* generate code to move valueoffset,eaoffset(eareg) */
    switch(eaid->size) {
     case sz_byte: assemble(0xC6); break;
     case sz_word: assemble(0x66); /* fall through */
     case sz_long: assemble(0xC7); break;
    }
    if (eaid->address_reg == -2) { /* absolute or PC-relative */
        assemble(0x05);
        assemble_long(eaid->addr_const_off + address_space);
    } else {
        assemble(0x80 + eaid->address_reg);
        assemble_long(eaid->addr_const_off + address_space);
    }
    switch(eaid->size) {
     case sz_byte: assemble(eais->data_const_off); break;
     case sz_word: assemble_uword(eais->data_const_off); break;
     case sz_long: assemble_ulong(eais->data_const_off); break;
    }
    }
}

#define CE_STACK_SIZE 1000

static struct {
    struct register_mapping map;
    char *jmpoffs;
    uae_u32 address;
    int noflush:1;
} compile_exit_stack[CE_STACK_SIZE];

static int cesp;

static struct register_mapping current_exit_regmap;

static void generate_exit(struct register_mapping *map, int address)
{
    int i;

    if (map != NULL)
    sync_reg_cache (map, 1);
    assemble(0xB8); /* movl $new_pc,%eax */
    assemble_ulong(address);
    assemble(0xC3); /* RET */
}

static void copy_map_with_undo(struct register_mapping *dst,
                   struct register_mapping *src,
                   struct pid_undo *pud)
{
    int i;
    *dst = *src;
    for (i = 0; i < pud->used; i++) {
    int m68kr = pud->m68kr[i];
    int x86r = pud->x86r[i];
    int old_cr = dst->areg_map[m68kr];
    if (old_cr != -1) {
        dst->x86_cache_reg[old_cr] = -1;
    }
    dst->x86_cache_reg[x86r] = m68kr;
    dst->areg_map[m68kr] = x86r;
    dst->x86_cr_type[x86r] = 0;
    dst->x86_const_offset[x86r] = pud->offs[i];
    dst->x86_dirty[x86r] = pud->dirty[i];
    }
}

static void unlock_pud(struct register_mapping *map, struct pid_undo *pud)
{
    int i;
    for (i = 0; i < pud->used; i++) {
    compile_unlock_reg(map, pud->x86r[i]);
    }
}

static int exits_necessary;

static void generate_possible_exit(struct register_mapping *map,
                   struct ea_info *eai, int iip,
                   struct pid_undo *pud)
{
    struct register_mapping exit_regmap;

    if (!exits_necessary) {
    unlock_pud(map, pud);
    return;
    }

    compile_force_byteorder(map, eai->address_reg, BO_NORMAL, 0);
    switch (eai->address_reg) {
     case -1:
    /* EA doesn't refer to memory */
    break;
     case -2:
    /* Only a constant offset */
    eai->addr_const_off &= (1<<24)-1;
    if (!good_address_map[eai->addr_const_off]) {
        copy_map_with_undo(&exit_regmap, map, pud);
        generate_exit(&exit_regmap, insn_info[iip].address);
    }
    break;
     default:
    if (map->x86_verified[eai->address_reg])
        break;
    map->x86_verified[eai->address_reg] = 1;
    if (cesp == CE_STACK_SIZE) {
        copy_map_with_undo(&exit_regmap, map, pud);
        generate_exit(&exit_regmap, insn_info[iip].address);
        break;
    }
    copy_map_with_undo(&compile_exit_stack[cesp].map, map, pud);
    compile_exit_stack[cesp].address = insn_info[iip].address;
    assemble(0x80); assemble(0xB8 + eai->address_reg); /* cmpb $0, good_address_map(x86r) */
    assemble_long(good_address_map + eai->addr_const_off);
    assemble(0);
    assemble(0x0F); assemble(0x84); /* JE finish */
    compile_exit_stack[cesp].jmpoffs = compile_here();
    compile_exit_stack[cesp].noflush = 0;
    assemble_ulong(0);
    cesp++;
    break;
    }
    unlock_pud(map, pud);
}

static void finish_exits(void)
{
    int i;
    for (i = 0; i < cesp; i++) {
    char *exitpoint = compile_here();
    char *nextpoint;

    if (compile_exit_stack[i].noflush)
        generate_exit(NULL, compile_exit_stack[i].address);
    else
        generate_exit(&compile_exit_stack[i].map, compile_exit_stack[i].address);
    nextpoint = compile_here();
    compile_org(compile_exit_stack[i].jmpoffs);
    assemble_ulong(exitpoint - (compile_exit_stack[i].jmpoffs + 4));
    compile_org(nextpoint);
    }
}

static void finish_condjumps(int lastiip)
{
    int iip;
    char *lastptr = compile_here();
    for (iip = 0; iip < lastiip; iip++) {
    char *fillin = insn_info[iip].compiled_fillin;
    if (fillin != NULL) {
        compile_org(insn_info[iip].compiled_fillin);
        assemble_ulong(insn_info[insn_info[iip].jumps_to].compiled_jumpaddr - (fillin + 4));
    }
    }
    compile_org(lastptr);
}

#define CC_X_FROM_86C 1
#define CC_C_FROM_86C 2
#define CC_Z_FROM_86Z 4
#define CC_V_FROM_86V 8
#define CC_N_FROM_86N 16
#define CC_TEST_REG   32
#define CC_Z_FROM_86C 64
#define CC_SAHF       128
#define CC_TEST_CONST 256
#define CC_AFTER_RO   512
#define CC_AFTER_ROX  1024

static unsigned int cc_status;
static int cc_reg;
static uae_u32 cc_offset;
static wordsizes cc_size;

static void compile_do_cc_test_reg(struct register_mapping *map)
{
    compile_force_byteorder(map, cc_reg, BO_NORMAL, 1);
    if (cc_offset != 0)
    printf("Pull my finger\n");
    if (cc_size == sz_word) /* test ccreg */
    assemble(0x66);
    if (cc_size == sz_byte)
    assemble(0x84);
    else
    assemble(0x85);
    assemble(0xC0 + 9*cc_reg);
}

static int compile_flush_cc_cache(struct register_mapping *map, int status,
                  int live_at_end, int user_follows,
                  int user_live_at_end, int user_ccval)
{
    int status_for_user = 0;

    if (user_follows) {
    int need_for_user = 0;
    int user_flagmask = cc_flagmask_68k(user_ccval);

    if (user_flagmask & CC68K_C)
        need_for_user |= CC_C_FROM_86C;
    if (user_flagmask & CC68K_Z)
        need_for_user |= CC_Z_FROM_86Z;
    if (user_flagmask & CC68K_N)
        need_for_user |= CC_N_FROM_86N;
    if (user_flagmask & CC68K_V)
        need_for_user |= CC_V_FROM_86V;

    /* Check whether we can satisfy the user's needs in a simple way. */
    if ((need_for_user & status) == need_for_user)
        status_for_user = status;
    else if (user_flagmask == CC68K_Z && status == CC_Z_FROM_86C)
        status_for_user = status;
    else if (status == CC_TEST_REG && (user_flagmask & (CC68K_C|CC68K_V|CC68K_Z|CC68K_N)) != 0) {
        if (cc_reg == -2) {
        status_for_user = CC_TEST_CONST;
        } else {
        compile_do_cc_test_reg(map);
        status_for_user = status = (CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V);
        }
    } else if (status == CC_AFTER_RO) {
        /* We fake some information here... */
        if (user_flagmask == CC68K_C && (user_live_at_end & ~CC68K_C) == 0)
        status = status_for_user = CC_C_FROM_86C;
        else if (((user_flagmask | user_live_at_end) & (CC68K_C|CC68K_V)) == 0) {
        status = CC_TEST_REG; user_live_at_end = CC68K_Z|CC68K_N|CC68K_V;
        status_for_user = (CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V);
        } else
        status_for_user = CC_SAHF;
    } else if (status == CC_AFTER_ROX) {
        if (user_flagmask == CC68K_C && (user_live_at_end & ~(CC68K_C|CC68K_X)) == 0)
        status = status_for_user = CC_C_FROM_86C;
        else if (((user_flagmask | user_live_at_end) & (CC68K_C|CC68K_X|CC68K_V)) == 0) {
        status = CC_TEST_REG; user_live_at_end = CC68K_Z|CC68K_N|CC68K_V;
        status_for_user = (CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V);
        } else
        status_for_user = CC_SAHF;
    } else if (need_for_user != 0) {
        /* No way to handle it easily */
        status_for_user = CC_SAHF;
    }
    if (status_for_user != CC_SAHF)
        live_at_end = user_live_at_end;
    }

    /*
     * Now store the flags which are live at the end of this insn and set by
     * us into their home locations
     */
    if (status == CC_TEST_REG) {
    if ((live_at_end & (CC68K_C|CC68K_V|CC68K_Z|CC68K_N)) == 0)
        goto all_ok;

    if (cc_reg == -2) {
        uae_u8 f = 0;
        if (cc_size == sz_byte) {
        f |= (cc_offset & 0x80) ? 0x80 : 0;
        f |= (cc_offset & 0xFF) == 0 ? 0x40 : 0;
        } else if (cc_size == sz_byte) {
        f |= (cc_offset & 0x8000) ? 0x80 : 0;
        f |= (cc_offset & 0xFFFF) == 0 ? 0x40 : 0;
        } else {
        f |= (cc_offset & 0x80000000) ? 0x80 : 0;
        f |= (cc_offset & 0xFFFFFFFF) == 0 ? 0x40 : 0;
        }
        assemble(0xC7); assemble(0x05);
        assemble_long((char*)®flags);
        assemble_uword(f);
    } else {
        int tmpr = get_free_x86_register(map, ALL_X86_REGS);
        compile_do_cc_test_reg(map);

        /* pushfl; popl tmpr; movl tempr, regflags */
        assemble(0x9C); assemble(0x58+tmpr);
        compile_move_reg_to_mem_regoffs(-2, (uae_u32)®flags, tmpr, sz_long);
    }
    } else if (status == CC_Z_FROM_86C) {
    if ((live_at_end & CC68K_Z) != 0) {
        int tmpr = get_typed_x86_register(map, DATA_X86_REGS);
        assemble(0x9C);
        /* setnc tmpr; shl $6, tmpr; andb $~0x40, regflags; orb tmpr, regflags */
        assemble(0x0F); assemble(0x93); assemble(0xC0 + tmpr);
        assemble(0xC0); assemble(4*8 + 0xC0 + tmpr); assemble(6);
        assemble(0x80); assemble(0x05+0x20); assemble_long(®flags); assemble((uae_u8)~0x40);
        assemble(0x08); assemble(0x05+ tmpr*8); assemble_long(®flags);
        assemble(0x9D);
    }
    } else if (status == CC_AFTER_RO || status == CC_AFTER_ROX) {
    int tmpr = get_typed_x86_register(map, DATA_X86_REGS);
    assemble(0x9C);
    compile_do_cc_test_reg(map);
    /* pushfl; popl tmpr; andl $0xff,tmpr (mask out V flag which is cleared after rotates) */
    assemble(0x9C); assemble(0x58 + tmpr);
    assemble(0x81); assemble(0xC0 + tmpr + 8*4); assemble_ulong(0xFF);
    assemble(0x9D);
    /* adc $0, tmpr */
    assemble(0x80); assemble(0xC0 + tmpr + 8*2); assemble(0);
    compile_move_reg_to_mem_regoffs(-2, (uae_u32)®flags, tmpr, sz_long);
    if (status == CC_AFTER_ROX)
        compile_move_reg_to_mem_regoffs(-2, 4 + (uae_u32)®flags, tmpr, sz_long);
    } else if (status != 0) {
    assert((status & CC_TEST_REG) == 0);
    assert (status == (CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_X_FROM_86C | CC_V_FROM_86V)
        || status == (CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V)
        || status == CC_C_FROM_86C);

    if ((status & CC_X_FROM_86C) == 0)
        live_at_end &= ~CC68K_X;

    if (status == CC_C_FROM_86C && (live_at_end & CC68K_C) != 0)
        fprintf(stderr, "Shouldn't be needing C here!\n");
    else if (live_at_end) {
        if ((live_at_end & CC68K_X) == 0)
        status &= ~CC_X_FROM_86C;

        if (live_at_end) {
        if ((status & CC_X_FROM_86C) != 0 && live_at_end == CC68K_X) {
            /* SETC regflags + 4 */
            assemble(0x0F); assemble(0x92);
            assemble(0x05); assemble_long(4 + (uae_u32)®flags);
        } else {
            int tmpr = get_free_x86_register(map, ALL_X86_REGS);
            /* pushfl; popl tmpr; movl tempr, regflags */
            assemble(0x9C); assemble(0x58+tmpr);
            compile_move_reg_to_mem_regoffs(-2, (uae_u32)®flags, tmpr, sz_long);

            if (status & CC_X_FROM_86C) {
            compile_move_reg_to_mem_regoffs(-2, 4 + (uae_u32)®flags, tmpr, sz_word);
            }
        }
        }
    }
    }

    all_ok:
    return status_for_user;
}

static char *compile_condbranch(struct register_mapping *map, int iip,
                int new_cc_status)
{
    int cc = insn_info[iip].dp->cc;
    int flagsused = cc_flagmask_68k(cc);
    int flagsneeded = 0;
    char *undo_pointer = compile_here();

    if (flagsused & CC68K_C)
    flagsneeded |= CC_C_FROM_86C;
    if (flagsused & CC68K_Z)
    flagsneeded |= CC_Z_FROM_86Z;
    if (flagsused & CC68K_N)
    flagsneeded |= CC_N_FROM_86N;
    if (flagsused & CC68K_V)
    flagsneeded |= CC_V_FROM_86V;

    if (flagsneeded == 0)
    /* Fine */;
    else if (new_cc_status == CC_SAHF) {
    int tmpr = get_free_x86_register(map, ALL_X86_REGS);
    compile_move_reg_from_mem_regoffs(tmpr, -2, (uae_u32)®flags, sz_long);
    assemble(0x66); assemble(0x50+tmpr); assemble(0x66); assemble(0x9D);
    new_cc_status = CC_C_FROM_86C|CC_Z_FROM_86Z|CC_N_FROM_86N|CC_V_FROM_86V;
    } else if (new_cc_status == CC_TEST_CONST) {
    int n,z;
    switch(cc_size) {
     case sz_byte: n = ((uae_s8)cc_offset) < 0; z = ((uae_s8)cc_offset) == 0; break;
     case sz_word: n = ((uae_s16)cc_offset) < 0; z = ((uae_s16)cc_offset) == 0; break;
     case sz_long: n = ((uae_s32)cc_offset) < 0; z = ((uae_s32)cc_offset) == 0; break;
    }
#define Bcc_TRUE 0
#define Bcc_FALSE 1
    flagsneeded = 0;
    new_cc_status = 0;
    switch (cc) {
     case 2: cc = !z ? Bcc_TRUE : Bcc_FALSE; break; /* !CFLG && !ZFLG */
     case 3: cc = z ? Bcc_TRUE : Bcc_FALSE; break; /* CFLG || ZFLG */
     case 4: cc = Bcc_TRUE; break; /* !CFLG */
     case 5: cc = Bcc_FALSE; break; /* CFLG */
     case 6: cc = !z ? Bcc_TRUE : Bcc_FALSE; break; /* !ZFLG */
     case 7: cc = z ? Bcc_TRUE : Bcc_FALSE; break; /* ZFLG */
     case 8: cc = Bcc_TRUE; break; /* !VFLG */
     case 9: cc = Bcc_FALSE; break; /* VFLG */
     case 10:cc = !n ? Bcc_TRUE : Bcc_FALSE; break; /* !NFLG */
     case 11:cc = n ? Bcc_TRUE : Bcc_FALSE; break; /* NFLG */
     case 12:cc = !n ? Bcc_TRUE : Bcc_FALSE; break; /* NFLG == VFLG */
     case 13:cc = n ? Bcc_TRUE : Bcc_FALSE; break; /* NFLG != VFLG */
     case 14:cc = !n && !z ? Bcc_TRUE : Bcc_FALSE; break; /* !ZFLG && (NFLG == VFLG) */
     case 15:cc = n || z ? Bcc_TRUE : Bcc_FALSE; break; /* ZFLG || (NFLG != VFLG) */
    }
    } else if (new_cc_status == CC_Z_FROM_86C) {
    if (cc == 6 || cc == 7) {
        cc = (cc - 2) ^ 1;
        /* Fake... */
        flagsneeded = new_cc_status = CC_C_FROM_86C;
    } else if (cc != 0 && cc != 1)
        printf("Groan!\n");
    }

    if (cc == 1)
    return NULL;

    if ((flagsneeded & new_cc_status) == flagsneeded) {
    char *result;
    /* We can generate a simple branch */
    if (cc == 0)
        assemble(0xE9);
    else
        assemble(0x0F);
    switch(cc) {
     case 2: assemble(0x87); break;          /* HI */
     case 3: assemble(0x86); break;          /* LS */
     case 4: assemble(0x83); break;          /* CC */
     case 5: assemble(0x82); break;          /* CS */
     case 6: assemble(0x85); break;          /* NE */
     case 7: assemble(0x84); break;          /* EQ */
     case 8: assemble(0x81); break;          /* VC */
     case 9: assemble(0x80); break;          /* VS */
     case 10:assemble(0x89); break;          /* PL */
     case 11:assemble(0x88); break;          /* MI */
     case 12:assemble(0x8D); break;          /* GE */
     case 13:assemble(0x8C); break;          /* LT */
     case 14:assemble(0x8F); break;          /* GT */
     case 15:assemble(0x8E); break;          /* LE */
    }
    result = compile_here();
    assemble_ulong(0);
    return result;
    }
    printf("Uhhuh.\n");
    return NULL;
}

static void compile_handle_bcc(struct register_mapping *map, int iip,
                   int new_cc_status)
{
    insn_info[iip].compiled_fillin = compile_condbranch(map, iip, new_cc_status);
}

static void compile_handle_dbcc(struct register_mapping *map, int iip,
                int new_cc_status, int dreg)
{
    char *fillin1 = compile_condbranch(map, iip, new_cc_status);

    /* subw $1,dreg; jnc ... */
    assemble(0x66); assemble(0x83); assemble(0x05 + 5*8);
    assemble_long(regs.regs + dreg);
    assemble(1);
    assemble(0x0F); assemble(0x83);
    insn_info[iip].compiled_fillin = compile_here();
    assemble_ulong(0);
    if (fillin1 != NULL) {
    char *oldp = compile_here();
    compile_org(fillin1);
    assemble_ulong(oldp - (fillin1+4));
    compile_org(oldp);
    }
}

static void handle_bit_insns(struct register_mapping *map, struct ea_info *eainf,
                 int eaino_s, int eaino_d, instrmnem optype)
{
    struct ea_info *srcea = eainf + eaino_s, *dstea = eainf + eaino_d;
    int code = (optype == i_BTST ? 0
        : optype == i_BSET ? 1
        : optype == i_BCLR ? 2
        : /* optype == i_BCHG */ 3);

    compile_fetchea(map, eainf, eaino_s, 5);
    compile_fetchea(map, eainf, eaino_d, 3);

    if (srcea->data_reg != -2) {
    compile_force_byteorder(map, srcea->data_reg, BO_NORMAL, 0);
    remove_x86r_from_cache(map, srcea->data_reg, 0);
    /* andl $something,srcreg */
    assemble(0x83); assemble(0xC0 + 4*8 + srcea->data_reg);
    if (dstea->size == sz_byte)
        assemble(7);
    else
        assemble(31);
    } else
    if (dstea->size == sz_byte)
        srcea->data_const_off &= 7;
    else
        srcea->data_const_off &= 31;

    /* Areg isn't possible here */
    if (dstea->mode == Dreg && dstea->data_reg == -1) {
    if (srcea->data_reg == -2) {
        assemble(0x0F); assemble(0xBA); assemble(5 + 8*(4 + code));
        assemble_long(regs.regs + dstea->reg);
        assemble(srcea->data_const_off);
    } else {
        assemble(0x0F); assemble(0xA3 + 8*code);
        assemble(5 + srcea->data_reg*8);
        assemble_long(regs.regs + dstea->reg);
    }
    } else if (dstea->data_reg >= 0) {
    compile_force_byteorder(map, dstea->data_reg, BO_NORMAL, 0);
    if (srcea->data_reg == -2) {
        assemble(0x0F); assemble(0xBA); assemble(0xC0 + dstea->data_reg + 8*(4 + code));
        assemble(srcea->data_const_off);
    } else {
        assemble(0x0F); assemble(0xA3 + 8*code);
        assemble(0xC0 + dstea->data_reg + srcea->data_reg*8);
    }
    if (optype != i_BTST)
        map->x86_dirty[dstea->data_reg] = 1;
    } else {
    int addr_code = dstea->address_reg == -2 ? 5 : dstea->address_reg + 0x80;
    compile_force_byteorder(map, dstea->address_reg, BO_NORMAL, 0);
    /* We have an address in memory */
    if (dstea->data_reg != -1)
        printf("Things don't look good in handle_bit_insns\n");
    if (srcea->data_reg == -2) {
        assemble(0x0F); assemble(0xBA);
        assemble(addr_code + 8*(4 + code));
        assemble_long(address_space + dstea->addr_const_off);
        assemble(srcea->data_const_off);
    } else {
        assemble(0x0F); assemble(0xA3 + 8*code);
        assemble(addr_code + srcea->data_reg*8);
        assemble_long(address_space + dstea->addr_const_off);
    }

    }
    cc_status = CC_Z_FROM_86C;
}

static int do_rotshi = 1;

static void handle_rotshi(struct register_mapping *map, int iip,
              uae_u8 *realpc, uaecptr current_addr, struct pid_undo *pud)
{
    struct ea_info eai;
    int amode_reg = insn_info[iip].dp->sreg;
    int amode_mode = insn_info[iip].dp->smode;
    wordsizes size = insn_info[iip].dp->size;
    int shiftcount;
    int mnemo = insn_info[iip].dp->mnemo;
    int shiftcode;
    int locked_eax_for_sahf = 0;

    switch(mnemo) {
     case i_ASLW: shiftcount = 1; mnemo = i_ASL; break;
     case i_ASRW: shiftcount = 1; mnemo = i_ASR; break;
     case i_LSLW: shiftcount = 1; mnemo = i_LSL; break;
     case i_LSRW: shiftcount = 1; mnemo = i_LSR; break;
     case i_ROLW: shiftcount = 1; mnemo = i_ROL; break;
     case i_RORW: shiftcount = 1; mnemo = i_ROR; break;
     case i_ROXLW:shiftcount = 1; mnemo = i_ROXL;break;
     case i_ROXRW:shiftcount = 1; mnemo = i_ROXR;break;
     default:
    amode_reg = insn_info[iip].dp->dreg;
    amode_mode = insn_info[iip].dp->dmode;
    shiftcount = insn_info[iip].dp->sreg;
    break;
    }
    if ((insn_info[iip].flags_live_at_end & CC68K_V) != 0) {
    if (mnemo == i_ASL) {
        generate_exit(map, insn_info[iip].address);
        printf("Can't handle this shift\n");
        return;
    } else if (mnemo == i_ASR || mnemo == i_LSR || mnemo == i_LSL) {
        remove_x86r_from_cache(map, r_EAX, 1);
        locked_eax_for_sahf = 1;
        lock_reg(map, r_EAX, 2);
    }

    }
    if (mnemo == i_ROXR || mnemo == i_ROXL) {
    remove_x86r_from_cache(map, r_EAX, 1);
    lock_reg(map, r_EAX, 2);
    compile_move_reg_from_mem_regoffs(r_AH, -2, 4 + (uae_u32)®flags, sz_byte);
    }
    compile_prepare_undo(map, amode_mode, amode_reg, pud);
    compile_prepareea(map, amode_mode, amode_reg, size,
              &realpc, current_addr,
              &eai, 0, EA_LOAD|EA_STORE|EA_MODIFY, 1);

    generate_possible_exit(map, &eai, iip, pud);

    compile_fetchea(map, &eai, 0, 1);
    compile_force_byteorder(map, eai.data_reg, BO_NORMAL, 0);

    switch (mnemo) {
     case i_ASL:
    shiftcode = 4; cc_status = CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V | CC_X_FROM_86C;
    break;
     case i_LSL:
    shiftcode = 4; cc_status = CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V | CC_X_FROM_86C;
    break;
     case i_LSR:
    shiftcode = 5; cc_status = CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V | CC_X_FROM_86C;
    break;
     case i_ASR:
    shiftcode = 7; cc_status = CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V | CC_X_FROM_86C;
    break;
     case i_ROR:
    shiftcode = 1; cc_status = CC_AFTER_RO;
    break;
     case i_ROL:
    shiftcode = 0; cc_status = CC_AFTER_RO;
    break;
     case i_ROXL:
    shiftcode = 2; assemble(0x9E); /* SAHF */ cc_status = CC_AFTER_ROX; compile_unlock_reg(map, r_EAX);
    break;
     case i_ROXR:
    shiftcode = 3; assemble(0x9E); /* SAHF */ cc_status = CC_AFTER_ROX; compile_unlock_reg(map, r_EAX);
    break;
    }

    if (size == sz_word)
    assemble(0x66);
    assemble((shiftcount == 1 ? 0xD0 : 0xC0) + (size == sz_byte ? 0 : 1));
    assemble(shiftcode*8+0xC0 + eai.data_reg);
    if (shiftcount != 1) assemble(shiftcount);
    cc_offset = 0; cc_size = size; cc_reg = eai.data_reg;

    if (locked_eax_for_sahf) {
    /* The trick here is that the overflow flag isn't put into AH in SAHF */
    assemble(0x9E);
    assemble(0x0B); assemble(9*1 + 0xC0);
    assemble(0x9F);
    compile_unlock_reg(map, r_EAX);
    }
    compile_note_modify(map, &eai, 0);
}

static void handle_rotshi_variable(struct register_mapping *map, int iip,
                   uae_u8 *realpc, uaecptr current_addr,
                   struct pid_undo *pud)
{
    struct ea_info eais, eaid;
    int mnemo = insn_info[iip].dp->mnemo;
    int shiftcode;
    char *tmp1, *tmp2;
    int locked_eax_for_sahf = 0;

    remove_x86r_from_cache(map, r_ECX, 1);
    lock_reg(map, r_ECX, 2);

    if ((insn_info[iip].flags_live_at_end & CC68K_V) != 0) {
    if (mnemo == i_ASL) {
        generate_exit(map, insn_info[iip].address);
        printf("Can't handle this shift (var)\n");
        return;
    } else if (mnemo == i_ASR || mnemo == i_LSR || mnemo == i_LSL) {
        remove_x86r_from_cache(map, r_EAX, 1);
        locked_eax_for_sahf = 1;
        lock_reg(map, r_EAX, 2);
    }

    }
    if (mnemo == i_ROXR || mnemo == i_ROXL) {
    remove_x86r_from_cache(map, r_EAX, 1);
    lock_reg(map, r_EAX, 2);
    compile_move_reg_from_mem_regoffs(r_AH, -2, 4 + (uae_u32)®flags,
                      sz_byte);
    }
    /* Both src and dest are Dreg modes */
    compile_prepareea(map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg,
              sz_long, &realpc, current_addr,
              &eais, 0, EA_LOAD, 1);
    compile_prepareea(map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg,
              insn_info[iip].dp->size, &realpc, current_addr,
              &eaid, 0, EA_LOAD|EA_STORE|EA_MODIFY, 1);

    compile_fetchea(map, &eais, 0, 1);
    compile_fetchea(map, &eaid, 1, 1);
    compile_force_byteorder(map, eais.data_reg, BO_NORMAL, 0);
    compile_force_byteorder(map, eaid.data_reg, BO_NORMAL, 0);
    compile_move_reg_reg(r_ECX, eais.data_reg, sz_long);
    /* Test against zero, and test bit 6. If 1 <= count <= 31, we can do the
     * operation, otherwise, we have to exit */
    assemble(0xF6); assemble(0xC0 + r_ECX); assemble(0x1F);
    assemble(0x74); assemble(9);
    assemble(0xF6); assemble(0xC0 + r_ECX); assemble(0x20);

    assemble(0x0F); assemble(0x85); tmp1 = compile_here(); assemble_ulong(0);
    generate_exit(map, insn_info[iip].address);
    tmp2 = compile_here(); compile_org (tmp1); assemble_ulong((tmp2-tmp1) + 4); compile_org(tmp2);

    switch (mnemo) {
     case i_ASL:
    shiftcode = 4; cc_status = CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V | CC_X_FROM_86C;
    break;
     case i_LSL:
    shiftcode = 4; cc_status = CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V | CC_X_FROM_86C;
    break;
     case i_LSR:
    shiftcode = 5; cc_status = CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V | CC_X_FROM_86C;
    break;
     case i_ASR:
    shiftcode = 7; cc_status = CC_C_FROM_86C | CC_Z_FROM_86Z | CC_N_FROM_86N | CC_V_FROM_86V | CC_X_FROM_86C;
    break;
     case i_ROR:
    shiftcode = 1; cc_status = CC_AFTER_RO;
    break;
     case i_ROL:
    shiftcode = 0; cc_status = CC_AFTER_RO;
    break;
     case i_ROXL:
    shiftcode = 2; assemble(0x9E); /* SAHF */ cc_status = CC_AFTER_ROX; compile_unlock_reg(map, r_EAX);
    break;
     case i_ROXR:
    shiftcode = 3; assemble(0x9E); /* SAHF */ cc_status = CC_AFTER_ROX; compile_unlock_reg(map, r_EAX);
    break;
    }

    if (insn_info[iip].dp->size == sz_word)
    assemble(0x66);
    assemble(0xD2 + (insn_info[iip].dp->size == sz_byte ? 0 : 1));
    assemble(shiftcode*8+0xC0 + eaid.data_reg);
    cc_offset = 0; cc_size = insn_info[iip].dp->size; cc_reg = eaid.data_reg;

    if (locked_eax_for_sahf) {
    /* The trick here is that the overflow flag isn't put into AH in SAHF */
    assemble(0x9E);
    assemble(0x0B); assemble(9*1 + 0xC0);
    assemble(0x9F);
    compile_unlock_reg(map, r_EAX);
    }
    compile_note_modify(map, &eaid, 0);
    compile_unlock_reg(map, r_ECX);
}

static uae_u32 testmask = 0xF80000, testval = 0xF80000;

static int m68k_compile_block(struct hash_block *hb)
{
    int movem_extra = 0;
    int last_iip = m68k_scan_block(hb, &movem_extra);
    struct register_mapping map;
    int i, iip, szflag;
    uae_u8 *realpc_start = NULL;
    struct bb_info *current_bb;
    int cc_status_for_bcc = CC_SAHF;
    struct insn_reg_needs reg_needs_init;

    cesp = 0;

    if (n_compiled > n_max_comp)
    return 1;
    else if (n_compiled++ == n_max_comp)
    printf("X\n");

    cc_status = 0; compile_failure = 0;

    /* Kickstart ROM address? */
    if ((hb->he_first->addr & 0xF80000) != 0xF80000
    && 0 && !patched_syscalls)
    return 1;

    exits_necessary = ((hb->he_first->addr & 0xF80000) == 0xF80000 || !USER_PROGRAMS_BEHAVE);

    if (alloc_code (hb, last_iip + movem_extra) == NULL) {
    hb->allocfailed = 1;
    return 0;
    }
    compile_org(hb->compile_start);
    compile_last_addr = (char *)hb->compile_start + hb->alloclen;

    /* m68k_scan_block() will leave this all set up */
    current_bb = bb_stack;

    for (i = 0; i < 8; i++) {
    map.dreg_map[i] = map.areg_map[i] = -1;
    map.x86_dirty[i] = 0;
    map.x86_cache_reg[i] = -1;
    map.x86_cr_type[i] = 0;
    map.x86_const_offset[i] = 0;
    map.x86_verified[i] = 0;
    map.x86_byteorder[i] = BO_NORMAL;
    }

    reg_needs_init.checkpoint_no = 0;
    for (i = 0; i < 8; i++) {
    reg_needs_init.dreg_needed[i] = reg_needs_init.areg_needed[i] = -1;
    reg_needs_init.dreg_mask[i] = reg_needs_init.areg_mask[i] = ALL_X86_REGS;
    }

    for (iip = 0; iip < last_iip && !compile_failure; iip++) {
    uae_u8 *realpc;
    struct ea_info eainfo[8];
    uaecptr current_addr;
    struct pid_undo pub;
    struct insn_reg_needs this_reg_needs = reg_needs_init;

    /* Set up locks for a new insn. We don't bother to clear this
     * properly after compiling one insn. */
    for (i = 0; i < 8; i++) {
        map.x86_users[i] = i == r_ESP ? 1 : 0;
        map.x86_locked[i] = i == r_ESP ? 2 : 0;
    }

    pub.used = 0;
    current_addr = insn_info[iip].address + 2;

    if (iip == current_bb->first_iip) {
        sync_reg_cache(&map, 1);
        if (!quiet_compile)
        printf("Compiling %08lx\n", current_bb->h->addr);

        realpc_start = get_real_address(current_bb->h->addr);
        current_bb->h->execute = (code_execfunc)compile_here();
        current_bb->h->matchword = *(uae_u32 *)realpc_start;
        cc_status_for_bcc = CC_SAHF;
    }

    realpc = realpc_start + (current_addr - current_bb->h->addr);

    insn_info[iip].compiled_jumpaddr = compile_here();
    insn_info[iip].compiled_fillin = NULL;

    if (insn_info[iip].jump_target) {
        if (cesp == CE_STACK_SIZE) {
        generate_exit(NULL, insn_info[iip].address);
        compile_failure = 1;
        } else {
        assemble(0xFE); assemble(0x05 + 8*1); assemble_long(&nr_bbs_to_run);
        assemble(0x0F); assemble(0x84); /* JE finish */
        compile_exit_stack[cesp].noflush = 1;
        compile_exit_stack[cesp].address = current_bb->h;
        compile_exit_stack[cesp].jmpoffs = compile_here();
        assemble_ulong(0);
        cesp++;
        }
    }
    /*
     * This will sort out all insns we can't compile, including
     * jumps out of this block */
    if (insn_info[iip].stop_translation == 1) {
        generate_exit(&map, insn_info[iip].address);
        cc_status = 0;
    } else switch (insn_info[iip].dp->mnemo) {
     case i_NOP:
        cc_status = 0;
        if (!quiet_compile)
        printf("Compiling a NOP\n");
        break;

     case i_RTS:
        sync_reg_cache(&map, 1);
        lock_reg(&map, r_ECX, 2);
        lock_reg(&map, r_EBX, 2);
        {
        char *tmp1, *tmp2, *tmp3;

        /* fetch (A7) */
        assemble(0x8B); assemble(0x5 + r_EBX*8); assemble_long(regs.regs + 15);
        assemble(0x8B); assemble(0x80 + 9*r_EBX); assemble_long(address_space);
        assemble(0x0F); /* bswapl x86r */
        assemble(0xC8 + r_EBX);
        /* fetch jsr_num */
        assemble(0x8B); assemble(0x5 + r_ECX*8); assemble_long(&jsr_num);
        assemble(0x09); assemble(0xC0 + 9*r_ECX);
        assemble(0x0F); assemble(0x84); tmp1 = compile_here(); assemble_ulong(0);
        assemble(0xFF); assemble(1*8 + 0xC0 + r_ECX);
        /* cmpl %ebx,disp32(,%ecx,4) */
        assemble(0x39); assemble(0x04 + 8*r_EBX); assemble(0x8d);
        assemble_long(jsr_rets);
        assemble(0x0F); assemble(0x85); tmp2 = compile_here(); assemble_ulong(0);
        /* movl disp32(,%ecx,4),%ebx */
        assemble(0x8B); assemble(0x04 + 8*r_EBX); assemble(0x8d);
        assemble_long(jsr_hash);
        /* movl execute(%ebx), %ebx */
        assemble(0x8B); assemble(0x040 + 9*r_EBX); assemble((int)&((struct hash_entry *)0)->execute);
        assemble(0x09); assemble(0xC0 + 9*r_EBX);
        assemble(0x0F); assemble(0x85); tmp3 = compile_here(); assemble_ulong(0);
        compile_org(tmp1); assemble_ulong(tmp3 - tmp1);
        compile_org(tmp2); assemble_ulong(tmp3 - tmp2);
        compile_org(tmp3 + 4);
        generate_exit(&map, insn_info[iip].address);
        tmp1 = compile_here();
        compile_org(tmp3); assemble_ulong((tmp1-tmp3)-4);
        compile_org(tmp1);
        assemble(0x89); assemble(0x5 + r_ECX*8); assemble_long(&jsr_num);
        assemble(0x83); assemble(0x05 + 5*8); assemble_long(regs.regs + 15); assemble(-4);
        /* Off we go */
        assemble(0xFF); assemble(4*8 + 0xC0 + r_EBX);
        }
        break;

     case i_JMP:
        sync_reg_cache(&map, 1);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        {
        char *tmp1, *tmp2, *tmp3;

        struct hash_entry *tmph;
        if (eainfo[0].address_reg != -2 || (tmph = get_hash_for_func(eainfo[0].addr_const_off, 1)) == 0) {
            if (eainfo[0].address_reg != -2 && !quiet_compile)
            printf("Can't compile indirect JMP\n");
            generate_exit(&map, insn_info[iip].address);
            break;
        }
        /* check whether the destination has compiled code */
        assemble(0x8B); assemble(r_EBX*8 + 0x05); assemble_long(&(tmph->execute));
        assemble(0x09); assemble(0xC0 + 9*r_EBX);
        assemble(0x0F); assemble(0x85); tmp1 = compile_here(); assemble_ulong(0);
        generate_exit(&map, insn_info[iip].address);
        tmp2 = compile_here(); compile_org(tmp1);
        assemble_ulong((tmp2 - tmp1) - 4);
        compile_org(tmp2);
        /* Off we go */
        assemble(0xFF); assemble(4*8 + 0xC0 + r_EBX);
        }
        cc_status = 0;
        break;

     case i_JSR:
        sync_reg_cache(&map, 1);
        lock_reg(&map, r_ECX, 2);
        lock_reg(&map, r_EBX, 2);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        {
        char *tmp1, *tmp2, *tmp3;

        struct hash_entry *tmph;
        if (eainfo[0].address_reg != -2 || (tmph = get_hash_for_func(eainfo[0].addr_const_off, 1)) == 0) {
            if (eainfo[0].address_reg != -2 && !quiet_compile)
            printf("Can't compile indirect JSR\n");
            generate_exit(&map, insn_info[iip].address);
            break;
        }
        assert(iip + 1 < last_iip);
        assert(iip == current_bb->last_iip);
        /* check whether the destination has compiled code */
        assemble(0x8B); assemble(r_EBX*8 + 0x05); assemble_long(&(tmph->execute));
        assemble(0x09); assemble(0xC0 + 9*r_EBX);
        assemble(0x0F); assemble(0x84); tmp3 = compile_here(); assemble_ulong(0);
        /* check for stack overflow */
        assemble(0x8B); assemble(r_ECX*8 + 0x05); assemble_long(&jsr_num);
        assemble(0xF7); assemble(0xC0+r_ECX); assemble_ulong(MAX_JSRS);
        assemble(0x0F); assemble(0x84); tmp1 = compile_here(); assemble_ulong(0);
        generate_exit(&map, insn_info[iip].address);
        tmp2 = compile_here(); compile_org(tmp1); assemble_ulong((tmp2 - tmp1) - 4);
        compile_org(tmp3); assemble_ulong(tmp1-tmp3);
        compile_org(tmp2);
        /* movl $something,disp32(,%ecx,4) */
        assemble(0xC7); assemble(0x04); assemble(0x8d);
        assemble_long(jsr_rets); assemble_ulong(insn_info[iip+1].address);
        assemble(0xC7); assemble(0x04); assemble(0x8d);
        assemble_long(jsr_hash); assemble_long((current_bb + 1)->h);
        /* incl jsr_num */
        assemble(0xFF); assemble(0x05); assemble_long(&jsr_num);
        /* Put things on the 68k stack */
        assemble(0x83); assemble(0x05 + 5*8); assemble_long(regs.regs + 15); assemble(4);
        assemble(0x8B); assemble(r_ECX*8+ 0x05); assemble_long(regs.regs + 15);
        assemble(0xC7); assemble(0x80 + r_ECX); assemble_long(address_space);
        assemble_ulong_68k(insn_info[iip+1].address);
        /* Off we go */
        assemble(0xFF); assemble(4*8 + 0xC0 + r_EBX);
        }
        break;

     case i_BSR:
        sync_reg_cache(&map, 1);
        lock_reg(&map, r_ECX, 2);
        lock_reg(&map, r_EBX, 2);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        {
        char *tmp1, *tmp2, *tmp3;
        uaecptr dest = insn_info[iip].address + 2 + (uae_s32)eainfo[0].data_const_off;
        struct hash_entry *tmph;
        if ((tmph = get_hash_for_func(dest, 1)) == 0) {
            generate_exit(&map, insn_info[iip].address);
            break;
        }
        assert(iip + 1 < last_iip);
        assert(iip == current_bb->last_iip);

        /* check whether the destination has compiled code */
        assemble(0x8B); assemble(r_EBX*8 + 0x05); assemble_long(&(tmph->execute));
        assemble(0x09); assemble(0xC0 + 9*r_EBX);
        assemble(0x0F); assemble(0x84); tmp3 = compile_here(); assemble_ulong(0);
        /* check for stack overflow */
        assemble(0x8B); assemble(r_ECX*8 + 0x05); assemble_long(&jsr_num);
        assemble(0xF7); assemble(0xC0+r_ECX); assemble_ulong(MAX_JSRS);
        assemble(0x0F); assemble(0x84); tmp1 = compile_here(); assemble_ulong(0);
        generate_exit(&map, insn_info[iip].address);
        tmp2 = compile_here(); compile_org(tmp1); assemble_ulong((tmp2 - tmp1) - 4);
        compile_org(tmp3); assemble_ulong(tmp1-tmp3);
        compile_org(tmp2);
        /* movl $something,disp32(,%ecx,4) */
        assemble(0xC7); assemble(0x04); assemble(0x8d);
        assemble_long(jsr_rets); assemble_ulong(insn_info[iip+1].address);
        assemble(0xC7); assemble(0x04); assemble(0x8d);
        assemble_long(jsr_hash); assemble_long((current_bb + 1)->h);
        /* incl jsr_num */
        assemble(0xFF); assemble(0x05); assemble_long(&jsr_num);
        /* Put things on the 68k stack */
        assemble(0x83); assemble(0x05 + 5*8); assemble_long(regs.regs + 15); assemble(4);
        assemble(0x8B); assemble(r_ECX*8+ 0x05); assemble_long(regs.regs + 15);
        assemble(0xC7); assemble(0x80 + r_ECX); assemble_long(address_space);
        assemble_ulong_68k(insn_info[iip+1].address);
        /* Off we go */
        assemble(0xFF); assemble(4*8 + 0xC0 + r_EBX);
        }
        break;

     case i_Bcc:
        sync_reg_cache(&map, 0);
        compile_handle_bcc(&map, iip, cc_status_for_bcc);
        cc_status = 0;
        break;

     case i_DBcc:
        sync_reg_cache(&map, 0);
        remove_x86r_from_cache(&map, map.dreg_map[insn_info[iip].dp->sreg], 1);
        compile_handle_dbcc(&map, iip, cc_status_for_bcc,
                insn_info[iip].dp->sreg);
        cc_status = 0;
        break;
#if 0
     case i_Scc:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_STORE, 1);

        generate_possible_exit(&map, eainfo, iip, &pub);
        srcreg2 = get_;
        compile_note_modify(&map, eainfo, 0);

        cc_status = 0;
        break;
#endif
     case i_ADD:
     case i_SUB:
     case i_CMP:
     case i_CMPM:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepare_undo(&map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 1,
                  (insn_info[iip].dp->mnemo == i_ADD || insn_info[iip].dp->mnemo == i_SUB
                   ? EA_MODIFY | EA_LOAD | EA_STORE
                   : EA_LOAD | EA_STORE), 1);

        generate_possible_exit(&map, eainfo, iip, &pub);
        generate_possible_exit(&map, eainfo+1, iip, &pub);

        compile_loadeas(&map, eainfo, 0, 1, binop_alternatives, 0, 1);

        switch (insn_info[iip].dp->mnemo) {
         case i_ADD: compile_eas(&map, eainfo, 0, 1, 0); break;
         case i_SUB: compile_eas(&map, eainfo, 0, 1, 5); break;
         case i_CMP: case i_CMPM: compile_eas(&map, eainfo, 0, 1, 7); break;
        }

        if (insn_info[iip].dp->mnemo != i_CMP && insn_info[iip].dp->mnemo != i_CMPM)
        compile_note_modify(&map, eainfo, 1);
        switch (insn_info[iip].dp->mnemo) {
         case i_ADD:
         case i_SUB:
        cc_status = CC_X_FROM_86C | CC_Z_FROM_86Z | CC_C_FROM_86C | CC_V_FROM_86V | CC_N_FROM_86N;
        break;
         case i_CMP:
         case i_CMPM:
        cc_status = CC_Z_FROM_86Z | CC_C_FROM_86C | CC_V_FROM_86V | CC_N_FROM_86N;
        break;
        }
        break;

     case i_ADDX:
     case i_SUBX:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepare_undo(&map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 1, EA_MODIFY | EA_LOAD | EA_STORE, 1);

        generate_possible_exit(&map, eainfo, iip, &pub);
        generate_possible_exit(&map, eainfo+1, iip, &pub);

        compile_loadeas(&map, eainfo, 0, 1, binop_alternatives, 0, 1);

        /* bt $0, regflags+4 ; get carry */
        assemble(0x0F); assemble(0xBA); assemble(0x5+4*8);
        assemble_ulong(4 + (uae_u32)®flags); assemble(0);

        switch (insn_info[iip].dp->mnemo) {
         case i_ADDX: compile_eas(&map, eainfo, 0, 1, 2); break;
         case i_SUBX: compile_eas(&map, eainfo, 0, 1, 3); break;
        }
        compile_note_modify(&map, eainfo, 1);

        if (insn_info[iip].flags_live_at_end & CC68K_Z) {
        /* Darn. */
        int tmpr = get_free_x86_register(&map, ALL_X86_REGS);
        /* pushfl; popl tmpr */
        assemble(0x9C); assemble(0x58+tmpr);
        /* Magic! */
        /* andl tmpr, regflags; andl $~0x40,tmpr; orl tmpr, regflags */
        assemble(0x21); assemble(0x05 + 8*tmpr); assemble_long(®flags);
        assemble(0x81); assemble(0xC0 + 8*4 + tmpr); assemble_ulong(~0x40);
        assemble(0x09); assemble(0x05 + 8*tmpr); assemble_long(®flags);
        compile_move_reg_to_mem_regoffs(-2, 4 + (uae_u32)®flags, tmpr, sz_long);
        cc_status = 0;
        } else {
        /* Lies! */
        cc_status = CC_X_FROM_86C | CC_Z_FROM_86Z |CC_C_FROM_86C |CC_V_FROM_86V |CC_N_FROM_86N;
        }
        break;

     case i_MULU:
     case i_MULS:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepare_undo(&map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 1, EA_MODIFY | EA_LOAD | EA_STORE, 1);

        generate_possible_exit(&map, eainfo, iip, &pub);
        generate_possible_exit(&map, eainfo+1, iip, &pub);

        compile_loadeas(&map, eainfo, 0, 1, regonly_alternatives, 0, 1);

        /* Extend the regs properly */
        remove_x86r_from_cache(&map, eainfo[0].data_reg, 0);
        switch (insn_info[iip].dp->mnemo) {
         case i_MULU:
        assemble(0x81); assemble(0xC0+4*8 + eainfo[0].data_reg); assemble_ulong(0xFFFF);
        assemble(0x81); assemble(0xC0+4*8 + eainfo[1].data_reg); assemble_ulong(0xFFFF);
        break;
         case i_MULS:
        assemble(0x0F); assemble(0xBF); assemble(0xC0 + 9*eainfo[0].data_reg);
        assemble(0x0F); assemble(0xBF); assemble(0xC0 + 9*eainfo[1].data_reg);
        break;
        }
        /* and multiply */
        assemble(0x0F); assemble(0xAF); assemble(0xC0 + 8*eainfo[1].data_reg + eainfo[0].data_reg);
        compile_note_modify(&map, eainfo, 1);
        cc_status = CC_TEST_REG;
        cc_reg = eainfo[1].data_reg;
        cc_offset = 0;
        cc_size = sz_long;
        break;

     case i_ADDA:
     case i_SUBA:
     case i_CMPA:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepare_undo(&map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  sz_long, &realpc, current_addr,
                  eainfo, 1,
                  (insn_info[iip].dp->mnemo == i_ADDA || insn_info[iip].dp->mnemo == i_SUBA
                   ? EA_MODIFY | EA_LOAD | EA_STORE
                   : EA_LOAD | EA_STORE),
                  1);

        generate_possible_exit(&map, eainfo, iip, &pub);

        compile_loadeas(&map, eainfo, 0, 1,
                insn_info[iip].dp->size == sz_word ? binop_worda_alternatives : binop_alternatives,
                0, 1);

        if (insn_info[iip].dp->size == sz_word) {
        remove_x86r_from_cache(&map, eainfo[0].data_reg, 0);
        compile_extend_long(&map, eainfo[0].data_reg, sz_word);
        }
        eainfo[0].size = sz_long;

        switch (insn_info[iip].dp->mnemo) {
         case i_ADDA: compile_eas(&map, eainfo, 0, 1, 0); break;
         case i_SUBA: compile_eas(&map, eainfo, 0, 1, 5); break;
         case i_CMPA: compile_eas(&map, eainfo, 0, 1, 7); break;
        }

        if (insn_info[iip].dp->mnemo == i_CMPA) {
        cc_status = CC_Z_FROM_86Z |CC_C_FROM_86C |CC_V_FROM_86V |CC_N_FROM_86N;
        } else {
        compile_note_modify(&map, eainfo, 1);
        cc_status = 0;
        }
        break;

     case i_MOVE:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepare_undo(&map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 1, EA_STORE, 1);

        generate_possible_exit(&map, eainfo, iip, &pub);
        generate_possible_exit(&map, eainfo + 1, iip, &pub);

        compile_loadeas(&map, eainfo, 0, 1, binop_alternatives, 1, 0);
        compile_storeea(&map, eainfo, 0, 1);

        if (eainfo[0].data_reg == -2) {
        cc_status = CC_TEST_REG;
        cc_reg = -2;
        cc_offset = eainfo[0].data_const_off;
        } else if (eainfo[0].data_reg == -1) {
        if (eainfo[1].data_reg == -1)
            printf("Don't know where to get flags from\n");
        cc_status = CC_TEST_REG;
        cc_offset = 0;
        cc_reg = eainfo[1].data_reg;
        } else {
        cc_status = CC_TEST_REG;
        cc_reg = eainfo[0].data_reg;
        cc_offset = 0;
        }
        cc_size = eainfo[0].size;

        break;

     case i_MOVEA:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepare_undo(&map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  sz_long, &realpc, current_addr,
                  eainfo, 1, EA_STORE, 1);

        generate_possible_exit(&map, eainfo, iip, &pub);

        compile_loadeas(&map, eainfo, 0, 1,
                insn_info[iip].dp->size == sz_word ? binop_worda_alternatives : binop_alternatives,
                0, 0);

        if (insn_info[iip].dp->size == sz_word) {
        remove_x86r_from_cache(&map, eainfo[0].data_reg, 0);
        compile_extend_long(&map, eainfo[0].data_reg, sz_word);
        }
        eainfo[0].size = sz_long;

        compile_storeea(&map, eainfo, 0, 1);

        cc_status = 0;
        break;

     case i_EXG:
        if (insn_info[iip].dp->smode != insn_info[iip].dp->dmode
        || insn_info[iip].dp->sreg != insn_info[iip].dp->dreg)
        {
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  sz_long, &realpc, current_addr,
                  eainfo, 0, EA_LOAD|EA_STORE, 1);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  sz_long, &realpc, current_addr,
                  eainfo, 1, EA_LOAD|EA_STORE, 1);

        compile_loadeas(&map, eainfo, 0, 1, regonly_alternatives, 0, 1);
        compile_storeea(&map, eainfo, 1, 0);
        compile_storeea(&map, eainfo, 0, 1);
        }

        cc_status = 0;
        break;

     case i_LINK:
        compile_prepare_undo(&map, Apdi, 7, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  sz_long, &realpc, current_addr,
                  eainfo, 0, EA_LOAD|EA_STORE, 1);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  sz_long, &realpc, current_addr,
                  eainfo, 1, EA_LOAD, 1);
        compile_prepareea(&map, Apdi, 7, sz_long, &realpc, current_addr,
                  eainfo, 2, EA_STORE, 1);

        generate_possible_exit(&map, eainfo+2, iip, &pub);

        compile_fetchea(&map, eainfo, 0, 1);
        /* we know this is a constant - no need to fetch it*/
        /* compile_fetchea(&map, eainfo, 1); */
        compile_storeea(&map, eainfo, 0, 2); /* An -> -(A7) */

        compile_prepareea(&map, Areg, 7, sz_long, &realpc, current_addr,
                  eainfo, 3, EA_STORE, 1);
        compile_fetchea(&map, eainfo, 3, 1);
        compile_storeea(&map, eainfo, 3, 0); /* A7 -> An */

        /* @@@ 020 */
        compile_prepareea(&map, Areg, 7, sz_long, &realpc, current_addr,
                  eainfo, 4, EA_LOAD, 1);
        compile_prepareea(&map, Areg, 7, sz_long, &realpc, current_addr,
                  eainfo, 5, EA_STORE, 1);
        compile_fetchea(&map, eainfo, 4, 1);
        eainfo[4].data_const_off += (uae_s16)eainfo[1].data_const_off;
        compile_storeea(&map, eainfo, 4, 5); /* A7+off -> A7 */
        cc_status = 0;
        break;

     case i_UNLK:
        compile_prepareea(&map, Areg,
                  insn_info[iip].dp->sreg,
                  sz_long, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        compile_prepareea(&map, Areg, 7, sz_long, &realpc, current_addr,
                  eainfo, 1, EA_STORE, 1);

        generate_possible_exit(&map, eainfo + 0, iip, &pub);

        compile_fetchea(&map, eainfo, 0, 1);
        compile_storeea(&map, eainfo, 0, 1);

        /* The Apdi could of course point to a non-memory area, but undos
         * are difficult here, and anyway: which program does evil hacks
         * with UNLK? */
        compile_prepareea(&map, Aipi, 7, sz_long, &realpc, current_addr,
                  eainfo, 2, EA_LOAD, 1);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  sz_long, &realpc, current_addr,
                  eainfo, 3, EA_STORE, 1);
        compile_fetchea(&map, eainfo, 2, 1);
        compile_storeea(&map, eainfo, 2, 3);

        cc_status = 0;
        break;

     case i_OR:
     case i_AND:
     case i_EOR:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepare_undo(&map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 1, EA_MODIFY | EA_LOAD | EA_STORE, 1);

        generate_possible_exit(&map, eainfo, iip, &pub);
        generate_possible_exit(&map, eainfo + 1, iip, &pub);

        compile_loadeas(&map, eainfo, 0, 1, binop_alternatives, 0, 1);

        switch (insn_info[iip].dp->mnemo) {
         case i_AND: compile_eas(&map, eainfo, 0, 1, 4); break;
         case i_EOR: compile_eas(&map, eainfo, 0, 1, 6); break;
         case i_OR:  compile_eas(&map, eainfo, 0, 1, 1); break;
        }

        compile_note_modify(&map, eainfo, 1);
        cc_status = CC_Z_FROM_86Z | CC_C_FROM_86C | CC_V_FROM_86V | CC_N_FROM_86N;
        break;

     case i_TST:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);

        generate_possible_exit(&map, eainfo, iip, &pub);

        compile_fetchea(&map, eainfo, 0, 1);
        cc_status = CC_TEST_REG;
        cc_reg = eainfo[0].data_reg;
        cc_offset = 0;
        cc_size = eainfo[0].size;
        break;

     case i_CLR:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_STORE, 1);
        compile_prepareea(&map, immi, 0, sz_long, &realpc, current_addr,
                  eainfo, 1, EA_LOAD, 1);
        generate_possible_exit(&map, eainfo + 0, iip, &pub);
        compile_loadeas(&map, eainfo, 1, 0, binop_alternatives, 1, 0);
        compile_storeea(&map, eainfo, 1, 0);

        cc_status = CC_TEST_REG;
        cc_reg = -2;
        cc_offset = 0;
        cc_size = eainfo[0].size;
        break;

     case i_EXT:
        /* No exits, no undo - this is always a Dreg; fetchea will get it in a reg
         * without offset */
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size == sz_long ? sz_word : sz_byte,
                  &realpc, current_addr,
                  eainfo, 0, EA_LOAD|EA_STORE, 1);
        compile_fetchea(&map, eainfo, 0, 1);
        compile_force_byteorder(&map, eainfo[0].data_reg, BO_NORMAL, 0);

        if (insn_info[iip].dp->size == sz_word)
        assemble(0x66);
        assemble(0x0F);
        if (insn_info[iip].dp->size == sz_long)
        assemble(0xBF);
        else
        assemble(0xBE);

        assemble(0xC0 + 9*eainfo[0].data_reg);
        map.x86_dirty[eainfo[0].data_reg] = 1;

        cc_status = CC_TEST_REG;
        cc_reg = eainfo[0].data_reg;
        cc_offset = 0;
        cc_size = eainfo[0].size;
        break;

     case i_NOT:
     case i_NEG:
        szflag = insn_info[iip].dp->size == sz_byte ? 0 : 1;

        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size,
                  &realpc, current_addr,
                  eainfo, 0, EA_LOAD|EA_STORE, 1);

        generate_possible_exit(&map, eainfo, iip, &pub);

        compile_fetchea(&map, eainfo, 0, 1);
        compile_force_byteorder(&map, eainfo[0].data_reg, BO_NORMAL, 0);

        if (insn_info[iip].dp->size == sz_word)
        assemble(0x66);
        assemble(0xF6 + szflag);

        assemble(0xC0 + eainfo[0].data_reg + 8*(insn_info[iip].dp->mnemo == i_NOT ? 2 : 3));
        compile_note_modify(&map, eainfo, 0);

        if (insn_info[iip].dp->mnemo == i_NEG)
        cc_status = CC_Z_FROM_86Z | CC_C_FROM_86C | CC_V_FROM_86V | CC_N_FROM_86N | CC_X_FROM_86C;
        else {
        cc_status = CC_TEST_REG;
        cc_reg = eainfo[0].data_reg;
        cc_offset = 0;
        cc_size = eainfo[0].size;
        }
        break;

     case i_SWAP:
        /* No exits, no undo - this is always a Dreg; fetchea will get it in a reg
         * without offset */
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg, sz_long,
                  &realpc, current_addr,
                  eainfo, 0, EA_LOAD|EA_STORE, 1);

        compile_fetchea(&map, eainfo, 0, 1);
        compile_force_byteorder(&map, eainfo[0].data_reg, BO_NORMAL, 0);

        /* roll $16, srcreg */
        assemble(0xC1); assemble(0xC0 + eainfo[0].data_reg); assemble(16);

        /* @@@ un-shortcut */
        map.x86_dirty[eainfo[0].data_reg] = 1;

        cc_status = CC_TEST_REG;
        cc_reg = eainfo[0].data_reg;
        cc_offset = 0;
        cc_size = eainfo[0].size;
        break;

     case i_LEA:
        /* No exits necessary here: never touches memory */
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, 0, 1);
        eainfo[0].data_reg = eainfo[0].address_reg;
        eainfo[0].data_const_off = eainfo[0].addr_const_off;
        eainfo[0].address_reg = -1;
        compile_get_excl_lock(&map, eainfo + 0);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  sz_long, &realpc, current_addr,
                  eainfo, 1, EA_STORE, 1);
        compile_storeea(&map, eainfo, 0, 1);
        cc_status = 0;
        break;

     case i_PEA:
        compile_prepare_undo(&map, Apdi, 7, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, 0, 1);
        eainfo[0].data_reg = eainfo[0].address_reg;
        eainfo[0].data_const_off = eainfo[0].addr_const_off;
        eainfo[0].address_reg = -1;
        compile_get_excl_lock(&map, eainfo + 0);
        compile_prepareea(&map, Apdi, 7, sz_long, &realpc, current_addr,
                  eainfo, 1, EA_STORE, 1);

        generate_possible_exit(&map, eainfo+1, iip, &pub);
        compile_storeea(&map, eainfo, 0, 1);

        cc_status = 0;
        break;

     case i_MVMEL:
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  sz_word, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        sync_reg_cache(&map, 0);
        {
        /* Scratch 0 holds the registers while they are being moved
         * from/to memory. Scratch 1 points at regs.d. Scratch 2
         * points at the base addr in memory where to fetch data
         * from.
         */
        int scratch0, scratch1, scratch2;
        uae_u16 mask = eainfo[0].data_const_off;
        int bits = count_bits(mask);
        int size = insn_info[iip].dp->size == sz_long ? 4 : 2;
        int i;
        uae_u8 x86amode;
        uae_u32 current_offs = 0;

        compile_prepare_undo(&map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg, &pub);
        /* !!! Note current_addr + 2 here! */
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  insn_info[iip].dp->size, &realpc, current_addr + 2,
                  eainfo, 1, EA_LOAD, bits);

        generate_possible_exit(&map, eainfo + 1, iip, &pub);

        scratch0 = get_free_x86_register(&map, ADDRESS_X86_REGS);
        lock_reg(&map, scratch0, 2);
        scratch1 = get_free_x86_register(&map, ADDRESS_X86_REGS);
        lock_reg(&map, scratch1, 2);
        scratch2 = get_free_x86_register(&map, ADDRESS_X86_REGS);
        lock_reg(&map, scratch2, 2);
        compile_force_byteorder(&map, eainfo[1].address_reg, BO_NORMAL, 0);

        compile_lea_reg_with_offset(scratch1, -2, (uae_u32)regs.regs);
        compile_lea_reg_with_offset(scratch2, eainfo[1].address_reg,
                        (uae_u32)(address_space + eainfo[1].addr_const_off));

        for (i = 0; i < 16; i++) {
            int r68k = i;
            int *cache68k = i < 8 ? map.dreg_map : map.areg_map;
            if (mask & 1
            && (i < 8
                || insn_info[iip].dp->dmode != Aipi
                || (r68k & 7) != insn_info[iip].dp->dreg)) {
            int tmpr = cache68k[r68k & 7];

            if (tmpr != -1) {
                cache68k[r68k & 7] = -1;
                map.x86_cache_reg[tmpr] = -1;
            }
            compile_move_reg_from_mem_regoffs(scratch0, scratch2,
                              current_offs, insn_info[iip].dp->size);
            if (size == 2) {
                assemble(0x66); /* rolw $8,scratch0 */
                assemble(0xC1);
                assemble(0xC0 + scratch0);
                assemble(8);
                assemble(0x0F); assemble(0xBF); /* extend */
                assemble(0xC0 + 9*scratch0);
            } else {
                assemble(0x0F); /* bswapl scratch0 */
                assemble(0xC8 + scratch0);
            }
            compile_move_reg_to_mem_regoffs(scratch1, (char *)(regs.regs + r68k) - (char *)regs.regs,
                            scratch0, sz_long);
            }
            if (mask & 1)
            current_offs += size;
            mask >>= 1;
        }
        }
        cc_status = 0;
        break;

     case i_MVMLE:
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  sz_word, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        sync_reg_cache(&map, 0);
        {
        int scratch0,scratch1,scratch2;
        uae_u16 mask = eainfo[0].data_const_off;
        int bits = count_bits(mask);
        int size = insn_info[iip].dp->size == sz_long ? 4 : 2;
        int i;
        uae_u8 x86amode;
        uae_u32 current_offs = 0;
        int addrareg = -1;
        if (insn_info[iip].dp->dmode == Aind
            || insn_info[iip].dp->dmode == Apdi
            || insn_info[iip].dp->dmode == Aipi
            || insn_info[iip].dp->dmode == Ad16
            || insn_info[iip].dp->dmode == Ad8r)
        {
            addrareg = get_and_lock_68k_reg(&map, insn_info[iip].dp->dreg, 0, ADDRESS_X86_REGS, 1, 2);
            compile_force_byteorder(&map, addrareg, BO_NORMAL, 0);
        }
        if (insn_info[iip].dp->dmode == Apdi)
            mask = bitswap(mask);
        /* !!! Note current_addr + 2 here! */
        compile_prepare_undo(&map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  insn_info[iip].dp->size, &realpc, current_addr + 2,
                  eainfo, 1, EA_STORE, bits);

        generate_possible_exit(&map, eainfo + 1, iip, &pub);

        scratch0 = get_free_x86_register(&map, ADDRESS_X86_REGS);
        lock_reg(&map, scratch0, 2);
        scratch1 = get_free_x86_register(&map, ADDRESS_X86_REGS);
        lock_reg(&map, scratch1, 2);
        scratch2 = get_free_x86_register(&map, ADDRESS_X86_REGS);
        lock_reg(&map, scratch2, 2);

        compile_force_byteorder(&map, eainfo[1].address_reg, BO_NORMAL, 0);

        compile_lea_reg_with_offset(scratch1, -2, (uae_u32)regs.regs);
        compile_lea_reg_with_offset(scratch2, eainfo[1].address_reg,
                        (uae_u32)(address_space + eainfo[1].addr_const_off));

        for (i = 0; i < 16; i++) {
            int r68k = i;
            if (mask & 1) {
            /* move from 68k reg */
            if (i < 8 || (i & 7) != insn_info[iip].dp->dreg || addrareg == -1) {
                compile_move_reg_from_mem_regoffs(scratch0, scratch1, (char *)(regs.regs + r68k) - (char *)regs.regs,
                                  sz_long);
            } else {
                assemble(0x8B); assemble(0xC0 + 8*scratch0 + addrareg);
            }

            if (size == 2) {
                assemble(0x66); /* rolw $8,scratch0 */
                assemble(0xC1);
                assemble(0xC0 + scratch0); assemble(8);
            } else {
                assemble(0x0F); /* bswapl scratch0 */
                assemble(0xC8 + scratch0);
            }
            compile_move_reg_to_mem_regoffs(scratch2, current_offs,
                            scratch0, insn_info[iip].dp->size);
            }
            if (mask & 1)
            current_offs += size;
            mask >>= 1;
        }
        }
        cc_status = 0;
        break;
#if 1
     case i_BTST:
     case i_BSET:
     case i_BCLR:
     case i_BCHG:
        compile_prepare_undo(&map, insn_info[iip].dp->smode, insn_info[iip].dp->sreg, &pub);
        compile_prepare_undo(&map, insn_info[iip].dp->dmode, insn_info[iip].dp->dreg, &pub);
        compile_prepareea(&map, insn_info[iip].dp->smode,
                  insn_info[iip].dp->sreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 0, EA_LOAD, 1);
        compile_prepareea(&map, insn_info[iip].dp->dmode,
                  insn_info[iip].dp->dreg,
                  insn_info[iip].dp->size, &realpc, current_addr,
                  eainfo, 1, 0, 1);

        generate_possible_exit(&map, eainfo, iip, &pub);
        generate_possible_exit(&map, eainfo + 1, iip, &pub);

        handle_bit_insns(&map, eainfo, 0, 1, insn_info[iip].dp->mnemo);
        break;

     case i_ASL: case i_ASR: case i_LSL: case i_LSR:
     case i_ROL: case i_ROR: case i_ROXL:case i_ROXR:
        if (insn_info[iip].dp->smode == Dreg && do_rotshi) {
        handle_rotshi_variable(&map, iip, realpc, current_addr, &pub);
        break;
        }
        /* fall through */
     case i_ASLW: case i_ASRW: case i_LSLW: case i_LSRW:
     case i_ROLW: case i_RORW: case i_ROXLW:case i_ROXRW:
        if (do_rotshi) {
        handle_rotshi(&map, iip, realpc, current_addr, &pub);
        break;
        }
#endif
     default:
        generate_exit(&map, insn_info[iip].address); cc_status = 0;
        break;
    }
    if (insn_info[iip].ccuser_follows)
        cc_status_for_bcc = compile_flush_cc_cache(&map, cc_status,
                   insn_info[iip].flags_live_at_end,
                   1, insn_info[iip+1].flags_live_at_end,
                   insn_info[iip+1].dp->cc);
    else
        cc_status_for_bcc = compile_flush_cc_cache(&map, cc_status,
                   insn_info[iip].flags_live_at_end,
                   0, 0, 0);

    if (iip == current_bb->last_iip) {
        current_bb++;
    }
    }
    if (compile_failure)
    goto oops;

    /* Compile all exits that we prepared earlier */
    finish_exits();
    if (compile_failure)
    goto oops;
    finish_condjumps(last_iip);
    {
    int needed_len = compile_here() - hb->compile_start;
    int allocsize = (needed_len + PAGE_SUBUNIT - 1) & ~(PAGE_SUBUNIT-1);
    uae_u32 allocmask;
    int allocbits;

    allocbits = (allocsize >> SUBUNIT_ORDER);
    allocmask = (1 << allocbits) - 1;
    while ((allocmask & hb->page_allocmask) != allocmask)
        allocmask <<= 1;
    if ((hb->page_allocmask & ~allocmask) != 0 && !quiet_compile)
        fprintf(stderr, "Gaining some bits: %08lx\n", hb->page_allocmask & ~allocmask);
    hb->cpage->allocmask &= ~hb->page_allocmask;
    hb->page_allocmask = allocmask;
    hb->cpage->allocmask |= allocmask;
    }
    return 0;

    oops:
    if (1 || !quiet_compile)
    fprintf(stderr, "Compile failed!\n");
    hb->cpage->allocmask &= ~hb->page_allocmask;
    hb->cpage = NULL;
    hb->untranslatable = 1;
    {
    struct hash_entry *h = hb->he_first;

    do {
        h->execute = NULL;
        h = h->next_same_block;
    } while (h != hb->he_first);
    }
    return 1;
}

void compiler_init(void)
{
    code_init();
    hash_init();
    jsr_stack_init();
}

/*
 * Why do compilers always have to be so complicated? And I thought GCC was
 * a mess...
 */

#endif /* USE_COMPILER */