X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fnew_dynarec.c;h=e0cff62ca622b1b5b04dd942626a76e89d4ab5f0;hp=21e7feda9b4051900b1085b782837c6fea6caac3;hb=e3c6bdb5e46f72f063bb7f588da6588ac1893b17;hpb=9f51b4b9aa4fffa7b9411ca274f5dfb179a6a30a

diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c
index 21e7feda..e0cff62c 100644
--- a/libpcsxcore/new_dynarec/new_dynarec.c
+++ b/libpcsxcore/new_dynarec/new_dynarec.c
@@ -23,8 +23,30 @@
 #include <assert.h>
 #include <errno.h>
 #include <sys/mman.h>
+#ifdef __MACH__
+#include <libkern/OSCacheControl.h>
+#endif
+#ifdef _3DS
+#include <3ds_utils.h>
+#endif
+#ifdef VITA
+#include <psp2/kernel/sysmem.h>
+static int sceBlock;
+#endif
 
-#include "emu_if.h" //emulator interface
+#include "new_dynarec_config.h"
+#include "../psxhle.h"
+#include "../psxinterpreter.h"
+#include "../gte.h"
+#include "emu_if.h" // emulator interface
+
+#define noinline __attribute__((noinline,noclone))
+#ifndef ARRAY_SIZE
+#define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
+#endif
+#ifndef min
+#define min(a, b) ((b) < (a) ? (b) : (a))
+#endif
 
 //#define DISASM
 //#define assem_debug printf
@@ -41,33 +63,56 @@
 #ifdef __arm__
 #include "assem_arm.h"
 #endif
-
-#ifdef __BLACKBERRY_QNX__
-#undef __clear_cache
-#define __clear_cache(start,end) msync(start, (size_t)((void*)end - (void*)start), MS_SYNC | MS_CACHE_ONLY | MS_INVALIDATE_ICACHE);
-#elif defined(__MACH__)
-#include <libkern/OSCacheControl.h>
-#define __clear_cache mach_clear_cache
-static void __clear_cache(void *start, void *end) {
-  size_t len = (char *)end - (char *)start;
-  sys_dcache_flush(start, len);
-  sys_icache_invalidate(start, len);
-}
+#ifdef __aarch64__
+#include "assem_arm64.h"
 #endif
 
+#define RAM_SIZE 0x200000
 #define MAXBLOCK 4096
 #define MAX_OUTPUT_BLOCK_SIZE 262144
 
+struct ndrc_mem
+{
+  u_char translation_cache[1 << TARGET_SIZE_2];
+  struct
+  {
+    struct tramp_insns ops[2048 / sizeof(struct tramp_insns)];
+    const void *f[2048 / sizeof(void *)];
+  } tramp;
+};
+
+#ifdef BASE_ADDR_DYNAMIC
+static struct ndrc_mem *ndrc;
+#else
+static struct ndrc_mem ndrc_ __attribute__((aligned(4096)));
+static struct ndrc_mem *ndrc = &ndrc_;
+#endif
+
+// stubs
+enum stub_type {
+  CC_STUB = 1,
+  FP_STUB = 2,
+  LOADB_STUB = 3,
+  LOADH_STUB = 4,
+  LOADW_STUB = 5,
+  LOADD_STUB = 6,
+  LOADBU_STUB = 7,
+  LOADHU_STUB = 8,
+  STOREB_STUB = 9,
+  STOREH_STUB = 10,
+  STOREW_STUB = 11,
+  STORED_STUB = 12,
+  STORELR_STUB = 13,
+  INVCODE_STUB = 14,
+};
+
 struct regstat
 {
   signed char regmap_entry[HOST_REGS];
   signed char regmap[HOST_REGS];
-  uint64_t was32;
-  uint64_t is32;
   uint64_t wasdirty;
   uint64_t dirty;
   uint64_t u;
-  uint64_t uu;
   u_int wasconst;
   u_int isconst;
   u_int loadedconst;             // host regs that have constants loaded
@@ -83,22 +128,52 @@ struct ll_entry
   struct ll_entry *next;
 };
 
-  u_int start;
-  u_int *source;
-  char insn[MAXBLOCK][10];
-  u_char itype[MAXBLOCK];
-  u_char opcode[MAXBLOCK];
-  u_char opcode2[MAXBLOCK];
-  u_char bt[MAXBLOCK];
-  u_char rs1[MAXBLOCK];
-  u_char rs2[MAXBLOCK];
-  u_char rt1[MAXBLOCK];
-  u_char rt2[MAXBLOCK];
-  u_char us1[MAXBLOCK];
-  u_char us2[MAXBLOCK];
-  u_char dep1[MAXBLOCK];
-  u_char dep2[MAXBLOCK];
-  u_char lt1[MAXBLOCK];
+struct ht_entry
+{
+  u_int vaddr[2];
+  void *tcaddr[2];
+};
+
+struct code_stub
+{
+  enum stub_type type;
+  void *addr;
+  void *retaddr;
+  u_int a;
+  uintptr_t b;
+  uintptr_t c;
+  u_int d;
+  u_int e;
+};
+
+struct link_entry
+{
+  void *addr;
+  u_int target;
+  u_int ext;
+};
+
+  // used by asm:
+  u_char *out;
+  struct ht_entry hash_table[65536]  __attribute__((aligned(16)));
+  struct ll_entry *jump_in[4096] __attribute__((aligned(16)));
+  struct ll_entry *jump_dirty[4096];
+
+  static struct ll_entry *jump_out[4096];
+  static u_int start;
+  static u_int *source;
+  static char insn[MAXBLOCK][10];
+  static u_char itype[MAXBLOCK];
+  static u_char opcode[MAXBLOCK];
+  static u_char opcode2[MAXBLOCK];
+  static u_char bt[MAXBLOCK];
+  static u_char rs1[MAXBLOCK];
+  static u_char rs2[MAXBLOCK];
+  static u_char rt1[MAXBLOCK];
+  static u_char rt2[MAXBLOCK];
+  static u_char dep1[MAXBLOCK];
+  static u_char dep2[MAXBLOCK];
+  static u_char lt1[MAXBLOCK];
   static uint64_t gte_rs[MAXBLOCK]; // gte: 32 data and 32 ctl regs
   static uint64_t gte_rt[MAXBLOCK];
   static uint64_t gte_unneeded[MAXBLOCK];
@@ -107,65 +182,68 @@ struct ll_entry
   static u_int smrv_weak; // same, but somewhat less likely
   static u_int smrv_strong_next; // same, but after current insn executes
   static u_int smrv_weak_next;
-  int imm[MAXBLOCK];
-  u_int ba[MAXBLOCK];
-  char likely[MAXBLOCK];
-  char is_ds[MAXBLOCK];
-  char ooo[MAXBLOCK];
-  uint64_t unneeded_reg[MAXBLOCK];
-  uint64_t unneeded_reg_upper[MAXBLOCK];
-  uint64_t branch_unneeded_reg[MAXBLOCK];
-  uint64_t branch_unneeded_reg_upper[MAXBLOCK];
-  uint64_t pr32[MAXBLOCK];
-  signed char regmap_pre[MAXBLOCK][HOST_REGS];
-  static uint64_t current_constmap[HOST_REGS];
-  static uint64_t constmap[MAXBLOCK][HOST_REGS];
+  static int imm[MAXBLOCK];
+  static u_int ba[MAXBLOCK];
+  static char likely[MAXBLOCK];
+  static char is_ds[MAXBLOCK];
+  static char ooo[MAXBLOCK];
+  static uint64_t unneeded_reg[MAXBLOCK];
+  static uint64_t branch_unneeded_reg[MAXBLOCK];
+  static signed char regmap_pre[MAXBLOCK][HOST_REGS]; // pre-instruction i?
+  // contains 'real' consts at [i] insn, but may differ from what's actually
+  // loaded in host reg as 'final' value is always loaded, see get_final_value()
+  static uint32_t current_constmap[HOST_REGS];
+  static uint32_t constmap[MAXBLOCK][HOST_REGS];
   static struct regstat regs[MAXBLOCK];
   static struct regstat branch_regs[MAXBLOCK];
-  signed char minimum_free_regs[MAXBLOCK];
-  u_int needed_reg[MAXBLOCK];
-  u_int wont_dirty[MAXBLOCK];
-  u_int will_dirty[MAXBLOCK];
-  int ccadj[MAXBLOCK];
-  int slen;
-  u_int instr_addr[MAXBLOCK];
-  u_int link_addr[MAXBLOCK][3];
-  int linkcount;
-  u_int stubs[MAXBLOCK*3][8];
-  int stubcount;
-  u_int literals[1024][2];
-  int literalcount;
-  int is_delayslot;
-  int cop1_usable;
-  u_char *out;
-  struct ll_entry *jump_in[4096] __attribute__((aligned(16)));
-  struct ll_entry *jump_out[4096];
-  struct ll_entry *jump_dirty[4096];
-  u_int hash_table[65536][4]  __attribute__((aligned(16)));
-  char shadow[1048576]  __attribute__((aligned(16)));
-  void *copy;
-  int expirep;
-  int new_dynarec_did_compile;
-  int new_dynarec_hacks;
-  u_int stop_after_jal;
+  static signed char minimum_free_regs[MAXBLOCK];
+  static u_int needed_reg[MAXBLOCK];
+  static u_int wont_dirty[MAXBLOCK];
+  static u_int will_dirty[MAXBLOCK];
+  static int ccadj[MAXBLOCK];
+  static int slen;
+  static void *instr_addr[MAXBLOCK];
+  static struct link_entry link_addr[MAXBLOCK];
+  static int linkcount;
+  static struct code_stub stubs[MAXBLOCK*3];
+  static int stubcount;
+  static u_int literals[1024][2];
+  static int literalcount;
+  static int is_delayslot;
+  static char shadow[1048576]  __attribute__((aligned(16)));
+  static void *copy;
+  static int expirep;
+  static u_int stop_after_jal;
 #ifndef RAM_FIXED
-  static u_int ram_offset;
+  static uintptr_t ram_offset;
 #else
-  static const u_int ram_offset=0;
+  static const uintptr_t ram_offset=0;
 #endif
+
+  int new_dynarec_hacks;
+  int new_dynarec_hacks_pergame;
+  int new_dynarec_did_compile;
+
+  #define HACK_ENABLED(x) ((new_dynarec_hacks | new_dynarec_hacks_pergame) & (x))
+
+  extern int cycle_count; // ... until end of the timeslice, counts -N -> 0
+  extern int last_count;  // last absolute target, often = next_interupt
+  extern int pcaddr;
+  extern int pending_exception;
+  extern int branch_target;
+  extern uintptr_t mini_ht[32][2];
   extern u_char restore_candidate[512];
-  extern int cycle_count;
 
   /* registers that may be allocated */
   /* 1-31 gpr */
-#define HIREG 32 // hi
-#define LOREG 33 // lo
-#define FSREG 34 // FPU status (FCSR)
+#define LOREG 32 // lo
+#define HIREG 33 // hi
+//#define FSREG 34 // FPU status (FCSR)
 #define CSREG 35 // Coprocessor status
 #define CCREG 36 // Cycle count
 #define INVCP 37 // Pointer to invalid_code
 //#define MMREG 38 // Pointer to memory_map
-#define ROREG 39 // ram offset (if rdram!=0x80000000)
+//#define ROREG 39 // ram offset (if rdram!=0x80000000)
 #define TEMPREG 40
 #define FTEMP 40 // FPU temporary register
 #define PTEMP 41 // Prefetch temporary register
@@ -199,10 +277,10 @@ struct ll_entry
 #define COP0 15   // Coprocessor 0
 #define COP1 16   // Coprocessor 1
 #define C1LS 17   // Coprocessor 1 load/store
-#define FJUMP 18  // Conditional branch (floating point)
-#define FLOAT 19  // Floating point unit
-#define FCONV 20  // Convert integer to float
-#define FCOMP 21  // Floating point compare (sets FSREG)
+//#define FJUMP 18  // Conditional branch (floating point)
+//#define FLOAT 19  // Floating point unit
+//#define FCONV 20  // Convert integer to float
+//#define FCOMP 21  // Floating point compare (sets FSREG)
 #define SYSCALL 22// SYSCALL
 #define OTHER 23  // Other
 #define SPAN 24   // Branch/delay slot spans 2 pages
@@ -213,29 +291,16 @@ struct ll_entry
 #define C2OP 29   // Coprocessor 2 operation
 #define INTCALL 30// Call interpreter to handle rare corner cases
 
-  /* stubs */
-#define CC_STUB 1
-#define FP_STUB 2
-#define LOADB_STUB 3
-#define LOADH_STUB 4
-#define LOADW_STUB 5
-#define LOADD_STUB 6
-#define LOADBU_STUB 7
-#define LOADHU_STUB 8
-#define STOREB_STUB 9
-#define STOREH_STUB 10
-#define STOREW_STUB 11
-#define STORED_STUB 12
-#define STORELR_STUB 13
-#define INVCODE_STUB 14
-
   /* branch codes */
 #define TAKEN 1
 #define NOTTAKEN 2
 #define NULLDS 3
 
+#define DJT_1 (void *)1l // no function, just a label in assem_debug log
+#define DJT_2 (void *)2l
+
 // asm linkage
-int new_recompile_block(int addr);
+int new_recompile_block(u_int addr);
 void *get_addr_ht(u_int vaddr);
 void invalidate_block(u_int block);
 void invalidate_addr(u_int addr);
@@ -243,37 +308,171 @@ void remove_hash(int vaddr);
 void dyna_linker();
 void dyna_linker_ds();
 void verify_code();
-void verify_code_vm();
 void verify_code_ds();
 void cc_interrupt();
 void fp_exception();
 void fp_exception_ds();
-void jump_syscall_hle();
-void jump_hlecall();
-void jump_intcall();
+void jump_to_new_pc();
+void call_gteStall();
 void new_dyna_leave();
 
 // Needed by assembler
-void wb_register(signed char r,signed char regmap[],uint64_t dirty,uint64_t is32);
-void wb_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty);
-void wb_needed_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr);
-void load_all_regs(signed char i_regmap[]);
-void load_needed_regs(signed char i_regmap[],signed char next_regmap[]);
-void load_regs_entry(int t);
-void load_all_consts(signed char regmap[],int is32,u_int dirty,int i);
+static void wb_register(signed char r,signed char regmap[],uint64_t dirty);
+static void wb_dirtys(signed char i_regmap[],uint64_t i_dirty);
+static void wb_needed_dirtys(signed char i_regmap[],uint64_t i_dirty,int addr);
+static void load_all_regs(signed char i_regmap[]);
+static void load_needed_regs(signed char i_regmap[],signed char next_regmap[]);
+static void load_regs_entry(int t);
+static void load_all_consts(signed char regmap[],u_int dirty,int i);
+static u_int get_host_reglist(const signed char *regmap);
+
+static int verify_dirty(const u_int *ptr);
+static int get_final_value(int hr, int i, int *value);
+static void add_stub(enum stub_type type, void *addr, void *retaddr,
+  u_int a, uintptr_t b, uintptr_t c, u_int d, u_int e);
+static void add_stub_r(enum stub_type type, void *addr, void *retaddr,
+  int i, int addr_reg, const struct regstat *i_regs, int ccadj, u_int reglist);
+static void add_to_linker(void *addr, u_int target, int ext);
+static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override);
+static void *get_direct_memhandler(void *table, u_int addr,
+  enum stub_type type, uintptr_t *addr_host);
+static void cop2_call_stall_check(u_int op, int i, const struct regstat *i_regs, u_int reglist);
+static void pass_args(int a0, int a1);
+static void emit_far_jump(const void *f);
+static void emit_far_call(const void *f);
+
+static void mprotect_w_x(void *start, void *end, int is_x)
+{
+#ifdef NO_WRITE_EXEC
+  #if defined(VITA)
+  // *Open* enables write on all memory that was
+  // allocated by sceKernelAllocMemBlockForVM()?
+  if (is_x)
+    sceKernelCloseVMDomain();
+  else
+    sceKernelOpenVMDomain();
+  #else
+  u_long mstart = (u_long)start & ~4095ul;
+  u_long mend = (u_long)end;
+  if (mprotect((void *)mstart, mend - mstart,
+               PROT_READ | (is_x ? PROT_EXEC : PROT_WRITE)) != 0)
+    SysPrintf("mprotect(%c) failed: %s\n", is_x ? 'x' : 'w', strerror(errno));
+  #endif
+#endif
+}
+
+static void start_tcache_write(void *start, void *end)
+{
+  mprotect_w_x(start, end, 0);
+}
+
+static void end_tcache_write(void *start, void *end)
+{
+#if defined(__arm__) || defined(__aarch64__)
+  size_t len = (char *)end - (char *)start;
+  #if   defined(__BLACKBERRY_QNX__)
+  msync(start, len, MS_SYNC | MS_CACHE_ONLY | MS_INVALIDATE_ICACHE);
+  #elif defined(__MACH__)
+  sys_cache_control(kCacheFunctionPrepareForExecution, start, len);
+  #elif defined(VITA)
+  sceKernelSyncVMDomain(sceBlock, start, len);
+  #elif defined(_3DS)
+  ctr_flush_invalidate_cache();
+  #elif defined(__aarch64__)
+  // as of 2021, __clear_cache() is still broken on arm64
+  // so here is a custom one :(
+  clear_cache_arm64(start, end);
+  #else
+  __clear_cache(start, end);
+  #endif
+  (void)len;
+#endif
+
+  mprotect_w_x(start, end, 1);
+}
 
-int tracedebug=0;
+static void *start_block(void)
+{
+  u_char *end = out + MAX_OUTPUT_BLOCK_SIZE;
+  if (end > ndrc->translation_cache + sizeof(ndrc->translation_cache))
+    end = ndrc->translation_cache + sizeof(ndrc->translation_cache);
+  start_tcache_write(out, end);
+  return out;
+}
+
+static void end_block(void *start)
+{
+  end_tcache_write(start, out);
+}
+
+// also takes care of w^x mappings when patching code
+static u_int needs_clear_cache[1<<(TARGET_SIZE_2-17)];
+
+static void mark_clear_cache(void *target)
+{
+  uintptr_t offset = (u_char *)target - ndrc->translation_cache;
+  u_int mask = 1u << ((offset >> 12) & 31);
+  if (!(needs_clear_cache[offset >> 17] & mask)) {
+    char *start = (char *)((uintptr_t)target & ~4095l);
+    start_tcache_write(start, start + 4095);
+    needs_clear_cache[offset >> 17] |= mask;
+  }
+}
+
+// Clearing the cache is rather slow on ARM Linux, so mark the areas
+// that need to be cleared, and then only clear these areas once.
+static void do_clear_cache(void)
+{
+  int i, j;
+  for (i = 0; i < (1<<(TARGET_SIZE_2-17)); i++)
+  {
+    u_int bitmap = needs_clear_cache[i];
+    if (!bitmap)
+      continue;
+    for (j = 0; j < 32; j++)
+    {
+      u_char *start, *end;
+      if (!(bitmap & (1<<j)))
+        continue;
+
+      start = ndrc->translation_cache + i*131072 + j*4096;
+      end = start + 4095;
+      for (j++; j < 32; j++) {
+        if (!(bitmap & (1<<j)))
+          break;
+        end += 4096;
+      }
+      end_tcache_write(start, end);
+    }
+    needs_clear_cache[i] = 0;
+  }
+}
 
 //#define DEBUG_CYCLE_COUNT 1
 
 #define NO_CYCLE_PENALTY_THR 12
 
 int cycle_multiplier; // 100 for 1.0
+int cycle_multiplier_override;
 
 static int CLOCK_ADJUST(int x)
 {
+  int m = cycle_multiplier_override
+        ? cycle_multiplier_override : cycle_multiplier;
   int s=(x>>31)|1;
-  return (x * cycle_multiplier + s * 50) / 100;
+  return (x * m + s * 50) / 100;
+}
+
+// is the op an unconditional jump?
+static int is_ujump(int i)
+{
+  return itype[i] == UJUMP || itype[i] == RJUMP
+    || (source[i] >> 16) == 0x1000; // beq r0, r0, offset // b offset
+}
+
+static int is_jump(int i)
+{
+  return itype[i] == RJUMP || itype[i] == UJUMP || itype[i] == CJUMP || itype[i] == SJUMP;
 }
 
 static u_int get_page(u_int vaddr)
@@ -292,9 +491,29 @@ static u_int get_vpage(u_int vaddr)
   return get_page(vaddr);
 }
 
+static struct ht_entry *hash_table_get(u_int vaddr)
+{
+  return &hash_table[((vaddr>>16)^vaddr)&0xFFFF];
+}
+
+static void hash_table_add(struct ht_entry *ht_bin, u_int vaddr, void *tcaddr)
+{
+  ht_bin->vaddr[1] = ht_bin->vaddr[0];
+  ht_bin->tcaddr[1] = ht_bin->tcaddr[0];
+  ht_bin->vaddr[0] = vaddr;
+  ht_bin->tcaddr[0] = tcaddr;
+}
+
+// some messy ari64's code, seems to rely on unsigned 32bit overflow
+static int doesnt_expire_soon(void *tcaddr)
+{
+  u_int diff = (u_int)((u_char *)tcaddr - out) << (32-TARGET_SIZE_2);
+  return diff > (u_int)(0x60000000 + (MAX_OUTPUT_BLOCK_SIZE << (32-TARGET_SIZE_2)));
+}
+
 // Get address from virtual address
 // This is called from the recompiled JR/JALR instructions
-void *get_addr(u_int vaddr)
+void noinline *get_addr(u_int vaddr)
 {
   u_int page=get_page(vaddr);
   u_int vpage=get_vpage(vaddr);
@@ -303,12 +522,8 @@ void *get_addr(u_int vaddr)
   head=jump_in[page];
   while(head!=NULL) {
     if(head->vaddr==vaddr) {
-  //printf("TRACE: count=%d next=%d (get_addr match %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr);
-      int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
-      ht_bin[3]=ht_bin[1];
-      ht_bin[2]=ht_bin[0];
-      ht_bin[1]=(int)head->addr;
-      ht_bin[0]=vaddr;
+  //printf("TRACE: count=%d next=%d (get_addr match %x: %p)\n",Count,next_interupt,vaddr,head->addr);
+      hash_table_add(hash_table_get(vaddr), vaddr, head->addr);
       return head->addr;
     }
     head=head->next;
@@ -316,10 +531,10 @@ void *get_addr(u_int vaddr)
   head=jump_dirty[vpage];
   while(head!=NULL) {
     if(head->vaddr==vaddr) {
-      //printf("TRACE: count=%d next=%d (get_addr match dirty %x: %x)\n",Count,next_interupt,vaddr,(int)head->addr);
+      //printf("TRACE: count=%d next=%d (get_addr match dirty %x: %p)\n",Count,next_interupt,vaddr,head->addr);
       // Don't restore blocks which are about to expire from the cache
-      if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2)))
-      if(verify_dirty(head->addr)) {
+      if (doesnt_expire_soon(head->addr))
+      if (verify_dirty(head->addr)) {
         //printf("restore candidate: %x (%d) d=%d\n",vaddr,page,invalid_code[vaddr>>12]);
         invalid_code[vaddr>>12]=0;
         inv_code_start=inv_code_end=~0;
@@ -327,17 +542,12 @@ void *get_addr(u_int vaddr)
           restore_candidate[vpage>>3]|=1<<(vpage&7);
         }
         else restore_candidate[page>>3]|=1<<(page&7);
-        int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
-        if(ht_bin[0]==vaddr) {
-          ht_bin[1]=(int)head->addr; // Replace existing entry
-        }
+        struct ht_entry *ht_bin = hash_table_get(vaddr);
+        if (ht_bin->vaddr[0] == vaddr)
+          ht_bin->tcaddr[0] = head->addr; // Replace existing entry
         else
-        {
-          ht_bin[3]=ht_bin[1];
-          ht_bin[2]=ht_bin[0];
-          ht_bin[1]=(int)head->addr;
-          ht_bin[0]=vaddr;
-        }
+          hash_table_add(ht_bin, vaddr, head->addr);
+
         return head->addr;
       }
     }
@@ -359,9 +569,9 @@ void *get_addr(u_int vaddr)
 void *get_addr_ht(u_int vaddr)
 {
   //printf("TRACE: count=%d next=%d (get_addr_ht %x)\n",Count,next_interupt,vaddr);
-  int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
-  if(ht_bin[0]==vaddr) return (void *)ht_bin[1];
-  if(ht_bin[2]==vaddr) return (void *)ht_bin[3];
+  const struct ht_entry *ht_bin = hash_table_get(vaddr);
+  if (ht_bin->vaddr[0] == vaddr) return ht_bin->tcaddr[0];
+  if (ht_bin->vaddr[1] == vaddr) return ht_bin->tcaddr[1];
   return get_addr(vaddr);
 }
 
@@ -371,7 +581,7 @@ void clear_all_regs(signed char regmap[])
   for (hr=0;hr<HOST_REGS;hr++) regmap[hr]=-1;
 }
 
-signed char get_reg(signed char regmap[],int r)
+static signed char get_reg(const signed char regmap[],int r)
 {
   int hr;
   for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&&regmap[hr]==r) return hr;
@@ -379,7 +589,7 @@ signed char get_reg(signed char regmap[],int r)
 }
 
 // Find a register that is available for two consecutive cycles
-signed char get_reg2(signed char regmap1[],signed char regmap2[],int r)
+static signed char get_reg2(signed char regmap1[], const signed char regmap2[], int r)
 {
   int hr;
   for (hr=0;hr<HOST_REGS;hr++) if(hr!=EXCLUDE_REG&&regmap1[hr]==r&&regmap2[hr]==r) return hr;
@@ -410,24 +620,7 @@ void dirty_reg(struct regstat *cur,signed char reg)
   }
 }
 
-// If we dirty the lower half of a 64 bit register which is now being
-// sign-extended, we need to dump the upper half.
-// Note: Do this only after completion of the instruction, because
-// some instructions may need to read the full 64-bit value even if
-// overwriting it (eg SLTI, DSRA32).
-static void flush_dirty_uppers(struct regstat *cur)
-{
-  int hr,reg;
-  for (hr=0;hr<HOST_REGS;hr++) {
-    if((cur->dirty>>hr)&1) {
-      reg=cur->regmap[hr];
-      if(reg>=64)
-        if((cur->is32>>(reg&63))&1) cur->regmap[hr]=-1;
-    }
-  }
-}
-
-void set_const(struct regstat *cur,signed char reg,uint64_t value)
+static void set_const(struct regstat *cur, signed char reg, uint32_t value)
 {
   int hr;
   if(!reg) return;
@@ -436,14 +629,10 @@ void set_const(struct regstat *cur,signed char reg,uint64_t value)
       cur->isconst|=1<<hr;
       current_constmap[hr]=value;
     }
-    else if((cur->regmap[hr]^64)==reg) {
-      cur->isconst|=1<<hr;
-      current_constmap[hr]=value>>32;
-    }
   }
 }
 
-void clear_const(struct regstat *cur,signed char reg)
+static void clear_const(struct regstat *cur, signed char reg)
 {
   int hr;
   if(!reg) return;
@@ -454,7 +643,7 @@ void clear_const(struct regstat *cur,signed char reg)
   }
 }
 
-int is_const(struct regstat *cur,signed char reg)
+static int is_const(struct regstat *cur, signed char reg)
 {
   int hr;
   if(reg<0) return 0;
@@ -466,7 +655,8 @@ int is_const(struct regstat *cur,signed char reg)
   }
   return 0;
 }
-uint64_t get_const(struct regstat *cur,signed char reg)
+
+static uint32_t get_const(struct regstat *cur, signed char reg)
 {
   int hr;
   if(!reg) return 0;
@@ -476,7 +666,7 @@ uint64_t get_const(struct regstat *cur,signed char reg)
     }
   }
   SysPrintf("Unknown constant in r%d\n",reg);
-  exit(1);
+  abort();
 }
 
 // Least soon needed registers
@@ -492,7 +682,7 @@ void lsn(u_char hsn[], int i, int *preferred_reg)
       j=slen-i-1;
       break;
     }
-    if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000)
+    if (is_ujump(i+j))
     {
       // Don't go past an unconditonal jump
       j++;
@@ -516,7 +706,7 @@ void lsn(u_char hsn[], int i, int *preferred_reg)
       hsn[INVCP]=j;
     }
     #endif
-    if(i+j>=0&&(itype[i+j]==UJUMP||itype[i+j]==CJUMP||itype[i+j]==SJUMP||itype[i+j]==FJUMP))
+    if(i+j>=0&&(itype[i+j]==UJUMP||itype[i+j]==CJUMP||itype[i+j]==SJUMP))
     {
       hsn[CCREG]=j;
       b=j;
@@ -540,7 +730,7 @@ void lsn(u_char hsn[], int i, int *preferred_reg)
     // TODO: preferred register based on backward branch
   }
   // Delay slot should preferably not overwrite branch conditions or cycle count
-  if(i>0&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP)) {
+  if (i > 0 && is_jump(i-1)) {
     if(rs1[i-1]) if(hsn[rs1[i-1]]>1) hsn[rs1[i-1]]=1;
     if(rs2[i-1]) if(hsn[rs2[i-1]]>1) hsn[rs2[i-1]]=1;
     hsn[CCREG]=1;
@@ -575,7 +765,7 @@ int needed_again(int r, int i)
   int b=-1;
   int rn=10;
 
-  if(i>0&&(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000))
+  if (i > 0 && is_ujump(i-1))
   {
     if(ba[i-1]<start || ba[i-1]>start+slen*4-4)
       return 0; // Don't need any registers if exiting the block
@@ -586,7 +776,7 @@ int needed_again(int r, int i)
       j=slen-i-1;
       break;
     }
-    if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000)
+    if (is_ujump(i+j))
     {
       // Don't go past an unconditonal jump
       j++;
@@ -602,7 +792,7 @@ int needed_again(int r, int i)
     if(rs1[i+j]==r) rn=j;
     if(rs2[i+j]==r) rn=j;
     if((unneeded_reg[i+j]>>r)&1) rn=10;
-    if(i+j>=0&&(itype[i+j]==UJUMP||itype[i+j]==CJUMP||itype[i+j]==SJUMP||itype[i+j]==FJUMP))
+    if(i+j>=0&&(itype[i+j]==UJUMP||itype[i+j]==CJUMP||itype[i+j]==SJUMP))
     {
       b=j;
     }
@@ -627,6 +817,7 @@ int needed_again(int r, int i)
     }
   }*/
   if(rn<10) return 1;
+  (void)b;
   return 0;
 }
 
@@ -641,7 +832,7 @@ int loop_reg(int i, int r, int hr)
       j=slen-i-1;
       break;
     }
-    if(itype[i+j]==UJUMP||itype[i+j]==RJUMP||(source[i+j]>>16)==0x1000)
+    if (is_ujump(i+j))
     {
       // Don't go past an unconditonal jump
       j++;
@@ -650,14 +841,14 @@ int loop_reg(int i, int r, int hr)
   }
   k=0;
   if(i>0){
-    if(itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP)
+    if(itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP)
       k--;
   }
   for(;k<j;k++)
   {
-    if(r<64&&((unneeded_reg[i+k]>>r)&1)) return hr;
-    if(r>64&&((unneeded_reg_upper[i+k]>>r)&1)) return hr;
-    if(i+k>=0&&(itype[i+k]==UJUMP||itype[i+k]==CJUMP||itype[i+k]==SJUMP||itype[i+k]==FJUMP))
+    assert(r < 64);
+    if((unneeded_reg[i+k]>>r)&1) return hr;
+    if(i+k>=0&&(itype[i+k]==UJUMP||itype[i+k]==CJUMP||itype[i+k]==SJUMP))
     {
       if(ba[i+k]>=start && ba[i+k]<(start+i*4))
       {
@@ -696,6 +887,66 @@ void alloc_all(struct regstat *cur,int i)
   }
 }
 
+#ifndef NDEBUG
+static int host_tempreg_in_use;
+
+static void host_tempreg_acquire(void)
+{
+  assert(!host_tempreg_in_use);
+  host_tempreg_in_use = 1;
+}
+
+static void host_tempreg_release(void)
+{
+  host_tempreg_in_use = 0;
+}
+#else
+static void host_tempreg_acquire(void) {}
+static void host_tempreg_release(void) {}
+#endif
+
+#ifdef DRC_DBG
+extern void gen_interupt();
+extern void do_insn_cmp();
+#define FUNCNAME(f) { f, " " #f }
+static const struct {
+  void *addr;
+  const char *name;
+} function_names[] = {
+  FUNCNAME(cc_interrupt),
+  FUNCNAME(gen_interupt),
+  FUNCNAME(get_addr_ht),
+  FUNCNAME(get_addr),
+  FUNCNAME(jump_handler_read8),
+  FUNCNAME(jump_handler_read16),
+  FUNCNAME(jump_handler_read32),
+  FUNCNAME(jump_handler_write8),
+  FUNCNAME(jump_handler_write16),
+  FUNCNAME(jump_handler_write32),
+  FUNCNAME(invalidate_addr),
+  FUNCNAME(jump_to_new_pc),
+  FUNCNAME(call_gteStall),
+  FUNCNAME(new_dyna_leave),
+  FUNCNAME(pcsx_mtc0),
+  FUNCNAME(pcsx_mtc0_ds),
+  FUNCNAME(do_insn_cmp),
+#ifdef __arm__
+  FUNCNAME(verify_code),
+#endif
+};
+
+static const char *func_name(const void *a)
+{
+  int i;
+  for (i = 0; i < sizeof(function_names)/sizeof(function_names[0]); i++)
+    if (function_names[i].addr == a)
+      return function_names[i].name;
+  return "";
+}
+#else
+#define func_name(x) ""
+#endif
+
 #ifdef __i386__
 #include "assem_x86.c"
 #endif
@@ -705,6 +956,51 @@ void alloc_all(struct regstat *cur,int i)
 #ifdef __arm__
 #include "assem_arm.c"
 #endif
+#ifdef __aarch64__
+#include "assem_arm64.c"
+#endif
+
+static void *get_trampoline(const void *f)
+{
+  size_t i;
+
+  for (i = 0; i < ARRAY_SIZE(ndrc->tramp.f); i++) {
+    if (ndrc->tramp.f[i] == f || ndrc->tramp.f[i] == NULL)
+      break;
+  }
+  if (i == ARRAY_SIZE(ndrc->tramp.f)) {
+    SysPrintf("trampoline table is full, last func %p\n", f);
+    abort();
+  }
+  if (ndrc->tramp.f[i] == NULL) {
+    start_tcache_write(&ndrc->tramp.f[i], &ndrc->tramp.f[i + 1]);
+    ndrc->tramp.f[i] = f;
+    end_tcache_write(&ndrc->tramp.f[i], &ndrc->tramp.f[i + 1]);
+  }
+  return &ndrc->tramp.ops[i];
+}
+
+static void emit_far_jump(const void *f)
+{
+  if (can_jump_or_call(f)) {
+    emit_jmp(f);
+    return;
+  }
+
+  f = get_trampoline(f);
+  emit_jmp(f);
+}
+
+static void emit_far_call(const void *f)
+{
+  if (can_jump_or_call(f)) {
+    emit_call(f);
+    return;
+  }
+
+  f = get_trampoline(f);
+  emit_call(f);
+}
 
 // Add virtual address mapping to linked list
 void ll_add(struct ll_entry **head,int vaddr,void *addr)
@@ -729,39 +1025,39 @@ void ll_add_flags(struct ll_entry **head,int vaddr,u_int reg_sv_flags,void *addr
 // but don't return addresses which are about to expire from the cache
 void *check_addr(u_int vaddr)
 {
-  u_int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
-  if(ht_bin[0]==vaddr) {
-    if(((ht_bin[1]-MAX_OUTPUT_BLOCK_SIZE-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2)))
-      if(isclean(ht_bin[1])) return (void *)ht_bin[1];
-  }
-  if(ht_bin[2]==vaddr) {
-    if(((ht_bin[3]-MAX_OUTPUT_BLOCK_SIZE-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2)))
-      if(isclean(ht_bin[3])) return (void *)ht_bin[3];
+  struct ht_entry *ht_bin = hash_table_get(vaddr);
+  size_t i;
+  for (i = 0; i < ARRAY_SIZE(ht_bin->vaddr); i++) {
+    if (ht_bin->vaddr[i] == vaddr)
+      if (doesnt_expire_soon((u_char *)ht_bin->tcaddr[i] - MAX_OUTPUT_BLOCK_SIZE))
+        if (isclean(ht_bin->tcaddr[i]))
+          return ht_bin->tcaddr[i];
   }
   u_int page=get_page(vaddr);
   struct ll_entry *head;
   head=jump_in[page];
-  while(head!=NULL) {
-    if(head->vaddr==vaddr) {
-      if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) {
+  while (head != NULL) {
+    if (head->vaddr == vaddr) {
+      if (doesnt_expire_soon(head->addr)) {
         // Update existing entry with current address
-        if(ht_bin[0]==vaddr) {
-          ht_bin[1]=(int)head->addr;
+        if (ht_bin->vaddr[0] == vaddr) {
+          ht_bin->tcaddr[0] = head->addr;
           return head->addr;
         }
-        if(ht_bin[2]==vaddr) {
-          ht_bin[3]=(int)head->addr;
+        if (ht_bin->vaddr[1] == vaddr) {
+          ht_bin->tcaddr[1] = head->addr;
           return head->addr;
         }
         // Insert into hash table with low priority.
         // Don't evict existing entries, as they are probably
         // addresses that are being accessed frequently.
-        if(ht_bin[0]==-1) {
-          ht_bin[1]=(int)head->addr;
-          ht_bin[0]=vaddr;
-        }else if(ht_bin[2]==-1) {
-          ht_bin[3]=(int)head->addr;
-          ht_bin[2]=vaddr;
+        if (ht_bin->vaddr[0] == -1) {
+          ht_bin->vaddr[0] = vaddr;
+          ht_bin->tcaddr[0] = head->addr;
+        }
+        else if (ht_bin->vaddr[1] == -1) {
+          ht_bin->vaddr[1] = vaddr;
+          ht_bin->tcaddr[1] = head->addr;
         }
         return head->addr;
       }
@@ -774,25 +1070,27 @@ void *check_addr(u_int vaddr)
 void remove_hash(int vaddr)
 {
   //printf("remove hash: %x\n",vaddr);
-  int *ht_bin=hash_table[(((vaddr)>>16)^vaddr)&0xFFFF];
-  if(ht_bin[2]==vaddr) {
-    ht_bin[2]=ht_bin[3]=-1;
+  struct ht_entry *ht_bin = hash_table_get(vaddr);
+  if (ht_bin->vaddr[1] == vaddr) {
+    ht_bin->vaddr[1] = -1;
+    ht_bin->tcaddr[1] = NULL;
   }
-  if(ht_bin[0]==vaddr) {
-    ht_bin[0]=ht_bin[2];
-    ht_bin[1]=ht_bin[3];
-    ht_bin[2]=ht_bin[3]=-1;
+  if (ht_bin->vaddr[0] == vaddr) {
+    ht_bin->vaddr[0] = ht_bin->vaddr[1];
+    ht_bin->tcaddr[0] = ht_bin->tcaddr[1];
+    ht_bin->vaddr[1] = -1;
+    ht_bin->tcaddr[1] = NULL;
   }
 }
 
-void ll_remove_matching_addrs(struct ll_entry **head,int addr,int shift)
+void ll_remove_matching_addrs(struct ll_entry **head,uintptr_t addr,int shift)
 {
   struct ll_entry *next;
   while(*head) {
-    if(((u_int)((*head)->addr)>>shift)==(addr>>shift) ||
-       ((u_int)((*head)->addr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift))
+    if(((uintptr_t)((*head)->addr)>>shift)==(addr>>shift) ||
+       ((uintptr_t)((*head)->addr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift))
     {
-      inv_debug("EXP: Remove pointer to %x (%x)\n",(int)(*head)->addr,(*head)->vaddr);
+      inv_debug("EXP: Remove pointer to %p (%x)\n",(*head)->addr,(*head)->vaddr);
       remove_hash((*head)->vaddr);
       next=(*head)->next;
       free(*head);
@@ -810,7 +1108,7 @@ void ll_clear(struct ll_entry **head)
 {
   struct ll_entry *cur;
   struct ll_entry *next;
-  if(cur=*head) {
+  if((cur=*head)) {
     *head=0;
     while(cur) {
       next=cur->next;
@@ -821,26 +1119,25 @@ void ll_clear(struct ll_entry **head)
 }
 
 // Dereference the pointers and remove if it matches
-void ll_kill_pointers(struct ll_entry *head,int addr,int shift)
+static void ll_kill_pointers(struct ll_entry *head,uintptr_t addr,int shift)
 {
   while(head) {
-    int ptr=get_pointer(head->addr);
-    inv_debug("EXP: Lookup pointer to %x at %x (%x)\n",(int)ptr,(int)head->addr,head->vaddr);
+    uintptr_t ptr = (uintptr_t)get_pointer(head->addr);
+    inv_debug("EXP: Lookup pointer to %lx at %p (%x)\n",(long)ptr,head->addr,head->vaddr);
     if(((ptr>>shift)==(addr>>shift)) ||
        (((ptr-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(addr>>shift)))
     {
-      inv_debug("EXP: Kill pointer at %x (%x)\n",(int)head->addr,head->vaddr);
-      u_int host_addr=(u_int)kill_pointer(head->addr);
-      #ifdef __arm__
-        needs_clear_cache[(host_addr-(u_int)BASE_ADDR)>>17]|=1<<(((host_addr-(u_int)BASE_ADDR)>>12)&31);
-      #endif
+      inv_debug("EXP: Kill pointer at %p (%x)\n",head->addr,head->vaddr);
+      void *host_addr=find_extjump_insn(head->addr);
+      mark_clear_cache(host_addr);
+      set_jump_target(host_addr, head->addr);
     }
     head=head->next;
   }
 }
 
 // This is called when we write to a compiled block (see do_invstub)
-void invalidate_page(u_int page)
+static void invalidate_page(u_int page)
 {
   struct ll_entry *head;
   struct ll_entry *next;
@@ -856,11 +1153,10 @@ void invalidate_page(u_int page)
   head=jump_out[page];
   jump_out[page]=0;
   while(head!=NULL) {
-    inv_debug("INVALIDATE: kill pointer to %x (%x)\n",head->vaddr,(int)head->addr);
-    u_int host_addr=(u_int)kill_pointer(head->addr);
-    #ifdef __arm__
-      needs_clear_cache[(host_addr-(u_int)BASE_ADDR)>>17]|=1<<(((host_addr-(u_int)BASE_ADDR)>>12)&31);
-    #endif
+    inv_debug("INVALIDATE: kill pointer to %x (%p)\n",head->vaddr,head->addr);
+    void *host_addr=find_extjump_insn(head->addr);
+    mark_clear_cache(host_addr);
+    set_jump_target(host_addr, head->addr);
     next=head->next;
     free(head);
     head=next;
@@ -882,9 +1178,7 @@ static void invalidate_block_range(u_int block, u_int first, u_int last)
   for(first=page+1;first<last;first++) {
     invalidate_page(first);
   }
-  #ifdef __arm__
-    do_clear_cache();
-  #endif
+  do_clear_cache();
 
   // Don't trap writes
   invalid_code[block]=1;
@@ -906,14 +1200,14 @@ void invalidate_block(u_int block)
   head=jump_dirty[vpage];
   //printf("page=%d vpage=%d\n",page,vpage);
   while(head!=NULL) {
-    u_int start,end;
     if(vpage>2047||(head->vaddr>>12)==block) { // Ignore vaddr hash collision
-      get_bounds((int)head->addr,&start,&end);
-      //printf("start: %x end: %x\n",start,end);
-      if(page<2048&&start>=(u_int)rdram&&end<(u_int)rdram+RAM_SIZE) {
-        if(((start-(u_int)rdram)>>12)<=page&&((end-1-(u_int)rdram)>>12)>=page) {
-          if((((start-(u_int)rdram)>>12)&2047)<first) first=((start-(u_int)rdram)>>12)&2047;
-          if((((end-1-(u_int)rdram)>>12)&2047)>last) last=((end-1-(u_int)rdram)>>12)&2047;
+      u_char *start, *end;
+      get_bounds(head->addr, &start, &end);
+      //printf("start: %p end: %p\n", start, end);
+      if (page < 2048 && start >= rdram && end < rdram+RAM_SIZE) {
+        if (((start-rdram)>>12) <= page && ((end-1-rdram)>>12) >= page) {
+          if ((((start-rdram)>>12)&2047) < first) first = ((start-rdram)>>12)&2047;
+          if ((((end-1-rdram)>>12)&2047) > last)  last = ((end-1-rdram)>>12)&2047;
         }
       }
     }
@@ -944,12 +1238,11 @@ void invalidate_addr(u_int addr)
     }
     for(;pg1<=page;pg1++) {
       for(head=jump_dirty[pg1];head!=NULL;head=head->next) {
-        u_int start,end;
-        get_bounds((int)head->addr,&start,&end);
-        if(ram_offset) {
-          start-=ram_offset;
-          end-=ram_offset;
-        }
+        u_char *start_h, *end_h;
+        u_int start, end;
+        get_bounds(head->addr, &start_h, &end_h);
+        start = (uintptr_t)start_h - ram_offset;
+        end = (uintptr_t)end_h - ram_offset;
         if(start<=addr_main&&addr_main<end) {
           if(start<addr_min) addr_min=start;
           if(end>addr_max) addr_max=end;
@@ -982,9 +1275,9 @@ void invalidate_addr(u_int addr)
 
 // This is called when loading a save state.
 // Anything could have changed, so invalidate everything.
-void invalidate_all_pages()
+void invalidate_all_pages(void)
 {
-  u_int page,n;
+  u_int page;
   for(page=0;page<4096;page++)
     invalidate_page(page);
   for(page=0;page<1048576;page++)
@@ -992,24 +1285,34 @@ void invalidate_all_pages()
       restore_candidate[(page&2047)>>3]|=1<<(page&7);
       restore_candidate[((page&2047)>>3)+256]|=1<<(page&7);
     }
-  #ifdef __arm__
-  __clear_cache((void *)BASE_ADDR,(void *)BASE_ADDR+(1<<TARGET_SIZE_2));
-  #endif
   #ifdef USE_MINI_HT
   memset(mini_ht,-1,sizeof(mini_ht));
   #endif
+  do_clear_cache();
+}
+
+static void do_invstub(int n)
+{
+  literal_pool(20);
+  u_int reglist=stubs[n].a;
+  set_jump_target(stubs[n].addr, out);
+  save_regs(reglist);
+  if(stubs[n].b!=0) emit_mov(stubs[n].b,0);
+  emit_far_call(invalidate_addr);
+  restore_regs(reglist);
+  emit_jmp(stubs[n].retaddr); // return address
 }
 
 // Add an entry to jump_out after making a link
+// src should point to code by emit_extjump2()
 void add_link(u_int vaddr,void *src)
 {
   u_int page=get_page(vaddr);
-  inv_debug("add_link: %x -> %x (%d)\n",(int)src,vaddr,page);
-  int *ptr=(int *)(src+4);
-  assert((*ptr&0x0fff0000)==0x059f0000);
+  inv_debug("add_link: %p -> %x (%d)\n",src,vaddr,page);
+  check_extjump2(src);
   ll_add(jump_out+page,vaddr,src);
-  //int ptr=get_pointer(src);
-  //inv_debug("add_link: Pointer is to %x\n",(int)ptr);
+  //void *ptr=get_pointer(src);
+  //inv_debug("add_link: Pointer is to %p\n",ptr);
 }
 
 // If a code block was found to be unmodified (bit was set in
@@ -1024,15 +1327,15 @@ void clean_blocks(u_int page)
   while(head!=NULL) {
     if(!invalid_code[head->vaddr>>12]) {
       // Don't restore blocks which are about to expire from the cache
-      if((((u_int)head->addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) {
-        u_int start,end;
-        if(verify_dirty((int)head->addr)) {
-          //printf("Possibly Restore %x (%x)\n",head->vaddr, (int)head->addr);
+      if (doesnt_expire_soon(head->addr)) {
+        if(verify_dirty(head->addr)) {
+          u_char *start, *end;
+          //printf("Possibly Restore %x (%p)\n",head->vaddr, head->addr);
           u_int i;
           u_int inv=0;
-          get_bounds((int)head->addr,&start,&end);
-          if(start-(u_int)rdram<RAM_SIZE) {
-            for(i=(start-(u_int)rdram+0x80000000)>>12;i<=(end-1-(u_int)rdram+0x80000000)>>12;i++) {
+          get_bounds(head->addr, &start, &end);
+          if (start - rdram < RAM_SIZE) {
+            for (i = (start-rdram+0x80000000)>>12; i <= (end-1-rdram+0x80000000)>>12; i++) {
               inv|=invalid_code[i];
             }
           }
@@ -1040,20 +1343,18 @@ void clean_blocks(u_int page)
             inv=1;
           }
           if(!inv) {
-            void * clean_addr=(void *)get_clean_addr((int)head->addr);
-            if((((u_int)clean_addr-(u_int)out)<<(32-TARGET_SIZE_2))>0x60000000+(MAX_OUTPUT_BLOCK_SIZE<<(32-TARGET_SIZE_2))) {
+            void *clean_addr = get_clean_addr(head->addr);
+            if (doesnt_expire_soon(clean_addr)) {
               u_int ppage=page;
-              inv_debug("INV: Restored %x (%x/%x)\n",head->vaddr, (int)head->addr, (int)clean_addr);
+              inv_debug("INV: Restored %x (%p/%p)\n",head->vaddr, head->addr, clean_addr);
               //printf("page=%x, addr=%x\n",page,head->vaddr);
               //assert(head->vaddr>>12==(page|0x80000));
               ll_add_flags(jump_in+ppage,head->vaddr,head->reg_sv_flags,clean_addr);
-              int *ht_bin=hash_table[((head->vaddr>>16)^head->vaddr)&0xFFFF];
-              if(ht_bin[0]==head->vaddr) {
-                ht_bin[1]=(int)clean_addr; // Replace existing entry
-              }
-              if(ht_bin[2]==head->vaddr) {
-                ht_bin[3]=(int)clean_addr; // Replace existing entry
-              }
+              struct ht_entry *ht_bin = hash_table_get(head->vaddr);
+              if (ht_bin->vaddr[0] == head->vaddr)
+                ht_bin->tcaddr[0] = clean_addr; // Replace existing entry
+              if (ht_bin->vaddr[1] == head->vaddr)
+                ht_bin->tcaddr[1] = clean_addr; // Replace existing entry
             }
           }
         }
@@ -1063,25 +1364,252 @@ void clean_blocks(u_int page)
   }
 }
 
+/* Register allocation */
 
-void mov_alloc(struct regstat *current,int i)
+// Note: registers are allocated clean (unmodified state)
+// if you intend to modify the register, you must call dirty_reg().
+static void alloc_reg(struct regstat *cur,int i,signed char reg)
 {
-  // Note: Don't need to actually alloc the source registers
-  if((~current->is32>>rs1[i])&1) {
-    //alloc_reg64(current,i,rs1[i]);
-    alloc_reg64(current,i,rt1[i]);
-    current->is32&=~(1LL<<rt1[i]);
-  } else {
-    //alloc_reg(current,i,rs1[i]);
-    alloc_reg(current,i,rt1[i]);
-    current->is32|=(1LL<<rt1[i]);
+  int r,hr;
+  int preferred_reg = (reg&7);
+  if(reg==CCREG) preferred_reg=HOST_CCREG;
+  if(reg==PTEMP||reg==FTEMP) preferred_reg=12;
+
+  // Don't allocate unused registers
+  if((cur->u>>reg)&1) return;
+
+  // see if it's already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(cur->regmap[hr]==reg) return;
+  }
+
+  // Keep the same mapping if the register was already allocated in a loop
+  preferred_reg = loop_reg(i,reg,preferred_reg);
+
+  // Try to allocate the preferred register
+  if(cur->regmap[preferred_reg]==-1) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+  r=cur->regmap[preferred_reg];
+  assert(r < 64);
+  if((cur->u>>r)&1) {
+    cur->regmap[preferred_reg]=reg;
+    cur->dirty&=~(1<<preferred_reg);
+    cur->isconst&=~(1<<preferred_reg);
+    return;
+  }
+
+  // Clear any unneeded registers
+  // We try to keep the mapping consistent, if possible, because it
+  // makes branches easier (especially loops).  So we try to allocate
+  // first (see above) before removing old mappings.  If this is not
+  // possible then go ahead and clear out the registers that are no
+  // longer needed.
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      assert(r < 64);
+      if((cur->u>>r)&1) {cur->regmap[hr]=-1;break;}
+    }
+  }
+  // Try to allocate any available register, but prefer
+  // registers that have not been used recently.
+  if(i>0) {
+    for(hr=0;hr<HOST_REGS;hr++) {
+      if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+        if(regs[i-1].regmap[hr]!=rs1[i-1]&&regs[i-1].regmap[hr]!=rs2[i-1]&&regs[i-1].regmap[hr]!=rt1[i-1]&&regs[i-1].regmap[hr]!=rt2[i-1]) {
+          cur->regmap[hr]=reg;
+          cur->dirty&=~(1<<hr);
+          cur->isconst&=~(1<<hr);
+          return;
+        }
+      }
+    }
+  }
+  // Try to allocate any available register
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",cur->regmap[0],cur->regmap[1],cur->regmap[2],cur->regmap[3],cur->regmap[5],cur->regmap[6],cur->regmap[7]);
+  //printf("hsn(%x): %d %d %d %d %d %d %d\n",start+i*4,hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      // Alloc preferred register if available
+      if(hsn[r=cur->regmap[preferred_reg]&63]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          // Evict both parts of a 64-bit register
+          if((cur->regmap[hr]&63)==r) {
+            cur->regmap[hr]=-1;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+          }
+        }
+        cur->regmap[preferred_reg]=reg;
+        return;
+      }
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||j<hsn[CCREG]) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
+  }
+  SysPrintf("This shouldn't happen (alloc_reg)");abort();
+}
+
+// Allocate a temporary register.  This is done without regard to
+// dirty status or whether the register we request is on the unneeded list
+// Note: This will only allocate one register, even if called multiple times
+static void alloc_reg_temp(struct regstat *cur,int i,signed char reg)
+{
+  int r,hr;
+  int preferred_reg = -1;
+
+  // see if it's already allocated
+  for(hr=0;hr<HOST_REGS;hr++)
+  {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==reg) return;
+  }
+
+  // Try to allocate any available register
+  for(hr=HOST_REGS-1;hr>=0;hr--) {
+    if(hr!=EXCLUDE_REG&&cur->regmap[hr]==-1) {
+      cur->regmap[hr]=reg;
+      cur->dirty&=~(1<<hr);
+      cur->isconst&=~(1<<hr);
+      return;
+    }
+  }
+
+  // Find an unneeded register
+  for(hr=HOST_REGS-1;hr>=0;hr--)
+  {
+    r=cur->regmap[hr];
+    if(r>=0) {
+      assert(r < 64);
+      if((cur->u>>r)&1) {
+        if(i==0||((unneeded_reg[i-1]>>r)&1)) {
+          cur->regmap[hr]=reg;
+          cur->dirty&=~(1<<hr);
+          cur->isconst&=~(1<<hr);
+          return;
+        }
+      }
+    }
+  }
+
+  // Ok, now we have to evict someone
+  // Pick a register we hopefully won't need soon
+  // TODO: we might want to follow unconditional jumps here
+  // TODO: get rid of dupe code and make this into a function
+  u_char hsn[MAXREG+1];
+  memset(hsn,10,sizeof(hsn));
+  int j;
+  lsn(hsn,i,&preferred_reg);
+  //printf("hsn: %d %d %d %d %d %d %d\n",hsn[cur->regmap[0]&63],hsn[cur->regmap[1]&63],hsn[cur->regmap[2]&63],hsn[cur->regmap[3]&63],hsn[cur->regmap[5]&63],hsn[cur->regmap[6]&63],hsn[cur->regmap[7]&63]);
+  if(i>0) {
+    // Don't evict the cycle count at entry points, otherwise the entry
+    // stub will have to write it.
+    if(bt[i]&&hsn[CCREG]>2) hsn[CCREG]=2;
+    if(i>1&&hsn[CCREG]>2&&(itype[i-2]==RJUMP||itype[i-2]==UJUMP||itype[i-2]==CJUMP||itype[i-2]==SJUMP)) hsn[CCREG]=2;
+    for(j=10;j>=3;j--)
+    {
+      for(r=1;r<=MAXREG;r++)
+      {
+        if(hsn[r]==j&&r!=rs1[i-1]&&r!=rs2[i-1]&&r!=rt1[i-1]&&r!=rt2[i-1]) {
+          for(hr=0;hr<HOST_REGS;hr++) {
+            if(hr!=HOST_CCREG||hsn[CCREG]>2) {
+              if(cur->regmap[hr]==r) {
+                cur->regmap[hr]=reg;
+                cur->dirty&=~(1<<hr);
+                cur->isconst&=~(1<<hr);
+                return;
+              }
+            }
+          }
+        }
+      }
+    }
+  }
+  for(j=10;j>=0;j--)
+  {
+    for(r=1;r<=MAXREG;r++)
+    {
+      if(hsn[r]==j) {
+        for(hr=0;hr<HOST_REGS;hr++) {
+          if(cur->regmap[hr]==r) {
+            cur->regmap[hr]=reg;
+            cur->dirty&=~(1<<hr);
+            cur->isconst&=~(1<<hr);
+            return;
+          }
+        }
+      }
+    }
   }
+  SysPrintf("This shouldn't happen");abort();
+}
+
+static void mov_alloc(struct regstat *current,int i)
+{
+  // Note: Don't need to actually alloc the source registers
+  //alloc_reg(current,i,rs1[i]);
+  alloc_reg(current,i,rt1[i]);
+
   clear_const(current,rs1[i]);
   clear_const(current,rt1[i]);
   dirty_reg(current,rt1[i]);
 }
 
-void shiftimm_alloc(struct regstat *current,int i)
+static void shiftimm_alloc(struct regstat *current,int i)
 {
   if(opcode2[i]<=0x3) // SLL/SRL/SRA
   {
@@ -1089,7 +1617,6 @@ void shiftimm_alloc(struct regstat *current,int i)
       if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
       else lt1[i]=rs1[i];
       alloc_reg(current,i,rt1[i]);
-      current->is32|=1LL<<rt1[i];
       dirty_reg(current,rt1[i]);
       if(is_const(current,rs1[i])) {
         int v=get_const(current,rs1[i]);
@@ -1108,48 +1635,23 @@ void shiftimm_alloc(struct regstat *current,int i)
 
   if(opcode2[i]>=0x38&&opcode2[i]<=0x3b) // DSLL/DSRL/DSRA
   {
-    if(rt1[i]) {
-      if(rs1[i]) alloc_reg64(current,i,rs1[i]);
-      alloc_reg64(current,i,rt1[i]);
-      current->is32&=~(1LL<<rt1[i]);
-      dirty_reg(current,rt1[i]);
-    }
+    assert(0);
   }
   if(opcode2[i]==0x3c) // DSLL32
   {
-    if(rt1[i]) {
-      if(rs1[i]) alloc_reg(current,i,rs1[i]);
-      alloc_reg64(current,i,rt1[i]);
-      current->is32&=~(1LL<<rt1[i]);
-      dirty_reg(current,rt1[i]);
-    }
+    assert(0);
   }
   if(opcode2[i]==0x3e) // DSRL32
   {
-    if(rt1[i]) {
-      alloc_reg64(current,i,rs1[i]);
-      if(imm[i]==32) {
-        alloc_reg64(current,i,rt1[i]);
-        current->is32&=~(1LL<<rt1[i]);
-      } else {
-        alloc_reg(current,i,rt1[i]);
-        current->is32|=1LL<<rt1[i];
-      }
-      dirty_reg(current,rt1[i]);
-    }
+    assert(0);
   }
   if(opcode2[i]==0x3f) // DSRA32
   {
-    if(rt1[i]) {
-      alloc_reg64(current,i,rs1[i]);
-      alloc_reg(current,i,rt1[i]);
-      current->is32|=1LL<<rt1[i];
-      dirty_reg(current,rt1[i]);
-    }
+    assert(0);
   }
 }
 
-void shift_alloc(struct regstat *current,int i)
+static void shift_alloc(struct regstat *current,int i)
 {
   if(rt1[i]) {
     if(opcode2[i]<=0x07) // SLLV/SRLV/SRAV
@@ -1161,17 +1663,8 @@ void shift_alloc(struct regstat *current,int i)
         alloc_reg_temp(current,i,-1);
         minimum_free_regs[i]=1;
       }
-      current->is32|=1LL<<rt1[i];
     } else { // DSLLV/DSRLV/DSRAV
-      if(rs1[i]) alloc_reg64(current,i,rs1[i]);
-      if(rs2[i]) alloc_reg(current,i,rs2[i]);
-      alloc_reg64(current,i,rt1[i]);
-      current->is32&=~(1LL<<rt1[i]);
-      if(opcode2[i]==0x16||opcode2[i]==0x17) // DSRLV and DSRAV need a temporary register
-      {
-        alloc_reg_temp(current,i,-1);
-        minimum_free_regs[i]=1;
-      }
+      assert(0);
     }
     clear_const(current,rs1[i]);
     clear_const(current,rs2[i]);
@@ -1180,7 +1673,7 @@ void shift_alloc(struct regstat *current,int i)
   }
 }
 
-void alu_alloc(struct regstat *current,int i)
+static void alu_alloc(struct regstat *current,int i)
 {
   if(opcode2[i]>=0x20&&opcode2[i]<=0x23) { // ADD/ADDU/SUB/SUBU
     if(rt1[i]) {
@@ -1194,22 +1687,13 @@ void alu_alloc(struct regstat *current,int i)
       }
       alloc_reg(current,i,rt1[i]);
     }
-    current->is32|=1LL<<rt1[i];
   }
   if(opcode2[i]==0x2a||opcode2[i]==0x2b) { // SLT/SLTU
     if(rt1[i]) {
-      if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1))
-      {
-        alloc_reg64(current,i,rs1[i]);
-        alloc_reg64(current,i,rs2[i]);
-        alloc_reg(current,i,rt1[i]);
-      } else {
-        alloc_reg(current,i,rs1[i]);
-        alloc_reg(current,i,rs2[i]);
-        alloc_reg(current,i,rt1[i]);
-      }
+      alloc_reg(current,i,rs1[i]);
+      alloc_reg(current,i,rs2[i]);
+      alloc_reg(current,i,rt1[i]);
     }
-    current->is32|=1LL<<rt1[i];
   }
   if(opcode2[i]>=0x24&&opcode2[i]<=0x27) { // AND/OR/XOR/NOR
     if(rt1[i]) {
@@ -1223,78 +1707,10 @@ void alu_alloc(struct regstat *current,int i)
         if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg(current,i,rs2[i]);
       }
       alloc_reg(current,i,rt1[i]);
-      if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1))
-      {
-        if(!((current->uu>>rt1[i])&1)) {
-          alloc_reg64(current,i,rt1[i]);
-        }
-        if(get_reg(current->regmap,rt1[i]|64)>=0) {
-          if(rs1[i]&&rs2[i]) {
-            alloc_reg64(current,i,rs1[i]);
-            alloc_reg64(current,i,rs2[i]);
-          }
-          else
-          {
-            // Is is really worth it to keep 64-bit values in registers?
-            #ifdef NATIVE_64BIT
-            if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg64(current,i,rs1[i]);
-            if(rs2[i]&&needed_again(rs2[i],i)) alloc_reg64(current,i,rs2[i]);
-            #endif
-          }
-        }
-        current->is32&=~(1LL<<rt1[i]);
-      } else {
-        current->is32|=1LL<<rt1[i];
-      }
     }
   }
   if(opcode2[i]>=0x2c&&opcode2[i]<=0x2f) { // DADD/DADDU/DSUB/DSUBU
-    if(rt1[i]) {
-      if(rs1[i]&&rs2[i]) {
-        if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) {
-          alloc_reg64(current,i,rs1[i]);
-          alloc_reg64(current,i,rs2[i]);
-          alloc_reg64(current,i,rt1[i]);
-        } else {
-          alloc_reg(current,i,rs1[i]);
-          alloc_reg(current,i,rs2[i]);
-          alloc_reg(current,i,rt1[i]);
-        }
-      }
-      else {
-        alloc_reg(current,i,rt1[i]);
-        if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) {
-          // DADD used as move, or zeroing
-          // If we have a 64-bit source, then make the target 64 bits too
-          if(rs1[i]&&!((current->is32>>rs1[i])&1)) {
-            if(get_reg(current->regmap,rs1[i])>=0) alloc_reg64(current,i,rs1[i]);
-            alloc_reg64(current,i,rt1[i]);
-          } else if(rs2[i]&&!((current->is32>>rs2[i])&1)) {
-            if(get_reg(current->regmap,rs2[i])>=0) alloc_reg64(current,i,rs2[i]);
-            alloc_reg64(current,i,rt1[i]);
-          }
-          if(opcode2[i]>=0x2e&&rs2[i]) {
-            // DSUB used as negation - 64-bit result
-            // If we have a 32-bit register, extend it to 64 bits
-            if(get_reg(current->regmap,rs2[i])>=0) alloc_reg64(current,i,rs2[i]);
-            alloc_reg64(current,i,rt1[i]);
-          }
-        }
-      }
-      if(rs1[i]&&rs2[i]) {
-        current->is32&=~(1LL<<rt1[i]);
-      } else if(rs1[i]) {
-        current->is32&=~(1LL<<rt1[i]);
-        if((current->is32>>rs1[i])&1)
-          current->is32|=1LL<<rt1[i];
-      } else if(rs2[i]) {
-        current->is32&=~(1LL<<rt1[i]);
-        if((current->is32>>rs2[i])&1)
-          current->is32|=1LL<<rt1[i];
-      } else {
-        current->is32|=1LL<<rt1[i];
-      }
-    }
+    assert(0);
   }
   clear_const(current,rs1[i]);
   clear_const(current,rs2[i]);
@@ -1302,36 +1718,19 @@ void alu_alloc(struct regstat *current,int i)
   dirty_reg(current,rt1[i]);
 }
 
-void imm16_alloc(struct regstat *current,int i)
+static void imm16_alloc(struct regstat *current,int i)
 {
   if(rs1[i]&&needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
   else lt1[i]=rs1[i];
   if(rt1[i]) alloc_reg(current,i,rt1[i]);
   if(opcode[i]==0x18||opcode[i]==0x19) { // DADDI/DADDIU
-    current->is32&=~(1LL<<rt1[i]);
-    if(!((current->uu>>rt1[i])&1)||get_reg(current->regmap,rt1[i]|64)>=0) {
-      // TODO: Could preserve the 32-bit flag if the immediate is zero
-      alloc_reg64(current,i,rt1[i]);
-      alloc_reg64(current,i,rs1[i]);
-    }
-    clear_const(current,rs1[i]);
-    clear_const(current,rt1[i]);
+    assert(0);
   }
   else if(opcode[i]==0x0a||opcode[i]==0x0b) { // SLTI/SLTIU
-    if((~current->is32>>rs1[i])&1) alloc_reg64(current,i,rs1[i]);
-    current->is32|=1LL<<rt1[i];
     clear_const(current,rs1[i]);
     clear_const(current,rt1[i]);
   }
   else if(opcode[i]>=0x0c&&opcode[i]<=0x0e) { // ANDI/ORI/XORI
-    if(((~current->is32>>rs1[i])&1)&&opcode[i]>0x0c) {
-      if(rs1[i]!=rt1[i]) {
-        if(needed_again(rs1[i],i)) alloc_reg64(current,i,rs1[i]);
-        alloc_reg64(current,i,rt1[i]);
-        current->is32&=~(1LL<<rt1[i]);
-      }
-    }
-    else current->is32|=1LL<<rt1[i]; // ANDI clears upper bits
     if(is_const(current,rs1[i])) {
       int v=get_const(current,rs1[i]);
       if(opcode[i]==0x0c) set_const(current,rt1[i],v&imm[i]);
@@ -1346,16 +1745,14 @@ void imm16_alloc(struct regstat *current,int i)
       set_const(current,rt1[i],v+imm[i]);
     }
     else clear_const(current,rt1[i]);
-    current->is32|=1LL<<rt1[i];
   }
   else {
-    set_const(current,rt1[i],((long long)((short)imm[i]))<<16); // LUI
-    current->is32|=1LL<<rt1[i];
+    set_const(current,rt1[i],imm[i]<<16); // LUI
   }
   dirty_reg(current,rt1[i]);
 }
 
-void load_alloc(struct regstat *current,int i)
+static void load_alloc(struct regstat *current,int i)
 {
   clear_const(current,rt1[i]);
   //if(rs1[i]!=rt1[i]&&needed_again(rs1[i],i)) clear_const(current,rs1[i]); // Does this help or hurt?
@@ -1366,18 +1763,12 @@ void load_alloc(struct regstat *current,int i)
     assert(get_reg(current->regmap,rt1[i])>=0);
     if(opcode[i]==0x27||opcode[i]==0x37) // LWU/LD
     {
-      current->is32&=~(1LL<<rt1[i]);
-      alloc_reg64(current,i,rt1[i]);
+      assert(0);
     }
     else if(opcode[i]==0x1A||opcode[i]==0x1B) // LDL/LDR
     {
-      current->is32&=~(1LL<<rt1[i]);
-      alloc_reg64(current,i,rt1[i]);
-      alloc_all(current,i);
-      alloc_reg64(current,i,FTEMP);
-      minimum_free_regs[i]=HOST_REGS;
+      assert(0);
     }
-    else current->is32|=1LL<<rt1[i];
     dirty_reg(current,rt1[i]);
     // LWL/LWR need a temporary register for the old value
     if(opcode[i]==0x22||opcode[i]==0x26)
@@ -1399,9 +1790,7 @@ void load_alloc(struct regstat *current,int i)
     minimum_free_regs[i]=1;
     if(opcode[i]==0x1A||opcode[i]==0x1B) // LDL/LDR
     {
-      alloc_all(current,i);
-      alloc_reg64(current,i,FTEMP);
-      minimum_free_regs[i]=HOST_REGS;
+      assert(0);
     }
   }
 }
@@ -1413,8 +1802,7 @@ void store_alloc(struct regstat *current,int i)
   if(needed_again(rs1[i],i)) alloc_reg(current,i,rs1[i]);
   alloc_reg(current,i,rs2[i]);
   if(opcode[i]==0x2c||opcode[i]==0x2d||opcode[i]==0x3f) { // 64-bit SDL/SDR/SD
-    alloc_reg64(current,i,rs2[i]);
-    if(rs2[i]) alloc_reg(current,i,FTEMP);
+    assert(0);
   }
   #if defined(HOST_IMM8)
   // On CPUs without 32-bit immediates we need a pointer to invalid_code
@@ -1436,7 +1824,7 @@ void c1ls_alloc(struct regstat *current,int i)
   alloc_reg(current,i,CSREG); // Status
   alloc_reg(current,i,FTEMP);
   if(opcode[i]==0x35||opcode[i]==0x3d) { // 64-bit LDC1/SDC1
-    alloc_reg64(current,i,FTEMP);
+    assert(0);
   }
   #if defined(HOST_IMM8)
   // On CPUs without 32-bit immediates we need a pointer to invalid_code
@@ -1485,27 +1873,12 @@ void multdiv_alloc(struct regstat *current,int i)
       alloc_reg(current,i,LOREG);
       alloc_reg(current,i,rs1[i]);
       alloc_reg(current,i,rs2[i]);
-      current->is32|=1LL<<HIREG;
-      current->is32|=1LL<<LOREG;
       dirty_reg(current,HIREG);
       dirty_reg(current,LOREG);
     }
     else // 64-bit
     {
-      current->u&=~(1LL<<HIREG);
-      current->u&=~(1LL<<LOREG);
-      current->uu&=~(1LL<<HIREG);
-      current->uu&=~(1LL<<LOREG);
-      alloc_reg64(current,i,HIREG);
-      //if(HOST_REGS>10) alloc_reg64(current,i,LOREG);
-      alloc_reg64(current,i,rs1[i]);
-      alloc_reg64(current,i,rs2[i]);
-      alloc_all(current,i);
-      current->is32&=~(1LL<<HIREG);
-      current->is32&=~(1LL<<LOREG);
-      dirty_reg(current,HIREG);
-      dirty_reg(current,LOREG);
-      minimum_free_regs[i]=HOST_REGS;
+      assert(0);
     }
   }
   else
@@ -1515,8 +1888,6 @@ void multdiv_alloc(struct regstat *current,int i)
     // The result is undefined, we return zero.
     alloc_reg(current,i,HIREG);
     alloc_reg(current,i,LOREG);
-    current->is32|=1LL<<HIREG;
-    current->is32|=1LL<<LOREG;
     dirty_reg(current,HIREG);
     dirty_reg(current,LOREG);
   }
@@ -1531,7 +1902,6 @@ void cop0_alloc(struct regstat *current,int i)
       clear_const(current,rt1[i]);
       alloc_all(current,i);
       alloc_reg(current,i,rt1[i]);
-      current->is32|=1LL<<rt1[i];
       dirty_reg(current,rt1[i]);
     }
   }
@@ -1557,66 +1927,39 @@ void cop0_alloc(struct regstat *current,int i)
   minimum_free_regs[i]=HOST_REGS;
 }
 
-void cop1_alloc(struct regstat *current,int i)
+static void cop2_alloc(struct regstat *current,int i)
 {
-  alloc_reg(current,i,CSREG); // Load status
-  if(opcode2[i]<3) // MFC1/DMFC1/CFC1
+  if (opcode2[i] < 3) // MFC2/CFC2
   {
+    alloc_cc(current,i); // for stalls
+    dirty_reg(current,CCREG);
     if(rt1[i]){
       clear_const(current,rt1[i]);
-      if(opcode2[i]==1) {
-        alloc_reg64(current,i,rt1[i]); // DMFC1
-        current->is32&=~(1LL<<rt1[i]);
-      }else{
-        alloc_reg(current,i,rt1[i]); // MFC1/CFC1
-        current->is32|=1LL<<rt1[i];
-      }
+      alloc_reg(current,i,rt1[i]);
       dirty_reg(current,rt1[i]);
     }
-    alloc_reg_temp(current,i,-1);
   }
-  else if(opcode2[i]>3) // MTC1/DMTC1/CTC1
+  else if (opcode2[i] > 3) // MTC2/CTC2
   {
     if(rs1[i]){
       clear_const(current,rs1[i]);
-      if(opcode2[i]==5)
-        alloc_reg64(current,i,rs1[i]); // DMTC1
-      else
-        alloc_reg(current,i,rs1[i]); // MTC1/CTC1
-      alloc_reg_temp(current,i,-1);
+      alloc_reg(current,i,rs1[i]);
     }
     else {
       current->u&=~1LL;
       alloc_reg(current,i,0);
-      alloc_reg_temp(current,i,-1);
     }
   }
-  minimum_free_regs[i]=1;
-}
-void fconv_alloc(struct regstat *current,int i)
-{
-  alloc_reg(current,i,CSREG); // Load status
-  alloc_reg_temp(current,i,-1);
-  minimum_free_regs[i]=1;
-}
-void float_alloc(struct regstat *current,int i)
-{
-  alloc_reg(current,i,CSREG); // Load status
   alloc_reg_temp(current,i,-1);
   minimum_free_regs[i]=1;
 }
+
 void c2op_alloc(struct regstat *current,int i)
 {
+  alloc_cc(current,i); // for stalls
+  dirty_reg(current,CCREG);
   alloc_reg_temp(current,i,-1);
 }
-void fcomp_alloc(struct regstat *current,int i)
-{
-  alloc_reg(current,i,CSREG); // Load status
-  alloc_reg(current,i,FSREG); // Load flags
-  dirty_reg(current,FSREG); // Flag will be modified
-  alloc_reg_temp(current,i,-1);
-  minimum_free_regs[i]=1;
-}
 
 void syscall_alloc(struct regstat *current,int i)
 {
@@ -1634,11 +1977,10 @@ void delayslot_alloc(struct regstat *current,int i)
     case CJUMP:
     case SJUMP:
     case RJUMP:
-    case FJUMP:
     case SYSCALL:
     case HLECALL:
     case SPAN:
-      assem_debug("jump in the delay slot.  this shouldn't happen.\n");//exit(1);
+      assem_debug("jump in the delay slot.  this shouldn't happen.\n");//abort();
       SysPrintf("Disabled speculative precompilation\n");
       stop_after_jal=1;
       break;
@@ -1672,8 +2014,9 @@ void delayslot_alloc(struct regstat *current,int i)
       cop0_alloc(current,i);
       break;
     case COP1:
+      break;
     case COP2:
-      cop1_alloc(current,i);
+      cop2_alloc(current,i);
       break;
     case C1LS:
       c1ls_alloc(current,i);
@@ -1681,15 +2024,6 @@ void delayslot_alloc(struct regstat *current,int i)
     case C2LS:
       c2ls_alloc(current,i);
       break;
-    case FCONV:
-      fconv_alloc(current,i);
-      break;
-    case FLOAT:
-      float_alloc(current,i);
-      break;
-    case FCOMP:
-      fcomp_alloc(current,i);
-      break;
     case C2OP:
       c2op_alloc(current,i);
       break;
@@ -1723,124 +2057,93 @@ static void pagespan_alloc(struct regstat *current,int i)
   {
     if(rs1[i]) alloc_reg(current,i,rs1[i]);
     if(rs2[i]) alloc_reg(current,i,rs2[i]);
-    if(!((current->is32>>rs1[i])&(current->is32>>rs2[i])&1))
-    {
-      if(rs1[i]) alloc_reg64(current,i,rs1[i]);
-      if(rs2[i]) alloc_reg64(current,i,rs2[i]);
-    }
   }
   else
   if((opcode[i]&0x2E)==6) // BLEZ/BGTZ/BLEZL/BGTZL
   {
     if(rs1[i]) alloc_reg(current,i,rs1[i]);
-    if(!((current->is32>>rs1[i])&1))
-    {
-      if(rs1[i]) alloc_reg64(current,i,rs1[i]);
-    }
-  }
-  else
-  if(opcode[i]==0x11) // BC1
-  {
-    alloc_reg(current,i,FSREG);
-    alloc_reg(current,i,CSREG);
   }
   //else ...
 }
 
-add_stub(int type,int addr,int retaddr,int a,int b,int c,int d,int e)
+static void add_stub(enum stub_type type, void *addr, void *retaddr,
+  u_int a, uintptr_t b, uintptr_t c, u_int d, u_int e)
 {
-  stubs[stubcount][0]=type;
-  stubs[stubcount][1]=addr;
-  stubs[stubcount][2]=retaddr;
-  stubs[stubcount][3]=a;
-  stubs[stubcount][4]=b;
-  stubs[stubcount][5]=c;
-  stubs[stubcount][6]=d;
-  stubs[stubcount][7]=e;
+  assert(stubcount < ARRAY_SIZE(stubs));
+  stubs[stubcount].type = type;
+  stubs[stubcount].addr = addr;
+  stubs[stubcount].retaddr = retaddr;
+  stubs[stubcount].a = a;
+  stubs[stubcount].b = b;
+  stubs[stubcount].c = c;
+  stubs[stubcount].d = d;
+  stubs[stubcount].e = e;
   stubcount++;
 }
 
+static void add_stub_r(enum stub_type type, void *addr, void *retaddr,
+  int i, int addr_reg, const struct regstat *i_regs, int ccadj, u_int reglist)
+{
+  add_stub(type, addr, retaddr, i, addr_reg, (uintptr_t)i_regs, ccadj, reglist);
+}
+
 // Write out a single register
-void wb_register(signed char r,signed char regmap[],uint64_t dirty,uint64_t is32)
+static void wb_register(signed char r,signed char regmap[],uint64_t dirty)
 {
   int hr;
   for(hr=0;hr<HOST_REGS;hr++) {
     if(hr!=EXCLUDE_REG) {
       if((regmap[hr]&63)==r) {
         if((dirty>>hr)&1) {
-          if(regmap[hr]<64) {
-            emit_storereg(r,hr);
-          }else{
-            emit_storereg(r|64,hr);
-          }
+          assert(regmap[hr]<64);
+          emit_storereg(r,hr);
         }
       }
     }
   }
 }
 
-int mchecksum()
+static void wb_valid(signed char pre[],signed char entry[],u_int dirty_pre,u_int dirty,uint64_t u)
 {
-  //if(!tracedebug) return 0;
-  int i;
-  int sum=0;
-  for(i=0;i<2097152;i++) {
-    unsigned int temp=sum;
-    sum<<=1;
-    sum|=(~temp)>>31;
-    sum^=((u_int *)rdram)[i];
+  //if(dirty_pre==dirty) return;
+  int hr,reg;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(hr!=EXCLUDE_REG) {
+      reg=pre[hr];
+      if(((~u)>>(reg&63))&1) {
+        if(reg>0) {
+          if(((dirty_pre&~dirty)>>hr)&1) {
+            if(reg>0&&reg<34) {
+              emit_storereg(reg,hr);
+            }
+            else if(reg>=64) {
+              assert(0);
+            }
+          }
+        }
+      }
+    }
   }
-  return sum;
-}
-int rchecksum()
-{
-  int i;
-  int sum=0;
-  for(i=0;i<64;i++)
-    sum^=((u_int *)reg)[i];
-  return sum;
-}
-void rlist()
-{
-  int i;
-  printf("TRACE: ");
-  for(i=0;i<32;i++)
-    printf("r%d:%8x%8x ",i,((int *)(reg+i))[1],((int *)(reg+i))[0]);
-  printf("\n");
-}
-
-void enabletrace()
-{
-  tracedebug=1;
 }
 
-void memdebug(int i)
+// trashes r2
+static void pass_args(int a0, int a1)
 {
-  //printf("TRACE: count=%d next=%d (checksum %x) lo=%8x%8x\n",Count,next_interupt,mchecksum(),(int)(reg[LOREG]>>32),(int)reg[LOREG]);
-  //printf("TRACE: count=%d next=%d (rchecksum %x)\n",Count,next_interupt,rchecksum());
-  //rlist();
-  //if(tracedebug) {
-  //if(Count>=-2084597794) {
-  if((signed int)Count>=-2084597794&&(signed int)Count<0) {
-  //if(0) {
-    printf("TRACE: count=%d next=%d (checksum %x)\n",Count,next_interupt,mchecksum());
-    //printf("TRACE: count=%d next=%d (checksum %x) Status=%x\n",Count,next_interupt,mchecksum(),Status);
-    //printf("TRACE: count=%d next=%d (checksum %x) hi=%8x%8x\n",Count,next_interupt,mchecksum(),(int)(reg[HIREG]>>32),(int)reg[HIREG]);
-    rlist();
-    #ifdef __i386__
-    printf("TRACE: %x\n",(&i)[-1]);
-    #endif
-    #ifdef __arm__
-    int j;
-    printf("TRACE: %x \n",(&j)[10]);
-    printf("TRACE: %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x %x\n",(&j)[1],(&j)[2],(&j)[3],(&j)[4],(&j)[5],(&j)[6],(&j)[7],(&j)[8],(&j)[9],(&j)[10],(&j)[11],(&j)[12],(&j)[13],(&j)[14],(&j)[15],(&j)[16],(&j)[17],(&j)[18],(&j)[19],(&j)[20]);
-    #endif
-    //fflush(stdout);
+  if(a0==1&&a1==0) {
+    // must swap
+    emit_mov(a0,2); emit_mov(a1,1); emit_mov(2,0);
+  }
+  else if(a0!=0&&a1==0) {
+    emit_mov(a1,1);
+    if (a0>=0) emit_mov(a0,0);
+  }
+  else {
+    if(a0>=0&&a0!=0) emit_mov(a0,0);
+    if(a1>=0&&a1!=1) emit_mov(a1,1);
   }
-  //printf("TRACE: %x\n",(&i)[-1]);
 }
 
-void alu_assemble(int i,struct regstat *i_regs)
+static void alu_assemble(int i,struct regstat *i_regs)
 {
   if(opcode2[i]>=0x20&&opcode2[i]<=0x23) { // ADD/ADDU/SUB/SUBU
     if(rt1[i]) {
@@ -1874,113 +2177,12 @@ void alu_assemble(int i,struct regstat *i_regs)
     }
   }
   if(opcode2[i]>=0x2c&&opcode2[i]<=0x2f) { // DADD/DADDU/DSUB/DSUBU
-    if(rt1[i]) {
-      signed char s1l,s2l,s1h,s2h,tl,th;
-      tl=get_reg(i_regs->regmap,rt1[i]);
-      th=get_reg(i_regs->regmap,rt1[i]|64);
-      if(tl>=0) {
-        s1l=get_reg(i_regs->regmap,rs1[i]);
-        s2l=get_reg(i_regs->regmap,rs2[i]);
-        s1h=get_reg(i_regs->regmap,rs1[i]|64);
-        s2h=get_reg(i_regs->regmap,rs2[i]|64);
-        if(rs1[i]&&rs2[i]) {
-          assert(s1l>=0);
-          assert(s2l>=0);
-          if(opcode2[i]&2) emit_subs(s1l,s2l,tl);
-          else emit_adds(s1l,s2l,tl);
-          if(th>=0) {
-            #ifdef INVERTED_CARRY
-            if(opcode2[i]&2) {if(s1h!=th) emit_mov(s1h,th);emit_sbb(th,s2h);}
-            #else
-            if(opcode2[i]&2) emit_sbc(s1h,s2h,th);
-            #endif
-            else emit_add(s1h,s2h,th);
-          }
-        }
-        else if(rs1[i]) {
-          if(s1l>=0) emit_mov(s1l,tl);
-          else emit_loadreg(rs1[i],tl);
-          if(th>=0) {
-            if(s1h>=0) emit_mov(s1h,th);
-            else emit_loadreg(rs1[i]|64,th);
-          }
-        }
-        else if(rs2[i]) {
-          if(s2l>=0) {
-            if(opcode2[i]&2) emit_negs(s2l,tl);
-            else emit_mov(s2l,tl);
-          }
-          else {
-            emit_loadreg(rs2[i],tl);
-            if(opcode2[i]&2) emit_negs(tl,tl);
-          }
-          if(th>=0) {
-            #ifdef INVERTED_CARRY
-            if(s2h>=0) emit_mov(s2h,th);
-            else emit_loadreg(rs2[i]|64,th);
-            if(opcode2[i]&2) {
-              emit_adcimm(-1,th); // x86 has inverted carry flag
-              emit_not(th,th);
-            }
-            #else
-            if(opcode2[i]&2) {
-              if(s2h>=0) emit_rscimm(s2h,0,th);
-              else {
-                emit_loadreg(rs2[i]|64,th);
-                emit_rscimm(th,0,th);
-              }
-            }else{
-              if(s2h>=0) emit_mov(s2h,th);
-              else emit_loadreg(rs2[i]|64,th);
-            }
-            #endif
-          }
-        }
-        else {
-          emit_zeroreg(tl);
-          if(th>=0) emit_zeroreg(th);
-        }
-      }
-    }
+    assert(0);
   }
   if(opcode2[i]==0x2a||opcode2[i]==0x2b) { // SLT/SLTU
     if(rt1[i]) {
-      signed char s1l,s1h,s2l,s2h,t;
-      if(!((i_regs->was32>>rs1[i])&(i_regs->was32>>rs2[i])&1))
+      signed char s1l,s2l,t;
       {
-        t=get_reg(i_regs->regmap,rt1[i]);
-        //assert(t>=0);
-        if(t>=0) {
-          s1l=get_reg(i_regs->regmap,rs1[i]);
-          s1h=get_reg(i_regs->regmap,rs1[i]|64);
-          s2l=get_reg(i_regs->regmap,rs2[i]);
-          s2h=get_reg(i_regs->regmap,rs2[i]|64);
-          if(rs2[i]==0) // rx<r0
-          {
-            assert(s1h>=0);
-            if(opcode2[i]==0x2a) // SLT
-              emit_shrimm(s1h,31,t);
-            else // SLTU (unsigned can not be less than zero)
-              emit_zeroreg(t);
-          }
-          else if(rs1[i]==0) // r0<rx
-          {
-            assert(s2h>=0);
-            if(opcode2[i]==0x2a) // SLT
-              emit_set_gz64_32(s2h,s2l,t);
-            else // SLTU (set if not zero)
-              emit_set_nz64_32(s2h,s2l,t);
-          }
-          else {
-            assert(s1l>=0);assert(s1h>=0);
-            assert(s2l>=0);assert(s2h>=0);
-            if(opcode2[i]==0x2a) // SLT
-              emit_set_if_less64_32(s1h,s1l,s2h,s2l,t);
-            else // SLTU
-              emit_set_if_carry64_32(s1h,s1l,s2h,s2l,t);
-          }
-        }
-      } else {
         t=get_reg(i_regs->regmap,rt1[i]);
         //assert(t>=0);
         if(t>=0) {
@@ -1988,10 +2190,11 @@ void alu_assemble(int i,struct regstat *i_regs)
           s2l=get_reg(i_regs->regmap,rs2[i]);
           if(rs2[i]==0) // rx<r0
           {
-            assert(s1l>=0);
-            if(opcode2[i]==0x2a) // SLT
+            if(opcode2[i]==0x2a&&rs1[i]!=0) { // SLT
+              assert(s1l>=0);
               emit_shrimm(s1l,31,t);
-            else // SLTU (unsigned can not be less than zero)
+            }
+            else // SLTU (unsigned can not be less than zero, 0<0)
               emit_zeroreg(t);
           }
           else if(rs1[i]==0) // r0<rx
@@ -2015,101 +2218,9 @@ void alu_assemble(int i,struct regstat *i_regs)
   }
   if(opcode2[i]>=0x24&&opcode2[i]<=0x27) { // AND/OR/XOR/NOR
     if(rt1[i]) {
-      signed char s1l,s1h,s2l,s2h,th,tl;
+      signed char s1l,s2l,tl;
       tl=get_reg(i_regs->regmap,rt1[i]);
-      th=get_reg(i_regs->regmap,rt1[i]|64);
-      if(!((i_regs->was32>>rs1[i])&(i_regs->was32>>rs2[i])&1)&&th>=0)
-      {
-        assert(tl>=0);
-        if(tl>=0) {
-          s1l=get_reg(i_regs->regmap,rs1[i]);
-          s1h=get_reg(i_regs->regmap,rs1[i]|64);
-          s2l=get_reg(i_regs->regmap,rs2[i]);
-          s2h=get_reg(i_regs->regmap,rs2[i]|64);
-          if(rs1[i]&&rs2[i]) {
-            assert(s1l>=0);assert(s1h>=0);
-            assert(s2l>=0);assert(s2h>=0);
-            if(opcode2[i]==0x24) { // AND
-              emit_and(s1l,s2l,tl);
-              emit_and(s1h,s2h,th);
-            } else
-            if(opcode2[i]==0x25) { // OR
-              emit_or(s1l,s2l,tl);
-              emit_or(s1h,s2h,th);
-            } else
-            if(opcode2[i]==0x26) { // XOR
-              emit_xor(s1l,s2l,tl);
-              emit_xor(s1h,s2h,th);
-            } else
-            if(opcode2[i]==0x27) { // NOR
-              emit_or(s1l,s2l,tl);
-              emit_or(s1h,s2h,th);
-              emit_not(tl,tl);
-              emit_not(th,th);
-            }
-          }
-          else
-          {
-            if(opcode2[i]==0x24) { // AND
-              emit_zeroreg(tl);
-              emit_zeroreg(th);
-            } else
-            if(opcode2[i]==0x25||opcode2[i]==0x26) { // OR/XOR
-              if(rs1[i]){
-                if(s1l>=0) emit_mov(s1l,tl);
-                else emit_loadreg(rs1[i],tl);
-                if(s1h>=0) emit_mov(s1h,th);
-                else emit_loadreg(rs1[i]|64,th);
-              }
-              else
-              if(rs2[i]){
-                if(s2l>=0) emit_mov(s2l,tl);
-                else emit_loadreg(rs2[i],tl);
-                if(s2h>=0) emit_mov(s2h,th);
-                else emit_loadreg(rs2[i]|64,th);
-              }
-              else{
-                emit_zeroreg(tl);
-                emit_zeroreg(th);
-              }
-            } else
-            if(opcode2[i]==0x27) { // NOR
-              if(rs1[i]){
-                if(s1l>=0) emit_not(s1l,tl);
-                else{
-                  emit_loadreg(rs1[i],tl);
-                  emit_not(tl,tl);
-                }
-                if(s1h>=0) emit_not(s1h,th);
-                else{
-                  emit_loadreg(rs1[i]|64,th);
-                  emit_not(th,th);
-                }
-              }
-              else
-              if(rs2[i]){
-                if(s2l>=0) emit_not(s2l,tl);
-                else{
-                  emit_loadreg(rs2[i],tl);
-                  emit_not(tl,tl);
-                }
-                if(s2h>=0) emit_not(s2h,th);
-                else{
-                  emit_loadreg(rs2[i]|64,th);
-                  emit_not(th,th);
-                }
-              }
-              else {
-                emit_movimm(-1,tl);
-                emit_movimm(-1,th);
-              }
-            }
-          }
-        }
-      }
-      else
       {
-        // 32 bit
         if(tl>=0) {
           s1l=get_reg(i_regs->regmap,rs1[i]);
           s2l=get_reg(i_regs->regmap,rs2[i]);
@@ -2216,24 +2327,15 @@ void imm16_assemble(int i,struct regstat *i_regs)
   }
   if(opcode[i]==0x18||opcode[i]==0x19) { // DADDI/DADDIU
     if(rt1[i]) {
-      signed char sh,sl,th,tl;
-      th=get_reg(i_regs->regmap,rt1[i]|64);
+      signed char sl,tl;
       tl=get_reg(i_regs->regmap,rt1[i]);
-      sh=get_reg(i_regs->regmap,rs1[i]|64);
       sl=get_reg(i_regs->regmap,rs1[i]);
       if(tl>=0) {
         if(rs1[i]) {
-          assert(sh>=0);
           assert(sl>=0);
-          if(th>=0) {
-            emit_addimm64_32(sh,sl,imm[i],th,tl);
-          }
-          else {
-            emit_addimm(sl,imm[i],tl);
-          }
+          emit_addimm(sl,imm[i],tl);
         } else {
           emit_movimm(imm[i],tl);
-          if(th>=0) emit_movimm(((signed int)imm[i])>>31,th);
         }
       }
     }
@@ -2241,15 +2343,12 @@ void imm16_assemble(int i,struct regstat *i_regs)
   else if(opcode[i]==0x0a||opcode[i]==0x0b) { // SLTI/SLTIU
     if(rt1[i]) {
       //assert(rs1[i]!=0); // r0 might be valid, but it's probably a bug
-      signed char sh,sl,t;
+      signed char sl,t;
       t=get_reg(i_regs->regmap,rt1[i]);
-      sh=get_reg(i_regs->regmap,rs1[i]|64);
       sl=get_reg(i_regs->regmap,rs1[i]);
       //assert(t>=0);
       if(t>=0) {
         if(rs1[i]>0) {
-          if(sh<0) assert((i_regs->was32>>rs1[i])&1);
-          if(sh<0||((i_regs->was32>>rs1[i])&1)) {
             if(opcode[i]==0x0a) { // SLTI
               if(sl<0) {
                 if(i_regs->regmap_entry[t]!=rs1[i]) emit_loadreg(rs1[i],t);
@@ -2266,13 +2365,6 @@ void imm16_assemble(int i,struct regstat *i_regs)
                 emit_sltiu32(sl,imm[i],t);
               }
             }
-          }else{ // 64-bit
-            assert(sl>=0);
-            if(opcode[i]==0x0a) // SLTI
-              emit_slti64_32(sh,sl,imm[i],t);
-            else // SLTIU
-              emit_sltiu64_32(sh,sl,imm[i],t);
-          }
         }else{
           // SLTI(U) with r0 is just stupid,
           // nonetheless examples can be found
@@ -2290,10 +2382,8 @@ void imm16_assemble(int i,struct regstat *i_regs)
   }
   else if(opcode[i]>=0x0c&&opcode[i]<=0x0e) { // ANDI/ORI/XORI
     if(rt1[i]) {
-      signed char sh,sl,th,tl;
-      th=get_reg(i_regs->regmap,rt1[i]|64);
+      signed char sl,tl;
       tl=get_reg(i_regs->regmap,rt1[i]);
-      sh=get_reg(i_regs->regmap,rs1[i]|64);
       sl=get_reg(i_regs->regmap,rs1[i]);
       if(tl>=0 && !((i_regs->isconst>>tl)&1)) {
         if(opcode[i]==0x0c) //ANDI
@@ -2311,7 +2401,6 @@ void imm16_assemble(int i,struct regstat *i_regs)
           }
           else
             emit_zeroreg(tl);
-          if(th>=0) emit_zeroreg(th);
         }
         else
         {
@@ -2319,35 +2408,29 @@ void imm16_assemble(int i,struct regstat *i_regs)
             if(sl<0) {
               if(i_regs->regmap_entry[tl]!=rs1[i]) emit_loadreg(rs1[i],tl);
             }
-            if(th>=0) {
-              if(sh<0) {
-                emit_loadreg(rs1[i]|64,th);
+            if(opcode[i]==0x0d) { // ORI
+              if(sl<0) {
+                emit_orimm(tl,imm[i],tl);
               }else{
-                emit_mov(sh,th);
+                if(!((i_regs->wasconst>>sl)&1))
+                  emit_orimm(sl,imm[i],tl);
+                else
+                  emit_movimm(constmap[i][sl]|imm[i],tl);
               }
             }
-            if(opcode[i]==0x0d) //ORI
-            if(sl<0) {
-              emit_orimm(tl,imm[i],tl);
-            }else{
-              if(!((i_regs->wasconst>>sl)&1))
-                emit_orimm(sl,imm[i],tl);
-              else
-                emit_movimm(constmap[i][sl]|imm[i],tl);
-            }
-            if(opcode[i]==0x0e) //XORI
-            if(sl<0) {
-              emit_xorimm(tl,imm[i],tl);
-            }else{
-              if(!((i_regs->wasconst>>sl)&1))
-                emit_xorimm(sl,imm[i],tl);
-              else
-                emit_movimm(constmap[i][sl]^imm[i],tl);
+            if(opcode[i]==0x0e) { // XORI
+              if(sl<0) {
+                emit_xorimm(tl,imm[i],tl);
+              }else{
+                if(!((i_regs->wasconst>>sl)&1))
+                  emit_xorimm(sl,imm[i],tl);
+                else
+                  emit_movimm(constmap[i][sl]^imm[i],tl);
+              }
             }
           }
           else {
             emit_movimm(imm[i],tl);
-            if(th>=0) emit_zeroreg(th);
           }
         }
       }
@@ -2396,126 +2479,225 @@ void shiftimm_assemble(int i,struct regstat *i_regs)
   }
   if(opcode2[i]>=0x38&&opcode2[i]<=0x3b) // DSLL/DSRL/DSRA
   {
-    if(rt1[i]) {
-      signed char sh,sl,th,tl;
-      th=get_reg(i_regs->regmap,rt1[i]|64);
-      tl=get_reg(i_regs->regmap,rt1[i]);
-      sh=get_reg(i_regs->regmap,rs1[i]|64);
-      sl=get_reg(i_regs->regmap,rs1[i]);
-      if(tl>=0) {
-        if(rs1[i]==0)
-        {
-          emit_zeroreg(tl);
-          if(th>=0) emit_zeroreg(th);
-        }
-        else
-        {
-          assert(sl>=0);
-          assert(sh>=0);
-          if(imm[i]) {
-            if(opcode2[i]==0x38) // DSLL
-            {
-              if(th>=0) emit_shldimm(sh,sl,imm[i],th);
-              emit_shlimm(sl,imm[i],tl);
-            }
-            if(opcode2[i]==0x3a) // DSRL
-            {
-              emit_shrdimm(sl,sh,imm[i],tl);
-              if(th>=0) emit_shrimm(sh,imm[i],th);
-            }
-            if(opcode2[i]==0x3b) // DSRA
-            {
-              emit_shrdimm(sl,sh,imm[i],tl);
-              if(th>=0) emit_sarimm(sh,imm[i],th);
-            }
-          }else{
-            // Shift by zero
-            if(sl!=tl) emit_mov(sl,tl);
-            if(th>=0&&sh!=th) emit_mov(sh,th);
-          }
-        }
-      }
-    }
+    assert(0);
   }
   if(opcode2[i]==0x3c) // DSLL32
   {
-    if(rt1[i]) {
-      signed char sl,tl,th;
-      tl=get_reg(i_regs->regmap,rt1[i]);
-      th=get_reg(i_regs->regmap,rt1[i]|64);
-      sl=get_reg(i_regs->regmap,rs1[i]);
-      if(th>=0||tl>=0){
-        assert(tl>=0);
-        assert(th>=0);
-        assert(sl>=0);
-        emit_mov(sl,th);
-        emit_zeroreg(tl);
-        if(imm[i]>32)
-        {
-          emit_shlimm(th,imm[i]&31,th);
-        }
-      }
-    }
+    assert(0);
   }
   if(opcode2[i]==0x3e) // DSRL32
   {
-    if(rt1[i]) {
-      signed char sh,tl,th;
-      tl=get_reg(i_regs->regmap,rt1[i]);
-      th=get_reg(i_regs->regmap,rt1[i]|64);
-      sh=get_reg(i_regs->regmap,rs1[i]|64);
-      if(tl>=0){
-        assert(sh>=0);
-        emit_mov(sh,tl);
-        if(th>=0) emit_zeroreg(th);
-        if(imm[i]>32)
-        {
-          emit_shrimm(tl,imm[i]&31,tl);
-        }
-      }
-    }
+    assert(0);
   }
   if(opcode2[i]==0x3f) // DSRA32
   {
-    if(rt1[i]) {
-      signed char sh,tl;
-      tl=get_reg(i_regs->regmap,rt1[i]);
-      sh=get_reg(i_regs->regmap,rs1[i]|64);
-      if(tl>=0){
-        assert(sh>=0);
-        emit_mov(sh,tl);
-        if(imm[i]>32)
-        {
-          emit_sarimm(tl,imm[i]&31,tl);
-        }
-      }
-    }
+    assert(0);
   }
 }
 
 #ifndef shift_assemble
-void shift_assemble(int i,struct regstat *i_regs)
+static void shift_assemble(int i,struct regstat *i_regs)
 {
-  printf("Need shift_assemble for this architecture.\n");
-  exit(1);
+  signed char s,t,shift;
+  if (rt1[i] == 0)
+    return;
+  assert(opcode2[i]<=0x07); // SLLV/SRLV/SRAV
+  t = get_reg(i_regs->regmap, rt1[i]);
+  s = get_reg(i_regs->regmap, rs1[i]);
+  shift = get_reg(i_regs->regmap, rs2[i]);
+  if (t < 0)
+    return;
+
+  if(rs1[i]==0)
+    emit_zeroreg(t);
+  else if(rs2[i]==0) {
+    assert(s>=0);
+    if(s!=t) emit_mov(s,t);
+  }
+  else {
+    host_tempreg_acquire();
+    emit_andimm(shift,31,HOST_TEMPREG);
+    switch(opcode2[i]) {
+    case 4: // SLLV
+      emit_shl(s,HOST_TEMPREG,t);
+      break;
+    case 6: // SRLV
+      emit_shr(s,HOST_TEMPREG,t);
+      break;
+    case 7: // SRAV
+      emit_sar(s,HOST_TEMPREG,t);
+      break;
+    default:
+      assert(0);
+    }
+    host_tempreg_release();
+  }
 }
+
 #endif
 
-void load_assemble(int i,struct regstat *i_regs)
+enum {
+  MTYPE_8000 = 0,
+  MTYPE_8020,
+  MTYPE_0000,
+  MTYPE_A000,
+  MTYPE_1F80,
+};
+
+static int get_ptr_mem_type(u_int a)
+{
+  if(a < 0x00200000) {
+    if(a<0x1000&&((start>>20)==0xbfc||(start>>24)==0xa0))
+      // return wrong, must use memhandler for BIOS self-test to pass
+      // 007 does similar stuff from a00 mirror, weird stuff
+      return MTYPE_8000;
+    return MTYPE_0000;
+  }
+  if(0x1f800000 <= a && a < 0x1f801000)
+    return MTYPE_1F80;
+  if(0x80200000 <= a && a < 0x80800000)
+    return MTYPE_8020;
+  if(0xa0000000 <= a && a < 0xa0200000)
+    return MTYPE_A000;
+  return MTYPE_8000;
+}
+
+static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override)
+{
+  void *jaddr = NULL;
+  int type=0;
+  int mr=rs1[i];
+  if(((smrv_strong|smrv_weak)>>mr)&1) {
+    type=get_ptr_mem_type(smrv[mr]);
+    //printf("set %08x @%08x r%d %d\n", smrv[mr], start+i*4, mr, type);
+  }
+  else {
+    // use the mirror we are running on
+    type=get_ptr_mem_type(start);
+    //printf("set nospec   @%08x r%d %d\n", start+i*4, mr, type);
+  }
+
+  if(type==MTYPE_8020) { // RAM 80200000+ mirror
+    host_tempreg_acquire();
+    emit_andimm(addr,~0x00e00000,HOST_TEMPREG);
+    addr=*addr_reg_override=HOST_TEMPREG;
+    type=0;
+  }
+  else if(type==MTYPE_0000) { // RAM 0 mirror
+    host_tempreg_acquire();
+    emit_orimm(addr,0x80000000,HOST_TEMPREG);
+    addr=*addr_reg_override=HOST_TEMPREG;
+    type=0;
+  }
+  else if(type==MTYPE_A000) { // RAM A mirror
+    host_tempreg_acquire();
+    emit_andimm(addr,~0x20000000,HOST_TEMPREG);
+    addr=*addr_reg_override=HOST_TEMPREG;
+    type=0;
+  }
+  else if(type==MTYPE_1F80) { // scratchpad
+    if (psxH == (void *)0x1f800000) {
+      host_tempreg_acquire();
+      emit_xorimm(addr,0x1f800000,HOST_TEMPREG);
+      emit_cmpimm(HOST_TEMPREG,0x1000);
+      host_tempreg_release();
+      jaddr=out;
+      emit_jc(0);
+    }
+    else {
+      // do the usual RAM check, jump will go to the right handler
+      type=0;
+    }
+  }
+
+  if(type==0)
+  {
+    emit_cmpimm(addr,RAM_SIZE);
+    jaddr=out;
+    #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
+    // Hint to branch predictor that the branch is unlikely to be taken
+    if(rs1[i]>=28)
+      emit_jno_unlikely(0);
+    else
+    #endif
+      emit_jno(0);
+    if(ram_offset!=0) {
+      host_tempreg_acquire();
+      emit_addimm(addr,ram_offset,HOST_TEMPREG);
+      addr=*addr_reg_override=HOST_TEMPREG;
+    }
+  }
+
+  return jaddr;
+}
+
+// return memhandler, or get directly accessable address and return 0
+static void *get_direct_memhandler(void *table, u_int addr,
+  enum stub_type type, uintptr_t *addr_host)
+{
+  uintptr_t l1, l2 = 0;
+  l1 = ((uintptr_t *)table)[addr>>12];
+  if ((l1 & (1ul << (sizeof(l1)*8-1))) == 0) {
+    uintptr_t v = l1 << 1;
+    *addr_host = v + addr;
+    return NULL;
+  }
+  else {
+    l1 <<= 1;
+    if (type == LOADB_STUB || type == LOADBU_STUB || type == STOREB_STUB)
+      l2 = ((uintptr_t *)l1)[0x1000/4 + 0x1000/2 + (addr&0xfff)];
+    else if (type == LOADH_STUB || type == LOADHU_STUB || type == STOREH_STUB)
+      l2=((uintptr_t *)l1)[0x1000/4 + (addr&0xfff)/2];
+    else
+      l2=((uintptr_t *)l1)[(addr&0xfff)/4];
+    if ((l2 & (1<<31)) == 0) {
+      uintptr_t v = l2 << 1;
+      *addr_host = v + (addr&0xfff);
+      return NULL;
+    }
+    return (void *)(l2 << 1);
+  }
+}
+
+static u_int get_host_reglist(const signed char *regmap)
 {
-  int s,th,tl,addr,map=-1;
+  u_int reglist = 0, hr;
+  for (hr = 0; hr < HOST_REGS; hr++) {
+    if (hr != EXCLUDE_REG && regmap[hr] >= 0)
+      reglist |= 1 << hr;
+  }
+  return reglist;
+}
+
+static u_int reglist_exclude(u_int reglist, int r1, int r2)
+{
+  if (r1 >= 0)
+    reglist &= ~(1u << r1);
+  if (r2 >= 0)
+    reglist &= ~(1u << r2);
+  return reglist;
+}
+
+// find a temp caller-saved register not in reglist (so assumed to be free)
+static int reglist_find_free(u_int reglist)
+{
+  u_int free_regs = ~reglist & CALLER_SAVE_REGS;
+  if (free_regs == 0)
+    return -1;
+  return __builtin_ctz(free_regs);
+}
+
+static void load_assemble(int i, const struct regstat *i_regs)
+{
+  int s,tl,addr;
   int offset;
-  int jaddr=0;
+  void *jaddr=0;
   int memtarget=0,c=0;
-  int fastload_reg_override=0;
-  u_int hr,reglist=0;
-  th=get_reg(i_regs->regmap,rt1[i]|64);
+  int fastio_reg_override=-1;
+  u_int reglist=get_host_reglist(i_regs->regmap);
   tl=get_reg(i_regs->regmap,rt1[i]);
   s=get_reg(i_regs->regmap,rs1[i]);
   offset=imm[i];
-  for(hr=0;hr<HOST_REGS;hr++) {
-    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
-  }
   if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
   if(s>=0) {
     c=(i_regs->wasconst>>s)&1;
@@ -2524,9 +2706,9 @@ void load_assemble(int i,struct regstat *i_regs)
     }
   }
   //printf("load_assemble: c=%d\n",c);
-  //if(c) printf("load_assemble: const=%x\n",(int)constmap[i][s]+offset);
+  //if(c) printf("load_assemble: const=%lx\n",(long)constmap[i][s]+offset);
   // FIXME: Even if the load is a NOP, we should check for pagefaults...
-  if(tl<0&&(!c||(((u_int)constmap[i][s]+offset)>>16)==0x1f80)
+  if((tl<0&&(!c||(((u_int)constmap[i][s]+offset)>>16)==0x1f80))
     ||rt1[i]==0) {
       // could be FIFO, must perform the read
       // ||dummy read
@@ -2539,53 +2721,37 @@ void load_assemble(int i,struct regstat *i_regs)
   //if(tl<0) tl=get_reg(i_regs->regmap,-1);
  if(tl>=0) {
   //printf("load_assemble: c=%d\n",c);
-  //if(c) printf("load_assemble: const=%x\n",(int)constmap[i][s]+offset);
+  //if(c) printf("load_assemble: const=%lx\n",(long)constmap[i][s]+offset);
   assert(tl>=0); // Even if the load is a NOP, we must check for pagefaults and I/O
   reglist&=~(1<<tl);
-  if(th>=0) reglist&=~(1<<th);
   if(!c) {
-    #ifdef RAM_OFFSET
-    map=get_reg(i_regs->regmap,ROREG);
-    if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG);
-    #endif
     #ifdef R29_HACK
     // Strmnnrmn's speed hack
     if(rs1[i]!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
     #endif
     {
-      jaddr=emit_fastpath_cmp_jump(i,addr,&fastload_reg_override);
+      jaddr=emit_fastpath_cmp_jump(i,addr,&fastio_reg_override);
     }
   }
   else if(ram_offset&&memtarget) {
+    host_tempreg_acquire();
     emit_addimm(addr,ram_offset,HOST_TEMPREG);
-    fastload_reg_override=HOST_TEMPREG;
+    fastio_reg_override=HOST_TEMPREG;
   }
   int dummy=(rt1[i]==0)||(tl!=get_reg(i_regs->regmap,rt1[i])); // ignore loads to r0 and unneeded reg
   if (opcode[i]==0x20) { // LB
     if(!c||memtarget) {
       if(!dummy) {
-        #ifdef HOST_IMM_ADDR32
-        if(c)
-          emit_movsbl_tlb((constmap[i][s]+offset)^3,map,tl);
-        else
-        #endif
         {
-          //emit_xorimm(addr,3,tl);
-          //emit_movsbl_indexed((int)rdram-0x80000000,tl,tl);
           int x=0,a=tl;
-#ifdef BIG_ENDIAN_MIPS
-          if(!c) emit_xorimm(addr,3,tl);
-          else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
-#else
           if(!c) a=addr;
-#endif
-          if(fastload_reg_override) a=fastload_reg_override;
+          if(fastio_reg_override>=0) a=fastio_reg_override;
 
-          emit_movsbl_indexed_tlb(x,a,map,tl);
+          emit_movsbl_indexed(x,a,tl);
         }
       }
       if(jaddr)
-        add_stub(LOADB_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+        add_stub_r(LOADB_STUB,jaddr,out,i,addr,i_regs,ccadj[i],reglist);
     }
     else
       inline_readstub(LOADB_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
@@ -2593,36 +2759,13 @@ void load_assemble(int i,struct regstat *i_regs)
   if (opcode[i]==0x21) { // LH
     if(!c||memtarget) {
       if(!dummy) {
-        #ifdef HOST_IMM_ADDR32
-        if(c)
-          emit_movswl_tlb((constmap[i][s]+offset)^2,map,tl);
-        else
-        #endif
-        {
-          int x=0,a=tl;
-#ifdef BIG_ENDIAN_MIPS
-          if(!c) emit_xorimm(addr,2,tl);
-          else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
-#else
-          if(!c) a=addr;
-#endif
-          if(fastload_reg_override) a=fastload_reg_override;
-          //#ifdef
-          //emit_movswl_indexed_tlb(x,tl,map,tl);
-          //else
-          if(map>=0) {
-            emit_movswl_indexed(x,a,tl);
-          }else{
-            #if 1 //def RAM_OFFSET
-            emit_movswl_indexed(x,a,tl);
-            #else
-            emit_movswl_indexed((int)rdram-0x80000000+x,a,tl);
-            #endif
-          }
-        }
+        int x=0,a=tl;
+        if(!c) a=addr;
+        if(fastio_reg_override>=0) a=fastio_reg_override;
+        emit_movswl_indexed(x,a,tl);
       }
       if(jaddr)
-        add_stub(LOADH_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+        add_stub_r(LOADH_STUB,jaddr,out,i,addr,i_regs,ccadj[i],reglist);
     }
     else
       inline_readstub(LOADH_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
@@ -2631,17 +2774,11 @@ void load_assemble(int i,struct regstat *i_regs)
     if(!c||memtarget) {
       if(!dummy) {
         int a=addr;
-        if(fastload_reg_override) a=fastload_reg_override;
-        //emit_readword_indexed((int)rdram-0x80000000,addr,tl);
-        #ifdef HOST_IMM_ADDR32
-        if(c)
-          emit_readword_tlb(constmap[i][s]+offset,map,tl);
-        else
-        #endif
-        emit_readword_indexed_tlb(0,a,map,tl);
+        if(fastio_reg_override>=0) a=fastio_reg_override;
+        emit_readword_indexed(0,a,tl);
       }
       if(jaddr)
-        add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+        add_stub_r(LOADW_STUB,jaddr,out,i,addr,i_regs,ccadj[i],reglist);
     }
     else
       inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
@@ -2649,28 +2786,14 @@ void load_assemble(int i,struct regstat *i_regs)
   if (opcode[i]==0x24) { // LBU
     if(!c||memtarget) {
       if(!dummy) {
-        #ifdef HOST_IMM_ADDR32
-        if(c)
-          emit_movzbl_tlb((constmap[i][s]+offset)^3,map,tl);
-        else
-        #endif
-        {
-          //emit_xorimm(addr,3,tl);
-          //emit_movzbl_indexed((int)rdram-0x80000000,tl,tl);
-          int x=0,a=tl;
-#ifdef BIG_ENDIAN_MIPS
-          if(!c) emit_xorimm(addr,3,tl);
-          else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
-#else
-          if(!c) a=addr;
-#endif
-          if(fastload_reg_override) a=fastload_reg_override;
+        int x=0,a=tl;
+        if(!c) a=addr;
+        if(fastio_reg_override>=0) a=fastio_reg_override;
 
-          emit_movzbl_indexed_tlb(x,a,map,tl);
-        }
+        emit_movzbl_indexed(x,a,tl);
       }
       if(jaddr)
-        add_stub(LOADBU_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+        add_stub_r(LOADBU_STUB,jaddr,out,i,addr,i_regs,ccadj[i],reglist);
     }
     else
       inline_readstub(LOADBU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
@@ -2678,133 +2801,120 @@ void load_assemble(int i,struct regstat *i_regs)
   if (opcode[i]==0x25) { // LHU
     if(!c||memtarget) {
       if(!dummy) {
-        #ifdef HOST_IMM_ADDR32
-        if(c)
-          emit_movzwl_tlb((constmap[i][s]+offset)^2,map,tl);
-        else
-        #endif
-        {
-          int x=0,a=tl;
-#ifdef BIG_ENDIAN_MIPS
-          if(!c) emit_xorimm(addr,2,tl);
-          else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
-#else
-          if(!c) a=addr;
-#endif
-          if(fastload_reg_override) a=fastload_reg_override;
-          //#ifdef
-          //emit_movzwl_indexed_tlb(x,tl,map,tl);
-          //#else
-          if(map>=0) {
-            emit_movzwl_indexed(x,a,tl);
-          }else{
-            #if 1 //def RAM_OFFSET
-            emit_movzwl_indexed(x,a,tl);
-            #else
-            emit_movzwl_indexed((int)rdram-0x80000000+x,a,tl);
-            #endif
-          }
-        }
+        int x=0,a=tl;
+        if(!c) a=addr;
+        if(fastio_reg_override>=0) a=fastio_reg_override;
+        emit_movzwl_indexed(x,a,tl);
       }
       if(jaddr)
-        add_stub(LOADHU_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+        add_stub_r(LOADHU_STUB,jaddr,out,i,addr,i_regs,ccadj[i],reglist);
     }
     else
       inline_readstub(LOADHU_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
   }
   if (opcode[i]==0x27) { // LWU
-    assert(th>=0);
-    if(!c||memtarget) {
-      if(!dummy) {
-        int a=addr;
-        if(fastload_reg_override) a=fastload_reg_override;
-        //emit_readword_indexed((int)rdram-0x80000000,addr,tl);
-        #ifdef HOST_IMM_ADDR32
-        if(c)
-          emit_readword_tlb(constmap[i][s]+offset,map,tl);
-        else
-        #endif
-        emit_readword_indexed_tlb(0,a,map,tl);
-      }
-      if(jaddr)
-        add_stub(LOADW_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
-    }
-    else {
-      inline_readstub(LOADW_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
-    }
-    emit_zeroreg(th);
+    assert(0);
   }
   if (opcode[i]==0x37) { // LD
-    if(!c||memtarget) {
-      if(!dummy) {
-        int a=addr;
-        if(fastload_reg_override) a=fastload_reg_override;
-        //if(th>=0) emit_readword_indexed((int)rdram-0x80000000,addr,th);
-        //emit_readword_indexed((int)rdram-0x7FFFFFFC,addr,tl);
-        #ifdef HOST_IMM_ADDR32
-        if(c)
-          emit_readdword_tlb(constmap[i][s]+offset,map,th,tl);
-        else
-        #endif
-        emit_readdword_indexed_tlb(0,a,map,th,tl);
-      }
-      if(jaddr)
-        add_stub(LOADD_STUB,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
-    }
-    else
-      inline_readstub(LOADD_STUB,i,constmap[i][s]+offset,i_regs->regmap,rt1[i],ccadj[i],reglist);
+    assert(0);
   }
  }
-  //emit_storereg(rt1[i],tl); // DEBUG
-  //if(opcode[i]==0x23)
-  //if(opcode[i]==0x24)
-  //if(opcode[i]==0x23||opcode[i]==0x24)
-  /*if(opcode[i]==0x21||opcode[i]==0x23||opcode[i]==0x24)
-  {
-    //emit_pusha();
-    save_regs(0x100f);
-        emit_readword((int)&last_count,ECX);
-        #ifdef __i386__
-        if(get_reg(i_regs->regmap,CCREG)<0)
-          emit_loadreg(CCREG,HOST_CCREG);
-        emit_add(HOST_CCREG,ECX,HOST_CCREG);
-        emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
-        emit_writeword(HOST_CCREG,(int)&Count);
-        #endif
-        #ifdef __arm__
-        if(get_reg(i_regs->regmap,CCREG)<0)
-          emit_loadreg(CCREG,0);
-        else
-          emit_mov(HOST_CCREG,0);
-        emit_add(0,ECX,0);
-        emit_addimm(0,2*ccadj[i],0);
-        emit_writeword(0,(int)&Count);
-        #endif
-    emit_call((int)memdebug);
-    //emit_popa();
-    restore_regs(0x100f);
-  }/**/
+ if (fastio_reg_override == HOST_TEMPREG)
+   host_tempreg_release();
 }
 
 #ifndef loadlr_assemble
-void loadlr_assemble(int i,struct regstat *i_regs)
+static void loadlr_assemble(int i, const struct regstat *i_regs)
 {
-  printf("Need loadlr_assemble for this architecture.\n");
-  exit(1);
+  int s,tl,temp,temp2,addr;
+  int offset;
+  void *jaddr=0;
+  int memtarget=0,c=0;
+  int fastio_reg_override=-1;
+  u_int reglist=get_host_reglist(i_regs->regmap);
+  tl=get_reg(i_regs->regmap,rt1[i]);
+  s=get_reg(i_regs->regmap,rs1[i]);
+  temp=get_reg(i_regs->regmap,-1);
+  temp2=get_reg(i_regs->regmap,FTEMP);
+  addr=get_reg(i_regs->regmap,AGEN1+(i&1));
+  assert(addr<0);
+  offset=imm[i];
+  reglist|=1<<temp;
+  if(offset||s<0||c) addr=temp2;
+  else addr=s;
+  if(s>=0) {
+    c=(i_regs->wasconst>>s)&1;
+    if(c) {
+      memtarget=((signed int)(constmap[i][s]+offset))<(signed int)0x80000000+RAM_SIZE;
+    }
+  }
+  if(!c) {
+    emit_shlimm(addr,3,temp);
+    if (opcode[i]==0x22||opcode[i]==0x26) {
+      emit_andimm(addr,0xFFFFFFFC,temp2); // LWL/LWR
+    }else{
+      emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR
+    }
+    jaddr=emit_fastpath_cmp_jump(i,temp2,&fastio_reg_override);
+  }
+  else {
+    if(ram_offset&&memtarget) {
+      host_tempreg_acquire();
+      emit_addimm(temp2,ram_offset,HOST_TEMPREG);
+      fastio_reg_override=HOST_TEMPREG;
+    }
+    if (opcode[i]==0x22||opcode[i]==0x26) {
+      emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
+    }else{
+      emit_movimm(((constmap[i][s]+offset)<<3)&56,temp); // LDL/LDR
+    }
+  }
+  if (opcode[i]==0x22||opcode[i]==0x26) { // LWL/LWR
+    if(!c||memtarget) {
+      int a=temp2;
+      if(fastio_reg_override>=0) a=fastio_reg_override;
+      emit_readword_indexed(0,a,temp2);
+      if(fastio_reg_override==HOST_TEMPREG) host_tempreg_release();
+      if(jaddr) add_stub_r(LOADW_STUB,jaddr,out,i,temp2,i_regs,ccadj[i],reglist);
+    }
+    else
+      inline_readstub(LOADW_STUB,i,(constmap[i][s]+offset)&0xFFFFFFFC,i_regs->regmap,FTEMP,ccadj[i],reglist);
+    if(rt1[i]) {
+      assert(tl>=0);
+      emit_andimm(temp,24,temp);
+      if (opcode[i]==0x22) // LWL
+        emit_xorimm(temp,24,temp);
+      host_tempreg_acquire();
+      emit_movimm(-1,HOST_TEMPREG);
+      if (opcode[i]==0x26) {
+        emit_shr(temp2,temp,temp2);
+        emit_bic_lsr(tl,HOST_TEMPREG,temp,tl);
+      }else{
+        emit_shl(temp2,temp,temp2);
+        emit_bic_lsl(tl,HOST_TEMPREG,temp,tl);
+      }
+      host_tempreg_release();
+      emit_or(temp2,tl,tl);
+    }
+    //emit_storereg(rt1[i],tl); // DEBUG
+  }
+  if (opcode[i]==0x1A||opcode[i]==0x1B) { // LDL/LDR
+    assert(0);
+  }
 }
 #endif
 
-void store_assemble(int i,struct regstat *i_regs)
+void store_assemble(int i, const struct regstat *i_regs)
 {
-  int s,th,tl,map=-1;
+  int s,tl;
   int addr,temp;
   int offset;
-  int jaddr=0,jaddr2,type;
+  void *jaddr=0;
+  enum stub_type type;
   int memtarget=0,c=0;
   int agr=AGEN1+(i&1);
-  int faststore_reg_override=0;
-  u_int hr,reglist=0;
-  th=get_reg(i_regs->regmap,rs2[i]|64);
+  int fastio_reg_override=-1;
+  u_int reglist=get_host_reglist(i_regs->regmap);
   tl=get_reg(i_regs->regmap,rs2[i]);
   s=get_reg(i_regs->regmap,rs1[i]);
   temp=get_reg(i_regs->regmap,agr);
@@ -2818,90 +2928,57 @@ void store_assemble(int i,struct regstat *i_regs)
   }
   assert(tl>=0);
   assert(temp>=0);
-  for(hr=0;hr<HOST_REGS;hr++) {
-    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
-  }
   if(i_regs->regmap[HOST_CCREG]==CCREG) reglist&=~(1<<HOST_CCREG);
   if(offset||s<0||c) addr=temp;
   else addr=s;
   if(!c) {
-    jaddr=emit_fastpath_cmp_jump(i,addr,&faststore_reg_override);
+    jaddr=emit_fastpath_cmp_jump(i,addr,&fastio_reg_override);
   }
   else if(ram_offset&&memtarget) {
+    host_tempreg_acquire();
     emit_addimm(addr,ram_offset,HOST_TEMPREG);
-    faststore_reg_override=HOST_TEMPREG;
+    fastio_reg_override=HOST_TEMPREG;
   }
 
   if (opcode[i]==0x28) { // SB
     if(!c||memtarget) {
       int x=0,a=temp;
-#ifdef BIG_ENDIAN_MIPS
-      if(!c) emit_xorimm(addr,3,temp);
-      else x=((constmap[i][s]+offset)^3)-(constmap[i][s]+offset);
-#else
       if(!c) a=addr;
-#endif
-      if(faststore_reg_override) a=faststore_reg_override;
-      //emit_writebyte_indexed(tl,(int)rdram-0x80000000,temp);
-      emit_writebyte_indexed_tlb(tl,x,a,map,a);
+      if(fastio_reg_override>=0) a=fastio_reg_override;
+      emit_writebyte_indexed(tl,x,a);
     }
     type=STOREB_STUB;
   }
   if (opcode[i]==0x29) { // SH
     if(!c||memtarget) {
       int x=0,a=temp;
-#ifdef BIG_ENDIAN_MIPS
-      if(!c) emit_xorimm(addr,2,temp);
-      else x=((constmap[i][s]+offset)^2)-(constmap[i][s]+offset);
-#else
       if(!c) a=addr;
-#endif
-      if(faststore_reg_override) a=faststore_reg_override;
-      //#ifdef
-      //emit_writehword_indexed_tlb(tl,x,temp,map,temp);
-      //#else
-      if(map>=0) {
-        emit_writehword_indexed(tl,x,a);
-      }else
-        //emit_writehword_indexed(tl,(int)rdram-0x80000000+x,a);
-        emit_writehword_indexed(tl,x,a);
+      if(fastio_reg_override>=0) a=fastio_reg_override;
+      emit_writehword_indexed(tl,x,a);
     }
     type=STOREH_STUB;
   }
   if (opcode[i]==0x2B) { // SW
     if(!c||memtarget) {
       int a=addr;
-      if(faststore_reg_override) a=faststore_reg_override;
-      //emit_writeword_indexed(tl,(int)rdram-0x80000000,addr);
-      emit_writeword_indexed_tlb(tl,0,a,map,temp);
+      if(fastio_reg_override>=0) a=fastio_reg_override;
+      emit_writeword_indexed(tl,0,a);
     }
     type=STOREW_STUB;
   }
   if (opcode[i]==0x3F) { // SD
-    if(!c||memtarget) {
-      int a=addr;
-      if(faststore_reg_override) a=faststore_reg_override;
-      if(rs2[i]) {
-        assert(th>=0);
-        //emit_writeword_indexed(th,(int)rdram-0x80000000,addr);
-        //emit_writeword_indexed(tl,(int)rdram-0x7FFFFFFC,addr);
-        emit_writedword_indexed_tlb(th,tl,0,a,map,temp);
-      }else{
-        // Store zero
-        //emit_writeword_indexed(tl,(int)rdram-0x80000000,temp);
-        //emit_writeword_indexed(tl,(int)rdram-0x7FFFFFFC,temp);
-        emit_writedword_indexed_tlb(tl,tl,0,a,map,temp);
-      }
-    }
+    assert(0);
     type=STORED_STUB;
   }
+  if(fastio_reg_override==HOST_TEMPREG)
+    host_tempreg_release();
   if(jaddr) {
     // PCSX store handlers don't check invcode again
     reglist|=1<<addr;
-    add_stub(type,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+    add_stub_r(type,jaddr,out,i,addr,i_regs,ccadj[i],reglist);
     jaddr=0;
   }
-  if(!(i_regs->waswritten&(1<<rs1[i]))&&!(new_dynarec_hacks&NDHACK_NO_SMC_CHECK)) {
+  if(!(i_regs->waswritten&(1<<rs1[i])) && !HACK_ENABLED(NDHACK_NO_SMC_CHECK)) {
     if(!c||memtarget) {
       #ifdef DESTRUCTIVE_SHIFT
       // The x86 shift operation is 'destructive'; it overwrites the
@@ -2913,88 +2990,52 @@ void store_assemble(int i,struct regstat *i_regs)
       assert(ir>=0);
       emit_cmpmem_indexedsr12_reg(ir,addr,1);
       #else
-      emit_cmpmem_indexedsr12_imm((int)invalid_code,addr,1);
+      emit_cmpmem_indexedsr12_imm(invalid_code,addr,1);
       #endif
       #if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
       emit_callne(invalidate_addr_reg[addr]);
       #else
-      jaddr2=(int)out;
+      void *jaddr2 = out;
       emit_jne(0);
-      add_stub(INVCODE_STUB,jaddr2,(int)out,reglist|(1<<HOST_CCREG),addr,0,0,0);
+      add_stub(INVCODE_STUB,jaddr2,out,reglist|(1<<HOST_CCREG),addr,0,0,0);
       #endif
     }
   }
   u_int addr_val=constmap[i][s]+offset;
   if(jaddr) {
-    add_stub(type,jaddr,(int)out,i,addr,(int)i_regs,ccadj[i],reglist);
+    add_stub_r(type,jaddr,out,i,addr,i_regs,ccadj[i],reglist);
   } else if(c&&!memtarget) {
     inline_writestub(type,i,addr_val,i_regs->regmap,rs2[i],ccadj[i],reglist);
   }
   // basic current block modification detection..
   // not looking back as that should be in mips cache already
+  // (see Spyro2 title->attract mode)
   if(c&&start+i*4<addr_val&&addr_val<start+slen*4) {
     SysPrintf("write to %08x hits block %08x, pc=%08x\n",addr_val,start,start+i*4);
     assert(i_regs->regmap==regs[i].regmap); // not delay slot
     if(i_regs->regmap==regs[i].regmap) {
-      load_all_consts(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty,i);
-      wb_dirtys(regs[i].regmap_entry,regs[i].was32,regs[i].wasdirty);
+      load_all_consts(regs[i].regmap_entry,regs[i].wasdirty,i);
+      wb_dirtys(regs[i].regmap_entry,regs[i].wasdirty);
       emit_movimm(start+i*4+4,0);
-      emit_writeword(0,(int)&pcaddr);
-      emit_jmp((int)do_interrupt);
+      emit_writeword(0,&pcaddr);
+      emit_addimm(HOST_CCREG,2,HOST_CCREG);
+      emit_far_call(get_addr_ht);
+      emit_jmpreg(0);
     }
   }
-  //if(opcode[i]==0x2B || opcode[i]==0x3F)
-  //if(opcode[i]==0x2B || opcode[i]==0x28)
-  //if(opcode[i]==0x2B || opcode[i]==0x29)
-  //if(opcode[i]==0x2B)
-  /*if(opcode[i]==0x2B || opcode[i]==0x28 || opcode[i]==0x29 || opcode[i]==0x3F)
-  {
-    #ifdef __i386__
-    emit_pusha();
-    #endif
-    #ifdef __arm__
-    save_regs(0x100f);
-    #endif
-        emit_readword((int)&last_count,ECX);
-        #ifdef __i386__
-        if(get_reg(i_regs->regmap,CCREG)<0)
-          emit_loadreg(CCREG,HOST_CCREG);
-        emit_add(HOST_CCREG,ECX,HOST_CCREG);
-        emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
-        emit_writeword(HOST_CCREG,(int)&Count);
-        #endif
-        #ifdef __arm__
-        if(get_reg(i_regs->regmap,CCREG)<0)
-          emit_loadreg(CCREG,0);
-        else
-          emit_mov(HOST_CCREG,0);
-        emit_add(0,ECX,0);
-        emit_addimm(0,2*ccadj[i],0);
-        emit_writeword(0,(int)&Count);
-        #endif
-    emit_call((int)memdebug);
-    #ifdef __i386__
-    emit_popa();
-    #endif
-    #ifdef __arm__
-    restore_regs(0x100f);
-    #endif
-  }/**/
 }
 
-void storelr_assemble(int i,struct regstat *i_regs)
+static void storelr_assemble(int i, const struct regstat *i_regs)
 {
-  int s,th,tl;
+  int s,tl;
   int temp;
-  int temp2;
   int offset;
-  int jaddr=0,jaddr2;
-  int case1,case2,case3;
-  int done0,done1,done2;
+  void *jaddr=0;
+  void *case1, *case2, *case3;
+  void *done0, *done1, *done2;
   int memtarget=0,c=0;
   int agr=AGEN1+(i&1);
-  u_int hr,reglist=0;
-  th=get_reg(i_regs->regmap,rs2[i]|64);
+  u_int reglist=get_host_reglist(i_regs->regmap);
   tl=get_reg(i_regs->regmap,rs2[i]);
   s=get_reg(i_regs->regmap,rs1[i]);
   temp=get_reg(i_regs->regmap,agr);
@@ -3007,64 +3048,47 @@ void storelr_assemble(int i,struct regstat *i_regs)
     }
   }
   assert(tl>=0);
-  for(hr=0;hr<HOST_REGS;hr++) {
-    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
-  }
   assert(temp>=0);
   if(!c) {
     emit_cmpimm(s<0||offset?temp:s,RAM_SIZE);
     if(!offset&&s!=temp) emit_mov(s,temp);
-    jaddr=(int)out;
+    jaddr=out;
     emit_jno(0);
   }
   else
   {
     if(!memtarget||!rs1[i]) {
-      jaddr=(int)out;
+      jaddr=out;
       emit_jmp(0);
     }
   }
-  #ifdef RAM_OFFSET
-  int map=get_reg(i_regs->regmap,ROREG);
-  if(map<0) emit_loadreg(ROREG,map=HOST_TEMPREG);
-  #else
-  if((u_int)rdram!=0x80000000)
-    emit_addimm_no_flags((u_int)rdram-(u_int)0x80000000,temp);
-  #endif
+  if(ram_offset)
+    emit_addimm_no_flags(ram_offset,temp);
 
   if (opcode[i]==0x2C||opcode[i]==0x2D) { // SDL/SDR
-    temp2=get_reg(i_regs->regmap,FTEMP);
-    if(!rs2[i]) temp2=th=tl;
+    assert(0);
   }
 
-#ifndef BIG_ENDIAN_MIPS
-    emit_xorimm(temp,3,temp);
-#endif
+  emit_xorimm(temp,3,temp);
   emit_testimm(temp,2);
-  case2=(int)out;
+  case2=out;
   emit_jne(0);
   emit_testimm(temp,1);
-  case1=(int)out;
+  case1=out;
   emit_jne(0);
   // 0
   if (opcode[i]==0x2A) { // SWL
     emit_writeword_indexed(tl,0,temp);
   }
-  if (opcode[i]==0x2E) { // SWR
+  else if (opcode[i]==0x2E) { // SWR
     emit_writebyte_indexed(tl,3,temp);
   }
-  if (opcode[i]==0x2C) { // SDL
-    emit_writeword_indexed(th,0,temp);
-    if(rs2[i]) emit_mov(tl,temp2);
-  }
-  if (opcode[i]==0x2D) { // SDR
-    emit_writebyte_indexed(tl,3,temp);
-    if(rs2[i]) emit_shldimm(th,tl,24,temp2);
-  }
-  done0=(int)out;
+  else
+    assert(0);
+  done0=out;
   emit_jmp(0);
   // 1
-  set_jump_target(case1,(int)out);
+  set_jump_target(case1, out);
   if (opcode[i]==0x2A) { // SWL
     // Write 3 msb into three least significant bytes
     if(rs2[i]) emit_rorimm(tl,8,tl);
@@ -3073,30 +3097,16 @@ void storelr_assemble(int i,struct regstat *i_regs)
     emit_writebyte_indexed(tl,1,temp);
     if(rs2[i]) emit_rorimm(tl,8,tl);
   }
-  if (opcode[i]==0x2E) { // SWR
-    // Write two lsb into two most significant bytes
-    emit_writehword_indexed(tl,1,temp);
-  }
-  if (opcode[i]==0x2C) { // SDL
-    if(rs2[i]) emit_shrdimm(tl,th,8,temp2);
-    // Write 3 msb into three least significant bytes
-    if(rs2[i]) emit_rorimm(th,8,th);
-    emit_writehword_indexed(th,-1,temp);
-    if(rs2[i]) emit_rorimm(th,16,th);
-    emit_writebyte_indexed(th,1,temp);
-    if(rs2[i]) emit_rorimm(th,8,th);
-  }
-  if (opcode[i]==0x2D) { // SDR
-    if(rs2[i]) emit_shldimm(th,tl,16,temp2);
+  else if (opcode[i]==0x2E) { // SWR
     // Write two lsb into two most significant bytes
     emit_writehword_indexed(tl,1,temp);
   }
-  done1=(int)out;
+  done1=out;
   emit_jmp(0);
   // 2
-  set_jump_target(case2,(int)out);
+  set_jump_target(case2, out);
   emit_testimm(temp,1);
-  case3=(int)out;
+  case3=out;
   emit_jne(0);
   if (opcode[i]==0x2A) { // SWL
     // Write two msb into two least significant bytes
@@ -3104,128 +3114,401 @@ void storelr_assemble(int i,struct regstat *i_regs)
     emit_writehword_indexed(tl,-2,temp);
     if(rs2[i]) emit_rorimm(tl,16,tl);
   }
-  if (opcode[i]==0x2E) { // SWR
-    // Write 3 lsb into three most significant bytes
-    emit_writebyte_indexed(tl,-1,temp);
-    if(rs2[i]) emit_rorimm(tl,8,tl);
-    emit_writehword_indexed(tl,0,temp);
-    if(rs2[i]) emit_rorimm(tl,24,tl);
-  }
-  if (opcode[i]==0x2C) { // SDL
-    if(rs2[i]) emit_shrdimm(tl,th,16,temp2);
-    // Write two msb into two least significant bytes
-    if(rs2[i]) emit_rorimm(th,16,th);
-    emit_writehword_indexed(th,-2,temp);
-    if(rs2[i]) emit_rorimm(th,16,th);
-  }
-  if (opcode[i]==0x2D) { // SDR
-    if(rs2[i]) emit_shldimm(th,tl,8,temp2);
+  else if (opcode[i]==0x2E) { // SWR
     // Write 3 lsb into three most significant bytes
     emit_writebyte_indexed(tl,-1,temp);
     if(rs2[i]) emit_rorimm(tl,8,tl);
     emit_writehword_indexed(tl,0,temp);
     if(rs2[i]) emit_rorimm(tl,24,tl);
   }
-  done2=(int)out;
+  done2=out;
   emit_jmp(0);
   // 3
-  set_jump_target(case3,(int)out);
+  set_jump_target(case3, out);
   if (opcode[i]==0x2A) { // SWL
     // Write msb into least significant byte
     if(rs2[i]) emit_rorimm(tl,24,tl);
     emit_writebyte_indexed(tl,-3,temp);
     if(rs2[i]) emit_rorimm(tl,8,tl);
   }
-  if (opcode[i]==0x2E) { // SWR
-    // Write entire word
-    emit_writeword_indexed(tl,-3,temp);
-  }
-  if (opcode[i]==0x2C) { // SDL
-    if(rs2[i]) emit_shrdimm(tl,th,24,temp2);
-    // Write msb into least significant byte
-    if(rs2[i]) emit_rorimm(th,24,th);
-    emit_writebyte_indexed(th,-3,temp);
-    if(rs2[i]) emit_rorimm(th,8,th);
-  }
-  if (opcode[i]==0x2D) { // SDR
-    if(rs2[i]) emit_mov(th,temp2);
+  else if (opcode[i]==0x2E) { // SWR
     // Write entire word
     emit_writeword_indexed(tl,-3,temp);
   }
-  set_jump_target(done0,(int)out);
-  set_jump_target(done1,(int)out);
-  set_jump_target(done2,(int)out);
-  if (opcode[i]==0x2C) { // SDL
-    emit_testimm(temp,4);
-    done0=(int)out;
-    emit_jne(0);
-    emit_andimm(temp,~3,temp);
-    emit_writeword_indexed(temp2,4,temp);
-    set_jump_target(done0,(int)out);
-  }
-  if (opcode[i]==0x2D) { // SDR
-    emit_testimm(temp,4);
-    done0=(int)out;
-    emit_jeq(0);
-    emit_andimm(temp,~3,temp);
-    emit_writeword_indexed(temp2,-4,temp);
-    set_jump_target(done0,(int)out);
-  }
+  set_jump_target(done0, out);
+  set_jump_target(done1, out);
+  set_jump_target(done2, out);
   if(!c||!memtarget)
-    add_stub(STORELR_STUB,jaddr,(int)out,i,(int)i_regs,temp,ccadj[i],reglist);
-  if(!(i_regs->waswritten&(1<<rs1[i]))&&!(new_dynarec_hacks&NDHACK_NO_SMC_CHECK)) {
-    #ifdef RAM_OFFSET
-    int map=get_reg(i_regs->regmap,ROREG);
-    if(map<0) map=HOST_TEMPREG;
-    gen_orig_addr_w(temp,map);
-    #else
-    emit_addimm_no_flags((u_int)0x80000000-(u_int)rdram,temp);
-    #endif
+    add_stub_r(STORELR_STUB,jaddr,out,i,temp,i_regs,ccadj[i],reglist);
+  if(!(i_regs->waswritten&(1<<rs1[i])) && !HACK_ENABLED(NDHACK_NO_SMC_CHECK)) {
+    emit_addimm_no_flags(-ram_offset,temp);
     #if defined(HOST_IMM8)
     int ir=get_reg(i_regs->regmap,INVCP);
     assert(ir>=0);
     emit_cmpmem_indexedsr12_reg(ir,temp,1);
     #else
-    emit_cmpmem_indexedsr12_imm((int)invalid_code,temp,1);
+    emit_cmpmem_indexedsr12_imm(invalid_code,temp,1);
     #endif
     #if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
     emit_callne(invalidate_addr_reg[temp]);
     #else
-    jaddr2=(int)out;
+    void *jaddr2 = out;
     emit_jne(0);
-    add_stub(INVCODE_STUB,jaddr2,(int)out,reglist|(1<<HOST_CCREG),temp,0,0,0);
+    add_stub(INVCODE_STUB,jaddr2,out,reglist|(1<<HOST_CCREG),temp,0,0,0);
     #endif
   }
-  /*
-    emit_pusha();
-    //save_regs(0x100f);
-        emit_readword((int)&last_count,ECX);
-        if(get_reg(i_regs->regmap,CCREG)<0)
-          emit_loadreg(CCREG,HOST_CCREG);
-        emit_add(HOST_CCREG,ECX,HOST_CCREG);
-        emit_addimm(HOST_CCREG,2*ccadj[i],HOST_CCREG);
-        emit_writeword(HOST_CCREG,(int)&Count);
-    emit_call((int)memdebug);
-    emit_popa();
-    //restore_regs(0x100f);
-  /**/
 }
 
-void c1ls_assemble(int i,struct regstat *i_regs)
+static void cop0_assemble(int i,struct regstat *i_regs)
+{
+  if(opcode2[i]==0) // MFC0
+  {
+    signed char t=get_reg(i_regs->regmap,rt1[i]);
+    u_int copr=(source[i]>>11)&0x1f;
+    //assert(t>=0); // Why does this happen?  OOT is weird
+    if(t>=0&&rt1[i]!=0) {
+      emit_readword(&reg_cop0[copr],t);
+    }
+  }
+  else if(opcode2[i]==4) // MTC0
+  {
+    signed char s=get_reg(i_regs->regmap,rs1[i]);
+    char copr=(source[i]>>11)&0x1f;
+    assert(s>=0);
+    wb_register(rs1[i],i_regs->regmap,i_regs->dirty);
+    if(copr==9||copr==11||copr==12||copr==13) {
+      emit_readword(&last_count,HOST_TEMPREG);
+      emit_loadreg(CCREG,HOST_CCREG); // TODO: do proper reg alloc
+      emit_add(HOST_CCREG,HOST_TEMPREG,HOST_CCREG);
+      emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG);
+      emit_writeword(HOST_CCREG,&Count);
+    }
+    // What a mess.  The status register (12) can enable interrupts,
+    // so needs a special case to handle a pending interrupt.
+    // The interrupt must be taken immediately, because a subsequent
+    // instruction might disable interrupts again.
+    if(copr==12||copr==13) {
+      if (is_delayslot) {
+        // burn cycles to cause cc_interrupt, which will
+        // reschedule next_interupt. Relies on CCREG from above.
+        assem_debug("MTC0 DS %d\n", copr);
+        emit_writeword(HOST_CCREG,&last_count);
+        emit_movimm(0,HOST_CCREG);
+        emit_storereg(CCREG,HOST_CCREG);
+        emit_loadreg(rs1[i],1);
+        emit_movimm(copr,0);
+        emit_far_call(pcsx_mtc0_ds);
+        emit_loadreg(rs1[i],s);
+        return;
+      }
+      emit_movimm(start+i*4+4,HOST_TEMPREG);
+      emit_writeword(HOST_TEMPREG,&pcaddr);
+      emit_movimm(0,HOST_TEMPREG);
+      emit_writeword(HOST_TEMPREG,&pending_exception);
+    }
+    if(s==HOST_CCREG)
+      emit_loadreg(rs1[i],1);
+    else if(s!=1)
+      emit_mov(s,1);
+    emit_movimm(copr,0);
+    emit_far_call(pcsx_mtc0);
+    if(copr==9||copr==11||copr==12||copr==13) {
+      emit_readword(&Count,HOST_CCREG);
+      emit_readword(&next_interupt,HOST_TEMPREG);
+      emit_addimm(HOST_CCREG,-CLOCK_ADJUST(ccadj[i]),HOST_CCREG);
+      emit_sub(HOST_CCREG,HOST_TEMPREG,HOST_CCREG);
+      emit_writeword(HOST_TEMPREG,&last_count);
+      emit_storereg(CCREG,HOST_CCREG);
+    }
+    if(copr==12||copr==13) {
+      assert(!is_delayslot);
+      emit_readword(&pending_exception,14);
+      emit_test(14,14);
+      void *jaddr = out;
+      emit_jeq(0);
+      emit_readword(&pcaddr, 0);
+      emit_addimm(HOST_CCREG,2,HOST_CCREG);
+      emit_far_call(get_addr_ht);
+      emit_jmpreg(0);
+      set_jump_target(jaddr, out);
+    }
+    emit_loadreg(rs1[i],s);
+  }
+  else
+  {
+    assert(opcode2[i]==0x10);
+    //if((source[i]&0x3f)==0x10) // RFE
+    {
+      emit_readword(&Status,0);
+      emit_andimm(0,0x3c,1);
+      emit_andimm(0,~0xf,0);
+      emit_orrshr_imm(1,2,0);
+      emit_writeword(0,&Status);
+    }
+  }
+}
+
+static void cop1_unusable(int i,struct regstat *i_regs)
+{
+  // XXX: should just just do the exception instead
+  //if(!cop1_usable)
+  {
+    void *jaddr=out;
+    emit_jmp(0);
+    add_stub_r(FP_STUB,jaddr,out,i,0,i_regs,is_delayslot,0);
+  }
+}
+
+static void cop1_assemble(int i,struct regstat *i_regs)
+{
+  cop1_unusable(i, i_regs);
+}
+
+static void c1ls_assemble(int i,struct regstat *i_regs)
+{
+  cop1_unusable(i, i_regs);
+}
+
+// FP_STUB
+static void do_cop1stub(int n)
+{
+  literal_pool(256);
+  assem_debug("do_cop1stub %x\n",start+stubs[n].a*4);
+  set_jump_target(stubs[n].addr, out);
+  int i=stubs[n].a;
+//  int rs=stubs[n].b;
+  struct regstat *i_regs=(struct regstat *)stubs[n].c;
+  int ds=stubs[n].d;
+  if(!ds) {
+    load_all_consts(regs[i].regmap_entry,regs[i].wasdirty,i);
+    //if(i_regs!=&regs[i]) printf("oops: regs[i]=%x i_regs=%x",(int)&regs[i],(int)i_regs);
+  }
+  //else {printf("fp exception in delay slot\n");}
+  wb_dirtys(i_regs->regmap_entry,i_regs->wasdirty);
+  if(regs[i].regmap_entry[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
+  emit_movimm(start+(i-ds)*4,EAX); // Get PC
+  emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // CHECK: is this right?  There should probably be an extra cycle...
+  emit_far_jump(ds?fp_exception_ds:fp_exception);
+}
+
+static int cop2_is_stalling_op(int i, int *cycles)
+{
+  if (opcode[i] == 0x3a) { // SWC2
+    *cycles = 0;
+    return 1;
+  }
+  if (itype[i] == COP2 && (opcode2[i] == 0 || opcode2[i] == 2)) { // MFC2/CFC2
+    *cycles = 0;
+    return 1;
+  }
+  if (itype[i] == C2OP) {
+    *cycles = gte_cycletab[source[i] & 0x3f];
+    return 1;
+  }
+  // ... what about MTC2/CTC2/LWC2?
+  return 0;
+}
+
+#if 0
+static void log_gte_stall(int stall, u_int cycle)
+{
+  if ((u_int)stall <= 44)
+    printf("x    stall %2d %u\n", stall, cycle + last_count);
+ if (cycle + last_count > 1215348544) exit(1);
+}
+
+static void emit_log_gte_stall(int i, int stall, u_int reglist)
+{
+  save_regs(reglist);
+  if (stall > 0)
+    emit_movimm(stall, 0);
+  else
+    emit_mov(HOST_TEMPREG, 0);
+  emit_addimm(HOST_CCREG, CLOCK_ADJUST(ccadj[i]), 1);
+  emit_far_call(log_gte_stall);
+  restore_regs(reglist);
+}
+#endif
+
+static void cop2_call_stall_check(u_int op, int i, const struct regstat *i_regs, u_int reglist)
+{
+  int j = i, other_gte_op_cycles = -1, stall = -MAXBLOCK, cycles_passed;
+  int rtmp = reglist_find_free(reglist);
+
+  if (HACK_ENABLED(NDHACK_GTE_NO_STALL))
+    return;
+  //assert(get_reg(i_regs->regmap, CCREG) == HOST_CCREG);
+  if (get_reg(i_regs->regmap, CCREG) != HOST_CCREG) {
+    // happens occasionally... cc evicted? Don't bother then
+    //printf("no cc %08x\n", start + i*4);
+    return;
+  }
+  if (!bt[i]) {
+    for (j = i - 1; j >= 0; j--) {
+      //if (is_ds[j]) break;
+      if (cop2_is_stalling_op(j, &other_gte_op_cycles) || bt[j])
+        break;
+    }
+  }
+  cycles_passed = CLOCK_ADJUST(ccadj[i] - ccadj[j]);
+  if (other_gte_op_cycles >= 0)
+    stall = other_gte_op_cycles - cycles_passed;
+  else if (cycles_passed >= 44)
+    stall = 0; // can't stall
+  if (stall == -MAXBLOCK && rtmp >= 0) {
+    // unknown stall, do the expensive runtime check
+    assem_debug("; cop2_call_stall_check\n");
+#if 0 // too slow
+    save_regs(reglist);
+    emit_movimm(gte_cycletab[op], 0);
+    emit_addimm(HOST_CCREG, CLOCK_ADJUST(ccadj[i]), 1);
+    emit_far_call(call_gteStall);
+    restore_regs(reglist);
+#else
+    host_tempreg_acquire();
+    emit_readword(&psxRegs.gteBusyCycle, rtmp);
+    emit_addimm(rtmp, -CLOCK_ADJUST(ccadj[i]), rtmp);
+    emit_sub(rtmp, HOST_CCREG, HOST_TEMPREG);
+    emit_cmpimm(HOST_TEMPREG, 44);
+    emit_cmovb_reg(rtmp, HOST_CCREG);
+    //emit_log_gte_stall(i, 0, reglist);
+    host_tempreg_release();
+#endif
+  }
+  else if (stall > 0) {
+    //emit_log_gte_stall(i, stall, reglist);
+    emit_addimm(HOST_CCREG, stall, HOST_CCREG);
+  }
+
+  // save gteBusyCycle, if needed
+  if (gte_cycletab[op] == 0)
+    return;
+  other_gte_op_cycles = -1;
+  for (j = i + 1; j < slen; j++) {
+    if (cop2_is_stalling_op(j, &other_gte_op_cycles))
+      break;
+    if (is_jump(j)) {
+      // check ds
+      if (j + 1 < slen && cop2_is_stalling_op(j + 1, &other_gte_op_cycles))
+        j++;
+      break;
+    }
+  }
+  if (other_gte_op_cycles >= 0)
+    // will handle stall when assembling that op
+    return;
+  cycles_passed = CLOCK_ADJUST(ccadj[min(j, slen -1)] - ccadj[i]);
+  if (cycles_passed >= 44)
+    return;
+  assem_debug("; save gteBusyCycle\n");
+  host_tempreg_acquire();
+#if 0
+  emit_readword(&last_count, HOST_TEMPREG);
+  emit_add(HOST_TEMPREG, HOST_CCREG, HOST_TEMPREG);
+  emit_addimm(HOST_TEMPREG, CLOCK_ADJUST(ccadj[i]), HOST_TEMPREG);
+  emit_addimm(HOST_TEMPREG, gte_cycletab[op]), HOST_TEMPREG);
+  emit_writeword(HOST_TEMPREG, &psxRegs.gteBusyCycle);
+#else
+  emit_addimm(HOST_CCREG, CLOCK_ADJUST(ccadj[i]) + gte_cycletab[op], HOST_TEMPREG);
+  emit_writeword(HOST_TEMPREG, &psxRegs.gteBusyCycle);
+#endif
+  host_tempreg_release();
+}
+
+static void cop2_get_dreg(u_int copr,signed char tl,signed char temp)
+{
+  switch (copr) {
+    case 1:
+    case 3:
+    case 5:
+    case 8:
+    case 9:
+    case 10:
+    case 11:
+      emit_readword(&reg_cop2d[copr],tl);
+      emit_signextend16(tl,tl);
+      emit_writeword(tl,&reg_cop2d[copr]); // hmh
+      break;
+    case 7:
+    case 16:
+    case 17:
+    case 18:
+    case 19:
+      emit_readword(&reg_cop2d[copr],tl);
+      emit_andimm(tl,0xffff,tl);
+      emit_writeword(tl,&reg_cop2d[copr]);
+      break;
+    case 15:
+      emit_readword(&reg_cop2d[14],tl); // SXY2
+      emit_writeword(tl,&reg_cop2d[copr]);
+      break;
+    case 28:
+    case 29:
+      c2op_mfc2_29_assemble(tl,temp);
+      break;
+    default:
+      emit_readword(&reg_cop2d[copr],tl);
+      break;
+  }
+}
+
+static void cop2_put_dreg(u_int copr,signed char sl,signed char temp)
 {
-  cop1_unusable(i, i_regs);
+  switch (copr) {
+    case 15:
+      emit_readword(&reg_cop2d[13],temp);  // SXY1
+      emit_writeword(sl,&reg_cop2d[copr]);
+      emit_writeword(temp,&reg_cop2d[12]); // SXY0
+      emit_readword(&reg_cop2d[14],temp);  // SXY2
+      emit_writeword(sl,&reg_cop2d[14]);
+      emit_writeword(temp,&reg_cop2d[13]); // SXY1
+      break;
+    case 28:
+      emit_andimm(sl,0x001f,temp);
+      emit_shlimm(temp,7,temp);
+      emit_writeword(temp,&reg_cop2d[9]);
+      emit_andimm(sl,0x03e0,temp);
+      emit_shlimm(temp,2,temp);
+      emit_writeword(temp,&reg_cop2d[10]);
+      emit_andimm(sl,0x7c00,temp);
+      emit_shrimm(temp,3,temp);
+      emit_writeword(temp,&reg_cop2d[11]);
+      emit_writeword(sl,&reg_cop2d[28]);
+      break;
+    case 30:
+      emit_xorsar_imm(sl,sl,31,temp);
+#if defined(HAVE_ARMV5) || defined(__aarch64__)
+      emit_clz(temp,temp);
+#else
+      emit_movs(temp,HOST_TEMPREG);
+      emit_movimm(0,temp);
+      emit_jeq((int)out+4*4);
+      emit_addpl_imm(temp,1,temp);
+      emit_lslpls_imm(HOST_TEMPREG,1,HOST_TEMPREG);
+      emit_jns((int)out-2*4);
+#endif
+      emit_writeword(sl,&reg_cop2d[30]);
+      emit_writeword(temp,&reg_cop2d[31]);
+      break;
+    case 31:
+      break;
+    default:
+      emit_writeword(sl,&reg_cop2d[copr]);
+      break;
+  }
 }
 
-void c2ls_assemble(int i,struct regstat *i_regs)
+static void c2ls_assemble(int i, const struct regstat *i_regs)
 {
   int s,tl;
   int ar;
   int offset;
   int memtarget=0,c=0;
-  int jaddr2=0,jaddr3,type;
+  void *jaddr2=NULL;
+  enum stub_type type;
   int agr=AGEN1+(i&1);
-  int fastio_reg_override=0;
-  u_int hr,reglist=0;
+  int fastio_reg_override=-1;
+  u_int reglist=get_host_reglist(i_regs->regmap);
   u_int copr=(source[i]>>16)&0x1f;
   s=get_reg(i_regs->regmap,rs1[i]);
   tl=get_reg(i_regs->regmap,FTEMP);
@@ -3233,9 +3516,6 @@ void c2ls_assemble(int i,struct regstat *i_regs)
   assert(rs1[i]>0);
   assert(tl>=0);
 
-  for(hr=0;hr<HOST_REGS;hr++) {
-    if(i_regs->regmap[hr]>=0) reglist|=1<<hr;
-  }
   if(i_regs->regmap[HOST_CCREG]==CCREG)
     reglist&=~(1<<HOST_CCREG);
 
@@ -3253,14 +3533,15 @@ void c2ls_assemble(int i,struct regstat *i_regs)
   assert(ar>=0);
 
   if (opcode[i]==0x3a) { // SWC2
-    cop2_get_dreg(copr,tl,HOST_TEMPREG);
+    cop2_call_stall_check(0, i, i_regs, reglist_exclude(reglist, tl, -1));
+    cop2_get_dreg(copr,tl,-1);
     type=STOREW_STUB;
   }
   else
     type=LOADW_STUB;
 
   if(c&&!memtarget) {
-    jaddr2=(int)out;
+    jaddr2=out;
     emit_jmp(0); // inline_readstub/inline_writestub?
   }
   else {
@@ -3268,16 +3549,13 @@ void c2ls_assemble(int i,struct regstat *i_regs)
       jaddr2=emit_fastpath_cmp_jump(i,ar,&fastio_reg_override);
     }
     else if(ram_offset&&memtarget) {
+      host_tempreg_acquire();
       emit_addimm(ar,ram_offset,HOST_TEMPREG);
       fastio_reg_override=HOST_TEMPREG;
     }
     if (opcode[i]==0x32) { // LWC2
-      #ifdef HOST_IMM_ADDR32
-      if(c) emit_readword_tlb(constmap[i][s]+offset,-1,tl);
-      else
-      #endif
       int a=ar;
-      if(fastio_reg_override) a=fastio_reg_override;
+      if(fastio_reg_override>=0) a=fastio_reg_override;
       emit_readword_indexed(0,a,tl);
     }
     if (opcode[i]==0x3a) { // SWC2
@@ -3285,112 +3563,331 @@ void c2ls_assemble(int i,struct regstat *i_regs)
       if(!offset&&!c&&s>=0) emit_mov(s,ar);
       #endif
       int a=ar;
-      if(fastio_reg_override) a=fastio_reg_override;
+      if(fastio_reg_override>=0) a=fastio_reg_override;
       emit_writeword_indexed(tl,0,a);
     }
   }
+  if(fastio_reg_override==HOST_TEMPREG)
+    host_tempreg_release();
   if(jaddr2)
-    add_stub(type,jaddr2,(int)out,i,ar,(int)i_regs,ccadj[i],reglist);
+    add_stub_r(type,jaddr2,out,i,ar,i_regs,ccadj[i],reglist);
   if(opcode[i]==0x3a) // SWC2
-  if(!(i_regs->waswritten&(1<<rs1[i]))&&!(new_dynarec_hacks&NDHACK_NO_SMC_CHECK)) {
+  if(!(i_regs->waswritten&(1<<rs1[i])) && !HACK_ENABLED(NDHACK_NO_SMC_CHECK)) {
 #if defined(HOST_IMM8)
     int ir=get_reg(i_regs->regmap,INVCP);
     assert(ir>=0);
     emit_cmpmem_indexedsr12_reg(ir,ar,1);
 #else
-    emit_cmpmem_indexedsr12_imm((int)invalid_code,ar,1);
+    emit_cmpmem_indexedsr12_imm(invalid_code,ar,1);
 #endif
     #if defined(HAVE_CONDITIONAL_CALL) && !defined(DESTRUCTIVE_SHIFT)
     emit_callne(invalidate_addr_reg[ar]);
     #else
-    jaddr3=(int)out;
+    void *jaddr3 = out;
     emit_jne(0);
-    add_stub(INVCODE_STUB,jaddr3,(int)out,reglist|(1<<HOST_CCREG),ar,0,0,0);
+    add_stub(INVCODE_STUB,jaddr3,out,reglist|(1<<HOST_CCREG),ar,0,0,0);
     #endif
   }
   if (opcode[i]==0x32) { // LWC2
+    host_tempreg_acquire();
     cop2_put_dreg(copr,tl,HOST_TEMPREG);
+    host_tempreg_release();
+  }
+}
+
+static void cop2_assemble(int i, const struct regstat *i_regs)
+{
+  u_int copr = (source[i]>>11) & 0x1f;
+  signed char temp = get_reg(i_regs->regmap, -1);
+
+  if (opcode2[i] == 0 || opcode2[i] == 2) { // MFC2/CFC2
+    if (!HACK_ENABLED(NDHACK_GTE_NO_STALL)) {
+      signed char tl = get_reg(i_regs->regmap, rt1[i]);
+      u_int reglist = reglist_exclude(get_host_reglist(i_regs->regmap), tl, temp);
+      cop2_call_stall_check(0, i, i_regs, reglist);
+    }
+  }
+  if (opcode2[i]==0) { // MFC2
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    if(tl>=0&&rt1[i]!=0)
+      cop2_get_dreg(copr,tl,temp);
+  }
+  else if (opcode2[i]==4) { // MTC2
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    cop2_put_dreg(copr,sl,temp);
   }
+  else if (opcode2[i]==2) // CFC2
+  {
+    signed char tl=get_reg(i_regs->regmap,rt1[i]);
+    if(tl>=0&&rt1[i]!=0)
+      emit_readword(&reg_cop2c[copr],tl);
+  }
+  else if (opcode2[i]==6) // CTC2
+  {
+    signed char sl=get_reg(i_regs->regmap,rs1[i]);
+    switch(copr) {
+      case 4:
+      case 12:
+      case 20:
+      case 26:
+      case 27:
+      case 29:
+      case 30:
+        emit_signextend16(sl,temp);
+        break;
+      case 31:
+        c2op_ctc2_31_assemble(sl,temp);
+        break;
+      default:
+        temp=sl;
+        break;
+    }
+    emit_writeword(temp,&reg_cop2c[copr]);
+    assert(sl>=0);
+  }
+}
+
+static void do_unalignedwritestub(int n)
+{
+  assem_debug("do_unalignedwritestub %x\n",start+stubs[n].a*4);
+  literal_pool(256);
+  set_jump_target(stubs[n].addr, out);
+
+  int i=stubs[n].a;
+  struct regstat *i_regs=(struct regstat *)stubs[n].c;
+  int addr=stubs[n].b;
+  u_int reglist=stubs[n].e;
+  signed char *i_regmap=i_regs->regmap;
+  int temp2=get_reg(i_regmap,FTEMP);
+  int rt;
+  rt=get_reg(i_regmap,rs2[i]);
+  assert(rt>=0);
+  assert(addr>=0);
+  assert(opcode[i]==0x2a||opcode[i]==0x2e); // SWL/SWR only implemented
+  reglist|=(1<<addr);
+  reglist&=~(1<<temp2);
+
+#if 1
+  // don't bother with it and call write handler
+  save_regs(reglist);
+  pass_args(addr,rt);
+  int cc=get_reg(i_regmap,CCREG);
+  if(cc<0)
+    emit_loadreg(CCREG,2);
+  emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d+1),2);
+  emit_far_call((opcode[i]==0x2a?jump_handle_swl:jump_handle_swr));
+  emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d+1),cc<0?2:cc);
+  if(cc<0)
+    emit_storereg(CCREG,2);
+  restore_regs(reglist);
+  emit_jmp(stubs[n].retaddr); // return address
+#else
+  emit_andimm(addr,0xfffffffc,temp2);
+  emit_writeword(temp2,&address);
+
+  save_regs(reglist);
+  emit_shrimm(addr,16,1);
+  int cc=get_reg(i_regmap,CCREG);
+  if(cc<0) {
+    emit_loadreg(CCREG,2);
+  }
+  emit_movimm((u_int)readmem,0);
+  emit_addimm(cc<0?2:cc,2*stubs[n].d+2,2);
+  emit_call((int)&indirect_jump_indexed);
+  restore_regs(reglist);
+
+  emit_readword(&readmem_dword,temp2);
+  int temp=addr; //hmh
+  emit_shlimm(addr,3,temp);
+  emit_andimm(temp,24,temp);
+  if (opcode[i]==0x2a) // SWL
+    emit_xorimm(temp,24,temp);
+  emit_movimm(-1,HOST_TEMPREG);
+  if (opcode[i]==0x2a) { // SWL
+    emit_bic_lsr(temp2,HOST_TEMPREG,temp,temp2);
+    emit_orrshr(rt,temp,temp2);
+  }else{
+    emit_bic_lsl(temp2,HOST_TEMPREG,temp,temp2);
+    emit_orrshl(rt,temp,temp2);
+  }
+  emit_readword(&address,addr);
+  emit_writeword(temp2,&word);
+  //save_regs(reglist); // don't need to, no state changes
+  emit_shrimm(addr,16,1);
+  emit_movimm((u_int)writemem,0);
+  //emit_call((int)&indirect_jump_indexed);
+  emit_mov(15,14);
+  emit_readword_dualindexedx4(0,1,15);
+  emit_readword(&Count,HOST_TEMPREG);
+  emit_readword(&next_interupt,2);
+  emit_addimm(HOST_TEMPREG,-2*stubs[n].d-2,HOST_TEMPREG);
+  emit_writeword(2,&last_count);
+  emit_sub(HOST_TEMPREG,2,cc<0?HOST_TEMPREG:cc);
+  if(cc<0) {
+    emit_storereg(CCREG,HOST_TEMPREG);
+  }
+  restore_regs(reglist);
+  emit_jmp(stubs[n].retaddr); // return address
+#endif
 }
 
 #ifndef multdiv_assemble
 void multdiv_assemble(int i,struct regstat *i_regs)
 {
   printf("Need multdiv_assemble for this architecture.\n");
-  exit(1);
+  abort();
 }
 #endif
 
-void mov_assemble(int i,struct regstat *i_regs)
+static void mov_assemble(int i,struct regstat *i_regs)
 {
   //if(opcode2[i]==0x10||opcode2[i]==0x12) { // MFHI/MFLO
   //if(opcode2[i]==0x11||opcode2[i]==0x13) { // MTHI/MTLO
   if(rt1[i]) {
-    signed char sh,sl,th,tl;
-    th=get_reg(i_regs->regmap,rt1[i]|64);
+    signed char sl,tl;
     tl=get_reg(i_regs->regmap,rt1[i]);
     //assert(tl>=0);
     if(tl>=0) {
-      sh=get_reg(i_regs->regmap,rs1[i]|64);
       sl=get_reg(i_regs->regmap,rs1[i]);
       if(sl>=0) emit_mov(sl,tl);
       else emit_loadreg(rs1[i],tl);
-      if(th>=0) {
-        if(sh>=0) emit_mov(sh,th);
-        else emit_loadreg(rs1[i]|64,th);
-      }
     }
   }
 }
 
-#ifndef fconv_assemble
-void fconv_assemble(int i,struct regstat *i_regs)
+// call interpreter, exception handler, things that change pc/regs/cycles ...
+static void call_c_cpu_handler(int i, const struct regstat *i_regs, u_int pc, void *func)
 {
-  printf("Need fconv_assemble for this architecture.\n");
-  exit(1);
+  signed char ccreg=get_reg(i_regs->regmap,CCREG);
+  assert(ccreg==HOST_CCREG);
+  assert(!is_delayslot);
+  (void)ccreg;
+
+  emit_movimm(pc,3); // Get PC
+  emit_readword(&last_count,2);
+  emit_writeword(3,&psxRegs.pc);
+  emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // XXX
+  emit_add(2,HOST_CCREG,2);
+  emit_writeword(2,&psxRegs.cycle);
+  emit_far_call(func);
+  emit_far_jump(jump_to_new_pc);
 }
-#endif
 
-#if 0
-void float_assemble(int i,struct regstat *i_regs)
+static void syscall_assemble(int i,struct regstat *i_regs)
 {
-  printf("Need float_assemble for this architecture.\n");
-  exit(1);
+  emit_movimm(0x20,0); // cause code
+  emit_movimm(0,1);    // not in delay slot
+  call_c_cpu_handler(i,i_regs,start+i*4,psxException);
 }
-#endif
 
-void syscall_assemble(int i,struct regstat *i_regs)
+static void hlecall_assemble(int i,struct regstat *i_regs)
 {
-  signed char ccreg=get_reg(i_regs->regmap,CCREG);
-  assert(ccreg==HOST_CCREG);
-  assert(!is_delayslot);
-  emit_movimm(start+i*4,EAX); // Get PC
-  emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // CHECK: is this right?  There should probably be an extra cycle...
-  emit_jmp((int)jump_syscall_hle); // XXX
+  void *hlefunc = psxNULL;
+  uint32_t hleCode = source[i] & 0x03ffffff;
+  if (hleCode < ARRAY_SIZE(psxHLEt))
+    hlefunc = psxHLEt[hleCode];
+
+  call_c_cpu_handler(i,i_regs,start+i*4+4,hlefunc);
 }
 
-void hlecall_assemble(int i,struct regstat *i_regs)
+static void intcall_assemble(int i,struct regstat *i_regs)
 {
-  signed char ccreg=get_reg(i_regs->regmap,CCREG);
-  assert(ccreg==HOST_CCREG);
-  assert(!is_delayslot);
-  emit_movimm(start+i*4+4,0); // Get PC
-  emit_movimm((int)psxHLEt[source[i]&7],1);
-  emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG); // XXX
-  emit_jmp((int)jump_hlecall);
+  call_c_cpu_handler(i,i_regs,start+i*4,execI);
 }
 
-void intcall_assemble(int i,struct regstat *i_regs)
+static void speculate_mov(int rs,int rt)
 {
-  signed char ccreg=get_reg(i_regs->regmap,CCREG);
-  assert(ccreg==HOST_CCREG);
-  assert(!is_delayslot);
-  emit_movimm(start+i*4,0); // Get PC
-  emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]),HOST_CCREG);
-  emit_jmp((int)jump_intcall);
+  if(rt!=0) {
+    smrv_strong_next|=1<<rt;
+    smrv[rt]=smrv[rs];
+  }
+}
+
+static void speculate_mov_weak(int rs,int rt)
+{
+  if(rt!=0) {
+    smrv_weak_next|=1<<rt;
+    smrv[rt]=smrv[rs];
+  }
+}
+
+static void speculate_register_values(int i)
+{
+  if(i==0) {
+    memcpy(smrv,psxRegs.GPR.r,sizeof(smrv));
+    // gp,sp are likely to stay the same throughout the block
+    smrv_strong_next=(1<<28)|(1<<29)|(1<<30);
+    smrv_weak_next=~smrv_strong_next;
+    //printf(" llr %08x\n", smrv[4]);
+  }
+  smrv_strong=smrv_strong_next;
+  smrv_weak=smrv_weak_next;
+  switch(itype[i]) {
+    case ALU:
+      if     ((smrv_strong>>rs1[i])&1) speculate_mov(rs1[i],rt1[i]);
+      else if((smrv_strong>>rs2[i])&1) speculate_mov(rs2[i],rt1[i]);
+      else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]);
+      else if((smrv_weak>>rs2[i])&1) speculate_mov_weak(rs2[i],rt1[i]);
+      else {
+        smrv_strong_next&=~(1<<rt1[i]);
+        smrv_weak_next&=~(1<<rt1[i]);
+      }
+      break;
+    case SHIFTIMM:
+      smrv_strong_next&=~(1<<rt1[i]);
+      smrv_weak_next&=~(1<<rt1[i]);
+      // fallthrough
+    case IMM16:
+      if(rt1[i]&&is_const(&regs[i],rt1[i])) {
+        int value,hr=get_reg(regs[i].regmap,rt1[i]);
+        if(hr>=0) {
+          if(get_final_value(hr,i,&value))
+               smrv[rt1[i]]=value;
+          else smrv[rt1[i]]=constmap[i][hr];
+          smrv_strong_next|=1<<rt1[i];
+        }
+      }
+      else {
+        if     ((smrv_strong>>rs1[i])&1) speculate_mov(rs1[i],rt1[i]);
+        else if((smrv_weak>>rs1[i])&1) speculate_mov_weak(rs1[i],rt1[i]);
+      }
+      break;
+    case LOAD:
+      if(start<0x2000&&(rt1[i]==26||(smrv[rt1[i]]>>24)==0xa0)) {
+        // special case for BIOS
+        smrv[rt1[i]]=0xa0000000;
+        smrv_strong_next|=1<<rt1[i];
+        break;
+      }
+      // fallthrough
+    case SHIFT:
+    case LOADLR:
+    case MOV:
+      smrv_strong_next&=~(1<<rt1[i]);
+      smrv_weak_next&=~(1<<rt1[i]);
+      break;
+    case COP0:
+    case COP2:
+      if(opcode2[i]==0||opcode2[i]==2) { // MFC/CFC
+        smrv_strong_next&=~(1<<rt1[i]);
+        smrv_weak_next&=~(1<<rt1[i]);
+      }
+      break;
+    case C2LS:
+      if (opcode[i]==0x32) { // LWC2
+        smrv_strong_next&=~(1<<rt1[i]);
+        smrv_weak_next&=~(1<<rt1[i]);
+      }
+      break;
+  }
+#if 0
+  int r=4;
+  printf("x %08x %08x %d %d c %08x %08x\n",smrv[r],start+i*4,
+    ((smrv_strong>>r)&1),(smrv_weak>>r)&1,regs[i].isconst,regs[i].wasconst);
+#endif
 }
 
-void ds_assemble(int i,struct regstat *i_regs)
+static void ds_assemble(int i,struct regstat *i_regs)
 {
   speculate_register_values(i);
   is_delayslot=1;
@@ -3423,12 +3920,6 @@ void ds_assemble(int i,struct regstat *i_regs)
       c2ls_assemble(i,i_regs);break;
     case C2OP:
       c2op_assemble(i,i_regs);break;
-    case FCONV:
-      fconv_assemble(i,i_regs);break;
-    case FLOAT:
-      float_assemble(i,i_regs);break;
-    case FCOMP:
-      fcomp_assemble(i,i_regs);break;
     case MULTDIV:
       multdiv_assemble(i,i_regs);break;
     case MOV:
@@ -3441,36 +3932,23 @@ void ds_assemble(int i,struct regstat *i_regs)
     case RJUMP:
     case CJUMP:
     case SJUMP:
-    case FJUMP:
       SysPrintf("Jump in the delay slot.  This is probably a bug.\n");
   }
   is_delayslot=0;
 }
 
 // Is the branch target a valid internal jump?
-int internal_branch(uint64_t i_is32,int addr)
+static int internal_branch(int addr)
 {
   if(addr&1) return 0; // Indirect (register) jump
   if(addr>=start && addr<start+slen*4-4)
   {
-    //int t=(addr-start)>>2;
-    // Delay slots are not valid branch targets
-    //if(t>0&&(itype[t-1]==RJUMP||itype[t-1]==UJUMP||itype[t-1]==CJUMP||itype[t-1]==SJUMP||itype[t-1]==FJUMP)) return 0;
-    // 64 -> 32 bit transition requires a recompile
-    /*if(is32[t]&~unneeded_reg_upper[t]&~i_is32)
-    {
-      if(requires_32bit[t]&~i_is32) printf("optimizable: no\n");
-      else printf("optimizable: yes\n");
-    }*/
-    //if(is32[t]&~unneeded_reg_upper[t]&~i_is32) return 0;
     return 1;
   }
   return 0;
 }
 
-#ifndef wb_invalidate
-void wb_invalidate(signed char pre[],signed char entry[],uint64_t dirty,uint64_t is32,
-  uint64_t u,uint64_t uu)
+static void wb_invalidate(signed char pre[],signed char entry[],uint64_t dirty,uint64_t u)
 {
   int hr;
   for(hr=0;hr<HOST_REGS;hr++) {
@@ -3479,19 +3957,9 @@ void wb_invalidate(signed char pre[],signed char entry[],uint64_t dirty,uint64_t
         if(pre[hr]>=0) {
           if((dirty>>hr)&1) {
             if(get_reg(entry,pre[hr])<0) {
-              if(pre[hr]<64) {
-                if(!((u>>pre[hr])&1)) {
-                  emit_storereg(pre[hr],hr);
-                  if( ((is32>>pre[hr])&1) && !((uu>>pre[hr])&1) ) {
-                    emit_sarimm(hr,31,hr);
-                    emit_storereg(pre[hr]|64,hr);
-                  }
-                }
-              }else{
-                if(!((uu>>(pre[hr]&63))&1) && !((is32>>(pre[hr]&63))&1)) {
-                  emit_storereg(pre[hr],hr);
-                }
-              }
+              assert(pre[hr]<64);
+              if(!((u>>pre[hr])&1))
+                emit_storereg(pre[hr],hr);
             }
           }
         }
@@ -3512,12 +3980,11 @@ void wb_invalidate(signed char pre[],signed char entry[],uint64_t dirty,uint64_t
     }
   }
 }
-#endif
 
 // Load the specified registers
 // This only loads the registers given as arguments because
 // we don't want to load things that will be overwritten
-void load_regs(signed char entry[],signed char regmap[],int is32,int rs1,int rs2)
+static void load_regs(signed char entry[],signed char regmap[],int rs1,int rs2)
 {
   int hr;
   // Load 32-bit regs
@@ -3537,28 +4004,6 @@ void load_regs(signed char entry[],signed char regmap[],int is32,int rs1,int rs2
       }
     }
   }
-  //Load 64-bit regs
-  for(hr=0;hr<HOST_REGS;hr++) {
-    if(hr!=EXCLUDE_REG&&regmap[hr]>=0) {
-      if(entry[hr]!=regmap[hr]) {
-        if(regmap[hr]-64==rs1||regmap[hr]-64==rs2)
-        {
-          assert(regmap[hr]!=64);
-          if((is32>>(regmap[hr]&63))&1) {
-            int lr=get_reg(regmap,regmap[hr]-64);
-            if(lr>=0)
-              emit_sarimm(lr,31,hr);
-            else
-              emit_loadreg(regmap[hr],hr);
-          }
-          else
-          {
-            emit_loadreg(regmap[hr],hr);
-          }
-        }
-      }
-    }
-  }
 }
 
 // Load registers prior to the start of a loop
@@ -3648,9 +4093,6 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[])
             }else if (opcode[i]==0x1a||opcode[i]==0x1b) {
               emit_movimm((constmap[i][rs]+offset)&0xFFFFFFF8,ra); // LDL/LDR
             }else{
-              #ifdef HOST_IMM_ADDR32
-              if((itype[i]!=LOAD&&(opcode[i]&0x3b)!=0x31&&(opcode[i]&0x3b)!=0x32)) // LWC1/LDC1/LWC2/LDC2
-              #endif
               emit_movimm(constmap[i][rs]+offset,ra);
               regs[i].loadedconst|=1<<ra;
             }
@@ -3682,9 +4124,6 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[])
         }else if (opcode[i+1]==0x1a||opcode[i+1]==0x1b) {
           emit_movimm((constmap[i+1][rs]+offset)&0xFFFFFFF8,ra); // LDL/LDR
         }else{
-          #ifdef HOST_IMM_ADDR32
-          if((itype[i+1]!=LOAD&&(opcode[i+1]&0x3b)!=0x31&&(opcode[i+1]&0x3b)!=0x32)) // LWC1/LDC1/LWC2/LDC2
-          #endif
           emit_movimm(constmap[i+1][rs]+offset,ra);
           regs[i+1].loadedconst|=1<<ra;
         }
@@ -3703,7 +4142,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[])
   }
 }
 
-int get_final_value(int hr, int i, int *value)
+static int get_final_value(int hr, int i, int *value)
 {
   int reg=regs[i].regmap[hr];
   while(i<slen-1) {
@@ -3731,23 +4170,20 @@ int get_final_value(int hr, int i, int *value)
       {
         // Precompute load address
         *value=constmap[i][hr]+imm[i+1];
-        //printf("c=%x imm=%x\n",(int)constmap[i][hr],imm[i+1]);
+        //printf("c=%x imm=%lx\n",(long)constmap[i][hr],imm[i+1]);
         return 1;
       }
     }
   }
   *value=constmap[i][hr];
-  //printf("c=%x\n",(int)constmap[i][hr]);
+  //printf("c=%lx\n",(long)constmap[i][hr]);
   if(i==slen-1) return 1;
-  if(reg<64) {
-    return !((unneeded_reg[i+1]>>reg)&1);
-  }else{
-    return !((unneeded_reg_upper[i+1]>>reg)&1);
-  }
+  assert(reg < 64);
+  return !((unneeded_reg[i+1]>>reg)&1);
 }
 
 // Load registers with known constants
-void load_consts(signed char pre[],signed char regmap[],int is32,int i)
+static void load_consts(signed char pre[],signed char regmap[],int i)
 {
   int hr,hr2;
   // propagate loaded constant flags
@@ -3767,7 +4203,8 @@ void load_consts(signed char pre[],signed char regmap[],int is32,int i)
     if(hr!=EXCLUDE_REG&&regmap[hr]>=0) {
       //if(entry[hr]!=regmap[hr]) {
       if(!((regs[i].loadedconst>>hr)&1)) {
-        if(((regs[i].isconst>>hr)&1)&&regmap[hr]<64&&regmap[hr]>0) {
+        assert(regmap[hr]<64);
+        if(((regs[i].isconst>>hr)&1)&&regmap[hr]>0) {
           int value,similar=0;
           if(get_final_value(hr,i,&value)) {
             // see if some other register has similar value
@@ -3798,41 +4235,16 @@ void load_consts(signed char pre[],signed char regmap[],int is32,int i)
       }
     }
   }
-  // Load 64-bit regs
-  for(hr=0;hr<HOST_REGS;hr++) {
-    if(hr!=EXCLUDE_REG&&regmap[hr]>=0) {
-      //if(entry[hr]!=regmap[hr]) {
-      if(i==0||!((regs[i-1].isconst>>hr)&1)||pre[hr]!=regmap[hr]||bt[i]) {
-        if(((regs[i].isconst>>hr)&1)&&regmap[hr]>64) {
-          if((is32>>(regmap[hr]&63))&1) {
-            int lr=get_reg(regmap,regmap[hr]-64);
-            assert(lr>=0);
-            emit_sarimm(lr,31,hr);
-          }
-          else
-          {
-            int value;
-            if(get_final_value(hr,i,&value)) {
-              if(value==0) {
-                emit_zeroreg(hr);
-              }
-              else {
-                emit_movimm(value,hr);
-              }
-            }
-          }
-        }
-      }
-    }
-  }
 }
-void load_all_consts(signed char regmap[],int is32,u_int dirty,int i)
+
+void load_all_consts(signed char regmap[], u_int dirty, int i)
 {
   int hr;
   // Load 32-bit regs
   for(hr=0;hr<HOST_REGS;hr++) {
     if(hr!=EXCLUDE_REG&&regmap[hr]>=0&&((dirty>>hr)&1)) {
-      if(((regs[i].isconst>>hr)&1)&&regmap[hr]<64&&regmap[hr]>0) {
+      assert(regmap[hr] < 64);
+      if(((regs[i].isconst>>hr)&1)&&regmap[hr]>0) {
         int value=constmap[i][hr];
         if(value==0) {
           emit_zeroreg(hr);
@@ -3843,32 +4255,10 @@ void load_all_consts(signed char regmap[],int is32,u_int dirty,int i)
       }
     }
   }
-  // Load 64-bit regs
-  for(hr=0;hr<HOST_REGS;hr++) {
-    if(hr!=EXCLUDE_REG&&regmap[hr]>=0&&((dirty>>hr)&1)) {
-      if(((regs[i].isconst>>hr)&1)&&regmap[hr]>64) {
-        if((is32>>(regmap[hr]&63))&1) {
-          int lr=get_reg(regmap,regmap[hr]-64);
-          assert(lr>=0);
-          emit_sarimm(lr,31,hr);
-        }
-        else
-        {
-          int value=constmap[i][hr];
-          if(value==0) {
-            emit_zeroreg(hr);
-          }
-          else {
-            emit_movimm(value,hr);
-          }
-        }
-      }
-    }
-  }
 }
 
 // Write out all dirty registers (except cycle count)
-void wb_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty)
+static void wb_dirtys(signed char i_regmap[],uint64_t i_dirty)
 {
   int hr;
   for(hr=0;hr<HOST_REGS;hr++) {
@@ -3876,22 +4266,18 @@ void wb_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty)
       if(i_regmap[hr]>0) {
         if(i_regmap[hr]!=CCREG) {
           if((i_dirty>>hr)&1) {
-            if(i_regmap[hr]<64) {
-              emit_storereg(i_regmap[hr],hr);
-            }else{
-              if( !((i_is32>>(i_regmap[hr]&63))&1) ) {
-                emit_storereg(i_regmap[hr],hr);
-              }
-            }
+            assert(i_regmap[hr]<64);
+            emit_storereg(i_regmap[hr],hr);
           }
         }
       }
     }
   }
 }
+
 // Write out dirty registers that we need to reload (pair with load_needed_regs)
 // This writes the registers not written by store_regs_bt
-void wb_needed_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+void wb_needed_dirtys(signed char i_regmap[],uint64_t i_dirty,int addr)
 {
   int hr;
   int t=(addr-start)>>2;
@@ -3899,15 +4285,10 @@ void wb_needed_dirtys(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,in
     if(hr!=EXCLUDE_REG) {
       if(i_regmap[hr]>0) {
         if(i_regmap[hr]!=CCREG) {
-          if(i_regmap[hr]==regs[t].regmap_entry[hr] && ((regs[t].dirty>>hr)&1) && !(((i_is32&~regs[t].was32&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)) {
+          if(i_regmap[hr]==regs[t].regmap_entry[hr] && ((regs[t].dirty>>hr)&1)) {
             if((i_dirty>>hr)&1) {
-              if(i_regmap[hr]<64) {
-                emit_storereg(i_regmap[hr],hr);
-              }else{
-                if( !((i_is32>>(i_regmap[hr]&63))&1) ) {
-                  emit_storereg(i_regmap[hr],hr);
-                }
-              }
+              assert(i_regmap[hr]<64);
+              emit_storereg(i_regmap[hr],hr);
             }
           }
         }
@@ -3975,58 +4356,23 @@ void load_regs_entry(int t)
       }
     }
   }
-  // Load 64-bit regs
-  for(hr=0;hr<HOST_REGS;hr++) {
-    if(regs[t].regmap_entry[hr]>=64&&regs[t].regmap_entry[hr]<TEMPREG+64) {
-      assert(regs[t].regmap_entry[hr]!=64);
-      if((regs[t].was32>>(regs[t].regmap_entry[hr]&63))&1) {
-        int lr=get_reg(regs[t].regmap_entry,regs[t].regmap_entry[hr]-64);
-        if(lr<0) {
-          emit_loadreg(regs[t].regmap_entry[hr],hr);
-        }
-        else
-        {
-          emit_sarimm(lr,31,hr);
-        }
-      }
-      else
-      {
-        emit_loadreg(regs[t].regmap_entry[hr],hr);
-      }
-    }
-  }
 }
 
 // Store dirty registers prior to branch
-void store_regs_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+void store_regs_bt(signed char i_regmap[],uint64_t i_dirty,int addr)
 {
-  if(internal_branch(i_is32,addr))
+  if(internal_branch(addr))
   {
     int t=(addr-start)>>2;
     int hr;
     for(hr=0;hr<HOST_REGS;hr++) {
       if(hr!=EXCLUDE_REG) {
         if(i_regmap[hr]>0 && i_regmap[hr]!=CCREG) {
-          if(i_regmap[hr]!=regs[t].regmap_entry[hr] || !((regs[t].dirty>>hr)&1) || (((i_is32&~regs[t].was32&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)) {
+          if(i_regmap[hr]!=regs[t].regmap_entry[hr] || !((regs[t].dirty>>hr)&1)) {
             if((i_dirty>>hr)&1) {
-              if(i_regmap[hr]<64) {
-                if(!((unneeded_reg[t]>>i_regmap[hr])&1)) {
-                  emit_storereg(i_regmap[hr],hr);
-                  if( ((i_is32>>i_regmap[hr])&1) && !((unneeded_reg_upper[t]>>i_regmap[hr])&1) ) {
-                    #ifdef DESTRUCTIVE_WRITEBACK
-                    emit_sarimm(hr,31,hr);
-                    emit_storereg(i_regmap[hr]|64,hr);
-                    #else
-                    emit_sarimm(hr,31,HOST_TEMPREG);
-                    emit_storereg(i_regmap[hr]|64,HOST_TEMPREG);
-                    #endif
-                  }
-                }
-              }else{
-                if( !((i_is32>>(i_regmap[hr]&63))&1) && !((unneeded_reg_upper[t]>>(i_regmap[hr]&63))&1) ) {
-                  emit_storereg(i_regmap[hr],hr);
-                }
-              }
+              assert(i_regmap[hr]<64);
+              if(!((unneeded_reg[t]>>i_regmap[hr])&1))
+                emit_storereg(i_regmap[hr],hr);
             }
           }
         }
@@ -4036,15 +4382,15 @@ void store_regs_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int a
   else
   {
     // Branch out of this block, write out all dirty regs
-    wb_dirtys(i_regmap,i_is32,i_dirty);
+    wb_dirtys(i_regmap,i_dirty);
   }
 }
 
 // Load all needed registers for branch target
-void load_regs_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+static void load_regs_bt(signed char i_regmap[],uint64_t i_dirty,int addr)
 {
   //if(addr>=start && addr<(start+slen*4))
-  if(internal_branch(i_is32,addr))
+  if(internal_branch(addr))
   {
     int t=(addr-start)>>2;
     int hr;
@@ -4058,11 +4404,7 @@ void load_regs_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int ad
     // Load 32-bit regs
     for(hr=0;hr<HOST_REGS;hr++) {
       if(hr!=EXCLUDE_REG&&regs[t].regmap_entry[hr]>=0&&regs[t].regmap_entry[hr]<TEMPREG) {
-        #ifdef DESTRUCTIVE_WRITEBACK
-        if(i_regmap[hr]!=regs[t].regmap_entry[hr] || ( !((regs[t].dirty>>hr)&1) && ((i_dirty>>hr)&1) && (((i_is32&~unneeded_reg_upper[t])>>i_regmap[hr])&1) ) || (((i_is32&~regs[t].was32&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)) {
-        #else
-        if(i_regmap[hr]!=regs[t].regmap_entry[hr] ) {
-        #endif
+        if(i_regmap[hr]!=regs[t].regmap_entry[hr]) {
           if(regs[t].regmap_entry[hr]==0) {
             emit_zeroreg(hr);
           }
@@ -4073,37 +4415,10 @@ void load_regs_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int ad
         }
       }
     }
-    //Load 64-bit regs
-    for(hr=0;hr<HOST_REGS;hr++) {
-      if(hr!=EXCLUDE_REG&&regs[t].regmap_entry[hr]>=64&&regs[t].regmap_entry[hr]<TEMPREG+64) {
-        if(i_regmap[hr]!=regs[t].regmap_entry[hr]) {
-          assert(regs[t].regmap_entry[hr]!=64);
-          if((i_is32>>(regs[t].regmap_entry[hr]&63))&1) {
-            int lr=get_reg(regs[t].regmap_entry,regs[t].regmap_entry[hr]-64);
-            if(lr<0) {
-              emit_loadreg(regs[t].regmap_entry[hr],hr);
-            }
-            else
-            {
-              emit_sarimm(lr,31,hr);
-            }
-          }
-          else
-          {
-            emit_loadreg(regs[t].regmap_entry[hr],hr);
-          }
-        }
-        else if((i_is32>>(regs[t].regmap_entry[hr]&63))&1) {
-          int lr=get_reg(regs[t].regmap_entry,regs[t].regmap_entry[hr]-64);
-          assert(lr>=0);
-          emit_sarimm(lr,31,hr);
-        }
-      }
-    }
   }
 }
 
-int match_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
+static int match_bt(signed char i_regmap[],uint64_t i_dirty,int addr)
 {
   if(addr>=start && addr<start+slen*4-4)
   {
@@ -4130,8 +4445,7 @@ int match_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
             }
             else if(i_regmap[hr]>=64&&i_regmap[hr]<TEMPREG+64)
             {
-              if(!((unneeded_reg_upper[t]>>(i_regmap[hr]&63))&1))
-                return 0;
+              assert(0);
             }
           }
         }
@@ -4149,17 +4463,11 @@ int match_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
               }
             }
           }
-          if((((regs[t].was32^i_is32)&~unneeded_reg_upper[t])>>(i_regmap[hr]&63))&1)
-          {
-            //printf("%x: is32 no match\n",addr);
-            return 0;
-          }
         }
       }
     }
-    //if(is32[t]&~unneeded_reg_upper[t]&~i_is32) return 0;
     // Delay slots are not valid branch targets
-    //if(t>0&&(itype[t-1]==RJUMP||itype[t-1]==UJUMP||itype[t-1]==CJUMP||itype[t-1]==SJUMP||itype[t-1]==FJUMP)) return 0;
+    //if(t>0&&(itype[t-1]==RJUMP||itype[t-1]==UJUMP||itype[t-1]==CJUMP||itype[t-1]==SJUMP)) return 0;
     // Delay slots require additional processing, so do not match
     if(is_ds[t]) return 0;
   }
@@ -4186,20 +4494,58 @@ int match_bt(signed char i_regmap[],uint64_t i_is32,uint64_t i_dirty,int addr)
   return 1;
 }
 
+#ifdef DRC_DBG
+static void drc_dbg_emit_do_cmp(int i)
+{
+  extern void do_insn_cmp();
+  //extern int cycle;
+  u_int hr, reglist = get_host_reglist(regs[i].regmap);
+
+  assem_debug("//do_insn_cmp %08x\n", start+i*4);
+  save_regs(reglist);
+  // write out changed consts to match the interpreter
+  if (i > 0 && !bt[i]) {
+    for (hr = 0; hr < HOST_REGS; hr++) {
+      int reg = regs[i-1].regmap[hr];
+      if (hr == EXCLUDE_REG || reg < 0)
+        continue;
+      if (!((regs[i-1].isconst >> hr) & 1))
+        continue;
+      if (i > 1 && reg == regs[i-2].regmap[hr] && constmap[i-1][hr] == constmap[i-2][hr])
+        continue;
+      emit_movimm(constmap[i-1][hr],0);
+      emit_storereg(reg, 0);
+    }
+  }
+  emit_movimm(start+i*4,0);
+  emit_writeword(0,&pcaddr);
+  emit_far_call(do_insn_cmp);
+  //emit_readword(&cycle,0);
+  //emit_addimm(0,2,0);
+  //emit_writeword(0,&cycle);
+  (void)get_reg2;
+  restore_regs(reglist);
+  assem_debug("\\\\do_insn_cmp\n");
+}
+#else
+#define drc_dbg_emit_do_cmp(x)
+#endif
+
 // Used when a branch jumps into the delay slot of another branch
-void ds_assemble_entry(int i)
+static void ds_assemble_entry(int i)
 {
   int t=(ba[i]-start)>>2;
-  if(!instr_addr[t]) instr_addr[t]=(u_int)out;
+  if (!instr_addr[t])
+    instr_addr[t] = out;
   assem_debug("Assemble delay slot at %x\n",ba[i]);
   assem_debug("<->\n");
+  drc_dbg_emit_do_cmp(t);
   if(regs[t].regmap_entry[HOST_CCREG]==CCREG&&regs[t].regmap[HOST_CCREG]!=CCREG)
-    wb_register(CCREG,regs[t].regmap_entry,regs[t].wasdirty,regs[t].was32);
-  load_regs(regs[t].regmap_entry,regs[t].regmap,regs[t].was32,rs1[t],rs2[t]);
+    wb_register(CCREG,regs[t].regmap_entry,regs[t].wasdirty);
+  load_regs(regs[t].regmap_entry,regs[t].regmap,rs1[t],rs2[t]);
   address_generation(t,&regs[t],regs[t].regmap_entry);
   if(itype[t]==STORE||itype[t]==STORELR||(opcode[t]&0x3b)==0x39||(opcode[t]&0x3b)==0x3a)
-    load_regs(regs[t].regmap_entry,regs[t].regmap,regs[t].was32,INVCP,INVCP);
-  cop1_usable=0;
+    load_regs(regs[t].regmap_entry,regs[t].regmap,INVCP,INVCP);
   is_delayslot=0;
   switch(itype[t]) {
     case ALU:
@@ -4230,12 +4576,6 @@ void ds_assemble_entry(int i)
       c2ls_assemble(t,&regs[t]);break;
     case C2OP:
       c2op_assemble(t,&regs[t]);break;
-    case FCONV:
-      fconv_assemble(t,&regs[t]);break;
-    case FLOAT:
-      float_assemble(t,&regs[t]);break;
-    case FCOMP:
-      fcomp_assemble(t,&regs[t]);break;
     case MULTDIV:
       multdiv_assemble(t,&regs[t]);break;
     case MOV:
@@ -4248,32 +4588,48 @@ void ds_assemble_entry(int i)
     case RJUMP:
     case CJUMP:
     case SJUMP:
-    case FJUMP:
       SysPrintf("Jump in the delay slot.  This is probably a bug.\n");
   }
-  store_regs_bt(regs[t].regmap,regs[t].is32,regs[t].dirty,ba[i]+4);
-  load_regs_bt(regs[t].regmap,regs[t].is32,regs[t].dirty,ba[i]+4);
-  if(internal_branch(regs[t].is32,ba[i]+4))
+  store_regs_bt(regs[t].regmap,regs[t].dirty,ba[i]+4);
+  load_regs_bt(regs[t].regmap,regs[t].dirty,ba[i]+4);
+  if(internal_branch(ba[i]+4))
     assem_debug("branch: internal\n");
   else
     assem_debug("branch: external\n");
-  assert(internal_branch(regs[t].is32,ba[i]+4));
-  add_to_linker((int)out,ba[i]+4,internal_branch(regs[t].is32,ba[i]+4));
+  assert(internal_branch(ba[i]+4));
+  add_to_linker(out,ba[i]+4,internal_branch(ba[i]+4));
   emit_jmp(0);
 }
 
+static void emit_extjump(void *addr, u_int target)
+{
+  emit_extjump2(addr, target, dyna_linker);
+}
+
+static void emit_extjump_ds(void *addr, u_int target)
+{
+  emit_extjump2(addr, target, dyna_linker_ds);
+}
+
+// Load 2 immediates optimizing for small code size
+static void emit_mov2imm_compact(int imm1,u_int rt1,int imm2,u_int rt2)
+{
+  emit_movimm(imm1,rt1);
+  emit_movimm_from(imm1,rt1,imm2,rt2);
+}
+
 void do_cc(int i,signed char i_regmap[],int *adj,int addr,int taken,int invert)
 {
   int count;
-  int jaddr;
-  int idle=0;
+  void *jaddr;
+  void *idle=NULL;
   int t=0;
   if(itype[i]==RJUMP)
   {
     *adj=0;
   }
   //if(ba[i]>=start && ba[i]<(start+slen*4))
-  if(internal_branch(branch_regs[i].is32,ba[i]))
+  if(internal_branch(ba[i]))
   {
     t=(ba[i]-start)>>2;
     if(is_ds[t]) *adj=-1; // Branch into delay slot adds an extra cycle
@@ -4287,89 +4643,86 @@ void do_cc(int i,signed char i_regmap[],int *adj,int addr,int taken,int invert)
   if(taken==TAKEN && i==(ba[i]-start)>>2 && source[i+1]==0) {
     // Idle loop
     if(count&1) emit_addimm_and_set_flags(2*(count+2),HOST_CCREG);
-    idle=(int)out;
+    idle=out;
     //emit_subfrommem(&idlecount,HOST_CCREG); // Count idle cycles
     emit_andimm(HOST_CCREG,3,HOST_CCREG);
-    jaddr=(int)out;
+    jaddr=out;
     emit_jmp(0);
   }
   else if(*adj==0||invert) {
     int cycles=CLOCK_ADJUST(count+2);
     // faster loop HACK
+#if 0
     if (t&&*adj) {
       int rel=t-i;
       if(-NO_CYCLE_PENALTY_THR<rel&&rel<0)
         cycles=CLOCK_ADJUST(*adj)+count+2-*adj;
     }
+#endif
     emit_addimm_and_set_flags(cycles,HOST_CCREG);
-    jaddr=(int)out;
+    jaddr=out;
     emit_jns(0);
   }
   else
   {
     emit_cmpimm(HOST_CCREG,-CLOCK_ADJUST(count+2));
-    jaddr=(int)out;
+    jaddr=out;
     emit_jns(0);
   }
-  add_stub(CC_STUB,jaddr,idle?idle:(int)out,(*adj==0||invert||idle)?0:(count+2),i,addr,taken,0);
+  add_stub(CC_STUB,jaddr,idle?idle:out,(*adj==0||invert||idle)?0:(count+2),i,addr,taken,0);
 }
 
-void do_ccstub(int n)
+static void do_ccstub(int n)
 {
   literal_pool(256);
-  assem_debug("do_ccstub %x\n",start+stubs[n][4]*4);
-  set_jump_target(stubs[n][1],(int)out);
-  int i=stubs[n][4];
-  if(stubs[n][6]==NULLDS) {
+  assem_debug("do_ccstub %x\n",start+(u_int)stubs[n].b*4);
+  set_jump_target(stubs[n].addr, out);
+  int i=stubs[n].b;
+  if(stubs[n].d==NULLDS) {
     // Delay slot instruction is nullified ("likely" branch)
-    wb_dirtys(regs[i].regmap,regs[i].is32,regs[i].dirty);
+    wb_dirtys(regs[i].regmap,regs[i].dirty);
   }
-  else if(stubs[n][6]!=TAKEN) {
-    wb_dirtys(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty);
+  else if(stubs[n].d!=TAKEN) {
+    wb_dirtys(branch_regs[i].regmap,branch_regs[i].dirty);
   }
   else {
-    if(internal_branch(branch_regs[i].is32,ba[i]))
-      wb_needed_dirtys(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+    if(internal_branch(ba[i]))
+      wb_needed_dirtys(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
   }
-  if(stubs[n][5]!=-1)
+  if(stubs[n].c!=-1)
   {
     // Save PC as return address
-    emit_movimm(stubs[n][5],EAX);
-    emit_writeword(EAX,(int)&pcaddr);
+    emit_movimm(stubs[n].c,EAX);
+    emit_writeword(EAX,&pcaddr);
   }
   else
   {
     // Return address depends on which way the branch goes
-    if(itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    if(itype[i]==CJUMP||itype[i]==SJUMP)
     {
       int s1l=get_reg(branch_regs[i].regmap,rs1[i]);
-      int s1h=get_reg(branch_regs[i].regmap,rs1[i]|64);
       int s2l=get_reg(branch_regs[i].regmap,rs2[i]);
-      int s2h=get_reg(branch_regs[i].regmap,rs2[i]|64);
       if(rs1[i]==0)
       {
-        s1l=s2l;s1h=s2h;
-        s2l=s2h=-1;
+        s1l=s2l;
+        s2l=-1;
       }
       else if(rs2[i]==0)
       {
-        s2l=s2h=-1;
-      }
-      if((branch_regs[i].is32>>rs1[i])&(branch_regs[i].is32>>rs2[i])&1) {
-        s1h=s2h=-1;
+        s2l=-1;
       }
       assert(s1l>=0);
       #ifdef DESTRUCTIVE_WRITEBACK
       if(rs1[i]) {
-        if((branch_regs[i].dirty>>s1l)&(branch_regs[i].is32>>rs1[i])&1)
+        if((branch_regs[i].dirty>>s1l)&&1)
           emit_loadreg(rs1[i],s1l);
       }
       else {
-        if((branch_regs[i].dirty>>s1l)&(branch_regs[i].is32>>rs2[i])&1)
+        if((branch_regs[i].dirty>>s1l)&1)
           emit_loadreg(rs2[i],s1l);
       }
       if(s2l>=0)
-        if((branch_regs[i].dirty>>s2l)&(branch_regs[i].is32>>rs2[i])&1)
+        if((branch_regs[i].dirty>>s2l)&1)
           emit_loadreg(rs2[i],s2l);
       #endif
       int hr=0;
@@ -4411,46 +4764,28 @@ void do_ccstub(int n)
       if((opcode[i]&0x2f)==4) // BEQ
       {
         #ifdef HAVE_CMOV_IMM
-        if(s1h<0) {
-          if(s2l>=0) emit_cmp(s1l,s2l);
-          else emit_test(s1l,s1l);
-          emit_cmov2imm_e_ne_compact(ba[i],start+i*4+8,addr);
-        }
-        else
+        if(s2l>=0) emit_cmp(s1l,s2l);
+        else emit_test(s1l,s1l);
+        emit_cmov2imm_e_ne_compact(ba[i],start+i*4+8,addr);
+        #else
+        emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
+        if(s2l>=0) emit_cmp(s1l,s2l);
+        else emit_test(s1l,s1l);
+        emit_cmovne_reg(alt,addr);
         #endif
-        {
-          emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
-          if(s1h>=0) {
-            if(s2h>=0) emit_cmp(s1h,s2h);
-            else emit_test(s1h,s1h);
-            emit_cmovne_reg(alt,addr);
-          }
-          if(s2l>=0) emit_cmp(s1l,s2l);
-          else emit_test(s1l,s1l);
-          emit_cmovne_reg(alt,addr);
-        }
       }
       if((opcode[i]&0x2f)==5) // BNE
       {
         #ifdef HAVE_CMOV_IMM
-        if(s1h<0) {
-          if(s2l>=0) emit_cmp(s1l,s2l);
-          else emit_test(s1l,s1l);
-          emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
-        }
-        else
+        if(s2l>=0) emit_cmp(s1l,s2l);
+        else emit_test(s1l,s1l);
+        emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
+        #else
+        emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
+        if(s2l>=0) emit_cmp(s1l,s2l);
+        else emit_test(s1l,s1l);
+        emit_cmovne_reg(alt,addr);
         #endif
-        {
-          emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
-          if(s1h>=0) {
-            if(s2h>=0) emit_cmp(s1h,s2h);
-            else emit_test(s1h,s1h);
-            emit_cmovne_reg(alt,addr);
-          }
-          if(s2l>=0) emit_cmp(s1l,s2l);
-          else emit_test(s1l,s1l);
-          emit_cmovne_reg(alt,addr);
-        }
       }
       if((opcode[i]&0x2f)==6) // BLEZ
       {
@@ -4458,13 +4793,7 @@ void do_ccstub(int n)
         //emit_movimm(start+i*4+8,addr);
         emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
         emit_cmpimm(s1l,1);
-        if(s1h>=0) emit_mov(addr,ntaddr);
         emit_cmovl_reg(alt,addr);
-        if(s1h>=0) {
-          emit_test(s1h,s1h);
-          emit_cmovne_reg(ntaddr,addr);
-          emit_cmovs_reg(alt,addr);
-        }
       }
       if((opcode[i]&0x2f)==7) // BGTZ
       {
@@ -4472,21 +4801,14 @@ void do_ccstub(int n)
         //emit_movimm(start+i*4+8,ntaddr);
         emit_mov2imm_compact(ba[i],addr,start+i*4+8,ntaddr);
         emit_cmpimm(s1l,1);
-        if(s1h>=0) emit_mov(addr,alt);
         emit_cmovl_reg(ntaddr,addr);
-        if(s1h>=0) {
-          emit_test(s1h,s1h);
-          emit_cmovne_reg(alt,addr);
-          emit_cmovs_reg(ntaddr,addr);
-        }
       }
       if((opcode[i]==1)&&(opcode2[i]&0x2D)==0) // BLTZ
       {
         //emit_movimm(ba[i],alt);
         //emit_movimm(start+i*4+8,addr);
         emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
-        if(s1h>=0) emit_test(s1h,s1h);
-        else emit_test(s1l,s1l);
+        emit_test(s1l,s1l);
         emit_cmovs_reg(alt,addr);
       }
       if((opcode[i]==1)&&(opcode2[i]&0x2D)==1) // BGEZ
@@ -4494,8 +4816,7 @@ void do_ccstub(int n)
         //emit_movimm(ba[i],addr);
         //emit_movimm(start+i*4+8,alt);
         emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
-        if(s1h>=0) emit_test(s1h,s1h);
-        else emit_test(s1l,s1l);
+        emit_test(s1l,s1l);
         emit_cmovs_reg(alt,addr);
       }
       if(opcode[i]==0x11 && opcode2[i]==0x08 ) {
@@ -4516,7 +4837,7 @@ void do_ccstub(int n)
           emit_cmovne_reg(alt,addr);
         }
       }
-      emit_writeword(addr,(int)&pcaddr);
+      emit_writeword(addr,&pcaddr);
     }
     else
     if(itype[i]==RJUMP)
@@ -4525,70 +4846,46 @@ void do_ccstub(int n)
       if(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]) {
         r=get_reg(branch_regs[i].regmap,RTEMP);
       }
-      emit_writeword(r,(int)&pcaddr);
+      emit_writeword(r,&pcaddr);
     }
-    else {SysPrintf("Unknown branch type in do_ccstub\n");exit(1);}
+    else {SysPrintf("Unknown branch type in do_ccstub\n");abort();}
   }
   // Update cycle count
   assert(branch_regs[i].regmap[HOST_CCREG]==CCREG||branch_regs[i].regmap[HOST_CCREG]==-1);
-  if(stubs[n][3]) emit_addimm(HOST_CCREG,CLOCK_ADJUST((int)stubs[n][3]),HOST_CCREG);
-  emit_call((int)cc_interrupt);
-  if(stubs[n][3]) emit_addimm(HOST_CCREG,-CLOCK_ADJUST((int)stubs[n][3]),HOST_CCREG);
-  if(stubs[n][6]==TAKEN) {
-    if(internal_branch(branch_regs[i].is32,ba[i]))
+  if(stubs[n].a) emit_addimm(HOST_CCREG,CLOCK_ADJUST((signed int)stubs[n].a),HOST_CCREG);
+  emit_far_call(cc_interrupt);
+  if(stubs[n].a) emit_addimm(HOST_CCREG,-CLOCK_ADJUST((signed int)stubs[n].a),HOST_CCREG);
+  if(stubs[n].d==TAKEN) {
+    if(internal_branch(ba[i]))
       load_needed_regs(branch_regs[i].regmap,regs[(ba[i]-start)>>2].regmap_entry);
     else if(itype[i]==RJUMP) {
       if(get_reg(branch_regs[i].regmap,RTEMP)>=0)
-        emit_readword((int)&pcaddr,get_reg(branch_regs[i].regmap,RTEMP));
+        emit_readword(&pcaddr,get_reg(branch_regs[i].regmap,RTEMP));
       else
         emit_loadreg(rs1[i],get_reg(branch_regs[i].regmap,rs1[i]));
     }
-  }else if(stubs[n][6]==NOTTAKEN) {
+  }else if(stubs[n].d==NOTTAKEN) {
     if(i<slen-2) load_needed_regs(branch_regs[i].regmap,regmap_pre[i+2]);
     else load_all_regs(branch_regs[i].regmap);
-  }else if(stubs[n][6]==NULLDS) {
+  }else if(stubs[n].d==NULLDS) {
     // Delay slot instruction is nullified ("likely" branch)
     if(i<slen-2) load_needed_regs(regs[i].regmap,regmap_pre[i+2]);
     else load_all_regs(regs[i].regmap);
   }else{
     load_all_regs(branch_regs[i].regmap);
   }
-  emit_jmp(stubs[n][2]); // return address
-
-  /* This works but uses a lot of memory...
-  emit_readword((int)&last_count,ECX);
-  emit_add(HOST_CCREG,ECX,EAX);
-  emit_writeword(EAX,(int)&Count);
-  emit_call((int)gen_interupt);
-  emit_readword((int)&Count,HOST_CCREG);
-  emit_readword((int)&next_interupt,EAX);
-  emit_readword((int)&pending_exception,EBX);
-  emit_writeword(EAX,(int)&last_count);
-  emit_sub(HOST_CCREG,EAX,HOST_CCREG);
-  emit_test(EBX,EBX);
-  int jne_instr=(int)out;
-  emit_jne(0);
-  if(stubs[n][3]) emit_addimm(HOST_CCREG,-2*stubs[n][3],HOST_CCREG);
-  load_all_regs(branch_regs[i].regmap);
-  emit_jmp(stubs[n][2]); // return address
-  set_jump_target(jne_instr,(int)out);
-  emit_readword((int)&pcaddr,EAX);
-  // Call get_addr_ht instead of doing the hash table here.
-  // This code is executed infrequently and takes up a lot of space
-  // so smaller is better.
-  emit_storereg(CCREG,HOST_CCREG);
-  emit_pushreg(EAX);
-  emit_call((int)get_addr_ht);
-  emit_loadreg(CCREG,HOST_CCREG);
-  emit_addimm(ESP,4,ESP);
-  emit_jmpreg(EAX);*/
-}
-
-add_to_linker(int addr,int target,int ext)
-{
-  link_addr[linkcount][0]=addr;
-  link_addr[linkcount][1]=target;
-  link_addr[linkcount][2]=ext;
+  if (stubs[n].retaddr)
+    emit_jmp(stubs[n].retaddr);
+  else
+    do_jump_vaddr(stubs[n].e);
+}
+
+static void add_to_linker(void *addr, u_int target, int ext)
+{
+  assert(linkcount < ARRAY_SIZE(link_addr));
+  link_addr[linkcount].addr = addr;
+  link_addr[linkcount].target = target;
+  link_addr[linkcount].ext = ext;
   linkcount++;
 }
 
@@ -4602,7 +4899,7 @@ static void ujump_assemble_write_ra(int i)
   return_address=start+i*4+8;
   if(rt>=0) {
     #ifdef USE_MINI_HT
-    if(internal_branch(branch_regs[i].is32,return_address)&&rt1[i+1]!=31) {
+    if(internal_branch(return_address)&&rt1[i+1]!=31) {
       int temp=-1; // note: must be ds-safe
       #ifdef HOST_TEMPREG
       temp=HOST_TEMPREG;
@@ -4616,20 +4913,19 @@ static void ujump_assemble_write_ra(int i)
       #ifdef REG_PREFETCH
       if(temp>=0)
       {
-        if(i_regmap[temp]!=PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp);
+        if(i_regmap[temp]!=PTEMP) emit_movimm((uintptr_t)hash_table_get(return_address),temp);
       }
       #endif
       emit_movimm(return_address,rt); // PC into link register
       #ifdef IMM_PREFETCH
-      emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+      emit_prefetch(hash_table_get(return_address));
       #endif
     }
   }
 }
 
-void ujump_assemble(int i,struct regstat *i_regs)
+static void ujump_assemble(int i,struct regstat *i_regs)
 {
-  signed char *i_regmap=i_regs->regmap;
   int ra_done=0;
   if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
   address_generation(i+1,i_regs,regs[i].regmap_entry);
@@ -4637,9 +4933,10 @@ void ujump_assemble(int i,struct regstat *i_regs)
   int temp=get_reg(branch_regs[i].regmap,PTEMP);
   if(rt1[i]==31&&temp>=0)
   {
+    signed char *i_regmap=i_regs->regmap;
     int return_address=start+i*4+8;
     if(get_reg(branch_regs[i].regmap,31)>0)
-    if(i_regmap[temp]==PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp);
+    if(i_regmap[temp]==PTEMP) emit_movimm((uintptr_t)hash_table_get(return_address),temp);
   }
   #endif
   if(rt1[i]==31&&(rt1[i]==rs1[i+1]||rt1[i]==rs2[i+1])) {
@@ -4648,33 +4945,30 @@ void ujump_assemble(int i,struct regstat *i_regs)
   }
   ds_assemble(i+1,i_regs);
   uint64_t bc_unneeded=branch_regs[i].u;
-  uint64_t bc_unneeded_upper=branch_regs[i].uu;
   bc_unneeded|=1|(1LL<<rt1[i]);
-  bc_unneeded_upper|=1|(1LL<<rt1[i]);
-  wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                bc_unneeded,bc_unneeded_upper);
-  load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+  wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,bc_unneeded);
+  load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,CCREG);
   if(!ra_done&&rt1[i]==31)
     ujump_assemble_write_ra(i);
   int cc,adj;
   cc=get_reg(branch_regs[i].regmap,CCREG);
   assert(cc==HOST_CCREG);
-  store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+  store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
   #ifdef REG_PREFETCH
   if(rt1[i]==31&&temp>=0) emit_prefetchreg(temp);
   #endif
   do_cc(i,branch_regs[i].regmap,&adj,ba[i],TAKEN,0);
   if(adj) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc);
-  load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
-  if(internal_branch(branch_regs[i].is32,ba[i]))
+  load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
+  if(internal_branch(ba[i]))
     assem_debug("branch: internal\n");
   else
     assem_debug("branch: external\n");
-  if(internal_branch(branch_regs[i].is32,ba[i])&&is_ds[(ba[i]-start)>>2]) {
+  if(internal_branch(ba[i])&&is_ds[(ba[i]-start)>>2]) {
     ds_assemble_entry(i);
   }
   else {
-    add_to_linker((int)out,ba[i],internal_branch(branch_regs[i].is32,ba[i]));
+    add_to_linker(out,ba[i],internal_branch(ba[i]));
     emit_jmp(0);
   }
 }
@@ -4691,20 +4985,19 @@ static void rjump_assemble_write_ra(int i)
   #ifdef REG_PREFETCH
   if(temp>=0)
   {
-    if(i_regmap[temp]!=PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp);
+    if(i_regmap[temp]!=PTEMP) emit_movimm((uintptr_t)hash_table_get(return_address),temp);
   }
   #endif
   emit_movimm(return_address,rt); // PC into link register
   #ifdef IMM_PREFETCH
-  emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+  emit_prefetch(hash_table_get(return_address));
   #endif
 }
 
-void rjump_assemble(int i,struct regstat *i_regs)
+static void rjump_assemble(int i,struct regstat *i_regs)
 {
-  signed char *i_regmap=i_regs->regmap;
   int temp;
-  int rs,cc,adj;
+  int rs,cc;
   int ra_done=0;
   rs=get_reg(branch_regs[i].regmap,rs1[i]);
   assert(rs>=0);
@@ -4721,8 +5014,9 @@ void rjump_assemble(int i,struct regstat *i_regs)
   if(rt1[i]==31)
   {
     if((temp=get_reg(branch_regs[i].regmap,PTEMP))>=0) {
+      signed char *i_regmap=i_regs->regmap;
       int return_address=start+i*4+8;
-      if(i_regmap[temp]==PTEMP) emit_movimm((int)hash_table[((return_address>>16)^return_address)&0xFFFF],temp);
+      if(i_regmap[temp]==PTEMP) emit_movimm((uintptr_t)hash_table_get(return_address),temp);
     }
   }
   #endif
@@ -4738,17 +5032,15 @@ void rjump_assemble(int i,struct regstat *i_regs)
   }
   ds_assemble(i+1,i_regs);
   uint64_t bc_unneeded=branch_regs[i].u;
-  uint64_t bc_unneeded_upper=branch_regs[i].uu;
   bc_unneeded|=1|(1LL<<rt1[i]);
-  bc_unneeded_upper|=1|(1LL<<rt1[i]);
   bc_unneeded&=~(1LL<<rs1[i]);
-  wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                bc_unneeded,bc_unneeded_upper);
-  load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],CCREG);
+  wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,bc_unneeded);
+  load_regs(regs[i].regmap,branch_regs[i].regmap,rs1[i],CCREG);
   if(!ra_done&&rt1[i]!=0)
     rjump_assemble_write_ra(i);
   cc=get_reg(branch_regs[i].regmap,CCREG);
   assert(cc==HOST_CCREG);
+  (void)cc;
   #ifdef USE_MINI_HT
   int rh=get_reg(branch_regs[i].regmap,RHASH);
   int ht=get_reg(branch_regs[i].regmap,RHTBL);
@@ -4758,9 +5050,9 @@ void rjump_assemble(int i,struct regstat *i_regs)
     do_rhash(rs,rh);
   }
   #endif
-  store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,-1);
+  store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,-1);
   #ifdef DESTRUCTIVE_WRITEBACK
-  if((branch_regs[i].dirty>>rs)&(branch_regs[i].is32>>rs1[i])&1) {
+  if((branch_regs[i].dirty>>rs)&1) {
     if(rs1[i]!=rt1[i+1]&&rs1[i]!=rt2[i+1]) {
       emit_loadreg(rs1[i],rs);
     }
@@ -4778,13 +5070,13 @@ void rjump_assemble(int i,struct regstat *i_regs)
   //if(adj) emit_addimm(cc,2*(ccadj[i]+2-adj),cc); // ??? - Shouldn't happen
   //assert(adj==0);
   emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG);
-  add_stub(CC_STUB,(int)out,jump_vaddr_reg[rs],0,i,-1,TAKEN,0);
+  add_stub(CC_STUB,out,NULL,0,i,-1,TAKEN,rs);
   if(itype[i+1]==COP0&&(source[i+1]&0x3f)==0x10)
     // special case for RFE
     emit_jmp(0);
   else
     emit_jns(0);
-  //load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,-1);
+  //load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,-1);
   #ifdef USE_MINI_HT
   if(rs1[i]==31) {
     do_miniht_jump(rs,rh,ht);
@@ -4792,76 +5084,40 @@ void rjump_assemble(int i,struct regstat *i_regs)
   else
   #endif
   {
-    //if(rs!=EAX) emit_mov(rs,EAX);
-    //emit_jmp((int)jump_vaddr_eax);
-    emit_jmp(jump_vaddr_reg[rs]);
-  }
-  /* Check hash table
-  temp=!rs;
-  emit_mov(rs,temp);
-  emit_shrimm(rs,16,rs);
-  emit_xor(temp,rs,rs);
-  emit_movzwl_reg(rs,rs);
-  emit_shlimm(rs,4,rs);
-  emit_cmpmem_indexed((int)hash_table,rs,temp);
-  emit_jne((int)out+14);
-  emit_readword_indexed((int)hash_table+4,rs,rs);
-  emit_jmpreg(rs);
-  emit_cmpmem_indexed((int)hash_table+8,rs,temp);
-  emit_addimm_no_flags(8,rs);
-  emit_jeq((int)out-17);
-  // No hit on hash table, call compiler
-  emit_pushreg(temp);
-//DEBUG >
-#ifdef DEBUG_CYCLE_COUNT
-  emit_readword((int)&last_count,ECX);
-  emit_add(HOST_CCREG,ECX,HOST_CCREG);
-  emit_readword((int)&next_interupt,ECX);
-  emit_writeword(HOST_CCREG,(int)&Count);
-  emit_sub(HOST_CCREG,ECX,HOST_CCREG);
-  emit_writeword(ECX,(int)&last_count);
-#endif
-//DEBUG <
-  emit_storereg(CCREG,HOST_CCREG);
-  emit_call((int)get_addr);
-  emit_loadreg(CCREG,HOST_CCREG);
-  emit_addimm(ESP,4,ESP);
-  emit_jmpreg(EAX);*/
+    do_jump_vaddr(rs);
+  }
   #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
   if(rt1[i]!=31&&i<slen-2&&(((u_int)out)&7)) emit_mov(13,13);
   #endif
 }
 
-void cjump_assemble(int i,struct regstat *i_regs)
+static void cjump_assemble(int i,struct regstat *i_regs)
 {
   signed char *i_regmap=i_regs->regmap;
   int cc;
   int match;
-  match=match_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+  match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
   assem_debug("match=%d\n",match);
-  int s1h,s1l,s2h,s2l;
-  int prev_cop1_usable=cop1_usable;
+  int s1l,s2l;
   int unconditional=0,nop=0;
-  int only32=0;
   int invert=0;
-  int internal=internal_branch(branch_regs[i].is32,ba[i]);
+  int internal=internal_branch(ba[i]);
   if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
   if(!match) invert=1;
   #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
   if(i>(ba[i]-start)>>2) invert=1;
   #endif
+  #ifdef __aarch64__
+  invert=1; // because of near cond. branches
+  #endif
 
   if(ooo[i]) {
     s1l=get_reg(branch_regs[i].regmap,rs1[i]);
-    s1h=get_reg(branch_regs[i].regmap,rs1[i]|64);
     s2l=get_reg(branch_regs[i].regmap,rs2[i]);
-    s2h=get_reg(branch_regs[i].regmap,rs2[i]|64);
   }
   else {
     s1l=get_reg(i_regmap,rs1[i]);
-    s1h=get_reg(i_regmap,rs1[i]|64);
     s2l=get_reg(i_regmap,rs2[i]);
-    s2h=get_reg(i_regmap,rs2[i]|64);
   }
   if(rs1[i]==0&&rs2[i]==0)
   {
@@ -4874,17 +5130,12 @@ void cjump_assemble(int i,struct regstat *i_regs)
   }
   else if(rs1[i]==0)
   {
-    s1l=s2l;s1h=s2h;
-    s2l=s2h=-1;
-    only32=(regs[i].was32>>rs2[i])&1;
+    s1l=s2l;
+    s2l=-1;
   }
   else if(rs2[i]==0)
   {
-    s2l=s2h=-1;
-    only32=(regs[i].was32>>rs1[i])&1;
-  }
-  else {
-    only32=(regs[i].was32>>rs1[i])&(regs[i].was32>>rs2[i])&1;
+    s2l=-1;
   }
 
   if(ooo[i]) {
@@ -4894,26 +5145,22 @@ void cjump_assemble(int i,struct regstat *i_regs)
     ds_assemble(i+1,i_regs);
     int adj;
     uint64_t bc_unneeded=branch_regs[i].u;
-    uint64_t bc_unneeded_upper=branch_regs[i].uu;
     bc_unneeded&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
-    bc_unneeded_upper&=~((1LL<<us1[i])|(1LL<<us2[i]));
     bc_unneeded|=1;
-    bc_unneeded_upper|=1;
-    wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                  bc_unneeded,bc_unneeded_upper);
-    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],rs2[i]);
-    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+    wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,bc_unneeded);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,rs1[i],rs2[i]);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,CCREG);
     cc=get_reg(branch_regs[i].regmap,CCREG);
     assert(cc==HOST_CCREG);
     if(unconditional)
-      store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+      store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
     //do_cc(i,branch_regs[i].regmap,&adj,unconditional?ba[i]:-1,unconditional);
     //assem_debug("cycle count (adj)\n");
     if(unconditional) {
       do_cc(i,branch_regs[i].regmap,&adj,ba[i],TAKEN,0);
       if(i!=(ba[i]-start)>>2 || source[i+1]!=0) {
         if(adj) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc);
-        load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+        load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
         if(internal)
           assem_debug("branch: internal\n");
         else
@@ -4922,7 +5169,7 @@ void cjump_assemble(int i,struct regstat *i_regs)
           ds_assemble_entry(i);
         }
         else {
-          add_to_linker((int)out,ba[i],internal);
+          add_to_linker(out,ba[i],internal);
           emit_jmp(0);
         }
         #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
@@ -4932,51 +5179,14 @@ void cjump_assemble(int i,struct regstat *i_regs)
     }
     else if(nop) {
       emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc);
-      int jaddr=(int)out;
+      void *jaddr=out;
       emit_jns(0);
-      add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+      add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
     }
     else {
-      int taken=0,nottaken=0,nottaken1=0;
+      void *taken = NULL, *nottaken = NULL, *nottaken1 = NULL;
       do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert);
       if(adj&&!invert) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc);
-      if(!only32)
-      {
-        assert(s1h>=0);
-        if(opcode[i]==4) // BEQ
-        {
-          if(s2h>=0) emit_cmp(s1h,s2h);
-          else emit_test(s1h,s1h);
-          nottaken1=(int)out;
-          emit_jne(1);
-        }
-        if(opcode[i]==5) // BNE
-        {
-          if(s2h>=0) emit_cmp(s1h,s2h);
-          else emit_test(s1h,s1h);
-          if(invert) taken=(int)out;
-          else add_to_linker((int)out,ba[i],internal);
-          emit_jne(0);
-        }
-        if(opcode[i]==6) // BLEZ
-        {
-          emit_test(s1h,s1h);
-          if(invert) taken=(int)out;
-          else add_to_linker((int)out,ba[i],internal);
-          emit_js(0);
-          nottaken1=(int)out;
-          emit_jne(1);
-        }
-        if(opcode[i]==7) // BGTZ
-        {
-          emit_test(s1h,s1h);
-          nottaken1=(int)out;
-          emit_js(1);
-          if(invert) taken=(int)out;
-          else add_to_linker((int)out,ba[i],internal);
-          emit_jne(0);
-        }
-      } // if(!only32)
 
       //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
       assert(s1l>=0);
@@ -4985,10 +5195,10 @@ void cjump_assemble(int i,struct regstat *i_regs)
         if(s2l>=0) emit_cmp(s1l,s2l);
         else emit_test(s1l,s1l);
         if(invert){
-          nottaken=(int)out;
-          emit_jne(1);
+          nottaken=out;
+          emit_jne(DJT_1);
         }else{
-          add_to_linker((int)out,ba[i],internal);
+          add_to_linker(out,ba[i],internal);
           emit_jeq(0);
         }
       }
@@ -4997,10 +5207,10 @@ void cjump_assemble(int i,struct regstat *i_regs)
         if(s2l>=0) emit_cmp(s1l,s2l);
         else emit_test(s1l,s1l);
         if(invert){
-          nottaken=(int)out;
-          emit_jeq(1);
+          nottaken=out;
+          emit_jeq(DJT_1);
         }else{
-          add_to_linker((int)out,ba[i],internal);
+          add_to_linker(out,ba[i],internal);
           emit_jne(0);
         }
       }
@@ -5008,10 +5218,10 @@ void cjump_assemble(int i,struct regstat *i_regs)
       {
         emit_cmpimm(s1l,1);
         if(invert){
-          nottaken=(int)out;
-          emit_jge(1);
+          nottaken=out;
+          emit_jge(DJT_1);
         }else{
-          add_to_linker((int)out,ba[i],internal);
+          add_to_linker(out,ba[i],internal);
           emit_jl(0);
         }
       }
@@ -5019,31 +5229,31 @@ void cjump_assemble(int i,struct regstat *i_regs)
       {
         emit_cmpimm(s1l,1);
         if(invert){
-          nottaken=(int)out;
-          emit_jl(1);
+          nottaken=out;
+          emit_jl(DJT_1);
         }else{
-          add_to_linker((int)out,ba[i],internal);
+          add_to_linker(out,ba[i],internal);
           emit_jge(0);
         }
       }
       if(invert) {
-        if(taken) set_jump_target(taken,(int)out);
+        if(taken) set_jump_target(taken, out);
         #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
         if(match&&(!internal||!is_ds[(ba[i]-start)>>2])) {
           if(adj) {
             emit_addimm(cc,-CLOCK_ADJUST(adj),cc);
-            add_to_linker((int)out,ba[i],internal);
+            add_to_linker(out,ba[i],internal);
           }else{
             emit_addnop(13);
-            add_to_linker((int)out,ba[i],internal*2);
+            add_to_linker(out,ba[i],internal*2);
           }
           emit_jmp(0);
         }else
         #endif
         {
           if(adj) emit_addimm(cc,-CLOCK_ADJUST(adj),cc);
-          store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
-          load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+          store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
+          load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
           if(internal)
             assem_debug("branch: internal\n");
           else
@@ -5052,14 +5262,14 @@ void cjump_assemble(int i,struct regstat *i_regs)
             ds_assemble_entry(i);
           }
           else {
-            add_to_linker((int)out,ba[i],internal);
+            add_to_linker(out,ba[i],internal);
             emit_jmp(0);
           }
         }
-        set_jump_target(nottaken,(int)out);
+        set_jump_target(nottaken, out);
       }
 
-      if(nottaken1) set_jump_target(nottaken1,(int)out);
+      if(nottaken1) set_jump_target(nottaken1, out);
       if(adj) {
         if(!invert) emit_addimm(cc,CLOCK_ADJUST(adj),cc);
       }
@@ -5071,90 +5281,50 @@ void cjump_assemble(int i,struct regstat *i_regs)
     //if(likely[i]) printf("IOL\n");
     //else
     //printf("IOE\n");
-    int taken=0,nottaken=0,nottaken1=0;
+    void *taken = NULL, *nottaken = NULL, *nottaken1 = NULL;
     if(!unconditional&&!nop) {
-      if(!only32)
-      {
-        assert(s1h>=0);
-        if((opcode[i]&0x2f)==4) // BEQ
-        {
-          if(s2h>=0) emit_cmp(s1h,s2h);
-          else emit_test(s1h,s1h);
-          nottaken1=(int)out;
-          emit_jne(2);
-        }
-        if((opcode[i]&0x2f)==5) // BNE
-        {
-          if(s2h>=0) emit_cmp(s1h,s2h);
-          else emit_test(s1h,s1h);
-          taken=(int)out;
-          emit_jne(1);
-        }
-        if((opcode[i]&0x2f)==6) // BLEZ
-        {
-          emit_test(s1h,s1h);
-          taken=(int)out;
-          emit_js(1);
-          nottaken1=(int)out;
-          emit_jne(2);
-        }
-        if((opcode[i]&0x2f)==7) // BGTZ
-        {
-          emit_test(s1h,s1h);
-          nottaken1=(int)out;
-          emit_js(2);
-          taken=(int)out;
-          emit_jne(1);
-        }
-      } // if(!only32)
-
       //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
       assert(s1l>=0);
       if((opcode[i]&0x2f)==4) // BEQ
       {
         if(s2l>=0) emit_cmp(s1l,s2l);
         else emit_test(s1l,s1l);
-        nottaken=(int)out;
-        emit_jne(2);
+        nottaken=out;
+        emit_jne(DJT_2);
       }
       if((opcode[i]&0x2f)==5) // BNE
       {
         if(s2l>=0) emit_cmp(s1l,s2l);
         else emit_test(s1l,s1l);
-        nottaken=(int)out;
-        emit_jeq(2);
+        nottaken=out;
+        emit_jeq(DJT_2);
       }
       if((opcode[i]&0x2f)==6) // BLEZ
       {
         emit_cmpimm(s1l,1);
-        nottaken=(int)out;
-        emit_jge(2);
+        nottaken=out;
+        emit_jge(DJT_2);
       }
       if((opcode[i]&0x2f)==7) // BGTZ
       {
         emit_cmpimm(s1l,1);
-        nottaken=(int)out;
-        emit_jl(2);
+        nottaken=out;
+        emit_jl(DJT_2);
       }
     } // if(!unconditional)
     int adj;
     uint64_t ds_unneeded=branch_regs[i].u;
-    uint64_t ds_unneeded_upper=branch_regs[i].uu;
     ds_unneeded&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
-    ds_unneeded_upper&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
-    if((~ds_unneeded_upper>>rt1[i+1])&1) ds_unneeded_upper&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
     ds_unneeded|=1;
-    ds_unneeded_upper|=1;
     // branch taken
     if(!nop) {
-      if(taken) set_jump_target(taken,(int)out);
+      if(taken) set_jump_target(taken, out);
       assem_debug("1:\n");
-      wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                    ds_unneeded,ds_unneeded_upper);
+      wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,ds_unneeded);
       // load regs
-      load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
+      load_regs(regs[i].regmap,branch_regs[i].regmap,rs1[i+1],rs2[i+1]);
       address_generation(i+1,&branch_regs[i],0);
-      load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,INVCP);
+      load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,INVCP);
       ds_assemble(i+1,&branch_regs[i]);
       cc=get_reg(branch_regs[i].regmap,CCREG);
       if(cc==-1) {
@@ -5162,11 +5332,11 @@ void cjump_assemble(int i,struct regstat *i_regs)
         // CHECK: Is the following instruction (fall thru) allocated ok?
       }
       assert(cc==HOST_CCREG);
-      store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+      store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
       do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
       assem_debug("cycle count (adj)\n");
       if(adj) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc);
-      load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+      load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
       if(internal)
         assem_debug("branch: internal\n");
       else
@@ -5175,22 +5345,20 @@ void cjump_assemble(int i,struct regstat *i_regs)
         ds_assemble_entry(i);
       }
       else {
-        add_to_linker((int)out,ba[i],internal);
+        add_to_linker(out,ba[i],internal);
         emit_jmp(0);
       }
     }
     // branch not taken
-    cop1_usable=prev_cop1_usable;
     if(!unconditional) {
-      if(nottaken1) set_jump_target(nottaken1,(int)out);
-      set_jump_target(nottaken,(int)out);
+      if(nottaken1) set_jump_target(nottaken1, out);
+      set_jump_target(nottaken, out);
       assem_debug("2:\n");
       if(!likely[i]) {
-        wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                      ds_unneeded,ds_unneeded_upper);
-        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
+        wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,ds_unneeded);
+        load_regs(regs[i].regmap,branch_regs[i].regmap,rs1[i+1],rs2[i+1]);
         address_generation(i+1,&branch_regs[i],0);
-        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+        load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,CCREG);
         ds_assemble(i+1,&branch_regs[i]);
       }
       cc=get_reg(branch_regs[i].regmap,CCREG);
@@ -5198,52 +5366,51 @@ void cjump_assemble(int i,struct regstat *i_regs)
         // Cycle count isn't in a register, temporarily load it then write it out
         emit_loadreg(CCREG,HOST_CCREG);
         emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG);
-        int jaddr=(int)out;
+        void *jaddr=out;
         emit_jns(0);
-        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+        add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
         emit_storereg(CCREG,HOST_CCREG);
       }
       else{
         cc=get_reg(i_regmap,CCREG);
         assert(cc==HOST_CCREG);
         emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc);
-        int jaddr=(int)out;
+        void *jaddr=out;
         emit_jns(0);
-        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
+        add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
       }
     }
   }
 }
 
-void sjump_assemble(int i,struct regstat *i_regs)
+static void sjump_assemble(int i,struct regstat *i_regs)
 {
   signed char *i_regmap=i_regs->regmap;
   int cc;
   int match;
-  match=match_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+  match=match_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
   assem_debug("smatch=%d\n",match);
-  int s1h,s1l;
-  int prev_cop1_usable=cop1_usable;
+  int s1l;
   int unconditional=0,nevertaken=0;
-  int only32=0;
   int invert=0;
-  int internal=internal_branch(branch_regs[i].is32,ba[i]);
+  int internal=internal_branch(ba[i]);
   if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
   if(!match) invert=1;
   #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
   if(i>(ba[i]-start)>>2) invert=1;
   #endif
+  #ifdef __aarch64__
+  invert=1; // because of near cond. branches
+  #endif
 
   //if(opcode2[i]>=0x10) return; // FIXME (BxxZAL)
   //assert(opcode2[i]<0x10||rs1[i]==0); // FIXME (BxxZAL)
 
   if(ooo[i]) {
     s1l=get_reg(branch_regs[i].regmap,rs1[i]);
-    s1h=get_reg(branch_regs[i].regmap,rs1[i]|64);
   }
   else {
     s1l=get_reg(i_regmap,rs1[i]);
-    s1h=get_reg(i_regmap,rs1[i]|64);
   }
   if(rs1[i]==0)
   {
@@ -5255,9 +5422,6 @@ void sjump_assemble(int i,struct regstat *i_regs)
     //assert(opcode2[i]!=0x10);
     //assert(opcode2[i]!=0x12);
   }
-  else {
-    only32=(regs[i].was32>>rs1[i])&1;
-  }
 
   if(ooo[i]) {
     // Out of order execution (delay slot first)
@@ -5266,15 +5430,11 @@ void sjump_assemble(int i,struct regstat *i_regs)
     ds_assemble(i+1,i_regs);
     int adj;
     uint64_t bc_unneeded=branch_regs[i].u;
-    uint64_t bc_unneeded_upper=branch_regs[i].uu;
     bc_unneeded&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
-    bc_unneeded_upper&=~((1LL<<us1[i])|(1LL<<us2[i]));
     bc_unneeded|=1;
-    bc_unneeded_upper|=1;
-    wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                  bc_unneeded,bc_unneeded_upper);
-    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],rs1[i]);
-    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+    wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,bc_unneeded);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,rs1[i],rs1[i]);
+    load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,CCREG);
     if(rt1[i]==31) {
       int rt,return_address;
       rt=get_reg(branch_regs[i].regmap,31);
@@ -5284,21 +5444,21 @@ void sjump_assemble(int i,struct regstat *i_regs)
         return_address=start+i*4+8;
         emit_movimm(return_address,rt); // PC into link register
         #ifdef IMM_PREFETCH
-        if(!nevertaken) emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+        if(!nevertaken) emit_prefetch(hash_table_get(return_address));
         #endif
       }
     }
     cc=get_reg(branch_regs[i].regmap,CCREG);
     assert(cc==HOST_CCREG);
     if(unconditional)
-      store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+      store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
     //do_cc(i,branch_regs[i].regmap,&adj,unconditional?ba[i]:-1,unconditional);
     assem_debug("cycle count (adj)\n");
     if(unconditional) {
       do_cc(i,branch_regs[i].regmap,&adj,ba[i],TAKEN,0);
       if(i!=(ba[i]-start)>>2 || source[i+1]!=0) {
         if(adj) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc);
-        load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+        load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
         if(internal)
           assem_debug("branch: internal\n");
         else
@@ -5307,7 +5467,7 @@ void sjump_assemble(int i,struct regstat *i_regs)
           ds_assemble_entry(i);
         }
         else {
-          add_to_linker((int)out,ba[i],internal);
+          add_to_linker(out,ba[i],internal);
           emit_jmp(0);
         }
         #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
@@ -5317,51 +5477,24 @@ void sjump_assemble(int i,struct regstat *i_regs)
     }
     else if(nevertaken) {
       emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc);
-      int jaddr=(int)out;
+      void *jaddr=out;
       emit_jns(0);
-      add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+      add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
     }
     else {
-      int nottaken=0;
+      void *nottaken = NULL;
       do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert);
       if(adj&&!invert) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc);
-      if(!only32)
-      {
-        assert(s1h>=0);
-        if((opcode2[i]&0xf)==0) // BLTZ/BLTZAL
-        {
-          emit_test(s1h,s1h);
-          if(invert){
-            nottaken=(int)out;
-            emit_jns(1);
-          }else{
-            add_to_linker((int)out,ba[i],internal);
-            emit_js(0);
-          }
-        }
-        if((opcode2[i]&0xf)==1) // BGEZ/BLTZAL
-        {
-          emit_test(s1h,s1h);
-          if(invert){
-            nottaken=(int)out;
-            emit_js(1);
-          }else{
-            add_to_linker((int)out,ba[i],internal);
-            emit_jns(0);
-          }
-        }
-      } // if(!only32)
-      else
       {
         assert(s1l>=0);
         if((opcode2[i]&0xf)==0) // BLTZ/BLTZAL
         {
           emit_test(s1l,s1l);
           if(invert){
-            nottaken=(int)out;
-            emit_jns(1);
+            nottaken=out;
+            emit_jns(DJT_1);
           }else{
-            add_to_linker((int)out,ba[i],internal);
+            add_to_linker(out,ba[i],internal);
             emit_js(0);
           }
         }
@@ -5369,32 +5502,32 @@ void sjump_assemble(int i,struct regstat *i_regs)
         {
           emit_test(s1l,s1l);
           if(invert){
-            nottaken=(int)out;
-            emit_js(1);
+            nottaken=out;
+            emit_js(DJT_1);
           }else{
-            add_to_linker((int)out,ba[i],internal);
+            add_to_linker(out,ba[i],internal);
             emit_jns(0);
           }
         }
-      } // if(!only32)
+      }
 
       if(invert) {
         #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
         if(match&&(!internal||!is_ds[(ba[i]-start)>>2])) {
           if(adj) {
             emit_addimm(cc,-CLOCK_ADJUST(adj),cc);
-            add_to_linker((int)out,ba[i],internal);
+            add_to_linker(out,ba[i],internal);
           }else{
             emit_addnop(13);
-            add_to_linker((int)out,ba[i],internal*2);
+            add_to_linker(out,ba[i],internal*2);
           }
           emit_jmp(0);
         }else
         #endif
         {
           if(adj) emit_addimm(cc,-CLOCK_ADJUST(adj),cc);
-          store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
-          load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+          store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
+          load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
           if(internal)
             assem_debug("branch: internal\n");
           else
@@ -5403,11 +5536,11 @@ void sjump_assemble(int i,struct regstat *i_regs)
             ds_assemble_entry(i);
           }
           else {
-            add_to_linker((int)out,ba[i],internal);
+            add_to_linker(out,ba[i],internal);
             emit_jmp(0);
           }
         }
-        set_jump_target(nottaken,(int)out);
+        set_jump_target(nottaken, out);
       }
 
       if(adj) {
@@ -5419,7 +5552,7 @@ void sjump_assemble(int i,struct regstat *i_regs)
   {
     // In-order execution (branch first)
     //printf("IOE\n");
-    int nottaken=0;
+    void *nottaken = NULL;
     if(rt1[i]==31) {
       int rt,return_address;
       rt=get_reg(branch_regs[i].regmap,31);
@@ -5428,62 +5561,38 @@ void sjump_assemble(int i,struct regstat *i_regs)
         return_address=start+i*4+8;
         emit_movimm(return_address,rt); // PC into link register
         #ifdef IMM_PREFETCH
-        emit_prefetch(hash_table[((return_address>>16)^return_address)&0xFFFF]);
+        emit_prefetch(hash_table_get(return_address));
         #endif
       }
     }
     if(!unconditional) {
       //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
-      if(!only32)
-      {
-        assert(s1h>=0);
-        if((opcode2[i]&0x0d)==0) // BLTZ/BLTZL/BLTZAL/BLTZALL
-        {
-          emit_test(s1h,s1h);
-          nottaken=(int)out;
-          emit_jns(1);
-        }
-        if((opcode2[i]&0x0d)==1) // BGEZ/BGEZL/BGEZAL/BGEZALL
-        {
-          emit_test(s1h,s1h);
-          nottaken=(int)out;
-          emit_js(1);
-        }
-      } // if(!only32)
-      else
-      {
         assert(s1l>=0);
         if((opcode2[i]&0x0d)==0) // BLTZ/BLTZL/BLTZAL/BLTZALL
         {
           emit_test(s1l,s1l);
-          nottaken=(int)out;
-          emit_jns(1);
+          nottaken=out;
+          emit_jns(DJT_1);
         }
         if((opcode2[i]&0x0d)==1) // BGEZ/BGEZL/BGEZAL/BGEZALL
         {
           emit_test(s1l,s1l);
-          nottaken=(int)out;
-          emit_js(1);
+          nottaken=out;
+          emit_js(DJT_1);
         }
-      }
     } // if(!unconditional)
     int adj;
     uint64_t ds_unneeded=branch_regs[i].u;
-    uint64_t ds_unneeded_upper=branch_regs[i].uu;
     ds_unneeded&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
-    ds_unneeded_upper&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
-    if((~ds_unneeded_upper>>rt1[i+1])&1) ds_unneeded_upper&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
     ds_unneeded|=1;
-    ds_unneeded_upper|=1;
     // branch taken
     if(!nevertaken) {
       //assem_debug("1:\n");
-      wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                    ds_unneeded,ds_unneeded_upper);
+      wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,ds_unneeded);
       // load regs
-      load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
+      load_regs(regs[i].regmap,branch_regs[i].regmap,rs1[i+1],rs2[i+1]);
       address_generation(i+1,&branch_regs[i],0);
-      load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,INVCP);
+      load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,INVCP);
       ds_assemble(i+1,&branch_regs[i]);
       cc=get_reg(branch_regs[i].regmap,CCREG);
       if(cc==-1) {
@@ -5491,11 +5600,11 @@ void sjump_assemble(int i,struct regstat *i_regs)
         // CHECK: Is the following instruction (fall thru) allocated ok?
       }
       assert(cc==HOST_CCREG);
-      store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+      store_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
       do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
       assem_debug("cycle count (adj)\n");
       if(adj) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc);
-      load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
+      load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,ba[i]);
       if(internal)
         assem_debug("branch: internal\n");
       else
@@ -5504,226 +5613,19 @@ void sjump_assemble(int i,struct regstat *i_regs)
         ds_assemble_entry(i);
       }
       else {
-        add_to_linker((int)out,ba[i],internal);
+        add_to_linker(out,ba[i],internal);
         emit_jmp(0);
       }
     }
     // branch not taken
-    cop1_usable=prev_cop1_usable;
     if(!unconditional) {
-      set_jump_target(nottaken,(int)out);
-      assem_debug("1:\n");
-      if(!likely[i]) {
-        wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                      ds_unneeded,ds_unneeded_upper);
-        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
-        address_generation(i+1,&branch_regs[i],0);
-        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
-        ds_assemble(i+1,&branch_regs[i]);
-      }
-      cc=get_reg(branch_regs[i].regmap,CCREG);
-      if(cc==-1&&!likely[i]) {
-        // Cycle count isn't in a register, temporarily load it then write it out
-        emit_loadreg(CCREG,HOST_CCREG);
-        emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG);
-        int jaddr=(int)out;
-        emit_jns(0);
-        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
-        emit_storereg(CCREG,HOST_CCREG);
-      }
-      else{
-        cc=get_reg(i_regmap,CCREG);
-        assert(cc==HOST_CCREG);
-        emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc);
-        int jaddr=(int)out;
-        emit_jns(0);
-        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
-      }
-    }
-  }
-}
-
-void fjump_assemble(int i,struct regstat *i_regs)
-{
-  signed char *i_regmap=i_regs->regmap;
-  int cc;
-  int match;
-  match=match_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
-  assem_debug("fmatch=%d\n",match);
-  int fs,cs;
-  int eaddr;
-  int invert=0;
-  int internal=internal_branch(branch_regs[i].is32,ba[i]);
-  if(i==(ba[i]-start)>>2) assem_debug("idle loop\n");
-  if(!match) invert=1;
-  #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
-  if(i>(ba[i]-start)>>2) invert=1;
-  #endif
-
-  if(ooo[i]) {
-    fs=get_reg(branch_regs[i].regmap,FSREG);
-    address_generation(i+1,i_regs,regs[i].regmap_entry); // Is this okay?
-  }
-  else {
-    fs=get_reg(i_regmap,FSREG);
-  }
-
-  // Check cop1 unusable
-  if(!cop1_usable) {
-    cs=get_reg(i_regmap,CSREG);
-    assert(cs>=0);
-    emit_testimm(cs,0x20000000);
-    eaddr=(int)out;
-    emit_jeq(0);
-    add_stub(FP_STUB,eaddr,(int)out,i,cs,(int)i_regs,0,0);
-    cop1_usable=1;
-  }
-
-  if(ooo[i]) {
-    // Out of order execution (delay slot first)
-    //printf("OOOE\n");
-    ds_assemble(i+1,i_regs);
-    int adj;
-    uint64_t bc_unneeded=branch_regs[i].u;
-    uint64_t bc_unneeded_upper=branch_regs[i].uu;
-    bc_unneeded&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
-    bc_unneeded_upper&=~((1LL<<us1[i])|(1LL<<us2[i]));
-    bc_unneeded|=1;
-    bc_unneeded_upper|=1;
-    wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                  bc_unneeded,bc_unneeded_upper);
-    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i],rs1[i]);
-    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
-    cc=get_reg(branch_regs[i].regmap,CCREG);
-    assert(cc==HOST_CCREG);
-    do_cc(i,branch_regs[i].regmap,&adj,-1,0,invert);
-    assem_debug("cycle count (adj)\n");
-    if(1) {
-      int nottaken=0;
-      if(adj&&!invert) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc);
-      if(1) {
-        assert(fs>=0);
-        emit_testimm(fs,0x800000);
-        if(source[i]&0x10000) // BC1T
-        {
-          if(invert){
-            nottaken=(int)out;
-            emit_jeq(1);
-          }else{
-            add_to_linker((int)out,ba[i],internal);
-            emit_jne(0);
-          }
-        }
-        else // BC1F
-          if(invert){
-            nottaken=(int)out;
-            emit_jne(1);
-          }else{
-            add_to_linker((int)out,ba[i],internal);
-            emit_jeq(0);
-          }
-        {
-        }
-      } // if(!only32)
-
-      if(invert) {
-        if(adj) emit_addimm(cc,-CLOCK_ADJUST(adj),cc);
-        #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
-        else if(match) emit_addnop(13);
-        #endif
-        store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
-        load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
-        if(internal)
-          assem_debug("branch: internal\n");
-        else
-          assem_debug("branch: external\n");
-        if(internal&&is_ds[(ba[i]-start)>>2]) {
-          ds_assemble_entry(i);
-        }
-        else {
-          add_to_linker((int)out,ba[i],internal);
-          emit_jmp(0);
-        }
-        set_jump_target(nottaken,(int)out);
-      }
-
-      if(adj) {
-        if(!invert) emit_addimm(cc,CLOCK_ADJUST(adj),cc);
-      }
-    } // (!unconditional)
-  } // if(ooo)
-  else
-  {
-    // In-order execution (branch first)
-    //printf("IOE\n");
-    int nottaken=0;
-    if(1) {
-      //printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
-      if(1) {
-        assert(fs>=0);
-        emit_testimm(fs,0x800000);
-        if(source[i]&0x10000) // BC1T
-        {
-          nottaken=(int)out;
-          emit_jeq(1);
-        }
-        else // BC1F
-        {
-          nottaken=(int)out;
-          emit_jne(1);
-        }
-      }
-    } // if(!unconditional)
-    int adj;
-    uint64_t ds_unneeded=branch_regs[i].u;
-    uint64_t ds_unneeded_upper=branch_regs[i].uu;
-    ds_unneeded&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
-    ds_unneeded_upper&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
-    if((~ds_unneeded_upper>>rt1[i+1])&1) ds_unneeded_upper&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
-    ds_unneeded|=1;
-    ds_unneeded_upper|=1;
-    // branch taken
-    //assem_debug("1:\n");
-    wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                  ds_unneeded,ds_unneeded_upper);
-    // load regs
-    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
-    address_generation(i+1,&branch_regs[i],0);
-    load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,INVCP);
-    ds_assemble(i+1,&branch_regs[i]);
-    cc=get_reg(branch_regs[i].regmap,CCREG);
-    if(cc==-1) {
-      emit_loadreg(CCREG,cc=HOST_CCREG);
-      // CHECK: Is the following instruction (fall thru) allocated ok?
-    }
-    assert(cc==HOST_CCREG);
-    store_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
-    do_cc(i,i_regmap,&adj,ba[i],TAKEN,0);
-    assem_debug("cycle count (adj)\n");
-    if(adj) emit_addimm(cc,CLOCK_ADJUST(ccadj[i]+2-adj),cc);
-    load_regs_bt(branch_regs[i].regmap,branch_regs[i].is32,branch_regs[i].dirty,ba[i]);
-    if(internal)
-      assem_debug("branch: internal\n");
-    else
-      assem_debug("branch: external\n");
-    if(internal&&is_ds[(ba[i]-start)>>2]) {
-      ds_assemble_entry(i);
-    }
-    else {
-      add_to_linker((int)out,ba[i],internal);
-      emit_jmp(0);
-    }
-
-    // branch not taken
-    if(1) { // <- FIXME (don't need this)
-      set_jump_target(nottaken,(int)out);
+      set_jump_target(nottaken, out);
       assem_debug("1:\n");
       if(!likely[i]) {
-        wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,regs[i].is32,
-                      ds_unneeded,ds_unneeded_upper);
-        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,rs1[i+1],rs2[i+1]);
+        wb_invalidate(regs[i].regmap,branch_regs[i].regmap,regs[i].dirty,ds_unneeded);
+        load_regs(regs[i].regmap,branch_regs[i].regmap,rs1[i+1],rs2[i+1]);
         address_generation(i+1,&branch_regs[i],0);
-        load_regs(regs[i].regmap,branch_regs[i].regmap,regs[i].was32,CCREG,CCREG);
+        load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,CCREG);
         ds_assemble(i+1,&branch_regs[i]);
       }
       cc=get_reg(branch_regs[i].regmap,CCREG);
@@ -5731,18 +5633,18 @@ void fjump_assemble(int i,struct regstat *i_regs)
         // Cycle count isn't in a register, temporarily load it then write it out
         emit_loadreg(CCREG,HOST_CCREG);
         emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG);
-        int jaddr=(int)out;
+        void *jaddr=out;
         emit_jns(0);
-        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,NOTTAKEN,0);
+        add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
         emit_storereg(CCREG,HOST_CCREG);
       }
       else{
         cc=get_reg(i_regmap,CCREG);
         assert(cc==HOST_CCREG);
         emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc);
-        int jaddr=(int)out;
+        void *jaddr=out;
         emit_jns(0);
-        add_stub(CC_STUB,jaddr,(int)out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
+        add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
       }
     }
   }
@@ -5751,27 +5653,21 @@ void fjump_assemble(int i,struct regstat *i_regs)
 static void pagespan_assemble(int i,struct regstat *i_regs)
 {
   int s1l=get_reg(i_regs->regmap,rs1[i]);
-  int s1h=get_reg(i_regs->regmap,rs1[i]|64);
   int s2l=get_reg(i_regs->regmap,rs2[i]);
-  int s2h=get_reg(i_regs->regmap,rs2[i]|64);
-  void *nt_branch=NULL;
-  int taken=0;
-  int nottaken=0;
+  void *taken = NULL;
+  void *nottaken = NULL;
   int unconditional=0;
   if(rs1[i]==0)
   {
-    s1l=s2l;s1h=s2h;
-    s2l=s2h=-1;
+    s1l=s2l;
+    s2l=-1;
   }
   else if(rs2[i]==0)
   {
-    s2l=s2h=-1;
-  }
-  if((i_regs->is32>>rs1[i])&(i_regs->is32>>rs2[i])&1) {
-    s1h=s2h=-1;
+    s2l=-1;
   }
   int hr=0;
-  int addr,alt,ntaddr;
+  int addr=-1,alt=-1,ntaddr=-1;
   if(i_regs->regmap[HOST_BTREG]<0) {addr=HOST_BTREG;}
   else {
     while(hr<HOST_REGS)
@@ -5810,7 +5706,7 @@ static void pagespan_assemble(int i,struct regstat *i_regs)
   }
   assert(hr<HOST_REGS);
   if((opcode[i]&0x2e)==4||opcode[i]==0x11) { // BEQ/BNE/BEQL/BNEL/BC1
-    load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,CCREG,CCREG);
+    load_regs(regs[i].regmap_entry,regs[i].regmap,CCREG,CCREG);
   }
   emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG);
   if(opcode[i]==2) // J
@@ -5841,7 +5737,7 @@ static void pagespan_assemble(int i,struct regstat *i_regs)
     }
     else
     #ifdef HAVE_CMOV_IMM
-    if(s1h<0) {
+    if(1) {
       if(s2l>=0) emit_cmp(s1l,s2l);
       else emit_test(s1l,s1l);
       emit_cmov2imm_e_ne_compact(ba[i],start+i*4+8,addr);
@@ -5851,11 +5747,6 @@ static void pagespan_assemble(int i,struct regstat *i_regs)
     {
       assert(s1l>=0);
       emit_mov2imm_compact(ba[i],addr,start+i*4+8,alt);
-      if(s1h>=0) {
-        if(s2h>=0) emit_cmp(s1h,s2h);
-        else emit_test(s1h,s1h);
-        emit_cmovne_reg(alt,addr);
-      }
       if(s2l>=0) emit_cmp(s1l,s2l);
       else emit_test(s1l,s1l);
       emit_cmovne_reg(alt,addr);
@@ -5864,77 +5755,44 @@ static void pagespan_assemble(int i,struct regstat *i_regs)
   if((opcode[i]&0x3f)==5) // BNE
   {
     #ifdef HAVE_CMOV_IMM
-    if(s1h<0) {
-      if(s2l>=0) emit_cmp(s1l,s2l);
-      else emit_test(s1l,s1l);
-      emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
-    }
-    else
+    if(s2l>=0) emit_cmp(s1l,s2l);
+    else emit_test(s1l,s1l);
+    emit_cmov2imm_e_ne_compact(start+i*4+8,ba[i],addr);
+    #else
+    assert(s1l>=0);
+    emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
+    if(s2l>=0) emit_cmp(s1l,s2l);
+    else emit_test(s1l,s1l);
+    emit_cmovne_reg(alt,addr);
     #endif
-    {
-      assert(s1l>=0);
-      emit_mov2imm_compact(start+i*4+8,addr,ba[i],alt);
-      if(s1h>=0) {
-        if(s2h>=0) emit_cmp(s1h,s2h);
-        else emit_test(s1h,s1h);
-        emit_cmovne_reg(alt,addr);
-      }
-      if(s2l>=0) emit_cmp(s1l,s2l);
-      else emit_test(s1l,s1l);
-      emit_cmovne_reg(alt,addr);
-    }
   }
   if((opcode[i]&0x3f)==0x14) // BEQL
   {
-    if(s1h>=0) {
-      if(s2h>=0) emit_cmp(s1h,s2h);
-      else emit_test(s1h,s1h);
-      nottaken=(int)out;
-      emit_jne(0);
-    }
     if(s2l>=0) emit_cmp(s1l,s2l);
     else emit_test(s1l,s1l);
-    if(nottaken) set_jump_target(nottaken,(int)out);
-    nottaken=(int)out;
+    if(nottaken) set_jump_target(nottaken, out);
+    nottaken=out;
     emit_jne(0);
   }
   if((opcode[i]&0x3f)==0x15) // BNEL
   {
-    if(s1h>=0) {
-      if(s2h>=0) emit_cmp(s1h,s2h);
-      else emit_test(s1h,s1h);
-      taken=(int)out;
-      emit_jne(0);
-    }
     if(s2l>=0) emit_cmp(s1l,s2l);
     else emit_test(s1l,s1l);
-    nottaken=(int)out;
+    nottaken=out;
     emit_jeq(0);
-    if(taken) set_jump_target(taken,(int)out);
+    if(taken) set_jump_target(taken, out);
   }
   if((opcode[i]&0x3f)==6) // BLEZ
   {
     emit_mov2imm_compact(ba[i],alt,start+i*4+8,addr);
     emit_cmpimm(s1l,1);
-    if(s1h>=0) emit_mov(addr,ntaddr);
     emit_cmovl_reg(alt,addr);
-    if(s1h>=0) {
-      emit_test(s1h,s1h);
-      emit_cmovne_reg(ntaddr,addr);
-      emit_cmovs_reg(alt,addr);
-    }
   }
   if((opcode[i]&0x3f)==7) // BGTZ
   {
     emit_mov2imm_compact(ba[i],addr,start+i*4+8,ntaddr);
     emit_cmpimm(s1l,1);
-    if(s1h>=0) emit_mov(addr,alt);
     emit_cmovl_reg(ntaddr,addr);
-    if(s1h>=0) {
-      emit_test(s1h,s1h);
-      emit_cmovne_reg(alt,addr);
-      emit_cmovs_reg(ntaddr,addr);
-    }
   }
   if((opcode[i]&0x3f)==0x16) // BLEZL
   {
@@ -5963,19 +5821,19 @@ static void pagespan_assemble(int i,struct regstat *i_regs)
     if((source[i]&0x30000)==0x20000) // BC1FL
     {
       emit_testimm(s1l,0x800000);
-      nottaken=(int)out;
+      nottaken=out;
       emit_jne(0);
     }
     if((source[i]&0x30000)==0x30000) // BC1TL
     {
       emit_testimm(s1l,0x800000);
-      nottaken=(int)out;
+      nottaken=out;
       emit_jeq(0);
     }
   }
 
   assert(i_regs->regmap[HOST_CCREG]==CCREG);
-  wb_dirtys(regs[i].regmap,regs[i].is32,regs[i].dirty);
+  wb_dirtys(regs[i].regmap,regs[i].dirty);
   if(likely[i]||unconditional)
   {
     emit_movimm(ba[i],HOST_BTREG);
@@ -5989,27 +5847,27 @@ static void pagespan_assemble(int i,struct regstat *i_regs)
   int target_addr=start+i*4+5;
   void *stub=out;
   void *compiled_target_addr=check_addr(target_addr);
-  emit_extjump_ds((int)branch_addr,target_addr);
+  emit_extjump_ds(branch_addr, target_addr);
   if(compiled_target_addr) {
-    set_jump_target((int)branch_addr,(int)compiled_target_addr);
+    set_jump_target(branch_addr, compiled_target_addr);
     add_link(target_addr,stub);
   }
-  else set_jump_target((int)branch_addr,(int)stub);
+  else set_jump_target(branch_addr, stub);
   if(likely[i]) {
     // Not-taken path
-    set_jump_target((int)nottaken,(int)out);
-    wb_dirtys(regs[i].regmap,regs[i].is32,regs[i].dirty);
+    set_jump_target(nottaken, out);
+    wb_dirtys(regs[i].regmap,regs[i].dirty);
     void *branch_addr=out;
     emit_jmp(0);
     int target_addr=start+i*4+8;
     void *stub=out;
     void *compiled_target_addr=check_addr(target_addr);
-    emit_extjump_ds((int)branch_addr,target_addr);
+    emit_extjump_ds(branch_addr, target_addr);
     if(compiled_target_addr) {
-      set_jump_target((int)branch_addr,(int)compiled_target_addr);
+      set_jump_target(branch_addr, compiled_target_addr);
       add_link(target_addr,stub);
     }
-    else set_jump_target((int)branch_addr,(int)stub);
+    else set_jump_target(branch_addr, stub);
   }
 }
 
@@ -6025,14 +5883,13 @@ static void pagespan_ds()
   ll_add(jump_in+page,vaddr,(void *)out);
   assert(regs[0].regmap_entry[HOST_CCREG]==CCREG);
   if(regs[0].regmap[HOST_CCREG]!=CCREG)
-    wb_register(CCREG,regs[0].regmap_entry,regs[0].wasdirty,regs[0].was32);
+    wb_register(CCREG,regs[0].regmap_entry,regs[0].wasdirty);
   if(regs[0].regmap[HOST_BTREG]!=BTREG)
-    emit_writeword(HOST_BTREG,(int)&branch_target);
-  load_regs(regs[0].regmap_entry,regs[0].regmap,regs[0].was32,rs1[0],rs2[0]);
+    emit_writeword(HOST_BTREG,&branch_target);
+  load_regs(regs[0].regmap_entry,regs[0].regmap,rs1[0],rs2[0]);
   address_generation(0,&regs[0],regs[0].regmap_entry);
   if(itype[0]==STORE||itype[0]==STORELR||(opcode[0]&0x3b)==0x39||(opcode[0]&0x3b)==0x3a)
-    load_regs(regs[0].regmap_entry,regs[0].regmap,regs[0].was32,INVCP,INVCP);
-  cop1_usable=0;
+    load_regs(regs[0].regmap_entry,regs[0].regmap,INVCP,INVCP);
   is_delayslot=0;
   switch(itype[0]) {
     case ALU:
@@ -6063,12 +5920,6 @@ static void pagespan_ds()
       c2ls_assemble(0,&regs[0]);break;
     case C2OP:
       c2op_assemble(0,&regs[0]);break;
-    case FCONV:
-      fconv_assemble(0,&regs[0]);break;
-    case FLOAT:
-      float_assemble(0,&regs[0]);break;
-    case FCOMP:
-      fcomp_assemble(0,&regs[0]);break;
     case MULTDIV:
       multdiv_assemble(0,&regs[0]);break;
     case MOV:
@@ -6081,55 +5932,54 @@ static void pagespan_ds()
     case RJUMP:
     case CJUMP:
     case SJUMP:
-    case FJUMP:
       SysPrintf("Jump in the delay slot.  This is probably a bug.\n");
   }
   int btaddr=get_reg(regs[0].regmap,BTREG);
   if(btaddr<0) {
     btaddr=get_reg(regs[0].regmap,-1);
-    emit_readword((int)&branch_target,btaddr);
+    emit_readword(&branch_target,btaddr);
   }
   assert(btaddr!=HOST_CCREG);
   if(regs[0].regmap[HOST_CCREG]!=CCREG) emit_loadreg(CCREG,HOST_CCREG);
 #ifdef HOST_IMM8
+  host_tempreg_acquire();
   emit_movimm(start+4,HOST_TEMPREG);
   emit_cmp(btaddr,HOST_TEMPREG);
+  host_tempreg_release();
 #else
   emit_cmpimm(btaddr,start+4);
 #endif
-  int branch=(int)out;
+  void *branch = out;
   emit_jeq(0);
-  store_regs_bt(regs[0].regmap,regs[0].is32,regs[0].dirty,-1);
-  emit_jmp(jump_vaddr_reg[btaddr]);
-  set_jump_target(branch,(int)out);
-  store_regs_bt(regs[0].regmap,regs[0].is32,regs[0].dirty,start+4);
-  load_regs_bt(regs[0].regmap,regs[0].is32,regs[0].dirty,start+4);
+  store_regs_bt(regs[0].regmap,regs[0].dirty,-1);
+  do_jump_vaddr(btaddr);
+  set_jump_target(branch, out);
+  store_regs_bt(regs[0].regmap,regs[0].dirty,start+4);
+  load_regs_bt(regs[0].regmap,regs[0].dirty,start+4);
 }
 
 // Basic liveness analysis for MIPS registers
 void unneeded_registers(int istart,int iend,int r)
 {
   int i;
-  uint64_t u,uu,gte_u,b,bu,gte_bu;
-  uint64_t temp_u,temp_uu,temp_gte_u=0;
-  uint64_t tdep;
+  uint64_t u,gte_u,b,gte_b;
+  uint64_t temp_u,temp_gte_u=0;
   uint64_t gte_u_unknown=0;
-  if(new_dynarec_hacks&NDHACK_GTE_UNNEEDED)
+  if (HACK_ENABLED(NDHACK_GTE_UNNEEDED))
     gte_u_unknown=~0ll;
   if(iend==slen-1) {
-    u=1;uu=1;
+    u=1;
     gte_u=gte_u_unknown;
   }else{
-    u=unneeded_reg[iend+1];
-    uu=unneeded_reg_upper[iend+1];
-    u=1;uu=1;
+    //u=unneeded_reg[iend+1];
+    u=1;
     gte_u=gte_unneeded[iend+1];
   }
 
   for (i=iend;i>=istart;i--)
   {
     //printf("unneeded registers i=%d (%d,%d) r=%d\n",i,istart,iend,r);
-    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP)
     {
       // If subroutine call, flag return address as a possible branch target
       if(rt1[i]==31 && i<slen-2) bt[i+2]=1;
@@ -6138,39 +5988,12 @@ void unneeded_registers(int istart,int iend,int r)
       {
         // Branch out of this block, flush all regs
         u=1;
-        uu=1;
         gte_u=gte_u_unknown;
-        /* Hexagon hack
-        if(itype[i]==UJUMP&&rt1[i]==31)
-        {
-          uu=u=0x300C00F; // Discard at, v0-v1, t6-t9
-        }
-        if(itype[i]==RJUMP&&rs1[i]==31)
-        {
-          uu=u=0x300C0F3; // Discard at, a0-a3, t6-t9
-        }
-        if(start>0x80000400&&start<0x80000000+RAM_SIZE) {
-          if(itype[i]==UJUMP&&rt1[i]==31)
-          {
-            //uu=u=0x30300FF0FLL; // Discard at, v0-v1, t0-t9, lo, hi
-            uu=u=0x300FF0F; // Discard at, v0-v1, t0-t9
-          }
-          if(itype[i]==RJUMP&&rs1[i]==31)
-          {
-            //uu=u=0x30300FFF3LL; // Discard at, a0-a3, t0-t9, lo, hi
-            uu=u=0x300FFF3; // Discard at, a0-a3, t0-t9
-          }
-        }*/
         branch_unneeded_reg[i]=u;
-        branch_unneeded_reg_upper[i]=uu;
         // Merge in delay slot
-        tdep=(~uu>>rt1[i+1])&1;
         u|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
-        uu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
         u&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
-        uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
-        uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
-        u|=1;uu|=1;
+        u|=1;
         gte_u|=gte_rt[i+1];
         gte_u&=~gte_rs[i+1];
         // If branch is "likely" (and conditional)
@@ -6178,13 +6001,11 @@ void unneeded_registers(int istart,int iend,int r)
         if(likely[i]) {
           if(i<slen-1) {
             u&=unneeded_reg[i+2];
-            uu&=unneeded_reg_upper[i+2];
             gte_u&=gte_unneeded[i+2];
           }
           else
           {
             u=1;
-            uu=1;
             gte_u=gte_u_unknown;
           }
         }
@@ -6195,25 +6016,20 @@ void unneeded_registers(int istart,int iend,int r)
         bt[(ba[i]-start)>>2]=1;
         if(ba[i]<=start+i*4) {
           // Backward branch
-          if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
+          if(is_ujump(i))
           {
             // Unconditional branch
-            temp_u=1;temp_uu=1;
+            temp_u=1;
             temp_gte_u=0;
           } else {
             // Conditional branch (not taken case)
             temp_u=unneeded_reg[i+2];
-            temp_uu=unneeded_reg_upper[i+2];
             temp_gte_u&=gte_unneeded[i+2];
           }
           // Merge in delay slot
-          tdep=(~temp_uu>>rt1[i+1])&1;
           temp_u|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
-          temp_uu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
           temp_u&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
-          temp_uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
-          temp_uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
-          temp_u|=1;temp_uu|=1;
+          temp_u|=1;
           temp_gte_u|=gte_rt[i+1];
           temp_gte_u&=~gte_rs[i+1];
           // If branch is "likely" (and conditional)
@@ -6221,27 +6037,20 @@ void unneeded_registers(int istart,int iend,int r)
           if(likely[i]) {
             if(i<slen-1) {
               temp_u&=unneeded_reg[i+2];
-              temp_uu&=unneeded_reg_upper[i+2];
               temp_gte_u&=gte_unneeded[i+2];
             }
             else
             {
               temp_u=1;
-              temp_uu=1;
               temp_gte_u=gte_u_unknown;
             }
           }
-          tdep=(~temp_uu>>rt1[i])&1;
           temp_u|=(1LL<<rt1[i])|(1LL<<rt2[i]);
-          temp_uu|=(1LL<<rt1[i])|(1LL<<rt2[i]);
           temp_u&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
-          temp_uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
-          temp_uu&=~((tdep<<dep1[i])|(tdep<<dep2[i]));
-          temp_u|=1;temp_uu|=1;
+          temp_u|=1;
           temp_gte_u|=gte_rt[i];
           temp_gte_u&=~gte_rs[i];
           unneeded_reg[i]=temp_u;
-          unneeded_reg_upper[i]=temp_uu;
           gte_unneeded[i]=temp_gte_u;
           // Only go three levels deep.  This recursion can take an
           // excessive amount of time if there are a lot of nested loops.
@@ -6249,81 +6058,49 @@ void unneeded_registers(int istart,int iend,int r)
             unneeded_registers((ba[i]-start)>>2,i-1,r+1);
           }else{
             unneeded_reg[(ba[i]-start)>>2]=1;
-            unneeded_reg_upper[(ba[i]-start)>>2]=1;
             gte_unneeded[(ba[i]-start)>>2]=gte_u_unknown;
           }
         } /*else*/ if(1) {
-          if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
+          if (is_ujump(i))
           {
             // Unconditional branch
             u=unneeded_reg[(ba[i]-start)>>2];
-            uu=unneeded_reg_upper[(ba[i]-start)>>2];
             gte_u=gte_unneeded[(ba[i]-start)>>2];
             branch_unneeded_reg[i]=u;
-            branch_unneeded_reg_upper[i]=uu;
-        //u=1;
-        //uu=1;
-        //branch_unneeded_reg[i]=u;
-        //branch_unneeded_reg_upper[i]=uu;
             // Merge in delay slot
-            tdep=(~uu>>rt1[i+1])&1;
             u|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
-            uu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
             u&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
-            uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
-            uu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
-            u|=1;uu|=1;
+            u|=1;
             gte_u|=gte_rt[i+1];
             gte_u&=~gte_rs[i+1];
           } else {
             // Conditional branch
             b=unneeded_reg[(ba[i]-start)>>2];
-            bu=unneeded_reg_upper[(ba[i]-start)>>2];
-            gte_bu=gte_unneeded[(ba[i]-start)>>2];
+            gte_b=gte_unneeded[(ba[i]-start)>>2];
             branch_unneeded_reg[i]=b;
-            branch_unneeded_reg_upper[i]=bu;
-        //b=1;
-        //bu=1;
-        //branch_unneeded_reg[i]=b;
-        //branch_unneeded_reg_upper[i]=bu;
             // Branch delay slot
-            tdep=(~uu>>rt1[i+1])&1;
             b|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
-            bu|=(1LL<<rt1[i+1])|(1LL<<rt2[i+1]);
             b&=~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
-            bu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
-            bu&=~((tdep<<dep1[i+1])|(tdep<<dep2[i+1]));
-            b|=1;bu|=1;
-            gte_bu|=gte_rt[i+1];
-            gte_bu&=~gte_rs[i+1];
+            b|=1;
+            gte_b|=gte_rt[i+1];
+            gte_b&=~gte_rs[i+1];
             // If branch is "likely" then we skip the
             // delay slot on the fall-thru path
             if(likely[i]) {
               u=b;
-              uu=bu;
-              gte_u=gte_bu;
+              gte_u=gte_b;
               if(i<slen-1) {
                 u&=unneeded_reg[i+2];
-                uu&=unneeded_reg_upper[i+2];
                 gte_u&=gte_unneeded[i+2];
-        //u=1;
-        //uu=1;
               }
             } else {
               u&=b;
-              uu&=bu;
-              gte_u&=gte_bu;
-        //u=1;
-        //uu=1;
+              gte_u&=gte_b;
             }
             if(i<slen-1) {
               branch_unneeded_reg[i]&=unneeded_reg[i+2];
-              branch_unneeded_reg_upper[i]&=unneeded_reg_upper[i+2];
-        //branch_unneeded_reg[i]=1;
-        //branch_unneeded_reg_upper[i]=1;
             } else {
               branch_unneeded_reg[i]=1;
-              branch_unneeded_reg_upper[i]=1;
             }
           }
         }
@@ -6333,38 +6110,28 @@ void unneeded_registers(int istart,int iend,int r)
     {
       // SYSCALL instruction (software interrupt)
       u=1;
-      uu=1;
     }
     else if(itype[i]==COP0 && (source[i]&0x3f)==0x18)
     {
       // ERET instruction (return from interrupt)
       u=1;
-      uu=1;
     }
-    //u=uu=1; // DEBUG
-    tdep=(~uu>>rt1[i])&1;
+    //u=1; // DEBUG
     // Written registers are unneeded
     u|=1LL<<rt1[i];
     u|=1LL<<rt2[i];
-    uu|=1LL<<rt1[i];
-    uu|=1LL<<rt2[i];
     gte_u|=gte_rt[i];
     // Accessed registers are needed
     u&=~(1LL<<rs1[i]);
     u&=~(1LL<<rs2[i]);
-    uu&=~(1LL<<us1[i]);
-    uu&=~(1LL<<us2[i]);
     gte_u&=~gte_rs[i];
     if(gte_rs[i]&&rt1[i]&&(unneeded_reg[i+1]&(1ll<<rt1[i])))
       gte_u|=gte_rs[i]&gte_unneeded[i+1]; // MFC2/CFC2 to dead register, unneeded
     // Source-target dependencies
-    uu&=~(tdep<<dep1[i]);
-    uu&=~(tdep<<dep2[i]);
     // R0 is always unneeded
-    u|=1;uu|=1;
+    u|=1;
     // Save it
     unneeded_reg[i]=u;
-    unneeded_reg_upper[i]=uu;
     gte_unneeded[i]=gte_u;
     /*
     printf("ur (%d,%d) %x: ",istart,iend,start+i*4);
@@ -6377,19 +6144,8 @@ void unneeded_registers(int istart,int iend,int r)
         else printf(" r%d",r);
       }
     }
-    printf(" UU:");
-    for(r=1;r<=CCREG;r++) {
-      if(((unneeded_reg_upper[i]&~unneeded_reg[i])>>r)&1) {
-        if(r==HIREG) printf(" HI");
-        else if(r==LOREG) printf(" LO");
-        else printf(" r%d",r);
-      }
-    }
-    printf("\n");*/
-  }
-  for (i=iend;i>=istart;i--)
-  {
-    unneeded_reg_upper[i]=branch_unneeded_reg_upper[i]=-1LL;
+    printf("\n");
+    */
   }
 }
 
@@ -6410,12 +6166,12 @@ void clean_registers(int istart,int iend,int wr)
   }
   for (i=iend;i>=istart;i--)
   {
-    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP)
     {
       if(ba[i]<start || ba[i]>=(start+slen*4))
       {
         // Branch out of this block, flush all regs
-        if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
+        if (is_ujump(i))
         {
           // Unconditional branch
           will_dirty_i=0;
@@ -6495,7 +6251,7 @@ void clean_registers(int istart,int iend,int wr)
         // Internal branch
         if(ba[i]<=start+i*4) {
           // Backward branch
-          if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
+          if (is_ujump(i))
           {
             // Unconditional branch
             temp_will_dirty=0;
@@ -6592,7 +6348,7 @@ void clean_registers(int istart,int iend,int wr)
         }
         /*else*/ if(1)
         {
-          if(itype[i]==RJUMP||itype[i]==UJUMP||(source[i]>>16)==0x1000)
+          if (is_ujump(i))
           {
             // Unconditional branch
             will_dirty_i=0;
@@ -6730,7 +6486,7 @@ void clean_registers(int istart,int iend,int wr)
         if((regs[i].regmap[r]&63)==rt2[i]) wont_dirty_i|=1<<r;
         if(regs[i].regmap[r]==CCREG) wont_dirty_i|=1<<r;
         if(i>istart) {
-          if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=FJUMP)
+          if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP)
           {
             // Don't store a register immediately after writing it,
             // may prevent dual-issue.
@@ -6753,18 +6509,18 @@ void clean_registers(int istart,int iend,int wr)
       }
       printf("\n");*/
 
-      //if(i==istart||(itype[i-1]!=RJUMP&&itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=FJUMP)) {
+      //if(i==istart||(itype[i-1]!=RJUMP&&itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP)) {
         regs[i].dirty|=will_dirty_i;
         #ifndef DESTRUCTIVE_WRITEBACK
         regs[i].dirty&=wont_dirty_i;
-        if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+        if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP)
         {
-          if(i<iend-1&&itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000) {
+          if (i < iend-1 && !is_ujump(i)) {
             for(r=0;r<HOST_REGS;r++) {
               if(r!=EXCLUDE_REG) {
                 if(regs[i].regmap[r]==regmap_pre[i+2][r]) {
                   regs[i+2].wasdirty&=wont_dirty_i|~(1<<r);
-                }else {/*printf("i: %x (%d) mismatch(+2): %d\n",start+i*4,i,r);/*assert(!((wont_dirty_i>>r)&1));*/}
+                }else {/*printf("i: %x (%d) mismatch(+2): %d\n",start+i*4,i,r);assert(!((wont_dirty_i>>r)&1));*/}
               }
             }
           }
@@ -6776,7 +6532,7 @@ void clean_registers(int istart,int iend,int wr)
               if(r!=EXCLUDE_REG) {
                 if(regs[i].regmap[r]==regmap_pre[i+1][r]) {
                   regs[i+1].wasdirty&=wont_dirty_i|~(1<<r);
-                }else {/*printf("i: %x (%d) mismatch(+1): %d\n",start+i*4,i,r);/*assert(!((wont_dirty_i>>r)&1));*/}
+                }else {/*printf("i: %x (%d) mismatch(+1): %d\n",start+i*4,i,r);assert(!((wont_dirty_i>>r)&1));*/}
               }
             }
           }
@@ -6819,7 +6575,7 @@ void clean_registers(int istart,int iend,int wr)
             wont_dirty_i|=((unneeded_reg[i]>>(regmap_pre[i][r]&63))&1)<<r;
           } else {
             wont_dirty_i|=1<<r;
-            /*printf("i: %x (%d) mismatch: %d\n",start+i*4,i,r);/*assert(!((will_dirty>>r)&1));*/
+            /*printf("i: %x (%d) mismatch: %d\n",start+i*4,i,r);assert(!((will_dirty>>r)&1));*/
           }
         }
       }
@@ -6839,8 +6595,6 @@ void disassemble_inst(int i)
         printf (" %x: %s r%d,r%d,%8x\n",start+i*4,insn[i],rs1[i],rs2[i],i?start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14):*ba);break;
       case SJUMP:
         printf (" %x: %s r%d,%8x\n",start+i*4,insn[i],rs1[i],start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14));break;
-      case FJUMP:
-        printf (" %x: %s %8x\n",start+i*4,insn[i],ba[i]);break;
       case RJUMP:
         if (opcode[i]==0x9&&rt1[i]!=31)
           printf (" %x: %s r%d,r%d\n",start+i*4,insn[i],rt1[i],rs1[i]);
@@ -6922,32 +6676,46 @@ static void disassemble_inst(int i) {}
 
 #define DRC_TEST_VAL 0x74657374
 
-static int new_dynarec_test(void)
+static void new_dynarec_test(void)
 {
-  int (*testfunc)(void) = (void *)out;
-  int ret;
-  emit_movimm(DRC_TEST_VAL,0); // test
-  emit_jmpreg(14);
-  literal_pool(0);
-#ifdef __arm__
-  __clear_cache((void *)testfunc, out);
-#endif
-  SysPrintf("testing if we can run recompiled code..\n");
-  ret = testfunc();
-  if (ret == DRC_TEST_VAL)
+  int (*testfunc)(void);
+  void *beginning;
+  int ret[2];
+  size_t i;
+
+  // check structure linkage
+  if ((u_char *)rcnts - (u_char *)&psxRegs != sizeof(psxRegs))
+  {
+    SysPrintf("linkage_arm* miscompilation/breakage detected.\n");
+  }
+
+  SysPrintf("testing if we can run recompiled code...\n");
+  ((volatile u_int *)out)[0]++; // make cache dirty
+
+  for (i = 0; i < ARRAY_SIZE(ret); i++) {
+    out = ndrc->translation_cache;
+    beginning = start_block();
+    emit_movimm(DRC_TEST_VAL + i, 0); // test
+    emit_ret();
+    literal_pool(0);
+    end_block(beginning);
+    testfunc = beginning;
+    ret[i] = testfunc();
+  }
+
+  if (ret[0] == DRC_TEST_VAL && ret[1] == DRC_TEST_VAL + 1)
     SysPrintf("test passed.\n");
   else
-    SysPrintf("test failed: %08x\n", ret);
-  out=(u_char *)BASE_ADDR;
-  return ret == DRC_TEST_VAL;
+    SysPrintf("test failed, will likely crash soon (r=%08x %08x)\n", ret[0], ret[1]);
+  out = ndrc->translation_cache;
 }
 
 // clear the state completely, instead of just marking
 // things invalid like invalidate_all_pages() does
-void new_dynarec_clear_full()
+void new_dynarec_clear_full(void)
 {
   int n;
-  out=(u_char *)BASE_ADDR;
+  out = ndrc->translation_cache;
   memset(invalid_code,1,sizeof(invalid_code));
   memset(hash_table,0xff,sizeof(hash_table));
   memset(mini_ht,-1,sizeof(mini_ht));
@@ -6965,23 +6733,41 @@ void new_dynarec_clear_full()
   for(n=0;n<4096;n++) ll_clear(jump_dirty+n);
 }
 
-void new_dynarec_init()
+void new_dynarec_init(void)
 {
   SysPrintf("Init new dynarec\n");
-  out=(u_char *)BASE_ADDR;
-#if BASE_ADDR_FIXED
-  if (mmap (out, 1<<TARGET_SIZE_2,
+
+#ifdef BASE_ADDR_DYNAMIC
+  #ifdef VITA
+  sceBlock = sceKernelAllocMemBlockForVM("code", 1 << TARGET_SIZE_2);
+  if (sceBlock < 0)
+    SysPrintf("sceKernelAllocMemBlockForVM failed\n");
+  int ret = sceKernelGetMemBlockBase(sceBlock, (void **)&ndrc);
+  if (ret < 0)
+    SysPrintf("sceKernelGetMemBlockBase failed\n");
+  #else
+  uintptr_t desired_addr = 0;
+  #ifdef __ELF__
+  extern char _end;
+  desired_addr = ((uintptr_t)&_end + 0xffffff) & ~0xffffffl;
+  #endif
+  ndrc = mmap((void *)desired_addr, sizeof(*ndrc),
             PROT_READ | PROT_WRITE | PROT_EXEC,
-            MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
-            -1, 0) <= 0) {
+            MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+  if (ndrc == MAP_FAILED) {
     SysPrintf("mmap() failed: %s\n", strerror(errno));
+    abort();
   }
+  #endif
 #else
+  #ifndef NO_WRITE_EXEC
   // not all systems allow execute in data segment by default
-  if (mprotect(out, 1<<TARGET_SIZE_2, PROT_READ | PROT_WRITE | PROT_EXEC) != 0)
+  if (mprotect(ndrc, sizeof(ndrc->translation_cache) + sizeof(ndrc->tramp.ops),
+               PROT_READ | PROT_WRITE | PROT_EXEC) != 0)
     SysPrintf("mprotect() failed: %s\n", strerror(errno));
+  #endif
 #endif
-  int n;
+  out = ndrc->translation_cache;
   cycle_multiplier=200;
   new_dynarec_clear_full();
 #ifdef HOST_IMM8
@@ -6991,18 +6777,24 @@ void new_dynarec_init()
   arch_init();
   new_dynarec_test();
 #ifndef RAM_FIXED
-  ram_offset=(u_int)rdram-0x80000000;
+  ram_offset=(uintptr_t)rdram-0x80000000;
 #endif
   if (ram_offset!=0)
     SysPrintf("warning: RAM is not directly mapped, performance will suffer\n");
 }
 
-void new_dynarec_cleanup()
+void new_dynarec_cleanup(void)
 {
   int n;
-  #if BASE_ADDR_FIXED
-  if (munmap ((void *)BASE_ADDR, 1<<TARGET_SIZE_2) < 0) {SysPrintf("munmap() failed\n");}
+#ifdef BASE_ADDR_DYNAMIC
+  #ifdef VITA
+  sceKernelFreeMemBlock(sceBlock);
+  sceBlock = -1;
+  #else
+  if (munmap(ndrc, sizeof(*ndrc)) < 0)
+    SysPrintf("munmap() failed\n");
   #endif
+#endif
   for(n=0;n<4096;n++) ll_clear(jump_in+n);
   for(n=0;n<4096;n++) ll_clear(jump_out+n);
   for(n=0;n<4096;n++) ll_clear(jump_dirty+n);
@@ -7013,23 +6805,33 @@ void new_dynarec_cleanup()
 
 static u_int *get_source_start(u_int addr, u_int *limit)
 {
+  if (!HACK_ENABLED(NDHACK_OVERRIDE_CYCLE_M))
+    cycle_multiplier_override = 0;
+
   if (addr < 0x00200000 ||
-    (0xa0000000 <= addr && addr < 0xa0200000)) {
+    (0xa0000000 <= addr && addr < 0xa0200000))
+  {
     // used for BIOS calls mostly?
     *limit = (addr&0xa0000000)|0x00200000;
-    return (u_int *)((u_int)rdram + (addr&0x1fffff));
+    return (u_int *)(rdram + (addr&0x1fffff));
   }
   else if (!Config.HLE && (
     /* (0x9fc00000 <= addr && addr < 0x9fc80000) ||*/
-    (0xbfc00000 <= addr && addr < 0xbfc80000))) {
-    // BIOS
+    (0xbfc00000 <= addr && addr < 0xbfc80000)))
+  {
+    // BIOS. The multiplier should be much higher as it's uncached 8bit mem,
+    // but timings in PCSX are too tied to the interpreter's BIAS
+    if (!HACK_ENABLED(NDHACK_OVERRIDE_CYCLE_M))
+      cycle_multiplier_override = 200;
+
     *limit = (addr & 0xfff00000) | 0x80000;
-    return (u_int *)((u_int)psxR + (addr&0x7ffff));
+    return (u_int *)((u_char *)psxR + (addr&0x7ffff));
   }
   else if (addr >= 0x80000000 && addr < 0x80000000+RAM_SIZE) {
     *limit = (addr & 0x80600000) + 0x00200000;
-    return (u_int *)((u_int)rdram + (addr&0x1fffff));
+    return (u_int *)(rdram + (addr&0x1fffff));
   }
+  return NULL;
 }
 
 static u_int scan_for_ret(u_int addr)
@@ -7047,6 +6849,7 @@ static u_int scan_for_ret(u_int addr)
     if (*mem == 0x03e00008) // jr $ra
       return addr + 8;
   }
+  return addr;
 }
 
 struct savestate_block {
@@ -7070,7 +6873,7 @@ int new_dynarec_save_blocks(void *save, int size)
   u_int addr;
 
   o = 0;
-  for (p = 0; p < sizeof(jump_in) / sizeof(jump_in[0]); p++) {
+  for (p = 0; p < ARRAY_SIZE(jump_in); p++) {
     bcnt = 0;
     for (head = jump_in[p]; head != NULL; head = head->next) {
       tmp_blocks[bcnt].addr = head->vaddr;
@@ -7131,22 +6934,16 @@ void new_dynarec_load_blocks(const void *save, int size)
   memcpy(&psxRegs.GPR, regs_save, sizeof(regs_save));
 }
 
-int new_recompile_block(int addr)
+int new_recompile_block(u_int addr)
 {
   u_int pagelimit = 0;
   u_int state_rflags = 0;
   int i;
 
-  assem_debug("NOTCOMPILED: addr = %x -> %x\n", (int)addr, (int)out);
-  //printf("NOTCOMPILED: addr = %x -> %x\n", (int)addr, (int)out);
+  assem_debug("NOTCOMPILED: addr = %x -> %p\n", addr, out);
   //printf("TRACE: count=%d next=%d (compile %x)\n",Count,next_interupt,addr);
   //if(debug)
-  //printf("TRACE: count=%d next=%d (checksum %x)\n",Count,next_interupt,mchecksum());
   //printf("fpu mapping=%x enabled=%x\n",(Status & 0x04000000)>>26,(Status & 0x20000000)>>29);
-  /*if(Count>=312978186) {
-    rlist();
-  }*/
-  //rlist();
 
   // this is just for speculation
   for (i = 1; i < 32; i++) {
@@ -7155,21 +6952,20 @@ int new_recompile_block(int addr)
   }
 
   start = (u_int)addr&~3;
-  //assert(((u_int)addr&1)==0);
+  //assert(((u_int)addr&1)==0); // start-in-delay-slot flag
   new_dynarec_did_compile=1;
   if (Config.HLE && start == 0x80001000) // hlecall
   {
     // XXX: is this enough? Maybe check hleSoftCall?
-    u_int beginning=(u_int)out;
+    void *beginning=start_block();
     u_int page=get_page(start);
+
     invalid_code[start>>12]=0;
     emit_movimm(start,0);
-    emit_writeword(0,(int)&pcaddr);
-    emit_jmp((int)new_dyna_leave);
+    emit_writeword(0,&pcaddr);
+    emit_far_jump(new_dyna_leave);
     literal_pool(0);
-#ifdef __arm__
-    __clear_cache((void *)beginning,out);
-#endif
+    end_block(beginning);
     ll_add_flags(jump_in+page,start,state_rflags,(void *)beginning);
     return 0;
   }
@@ -7177,7 +6973,7 @@ int new_recompile_block(int addr)
   source = get_source_start(start, &pagelimit);
   if (source == NULL) {
     SysPrintf("Compile at bogus memory address: %08x\n", addr);
-    exit(1);
+    abort();
   }
 
   /* Pass 1: disassemble */
@@ -7304,133 +7100,14 @@ int new_recompile_block(int addr)
         switch(op2)
         {
           case 0x00: strcpy(insn[i],"MFC0"); type=COP0; break;
+          case 0x02: strcpy(insn[i],"CFC0"); type=COP0; break;
           case 0x04: strcpy(insn[i],"MTC0"); type=COP0; break;
-          case 0x10: strcpy(insn[i],"tlb"); type=NI;
-          switch(source[i]&0x3f)
-          {
-            case 0x01: strcpy(insn[i],"TLBR"); type=COP0; break;
-            case 0x02: strcpy(insn[i],"TLBWI"); type=COP0; break;
-            case 0x06: strcpy(insn[i],"TLBWR"); type=COP0; break;
-            case 0x08: strcpy(insn[i],"TLBP"); type=COP0; break;
-            case 0x10: strcpy(insn[i],"RFE"); type=COP0; break;
-            //case 0x18: strcpy(insn[i],"ERET"); type=COP0; break;
-          }
+          case 0x06: strcpy(insn[i],"CTC0"); type=COP0; break;
+          case 0x10: strcpy(insn[i],"RFE"); type=COP0; break;
         }
         break;
-      case 0x11: strcpy(insn[i],"cop1"); type=NI;
+      case 0x11: strcpy(insn[i],"cop1"); type=COP1;
         op2=(source[i]>>21)&0x1f;
-        switch(op2)
-        {
-          case 0x00: strcpy(insn[i],"MFC1"); type=COP1; break;
-          case 0x01: strcpy(insn[i],"DMFC1"); type=COP1; break;
-          case 0x02: strcpy(insn[i],"CFC1"); type=COP1; break;
-          case 0x04: strcpy(insn[i],"MTC1"); type=COP1; break;
-          case 0x05: strcpy(insn[i],"DMTC1"); type=COP1; break;
-          case 0x06: strcpy(insn[i],"CTC1"); type=COP1; break;
-          case 0x08: strcpy(insn[i],"BC1"); type=FJUMP;
-          switch((source[i]>>16)&0x3)
-          {
-            case 0x00: strcpy(insn[i],"BC1F"); break;
-            case 0x01: strcpy(insn[i],"BC1T"); break;
-            case 0x02: strcpy(insn[i],"BC1FL"); break;
-            case 0x03: strcpy(insn[i],"BC1TL"); break;
-          }
-          break;
-          case 0x10: strcpy(insn[i],"C1.S"); type=NI;
-          switch(source[i]&0x3f)
-          {
-            case 0x00: strcpy(insn[i],"ADD.S"); type=FLOAT; break;
-            case 0x01: strcpy(insn[i],"SUB.S"); type=FLOAT; break;
-            case 0x02: strcpy(insn[i],"MUL.S"); type=FLOAT; break;
-            case 0x03: strcpy(insn[i],"DIV.S"); type=FLOAT; break;
-            case 0x04: strcpy(insn[i],"SQRT.S"); type=FLOAT; break;
-            case 0x05: strcpy(insn[i],"ABS.S"); type=FLOAT; break;
-            case 0x06: strcpy(insn[i],"MOV.S"); type=FLOAT; break;
-            case 0x07: strcpy(insn[i],"NEG.S"); type=FLOAT; break;
-            case 0x08: strcpy(insn[i],"ROUND.L.S"); type=FCONV; break;
-            case 0x09: strcpy(insn[i],"TRUNC.L.S"); type=FCONV; break;
-            case 0x0A: strcpy(insn[i],"CEIL.L.S"); type=FCONV; break;
-            case 0x0B: strcpy(insn[i],"FLOOR.L.S"); type=FCONV; break;
-            case 0x0C: strcpy(insn[i],"ROUND.W.S"); type=FCONV; break;
-            case 0x0D: strcpy(insn[i],"TRUNC.W.S"); type=FCONV; break;
-            case 0x0E: strcpy(insn[i],"CEIL.W.S"); type=FCONV; break;
-            case 0x0F: strcpy(insn[i],"FLOOR.W.S"); type=FCONV; break;
-            case 0x21: strcpy(insn[i],"CVT.D.S"); type=FCONV; break;
-            case 0x24: strcpy(insn[i],"CVT.W.S"); type=FCONV; break;
-            case 0x25: strcpy(insn[i],"CVT.L.S"); type=FCONV; break;
-            case 0x30: strcpy(insn[i],"C.F.S"); type=FCOMP; break;
-            case 0x31: strcpy(insn[i],"C.UN.S"); type=FCOMP; break;
-            case 0x32: strcpy(insn[i],"C.EQ.S"); type=FCOMP; break;
-            case 0x33: strcpy(insn[i],"C.UEQ.S"); type=FCOMP; break;
-            case 0x34: strcpy(insn[i],"C.OLT.S"); type=FCOMP; break;
-            case 0x35: strcpy(insn[i],"C.ULT.S"); type=FCOMP; break;
-            case 0x36: strcpy(insn[i],"C.OLE.S"); type=FCOMP; break;
-            case 0x37: strcpy(insn[i],"C.ULE.S"); type=FCOMP; break;
-            case 0x38: strcpy(insn[i],"C.SF.S"); type=FCOMP; break;
-            case 0x39: strcpy(insn[i],"C.NGLE.S"); type=FCOMP; break;
-            case 0x3A: strcpy(insn[i],"C.SEQ.S"); type=FCOMP; break;
-            case 0x3B: strcpy(insn[i],"C.NGL.S"); type=FCOMP; break;
-            case 0x3C: strcpy(insn[i],"C.LT.S"); type=FCOMP; break;
-            case 0x3D: strcpy(insn[i],"C.NGE.S"); type=FCOMP; break;
-            case 0x3E: strcpy(insn[i],"C.LE.S"); type=FCOMP; break;
-            case 0x3F: strcpy(insn[i],"C.NGT.S"); type=FCOMP; break;
-          }
-          break;
-          case 0x11: strcpy(insn[i],"C1.D"); type=NI;
-          switch(source[i]&0x3f)
-          {
-            case 0x00: strcpy(insn[i],"ADD.D"); type=FLOAT; break;
-            case 0x01: strcpy(insn[i],"SUB.D"); type=FLOAT; break;
-            case 0x02: strcpy(insn[i],"MUL.D"); type=FLOAT; break;
-            case 0x03: strcpy(insn[i],"DIV.D"); type=FLOAT; break;
-            case 0x04: strcpy(insn[i],"SQRT.D"); type=FLOAT; break;
-            case 0x05: strcpy(insn[i],"ABS.D"); type=FLOAT; break;
-            case 0x06: strcpy(insn[i],"MOV.D"); type=FLOAT; break;
-            case 0x07: strcpy(insn[i],"NEG.D"); type=FLOAT; break;
-            case 0x08: strcpy(insn[i],"ROUND.L.D"); type=FCONV; break;
-            case 0x09: strcpy(insn[i],"TRUNC.L.D"); type=FCONV; break;
-            case 0x0A: strcpy(insn[i],"CEIL.L.D"); type=FCONV; break;
-            case 0x0B: strcpy(insn[i],"FLOOR.L.D"); type=FCONV; break;
-            case 0x0C: strcpy(insn[i],"ROUND.W.D"); type=FCONV; break;
-            case 0x0D: strcpy(insn[i],"TRUNC.W.D"); type=FCONV; break;
-            case 0x0E: strcpy(insn[i],"CEIL.W.D"); type=FCONV; break;
-            case 0x0F: strcpy(insn[i],"FLOOR.W.D"); type=FCONV; break;
-            case 0x20: strcpy(insn[i],"CVT.S.D"); type=FCONV; break;
-            case 0x24: strcpy(insn[i],"CVT.W.D"); type=FCONV; break;
-            case 0x25: strcpy(insn[i],"CVT.L.D"); type=FCONV; break;
-            case 0x30: strcpy(insn[i],"C.F.D"); type=FCOMP; break;
-            case 0x31: strcpy(insn[i],"C.UN.D"); type=FCOMP; break;
-            case 0x32: strcpy(insn[i],"C.EQ.D"); type=FCOMP; break;
-            case 0x33: strcpy(insn[i],"C.UEQ.D"); type=FCOMP; break;
-            case 0x34: strcpy(insn[i],"C.OLT.D"); type=FCOMP; break;
-            case 0x35: strcpy(insn[i],"C.ULT.D"); type=FCOMP; break;
-            case 0x36: strcpy(insn[i],"C.OLE.D"); type=FCOMP; break;
-            case 0x37: strcpy(insn[i],"C.ULE.D"); type=FCOMP; break;
-            case 0x38: strcpy(insn[i],"C.SF.D"); type=FCOMP; break;
-            case 0x39: strcpy(insn[i],"C.NGLE.D"); type=FCOMP; break;
-            case 0x3A: strcpy(insn[i],"C.SEQ.D"); type=FCOMP; break;
-            case 0x3B: strcpy(insn[i],"C.NGL.D"); type=FCOMP; break;
-            case 0x3C: strcpy(insn[i],"C.LT.D"); type=FCOMP; break;
-            case 0x3D: strcpy(insn[i],"C.NGE.D"); type=FCOMP; break;
-            case 0x3E: strcpy(insn[i],"C.LE.D"); type=FCOMP; break;
-            case 0x3F: strcpy(insn[i],"C.NGT.D"); type=FCOMP; break;
-          }
-          break;
-          case 0x14: strcpy(insn[i],"C1.W"); type=NI;
-          switch(source[i]&0x3f)
-          {
-            case 0x20: strcpy(insn[i],"CVT.S.W"); type=FCONV; break;
-            case 0x21: strcpy(insn[i],"CVT.D.W"); type=FCONV; break;
-          }
-          break;
-          case 0x15: strcpy(insn[i],"C1.L"); type=NI;
-          switch(source[i]&0x3f)
-          {
-            case 0x20: strcpy(insn[i],"CVT.S.L"); type=FCONV; break;
-            case 0x21: strcpy(insn[i],"CVT.D.L"); type=FCONV; break;
-          }
-          break;
-        }
         break;
 #if 0
       case 0x14: strcpy(insn[i],"BEQL"); type=CJUMP; break;
@@ -7478,7 +7155,7 @@ int new_recompile_block(int addr)
 #endif
       case 0x12: strcpy(insn[i],"COP2"); type=NI;
         op2=(source[i]>>21)&0x1f;
-        //if (op2 & 0x10) {
+        //if (op2 & 0x10)
         if (source[i]&0x3f) { // use this hack to support old savestates with patched gte insns
           if (gte_handlers[source[i]&0x3f]!=NULL) {
             if (gte_regnames[source[i]&0x3f]!=NULL)
@@ -7507,8 +7184,6 @@ int new_recompile_block(int addr)
     opcode2[i]=op2;
     /* Get registers/immediates */
     lt1[i]=0;
-    us1[i]=0;
-    us2[i]=0;
     dep1[i]=0;
     dep2[i]=0;
     gte_rs[i]=gte_rt[i]=0;
@@ -7527,7 +7202,6 @@ int new_recompile_block(int addr)
         rt1[i]=0;
         rt2[i]=0;
         imm[i]=(short)source[i];
-        if(op==0x2c||op==0x2d||op==0x3f) us1[i]=rs2[i]; // 64-bit SDL/SDR/SD
         break;
       case LOADLR:
         // LWL/LWR only load part of the register,
@@ -7537,7 +7211,6 @@ int new_recompile_block(int addr)
         rt1[i]=(source[i]>>16)&0x1f;
         rt2[i]=0;
         imm[i]=(short)source[i];
-        if(op==0x1a||op==0x1b) us1[i]=rs2[i]; // LDR/LDL
         if(op==0x26) dep1[i]=rt1[i]; // LWR
         break;
       case IMM16:
@@ -7551,8 +7224,6 @@ int new_recompile_block(int addr)
         }else{
           imm[i]=(short)source[i];
         }
-        if(op==0x18||op==0x19) us1[i]=rs1[i]; // DADDI/DADDIU
-        if(op==0x0a||op==0x0b) us1[i]=rs1[i]; // SLTI/SLTIU
         if(op==0x0d||op==0x0e) dep1[i]=rs1[i]; // ORI/XORI
         break;
       case UJUMP:
@@ -7585,8 +7256,6 @@ int new_recompile_block(int addr)
         if(op&2) { // BGTZ/BLEZ
           rs2[i]=0;
         }
-        us1[i]=rs1[i];
-        us2[i]=rs2[i];
         likely[i]=op>>4;
         break;
       case SJUMP:
@@ -7594,29 +7263,18 @@ int new_recompile_block(int addr)
         rs2[i]=CCREG;
         rt1[i]=0;
         rt2[i]=0;
-        us1[i]=rs1[i];
         if(op2&0x10) { // BxxAL
           rt1[i]=31;
           // NOTE: If the branch is not taken, r31 is still overwritten
         }
         likely[i]=(op2&2)>>1;
         break;
-      case FJUMP:
-        rs1[i]=FSREG;
-        rs2[i]=CSREG;
-        rt1[i]=0;
-        rt2[i]=0;
-        likely[i]=((source[i])>>17)&1;
-        break;
       case ALU:
         rs1[i]=(source[i]>>21)&0x1f; // source
         rs2[i]=(source[i]>>16)&0x1f; // subtract amount
         rt1[i]=(source[i]>>11)&0x1f; // destination
         rt2[i]=0;
-        if(op2==0x2a||op2==0x2b) { // SLT/SLTU
-          us1[i]=rs1[i];us2[i]=rs2[i];
-        }
-        else if(op2>=0x24&&op2<=0x27) { // AND/OR/XOR/NOR
+        if(op2>=0x24&&op2<=0x27) { // AND/OR/XOR/NOR
           dep1[i]=rs1[i];dep2[i]=rs2[i];
         }
         else if(op2>=0x2c&&op2<=0x2f) { // DADD/DSUB
@@ -7628,9 +7286,6 @@ int new_recompile_block(int addr)
         rs2[i]=(source[i]>>16)&0x1f; // divisor
         rt1[i]=HIREG;
         rt2[i]=LOREG;
-        if (op2>=0x1c&&op2<=0x1f) { // DMULT/DMULTU/DDIV/DDIVU
-          us1[i]=rs1[i];us2[i]=rs2[i];
-        }
         break;
       case MOV:
         rs1[i]=0;
@@ -7650,8 +7305,6 @@ int new_recompile_block(int addr)
         rs2[i]=(source[i]>>21)&0x1f; // shift amount
         rt1[i]=(source[i]>>11)&0x1f; // destination
         rt2[i]=0;
-        // DSLLV/DSRLV/DSRAV are 64-bit
-        if(op2>=0x14&&op2<=0x17) us1[i]=rs1[i];
         break;
       case SHIFTIMM:
         rs1[i]=(source[i]>>16)&0x1f;
@@ -7661,16 +7314,14 @@ int new_recompile_block(int addr)
         imm[i]=(source[i]>>6)&0x1f;
         // DSxx32 instructions
         if(op2>=0x3c) imm[i]|=0x20;
-        // DSLL/DSRL/DSRA/DSRA32/DSRL32 but not DSLL32 require 64-bit source
-        if(op2>=0x38&&op2!=0x3c) us1[i]=rs1[i];
         break;
       case COP0:
         rs1[i]=0;
         rs2[i]=0;
         rt1[i]=0;
         rt2[i]=0;
-        if(op2==0) rt1[i]=(source[i]>>16)&0x1F; // MFC0
-        if(op2==4) rs1[i]=(source[i]>>16)&0x1F; // MTC0
+        if(op2==0||op2==2) rt1[i]=(source[i]>>16)&0x1F; // MFC0/CFC0
+        if(op2==4||op2==6) rs1[i]=(source[i]>>16)&0x1F; // MTC0/CTC0
         if(op2==4&&((source[i]>>11)&0x1f)==12) rt2[i]=CSREG; // Status
         if(op2==16) if((source[i]&0x3f)==0x18) rs2[i]=CCREG; // ERET
         break;
@@ -7681,7 +7332,6 @@ int new_recompile_block(int addr)
         rt2[i]=0;
         if(op2<3) rt1[i]=(source[i]>>16)&0x1F; // MFC1/DMFC1/CFC1
         if(op2>3) rs1[i]=(source[i]>>16)&0x1F; // MTC1/DMTC1/CTC1
-        if(op2==5) us1[i]=rs1[i]; // DMTC1
         rs2[i]=CSREG;
         break;
       case COP2:
@@ -7732,19 +7382,6 @@ int new_recompile_block(int addr)
           else gte_rs[i]|=3ll<<(v*2);
         }
         break;
-      case FLOAT:
-      case FCONV:
-        rs1[i]=0;
-        rs2[i]=CSREG;
-        rt1[i]=0;
-        rt2[i]=0;
-        break;
-      case FCOMP:
-        rs1[i]=FSREG;
-        rs2[i]=CSREG;
-        rt1[i]=FSREG;
-        rt2[i]=0;
-        break;
       case SYSCALL:
       case HLECALL:
       case INTCALL:
@@ -7766,13 +7403,13 @@ int new_recompile_block(int addr)
       ba[i]=start+i*4+8; // Ignore never taken branch
     else if(type==SJUMP&&rs1[i]==0&&!(op2&1))
       ba[i]=start+i*4+8; // Ignore never taken branch
-    else if(type==CJUMP||type==SJUMP||type==FJUMP)
+    else if(type==CJUMP||type==SJUMP)
       ba[i]=start+i*4+4+((signed int)((unsigned int)source[i]<<16)>>14);
     else ba[i]=-1;
-    if(i>0&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP)) {
+    if (i > 0 && is_jump(i-1)) {
       int do_in_intrp=0;
       // branch in delay slot?
-      if(type==RJUMP||type==UJUMP||type==CJUMP||type==SJUMP||type==FJUMP) {
+      if(type==RJUMP||type==UJUMP||type==CJUMP||type==SJUMP) {
         // don't handle first branch and call interpreter if it's hit
         SysPrintf("branch in delay slot @%08x (%08x)\n", addr + i*4, addr);
         do_in_intrp=1;
@@ -7787,7 +7424,7 @@ int new_recompile_block(int addr)
           bt[t+1]=1; // expected return from interpreter
         }
         else if(i>=2&&rt1[i-2]==2&&rt1[i]==2&&rs1[i]!=2&&rs2[i]!=2&&rs1[i-1]!=2&&rs2[i-1]!=2&&
-              !(i>=3&&(itype[i-3]==RJUMP||itype[i-3]==UJUMP||itype[i-3]==CJUMP||itype[i-3]==SJUMP))) {
+              !(i>=3&&is_jump(i-3))) {
           // v0 overwrite like this is a sign of trouble, bail out
           SysPrintf("v0 overwrite @%08x (%08x)\n", addr + i*4, addr);
           do_in_intrp=1;
@@ -7803,7 +7440,7 @@ int new_recompile_block(int addr)
       }
     }
     /* Is this the end of the block? */
-    if(i>0&&(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)) {
+    if (i > 0 && is_ujump(i-1)) {
       if(rt1[i-1]==0) { // Continue past subroutine call (JAL)
         done=2;
       }
@@ -7839,7 +7476,7 @@ int new_recompile_block(int addr)
     }
   }
   slen=i;
-  if(itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==RJUMP||itype[i-1]==FJUMP) {
+  if(itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==RJUMP) {
     if(start+i*4==pagelimit) {
       itype[i-1]=SPAN;
     }
@@ -7853,10 +7490,8 @@ int new_recompile_block(int addr)
   /* Pass 3 - Register allocation */
 
   struct regstat current; // Current register allocations/status
-  current.is32=1;
   current.dirty=0;
   current.u=unneeded_reg[0];
-  current.uu=unneeded_reg_upper[0];
   clear_all_regs(current.regmap);
   alloc_reg(&current,0,CCREG);
   dirty_reg(&current,CCREG);
@@ -7873,7 +7508,6 @@ int new_recompile_block(int addr)
     bt[1]=1;
     ds=1;
     unneeded_reg[0]=1;
-    unneeded_reg_upper[0]=1;
     current.regmap[HOST_BTREG]=BTREG;
   }
 
@@ -7890,53 +7524,24 @@ int new_recompile_block(int addr)
       current.isconst=0;
       current.waswritten=0;
     }
-    if(i>1)
-    {
-      if((opcode[i-2]&0x2f)==0x05) // BNE/BNEL
-      {
-        if(rs1[i-2]==0||rs2[i-2]==0)
-        {
-          if(rs1[i-2]) {
-            current.is32|=1LL<<rs1[i-2];
-            int hr=get_reg(current.regmap,rs1[i-2]|64);
-            if(hr>=0) current.regmap[hr]=-1;
-          }
-          if(rs2[i-2]) {
-            current.is32|=1LL<<rs2[i-2];
-            int hr=get_reg(current.regmap,rs2[i-2]|64);
-            if(hr>=0) current.regmap[hr]=-1;
-          }
-        }
-      }
-    }
-    current.is32=-1LL;
 
     memcpy(regmap_pre[i],current.regmap,sizeof(current.regmap));
     regs[i].wasconst=current.isconst;
-    regs[i].was32=current.is32;
     regs[i].wasdirty=current.dirty;
     regs[i].loadedconst=0;
-    if(itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=RJUMP&&itype[i]!=FJUMP) {
+    if(itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=RJUMP) {
       if(i+1<slen) {
         current.u=unneeded_reg[i+1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
-        current.uu=unneeded_reg_upper[i+1]&~((1LL<<us1[i])|(1LL<<us2[i]));
-        if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
         current.u|=1;
-        current.uu|=1;
       } else {
         current.u=1;
-        current.uu=1;
       }
     } else {
       if(i+1<slen) {
         current.u=branch_unneeded_reg[i]&~((1LL<<rs1[i+1])|(1LL<<rs2[i+1]));
-        current.uu=branch_unneeded_reg_upper[i]&~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
-        if((~current.uu>>rt1[i+1])&1) current.uu&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
         current.u&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
-        current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
         current.u|=1;
-        current.uu|=1;
-      } else { SysPrintf("oops, branch at end of block with no delay slot\n");exit(1); }
+      } else { SysPrintf("oops, branch at end of block with no delay slot\n");abort(); }
     }
     is_ds[i]=ds;
     if(ds) {
@@ -7944,27 +7549,19 @@ int new_recompile_block(int addr)
       // ...but we need to alloc it in case something jumps here
       if(i+1<slen) {
         current.u=branch_unneeded_reg[i-1]&unneeded_reg[i+1];
-        current.uu=branch_unneeded_reg_upper[i-1]&unneeded_reg_upper[i+1];
       }else{
         current.u=branch_unneeded_reg[i-1];
-        current.uu=branch_unneeded_reg_upper[i-1];
       }
       current.u&=~((1LL<<rs1[i])|(1LL<<rs2[i]));
-      current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
-      if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
       current.u|=1;
-      current.uu|=1;
       struct regstat temp;
       memcpy(&temp,&current,sizeof(current));
       temp.wasdirty=temp.dirty;
-      temp.was32=temp.is32;
       // TODO: Take into account unconditional branches, as below
       delayslot_alloc(&temp,i);
       memcpy(regs[i].regmap,temp.regmap,sizeof(temp.regmap));
       regs[i].wasdirty=temp.wasdirty;
-      regs[i].was32=temp.was32;
       regs[i].dirty=temp.dirty;
-      regs[i].is32=temp.is32;
       regs[i].isconst=0;
       regs[i].wasconst=0;
       current.isconst=0;
@@ -7978,7 +7575,7 @@ int new_recompile_block(int addr)
           }
           else
           {
-            if(r<64){
+              assert(r < 64);
               if((current.u>>r)&1) {
                 regs[i].regmap_entry[hr]=-1;
                 regs[i].regmap[hr]=-1;
@@ -7986,16 +7583,6 @@ int new_recompile_block(int addr)
                 //current.regmap[hr]=-1;
               }else
                 regs[i].regmap_entry[hr]=r;
-            }
-            else {
-              if((current.uu>>(r&63))&1) {
-                regs[i].regmap_entry[hr]=-1;
-                regs[i].regmap[hr]=-1;
-                //Don't clear regs in the delay slot as the branch might need them
-                //current.regmap[hr]=-1;
-              }else
-                regs[i].regmap_entry[hr]=r;
-            }
           }
         } else {
           // First instruction expects CCREG to be allocated
@@ -8023,7 +7610,6 @@ int new_recompile_block(int addr)
             #ifdef REG_PREFETCH
             alloc_reg(&current,i,PTEMP);
             #endif
-            //current.is32|=1LL<<rt1[i];
           }
           ooo[i]=1;
           delayslot_alloc(&current,i+1);
@@ -8053,9 +7639,7 @@ int new_recompile_block(int addr)
             #ifdef USE_MINI_HT
             if(rs1[i]==31) { // JALR
               alloc_reg(&current,i,RHASH);
-              #ifndef HOST_IMM_ADDR32
               alloc_reg(&current,i,RHTBL);
-              #endif
             }
             #endif
             delayslot_alloc(&current,i+1);
@@ -8085,11 +7669,6 @@ int new_recompile_block(int addr)
             dirty_reg(&current,CCREG);
             if(rs1[i]) alloc_reg(&current,i,rs1[i]);
             if(rs2[i]) alloc_reg(&current,i,rs2[i]);
-            if(!((current.is32>>rs1[i])&(current.is32>>rs2[i])&1))
-            {
-              if(rs1[i]) alloc_reg64(&current,i,rs1[i]);
-              if(rs2[i]) alloc_reg64(&current,i,rs2[i]);
-            }
             if((rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1]))||
                (rs2[i]&&(rs2[i]==rt1[i+1]||rs2[i]==rt2[i+1]))) {
               // The delay slot overwrites one of our conditions.
@@ -8099,11 +7678,6 @@ int new_recompile_block(int addr)
               regs[i].wasconst=0;
               if(rs1[i]) alloc_reg(&current,i,rs1[i]);
               if(rs2[i]) alloc_reg(&current,i,rs2[i]);
-              if(!((current.is32>>rs1[i])&(current.is32>>rs2[i])&1))
-              {
-                if(rs1[i]) alloc_reg64(&current,i,rs1[i]);
-                if(rs2[i]) alloc_reg64(&current,i,rs2[i]);
-              }
             }
             else
             {
@@ -8117,10 +7691,6 @@ int new_recompile_block(int addr)
             alloc_cc(&current,i);
             dirty_reg(&current,CCREG);
             alloc_reg(&current,i,rs1[i]);
-            if(!(current.is32>>rs1[i]&1))
-            {
-              alloc_reg64(&current,i,rs1[i]);
-            }
             if(rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1])) {
               // The delay slot overwrites one of our conditions.
               // Allocate the branch condition registers instead.
@@ -8128,10 +7698,6 @@ int new_recompile_block(int addr)
               current.wasconst=0;
               regs[i].wasconst=0;
               if(rs1[i]) alloc_reg(&current,i,rs1[i]);
-              if(!((current.is32>>rs1[i])&1))
-              {
-                if(rs1[i]) alloc_reg64(&current,i,rs1[i]);
-              }
             }
             else
             {
@@ -8150,11 +7716,6 @@ int new_recompile_block(int addr)
             dirty_reg(&current,CCREG);
             alloc_reg(&current,i,rs1[i]);
             alloc_reg(&current,i,rs2[i]);
-            if(!((current.is32>>rs1[i])&(current.is32>>rs2[i])&1))
-            {
-              alloc_reg64(&current,i,rs1[i]);
-              alloc_reg64(&current,i,rs2[i]);
-            }
           }
           else
           if((opcode[i]&0x3E)==0x16) // BLEZL/BGTZL
@@ -8165,10 +7726,6 @@ int new_recompile_block(int addr)
             alloc_cc(&current,i);
             dirty_reg(&current,CCREG);
             alloc_reg(&current,i,rs1[i]);
-            if(!(current.is32>>rs1[i]&1))
-            {
-              alloc_reg64(&current,i,rs1[i]);
-            }
           }
           ds=1;
           //current.isconst=0;
@@ -8185,17 +7742,12 @@ int new_recompile_block(int addr)
             alloc_cc(&current,i);
             dirty_reg(&current,CCREG);
             alloc_reg(&current,i,rs1[i]);
-            if(!(current.is32>>rs1[i]&1))
-            {
-              alloc_reg64(&current,i,rs1[i]);
-            }
             if (rt1[i]==31) { // BLTZAL/BGEZAL
               alloc_reg(&current,i,31);
               dirty_reg(&current,31);
               //#ifdef REG_PREFETCH
               //alloc_reg(&current,i,PTEMP);
               //#endif
-              //current.is32|=1LL<<rt1[i];
             }
             if((rs1[i]&&(rs1[i]==rt1[i+1]||rs1[i]==rt2[i+1])) // The delay slot overwrites the branch condition.
                ||(rt1[i]==31&&(rs1[i+1]==31||rs2[i+1]==31||rt1[i+1]==31||rt2[i+1]==31))) { // DS touches $ra
@@ -8204,10 +7756,6 @@ int new_recompile_block(int addr)
               current.wasconst=0;
               regs[i].wasconst=0;
               if(rs1[i]) alloc_reg(&current,i,rs1[i]);
-              if(!((current.is32>>rs1[i])&1))
-              {
-                if(rs1[i]) alloc_reg64(&current,i,rs1[i]);
-              }
             }
             else
             {
@@ -8225,55 +7773,10 @@ int new_recompile_block(int addr)
             alloc_cc(&current,i);
             dirty_reg(&current,CCREG);
             alloc_reg(&current,i,rs1[i]);
-            if(!(current.is32>>rs1[i]&1))
-            {
-              alloc_reg64(&current,i,rs1[i]);
-            }
           }
           ds=1;
           //current.isconst=0;
           break;
-        case FJUMP:
-          current.isconst=0;
-          current.wasconst=0;
-          regs[i].wasconst=0;
-          if(likely[i]==0) // BC1F/BC1T
-          {
-            // TODO: Theoretically we can run out of registers here on x86.
-            // The delay slot can allocate up to six, and we need to check
-            // CSREG before executing the delay slot.  Possibly we can drop
-            // the cycle count and then reload it after checking that the
-            // FPU is in a usable state, or don't do out-of-order execution.
-            alloc_cc(&current,i);
-            dirty_reg(&current,CCREG);
-            alloc_reg(&current,i,FSREG);
-            alloc_reg(&current,i,CSREG);
-            if(itype[i+1]==FCOMP) {
-              // The delay slot overwrites the branch condition.
-              // Allocate the branch condition registers instead.
-              alloc_cc(&current,i);
-              dirty_reg(&current,CCREG);
-              alloc_reg(&current,i,CSREG);
-              alloc_reg(&current,i,FSREG);
-            }
-            else {
-              ooo[i]=1;
-              delayslot_alloc(&current,i+1);
-              alloc_reg(&current,i+1,CSREG);
-            }
-          }
-          else
-          // Don't alloc the delay slot yet because we might not execute it
-          if(likely[i]) // BC1FL/BC1TL
-          {
-            alloc_cc(&current,i);
-            dirty_reg(&current,CCREG);
-            alloc_reg(&current,i,CSREG);
-            alloc_reg(&current,i,FSREG);
-          }
-          ds=1;
-          current.isconst=0;
-          break;
         case IMM16:
           imm16_alloc(&current,i);
           break;
@@ -8304,8 +7807,9 @@ int new_recompile_block(int addr)
           cop0_alloc(&current,i);
           break;
         case COP1:
+          break;
         case COP2:
-          cop1_alloc(&current,i);
+          cop2_alloc(&current,i);
           break;
         case C1LS:
           c1ls_alloc(&current,i);
@@ -8316,15 +7820,6 @@ int new_recompile_block(int addr)
         case C2OP:
           c2op_alloc(&current,i);
           break;
-        case FCONV:
-          fconv_alloc(&current,i);
-          break;
-        case FLOAT:
-          float_alloc(&current,i);
-          break;
-        case FCOMP:
-          fcomp_alloc(&current,i);
-          break;
         case SYSCALL:
         case HLECALL:
         case INTCALL:
@@ -8335,24 +7830,10 @@ int new_recompile_block(int addr)
           break;
       }
 
-      // Drop the upper half of registers that have become 32-bit
-      current.uu|=current.is32&((1LL<<rt1[i])|(1LL<<rt2[i]));
-      if(itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=RJUMP&&itype[i]!=FJUMP) {
-        current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
-        if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
-        current.uu|=1;
-      } else {
-        current.uu|=current.is32&((1LL<<rt1[i+1])|(1LL<<rt2[i+1]));
-        current.uu&=~((1LL<<us1[i+1])|(1LL<<us2[i+1]));
-        if((~current.uu>>rt1[i+1])&1) current.uu&=~((1LL<<dep1[i+1])|(1LL<<dep2[i+1]));
-        current.uu&=~((1LL<<us1[i])|(1LL<<us2[i]));
-        current.uu|=1;
-      }
-
       // Create entry (branch target) regmap
       for(hr=0;hr<HOST_REGS;hr++)
       {
-        int r,or,er;
+        int r,or;
         r=current.regmap[hr];
         if(r>=0) {
           if(r!=regmap_pre[i][hr]) {
@@ -8376,7 +7857,8 @@ int new_recompile_block(int addr)
               regs[i].regmap_entry[hr]=0;
             }
             else
-            if(r<64){
+            {
+              assert(r<64);
               if((current.u>>r)&1) {
                 regs[i].regmap_entry[hr]=-1;
                 //regs[i].regmap[hr]=-1;
@@ -8384,14 +7866,6 @@ int new_recompile_block(int addr)
               }else
                 regs[i].regmap_entry[hr]=r;
             }
-            else {
-              if((current.uu>>(r&63))&1) {
-                regs[i].regmap_entry[hr]=-1;
-                //regs[i].regmap[hr]=-1;
-                current.regmap[hr]=-1;
-              }else
-                regs[i].regmap_entry[hr]=r;
-            }
           }
         } else {
           // Branches expect CCREG to be allocated at the target
@@ -8414,7 +7888,6 @@ int new_recompile_block(int addr)
     /* Branch post-alloc */
     if(i>0)
     {
-      current.was32=current.is32;
       current.wasdirty=current.dirty;
       switch(itype[i-1]) {
         case UJUMP:
@@ -8422,41 +7895,35 @@ int new_recompile_block(int addr)
           branch_regs[i-1].isconst=0;
           branch_regs[i-1].wasconst=0;
           branch_regs[i-1].u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i-1])|(1LL<<rs2[i-1]));
-          branch_regs[i-1].uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i-1])|(1LL<<us2[i-1]));
           alloc_cc(&branch_regs[i-1],i-1);
           dirty_reg(&branch_regs[i-1],CCREG);
           if(rt1[i-1]==31) { // JAL
             alloc_reg(&branch_regs[i-1],i-1,31);
             dirty_reg(&branch_regs[i-1],31);
-            branch_regs[i-1].is32|=1LL<<31;
           }
           memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
-          memcpy(constmap[i],constmap[i-1],sizeof(current_constmap));
+          memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
           break;
         case RJUMP:
           memcpy(&branch_regs[i-1],&current,sizeof(current));
           branch_regs[i-1].isconst=0;
           branch_regs[i-1].wasconst=0;
           branch_regs[i-1].u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i-1])|(1LL<<rs2[i-1]));
-          branch_regs[i-1].uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i-1])|(1LL<<us2[i-1]));
           alloc_cc(&branch_regs[i-1],i-1);
           dirty_reg(&branch_regs[i-1],CCREG);
           alloc_reg(&branch_regs[i-1],i-1,rs1[i-1]);
           if(rt1[i-1]!=0) { // JALR
             alloc_reg(&branch_regs[i-1],i-1,rt1[i-1]);
             dirty_reg(&branch_regs[i-1],rt1[i-1]);
-            branch_regs[i-1].is32|=1LL<<rt1[i-1];
           }
           #ifdef USE_MINI_HT
           if(rs1[i-1]==31) { // JALR
             alloc_reg(&branch_regs[i-1],i-1,RHASH);
-            #ifndef HOST_IMM_ADDR32
             alloc_reg(&branch_regs[i-1],i-1,RHTBL);
-            #endif
           }
           #endif
           memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
-          memcpy(constmap[i],constmap[i-1],sizeof(current_constmap));
+          memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
           break;
         case CJUMP:
           if((opcode[i-1]&0x3E)==4) // BEQ/BNE
@@ -8468,31 +7935,22 @@ int new_recompile_block(int addr)
               // The delay slot overwrote one of our conditions
               // Delay slot goes after the test (in order)
               current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
-              current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i]));
-              if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
               current.u|=1;
-              current.uu|=1;
               delayslot_alloc(&current,i);
               current.isconst=0;
             }
             else
             {
               current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i-1])|(1LL<<rs2[i-1]));
-              current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i-1])|(1LL<<us2[i-1]));
               // Alloc the branch condition registers
               if(rs1[i-1]) alloc_reg(&current,i-1,rs1[i-1]);
               if(rs2[i-1]) alloc_reg(&current,i-1,rs2[i-1]);
-              if(!((current.is32>>rs1[i-1])&(current.is32>>rs2[i-1])&1))
-              {
-                if(rs1[i-1]) alloc_reg64(&current,i-1,rs1[i-1]);
-                if(rs2[i-1]) alloc_reg64(&current,i-1,rs2[i-1]);
-              }
             }
             memcpy(&branch_regs[i-1],&current,sizeof(current));
             branch_regs[i-1].isconst=0;
             branch_regs[i-1].wasconst=0;
             memcpy(&branch_regs[i-1].regmap_entry,&current.regmap,sizeof(current.regmap));
-            memcpy(constmap[i],constmap[i-1],sizeof(current_constmap));
+            memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
           }
           else
           if((opcode[i-1]&0x3E)==6) // BLEZ/BGTZ
@@ -8503,29 +7961,21 @@ int new_recompile_block(int addr)
               // The delay slot overwrote the branch condition
               // Delay slot goes after the test (in order)
               current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
-              current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i]));
-              if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
               current.u|=1;
-              current.uu|=1;
               delayslot_alloc(&current,i);
               current.isconst=0;
             }
             else
             {
               current.u=branch_unneeded_reg[i-1]&~(1LL<<rs1[i-1]);
-              current.uu=branch_unneeded_reg_upper[i-1]&~(1LL<<us1[i-1]);
               // Alloc the branch condition register
               alloc_reg(&current,i-1,rs1[i-1]);
-              if(!(current.is32>>rs1[i-1]&1))
-              {
-                alloc_reg64(&current,i-1,rs1[i-1]);
-              }
             }
             memcpy(&branch_regs[i-1],&current,sizeof(current));
             branch_regs[i-1].isconst=0;
             branch_regs[i-1].wasconst=0;
             memcpy(&branch_regs[i-1].regmap_entry,&current.regmap,sizeof(current.regmap));
-            memcpy(constmap[i],constmap[i-1],sizeof(current_constmap));
+            memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
           }
           else
           // Alloc the delay slot in case the branch is taken
@@ -8533,8 +7983,6 @@ int new_recompile_block(int addr)
           {
             memcpy(&branch_regs[i-1],&current,sizeof(current));
             branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
-            branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
-            if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
             alloc_cc(&branch_regs[i-1],i);
             dirty_reg(&branch_regs[i-1],CCREG);
             delayslot_alloc(&branch_regs[i-1],i);
@@ -8548,8 +7996,6 @@ int new_recompile_block(int addr)
           {
             memcpy(&branch_regs[i-1],&current,sizeof(current));
             branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
-            branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
-            if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
             alloc_cc(&branch_regs[i-1],i);
             dirty_reg(&branch_regs[i-1],CCREG);
             delayslot_alloc(&branch_regs[i-1],i);
@@ -8569,29 +8015,21 @@ int new_recompile_block(int addr)
               // The delay slot overwrote the branch condition
               // Delay slot goes after the test (in order)
               current.u=branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i]));
-              current.uu=branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i]));
-              if((~current.uu>>rt1[i])&1) current.uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]));
               current.u|=1;
-              current.uu|=1;
               delayslot_alloc(&current,i);
               current.isconst=0;
             }
             else
             {
               current.u=branch_unneeded_reg[i-1]&~(1LL<<rs1[i-1]);
-              current.uu=branch_unneeded_reg_upper[i-1]&~(1LL<<us1[i-1]);
               // Alloc the branch condition register
               alloc_reg(&current,i-1,rs1[i-1]);
-              if(!(current.is32>>rs1[i-1]&1))
-              {
-                alloc_reg64(&current,i-1,rs1[i-1]);
-              }
             }
             memcpy(&branch_regs[i-1],&current,sizeof(current));
             branch_regs[i-1].isconst=0;
             branch_regs[i-1].wasconst=0;
             memcpy(&branch_regs[i-1].regmap_entry,&current.regmap,sizeof(current.regmap));
-            memcpy(constmap[i],constmap[i-1],sizeof(current_constmap));
+            memcpy(constmap[i],constmap[i-1],sizeof(constmap[i]));
           }
           else
           // Alloc the delay slot in case the branch is taken
@@ -8599,8 +8037,6 @@ int new_recompile_block(int addr)
           {
             memcpy(&branch_regs[i-1],&current,sizeof(current));
             branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
-            branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
-            if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
             alloc_cc(&branch_regs[i-1],i);
             dirty_reg(&branch_regs[i-1],CCREG);
             delayslot_alloc(&branch_regs[i-1],i);
@@ -8613,54 +8049,15 @@ int new_recompile_block(int addr)
           if(opcode2[i-1]&0x10) { // BxxZAL
             alloc_reg(&branch_regs[i-1],i-1,31);
             dirty_reg(&branch_regs[i-1],31);
-            branch_regs[i-1].is32|=1LL<<31;
-          }
-          break;
-        case FJUMP:
-          if(likely[i-1]==0) // BC1F/BC1T
-          {
-            alloc_cc(&current,i-1);
-            dirty_reg(&current,CCREG);
-            if(itype[i]==FCOMP) {
-              // The delay slot overwrote the branch condition
-              // Delay slot goes after the test (in order)
-              delayslot_alloc(&current,i);
-              current.isconst=0;
-            }
-            else
-            {
-              current.u=branch_unneeded_reg[i-1]&~(1LL<<rs1[i-1]);
-              current.uu=branch_unneeded_reg_upper[i-1]&~(1LL<<us1[i-1]);
-              // Alloc the branch condition register
-              alloc_reg(&current,i-1,FSREG);
-            }
-            memcpy(&branch_regs[i-1],&current,sizeof(current));
-            memcpy(&branch_regs[i-1].regmap_entry,&current.regmap,sizeof(current.regmap));
-          }
-          else // BC1FL/BC1TL
-          {
-            // Alloc the delay slot in case the branch is taken
-            memcpy(&branch_regs[i-1],&current,sizeof(current));
-            branch_regs[i-1].u=(branch_unneeded_reg[i-1]&~((1LL<<rs1[i])|(1LL<<rs2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
-            branch_regs[i-1].uu=(branch_unneeded_reg_upper[i-1]&~((1LL<<us1[i])|(1LL<<us2[i])|(1LL<<rt1[i])|(1LL<<rt2[i])))|1;
-            if((~branch_regs[i-1].uu>>rt1[i])&1) branch_regs[i-1].uu&=~((1LL<<dep1[i])|(1LL<<dep2[i]))|1;
-            alloc_cc(&branch_regs[i-1],i);
-            dirty_reg(&branch_regs[i-1],CCREG);
-            delayslot_alloc(&branch_regs[i-1],i);
-            branch_regs[i-1].isconst=0;
-            alloc_reg(&current,i,CCREG); // Not taken path
-            dirty_reg(&current,CCREG);
-            memcpy(&branch_regs[i-1].regmap_entry,&branch_regs[i-1].regmap,sizeof(current.regmap));
           }
           break;
       }
 
-      if(itype[i-1]==UJUMP||itype[i-1]==RJUMP||(source[i-1]>>16)==0x1000)
+      if (is_ujump(i-1))
       {
         if(rt1[i-1]==31) // JAL/JALR
         {
           // Subroutine call will return here, don't alloc any registers
-          current.is32=1;
           current.dirty=0;
           clear_all_regs(current.regmap);
           alloc_reg(&current,i,CCREG);
@@ -8669,7 +8066,6 @@ int new_recompile_block(int addr)
         else if(i+1<slen)
         {
           // Internal branch will jump here, match registers to caller
-          current.is32=0x3FFFFFFFFLL;
           current.dirty=0;
           clear_all_regs(current.regmap);
           alloc_reg(&current,i,CCREG);
@@ -8678,7 +8074,6 @@ int new_recompile_block(int addr)
           {
             if(ba[j]==start+i*4+4) {
               memcpy(current.regmap,branch_regs[j].regmap,sizeof(current.regmap));
-              current.is32=branch_regs[j].is32;
               current.dirty=branch_regs[j].dirty;
               break;
             }
@@ -8689,7 +8084,6 @@ int new_recompile_block(int addr)
                 if(current.regmap[hr]!=branch_regs[j].regmap[hr]) {
                   current.regmap[hr]=-1;
                 }
-                current.is32&=branch_regs[j].is32;
                 current.dirty&=branch_regs[j].dirty;
               }
             }
@@ -8701,19 +8095,17 @@ int new_recompile_block(int addr)
 
     // Count cycles in between branches
     ccadj[i]=cc;
-    if(i>0&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i-1]==FJUMP||itype[i]==SYSCALL||itype[i]==HLECALL))
+    if(i>0&&(itype[i-1]==RJUMP||itype[i-1]==UJUMP||itype[i-1]==CJUMP||itype[i-1]==SJUMP||itype[i]==SYSCALL||itype[i]==HLECALL))
     {
       cc=0;
     }
 #if !defined(DRC_DBG)
     else if(itype[i]==C2OP&&gte_cycletab[source[i]&0x3f]>2)
     {
-      // GTE runs in parallel until accessed, divide by 2 for a rough guess
-      cc+=gte_cycletab[source[i]&0x3f]/2;
-    }
-    else if(/*itype[i]==LOAD||itype[i]==STORE||*/itype[i]==C1LS) // load,store causes weird timing issues
-    {
-      cc+=2; // 2 cycle penalty (after CLOCK_DIVIDER)
+      // this should really be removed since the real stalls have been implemented,
+      // but doing so causes sizeable perf regression against the older version
+      u_int gtec = gte_cycletab[source[i] & 0x3f];
+      cc += HACK_ENABLED(NDHACK_GTE_NO_STALL) ? gtec/2 : 2;
     }
     else if(i>1&&itype[i]==STORE&&itype[i-1]==STORE&&itype[i-2]==STORE&&!bt[i])
     {
@@ -8721,7 +8113,8 @@ int new_recompile_block(int addr)
     }
     else if(itype[i]==C2LS)
     {
-      cc+=4;
+      // same as with C2OP
+      cc += HACK_ENABLED(NDHACK_GTE_NO_STALL) ? 4 : 2;
     }
 #endif
     else
@@ -8729,12 +8122,10 @@ int new_recompile_block(int addr)
       cc++;
     }
 
-    flush_dirty_uppers(&current);
     if(!is_ds[i]) {
-      regs[i].is32=current.is32;
       regs[i].dirty=current.dirty;
       regs[i].isconst=current.isconst;
-      memcpy(constmap[i],current_constmap,sizeof(current_constmap));
+      memcpy(constmap[i],current_constmap,sizeof(constmap[i]));
     }
     for(hr=0;hr<HOST_REGS;hr++) {
       if(hr!=EXCLUDE_REG&&regs[i].regmap[hr]>=0) {
@@ -8754,7 +8145,7 @@ int new_recompile_block(int addr)
   for (i=slen-1;i>=0;i--)
   {
     int hr;
-    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP)
     {
       if(ba[i]<start || ba[i]>=(start+slen*4))
       {
@@ -8775,7 +8166,7 @@ int new_recompile_block(int addr)
         }
       }
       // Conditional branch may need registers for following instructions
-      if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
+      if (!is_ujump(i))
       {
         if(i<slen-2) {
           nr|=needed_reg[i+2];
@@ -8798,22 +8189,10 @@ int new_recompile_block(int addr)
           if(rt1[i+1]&&rt1[i+1]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
           if(rt2[i+1]&&rt2[i+1]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
         }
-        if(us1[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
-        if(us2[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
         if(rs1[i+1]==regmap_pre[i][hr]) nr|=1<<hr;
         if(rs2[i+1]==regmap_pre[i][hr]) nr|=1<<hr;
-        if(us1[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
-        if(us2[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
         if(rs1[i+1]==regs[i].regmap_entry[hr]) nr|=1<<hr;
         if(rs2[i+1]==regs[i].regmap_entry[hr]) nr|=1<<hr;
-        if(dep1[i+1]&&!((unneeded_reg_upper[i]>>dep1[i+1])&1)) {
-          if(dep1[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
-          if(dep2[i+1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
-        }
-        if(dep2[i+1]&&!((unneeded_reg_upper[i]>>dep2[i+1])&1)) {
-          if(dep1[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
-          if(dep2[i+1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
-        }
         if(itype[i+1]==STORE || itype[i+1]==STORELR || (opcode[i+1]&0x3b)==0x39 || (opcode[i+1]&0x3b)==0x3a) {
           if(regmap_pre[i][hr]==INVCP) nr|=1<<hr;
           if(regs[i].regmap_entry[hr]==INVCP) nr|=1<<hr;
@@ -8848,22 +8227,10 @@ int new_recompile_block(int addr)
       if(rt2[i]&&rt2[i]==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
       if(FTEMP==(regs[i].regmap[hr]&63)) nr&=~(1<<hr);
       // Source registers are needed
-      if(us1[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
-      if(us2[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
       if(rs1[i]==regmap_pre[i][hr]) nr|=1<<hr;
       if(rs2[i]==regmap_pre[i][hr]) nr|=1<<hr;
-      if(us1[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
-      if(us2[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
       if(rs1[i]==regs[i].regmap_entry[hr]) nr|=1<<hr;
       if(rs2[i]==regs[i].regmap_entry[hr]) nr|=1<<hr;
-      if(dep1[i]&&!((unneeded_reg_upper[i]>>dep1[i])&1)) {
-        if(dep1[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
-        if(dep1[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
-      }
-      if(dep2[i]&&!((unneeded_reg_upper[i]>>dep2[i])&1)) {
-        if(dep2[i]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
-        if(dep2[i]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
-      }
       if(itype[i]==STORE || itype[i]==STORELR || (opcode[i]&0x3b)==0x39 || (opcode[i]&0x3b)==0x3a) {
         if(regmap_pre[i][hr]==INVCP) nr|=1<<hr;
         if(regs[i].regmap_entry[hr]==INVCP) nr|=1<<hr;
@@ -8873,20 +8240,18 @@ int new_recompile_block(int addr)
       // But do so if this is a branch target, otherwise we
       // might have to load the register before the branch.
       if(i>0&&!bt[i]&&((regs[i].wasdirty>>hr)&1)) {
-        if((regmap_pre[i][hr]>0&&regmap_pre[i][hr]<64&&!((unneeded_reg[i]>>regmap_pre[i][hr])&1)) ||
-           (regmap_pre[i][hr]>64&&!((unneeded_reg_upper[i]>>(regmap_pre[i][hr]&63))&1)) ) {
+        if((regmap_pre[i][hr]>0&&!((unneeded_reg[i]>>regmap_pre[i][hr])&1))) {
           if(rt1[i-1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
           if(rt2[i-1]==(regmap_pre[i][hr]&63)) nr|=1<<hr;
         }
-        if((regs[i].regmap_entry[hr]>0&&regs[i].regmap_entry[hr]<64&&!((unneeded_reg[i]>>regs[i].regmap_entry[hr])&1)) ||
-           (regs[i].regmap_entry[hr]>64&&!((unneeded_reg_upper[i]>>(regs[i].regmap_entry[hr]&63))&1)) ) {
+        if((regs[i].regmap_entry[hr]>0&&!((unneeded_reg[i]>>regs[i].regmap_entry[hr])&1))) {
           if(rt1[i-1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
           if(rt2[i-1]==(regs[i].regmap_entry[hr]&63)) nr|=1<<hr;
         }
       }
     }
     // Cycle count is needed at branches.  Assume it is needed at the target too.
-    if(i==0||bt[i]||itype[i]==CJUMP||itype[i]==FJUMP||itype[i]==SPAN) {
+    if(i==0||bt[i]||itype[i]==CJUMP||itype[i]==SPAN) {
       if(regmap_pre[i][HOST_CCREG]==CCREG) nr|=1<<HOST_CCREG;
       if(regs[i].regmap_entry[HOST_CCREG]==CCREG) nr|=1<<HOST_CCREG;
     }
@@ -8902,7 +8267,7 @@ int new_recompile_block(int addr)
            (regs[i].regmap[hr]&63)!=rt1[i] && (regs[i].regmap[hr]&63)!=rt2[i] &&
            (regs[i].regmap[hr]&63)!=PTEMP && (regs[i].regmap[hr]&63)!=CCREG)
         {
-          if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
+          if (!is_ujump(i))
           {
             if(likely[i]) {
               regs[i].regmap[hr]=-1;
@@ -8914,14 +8279,9 @@ int new_recompile_block(int addr)
             }
           }
         }
-        if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+        if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP)
         {
-          int d1=0,d2=0,map=0,temp=0;
-          if(get_reg(regs[i].regmap,rt1[i+1]|64)>=0||get_reg(branch_regs[i].regmap,rt1[i+1]|64)>=0)
-          {
-            d1=dep1[i+1];
-            d2=dep2[i+1];
-          }
+          int map=0,temp=0;
           if(itype[i+1]==STORE || itype[i+1]==STORELR ||
              (opcode[i+1]&0x3b)==0x39 || (opcode[i+1]&0x3b)==0x3a) { // SWC1/SDC1 || SWC2/SDC2
             map=INVCP;
@@ -8932,8 +8292,6 @@ int new_recompile_block(int addr)
           if((regs[i].regmap[hr]&63)!=rs1[i] && (regs[i].regmap[hr]&63)!=rs2[i] &&
              (regs[i].regmap[hr]&63)!=rt1[i] && (regs[i].regmap[hr]&63)!=rt2[i] &&
              (regs[i].regmap[hr]&63)!=rt1[i+1] && (regs[i].regmap[hr]&63)!=rt2[i+1] &&
-             (regs[i].regmap[hr]^64)!=us1[i+1] && (regs[i].regmap[hr]^64)!=us2[i+1] &&
-             (regs[i].regmap[hr]^64)!=d1 && (regs[i].regmap[hr]^64)!=d2 &&
              regs[i].regmap[hr]!=rs1[i+1] && regs[i].regmap[hr]!=rs2[i+1] &&
              (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=PTEMP &&
              regs[i].regmap[hr]!=RHASH && regs[i].regmap[hr]!=RHTBL &&
@@ -8945,8 +8303,6 @@ int new_recompile_block(int addr)
             if((branch_regs[i].regmap[hr]&63)!=rs1[i] && (branch_regs[i].regmap[hr]&63)!=rs2[i] &&
                (branch_regs[i].regmap[hr]&63)!=rt1[i] && (branch_regs[i].regmap[hr]&63)!=rt2[i] &&
                (branch_regs[i].regmap[hr]&63)!=rt1[i+1] && (branch_regs[i].regmap[hr]&63)!=rt2[i+1] &&
-               (branch_regs[i].regmap[hr]^64)!=us1[i+1] && (branch_regs[i].regmap[hr]^64)!=us2[i+1] &&
-               (branch_regs[i].regmap[hr]^64)!=d1 && (branch_regs[i].regmap[hr]^64)!=d2 &&
                branch_regs[i].regmap[hr]!=rs1[i+1] && branch_regs[i].regmap[hr]!=rs2[i+1] &&
                (branch_regs[i].regmap[hr]&63)!=temp && branch_regs[i].regmap[hr]!=PTEMP &&
                branch_regs[i].regmap[hr]!=RHASH && branch_regs[i].regmap[hr]!=RHTBL &&
@@ -8955,7 +8311,7 @@ int new_recompile_block(int addr)
             {
               branch_regs[i].regmap[hr]=-1;
               branch_regs[i].regmap_entry[hr]=-1;
-              if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000)
+              if (!is_ujump(i))
               {
                 if(!likely[i]&&i<slen-2) {
                   regmap_pre[i+2][hr]=-1;
@@ -8970,12 +8326,7 @@ int new_recompile_block(int addr)
           // Non-branch
           if(i>0)
           {
-            int d1=0,d2=0,map=-1,temp=-1;
-            if(get_reg(regs[i].regmap,rt1[i]|64)>=0)
-            {
-              d1=dep1[i];
-              d2=dep2[i];
-            }
+            int map=-1,temp=-1;
             if(itype[i]==STORE || itype[i]==STORELR ||
                       (opcode[i]&0x3b)==0x39 || (opcode[i]&0x3b)==0x3a) { // SWC1/SDC1 || SWC2/SDC2
               map=INVCP;
@@ -8984,16 +8335,14 @@ int new_recompile_block(int addr)
                itype[i]==C1LS || itype[i]==C2LS)
               temp=FTEMP;
             if((regs[i].regmap[hr]&63)!=rt1[i] && (regs[i].regmap[hr]&63)!=rt2[i] &&
-               (regs[i].regmap[hr]^64)!=us1[i] && (regs[i].regmap[hr]^64)!=us2[i] &&
-               (regs[i].regmap[hr]^64)!=d1 && (regs[i].regmap[hr]^64)!=d2 &&
                regs[i].regmap[hr]!=rs1[i] && regs[i].regmap[hr]!=rs2[i] &&
                (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=map &&
                (itype[i]!=SPAN||regs[i].regmap[hr]!=CCREG))
             {
               if(i<slen-1&&!is_ds[i]) {
-                if(regmap_pre[i+1][hr]!=-1 || regs[i].regmap[hr]!=-1)
+                assert(regs[i].regmap[hr]<64);
+                if(regmap_pre[i+1][hr]!=-1 || regs[i].regmap[hr]>0)
                 if(regmap_pre[i+1][hr]!=regs[i].regmap[hr])
-                if(regs[i].regmap[hr]<64||!((regs[i].was32>>(regs[i].regmap[hr]&63))&1))
                 {
                   SysPrintf("fail: %x (%d %d!=%d)\n",start+i*4,hr,regmap_pre[i+1][hr],regs[i].regmap[hr]);
                   assert(regmap_pre[i+1][hr]==regs[i].regmap[hr]);
@@ -9007,8 +8356,8 @@ int new_recompile_block(int addr)
             }
           }
         }
-      }
-    }
+      } // if needed
+    } // for hr
   }
 
   /* Pass 5 - Pre-allocate registers */
@@ -9021,27 +8370,21 @@ int new_recompile_block(int addr)
   clear_all_regs(f_regmap);
   for(i=0;i<slen-1;i++)
   {
-    if(itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    if(itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP)
     {
       if(ba[i]>=start && ba[i]<(start+i*4))
       if(itype[i+1]==NOP||itype[i+1]==MOV||itype[i+1]==ALU
       ||itype[i+1]==SHIFTIMM||itype[i+1]==IMM16||itype[i+1]==LOAD
       ||itype[i+1]==STORE||itype[i+1]==STORELR||itype[i+1]==C1LS
-      ||itype[i+1]==SHIFT||itype[i+1]==COP1||itype[i+1]==FLOAT
-      ||itype[i+1]==FCOMP||itype[i+1]==FCONV
+      ||itype[i+1]==SHIFT||itype[i+1]==COP1
       ||itype[i+1]==COP2||itype[i+1]==C2LS||itype[i+1]==C2OP)
       {
         int t=(ba[i]-start)>>2;
-        if(t>0&&(itype[t-1]!=UJUMP&&itype[t-1]!=RJUMP&&itype[t-1]!=CJUMP&&itype[t-1]!=SJUMP&&itype[t-1]!=FJUMP)) // loop_preload can't handle jumps into delay slots
+        if(t>0&&(itype[t-1]!=UJUMP&&itype[t-1]!=RJUMP&&itype[t-1]!=CJUMP&&itype[t-1]!=SJUMP)) // loop_preload can't handle jumps into delay slots
         if(t<2||(itype[t-2]!=UJUMP&&itype[t-2]!=RJUMP)||rt1[t-2]!=31) // call/ret assumes no registers allocated
         for(hr=0;hr<HOST_REGS;hr++)
         {
-          if(regs[i].regmap[hr]>64) {
-            if(!((regs[i].dirty>>hr)&1))
-              f_regmap[hr]=regs[i].regmap[hr];
-            else f_regmap[hr]=-1;
-          }
-          else if(regs[i].regmap[hr]>=0) {
+          if(regs[i].regmap[hr]>=0) {
             if(f_regmap[hr]!=regs[i].regmap[hr]) {
               // dealloc old register
               int n;
@@ -9053,12 +8396,7 @@ int new_recompile_block(int addr)
               f_regmap[hr]=regs[i].regmap[hr];
             }
           }
-          if(branch_regs[i].regmap[hr]>64) {
-            if(!((branch_regs[i].dirty>>hr)&1))
-              f_regmap[hr]=branch_regs[i].regmap[hr];
-            else f_regmap[hr]=-1;
-          }
-          else if(branch_regs[i].regmap[hr]>=0) {
+          if(branch_regs[i].regmap[hr]>=0) {
             if(f_regmap[hr]!=branch_regs[i].regmap[hr]) {
               // dealloc old register
               int n;
@@ -9095,15 +8433,7 @@ int new_recompile_block(int addr)
               {
                 //printf("Test %x -> %x, %x %d/%d\n",start+i*4,ba[i],start+j*4,hr,r);
                 if(r<34&&((unneeded_reg[j]>>r)&1)) break;
-                if(r>63&&((unneeded_reg_upper[j]>>(r&63))&1)) break;
-                if(r>63) {
-                  // NB This can exclude the case where the upper-half
-                  // register is lower numbered than the lower-half
-                  // register.  Not sure if it's worth fixing...
-                  if(get_reg(regs[j].regmap,r&63)<0) break;
-                  if(get_reg(regs[j].regmap_entry,r&63)<0) break;
-                  if(regs[j].is32&(1LL<<(r&63))) break;
-                }
+                assert(r < 64);
                 if(regs[j].regmap[hr]==f_regmap[hr]&&(f_regmap[hr]&63)<TEMPREG) {
                   //printf("Hit %x -> %x, %x %d/%d\n",start+i*4,ba[i],start+j*4,hr,r);
                   int k;
@@ -9123,7 +8453,7 @@ int new_recompile_block(int addr)
                         //printf("no-match due to different register\n");
                         break;
                       }
-                      if(itype[k-2]==UJUMP||itype[k-2]==RJUMP||itype[k-2]==CJUMP||itype[k-2]==SJUMP||itype[k-2]==FJUMP) {
+                      if(itype[k-2]==UJUMP||itype[k-2]==RJUMP||itype[k-2]==CJUMP||itype[k-2]==SJUMP) {
                         //printf("no-match due to branch\n");
                         break;
                       }
@@ -9131,22 +8461,9 @@ int new_recompile_block(int addr)
                       if(k>2&&(itype[k-3]==UJUMP||itype[k-3]==RJUMP)&&rt1[k-3]==31) {
                         break;
                       }
-                      if(r>63) {
-                        // NB This can exclude the case where the upper-half
-                        // register is lower numbered than the lower-half
-                        // register.  Not sure if it's worth fixing...
-                        if(get_reg(regs[k-1].regmap,r&63)<0) break;
-                        if(regs[k-1].is32&(1LL<<(r&63))) break;
-                      }
+                      assert(r < 64);
                       k--;
                     }
-                    if(i<slen-1) {
-                      if((regs[k].is32&(1LL<<f_regmap[hr]))!=
-                        (regs[i+2].was32&(1LL<<f_regmap[hr]))) {
-                        //printf("bad match after branch\n");
-                        break;
-                      }
-                    }
                     if(regs[k-1].regmap[hr]==f_regmap[hr]&&regmap_pre[k][hr]==f_regmap[hr]) {
                       //printf("Extend r%d, %x ->\n",hr,start+k*4);
                       while(k<i) {
@@ -9185,12 +8502,10 @@ int new_recompile_block(int addr)
                       branch_regs[i].dirty|=(1<<hr)&regs[i].dirty;
                       branch_regs[i].wasconst&=~(1<<hr);
                       branch_regs[i].isconst&=~(1<<hr);
-                      if(itype[i]!=RJUMP&&itype[i]!=UJUMP&&(source[i]>>16)!=0x1000) {
+                      if (!is_ujump(i)) {
                         regmap_pre[i+2][hr]=f_regmap[hr];
                         regs[i+2].wasdirty&=~(1<<hr);
                         regs[i+2].wasdirty|=(1<<hr)&regs[i].dirty;
-                        assert((branch_regs[i].is32&(1LL<<f_regmap[hr]))==
-                          (regs[i+2].was32&(1LL<<f_regmap[hr])));
                       }
                     }
                   }
@@ -9202,17 +8517,15 @@ int new_recompile_block(int addr)
                     regs[k].dirty&=~(1<<hr);
                     regs[k].wasconst&=~(1<<hr);
                     regs[k].isconst&=~(1<<hr);
-                    if(itype[k]==UJUMP||itype[k]==RJUMP||itype[k]==CJUMP||itype[k]==SJUMP||itype[k]==FJUMP) {
+                    if(itype[k]==UJUMP||itype[k]==RJUMP||itype[k]==CJUMP||itype[k]==SJUMP) {
                       branch_regs[k].regmap_entry[hr]=f_regmap[hr];
                       branch_regs[k].regmap[hr]=f_regmap[hr];
                       branch_regs[k].dirty&=~(1<<hr);
                       branch_regs[k].wasconst&=~(1<<hr);
                       branch_regs[k].isconst&=~(1<<hr);
-                      if(itype[k]!=RJUMP&&itype[k]!=UJUMP&&(source[k]>>16)!=0x1000) {
+                      if (!is_ujump(k)) {
                         regmap_pre[k+2][hr]=f_regmap[hr];
                         regs[k+2].wasdirty&=~(1<<hr);
-                        assert((branch_regs[k].is32&(1LL<<f_regmap[hr]))==
-                          (regs[k+2].was32&(1LL<<f_regmap[hr])));
                       }
                     }
                     else
@@ -9232,16 +8545,12 @@ int new_recompile_block(int addr)
                   //printf("no-match due to different register\n");
                   break;
                 }
-                if((regs[j+1].is32&(1LL<<f_regmap[hr]))!=(regs[j].is32&(1LL<<f_regmap[hr]))) {
-                  //printf("32/64 mismatch %x %d\n",start+j*4,hr);
-                  break;
-                }
-                if(itype[j]==UJUMP||itype[j]==RJUMP||(source[j]>>16)==0x1000)
+                if (is_ujump(j))
                 {
                   // Stop on unconditional branch
                   break;
                 }
-                if(itype[j]==CJUMP||itype[j]==SJUMP||itype[j]==FJUMP)
+                if(itype[j]==CJUMP||itype[j]==SJUMP)
                 {
                   if(ooo[j]) {
                     if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j+1])
@@ -9259,17 +8568,7 @@ int new_recompile_block(int addr)
                   //printf("No free regs for store %x\n",start+j*4);
                   break;
                 }
-                if(f_regmap[hr]>=64) {
-                  if(regs[j].is32&(1LL<<(f_regmap[hr]&63))) {
-                    break;
-                  }
-                  else
-                  {
-                    if(get_reg(regs[j].regmap,f_regmap[hr]&63)<0) {
-                      break;
-                    }
-                  }
-                }
+                assert(f_regmap[hr]<64);
               }
             }
           }
@@ -9280,11 +8579,7 @@ int new_recompile_block(int addr)
       for(hr=0;hr<HOST_REGS;hr++)
       {
         if(hr!=EXCLUDE_REG) {
-          if(regs[i].regmap[hr]>64) {
-            if(!((regs[i].dirty>>hr)&1))
-              f_regmap[hr]=regs[i].regmap[hr];
-          }
-          else if(regs[i].regmap[hr]>=0) {
+          if(regs[i].regmap[hr]>=0) {
             if(f_regmap[hr]!=regs[i].regmap[hr]) {
               // dealloc old register
               int n;
@@ -9355,199 +8650,18 @@ int new_recompile_block(int addr)
       }
       if(itype[i]!=STORE&&itype[i]!=STORELR&&itype[i]!=C1LS&&itype[i]!=SHIFT&&
          itype[i]!=NOP&&itype[i]!=MOV&&itype[i]!=ALU&&itype[i]!=SHIFTIMM&&
-         itype[i]!=IMM16&&itype[i]!=LOAD&&itype[i]!=COP1&&itype[i]!=FLOAT&&
-         itype[i]!=FCONV&&itype[i]!=FCOMP)
+         itype[i]!=IMM16&&itype[i]!=LOAD&&itype[i]!=COP1)
       {
         memcpy(f_regmap,regs[i].regmap,sizeof(f_regmap));
       }
     }
   }
 
-  // Cache memory offset or tlb map pointer if a register is available
-  #ifndef HOST_IMM_ADDR32
-  #ifndef RAM_OFFSET
-  if(0)
-  #endif
-  {
-    int earliest_available[HOST_REGS];
-    int loop_start[HOST_REGS];
-    int score[HOST_REGS];
-    int end[HOST_REGS];
-    int reg=ROREG;
-
-    // Init
-    for(hr=0;hr<HOST_REGS;hr++) {
-      score[hr]=0;earliest_available[hr]=0;
-      loop_start[hr]=MAXBLOCK;
-    }
-    for(i=0;i<slen-1;i++)
-    {
-      // Can't do anything if no registers are available
-      if(count_free_regs(regs[i].regmap)<=minimum_free_regs[i]) {
-        for(hr=0;hr<HOST_REGS;hr++) {
-          score[hr]=0;earliest_available[hr]=i+1;
-          loop_start[hr]=MAXBLOCK;
-        }
-      }
-      if(itype[i]==UJUMP||itype[i]==RJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP) {
-        if(!ooo[i]) {
-          if(count_free_regs(branch_regs[i].regmap)<=minimum_free_regs[i+1]) {
-            for(hr=0;hr<HOST_REGS;hr++) {
-              score[hr]=0;earliest_available[hr]=i+1;
-              loop_start[hr]=MAXBLOCK;
-            }
-          }
-        }else{
-          if(count_free_regs(regs[i].regmap)<=minimum_free_regs[i+1]) {
-            for(hr=0;hr<HOST_REGS;hr++) {
-              score[hr]=0;earliest_available[hr]=i+1;
-              loop_start[hr]=MAXBLOCK;
-            }
-          }
-        }
-      }
-      // Mark unavailable registers
-      for(hr=0;hr<HOST_REGS;hr++) {
-        if(regs[i].regmap[hr]>=0) {
-          score[hr]=0;earliest_available[hr]=i+1;
-          loop_start[hr]=MAXBLOCK;
-        }
-        if(itype[i]==UJUMP||itype[i]==RJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP) {
-          if(branch_regs[i].regmap[hr]>=0) {
-            score[hr]=0;earliest_available[hr]=i+2;
-            loop_start[hr]=MAXBLOCK;
-          }
-        }
-      }
-      // No register allocations after unconditional jumps
-      if(itype[i]==UJUMP||itype[i]==RJUMP||(source[i]>>16)==0x1000)
-      {
-        for(hr=0;hr<HOST_REGS;hr++) {
-          score[hr]=0;earliest_available[hr]=i+2;
-          loop_start[hr]=MAXBLOCK;
-        }
-        i++; // Skip delay slot too
-        //printf("skip delay slot: %x\n",start+i*4);
-      }
-      else
-      // Possible match
-      if(itype[i]==LOAD||itype[i]==LOADLR||
-         itype[i]==STORE||itype[i]==STORELR||itype[i]==C1LS) {
-        for(hr=0;hr<HOST_REGS;hr++) {
-          if(hr!=EXCLUDE_REG) {
-            end[hr]=i-1;
-            for(j=i;j<slen-1;j++) {
-              if(regs[j].regmap[hr]>=0) break;
-              if(itype[j]==UJUMP||itype[j]==RJUMP||itype[j]==CJUMP||itype[j]==SJUMP||itype[j]==FJUMP) {
-                if(branch_regs[j].regmap[hr]>=0) break;
-                if(ooo[j]) {
-                  if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j+1]) break;
-                }else{
-                  if(count_free_regs(branch_regs[j].regmap)<=minimum_free_regs[j+1]) break;
-                }
-              }
-              else if(count_free_regs(regs[j].regmap)<=minimum_free_regs[j]) break;
-              if(itype[j]==UJUMP||itype[j]==RJUMP||itype[j]==CJUMP||itype[j]==SJUMP||itype[j]==FJUMP) {
-                int t=(ba[j]-start)>>2;
-                if(t<j&&t>=earliest_available[hr]) {
-                  if(t==1||(t>1&&itype[t-2]!=UJUMP&&itype[t-2]!=RJUMP)||(t>1&&rt1[t-2]!=31)) { // call/ret assumes no registers allocated
-                    // Score a point for hoisting loop invariant
-                    if(t<loop_start[hr]) loop_start[hr]=t;
-                    //printf("set loop_start: i=%x j=%x (%x)\n",start+i*4,start+j*4,start+t*4);
-                    score[hr]++;
-                    end[hr]=j;
-                  }
-                }
-                else if(t<j) {
-                  if(regs[t].regmap[hr]==reg) {
-                    // Score a point if the branch target matches this register
-                    score[hr]++;
-                    end[hr]=j;
-                  }
-                }
-                if(itype[j+1]==LOAD||itype[j+1]==LOADLR||
-                   itype[j+1]==STORE||itype[j+1]==STORELR||itype[j+1]==C1LS) {
-                  score[hr]++;
-                  end[hr]=j;
-                }
-              }
-              if(itype[j]==UJUMP||itype[j]==RJUMP||(source[j]>>16)==0x1000)
-              {
-                // Stop on unconditional branch
-                break;
-              }
-              else
-              if(itype[j]==LOAD||itype[j]==LOADLR||
-                 itype[j]==STORE||itype[j]==STORELR||itype[j]==C1LS) {
-                score[hr]++;
-                end[hr]=j;
-              }
-            }
-          }
-        }
-        // Find highest score and allocate that register
-        int maxscore=0;
-        for(hr=0;hr<HOST_REGS;hr++) {
-          if(hr!=EXCLUDE_REG) {
-            if(score[hr]>score[maxscore]) {
-              maxscore=hr;
-              //printf("highest score: %d %d (%x->%x)\n",score[hr],hr,start+i*4,start+end[hr]*4);
-            }
-          }
-        }
-        if(score[maxscore]>1)
-        {
-          if(i<loop_start[maxscore]) loop_start[maxscore]=i;
-          for(j=loop_start[maxscore];j<slen&&j<=end[maxscore];j++) {
-            //if(regs[j].regmap[maxscore]>=0) {printf("oops: %x %x was %d=%d\n",loop_start[maxscore]*4+start,j*4+start,maxscore,regs[j].regmap[maxscore]);}
-            assert(regs[j].regmap[maxscore]<0);
-            if(j>loop_start[maxscore]) regs[j].regmap_entry[maxscore]=reg;
-            regs[j].regmap[maxscore]=reg;
-            regs[j].dirty&=~(1<<maxscore);
-            regs[j].wasconst&=~(1<<maxscore);
-            regs[j].isconst&=~(1<<maxscore);
-            if(itype[j]==UJUMP||itype[j]==RJUMP||itype[j]==CJUMP||itype[j]==SJUMP||itype[j]==FJUMP) {
-              branch_regs[j].regmap[maxscore]=reg;
-              branch_regs[j].wasdirty&=~(1<<maxscore);
-              branch_regs[j].dirty&=~(1<<maxscore);
-              branch_regs[j].wasconst&=~(1<<maxscore);
-              branch_regs[j].isconst&=~(1<<maxscore);
-              if(itype[j]!=RJUMP&&itype[j]!=UJUMP&&(source[j]>>16)!=0x1000) {
-                regmap_pre[j+2][maxscore]=reg;
-                regs[j+2].wasdirty&=~(1<<maxscore);
-              }
-              // loop optimization (loop_preload)
-              int t=(ba[j]-start)>>2;
-              if(t==loop_start[maxscore]) {
-                if(t==1||(t>1&&itype[t-2]!=UJUMP&&itype[t-2]!=RJUMP)||(t>1&&rt1[t-2]!=31)) // call/ret assumes no registers allocated
-                  regs[t].regmap_entry[maxscore]=reg;
-              }
-            }
-            else
-            {
-              if(j<1||(itype[j-1]!=RJUMP&&itype[j-1]!=UJUMP&&itype[j-1]!=CJUMP&&itype[j-1]!=SJUMP&&itype[j-1]!=FJUMP)) {
-                regmap_pre[j+1][maxscore]=reg;
-                regs[j+1].wasdirty&=~(1<<maxscore);
-              }
-            }
-          }
-          i=j-1;
-          if(itype[j-1]==RJUMP||itype[j-1]==UJUMP||itype[j-1]==CJUMP||itype[j-1]==SJUMP||itype[j-1]==FJUMP) i++; // skip delay slot
-          for(hr=0;hr<HOST_REGS;hr++) {
-            score[hr]=0;earliest_available[hr]=i+i;
-            loop_start[hr]=MAXBLOCK;
-          }
-        }
-      }
-    }
-  }
-  #endif
-
   // This allocates registers (if possible) one instruction prior
   // to use, which can avoid a load-use penalty on certain CPUs.
   for(i=0;i<slen-1;i++)
   {
-    if(!i||(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP))
+    if(!i||(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP))
     {
       if(!bt[i+1])
       {
@@ -9708,7 +8822,7 @@ int new_recompile_block(int addr)
   /* Pass 7 - Identify 32-bit registers */
   for (i=slen-1;i>=0;i--)
   {
-    if(itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+    if(itype[i]==CJUMP||itype[i]==SJUMP)
     {
       // Conditional branch
       if((source[i]>>16)!=0x1000&&i<slen-2) {
@@ -9743,6 +8857,7 @@ int new_recompile_block(int addr)
     #ifdef __arm__
     printf("pre: r0=%d r1=%d r2=%d r3=%d r4=%d r5=%d r6=%d r7=%d r8=%d r9=%d r10=%d r12=%d\n",regmap_pre[i][0],regmap_pre[i][1],regmap_pre[i][2],regmap_pre[i][3],regmap_pre[i][4],regmap_pre[i][5],regmap_pre[i][6],regmap_pre[i][7],regmap_pre[i][8],regmap_pre[i][9],regmap_pre[i][10],regmap_pre[i][12]);
     #endif
+    #if defined(__i386__) || defined(__x86_64__)
     printf("needs: ");
     if(needed_reg[i]&1) printf("eax ");
     if((needed_reg[i]>>1)&1) printf("ecx ");
@@ -9752,7 +8867,6 @@ int new_recompile_block(int addr)
     if((needed_reg[i]>>6)&1) printf("esi ");
     if((needed_reg[i]>>7)&1) printf("edi ");
     printf("\n");
-    #if defined(__i386__) || defined(__x86_64__)
     printf("entry: eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d\n",regs[i].regmap_entry[0],regs[i].regmap_entry[1],regs[i].regmap_entry[2],regs[i].regmap_entry[3],regs[i].regmap_entry[5],regs[i].regmap_entry[6],regs[i].regmap_entry[7]);
     printf("dirty: ");
     if(regs[i].wasdirty&1) printf("eax ");
@@ -9811,31 +8925,23 @@ int new_recompile_block(int addr)
     if(regs[i].isconst) {
       printf("constants: ");
       #if defined(__i386__) || defined(__x86_64__)
-      if(regs[i].isconst&1) printf("eax=%x ",(int)constmap[i][0]);
-      if((regs[i].isconst>>1)&1) printf("ecx=%x ",(int)constmap[i][1]);
-      if((regs[i].isconst>>2)&1) printf("edx=%x ",(int)constmap[i][2]);
-      if((regs[i].isconst>>3)&1) printf("ebx=%x ",(int)constmap[i][3]);
-      if((regs[i].isconst>>5)&1) printf("ebp=%x ",(int)constmap[i][5]);
-      if((regs[i].isconst>>6)&1) printf("esi=%x ",(int)constmap[i][6]);
-      if((regs[i].isconst>>7)&1) printf("edi=%x ",(int)constmap[i][7]);
+      if(regs[i].isconst&1) printf("eax=%x ",(u_int)constmap[i][0]);
+      if((regs[i].isconst>>1)&1) printf("ecx=%x ",(u_int)constmap[i][1]);
+      if((regs[i].isconst>>2)&1) printf("edx=%x ",(u_int)constmap[i][2]);
+      if((regs[i].isconst>>3)&1) printf("ebx=%x ",(u_int)constmap[i][3]);
+      if((regs[i].isconst>>5)&1) printf("ebp=%x ",(u_int)constmap[i][5]);
+      if((regs[i].isconst>>6)&1) printf("esi=%x ",(u_int)constmap[i][6]);
+      if((regs[i].isconst>>7)&1) printf("edi=%x ",(u_int)constmap[i][7]);
       #endif
-      #ifdef __arm__
-      if(regs[i].isconst&1) printf("r0=%x ",(int)constmap[i][0]);
-      if((regs[i].isconst>>1)&1) printf("r1=%x ",(int)constmap[i][1]);
-      if((regs[i].isconst>>2)&1) printf("r2=%x ",(int)constmap[i][2]);
-      if((regs[i].isconst>>3)&1) printf("r3=%x ",(int)constmap[i][3]);
-      if((regs[i].isconst>>4)&1) printf("r4=%x ",(int)constmap[i][4]);
-      if((regs[i].isconst>>5)&1) printf("r5=%x ",(int)constmap[i][5]);
-      if((regs[i].isconst>>6)&1) printf("r6=%x ",(int)constmap[i][6]);
-      if((regs[i].isconst>>7)&1) printf("r7=%x ",(int)constmap[i][7]);
-      if((regs[i].isconst>>8)&1) printf("r8=%x ",(int)constmap[i][8]);
-      if((regs[i].isconst>>9)&1) printf("r9=%x ",(int)constmap[i][9]);
-      if((regs[i].isconst>>10)&1) printf("r10=%x ",(int)constmap[i][10]);
-      if((regs[i].isconst>>12)&1) printf("r12=%x ",(int)constmap[i][12]);
+      #if defined(__arm__) || defined(__aarch64__)
+      int r;
+      for (r = 0; r < ARRAY_SIZE(constmap[i]); r++)
+        if ((regs[i].isconst >> r) & 1)
+          printf(" r%d=%x", r, (u_int)constmap[i][r]);
       #endif
       printf("\n");
     }
-    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP) {
+    if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP) {
       #if defined(__i386__) || defined(__x86_64__)
       printf("branch(%d): eax=%d ecx=%d edx=%d ebx=%d ebp=%d esi=%d edi=%d dirty: ",i,branch_regs[i].regmap[0],branch_regs[i].regmap[1],branch_regs[i].regmap[2],branch_regs[i].regmap[3],branch_regs[i].regmap[5],branch_regs[i].regmap[6],branch_regs[i].regmap[7]);
       if(branch_regs[i].dirty&1) printf("eax ");
@@ -9868,28 +8974,31 @@ int new_recompile_block(int addr)
   /* Pass 8 - Assembly */
   linkcount=0;stubcount=0;
   ds=0;is_delayslot=0;
-  cop1_usable=0;
-  uint64_t is32_pre=0;
   u_int dirty_pre=0;
-  u_int beginning=(u_int)out;
+  void *beginning=start_block();
   if((u_int)addr&1) {
     ds=1;
     pagespan_ds();
   }
-  u_int instr_addr0_override=0;
+  void *instr_addr0_override = NULL;
 
   if (start == 0x80030000) {
-    // nasty hack for fastbios thing
+    // nasty hack for the fastbios thing
     // override block entry to this code
-    instr_addr0_override=(u_int)out;
+    instr_addr0_override = out;
     emit_movimm(start,0);
     // abuse io address var as a flag that we
     // have already returned here once
-    emit_readword((int)&address,1);
-    emit_writeword(0,(int)&pcaddr);
-    emit_writeword(0,(int)&address);
+    emit_readword(&address,1);
+    emit_writeword(0,&pcaddr);
+    emit_writeword(0,&address);
     emit_cmp(0,1);
-    emit_jne((int)new_dyna_leave);
+    #ifdef __aarch64__
+    emit_jeq(out + 4*2);
+    emit_far_jump(new_dyna_leave);
+    #else
+    emit_jne(new_dyna_leave);
+    #endif
   }
   for(i=0;i<slen;i++)
   {
@@ -9898,65 +9007,62 @@ int new_recompile_block(int addr)
     if(ds) {
       ds=0; // Skip delay slot
       if(bt[i]) assem_debug("OOPS - branch into delay slot\n");
-      instr_addr[i]=0;
+      instr_addr[i] = NULL;
     } else {
       speculate_register_values(i);
       #ifndef DESTRUCTIVE_WRITEBACK
-      if(i<2||(itype[i-2]!=UJUMP&&itype[i-2]!=RJUMP&&(source[i-2]>>16)!=0x1000))
+      if (i < 2 || !is_ujump(i-2))
       {
-        wb_valid(regmap_pre[i],regs[i].regmap_entry,dirty_pre,regs[i].wasdirty,is32_pre,
-              unneeded_reg[i],unneeded_reg_upper[i]);
+        wb_valid(regmap_pre[i],regs[i].regmap_entry,dirty_pre,regs[i].wasdirty,unneeded_reg[i]);
       }
-      if((itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)&&!likely[i]) {
-        is32_pre=branch_regs[i].is32;
+      if((itype[i]==CJUMP||itype[i]==SJUMP)&&!likely[i]) {
         dirty_pre=branch_regs[i].dirty;
       }else{
-        is32_pre=regs[i].is32;
         dirty_pre=regs[i].dirty;
       }
       #endif
       // write back
-      if(i<2||(itype[i-2]!=UJUMP&&itype[i-2]!=RJUMP&&(source[i-2]>>16)!=0x1000))
+      if (i < 2 || !is_ujump(i-2))
       {
-        wb_invalidate(regmap_pre[i],regs[i].regmap_entry,regs[i].wasdirty,regs[i].was32,
-                      unneeded_reg[i],unneeded_reg_upper[i]);
+        wb_invalidate(regmap_pre[i],regs[i].regmap_entry,regs[i].wasdirty,unneeded_reg[i]);
         loop_preload(regmap_pre[i],regs[i].regmap_entry);
       }
       // branch target entry point
-      instr_addr[i]=(u_int)out;
+      instr_addr[i] = out;
       assem_debug("<->\n");
+      drc_dbg_emit_do_cmp(i);
+
       // load regs
       if(regs[i].regmap_entry[HOST_CCREG]==CCREG&&regs[i].regmap[HOST_CCREG]!=CCREG)
-        wb_register(CCREG,regs[i].regmap_entry,regs[i].wasdirty,regs[i].was32);
-      load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs1[i],rs2[i]);
+        wb_register(CCREG,regs[i].regmap_entry,regs[i].wasdirty);
+      load_regs(regs[i].regmap_entry,regs[i].regmap,rs1[i],rs2[i]);
       address_generation(i,&regs[i],regs[i].regmap_entry);
-      load_consts(regmap_pre[i],regs[i].regmap,regs[i].was32,i);
-      if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP||itype[i]==FJUMP)
+      load_consts(regmap_pre[i],regs[i].regmap,i);
+      if(itype[i]==RJUMP||itype[i]==UJUMP||itype[i]==CJUMP||itype[i]==SJUMP)
       {
         // Load the delay slot registers if necessary
         if(rs1[i+1]!=rs1[i]&&rs1[i+1]!=rs2[i]&&(rs1[i+1]!=rt1[i]||rt1[i]==0))
-          load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs1[i+1],rs1[i+1]);
+          load_regs(regs[i].regmap_entry,regs[i].regmap,rs1[i+1],rs1[i+1]);
         if(rs2[i+1]!=rs1[i+1]&&rs2[i+1]!=rs1[i]&&rs2[i+1]!=rs2[i]&&(rs2[i+1]!=rt1[i]||rt1[i]==0))
-          load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs2[i+1],rs2[i+1]);
+          load_regs(regs[i].regmap_entry,regs[i].regmap,rs2[i+1],rs2[i+1]);
         if(itype[i+1]==STORE||itype[i+1]==STORELR||(opcode[i+1]&0x3b)==0x39||(opcode[i+1]&0x3b)==0x3a)
-          load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,INVCP,INVCP);
+          load_regs(regs[i].regmap_entry,regs[i].regmap,INVCP,INVCP);
       }
       else if(i+1<slen)
       {
         // Preload registers for following instruction
         if(rs1[i+1]!=rs1[i]&&rs1[i+1]!=rs2[i])
           if(rs1[i+1]!=rt1[i]&&rs1[i+1]!=rt2[i])
-            load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs1[i+1],rs1[i+1]);
+            load_regs(regs[i].regmap_entry,regs[i].regmap,rs1[i+1],rs1[i+1]);
         if(rs2[i+1]!=rs1[i+1]&&rs2[i+1]!=rs1[i]&&rs2[i+1]!=rs2[i])
           if(rs2[i+1]!=rt1[i]&&rs2[i+1]!=rt2[i])
-            load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,rs2[i+1],rs2[i+1]);
+            load_regs(regs[i].regmap_entry,regs[i].regmap,rs2[i+1],rs2[i+1]);
       }
       // TODO: if(is_ooo(i)) address_generation(i+1);
-      if(itype[i]==CJUMP||itype[i]==FJUMP)
-        load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,CCREG,CCREG);
+      if(itype[i]==CJUMP)
+        load_regs(regs[i].regmap_entry,regs[i].regmap,CCREG,CCREG);
       if(itype[i]==STORE||itype[i]==STORELR||(opcode[i]&0x3b)==0x39||(opcode[i]&0x3b)==0x3a)
-        load_regs(regs[i].regmap_entry,regs[i].regmap,regs[i].was32,INVCP,INVCP);
-      if(bt[i]) cop1_usable=0;
+        load_regs(regs[i].regmap_entry,regs[i].regmap,INVCP,INVCP);
       // assemble
       switch(itype[i]) {
         case ALU:
@@ -9987,12 +9093,6 @@ int new_recompile_block(int addr)
           c2ls_assemble(i,&regs[i]);break;
         case C2OP:
           c2op_assemble(i,&regs[i]);break;
-        case FCONV:
-          fconv_assemble(i,&regs[i]);break;
-        case FLOAT:
-          float_assemble(i,&regs[i]);break;
-        case FCOMP:
-          fcomp_assemble(i,&regs[i]);break;
         case MULTDIV:
           multdiv_assemble(i,&regs[i]);break;
         case MOV:
@@ -10011,53 +9111,51 @@ int new_recompile_block(int addr)
           cjump_assemble(i,&regs[i]);ds=1;break;
         case SJUMP:
           sjump_assemble(i,&regs[i]);ds=1;break;
-        case FJUMP:
-          fjump_assemble(i,&regs[i]);ds=1;break;
         case SPAN:
           pagespan_assemble(i,&regs[i]);break;
       }
-      if(itype[i]==UJUMP||itype[i]==RJUMP||(source[i]>>16)==0x1000)
+      if (is_ujump(i))
         literal_pool(1024);
       else
         literal_pool_jumpover(256);
     }
   }
-  //assert(itype[i-2]==UJUMP||itype[i-2]==RJUMP||(source[i-2]>>16)==0x1000);
+  //assert(is_ujump(i-2));
   // If the block did not end with an unconditional branch,
   // add a jump to the next instruction.
   if(i>1) {
-    if(itype[i-2]!=UJUMP&&itype[i-2]!=RJUMP&&(source[i-2]>>16)!=0x1000&&itype[i-1]!=SPAN) {
-      assert(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP);
+    if(!is_ujump(i-2)&&itype[i-1]!=SPAN) {
+      assert(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP);
       assert(i==slen);
-      if(itype[i-2]!=CJUMP&&itype[i-2]!=SJUMP&&itype[i-2]!=FJUMP) {
-        store_regs_bt(regs[i-1].regmap,regs[i-1].is32,regs[i-1].dirty,start+i*4);
+      if(itype[i-2]!=CJUMP&&itype[i-2]!=SJUMP) {
+        store_regs_bt(regs[i-1].regmap,regs[i-1].dirty,start+i*4);
         if(regs[i-1].regmap[HOST_CCREG]!=CCREG)
           emit_loadreg(CCREG,HOST_CCREG);
         emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i-1]+1),HOST_CCREG);
       }
       else if(!likely[i-2])
       {
-        store_regs_bt(branch_regs[i-2].regmap,branch_regs[i-2].is32,branch_regs[i-2].dirty,start+i*4);
+        store_regs_bt(branch_regs[i-2].regmap,branch_regs[i-2].dirty,start+i*4);
         assert(branch_regs[i-2].regmap[HOST_CCREG]==CCREG);
       }
       else
       {
-        store_regs_bt(regs[i-2].regmap,regs[i-2].is32,regs[i-2].dirty,start+i*4);
+        store_regs_bt(regs[i-2].regmap,regs[i-2].dirty,start+i*4);
         assert(regs[i-2].regmap[HOST_CCREG]==CCREG);
       }
-      add_to_linker((int)out,start+i*4,0);
+      add_to_linker(out,start+i*4,0);
       emit_jmp(0);
     }
   }
   else
   {
     assert(i>0);
-    assert(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP&&itype[i-1]!=FJUMP);
-    store_regs_bt(regs[i-1].regmap,regs[i-1].is32,regs[i-1].dirty,start+i*4);
+    assert(itype[i-1]!=UJUMP&&itype[i-1]!=CJUMP&&itype[i-1]!=SJUMP&&itype[i-1]!=RJUMP);
+    store_regs_bt(regs[i-1].regmap,regs[i-1].dirty,start+i*4);
     if(regs[i-1].regmap[HOST_CCREG]!=CCREG)
       emit_loadreg(CCREG,HOST_CCREG);
     emit_addimm(HOST_CCREG,CLOCK_ADJUST(ccadj[i-1]+1),HOST_CCREG);
-    add_to_linker((int)out,start+i*4,0);
+    add_to_linker(out,start+i*4,0);
     emit_jmp(0);
   }
 
@@ -10065,7 +9163,7 @@ int new_recompile_block(int addr)
   // Stubs
   for(i=0;i<stubcount;i++)
   {
-    switch(stubs[i][0])
+    switch(stubs[i].type)
     {
       case LOADB_STUB:
       case LOADH_STUB:
@@ -10096,29 +9194,30 @@ int new_recompile_block(int addr)
   /* Pass 9 - Linker */
   for(i=0;i<linkcount;i++)
   {
-    assem_debug("%8x -> %8x\n",link_addr[i][0],link_addr[i][1]);
+    assem_debug("%p -> %8x\n",link_addr[i].addr,link_addr[i].target);
     literal_pool(64);
-    if(!link_addr[i][2])
+    if (!link_addr[i].ext)
     {
-      void *stub=out;
-      void *addr=check_addr(link_addr[i][1]);
-      emit_extjump(link_addr[i][0],link_addr[i][1]);
-      if(addr) {
-        set_jump_target(link_addr[i][0],(int)addr);
-        add_link(link_addr[i][1],stub);
+      void *stub = out;
+      void *addr = check_addr(link_addr[i].target);
+      emit_extjump(link_addr[i].addr, link_addr[i].target);
+      if (addr) {
+        set_jump_target(link_addr[i].addr, addr);
+        add_link(link_addr[i].target,stub);
       }
-      else set_jump_target(link_addr[i][0],(int)stub);
+      else
+        set_jump_target(link_addr[i].addr, stub);
     }
     else
     {
       // Internal branch
-      int target=(link_addr[i][1]-start)>>2;
+      int target=(link_addr[i].target-start)>>2;
       assert(target>=0&&target<slen);
       assert(instr_addr[target]);
       //#ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
-      //set_jump_target_fillslot(link_addr[i][0],instr_addr[target],link_addr[i][2]>>1);
+      //set_jump_target_fillslot(link_addr[i].addr,instr_addr[target],link_addr[i].ext>>1);
       //#else
-      set_jump_target(link_addr[i][0],instr_addr[target]);
+      set_jump_target(link_addr[i].addr, instr_addr[target]);
       //#endif
     }
   }
@@ -10135,22 +9234,20 @@ int new_recompile_block(int addr)
         u_int vpage=get_vpage(vaddr);
         literal_pool(256);
         {
-          assem_debug("%8x (%d) <- %8x\n",instr_addr[i],i,start+i*4);
+          assem_debug("%p (%d) <- %8x\n",instr_addr[i],i,start+i*4);
           assem_debug("jump_in: %x\n",start+i*4);
-          ll_add(jump_dirty+vpage,vaddr,(void *)out);
-          int entry_point=do_dirty_stub(i);
-          ll_add_flags(jump_in+page,vaddr,state_rflags,(void *)entry_point);
+          ll_add(jump_dirty+vpage,vaddr,out);
+          void *entry_point = do_dirty_stub(i);
+          ll_add_flags(jump_in+page,vaddr,state_rflags,entry_point);
           // If there was an existing entry in the hash table,
           // replace it with the new address.
           // Don't add new entries.  We'll insert the
           // ones that actually get used in check_addr().
-          int *ht_bin=hash_table[((vaddr>>16)^vaddr)&0xFFFF];
-          if(ht_bin[0]==vaddr) {
-            ht_bin[1]=entry_point;
-          }
-          if(ht_bin[2]==vaddr) {
-            ht_bin[3]=entry_point;
-          }
+          struct ht_entry *ht_bin = hash_table_get(vaddr);
+          if (ht_bin->vaddr[0] == vaddr)
+            ht_bin->tcaddr[0] = entry_point;
+          if (ht_bin->vaddr[1] == vaddr)
+            ht_bin->tcaddr[1] = entry_point;
         }
       }
     }
@@ -10161,18 +9258,17 @@ int new_recompile_block(int addr)
   // Align code
   if(((u_int)out)&7) emit_addnop(13);
   #endif
-  assert((u_int)out-beginning<MAX_OUTPUT_BLOCK_SIZE);
-  //printf("shadow buffer: %x-%x\n",(int)copy,(int)copy+slen*4);
+  assert(out - (u_char *)beginning < MAX_OUTPUT_BLOCK_SIZE);
+  //printf("shadow buffer: %p-%p\n",copy,(u_char *)copy+slen*4);
   memcpy(copy,source,slen*4);
   copy+=slen*4;
 
-  #ifdef __arm__
-  __clear_cache((void *)beginning,out);
-  #endif
+  end_block(beginning);
 
   // If we're within 256K of the end of the buffer,
   // start over from the beginning. (Is 256K enough?)
-  if((u_int)out>(u_int)BASE_ADDR+(1<<TARGET_SIZE_2)-MAX_OUTPUT_BLOCK_SIZE) out=(u_char *)BASE_ADDR;
+  if (out > ndrc->translation_cache + sizeof(ndrc->translation_cache) - MAX_OUTPUT_BLOCK_SIZE)
+    out = ndrc->translation_cache;
 
   // Trap writes to any of the pages we compiled
   for(i=start>>12;i<=(start+slen*4)>>12;i++) {
@@ -10189,11 +9285,11 @@ int new_recompile_block(int addr)
 
   /* Pass 10 - Free memory by expiring oldest blocks */
 
-  int end=((((int)out-(int)BASE_ADDR)>>(TARGET_SIZE_2-16))+16384)&65535;
+  int end=(((out-ndrc->translation_cache)>>(TARGET_SIZE_2-16))+16384)&65535;
   while(expirep!=end)
   {
     int shift=TARGET_SIZE_2-3; // Divide into 8 blocks
-    int base=(int)BASE_ADDR+((expirep>>13)<<shift); // Base address of this block
+    uintptr_t base=(uintptr_t)ndrc->translation_cache+((expirep>>13)<<shift); // Base address of this block
     inv_debug("EXP: Phase %d\n",expirep);
     switch((expirep>>11)&3)
     {
@@ -10212,27 +9308,27 @@ int new_recompile_block(int addr)
       case 2:
         // Clear hash table
         for(i=0;i<32;i++) {
-          int *ht_bin=hash_table[((expirep&2047)<<5)+i];
-          if((ht_bin[3]>>shift)==(base>>shift) ||
-             ((ht_bin[3]-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(base>>shift)) {
-            inv_debug("EXP: Remove hash %x -> %x\n",ht_bin[2],ht_bin[3]);
-            ht_bin[2]=ht_bin[3]=-1;
-          }
-          if((ht_bin[1]>>shift)==(base>>shift) ||
-             ((ht_bin[1]-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(base>>shift)) {
-            inv_debug("EXP: Remove hash %x -> %x\n",ht_bin[0],ht_bin[1]);
-            ht_bin[0]=ht_bin[2];
-            ht_bin[1]=ht_bin[3];
-            ht_bin[2]=ht_bin[3]=-1;
+          struct ht_entry *ht_bin = &hash_table[((expirep&2047)<<5)+i];
+          if (((uintptr_t)ht_bin->tcaddr[1]>>shift) == (base>>shift) ||
+             (((uintptr_t)ht_bin->tcaddr[1]-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(base>>shift)) {
+            inv_debug("EXP: Remove hash %x -> %p\n",ht_bin->vaddr[1],ht_bin->tcaddr[1]);
+            ht_bin->vaddr[1] = -1;
+            ht_bin->tcaddr[1] = NULL;
+          }
+          if (((uintptr_t)ht_bin->tcaddr[0]>>shift) == (base>>shift) ||
+             (((uintptr_t)ht_bin->tcaddr[0]-MAX_OUTPUT_BLOCK_SIZE)>>shift)==(base>>shift)) {
+            inv_debug("EXP: Remove hash %x -> %p\n",ht_bin->vaddr[0],ht_bin->tcaddr[0]);
+            ht_bin->vaddr[0] = ht_bin->vaddr[1];
+            ht_bin->tcaddr[0] = ht_bin->tcaddr[1];
+            ht_bin->vaddr[1] = -1;
+            ht_bin->tcaddr[1] = NULL;
           }
         }
         break;
       case 3:
         // Clear jump_out
-        #ifdef __arm__
         if((expirep&2047)==0)
           do_clear_cache();
-        #endif
         ll_remove_matching_addrs(jump_out+(expirep&2047),base,shift);
         ll_remove_matching_addrs(jump_out+2048+(expirep&2047),base,shift);
         break;