X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fassem_arm64.c;h=17517eff7c65925772b3046cc5bb864adc3991fd;hp=a0c628b5892242ff230593bf9e5f6160bd212a2c;hb=e3c6bdb5e46f72f063bb7f588da6588ac1893b17;hpb=d1e4ebd9988a9a5d9fb38b89f19e24b9ab6029d7

diff --git a/libpcsxcore/new_dynarec/assem_arm64.c b/libpcsxcore/new_dynarec/assem_arm64.c
index a0c628b5..17517eff 100644
--- a/libpcsxcore/new_dynarec/assem_arm64.c
+++ b/libpcsxcore/new_dynarec/assem_arm64.c
@@ -20,16 +20,9 @@
  *   51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.          *
  * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
 
+#include "pcnt.h"
 #include "arm_features.h"
 
-#if   defined(BASE_ADDR_FIXED)
-#elif defined(BASE_ADDR_DYNAMIC)
-u_char *translation_cache;
-#else
-u_char translation_cache[1 << TARGET_SIZE_2] __attribute__((aligned(4096)));
-#endif
-static u_int needs_clear_cache[1<<(TARGET_SIZE_2-17)];
-
 #define CALLER_SAVE_REGS 0x0007ffff
 
 #define unused __attribute__((unused))
@@ -43,25 +36,26 @@ static void set_jump_target(void *addr, void *target)
   u_int *ptr = addr;
   intptr_t offset = (u_char *)target - (u_char *)addr;
 
-  if((*ptr&0xFC000000)==0x14000000) {
+  if ((*ptr&0xFC000000) == 0x14000000) { // b
     assert(offset>=-134217728LL&&offset<134217728LL);
     *ptr=(*ptr&0xFC000000)|((offset>>2)&0x3ffffff);
   }
-  else if((*ptr&0xff000000)==0x54000000) {
+  else if ((*ptr&0xff000000) == 0x54000000 // b.cond
+        || (*ptr&0x7e000000) == 0x34000000) { // cbz/cbnz
     // Conditional branch are limited to +/- 1MB
     // block max size is 256k so branching beyond the +/- 1MB limit
     // should only happen when jumping to an already compiled block (see add_link)
     // a workaround would be to do a trampoline jump via a stub at the end of the block
-    assert(offset>=-1048576LL&&offset<1048576LL);
+    assert(-1048576 <= offset && offset < 1048576);
     *ptr=(*ptr&0xFF00000F)|(((offset>>2)&0x7ffff)<<5);
   }
-  else if((*ptr&0x9f000000)==0x10000000) { //adr
+  else if((*ptr&0x9f000000)==0x10000000) { // adr
     // generated by do_miniht_insert
     assert(offset>=-1048576LL&&offset<1048576LL);
     *ptr=(*ptr&0x9F00001F)|(offset&0x3)<<29|((offset>>2)&0x7ffff)<<5;
   }
   else
-    assert(0); // should not happen
+    abort(); // should not happen
 }
 
 // from a pointer to external jump stub (which was produced by emit_extjump2)
@@ -75,62 +69,21 @@ static void *find_extjump_insn(void *stub)
 }
 
 // find where external branch is liked to using addr of it's stub:
-// get address that insn one after stub loads (dyna_linker arg1),
+// get address that the stub loads (dyna_linker arg1),
 // treat it as a pointer to branch insn,
 // return addr where that branch jumps to
 static void *get_pointer(void *stub)
 {
   int *i_ptr = find_extjump_insn(stub);
-  assert((*i_ptr&0xfc000000) == 0x14000000); // b
-  return (u_char *)i_ptr+(((signed int)(*i_ptr<<6)>>6)<<2);
-}
-
-// Find the "clean" entry point from a "dirty" entry point
-// by skipping past the call to verify_code
-static void *get_clean_addr(void *addr)
-{
+  if ((*i_ptr&0xfc000000) == 0x14000000)  // b
+    return i_ptr + ((signed int)(*i_ptr<<6)>>6);
+  if ((*i_ptr&0xff000000) == 0x54000000     // b.cond
+      || (*i_ptr&0x7e000000) == 0x34000000) // cbz/cbnz
+    return i_ptr + ((signed int)(*i_ptr<<8)>>13);
   assert(0);
   return NULL;
 }
 
-static int verify_dirty(u_int *ptr)
-{
-  assert(0);
-  return 0;
-}
-
-static int isclean(void *addr)
-{
-  u_int *ptr = addr;
-  return (*ptr >> 24) != 0x58; // the only place ldr (literal) is used
-}
-
-static uint64_t get_from_ldr_literal(const u_int *i)
-{
-  signed int ofs;
-  assert((i[0] & 0xff000000) == 0x58000000);
-  ofs = i[0] << 8;
-  ofs >>= 5+8;
-  return *(uint64_t *)(i + ofs);
-}
-
-static uint64_t get_from_movz(const u_int *i)
-{
-  assert((i[0] & 0x7fe00000) == 0x52800000);
-  return (i[0] >> 5) & 0xffff;
-}
-
-// get source that block at addr was compiled from (host pointers)
-static void get_bounds(void *addr, u_char **start, u_char **end)
-{
-  const u_int *ptr = addr;
-  assert((ptr[0] & 0xff00001f) == 0x58000001); // ldr x1, source
-  assert((ptr[1] & 0xff00001f) == 0x58000002); // ldr x2, copy
-  assert((ptr[2] & 0xffe0001f) == 0x52800003); // movz w3, #slen*4
-  *start = (u_char *)get_from_ldr_literal(&ptr[0]);
-  *end = *start + get_from_movz(&ptr[2]);
-}
-
 // Allocate a specific ARM register.
 static void alloc_arm_reg(struct regstat *cur,int i,signed char reg,int hr)
 {
@@ -208,6 +161,13 @@ static u_int rm_rd(u_int rm, u_int rd)
 }
 */
 
+static u_int rn_rd(u_int rn, u_int rd)
+{
+  assert(rn < 31);
+  assert(rd < 31);
+  return (rn << 5) | rd;
+}
+
 static u_int rm_rn_rd(u_int rm, u_int rn, u_int rd)
 {
   assert(rm < 32);
@@ -216,6 +176,12 @@ static u_int rm_rn_rd(u_int rm, u_int rn, u_int rd)
   return (rm << 16) | (rn << 5) | rd;
 }
 
+static u_int rm_ra_rn_rd(u_int rm, u_int ra, u_int rn, u_int rd)
+{
+  assert(ra < 32);
+  return rm_rn_rd(rm, rn, rd) | (ra << 10);
+}
+
 static u_int imm7_rt2_rn_rt(u_int imm7, u_int rt2, u_int rn, u_int rt)
 {
   assert(imm7 < 0x80);
@@ -304,7 +270,7 @@ static uint32_t is_mask(u_int value)
 // non-empty sequence of ones (possibly rotated) with the remainder zero.
 static uint32_t is_rotated_mask(u_int value)
 {
-  if (value == 0)
+  if (value == 0 || value == ~0)
     return 0;
   if (is_mask((value - 1) | value))
     return 1;
@@ -328,11 +294,11 @@ static void gen_logical_imm(u_int value, u_int *immr, u_int *imms)
     lzeros = __builtin_clz(value);
     tzeros = __builtin_ctz(value);
     ones = 32 - lzeros - tzeros;
-    *immr = 31 - tzeros;
+    *immr = lzeros;
     *imms = 31 - ones;
     return;
   }
-  assert(0);
+  abort();
 }
 
 static void emit_mov(u_int rs, u_int rt)
@@ -347,13 +313,6 @@ static void emit_mov64(u_int rs, u_int rt)
   output_w32(0xaa000000 | rm_rn_rd(rs, WZR, rt));
 }
 
-static void emit_movs(u_int rs, u_int rt)
-{
-  assert(0); // misleading
-  assem_debug("movs %s,%s\n", regname[rt], regname[rs]);
-  output_w32(0x31000000 | imm12_rn_rd(0, rs, rt));
-}
-
 static void emit_add(u_int rs1, u_int rs2, u_int rt)
 {
   assem_debug("add %s,%s,%s\n", regname[rt], regname[rs1], regname[rs2]);
@@ -366,16 +325,9 @@ static void emit_add64(u_int rs1, u_int rs2, u_int rt)
   output_w32(0x8b000000 | rm_rn_rd(rs2, rs1, rt));
 }
 
-#pragma GCC diagnostic ignored "-Wunused-function"
-static void emit_adds(u_int rs1, u_int rs2, u_int rt)
-{
-  assem_debug("adds %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
-  output_w32(0x2b000000 | rm_rn_rd(rs2, rs1, rt));
-}
-
 static void emit_adds64(u_int rs1, u_int rs2, u_int rt)
 {
-  assem_debug("adds %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  assem_debug("adds %s,%s,%s\n",regname64[rt],regname64[rs1],regname64[rs2]);
   output_w32(0xab000000 | rm_rn_rd(rs2, rs1, rt));
 }
 
@@ -391,6 +343,12 @@ static void emit_sub(u_int rs1, u_int rs2, u_int rt)
   output_w32(0x4b000000 | rm_imm6_rn_rd(rs2, 0, rs1, rt));
 }
 
+static void emit_sub_asrimm(u_int rs1, u_int rs2, u_int shift, u_int rt)
+{
+  assem_debug("sub %s,%s,%s,asr #%u\n",regname[rt],regname[rs1],regname[rs2],shift);
+  output_w32(0x4b800000 | rm_imm6_rn_rd(rs2, shift, rs1, rt));
+}
+
 static void emit_movz(u_int imm, u_int rt)
 {
   assem_debug("movz %s,#%#x\n", regname[rt], imm);
@@ -424,7 +382,7 @@ static void emit_movk(u_int imm,u_int rt)
 static void emit_movk_lsl16(u_int imm,u_int rt)
 {
   assert(imm<65536);
-  assem_debug("movk %s, #%#x, lsl #16\n", regname[rt], imm);
+  assem_debug("movk %s,#%#x,lsl #16\n", regname[rt], imm);
   output_w32(0x72a00000 | imm16_rd(imm, rt));
 }
 
@@ -463,7 +421,7 @@ static void emit_readword(void *addr, u_int rt)
     output_w32(0xb9400000 | imm12_rn_rd(offset >> 2, FP, rt));
   }
   else
-    assert(0);
+    abort();
 }
 
 static void emit_readdword(void *addr, u_int rt)
@@ -473,6 +431,17 @@ static void emit_readdword(void *addr, u_int rt)
     assem_debug("ldr %s,[x%d+%#lx]\n", regname64[rt], FP, offset);
     output_w32(0xf9400000 | imm12_rn_rd(offset >> 3, FP, rt));
   }
+  else
+    abort();
+}
+
+static void emit_readshword(void *addr, u_int rt)
+{
+  uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
+  if (!(offset & 1) && offset <= 8190) {
+    assem_debug("ldrsh %s,[x%d+%#lx]\n", regname[rt], FP, offset);
+    output_w32(0x79c00000 | imm12_rn_rd(offset >> 1, FP, rt));
+  }
   else
     assert(0);
 }
@@ -516,10 +485,10 @@ static void emit_writedword(u_int rt, void *addr)
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
   if (!(offset & 7) && offset <= 32760) {
     assem_debug("str %s,[x%d+%#lx]\n", regname64[rt], FP, offset);
-    output_w32(0xf9000000 | imm12_rn_rd(offset >> 2, FP, rt));
+    output_w32(0xf9000000 | imm12_rn_rd(offset >> 3, FP, rt));
   }
   else
-    assert(0);
+    abort();
 }
 
 static void emit_storereg(u_int r, u_int hr)
@@ -547,13 +516,7 @@ static void emit_testimm(u_int rs, u_int imm)
   assem_debug("tst %s,#%#x\n", regname[rs], imm);
   assert(is_rotated_mask(imm)); // good enough for PCSX
   gen_logical_imm(imm, &immr, &imms);
-  output_w32(0xb9000000 | n_immr_imms_rn_rd(0, immr, imms, rs, WZR));
-}
-
-static void emit_testeqimm(u_int rs,int imm)
-{
-  assem_debug("tsteq %s,$%d\n",regname[rs],imm);
-  assert(0); // TODO eliminate emit_testeqimm
+  output_w32(0x72000000 | n_immr_imms_rn_rd(0, immr, imms, rs, WZR));
 }
 
 static void emit_not(u_int rs,u_int rt)
@@ -562,12 +525,6 @@ static void emit_not(u_int rs,u_int rt)
   output_w32(0x2a200000 | rm_rn_rd(rs, WZR, rt));
 }
 
-static void emit_mvnmi(u_int rs,u_int rt)
-{
-  assem_debug("mvnmi %s,%s\n",regname[rt],regname[rs]);
-  assert(0); // eliminate
-}
-
 static void emit_and(u_int rs1,u_int rs2,u_int rt)
 {
   assem_debug("and %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
@@ -580,6 +537,12 @@ static void emit_or(u_int rs1,u_int rs2,u_int rt)
   output_w32(0x2a000000 | rm_rn_rd(rs2, rs1, rt));
 }
 
+static void emit_bic(u_int rs1,u_int rs2,u_int rt)
+{
+  assem_debug("bic %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0x0a200000 | rm_rn_rd(rs2, rs1, rt));
+}
+
 static void emit_orrshl_imm(u_int rs,u_int imm,u_int rt)
 {
   assem_debug("orr %s,%s,%s,lsl #%d\n",regname[rt],regname[rt],regname[rs],imm);
@@ -592,12 +555,24 @@ static void emit_orrshr_imm(u_int rs,u_int imm,u_int rt)
   output_w32(0x2a400000 | rm_imm6_rn_rd(rs, imm, rt, rt));
 }
 
+static void emit_bicsar_imm(u_int rs,u_int imm,u_int rt)
+{
+  assem_debug("bic %s,%s,%s,asr #%d\n",regname[rt],regname[rt],regname[rs],imm);
+  output_w32(0x0aa00000 | rm_imm6_rn_rd(rs, imm, rt, rt));
+}
+
 static void emit_xor(u_int rs1,u_int rs2,u_int rt)
 {
   assem_debug("eor %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
   output_w32(0x4a000000 | rm_rn_rd(rs2, rs1, rt));
 }
 
+static void emit_xorsar_imm(u_int rs1, u_int rs2, u_int imm, u_int rt)
+{
+  assem_debug("eor %s,%s,%s,asr #%d\n",regname[rt],regname[rs1],regname[rs2],imm);
+  output_w32(0x4a800000 | rm_imm6_rn_rd(rs2, imm, rs1, rt));
+}
+
 static void emit_addimm_s(u_int s, u_int is64, u_int rs, uintptr_t imm, u_int rt)
 {
   unused const char *st = s ? "s" : "";
@@ -608,7 +583,7 @@ static void emit_addimm_s(u_int s, u_int is64, u_int rs, uintptr_t imm, u_int rt
     output_w32(0x11000000 | is64 | s | imm12_rn_rd(imm, rs, rt));
   }
   else if (-imm < 4096) {
-    assem_debug("sub%s %s,%s,%#lx\n", st, regname[rt], regname[rs], imm);
+    assem_debug("sub%s %s,%s,%#lx\n", st, regname[rt], regname[rs], -imm);
     output_w32(0x51000000 | is64 | s | imm12_rn_rd(-imm, rs, rt));
   }
   else if (imm < 16777216) {
@@ -616,7 +591,7 @@ static void emit_addimm_s(u_int s, u_int is64, u_int rs, uintptr_t imm, u_int rt
     output_w32(0x11400000 | is64 | imm12_rn_rd(imm >> 12, rs, rt));
     if ((imm & 0xfff) || s) {
       assem_debug("add%s %s,%s,#%#lx\n",st,regname[rt],regname[rs],imm&0xfff);
-      output_w32(0x11000000 | is64 | s | imm12_rn_rd(imm, rt, rt));
+      output_w32(0x11000000 | is64 | s | imm12_rn_rd(imm & 0xfff, rt, rt));
     }
   }
   else if (-imm < 16777216) {
@@ -628,7 +603,7 @@ static void emit_addimm_s(u_int s, u_int is64, u_int rs, uintptr_t imm, u_int rt
     }
   }
   else
-    assert(0);
+    abort();
 }
 
 static void emit_addimm(u_int rs, uintptr_t imm, u_int rt)
@@ -720,15 +695,16 @@ static void emit_shlimm(u_int rs,u_int imm,u_int rt)
   output_w32(0x53000000 | n_immr_imms_rn_rd(0, (31-imm)+1, 31-imm, rs, rt));
 }
 
-static unused void emit_lslpls_imm(u_int rs,int imm,u_int rt)
+static void emit_shrimm(u_int rs,u_int imm,u_int rt)
 {
-  assert(0); // eliminate
+  assem_debug("lsr %s,%s,#%d\n",regname[rt],regname[rs],imm);
+  output_w32(0x53000000 | n_immr_imms_rn_rd(0, imm, 31, rs, rt));
 }
 
-static void emit_shrimm(u_int rs,u_int imm,u_int rt)
+static void emit_shrimm64(u_int rs,u_int imm,u_int rt)
 {
   assem_debug("lsr %s,%s,#%d\n",regname[rt],regname[rs],imm);
-  output_w32(0x53000000 | n_immr_imms_rn_rd(0, imm, 31, rs, rt));
+  output_w32(0xd3400000 | n_immr_imms_rn_rd(0, imm, 63, rs, rt));
 }
 
 static void emit_sarimm(u_int rs,u_int imm,u_int rt)
@@ -739,7 +715,7 @@ static void emit_sarimm(u_int rs,u_int imm,u_int rt)
 
 static void emit_rorimm(u_int rs,u_int imm,u_int rt)
 {
-  assem_debug("ror %s,%s,#%d",regname[rt],regname[rs],imm);
+  assem_debug("ror %s,%s,#%d\n",regname[rt],regname[rs],imm);
   output_w32(0x13800000 | rm_imm6_rn_rd(rs, imm, rs, rt));
 }
 
@@ -751,7 +727,7 @@ static void emit_signextend16(u_int rs, u_int rt)
 
 static void emit_shl(u_int rs,u_int rshift,u_int rt)
 {
-  assem_debug("lsl %s,%s,%s",regname[rt],regname[rs],regname[rshift]);
+  assem_debug("lsl %s,%s,%s\n",regname[rt],regname[rs],regname[rshift]);
   output_w32(0x1ac02000 | rm_rn_rd(rshift, rs, rt));
 }
 
@@ -778,7 +754,7 @@ static void emit_cmpimm(u_int rs, u_int imm)
     output_w32(0x31000000 | imm12_rn_rd(-imm, rs, WZR));
   }
   else if (imm < 16777216 && !(imm & 0xfff)) {
-    assem_debug("cmp %s,#%#x,lsl #12\n", regname[rs], imm >> 12);
+    assem_debug("cmp %s,#%#x\n", regname[rs], imm);
     output_w32(0x71400000 | imm12_rn_rd(imm >> 12, rs, WZR));
   }
   else {
@@ -819,9 +795,10 @@ static void emit_cmovb_imm(int imm,u_int rt)
   emit_cmov_imm(COND_CC, COND_CS, imm, rt);
 }
 
-static void emit_cmovs_imm(int imm,u_int rt)
+static void emit_cmoveq_reg(u_int rs,u_int rt)
 {
-  emit_cmov_imm(COND_MI, COND_PL, imm, rt);
+  assem_debug("csel %s,%s,%s,eq\n",regname[rt],regname[rs],regname[rt]);
+  output_w32(0x1a800000 | (COND_EQ << 12) | rm_rn_rd(rt, rs, rt));
 }
 
 static void emit_cmovne_reg(u_int rs,u_int rt)
@@ -836,12 +813,24 @@ static void emit_cmovl_reg(u_int rs,u_int rt)
   output_w32(0x1a800000 | (COND_LT << 12) | rm_rn_rd(rt, rs, rt));
 }
 
+static void emit_cmovb_reg(u_int rs,u_int rt)
+{
+  assem_debug("csel %s,%s,%s,cc\n",regname[rt],regname[rs],regname[rt]);
+  output_w32(0x1a800000 | (COND_CC << 12) | rm_rn_rd(rt, rs, rt));
+}
+
 static void emit_cmovs_reg(u_int rs,u_int rt)
 {
   assem_debug("csel %s,%s,%s,mi\n",regname[rt],regname[rs],regname[rt]);
   output_w32(0x1a800000 | (COND_MI << 12) | rm_rn_rd(rt, rs, rt));
 }
 
+static void emit_csinvle_reg(u_int rs1,u_int rs2,u_int rt)
+{
+  assem_debug("csinv %s,%s,%s,le\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0x5a800000 | (COND_LE << 12) | rm_rn_rd(rs2, rs1, rt));
+}
+
 static void emit_slti32(u_int rs,int imm,u_int rt)
 {
   if(rs!=rt) emit_zeroreg(rt);
@@ -898,6 +887,12 @@ static void emit_set_if_carry32(u_int rs1, u_int rs2, u_int rt)
   emit_cmovb_imm(1,rt);
 }
 
+static int can_jump_or_call(const void *a)
+{
+  intptr_t diff = (u_char *)a - out;
+  return (-134217728 <= diff && diff <= 134217727);
+}
+
 static void emit_call(const void *a)
 {
   intptr_t diff = (u_char *)a - out;
@@ -906,7 +901,7 @@ static void emit_call(const void *a)
   if (-134217728 <= diff && diff <= 134217727)
     output_w32(0x94000000 | ((diff >> 2) & 0x03ffffff));
   else
-    assert(0);
+    abort();
 }
 
 static void emit_jmp(const void *a)
@@ -972,16 +967,23 @@ static void emit_jc(const void *a)
   output_w32(0x54000000 | (offset << 5) | COND_CS);
 }
 
-static void emit_jcc(const void *a)
+static void emit_cb(u_int isnz, u_int is64, const void *a, u_int r)
 {
-  assem_debug("bcc %p\n", a);
+  assem_debug("cb%sz %s,%p\n", isnz?"n":"", is64?regname64[r]:regname[r], a);
   u_int offset = genjmpcc(a);
-  output_w32(0x54000000 | (offset << 5) | COND_CC);
+  is64 = is64 ? 0x80000000 : 0;
+  isnz = isnz ? 0x01000000 : 0;
+  output_w32(0x34000000 | is64 | isnz | imm19_rt(offset, r));
+}
+
+static void emit_cbz(const void *a, u_int r)
+{
+  emit_cb(0, 0, a, r);
 }
 
 static void emit_jmpreg(u_int r)
 {
-  assem_debug("br %s", regname64[r]);
+  assem_debug("br %s\n", regname64[r]);
   output_w32(0xd61f0000 | rm_rn_rd(0, r, 0));
 }
 
@@ -1000,10 +1002,21 @@ static void emit_adr(void *addr, u_int rt)
 {
   intptr_t offset = (u_char *)addr - out;
   assert(-1048576 <= offset && offset < 1048576);
+  assert(rt < 31);
   assem_debug("adr x%d,#%#lx\n", rt, offset);
   output_w32(0x10000000 | ((offset&0x3) << 29) | (((offset>>2)&0x7ffff) << 5) | rt);
 }
 
+static void emit_adrp(void *addr, u_int rt)
+{
+  intptr_t offset = ((intptr_t)addr & ~0xfffl) - ((intptr_t)out & ~0xfffl);
+  assert(-4294967296l <= offset && offset < 4294967296l);
+  assert(rt < 31);
+  offset >>= 12;
+  assem_debug("adrp %s,#%#lx(000)\n",regname64[rt],offset);
+  output_w32(0x90000000 | ((offset&0x3)<<29) | (((offset>>2)&0x7ffff)<<5) | rt);
+}
+
 static void emit_readword_indexed(int offset, u_int rs, u_int rt)
 {
   assem_debug("ldur %s,[%s+%#x]\n",regname[rt],regname64[rs],offset);
@@ -1095,55 +1108,80 @@ static void emit_movzwl_indexed(int offset, u_int rs, u_int rt)
 
 static void emit_writeword_indexed(u_int rt, int offset, u_int rs)
 {
-  assem_debug("str %s,[%s+%#x]\n", regname[rt], regname[rs], offset);
-  if (!(offset & 3) && offset <= 16380)
+  if (!(offset & 3) && (u_int)offset <= 16380) {
+    assem_debug("str %s,[%s+%#x]\n", regname[rt], regname[rs], offset);
     output_w32(0xb9000000 | imm12_rn_rd(offset >> 2, rs, rt));
+  }
+  else if (-256 <= offset && offset < 256) {
+    assem_debug("stur %s,[%s+%#x]\n", regname[rt], regname[rs], offset);
+    output_w32(0xb8000000 | imm9_rn_rt(offset & 0x1ff, rs, rt));
+  }
   else
     assert(0);
 }
 
 static void emit_writehword_indexed(u_int rt, int offset, u_int rs)
 {
-  assem_debug("strh %s,[%s+%#x]\n", regname[rt], regname[rs], offset);
-  if (!(offset & 1) && offset <= 8190)
+  if (!(offset & 1) && (u_int)offset <= 8190) {
+    assem_debug("strh %s,[%s+%#x]\n", regname[rt], regname64[rs], offset);
     output_w32(0x79000000 | imm12_rn_rd(offset >> 1, rs, rt));
+  }
+  else if (-256 <= offset && offset < 256) {
+    assem_debug("sturh %s,[%s+%#x]\n", regname[rt], regname64[rs], offset);
+    output_w32(0x78000000 | imm9_rn_rt(offset & 0x1ff, rs, rt));
+  }
   else
     assert(0);
 }
 
 static void emit_writebyte_indexed(u_int rt, int offset, u_int rs)
 {
-  assem_debug("strb %s,[%s+%#x]\n", regname[rt], regname[rs], offset);
-  if ((u_int)offset < 4096)
+  if ((u_int)offset < 4096) {
+    assem_debug("strb %s,[%s+%#x]\n", regname[rt], regname64[rs], offset);
     output_w32(0x39000000 | imm12_rn_rd(offset, rs, rt));
+  }
+  else if (-256 <= offset && offset < 256) {
+    assem_debug("sturb %s,[%s+%#x]\n", regname[rt], regname64[rs], offset);
+    output_w32(0x38000000 | imm9_rn_rt(offset & 0x1ff, rs, rt));
+  }
   else
     assert(0);
 }
 
-static void emit_umull(u_int rs1, u_int rs2, u_int hi, u_int lo)
+static void emit_umull(u_int rs1, u_int rs2, u_int rt)
 {
-  assem_debug("umull %s, %s, %s, %s\n",regname[lo],regname[hi],regname[rs1],regname[rs2]);
-  assert(rs1<16);
-  assert(rs2<16);
-  assert(hi<16);
-  assert(lo<16);
-  assert(0);
+  assem_debug("umull %s,%s,%s\n",regname64[rt],regname[rs1],regname[rs2]);
+  output_w32(0x9ba00000 | rm_ra_rn_rd(rs2, WZR, rs1, rt));
 }
 
-static void emit_smull(u_int rs1, u_int rs2, u_int hi, u_int lo)
+static void emit_smull(u_int rs1, u_int rs2, u_int rt)
 {
-  assem_debug("smull %s, %s, %s, %s\n",regname[lo],regname[hi],regname[rs1],regname[rs2]);
-  assert(rs1<16);
-  assert(rs2<16);
-  assert(hi<16);
-  assert(lo<16);
-  assert(0);
+  assem_debug("smull %s,%s,%s\n",regname64[rt],regname[rs1],regname[rs2]);
+  output_w32(0x9b200000 | rm_ra_rn_rd(rs2, WZR, rs1, rt));
 }
 
-static void emit_clz(u_int rs,u_int rt)
+static void emit_msub(u_int rs1, u_int rs2, u_int rs3, u_int rt)
+{
+  assem_debug("msub %s,%s,%s,%s\n",regname[rt],regname[rs1],regname[rs2],regname[rs3]);
+  output_w32(0x1b008000 | rm_ra_rn_rd(rs2, rs3, rs1, rt));
+}
+
+static void emit_sdiv(u_int rs1, u_int rs2, u_int rt)
+{
+  assem_debug("sdiv %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0x1ac00c00 | rm_rn_rd(rs2, rs1, rt));
+}
+
+static void emit_udiv(u_int rs1, u_int rs2, u_int rt)
+{
+  assem_debug("udiv %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0x1ac00800 | rm_rn_rd(rs2, rs1, rt));
+}
+
+static void emit_clz(u_int rs, u_int rt)
 {
   assem_debug("clz %s,%s\n",regname[rt],regname[rs]);
-  assert(0);
+  output_w32(0x5ac01000 | rn_rd(rs, rt));
 }
 
 // special case for checking invalid_code
@@ -1151,28 +1189,23 @@ static void emit_cmpmem_indexedsr12_reg(u_int rbase, u_int r, u_int imm)
 {
   host_tempreg_acquire();
   emit_shrimm(r, 12, HOST_TEMPREG);
-  assem_debug("ldrb %s,[%s,%s]",regname[HOST_TEMPREG],regname64[rbase],regname64[HOST_TEMPREG]);
-  output_w32(0x38606800 | rm_rn_rd(HOST_TEMPREG, rbase, HOST_TEMPREG));
+  assem_debug("ldrb %s,[%s,%s,uxtw]\n",regname[HOST_TEMPREG],regname64[rbase],regname[HOST_TEMPREG]);
+  output_w32(0x38604800 | rm_rn_rd(HOST_TEMPREG, rbase, HOST_TEMPREG));
   emit_cmpimm(HOST_TEMPREG, imm);
   host_tempreg_release();
 }
 
-static void emit_orrne_imm(u_int rs,int imm,u_int rt)
-{
-  assem_debug("orrne %s,%s,#%#x\n",regname[rt],regname[rs],imm);
-  assert(0);
-}
-
-static void emit_andne_imm(u_int rs,int imm,u_int rt)
+// special for loadlr_assemble, rs2 is destroyed
+static void emit_bic_lsl(u_int rs1,u_int rs2,u_int shift,u_int rt)
 {
-  assem_debug("andne %s,%s,#%#x\n",regname[rt],regname[rs],imm);
-  assert(0);
+  emit_shl(rs2, shift, rs2);
+  emit_bic(rs1, rs2, rt);
 }
 
-static unused void emit_addpl_imm(u_int rs,int imm,u_int rt)
+static void emit_bic_lsr(u_int rs1,u_int rs2,u_int shift,u_int rt)
 {
-  assem_debug("addpl %s,%s,#%#x\n",regname[rt],regname[rs],imm);
-  assert(0);
+  emit_shr(rs2, shift, rs2);
+  emit_bic(rs1, rs2, rt);
 }
 
 static void emit_loadlp_ofs(u_int ofs, u_int rt)
@@ -1266,7 +1299,7 @@ static void emit_extjump2(u_char *addr, u_int target, void *linker)
   // addr is in the current recompiled block (max 256k)
   // offset shouldn't exceed +/-1MB
   emit_adr(addr, 1);
-  emit_jmp(linker);
+  emit_far_jump(linker);
 }
 
 static void check_extjump2(void *src)
@@ -1280,9 +1313,11 @@ static void check_extjump2(void *src)
 static void emit_movimm_from(u_int rs_val, u_int rs, u_int rt_val, u_int rt)
 {
   int diff = rt_val - rs_val;
-  if ((-4096 <= diff && diff < 4096)
-      || (-16777216 <= diff && diff < 16777216 && !(diff & 0xfff)))
+  if ((-4096 < diff && diff < 4096)
+      || (-16777216 < diff && diff < 16777216 && !(diff & 0xfff)))
     emit_addimm(rs, diff, rt);
+  else if (rt_val == ~rs_val)
+    emit_not(rs, rt);
   else if (is_rotated_mask(rs_val ^ rt_val))
     emit_xorimm(rs, rs_val ^ rt_val, rt);
   else
@@ -1293,8 +1328,9 @@ static void emit_movimm_from(u_int rs_val, u_int rs, u_int rt_val, u_int rt)
 static int is_similar_value(u_int v1, u_int v2)
 {
   int diff = v1 - v2;
-  return (-4096 <= diff && diff < 4096)
-    || (-16777216 <= diff && diff < 16777216 && !(diff & 0xfff))
+  return (-4096 < diff && diff < 4096)
+    || (-16777216 < diff && diff < 16777216 && !(diff & 0xfff))
+    || v1 == ~v2
     || is_rotated_mask(v1 ^ v2);
 }
 
@@ -1326,7 +1362,7 @@ static void loadstore_extend(enum stub_type type, u_int rs, u_int rt)
     case STOREH_STUB: emit_ubfm(rs, 15, rt); break;
     case LOADW_STUB:  
     case STOREW_STUB: if (rs != rt) emit_mov(rs, rt); break;
-    default: assert(0);
+    default:          assert(0);
   }
 }
 
@@ -1381,7 +1417,7 @@ static void do_readstub(int n)
       case LOADH_STUB:  emit_ldrsh_dualindexed(temp2,rs,rt); break;
       case LOADHU_STUB: emit_ldrh_dualindexed(temp2,rs,rt); break;
       case LOADW_STUB:  emit_ldr_dualindexed(temp2,rs,rt); break;
-      default: assert(0);
+      default:          assert(0);
     }
   }
   if(regs_saved) {
@@ -1406,8 +1442,8 @@ static void do_readstub(int n)
   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_call(handler);
+  emit_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d),2);
+  emit_far_call(handler);
   // (no cycle reload after read)
   if(itype[i]==C1LS||itype[i]==C2LS||(rt>=0&&rt1[i]!=0)) {
     loadstore_extend(type,0,rt);
@@ -1418,7 +1454,8 @@ static void do_readstub(int n)
   emit_jmp(stubs[n].retaddr);
 }
 
-static void inline_readstub(enum stub_type type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
+static void inline_readstub(enum stub_type type, int i, u_int addr,
+  const signed char regmap[], int target, int adj, u_int reglist)
 {
   int rs=get_reg(regmap,target);
   int rt=get_reg(regmap,target);
@@ -1428,7 +1465,7 @@ static void inline_readstub(enum stub_type type, int i, u_int addr, signed char
   uintptr_t host_addr = 0;
   void *handler;
   int cc=get_reg(regmap,CCREG);
-  //if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj+1),cc,target?rs:-1,rt))
+  //if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj),cc,target?rs:-1,rt))
   //  return;
   handler = get_direct_memhandler(mem_rtab, addr, type, &host_addr);
   if (handler == NULL) {
@@ -1469,13 +1506,16 @@ static void inline_readstub(enum stub_type type, int i, u_int addr, signed char
     emit_mov(rs,0);
   if(cc<0)
     emit_loadreg(CCREG,2);
-  emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj+1),2);
-  if(is_dynamic)
-    emit_readdword(&mem_rtab,1);
+  emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj),2);
+  if(is_dynamic) {
+    uintptr_t l1 = ((uintptr_t *)mem_rtab)[addr>>12] << 1;
+    emit_adrp((void *)l1, 1);
+    emit_addimm64(1, l1 & 0xfff, 1);
+  }
   else
-    emit_call(do_memhandler_pre);
+    emit_far_call(do_memhandler_pre);
 
-  emit_call(handler);
+  emit_far_call(handler);
 
   // (no cycle reload after read)
   if(rt>=0&&rt1[i]!=0)
@@ -1551,7 +1591,7 @@ static void do_writestub(int n)
     case STOREB_STUB: handler=jump_handler_write8; break;
     case STOREH_STUB: handler=jump_handler_write16; break;
     case STOREW_STUB: handler=jump_handler_write32; break;
-    default: assert(0);
+    default:          assert(0);
   }
   assert(handler);
   pass_args(rs,rt);
@@ -1562,10 +1602,10 @@ static void do_writestub(int n)
   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_addimm(cc<0?2:cc,CLOCK_ADJUST((int)stubs[n].d),2);
   // returns new cycle_count
-  emit_call(handler);
-  emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d+1),cc<0?2:cc);
+  emit_far_call(handler);
+  emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d),cc<0?2:cc);
   if(cc<0)
     emit_storereg(CCREG,2);
   if(restore_jump)
@@ -1574,7 +1614,8 @@ static void do_writestub(int n)
   emit_jmp(stubs[n].retaddr);
 }
 
-static void inline_writestub(enum stub_type type, int i, u_int addr, signed char regmap[], int target, int adj, u_int reglist)
+static void inline_writestub(enum stub_type type, int i, u_int addr,
+  const signed char regmap[], int target, int adj, u_int reglist)
 {
   int rs = get_reg(regmap,-1);
   int rt = get_reg(regmap,target);
@@ -1605,21 +1646,47 @@ static void inline_writestub(enum stub_type type, int i, u_int addr, signed char
   cc = cc_use = get_reg(regmap, CCREG);
   if (cc < 0)
     emit_loadreg(CCREG, (cc_use = 2));
-  emit_addimm(cc_use, CLOCK_ADJUST(adj+1), 2);
+  emit_addimm(cc_use, CLOCK_ADJUST(adj), 2);
 
-  emit_call(do_memhandler_pre);
-  emit_call(handler);
-  emit_call(do_memhandler_post);
-  emit_addimm(0, -CLOCK_ADJUST(adj+1), cc_use);
+  emit_far_call(do_memhandler_pre);
+  emit_far_call(handler);
+  emit_far_call(do_memhandler_post);
+  emit_addimm(0, -CLOCK_ADJUST(adj), cc_use);
   if (cc < 0)
     emit_storereg(CCREG, cc_use);
   restore_regs(reglist);
 }
 
-static void do_unalignedwritestub(int n)
+static int verify_code_arm64(const void *source, const void *copy, u_int size)
 {
-  assem_debug("do_unalignedwritestub %x\n",start+stubs[n].a*4);
-  assert(0);
+  int ret = memcmp(source, copy, size);
+  //printf("%s %p,%#x = %d\n", __func__, source, size, ret);
+  return ret;
+}
+
+// this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr
+static void do_dirty_stub_base(u_int vaddr)
+{
+  assert(slen <= MAXBLOCK);
+  emit_loadlp_ofs(0, 0); // ldr x1, source
+  emit_loadlp_ofs(0, 1); // ldr x2, copy
+  emit_movz(slen*4, 2);
+  emit_far_call(verify_code_arm64);
+  void *jmp = out;
+  emit_cbz(0, 0);
+  emit_movz(vaddr & 0xffff, 0);
+  emit_movk_lsl16(vaddr >> 16, 0);
+  emit_far_call(get_addr);
+  emit_jmpreg(0);
+  set_jump_target(jmp, out);
+}
+
+static void assert_dirty_stub(const u_int *ptr)
+{
+  assert((ptr[0] & 0xff00001f) == 0x58000000); // ldr x0, source
+  assert((ptr[1] & 0xff00001f) == 0x58000001); // ldr x1, copy
+  assert((ptr[2] & 0xffe0001f) == 0x52800002); // movz w2, #slen*4
+  assert( ptr[8]               == 0xd61f0000); // br x0
 }
 
 static void set_loadlp(u_int *loadl, void *lit)
@@ -1631,17 +1698,6 @@ static void set_loadlp(u_int *loadl, void *lit)
   *loadl |= (ofs >> 2) << 5;
 }
 
-// this output is parsed by verify_dirty, get_bounds, isclean, get_clean_addr
-static void do_dirty_stub_emit_args(u_int arg0)
-{
-  assert(slen <= MAXBLOCK);
-  emit_loadlp_ofs(0, 1); // ldr x1, source
-  emit_loadlp_ofs(0, 2); // ldr x2, copy
-  emit_movz(slen*4, 3);
-  emit_movz(arg0 & 0xffff, 0);
-  emit_movk_lsl16(arg0 >> 16, 0);
-}
-
 static void do_dirty_stub_emit_literals(u_int *loadlps)
 {
   set_loadlp(&loadlps[0], out);
@@ -1654,8 +1710,7 @@ static void *do_dirty_stub(int i)
 {
   assem_debug("do_dirty_stub %x\n",start+i*4);
   u_int *loadlps = (void *)out;
-  do_dirty_stub_emit_args(start + i*4);
-  emit_call(verify_code);
+  do_dirty_stub_base(start + i*4);
   void *entry = out;
   load_regs_entry(i);
   if (entry == out)
@@ -1665,38 +1720,259 @@ static void *do_dirty_stub(int i)
   return entry;
 }
 
-static void do_dirty_stub_ds()
+static void do_dirty_stub_ds(void)
 {
-  do_dirty_stub_emit_args(start + 1);
   u_int *loadlps = (void *)out;
-  emit_call(verify_code_ds);
+  do_dirty_stub_base(start + 1);
+  void *lit_jumpover = out;
   emit_jmp(out + 8*2);
   do_dirty_stub_emit_literals(loadlps);
+  set_jump_target(lit_jumpover, out);
+}
+
+static uint64_t get_from_ldr_literal(const u_int *i)
+{
+  signed int ofs;
+  assert((i[0] & 0xff000000) == 0x58000000);
+  ofs = i[0] << 8;
+  ofs >>= 5+8;
+  return *(uint64_t *)(i + ofs);
+}
+
+static uint64_t get_from_movz(const u_int *i)
+{
+  assert((i[0] & 0x7fe00000) == 0x52800000);
+  return (i[0] >> 5) & 0xffff;
+}
+
+// Find the "clean" entry point from a "dirty" entry point
+// by skipping past the call to verify_code
+static void *get_clean_addr(u_int *addr)
+{
+  assert_dirty_stub(addr);
+  return addr + 9;
+}
+
+static int verify_dirty(const u_int *ptr)
+{
+  const void *source, *copy;
+  u_int len;
+  assert_dirty_stub(ptr);
+  source = (void *)get_from_ldr_literal(&ptr[0]); // ldr x1, source
+  copy   = (void *)get_from_ldr_literal(&ptr[1]); // ldr x1, copy
+  len = get_from_movz(&ptr[2]);                   // movz w3, #slen*4
+  return !memcmp(source, copy, len);
+}
+
+static int isclean(void *addr)
+{
+  const u_int *ptr = addr;
+  if ((*ptr >> 24) == 0x58) { // the only place ldr (literal) is used
+    assert_dirty_stub(ptr);
+    return 0;
+  }
+  return 1;
+}
+
+// get source that block at addr was compiled from (host pointers)
+static void get_bounds(void *addr, u_char **start, u_char **end)
+{
+  const u_int *ptr = addr;
+  assert_dirty_stub(ptr);
+  *start = (u_char *)get_from_ldr_literal(&ptr[0]); // ldr x1, source
+  *end = *start + get_from_movz(&ptr[2]);           // movz w3, #slen*4
 }
 
 /* Special assem */
 
-#define shift_assemble shift_assemble_arm64
+static void c2op_prologue(u_int op, int i, const struct regstat *i_regs, u_int reglist)
+{
+  save_load_regs_all(1, reglist);
+  cop2_call_stall_check(op, i, i_regs, 0);
+#ifdef PCNT
+  emit_movimm(op, 0);
+  emit_far_call(pcnt_gte_start);
+#endif
+  // pointer to cop2 regs
+  emit_addimm64(FP, (u_char *)&psxRegs.CP2D.r[0] - (u_char *)&dynarec_local, 0);
+}
 
-static void shift_assemble_arm64(int i,struct regstat *i_regs)
+static void c2op_epilogue(u_int op,u_int reglist)
 {
-  assert(0);
+#ifdef PCNT
+  emit_movimm(op, 0);
+  emit_far_call(pcnt_gte_end);
+#endif
+  save_load_regs_all(0, reglist);
 }
-#define loadlr_assemble loadlr_assemble_arm64
 
-static void loadlr_assemble_arm64(int i,struct regstat *i_regs)
+static void c2op_assemble(int i, const struct regstat *i_regs)
 {
-  assert(0);
+  u_int c2op=source[i]&0x3f;
+  u_int hr,reglist_full=0,reglist;
+  int need_flags,need_ir;
+  for(hr=0;hr<HOST_REGS;hr++) {
+    if(i_regs->regmap[hr]>=0) reglist_full|=1<<hr;
+  }
+  reglist=reglist_full&CALLER_SAVE_REGS;
+
+  if (gte_handlers[c2op]!=NULL) {
+    need_flags=!(gte_unneeded[i+1]>>63); // +1 because of how liveness detection works
+    need_ir=(gte_unneeded[i+1]&0xe00)!=0xe00;
+    assem_debug("gte op %08x, unneeded %016lx, need_flags %d, need_ir %d\n",
+      source[i],gte_unneeded[i+1],need_flags,need_ir);
+    if(HACK_ENABLED(NDHACK_GTE_NO_FLAGS))
+      need_flags=0;
+    //int shift = (source[i] >> 19) & 1;
+    //int lm = (source[i] >> 10) & 1;
+    switch(c2op) {
+      default:
+        (void)need_ir;
+        c2op_prologue(c2op, i, i_regs, reglist);
+        emit_movimm(source[i],1); // opcode
+        emit_writeword(1,&psxRegs.code);
+        emit_far_call(need_flags?gte_handlers[c2op]:gte_handlers_nf[c2op]);
+        break;
+    }
+    c2op_epilogue(c2op,reglist);
+  }
 }
 
-static void c2op_assemble(int i,struct regstat *i_regs)
+static void c2op_ctc2_31_assemble(signed char sl, signed char temp)
 {
-  assert(0);
+  //value = value & 0x7ffff000;
+  //if (value & 0x7f87e000) value |= 0x80000000;
+  emit_andimm(sl, 0x7fffe000, temp);
+  emit_testimm(temp, 0xff87ffff);
+  emit_andimm(sl, 0x7ffff000, temp);
+  host_tempreg_acquire();
+  emit_orimm(temp, 0x80000000, HOST_TEMPREG);
+  emit_cmovne_reg(HOST_TEMPREG, temp);
+  host_tempreg_release();
+  assert(0); // testing needed
+}
+
+static void do_mfc2_31_one(u_int copr,signed char temp)
+{
+  emit_readshword(&reg_cop2d[copr],temp);
+  emit_bicsar_imm(temp,31,temp);
+  emit_cmpimm(temp,0xf80);
+  emit_csinvle_reg(temp,WZR,temp); // if (temp > 0xf80) temp = ~0;
+  emit_andimm(temp,0xf80,temp);
+}
+
+static void c2op_mfc2_29_assemble(signed char tl, signed char temp)
+{
+  if (temp < 0) {
+    host_tempreg_acquire();
+    temp = HOST_TEMPREG;
+  }
+  do_mfc2_31_one(9,temp);
+  emit_shrimm(temp,7,tl);
+  do_mfc2_31_one(10,temp);
+  emit_orrshr_imm(temp,2,tl);
+  do_mfc2_31_one(11,temp);
+  emit_orrshl_imm(temp,3,tl);
+  emit_writeword(tl,&reg_cop2d[29]);
+
+  if (temp == HOST_TEMPREG)
+    host_tempreg_release();
 }
 
 static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
 {
-  assert(0);
+  //  case 0x18: MULT
+  //  case 0x19: MULTU
+  //  case 0x1A: DIV
+  //  case 0x1B: DIVU
+  if(rs1[i]&&rs2[i])
+  {
+    switch(opcode2[i])
+    {
+    case 0x18: // MULT
+    case 0x19: // MULTU
+      {
+        signed char m1=get_reg(i_regs->regmap,rs1[i]);
+        signed char m2=get_reg(i_regs->regmap,rs2[i]);
+        signed char hi=get_reg(i_regs->regmap,HIREG);
+        signed char lo=get_reg(i_regs->regmap,LOREG);
+        assert(m1>=0);
+        assert(m2>=0);
+        assert(hi>=0);
+        assert(lo>=0);
+
+        if(opcode2[i]==0x18) // MULT
+          emit_smull(m1,m2,hi);
+        else                 // MULTU
+          emit_umull(m1,m2,hi);
+
+        emit_mov(hi,lo);
+        emit_shrimm64(hi,32,hi);
+        break;
+      }
+    case 0x1A: // DIV
+    case 0x1B: // DIVU
+      {
+        signed char numerator=get_reg(i_regs->regmap,rs1[i]);
+        signed char denominator=get_reg(i_regs->regmap,rs2[i]);
+        signed char quotient=get_reg(i_regs->regmap,LOREG);
+        signed char remainder=get_reg(i_regs->regmap,HIREG);
+        assert(numerator>=0);
+        assert(denominator>=0);
+        assert(quotient>=0);
+        assert(remainder>=0);
+
+        if (opcode2[i] == 0x1A) // DIV
+          emit_sdiv(numerator,denominator,quotient);
+        else                    // DIVU
+          emit_udiv(numerator,denominator,quotient);
+        emit_msub(quotient,denominator,numerator,remainder);
+
+        // div 0 quotient (remainder is already correct)
+        host_tempreg_acquire();
+        if (opcode2[i] == 0x1A) // DIV
+          emit_sub_asrimm(0,numerator,31,HOST_TEMPREG);
+        else
+          emit_movimm(~0,HOST_TEMPREG);
+        emit_test(denominator,denominator);
+        emit_cmoveq_reg(HOST_TEMPREG,quotient);
+        host_tempreg_release();
+        break;
+      }
+    default:
+      assert(0);
+    }
+  }
+  else
+  {
+    signed char hr=get_reg(i_regs->regmap,HIREG);
+    signed char lr=get_reg(i_regs->regmap,LOREG);
+    if ((opcode2[i]==0x1A || opcode2[i]==0x1B) && rs2[i]==0) // div 0
+    {
+      if (rs1[i]) {
+        signed char numerator = get_reg(i_regs->regmap, rs1[i]);
+        assert(numerator >= 0);
+        if (hr >= 0)
+          emit_mov(numerator,hr);
+        if (lr >= 0) {
+          if (opcode2[i] == 0x1A) // DIV
+            emit_sub_asrimm(0,numerator,31,lr);
+          else
+            emit_movimm(~0,lr);
+        }
+      }
+      else {
+        if (hr >= 0) emit_zeroreg(hr);
+        if (lr >= 0) emit_movimm(~0,lr);
+      }
+    }
+    else
+    {
+      // Multiply by zero is zero.
+      if (hr >= 0) emit_zeroreg(hr);
+      if (lr >= 0) emit_zeroreg(lr);
+    }
+  }
 }
 #define multdiv_assemble multdiv_assemble_arm64
 
@@ -1704,7 +1980,7 @@ static void do_jump_vaddr(u_int rs)
 {
   if (rs != 0)
     emit_mov(rs, 0);
-  emit_call(get_addr_ht);
+  emit_far_call(get_addr_ht);
   emit_jmpreg(0);
 }
 
@@ -1748,51 +2024,60 @@ static void do_miniht_insert(u_int return_address,u_int rt,int temp) {
   emit_writeword(rt,&mini_ht[(return_address&0xFF)>>3][0]);
 }
 
-static void mark_clear_cache(void *target)
+static void clear_cache_arm64(char *start, char *end)
 {
-  u_long offset = (u_char *)target - translation_cache;
-  u_int mask = 1u << ((offset >> 12) & 31);
-  if (!(needs_clear_cache[offset >> 17] & mask)) {
-    char *start = (char *)((u_long)target & ~4095ul);
-    start_tcache_write(start, start + 4096);
-    needs_clear_cache[offset >> 17] |= mask;
+  // Don't rely on GCC's __clear_cache implementation, as it caches
+  // icache/dcache cache line sizes, that can vary between cores on
+  // big.LITTLE architectures.
+  uint64_t addr, ctr_el0;
+  static size_t icache_line_size = 0xffff, dcache_line_size = 0xffff;
+  size_t isize, dsize;
+
+  __asm__ volatile("mrs %0, ctr_el0" : "=r"(ctr_el0));
+  isize = 4 << ((ctr_el0 >> 0) & 0xf);
+  dsize = 4 << ((ctr_el0 >> 16) & 0xf);
+
+  // use the global minimum cache line size
+  icache_line_size = isize = icache_line_size < isize ? icache_line_size : isize;
+  dcache_line_size = dsize = dcache_line_size < dsize ? dcache_line_size : dsize;
+
+  /* If CTR_EL0.IDC is enabled, Data cache clean to the Point of Unification is
+     not required for instruction to data coherence.  */
+  if ((ctr_el0 & (1 << 28)) == 0x0) {
+    addr = (uint64_t)start & ~(uint64_t)(dsize - 1);
+    for (; addr < (uint64_t)end; addr += dsize)
+      // use "civac" instead of "cvau", as this is the suggested workaround for
+      // Cortex-A53 errata 819472, 826319, 827319 and 824069.
+      __asm__ volatile("dc civac, %0" : : "r"(addr) : "memory");
   }
-}
+  __asm__ volatile("dsb ish" : : : "memory");
 
-// 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()
-{
-  int i,j;
-  for (i=0;i<(1<<(TARGET_SIZE_2-17));i++)
-  {
-    u_int bitmap=needs_clear_cache[i];
-    if(bitmap) {
-      u_char *start, *end;
-      for(j=0;j<32;j++)
-      {
-        if(bitmap&(1<<j)) {
-          start=translation_cache+i*131072+j*4096;
-          end=start+4095;
-          j++;
-          while(j<32) {
-            if(bitmap&(1<<j)) {
-              end+=4096;
-              j++;
-            }else{
-              end_tcache_write(start, end);
-              break;
-            }
-          }
-        }
-      }
-      needs_clear_cache[i]=0;
-    }
+  /* If CTR_EL0.DIC is enabled, Instruction cache cleaning to the Point of
+     Unification is not required for instruction to data coherence.  */
+  if ((ctr_el0 & (1 << 29)) == 0x0) {
+    addr = (uint64_t)start & ~(uint64_t)(isize - 1);
+    for (; addr < (uint64_t)end; addr += isize)
+      __asm__ volatile("ic ivau, %0" : : "r"(addr) : "memory");
+
+    __asm__ volatile("dsb ish" : : : "memory");
   }
+
+  __asm__ volatile("isb" : : : "memory");
 }
 
 // CPU-architecture-specific initialization
-static void arch_init() {
+static void arch_init(void)
+{
+  uintptr_t diff = (u_char *)&ndrc->tramp.f - (u_char *)&ndrc->tramp.ops;
+  struct tramp_insns *ops = ndrc->tramp.ops;
+  size_t i;
+  assert(!(diff & 3));
+  start_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops));
+  for (i = 0; i < ARRAY_SIZE(ndrc->tramp.ops); i++) {
+    ops[i].ldr = 0x58000000 | imm19_rt(diff >> 2, 17); // ldr x17, [=val]
+    ops[i].br  = 0xd61f0000 | rm_rn_rd(0, 17, 0);      // br x17
+  }
+  end_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops));
 }
 
 // vim:shiftwidth=2:expandtab