X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=libpcsxcore%2Fnew_dynarec%2Fassem_arm64.c;h=0a29eaf4d2d323302ccfd3a00f202e86066994d6;hb=de6dbc5289bc76996ad4e3133e0eeb90e13b2ed4;hp=63c6866b3dacd93a6726273fc4130a3754ba809b;hpb=39b71d9abccb93457f17df887e50507a91390a78;p=pcsx_rearmed.git

diff --git a/libpcsxcore/new_dynarec/assem_arm64.c b/libpcsxcore/new_dynarec/assem_arm64.c
index 63c6866b..0a29eaf4 100644
--- a/libpcsxcore/new_dynarec/assem_arm64.c
+++ b/libpcsxcore/new_dynarec/assem_arm64.c
@@ -23,17 +23,10 @@
 #include "pcnt.h"
 #include "arm_features.h"
 
-#define CALLER_SAVE_REGS 0x0007ffff
-
-#define unused __attribute__((unused))
-
-void do_memhandler_pre();
-void do_memhandler_post();
-
 /* Linker */
 static void set_jump_target(void *addr, void *target)
 {
-  u_int *ptr = addr;
+  u_int *ptr = NDRC_WRITE_OFFSET(addr);
   intptr_t offset = (u_char *)target - (u_char *)addr;
 
   if ((*ptr&0xFC000000) == 0x14000000) { // b
@@ -47,7 +40,7 @@ static void set_jump_target(void *addr, void *target)
     // should only happen when jumping to an already compiled block (see add_jump_out)
     // a workaround would be to do a trampoline jump via a stub at the end of the block
     assert(-1048576 <= offset && offset < 1048576);
-    *ptr=(*ptr&0xFF00000F)|(((offset>>2)&0x7ffff)<<5);
+    *ptr=(*ptr&0xFF00001F)|(((offset>>2)&0x7ffff)<<5);
   }
   else if((*ptr&0x9f000000)==0x10000000) { // adr
     // generated by do_miniht_insert
@@ -68,6 +61,7 @@ static void *find_extjump_insn(void *stub)
   return ptr + offset / 4;
 }
 
+#if 0
 // find where external branch is liked to using addr of it's stub:
 // get address that the stub loads (dyna_linker arg1),
 // treat it as a pointer to branch insn,
@@ -83,6 +77,7 @@ static void *get_pointer(void *stub)
   assert(0);
   return NULL;
 }
+#endif
 
 // Allocate a specific ARM register.
 static void alloc_arm_reg(struct regstat *cur,int i,signed char reg,int hr)
@@ -142,25 +137,10 @@ static unused const char *condname[16] = {
 
 static void output_w32(u_int word)
 {
-  *((u_int *)out) = word;
+  *((u_int *)NDRC_WRITE_OFFSET(out)) = word;
   out += 4;
 }
 
-static void output_w64(uint64_t dword)
-{
-  *((uint64_t *)out) = dword;
-  out+=8;
-}
-
-/*
-static u_int rm_rd(u_int rm, u_int rd)
-{
-  assert(rm < 31);
-  assert(rd < 31);
-  return (rm << 16) | rd;
-}
-*/
-
 static u_int rn_rd(u_int rn, u_int rd)
 {
   assert(rn < 31);
@@ -319,6 +299,12 @@ static void emit_add(u_int rs1, u_int rs2, u_int rt)
   output_w32(0x0b000000 | rm_rn_rd(rs2, rs1, rt));
 }
 
+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_add64(u_int rs1, u_int rs2, u_int rt)
 {
   assem_debug("add %s,%s,%s\n", regname64[rt], regname64[rs1], regname64[rs2]);
@@ -332,19 +318,37 @@ static void emit_adds64(u_int rs1, u_int rs2, u_int rt)
 }
 #define emit_adds_ptr emit_adds64
 
+static void emit_add_lsrimm(u_int rs1, u_int rs2, u_int shift, u_int rt)
+{
+  assem_debug("add %s,%s,%s,lsr #%u\n",regname[rt],regname[rs1],regname[rs2],shift);
+  output_w32(0x0b400000 | rm_imm6_rn_rd(rs2, shift, rs1, rt));
+}
+
 static void emit_neg(u_int rs, u_int rt)
 {
   assem_debug("neg %s,%s\n",regname[rt],regname[rs]);
   output_w32(0x4b000000 | rm_rn_rd(rs, WZR, rt));
 }
 
+static void emit_negs(u_int rs, u_int rt)
+{
+  assem_debug("negs %s,%s\n",regname[rt],regname[rs]);
+  output_w32(0x6b000000 | rm_rn_rd(rs, WZR, rt));
+}
+
 static void emit_sub(u_int rs1, u_int rs2, u_int rt)
 {
   assem_debug("sub %s,%s,%s\n", regname[rt], regname[rs1], regname[rs2]);
   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)
+static void emit_subs(u_int rs1, u_int rs2, u_int rt)
+{
+  assem_debug("subs %s,%s,%s\n", regname[rt], regname[rs1], regname[rs2]);
+  output_w32(0x6b000000 | rm_imm6_rn_rd(rs2, 0, rs1, rt));
+}
+
+static unused 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));
@@ -414,11 +418,32 @@ static void emit_movimm(u_int imm, u_int rt)
   }
 }
 
+static void emit_movimm64(uint64_t imm, u_int rt)
+{
+  u_int shift, op, imm16, insns = 0;
+  for (shift = 0; shift < 4; shift++) {
+    imm16 = (imm >> shift * 16) & 0xffff;
+    if (!imm16)
+      continue;
+    op = insns ? 0xf2800000 : 0xd2800000;
+    assem_debug("mov%c %s,#%#x", insns ? 'k' : 'z', regname64[rt], imm16);
+    if (shift)
+      assem_debug(",lsl #%u", shift * 16);
+    assem_debug("\n");
+    output_w32(op | (shift << 21) | imm16_rd(imm16, rt));
+    insns++;
+  }
+  if (!insns) {
+    assem_debug("movz %s,#0\n", regname64[rt]);
+    output_w32(0xd2800000 | imm16_rd(0, rt));
+  }
+}
+
 static void emit_readword(void *addr, u_int rt)
 {
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
   if (!(offset & 3) && offset <= 16380) {
-    assem_debug("ldr %s,[x%d+%#lx]\n", regname[rt], FP, offset);
+    assem_debug("ldr %s,[x%d+%#lx]%s\n", regname[rt], FP, offset, fpofs_name(offset));
     output_w32(0xb9400000 | imm12_rn_rd(offset >> 2, FP, rt));
   }
   else
@@ -429,7 +454,7 @@ static void emit_readdword(void *addr, u_int rt)
 {
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
   if (!(offset & 7) && offset <= 32760) {
-    assem_debug("ldr %s,[x%d+%#lx]\n", regname64[rt], FP, offset);
+    assem_debug("ldr %s,[x%d+%#lx]%s\n", regname64[rt], FP, offset, fpofs_name(offset));
     output_w32(0xf9400000 | imm12_rn_rd(offset >> 3, FP, rt));
   }
   else
@@ -451,18 +476,20 @@ static void emit_readshword(void *addr, u_int rt)
 static void emit_loadreg(u_int r, u_int hr)
 {
   int is64 = 0;
-  assert(r < 64);
   if (r == 0)
     emit_zeroreg(hr);
   else {
-    void *addr = &psxRegs.GPR.r[r];
+    void *addr;
     switch (r) {
     //case HIREG: addr = &hi; break;
     //case LOREG: addr = &lo; break;
     case CCREG: addr = &cycle_count; break;
-    case CSREG: addr = &Status; break;
     case INVCP: addr = &invc_ptr; is64 = 1; break;
-    default: assert(r < 34); break;
+    case ROREG: addr = &ram_offset; is64 = 1; break;
+    default:
+      assert(r < 34);
+      addr = &psxRegs.GPR.r[r];
+      break;
     }
     if (is64)
       emit_readdword(addr, hr);
@@ -475,7 +502,7 @@ static void emit_writeword(u_int rt, void *addr)
 {
   uintptr_t offset = (u_char *)addr - (u_char *)&dynarec_local;
   if (!(offset & 3) && offset <= 16380) {
-    assem_debug("str %s,[x%d+%#lx]\n", regname[rt], FP, offset);
+    assem_debug("str %s,[x%d+%#lx]%s\n", regname[rt], FP, offset, fpofs_name(offset));
     output_w32(0xb9000000 | imm12_rn_rd(offset >> 2, FP, rt));
   }
   else
@@ -486,7 +513,7 @@ 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);
+    assem_debug("str %s,[x%d+%#lx]%s\n", regname64[rt], FP, offset, fpofs_name(offset));
     output_w32(0xf9000000 | imm12_rn_rd(offset >> 3, FP, rt));
   }
   else
@@ -557,6 +584,12 @@ 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_orn_asrimm(u_int rs1, u_int rs2, u_int shift, u_int rt)
+{
+  assem_debug("orn %s,%s,%s,asr #%u\n",regname[rt],regname[rs1],regname[rs2],shift);
+  output_w32(0x2aa00000 | rm_imm6_rn_rd(rs2, shift, rs1, 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);
@@ -588,28 +621,43 @@ static void emit_addimm_s(u_int s, u_int is64, u_int rs, uintptr_t imm, u_int rt
     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) {
-    assem_debug("add %s,%s,#%#lx\n",regname[rt],regname[rt],imm&0xfff000);
-    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 & 0xfff, rt, rt));
+  else if (imm < 16777216 && (!(imm & 0xfff) || !s)) {
+    assem_debug("add%s %s,%s,#%#lx\n", st, regname[rt], regname[rs], imm&0xfff000);
+    output_w32(0x11400000 | is64 | s | imm12_rn_rd(imm >> 12, rs, rt));
+    if (imm & 0xfff) {
+      assem_debug("add %s,%s,#%#lx\n", regname[rt], regname[rt], imm&0xfff);
+      output_w32(0x11000000 | is64 | imm12_rn_rd(imm & 0xfff, rt, rt));
     }
   }
-  else if (-imm < 16777216) {
-    assem_debug("sub %s,%s,#%#lx\n",regname[rt],regname[rt],-imm&0xfff000);
-    output_w32(0x51400000 | is64 | imm12_rn_rd(-imm >> 12, rs, rt));
-    if ((imm & 0xfff) || s) {
-      assem_debug("sub%s %s,%s,#%#lx\n",st,regname[rt],regname[rs],-imm&0xfff);
-      output_w32(0x51000000 | is64 | s | imm12_rn_rd(-imm & 0xfff, rt, rt));
+  else if (-imm < 16777216 && (!(-imm & 0xfff) || !s)) {
+    assem_debug("sub%s %s,%s,#%#lx\n", st, regname[rt], regname[rs], -imm&0xfff000);
+    output_w32(0x51400000 | is64 | s | imm12_rn_rd(-imm >> 12, rs, rt));
+    if (-imm & 0xfff) {
+      assem_debug("sub %s,%s,#%#lx\n", regname[rt], regname[rt], -imm&0xfff);
+      output_w32(0x51000000 | is64 | imm12_rn_rd(-imm & 0xfff, rt, rt));
     }
   }
-  else
-    abort();
+  else {
+    u_int tmp = rt;
+    assert(!is64);
+    if (rs == rt) {
+      host_tempreg_acquire();
+      tmp = HOST_TEMPREG;
+    }
+    emit_movimm(imm, tmp);
+    assem_debug("add%s %s,%s,%s\n", st, regname[rt], regname[rs], regname[tmp]);
+    output_w32(0x0b000000 | s | rm_rn_rd(rs, tmp, rt));
+    if (tmp == HOST_TEMPREG)
+      host_tempreg_release();
+  }
 }
 
 static void emit_addimm(u_int rs, uintptr_t imm, u_int rt)
 {
+  if (imm == 0) {
+    emit_mov(rs, rt);
+    return;
+  }
   emit_addimm_s(0, 0, rs, imm, rt);
 }
 
@@ -618,14 +666,19 @@ static void emit_addimm64(u_int rs, uintptr_t imm, u_int rt)
   emit_addimm_s(0, 1, rs, imm, rt);
 }
 
+static void emit_addimm_ptr(u_int rs, uintptr_t imm, u_int rt)
+{
+  emit_addimm64(rs, imm, rt);
+}
+
 static void emit_addimm_and_set_flags(int imm, u_int rt)
 {
   emit_addimm_s(1, 0, rt, imm, rt);
 }
 
-static void emit_addimm_no_flags(u_int imm,u_int rt)
+static void emit_addimm_and_set_flags3(u_int rs, int imm, u_int rt)
 {
-  emit_addimm(rt,imm,rt);
+  emit_addimm_s(1, 0, rs, imm, rt);
 }
 
 static void emit_logicop_imm(u_int op, u_int rs, u_int imm, u_int rt)
@@ -833,6 +886,12 @@ static void emit_csinvle_reg(u_int rs1,u_int rs2,u_int rt)
   output_w32(0x5a800000 | (COND_LE << 12) | rm_rn_rd(rs2, rs1, rt));
 }
 
+static void emit_csinvne_reg(u_int rs1,u_int rs2,u_int rt)
+{
+  assem_debug("csinv %s,%s,%s,ne\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0x5a800000 | (COND_NE << 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);
@@ -855,6 +914,12 @@ static void emit_cmp(u_int rs,u_int rt)
   output_w32(0x6b000000 | rm_rn_rd(rt, rs, WZR));
 }
 
+static void emit_cmpcs(u_int rs,u_int rt)
+{
+  assem_debug("ccmp %s,%s,#0,cs\n",regname[rs],regname[rt]);
+  output_w32(0x7a400000 | (COND_CS << 12) | rm_rn_rd(rt, rs, 0));
+}
+
 static void emit_set_gz32(u_int rs, u_int rt)
 {
   //assem_debug("set_gz32\n");
@@ -955,6 +1020,13 @@ static void emit_jge(const void *a)
   output_w32(0x54000000 | (offset << 5) | COND_GE);
 }
 
+static void emit_jo(const void *a)
+{
+  assem_debug("bvs %p\n", a);
+  u_int offset = genjmpcc(a);
+  output_w32(0x54000000 | (offset << 5) | COND_VS);
+}
+
 static void emit_jno(const void *a)
 {
   assem_debug("bvc %p\n", a);
@@ -978,9 +1050,11 @@ static void emit_cb(u_int isnz, u_int is64, const void *a, u_int r)
   output_w32(0x34000000 | is64 | isnz | imm19_rt(offset, r));
 }
 
-static void emit_cbz(const void *a, u_int r)
+static void *emit_cbz(u_int r, const void *a)
 {
+  void *ret = out;
   emit_cb(0, 0, a, r);
+  return ret;
 }
 
 static void emit_jmpreg(u_int r)
@@ -1188,14 +1262,11 @@ static void emit_clz(u_int rs, u_int rt)
 }
 
 // special case for checking invalid_code
-static void emit_cmpmem_indexedsr12_reg(u_int rbase, u_int r, u_int imm)
+static void emit_ldrb_indexedsr12_reg(u_int rbase, u_int r, u_int rt)
 {
-  host_tempreg_acquire();
-  emit_shrimm(r, 12, 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();
+  emit_shrimm(r, 12, rt);
+  assem_debug("ldrb %s,[%s,%s,uxtw]\n",regname[rt],regname64[rbase],regname[rt]);
+  output_w32(0x38604800 | rm_rn_rd(rt, rbase, rt));
 }
 
 // special for loadlr_assemble, rs2 is destroyed
@@ -1211,11 +1282,6 @@ static void emit_bic_lsr(u_int rs1,u_int rs2,u_int shift,u_int rt)
   emit_bic(rs1, rs2, rt);
 }
 
-static void emit_loadlp_ofs(u_int ofs, u_int rt)
-{
-  output_w32(0x58000000 | imm19_rt(ofs, rt));
-}
-
 static void emit_ldst(int is_st, int is64, u_int rt, u_int rn, u_int ofs)
 {
   u_int op = 0xb9000000;
@@ -1292,7 +1358,7 @@ static void literal_pool_jumpover(int n)
 }
 
 // parsed by get_pointer, find_extjump_insn
-static void emit_extjump2(u_char *addr, u_int target, void *linker)
+static void emit_extjump(u_char *addr, u_int target)
 {
   assert(((addr[3]&0xfc)==0x14) || ((addr[3]&0xff)==0x54)); // b or b.cond
 
@@ -1302,7 +1368,7 @@ 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_far_jump(linker);
+  emit_far_jump(dyna_linker);
 }
 
 static void check_extjump2(void *src)
@@ -1337,7 +1403,18 @@ static int is_similar_value(u_int v1, u_int v2)
     || is_rotated_mask(v1 ^ v2);
 }
 
-// trashes r2
+static void emit_movimm_from64(u_int rs_val, u_int rs, uintptr_t rt_val, u_int rt)
+{
+  if (rt_val < 0x100000000ull) {
+    emit_movimm_from(rs_val, rs, rt_val, rt);
+    return;
+  }
+  // just move the whole thing. At least on Linux all addresses
+  // seem to be 48bit, so 3 insns - not great not terrible
+  emit_movimm64(rt_val, rt);
+}
+
+// trashes x2
 static void pass_args64(u_int a0, u_int a1)
 {
   if(a0==1&&a1==0) {
@@ -1383,7 +1460,7 @@ static void do_readstub(int n)
   u_int reglist = stubs[n].e;
   const signed char *i_regmap = i_regs->regmap;
   int rt;
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS||dops[i].itype==LOADLR) {
+  if(dops[i].itype==C2LS||dops[i].itype==LOADLR) {
     rt=get_reg(i_regmap,FTEMP);
   }else{
     rt=get_reg(i_regmap,dops[i].rt1);
@@ -1413,7 +1490,7 @@ static void do_readstub(int n)
   emit_adds64(temp2,temp2,temp2);
   handler_jump=out;
   emit_jc(0);
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
+  if(dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
     switch(type) {
       case LOADB_STUB:  emit_ldrsb_dualindexed(temp2,rs,rt); break;
       case LOADBU_STUB: emit_ldrb_dualindexed(temp2,rs,rt); break;
@@ -1445,10 +1522,10 @@ 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),2);
+  emit_addimm(cc<0?2:cc,(int)stubs[n].d,2);
   emit_far_call(handler);
   // (no cycle reload after read)
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
+  if(dops[i].itype==C2LS||(rt>=0&&dops[i].rt1!=0)) {
     loadstore_extend(type,0,rt);
   }
   if(restore_jump)
@@ -1460,37 +1537,33 @@ static void do_readstub(int n)
 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);
-  if(rs<0) rs=get_reg(regmap,-1);
-  assert(rs>=0);
+  int ra = cinfo[i].addr;
+  int rt = get_reg(regmap, target);
+  assert(ra >= 0);
   u_int is_dynamic=0;
   uintptr_t host_addr = 0;
   void *handler;
   int cc=get_reg(regmap,CCREG);
-  //if(pcsx_direct_read(type,addr,CLOCK_ADJUST(adj),cc,target?rs:-1,rt))
+  //if(pcsx_direct_read(type,addr,adj,cc,target?ra:-1,rt))
   //  return;
   handler = get_direct_memhandler(mem_rtab, addr, type, &host_addr);
   if (handler == NULL) {
     if(rt<0||dops[i].rt1==0)
       return;
-    if (addr != host_addr) {
-      if (host_addr >= 0x100000000ull)
-        abort(); // ROREG not implemented
-      emit_movimm_from(addr, rs, host_addr, rs);
-    }
+    if (addr != host_addr)
+      emit_movimm_from64(addr, ra, host_addr, ra);
     switch(type) {
-      case LOADB_STUB:  emit_movsbl_indexed(0,rs,rt); break;
-      case LOADBU_STUB: emit_movzbl_indexed(0,rs,rt); break;
-      case LOADH_STUB:  emit_movswl_indexed(0,rs,rt); break;
-      case LOADHU_STUB: emit_movzwl_indexed(0,rs,rt); break;
-      case LOADW_STUB:  emit_readword_indexed(0,rs,rt); break;
+      case LOADB_STUB:  emit_movsbl_indexed(0,ra,rt); break;
+      case LOADBU_STUB: emit_movzbl_indexed(0,ra,rt); break;
+      case LOADH_STUB:  emit_movswl_indexed(0,ra,rt); break;
+      case LOADHU_STUB: emit_movzwl_indexed(0,ra,rt); break;
+      case LOADW_STUB:  emit_readword_indexed(0,ra,rt); break;
       default:          assert(0);
     }
     return;
   }
-  is_dynamic=pcsxmem_is_handler_dynamic(addr);
-  if(is_dynamic) {
+  is_dynamic = pcsxmem_is_handler_dynamic(addr);
+  if (is_dynamic) {
     if(type==LOADB_STUB||type==LOADBU_STUB)
       handler=jump_handler_read8;
     if(type==LOADH_STUB||type==LOADHU_STUB)
@@ -1505,15 +1578,20 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
   save_regs(reglist);
   if(target==0)
     emit_movimm(addr,0);
-  else if(rs!=0)
-    emit_mov(rs,0);
+  else if(ra!=0)
+    emit_mov(ra,0);
   if(cc<0)
     emit_loadreg(CCREG,2);
-  emit_addimm(cc<0?2:cc,CLOCK_ADJUST(adj),2);
+  emit_addimm(cc<0?2:cc,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);
+    intptr_t offset = (l1 & ~0xfffl) - ((intptr_t)out & ~0xfffl);
+    if (-4294967296l <= offset && offset < 4294967296l) {
+      emit_adrp((void *)l1, 1);
+      emit_addimm64(1, l1 & 0xfff, 1);
+    }
+    else
+      emit_movimm64(l1, 1);
   }
   else
     emit_far_call(do_memhandler_pre);
@@ -1537,7 +1615,7 @@ static void do_writestub(int n)
   u_int reglist=stubs[n].e;
   signed char *i_regmap=i_regs->regmap;
   int rt,r;
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS) {
+  if(dops[i].itype==C2LS) {
     rt=get_reg(i_regmap,r=FTEMP);
   }else{
     rt=get_reg(i_regmap,r=dops[i].rs2);
@@ -1586,7 +1664,6 @@ static void do_writestub(int n)
     emit_jmp(stubs[n].retaddr); // return address (invcode check)
   set_jump_target(handler_jump, out);
 
-  // TODO FIXME: regalloc should prefer callee-saved regs
   if(!regs_saved)
     save_regs(reglist);
   void *handler=NULL;
@@ -1605,10 +1682,10 @@ 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),2);
+  emit_addimm(cc<0?2:cc,(int)stubs[n].d,2);
   // returns new cycle_count
   emit_far_call(handler);
-  emit_addimm(0,-CLOCK_ADJUST((int)stubs[n].d),cc<0?2:cc);
+  emit_addimm(0,-(int)stubs[n].d,cc<0?2:cc);
   if(cc<0)
     emit_storereg(CCREG,2);
   if(restore_jump)
@@ -1620,22 +1697,19 @@ static void do_writestub(int n)
 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 ra = cinfo[i].addr;
   int rt = get_reg(regmap,target);
-  assert(rs >= 0);
+  assert(ra >= 0);
   assert(rt >= 0);
   uintptr_t host_addr = 0;
   void *handler = get_direct_memhandler(mem_wtab, addr, type, &host_addr);
   if (handler == NULL) {
-    if (addr != host_addr) {
-      if (host_addr >= 0x100000000ull)
-        abort(); // ROREG not implemented
-      emit_movimm_from(addr, rs, host_addr, rs);
-    }
+    if (addr != host_addr)
+      emit_movimm_from64(addr, ra, host_addr, ra);
     switch (type) {
-      case STOREB_STUB: emit_writebyte_indexed(rt, 0, rs); break;
-      case STOREH_STUB: emit_writehword_indexed(rt, 0, rs); break;
-      case STOREW_STUB: emit_writeword_indexed(rt, 0, rs); break;
+      case STOREB_STUB: emit_writebyte_indexed(rt, 0, ra); break;
+      case STOREH_STUB: emit_writehword_indexed(rt, 0, ra); break;
+      case STOREW_STUB: emit_writeword_indexed(rt, 0, ra); break;
       default:          assert(0);
     }
     return;
@@ -1643,149 +1717,23 @@ static void inline_writestub(enum stub_type type, int i, u_int addr,
 
   // call a memhandler
   save_regs(reglist);
-  emit_writeword(rs, &address); // some handlers still need it
+  emit_writeword(ra, &address); // some handlers still need it
   loadstore_extend(type, rt, 0);
   int cc, cc_use;
   cc = cc_use = get_reg(regmap, CCREG);
   if (cc < 0)
     emit_loadreg(CCREG, (cc_use = 2));
-  emit_addimm(cc_use, CLOCK_ADJUST(adj), 2);
+  emit_addimm(cc_use, adj, 2);
 
   emit_far_call(do_memhandler_pre);
   emit_far_call(handler);
   emit_far_call(do_memhandler_post);
-  emit_addimm(0, -CLOCK_ADJUST(adj), cc_use);
+  emit_addimm(0, -adj, cc_use);
   if (cc < 0)
     emit_storereg(CCREG, cc_use);
   restore_regs(reglist);
 }
 
-static int verify_code_arm64(const void *source, const void *copy, u_int size)
-{
-  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, u_int source_len)
-{
-  assert(source_len <= MAXBLOCK*4);
-  emit_loadlp_ofs(0, 0); // ldr x1, source
-  emit_loadlp_ofs(0, 1); // ldr x2, copy
-  emit_movz(source_len, 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, #source_len
-  assert( ptr[8]               == 0xd61f0000); // br x0
-}
-
-static void set_loadlp(u_int *loadl, void *lit)
-{
-  uintptr_t ofs = (u_char *)lit - (u_char *)loadl;
-  assert((*loadl & ~0x1f) == 0x58000000);
-  assert((ofs & 3) == 0);
-  assert(ofs < 0x100000);
-  *loadl |= (ofs >> 2) << 5;
-}
-
-static void do_dirty_stub_emit_literals(u_int *loadlps)
-{
-  set_loadlp(&loadlps[0], out);
-  output_w64((uintptr_t)source);
-  set_loadlp(&loadlps[1], out);
-  output_w64((uintptr_t)copy);
-}
-
-static void *do_dirty_stub(int i, u_int source_len)
-{
-  assem_debug("do_dirty_stub %x\n",start+i*4);
-  u_int *loadlps = (void *)out;
-  do_dirty_stub_base(start + i*4, source_len);
-  void *entry = out;
-  load_regs_entry(i);
-  if (entry == out)
-    entry = instr_addr[i];
-  emit_jmp(instr_addr[i]);
-  do_dirty_stub_emit_literals(loadlps);
-  return entry;
-}
-
-static void do_dirty_stub_ds(u_int source_len)
-{
-  u_int *loadlps = (void *)out;
-  do_dirty_stub_base(start + 1, source_len);
-  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, #source_len
-  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, #source_len
-}
-
 /* Special assem */
 
 static void c2op_prologue(u_int op, int i, const struct regstat *i_regs, u_int reglist)
@@ -1882,7 +1830,7 @@ static void c2op_mfc2_29_assemble(signed char tl, signed char temp)
     host_tempreg_release();
 }
 
-static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
+static void multdiv_assemble_arm64(int i, const struct regstat *i_regs)
 {
   //  case 0x18: MULT
   //  case 0x19: MULTU
@@ -1933,8 +1881,10 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
 
         // div 0 quotient (remainder is already correct)
         host_tempreg_acquire();
-        if (dops[i].opcode2 == 0x1A) // DIV
-          emit_sub_asrimm(0,numerator,31,HOST_TEMPREG);
+        if (dops[i].opcode2 == 0x1A) { // DIV
+          emit_add_lsrimm(WZR,numerator,31,HOST_TEMPREG);
+          emit_orn_asrimm(HOST_TEMPREG,numerator,31,HOST_TEMPREG);
+        }
         else
           emit_movimm(~0,HOST_TEMPREG);
         emit_test(denominator,denominator);
@@ -1958,8 +1908,10 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
         if (hr >= 0)
           emit_mov(numerator,hr);
         if (lr >= 0) {
-          if (dops[i].opcode2 == 0x1A) // DIV
-            emit_sub_asrimm(0,numerator,31,lr);
+          if (dops[i].opcode2 == 0x1A) { // DIV
+            emit_add_lsrimm(WZR,numerator,31,lr);
+            emit_orn_asrimm(lr,numerator,31,lr);
+          }
           else
             emit_movimm(~0,lr);
         }
@@ -1969,6 +1921,17 @@ static void multdiv_assemble_arm64(int i,struct regstat *i_regs)
         if (lr >= 0) emit_movimm(~0,lr);
       }
     }
+    else if ((dops[i].opcode2==0x1A || dops[i].opcode2==0x1B) && dops[i].rs1==0)
+    {
+      signed char denominator = get_reg(i_regs->regmap, dops[i].rs2);
+      assert(denominator >= 0);
+      if (hr >= 0) emit_zeroreg(hr);
+      if (lr >= 0) {
+        emit_zeroreg(lr);
+        emit_test(denominator, denominator);
+        emit_csinvne_reg(lr, lr, lr);
+      }
+    }
     else
     {
       // Multiply by zero is zero.
@@ -1983,7 +1946,7 @@ static void do_jump_vaddr(u_int rs)
 {
   if (rs != 0)
     emit_mov(rs, 0);
-  emit_far_call(get_addr_ht);
+  emit_far_call(ndrc_get_addr_ht);
   emit_jmpreg(0);
 }
 
@@ -2027,7 +1990,7 @@ 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 clear_cache_arm64(char *start, char *end)
+static unused void clear_cache_arm64(char *start, char *end)
 {
   // Don't rely on GCC's __clear_cache implementation, as it caches
   // icache/dcache cache line sizes, that can vary between cores on
@@ -2072,7 +2035,7 @@ static void clear_cache_arm64(char *start, char *end)
 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;
+  struct tramp_insns *ops = NDRC_WRITE_OFFSET(ndrc->tramp.ops);
   size_t i;
   assert(!(diff & 3));
   start_tcache_write(ops, (u_char *)ops + sizeof(ndrc->tramp.ops));