From: notaz <notasas@gmail.com>
Date: Fri, 26 Nov 2021 23:28:08 +0000 (+0200)
Subject: drc: get rid of RAM_FIXED, revive ROREG
X-Git-Tag: r23~56
X-Git-Url: https://notaz.gp2x.de/cgi-bin/gitweb.cgi?p=pcsx_rearmed.git;a=commitdiff_plain;h=37387d8b2b8b9705ca42bd7582ed48d88aeafb9b

drc: get rid of RAM_FIXED, revive ROREG

should work better on devices that can't map memory at desired locations
---

diff --git a/configure b/configure
index 7b300259..1df9aac0 100755
--- a/configure
+++ b/configure
@@ -45,7 +45,6 @@ sound_driver_list="oss alsa pulseaudio sdl libretro"
 sound_drivers=""
 plugins="plugins/spunull/spunull.so \
 plugins/dfxvideo/gpu_peops.so plugins/gpu_unai/gpu_unai.so plugins/gpu_senquack/gpu_senquack.so"
-ram_fixed="no"
 drc_cache_base="no"
 have_armv5=""
 have_armv6=""
@@ -89,21 +88,18 @@ set_platform()
     ;;
   pandora)
     sound_drivers="oss alsa"
-    ram_fixed="yes"
     drc_cache_base="yes"
     optimize_cortexa8="yes"
     have_arm_neon="yes"
     need_xlib="yes"
     ;;
   maemo)
-    ram_fixed="yes"
     drc_cache_base="yes"
     optimize_cortexa8="yes"
     have_arm_neon="yes"
     ;;
   caanoo)
     sound_drivers="oss"
-    ram_fixed="yes"
     drc_cache_base="yes"
     optimize_arm926ej="yes"
     need_warm="yes"
@@ -297,10 +293,6 @@ if [ "$ARCH" != "arm" -o "$have_armv6" = "yes" ]; then
   PLUGIN_CFLAGS="$PLUGIN_CFLAGS -fPIC"
 fi
 
-if [ "$ram_fixed" = "yes" ]; then
-  CFLAGS="$CFLAGS -DRAM_FIXED"
-fi
-
 case "$platform" in
 generic)
   need_sdl="yes"
diff --git a/libpcsxcore/new_dynarec/assem_arm.c b/libpcsxcore/new_dynarec/assem_arm.c
index c7bd5cc2..87490c61 100644
--- a/libpcsxcore/new_dynarec/assem_arm.c
+++ b/libpcsxcore/new_dynarec/assem_arm.c
@@ -543,6 +543,7 @@ static void emit_loadreg(int r, int hr)
     case CCREG: addr = (int)&cycle_count; break;
     case CSREG: addr = (int)&Status; break;
     case INVCP: addr = (int)&invc_ptr; break;
+    case ROREG: addr = (int)&ram_offset; break;
     default: assert(r < 34); break;
     }
     u_int offset = addr-(u_int)&dynarec_local;
@@ -706,11 +707,6 @@ static void emit_addimm_and_set_flags(int imm,int rt)
   }
 }
 
-static void emit_addimm_no_flags(u_int imm,u_int rt)
-{
-  emit_addimm(rt,imm,rt);
-}
-
 static void emit_addnop(u_int r)
 {
   assert(r<16);
@@ -1181,30 +1177,72 @@ static void emit_ldrcc_dualindexed(int rs1, int rs2, int rt)
   output_w32(0x37900000|rd_rn_rm(rt,rs1,rs2));
 }
 
+static void emit_ldrb_dualindexed(int rs1, int rs2, int rt)
+{
+  assem_debug("ldrb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe7d00000|rd_rn_rm(rt,rs1,rs2));
+}
+
 static void emit_ldrccb_dualindexed(int rs1, int rs2, int rt)
 {
   assem_debug("ldrccb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
   output_w32(0x37d00000|rd_rn_rm(rt,rs1,rs2));
 }
 
+static void emit_ldrsb_dualindexed(int rs1, int rs2, int rt)
+{
+  assem_debug("ldrsb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe19000d0|rd_rn_rm(rt,rs1,rs2));
+}
+
 static void emit_ldrccsb_dualindexed(int rs1, int rs2, int rt)
 {
   assem_debug("ldrccsb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
   output_w32(0x319000d0|rd_rn_rm(rt,rs1,rs2));
 }
 
+static void emit_ldrh_dualindexed(int rs1, int rs2, int rt)
+{
+  assem_debug("ldrh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe19000b0|rd_rn_rm(rt,rs1,rs2));
+}
+
 static void emit_ldrcch_dualindexed(int rs1, int rs2, int rt)
 {
   assem_debug("ldrcch %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
   output_w32(0x319000b0|rd_rn_rm(rt,rs1,rs2));
 }
 
+static void emit_ldrsh_dualindexed(int rs1, int rs2, int rt)
+{
+  assem_debug("ldrsh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe19000f0|rd_rn_rm(rt,rs1,rs2));
+}
+
 static void emit_ldrccsh_dualindexed(int rs1, int rs2, int rt)
 {
   assem_debug("ldrccsh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
   output_w32(0x319000f0|rd_rn_rm(rt,rs1,rs2));
 }
 
+static void emit_str_dualindexed(int rs1, int rs2, int rt)
+{
+  assem_debug("str %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe7800000|rd_rn_rm(rt,rs1,rs2));
+}
+
+static void emit_strb_dualindexed(int rs1, int rs2, int rt)
+{
+  assem_debug("strb %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe7c00000|rd_rn_rm(rt,rs1,rs2));
+}
+
+static void emit_strh_dualindexed(int rs1, int rs2, int rt)
+{
+  assem_debug("strh %s,%s,%s\n",regname[rt],regname[rs1],regname[rs2]);
+  output_w32(0xe18000b0|rd_rn_rm(rt,rs1,rs2));
+}
+
 static void emit_movsbl_indexed(int offset, int rs, int rt)
 {
   assert(offset>-256&&offset<256);
diff --git a/libpcsxcore/new_dynarec/assem_arm64.c b/libpcsxcore/new_dynarec/assem_arm64.c
index 63c6866b..431805d4 100644
--- a/libpcsxcore/new_dynarec/assem_arm64.c
+++ b/libpcsxcore/new_dynarec/assem_arm64.c
@@ -462,6 +462,7 @@ static void emit_loadreg(u_int r, u_int hr)
     case CCREG: addr = &cycle_count; break;
     case CSREG: addr = &Status; break;
     case INVCP: addr = &invc_ptr; is64 = 1; break;
+    case ROREG: addr = &ram_offset; is64 = 1; break;
     default: assert(r < 34); break;
     }
     if (is64)
@@ -623,11 +624,6 @@ 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)
-{
-  emit_addimm(rt,imm,rt);
-}
-
 static void emit_logicop_imm(u_int op, u_int rs, u_int imm, u_int rt)
 {
   const char *names[] = { "and", "orr", "eor", "ands" };
@@ -1337,7 +1333,27 @@ 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
+  assem_debug("movz %s,#%#lx\n", regname64[rt], rt_val & 0xffff);
+  output_w32(0xd2800000 | imm16_rd(rt_val & 0xffff, rt));
+  assem_debug("movk %s,#%#lx,lsl #16\n", regname64[rt], (rt_val >> 16) & 0xffff);
+  output_w32(0xf2a00000 | imm16_rd((rt_val >> 16) & 0xffff, rt));
+  assem_debug("movk %s,#%#lx,lsl #32\n", regname64[rt], (rt_val >> 32) & 0xffff);
+  output_w32(0xf2c00000 | imm16_rd((rt_val >> 32) & 0xffff, rt));
+  if (rt_val >> 48) {
+    assem_debug("movk %s,#%#lx,lsl #48\n", regname64[rt], (rt_val >> 48) & 0xffff);
+    output_w32(0xf2e00000 | imm16_rd((rt_val >> 48) & 0xffff, rt));
+  }
+}
+
+// trashes x2
 static void pass_args64(u_int a0, u_int a1)
 {
   if(a0==1&&a1==0) {
@@ -1474,11 +1490,8 @@ static void inline_readstub(enum stub_type type, int i, u_int 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, rs, host_addr, rs);
     switch(type) {
       case LOADB_STUB:  emit_movsbl_indexed(0,rs,rt); break;
       case LOADBU_STUB: emit_movzbl_indexed(0,rs,rt); break;
@@ -1489,8 +1502,8 @@ static void inline_readstub(enum stub_type type, int i, u_int addr,
     }
     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)
@@ -1627,11 +1640,8 @@ static void inline_writestub(enum stub_type type, int i, u_int addr,
   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, rs, host_addr, rs);
     switch (type) {
       case STOREB_STUB: emit_writebyte_indexed(rt, 0, rs); break;
       case STOREH_STUB: emit_writehword_indexed(rt, 0, rs); break;
diff --git a/libpcsxcore/new_dynarec/emu_if.h b/libpcsxcore/new_dynarec/emu_if.h
index 36cc275f..30cb9ef6 100644
--- a/libpcsxcore/new_dynarec/emu_if.h
+++ b/libpcsxcore/new_dynarec/emu_if.h
@@ -95,8 +95,4 @@ void pcsx_mtc0_ds(u32 reg, u32 val);
 /* misc */
 extern void SysPrintf(const char *fmt, ...);
 
-#ifdef RAM_FIXED
-#define rdram ((u_char *)0x80000000)
-#else
 #define rdram ((u_char *)psxM)
-#endif
diff --git a/libpcsxcore/new_dynarec/linkage_arm.S b/libpcsxcore/new_dynarec/linkage_arm.S
index 970d91c7..1d8cefaa 100644
--- a/libpcsxcore/new_dynarec/linkage_arm.S
+++ b/libpcsxcore/new_dynarec/linkage_arm.S
@@ -85,7 +85,7 @@ DRC_VAR(psxH_ptr, 4)
 DRC_VAR(zeromem_ptr, 4)
 DRC_VAR(invc_ptr, 4)
 DRC_VAR(scratch_buf_ptr, 4)
-@DRC_VAR(align1, 8) /* unused/alignment */
+DRC_VAR(ram_offset, 4)
 DRC_VAR(mini_ht, 256)
 DRC_VAR(restore_candidate, 512)
 
diff --git a/libpcsxcore/new_dynarec/linkage_arm64.S b/libpcsxcore/new_dynarec/linkage_arm64.S
index 6f32fa3b..7df82b4e 100644
--- a/libpcsxcore/new_dynarec/linkage_arm64.S
+++ b/libpcsxcore/new_dynarec/linkage_arm64.S
@@ -77,7 +77,7 @@ DRC_VAR(psxH_ptr, 8)
 DRC_VAR(invc_ptr, 8)
 DRC_VAR(zeromem_ptr, 8)
 DRC_VAR(scratch_buf_ptr, 8)
-#DRC_VAR(align1, 16) /* unused/alignment */
+DRC_VAR(ram_offset, 8)
 DRC_VAR(mini_ht, 256)
 DRC_VAR(restore_candidate, 512)
 
diff --git a/libpcsxcore/new_dynarec/linkage_offsets.h b/libpcsxcore/new_dynarec/linkage_offsets.h
index 916bb1a8..e9bb3abd 100644
--- a/libpcsxcore/new_dynarec/linkage_offsets.h
+++ b/libpcsxcore/new_dynarec/linkage_offsets.h
@@ -37,8 +37,8 @@
 #define LO_invc_ptr		(LO_zeromem_ptr + PTRSZ)
 #define LO_scratch_buf_ptr	(LO_invc_ptr + PTRSZ)
 #define LO_saved_lr		(LO_scratch_buf_ptr + PTRSZ)
-#define LO_align1		(LO_saved_lr + PTRSZ)
-#define LO_mini_ht		(LO_align1 + PTRSZ)
+#define LO_ram_offset		(LO_saved_lr + PTRSZ)
+#define LO_mini_ht		(LO_ram_offset + PTRSZ)
 #define LO_restore_candidate	(LO_mini_ht + PTRSZ*32*2)
 #define LO_dynarec_local_size	(LO_restore_candidate + 512)
 
diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c
index 910046f9..921a2ed1 100644
--- a/libpcsxcore/new_dynarec/new_dynarec.c
+++ b/libpcsxcore/new_dynarec/new_dynarec.c
@@ -176,6 +176,8 @@ static struct decoded_insn
   u_char is_ds:1;
   u_char is_jump:1;
   u_char is_ujump:1;
+  u_char is_load:1;
+  u_char is_store:1;
 } dops[MAXBLOCK];
 
   // used by asm:
@@ -226,11 +228,6 @@ static struct decoded_insn
   static int expirep;
   static u_int stop_after_jal;
   static u_int f1_hack; // 0 - off, ~0 - capture address, else addr
-#ifndef RAM_FIXED
-  static uintptr_t ram_offset;
-#else
-  static const uintptr_t ram_offset=0;
-#endif
 
   int new_dynarec_hacks;
   int new_dynarec_hacks_pergame;
@@ -244,6 +241,7 @@ static struct decoded_insn
   extern int pcaddr;
   extern int pending_exception;
   extern int branch_target;
+  extern uintptr_t ram_offset;
   extern uintptr_t mini_ht[32][2];
   extern u_char restore_candidate[512];
 
@@ -256,7 +254,7 @@ static struct decoded_insn
 #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
@@ -346,7 +344,8 @@ static void add_stub(enum stub_type type, void *addr, void *retaddr,
 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 *emit_fastpath_cmp_jump(int i, const struct regstat *i_regs,
+  int addr, int *offset_reg, 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_do_stall_check(u_int op, int i, const struct regstat *i_regs, u_int reglist);
@@ -707,11 +706,12 @@ void lsn(u_char hsn[], int i, int *preferred_reg)
       hsn[dops[i+j].rs1]=j;
       hsn[dops[i+j].rs2]=j;
     }
+    if (ram_offset && (dops[i+j].is_load || dops[i+j].is_store))
+      hsn[ROREG] = j;
     // On some architectures stores need invc_ptr
     #if defined(HOST_IMM8)
-    if(dops[i+j].itype==STORE || dops[i+j].itype==STORELR || (dops[i+j].opcode&0x3b)==0x39 || (dops[i+j].opcode&0x3b)==0x3a) {
-      hsn[INVCP]=j;
-    }
+    if (dops[i+j].is_store)
+      hsn[INVCP] = j;
     #endif
     if(i+j>=0&&(dops[i+j].itype==UJUMP||dops[i+j].itype==CJUMP||dops[i+j].itype==SJUMP))
     {
@@ -746,7 +746,7 @@ void lsn(u_char hsn[], int i, int *preferred_reg)
     hsn[RHTBL]=1;
   }
   // Coprocessor load/store needs FTEMP, even if not declared
-  if(dops[i].itype==C1LS||dops[i].itype==C2LS) {
+  if(dops[i].itype==C2LS) {
     hsn[FTEMP]=0;
   }
   // Load L/R also uses FTEMP as a temporary register
@@ -1756,7 +1756,10 @@ static void load_alloc(struct regstat *current,int i)
   clear_const(current,dops[i].rt1);
   //if(dops[i].rs1!=dops[i].rt1&&needed_again(dops[i].rs1,i)) clear_const(current,dops[i].rs1); // Does this help or hurt?
   if(!dops[i].rs1) current->u&=~1LL; // Allow allocating r0 if it's the source register
-  if(needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1);
+  if (needed_again(dops[i].rs1, i))
+    alloc_reg(current, i, dops[i].rs1);
+  if (ram_offset)
+    alloc_reg(current, i, ROREG);
   if(dops[i].rt1&&!((current->u>>dops[i].rt1)&1)) {
     alloc_reg(current,i,dops[i].rt1);
     assert(get_reg(current->regmap,dops[i].rt1)>=0);
@@ -1803,9 +1806,11 @@ void store_alloc(struct regstat *current,int i)
   if(dops[i].opcode==0x2c||dops[i].opcode==0x2d||dops[i].opcode==0x3f) { // 64-bit SDL/SDR/SD
     assert(0);
   }
+  if (ram_offset)
+    alloc_reg(current, i, ROREG);
   #if defined(HOST_IMM8)
   // On CPUs without 32-bit immediates we need a pointer to invalid_code
-  else alloc_reg(current,i,INVCP);
+  alloc_reg(current, i, INVCP);
   #endif
   if(dops[i].opcode==0x2a||dops[i].opcode==0x2e||dops[i].opcode==0x2c||dops[i].opcode==0x2d) { // SWL/SWL/SDL/SDR
     alloc_reg(current,i,FTEMP);
@@ -1817,21 +1822,8 @@ void store_alloc(struct regstat *current,int i)
 
 void c1ls_alloc(struct regstat *current,int i)
 {
-  //clear_const(current,dops[i].rs1); // FIXME
   clear_const(current,dops[i].rt1);
-  if(needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1);
   alloc_reg(current,i,CSREG); // Status
-  alloc_reg(current,i,FTEMP);
-  if(dops[i].opcode==0x35||dops[i].opcode==0x3d) { // 64-bit LDC1/SDC1
-    assert(0);
-  }
-  #if defined(HOST_IMM8)
-  // On CPUs without 32-bit immediates we need a pointer to invalid_code
-  else if((dops[i].opcode&0x3b)==0x39) // SWC1/SDC1
-    alloc_reg(current,i,INVCP);
-  #endif
-  // We need a temporary register for address generation
-  alloc_reg_temp(current,i,-1);
 }
 
 void c2ls_alloc(struct regstat *current,int i)
@@ -1839,9 +1831,11 @@ void c2ls_alloc(struct regstat *current,int i)
   clear_const(current,dops[i].rt1);
   if(needed_again(dops[i].rs1,i)) alloc_reg(current,i,dops[i].rs1);
   alloc_reg(current,i,FTEMP);
+  if (ram_offset)
+    alloc_reg(current, i, ROREG);
   #if defined(HOST_IMM8)
   // On CPUs without 32-bit immediates we need a pointer to invalid_code
-  if((dops[i].opcode&0x3b)==0x3a) // SWC2/SDC2
+  if (dops[i].opcode == 0x3a) // SWC2
     alloc_reg(current,i,INVCP);
   #endif
   // We need a temporary register for address generation
@@ -2562,11 +2556,25 @@ static int get_ptr_mem_type(u_int a)
   return MTYPE_8000;
 }
 
-static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override)
+static int get_ro_reg(const struct regstat *i_regs, int host_tempreg_free)
+{
+  int r = get_reg(i_regs->regmap, ROREG);
+  if (r < 0 && host_tempreg_free) {
+    host_tempreg_acquire();
+    emit_loadreg(ROREG, r = HOST_TEMPREG);
+  }
+  if (r < 0)
+    abort();
+  return r;
+}
+
+static void *emit_fastpath_cmp_jump(int i, const struct regstat *i_regs,
+  int addr, int *offset_reg, int *addr_reg_override)
 {
   void *jaddr = NULL;
-  int type=0;
-  int mr=dops[i].rs1;
+  int type = 0;
+  int mr = dops[i].rs1;
+  *offset_reg = -1;
   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);
@@ -2610,22 +2618,19 @@ static void *emit_fastpath_cmp_jump(int i,int addr,int *addr_reg_override)
     }
   }
 
-  if(type==0)
+  if (type == 0) // need ram check
   {
     emit_cmpimm(addr,RAM_SIZE);
-    jaddr=out;
+    jaddr = out;
     #ifdef CORTEX_A8_BRANCH_PREDICTION_HACK
     // Hint to branch predictor that the branch is unlikely to be taken
-    if(dops[i].rs1>=28)
+    if (dops[i].rs1 >= 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;
-    }
+    if (ram_offset != 0)
+      *offset_reg = get_ro_reg(i_regs, 0);
   }
 
   return jaddr;
@@ -2688,13 +2693,56 @@ static int reglist_find_free(u_int reglist)
   return __builtin_ctz(free_regs);
 }
 
+static void do_load_word(int a, int rt, int offset_reg)
+{
+  if (offset_reg >= 0)
+    emit_ldr_dualindexed(offset_reg, a, rt);
+  else
+    emit_readword_indexed(0, a, rt);
+}
+
+static void do_store_word(int a, int ofs, int rt, int offset_reg, int preseve_a)
+{
+  if (offset_reg < 0) {
+    emit_writeword_indexed(rt, ofs, a);
+    return;
+  }
+  if (ofs != 0)
+    emit_addimm(a, ofs, a);
+  emit_str_dualindexed(offset_reg, a, rt);
+  if (ofs != 0 && preseve_a)
+    emit_addimm(a, -ofs, a);
+}
+
+static void do_store_hword(int a, int ofs, int rt, int offset_reg, int preseve_a)
+{
+  if (offset_reg < 0) {
+    emit_writehword_indexed(rt, ofs, a);
+    return;
+  }
+  if (ofs != 0)
+    emit_addimm(a, ofs, a);
+  emit_strh_dualindexed(offset_reg, a, rt);
+  if (ofs != 0 && preseve_a)
+    emit_addimm(a, -ofs, a);
+}
+
+static void do_store_byte(int a, int rt, int offset_reg)
+{
+  if (offset_reg >= 0)
+    emit_strb_dualindexed(offset_reg, a, rt);
+  else
+    emit_writebyte_indexed(rt, 0, a);
+}
+
 static void load_assemble(int i, const struct regstat *i_regs)
 {
   int s,tl,addr;
   int offset;
   void *jaddr=0;
   int memtarget=0,c=0;
-  int fastio_reg_override=-1;
+  int offset_reg = -1;
+  int fastio_reg_override = -1;
   u_int reglist=get_host_reglist(i_regs->regmap);
   tl=get_reg(i_regs->regmap,dops[i].rt1);
   s=get_reg(i_regs->regmap,dops[i].rs1);
@@ -2731,96 +2779,110 @@ static void load_assemble(int i, const struct regstat *i_regs)
     if(dops[i].rs1!=29||start<0x80001000||start>=0x80000000+RAM_SIZE)
     #endif
     {
-      jaddr=emit_fastpath_cmp_jump(i,addr,&fastio_reg_override);
+      jaddr = emit_fastpath_cmp_jump(i, i_regs, addr,
+                &offset_reg, &fastio_reg_override);
     }
   }
-  else if(ram_offset&&memtarget) {
-    host_tempreg_acquire();
-    emit_addimm(addr,ram_offset,HOST_TEMPREG);
-    fastio_reg_override=HOST_TEMPREG;
+  else if (ram_offset && memtarget) {
+    offset_reg = get_ro_reg(i_regs, 0);
   }
   int dummy=(dops[i].rt1==0)||(tl!=get_reg(i_regs->regmap,dops[i].rt1)); // ignore loads to r0 and unneeded reg
-  if (dops[i].opcode==0x20) { // LB
+  switch (dops[i].opcode) {
+  case 0x20: // LB
     if(!c||memtarget) {
       if(!dummy) {
-        {
-          int x=0,a=tl;
-          if(!c) a=addr;
-          if(fastio_reg_override>=0) a=fastio_reg_override;
+        int a = tl;
+        if (!c) a = addr;
+        if (fastio_reg_override >= 0)
+          a = fastio_reg_override;
 
-          emit_movsbl_indexed(x,a,tl);
-        }
+        if (offset_reg >= 0)
+          emit_ldrsb_dualindexed(offset_reg, a, tl);
+        else
+          emit_movsbl_indexed(0, a, tl);
       }
       if(jaddr)
         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,dops[i].rt1,ccadj[i],reglist);
-  }
-  if (dops[i].opcode==0x21) { // LH
+    break;
+  case 0x21: // LH
     if(!c||memtarget) {
       if(!dummy) {
-        int x=0,a=tl;
-        if(!c) a=addr;
-        if(fastio_reg_override>=0) a=fastio_reg_override;
-        emit_movswl_indexed(x,a,tl);
+        int a = tl;
+        if (!c) a = addr;
+        if (fastio_reg_override >= 0)
+          a = fastio_reg_override;
+        if (offset_reg >= 0)
+          emit_ldrsh_dualindexed(offset_reg, a, tl);
+        else
+          emit_movswl_indexed(0, a, tl);
       }
       if(jaddr)
         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,dops[i].rt1,ccadj[i],reglist);
-  }
-  if (dops[i].opcode==0x23) { // LW
+    break;
+  case 0x23: // LW
     if(!c||memtarget) {
       if(!dummy) {
-        int a=addr;
-        if(fastio_reg_override>=0) a=fastio_reg_override;
-        emit_readword_indexed(0,a,tl);
+        int a = addr;
+        if (fastio_reg_override >= 0)
+          a = fastio_reg_override;
+        do_load_word(a, tl, offset_reg);
       }
       if(jaddr)
         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,dops[i].rt1,ccadj[i],reglist);
-  }
-  if (dops[i].opcode==0x24) { // LBU
+    break;
+  case 0x24: // LBU
     if(!c||memtarget) {
       if(!dummy) {
-        int x=0,a=tl;
-        if(!c) a=addr;
-        if(fastio_reg_override>=0) a=fastio_reg_override;
+        int a = tl;
+        if (!c) a = addr;
+        if (fastio_reg_override >= 0)
+          a = fastio_reg_override;
 
-        emit_movzbl_indexed(x,a,tl);
+        if (offset_reg >= 0)
+          emit_ldrb_dualindexed(offset_reg, a, tl);
+        else
+          emit_movzbl_indexed(0, a, tl);
       }
       if(jaddr)
         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,dops[i].rt1,ccadj[i],reglist);
-  }
-  if (dops[i].opcode==0x25) { // LHU
+    break;
+  case 0x25: // LHU
     if(!c||memtarget) {
       if(!dummy) {
-        int x=0,a=tl;
-        if(!c) a=addr;
-        if(fastio_reg_override>=0) a=fastio_reg_override;
-        emit_movzwl_indexed(x,a,tl);
+        int a = tl;
+        if(!c) a = addr;
+        if (fastio_reg_override >= 0)
+          a = fastio_reg_override;
+        if (offset_reg >= 0)
+          emit_ldrh_dualindexed(offset_reg, a, tl);
+        else
+          emit_movzwl_indexed(0, a, tl);
       }
       if(jaddr)
         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,dops[i].rt1,ccadj[i],reglist);
-  }
-  if (dops[i].opcode==0x27) { // LWU
-    assert(0);
-  }
-  if (dops[i].opcode==0x37) { // LD
+    break;
+  case 0x27: // LWU
+  case 0x37: // LD
+  default:
     assert(0);
   }
  }
- if (fastio_reg_override == HOST_TEMPREG)
+ if (fastio_reg_override == HOST_TEMPREG || offset_reg == HOST_TEMPREG)
    host_tempreg_release();
 }
 
@@ -2831,7 +2893,8 @@ static void loadlr_assemble(int i, const struct regstat *i_regs)
   int offset;
   void *jaddr=0;
   int memtarget=0,c=0;
-  int fastio_reg_override=-1;
+  int offset_reg = -1;
+  int fastio_reg_override = -1;
   u_int reglist=get_host_reglist(i_regs->regmap);
   tl=get_reg(i_regs->regmap,dops[i].rt1);
   s=get_reg(i_regs->regmap,dops[i].rs1);
@@ -2856,13 +2919,12 @@ static void loadlr_assemble(int i, const struct regstat *i_regs)
     }else{
       emit_andimm(addr,0xFFFFFFF8,temp2); // LDL/LDR
     }
-    jaddr=emit_fastpath_cmp_jump(i,temp2,&fastio_reg_override);
+    jaddr = emit_fastpath_cmp_jump(i, i_regs, temp2,
+              &offset_reg, &fastio_reg_override);
   }
   else {
-    if(ram_offset&&memtarget) {
-      host_tempreg_acquire();
-      emit_addimm(temp2,ram_offset,HOST_TEMPREG);
-      fastio_reg_override=HOST_TEMPREG;
+    if (ram_offset && memtarget) {
+      offset_reg = get_ro_reg(i_regs, 0);
     }
     if (dops[i].opcode==0x22||dops[i].opcode==0x26) {
       emit_movimm(((constmap[i][s]+offset)<<3)&24,temp); // LWL/LWR
@@ -2872,10 +2934,12 @@ static void loadlr_assemble(int i, const struct regstat *i_regs)
   }
   if (dops[i].opcode==0x22||dops[i].opcode==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();
+      int a = temp2;
+      if (fastio_reg_override >= 0)
+        a = fastio_reg_override;
+      do_load_word(a, temp2, offset_reg);
+      if (fastio_reg_override == HOST_TEMPREG || offset_reg == HOST_TEMPREG)
+        host_tempreg_release();
       if(jaddr) add_stub_r(LOADW_STUB,jaddr,out,i,temp2,i_regs,ccadj[i],reglist);
     }
     else
@@ -2905,16 +2969,17 @@ static void loadlr_assemble(int i, const struct regstat *i_regs)
 }
 #endif
 
-void store_assemble(int i, const struct regstat *i_regs)
+static void store_assemble(int i, const struct regstat *i_regs)
 {
   int s,tl;
   int addr,temp;
   int offset;
   void *jaddr=0;
-  enum stub_type type;
+  enum stub_type type=0;
   int memtarget=0,c=0;
   int agr=AGEN1+(i&1);
-  int fastio_reg_override=-1;
+  int offset_reg = -1;
+  int fastio_reg_override = -1;
   u_int reglist=get_host_reglist(i_regs->regmap);
   tl=get_reg(i_regs->regmap,dops[i].rs2);
   s=get_reg(i_regs->regmap,dops[i].rs1);
@@ -2932,46 +2997,49 @@ void store_assemble(int i, const struct regstat *i_regs)
   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,&fastio_reg_override);
+  if (!c) {
+    jaddr = emit_fastpath_cmp_jump(i, i_regs, addr,
+              &offset_reg, &fastio_reg_override);
   }
-  else if(ram_offset&&memtarget) {
-    host_tempreg_acquire();
-    emit_addimm(addr,ram_offset,HOST_TEMPREG);
-    fastio_reg_override=HOST_TEMPREG;
+  else if (ram_offset && memtarget) {
+    offset_reg = get_ro_reg(i_regs, 0);
   }
 
-  if (dops[i].opcode==0x28) { // SB
+  switch (dops[i].opcode) {
+  case 0x28: // SB
     if(!c||memtarget) {
-      int x=0,a=temp;
-      if(!c) a=addr;
-      if(fastio_reg_override>=0) a=fastio_reg_override;
-      emit_writebyte_indexed(tl,x,a);
-    }
-    type=STOREB_STUB;
-  }
-  if (dops[i].opcode==0x29) { // SH
+      int a = temp;
+      if (!c) a = addr;
+      if (fastio_reg_override >= 0)
+        a = fastio_reg_override;
+      do_store_byte(a, tl, offset_reg);
+    }
+    type = STOREB_STUB;
+    break;
+  case 0x29: // SH
     if(!c||memtarget) {
-      int x=0,a=temp;
-      if(!c) a=addr;
-      if(fastio_reg_override>=0) a=fastio_reg_override;
-      emit_writehword_indexed(tl,x,a);
-    }
-    type=STOREH_STUB;
-  }
-  if (dops[i].opcode==0x2B) { // SW
+      int a = temp;
+      if (!c) a = addr;
+      if (fastio_reg_override >= 0)
+        a = fastio_reg_override;
+      do_store_hword(a, 0, tl, offset_reg, 1);
+    }
+    type = STOREH_STUB;
+    break;
+  case 0x2B: // SW
     if(!c||memtarget) {
-      int a=addr;
-      if(fastio_reg_override>=0) a=fastio_reg_override;
-      emit_writeword_indexed(tl,0,a);
-    }
-    type=STOREW_STUB;
-  }
-  if (dops[i].opcode==0x3F) { // SD
+      int a = addr;
+      if (fastio_reg_override >= 0)
+        a = fastio_reg_override;
+      do_store_word(a, 0, tl, offset_reg, 1);
+    }
+    type = STOREW_STUB;
+    break;
+  case 0x3F: // SD
+  default:
     assert(0);
-    type=STORED_STUB;
   }
-  if(fastio_reg_override==HOST_TEMPREG)
+  if (fastio_reg_override == HOST_TEMPREG || offset_reg == HOST_TEMPREG)
     host_tempreg_release();
   if(jaddr) {
     // PCSX store handlers don't check invcode again
@@ -3032,10 +3100,11 @@ static void storelr_assemble(int i, const struct regstat *i_regs)
   int temp;
   int offset;
   void *jaddr=0;
-  void *case1, *case2, *case3;
+  void *case1, *case23, *case3;
   void *done0, *done1, *done2;
   int memtarget=0,c=0;
   int agr=AGEN1+(i&1);
+  int offset_reg = -1;
   u_int reglist=get_host_reglist(i_regs->regmap);
   tl=get_reg(i_regs->regmap,dops[i].rs2);
   s=get_reg(i_regs->regmap,dops[i].rs1);
@@ -3063,86 +3132,85 @@ static void storelr_assemble(int i, const struct regstat *i_regs)
       emit_jmp(0);
     }
   }
-  if(ram_offset)
-    emit_addimm_no_flags(ram_offset,temp);
+  if (ram_offset)
+    offset_reg = get_ro_reg(i_regs, 0);
 
   if (dops[i].opcode==0x2C||dops[i].opcode==0x2D) { // SDL/SDR
     assert(0);
   }
 
-  emit_xorimm(temp,3,temp);
   emit_testimm(temp,2);
-  case2=out;
+  case23=out;
   emit_jne(0);
   emit_testimm(temp,1);
   case1=out;
   emit_jne(0);
   // 0
-  if (dops[i].opcode==0x2A) { // SWL
-    emit_writeword_indexed(tl,0,temp);
+  if (dops[i].opcode == 0x2A) { // SWL
+    // Write msb into least significant byte
+    if (dops[i].rs2) emit_rorimm(tl, 24, tl);
+    do_store_byte(temp, tl, offset_reg);
+    if (dops[i].rs2) emit_rorimm(tl, 8, tl);
   }
-  else if (dops[i].opcode==0x2E) { // SWR
-    emit_writebyte_indexed(tl,3,temp);
+  else if (dops[i].opcode == 0x2E) { // SWR
+    // Write entire word
+    do_store_word(temp, 0, tl, offset_reg, 1);
   }
-  else
-    assert(0);
-  done0=out;
+  done0 = out;
   emit_jmp(0);
   // 1
   set_jump_target(case1, out);
-  if (dops[i].opcode==0x2A) { // SWL
-    // Write 3 msb into three least significant bytes
-    if(dops[i].rs2) emit_rorimm(tl,8,tl);
-    emit_writehword_indexed(tl,-1,temp);
-    if(dops[i].rs2) emit_rorimm(tl,16,tl);
-    emit_writebyte_indexed(tl,1,temp);
-    if(dops[i].rs2) emit_rorimm(tl,8,tl);
+  if (dops[i].opcode == 0x2A) { // SWL
+    // Write two msb into two least significant bytes
+    if (dops[i].rs2) emit_rorimm(tl, 16, tl);
+    do_store_hword(temp, -1, tl, offset_reg, 0);
+    if (dops[i].rs2) emit_rorimm(tl, 16, tl);
   }
-  else if (dops[i].opcode==0x2E) { // SWR
-    // Write two lsb into two most significant bytes
-    emit_writehword_indexed(tl,1,temp);
+  else if (dops[i].opcode == 0x2E) { // SWR
+    // Write 3 lsb into three most significant bytes
+    do_store_byte(temp, tl, offset_reg);
+    if (dops[i].rs2) emit_rorimm(tl, 8, tl);
+    do_store_hword(temp, 1, tl, offset_reg, 0);
+    if (dops[i].rs2) emit_rorimm(tl, 24, tl);
   }
   done1=out;
   emit_jmp(0);
-  // 2
-  set_jump_target(case2, out);
+  // 2,3
+  set_jump_target(case23, out);
   emit_testimm(temp,1);
-  case3=out;
+  case3 = out;
   emit_jne(0);
+  // 2
   if (dops[i].opcode==0x2A) { // SWL
-    // Write two msb into two least significant bytes
-    if(dops[i].rs2) emit_rorimm(tl,16,tl);
-    emit_writehword_indexed(tl,-2,temp);
-    if(dops[i].rs2) emit_rorimm(tl,16,tl);
+    // Write 3 msb into three least significant bytes
+    if (dops[i].rs2) emit_rorimm(tl, 8, tl);
+    do_store_hword(temp, -2, tl, offset_reg, 1);
+    if (dops[i].rs2) emit_rorimm(tl, 16, tl);
+    do_store_byte(temp, tl, offset_reg);
+    if (dops[i].rs2) emit_rorimm(tl, 8, tl);
   }
-  else if (dops[i].opcode==0x2E) { // SWR
-    // Write 3 lsb into three most significant bytes
-    emit_writebyte_indexed(tl,-1,temp);
-    if(dops[i].rs2) emit_rorimm(tl,8,tl);
-    emit_writehword_indexed(tl,0,temp);
-    if(dops[i].rs2) emit_rorimm(tl,24,tl);
+  else if (dops[i].opcode == 0x2E) { // SWR
+    // Write two lsb into two most significant bytes
+    do_store_hword(temp, 0, tl, offset_reg, 1);
   }
-  done2=out;
+  done2 = out;
   emit_jmp(0);
   // 3
   set_jump_target(case3, out);
-  if (dops[i].opcode==0x2A) { // SWL
-    // Write msb into least significant byte
-    if(dops[i].rs2) emit_rorimm(tl,24,tl);
-    emit_writebyte_indexed(tl,-3,temp);
-    if(dops[i].rs2) emit_rorimm(tl,8,tl);
+  if (dops[i].opcode == 0x2A) { // SWL
+    do_store_word(temp, -3, tl, offset_reg, 0);
   }
-  else if (dops[i].opcode==0x2E) { // SWR
-    // Write entire word
-    emit_writeword_indexed(tl,-3,temp);
+  else if (dops[i].opcode == 0x2E) { // SWR
+    do_store_byte(temp, tl, offset_reg);
   }
   set_jump_target(done0, out);
   set_jump_target(done1, out);
   set_jump_target(done2, out);
+  if (offset_reg == HOST_TEMPREG)
+    host_tempreg_release();
   if(!c||!memtarget)
     add_stub_r(STORELR_STUB,jaddr,out,i,temp,i_regs,ccadj[i],reglist);
   if(!(i_regs->waswritten&(1<<dops[i].rs1)) && !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);
@@ -3599,7 +3667,8 @@ static void c2ls_assemble(int i, const struct regstat *i_regs)
   void *jaddr2=NULL;
   enum stub_type type;
   int agr=AGEN1+(i&1);
-  int fastio_reg_override=-1;
+  int offset_reg = -1;
+  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,dops[i].rs1);
@@ -3639,28 +3708,35 @@ static void c2ls_assemble(int i, const struct regstat *i_regs)
   }
   else {
     if(!c) {
-      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 (dops[i].opcode==0x32) { // LWC2
-      int a=ar;
-      if(fastio_reg_override>=0) a=fastio_reg_override;
-      emit_readword_indexed(0,a,tl);
+      jaddr2 = emit_fastpath_cmp_jump(i, i_regs, ar,
+                &offset_reg, &fastio_reg_override);
+    }
+    else if (ram_offset && memtarget) {
+      offset_reg = get_ro_reg(i_regs, 0);
+    }
+    switch (dops[i].opcode) {
+    case 0x32: { // LWC2
+      int a = ar;
+      if (fastio_reg_override >= 0)
+        a = fastio_reg_override;
+      do_load_word(a, tl, offset_reg);
+      break;
     }
-    if (dops[i].opcode==0x3a) { // SWC2
+    case 0x3a: { // SWC2
       #ifdef DESTRUCTIVE_SHIFT
       if(!offset&&!c&&s>=0) emit_mov(s,ar);
       #endif
-      int a=ar;
-      if(fastio_reg_override>=0) a=fastio_reg_override;
-      emit_writeword_indexed(tl,0,a);
+      int a = ar;
+      if (fastio_reg_override >= 0)
+        a = fastio_reg_override;
+      do_store_word(a, 0, tl, offset_reg, 1);
+      break;
+    }
+    default:
+      assert(0);
     }
   }
-  if(fastio_reg_override==HOST_TEMPREG)
+  if (fastio_reg_override == HOST_TEMPREG || offset_reg == HOST_TEMPREG)
     host_tempreg_release();
   if(jaddr2)
     add_stub_r(type,jaddr2,out,i,ar,i_regs,ccadj[i],reglist);
@@ -4089,7 +4165,7 @@ static void loop_preload(signed char pre[],signed char entry[])
 // goes to AGEN for writes, FTEMP for LOADLR and cop1/2 loads
 void address_generation(int i,struct regstat *i_regs,signed char entry[])
 {
-  if(dops[i].itype==LOAD||dops[i].itype==LOADLR||dops[i].itype==STORE||dops[i].itype==STORELR||dops[i].itype==C1LS||dops[i].itype==C2LS) {
+  if (dops[i].is_load || dops[i].is_store) {
     int ra=-1;
     int agr=AGEN1+(i&1);
     if(dops[i].itype==LOAD) {
@@ -4104,7 +4180,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[])
       ra=get_reg(i_regs->regmap,agr);
       if(ra<0) ra=get_reg(i_regs->regmap,-1);
     }
-    if(dops[i].itype==C1LS||dops[i].itype==C2LS) {
+    if(dops[i].itype==C2LS) {
       if ((dops[i].opcode&0x3b)==0x31||(dops[i].opcode&0x3b)==0x32) // LWC1/LDC1/LWC2/LDC2
         ra=get_reg(i_regs->regmap,FTEMP);
       else { // SWC1/SDC1/SWC2/SDC2
@@ -4158,7 +4234,7 @@ void address_generation(int i,struct regstat *i_regs,signed char entry[])
     }
   }
   // Preload constants for next instruction
-  if(dops[i+1].itype==LOAD||dops[i+1].itype==LOADLR||dops[i+1].itype==STORE||dops[i+1].itype==STORELR||dops[i+1].itype==C1LS||dops[i+1].itype==C2LS) {
+  if (dops[i+1].is_load || dops[i+1].is_store) {
     int agr,ra;
     // Actual address
     agr=AGEN1+((i+1)&1);
@@ -4593,7 +4669,9 @@ static void ds_assemble_entry(int i)
     wb_register(CCREG,regs[t].regmap_entry,regs[t].wasdirty);
   load_regs(regs[t].regmap_entry,regs[t].regmap,dops[t].rs1,dops[t].rs2);
   address_generation(t,&regs[t],regs[t].regmap_entry);
-  if(dops[t].itype==STORE||dops[t].itype==STORELR||(dops[t].opcode&0x3b)==0x39||(dops[t].opcode&0x3b)==0x3a)
+  if (ram_offset && (dops[t].is_load || dops[t].is_store))
+    load_regs(regs[t].regmap_entry,regs[t].regmap,ROREG,ROREG);
+  if (dops[t].is_store)
     load_regs(regs[t].regmap_entry,regs[t].regmap,INVCP,INVCP);
   is_delayslot=0;
   switch(dops[t].itype) {
@@ -5372,6 +5450,8 @@ static void cjump_assemble(int i,struct regstat *i_regs)
       // load regs
       load_regs(regs[i].regmap,branch_regs[i].regmap,dops[i+1].rs1,dops[i+1].rs2);
       address_generation(i+1,&branch_regs[i],0);
+      if (ram_offset)
+        load_regs(regs[i].regmap,branch_regs[i].regmap,ROREG,ROREG);
       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);
@@ -5403,9 +5483,12 @@ static void cjump_assemble(int i,struct regstat *i_regs)
       set_jump_target(nottaken, out);
       assem_debug("2:\n");
       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,dops[i+1].rs1,dops[i+1].rs2);
       address_generation(i+1,&branch_regs[i],0);
-      load_regs(regs[i].regmap,branch_regs[i].regmap,CCREG,CCREG);
+      if (ram_offset)
+        load_regs(regs[i].regmap,branch_regs[i].regmap,ROREG,ROREG);
+      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) {
@@ -5638,6 +5721,8 @@ static void sjump_assemble(int i,struct regstat *i_regs)
       // load regs
       load_regs(regs[i].regmap,branch_regs[i].regmap,dops[i+1].rs1,dops[i+1].rs2);
       address_generation(i+1,&branch_regs[i],0);
+      if (ram_offset)
+        load_regs(regs[i].regmap,branch_regs[i].regmap,ROREG,ROREG);
       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);
@@ -5916,7 +6001,9 @@ static void pagespan_ds()
     emit_writeword(HOST_BTREG,&branch_target);
   load_regs(regs[0].regmap_entry,regs[0].regmap,dops[0].rs1,dops[0].rs2);
   address_generation(0,&regs[0],regs[0].regmap_entry);
-  if(dops[0].itype==STORE||dops[0].itype==STORELR||(dops[0].opcode&0x3b)==0x39||(dops[0].opcode&0x3b)==0x3a)
+  if (ram_offset && (dops[0].is_load || dops[0].is_store))
+    load_regs(regs[0].regmap_entry,regs[0].regmap,ROREG,ROREG);
+  if (dops[0].is_store)
     load_regs(regs[0].regmap_entry,regs[0].regmap,INVCP,INVCP);
   is_delayslot=0;
   switch(dops[0].itype) {
@@ -6755,9 +6842,7 @@ void new_dynarec_init(void)
 #endif
   arch_init();
   new_dynarec_test();
-#ifndef RAM_FIXED
   ram_offset=(uintptr_t)rdram-0x80000000;
-#endif
   if (ram_offset!=0)
     SysPrintf("warning: RAM is not directly mapped, performance will suffer\n");
 }
@@ -7409,6 +7494,8 @@ int new_recompile_block(u_int addr)
 
     dops[i].is_jump = (dops[i].itype == RJUMP || dops[i].itype == UJUMP || dops[i].itype == CJUMP || dops[i].itype == SJUMP);
     dops[i].is_ujump = (dops[i].itype == RJUMP || dops[i].itype == UJUMP); // || (source[i] >> 16) == 0x1000 // beq r0,r0
+    dops[i].is_load = (dops[i].itype == LOAD || dops[i].itype == LOADLR || op == 0x32); // LWC2
+    dops[i].is_store = (dops[i].itype == STORE || dops[i].itype == STORELR || op == 0x3a); // SWC2
 
     /* messy cases to just pass over to the interpreter */
     if (i > 0 && dops[i-1].is_jump) {
@@ -8213,7 +8300,11 @@ int new_recompile_block(u_int addr)
         if(dops[i+1].rs2==regmap_pre[i][hr]) nr|=1<<hr;
         if(dops[i+1].rs1==regs[i].regmap_entry[hr]) nr|=1<<hr;
         if(dops[i+1].rs2==regs[i].regmap_entry[hr]) nr|=1<<hr;
-        if(dops[i+1].itype==STORE || dops[i+1].itype==STORELR || (dops[i+1].opcode&0x3b)==0x39 || (dops[i+1].opcode&0x3b)==0x3a) {
+        if(ram_offset && (dops[i+1].is_load || dops[i+1].is_store)) {
+          if(regmap_pre[i][hr]==ROREG) nr|=1<<hr;
+          if(regs[i].regmap_entry[hr]==ROREG) nr|=1<<hr;
+        }
+        if(dops[i+1].is_store) {
           if(regmap_pre[i][hr]==INVCP) nr|=1<<hr;
           if(regs[i].regmap_entry[hr]==INVCP) nr|=1<<hr;
         }
@@ -8251,7 +8342,11 @@ int new_recompile_block(u_int addr)
       if(dops[i].rs2==regmap_pre[i][hr]) nr|=1<<hr;
       if(dops[i].rs1==regs[i].regmap_entry[hr]) nr|=1<<hr;
       if(dops[i].rs2==regs[i].regmap_entry[hr]) nr|=1<<hr;
-      if(dops[i].itype==STORE || dops[i].itype==STORELR || (dops[i].opcode&0x3b)==0x39 || (dops[i].opcode&0x3b)==0x3a) {
+      if(ram_offset && (dops[i].is_load || dops[i].is_store)) {
+        if(regmap_pre[i][hr]==ROREG) nr|=1<<hr;
+        if(regs[i].regmap_entry[hr]==ROREG) nr|=1<<hr;
+      }
+      if(dops[i].is_store) {
         if(regmap_pre[i][hr]==INVCP) nr|=1<<hr;
         if(regs[i].regmap_entry[hr]==INVCP) nr|=1<<hr;
       }
@@ -8285,14 +8380,13 @@ int new_recompile_block(u_int addr)
         if(regs[i].regmap_entry[hr]!=CCREG) regs[i].regmap_entry[hr]=-1;
         if(dops[i].is_jump)
         {
-          int map=0,temp=0;
-          if(dops[i+1].itype==STORE || dops[i+1].itype==STORELR ||
-             (dops[i+1].opcode&0x3b)==0x39 || (dops[i+1].opcode&0x3b)==0x3a) { // SWC1/SDC1 || SWC2/SDC2
-            map=INVCP;
-          }
-          if(dops[i+1].itype==LOADLR || dops[i+1].itype==STORELR ||
-             dops[i+1].itype==C1LS || dops[i+1].itype==C2LS)
-            temp=FTEMP;
+          int map1 = 0, map2 = 0, temp = 0; // or -1 ??
+          if (dops[i+1].is_load || dops[i+1].is_store)
+            map1 = ROREG;
+          if (dops[i+1].is_store)
+            map2 = INVCP;
+          if(dops[i+1].itype==LOADLR || dops[i+1].itype==STORELR || dops[i+1].itype==C2LS)
+            temp = FTEMP;
           if((regs[i].regmap[hr]&63)!=dops[i].rs1 && (regs[i].regmap[hr]&63)!=dops[i].rs2 &&
              (regs[i].regmap[hr]&63)!=dops[i].rt1 && (regs[i].regmap[hr]&63)!=dops[i].rt2 &&
              (regs[i].regmap[hr]&63)!=dops[i+1].rt1 && (regs[i].regmap[hr]&63)!=dops[i+1].rt2 &&
@@ -8300,7 +8394,7 @@ int new_recompile_block(u_int addr)
              (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=PTEMP &&
              regs[i].regmap[hr]!=RHASH && regs[i].regmap[hr]!=RHTBL &&
              regs[i].regmap[hr]!=RTEMP && regs[i].regmap[hr]!=CCREG &&
-             regs[i].regmap[hr]!=map )
+             regs[i].regmap[hr]!=map1 && regs[i].regmap[hr]!=map2)
           {
             regs[i].regmap[hr]=-1;
             regs[i].isconst&=~(1<<hr);
@@ -8311,7 +8405,7 @@ int new_recompile_block(u_int addr)
                (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 &&
                branch_regs[i].regmap[hr]!=RTEMP && branch_regs[i].regmap[hr]!=CCREG &&
-               branch_regs[i].regmap[hr]!=map)
+               branch_regs[i].regmap[hr]!=map1 && branch_regs[i].regmap[hr]!=map2)
             {
               branch_regs[i].regmap[hr]=-1;
               branch_regs[i].regmap_entry[hr]=-1;
@@ -8330,17 +8424,16 @@ int new_recompile_block(u_int addr)
           // Non-branch
           if(i>0)
           {
-            int map=-1,temp=-1;
-            if(dops[i].itype==STORE || dops[i].itype==STORELR ||
-                      (dops[i].opcode&0x3b)==0x39 || (dops[i].opcode&0x3b)==0x3a) { // SWC1/SDC1 || SWC2/SDC2
-              map=INVCP;
-            }
-            if(dops[i].itype==LOADLR || dops[i].itype==STORELR ||
-               dops[i].itype==C1LS || dops[i].itype==C2LS)
-              temp=FTEMP;
+            int map1 = -1, map2 = -1, temp=-1;
+            if (dops[i].is_load || dops[i].is_store)
+              map1 = ROREG;
+            if (dops[i].is_store)
+              map2 = INVCP;
+            if (dops[i].itype==LOADLR || dops[i].itype==STORELR || dops[i].itype==C2LS)
+              temp = FTEMP;
             if((regs[i].regmap[hr]&63)!=dops[i].rt1 && (regs[i].regmap[hr]&63)!=dops[i].rt2 &&
                regs[i].regmap[hr]!=dops[i].rs1 && regs[i].regmap[hr]!=dops[i].rs2 &&
-               (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=map &&
+               (regs[i].regmap[hr]&63)!=temp && regs[i].regmap[hr]!=map1 && regs[i].regmap[hr]!=map2 &&
                (dops[i].itype!=SPAN||regs[i].regmap[hr]!=CCREG))
             {
               if(i<slen-1&&!dops[i].is_ds) {
@@ -9049,7 +9142,9 @@ int new_recompile_block(u_int addr)
           load_regs(regs[i].regmap_entry,regs[i].regmap,dops[i+1].rs1,dops[i+1].rs1);
         if(dops[i+1].rs2!=dops[i+1].rs1&&dops[i+1].rs2!=dops[i].rs1&&dops[i+1].rs2!=dops[i].rs2&&(dops[i+1].rs2!=dops[i].rt1||dops[i].rt1==0))
           load_regs(regs[i].regmap_entry,regs[i].regmap,dops[i+1].rs2,dops[i+1].rs2);
-        if(dops[i+1].itype==STORE||dops[i+1].itype==STORELR||(dops[i+1].opcode&0x3b)==0x39||(dops[i+1].opcode&0x3b)==0x3a)
+        if (ram_offset && (dops[i+1].is_load || dops[i+1].is_store))
+          load_regs(regs[i].regmap_entry,regs[i].regmap,ROREG,ROREG);
+        if (dops[i+1].is_store)
           load_regs(regs[i].regmap_entry,regs[i].regmap,INVCP,INVCP);
       }
       else if(i+1<slen)
@@ -9063,9 +9158,11 @@ int new_recompile_block(u_int addr)
             load_regs(regs[i].regmap_entry,regs[i].regmap,dops[i+1].rs2,dops[i+1].rs2);
       }
       // TODO: if(is_ooo(i)) address_generation(i+1);
-      if(dops[i].itype==CJUMP)
+      if (dops[i].itype == CJUMP)
         load_regs(regs[i].regmap_entry,regs[i].regmap,CCREG,CCREG);
-      if(dops[i].itype==STORE||dops[i].itype==STORELR||(dops[i].opcode&0x3b)==0x39||(dops[i].opcode&0x3b)==0x3a)
+      if (ram_offset && (dops[i].is_load || dops[i].is_store))
+        load_regs(regs[i].regmap_entry,regs[i].regmap,ROREG,ROREG);
+      if (dops[i].is_store)
         load_regs(regs[i].regmap_entry,regs[i].regmap,INVCP,INVCP);
       // assemble
       switch(dops[i].itype) {
@@ -9352,6 +9449,9 @@ int new_recompile_block(u_int addr)
     }
     expirep=(expirep+1)&65535;
   }
+#ifdef ASSEM_PRINT
+  fflush(stdout);
+#endif
   return 0;
 }
 
diff --git a/libpcsxcore/new_dynarec/patches/trace_drc_chk b/libpcsxcore/new_dynarec/patches/trace_drc_chk
index eca104d6..93ca5980 100644
--- a/libpcsxcore/new_dynarec/patches/trace_drc_chk
+++ b/libpcsxcore/new_dynarec/patches/trace_drc_chk
@@ -1,8 +1,8 @@
 diff --git a/libpcsxcore/new_dynarec/linkage_arm.S b/libpcsxcore/new_dynarec/linkage_arm.S
-index bbc52c3..83c5b08 100644
+index 1d8cefa..528929f 100644
 --- a/libpcsxcore/new_dynarec/linkage_arm.S
 +++ b/libpcsxcore/new_dynarec/linkage_arm.S
-@@ -437,7 +437,7 @@ FUNCTION(cc_interrupt):
+@@ -438,7 +438,7 @@ FUNCTION(cc_interrupt):
  	str	r1, [fp, #LO_pending_exception]
  	and	r2, r2, r10, lsr #17
  	add	r3, fp, #LO_restore_candidate
@@ -11,7 +11,7 @@ index bbc52c3..83c5b08 100644
  @@	str	r10, [fp, #LO_reg_cop0+36]	/* Count */
  	ldr	r4, [r2, r3]
  	mov	r10, lr
-@@ -527,7 +527,7 @@ FUNCTION(new_dyna_leave):
+@@ -528,7 +528,7 @@ FUNCTION(new_dyna_leave):
  	ldr	r0, [fp, #LO_last_count]
  	add	r12, fp, #28
  	add	r10, r0, r10
@@ -20,7 +20,7 @@ index bbc52c3..83c5b08 100644
  	ldmfd	sp!, {r4, r5, r6, r7, r8, r9, sl, fp, ip, pc}
  	.size	new_dyna_leave, .-new_dyna_leave
  
-@@ -644,7 +644,7 @@ FUNCTION(new_dyna_start):
+@@ -645,7 +645,7 @@ FUNCTION(new_dyna_start):
  	\readop	r0, [r1, r3, lsl #\tab_shift]
  .endif
  	movcc	pc, lr
@@ -29,7 +29,7 @@ index bbc52c3..83c5b08 100644
  	bx	r1
  .endm
  
-@@ -679,7 +679,7 @@ FUNCTION(jump_handler_read32):
+@@ -680,7 +680,7 @@ FUNCTION(jump_handler_read32):
  	mov     r0, r1
  	add	r2, r2, r12
  	push	{r2, lr}
@@ -38,7 +38,7 @@ index bbc52c3..83c5b08 100644
  	blx	r3
  
  	ldr	r0, [fp, #LO_next_interupt]
-@@ -707,7 +707,7 @@ FUNCTION(jump_handler_write_h):
+@@ -708,7 +708,7 @@ FUNCTION(jump_handler_write_h):
  	add	r2, r2, r12
  	mov     r0, r1
  	push	{r2, lr}
@@ -48,10 +48,10 @@ index bbc52c3..83c5b08 100644
  
  	ldr	r0, [fp, #LO_next_interupt]
 diff --git a/libpcsxcore/new_dynarec/linkage_arm64.S b/libpcsxcore/new_dynarec/linkage_arm64.S
-index 444545c..031cee2 100644
+index 7df82b4..79298e4 100644
 --- a/libpcsxcore/new_dynarec/linkage_arm64.S
 +++ b/libpcsxcore/new_dynarec/linkage_arm64.S
-@@ -119,7 +119,7 @@ FUNCTION(cc_interrupt):
+@@ -123,7 +123,7 @@ FUNCTION(cc_interrupt):
  	str	wzr, [rFP, #LO_pending_exception]
  	and	w2, w2, rCC, lsr #17
  	add	x3, rFP, #LO_restore_candidate
@@ -60,7 +60,7 @@ index 444545c..031cee2 100644
  #	str	rCC, [rFP, #LO_reg_cop0+36]	/* Count */
  	ldr	w19, [x3, w2, uxtw]
  	mov	x21, lr
-@@ -227,7 +227,7 @@ FUNCTION(new_dyna_start):
+@@ -231,7 +231,7 @@ FUNCTION(new_dyna_start):
  FUNCTION(new_dyna_leave):
  	ldr	w0,  [rFP, #LO_last_count]
  	add	rCC, rCC, w0
@@ -69,7 +69,7 @@ index 444545c..031cee2 100644
  	ldp	x19, x20, [sp, #16*1]
  	ldp	x21, x22, [sp, #16*2]
  	ldp	x23, x24, [sp, #16*3]
-@@ -245,7 +245,7 @@ FUNCTION(new_dyna_leave):
+@@ -249,7 +249,7 @@ FUNCTION(new_dyna_leave):
  	/* w0 = adddr/data, x1 = rhandler, w2 = cycles, x3 = whandler */
  	ldr	w4, [rFP, #LO_last_count]
  	add	w4, w4, w2
@@ -79,24 +79,10 @@ index 444545c..031cee2 100644
  
  .macro memhandler_post
 diff --git a/libpcsxcore/new_dynarec/new_dynarec.c b/libpcsxcore/new_dynarec/new_dynarec.c
-index a1d7f6a..3960f3b 100644
+index 2f77516..21481bc 100644
 --- a/libpcsxcore/new_dynarec/new_dynarec.c
 +++ b/libpcsxcore/new_dynarec/new_dynarec.c
-@@ -44,10 +44,10 @@ static int sceBlock;
- #define ARRAY_SIZE(x) (sizeof(x) / sizeof(x[0]))
- #endif
- 
--//#define DISASM
--//#define assem_debug printf
-+#define DISASM
-+#define assem_debug printf
- //#define inv_debug printf
--#define assem_debug(...)
-+//#define assem_debug(...)
- #define inv_debug(...)
- 
- #ifdef __i386__
-@@ -489,6 +489,9 @@ static int doesnt_expire_soon(void *tcaddr)
+@@ -521,6 +521,9 @@ static int doesnt_expire_soon(void *tcaddr)
  // This is called from the recompiled JR/JALR instructions
  void noinline *get_addr(u_int vaddr)
  {
@@ -106,8 +92,8 @@ index a1d7f6a..3960f3b 100644
    u_int page=get_page(vaddr);
    u_int vpage=get_vpage(vaddr);
    struct ll_entry *head;
-@@ -4485,13 +4492,15 @@ void do_cc(int i,signed char i_regmap[],int *adj,int addr,int taken,int invert)
-     }
+@@ -4790,13 +4793,15 @@ void do_cc(int i,signed char i_regmap[],int *adj,int addr,int taken,int invert)
+ #endif
      emit_addimm_and_set_flags(cycles,HOST_CCREG);
      jaddr=out;
 -    emit_jns(0);
@@ -124,7 +110,7 @@ index a1d7f6a..3960f3b 100644
    }
    add_stub(CC_STUB,jaddr,idle?idle:out,(*adj==0||invert||idle)?0:(count+2),i,addr,taken,0);
  }
-@@ -4899,7 +4908,8 @@ static void rjump_assemble(int i,struct regstat *i_regs)
+@@ -5204,7 +5209,8 @@ static void rjump_assemble(int i,struct regstat *i_regs)
      // special case for RFE
      emit_jmp(0);
    else
@@ -133,8 +119,8 @@ index a1d7f6a..3960f3b 100644
 +   emit_jmp(0);
    //load_regs_bt(branch_regs[i].regmap,branch_regs[i].dirty,-1);
    #ifdef USE_MINI_HT
-   if(rs1[i]==31) {
-@@ -5004,7 +5014,8 @@ static void cjump_assemble(int i,struct regstat *i_regs)
+   if(dops[i].rs1==31) {
+@@ -5309,7 +5315,8 @@ static void cjump_assemble(int i,struct regstat *i_regs)
      else if(nop) {
        emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc);
        void *jaddr=out;
@@ -144,7 +130,7 @@ index a1d7f6a..3960f3b 100644
        add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
      }
      else {
-@@ -5191,7 +5202,8 @@ static void cjump_assemble(int i,struct regstat *i_regs)
+@@ -5496,7 +5503,8 @@ static void cjump_assemble(int i,struct regstat *i_regs)
          emit_loadreg(CCREG,HOST_CCREG);
          emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG);
          void *jaddr=out;
@@ -154,17 +140,17 @@ index a1d7f6a..3960f3b 100644
          add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
          emit_storereg(CCREG,HOST_CCREG);
        }
-@@ -5200,7 +5212,8 @@ static void cjump_assemble(int i,struct regstat *i_regs)
+@@ -5505,7 +5513,8 @@ static void cjump_assemble(int i,struct regstat *i_regs)
          assert(cc==HOST_CCREG);
          emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc);
          void *jaddr=out;
 -        emit_jns(0);
 +        //emit_jns(0);
 +       emit_jmp(0);
-         add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
+         add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
        }
      }
-@@ -5302,7 +5315,8 @@ static void sjump_assemble(int i,struct regstat *i_regs)
+@@ -5607,7 +5616,8 @@ static void sjump_assemble(int i,struct regstat *i_regs)
      else if(nevertaken) {
        emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc);
        void *jaddr=out;
@@ -174,7 +160,7 @@ index a1d7f6a..3960f3b 100644
        add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
      }
      else {
-@@ -5458,7 +5472,8 @@ static void sjump_assemble(int i,struct regstat *i_regs)
+@@ -5763,7 +5773,8 @@ static void sjump_assemble(int i,struct regstat *i_regs)
          emit_loadreg(CCREG,HOST_CCREG);
          emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),HOST_CCREG);
          void *jaddr=out;
@@ -184,17 +170,17 @@ index a1d7f6a..3960f3b 100644
          add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
          emit_storereg(CCREG,HOST_CCREG);
        }
-@@ -5467,7 +5482,8 @@ static void sjump_assemble(int i,struct regstat *i_regs)
+@@ -5772,7 +5783,8 @@ static void sjump_assemble(int i,struct regstat *i_regs)
          assert(cc==HOST_CCREG);
          emit_addimm_and_set_flags(CLOCK_ADJUST(ccadj[i]+2),cc);
          void *jaddr=out;
 -        emit_jns(0);
 +        //emit_jns(0);
 +       emit_jmp(0);
-         add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,likely[i]?NULLDS:NOTTAKEN,0);
+         add_stub(CC_STUB,jaddr,out,0,i,start+i*4+8,NOTTAKEN,0);
        }
      }
-@@ -5955,7 +5971,7 @@ void unneeded_registers(int istart,int iend,int r)
+@@ -6211,7 +6223,7 @@ void unneeded_registers(int istart,int iend,int r)
      // R0 is always unneeded
      u|=1;
      // Save it
@@ -203,15 +189,15 @@ index a1d7f6a..3960f3b 100644
      gte_unneeded[i]=gte_u;
      /*
      printf("ur (%d,%d) %x: ",istart,iend,start+i*4);
-@@ -8474,6 +8491,7 @@ int new_recompile_block(u_int addr)
+@@ -8756,6 +8768,7 @@ int new_recompile_block(u_int addr)
  
    // This allocates registers (if possible) one instruction prior
    // to use, which can avoid a load-use penalty on certain CPUs.
 +#if 0
    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))
-@@ -8630,6 +8648,7 @@ int new_recompile_block(u_int addr)
+     if (!i || !dops[i-1].is_jump)
+@@ -8912,6 +8925,7 @@ int new_recompile_block(u_int addr)
        }
      }
    }
@@ -219,36 +205,26 @@ index a1d7f6a..3960f3b 100644
  
    /* Pass 6 - Optimize clean/dirty state */
    clean_registers(0,slen-1,1);
-@@ -8929,6 +8948,12 @@ int new_recompile_block(u_int addr)
+@@ -9217,6 +9231,11 @@ int new_recompile_block(u_int addr)
          case SPAN:
            pagespan_assemble(i,&regs[i]);break;
        }
-+
 +#ifdef DRC_DBG
-+      // write-out non-consts, consts are likely different because of get_final_value()
-+      if(itype[i]!=UJUMP&&itype[i]!=CJUMP&&itype[i]!=SJUMP&&itype[i]!=RJUMP)
-+        wb_dirtys(regs[i].regmap,regs[i].dirty&~regs[i].loadedconst);
++       // write-out non-consts, consts are likely different because of get_final_value()
++       if (!dops[i].is_jump)
++         wb_dirtys(regs[i].regmap,regs[i].dirty&~regs[i].loadedconst);
 +#endif
-       if(itype[i]==UJUMP||itype[i]==RJUMP||(source[i]>>16)==0x1000)
+       if (dops[i].is_ujump)
          literal_pool(1024);
        else
-@@ -9037,7 +9062,7 @@ int new_recompile_block(u_int addr)
-     }
-   }
-   // External Branch Targets (jump_in)
--  if(copy+slen*4>(void *)shadow+sizeof(shadow)) copy=shadow;
-+  if(copy+slen*4>(void *)shadow+sizeof(shadow)) {copy=shadow;printf("shadow overflow\n");}
-   for(i=0;i<slen;i++)
-   {
-     if(bt[i]||i==0)
-@@ -9150,6 +9175,10 @@ int new_recompile_block(u_int addr)
-     }
-     expirep=(expirep+1)&65535;
+@@ -9451,6 +9470,10 @@ int new_recompile_block(u_int addr)
    }
+ #ifdef ASSEM_PRINT
+   fflush(stdout);
++#endif
 +#ifdef DRC_DBG
 +printf("new_recompile_block done\n");
 +fflush(stdout);
-+#endif
+ #endif
    return 0;
  }
- 
diff --git a/libpcsxcore/new_dynarec/patches/trace_intr b/libpcsxcore/new_dynarec/patches/trace_intr
index 9538bd08..76521872 100644
--- a/libpcsxcore/new_dynarec/patches/trace_intr
+++ b/libpcsxcore/new_dynarec/patches/trace_intr
@@ -1,8 +1,8 @@
 diff --git a/libpcsxcore/new_dynarec/emu_if.c b/libpcsxcore/new_dynarec/emu_if.c
-index 2df259b..2a15e6e 100644
+index 8c96504..1648b8f 100644
 --- a/libpcsxcore/new_dynarec/emu_if.c
 +++ b/libpcsxcore/new_dynarec/emu_if.c
-@@ -417,13 +417,17 @@ static void ari64_shutdown()
+@@ -424,13 +424,17 @@ static void ari64_shutdown()
  {
  	new_dynarec_cleanup();
  	new_dyna_pcsx_mem_shutdown();
@@ -20,9 +20,9 @@ index 2df259b..2a15e6e 100644
 +	intExecuteT,
 +	intExecuteBlockT,
  	ari64_clear,
- #ifdef ICACHE_EMULATION
  	ari64_notify,
-@@ -489,7 +493,7 @@ static u32 memcheck_read(u32 a)
+ 	ari64_apply_config,
+@@ -499,7 +503,7 @@ static u32 memcheck_read(u32 a)
  	return *(u32 *)(psxM + (a & 0x1ffffc));
  }
  
@@ -80,10 +80,10 @@ index dbcb989..0716f5e 100644
  #ifdef PSXHW_LOG
  			PSXHW_LOG("*Unknown 32bit write at address %x value %x\n", add, value);
 diff --git a/libpcsxcore/psxinterpreter.c b/libpcsxcore/psxinterpreter.c
-index 61c60ed..0fa5283 100644
+index f7898e9..176a0f7 100644
 --- a/libpcsxcore/psxinterpreter.c
 +++ b/libpcsxcore/psxinterpreter.c
-@@ -511,8 +511,9 @@ static void doBranch(u32 tar) {
+@@ -464,8 +464,9 @@ static void doBranch(u32 tar) {
  	debugI();
  
  	psxRegs.pc += 4;
@@ -94,7 +94,7 @@ index 61c60ed..0fa5283 100644
  	// check for load delay
  	tmp = psxRegs.code >> 26;
  	switch (tmp) {
-@@ -546,13 +547,15 @@ static void doBranch(u32 tar) {
+@@ -499,13 +500,15 @@ static void doBranch(u32 tar) {
  			}
  			break;
  	}
@@ -111,8 +111,8 @@ index 61c60ed..0fa5283 100644
  }
  
  /*********************************************************
-@@ -635,12 +638,13 @@ void psxMULTU() {
- 	psxRegs.GPR.n.hi = (u32)((res >> 32) & 0xffffffff);
+@@ -615,12 +618,13 @@ void psxMULTU_stall() {
+ 	psxMULTU();
  }
  
 +#define doBranchNotTaken() do { psxRegs.cycle -= BIAS; execI(); psxBranchTest(); psxRegs.cycle += BIAS; } while(0)
@@ -127,7 +127,7 @@ index 61c60ed..0fa5283 100644
  
  void psxBGEZ()   { RepZBranchi32(>=) }      // Branch if Rs >= 0
  void psxBGEZAL() { RepZBranchLinki32(>=) }  // Branch if Rs >= 0 and link
-@@ -710,7 +714,7 @@ void psxRFE() {
+@@ -702,7 +706,7 @@ void psxRFE() {
  * Register branch logic                                  *
  * Format:  OP rs, rt, offset                             *
  *********************************************************/
@@ -136,7 +136,7 @@ index 61c60ed..0fa5283 100644
  
  void psxBEQ() {	RepBranchi32(==) }  // Branch if Rs == Rt
  void psxBNE() {	RepBranchi32(!=) }  // Branch if Rs != Rt
-@@ -894,6 +898,9 @@ void MTC0(int reg, u32 val) {
+@@ -886,6 +890,9 @@ void MTC0(int reg, u32 val) {
  		case 12: // Status
  			psxRegs.CP0.r[12] = val;
  			psxTestSWInts();
@@ -146,7 +146,7 @@ index 61c60ed..0fa5283 100644
  			break;
  
  		case 13: // Cause
-@@ -1056,6 +1063,23 @@ void intExecuteBlock() {
+@@ -1027,6 +1034,23 @@ void intExecuteBlock() {
  	while (!branch2) execI();
  }
  
@@ -171,10 +171,10 @@ index 61c60ed..0fa5283 100644
  }
  
 diff --git a/libpcsxcore/psxmem.c b/libpcsxcore/psxmem.c
-index c09965d..135a5d0 100644
+index 04aeec2..1242653 100644
 --- a/libpcsxcore/psxmem.c
 +++ b/libpcsxcore/psxmem.c
-@@ -219,11 +219,13 @@ void psxMemShutdown() {
+@@ -217,11 +217,13 @@ void psxMemShutdown() {
  }
  
  static int writeok = 1;
@@ -188,7 +188,7 @@ index c09965d..135a5d0 100644
  	t = mem >> 16;
  	if (t == 0x1f80 || t == 0x9f80 || t == 0xbf80) {
  		if ((mem & 0xffff) < 0x400)
-@@ -249,6 +251,7 @@ u16 psxMemRead16(u32 mem) {
+@@ -247,6 +249,7 @@ u16 psxMemRead16(u32 mem) {
  	char *p;
  	u32 t;
  
@@ -196,7 +196,7 @@ index c09965d..135a5d0 100644
  	t = mem >> 16;
  	if (t == 0x1f80 || t == 0x9f80 || t == 0xbf80) {
  		if ((mem & 0xffff) < 0x400)
-@@ -274,6 +277,7 @@ u32 psxMemRead32(u32 mem) {
+@@ -272,6 +275,7 @@ u32 psxMemRead32(u32 mem) {
  	char *p;
  	u32 t;
  
@@ -204,7 +204,7 @@ index c09965d..135a5d0 100644
  	t = mem >> 16;
  	if (t == 0x1f80 || t == 0x9f80 || t == 0xbf80) {
  		if ((mem & 0xffff) < 0x400)
-@@ -299,6 +303,7 @@ void psxMemWrite8(u32 mem, u8 value) {
+@@ -297,6 +301,7 @@ void psxMemWrite8(u32 mem, u8 value) {
  	char *p;
  	u32 t;
  
@@ -212,7 +212,7 @@ index c09965d..135a5d0 100644
  	t = mem >> 16;
  	if (t == 0x1f80 || t == 0x9f80 || t == 0xbf80) {
  		if ((mem & 0xffff) < 0x400)
-@@ -326,6 +331,7 @@ void psxMemWrite16(u32 mem, u16 value) {
+@@ -324,6 +329,7 @@ void psxMemWrite16(u32 mem, u16 value) {
  	char *p;
  	u32 t;
  
@@ -220,7 +220,7 @@ index c09965d..135a5d0 100644
  	t = mem >> 16;
  	if (t == 0x1f80 || t == 0x9f80 || t == 0xbf80) {
  		if ((mem & 0xffff) < 0x400)
-@@ -353,6 +359,7 @@ void psxMemWrite32(u32 mem, u32 value) {
+@@ -351,6 +357,7 @@ void psxMemWrite32(u32 mem, u32 value) {
  	char *p;
  	u32 t;
  
diff --git a/libpcsxcore/psxmem.c b/libpcsxcore/psxmem.c
index 52d27593..04aeec27 100644
--- a/libpcsxcore/psxmem.c
+++ b/libpcsxcore/psxmem.c
@@ -136,10 +136,8 @@ int psxMemInit() {
 	memset(psxMemWLUT, 0, 0x10000 * sizeof(void *));
 
 	psxM = psxMap(0x80000000, 0x00210000, 1, MAP_TAG_RAM);
-#ifndef RAM_FIXED
 	if (psxM == NULL)
 		psxM = psxMap(0x77000000, 0x00210000, 0, MAP_TAG_RAM);
-#endif
 	if (psxM == NULL) {
 		SysMessage(_("mapping main RAM failed"));
 		return -1;