diff --git a/bsnes.cfg b/bsnes.cfg
index 48be699b..78af5efa 100644
--- a/bsnes.cfg
+++ b/bsnes.cfg
@@ -14,7 +14,11 @@ snes.video_color_adjust_mode = 0
 
 # Mutes SNES audio output when enabled
 # (default = true)
-snes.mute = true
+snes.mute = false
+
+# Regulate speed to 60hz (NTSC) / 50hz (PAL)
+# (default = true)
+system.regulate_speed = true
 
 # Video mode
 #   0 = 256x224w
diff --git a/bsnes.exe b/bsnes.exe
index 1fe6a58a..8514825d 100644
Binary files a/bsnes.exe and b/bsnes.exe differ
diff --git a/src/apu/bapu/bapu.cpp b/src/apu/bapu/bapu.cpp
index 7395bce2..44501483 100644
--- a/src/apu/bapu/bapu.cpp
+++ b/src/apu/bapu/bapu.cpp
@@ -24,13 +24,13 @@ uint8 r;
       break;
     case 0xf3: //DSPDATA
     //0x80-0xff is a read-only mirror of 0x00-0x7f
-      r = dsp->read(status.dsp_addr & 0x7f);
+      r = r_dsp->read(status.dsp_addr & 0x7f);
       break;
     case 0xf4: //CPUIO0
     case 0xf5: //CPUIO1
     case 0xf6: //CPUIO2
     case 0xf7: //CPUIO3
-      r = cpu->port_read(addr & 3);
+      r = r_cpu->port_read(addr & 3);
       break;
     case 0xf8: //???
     case 0xf9: //??? -- Mapped to SPCRAM
@@ -63,7 +63,10 @@ uint8 r;
       r = spcram[addr];
     }
   }
+
+#ifdef DEBUGGER
   snes->notify(SNES::SPCRAM_READ, addr, r);
+#endif
   return r;
 }
 
@@ -78,12 +81,12 @@ void bAPU::spcram_write(uint16 addr, uint8 value) {
     //one-time clearing of APU port read registers,
     //emulated by simulating CPU writes of 0x00
       if(value & 0x20) {
-        cpu->port_write(2, 0x00);
-        cpu->port_write(3, 0x00);
+        r_cpu->port_write(2, 0x00);
+        r_cpu->port_write(3, 0x00);
       }
       if(value & 0x10) {
-        cpu->port_write(0, 0x00);
-        cpu->port_write(1, 0x00);
+        r_cpu->port_write(0, 0x00);
+        r_cpu->port_write(1, 0x00);
       }
 
     //0->1 transistion resets timers
@@ -111,7 +114,7 @@ void bAPU::spcram_write(uint16 addr, uint8 value) {
     case 0xf3: //DSPDATA
     //0x80-0xff is a read-only mirror of 0x00-0x7f
       if(status.dsp_addr < 0x80) {
-        dsp->write(status.dsp_addr & 0x7f, value);
+        r_dsp->write(status.dsp_addr & 0x7f, value);
       }
       break;
     case 0xf4: //CPUIO0
@@ -143,7 +146,10 @@ void bAPU::spcram_write(uint16 addr, uint8 value) {
   //even if the iplrom is enabled.
     spcram[addr] = value;
   }
+
+#ifdef DEBUGGER
   snes->notify(SNES::SPCRAM_WRITE, addr, value);
+#endif
 }
 
 uint8 bAPU::port_read(uint8 port) {
diff --git a/src/apu/bapu/bapu_exec.cpp b/src/apu/bapu/bapu_exec.cpp
index 37235466..25c92029 100644
--- a/src/apu/bapu/bapu_exec.cpp
+++ b/src/apu/bapu/bapu_exec.cpp
@@ -1,18 +1,22 @@
 void bAPU::exec_cycle() {
-uint8 op;
-  if(status.cycle_pos == 0) {
-    op = spcram_read(regs.pc);
-    snes->notify(SNES::APU_EXEC_OPCODE_BEGIN);
-    status.opcode = op_read();
-    status.cycle_pos = 1;
-    add_cycles(1);
-  } else {
+  if(status.cycle_pos) {
     (this->*optbl[status.opcode])();
     add_cycles(1);
     if(status.cycle_pos == 0) {
+    #ifdef DEBUGGER
       snes->notify(SNES::APU_EXEC_OPCODE_END);
+    #endif
     }
+    return;
   }
+
+//on first cycle?
+#ifdef DEBUGGER
+  snes->notify(SNES::APU_EXEC_OPCODE_BEGIN);
+#endif
+  status.opcode = op_read();
+  status.cycle_pos = 1;
+  add_cycles(1);
 }
 
 //only return true when we are on an opcode edge
diff --git a/src/apu/bapuskip/bapuskip.cpp b/src/apu/bapuskip/bapuskip.cpp
deleted file mode 100644
index 597a93bd..00000000
--- a/src/apu/bapuskip/bapuskip.cpp
+++ /dev/null
@@ -1,171 +0,0 @@
-#include "../../base.h"
-
-uint8 bAPUSkip::spcram_read (uint16 addr) { return 0xff; }
-void  bAPUSkip::spcram_write(uint16 addr, uint8 value) {}
-
-/*
-  This routine is very serious. It will eat holes through
-  the ROM to skip APU test conditions. Or in other words,
-  it will disable and/or force branches when neccesary.
-
-  It can very easily break or corrupt a game and prevent it
-  from being playable until the ROM is reloaded (ROM writes
-  are only performed in memory, of course).
-
-  However, this kind of brute force approach is required to
-  get many games playable without proper SPC700 emulation.
-*/
-uint8 bAPUSkip::port_read(uint8 port) {
-  port &= 3;
-_apu_port *p = &apu_port[port];
-int i, x, y, z, t;
-uint32 addr;
-
-  addr = cpu->regs.pc.d;
-  p->read_addr[p->read_pos & 31] = addr;
-
-//- lda $214x
-//  cmp $214x
-//  bne -
-//  cmp ???
-//  beq/bne -
-__test1:
-//look for an lda/cmp read pattern
-  if(addr == p->read_addr[(p->read_pos - 1) & 31])goto __test2;
-  if(addr != p->read_addr[(p->read_pos - 2) & 31])goto __test2;
-  if(addr == p->read_addr[(p->read_pos - 3) & 31])goto __test2;
-  if(addr != p->read_addr[(p->read_pos - 4) & 31])goto __test2;
-  if(p->read_addr[(p->read_pos - 1) & 31] != p->read_addr[(p->read_pos - 3) & 31])goto __test2;
-//try and find compare opcode
-  for(i=0;i<24;i++) {
-    x = mem_bus->read(addr + i);
-    if(x == OP_CMP_CONST || x == OP_CPX_CONST || x == OP_CPY_CONST)break;
-    if(x == OP_CMP_ADDR  || x == OP_CPX_ADDR  || x == OP_CPY_ADDR) break;
-    if(x == OP_CMP_LONG)break;
-  }
-  if(i == 24)goto __test2;
-//seek to next opcode
-  if(x == OP_CMP_CONST) {
-    i += (cpu->regs.p.m)?2:3;
-  } else if(x == OP_CPX_CONST || x == OP_CPY_CONST) {
-    i += (cpu->regs.p.x)?2:3;
-  } else if(x == OP_CMP_ADDR || x == OP_CPX_ADDR || x == OP_CPY_ADDR) {
-    i += 3;
-  } else { //(x == OP_CMP_LONG) {
-    i += 4;
-  }
-  x = mem_bus->read(addr + i);
-  if(x == OP_BNE) {
-    mem_bus->cart->write_protect(false);
-    mem_bus->write(addr + i,     OP_NOP);
-    mem_bus->write(addr + i + 1, OP_NOP);
-    mem_bus->cart->write_protect(true);
-  } else if(x == OP_BEQ) {
-    mem_bus->cart->write_protect(false);
-    mem_bus->write(addr + i, OP_BRA);
-    mem_bus->cart->write_protect(true);
-  } else goto __test2;
-  goto __pass;
-
-//- lda $214x
-//  cmp ???
-//  beq/bne -
-__test2:
-//look for a repeated read pattern
-  if(addr != p->read_addr[(p->read_pos - 1) & 31])goto __test3;
-  if(addr != p->read_addr[(p->read_pos - 2) & 31])goto __test3;
-  if(addr != p->read_addr[(p->read_pos - 3) & 31])goto __test3;
-  if(addr != p->read_addr[(p->read_pos - 4) & 31])goto __test3;
-  if(addr != p->read_addr[(p->read_pos - 5) & 31])goto __test3;
-  if(addr != p->read_addr[(p->read_pos - 6) & 31])goto __test3;
-  if(addr != p->read_addr[(p->read_pos - 7) & 31])goto __test3;
-  if(addr != p->read_addr[(p->read_pos - 8) & 31])goto __test3;
-//try and find compare opcode
-  for(i=0;i<24;i++) {
-    x = mem_bus->read(addr + i);
-    if(x == OP_CMP_CONST || x == OP_CPX_CONST || x == OP_CPY_CONST)break;
-    if(x == OP_CMP_ADDR  || x == OP_CPX_ADDR  || x == OP_CPY_ADDR) break;
-    if(x == OP_CMP_LONG)break;
-  }
-  if(i == 24)goto __test3;
-//seek to next opcode
-  if(x == OP_CMP_CONST) {
-    i += (cpu->regs.p.m)?2:3;
-  } else if(x == OP_CPX_CONST || x == OP_CPY_CONST) {
-    i += (cpu->regs.p.x)?2:3;
-  } else if(x == OP_CMP_ADDR || x == OP_CPX_ADDR || x == OP_CPY_ADDR) {
-    i += 3;
-  } else if(x == OP_CMP_LONG) {
-    i += 4;
-  }
-  x = mem_bus->read(addr + i);
-  if(x == OP_BNE) {
-    mem_bus->cart->write_protect(false);
-    mem_bus->write(addr + i,     OP_NOP);
-    mem_bus->write(addr + i + 1, OP_NOP);
-    mem_bus->cart->write_protect(true);
-  } else if(x == OP_BEQ) {
-    mem_bus->cart->write_protect(false);
-    mem_bus->write(addr + i, OP_BRA);
-    mem_bus->cart->write_protect(true);
-  } else goto __test3;
-  goto __pass;
-
-//fallback
-__test3:
-  if(p->pos < 4) {
-    if(!(port & 1)) {
-      p->value = cpu->regs.a.l;
-    } else {
-      p->value = cpu->regs.a.h;
-    }
-  } else if(p->pos < 8) {
-    if(!(port & 1)) {
-      p->value = cpu->regs.x.l;
-    } else {
-      p->value = cpu->regs.x.h;
-    }
-  } else if(p->pos < 12) {
-    if(!(port & 1)) {
-      p->value = cpu->regs.y.l;
-    } else {
-      p->value = cpu->regs.y.h;
-    }
-  } else if(p->pos < 16) {
-    p->value = rand();
-  }
-  if(++p->pos == 16)p->pos = 0;
-
-__pass:
-  p->read_pos++;
-  p->read_pos &= 31;
-  return p->value;
-}
-
-void bAPUSkip::port_write(uint8 port, uint8 value) {
-  port &= 3;
-  apu_port[port].value = value;
-}
-
-void bAPUSkip::run() {
-  snes->notify(SNES::APU_EXEC_OPCODE_BEGIN);
-  snes->notify(SNES::APU_EXEC_OPCODE_END);
-}
-
-void bAPUSkip::power() {
-  reset();
-}
-
-void bAPUSkip::reset() {
-  regs.a  = 0x00;
-  regs.x  = 0x00;
-  regs.y  = 0x00;
-  regs.sp = 0x00;
-  regs.p  = 0x02;
-  regs.pc = 0xffc0;
-
-  memset(&apu_port[0], 0, sizeof(_apu_port));
-  memset(&apu_port[1], 0, sizeof(_apu_port));
-  memset(&apu_port[2], 0, sizeof(_apu_port));
-  memset(&apu_port[3], 0, sizeof(_apu_port));
-}
diff --git a/src/apu/bapuskip/bapuskip.h b/src/apu/bapuskip/bapuskip.h
deleted file mode 100644
index 63b1da50..00000000
--- a/src/apu/bapuskip/bapuskip.h
+++ /dev/null
@@ -1,37 +0,0 @@
-class bAPUSkip : public APU {
-private:
-uint8 spcram[65536];
-
-public:
-
-struct _apu_port {
-  uint8  value;
-  uint8  step, pos;
-  uint32 read_addr[32], read_pos;
-}apu_port[4];
-
-enum {
-  OP_CMP_CONST = 0xc9,
-  OP_CPX_CONST = 0xe0,
-  OP_CPY_CONST = 0xc0,
-  OP_CMP_ADDR  = 0xcd,
-  OP_CPX_ADDR  = 0xec,
-  OP_CPY_ADDR  = 0xcc,
-  OP_CMP_LONG  = 0xcf,
-  OP_BNE       = 0xd0,
-  OP_BEQ       = 0xf0,
-  OP_BRA       = 0x80,
-  OP_NOP       = 0xea
-};
-
-  uint8  spcram_read (uint16 addr);
-  void   spcram_write(uint16 addr, uint8 value);
-  uint8  port_read   (uint8 port);
-  void   port_write  (uint8 port, uint8 value);
-
-  uint8 *get_spcram_handle() { return spcram; }
-  void   run();
-  uint32 cycles_executed() { return 12 * 24; }
-  void   power();
-  void   reset();
-};
diff --git a/src/base.h b/src/base.h
index 3141c3a0..7e00bb59 100644
--- a/src/base.h
+++ b/src/base.h
@@ -1,3 +1,6 @@
+//#define DEBUGGER
+//#define POLYMORPHISM
+
 //this should be declared in the port-specific makefiles
 //#define ARCH_LSB
 //#define ARCH_MSB
diff --git a/src/chip/sdd1/sdd1.cpp b/src/chip/sdd1/sdd1.cpp
index c26694bb..e7e16098 100644
--- a/src/chip/sdd1/sdd1.cpp
+++ b/src/chip/sdd1/sdd1.cpp
@@ -6,7 +6,7 @@ void SDD1::init() {
 
 void SDD1::enable() {
   for(int i=0x4800;i<=0x4807;i++) {
-    mem_bus->set_mmio_mapper(i, mmio);
+    r_mem->set_mmio_mapper(i, mmio);
   }
 }
 
@@ -47,7 +47,7 @@ uint8 SDD1::mmio_read(uint16 addr) {
   case 0x4807:return (sdd1.index[3] >> 20) & 7;
   }
 
-  return cpu->regs.mdr;
+  return r_cpu->regs.mdr;
 }
 
 void SDD1::mmio_write(uint16 addr, uint8 data) {
diff --git a/src/chip/sdd1/sdd1emu.cpp b/src/chip/sdd1/sdd1emu.cpp
index c98506ee..fdb6aaca 100644
--- a/src/chip/sdd1/sdd1emu.cpp
+++ b/src/chip/sdd1/sdd1emu.cpp
@@ -28,7 +28,7 @@ understood.
 
 ************************************************************************/
 
-#define SDD1_read(__addr) (mem_bus->read(__addr))
+#define SDD1_read(__addr) (r_mem->read(__addr))
 
 ////////////////////////////////////////////////////
 
diff --git a/src/chip/srtc/srtc.cpp b/src/chip/srtc/srtc.cpp
index 590037ab..aa59e006 100644
--- a/src/chip/srtc/srtc.cpp
+++ b/src/chip/srtc/srtc.cpp
@@ -78,8 +78,8 @@ void SRTC::init() {
 }
 
 void SRTC::enable() {
-  mem_bus->set_mmio_mapper(0x2800, mmio);
-  mem_bus->set_mmio_mapper(0x2801, mmio);
+  r_mem->set_mmio_mapper(0x2800, mmio);
+  r_mem->set_mmio_mapper(0x2801, mmio);
 }
 
 void SRTC::power() {
@@ -176,7 +176,7 @@ uint8 SRTCMMIO::read(uint32 addr) {
   case 0x2800:return srtc->read();
   }
 
-  return cpu->regs.mdr;
+  return r_cpu->regs.mdr;
 }
 
 void SRTCMMIO::write(uint32 addr, uint8 value) {
diff --git a/src/cpu/bcpu/bcpu.cpp b/src/cpu/bcpu/bcpu.cpp
index 3d92c2aa..f6706b49 100644
--- a/src/cpu/bcpu/bcpu.cpp
+++ b/src/cpu/bcpu/bcpu.cpp
@@ -85,7 +85,7 @@ void bCPU::power() {
 
 void bCPU::reset() {
 //reset vector location
-  regs.pc  = mem_bus->read(0xfffc) | (mem_bus->read(0xfffd) << 8);
+  regs.pc  = r_mem->read(0xfffc) | (r_mem->read(0xfffd) << 8);
 
 //registers are not fully reset by SNES
   regs.x.h = 0x00;
@@ -161,22 +161,25 @@ void bCPU::cpu_io() {
   status.cycle_count = 6;
   pre_exec_cycle();
   add_cycles(6);
+  cycle_edge();
 }
 
 uint8 bCPU::mem_read(uint32 addr) {
-  status.cycle_count = mem_bus->speed(addr);
+  status.cycle_count = r_mem->speed(addr);
   pre_exec_cycle();
   add_cycles(status.cycle_count - 4);
-  regs.mdr = mem_bus->read(addr);
+  regs.mdr = r_mem->read(addr);
   add_cycles(4);
+  cycle_edge();
   return regs.mdr;
 }
 
 void bCPU::mem_write(uint32 addr, uint8 value) {
-  status.cycle_count = mem_bus->speed(addr);
+  status.cycle_count = r_mem->speed(addr);
   pre_exec_cycle();
   add_cycles(status.cycle_count);
-  mem_bus->write(addr, value);
+  r_mem->write(addr, value);
+  cycle_edge();
 }
 
 uint32 bCPU::op_addr(uint8 mode, uint32 addr) {
diff --git a/src/cpu/bcpu/bcpu_dma.cpp b/src/cpu/bcpu/bcpu_dma.cpp
index dc9ccbc0..2fad96eb 100644
--- a/src/cpu/bcpu/bcpu_dma.cpp
+++ b/src/cpu/bcpu/bcpu_dma.cpp
@@ -25,10 +25,10 @@ uint8 x;
   if(sdd1->dma_active() == true) {
     x = sdd1->dma_read();
   } else {
-    x = mem_bus->read(dma_addr(i));
+    x = r_mem->read(dma_addr(i));
   }
 
-  mem_bus->write(0x2100 | ((channel[i].destaddr + index) & 0xff), x);
+  r_mem->write(0x2100 | ((channel[i].destaddr + index) & 0xff), x);
 
   add_cycles(8);
   channel[i].xfersize--;
@@ -36,8 +36,8 @@ uint8 x;
 
 void bCPU::dma_mmiotocpu(uint8 i, uint8 index) {
 uint8 x;
-  x = mem_bus->read(0x2100 | ((channel[i].destaddr + index) & 0xff));
-  mem_bus->write(dma_addr(i), x);
+  x = r_mem->read(0x2100 | ((channel[i].destaddr + index) & 0xff));
+  r_mem->write(dma_addr(i), x);
 
   add_cycles(8);
   channel[i].xfersize--;
@@ -113,21 +113,21 @@ uint16 index;
 
 uint8 bCPU::hdma_read(uint8 i) {
   if(channel[i].direction == DMA_MMIOTOCPU) {
-    return mem_bus->read(hdma_mmio(i));
+    return r_mem->read(hdma_mmio(i));
   } else if(!channel[i].hdma_indirect) {
-    return mem_bus->read(hdma_addr(i));
+    return r_mem->read(hdma_addr(i));
   } else {
-    return mem_bus->read(hdma_iaddr(i));
+    return r_mem->read(hdma_iaddr(i));
   }
 }
 
 void bCPU::hdma_write(uint8 i, uint8 x) {
   if(channel[i].direction == DMA_CPUTOMMIO) {
-    mem_bus->write(hdma_mmio(i), x);
+    r_mem->write(hdma_mmio(i), x);
   } else if(!channel[i].hdma_indirect) {
-    mem_bus->write(hdma_addr(i), x);
+    r_mem->write(hdma_addr(i), x);
   } else {
-    mem_bus->write(hdma_iaddr(i), x);
+    r_mem->write(hdma_iaddr(i), x);
   }
 
   add_cycles(8);
@@ -135,12 +135,12 @@ void bCPU::hdma_write(uint8 i, uint8 x) {
 }
 
 void bCPU::hdma_update(uint8 i) {
-  channel[i].hdma_line_counter = mem_bus->read(hdma_addr(i));
+  channel[i].hdma_line_counter = r_mem->read(hdma_addr(i));
   add_cycles(8);
   hdma_add_cycles(8);
 
   if(channel[i].hdma_indirect) {
-    channel[i].hdma_iaddr = mem_bus->read(hdma_addr(i)) << 8;
+    channel[i].hdma_iaddr = r_mem->read(hdma_addr(i)) << 8;
     add_cycles(8);
     hdma_add_cycles(8);
   }
@@ -155,7 +155,7 @@ void bCPU::hdma_update(uint8 i) {
 
   if(channel[i].hdma_indirect) {
     channel[i].hdma_iaddr >>= 8;
-    channel[i].hdma_iaddr  |= mem_bus->read(hdma_addr(i)) << 8;
+    channel[i].hdma_iaddr  |= r_mem->read(hdma_addr(i)) << 8;
     add_cycles(8);
     hdma_add_cycles(8);
   }
diff --git a/src/cpu/bcpu/bcpu_exec.cpp b/src/cpu/bcpu/bcpu_exec.cpp
index e0eb9b40..6a961442 100644
--- a/src/cpu/bcpu/bcpu_exec.cpp
+++ b/src/cpu/bcpu/bcpu_exec.cpp
@@ -133,6 +133,7 @@ static int z;
     break;
   case DMASTATE_RUN:
     dma_run(); //updates status.dma_cycle_count
+    cycle_edge();
     break;
   case DMASTATE_CPUSYNC:
     exec_cycle();
@@ -147,18 +148,22 @@ void bCPU::exec_cycle() {
     return;
   }
 
-//on first cycle?
-  if(status.cycle_pos == 0) {
-    snes->notify(SNES::CPU_EXEC_OPCODE_BEGIN);
-    status.opcode = op_read();
-    status.cycle_pos = 1;
+  if(status.cycle_pos) {
+    (this->*optbl[status.opcode])();
+  #ifdef DEBUGGER
+    if(status.cycle_pos == 0) {
+      snes->notify(SNES::CPU_EXEC_OPCODE_END);
+    }
+  #endif
     return;
   }
 
-  (this->*optbl[status.opcode])();
-  if(status.cycle_pos == 0) {
-    snes->notify(SNES::CPU_EXEC_OPCODE_END);
-  }
+//on first cycle?
+#ifdef DEBUGGER
+  snes->notify(SNES::CPU_EXEC_OPCODE_BEGIN);
+#endif
+  status.opcode = op_read();
+  status.cycle_pos = 1;
 }
 
 //only return true when we are on an opcode edge
diff --git a/src/cpu/bcpu/bcpu_int.cpp b/src/cpu/bcpu/bcpu_int.cpp
index 9f47c88d..88336745 100644
--- a/src/cpu/bcpu/bcpu_int.cpp
+++ b/src/cpu/bcpu/bcpu_int.cpp
@@ -13,7 +13,7 @@ void bCPU::irq_run() {
 //WDC documentation is incorrect, first cycle
 //is a memory read fetch from PBR:PC
   switch(status.cycle_pos++) {
-  case 0: add_cycles(mem_bus->speed(regs.pc.d)); break;
+  case 0: add_cycles(r_mem->speed(regs.pc.d));   break;
   case 1: add_cycles(6);                         break;
   case 2: stack_write(regs.pc.b);                break;
   case 3: stack_write(regs.pc.h);                break;
@@ -25,8 +25,10 @@ void bCPU::irq_run() {
     regs.pc.w = rd.w;
     regs.p.i  = 1;
     regs.p.d  = 0;
+  #ifdef DEBUGGER
   //let debugger know the new IRQ opcode address
     snes->notify(SNES::CPU_EXEC_OPCODE_END);
+  #endif
     status.cycle_pos = 0;
     run_state.irq = false;
     break;
diff --git a/src/cpu/bcpu/bcpu_mmio.cpp b/src/cpu/bcpu/bcpu_mmio.cpp
index fbcb0349..8d833cbe 100644
--- a/src/cpu/bcpu/bcpu_mmio.cpp
+++ b/src/cpu/bcpu/bcpu_mmio.cpp
@@ -37,7 +37,7 @@ void bCPU::mmio_reset() {
 //WMDATA
 uint8 bCPU::mmio_r2180() {
 uint8 r;
-  r = mem_bus->read(0x7e0000 | status.wram_addr);
+  r = r_mem->read(0x7e0000 | status.wram_addr);
   status.wram_addr++;
   status.wram_addr &= 0x01ffff;
   return r;
@@ -324,7 +324,7 @@ uint8 bCPUMMIO::read(uint32 addr) {
 uint i;
 //APU
   if(addr >= 0x2140 && addr <= 0x217f) {
-    return apu->port_read(addr & 3);
+    return r_apu->port_read(addr & 3);
   }
 
 //HDMA
@@ -374,7 +374,7 @@ uint i;
 
 //WMDATA
 void bCPU::mmio_w2180(uint8 value) {
-  mem_bus->write(0x7e0000 | status.wram_addr, value);
+  r_mem->write(0x7e0000 | status.wram_addr, value);
   status.wram_addr++;
   status.wram_addr &= 0x01ffff;
 }
@@ -441,7 +441,7 @@ void bCPU::mmio_w4200(uint8 value) {
 //WRIO
 void bCPU::mmio_w4201(uint8 value) {
   if((status.pio & 0x80) && !(value & 0x80)) {
-    ppu->latch_counters();
+    r_ppu->latch_counters();
   }
   status.pio = value;
 }
@@ -525,7 +525,7 @@ void bCPU::mmio_w420c(uint8 value) {
 
 //MEMSEL
 void bCPU::mmio_w420d(uint8 value) {
-  mem_bus->set_speed(value & 1);
+  r_mem->set_speed(value & 1);
 }
 
 //DMAPx
diff --git a/src/cpu/bcpu/bcpu_timing.cpp b/src/cpu/bcpu/bcpu_timing.cpp
index 93eed40e..2ca28391 100644
--- a/src/cpu/bcpu/bcpu_timing.cpp
+++ b/src/cpu/bcpu/bcpu_timing.cpp
@@ -239,6 +239,24 @@ uint32 r = status.cycles_executed;
   return r;
 }
 
+void bCPU::cycle_edge() {
+  if(time.hdmainit_triggered == false) {
+    if(time.hc >= time.hdmainit_trigger_pos || time.v) {
+      time.hdmainit_triggered = true;
+      hdmainit_activate();
+    }
+  }
+
+  if(time.hdma_triggered == false) {
+    if(time.v <= (overscan() ? 239 : 224)) {
+      if(time.hc >= 1106) {
+        time.hdma_triggered = true;
+        hdma_activate();
+      }
+    }
+  }
+}
+
 void bCPU::add_cycles(int cycles) {
   status.cycles_executed += cycles;
 
@@ -253,23 +271,16 @@ void bCPU::add_cycles(int cycles) {
 
     if(time.v == 0) {
       frame();
-      ppu->frame();
+      r_ppu->frame();
       snes->frame();
     }
 
     scanline();
-    ppu->scanline();
+    r_ppu->scanline();
     snes->scanline();
     time.line_rendered = false;
   }
 
-  if(time.hdmainit_triggered == false) {
-    if(time.hc + cycles >= time.hdmainit_trigger_pos || time.v) {
-      time.hdmainit_triggered = true;
-      hdmainit_activate();
-    }
-  }
-
   if(time.dram_refreshed == false) {
     if(time.hc + cycles >= time.dram_refresh_pos) {
       time.dram_refreshed = true;
@@ -296,16 +307,7 @@ void bCPU::add_cycles(int cycles) {
   //therefore, wait a few dots before rendering the scanline
     if(time.hc + cycles >= (48 * 4)) {
       time.line_rendered = true;
-      ppu->render_scanline();
-    }
-  }
-
-  if(time.hdma_triggered == false) {
-    if(time.v <= (overscan() ? 239 : 224)) {
-      if(time.hc + cycles >= 1106) {
-        time.hdma_triggered = true;
-        hdma_activate();
-      }
+      r_ppu->render_scanline();
     }
   }
 
diff --git a/src/cpu/bcpu/bcpu_timing.h b/src/cpu/bcpu/bcpu_timing.h
index bb70a3de..71a5e677 100644
--- a/src/cpu/bcpu/bcpu_timing.h
+++ b/src/cpu/bcpu/bcpu_timing.h
@@ -69,5 +69,6 @@ inline uint8  dma_counter();
 
 inline void   inc_vcounter();
 inline uint16 get_hcounter();
+inline void   cycle_edge();
 inline void   add_cycles(int cycles);
 inline void   time_reset();
diff --git a/src/cpu/cpuregs.h b/src/cpu/cpuregs.h
index 8d6447bd..bf5207cb 100644
--- a/src/cpu/cpuregs.h
+++ b/src/cpu/cpuregs.h
@@ -41,7 +41,7 @@ union {
   struct { uint8 l, h; };
 #else
   struct { uint8 h, l; };
-#endif;
+#endif
 };
 
   CPUReg16() { w = 0; }
diff --git a/src/cpu/dcpu.cpp b/src/cpu/dcpu.cpp
index be6896d0..8a3952ae 100644
--- a/src/cpu/dcpu.cpp
+++ b/src/cpu/dcpu.cpp
@@ -38,23 +38,23 @@ uint32 r = 0;
     break;
   case OPTYPE_IDP:
     addr = (regs.d + (addr & 0xffff)) & 0xffff;
-    r = (regs.db << 16) + mem_bus->read_word(addr);
+    r = (regs.db << 16) + r_mem->read_word(addr);
     break;
   case OPTYPE_IDPX:
     addr = (regs.d + regs.x + (addr & 0xffff)) & 0xffff;
-    r = (regs.db << 16) + mem_bus->read_word(addr);
+    r = (regs.db << 16) + r_mem->read_word(addr);
     break;
   case OPTYPE_IDPY:
     addr = (regs.d + (addr & 0xffff)) & 0xffff;
-    r = (regs.db << 16) + mem_bus->read_word(addr) + regs.y;
+    r = (regs.db << 16) + r_mem->read_word(addr) + regs.y;
     break;
   case OPTYPE_ILDP:
     addr = (regs.d + (addr & 0xffff)) & 0xffff;
-    r = mem_bus->read_long(addr);
+    r = r_mem->read_long(addr);
     break;
   case OPTYPE_ILDPY:
     addr = (regs.d + (addr & 0xffff)) & 0xffff;
-    r = mem_bus->read_long(addr) + regs.y;
+    r = r_mem->read_long(addr) + regs.y;
     break;
   case OPTYPE_ADDR:
     r = (regs.db << 16) + (addr & 0xffff);
@@ -88,7 +88,7 @@ uint32 r = 0;
     break;
   case OPTYPE_ISRY:
     addr = (regs.s + (addr & 0xff)) & 0xffff;
-    r = (regs.db << 16) + mem_bus->read_word(addr) + regs.y;
+    r = (regs.db << 16) + r_mem->read_word(addr) + regs.y;
     break;
   }
   return r;
@@ -109,10 +109,10 @@ static CPUReg24 pc;
   pc.d = regs.pc.d;
   sprintf(s, "%0.6x ", pc.d);
 
-  op  = mem_bus->read(pc.d); pc.w++;
-  op0 = mem_bus->read(pc.d); pc.w++;
-  op1 = mem_bus->read(pc.d); pc.w++;
-  op2 = mem_bus->read(pc.d);
+  op  = r_mem->read(pc.d); pc.w++;
+  op0 = r_mem->read(pc.d); pc.w++;
+  op1 = r_mem->read(pc.d); pc.w++;
+  op2 = r_mem->read(pc.d);
 
   switch(op) {
   case 0x00:sprintf(t, "brk #$%0.2x               ", op0);break;
@@ -457,7 +457,7 @@ static uint8 op_len_tbl[256] = {
     return 0;
   }
 
-  op = mem_bus->read(regs.pc.d);
+  op = r_mem->read(regs.pc.d);
   len = op_len_tbl[op];
   if(len == 5)return (regs.p.m)?2:3;
   if(len == 6)return (regs.p.x)?2:3;
diff --git a/src/dsp/bdsp/bdsp.cpp b/src/dsp/bdsp/bdsp.cpp
index d973ce81..18a91872 100644
--- a/src/dsp/bdsp/bdsp.cpp
+++ b/src/dsp/bdsp/bdsp.cpp
@@ -183,7 +183,7 @@ int i, v, n;
 
 void bDSP::power() {
 int v;
-  spcram = apu->get_spcram_handle();
+  spcram = r_apu->get_spcram_handle();
   memset(dspram, 0x00, 128);
 
   for(v=0;v<8;v++) {
diff --git a/src/interface.h b/src/interface.h
index b8729189..28fc434a 100644
--- a/src/interface.h
+++ b/src/interface.h
@@ -1,49 +1,81 @@
-#define BSNES_VERSION "0.013 wip2"
+#define BSNES_VERSION "0.014"
 #define BSNES_TITLE "bsnes v" BSNES_VERSION
 
+#ifdef POLYMORPHISM
+  #define ref(x) x
+#else
+  #define ref(x) (&x)
+#endif
+#define deref(x) __##x
+
+#define r_mem ref(deref(mem))
+#define r_cpu ref(deref(cpu))
+#define r_apu ref(deref(apu))
+#define r_dsp ref(deref(dsp))
+#define r_ppu ref(deref(ppu))
+
 #include "reader/reader.h"
 
 #include "memory/memory.h"
 #include "memory/bmemory/bmemory.h"
-extern MMIO mmio_unmapped;
-extern MemBus *mem_bus;
 
 #include "cpu/cpu.h"
 #include "cpu/bcpu/bcpu.h"
-extern CPU *cpu;
 
 #include "apu/apu.h"
 #include "apu/bapu/bapu.h"
-#include "apu/bapuskip/bapuskip.h"
-extern APU *apu;
 
 #include "dsp/dsp.h"
 #include "dsp/bdsp/bdsp.h"
-extern DSP *dsp;
 
 #include "ppu/ppu.h"
 #include "ppu/bppu/bppu.h"
-extern PPU *ppu;
 
 #include "snes/snes.h"
-extern SNES *snes;
 
 #include "chip/srtc/srtc.h"
 #include "chip/sdd1/sdd1.h"
+
+extern MMIO mmio_unmapped;
+#ifdef POLYMORPHISM
+  extern MemBus *deref(mem);
+  extern CPU    *deref(cpu);
+  extern APU    *deref(apu);
+  extern DSP    *deref(dsp);
+  extern PPU    *deref(ppu);
+#else
+  extern bMemBus deref(mem);
+  extern bCPU    deref(cpu);
+  extern bAPU    deref(apu);
+  extern bDSP    deref(dsp);
+  extern bPPU    deref(ppu);
+  #endif
+extern SNES *snes;
+
 extern SRTC *srtc;
 extern SDD1 *sdd1;
 
 #include "config/config.h"
 
 #ifdef INTERFACE_MAIN
-#include "config/config.cpp"
-  MemBus *mem_bus;
-  CPU    *cpu;
-  APU    *apu;
-  DSP    *dsp;
-  PPU    *ppu;
-  SNES   *snes;
 
-  SRTC   *srtc;
-  SDD1   *sdd1;
+#include "config/config.cpp"
+#ifdef POLYMORPHISM
+  MemBus *deref(mem);
+  CPU    *deref(cpu);
+  APU    *deref(apu);
+  DSP    *deref(dsp);
+  PPU    *deref(ppu);
+#else
+  bMemBus deref(mem);
+  bCPU    deref(cpu);
+  bAPU    deref(apu);
+  bDSP    deref(dsp);
+  bPPU    deref(ppu);
+#endif
+SNES *snes;
+
+SRTC *srtc;
+SDD1 *sdd1;
+
 #endif
diff --git a/src/lib/libconfig.cpp b/src/lib/libconfig.cpp
index 949e7c47..f772619e 100644
--- a/src/lib/libconfig.cpp
+++ b/src/lib/libconfig.cpp
@@ -11,7 +11,6 @@ uint Setting::get() {
 }
 
 void Setting::set(uint _data) {
-printf("%s %d\n", name, _data);
   data = _data;
 
   if(type != DEC && type != HEX) {
diff --git a/src/memory/bmemory/bmemory.cpp b/src/memory/bmemory/bmemory.cpp
index 010459a3..de003674 100644
--- a/src/memory/bmemory/bmemory.cpp
+++ b/src/memory/bmemory/bmemory.cpp
@@ -15,7 +15,7 @@ uint8  mapper, region;
   rom_size = rf->size();
 
   if(rom_size < 32768) {
-    free(rom_image);
+    zerofree(rom_image);
     return false;
   }
 
@@ -77,7 +77,7 @@ end:
   region = rom[index + 0x19];
 
   dprintf("* Image Name : \"%s\"", cart_title);
-  dprintf("* Region     : %s", (region <= 1)?"NTSC":"PAL");
+  dprintf("* Region     : %s", (region <= 1) ? "NTSC" : "PAL");
   dprintf("* MAD        : %0.2x", mapper);
   dprintf("* SRAM Size  : %dkb", sram_size / 1024);
   dprintf("* Reset:%0.4x NMI:%0.4x IRQ:%0.4x BRK[n]:%0.4x COP[n]:%0.4x BRK[e]:%0.4x COP[e]:%0.4x",
@@ -206,7 +206,9 @@ static uint32 r;
     break;
   }
 
+#ifdef DEBUGGER
   snes->notify(SNES::MEM_READ, addr, r);
+#endif
   return r;
 }
 
@@ -245,7 +247,9 @@ void bMemBus::write(uint32 addr, uint8 value) {
     break;
   }
 
+#ifdef DEBUGGER
   snes->notify(SNES::MEM_WRITE, addr, value);
+#endif
 }
 
 void bMemBus::power() {
diff --git a/src/memory/memory.cpp b/src/memory/memory.cpp
index ed894abd..c305ac2d 100644
--- a/src/memory/memory.cpp
+++ b/src/memory/memory.cpp
@@ -79,7 +79,7 @@ void Memory::write_long(uint32 addr, uint32 data, uint wrap) {
 }
 
 MMIO  mmio_unmapped;
-uint8 MMIO::read (uint32 addr) { return cpu->regs.mdr; }
+uint8 MMIO::read (uint32 addr) { return r_cpu->regs.mdr; }
 void  MMIO::write(uint32 addr, uint8 value) {}
 
 uint8 MemBus::calc_speed(uint32 addr, bool fast) {
@@ -92,10 +92,6 @@ uint8 MemBus::calc_speed(uint32 addr, bool fast) {
   return 8;
 }
 
-uint8 MemBus::speed(uint32 addr) {
-  return speed_table[addr >> 9];
-}
-
 void MemBus::set_speed(bool fast) {
   fastROM = fast;
 
diff --git a/src/memory/memory.h b/src/memory/memory.h
index 01424c13..7c195169 100644
--- a/src/memory/memory.h
+++ b/src/memory/memory.h
@@ -43,8 +43,10 @@ uint8 *speed_table,
        speed_table_fastrom[32768];
   inline uint8 calc_speed(uint32 addr, bool fast);
 public:
-  uint8 speed(uint32 addr);
-  void  set_speed(bool fast);
+  inline uint8 speed(uint32 addr) {
+    return speed_table[addr >> 9];
+  }
+  void set_speed(bool fast);
 
   virtual bool load_cart(Reader *rf) = 0;
   virtual bool load_sram(Reader *rf) = 0;
diff --git a/src/ppu/bppu/Copy of bppu_render_addsub.cpp b/src/ppu/bppu/Copy of bppu_render_addsub.cpp
new file mode 100644
index 00000000..2a216632
--- /dev/null
+++ b/src/ppu/bppu/Copy of bppu_render_addsub.cpp	
@@ -0,0 +1,64 @@
+inline uint16 bPPU::addsub_pixels(uint32 cdest, uint32 csrc) {
+  if(!regs.color_mode) {
+  //add
+    cdest = ((cdest << 16) | cdest) & 0x03e07c1f;
+    csrc  = ((csrc  << 16) | csrc)  & 0x03e07c1f;
+    cdest += csrc;
+
+    if(regs.color_halve) {
+      cdest >>= 1;
+    } else {
+      if(cdest & 0x04000000)cdest |= 0x03e00000;
+      if(cdest & 0x00008000)cdest |= 0x00007c00;
+      if(cdest & 0x00000020)cdest |= 0x0000001f;
+    }
+
+    cdest &= 0x03e07c1f;
+    return (cdest >> 16) | cdest;
+  } else {
+  //subtract
+    if((cdest & 0x7c00) < (csrc & 0x7c00))cdest &= ~0x7c00;
+    else cdest -= (csrc & 0x7c00);
+    if((cdest & 0x03e0) < (csrc & 0x03e0))cdest &= ~0x03e0;
+    else cdest -= (csrc & 0x03e0);
+    if((cdest & 0x001f) < (csrc & 0x001f))cdest &= ~0x001f;
+    else cdest -= (csrc & 0x001f);
+
+    return (!regs.color_halve) ? cdest : ((cdest & 0x7bde) >> 1);
+  }
+}
+
+inline uint16 bPPU::addsub_pixel(uint32 cdest) {
+uint32 csrc = regs.color_rgb;
+  if(!regs.color_mode) {
+  //add
+    cdest = ((cdest << 16) | cdest) & 0x03e07c1f;
+    csrc  = ((csrc  << 16) | csrc)  & 0x03e07c1f;
+    cdest += csrc;
+
+    if(regs.color_halve && !regs.addsub_mode) {
+      cdest >>= 1;
+    } else {
+      if(cdest & 0x04000000)cdest |= 0x03e00000;
+      if(cdest & 0x00008000)cdest |= 0x00007c00;
+      if(cdest & 0x00000020)cdest |= 0x0000001f;
+    }
+
+    cdest &= 0x03e07c1f;
+    return (cdest >> 16) | cdest;
+  } else {
+  //subtract
+    if((cdest & 0x7c00) < (csrc & 0x7c00))cdest &= ~0x7c00;
+    else cdest -= (csrc & 0x7c00);
+    if((cdest & 0x03e0) < (csrc & 0x03e0))cdest &= ~0x03e0;
+    else cdest -= (csrc & 0x03e0);
+    if((cdest & 0x001f) < (csrc & 0x001f))cdest &= ~0x001f;
+    else cdest -= (csrc & 0x001f);
+
+    if(regs.color_halve && !regs.addsub_mode) {
+      return (cdest & 0x7bde) >> 1;
+    }
+
+    return cdest;
+  }
+}
diff --git a/src/ppu/bppu/Copy of bppu_render_line.cpp b/src/ppu/bppu/Copy of bppu_render_line.cpp
new file mode 100644
index 00000000..58ac93f8
--- /dev/null
+++ b/src/ppu/bppu/Copy of bppu_render_line.cpp	
@@ -0,0 +1,82 @@
+inline uint16 bPPU::get_palette(uint8 index) {
+  return read16(cgram, index << 1);
+}
+
+inline uint16 bPPU::get_direct_color(uint8 p, uint8 t) {
+//p = 00000bgr <palette data>
+//t = BBGGGRRR <tilemap data>
+//r = 0BBb00GGGg0RRRr0 <return data>
+  return ((t & 7) << 2) | ((p & 1) << 1) |
+    (((t >> 3) & 7) << 7) | (((p >> 1) & 1) << 6) |
+    ((t >> 6) << 13) | ((p >> 2) << 12);
+}
+
+inline uint16 bPPU::get_pixel(int x) {
+_pixel *p = &pixel_cache[x];
+uint16 r, src_back = get_palette(0);
+  if(p->bg_main && p->bg_sub) {
+    if(!p->color_exempt && regs.color_enabled[p->bg_main & 0x7f] && window_cache[COL].sub[x]) {
+      if(regs.addsub_mode) {
+        r = addsub_pixels(p->src_main, p->src_sub);
+      } else {
+        r = addsub_pixel(p->src_main);
+      }
+    } else {
+      r = p->src_main;
+    }
+  } else if(p->bg_main) {
+    if(!p->color_exempt && regs.color_enabled[p->bg_main & 0x7f] && window_cache[COL].sub[x]) {
+      r = addsub_pixel(p->src_main);
+    } else {
+      r = p->src_main;
+    }
+  } else if(p->bg_sub) {
+    if(regs.color_enabled[BACK]) {
+      if(window_cache[COL].sub[x]) {
+        if(regs.addsub_mode) {
+          r = addsub_pixels(src_back, p->src_sub);
+        } else {
+          r = addsub_pixel(src_back);
+        }
+      } else {
+        r = src_back;
+      }
+    } else {
+      r = src_back; //was 0x0000 -- possibly another condition here?
+    }
+  } else {
+    if(window_cache[COL].main[x]) {
+      if(regs.color_enabled[BACK] && window_cache[COL].sub[x]) {
+        r = addsub_pixel(src_back);
+      } else {
+        r = src_back;
+      }
+    } else {
+      if(regs.color_enabled[BACK] && window_cache[COL].sub[x]) {
+        r = (!regs.color_mode) ? regs.color_rgb : 0x0000;
+      } else {
+        r = 0x0000;
+      }
+    }
+  }
+  return r;
+}
+
+inline void bPPU::render_line_output() {
+uint16 r, x;
+uint16 *ptr    = (uint16*)output + (line.y * 1024) +
+                 ((line.interlace && line.interlace_field) ? 512 : 0);
+uint16 *ltable = (uint16*)light_table + (regs.display_brightness << 15);
+
+  if(line.width == 256) {
+    for(x=0;x<256;x++) {
+      r = get_pixel(x);
+      *ptr++ = *(ltable + r);
+    }
+  } else {
+    for(x=0;x<512;x++) {
+      r = get_pixel(x);
+      *ptr++ = *(ltable + r);
+    }
+  }
+}
diff --git a/src/ppu/bppu/bppu.cpp b/src/ppu/bppu/bppu.cpp
index e67a663f..b0d1af9b 100644
--- a/src/ppu/bppu/bppu.cpp
+++ b/src/ppu/bppu/bppu.cpp
@@ -5,11 +5,11 @@
 void bPPU::run() {}
 
 void bPPU::scanline() {
-  line.y               = cpu->vcounter();
+  line.y               = r_cpu->vcounter();
   line.width           = (regs.bg_mode == 5 || regs.bg_mode == 6) ? 512 : 256;
   line.hires           = (regs.bg_mode == 5 || regs.bg_mode == 6);
-  line.interlace       = cpu->interlace();
-  line.interlace_field = cpu->interlace_field();
+  line.interlace       = r_cpu->interlace();
+  line.interlace_field = r_cpu->interlace_field();
 
   if(line.y == 0) {
   //RTO flag reset
@@ -26,7 +26,7 @@ void bPPU::scanline() {
     }
   }
 
-  if(line.y == (cpu->overscan() ? 239 : 224) && regs.display_disabled == false) {
+  if(line.y == (r_cpu->overscan() ? 239 : 224) && regs.display_disabled == false) {
   //OAM address reset
     regs.oam_addr = ((regs.oam_addrh << 8) | regs.oam_addrl) << 1;
   }
@@ -35,17 +35,18 @@ void bPPU::scanline() {
 void bPPU::render_scanline() {
   if(status.render_output == false)return;
 
-  if(line.y > 0 && line.y < (cpu->overscan() ? 239 : 224)) {
+  if(line.y > 0 && line.y < (r_cpu->overscan() ? 239 : 224)) {
     render_line();
   }
 }
 
 void bPPU::frame() {
   PPU::frame();
-  snes->notify(SNES::RENDER_FRAME);
 }
 
 void bPPU::power() {
+  PPU::power();
+
   memset(vram,  0, 65536);
   memset(oam,   0,   544);
   memset(cgram, 0,   512);
@@ -56,6 +57,7 @@ void bPPU::power() {
 }
 
 void bPPU::reset() {
+  PPU::reset();
   frame();
 
   memset(sprite_list, 0, sizeof(sprite_list));
@@ -228,9 +230,10 @@ void bPPU::reset() {
   regs.color_enabled[BG1]  = false;
 
 //$2132
-  regs.color_r = 0x00;
-  regs.color_g = 0x00;
-  regs.color_b = 0x00;
+  regs.color_r   = 0x00;
+  regs.color_g   = 0x00;
+  regs.color_b   = 0x00;
+  regs.color_rgb = 0x0000;
 
 //$2133
   regs.mode7_extbg = false;
@@ -262,20 +265,26 @@ void bPPU::reset() {
 uint8 bPPU::vram_read(uint16 addr) {
 uint8 r;
   r = vram[addr];
+#ifdef DEBUGGER
   snes->notify(SNES::VRAM_READ, addr, r);
+#endif
   return r;
 }
 
 void bPPU::vram_write(uint16 addr, uint8 value) {
   vram[addr] = value;
+#ifdef DEBUGGER
   snes->notify(SNES::VRAM_WRITE, addr, value);
+#endif
 }
 
 uint8 bPPU::oam_read(uint16 addr) {
 uint8 r;
   if(addr >= 0x0200)addr = 0x0200 | (addr & 31);
   r = oam[addr];
+#ifdef DEBUGGER
   snes->notify(SNES::OAM_READ, addr, r);
+#endif
   return r;
 }
 
@@ -283,7 +292,9 @@ void bPPU::oam_write(uint16 addr, uint8 value) {
   if(addr >= 0x0200)addr = 0x0200 | (addr & 31);
   oam[addr] = value;
   update_sprite_list(addr);
+#ifdef DEBUGGER
   snes->notify(SNES::OAM_WRITE, addr, value);
+#endif
 }
 
 uint8 bPPU::cgram_read(uint16 addr) {
@@ -293,7 +304,9 @@ uint8 r;
   if(addr & 1) {
     r &= 0x7f;
   }
+#ifdef DEBUGGER
   snes->notify(SNES::CGRAM_READ, addr, r);
+#endif
   return r;
 }
 
@@ -303,7 +316,9 @@ void bPPU::cgram_write(uint16 addr, uint8 value) {
     value &= 0x7f;
   }
   cgram[addr] = value;
+#ifdef DEBUGGER
   snes->notify(SNES::CGRAM_WRITE, addr, value);
+#endif
 }
 
 bPPU::bPPU() {
diff --git a/src/ppu/bppu/bppu.h b/src/ppu/bppu/bppu.h
index ce4956c3..6223b8db 100644
--- a/src/ppu/bppu/bppu.h
+++ b/src/ppu/bppu/bppu.h
@@ -127,6 +127,7 @@ struct {
 
 //$2132
   uint8  color_r, color_g, color_b;
+  uint16 color_rgb;
 
 //$2133
   bool   mode7_extbg;
diff --git a/src/ppu/bppu/bppu_mmio.cpp b/src/ppu/bppu/bppu_mmio.cpp
index 0bb916bd..fcd3ddbb 100644
--- a/src/ppu/bppu/bppu_mmio.cpp
+++ b/src/ppu/bppu/bppu_mmio.cpp
@@ -1,6 +1,6 @@
 void bPPU::latch_counters() {
-  regs.hcounter = cpu->hcounter();
-  regs.vcounter = cpu->vcounter();
+  regs.hcounter = r_cpu->hcounter();
+  regs.vcounter = r_cpu->vcounter();
   regs.counters_latched = true;
 }
 
@@ -21,20 +21,20 @@ bool bPPU::vram_can_read() {
     return true;
   }
 
-uint16 v  = cpu->vcounter();
-uint16 hc = cpu->hcycles();
+uint16 v  = r_cpu->vcounter();
+uint16 hc = r_cpu->hcycles();
 uint16 ls;
-  if(cpu->interlace() && !cpu->interlace_field()) {
-    ls = cpu->region_scanlines();
+  if(r_cpu->interlace() && !r_cpu->interlace_field()) {
+    ls = r_cpu->region_scanlines();
   } else {
-    ls = cpu->region_scanlines() - 1;
+    ls = r_cpu->region_scanlines() - 1;
   }
 
   if(v == ls && hc == 1362)return false;
 
-  if(v < (cpu->overscan() ? 239 : 224))return false;
+  if(v < (r_cpu->overscan() ? 239 : 224))return false;
 
-  if(v == (cpu->overscan() ? 239 : 224)) {
+  if(v == (r_cpu->overscan() ? 239 : 224)) {
     if(hc == 1362)return true;
     return false;
   }
@@ -47,17 +47,17 @@ bool bPPU::vram_can_write(uint8 &value) {
     return true;
   }
 
-uint16 v  = cpu->vcounter();
-uint16 hc = cpu->hcycles();
+uint16 v  = r_cpu->vcounter();
+uint16 hc = r_cpu->hcycles();
   if(v == 0) {
     if(hc <= 4)return true;
-    if(hc == 6) { value = cpu->regs.mdr; return true; }
+    if(hc == 6) { value = r_cpu->regs.mdr; return true; }
     return false;
   }
 
-  if(v < (cpu->overscan() ? 240 : 225))return false;
+  if(v < (r_cpu->overscan() ? 240 : 225))return false;
 
-  if(v == (cpu->overscan() ? 240 : 225)) {
+  if(v == (r_cpu->overscan() ? 240 : 225)) {
     if(hc <= 4)return false;
     return true;
   }
@@ -542,6 +542,8 @@ void bPPU::mmio_w2132(uint8 value) {
   if(value & 0x80)regs.color_b = value & 0x1f;
   if(value & 0x40)regs.color_g = value & 0x1f;
   if(value & 0x20)regs.color_r = value & 0x1f;
+
+  regs.color_rgb = (regs.color_r) | (regs.color_g << 5) | (regs.color_b << 10);
 }
 
 //SETINI
@@ -552,8 +554,8 @@ void bPPU::mmio_w2133(uint8 value) {
   regs.oam_halve   = !!(value & 0x02);
   regs.interlace   = !!(value & 0x01);
 
-  cpu->set_overscan(regs.overscan);
-  cpu->set_interlace(regs.interlace);
+  r_cpu->set_overscan(regs.overscan);
+  r_cpu->set_interlace(regs.interlace);
 }
 
 //MPYL
@@ -582,10 +584,10 @@ uint32 r;
 
 //SLHV
 uint8 bPPU::mmio_r2137() {
-  if(cpu->pio_status() & 0x80) {
+  if(r_cpu->pio_status() & 0x80) {
     latch_counters();
   }
-  return cpu->regs.mdr;
+  return r_cpu->regs.mdr;
 }
 
 //OAMDATAREAD
@@ -689,8 +691,8 @@ uint8 r = 0x00;
   regs.latch_hcounter = 0;
   regs.latch_vcounter = 0;
 
-  r |= cpu->interlace_field() << 7;
-  if(!(cpu->pio_status() & 0x80)) {
+  r |= r_cpu->interlace_field() << 7;
+  if(!(r_cpu->pio_status() & 0x80)) {
     r |= 0x40;
   } else if(regs.counters_latched == true) {
     r |= 0x40;
@@ -738,7 +740,7 @@ uint8 bPPUMMIO::read(uint32 addr) {
   case 0x213f:return ppu->mmio_r213f(); //STAT78
   }
 
-  return cpu->regs.mdr;
+  return r_cpu->regs.mdr;
 }
 
 void bPPUMMIO::write(uint32 addr, uint8 value) {
diff --git a/src/ppu/bppu/bppu_render.cpp b/src/ppu/bppu/bppu_render.cpp
index 05618f3f..8edac47c 100644
--- a/src/ppu/bppu/bppu_render.cpp
+++ b/src/ppu/bppu/bppu_render.cpp
@@ -20,28 +20,25 @@ inline void bPPU::render_line_mode0() {
 }
 
 /*
-Mode 1 (pri=0): ->
-     1,    2,    3,    4,    5,    6,    7,    8,    9,   10
-  BG3B, OAM0, BG3A, OAM1, BG2B, BG1B, OAM2, BG2A, BG1A, OAM3
-
 Mode 1 (pri=1): ->
      1,    2,    3,    4,    5,    6,    7,    8,    9,   10
   BG3B, OAM0, OAM1, BG2B, BG1B, OAM2, BG2A, BG1A, OAM3, BG3A
+
+Mode 1 (pri=0): ->
+     1,    2,    3,    4,    5,    6,    7,    8,    9,   10
+  BG3B, OAM0, BG3A, OAM1, BG2B, BG1B, OAM2, BG2A, BG1A, OAM3
 */
 inline void bPPU::render_line_mode1() {
-  switch(regs.bg3_priority) {
-  case 0:
-    render_line_bg(BG1, COLORDEPTH_16, 6,  9);
-    render_line_bg(BG2, COLORDEPTH_16, 5,  8);
-    render_line_bg(BG3, COLORDEPTH_4,  1,  3);
-    render_line_oam(2, 4, 7, 10);
-    break;
-  case 1:
+  if(regs.bg3_priority) {
     render_line_bg(BG1, COLORDEPTH_16, 5,  8);
     render_line_bg(BG2, COLORDEPTH_16, 4,  7);
     render_line_bg(BG3, COLORDEPTH_4,  1, 10);
     render_line_oam(2, 3, 6, 9);
-    break;
+  } else {
+    render_line_bg(BG1, COLORDEPTH_16, 6,  9);
+    render_line_bg(BG2, COLORDEPTH_16, 5,  8);
+    render_line_bg(BG3, COLORDEPTH_4,  1,  3);
+    render_line_oam(2, 4, 7, 10);
   }
 }
 
@@ -121,7 +118,7 @@ inline void bPPU::render_line_mode7() {
 
 void bPPU::render_line() {
   if(regs.display_disabled == true) {
-    memset(snes->get_ppu_output_handle(), 0, 1024);
+    memset(output + (line.y * 1024), 0, 1024);
     return;
   }
 
diff --git a/src/ppu/bppu/bppu_render.h b/src/ppu/bppu/bppu_render.h
index 5143f21d..0e9b3c24 100644
--- a/src/ppu/bppu/bppu_render.h
+++ b/src/ppu/bppu/bppu_render.h
@@ -64,24 +64,25 @@ struct oam_tileitem {
 } oam_tilelist[34];
 
 enum { OAM_PRI_NONE = 4 };
-uint8 oam_line_pal[512], oam_line_pri[512];
+uint8 oam_line_pal[256], oam_line_pri[256];
 
 bool is_sprite_on_scanline();
 void load_oam_tiles();
 void render_oam_tile(int tile_num);
 void render_line_oam(uint8 pri0_pos, uint8 pri1_pos, uint8 pri2_pos, uint8 pri3_pos);
+void render_line_oam_lores(uint8 pri0_pos, uint8 pri1_pos, uint8 pri2_pos, uint8 pri3_pos);
+void render_line_oam_hires(uint8 pri0_pos, uint8 pri1_pos, uint8 pri2_pos, uint8 pri3_pos);
 
 //bppu_render_mode7.cpp
 void render_line_mode7(uint8 bg, uint8 pri0_pos, uint8 pri1_pos);
 
 //bppu_render_addsub.cpp
-inline uint16 addsub_pixels(uint32 cdest, uint32 csrc);
-inline uint16 addsub_pixel (uint32 cdest);
+inline uint16 addsub_pixels(uint32 x);
 
 //bppu_render_line.cpp
 enum { BLENDTYPE_BACK = 0, BLENDTYPE_MAIN = 1, BLENDTYPE_SUB = 2, BLENDTYPE_COMBINE = 3 };
 
 inline uint16 get_palette(uint8 index);
 inline uint16 get_direct_color(uint8 p, uint8 t);
-inline uint16 get_pixel(int x);
+inline uint16 get_pixel(uint32 x);
 inline void render_line_output();
diff --git a/src/ppu/bppu/bppu_render_addsub.cpp b/src/ppu/bppu/bppu_render_addsub.cpp
index 43bea198..835d4d8a 100644
--- a/src/ppu/bppu/bppu_render_addsub.cpp
+++ b/src/ppu/bppu/bppu_render_addsub.cpp
@@ -1,92 +1,39 @@
-inline uint16 bPPU::addsub_pixels(uint32 cdest, uint32 csrc) {
-int r, g, b;
-uint16 res;
-  switch(regs.color_mode) {
-  case 0: //COLORMODE_ADD:
-    if(regs.color_halve == true) {
-      cdest = ((cdest << 16) | cdest) & 0x03e07c1f;
-      csrc  = ((csrc  << 16) | csrc)  & 0x03e07c1f;
-      cdest += csrc;
-      cdest >>= 1;
-      cdest &= 0x03e07c1f;
-      return (cdest >> 16) | cdest;
-    } else {
-      cdest = ((cdest << 16) | cdest) & 0x03e07c1f;
-      csrc  = ((csrc  << 16) | csrc)  & 0x03e07c1f;
-      cdest += csrc;
-      if(cdest & 0x04000000)cdest |= 0x03e00000;
-      if(cdest & 0x00008000)cdest |= 0x00007c00;
-      if(cdest & 0x00000020)cdest |= 0x0000001f;
-      cdest &= 0x03e07c1f;
-      return (cdest >> 16) | cdest;
+inline uint16 bPPU::addsub_pixels(uint32 x) {
+uint32 cdest = pixel_cache[x].src_main,
+       csrc  = pixel_cache[x].src_sub;
+bool   halve = false;
+  if(regs.color_halve && window_cache[COL].main[x]) {
+    if(regs.addsub_mode && pixel_cache[x].bg_sub == BACK);
+    else {
+      halve = true;
     }
-    break;
-  case 1: //COLORMODE_SUB:
-    if(regs.color_halve == true) {
-      if((cdest & 0x7c00) < (csrc & 0x7c00))cdest &= ~0x7c00;
-      else cdest -= (csrc & 0x7c00);
-      if((cdest & 0x03e0) < (csrc & 0x03e0))cdest &= ~0x03e0;
-      else cdest -= (csrc & 0x03e0);
-      if((cdest & 0x001f) < (csrc & 0x001f))cdest &= ~0x001f;
-      else cdest -= (csrc & 0x001f);
-      return (cdest & 0x7bde) >> 1;
-    } else {
-      if((cdest & 0x7c00) < (csrc & 0x7c00))cdest &= ~0x7c00;
-      else cdest -= (csrc & 0x7c00);
-      if((cdest & 0x03e0) < (csrc & 0x03e0))cdest &= ~0x03e0;
-      else cdest -= (csrc & 0x03e0);
-      if((cdest & 0x001f) < (csrc & 0x001f))cdest &= ~0x001f;
-      else cdest -= (csrc & 0x001f);
-      return cdest;
-    }
-    break;
   }
-  return 0x0000; //prevent annoying warning message
-}
 
-inline uint16 bPPU::addsub_pixel(uint32 cdest) {
-int r, g, b;
-uint32 csrc  = (regs.color_r) | (regs.color_g << 5) | (regs.color_b << 10);
-uint16 res;
-  switch(regs.color_mode) {
-  case 0: //COLORMODE_ADD:
-    if(regs.color_halve == true && regs.addsub_mode == 0) {
-      cdest = ((cdest << 16) | cdest) & 0x03e07c1f;
-      csrc  = ((csrc  << 16) | csrc)  & 0x03e07c1f;
-      cdest += csrc;
-      cdest >>= 1;
-      cdest &= 0x03e07c1f;
-      return (cdest >> 16) | cdest;
-    } else {
-      cdest = ((cdest << 16) | cdest) & 0x03e07c1f;
-      csrc  = ((csrc  << 16) | csrc)  & 0x03e07c1f;
-      cdest += csrc;
+  if(!regs.color_mode) {
+  //add
+    cdest = ((cdest << 16) | cdest) & 0x03e07c1f;
+    csrc  = ((csrc  << 16) | csrc)  & 0x03e07c1f;
+    cdest += csrc;
+
+    if(!halve) {
       if(cdest & 0x04000000)cdest |= 0x03e00000;
       if(cdest & 0x00008000)cdest |= 0x00007c00;
       if(cdest & 0x00000020)cdest |= 0x0000001f;
-      cdest &= 0x03e07c1f;
-      return (cdest >> 16) | cdest;
-    }
-    break;
-  case 1: //COLORMODE_SUB:
-    if(regs.color_halve == true && regs.addsub_mode == 0) {
-      if((cdest & 0x7c00) < (csrc & 0x7c00))cdest &= ~0x7c00;
-      else cdest -= (csrc & 0x7c00);
-      if((cdest & 0x03e0) < (csrc & 0x03e0))cdest &= ~0x03e0;
-      else cdest -= (csrc & 0x03e0);
-      if((cdest & 0x001f) < (csrc & 0x001f))cdest &= ~0x001f;
-      else cdest -= (csrc & 0x001f);
-      return (cdest & 0x7bde) >> 1;
     } else {
-      if((cdest & 0x7c00) < (csrc & 0x7c00))cdest &= ~0x7c00;
-      else cdest -= (csrc & 0x7c00);
-      if((cdest & 0x03e0) < (csrc & 0x03e0))cdest &= ~0x03e0;
-      else cdest -= (csrc & 0x03e0);
-      if((cdest & 0x001f) < (csrc & 0x001f))cdest &= ~0x001f;
-      else cdest -= (csrc & 0x001f);
-      return cdest;
+      cdest >>= 1;
     }
-    break;
+
+    cdest &= 0x03e07c1f;
+    return (cdest >> 16) | cdest;
+  } else {
+  //subtract
+    if((cdest & 0x7c00) < (csrc & 0x7c00))cdest &= ~0x7c00;
+    else cdest -= (csrc & 0x7c00);
+    if((cdest & 0x03e0) < (csrc & 0x03e0))cdest &= ~0x03e0;
+    else cdest -= (csrc & 0x03e0);
+    if((cdest & 0x001f) < (csrc & 0x001f))cdest &= ~0x001f;
+    else cdest -= (csrc & 0x001f);
+
+    return (!halve) ? cdest : ((cdest & 0x7bde) >> 1);
   }
-  return 0x0000; //prevent annoying warning message
 }
diff --git a/src/ppu/bppu/bppu_render_bg.cpp b/src/ppu/bppu/bppu_render_bg.cpp
index 87c54b5c..9b0bd8d6 100644
--- a/src/ppu/bppu/bppu_render_bg.cpp
+++ b/src/ppu/bppu/bppu_render_bg.cpp
@@ -43,6 +43,7 @@ uint16 opt_valid_bit  = (bg == BG1) ? 0x2000 : ((bg == BG2) ? 0x4000 : 0x0000);
 uint8  bgpal_index    = (regs.bg_mode == 0) ? (bg << 5) : 0;
 
 uint8  pal_size       = (color_depth == 2) ? 256 : ((color_depth == 1) ? 16 : 4);
+uint16 tile_mask      = (color_depth == 2) ? 0x03ff : ((color_depth == 1) ? 0x07ff : 0x0fff);
 //4 + color_depth = >>(4-6) -- / {16, 32, 64 } bytes/tile
 //index is a tile number count to add to base tile number
 uint   tiledata_index = regs.bg_tdaddr[bg] >> (4 + color_depth);
@@ -145,6 +146,7 @@ int32 prev_x = -1, prev_y = -1;
 
       tile_num &= 0x03ff;
       tile_num += tiledata_index;
+      tile_num &= tile_mask;
 
       if(bg_td_state[tile_num] == 1) {
         render_bg_tile(color_depth, tile_num);
@@ -162,7 +164,7 @@ int32 prev_x = -1, prev_y = -1;
     xpos = mosaic_x & 7;
     if(mirror_x)xpos ^= 7; //invert x tile pos
     col = *(tile_ptr + xpos);
-    if(col && window_cache[COL].main[x]) {
+    if(col) {
       if(regs.direct_color == true && bg == BG1 && (regs.bg_mode == 3 || regs.bg_mode == 4)) {
         col = get_direct_color(pal_num, col);
       } else {
diff --git a/src/ppu/bppu/bppu_render_cache.cpp b/src/ppu/bppu/bppu_render_cache.cpp
index aa57f3cf..0557f711 100644
--- a/src/ppu/bppu/bppu_render_cache.cpp
+++ b/src/ppu/bppu/bppu_render_cache.cpp
@@ -118,7 +118,7 @@ void bPPU::init_tiledata_cache() {
 void bPPU::clear_tiledata_cache() {
   memset(bg_tiledata[TILE_2BIT],       0, 262144);
   memset(bg_tiledata[TILE_4BIT],       0, 131072);
-  memset(bg_tiledata[TILE_4BIT],       0,  65536);
+  memset(bg_tiledata[TILE_8BIT],       0,  65536);
   memset(bg_tiledata_state[TILE_2BIT], 0,   4096);
   memset(bg_tiledata_state[TILE_4BIT], 0,   2048);
   memset(bg_tiledata_state[TILE_8BIT], 0,   1024);
diff --git a/src/ppu/bppu/bppu_render_line.cpp b/src/ppu/bppu/bppu_render_line.cpp
index 6293c76c..3e254660 100644
--- a/src/ppu/bppu/bppu_render_line.cpp
+++ b/src/ppu/bppu/bppu_render_line.cpp
@@ -11,71 +11,59 @@ inline uint16 bPPU::get_direct_color(uint8 p, uint8 t) {
     ((t >> 6) << 13) | ((p >> 2) << 12);
 }
 
-inline uint16 bPPU::get_pixel(int x) {
+inline uint16 bPPU::get_pixel(uint32 x) {
 _pixel *p = &pixel_cache[x];
-uint16 _r, src_back = get_palette(0);
-  if(p->bg_main && p->bg_sub) {
-    if(p->color_exempt == false && regs.color_enabled[p->bg_main & 0x7f] && window_cache[COL].sub[x]) {
-      if(regs.addsub_mode) {
-        _r = addsub_pixels(p->src_main, p->src_sub);
-      } else {
-        _r = addsub_pixel(p->src_main);
-      }
-    } else {
-      _r = p->src_main;
-    }
-  } else if(p->bg_main) {
-    if(p->color_exempt == false && regs.color_enabled[p->bg_main & 0x7f] && window_cache[COL].sub[x]) {
-      _r = addsub_pixel(p->src_main);
-    } else {
-      _r = p->src_main;
-    }
-  } else if(p->bg_sub) {
-    if(regs.color_enabled[BACK]) {
-      if(window_cache[COL].sub[x]) {
-        if(regs.addsub_mode) {
-          _r = addsub_pixels(src_back, p->src_sub);
-        } else {
-          _r = addsub_pixel(src_back);
-        }
-      } else {
-        _r = src_back;
-      }
-    } else {
-      _r = src_back; //was 0x0000 -- possibly another condition here?
-    }
+  if(!p->bg_main) {
+    p->bg_main      = BACK;
+    p->src_main     = get_palette(0);
+    p->color_exempt = false;
   } else {
-    if(window_cache[COL].main[x]) {
-      if(regs.color_enabled[BACK] && window_cache[COL].sub[x]) {
-        _r = addsub_pixel(src_back);
-      } else {
-        _r = src_back;
-      }
-    } else {
-      _r = 0x0000;
-    }
+    p->bg_main &= 0x7f;
   }
-  return _r;
+
+  if(!p->bg_sub) {
+    p->bg_sub  = BACK;
+    p->src_sub = regs.color_rgb;
+  } else {
+    p->bg_sub &= 0x7f;
+  }
+
+  if(!regs.addsub_mode) {
+    p->bg_sub  = BACK;
+    p->src_sub = regs.color_rgb;
+  }
+
+  if(!window_cache[COL].main[x]) {
+    if(!window_cache[COL].sub[x]) {
+      return 0x0000;
+    }
+  //p->bg_main remains the same, even when the main color window
+  //masks out the color. this is needed for regs.color_enabled[p->bg_main]
+  //test below. illusion of gaia relies on this behavior for its load menu.
+    p->src_main = 0x0000;
+  }
+
+  if(!p->color_exempt && regs.color_enabled[p->bg_main] && window_cache[COL].sub[x]) {
+    return addsub_pixels(x);
+  }
+  return p->src_main;
 }
 
 inline void bPPU::render_line_output() {
-int x;
-uint16 _r;
-uint16 *ptr;
-  ptr = (uint16*)snes->get_ppu_output_handle();
-
-uint16 *ltable;
-  ltable = (uint16*)light_table + (regs.display_brightness << 15);
+uint16 r, x;
+uint16 *ptr    = (uint16*)output + (line.y * 1024) +
+                 ((line.interlace && line.interlace_field) ? 512 : 0);
+uint16 *ltable = (uint16*)light_table + (regs.display_brightness << 15);
 
   if(line.width == 256) {
     for(x=0;x<256;x++) {
-      _r = get_pixel(x);
-      *ptr++ = *(ltable + _r);
+      r = get_pixel(x);
+      *ptr++ = *(ltable + r);
     }
   } else {
     for(x=0;x<512;x++) {
-      _r = get_pixel(x);
-      *ptr++ = *(ltable + _r);
+      r = get_pixel(x);
+      *ptr++ = *(ltable + r);
     }
   }
 }
diff --git a/src/ppu/bppu/bppu_render_mode7.cpp b/src/ppu/bppu/bppu_render_mode7.cpp
index 4c9f7d2a..6e0797eb 100644
--- a/src/ppu/bppu/bppu_render_mode7.cpp
+++ b/src/ppu/bppu/bppu_render_mode7.cpp
@@ -120,29 +120,27 @@ int32 psy = ((c * CLIP(hofs - cx)) & ~63) + ((d * CLIP(vofs - cy)) & ~63) + ((d
       _x = x;
     }
 
-    if(window_cache[COL].main[_x]) {
-    uint32 col;
-      if(regs.direct_color == true && bg == BG1) {
-      //direct color mode does not apply to bg2, as it is only 128 colors...
-        col = get_direct_color(0, palette);
-      } else {
-        col = get_palette(palette);
-      }
+  uint32 col;
+    if(regs.direct_color == true && bg == BG1) {
+    //direct color mode does not apply to bg2, as it is only 128 colors...
+      col = get_direct_color(0, palette);
+    } else {
+      col = get_palette(palette);
+    }
 
-      if(regs.bg_enabled[bg] == true && !wt_main[_x]) {
-        if(pixel_cache[_x].pri_main < _pri) {
-          pixel_cache[_x].pri_main = _pri;
-          pixel_cache[_x].bg_main  = 0x80 | bg;
-          pixel_cache[_x].src_main = col;
-          pixel_cache[_x].color_exempt = false;
-        }
+    if(regs.bg_enabled[bg] == true && !wt_main[_x]) {
+      if(pixel_cache[_x].pri_main < _pri) {
+        pixel_cache[_x].pri_main = _pri;
+        pixel_cache[_x].bg_main  = 0x80 | bg;
+        pixel_cache[_x].src_main = col;
+        pixel_cache[_x].color_exempt = false;
       }
-      if(regs.bgsub_enabled[bg] == true && !wt_sub[_x]) {
-        if(pixel_cache[_x].pri_sub < _pri) {
-          pixel_cache[_x].pri_sub = _pri;
-          pixel_cache[_x].bg_sub  = 0x80 | bg;
-          pixel_cache[_x].src_sub = col;
-        }
+    }
+    if(regs.bgsub_enabled[bg] == true && !wt_sub[_x]) {
+      if(pixel_cache[_x].pri_sub < _pri) {
+        pixel_cache[_x].pri_sub = _pri;
+        pixel_cache[_x].bg_sub  = 0x80 | bg;
+        pixel_cache[_x].src_sub = col;
       }
     }
   }
diff --git a/src/ppu/bppu/bppu_render_oam.cpp b/src/ppu/bppu/bppu_render_oam.cpp
index e6a3b0db..a1177b40 100644
--- a/src/ppu/bppu/bppu_render_oam.cpp
+++ b/src/ppu/bppu/bppu_render_oam.cpp
@@ -63,10 +63,10 @@ uint16 width, height;
     addr &= 0x001f;
     z = oam[0x0200 + addr];
     i = addr << 2;
-    sprite_list[i    ].x = ((z & 0x01)?0x0100:0x0000) + (sprite_list[i    ].x & 0xff);
-    sprite_list[i + 1].x = ((z & 0x04)?0x0100:0x0000) + (sprite_list[i + 1].x & 0xff);
-    sprite_list[i + 2].x = ((z & 0x10)?0x0100:0x0000) + (sprite_list[i + 2].x & 0xff);
-    sprite_list[i + 3].x = ((z & 0x40)?0x0100:0x0000) + (sprite_list[i + 3].x & 0xff);
+    sprite_list[i    ].x = ((z & 0x01) ? 256 : 0) + (sprite_list[i    ].x & 255);
+    sprite_list[i + 1].x = ((z & 0x04) ? 256 : 0) + (sprite_list[i + 1].x & 255);
+    sprite_list[i + 2].x = ((z & 0x10) ? 256 : 0) + (sprite_list[i + 2].x & 255);
+    sprite_list[i + 3].x = ((z & 0x40) ? 256 : 0) + (sprite_list[i + 3].x & 255);
     get_sprite_size(i,     !!(z & 0x02));
     get_sprite_size(i + 1, !!(z & 0x08));
     get_sprite_size(i + 2, !!(z & 0x20));
@@ -90,7 +90,7 @@ uint16 addr = 0x0200;
 bool bPPU::is_sprite_on_scanline() {
 //if sprite is entirely offscreen and doesn't wrap around to the left side of the screen,
 //then it is not counted. 256 is correct, and not 255 -- as one might first expect
-  if(spr->x > 256 && (spr->x + spr->width) < 512 && line.width != 512)return false;
+  if(spr->x > 256 && (spr->x + spr->width) < 512)return false;
 
   if(regs.oam_halve == false) {
     if(line.y >= spr->y && line.y < (spr->y + spr->height)) {
@@ -110,19 +110,10 @@ bool bPPU::is_sprite_on_scanline() {
 }
 
 void bPPU::load_oam_tiles() {
-uint16 tile_width;
-  tile_width = spr->width >> 3;
+uint16 tile_width = spr->width >> 3;
+int x = spr->x;
+int y = (spr->vflip) ? ((spr->height - 1) - (line.y - spr->y)) : (line.y - spr->y);
 
-int x, y, chr, nameselect_index;
-  x = spr->x;
-  if(line.width == 512)x <<= 1;
-  x &= 511;
-
-  if(spr->vflip) {
-    y = ((spr->height - 1) - (line.y - spr->y));
-  } else {
-    y = (line.y - spr->y);
-  }
 //todo: double-check code below. seems that interlace_field
 //should be added to hires 512x448 sprites as well, and not
 //just when oam_halve is enabled...
@@ -132,40 +123,40 @@ int x, y, chr, nameselect_index;
       y += line.interlace_field;
     }
   }
+
+  x &= 511;
   y &= 255;
 
-  chr = spr->character;
+uint16 tdaddr = regs.oam_tdaddr;
+uint16 chrx   = (spr->character     ) & 15;
+uint16 chry   = (spr->character >> 4) & 15;
   if(spr->use_nameselect == true) {
-    chr += 256;
-    nameselect_index = regs.oam_nameselect << 13;
-  } else {
-    nameselect_index = 0x0000;
+    tdaddr += (256 * 32) + (regs.oam_nameselect << 13);
   }
-  chr += (y >> 3) << 4;
+  chry  += (y >> 3);
+  chry  &= 15;
+  chry <<= 4;
 
-int i, n, mx, pos, z;
+int i, n, sx, mx, pos;
   for(i=0;i<tile_width;i++) {
-    z  = x;
-    z += (i << ((line.width == 512) ? 4 : 3));
-    z &= 511;
+    sx  = x;
+    sx += i << 3;
+    sx &= 511;
+
   //ignore sprites that are offscreen
   //sprites at 256 are still counted, even though they aren't visible onscreen
-    if(z >= 257 && (z + 7) < 512 && line.width != 512)continue;
+    if(sx >= 257 && (sx + 7) < 512)continue;
 
     if(regs.oam_tilecount++ > 34)break;
     n = regs.oam_tilecount - 1;
-    oam_tilelist[n].x     = z;
+    oam_tilelist[n].x     = sx;
     oam_tilelist[n].y     = y;
     oam_tilelist[n].pri   = spr->priority;
     oam_tilelist[n].pal   = (spr->palette << 4) + 128;
     oam_tilelist[n].hflip = spr->hflip;
 
-    if(oam_tilelist[n].hflip) {
-      mx = (tile_width - 1) - i;
-    } else {
-      mx = i;
-    }
-    pos = regs.oam_tdaddr + ((chr + mx) << 5) + ((y & 7) << 1) + nameselect_index;
+    mx  = (oam_tilelist[n].hflip) ? ((tile_width - 1) - i) : i;
+    pos = tdaddr + ((chry + ((chrx + mx) & 15)) << 5);
     oam_tilelist[n].tile = (pos >> 5) & 0x07ff;
   }
 }
@@ -182,38 +173,26 @@ oam_tileitem *t = &oam_tilelist[tile_num];
 
 int x, sx, col;
   sx = t->x;
-//tile_ptr = tiledata + (tile * (8_width * 8_height)) + ((y & 7_height_mask) * 8_width);
   tile_ptr = (uint8*)oam_td + (t->tile << 6) + ((t->y & 7) << 3);
   for(x=0;x<8;x++) {
     sx &= 511;
-    if(sx < line.width) {
-      col = *(tile_ptr + ((t->hflip)?7-x:x));
+    if(sx < 256) {
+      col = *(tile_ptr + ((t->hflip) ? (7 - x) : x));
       if(col) {
         col += t->pal;
         oam_line_pal[sx] = col;
         oam_line_pri[sx] = t->pri;
-        if(line.width == 512) {
-          oam_line_pal[sx + 1] = col;
-          oam_line_pri[sx + 1] = t->pri;
-        }
       }
     }
-    sx += (line.width == 512) ? 2 : 1;
+    sx++;
   }
 }
 
 void bPPU::render_line_oam(uint8 pri0_pos, uint8 pri1_pos, uint8 pri2_pos, uint8 pri3_pos) {
-int s, x;
-bool _bg_enabled    = regs.bg_enabled[OAM];
-bool _bgsub_enabled = regs.bgsub_enabled[OAM];
-
-  build_window_tables(OAM);
-uint8 *wt_main = window_cache[OAM].main;
-uint8 *wt_sub  = window_cache[OAM].sub;
-
+int s;
   regs.oam_itemcount = 0;
   regs.oam_tilecount = 0;
-  memset(oam_line_pri, OAM_PRI_NONE, 512);
+  memset(oam_line_pri, OAM_PRI_NONE, 256);
 
   memset(oam_itemlist, 0xff, 32);
   for(s=0;s<128;s++) {
@@ -239,34 +218,105 @@ uint8 *wt_sub  = window_cache[OAM].sub;
   regs.time_over  |= (regs.oam_tilecount > 34);
   regs.range_over |= (regs.oam_itemcount > 32);
 
-  if(_bg_enabled == false && _bgsub_enabled == false)return;
+  if(regs.bg_enabled[OAM] == false && regs.bgsub_enabled[OAM] == false)return;
 
-int _pri;
-  for(x=0;x<line.width;x++) {
+  if(line.width == 256) {
+    render_line_oam_lores(pri0_pos, pri1_pos, pri2_pos, pri3_pos);
+  } else {
+    render_line_oam_hires(pri0_pos, pri1_pos, pri2_pos, pri3_pos);
+  }
+}
+
+void bPPU::render_line_oam_lores(uint8 pri0_pos, uint8 pri1_pos, uint8 pri2_pos, uint8 pri3_pos) {
+bool bg_enabled    = regs.bg_enabled[OAM];
+bool bgsub_enabled = regs.bgsub_enabled[OAM];
+
+  build_window_tables(OAM);
+uint8 *wt_main = window_cache[OAM].main;
+uint8 *wt_sub  = window_cache[OAM].sub;
+
+int pri;
+  for(int x=0;x<256;x++) {
     if(oam_line_pri[x] == OAM_PRI_NONE)continue;
 
     switch(oam_line_pri[x]) {
-    case 0:_pri = pri0_pos;break;
-    case 1:_pri = pri1_pos;break;
-    case 2:_pri = pri2_pos;break;
-    case 3:_pri = pri3_pos;break;
+    case 0:pri = pri0_pos;break;
+    case 1:pri = pri1_pos;break;
+    case 2:pri = pri2_pos;break;
+    case 3:pri = pri3_pos;break;
     }
 
-    if(window_cache[COL].main[x]) {
-      if(_bg_enabled == true && !wt_main[x]) {
-        if(pixel_cache[x].pri_main < _pri) {
-          pixel_cache[x].pri_main = _pri;
-          pixel_cache[x].bg_main  = PC_OAM;
-          pixel_cache[x].src_main = get_palette(oam_line_pal[x]);
-          pixel_cache[x].color_exempt = (oam_line_pal[x] < 192);
-        }
+    if(bg_enabled == true && !wt_main[x]) {
+      if(pixel_cache[x].pri_main < pri) {
+        pixel_cache[x].pri_main = pri;
+        pixel_cache[x].bg_main  = PC_OAM;
+        pixel_cache[x].src_main = get_palette(oam_line_pal[x]);
+        pixel_cache[x].color_exempt = (oam_line_pal[x] < 192);
       }
-      if(_bgsub_enabled == true && !wt_sub[x]) {
-        if(pixel_cache[x].pri_sub < _pri) {
-          pixel_cache[x].pri_sub = _pri;
-          pixel_cache[x].bg_sub  = PC_OAM;
-          pixel_cache[x].src_sub = get_palette(oam_line_pal[x]);
-        }
+    }
+
+    if(bgsub_enabled == true && !wt_sub[x]) {
+      if(pixel_cache[x].pri_sub < pri) {
+        pixel_cache[x].pri_sub = pri;
+        pixel_cache[x].bg_sub  = PC_OAM;
+        pixel_cache[x].src_sub = get_palette(oam_line_pal[x]);
+      }
+    }
+  }
+}
+
+void bPPU::render_line_oam_hires(uint8 pri0_pos, uint8 pri1_pos, uint8 pri2_pos, uint8 pri3_pos) {
+bool bg_enabled    = regs.bg_enabled[OAM];
+bool bgsub_enabled = regs.bgsub_enabled[OAM];
+
+  build_window_tables(OAM);
+uint8 *wt_main = window_cache[OAM].main;
+uint8 *wt_sub  = window_cache[OAM].sub;
+
+int pri, sx;
+  for(int x=0;x<256;x++) {
+    if(oam_line_pri[x] == OAM_PRI_NONE)continue;
+
+    switch(oam_line_pri[x]) {
+    case 0:pri = pri0_pos;break;
+    case 1:pri = pri1_pos;break;
+    case 2:pri = pri2_pos;break;
+    case 3:pri = pri3_pos;break;
+    }
+
+    sx = x << 1;
+    if(bg_enabled == true && !wt_main[sx]) {
+      if(pixel_cache[sx].pri_main < pri) {
+        pixel_cache[sx].pri_main = pri;
+        pixel_cache[sx].bg_main  = PC_OAM;
+        pixel_cache[sx].src_main = get_palette(oam_line_pal[x]);
+        pixel_cache[sx].color_exempt = (oam_line_pal[x] < 192);
+      }
+    }
+
+    if(bgsub_enabled == true && !wt_sub[sx]) {
+      if(pixel_cache[sx].pri_sub < pri) {
+        pixel_cache[sx].pri_sub = pri;
+        pixel_cache[sx].bg_sub  = PC_OAM;
+        pixel_cache[sx].src_sub = get_palette(oam_line_pal[x]);
+      }
+    }
+
+    sx++;
+    if(bg_enabled == true && !wt_main[sx]) {
+      if(pixel_cache[sx].pri_main < pri) {
+        pixel_cache[sx].pri_main = pri;
+        pixel_cache[sx].bg_main  = PC_OAM;
+        pixel_cache[sx].src_main = get_palette(oam_line_pal[x]);
+        pixel_cache[sx].color_exempt = (oam_line_pal[x] < 192);
+      }
+    }
+
+    if(bgsub_enabled == true && !wt_sub[sx]) {
+      if(pixel_cache[sx].pri_sub < pri) {
+        pixel_cache[sx].pri_sub = pri;
+        pixel_cache[sx].bg_sub  = PC_OAM;
+        pixel_cache[sx].src_sub = get_palette(oam_line_pal[x]);
       }
     }
   }
diff --git a/src/ppu/ppu.cpp b/src/ppu/ppu.cpp
index f93b0fbf..e4339a6c 100644
--- a/src/ppu/ppu.cpp
+++ b/src/ppu/ppu.cpp
@@ -2,14 +2,16 @@
 
 void PPU::get_scanline_info(scanline_info *info) {
   info->hires     = scanline_is_hires();
-  info->interlace = cpu->interlace();
+  info->interlace = r_cpu->interlace();
 }
 
 void PPU::enable_renderer(bool r) { status.render_output = r; }
 bool PPU::renderer_enabled() { return status.render_output; }
 
 void PPU::frame() {
-static fr = 0, fe = 0;
+  status.frame_executed = true;
+
+static int32 fr = 0, fe = 0;
 static time_t prev, curr;
   fe++;
   if(status.render_output)fr++;
@@ -24,7 +26,16 @@ static time_t prev, curr;
   prev = curr;
 }
 
+void PPU::power() {}
+
+void PPU::reset() {
+  memset(output, 0, 512 * 480 * sizeof(uint16));
+}
+
 PPU::PPU() {
+  output = (uint16*)malloc(512 * 480 * sizeof(uint16));
+  memset(output, 0, 512 * 480 * sizeof(uint16));
+
   status.render_output   = true;
   status.frames_updated  = false;
   status.frames_rendered = 0;
@@ -35,4 +46,6 @@ PPU::PPU() {
   mmio = &mmio_unmapped;
 }
 
-PPU::~PPU() {}
+PPU::~PPU() {
+  zerofree(output);
+}
diff --git a/src/ppu/ppu.h b/src/ppu/ppu.h
index 7065473d..6def0e52 100644
--- a/src/ppu/ppu.h
+++ b/src/ppu/ppu.h
@@ -1,11 +1,13 @@
 class PPU {
 public:
+uint16 *output;
 
 //this struct should be read-only to
 //functions outside of this class
 struct {
   bool   render_output;
 
+  bool   frame_executed;
   bool   frames_updated;
   uint32 frames_rendered;
   uint32 frames_executed;
@@ -43,8 +45,8 @@ struct scanline_info {
   virtual void  scanline() = 0;
   virtual void  render_scanline() = 0;
   virtual void  frame();
-  virtual void  power() = 0;
-  virtual void  reset() = 0;
+  virtual void  power();
+  virtual void  reset();
   virtual void  enable_renderer(bool r);
   virtual bool  renderer_enabled();
 
diff --git a/src/sdl/Makefile b/src/sdl/Makefile
index 16b9d601..c80c0168 100644
--- a/src/sdl/Makefile
+++ b/src/sdl/Makefile
@@ -5,7 +5,7 @@ OBJS = sdlmain.o \
        reader.o \
        memory.o bmemory.o \
        cpu.o bcpu.o \
-       apu.o bapu.o bapuskip.o \
+       apu.o bapu.o \
        bdsp.o \
        ppu.o bppu.o \
        snes.o \
@@ -58,9 +58,6 @@ apu.o: ../apu/*
 bapu.o: ../apu/bapu/*
 	$(CC) $(CFLAGS) -c ../apu/bapu/bapu.cpp
 
-bapuskip.o: ../apu/bapuskip/*
-	$(CC) $(CFLAGS) -c ../apu/bapuskip/bapuskip.cpp
-
 ###########
 ### dsp ###
 ###########
diff --git a/src/sdl/Makefile.win32 b/src/sdl/Makefile.win32
index 1bc07ca4..3f83052b 100644
--- a/src/sdl/Makefile.win32
+++ b/src/sdl/Makefile.win32
@@ -1,11 +1,11 @@
 CC = cl
-CFLAGS = /nologo /O2 /Ogityb2 /Gr /Gs /DARCH_LSB
+CFLAGS = /nologo /O2 /wd4996
 OBJS = sdlmain.obj \
        libstring.obj libconfig.obj \
        reader.obj \
        memory.obj bmemory.obj \
        cpu.obj bcpu.obj \
-       apu.obj bapu.obj bapuskip.obj \
+       apu.obj bapu.obj \
        bdsp.obj \
        ppu.obj bppu.obj \
        snes.obj \
@@ -59,9 +59,6 @@ apu.obj: ../apu/*
 bapu.obj: ../apu/bapu/*
 	$(CC) $(CFLAGS) /c ../apu/bapu/bapu.cpp
 
-bapuskip.obj: ../apu/bapuskip/*
-	$(CC) $(CFLAGS) /c ../apu/bapuskip/bapuskip.cpp
-
 ###########
 ### dsp ###
 ###########
diff --git a/src/sdl/bsnes.cpp b/src/sdl/bsnes.cpp
index 1c2482b0..11188844 100644
--- a/src/sdl/bsnes.cpp
+++ b/src/sdl/bsnes.cpp
@@ -6,11 +6,7 @@ void bSNES::run() {
 
   switch(run_status) {
   case RUN:
-    while(update_frame == false) {
-      SNES::run();
-    }
-    update_frame = false;
-    render();
+    SNES::runtoframe();
     return;
   case STOP:
     break;
@@ -18,13 +14,13 @@ void bSNES::run() {
 }
 
 void bSNES::video_run() {
-  if(ppu->status.frames_updated) {
+  if(r_ppu->status.frames_updated) {
   char s[512], t[512];
-    ppu->status.frames_updated = false;
+    r_ppu->status.frames_updated = false;
 //  if((bool)config::gui.show_fps == true) {
-      sprintf(s, "%s : %d fps", BSNES_TITLE, ppu->status.frames_executed);
+      sprintf(s, "%s : %d fps", BSNES_TITLE, r_ppu->status.frames_executed);
 //    if(w_main->frameskip != 0) {
-//      sprintf(t, " (%d frames)", ppu->status.frames_rendered);
+//      sprintf(t, " (%d frames)", r_ppu->status.frames_rendered);
 //      strcat(s, t);
 //    }
       SDL_WM_SetCaption(s, 0);
@@ -34,7 +30,28 @@ void bSNES::video_run() {
   render();
 }
 
-void bSNES::sound_run() {}
+void bSNES::sound_run(uint32 data) {}
+
+uint16 *bSNES::video_lock(uint32 &pitch) {
+  if(SDL_MUSTLOCK(screen)) {
+    SDL_LockSurface(screen);
+  }
+  if(SDL_MUSTLOCK(backbuffer)) {
+    SDL_LockSurface(backbuffer);
+  }
+
+  pitch = backbuffer->pitch;
+  return (uint16*)backbuffer->pixels;
+}
+
+void bSNES::video_unlock() {
+  if(SDL_MUSTLOCK(backbuffer)) {
+    SDL_UnlockSurface(backbuffer);
+  }
+  if(SDL_MUSTLOCK(screen)) {
+    SDL_UnlockSurface(screen);
+  }
+}
 
 /***********************
  *** Input functions ***
@@ -93,15 +110,8 @@ bJoypad::bJoypad() {
   l = r = select = start = false;
 }
 
-void bSNES::notify(uint32 message, uint32 param1, uint32 param2) {
-  switch(message) {
-  case RENDER_FRAME:
-    update_frame = true;
-    break;
-  }
-}
+void bSNES::notify(uint32 message, uint32 param1, uint32 param2) {}
 
 bSNES::bSNES() {
-  run_status   = STOP;
-  update_frame = false;
+  run_status = STOP;
 }
diff --git a/src/sdl/bsnes.h b/src/sdl/bsnes.h
index de5af083..74a08e18 100644
--- a/src/sdl/bsnes.h
+++ b/src/sdl/bsnes.h
@@ -9,17 +9,19 @@ bool l, r, select, start;
 class bSNES : public SNES {
 private:
 uint32 run_status;
-bool update_frame;
 bJoypad joypad1, joypad2;
 
 public:
 enum { STOP = 0, RUN };
-  void   run();
-  void   video_run();
-  void   sound_run();
+  void    run();
+  void    video_run();
+  void    sound_run(uint32 data);
 
-  void   set_status(uint32 new_status);
-  uint32 get_status();
+  uint16 *video_lock(uint32 &data);
+  void    video_unlock();
+
+  void    set_status(uint32 new_status);
+  uint32  get_status();
 
 //input functions
   void poll_input(uint8 type);
diff --git a/src/sdl/clean.bat b/src/sdl/clean.bat
index 81715981..74d64470 100644
--- a/src/sdl/clean.bat
+++ b/src/sdl/clean.bat
@@ -1 +1,3 @@
 @nmake /NOLOGO /f Makefile.win32 clean
+@del output.wav
+@del bsnes_sdl.exe
diff --git a/src/sdl/render.cpp b/src/sdl/render.cpp
index 38d4da84..9848f18a 100644
--- a/src/sdl/render.cpp
+++ b/src/sdl/render.cpp
@@ -1,59 +1,11 @@
-void render16() {
-uint16 *dest, *src;
-uint32 pitch;
-int x, y;
-
+void render() {
 SNES::video_info vi;
   snes->get_video_info(&vi);
 
-  pitch = (backbuffer->pitch >> 1);
-  dest  = (uint16*)backbuffer->pixels;
-  src   = (uint16*)vi.data;
-
-  if(vi.width == 256 && vi.height == 224) {
-    for(y=0;y<224;y++) {
-      memcpy(dest, src, 512);
-      dest += pitch;
-      src  += 256;
-    }
-  } else if(vi.width == 512 && vi.height == 224) {
-    for(y=0;y<224;y++) {
-      memcpy(dest, src, 1024);
-      dest += pitch;
-      src  += 512;
-    }
-  } else if(vi.width == 256 && vi.height == 448) {
-    for(y=0;y<448;y++) {
-      memcpy(dest, src, 512);
-      dest += pitch;
-      src  += 256;
-    }
-  } else if(vi.width == 512 && vi.height == 448) {
-    for(y=0;y<448;y++) {
-      memcpy(dest, src, 1024);
-      dest += pitch;
-      src  += 512;
-    }
-  }
-
-
   screen_info.rs.x = 0;
   screen_info.rs.y = (vi.height == 224) ?   1 :   2;
   screen_info.rs.w = vi.width;
   screen_info.rs.h = (vi.height == 224) ? 223 : 446;
-}
-
-void render32() {}
-
-void render() {
-  if(SDL_MUSTLOCK(screen)) {
-    SDL_LockSurface(screen);
-  }
-  if(SDL_MUSTLOCK(backbuffer)) {
-    SDL_LockSurface(backbuffer);
-  }
-
-  render16();
 
 //documentation says not to use this, but it's rather ridiculous that a graphics
 //library wouldn't support simple image scaling... so let's use it anyway and see
@@ -61,12 +13,5 @@ void render() {
   SDL_SoftStretch(backbuffer, &screen_info.rs, screen, &screen_info.rd);
 //SDL_BlitSurface(backbuffer, &screen_info.rs, screen, &screen_info.rd);
 
-  if(SDL_MUSTLOCK(backbuffer)) {
-    SDL_UnlockSurface(backbuffer);
-  }
-  if(SDL_MUSTLOCK(screen)) {
-    SDL_UnlockSurface(screen);
-  }
-
   SDL_UpdateRect(screen, screen_info.rd.x, screen_info.rd.y, screen_info.rd.w, screen_info.rd.h);
 }
diff --git a/src/sdl/rom.cpp b/src/sdl/rom.cpp
index f5beae7a..46897fcb 100644
--- a/src/sdl/rom.cpp
+++ b/src/sdl/rom.cpp
@@ -27,14 +27,14 @@ FileReader *rf = new FileReader();
     alert("Error loading image file [%s]!", rom_fn);
     return false;
   }
-  mem_bus->load_cart(static_cast<Reader*>(rf));
+  r_mem->load_cart(static_cast<Reader*>(rf));
   rf->close();
 
 CartInfo ci;
-  mem_bus->get_cartinfo(&ci);
+  r_mem->get_cartinfo(&ci);
   if(ci.sram_size != 0) {
     rf->open(sram_fn);
-    mem_bus->load_sram(static_cast<Reader*>(rf));
+    r_mem->load_sram(static_cast<Reader*>(rf));
     rf->close();
   }
 
@@ -49,25 +49,25 @@ void ROMImage::unload() {
 
 FileWriter *wf;
 CartInfo ci;
-  mem_bus->get_cartinfo(&ci);
+  r_mem->get_cartinfo(&ci);
   if(ci.sram_size != 0) {
     wf = new FileWriter();
     wf->open(sram_fn);
-    mem_bus->save_sram(static_cast<Writer*>(wf));
+    r_mem->save_sram(static_cast<Writer*>(wf));
     wf->close();
     delete(wf);
   }
 
   file_loaded = false;
 
-  mem_bus->unload_cart();
+  r_mem->unload_cart();
 }
 
 void ROMImage::select(char *fn) {
 int i;
   if(file_loaded == true)return;
 
-/* Remove quotes */
+//remove quotes
   if(fn[0] == '\"') {
     strcpy(rom_fn, fn + 1);
     rom_fn[strlen(rom_fn) - 1] = 0;
diff --git a/src/sdl/sdlmain.cpp b/src/sdl/sdlmain.cpp
index 76d69354..d371da97 100644
--- a/src/sdl/sdlmain.cpp
+++ b/src/sdl/sdlmain.cpp
@@ -44,13 +44,15 @@ va_list args;
 }
 
 void init_snes() {
-  mem_bus = new bMemBus();
-  cpu     = new bCPU();
-  apu     = new bAPU();
-  dsp     = new bDSP();
-  ppu     = new bPPU();
-  snes    = new bSNES();
-  bsnes   = static_cast<bSNES*>(snes);
+#ifdef POLYMORPHISM
+  deref(mem) = new bMemBus();
+  deref(cpu) = new bCPU();
+  deref(apu) = new bAPU();
+  deref(dsp) = new bDSP();
+  deref(ppu) = new bPPU();
+#endif
+  snes  = new bSNES();
+  bsnes = static_cast<bSNES*>(snes);
 
   snes->init();
 
@@ -59,18 +61,18 @@ void init_snes() {
 //play audio in real-time while sound output
 //isn't available.
   snes->log_audio_enable("output.wav");
-
-  snes->set_playback_buffer_size(2000);
 }
 
 void term_snes() {
   snes->term();
-
-  if(mem_bus) { delete(static_cast<bMemBus*>(mem_bus)); mem_bus = 0; }
-  if(cpu)     { delete(static_cast<bCPU*>   (cpu));     cpu     = 0; }
-  if(apu)     { delete(static_cast<bAPU*>   (apu));     apu     = 0; }
-  if(ppu)     { delete(static_cast<bPPU*>   (ppu));     ppu     = 0; }
-  if(snes)    { delete(static_cast<bSNES*>  (snes));    snes    = 0; }
+#ifdef POLYMORPHISM
+  if(deref(mem)) { delete static_cast<bMemBus*>(deref(mem)); deref(mem) = 0; }
+  if(deref(cpu)) { delete static_cast<bCPU*>   (deref(cpu)); deref(cpu) = 0; }
+  if(deref(apu)) { delete static_cast<bAPU*>   (deref(apu)); deref(apu) = 0; }
+  if(deref(dsp)) { delete static_cast<bDSP*>   (deref(dsp)); deref(dsp) = 0; }
+  if(deref(ppu)) { delete static_cast<bPPU*>   (deref(ppu)); deref(ppu) = 0; }
+#endif
+  if(snes) { delete(static_cast<bSNES*>(snes)); snes = 0; }
 }
 
 void center_window() {
@@ -133,7 +135,7 @@ SDL_Event event;
   atexit(SDL_Quit);
   set_window_info();
   screen = SDL_SetVideoMode(config::video.display_width, config::video.display_height, 16,
-    SDL_SWSURFACE | ((config::video.fullscreen)?SDL_FULLSCREEN:0));
+    SDL_SWSURFACE | ((config::video.fullscreen) ? SDL_FULLSCREEN : 0));
   if(!screen)     { alert("Failed to initialize SDL"); goto _end; }
   backbuffer = SDL_CreateRGBSurface(SDL_SWSURFACE, 512, 448, 16, 0xf800, 0x07e0, 0x001f, 0x0000);
   if(!backbuffer) { alert("Failed to initialize SDL"); goto _end; }
@@ -160,7 +162,7 @@ int cursor_status;
           snes->capture_screenshot();
           break;
         case SDLK_F10: //toggle cursor display
-          cursor_status = (SDL_ShowCursor(SDL_QUERY) == SDL_ENABLE)?SDL_DISABLE:SDL_ENABLE;
+          cursor_status = (SDL_ShowCursor(SDL_QUERY) == SDL_ENABLE) ? SDL_DISABLE : SDL_ENABLE;
           SDL_ShowCursor(cursor_status);
           break;
         case SDLK_F11: //only supported on X11
diff --git a/src/snes/snes.cpp b/src/snes/snes.cpp
index 9370b905..04ca776e 100644
--- a/src/snes/snes.cpp
+++ b/src/snes/snes.cpp
@@ -5,13 +5,12 @@
 #include "snes_input.cpp"
 
 void SNES::run() {
-uint32 cycles, r;
   if(apusync.cycles < 0) {
-    cpu->run();
-    apusync.cycles += apusync.apu_multbl[cpu->cycles_executed()];
+    r_cpu->run();
+    apusync.cycles += apusync.apu_multbl[r_cpu->cycles_executed()];
   } else {
-    apu->run();
-    cycles = apu->cycles_executed();
+    r_apu->run();
+  uint32 cycles = r_apu->cycles_executed();
     apusync.dsp += cycles;
     apusync.cycles -= apusync.cpu_multbl[cycles];
 
@@ -19,11 +18,18 @@ uint32 cycles, r;
   //24576000(Sound clock crystal) / 32000(DSP) = 768crystal/dsp ticks
     while(apusync.dsp >= 768) {
       apusync.dsp -= 768;
-      audio_update(dsp->run());
+      audio_update(r_dsp->run());
     }
   }
 }
 
+void SNES::runtoframe() {
+  while(r_ppu->status.frame_executed == false) {
+    SNES::run();
+  }
+  r_ppu->status.frame_executed = false;
+}
+
 void SNES::init() {
   srtc = new SRTC();
   sdd1 = new SDD1();
@@ -40,48 +46,44 @@ void SNES::term() {
 }
 
 void SNES::power() {
-  cpu->power();
-  apu->power();
-  dsp->power();
-  ppu->power();
-  mem_bus->power();
+  r_cpu->power();
+  r_apu->power();
+  r_dsp->power();
+  r_ppu->power();
+  r_mem->power();
 
   srtc->power();
   sdd1->power();
 
 int i;
-  mem_bus->flush_mmio_mappers();
-  for(i=0x2100;i<=0x213f;i++)mem_bus->set_mmio_mapper(i, ppu->mmio);
-  for(i=0x2140;i<=0x217f;i++)mem_bus->set_mmio_mapper(i, cpu->mmio);
-  for(i=0x2180;i<=0x2183;i++)mem_bus->set_mmio_mapper(i, cpu->mmio);
+  r_mem->flush_mmio_mappers();
+  for(i=0x2100;i<=0x213f;i++)r_mem->set_mmio_mapper(i, r_ppu->mmio);
+  for(i=0x2140;i<=0x217f;i++)r_mem->set_mmio_mapper(i, r_cpu->mmio);
+  for(i=0x2180;i<=0x2183;i++)r_mem->set_mmio_mapper(i, r_cpu->mmio);
 //input
-  mem_bus->set_mmio_mapper(0x4016, cpu->mmio);
-  mem_bus->set_mmio_mapper(0x4017, cpu->mmio);
-  for(i=0x4200;i<=0x421f;i++)mem_bus->set_mmio_mapper(i, cpu->mmio);
-  for(i=0x4300;i<=0x437f;i++)mem_bus->set_mmio_mapper(i, cpu->mmio);
+  r_mem->set_mmio_mapper(0x4016, r_cpu->mmio);
+  r_mem->set_mmio_mapper(0x4017, r_cpu->mmio);
+  for(i=0x4200;i<=0x421f;i++)r_mem->set_mmio_mapper(i, r_cpu->mmio);
+  for(i=0x4300;i<=0x437f;i++)r_mem->set_mmio_mapper(i, r_cpu->mmio);
 
   srtc->enable();
   sdd1->enable();
 
-  memset(video.data,     0, 512 * 448 * sizeof(uint32));
-  memset(video.ppu_data, 0, 512 * 480 * sizeof(uint16));
   video_update();
 }
 
 void SNES::reset() {
   apusync.cycles = -apusync.cpu_multbl[32];
 
-  cpu->reset();
-  apu->reset();
-  dsp->reset();
-  ppu->reset();
-  mem_bus->reset();
+  r_cpu->reset();
+  r_apu->reset();
+  r_dsp->reset();
+  r_ppu->reset();
+  r_mem->reset();
 
   srtc->reset();
   sdd1->reset();
 
-  memset(video.data,     0, 512 * 448 * sizeof(uint32));
-  memset(video.ppu_data, 0, 512 * 480 * sizeof(uint16));
   video_update();
 }
 
@@ -154,6 +156,4 @@ SNES::SNES() {
 
   snes_region = NTSC;
   update_timing();
-
-  dsp_buffer.data = 0;
 }
diff --git a/src/snes/snes.h b/src/snes/snes.h
index 4524c304..db2a31f5 100644
--- a/src/snes/snes.h
+++ b/src/snes/snes.h
@@ -19,7 +19,8 @@ public:
 enum { NTSC = 0, PAL = 1 };
 
 //system functions
-  virtual void run();
+  virtual inline void run();
+  virtual inline void runtoframe();
   virtual void init();
   virtual void term();
   virtual void power();
@@ -39,7 +40,6 @@ enum { NTSC = 0, PAL = 1 };
 //debugging functions
 enum {
   NO_ACTION = 0,
-  RENDER_FRAME,
   CPU_EXEC_OPCODE_BEGIN, CPU_EXEC_OPCODE_END,
   APU_EXEC_OPCODE_BEGIN, APU_EXEC_OPCODE_END,
   MEM_READ,    MEM_WRITE,
diff --git a/src/snes/snes_audio.cpp b/src/snes/snes_audio.cpp
index 5f10e603..eb2787e2 100644
--- a/src/snes/snes_audio.cpp
+++ b/src/snes/snes_audio.cpp
@@ -1,25 +1,3 @@
-void SNES::set_playback_buffer_size(uint32 buffer_size) {
-  if(dsp_buffer.data) {
-    free(dsp_buffer.data);
-    dsp_buffer.data = 0;
-  }
-
-//* 2 is for left/right channel data
-  dsp_buffer.data = (uint16*)malloc(buffer_size * sizeof(uint16) * 2);
-  memset(dsp_buffer.data, 0, buffer_size * sizeof(uint16) * 2);
-
-  dsp_buffer.size = buffer_size;
-  dsp_buffer.pos  = 0;
-}
-
-uint32 SNES::get_playback_buffer_pos() {
-  return dsp_buffer.pos;
-}
-
-uint16 *SNES::get_playback_buffer() {
-  return dsp_buffer.data;
-}
-
 void SNES::audio_update(uint32 data) {
   if(pcmfp) {
     fputc(data,       pcmfp);
@@ -30,11 +8,7 @@ void SNES::audio_update(uint32 data) {
 
   if((bool)config::snes.mute == true)data = 0x0000;
 
-  dsp_buffer.data[dsp_buffer.pos++] = (data)       & 0xffff;
-  dsp_buffer.data[dsp_buffer.pos++] = (data >> 16) & 0xffff;
-  dsp_buffer.pos %= dsp_buffer.size;
-
-  sound_run();
+  sound_run(data);
 }
 
 void SNES::log_audio_enable(const char *fn) {
diff --git a/src/snes/snes_audio.h b/src/snes/snes_audio.h
index 4ba04c88..1442e57c 100644
--- a/src/snes/snes_audio.h
+++ b/src/snes/snes_audio.h
@@ -1,15 +1,5 @@
-struct {
-  uint16 *data;
-  uint32  size, pos;
-} dsp_buffer;
-
 FILE *pcmfp;
 
-//buffer_size is in samples
-  void    set_playback_buffer_size(uint32 buffer_size);
-  uint32  get_playback_buffer_pos();
-  uint16 *get_playback_buffer();
-
 //if a filename is not specified, one will be generated
 //automatically ("audio%0.3d.wav")
   void    log_audio_enable(const char *fn = 0);
@@ -19,4 +9,4 @@ FILE *pcmfp;
   void    audio_init();
   void    audio_term();
 
-  virtual void sound_run() = 0;
+  virtual void sound_run(uint32 data) = 0;
diff --git a/src/snes/snes_video.cpp b/src/snes/snes_video.cpp
index eb79e5f6..dbdd1bc7 100644
--- a/src/snes/snes_video.cpp
+++ b/src/snes/snes_video.cpp
@@ -8,7 +8,7 @@ const uint8 SNES::color_curve_table[32] = {
 };
 
 void SNES::update_color_lookup_table() {
-int i, l, r, g, b;
+int32 i, l, r, g, b;
 double lr = 0.2126, lg = 0.7152, lb = 0.0722; //luminance
 uint32 col;
 
@@ -29,7 +29,7 @@ uint32 col;
     }
 
     if((int)config::snes.video_color_adjust_mode == VCA_GRAYSCALE) {
-      l = int((double)r * lr) + ((double)g * lg) + ((double)b * lb);
+      l = int32(((double)r * lr) + ((double)g * lg) + ((double)b * lb));
       if(l <   0)l =   0;
       if(l > 255)l = 255;
       r = g = b = l;
@@ -92,7 +92,6 @@ void SNES::update_video_format() {
 }
 
 void SNES::get_video_info(video_info *info) {
-  info->data = video.data;
   info->mode = video.mode;
 
   switch(video.mode) {
@@ -129,10 +128,10 @@ void SNES::get_video_info(video_info *info) {
   }
 }
 
-void SNES::video_update_256x224(uint16 *src) {
+void SNES::video_update_256x224() {
 int x, y;
-uint16 *dest;
-  dest = video.data;
+uint16 *src  = video.ppu_data;
+uint16 *dest = video.data;
 
   for(y=0;y<224;y++) {
     if(video_frame[y].hires == false) {
@@ -147,13 +146,14 @@ uint16 *dest;
       }
       src += 512;
     }
+    dest += video.pitch - 256;
   }
 }
 
-void SNES::video_update_512x224(uint16 *src) {
+void SNES::video_update_512x224() {
 int x, y;
-uint16 *dest;
-  dest = video.data;
+uint16 *src  = video.ppu_data;
+uint16 *dest = video.data;
 
   for(y=0;y<224;y++) {
     if(video_frame[y].hires == false) {
@@ -168,31 +168,41 @@ uint16 *dest;
       }
       src += 512;
     }
+    dest += video.pitch - 512;
   }
 }
 
-void SNES::video_update_256x448(uint16 *src) {
+void SNES::video_update_256x448() {
 int x, y;
-uint16 *dest;
-bool field = !cpu->interlace_field();
-  dest = video.data;
+uint16 *src   = video.ppu_data;
+uint16 *dest  = video.data;
+bool    field = !r_cpu->interlace_field();
 
   for(y=0;y<224;y++) {
     if(video_frame[y].interlace == false) {
       if(video_frame[y].hires == false) {
-        for(x=0;x<512;x++) {
-          *dest++ = color_lookup_table[*(src + (uint8)x)];
+        for(x=0;x<256;x++) {
+          *dest++ = color_lookup_table[*(src + x)];
         }
-        src += 1024;
+        dest += video.pitch - 256;
+        for(x=0;x<256;x++) {
+          *dest++ = color_lookup_table[*(src + x)];
+        }
+        dest += video.pitch - 256;
       } else {
-        for(x=0;x<512;x++) {
-          *dest++ = color_lookup_table[*(src + ((x & 255) << 1))];
+        for(x=0;x<256;x++) {
+          *dest++ = color_lookup_table[*(src + (x << 1))];
         }
-        src += 1024;
+        dest += video.pitch - 256;
+        for(x=0;x<256;x++) {
+          *dest++ = color_lookup_table[*(src + (x << 1))];
+        }
+        dest += video.pitch - 256;
       }
+      src += 1024;
     } else {
       if(field) {
-        dest += 256;
+        dest += video.pitch;
         src  += 512;
       }
 
@@ -207,38 +217,49 @@ bool field = !cpu->interlace_field();
           src += 2;
         }
       }
+      dest += video.pitch - 256;
 
       if(!field) {
-        dest += 256;
+        dest += video.pitch;
         src  += 512;
       }
     }
   }
 }
 
-void SNES::video_update_512x448(uint16 *src) {
+void SNES::video_update_512x448() {
 int x, y;
-uint16 *dest;
-bool field = !cpu->interlace_field();
-  dest = video.data;
+uint16 *src   = video.ppu_data;
+uint16 *dest  = video.data;
+bool    field = !r_cpu->interlace_field();
 
   for(y=0;y<224;y++) {
     if(video_frame[y].interlace == false) {
       if(video_frame[y].hires == false) {
-        for(x=0;x<512;x++) {
-          *dest++ = color_lookup_table[*(src + (uint8)x)];
-          *dest++ = color_lookup_table[*(src + (uint8)x)];
+        for(x=0;x<256;x++) {
+          *dest++ = color_lookup_table[*(src + x)];
+          *dest++ = color_lookup_table[*(src + x)];
         }
-        src += 1024;
+        dest += video.pitch - 512;
+        for(x=0;x<256;x++) {
+          *dest++ = color_lookup_table[*(src + x)];
+          *dest++ = color_lookup_table[*(src + x)];
+        }
+        dest += video.pitch - 512;
       } else {
-        for(x=0;x<1024;x++) {
-          *dest++ = color_lookup_table[*(src + (x & 511))];
+        for(x=0;x<512;x++) {
+          *dest++ = color_lookup_table[*(src + x)];
         }
-        src += 1024;
+        dest += video.pitch - 512;
+        for(x=0;x<512;x++) {
+          *dest++ = color_lookup_table[*(src + x)];
+        }
+        dest += video.pitch - 512;
       }
+      src += 1024;
     } else {
       if(field) {
-        dest += 512;
+        dest += video.pitch;
         src  += 512;
       }
 
@@ -253,9 +274,10 @@ bool field = !cpu->interlace_field();
           *dest++ = color_lookup_table[*src++];
         }
       }
+      dest += video.pitch - 512;
 
       if(!field) {
-        dest += 512;
+        dest += video.pitch;
         src  += 512;
       }
     }
@@ -263,25 +285,30 @@ bool field = !cpu->interlace_field();
 }
 
 void SNES::video_update() {
-  if(ppu->renderer_enabled()) {
+  if(r_ppu->renderer_enabled()) {
     if(video.format_changed == true) {
       update_video_format();
     }
 
-  uint16 *src = (uint16*)video.ppu_data + ((int(cpu->overscan()) << 3) * 1024);
-    switch(video.mode) {
-    case VM_256x224:video_update_256x224(src);break;
-    case VM_512x224:video_update_512x224(src);break;
-    case VM_256x448:video_update_256x448(src);break;
-    case VM_512x448:video_update_512x448(src);break;
-    case VM_VARIABLE:
-      switch(int(video.frame_hires) | (int(video.frame_interlace) << 1)) {
-      case 0:video_update_256x224(src);break;
-      case 1:video_update_512x224(src);break;
-      case 2:video_update_256x448(src);break;
-      case 3:video_update_512x448(src);break;
+    video.data     = (uint16*)video_lock(video.pitch);
+    video.ppu_data = (uint16*)r_ppu->output + (int(r_cpu->overscan()) << 13);
+    video.pitch  >>= 1;
+    if(video.data) {
+      switch(video.mode) {
+      case VM_256x224:video_update_256x224();break;
+      case VM_512x224:video_update_512x224();break;
+      case VM_256x448:video_update_256x448();break;
+      case VM_512x448:video_update_512x448();break;
+      case VM_VARIABLE:
+        switch(int(video.frame_hires) | (int(video.frame_interlace) << 1)) {
+        case 0:video_update_256x224();break;
+        case 1:video_update_512x224();break;
+        case 2:video_update_256x448();break;
+        case 3:video_update_512x448();break;
+        }
+        break;
       }
-      break;
+      video_unlock();
     }
 
   //SNES::capture_screenshot() was called by emulation interface
@@ -298,13 +325,13 @@ void SNES::video_update() {
 }
 
 void SNES::video_scanline() {
-int y = cpu->vcounter();
-int o = int(cpu->overscan()) << 3;
+int y = r_cpu->vcounter();
+int o = int(r_cpu->overscan()) << 3;
   if(y <= (0 + o) || y >= (224 + o))return;
   y -= o;
 
 PPU::scanline_info si;
-  ppu->get_scanline_info(&si);
+  r_ppu->get_scanline_info(&si);
 
   video_frame[y].hires     = si.hires;
   video_frame[y].interlace = si.interlace;
@@ -313,21 +340,10 @@ PPU::scanline_info si;
   video.frame_interlace   |= si.interlace;
 }
 
-uint16 *SNES::get_ppu_output_handle() {
-  return (uint16*)(video.ppu_data +
-    (cpu->vcounter() * 1024) +
-    ((cpu->interlace() && cpu->interlace_field())?512:0));
-}
-
 void SNES::video_init() {
 int i, c;
   video.format_changed = false;
 
-  video.data     = (uint16*)malloc(512 * 448 * sizeof(uint32));
-  video.ppu_data = (uint16*)malloc(512 * 480 * sizeof(uint16));
-  memset(video.data,     0, 512 * 448 * sizeof(uint32));
-  memset(video.ppu_data, 0, 512 * 480 * sizeof(uint16));
-
   for(i=0;i<224;i++) {
     video_frame[i].hires     = false;
     video_frame[i].interlace = false;
diff --git a/src/snes/snes_video.h b/src/snes/snes_video.h
index df89b25d..02c1eb73 100644
--- a/src/snes/snes_video.h
+++ b/src/snes/snes_video.h
@@ -29,6 +29,7 @@ struct {
   uint16 *data, *ppu_data;
   uint8   mode;
   uint8   depth;
+  uint32  pitch;
 
   bool    frame_hires, frame_interlace;
 
@@ -40,21 +41,20 @@ struct {
 } video_frame[224];
 
 struct video_info {
-  uint16 *data;
-  uint8   mode;
-  uint32  width, height;
+  uint32  mode, width, height;
 };
 
 //public functions
   void capture_screenshot();
   void update_color_lookup_table();
 
-  virtual void  set_video_format(uint8 mode, uint8 depth);
-  virtual void  get_video_info(video_info *info);
-  virtual void  video_run() = 0;
+  virtual void    set_video_format(uint8 mode, uint8 depth);
+  virtual void    get_video_info(video_info *info);
+  virtual void    video_run() = 0;
+  virtual uint16 *video_lock(uint32 &pitch) = 0;
+  virtual void    video_unlock() = 0;
 
 //private functions
-  uint16 *get_ppu_output_handle(); //used by PPU only
 private:
 //when a screenshot is requested, wait until the frame
 //has finished rendering, so we can tell the image size
@@ -62,10 +62,10 @@ bool flag_output_screenshot;
   uint16 to_rgb555(uint32 color);
   void   output_screenshot();
   void   update_video_format();
-  void   video_update_256x224(uint16 *src);
-  void   video_update_512x224(uint16 *src);
-  void   video_update_256x448(uint16 *src);
-  void   video_update_512x448(uint16 *src);
+  void   video_update_256x224();
+  void   video_update_512x224();
+  void   video_update_256x448();
+  void   video_update_512x448();
   void   video_update();
   void   video_scanline();
   void   video_init();
diff --git a/src/snes/snes_video_ex.cpp b/src/snes/snes_video_ex.cpp
index 7a899f72..4de1eada 100644
--- a/src/snes/snes_video_ex.cpp
+++ b/src/snes/snes_video_ex.cpp
@@ -5,25 +5,13 @@ void SNES::capture_screenshot() {
 //used to convert pixel data to write to rgb555 format
 //bitmap image via SNES::output_screenshot() function
 uint16 SNES::to_rgb555(uint32 color) {
-  if(video.depth == 15) {
-  //rgb555
-    return color & 0x7fff;
-  }
-
-  if(video.depth == 16) {
-  //rgb565->rgb555
-    return ((color >> 1) & 0x7fe0) | (color & 0x001f);
-  }
-
-  if(video.depth == 24 || video.depth == 32) {
-  //rgb888->rgb555
-    return ((color >> 9) & 0x7c00) | ((color >> 6) & 0x03e0) | ((color >> 3) & 0x001f);
-  }
-
-//unsupported color depth
-  return color;
+//bgr555->rgb555
+  return ((color & 0x7c00) >> 10) | (color & 0x03e0) | ((color & 0x001f) << 10);
 }
 
+//this routine isn't perfect... it will fail if the video frame
+//mixes resolutions (e.g. the top half is 256x224, and the bottom
+//half is 512x224, etc.)
 void SNES::output_screenshot() {
 FILE *fp;
 char fn[256];
@@ -95,7 +83,7 @@ int x, y;
 uint16 c;
   for(y=height;y>=1;y--) {
     for(x=0;x<width;x++) {
-      c = to_rgb555(video.data[y * width + x]);
+      c = to_rgb555(video.ppu_data[y * 1024 + x]);
       fputc(c,      fp);
       fputc(c >> 8, fp);
     }
diff --git a/src/win/Makefile b/src/win/Makefile
index e4e01a22..ec12082c 100644
--- a/src/win/Makefile
+++ b/src/win/Makefile
@@ -1,11 +1,11 @@
 CC = cl
-CFLAGS = /nologo /O2 /Ogityb2 /Gr /Gs /DARCH_LSB
+CFLAGS = /nologo /O2 /wd4996
 OBJS = winmain.obj \
        libstring.obj libconfig.obj \
        reader.obj \
        memory.obj bmemory.obj \
        cpu.obj bcpu.obj \
-       apu.obj bapu.obj bapuskip.obj \
+       apu.obj bapu.obj \
        bdsp.obj \
        ppu.obj bppu.obj \
        snes.obj \
@@ -15,9 +15,12 @@ LIBS = kernel32.lib user32.lib gdi32.lib comdlg32.lib ddraw.lib dsound.lib dxgui
 all: $(OBJS)
 	rc /r /fobsnes.res bsnes.rc
 	$(CC) /Febsnes.exe $(CFLAGS) $(OBJS) bsnes.res $(LIBS)
+#/link /PGD:bsnes.pgd /LTCG:PGOPTIMIZE
 
 clean:
 	del *.obj
+	del *.pgd
+	del *.pgc
 
 ######################
 ### win32-specific ###
@@ -60,9 +63,6 @@ apu.obj: ../apu/*
 bapu.obj: ../apu/bapu/*
 	$(CC) $(CFLAGS) /c ../apu/bapu/bapu.cpp
 
-bapuskip.obj: ../apu/bapuskip/*
-	$(CC) $(CFLAGS) /c ../apu/bapuskip/bapuskip.cpp
-
 ###########
 ### dsp ###
 ###########
diff --git a/src/win/bsnes.cpp b/src/win/bsnes.cpp
index 18c54999..826e1128 100644
--- a/src/win/bsnes.cpp
+++ b/src/win/bsnes.cpp
@@ -1,3 +1,12 @@
+void bSNES::power() {
+  ds_sound->init();
+  SNES::power();
+}
+
+void bSNES::reset() {
+  SNES::reset();
+}
+
 void bSNES::set_status(uint32 new_status) {
 uint8 cpu_op;
   run_status = new_status;
@@ -12,7 +21,7 @@ uint8 cpu_op;
     status.cpu_ran = false;
     break;
   case RUNTOCPUPROCEED:
-    cpu_op = mem_bus->read(cpu->regs.pc.d);
+    cpu_op = r_mem->read(r_cpu->regs.pc.d);
 
     if(cpu_op == 0x10 || //bpl rel
        cpu_op == 0x30 || //bmi rel
@@ -27,7 +36,7 @@ uint8 cpu_op;
        cpu_op == 0xfc    //jsr (addr,x)
     ) {
       w_console->is_running(true);
-      status.cpu_stop_pos = (cpu->regs.pc.b << 16) | ((cpu->regs.pc.d + cpu->opcode_length()) & 0xffff);
+      status.cpu_stop_pos = (r_cpu->regs.pc.b << 16) | ((r_cpu->regs.pc.d + r_cpu->opcode_length()) & 0xffff);
     } else {
       status.cpu_ran = false;
       run_status     = RUNTOCPUSTEP;
@@ -48,10 +57,7 @@ void bSNES::run() {
 
   switch(run_status) {
   case RUN:
-    while(update_frame == false) {
-      SNES::run();
-    }
-    update_frame = false;
+    SNES::runtoframe();
     return;
   case STOP:
     break;
@@ -68,8 +74,8 @@ void bSNES::run() {
     break;
   case RUNTOFRAME:
     SNES::run();
-    if(update_frame == true) {
-      update_frame = false;
+    if(r_ppu->status.frame_executed == true) {
+      r_ppu->status.frame_executed = false;
       set_status(STOP);
       disassemble_apu_op();
       disassemble_cpu_op();
@@ -89,7 +95,7 @@ void bSNES::run() {
     break;
   case RUNTOCPUPROCEED:
     SNES::run();
-    if(cpu->in_opcode() == false && status.cpu_stop_pos == cpu->regs.pc.d) {
+    if(r_cpu->in_opcode() == false && status.cpu_stop_pos == r_cpu->regs.pc.d) {
       set_status(STOP);
       disassemble_cpu_op();
     } else if(w_bp->hit() == true) {
@@ -114,13 +120,13 @@ void bSNES::run() {
 }
 
 void bSNES::video_run() {
-  if(ppu->status.frames_updated) {
+  if(r_ppu->status.frames_updated) {
   char s[512], t[512];
-    ppu->status.frames_updated = false;
+    r_ppu->status.frames_updated = false;
     if((bool)config::gui.show_fps == true) {
-      sprintf(s, "%s : %d fps", BSNES_TITLE, ppu->status.frames_executed);
+      sprintf(s, "%s : %d fps", BSNES_TITLE, r_ppu->status.frames_executed);
       if(w_main->frameskip != 0) {
-        sprintf(t, " (%d frames)", ppu->status.frames_rendered);
+        sprintf(t, " (%d frames)", r_ppu->status.frames_rendered);
         strcat(s, t);
       }
       SetWindowText(w_main->hwnd, s);
@@ -129,17 +135,43 @@ void bSNES::video_run() {
 
   w_main->frameskip_pos++;
   w_main->frameskip_pos %= (w_main->frameskip + 1);
-  if(ppu->renderer_enabled())dd_renderer->update();
-  ppu->enable_renderer(w_main->frameskip_pos == 0);
+  if(r_ppu->renderer_enabled())dd_renderer->update();
+  r_ppu->enable_renderer(w_main->frameskip_pos == 0);
 }
 
-void bSNES::sound_run() {
-  ds_sound->run();
+void bSNES::sound_run(uint32 data) {
+  ds_sound->run(data);
+}
+
+/***********************
+ *** Video functions ***
+ ***********************/
+uint16 *bSNES::video_lock(uint32 &pitch) {
+  return dd_renderer->lock(pitch);
+}
+
+void bSNES::video_unlock() {
+  dd_renderer->unlock();
 }
 
 /***********************
  *** Input functions ***
  ***********************/
+void bSNES::clear_input() {
+  joypad1.up     = joypad2.up =
+  joypad1.down   = joypad2.down =
+  joypad1.left   = joypad2.left =
+  joypad1.right  = joypad2.right =
+  joypad1.a      = joypad2.a =
+  joypad1.b      = joypad2.b =
+  joypad1.x      = joypad2.x =
+  joypad1.y      = joypad2.y =
+  joypad1.l      = joypad2.l =
+  joypad1.r      = joypad2.r =
+  joypad1.select = joypad2.select =
+  joypad1.start  = joypad2.start = 0;
+}
+
 void bSNES::poll_input(uint8 type) {
 //only capture input when main window has focus
   if(GetForegroundWindow() == w_main->hwnd) {
@@ -279,27 +311,27 @@ uint8  r = 0x00;
       if(a >= 0x2000 && a <= 0x5fff) {
         r = 0x00;
       } else {
-        r = mem_bus->read(addr);
+        r = r_mem->read(addr);
       }
     } else {
-      r = mem_bus->read(addr);
+      r = r_mem->read(addr);
     }
     break;
   case SPCRAM:
     addr &= 0xffff;
-    r = apu->spcram_read(addr);
+    r = r_apu->spcram_read(addr);
     break;
   case VRAM:
     addr &= 0xffff;
-    r = ppu->vram_read(addr);
+    r = r_ppu->vram_read(addr);
     break;
   case OAM:
     addr &= 0x03ff;
-    r = ppu->oam_read(addr);
+    r = r_ppu->oam_read(addr);
     break;
   case CGRAM:
     addr &= 0x01ff;
-    r = ppu->cgram_read(addr);
+    r = r_ppu->cgram_read(addr);
     break;
   }
   debug_command = false;
@@ -311,38 +343,31 @@ void bSNES::write(uint8 type, uint32 addr, uint8 value) {
   switch(type) {
   case DRAM:
     addr &= 0xffffff;
-    mem_bus->cart->write_protect(false);
-    mem_bus->write(addr, value);
-    mem_bus->cart->write_protect(true);
+    r_mem->cart->write_protect(false);
+    r_mem->write(addr, value);
+    r_mem->cart->write_protect(true);
     break;
   case SPCRAM:
     addr &= 0xffff;
-    apu->spcram_write(addr, value);
+    r_apu->spcram_write(addr, value);
     break;
   case VRAM:
     addr &= 0xffff;
-    ppu->vram_write(addr, value);
+    r_ppu->vram_write(addr, value);
     break;
   case OAM:
     addr &= 0x03ff;
-    ppu->oam_write(addr, value);
+    r_ppu->oam_write(addr, value);
     break;
   case CGRAM:
     addr &= 0x01ff;
-    ppu->cgram_write(addr, value);
+    r_ppu->cgram_write(addr, value);
     break;
   }
   debug_command = false;
 }
 
 void bSNES::notify(uint32 message, uint32 param1, uint32 param2) {
-//system messages
-  switch(message) {
-  case RENDER_FRAME:
-    update_frame = true;
-    break;
-  }
-
 //debugging messages
   if(is_debugger_enabled == false)return;
 
@@ -353,14 +378,14 @@ void bSNES::notify(uint32 message, uint32 param1, uint32 param2) {
     status.cpu_ran = true;
     status.cpu_trace_pos++;
   //test next opcode for breakpoint
-    w_bp->test(message, cpu->regs.pc.d, 0);
+    w_bp->test(message, r_cpu->regs.pc.d, 0);
     disassemble_cpu_op();
     break;
   case APU_EXEC_OPCODE_BEGIN:
     break;
   case APU_EXEC_OPCODE_END:
     status.apu_ran = true;
-    w_bp->test(message, apu->regs.pc, 0);
+    w_bp->test(message, r_apu->regs.pc, 0);
     disassemble_apu_op();
     break;
   case MEM_READ:
@@ -395,7 +420,7 @@ char t[512];
 //don't disassemble opcodes that won't be printed to console/traced to log anyway
   if(!w_console->can_write(Console::CPU_MESSAGE) && !w_console->tracing_enabled)return;
 
-  cpu->disassemble_opcode(t);
+  r_cpu->disassemble_opcode(t);
   w_console->write(t, Console::CPU_MESSAGE);
 }
 
@@ -405,7 +430,7 @@ char t[512];
 
   if(!w_console->can_write(Console::APU_MESSAGE) && !w_console->tracing_enabled)return;
 
-  apu->disassemble_opcode(t);
+  r_apu->disassemble_opcode(t);
   w_console->write(t, Console::APU_MESSAGE);
 }
 
@@ -540,7 +565,6 @@ uint32 i, style;
 bSNES::bSNES() {
   run_status    = STOP;
   debug_command = false;
-  update_frame  = false;
 
   debugger_disable();
 }
diff --git a/src/win/bsnes.h b/src/win/bsnes.h
index d0f3340a..054246ca 100644
--- a/src/win/bsnes.h
+++ b/src/win/bsnes.h
@@ -18,8 +18,6 @@ bool is_debugger_activated;
 bool debug_command;
 uint32 run_status;
 
-bool update_frame;
-
 bJoypad joypad1, joypad2;
 
 public:
@@ -35,14 +33,22 @@ enum {
   RUNTOAPUSTEP
 };
 enum { DRAM = 0, SPCRAM = 1, VRAM = 2, OAM = 3, CGRAM = 4 };
+  void   power();
+  void   reset();
+
   void   run();
   void   video_run();
-  void   sound_run();
+  void   sound_run(uint32 data);
 
   void   set_status(uint32 new_status);
   uint32 get_status();
 
+//video functions
+  uint16 *video_lock(uint32 &pitch);
+  void    video_unlock();
+
 //input functions
+  void clear_input();
   void poll_input(uint8 type);
   bool get_input_status(uint8 device, uint8 button);
 
diff --git a/src/win/bsnes.lnk b/src/win/bsnes.lnk
index ce73821e..b3f084ab 100644
Binary files a/src/win/bsnes.lnk and b/src/win/bsnes.lnk differ
diff --git a/src/win/config.cpp b/src/win/config.cpp
index c0a92860..8d238f06 100644
--- a/src/win/config.cpp
+++ b/src/win/config.cpp
@@ -1,5 +1,10 @@
 namespace config {
 
+struct System {
+  static Setting regulate_speed;
+} system;
+Setting System::regulate_speed(&config_file, "system.regulate_speed", "Regulate speed to 60hz (NTSC) / 50hz (PAL)", true, Setting::TRUE_FALSE);
+
 struct Video {
   static Setting mode, use_vram, vblank;
 } video;
diff --git a/src/win/dd_renderer.cpp b/src/win/dd_renderer.cpp
index 47fae633..7dc31009 100644
--- a/src/win/dd_renderer.cpp
+++ b/src/win/dd_renderer.cpp
@@ -176,66 +176,19 @@ int rx, ry;
   }
 }
 
-#include "dd_renderer16.cpp"
-void DDRenderer::update16() {
-HRESULT hr;
-  hr = lpddsb->Lock(0, &ddsd, DDLOCK_WAIT, 0);
-  if(hr != DD_OK)return;
-
-  set_source_window();
-
-  if       (vi.width == 256 && vi.height == 224) {
-    update16_256x224();
-  } else if(vi.width == 512 && vi.height == 224) {
-    update16_512x224();
-  } else if(vi.width == 256 && vi.height == 448) {
-    update16_256x448();
-  } else if(vi.width == 512 && vi.height == 448) {
-    update16_512x448();
-  }
-
-  lpddsb->Unlock(0);
+uint16 *DDRenderer::lock(uint32 &pitch) {
+  if(lpddsb->Lock(0, &ddsd, DDLOCK_WAIT, 0) != DD_OK)return 0;
+  pitch = ddsd.lPitch;
+  return (uint16*)ddsd.lpSurface;
 }
 
-//#include "dd_renderer32.cpp"
-/*
-void DDRenderer::update32() {
-HRESULT hr;
-  hr = lpddsb->Lock(0, &ddsd, DDLOCK_WAIT, 0);
-  if(hr != DD_OK)return;
-
-  set_source_window();
-  if(ppu->output->hires == false) {
-    if(ppu->output->interlace == false) {
-      update32_256x224();
-    } else {
-      update32_256x448();
-    }
-  } else {
-    if(ppu->output->interlace == false) {
-      update32_512x224();
-    } else {
-      update32_512x448();
-    }
-  }
-
+void DDRenderer::unlock() {
   lpddsb->Unlock(0);
 }
-*/
 
 void DDRenderer::update() {
   snes->get_video_info(&vi);
-
-  switch(color_depth) {
-  case 15:
-  case 16:
-    update16();
-    break;
-  case 32:
-  //update32();
-    break;
-  }
-
+  set_source_window();
   redraw();
 }
 
diff --git a/src/win/dd_renderer.h b/src/win/dd_renderer.h
index b98127ec..626dc373 100644
--- a/src/win/dd_renderer.h
+++ b/src/win/dd_renderer.h
@@ -14,22 +14,15 @@ bool  fullscreen;
 int   width, height; //used for fullscreen mode clipping only
 uint8 color_depth;
 SNES::video_info vi; //initialized each frame at start of update()
+  uint16 *lock(uint32 &pitch);
+  void    unlock();
+
   void set_window(HWND hwnd_handle);
   void create_backbuffer();
   void to_windowed();
   void to_fullscreen(int _width, int _height);
   void set_source_window();
   void redraw();
-  inline void update16_256x224();
-  inline void update16_512x224();
-  inline void update16_256x448();
-  inline void update16_512x448();
-  void update16();
-//inline void update32_256x224();
-//inline void update32_512x224();
-//inline void update32_256x448();
-//inline void update32_512x448();
-//void update32();
   void update();
   void destroy();
 
diff --git a/src/win/dd_renderer16.cpp b/src/win/dd_renderer16.cpp
deleted file mode 100644
index 6a4fef80..00000000
--- a/src/win/dd_renderer16.cpp
+++ /dev/null
@@ -1,79 +0,0 @@
-inline void DDRenderer::update16_256x224() {
-uint16 *src, *dest;
-uint32 pitch;
-int x, y;
-  dest = (uint16*)ddsd.lpSurface;
-  src  = (uint16*)vi.data;
-
-#ifdef USE_X86_ASM
-  pitch = (ddsd.lPitch) - 512;
-  __asm {
-    mov edi,dest
-    mov esi,src
-    mov edx,224
-  ly:
-    mov ecx,32
-  lx:
-    movsd
-    movsd
-    movsd
-    movsd
-    loopnz lx
-    add edi,pitch
-    dec edx
-    jnz ly
-  }
-#else
-  pitch = (ddsd.lPitch >> 1);
-  for(y=0;y<224;y++) {
-    memcpy(dest, src, 512);
-    dest += pitch;
-    src  += 256;
-  }
-#endif
-}
-
-inline void DDRenderer::update16_512x224() {
-uint16 *src, *dest;
-uint32 pitch;
-int x, y;
-  pitch = (ddsd.lPitch >> 1);
-  dest  = (uint16*)ddsd.lpSurface;
-  src   = (uint16*)vi.data;
-
-  for(y=0;y<224;y++) {
-    memcpy(dest, src, 1024);
-    dest += pitch;
-    src  += 512;
-  }
-}
-
-inline void DDRenderer::update16_256x448() {
-uint16 *src, *dest;
-uint32 pitch;
-int x, y;
-  pitch = (ddsd.lPitch >> 1);
-  dest  = (uint16*)ddsd.lpSurface;
-  src   = (uint16*)vi.data;
-
-  for(y=0;y<448;y++) {
-    memcpy(dest, src, 512);
-    dest += pitch;
-    src  += 256;
-  }
-}
-
-inline void DDRenderer::update16_512x448() {
-uint16 *src, *dest;
-uint32 pitch;
-int x, y;
-  pitch = (ddsd.lPitch >> 1);
-  dest  = (uint16*)ddsd.lpSurface;
-  src   = (uint16*)vi.data;
-
-  for(y=0;y<448;y++) {
-    memcpy(dest, src, 1024);
-    dest += pitch;
-    src  += 512;
-  }
-}
diff --git a/src/win/dd_renderer32.cpp b/src/win/dd_renderer32.cpp
deleted file mode 100644
index 6b4893ea..00000000
--- a/src/win/dd_renderer32.cpp
+++ /dev/null
@@ -1,123 +0,0 @@
-inline void DDRenderer::update32_256x224() {
-uint16 *src;
-uint32 *dest;
-uint32 pitch;
-int x, y;
-  src   = (uint16*)ppu->output->buffer + (1 << 10);
-  dest  = (uint32*)ddsd.lpSurface;
-  pitch = (ddsd.lPitch >> 2) - 256;
-
-int overscan_adjust = 0;
-  if(cpu->overscan() == true) {
-    src += 7 << 10;
-    overscan_adjust = 7;
-  }
-
-  for(y=1+overscan_adjust;y<224+overscan_adjust;y++) {
-    x = 256;
-    while(x--) {
-      *dest++ = color_lookup_table[*src];
-      src += 2;
-    }
-    dest += pitch;
-    src  += 512;
-  }
-}
-
-inline void DDRenderer::update32_256x448() {
-uint16 *src;
-uint32 *dest;
-uint32 pitch;
-int x, y;
-  src   = (uint16*)ppu->output->buffer + (1 << 10);
-  dest  = (uint32*)ddsd.lpSurface;
-  pitch = (ddsd.lPitch >> 2) - 256;
-
-int overscan_adjust = 0;
-  if(cpu->overscan() == true) {
-    src += 7 << 10;
-    overscan_adjust = 14;
-  }
-
-  for(y=2+overscan_adjust;y<448+overscan_adjust;y++) {
-    x = 256;
-    while(x--) {
-      *dest++ = color_lookup_table[*src];
-      src += 2;
-    }
-    dest += pitch;
-    if(ppu->output->line[y >> 1].interlace == false) {
-      src += (y & 1)?512:-512;
-    }
-  }
-}
-
-inline void DDRenderer::update32_512x224() {
-uint16 *src;
-uint32 *dest;
-uint32 pitch;
-int x, y;
-  src   = (uint16*)ppu->output->buffer + (1 << 10);
-  dest  = (uint32*)ddsd.lpSurface;
-  pitch = (ddsd.lPitch >> 2) - 512;
-
-int overscan_adjust = 0;
-  if(cpu->overscan() == true) {
-    src += 7 << 10;
-    overscan_adjust = 7;
-  }
-
-  for(y=1+overscan_adjust;y<224+overscan_adjust;y++) {
-    if(ppu->output->line[y].hires == true) {
-      x = 512;
-      while(x--) {
-        *dest++ = color_lookup_table[*src++];
-      }
-    } else {
-      x = 256;
-      while(x--) {
-        *dest++ = color_lookup_table[*src];
-        *dest++ = color_lookup_table[*src];
-        src += 2;
-      }
-    }
-    dest += pitch;
-    src  += 512;
-  }
-}
-
-inline void DDRenderer::update32_512x448() {
-uint16 *src;
-uint32 *dest;
-uint32 pitch;
-int x, y;
-  src   = (uint16*)ppu->output->buffer + (1 << 10);
-  dest  = (uint32*)ddsd.lpSurface;
-  pitch = (ddsd.lPitch >> 2) - 512;
-
-int overscan_adjust = 0;
-  if(cpu->overscan() == true) {
-    src += 7 << 10;
-    overscan_adjust = 14;
-  }
-
-  for(y=2+overscan_adjust;y<448+overscan_adjust;y++) {
-    if(ppu->output->line[y >> 1].hires == true) {
-      x = 512;
-      while(x--) {
-        *dest++ = color_lookup_table[*src++];
-      }
-    } else {
-      x = 256;
-      while(x--) {
-        *dest++ = color_lookup_table[*src];
-        *dest++ = color_lookup_table[*src];
-        src += 2;
-      }
-    }
-    dest += pitch;
-    if(ppu->output->line[y >> 1].interlace == false) {
-      src += (y & 1)?512:-512;
-    }
-  }
-}
diff --git a/src/win/ds_sound.cpp b/src/win/ds_sound.cpp
index ab6862e8..25af846e 100644
--- a/src/win/ds_sound.cpp
+++ b/src/win/ds_sound.cpp
@@ -1,77 +1,116 @@
-DSSound::DSSound() {
-  data.buffer = 0;
-  data.lpos = data.lsample = data.lbuffer = 0;
-}
+void DSSound::run(uint32 sample) {
+  data.buffer[data.buffer_pos++] = sample;
 
-void DSSound::run() {
-  if(snes->get_playback_buffer_pos() != 0)return;
+  if(data.buffer_pos >= data.samples_per_frame) {
+  uint32 pos, size;
+  void  *buffer;
+    if((bool)config::system.regulate_speed == true) {
+      do {
+        dsb_b->GetCurrentPosition(&pos, 0);
+        data.read_buffer = pos / data.buffer_size;
+      } while(data.read_buffer == data.prev_buffer);
+    }
 
-uint32 pos, status;
+    data.prev_buffer = data.read_buffer;
+    data.read_buffer++;
+    data.read_buffer &= 7;
 
-//dsb_b[0]->SetFrequency(22050);
+    pos = (data.read_buffer + 1) & 7;
 
-//do {
-//  dsb_b[0]->GetStatus(&status);
-//} while(status & DSBSTATUS_PLAYING);
+    if(dsb_b->Lock(pos * data.buffer_size,
+    data.buffer_size, &buffer, &size, 0, 0, 0) == DS_OK) {
+      memcpy(buffer, data.buffer, data.buffer_size);
+      dsb_b->Unlock(buffer, size, 0, 0);
+    }
 
-  dsb_b[0]->Lock(0, DSP_BUFFER_SIZE * 4, &dslb, &dslbs, 0, 0, 0);
-  memcpy(dslb, snes->get_playback_buffer(), DSP_BUFFER_SIZE * 4);
-  dsb_b[0]->Unlock(dslb, dslbs, 0, 0);
-
-  dsb_b[0]->SetCurrentPosition(0);
-
-//has the buffer stopped (possibly due to running too fast)?
-  dsb_b[0]->GetStatus(&status);
-  if(!(status & DSBSTATUS_PLAYING)) {
-    dsb_b[0]->Play(0, 0, 0);
+    data.buffer_pos = 0;
   }
 }
 
+void DSSound::clear() {
+  data.read_buffer = 0;
+  data.prev_buffer = 0;
+  data.buffer_pos  = 0;
+  memset(data.buffer, 0, 2048 * 4);
+
+  if(!dsb_b)return;
+
+  dsb_b->Stop();
+  dsb_b->SetCurrentPosition(0);
+
+uint32 size;
+void  *buffer;
+  dsb_b->Lock(0, data.buffer_size * 8, &buffer, &size, 0, 0, 0);
+  memset(buffer, 0, data.buffer_size * 8);
+  dsb_b->Unlock(buffer, size, 0, 0);
+
+  dsb_b->Play(0, 0, DSBPLAY_LOOPING);
+}
+
 void DSSound::init() {
+  clear();
+  term();
+
+  data.read_buffer       = 0;
+  data.prev_buffer       = 0;
+  data.buffer_pos        = 0;
+  data.samples_per_frame = DSP_FREQ / ((snes->region() == SNES::NTSC) ? 60 : 50);
+  data.buffer_size       = data.samples_per_frame * 4;
+
   DirectSoundCreate(0, &ds, 0);
   ds->SetCooperativeLevel(w_main->hwnd, DSSCL_PRIORITY);
 
   memset(&dsbd, 0, sizeof(dsbd));
-  dsbd.dwSize = sizeof(dsbd);
-  dsbd.dwFlags = DSBCAPS_PRIMARYBUFFER;
+  dsbd.dwSize        = sizeof(dsbd);
+  dsbd.dwFlags       = DSBCAPS_PRIMARYBUFFER;
   dsbd.dwBufferBytes = 0;
-  dsbd.lpwfxFormat = 0;
+  dsbd.lpwfxFormat   = 0;
   ds->CreateSoundBuffer(&dsbd, &dsb_p, 0);
 
   memset(&wfx, 0, sizeof(wfx));
-  wfx.wFormatTag = WAVE_FORMAT_PCM;
-  wfx.nChannels = 2;
-  wfx.nSamplesPerSec = 32000;
-  wfx.wBitsPerSample = 16;
-  wfx.nBlockAlign = wfx.wBitsPerSample / 8 * wfx.nChannels;
+  wfx.wFormatTag      = WAVE_FORMAT_PCM;
+  wfx.nChannels       = 2;
+  wfx.nSamplesPerSec  = DSP_FREQ;
+  wfx.wBitsPerSample  = 16;
+  wfx.nBlockAlign     = wfx.wBitsPerSample / 8 * wfx.nChannels;
   wfx.nAvgBytesPerSec = wfx.nSamplesPerSec * wfx.nBlockAlign;
   dsb_p->SetFormat(&wfx);
 
-  sample_size = (16 / 8) * 2;
-  buffer_size = DSP_BUFFER_SIZE * sample_size;
-  buffer_pos  = 0;
-
-  dsb_b = new LPDIRECTSOUNDBUFFER[1];
   memset(&dsbd, 0, sizeof(dsbd));
-  dsbd.dwSize = sizeof(dsbd);
-  dsbd.dwFlags = DSBCAPS_CTRLFREQUENCY | DSBCAPS_GLOBALFOCUS;
-  dsbd.dwBufferBytes = buffer_size;
+  dsbd.dwSize          = sizeof(dsbd);
+  dsbd.dwFlags         = DSBCAPS_GETCURRENTPOSITION2 | DSBCAPS_CTRLFREQUENCY |
+                         DSBCAPS_GLOBALFOCUS | DSBCAPS_LOCSOFTWARE;
+  dsbd.dwBufferBytes   = data.buffer_size * 8;
   dsbd.guid3DAlgorithm = GUID_NULL;
-  dsbd.lpwfxFormat = &wfx;
-  ds->CreateSoundBuffer(&dsbd, &dsb_b[0], 0);
-  ds->CreateSoundBuffer(&dsbd, &dsb_b[1], 0);
-  dsb_b[0]->SetFrequency(32000);
-  dsb_b[1]->SetFrequency(32000);
+  dsbd.lpwfxFormat     = &wfx;
+  ds->CreateSoundBuffer(&dsbd, &dsb_b, 0);
+  dsb_b->SetFrequency(DSP_FREQ);
+  dsb_b->SetCurrentPosition(0);
 
-  dsb_b[0]->Lock(0, buffer_size, &dslb, &dslbs, 0, 0, 0);
-  memset(dslb, 0, buffer_size);
-  dsb_b[0]->Unlock(dslb, dslbs, 0, 0);
+uint32 size;
+void  *buffer;
+  dsb_b->Lock(0, data.buffer_size * 8, &buffer, &size, 0, 0, 0);
+  memset(buffer, 0, data.buffer_size * 8);
+  dsb_b->Unlock(buffer, size, 0, 0);
 
-  dsb_b[1]->Lock(0, buffer_size, &dslb, &dslbs, 0, 0, 0);
-  memset(dslb, 0, buffer_size);
-  dsb_b[1]->Unlock(dslb, dslbs, 0, 0);
-
-  dsb_b[0]->Play(0, 0, 0);
-
-  buffer_pos = 0;
+  data.read_buffer = 0;
+  dsb_b->Play(0, 0, DSBPLAY_LOOPING);
+}
+
+void DSSound::term() {
+  if(dsb_b) {
+    dsb_b->Stop();
+    dsb_b->Release();
+    dsb_b = 0;
+  }
+
+  if(dsb_p) {
+    dsb_p->Release();
+    dsb_p = 0;
+  }
+
+  if(ds) {
+    ds->Release();
+    ds = 0;
+  }
 }
diff --git a/src/win/ds_sound.h b/src/win/ds_sound.h
index d580facb..4e2f9f5d 100644
--- a/src/win/ds_sound.h
+++ b/src/win/ds_sound.h
@@ -1,30 +1,28 @@
 #include <dsound.h>
 
+#define DSP_FREQ 32000
+
 class DSSound {
 public:
 LPDIRECTSOUND ds;
-LPDIRECTSOUNDBUFFER dsb_p, *dsb_b;
+LPDIRECTSOUNDBUFFER dsb_p, dsb_b;
 DSBUFFERDESC dsbd;
 WAVEFORMATEX wfx;
-uint32 sample_size;
 uint32 buffer_size;
-uint32 buffer_pos;
-
-void  *dslb;
-uint32 dslbs;
 
 public:
-  void run();
+  void run(uint32 sample);
+  void clear();
   void init();
-
-  DSSound();
+  void term();
 };
 
 struct {
-  uint32 *buffer;
-  uint32 lpos;
-  uint32 lsample;
-  bool   lbuffer;
+  uint32 buffer[2048];
+  uint8  read_buffer, prev_buffer;
+  uint32 buffer_pos, buffer_size;
+
+  uint32 samples_per_frame;
 } data;
 
 DSSound *ds_sound;
diff --git a/src/win/rom.cpp b/src/win/rom.cpp
index e2a33fb5..7057affd 100644
--- a/src/win/rom.cpp
+++ b/src/win/rom.cpp
@@ -27,14 +27,14 @@ FileReader *rf = new FileReader();
     alert("Error loading image file [%s]!", rom_fn);
     return false;
   }
-  mem_bus->load_cart(static_cast<Reader*>(rf));
+  r_mem->load_cart(static_cast<Reader*>(rf));
   rf->close();
 
 CartInfo ci;
-  mem_bus->get_cartinfo(&ci);
+  r_mem->get_cartinfo(&ci);
   if(ci.sram_size != 0) {
     rf->open(sram_fn);
-    mem_bus->load_sram(static_cast<Reader*>(rf));
+    r_mem->load_sram(static_cast<Reader*>(rf));
     rf->close();
   }
 
@@ -50,11 +50,11 @@ void ROMImage::unload() {
 
 FileWriter *wf;
 CartInfo ci;
-  mem_bus->get_cartinfo(&ci);
+  r_mem->get_cartinfo(&ci);
   if(ci.sram_size != 0) {
     wf = new FileWriter();
     wf->open(sram_fn);
-    mem_bus->save_sram(static_cast<Writer*>(wf));
+    r_mem->save_sram(static_cast<Writer*>(wf));
     wf->close();
     delete(wf);
   }
@@ -62,7 +62,7 @@ CartInfo ci;
   file_loaded = false;
   bsnes->debugger_deactivate();
 
-  mem_bus->unload_cart();
+  r_mem->unload_cart();
 }
 
 void ROMImage::select(char *fn) {
diff --git a/src/win/ui.cpp b/src/win/ui.cpp
index 46ae9c34..d48ca202 100644
--- a/src/win/ui.cpp
+++ b/src/win/ui.cpp
@@ -54,7 +54,6 @@ void init_ui1() {
   SetFocus(w_main->hwnd);
   dd_renderer->set_window(w_main->hwnd);
   dd_renderer->to_windowed();
-  ds_sound->init();
 
   w_main->show_menu();
   w_main->set_video_mode(config::video.mode);
diff --git a/src/win/ui.h b/src/win/ui.h
index 9619a420..8ca69fc5 100644
--- a/src/win/ui.h
+++ b/src/win/ui.h
@@ -14,6 +14,7 @@ enum {
   MENU_FILE_RESET,
   MENU_FILE_POWER,
   MENU_FILE_EXIT,
+  MENU_SETTINGS_REGULATE_SPEED,
   MENU_SETTINGS_FRAMESKIP_OFF,
   MENU_SETTINGS_FRAMESKIP_1,
   MENU_SETTINGS_FRAMESKIP_2,
@@ -251,7 +252,7 @@ bool   auto_update; //update memory window whenever visible value is written to
   uint8 read_byte(uint32 addr);
   void  write_byte(uint32 addr, uint8 value);
   void  refresh(uint32 type = null, uint32 addr = 0);
-  void  export(uint32 type);
+  void  export_data(uint32 type);
 }*w_memory = 0;
 
 class InputConfig : public Window {
diff --git a/src/win/ui_console.cpp b/src/win/ui_console.cpp
index 52cf61fc..09b01dd0 100644
--- a/src/win/ui_console.cpp
+++ b/src/win/ui_console.cpp
@@ -29,7 +29,7 @@ HDC hdc;
       break;
     case CONSOLE_CPUPROCEED:
       if(bsnes->get_status() == bSNES::STOP) {
-        if(cpu->in_opcode() == true) {
+        if(r_cpu->in_opcode() == true) {
           dprintf("* CPU within opcode, proceed aborted");
         } else {
           bsnes->set_status(bSNES::RUNTOCPUPROCEED);
@@ -38,10 +38,10 @@ HDC hdc;
       break;
     case CONSOLE_CPUSKIP:
       if(bsnes->get_status() == bSNES::STOP) {
-        if(cpu->in_opcode() == true) {
+        if(r_cpu->in_opcode() == true) {
           dprintf("* CPU within opcode, skip aborted");
         } else {
-          cpu->regs.pc.w += cpu->opcode_length();
+          r_cpu->regs.pc.w += r_cpu->opcode_length();
           bsnes->disassemble_cpu_op();
         }
       }
@@ -57,15 +57,15 @@ HDC hdc;
       break;
     case CONSOLE_CPUDISABLE:
       if(bsnes->get_status() == bSNES::STOP) {
-        if(cpu->in_opcode() == true) {
+        if(r_cpu->in_opcode() == true) {
           dprintf("* CPU within opcode, disable aborted");
         } else {
-          addr = cpu->regs.pc.d;
-          len = cpu->opcode_length();
+          addr = r_cpu->regs.pc.d;
+          len = r_cpu->opcode_length();
           for(i=0;i<len;i++) {
             bsnes->write(bSNES::DRAM, (addr & 0xff0000) | ((addr + i) & 0xffff), 0xea);
           }
-        //cpu->regs.pc.w += len;
+        //r_cpu->regs.pc.w += len;
           bsnes->disassemble_cpu_op();
         }
       }
@@ -123,23 +123,23 @@ HDC hdc;
       value = strhex(t);
       pos = SendDlgItemMessage(hwnd, CONSOLE_CFGREGTYPE, CB_GETCURSEL, 0, 0);
       if(pos == 0) { //Set CPU register
-        if(cpu->in_opcode() == true) {
+        if(r_cpu->in_opcode() == true) {
           dprintf("* CPU within opcode, register set aborted");
         } else {
           pos = SendDlgItemMessage(hwnd, CONSOLE_CFGREGNUM, CB_GETCURSEL, 0, 0);
           switch(pos) {
-          case 0:cpu->regs.a.w  = value;break;
-          case 1:cpu->regs.x.w  = value;break;
-          case 2:cpu->regs.y.w  = value;break;
-          case 3:cpu->regs.s.w  = value;break;
-          case 4:cpu->regs.d.w  = value;break;
-          case 5:cpu->regs.db   = value;break;
-          case 6:cpu->regs.p    = value;break;
-          case 7:cpu->regs.e    = value;break;
-          case 8:cpu->regs.pc.d = value;break;
+          case 0:r_cpu->regs.a.w  = value;break;
+          case 1:r_cpu->regs.x.w  = value;break;
+          case 2:r_cpu->regs.y.w  = value;break;
+          case 3:r_cpu->regs.s.w  = value;break;
+          case 4:r_cpu->regs.d.w  = value;break;
+          case 5:r_cpu->regs.db   = value;break;
+          case 6:r_cpu->regs.p    = value;break;
+          case 7:r_cpu->regs.e    = value;break;
+          case 8:r_cpu->regs.pc.d = value;break;
           }
         //these bits can never be clear in emulation mode
-          if(cpu->regs.e)cpu->regs.p |= 0x30;
+          if(r_cpu->regs.e)r_cpu->regs.p |= 0x30;
           bsnes->disassemble_cpu_op();
         }
       } else { //Set APU register
@@ -229,10 +229,12 @@ int sl = strlen(s);
     memset(t + sl, 0x20, 80 - sl);
   }
   t[80] = 0;
+
 //only allow ascii characters. other characters will force the
 //font to change to one that supports non-ascii characters,
 //which will break the line highlighting and alignment of text
-  for(int i=0;i<80;i++) {
+int i;
+  for(i=0;i<80;i++) {
     if(t[i] & 0x80)t[i] = '?';
   }
 
@@ -281,13 +283,13 @@ static uint8 linecol[4] = { 1, 2, 3 };
 
   strcpy(s, "");
   sprintf(t, "V:%3d H:%3d HC:%4d I:%d IF:%d O:%d",
-    cpu->vcounter(), cpu->hcounter(), cpu->hcycles(),
-    cpu->interlace(), cpu->interlace_field(), cpu->overscan());
+    r_cpu->vcounter(), r_cpu->hcounter(), r_cpu->hcycles(),
+    r_cpu->interlace(), r_cpu->interlace_field(), r_cpu->overscan());
   strcat(s, t);
   if(1) { //config::apu.enabled
     sprintf(t, " -- CPU[$%0.2x,$%0.2x,$%0.2x,$%0.2x]<>APU[$%0.2x,$%0.2x,$%0.2x,$%0.2x]",
-      cpu->port_read(0), cpu->port_read(1), cpu->port_read(2), cpu->port_read(3),
-      apu->port_read(0), apu->port_read(1), apu->port_read(2), apu->port_read(3)
+      r_cpu->port_read(0), r_cpu->port_read(1), r_cpu->port_read(2), r_cpu->port_read(3),
+      r_apu->port_read(0), r_apu->port_read(1), r_apu->port_read(2), r_apu->port_read(3)
     );
     strcat(s, t);
   }
diff --git a/src/win/ui_main.cpp b/src/win/ui_main.cpp
index 0ad3aca6..142aa3cd 100644
--- a/src/win/ui_main.cpp
+++ b/src/win/ui_main.cpp
@@ -135,6 +135,16 @@ long __stdcall wndproc_main(HWND hwnd, UINT msg, WPARAM wparam, LPARAM lparam) {
 time_t timeout;
 int i;
   switch(msg) {
+  case WM_ENTERMENULOOP:
+    ds_sound->clear();
+    break;
+  case WM_EXITMENULOOP:
+    timeout = time(0);
+    while(difftime(time(0), timeout) < 3) {
+      if(!KeyDown(VK_RETURN))break;
+    }
+    bsnes->clear_input();
+    break;
   case WM_KEYDOWN:
     if(wparam == VK_ESCAPE) {
       if(GetMenu(w_main->hwnd) == NULL) {
@@ -146,13 +156,6 @@ int i;
     }
     break;
   case WM_COMMAND:
-//below code fails because it is triggered after snes->poll_input()...
-//unsure how to fix this...
-//  timeout = time(NULL);
-//  while(difftime(time(NULL), timeout) < 5) {
-//    if(!KeyDown(VK_RETURN))break;
-//  }
-
     switch(LOWORD(wparam)) {
     case MENU_FILE_LOAD:
       w_main->menu_load();
@@ -175,6 +178,11 @@ int i;
     case MENU_FILE_EXIT:
       PostQuitMessage(0);
       break;
+    case MENU_SETTINGS_REGULATE_SPEED:
+      config::system.regulate_speed.toggle();
+      CheckMenuItem(w_main->hmenu, MENU_SETTINGS_REGULATE_SPEED,
+        (config::system.regulate_speed)?MF_CHECKED:MF_UNCHECKED);
+      break;
     case MENU_SETTINGS_FRAMESKIP_OFF:
     case MENU_SETTINGS_FRAMESKIP_1:
     case MENU_SETTINGS_FRAMESKIP_2:
@@ -314,6 +322,7 @@ HMENU hsubmenu, hbranchmenu;
   AppendMenu(hmenu, MF_STRING | MF_POPUP, (unsigned int)hsubmenu, "&File");
 
   hsubmenu = CreatePopupMenu();
+  AppendMenu(hsubmenu, MF_STRING, MENU_SETTINGS_REGULATE_SPEED, "&Regulate Speed");
 
   hbranchmenu = CreatePopupMenu();
   AppendMenu(hbranchmenu, MF_STRING, MENU_SETTINGS_FRAMESKIP_OFF, "Off");
@@ -356,8 +365,11 @@ HMENU hsubmenu, hbranchmenu;
   AppendMenu(hbranchmenu, MF_STRING, MENU_SETTINGS_INPUTCFG_JOYPAD2, "Joypad 2");
   AppendMenu(hsubmenu, MF_STRING | MF_POPUP, (unsigned int)hbranchmenu, "&Configure Input Devices");
 
+#ifdef DEBUGGER
   AppendMenu(hsubmenu, MF_SEPARATOR, 0, "");
   AppendMenu(hsubmenu, MF_STRING, MENU_SETTINGS_DEBUGGER, "&Debugger");
+#endif
+
   AppendMenu(hmenu, MF_STRING | MF_POPUP, (unsigned int)hsubmenu, "&Settings");
 
   hsubmenu = CreatePopupMenu();
@@ -367,6 +379,7 @@ HMENU hsubmenu, hbranchmenu;
   AppendMenu(hsubmenu, MF_STRING, MENU_MISC_ABOUT, "&About...");
   AppendMenu(hmenu, MF_STRING | MF_POPUP, (unsigned int)hsubmenu, "&Misc");
 
+  CheckMenuItem(hmenu, MENU_SETTINGS_REGULATE_SPEED, (config::system.regulate_speed)?MF_CHECKED:MF_UNCHECKED);
   CheckMenuItem(hmenu, MENU_SETTINGS_USEVRAM,    (config::video.use_vram)?MF_CHECKED:MF_UNCHECKED);
   CheckMenuItem(hmenu, MENU_SETTINGS_VBLANK,     (config::video.vblank)?MF_CHECKED:MF_UNCHECKED);
   CheckMenuItem(hmenu, MENU_SETTINGS_COLORADJUST_COLORCURVE, (config::snes.video_color_curve)?MF_CHECKED:MF_UNCHECKED);
diff --git a/src/win/ui_memory.cpp b/src/win/ui_memory.cpp
index 380d9fef..a877778b 100644
--- a/src/win/ui_memory.cpp
+++ b/src/win/ui_memory.cpp
@@ -154,7 +154,7 @@ HDC hdc;
       break;
     case MEMORYEDITOR_FEXPORT:
       pos = SendDlgItemMessage(hwnd, MEMORYEDITOR_FSOURCE, CB_GETCURSEL, 0, 0);
-      w_memory->export(pos);
+      w_memory->export_data(pos);
       break;
     case MEMORYEDITOR_AUTOUPDATE:
       if(w_memory->auto_update == false) {
@@ -186,7 +186,7 @@ HDC hdc;
   return DefWindowProc(hwnd, msg, wparam, lparam);
 }
 
-void MemoryEditor::export(uint32 type) {
+void MemoryEditor::export_data(uint32 type) {
 FILE *fp;
 int i, x;
   if(type == 0) { //DRAM
@@ -233,12 +233,12 @@ int i, x;
     }
     fclose(fp);
   } else if(type == 6) { //All
-    export(0);
-    export(1);
-    export(2);
-    export(3);
-    export(4);
-    export(5);
+    export_data(0);
+    export_data(1);
+    export_data(2);
+    export_data(3);
+    export_data(4);
+    export_data(5);
   }
 }
 
diff --git a/src/win/winmain.cpp b/src/win/winmain.cpp
index 7778e009..45d8f349 100644
--- a/src/win/winmain.cpp
+++ b/src/win/winmain.cpp
@@ -1,15 +1,10 @@
 #define INTERFACE_MAIN
 
-//requires visual c++
-#define USE_X86_ASM
-
 #include "winmain.h"
 #include "../base.h"
 
 #include "config.cpp"
 
-#define DSP_BUFFER_SIZE 8000
-
 #include "bsnes.h"
 #include "ui.h"
 #include "dd_renderer.h"
@@ -22,44 +17,47 @@
 #include "ui.cpp"
 
 void init_snes() {
-  mem_bus = new bMemBus();
-  cpu     = new bCPU();
-  apu     = new bAPU();
-  dsp     = new bDSP();
-  ppu     = new bPPU();
-  snes    = new bSNES();
-  bsnes   = static_cast<bSNES*>(snes);
+#ifdef POLYMORPHISM
+  deref(mem) = new bMemBus();
+  deref(cpu) = new bCPU();
+  deref(apu) = new bAPU();
+  deref(dsp) = new bDSP();
+  deref(ppu) = new bPPU();
+#endif
+  snes  = new bSNES();
+  bsnes = static_cast<bSNES*>(snes);
 
   snes->init();
-  snes->set_playback_buffer_size(DSP_BUFFER_SIZE);
 }
 
 void term_snes() {
   snes->term();
 
+#ifdef POLYMORPHISM
 //static casting is neccesary to call derived class deconstructor...
-  if(mem_bus) {
-    delete(static_cast<bMemBus*>(mem_bus));
-    mem_bus = 0;
+  if(deref(mem)) {
+    delete static_cast<bMemBus*>(deref(mem));
+    deref(mem) = 0;
   }
-  if(cpu) {
-    delete(static_cast<bCPU*>(cpu));
-    cpu = 0;
+  if(deref(cpu)) {
+    delete static_cast<bCPU*>(deref(cpu));
+    deref(cpu) = 0;
   }
-  if(apu) {
-    delete(static_cast<bAPU*>(apu));
-    apu = 0;
+  if(deref(apu)) {
+    delete static_cast<bAPU*>(deref(apu));
+    deref(apu) = 0;
   }
-  if(dsp) {
-    delete(static_cast<bDSP*>(dsp));
-    dsp = 0;
+  if(deref(dsp)) {
+    delete static_cast<bDSP*>(deref(dsp));
+    deref(dsp) = 0;
   }
-  if(ppu) {
-    delete(static_cast<bPPU*>(ppu));
-    ppu = 0;
+  if(deref(ppu)) {
+    delete static_cast<bPPU*>(deref(ppu));
+    deref(ppu) = 0;
   }
+#endif
   if(snes) {
-    delete(static_cast<bSNES*>(snes));
+    delete static_cast<bSNES*>(snes);
     snes = 0;
   }
 }