diff --git a/bsnes/base/base.hpp b/bsnes/base/base.hpp
index aa168cdc..cb1edb92 100755
--- a/bsnes/base/base.hpp
+++ b/bsnes/base/base.hpp
@@ -1,7 +1,7 @@
 #ifndef BASE_HPP
 #define BASE_HPP
 
-static const char Version[] = "087.15";
+static const char Version[] = "087.16";
 
 #include <nall/platform.hpp>
 #include <nall/algorithm.hpp>
diff --git a/bsnes/gba/cpu/cpu.cpp b/bsnes/gba/cpu/cpu.cpp
index ecd212f7..909bdeac 100755
--- a/bsnes/gba/cpu/cpu.cpp
+++ b/bsnes/gba/cpu/cpu.cpp
@@ -4,6 +4,7 @@ namespace GBA {
 
 #include "registers.cpp"
 #include "mmio.cpp"
+#include "dma.cpp"
 CPU cpu;
 
 void CPU::Enter() { cpu.enter(); }
@@ -28,6 +29,7 @@ void CPU::enter() {
       regs.mode = Registers::Mode::Normal;
     }
 
+    dma_run();
     exec();
   }
 }
@@ -72,6 +74,10 @@ void CPU::power() {
   regs.mode = Registers::Mode::Normal;
   regs.memory.control = 0x0d000020;
 
+  pending.dma.vblank = 0;
+  pending.dma.hblank = 0;
+  pending.dma.hdma = 0;
+
   for(unsigned n = 0x0b0; n <= 0x0df; n++) bus.mmio[n] = this;  //DMA
   for(unsigned n = 0x100; n <= 0x10f; n++) bus.mmio[n] = this;  //Timers
   for(unsigned n = 0x130; n <= 0x133; n++) bus.mmio[n] = this;  //Keypad
diff --git a/bsnes/gba/cpu/cpu.hpp b/bsnes/gba/cpu/cpu.hpp
index 3fb53966..83fb0968 100755
--- a/bsnes/gba/cpu/cpu.hpp
+++ b/bsnes/gba/cpu/cpu.hpp
@@ -2,6 +2,7 @@ struct CPU : Processor::ARM, Thread, MMIO {
   StaticMemory iwram;
   StaticMemory ewram;
   #include "registers.hpp"
+  #include "state.hpp"
 
   static void Enter();
   void enter();
@@ -14,6 +15,9 @@ struct CPU : Processor::ARM, Thread, MMIO {
   uint8 read(uint32 addr);
   void write(uint32 addr, uint8 byte);
 
+  void dma_run();
+  void dma_transfer(uint2 channel);
+
   CPU();
 };
 
diff --git a/bsnes/gba/cpu/dma.cpp b/bsnes/gba/cpu/dma.cpp
new file mode 100755
index 00000000..d4b5b1f3
--- /dev/null
+++ b/bsnes/gba/cpu/dma.cpp
@@ -0,0 +1,45 @@
+void CPU::dma_run() {
+  for(unsigned n = 0; n < 4; n++) {
+    if(regs.dma[n].control.enable == false) continue;
+    switch(regs.dma[n].control.timingmode) {
+    case 0: break;
+    case 1: if(pending.dma.vblank == false) continue; break;
+    case 2: if(pending.dma.hblank == false) continue; break;
+    case 3: if(pending.dma.hdma == false || n != 3) continue; break;
+    }
+    dma_transfer(n);
+  }
+
+  pending.dma.vblank = false;
+  pending.dma.hblank = false;
+  pending.dma.hdma   = false;
+}
+
+void CPU::dma_transfer(uint2 n) {
+  auto &channel = regs.dma[n];
+
+  unsigned size = channel.control.size ? Word : Half;
+  unsigned seek = channel.control.size ? 4 : 2;
+  uint16 length = channel.length;
+
+  channel.basetarget = channel.target;
+  do {
+    uint32 word = bus.read(channel.source, size);
+    bus.write(channel.target, size, word);
+
+    switch(channel.control.sourcemode) {
+    case 0: channel.source += seek; break;
+    case 1: channel.source -= seek; break;
+    }
+
+    switch(channel.control.targetmode) {
+    case 0: channel.target += seek; break;
+    case 1: channel.target -= seek; break;
+    case 3: channel.target += seek; break;
+    }
+  } while(--length);
+  if(channel.control.targetmode == 3) channel.target = channel.basetarget;
+
+  channel.control.enable = false;
+  if(channel.control.irq) regs.irq.flag.dma[n] = 1;
+}
diff --git a/bsnes/gba/cpu/registers.cpp b/bsnes/gba/cpu/registers.cpp
index e37aa269..40150c01 100755
--- a/bsnes/gba/cpu/registers.cpp
+++ b/bsnes/gba/cpu/registers.cpp
@@ -5,7 +5,7 @@ CPU::Registers::DMA::Control::operator uint16() const {
   | (repeat     <<  9)
   | (size       << 10)
   | (drq        << 11)
-  | (timing     << 12)
+  | (timingmode << 12)
   | (irq        << 14)
   | (enable     << 15)
   );
@@ -17,7 +17,7 @@ uint16 CPU::Registers::DMA::Control::operator=(uint16 source) {
   repeat     = source >>  9;
   size       = source >> 10;
   drq        = source >> 11;
-  timing     = source >> 12;
+  timingmode = source >> 12;
   irq        = source >> 14;
   enable     = source >> 15;
   return operator uint16();
@@ -78,15 +78,15 @@ CPU::Registers::Interrupt::operator uint16() const {
     (vblank       <<  0)
   | (hblank       <<  1)
   | (vcoincidence <<  2)
-  | (timer0       <<  3)
-  | (timer1       <<  4)
-  | (timer2       <<  5)
-  | (timer3       <<  6)
+  | (timer[0]     <<  3)
+  | (timer[1]     <<  4)
+  | (timer[2]     <<  5)
+  | (timer[3]     <<  6)
   | (serial       <<  7)
-  | (dma0         <<  8)
-  | (dma1         <<  9)
-  | (dma2         << 10)
-  | (dma3         << 11)
+  | (dma[0]       <<  8)
+  | (dma[1]       <<  9)
+  | (dma[2]       << 10)
+  | (dma[3]       << 11)
   | (keypad       << 12)
   | (cartridge    << 13)
   );
@@ -96,15 +96,15 @@ uint16 CPU::Registers::Interrupt::operator=(uint16 source) {
   vblank       = source & (1 <<  0);
   hblank       = source & (1 <<  1);
   vcoincidence = source & (1 <<  2);
-  timer0       = source & (1 <<  3);
-  timer1       = source & (1 <<  4);
-  timer2       = source & (1 <<  5);
-  timer3       = source & (1 <<  6);
+  timer[0]     = source & (1 <<  3);
+  timer[1]     = source & (1 <<  4);
+  timer[2]     = source & (1 <<  5);
+  timer[3]     = source & (1 <<  6);
   serial       = source & (1 <<  7);
-  dma0         = source & (1 <<  8);
-  dma1         = source & (1 <<  9);
-  dma2         = source & (1 << 10);
-  dma3         = source & (1 << 11);
+  dma[0]       = source & (1 <<  8);
+  dma[1]       = source & (1 <<  9);
+  dma[2]       = source & (1 << 10);
+  dma[3]       = source & (1 << 11);
   keypad       = source & (1 << 12);
   cartridge    = source & (1 << 13);
   return operator uint16();
diff --git a/bsnes/gba/cpu/registers.hpp b/bsnes/gba/cpu/registers.hpp
index eb5db46d..21f3a7ba 100755
--- a/bsnes/gba/cpu/registers.hpp
+++ b/bsnes/gba/cpu/registers.hpp
@@ -9,7 +9,7 @@ struct Registers {
       uint1 repeat;
       uint1 size;
       uint1 drq;
-      uint2 timing;
+      uint2 timingmode;
       uint1 irq;
       uint1 enable;
 
@@ -17,6 +17,9 @@ struct Registers {
       uint16 operator=(uint16 source);
       DMA& operator=(const DMA&) = delete;
     } control;
+
+    //internal
+    uint32 basetarget;
   } dma[4];
 
   struct TimerControl {
@@ -64,15 +67,9 @@ struct Registers {
     bool vblank;
     bool hblank;
     bool vcoincidence;
-    bool timer0;
-    bool timer1;
-    bool timer2;
-    bool timer3;
+    bool timer[4];
     bool serial;
-    bool dma0;
-    bool dma1;
-    bool dma2;
-    bool dma3;
+    bool dma[4];
     bool keypad;
     bool cartridge;
 
diff --git a/bsnes/gba/cpu/state.hpp b/bsnes/gba/cpu/state.hpp
new file mode 100755
index 00000000..4926a3d8
--- /dev/null
+++ b/bsnes/gba/cpu/state.hpp
@@ -0,0 +1,7 @@
+struct Pending {
+  struct DMA {
+    bool vblank;
+    bool hblank;
+    bool hdma;
+  } dma;
+} pending;
diff --git a/bsnes/gba/gba.hpp b/bsnes/gba/gba.hpp
index 36788a54..77f5db26 100755
--- a/bsnes/gba/gba.hpp
+++ b/bsnes/gba/gba.hpp
@@ -13,7 +13,7 @@ namespace GBA {
 
 /*
   bgba - Game Boy Advance emulator
-  author: byuu
+  authors: byuu, Cydrak
   license: GPLv3
   project started: 2012-03-19
 */
diff --git a/bsnes/gba/ppu/background.cpp b/bsnes/gba/ppu/background.cpp
new file mode 100755
index 00000000..3a8af01f
--- /dev/null
+++ b/bsnes/gba/ppu/background.cpp
@@ -0,0 +1,72 @@
+void PPU::render_backgrounds() {
+  if(regs.control.bgmode == 0) {
+    render_background_linear(0);
+    render_background_linear(1);
+    render_background_linear(2);
+    render_background_linear(3);
+  }
+
+  if(regs.control.bgmode == 1) {
+    render_background_linear(0);
+    render_background_linear(1);
+  //render_background_affine(2);
+  }
+
+  if(regs.control.bgmode == 2) {
+  //render_background_affine(2);
+  //render_background_affine(3);
+  }
+}
+
+void PPU::render_background_linear(unsigned bgnumber) {
+  for(unsigned n = 0; n < 240; n++) pixel[bgnumber][n].exists = false;
+  if(regs.control.enablebg[bgnumber] == false) return;
+
+  auto &bg = regs.bg[bgnumber];
+  uint9 voffset = regs.vcounter + bg.voffset;
+  uint9 hoffset = bg.hoffset;
+
+  unsigned basemap = bg.control.screenbaseblock    << 11;
+  unsigned basechr = bg.control.characterbaseblock << 14;
+  unsigned px = hoffset & 7, py = voffset & 7;
+
+  Tile tile;
+  uint8 data[8];
+
+  for(unsigned x = 0; x < 240; x++) {
+    if(x == 0 || px & 8) {
+      px &= 7;
+
+      unsigned tx = hoffset / 8, ty = voffset / 8;
+      unsigned offset = (ty & 31) * 32 + (tx & 31);
+      if(bg.control.screensize & 1) if(tx & 32) offset += 32 * 32;
+      if(bg.control.screensize & 2) if(ty & 32) offset += 32 * 32 * (1 + (bg.control.screensize & 1));
+      offset = basemap + offset * 2;
+      uint16 mapdata = vram.read(offset, Half);
+
+      tile.character = mapdata >>  0;
+      tile.hflip     = mapdata >> 10;
+      tile.vflip     = mapdata >> 11;
+      tile.palette   = mapdata >> 12;
+
+      if(bg.control.colormode == 0) {
+        offset = basechr + tile.character * 32 + (py ^ (tile.vflip ? 7 : 0)) * 4;
+        uint32 word = vram.read(offset, Word);
+        for(unsigned n = 0; n < 8; n++) data[n] = (word >> (n * 4)) & 15;
+      } else {
+        offset = basechr + tile.character * 64 + (py ^ (tile.vflip ? 7 : 0)) * 8;
+        uint32 wordlo = vram.read(offset + 0, Word);
+        uint32 wordhi = vram.read(offset + 4, Word);
+        for(unsigned n = 0; n < 4; n++) data[0 + n] = (wordlo >> (n * 8)) & 255;
+        for(unsigned n = 0; n < 4; n++) data[4 + n] = (wordhi >> (n * 8)) & 255;
+      }
+    }
+
+    hoffset++;
+    uint8 color = data[px++ ^ (tile.hflip ? 7 : 0)];
+    if(color == 0) continue;  //transparent
+
+    if(bg.control.colormode == 0) pixel[bgnumber][x] = { true, palette(tile.palette * 16 + color), bg.control.priority };
+    if(bg.control.colormode == 1) pixel[bgnumber][x] = { true, palette(color), bg.control.priority };
+  }
+}
diff --git a/bsnes/gba/ppu/object.cpp b/bsnes/gba/ppu/object.cpp
index 71493a48..3164f005 100755
--- a/bsnes/gba/ppu/object.cpp
+++ b/bsnes/gba/ppu/object.cpp
@@ -1,5 +1,6 @@
 void PPU::render_objects() {
-  for(unsigned n = 0; n < 240; n++) pixel[n].exists = false;
+  for(unsigned n = 0; n < 240; n++) pixel[4][n].exists = false;
+  if(regs.control.enableobj == false) return;
 
   for(unsigned n = 0; n < 128; n++) {
     auto &obj = object[n];
@@ -72,8 +73,8 @@ void PPU::render_object_linear(Object &obj) {
     if(obj.colors == 0) color = (px & 1) ? color >> 4 : color & 15;
     if(color == 0) continue;  //transparent
 
-    if(obj.colors == 0) pixel[sx] = { true, palette(256 + obj.palette * 16 + color), obj.priority };
-    if(obj.colors == 1) pixel[sx] = { true, palette(256 + color), obj.priority };
+    if(obj.colors == 0) pixel[4][sx] = { true, palette(256 + obj.palette * 16 + color), obj.priority };
+    if(obj.colors == 1) pixel[4][sx] = { true, palette(256 + color), obj.priority };
   }
 }
 
diff --git a/bsnes/gba/ppu/ppu.cpp b/bsnes/gba/ppu/ppu.cpp
index 1ad26b58..585445ef 100755
--- a/bsnes/gba/ppu/ppu.cpp
+++ b/bsnes/gba/ppu/ppu.cpp
@@ -13,6 +13,7 @@
 namespace GBA {
 
 #include "registers.cpp"
+#include "background.cpp"
 #include "object.cpp"
 #include "screen.cpp"
 #include "mmio.cpp"
@@ -85,6 +86,7 @@ void PPU::scanline() {
 
   if(regs.vcounter == 160) {
     if(regs.status.irqvblank) cpu.regs.irq.flag.vblank = 1;
+    cpu.pending.dma.vblank = true;
   }
 
   if(regs.status.irqvcoincidence) {
@@ -92,17 +94,21 @@ void PPU::scanline() {
   }
 
   if(regs.vcounter < 160) {
+    render_backgrounds();
     render_objects();
     render_screen();
   }
 
-  step(256 * 4);
+  step(1024);
   regs.status.hblank = 1;
   if(regs.status.irqhblank) cpu.regs.irq.flag.hblank = 1;
+  cpu.pending.dma.hblank = true;
 
-  step( 52 * 4);
+  step(200);
   regs.status.hblank = 0;
+  cpu.pending.dma.hdma = true;
 
+  step(8);
   if(++regs.vcounter == 228) regs.vcounter = 0;
 }
 
diff --git a/bsnes/gba/ppu/ppu.hpp b/bsnes/gba/ppu/ppu.hpp
index 36f1c18e..6dd6bc26 100755
--- a/bsnes/gba/ppu/ppu.hpp
+++ b/bsnes/gba/ppu/ppu.hpp
@@ -17,6 +17,9 @@ struct PPU : Thread, MMIO {
   uint8 read(uint32 addr);
   void write(uint32 addr, uint8 byte);
 
+  void render_backgrounds();
+  void render_background_linear(unsigned bgnumber);
+
   void render_objects();
   void render_object_linear(Object&);
   void render_object_affine(Object&);
diff --git a/bsnes/gba/ppu/screen.cpp b/bsnes/gba/ppu/screen.cpp
index 1a51b64c..80543e3d 100755
--- a/bsnes/gba/ppu/screen.cpp
+++ b/bsnes/gba/ppu/screen.cpp
@@ -9,7 +9,14 @@ void PPU::render_screen() {
   uint16 *line = output + regs.vcounter * 240;
 
   for(unsigned x = 0; x < 240; x++) {
-    if(pixel[x].exists) line[x] = pixel[x].color;
-    else line[x] = palette(0) & 0x7fff;
+    uint15 color = palette(0) & 0x7fff;
+    for(signed p = 3; p >= 0; p--) {
+      if(pixel[3][x].exists && pixel[3][x].priority == p) color = pixel[3][x].color;
+      if(pixel[2][x].exists && pixel[2][x].priority == p) color = pixel[2][x].color;
+      if(pixel[1][x].exists && pixel[1][x].priority == p) color = pixel[1][x].color;
+      if(pixel[0][x].exists && pixel[0][x].priority == p) color = pixel[0][x].color;
+      if(pixel[4][x].exists && pixel[4][x].priority == p) color = pixel[4][x].color;
+    }
+    line[x] = color;
   }
 }
diff --git a/bsnes/gba/ppu/state.hpp b/bsnes/gba/ppu/state.hpp
index 1234accb..37257320 100755
--- a/bsnes/gba/ppu/state.hpp
+++ b/bsnes/gba/ppu/state.hpp
@@ -2,7 +2,7 @@ struct Pixel {
   bool exists;
   uint15 color;
   uint2 priority;
-} pixel[256];
+} pixel[5][256];
 
 struct Object {
   uint8  y;
@@ -27,3 +27,10 @@ struct Object {
   unsigned width;
   unsigned height;
 } object[128];
+
+struct Tile {
+  uint10 character;
+  uint1  hflip;
+  uint1  vflip;
+  uint4  palette;
+};
diff --git a/bsnes/processor/arm/arm.cpp b/bsnes/processor/arm/arm.cpp
index 49844bcd..1a9a35f1 100755
--- a/bsnes/processor/arm/arm.cpp
+++ b/bsnes/processor/arm/arm.cpp
@@ -16,7 +16,6 @@ void ARM::power() {
   crash = false;
   r(15).modify = [&] {
     pipeline.reload = true;
-    r(15).data &= cpsr().t ? ~1 : ~3;
   };
 
   trace = false;
diff --git a/bsnes/processor/arm/instructions-arm.cpp b/bsnes/processor/arm/instructions-arm.cpp
index 0d31db6a..9807dda2 100755
--- a/bsnes/processor/arm/instructions-arm.cpp
+++ b/bsnes/processor/arm/instructions-arm.cpp
@@ -5,8 +5,8 @@ void ARM::arm_step() {
     pipeline.reload = false;
     r(15).data &= ~3;
 
-    pipeline.fetch.address = r(15);
-    pipeline.fetch.instruction = read(r(15), Word);
+    pipeline.fetch.address = r(15) & ~3;
+    pipeline.fetch.instruction = read(pipeline.fetch.address, Word);
 
     pipeline_step();
     step(2);
diff --git a/bsnes/processor/arm/instructions-thumb.cpp b/bsnes/processor/arm/instructions-thumb.cpp
index c7320a4a..db9e4f89 100755
--- a/bsnes/processor/arm/instructions-thumb.cpp
+++ b/bsnes/processor/arm/instructions-thumb.cpp
@@ -5,8 +5,8 @@ void ARM::thumb_step() {
     pipeline.reload = false;
     r(15).data &= ~1;
 
-    pipeline.fetch.address = r(15);
-    pipeline.fetch.instruction = read(r(15), Half);
+    pipeline.fetch.address = r(15) & ~1;
+    pipeline.fetch.instruction = read(pipeline.fetch.address, Half);
 
     pipeline_step();
     step(1);
diff --git a/bsnes/processor/arm/registers.cpp b/bsnes/processor/arm/registers.cpp
index 183a77ef..9292be3f 100755
--- a/bsnes/processor/arm/registers.cpp
+++ b/bsnes/processor/arm/registers.cpp
@@ -66,12 +66,12 @@ void ARM::pipeline_step() {
 
   if(cpsr().t == 0) {
     r(15).data += 4;
-    pipeline.fetch.address = r(15);
-    pipeline.fetch.instruction = read(r(15), Word);
+    pipeline.fetch.address = r(15) & ~3;
+    pipeline.fetch.instruction = read(pipeline.fetch.address, Word);
   } else {
     r(15).data += 2;
-    pipeline.fetch.address = r(15);
-    pipeline.fetch.instruction = read(r(15), Half);
+    pipeline.fetch.address = r(15) & ~1;
+    pipeline.fetch.instruction = read(pipeline.fetch.address, Half);
   }
 }