diff --git a/example/aica/aica.cpp b/example/aica/aica.cpp
index aae4850..f1bd373 100644
--- a/example/aica/aica.cpp
+++ b/example/aica/aica.cpp
@@ -35,10 +35,10 @@ void main()
 
   wait(); aica_sound.common.afsel_mslc_mobuf = aica::afsel_mslc_mobuf::MSLC(0);
   serial::string("mrwinh: ");
-  wait_read();
+  wait();
   serial::integer<uint8_t>(aica_sound.common.MRWINH());
   while (1) {
-    wait_read();
+    wait();
     serial::string("sgc: ");
     serial::integer<uint8_t>(aica_sound.common.SGC(), ' ');
     serial::string("; ca: ");
diff --git a/example/aica/aica.mk b/example/aica/aica.mk
index 79e789f..f2bc80c 100644
--- a/example/aica/aica.mk
+++ b/example/aica/aica.mk
@@ -9,7 +9,8 @@ example/aica/aica.elf: $(START_OBJ) $(AICA_OBJ)
 AICA_XM_OBJ = \
 	example/aica/aica_xm.o \
 	sh7091/serial.o \
-	example/arm/xm.bin.o
+	example/arm/xm.bin.o \
+	example/arm/channel.bin.o
 
 example/aica/aica_xm.elf: LDSCRIPT = $(LIB)/main.lds
 example/aica/aica_xm.elf: $(START_OBJ) $(AICA_XM_OBJ)
diff --git a/example/aica/aica_xm.cpp b/example/aica/aica_xm.cpp
index 4d1962f..b96743e 100644
--- a/example/aica/aica_xm.cpp
+++ b/example/aica/aica_xm.cpp
@@ -3,9 +3,16 @@
 #include "systembus.hpp"
 #include "systembus_bits.hpp"
 #include "aica/aica.hpp"
+#include "sh7091/sh7091.hpp"
+#include "sh7091/sh7091_bits.hpp"
+
+#include "assert.h"
 
 #include "example/arm/xm.bin.h"
 
+extern void * _binary_start __asm("_binary_example_arm_channel_bin_start");
+extern void * _binary_size __asm("_binary_example_arm_channel_bin_size");
+
 void wait()
 {
   uint32_t ffst = system.FFST;
@@ -16,35 +23,97 @@ void wait()
   };
 }
 
+constexpr uint32_t dma_address_mask = 0x1fffffe0;
+
+void g2_aica_dma(uint32_t g2_address, uint32_t system_address, int length)
+{
+  using namespace dmac;
+
+  length = (length + 31) & (~31);
+
+  // is DMAOR needed?
+  sh7091.DMAC.DMAOR = dmaor::ddt::on_demand_data_transfer_mode       /* on-demand data transfer mode */
+                    | dmaor::pr::ch2_ch0_ch1_ch3                     /* priority mode; CH2 > CH0 > CH1 > CH3 */
+                    | dmaor::dme::operation_enabled_on_all_channels; /* DMAC master enable */
+
+
+  g2_if.G2APRO = 0x4659007f; // disable protection
+
+  g2_if.ADSTAG = dma_address_mask & g2_address; // G2 address
+  g2_if.ADSTAR = dma_address_mask & system_address; // system memory address
+  g2_if.ADLEN = length;
+  g2_if.ADDIR = 0; // from root bus to G2 device
+  g2_if.ADTSEL = 0; // CPU controlled trigger
+  g2_if.ADEN = 1; // enable G2-DMA
+  g2_if.ADST = 1; // start G2-DMA
+}
+
+void g2_aica_dma_wait_complete()
+{
+  // wait for maple DMA completion
+  while ((system.ISTNRM & istnrm::end_of_dma_aica_dma) == 0);
+  system.ISTNRM = istnrm::end_of_dma_aica_dma;
+}
+
+uint8_t __attribute__((aligned(32))) zero[32768] = {0};
+
 void main()
 {
   serial::init(0);
-  const uint32_t * binary = reinterpret_cast<uint32_t *>(&_binary_example_arm_xm_bin_start);
-  const uint32_t binary_size = reinterpret_cast<uint32_t>(&_binary_example_arm_xm_bin_size);
+
+  const int start = reinterpret_cast<int>(&_binary_start);
+  const int size = reinterpret_cast<int>(&_binary_size);
 
   wait(); aica_sound.common.vreg_armrst = aica::vreg_armrst::ARMRST(1);
   wait(); aica_sound.common.dmea0_mrwinh = aica::dmea0_mrwinh::MRWINH(0b0111);
-  for (uint32_t i = 0; i < binary_size / 4; i++) {
-    // copy
-    while (aica_wave_memory[i] != binary[i]) {
-      wait();
-      aica_wave_memory[i] = binary[i];
+  system.ISTNRM = istnrm::end_of_dma_aica_dma;
+
+  g2_aica_dma((uint32_t)&aica_sound, (int)zero, 32768);
+  g2_aica_dma_wait_complete();
+  assert(g2_if.ADST == 0);
+  g2_aica_dma((uint32_t)aica_wave_memory, start, size);
+  g2_aica_dma_wait_complete();
+  assert(g2_if.ADST == 0);
+
+  for (int i = 0; i < size / 4; i++) {
+    wait();
+    assert(aica_wave_memory[i] == ((uint32_t*)start)[i]);
+  }
+
+  wait(); aica_sound.common.dmea0_mrwinh = aica::dmea0_mrwinh::MRWINH(0b0001);
+
+  wait(); aica_sound.channel[0].KYONB(1);
+  wait(); aica_sound.channel[0].LPCTL(1);
+  wait(); aica_sound.channel[0].PCMS(0);
+  wait(); aica_sound.channel[0].LSA(0);
+  wait(); aica_sound.channel[0].LEA(44100);
+  wait(); aica_sound.channel[0].D2R(0x0);
+  wait(); aica_sound.channel[0].D1R(0x0);
+  wait(); aica_sound.channel[0].RR(0x0);
+  wait(); aica_sound.channel[0].AR(0x1f);
+
+  wait(); aica_sound.channel[0].OCT(0);
+  wait(); aica_sound.channel[0].FNS(0);
+  wait(); aica_sound.channel[0].DISDL(0xf);
+  wait(); aica_sound.channel[0].DIPAN(0x0);
+
+  wait(); aica_sound.channel[0].Q(0b00100);
+  wait(); aica_sound.channel[0].TL(0);
+  wait(); aica_sound.channel[0].LPOFF(1);
+
+  wait(); aica_sound.common.mono_mem8mb_dac18b_ver_mvol =
+      aica::mono_mem8mb_dac18b_ver_mvol::MONO(0)   // enable panpots
+    | aica::mono_mem8mb_dac18b_ver_mvol::MEM8MB(0) // 16Mbit SDRAM
+    | aica::mono_mem8mb_dac18b_ver_mvol::DAC18B(0) // 16-bit DAC
+    | aica::mono_mem8mb_dac18b_ver_mvol::MVOL(0xf) // 15/15 volume
+    ;
+
+  wait(); aica_sound.channel[0].SA(44100 * 2);
+  wait(); aica_sound.channel[0].KYONEX(1);
+
+  while (1) {
+    for (int i = 0; i < 10000000; i++) {
+      asm volatile ("nop");
     }
   }
-  wait(); aica_sound.common.dmea0_mrwinh = aica::dmea0_mrwinh::MRWINH(0b0001);
-  wait(); aica_sound.common.vreg_armrst = aica::vreg_armrst::ARMRST(0);
-
-  wait(); aica_sound.common.afsel_mslc_mobuf = aica::afsel_mslc_mobuf::MSLC(0);
-  serial::string("mrwinh: ");
-  wait();
-  serial::integer<uint8_t>(aica_sound.common.MRWINH());
-  int last_dram = -1;
-  while (1) {
-    wait();
-    int read = aica_wave_memory[0];
-    if (read != last_dram) {
-      serial::integer<uint32_t>(read);
-    }
-    last_dram = read;
-  };
 }
diff --git a/example/arm/Makefile b/example/arm/Makefile
index 65ff063..66dbf59 100644
--- a/example/arm/Makefile
+++ b/example/arm/Makefile
@@ -2,7 +2,7 @@ MAKEFILE_PATH := $(abspath $(lastword $(MAKEFILE_LIST)))
 DIR := $(dir $(MAKEFILE_PATH))
 
 LIB ?= .
-OPT ?= -O1
+OPT ?= -Og
 GENERATED ?=
 
 AARCH = -march=armv4 -mlittle-endian
diff --git a/example/arm/xm.cpp b/example/arm/xm.cpp
index 92d9e21..ae9e756 100644
--- a/example/arm/xm.cpp
+++ b/example/arm/xm.cpp
@@ -11,9 +11,12 @@ struct xm_state {
   xm_header_t * header;
   xm_pattern_header_t * pattern_header[max_patterns];
   xm_instrument_header_t * instrument_header[max_instruments];
-  xm_sample_header_t * sample_header; // array
+  xm_sample_header_t * sample_header[max_instruments]; // array
+  int sample_data_address[max_instruments];
 };
 
+uint8_t __attribute__((aligned(2))) sample_data[512 * 1024];
+
 static xm_state xm = {0};
 
 int s16(void * buf)
@@ -30,20 +33,6 @@ int s32(void * buf)
   return v;
 }
 
-int xm_samples_init(int buf, int offset, int number_of_samples)
-{
-  xm_sample_header_t * sample_header[number_of_samples];
-  for (int i = 0; i < number_of_samples; i++) {
-    sample_header[i] = (xm_sample_header_t *)(buf + offset);
-    offset += (sizeof (xm_sample_header_t));
-  }
-
-  for (int i = 0; i < number_of_samples; i++) {
-    offset += s32(&sample_header[i]->sample_length);
-  }
-  return offset;
-}
-
 void print(int i)
 {
   for (int i = 0; i < 100000; i++) {
@@ -52,6 +41,66 @@ void print(int i)
   dram[0] = i;
 }
 
+int unpack_sample(int buf, int offset, int sample_ix, xm_sample_header_t * sample_header, int sample_data_ix)
+{
+  int size = s32(&sample_header->sample_length);
+  if (sample_header->type & (1 << 4)) { // 16-bit samples
+    int num_samples = size / 2;
+    int old = 0;
+    volatile int16_t * out = (volatile int16_t *)(&sample_data[sample_data_ix]);
+    print((int)out);
+    int16_t * in = (int16_t *)(buf + offset);
+    for (int i = 0; i < num_samples; i++) {
+      old += s16(&in[i]);
+      while (((uint16_t)out[i]) != ((uint16_t)old)) {
+        out[i] = old;
+        print((uint16_t)old);
+        print((uint16_t)out[i]);
+      }
+    }
+  } else { // 8-bit
+    int num_samples = size;
+    int old = 0;
+    volatile int8_t * out = (volatile int8_t *)(&sample_data[sample_data_ix]);
+    int8_t * in = (int8_t *)(buf + offset);
+    for (int i = 0; i < num_samples; i++) {
+      old += in[i];
+      out[i] = old;
+    }
+  }
+
+  if (size & 1) {
+    size += 1;
+  }
+
+  return size;
+}
+
+static int sample_data_ix;
+
+int xm_samples_init(int buf, int offset, int instrument_ix, int number_of_samples)
+{
+  xm_sample_header_t * sample_header[number_of_samples];
+  xm.sample_header[instrument_ix] = (xm_sample_header_t *)(buf + offset);
+  for (int i = 0; i < number_of_samples; i++) {
+    sample_header[i] = (xm_sample_header_t *)(buf + offset);
+    offset += (sizeof (xm_sample_header_t));
+  }
+
+  if (number_of_samples > 0) {
+    print(instrument_ix);
+    print(sample_data_ix);
+  }
+  for (int i = 0; i < number_of_samples; i++) {
+    if (s32(&sample_header[i]->sample_length) > 0) {
+      xm.sample_data_address[instrument_ix] = (int)(&sample_data[sample_data_ix]);
+      sample_data_ix += unpack_sample(buf, offset, instrument_ix, sample_header[i], sample_data_ix);
+    }
+    offset += s32(&sample_header[i]->sample_length);
+  }
+  return offset;
+}
+
 void xm_init()
 {
   int buf = (int)(&_binary_xm_milkypack01_xm_start);
@@ -74,6 +123,8 @@ void xm_init()
     print(offset);
   }
 
+  print(0xaaaaaaaa);
+  sample_data_ix = 0;
   int number_of_instruments = s16(&xm.header->number_of_instruments);
   for (int i = 0; i < number_of_instruments; i++) {
     xm_instrument_header_t * instrument_header = (xm_instrument_header_t *)(buf + offset);
@@ -82,7 +133,7 @@ void xm_init()
     offset += s32(&instrument_header->instrument_size);
 
     int number_of_samples = s16(&instrument_header->number_of_samples);
-    offset = xm_samples_init(buf, offset, number_of_samples);
+    offset = xm_samples_init(buf, offset, i, number_of_samples);
   }
 
   print(0x11223344);
@@ -112,7 +163,7 @@ void main()
   aica_sound.channel[0].RR(0x0);
   aica_sound.channel[0].AR(0x1f);
 
-  aica_sound.channel[0].OCT(0);
+  aica_sound.channel[0].OCT(-1);
   aica_sound.channel[0].FNS(0);
   aica_sound.channel[0].DISDL(0xf);
   aica_sound.channel[0].DIPAN(0x0);
@@ -128,17 +179,26 @@ void main()
     | aica::mono_mem8mb_dac18b_ver_mvol::MVOL(0xf) // 15/15 volume
     ;
 
-  uint32_t segment = 0;
-
   constexpr uint32_t timer_a_interrupt = (1 << 6);
   aica_sound.common.scire = timer_a_interrupt;
-  bool started = 0;
 
   xm_init();
 
+  print(xm.sample_data_address[0]);
+  aica_sound.channel[0].SA(xm.sample_data_address[0]);
+  int lsa = xm.sample_header[0]->sample_loop_start / 2;
+  int lea = xm.sample_header[0]->sample_loop_length / 2;
+  print(lsa);
+  print(lea);
+  aica_sound.channel[0].LSA(lsa);
+  aica_sound.channel[0].LEA(lsa + lea);
+  aica_sound.channel[0].KYONB(1);
+  aica_sound.channel[0].KYONEX(1);
+
   while (1) {
+    /*
     if (!started || (aica_sound.common.SCIPD() & timer_a_interrupt)) {
-      //aica_sound.channel[0].SA(next_sa);
+      aica_sound.channel[0].SA();
       aica_sound.common.tactl_tima =
           aica::tactl_tima::TACTL(0)  // increment once every 128 samples
         | aica::tactl_tima::TIMA(256 - 128) // interrupt after 128 counts
@@ -148,5 +208,6 @@ void main()
 
       aica_sound.common.scire = timer_a_interrupt;
     }
+    */
   }
 }