cartridge: add adpcm example

2024-09-19 21:04:44 -05:00 · 2024-09-19 21:04:44 -05:00 · f275732dfc
commit f275732dfc
parent f18944e523
4 changed files with 410 additions and 1 deletions
--- a/cartridge/Makefile
+++ b/cartridge/Makefile
@ -8,7 +8,7 @@ CARCH = -m2 -mb

 ### general flags ###

-OPT ?= -O3
+OPT ?= -O2

 DEBUG = -g -gdwarf-4

@ -63,6 +63,12 @@ OBJDUMP = $(TARGET)objdump
 		--rename-section .data=.rom.$(basename $@) \
 		$< $@

+%.adpcm.o: %.adpcm
+	$(OBJCOPY) \
+		-I binary -O elf32-sh -B sh2 \
+		--rename-section .data=.rom.$(basename $@) \
+		$< $@
+
 ### object files ###

 SYS_IP_OBJ += $(LIB)/ip/sys_id.o
@ -83,6 +89,10 @@ CARTRIDGE_OBJ = main.o wc3.pcm.o
 cartridge.elf: LDSCRIPT = $(LIB)/cartridge.lds
 cartridge.elf: $(SYS_IP_OBJ) $(COMMON_OBJ) $(CARTRIDGE_OBJ)

+ADPCM_OBJ = adpcm.o ecclesia.adpcm.o
+adpcm.elf: LDSCRIPT = $(LIB)/cartridge.lds
+adpcm.elf: $(SYS_IP_OBJ) $(COMMON_OBJ) $(ADPCM_OBJ)
+
 .SUFFIXES:
 .INTERMEDIATE:
 .SECONDARY:
--- a/cartridge/adpcm.cpp
+++ b/cartridge/adpcm.cpp
@ -0,0 +1,220 @@
+#include <stdint.h>
+
+#include "vdp1.h"
+#include "vdp2.h"
+#include "scsp.h"
+#include "smpc.h"
+
+#include "../common/copy.hpp"
+
+extern void * _ecclesia_adpcm_start __asm("_binary_ecclesia_adpcm_start");
+extern void * _ecclesia_adpcm_size __asm("_binary_ecclesia_adpcm_size");
+
+// table of index changes
+static const int8_t index_table[16] = {
+    -1, -1, -1, -1, 2, 4, 6, 8,
+    -1, -1, -1, -1, 2, 4, 6, 8
+};
+
+// quantizer lookup table
+static const int16_t step_size_table[89] = {
+     7      , 8     , 9     , 10    , 11    , 12    , 13    , 14    ,
+     16     , 17    , 19    , 21    , 23    , 25    , 28    , 31    ,
+     34     , 37    , 41    , 45    , 50    , 55    , 60    , 66    ,
+     73     , 80    , 88    , 97    , 107   , 118   , 130   , 143   ,
+     157    , 173   , 190   , 209   , 230   , 253   , 279   , 307   ,
+     337    , 371   , 408   , 449   , 494   , 544   , 598   , 658   ,
+     724    , 796   , 876   , 963   , 1060  , 1166  , 1282  , 1411  ,
+     1552   , 1707  , 1878  , 2066  , 2272  , 2499  , 2749  , 3024  ,
+     3327   , 3660  , 4026  , 4428  , 4871  , 5358  , 5894  , 6484  ,
+     7132   , 7845  , 8630  , 9493  , 10442 , 11487 , 12635 , 13899 ,
+     15289  , 16818 , 18500 , 20350 , 22385 , 24623 , 27086 , 29794 ,
+     32767
+};
+
+
+struct adpcm_state {
+  int32_t predicted_sample;
+  int32_t index;
+  int32_t step_size;
+};
+
+static struct adpcm_state state;
+
+void __attribute__ ((noinline)) init_state()
+{
+  state.predicted_sample = 0;
+  state.index = 0;
+  state.step_size = 7;
+}
+
+int16_t decode_sample(int sample)
+{
+  int difference = 0;
+  if (sample & 0b100)
+    difference += state.step_size;
+  if (sample & 0b010)
+    difference += state.step_size >> 1;
+  if (sample & 0b001)
+    difference += state.step_size >> 2;
+  difference += state.step_size >> 3;
+
+  if (sample & 0b1000) {
+    difference = -difference;
+  }
+
+  state.predicted_sample += difference;
+  if (state.predicted_sample > 32767)
+    state.predicted_sample = 32767;
+  if (state.predicted_sample < -32768)
+    state.predicted_sample = -32768;
+
+  state.index += index_table[sample];
+  if (state.index < 0)
+    state.index = 0;
+  if (state.index > 88)
+    state.index = 88;
+  state.step_size = step_size_table[state.index];
+
+  return state.predicted_sample;
+}
+
+void snd_init()
+{
+  while ((smpc.reg.SF & 1) != 0);
+  smpc.reg.SF = 1;
+  smpc.reg.COMREG = COMREG__SNDOFF;
+  while (smpc.reg.OREG[31].val != OREG31__SNDOFF);
+
+  scsp.reg.ctrl.MIXER = MIXER__MEM4MB;
+
+  volatile uint32_t * dsp_steps = reinterpret_cast<volatile uint32_t *>(&(scsp.reg.dsp.STEP[0].MPRO[0]));
+  fill<volatile uint32_t>(dsp_steps, 0, (sizeof (scsp.reg.dsp.STEP)));
+
+  for (int i = 0; i < 32; i++) {
+    scsp.reg.slot[i].SA = 0;
+    scsp.reg.slot[i].MIXER = 0;
+  }
+
+  scsp.reg.ctrl.MIXER = MIXER__MEM4MB | MIXER__MVOL(0xf);
+}
+
+void __attribute__ ((noinline)) decode_chunk(volatile uint16_t * dst, const uint8_t * src, int samples)
+{
+  for (int i = 0; i < samples / 2; i++) {
+    uint8_t b = src[i];
+    int nib0 = (b >> 0) & 0xf;
+    int nib1 = (b >> 4) & 0xf;
+    dst[i * 2 + 0] = decode_sample(nib0);
+    dst[i * 2 + 1] = decode_sample(nib1);
+  }
+}
+
+void snd_step()
+{
+  const uint8_t * buf = reinterpret_cast<uint8_t*>(&_ecclesia_adpcm_start);
+  const uint32_t size = reinterpret_cast<uint32_t>(&_ecclesia_adpcm_size);
+  constexpr uint32_t chunk_samples = 16384 / 2;
+  decode_chunk(&scsp.ram.u16[0], buf, chunk_samples);
+
+  scsp_slot& slot = scsp.reg.slot[0];
+  // start address (bytes)
+  slot.SA = SA__KYONB | SA__LPCTL__NORMAL | SA__SA(0); // kx kb sbctl[1:0] ssctl[1:0] lpctl[1:0] 8b sa[19:0]
+  slot.LSA = 0; // loop start address (samples)
+  slot.LEA = chunk_samples * 2; // loop end address (samples)
+  //slot.EG = EG__EGHOLD; // d2r d1r ho ar krs dl rr
+  slot.EG = EG__AR(0x1F) | EG__D1R(0x00) | EG__D2R(0x00) | EG__RR(0x1F);
+  slot.FM = 0; // stwinh sdir tl mdl mdxsl mdysl
+  slot.PITCH = PITCH__OCT(-1) | PITCH__FNS(0); // oct fns
+  slot.LFO = 0; // lfof plfows
+  slot.MIXER = MIXER__DISDL(0b110); // disdl dipan efsdl efpan
+
+  slot.LOOP |= LOOP__KYONEX;
+
+  uint32_t offset = 1;
+  uint32_t chunk = 1;
+
+  constexpr uint32_t timer_a_interrupt = (1 << 6);
+  scsp.reg.ctrl.TIMA = TIMA__TACTL(7)
+                     | TIMA__TIMA(128);
+  scsp.reg.ctrl.SCIRE = timer_a_interrupt;
+
+  while (1) {
+    decode_chunk(&scsp.ram.u16[chunk_samples * chunk],
+		 &buf[(chunk_samples * offset) / 2],
+		 chunk_samples);
+    chunk = !chunk;
+    offset = offset + 1;
+    if (chunk_samples * offset > size * 2) {
+      offset = 0;
+    }
+
+    /*
+    uint32_t sample = 0;
+    const uint32_t target = chunk_samples * 2;
+    constexpr uint32_t sample_interval = (1 << 10);
+    while (sample < target) {
+      scsp.reg.ctrl.SCIRE = sample_interval;
+      while (!(scsp.reg.ctrl.SCIPD & sample_interval));
+      sample++;
+    }
+    */
+    while (!(scsp.reg.ctrl.SCIPD & timer_a_interrupt));
+    scsp.reg.ctrl.TIMA = TIMA__TACTL(7)
+                       | TIMA__TIMA(128);
+    scsp.reg.ctrl.SCIRE = timer_a_interrupt;
+  }
+}
+
+extern "C"
+void main() __attribute__((section(".text.main")));
+void main()
+{
+  // DISP: Please make sure to change this bit from 0 to 1 during V blank.
+  vdp2.reg.TVMD = ( TVMD__DISP | TVMD__LSMD__NON_INTERLACE
+                  | TVMD__VRESO__240 | TVMD__HRESO__NORMAL_320);
+
+  // VDP2 User's Manual:
+  // "When sprite data is in an RGB format, sprite register 0 is selected"
+  // "When the value of a priority number is 0h, it is read as transparent"
+  //
+  // The power-on value of PRISA is zero. Set the priority for sprite register 0
+  // to some number greater than zero, so that the color data is not interpreted
+  // as "transparent".
+  vdp2.reg.PRISA = PRISA__S0PRIN(1); // Sprite register 0 Priority Number
+
+  /* TVM settings must be performed from the second H-blank IN interrupt after the
+  V-blank IN interrupt to the H-blank IN interrupt immediately after the V-blank
+  OUT interrupt. */
+  // "normal" display resolution, 16 bits per pixel, 512x256 framebuffer
+  vdp1.reg.TVMR = TVMR__TVM__NORMAL;
+
+  // swap framebuffers every 1 cycle; non-interlace
+  vdp1.reg.FBCR = 0;
+
+  // erase upper-left coordinate
+  vdp1.reg.EWLR = EWLR__16BPP_X1(0) | EWLR__Y1(0);
+
+  // erase lower-right coordinate
+  vdp1.reg.EWRR = EWRR__16BPP_X3(319) | EWRR__Y3(239);
+
+  // during a framebuffer erase cycle, write the color "black" to each pixel
+  const uint16_t black = 1 << 15 | 0x00ff;
+  vdp1.reg.EWDR = black;
+
+  vdp1.vram.cmd[0].CTRL = CTRL__END;
+
+  init_state();
+  snd_init();
+  snd_step();
+
+  const uint16_t blue = 1 << 15 | 0x7f00;
+  vdp1.reg.EWDR = blue;
+
+  vdp2.reg.BGON = 0;
+
+  // start drawing (execute the command list) on every frame
+  vdp1.reg.PTMR = PTMR__PTM__FRAME_CHANGE;
+
+  while (1);
+}
--- a/cartridge/adpcm.py
+++ b/cartridge/adpcm.py
@ -0,0 +1,179 @@
+from dataclasses import dataclass
+import struct
+
+# Table of index changes
+index_table = [
+    -1, -1, -1, -1, 2, 4, 6, 8,
+    -1, -1, -1, -1, 2, 4, 6, 8
+]
+assert len(index_table) == 16
+
+# quantizer lookup table
+step_size_table = [
+     7      , 8     , 9     , 10    , 11    , 12    , 13    , 14    ,
+     16     , 17    , 19    , 21    , 23    , 25    , 28    , 31    ,
+     34     , 37    , 41    , 45    , 50    , 55    , 60    , 66    ,
+     73     , 80    , 88    , 97    , 107   , 118   , 130   , 143   ,
+     157    , 173   , 190   , 209   , 230   , 253   , 279   , 307   ,
+     337    , 371   , 408   , 449   , 494   , 544   , 598   , 658   ,
+     724    , 796   , 876   , 963   , 1060  , 1166  , 1282  , 1411  ,
+     1552   , 1707  , 1878  , 2066  , 2272  , 2499  , 2749  , 3024  ,
+     3327   , 3660  , 4026  , 4428  , 4871  , 5358  , 5894  , 6484  ,
+     7132   , 7845  , 8630  , 9493  , 10442 , 11487 , 12635 , 13899 ,
+     15289  , 16818 , 18500 , 20350 , 22385 , 24623 , 27086 , 29794 ,
+     32767
+]
+assert len(step_size_table) == 89
+
+@dataclass
+class ADPCMState:
+    predicted_sample: int
+    index: int
+    step_size: int
+
+def encode_sample(state, sample):
+    assert sample >= -32768 and sample <= 32767
+    difference = sample - state.predicted_sample
+    if difference >= 0:
+        new_sample = 0b0000
+    else:
+        new_sample = 0b1000
+        difference = -difference
+    mask = 0b100
+
+    temp_step_size = state.step_size
+    # division through repeated subtraction
+    for i in range(3):
+        if difference >= temp_step_size:
+            new_sample |= mask
+            difference -= temp_step_size
+        temp_step_size >>= 1
+        mask >>= 1
+
+    difference = 0
+    if new_sample & 0b100:
+        difference += state.step_size
+    if new_sample & 0b010:
+        difference += state.step_size >> 1
+    if new_sample & 0b001:
+        difference += state.step_size >> 2
+    difference += state.step_size >> 3
+
+    if new_sample & 0b1000:
+        difference = -difference
+
+    state.predicted_sample += difference
+    if state.predicted_sample > 32767:
+        state.predicted_sample = 32767
+    elif state.predicted_sample < -32768:
+        state.predicted_sample = -32768
+
+    state.index += index_table[new_sample]
+    if state.index < 0:
+        state.index = 0
+    elif state.index > 88:
+        state.index = 88
+
+    state.step_size = step_size_table[state.index]
+
+    return new_sample
+
+def sign_extend(value, bits):
+    sign_bit = 1 << (bits - 1)
+    return (value & (sign_bit - 1)) - (value & sign_bit)
+
+def test_encode_sample():
+    sample = sign_extend(0x873f, 16)
+    state = ADPCMState(
+        predicted_sample = sign_extend(0x8700, 16),
+        index = 24,
+        step_size = 73
+    )
+
+    new_sample = encode_sample(state, sample)
+    assert new_sample == 3
+    assert (state.predicted_sample & 0xffff) == 0x873f, hex(state.predicted_sample)
+    assert state.index == 23
+    assert state.step_size == step_size_table[23]
+
+test_encode_sample()
+
+def decode_sample(state, sample):
+    difference = 0
+    if sample & 0b100:
+        difference += state.step_size
+    if sample & 0b010:
+        difference += state.step_size >> 1
+    if sample & 0b001:
+        difference += state.step_size >> 2
+    difference += state.step_size >> 3
+
+    if sample & 0b1000:
+        difference = -difference
+
+    state.predicted_sample += difference
+    if state.predicted_sample > 32767:
+        state.predicted_sample = 32767
+    elif state.predicted_sample < -32768:
+        state.predicted_sample = -32768
+
+    state.index += index_table[sample]
+    if state.index < 0:
+        state.index = 0
+    elif state.index > 88:
+        state.index = 88
+    state.step_size = step_size_table[state.index]
+
+    return state.predicted_sample
+
+def test_decode_sample():
+    sample = 3
+    state = ADPCMState(
+        predicted_sample = sign_extend(0x8700, 16),
+        index = 24,
+        step_size = 73
+    )
+
+    new_sample = decode_sample(state, sample)
+    assert new_sample == sign_extend(0x873f, 16)
+    assert state.index == 23
+    assert state.step_size == step_size_table[23]
+
+test_decode_sample()
+
+with open('ecclesia.pcm', 'rb') as f:
+    pcm = f.read()
+
+def nibbles(buf):
+    state = ADPCMState(
+        0, 0, 7
+    )
+    byte = 0
+    for i in range(len(buf) // 2):
+        sample, = struct.unpack('>h', buf[i * 2:i * 2+2])
+        new_sample = encode_sample(state, sample)
+        assert new_sample >= 0 and new_sample <= 15
+        if i % 2 == 0:
+            byte = new_sample
+        else:
+            byte |= (new_sample << 4)
+            yield byte
+
+with open('ecclesia.adpcm', 'wb') as f:
+    f.write(bytes(nibbles(pcm)))
+
+with open('ecclesia.adpcm', 'rb') as f:
+    adpcm = f.read()
+
+def unnibbles(buf):
+    state = ADPCMState(
+        0, 0, 7
+    )
+    for i in range(len(buf) * 2):
+        nib = (buf[i // 2] >> (4 * (i % 2))) & 0xf
+        new_sample = decode_sample(state, nib)
+        yield struct.pack('>h', new_sample)
+
+with open('ecclesia.adpcm.pcm', 'wb') as f:
+    for b in unnibbles(adpcm):
+        f.write(b)
--- a/cartridge/ecclesia.adpcm
+++ b/cartridge/ecclesia.adpcm