2taud converters refactoring

Implemented in it2taud.py:
- _parse_it_pf_envelope_raw() (it2taud.py:677) — parses IT's third envelope at IMPI+0x1D4, keeping all 25 nodes (no decimation), distinguishing pitch vs filter mode via flag bit 7.
- _env_value_at() — tick-time linear interpolation honouring env-loop wrap.
- _clone_sample(), _plan_baked_length() — sample copy and entry + N×loop_len length planner (N up to 16).
- _bake_pitch_envelope() — time-varying linear-interpolated resampling, rate = 2^(env_v/12).
- _bake_filter_envelope() — RBJ 2-pole resonant LP biquad with time-varying coefficients; cutoff mapped 110 Hz (env_v=−32) → ~28 kHz (env_v=+32), Q from inst.ifr ∈ [0.5, 6.0].
- ITInstrument extended with pf_nodes, pf_flags, ifc, ifr. parse_instruments() reads IFC/IFR at IMPI+0x39/0x3A and pf envelope at IMPI+0x1D4.
- assemble_taud() use_instruments branch now substitutes baked copies in the proxy[] list (originals in samples[] stay intact).
- --no-pf-envelope CLI flag for A/B testing; module docstring updated.

assets/disk0/tvdos/bin/taut.js

@@ -196,7 +196,7 @@ sym:[`C${sym.accnull}`,`C${sym.sharp}`,`D${sym.accnull}`,`D${sym.sharp}`,`E${sym
 sym:[`C${sym.accnull}`,`C${sym.sharp}`,`D${sym.accnull}`,`D${sym.sharp}`,`E${sym.accnull}`,`F${sym.accnull}`,`F${sym.sharp}`,`G${sym.accnull}`,`G${sym.sharp}`,`A${sym.accnull}`,`A${sym.sharp}`,`B${sym.accnull}`]},
 10122:{index:10122,name:"Pythagorean Augmented Fourth", table:[0x0,0x134,0x2B8,0x3EC,0x570,0x6A4,0x828,0x95C,0xA90,0xC14,0xD48,0xECC],
 sym:[`C${sym.accnull}`,`C${sym.sharp}`,`D${sym.accnull}`,`D${sym.sharp}`,`E${sym.accnull}`,`F${sym.accnull}`,`F${sym.sharp}`,`G${sym.accnull}`,`G${sym.sharp}`,`A${sym.accnull}`,`A${sym.sharp}`,`B${sym.accnull}`]},
-10123:{index:10123,name:"Shi'er lu",                         table:[0x0,0x184,0x2B8,0x43C,0x570,0x6F4,0x828,0x95C,0xAE0,0xC14,0xD98,0xECC],
+10123:{index:10123,name:"\u00FC\u00FD\u00FE",         table:[0x0,0x184,0x2B8,0x43C,0x570,0x6F4,0x828,0x95C,0xAE0,0xC14,0xD98,0xECC],
 sym:[` \u00E0\u00E1`,` \u00E2\u00E3`,` \u00E4\u00E5`,` \u00E6\u00E7`,` \u00E8\u00E9`,` \u00EA\u00EB`,` \u00EC\u00ED`,` \u00EE\u00EF`,` \u00F0\u00F1`,` \u00F2\u00F3`,` \u00F4\u00F5`,` \u00F6\u00F7`]},
 
 
@@ -218,6 +218,7 @@ const colBackPtn = 255
 let PITCH_PRESET_IDX = 240 // TODO read from the Project Data section of the .taud
 let beatDivPrimary = 4 // TODO read from the Project Data section of the .taud
 let beatDivSecondary = 16
+let hasUnsavedChanges = false
 
 // pitchSymLut[pitchInOct] = [symString, octaveOffset]
 // octaveOffset is 1 when pitchInOct is closer to the next octave's root (wraps up) than to any table entry.
@@ -675,7 +676,7 @@ function drawStatusBar() {
 
     // beat indicator
     let beatCursorRow = cursorRow
-    while (beatCursorRow > beatDivSecondary) { beatCursorRow -= beatDivSecondary } // test this behaviour with primary=4, secondary=22 or something
+    while (beatCursorRow >= beatDivSecondary) { beatCursorRow -= beatDivSecondary }
     let beatInd = (playbackMode != PLAYMODE_NONE && beatCursorRow % beatDivPrimary < (beatDivPrimary >>> 1)) ?
      ((beatCursorRow % beatDivSecondary < (beatDivPrimary >>> 1)) ? '\u00846u' : '\u00847u') :
      ''
@@ -850,18 +851,22 @@ function drawControlHint() {
         [`\u008428u\u008429u`,'Nav'],
         [`Pg\u008418u`,'Cue'],
     ['sep'],
-        ['WER','ViewMode'],
+        ['WER','View'],
+    ['sep'],
+        ['Sp','Edit'],
     ['sep'],
         ['m','Mute'],
         ['s','Solo'],
     ['sep'],
-        ['Tab','Panel'],
+        ['Tab','Panel']
 //    ['sep'],
 //        ['q','Quit'],
     ]
     let hintElemOrders = [
         [`\u008428u\u008429u`,'Nav'],
         [`Ent`,'Go to cue'],
+    ['sep'],
+        ['Sp','Edit'],
     ['sep'],
         ['Tab','Panel'],
 //    ['sep'],
@@ -871,14 +876,84 @@ function drawControlHint() {
     let hintElemPatterns = [
         [`\u008428u\u008429u`,'Nav'],
         [`Pg\u008418u`,'Ptn'],
+    ['sep'],
+        ['Sp','Edit'],
     ['sep'],
         ['Tab','Panel'],
 //    ['sep'],
 //        ['q','Quit'],
     ]
 
+    let hintElemEditNoteValue = [ // only enabled in viewmode 'E' or in pattern editor
+        [`\u008428u\u008429u`,'Nav'],
+        [`Pg\u008418u`,'Cue'],
+    ['sep'],
+        [`A${sym.doubledot}G`,'Note'],
+        [`0${sym.doubledot}9`,'Oct'],
+        ['[]',`Tone\u008418u`],
+    ['sep'],
+        ['#',sym.sharp],
+        ['@','Acc'],
+    ['sep'],
+        ['=','KOff'],
+        ['^','KCut'],
+//    ['sep'],
+//        ['Sp','ExitEdit'],
+    ]
+    let hintElemEditInstValue = [
+        [`\u008428u\u008429u`,'Nav'],
+        [`Pg\u008418u`,'Cue'],
+    ['sep'],
+        [`0${sym.doubledot}9 A${sym.doubledot}F`,'Instrument'],
+    ['sep'],
+        ['Sp','ExitEdit'],
+    ]
+    let hintElemEditVolEff = [
+        [`\u008428u\u008429u`,'Nav'],
+        [`Pg\u008418u`,'Cue'],
+    ['sep'],
+        ['h','Set'],
+        ['j','SlideDn'],
+        ['k','SlideUp'],
+        ['u','FineDn'],
+        ['i','FineUp'],
+        [`0${sym.doubledot}9 A${sym.doubledot}F`,'Val'],
+//    ['sep'],
+//        ['Sp','ExitEdit'],
+    ]
+    let hintElemEditPanEff = [
+        [`\u008428u\u008429u`,'Nav'],
+        [`Pg\u008418u`,'Cue'],
+    ['sep'],
+        ['h','Set'],
+        ['j','SlideL'],
+        ['k','SlideR'],
+        ['u','FineL'],
+        ['i','FineR'],
+        [`0${sym.doubledot}9 A${sym.doubledot}F`,'Val'],
+//    ['sep'],
+//        ['Sp','ExitEdit'],
+    ]
+    let hintElemEditFxSym = [
+        [`\u008428u\u008429u`,'Nav'],
+        [`Pg\u008418u`,'Cue'],
+    ['sep'],
+        [`0${sym.doubledot}9 A${sym.doubledot}F`,`FxSym`],
+    ['sep'],
+        ['Sp','ExitEdit'],
+    ]
+    let hintElemEditFxVal = [
+        [`\u008428u\u008429u`,'Nav'],
+        [`Pg\u008418u`,'Cue'],
+    ['sep'],
+        [`0${sym.doubledot}9 A${sym.doubledot}F`,`FxVal`],
+    ['sep'],
+        ['Sp','ExitEdit'],
+    ]
+
     const hintElemExternal = [['Tab','Panel']]
     let hintElems = [hintElemTimeline, hintElemOrders, hintElemPatterns, hintElemExternal, hintElemExternal, hintElemExternal, hintElemExternal]
+    let hintElemPat = [hintElemEditNoteValue, hintElemEditInstValue, hintElemEditVolEff, hintElemEditPanEff, hintElemEditFxSym, hintElemEditFxVal]
 
     // erase current line
     con.move(SCRH, 1)
@@ -1837,9 +1912,36 @@ function makeExternalPanelDraw(progName) {
 function drawProjectContents(wo) {
     fillLine(PTNVIEW_OFFSET_Y - 1, colVoiceHdr, 255)
     for (let y = PTNVIEW_OFFSET_Y; y < SCRH; y++) fillLine(y, colBackPtn, 255)
-    con.move(PTNVIEW_OFFSET_Y + 2, 3)
+
-    con.color_pair(colStatus, 255)
+    let mixerflag = initialTrackerMixerflags
-    print('[Project settings — not yet implemented]')
+    let flagstrbuf = ''
+    let flagstr = [
+        ['Linear pan','Equal-energy pan'],
+        ['Linear tone','Amiga tone'],
+    ]
+    for (let i = 0; i < flagstr.length; i++) {
+        let s = flagstr[i][(mixerflag >>> i) & 1 != 0]
+        if (i > 0) flagstrbuf += ', ';
+        flagstrbuf += s
+    }
+
+
+    let projMeta = {
+        Filename: fullPathObj.string.split('\\').last(),
+        Patterns: `${song.numPats}/4095 ($${song.numPats.hex03()})`,
+        Cues: `${song.lastActiveCue}/1024 ($${song.lastActiveCue.hex03()})`,
+        Notation: pitchTablePresets[PITCH_PRESET_IDX].name,
+        Flags: `${flagstrbuf} ($${mixerflag.hex02()})`,
+    }
+
+    Object.entries(projMeta).forEach(([key, value], index) => {
+        con.move(PTNVIEW_OFFSET_Y + index, 2)
+        con.color_pair(colStatus, 255); print(key)
+        con.move(PTNVIEW_OFFSET_Y + index, 12)
+        con.color_pair(colVoiceHdr, colBLACK); print(value)
+    })
+
+    con.color_pair(colStatus, 255) // reset colour
 }
 function externalPanelInput(wo, event) {}
 
@@ -2148,7 +2250,7 @@ function drawGotoPopup(popup, buf) {
     const promptStr = prompts[currentPanel] || 'Number:'
 
     con.move(popup.y + 2, popup.x + 2)
-    con.color_pair(colStatus, colPopupBack)
+    con.color_pair(colWHITE, colPopupBack)
     print(promptStr + ' ')
     con.color_pair(230, 240)
     print('[' + buf.padEnd(3, '_') + ']')
@@ -2171,8 +2273,8 @@ function applyGoto(num) {
 }
 
 function openConfirmQuit() {
-    const pw = 25
+    const pw = 25 + hasUnsavedChanges * 4
-    const ph = 5
+    const ph = 5 + hasUnsavedChanges
     const px = ((SCRW - pw) / 2 | 0) + 1
     const py = ((SCRH - ph) / 2 | 0)
 
@@ -2184,11 +2286,17 @@ function openConfirmQuit() {
     popup.drawFrame()
 
     con.move(py + 2, px + 2)
-    con.color_pair(colStatus, colPopupBack)
+    con.color_pair(colWHITE, colPopupBack)
     print('Exit Microtone? ')
     con.color_pair(230, 240)
     print('[Y/N]')
 
+    if (hasUnsavedChanges) {
+        con.move(py + 3, px + 2)
+        con.color_pair(colWHITE, colPopupBack)
+        print('You have unsaved changes.')
+    }
+
     con.color_pair(colStatus, 255) // reset colour
 
     let result = false
@@ -2261,6 +2369,7 @@ resetAudioDevice()
 taud.uploadTaudFile(fullPathObj.full, 0, PLAYHEAD)
 audio.setMasterVolume(PLAYHEAD, 255)
 audio.setMasterPan(PLAYHEAD, 128)
+const initialTrackerMixerflags = audio.getTrackerMixerFlags(PLAYHEAD)
 
 function isExternalPanel(p) {
     return p === VIEW_SAMPLES || p === VIEW_INSTRMNT || p === VIEW_FILE

assets/disk0/tvdos/include/taud.mjs

@@ -43,9 +43,9 @@ function _pokeU32LE(ptr, off, v) {
  *
  * @param inFile             Full path with drive letter, e.g. "A:/music/song.taud"
  * @param songIndex          0-based index of the song in the SONG TABLE
- * @param targetPlaydataSlot Playhead number (0-3) to configure
+ * @param playhead Playhead number (0-3) to configure
  */
-function uploadTaudFile(inFile, songIndex, targetPlaydataSlot) {
+function uploadTaudFile(inFile, songIndex, playhead) {
     const drive    = inFile[0].toUpperCase()
     const diskPath = inFile.substring(2)
 
@@ -107,6 +107,7 @@ function uploadTaudFile(inFile, songIndex, targetPlaydataSlot) {
     let numPatsHi  = sys.peek(filePtr + entryOff + 6) & 0xFF
     let bpmStored  = sys.peek(filePtr + entryOff + 7) & 0xFF
     let tickRate   = sys.peek(filePtr + entryOff + 8) & 0xFF
+    let mixerflags = sys.peek(filePtr + entryOff + 15) & 0xFF
 
     let bpm        = bpmStored + 24
     let patsToLoad = numPatsLo | (numPatsHi << 8)
@@ -130,9 +131,10 @@ function uploadTaudFile(inFile, songIndex, targetPlaydataSlot) {
     }
 
     // -- 8. Configure playhead ------------------------------------------------
-    audio.setTrackerMode(targetPlaydataSlot)
+    audio.setTrackerMode(playhead)
-    audio.setBPM(targetPlaydataSlot, bpm)
+    audio.setBPM(playhead, bpm)
-    audio.setTickRate(targetPlaydataSlot, tickRate > 0 ? tickRate : 6)
+    audio.setTickRate(playhead, tickRate > 0 ? tickRate : 6)
+    audio.setTrackerMixerFlags(playhead, mixerflags)
 
 
     fileHandle.close()

it2taud.py

@@ -3,6 +3,7 @@
 
 Usage:
     python3 it2taud.py input.it output.taud [-v] [--no-decompress]
+                                            [--no-pf-envelope]
 
 Limits:
     - Up to 20 IT channels (excess dropped; hard error if chunk count
@@ -11,9 +12,17 @@ Limits:
       Taud patterns. Pattern-loop (SBx) crossing a chunk boundary is
       warned; B/C effects are remapped to new cue indices.
     - IT2.14/IT2.15 compressed samples are decoded unless --no-decompress.
-    - IT instrument volume/pan envelopes (up to 8 nodes, sustain loops) are
+    - IT instrument volume/pan envelopes (up to 12 nodes, sustain loops) are
       converted to Taud format. NNA actions are ignored. Each IT instrument
       resolves to its C-5 canonical sample.
+    - IT pitch/filter envelope (no Taud equivalent) is BAKED onto a per-
+      instrument copy of the canonical sample (--no-pf-envelope to disable).
+      Pitch mode uses time-varying linear-interpolated resampling; filter
+      mode uses a 2-pole resonant low-pass biquad (RBJ cookbook),
+      approximate to IT's actual filter. Looped samples are rendered as
+      `entry + N×loop_len` with the loop reapplied to the tail. Caveat:
+      the envelope is locked to the sample's playback rate, so playing
+      the instrument an octave up advances the envelope twice as fast.
     - AdLib / OPL instruments are skipped.
 
 Effect support:
@@ -33,11 +42,22 @@ import math
 import struct
 import sys
 
-VERBOSE = False
+from taud_common import (
-
+    set_verbose, vprint,
-def vprint(*a, **kw):
+    TAUD_MAGIC, TAUD_VERSION, TAUD_HEADER_SIZE, TAUD_SONG_ENTRY,
-    if VERBOSE:
+    SAMPLEBIN_SIZE, INSTBIN_SIZE, SAMPLEINST_SIZE,
-        print(*a, **kw, file=sys.stderr)
+    PATTERN_ROWS, PATTERN_BYTES, NUM_PATTERNS_MAX, NUM_CUES, CUE_SIZE, NUM_VOICES,
+    NOTE_NOP, NOTE_KEYOFF, NOTE_CUT, TAUD_C3,
+    TOP_NONE, TOP_A, TOP_B, TOP_C, TOP_D, TOP_E, TOP_F, TOP_G, TOP_H, TOP_I,
+    TOP_J, TOP_K, TOP_L, TOP_O, TOP_Q, TOP_R, TOP_S, TOP_T, TOP_U, TOP_V, TOP_Y,
+    SEL_SET, SEL_UP, SEL_DOWN, SEL_FINE,
+    EFF_A, EFF_B, EFF_C, EFF_D, EFF_E, EFF_F, EFF_G, EFF_H, EFF_I, EFF_J,
+    EFF_K, EFF_L, EFF_M, EFF_N, EFF_O, EFF_P, EFF_Q, EFF_R, EFF_S, EFF_T,
+    EFF_U, EFF_V, EFF_W, EFF_X, EFF_Y, EFF_Z,
+    J_SEMI_TABLE,
+    d_arg_to_col, resample_linear, encode_cue, deduplicate_patterns,
+    normalise_sample,
+)
 
 
 # ── IT constants ─────────────────────────────────────────────────────────────
@@ -82,14 +102,6 @@ VC_VIB_LO,    VC_VIB_HI    = 203, 212   # vibrato       H (depth 1..10)
 
 VC_TPORTA_TABLE = (0, 1, 4, 8, 16, 32, 64, 96, 128, 255)
 
-# IT effect letters (1-based, same numbering as S3M so we can reuse encode_effect)
-EFF_A = 1;  EFF_B = 2;  EFF_C = 3;  EFF_D = 4;  EFF_E = 5
-EFF_F = 6;  EFF_G = 7;  EFF_H = 8;  EFF_I = 9;  EFF_J = 10
-EFF_K = 11; EFF_L = 12; EFF_M = 13; EFF_N = 14; EFF_O = 15
-EFF_P = 16; EFF_Q = 17; EFF_R = 18; EFF_S = 19; EFF_T = 20
-EFF_U = 21; EFF_V = 22; EFF_W = 23; EFF_X = 24; EFF_Y = 25
-EFF_Z = 26
-
 # IT effects that recall last non-zero arg (per-effect-private, with cohort exceptions).
 # V (Set Global Volume) recalls in IT compat mode — the first V $00 resolves to the
 # header's global_vol, not literal 0. Without this, songs starting with V $00 silence.
@@ -100,40 +112,10 @@ IT_MEM_EFFECTS = frozenset({
 })
 
 
-# ── Taud constants ────────────────────────────────────────────────────────────
+# ── Taud constants (it-specific) ──────────────────────────────────────────────
 
-TAUD_MAGIC       = bytes([0x1F,0x54,0x53,0x56,0x4D,0x61,0x75,0x64])
-TAUD_VERSION     = 1
-TAUD_HEADER_SIZE = 32
-TAUD_SONG_ENTRY  = 16
-SAMPLEBIN_SIZE   = 770048
-INSTBIN_SIZE     = 16384
-SAMPLEINST_SIZE  = SAMPLEBIN_SIZE + INSTBIN_SIZE
-PATTERN_ROWS     = 64
-PATTERN_BYTES    = PATTERN_ROWS * 8
-NUM_PATTERNS_MAX = 4095
-NUM_CUES         = 1024
-CUE_SIZE         = 32
-NUM_VOICES       = 20
 SIGNATURE        = b'it2taud/TSVM  '   # 14 bytes
 
-NOTE_NOP    = 0xFFFF
-NOTE_KEYOFF = 0x0000
-NOTE_CUT    = 0xFFFE
-TAUD_C3     = 0x4000
-
-TOP_NONE = 0x00
-TOP_A    = 0x0A; TOP_B = 0x0B; TOP_C = 0x0C; TOP_D = 0x0D
-TOP_E    = 0x0E; TOP_F = 0x0F; TOP_G = 0x10; TOP_H = 0x11
-TOP_I    = 0x12; TOP_J = 0x13; TOP_K = 0x14; TOP_L = 0x15
-TOP_O    = 0x18; TOP_Q = 0x1A; TOP_R = 0x1B; TOP_S = 0x1C
-TOP_T    = 0x1D; TOP_U = 0x1E; TOP_V = 0x1F; TOP_Y = 0x22
-
-SEL_SET  = 0; SEL_UP = 1; SEL_DOWN = 2; SEL_FINE = 3
-
-J_SEMI_TABLE = [0x00, 0x01, 0x03, 0x04, 0x05, 0x07, 0x08, 0x09,
-                0x0B, 0x0C, 0x0D, 0x0F, 0x10, 0x11, 0x13, 0x14]
-
 # ThreeFiveMiniUfloat LUT — 256 entries, seconds 0.0..126.0 (must match Kotlin)
 _MINUFLOAT_LUT = [
     0.0, 0.03125, 0.0625, 0.09375, 0.125, 0.15625, 0.1875, 0.21875,
@@ -390,40 +372,6 @@ def it214_decompress(blob: bytes, smp_offset: int, num_samples: int,
         return bytes(s & 0xFF for s in out_samples)
 
 
-# ── Sample normaliser (same as s3m2taud but signed derived from cvt byte) ────
-
-def _normalise_sample(raw: bytes, signed: bool, is_16bit: bool,
-                      is_stereo: bool, name: str) -> bytes:
-    """Return unsigned 8-bit mono sample bytes."""
-    out = []
-    stride = (2 if is_16bit else 1) * (2 if is_stereo else 1)
-    i = 0
-    while i + stride <= len(raw):
-        if is_16bit:
-            if is_stereo:
-                l16 = struct.unpack_from('<h', raw, i)[0]
-                r16 = struct.unpack_from('<h', raw, i+2)[0]
-                s = (l16 + r16) >> 1
-            else:
-                s = struct.unpack_from('<h', raw, i)[0]
-            v = (s >> 8) + 128
-        else:
-            if is_stereo:
-                l8 = raw[i]; r8 = raw[i+1]
-                raw_s = (l8 + r8) // 2
-            else:
-                raw_s = raw[i]
-            if signed:
-                v = (raw_s ^ 0x80) & 0xFF
-            else:
-                v = raw_s
-        out.append(v & 0xFF)
-        i += stride
-    if is_16bit or is_stereo:
-        vprint(f"  info: '{name}' converted to unsigned 8-bit mono ({len(out)} samples)")
-    return bytes(out)
-
-
 # ── IT sample parser ──────────────────────────────────────────────────────────
 
 class ITSample:
@@ -480,7 +428,7 @@ def parse_samples(data: bytes, h: ITHeader, decompress: bool) -> list:
                         is_it215 = bool(s.cvt & 0x04)
                         raw = it214_decompress(data, s.smp_point, s.length,
                                                s.is_16bit, is_it215)
-                        s.sample_data = _normalise_sample(raw, True,
+                        s.sample_data = normalise_sample(raw, True,
                                                           s.is_16bit, s.is_stereo, s.name)
                         s.length = len(s.sample_data)
                         s.loop_beg = min(s.loop_beg, s.length)
@@ -494,7 +442,7 @@ def parse_samples(data: bytes, h: ITHeader, decompress: bool) -> list:
                     s.sample_data = bytes(min(s.length, 256))
                 else:
                     raw = data[s.smp_point : s.smp_point + byte_len]
-                    s.sample_data = _normalise_sample(raw, s.is_signed,
+                    s.sample_data = normalise_sample(raw, s.is_signed,
                                                       s.is_16bit, s.is_stereo, s.name)
                     s.length    = len(s.sample_data)
                     s.loop_beg  = min(s.loop_beg, s.length)
@@ -503,13 +451,206 @@ def parse_samples(data: bytes, h: ITHeader, decompress: bool) -> list:
     return samples
 
 
+# ── Pitch / filter envelope baker ─────────────────────────────────────────────
+# IT instruments carry a third envelope (pitch or filter, distinguished by
+# flag bit 7) that has no Taud equivalent. We render its effect onto a
+# per-instrument copy of the canonical sample so the instrument can play
+# through Taud's normal sample path with the modulation already baked in.
+#
+# Caveat: the baked envelope's time axis is locked to the sample's playback
+# rate, so playing the instrument an octave up advances the envelope twice
+# as fast (IT's envelope is wall-clock-time, regardless of note pitch). For
+# typical drum/lead use cases this is musically acceptable.
+
+def _clone_sample(src: 'ITSample') -> 'ITSample':
+    dst = ITSample()
+    for slot in ITSample.__slots__:
+        setattr(dst, slot, getattr(src, slot))
+    dst.sample_data = bytes(src.sample_data)
+    return dst
+
+
+def _plan_baked_length(src: 'ITSample', nodes: list, flags: dict,
+                       ticks_per_sec: float) -> tuple:
+    """Determine baked output length and loop boundaries.
+
+    Non-looped src → (src.length, 0, 0).
+    Looped src    → (entry + N×loop_len, entry, entry + N×loop_len),
+                    with N the smallest integer that covers the envelope's
+                    active duration in seconds. N is clamped to [1, 16].
+    """
+    if not nodes:
+        return src.length, 0, 0
+    last_tick = nodes[-1][1]
+    env_dur_sec = last_tick / ticks_per_sec if ticks_per_sec > 0 else 0.0
+    env_dur_samples = int(env_dur_sec * src.c5_speed)
+
+    if not src.has_loop or src.loop_end <= src.loop_beg:
+        return src.length, 0, 0
+
+    entry = max(0, src.loop_beg)
+    loop_len = src.loop_end - src.loop_beg
+    if loop_len <= 0:
+        return src.length, 0, 0
+
+    samples_needed = max(0, env_dur_samples - entry)
+    n = max(1, (samples_needed + loop_len - 1) // loop_len)
+    n = min(n, 16)
+    out_len = entry + n * loop_len
+    return out_len, entry, out_len
+
+
+def _read_src_sample(sd: bytes, pos: float, src: 'ITSample') -> float:
+    """Linear-interpolated read of `sd` (unsigned u8 PCM) at fractional `pos`,
+    returning a signed float in roughly [-128, +127]. Honours `src.has_loop`
+    by wrapping into [loop_beg, loop_end) once `pos >= loop_end`. Past the
+    end of a non-looped sample, returns 0.0 (silence)."""
+    if not sd:
+        return 0.0
+    if src.has_loop and src.loop_end > src.loop_beg:
+        if pos >= src.loop_end:
+            span = src.loop_end - src.loop_beg
+            pos = src.loop_beg + ((pos - src.loop_beg) % span)
+    if pos < 0.0:
+        return 0.0
+    if pos >= len(sd) - 1:
+        if pos >= len(sd):
+            return 0.0
+        return float(sd[len(sd) - 1] - 128)
+    i0 = int(pos)
+    frac = pos - i0
+    a = sd[i0] - 128
+    b = sd[i0 + 1] - 128
+    return a + (b - a) * frac
+
+
+def _bake_pitch_envelope(src: 'ITSample', nodes: list, flags: dict,
+                         ticks_per_sec: float) -> 'ITSample':
+    if not src.sample_data:
+        return src
+    out_len, lb, le = _plan_baked_length(src, nodes, flags, ticks_per_sec)
+    tick_per_sample = ticks_per_sec / max(1, src.c5_speed)
+    sd = src.sample_data
+    out = bytearray(out_len)
+    read_pos = 0.0
+    t_tick = 0.0
+    for i in range(out_len):
+        env_v = _env_value_at(t_tick, nodes, flags)
+        rate = 2.0 ** (env_v / 12.0)
+        v = _read_src_sample(sd, read_pos, src)
+        b = int(round(v)) + 128
+        if b < 0: b = 0
+        elif b > 255: b = 255
+        out[i] = b
+        read_pos += rate
+        t_tick += tick_per_sample
+
+    dst = _clone_sample(src)
+    dst.sample_data = bytes(out)
+    dst.length = out_len
+    dst.loop_beg = lb
+    dst.loop_end = le
+    if lb < le:
+        dst.has_loop = True
+        dst.flags = (dst.flags | IT_SMP_LOOP) & ~IT_SMP_PINGPONG
+    else:
+        dst.has_loop = False
+        dst.flags = dst.flags & ~(IT_SMP_LOOP | IT_SMP_PINGPONG)
+    return dst
+
+
+def _bake_filter_envelope(src: 'ITSample', nodes: list, flags: dict,
+                          ticks_per_sec: float, ifr: int) -> 'ITSample':
+    """Time-varying 2-pole resonant low-pass biquad (RBJ cookbook).
+    Approximates IT's filter; not bit-exact to IT's Pentium routine."""
+    if not src.sample_data:
+        return src
+    out_len, lb, le = _plan_baked_length(src, nodes, flags, ticks_per_sec)
+    tick_per_sample = ticks_per_sec / max(1, src.c5_speed)
+    sr = max(1, src.c5_speed)
+    nyq = sr * 0.5 - 1.0
+
+    # Resonance (0..127) → Q ∈ [0.5, 6.0]
+    Q = 0.5 + (max(0, min(ifr, 127)) / 127.0) * 5.5
+
+    sd = src.sample_data
+    out = bytearray(out_len)
+    x1 = x2 = 0.0
+    y1 = y2 = 0.0
+    t_tick = 0.0
+    src_len = len(sd)
+    for i in range(out_len):
+        env_v = _env_value_at(t_tick, nodes, flags)
+        # Map env value -32..+32 to cutoff: 110 Hz (closed) to ~28 kHz (open)
+        cutoff = 110.0 * (2.0 ** ((env_v + 32.0) * 0.125))
+        if cutoff > nyq: cutoff = nyq
+        if cutoff < 1.0: cutoff = 1.0
+        w0 = 2.0 * math.pi * cutoff / sr
+        cosw = math.cos(w0)
+        sinw = math.sin(w0)
+        alpha = sinw / (2.0 * Q)
+        b0 = (1.0 - cosw) * 0.5
+        b1 = 1.0 - cosw
+        b2 = b0
+        a0 = 1.0 + alpha
+        a1 = -2.0 * cosw
+        a2 = 1.0 - alpha
+
+        # Read source with looping
+        in_pos = i
+        if src.has_loop and src.loop_end > src.loop_beg and in_pos >= src.loop_end:
+            span = src.loop_end - src.loop_beg
+            in_pos = src.loop_beg + ((in_pos - src.loop_beg) % span)
+        if 0 <= in_pos < src_len:
+            x0 = float(sd[in_pos] - 128)
+        else:
+            x0 = 0.0
+
+        y0 = (b0 * x0 + b1 * x1 + b2 * x2 - a1 * y1 - a2 * y2) / a0
+        x2 = x1; x1 = x0
+        y2 = y1; y1 = y0
+
+        b = int(round(y0)) + 128
+        if b < 0: b = 0
+        elif b > 255: b = 255
+        out[i] = b
+        t_tick += tick_per_sample
+
+    dst = _clone_sample(src)
+    dst.sample_data = bytes(out)
+    dst.length = out_len
+    dst.loop_beg = lb
+    dst.loop_end = le
+    if lb < le:
+        dst.has_loop = True
+        dst.flags = (dst.flags | IT_SMP_LOOP) & ~IT_SMP_PINGPONG
+    else:
+        dst.has_loop = False
+        dst.flags = dst.flags & ~(IT_SMP_LOOP | IT_SMP_PINGPONG)
+    return dst
+
+
+def _bake_pf_envelope(src: 'ITSample', inst, ticks_per_sec: float) -> 'ITSample':
+    if not inst.pf_nodes or not inst.pf_flags or not inst.pf_flags.get('enabled'):
+        return src
+    if inst.pf_flags['is_filter']:
+        return _bake_filter_envelope(src, inst.pf_nodes, inst.pf_flags,
+                                     ticks_per_sec, inst.ifr)
+    return _bake_pitch_envelope(src, inst.pf_nodes, inst.pf_flags, ticks_per_sec)
+
+
 # ── IT instrument parser ──────────────────────────────────────────────────────
 
 class ITInstrument:
     __slots__ = ('name', 'dfp', 'gv', 'canonical_sample', 'canonical_volume',
-                 'vol_envelope', 'pan_envelope', 'vol_env_sustain', 'pan_env_sustain')
+                 'vol_envelope', 'pan_envelope', 'vol_env_sustain', 'pan_env_sustain',
-    # vol_envelope / pan_envelope: list of 8 (value, minifloat_idx) tuples, or None
+                 'pf_nodes', 'pf_flags', 'ifc', 'ifr')
+    # vol_envelope / pan_envelope: list of 12 (value, minifloat_idx) tuples, or None
     # vol_env_sustain / pan_env_sustain: int (0 = disabled, else (end<<3)|start)
+    # pf_nodes: raw list of (int8 value, uint16 tick) tuples (up to 25), or None
+    # pf_flags: dict {enabled, has_env_loop, has_sus_loop, lpb, lpe, slb, sle,
+    #                 is_filter, carry}, or None
+    # ifc / ifr: initial filter cutoff / resonance (0..127, or 0 if not set)
 
 def parse_instruments(data: bytes, h: ITHeader) -> list:
     insts = []
@@ -542,29 +683,40 @@ def parse_instruments(data: bytes, h: ITHeader) -> list:
         inst.canonical_sample = c5_smp   # 1-based sample index, 0 = none
         inst.canonical_volume  = min(inst.gv, 64)
 
+        # Initial filter cutoff/resonance (high bit = enabled, low 7 bits = value)
+        ifc_raw = data[ptr + 0x39]
+        ifr_raw = data[ptr + 0x3A]
+        inst.ifc = ifc_raw & 0x7F if (ifc_raw & 0x80) else 0
+        inst.ifr = ifr_raw & 0x7F if (ifr_raw & 0x80) else 0
+
         # Parse IT envelopes (new-format only, ≥cmwt 0x200)
-        # Vol envelope at ptr+0x130; pan envelope at ptr+0x182
+        # Vol envelope at ptr+0x130; pan envelope at ptr+0x182; pf envelope at ptr+0x1D4
         ticks_per_sec = max(h.initial_tempo * 2.0 / 5.0, 1.0)   # tick rate = bpm×2/5 (50 Hz at 125 BPM); speed is ticks-per-row, irrelevant here
         inst.vol_envelope, inst.vol_env_sustain = _parse_it_envelope(
             data, ptr + 0x130, False, ticks_per_sec)
         inst.pan_envelope, inst.pan_env_sustain = _parse_it_envelope(
             data, ptr + 0x182, True,  ticks_per_sec)
+        inst.pf_nodes, inst.pf_flags = _parse_it_pf_envelope_raw(
+            data, ptr + 0x1D4)
         insts.append(inst)
     return insts
 
 
 def _parse_it_envelope(data: bytes, env_ptr: int, is_pan: bool,
                        ticks_per_sec: float) -> tuple:
-    """Parse one IT envelope block into 8 Taud (value, minifloat_idx) points.
+    """Parse one IT envelope block into 12 Taud (value, minifloat_idx) points.
 
     Returns (points_list, sustain_byte) where points_list is a list of
-    8 (value, minifloat_idx) tuples, or None if envelope not enabled.
+    12 (value, minifloat_idx) tuples, or None if envelope not enabled.
-    sustain_byte: bit7=enabled, bits[5:3]=end_idx, bits[2:0]=start_idx; 0=disabled.
+    sustain_byte: bit7=enabled (u), bit6=sustain (t: 1=breaks on key-off,
+    0=loops forever), bits[5:3]=end_idx, bits[2:0]=start_idx; 0=disabled.
+    Note: sustain byte still uses 3-bit indices (0..7), so loop nodes
+    referencing indices 8..11 cannot be encoded and fall back to no loop.
 
-    IT has two loop types: envelope loop (always on) and sustain loop (breaks on
+    IT has two loop types: envelope loop (continues forever) and sustain loop
-    key-off). Taud only has sustain loop semantics. Priority: sustain > env loop.
+    (breaks on key-off). Taud distinguishes them via the 't' flag. Priority
-    When slb==sle the AudioAdapter holds at that node (no cycling); for slb!=sle
+    when both exist: sustain (because IT plays sustain while held, then env
-    it cycles between them.
+    loop after release; Taud can only express one).
     """
     if env_ptr + 82 > len(data):
         return None, 0
@@ -580,17 +732,20 @@ def _parse_it_envelope(data: bytes, env_ptr: int, is_pan: bool,
     has_env_loop = bool(flags & 0x02)
     has_sus_loop = bool(flags & 0x04)
 
-    # Choose which IT loop to map to Taud sustain (priority: sus > env)
+    # Choose which IT loop to map to Taud (priority: sus > env). The 't' flag
+    # distinguishes them: t=1 for sustain (breaks on key-off), t=0 for env loop.
     if has_sus_loop:
         use_lb, use_le = it_slb, it_sle
         has_loop = True
+        is_sustain = True
     elif has_env_loop:
         use_lb, use_le = it_lpb, it_lpe
         has_loop = True
-        vprint(f"    envelope loop mapped as sustain loop (approximation: breaks on key-off)")
+        is_sustain = False
     else:
         use_lb = use_le = -1
         has_loop = False
+        is_sustain = False
 
     # Read IT nodes: (int8 value, uint16 tick_pos LE)
     nodes = []
@@ -605,23 +760,26 @@ def _parse_it_envelope(data: bytes, env_ptr: int, is_pan: bool,
     if not nodes:
         return None, 0
 
-    # Decimate to 8 nodes if needed, preserving first/last
+    # Decimate to 12 nodes if needed, preserving first/last
-    decimated = len(nodes) > 8
+    decimated = len(nodes) > 12
     if not decimated:
         selected = nodes[:]
-        if has_loop and use_lb < len(selected) and use_le < len(selected):
+        # Sustain byte encodes 3-bit indices; loop nodes ≥8 cannot be referenced.
+        if has_loop and use_lb < min(len(selected), 8) and use_le < min(len(selected), 8):
             taud_slb, taud_sle = use_lb, use_le
         else:
             taud_slb = taud_sle = -1
+            if has_loop and (use_lb >= 8 or use_le >= 8):
+                vprint(f"    loop indices ≥8 cannot be encoded in 3-bit sustain field")
     else:
-        selected = [nodes[round(k * (len(nodes) - 1) / 7)] for k in range(8)]
+        selected = [nodes[round(k * (len(nodes) - 1) / 11)] for k in range(12)]
         taud_slb = taud_sle = -1   # loop indices lost in decimation
         if has_loop:
-            vprint(f"    loop indices lost due to decimation ({len(nodes)} nodes → 8)")
+            vprint(f"    loop indices lost due to decimation ({len(nodes)} nodes → 12)")
 
-    # Build 8 Taud envelope points with delta-time minifloats
+    # Build 12 Taud envelope points with delta-time minifloats
     points = []
-    for k in range(8):
+    for k in range(12):
         if k < len(selected):
             val, tick = selected[k]
             if is_pan:
@@ -640,16 +798,99 @@ def _parse_it_envelope(data: bytes, env_ptr: int, is_pan: bool,
             mf_idx   = 0
         points.append((taud_val, mf_idx))
 
-    # Build sustain byte: bit7=enabled, bits[5:3]=end_idx, bits[2:0]=start_idx.
+    # Build sustain byte: bit7=enable (u), bit6=sustain (t), bits[5:3]=end,
-    # 0 = disabled (no bit7). All (slb, sle) pairs including (0,0) are valid when bit7=1.
+    # bits[2:0]=start. 0=disabled. t=1 → breaks on key-off (IT sustain loop);
+    # t=0 → loops forever (IT envelope loop).
     if taud_slb >= 0 and taud_sle >= 0:
-        sus_byte = 0x80 | ((taud_sle & 7) << 3) | (taud_slb & 7)
+        t_bit = 0x40 if is_sustain else 0x00
+        sus_byte = 0x80 | t_bit | ((taud_sle & 7) << 3) | (taud_slb & 7)
     else:
         sus_byte = 0
 
     return points, sus_byte
 
 
+def _parse_it_pf_envelope_raw(data: bytes, env_ptr: int) -> tuple:
+    """Parse the IT pitch/filter envelope keeping all nodes (no decimation).
+
+    Returns (nodes, flags) where:
+      nodes: list of (int8 value, uint16 tick) tuples (up to 25 entries),
+             values in -32..+32 semitones (pitch) or filter modulation units.
+      flags: dict {enabled, has_env_loop, has_sus_loop, lpb, lpe, slb, sle,
+                   is_filter, carry}
+    Returns (None, None) if the envelope is not enabled or out of range.
+    """
+    if env_ptr + 82 > len(data):
+        return None, None
+    flags_byte = data[env_ptr]
+    if not (flags_byte & 0x01):
+        return None, None
+    num_nodes = max(1, min(data[env_ptr + 1], 25))
+    flags = {
+        'enabled':      True,
+        'has_env_loop': bool(flags_byte & 0x02),
+        'has_sus_loop': bool(flags_byte & 0x04),
+        'carry':        bool(flags_byte & 0x08),
+        'is_filter':    bool(flags_byte & 0x80),
+        'lpb': data[env_ptr + 2],
+        'lpe': data[env_ptr + 3],
+        'slb': data[env_ptr + 4],
+        'sle': data[env_ptr + 5],
+    }
+    nodes = []
+    for n in range(num_nodes):
+        nptr = env_ptr + 6 + n * 3
+        if nptr + 2 >= len(data):
+            break
+        val  = struct.unpack_from('b', data, nptr)[0]
+        tick = struct.unpack_from('<H', data, nptr + 1)[0]
+        nodes.append((val, tick))
+    if not nodes:
+        return None, None
+    # Clamp loop indices into valid range
+    for k in ('lpb', 'lpe', 'slb', 'sle'):
+        flags[k] = min(flags[k], len(nodes) - 1)
+    return nodes, flags
+
+
+def _env_value_at(t_tick: float, nodes: list, flags: dict) -> float:
+    """Return interpolated envelope value at time `t_tick` (IT envelope ticks).
+
+    Honours env-loop wrap. Sus-loop is treated as 'play through once then hold
+    last value' (no key-off model). Returns nodes[0].value if t_tick is before
+    the first node, nodes[-1].value if past the last node and no env loop.
+    """
+    if not nodes:
+        return 0.0
+    first_tick = nodes[0][1]
+    if t_tick <= first_tick:
+        return float(nodes[0][0])
+
+    if flags['has_env_loop']:
+        lpb_tick = nodes[flags['lpb']][1]
+        lpe_tick = nodes[flags['lpe']][1]
+        loop_span = lpe_tick - lpb_tick
+        if loop_span > 0 and t_tick >= lpe_tick:
+            t_tick = lpb_tick + ((t_tick - lpb_tick) % loop_span)
+        elif loop_span <= 0 and t_tick >= lpe_tick:
+            return float(nodes[flags['lpe']][0])
+
+    last_tick = nodes[-1][1]
+    if t_tick >= last_tick:
+        return float(nodes[-1][0])
+
+    # Linear interpolate between bracketing nodes
+    for k in range(len(nodes) - 1):
+        a_val, a_t = nodes[k]
+        b_val, b_t = nodes[k + 1]
+        if a_t <= t_tick <= b_t:
+            if b_t == a_t:
+                return float(a_val)
+            frac = (t_tick - a_t) / (b_t - a_t)
+            return a_val + (b_val - a_val) * frac
+    return float(nodes[-1][0])
+
+
 # ── IT pattern parser ─────────────────────────────────────────────────────────
 
 class ITRow:
@@ -783,21 +1024,6 @@ def decode_volcol(vc: int):
 
 # ── Effect translator ─────────────────────────────────────────────────────────
 
-def _d_arg_to_col(arg: int):
-    if arg == 0:
-        return None
-    hi = (arg >> 4) & 0xF
-    lo = arg & 0xF
-    if hi == 0xF and lo > 0:
-        return (SEL_FINE, lo & 0x1F)
-    if lo == 0xF and hi > 0:
-        return (SEL_FINE, (hi & 0x1F) | 0x20)
-    if hi > 0 and lo == 0:
-        return (SEL_UP, hi)
-    if lo > 0 and hi == 0:
-        return (SEL_DOWN, lo)
-    return (SEL_UP, hi)
-
 def encode_effect_it(cmd: int, arg: int, ch: int = 0, row: int = 0) -> tuple:
     """Return (taud_op, taud_arg16, vol_override, pan_override).
 
@@ -851,22 +1077,22 @@ def encode_effect_it(cmd: int, arg: int, ch: int = 0, row: int = 0) -> tuple:
         return (TOP_J, (J_SEMI_TABLE[hi_semi] << 8) | J_SEMI_TABLE[lo_semi], None, None)
 
     if cmd == EFF_K:
-        return (TOP_H, 0x0000, _d_arg_to_col(arg), None)
+        return (TOP_H, 0x0000, d_arg_to_col(arg), None)
 
     if cmd == EFF_L:
-        return (TOP_G, 0x0000, _d_arg_to_col(arg), None)
+        return (TOP_G, 0x0000, d_arg_to_col(arg), None)
 
     if cmd == EFF_M:
         return (TOP_NONE, 0, (SEL_SET, min(arg, 0x3F)), None)
 
     if cmd == EFF_N:
-        return (TOP_NONE, 0, _d_arg_to_col(arg), None)
+        return (TOP_NONE, 0, d_arg_to_col(arg), None)
 
     if cmd == EFF_O:
         return (TOP_O, (arg & 0xFF) << 8, None, None)
 
     if cmd == EFF_P:
-        return (TOP_NONE, 0, None, _d_arg_to_col(arg))
+        return (TOP_NONE, 0, None, d_arg_to_col(arg))
 
     if cmd == EFF_Q:
         return (TOP_Q, (arg & 0xFF) << 8, None, None)
@@ -1108,26 +1334,13 @@ def _find_post_pat_cue(pi: int, order_list: list, chunk_map: list,
 
 # ── Sample / instrument bin (same as s3m2taud) ────────────────────────────────
 
-def _resample_linear(data: bytes, ratio: float) -> bytes:
-    if not data:
-        return data
-    n_out = max(1, int(len(data) * ratio))
-    out   = bytearray(n_out)
-    for i in range(n_out):
-        src  = i / ratio
-        i0   = int(src)
-        frac = src - i0
-        i1   = min(i0 + 1, len(data) - 1)
-        v    = data[i0] * (1.0 - frac) + data[i1] * frac
-        out[i] = int(v + 0.5) & 0xFF
-    return bytes(out)
-
 def build_sample_inst_bin_it(samples_or_proxy: list,
                               envelopes_by_slot: dict = None) -> tuple:
     """samples_or_proxy: list of ITSample | None, indexed 1-based (index 0 unused).
 
-    envelopes_by_slot: optional dict mapping taud_slot → (vol_env, vol_sus, pan_env, pan_sus)
+    envelopes_by_slot: optional dict mapping taud_slot → (vol_env, vol_sus, pan_env, pan_sus, inst_gv)
-    where vol_env/pan_env are lists of 8 (value, minifloat_idx) tuples (or None).
+    where vol_env/pan_env are lists of 12 (value, minifloat_idx) tuples (or None),
+    and inst_gv is instrument global volume (0..255, byte 15).
 
     Returns (bin_bytes[SAMPLEINST_SIZE], offsets_dict).
     """
@@ -1140,7 +1353,7 @@ def build_sample_inst_bin_it(samples_or_proxy: list,
         ratio = SAMPLEBIN_SIZE / total
         vprint(f"  info: sample bin overflow ({total} bytes); resampling all by {ratio:.4f}")
         for _, s in pcm_list:
-            new_data    = _resample_linear(s.sample_data, ratio)
+            new_data    = resample_linear(s.sample_data, ratio)
             s.sample_data  = new_data
             s.length       = len(new_data)
             s.loop_beg     = max(0, int(s.loop_beg * ratio))
@@ -1191,30 +1404,33 @@ def build_sample_inst_bin_it(samples_or_proxy: list,
         struct.pack_into('<H', inst_bin, base + 10, le)
         inst_bin[base + 12] = flags_byte
 
-        # Write envelope data (new 8-point format)
+        # Write envelope data (12-point format: vol at +16..+39, pan at +40..+63)
         env_data = envelopes_by_slot.get(taud_idx) if envelopes_by_slot else None
         if env_data and env_data[0]:
-            vol_env, vol_sus, pan_env, pan_sus = env_data
+            vol_env, vol_sus, pan_env, pan_sus, inst_gv = env_data
             inst_bin[base + 13] = vol_sus & 0xFF
             inst_bin[base + 14] = pan_sus & 0xFF
-            for k, (val, mf) in enumerate(vol_env[:8]):
+            inst_bin[base + 15] = inst_gv & 0xFF
+            for k, (val, mf) in enumerate(vol_env[:12]):
                 inst_bin[base + 16 + k*2]     = val & 0xFF
                 inst_bin[base + 16 + k*2 + 1] = mf  & 0xFF
             if pan_env:
-                for k, (val, mf) in enumerate(pan_env[:8]):
+                for k, (val, mf) in enumerate(pan_env[:12]):
-                    inst_bin[base + 32 + k*2]     = val & 0xFF
+                    inst_bin[base + 40 + k*2]     = val & 0xFF
-                    inst_bin[base + 32 + k*2 + 1] = mf  & 0xFF
+                    inst_bin[base + 40 + k*2 + 1] = mf  & 0xFF
             else:
-                for k in range(8):
+                for k in range(12):
-                    inst_bin[base + 32 + k*2]     = 0x80  # pan centre
+                    inst_bin[base + 40 + k*2]     = 0x80  # pan centre
-                    inst_bin[base + 32 + k*2 + 1] = 0x00  # hold
+                    inst_bin[base + 40 + k*2 + 1] = 0x00  # hold
         else:
-            # No instrument envelope: single-point vol, neutral pan
+            # No instrument envelope: single-point vol, neutral pan, full gv
+            inst_gv = env_data[4] if env_data else 255
+            inst_bin[base + 15] = inst_gv & 0xFF
             inst_bin[base + 16] = min(s.vol, 63)   # value 0-63
             inst_bin[base + 17] = 0                 # offset 0 = hold
-            for k in range(8):
+            for k in range(12):
-                inst_bin[base + 32 + k*2]     = 0x80  # pan centre
+                inst_bin[base + 40 + k*2]     = 0x80  # pan centre
-                inst_bin[base + 32 + k*2 + 1] = 0x00  # hold
+                inst_bin[base + 40 + k*2 + 1] = 0x00  # hold
         vprint(f"  instrument[{taud_idx}] '{s.name}' ptr:{ptr} c5spd:{s.c5_speed}")
 
     return bytes(sample_bin) + bytes(inst_bin), offsets
@@ -1335,32 +1551,6 @@ def build_pattern_it(chunk_grid: list, ch_idx: int, default_pan: int,
     return bytes(out)
 
 
-# ── Cue sheet helpers (adapted from s3m2taud) ─────────────────────────────────
-
-def _encode_cue(patterns12: list, instruction: int) -> bytearray:
-    pats = list(patterns12) + [0xFFF] * NUM_VOICES
-    pats = pats[:NUM_VOICES]
-    entry = bytearray(CUE_SIZE)
-    for i in range(10):
-        v0, v1 = pats[i*2], pats[i*2+1]
-        entry[i]      = ((v0 & 0xF) << 4) | (v1 & 0xF)
-        entry[10 + i] = (((v0 >> 4) & 0xF) << 4) | ((v1 >> 4) & 0xF)
-        entry[20 + i] = (((v0 >> 8) & 0xF) << 4) | ((v1 >> 8) & 0xF)
-    entry[30] = instruction & 0xFF
-    return entry
-
-def deduplicate_patterns(pat_bin: bytes, num_pats: int) -> tuple:
-    seen = {}; remap = {}; canonical = []
-    for i in range(num_pats):
-        pat = pat_bin[i*PATTERN_BYTES:(i+1)*PATTERN_BYTES]
-        if pat in seen:
-            remap[i] = seen[pat]
-        else:
-            ci = len(canonical)
-            seen[pat] = ci; remap[i] = ci; canonical.append(pat)
-    return b''.join(canonical), remap, len(canonical)
-
-
 # ── Main assembly ─────────────────────────────────────────────────────────────
 
 def find_initial_bpm_speed(patterns_rows: list, order_list: list,
@@ -1406,7 +1596,8 @@ def _active_channels(h: ITHeader, patterns_rows: list) -> list:
     return active
 
 def assemble_taud(h: ITHeader, samples: list, instruments: list,
-                  patterns_rows: list, decompress: bool) -> bytes:
+                  patterns_rows: list, decompress: bool,
+                  no_pf_envelope: bool = False) -> bytes:
     # ── Resolve IT recalls ───────────────────────────────────────────────────
     vprint("  resolving IT recalls…")
     resolve_it_recalls(patterns_rows, h.order_list, 64, h.link_gef,
@@ -1471,9 +1662,11 @@ def assemble_taud(h: ITHeader, samples: list, instruments: list,
     if h.use_instruments:
         # Build a proxy sample list where Taud inst slot = IT inst index,
         # resolved to the canonical sample. Slot 0 unused.
+        ticks_per_sec = max(h.initial_tempo * 2.0 / 5.0, 1.0)
         proxy = [None] * (max(len(instruments), 256) + 1)
         inst_vols = {}
         envelopes_by_slot = {}
+        bake_count = 0
         for ii, inst in enumerate(instruments):
             taud_slot = ii + 1
             if taud_slot >= 256: break
@@ -1481,13 +1674,29 @@ def assemble_taud(h: ITHeader, samples: list, instruments: list,
             si = inst.canonical_sample - 1   # 0-based sample index
             if si < 0 or si >= len(samples) or samples[si] is None:
                 continue
-            proxy[taud_slot] = samples[si]
+            src_smp = samples[si]
+            if (not no_pf_envelope and inst.pf_nodes
+                    and inst.pf_flags and inst.pf_flags.get('enabled')):
+                baked = _bake_pf_envelope(src_smp, inst, ticks_per_sec)
+                if baked is not src_smp:
+                    bake_count += 1
+                    mode = 'filter' if inst.pf_flags['is_filter'] else 'pitch'
+                    vprint(f"  baked pf envelope on inst[{ii+1}] '{inst.name}' "
+                           f"(mode={mode}, src_len={src_smp.length}, "
+                           f"out_len={baked.length})")
+                src_smp = baked
+            proxy[taud_slot] = src_smp
             vol64 = min(inst.canonical_volume, 64)
             inst_vols[taud_slot] = min(vol64, 0x3F)
+            # IT global volume range is 0..128; rescale to Taud's 0..255.
+            inst_gv_255 = min(255, round(inst.gv * 255 / 128))
             envelopes_by_slot[taud_slot] = (
                 inst.vol_envelope, inst.vol_env_sustain,
                 inst.pan_envelope, inst.pan_env_sustain,
+                inst_gv_255,
             )
+        if bake_count:
+            vprint(f"  pf envelope baking: {bake_count} instrument(s)")
         sampleinst_raw, _ = build_sample_inst_bin_it(proxy, envelopes_by_slot)
     else:
         # Samples referenced directly; proxy is samples list (0-based, slot 0 unused)
@@ -1538,14 +1747,14 @@ def assemble_taud(h: ITHeader, samples: list, instruments: list,
     song_offset = TAUD_HEADER_SIZE + comp_size + TAUD_SONG_ENTRY
     sheet       = bytearray(NUM_CUES * CUE_SIZE)
     for c in range(NUM_CUES):
-        sheet[c*CUE_SIZE:c*CUE_SIZE+CUE_SIZE] = _encode_cue([], 0)
+        sheet[c*CUE_SIZE:c*CUE_SIZE+CUE_SIZE] = encode_cue([], 0)
 
     last_active = -1
     for cue_idx, ci in enumerate(taud_cue_list):
         if cue_idx >= NUM_CUES: break
         base_pat = cue_idx * C
         pats = [pat_remap[base_pat + vi] for vi in range(C)]
-        sheet[cue_idx*CUE_SIZE:(cue_idx+1)*CUE_SIZE] = _encode_cue(pats, 0)
+        sheet[cue_idx*CUE_SIZE:(cue_idx+1)*CUE_SIZE] = encode_cue(pats, 0)
         last_active = cue_idx
 
     if last_active >= 0:
@@ -1581,7 +1790,6 @@ def assemble_taud(h: ITHeader, samples: list, instruments: list,
 # ── Main ──────────────────────────────────────────────────────────────────────
 
 def main():
-    global VERBOSE
     ap = argparse.ArgumentParser(description=__doc__,
                                  formatter_class=argparse.RawDescriptionHelpFormatter)
     ap.add_argument('input',  help='Input .it file')
@@ -1589,8 +1797,10 @@ def main():
     ap.add_argument('-v', '--verbose', action='store_true')
     ap.add_argument('--no-decompress', action='store_true',
                     help='Treat compressed IT samples as silent (debug)')
+    ap.add_argument('--no-pf-envelope', action='store_true',
+                    help='Skip baking IT pitch/filter envelope onto sample copies')
     args = ap.parse_args()
-    VERBOSE = args.verbose
+    set_verbose(args.verbose)
 
     with open(args.input, 'rb') as f:
         data = f.read()
@@ -1608,13 +1818,14 @@ def main():
     patterns_rows = parse_patterns(data, h)
 
     taud = assemble_taud(h, samples, instruments, patterns_rows,
-                         decompress=not args.no_decompress)
+                         decompress=not args.no_decompress,
+                         no_pf_envelope=args.no_pf_envelope)
 
     with open(args.output, 'wb') as f:
         f.write(taud)
 
     print(f"wrote {len(taud)} bytes to '{args.output}'")
-    if VERBOSE:
+    if args.verbose:
         print(f"  magic ok: {taud[:8].hex()}", file=sys.stderr)
         sig_off = TAUD_HEADER_SIZE - 14
         print(f"  signature: {taud[sig_off:sig_off+14]}", file=sys.stderr)

mod2taud.py

@@ -0,0 +1,731 @@
+#!/usr/bin/env python3
+"""mod2taud.py — Convert ProTracker (.mod) to TSVM Taud (.taud)
+
+Usage:
+    python3 mod2taud.py input.mod output.taud [-v]
+
+Limits:
+    - Up to 20 MOD channels (excess disabled; hard error if pattern count
+      × channel count > 4095).
+    - Sample bin is 770048 bytes; if all samples together exceed this, every
+      sample is globally resampled down (with c2spd adjusted) so pitch is
+      preserved.
+
+Effect support:
+    Full PT effect dispatch per TAUD_NOTE_EFFECTS.md "ProTracker to Taud
+    conversion table". PT recalls (effect $00 args) are eagerly resolved
+    per channel using PT's per-effect private memory model. Cxx folds
+    into the volume column (0.$xx). Axy / EAx / EBx fold into the volume
+    column. 8xx and E8x fold into the pan column. Periods convert to Taud
+    units via log2 against PT period 428 (≡ Taud C3). Sample finetune is
+    pre-baked into the per-instrument c2spd. Amiga-mode flag is set in
+    the song-table flags byte so the engine applies coarse pitch slides
+    in period space.
+"""
+
+import argparse
+import gzip
+import math
+import struct
+import sys
+
+from taud_common import (
+    set_verbose, vprint,
+    TAUD_MAGIC, TAUD_VERSION, TAUD_HEADER_SIZE, TAUD_SONG_ENTRY,
+    SAMPLEBIN_SIZE, INSTBIN_SIZE, SAMPLEINST_SIZE,
+    PATTERN_ROWS, PATTERN_BYTES, NUM_PATTERNS_MAX, NUM_CUES, CUE_SIZE, NUM_VOICES,
+    NOTE_NOP, NOTE_KEYOFF, NOTE_CUT, TAUD_C3,
+    TOP_NONE, TOP_A, TOP_B, TOP_C, TOP_D, TOP_E, TOP_F, TOP_G, TOP_H, TOP_I,
+    TOP_J, TOP_K, TOP_L, TOP_O, TOP_Q, TOP_R, TOP_S, TOP_T, TOP_U, TOP_V, TOP_Y,
+    SEL_SET, SEL_UP, SEL_DOWN, SEL_FINE,
+    J_SEMI_TABLE,
+    d_arg_to_col, resample_linear, encode_cue, deduplicate_patterns,
+)
+
+
+# ── MOD constants ────────────────────────────────────────────────────────────
+
+MOD_NUM_SAMPLES = 31
+MOD_PATTERN_ROWS = 64
+
+# PT effect numbers (single hex digit 0..F). Effect $E uses sub-nibbles.
+PT_E_BASE = 0xE
+PT_F = 0xF
+
+# PT effects that have private memory and therefore recall their last
+# non-zero argument when re-issued with $00. Top-level effects:
+PT_MEM_TOP = frozenset({0x1, 0x2, 0x3, 0x4, 0x5, 0x6, 0x7, 0xA})
+# E sub-effects with memory (key is sub-nibble of the E command):
+PT_MEM_E_SUB = frozenset({0x1, 0x2, 0xA, 0xB})
+
+
+# ── Taud constants (mod-specific) ────────────────────────────────────────────
+
+SIGNATURE        = b"mod2taud/TSVM "    # 14 bytes
+
+# PT period 428 (PT "C-2") corresponds to OpenMPT/IT C-4 which s3m2taud
+# anchors to Taud C3 (0x4000). We use the same anchor so MOD/S3M imports
+# share a pitch reference.
+PT_REFERENCE_PERIOD = 428.0
+
+
+# ── MOD parser ───────────────────────────────────────────────────────────────
+
+class ModSample:
+    __slots__ = ('name','length','finetune','volume','loop_begin','loop_end',
+                 'sample_data','c2spd','flags')
+
+class ModRow:
+    __slots__ = ('period','inst','effect','effect_arg','vol_set')
+    def __init__(self):
+        self.period     = 0      # 0 = empty / no trigger
+        self.inst       = 0      # 0 = no instrument set
+        self.effect     = 0      # PT effect digit ($0..$F)
+        self.effect_arg = 0
+        # PT has no volume column; Cxx folds into vol_set during parsing.
+        # -1 = no explicit volume.
+        self.vol_set    = -1
+
+
+def _parse_magic(magic: bytes) -> int:
+    """Return number of channels declared by the 4-byte MOD magic."""
+    if magic in (b'M.K.', b'M!K!', b'FLT4', b'M&K!', b'N.T.'):
+        return 4
+    if magic == b'FLT8':
+        return 8
+    if magic == b'OCTA' or magic == b'CD81':
+        return 8
+    # xCHN (1..9 channels)
+    if len(magic) == 4 and magic[1:] == b'CHN' and 0x31 <= magic[0] <= 0x39:
+        return magic[0] - 0x30
+    # xxCH (10..32 channels)
+    if len(magic) == 4 and magic[2:] == b'CH' and magic[:2].isdigit():
+        return int(magic[:2].decode('ascii'))
+    # xxCN (e.g., 16CN — rare)
+    if len(magic) == 4 and magic[2:] == b'CN' and magic[:2].isdigit():
+        return int(magic[:2].decode('ascii'))
+    return 0
+
+
+def parse_mod(data: bytes):
+    if len(data) < 0x43C:
+        sys.exit("error: file too short to be a ProTracker module")
+
+    title = data[0x00:0x14].rstrip(b'\x00').decode('latin-1', errors='replace')
+
+    # 31 sample headers
+    samples = []
+    for i in range(MOD_NUM_SAMPLES):
+        base = 0x14 + i * 30
+        s = ModSample()
+        s.name        = data[base:base+22].rstrip(b'\x00').decode('latin-1', errors='replace')
+        s.length      = struct.unpack_from('>H', data, base + 22)[0] * 2
+        s.finetune    = data[base + 24] & 0x0F     # signed nibble 0..15
+        s.volume      = data[base + 25]            # 0..64
+        s.loop_begin  = struct.unpack_from('>H', data, base + 26)[0] * 2
+        loop_len_w    = struct.unpack_from('>H', data, base + 28)[0]
+        loop_len      = loop_len_w * 2
+        s.loop_end    = s.loop_begin + loop_len
+        # Flag bit 0 = looped (loop_len > 2 by convention; loop_len_w == 1 means no loop)
+        s.flags       = 1 if loop_len_w > 1 else 0
+        if not s.flags:
+            s.loop_begin = 0
+            s.loop_end   = 0
+        s.sample_data = b''
+        s.c2spd       = round(8363.0 * (2.0 ** (_signed4(s.finetune) / 96.0)))
+        samples.append(s)
+
+    song_length = data[0x3B6]
+    # 0x3B7 = restart byte (unused by us)
+    order_table = list(data[0x3B8:0x438])
+
+    magic = data[0x438:0x43C]
+    n_channels = _parse_magic(magic)
+    if n_channels == 0:
+        # Some very old MODs have only 15 samples and no magic. Detect 15-sample MOD.
+        # Header is 0x14 (title) + 15*30 (samples) = 0x14 + 0x1C2 = 0x1D6.
+        # Order table at 0x1D6, then 0x1D6+0x80 = 0x256, then patterns directly.
+        # We don't auto-detect that; require a magic.
+        sys.exit(f"error: unrecognised MOD magic {magic!r} at 0x438; "
+                 f"expected M.K., M!K!, FLT4, FLT8, xCHN or xxCH")
+
+    # Order list: only the first song_length entries are part of the song.
+    # Pattern count = 1 + max(order_table[0..127]) (scan all 128).
+    n_patterns = 1 + max(order_table)
+
+    pat_data_off = 0x43C
+    cell_size    = 4
+    pattern_size = MOD_PATTERN_ROWS * n_channels * cell_size
+
+    # Parse patterns
+    patterns = []   # patterns[pat_idx][channel][row] -> ModRow
+    for pi in range(n_patterns):
+        grid = [[ModRow() for _ in range(MOD_PATTERN_ROWS)] for _ in range(n_channels)]
+        base = pat_data_off + pi * pattern_size
+        if base + pattern_size > len(data):
+            vprint(f"  warning: pattern {pi} truncated; padding with empty rows")
+            patterns.append(grid)
+            continue
+        for r in range(MOD_PATTERN_ROWS):
+            row_off = base + r * n_channels * cell_size
+            for ch in range(n_channels):
+                cell_off = row_off + ch * cell_size
+                b0 = data[cell_off]
+                b1 = data[cell_off + 1]
+                b2 = data[cell_off + 2]
+                b3 = data[cell_off + 3]
+                period = ((b0 & 0x0F) << 8) | b1
+                inst   = (b0 & 0xF0) | ((b2 >> 4) & 0x0F)
+                effect = b2 & 0x0F
+                arg    = b3
+                cell = grid[ch][r]
+                cell.period     = period
+                cell.inst       = inst
+                cell.effect     = effect
+                cell.effect_arg = arg
+        patterns.append(grid)
+
+    # Sample data follows pattern data
+    sample_off = pat_data_off + n_patterns * pattern_size
+    for s in samples:
+        if s.length == 0:
+            continue
+        n = min(s.length, max(0, len(data) - sample_off))
+        if n <= 0:
+            break
+        raw = data[sample_off:sample_off + n]
+        # PT samples are signed 8-bit; convert to unsigned by XOR 0x80.
+        s.sample_data = bytes((b ^ 0x80) for b in raw)
+        s.length      = len(s.sample_data)
+        if s.flags:
+            s.loop_begin = min(s.loop_begin, s.length)
+            s.loop_end   = min(s.loop_end,   s.length)
+        sample_off += n
+
+    return {
+        'title':      title,
+        'samples':    samples,
+        'order_list': order_table[:song_length],
+        'order_full': order_table,
+        'n_channels': n_channels,
+        'n_patterns': n_patterns,
+        'patterns':   patterns,
+        'magic':      magic,
+    }
+
+
+def _signed4(nibble: int) -> int:
+    """Convert a 4-bit unsigned nibble to signed -8..+7."""
+    return nibble - 16 if nibble >= 8 else nibble
+
+
+# ── Note encoding (period → Taud) ────────────────────────────────────────────
+
+def period_to_taud_note(period: int) -> int:
+    if period <= 0:
+        return NOTE_NOP
+    val = round(TAUD_C3 + 4096.0 * math.log2(PT_REFERENCE_PERIOD / period))
+    return max(1, min(0xFFFD, val))
+
+
+# ── PT effect → Taud effect ──────────────────────────────────────────────────
+
+def encode_effect(cmd: int, arg: int, ch: int = 0, row: int = 0) -> tuple:
+    """Return (taud_op, taud_arg16, vol_override, pan_override).
+
+    The caller is responsible for resolving PT zero-arg recalls before this
+    point — see resolve_pt_recalls(). cmd is the raw PT digit ($0..$F).
+    """
+    # $0 with arg 0 is a true no-op; $0 with arg != 0 is arpeggio.
+    if cmd == 0x0:
+        if arg == 0:
+            return (TOP_NONE, 0, None, None)
+        hi = (arg >> 4) & 0xF
+        lo = arg & 0xF
+        return (TOP_J, (J_SEMI_TABLE[hi] << 8) | J_SEMI_TABLE[lo], None, None)
+
+    if cmd == 0x1:
+        return (TOP_F, round(arg * 64 / 3) & 0xFFFF, None, None)
+
+    if cmd == 0x2:
+        return (TOP_E, round(arg * 64 / 3) & 0xFFFF, None, None)
+
+    if cmd == 0x3:
+        return (TOP_G, round(arg * 64 / 3) & 0xFFFF, None, None)
+
+    if cmd == 0x4:
+        hi = (arg >> 4) & 0xF
+        lo = arg & 0xF
+        return (TOP_H, ((hi * 0x11) << 8) | (lo * 0x11), None, None)
+
+    if cmd == 0x5:
+        # Tone porta + vol slide → Taud L (engine splits internally).
+        return (TOP_G, 0x0000, d_arg_to_col(arg), None)
+
+    if cmd == 0x6:
+        # Vibrato + vol slide → Taud K.
+        return (TOP_H, 0x0000, d_arg_to_col(arg), None)
+
+    if cmd == 0x7:
+        hi = (arg >> 4) & 0xF
+        lo = arg & 0xF
+        return (TOP_R, ((hi * 0x11) << 8) | (lo * 0x11), None, None)
+
+    if cmd == 0x8:
+        # PT 8xx is fine pan (or unused/sync in some trackers). Map to pan
+        # column 0.$yy where yy is the upper 6 bits of the 8-bit pan.
+        return (TOP_NONE, 0, None, (SEL_SET, (arg >> 2) & 0x3F))
+
+    if cmd == 0x9:
+        return (TOP_O, (arg & 0xFF) << 8, None, None)
+
+    if cmd == 0xA:
+        return (TOP_NONE, 0, d_arg_to_col(arg), None)
+
+    if cmd == 0xB:
+        return (TOP_B, arg & 0xFF, None, None)
+
+    if cmd == 0xC:
+        # Caller folds Cxx into vol_set during parsing; this branch is a
+        # safety net in case a Cxx slips through.
+        return (TOP_NONE, 0, (SEL_SET, min(arg, 0x3F)), None)
+
+    if cmd == 0xD:
+        # PT pattern break is BCD on disk.
+        bcd_row = ((arg >> 4) & 0xF) * 10 + (arg & 0xF)
+        if bcd_row >= PATTERN_ROWS:
+            bcd_row = 0
+        return (TOP_C, bcd_row & 0xFF, None, None)
+
+    if cmd == 0xE:
+        sub = (arg >> 4) & 0xF
+        x   = arg & 0xF
+        if sub == 0x0:
+            # E0x = filter on/off (Amiga LED filter); no Taud equivalent.
+            return (TOP_NONE, 0, None, None)
+        if sub == 0x1:
+            # Fine pitch up.
+            return (TOP_F, 0xF000 | (round(x * 16 / 3) & 0xFFF), None, None)
+        if sub == 0x2:
+            # Fine pitch down.
+            return (TOP_E, 0xF000 | (round(x * 16 / 3) & 0xFFF), None, None)
+        if sub == 0x3:
+            return (TOP_S, 0x1000 | (x << 8), None, None)
+        if sub == 0x4:
+            return (TOP_S, 0x3000 | (x << 8), None, None)
+        if sub == 0x5:
+            return (TOP_S, 0x2000 | (x << 8), None, None)
+        if sub == 0x6:
+            return (TOP_S, 0xB000 | (x << 8), None, None)
+        if sub == 0x7:
+            return (TOP_S, 0x4000 | (x << 8), None, None)
+        if sub == 0x8:
+            # Coarse pan (4-bit). Map nibble 0..15 to pan 0..63 via × 4.2.
+            return (TOP_NONE, 0, None, (SEL_SET, round(x * 4.2)))
+        if sub == 0x9:
+            return (TOP_Q, (x & 0xF) << 8, None, None)
+        if sub == 0xA:
+            # Fine vol slide up.
+            return (TOP_NONE, 0, (SEL_FINE, (x & 0xF) | 0x20), None)
+        if sub == 0xB:
+            # Fine vol slide down.
+            return (TOP_NONE, 0, (SEL_FINE, x & 0xF), None)
+        if sub == 0xC:
+            return (TOP_S, 0xC000 | (x << 8), None, None)
+        if sub == 0xD:
+            return (TOP_S, 0xD000 | (x << 8), None, None)
+        if sub == 0xE:
+            return (TOP_S, 0xE000 | (x << 8), None, None)
+        if sub == 0xF:
+            return (TOP_S, 0xF000 | (x << 8), None, None)
+        return (TOP_NONE, 0, None, None)
+
+    if cmd == 0xF:
+        if arg < 0x20:
+            if arg == 0:
+                return (TOP_NONE, 0, None, None)
+            return (TOP_A, (arg & 0xFF) << 8, None, None)
+        return (TOP_T, ((arg - 0x18) & 0xFF) << 8, None, None)
+
+    return (TOP_NONE, 0, None, None)
+
+
+def resolve_pt_recalls(patterns: list, order_list: list, n_channels: int) -> None:
+    """In-place: replace PT zero-arg recalls with each effect's last non-zero arg.
+
+    PT memory is per-effect-private. Walking patterns in order-list order,
+    we track each channel's last non-zero arg per memorising effect and
+    rewrite recall args to make them explicit.
+    """
+    # mem[ch][key] = last_non_zero_arg
+    # key is either an int (top-level 0..F) or a tuple ('E', sub) for E-subs.
+    mem = [dict() for _ in range(n_channels)]
+    for order in order_list:
+        if order >= 0xFF:
+            break
+        if order >= len(patterns):
+            continue
+        grid = patterns[order]
+        for r in range(MOD_PATTERN_ROWS):
+            for ch in range(n_channels):
+                if ch >= len(grid):
+                    continue
+                row = grid[ch][r]
+                cmd = row.effect
+                arg = row.effect_arg
+                if cmd in PT_MEM_TOP:
+                    if arg == 0:
+                        row.effect_arg = mem[ch].get(cmd, 0)
+                    else:
+                        mem[ch][cmd] = arg
+                elif cmd == 0xE:
+                    sub = (arg >> 4) & 0xF
+                    x   = arg & 0xF
+                    if sub in PT_MEM_E_SUB:
+                        key = ('E', sub)
+                        if x == 0:
+                            recalled = mem[ch].get(key, 0)
+                            row.effect_arg = (sub << 4) | (recalled & 0xF)
+                        else:
+                            mem[ch][key] = x
+
+
+# ── Sample resampling and Taud sample/instrument bin (port of s3m2taud) ──────
+
+def build_sample_inst_bin(samples: list) -> tuple:
+    """Returns (bin_bytes[786432], offsets_dict). 1-based indexing."""
+    pcm = [(i, s) for i, s in enumerate(samples) if s.sample_data]
+
+    total = sum(len(s.sample_data) for _, s in pcm)
+    ratio = 1.0
+    if total > SAMPLEBIN_SIZE:
+        ratio = SAMPLEBIN_SIZE / total
+        vprint(f"  info: sample bin overflow ({total} bytes); resampling all by {ratio:.4f}")
+        for _, s in pcm:
+            new_data    = resample_linear(s.sample_data, ratio)
+            s.sample_data = new_data
+            s.length      = len(new_data)
+            s.loop_begin  = max(0, int(s.loop_begin * ratio))
+            s.loop_end    = max(0, min(int(s.loop_end * ratio), s.length))
+            s.c2spd       = max(1, int(s.c2spd * ratio))
+
+    sample_bin = bytearray(SAMPLEBIN_SIZE)
+    offsets    = {}
+    pos        = 0
+    for idx, s in pcm:
+        n = min(len(s.sample_data), SAMPLEBIN_SIZE - pos)
+        if n <= 0:
+            vprint(f"  warning: sample bin full, dropping '{s.name}'")
+            offsets[idx] = 0
+            s.length = 0
+            continue
+        sample_bin[pos:pos+n] = s.sample_data[:n]
+        offsets[idx] = pos
+        if n < len(s.sample_data):
+            vprint(f"  warning: '{s.name}' truncated from {len(s.sample_data)} to {n}")
+            s.length   = n
+            s.loop_end = min(s.loop_end, n)
+        pos += n
+
+    inst_bin = bytearray(INSTBIN_SIZE)
+    for i, s in enumerate(samples):
+        taud_idx = i + 1     # 1-based instrument number
+        if i >= 256:
+            break
+        if not s.sample_data:
+            continue
+        ptr      = offsets.get(i, 0)
+        ptr_lo   = ptr & 0xFFFF
+        ptr_hi   = (ptr >> 16)
+        s_len    = min(s.length, 65535)
+        c2spd    = min(s.c2spd, 65535)
+        ps       = 0
+        ls       = min(s.loop_begin, 65535)
+        le       = min(s.loop_end,   65535)
+        loop_mode = 1 if (s.flags & 1) else 0
+        flags_byte = (ptr_hi << 4) | (loop_mode & 0x3)
+
+        base = taud_idx * 64
+        struct.pack_into('<H', inst_bin, base + 0,  ptr_lo)
+        struct.pack_into('<H', inst_bin, base + 2,  s_len)
+        struct.pack_into('<H', inst_bin, base + 4,  c2spd)
+        struct.pack_into('<H', inst_bin, base + 6,  ps)
+        struct.pack_into('<H', inst_bin, base + 8,  ls)
+        struct.pack_into('<H', inst_bin, base + 10, le)
+        inst_bin[base + 12] = flags_byte
+        inst_bin[base + 15] = 0xFF                 # global volume — full
+        env_vol = min(s.volume, 63)
+        inst_bin[base + 16] = env_vol              # envelope hold value
+        inst_bin[base + 17] = 0                    # offset minifloat = 0 → hold
+
+        vprint(f"  instrument[{taud_idx}] '{s.name}' ptr={ptr} c2spd={s.c2spd} "
+               f"vol={s.volume} loop=({ls},{le},{'on' if loop_mode else 'off'})")
+
+    return bytes(sample_bin) + bytes(inst_bin), offsets
+
+
+# ── Pattern build ────────────────────────────────────────────────────────────
+
+# PT hard-pans channels in LRRL order: 0=L 1=R 2=R 3=L (and tile for >4).
+def _default_channel_pan(ch_idx: int) -> int:
+    side = (ch_idx % 4)
+    return 16 if side in (0, 3) else 47
+
+
+def build_pattern(grid: list, ch_idx: int, default_pan: int,
+                  inst_vols: dict) -> bytes:
+    """Build a 512-byte Taud pattern for one MOD channel.
+
+    Volume column rules (mirrors s3m2taud):
+      explicit Cxx vol > note-trigger inst default > instrument-only retrigger
+      recall > vol_override from effect > no-op.
+    """
+    out = bytearray(PATTERN_BYTES)
+    rows = grid[ch_idx] if ch_idx < len(grid) else [ModRow()] * MOD_PATTERN_ROWS
+    last_inst = 0
+    last_period = 0
+    last_vol  = None
+    for r, row in enumerate(rows[:MOD_PATTERN_ROWS]):
+        note_taud = period_to_taud_note(row.period)
+        note_triggers = (row.period > 0)
+
+        if row.inst > 0:
+            last_inst = row.inst
+
+        retrigger = (row.inst > 0
+                     and row.period == 0
+                     and last_period > 0)
+
+        op, arg, vol_override, pan_override = encode_effect(
+            row.effect, row.effect_arg, ch_idx, r)
+
+        # ── Volume column ──
+        if row.vol_set >= 0:
+            vol_sel, vol_value = SEL_SET, min(row.vol_set, 0x3F)
+            if vol_override is not None and vol_override[0] != SEL_SET:
+                vprint(f"    ch{ch_idx} row{r}: dropped vol slide "
+                       f"(cell already carries explicit Cxx volume)")
+        elif note_triggers and last_inst > 0:
+            vol_sel = SEL_SET
+            vol_value = inst_vols.get(last_inst, 0x3F)
+        elif retrigger and last_vol is not None:
+            vol_sel, vol_value = SEL_SET, last_vol
+        elif vol_override is not None:
+            vol_sel, vol_value = vol_override
+        else:
+            vol_sel, vol_value = SEL_FINE, 0
+
+        if note_triggers:
+            last_period = row.period
+        if vol_sel == SEL_SET:
+            last_vol = vol_value
+
+        # ── Pan column ──
+        if pan_override is not None:
+            pan_sel, pan_value = pan_override
+        elif r == 0:
+            pan_sel, pan_value = SEL_SET, default_pan & 0x3F
+        else:
+            pan_sel, pan_value = SEL_FINE, 0
+
+        vol_byte = (vol_value & 0x3F) | ((vol_sel & 0x3) << 6)
+        pan_byte = (pan_value & 0x3F) | ((pan_sel & 0x3) << 6)
+
+        base = r * 8
+        struct.pack_into('<H', out, base + 0, note_taud)
+        out[base + 2] = row.inst & 0xFF
+        out[base + 3] = vol_byte
+        out[base + 4] = pan_byte
+        out[base + 5] = op & 0xFF
+        struct.pack_into('<H', out, base + 6, arg & 0xFFFF)
+    return bytes(out)
+
+
+def build_cue_sheet(order_list: list, n_pats_mod: int, n_channels: int,
+                    pat_remap: dict = None) -> bytes:
+    sheet = bytearray(NUM_CUES * CUE_SIZE)
+    for c in range(NUM_CUES):
+        sheet[c*CUE_SIZE : c*CUE_SIZE+CUE_SIZE] = encode_cue([], 0)
+
+    cue_idx = 0
+    last_active = -1
+    for order in order_list:
+        if order == 0xFF or cue_idx >= NUM_CUES:
+            break
+        if order == 0xFE:
+            continue
+        if order >= n_pats_mod:
+            continue
+        orig = [order * n_channels + v for v in range(n_channels)]
+        pats = [pat_remap[p] if pat_remap else p for p in orig]
+        sheet[cue_idx*CUE_SIZE : cue_idx*CUE_SIZE+CUE_SIZE] = encode_cue(pats, 0)
+        last_active = cue_idx
+        cue_idx += 1
+
+    if last_active >= 0:
+        sheet[last_active * CUE_SIZE + 30] = 0x01
+    elif cue_idx < NUM_CUES:
+        sheet[30] = 0x01
+
+    return bytes(sheet)
+
+
+def find_initial_bpm_speed(patterns: list, order_list: list) -> tuple:
+    """Scan first pattern in order for Fxx in row 0 of any channel."""
+    speed = 6
+    tempo = 125
+    for order in order_list:
+        if order >= 0xFF:
+            break
+        if order >= len(patterns):
+            continue
+        grid = patterns[order]
+        for ch_rows in grid:
+            row = ch_rows[0]
+            if row.effect == 0xF and row.effect_arg > 0:
+                if row.effect_arg < 0x20:
+                    speed = row.effect_arg
+                else:
+                    tempo = row.effect_arg
+        break
+    return speed, tempo
+
+
+def assemble_taud(mod: dict) -> bytes:
+    samples    = mod['samples']
+    patterns   = mod['patterns']
+    order_list = mod['order_list']
+    n_channels = mod['n_channels']
+    n_patterns = mod['n_patterns']
+
+    if n_channels > NUM_VOICES:
+        vprint(f"  warning: MOD has {n_channels} channels; truncating to {NUM_VOICES}")
+        n_channels = NUM_VOICES
+
+    if n_patterns * n_channels > NUM_PATTERNS_MAX:
+        sys.exit(
+            f"error: {n_patterns} MOD patterns × {n_channels} channels = "
+            f"{n_patterns*n_channels} > {NUM_PATTERNS_MAX} Taud pattern limit.\n"
+            f"  Reduce the MOD to ≤ {NUM_PATTERNS_MAX // max(n_channels,1)} patterns."
+        )
+
+    vprint(f"  channels: {n_channels}, mod patterns: {n_patterns}, "
+           f"taud patterns: {n_patterns * n_channels}")
+
+    # Fold Cxx into row.vol_set so the volume column carries explicit set-volume.
+    # This is done in-place before recall resolution so Cxx with arg 0 still
+    # resolves to vol 0 (silence) rather than recalling another effect's memory.
+    for grid in patterns:
+        for ch in range(min(n_channels, len(grid))):
+            for row in grid[ch]:
+                if row.effect == 0xC:
+                    row.vol_set = min(row.effect_arg, 0x3F)
+                    row.effect = 0
+                    row.effect_arg = 0
+
+    vprint("  resolving PT per-effect recalls…")
+    resolve_pt_recalls(patterns, order_list, n_channels)
+
+    vprint("  building sample/instrument bin…")
+    sampleinst_raw, _offsets = build_sample_inst_bin(samples)
+    assert len(sampleinst_raw) == SAMPLEINST_SIZE
+
+    compressed = gzip.compress(sampleinst_raw, compresslevel=9, mtime=0)
+    comp_size  = len(compressed)
+    vprint(f"  sample+inst bin: {SAMPLEINST_SIZE} → {comp_size} bytes (gzip)")
+
+    speed, tempo = find_initial_bpm_speed(patterns, order_list)
+    tempo = max(24, min(280, tempo))
+    bpm_stored = (tempo - 24) & 0xFF
+    vprint(f"  initial speed={speed}, tempo(BPM)={tempo}")
+
+    song_offset = TAUD_HEADER_SIZE + comp_size + TAUD_SONG_ENTRY
+
+    sig = (SIGNATURE + b' ' * 14)[:14]
+    header = (
+        TAUD_MAGIC +
+        bytes([TAUD_VERSION, 1]) +
+        struct.pack('<I', comp_size) +
+        b'\x00\x00\x00\x00' +
+        sig
+    )
+    assert len(header) == TAUD_HEADER_SIZE
+
+    vprint("  building pattern bin…")
+    inst_vols = {
+        i + 1: min(s.volume, 0x3F)
+        for i, s in enumerate(samples)
+        if s.sample_data
+    }
+    pat_bin = bytearray()
+    for pi in range(n_patterns):
+        grid = patterns[pi]
+        for ch in range(n_channels):
+            default_pan = _default_channel_pan(ch)
+            pat_bin += build_pattern(grid, ch, default_pan, inst_vols)
+    assert len(pat_bin) == n_patterns * n_channels * PATTERN_BYTES
+
+    vprint("  deduplicating patterns…")
+    orig_count = n_patterns * n_channels
+    pat_bin, pat_remap, num_taud_pats = deduplicate_patterns(bytes(pat_bin), orig_count)
+    vprint(f"  patterns: {orig_count} → {num_taud_pats} unique "
+           f"({orig_count - num_taud_pats} deduplicated)")
+
+    # ProTracker is Amiga-period-based by definition, so we set the f bit so
+    # the engine applies coarse pitch slides in period space (recovers PT's
+    # characteristic non-linear pitch character).
+    flags_byte = 0x02
+    song_table = struct.pack('<IBHBBHfB',
+        song_offset,
+        n_channels,
+        num_taud_pats,
+        bpm_stored,
+        speed,
+        0xA000,
+        8363.0,
+        flags_byte,
+    )
+    assert len(song_table) == TAUD_SONG_ENTRY
+
+    vprint("  building cue sheet…")
+    cue_sheet = build_cue_sheet(order_list, n_patterns, n_channels, pat_remap)
+    assert len(cue_sheet) == NUM_CUES * CUE_SIZE
+
+    return header + compressed + song_table + bytes(pat_bin) + cue_sheet
+
+
+# ── Main ─────────────────────────────────────────────────────────────────────
+
+def main():
+    ap = argparse.ArgumentParser(description=__doc__,
+                                 formatter_class=argparse.RawDescriptionHelpFormatter)
+    ap.add_argument('input',  help='Input .mod file')
+    ap.add_argument('output', help='Output .taud file')
+    ap.add_argument('-v', '--verbose', action='store_true',
+                    help='Print conversion details to stderr')
+    args = ap.parse_args()
+
+    set_verbose(args.verbose)
+
+    with open(args.input, 'rb') as f:
+        data = f.read()
+
+    vprint(f"parsing '{args.input}' ({len(data)} bytes)…")
+    mod = parse_mod(data)
+    vprint(f"  title: '{mod['title']}'")
+    vprint(f"  magic: {mod['magic']!r} ({mod['n_channels']} channels)")
+    vprint(f"  orders={len(mod['order_list'])}, patterns={mod['n_patterns']}, "
+           f"samples={sum(1 for s in mod['samples'] if s.sample_data)}")
+
+    taud = assemble_taud(mod)
+
+    with open(args.output, 'wb') as f:
+        f.write(taud)
+
+    print(f"wrote {len(taud)} bytes to '{args.output}'")
+    if args.verbose:
+        print(f"  magic ok: {taud[:8].hex()}", file=sys.stderr)
+
+
+if __name__ == '__main__':
+    main()

s3m2taud.py

@@ -29,11 +29,22 @@ import math
 import struct
 import sys
 
-VERBOSE = False
+from taud_common import (
-
+    set_verbose, vprint,
-def vprint(*a, **kw):
+    TAUD_MAGIC, TAUD_VERSION, TAUD_HEADER_SIZE, TAUD_SONG_ENTRY,
-    if VERBOSE:
+    SAMPLEBIN_SIZE, INSTBIN_SIZE, SAMPLEINST_SIZE,
-        print(*a, **kw, file=sys.stderr)
+    PATTERN_ROWS, PATTERN_BYTES, NUM_PATTERNS_MAX, NUM_CUES, CUE_SIZE, NUM_VOICES,
+    NOTE_NOP, NOTE_KEYOFF, NOTE_CUT, TAUD_C3,
+    TOP_NONE, TOP_A, TOP_B, TOP_C, TOP_D, TOP_E, TOP_F, TOP_G, TOP_H, TOP_I,
+    TOP_J, TOP_K, TOP_L, TOP_O, TOP_Q, TOP_R, TOP_S, TOP_T, TOP_U, TOP_V, TOP_Y,
+    SEL_SET, SEL_UP, SEL_DOWN, SEL_FINE,
+    EFF_A, EFF_B, EFF_C, EFF_D, EFF_E, EFF_F, EFF_G, EFF_H, EFF_I, EFF_J,
+    EFF_K, EFF_L, EFF_M, EFF_N, EFF_O, EFF_P, EFF_Q, EFF_R, EFF_S, EFF_T,
+    EFF_U, EFF_V, EFF_W, EFF_X, EFF_Y, EFF_Z,
+    J_SEMI_TABLE,
+    d_arg_to_col, resample_linear, encode_cue, deduplicate_patterns,
+    normalise_sample,
+)
 
 
 # ── S3M constants ────────────────────────────────────────────────────────────
@@ -45,92 +56,8 @@ S3M_NOTE_OFF     = 0xFE
 S3M_ORDER_SKIP   = 0xFE
 S3M_ORDER_END    = 0xFF
 
-# S3M effect letters (1-based: 1='A', 2='B', …)
-EFF_A = 1   # set speed
-EFF_B = 2   # jump to order
-EFF_C = 3   # pattern break
-EFF_D = 4   # volume slide
-EFF_E = 5   # porta down
-EFF_F = 6   # porta up
-EFF_G = 7   # tone porta
-EFF_H = 8   # vibrato
-EFF_I = 9   # tremor
-EFF_J = 10  # arpeggio
-EFF_K = 11  # vibrato+volslide
-EFF_L = 12  # porta+volslide
-EFF_M = 13  # channel vol
-EFF_N = 14  # chan vol slide
-EFF_O = 15  # sample offset
-EFF_P = 16  # pan slide
-EFF_Q = 17  # retrigger
-EFF_R = 18  # tremolo
-EFF_S = 19  # special (sub-cmds)
-EFF_T = 20  # set BPM
-EFF_U = 21  # fine vibrato
-EFF_V = 22  # global vol
-EFF_W = 23  # global vol slide
-EFF_X = 24  # set pan
-EFF_Y = 25  # panbrello
-EFF_Z = 26  # sync
-
-
-# ── Taud constants ───────────────────────────────────────────────────────────
-
-TAUD_MAGIC       = bytes([0x1F,0x54,0x53,0x56,0x4D,0x61,0x75,0x64])
-TAUD_VERSION     = 1
-TAUD_HEADER_SIZE = 32    # magic(8)+ver(1)+numSongs(1)+compSize(4)+rsvd(2)+sig(16)
-TAUD_SONG_ENTRY  = 16   # offset(4)+voices(1)+pats(2)+bpm(1)+tick(1)+basenote(2)+basefreq(4)+flags(1)
-SAMPLEBIN_SIZE   = 770048
-INSTBIN_SIZE     = 16384   # 256 instruments × 64 bytes
-SAMPLEINST_SIZE  = SAMPLEBIN_SIZE + INSTBIN_SIZE  # 786432
-PATTERN_ROWS     = 64
-PATTERN_BYTES    = PATTERN_ROWS * 8   # 512
-NUM_PATTERNS_MAX = 4095
-NUM_CUES         = 1024
-CUE_SIZE         = 32   # packed 12-bit×20 voices + instruction + pad
-NUM_VOICES       = 20
 SIGNATURE        = b"s3m2taud/TSVM "    # 14 bytes
 
-# Taud note constants
-NOTE_NOP    = 0xFFFF
-NOTE_KEYOFF = 0x0000
-NOTE_CUT    = 0xFFFE
-TAUD_C3     = 0x4000
-
-# Taud effect opcode bytes (base-36: 0..9 → 0x00..0x09, A..Z → 0x0A..0x23)
-TOP_NONE = 0x00
-TOP_A    = 0x0A   # set tick speed
-TOP_B    = 0x0B   # jump to order
-TOP_C    = 0x0C   # break to row
-TOP_D    = 0x0D   # volume slide
-TOP_E    = 0x0E   # pitch slide down
-TOP_F    = 0x0F   # pitch slide up
-TOP_G    = 0x10   # tone porta
-TOP_H    = 0x11   # vibrato
-TOP_I    = 0x12   # tremor
-TOP_J    = 0x13   # microtonal arpeggio
-TOP_K    = 0x14   # vibrato + vol slide (engine no-op; converter splits)
-TOP_L    = 0x15   # tone porta + vol slide (engine no-op; converter splits)
-TOP_O    = 0x18   # sample offset
-TOP_Q    = 0x1A   # retrigger
-TOP_R    = 0x1B   # tremolo
-TOP_S    = 0x1C   # sub-effects
-TOP_T    = 0x1D   # tempo set/slide
-TOP_U    = 0x1E   # fine vibrato
-TOP_V    = 0x1F   # global volume
-TOP_Y    = 0x22   # panbrello
-
-# Volume / pan column selectors (2-bit field, packed into top of vol/pan byte).
-SEL_SET  = 0   # 6-bit value: set vol / pan
-SEL_UP   = 1   # 6-bit per-tick slide up / right
-SEL_DOWN = 2   # 6-bit per-tick slide down / left
-SEL_FINE = 3   # 1-bit dir + 5-bit magnitude, fired on tick 0
-
-# 12-TET semitone → Taud J-arpeggio byte (high byte of pitch delta).
-# byte = round(semitone * 4096 / 12 / 256) = round(semitone * 4 / 3).
-J_SEMI_TABLE = [0x00, 0x01, 0x03, 0x04, 0x05, 0x07, 0x08, 0x09,
-                0x0B, 0x0C, 0x0D, 0x0F, 0x10, 0x11, 0x13, 0x14]
-
 # ST3's single shared memory slot backs these effects.
 ST3_SHARED_EFFECTS = frozenset({
     EFF_D, EFF_E, EFF_F, EFF_I, EFF_J, EFF_K, EFF_L, EFF_Q, EFF_R, EFF_S
@@ -234,7 +161,7 @@ def parse_instruments(data: bytes, h: S3MHeader) -> list:
                 inst.sample_data = bytes(min(sample_len, 256))
             else:
                 raw = data[sample_off:sample_off + sample_len]
-                inst.sample_data = _normalise_sample(raw, inst.signed, is_16bit, is_stereo, inst.name)
+                inst.sample_data = normalise_sample(raw, inst.signed, is_16bit, is_stereo, inst.name)
                 inst.length      = len(inst.sample_data)
                 inst.loop_begin  = min(inst.loop_begin, inst.length)
                 inst.loop_end    = min(inst.loop_end,   inst.length)
@@ -242,37 +169,6 @@ def parse_instruments(data: bytes, h: S3MHeader) -> list:
     return insts
 
 
-def _normalise_sample(raw: bytes, signed: bool, is_16bit: bool, is_stereo: bool, name: str) -> bytes:
-    """Return unsigned 8-bit mono sample bytes."""
-    out = []
-    stride = (2 if is_16bit else 1) * (2 if is_stereo else 1)
-    i = 0
-    while i + stride <= len(raw):
-        if is_16bit:
-            if is_stereo:
-                l16 = struct.unpack_from('<h', raw, i)[0]
-                r16 = struct.unpack_from('<h', raw, i+2)[0]
-                s = (l16 + r16) >> 1
-            else:
-                s = struct.unpack_from('<h', raw, i)[0]
-            v = (s >> 8) + 128
-        else:
-            if is_stereo:
-                l8 = raw[i]; r8 = raw[i+1]
-                raw_s = (l8 + r8) // 2
-            else:
-                raw_s = raw[i]
-            if signed:
-                v = ((raw_s ^ 0x80) & 0xFF)   # signed→unsigned
-            else:
-                v = raw_s
-        out.append(v & 0xFF)
-        i += stride
-    if is_16bit or is_stereo:
-        vprint(f"  info: '{name}' converted to unsigned 8-bit mono ({len(out)} samples)")
-    return bytes(out)
-
-
 # ── S3M pattern parser ───────────────────────────────────────────────────────
 
 class S3MRow:
@@ -339,28 +235,6 @@ def encode_note(s3m_note: int) -> int:
     return max(1, min(0xFFFD, val))
 
 
-def _d_arg_to_col(arg: int):
-    """Convert an ST3 D-style two-nibble vol/pan slide arg into a column override.
-
-    Returns (selector, value) or None for no-op. Volume column treats
-    selector 1 as up / 2 as down; pan column reuses 1 = right, 2 = left.
-    """
-    if arg == 0:
-        return None
-    hi = (arg >> 4) & 0xF
-    lo = arg & 0xF
-    if hi == 0xF and lo > 0:
-        return (SEL_FINE, lo & 0x1F)              # fine slide down (dir bit 0)
-    if lo == 0xF and hi > 0:
-        return (SEL_FINE, (hi & 0x1F) | 0x20)     # fine slide up (dir bit 1)
-    if hi > 0 and lo == 0:
-        return (SEL_UP, hi)
-    if lo > 0 and hi == 0:
-        return (SEL_DOWN, lo)
-    # Both nibbles non-zero, neither $F → ambiguous; ST3 prefers up.
-    return (SEL_UP, hi)
-
-
 def encode_effect(cmd: int, arg: int, ch: int = 0, row: int = 0) -> tuple:
     """Return (taud_op, taud_arg16, vol_override, pan_override).
 
@@ -417,23 +291,23 @@ def encode_effect(cmd: int, arg: int, ch: int = 0, row: int = 0) -> tuple:
     if cmd == EFF_K:
         # K = vibrato continuation + vol slide; engine treats K as no-op.
         # Split into: H $0000 (recall vibrato from HU memory) + vol-col slide.
-        return (TOP_H, 0x0000, _d_arg_to_col(arg), None)
+        return (TOP_H, 0x0000, d_arg_to_col(arg), None)
 
     if cmd == EFF_L:
         # L = tone-porta continuation + vol slide; split similarly.
-        return (TOP_G, 0x0000, _d_arg_to_col(arg), None)
+        return (TOP_G, 0x0000, d_arg_to_col(arg), None)
 
     if cmd == EFF_M:
         return (TOP_NONE, 0, (SEL_SET, min(arg, 0x3F)), None)
 
     if cmd == EFF_N:
-        return (TOP_NONE, 0, _d_arg_to_col(arg), None)
+        return (TOP_NONE, 0, d_arg_to_col(arg), None)
 
     if cmd == EFF_O:
         return (TOP_O, (arg & 0xFF) << 8, None, None)
 
     if cmd == EFF_P:
-        return (TOP_NONE, 0, None, _d_arg_to_col(arg))
+        return (TOP_NONE, 0, None, d_arg_to_col(arg))
 
     if cmd == EFF_Q:
         return (TOP_Q, (arg & 0xFF) << 8, None, None)
@@ -541,22 +415,6 @@ def warn_st3_quirks(patterns: list, order_list: list, num_channels: int) -> None
 
 # ── Taud builders ────────────────────────────────────────────────────────────
 
-def _resample_linear(data: bytes, ratio: float) -> bytes:
-    """Resample bytes by ratio (< 1 = downsample) using linear interpolation."""
-    if not data:
-        return data
-    n_out = max(1, int(len(data) * ratio))
-    out   = bytearray(n_out)
-    for i in range(n_out):
-        src = i / ratio
-        i0  = int(src)
-        frac = src - i0
-        i1   = min(i0 + 1, len(data) - 1)
-        v    = data[i0] * (1.0 - frac) + data[i1] * frac
-        out[i] = int(v + 0.5) & 0xFF
-    return bytes(out)
-
-
 def build_sample_inst_bin(instruments: list) -> tuple:
     """
     Returns (bin_bytes[786432], offsets_list, updated_insts).
@@ -571,7 +429,7 @@ def build_sample_inst_bin(instruments: list) -> tuple:
         ratio = SAMPLEBIN_SIZE / total
         vprint(f"  info: sample bin overflow ({total} bytes); resampling all by {ratio:.4f}")
         for _, inst in pcm_insts:
-            new_data = _resample_linear(inst.sample_data, ratio)
+            new_data = resample_linear(inst.sample_data, ratio)
             old_len  = len(inst.sample_data)
             inst.sample_data  = new_data
             inst.length       = len(new_data)
@@ -624,6 +482,7 @@ def build_sample_inst_bin(instruments: list) -> tuple:
         struct.pack_into('<H', inst_bin, base + 8,  ls)
         struct.pack_into('<H', inst_bin, base + 10, le)
         inst_bin[base + 12] = flags_byte
+        inst_bin[base + 15] = 0xFF      # instrument global volume (S3M has none → full)
         # Volume envelope: hold at instrument volume (clamped to 0x3F)
         env_vol = min(inst.volume, 63)
         inst_bin[base + 16] = env_vol   # volume
@@ -733,48 +592,13 @@ def build_pattern(s3m_grid: list, ch_idx: int, default_pan: int,
     return bytes(out)
 
 
-def deduplicate_patterns(pat_bin: bytes, num_pats: int) -> tuple:
-    """
-    Consolidate identical 512-byte Taud patterns into a single copy.
-    Returns (deduped_bin, remap, num_unique) where remap[original_idx] = canonical_idx.
-    """
-    seen = {}       # pattern_bytes -> canonical_index
-    remap = {}      # original_index -> canonical_index
-    canonical = []
-    for i in range(num_pats):
-        pat = pat_bin[i * PATTERN_BYTES : (i + 1) * PATTERN_BYTES]
-        if pat in seen:
-            remap[i] = seen[pat]
-        else:
-            ci = len(canonical)
-            seen[pat] = ci
-            remap[i] = ci
-            canonical.append(pat)
-    return b''.join(canonical), remap, len(canonical)
-
-
-def _encode_cue(patterns12: list, instruction: int) -> bytearray:
-    """Encode a 32-byte cue entry for up to 20 voices with 12-bit pattern numbers."""
-    # patterns12: list of up to NUM_VOICES 12-bit values (0xFFF = disabled)
-    pats = list(patterns12) + [0xFFF] * NUM_VOICES
-    pats = pats[:NUM_VOICES]
-    entry = bytearray(CUE_SIZE)
-    for i in range(10):      # 10 bytes: 2 voices per byte
-        v0, v1 = pats[i*2], pats[i*2+1]
-        entry[i]      = ((v0 & 0xF) << 4) | (v1 & 0xF)        # low nybbles
-        entry[10 + i] = (((v0 >> 4) & 0xF) << 4) | ((v1 >> 4) & 0xF)  # mid nybbles
-        entry[20 + i] = (((v0 >> 8) & 0xF) << 4) | ((v1 >> 8) & 0xF)  # high nybbles
-    entry[30] = instruction & 0xFF
-    return entry
-
-
 def build_cue_sheet(order_list: list, num_pats_s3m: int, num_channels: int,
                     pat_remap: dict = None) -> bytes:
     """Build the 1024×32-byte cue sheet with 12-bit packed pattern numbers."""
     sheet = bytearray(NUM_CUES * CUE_SIZE)
     # Fill entire sheet with the "all disabled" cue (patterns=0xFFF, instr=0)
     for c in range(NUM_CUES):
-        sheet[c*CUE_SIZE : c*CUE_SIZE+CUE_SIZE] = _encode_cue([], 0)
+        sheet[c*CUE_SIZE : c*CUE_SIZE+CUE_SIZE] = encode_cue([], 0)
 
     cue_idx = 0
     last_active = -1
@@ -785,7 +609,7 @@ def build_cue_sheet(order_list: list, num_pats_s3m: int, num_channels: int,
             continue
         orig = [order * num_channels + v for v in range(num_channels)]
         pats = [pat_remap[p] if pat_remap else p for p in orig]
-        sheet[cue_idx*CUE_SIZE : cue_idx*CUE_SIZE+CUE_SIZE] = _encode_cue(pats, 0)
+        sheet[cue_idx*CUE_SIZE : cue_idx*CUE_SIZE+CUE_SIZE] = encode_cue(pats, 0)
         last_active = cue_idx
         cue_idx += 1
 
@@ -925,7 +749,6 @@ def assemble_taud(h: S3MHeader, instruments: list, patterns: list) -> bytes:
 # ── Main ─────────────────────────────────────────────────────────────────────
 
 def main():
-    global VERBOSE
     ap = argparse.ArgumentParser(description=__doc__,
                                  formatter_class=argparse.RawDescriptionHelpFormatter)
     ap.add_argument('input',  help='Input .s3m file')
@@ -934,7 +757,7 @@ def main():
                     help='Print conversion details to stderr')
     args = ap.parse_args()
 
-    VERBOSE = args.verbose
+    set_verbose(args.verbose)
 
     with open(args.input, 'rb') as f:
         data = f.read()
@@ -953,7 +776,7 @@ def main():
         f.write(taud)
 
     print(f"wrote {len(taud)} bytes to '{args.output}'")
-    if VERBOSE:
+    if args.verbose:
         print(f"  magic ok: {taud[:8].hex()}", file=sys.stderr)
 
 

taud_common.py

@@ -0,0 +1,197 @@
+"""taud_common.py — Shared constants and helpers for *2taud converters.
+
+Imported by s3m2taud.py, it2taud.py, and mod2taud.py. Holds the Taud
+container constants, the effect-letter index table, and the small set
+of helpers (sample resampler, vol/pan column packer, cue encoder,
+pattern deduper, sample normaliser) that all three converters used to
+duplicate verbatim.
+"""
+
+import struct
+import sys
+
+
+# ── Verbose logging (shared across converters via set_verbose) ───────────────
+
+VERBOSE = False
+
+def set_verbose(b: bool) -> None:
+    global VERBOSE
+    VERBOSE = bool(b)
+
+def vprint(*a, **kw) -> None:
+    if VERBOSE:
+        print(*a, **kw, file=sys.stderr)
+
+
+# ── Taud container constants ─────────────────────────────────────────────────
+
+TAUD_MAGIC       = bytes([0x1F,0x54,0x53,0x56,0x4D,0x61,0x75,0x64])
+TAUD_VERSION     = 1
+TAUD_HEADER_SIZE = 32       # magic(8)+ver(1)+numSongs(1)+compSize(4)+rsvd(4)+sig(14)
+TAUD_SONG_ENTRY  = 16       # offset(4)+voices(1)+pats(2)+bpm(1)+tick(1)+basenote(2)+basefreq(4)+flags(1)
+SAMPLEBIN_SIZE   = 770048
+INSTBIN_SIZE     = 16384    # 256 instruments × 64 bytes
+SAMPLEINST_SIZE  = SAMPLEBIN_SIZE + INSTBIN_SIZE
+PATTERN_ROWS     = 64
+PATTERN_BYTES    = PATTERN_ROWS * 8     # 512
+NUM_PATTERNS_MAX = 4095
+NUM_CUES         = 1024
+CUE_SIZE         = 32
+NUM_VOICES       = 20
+
+# Note word sentinels
+NOTE_NOP    = 0xFFFF
+NOTE_KEYOFF = 0x0000
+NOTE_CUT    = 0xFFFE
+TAUD_C3     = 0x4000
+
+# Taud effect opcodes (base-36: 0..9 → 0x00..0x09, A..Z → 0x0A..0x23)
+TOP_NONE = 0x00
+TOP_A    = 0x0A
+TOP_B    = 0x0B
+TOP_C    = 0x0C
+TOP_D    = 0x0D
+TOP_E    = 0x0E
+TOP_F    = 0x0F
+TOP_G    = 0x10
+TOP_H    = 0x11
+TOP_I    = 0x12
+TOP_J    = 0x13
+TOP_K    = 0x14
+TOP_L    = 0x15
+TOP_O    = 0x18
+TOP_Q    = 0x1A
+TOP_R    = 0x1B
+TOP_S    = 0x1C
+TOP_T    = 0x1D
+TOP_U    = 0x1E
+TOP_V    = 0x1F
+TOP_Y    = 0x22
+
+# Volume / pan column selectors (2-bit field at top of vol/pan byte)
+SEL_SET  = 0     # 6-bit value: set vol / pan
+SEL_UP   = 1     # 6-bit per-tick slide up / right
+SEL_DOWN = 2     # 6-bit per-tick slide down / left
+SEL_FINE = 3     # 1-bit dir + 5-bit magnitude, fired on tick 0
+
+# 12-TET semitone → Taud J-arpeggio byte (high byte of pitch delta).
+# byte = round(semitone * 4096 / 12 / 256) = round(semitone * 4 / 3).
+J_SEMI_TABLE = [0x00, 0x01, 0x03, 0x04, 0x05, 0x07, 0x08, 0x09,
+                0x0B, 0x0C, 0x0D, 0x0F, 0x10, 0x11, 0x13, 0x14]
+
+# Effect-letter indices (1-based; A=1..Z=26). Shared by s3m2taud and it2taud.
+EFF_A = 1;  EFF_B = 2;  EFF_C = 3;  EFF_D = 4;  EFF_E = 5
+EFF_F = 6;  EFF_G = 7;  EFF_H = 8;  EFF_I = 9;  EFF_J = 10
+EFF_K = 11; EFF_L = 12; EFF_M = 13; EFF_N = 14; EFF_O = 15
+EFF_P = 16; EFF_Q = 17; EFF_R = 18; EFF_S = 19; EFF_T = 20
+EFF_U = 21; EFF_V = 22; EFF_W = 23; EFF_X = 24; EFF_Y = 25
+EFF_Z = 26
+
+
+# ── Helpers ──────────────────────────────────────────────────────────────────
+
+def d_arg_to_col(arg: int):
+    """Convert a two-nibble D-style vol/pan slide arg into a column override.
+
+    Returns (selector, value) or None for no-op. Volume column treats
+    selector 1 as up / 2 as down; pan column reuses 1 = right, 2 = left.
+    Both-nibbles-non-zero (and neither $F) is ambiguous; ST3/PT/IT all
+    prefer up.
+    """
+    if arg == 0:
+        return None
+    hi = (arg >> 4) & 0xF
+    lo = arg & 0xF
+    if hi == 0xF and lo > 0:
+        return (SEL_FINE, lo & 0x1F)              # fine slide down (dir bit 0)
+    if lo == 0xF and hi > 0:
+        return (SEL_FINE, (hi & 0x1F) | 0x20)     # fine slide up (dir bit 1)
+    if hi > 0 and lo == 0:
+        return (SEL_UP, hi)
+    if lo > 0 and hi == 0:
+        return (SEL_DOWN, lo)
+    return (SEL_UP, hi)
+
+
+def resample_linear(data: bytes, ratio: float) -> bytes:
+    """Resample bytes by ratio (< 1 = downsample) using linear interpolation."""
+    if not data:
+        return data
+    n_out = max(1, int(len(data) * ratio))
+    out   = bytearray(n_out)
+    for i in range(n_out):
+        src  = i / ratio
+        i0   = int(src)
+        frac = src - i0
+        i1   = min(i0 + 1, len(data) - 1)
+        v    = data[i0] * (1.0 - frac) + data[i1] * frac
+        out[i] = int(v + 0.5) & 0xFF
+    return bytes(out)
+
+
+def encode_cue(patterns12: list, instruction: int) -> bytearray:
+    """Encode a 32-byte cue entry for up to 20 voices with 12-bit pattern numbers."""
+    pats = list(patterns12) + [0xFFF] * NUM_VOICES
+    pats = pats[:NUM_VOICES]
+    entry = bytearray(CUE_SIZE)
+    for i in range(10):      # 10 bytes: 2 voices per byte
+        v0, v1 = pats[i*2], pats[i*2+1]
+        entry[i]      = ((v0 & 0xF) << 4) | (v1 & 0xF)               # low nybbles
+        entry[10 + i] = (((v0 >> 4) & 0xF) << 4) | ((v1 >> 4) & 0xF) # mid nybbles
+        entry[20 + i] = (((v0 >> 8) & 0xF) << 4) | ((v1 >> 8) & 0xF) # high nybbles
+    entry[30] = instruction & 0xFF
+    return entry
+
+
+def deduplicate_patterns(pat_bin: bytes, num_pats: int) -> tuple:
+    """Consolidate identical 512-byte Taud patterns into a single copy.
+
+    Returns (deduped_bin, remap, num_unique) where remap[original_idx] =
+    canonical_idx.
+    """
+    seen = {}
+    remap = {}
+    canonical = []
+    for i in range(num_pats):
+        pat = pat_bin[i * PATTERN_BYTES : (i + 1) * PATTERN_BYTES]
+        if pat in seen:
+            remap[i] = seen[pat]
+        else:
+            ci = len(canonical)
+            seen[pat] = ci
+            remap[i] = ci
+            canonical.append(pat)
+    return b''.join(canonical), remap, len(canonical)
+
+
+def normalise_sample(raw: bytes, signed: bool, is_16bit: bool,
+                     is_stereo: bool, name: str) -> bytes:
+    """Return unsigned 8-bit mono sample bytes, downmixing/depthing as needed."""
+    out = []
+    stride = (2 if is_16bit else 1) * (2 if is_stereo else 1)
+    i = 0
+    while i + stride <= len(raw):
+        if is_16bit:
+            if is_stereo:
+                l16 = struct.unpack_from('<h', raw, i)[0]
+                r16 = struct.unpack_from('<h', raw, i+2)[0]
+                s = (l16 + r16) >> 1
+            else:
+                s = struct.unpack_from('<h', raw, i)[0]
+            v = (s >> 8) + 128
+        else:
+            if is_stereo:
+                l8 = raw[i]; r8 = raw[i+1]
+                raw_s = (l8 + r8) // 2
+            else:
+                raw_s = raw[i]
+            if signed:
+                v = (raw_s ^ 0x80) & 0xFF
+            else:
+                v = raw_s
+        out.append(v & 0xFF)
+        i += stride
+    if is_16bit or is_stereo:
+        vprint(f"  info: '{name}' converted to unsigned 8-bit mono ({len(out)} samples)")
+    return bytes(out)

terranmon.txt

@@ -2005,24 +2005,29 @@ Instrument bin: Registry for 256 instruments, formatted as:
         0b hhhh 00pp
             h: sample pointer high bit
            pp: loop mode. 0-no loop, 1-loop, 2-backandforth, 3-oneshot (ignores note length unless overridden by other notes)
-    Bit8   Volume envelope sustain loops
+    Bit8   Volume envelope sustain/loops
-        0b u0 eee sss
+        * Sustain is implemented by enabling 't' flag. FastTracker has no 'Sus Loop' but only 'Sus Point'; use same value for start and end index
-            s: sustain loop start index
+        0b ut eee sss
-            e: sustain loop end index
+            s: sustain/loop start index
-            u: set to enable the loop
+            e: sustain/loop end index
-    Bit8   Panning envelope sustain loops
+            t: the loop must sustain (key-off escapes the loop)
-        0b u0 eee sss
+            u: set to enable the sustain/loop
-            s: sustain loop start index
+    Bit8   Panning envelope sustain/loops
-            e: sustain loop end index
+        * Sustain is implemented by enabling 't' flag
-            u: set to enable the loop
+        0b ut eee sss
-    Bit8   Reserved
+            s: sustain/loop start index
-  Bit16x8 Volume envelopes
+            e: sustain/loop end index
+            t: the loop must sustain (key-off escapes the loop)
+            u: set to enable the sustain/loop
+    Uint8  Instrument Global Volume (0..255)
+        * ImpulseTracker has range of 0..128; multiply by (255/128) then round to int
+        * FastTracker2 has range of 0..64; multiply by (255/64) then round to int
+    Bit16x12 Volume envelopes
        Byte 1: Volume (00..3F)
        Byte 2: Time until the next point, in seconds (3.5 Unsigned Minifloat). 0 = hold at this point indefinitely.
-  Bit16x8 Panning envelopes
+    Bit16x12 Panning envelopes
       Byte 1: Pan (00..FF)
       Byte 2: Time until the next point, in seconds (3.5 Unsigned Minifloat). 0 = hold at this point indefinitely.
-  Bit16x8 Reserved
 
 Play Data: play data are series of tracker-like instructions, visualised as:
 

tsvm_core/src/net/torvald/tsvm/AudioJSR223Delegate.kt

@@ -131,6 +131,14 @@ class AudioJSR223Delegate(private val vm: VM) {
         }
     }
 
+    fun setTrackerMixerFlags(playhead: Int, flags: Int) {
+        getFirstSnd()?.playheads?.get(playhead)?.initialGlobalFlags = flags
+    }
+
+    fun getTrackerMixerFlags(playhead: Int): Int? {
+        return getFirstSnd()?.playheads?.get(playhead)?.initialGlobalFlags
+    }
+
     fun putPcmDataByPtr(playhead: Int, ptr: Int, length: Int, destOffset: Int) {
         getFirstSnd()?.let {
             val vkMult = if (ptr >= 0) 1 else -1

tsvm_core/src/net/torvald/tsvm/peripheral/AudioAdapter.kt

@@ -1184,17 +1184,26 @@ class AudioAdapter(val vm: VM) : PeriBase(VM.PERITYPE_SOUND) {
 
     private fun advanceEnvelope(voice: Voice, inst: TaudInst, tickSec: Double) {
         // Volume envelope
-        // sustain byte: bit7=enabled, bits[5:3]=end_idx, bits[2:0]=start_idx
+        // sustain byte: bit7=enable (u), bit6=sustain (t: 1=breaks on key-off,
-        val vSus      = inst.volEnvSustain
+        // 0=loops forever), bits[5:3]=end_idx, bits[2:0]=start_idx
-        val vSusOn    = (vSus and 0x80) != 0 && !voice.keyOff
+        val vSus       = inst.volEnvSustain
-        val vSusStart = vSus and 7
+        val vEnabled   = (vSus and 0x80) != 0
-        val vSusEnd   = (vSus ushr 3) and 7
+        val vIsSustain = (vSus and 0x40) != 0
+        // Loop is "active" when enabled AND (it's a forever-loop OR key not yet released)
+        val vSusOn     = vEnabled && (!vIsSustain || !voice.keyOff)
+        val vSusStart  = vSus and 7
+        val vSusEnd    = (vSus ushr 3) and 7
 
-        if (voice.envIndex >= 7) {
+        if (vSusOn && voice.envIndex == vSusEnd && vSusStart == vSusEnd) {
-            voice.envVolume = (inst.volEnvelopes[7].value / 63.0).coerceIn(0.0, 1.0)
-        } else if (vSusOn && voice.envIndex == vSusEnd && vSusStart == vSusEnd) {
             // slb == sle: hold at this node until key-off (no cycling)
             voice.envVolume = (inst.volEnvelopes[voice.envIndex].value / 63.0).coerceIn(0.0, 1.0)
+        } else if (vSusOn && voice.envIndex == vSusEnd) {
+            // At sustain-loop end: snap back to start regardless of stored offset.
+            voice.envTimeSec = 0.0
+            voice.envIndex = vSusStart
+            voice.envVolume = (inst.volEnvelopes[voice.envIndex].value / 63.0).coerceIn(0.0, 1.0)
+        } else if (voice.envIndex >= 11) {
+            voice.envVolume = (inst.volEnvelopes[11].value / 63.0).coerceIn(0.0, 1.0)
         } else {
             val vOffset = inst.volEnvelopes[voice.envIndex].offset.toDouble()
             if (vOffset == 0.0) {
@@ -1204,12 +1213,12 @@ class AudioAdapter(val vm: VM) : PeriBase(VM.PERITYPE_SOUND) {
                 if (voice.envTimeSec >= vOffset) {
                     voice.envTimeSec -= vOffset
                     val nextIdx = if (vSusOn && voice.envIndex == vSusEnd) vSusStart
-                                  else (voice.envIndex + 1).coerceAtMost(7)
+                                  else (voice.envIndex + 1).coerceAtMost(11)
                     voice.envIndex = nextIdx
                     voice.envVolume = (inst.volEnvelopes[voice.envIndex].value / 63.0).coerceIn(0.0, 1.0)
                 } else {
                     val cur = (inst.volEnvelopes[voice.envIndex].value / 63.0).coerceIn(0.0, 1.0)
-                    val nxt = (inst.volEnvelopes[(voice.envIndex + 1).coerceAtMost(7)].value / 63.0).coerceIn(0.0, 1.0)
+                    val nxt = (inst.volEnvelopes[(voice.envIndex + 1).coerceAtMost(11)].value / 63.0).coerceIn(0.0, 1.0)
                     voice.envVolume = cur + (nxt - cur) * (voice.envTimeSec / vOffset)
                 }
             }
@@ -1217,16 +1226,24 @@ class AudioAdapter(val vm: VM) : PeriBase(VM.PERITYPE_SOUND) {
 
         // Pan envelope (only when active for this instrument)
         if (!voice.hasPanEnv) return
-        val pSus      = inst.panEnvSustain
+        val pSus       = inst.panEnvSustain
-        val pSusOn    = (pSus and 0x80) != 0 && !voice.keyOff
+        val pEnabled   = (pSus and 0x80) != 0
-        val pSusStart = pSus and 7
+        val pIsSustain = (pSus and 0x40) != 0
-        val pSusEnd   = (pSus ushr 3) and 7
+        val pSusOn     = pEnabled && (!pIsSustain || !voice.keyOff)
+        val pSusStart  = pSus and 7
+        val pSusEnd    = (pSus ushr 3) and 7
 
-        if (voice.envPanIndex >= 7) {
+        if (pSusOn && voice.envPanIndex == pSusEnd && pSusStart == pSusEnd) {
-            voice.envPan = inst.panEnvelopes[7].value / 255.0
-        } else if (pSusOn && voice.envPanIndex == pSusEnd && pSusStart == pSusEnd) {
             // slb == sle: hold at this pan node until key-off
             voice.envPan = inst.panEnvelopes[voice.envPanIndex].value / 255.0
+        } else if (pSusOn && voice.envPanIndex == pSusEnd) {
+            // At sustain-loop end: snap back to start regardless of stored offset
+            // (encoder writes mf=0 on the last node by convention).
+            voice.envPanTimeSec = 0.0
+            voice.envPanIndex = pSusStart
+            voice.envPan = inst.panEnvelopes[voice.envPanIndex].value / 255.0
+        } else if (voice.envPanIndex >= 11) {
+            voice.envPan = inst.panEnvelopes[11].value / 255.0
         } else {
             val pOffset = inst.panEnvelopes[voice.envPanIndex].offset.toDouble()
             if (pOffset == 0.0) {
@@ -1236,12 +1253,12 @@ class AudioAdapter(val vm: VM) : PeriBase(VM.PERITYPE_SOUND) {
                 if (voice.envPanTimeSec >= pOffset) {
                     voice.envPanTimeSec -= pOffset
                     val nextIdx = if (pSusOn && voice.envPanIndex == pSusEnd) pSusStart
-                                  else (voice.envPanIndex + 1).coerceAtMost(7)
+                                  else (voice.envPanIndex + 1).coerceAtMost(11)
                     voice.envPanIndex = nextIdx
                     voice.envPan = inst.panEnvelopes[voice.envPanIndex].value / 255.0
                 } else {
                     val cur = inst.panEnvelopes[voice.envPanIndex].value / 255.0
-                    val nxt = inst.panEnvelopes[(voice.envPanIndex + 1).coerceAtMost(7)].value / 255.0
+                    val nxt = inst.panEnvelopes[(voice.envPanIndex + 1).coerceAtMost(11)].value / 255.0
                     voice.envPan = cur + (nxt - cur) * (voice.envPanTimeSec / pOffset)
                 }
             }
@@ -1816,8 +1833,10 @@ class AudioAdapter(val vm: VM) : PeriBase(VM.PERITYPE_SOUND) {
             val gvol = playhead.globalVolume / 255.0
             for (voice in ts.voices) {
                 if (!voice.active || voice.muted) continue
-                val s = fetchTrackerSample(voice, instruments[voice.instrumentId])
+                val voiceInst = instruments[voice.instrumentId]
-                val vol = voice.envVolume * voice.rowVolume / 63.0 * gvol * playhead.masterVolume / 255.0
+                val s = fetchTrackerSample(voice, voiceInst)
+                val instGv = voiceInst.instGlobalVolume / 255.0
+                val vol = voice.envVolume * voice.rowVolume / 63.0 * gvol * instGv * playhead.masterVolume / 255.0
                 val pan = if (voice.hasPanEnv) {
                     val envPanRaw = (voice.envPan * 255.0).roundToInt().coerceIn(0, 255)
                     (voice.channelPan + envPanRaw - 128).coerceIn(0, 255)
@@ -2314,14 +2333,15 @@ class AudioAdapter(val vm: VM) : PeriBase(VM.PERITYPE_SOUND) {
         var sampleLoopEnd: Int,
         // flags
         var loopMode: Int,
-        var volEnvSustain: Int,           // byte 13: 00 eee sss (0 = no sustain loop)
+        var volEnvSustain: Int,           // byte 13: ut eee sss (u=enable, t=sustain (1=breaks on key-off, 0=loops forever))
-        var panEnvSustain: Int,           // byte 14: 00 eee sss (0 = no sustain loop)
+        var panEnvSustain: Int,           // byte 14: ut eee sss (u=enable, t=sustain (1=breaks on key-off, 0=loops forever))
-        var volEnvelopes: Array<TaudInstEnvPoint>, // 8 points, value 0x00-0x3F
+        var instGlobalVolume: Int,        // byte 15: instrument global volume (0..255, 255 = unity)
-        var panEnvelopes: Array<TaudInstEnvPoint>  // 8 points, value 0x00-0xFF (0x80 = centre)
+        var volEnvelopes: Array<TaudInstEnvPoint>, // 12 points, value 0x00-0x3F
+        var panEnvelopes: Array<TaudInstEnvPoint>  // 12 points, value 0x00-0xFF (0x80 = centre)
     ) {
-        constructor(index: Int) : this(index, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+        constructor(index: Int) : this(index, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0xFF,
-            Array(8) { TaudInstEnvPoint(0x3F, ThreeFiveMiniUfloat(0)) },
+            Array(12) { TaudInstEnvPoint(0x3F, ThreeFiveMiniUfloat(0)) },
-            Array(8) { TaudInstEnvPoint(0x80, ThreeFiveMiniUfloat(0)) })
+            Array(12) { TaudInstEnvPoint(0x80, ThreeFiveMiniUfloat(0)) })
 
         // Funk repeat (S$Fx00) bit-mask — non-destructive XOR overlay across the loop region.
         // Lazily allocated; a 1-bit flips the byte, a 0-bit leaves it intact.
@@ -2361,12 +2381,11 @@ class AudioAdapter(val vm: VM) : PeriBase(VM.PERITYPE_SOUND) {
             12 -> (samplePtr.ushr(16).and(15).shl(4) or loopMode.and(3)).toByte()
             13 -> volEnvSustain.toByte()
             14 -> panEnvSustain.toByte()
-            15 -> 0
+            15 -> instGlobalVolume.toByte()
-            in 16..30 step 2 -> volEnvelopes[(offset - 16) / 2].value.toByte()
+            in 16..38 step 2 -> volEnvelopes[(offset - 16) / 2].value.toByte()
-            in 17..31 step 2 -> volEnvelopes[(offset - 17) / 2].offset.index.toByte()
+            in 17..39 step 2 -> volEnvelopes[(offset - 17) / 2].offset.index.toByte()
-            in 32..46 step 2 -> panEnvelopes[(offset - 32) / 2].value.toByte()
+            in 40..62 step 2 -> panEnvelopes[(offset - 40) / 2].value.toByte()
-            in 33..47 step 2 -> panEnvelopes[(offset - 33) / 2].offset.index.toByte()
+            in 41..63 step 2 -> panEnvelopes[(offset - 41) / 2].offset.index.toByte()
-            in 48..63 -> 0
             else -> throw InternalError("Bad offset $offset")
         }
 
@@ -2396,13 +2415,12 @@ class AudioAdapter(val vm: VM) : PeriBase(VM.PERITYPE_SOUND) {
             }
             13 -> { volEnvSustain = byte }
             14 -> { panEnvSustain = byte }
-            15 -> {}
+            15 -> { instGlobalVolume = byte and 0xFF }
 
-            in 16..30 step 2 -> volEnvelopes[(offset - 16) / 2].value = byte
+            in 16..38 step 2 -> volEnvelopes[(offset - 16) / 2].value = byte
-            in 17..31 step 2 -> volEnvelopes[(offset - 17) / 2].offset = ThreeFiveMiniUfloat(byte)
+            in 17..39 step 2 -> volEnvelopes[(offset - 17) / 2].offset = ThreeFiveMiniUfloat(byte)
-            in 32..46 step 2 -> panEnvelopes[(offset - 32) / 2].value = byte
+            in 40..62 step 2 -> panEnvelopes[(offset - 40) / 2].value = byte
-            in 33..47 step 2 -> panEnvelopes[(offset - 33) / 2].offset = ThreeFiveMiniUfloat(byte)
+            in 41..63 step 2 -> panEnvelopes[(offset - 41) / 2].offset = ThreeFiveMiniUfloat(byte)
-            in 48..63 -> {}
             else -> throw InternalError("Bad offset $offset")
         }
     }