diff --git a/README.md b/README.md
index fc1000a..b326b75 100644
--- a/README.md
+++ b/README.md
@@ -122,6 +122,12 @@ solve this problem.
 
 ## Changelog
 
+### V2.x (unreleased)
+- Split wave generation out of main file, reorganised content into header files.
+- Optimized some more code paths for Raspberry Casket replayer.
+- In the wave generator optimized away a table (32 words), replacement code is even smaller!
+- Added Presto player draft.
+
 ### V1.x (unreleased)
 - Fixed a bug regarding the copper output mode with looping waves having a loop-offset.
 - Fixed wrong register use on triggering waves regarding the loop offset.
diff --git a/binaries/raspberry_casket.bin b/binaries/raspberry_casket.bin
index 2f641c2..3b88e85 100644
Binary files a/binaries/raspberry_casket.bin and b/binaries/raspberry_casket.bin differ
diff --git a/binaries/readme.txt b/binaries/readme.txt
index 6ae6124..39bbd13 100644
--- a/binaries/readme.txt
+++ b/binaries/readme.txt
@@ -2,4 +2,4 @@ This file "raspberry_casket.bin" can replace the original "player.bin"
 provided that you used the 16 KB + 16 KB allocation for player
 variables as in the original player (you need only about 2 + 12 KB).
 
-It has been assembled from src/drop_in_replacement.asm (version V1.1).
\ No newline at end of file
+It has been assembled from src/drop_in_replacement.asm (version V2.0).
\ No newline at end of file
diff --git a/src/drop_in_replacement.asm b/src/drop_in_replacement.asm
index 92e5789..d849a2d 100644
--- a/src/drop_in_replacement.asm
+++ b/src/drop_in_replacement.asm
@@ -1,4 +1,4 @@
-; vasmm68k_mot.exe -devpac -Fbin -o ..\binaries\raspberry_casket.bin -opt-allbra drop_in_replacement.asm
+; vasmm68k_mot.exe -I..\includes -devpac -Fbin -o ..\binaries\raspberry_casket.bin -opt-allbra drop_in_replacement.asm
 
 PRETRACKER_SUPPORT_V1_5 = 0
 PRETRACKER_PARANOIA_MODE = 0
diff --git a/src/raspberry_casket.asm b/src/raspberry_casket.asm
index 2e72a22..6c03816 100755
--- a/src/raspberry_casket.asm
+++ b/src/raspberry_casket.asm
@@ -1,5 +1,5 @@
 ;--------------------------------------------------------------------
-; Raspberry Casket Player V1.1 (30-May-2023)
+; Raspberry Casket Player V2.x (26-Jul-2023)
 ; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 ;
 ; Provided by Chris 'platon42' Hodges <chrisly@platon42.de>
@@ -80,14 +80,14 @@
 ; Size
 ; ~~~~
 ; The original C compiled code was... just bad. The new binary is
-; about 1/3 of the original one.
+; about 1/3rd of the original one.
 ;
 ; The code has been also optimized in a way that it compresses better.
 ; The original code compressed with Blueberry's Shrinkler goes from
 ; 18052 bytes down to 9023 bytes.
 ;
 ; Raspberry Casket, depending on the features compiled in, is about
-; 6216 bytes and goes down to ~4348 bytes (in isolation).
+; 5992 bytes and goes down to ~4238 bytes (in isolation).
 ;
 ; So this means that the optimization is not just "on the outside".
 ;
@@ -254,398 +254,22 @@ PRETRACKER_DONT_TRASH_REGS = 1
 PRETRACKER_COPPER_OUTPUT = 0 ; 0 = standard CPU wait, 1 = Copperlist
         ENDC
 
-; Pretracker song format description:
-;
-; $0000  4: PRT<version> ($19 (V0.x), $1b (V1.0), $1e (V1.5))
-; $0004  4: File offset to position data (POSD)
-; $0008  4: File offset to pattern data (PATT)
-; $000C  4: File offset to instruments (INST)
-; $0010  4: File offset to waves (WAVE)
-; $0014 20: Songname
-; $0028 20: Author
-; $003C  1: Restart position for song (<=V1.0)
-; $003D  1: Number of patterns (<=V1.0)
-; $003E  1: Songlength in patterns (<=V1.0)
-; $003F  1: Number of steps per pattern (<=V1.0)
-; $0040  1: Number of instruments (<=V1.0), $40 for V1.5
-; $0041  1: Number of waves
-; $0042 24: Wave generation ordering (>=V1.0)
-; $005a  1: Number of subsongs (>=V1.5)
-; [...]
-; Position data (POSD):
-; - Per subsong (>=V1.5)
-;   - 1: Restart pos
-;   - 1: #patterns
-;   - 1: #numsteps
-;   - 1: songlength
-;   - 4: relative pattern offset in pattern data
-; - Positions x 4 x [Pattern number (byte), pitch shift (signed byte)]
-;
-; Pattern data (PATT):
-; Each pattern line consists of three bytes:
-;   - 1 Bit : Bit 4 of Inst Number
-;   - 1 Bit : ARP instead of effect
-;   - 6 Bits: Pitch ($01 is C-0, $3d is NOTE OFF)
-;
-;   - 4 Bits: Bit 0-3 of Inst Number
-;   - 4 Bits: Effect command
-;   - 8 Bits: Effect data
-; - Patterns x steps x 3 bytes
-;
-; Unknown data after pattern data: (12 10 00 00 00 00)
-;
-; Instrument definitions (INST):
-; - 32 (!) x Null terminated string (or 23 chars max) (for V1.5 this is 64 strings)
-; - For each instrument: Instrument Info (ININ)
-;   - $00  1: Vibrato Delay
-;   - $01  1: Vibrato Depth
-;   - $02  1: Vibrato Speed (-1)
-;   - $03  1: ADSR Attack
-;   - $04  1: ADSR Decay
-;   - $05  1: ADSR Sustain
-;   - $06  1: ADSR Release
-;   - $07  1: Number of Inst Pattern steps
-;
-; - For each instrument:
-;   - 1 Byte: number of steps
-;   - 3 Bytes x number of steps: Inst Pattern (IPTT)
-;
-; Inst pattern data (IPTT):
-; Each pattern line consists of three bytes:
-;   - 1 Bit  : Next note stitched to this one
-;   - 1 Bit  : Fixed Pitch Note
-;   - 6 Bits : Pitch ($01 is C-0)
-;   - 4 Bits : unused?
-;   - 12 Bits: Effect
-; - Patterns x steps x 3 bytes
-;
-; Wave definitions (WAVE):
-; - 24 (!) x Null terminated string (or 23 chars max)
-; - Optional padding to even address, if necessary
-; - For each wave:
-;   - 42 Bytes: Wave info structure (see definition below)
-
-; ----------------------------------------
-; Some constants for clarity
-
-MAX_VOLUME      = $40
-MAX_SPEED       = $2f
-MAX_WAVES       = 24
-MAX_INSTRUMENTS = 32
-MAX_TRACK_DELAY = 32
-NOTE_OFF_PITCH  = $3d
-NOTES_IN_OCTAVE = 12
-NUM_CHANNELS    = 4     ; yes, you can reduce the number of channels if you want
-
-; ----------------------------------------
-; Pretracker file structures
-
-; Pattern data (PATT and IPTT)
-pdb_pitch_ctrl      = 0
-pdb_inst_effect     = 1 ; for normal pattern data
-pdb_effect_cmd      = 1 ; for inst pattern
-pdb_effect_data     = 2
-
-; Pattern pos data (POSD)
-ppd_pat_num         = 0
-ppd_pat_shift       = 1
-
-; Instrument Info (ININ)
-ii_vibrato_delay    = 0
-ii_vibrato_depth    = 1
-ii_vibrato_speed    = 2
-ii_adsr_attack      = 3
-ii_adsr_decay       = 4
-ii_adsr_sustain     = 5
-ii_adsr_release     = 6
-ii_pattern_steps    = 7
-ii_SIZEOF           = 8
-
-; Wave Info (WAVE)
-wi_loop_start_w    = $00
-wi_loop_end_w      = $02
-wi_subloop_len_w   = $04
-wi_allow_9xx_b     = $06
-wi_subloop_wait_b  = $07
-wi_subloop_step_w  = $08
-wi_chipram_w       = $0a
-wi_loop_offset_w   = $0c
-wi_chord_note1_b   = $0e
-wi_chord_note2_b   = $0f
-wi_chord_note3_b   = $10
-wi_chord_shift_b   = $11
-wi_osc_unknown_b   = $12 ; always $00? (unused in code?)
-wi_osc_phase_spd_b = $13
-wi_flags_b         = $14 ; bit 0/1: osc type, bit 2: needs extra octaves, bit 3: boost, bit 4: pitch linear, bit 5: vol fast
-wi_osc_phase_min_b = $15
-wi_osc_phase_max_b = $16
-wi_osc_basenote_b  = $17
-wi_osc_gain_b      = $18
-wi_sam_len_b       = $19 ; in multiples of 128, zero-based (0 == 128)
-wi_mix_wave_b      = $1a
-wi_vol_attack_b    = $1b
-wi_vol_delay_b     = $1c
-wi_vol_decay_b     = $1d
-wi_vol_sustain_b   = $1e
-wi_flt_type_b      = $1f ; 1=lowpass, 2=highpass, 3=bandpass, 4=notch
-wi_flt_resonance_b = $20
-wi_pitch_ramp_b    = $21
-wi_flt_start_b     = $22
-wi_flt_min_b       = $23
-wi_flt_max_b       = $24
-wi_flt_speed_b     = $25
-wi_mod_params_l    = $26
-wi_mod_wetness_b   = $26
-wi_mod_length_b    = $27
-wi_mod_predelay_b  = $28
-wi_mod_density_b   = $29 ; (1-7), unisono (bits 3/4) and post bit 5
-wi_SIZEOF          = $2a
-
-; ----------------------------------------
-; Unpacked Instrument Info (addition to player for faster processing)
-                    rsreset
-uii_vibrato_delay   rs.w    1
-uii_vibrato_depth   rs.w    1
-uii_vibrato_speed   rs.w    1
-uii_adsr_release    rs.b    1
-                    rs.b    1 ; dummy
-uii_adsr_attack     rs.w    1
-uii_adsr_decay      rs.w    1
-uii_adsr_sustain    rs.w    1
-uii_pattern_steps   rs.b    1
-                    rs.b    1 ; padding
-uii_SIZEOF          rs.b    0
-
-; ----------------------------------------
-; MySong offsets
-                        rsreset
-sv_waveinfo_table       rs.l    MAX_WAVES ; 24 pointers to wave infos to avoid mulu
-sv_inst_patterns_table  rs.l    MAX_INSTRUMENTS ; 32 pointers to pattern data
-                        ; --- 127 byte displacement limit ---
-sv_wavelength_table     rs.l    MAX_WAVES ; 24 longwords to sample lengths (standard octave) (NEW)
-sv_wavetotal_table      rs.l    MAX_WAVES ; 24 longwords to sample lengths for all octaves (NEW)
-sv_wavegen_order_table  rs.b    MAX_WAVES ; 24 bytes
-sv_num_waves_b          rs.b    1
-sv_num_steps_b          rs.b    1
-sv_patterns_ptr         rs.l    1
-sv_curr_pat_pos_w       rs.w    1 ; only byte used FIXME why is this part of MySong? Should be in Player
-sv_pat_pos_len_w        rs.w    1 ; only byte used
-sv_pat_restart_pos_w    rs.w    1 ; only byte used
-sv_pos_data_adr         rs.l    1
-sv_waveinfo_ptr         rs.l    1 ; base pointer of wave info
-sv_pattern_table        rs.l    256
-sv_inst_infos_table     rs.b    MAX_INSTRUMENTS*uii_SIZEOF
-sv_SIZEOF               rs.b    0
-
-; ----------------------------------------
-; channel output data (part of pcd structure below)
-                rsreset
-ocd_sam_ptr     rs.l    1   ; 0
-ocd_length      rs.w    1   ; 4
-ocd_loop_offset rs.w    1   ; 6
-ocd_period      rs.w    1   ; 8
-ocd_volume      rs.b    1   ; 10
-ocd_trigger     rs.b    1   ; 11 needs to be after volume
-ocd_unused      rs.l    1   ; 12 unused, but makes the structure an even 16 bytes
-ocd_SIZEOF      rs.b    0
-
-; channel structure (part of pv structure)
-                            rsreset
-; DO NOT CHANGE ORDER -- OPTIMIZED CLEARING
-pcd_pat_portamento_dest_w   rs.w    1 ; portamento destination pitch
-pcd_pat_pitch_slide_w       rs.w    1
-
-pcd_pat_vol_ramp_speed_b    rs.b    1
-pcd_pat_2nd_inst_num4_b     rs.b    1
-pcd_pat_2nd_inst_delay_b    rs.b    1
-pcd_wave_offset_b           rs.b    1
-
-pcd_inst_pitch_slide_w      rs.w    1
-pcd_inst_sel_arp_note_w     rs.w    1
-
-pcd_inst_note_pitch_w       rs.w    1
-pcd_inst_curr_port_pitch_w  rs.w    1
-
-pcd_inst_line_ticks_b       rs.b    1
-pcd_inst_pitch_pinned_b     rs.b    1
-pcd_inst_vol_slide_b        rs.b    1
-pcd_inst_step_pos_b         rs.b    1
-
-pcd_inst_wave_num_w         rs.w    1 ; current wave number (1 based) (lower byte used)
-
-pcd_track_delay_offset_b    rs.b    1 ; $ff = no track delay
-pcd_inst_speed_stop_b       rs.b    1 ; speed byte, $ff stops processing
-pcd_inst_pitch_w            rs.w    1
-
-pcd_inst_vol_w              rs.w    1
-pcd_loaded_inst_vol_b       rs.b    1
-pcd_pat_vol_b               rs.b    1 ; Multiplied with volume of instrument.
-; DO NOT CHANGE ORDER -- OPTIMIZED CLEARING END
-
-pcd_arp_notes_l             rs.b    0
-pcd_arp_note_1_b            rs.b    1
-pcd_arp_note_2_b            rs.b    1
-pcd_arp_note_3_b            rs.b    1
-                            rs.b    1 ; gets cleared
-
-pcd_last_trigger_pos_w      rs.w    1 ; I think this makes sure that we don't retrigger the note on same pos
-
-pcd_pat_portamento_speed_b  rs.b    1
-pcd_pat_adsr_rel_delay_b    rs.b    1 ; counts down until adsr release. Seems unused?
-pcd_note_off_delay_b        rs.b    1 ; time before note is released ($ff = disabled)
-pcd_inst_pattern_steps_b    rs.b    1 ; number of steps in instrument pattern
-
-pcd_note_delay_b            rs.b    1 ; $ff = no note delay
-pcd_track_delay_steps_b     rs.b    1 ; $00 = no track delay, $ff = stop track delay (this is for the next channel!)
-pcd_track_delay_vol16_b     rs.b    1
-pcd_track_init_delay_b      rs.b    1 ; number of frames to ignore the delay
-
-pcd_inst_num4_w             rs.w    1 ; current instrument number * 4
-;pcd_inst_new_step_w         rs.w    1 ; seems to be unused
-pcd_inst_subloop_wait_w     rs.w    1
-pcd_inst_loop_offset_w      rs.w    1
-pcd_inst_info_ptr           rs.l    1 ; pointer to currently active instrument
-
-pcd_waveinfo_ptr            rs.l    1 ; pointer to currently active waveinfo
-pcd_channel_mask_b          rs.b    1
-pcd_channel_num_b           rs.b    1
-pcd_adsr_phase_w            rs.w    1 ; 0=attack, 1=decay, 2=sustain, 3=release ! do not change order
-pcd_adsr_volume_w           rs.w    1 ; 0 for restart / $400 (word only) ! do not change order
-pcd_adsr_phase_speed_b      rs.b    1
-pcd_inst_ping_pong_dir_b    rs.b    1 ; direction of ping-pong (-1 == $00 / +1 = $ff)
-pcd_adsr_pos_w              rs.w    1 ; pos in adsr curve
-pcd_adsr_vol64_w            rs.w    1 ; some adsr volume
-
-pcd_new_inst_num_b          rs.b    1 ; load new instrument (number) ! do not change order
-                            rs.b    1 ; gets cleared
-pcd_vibrato_pos_w           rs.w    1 ;
-pcd_vibrato_delay_w         rs.w    1 ; is a byte value ! do not change order
-pcd_vibrato_depth_w         rs.w    1 ; is a byte value ! do not change order
-pcd_vibrato_speed_w         rs.w    1 ; is a byte value ! do not change order
-pcd_adsr_release_b          rs.b    1 ; is a byte value ! do not change order
-                            rs.b    1 ; padding will be overwritten!
-
-pcd_out_base                rs.b    ocd_SIZEOF
-pcd_track_delay_buffer      rs.b    MAX_TRACK_DELAY*ocd_SIZEOF
-pcd_SIZEOF                  rs.b    0
-
-pcd_out_ptr_l       = pcd_out_base+ocd_sam_ptr
-pcd_out_len_w       = pcd_out_base+ocd_length
-pcd_out_lof_w       = pcd_out_base+ocd_loop_offset
-pcd_out_per_w       = pcd_out_base+ocd_period
-pcd_out_vol_b       = pcd_out_base+ocd_volume
-pcd_out_trg_b       = pcd_out_base+ocd_trigger
-pcd_out_unused_l    = pcd_out_base+ocd_unused          ; copied for track delay, but not used?
-
-                rsreset
-owb_saw_waves   rs.b    128
-owb_sqr_waves   rs.b    128
-owb_tri_waves   rs.b    128
-owb_wave_length rs.b    1
-owb_SIZEOF      rs.b    0
-
-; ----------------------------------------
-; MyPlayer global variables (not bound to channel)
-                        rsreset
-; DO NOT CHANGE ORDER -- OPTIMIZED INIT
-pv_pat_curr_row_b       rs.b    1 ; current step
-pv_next_pat_row_b       rs.b    1
-pv_next_pat_pos_b       rs.b    1
-pv_pat_speed_even_b     rs.b    1 ; even shuffle speed
-
-pv_pat_speed_odd_b      rs.b    1 ; odd shuffle speed
-pv_pat_line_ticks_b     rs.b    1
-pv_pat_stopped_b        rs.b    1 ; 0 = stop, $ff = run
-pv_songend_detected_b   rs.b    1
-; DO NOT CHANGE ORDER -- OPTIMIZED INIT END
-
-pv_loop_pattern_b       rs.b    1 ; repeat current pattern, do not advance
-                        rs.b    1 ; padding
-
-pv_trigger_mask_w       rs.w    1
-
-pv_my_song              rs.l    1
-pv_sample_buffer_ptr    rs.l    1 ; pointer to start of sample buffer
-pv_copperlist_ptr       rs.l    1
-pv_wave_sample_table    rs.l    MAX_WAVES ; 24 pointers to sample starts
-pv_period_table         rs.w    16*NOTES_IN_OCTAVE*3
-                        ; --- 127 byte displacement limit ---
-pv_channeldata          rs.b    NUM_CHANNELS*pcd_SIZEOF
-
-                        IFNE    PRETRACKER_VOLUME_TABLE
-pv_osc_buffers          rs.b    0 ; reuse space of volume_table, which is bigger than NOTES_IN_OCTAVE*owb_SIZEOF
-pv_volume_table         rs.b    (MAX_VOLUME+1)*MAX_VOLUME*2
-                        ELSE
-pv_osc_buffers          rs.b    NOTES_IN_OCTAVE*owb_SIZEOF
-                        ENDC
-
-pv_precalc_sample_size  rs.l    1
-pv_precalc_progress_ptr rs.l    1
-pv_wg_wave_ord_num_w    rs.w    1
-                        IFNE    PRETRACKER_PARANOIA_MODE    ; same wave for mixing cannot be selected in Pretracker
-pv_wg_curr_wave_num_b   rs.b    1
-                        rs.b    1
-                        ENDC
-pv_wg_curr_sample_ptr   rs.l    1
-pv_wg_curr_samend_ptr   rs.l    1
-pv_wg_curr_sample_len_w rs.w    1
-pv_wg_chord_note_num_b  rs.b    1 ; don't change order
-pv_wg_unisono_run_b     rs.b    1 ; don't change order
-pv_wg_chord_flag_w      rs.w    1
-pv_wg_chord_pitches     rs.l    1
-pv_wg_osc_speed_l       rs.l    1
-pv_wg_flt_taps          rs.w    4
-pv_SIZEOF               rs.b    0
-
 ;--------------------------------------------------------------------
 
-        include "../includes/hardware/custom.i"
-        include "../includes/hardware/dmabits.i"
-
-CLIPTO8BIT MACRO
-        cmpi.w  #-$80,\1
-        bge.s   .nominclip\@
-        moveq.l #-$80,\1
-.nominclip\@
-        cmpi.w  #$7F,\1
-        ble.s   .nomaxclip\@
-        moveq.l #$7F,\1
-.nomaxclip\@
-        ENDM
-
-CLIPORTRUNC8BIT MACRO
-        beq.s   .unboosted\@
-        asr.l    #6,\1
-        cmpi.w  #-$80,\1
-        bge.s   .nominclip\@
-        moveq.l #-$80,\1
-.nominclip\@
-        cmpi.w  #$7F,\1
-        ble.s   .nomaxclip\@
-        moveq.l #$7F,\1
-        bra.s   .nomaxclip\@
-.unboosted\@
-        asr.l   #8,\1
-.nomaxclip\@
-        ENDM
-
-CLIPTO8BITAFTERADD MACRO
-        bvc.s   .noclip\@
-        spl     \1
-        eor.b   #$7f,\1
-.noclip\@
-        ENDM
+        include "raspberry_casket.i"
+        include "hardware/custom.i"
+        include "hardware/dmabits.i"
 
 ;--------------------------------------------------------------------
 ; Code starts here
 
         IFNE    PRETRACKER_KEEP_JUMP_TABLE
-pre_FuncTable
-        dc.l     pre_SongInit-pre_FuncTable
-        dc.l     pre_PlayerInit-pre_FuncTable
-        dc.l     pre_PlayerTick-pre_FuncTable
+pre_FuncTable:
+        dc.l    pre_SongInit-pre_FuncTable
+        dc.l    pre_PlayerInit-pre_FuncTable
+        dc.l    pre_PlayerTick-pre_FuncTable
         ENDC
-        ;dc.b     '$VER: Raspberry Casket 1.1',0
+        ;dc.b    '$VER: Raspberry Casket 1.1',0
         ;even
 
         IFNE    PRETRACKER_COPPER_OUTPUT
@@ -727,8 +351,8 @@ pre_SongInit:
         moveq.l #0,d0
         move.l  $0000(a2),d1
         move.b  d1,d2
-        clr.b   d1
-        cmpi.l  #$50525400,d1                   ; "PRE"-Text
+        move.b  d0,d1
+        cmp.l   #$50525400,d1                   ; "PRE"-Text
         bne     .error
         moveq.l #MAX_INSTRUMENTS-1,d7           ; notice there's one extra name (available in 1.5, but not usable)!
         IFNE    PRETRACKER_SUPPORT_V1_5
@@ -742,7 +366,7 @@ pre_SongInit:
         moveq.l #2*MAX_INSTRUMENTS-1,d7         ; v1.5 has 32 slots (the other ones used for sfx)
 .nopatchv15
         ELSE
-        cmpi.b  #$1b,d2
+        cmp.b   #$1b,d2
         bgt     .error
         ENDC
 
@@ -933,6 +557,8 @@ pre_PlayerInit:
         move.w  #pv_SIZEOF,d0
         bsr     pre_MemClr     ; keeps a1 unchanged!
 
+        move.l  a6,pv_my_song(a4)
+
 ; ----------------------------------------
 ; proposed register assignment:
 ; a0 = sample output / scratch
@@ -941,7 +567,6 @@ pre_PlayerInit:
 ; a4 = MyPlayer
 ; a6 = MySong
 
-        move.l  a2,pv_my_song(a4)
         IFNE    PRETRACKER_PROGRESS_SUPPORT
         move.l  a3,pv_precalc_progress_ptr(a4)
         ENDC
@@ -951,9 +576,10 @@ pre_PlayerInit:
         beq.s   .hasnosamplebuffer  ; PARANOIA
         ENDC
 
-        clr.w   (a1)+               ; empty sample
         moveq.l #0,d7
+        move.w  d7,(a1)+               ; empty sample
         move.b  sv_num_waves_b(a6),d7
+        move.w  d7,pv_wg_wave_counter_w(a4)
         IFNE    PRETRACKER_PARANOIA_MODE
         beq.s   .hasnosamplebuffer  ; PARANOIA
         ENDC
@@ -1022,7 +648,9 @@ pre_PlayerInit:
 .chaninitloop2
         move.b  #MAX_VOLUME,pcd_pat_vol_b(a0)
         st      pcd_track_delay_offset_b(a0)
-        move.l  a1,pcd_waveinfo_ptr(a0)
+        IFEQ    PRETRACKER_BUGFIX_CODE
+        move.l  a1,pcd_waveinfo_ptr(a0)             ; we should actually have no wave selected
+        ENDC
         move.w  #3,pcd_adsr_phase_w(a0)
 
         move.l  pv_sample_buffer_ptr(a4),pcd_out_ptr_l(a0)
@@ -1034,1137 +662,20 @@ pre_PlayerInit:
         lea     pcd_SIZEOF(a0),a0
         dbra    d7,.chaninitloop2
 
-; ----------------------------------------
-
-        lea     pre_log12_table(pc),a0   ; 128, 121, 114, 107, 102, 96, 90, 85, 80, 76, 72, 67
-        lea     pv_osc_buffers+owb_sqr_waves(a4),a1
-        moveq.l #NOTES_IN_OCTAVE-1,d7
-.noteloop
-        moveq.l #0,d6
-        move.l  d6,a3               ; tabpos
-        move.b  (a0)+,d6            ; period
-        move.b  d6,owb_wave_length-owb_sqr_waves(a1)
-
-        move.l  #$ff00,d5
-        divu    d6,d5               ; frac increment
-
-        move.w  d6,d4
-        lsr.w   #1,d4               ; half-period
-        move.w  d4,d3
-        lsr.w   #1,d3               ; quarter-period
-
-        moveq.l #0,d0               ; acc
-        lea     (a1,d6.w),a2
-        lea     owb_tri_waves-owb_sqr_waves(a2),a2
-        suba.w  d3,a2
-.notewaveloop
-        move.w  d0,d2
-        lsr.w   #8,d2
-
-        moveq.l #$7f,d1
-        sub.b   d2,d1
-        move.b  d1,owb_saw_waves-owb_sqr_waves(a1,a3.w)
-
-        add.b   d2,d2
-        cmp.w   a3,d3               ; tabpos == negquarter
-        bne.s   .nowrapback
-        suba.w  d6,a2               ; go back to start of period
-.nowrapback
-        cmp.w   d0,a3
-        ble.s   .otherhalf
-        moveq.l #$7f,d1
-        sub.b   d2,d1
-        move.b  d1,(a2)+
-        bra.s   .clip80
-
-.otherhalf
-        add.b   #$80,d2
-        move.b  d2,(a2)+
-        moveq.l #$7f,d2
-        cmp.w   a3,d4
-        bne.s   .noclip80
-.clip80 moveq.l #-$80,d2
-.noclip80
-        move.b  d2,owb_sqr_waves-owb_sqr_waves(a1,a3.w)
-
-        add.w   d5,d0               ; increment acc by frac
-        addq.w  #1,a3               ; increment pos
-
-        cmp.w   a3,d6
-        bne.s   .notewaveloop
-
-        lea     owb_SIZEOF(a1),a1
-        dbra    d7,.noteloop
-
 ; ----------------------------------------
 
 .audioinit
+        IFND    PRESTO_SKIP_OUTPUT
         bset    #1,$BFE001          ; filter off
+        ENDC
 
         IFNE    PRETRACKER_PARANOIA_MODE
         tst.b   sv_num_waves_b(a6)
         beq     .earlyexit          ; PARANOIA
         ENDC
 
-; ----------------------------------------
-; proposed register assignment:
-; a0 = sample output
-; a1 = scratch
-; a3 = songinfo
-; a4 = MyPlayer
-; a6 = MySong / waveinfo
-
-.wavegenloop
-        ;movea.l  pv_my_song(a4),a6
         lea     sv_wavegen_order_table(a6),a1
-        move.w  pv_wg_wave_ord_num_w(a4),d0         ; counter through all 24 waves
-        moveq.l #0,d1
-        move.b  (a1,d0.w),d1        ; apply wave order redirection
-        IFNE    PRETRACKER_PARANOIA_MODE    ; same wave for mixing cannot be selected in Pretracker
-        move.b  d1,pv_wg_curr_wave_num_b(a4)
-        ENDC
-        lsl.w   #2,d1
-        move.l  pv_wave_sample_table(a4,d1.w),a0
-        move.l  a0,pv_wg_curr_sample_ptr(a4)
-
-        add.w   #sv_wavelength_table,d1
-        adda.w  d1,a6
-        move.l  sv_wavelength_table-sv_wavelength_table(a6),d0
-        move.w  d0,pv_wg_curr_sample_len_w(a4)
-        IFNE    PRETRACKER_PROGRESS_SUPPORT
-        move.l  sv_wavetotal_table-sv_wavelength_table(a6),pv_precalc_sample_size(a4)
-        ENDC
-        move.l  sv_waveinfo_table-sv_wavelength_table(a6),a3
-
-        ; clear sample data (a0 and d0 from above)
-        bsr     pre_MemClr
-        move.l  a0,pv_wg_curr_samend_ptr(a4)
-
-        ; read out chord information
-        lea     wi_chord_note1_b(a3),a1
-        move.b  (a1)+,d1
-        move.b  (a1)+,d2
-        move.b  (a1)+,d3
-
-        moveq.l #0,d4
-        move.b  d1,d4
-        or.b    d2,d4
-        or.b    d3,d4
-        seq     d4
-        neg.b   d4
-        move.w  d4,pv_wg_chord_flag_w(a4)   ; has chord flag (0/1)
-
-        move.b  wi_osc_basenote_b(a3),d0
-        add.b   d0,d1
-        add.b   d0,d2
-        add.b   d0,d3
-        lea     pv_wg_chord_pitches(a4),a1
-        move.b  d0,(a1)+
-        move.b  d1,(a1)+
-        move.b  d2,(a1)+
-        move.b  d3,(a1)+
-
-        suba.l  a5,a5
-        clr.w   pv_wg_chord_note_num_b(a4)  ; and pv_wg_chord_note_num_b
-
-.wavegen_chordloop
-        lea     pv_wg_chord_pitches(a4),a1
-        moveq.l #0,d1
-        move.b  pv_wg_chord_note_num_b(a4),d1   ; chord note counter
-        move.b  (a1,d1.w),d0    ; get chord note
-        ext.w   d0
-
-        tst.w   d1
-        beq.s   .base_note_is_never_skipped ; is base note?
-        cmp.b   wi_osc_basenote_b(a3),d0
-        beq     .wave_gen_tone_done      ; skip chord notes that are same as base note
-.base_note_is_never_skipped
-        moveq.l #0,d1
-        moveq.l #NOTES_IN_OCTAVE,d2
-        move.w  d0,d1
-        move.w  d0,a2           ; save base note, used later (much later, noise generator)!
-        add.w   #NOTES_IN_OCTAVE*NOTES_IN_OCTAVE,d1 ; make sure we don't run into negative modulo
-        divu    d2,d1
-        swap    d1
-        move.w  d1,d0           ; note within octave
-        swap    d1
-        sub.w   d2,d1           ; restore octave
-
-        mulu    #owb_SIZEOF,d0
-        lea     (a4,d0.w),a1
-
-        lea     pv_osc_buffers+owb_saw_waves(a1),a6
-        moveq.l #3,d0
-        and.b   wi_flags_b(a3),d0
-        beq.s   .osc_selected
-        lea     owb_tri_waves-owb_saw_waves(a6),a6
-        subq.b  #1,d0
-        beq.s   .osc_selected
-        lea     owb_sqr_waves-owb_tri_waves(a6),a6
-        subq.b  #1,d0
-        beq.s   .osc_selected
-        suba.l  a6,a6           ; noise selected
-.osc_selected
-
-        move.l  #$8000,d6
-        move.w  d1,d0           ; check octave shift
-        bgt.s   .shiftleft
-        beq.s   .contshift
-;.shiftright
-        move.w  d0,d3
-        neg.w   d3
-        asr.l   d3,d6
-        bra.s   .contshift
-.shiftleft
-        lsl.l   d0,d6
-.contshift
-
-; ----------------------------------------
-; pitch ramp
-        move.b  wi_pitch_ramp_b(a3),d3
-        ext.w   d3
-        ext.l   d3
-        btst    #4,wi_flags_b(a3)   ; pitch linear flag
-        beq.s   .pitch_not_linear
-        tst.b   d3
-        bgt.s   .pitch_ramp_positive
-
-        lsl.l   d0,d3
-        add.l   d3,d3
-        bra.s   .pitch_ramp_cont
-
-.pitch_not_linear
-        tst.b   d3
-        ble.s   .pitch_ramp_cont
-.pitch_ramp_positive
-        muls    d3,d3
-.pitch_ramp_cont
-        move.l  d3,d2
-        lsl.l   #8,d2
-        lsl.l   #2,d2
-
-        moveq.l #0,d7
-        move.b  pv_osc_buffers+owb_wave_length(a1),d7   ; get period
-        moveq.l #15,d5
-        lsl.l   d5,d7
-
-        sub.w   d0,d5       ; 15-octave
-        lsl.w   #3,d5
-
-        moveq.l #0,d3
-        move.b  wi_osc_phase_min_b(a3),d3
-        mulu    d5,d3
-        lsl.l   #6,d3
-
-        moveq.l #0,d0
-        move.b  wi_osc_phase_max_b(a3),d0
-        mulu    d5,d0
-        lsl.l   #6,d0
-        move.l  d0,a5
-
-        moveq.l #0,d5
-        move.b  wi_osc_phase_spd_b(a3),d5
-        lsl.l   #8,d5
-        lsl.l   #3,d5
-
-        cmp.l   d3,d0
-        bge.s   .osc_with_positive_phase_speed
-        neg.l   d5
-
-        movea.l d3,a5
-        bra.s   .osc_continue
-.osc_with_positive_phase_speed
-        move.l  d3,d0
-
-.osc_continue
-        move.l  d0,pv_wg_osc_speed_l(a4)
-
-        ; d0 = d6 * chord_shift * chordnum + d6 * phase_min = d6 * (chord_shift * chordnum + phase_min)
-
-        moveq.l #0,d4
-        move.b  wi_chord_shift_b(a3),d4
-        move.w  pv_wg_chord_flag_w(a4),d0
-        add.b   pv_wg_chord_note_num_b(a4),d0
-        mulu    d0,d4
-        moveq.l #0,d0
-        move.b  wi_osc_phase_min_b(a3),d0
-        add.w   d0,d4
-        move.l  d6,d0
-        lsr.l   #4,d0
-        lsl.l   #4,d4
-        mulu    d4,d0
-
-        cmp.l   d7,d0
-        ble.s   .lbC002516
-.lbC002510
-        sub.l   d7,d0
-        cmp.l   d7,d0
-        bgt.s   .lbC002510
-.lbC002516
-        move.l  a6,d4
-        bne     .no_noise
-
-; ----------------------------------------
-; d1 = octave
-; d5 = osc phase speed
-; a2 = base note
-.gen_noise
-        IFNE    PRETRACKER_PARANOIA_MODE
-        tst.w   pv_wg_curr_sample_len_w(a4)
-        beq     .wave_gen_tone_done
-        ENDC
-
-        move.l  #$8000,d5
-        move.l  d5,a5
-        tst.w   d1
-        bge.s   .gen_noise_positive_octave
-
-.gen_noise_negative_octave
-        moveq.l #NOTES_IN_OCTAVE,d1
-        move.l  a2,d3
-        neg.l   d3
-        move.l  d3,d0
-        divu    d1,d0
-        IFNE    PRETRACKER_PARANOIA_MODE
-        bvs.s   .divisionoverflow
-        ENDC
-        addq.w  #1,d0
-        lsr.w   d0,d5
-        moveq.l #0,d0
-        move.w  d5,d0
-        divu    d1,d0
-        moveq.l #0,d4
-        move.w  d0,d4
-        IFNE    PRETRACKER_PARANOIA_MODE
-        bra.s   .returnfromoverflow
-.divisionoverflow
-        move.l  #$AAA,d4        ; some dummy value, I would expect
-.returnfromoverflow
-        ENDC
-        moveq.l #0,d0
-        move.w  d3,d0
-.cheap_mod12
-        sub.w   d1,d0
-        bpl.s   .cheap_mod12
-        neg.w   d0
-        mulu    d4,d0
-        add.l   d0,d5
-
-.gen_noise_positive_octave
-        moveq.l #0,d0
-        move.b  wi_osc_phase_min_b(a3),d0
-        moveq.l #0,d1
-        move.b  wi_chord_shift_b(a3),d1
-        add.w   d1,d0
-        addq.w  #1,d0
-
-        movea.l d5,a1
-        movea.l pv_wg_curr_sample_ptr(a4),a0
-        moveq.l #0,d1
-        moveq.l #0,d6
-        moveq.l #0,d3
-        move.b  wi_osc_gain_b(a3),d3
-
-.gen_noise_outerloop
-        move.w  d0,d1           ; random noise generator
-        lsl.w   #8,d1
-        lsl.w   #5,d1
-        eor.w   d1,d0
-
-        move.w  d0,d1
-        lsr.w   #8,d1
-        lsr.w   #1,d1
-        eor.w   d1,d0
-
-        move.w  d0,d1
-        lsl.w   #7,d1
-        eor.w   d1,d0
-
-        cmpa.l  #$8000,a5       ; if symmetrical
-        beq.s   .gen_noise_centered
-        ; FIXME what does this do? d4 = (a5 - $8000) d1 = (a5 + $7fff)&$ffff8000 -> d4 - d1 == ((a5 - $8000) - ((a5 + $7fff)&$ffff8000)
-        move.l  a5,d4
-        addi.l  #$FFFF8000,d4
-        move.l  a5,d1
-        addi.l  #$FFFF7FFF,d1
-        andi.w  #$8000,d1
-        movea.l d4,a5
-        suba.l  d1,a5
-.gen_noise_centered
-
-        move.b  d0,d1
-        ext.w   d1
-        move.w  d3,d4
-        muls    d1,d4
-        move.l  d4,d1
-        asr.l   #7,d1
-        CLIPTO8BIT d1
-        add.b   (a0),d1
-
-.gen_noise_innerloop
-        move.b  d1,(a0)+
-        adda.l  a1,a5
-        tst.l   d2
-        beq.s   .lbC002BAC
-        add.l   d2,d6
-        move.l  d6,d4
-        asr.l   #8,d4
-        asr.l   #2,d4
-        add.l   d5,d4
-        movea.l d4,a1
-
-        btst    #4,wi_flags_b(a3)   ; pitch linear flag
-        beq.s   .noise_nonlinear_pitch
-        move.l  d2,d4
-        asr.l   #7,d4
-        sub.l   d4,d2
-.noise_nonlinear_pitch
-
-        cmpa.w  #$1FF,a1
-        bgt.s   .lbC002BAC
-        moveq.l #0,d2
-        move.l  d2,a5
-        ;suba.l   a5,a5
-        movea.w #$200,a1
-.lbC002BAC
-        cmpa.l  pv_wg_curr_samend_ptr(a4),a0
-        beq     .wave_gen_tone_done
-        cmpa.w  #$7FFF,a5
-        ble.s   .gen_noise_innerloop
-        bra     .gen_noise_outerloop
-
-.no_noise
-        move.l  d6,d1
-        tst.b   pv_wg_unisono_run_b(a4)
-        beq.s   .is_not_in_unisono
-
-        moveq.l #3<<3,d1
-        and.b   wi_mod_density_b(a3),d1
-        lsr.b   #3,d1
-        moveq.l #9,d4
-        sub.b   d1,d4
-
-        move.l  d6,d1
-        asr.l   d4,d1
-        add.l   d6,d1
-.is_not_in_unisono
-
-        IFNE    PRETRACKER_PARANOIA_MODE
-        tst.w   pv_wg_curr_sample_len_w(a4)
-        beq     .wave_gen_tone_done
-        ENDC
-
-; ----------------------------------------
-; chord gen
-; in: d0/d1/d2/d3/d5/d7
-; in: a0
-        move.l  d3,a1
-        movea.l pv_wg_curr_sample_ptr(a4),a0
-        suba.l  a2,a2
-
-.chordtoneloop
-        move.l  d0,d4
-        sub.l   a1,d4
-        bpl.s   .noclip_osc_phase
-        moveq.l #0,d4
-.noclip_osc_phase
-        asr.l   #8,d4
-        asr.l   #7,d4
-        move.b  (a6,d4.w),d4        ; fetch precalced sample
-        ext.w   d4
-
-        moveq.l #0,d3
-        move.b  wi_osc_gain_b(a3),d3
-        muls    d4,d3
-        asr.w   #7,d3
-
-        move.b  (a0),d4
-        ext.w   d4
-        add.w   d3,d4
-        CLIPTO8BIT d4
-        move.b  d4,(a0)+
-
-        add.l   d1,d0
-        cmp.l   d7,d0
-        blt.s   .lbC0025A2
-        sub.l   d7,d0
-        add.l   d5,a1
-        cmp.l   a5,a1
-        blt.s   .lbC00259A
-        neg.l   d5
-        move.l  a5,a1
-.lbC00259A
-        cmp.l   pv_wg_osc_speed_l(a4),a1
-        bgt.s   .lbC0025A2
-        neg.l   d5
-        move.l  pv_wg_osc_speed_l(a4),a1
-.lbC0025A2
-        tst.l   d2
-        beq.s   .chordtone_done
-        adda.l  d2,a2
-        move.l  a2,d1
-        asr.l   #8,d1
-        asr.l   #2,d1
-        add.l   d6,d1
-
-        btst    #4,wi_flags_b(a3)   ; pitch linear flag
-        beq.s   .no_linear_pitch
-        move.l  d2,d4
-        asr.l   #7,d4
-        sub.l   d4,d2
-.no_linear_pitch
-        cmp.l   d7,d1
-        bcs.s   .chordtone_done
-        moveq.l #0,d2
-        moveq.l #0,d1
-.chordtone_done
-        cmpa.l  pv_wg_curr_samend_ptr(a4),a0
-        bne.s   .chordtoneloop
-
-.wave_gen_tone_done
-        addq.b  #1,pv_wg_chord_note_num_b(a4)
-        cmp.b   #4,pv_wg_chord_note_num_b(a4)
-        bne     .wavegen_chordloop
-
-        moveq.l #3<<3,d0
-        and.b   wi_mod_density_b(a3),d0 ; unisono
-        beq.s   .chords_done
-        move.l  a6,d1
-        beq.s   .chords_done
-        tst.b   pv_wg_unisono_run_b(a4)
-        bne.s   .chords_done
-        move.w  #$0001,pv_wg_chord_note_num_b(a4) ; sets also pv_wg_unisono_run_b
-        bra     .wavegen_chordloop
-
-.chords_done
-; ----------------------------------------
-; filters
-; proposed register assignment:
-; a0 = sample output
-; a1 = end of filter chunk
-; a2 = filter func
-; d7/a6 = filter start
-; a4 = MyPlayer
-; a5 = unused
-; a3 = waveinfo
-; d3/d4/d5/d6 = filter taps
-; d0/d1/d7 = scratch
-
-        moveq.l #0,d0
-        move.b  wi_flt_type_b(a3),d0
-        beq     .filter_done
-
-        IFNE    PRETRACKER_PARANOIA_MODE
-        tst.w   pv_wg_curr_sample_len_w(a4)
-        beq     .filter_done
-        ENDC
-
-        add.w   d0,d0
-        lea     .filterfunc_jmptable(pc),a2
-        add.w   -1*2(a2,d0.w),a2
-
-        moveq.l #0,d4       ; filter tap values
-        moveq.l #0,d5       ; filter tap values
-        movem.l d4-d5,pv_wg_flt_taps(a4)
-
-        lea     wi_flt_start_b(a3),a0
-        moveq.l #0,d0
-        move.b  (a0)+,d0    ; wi_flt_start_b
-        lsl.w   #8,d0
-
-        move.b  (a0)+,d4    ; wi_flt_min_b
-        lsl.w   #8,d4       ; flt_min*256
-
-        move.b  (a0)+,d5    ; wi_flt_max_b
-        lsl.w   #8,d5       ; flt_max*256
-
-        move.b  (a0)+,d3    ; wi_flt_speed_b
-        ext.w   d3
-        ext.l   d3          ; flt_speed*128
-        lsl.l   #7,d3
-
-        movea.l pv_wg_curr_sample_ptr(a4),a0
-
-.entry_to_filter_loop
-        move.l  d0,a6
-        move.l  d3,d1       ; flt_speed_b*128
-        adda.l  d1,a6       ; suppress M68kUnexpectedConditionalInstruction
-        bgt.s   .filter_speed_pos
-
-.filter_speed_neg
-        move.l  d4,d1       ; flt_min*256
-        cmp.l   d1,d0
-        blt.s   .lbC002790
-        cmp.l   d1,a6
-        bgt.s   .lbC002936
-        move.l  d1,a6
-        cmp.l   d5,d1       ; flt_max*256
-        beq.s   .filter_load_min
-        neg.l   d3          ; flt_speed_b*128
-        bra.s   .filter_load_min
-
-.filterfunc_jmptable
-        dc.w    .lowpassfilter-.filterfunc_jmptable
-        dc.w    .highpassfilter-.filterfunc_jmptable
-        dc.w    .bandpassfilter-.filterfunc_jmptable
-        dc.w    .notchfilter-.filterfunc_jmptable
-
-.lbC002790
-        tst.l   d0
-        blt.s   .lbC002936
-        move.l  a6,d7
-        bgt.s   .lbC002936
-        neg.l   d3          ; flt_speed_b*128
-        move.w  #$FF,d2
-        move.w  d2,d1
-        sub.l   a6,a6
-        bra.s   .filter_cont
-
-.filter_speed_pos
-        cmp.l   d5,d0       ; flt_max*256
-        bgt.s   .lbC002D2A
-        cmp.l   d5,a6       ; flt_max*256
-        blt.s   .lbC002936
-        move.l  d4,d2       ; flt_min*256
-        cmp.l   d5,d2       ; flt_max*256
-        beq.s   .filter_load_max_no_flip
-        neg.l   d3          ; flt_speed_b*128
-        move.l  d5,a6       ; flt_max*256
-        bra.s   .filter_load_max
-
-.lbC002D2A
-        cmpi.l  #$FF00,d0
-        bgt.s   .lbC002936
-        cmp.l   #$FEFF,a6
-        ble.s   .lbC002936
-        neg.l   d3          ; flt_speed_b*128
-        moveq.l #0,d2
-        move.l  #$FF00,a6
-        bra.s   .filter_cont
-
-.lbC002936
-        move.w  a6,d2
-        lsr.w   #8,d2
-        not.b   d2
-        bra.s   .filter_cont
-
-.filter_load_max_no_flip
-        movea.l d2,a6
-.filter_load_max
-        moveq.l #0,d2
-        move.b  wi_flt_max_b(a3),d2
-        not.b   d2
-        bra.s   .filter_cont
-
-.filter_load_min
-        moveq.l #0,d2
-        move.b  wi_flt_min_b(a3),d2
-        not.b   d2
-
-.filter_cont
-        btst    #0,wi_flt_type_b(a3)
-        bne.s   .not_notch_or_highpass
-.highpass_or_notch          ; entered for 2 or 4
-        not.b   d2
-.not_notch_or_highpass      ; entered for 1 or 3
-        move.w  d2,d0
-        add.w   d0,d0
-
-        moveq.l #0,d7
-        move.b  wi_flt_resonance_b(a3),d7
-        beq.s   .filter_no_resonance
-        move.w  d2,d1
-        ext.l   d1
-        lsl.l   #8,d1
-
-        moveq.l #$B6/2,d0
-        sub.w   d7,d0
-        add.w   d0,d0
-
-        cmpi.w  #$36,d0
-        bge.s   .filter_no_clip_resonance
-        moveq.l #$36,d0
-.filter_no_clip_resonance
-        divu    d0,d1
-        move.w  d2,d0
-        add.w   d1,d0
-.filter_no_resonance
-        lea     $40(a0),a1          ; end of sample chunk
-
-        movem.l d3-d5,-(sp)
-        movem.w pv_wg_flt_taps(a4),d3-d6
-
-        ; d0/d2 relevant for inner loop
-.filter_innerloop
-        move.b  (a0),d1
-        ext.w   d1
-
-        move.w  d4,d7
-        sub.w   d5,d7
-        muls    d0,d7
-        asr.l   #8,d7
-        sub.w   d4,d7
-
-        add.w   d1,d7
-        muls    d2,d7
-        asr.l   #8,d7
-        add.w   d7,d4
-        move.w  d4,d7
-
-        sub.w   d5,d7
-        muls    d2,d7
-        asr.l   #8,d7
-        add.w   d7,d5
-        move.w  d5,d7
-
-        sub.w   d6,d7
-        muls    d2,d7
-        asr.l   #8,d7
-        add.w   d7,d6
-        move.w  d6,d7
-
-        sub.w   d3,d7
-        muls    d2,d7
-        asr.l   #8,d7
-        add.w   d7,d3
-        move.w  d3,d7
-
-        jmp     (a2)
-
-.highpassfilter
-        sub.w   d1,d7
-        bra.s   .filterclipresult
-
-.bandpassfilter
-        sub.w   d4,d7
-        sub.w   d5,d7
-        sub.w   d6,d7
-        asr.w   #1,d7
-        bra.s   .filterclipresult
-
-.notchfilter
-        sub.w   d4,d7
-        neg.w   d7
-
-.lowpassfilter
-.filterclipresult
-        CLIPTO8BIT d7
-.filter_outputbyte
-        move.b  d7,(a0)+
-        cmpa.l  a0,a1
-        bne.s   .filter_innerloop
-
-.filterloop_end_test
-        movem.w d3-d6,pv_wg_flt_taps(a4)
-        movem.l (sp)+,d3-d5
-
-        cmpa.l  pv_wg_curr_samend_ptr(a4),a0
-        bhs.s   .filter_done
-        move.l  a6,d0
-        bra     .entry_to_filter_loop
-
-.filter_done
-; ----------------------------------------
-; Optional Pre-Modulator
-        btst    #5,wi_mod_density_b(a3) ; post bit
-        bne.s   .nopremodulator
-        bsr     pre_Modulator
-.nopremodulator
-
-; ----------------------------------------
-; start with volume envelope
-; a0 = output sample buffer
-; d0 = scratch (e.g. sample)
-; d1 = increment for attack phase
-; d3 = current volume for attack and decay phases
-; d4 = remaining sample length - 1
-; a3 = wave info
-
-.vol_do_envelope
-        movea.l pv_wg_curr_sample_ptr(a4),a0     ; load buffer pointer
-        move.w  pv_wg_curr_sample_len_w(a4),d4     ; load length
-        IFNE    PRETRACKER_PARANOIA_MODE
-        beq     .vol_envelope_finished ; paranoia
-        ENDC
-        subq.w  #1,d4           ; we use length-1, <0 is end
-
-        moveq.l #2,d1           ; turns into $20000 through swap
-        moveq.l #0,d0
-        move.b  wi_vol_attack_b(a3),d0
-        bne.s   .has_attack_volume
-        cmpi.b  #$FF,wi_vol_sustain_b(a3)
-        beq     .vol_envelope_finished
-        ; no attack but not full sustain -> go to delay
-        ;move.l  #$100<<16,d3
-        bra.s   .vol_skip_attack
-
-.vol_avoid_overflow_with_1
-        moveq.l #1,d1
-        bra.s   .cont_vol_envelope
-.has_attack_volume
-        moveq.l #0,d3
-        cmp.w   #1,d0
-        beq.s   .cont_vol_envelope
-        cmp.w   #2,d0
-        beq.s   .vol_avoid_overflow_with_1
-        swap    d1                  ; turn into $20000
-        divu    d0,d1
-        swap    d1
-        clr.w   d1
-        ; swap is done below
-.cont_vol_envelope
-        swap    d1                  ; move to high word (should be max $20000 then)
-        btst    #5,wi_flags_b(a3)   ; vol fast flag
-        beq.s   .vol_no_fast
-        lsl.l   #4,d1               ; multiply speed by 16
-.vol_no_fast
-        add.l   d1,d3               ; increase volume
-        cmpi.l  #$FFFFFF,d3
-        ble.s   .vol_do_attack      ; first step overshooting?
-.vol_skip_attack
-        btst    #3,wi_flags_b(a3)   ; boost flag
-        bne.s   .vol_delay_boosted
-        bra.s   .vol_delay_normal
-
-.vol_do_attack
-        btst    #3,wi_flags_b(a3)   ; boost flag
-        bne.s   .vol_attack_boosted
-
-; ----------------------------------------
-; attack phase with volume boosted
-.vol_attack_normal
-.vol_attack_normal_loop
-        move.b  (a0),d0
-        ext.w   d0
-        swap    d3
-        muls    d3,d0
-        swap    d3
-        asr.w   #8,d0
-        move.b  d0,(a0)+
-
-        subq.w  #1,d4
-        bmi     .vol_envelope_finished
-        add.l   d1,d3               ; increase volume
-        cmpi.l  #$FFFFFF,d3
-        ble.s   .vol_attack_normal_loop
-
-; ----------------------------------------
-; delay phase (normal)
-
-.vol_delay_normal   ; moved this label two inst up, didn't make sense there
-        moveq.l #0,d0
-        move.b  wi_vol_delay_b(a3),d0
-        lsl.w   #4,d0
-
-        IFNE    PRETRACKER_FASTER_CODE
-        ; skip the delay -- we don't change the volume for this section
-        addq.w  #1,d0
-        sub.w   d0,d4
-        bmi     .vol_envelope_finished
-        lea     1(a0,d0.w),a0
-        ELSE
-        lea     2(a0,d0.w),a1
-
-        move.w  #$FF,d3        ; FIXME I don't think that this is quite right. Shouldn't the max volume NOT change the value?
-.vol_delay_normal_loop
-        move.b  (a0),d0
-        IFNE    1
-        ext.w   d0
-        muls    d3,d0
-        asr.w   #8,d0
-        ELSE
-        ; this should be the same as above (*(256-1))
-        spl     d3
-        add.b   d3,d0
-        ENDC
-        move.b  d0,(a0)+
-
-        cmpa.l  a1,a0
-        dbeq    d4,.vol_delay_normal_loop
-        bne     .vol_envelope_finished
-        ENDC
-
-        bra.s   .vol_delay_end_reached
-
-; ----------------------------------------
-; attack with volume boosted
-
-.vol_attack_boosted
-.vol_attack_boosted_loop
-        move.b  (a0),d0
-        ext.w   d0
-        swap    d3
-        muls    d3,d0
-        swap    d3
-        asr.w   #6,d0
-        CLIPTO8BIT d0
-        move.b  d0,(a0)+
-
-        subq.w  #1,d4
-        bmi     .vol_envelope_finished
-        add.l   d1,d3
-        cmpi.l  #$FFFFFF,d3
-        ble.s   .vol_attack_boosted_loop
-
-; ----------------------------------------
-; delay with max volume boosted
-
-.vol_delay_boosted
-        moveq.l #0,d0
-        move.b  wi_vol_delay_b(a3),d0
-        lsl.w   #4,d0
-
-        lea     2(a0,d0.w),a1
-
-        IFNE    PRETRACKER_FASTER_CODE
-.vol_delay_boosted_loop
-        move.b  (a0),d0
-        add.b   d0,d0
-        CLIPTO8BITAFTERADD d0
-        add.b   d0,d0
-        CLIPTO8BITAFTERADD d0
-        ELSE
-        move.w  #$FF,d3        ; FIXME I don't think that this is quite right. It should be $100 to boost by full volume
-.vol_delay_boosted_loop
-        move.b  (a0),d0
-        ext.w   d0
-        muls    d3,d0
-        asr.w   #6,d0
-        CLIPTO8BIT d0
-        ENDC
-        move.b  d0,(a0)+
-
-        cmpa.l  a1,a0
-        dbeq    d4,.vol_delay_boosted_loop
-        bne     .vol_envelope_finished
-
-.vol_delay_end_reached
-        subq.w  #1,d4
-
-; ----------------------------------------
-; decay phase
-; d0 = scratch
-; d1 = current volume decrement
-; d2 = table index boundary
-; d3 = 16:16 decay pos
-; d4 = sample length counter
-; d5 = volume
-; d6 = scratch
-; d7 = decay increment
-; a0 = sample pointer
-; a1 = (current) roll off table pointer (points to upper bound)
-; a2 = lower bound
-
-.vol_do_decay
-        moveq.l #0,d3
-        move.b  wi_vol_decay_b(a3),d3
-        beq     .vol_do_sustain
-        move.w  d3,d7
-        mulu    d7,d7
-        lsr.w   #2,d7
-        add.w   d3,d7       ; d7 = (d3^2)/4+d3 (<= 16511)
-
-        btst    #5,wi_flags_b(a3)   ; vol fast flag
-        beq.s   .vol_decay_not_fast
-        moveq.l #0,d3       ; will cause a5=$400,a2=$200,d2=0, decay value has no effect on the ramp used
-.vol_decay_not_fast
-        lsl.w   #8,d3
-        lsl.l   #4,d3
-        move.l  d3,d2
-        swap    d2
-        lea     pre_roll_off_table(pc),a1
-        add.w   d2,d2
-        adda.w  d2,a1
-        move.w  (a1)+,a2    ; first index in table
-        lsr.w   #1,d2       ; update next boundary
-
-        moveq.l #0,d1       ; current volume decrement
-        moveq.l #0,d5
-        not.w   d5          ; set maximum volume
-.vol_decay_normal_loop
-        add.l   d7,d3       ; increment position in decay
-        swap    d3
-
-        cmp.w   #$8e,d3     ; pos in table where it makes no sense to do lerp anymore
-        bhi.s   .vol_keep_voldec
-        cmp.w   d2,d3
-        bls.s   .vol_do_lerp
-
-.vol_lerp_next_section
-        move.w  (a1),a2
-        IFNE    PRETRACKER_BUGFIX_CODE
-        lea     pre_roll_off_table+2(pc),a1 ; Take the right boundary value
-        ELSE
-        lea     pre_roll_off_table(pc),a1   ; This will set a wrong boundary and thus plateau the slope decay speed
-        ENDC
-        add.w   d3,a1
-        add.w   d3,a1
-
-        move.w  d3,d2       ; update next boundary
-
-.vol_do_lerp        ; ((lowerbound-upperbound)*(d3<<8))>>8 + upperbound
-        move.w  a2,d1
-        move.w  (a1),d0
-        sub.w   d1,d0               ; delta between lower and upper bound (negative value)
-        beq.s   .vol_skip_lerp
-
-        swap    d3
-        move.w  d3,d1
-        lsr.w   #8,d1
-        muls    d0,d1
-        asr.l   #8,d1
-        add.w   a2,d1
-        swap    d3
-
-.vol_keep_voldec
-.vol_skip_lerp
-        swap    d3
-        sub.w   d1,d5
-        bls.s   .vol_do_sustain
-
-        move.w  d5,d6
-        lsr.w   #8,d6
-
-        cmp.b   wi_vol_sustain_b(a3),d6
-        bls.s   .vol_do_sustain
-
-        move.b  (a0),d0
-        ext.w   d0
-        muls    d6,d0
-        btst    #3,wi_flags_b(a3)       ; boost flag
-        CLIPORTRUNC8BIT d0
-        move.b  d0,(a0)+
-
-        dbra    d4,.vol_decay_normal_loop
-        bra.s   .vol_envelope_finished
-
-; ----------------------------------------
-; sustain phase
-.vol_do_sustain
-        moveq.l #0,d3
-        move.b  wi_vol_sustain_b(a3),d3
-        beq.s   .vol_sustain_silence
-
-        btst    #3,wi_flags_b(a3)       ; boost flag
-        beq.s   .vol_sustain_normal
-.vol_sustain_boosted
-.vol_sustain_boosted_loop
-        move.b  (a0),d0
-        ext.w   d0
-        muls    d3,d0
-        asr.w   #6,d0
-
-        CLIPTO8BIT d0
-        move.b  d0,(a0)+
-        dbra    d4,.vol_sustain_boosted_loop
-        bra.s   .vol_envelope_finished
-
-.vol_sustain_silence
-        moveq.l #0,d0
-.vol_sustain_silence_loop
-        move.b  d0,(a0)+
-        dbra    d4,.vol_sustain_silence_loop
-        bra.s   .vol_envelope_finished
-
-.vol_sustain_normal
-        IFNE    PRETRACKER_FASTER_CODE
-        cmp.b   #$ff,d3
-        beq.s   .vol_envelope_finished
-        ENDC
-
-.vol_sustain_normal_loop
-        move.b  (a0),d0
-        ext.w   d0
-        muls    d3,d0
-        asr.w   #8,d0
-        move.b  d0,(a0)+
-        dbra    d4,.vol_sustain_normal_loop
-
-.vol_envelope_finished
-
-; ----------------------------------------
-; Optional Post-Modulator
-
-        btst    #5,wi_mod_density_b(a3) ; post bit
-        beq.s   .nopostmodulator
-        bsr     pre_Modulator
-.nopostmodulator
-
-; ----------------------------------------
-; wave mixing (removed some code here that was doing nothing as result
-; because below higher octaves code would overwrite it anyway).
-        movea.l pv_my_song(a4),a6
-
-        moveq.l #0,d0
-        move.b  wi_mix_wave_b(a3),d0
-        beq.s   .mix_no_wave_mixing         ; no mixing selected
-        subq.b  #1,d0
-        IFNE    PRETRACKER_PARANOIA_MODE    ; same wave for mixing cannot be selected in Pretracker
-        cmp.b   pv_wg_curr_wave_num_b(a4),d0
-        beq     .mix_no_wave_mixing         ; same wave number!
-        ENDC
-
-        lsl.w   #2,d0
-        move.l  pv_wave_sample_table(a4,d0.w),a1
-        add.w   #sv_wavelength_table,d0
-        move.l  (a6,d0.w),d3
-
-        move.w  pv_wg_curr_sample_len_w(a4),d4 ; length of the sample to mix to
-        cmp.w   d3,d4
-        ble.s   .mix_picked_shorter
-        move.w  d3,d4
-.mix_picked_shorter
-
-        move.l  pv_wg_curr_sample_ptr(a4),a0
-
-        subq.w  #1,d4
-.mix_mixloop
-        move.b  (a0),d0
-        add.b   (a1)+,d0
-        CLIPTO8BITAFTERADD d0
-        move.b  d0,(a0)+
-        dbra    d4,.mix_mixloop
-.mix_no_wave_mixing
-
-; ----------------------------------------
-; create higher octaves (this has been massively shortened)
-
-        btst    #2,wi_flags_b(a3)
-        beq.s   .oct_has_no_extra_octaves
-
-        moveq.l #0,d4
-        move.b  wi_sam_len_b(a3),d4 ; length of the sample
-        IFNE    PRETRACKER_PARANOIA_MODE
-        beq.s   .oct_has_no_extra_octaves
-        ENDC
-        addq.w  #1,d4
-        lsl.w   #7,d4
-
-        movea.l pv_wg_curr_sample_ptr(a4),a1
-        lea     (a1,d4.l),a0    ; needs to be .l due to 32678 max length
-
-        mulu    #7,d4
-        lsr.l   #3,d4           ; * 0.875
-        subq.w  #1,d4
-.oct_downsample_loop
-        move.b  (a1),(a0)+
-        addq.l  #2,a1
-        dbra    d4,.oct_downsample_loop
-
-.oct_has_no_extra_octaves
-; ----------------------------------------
-        IFNE    PRETRACKER_PROGRESS_SUPPORT
-        move.l  pv_precalc_progress_ptr(a4),d0
-        beq.s   .no_progress_out
-        move.l  d0,a0
-        move.l  pv_precalc_sample_size(a4),d0
-        add.l   d0,(a0)
-.no_progress_out
-        ENDC
-        move.b  sv_num_waves_b(a6),d0
-        addq.b  #1,pv_wg_wave_ord_num_w+1(a4)
-        cmp.b   pv_wg_wave_ord_num_w+1(a4),d0
-        bgt     .wavegenloop
+        bsr.s   pre_WaveGen
 
 ; ----------------------------------------
         IFNE    PRETRACKER_VOLUME_TABLE
@@ -2191,109 +702,7 @@ pre_PlayerInit:
         ENDC
         rts
 
-;--------------------------------------------------------------------
-; a3: waveinfo
-;
-; d6: wetness (word)
-; uses all data registers and a0-a2 (a3 unchanged)
-pre_Modulator:
-        tst.b   wi_mod_wetness_b(a3)
-        beq.s   .earlyexit
-        moveq.l #7,d0
-        and.b   wi_mod_density_b(a3),d0
-        bne.s   .has_density
-.earlyexit
-        rts
-.has_density
-        move.l  pv_wg_curr_sample_ptr(a4),a0
-        move.w  pv_wg_curr_sample_len_w(a4),d4
-        IFNE    PRETRACKER_PARANOIA_MODE
-        bne.s   .not_zero
-        rts
-.not_zero
-        ENDC
-        moveq.l #0,d6
-        move.b  wi_mod_wetness_b(a3),d6
-
-        moveq.l #0,d7
-
-        lea     pre_modulator_ramp_8(pc),a1
-        lea     pre_ramp_up_down_32(pc),a2
-.loop   moveq.l #0,d5
-        move.b  wi_mod_length_b(a3),d5
-        mulu    (a1)+,d5            ; result is a long value
-
-        moveq.l #0,d1
-        move.b  wi_mod_predelay_b(a3),d1
-
-        btst    #5,wi_mod_density_b(a3) ; post-modulator?
-        bne.s   .factor1_256
-        ; factor 1/4 and 64
-        lsr.l   #2,d5
-        lsl.l   #6,d1
-        bra.s   .cont
-
-.factor1_256
-        lsl.w   #8,d1
-.cont   add.l   d1,d5               ; sum up length and predelay
-
-        moveq.l #0,d2
-        moveq.l #0,d3
-
-.innerloop
-        add.w   d7,d3
-        addq.w  #8,d3
-        moveq.l #0,d1
-        move.w  d3,d1
-        lsr.w   #8,d1
-        lsr.w   #3,d1
-        add.w   d1,d1
-        move.w  (a2,d1.w),d1        ; 4 bit value
-        add.l   d5,d1
-        lsr.l   #6,d1
-
-        move.w  d2,d0
-        sub.w   d1,d0
-        bmi.s   .isneg
-
-        move.b  (a0,d0.w),d1
-        ext.w   d1
-        btst    #0,d7
-        beq.s   .keep_dc
-        neg.w   d1
-.keep_dc
-
-        muls    d6,d1
-        asr.w   #8,d1
-        add.b   (a0,d2.w),d1
-
-        CLIPTO8BITAFTERADD d1
-        move.b  d1,(a0,d2.w)
-.isneg
-        addq.w  #1,d2
-        cmp.w   d4,d2
-        bcs.s   .innerloop
-
-.restartloop
-        addq.w  #1,d7
-        moveq.l #7,d0
-        and.b   wi_mod_density_b(a3),d0
-        cmp.w   d0,d7
-        bcs.s   .loop
-        rts
-
-;--------------------------------------------------------------------
-
-pre_MemClr
-        lsr.w   #1,d0
-        subq.w  #1,d0
-        bmi.s   .skipmemclr
-        moveq.l #0,d1
-.fillmemloop
-        move.w  d1,(a0)+
-        dbra    d0,.fillmemloop
-.skipmemclr
-        rts
+        include "raspberry_casket_wavegen.asm"
 
 ;********************************************************************
 ;--------------------------------------------------------------------
@@ -2392,7 +801,7 @@ pre_PlayerTick:
         move.l  d1,(a1)+    ; pcd_inst_line_ticks_b, pcd_inst_pitch_pinned_b, pcd_inst_vol_slide_b, pcd_inst_step_pos_b
 
         subq.b  #1,d1
-        move.w  d1,(a1)+    ; pcd_inst_wave_num_w
+        move.w  d1,(a1)+    ; pcd_inst_wave_num4_w
 
         move.l  #$ff010010,(a1)+    ; pcd_track_delay_offset_b, pcd_inst_speed_stop_b, pcd_inst_pitch_w
         move.l  #(MAX_VOLUME<<16)|(MAX_VOLUME<<8)|MAX_VOLUME,(a1)+  ; pcd_inst_vol_w / pcd_loaded_inst_vol_b / pcd_pat_vol_b
@@ -2467,7 +876,7 @@ pre_PlayerTick:
 .no_note_delay
         moveq.l #0,d5
         move.b  pcd_channel_num_b(a5),d5
-        move.w  sv_curr_pat_pos_w(a6),d2
+        move.w  pv_curr_pat_pos_w(a4),d2
         add.w   d2,d2
         add.w   d2,d2          ; *4
         add.w   d5,d2
@@ -2500,15 +909,15 @@ pre_PlayerTick:
         add.w   d2,d2
         add.w   d2,a0                       ; pattern data
 
-        moveq.l #15,d2
         move.b  pdb_inst_effect(a0),d4      ; instrument and command byte
+        moveq.l #15,d2
         and.w   d4,d2
         lsr.w   #4,d4                       ; instrument nr bits 0-4
 
         moveq.l #0,d5
         move.b  pdb_effect_data(a0),d5
 
-        cmpi.b  #$e,d2
+        cmp.b   #$e,d2
         bne.s   .pat_exy_cmd_cont
         ; handle $exy commands
         tst.b   pcd_note_delay_b(a5)
@@ -2544,6 +953,8 @@ pre_PlayerTick:
 ; read out pattern editor data
 
         moveq.l #0,d6                   ; clear arp flag
+        move.b  d6,pcd_pat_vol_ramp_speed_b(a5)
+        move.w  d6,pcd_pat_pitch_slide_w(a5)
         move.b  pdb_pitch_ctrl(a0),d3   ; pitch and control byte
         bpl.s   .noselinst16plus
         add.w   #16,d4                  ; add high bit of instrument number
@@ -2570,49 +981,39 @@ pre_PlayerTick:
 
 ; d2 = effect cmd, d3 = pitch_ctrl, d4 = inst number, d5 = effect data, d7 = pitch
 
-        andi.b  #$40,d3     ; ARP bit
-        bne.s   .has_arp_note
-.no_arp_note
+        and.w   #$40,d3     ; ARP bit
+        bne.s   .is_an_arp_note ; d3 is zero in this case, no no 2nd inst number
+
+        ; d3.l must be 0 in this case
+        ; normal note, not an ARP node
         tst.b   d2
         bne.s   .arp_processing_done
+
+        ; d2.l is guaranteed to be zero
+
         ; 0xx: play second instrument
         tst.b   d5
         beq.s   .no_effect
-        moveq.l #15,d1      ; FIXME seems like it only supports the lower 15 instruments
-        and.b   d5,d1
-        add.w   d1,d1
-        add.w   d1,d1
+
+        move.w  d4,d3       ; 1st instrument num
+
+        moveq.l #15,d4      ; FIXME seems like it only supports the lower 15 instruments
+        and.w   d5,d4
+        add.w   d4,d4
+        add.w   d4,d4       ; 2nd instrument from pattern effect
 
         tst.b   d7
-        bne.s   .play_2nd_inst_without_trigger
+        bne.s   .arp_processing_done
 
-.play_2nd_inst_with_no_pitch
+        ; play 2nd inst without (new) pitch
         addq.w  #1,d0       ; pattern pitch shift
         lsl.w   #4,d0
+        bra.s   .check_for_2nd_instrument
 
-        tst.b   d1
-        beq.s   .clear_note_arp
-
-        moveq.l #0,d2
-        move.w  d4,d3
-        move.w  d1,d4
-        bra.s   .trigger_new_instrument
-
-.play_2nd_inst_without_trigger
-        move.w  d4,d3       ; 1st instrument num
-        move.w  d1,d4       ; 2nd instrument num
-        bra.s   .arp_processing_done
-
-.clear_note_arp
-        clr.l   pcd_arp_notes_l(a5)
-        move.b  d7,d2
-        move.b  d4,d3
-        bra     .clear_portamento
-
-.has_arp_note
+.is_an_arp_note
         move.b  d2,d3
         or.b    d5,d3
-        beq.s   .all_arp_notes_zero     ; if we go there, d3 MUST be 0 already
+        beq.s   .all_arp_notes_zero     ; if we branch there, both d2 and d3 MUST be 0 already
 
         move.b  d2,pcd_arp_note_1_b(a5)
 
@@ -2624,40 +1025,32 @@ pre_PlayerTick:
         and.b   d5,d2
         move.b  d2,pcd_arp_note_3_b(a5)
 
-.all_arp_notes_zero
-        moveq.l #1,d6
-.no_effect
         moveq.l #0,d2   ; make sure we don't get a random command here
+
+.all_arp_notes_zero
+        moveq.l #1,d6   ; set ARP flag
+.no_effect
         moveq.l #0,d3
 
 ; ----------------------------------------
 ; d2 = effect cmd, d3 = alt inst number (or 0), d4 = inst number, d5 = effect cmd, d6 = ARP flag, d7 = pitch
 
 .arp_processing_done
-        cmpi.b  #NOTE_OFF_PITCH,d7
-        bne.s   .has_no_note_off
+        cmp.b   #NOTE_OFF_PITCH,d7
+        beq.s   .release_note
 
-.release_note
-        ; release note
-        ; FIXME we have the identical code (different regs) three times (one is inactive)
-        move.w  pcd_adsr_volume_w(a5),d4
-        asr.w   #6,d4
-        move.w  d4,pcd_adsr_vol64_w(a5)
-        moveq.l #16,d7
-        move.w  d7,pcd_adsr_pos_w(a5)
-        sub.w   d4,d7
-        lsr.w   #1,d7
-        add.b   pcd_adsr_release_b(a5),d7
-        move.b  d7,pcd_adsr_phase_speed_b(a5)
-        move.w  #3,pcd_adsr_phase_w(a5)
-        bra     .start_patt_effect_handling
+        tst.b   d7
+        beq.s   .start_patt_effect_handling
 
-.has_inst_number
-        add.w   d7,d0       ; pattern pitch shift?
+        add.w   d7,d0               ; pattern pitch shift
         lsl.w   #4,d0
 
-        cmpi.b  #3,d2       ; is command portamento?
-        beq.s   .pat_set_portamento
+        cmp.b   #3,d2       ; is command portamento?
+        beq.s   .cont_after_inst_trigger
+
+.check_for_2nd_instrument
+        tst.b   d4
+        beq.s   .cont_after_inst_trigger
 
 .trigger_new_instrument
         move.w  d4,d1
@@ -2681,39 +1074,39 @@ pre_PlayerTick:
         move.l  d1,(a1)+    ; pcd_inst_line_ticks_b, pcd_inst_pitch_pinned_b, pcd_inst_vol_slide_b, pcd_inst_step_pos_b
 
         subq.b  #1,d1
-        move.w  d1,(a1)+    ; pcd_inst_wave_num_w
+        move.w  d1,(a1)+    ; pcd_inst_wave_num4_w
 
         move.l  #$ff010010,(a1)+    ; pcd_track_delay_offset_b, pcd_inst_speed_stop_b, pcd_inst_pitch_w
         move.l  #(MAX_VOLUME<<16)|(MAX_VOLUME<<8)|MAX_VOLUME,(a1)+  ; pcd_inst_vol_w / pcd_loaded_inst_vol_b / pcd_pat_vol_b
 
         bra.s   .cont_after_inst_trigger
 
-.has_no_note_off
-        tst.b   d7
-        beq.s   .start_patt_effect_handling
-        tst.b   d4
-        bne.s   .has_inst_number
-
-        add.w   d7,d0               ; pattern pitch shift
-        lsl.w   #4,d0
+.release_note
+        ; FIXME we have the identical code (different regs) three times (one is inactive)
+        move.w  pcd_adsr_volume_w(a5),d4
+        asr.w   #6,d4
+        move.w  d4,pcd_adsr_vol64_w(a5)
+        moveq.l #16,d7
+        move.w  d7,pcd_adsr_pos_w(a5)
+        sub.w   d4,d7
+        lsr.w   #1,d7
+        add.b   pcd_adsr_release_b(a5),d7
+        move.b  d7,pcd_adsr_phase_speed_b(a5)
+        move.w  #3,pcd_adsr_phase_w(a5)
+        bra.s   .start_patt_effect_handling
 
 .cont_after_inst_trigger
         tst.b   d6                  ; has ARP?
-        bne.s   .lbC001852
-        bra.s   .lbC00193C
+        bne.s   .has_arp_check_portamento
 
-.pat_set_portamento
-        tst.b   d6
-        bne.s   .pat_new_portamento
-.lbC00193C
-        clr.l   pcd_arp_notes_l(a5)
-        moveq.l #0,d6
+.clear_arp_notes
+        move.l  d6,pcd_arp_notes_l(a5) ; the whole ARP flag in d6 must be 0
 
-.lbC001852
-        cmpi.b  #3,d2               ; is command portamento?
+.has_arp_check_portamento
+        cmp.b   #3,d2               ; is command portamento?
         beq.s   .pat_new_portamento
 
-.clear_portamento
+        ; clear portamento
         move.w  #$10,pcd_inst_pitch_w(a5)
         move.w  d0,pcd_inst_curr_port_pitch_w(a5)
         clr.w   pcd_pat_portamento_dest_w(a5)
@@ -2727,9 +1120,10 @@ pre_PlayerTick:
         add.w   d1,pcd_inst_curr_port_pitch_w(a5)
         clr.w   pcd_inst_pitch_w(a5)
         tst.b   d5
-        beq.s   .pat_keep_old_portamento
+        beq.s   .execute_pattern_command
         move.b  d5,pcd_pat_portamento_speed_b(a5)
 .pat_keep_old_portamento
+        bra.s   .execute_pattern_command
 
 ; ----------------------------------------
 .start_patt_effect_handling
@@ -2742,10 +1136,9 @@ pre_PlayerTick:
         move.b  d3,pcd_pat_2nd_inst_delay_b(a5)
 
 .has_no_second_inst
-        clr.b   pcd_pat_vol_ramp_speed_b(a5)
-        clr.w   pcd_pat_pitch_slide_w(a5)
-        tst.b   d6
+        tst.b   d6                  ; ARP bit is set, cannot have a command
         bne     .pat_play_cont
+.execute_pattern_command
         add.w   d2,d2
         move.w  .pattern_command_jmptable(pc,d2.w),d2
         jmp     .pattern_command_jmptable(pc,d2.w)
@@ -2754,7 +1147,7 @@ pre_PlayerTick:
         dc.w    .pat_play_nop-.pattern_command_jmptable
         dc.w    .pat_slide_up-.pattern_command_jmptable
         dc.w    .pat_slide_down-.pattern_command_jmptable
-        dc.w    .pat_play_nop-.pattern_command_jmptable
+        dc.w    .pat_play_nop-.pattern_command_jmptable     ; portamento is handled above
         dc.w    .pat_set_vibrato-.pattern_command_jmptable
         dc.w    .pat_set_track_delay-.pattern_command_jmptable
         dc.w    .pat_play_nop-.pattern_command_jmptable
@@ -2817,17 +1210,33 @@ pre_PlayerTick:
         beq.s   .pat_play_cont  ; we are at channel 3 -- track delay not available here
 
         lea     pcd_SIZEOF+pcd_track_delay_buffer+ocd_volume(a5),a1
+        moveq.l #0,d3
         moveq.l #MAX_TRACK_DELAY-1,d2
 .clr_track_delay_buffer_loop
-        clr.b   (a1)            ; ocd_volume
+        move.b  d3,(a1)         ; ocd_volume
         lea     ocd_SIZEOF(a1),a1
         dbra    d2,.clr_track_delay_buffer_loop
 
         tst.b   d5
         bne.s   .pat_track_delay_set
+        IFNE    PRETRACKER_BUGFIX_CODE      ; clearing track delay when it already was cleared will overwrite the note needlessly
+        tst.b   pcd_track_delay_steps_b(a5)
+        beq.s   .pat_play_cont
+        ENDC
         st      pcd_track_delay_steps_b(a5)
         bra.s   .handle_track_delay
 
+.pat_track_delay_set
+        moveq.l #15,d2
+        and.b   d5,d2
+        add.b   d2,d2
+        move.b  d2,pcd_track_delay_steps_b(a5)
+;        subq.b  #1,d2
+;        move.b  d2,pcd_SIZEOF+pcd_track_init_delay_b(a5)
+        lsr.b   #4,d5
+        move.b  d5,pcd_track_delay_vol16_b(a5)
+        bra.s   .pat_play_cont
+
 ; ----------------------------------------
 .pat_volume_ramp
         tst.b   d5
@@ -2846,17 +1255,6 @@ pre_PlayerTick:
         move.b  d5,pcd_pat_vol_ramp_speed_b(a5)
         bra.s   .pat_play_cont
 
-.pat_track_delay_set
-        moveq.l #15,d2
-        and.b   d5,d2
-        add.b   d2,d2
-        move.b  d2,pcd_track_delay_steps_b(a5)
-;        subq.b  #1,d2
-;        move.b  d2,pcd_SIZEOF+pcd_track_init_delay_b(a5)
-        lsr.b   #4,d5
-        move.b  d5,pcd_track_delay_vol16_b(a5)
-        bra.s   .pat_play_cont
-
 ; ----------------------------------------
 .pat_slide_down
         neg.w   d5
@@ -2924,7 +1322,7 @@ pre_PlayerTick:
         move.b  pv_pat_curr_row_b(a4),d0
         addq.b  #1,d0                       ; normal step increment
 
-        move.w  sv_curr_pat_pos_w(a6),d3    ; current song position
+        move.w  pv_curr_pat_pos_w(a4),d3    ; current song position
 
         move.b  pv_next_pat_row_b(a4),d2    ; $ff means no pattern break
         bmi.s   .no_pattern_break
@@ -2977,7 +1375,7 @@ pre_PlayerTick:
         ENDC
 .no_restart_song
         move.b  d0,pv_pat_curr_row_b(a4)
-        move.w  d3,sv_curr_pat_pos_w(a6)
+        move.w  d3,pv_curr_pat_pos_w(a4)
 
         move.b  pv_pat_speed_even_b(a4),d1
         lsr.b   #1,d0
@@ -3133,6 +1531,7 @@ pre_PlayerTick:
 ; d0 = current step / next step
 ; d5 = command parameter data / scratch
 ; d2 = note stitched flag
+; d3 = scratch
 ; d6 = scratch
 ; ----------------------------------------
 .inst_select_wave
@@ -3141,28 +1540,33 @@ pre_PlayerTick:
         bhs     .inst_cmd_cont_next
         add.w   d5,d5
         add.w   d5,d5
-        cmp.w   pcd_inst_wave_num_w(a5),d5
+        cmp.w   pcd_inst_wave_num4_w(a5),d5
         beq     .inst_cmd_cont_next
-        move.w  d5,pcd_inst_wave_num_w(a5)
+        move.w  d5,pcd_inst_wave_num4_w(a5)
         move.l  sv_waveinfo_table(a6,d5.w),a3
         move.l  a3,pcd_waveinfo_ptr(a5)
 
-        move.b  pcd_channel_mask_b(a5),d3
-        move.b  d3,pcd_out_trg_b(a5)
-        or.b    d3,pv_trigger_mask_w+1(a4)
-
         move.l  pv_wave_sample_table(a4,d5.w),a1
+
+        pea     .inst_cmd_cont_next(pc)
+.inst_select_wave_subroutine
+        move.w  wi_chipram_w(a3),d5
+
+        move.b  pcd_channel_mask_b(a5),d3
+        or.b    d3,pv_trigger_mask_w+1(a4)
+        move.b  d3,pcd_out_trg_b(a5)
+
+        moveq.l #0,d3
+
         tst.w   d6
         bne.s   .inst_select_wave_nosync
 
-        moveq.l #0,d3
         move.w  wi_loop_offset_w(a3),d6     ; is unlikely 32768
-        move.w  wi_subloop_len_w(a3),d5
+        tst.w   wi_subloop_len_w(a3)
         bne.s   .inst_set_wave_has_subloop
-
+.inst_set_wave_has_no_subloop
         adda.w  d6,a1                       ; add loop offset
 
-        move.w  wi_chipram_w(a3),d5
         sub.w   d6,d5
         ;cmp.w   #1,d5 ; not necessary as increases in steps of 2
         bhi.s   .inst_set_wave_has_min_length
@@ -3174,6 +1578,7 @@ pre_PlayerTick:
 .inst_set_wave_has_subloop
         move.w  d3,pcd_out_lof_w(a5)
         move.l  a1,pcd_out_ptr_l(a5)
+        move.w  d6,pcd_inst_loop_offset_w(a5)
 
         st      pcd_inst_ping_pong_dir_b(a5)
         moveq.l #0,d5
@@ -3181,16 +1586,19 @@ pre_PlayerTick:
         addq.w  #1,d5
         move.w  d5,pcd_inst_subloop_wait_w(a5)
 
-        move.w  d6,pcd_inst_loop_offset_w(a5)
-        bra     .inst_cmd_cont_next
+        rts
 
 ; ----------------------------------------
 .inst_select_wave_nosync
-        move.w  wi_chipram_w(a3),d5
-        moveq.l #0,d3
+        IFNE    PRETRACKER_BUGFIX_CODE
+        move.w  wi_loop_offset_w(a3),d6
+        tst.w   wi_subloop_len_w(a3)
+        beq.s   .inst_set_wave_has_no_subloop
+        ELSE
         tst.w   wi_subloop_len_w(a3)
         bne.s   .inst_set_wave_ns_has_subloop
 
+        ; note that nosync on non-looping waves doesn't effectivly do anything special
         move.w  wi_loop_offset_w(a3),d6
         adda.w  d6,a1
 
@@ -3202,21 +1610,27 @@ pre_PlayerTick:
         move.w  d5,pcd_out_len_w(a5)
 
         moveq.l #-1,d3
+        ENDC
 .inst_set_wave_ns_has_subloop
         move.w  d3,pcd_out_lof_w(a5)
         move.l  a1,pcd_out_ptr_l(a5)
 
+        IFEQ    PRETRACKER_BUGFIX_CODE
         move.w  pcd_inst_loop_offset_w(a5),d6
         cmp.w   d6,d5
+        ELSE
+        cmp.w   pcd_inst_loop_offset_w(a5),d5
+        ENDC
         bhs.s   .inst_set_wave_ns_keep_pp
         st      pcd_inst_ping_pong_dir_b(a5)
 .inst_set_wave_ns_keep_pp
-        moveq.l #0,d5
-        move.w  d5,pcd_inst_subloop_wait_w(a5)
+        clr.w   pcd_inst_subloop_wait_w(a5)
 
+        IFEQ    PRETRACKER_BUGFIX_CODE          ; This adds an extra jump to the wave that's not desired
         sub.w   wi_subloop_step_w(a3),d6
         move.w  d6,pcd_inst_loop_offset_w(a5)
-        bra     .inst_cmd_cont_next
+        ENDC
+        rts
 
 ; ----------------------------------------
 .inst_slide_down
@@ -3364,54 +1778,18 @@ pre_PlayerTick:
         move.b  d0,pcd_inst_step_pos_b(a5)  ; update inst step pos
 
 .inst_still_ticking
-        tst.b   pcd_inst_wave_num_w+1(a5)
+        tst.b   pcd_inst_wave_num4_w+1(a5)
         bpl.s   .inst_wave_selected
 
-.inst_no_wave_selected
-        moveq.l #0,d3
-        move.w  d3,pcd_inst_wave_num_w(a5)  ; FIXME why set to wave 0? just asking, because this is zero based
+.inst_no_wave_selected                      ; FIXME this code is dubious at best -- it selects wave 0 if no wave was selected before
+        clr.w   pcd_inst_wave_num4_w(a5)
         move.l  sv_waveinfo_ptr(a6),a3
         move.l  a3,pcd_waveinfo_ptr(a5)
 
-        move.b  pcd_channel_mask_b(a5),d0
-        move.b  d0,pcd_out_trg_b(a5)
-        or.b    d0,pv_trigger_mask_w+1(a4)
-
-        tst.w   wi_subloop_len_w(a3)
-        bne.s   .inst_nowave_has_subloop
-
-        move.w  wi_loop_offset_w(a3),d5
         move.l  pv_wave_sample_table(a4),a1
-        adda.w  d5,a1
-        move.l  a1,pcd_out_ptr_l(a5)
 
-        move.w  wi_chipram_w(a3),d0
-        move.w  d0,d6
-        subq.w  #1,d6
-        cmp.w   d5,d6
-        ble.s   .lbC00118E
-        sub.w   d5,d0
-        bra.s   .lbC000F48
-.lbC00118E
-        moveq.l #2,d0
-.lbC000F48
-        move.w  d0,pcd_out_len_w(a5)
-        subq.w  #1,d3
-        bra.s   .inst_set_no_lof
-
-.inst_nowave_has_subloop
-        move.l  pv_wave_sample_table(a4),pcd_out_ptr_l(a5)
-
-        move.w  wi_loop_offset_w(a3),d5
-.inst_set_no_lof
-        move.w  d3,pcd_out_lof_w(a5)
-        moveq.l #0,d0
-        move.b  wi_subloop_wait_b(a3),d0
-        addq.w  #1,d0
-        move.w  d0,pcd_inst_subloop_wait_w(a5)
-
-        move.w  d5,pcd_inst_loop_offset_w(a5)
-        st      pcd_inst_ping_pong_dir_b(a5)
+        moveq.l #0,d6
+        bsr     .inst_select_wave_subroutine
 
 .inst_wave_selected
         cmpi.b  #$FF,d2
@@ -3419,16 +1797,18 @@ pre_PlayerTick:
         subq.b  #1,d2
         move.b  d2,pcd_inst_line_ticks_b(a5)
 
-.inst_no_inst_active
 .inst_pat_loop_exit3
+.inst_no_inst_active
 
 ; ----------------------------------------
-
-; expects d2 = inst num
 ; a5 = channel
         move.w  pcd_inst_vol_w(a5),d1
         tst.b   pcd_new_inst_num_b(a5)
         bne.s   .load_instrument
+        IFNE    PRETRACKER_BUGFIX_CODE
+        move.l  a3,d3
+        beq.s   .no_inst_selected
+        ENDC
 
 .dont_load_instrument
         move.w  pcd_adsr_volume_w(a5),d2
@@ -3595,7 +1975,7 @@ pre_PlayerTick:
         move.w  d5,pcd_inst_subloop_wait_w(a5)
 
         tst.b   d4
-        bpl.s   .wave_is_moving_backwards
+        bpl.s   .loop_is_moving_backwards
         add.w   d2,d1                   ; increment offset in forward direction one step
 
         move.w  d1,d4
@@ -3614,7 +1994,7 @@ pre_PlayerTick:
         clr.b   pcd_inst_ping_pong_dir_b(a5) ; mark going backwards
         bra.s   .wave_loop_dir_changed
 
-.wave_is_moving_backwards
+.loop_is_moving_backwards
         sub.w   d2,d1                   ; decrement offset in backward direction one step
 
         move.w  wi_loop_start_w(a3),d4
@@ -3644,7 +2024,7 @@ pre_PlayerTick:
 .done_lof_calc
         move.w  d4,pcd_out_lof_w(a5)
 
-        move.w  pcd_inst_wave_num_w(a5),d1  ; FIXME can we move this to wave loading?
+        move.w  pcd_inst_wave_num4_w(a5),d1 ; FIXME can we move this to wave loading?
         move.l  pv_wave_sample_table(a4,d1.w),pcd_out_ptr_l(a5)
         bra.s   .loop_handling_done
 
@@ -3661,8 +2041,8 @@ pre_PlayerTick:
         clr.b   pcd_wave_offset_b(a5)
 
         move.b  pcd_channel_mask_b(a5),d2   ; trigger output
-        move.b  d2,pcd_out_trg_b(a5)
         or.b    d2,pv_trigger_mask_w+1(a4)
+        move.b  d2,pcd_out_trg_b(a5)
 
         move.w  wi_chipram_w(a3),d2
         sub.w   d1,d2
@@ -3676,7 +2056,7 @@ pre_PlayerTick:
 .waveoffset_is_not_past_end
         move.w  d2,pcd_out_len_w(a5)
 
-        move.w  pcd_inst_wave_num_w(a5),d2
+        move.w  pcd_inst_wave_num4_w(a5),d2
         add.l   pv_wave_sample_table(a4,d2.w),d1
         move.l  d1,pcd_out_ptr_l(a5)
 
@@ -3745,25 +2125,26 @@ pre_PlayerTick:
         lsr.w   #6,d5
         lea     pre_octave_select_table(pc),a1
         moveq.l #0,d1
-        move.b  (a1,d5.w),d1
+        move.b  (a1,d5.w),d1                ; higher octave 1-3
+
+        move.l  pcd_out_ptr_l(a5),d4
         move.w  pcd_out_lof_w(a5),d7
         addq.w  #1,d7   ; compare to $ffff
         beq.s   .has_no_loop_offset
         subq.w  #1,d7
         lsr.w   d1,d7
-        move.w  d7,pcd_out_lof_w(a5)
+        move.w  d7,pcd_out_lof_w(a5)        ; halve/quarter/eighth loop offset
         lsr.w   d1,d3
-        move.w  d3,pcd_out_len_w(a5)
-        bra.s   .lbC0005C0
+        move.w  d3,pcd_out_len_w(a5)        ; halve/quarter/eighth loop length
+        bra.s   .cont_after_loop_fix
 
 .has_no_loop_offset
         tst.b   pcd_out_trg_b(a5)
         beq.s   .no_retrigger_new
 
-        move.w  pcd_inst_wave_num_w(a5),d7
+        move.w  pcd_inst_wave_num4_w(a5),d7
         movea.l pv_wave_sample_table(a4,d7.w),a3
 
-        move.l  pcd_out_ptr_l(a5),d4
         sub.l   a3,d4
         move.w  d3,d6
         add.w   d4,d6
@@ -3785,18 +2166,16 @@ pre_PlayerTick:
 
         lsr.w   d1,d4
         add.l   a3,d4
-        move.l  d4,pcd_out_ptr_l(a5)
 
-.lbC0005C0
-        subq.b  #1,d1
-        bmi.s   .is_normal_octave
+.cont_after_loop_fix
+        subq.w  #1,d1
+        ;bmi.s   .is_normal_octave          ; this should never happen -- d1 is at least 1
         ; find offset in sample buffer for the right octave
-        moveq.l #0,d4
 .movetoloopposloop
         add.l   d2,d4
         lsr.w   #1,d2
         dbra    d1,.movetoloopposloop
-        add.l   d4,pcd_out_ptr_l(a5)
+        move.l  d4,pcd_out_ptr_l(a5)        ; move trigger start pos to right octave wave
 
 .no_retrigger_new
 .is_normal_octave
@@ -3817,6 +2196,7 @@ pre_PlayerTick:
 .noclippitchlow2
         tst.b   pcd_out_trg_b(a5)
         beq.s   .wasnottriggered
+        ; this code seems to move the sample start to "loop offset" for first trigger
         moveq.l #0,d0
         move.w  pcd_out_lof_w(a5),d0
         addq.w  #1,d0       ; compare to $ffff
@@ -3826,19 +2206,28 @@ pre_PlayerTick:
         clr.w   pcd_out_lof_w(a5)
 .wasnottriggered
 .hasnoloop2
-        cmp.w   pcd_last_trigger_pos_w(a5),d3
-        beq.s   .hassamesampos
-        move.w  d3,pcd_last_trigger_pos_w(a5)
+        ; this code is probably here to ensure triggering the wave when the octave sample changes
+        cmp.w   pcd_last_trigger_length_w(a5),d3
+        beq.s   .hassamesamlen
+        move.w  d3,pcd_last_trigger_length_w(a5)
         move.b  pcd_channel_mask_b(a5),d3
-        move.b  d3,pcd_out_trg_b(a5)
+        IFND    PRESTO_SKIP_OUTPUT
         or.b    d3,pv_trigger_mask_w+1(a4)
-.hassamesampos
+        ELSE
+        cmp.b   pcd_out_trg_b(a5),d3
+        beq.s   .hassamesamlen
+        or.b    d3,pv_trigger_mask_w+1(a4)
+        tas     d3      ; we need to mark the first length loading, without trigger that is actually not a real trigger, so we can filter it
+        ENDC
+        move.b  d3,pcd_out_trg_b(a5)
+.hassamesamlen
         add.w   d6,d6
         move.w  pv_period_table(a4,d6.w),pcd_out_per_w(a5)
 
 ; ----------------------------------------
 ; track delay handling
 
+.no_inst_selected
         move.b  pcd_track_delay_steps_b(a5),d3
         beq     .incrementchannel       ; no track delay
 
@@ -3924,7 +2313,7 @@ pre_PlayerTick:
         move.l  (a1)+,(a3)+             ; ocd_sam_ptr
         move.l  (a1)+,(a3)+             ; ocd_length/ocd_loop_offset
         move.l  (a1)+,(a3)+             ; ocd_period/ocd_volume/ocd_trigger
-        ;move.l   (a1)+,(a3)+            ; this is never used
+        ;move.l  (a1)+,(a3)+             ; this is never used
 
         ;clr.b   ocd_volume-ocd_unused(a1) ; does not seem to bother
 
@@ -3949,6 +2338,7 @@ pre_PlayerTick:
         ; so this changed a lot from the original routine
         move.w  pv_trigger_mask_w(a4),d2
 
+        IFND    PRESTO_SKIP_OUTPUT
         IFNE    PRETRACKER_COPPER_OUTPUT
         move.l  pv_copperlist_ptr(a4),d0
         beq     .skipcopperlist
@@ -3988,8 +2378,8 @@ pre_PlayerTick:
 
 .is_looping_sample
         move.l  ocd_sam_ptr(a0),d0
-        add.l   d2,d0
-        move.l  d0,d6
+        add.l   d2,d0                       ; add loop offset to sample start
+        move.l  d0,d6                       ; make a copy for loop start
 
         move.w  ocd_length(a0),d4
         lsr.w   #1,d4
@@ -3997,7 +2387,7 @@ pre_PlayerTick:
         tst.b   ocd_trigger(a0)
         beq.s   .setptrvolper
         move.b  d5,ocd_trigger(a0)
-        sub.l   d2,d0
+        sub.l   d2,d0                       ; if triggered, deduct loop offset so we are back at sample start
 .setptrvolper
         move.w  d0,1*4(a1)  ; ac_ptr (lo)
         swap    d0
@@ -4098,6 +2488,7 @@ pre_PlayerTick:
 
 .skiprasterwait
         ENDC
+        ENDC
 
         IFNE    PRETRACKER_DONT_TRASH_REGS
         movem.l (sp)+,d2-d7/a2-a6
@@ -4106,12 +2497,26 @@ pre_PlayerTick:
 
 ;--------------------------------------------------------------------
 ; table data currently about 594 bytes
+        ; Tables used by WaveGen
+pre_roll_off_table: ; used by WaveGen
+        dc.w    $400,$200,$180,$140,$100,$C0,$A0,$80,$78,$74,$6E
+        dc.w    $69,$64,$5A,$46,$40,$38,$30,$28,$20,$1F,$1E,$1D
+        dc.w    $1C,$1B,$1A,$19,$18,$17,$16,$15,$14,$13,$12,$11
+        dc.w    $10,15,14,13,13,12,12,11,11,10,10,9,9,8,8,8,8,7,7
+        dc.w    7,7,6,6,6,6,5,5,5,5,4,4,4,4,4,4,4,4,4,4,3,4,4,3,4
+        dc.w    4,3,4,3,4,3,4,3,4,3,3,3,3,3,3,3,3,3,2,3,3,3,2,3,3
+        dc.w    2,3,3,2,3,3,2,3,2,3,2,3,2,3,2,3,2,2,2,2,2,2,2,2,1
+        dc.w    2,1,2,1,2,1,2,1,1,2,1,1,1,2,1
 
-; I assume this is a log table for freq distances within an octave
+; I assume this is a log table for freq distances within an octave ; used by WaveGen
 pre_log12_table:
         dc.b    $400000/$8000,$400000/$871D,$400000/$8F2F,$400000/$97B7,$400000/$9FC4,$400000/$A9DE
         dc.b    $400000/$B505,$400000/$BF49,$400000/$CB31,$400000/$D645,$400000/$E215,$400000/$F1A0
 
+pre_modulator_ramp_8: ; used by WaveGen
+        ;dc.w     77,293,539,1079,1337,1877,2431,3031 ; the 1079 value is strange (938 better?)
+        dc.w     $4D,$125,$21B,$437,$539,$755,$96D,$BD7
+
 ; linear then steep quadratic slope
 pre_vib_speed_table:
         dc.b    2,3,4,5,6,7,8,9,10,11,12,13,14,20,40,80
@@ -4131,16 +2536,6 @@ pre_fast_roll_off_16:
         dc.w    $400,$200,$80,$64,$50,$40,$30,$20,$10,14,12,10,8
         dc.w    4,2,1
 
-pre_roll_off_table:
-        dc.w    $400,$200,$180,$140,$100,$C0,$A0,$80,$78,$74,$6E
-        dc.w    $69,$64,$5A,$46,$40,$38,$30,$28,$20,$1F,$1E,$1D
-        dc.w    $1C,$1B,$1A,$19,$18,$17,$16,$15,$14,$13,$12,$11
-        dc.w    $10,15,14,13,13,12,12,11,11,10,10,9,9,8,8,8,8,7,7
-        dc.w    7,7,6,6,6,6,5,5,5,5,4,4,4,4,4,4,4,4,4,4,3,4,4,3,4
-        dc.w    4,3,4,3,4,3,4,3,4,3,3,3,3,3,3,3,3,3,2,3,3,3,2,3,3
-        dc.w    2,3,3,2,3,3,2,3,2,3,2,3,2,3,2,3,2,2,2,2,2,2,2,2,1
-        dc.w    2,1,2,1,2,1,2,1,1,2,1,1,1,2,1
-
 pre_octave_note_offset_table:
         dc.b    1*NOTES_IN_OCTAVE*4,1*NOTES_IN_OCTAVE*4,1*NOTES_IN_OCTAVE*4
         dc.b    2*NOTES_IN_OCTAVE*4,2*NOTES_IN_OCTAVE*4,2*NOTES_IN_OCTAVE*4
@@ -4179,14 +2574,6 @@ pre_minus4plus4_table:
         dc.b     $FC,$FB,$FF,1,2,3,4,0
         ENDC
 
-pre_ramp_up_down_32:
-        dc.w     0*32,1*32,2*32,3*32,4*32,5*32,6*32,7*32,8*32,9*32,10*32,11*32,12*32,13*32,14*32,15*32
-        dc.w     15*32,14*32,13*32,12*32,11*32,10*32,9*32,8*32,7*32,6*32,5*32,4*32,3*32,2*32,1*32,0*32
-
-pre_modulator_ramp_8:
-        ;dc.w     77,293,539,1079,1337,1877,2431,3031 ; the 1079 value is strange (938 better?)
-        dc.w     $4D,$125,$21B,$437,$539,$755,$96D,$BD7
-
 pre_period_table:
         dc.w     $350,$320,$2F2,$2C8,$2A0,$279,$256,$236,$216,$1F8,$1DC,$1C0
         dc.w     $1A8,$190,$179,$164,$151,$13E,$12C,$11B,$10B,$FC,$EE,$E0
diff --git a/src/raspberry_casket.i b/src/raspberry_casket.i
new file mode 100644
index 0000000..09fc2c6
--- /dev/null
+++ b/src/raspberry_casket.i
@@ -0,0 +1,344 @@
+; Pretracker song format description:
+;
+; $0000  4: PRT<version> ($19 (V0.x), $1b (V1.0), $1e (V1.5))
+; $0004  4: File offset to position data (POSD)
+; $0008  4: File offset to pattern data (PATT)
+; $000C  4: File offset to instruments (INST)
+; $0010  4: File offset to waves (WAVE)
+; $0014 20: Songname
+; $0028 20: Author
+; $003C  1: Restart position for song (<=V1.0)
+; $003D  1: Number of patterns (<=V1.0)
+; $003E  1: Songlength in patterns (<=V1.0)
+; $003F  1: Number of steps per pattern (<=V1.0)
+; $0040  1: Number of instruments (<=V1.0), $40 for V1.5
+; $0041  1: Number of waves
+; $0042 24: Wave generation ordering (>=V1.0)
+; $005a  1: Number of subsongs (>=V1.5)
+; [...]
+; Position data (POSD):
+; - Per subsong (>=V1.5)
+;   - 1: Restart pos
+;   - 1: #patterns
+;   - 1: #numsteps
+;   - 1: songlength
+;   - 4: relative pattern offset in pattern data
+; - Positions x 4 x [Pattern number (byte), pitch shift (signed byte)]
+;
+; Pattern data (PATT):
+; Each pattern line consists of three bytes:
+;   - 1 Bit : Bit 4 of Inst Number
+;   - 1 Bit : ARP instead of effect
+;   - 6 Bits: Pitch ($01 is C-0, $3d is NOTE OFF)
+;
+;   - 4 Bits: Bit 0-3 of Inst Number
+;   - 4 Bits: Effect command
+;   - 8 Bits: Effect data
+; - Patterns x steps x 3 bytes
+;
+; Unknown data after pattern data: (12 10 00 00 00 00)
+;
+; Instrument definitions (INST):
+; - 32 (!) x Null terminated string (or 23 chars max) (for V1.5 this is 64 strings)
+; - For each instrument: Instrument Info (ININ)
+;   - $00  1: Vibrato Delay
+;   - $01  1: Vibrato Depth
+;   - $02  1: Vibrato Speed (-1)
+;   - $03  1: ADSR Attack
+;   - $04  1: ADSR Decay
+;   - $05  1: ADSR Sustain
+;   - $06  1: ADSR Release
+;   - $07  1: Number of Inst Pattern steps
+;
+; - For each instrument:
+;   - 1 Byte: number of steps
+;   - 3 Bytes x number of steps: Inst Pattern (IPTT)
+;
+; Inst pattern data (IPTT):
+; Each pattern line consists of three bytes:
+;   - 1 Bit  : Next note stitched to this one
+;   - 1 Bit  : Fixed Pitch Note
+;   - 6 Bits : Pitch ($01 is C-0)
+;   - 4 Bits : unused?
+;   - 12 Bits: Effect
+; - Patterns x steps x 3 bytes
+;
+; Wave definitions (WAVE):
+; - 24 (!) x Null terminated string (or 23 chars max)
+; - Optional padding to even address, if necessary
+; - For each wave:
+;   - 42 Bytes: Wave info structure (see definition below)
+
+; ----------------------------------------
+; Some constants for clarity
+
+MAX_VOLUME      = $40
+MAX_SPEED       = $2f
+MAX_WAVES       = 24
+MAX_INSTRUMENTS = 32
+MAX_TRACK_DELAY = 32
+NOTE_OFF_PITCH  = $3d
+NOTES_IN_OCTAVE = 12
+NUM_CHANNELS    = 4     ; yes, you can reduce the number of channels if you want
+
+; ----------------------------------------
+; Pretracker file structures
+
+; Pattern data (PATT and IPTT)
+pdb_pitch_ctrl      = 0
+pdb_inst_effect     = 1 ; for normal pattern data
+pdb_effect_cmd      = 1 ; for inst pattern
+pdb_effect_data     = 2
+
+; Pattern pos data (POSD)
+ppd_pat_num         = 0
+ppd_pat_shift       = 1
+
+; Instrument Info (ININ)
+ii_vibrato_delay    = 0
+ii_vibrato_depth    = 1
+ii_vibrato_speed    = 2
+ii_adsr_attack      = 3
+ii_adsr_decay       = 4
+ii_adsr_sustain     = 5
+ii_adsr_release     = 6
+ii_pattern_steps    = 7
+ii_SIZEOF           = 8
+
+; Wave Info (WAVE)
+wi_loop_start_w    = $00
+wi_loop_end_w      = $02
+wi_subloop_len_w   = $04
+wi_allow_9xx_b     = $06
+wi_subloop_wait_b  = $07
+wi_subloop_step_w  = $08
+wi_chipram_w       = $0a
+wi_loop_offset_w   = $0c
+wi_chord_note1_b   = $0e
+wi_chord_note2_b   = $0f
+wi_chord_note3_b   = $10
+wi_chord_shift_b   = $11
+wi_osc_unknown_b   = $12 ; always $00? (unused in code?)
+wi_osc_phase_spd_b = $13
+wi_flags_b         = $14 ; bit 0/1: osc type, bit 2: needs extra octaves, bit 3: boost, bit 4: pitch linear, bit 5: vol fast
+wi_osc_phase_min_b = $15
+wi_osc_phase_max_b = $16
+wi_osc_basenote_b  = $17
+wi_osc_gain_b      = $18
+wi_sam_len_b       = $19 ; in multiples of 128, zero-based (0 == 128)
+wi_mix_wave_b      = $1a
+wi_vol_attack_b    = $1b
+wi_vol_delay_b     = $1c
+wi_vol_decay_b     = $1d
+wi_vol_sustain_b   = $1e
+wi_flt_type_b      = $1f ; 1=lowpass, 2=highpass, 3=bandpass, 4=notch
+wi_flt_resonance_b = $20
+wi_pitch_ramp_b    = $21
+wi_flt_start_b     = $22
+wi_flt_min_b       = $23
+wi_flt_max_b       = $24
+wi_flt_speed_b     = $25
+wi_mod_params_l    = $26
+wi_mod_wetness_b   = $26
+wi_mod_length_b    = $27
+wi_mod_predelay_b  = $28
+wi_mod_density_b   = $29 ; (1-7), unisono (bits 3/4) and post bit 5
+wi_SIZEOF          = $2a
+
+; ----------------------------------------
+; Unpacked Instrument Info (addition to player for faster processing)
+                    rsreset
+uii_vibrato_delay   rs.w    1
+uii_vibrato_depth   rs.w    1
+uii_vibrato_speed   rs.w    1
+uii_adsr_release    rs.b    1
+                    rs.b    1 ; dummy
+uii_adsr_attack     rs.w    1
+uii_adsr_decay      rs.w    1
+uii_adsr_sustain    rs.w    1
+uii_pattern_steps   rs.b    1
+                    rs.b    1 ; padding
+uii_SIZEOF          rs.b    0
+
+; ----------------------------------------
+; MySong offsets
+                        rsreset
+sv_waveinfo_table       rs.l    MAX_WAVES ; 24 pointers to wave infos to avoid mulu
+sv_inst_patterns_table  rs.l    MAX_INSTRUMENTS ; 32 pointers to pattern data
+                        ; --- 127 byte displacement limit ---
+sv_wavelength_table     rs.l    MAX_WAVES ; 24 longwords to sample lengths (standard octave) (NEW)
+sv_wavetotal_table      rs.l    MAX_WAVES ; 24 longwords to sample lengths for all octaves (NEW)
+sv_wavegen_order_table  rs.b    MAX_WAVES ; 24 bytes
+sv_num_waves_b          rs.b    1
+sv_num_steps_b          rs.b    1
+sv_patterns_ptr         rs.l    1
+sv_pat_pos_len_w        rs.w    1 ; only byte used
+sv_pat_restart_pos_w    rs.w    1 ; only byte used
+sv_pos_data_adr         rs.l    1
+sv_waveinfo_ptr         rs.l    1 ; base pointer of wave info
+sv_pattern_table        rs.l    256
+sv_inst_infos_table     rs.b    MAX_INSTRUMENTS*uii_SIZEOF
+sv_SIZEOF               rs.b    0
+
+; ----------------------------------------
+; channel output data (part of pcd structure below)
+                rsreset
+ocd_sam_ptr     rs.l    1   ; 0
+ocd_length      rs.w    1   ; 4
+ocd_loop_offset rs.w    1   ; 6
+ocd_period      rs.w    1   ; 8
+ocd_volume      rs.b    1   ; 10
+ocd_trigger     rs.b    1   ; 11 needs to be after volume
+ocd_unused      rs.l    1   ; 12 unused, but makes the structure an even 16 bytes
+ocd_SIZEOF      rs.b    0
+
+; channel structure (part of pv structure)
+                            rsreset
+; DO NOT CHANGE ORDER -- OPTIMIZED CLEARING
+pcd_pat_portamento_dest_w   rs.w    1 ; portamento destination pitch
+pcd_pat_pitch_slide_w       rs.w    1
+
+pcd_pat_vol_ramp_speed_b    rs.b    1
+pcd_pat_2nd_inst_num4_b     rs.b    1
+pcd_pat_2nd_inst_delay_b    rs.b    1
+pcd_wave_offset_b           rs.b    1
+
+pcd_inst_pitch_slide_w      rs.w    1
+pcd_inst_sel_arp_note_w     rs.w    1
+
+pcd_inst_note_pitch_w       rs.w    1
+pcd_inst_curr_port_pitch_w  rs.w    1
+
+pcd_inst_line_ticks_b       rs.b    1
+pcd_inst_pitch_pinned_b     rs.b    1
+pcd_inst_vol_slide_b        rs.b    1
+pcd_inst_step_pos_b         rs.b    1
+
+pcd_inst_wave_num4_w        rs.w    1 ; current wave number (1 based, times 4) (lower byte used)
+
+pcd_track_delay_offset_b    rs.b    1 ; $ff = no track delay
+pcd_inst_speed_stop_b       rs.b    1 ; speed byte, $ff stops processing
+pcd_inst_pitch_w            rs.w    1
+
+pcd_inst_vol_w              rs.w    1
+pcd_loaded_inst_vol_b       rs.b    1
+pcd_pat_vol_b               rs.b    1 ; Multiplied with volume of instrument.
+; DO NOT CHANGE ORDER -- OPTIMIZED CLEARING END
+
+pcd_arp_notes_l             rs.b    0
+pcd_arp_note_1_b            rs.b    1
+pcd_arp_note_2_b            rs.b    1
+pcd_arp_note_3_b            rs.b    1
+                            rs.b    1 ; gets cleared
+
+pcd_last_trigger_length_w   rs.w    1 ; I think this makes sure that we trigger the note if the length changes
+
+pcd_pat_portamento_speed_b  rs.b    1
+pcd_pat_adsr_rel_delay_b    rs.b    1 ; counts down until adsr release. Seems unused?
+pcd_note_off_delay_b        rs.b    1 ; time before note is released ($ff = disabled)
+pcd_inst_pattern_steps_b    rs.b    1 ; number of steps in instrument pattern
+
+pcd_note_delay_b            rs.b    1 ; $ff = no note delay
+pcd_track_delay_steps_b     rs.b    1 ; $00 = no track delay, $ff = stop track delay (this is for the next channel!)
+pcd_track_delay_vol16_b     rs.b    1
+pcd_track_init_delay_b      rs.b    1 ; number of frames to ignore the delay
+
+pcd_inst_num4_w             rs.w    1 ; current instrument number * 4
+;pcd_inst_new_step_w         rs.w    1 ; seems to be unused
+pcd_inst_subloop_wait_w     rs.w    1
+pcd_inst_loop_offset_w      rs.w    1
+pcd_inst_info_ptr           rs.l    1 ; pointer to currently active instrument
+
+pcd_waveinfo_ptr            rs.l    1 ; pointer to currently active waveinfo
+pcd_channel_mask_b          rs.b    1
+pcd_channel_num_b           rs.b    1
+pcd_adsr_phase_w            rs.w    1 ; 0=attack, 1=decay, 2=sustain, 3=release ! do not change order
+pcd_adsr_volume_w           rs.w    1 ; 0 for restart / $400 (word only) ! do not change order
+pcd_adsr_phase_speed_b      rs.b    1
+pcd_inst_ping_pong_dir_b    rs.b    1 ; direction of ping-pong (-1 == $00 / +1 = $ff)
+pcd_adsr_pos_w              rs.w    1 ; pos in adsr curve
+pcd_adsr_vol64_w            rs.w    1 ; some adsr volume
+
+pcd_new_inst_num_b          rs.b    1 ; load new instrument (number) ! do not change order
+                            rs.b    1 ; gets cleared
+pcd_vibrato_pos_w           rs.w    1 ;
+pcd_vibrato_delay_w         rs.w    1 ; is a byte value ! do not change order
+pcd_vibrato_depth_w         rs.w    1 ; is a byte value ! do not change order
+pcd_vibrato_speed_w         rs.w    1 ; is a byte value ! do not change order
+pcd_adsr_release_b          rs.b    1 ; is a byte value ! do not change order
+                            rs.b    1 ; padding will be overwritten!
+
+pcd_out_base                rs.b    ocd_SIZEOF
+pcd_track_delay_buffer      rs.b    MAX_TRACK_DELAY*ocd_SIZEOF
+pcd_SIZEOF                  rs.b    0
+
+pcd_out_ptr_l       = pcd_out_base+ocd_sam_ptr
+pcd_out_len_w       = pcd_out_base+ocd_length
+pcd_out_lof_w       = pcd_out_base+ocd_loop_offset
+pcd_out_per_w       = pcd_out_base+ocd_period
+pcd_out_vol_b       = pcd_out_base+ocd_volume
+pcd_out_trg_b       = pcd_out_base+ocd_trigger
+pcd_out_unused_l    = pcd_out_base+ocd_unused          ; copied for track delay, but not used?
+
+                rsreset
+owb_saw_waves   rs.b    128
+owb_sqr_waves   rs.b    128
+owb_tri_waves   rs.b    128
+owb_wave_length rs.b    1
+owb_SIZEOF      rs.b    0
+
+; ----------------------------------------
+; MyPlayer global variables (not bound to channel)
+                        rsreset
+; DO NOT CHANGE ORDER -- OPTIMIZED INIT
+pv_pat_curr_row_b       rs.b    1 ; current step
+pv_next_pat_row_b       rs.b    1
+pv_next_pat_pos_b       rs.b    1
+pv_pat_speed_even_b     rs.b    1 ; even shuffle speed
+
+pv_pat_speed_odd_b      rs.b    1 ; odd shuffle speed
+pv_pat_line_ticks_b     rs.b    1
+pv_pat_stopped_b        rs.b    1 ; 0 = stop, $ff = run
+pv_songend_detected_b   rs.b    1
+; DO NOT CHANGE ORDER -- OPTIMIZED INIT END
+
+pv_curr_pat_pos_w       rs.w    1 ; only byte used
+pv_loop_pattern_b       rs.b    1 ; repeat current pattern, do not advance
+                        rs.b    1 ; padding
+
+pv_trigger_mask_w       rs.w    1
+
+pv_my_song              rs.l    1
+pv_copperlist_ptr       rs.l    1
+pv_sample_buffer_ptr    rs.l    1 ; pointer to start of sample buffer
+pv_wave_sample_table    rs.l    MAX_WAVES ; 24 pointers to sample starts
+pv_period_table         rs.w    16*NOTES_IN_OCTAVE*3
+                        ; --- 127 byte displacement limit ---
+pv_channeldata          rs.b    NUM_CHANNELS*pcd_SIZEOF
+
+                        IFNE    PRETRACKER_VOLUME_TABLE
+pv_osc_buffers          rs.b    0 ; reuse space of volume_table, which is bigger than NOTES_IN_OCTAVE*owb_SIZEOF
+pv_volume_table         rs.b    (MAX_VOLUME+1)*MAX_VOLUME*2
+                        ELSE
+pv_osc_buffers          rs.b    NOTES_IN_OCTAVE*owb_SIZEOF
+                        ENDC
+
+pv_precalc_sample_size  rs.l    1
+pv_precalc_progress_ptr rs.l    1
+pv_wg_wave_counter_w    rs.w    1
+                        IFNE    PRETRACKER_PARANOIA_MODE    ; same wave for mixing cannot be selected in Pretracker
+pv_wg_curr_wave_num_b   rs.b    1
+                        rs.b    1
+                        ENDC
+pv_wg_curr_sample_ptr   rs.l    1
+pv_wg_curr_samend_ptr   rs.l    1
+pv_wg_curr_sample_len_l rs.w    1
+pv_wg_curr_sample_len_w rs.w    1
+pv_wg_chord_note_num_b  rs.b    1 ; don't change order
+pv_wg_unisono_run_b     rs.b    1 ; don't change order
+pv_wg_chord_flag_w      rs.w    1
+pv_wg_chord_pitches     rs.l    1
+pv_wg_osc_speed_l       rs.l    1
+pv_wg_flt_taps          rs.w    4
+pv_SIZEOF               rs.b    0
+
diff --git a/src/raspberry_casket_wavegen.asm b/src/raspberry_casket_wavegen.asm
new file mode 100644
index 0000000..b41e2fa
--- /dev/null
+++ b/src/raspberry_casket_wavegen.asm
@@ -0,0 +1,1297 @@
+
+CLIPTO8BIT MACRO
+        cmpi.w  #-$80,\1
+        bge.s   .nominclip\@
+        moveq.l #-$80,\1
+.nominclip\@
+        cmpi.w  #$7F,\1
+        ble.s   .nomaxclip\@
+        moveq.l #$7F,\1
+.nomaxclip\@
+        ENDM
+
+CLIPORTRUNC8BIT MACRO
+        beq.s   .unboosted\@
+        asr.l    #6,\1
+        cmpi.w  #-$80,\1
+        bge.s   .nominclip\@
+        moveq.l #-$80,\1
+.nominclip\@
+        cmpi.w  #$7F,\1
+        ble.s   .nomaxclip\@
+        moveq.l #$7F,\1
+        bra.s   .nomaxclip\@
+.unboosted\@
+        asr.l   #8,\1
+.nomaxclip\@
+        ENDM
+
+CLIPTO8BITAFTERADD MACRO
+        bvc.s   .noclip\@
+        spl     \1
+        eor.b   #$7f,\1
+.noclip\@
+        ENDM
+
+; ----------------------------------------
+; proposed register assignment:
+; a1 = Wave table order
+; a4 = MyPlayer
+; a6 = MySong / waveinfo
+pre_WaveGen:
+; ----------------------------------------
+
+        lea     pre_log12_table(pc),a0   ; 128, 121, 114, 107, 102, 96, 90, 85, 80, 76, 72, 67
+        lea     pv_osc_buffers+owb_sqr_waves(a4),a3
+        moveq.l #NOTES_IN_OCTAVE-1,d7
+.noteloop
+        swap    d7
+        moveq.l #0,d6
+        move.w  d6,d7               ; tabpos
+        move.b  (a0)+,d6            ; period
+        move.b  d6,owb_wave_length-owb_sqr_waves(a3)
+
+        move.l  #$ff00,d5
+        divu    d6,d5               ; frac increment
+
+        move.w  d6,d4
+        lsr.w   #1,d4               ; half-period
+        move.w  d4,d3
+        lsr.w   #1,d3               ; quarter-period
+
+        moveq.l #0,d0               ; acc
+        lea     (a3,d6.w),a2
+        lea     owb_tri_waves-owb_sqr_waves(a2),a2
+        suba.w  d3,a2
+.notewaveloop
+        move.w  d0,d2
+        lsr.w   #8,d2
+
+        moveq.l #$7f,d1
+        sub.b   d2,d1
+        move.b  d1,owb_saw_waves-owb_sqr_waves(a3,d7.w)
+
+        add.b   d2,d2
+        cmp.w   d7,d3               ; tabpos == negquarter
+        bne.s   .nowrapback
+        suba.w  d6,a2               ; go back to start of period
+.nowrapback
+        cmp.w   d0,d7
+        ble.s   .otherhalf
+        moveq.l #$7f,d1
+        sub.b   d2,d1
+        move.b  d1,(a2)+
+        bra.s   .clip80
+
+.otherhalf
+        add.b   #$80,d2
+        move.b  d2,(a2)+
+        moveq.l #$7f,d2
+        cmp.w   d7,d4
+        bne.s   .noclip80
+.clip80 moveq.l #-$80,d2
+.noclip80
+        move.b  d2,owb_sqr_waves-owb_sqr_waves(a3,d7.w)
+
+        add.w   d5,d0               ; increment acc by frac
+        addq.w  #1,d7               ; increment pos
+
+        cmp.w   d7,d6
+        bne.s   .notewaveloop
+
+        swap    d7
+        lea     owb_SIZEOF(a3),a3
+        dbra    d7,.noteloop
+
+; ----------------------------------------
+; proposed register assignment:
+; a0 = sample output
+; a1 = scratch
+; a3 = waveinfo
+; a4 = MyPlayer
+; a6 = MySong / waveinfo
+.wavegenloop
+        ;movea.l  pv_my_song(a4),a6
+        moveq.l #0,d1
+        move.b  (a1)+,d1        ; apply wave order redirection
+        move.l  a1,-(sp)        ; what used to be sv_wavegen_order_table
+        IFNE    PRETRACKER_PARANOIA_MODE    ; same wave for mixing cannot be selected in Pretracker
+        move.b  d1,pv_wg_curr_wave_num_b(a4)
+        ENDC
+        lsl.w   #2,d1
+        move.l  pv_wave_sample_table(a4,d1.w),a0
+        move.l  a0,pv_wg_curr_sample_ptr(a4)
+
+        IFND    PRESTO_UNIFIED_STRUCT
+        add.w   #sv_wavelength_table,d1
+        adda.w  d1,a6
+        move.l  sv_wavelength_table-sv_wavelength_table(a6),d0
+        move.l  d0,pv_wg_curr_sample_len_l(a4)
+        IFNE    PRETRACKER_PROGRESS_SUPPORT
+        move.l  sv_wavetotal_table-sv_wavelength_table(a6),pv_precalc_sample_size(a4)
+        ENDC
+        move.l  sv_waveinfo_table-sv_wavelength_table(a6),a3
+        ELSE    ; PRESTO_UNIFIED_STRUCT
+        ; Presto has no MySong structure
+        lea     pv_wavelength_table(a4,d1.w),a6
+
+        ; copied from raspberry_casket
+        moveq.l #0,d0
+        move.b  wi_sam_len_b(a3),d0
+        addq.w  #1,d0
+        lsl.w   #7,d0
+        move.l  d0,pv_wavelength_table-pv_wavelength_table(a6)
+        move.l  d0,d1
+        btst    #2,wi_flags_b(a3)
+        beq.s   .onlythreeocts
+        mulu    #15,d1
+        lsr.l   #3,d1           ; * 1.875
+.onlythreeocts
+        move.l  d1,pv_wavetotal_table-pv_wavelength_table(a6)
+        ; end stuff
+        move.l  d0,pv_wg_curr_sample_len_l(a4)
+        IFNE    PRETRACKER_PROGRESS_SUPPORT
+        move.l  d1,pv_precalc_sample_size(a4)
+        ENDC
+        move.l  pv_waveinfo_table-pv_wavelength_table(a6),a3
+        ENDC    ; PRESTO_UNIFIED_STRUCT
+
+        ; clear sample data (a0 and d0 from above)
+        bsr     pre_MemClr
+        move.l  a0,pv_wg_curr_samend_ptr(a4)
+
+        ; read out chord information
+        lea     wi_chord_note1_b(a3),a1
+        move.b  (a1)+,d1
+        move.b  (a1)+,d2
+        move.b  (a1)+,d3
+
+        moveq.l #0,d4
+        move.b  d1,d4
+        or.b    d2,d4
+        or.b    d3,d4
+        seq     d4
+        neg.b   d4
+        move.w  d4,pv_wg_chord_flag_w(a4)   ; has chord flag (0/1)
+
+        move.b  wi_osc_basenote_b(a3),d0
+        add.b   d0,d1
+        add.b   d0,d2
+        add.b   d0,d3
+        lea     pv_wg_chord_pitches(a4),a1
+        move.b  d0,(a1)+
+        move.b  d1,(a1)+
+        move.b  d2,(a1)+
+        move.b  d3,(a1)+
+
+        suba.l  a5,a5
+        clr.w   pv_wg_chord_note_num_b(a4)  ; and pv_wg_chord_note_num_b
+
+.wavegen_chordloop
+        lea     pv_wg_chord_pitches(a4),a1
+        moveq.l #0,d1
+        move.b  pv_wg_chord_note_num_b(a4),d1   ; chord note counter
+        move.b  (a1,d1.w),d0    ; get chord note
+        ext.w   d0
+
+        tst.w   d1
+        beq.s   .base_note_is_never_skipped ; is base note?
+        cmp.b   wi_osc_basenote_b(a3),d0
+        beq     .wave_gen_tone_done      ; skip chord notes that are same as base note
+.base_note_is_never_skipped
+        moveq.l #0,d1
+        moveq.l #NOTES_IN_OCTAVE,d2
+        move.w  d0,d1
+        move.w  d0,a2           ; save base note, used later (much later, noise generator)!
+        add.w   #NOTES_IN_OCTAVE*NOTES_IN_OCTAVE,d1 ; make sure we don't run into negative modulo
+        divu    d2,d1
+        swap    d1
+        move.w  d1,d0           ; note within octave
+        swap    d1
+        sub.w   d2,d1           ; restore octave
+
+        mulu    #owb_SIZEOF,d0
+        lea     (a4,d0.w),a1
+
+        lea     pv_osc_buffers+owb_saw_waves(a1),a6
+        moveq.l #3,d0
+        and.b   wi_flags_b(a3),d0
+        beq.s   .osc_selected
+        lea     owb_tri_waves-owb_saw_waves(a6),a6
+        subq.b  #1,d0
+        beq.s   .osc_selected
+        lea     owb_sqr_waves-owb_tri_waves(a6),a6
+        subq.b  #1,d0
+        beq.s   .osc_selected
+        suba.l  a6,a6           ; noise selected
+.osc_selected
+
+        move.l  #$8000,d6
+        move.w  d1,d0           ; check octave shift
+        bgt.s   .shiftleft
+        beq.s   .contshift
+;.shiftright
+        move.w  d0,d3
+        neg.w   d3
+        asr.l   d3,d6
+        bra.s   .contshift
+.shiftleft
+        lsl.l   d0,d6
+.contshift
+
+; ----------------------------------------
+; pitch ramp
+        move.b  wi_pitch_ramp_b(a3),d3
+        ext.w   d3
+        ext.l   d3
+        btst    #4,wi_flags_b(a3)   ; pitch linear flag
+        beq.s   .pitch_not_linear
+        tst.b   d3
+        bgt.s   .pitch_ramp_positive
+
+        lsl.l   d0,d3
+        add.l   d3,d3
+        bra.s   .pitch_ramp_cont
+
+.pitch_not_linear
+        tst.b   d3
+        ble.s   .pitch_ramp_cont
+.pitch_ramp_positive
+        muls    d3,d3
+.pitch_ramp_cont
+        move.l  d3,d2
+        lsl.l   #8,d2
+        lsl.l   #2,d2
+
+        moveq.l #0,d7
+        move.b  pv_osc_buffers+owb_wave_length(a1),d7   ; get period
+        moveq.l #15,d5
+        lsl.l   d5,d7
+
+        sub.w   d0,d5       ; 15-octave
+        lsl.w   #3,d5
+
+        moveq.l #0,d3
+        move.b  wi_osc_phase_min_b(a3),d3
+        mulu    d5,d3
+        lsl.l   #6,d3
+
+        moveq.l #0,d0
+        move.b  wi_osc_phase_max_b(a3),d0
+        mulu    d5,d0
+        lsl.l   #6,d0
+        move.l  d0,a5
+
+        moveq.l #0,d5
+        move.b  wi_osc_phase_spd_b(a3),d5
+        lsl.l   #8,d5
+        lsl.l   #3,d5
+
+        cmp.l   d3,d0
+        bge.s   .osc_with_positive_phase_speed
+        neg.l   d5
+
+        movea.l d3,a5
+        bra.s   .osc_continue
+.osc_with_positive_phase_speed
+        move.l  d3,d0
+
+.osc_continue
+        move.l  d0,pv_wg_osc_speed_l(a4)
+
+        ; d0 = d6 * chord_shift * chordnum + d6 * phase_min = d6 * (chord_shift * chordnum + phase_min)
+
+        moveq.l #0,d4
+        move.b  wi_chord_shift_b(a3),d4
+        move.w  pv_wg_chord_flag_w(a4),d0
+        add.b   pv_wg_chord_note_num_b(a4),d0
+        mulu    d0,d4
+        moveq.l #0,d0
+        move.b  wi_osc_phase_min_b(a3),d0
+        add.w   d0,d4
+        move.l  d6,d0
+        lsr.l   #4,d0
+        lsl.l   #4,d4
+        mulu    d4,d0
+
+        cmp.l   d7,d0
+        ble.s   .lbC002516
+.lbC002510
+        sub.l   d7,d0
+        cmp.l   d7,d0
+        bgt.s   .lbC002510
+.lbC002516
+        move.l  a6,d4
+        bne     .no_noise
+
+; ----------------------------------------
+; d1 = octave
+; d2 = pitch ramping value
+; d5 = osc phase speed
+; a2 = base note
+.gen_noise
+        IFNE    PRETRACKER_PARANOIA_MODE
+        tst.w   pv_wg_curr_sample_len_w(a4)
+        beq     .wave_gen_tone_done
+        ENDC
+
+        move.l  #$8000,d5
+        move.l  d5,a5
+        tst.w   d1
+        bge.s   .gen_noise_positive_octave
+
+.gen_noise_negative_octave
+        moveq.l #NOTES_IN_OCTAVE,d1
+        move.l  a2,d3
+        neg.l   d3
+        move.l  d3,d0
+        divu    d1,d0
+        IFNE    PRETRACKER_PARANOIA_MODE
+        bvs.s   .divisionoverflow
+        ENDC
+        addq.w  #1,d0
+        lsr.w   d0,d5
+        moveq.l #0,d0
+        move.w  d5,d0
+        divu    d1,d0
+        moveq.l #0,d4
+        move.w  d0,d4
+        IFNE    PRETRACKER_PARANOIA_MODE
+        bra.s   .returnfromoverflow
+.divisionoverflow
+        move.l  #$AAA,d4        ; some dummy value, I would expect
+.returnfromoverflow
+        ENDC
+        moveq.l #0,d0
+        move.w  d3,d0
+.cheap_mod12
+        sub.w   d1,d0
+        bpl.s   .cheap_mod12
+        neg.w   d0
+        mulu    d4,d0
+        add.l   d0,d5
+
+.gen_noise_positive_octave
+        moveq.l #0,d0
+        move.b  wi_osc_phase_min_b(a3),d0
+        moveq.l #0,d1
+        move.b  wi_chord_shift_b(a3),d1
+        add.w   d1,d0
+        addq.w  #1,d0
+
+        movea.l d5,a1
+        movea.l pv_wg_curr_sample_ptr(a4),a0
+        moveq.l #0,d1
+        moveq.l #0,d6
+        moveq.l #0,d3
+        move.b  wi_osc_gain_b(a3),d3
+
+.gen_noise_outerloop
+        move.w  d0,d1           ; random noise generator
+        lsl.w   #8,d1
+        lsl.w   #5,d1
+        eor.w   d1,d0
+
+        move.w  d0,d1
+        lsr.w   #8,d1
+        lsr.w   #1,d1
+        eor.w   d1,d0
+
+        move.w  d0,d1
+        lsl.w   #7,d1
+        eor.w   d1,d0
+
+        cmpa.l  #$8000,a5       ; if symmetrical
+        beq.s   .gen_noise_centered
+        ; FIXME what does this do? d4 = (a5 - $8000) d1 = (a5 + $7fff)&$ffff8000 -> d4 - d1 == ((a5 - $8000) - ((a5 + $7fff)&$ffff8000)
+        move.l  a5,d4
+        addi.l  #$FFFF8000,d4
+        move.l  a5,d1
+        addi.l  #$FFFF7FFF,d1
+        andi.w  #$8000,d1
+        movea.l d4,a5
+        suba.l  d1,a5
+.gen_noise_centered
+
+        move.b  d0,d1
+        ext.w   d1
+        move.w  d3,d4
+        muls    d1,d4
+        move.l  d4,d1
+        asr.l   #7,d1
+        CLIPTO8BIT d1
+        add.b   (a0),d1
+
+.gen_noise_innerloop
+        move.b  d1,(a0)+
+        adda.l  a1,a5
+        tst.l   d2
+        beq.s   .gen_noise_no_pitch_ramping
+        add.l   d2,d6
+        move.l  d6,d4
+        asr.l   #8,d4
+        asr.l   #2,d4
+        add.l   d5,d4
+        movea.l d4,a1
+
+        btst    #4,wi_flags_b(a3)   ; pitch linear flag
+        beq.s   .noise_nonlinear_pitch
+        move.l  d2,d4
+        asr.l   #7,d4
+        sub.l   d4,d2
+.noise_nonlinear_pitch
+
+        cmpa.w  #$1FF,a1
+        bgt.s   .gen_noise_no_pitch_ramping
+        moveq.l #0,d2
+        move.l  d2,a5
+        ;suba.l   a5,a5
+        movea.w #$200,a1
+.gen_noise_no_pitch_ramping
+        cmpa.l  pv_wg_curr_samend_ptr(a4),a0
+        beq     .wave_gen_tone_done
+        cmpa.w  #$7FFF,a5
+        ble.s   .gen_noise_innerloop
+        bra     .gen_noise_outerloop
+
+.no_noise
+        move.l  d6,d1
+        tst.b   pv_wg_unisono_run_b(a4)
+        beq.s   .is_not_in_unisono
+
+        moveq.l #3<<3,d1
+        and.b   wi_mod_density_b(a3),d1
+        lsr.b   #3,d1
+        moveq.l #9,d4
+        sub.b   d1,d4
+
+        move.l  d6,d1
+        asr.l   d4,d1
+        add.l   d6,d1
+.is_not_in_unisono
+
+        IFNE    PRETRACKER_PARANOIA_MODE
+        tst.w   pv_wg_curr_sample_len_w(a4)
+        beq     .wave_gen_tone_done
+        ENDC
+
+; ----------------------------------------
+; chord gen
+; in: d0/d1/d2/d3/d5/d7
+; in: a0
+        move.l  d3,a1
+        movea.l pv_wg_curr_sample_ptr(a4),a0
+        suba.l  a2,a2
+
+.chordtoneloop
+        move.l  d0,d4
+        sub.l   a1,d4
+        bpl.s   .noclip_osc_phase
+        moveq.l #0,d4
+.noclip_osc_phase
+        asr.l   #8,d4
+        asr.l   #7,d4
+        move.b  (a6,d4.w),d4        ; fetch precalced sample
+        ext.w   d4
+
+        moveq.l #0,d3
+        move.b  wi_osc_gain_b(a3),d3
+        muls    d4,d3
+        asr.w   #7,d3
+
+        move.b  (a0),d4
+        ext.w   d4
+        add.w   d3,d4
+        CLIPTO8BIT d4
+        move.b  d4,(a0)+
+
+        add.l   d1,d0
+        cmp.l   d7,d0
+        blt.s   .lbC0025A2
+        sub.l   d7,d0
+        add.l   d5,a1
+        cmp.l   a5,a1
+        blt.s   .lbC00259A
+        neg.l   d5
+        move.l  a5,a1
+.lbC00259A
+        cmp.l   pv_wg_osc_speed_l(a4),a1
+        bgt.s   .lbC0025A2
+        neg.l   d5
+        move.l  pv_wg_osc_speed_l(a4),a1
+.lbC0025A2
+        tst.l   d2
+        beq.s   .chordtone_done
+        adda.l  d2,a2
+        move.l  a2,d1
+        asr.l   #8,d1
+        asr.l   #2,d1
+        add.l   d6,d1
+
+        btst    #4,wi_flags_b(a3)   ; pitch linear flag
+        beq.s   .no_linear_pitch
+        move.l  d2,d4
+        asr.l   #7,d4
+        sub.l   d4,d2
+.no_linear_pitch
+        cmp.l   d7,d1
+        bcs.s   .chordtone_done
+        moveq.l #0,d2
+        moveq.l #0,d1
+.chordtone_done
+        cmpa.l  pv_wg_curr_samend_ptr(a4),a0
+        bne.s   .chordtoneloop
+
+.wave_gen_tone_done
+        addq.b  #1,pv_wg_chord_note_num_b(a4)
+        cmp.b   #4,pv_wg_chord_note_num_b(a4)
+        bne     .wavegen_chordloop
+
+        moveq.l #3<<3,d0
+        and.b   wi_mod_density_b(a3),d0 ; unisono
+        beq.s   .chords_done
+        move.l  a6,d1
+        beq.s   .chords_done
+        tst.b   pv_wg_unisono_run_b(a4)
+        bne.s   .chords_done
+        move.w  #$0001,pv_wg_chord_note_num_b(a4) ; sets also pv_wg_unisono_run_b
+        bra     .wavegen_chordloop
+
+.chords_done
+; ----------------------------------------
+; filters
+; proposed register assignment:
+; a0 = sample output
+; a1 = end of filter chunk
+; a2 = filter func
+; d7/a6 = filter start
+; a4 = MyPlayer
+; a5 = unused
+; a3 = waveinfo
+; d3/d4/d5/d6 = filter taps
+; d0/d1/d7 = scratch
+
+        moveq.l #0,d0
+        move.b  wi_flt_type_b(a3),d0
+        beq     .filter_done
+
+        IFNE    PRETRACKER_PARANOIA_MODE
+        tst.w   pv_wg_curr_sample_len_w(a4)
+        beq     .filter_done
+        ENDC
+
+        add.w   d0,d0
+        lea     .filterfunc_jmptable(pc),a2
+        add.w   -1*2(a2,d0.w),a2
+
+        moveq.l #0,d4       ; filter tap values
+        moveq.l #0,d5       ; filter tap values
+        movem.l d4-d5,pv_wg_flt_taps(a4)
+
+        lea     wi_flt_start_b(a3),a0
+        moveq.l #0,d0
+        move.b  (a0)+,d0    ; wi_flt_start_b
+        lsl.w   #8,d0
+
+        move.b  (a0)+,d4    ; wi_flt_min_b
+        lsl.w   #8,d4       ; flt_min*256
+
+        move.b  (a0)+,d5    ; wi_flt_max_b
+        lsl.w   #8,d5       ; flt_max*256
+
+        move.b  (a0)+,d3    ; wi_flt_speed_b
+        ext.w   d3
+        ext.l   d3          ; flt_speed*128
+        lsl.l   #7,d3
+
+        movea.l pv_wg_curr_sample_ptr(a4),a0
+
+.entry_to_filter_loop
+        move.l  d0,a6
+        move.l  d3,d1       ; flt_speed_b*128
+        adda.l  d1,a6       ; suppress M68kUnexpectedConditionalInstruction
+        bgt.s   .filter_speed_pos
+
+.filter_speed_neg
+        move.l  d4,d1       ; flt_min*256
+        cmp.l   d1,d0
+        blt.s   .lbC002790
+        cmp.l   d1,a6
+        bgt.s   .lbC002936
+        move.l  d1,a6
+        cmp.l   d5,d1       ; flt_max*256
+        beq.s   .filter_load_min
+        neg.l   d3          ; flt_speed_b*128
+        bra.s   .filter_load_min
+
+.filterfunc_jmptable
+        dc.w    .lowpassfilter-.filterfunc_jmptable
+        dc.w    .highpassfilter-.filterfunc_jmptable
+        dc.w    .bandpassfilter-.filterfunc_jmptable
+        dc.w    .notchfilter-.filterfunc_jmptable
+
+.lbC002790
+        tst.l   d0
+        blt.s   .lbC002936
+        move.l  a6,d7
+        bgt.s   .lbC002936
+        neg.l   d3          ; flt_speed_b*128
+        move.w  #$FF,d2
+        move.w  d2,d1
+        sub.l   a6,a6
+        bra.s   .filter_cont
+
+.filter_speed_pos
+        cmp.l   d5,d0       ; flt_max*256
+        bgt.s   .lbC002D2A
+        cmp.l   d5,a6       ; flt_max*256
+        blt.s   .lbC002936
+        move.l  d4,d2       ; flt_min*256
+        cmp.l   d5,d2       ; flt_max*256
+        beq.s   .filter_load_max_no_flip
+        neg.l   d3          ; flt_speed_b*128
+        move.l  d5,a6       ; flt_max*256
+        bra.s   .filter_load_max
+
+.lbC002D2A
+        cmpi.l  #$FF00,d0
+        bgt.s   .lbC002936
+        cmp.l   #$FEFF,a6
+        ble.s   .lbC002936
+        neg.l   d3          ; flt_speed_b*128
+        moveq.l #0,d2
+        move.l  #$FF00,a6
+        bra.s   .filter_cont
+
+.lbC002936
+        move.w  a6,d2
+        lsr.w   #8,d2
+        not.b   d2
+        bra.s   .filter_cont
+
+.filter_load_max_no_flip
+        movea.l d2,a6
+.filter_load_max
+        moveq.l #0,d2
+        move.b  wi_flt_max_b(a3),d2
+        not.b   d2
+        bra.s   .filter_cont
+
+.filter_load_min
+        moveq.l #0,d2
+        move.b  wi_flt_min_b(a3),d2
+        not.b   d2
+
+.filter_cont
+        btst    #0,wi_flt_type_b(a3)
+        bne.s   .not_notch_or_highpass
+.highpass_or_notch          ; entered for 2 or 4
+        not.b   d2
+.not_notch_or_highpass      ; entered for 1 or 3
+        move.w  d2,d0
+        add.w   d0,d0
+
+        moveq.l #0,d7
+        move.b  wi_flt_resonance_b(a3),d7
+        beq.s   .filter_no_resonance
+        move.w  d2,d1
+        ext.l   d1
+        lsl.l   #8,d1
+
+        moveq.l #$B6/2,d0
+        sub.w   d7,d0
+        add.w   d0,d0
+
+        cmpi.w  #$36,d0
+        bge.s   .filter_no_clip_resonance
+        moveq.l #$36,d0
+.filter_no_clip_resonance
+        divu    d0,d1
+        move.w  d2,d0
+        add.w   d1,d0
+.filter_no_resonance
+        lea     $40(a0),a1          ; end of sample chunk
+
+        movem.l d3-d5,-(sp)
+        movem.w pv_wg_flt_taps(a4),d3-d6
+
+        ; d0/d2 relevant for inner loop
+.filter_innerloop
+        move.b  (a0),d1
+        ext.w   d1
+
+        move.w  d4,d7
+        sub.w   d5,d7
+        muls    d0,d7
+        asr.l   #8,d7
+        sub.w   d4,d7
+
+        add.w   d1,d7
+        muls    d2,d7
+        asr.l   #8,d7
+        add.w   d7,d4
+        move.w  d4,d7
+
+        sub.w   d5,d7
+        muls    d2,d7
+        asr.l   #8,d7
+        add.w   d7,d5
+        move.w  d5,d7
+
+        sub.w   d6,d7
+        muls    d2,d7
+        asr.l   #8,d7
+        add.w   d7,d6
+        move.w  d6,d7
+
+        sub.w   d3,d7
+        muls    d2,d7
+        asr.l   #8,d7
+        add.w   d7,d3
+        move.w  d3,d7
+
+        jmp     (a2)
+
+.highpassfilter
+        sub.w   d1,d7
+        bra.s   .filterclipresult
+
+.bandpassfilter
+        sub.w   d4,d7
+        sub.w   d5,d7
+        sub.w   d6,d7
+        asr.w   #1,d7
+        bra.s   .filterclipresult
+
+.notchfilter
+        sub.w   d4,d7
+        neg.w   d7
+
+.lowpassfilter
+.filterclipresult
+        CLIPTO8BIT d7
+.filter_outputbyte
+        move.b  d7,(a0)+
+        cmpa.l  a0,a1
+        bne.s   .filter_innerloop
+
+.filterloop_end_test
+        movem.w d3-d6,pv_wg_flt_taps(a4)
+        movem.l (sp)+,d3-d5
+
+        cmpa.l  pv_wg_curr_samend_ptr(a4),a0
+        bhs.s   .filter_done
+        move.l  a6,d0
+        bra     .entry_to_filter_loop
+
+.filter_done
+; ----------------------------------------
+; Optional Pre-Modulator
+        btst    #5,wi_mod_density_b(a3) ; post bit
+        bne.s   .nopremodulator
+        bsr     pre_Modulator
+.nopremodulator
+
+; ----------------------------------------
+; start with volume envelope
+; a0 = output sample buffer
+; d0 = scratch (e.g. sample)
+; d1 = increment for attack phase
+; d3 = current volume for attack and decay phases
+; d4 = remaining sample length - 1
+; a3 = wave info
+
+.vol_do_envelope
+        move.l  pv_wg_curr_sample_ptr(a4),a0    ; load buffer pointer
+        move.w  pv_wg_curr_sample_len_w(a4),d4  ; load length
+        IFNE    PRETRACKER_PARANOIA_MODE
+        beq     .vol_envelope_finished ; paranoia
+        ENDC
+        subq.w  #1,d4           ; we use length-1, <0 is end
+
+        moveq.l #2,d1           ; turns into $20000 through swap
+        moveq.l #0,d0
+        move.b  wi_vol_attack_b(a3),d0
+        bne.s   .has_attack_volume
+        cmpi.b  #$FF,wi_vol_sustain_b(a3)
+        beq     .vol_envelope_finished
+        ; no attack but not full sustain -> go to delay
+        ;move.l  #$100<<16,d3
+        bra.s   .vol_skip_attack
+
+.vol_avoid_overflow_with_1
+        moveq.l #1,d1
+        bra.s   .cont_vol_envelope
+.has_attack_volume
+        moveq.l #0,d3
+        cmp.w   #1,d0
+        beq.s   .cont_vol_envelope
+        cmp.w   #2,d0
+        beq.s   .vol_avoid_overflow_with_1
+        swap    d1                  ; turn into $20000
+        divu    d0,d1
+        swap    d1
+        clr.w   d1
+        ; swap is done below
+.cont_vol_envelope
+        swap    d1                  ; move to high word (should be max $20000 then)
+        btst    #5,wi_flags_b(a3)   ; vol fast flag
+        beq.s   .vol_no_fast
+        lsl.l   #4,d1               ; multiply speed by 16
+.vol_no_fast
+        add.l   d1,d3               ; increase volume
+        cmpi.l  #$FFFFFF,d3
+        ble.s   .vol_do_attack      ; first step overshooting?
+.vol_skip_attack
+        btst    #3,wi_flags_b(a3)   ; boost flag
+        bne.s   .vol_delay_boosted
+        bra.s   .vol_delay_normal
+
+.vol_do_attack
+        btst    #3,wi_flags_b(a3)   ; boost flag
+        bne.s   .vol_attack_boosted
+
+; ----------------------------------------
+; attack phase with volume boosted
+.vol_attack_normal
+.vol_attack_normal_loop
+        move.b  (a0),d0
+        ext.w   d0
+        swap    d3
+        muls    d3,d0
+        swap    d3
+        asr.w   #8,d0
+        move.b  d0,(a0)+
+
+        subq.w  #1,d4
+        bmi     .vol_envelope_finished
+        add.l   d1,d3               ; increase volume
+        cmpi.l  #$FFFFFF,d3
+        ble.s   .vol_attack_normal_loop
+
+; ----------------------------------------
+; delay phase (normal)
+
+.vol_delay_normal   ; moved this label two inst up, didn't make sense there
+        moveq.l #0,d0
+        move.b  wi_vol_delay_b(a3),d0
+        lsl.w   #4,d0
+
+        IFNE    PRETRACKER_FASTER_CODE
+        ; skip the delay -- we don't change the volume for this section
+        addq.w  #1,d0
+        sub.w   d0,d4
+        bmi     .vol_envelope_finished
+        lea     1(a0,d0.w),a0
+        ELSE
+        lea     2(a0,d0.w),a1
+
+        move.w  #$FF,d3        ; FIXME I don't think that this is quite right. Shouldn't the max volume NOT change the value?
+.vol_delay_normal_loop
+        move.b  (a0),d0
+        IFNE    1
+        ext.w   d0
+        muls    d3,d0
+        asr.w   #8,d0
+        ELSE
+        ; this should be the same as above (*(256-1))
+        spl     d3
+        add.b   d3,d0
+        ENDC
+        move.b  d0,(a0)+
+
+        cmpa.l  a1,a0
+        dbeq    d4,.vol_delay_normal_loop
+        bne     .vol_envelope_finished
+        ENDC
+
+        bra.s   .vol_delay_end_reached
+
+; ----------------------------------------
+; attack with volume boosted
+
+.vol_attack_boosted
+.vol_attack_boosted_loop
+        move.b  (a0),d0
+        ext.w   d0
+        swap    d3
+        muls    d3,d0
+        swap    d3
+        asr.w   #6,d0
+        CLIPTO8BIT d0
+        move.b  d0,(a0)+
+
+        subq.w  #1,d4
+        bmi     .vol_envelope_finished
+        add.l   d1,d3
+        cmpi.l  #$FFFFFF,d3
+        ble.s   .vol_attack_boosted_loop
+
+; ----------------------------------------
+; delay with max volume boosted
+
+.vol_delay_boosted
+        moveq.l #0,d0
+        move.b  wi_vol_delay_b(a3),d0
+        lsl.w   #4,d0
+
+        lea     2(a0,d0.w),a1
+
+        IFNE    PRETRACKER_FASTER_CODE
+.vol_delay_boosted_loop
+        move.b  (a0),d0
+        add.b   d0,d0
+        CLIPTO8BITAFTERADD d0
+        add.b   d0,d0
+        CLIPTO8BITAFTERADD d0
+        ELSE
+        move.w  #$FF,d3        ; FIXME I don't think that this is quite right. It should be $100 to boost by full volume
+.vol_delay_boosted_loop
+        move.b  (a0),d0
+        ext.w   d0
+        muls    d3,d0
+        asr.w   #6,d0
+        CLIPTO8BIT d0
+        ENDC
+        move.b  d0,(a0)+
+
+        cmpa.l  a1,a0
+        dbeq    d4,.vol_delay_boosted_loop
+        bne     .vol_envelope_finished
+
+.vol_delay_end_reached
+        subq.w  #1,d4
+
+; ----------------------------------------
+; decay phase
+; d0 = scratch
+; d1 = current volume decrement
+; d2 = table index boundary
+; d3 = 16:16 decay pos
+; d4 = sample length counter
+; d5 = volume
+; d6 = scratch
+; d7 = decay increment
+; a0 = sample pointer
+; a1 = (current) roll off table pointer (points to upper bound)
+; a2 = lower bound
+
+.vol_do_decay
+        moveq.l #0,d3
+        move.b  wi_vol_decay_b(a3),d3
+        beq     .vol_do_sustain
+        move.w  d3,d7
+        mulu    d7,d7
+        lsr.w   #2,d7
+        add.w   d3,d7       ; d7 = (d3^2)/4+d3 (<= 16511)
+
+        btst    #5,wi_flags_b(a3)   ; vol fast flag
+        beq.s   .vol_decay_not_fast
+        moveq.l #0,d3       ; will cause a5=$400,a2=$200,d2=0, decay value has no effect on the ramp used
+.vol_decay_not_fast
+        lsl.w   #8,d3
+        lsl.l   #4,d3
+        move.l  d3,d2
+        swap    d2
+        lea     pre_roll_off_table(pc),a1
+        add.w   d2,d2
+        adda.w  d2,a1
+        move.w  (a1)+,a2    ; first index in table
+        lsr.w   #1,d2       ; update next boundary
+
+        moveq.l #0,d1       ; current volume decrement
+        moveq.l #0,d5
+        not.w   d5          ; set maximum volume
+.vol_decay_normal_loop
+        add.l   d7,d3       ; increment position in decay
+        swap    d3
+
+        cmp.w   #$8e,d3     ; pos in table where it makes no sense to do lerp anymore
+        bhi.s   .vol_keep_voldec
+        cmp.w   d2,d3
+        bls.s   .vol_do_lerp
+
+.vol_lerp_next_section
+        move.w  (a1),a2
+        IFNE    PRETRACKER_BUGFIX_CODE
+        lea     pre_roll_off_table+2(pc),a1 ; Take the right boundary value
+        ELSE
+        lea     pre_roll_off_table(pc),a1   ; This will set a wrong boundary and thus plateau the slope decay speed
+        ENDC
+        add.w   d3,a1
+        add.w   d3,a1
+
+        move.w  d3,d2       ; update next boundary
+
+.vol_do_lerp        ; ((lowerbound-upperbound)*(d3<<8))>>8 + upperbound
+        move.w  a2,d1
+        move.w  (a1),d0
+        sub.w   d1,d0               ; delta between lower and upper bound (negative value)
+        beq.s   .vol_skip_lerp
+
+        swap    d3
+        move.w  d3,d1
+        lsr.w   #8,d1
+        muls    d0,d1
+        asr.l   #8,d1
+        add.w   a2,d1
+        swap    d3
+
+.vol_keep_voldec
+.vol_skip_lerp
+        swap    d3
+        sub.w   d1,d5
+        bls.s   .vol_do_sustain
+
+        move.w  d5,d6
+        lsr.w   #8,d6
+
+        cmp.b   wi_vol_sustain_b(a3),d6
+        bls.s   .vol_do_sustain
+
+        move.b  (a0),d0
+        ext.w   d0
+        muls    d6,d0
+        btst    #3,wi_flags_b(a3)       ; boost flag
+        CLIPORTRUNC8BIT d0
+        move.b  d0,(a0)+
+
+        dbra    d4,.vol_decay_normal_loop
+        bra.s   .vol_envelope_finished
+
+; ----------------------------------------
+; sustain phase
+.vol_do_sustain
+        moveq.l #0,d3
+        move.b  wi_vol_sustain_b(a3),d3
+        beq.s   .vol_sustain_silence
+
+        btst    #3,wi_flags_b(a3)       ; boost flag
+        beq.s   .vol_sustain_normal
+.vol_sustain_boosted
+.vol_sustain_boosted_loop
+        move.b  (a0),d0
+        ext.w   d0
+        muls    d3,d0
+        asr.w   #6,d0
+
+        CLIPTO8BIT d0
+        move.b  d0,(a0)+
+        dbra    d4,.vol_sustain_boosted_loop
+        bra.s   .vol_envelope_finished
+
+.vol_sustain_silence
+        moveq.l #0,d0
+.vol_sustain_silence_loop
+        move.b  d0,(a0)+
+        dbra    d4,.vol_sustain_silence_loop
+        bra.s   .vol_envelope_finished
+
+.vol_sustain_normal
+        IFNE    PRETRACKER_FASTER_CODE
+        cmp.b   #$ff,d3
+        beq.s   .vol_envelope_finished
+        ENDC
+
+.vol_sustain_normal_loop
+        move.b  (a0),d0
+        ext.w   d0
+        muls    d3,d0
+        asr.w   #8,d0
+        move.b  d0,(a0)+
+        dbra    d4,.vol_sustain_normal_loop
+
+.vol_envelope_finished
+
+; ----------------------------------------
+; Optional Post-Modulator
+
+        btst    #5,wi_mod_density_b(a3) ; post bit
+        beq.s   .nopostmodulator
+        bsr     pre_Modulator
+.nopostmodulator
+
+; ----------------------------------------
+; wave mixing (removed some code here that was doing nothing as result
+; because below higher octaves code would overwrite it anyway).
+        IFND    PRESTO_UNIFIED_STRUCT
+        movea.l pv_my_song(a4),a6
+        ENDC
+
+        moveq.l #0,d0
+        move.b  wi_mix_wave_b(a3),d0
+        beq.s   .mix_no_wave_mixing         ; no mixing selected
+        subq.b  #1,d0
+        IFNE    PRETRACKER_PARANOIA_MODE    ; same wave for mixing cannot be selected in Pretracker
+        cmp.b   pv_wg_curr_wave_num_b(a4),d0
+        beq     .mix_no_wave_mixing         ; same wave number!
+        ENDC
+
+        lsl.w   #2,d0
+        move.l  pv_wave_sample_table(a4,d0.w),a1
+        IFND    PRESTO_UNIFIED_STRUCT
+        add.w   #sv_wavelength_table,d0
+        move.l  (a6,d0.w),d3
+        ELSE
+        move.l  pv_wavelength_table(a4,d0.w),d3
+        ENDC
+
+        move.w  pv_wg_curr_sample_len_w(a4),d4 ; length of the sample to mix to
+        cmp.w   d3,d4
+        ble.s   .mix_picked_shorter
+        move.w  d3,d4
+.mix_picked_shorter
+
+        move.l  pv_wg_curr_sample_ptr(a4),a0
+
+        subq.w  #1,d4
+.mix_mixloop
+        move.b  (a0),d0
+        add.b   (a1)+,d0
+        CLIPTO8BITAFTERADD d0
+        move.b  d0,(a0)+
+        dbra    d4,.mix_mixloop
+.mix_no_wave_mixing
+
+; ----------------------------------------
+; create higher octaves (this has been massively shortened)
+
+        btst    #2,wi_flags_b(a3)
+        beq.s   .oct_has_no_extra_octaves
+
+        move.l  pv_wg_curr_sample_len_l(a4),d4
+        IFNE    PRETRACKER_PARANOIA_MODE
+        beq.s   .oct_has_no_extra_octaves
+        ENDC
+
+        movea.l pv_wg_curr_sample_ptr(a4),a1
+        lea     (a1,d4.l),a0    ; needs to be .l due to 32678 max length
+
+        mulu    #7,d4
+        lsr.l   #3,d4           ; * 0.875
+        subq.w  #1,d4
+.oct_downsample_loop
+        move.b  (a1),(a0)+
+        addq.l  #2,a1
+        dbra    d4,.oct_downsample_loop
+
+.oct_has_no_extra_octaves
+; ----------------------------------------
+        IFNE    PRETRACKER_PROGRESS_SUPPORT
+        move.l  pv_precalc_progress_ptr(a4),d0
+        beq.s   .no_progress_out
+        move.l  d0,a0
+        move.l  pv_precalc_sample_size(a4),d0
+        add.l   d0,(a0)
+.no_progress_out
+        ENDC
+        move.l  (sp)+,a1
+        subq.w  #1,pv_wg_wave_counter_w(a4)
+        bgt     .wavegenloop
+
+        rts
+
+;--------------------------------------------------------------------
+; a3: waveinfo
+;
+; d6: wetness (word)
+; uses all data registers and a0-a2 (a3 unchanged)
+pre_Modulator:
+        tst.b   wi_mod_wetness_b(a3)
+        beq.s   .earlyexit
+        moveq.l #7,d0
+        and.b   wi_mod_density_b(a3),d0
+        bne.s   .has_density
+.earlyexit
+        rts
+.has_density
+        move.l  pv_wg_curr_sample_ptr(a4),a0
+        move.w  pv_wg_curr_sample_len_w(a4),d4
+        IFNE    PRETRACKER_PARANOIA_MODE
+        bne.s   .not_zero
+        rts
+.not_zero
+        ENDC
+        moveq.l #0,d6
+        move.b  wi_mod_wetness_b(a3),d6
+
+        moveq.l #0,d7
+
+        lea     pre_modulator_ramp_8(pc),a1
+.loop   moveq.l #0,d5
+        move.b  wi_mod_length_b(a3),d5
+        mulu    (a1)+,d5            ; result is a long value
+
+        moveq.l #0,d1
+        move.b  wi_mod_predelay_b(a3),d1
+
+        btst    #5,wi_mod_density_b(a3) ; post-modulator?
+        bne.s   .factor1_256
+        ; factor 1/4 and 64
+        lsr.l   #2,d5
+        lsl.l   #6,d1
+        bra.s   .cont
+
+.factor1_256
+        lsl.w   #8,d1
+.cont   add.l   d1,d5               ; sum up length and predelay
+
+        moveq.l #0,d2
+        moveq.l #0,d3
+
+.innerloop
+        add.w   d7,d3
+        addq.w  #8,d3
+        smi     d1
+        ext.w   d1
+        eor.w   d3,d1
+        lsr.w   #6,d1               ; 4 bit key is bits 14 to 11, needs to be 8 to 5
+        and.w   #15<<5,d1
+        ext.l   d1
+
+        add.l   d5,d1
+        lsr.l   #6,d1
+
+        move.w  d2,d0
+        sub.w   d1,d0
+        bmi.s   .isneg
+
+        move.b  (a0,d0.w),d1
+        ext.w   d1
+        btst    #0,d7
+        beq.s   .keep_dc
+        neg.w   d1
+.keep_dc
+
+        muls    d6,d1
+        asr.w   #8,d1
+        add.b   (a0,d2.w),d1
+
+        CLIPTO8BITAFTERADD d1
+        move.b  d1,(a0,d2.w)
+.isneg
+        addq.w  #1,d2
+        cmp.w   d4,d2
+        bcs.s   .innerloop
+
+.restartloop
+        addq.w  #1,d7
+        moveq.l #7,d0
+        and.b   wi_mod_density_b(a3),d0
+        cmp.w   d0,d7
+        bcs.s   .loop
+        rts
+
+
+;--------------------------------------------------------------------
+
+pre_MemClr
+        lsr.w   #1,d0
+        subq.w  #1,d0
+        bmi.s   .skipmemclr
+        moveq.l #0,d1
+.fillmemloop
+        move.w  d1,(a0)+
+        dbra    d0,.fillmemloop
+.skipmemclr
+        rts