FANCO v6 — SHORT SOUND AND ORDINARY CODE
======================================

Version boundary: import v6/assembler.ts and use `.version 6`.
Old version-05 bytes do not run under the new encoding/map.

One write to play a factory instrument
-------------------------------------
    mov.b [0x10f], 69      ; A4, 440 Hz, gate on, envelope hit
    mov.b [0x10f], 69      ; another hit, even though the value is unchanged
    mov.b [0x10f], 255     ; release

Each instruction is five bytes in the current v6 assembler. NOTE uses the
existing fixed patch; it never chooses or rewrites an instrument. With legato
flag bit4, held-note pitch changes skip the hit. Action ports read idle FF.
Commands are consumed at the next control tick; repeated writes within that
tick collapse to the latest, rather than creating an unbounded sound queue.

The CPU must keep the host running; a program that HALTs ends its machine
session. A minimal sustained example uses a timer/WAIT loop:

    .version 6
    .sound on
    mov.w [0x10], irq
    mov.w [0x12], 50
    mov.b [0x10f], 69
    idle: wait
    bra idle
    irq: reti

Low-level control remains straightforward
----------------------------------------
Generator n starts at 0100+n*10 hex. Its first seven bytes are frequency word,
pulse word, wave/control, AD and SR. They are typed parameters of the same
kernel, not a call to a separate SID engine. A raw-frequency instrument:

    mov.w [0x100], 0x1d44
    mov.w [0x102], 0x0800
    mov.b [0x105], 0x00
    mov.b [0x106], 0xf0
    mov.b [0x104], 0x41    ; pulse + gate

Modulation, custom exponential/looping envelopes, self-feedback and effects
can be layered without replacing these basic fields. Inaudible generators
can modulate audible ones. The eight-slot mixer/pan/filter/send bank is
separate from synthesis configuration.

General CPU savings, not a song opcode
-------------------------------------
    mov b,a               ; one byte
    movzx a,[p+18]        ; two-byte byte-to-word record load
    mov.b [p+6],a.lo      ; two-byte record store
    mov.b [p+7],0         ; three-byte record initialization
    lea r,[p+18]          ; three-byte flag-neutral effective address

MOV/loads do not set flags. Explicit CMP or arithmetic produces N/Z/C/V.
For a loop, do not branch on a stale flag after a load:

    loop: movzx a,[p++]
    cmp a,0
    bz done
    ; use a
    bra loop
    done: ret

The current CPU has useful pointer/displacement, byte/word, immediate and
full escape forms. No firmware pattern/music decoder or score data is used.

Fixed point and multiword math
-----------------------------
    smulshift a,b,15      ; signed Q1.15 multiply, floor rounding, four bytes
    smulshift x,y,16      ; signed high product, four bytes
    umulhi x,y           ; unsigned high product, three bytes
    add a,x              ; low-word addition sets carry
    adc b,y              ; high-word addition includes carry, three bytes
    sub a,x              ; low-word subtraction sets borrow
    sbc b,y              ; high-word subtraction includes borrow, three bytes

Math destinations are a..sp and sources a..pc (word registers only).
SMULSHIFT accepts 0..31; V reports signed overflow after shifting, C clears.
UMULHI clears C/V. ADC/SBC set ordinary N/Z/C/V, including incoming carry
or borrow in the exact sum/difference. MOV preserves carry between limbs.
For a full unsigned product, copy the original destination before MUL and
use UMULHI on the copy. No fixed-point scale, table or scene is in firmware.
DIV/MOD remain unsigned; signed/wide division is not added speculatively.

    abs x                ; signed magnitude, three bytes, no scratch register
    smax x,y             ; signed maximum, three bytes
    smin x,y             ; signed minimum, three bytes

These operations set N/Z from the result and clear C/V, except ABS(-32768)
returns 8000 hex with N/V set. SMIN/SMAX compare signed words, not magnitudes.
They are useful for clipping, distance fields and ordinary integer intervals.

Observed complete-program budget
--------------------------------
Including all nineteen selections, touch/keyboard controls, visible tiles,
all musical/effect data and image framing:

    Original C64 payload      4,039 bytes
    v5 full native            5,120 bytes
    v6 basic-alias reference   4,126 bytes
    v6 native                 4,145 bytes

The v6 image is about 2.6% larger than the original. General compact CPU
encodings supply most of the savings; the regular native bank and simple
50-Hz timer remove staging/conversion overhead. These are actual assembled
sizes, not transistor-equivalence or an exact C64 waveform claim. New timing
is slightly slower than PAL, and SIREN retains its native synthesis/effects.
Whole-program code/data count includes the player, never hides it in the VM.