Retrointernals

Boulder Dash Theme Score

OK, I got the idea to score the Boulder Dash music into my head. I had a look around to see if anyone else had already done so, and they had, kind of. The man himself has a crack here. There are other scores out there, but most of them seem to be arrangements of the theme and not faithful representations. Assuring the world has access to accurate sheet music for this corker might keep me out of trouble for a bit.

Congratulations. I don’t give a shit!

Ah… It’s like that is it?

A PDF of the score can be found here

More of an SVG man? We can accommodate

MIDI? sure

A GitHub repository with all the code

Overview

The theme music use two of the SID chip’s three voices. The data for the tune is exactly 256 bytes long (not including the note to SID frequency value table). The routine doesn’t support rests and voice 1 can only have notes of a single duration. Voice 2 is similarly limited but it’s envelope generator is set such that two or more consecutive notes of the same pitch blur into a single note, so notes of different durations are supported. The tune data is simply 128 pairs of note numbers, the first value of the pair for one voice and the second for the other. If we notate the note duration as a quaver (a quarter note) the tune is 16 bars long (4/4 time) repeated twice. I think it’s in F minor, and I’ve notated it as such, but I’m no musician.

Tuning

I’m going to start by figuring out the tuning. The table which maps notes to SID frequency values looks like this.

MusicNoteToFreqTable:
8317   dc 02 0a 03  3a 03 6c 03  a0 03 d2 03  12 04 4c 04
8327   92 04 d6 04  20 05 6e 05  b8 05 14 06  74 06 d8 06
8337   40 07 a4 07  24 08 98 08  24 09 ac 09  40 0a dc 0a
8347   70 0b 28 0c  e8 0c b0 0d  80 0e 48 0f  48 10 30 11
8357   48 12 58 13  80 14 b8 15  e0 16 50 18  d0 19 60 1b
8367   00 1d 90 1e  90 20 60 22  90 24 b0 26  00 29 70 2b
8377   c0 2d

Each entry is two bytes long. We need to do two things here:

1. Map from SID frequency value to a frequency.
2. Map from a frequency to a note.

We’ll take one at a time then tie it all together.

Mapping from a SID frequency value to a frequency

I Googled a few resources when trying to figure out how to do this. I did briefly consider getting an oscilloscope and verifying what I’d learnt, but I decided that was probably overkill. We’ll interpret consistency across multiple sources as verification enough. Anyway, here are some links if you’re interested in reading up on this:

The SID (by Stephen L. Judd)

How to calculate SID frequency tables By FTC/HT

Excact [sic] Vertical Frequency / Refresh Rate

I hacked together the Python code below to get the frequencies of the notes. I live in Australia, so we had PAL machines. I’m only going to consider the PAL case but I’ve included the code for NTSC machines.

#!/usr/bin/env python3

from math import log

bd_sid_values = [
0xdc, 0x02, 0x0a, 0x03,  0x3a, 0x03, 0x6c, 0x03,  0xa0, 0x03, 0xd2, 0x03,  0x12, 0x04, 0x4c, 0x04,
0x92, 0x04, 0xd6, 0x04,  0x20, 0x05, 0x6e, 0x05,  0xb8, 0x05, 0x14, 0x06,  0x74, 0x06, 0xd8, 0x06,
0x40, 0x07, 0xa4, 0x07,  0x24, 0x08, 0x98, 0x08,  0x24, 0x09, 0xac, 0x09,  0x40, 0x0a, 0xdc, 0x0a,
0x70, 0x0b, 0x28, 0x0c,  0xe8, 0x0c, 0xb0, 0x0d,  0x80, 0x0e, 0x48, 0x0f,  0x48, 0x10, 0x30, 0x11,
0x48, 0x12, 0x58, 0x13,  0x80, 0x14, 0xb8, 0x15,  0xe0, 0x16, 0x50, 0x18,  0xd0, 0x19, 0x60, 0x1b,
0x00, 0x1d, 0x90, 0x1e,  0x90, 0x20, 0x60, 0x22,  0x90, 0x24, 0xb0, 0x26,  0x00, 0x29, 0x70, 0x2b,
0xc0, 0x2d]

def sid_frequencies():
    i = iter(bd_sid_values)
    try:
        while True:
            yield next(i)+next(i)*256
    except StopIteration:
        return

pal_const =  (256**3)/985248
ntsc_const = (256**3)/1022727

def reg_to_freq_pal(reg):
    return reg/pal_const

def reg_to_freq_ntsc(reg):
    return reg/ntsc_const

def cents_from(f, ref):
    return 1200*log(f/ref, 2)

print("PAL\n---")
for f in sid_frequencies():
    f = reg_to_freq_pal(f)
    print(reg_to_freq_pal(f))

Here’s the output:

PAL
---
42.986961364746094
45.68832778930664
48.507144927978516
51.44341278076171
54.49713134765624
57.433399200439446
61.191822052001946
64.59789276123047
68.70866775512695
72.70199203491211
77.04766845703125
81.62824630737305
85.97392272949219
91.37665557861328
97.01428985595703
102.88682556152342
108.99426269531249
114.86679840087889
122.38364410400389
129.19578552246094
137.4173355102539
145.40398406982422
154.0953369140625
163.2564926147461
171.94784545898438
182.75331115722656
194.02857971191406
205.77365112304685
217.98852539062497
229.73359680175778
244.76728820800778
258.3915710449219
274.8346710205078
290.80796813964844
308.190673828125
326.5129852294922
343.89569091796875
365.5066223144531
388.0571594238281
411.5473022460937
435.97705078124994
459.46719360351557
489.53457641601557
516.7831420898438
549.6693420410156
581.6159362792969
616.38134765625
653.0259704589844
687.7913818359375

The closest value to A440 is 435.97705078124994 which is 16 cents below. A perceptible distance. If we use A435 instead this is still the closest value but this time 4 cents higher. That’s more like it. We’ll use this entry as our A4. Here is some info on cents and A440 for the curious.

Maping from a frequency to a note

This requires some understanding of the maths behind musical tunings. When you start looking into this you can’t help but wonder how deep the rabbit hole goes. For our purposes we’ll only concern ourselves with the equal temperament tuning system. To cut a long story short the following Python code will give us a note number measured in semitones from the reference frequency a4.

base = 2**(1/12)
a4 = 435.97705078124994

def freq_to_note(f):
    return log(f/a4, base)

Tieing it all together

In short this:

#!/usr/bin/env python3

from math import log, floor

bd_sid_values = [
0xdc, 0x02, 0x0a, 0x03,  0x3a, 0x03, 0x6c, 0x03,  0xa0, 0x03, 0xd2, 0x03,  0x12, 0x04, 0x4c, 0x04,
0x92, 0x04, 0xd6, 0x04,  0x20, 0x05, 0x6e, 0x05,  0xb8, 0x05, 0x14, 0x06,  0x74, 0x06, 0xd8, 0x06,
0x40, 0x07, 0xa4, 0x07,  0x24, 0x08, 0x98, 0x08,  0x24, 0x09, 0xac, 0x09,  0x40, 0x0a, 0xdc, 0x0a,
0x70, 0x0b, 0x28, 0x0c,  0xe8, 0x0c, 0xb0, 0x0d,  0x80, 0x0e, 0x48, 0x0f,  0x48, 0x10, 0x30, 0x11,
0x48, 0x12, 0x58, 0x13,  0x80, 0x14, 0xb8, 0x15,  0xe0, 0x16, 0x50, 0x18,  0xd0, 0x19, 0x60, 0x1b,
0x00, 0x1d, 0x90, 0x1e,  0x90, 0x20, 0x60, 0x22,  0x90, 0x24, 0xb0, 0x26,  0x00, 0x29, 0x70, 0x2b,
0xc0, 0x2d]

def sid_frequencies():
    i = iter(bd_sid_values)
    try:
        while True:
            yield next(i)+next(i)*256
    except StopIteration:
        return

def note_to_sid(n):
    return bd_sid_values[n*2]+bd_sid_values[n*2+1]*256

names_sharp=['a{}','a{}♯','b{}','c{}','c{}♯','d{}','d{}♯','e{}','f{}','f{}♯','g{}','g{}♯']
names_flat=['a{}','b{}♭','b{}','c{}','d{}♭','d{}','e{}♭','e{}','f{}','g{}♭','g{}','a♭{}']

def index_to_name(i, sharp):
    octave = floor((i-3)/12)+5
    names = names_sharp if sharp else names_flat
    return names[i%12].format(octave)

pal_const =  (256**3)/985248
ntsc_const = (256**3)/1022727
a4 = 435.97705078124994
base = 2**(1/12)

def reg_to_freq_pal(reg):
    return reg/pal_const

def reg_to_freq_ntsc(reg):
    return reg/ntsc_const

def freq_to_note(f):
    return log(f/a4, base)

bd_val = 0x0a
for sid in sid_frequencies():
    freq = reg_to_freq_pal(sid)
    idx = round(freq_to_note(freq))
    print("${:02x} : {}".format(bd_val, index_to_name(idx, True)))
    bd_val += 1
print()

Gives us the note names that correspond to the note numbering Boulder Dash uses:

$0a : f1
$0b : f1♯
$0c : g1
$0d : g1♯
$0e : a1
$0f : a1♯
$10 : b1
$11 : c2
$12 : c2♯
$13 : d2
$14 : d2♯
$15 : e2
$16 : f2
$17 : f2♯
$18 : g2
$19 : g2♯
$1a : a2
$1b : a2♯
$1c : b2
$1d : c3
$1e : c3♯
$1f : d3
$20 : d3♯
$21 : e3
$22 : f3
$23 : f3♯
$24 : g3
$25 : g3♯
$26 : a3
$27 : a3♯
$28 : b3
$29 : c4
$2a : c4♯
$2b : d4
$2c : d4♯
$2d : e4
$2e : f4
$2f : f4♯
$30 : g4
$31 : g4♯
$32 : a4
$33 : a4♯
$34 : b4
$35 : c5
$36 : c5♯
$37 : d5
$38 : d5♯
$39 : e5
$3a : f5

In music notation:

Note values

The Tune

Here’s the data for the tune:

5fe8   16 22 1d 26  22 29 25 2e  14 24 1f 27  20 29 27 30
5ff8   12 2a 12 2c  1e 2e 12 31  20 2c 33 37  21 2d 31 35
6008   16 22 16 2e  16 1d 16 24  14 20 14 30  14 24 14 20
6018   16 22 16 2e  16 1d 16 24  1e 2a 1e 3a  1e 2e 1e 2a
6028   14 20 14 2c  14 1b 14 22  1c 28 1c 38  1c 2c 1c 28
6038   11 1d 29 2d  11 1f 29 2e  0f 27 0f 27  16 33 16 27
6048   16 2e 16 2e  16 2e 16 2e  22 2e 22 2e  16 2e 16 2e
6058   14 2e 14 2e  14 2e 14 2e  20 2e 20 2e  14 2e 14 2e
6068   16 2e 32 2e  16 2e 33 2e  22 2e 32 2e  16 2e 33 2e
6078   14 2e 32 2e  14 2e 33 2e  20 2c 30 2c  14 2c 31 2c
6088   16 2e 16 3a  16 2e 35 38  22 2e 22 37  16 2e 31 35
6098   14 2c 14 38  14 2c 14 38  20 2c 20 33  14 2c 14 38
60a8   16 2e 32 2e  16 2e 33 2e  22 2e 32 2e  16 2e 33 2e
60b8   14 2e 32 2e  14 2e 33 2e  20 2c 30 2c  14 2c 31 2c
60c8   2e 32 29 2e  26 29 22 26  2c 30 27 2c  24 27 14 20
60d8   35 32 32 2e  2e 29 29 26  27 30 24 2c  20 27 14 20

Now we’ll take a look at the code which fetches the note data, looks up the SID values for the note and writes the values to the SID chip’s frequency registers:

83ca   ae 00 98   LDX MusicDataIndex
83cd   ee 00 98   INC MusicDataIndex
83d0   ee 00 98   INC MusicDataIndex
83d3   bd e9 5f   LDA MusicData+1,X
83d6   0a         ASL A
83d7   a8         TAY 
83d8   b9 03 83   LDA MusicNoteToFreqTable-$14,Y
83db   8d 00 d4   STA Sid_Voice1FreqLo
83de   b9 04 83   LDA MusicNoteToFreqTable-$13,Y
83e1   8d 01 d4   STA Sid_Voice1FreqHi
83e4   bd e8 5f   LDA MusicData,X
83e7   0a         ASL A
83e8   a8         TAY 
83e9   b9 03 83   LDA MusicNoteToFreqTable-$14,Y
83ec   8d 07 d4   STA Sid_Voice2FreqLo
83ef   b9 04 83   LDA MusicNoteToFreqTable-$13,Y
83f2   8d 08 d4   STA Sid_Voice2FreqHi

Again, the music data this is simply 128 pairs of note values, one for each voice. You can see the first number of the pair is for voice 2 and the second for voice 1. Also the first note value is $0a which makes the last $3a.

We have all the pieces we need to dump the notes for both voices. Let’s snap some blocks together:

#!/usr/bin/env python3

from math import log, floor

bd_sid_values = [
0xdc, 0x02, 0x0a, 0x03,  0x3a, 0x03, 0x6c, 0x03,  0xa0, 0x03, 0xd2, 0x03,  0x12, 0x04, 0x4c, 0x04,
0x92, 0x04, 0xd6, 0x04,  0x20, 0x05, 0x6e, 0x05,  0xb8, 0x05, 0x14, 0x06,  0x74, 0x06, 0xd8, 0x06,
0x40, 0x07, 0xa4, 0x07,  0x24, 0x08, 0x98, 0x08,  0x24, 0x09, 0xac, 0x09,  0x40, 0x0a, 0xdc, 0x0a,
0x70, 0x0b, 0x28, 0x0c,  0xe8, 0x0c, 0xb0, 0x0d,  0x80, 0x0e, 0x48, 0x0f,  0x48, 0x10, 0x30, 0x11,
0x48, 0x12, 0x58, 0x13,  0x80, 0x14, 0xb8, 0x15,  0xe0, 0x16, 0x50, 0x18,  0xd0, 0x19, 0x60, 0x1b,
0x00, 0x1d, 0x90, 0x1e,  0x90, 0x20, 0x60, 0x22,  0x90, 0x24, 0xb0, 0x26,  0x00, 0x29, 0x70, 0x2b,
0xc0, 0x2d]

def sid_frequencies():
    i = iter(bd_sid_values)
    try:
        while True:
            yield next(i)+next(i)*256
    except StopIteration:
        return

def note_to_sid(n):
    return bd_sid_values[n*2]+bd_sid_values[n*2+1]*256

names_sharp=['a{}','a{}♯','b{}','c{}','c{}♯','d{}','d{}♯','e{}','f{}','f{}♯','g{}','g{}♯']
names_flat=['a{}','b{}♭','b{}','c{}','d{}♭','d{}','e{}♭','e{}','f{}','g{}♭','g{}','a♭{}']

def index_to_name(i, sharp):
    octave = floor((i-3)/12)+5
    names = names_sharp if sharp else names_flat
    return names[i%12].format(octave)

bd_music = [
0x16, 0x22, 0x1d, 0x26,  0x22, 0x29, 0x25, 0x2e,  0x14, 0x24, 0x1f, 0x27,  0x20, 0x29, 0x27, 0x30,
0x12, 0x2a, 0x12, 0x2c,  0x1e, 0x2e, 0x12, 0x31,  0x20, 0x2c, 0x33, 0x37,  0x21, 0x2d, 0x31, 0x35,
0x16, 0x22, 0x16, 0x2e,  0x16, 0x1d, 0x16, 0x24,  0x14, 0x20, 0x14, 0x30,  0x14, 0x24, 0x14, 0x20,
0x16, 0x22, 0x16, 0x2e,  0x16, 0x1d, 0x16, 0x24,  0x1e, 0x2a, 0x1e, 0x3a,  0x1e, 0x2e, 0x1e, 0x2a,
0x14, 0x20, 0x14, 0x2c,  0x14, 0x1b, 0x14, 0x22,  0x1c, 0x28, 0x1c, 0x38,  0x1c, 0x2c, 0x1c, 0x28,
0x11, 0x1d, 0x29, 0x2d,  0x11, 0x1f, 0x29, 0x2e,  0x0f, 0x27, 0x0f, 0x27,  0x16, 0x33, 0x16, 0x27,
0x16, 0x2e, 0x16, 0x2e,  0x16, 0x2e, 0x16, 0x2e,  0x22, 0x2e, 0x22, 0x2e,  0x16, 0x2e, 0x16, 0x2e,
0x14, 0x2e, 0x14, 0x2e,  0x14, 0x2e, 0x14, 0x2e,  0x20, 0x2e, 0x20, 0x2e,  0x14, 0x2e, 0x14, 0x2e,
0x16, 0x2e, 0x32, 0x2e,  0x16, 0x2e, 0x33, 0x2e,  0x22, 0x2e, 0x32, 0x2e,  0x16, 0x2e, 0x33, 0x2e,
0x14, 0x2e, 0x32, 0x2e,  0x14, 0x2e, 0x33, 0x2e,  0x20, 0x2c, 0x30, 0x2c,  0x14, 0x2c, 0x31, 0x2c,
0x16, 0x2e, 0x16, 0x3a,  0x16, 0x2e, 0x35, 0x38,  0x22, 0x2e, 0x22, 0x37,  0x16, 0x2e, 0x31, 0x35,
0x14, 0x2c, 0x14, 0x38,  0x14, 0x2c, 0x14, 0x38,  0x20, 0x2c, 0x20, 0x33,  0x14, 0x2c, 0x14, 0x38,
0x16, 0x2e, 0x32, 0x2e,  0x16, 0x2e, 0x33, 0x2e,  0x22, 0x2e, 0x32, 0x2e,  0x16, 0x2e, 0x33, 0x2e,
0x14, 0x2e, 0x32, 0x2e,  0x14, 0x2e, 0x33, 0x2e,  0x20, 0x2c, 0x30, 0x2c,  0x14, 0x2c, 0x31, 0x2c,
0x2e, 0x32, 0x29, 0x2e,  0x26, 0x29, 0x22, 0x26,  0x2c, 0x30, 0x27, 0x2c,  0x24, 0x27, 0x14, 0x20,
0x35, 0x32, 0x32, 0x2e,  0x2e, 0x29, 0x29, 0x26,  0x27, 0x30, 0x24, 0x2c,  0x20, 0x27, 0x14, 0x20]

def voice1():
    i = iter(bd_music)
    try:
        while True:
            next(i)
            yield next(i)-10
    except StopIteration:
        return

def voice2():
    i = iter(bd_music)
    try:
        while True:
            yield next(i)-10
            next(i)
    except StopIteration:
        return

pal_const =  (256**3)/985248
ntsc_const = (256**3)/1022727
a4 = 435.97705078124994
base = 2**(1/12)

def reg_to_freq_pal(reg):
    return reg/pal_const

def reg_to_freq_ntsc(reg):
    return reg/ntsc_const

def freq_to_note(f):
    return log(f/a4, base)

v1 = ""
for n in voice1():
    sid = note_to_sid(n)
    f = reg_to_freq_pal(sid)
    i = round(freq_to_note(f))
    v1 += index_to_name(i, True)+" "

v2 = ""
for n in voice2():
    sid = note_to_sid(n)
    f = reg_to_freq_pal(sid)
    i = round(freq_to_note(f))
    v2 += index_to_name(i, True)+" "

print("Voice 1")
print("-------")
print(v1)
print()
print("Voice 2")
print("-------")
print(v2)

Which spits out this:

Voice 1
-------
f3 a3 c4 f4 g3 a3♯ c4 g4 c4♯ d4♯ f4 g4♯ d4♯ d5 e4 c5 f3 f4 c3 g3 d3♯ g4 g3
d3♯ f3 f4 c3 g3 c4♯ f5 f4 c4♯ d3♯ d4♯ a2♯ f3 b3 d5♯ d4♯ b3 c3 e4 d3 f4 a3♯
a3♯ a4♯ a3♯ f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4
f4 f4 f4 f4 f4 f4 f4 d4♯ d4♯ d4♯ d4♯ f4 f5 f4 d5♯ f4 d5 f4 c5 d4♯ d5♯ d4♯
d5♯ d4♯ a4♯ d4♯ d5♯ f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 d4♯ d4♯ d4♯ d4♯ a4
f4 c4 a3 g4 d4♯ a3♯ d3♯ a4 f4 c4 a3 g4 d4♯ a3♯ d3♯

Voice 2
-------
f2 c3 f3 g3♯ d2♯ d3 d3♯ a3♯ c2♯ c2♯ c3♯ c2♯ d3♯ a4♯ e3 g4♯ f2 f2 f2 f2 d2♯
d2♯ d2♯ d2♯ f2 f2 f2 f2 c3♯ c3♯ c3♯ c3♯ d2♯ d2♯ d2♯ d2♯ b2 b2 b2 b2 c2 c4
c2 c4 a1♯ a1♯ f2 f2 f2 f2 f2 f2 f3 f3 f2 f2 d2♯ d2♯ d2♯ d2♯ d3♯ d3♯ d2♯ d2♯
f2 a4 f2 a4♯ f3 a4 f2 a4♯ d2♯ a4 d2♯ a4♯ d3♯ g4 d2♯ g4♯ f2 f2 f2 c5 f3 f3
f2 g4♯ d2♯ d2♯ d2♯ d2♯ d3♯ d3♯ d2♯ d2♯ f2 a4 f2 a4♯ f3 a4 f2 a4♯ d2♯ a4 d2♯
a4♯ d3♯ g4 d2♯ g4♯ f4 c4 a3 f3 d4♯ a3♯ g3 d2♯ c5 a4 f4 c4 a3♯ g3 d3♯ d2♯

What key is it in?

A fair question. At first I tried to stare down the notes until they buckled and gave up the goods. My knowledge of music theory is limited however and any candidate key I picked still had more accidentals than I felt comfortable with. I decided to whip up some code which compares the notes in the theme to all of the major keys and counts up the accidentals. With a bit of luck this would narrow down the candidates considerably. This article is getting a little code heavy and there’s a bit of repetition, so I’ll just show the additional code.

Firstly I added code the count the notes in the melody. All octaves of a note are counted as the same note (so a1 and a2 just count as a).

class Spectrum(object):
    def __init__(self):
        self.notes = [0]*12

    def add(self, n):
        self.notes[n%12] += 1

    def __str__(self):
        s = ""
        for i in range(0, 12):
            s += names[i].ljust(2)+" : "+str(self.notes[i])+"\n"
        return s

Next a class to generate the notes in all of the major keys (and the notes which aren’t in them):

class Keys(object):
    def __init__(self):
        maj = [0, 2, 2, 1, 2, 2, 2]
        self.keys = {}
        for t in range(0, 12):
            s = []
            last = t
            for i in maj:
                s.append((last+i)%12)
                last += i
            self.keys[t] = s

    def notes_in(self, k):
        return self.keys[k]

    def notes_not_in(self, k):
        return [n for n in range(0, 12) if n not in self.keys[k]]

So we get a Spectrum (s below) object, feed it all the notes then run this:

names = ['a', 'a♯', 'b', 'c', 'c♯', 'd', 'd♯', 'e', 'f', 'f♯', 'g', 'g♯']
keys = Keys()
for k in range(0, 12):
    acc = 0
    for nkn in keys.notes_not_in(k):
        acc += s.notes[nkn]
    print(names[k].ljust(2)+" : "+str(acc))

Here’s the results:

a  : 213
a♯ : 26
b  : 142
c  : 105
c♯ : 38
d  : 207
d♯ : 33
e  : 150
f  : 90
f♯ : 50
g  : 200
g♯ : 26

We’ve got two stand outs here: a♯ major and g# major (normally you’d call these B♭ and A♭ major). I actually think it sounds minor. The relative minor keys which correspond are G minor and F minor. Of these I went with F minor after some more note gazing.

Given this key the accidentals should be flats:

Voice 1
-------
f3 a3 c4 f4 g3 b3♭ c4 g4 d4♭ e4♭ f4 a♭4 e4♭ d5 e4 c5 f3 f4 c3 g3 e3♭ g4 g3 e3♭ f3 f4 c3 g3
d4♭ f5 f4 d4♭ e3♭ e4♭ b2♭ f3 b3 e5♭ e4♭ b3 c3 e4 d3 f4 b3♭ b3♭ b4♭ b3♭ f4 f4 f4 f4 f4 f4 f4
f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 e4♭ e4♭ e4♭ e4♭ f4 f5 f4 e5♭
f4 d5 f4 c5 e4♭ e5♭ e4♭ e5♭ e4♭ b4♭ e4♭ e5♭ f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 f4 e4♭ e4♭ e4♭
a4 f4 c4 a3 g4 e4♭ b3♭ e3♭ a4 f4 c4 a3 g4 e4♭ b3♭ e3♭ 

Voice 2
-------
f2 c3 f3 a♭3 e2♭ d3 e3♭ b3♭ d2♭ d2♭ d3♭ d2♭ e3♭ b4♭ e3 a♭4 f2 f2 f2 f2 e2♭ e2♭ e2♭ e2♭ f2 f2
f2 f2 d3♭ d3♭ d3♭ d3♭ e2♭ e2♭ e2♭ e2♭ b2 b2 b2 b2 c2 c4 c2 c4 b1♭ b1♭ f2 f2 f2 f2 f2 f2 f3
f3 f2 f2 e2♭ e2♭ e2♭ e2♭ e3♭ e3♭ e2♭ e2♭ f2 a4 f2 b4♭ f3 a4 f2 b4♭ e2♭ a4 e2♭ b4♭ e3♭ g4 e2♭
a♭4 f2 f2 f2 c5 f3 f3 f2 a♭4 e2♭ e2♭ e2♭ e2♭ e3♭ e3♭ e2♭ e2♭ f2 a4 f2 b4♭ f3 a4 f2 b4♭ e2♭
a4 e2♭ b4♭ e3♭ g4 e2♭ a♭4 f4 c4 a3 f3 e4♭ b3♭ g3 e2♭ c5 a4 f4 c4 b3♭ g3 e3♭ e2♭ 

Tempo

The music routine is called twice per frame but only does anything of consequence every other call. There’s some weirdness going on there. Why not just call the routine once per frame instead of calling it twice and having it do nothing half of the time? Function calls aren’t free.

MusicTickRoutine:
83ac   a5 c1      LDA TitleScreenFrameCountMod4
83ae   29 01      AND #$01
83b0   d0 01      BNE $83b3
83b2   60         RTS 

So on every other call:

83b3   ad 0c 98   LDA MusicTickRoutine__Voice1SustainLevel
83b6   c9 a0      CMP #$a0
83b8   d0 43      BNE __MusicTickRoutine__TweakCurrentNote
__MusicTickRoutine__NextNote:
; Set voice 1&2 to triangle wave.
83ba   a9 10      LDA #$10
83bc   8d 04 d4   STA Sid_Voice1Ctrl
83bf   8d 0b d4   STA Sid_Voice2Ctrl
 .
 .
 .
__MusicTickRoutine__TweakCurrentNote:
83fd   ad 0c 98   LDA MusicTickRoutine__Voice1SustainLevel
 .
 .
 .
8413   ad 0c 98   LDA MusicTickRoutine__Voice1SustainLevel
8416   18         CLC 
8417   69 01      ADC #$01
8419   29 a7      AND #$a7
841b   8d 0c 98   STA MusicTickRoutine__Voice1SustainLevel
 .
 .
 .

OK, there’s some weirdness going on here too. Whether the routine advances to the next note (actually the next note for each of the two voices used) or just modulates the amplitude of voice one is determined by MusicTickRoutine__Voice1SustainLevel. This is initialised to $a0 before the music starts and a new note is only played when it’s $a0. The lower three bits are incremented every other call but the $a in the high nybble is entirely pointless. The upshot of all this is the song advances to the next note every 8 frames. We’ll use quavers (eighths notes) for the transcription. Given that PAL has around 50 frames per second this gives us a BPM of 187.5. We’ll call it 188.

Notes on a staff bitch!

OK, OK, I’m getting to it. Writing music engraving software from scratch would obviously be a major undertaking and breaking out my ruler is just offensive. I fired up Google and starting searching for some free software that could do the deed. Something that had an interactive user interface was not a requirement; I needed it to consume a text based file that my software could generate. Initially I came across VexFlow. It had limitations, but while reading various message boards trying to figure out how to resolve these issues I came discovered LilyPond. Holy shit! The stuff they give away for free these days. The other building block I needed was a Python based templating system. My go to for this kind of thing is Jinja2. So firstly we need the notes in a form that LilyPond can consume. I added this code:

lilynames_sharp = ['a', 'as', 'b', 'c', 'cs', 'd', 'ds', 'e','f', 'fs', 'g', 'gs']
lilynames_flat  = ['a', 'bf', 'b', 'c', 'df', 'd', 'ef', 'e','f', 'gf', 'g', 'af']

def index_to_lily(i, sharp):
    names = lilynames_sharp if sharp else lilynames_flat
    octave = floor((i-3)/12)+2
    if octave<0:
        octave = ','*-octave
    else:
        octave = "'"*octave
    return names[i%12]+octave

See here and there for information on these note naming conventions.

Next we need a skeletal Lilipond file and to hook up Jinja2 to fill in the blanks. The hooking up side of things is pretty simple:

# import jinja2

def render(env, template_name, **template_vars):
    template = env.get_template(template_name)
    return template.render(**template_vars)

with open("Boulder_Dash.ly", "w", encoding='utf-8') as of:
    v1 = ""
    for n in voice1():
        sid = note_to_sid(n)
        f = reg_to_freq_pal(sid)
        i = round(freq_to_note(f))
        v1 += index_to_lily(i, False)+"8 "
    v2 = ""
    last_note = 0
    for n in voice2():
        sid = note_to_sid(n)
        f = reg_to_freq_pal(sid)
        i = round(freq_to_note(f))
        s.add(i)
        if i==last_note:
            v2 += "~"
        v2 += " "
        v2 += index_to_lily(i, False)+"8"
        last_note = i

    env = jinja2.Environment(loader=jinja2.FileSystemLoader('.'))
    env.globals['voice1'] = v1
    env.globals['voice2'] = v2
    env.globals['key'] = r"\key f \minor"
    env.globals['tempo'] = r"\tempo 4 = 188"
    s = render(env, 'bd.ly')
    of.write(s)

And finally our skeleton:

\version "2.16.0"

\header {
  title = "Boulder Dash (C64)"
  composer = "Peter Liepa"
  arranger = "Transcribed by Stephen Hewitt"
  tagline = "" % removed
}

\language english

\score
{
    \new PianoStaff \with { instrumentName = "SID chip" }
    <<
        \new Staff = "up"
        {
            <<
                \new Voice = "voice1"
                {
                    \voiceOne
                    {{key}}
                    {{tempo}}
                    \set midiInstrument = #"acoustic guitar (steel)"
                    \autochange
                    \repeat volta 2
                    {
                        {{voice1}}
                    }
                }

                \new Voice = "voice2"
                {
                    \voiceTwo
                    {{key}}
                    \set midiInstrument = #"electric guitar (jazz)"
                    \autochange
                    \repeat volta 2
                    {

                        {{voice2}}
                    }
                }
            >>
        }

        \new Staff = "down"
        {
            \clef bass
            {{key}}
        }
    >>

    \layout { }
    \midi
    {
        \context
        {
            \Staff
            \remove "Staff_performer"
        }
        \context
        {
            \Voice
            \consists "Staff_performer"
        }
    }
}

This gives us a PDF and a MIDI file.

Boulder Dash music