CHAPTER 3 : CODING SOUND
This little part is aimed at the DMA-sound of the STe and newer machines.
I diliberately chose to skip the Yamaha soundchip, because this is
actually a bit more diffcult than the DMA-chip. The old Yammy can only
produce crap sound anyway... (though what some people got out of that thing
is incredible!!) Take a peek at chapter 11 to find out a bit more about the
Yammy there.
Let's say we want something like a sample of an explosion coming out of
your speakers/monitor. What do you need:
* a signed 8-bit sample at 12.5, 25 or 50 KHz
* an STe or better (TT/Falcon)
* a piece of assembly code to control the DMA chip
I just suppose you all have the required machine, otherwise you can just
skip this chapter. You can get samples anywhere and convert them to the
above specifications. Now you need to include the sample in your code like
this:
sample: INCBIN D:\SOUNDS\EXPLO.SPL
^ ^ ^
| | |
labelname actual command path/name of sample
This will include the binairy data of the file into the assembled
executable. It's quite easy and you can do it with pictures, sprites and
text too.
OK, now you need the rest of the code. First you need to include the
supervisor-mode routines listed in chapter 1. In order to access the DMA-
chip you'll need to be in supervisor-mode.
Now comes the important part. The actual controlling of the DMA-chip and
getting it to play our sample. The chip is represented by various hardware-
registers that are located in high memory. I'll explain some here:
$ffff8900: DMA-control register. This is a word that contains bits that can
be set that let the chip begin to play a sample or loop a sample. Here's
the exact layout:
+------- bit 1: loop sample (1=on 0=off)
|+------ bit 0: play sample (1=on 0=off)
||
+----------------+
|iiiixxxxxxxxxxlp|
+----------------+
||||
bits 12 to 15:
interrupt thingies (not really interesting for us)
Note: I've not included the falcon-extended bits. I just want to show you
the basics and not all the details.
$ffff8921: sound-mode control register. This is a byte that contains the
bits that control the replay frequency and stereo/mono stuff.
++--11 = 50 KHz
++--10 = 25 KHz
++--01 = 12.5 KHz
++--00 = 6.25 KHz
||
+--------+
|sxxxxxff|
+--------+
|
bit 7: stereo/mono select (1=mono 0=stereo)
$ffff8903: frame start address - high byte: contains bits 16 to 23 of the
24 bits sample beginaddress.
$ffff8905: frame start address - mid byte: Same as high byte only for bits
8 to 15.
$ffff8907: frame start address - low byte: Same as high byte only for bits
0 to 7.
$ffff890f: frame end address - high byte: contains bits 16 to 23 of the
24 bits sample endaddress.
$ffff8911: frame end address - mid byte: Same as high byte only for bits
8 to 15.
$ffff8913: frame end address - low byte: Same as high byte only for bits
0 to 7.
Now that I've explained all the usefull registers let's get on with the
code. First I need to note that in order to play the sample you need to give
the frame-end-address-register the endaddress of your sample. You get the
address by doing this:
sample: INCBIN D:\SOUND\ARSEWIPE.SPL
end:
That's right! Simply put a label behind your incbin and there you have it!
Now let's get on with the rest of the code:
* First we set the sound-mode control register. We set it to mono, 25 KHz.
* We set bit 7, set the bit 1 and clear bit 0.
andi.b #%11111110,$ffff8921
ori.b #%10000010,$ffff8921
* Then we set the startaddress
move.l #sample,d0 * Move startaddress into d0.
swap d0 * Bring the high byte in d0.b.
* I moved the high byte into d0.b because first of all we MUST do the high
* byte!
move.b d0,$ffff8903 * Move the high byte of the address.
* Then we do the mid byte of the address.
rol.l #8,d0 * Bring the mid byte in d0.b.
move.b d0,$ffff8905 * Move the mid byte of the address.
* Finally, we do the low byte.
rol.l #8,d0 * Bring the low byte in d0.b.
move.b d0,$ffff8907 * Move the low byte of the address.
* OK, the startaddress is done, now for the endaddress.
* It's basicly the same as the startaddress, but with a different label and
* different registers.
move.l #end,d0 * Move endaddress into d0.
swap d0 * Bring the high byte in d0.b.
* I moved the high byte into d0.b because first of all we MUST do the high
* byte!
move.b d0,$ffff890f * Move the high byte of the address
* Then we do the mid byte of the address.
rol.l #8,d0 * Bring the mid byte in d0.b.
move.b d0,$ffff8911 * Move the mid byte of the address.
* Finally, we do the low byte.
rol.l #8,d0 * Bring the low byte in d0.b.
move.b d0,$ffff8913 * Move the low byte of the address.
* Now to trigger the playback of the sample.
* We set bit 0 to trigger playback and clear bit 1 to turn looping off.
andi.w #%1111111111111101,$ffff8900
ori.w #%0000000000000001,$ffff8900
That's it! You replayed a sample. This is so simple! Just compare it to
a system like a PC.
About the order of high/mid/low bytes: the DMA-chip expects you to first
do the high one and later on the lower ones. Otherwise it just won't work!
There you have it. But now I'm gonna go a bit deeper into the whole thing.
If you think you're up to editing samples with programs you can continue
reading....
OK, that's all very nice, but what if we want two or maybe four channel
sound just like in those groovy MOD-thingies. Then you need to mix a few
samples together. What you do is average every byte in the samples and
make a new sample of it.
Hold on, stop, stop, whats goin' on here! Oh, you don't understand, I'll
explain a bit better. Every sample consists of load of bytes. Each byte
represents the heigth of the soundwave at a certain moment. Like this:
This is a normal soundwave(in a funky ASCII drawing):
-- ^
/ \ | amplitudo (heigth of soundwave)
/ \ |
|------|------ ---> time
\ /
\ /
--
Now if you sample this you get the following bytes:
$00,$28,$50,$7f,$7f,$50,$28,$00,-$28,-$50,-$80,-$80,-$50,-$28,$00
Every byte simply represents the amplitudo at each interval. OK, that's the
basics of sampling. But now you want to mix. The only thing you need to do
now is add the amplitudo's of the samples to eachother and divide them
through the number of channels your using. Suppose you wanted to mix two
channels together:
amplitudo of sample 1 at interval 0 = $34
amplitudo of sample 2 at interval 0 = $1a
Now you add them together: $34 + $1a = $4d (just some hexa-calculation)
(if you don't know what that)
(is, please skip this theory!)
Now divide it by the number of channels: $4d/2 = $27
Now do this for every byte in the samples and you have your mixed sample
that you can easily playback with your DMA. The practical aproach to this
would look like:
move.l #sample1,a0 * Move startaddress of spl1 into a0.
move.l #sample2,a1 * Move startaddress of spl2 into a1.
move.l #d_sample,a2 * Move " of destination spl into a2.
move.w #10000-1,d7 * Suppose the samples have 10000 bytes.
moveq #0,d1 * Clear d1 for use as adding register.
moveq #0,d0 * Clear d0 for use as adding register.
loop: move.b (a0)+,d0 * Put amplitudo of spl1 in d0.
move.b (a1)+,d1 * Put amplitudo of spl2 in d1.
add.b d1,d0 * Add them.
lsr.w #1,d0 * Divide it by 2 (shift 1 bit right).
move.b d0,(a2)+ * Move result in destination sample.
dbra d7,loop * Loop 10000 times.
Now you can put your playback routine underneath this and give it the
start/end-addresses of d_sample and there you have it!!
Some of you might be curious why I clear the registers first. Well, when
you add like $60 and $40 it should make $a0, right? Well, a byte cannot
hold this! You simply need an extra bit. But if this bit should happen to be
1 you'd get wrong results:
This is the situation with bit 8 cleared:
$60 + $40 = $a0
$a0/2= $50
This is the situation when bit 8 accidently isn't cleared:
bit 8 set!!
|
$160 + $40 = $1a0
$1a0/2 = $208 (The byte can only contain the
first two digits so you get $08!!)
As you can see the accidental being set of this can be catastrophic! Now
I've explained it... Phew, That's I guess.. I could have explained more
about frequencies, but would take up even more space and time. I'd like to
keep it a bit more basical. The summary:
Amplitudo: heigth of a soundwave at a certain interval. In a
sample this is represented with a numerical value.
DMA-chip: Soundchip in STe or higher that can play one 8-bit
stereo or mono sample at a time
DMA-registers: Registers you control the DMA-chip with. They are
located in high memory and you need supervisor-mode
to access them.
Funky ASCII drawing: A bunch of ASCII characters put next to eachother by
somekind of moron who calls it a drawing or even
'art'.
Incbin: Easy way to include binairy data (sample, text, etc.)
into your executable file (PRG,TOS,APP,etc.)
Mixing: Making more samples in to one sample by averaging
all the bytes.
Sample: A string of amplitudo values.
YM2149: Also known as the "Yammy". The oldest soundchip of the
ST. Basicly outdated when the ST was released. Can
produce some blips and blops and other noise.