The Bloody SPC-700
A try to stumble into the inner secret of a nasty chip.
By Antitrack exclusively for the FAMIDEV development group.
* The SPC 700 is a very stupid sound chip with about the worst handling
that you have seen in your lifetime.
* This chip is a co processor. He has a quite large instruction set
(contrary to the Amiga's COPPER, who has a very small one) and 64KB
RAM memory, of which you can use atleast 32KB. (or so)
* All program and data that is supposed to be run by this chip must
be' moved to the SPC's own ram with a small loop that pokes each byte of
your SPC assembler program and (e.g. sample-)data into four memory
locations : $2140 - $2143. They are your only chance to communicate
with the SPC.
* These four memory locations have different meanings for read and
if you read (LDA) $2140, you get the data from memory loc. 00f4 (or
so) of the sound chip.
* On power-on, the SPC 700 jumps (much like the main processor) to a
very small ROM area that resides from $ffc0 to $ffff inside the SPC.
(This chip REALLY follows the black box principle, eh...) This
program at $ffc0 is waiting to get the data in the right format on his
input ports at $00f4/5/6/7 , which are $2140/1/2/3 from the 65c816's (e.g.
your's ) point of view.
* Your main program will therefore have to follow the SPC's
conditions and poke all the program and data for the SPC into 2140/1/2/3 in a
* When transmission is completed, you will also have transmitted the
start address of your SPC code, and the SPC will start to execute your
Q: How do I move my program and data to the SPC then, what format do
I have to use?
A: First, your SPC data/code has to be moved from ROM to the extra
RAM at e.g. $7f0000 . Dont ask me why it has to be in RAM, probably it doesnt
but all the existing routines that send data to the SPC do something like
Your data/code has to be in groups which I will call "chunks". A
alid chunk looks like that:
first word: number of bytes to transmit to SPC
sec. word : start address where to move data to the SPC | one chunk
byte 4-???? : your data/code -+
You can have as many chunks as you want to , but the last chunk must
be like that:
first word : $0000
second word: Start address of your code.
Q: So if you are right, this means: After I transmitted all my code
and data, and my own SPC code takes over the control, I might encounter
problems if my SPC program has to communicate with the outer world (the
65c816).What if the main program wants to change sounds? What if a background
melody shall always play on two voices, and extra two voices will be used for
sound effects whenever the player sprite e.g. picks up an object?
A: That is sure a point. Your own code will have to look at memory
locations $00f4/00f5/00f6/00f7 , because they are the only accessible from
outside at $2140/1/2/3. The easiest way would be: As soon as any of $f4-$f7
change, jump into the Boot ROM at $ffc0 (?) so the SPC is executing his
receive routine again. Then you *probably* can send another SPC chunk with new
sound and code to the SPC....
Q: This only helps if a complete new tune is to be played, this
doesnt help if a melody using two voices shall still remain....
A: Thats true. The best approach is to send own command bytes to the
SPC and your SPC code has to check out $f4-$f7 constantly and react to it.....
A command byte like $00 could mean: sound off,
$01 : play tune 1
$0f : play tune $0f
$10 : play jingle (fx) 01
$ff : jump to $ffc0 (??) the receive
Q: is there another approach?
A: Yes there is. As you probably know, all important addresses of the
SPC 700 reside inside its own RAM's zeropage:
Address / register
0000 Volume left
0001 Volume right
0002 Pitch low
0003 Pitch high (The total 14 bits of pitch
0004 SRCN Designates source number from 0-
0005 ADSR 1
0006 ADSR 2
0007 GAIN Envelope can be freely designated by
0008 ENVX Present val of envelope with DSP
0009 VALX Present wave height val
(and so on...)
Your approach would be to move only sample data there, and/or (lots
small chunks of data with a target address in the zeropage, and a
address of e.g. $ffc0. The small chunks would access zeropage
for the volume etc and thus result in tones; if this is done every
you might end up with a music player quite similar to the C64 styled
Q: So anyway, in what format exactly do I have to move data to the
A: I have the following source code for you, but let me explain it a
BEFORE you start to dig into it.
I've already mentioned the general "chunk" format. The loop does the
- move ram destination address to $2142/3 (akku: 16 bit)
- move either #$00 or #$01 into 2141, this depends if you have more
bytes of data for the SPC;
- first time (first chunk you transmit): move constant #$cc into 2140
- loop: poke each byte that you want to be transmitted into 2140
the higher 7-15 bits of your accu-word contain the number of bytes
moved (e.g. 00 on the start)
- cmp $2140 with this number of bytes already moved (lower 8 bits of
number only!) and wait if its not equal.
- until the loop is over.
- for the next chunk header this is repeated, but not #$cc is moved
2140 but "nn" (lobyte of number of bytes moved) +3 or +6 if it was
+3 was used.
move #$0400 to 2142 /word access
move #$01 to 2141
move #$cc to 2140
move "gg00" to 2140 where
"gg" is the first real code/data
wait till 2140 is #$00
move hh01 to 2140 where "hh"
is the second byte of code or
data for SPC
wait till 2140 is #$01
move ii02 to 2140 where "ii"
is the 3rd byte of data for the
wait till 2140 is #$02
lets say "ii" was the last byte.
Now we add #$04 (3+carry) to
(#$02 being the number-1 of how many bytes we moved to the
will push it onto the stack), now :
fetch the next header , poke target
RAM address into $2142
poke 00 or 01 into 2141 depending of how many bytes to send,
poke #$06 into 2140 (06 : number of bytes sent from last chunk-
1 + 3 )
I think I got this scheme pretty much right this time. Now, is PLEASE
going to donate their home-brewed SPC dis/assemblers to me? Oh pretty
I hate silent SNES's ! :)
Source code follows, reassembled from a PAN/Baseline demo "xmas wish
; entry to the code starts here
#$30 ; x y a set to 8 bit length
LDA #$FF ; ff into audio0w (write)
REP #$10 ; x,y: 16 bit length
l0DB5B LDA $018000,X ; move rom music data to ram at $7f0000
LDA $028000,X ; move rom music data to ram at $7f0000
LDA #$80 ; screen on , probably not important at all
LDA #$00 ; 00fd/00fe/00ff point to the data that is
STA $00FD ; in ram at $7f0000
STZ $4200 ; disable nmi and timer h/v count
SEI ; disable irq
l0DBCD ; unknown sub routine, labeled "RESTART"
#$30 ; all regs 8 bit
l0DB8B LDA $2140 ; wait for reply from sound chip ?
LDA #$E0 ; audio3w ?
LDA #$FF ; send data to sound chip ?
STA $2142 ; $ffe0 this could be an address within the
; sound chip ROM between $ffc0 and $ffff
; ROM mask.......
LDA #$01 ; send data to sound chip ?
LDA #$01 ; send data to sound chip ?
l0DBA4 LDA $2140
; wait for reply from sound chip ?
CMP #$01 ; what a fuck of a protocol .... :(
l0DBAB LDA $2140
; wait again for reply from soundchip ?
$0207 ; aha ... move $0207 to sound
STA $2141 ; probably sound number selector
STA $2140 ; send data to sound chip
l0DBBD LDA $2140 ; wait until sound chip accepted data?
l0DBC4 LDA $2140 ; wait for reply ?
; labeled "RESTART" by pan/ATX
JSR l0DBD8 ;
LDA #$00 ; 00 into audio0w
REP #$30 ; a,x,y 16 bit regs
LDY #$0000 ; needed first time at lda [$fd],y :
pointer to ram
l0DBE1 CMP $2140 ; wait for sound chip $2140/2141 ?
SEP #$20 ; akku 8 bit
BRA l0DC12 ; oh well, another mystery :-)
; jump here if overflow is set e.g. if more than $0100 data to
l0DBEC LDA [$FD],Y ; get data from ram pointer
INY ; the accumulator is about to get "xx00"
XBA ; /"xx" is the byte from [fd],y (first
LDA #$00 ; /and resides into bit 15-7 of accu,
and 00 is
BRA l0DBFF ; /#$00 (8bit number of bytes already
; accu is now "nn??" ?? is old data from
LDA [$FD],Y ; accu is now "nnxx" with xx the newest
INY ; /for
XBA ; accu is now "xxnn"
l0DBF9 CMP $2140 ; wait for sound chip to reply with "nn" !!
INC A ; increment number of bytes that were
; accu is now "xxnn" with newest val for
l0DBFF REP #$20
; akku 16 bit
STA $2140 ; poke "xxnn" to soundchip. xx is actual
SEP #$20 ; akku 8 bit ! nn is the 8-bit cutted
number of bytes
DEX ! which were already sent!!
BNE l0DBF4 ; as many times as xreg says...
l0DC09 CMP $2140
; byte "nn" will be replied from the SPC if
BNE l0DC09 ; received correctly!
l0DC0E ADC #$03 ; compare accu with #$fb ADC WILL ADD #$04
; CARRY IS ALWAYS SET AFTER THE CMP!!!
BEQ l0DC0E ; if accu was $fb then accu := $03 . (what
; push value accu+$04 to stack (or
REP #$20 ; accu = 16 bit
LDA [$FD],Y ; get ram data 2 bytes
INY ; point to next word
TAX ; x:=a : number of bytes to transmit
LDA [$FD],Y ; get ram data
STA $2142 ; audio2w : possibly the dest. area in the
SEP #$20 ; accu 8 bit
CPX #$0100 ; set carry if first ram data was >= 0100
lda #$00 ;
STA $2141 ; if ram data >= 0100, poke "1" into reg 1
ADC #$7F ; SET OVERFLOW FLAG IF X>=$0100 !!!! (nice
STA $2140 ; $cc in the first case , nn+4 on all later
l0DC32 CMP $2140 ; wait
for snd chip reply
BVS l0DBEC ; if there were more than $0100 data for the
; move them where they R supposed to belong
STA $2140 ; same shit, never been jumped into
l0DC3F CMP $2140
; also lets look at 7f0000: the first few bytes at 7f0000 are:
7f0000: b7 0e 00 04 20 cd cf bd e8 00 5d af c8 f0 d0 fb 5d d5 00 01
d5 00 02
b7 0e should be number of bytes to transmit, 0400 the destination
at this point I really need an SPC dis/assembler..... :(((
Okay well my first source was incompetent, sure thing. But I think I
solve a lot of questions meanwhile.