SID File Format

sid.txt

                          ===========================
                          SID FILE FORMAT DESCRIPTION
                          ===========================

AUTHORS:
Michael Schwendt (PSID v1 and v2)
Simon White (PSID v2NG, RSID)
Dag Lem (PSID v2NG)
Wilfred Bos (PSID v3, RSID v3, PSID v4, RSID v4)
LaLa (This document)


INTRODUCTION
============

This document describes the SID file format used for SID tunes in the HVSC
(High Voltage SID Collection - http://hvsc.de). It is based mostly on
Michael Schwendt's document that describes the file format and the PSID v2NG
extensions described by Simon White and Dag Lem.

SID files use the .sid file extension.

Since PSID v2 is simply an extension of PSID v1, PSID v2NG is an extension of
PSID v2, RSID is a restricted version of PSID v2NG, PSID v3/v4 and RSID v3/v4
are extensions of PSID v2NG and RSID v2, all of the formats are discussed
together below. RSID in specific is discussed in detail under the 'magicID'
field description.

The information presented here targets programmers or other people with
reasonable background. It is not suitable for newbies who have never before
used a machine code monitor, a disassembler, or a hexadecimal editor.


The SID file header v1
======================

The detailed structure of the SID header looks like the following. Header
offsets are in hexadecimal notation. Other integer values are decimal unless
explicitly marked otherwise. Any stored integer values are in big-endian
format:

+00    magicID: 'PSID' or 'RSID'

This is a four byte long ASCII character string containing the value
0x50534944 or 0x52534944. 'RSID' (Real SID) denotes that the file strictly
requires a true Commodore-64 environment to run properly. 'PSID' files will
generally run trouble-free on older PlaySID and libsidplay1 based emulators,
too.

Some words about the Real C64 SID file format (RSID):

The RSID format was designed to contain tunes that are not PlaySID compatible,
but strictly require a real C64 environment to run. Tunes that are multi-speed
and/or contain samples and/or use additional interrupt sources or do busy
looping will cause older SID emulators to lock up or play very wrongly (if at
all).

By using the name RSID for such rips all existing SID emulators will reject
these tunes safely until they can be upgraded to cope with the additional
requirements.

Due to the nature of these tunes, every effort must be made to make sure they
are directly runnable on an actual C64 computer. As such the tunes will only
be presented with the default C64 power-on environment and expected to
configure and use all hardware appropriately.

RSID is based on PSIDv2NG with the following modifications:

magicID = RSID
version = 2, 3 and 4 only
loadAddress = 0 (reserved)
playAddress = 0 (reserved)
speed = 0 (reserved)
psidSpecific flag is called C64BASIC flag

The above fields MUST be checked and if any differ from the above then the
tune MUST be rejected. The definitions above will force tunes to contain
proper hardware configuration code and install valid interrupt handlers.

See section "The SID file environment" mentioned later in this document for the
default C64 power-on environment for each SID file format.

+04    WORD version

Available version number can be 0001, 0002, 0003 or 0004. Headers of version 2,
3 and 4 provide additional fields. RSID and PSID v2NG files must have 0002,
0003 or 0004 here.

+06    WORD dataOffset

This is the offset from the start of the file to the C64 binary data area.
Because of the fixed size of the header, this is either 0x0076 for version 1
and 0x007C for version 2, 3 and 4.

+08    WORD loadAddress

The C64 memory location where to put the C64 data. 0 means the data are in
original C64 binary file format, i.e. the first two bytes of the data contain
the little-endian load address (low byte, high byte). This must always be true
for RSID files. Furthermore, the actual load address must NOT be less than
$07E8 in RSID files.

You must be absolutely sure what to enter here. There is no way to detect
automatically whether the first two bytes in a C64 data file are meant to be a
load address or some arbitrary bytes of code or data. Unless your C64 file is
not a normal binary file and thus has no load address in front, you need not
enter anything else than 0 here. The SID tune will not play if you specify a
load address which is present in the C64 file already.

Normal C64 binary data files have a load address in their first two bytes, so
they can be loaded to a pre-defined destination address by executing
LOAD"FILE",8,1, for instance. If a load address is explicitly specified in the
sidtune info file, some sidtune converters and utilities conjecture that the
C64 data don't have a load address in their first two bytes. Hence, the
explicit load address from the info file is moved in front of the C64 data to
create a valid C64 binary file which can be easily loaded on a C64, too. If
that C64 file were to be saved, it would contain two superfluous data bytes at
offset 2 if an original load address had been in the first two bytes of the
old file. This process of adding a duplicate load address can be repeated. The
file loader strips off the first two bytes (the used load address) and puts
the rest of the file contents (including the now obsolete load address at file
offset 2) into memory. If the new load address is the same than the old one
the two added bytes cause the whole data to be displaced by two bytes, which
most likely results in malfunctioning code. Also, superfluous bytes in memory
then can confuse disassemblers which start at the beginning of the file or
memory buffer.

+0A    WORD initAddress

The start address of the machine code subroutine that initializes a song,
accepting the contents of the 8-bit 6510 Accumulator as the song number
parameter. 0 means the address is equal to the effective load address.

In RSID files initAddress must never point to a ROM area ($A000-$BFFF or
$D000-$FFFF) or be lower than $07E8. Also, if the C64 BASIC flag is set,
initAddress must be 0.

+0C    WORD playAddress

The start address of the machine code subroutine that can be called frequently
to produce a continuous sound. 0 means the initialization subroutine is
expected to install an interrupt handler, which then calls the music player at
some place. This must always be true for RSID files.

+0E    WORD songs

The number of songs (or sound effects) that can be initialized by calling the
init address. The minimum is 1. The maximum is 256.

+10    WORD startSong

The song number to be played by default. This value is optional. It often
specifies the first song you would hear upon starting the program is has been
taken from. It has a default of 1.

+12    LONGWORD speed

This is a 32 bit big endian number.

For version 1 and 2 and for version 2NG, 3 and 4 with PlaySID specific flag
(+76) set, the 'speed' should be handled as follows:

Each bit in 'speed' specifies the speed for the corresponding tune number,
i.e. bit 0 specifies the speed for tune 1. If there are more than 32 tunes,
the speed specified for tune 32 is the same as tune 1, for tune 33 it is the
same as tune 2, etc.

For version 2NG, 3 and 4 with PlaySID specific flag (+76) cleared, the 'speed'
should be handled as follows:

Each bit in 'speed' specifies the speed for the corresponding tune number,
i.e. bit 0 specifies the speed for tune 1. If there are more than 32 tunes,
the speed specified for tune 32 is also used for all higher numbered tunes.

For all version counts:

A 0 bit specifies vertical blank interrupt (50Hz PAL, 60Hz NTSC), and a 1 bit
specifies CIA 1 timer interrupt (default 60Hz).

Surplus bits in 'speed' should be set to 0.

For RSID files 'speed' must always be set to 0.

Note that if 'play' = 0, the bits in 'speed' should still be set for backwards
compatibility with older SID players. New SID players running in a C64
environment will ignore the speed bits in this case.

WARNING: This field does not work in PlaySID for Amiga like it was intended,
therefore the above is a redefinition of the original 'speed' field in SID
v2NG! See also the 'clock' (video standard) field described below for 'flags'.

+16    ``<name>''
+36    ``<author>''
+56    ``<released>'' (once known as ``<copyright>'')

These are 32 byte long ASCII character strings. Upon evaluating the header,
these fields may hold a character string of 32 bytes which is not zero
terminated. For less than 32 characters the string should be zero terminated.
The maximum number of available free characters is 32.

+76    <data>

Version 1 of the SID header is complete at this point. The binary C64 data
starts here.


The SID file header v2, v3 and v4
=================================

Version 2, 3 and 4 of the header incorporates the v1 header fields and provides
additional fields. Some of these are actually v2NG, v3 or v4 specific - those
are noted below.

+76    WORD flags

This is a 16 bit big endian number containing the following bitfields:

- Bit 0 specifies format of the binary data (musPlayer):
  0 = built-in music player,
  1 = Compute!'s Sidplayer MUS data, music player must be merged.

If this bit is set, the appended binary data are in Compute!'s Sidplayer MUS
format, and does not contain a built-in music player. An external player
machine code must be merged to replay such a sidtune.

- Bit 1 specifies whether the tune is PlaySID specific, e.g. uses PlaySID
  samples (psidSpecific):
  0 = C64 compatible,
  1 = PlaySID specific (PSID v2NG, v3, v4)
  1 = C64 BASIC flag (RSID)

This is a v2NG and RSID specific field.

PlaySID samples were invented to facilitate playback of C64 volume register
samples with the original Amiga PlaySID software. PlaySID samples made samples
a reality on slow Amiga hardware with a player that was updated only once a
frame.

Unfortunately, converting C64 volume samples to PlaySID samples means that
they can no longer be played on a C64, and furthermore the conversion might
potentially break the non-sample part of a tune if the timing between writes
to the SID registers is at all altered. This follows from the ADSR bugs in the
SID chip.

Today, the speed of common hardware and the sophistication of the SID players
is such that there is little need for PlaySID samples. However, with all the
PlaySID sample PSIDs in existence there's a need to differentiate between SID
files containing only original C64 code and PSID files containing PlaySID
samples or having other PlaySID specific issues. As stated above, bit 1 in
'flags' is reserved for this purpose.

Since RSID files do not have the need for PlaySID samples, this flag is used
for a different purpose: tunes that include a BASIC executable portion will
be played (with the BASIC portion executed) if the C64 BASIC flag is set. At
the same time, initAddress must be 0.

- Bits 2-3 specify the video standard (clock):
  00 = Unknown,
  01 = PAL,
  10 = NTSC,
  11 = PAL and NTSC.

This is a v2NG specific field.

As can be seen from the 'speed' field, it is not possible to specify NTSC C64
playback. This is unfortunate, since the different clock speeds means that a
tune written for the NTSC C64 will be slightly detuned if played back on a PAL
C64. Furthermore, NTSC C64 tunes driven by a vertical blank interrupt have to
be converted to use the CIA 1 timer to fit into this scheme. This can cause
severe problems, as the NTSC refresh rate is once every 17045 cycles, while
the CIA 1 timer A is latched with 17095 cycles. Apart from the difference in
timing itself, the SID ADSR bugs can actually break the tune.

The 'clock' (video standard) field was introduced to circumvent this problem.

- Bits 4-5 specify the SID version (sidModel):
  00 = Unknown,
  01 = MOS6581,
  10 = MOS8580,
  11 = MOS6581 and MOS8580.

This is a v2NG specific field.

- Bits 6-7 specify the SID version (sidModel) of the second SID:
  00 = Unknown,
  01 = MOS6581,
  10 = MOS8580,
  11 = MOS6581 and MOS8580.

This is a v3 specific field.

If bits 6-7 are set to Unknown then the second SID will be set to the same SID
model as the first SID.

- Bits 8-9 specify the SID version (sidModel) of the third SID:
  00 = Unknown,
  01 = MOS6581,
  10 = MOS8580,
  11 = MOS6581 and MOS8580.

This is a v4 specific field.

If bits 8-9 are set to Unknown then the third SID will be set to the same SID
model as the first SID.

The MOS6581 and the MOS8580 have three notable differences. First, combined
waveforms are generally louder on a MOS8580, to the extent that some
combinations that are clearly audible on a MOS8580 are completely silent on a
MOS6581. Second, the internal DC levels in the MOS8580 are so small that
software or hardware tricks must be used to play volume samples. Third, the
MOS8580 analog filter has totally different characteristics from the MOS6581
analog filter.

To ensure that music specifically written for one of the two SID versions can
be played back correctly, bits 4-9 in 'flags' are used as stated above.

- Bits 10-15 are reserved and should be set to 0.

+78    BYTE startPage (relocStartPage)

This is a v2NG specific field.

This is an 8 bit number. If 'startPage' is 0, the SID file is clean, i.e. it
does not write outside its data range within the driver ranges. In this case
the largest free memory range can be determined from the start address and the
data length of the SID binary data. If 'startPage' is 0xFF, there is not even
a single free page, and driver relocation is impossible. Otherwise,
'startPage' specifies the start page of the single largest free memory range
within the driver ranges. For example, if 'startPage' is 0x1E, this free
memory range starts at $1E00.

+79    BYTE pageLength (relocPages)

This is a v2NG specific field.

This is an 8 bit number indicating the number of free pages after 'startPage'.
If 'startPage' is not 0 or 0xFF, 'pageLength' is set to the number of free
pages starting at 'startPage'. If 'startPage' is 0 or 0xFF, 'pageLength' must
be set to 0.

The relocation range indicated by 'startPage' and 'pageLength' should never
overlap or encompass the load range of the C64 data. For RSID files, the
relocation range should also not overlap or encompass any of the ROM areas
($A000-$BFFF and $D000-$FFFF) or the reserved memory area ($0000-$03FF).

+7A    BYTE secondSIDAddress

This is a v3 specific field. For v2NG, it should be set to 0.

This is an 8 bit number indicating the address of the second SID. It specifies
the middle part of the address, $Dxx0, starting from value 0x42 for $D420 to
0xFE for $DFE0). Only even values are valid. Ranges 0x00-0x41 ($D000-$D410) and
0x80-0xDF ($D800-$DDF0) are invalid. Any invalid value means that no second SID
is used, like 0x00.

+7B    BYTE thirdSIDAddress

This is a v4 specific field. For v2NG and v3, it should be set to 0.

This is an 8 bit number indicating the address of the third SID. It specifies
the middle part of the address, $Dxx0, starting from value 0x42 for $D420 to
0xFE for $DFE0). Only even values are valid. Ranges 0x00-0x41 ($D000-$D410) and
0x80-0xDF ($D800-$DDF0) are invalid. Any invalid value means that no third SID
is used, like 0x00.
The address of the third SID cannot be the same as the second SID.

+7C    <data>

Version 2, 3 and 4 of the SID header ends here. This offset is the start of the
binary C64 data. See also 'loadAddress' for what the first 2 bytes of 'data'
might indicate.


The SID file environment
========================

Before the data of a SID file is loaded in memory of a C64, certain addresses
and chips must be initialized in order to play the SID tune correctly.

For RSID and PSID files the following address must be set for an emulated
environment:

$02A6 : depending on the PAL/NTSC flag in the SID file header, it is set to
        0x01 for PAL and set to 0x00 for NTSC.

On a real C64, when the speed flag is set to CIA, the value of $02A6 should be
read to setup the CIA timers accordingly (0x4025 for PAL and 0x4295 for NTSC).
When the speed flag is set to CIA, and a NTSC SID tune is played on a PAL
machine or a PAL SID tune is played on a NTSC machine, the player may set up a
CIA timer instead of VBI to run the tune at the correct speed.

The following CIA timer values can be used for:

NTSC tune on PAL machine: 0x3FFB (312 * 63 * 50 / 60 - 1)
PAL tune on NTSC machine: 0x5021 (263 * 65 * 60 / 50 - 1)

The default C64 environment for PSID files is as follows:

VIC           : IRQ set to any raster value less than 0x100. Enabled when
                speed flag is 0, otherwise disabled.
CIA 1 timer A : set to 60Hz (0x4025 for PAL and 0x4295 for NTSC) with the
                counter running. IRQs active when speed flag is 1, otherwise
                IRQs are disabled.
Other timers  : disabled and loaded with 0xFFFF.

When the init and play addresses are called the bank register value must be
written for every call and the value is calculated as follows:

if   address < $A000  -> 0x37 // I/O, Kernal-ROM, Basic-ROM 
else address < $D000  -> 0x36 // I/O, Kernal-ROM
else address >= $E000 -> 0x35 // I/O only
else                  -> 0x34 // RAM only

The default C64 environment for RSID files is as follows:

VIC           : IRQ set to raster 0x137, but not enabled.
CIA 1 timer A : set to 60Hz (0x4025 for PAL and 0x4295 for NTSC) with the
                counter running and IRQs active.
Other timers  : disabled and loaded with 0xFFFF.
Bank register : 0x37

A side effect of the bank register is that init MUST NOT be located under a
ROM/IO memory area (addresses $A000-$BFFF and $D000-$FFFF) or outside the
load image. Since every effort needs to be made to run the tune on a real
C64 the load address of the image MUST NOT be set lower than $07E8.

If the C64 BASIC flag is set, the value at $030C must be set with the song
number to be played (0x00 for song 1).