Clicks with "Ern0 - Gitar"

[mschwendt]

Clearly visible in a sample viewer and audible, too. It's been like that since v1.0.0.

[mschwendt]

Hunting starts with a look at the samples in Audacity. A bit after 2.870 one can see an usual spike that could be the culprit:

[mschwendt]

The samples around 2.880 and 2.900 are suspicious, too, because such a sudden decrease in volume can cause clicks.

BZR Player 2.0.89 is affected, too, but its TFMX support is based on UADE, so that makes this issue more interesting.

[mschwendt]

The original music is available as Danubius Replay by ERNO (sometimes called Ern0 or Erno, too) in the music demo Musication Vol.1 (Musicdisk) by Absolute!. https://janeway.exotica.org.uk/release.php?id=8799

Loaded that in WinUAE, and a quick re-rip of the MDAT and SMPL comes up with the same checksums as the Gitar track in some music collections. Except the first bytes of the TFMX header have been invalidated. The pointers to MDAT (0x1a0b2) and SMPL (0x1bfee) are found in the dwords at $cda0 and $cda4. The TFMX player used by that music demo uses those addresses.

[mschwendt]

Looking at the libtfmxaudiodecoder runtime debug output, the music starts with playing instrument 1.

What's peculiar in the instrument's macro script is the length of the sample. That's 20% of the samples data! From beginning to end it last several seconds depending on sample frequency. And the reason is it's not an instrument but the entire guitar backing music.

  0000 00 000000 DMAoff+Reset (stop sample & reset all)
  0001 02 000062 SetBegin    xxxxxx   sample-startadress
  0002 03 00608a SetLen      ..xxxx   sample-length
  0003 0d 00003f Addvol+note xx/fe/xx note/CONST./volume
  0004 09 250005 SetNote     xx/xxxx  note/detune
  0005 01 000000 DMAon (start sample at selected begin)
  0006 04 000000 Wait        ..xxxx   count (VBI's)
  0007 19 000000 -------Set one shot sample-------------
  0008 04 000000 Wait        ..xxxx   count (VBI's)
  0009 07 000000 -------------STOP----------------------

[mschwendt]

After almost three seconds, that guitar line is restarted via a new note. Shortly before the end of the specified sample data range, there's that spike. So, it's caused by simply replaying those samples. It's not a TFMX effect or such.

What else is suspicious? The SetBegin offset of the first instrument at 0x62. Typically, TFMX macro scripts start at 0 or 4 (after the first two silent sample words).

Redirecting the audio output to a WAV file in WinUAE, no click can be heard or seen. There's all sorts of crackling and noises while watching the music demo, but no such spike can be seen in a WAV editor. The lesser volume area around 2.880 to 2.900 is present, too.

[mschwendt]

In smpl.gitar that sample range is from 0x62 to 0x62+(0x608a*2)=0xc176.

At 0xc118, i.e. shortly before its end, a single sample 0x80 stands out. It's likely the one that causes the spike. Notice how all the other samples are lower and more of a smooth curve:

0000c110  24 0e 30 3e 32 32 4e 2c  80 28 1a 2c 2c 2c 2c 28  |$.0>22N,.(.,,,,(|
                                   ^^
0000c120  2c 32 3c 32 3e 3a 44 40  44 38 3a 44 3c 26 32 3e  |,2<2>:D@D8:D<&2>|
0000c130  4e 28 20 38 32 38 32 26  0e 30 40 34 34 4e 2c 44  |N( 8282&.0@44N,D|
0000c140  52 44 32 28 48 1e 32 32  44 4e 2a 32 22 32 0c 26  |RD2(H.22DN*2"2.&|
0000c150  1e 1e 2c 42 32 1e 26 2a  10 10 12 1a 3a 14 1e 42  |..,B2.&*....:..B|
0000c160  1c 28 4a 3e 28 3c 44 58  58 4c 44 44 44 28 38 32  |.(J>(<DXXLDDD(82|
0000c170  26 1e 32 20 3e 14 22 3a  32 32 32 32 14 28 28 1c  |&.2 >.":2222.((.|
0000c180  32 44 32 58 44 32 44 4e  48 12 28 20 1e 3a 58 32  |2D2XD2DNH.( .:X2|

[mschwendt]

Curiously, 0xc176-0xc118=0x5e is close to the 0x62 in the sample SetBegin offset. Coincidence? Or has the begin offset been modified in trial-and-error fashion while creating the music in TFMX-editor? If it started at 0 or 4, its end would be before that spike sample!

Also not sure whether that lesser volume range at the end belongs to the same sample data. Whereas TFMX instruments can start and stop anywhere within the SMPL data range, look what instrument 2 does!

  0000 00 000000 DMAoff+Reset (stop sample & reset all)
  0001 02 00c128 SetBegin    xxxxxx   sample-startadress
  0002 03 006074 SetLen      ..xxxx   sample-length
  0003 0d 00003f Addvol+note xx/fe/xx note/CONST./volume
  0004 09 250005 SetNote     xx/xxxx  note/detune
  0005 01 000000 DMAon (start sample at selected begin)
  0006 04 000000 Wait        ..xxxx   count (VBI's)
  0007 19 000000 -------Set one shot sample-------------
  0008 04 000000 Wait        ..xxxx   count (VBI's)
  0009 07 000000 -------------STOP----------------------

[mschwendt]

SetBegin 0xc128 is before the specified sample end in script 1, the aforementioned 0x62+0x608a*2=0xc176. That's odd!

Macro script 5 replays a similar range than script 1, and only the begin is relocated by 78 samples, potentially triggering a click (missing the end by just 16 samples!):

  0000 04 000002 Wait        ..xxxx   count (VBI's)
  0001 00 000000 DMAoff+Reset (stop sample & reset all)
  0002 02 000014 SetBegin    xxxxxx   sample-startadress
  0003 03 00608a SetLen      ..xxxx   sample-length
  0004 0d 00001c Addvol+note xx/fe/xx note/CONST./volume
  0005 09 250005 SetNote     xx/xxxx  note/detune
  0006 01 000000 DMAon (start sample at selected begin)
  0007 04 000000 Wait        ..xxxx   count (VBI's)
  0008 19 000000 -------Set one shot sample-------------
  0009 04 000000 Wait        ..xxxx   count (VBI's)
  000a 07 000000 -------------STOP----------------------```

[mschwendt]

As one conclusion, it could be a timing issue that doesn't show up in WinUAE because the original music demo is emulated. The CPU consumption between calls of the music player code. If player is called late during a frame, that specific sample is started late and doesn't reach its end when the player is called again next frame.

UADE can be cycle-accurate with regard to CPU time spent within the TFMX player code, but the music demo code is missing if replaying a ripped TFMX module.

[mschwendt]

FS-UAE on Linux is affected, too.

The original music demo crashes often already while loading. Restarting the currently playing music is anything but stable. It corrupts the memory contents somehow. Even with the default Magnetic Fields 4 song, among the symptoms is graphical corruption of the large SB Absolute! logo, finally a sudden freeze and crash.

The TFMX player used by the demo is v1.5 with immediate DMACON off/on.

[mschwendt]

WinUAE crashes similarly.

Restarting the music within the demo corrupts the sample data area, writing six zero bytes into seemingly random locations. At start of first playback, the data match the ripped smpl.gitar file, though.