vgmlpfnd
VGM loop finder is one of the VGM Tools, specifically one that helps with Trimming VGMs.
Trimming tends to be tricky; one reason is, songs often replay parts of the track as part of the track, so there's actually multiple loops in there. vgmlpfnd tries to find them all so you can choose which one is correct and then use vgm_trim to trim at the right place.
If you're lucky, this program will produce many !s for you, and all of them will be correct. Its finding abilities aren't perfect, nor does it always work with all tracks even within the same game. Moreover, it can be difficult to know when vgmlpfnd actually failed unless you know a song very well: when using the wrong loop point, the transition is very smooth most of the time.
That said, even if you don't get a working loop with only vgmlpfnd, it can sometimes help when combined with other trim-finding methods. See Mao's vgmlpfnd Method for an example.
Usage
vgmlpfnd insert_a.vgm StepSize MinimumNumberOfMatchingCommands StartPosition
You can also just drop a VGM on the program and be prompted for these inputs. Details follow:
Step Size
How many commands vgmlpfnd "steps" over while searching.
The default value is 1, a good value to use in most cases: that's the most accurate scanning, just reading all VGM commands. If you have a large VGM and the scanning is very slow, you can increase it to speed up the process at the cost of lower accuracy.
Using larger values may also be a good choice to test whether or not the loop finder works at all on the respective song.
Minimum Number of Matching Commands
vgmlpfnd will only detect a loop when a list of this many VGM commands are detected with the exact same values in the same order. Larger numbers like 4096 can reduce "noise" results, but will miss loops that complete with fewer commands. Smaller numbers like 300 might get you more false-positives but, if the track's loop is mere seconds long, it can be required to use them.
The minimum minimum number is 129; below this, too many commands for useful results will be matched. The default value is 1024.
Start Position
How many samples into the VGM to start searching.
The default, 0, is good for most cases when you're sure the vgmfile contains 1 track. (MAME recordings, for instance, always contain more than one track if they aren't split.) However, if you know the track has a very long intro, you can use a sample closer to the loop's start to save searching time.
Results, Explained
If you're lucky, you'll have a single result that's probably the right one. Here's an example:
Source Block Block Copy Copy Information Start Time Start Time Cmds Time 477258 00:10.82 2374215 00:53.84 26535 00:34.59 Done.
The values you need are Start from the Source Block and Start from the Block Copy. The first value means "where the loop starts" and the second one "where the copy starts", so actually where the original loop ended. You can then use vgm_trim with the inputs 0, 477258 and 2374215 in this case.
vgmlpfnd may report any (or none) of three notations in the results, one per line:
!is (usually) the perfect result and indicates a fully looping section that lasts until the end of the file. (There are more details on cases where this result is incorrect below.)fmeans a proper, full loop has been found, but it ends before the end of the VGM log (either due to another section following, or the notes being turned off before the VGM stops). This is still a good result. (Of course, it too should be double-checked.)emeans that a partial loop seems to be found which ends at the end of the file. If you don't have a proper loop and you get this result, you might consider recording a new vgm file which goes on a little longer (possibly even three or four+ loops).
When you don't get an f or a !, don't worry about it; just check the actual loop time of the track and compare that with the results. Or use values that are apart from each other by a lot.
Copy Information is how long the matched section is, which is often shorter than the total loop. Higher numbers in the Copy Information column may indicate the loop is more likely to really be a loop. Consider the following examples from Puyo Puyo (Genesis)'s Brave of Puyopuyo:
468902 00:10.63 2171155 00:49.23 4970 00:02.77 468902 00:10.63 3873400 01:27.83 4676 00:02.62 468902 00:10.63 5575647 02:06.43 4406 00:02.47
Configuration from Dr. Robotnik's Mean Bean Machine (and yes, the middle line is the correct one!):
1743634 00:39.54 4948219 01:52.20 1617 00:02.16 2186277 00:49.58 5390863 02:02.24 11042 00:16.35 3254474 01:13.80 6459059 02:26.46 5704 00:08.28
Mickey Mania's Whirlwind (recorded at 50hz):
328845 00:07.46 ! 1735571 00:39.36 21413 01:47.06 329198 00:07.46 610311 00:13.84 1230 00:06.37 329198 00:07.46 2017038 00:45.74 1230 00:06.37 329198 00:07.46 3423765 01:17.64 1230 00:06.37 329198 00:07.46 4830492 01:49.53 1230 00:06.37
Joe & Mac (Mega Drive)'s first level theme (50hz)
207634 00:04.71 ! 2233221 00:50.64 41549 00:59.43 207768 00:04.71 f 376476 00:08.54 2568 00:03.83 207768 00:04.71 2402162 00:54.47 2568 00:03.83 207768 00:04.71 4427850 01:40.40 2568 00:03.83
Chuck Rock II: Son of Chuck (Mega Drive)'s The Apple Tree (50hz):
78146 00:01.77 415760 00:09.43 1230 00:05.26 78146 00:01.77 ! 753375 00:17.08 20110 01:22.30 78146 00:01.77 1090989 00:24.74 1230 00:05.26 78146 00:01.77 1766218 00:40.05 1230 00:05.26 78146 00:01.77 2441447 00:55.36 1230 00:05.26 78146 00:01.77 3116676 01:10.67 1230 00:05.26 78146 00:01.77 3791905 01:25.98 1230 00:05.26
and its The Fruit Mountain (ditto):
403461 00:09.15 ! 2429261 00:55.09 64764 02:03.27 859756 00:19.50 1197372 00:27.15 1424 00:02.52 859756 00:19.50 3223059 01:13.09 1424 00:02.52 859756 00:19.50 5248747 01:59.02 1424 00:02.52 859756 00:19.50 7274434 02:44.95 1424 00:02.52
Meanwhile, an example of a wrong ! line from Mean Bean's Continue:
339390 00:07.70 ! 619010 00:14.04 12848 00:20.27
There are two reasons this can happen:
- When a song plays in the pattern "A A A B A A A B", the
!may show only the first "A" as a perfect loop. - The song plays "A B A B", but the only difference between "A" and "B" is in data that the loop finder ignores. This is the case here, as the loop finder ignores the YM2612 DAC due to performance reasons.
That said, now and then, values of a ! line may simply cause playback hiccups; the loop will be correct, except that for some reason there are clicks or dropped notes on loop. You can still use these to help you find values nearby that avoid this issue; sometimes only a little shifting is needed.
"Does vgmlpfnd work with...?"
Because vgmlpfnd relies on matching commands, small variances during loops can interfere with automatic loop detection. This means that how well it works with a VGM owes mainly to the used sound driver and sometimes to the track itself - not the sound chip(s). (Consider EarthWorm Jim 2: this tool could find the loop for The Moo Tango (with a long-enough recording) despite the GEMS driver being finder-unfriendly, but not so much other tracks.)
As a result, vgmlpfnd's success or failure throughout a soundtrack can't tell us any useful information, or even trivia, about chips, games, composers, or anything else. While it may feel exciting to discover a seeming pattern (like Krisalis games working far better at 50hz recordings than 60), it's unfortunately impossible to glean further patterns, tricks, or facts from it.
The good news, though, is with no meaning to be found you needn't worry about the "Rorshach inkblot" at all.