Use a shift instead of div to have bit exact output for MS Adpcm codec.

[dtzxporter]

This output now matches FFMPEG/ADPCMEncode.exe's output bit for bit.

[dedobbin]

Hey,

I played around a bit and wanted to run some files through symphonia-check, i ran into a right shift error though.

thread 'main' panicked at symphonia-codec-adpcm/src/codec_ms.rs:97:25:
attempt to shift right with overflow

tone4_adpmc_ms.wav.zip
The file i used. It's a tone generated in audacity and converted to adpcm_ms

[dtzxporter]

Should be fixed now.

[dedobbin]

Ah perfect! running the attached file through symphonia-check gives the same output as with master now.

[pdeljanov]

The reason why it overflowed is because evaluating 8 + signed_nibble * self.delta took precedence over the shift operation. So the amount shifted was not 8, rather, it was whatever that expression evaluated to, causing an overflow when it was >= 32.

The code below will work without the overhead of wrapping_shr.

let predictor = (((self.sample1 * self.coeff1) + (self.sample2 * self.coeff2)) >> 8) + signed_nibble * self.delta;