diff options
| author | Jean-Marc Valin <jmvalin@jmvalin.ca> | 2016-09-14 18:38:15 +0300 |
|---|---|---|
| committer | Jean-Marc Valin <jmvalin@jmvalin.ca> | 2016-09-14 18:41:10 +0300 |
| commit | e5ad26169502c1f0b067916e782c2b3fdea6fba9 (patch) | |
| tree | bcbbbb081db91cd6b45f8504d9da489dd83d7504 | |
| parent | 70e3c348228a28ea7f62def5ecbce71b975a517b (diff) | |
Prevents ridiculously large gains from causing inf/NaNs in float decoder
| -rw-r--r-- | celt/bands.c | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/celt/bands.c b/celt/bands.c index bbe8a4c3..05205554 100644 --- a/celt/bands.c +++ b/celt/bands.c @@ -226,7 +226,7 @@ void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, band_end = M*eBands[i+1]; lg = SATURATE16(ADD32(bandLogE[i], SHL32((opus_val32)eMeans[i],6))); #ifndef FIXED_POINT - g = celt_exp2(lg); + g = celt_exp2(MIN32(32.f, lg)); #else /* Handle the integer part of the log energy */ shift = 16-(lg>>DB_SHIFT); |