2x speedup of surround decoding on Cortex-A9.
Signed-off-by: Mans Rullgard <[email protected]>
---
libavcodec/vorbisdec.c | 17 ++++++++++++-----
1 file changed, 12 insertions(+), 5 deletions(-)
diff --git a/libavcodec/vorbisdec.c b/libavcodec/vorbisdec.c
index 08198d7..deaaaa2 100644
--- a/libavcodec/vorbisdec.c
+++ b/libavcodec/vorbisdec.c
@@ -1409,17 +1409,24 @@ static av_always_inline int
vorbis_residue_decode_internal(vorbis_context *vc,
}
} else if (vr_type == 2) {
- voffs = voffset;
+ unsigned voffs_div = FASTDIV(voffset, ch);
+ unsigned voffs_mod = voffset - voffs_div * ch;
for (k = 0; k < step; ++k) {
coffs = get_vlc2(gb, codebook.vlc.table,
codebook.nb_bits, 3) * dim;
- for (l = 0; l < dim; ++l, ++voffs) {
- vec[voffs / ch + (voffs % ch) * vlen]
+= codebook.codevectors[coffs + l]; // FPMATH FIXME use if and counter instead
of / and %
+ for (l = 0; l < dim; ++l) {
+ vec[voffs_div + voffs_mod * vlen] +=
+ codebook.codevectors[coffs + l];
av_dlog(NULL, " pass %d offs: %d curr:
%f change: %f cv offs.: %d+%d \n",
- pass, voffset / ch + (voffs %
ch) * vlen,
- vec[voffset / ch + (voffs %
ch) * vlen],
+ pass, voffs_div + voffs_mod *
vlen,
+ vec[voffs_div + voffs_mod *
vlen],
codebook.codevectors[coffs +
l], coffs, l);
+
+ if (++voffs_mod == ch) {
+ voffs_div++;
+ voffs_mod = 0;
+ }
}
}
}
--
1.7.10.2
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
