diff options
author | Andree Buschmann <AndreeBuschmann@t-online.de> | 2012-10-07 12:34:54 +0200 |
---|---|---|
committer | Nils Wallménius <nils@rockbox.org> | 2012-10-07 12:34:54 +0200 |
commit | 0ae8d16643fac1ba935a4e5303cc423655fb3ff7 (patch) | |
tree | 70ed2e31d4cebb3a8b9fdc0656c99de8ba2e8665 /lib/rbcodec/codecs/libopus/celt | |
parent | c7840e745e9359f8210e5112bef388c4e3157765 (diff) | |
download | rockbox-0ae8d16643fac1ba935a4e5303cc423655fb3ff7.tar.gz rockbox-0ae8d16643fac1ba935a4e5303cc423655fb3ff7.zip |
opus: replace copy loops with calls to memcpy
speeds up decoding of a 64kbps test file by 0.5Hz on h300 (cf)
0.9MHz on c200 (pp) and 0.2MHz on fuzev1 (amsv1)
Change-Id: Ib537c2393fa6dca0b61e4e9f80eef5e688c2c2bd
Signed-off-by: Nils Wallménius <nils@rockbox.org>
Diffstat (limited to 'lib/rbcodec/codecs/libopus/celt')
-rw-r--r-- | lib/rbcodec/codecs/libopus/celt/mdct.c | 22 |
1 files changed, 10 insertions, 12 deletions
diff --git a/lib/rbcodec/codecs/libopus/celt/mdct.c b/lib/rbcodec/codecs/libopus/celt/mdct.c index 8fc1b54db7..0f9af9eac4 100644 --- a/lib/rbcodec/codecs/libopus/celt/mdct.c +++ b/lib/rbcodec/codecs/libopus/celt/mdct.c @@ -305,12 +305,11 @@ void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scala kiss_fft_scalar * OPUS_RESTRICT yp1 = out+N4-overlap/2; const opus_val16 * OPUS_RESTRICT wp1 = window; const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; - for(i = 0; i< N4-overlap/2; i++) - { - *xp1 = *fp1; - xp1--; - fp1--; - } + + i = N4-overlap/2; + xp1 -= N4-overlap/2; + fp1 -= N4-overlap/2; + OPUS_COPY(xp1+1, fp1+1, N4-overlap/2); for(; i < N4; i++) { kiss_fft_scalar x1; @@ -327,12 +326,11 @@ void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scala kiss_fft_scalar * OPUS_RESTRICT yp2 = out+N-1-(N4-overlap/2); const opus_val16 * OPUS_RESTRICT wp1 = window; const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; - for(i = 0; i< N4-overlap/2; i++) - { - *xp2 = *fp2; - xp2++; - fp2++; - } + + i = N4-overlap/2; + OPUS_COPY(xp2, fp2, N4-overlap/2); + xp2 += N4-overlap/2; + fp2 += N4-overlap/2; for(; i < N4; i++) { kiss_fft_scalar x2; |