summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorAndree Buschmann <AndreeBuschmann@t-online.de>2010-07-28 18:15:53 +0000
committerAndree Buschmann <AndreeBuschmann@t-online.de>2010-07-28 18:15:53 +0000
commit8cc8bb9bde265959ad69b3fc5cc84201461af1fb (patch)
treedb676e1e8f18d50260a5ade242f3e4e0458ab707
parentf0aac62dd5bfd41c32ab375f8f03b8a1d648b919 (diff)
downloadrockbox-8cc8bb9bde265959ad69b3fc5cc84201461af1fb.tar.gz
rockbox-8cc8bb9bde265959ad69b3fc5cc84201461af1fb.tar.bz2
rockbox-8cc8bb9bde265959ad69b3fc5cc84201461af1fb.zip
Submit next part of FS#11498. Unroll loop for minor speedup of libwmapro on ARM (1%).
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@27595 a1c6a512-1295-4272-9138-f99709370657
-rw-r--r--apps/codecs/libwmapro/wmapro_math.h24
1 files changed, 15 insertions, 9 deletions
diff --git a/apps/codecs/libwmapro/wmapro_math.h b/apps/codecs/libwmapro/wmapro_math.h
index ba6421c9f5..6f8d6dbe0f 100644
--- a/apps/codecs/libwmapro/wmapro_math.h
+++ b/apps/codecs/libwmapro/wmapro_math.h
@@ -218,13 +218,13 @@ static inline void vector_fixmul_window(int32_t *dst, const int32_t *src0,
dst += len;
win += len;
src0+= len;
- for(i=-len, j=len-1; i<0; i++, j--) {
- int32_t s0 = src0[i];
- int32_t s1 = src1[j];
- int32_t wi = -win[i];
- int32_t wj = -win[j];
- dst[i] = fixmul31(s0, wj) - fixmul31(s1, wi);
- dst[j] = fixmul31(s0, wi) + fixmul31(s1, wj);
+ for(i=-len, j=len-1; i<0; i++, j--) {
+ int32_t s0 = src0[i]; /* s0 = src0[ 0 ... len-1] */
+ int32_t s1 = src1[j]; /* s1 = src1[2*len-1 ... len] */
+ int32_t wi = -win[i]; /* wi = -win[ 0 ... len-1] */
+ int32_t wj = -win[j]; /* wj = -win[2*len-1 ... len] */
+ dst[i] = fixmul31(s0, wj) - fixmul31(s1, wi); /* dst[ 0 ... len-1] */
+ dst[j] = fixmul31(s0, wi) + fixmul31(s1, wj); /* dst[2*len-1 ... len] */
}
}
#endif
@@ -232,9 +232,15 @@ static inline void vector_fixmul_window(int32_t *dst, const int32_t *src0,
static inline void vector_fixmul_scalar(int32_t *dst, const int32_t *src,
int32_t mul, int len)
{
+ /* len is _always_ a multiple of 4, because len is the difference of sfb's
+ * which themselves are always a multiple of 4. */
int i;
- for(i=0; i<len; i++)
- dst[i] = fixmul24(src[i], mul);
+ for (i=0; i<len; i+=4) {
+ dst[i ] = fixmul24(src[i ], mul);
+ dst[i+1] = fixmul24(src[i+1], mul);
+ dst[i+2] = fixmul24(src[i+2], mul);
+ dst[i+3] = fixmul24(src[i+3], mul);
+ }
}
static inline int av_clip(int a, int amin, int amax)