mirrors/gstreamer
mirrors/gstreamer
1
1
Fork 0
mirror of https://gitlab.freedesktop.org/gstreamer/gstreamer.git synced 2024-11-27 12:11:13 +00:00
Code Issues 1 Projects Releases Wiki Activity

audio-resampler: unroll some more loops

Browse source 
Unroll some loops.
This commit is contained in:
Wim Taymans 2016-03-02 11:40:15 +01:00
parent 90a41b81dc
commit 33855f0fe1
2 changed files with 14 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

19
gst-libs/gst/audio/audio-resampler-x86.h
View file

@ -171,9 +171,12 @@ inner_product_gint16_full_1_sse2 (gint16 * o, const gint16 * a,
sum = _mm_setzero_si128 ();
for (i = 0; i < len; i += 8) {
for (i = 0; i < len; i += 16) {
t = _mm_loadu_si128 ((__m128i *) (a + i));
sum = _mm_add_epi32 (sum, _mm_madd_epi16 (t, _mm_load_si128 ((__m128i *) (b + i))));
sum = _mm_add_epi32 (sum, _mm_madd_epi16 (t, _mm_load_si128 ((__m128i *) (b + i + 0))));
t = _mm_loadu_si128 ((__m128i *) (a + i + 8));
sum = _mm_add_epi32 (sum, _mm_madd_epi16 (t, _mm_load_si128 ((__m128i *) (b + i + 8))));
}
sum = _mm_add_epi32 (sum, _mm_shuffle_epi32 (sum, _MM_SHUFFLE (2, 3, 2, 3)));
sum = _mm_add_epi32 (sum, _mm_shuffle_epi32 (sum, _MM_SHUFFLE (1, 1, 1, 1)));
@ -197,10 +200,14 @@ inner_product_gint16_linear_1_sse2 (gint16 * o, const gint16 * a,
sum[0] = sum[1] = _mm_setzero_si128 ();
f = _mm_unpacklo_epi16 (f, sum[0]);
for (; i < len; i += 8) {
t = _mm_loadu_si128 ((__m128i *) (a + i));
sum[0] = _mm_add_epi32 (sum[0], _mm_madd_epi16 (t, _mm_load_si128 ((__m128i *) (c[0] + i))));
sum[1] = _mm_add_epi32 (sum[1], _mm_madd_epi16 (t, _mm_load_si128 ((__m128i *) (c[1] + i))));
for (; i < len; i += 16) {
t = _mm_loadu_si128 ((__m128i *) (a + i + 0));
sum[0] = _mm_add_epi32 (sum[0], _mm_madd_epi16 (t, _mm_load_si128 ((__m128i *) (c[0] + i + 0))));
sum[1] = _mm_add_epi32 (sum[1], _mm_madd_epi16 (t, _mm_load_si128 ((__m128i *) (c[1] + i + 0))));
t = _mm_loadu_si128 ((__m128i *) (a + i + 8));
sum[0] = _mm_add_epi32 (sum[0], _mm_madd_epi16 (t, _mm_load_si128 ((__m128i *) (c[0] + i + 8))));
sum[1] = _mm_add_epi32 (sum[1], _mm_madd_epi16 (t, _mm_load_si128 ((__m128i *) (c[1] + i + 8))));
}
sum[0] = _mm_srai_epi32 (sum[0], PRECISION_S16);
sum[1] = _mm_srai_epi32 (sum[1], PRECISION_S16);

2
gst-libs/gst/audio/audio-resampler.c
View file

@ -60,7 +60,7 @@ typedef void (*DeinterleaveFunc) (GstAudioResampler * resampler,
#define MEM_ALIGN(m,a) ((gint8 *)((guintptr)((gint8 *)(m) + ((a)-1)) & ~((a)-1)))
#define ALIGN 16
#define TAPS_OVERREAD 8
#define TAPS_OVERREAD 16
struct _GstAudioResampler
{