diff options
author | Zuxy Meng <zuxy.meng@gmail.com> | 2006-03-05 20:25:18 +0000 |
---|---|---|
committer | Guillaume Poirier <gpoirier@mplayerhq.hu> | 2006-03-05 20:25:18 +0000 |
commit | 2ffb22d2ad014e7fa8ee4c24263eb03abb9e86e5 (patch) | |
tree | 0108b52940ff24d5e18fbd4b56f5f1fc52cb2a61 | |
parent | d84f7c61eede0e5327c24e606097c271abc7fea3 (diff) | |
download | ffmpeg-streaming-2ffb22d2ad014e7fa8ee4c24263eb03abb9e86e5.zip ffmpeg-streaming-2ffb22d2ad014e7fa8ee4c24263eb03abb9e86e5.tar.gz |
use xorps instead of mulps to toggle the sign of a float, as suggested by Software Optimization Guide for AMD64 Processors.
Patch by Zuxy Meng < zuxy POIS meng AH gmail POIS com > OKed by Michael
Original thread:
Date: Mar 5, 2006 8:15 PM
Subject: [Ffmpeg-devel] [PATCH] Little optimization to fft_sse.c
Originally committed as revision 5112 to svn://svn.ffmpeg.org/ffmpeg/trunk
-rw-r--r-- | libavcodec/i386/fft_sse.c | 20 |
1 files changed, 10 insertions, 10 deletions
diff --git a/libavcodec/i386/fft_sse.c b/libavcodec/i386/fft_sse.c index 54851fb..6318482 100644 --- a/libavcodec/i386/fft_sse.c +++ b/libavcodec/i386/fft_sse.c @@ -23,14 +23,14 @@ #include <xmmintrin.h> -static const float p1p1p1m1[4] __attribute__((aligned(16))) = - { 1.0, 1.0, 1.0, -1.0 }; +static const int p1p1p1m1[4] __attribute__((aligned(16))) = + { 0, 0, 0, 1 << 31 }; -static const float p1p1m1p1[4] __attribute__((aligned(16))) = - { 1.0, 1.0, -1.0, 1.0 }; +static const int p1p1m1p1[4] __attribute__((aligned(16))) = + { 0, 0, 1 << 31, 0 }; -static const float p1p1m1m1[4] __attribute__((aligned(16))) = - { 1.0, 1.0, -1.0, -1.0 }; +static const int p1p1m1m1[4] __attribute__((aligned(16))) = + { 0, 0, 1 << 31, 1 << 31 }; #if 0 static void print_v4sf(const char *str, __m128 a) @@ -58,7 +58,6 @@ void ff_fft_calc_sse(FFTContext *s, FFTComplex *z) r = (__m128 *)&z[0]; c1 = *(__m128 *)p1p1m1m1; - c2 = *(__m128 *)p1p1p1m1; if (s->inverse) c2 = *(__m128 *)p1p1m1p1; else @@ -68,19 +67,20 @@ void ff_fft_calc_sse(FFTContext *s, FFTComplex *z) do { a = r[0]; b = _mm_shuffle_ps(a, a, _MM_SHUFFLE(1, 0, 3, 2)); - a = _mm_mul_ps(a, c1); + a = _mm_xor_ps(a, c1); /* do the pass 0 butterfly */ a = _mm_add_ps(a, b); a1 = r[1]; b = _mm_shuffle_ps(a1, a1, _MM_SHUFFLE(1, 0, 3, 2)); - a1 = _mm_mul_ps(a1, c1); + a1 = _mm_xor_ps(a1, c1); /* do the pass 0 butterfly */ b = _mm_add_ps(a1, b); /* multiply third by -i */ + /* by toggling the sign bit */ b = _mm_shuffle_ps(b, b, _MM_SHUFFLE(2, 3, 1, 0)); - b = _mm_mul_ps(b, c2); + b = _mm_xor_ps(b, c2); /* do the pass 1 butterfly */ r[0] = _mm_add_ps(a, b); |