fft_sse.c
Go to the documentation of this file.
1 /*
2  * FFT/MDCT transform with SSE optimizations
3  * Copyright (c) 2008 Loren Merritt
4  *
5  * This file is part of Libav.
6  *
7  * Libav is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * Libav is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "libavutil/x86_cpu.h"
23 #include "libavcodec/dsputil.h"
24 #include "fft.h"
25 #include "config.h"
26 
27 DECLARE_ASM_CONST(16, unsigned int, ff_m1m1m1m1)[4] =
28  { 1U << 31, 1U << 31, 1U << 31, 1U << 31 };
29 
30 void ff_fft_dispatch_sse(FFTComplex *z, int nbits);
31 void ff_fft_dispatch_interleave_sse(FFTComplex *z, int nbits);
32 void ff_fft_dispatch_interleave_avx(FFTComplex *z, int nbits);
33 
34 #if HAVE_AVX
36 {
38 }
39 #endif
40 
42 {
43  int n = 1 << s->nbits;
44 
46 
47  if(n <= 16) {
48  x86_reg i = -8*n;
49  __asm__ volatile(
50  "1: \n"
51  "movaps (%0,%1), %%xmm0 \n"
52  "movaps %%xmm0, %%xmm1 \n"
53  "unpcklps 16(%0,%1), %%xmm0 \n"
54  "unpckhps 16(%0,%1), %%xmm1 \n"
55  "movaps %%xmm0, (%0,%1) \n"
56  "movaps %%xmm1, 16(%0,%1) \n"
57  "add $32, %0 \n"
58  "jl 1b \n"
59  :"+r"(i)
60  :"r"(z+n)
61  :"memory"
62  );
63  }
64 }
65 
67 {
68  int n = 1 << s->nbits;
69  int i;
70  for(i=0; i<n; i+=2) {
71  __asm__ volatile(
72  "movaps %2, %%xmm0 \n"
73  "movlps %%xmm0, %0 \n"
74  "movhps %%xmm0, %1 \n"
75  :"=m"(s->tmp_buf[s->revtab[i]]),
76  "=m"(s->tmp_buf[s->revtab[i+1]])
77  :"m"(z[i])
78  );
79  }
80  memcpy(z, s->tmp_buf, n*sizeof(FFTComplex));
81 }
82 
83 void ff_imdct_calc_sse(FFTContext *s, FFTSample *output, const FFTSample *input)
84 {
85  x86_reg j, k;
86  long n = s->mdct_size;
87  long n4 = n >> 2;
88 
89  s->imdct_half(s, output + n4, input);
90 
91  j = -n;
92  k = n-16;
93  __asm__ volatile(
94  "movaps "MANGLE(ff_m1m1m1m1)", %%xmm7 \n"
95  "1: \n"
96  "movaps (%2,%1), %%xmm0 \n"
97  "movaps (%3,%0), %%xmm1 \n"
98  "shufps $0x1b, %%xmm0, %%xmm0 \n"
99  "shufps $0x1b, %%xmm1, %%xmm1 \n"
100  "xorps %%xmm7, %%xmm0 \n"
101  "movaps %%xmm1, (%3,%1) \n"
102  "movaps %%xmm0, (%2,%0) \n"
103  "sub $16, %1 \n"
104  "add $16, %0 \n"
105  "jl 1b \n"
106  :"+r"(j), "+r"(k)
107  :"r"(output+n4), "r"(output+n4*3)
108  XMM_CLOBBERS_ONLY("%xmm0", "%xmm1", "%xmm7")
109  );
110 }
111