• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  *  Copyright (c) 2013 The WebRTC project authors. All Rights Reserved.
3  *
4  *  Use of this source code is governed by a BSD-style license
5  *  that can be found in the LICENSE file in the root of the source
6  *  tree. An additional intellectual property rights grant can be found
7  *  in the file PATENTS.  All contributing project authors may
8  *  be found in the AUTHORS file in the root of the source tree.
9  *
10  */
11 
12 #include "dl/api/omxtypes.h"
13 #include "dl/sp/src/x86/x86SP_SSE_Math.h"
14 
x86SP_FFT_CToC_FC32_Inv_Radix4_ls_sse(const OMX_F32 * in,OMX_F32 * out,const OMX_F32 * twiddle,OMX_INT n)15 void x86SP_FFT_CToC_FC32_Inv_Radix4_ls_sse(
16     const OMX_F32 *in,
17     OMX_F32 *out,
18     const OMX_F32 *twiddle,
19     OMX_INT n) {
20   OMX_INT n_by_2 = n >> 1;
21   OMX_INT n_by_4 = n >> 2;
22   OMX_INT n_mul_2 = n << 1;
23   OMX_INT i;
24 
25   OMX_F32 *out0 = out;
26 
27   for (i = 0; i < n_by_2; i += 8) {
28     const OMX_F32 *tw1 = twiddle + i;
29     const OMX_F32 *tw2 = tw1 + i;
30     const OMX_F32 *tw3 = tw2 + i;
31     const OMX_F32 *in0 = in + (i << 1);
32     const OMX_F32 *in1 = in0 + 4;
33     const OMX_F32 *in2 = in1 + 4;
34     const OMX_F32 *in3 = in2 + 4;
35     OMX_F32 *out1 = out0 + n_by_4;
36     OMX_F32 *out2 = out1 + n_by_4;
37     OMX_F32 *out3 = out2 + n_by_4;
38 
39     VC v_tw1;
40     VC v_tw2;
41     VC v_tw3;
42     VC v_t0;
43     VC v_t1;
44     VC v_t2;
45     VC v_t3;
46     VC v_t4;
47     VC v_t5;
48     VC v_t6;
49     VC v_t7;
50 
51     v_tw1.real = _mm_set_ps(tw1[6], tw1[4], tw1[2], tw1[0]);
52     v_tw1.imag = _mm_set_ps(
53         tw1[6 + n_mul_2],
54         tw1[4 + n_mul_2],
55         tw1[2 + n_mul_2],
56         tw1[n_mul_2]);
57     v_tw2.real = _mm_set_ps(tw2[12], tw2[8], tw2[4], tw2[0]);
58     v_tw2.imag = _mm_set_ps(
59         tw2[12 + n_mul_2],
60         tw2[8 + n_mul_2],
61         tw2[4 + n_mul_2],
62         tw2[n_mul_2]);
63     v_tw3.real = _mm_set_ps(tw3[18], tw3[12], tw3[6], tw3[0]);
64     v_tw3.imag = _mm_set_ps(
65         tw3[18 + n_mul_2],
66         tw3[12 + n_mul_2],
67         tw3[6 + n_mul_2],
68         tw3[n_mul_2]);
69 
70     VC_LOAD_MATRIX_TRANSPOSE(&v_t0, &v_t1, &v_t2, &v_t3, in0, in1, in2, in3, n);
71 
72     RADIX4_INV_BUTTERFLY(&v_t4, &v_t5, &v_t6, &v_t7,
73                          &v_tw1, &v_tw2, &v_tw3,
74                          &v_t0, &v_t1, &v_t2, &v_t3);
75 
76     RADIX4_INV_BUTTERFLY_STORE(out0, out1, out2, out3,
77                                &v_t4, &v_t5, &v_t6, &v_t7, n);
78 
79     out0 += 4;
80   }
81 }
82