885
|
1 /*
|
|
2 * srfftp.h
|
|
3 *
|
|
4 * Copyright (C) Yuqing Deng <Yuqing_Deng@brown.edu> - April 2000
|
|
5 *
|
|
6 * 64 and 128 point split radix fft for ac3dec
|
|
7 *
|
|
8 * The algorithm is desribed in the book:
|
|
9 * "Computational Frameworks of the Fast Fourier Transform".
|
|
10 *
|
|
11 * The ideas and the the organization of code borrowed from djbfft written by
|
|
12 * D. J. Bernstein <djb@cr.py.to>. djbff can be found at
|
|
13 * http://cr.yp.to/djbfft.html.
|
|
14 *
|
|
15 * srfftp.h is free software; you can redistribute it and/or modify
|
|
16 * it under the terms of the GNU General Public License as published by
|
|
17 * the Free Software Foundation; either version 2, or (at your option)
|
|
18 * any later version.
|
|
19 *
|
|
20 * srfftp.h is distributed in the hope that it will be useful,
|
|
21 * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
22 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
23 * GNU General Public License for more details.
|
|
24 *
|
|
25 * You should have received a copy of the GNU General Public License
|
|
26 * along with GNU Make; see the file COPYING. If not, write to
|
|
27 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
|
|
28 *
|
|
29 * Modified for using AMD's 3DNow! - 3DNowEx(DSP)! SIMD operations
|
|
30 * by Nick Kurshev <nickols_k@mail.ru>
|
|
31 */
|
|
32
|
|
33 #ifndef SRFFTP_3DNOW_H__
|
|
34 #define SRFFTP_3DNOW_H__
|
|
35
|
|
36 #ifdef HAVE_3DNOWEX
|
|
37 #define TRANS_FILL_MM6_MM7_3DNOW()\
|
|
38 asm(\
|
|
39 "movl $-1, %%eax\n\t"\
|
|
40 "movd %%eax, %%mm7\n\t"\
|
|
41 "negl %%eax\n\t"\
|
|
42 "movd %%eax, %%mm6\n\t"\
|
|
43 "punpckldq %%mm6, %%mm7\n\t" /* -1.0 | 1.0 */\
|
|
44 "pi2fd %%mm7, %%mm7\n\t"\
|
|
45 "pswapd %%mm7, %%mm6\n\t"/* 1.0 | -1.0 */\
|
|
46 :::"eax","memory");
|
|
47 #else
|
|
48 #define TRANS_FILL_MM6_MM7_3DNOW()\
|
|
49 asm(\
|
|
50 "movl $-1, %%eax\n\t"\
|
|
51 "movd %%eax, %%mm7\n\t"\
|
|
52 "negl %%eax\n\t"\
|
|
53 "movd %%eax, %%mm6\n\t"\
|
|
54 "punpckldq %%mm6, %%mm7\n\t" /* -1.0 | 1.0 */\
|
|
55 "punpckldq %%mm7, %%mm6\n\t" /* 1.0 | -1.0 */\
|
|
56 "pi2fd %%mm7, %%mm7\n\t"\
|
|
57 "pi2fd %%mm6, %%mm6\n\t"\
|
|
58 :::"eax","memory");
|
|
59 #endif
|
|
60
|
|
61 #ifdef HAVE_3DNOWEX
|
|
62 #define PSWAP_MM(mm_base,mm_hlp) "pswapd "##mm_base","##mm_base" \n\t"
|
|
63 #else
|
|
64 #define PSWAP_MM(mm_base,mm_hlp)\
|
|
65 "movq "##mm_base","##mm_hlp" \n\t"\
|
|
66 "psrlq $32, "##mm_base"\n\t"\
|
|
67 "punpckldq "##mm_hlp","##mm_base"\n\t"
|
|
68 #endif
|
|
69
|
|
70 #define TRANSZERO_3DNOW(A0,A4,A8,A12) \
|
|
71 { \
|
|
72 asm volatile("femms":::"memory");\
|
|
73 TRANS_FILL_MM6_MM7_3DNOW()\
|
|
74 asm(\
|
|
75 "movq %4, %%mm0\n\t" /* mm0 = wTB[0]*/\
|
|
76 "movq %5, %%mm1\n\t" /* mm1 = wTB[k*2]*/ \
|
|
77 "movq %%mm0, %%mm5\n\t"/*u.re = wTB[0].re + wTB[k*2].re;*/\
|
|
78 "pfadd %%mm1, %%mm5\n\t"/*u.im = wTB[0].im + wTB[k*2].im; mm5 = u*/\
|
|
79 "pfmul %%mm6, %%mm0\n\t"/*mm0 = wTB[0].re | -wTB[0].im */\
|
|
80 "pfmul %%mm7, %%mm1\n\t"/*mm1 = -wTB[k*2].re | wTB[k*2].im */\
|
|
81 "pfadd %%mm1, %%mm0\n\t"/*v.im = wTB[0].re - wTB[k*2].re;*/\
|
|
82 "movq %%mm0, %%mm4\n\t"/*v.re =-wTB[0].im + wTB[k*2].im;*/\
|
|
83 PSWAP_MM("%%mm4","%%mm2")/* mm4 = v*/\
|
|
84 "movq %6, %%mm0\n\t" /* a1 = A0;*/\
|
|
85 "movq %%mm0, %%mm1\n\t"\
|
|
86 "pfadd %%mm5, %%mm0\n\t" /*A0 = a1 + u;*/\
|
|
87 "pfsub %%mm5, %%mm1\n\t" /*A1 = a1 - u;*/\
|
|
88 "movq %%mm0, %0\n\t"\
|
|
89 "movq %%mm1, %1\n\t"\
|
|
90 "movq %7, %%mm2\n\t" /* a1 = A4;*/\
|
|
91 "movq %%mm2, %%mm3\n\t"\
|
|
92 "pfadd %%mm4, %%mm2\n\t" /*A12 = a1 + v;*/\
|
|
93 "pfsub %%mm4, %%mm3\n\t" /*A4 = a1 - v;*/\
|
|
94 "movq %%mm2, %3\n\t"\
|
|
95 "movq %%mm3, %2"\
|
|
96 :"=m"(A0), "=m"(A8), "=m"(A4), "=m"(A12)\
|
|
97 :"m"(wTB[0]), "m"(wTB[k*2]), "0"(A0), "2"(A4)\
|
|
98 :"memory");\
|
|
99 asm volatile("femms":::"memory");\
|
|
100 }
|
|
101
|
|
102 #endif
|