Mercurial > mplayer.hg
comparison libao2/fir.h @ 3631:5f5189ac6a41
Added plugin for fractional resampling (alpha code)
author | anders |
---|---|
date | Thu, 20 Dec 2001 15:30:22 +0000 |
parents | |
children | d6f8feeac656 |
comparison
equal
deleted
inserted
replaced
3630:f24527fc1b79 | 3631:5f5189ac6a41 |
---|---|
1 /*============================================================================= | |
2 // | |
3 // This file is part of mplayer. | |
4 // | |
5 // mplayer is free software; you can redistribute it and/or modify | |
6 // it under the terms of the GNU General Public License as published by | |
7 // the Free Software Foundation; either version 2 of the License, or | |
8 // (at your option) any later version. | |
9 // | |
10 // mplayer is distributed in the hope that it will be useful, | |
11 // but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 // MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 // GNU General Public License for more details. | |
14 // | |
15 // You should have received a copy of the GNU General Public License | |
16 // along with mplayer; if not, write to the Free Software | |
17 // Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
18 // | |
19 // Copyright 2001 Anders Johansson ajh@atri.curtin.edu.au | |
20 // | |
21 //============================================================================= | |
22 */ | |
23 | |
24 #ifndef __FIR_H__ | |
25 #define __FIR_H__ | |
26 | |
27 /* 4, 8 and 16 tap FIR filters implemented using SSE instructions | |
28 int16_t* x Input data | |
29 int16_t* y Output value | |
30 int16_t* w Filter weights | |
31 | |
32 C function | |
33 for(int i = 0 ; i < L ; i++) | |
34 *y += w[i]*x[i]; | |
35 */ | |
36 | |
37 #ifdef HAVE_SSE | |
38 | |
39 // This block should be MMX only compatible, but it isn't... | |
40 #ifdef L4 | |
41 #define LOAD_QUE(x) \ | |
42 __asm __volatile("movq %0, %%mm2\n\t" \ | |
43 : \ | |
44 :"m"((x)[0]) \ | |
45 :"memory"); | |
46 #define SAVE_QUE(x) \ | |
47 __asm __volatile("movq %%mm2, %0\n\t" \ | |
48 :"=m"(x[0]) \ | |
49 : \ | |
50 :"memory"); | |
51 #define UPDATE_QUE(in) \ | |
52 __asm __volatile("psllq $16, %%mm2\n\t" \ | |
53 "pinsrw $0, %0,%%mm2\n\t" \ | |
54 : \ | |
55 :"m" ((in)[0]) \ | |
56 :"memory"); | |
57 #define FIR(x,w,y) \ | |
58 __asm __volatile("movq %%mm2, %%mm0\n\t" \ | |
59 "pmaddwd %1, %%mm0\n\t" \ | |
60 "movq %%mm0, %%mm1\n\t" \ | |
61 "psrlq $32, %%mm1\n\t" \ | |
62 "paddd %%mm0, %%mm1\n\t" \ | |
63 "movd %%mm1, %%esi\n\t" \ | |
64 "shrl $16, %%esi\n\t" \ | |
65 "movw %%si, %0\n\t" \ | |
66 : "=m" ((y)[0]) \ | |
67 : "m" ((w)[0]) \ | |
68 : "memory", "%esi"); | |
69 #endif /* L4 */ | |
70 | |
71 // It is possible to make the 8 bit filter a lot faster by using the | |
72 // 128 bit registers, feel free to optimize. | |
73 #ifdef L8 | |
74 #define LOAD_QUE(x) \ | |
75 __asm __volatile("movq %0, %%mm5\n\t" \ | |
76 "movq %1, %%mm4\n\t" \ | |
77 : \ | |
78 :"m"((x)[0]), \ | |
79 "m"((x)[4]) \ | |
80 :"memory"); | |
81 #define SAVE_QUE(x) \ | |
82 __asm __volatile("movq %%mm5, %0\n\t" \ | |
83 "movq %%mm4, %1\n\t" \ | |
84 :"=m"((x)[0]), \ | |
85 "=m"((x)[4]) \ | |
86 : \ | |
87 :"memory"); | |
88 | |
89 // Below operation could replace line 2 to 5 in macro below but can | |
90 // not cause of compiler bug ??? | |
91 // "pextrw $3, %%mm5,%%eax\n\t" | |
92 #define UPDATE_QUE(in) \ | |
93 __asm __volatile("psllq $16, %%mm4\n\t" \ | |
94 "movq %%mm5, %%mm0\n\t" \ | |
95 "psrlq $48, %%mm0\n\t" \ | |
96 "movd %%mm0, %%eax\n\t" \ | |
97 "pinsrw $0, %%eax,%%mm4\n\t" \ | |
98 "psllq $16, %%mm5\n\t" \ | |
99 "pinsrw $0, %0,%%mm5\n\t" \ | |
100 : \ | |
101 :"m" ((in)[0]) \ | |
102 :"memory", "%eax"); | |
103 #define FIR(x,w,y) \ | |
104 __asm __volatile("movq %%mm5, %%mm0\n\t" \ | |
105 "pmaddwd %1, %%mm0\n\t" \ | |
106 "movq %%mm4, %%mm1\n\t" \ | |
107 "pmaddwd %2, %%mm1\n\t" \ | |
108 "paddd %%mm1, %%mm0\n\t" \ | |
109 "movq %%mm0, %%mm1\n\t" \ | |
110 "psrlq $32, %%mm1\n\t" \ | |
111 "paddd %%mm0, %%mm1\n\t" \ | |
112 "movd %%mm1, %%esi\n\t" \ | |
113 "shrl $16, %%esi\n\t" \ | |
114 "movw %%si, %0\n\t" \ | |
115 : "=m" ((y)[0]) \ | |
116 : "m" ((w)[0]), \ | |
117 "m" ((w)[4]) \ | |
118 : "memory", "%esi"); | |
119 #endif /* L8 */ | |
120 | |
121 #else /* HAVE_SSE */ | |
122 | |
123 #define LOAD_QUE(x) | |
124 #define SAVE_QUE(x) | |
125 #define UPDATE_QUE(inm) \ | |
126 xi=(--xi)&(L-1); \ | |
127 x[xi]=x[xi+L]=*inm; | |
128 | |
129 #ifdef L4 | |
130 #define FIR(x,w,y) \ | |
131 y[0]=(w[0]*x[0]+w[1]*x[1]+w[2]*x[2]+w[3]*x[3]) >> 16; | |
132 #else | |
133 #define FIR(x,w,y){ \ | |
134 int16_t a = (w[0]*x[0]+w[1]*x[1]+w[2]*x[2]+w[3]*x[3]) >> 16; \ | |
135 int16_t b = (w[4]*x[4]+w[5]*x[5]+w[6]*x[6]+w[7]*x[7]) >> 16; \ | |
136 y[0] = a+b; \ | |
137 } | |
138 #endif /* L4 */ | |
139 | |
140 #endif /* HAVE_SSE */ | |
141 | |
142 #endif /* __FIR_H__ */ | |
143 | |
144 |