| blob | 8bc6bd3c5cd5686910d3c0ce2abb85b8431192c7 |
1 /***************************************************************************\r
2 * __________ __ ___.\r
3 * Open \______ \ ____ ____ | | _\_ |__ _______ ___\r
4 * Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ /\r
5 * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < <\r
6 * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \\r
7 * \/ \/ \/ \/ \/\r
8 * $Id$\r
9 *\r
10 * Copyright (C) 2008 by Andree Buschmann\r
11 *\r
12 * All files in this archive are subject to the GNU General Public License.\r
13 * See the file COPYING in the source tree root for full license agreement.\r
14 *\r
15 * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY\r
16 * KIND, either express or implied.\r
17 *\r
18 ****************************************************************************/\r
19 \r
20 #include "mpc_config.h"\r
21 \r
22 .section .text, "ax", %progbits\r
23 \r
24 /****************************************************************************\r
25 * void mpc_decoder_windowing_D(...)\r
26 * \r
27 * 2nd step within synthesis filter. Does the dewindowing.\r
28 * 32=32x32 multiplies (OPTIMIZE_FOR_SPEED)\r
29 * Uses pre-shifted V[] and D[] values. D[] will always be the second operand\r
30 * of mul/mla to achieve higher speed as D[] has lower amplitude than V[].\r
31 ****************************************************************************/\r
32 #if defined(OPTIMIZE_FOR_SPEED)\r
33 .align 2\r
34 .global mpc_decoder_windowing_D\r
35 .type mpc_decoder_windowing_D, %function\r
36 mpc_decoder_windowing_D:\r
37 /* r0 = Data[] */\r
38 /* r1 = V[] */\r
39 /* r2 = D[] */\r
40 /* lr = counter */\r
41 \r
42 stmfd sp!, {r4-r12, lr}\r
43 \r
44 mov lr, #32\r
45 .loop32:\r
46 ldmia r2!, { r3-r10 } /* load D[00..07] */\r
47 ldr r11, [r1] /* 0 */\r
48 mul r12, r11, r3 \r
49 ldr r11, [r1, #96*4] /* 1 */\r
50 mla r12, r11, r4, r12 \r
51 ldr r11, [r1, #128*4] /* 2 */\r
52 mla r12, r11, r5, r12 \r
53 ldr r11, [r1, #224*4] /* 3 */\r
54 mla r12, r11, r6, r12 \r
55 ldr r11, [r1, #256*4] /* 4 */\r
56 mla r12, r11, r7, r12 \r
57 ldr r11, [r1, #352*4] /* 5 */\r
58 mla r12, r11, r8, r12 \r
59 ldr r11, [r1, #384*4] /* 6 */\r
60 mla r12, r11, r9, r12 \r
61 ldr r11, [r1, #480*4] /* 7 */\r
62 mla r12, r11, r10, r12 \r
63 ldmia r2!, { r3-r10 } /* load D[08..15] */\r
64 ldr r11, [r1, #512*4] /* 8 */\r
65 mla r12, r11, r3, r12 \r
66 ldr r11, [r1, #608*4] /* 9 */\r
67 mla r12, r11, r4, r12 \r
68 ldr r11, [r1, #640*4] /* 10 */\r
69 mla r12, r11, r5, r12 \r
70 ldr r11, [r1, #736*4] /* 11 */\r
71 mla r12, r11, r6, r12 \r
72 ldr r11, [r1, #768*4] /* 12 */\r
73 mla r12, r11, r7, r12 \r
74 ldr r11, [r1, #864*4] /* 13 */\r
75 mla r12, r11, r8, r12 \r
76 ldr r11, [r1, #896*4] /* 14 */\r
77 mla r12, r11, r9, r12 \r
78 ldr r11, [r1, #992*4] /* 15 */\r
79 mla r12, r11, r10, r12\r
80 mov r12, r12, asr #1 /* post shift to compensate for pre-shifting */\r
81 str r12, [r0], #4 /* store Data */\r
82 add r1, r1, #4 /* V++ */\r
83 \r
84 subs lr, lr, #1\r
85 bgt .loop32\r
86 \r
87 ldmfd sp!, {r4-r12, pc}\r
88 .mpc_dewindowing_end:\r
89 .size mpc_decoder_windowing_D,.mpc_dewindowing_end-mpc_decoder_windowing_D\r
90 #else\r
91 /****************************************************************************\r
92 * void mpc_decoder_windowing_D(...)\r
93 * \r
94 * 2nd step within synthesis filter. Does the dewindowing.\r
95 * 64=32x32 multiplies\r
96 * Uses un-shifted D[]-values. D[] will always be the second operand of \r
97 * smull/smlal to achieve higher speed as D[] has lower amplitude than V[].\r
98 ****************************************************************************/\r
99 .align 2\r
100 .global mpc_decoder_windowing_D\r
101 .type mpc_decoder_windowing_D, %function\r
102 mpc_decoder_windowing_D:\r
103 /* r0 = Data[] */\r
104 /* r1 = V[] */\r
105 /* r2 = D[] */\r
106 /* lr = counter */\r
107 \r
108 stmfd sp!, {r4-r9, lr}\r
109 \r
110 mov lr, #32\r
111 .loop32:\r
112 ldmia r2!, { r3-r6 } /* load D[00..03] */\r
113 ldr r7, [r1] /* 0 */\r
114 smull r8, r9, r7, r3\r
115 ldr r7, [r1, #96*4] /* 1 */\r
116 smlal r8, r9, r7, r4\r
117 ldr r7, [r1, #128*4] /* 2 */\r
118 smlal r8, r9, r7, r5\r
119 ldr r7, [r1, #224*4] /* 3 */\r
120 smlal r8, r9, r7, r6\r
121 ldmia r2!, { r3-r6 } /* load D[04..07] */\r
122 ldr r7, [r1, #256*4] /* 4 */\r
123 smlal r8, r9, r7, r3\r
124 ldr r7, [r1, #352*4] /* 5 */\r
125 smlal r8, r9, r7, r4\r
126 ldr r7, [r1, #384*4] /* 6 */\r
127 smlal r8, r9, r7, r5\r
128 ldr r7, [r1, #480*4] /* 7 */\r
129 smlal r8, r9, r7, r6\r
130 ldmia r2!, { r3-r6 } /* load D[08..11] */\r
131 ldr r7, [r1, #512*4] /* 8 */\r
132 smlal r8, r9, r7, r3\r
133 ldr r7, [r1, #608*4] /* 9 */\r
134 smlal r8, r9, r7, r4\r
135 ldr r7, [r1, #640*4] /* 10 */\r
136 smlal r8, r9, r7, r5\r
137 ldr r7, [r1, #736*4] /* 11 */\r
138 smlal r8, r9, r7, r6\r
139 ldmia r2!, { r3-r6 } /* load D[12..15] */\r
140 ldr r7, [r1, #768*4] /* 12 */\r
141 smlal r8, r9, r7, r3\r
142 ldr r7, [r1, #864*4] /* 13 */\r
143 smlal r8, r9, r7, r4\r
144 ldr r7, [r1, #896*4] /* 14 */\r
145 smlal r8, r9, r7, r5\r
146 ldr r7, [r1, #992*4] /* 15 */\r
147 smlal r8, r9, r7, r6\r
148 mov r8, r8, lsr #16\r
149 orr r8, r8, r9, lsl #16 /* (lo>>16) || (hi<<16) */\r
150 str r8, [r0], #4 /* store Data */\r
151 add r1, r1, #4 /* V++ */\r
152 \r
153 subs lr, lr, #1\r
154 bgt .loop32\r
155 \r
156 ldmfd sp!, {r4-r9, pc}\r
157 .mpc_dewindowing_end:\r
158 .size mpc_decoder_windowing_D,.mpc_dewindowing_end-mpc_decoder_windowing_D\r
159 #endif\r