diff options
author | Michael Sevakis <jethead71@rockbox.org> | 2013-05-18 01:45:03 -0400 |
---|---|---|
committer | Michael Sevakis <jethead71@rockbox.org> | 2013-05-21 00:02:14 -0400 |
commit | 87021f7c0ac4620eafd185ff11905ee643f72b6c (patch) | |
tree | 03ae48f3d999cd8743af40cc5df933f64f6df2d2 /lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.c | |
parent | a17d6de5bc727b0bb55764ecef2605ae689e8dab (diff) | |
download | rockbox-87021f7c0ac4620eafd185ff11905ee643f72b6c.tar.gz rockbox-87021f7c0ac4620eafd185ff11905ee643f72b6c.zip |
SPC Codec: Refactor for CPU and clean up some things.
CPU optimization gets its own files in which to fill-in optimizable
routines.
Some pointless #if 0's for profiling need removal. Those macros are
empty if not profiling.
Force some functions that are undesirable to be force-inlined by the
compiler to be not inlined.
Change-Id: Ia7b7e45380d7efb20c9b1a4d52e05db3ef6bbaab
Diffstat (limited to 'lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.c')
-rw-r--r-- | lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.c | 198 |
1 files changed, 198 insertions, 0 deletions
diff --git a/lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.c b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.c new file mode 100644 index 0000000000..b0d14d157e --- /dev/null +++ b/lib/rbcodec/codecs/libspc/cpu/spc_dsp_coldfire.c | |||
@@ -0,0 +1,198 @@ | |||
1 | /*************************************************************************** | ||
2 | * __________ __ ___. | ||
3 | * Open \______ \ ____ ____ | | _\_ |__ _______ ___ | ||
4 | * Source | _// _ \_/ ___\| |/ /| __ \ / _ \ \/ / | ||
5 | * Jukebox | | ( <_> ) \___| < | \_\ ( <_> > < < | ||
6 | * Firmware |____|_ /\____/ \___ >__|_ \|___ /\____/__/\_ \ | ||
7 | * \/ \/ \/ \/ \/ | ||
8 | * $Id$ | ||
9 | * | ||
10 | * Copyright (C) 2007 Michael Sevakis (jhMikeS) | ||
11 | * | ||
12 | * This program is free software; you can redistribute it and/or | ||
13 | * modify it under the terms of the GNU General Public License | ||
14 | * as published by the Free Software Foundation; either version 2 | ||
15 | * of the License, or (at your option) any later version. | ||
16 | * | ||
17 | * This software is distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY | ||
18 | * KIND, either express or implied. | ||
19 | * | ||
20 | ****************************************************************************/ | ||
21 | #if SPC_NOINTERP | ||
22 | |||
23 | #define SPC_LINEAR_INTERP | ||
24 | static inline int linear_interp( int16_t const* samples, int32_t position ) | ||
25 | { | ||
26 | uint32_t f = position; | ||
27 | int32_t y0, y1; | ||
28 | |||
29 | /** | ||
30 | * output = y0 + f*y1 - f*y0 | ||
31 | */ | ||
32 | asm volatile ( | ||
33 | "move.l %[f], %[y1] \n" /* separate frac and whole */ | ||
34 | "and.l #0xfff, %[f] \n" | ||
35 | "asr.l %[sh], %[y1] \n" | ||
36 | "move.l 2(%[s], %[y1].l*2), %[y1] \n" /* y0=upper, y1=lower */ | ||
37 | "mac.w %[f]l, %[y1]l, %%acc0 \n" /* %acc0 = f*y1 */ | ||
38 | "msac.w %[f]l, %[y1]u, %%acc0 \n" /* %acc0 -= f*y0 */ | ||
39 | "swap %[y1] \n" /* separate out y0 and sign extend */ | ||
40 | "movea.w %[y1], %[y0] \n" | ||
41 | "movclr.l %%acc0, %[y1] \n" /* fetch, scale down, add y0 */ | ||
42 | "asr.l %[sh], %[y1] \n" /* output = y0 + (result >> 12) */ | ||
43 | "add.l %[y0], %[y1] \n" | ||
44 | : [f]"+d"(f), [y0]"=&a"(y0), [y1]"=&d"(y1) | ||
45 | : [s]"a"(samples), [sh]"d"(12)); | ||
46 | |||
47 | return y1; | ||
48 | } | ||
49 | |||
50 | #define SPC_LINEAR_AMP | ||
51 | static inline int linear_amp( struct voice_t* voice, int output, | ||
52 | int* amp_0, int* amp_1 ) | ||
53 | { | ||
54 | asm volatile ( | ||
55 | "mac.w %[out]l, %[envx]l, %%acc0" | ||
56 | : | ||
57 | : [out]"r"(output), [envx]"r"(voice->envx)); | ||
58 | asm volatile ( | ||
59 | "movclr.l %%acc0, %[out] \n" | ||
60 | "asr.l #8, %[out] \n" | ||
61 | "mac.l %[v0], %[out], %%acc0 \n" | ||
62 | "mac.l %[v1], %[out], %%acc1 \n" | ||
63 | "asr.l #3, %[out] \n" | ||
64 | : [out]"+r"(output) | ||
65 | : [v0]"r"((int) voice->volume [0]), | ||
66 | [v1]"r"((int) voice->volume [1])); | ||
67 | asm volatile ( | ||
68 | "movclr.l %%acc0, %[a0] \n" | ||
69 | "asr.l #3, %[a0] \n" | ||
70 | "movclr.l %%acc1, %[a1] \n" | ||
71 | "asr.l #3, %[a1] \n" | ||
72 | : [a0]"=d"(*amp_0), [a1]"=d"(*amp_1)); | ||
73 | |||
74 | return output; | ||
75 | } | ||
76 | |||
77 | #endif /* SPC_NOINTERP */ | ||
78 | |||
79 | |||
80 | #if !SPC_NOECHO | ||
81 | |||
82 | #define SPC_DSP_ECHO_APPLY | ||
83 | |||
84 | /* Echo filter history */ | ||
85 | static int32_t fir_buf[FIR_BUF_CNT] IBSS_ATTR_SPC | ||
86 | __attribute__(( aligned(FIR_BUF_ALIGN*1) )); | ||
87 | |||
88 | static inline void echo_init( struct Spc_Dsp* this ) | ||
89 | { | ||
90 | /* Initialize mask register with the buffer address mask */ | ||
91 | asm volatile ("move.l %0, %%mask" : : "i"(FIR_BUF_MASK)); | ||
92 | this->fir.ptr = fir_buf; | ||
93 | this->fir.hist_ptr = &fir_buf [1]; | ||
94 | ci->memset( fir_buf, 0, sizeof fir_buf ); | ||
95 | } | ||
96 | |||
97 | static inline void echo_apply( struct Spc_Dsp* this, uint8_t* echo_ptr, | ||
98 | int* out_0, int* out_1 ) | ||
99 | { | ||
100 | int t0, t1, t2; | ||
101 | |||
102 | t1 = swap_odd_even32( *(int32_t *)echo_ptr ); | ||
103 | |||
104 | /* Keep last 8 samples */ | ||
105 | *this->fir.ptr = t1; | ||
106 | this->fir.ptr = this->fir.hist_ptr; | ||
107 | |||
108 | asm volatile ( | ||
109 | "move.l (%[c]) , %[t2] \n" | ||
110 | "mac.w %[t1]u, %[t2]u, <<, (%[p])+&, %[t0], %%acc0 \n" | ||
111 | "mac.w %[t1]l, %[t2]u, <<, (%[p])& , %[t1], %%acc1 \n" | ||
112 | "mac.w %[t0]u, %[t2]l, << , %%acc0 \n" | ||
113 | "mac.w %[t0]l, %[t2]l, <<, 4(%[c]) , %[t2], %%acc1 \n" | ||
114 | "mac.w %[t1]u, %[t2]u, <<, 4(%[p])& , %[t0], %%acc0 \n" | ||
115 | "mac.w %[t1]l, %[t2]u, <<, 8(%[p])& , %[t1], %%acc1 \n" | ||
116 | "mac.w %[t0]u, %[t2]l, << , %%acc0 \n" | ||
117 | "mac.w %[t0]l, %[t2]l, <<, 8(%[c]) , %[t2], %%acc1 \n" | ||
118 | "mac.w %[t1]u, %[t2]u, <<, 12(%[p])& , %[t0], %%acc0 \n" | ||
119 | "mac.w %[t1]l, %[t2]u, <<, 16(%[p])& , %[t1], %%acc1 \n" | ||
120 | "mac.w %[t0]u, %[t2]l, << , %%acc0 \n" | ||
121 | "mac.w %[t0]l, %[t2]l, <<, 12(%[c]) , %[t2], %%acc1 \n" | ||
122 | "mac.w %[t1]u, %[t2]u, <<, 20(%[p])& , %[t0], %%acc0 \n" | ||
123 | "mac.w %[t1]l, %[t2]u, << , %%acc1 \n" | ||
124 | "mac.w %[t0]u, %[t2]l, << , %%acc0 \n" | ||
125 | "mac.w %[t0]l, %[t2]l, << , %%acc1 \n" | ||
126 | : [t0]"=&r"(t0), [t1]"+r"(t1), [t2]"=&r"(t2), | ||
127 | [p]"+a"(this->fir.hist_ptr) | ||
128 | : [c]"a"(this->fir.coeff)); | ||
129 | asm volatile ( | ||
130 | "movclr.l %%acc0, %[o0] \n" | ||
131 | "movclr.l %%acc1, %[o1] \n" | ||
132 | "mac.l %[ev0], %[o0], >>, %%acc2 \n" /* echo volume */ | ||
133 | "mac.l %[ev1], %[o1], >>, %%acc3 \n" | ||
134 | : [o0]"=&r"(*out_0), [o1]"=&r"(*out_1) | ||
135 | : [ev0]"r"((int) this->r.g.echo_volume_0), | ||
136 | [ev1]"r"((int) this->r.g.echo_volume_1)); | ||
137 | } | ||
138 | |||
139 | #define SPC_DSP_ECHO_FEEDBACK | ||
140 | static inline void echo_feedback( struct Spc_Dsp* this, uint8_t* echo_ptr, | ||
141 | int echo_0, int echo_1, int fb_0, int fb_1 ) | ||
142 | { | ||
143 | asm volatile ( | ||
144 | /* scale echo voices; saturate if overflow */ | ||
145 | "mac.l %[sh], %[e1] , %%acc1 \n" | ||
146 | "mac.l %[sh], %[e0] , %%acc0 \n" | ||
147 | /* add scaled output from FIR filter */ | ||
148 | "mac.l %[fb1], %[ef], <<, %%acc1 \n" | ||
149 | "mac.l %[fb0], %[ef], <<, %%acc0 \n" | ||
150 | : | ||
151 | : [e0]"d"(echo_0), [e1]"d"(echo_1), | ||
152 | [fb0]"r"(fb_0), [fb1]"r"(fb_1), | ||
153 | [ef]"r"((int)this->r.g.echo_feedback), | ||
154 | [sh]"r"(1 << 9)); | ||
155 | /* swap and fetch feedback results */ | ||
156 | int t0; | ||
157 | asm volatile( | ||
158 | "move.l #0x00ff00ff, %[t0] \n" | ||
159 | "movclr.l %%acc1, %[e1] \n" | ||
160 | "swap.w %[e1] \n" | ||
161 | "movclr.l %%acc0, %[e0] \n" | ||
162 | "move.w %[e1], %[e0] \n" | ||
163 | "and.l %[e0], %[t0] \n" | ||
164 | "eor.l %[t0], %[e0] \n" | ||
165 | "lsl.l #8, %[t0] \n" | ||
166 | "lsr.l #8, %[e0] \n" | ||
167 | "or.l %[e0], %[t0] \n" | ||
168 | : [e0]"=&d"(echo_0), [e1]"=&d"(echo_1), | ||
169 | [t0]"=&d"(t0)); | ||
170 | |||
171 | /* save final feedback into echo buffer */ | ||
172 | *(int32_t *)echo_ptr = t0; | ||
173 | } | ||
174 | |||
175 | #define SPC_DSP_GENERATE_OUTPUT | ||
176 | static inline void echo_output( struct Spc_Dsp* this, int global_muting, | ||
177 | int global_vol_0, int global_vol_1, int chans_0, int chans_1, | ||
178 | int fb_0, int fb_1, int* out_0, int* out_1 ) | ||
179 | { | ||
180 | asm volatile ( | ||
181 | "mac.l %[ch0], %[gv0], %%acc2 \n" /* global volume */ | ||
182 | "mac.l %[ch1], %[gv1], %%acc3 \n" | ||
183 | : | ||
184 | : [ch0]"r"(chans_0), [gv0]"r"(global_vol_0), | ||
185 | [ch1]"r"(chans_1), [gv1]"r"(global_vol_1)); | ||
186 | asm volatile ( | ||
187 | "movclr.l %%acc2, %[a0] \n" /* fetch mixed output */ | ||
188 | "movclr.l %%acc3, %[a1] \n" | ||
189 | "asr.l %[gm], %[a0] \n" /* scale by global_muting shift */ | ||
190 | "asr.l %[gm], %[a1] \n" | ||
191 | : [a0]"=&d"(*out_0), [a1]"=&d"(*out_1) | ||
192 | : [gm]"d"(global_muting)); | ||
193 | |||
194 | /* scaled echo is stored in %acc2 and %acc3 */ | ||
195 | (void)this; (void)fb_0; (void)fb_1; | ||
196 | } | ||
197 | |||
198 | #endif /* !SPC_NOECHO */ | ||