Mercurial > spc_convert
diff fast_dsp/SPC_DSP.cpp @ 0:e38dacceb958
initial import
| author | Robert McIntyre <rlm@mit.edu> |
|---|---|
| date | Fri, 21 Oct 2011 05:53:11 -0700 |
| parents | |
| children |
line wrap: on
line diff
1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000 1.2 +++ b/fast_dsp/SPC_DSP.cpp Fri Oct 21 05:53:11 2011 -0700 1.3 @@ -0,0 +1,703 @@ 1.4 +// snes_spc 0.9.0. http://www.slack.net/~ant/ 1.5 + 1.6 +#include "SPC_DSP.h" 1.7 + 1.8 +#include "blargg_endian.h" 1.9 +#include <string.h> 1.10 + 1.11 +/* Copyright (C) 2007 Shay Green. This module is free software; you 1.12 +can redistribute it and/or modify it under the terms of the GNU Lesser 1.13 +General Public License as published by the Free Software Foundation; either 1.14 +version 2.1 of the License, or (at your option) any later version. This 1.15 +module is distributed in the hope that it will be useful, but WITHOUT ANY 1.16 +WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS 1.17 +FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more 1.18 +details. You should have received a copy of the GNU Lesser General Public 1.19 +License along with this module; if not, write to the Free Software Foundation, 1.20 +Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ 1.21 + 1.22 +#include "blargg_source.h" 1.23 + 1.24 +#ifdef BLARGG_ENABLE_OPTIMIZER 1.25 + #include BLARGG_ENABLE_OPTIMIZER 1.26 +#endif 1.27 + 1.28 +#if INT_MAX < 0x7FFFFFFF 1.29 + #error "Requires that int type have at least 32 bits" 1.30 +#endif 1.31 + 1.32 + 1.33 +// TODO: add to blargg_endian.h 1.34 +#define GET_LE16SA( addr ) ((BOOST::int16_t) GET_LE16( addr )) 1.35 +#define GET_LE16A( addr ) GET_LE16( addr ) 1.36 +#define SET_LE16A( addr, data ) SET_LE16( addr, data ) 1.37 + 1.38 +static BOOST::uint8_t const initial_regs [SPC_DSP::register_count] = 1.39 +{ 1.40 + 0x45,0x8B,0x5A,0x9A,0xE4,0x82,0x1B,0x78,0x00,0x00,0xAA,0x96,0x89,0x0E,0xE0,0x80, 1.41 + 0x2A,0x49,0x3D,0xBA,0x14,0xA0,0xAC,0xC5,0x00,0x00,0x51,0xBB,0x9C,0x4E,0x7B,0xFF, 1.42 + 0xF4,0xFD,0x57,0x32,0x37,0xD9,0x42,0x22,0x00,0x00,0x5B,0x3C,0x9F,0x1B,0x87,0x9A, 1.43 + 0x6F,0x27,0xAF,0x7B,0xE5,0x68,0x0A,0xD9,0x00,0x00,0x9A,0xC5,0x9C,0x4E,0x7B,0xFF, 1.44 + 0xEA,0x21,0x78,0x4F,0xDD,0xED,0x24,0x14,0x00,0x00,0x77,0xB1,0xD1,0x36,0xC1,0x67, 1.45 + 0x52,0x57,0x46,0x3D,0x59,0xF4,0x87,0xA4,0x00,0x00,0x7E,0x44,0x9C,0x4E,0x7B,0xFF, 1.46 + 0x75,0xF5,0x06,0x97,0x10,0xC3,0x24,0xBB,0x00,0x00,0x7B,0x7A,0xE0,0x60,0x12,0x0F, 1.47 + 0xF7,0x74,0x1C,0xE5,0x39,0x3D,0x73,0xC1,0x00,0x00,0x7A,0xB3,0xFF,0x4E,0x7B,0xFF 1.48 +}; 1.49 + 1.50 +// if ( io < -32768 ) io = -32768; 1.51 +// if ( io > 32767 ) io = 32767; 1.52 +#define CLAMP16( io )\ 1.53 +{\ 1.54 + if ( (int16_t) io != io )\ 1.55 + io = (io >> 31) ^ 0x7FFF;\ 1.56 +} 1.57 + 1.58 +// Access global DSP register 1.59 +#define REG(n) m.regs [r_##n] 1.60 + 1.61 +// Access voice DSP register 1.62 +#define VREG(r,n) r [v_##n] 1.63 + 1.64 +#define WRITE_SAMPLES( l, r, out ) \ 1.65 +{\ 1.66 + out [0] = l;\ 1.67 + out [1] = r;\ 1.68 + out += 2;\ 1.69 + if ( out >= m.out_end )\ 1.70 + {\ 1.71 + check( out == m.out_end );\ 1.72 + check( m.out_end != &m.extra [extra_size] || \ 1.73 + (m.extra <= m.out_begin && m.extra < &m.extra [extra_size]) );\ 1.74 + out = m.extra;\ 1.75 + m.out_end = &m.extra [extra_size];\ 1.76 + }\ 1.77 +}\ 1.78 + 1.79 +void SPC_DSP::set_output( sample_t* out, int size ) 1.80 +{ 1.81 + require( (size & 1) == 0 ); // must be even 1.82 + if ( !out ) 1.83 + { 1.84 + out = m.extra; 1.85 + size = extra_size; 1.86 + } 1.87 + m.out_begin = out; 1.88 + m.out = out; 1.89 + m.out_end = out + size; 1.90 +} 1.91 + 1.92 +// Volume registers and efb are signed! Easy to forget int8_t cast. 1.93 +// Prefixes are to avoid accidental use of locals with same names. 1.94 + 1.95 +// Interleved gauss table (to improve cache coherency) 1.96 +// interleved_gauss [i] = gauss [(i & 1) * 256 + 255 - (i >> 1 & 0xFF)] 1.97 +static short const interleved_gauss [512] = 1.98 +{ 1.99 + 370,1305, 366,1305, 362,1304, 358,1304, 354,1304, 351,1304, 347,1304, 343,1303, 1.100 + 339,1303, 336,1303, 332,1302, 328,1302, 325,1301, 321,1300, 318,1300, 314,1299, 1.101 + 311,1298, 307,1297, 304,1297, 300,1296, 297,1295, 293,1294, 290,1293, 286,1292, 1.102 + 283,1291, 280,1290, 276,1288, 273,1287, 270,1286, 267,1284, 263,1283, 260,1282, 1.103 + 257,1280, 254,1279, 251,1277, 248,1275, 245,1274, 242,1272, 239,1270, 236,1269, 1.104 + 233,1267, 230,1265, 227,1263, 224,1261, 221,1259, 218,1257, 215,1255, 212,1253, 1.105 + 210,1251, 207,1248, 204,1246, 201,1244, 199,1241, 196,1239, 193,1237, 191,1234, 1.106 + 188,1232, 186,1229, 183,1227, 180,1224, 178,1221, 175,1219, 173,1216, 171,1213, 1.107 + 168,1210, 166,1207, 163,1205, 161,1202, 159,1199, 156,1196, 154,1193, 152,1190, 1.108 + 150,1186, 147,1183, 145,1180, 143,1177, 141,1174, 139,1170, 137,1167, 134,1164, 1.109 + 132,1160, 130,1157, 128,1153, 126,1150, 124,1146, 122,1143, 120,1139, 118,1136, 1.110 + 117,1132, 115,1128, 113,1125, 111,1121, 109,1117, 107,1113, 106,1109, 104,1106, 1.111 + 102,1102, 100,1098, 99,1094, 97,1090, 95,1086, 94,1082, 92,1078, 90,1074, 1.112 + 89,1070, 87,1066, 86,1061, 84,1057, 83,1053, 81,1049, 80,1045, 78,1040, 1.113 + 77,1036, 76,1032, 74,1027, 73,1023, 71,1019, 70,1014, 69,1010, 67,1005, 1.114 + 66,1001, 65, 997, 64, 992, 62, 988, 61, 983, 60, 978, 59, 974, 58, 969, 1.115 + 56, 965, 55, 960, 54, 955, 53, 951, 52, 946, 51, 941, 50, 937, 49, 932, 1.116 + 48, 927, 47, 923, 46, 918, 45, 913, 44, 908, 43, 904, 42, 899, 41, 894, 1.117 + 40, 889, 39, 884, 38, 880, 37, 875, 36, 870, 36, 865, 35, 860, 34, 855, 1.118 + 33, 851, 32, 846, 32, 841, 31, 836, 30, 831, 29, 826, 29, 821, 28, 816, 1.119 + 27, 811, 27, 806, 26, 802, 25, 797, 24, 792, 24, 787, 23, 782, 23, 777, 1.120 + 22, 772, 21, 767, 21, 762, 20, 757, 20, 752, 19, 747, 19, 742, 18, 737, 1.121 + 17, 732, 17, 728, 16, 723, 16, 718, 15, 713, 15, 708, 15, 703, 14, 698, 1.122 + 14, 693, 13, 688, 13, 683, 12, 678, 12, 674, 11, 669, 11, 664, 11, 659, 1.123 + 10, 654, 10, 649, 10, 644, 9, 640, 9, 635, 9, 630, 8, 625, 8, 620, 1.124 + 8, 615, 7, 611, 7, 606, 7, 601, 6, 596, 6, 592, 6, 587, 6, 582, 1.125 + 5, 577, 5, 573, 5, 568, 5, 563, 4, 559, 4, 554, 4, 550, 4, 545, 1.126 + 4, 540, 3, 536, 3, 531, 3, 527, 3, 522, 3, 517, 2, 513, 2, 508, 1.127 + 2, 504, 2, 499, 2, 495, 2, 491, 2, 486, 1, 482, 1, 477, 1, 473, 1.128 + 1, 469, 1, 464, 1, 460, 1, 456, 1, 451, 1, 447, 1, 443, 1, 439, 1.129 + 0, 434, 0, 430, 0, 426, 0, 422, 0, 418, 0, 414, 0, 410, 0, 405, 1.130 + 0, 401, 0, 397, 0, 393, 0, 389, 0, 385, 0, 381, 0, 378, 0, 374, 1.131 +}; 1.132 + 1.133 + 1.134 +//// Counters 1.135 + 1.136 +#define RATE( rate, div )\ 1.137 + (rate >= div ? rate / div * 8 - 1 : rate - 1) 1.138 + 1.139 +static unsigned const counter_mask [32] = 1.140 +{ 1.141 + RATE( 2,2), RATE(2048,4), RATE(1536,3), 1.142 + RATE(1280,5), RATE(1024,4), RATE( 768,3), 1.143 + RATE( 640,5), RATE( 512,4), RATE( 384,3), 1.144 + RATE( 320,5), RATE( 256,4), RATE( 192,3), 1.145 + RATE( 160,5), RATE( 128,4), RATE( 96,3), 1.146 + RATE( 80,5), RATE( 64,4), RATE( 48,3), 1.147 + RATE( 40,5), RATE( 32,4), RATE( 24,3), 1.148 + RATE( 20,5), RATE( 16,4), RATE( 12,3), 1.149 + RATE( 10,5), RATE( 8,4), RATE( 6,3), 1.150 + RATE( 5,5), RATE( 4,4), RATE( 3,3), 1.151 + RATE( 2,4), 1.152 + RATE( 1,4) 1.153 +}; 1.154 +#undef RATE 1.155 + 1.156 +inline void SPC_DSP::init_counter() 1.157 +{ 1.158 + // counters start out with this synchronization 1.159 + m.counters [0] = 1; 1.160 + m.counters [1] = 0; 1.161 + m.counters [2] = -0x20u; 1.162 + m.counters [3] = 0x0B; 1.163 + 1.164 + int n = 2; 1.165 + for ( int i = 1; i < 32; i++ ) 1.166 + { 1.167 + m.counter_select [i] = &m.counters [n]; 1.168 + if ( !--n ) 1.169 + n = 3; 1.170 + } 1.171 + m.counter_select [ 0] = &m.counters [0]; 1.172 + m.counter_select [30] = &m.counters [2]; 1.173 +} 1.174 + 1.175 +inline void SPC_DSP::run_counter( int i ) 1.176 +{ 1.177 + int n = m.counters [i]; 1.178 + if ( !(n-- & 7) ) 1.179 + n -= 6 - i; 1.180 + m.counters [i] = n; 1.181 +} 1.182 + 1.183 +#define READ_COUNTER( rate )\ 1.184 + (*m.counter_select [rate] & counter_mask [rate]) 1.185 + 1.186 + 1.187 +//// Emulation 1.188 + 1.189 +void SPC_DSP::run( int clock_count ) 1.190 +{ 1.191 + int new_phase = m.phase + clock_count; 1.192 + int count = new_phase >> 5; 1.193 + m.phase = new_phase & 31; 1.194 + if ( !count ) 1.195 + return; 1.196 + 1.197 + uint8_t* const ram = m.ram; 1.198 + uint8_t const* const dir = &ram [REG(dir) * 0x100]; 1.199 + int const slow_gaussian = (REG(pmon) >> 1) | REG(non); 1.200 + int const noise_rate = REG(flg) & 0x1F; 1.201 + 1.202 + // Global volume 1.203 + int mvoll = (int8_t) REG(mvoll); 1.204 + int mvolr = (int8_t) REG(mvolr); 1.205 + if ( mvoll * mvolr < m.surround_threshold ) 1.206 + mvoll = -mvoll; // eliminate surround 1.207 + 1.208 + do 1.209 + { 1.210 + // KON/KOFF reading 1.211 + if ( (m.every_other_sample ^= 1) != 0 ) 1.212 + { 1.213 + m.new_kon &= ~m.kon; 1.214 + m.kon = m.new_kon; 1.215 + m.t_koff = REG(koff); 1.216 + } 1.217 + 1.218 + run_counter( 1 ); 1.219 + run_counter( 2 ); 1.220 + run_counter( 3 ); 1.221 + 1.222 + // Noise 1.223 + if ( !READ_COUNTER( noise_rate ) ) 1.224 + { 1.225 + int feedback = (m.noise << 13) ^ (m.noise << 14); 1.226 + m.noise = (feedback & 0x4000) ^ (m.noise >> 1); 1.227 + } 1.228 + 1.229 + // Voices 1.230 + int pmon_input = 0; 1.231 + int main_out_l = 0; 1.232 + int main_out_r = 0; 1.233 + int echo_out_l = 0; 1.234 + int echo_out_r = 0; 1.235 + voice_t* v = m.voices; 1.236 + uint8_t* v_regs = m.regs; 1.237 + int vbit = 1; 1.238 + do 1.239 + { 1.240 + #define SAMPLE_PTR(i) GET_LE16A( &dir [VREG(v_regs,srcn) * 4 + i * 2] ) 1.241 + 1.242 + int brr_header = ram [v->brr_addr]; 1.243 + int kon_delay = v->kon_delay; 1.244 + 1.245 + // Pitch 1.246 + int pitch = GET_LE16A( &VREG(v_regs,pitchl) ) & 0x3FFF; 1.247 + if ( REG(pmon) & vbit ) 1.248 + pitch += ((pmon_input >> 5) * pitch) >> 10; 1.249 + 1.250 + // KON phases 1.251 + if ( --kon_delay >= 0 ) 1.252 + { 1.253 + v->kon_delay = kon_delay; 1.254 + 1.255 + // Get ready to start BRR decoding on next sample 1.256 + if ( kon_delay == 4 ) 1.257 + { 1.258 + v->brr_addr = SAMPLE_PTR( 0 ); 1.259 + v->brr_offset = 1; 1.260 + v->buf_pos = v->buf; 1.261 + brr_header = 0; // header is ignored on this sample 1.262 + } 1.263 + 1.264 + // Envelope is never run during KON 1.265 + v->env = 0; 1.266 + v->hidden_env = 0; 1.267 + 1.268 + // Disable BRR decoding until last three samples 1.269 + v->interp_pos = (kon_delay & 3 ? 0x4000 : 0); 1.270 + 1.271 + // Pitch is never added during KON 1.272 + pitch = 0; 1.273 + } 1.274 + 1.275 + int env = v->env; 1.276 + 1.277 + // Gaussian interpolation 1.278 + { 1.279 + int output = 0; 1.280 + VREG(v_regs,envx) = (uint8_t) (env >> 4); 1.281 + if ( env ) 1.282 + { 1.283 + // Make pointers into gaussian based on fractional position between samples 1.284 + int offset = (unsigned) v->interp_pos >> 3 & 0x1FE; 1.285 + short const* fwd = interleved_gauss + offset; 1.286 + short const* rev = interleved_gauss + 510 - offset; // mirror left half of gaussian 1.287 + 1.288 + int const* in = &v->buf_pos [(unsigned) v->interp_pos >> 12]; 1.289 + 1.290 + if ( !(slow_gaussian & vbit) ) // 99% 1.291 + { 1.292 + // Faster approximation when exact sample value isn't necessary for pitch mod 1.293 + output = (fwd [0] * in [0] + 1.294 + fwd [1] * in [1] + 1.295 + rev [1] * in [2] + 1.296 + rev [0] * in [3]) >> 11; 1.297 + output = (output * env) >> 11; 1.298 + } 1.299 + else 1.300 + { 1.301 + output = (int16_t) (m.noise * 2); 1.302 + if ( !(REG(non) & vbit) ) 1.303 + { 1.304 + output = (fwd [0] * in [0]) >> 11; 1.305 + output += (fwd [1] * in [1]) >> 11; 1.306 + output += (rev [1] * in [2]) >> 11; 1.307 + output = (int16_t) output; 1.308 + output += (rev [0] * in [3]) >> 11; 1.309 + 1.310 + CLAMP16( output ); 1.311 + output &= ~1; 1.312 + } 1.313 + output = (output * env) >> 11 & ~1; 1.314 + } 1.315 + 1.316 + // Output 1.317 + int l = output * v->volume [0]; 1.318 + int r = output * v->volume [1]; 1.319 + 1.320 + main_out_l += l; 1.321 + main_out_r += r; 1.322 + 1.323 + if ( REG(eon) & vbit ) 1.324 + { 1.325 + echo_out_l += l; 1.326 + echo_out_r += r; 1.327 + } 1.328 + } 1.329 + 1.330 + pmon_input = output; 1.331 + VREG(v_regs,outx) = (uint8_t) (output >> 8); 1.332 + } 1.333 + 1.334 + // Soft reset or end of sample 1.335 + if ( REG(flg) & 0x80 || (brr_header & 3) == 1 ) 1.336 + { 1.337 + v->env_mode = env_release; 1.338 + env = 0; 1.339 + } 1.340 + 1.341 + if ( m.every_other_sample ) 1.342 + { 1.343 + // KOFF 1.344 + if ( m.t_koff & vbit ) 1.345 + v->env_mode = env_release; 1.346 + 1.347 + // KON 1.348 + if ( m.kon & vbit ) 1.349 + { 1.350 + v->kon_delay = 5; 1.351 + v->env_mode = env_attack; 1.352 + REG(endx) &= ~vbit; 1.353 + } 1.354 + } 1.355 + 1.356 + // Envelope 1.357 + if ( !v->kon_delay ) 1.358 + { 1.359 + if ( v->env_mode == env_release ) // 97% 1.360 + { 1.361 + env -= 0x8; 1.362 + v->env = env; 1.363 + if ( env <= 0 ) 1.364 + { 1.365 + v->env = 0; 1.366 + goto skip_brr; // no BRR decoding for you! 1.367 + } 1.368 + } 1.369 + else // 3% 1.370 + { 1.371 + int rate; 1.372 + int const adsr0 = VREG(v_regs,adsr0); 1.373 + int env_data = VREG(v_regs,adsr1); 1.374 + if ( adsr0 >= 0x80 ) // 97% ADSR 1.375 + { 1.376 + if ( v->env_mode > env_decay ) // 89% 1.377 + { 1.378 + env--; 1.379 + env -= env >> 8; 1.380 + rate = env_data & 0x1F; 1.381 + 1.382 + // optimized handling 1.383 + v->hidden_env = env; 1.384 + if ( READ_COUNTER( rate ) ) 1.385 + goto exit_env; 1.386 + v->env = env; 1.387 + goto exit_env; 1.388 + } 1.389 + else if ( v->env_mode == env_decay ) 1.390 + { 1.391 + env--; 1.392 + env -= env >> 8; 1.393 + rate = (adsr0 >> 3 & 0x0E) + 0x10; 1.394 + } 1.395 + else // env_attack 1.396 + { 1.397 + rate = (adsr0 & 0x0F) * 2 + 1; 1.398 + env += rate < 31 ? 0x20 : 0x400; 1.399 + } 1.400 + } 1.401 + else // GAIN 1.402 + { 1.403 + int mode; 1.404 + env_data = VREG(v_regs,gain); 1.405 + mode = env_data >> 5; 1.406 + if ( mode < 4 ) // direct 1.407 + { 1.408 + env = env_data * 0x10; 1.409 + rate = 31; 1.410 + } 1.411 + else 1.412 + { 1.413 + rate = env_data & 0x1F; 1.414 + if ( mode == 4 ) // 4: linear decrease 1.415 + { 1.416 + env -= 0x20; 1.417 + } 1.418 + else if ( mode < 6 ) // 5: exponential decrease 1.419 + { 1.420 + env--; 1.421 + env -= env >> 8; 1.422 + } 1.423 + else // 6,7: linear increase 1.424 + { 1.425 + env += 0x20; 1.426 + if ( mode > 6 && (unsigned) v->hidden_env >= 0x600 ) 1.427 + env += 0x8 - 0x20; // 7: two-slope linear increase 1.428 + } 1.429 + } 1.430 + } 1.431 + 1.432 + // Sustain level 1.433 + if ( (env >> 8) == (env_data >> 5) && v->env_mode == env_decay ) 1.434 + v->env_mode = env_sustain; 1.435 + 1.436 + v->hidden_env = env; 1.437 + 1.438 + // unsigned cast because linear decrease going negative also triggers this 1.439 + if ( (unsigned) env > 0x7FF ) 1.440 + { 1.441 + env = (env < 0 ? 0 : 0x7FF); 1.442 + if ( v->env_mode == env_attack ) 1.443 + v->env_mode = env_decay; 1.444 + } 1.445 + 1.446 + if ( !READ_COUNTER( rate ) ) 1.447 + v->env = env; // nothing else is controlled by the counter 1.448 + } 1.449 + } 1.450 + exit_env: 1.451 + 1.452 + { 1.453 + // Apply pitch 1.454 + int old_pos = v->interp_pos; 1.455 + int interp_pos = (old_pos & 0x3FFF) + pitch; 1.456 + if ( interp_pos > 0x7FFF ) 1.457 + interp_pos = 0x7FFF; 1.458 + v->interp_pos = interp_pos; 1.459 + 1.460 + // BRR decode if necessary 1.461 + if ( old_pos >= 0x4000 ) 1.462 + { 1.463 + // Arrange the four input nybbles in 0xABCD order for easy decoding 1.464 + int nybbles = ram [(v->brr_addr + v->brr_offset) & 0xFFFF] * 0x100 + 1.465 + ram [(v->brr_addr + v->brr_offset + 1) & 0xFFFF]; 1.466 + 1.467 + // Advance read position 1.468 + int const brr_block_size = 9; 1.469 + int brr_offset = v->brr_offset; 1.470 + if ( (brr_offset += 2) >= brr_block_size ) 1.471 + { 1.472 + // Next BRR block 1.473 + int brr_addr = (v->brr_addr + brr_block_size) & 0xFFFF; 1.474 + assert( brr_offset == brr_block_size ); 1.475 + if ( brr_header & 1 ) 1.476 + { 1.477 + brr_addr = SAMPLE_PTR( 1 ); 1.478 + if ( !v->kon_delay ) 1.479 + REG(endx) |= vbit; 1.480 + } 1.481 + v->brr_addr = brr_addr; 1.482 + brr_offset = 1; 1.483 + } 1.484 + v->brr_offset = brr_offset; 1.485 + 1.486 + // Decode 1.487 + 1.488 + // 0: >>1 1: <<0 2: <<1 ... 12: <<11 13-15: >>4 <<11 1.489 + static unsigned char const shifts [16 * 2] = { 1.490 + 13,12,12,12,12,12,12,12,12,12,12, 12, 12, 16, 16, 16, 1.491 + 0, 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 11, 11, 11 1.492 + }; 1.493 + int const scale = brr_header >> 4; 1.494 + int const right_shift = shifts [scale]; 1.495 + int const left_shift = shifts [scale + 16]; 1.496 + 1.497 + // Write to next four samples in circular buffer 1.498 + int* pos = v->buf_pos; 1.499 + int* end; 1.500 + 1.501 + // Decode four samples 1.502 + for ( end = pos + 4; pos < end; pos++, nybbles <<= 4 ) 1.503 + { 1.504 + // Extract upper nybble and scale appropriately 1.505 + int s = ((int16_t) nybbles >> right_shift) << left_shift; 1.506 + 1.507 + // Apply IIR filter (8 is the most commonly used) 1.508 + int const filter = brr_header & 0x0C; 1.509 + int const p1 = pos [brr_buf_size - 1]; 1.510 + int const p2 = pos [brr_buf_size - 2] >> 1; 1.511 + if ( filter >= 8 ) 1.512 + { 1.513 + s += p1; 1.514 + s -= p2; 1.515 + if ( filter == 8 ) // s += p1 * 0.953125 - p2 * 0.46875 1.516 + { 1.517 + s += p2 >> 4; 1.518 + s += (p1 * -3) >> 6; 1.519 + } 1.520 + else // s += p1 * 0.8984375 - p2 * 0.40625 1.521 + { 1.522 + s += (p1 * -13) >> 7; 1.523 + s += (p2 * 3) >> 4; 1.524 + } 1.525 + } 1.526 + else if ( filter ) // s += p1 * 0.46875 1.527 + { 1.528 + s += p1 >> 1; 1.529 + s += (-p1) >> 5; 1.530 + } 1.531 + 1.532 + // Adjust and write sample 1.533 + CLAMP16( s ); 1.534 + s = (int16_t) (s * 2); 1.535 + pos [brr_buf_size] = pos [0] = s; // second copy simplifies wrap-around 1.536 + } 1.537 + 1.538 + if ( pos >= &v->buf [brr_buf_size] ) 1.539 + pos = v->buf; 1.540 + v->buf_pos = pos; 1.541 + } 1.542 + } 1.543 +skip_brr: 1.544 + // Next voice 1.545 + vbit <<= 1; 1.546 + v_regs += 0x10; 1.547 + v++; 1.548 + } 1.549 + while ( vbit < 0x100 ); 1.550 + 1.551 + // Echo position 1.552 + int echo_offset = m.echo_offset; 1.553 + uint8_t* const echo_ptr = &ram [(REG(esa) * 0x100 + echo_offset) & 0xFFFF]; 1.554 + if ( !echo_offset ) 1.555 + m.echo_length = (REG(edl) & 0x0F) * 0x800; 1.556 + echo_offset += 4; 1.557 + if ( echo_offset >= m.echo_length ) 1.558 + echo_offset = 0; 1.559 + m.echo_offset = echo_offset; 1.560 + 1.561 + // FIR 1.562 + int echo_in_l = GET_LE16SA( echo_ptr + 0 ); 1.563 + int echo_in_r = GET_LE16SA( echo_ptr + 2 ); 1.564 + 1.565 + int (*echo_hist_pos) [2] = m.echo_hist_pos; 1.566 + if ( ++echo_hist_pos >= &m.echo_hist [echo_hist_size] ) 1.567 + echo_hist_pos = m.echo_hist; 1.568 + m.echo_hist_pos = echo_hist_pos; 1.569 + 1.570 + echo_hist_pos [0] [0] = echo_hist_pos [8] [0] = echo_in_l; 1.571 + echo_hist_pos [0] [1] = echo_hist_pos [8] [1] = echo_in_r; 1.572 + 1.573 + #define CALC_FIR_( i, in ) ((in) * (int8_t) REG(fir + i * 0x10)) 1.574 + echo_in_l = CALC_FIR_( 7, echo_in_l ); 1.575 + echo_in_r = CALC_FIR_( 7, echo_in_r ); 1.576 + 1.577 + #define CALC_FIR( i, ch ) CALC_FIR_( i, echo_hist_pos [i + 1] [ch] ) 1.578 + #define DO_FIR( i )\ 1.579 + echo_in_l += CALC_FIR( i, 0 );\ 1.580 + echo_in_r += CALC_FIR( i, 1 ); 1.581 + DO_FIR( 0 ); 1.582 + DO_FIR( 1 ); 1.583 + DO_FIR( 2 ); 1.584 + #if defined (__MWERKS__) && __MWERKS__ < 0x3200 1.585 + __eieio(); // keeps compiler from stupidly "caching" things in memory 1.586 + #endif 1.587 + DO_FIR( 3 ); 1.588 + DO_FIR( 4 ); 1.589 + DO_FIR( 5 ); 1.590 + DO_FIR( 6 ); 1.591 + 1.592 + // Echo out 1.593 + if ( !(REG(flg) & 0x20) ) 1.594 + { 1.595 + int l = (echo_out_l >> 7) + ((echo_in_l * (int8_t) REG(efb)) >> 14); 1.596 + int r = (echo_out_r >> 7) + ((echo_in_r * (int8_t) REG(efb)) >> 14); 1.597 + 1.598 + // just to help pass more validation tests 1.599 + #if SPC_MORE_ACCURACY 1.600 + l &= ~1; 1.601 + r &= ~1; 1.602 + #endif 1.603 + 1.604 + CLAMP16( l ); 1.605 + CLAMP16( r ); 1.606 + 1.607 + SET_LE16A( echo_ptr + 0, l ); 1.608 + SET_LE16A( echo_ptr + 2, r ); 1.609 + } 1.610 + 1.611 + // Sound out 1.612 + int l = (main_out_l * mvoll + echo_in_l * (int8_t) REG(evoll)) >> 14; 1.613 + int r = (main_out_r * mvolr + echo_in_r * (int8_t) REG(evolr)) >> 14; 1.614 + 1.615 + CLAMP16( l ); 1.616 + CLAMP16( r ); 1.617 + 1.618 + if ( (REG(flg) & 0x40) ) 1.619 + { 1.620 + l = 0; 1.621 + r = 0; 1.622 + } 1.623 + 1.624 + sample_t* out = m.out; 1.625 + WRITE_SAMPLES( l, r, out ); 1.626 + m.out = out; 1.627 + } 1.628 + while ( --count ); 1.629 +} 1.630 + 1.631 + 1.632 +//// Setup 1.633 + 1.634 +void SPC_DSP::mute_voices( int mask ) 1.635 +{ 1.636 + m.mute_mask = mask; 1.637 + for ( int i = 0; i < voice_count; i++ ) 1.638 + { 1.639 + m.voices [i].enabled = (mask >> i & 1) - 1; 1.640 + update_voice_vol( i * 0x10 ); 1.641 + } 1.642 +} 1.643 + 1.644 +void SPC_DSP::init( void* ram_64k ) 1.645 +{ 1.646 + m.ram = (uint8_t*) ram_64k; 1.647 + mute_voices( 0 ); 1.648 + disable_surround( false ); 1.649 + set_output( 0, 0 ); 1.650 + reset(); 1.651 + 1.652 + #ifndef NDEBUG 1.653 + // be sure this sign-extends 1.654 + assert( (int16_t) 0x8000 == -0x8000 ); 1.655 + 1.656 + // be sure right shift preserves sign 1.657 + assert( (-1 >> 1) == -1 ); 1.658 + 1.659 + // check clamp macro 1.660 + int i; 1.661 + i = +0x8000; CLAMP16( i ); assert( i == +0x7FFF ); 1.662 + i = -0x8001; CLAMP16( i ); assert( i == -0x8000 ); 1.663 + 1.664 + blargg_verify_byte_order(); 1.665 + #endif 1.666 +} 1.667 + 1.668 +void SPC_DSP::soft_reset_common() 1.669 +{ 1.670 + require( m.ram ); // init() must have been called already 1.671 + 1.672 + m.noise = 0x4000; 1.673 + m.echo_hist_pos = m.echo_hist; 1.674 + m.every_other_sample = 1; 1.675 + m.echo_offset = 0; 1.676 + m.phase = 0; 1.677 + 1.678 + init_counter(); 1.679 +} 1.680 + 1.681 +void SPC_DSP::soft_reset() 1.682 +{ 1.683 + REG(flg) = 0xE0; 1.684 + soft_reset_common(); 1.685 +} 1.686 + 1.687 +void SPC_DSP::load( uint8_t const regs [register_count] ) 1.688 +{ 1.689 + memcpy( m.regs, regs, sizeof m.regs ); 1.690 + memset( &m.regs [register_count], 0, offsetof (state_t,ram) - register_count ); 1.691 + 1.692 + // Internal state 1.693 + int i; 1.694 + for ( i = voice_count; --i >= 0; ) 1.695 + { 1.696 + voice_t& v = m.voices [i]; 1.697 + v.brr_offset = 1; 1.698 + v.buf_pos = v.buf; 1.699 + } 1.700 + m.new_kon = REG(kon); 1.701 + 1.702 + mute_voices( m.mute_mask ); 1.703 + soft_reset_common(); 1.704 +} 1.705 + 1.706 +void SPC_DSP::reset() { load( initial_regs ); }