| 1 |
/*************************************************************************** |
| 2 |
* * |
| 3 |
* LinuxSampler - modular, streaming capable sampler * |
| 4 |
* * |
| 5 |
* Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck * |
| 6 |
* Copyright (C) 2005 - 2019 Christian Schoenebeck * |
| 7 |
* * |
| 8 |
* This program is free software; you can redistribute it and/or modify * |
| 9 |
* it under the terms of the GNU General Public License as published by * |
| 10 |
* the Free Software Foundation; either version 2 of the License, or * |
| 11 |
* (at your option) any later version. * |
| 12 |
* * |
| 13 |
* This program is distributed in the hope that it will be useful, * |
| 14 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of * |
| 15 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * |
| 16 |
* GNU General Public License for more details. * |
| 17 |
* * |
| 18 |
* You should have received a copy of the GNU General Public License * |
| 19 |
* along with this program; if not, write to the Free Software * |
| 20 |
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, * |
| 21 |
* MA 02111-1307 USA * |
| 22 |
***************************************************************************/ |
| 23 |
|
| 24 |
// Note: the assembly code is currently disabled, as it doesn't fit into |
| 25 |
// the new synthesis core introduced by LS 0.4.0 |
| 26 |
|
| 27 |
#ifndef __LS_RESAMPLER_H__ |
| 28 |
#define __LS_RESAMPLER_H__ |
| 29 |
|
| 30 |
#include "../../common/global_private.h" |
| 31 |
|
| 32 |
#ifndef USE_LINEAR_INTERPOLATION |
| 33 |
# define USE_LINEAR_INTERPOLATION 0 ///< we set this to 0 by default for using cubic interpolation instead (slightly slower, but much better audio quality than linear interpolation) |
| 34 |
#endif |
| 35 |
#if USE_LINEAR_INTERPOLATION |
| 36 |
# warning Linear interpolation was configured as resampling algorithm, which is deprecated! |
| 37 |
#endif |
| 38 |
|
| 39 |
namespace LinuxSampler { |
| 40 |
|
| 41 |
/** @brief Stereo sample point |
| 42 |
* |
| 43 |
* Encapsulates one stereo sample point, thus signal value for one |
| 44 |
* sample point for left and right channel. |
| 45 |
*/ |
| 46 |
struct stereo_sample_t { |
| 47 |
float left; |
| 48 |
float right; |
| 49 |
}; |
| 50 |
|
| 51 |
/** @brief Resampler Template |
| 52 |
* |
| 53 |
* This template provides pure C++ and MMX/SSE assembly implementations |
| 54 |
* for linear and cubic interpolation for pitching a mono or stereo |
| 55 |
* input signal. |
| 56 |
*/ |
| 57 |
template<bool INTERPOLATE,bool BITDEPTH24> |
| 58 |
class Resampler { |
| 59 |
public: |
| 60 |
inline static float GetNextSampleMonoCPP(sample_t* __restrict pSrc, double* __restrict Pos, float& Pitch) { |
| 61 |
if (INTERPOLATE) return Interpolate1StepMonoCPP(pSrc, Pos, Pitch); |
| 62 |
else { // no pitch, so no interpolation necessary |
| 63 |
int pos_int = (int) *Pos; |
| 64 |
*Pos += 1.0; |
| 65 |
return pSrc [pos_int]; |
| 66 |
} |
| 67 |
} |
| 68 |
|
| 69 |
inline static stereo_sample_t GetNextSampleStereoCPP(sample_t* __restrict pSrc, double* __restrict Pos, float& Pitch) { |
| 70 |
if (INTERPOLATE) return Interpolate1StepStereoCPP(pSrc, Pos, Pitch); |
| 71 |
else { // no pitch, so no interpolation necessary |
| 72 |
int pos_int = (int) *Pos; |
| 73 |
pos_int <<= 1; |
| 74 |
*Pos += 1.0; |
| 75 |
stereo_sample_t samplePoint; |
| 76 |
samplePoint.left = pSrc[pos_int]; |
| 77 |
samplePoint.right = pSrc[pos_int+1]; |
| 78 |
return samplePoint; |
| 79 |
} |
| 80 |
} |
| 81 |
|
| 82 |
#if 0 // CONFIG_ASM && ARCH_X86 |
| 83 |
inline static void GetNext4SamplesMonoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) { |
| 84 |
if (INTERPOLATE) Interpolate4StepsMonoMMXSSE(pSrc, Pos, Pitch); |
| 85 |
else { // no pitch, so no interpolation necessary |
| 86 |
const float __4f = 4.0f; |
| 87 |
__asm__ __volatile__ ( |
| 88 |
"movss (%1), %%xmm5 # load Pos\n\t" |
| 89 |
"cvtss2si %%xmm5, %%edi # int(Pos)\n\t" |
| 90 |
"addss %2, %%xmm5 # Pos += 4.0f\n\t" |
| 91 |
"movswl (%0,%%edi,2), %%eax # load sample 0\n\t" |
| 92 |
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 93 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 94 |
"movswl 2(%0,%%edi,2), %%edx # load sample 1\n\t" |
| 95 |
"cvtsi2ss %%edx, %%xmm2 # convert to float\n\t" |
| 96 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 97 |
"movss %%xmm5, (%1) # update Pos\n\t" |
| 98 |
"movswl 4(%0,%%edi,2), %%eax # load sample 2\n\t" |
| 99 |
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 100 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 101 |
"movswl 6(%0,%%edi,2), %%edx # load sample 3\n\t" |
| 102 |
"cvtsi2ss %%edx, %%xmm2 # convert to float\n\t" |
| 103 |
"shufps $0x1b, %%xmm2, %%xmm2 # swap to correct order\n\t" |
| 104 |
:: "r" (pSrc), "r" (Pos), "m" (__4f) |
| 105 |
: "%eax", "%edx", "%edi" |
| 106 |
); |
| 107 |
} |
| 108 |
} |
| 109 |
|
| 110 |
inline static void GetNext4SamplesStereoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) { |
| 111 |
if (INTERPOLATE) { |
| 112 |
Interpolate4StepsStereoMMXSSE(pSrc, Pos, Pitch); |
| 113 |
//EMMS; |
| 114 |
} else { // no pitch, so no interpolation necessary |
| 115 |
const float __4f = 4.0f; |
| 116 |
__asm__ __volatile__ ( |
| 117 |
"movss (%1), %%xmm5 # load Pos\n\t" |
| 118 |
"cvtss2si %%xmm5, %%edi # int(Pos)\n\t" |
| 119 |
"addss %2, %%xmm5 # Pos += 4.0f\n\t" |
| 120 |
"movswl (%0, %%edi,4), %%eax # load sample 0 (left)\n\t" |
| 121 |
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 122 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 123 |
"movss %%xmm5, (%1) # update Pos\n\t" |
| 124 |
"movswl 2(%0, %%edi,4), %%edx # load sample 0 (left)\n\t" |
| 125 |
"cvtsi2ss %%edx, %%xmm3 # convert to float\n\t" |
| 126 |
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 127 |
"movswl 4(%0, %%edi,4), %%eax # load sample 1 (left)\n\t" |
| 128 |
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 129 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 130 |
"movswl 6(%0, %%edi,4), %%edx # load sample 1 (right)\n\t" |
| 131 |
"cvtsi2ss %%edx, %%xmm3 # convert to float\n\t" |
| 132 |
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 133 |
"movswl 8(%0, %%edi,4), %%eax # load sample 2 (left)\n\t" |
| 134 |
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 135 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 136 |
"movswl 10(%0, %%edi,4), %%edx # load sample 2 (right)\n\t" |
| 137 |
"cvtsi2ss %%edx, %%xmm3 # convert to float\n\t" |
| 138 |
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 139 |
"movswl 12(%0, %%edi,4), %%eax # load sample 3 (left)\n\t" |
| 140 |
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 141 |
"shufps $0x1b, %%xmm2, %%xmm2 # swap to correct order\n\t" |
| 142 |
"movswl 14(%0, %%edi,4), %%edx # load sample 3 (right)\n\t" |
| 143 |
"cvtsi2ss %%edx, %%xmm3 # convert to float\n\t" |
| 144 |
"shufps $0x1b, %%xmm3, %%xmm3 # swap to correct order\n\t" |
| 145 |
:: "r" (pSrc), "r" (Pos), "m" (__4f) |
| 146 |
: "%eax", "%edx", "%edi" |
| 147 |
); |
| 148 |
} |
| 149 |
} |
| 150 |
#endif // CONFIG_ASM && ARCH_X86 |
| 151 |
|
| 152 |
protected: |
| 153 |
|
| 154 |
inline static int32_t getSample(sample_t* __restrict src, int pos) { |
| 155 |
if (BITDEPTH24) { |
| 156 |
pos *= 3; |
| 157 |
#if WORDS_BIGENDIAN |
| 158 |
unsigned char* p = (unsigned char*)src; |
| 159 |
return p[pos] << 8 | p[pos + 1] << 16 | p[pos + 2] << 24; |
| 160 |
#else |
| 161 |
// 24bit read optimization: |
| 162 |
// a misaligned 32bit read and subquent 8 bit shift is faster (on x86) than reading 3 single bytes and shifting them |
| 163 |
return (*((int32_t *)(&((char *)(src))[pos])))<<8; |
| 164 |
#endif |
| 165 |
} else { |
| 166 |
return src[pos]; |
| 167 |
} |
| 168 |
} |
| 169 |
|
| 170 |
inline static float Interpolate1StepMonoCPP(sample_t* __restrict pSrc, double* __restrict Pos, float& Pitch) { |
| 171 |
int pos_int = (int) *Pos; // integer position |
| 172 |
float pos_fract = *Pos - pos_int; // fractional part of position |
| 173 |
|
| 174 |
#if USE_LINEAR_INTERPOLATION |
| 175 |
int x1 = getSample(pSrc, pos_int); |
| 176 |
int x2 = getSample(pSrc, pos_int + 1); |
| 177 |
float samplePoint = (x1 + pos_fract * (x2 - x1)); |
| 178 |
#else // polynomial interpolation |
| 179 |
float xm1 = getSample(pSrc, pos_int); |
| 180 |
float x0 = getSample(pSrc, pos_int + 1); |
| 181 |
float x1 = getSample(pSrc, pos_int + 2); |
| 182 |
float x2 = getSample(pSrc, pos_int + 3); |
| 183 |
float a = (3.0f * (x0 - x1) - xm1 + x2) * 0.5f; |
| 184 |
float b = 2.0f * x1 + xm1 - (5.0f * x0 + x2) * 0.5f; |
| 185 |
float c = (x1 - xm1) * 0.5f; |
| 186 |
float samplePoint = (((a * pos_fract) + b) * pos_fract + c) * pos_fract + x0; |
| 187 |
#endif // USE_LINEAR_INTERPOLATION |
| 188 |
|
| 189 |
*Pos += Pitch; |
| 190 |
return samplePoint; |
| 191 |
} |
| 192 |
|
| 193 |
inline static stereo_sample_t Interpolate1StepStereoCPP(sample_t* __restrict pSrc, double* __restrict Pos, float& Pitch) { |
| 194 |
int pos_int = (int) *Pos; // integer position |
| 195 |
float pos_fract = *Pos - pos_int; // fractional part of position |
| 196 |
pos_int <<= 1; |
| 197 |
|
| 198 |
stereo_sample_t samplePoint; |
| 199 |
|
| 200 |
#if USE_LINEAR_INTERPOLATION |
| 201 |
// left channel |
| 202 |
int x1 = getSample(pSrc, pos_int); |
| 203 |
int x2 = getSample(pSrc, pos_int + 2); |
| 204 |
samplePoint.left = (x1 + pos_fract * (x2 - x1)); |
| 205 |
// right channel |
| 206 |
x1 = getSample(pSrc, pos_int + 1); |
| 207 |
x2 = getSample(pSrc, pos_int + 3); |
| 208 |
samplePoint.right = (x1 + pos_fract * (x2 - x1)); |
| 209 |
#else // polynomial interpolation |
| 210 |
// calculate left channel |
| 211 |
float xm1 = getSample(pSrc, pos_int); |
| 212 |
float x0 = getSample(pSrc, pos_int + 2); |
| 213 |
float x1 = getSample(pSrc, pos_int + 4); |
| 214 |
float x2 = getSample(pSrc, pos_int + 6); |
| 215 |
float a = (3.0f * (x0 - x1) - xm1 + x2) * 0.5f; |
| 216 |
float b = 2.0f * x1 + xm1 - (5.0f * x0 + x2) * 0.5f; |
| 217 |
float c = (x1 - xm1) * 0.5f; |
| 218 |
samplePoint.left = (((a * pos_fract) + b) * pos_fract + c) * pos_fract + x0; |
| 219 |
|
| 220 |
//calculate right channel |
| 221 |
xm1 = getSample(pSrc, pos_int + 1); |
| 222 |
x0 = getSample(pSrc, pos_int + 3); |
| 223 |
x1 = getSample(pSrc, pos_int + 5); |
| 224 |
x2 = getSample(pSrc, pos_int + 7); |
| 225 |
a = (3.0f * (x0 - x1) - xm1 + x2) * 0.5f; |
| 226 |
b = 2.0f * x1 + xm1 - (5.0f * x0 + x2) * 0.5f; |
| 227 |
c = (x1 - xm1) * 0.5f; |
| 228 |
samplePoint.right = (((a * pos_fract) + b) * pos_fract + c) * pos_fract + x0; |
| 229 |
#endif // USE_LINEAR_INTERPOLATION |
| 230 |
|
| 231 |
*Pos += Pitch; |
| 232 |
return samplePoint; |
| 233 |
} |
| 234 |
|
| 235 |
#if 0 // CONFIG_ASM && ARCH_X86 |
| 236 |
// TODO: no support for cubic interpolation yet |
| 237 |
inline static void Interpolate4StepsMonoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) { |
| 238 |
/* calculate playback position of each of the 4 samples by adding the associated pitch */ |
| 239 |
__asm__ __volatile__ ( |
| 240 |
"movss (%0),%%xmm0 # sample position of sample[0] -> xmm0[0]\n\t" |
| 241 |
"movss %1,%%xmm1 # copy pitch -> xmm1[0]\n\t" |
| 242 |
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 243 |
"addss %%xmm1,%%xmm0 # calculate sample position of sample[1]\n\t" |
| 244 |
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 245 |
"addss %%xmm1,%%xmm0 # calculate sample position of sample[2]\n\t" |
| 246 |
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 247 |
"addss %%xmm1,%%xmm0 # calculate sample position of sample[3]\n\t" |
| 248 |
"movss %%xmm0,%%xmm2 # xmm0[0] -> xmm2[0]\n\t" |
| 249 |
"addss %%xmm1,%%xmm2 # calculate initial sample position for the next 4-sample cycle\n\t" |
| 250 |
"movss %%xmm2,(%0) # update 'Pos'\n\t" |
| 251 |
"shufps $0x1b,%%xmm0,%%xmm0 # swap, so that xmm0[0]=sample pos 0, xmm0[1]=sample pos 1,...\n\t" |
| 252 |
"cvttps2pi %%xmm0,%%mm4 # int(xmm0[0-1]) -> mm4\n\t" |
| 253 |
"shufps $0xe4,%%xmm0,%%xmm1 # xmm0[2-3] -> xmm1[2-3]\n\t" |
| 254 |
"shufps $0x0e,%%xmm1,%%xmm1 # xmm1[2-3] -> xmm1[0-1]\n\t" |
| 255 |
"cvttps2pi %%xmm1,%%mm5 # int(xmm1[0-1]) -> mm5\n\t" |
| 256 |
"cvtpi2ps %%mm5,%%xmm1 # double(mm5) -> xmm1[0-1]\n\t" |
| 257 |
"shufps $0x44,%%xmm1,%%xmm1 # shift lower 2 FPs up to the upper 2 cells\n\t" |
| 258 |
"cvtpi2ps %%mm4,%%xmm1 # double(mm4) -> xmm1[0-1]\n\t" |
| 259 |
"subps %%xmm1,%%xmm0 # xmm0[1-3] = xmm0[1-3] - xmm1[1-3]\n\t" |
| 260 |
: |
| 261 |
: "r" (Pos), /* %0 */ |
| 262 |
"m" (Pitch) /* %1 */ |
| 263 |
: "%xmm0", /* holds fractional position (0.0 <= x < 1.0) of sample 0-3 at the end */ |
| 264 |
"%xmm1", /* holds integer position (back converted to SPFP) of sample 0-3 at the end */ |
| 265 |
"mm4", /* holds integer position of sample 0-1 at the end */ |
| 266 |
"mm5", /* holds integer position of sample 2-3 at the end */ |
| 267 |
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)" |
| 268 |
); |
| 269 |
/* get sample values of pSrc[pos_int] and pSrc[pos_int+1] of the 4 samples */ |
| 270 |
__asm__ __volatile__ ( |
| 271 |
"movd %%mm4,%%edi # sample position of sample 0\n\t" |
| 272 |
"psrlq $32,%%mm4 # mm4 >> 32\n\t" |
| 273 |
"movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 0)\n\t" |
| 274 |
"movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 0+1)\n\t" |
| 275 |
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 276 |
"cvtsi2ss %%ecx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 277 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 278 |
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 279 |
"movd %%mm4,%%edi # sample position of sample 1\n\t" |
| 280 |
"movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 1)\n\t" |
| 281 |
"movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 1+1)\n\t" |
| 282 |
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 283 |
"cvtsi2ss %%ecx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 284 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 285 |
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 286 |
"movd %%mm5,%%edi # sample position of sample 2\n\t" |
| 287 |
"psrlq $32,%%mm5 # mm5 >> 32\n\t" |
| 288 |
"movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 2)\n\t" |
| 289 |
"movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 2+1)\n\t" |
| 290 |
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 291 |
"cvtsi2ss %%ecx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 292 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 293 |
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 294 |
"movd %%mm5,%%edi # sample position of sample 2\n\t" |
| 295 |
"movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 3)\n\t" |
| 296 |
"movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 3+1)\n\t" |
| 297 |
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 298 |
"cvtsi2ss %%ecx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 299 |
"shufps $0x1b, %%xmm2, %%xmm2 # swap to correct order\n\t" |
| 300 |
"shufps $0x1b, %%xmm3, %%xmm3 # swap to correct order\n\t" |
| 301 |
: /* no output */ |
| 302 |
: "S" (pSrc) /* %0 - sample read position */ |
| 303 |
: "%eax", "%ecx", /*"%edx",*/ "%edi", |
| 304 |
"%xmm2", /* holds pSrc[int_pos] of the 4 samples at the end */ |
| 305 |
"%xmm3", /* holds pSrc[int_pos+1] of the 4 samples at the end */ |
| 306 |
"mm4", /* holds integer position of sample 0-1 at the end */ |
| 307 |
"mm5", /* holds integer position of sample 2-3 at the end */ |
| 308 |
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)" |
| 309 |
); |
| 310 |
/* linear interpolation of the 4 samples simultaniously */ |
| 311 |
__asm__ __volatile__ ( |
| 312 |
"subps %%xmm2,%%xmm3 # xmm3 = pSrc[pos_int+1] - pSrc[pos_int]\n\t" |
| 313 |
"mulps %%xmm0,%%xmm3 # xmm3 = pos_fract * (pSrc[pos_int+1] - pSrc[pos_int])\n\t" |
| 314 |
"addps %%xmm3,%%xmm2 # xmm2 = pSrc[pos_int] + (pos_fract * (pSrc[pos_int+1] - pSrc[pos_int]))\n\t" |
| 315 |
: /* no output */ |
| 316 |
: /* no input */ |
| 317 |
: "%xmm2" /* holds linear interpolated sample point (of all 4 samples) at the end */ |
| 318 |
); |
| 319 |
} |
| 320 |
|
| 321 |
// TODO: no support for cubic interpolation yet |
| 322 |
inline static void Interpolate4StepsStereoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) { |
| 323 |
/* calculate playback position of each of the 4 samples by adding the associated pitch */ |
| 324 |
__asm__ __volatile__ ( |
| 325 |
"movss (%0),%%xmm0 # sample position of sample[0] -> xmm0[0]\n\t" |
| 326 |
"movss %1,%%xmm1 # copy pitch -> xmm1[0]\n\t" |
| 327 |
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 328 |
"addss %%xmm1,%%xmm0 # calculate sample position of sample[1]\n\t" |
| 329 |
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 330 |
"addss %%xmm1,%%xmm0 # calculate sample position of sample[2]\n\t" |
| 331 |
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 332 |
"addss %%xmm1,%%xmm0 # calculate sample position of sample[3]\n\t" |
| 333 |
"movss %%xmm0,%%xmm2 # xmm0[0] -> xmm2[0]\n\t" |
| 334 |
"addss %%xmm1,%%xmm2 # calculate initial sample position for the next 4-sample cycle\n\t" |
| 335 |
"movss %%xmm2,(%0) # update 'Pos'\n\t" |
| 336 |
"shufps $0x1b,%%xmm0,%%xmm0 # swap, so that xmm0[0]=sample pos 0, xmm0[1]=sample pos 1,...\n\t" |
| 337 |
"cvttps2pi %%xmm0,%%mm4 # int(xmm0[0-1]) -> mm4\n\t" |
| 338 |
"shufps $0xe4,%%xmm0,%%xmm1 # xmm0[2-3] -> xmm1[2-3]\n\t" |
| 339 |
"shufps $0x0e,%%xmm1,%%xmm1 # xmm1[2-3] -> xmm1[0-1]\n\t" |
| 340 |
"cvttps2pi %%xmm1,%%mm5 # int(xmm1[0-1]) -> mm5\n\t" |
| 341 |
"cvtpi2ps %%mm5,%%xmm1 # double(mm5) -> xmm1[0-1]\n\t" |
| 342 |
"shufps $0x44,%%xmm1,%%xmm1 # shift lower 2 FPs up to the upper 2 cells\n\t" |
| 343 |
"cvtpi2ps %%mm4,%%xmm1 # double(mm4) -> xmm1[0-1]\n\t" |
| 344 |
"subps %%xmm1,%%xmm0 # xmm0[1-3] = xmm0[1-3] - xmm1[1-3]\n\t" |
| 345 |
: |
| 346 |
: "r" (Pos), /* %0 */ |
| 347 |
"m" (Pitch) /* %1 */ |
| 348 |
: "%xmm0", /* holds fractional position (0.0 <= x < 1.0) of sample 0-3 at the end */ |
| 349 |
"%xmm1", /* holds integer position (back converted to SPFP) of sample 0-3 at the end */ |
| 350 |
"mm4", /* holds integer position of sample 0-1 at the end */ |
| 351 |
"mm5", /* holds integer position of sample 2-3 at the end */ |
| 352 |
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)" |
| 353 |
); |
| 354 |
|
| 355 |
/* get sample values of pSrc[pos_int], pSrc[pos_int+1], pSrc[pos_int+2] and pSrc[pos_int+3] of the 4 samples */ |
| 356 |
__asm__ __volatile__ ( |
| 357 |
"xorl %%eax,%%eax # clear eax\n\t" |
| 358 |
"xorl %%edx,%%edx # clear edx\n\t" |
| 359 |
"movd %%mm4,%%edi # sample position of sample 0\n\t" |
| 360 |
"psrlq $32,%%mm4 # mm4 >> 32\n\t" |
| 361 |
"movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 0)\n\t" |
| 362 |
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 363 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 364 |
"movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 0+1)\n\t" |
| 365 |
"cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 366 |
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 367 |
"movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 0+2)\n\t" |
| 368 |
"cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t" |
| 369 |
"shufps $0x93, %%xmm4, %%xmm4 # shift up\n\t" |
| 370 |
"movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 0+3)\n\t" |
| 371 |
"cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t" |
| 372 |
"movd %%mm4,%%edi # sample position of sample 1\n\t" |
| 373 |
"shufps $0x93, %%xmm5, %%xmm5 # shift up\n\t" |
| 374 |
"movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 1)\n\t" |
| 375 |
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 376 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 377 |
"movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 1+1)\n\t" |
| 378 |
"cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 379 |
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 380 |
"movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 1+2)\n\t" |
| 381 |
"cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t" |
| 382 |
"shufps $0x93, %%xmm4, %%xmm4 # shift up\n\t" |
| 383 |
"movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 1+3)\n\t" |
| 384 |
"cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t" |
| 385 |
"movd %%mm5,%%edi # sample position of sample 2\n\t" |
| 386 |
"shufps $0x93, %%xmm5, %%xmm5 # shift up\n\t" |
| 387 |
"psrlq $32,%%mm5 # mm5 >> 32\n\t" |
| 388 |
"movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 2)\n\t" |
| 389 |
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 390 |
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 391 |
"movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 2+1)\n\t" |
| 392 |
"cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 393 |
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 394 |
"movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 2+2)\n\t" |
| 395 |
"cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t" |
| 396 |
"shufps $0x93, %%xmm4, %%xmm4 # shift up\n\t" |
| 397 |
"movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 2+3)\n\t" |
| 398 |
"cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t" |
| 399 |
"movd %%mm5,%%edi # sample position of sample 3\n\t" |
| 400 |
"shufps $0x93, %%xmm5, %%xmm5 # shift up\n\t" |
| 401 |
"movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 3)\n\t" |
| 402 |
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 403 |
"shufps $0x1b, %%xmm2, %%xmm2 # shift up\n\t" |
| 404 |
"movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 3+1)\n\t" |
| 405 |
"cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 406 |
"shufps $0x1b, %%xmm3, %%xmm3 # shift up\n\t" |
| 407 |
"movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 3+2)\n\t" |
| 408 |
"cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t" |
| 409 |
"shufps $0x1b, %%xmm4, %%xmm4 # swap to correct order\n\t" |
| 410 |
"movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 3+3)\n\t" |
| 411 |
"cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t" |
| 412 |
"shufps $0x1b, %%xmm5, %%xmm5 # swap to correct order\n\t" |
| 413 |
: /* no output */ |
| 414 |
: "S" (pSrc) /* %0 - sample read position */ |
| 415 |
: "%eax", "%edx", "%edi", |
| 416 |
"xmm2", /* holds pSrc[int_pos] of the 4 samples at the end */ |
| 417 |
"xmm3", /* holds pSrc[int_pos+1] of the 4 samples at the end */ |
| 418 |
"xmm4", /* holds pSrc[int_pos+2] of the 4 samples at the end */ |
| 419 |
"xmm5", /* holds pSrc[int_pos+3] of the 4 samples at the end */ |
| 420 |
"mm4", /* holds integer position of sample 0-1 at the end */ |
| 421 |
"mm5", /* holds integer position of sample 2-3 at the end */ |
| 422 |
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)" |
| 423 |
); |
| 424 |
/* linear interpolation of the 4 samples (left & right channel) simultaniously */ |
| 425 |
__asm__ __volatile__ ( |
| 426 |
"subps %%xmm2,%%xmm4 # xmm4 = pSrc[pos_int+2] - pSrc[pos_int] (left channel)\n\t" |
| 427 |
"mulps %%xmm0,%%xmm4 # xmm4 = pos_fract * (pSrc[pos_int+2] - pSrc[pos_int]) (left channel)\n\t" |
| 428 |
"addps %%xmm4,%%xmm2 # xmm2 = pSrc[pos_int] + (pos_fract * (pSrc[pos_int+2] - pSrc[pos_int])) (left channel)\n\t" |
| 429 |
"subps %%xmm3,%%xmm5 # xmm5 = pSrc[pos_int+3] - pSrc[pos_int+1] (right channel)\n\t" |
| 430 |
"mulps %%xmm0,%%xmm5 # xmm5 = pos_fract * (pSrc[pos_int+3] - pSrc[pos_int+1]) (right channel)\n\t" |
| 431 |
"addps %%xmm5,%%xmm3 # xmm3 = pSrc[pos_int+1] + (pos_fract * (pSrc[pos_int+3] - pSrc[pos_int+1])) (right channel)\n\t" |
| 432 |
: /* no output */ |
| 433 |
: /* no input */ |
| 434 |
: "%xmm2", /* holds linear interpolated sample of left channel (of all 4 samples) at the end */ |
| 435 |
"%xmm3" /* holds linear interpolated sample of right channel (of all 4 samples) at the end */ |
| 436 |
); |
| 437 |
} |
| 438 |
#endif // CONFIG_ASM && ARCH_X86 |
| 439 |
}; |
| 440 |
|
| 441 |
} // namespace LinuxSampler |
| 442 |
|
| 443 |
#endif // __LS_RESAMPLER_H__ |
Parent Directory
| 
Revision Log