| 1 |
schoenebeck |
320 |
/*************************************************************************** |
| 2 |
|
|
* * |
| 3 |
|
|
* LinuxSampler - modular, streaming capable sampler * |
| 4 |
|
|
* * |
| 5 |
|
|
* Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck * |
| 6 |
schoenebeck |
617 |
* Copyright (C) 2005 Christian Schoenebeck * |
| 7 |
schoenebeck |
320 |
* * |
| 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 |
|
|
#ifndef __LS_RESAMPLER_H__ |
| 25 |
|
|
#define __LS_RESAMPLER_H__ |
| 26 |
|
|
|
| 27 |
|
|
#include "../../common/global.h" |
| 28 |
|
|
|
| 29 |
|
|
// TODO: cubic interpolation is not yet supported by the MMX/SSE(1) version though |
| 30 |
|
|
#ifndef USE_LINEAR_INTERPOLATION |
| 31 |
|
|
# define USE_LINEAR_INTERPOLATION 1 ///< set to 0 if you prefer cubic interpolation (slower, better quality) |
| 32 |
|
|
#endif |
| 33 |
|
|
|
| 34 |
|
|
namespace LinuxSampler { |
| 35 |
|
|
|
| 36 |
schoenebeck |
563 |
/** @brief Stereo sample point |
| 37 |
|
|
* |
| 38 |
|
|
* Encapsulates one stereo sample point, thus signal value for one |
| 39 |
|
|
* sample point for left and right channel. |
| 40 |
|
|
*/ |
| 41 |
schoenebeck |
320 |
struct stereo_sample_t { |
| 42 |
|
|
float left; |
| 43 |
|
|
float right; |
| 44 |
|
|
}; |
| 45 |
|
|
|
| 46 |
schoenebeck |
563 |
/** @brief Resampler Template |
| 47 |
|
|
* |
| 48 |
|
|
* This template provides pure C++ and MMX/SSE assembly implementations |
| 49 |
|
|
* for linear and cubic interpolation for pitching a mono or stereo |
| 50 |
|
|
* input signal. |
| 51 |
|
|
*/ |
| 52 |
schoenebeck |
320 |
template<bool INTERPOLATE> |
| 53 |
|
|
class Resampler { |
| 54 |
|
|
public: |
| 55 |
|
|
inline static float GetNextSampleMonoCPP(sample_t* pSrc, double* Pos, float& Pitch) { |
| 56 |
|
|
if (INTERPOLATE) return Interpolate1StepMonoCPP(pSrc, Pos, Pitch); |
| 57 |
|
|
else { // no pitch, so no interpolation necessary |
| 58 |
|
|
int pos_int = (int) *Pos; |
| 59 |
|
|
*Pos += 1.0; |
| 60 |
|
|
return pSrc [pos_int]; |
| 61 |
|
|
} |
| 62 |
|
|
} |
| 63 |
|
|
|
| 64 |
|
|
inline static stereo_sample_t GetNextSampleStereoCPP(sample_t* pSrc, double* Pos, float& Pitch) { |
| 65 |
|
|
if (INTERPOLATE) return Interpolate1StepStereoCPP(pSrc, Pos, Pitch); |
| 66 |
|
|
else { // no pitch, so no interpolation necessary |
| 67 |
|
|
int pos_int = (int) *Pos; |
| 68 |
|
|
pos_int <<= 1; |
| 69 |
|
|
*Pos += 1.0; |
| 70 |
|
|
stereo_sample_t samplePoint; |
| 71 |
|
|
samplePoint.left = pSrc[pos_int]; |
| 72 |
|
|
samplePoint.right = pSrc[pos_int+1]; |
| 73 |
|
|
return samplePoint; |
| 74 |
|
|
} |
| 75 |
|
|
} |
| 76 |
|
|
|
| 77 |
schoenebeck |
617 |
#if CONFIG_ASM && ARCH_X86 |
| 78 |
schoenebeck |
320 |
inline static void GetNext4SamplesMonoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) { |
| 79 |
|
|
if (INTERPOLATE) Interpolate4StepsMonoMMXSSE(pSrc, Pos, Pitch); |
| 80 |
|
|
else { // no pitch, so no interpolation necessary |
| 81 |
|
|
const float __4f = 4.0f; |
| 82 |
|
|
__asm__ __volatile__ ( |
| 83 |
|
|
"movss (%1), %%xmm5 # load Pos\n\t" |
| 84 |
|
|
"cvtss2si %%xmm5, %%edi # int(Pos)\n\t" |
| 85 |
|
|
"addss %2, %%xmm5 # Pos += 4.0f\n\t" |
| 86 |
|
|
"movswl (%0,%%edi,2), %%eax # load sample 0\n\t" |
| 87 |
|
|
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 88 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 89 |
|
|
"movswl 2(%0,%%edi,2), %%edx # load sample 1\n\t" |
| 90 |
|
|
"cvtsi2ss %%edx, %%xmm2 # convert to float\n\t" |
| 91 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 92 |
|
|
"movss %%xmm5, (%1) # update Pos\n\t" |
| 93 |
|
|
"movswl 4(%0,%%edi,2), %%eax # load sample 2\n\t" |
| 94 |
|
|
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 95 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 96 |
|
|
"movswl 6(%0,%%edi,2), %%edx # load sample 3\n\t" |
| 97 |
|
|
"cvtsi2ss %%edx, %%xmm2 # convert to float\n\t" |
| 98 |
|
|
"shufps $0x1b, %%xmm2, %%xmm2 # swap to correct order\n\t" |
| 99 |
|
|
:: "r" (pSrc), "r" (Pos), "m" (__4f) |
| 100 |
|
|
: "%eax", "%edx", "%edi" |
| 101 |
|
|
); |
| 102 |
|
|
} |
| 103 |
|
|
} |
| 104 |
|
|
|
| 105 |
|
|
inline static void GetNext4SamplesStereoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) { |
| 106 |
|
|
if (INTERPOLATE) { |
| 107 |
|
|
Interpolate4StepsStereoMMXSSE(pSrc, Pos, Pitch); |
| 108 |
|
|
//EMMS; |
| 109 |
|
|
} else { // no pitch, so no interpolation necessary |
| 110 |
|
|
const float __4f = 4.0f; |
| 111 |
|
|
__asm__ __volatile__ ( |
| 112 |
|
|
"movss (%1), %%xmm5 # load Pos\n\t" |
| 113 |
|
|
"cvtss2si %%xmm5, %%edi # int(Pos)\n\t" |
| 114 |
|
|
"addss %2, %%xmm5 # Pos += 4.0f\n\t" |
| 115 |
|
|
"movswl (%0, %%edi,4), %%eax # load sample 0 (left)\n\t" |
| 116 |
|
|
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 117 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 118 |
|
|
"movss %%xmm5, (%1) # update Pos\n\t" |
| 119 |
|
|
"movswl 2(%0, %%edi,4), %%edx # load sample 0 (left)\n\t" |
| 120 |
|
|
"cvtsi2ss %%edx, %%xmm3 # convert to float\n\t" |
| 121 |
|
|
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 122 |
|
|
"movswl 4(%0, %%edi,4), %%eax # load sample 1 (left)\n\t" |
| 123 |
|
|
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 124 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 125 |
|
|
"movswl 6(%0, %%edi,4), %%edx # load sample 1 (right)\n\t" |
| 126 |
|
|
"cvtsi2ss %%edx, %%xmm3 # convert to float\n\t" |
| 127 |
|
|
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 128 |
|
|
"movswl 8(%0, %%edi,4), %%eax # load sample 2 (left)\n\t" |
| 129 |
|
|
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 130 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 131 |
|
|
"movswl 10(%0, %%edi,4), %%edx # load sample 2 (right)\n\t" |
| 132 |
|
|
"cvtsi2ss %%edx, %%xmm3 # convert to float\n\t" |
| 133 |
|
|
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 134 |
|
|
"movswl 12(%0, %%edi,4), %%eax # load sample 3 (left)\n\t" |
| 135 |
|
|
"cvtsi2ss %%eax, %%xmm2 # convert to float\n\t" |
| 136 |
|
|
"shufps $0x1b, %%xmm2, %%xmm2 # swap to correct order\n\t" |
| 137 |
|
|
"movswl 14(%0, %%edi,4), %%edx # load sample 3 (right)\n\t" |
| 138 |
|
|
"cvtsi2ss %%edx, %%xmm3 # convert to float\n\t" |
| 139 |
|
|
"shufps $0x1b, %%xmm3, %%xmm3 # swap to correct order\n\t" |
| 140 |
|
|
:: "r" (pSrc), "r" (Pos), "m" (__4f) |
| 141 |
|
|
: "%eax", "%edx", "%edi" |
| 142 |
|
|
); |
| 143 |
|
|
} |
| 144 |
|
|
} |
| 145 |
schoenebeck |
617 |
#endif // CONFIG_ASM && ARCH_X86 |
| 146 |
schoenebeck |
320 |
|
| 147 |
|
|
protected: |
| 148 |
|
|
|
| 149 |
|
|
inline static float Interpolate1StepMonoCPP(sample_t* pSrc, double* Pos, float& Pitch) { |
| 150 |
|
|
int pos_int = (int) *Pos; // integer position |
| 151 |
|
|
float pos_fract = *Pos - pos_int; // fractional part of position |
| 152 |
|
|
|
| 153 |
|
|
#if USE_LINEAR_INTERPOLATION |
| 154 |
|
|
float samplePoint = pSrc[pos_int] + pos_fract * (pSrc[pos_int+1] - pSrc[pos_int]); |
| 155 |
|
|
#else // polynomial interpolation |
| 156 |
|
|
float xm1 = pSrc[pos_int]; |
| 157 |
|
|
float x0 = pSrc[pos_int+1]; |
| 158 |
|
|
float x1 = pSrc[pos_int+2]; |
| 159 |
|
|
float x2 = pSrc[pos_int+3]; |
| 160 |
|
|
float a = (3.0f * (x0 - x1) - xm1 + x2) * 0.5f; |
| 161 |
|
|
float b = 2.0f * x1 + xm1 - (5.0f * x0 + x2) * 0.5f; |
| 162 |
|
|
float c = (x1 - xm1) * 0.5f; |
| 163 |
|
|
float samplePoint = (((a * pos_fract) + b) * pos_fract + c) * pos_fract + x0; |
| 164 |
|
|
#endif // USE_LINEAR_INTERPOLATION |
| 165 |
|
|
|
| 166 |
|
|
*Pos += Pitch; |
| 167 |
|
|
return samplePoint; |
| 168 |
|
|
} |
| 169 |
|
|
|
| 170 |
|
|
inline static stereo_sample_t Interpolate1StepStereoCPP(sample_t* pSrc, double* Pos, float& Pitch) { |
| 171 |
|
|
int pos_int = (int) *Pos; // integer position |
| 172 |
|
|
float pos_fract = *Pos - pos_int; // fractional part of position |
| 173 |
|
|
pos_int <<= 1; |
| 174 |
|
|
|
| 175 |
|
|
stereo_sample_t samplePoint; |
| 176 |
|
|
|
| 177 |
|
|
#if USE_LINEAR_INTERPOLATION |
| 178 |
|
|
// left channel |
| 179 |
|
|
samplePoint.left = pSrc[pos_int] + pos_fract * (pSrc[pos_int+2] - pSrc[pos_int]); |
| 180 |
|
|
// right channel |
| 181 |
|
|
samplePoint.right = pSrc[pos_int+1] + pos_fract * (pSrc[pos_int+3] - pSrc[pos_int+1]); |
| 182 |
|
|
#else // polynomial interpolation |
| 183 |
|
|
// calculate left channel |
| 184 |
|
|
float xm1 = pSrc[pos_int]; |
| 185 |
|
|
float x0 = pSrc[pos_int+2]; |
| 186 |
|
|
float x1 = pSrc[pos_int+4]; |
| 187 |
|
|
float x2 = pSrc[pos_int+6]; |
| 188 |
|
|
float a = (3.0f * (x0 - x1) - xm1 + x2) * 0.5f; |
| 189 |
|
|
float b = 2.0f * x1 + xm1 - (5.0f * x0 + x2) * 0.5f; |
| 190 |
|
|
float c = (x1 - xm1) * 0.5f; |
| 191 |
|
|
samplePoint.left = (((a * pos_fract) + b) * pos_fract + c) * pos_fract + x0; |
| 192 |
|
|
|
| 193 |
|
|
//calculate right channel |
| 194 |
|
|
xm1 = pSrc[pos_int+1]; |
| 195 |
|
|
x0 = pSrc[pos_int+3]; |
| 196 |
|
|
x1 = pSrc[pos_int+5]; |
| 197 |
|
|
x2 = pSrc[pos_int+7]; |
| 198 |
|
|
a = (3.0f * (x0 - x1) - xm1 + x2) * 0.5f; |
| 199 |
|
|
b = 2.0f * x1 + xm1 - (5.0f * x0 + x2) * 0.5f; |
| 200 |
|
|
c = (x1 - xm1) * 0.5f; |
| 201 |
|
|
samplePoint.right = (((a * pos_fract) + b) * pos_fract + c) * pos_fract + x0; |
| 202 |
|
|
#endif // USE_LINEAR_INTERPOLATION |
| 203 |
|
|
|
| 204 |
|
|
*Pos += Pitch; |
| 205 |
|
|
return samplePoint; |
| 206 |
|
|
} |
| 207 |
|
|
|
| 208 |
schoenebeck |
617 |
#if CONFIG_ASM && ARCH_X86 |
| 209 |
schoenebeck |
320 |
// TODO: no support for cubic interpolation yet |
| 210 |
|
|
inline static void Interpolate4StepsMonoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) { |
| 211 |
|
|
/* calculate playback position of each of the 4 samples by adding the associated pitch */ |
| 212 |
|
|
__asm__ __volatile__ ( |
| 213 |
|
|
"movss (%0),%%xmm0 # sample position of sample[0] -> xmm0[0]\n\t" |
| 214 |
|
|
"movss %1,%%xmm1 # copy pitch -> xmm1[0]\n\t" |
| 215 |
|
|
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 216 |
|
|
"addss %%xmm1,%%xmm0 # calculate sample position of sample[1]\n\t" |
| 217 |
|
|
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 218 |
|
|
"addss %%xmm1,%%xmm0 # calculate sample position of sample[2]\n\t" |
| 219 |
|
|
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 220 |
|
|
"addss %%xmm1,%%xmm0 # calculate sample position of sample[3]\n\t" |
| 221 |
|
|
"movss %%xmm0,%%xmm2 # xmm0[0] -> xmm2[0]\n\t" |
| 222 |
|
|
"addss %%xmm1,%%xmm2 # calculate initial sample position for the next 4-sample cycle\n\t" |
| 223 |
|
|
"movss %%xmm2,(%0) # update 'Pos'\n\t" |
| 224 |
|
|
"shufps $0x1b,%%xmm0,%%xmm0 # swap, so that xmm0[0]=sample pos 0, xmm0[1]=sample pos 1,...\n\t" |
| 225 |
|
|
"cvttps2pi %%xmm0,%%mm4 # int(xmm0[0-1]) -> mm4\n\t" |
| 226 |
|
|
"shufps $0xe4,%%xmm0,%%xmm1 # xmm0[2-3] -> xmm1[2-3]\n\t" |
| 227 |
|
|
"shufps $0x0e,%%xmm1,%%xmm1 # xmm1[2-3] -> xmm1[0-1]\n\t" |
| 228 |
|
|
"cvttps2pi %%xmm1,%%mm5 # int(xmm1[0-1]) -> mm5\n\t" |
| 229 |
|
|
"cvtpi2ps %%mm5,%%xmm1 # double(mm5) -> xmm1[0-1]\n\t" |
| 230 |
|
|
"shufps $0x44,%%xmm1,%%xmm1 # shift lower 2 FPs up to the upper 2 cells\n\t" |
| 231 |
|
|
"cvtpi2ps %%mm4,%%xmm1 # double(mm4) -> xmm1[0-1]\n\t" |
| 232 |
|
|
"subps %%xmm1,%%xmm0 # xmm0[1-3] = xmm0[1-3] - xmm1[1-3]\n\t" |
| 233 |
|
|
: |
| 234 |
|
|
: "r" (Pos), /* %0 */ |
| 235 |
|
|
"m" (Pitch) /* %1 */ |
| 236 |
|
|
: "%xmm0", /* holds fractional position (0.0 <= x < 1.0) of sample 0-3 at the end */ |
| 237 |
|
|
"%xmm1", /* holds integer position (back converted to SPFP) of sample 0-3 at the end */ |
| 238 |
|
|
"mm4", /* holds integer position of sample 0-1 at the end */ |
| 239 |
|
|
"mm5", /* holds integer position of sample 2-3 at the end */ |
| 240 |
|
|
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)" |
| 241 |
|
|
); |
| 242 |
|
|
/* get sample values of pSrc[pos_int] and pSrc[pos_int+1] of the 4 samples */ |
| 243 |
|
|
__asm__ __volatile__ ( |
| 244 |
|
|
"movd %%mm4,%%edi # sample position of sample 0\n\t" |
| 245 |
|
|
"psrlq $32,%%mm4 # mm4 >> 32\n\t" |
| 246 |
|
|
"movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 0)\n\t" |
| 247 |
|
|
"movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 0+1)\n\t" |
| 248 |
|
|
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 249 |
|
|
"cvtsi2ss %%ecx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 250 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 251 |
|
|
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 252 |
|
|
"movd %%mm4,%%edi # sample position of sample 1\n\t" |
| 253 |
|
|
"movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 1)\n\t" |
| 254 |
|
|
"movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 1+1)\n\t" |
| 255 |
|
|
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 256 |
|
|
"cvtsi2ss %%ecx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 257 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 258 |
|
|
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 259 |
|
|
"movd %%mm5,%%edi # sample position of sample 2\n\t" |
| 260 |
|
|
"psrlq $32,%%mm5 # mm5 >> 32\n\t" |
| 261 |
|
|
"movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 2)\n\t" |
| 262 |
|
|
"movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 2+1)\n\t" |
| 263 |
|
|
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 264 |
|
|
"cvtsi2ss %%ecx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 265 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 266 |
|
|
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 267 |
|
|
"movd %%mm5,%%edi # sample position of sample 2\n\t" |
| 268 |
|
|
"movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 3)\n\t" |
| 269 |
|
|
"movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 3+1)\n\t" |
| 270 |
|
|
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 271 |
|
|
"cvtsi2ss %%ecx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 272 |
|
|
"shufps $0x1b, %%xmm2, %%xmm2 # swap to correct order\n\t" |
| 273 |
|
|
"shufps $0x1b, %%xmm3, %%xmm3 # swap to correct order\n\t" |
| 274 |
|
|
: /* no output */ |
| 275 |
|
|
: "S" (pSrc) /* %0 - sample read position */ |
| 276 |
|
|
: "%eax", "%ecx", /*"%edx",*/ "%edi", |
| 277 |
|
|
"%xmm2", /* holds pSrc[int_pos] of the 4 samples at the end */ |
| 278 |
|
|
"%xmm3", /* holds pSrc[int_pos+1] of the 4 samples at the end */ |
| 279 |
|
|
"mm4", /* holds integer position of sample 0-1 at the end */ |
| 280 |
|
|
"mm5", /* holds integer position of sample 2-3 at the end */ |
| 281 |
|
|
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)" |
| 282 |
|
|
); |
| 283 |
|
|
/* linear interpolation of the 4 samples simultaniously */ |
| 284 |
|
|
__asm__ __volatile__ ( |
| 285 |
|
|
"subps %%xmm2,%%xmm3 # xmm3 = pSrc[pos_int+1] - pSrc[pos_int]\n\t" |
| 286 |
|
|
"mulps %%xmm0,%%xmm3 # xmm3 = pos_fract * (pSrc[pos_int+1] - pSrc[pos_int])\n\t" |
| 287 |
|
|
"addps %%xmm3,%%xmm2 # xmm2 = pSrc[pos_int] + (pos_fract * (pSrc[pos_int+1] - pSrc[pos_int]))\n\t" |
| 288 |
|
|
: /* no output */ |
| 289 |
|
|
: /* no input */ |
| 290 |
|
|
: "%xmm2" /* holds linear interpolated sample point (of all 4 samples) at the end */ |
| 291 |
|
|
); |
| 292 |
|
|
} |
| 293 |
|
|
|
| 294 |
|
|
// TODO: no support for cubic interpolation yet |
| 295 |
|
|
inline static void Interpolate4StepsStereoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) { |
| 296 |
|
|
/* calculate playback position of each of the 4 samples by adding the associated pitch */ |
| 297 |
|
|
__asm__ __volatile__ ( |
| 298 |
|
|
"movss (%0),%%xmm0 # sample position of sample[0] -> xmm0[0]\n\t" |
| 299 |
|
|
"movss %1,%%xmm1 # copy pitch -> xmm1[0]\n\t" |
| 300 |
|
|
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 301 |
|
|
"addss %%xmm1,%%xmm0 # calculate sample position of sample[1]\n\t" |
| 302 |
|
|
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 303 |
|
|
"addss %%xmm1,%%xmm0 # calculate sample position of sample[2]\n\t" |
| 304 |
|
|
"shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t" |
| 305 |
|
|
"addss %%xmm1,%%xmm0 # calculate sample position of sample[3]\n\t" |
| 306 |
|
|
"movss %%xmm0,%%xmm2 # xmm0[0] -> xmm2[0]\n\t" |
| 307 |
|
|
"addss %%xmm1,%%xmm2 # calculate initial sample position for the next 4-sample cycle\n\t" |
| 308 |
|
|
"movss %%xmm2,(%0) # update 'Pos'\n\t" |
| 309 |
|
|
"shufps $0x1b,%%xmm0,%%xmm0 # swap, so that xmm0[0]=sample pos 0, xmm0[1]=sample pos 1,...\n\t" |
| 310 |
|
|
"cvttps2pi %%xmm0,%%mm4 # int(xmm0[0-1]) -> mm4\n\t" |
| 311 |
|
|
"shufps $0xe4,%%xmm0,%%xmm1 # xmm0[2-3] -> xmm1[2-3]\n\t" |
| 312 |
|
|
"shufps $0x0e,%%xmm1,%%xmm1 # xmm1[2-3] -> xmm1[0-1]\n\t" |
| 313 |
|
|
"cvttps2pi %%xmm1,%%mm5 # int(xmm1[0-1]) -> mm5\n\t" |
| 314 |
|
|
"cvtpi2ps %%mm5,%%xmm1 # double(mm5) -> xmm1[0-1]\n\t" |
| 315 |
|
|
"shufps $0x44,%%xmm1,%%xmm1 # shift lower 2 FPs up to the upper 2 cells\n\t" |
| 316 |
|
|
"cvtpi2ps %%mm4,%%xmm1 # double(mm4) -> xmm1[0-1]\n\t" |
| 317 |
|
|
"subps %%xmm1,%%xmm0 # xmm0[1-3] = xmm0[1-3] - xmm1[1-3]\n\t" |
| 318 |
|
|
: |
| 319 |
|
|
: "r" (Pos), /* %0 */ |
| 320 |
|
|
"m" (Pitch) /* %1 */ |
| 321 |
|
|
: "%xmm0", /* holds fractional position (0.0 <= x < 1.0) of sample 0-3 at the end */ |
| 322 |
|
|
"%xmm1", /* holds integer position (back converted to SPFP) of sample 0-3 at the end */ |
| 323 |
|
|
"mm4", /* holds integer position of sample 0-1 at the end */ |
| 324 |
|
|
"mm5", /* holds integer position of sample 2-3 at the end */ |
| 325 |
|
|
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)" |
| 326 |
|
|
); |
| 327 |
|
|
|
| 328 |
|
|
/* get sample values of pSrc[pos_int], pSrc[pos_int+1], pSrc[pos_int+2] and pSrc[pos_int+3] of the 4 samples */ |
| 329 |
|
|
__asm__ __volatile__ ( |
| 330 |
|
|
"xorl %%eax,%%eax # clear eax\n\t" |
| 331 |
|
|
"xorl %%edx,%%edx # clear edx\n\t" |
| 332 |
|
|
"movd %%mm4,%%edi # sample position of sample 0\n\t" |
| 333 |
|
|
"psrlq $32,%%mm4 # mm4 >> 32\n\t" |
| 334 |
|
|
"movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 0)\n\t" |
| 335 |
|
|
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 336 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 337 |
|
|
"movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 0+1)\n\t" |
| 338 |
|
|
"cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 339 |
|
|
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 340 |
|
|
"movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 0+2)\n\t" |
| 341 |
|
|
"cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t" |
| 342 |
|
|
"shufps $0x93, %%xmm4, %%xmm4 # shift up\n\t" |
| 343 |
|
|
"movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 0+3)\n\t" |
| 344 |
|
|
"cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t" |
| 345 |
|
|
"movd %%mm4,%%edi # sample position of sample 1\n\t" |
| 346 |
|
|
"shufps $0x93, %%xmm5, %%xmm5 # shift up\n\t" |
| 347 |
|
|
"movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 1)\n\t" |
| 348 |
|
|
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 349 |
|
|
"shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t" |
| 350 |
|
|
"movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 1+1)\n\t" |
| 351 |
|
|
"cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 352 |
|
|
"shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t" |
| 353 |
|
|
"movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 1+2)\n\t" |
| 354 |
|
|
"cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t" |
| 355 |
|
|
"shufps $0x93, %%xmm4, %%xmm4 # shift up\n\t" |
| 356 |
|
|
"movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 1+3)\n\t" |
| 357 |
|
|
"cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t" |
| 358 |
|
|
"movd %%mm5,%%edi # sample position of sample 2\n\t" |
| 359 |
|
|
"shufps $0x93, %%xmm5, %%xmm5 # shift up\n\t" |
| 360 |
|
|
"psrlq $32,%%mm5 # mm5 >> 32\n\t" |
| 361 |
|
|
"movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 2)\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 2+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 2+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 2+3)\n\t" |
| 371 |
|
|
"cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t" |
| 372 |
|
|
"movd %%mm5,%%edi # sample position of sample 3\n\t" |
| 373 |
|
|
"shufps $0x93, %%xmm5, %%xmm5 # shift up\n\t" |
| 374 |
|
|
"movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 3)\n\t" |
| 375 |
|
|
"cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t" |
| 376 |
|
|
"shufps $0x1b, %%xmm2, %%xmm2 # shift up\n\t" |
| 377 |
|
|
"movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 3+1)\n\t" |
| 378 |
|
|
"cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t" |
| 379 |
|
|
"shufps $0x1b, %%xmm3, %%xmm3 # shift up\n\t" |
| 380 |
|
|
"movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 3+2)\n\t" |
| 381 |
|
|
"cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t" |
| 382 |
|
|
"shufps $0x1b, %%xmm4, %%xmm4 # swap to correct order\n\t" |
| 383 |
|
|
"movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 3+3)\n\t" |
| 384 |
|
|
"cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t" |
| 385 |
|
|
"shufps $0x1b, %%xmm5, %%xmm5 # swap to correct order\n\t" |
| 386 |
|
|
: /* no output */ |
| 387 |
|
|
: "S" (pSrc) /* %0 - sample read position */ |
| 388 |
|
|
: "%eax", "%edx", "%edi", |
| 389 |
|
|
"xmm2", /* holds pSrc[int_pos] of the 4 samples at the end */ |
| 390 |
|
|
"xmm3", /* holds pSrc[int_pos+1] of the 4 samples at the end */ |
| 391 |
|
|
"xmm4", /* holds pSrc[int_pos+2] of the 4 samples at the end */ |
| 392 |
|
|
"xmm5", /* holds pSrc[int_pos+3] of the 4 samples at the end */ |
| 393 |
|
|
"mm4", /* holds integer position of sample 0-1 at the end */ |
| 394 |
|
|
"mm5", /* holds integer position of sample 2-3 at the end */ |
| 395 |
|
|
"st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)" |
| 396 |
|
|
); |
| 397 |
|
|
/* linear interpolation of the 4 samples (left & right channel) simultaniously */ |
| 398 |
|
|
__asm__ __volatile__ ( |
| 399 |
|
|
"subps %%xmm2,%%xmm4 # xmm4 = pSrc[pos_int+2] - pSrc[pos_int] (left channel)\n\t" |
| 400 |
|
|
"mulps %%xmm0,%%xmm4 # xmm4 = pos_fract * (pSrc[pos_int+2] - pSrc[pos_int]) (left channel)\n\t" |
| 401 |
|
|
"addps %%xmm4,%%xmm2 # xmm2 = pSrc[pos_int] + (pos_fract * (pSrc[pos_int+2] - pSrc[pos_int])) (left channel)\n\t" |
| 402 |
|
|
"subps %%xmm3,%%xmm5 # xmm5 = pSrc[pos_int+3] - pSrc[pos_int+1] (right channel)\n\t" |
| 403 |
|
|
"mulps %%xmm0,%%xmm5 # xmm5 = pos_fract * (pSrc[pos_int+3] - pSrc[pos_int+1]) (right channel)\n\t" |
| 404 |
|
|
"addps %%xmm5,%%xmm3 # xmm3 = pSrc[pos_int+1] + (pos_fract * (pSrc[pos_int+3] - pSrc[pos_int+1])) (right channel)\n\t" |
| 405 |
|
|
: /* no output */ |
| 406 |
|
|
: /* no input */ |
| 407 |
|
|
: "%xmm2", /* holds linear interpolated sample of left channel (of all 4 samples) at the end */ |
| 408 |
|
|
"%xmm3" /* holds linear interpolated sample of right channel (of all 4 samples) at the end */ |
| 409 |
|
|
); |
| 410 |
|
|
} |
| 411 |
schoenebeck |
617 |
#endif // CONFIG_ASM && ARCH_X86 |
| 412 |
schoenebeck |
320 |
}; |
| 413 |
|
|
|
| 414 |
|
|
} // namespace LinuxSampler |
| 415 |
|
|
|
| 416 |
|
|
#endif // __LS_RESAMPLER_H__ |
Parent Directory
| 
Revision Log