/[svn]/linuxsampler/trunk/src/engines/common/Resampler.h
ViewVC logotype

Annotation of /linuxsampler/trunk/src/engines/common/Resampler.h

Parent Directory Parent Directory | Revision Log Revision Log

Revision 1259 - (hide annotations) (download) (as text)
Tue Jun 26 21:41:09 2007 UTC (16 years, 10 months ago) by schoenebeck
File MIME type: text/x-c++hdr
File size: 28176 byte(s) 
* reenabled assembly features support, at the moment only for
  enabling a fast denormal FPU mode (x86 platforms supporting SSE2)
* bumped version to 0.4.0.6cvs
1 schoenebeck 320 /***************************************************************************
2     * *
3     * LinuxSampler - modular, streaming capable sampler *
4     * *
5     * Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck *
6 schoenebeck 1259 * Copyright (C) 2005 - 2007 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 schoenebeck 1259 // 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 schoenebeck 320 #ifndef __LS_RESAMPLER_H__
28     #define __LS_RESAMPLER_H__
29    
30     #include "../../common/global.h"
31    
32     // TODO: cubic interpolation is not yet supported by the MMX/SSE(1) version though
33 persson 903 // TODO: cubic interpolation is not supported for 24 bit samples
34 schoenebeck 320 #ifndef USE_LINEAR_INTERPOLATION
35     # define USE_LINEAR_INTERPOLATION 1 ///< set to 0 if you prefer cubic interpolation (slower, better quality)
36     #endif
37    
38     namespace LinuxSampler {
39    
40 schoenebeck 563 /** @brief Stereo sample point
41     *
42     * Encapsulates one stereo sample point, thus signal value for one
43     * sample point for left and right channel.
44     */
45 schoenebeck 320 struct stereo_sample_t {
46     float left;
47     float right;
48     };
49    
50 schoenebeck 563 /** @brief Resampler Template
51     *
52     * This template provides pure C++ and MMX/SSE assembly implementations
53     * for linear and cubic interpolation for pitching a mono or stereo
54     * input signal.
55     */
56 persson 903 template<bool INTERPOLATE,bool BITDEPTH24>
57 schoenebeck 320 class Resampler {
58     public:
59     inline static float GetNextSampleMonoCPP(sample_t* pSrc, double* Pos, float& Pitch) {
60     if (INTERPOLATE) return Interpolate1StepMonoCPP(pSrc, Pos, Pitch);
61     else { // no pitch, so no interpolation necessary
62     int pos_int = (int) *Pos;
63     *Pos += 1.0;
64     return pSrc [pos_int];
65     }
66     }
67    
68     inline static stereo_sample_t GetNextSampleStereoCPP(sample_t* pSrc, double* Pos, float& Pitch) {
69     if (INTERPOLATE) return Interpolate1StepStereoCPP(pSrc, Pos, Pitch);
70     else { // no pitch, so no interpolation necessary
71     int pos_int = (int) *Pos;
72     pos_int <<= 1;
73     *Pos += 1.0;
74     stereo_sample_t samplePoint;
75     samplePoint.left = pSrc[pos_int];
76     samplePoint.right = pSrc[pos_int+1];
77     return samplePoint;
78     }
79     }
80    
81 schoenebeck 1259 #if 0 // CONFIG_ASM && ARCH_X86
82 schoenebeck 320 inline static void GetNext4SamplesMonoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) {
83     if (INTERPOLATE) Interpolate4StepsMonoMMXSSE(pSrc, Pos, Pitch);
84     else { // no pitch, so no interpolation necessary
85     const float __4f = 4.0f;
86     __asm__ __volatile__ (
87     "movss (%1), %%xmm5 # load Pos\n\t"
88     "cvtss2si %%xmm5, %%edi # int(Pos)\n\t"
89     "addss %2, %%xmm5 # Pos += 4.0f\n\t"
90     "movswl (%0,%%edi,2), %%eax # load sample 0\n\t"
91     "cvtsi2ss %%eax, %%xmm2 # convert to float\n\t"
92     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
93     "movswl 2(%0,%%edi,2), %%edx # load sample 1\n\t"
94     "cvtsi2ss %%edx, %%xmm2 # convert to float\n\t"
95     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
96     "movss %%xmm5, (%1) # update Pos\n\t"
97     "movswl 4(%0,%%edi,2), %%eax # load sample 2\n\t"
98     "cvtsi2ss %%eax, %%xmm2 # convert to float\n\t"
99     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
100     "movswl 6(%0,%%edi,2), %%edx # load sample 3\n\t"
101     "cvtsi2ss %%edx, %%xmm2 # convert to float\n\t"
102     "shufps $0x1b, %%xmm2, %%xmm2 # swap to correct order\n\t"
103     :: "r" (pSrc), "r" (Pos), "m" (__4f)
104     : "%eax", "%edx", "%edi"
105     );
106     }
107     }
108    
109     inline static void GetNext4SamplesStereoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) {
110     if (INTERPOLATE) {
111     Interpolate4StepsStereoMMXSSE(pSrc, Pos, Pitch);
112     //EMMS;
113     } else { // no pitch, so no interpolation necessary
114     const float __4f = 4.0f;
115     __asm__ __volatile__ (
116     "movss (%1), %%xmm5 # load Pos\n\t"
117     "cvtss2si %%xmm5, %%edi # int(Pos)\n\t"
118     "addss %2, %%xmm5 # Pos += 4.0f\n\t"
119     "movswl (%0, %%edi,4), %%eax # load sample 0 (left)\n\t"
120     "cvtsi2ss %%eax, %%xmm2 # convert to float\n\t"
121     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
122     "movss %%xmm5, (%1) # update Pos\n\t"
123     "movswl 2(%0, %%edi,4), %%edx # load sample 0 (left)\n\t"
124     "cvtsi2ss %%edx, %%xmm3 # convert to float\n\t"
125     "shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t"
126     "movswl 4(%0, %%edi,4), %%eax # load sample 1 (left)\n\t"
127     "cvtsi2ss %%eax, %%xmm2 # convert to float\n\t"
128     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
129     "movswl 6(%0, %%edi,4), %%edx # load sample 1 (right)\n\t"
130     "cvtsi2ss %%edx, %%xmm3 # convert to float\n\t"
131     "shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t"
132     "movswl 8(%0, %%edi,4), %%eax # load sample 2 (left)\n\t"
133     "cvtsi2ss %%eax, %%xmm2 # convert to float\n\t"
134     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
135     "movswl 10(%0, %%edi,4), %%edx # load sample 2 (right)\n\t"
136     "cvtsi2ss %%edx, %%xmm3 # convert to float\n\t"
137     "shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t"
138     "movswl 12(%0, %%edi,4), %%eax # load sample 3 (left)\n\t"
139     "cvtsi2ss %%eax, %%xmm2 # convert to float\n\t"
140     "shufps $0x1b, %%xmm2, %%xmm2 # swap to correct order\n\t"
141     "movswl 14(%0, %%edi,4), %%edx # load sample 3 (right)\n\t"
142     "cvtsi2ss %%edx, %%xmm3 # convert to float\n\t"
143     "shufps $0x1b, %%xmm3, %%xmm3 # swap to correct order\n\t"
144     :: "r" (pSrc), "r" (Pos), "m" (__4f)
145     : "%eax", "%edx", "%edi"
146     );
147     }
148     }
149 schoenebeck 617 #endif // CONFIG_ASM && ARCH_X86
150 schoenebeck 320
151     protected:
152    
153 persson 903 static int getSample(sample_t* src, int pos) {
154     if (BITDEPTH24) {
155     pos *= 3;
156     unsigned char* p = (unsigned char*)src;
157     return p[pos] << 8 | p[pos + 1] << 16 | p[pos + 2] << 24;
158     } else {
159     return src[pos];
160     }
161     }
162    
163 schoenebeck 320 inline static float Interpolate1StepMonoCPP(sample_t* pSrc, double* Pos, float& Pitch) {
164     int pos_int = (int) *Pos; // integer position
165     float pos_fract = *Pos - pos_int; // fractional part of position
166    
167     #if USE_LINEAR_INTERPOLATION
168 persson 903 int x1 = getSample(pSrc, pos_int);
169     int x2 = getSample(pSrc, pos_int + 1);
170     float samplePoint = (x1 + pos_fract * (x2 - x1));
171 schoenebeck 320 #else // polynomial interpolation
172     float xm1 = pSrc[pos_int];
173     float x0 = pSrc[pos_int+1];
174     float x1 = pSrc[pos_int+2];
175     float x2 = pSrc[pos_int+3];
176     float a = (3.0f * (x0 - x1) - xm1 + x2) * 0.5f;
177     float b = 2.0f * x1 + xm1 - (5.0f * x0 + x2) * 0.5f;
178     float c = (x1 - xm1) * 0.5f;
179     float samplePoint = (((a * pos_fract) + b) * pos_fract + c) * pos_fract + x0;
180     #endif // USE_LINEAR_INTERPOLATION
181    
182     *Pos += Pitch;
183     return samplePoint;
184     }
185    
186     inline static stereo_sample_t Interpolate1StepStereoCPP(sample_t* pSrc, double* Pos, float& Pitch) {
187     int pos_int = (int) *Pos; // integer position
188     float pos_fract = *Pos - pos_int; // fractional part of position
189     pos_int <<= 1;
190    
191     stereo_sample_t samplePoint;
192    
193     #if USE_LINEAR_INTERPOLATION
194     // left channel
195 persson 903 int x1 = getSample(pSrc, pos_int);
196     int x2 = getSample(pSrc, pos_int + 2);
197     samplePoint.left = (x1 + pos_fract * (x2 - x1));
198 schoenebeck 320 // right channel
199 persson 903 x1 = getSample(pSrc, pos_int + 1);
200     x2 = getSample(pSrc, pos_int + 3);
201     samplePoint.right = (x1 + pos_fract * (x2 - x1));
202 schoenebeck 320 #else // polynomial interpolation
203     // calculate left channel
204     float xm1 = pSrc[pos_int];
205     float x0 = pSrc[pos_int+2];
206     float x1 = pSrc[pos_int+4];
207     float x2 = pSrc[pos_int+6];
208     float a = (3.0f * (x0 - x1) - xm1 + x2) * 0.5f;
209     float b = 2.0f * x1 + xm1 - (5.0f * x0 + x2) * 0.5f;
210     float c = (x1 - xm1) * 0.5f;
211     samplePoint.left = (((a * pos_fract) + b) * pos_fract + c) * pos_fract + x0;
212    
213     //calculate right channel
214     xm1 = pSrc[pos_int+1];
215     x0 = pSrc[pos_int+3];
216     x1 = pSrc[pos_int+5];
217     x2 = pSrc[pos_int+7];
218     a = (3.0f * (x0 - x1) - xm1 + x2) * 0.5f;
219     b = 2.0f * x1 + xm1 - (5.0f * x0 + x2) * 0.5f;
220     c = (x1 - xm1) * 0.5f;
221     samplePoint.right = (((a * pos_fract) + b) * pos_fract + c) * pos_fract + x0;
222     #endif // USE_LINEAR_INTERPOLATION
223    
224     *Pos += Pitch;
225     return samplePoint;
226     }
227    
228 schoenebeck 1259 #if 0 // CONFIG_ASM && ARCH_X86
229 schoenebeck 320 // TODO: no support for cubic interpolation yet
230     inline static void Interpolate4StepsMonoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) {
231     /* calculate playback position of each of the 4 samples by adding the associated pitch */
232     __asm__ __volatile__ (
233     "movss (%0),%%xmm0 # sample position of sample[0] -> xmm0[0]\n\t"
234     "movss %1,%%xmm1 # copy pitch -> xmm1[0]\n\t"
235     "shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t"
236     "addss %%xmm1,%%xmm0 # calculate sample position of sample[1]\n\t"
237     "shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t"
238     "addss %%xmm1,%%xmm0 # calculate sample position of sample[2]\n\t"
239     "shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t"
240     "addss %%xmm1,%%xmm0 # calculate sample position of sample[3]\n\t"
241     "movss %%xmm0,%%xmm2 # xmm0[0] -> xmm2[0]\n\t"
242     "addss %%xmm1,%%xmm2 # calculate initial sample position for the next 4-sample cycle\n\t"
243     "movss %%xmm2,(%0) # update 'Pos'\n\t"
244     "shufps $0x1b,%%xmm0,%%xmm0 # swap, so that xmm0[0]=sample pos 0, xmm0[1]=sample pos 1,...\n\t"
245     "cvttps2pi %%xmm0,%%mm4 # int(xmm0[0-1]) -> mm4\n\t"
246     "shufps $0xe4,%%xmm0,%%xmm1 # xmm0[2-3] -> xmm1[2-3]\n\t"
247     "shufps $0x0e,%%xmm1,%%xmm1 # xmm1[2-3] -> xmm1[0-1]\n\t"
248     "cvttps2pi %%xmm1,%%mm5 # int(xmm1[0-1]) -> mm5\n\t"
249     "cvtpi2ps %%mm5,%%xmm1 # double(mm5) -> xmm1[0-1]\n\t"
250     "shufps $0x44,%%xmm1,%%xmm1 # shift lower 2 FPs up to the upper 2 cells\n\t"
251     "cvtpi2ps %%mm4,%%xmm1 # double(mm4) -> xmm1[0-1]\n\t"
252     "subps %%xmm1,%%xmm0 # xmm0[1-3] = xmm0[1-3] - xmm1[1-3]\n\t"
253     :
254     : "r" (Pos), /* %0 */
255     "m" (Pitch) /* %1 */
256     : "%xmm0", /* holds fractional position (0.0 <= x < 1.0) of sample 0-3 at the end */
257     "%xmm1", /* holds integer position (back converted to SPFP) of sample 0-3 at the end */
258     "mm4", /* holds integer position of sample 0-1 at the end */
259     "mm5", /* holds integer position of sample 2-3 at the end */
260     "st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)"
261     );
262     /* get sample values of pSrc[pos_int] and pSrc[pos_int+1] of the 4 samples */
263     __asm__ __volatile__ (
264     "movd %%mm4,%%edi # sample position of sample 0\n\t"
265     "psrlq $32,%%mm4 # mm4 >> 32\n\t"
266     "movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 0)\n\t"
267     "movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 0+1)\n\t"
268     "cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t"
269     "cvtsi2ss %%ecx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t"
270     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
271     "shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t"
272     "movd %%mm4,%%edi # sample position of sample 1\n\t"
273     "movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 1)\n\t"
274     "movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 1+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 %%mm5,%%edi # sample position of sample 2\n\t"
280     "psrlq $32,%%mm5 # mm5 >> 32\n\t"
281     "movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 2)\n\t"
282     "movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 2+1)\n\t"
283     "cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t"
284     "cvtsi2ss %%ecx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t"
285     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
286     "shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t"
287     "movd %%mm5,%%edi # sample position of sample 2\n\t"
288     "movswl (%0,%%edi,2),%%eax # pSrc[pos_int] (sample 3)\n\t"
289     "movswl 2(%0,%%edi,2),%%ecx # pSrc[pos_int] (sample 3+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 $0x1b, %%xmm2, %%xmm2 # swap to correct order\n\t"
293     "shufps $0x1b, %%xmm3, %%xmm3 # swap to correct order\n\t"
294     : /* no output */
295     : "S" (pSrc) /* %0 - sample read position */
296     : "%eax", "%ecx", /*"%edx",*/ "%edi",
297     "%xmm2", /* holds pSrc[int_pos] of the 4 samples at the end */
298     "%xmm3", /* holds pSrc[int_pos+1] of the 4 samples at the end */
299     "mm4", /* holds integer position of sample 0-1 at the end */
300     "mm5", /* holds integer position of sample 2-3 at the end */
301     "st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)"
302     );
303     /* linear interpolation of the 4 samples simultaniously */
304     __asm__ __volatile__ (
305     "subps %%xmm2,%%xmm3 # xmm3 = pSrc[pos_int+1] - pSrc[pos_int]\n\t"
306     "mulps %%xmm0,%%xmm3 # xmm3 = pos_fract * (pSrc[pos_int+1] - pSrc[pos_int])\n\t"
307     "addps %%xmm3,%%xmm2 # xmm2 = pSrc[pos_int] + (pos_fract * (pSrc[pos_int+1] - pSrc[pos_int]))\n\t"
308     : /* no output */
309     : /* no input */
310     : "%xmm2" /* holds linear interpolated sample point (of all 4 samples) at the end */
311     );
312     }
313    
314     // TODO: no support for cubic interpolation yet
315     inline static void Interpolate4StepsStereoMMXSSE(sample_t* pSrc, void* Pos, float& Pitch) {
316     /* calculate playback position of each of the 4 samples by adding the associated pitch */
317     __asm__ __volatile__ (
318     "movss (%0),%%xmm0 # sample position of sample[0] -> xmm0[0]\n\t"
319     "movss %1,%%xmm1 # copy pitch -> xmm1[0]\n\t"
320     "shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t"
321     "addss %%xmm1,%%xmm0 # calculate sample position of sample[1]\n\t"
322     "shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t"
323     "addss %%xmm1,%%xmm0 # calculate sample position of sample[2]\n\t"
324     "shufps $0x90,%%xmm0,%%xmm0 # shift up, but keep xmm0[0]\n\t"
325     "addss %%xmm1,%%xmm0 # calculate sample position of sample[3]\n\t"
326     "movss %%xmm0,%%xmm2 # xmm0[0] -> xmm2[0]\n\t"
327     "addss %%xmm1,%%xmm2 # calculate initial sample position for the next 4-sample cycle\n\t"
328     "movss %%xmm2,(%0) # update 'Pos'\n\t"
329     "shufps $0x1b,%%xmm0,%%xmm0 # swap, so that xmm0[0]=sample pos 0, xmm0[1]=sample pos 1,...\n\t"
330     "cvttps2pi %%xmm0,%%mm4 # int(xmm0[0-1]) -> mm4\n\t"
331     "shufps $0xe4,%%xmm0,%%xmm1 # xmm0[2-3] -> xmm1[2-3]\n\t"
332     "shufps $0x0e,%%xmm1,%%xmm1 # xmm1[2-3] -> xmm1[0-1]\n\t"
333     "cvttps2pi %%xmm1,%%mm5 # int(xmm1[0-1]) -> mm5\n\t"
334     "cvtpi2ps %%mm5,%%xmm1 # double(mm5) -> xmm1[0-1]\n\t"
335     "shufps $0x44,%%xmm1,%%xmm1 # shift lower 2 FPs up to the upper 2 cells\n\t"
336     "cvtpi2ps %%mm4,%%xmm1 # double(mm4) -> xmm1[0-1]\n\t"
337     "subps %%xmm1,%%xmm0 # xmm0[1-3] = xmm0[1-3] - xmm1[1-3]\n\t"
338     :
339     : "r" (Pos), /* %0 */
340     "m" (Pitch) /* %1 */
341     : "%xmm0", /* holds fractional position (0.0 <= x < 1.0) of sample 0-3 at the end */
342     "%xmm1", /* holds integer position (back converted to SPFP) of sample 0-3 at the end */
343     "mm4", /* holds integer position of sample 0-1 at the end */
344     "mm5", /* holds integer position of sample 2-3 at the end */
345     "st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)"
346     );
347    
348     /* get sample values of pSrc[pos_int], pSrc[pos_int+1], pSrc[pos_int+2] and pSrc[pos_int+3] of the 4 samples */
349     __asm__ __volatile__ (
350     "xorl %%eax,%%eax # clear eax\n\t"
351     "xorl %%edx,%%edx # clear edx\n\t"
352     "movd %%mm4,%%edi # sample position of sample 0\n\t"
353     "psrlq $32,%%mm4 # mm4 >> 32\n\t"
354     "movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 0)\n\t"
355     "cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t"
356     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
357     "movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 0+1)\n\t"
358     "cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t"
359     "shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t"
360     "movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 0+2)\n\t"
361     "cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t"
362     "shufps $0x93, %%xmm4, %%xmm4 # shift up\n\t"
363     "movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 0+3)\n\t"
364     "cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t"
365     "movd %%mm4,%%edi # sample position of sample 1\n\t"
366     "shufps $0x93, %%xmm5, %%xmm5 # shift up\n\t"
367     "movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 1)\n\t"
368     "cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t"
369     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
370     "movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 1+1)\n\t"
371     "cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t"
372     "shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t"
373     "movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 1+2)\n\t"
374     "cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t"
375     "shufps $0x93, %%xmm4, %%xmm4 # shift up\n\t"
376     "movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 1+3)\n\t"
377     "cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t"
378     "movd %%mm5,%%edi # sample position of sample 2\n\t"
379     "shufps $0x93, %%xmm5, %%xmm5 # shift up\n\t"
380     "psrlq $32,%%mm5 # mm5 >> 32\n\t"
381     "movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 2)\n\t"
382     "cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t"
383     "shufps $0x93, %%xmm2, %%xmm2 # shift up\n\t"
384     "movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 2+1)\n\t"
385     "cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t"
386     "shufps $0x93, %%xmm3, %%xmm3 # shift up\n\t"
387     "movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 2+2)\n\t"
388     "cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t"
389     "shufps $0x93, %%xmm4, %%xmm4 # shift up\n\t"
390     "movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 2+3)\n\t"
391     "cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t"
392     "movd %%mm5,%%edi # sample position of sample 3\n\t"
393     "shufps $0x93, %%xmm5, %%xmm5 # shift up\n\t"
394     "movswl (%0,%%edi,4),%%eax # pSrc[pos_int] (sample 3)\n\t"
395     "cvtsi2ss %%eax, %%xmm2 # pSrc[pos_int] -> xmm2[0]\n\t"
396     "shufps $0x1b, %%xmm2, %%xmm2 # shift up\n\t"
397     "movswl 2(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 3+1)\n\t"
398     "cvtsi2ss %%edx, %%xmm3 # pSrc[pos_int] -> xmm3[0]\n\t"
399     "shufps $0x1b, %%xmm3, %%xmm3 # shift up\n\t"
400     "movswl 4(%0,%%edi,4),%%eax # pSrc[pos_int] (sample 3+2)\n\t"
401     "cvtsi2ss %%eax, %%xmm4 # pSrc[pos_int] -> xmm4[0]\n\t"
402     "shufps $0x1b, %%xmm4, %%xmm4 # swap to correct order\n\t"
403     "movswl 6(%0,%%edi,4),%%edx # pSrc[pos_int] (sample 3+3)\n\t"
404     "cvtsi2ss %%edx, %%xmm5 # pSrc[pos_int] -> xmm5[0]\n\t"
405     "shufps $0x1b, %%xmm5, %%xmm5 # swap to correct order\n\t"
406     : /* no output */
407     : "S" (pSrc) /* %0 - sample read position */
408     : "%eax", "%edx", "%edi",
409     "xmm2", /* holds pSrc[int_pos] of the 4 samples at the end */
410     "xmm3", /* holds pSrc[int_pos+1] of the 4 samples at the end */
411     "xmm4", /* holds pSrc[int_pos+2] of the 4 samples at the end */
412     "xmm5", /* holds pSrc[int_pos+3] of the 4 samples at the end */
413     "mm4", /* holds integer position of sample 0-1 at the end */
414     "mm5", /* holds integer position of sample 2-3 at the end */
415     "st", "st(1)", "st(2)", "st(3)", "st(4)", "st(5)", "st(6)", "st(7)"
416     );
417     /* linear interpolation of the 4 samples (left & right channel) simultaniously */
418     __asm__ __volatile__ (
419     "subps %%xmm2,%%xmm4 # xmm4 = pSrc[pos_int+2] - pSrc[pos_int] (left channel)\n\t"
420     "mulps %%xmm0,%%xmm4 # xmm4 = pos_fract * (pSrc[pos_int+2] - pSrc[pos_int]) (left channel)\n\t"
421     "addps %%xmm4,%%xmm2 # xmm2 = pSrc[pos_int] + (pos_fract * (pSrc[pos_int+2] - pSrc[pos_int])) (left channel)\n\t"
422     "subps %%xmm3,%%xmm5 # xmm5 = pSrc[pos_int+3] - pSrc[pos_int+1] (right channel)\n\t"
423     "mulps %%xmm0,%%xmm5 # xmm5 = pos_fract * (pSrc[pos_int+3] - pSrc[pos_int+1]) (right channel)\n\t"
424     "addps %%xmm5,%%xmm3 # xmm3 = pSrc[pos_int+1] + (pos_fract * (pSrc[pos_int+3] - pSrc[pos_int+1])) (right channel)\n\t"
425     : /* no output */
426     : /* no input */
427     : "%xmm2", /* holds linear interpolated sample of left channel (of all 4 samples) at the end */
428     "%xmm3" /* holds linear interpolated sample of right channel (of all 4 samples) at the end */
429     );
430     }
431 schoenebeck 617 #endif // CONFIG_ASM && ARCH_X86
432 schoenebeck 320 };
433    
434     } // namespace LinuxSampler
435    
436     #endif // __LS_RESAMPLER_H__
  ViewVC Help
Powered by ViewVC