/[svn]/linuxsampler/trunk/src/engines/gig/Synthesizer.h
ViewVC logotype

Annotation of /linuxsampler/trunk/src/engines/gig/Synthesizer.h

Parent Directory Parent Directory | Revision Log Revision Log


Revision 325 - (hide annotations) (download) (as text)
Tue Dec 21 04:54:37 2004 UTC (19 years, 3 months ago) by senkov
File MIME type: text/x-c++hdr
File size: 20243 byte(s)
* Added some profiling capabilities, bugs.

1 schoenebeck 320 /***************************************************************************
2     * *
3     * LinuxSampler - modular, streaming capable sampler *
4     * *
5     * Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck *
6     * *
7     * This program is free software; you can redistribute it and/or modify *
8     * it under the terms of the GNU General Public License as published by *
9     * the Free Software Foundation; either version 2 of the License, or *
10     * (at your option) any later version. *
11     * *
12     * This program is distributed in the hope that it will be useful, *
13     * but WITHOUT ANY WARRANTY; without even the implied warranty of *
14     * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the *
15     * GNU General Public License for more details. *
16     * *
17     * You should have received a copy of the GNU General Public License *
18     * along with this program; if not, write to the Free Software *
19     * Foundation, Inc., 59 Temple Place, Suite 330, Boston, *
20     * MA 02111-1307 USA *
21     ***************************************************************************/
22    
23     #ifndef __LS_GIG_SYNTHESIZER_H__
24     #define __LS_GIG_SYNTHESIZER_H__
25    
26     #include "../../common/RTMath.h"
27     #include "../common/Resampler.h"
28     #include "../common/BiquadFilter.h"
29     #include "Filter.h"
30     #include "Voice.h"
31    
32     #define SYNTHESIS_MODE_SET_CONSTPITCH(iMode,bVal) if (bVal) iMode |= 0x01; else iMode &= 0xfe /* (un)set mode bit 0 */
33     #define SYNTHESIS_MODE_SET_LOOP(iMode,bVal) if (bVal) iMode |= 0x02; else iMode &= 0xfd /* (un)set mode bit 1 */
34     #define SYNTHESIS_MODE_SET_INTERPOLATE(iMode,bVal) if (bVal) iMode |= 0x04; else iMode &= 0xfb /* (un)set mode bit 2 */
35     #define SYNTHESIS_MODE_SET_FILTER(iMode,bVal) if (bVal) iMode |= 0x08; else iMode &= 0xf7 /* (un)set mode bit 3 */
36     #define SYNTHESIS_MODE_SET_CHANNELS(iMode,bVal) if (bVal) iMode |= 0x10; else iMode &= 0xef /* (un)set mode bit 4 */
37     #define SYNTHESIS_MODE_SET_IMPLEMENTATION(iMode,bVal) if (bVal) iMode |= 0x20; else iMode &= 0xdf /* (un)set mode bit 5 */
38 senkov 325 #define SYNTHESIS_MODE_SET_PROFILING(iMode,bVal) if (bVal) iMode |= 0x40; else iMode &= ~0x40 /* (un)set mode bit 6 */
39 schoenebeck 320
40     #define SYNTHESIS_MODE_GET_CONSTPITCH(iMode) iMode & 0x01
41     #define SYNTHESIS_MODE_GET_LOOP(iMode) iMode & 0x02
42     #define SYNTHESIS_MODE_GET_INTERPOLATE(iMode) iMode & 0x04
43     #define SYNTHESIS_MODE_GET_FILTER(iMode) iMode & 0x08
44     #define SYNTHESIS_MODE_GET_CHANNELS(iMode) iMode & 0x10
45     #define SYNTHESIS_MODE_GET_IMPLEMENTATION(iMode) iMode & 0x20
46    
47     // that's usually gig::Voice of course, but we make it a macro so we can
48     // include this code for our synthesis benchmark which uses fake data
49     // structures
50     #ifndef VOICE
51     # define VOICE Voice
52     #endif // VOICE
53    
54     namespace LinuxSampler { namespace gig {
55    
56 senkov 325 typedef void SynthesizeFragment_Fn(VOICE&, uint, sample_t*, uint);
57 schoenebeck 320
58     void* GetSynthesisFunction(const int SynthesisMode);
59 senkov 325 void RunSynthesisFunction(const int SynthesisMode, VOICE& voice, uint Samples, sample_t* pSrc, uint Skip);
60 schoenebeck 320
61     enum channels_t {
62     MONO,
63     STEREO
64     };
65    
66     template<implementation_t IMPLEMENTATION, channels_t CHANNELS, bool USEFILTER, bool INTERPOLATE, bool DOLOOP, bool CONSTPITCH>
67     class Synthesizer : public __RTMath<IMPLEMENTATION>, public LinuxSampler::Resampler<INTERPOLATE> {
68     public:
69     template<typename VOICE_T>
70 senkov 325 inline static void SynthesizeFragment(VOICE_T& Voice, uint Samples, sample_t* pSrc, uint i) {
71 schoenebeck 320 if (IMPLEMENTATION == ASM_X86_MMX_SSE) {
72     float fPos = (float) Voice.Pos;
73     SynthesizeFragment(Voice, Samples, pSrc, i, Voice.pSample->LoopPlayCount,
74     Voice.pSample->LoopStart,
75     Voice.pSample->LoopEnd,
76     Voice.pSample->LoopSize,
77     Voice.LoopCyclesLeft,
78     (void *)&fPos,
79     Voice.PitchBase,
80     Voice.PitchBend);
81     if (INTERPOLATE) EMMS;
82     Voice.Pos = (double) fPos;
83     } else {
84     SynthesizeFragment(Voice, Samples, pSrc, i, Voice.pSample->LoopPlayCount,
85     Voice.pSample->LoopStart,
86     Voice.pSample->LoopEnd,
87     Voice.pSample->LoopSize,
88     Voice.LoopCyclesLeft,
89     (void *)&Voice.Pos,
90     Voice.PitchBase,
91     Voice.PitchBend);
92     }
93     }
94    
95     //protected:
96    
97     template<typename VOICE_T>
98 senkov 325 inline static void SynthesizeFragment(VOICE_T& Voice, uint Samples, sample_t* pSrc, uint& i, uint& LoopPlayCount, uint LoopStart, uint LoopEnd, uint LoopSize, uint& LoopCyclesLeft, void* Pos, float& PitchBase, float& PitchBend) {
99 schoenebeck 320 const float loopEnd = Float(LoopEnd);
100     const float PBbyPB = Mul(PitchBase, PitchBend);
101     const float f_LoopStart = Float(LoopStart);
102     const float f_LoopSize = Float(LoopSize);
103     if (DOLOOP) {
104     if (LoopPlayCount) {
105     // render loop (loop count limited)
106     while (i < Samples && LoopCyclesLeft) {
107     if (CONSTPITCH) {
108 senkov 325 const uint processEnd = Min(Samples, i + DiffToLoopEnd(loopEnd,Pos, PBbyPB) + 1); //TODO: instead of +1 we could also round up
109 schoenebeck 320 while (i < processEnd) Synthesize(Voice, Pos, pSrc, i);
110     }
111     else Synthesize(Voice, Pos, pSrc, i);
112     if (WrapLoop(f_LoopStart, f_LoopSize, loopEnd, Pos)) LoopCyclesLeft--;
113     }
114     // render on without loop
115     while (i < Samples) Synthesize(Voice, Pos, pSrc, i);
116     }
117     else { // render loop (endless loop)
118     while (i < Samples) {
119     if (CONSTPITCH) {
120 senkov 325 const uint processEnd = Min(Samples, i + DiffToLoopEnd(loopEnd, Pos, PBbyPB) + 1); //TODO: instead of +1 we could also round up
121 schoenebeck 320 while (i < processEnd) Synthesize(Voice, Pos, pSrc, i);
122     }
123     else Synthesize(Voice, Pos, pSrc, i);
124     WrapLoop(f_LoopStart, f_LoopSize, loopEnd, Pos);
125     }
126     }
127     }
128     else { // no looping
129     while (i < Samples) { Synthesize(Voice, Pos, pSrc, i);}
130     }
131     }
132    
133     template<typename VOICE_T>
134 senkov 325 inline static void Synthesize(VOICE_T& Voice, void* Pos, sample_t* pSrc, uint& i) {
135 schoenebeck 320 Synthesize(pSrc, Pos,
136     Voice.pEngine->pSynthesisParameters[Event::destination_vco][i],
137     Voice.pEngine->pOutputLeft,
138     Voice.pEngine->pOutputRight,
139     i,
140     Voice.pEngine->pSynthesisParameters[Event::destination_vca],
141     &Voice.PanLeft,
142     &Voice.PanRight,
143     Voice.FilterLeft,
144     Voice.FilterRight,
145     Voice.pEngine->pBasicFilterParameters[i],
146     Voice.pEngine->pMainFilterParameters[i]);
147     }
148    
149     inline static int DiffToLoopEnd(const float& LoopEnd, const void* Pos, const float& Pitch) {
150     switch (IMPLEMENTATION) {
151     // pure C++ implementation (thus platform independent)
152     case CPP: {
153 senkov 325 return uint((LoopEnd - *((double *)Pos)) / Pitch);
154 schoenebeck 320 }
155     case ASM_X86_MMX_SSE: {
156     int result;
157     __asm__ __volatile__ (
158     "movss (%1), %%xmm0 #read loopend\n\t"
159     "subss (%2), %%xmm0 #sub pos\n\t"
160     "divss (%3), %%xmm0 #div by pitch\n\t"
161     "cvtss2si %%xmm0, %0 #convert to int\n\t"
162     : "=r" (result) /* %0 */
163     : "r" (&LoopEnd), /* %1 */
164     "r" (Pos), /* %2 */
165     "r" (&Pitch) /* %3 */
166     );
167     return result;
168     }
169     }
170     }
171    
172     inline static int WrapLoop(const float& LoopStart, const float& LoopSize, const float& LoopEnd, void* vPos) {
173     switch (IMPLEMENTATION) {
174     // pure C++ implementation (thus platform independent)
175     case CPP: {
176     double * Pos = (double *)vPos;
177     if (*Pos < LoopEnd) return 0;
178     *Pos = fmod(*Pos - LoopEnd, LoopSize) + LoopStart;
179     return 1;
180     }
181     case ASM_X86_MMX_SSE: {
182     int result;
183     __asm__ __volatile__ (
184     "movss (%2), %%xmm0 # load LoopEnd\n\t"
185     "movss (%1), %%xmm1 # load Pos\n\t"
186     "comiss %%xmm0, %%xmm1 # LoopEnd <> Pos\n\t"
187     "movl $0,%%eax # result = 0\n\t"
188     "jb 1f # jump if no work needs to be done\n\t"
189     "movss (%3), %%xmm2 # load LoopSize\n\t"
190     "subss %%xmm0, %%xmm1 # Pos - LoopEnd\n\t"
191     //now the fmodf
192     "movss %%xmm1, %%xmm3 # xmm3 = (Pos - LoopEnd)\n\t"
193     "divss %%xmm2, %%xmm1 # (Pos - LoopEnd) / LoopSize\n\t"
194     "cvttss2si %%xmm1, %%eax # convert to int\n\t"
195     "cvtsi2ss %%eax, %%xmm1 # convert back to float\n\t"
196     "movss (%4), %%xmm0 # load LoopStart\n\t"
197     "mulss %%xmm2, %%xmm1 # LoopSize * int((Pos-LoopEnd)/LoopSize)\n\t"
198     "subss %%xmm1, %%xmm3 # xmm2 = fmodf(Pos - LoopEnd, LoopSize)\n\t"
199     //done with fmodf
200     "addss %%xmm0, %%xmm3 # add LoopStart\n\t"
201     "movss %%xmm3, (%1) # update Pos\n\t"
202     "movl $1,%%eax # result = 1\n\t"
203     ".balign 16 \n\t"
204     "1:\n\t"
205     : "=a" (result) /* %0 */
206     : "r" (vPos), /* %1 */
207     "r" (&LoopEnd), /* %2 */
208     "r" (&LoopSize), /* %3 */
209     "r" (&LoopStart) /* %4 */
210     );
211     return result;
212     }
213     }
214     }
215    
216 senkov 325 inline static void Synthesize(sample_t* pSrc, void* Pos, float& Pitch, float* pOutL, float* pOutR, uint& i, float* Volume, float* PanL, float* PanR, Filter& FilterL, Filter& FilterR, biquad_param_t& bqBase, biquad_param_t& bqMain) {
217 schoenebeck 320 switch (IMPLEMENTATION) {
218     // pure C++ implementation (thus platform independent)
219     case CPP: {
220     switch (CHANNELS) {
221     case MONO: {
222     float samplePoint = GetNextSampleMonoCPP(pSrc, (double *)Pos, Pitch);
223     if (USEFILTER) samplePoint = FilterL.Apply(&bqBase, &bqMain, samplePoint);
224     pOutL[i] += samplePoint * Volume[i] * *PanL;
225     pOutR[i] += samplePoint * Volume[i] * *PanR;
226     i++;
227     break;
228     }
229     case STEREO: {
230     stereo_sample_t samplePoint = GetNextSampleStereoCPP(pSrc, (double *)Pos, Pitch);
231     if (USEFILTER) {
232     samplePoint.left = FilterL.Apply(&bqBase, &bqMain, samplePoint.left);
233     samplePoint.right = FilterR.Apply(&bqBase, &bqMain, samplePoint.right);
234     }
235     pOutL[i] += samplePoint.left * Volume[i] * *PanL;
236     pOutR[i] += samplePoint.right * Volume[i] * *PanR;
237     i++;
238     break;
239     }
240     }
241     break;
242     }
243     // Assembly optimization using the MMX & SSE(1) instruction set (thus only for x86)
244     case ASM_X86_MMX_SSE: {
245     const int ii = i & 0xfffffffc;
246     i += 4;
247     switch (CHANNELS) {
248     case MONO: {
249     GetNext4SamplesMonoMMXSSE(pSrc, (float *)Pos, Pitch); // outputs samples in xmm2
250     if (USEFILTER) {
251     /* prepare filter input */
252     __asm__ __volatile__ (
253     "movaps %xmm2,%xmm0"
254     );
255     FilterL.Apply4StepsSSE(&bqBase, &bqMain); // xmm0 input, xmm7 output
256     __asm__ __volatile__ (
257     "movaps %xmm7,%xmm2 # mono filter result -> xmm2"
258     );
259     }
260     /* apply panorama and volume factors */
261     __asm__ __volatile__ (
262     "movss (%1),%%xmm0 # load pan left\n\t"
263     "movss (%2),%%xmm1 # load pan right\n\t"
264     "movaps (%0),%%xmm4 # load vca\n\t"
265     "shufps $0x00,%%xmm0,%%xmm0 # copy pan left to the other 3 cells\n\t"
266     "shufps $0x00,%%xmm1,%%xmm1 # copy pan right to the other 3 cells\n\t"
267     "mulps %%xmm2,%%xmm0 # left = sample * pan_left\n\t"
268     "mulps %%xmm2,%%xmm1 # right = sample * pan_right\n\t"
269     "mulps %%xmm4,%%xmm0 # left = vca * (sample * pan_left)\n\t"
270     "mulps %%xmm4,%%xmm1 # right = vca * (sample * pan_right)\n\t"
271     : /* no output */
272     : "r" (&Volume[ii]), /* %0 */
273     "r" (PanL), /* %1 */
274     "r" (PanR) /* %2 */
275     : "xmm0", /* holds final left sample (for the 4 samples) at the end */
276     "xmm1" /* holds final right sample (for the 4 samples) at the end */
277     );
278     break;
279     }
280     case STEREO: {
281     GetNext4SamplesStereoMMXSSE(pSrc, (float *)Pos, Pitch); // outputs samples in xmm2 (left channel) and xmm3 (right channel)
282     if (USEFILTER) {
283     __asm__ __volatile__ (
284     "movaps %xmm2,%xmm0 # prepare left channel for filter\n\t"
285     "movaps %xmm3,%xmm1 # save right channel not to get overwritten by filter algorithms\n\t"
286     );
287     FilterL.Apply4StepsSSE(&bqBase, &bqMain); // xmm0 input, xmm7 output
288     __asm__ __volatile__ (
289     "movaps %xmm1,%xmm0 # prepare right channel for filter\n\t"
290     "movaps %xmm7,%xmm1 # save filter output for left channel\n\t"
291     );
292     FilterR.Apply4StepsSSE(&bqBase, &bqMain); // xmm0 input, xmm7 output
293     __asm__ __volatile__ (
294     "movaps %xmm1,%xmm2 # result left channel -> xmm2\n\t"
295     "movaps %xmm7,%xmm3 # result right channel -> xmm3\n\t"
296     );
297     }
298     /* apply panorama and volume factors */
299     __asm__ __volatile__ (
300     "movss (%1),%%xmm0 # load pan left\n\t"
301     "movss (%2),%%xmm1 # load pan right\n\t"
302     "movaps (%0),%%xmm4 # load vca\n\t"
303     "shufps $0x00,%%xmm0,%%xmm0 # copy pan left to the other 3 cells\n\t"
304     "shufps $0x00,%%xmm1,%%xmm1 # copy pan right to the other 3 cells\n\t"
305     "mulps %%xmm2,%%xmm0 # left = sample_left * pan_left\n\t"
306     "mulps %%xmm3,%%xmm1 # right = sample_right * pan_right\n\t"
307     "mulps %%xmm4,%%xmm0 # left = vca * (sample_left * pan_left)\n\t"
308     "mulps %%xmm4,%%xmm1 # right = vca * (sample_right * pan_right)\n\t"
309     : /* no output */
310     : "r" (&Volume[ii]), /* %0 */
311     "r" (PanL), /* %1 */
312     "r" (PanR) /* %2 */
313     );
314     break;
315     }
316     }
317     /* mix the 4 samples to the output channels */
318     __asm__ __volatile__ (
319     "addps (%0),%%xmm0 # mix calculated sample(s) to output left\n\t"
320     "movaps %%xmm0,(%0) # output to left channel\n\t"
321     "addps (%1),%%xmm1 # mix calculated sample(s) to output right\n\t"
322     "movaps %%xmm1,(%1) # output to right channel\n\t"
323     : /* no output */
324     : "r" (&pOutL[ii]), /* %0 - must be 16 byte aligned ! */
325     "r" (&pOutR[ii]) /* %1 - must be 16 byte aligned ! */
326     );
327     }
328     }
329     }
330     };
331    
332     }} // namespace LinuxSampler::gig
333    
334     #endif // __LS_GIG_SYNTHESIZER_H__

  ViewVC Help
Powered by ViewVC