3 |
* LinuxSampler - modular, streaming capable sampler * |
* LinuxSampler - modular, streaming capable sampler * |
4 |
* * |
* * |
5 |
* Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck * |
* Copyright (C) 2003, 2004 by Benno Senoner and Christian Schoenebeck * |
6 |
|
* Copyright (C) 2005 Christian Schoenebeck * |
7 |
* * |
* * |
8 |
* This program is free software; you can redistribute it and/or modify * |
* 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 * |
* it under the terms of the GNU General Public License as published by * |
65 |
STEREO |
STEREO |
66 |
}; |
}; |
67 |
|
|
68 |
|
/** @brief Main Synthesis algorithms for the gig::Engine |
69 |
|
* |
70 |
|
* Implementation of the main synthesis algorithms of the Gigasampler |
71 |
|
* format capable sampler engine. This means resampling / interpolation |
72 |
|
* for pitching the audio signal, looping, filter and amplification. |
73 |
|
*/ |
74 |
template<implementation_t IMPLEMENTATION, channels_t CHANNELS, bool USEFILTER, bool INTERPOLATE, bool DOLOOP, bool CONSTPITCH> |
template<implementation_t IMPLEMENTATION, channels_t CHANNELS, bool USEFILTER, bool INTERPOLATE, bool DOLOOP, bool CONSTPITCH> |
75 |
class Synthesizer : public __RTMath<IMPLEMENTATION>, public LinuxSampler::Resampler<INTERPOLATE> { |
class Synthesizer : public __RTMath<IMPLEMENTATION>, public LinuxSampler::Resampler<INTERPOLATE> { |
76 |
|
|
77 |
|
// declarations of derived functions (see "Name lookup, |
78 |
|
// templates, and accessing members of base classes" in |
79 |
|
// the gcc manual for an explanation of why this is |
80 |
|
// needed). |
81 |
|
using __RTMath<IMPLEMENTATION>::Mul; |
82 |
|
using __RTMath<IMPLEMENTATION>::Float; |
83 |
|
using LinuxSampler::Resampler<INTERPOLATE>::GetNextSampleMonoCPP; |
84 |
|
using LinuxSampler::Resampler<INTERPOLATE>::GetNextSampleStereoCPP; |
85 |
|
#if ARCH_X86 |
86 |
|
using LinuxSampler::Resampler<INTERPOLATE>::GetNext4SamplesMonoMMXSSE; |
87 |
|
using LinuxSampler::Resampler<INTERPOLATE>::GetNext4SamplesStereoMMXSSE; |
88 |
|
#endif |
89 |
|
|
90 |
public: |
public: |
91 |
|
/** |
92 |
|
* Render audio for the current fragment for the given voice. |
93 |
|
* This is the toplevel method of this class. |
94 |
|
*/ |
95 |
template<typename VOICE_T> |
template<typename VOICE_T> |
96 |
inline static void SynthesizeFragment(VOICE_T& Voice, uint Samples, sample_t* pSrc, uint i) { |
inline static void SynthesizeFragment(VOICE_T& Voice, uint Samples, sample_t* pSrc, uint i) { |
97 |
|
const float panLeft = Mul(Voice.PanLeft, Voice.pEngineChannel->GlobalPanLeft); |
98 |
|
const float panRight = Mul(Voice.PanRight, Voice.pEngineChannel->GlobalPanRight); |
99 |
if (IMPLEMENTATION == ASM_X86_MMX_SSE) { |
if (IMPLEMENTATION == ASM_X86_MMX_SSE) { |
100 |
float fPos = (float) Voice.Pos; |
float fPos = (float) Voice.Pos; |
101 |
SynthesizeFragment(Voice, Samples, pSrc, i, Voice.pSample->LoopPlayCount, |
SynthesizeFragment(Voice, Samples, pSrc, i, Voice.pSample->LoopPlayCount, |
105 |
Voice.LoopCyclesLeft, |
Voice.LoopCyclesLeft, |
106 |
(void *)&fPos, |
(void *)&fPos, |
107 |
Voice.PitchBase, |
Voice.PitchBase, |
108 |
Voice.PitchBend); |
Voice.PitchBend, |
109 |
|
&panLeft, &panRight); |
110 |
|
#if ARCH_X86 |
111 |
if (INTERPOLATE) EMMS; |
if (INTERPOLATE) EMMS; |
112 |
|
#endif |
113 |
Voice.Pos = (double) fPos; |
Voice.Pos = (double) fPos; |
114 |
} else { |
} else { |
115 |
SynthesizeFragment(Voice, Samples, pSrc, i, Voice.pSample->LoopPlayCount, |
SynthesizeFragment(Voice, Samples, pSrc, i, Voice.pSample->LoopPlayCount, |
119 |
Voice.LoopCyclesLeft, |
Voice.LoopCyclesLeft, |
120 |
(void *)&Voice.Pos, |
(void *)&Voice.Pos, |
121 |
Voice.PitchBase, |
Voice.PitchBase, |
122 |
Voice.PitchBend); |
Voice.PitchBend, |
123 |
|
&panLeft, &panRight); |
124 |
} |
} |
125 |
} |
} |
126 |
|
|
127 |
//protected: |
//protected: |
128 |
|
|
129 |
|
/** |
130 |
|
* Render audio for the current fragment for the given voice. |
131 |
|
* Will be called by the toplevel SynthesizeFragment() method. |
132 |
|
*/ |
133 |
template<typename VOICE_T> |
template<typename VOICE_T> |
134 |
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) { |
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, const float* PanLeft, const float* PanRight) { |
135 |
const float loopEnd = Float(LoopEnd); |
const float loopEnd = Float(LoopEnd); |
136 |
const float PBbyPB = Mul(PitchBase, PitchBend); |
const float PBbyPB = Mul(PitchBase, PitchBend); |
137 |
const float f_LoopStart = Float(LoopStart); |
const float f_LoopStart = Float(LoopStart); |
142 |
while (i < Samples && LoopCyclesLeft) { |
while (i < Samples && LoopCyclesLeft) { |
143 |
if (CONSTPITCH) { |
if (CONSTPITCH) { |
144 |
const uint processEnd = Min(Samples, i + DiffToLoopEnd(loopEnd,Pos, PBbyPB) + 1); //TODO: instead of +1 we could also round up |
const uint processEnd = Min(Samples, i + DiffToLoopEnd(loopEnd,Pos, PBbyPB) + 1); //TODO: instead of +1 we could also round up |
145 |
while (i < processEnd) Synthesize(Voice, Pos, pSrc, i); |
while (i < processEnd) Synthesize(Voice, Pos, pSrc, i, PanLeft, PanRight); |
146 |
} |
} |
147 |
else Synthesize(Voice, Pos, pSrc, i); |
else Synthesize(Voice, Pos, pSrc, i, PanLeft, PanRight); |
148 |
if (WrapLoop(f_LoopStart, f_LoopSize, loopEnd, Pos)) LoopCyclesLeft--; |
if (WrapLoop(f_LoopStart, f_LoopSize, loopEnd, Pos)) LoopCyclesLeft--; |
149 |
} |
} |
150 |
// render on without loop |
// render on without loop |
151 |
while (i < Samples) Synthesize(Voice, Pos, pSrc, i); |
while (i < Samples) Synthesize(Voice, Pos, pSrc, i, PanLeft, PanRight); |
152 |
} |
} |
153 |
else { // render loop (endless loop) |
else { // render loop (endless loop) |
154 |
while (i < Samples) { |
while (i < Samples) { |
155 |
if (CONSTPITCH) { |
if (CONSTPITCH) { |
156 |
const uint processEnd = Min(Samples, i + DiffToLoopEnd(loopEnd, Pos, PBbyPB) + 1); //TODO: instead of +1 we could also round up |
const uint processEnd = Min(Samples, i + DiffToLoopEnd(loopEnd, Pos, PBbyPB) + 1); //TODO: instead of +1 we could also round up |
157 |
while (i < processEnd) Synthesize(Voice, Pos, pSrc, i); |
while (i < processEnd) Synthesize(Voice, Pos, pSrc, i, PanLeft, PanRight); |
158 |
} |
} |
159 |
else Synthesize(Voice, Pos, pSrc, i); |
else Synthesize(Voice, Pos, pSrc, i, PanLeft, PanRight); |
160 |
WrapLoop(f_LoopStart, f_LoopSize, loopEnd, Pos); |
WrapLoop(f_LoopStart, f_LoopSize, loopEnd, Pos); |
161 |
} |
} |
162 |
} |
} |
163 |
} |
} |
164 |
else { // no looping |
else { // no looping |
165 |
while (i < Samples) { Synthesize(Voice, Pos, pSrc, i);} |
while (i < Samples) { Synthesize(Voice, Pos, pSrc, i, PanLeft, PanRight);} |
166 |
} |
} |
167 |
} |
} |
168 |
|
|
169 |
|
/** |
170 |
|
* Atomicly render a piece for the voice. For the C++ |
171 |
|
* implementation this means rendering exactly one sample |
172 |
|
* point, whereas for the MMX/SSE implementation this means |
173 |
|
* rendering 4 sample points. |
174 |
|
*/ |
175 |
template<typename VOICE_T> |
template<typename VOICE_T> |
176 |
inline static void Synthesize(VOICE_T& Voice, void* Pos, sample_t* pSrc, uint& i) { |
inline static void Synthesize(VOICE_T& Voice, void* Pos, sample_t* pSrc, uint& i, const float* PanLeft, const float* PanRight) { |
177 |
Synthesize(pSrc, Pos, |
Synthesize(pSrc, Pos, |
178 |
Voice.pEngine->pSynthesisParameters[Event::destination_vco][i], |
Voice.pEngine->pSynthesisParameters[Event::destination_vco][i], |
179 |
Voice.pEngine->pOutputLeft, |
Voice.pEngineChannel->pOutputLeft, |
180 |
Voice.pEngine->pOutputRight, |
Voice.pEngineChannel->pOutputRight, |
181 |
i, |
i, |
182 |
Voice.pEngine->pSynthesisParameters[Event::destination_vca], |
Voice.pEngine->pSynthesisParameters[Event::destination_vca], |
183 |
&Voice.PanLeft, |
PanLeft, |
184 |
&Voice.PanRight, |
PanRight, |
185 |
Voice.FilterLeft, |
Voice.FilterLeft, |
186 |
Voice.FilterRight, |
Voice.FilterRight, |
187 |
Voice.pEngine->pBasicFilterParameters[i], |
Voice.pEngine->pBasicFilterParameters[i], |
188 |
Voice.pEngine->pMainFilterParameters[i]); |
Voice.pEngine->pMainFilterParameters[i]); |
189 |
} |
} |
190 |
|
|
191 |
|
/** |
192 |
|
* Returns the difference to the sample's loop end. |
193 |
|
*/ |
194 |
inline static int DiffToLoopEnd(const float& LoopEnd, const void* Pos, const float& Pitch) { |
inline static int DiffToLoopEnd(const float& LoopEnd, const void* Pos, const float& Pitch) { |
195 |
switch (IMPLEMENTATION) { |
switch (IMPLEMENTATION) { |
196 |
// pure C++ implementation (thus platform independent) |
// pure C++ implementation (thus platform independent) |
216 |
} |
} |
217 |
} |
} |
218 |
|
|
219 |
|
/** |
220 |
|
* This method handles looping of the RAM playback part of the |
221 |
|
* sample, thus repositioning the playback position once the |
222 |
|
* loop limit was reached. Note: looping of the disk streaming |
223 |
|
* part is handled by libgig (ReadAndLoop() method which will |
224 |
|
* be called by the DiskThread). |
225 |
|
*/ |
226 |
inline static int WrapLoop(const float& LoopStart, const float& LoopSize, const float& LoopEnd, void* vPos) { |
inline static int WrapLoop(const float& LoopStart, const float& LoopSize, const float& LoopEnd, void* vPos) { |
227 |
switch (IMPLEMENTATION) { |
switch (IMPLEMENTATION) { |
228 |
// pure C++ implementation (thus platform independent) |
// pure C++ implementation (thus platform independent) |
234 |
} |
} |
235 |
#if ARCH_X86 |
#if ARCH_X86 |
236 |
case ASM_X86_MMX_SSE: { |
case ASM_X86_MMX_SSE: { |
237 |
int result; |
int result = 0; |
238 |
__asm__ __volatile__ ( |
__asm__ __volatile__ ( |
239 |
"movss (%2), %%xmm0 # load LoopEnd\n\t" |
"movss (%2), %%xmm0 # load LoopEnd\n\t" |
240 |
"movss (%1), %%xmm1 # load Pos\n\t" |
"movss (%1), %%xmm1 # load Pos\n\t" |
241 |
"comiss %%xmm0, %%xmm1 # LoopEnd <> Pos\n\t" |
"comiss %%xmm0, %%xmm1 # LoopEnd <> Pos\n\t" |
|
"movl $0,%%eax # result = 0\n\t" |
|
242 |
"jb 1f # jump if no work needs to be done\n\t" |
"jb 1f # jump if no work needs to be done\n\t" |
243 |
"movss (%3), %%xmm2 # load LoopSize\n\t" |
"movss (%3), %%xmm2 # load LoopSize\n\t" |
244 |
"subss %%xmm0, %%xmm1 # Pos - LoopEnd\n\t" |
"subss %%xmm0, %%xmm1 # Pos - LoopEnd\n\t" |
245 |
//now the fmodf |
//now the fmodf |
246 |
"movss %%xmm1, %%xmm3 # xmm3 = (Pos - LoopEnd)\n\t" |
"movss %%xmm1, %%xmm3 # xmm3 = (Pos - LoopEnd)\n\t" |
247 |
"divss %%xmm2, %%xmm1 # (Pos - LoopEnd) / LoopSize\n\t" |
"divss %%xmm2, %%xmm1 # (Pos - LoopEnd) / LoopSize\n\t" |
248 |
"cvttss2si %%xmm1, %%eax # convert to int\n\t" |
"cvttss2si %%xmm1, %2 # convert to int\n\t" |
249 |
"cvtsi2ss %%eax, %%xmm1 # convert back to float\n\t" |
"cvtsi2ss %2, %%xmm1 # convert back to float\n\t" |
250 |
"movss (%4), %%xmm0 # load LoopStart\n\t" |
"movss (%4), %%xmm0 # load LoopStart\n\t" |
251 |
"mulss %%xmm2, %%xmm1 # LoopSize * int((Pos-LoopEnd)/LoopSize)\n\t" |
"mulss %%xmm2, %%xmm1 # LoopSize * int((Pos-LoopEnd)/LoopSize)\n\t" |
252 |
"subss %%xmm1, %%xmm3 # xmm2 = fmodf(Pos - LoopEnd, LoopSize)\n\t" |
"subss %%xmm1, %%xmm3 # xmm2 = fmodf(Pos - LoopEnd, LoopSize)\n\t" |
253 |
//done with fmodf |
//done with fmodf |
254 |
"addss %%xmm0, %%xmm3 # add LoopStart\n\t" |
"addss %%xmm0, %%xmm3 # add LoopStart\n\t" |
255 |
"movss %%xmm3, (%1) # update Pos\n\t" |
"movss %%xmm3, (%1) # update Pos\n\t" |
256 |
"movl $1,%%eax # result = 1\n\t" |
"movl $1, (%0) # result = 1\n\t" |
257 |
".balign 16 \n\t" |
".balign 16 \n\t" |
258 |
"1:\n\t" |
"1:\n\t" |
259 |
: "=a" (result) /* %0 */ |
:: "r" (&result), /* %0 */ |
260 |
: "r" (vPos), /* %1 */ |
"r" (vPos), /* %1 */ |
261 |
"r" (&LoopEnd), /* %2 */ |
"r" (&LoopEnd), /* %2 */ |
262 |
"r" (&LoopSize), /* %3 */ |
"r" (&LoopSize), /* %3 */ |
263 |
"r" (&LoopStart) /* %4 */ |
"r" (&LoopStart) /* %4 */ |
268 |
} |
} |
269 |
} |
} |
270 |
|
|
271 |
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) { |
/** |
272 |
|
* Atomicly render a piece for the voice. For the C++ |
273 |
|
* implementation this means rendering exactly one sample |
274 |
|
* point, whereas for the MMX/SSE implementation this means |
275 |
|
* rendering 4 sample points. |
276 |
|
*/ |
277 |
|
inline static void Synthesize(sample_t* pSrc, void* Pos, float& Pitch, float* pOutL, float* pOutR, uint& i, float* Volume, const float* PanL, const float* PanR, Filter& FilterL, Filter& FilterR, biquad_param_t& bqBase, biquad_param_t& bqMain) { |
278 |
switch (IMPLEMENTATION) { |
switch (IMPLEMENTATION) { |
279 |
// pure C++ implementation (thus platform independent) |
// pure C++ implementation (thus platform independent) |
280 |
case CPP: { |
case CPP: { |