Mirrors/overte
3
0
Fork 0
mirror of https://github.com/overte-org/overte.git synced 2025-04-20 04:44:11 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

removed SIMD stuff from AudioMixer for all builds

Browse source
This commit is contained in:
wangyix 2014-07-15 14:43:08 -07:00
parent 871aa2790a
commit d5a30ff6ee
1 changed files with 2 additions and 109 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

111
assignment-client/src/audio/AudioMixer.cpp
View file

@ -231,29 +231,10 @@ void AudioMixer::addBufferToMixForListeningNodeWithBuffer(PositionalAudioRingBuf
delayBufferSample[0] = correctBufferSample[0] * weakChannelAmplitudeRatio;
delayBufferSample[1] = correctBufferSample[1] * weakChannelAmplitudeRatio;
#ifdef _WIN32
_clientSamples[s + goodChannelOffset] += correctBufferSample[0];
_clientSamples[s + goodChannelOffset + SINGLE_STEREO_OFFSET] += correctBufferSample[1];
_clientSamples[delayedChannelIndex] += delayBufferSample[0];
_clientSamples[delayedChannelIndex + SINGLE_STEREO_OFFSET] += delayBufferSample[1];
#else
__m64 bufferSamples = _mm_set_pi16(_clientSamples[s + goodChannelOffset],
_clientSamples[s + goodChannelOffset + SINGLE_STEREO_OFFSET],
_clientSamples[delayedChannelIndex],
_clientSamples[delayedChannelIndex + SINGLE_STEREO_OFFSET]);
__m64 addedSamples = _mm_set_pi16(correctBufferSample[0], correctBufferSample[1],
delayBufferSample[0], delayBufferSample[1]);
// perform the MMX add (with saturation) of two correct and delayed samples
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addedSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
// assign the results from the result of the mmx arithmetic
_clientSamples[s + goodChannelOffset] = shortResults[3];
_clientSamples[s + goodChannelOffset + SINGLE_STEREO_OFFSET] = shortResults[2];
_clientSamples[delayedChannelIndex] = shortResults[1];
_clientSamples[delayedChannelIndex + SINGLE_STEREO_OFFSET] = shortResults[0];
#endif
}
// The following code is pretty gross and redundant, but AFAIK it's the best way to avoid
@ -273,98 +254,10 @@ void AudioMixer::addBufferToMixForListeningNodeWithBuffer(PositionalAudioRingBuf
delayNextOutputStart = bufferStart + ringBufferSampleCapacity - numSamplesDelay;
}
int i = 0;
while (i + 3 < numSamplesDelay) {
// handle the first cases where we can MMX add four samples at once
for (int i = 0; i < numSamplesDelay; i++) {
int parentIndex = i * 2;
#ifdef _WIN32
_clientSamples[parentIndex + delayedChannelOffset] += delayNextOutputStart[i] * attenuationAndWeakChannelRatio;
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] += delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio;
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset] += delayNextOutputStart[i + 2] * attenuationAndWeakChannelRatio;
_clientSamples[parentIndex + TRIPLE_STEREO_OFFSET + delayedChannelOffset] += delayNextOutputStart[i + 3] * attenuationAndWeakChannelRatio;
#else
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset],
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset],
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset],
_clientSamples[parentIndex + TRIPLE_STEREO_OFFSET + delayedChannelOffset]);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 2] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 3] * attenuationAndWeakChannelRatio);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[2];
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[1];
_clientSamples[parentIndex + TRIPLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[0];
#endif
// push the index
i += 4;
}
int parentIndex = i * 2;
if (i + 2 < numSamplesDelay) {
// MMX add only three delayed samples
#ifdef _WIN32
_clientSamples[parentIndex + delayedChannelOffset] += delayNextOutputStart[i] * attenuationAndWeakChannelRatio;
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] += delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio;
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset] += delayNextOutputStart[i + 2] * attenuationAndWeakChannelRatio;
#else
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset],
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset],
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset],
0);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 2] * attenuationAndWeakChannelRatio,
0);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[2];
_clientSamples[parentIndex + DOUBLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[1];
#endif
} else if (i + 1 < numSamplesDelay) {
#ifdef _WIN32
_clientSamples[parentIndex + delayedChannelOffset] += delayNextOutputStart[i] * attenuationAndWeakChannelRatio;
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] += delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio;
#else
// MMX add two delayed samples
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset],
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset],
0, 0);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio,
delayNextOutputStart[i + 1] * attenuationAndWeakChannelRatio, 0, 0);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
_clientSamples[parentIndex + SINGLE_STEREO_OFFSET + delayedChannelOffset] = shortResults[2];
#endif
} else if (i < numSamplesDelay) {
// MMX add a single delayed sample
#ifdef _WIN32
_clientSamples[parentIndex + delayedChannelOffset] += delayNextOutputStart[i] * attenuationAndWeakChannelRatio;
#else
__m64 bufferSamples = _mm_set_pi16(_clientSamples[parentIndex + delayedChannelOffset], 0, 0, 0);
__m64 addSamples = _mm_set_pi16(delayNextOutputStart[i] * attenuationAndWeakChannelRatio, 0, 0, 0);
__m64 mmxResult = _mm_adds_pi16(bufferSamples, addSamples);
int16_t* shortResults = reinterpret_cast<int16_t*>(&mmxResult);
_clientSamples[parentIndex + delayedChannelOffset] = shortResults[3];
#endif
}
}
} else {