int OfdmGenerator::process(Buffer* const dataIn, Buffer* dataOut)
{
PDEBUG("OfdmGenerator::process(dataIn: %p, dataOut: %p)\n",
dataIn, dataOut);
dataOut->setLength(myNbSymbols * mySpacing * sizeof(complexf));
FFT_TYPE* in = reinterpret_cast<FFT_TYPE*>(dataIn->getData());
FFT_TYPE* out = reinterpret_cast<FFT_TYPE*>(dataOut->getData());
size_t sizeIn = dataIn->getLength() / sizeof(complexf);
size_t sizeOut = dataOut->getLength() / sizeof(complexf);
if (sizeIn != myNbSymbols * myNbCarriers) {
PDEBUG("Nb symbols: %zu\n", myNbSymbols);
PDEBUG("Nb carriers: %zu\n", myNbCarriers);
PDEBUG("Spacing: %zu\n", mySpacing);
PDEBUG("\n%zu != %zu\n", sizeIn, myNbSymbols * myNbCarriers);
throw std::runtime_error(
"OfdmGenerator::process input size not valid!");
}
if (sizeOut != myNbSymbols * mySpacing) {
PDEBUG("Nb symbols: %zu\n", myNbSymbols);
PDEBUG("Nb carriers: %zu\n", myNbCarriers);
PDEBUG("Spacing: %zu\n", mySpacing);
PDEBUG("\n%zu != %zu\n", sizeIn, myNbSymbols * mySpacing);
throw std::runtime_error(
"OfdmGenerator::process output size not valid!");
}
#if USE_FFTW
// No SIMD/no-SIMD distinction, it's too early to optimize anything
for (size_t i = 0; i < myNbSymbols; ++i) {
myFftIn[0][0] = 0;
myFftIn[0][1] = 0;
bzero(&myFftIn[myZeroDst], myZeroSize * sizeof(FFT_TYPE));
memcpy(&myFftIn[myPosDst], &in[myPosSrc],
myPosSize * sizeof(FFT_TYPE));
memcpy(&myFftIn[myNegDst], &in[myNegSrc],
myNegSize * sizeof(FFT_TYPE));
fftwf_execute(myFftPlan);
memcpy(out, myFftOut, mySpacing * sizeof(FFT_TYPE));
in += myNbCarriers;
out += mySpacing;
}
#else
# ifdef USE_SIMD
for (size_t i = 0, j = 0; i < sizeIn; ) {
// Pack 4 fft operations
typedef struct {
float r[4];
float i[4];
} fft_data;
assert(sizeof(FFT_TYPE) == sizeof(fft_data));
complexf *cplxIn = (complexf*)in;
complexf *cplxOut = (complexf*)out;
fft_data *dataBuffer = (fft_data*)myFftBuffer;
FFT_REAL(myFftBuffer[0]) = _mm_setzero_ps();
FFT_IMAG(myFftBuffer[0]) = _mm_setzero_ps();
for (size_t k = 0; k < myZeroSize; ++k) {
FFT_REAL(myFftBuffer[myZeroDst + k]) = _mm_setzero_ps();
FFT_IMAG(myFftBuffer[myZeroDst + k]) = _mm_setzero_ps();
}
for (int k = 0; k < 4; ++k) {
if (i < sizeIn) {
for (size_t l = 0; l < myPosSize; ++l) {
dataBuffer[myPosDst + l].r[k] = cplxIn[i + myPosSrc + l].real();
dataBuffer[myPosDst + l].i[k] = cplxIn[i + myPosSrc + l].imag();
}
for (size_t l = 0; l < myNegSize; ++l) {
dataBuffer[myNegDst + l].r[k] = cplxIn[i + myNegSrc + l].real();
dataBuffer[myNegDst + l].i[k] = cplxIn[i + myNegSrc + l].imag();
}
i += myNbCarriers;
}
else {
for (size_t l = 0; l < myNbCarriers; ++l) {
dataBuffer[l].r[k] = 0.0f;
dataBuffer[l].i[k] = 0.0f;
}
}
}
kiss_fft(myFftPlan, myFftBuffer, myFftBuffer);
for (int k = 0; k < 4; ++k) {
if (j < sizeOut) {
for (size_t l = 0; l < mySpacing; ++l) {
cplxOut[j + l] = complexf(dataBuffer[l].r[k], dataBuffer[l].i[k]);
}
j += mySpacing;
}
}
}
# else
for (size_t i = 0; i < myNbSymbols; ++i) {
FFT_REAL(myFftBuffer[0]) = 0;
FFT_IMAG(myFftBuffer[0]) = 0;
bzero(&myFftBuffer[myZeroDst], myZeroSize * sizeof(FFT_TYPE));
memcpy(&myFftBuffer[myPosDst], &in[myPosSrc],
myPosSize * sizeof(FFT_TYPE));
memcpy(&myFftBuffer[myNegDst], &in[myNegSrc],
myNegSize * sizeof(FFT_TYPE));
kiss_fft(myFftPlan, myFftBuffer, out);
in += myNbCarriers;
out += mySpacing;
}
# endif
#endif
return sizeOut;
}
int Resampler::process(Buffer* const dataIn, Buffer* dataOut)
{
PDEBUG("Resampler::process(dataIn: %p, dataOut: %p)\n",
dataIn, dataOut);
dataOut->setLength(dataIn->getLength() * L / M);
FFT_TYPE* in = reinterpret_cast<FFT_TYPE*>(dataIn->getData());
FFT_TYPE* out = reinterpret_cast<FFT_TYPE*>(dataOut->getData());
size_t sizeIn = dataIn->getLength() / sizeof(complexf);
for (size_t i = 0, j = 0; i < sizeIn; i += myFftSizeIn / 2, j += myFftSizeOut / 2) {
memcpy(myFftIn, myBufferIn, myFftSizeIn / 2 * sizeof(FFT_TYPE));
memcpy(myFftIn + (myFftSizeIn / 2), in + i, myFftSizeIn / 2 * sizeof(FFT_TYPE));
memcpy(myBufferIn, in + i, myFftSizeIn / 2 * sizeof(FFT_TYPE));
for (size_t k = 0; k < myFftSizeIn; ++k) {
FFT_REAL(myFftIn[k]) *= myWindow[k];
FFT_IMAG(myFftIn[k]) *= myWindow[k];
}
fftwf_execute(myFftPlan1);
if (myFftSizeOut > myFftSizeIn) {
memset(myBack, 0, myFftSizeOut * sizeof(FFT_TYPE));
memcpy(myBack, myFront, myFftSizeIn / 2 * sizeof(FFT_TYPE));
memcpy(&myBack[myFftSizeOut - (myFftSizeIn / 2)],
&myFront[myFftSizeIn / 2],
myFftSizeIn / 2 * sizeof(FFT_TYPE));
// Copy input Fs
FFT_REAL(myBack[myFftSizeIn / 2]) =
FFT_REAL(myFront[myFftSizeIn / 2]);
FFT_IMAG(myBack[myFftSizeIn / 2]) =
FFT_IMAG(myFront[myFftSizeIn / 2]);
} else {
memcpy(myBack, myFront, myFftSizeOut / 2 * sizeof(FFT_TYPE));
memcpy(&myBack[myFftSizeOut / 2],
&myFront[myFftSizeIn - (myFftSizeOut / 2)],
myFftSizeOut / 2 * sizeof(FFT_TYPE));
// Average output Fs from input
FFT_REAL(myBack[myFftSizeOut / 2]) +=
FFT_REAL(myFront[myFftSizeOut / 2]);
FFT_IMAG(myBack[myFftSizeOut / 2]) +=
FFT_IMAG(myFront[myFftSizeOut / 2]);
FFT_REAL(myBack[myFftSizeOut / 2]) *= 0.5f;
FFT_IMAG(myBack[myFftSizeOut / 2]) *= 0.5f;
}
for (size_t k = 0; k < myFftSizeOut; ++k) {
FFT_REAL(myBack[k]) *= myFactor;
FFT_IMAG(myBack[k]) *= myFactor;
}
fftwf_execute(myFftPlan2);
for (size_t k = 0; k < myFftSizeOut / 2; ++k) {
FFT_REAL(out[j + k]) = myBufferOut[k].real() + FFT_REAL(myFftOut[k]);
FFT_IMAG(out[j + k]) = myBufferOut[k].imag() + FFT_IMAG(myFftOut[k]);
}
memcpy(myBufferOut, myFftOut + (myFftSizeOut / 2), (myFftSizeOut / 2) * sizeof(FFT_TYPE));
}
return 1;
}
