IntHalfbandFilterSTIntrinsics< HBFilterOrder > Class Template Reference

#include <inthalfbandfiltersti.h>

Static Public Member Functions
static void	work (int32_t samples[HBFilterOrder][2], int32_t &iEvenAcc, int32_t &qEvenAcc, int32_t &iOddAcc, int32_t &qOddAcc)
static void	workNA (int ptr, int32_t samples[HBFilterOrder *2][2], int32_t &iEvenAcc, int32_t &qEvenAcc, int32_t &iOddAcc, int32_t &qOddAcc)

Detailed Description

template<uint32_t HBFilterOrder>
class IntHalfbandFilterSTIntrinsics< HBFilterOrder >

Definition at line 35 of file inthalfbandfiltersti.h.

Member Function Documentation

work()

template<uint32_t HBFilterOrder>

static void IntHalfbandFilterSTIntrinsics< HBFilterOrder >::work ( int32_t  samples[HBFilterOrder][2], int32_t &  iEvenAcc, int32_t &  qEvenAcc, int32_t &  iOddAcc, int32_t &  qOddAcc  )

inlinestatic

Definition at line 38 of file inthalfbandfiltersti.h.

References __attribute__, and i.

    {
#if defined(USE_SSE4_1)
        int a = HBFIRFilterTraits<HBFilterOrder>::hbOrder - 2; // tip
        int b = 0; // tail
        const int *h = (const int*) HBFIRFilterTraits<HBFilterOrder>::hbCoeffs;
        __m128i sum = _mm_setzero_si128();
        __m128i shh, sa, sb;
        int32_t sums[4] __attribute__ ((aligned (16)));

        for (int i = 0; i < HBFIRFilterTraits<HBFilterOrder>::hbOrder / 16; i++)
        {
            shh = _mm_set_epi32(h[4*i], h[4*i], h[4*i], h[4*i]);
            sa = _mm_load_si128((__m128i*) &(samples[a][0])); // Ei,Eq,Oi,Oq
            sb = _mm_load_si128((__m128i*) &(samples[b][0]));
            sum = _mm_add_epi32(sum, _mm_mullo_epi32(_mm_add_epi32(sa, sb), shh));
            a -= 2;
            b += 2;
            shh = _mm_set_epi32(h[4*i+1], h[4*i+1], h[4*i+1], h[4*i+1]);
            sa = _mm_load_si128((__m128i*) &(samples[a][0])); // Ei,Eq,Oi,Oq
            sb = _mm_load_si128((__m128i*) &(samples[b][0]));
            sum = _mm_add_epi32(sum, _mm_mullo_epi32(_mm_add_epi32(sa, sb), shh));
            a -= 2;
            b += 2;
            shh = _mm_set_epi32(h[4*i+2], h[4*i+2], h[4*i+2], h[4*i+2]);
            sa = _mm_load_si128((__m128i*) &(samples[a][0])); // Ei,Eq,Oi,Oq
            sb = _mm_load_si128((__m128i*) &(samples[b][0]));
            sum = _mm_add_epi32(sum, _mm_mullo_epi32(_mm_add_epi32(sa, sb), shh));
            a -= 2;
            b += 2;
            shh = _mm_set_epi32(h[4*i+3], h[4*i+3], h[4*i+3], h[4*i+3]);
            sa = _mm_load_si128((__m128i*) &(samples[a][0])); // Ei,Eq,Oi,Oq
            sb = _mm_load_si128((__m128i*) &(samples[b][0]));
            sum = _mm_add_epi32(sum, _mm_mullo_epi32(_mm_add_epi32(sa, sb), shh));
            a -= 2;
            b += 2;
        }

        // Extract values from sum vector
        _mm_store_si128((__m128i*) sums, sum);
        iEvenAcc = sums[0];
        qEvenAcc = sums[1];
        iOddAcc = sums[2];
        qOddAcc = sums[3];
#endif
    }

workNA()

template<uint32_t HBFilterOrder>

static void IntHalfbandFilterSTIntrinsics< HBFilterOrder >::workNA ( int  ptr, int32_t  samples[HBFilterOrder *2][2], int32_t &  iEvenAcc, int32_t &  qEvenAcc, int32_t &  iOddAcc, int32_t &  qOddAcc  )

inlinestatic

Definition at line 89 of file inthalfbandfiltersti.h.

References __attribute__, and i.

Referenced by IntHalfbandFilterST< HBFilterOrder >::doFIR().

    {
#if defined(USE_SSE4_1)
        int a = ptr + HBFIRFilterTraits<HBFilterOrder>::hbOrder - 2; // tip
        int b = ptr + 0; // tail
        const int *h = (const int*) HBFIRFilterTraits<HBFilterOrder>::hbCoeffs;
        __m128i sum = _mm_setzero_si128();
        __m128i shh, sa, sb;
        int32_t sums[4] __attribute__ ((aligned (16)));

        for (int i = 0; i < HBFIRFilterTraits<HBFilterOrder>::hbOrder / 16; i++)
        {
            shh = _mm_set_epi32(h[4*i], h[4*i], h[4*i], h[4*i]);
            sa = _mm_loadu_si128((__m128i*) &(samples[a][0])); // Ei,Eq,Oi,Oq
            sb = _mm_loadu_si128((__m128i*) &(samples[b][0]));
            sum = _mm_add_epi32(sum, _mm_mullo_epi32(_mm_add_epi32(sa, sb), shh));
            a -= 2;
            b += 2;
            shh = _mm_set_epi32(h[4*i+1], h[4*i+1], h[4*i+1], h[4*i+1]);
            sa = _mm_loadu_si128((__m128i*) &(samples[a][0])); // Ei,Eq,Oi,Oq
            sb = _mm_loadu_si128((__m128i*) &(samples[b][0]));
            sum = _mm_add_epi32(sum, _mm_mullo_epi32(_mm_add_epi32(sa, sb), shh));
            a -= 2;
            b += 2;
            shh = _mm_set_epi32(h[4*i+2], h[4*i+2], h[4*i+2], h[4*i+2]);
            sa = _mm_loadu_si128((__m128i*) &(samples[a][0])); // Ei,Eq,Oi,Oq
            sb = _mm_loadu_si128((__m128i*) &(samples[b][0]));
            sum = _mm_add_epi32(sum, _mm_mullo_epi32(_mm_add_epi32(sa, sb), shh));
            a -= 2;
            b += 2;
            shh = _mm_set_epi32(h[4*i+3], h[4*i+3], h[4*i+3], h[4*i+3]);
            sa = _mm_loadu_si128((__m128i*) &(samples[a][0])); // Ei,Eq,Oi,Oq
            sb = _mm_loadu_si128((__m128i*) &(samples[b][0]));
            sum = _mm_add_epi32(sum, _mm_mullo_epi32(_mm_add_epi32(sa, sb), shh));
            a -= 2;
            b += 2;
        }

        // Extract values from sum vector
        _mm_store_si128((__m128i*) sums, sum);
        iEvenAcc = sums[0];
        qEvenAcc = sums[1];
        iOddAcc = sums[2];
        qOddAcc = sums[3];
#endif
    }

Here is the caller graph for this function:

---

The documentation for this class was generated from the following file:

• sdrbase/dsp/inthalfbandfiltersti.h