void DoubleToComplex(double *srcI, double *srcQ, Complex *dst, const unsigned int len)
{
__m256d avxR_D, avxI_D, avxX_D, avxY_D, avxA_D, avxB_D;
__m128 avxA, avxB;
#if 1
__m256 avxD;
#endif
for (unsigned int i=0; i+4<=len; i+=4) {
avxR_D = _mm256_loadu_pd(srcI + i);
avxI_D = _mm256_loadu_pd(srcQ + i);
avxX_D = _mm256_unpacklo_pd(avxR_D, avxI_D); //swizzle
avxY_D = _mm256_unpackhi_pd(avxR_D, avxI_D);
avxA_D = _mm256_permute2f128_pd(avxX_D, avxY_D, 0x20);
avxB_D = _mm256_permute2f128_pd(avxX_D, avxY_D, 0x31);
avxA = _mm256_cvtpd_ps(avxA_D); //double to float
avxB = _mm256_cvtpd_ps(avxB_D);
#if 0
avxD = _mm256_castps128_ps256(avxA);
avxD = _mm256_insertf128_ps(avxD, avxB, 1);
_mm256_storeu_ps((float*)(dst+i), avxD);
#else
_mm_maskstore_ps((float*)(dst+i), _mm_set_epi32(SET_1, SET_1, SET_1, SET_1), avxA);
_mm_maskstore_ps((float*)(dst+i+2), _mm_set_epi32(SET_1, SET_1, SET_1, SET_1), avxB);
#endif
}
for (unsigned int i=len-(len & 0x03); i<len; ++i) {
dst[i].m_real = static_cast<float>(srcI[i]);
dst[i].m_imag = static_cast<float>(srcQ[i]);
}
}
inline void transpose_4x4block_AVX_64(double* A, double* B, const size_t lda,
const size_t ldb) {
__m256d row0 = _mm256_load_pd(&A[0*ldb]);
__m256d row1 = _mm256_load_pd(&A[1*ldb]);
__m256d row2 = _mm256_load_pd(&A[2*ldb]);
__m256d row3 = _mm256_load_pd(&A[3*ldb]);
__m256d tmp3, tmp2, tmp1, tmp0;
tmp0 = _mm256_unpacklo_pd(row0, row1);
tmp1 = _mm256_unpackhi_pd(row0, row1);
tmp2 = _mm256_unpacklo_pd(row2, row3);
tmp3 = _mm256_unpackhi_pd(row2, row3);
row0 = _mm256_permute2f128_pd(tmp0, tmp2, 0x20);
row1 = _mm256_permute2f128_pd(tmp1, tmp3, 0x20);
row2 = _mm256_permute2f128_pd(tmp0, tmp2, 0x31);
row3 = _mm256_permute2f128_pd(tmp1, tmp3, 0x31);
_mm256_store_pd(&B[0*lda], row0);
_mm256_store_pd(&B[1*lda], row1);
_mm256_store_pd(&B[2*lda], row2);
_mm256_store_pd(&B[3*lda], row3);
}
