28 #ifndef INCLUDE_VOLK_VOLK_AVX2_INTRINSICS_H_
29 #define INCLUDE_VOLK_VOLK_AVX2_INTRINSICS_H_
31 #include <immintrin.h>
35 const __m128i zeros = _mm_set1_epi8(0x00);
36 const __m128i sign_extract = _mm_set1_epi8(0x80);
37 const __m256i shuffle_mask = _mm256_setr_epi8(0xff,
69 __m256i sign_bits = _mm256_setzero_si256();
71 fbits = _mm_cmpgt_epi8(fbits, zeros);
72 fbits = _mm_and_si128(fbits, sign_extract);
73 sign_bits = _mm256_insertf128_si256(sign_bits, fbits, 0);
74 sign_bits = _mm256_insertf128_si256(sign_bits, fbits, 1);
75 sign_bits = _mm256_shuffle_epi8(sign_bits, shuffle_mask);
77 return _mm256_castsi256_ps(sign_bits);
90 llr0 = _mm256_xor_ps(llr0, sign_mask);
91 __m256 dst = _mm256_add_ps(llr0, llr1);
96 const __m256 cplxValue1)
98 const __m256i idx = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
99 const __m256 squared0 = _mm256_mul_ps(cplxValue0, cplxValue0);
100 const __m256 squared1 = _mm256_mul_ps(cplxValue1, cplxValue1);
101 const __m256 complex_result = _mm256_hadd_ps(squared0, squared1);
102 return _mm256_permutevar8x32_ps(complex_result, idx);
106 const __m256 symbols1,
107 const __m256 points0,
108 const __m256 points1,
116 const __m256 diff0 = _mm256_sub_ps(symbols0, points0);
117 const __m256 diff1 = _mm256_sub_ps(symbols1, points1);
119 return _mm256_mul_ps(norms, scalar);