3609     __m128i a16, b16, r16_1, r16_2;  in vmulq_s8()  local
3619     r16_2 = _mm_mullo_epi16 (a16, b16);  in vmulq_s8()
3621     r16_2 = _mm_shuffle_epi8 (r16_2, *(__m128i*)mask8_16_even_odd); //return to 8 bit  in vmulq_s8()
3623     return _mm_unpacklo_epi64(r16_1,  r16_2);  in vmulq_s8()
3640     __m128i maskff, a16, b16, r16_1, r16_2;  in vmulq_u8()  local
3652     r16_2 = _mm_mullo_epi16 (a16, b16);  in vmulq_u8()
3653     r16_2 = _mm_and_si128(r16_2, maskff); //to avoid saturation  in vmulq_u8()
3654     return _mm_packus_epi16 (r16_1,  r16_2);  in vmulq_u8()
3881     __m128i b16, c16, r16_1, a_2,r16_2;  in vmlaq_s8()  local
3894     r16_2 = _mm_mullo_epi16 (b16, c16);  in vmlaq_s8()
3895     r16_2 = _mm_shuffle_epi8 (r16_2, *(__m128i*) mask8_16_even_odd);  in vmlaq_s8()
3896     r16_2 = _mm_add_epi8(r16_2, a_2);  in vmlaq_s8()
3897     return _mm_unpacklo_epi64(r16_1,r16_2);  in vmlaq_s8()
3930     __m128i b16, c16, r16_1, a_2, r16_2;  in vmlaq_u8()  local
3943     r16_2 = _mm_mullo_epi16 (b16, c16);  in vmlaq_u8()
3944     r16_2 = _mm_shuffle_epi8 (r16_2, *(__m128i*) mask8_16_even_odd);  in vmlaq_u8()
3945     r16_2 = _mm_add_epi8(r16_2, a_2);  in vmlaq_u8()
3946     return _mm_unpacklo_epi64(r16_1,r16_2);  in vmlaq_u8()
4075     __m128i b16, c16, r16_1, a_2, r16_2;  in vmlsq_s8()  local
4088     r16_2 = _mm_mullo_epi16 (b16, c16);  in vmlsq_s8()
4089     r16_2 = _mm_shuffle_epi8 (r16_2, *(__m128i*) mask8_16_even_odd);  in vmlsq_s8()
4090     r16_2 = _mm_sub_epi8 (a_2, r16_2);  in vmlsq_s8()
4091     return _mm_unpacklo_epi64(r16_1,r16_2);  in vmlsq_s8()
4123     __m128i b16, c16, r16_1, a_2, r16_2;  in vmlsq_u8()  local
4136     r16_2 = _mm_mullo_epi16 (b16, c16);  in vmlsq_u8()
4137     r16_2 = _mm_shuffle_epi8 (r16_2, *(__m128i*) mask8_16_even_odd);  in vmlsq_u8()
4138     r16_2 = _mm_sub_epi8(a_2, r16_2);  in vmlsq_u8()
4139     return _mm_unpacklo_epi64(r16_1,r16_2);  in vmlsq_u8()
6422     __m128i r16_1, r16_2;  in vpaddlq_s8()  local
6425     r16_2 = _mm_shuffle_epi32 (a, _SWAP_HI_LOW32);  in vpaddlq_s8()
6426     r16_2 = _MM_CVTEPI8_EPI16 (r16_2);  in vpaddlq_s8()
6427     return _mm_hadd_epi16 (r16_1, r16_2);  in vpaddlq_s8()
6457     __m128i r16_1, r16_2;  in vpaddlq_u8()  local
6460     r16_2 = _mm_shuffle_epi32 (a, _SWAP_HI_LOW32);  in vpaddlq_u8()
6461     r16_2 = _MM_CVTEPU8_EPI16 (r16_2);  in vpaddlq_u8()
6462     return _mm_hadd_epi16 (r16_1, r16_2);  in vpaddlq_u8()