121      *  b0 b1 b2 b3  in ih264_ihadamard_scaling_4x4_sse42()
127     temp1 = _mm_unpackhi_epi32(src_r0, src_r1);                  //a2 b2 a3 b3  in ih264_ihadamard_scaling_4x4_sse42()
132     src_r3 = _mm_unpackhi_epi64(temp1, temp3);                    //a3 b3 c3 d3  in ih264_ihadamard_scaling_4x4_sse42()
152      *  a3 b3 c3 d3  in ih264_ihadamard_scaling_4x4_sse42()
155     temp2 = _mm_unpacklo_epi32(src_r2, src_r3);                  //a2 a3 b2 b3  in ih264_ihadamard_scaling_4x4_sse42()
159     src_r1 = _mm_unpackhi_epi64(temp0, temp2);                   //b0 b1 b2 b3  in ih264_ihadamard_scaling_4x4_sse42()
216     src = _mm_loadu_si128((__m128i *) pi2_src);         //a0 a1 a2 a3 b0 b1 b2 b3  in ih264_ihadamard_scaling_2x2_uv_sse42()
219     plane_1 = _mm_unpackhi_epi16(src, sign_reg);        //b0 b1 b2 b3 -- 32 bits  in ih264_ihadamard_scaling_2x2_uv_sse42()
221     temp0 = _mm_hadd_epi32(plane_0, plane_1);           //a0+a1 a2+a3 b0+b1 b2+b3  in ih264_ihadamard_scaling_2x2_uv_sse42()
222     temp1 = _mm_hsub_epi32(plane_0, plane_1);           //a0-a1 a2-a3 b0-b1 b2-b3  in ih264_ihadamard_scaling_2x2_uv_sse42()
223 …ane_0 = _mm_hadd_epi32(temp0, temp1);             //a0+a1+a2+a3 b0+b1+b2+b3 a0-a1+a2-a3 b0-b1+b2-b3  in ih264_ihadamard_scaling_2x2_uv_sse42()
224 …ane_1 = _mm_hsub_epi32(temp0, temp1);             //a0+a1-a2-a3 b0+b1-b2-b3 a0-a1-a2+a3 b0-b1-b2+b3  in ih264_ihadamard_scaling_2x2_uv_sse42()
225 …mp0 = _mm_unpacklo_epi32(plane_0, plane_1);       //a0+a1+a2+a3 a0+a1-a2-a3 b0+b1+b2+b3 b0+b1-b2-b3  in ih264_ihadamard_scaling_2x2_uv_sse42()
226 …mp1 = _mm_unpackhi_epi32(plane_0, plane_1);       //a0-a1+a2-a3 a0-a1-a2+a3 b0-b1+b2-b3 b0-b1-b2+b3  in ih264_ihadamard_scaling_2x2_uv_sse42()
229 …ane_1 = _mm_unpackhi_epi64(temp0, temp1);         //b0+b1+b2+b3 b0+b1-b2-b3 b0-b1+b2-b3 b0-b1-b2+b3  in ih264_ihadamard_scaling_2x2_uv_sse42()
232 …ane_1 = _mm_shuffle_epi32(plane_1, 0xd8);         //b0+b1+b2+b3 b0-b1+b2-b3 b0+b1-b2-b3 b0-b1-b2+b3  in ih264_ihadamard_scaling_2x2_uv_sse42()