98     __m128i temp0, temp1, temp2, temp3;  in ih264_ihadamard_scaling_4x4_sse42()  local
127     temp1 = _mm_unpackhi_epi32(src_r0, src_r1);                  //a2 b2 a3 b3  in ih264_ihadamard_scaling_4x4_sse42()
131     src_r2 = _mm_unpacklo_epi64(temp1, temp3);                    //a2 b2 c2 d2  in ih264_ihadamard_scaling_4x4_sse42()
132     src_r3 = _mm_unpackhi_epi64(temp1, temp3);                    //a3 b3 c3 d3  in ih264_ihadamard_scaling_4x4_sse42()
135     temp1 = _mm_add_epi32(src_r1, src_r2);  in ih264_ihadamard_scaling_4x4_sse42()
139     src_r0 = _mm_add_epi32(temp0, temp1);  in ih264_ihadamard_scaling_4x4_sse42()
141     src_r2 = _mm_sub_epi32(temp0, temp1);  in ih264_ihadamard_scaling_4x4_sse42()
156     temp1 = _mm_unpackhi_epi32(src_r0, src_r1);                  //c0 c1 d0 d1  in ih264_ihadamard_scaling_4x4_sse42()
160     src_r2 = _mm_unpacklo_epi64(temp1, temp3);                   //c0 c1 c2 c3  in ih264_ihadamard_scaling_4x4_sse42()
161     src_r3 = _mm_unpackhi_epi64(temp1, temp3);                   //d0 d1 d2 d3  in ih264_ihadamard_scaling_4x4_sse42()
164     temp1 = _mm_add_epi32(src_r1, src_r2);  in ih264_ihadamard_scaling_4x4_sse42()
168     src_r0 = _mm_add_epi32(temp0, temp1);  in ih264_ihadamard_scaling_4x4_sse42()
170     src_r2 = _mm_sub_epi32(temp0, temp1);  in ih264_ihadamard_scaling_4x4_sse42()
189         temp1 = _mm_add_epi32(src_r1, add_rshift);  in ih264_ihadamard_scaling_4x4_sse42()
193         src_r1 = _mm_srai_epi32(temp1, 6 - u4_qp_div_6);  in ih264_ihadamard_scaling_4x4_sse42()
211     __m128i src, plane_0, plane_1, temp0, temp1, sign_reg;  in ih264_ihadamard_scaling_2x2_uv_sse42()  local
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 …plane_0 = _mm_hadd_epi32(temp0, temp1);             //a0+a1+a2+a3 b0+b1+b2+b3 a0-a1+a2-a3 b0-b1+b2…  in ih264_ihadamard_scaling_2x2_uv_sse42()
224 …plane_1 = _mm_hsub_epi32(temp0, temp1);             //a0+a1-a2-a3 b0+b1-b2-b3 a0-a1-a2+a3 b0-b1-b2…  in ih264_ihadamard_scaling_2x2_uv_sse42()
226 …temp1 = _mm_unpackhi_epi32(plane_0, plane_1);       //a0-a1+a2-a3 a0-a1-a2+a3 b0-b1+b2-b3 b0-b1-b2…  in ih264_ihadamard_scaling_2x2_uv_sse42()
228 …plane_0 = _mm_unpacklo_epi64(temp0, temp1);         //a0+a1+a2+a3 a0+a1-a2-a3 a0-a1+a2-a3 a0-a1-a2…  in ih264_ihadamard_scaling_2x2_uv_sse42()
229 …plane_1 = _mm_unpackhi_epi64(temp0, temp1);         //b0+b1+b2+b3 b0+b1-b2-b3 b0-b1+b2-b3 b0-b1-b2…  in ih264_ihadamard_scaling_2x2_uv_sse42()
235 …temp1 = _mm_mullo_epi32(scale_val, plane_1);        //multiply by pu2_iscal_mat[0] * pu2_weigh_mat…  in ih264_ihadamard_scaling_2x2_uv_sse42()
238     temp1 = _mm_slli_epi32(temp1, u4_qp_div_6);  in ih264_ihadamard_scaling_2x2_uv_sse42()
241     temp1 = _mm_srai_epi32(temp1, 5);  in ih264_ihadamard_scaling_2x2_uv_sse42()
243     temp0 = _mm_packs_epi32(temp0, temp1);              //Final values are 16-bits only.  in ih264_ihadamard_scaling_2x2_uv_sse42()