98     __m128i temp0, temp1, temp2, temp3;  in ih264_ihadamard_scaling_4x4_sse42()  local
125     temp0 = _mm_unpacklo_epi32(src_r0, src_r1);                  //a0 b0 a1 b1  in ih264_ihadamard_scaling_4x4_sse42()
129     src_r0 = _mm_unpacklo_epi64(temp0, temp2);                    //a0 b0 c0 d0  in ih264_ihadamard_scaling_4x4_sse42()
130     src_r1 = _mm_unpackhi_epi64(temp0, temp2);                    //a1 b1 c1 d1  in ih264_ihadamard_scaling_4x4_sse42()
134     temp0 = _mm_add_epi32(src_r0, src_r3);  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()
154     temp0 = _mm_unpacklo_epi32(src_r0, src_r1);                  //a0 a1 b0 b1  in ih264_ihadamard_scaling_4x4_sse42()
158     src_r0 = _mm_unpacklo_epi64(temp0, temp2);                   //a0 a1 a2 a3  in ih264_ihadamard_scaling_4x4_sse42()
159     src_r1 = _mm_unpackhi_epi64(temp0, temp2);                   //b0 b1 b2 b3  in ih264_ihadamard_scaling_4x4_sse42()
163     temp0 = _mm_add_epi32(src_r0, src_r3);  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()
188         temp0 = _mm_add_epi32(src_r0, add_rshift);  in ih264_ihadamard_scaling_4x4_sse42()
192         src_r0 = _mm_srai_epi32(temp0, 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
221     temp0 = _mm_hadd_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()
225 …temp0 = _mm_unpacklo_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()
234 …temp0 = _mm_mullo_epi32(scale_val, plane_0);        //multiply by pu2_iscal_mat[0] * pu2_weigh_mat…  in ih264_ihadamard_scaling_2x2_uv_sse42()
237     temp0 = _mm_slli_epi32(temp0, u4_qp_div_6);  in ih264_ihadamard_scaling_2x2_uv_sse42()
240     temp0 = _mm_srai_epi32(temp0, 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()
245     _mm_storeu_si128((__m128i *) (&pi2_out[0]), temp0);  in ih264_ihadamard_scaling_2x2_uv_sse42()