1 /*
2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3 *
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 */
11
12 #include <emmintrin.h>
13 #include <stdio.h>
14
15 #include "av1/common/common.h"
16
av1_highbd_block_error_sse2(const tran_low_t * coeff,const tran_low_t * dqcoeff,intptr_t block_size,int64_t * ssz,int bps)17 int64_t av1_highbd_block_error_sse2(const tran_low_t *coeff,
18 const tran_low_t *dqcoeff,
19 intptr_t block_size, int64_t *ssz,
20 int bps) {
21 int i, j, test;
22 uint32_t temp[4];
23 __m128i max, min, cmp0, cmp1, cmp2, cmp3;
24 int64_t error = 0, sqcoeff = 0;
25 const int shift = 2 * (bps - 8);
26 const int rounding = shift > 0 ? 1 << (shift - 1) : 0;
27
28 for (i = 0; i < block_size; i += 8) {
29 // Load the data into xmm registers
30 __m128i mm_coeff = _mm_load_si128((__m128i *)(coeff + i));
31 __m128i mm_coeff2 = _mm_load_si128((__m128i *)(coeff + i + 4));
32 __m128i mm_dqcoeff = _mm_load_si128((__m128i *)(dqcoeff + i));
33 __m128i mm_dqcoeff2 = _mm_load_si128((__m128i *)(dqcoeff + i + 4));
34 // Check if any values require more than 15 bit
35 max = _mm_set1_epi32(0x3fff);
36 min = _mm_set1_epi32((int)0xffffc000);
37 cmp0 = _mm_xor_si128(_mm_cmpgt_epi32(mm_coeff, max),
38 _mm_cmplt_epi32(mm_coeff, min));
39 cmp1 = _mm_xor_si128(_mm_cmpgt_epi32(mm_coeff2, max),
40 _mm_cmplt_epi32(mm_coeff2, min));
41 cmp2 = _mm_xor_si128(_mm_cmpgt_epi32(mm_dqcoeff, max),
42 _mm_cmplt_epi32(mm_dqcoeff, min));
43 cmp3 = _mm_xor_si128(_mm_cmpgt_epi32(mm_dqcoeff2, max),
44 _mm_cmplt_epi32(mm_dqcoeff2, min));
45 test = _mm_movemask_epi8(
46 _mm_or_si128(_mm_or_si128(cmp0, cmp1), _mm_or_si128(cmp2, cmp3)));
47
48 if (!test) {
49 __m128i mm_diff, error_sse2, sqcoeff_sse2;
50 mm_coeff = _mm_packs_epi32(mm_coeff, mm_coeff2);
51 mm_dqcoeff = _mm_packs_epi32(mm_dqcoeff, mm_dqcoeff2);
52 mm_diff = _mm_sub_epi16(mm_coeff, mm_dqcoeff);
53 error_sse2 = _mm_madd_epi16(mm_diff, mm_diff);
54 sqcoeff_sse2 = _mm_madd_epi16(mm_coeff, mm_coeff);
55 _mm_storeu_si128((__m128i *)temp, error_sse2);
56 error = error + temp[0] + temp[1] + temp[2] + temp[3];
57 _mm_storeu_si128((__m128i *)temp, sqcoeff_sse2);
58 sqcoeff += temp[0] + temp[1] + temp[2] + temp[3];
59 } else {
60 for (j = 0; j < 8; j++) {
61 const int64_t diff = coeff[i + j] - dqcoeff[i + j];
62 error += diff * diff;
63 sqcoeff += (int64_t)coeff[i + j] * (int64_t)coeff[i + j];
64 }
65 }
66 }
67 assert(error >= 0 && sqcoeff >= 0);
68 error = (error + rounding) >> shift;
69 sqcoeff = (sqcoeff + rounding) >> shift;
70
71 *ssz = sqcoeff;
72 return error;
73 }
74