1 /*
2 * Loongson MMI optimizations for libjpeg-turbo
3 *
4 * Copyright (C) 2015, 2018, D. R. Commander. All Rights Reserved.
5 * Copyright (C) 2016-2017, Loongson Technology Corporation Limited, BeiJing.
6 * All Rights Reserved.
7 * Authors: ZhuChen <zhuchen@loongson.cn>
8 * CaiWanwei <caiwanwei@loongson.cn>
9 * SunZhangzhi <sunzhangzhi-cq@loongson.cn>
10 *
11 * Based on the x86 SIMD extension for IJG JPEG library
12 * Copyright (C) 1999-2006, MIYASAKA Masaru.
13 *
14 * This software is provided 'as-is', without any express or implied
15 * warranty. In no event will the authors be held liable for any damages
16 * arising from the use of this software.
17 *
18 * Permission is granted to anyone to use this software for any purpose,
19 * including commercial applications, and to alter it and redistribute it
20 * freely, subject to the following restrictions:
21 *
22 * 1. The origin of this software must not be misrepresented; you must not
23 * claim that you wrote the original software. If you use this software
24 * in a product, an acknowledgment in the product documentation would be
25 * appreciated but is not required.
26 * 2. Altered source versions must be plainly marked as such, and must not be
27 * misrepresented as being the original software.
28 * 3. This notice may not be removed or altered from any source distribution.
29 */
30
31 /* CHROMA DOWNSAMPLING */
32
33 #include "jsimd_mmi.h"
34 #include "jcsample.h"
35
36
jsimd_h2v2_downsample_mmi(JDIMENSION image_width,int max_v_samp_factor,JDIMENSION v_samp_factor,JDIMENSION width_in_blocks,JSAMPARRAY input_data,JSAMPARRAY output_data)37 void jsimd_h2v2_downsample_mmi(JDIMENSION image_width, int max_v_samp_factor,
38 JDIMENSION v_samp_factor,
39 JDIMENSION width_in_blocks,
40 JSAMPARRAY input_data, JSAMPARRAY output_data)
41 {
42 int inrow, outrow, outcol, bias;
43 JDIMENSION output_cols = width_in_blocks * DCTSIZE;
44 JSAMPROW inptr0, inptr1, outptr;
45 __m64 mm0, mm1, mm2, mm3, mm4, mm5, mm6 = 0.0, mm7;
46
47 expand_right_edge(input_data, max_v_samp_factor, image_width,
48 output_cols * 2);
49
50 bias = (1 << 17) + 1; /* 0x00020001 (bias pattern) */
51 mm7 = _mm_set1_pi32(bias); /* mm7={1, 2, 1, 2} */
52 mm6 = _mm_cmpeq_pi16(mm6, mm6);
53 mm6 = _mm_srli_pi16(mm6, BYTE_BIT); /* mm6={0xFF 0x00 0xFF 0x00 ..} */
54
55 for (inrow = 0, outrow = 0; outrow < v_samp_factor;
56 inrow += 2, outrow++) {
57
58 inptr0 = input_data[inrow];
59 inptr1 = input_data[inrow + 1];
60 outptr = output_data[outrow];
61
62 for (outcol = output_cols; outcol > 0;
63 outcol -= 8, inptr0 += 16, inptr1 += 16, outptr += 8) {
64
65 mm0 = _mm_load_si64((__m64 *)&inptr0[0]);
66 mm1 = _mm_load_si64((__m64 *)&inptr1[0]);
67 mm2 = _mm_load_si64((__m64 *)&inptr0[8]);
68 mm3 = _mm_load_si64((__m64 *)&inptr1[8]);
69
70 mm4 = mm0;
71 mm5 = mm1;
72 mm0 = _mm_and_si64(mm0, mm6);
73 mm4 = _mm_srli_pi16(mm4, BYTE_BIT);
74 mm1 = _mm_and_si64(mm1, mm6);
75 mm5 = _mm_srli_pi16(mm5, BYTE_BIT);
76 mm0 = _mm_add_pi16(mm0, mm4);
77 mm1 = _mm_add_pi16(mm1, mm5);
78
79 mm4 = mm2;
80 mm5 = mm3;
81 mm2 = _mm_and_si64(mm2, mm6);
82 mm4 = _mm_srli_pi16(mm4, BYTE_BIT);
83 mm3 = _mm_and_si64(mm3, mm6);
84 mm5 = _mm_srli_pi16(mm5, BYTE_BIT);
85 mm2 = _mm_add_pi16(mm2, mm4);
86 mm3 = _mm_add_pi16(mm3, mm5);
87
88 mm0 = _mm_add_pi16(mm0, mm1);
89 mm2 = _mm_add_pi16(mm2, mm3);
90 mm0 = _mm_add_pi16(mm0, mm7);
91 mm2 = _mm_add_pi16(mm2, mm7);
92 mm0 = _mm_srli_pi16(mm0, 2);
93 mm2 = _mm_srli_pi16(mm2, 2);
94
95 mm0 = _mm_packs_pu16(mm0, mm2);
96
97 _mm_store_si64((__m64 *)&outptr[0], mm0);
98 }
99 }
100 }
101