1 /*
2 * Loongson MMI optimizations for libjpeg-turbo
3 *
4 * Copyright (C) 2015, 2018-2019, 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;
43 JDIMENSION output_cols = width_in_blocks * DCTSIZE;
44 JSAMPROW inptr0, inptr1, outptr;
45 __m64 bias, mask = 0.0, thisavg, nextavg, avg;
46 __m64 this0o, this0e, this0, this0sum, next0o, next0e, next0, next0sum;
47 __m64 this1o, this1e, this1, this1sum, next1o, next1e, next1, next1sum;
48
49 expand_right_edge(input_data, max_v_samp_factor, image_width,
50 output_cols * 2);
51
52 bias = _mm_set1_pi32((1 << 17) + 1); /* 0x00020001 (32-bit bias pattern) */
53 /* bias={1, 2, 1, 2} (16-bit) */
54 mask = _mm_cmpeq_pi16(mask, mask);
55 mask = _mm_srli_pi16(mask, BYTE_BIT); /* {0xFF 0x00 0xFF 0x00 ..} */
56
57 for (inrow = 0, outrow = 0; outrow < v_samp_factor;
58 inrow += 2, outrow++) {
59
60 inptr0 = input_data[inrow];
61 inptr1 = input_data[inrow + 1];
62 outptr = output_data[outrow];
63
64 for (outcol = output_cols; outcol > 0;
65 outcol -= 8, inptr0 += 16, inptr1 += 16, outptr += 8) {
66
67 this0 = _mm_load_si64((__m64 *)&inptr0[0]);
68 this1 = _mm_load_si64((__m64 *)&inptr1[0]);
69 next0 = _mm_load_si64((__m64 *)&inptr0[8]);
70 next1 = _mm_load_si64((__m64 *)&inptr1[8]);
71
72 this0o = _mm_and_si64(this0, mask);
73 this0e = _mm_srli_pi16(this0, BYTE_BIT);
74 this1o = _mm_and_si64(this1, mask);
75 this1e = _mm_srli_pi16(this1, BYTE_BIT);
76 this0sum = _mm_add_pi16(this0o, this0e);
77 this1sum = _mm_add_pi16(this1o, this1e);
78
79 next0o = _mm_and_si64(next0, mask);
80 next0e = _mm_srli_pi16(next0, BYTE_BIT);
81 next1o = _mm_and_si64(next1, mask);
82 next1e = _mm_srli_pi16(next1, BYTE_BIT);
83 next0sum = _mm_add_pi16(next0o, next0e);
84 next1sum = _mm_add_pi16(next1o, next1e);
85
86 thisavg = _mm_add_pi16(this0sum, this1sum);
87 nextavg = _mm_add_pi16(next0sum, next1sum);
88 thisavg = _mm_add_pi16(thisavg, bias);
89 nextavg = _mm_add_pi16(nextavg, bias);
90 thisavg = _mm_srli_pi16(thisavg, 2);
91 nextavg = _mm_srli_pi16(nextavg, 2);
92
93 avg = _mm_packs_pu16(thisavg, nextavg);
94
95 _mm_store_si64((__m64 *)&outptr[0], avg);
96 }
97 }
98 }
99