1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
|
/*
* Copyright (c) 2021 - 2022 Samsung Electronics Co., Ltd. All rights reserved.
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#ifdef THORVG_AVX_VECTOR_SUPPORT
#include <immintrin.h>
#define N_32BITS_IN_128REG 4
#define N_32BITS_IN_256REG 8
static inline __m128i ALPHA_BLEND(__m128i c, __m128i a)
{
//1. set the masks for the A/G and R/B channels
auto AG = _mm_set1_epi32(0xff00ff00);
auto RB = _mm_set1_epi32(0x00ff00ff);
//2. mask the alpha vector - originally quartet [a, a, a, a]
auto aAG = _mm_and_si128(a, AG);
auto aRB = _mm_and_si128(a, RB);
//3. calculate the alpha blending of the 2nd and 4th channel
//- mask the color vector
//- multiply it by the masked alpha vector
//- add the correction to compensate bit shifting used instead of dividing by 255
//- shift bits - corresponding to division by 256
auto even = _mm_and_si128(c, RB);
even = _mm_mullo_epi16(even, aRB);
even =_mm_add_epi16(even, RB);
even = _mm_srli_epi16(even, 8);
//4. calculate the alpha blending of the 1st and 3rd channel:
//- mask the color vector
//- multiply it by the corresponding masked alpha vector and store the high bits of the result
//- add the correction to compensate division by 256 instead of by 255 (next step)
//- remove the low 8 bits to mimic the division by 256
auto odd = _mm_and_si128(c, AG);
odd = _mm_mulhi_epu16(odd, aAG);
odd = _mm_add_epi16(odd, RB);
odd = _mm_and_si128(odd, AG);
//5. the final result
return _mm_or_si128(odd, even);
}
static void avxRasterRGBA32(uint32_t *dst, uint32_t val, uint32_t offset, int32_t len)
{
//1. calculate how many iterations we need to cover the length
uint32_t iterations = len / N_32BITS_IN_256REG;
uint32_t avxFilled = iterations * N_32BITS_IN_256REG;
//2. set the beginning of the array
dst += offset;
//3. fill the octets
for (uint32_t i = 0; i < iterations; ++i, dst += N_32BITS_IN_256REG) {
_mm256_storeu_si256((__m256i*)dst, _mm256_set1_epi32(val));
}
//4. fill leftovers (in the first step we have to set the pointer to the place where the avx job is done)
int32_t leftovers = len - avxFilled;
while (leftovers--) *dst++ = val;
}
static bool avxRasterTranslucentRect(SwSurface* surface, const SwBBox& region, uint32_t color)
{
auto buffer = surface->buffer + (region.min.y * surface->stride) + region.min.x;
auto h = static_cast<uint32_t>(region.max.y - region.min.y);
auto w = static_cast<uint32_t>(region.max.x - region.min.x);
auto ialpha = 255 - static_cast<uint8_t>(_alpha(color));
auto avxColor = _mm_set1_epi32(color);
auto avxIalpha = _mm_set1_epi8(ialpha);
for (uint32_t y = 0; y < h; ++y) {
auto dst = &buffer[y * surface->stride];
//1. fill the not aligned memory (for 128-bit registers a 16-bytes alignment is required)
auto notAligned = ((uintptr_t)dst & 0xf) / 4;
if (notAligned) {
notAligned = (N_32BITS_IN_128REG - notAligned > w ? w : N_32BITS_IN_128REG - notAligned);
for (uint32_t x = 0; x < notAligned; ++x, ++dst) {
*dst = color + ALPHA_BLEND(*dst, ialpha);
}
}
//2. fill the aligned memory - N_32BITS_IN_128REG pixels processed at once
uint32_t iterations = (w - notAligned) / N_32BITS_IN_128REG;
uint32_t avxFilled = iterations * N_32BITS_IN_128REG;
auto avxDst = (__m128i*)dst;
for (uint32_t x = 0; x < iterations; ++x, ++avxDst) {
*avxDst = _mm_add_epi32(avxColor, ALPHA_BLEND(*avxDst, avxIalpha));
}
//3. fill the remaining pixels
int32_t leftovers = w - notAligned - avxFilled;
dst += avxFilled;
while (leftovers--) {
*dst = color + ALPHA_BLEND(*dst, ialpha);
dst++;
}
}
return true;
}
static bool avxRasterTranslucentRle(SwSurface* surface, const SwRleData* rle, uint32_t color)
{
auto span = rle->spans;
uint32_t src;
for (uint32_t i = 0; i < rle->size; ++i) {
auto dst = &surface->buffer[span->y * surface->stride + span->x];
if (span->coverage < 255) src = ALPHA_BLEND(color, span->coverage);
else src = color;
auto ialpha = 255 - static_cast<uint8_t>(_alpha(src));
//1. fill the not aligned memory (for 128-bit registers a 16-bytes alignment is required)
auto notAligned = ((uintptr_t)dst & 0xf) / 4;
if (notAligned) {
notAligned = (N_32BITS_IN_128REG - notAligned > span->len ? span->len : N_32BITS_IN_128REG - notAligned);
for (uint32_t x = 0; x < notAligned; ++x, ++dst) {
*dst = src + ALPHA_BLEND(*dst, ialpha);
}
}
//2. fill the aligned memory using avx - N_32BITS_IN_128REG pixels processed at once
//In order to avoid unneccessary avx variables declarations a check is made whether there are any iterations at all
uint32_t iterations = (span->len - notAligned) / N_32BITS_IN_128REG;
uint32_t avxFilled = 0;
if (iterations > 0) {
auto avxSrc = _mm_set1_epi32(src);
auto avxIalpha = _mm_set1_epi8(ialpha);
avxFilled = iterations * N_32BITS_IN_128REG;
auto avxDst = (__m128i*)dst;
for (uint32_t x = 0; x < iterations; ++x, ++avxDst) {
*avxDst = _mm_add_epi32(avxSrc, ALPHA_BLEND(*avxDst, avxIalpha));
}
}
//3. fill the remaining pixels
int32_t leftovers = span->len - notAligned - avxFilled;
dst += avxFilled;
while (leftovers--) {
*dst = src + ALPHA_BLEND(*dst, ialpha);
dst++;
}
++span;
}
return true;
}
#endif
|