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
|
/*
* Simd Library (http://ermig1979.github.io/Simd).
*
* Copyright (c) 2011-2022 Yermalayeu Ihar.
*
* 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.
*/
#include "Simd/SimdMemory.h"
#include "Simd/SimdLoadBlock.h"
#include "Simd/SimdStore.h"
#include "Simd/SimdUnpack.h"
namespace Simd
{
#ifdef SIMD_SSE41_ENABLE
namespace Sse41
{
namespace
{
struct Buffer
{
Buffer(size_t width)
{
_p = Allocate(sizeof(uint16_t) * 3 * width);
src0 = (uint16_t*)_p;
src1 = src0 + width;
src2 = src1 + width;
}
~Buffer()
{
Free(_p);
}
uint16_t * src0;
uint16_t * src1;
uint16_t * src2;
private:
void * _p;
};
}
const __m128i K8_01_02 = SIMD_MM_SET2_EPI8(0x01, 0x02);
template<int part> SIMD_INLINE __m128i BinomialSumUnpackedU8(__m128i a[3])
{
return _mm_add_epi16(_mm_maddubs_epi16(UnpackU8<part>(a[0], a[1]), K8_01_02), UnpackU8<part>(a[2]));
}
template<bool align> SIMD_INLINE void BlurCol(__m128i a[3], uint16_t * b)
{
Store<align>((__m128i*)b + 0, BinomialSumUnpackedU8<0>(a));
Store<align>((__m128i*)b + 1, BinomialSumUnpackedU8<1>(a));
}
template<bool align> SIMD_INLINE __m128i BlurRow16(const Buffer & buffer, size_t offset)
{
return DivideBy16(BinomialSum16(
Load<align>((__m128i*)(buffer.src0 + offset)),
Load<align>((__m128i*)(buffer.src1 + offset)),
Load<align>((__m128i*)(buffer.src2 + offset))));
}
template<bool align> SIMD_INLINE __m128i BlurRow(const Buffer & buffer, size_t offset)
{
return _mm_packus_epi16(BlurRow16<align>(buffer, offset), BlurRow16<align>(buffer, offset + HA));
}
template <bool align, size_t step> void GaussianBlur3x3(
const uint8_t * src, size_t srcStride, size_t width, size_t height, uint8_t * dst, size_t dstStride)
{
assert(step*width >= A);
if (align)
assert(Aligned(src) && Aligned(srcStride) && Aligned(step*width) && Aligned(dst) && Aligned(dstStride));
__m128i a[3];
size_t size = step*width;
size_t bodySize = Simd::AlignHi(size, A) - A;
Buffer buffer(Simd::AlignHi(size, A));
LoadNose3<align, step>(src + 0, a);
BlurCol<true>(a, buffer.src0 + 0);
for (size_t col = A; col < bodySize; col += A)
{
LoadBody3<align, step>(src + col, a);
BlurCol<true>(a, buffer.src0 + col);
}
LoadTail3<align, step>(src + size - A, a);
BlurCol<align>(a, buffer.src0 + size - A);
memcpy(buffer.src1, buffer.src0, sizeof(uint16_t)*size);
for (size_t row = 0; row < height; ++row, dst += dstStride)
{
const uint8_t *src2 = src + srcStride*(row + 1);
if (row >= height - 2)
src2 = src + srcStride*(height - 1);
LoadNose3<align, step>(src2 + 0, a);
BlurCol<true>(a, buffer.src2 + 0);
for (size_t col = A; col < bodySize; col += A)
{
LoadBody3<align, step>(src2 + col, a);
BlurCol<true>(a, buffer.src2 + col);
}
LoadTail3<align, step>(src2 + size - A, a);
BlurCol<align>(a, buffer.src2 + size - A);
for (size_t col = 0; col < bodySize; col += A)
Store<align>((__m128i*)(dst + col), BlurRow<true>(buffer, col));
Store<align>((__m128i*)(dst + size - A), BlurRow<align>(buffer, size - A));
Swap(buffer.src0, buffer.src2);
Swap(buffer.src0, buffer.src1);
}
}
template <bool align> void GaussianBlur3x3(const uint8_t * src, size_t srcStride, size_t width, size_t height,
size_t channelCount, uint8_t * dst, size_t dstStride)
{
assert(channelCount > 0 && channelCount <= 4);
switch (channelCount)
{
case 1: GaussianBlur3x3<align, 1>(src, srcStride, width, height, dst, dstStride); break;
case 2: GaussianBlur3x3<align, 2>(src, srcStride, width, height, dst, dstStride); break;
case 3: GaussianBlur3x3<align, 3>(src, srcStride, width, height, dst, dstStride); break;
case 4: GaussianBlur3x3<align, 4>(src, srcStride, width, height, dst, dstStride); break;
}
}
void GaussianBlur3x3(const uint8_t * src, size_t srcStride, size_t width, size_t height,
size_t channelCount, uint8_t * dst, size_t dstStride)
{
if (Aligned(src) && Aligned(srcStride) && Aligned(channelCount*width) && Aligned(dst) && Aligned(dstStride))
GaussianBlur3x3<true>(src, srcStride, width, height, channelCount, dst, dstStride);
else
GaussianBlur3x3<false>(src, srcStride, width, height, channelCount, dst, dstStride);
}
}
#endif
}
|
