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
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
|
/*
* 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/SimdResizer.h"
#include "Simd/SimdResizerCommon.h"
#include "Simd/SimdStore.h"
#include "Simd/SimdSet.h"
#include "Simd/SimdExtract.h"
namespace Simd
{
#ifdef SIMD_AVX2_ENABLE
namespace Avx2
{
ResizerByteBicubic::ResizerByteBicubic(const ResParam& param)
: Sse41::ResizerByteBicubic(param)
{
}
template<int N> __m256i LoadAx(const int8_t* ax);
template<> SIMD_INLINE __m256i LoadAx<1>(const int8_t* ax)
{
return _mm256_loadu_si256((__m256i*)ax);
}
template<> SIMD_INLINE __m256i LoadAx<2>(const int8_t* ax)
{
static const __m256i PERMUTE = SIMD_MM256_SETR_EPI32(0, 0, 1, 1, 2, 2, 3, 3);
return _mm256_permutevar8x32_epi32(_mm256_castsi128_si256(_mm_loadu_si128((__m128i*)ax)), PERMUTE);
}
template<> SIMD_INLINE __m256i LoadAx<3>(const int8_t* ax)
{
static const __m256i PERMUTE = SIMD_MM256_SETR_EPI32(0, 0, 0, 0, 1, 1, 1, 1);
return _mm256_permutevar8x32_epi32(_mm256_castsi128_si256(_mm_loadl_epi64((__m128i*)ax)), PERMUTE);
}
template<> SIMD_INLINE __m256i LoadAx<4>(const int8_t* ax)
{
static const __m256i PERMUTE = SIMD_MM256_SETR_EPI32(0, 0, 0, 0, 1, 1, 1, 1);
return _mm256_permutevar8x32_epi32(_mm256_castsi128_si256(_mm_loadl_epi64((__m128i*)ax)), PERMUTE);
}
template<int N> __m256i CubicSumX(const uint8_t* src, const int32_t* ix, __m256i ax, __m256i ay);
template<> SIMD_INLINE __m256i CubicSumX<1>(const uint8_t* src, const int32_t* ix, __m256i ax, __m256i ay)
{
__m256i _src = _mm256_i32gather_epi32((int32_t*)src, _mm256_loadu_si256((__m256i*)ix), 1);
return _mm256_madd_epi16(_mm256_maddubs_epi16(_src, ax), ay);
}
template<> SIMD_INLINE __m256i CubicSumX<2>(const uint8_t* src, const int32_t* ix, __m256i ax, __m256i ay)
{
static const __m256i SHUFFLE = SIMD_MM256_SETR_EPI8(
0x0, 0x2, 0x4, 0x6, 0x1, 0x3, 0x5, 0x7, 0x8, 0xA, 0xC, 0xE, 0x9, 0xB, 0xD, 0xF,
0x0, 0x2, 0x4, 0x6, 0x1, 0x3, 0x5, 0x7, 0x8, 0xA, 0xC, 0xE, 0x9, 0xB, 0xD, 0xF);
__m256i _src = _mm256_shuffle_epi8(_mm256_i32gather_epi64((long long*)src, _mm_loadu_si128((__m128i*)ix), 1), SHUFFLE);
return _mm256_madd_epi16(_mm256_maddubs_epi16(_src, ax), ay);
}
template<> SIMD_INLINE __m256i CubicSumX<3>(const uint8_t* src, const int32_t* ix, __m256i ax, __m256i ay)
{
static const __m256i SHUFFLE = SIMD_MM256_SETR_EPI8(
0x0, 0x3, 0x6, 0x9, 0x1, 0x4, 0x7, 0xA, 0x2, 0x5, 0x8, 0xB, -1, -1, -1, -1,
0x0, 0x3, 0x6, 0x9, 0x1, 0x4, 0x7, 0xA, 0x2, 0x5, 0x8, 0xB, -1, -1, -1, -1);
__m256i _src = _mm256_shuffle_epi8(Load<false>((__m128i*)(src + ix[0]), (__m128i*)(src + ix[1])), SHUFFLE);
return _mm256_madd_epi16(_mm256_maddubs_epi16(_src, ax), ay);
}
template<> SIMD_INLINE __m256i CubicSumX<4>(const uint8_t* src, const int32_t* ix, __m256i ax, __m256i ay)
{
static const __m256i SHUFFLE = SIMD_MM256_SETR_EPI8(
0x0, 0x4, 0x8, 0xC, 0x1, 0x5, 0x9, 0xD, 0x2, 0x6, 0xA, 0xE, 0x3, 0x7, 0xB, 0xF,
0x0, 0x4, 0x8, 0xC, 0x1, 0x5, 0x9, 0xD, 0x2, 0x6, 0xA, 0xE, 0x3, 0x7, 0xB, 0xF);
__m256i _src = _mm256_shuffle_epi8(Load<false>((__m128i*)(src + ix[0]), (__m128i*)(src + ix[1])), SHUFFLE);
return _mm256_madd_epi16(_mm256_maddubs_epi16(_src, ax), ay);
}
template <int N> SIMD_INLINE void StoreBicubicInt(__m256i val, uint8_t* dst)
{
*((int64_t*)dst) = Extract64i<0>(PackI16ToU8(PackI32ToI16(val, K_ZERO), K_ZERO));
}
template <> SIMD_INLINE void StoreBicubicInt<3>(__m256i val, uint8_t* dst)
{
static const __m128i SHUFFLE = SIMD_MM_SETR_EPI8(0x0, 0x1, 0x2, 0x4, 0x5, 0x6, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1);
__m128i u8 = _mm256_castsi256_si128(PackI16ToU8(PackI32ToI16(val, K_ZERO), K_ZERO));
*((int64_t*)dst) = Sse41::ExtractInt64<0>(_mm_shuffle_epi8(u8, SHUFFLE));
}
template <int N> SIMD_INLINE void BicubicInt(const uint8_t* src0, const uint8_t* src1, const uint8_t* src2, const uint8_t* src3, const int32_t* ix, const int8_t* ax, const __m256i* ay, uint8_t* dst)
{
static const __m256i ROUND = SIMD_MM256_SET1_EPI32(Base::BICUBIC_ROUND);
__m256i _ax = LoadAx<N>(ax);
__m256i say0 = CubicSumX<N>(src0 - N, ix, _ax, ay[0]);
__m256i say1 = CubicSumX<N>(src1 - N, ix, _ax, ay[1]);
__m256i say2 = CubicSumX<N>(src2 - N, ix, _ax, ay[2]);
__m256i say3 = CubicSumX<N>(src3 - N, ix, _ax, ay[3]);
__m256i sum = _mm256_add_epi32(_mm256_add_epi32(say0, say1), _mm256_add_epi32(say2, say3));
__m256i dst0 = _mm256_srai_epi32(_mm256_add_epi32(sum, ROUND), Base::BICUBIC_SHIFT);
StoreBicubicInt<N>(dst0, dst);
}
SIMD_INLINE __m256i CubicSumX1(const uint8_t* src, __m256i ix, __m256i ax, __m256i ay)
{
__m256i _src = _mm256_i32gather_epi32((int32_t*)(src - 1), ix, 1);
return _mm256_madd_epi16(_mm256_maddubs_epi16(_src, ax), ay);
}
template <> SIMD_INLINE void BicubicInt<1>(const uint8_t* src0, const uint8_t* src1, const uint8_t* src2, const uint8_t* src3, const int32_t* ix, const int8_t* ax, const __m256i* ay, uint8_t* dst)
{
static const __m256i ROUND = SIMD_MM256_SET1_EPI32(Base::BICUBIC_ROUND);
__m256i _ix = _mm256_loadu_si256((__m256i*)ix);
__m256i _ax = LoadAx<1>(ax);
__m256i say0 = CubicSumX1(src0, _ix, _ax, ay[0]);
__m256i say1 = CubicSumX1(src1, _ix, _ax, ay[1]);
__m256i say2 = CubicSumX1(src2, _ix, _ax, ay[2]);
__m256i say3 = CubicSumX1(src3, _ix, _ax, ay[3]);
__m256i sum = _mm256_add_epi32(_mm256_add_epi32(say0, say1), _mm256_add_epi32(say2, say3));
__m256i dst0 = _mm256_srai_epi32(_mm256_add_epi32(sum, ROUND), Base::BICUBIC_SHIFT);
StoreBicubicInt<1>(dst0, dst);
}
template<int N> void ResizerByteBicubic::RunS(const uint8_t* src, size_t srcStride, uint8_t* dst, size_t dstStride)
{
assert(_xn == 0 && _xt == _param.dstW);
size_t step = 4 / N * 2;
size_t body = AlignLoAny(_param.dstW - (N == 3 ? 1 : 0), step);
for (size_t dy = 0; dy < _param.dstH; dy++, dst += dstStride)
{
size_t sy = _iy[dy];
const uint8_t* src1 = src + sy * srcStride;
const uint8_t* src2 = src1 + srcStride;
const uint8_t* src0 = sy ? src1 - srcStride : src1;
const uint8_t* src3 = sy < _param.srcH - 2 ? src2 + srcStride : src2;
const int32_t* ay = _ay.data + dy * 4;
__m256i ays[4];
ays[0] = _mm256_set1_epi16(ay[0]);
ays[1] = _mm256_set1_epi16(ay[1]);
ays[2] = _mm256_set1_epi16(ay[2]);
ays[3] = _mm256_set1_epi16(ay[3]);
size_t dx = 0;
for (; dx < body; dx += step)
BicubicInt<N>(src0, src1, src2, src3, _ix.data + dx, _ax.data + dx * 4, ays, dst + dx * N);
for (; dx < _param.dstW; dx++)
Base::BicubicInt<N, -1, 2>(src0, src1, src2, src3, _ix[dx], _ax.data + dx * 4, ay, dst + dx * N);
}
}
template<int F> SIMD_INLINE void PixelCubicSumX(const uint8_t* src, const int32_t* ix, const int8_t* ax, int32_t* dst);
template<> SIMD_INLINE void PixelCubicSumX<1>(const uint8_t* src, const int32_t* ix, const int8_t* ax, int32_t* dst)
{
#if 1
__m128i src0 = _mm_setr_epi32(*(int32_t*)(src + ix[0]), *(int32_t*)(src + ix[1]), *(int32_t*)(src + ix[2]), *(int32_t*)(src + ix[3]));
__m128i src1 = _mm_setr_epi32(*(int32_t*)(src + ix[4]), *(int32_t*)(src + ix[5]), *(int32_t*)(src + ix[6]), *(int32_t*)(src + ix[7]));
__m256i _src = Set(src0, src1);
#else
__m256i _src = _mm256_i32gather_epi32((int32_t*)src, _mm256_loadu_si256((__m256i*)ix), 1);
#endif
__m256i _ax = _mm256_loadu_si256((__m256i*)ax);
_mm256_storeu_si256((__m256i*)dst, _mm256_madd_epi16(_mm256_maddubs_epi16(_src, _ax), K16_0001));
}
template<> SIMD_INLINE void PixelCubicSumX<2>(const uint8_t* src, const int32_t* ix, const int8_t* ax, int32_t* dst)
{
static const __m256i PERMUTE = SIMD_MM256_SETR_EPI32(0, 0, 1, 1, 2, 2, 3, 3);
__m256i _ax = _mm256_permutevar8x32_epi32(_mm256_castsi128_si256(_mm_loadu_si128((__m128i*)ax)), PERMUTE);
static const __m256i SHUFFLE = SIMD_MM256_SETR_EPI8(
0x0, 0x2, 0x4, 0x6, 0x1, 0x3, 0x5, 0x7, 0x8, 0xA, 0xC, 0xE, 0x9, 0xB, 0xD, 0xF,
0x0, 0x2, 0x4, 0x6, 0x1, 0x3, 0x5, 0x7, 0x8, 0xA, 0xC, 0xE, 0x9, 0xB, 0xD, 0xF);
__m256i _src = _mm256_shuffle_epi8(_mm256_i32gather_epi64((long long*)src, _mm_loadu_si128((__m128i*)ix), 1), SHUFFLE);
_mm256_storeu_si256((__m256i*)dst, _mm256_madd_epi16(_mm256_maddubs_epi16(_src, _ax), K16_0001));
}
template<> SIMD_INLINE void PixelCubicSumX<3>(const uint8_t* src, const int32_t* ix, const int8_t* ax, int32_t* dst)
{
static const __m256i PERM_A = SIMD_MM256_SETR_EPI32(0, 0, 0, 1, 1, 1, 0, 0);
__m256i _ax = _mm256_permutevar8x32_epi32(_mm256_castsi128_si256(_mm_loadl_epi64((__m128i*)ax)), PERM_A);
static const __m256i SHUFFLE = SIMD_MM256_SETR_EPI8(
0x0, 0x3, 0x6, 0x9, 0x1, 0x4, 0x7, 0xA, 0x2, 0x5, 0x8, 0xB, -1, -1, -1, -1,
0x0, 0x3, 0x6, 0x9, 0x1, 0x4, 0x7, 0xA, 0x2, 0x5, 0x8, 0xB, -1, -1, -1, -1);
static const __m256i PERM_B = SIMD_MM256_SETR_EPI32(0, 1, 2, 4, 5, 6, 0, 0);
__m256i _src = _mm256_permutevar8x32_epi32(_mm256_shuffle_epi8(Load<false>((__m128i*)(src + ix[0]), (__m128i*)(src + ix[1])), SHUFFLE), PERM_B);
_mm256_storeu_si256((__m256i*)dst, _mm256_madd_epi16(_mm256_maddubs_epi16(_src, _ax), K16_0001));
}
template<> SIMD_INLINE void PixelCubicSumX<4>(const uint8_t* src, const int32_t* ix, const int8_t* ax, int32_t* dst)
{
static const __m256i PERMUTE = SIMD_MM256_SETR_EPI32(0, 0, 0, 0, 1, 1, 1, 1);
__m256i _ax = _mm256_permutevar8x32_epi32(_mm256_castsi128_si256(_mm_loadl_epi64((__m128i*)ax)), PERMUTE);
static const __m256i SHUFFLE = SIMD_MM256_SETR_EPI8(
0x0, 0x4, 0x8, 0xC, 0x1, 0x5, 0x9, 0xD, 0x2, 0x6, 0xA, 0xE, 0x3, 0x7, 0xB, 0xF,
0x0, 0x4, 0x8, 0xC, 0x1, 0x5, 0x9, 0xD, 0x2, 0x6, 0xA, 0xE, 0x3, 0x7, 0xB, 0xF);
__m256i _src = _mm256_shuffle_epi8(Load<false>((__m128i*)(src + ix[0]), (__m128i*)(src + ix[1])), SHUFFLE);
_mm256_storeu_si256((__m256i*)dst, _mm256_madd_epi16(_mm256_maddubs_epi16(_src, _ax), K16_0001));
}
template<int N> SIMD_INLINE void RowCubicSumX(const uint8_t* src, size_t nose, size_t body, size_t tail, const int32_t* ix, const int8_t* ax, int32_t* dst)
{
size_t step = 4 / N * 2;
size_t bodyS = nose + AlignLoAny(body - nose, step);
size_t dx = 0;
for (; dx < nose; dx++, ax += 4, dst += N)
Base::PixelCubicSumX<N, 0, 2>(src + ix[dx], ax, dst);
for (; dx < bodyS; dx += step, ax += 4 * step, dst += N * step)
PixelCubicSumX<N>(src - N, ix + dx, ax, dst);
for (; dx < body; dx++, ax += 4, dst += N)
Base::PixelCubicSumX<N, -1, 2>(src + ix[dx], ax, dst);
for (; dx < tail; dx++, ax += 4, dst += N)
Base::PixelCubicSumX<N, -1, 1>(src + ix[dx], ax, dst);
}
SIMD_INLINE void BicubicRowInt(const int32_t* src0, const int32_t* src1, const int32_t* src2, const int32_t* src3, size_t n, const int32_t* ay, uint8_t* dst)
{
size_t nF = AlignLo(n, F);
size_t i = 0;
if (nF)
{
static const __m256i ROUND = SIMD_MM256_SET1_EPI32(Base::BICUBIC_ROUND);
__m256i ay0 = _mm256_set1_epi32(ay[0]);
__m256i ay1 = _mm256_set1_epi32(ay[1]);
__m256i ay2 = _mm256_set1_epi32(ay[2]);
__m256i ay3 = _mm256_set1_epi32(ay[3]);
for (; i < nF; i += F)
{
__m256i say0 = _mm256_mullo_epi32(_mm256_loadu_si256((__m256i*)(src0 + i)), ay0);
__m256i say1 = _mm256_mullo_epi32(_mm256_loadu_si256((__m256i*)(src1 + i)), ay1);
__m256i say2 = _mm256_mullo_epi32(_mm256_loadu_si256((__m256i*)(src2 + i)), ay2);
__m256i say3 = _mm256_mullo_epi32(_mm256_loadu_si256((__m256i*)(src3 + i)), ay3);
__m256i sum = _mm256_add_epi32(_mm256_add_epi32(say0, say1), _mm256_add_epi32(say2, say3));
__m256i dst0 = _mm256_srai_epi32(_mm256_add_epi32(sum, ROUND), Base::BICUBIC_SHIFT);
*((int64_t*)(dst + i)) = Extract64i<0>(PackI16ToU8(PackI32ToI16(dst0, K_ZERO), K_ZERO));
}
}
for (; i < n; ++i)
{
int32_t sum = ay[0] * src0[i] + ay[1] * src1[i] + ay[2] * src2[i] + ay[3] * src3[i];
dst[i] = Base::RestrictRange((sum + Base::BICUBIC_ROUND) >> Base::BICUBIC_SHIFT, 0, 255);
}
}
template<int N> void ResizerByteBicubic::RunB(const uint8_t* src, size_t srcStride, uint8_t* dst, size_t dstStride)
{
int32_t prev = -1;
for (size_t dy = 0; dy < _param.dstH; dy++, dst += dstStride)
{
int32_t sy = _iy[dy], next = prev;
for (int32_t curr = sy - 1, end = sy + 3; curr < end; ++curr)
{
if (curr < prev)
continue;
const uint8_t* ps = src + RestrictRange(curr, 0, (int)_param.srcH - 1) * srcStride;
int32_t* pb = _bx[(curr + 1) & 3].data;
RowCubicSumX<N>(ps, _xn, _xt, _param.dstW, _ix.data, _ax.data, pb);
next++;
}
prev = next;
const int32_t* ay = _ay.data + dy * 4;
int32_t* pb0 = _bx[(sy + 0) & 3].data;
int32_t* pb1 = _bx[(sy + 1) & 3].data;
int32_t* pb2 = _bx[(sy + 2) & 3].data;
int32_t* pb3 = _bx[(sy + 3) & 3].data;
BicubicRowInt(pb0, pb1, pb2, pb3, _bx[0].size, ay, dst);
}
}
void ResizerByteBicubic::Run(const uint8_t* src, size_t srcStride, uint8_t* dst, size_t dstStride)
{
bool sparse = _param.dstH * 3.0 <= _param.srcH;
Init(sparse);
switch (_param.channels)
{
case 1: sparse ? Sse41::ResizerByteBicubic::RunS<1>(src, srcStride, dst, dstStride) : RunB<1>(src, srcStride, dst, dstStride); return;
case 2: sparse ? RunS<2>(src, srcStride, dst, dstStride) : RunB<2>(src, srcStride, dst, dstStride); return;
case 3: sparse ? RunS<3>(src, srcStride, dst, dstStride) : RunB<3>(src, srcStride, dst, dstStride); return;
case 4: sparse ? RunS<4>(src, srcStride, dst, dstStride) : RunB<4>(src, srcStride, dst, dstStride); return;
default:
assert(0);
}
}
}
#endif //SIMD_AVX2_ENABLE
}
|
