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
// Copyright 2011 Google Inc. All Rights Reserved.
//
// Use of this source code is governed by a BSD-style license
// that can be found in the COPYING file in the root of the source
// tree. An additional intellectual property rights grant can be found
// in the file PATENTS. All contributing project authors may
// be found in the AUTHORS file in the root of the source tree.
// -----------------------------------------------------------------------------
//
// SSE2 version of YUV to RGB upsampling functions.
//
// Author: somnath@google.com (Somnath Banerjee)
#include "src/dsp/dsp.h"
#if defined(WEBP_USE_SSE2)
#include <assert.h>
#include <emmintrin.h>
#include <string.h>
#include "src/dsp/cpu.h"
#include "src/dsp/yuv.h"
#include "src/webp/decode.h"
#include "src/webp/types.h"
#ifdef FANCY_UPSAMPLING
// We compute (9*a + 3*b + 3*c + d + 8) / 16 as follows
// u = (9*a + 3*b + 3*c + d + 8) / 16
// = (a + (a + 3*b + 3*c + d) / 8 + 1) / 2
// = (a + m + 1) / 2
// where m = (a + 3*b + 3*c + d) / 8
// = ((a + b + c + d) / 2 + b + c) / 4
//
// Let's say k = (a + b + c + d) / 4.
// We can compute k as
// k = (s + t + 1) / 2 - ((a^d) | (b^c) | (s^t)) & 1
// where s = (a + d + 1) / 2 and t = (b + c + 1) / 2
//
// Then m can be written as
// m = (k + t + 1) / 2 - (((b^c) & (s^t)) | (k^t)) & 1
// Computes out = (k + in + 1) / 2 - ((ij & (s^t)) | (k^in)) & 1
#define GET_M(ij, in, out) \
do { \
const __m128i tmp0 = _mm_avg_epu8(k, (in)); /* (k + in + 1) / 2 */ \
const __m128i tmp1 = _mm_and_si128((ij), st); /* (ij) & (s^t) */ \
const __m128i tmp2 = _mm_xor_si128(k, (in)); /* (k^in) */ \
const __m128i tmp3 = \
_mm_or_si128(tmp1, tmp2); /* ((ij) & (s^t)) | (k^in) */ \
const __m128i tmp4 = _mm_and_si128(tmp3, one); /* & 1 -> lsb_correction */ \
(out) = _mm_sub_epi8(tmp0, tmp4); /* (k + in + 1) / 2 - lsb_correction */ \
} while (0)
// pack and store two alternating pixel rows
#define PACK_AND_STORE(a, b, da, db, out) \
do { \
const __m128i t_a = \
_mm_avg_epu8(a, da); /* (9a + 3b + 3c + d + 8) / 16 */ \
const __m128i t_b = \
_mm_avg_epu8(b, db); /* (3a + 9b + c + 3d + 8) / 16 */ \
const __m128i t_1 = _mm_unpacklo_epi8(t_a, t_b); \
const __m128i t_2 = _mm_unpackhi_epi8(t_a, t_b); \
_mm_store_si128(((__m128i*)(out)) + 0, t_1); \
_mm_store_si128(((__m128i*)(out)) + 1, t_2); \
} while (0)
// Loads 17 pixels each from rows r1 and r2 and generates 32 pixels.
#define UPSAMPLE_32PIXELS(r1, r2, out) \
do { \
const __m128i one = _mm_set1_epi8(1); \
const __m128i a = _mm_loadu_si128((const __m128i*)&(r1)[0]); \
const __m128i b = _mm_loadu_si128((const __m128i*)&(r1)[1]); \
const __m128i c = _mm_loadu_si128((const __m128i*)&(r2)[0]); \
const __m128i d = _mm_loadu_si128((const __m128i*)&(r2)[1]); \
\
const __m128i s = _mm_avg_epu8(a, d); /* s = (a + d + 1) / 2 */ \
const __m128i t = _mm_avg_epu8(b, c); /* t = (b + c + 1) / 2 */ \
const __m128i st = _mm_xor_si128(s, t); /* st = s^t */ \
\
const __m128i ad = _mm_xor_si128(a, d); /* ad = a^d */ \
const __m128i bc = _mm_xor_si128(b, c); /* bc = b^c */ \
\
const __m128i t1 = _mm_or_si128(ad, bc); /* (a^d) | (b^c) */ \
const __m128i t2 = _mm_or_si128(t1, st); /* (a^d) | (b^c) | (s^t) */ \
const __m128i t3 = _mm_and_si128(t2, one); /* (a^d) | (b^c) | (s^t) & 1 */ \
const __m128i t4 = _mm_avg_epu8(s, t); \
const __m128i k = _mm_sub_epi8(t4, t3); /* k = (a + b + c + d) / 4 */ \
__m128i diag1, diag2; \
\
GET_M(bc, t, diag1); /* diag1 = (a + 3b + 3c + d) / 8 */ \
GET_M(ad, s, diag2); /* diag2 = (3a + b + c + 3d) / 8 */ \
\
/* pack the alternate pixels */ \
PACK_AND_STORE(a, b, diag1, diag2, (out) + 0); /* store top */ \
PACK_AND_STORE(c, d, diag2, diag1, (out) + 2 * 32); /* store bottom */ \
} while (0)
// Turn the macro into a function for reducing code-size when non-critical
static void Upsample32Pixels_SSE2(const uint8_t* WEBP_RESTRICT const r1,
const uint8_t* WEBP_RESTRICT const r2,
uint8_t* WEBP_RESTRICT const out) {
UPSAMPLE_32PIXELS(r1, r2, out);
}
#define UPSAMPLE_LAST_BLOCK(tb, bb, num_pixels, out) \
{ \
uint8_t r1[17], r2[17]; \
memcpy(r1, (tb), (num_pixels)); \
memcpy(r2, (bb), (num_pixels)); \
/* replicate last byte */ \
memset(r1 + (num_pixels), r1[(num_pixels) - 1], 17 - (num_pixels)); \
memset(r2 + (num_pixels), r2[(num_pixels) - 1], 17 - (num_pixels)); \
/* using the shared function instead of the macro saves ~3k code size */ \
Upsample32Pixels_SSE2(r1, r2, out); \
}
#define CONVERT2RGB_32(FUNC, XSTEP, top_y, bottom_y, top_dst, bottom_dst, \
cur_x) \
do { \
FUNC##32_SSE2((top_y) + (cur_x), r_u, r_v, (top_dst) + (cur_x) * (XSTEP)); \
if ((bottom_y) != NULL) { \
FUNC##32_SSE2((bottom_y) + (cur_x), r_u + 64, r_v + 64, \
(bottom_dst) + (cur_x) * (XSTEP)); \
} \
} while (0)
#define SSE2_UPSAMPLE_FUNC(FUNC_NAME, FUNC, XSTEP) \
static void FUNC_NAME( \
const uint8_t* WEBP_RESTRICT top_y, \
const uint8_t* WEBP_RESTRICT bottom_y, \
const uint8_t* WEBP_RESTRICT top_u, const uint8_t* WEBP_RESTRICT top_v, \
const uint8_t* WEBP_RESTRICT cur_u, const uint8_t* WEBP_RESTRICT cur_v, \
uint8_t* WEBP_RESTRICT top_dst, uint8_t* WEBP_RESTRICT bottom_dst, \
int len) { \
int uv_pos, pos; \
/* 16byte-aligned array to cache reconstructed u and v */ \
uint8_t uv_buf[14 * 32 + 15] = {0}; \
uint8_t* const r_u = \
(uint8_t*)((uintptr_t)(uv_buf + 15) & ~(uintptr_t)15); \
uint8_t* const r_v = r_u + 32; \
\
assert(top_y != NULL); \
{ /* Treat the first pixel in regular way */ \
const int u_diag = ((top_u[0] + cur_u[0]) >> 1) + 1; \
const int v_diag = ((top_v[0] + cur_v[0]) >> 1) + 1; \
const int u0_t = (top_u[0] + u_diag) >> 1; \
const int v0_t = (top_v[0] + v_diag) >> 1; \
FUNC(top_y[0], u0_t, v0_t, top_dst); \
if (bottom_y != NULL) { \
const int u0_b = (cur_u[0] + u_diag) >> 1; \
const int v0_b = (cur_v[0] + v_diag) >> 1; \
FUNC(bottom_y[0], u0_b, v0_b, bottom_dst); \
} \
} \
/* For UPSAMPLE_32PIXELS, 17 u/v values must be read-able for each block \
*/ \
for (pos = 1, uv_pos = 0; pos + 32 + 1 <= len; pos += 32, uv_pos += 16) { \
UPSAMPLE_32PIXELS(top_u + uv_pos, cur_u + uv_pos, r_u); \
UPSAMPLE_32PIXELS(top_v + uv_pos, cur_v + uv_pos, r_v); \
CONVERT2RGB_32(FUNC, XSTEP, top_y, bottom_y, top_dst, bottom_dst, pos); \
} \
if (len > 1) { \
const int left_over = ((len + 1) >> 1) - (pos >> 1); \
uint8_t* const tmp_top_dst = r_u + 4 * 32; \
uint8_t* const tmp_bottom_dst = tmp_top_dst + 4 * 32; \
uint8_t* const tmp_top = tmp_bottom_dst + 4 * 32; \
uint8_t* const tmp_bottom = (bottom_y == NULL) ? NULL : tmp_top + 32; \
assert(left_over > 0); \
UPSAMPLE_LAST_BLOCK(top_u + uv_pos, cur_u + uv_pos, left_over, r_u); \
UPSAMPLE_LAST_BLOCK(top_v + uv_pos, cur_v + uv_pos, left_over, r_v); \
memcpy(tmp_top, top_y + pos, len - pos); \
if (bottom_y != NULL) memcpy(tmp_bottom, bottom_y + pos, len - pos); \
CONVERT2RGB_32(FUNC, XSTEP, tmp_top, tmp_bottom, tmp_top_dst, \
tmp_bottom_dst, 0); \
memcpy(top_dst + pos * (XSTEP), tmp_top_dst, (len - pos) * (XSTEP)); \
if (bottom_y != NULL) { \
memcpy(bottom_dst + pos * (XSTEP), tmp_bottom_dst, \
(len - pos) * (XSTEP)); \
} \
} \
}
// SSE2 variants of the fancy upsampler.
SSE2_UPSAMPLE_FUNC(UpsampleRgbaLinePair_SSE2, VP8YuvToRgba, 4)
SSE2_UPSAMPLE_FUNC(UpsampleBgraLinePair_SSE2, VP8YuvToBgra, 4)
#if !defined(WEBP_REDUCE_CSP)
SSE2_UPSAMPLE_FUNC(UpsampleRgbLinePair_SSE2, VP8YuvToRgb, 3)
SSE2_UPSAMPLE_FUNC(UpsampleBgrLinePair_SSE2, VP8YuvToBgr, 3)
SSE2_UPSAMPLE_FUNC(UpsampleArgbLinePair_SSE2, VP8YuvToArgb, 4)
SSE2_UPSAMPLE_FUNC(UpsampleRgba4444LinePair_SSE2, VP8YuvToRgba4444, 2)
SSE2_UPSAMPLE_FUNC(UpsampleRgb565LinePair_SSE2, VP8YuvToRgb565, 2)
#endif // WEBP_REDUCE_CSP
#undef GET_M
#undef PACK_AND_STORE
#undef UPSAMPLE_32PIXELS
#undef UPSAMPLE_LAST_BLOCK
#undef CONVERT2RGB
#undef CONVERT2RGB_32
#undef SSE2_UPSAMPLE_FUNC
//------------------------------------------------------------------------------
// Entry point
extern WebPUpsampleLinePairFunc WebPUpsamplers[/* MODE_LAST */];
extern void WebPInitUpsamplersSSE2(void);
WEBP_TSAN_IGNORE_FUNCTION void WebPInitUpsamplersSSE2(void) {
WebPUpsamplers[MODE_RGBA] = UpsampleRgbaLinePair_SSE2;
WebPUpsamplers[MODE_BGRA] = UpsampleBgraLinePair_SSE2;
WebPUpsamplers[MODE_rgbA] = UpsampleRgbaLinePair_SSE2;
WebPUpsamplers[MODE_bgrA] = UpsampleBgraLinePair_SSE2;
#if !defined(WEBP_REDUCE_CSP)
WebPUpsamplers[MODE_RGB] = UpsampleRgbLinePair_SSE2;
WebPUpsamplers[MODE_BGR] = UpsampleBgrLinePair_SSE2;
WebPUpsamplers[MODE_ARGB] = UpsampleArgbLinePair_SSE2;
WebPUpsamplers[MODE_Argb] = UpsampleArgbLinePair_SSE2;
WebPUpsamplers[MODE_RGB_565] = UpsampleRgb565LinePair_SSE2;
WebPUpsamplers[MODE_RGBA_4444] = UpsampleRgba4444LinePair_SSE2;
WebPUpsamplers[MODE_rgbA_4444] = UpsampleRgba4444LinePair_SSE2;
#endif // WEBP_REDUCE_CSP
}
#endif // FANCY_UPSAMPLING
//------------------------------------------------------------------------------
extern WebPYUV444Converter WebPYUV444Converters[/* MODE_LAST */];
extern void WebPInitYUV444ConvertersSSE2(void);
#define YUV444_FUNC(FUNC_NAME, CALL, CALL_C, XSTEP) \
extern void CALL_C( \
const uint8_t* WEBP_RESTRICT y, const uint8_t* WEBP_RESTRICT u, \
const uint8_t* WEBP_RESTRICT v, uint8_t* WEBP_RESTRICT dst, int len); \
static void FUNC_NAME( \
const uint8_t* WEBP_RESTRICT y, const uint8_t* WEBP_RESTRICT u, \
const uint8_t* WEBP_RESTRICT v, uint8_t* WEBP_RESTRICT dst, int len) { \
int i; \
const int max_len = len & ~31; \
for (i = 0; i < max_len; i += 32) { \
CALL(y + i, u + i, v + i, dst + i * (XSTEP)); \
} \
if (i < len) { /* C-fallback */ \
CALL_C(y + i, u + i, v + i, dst + i * (XSTEP), len - i); \
} \
}
YUV444_FUNC(Yuv444ToRgba_SSE2, VP8YuvToRgba32_SSE2, WebPYuv444ToRgba_C, 4)
YUV444_FUNC(Yuv444ToBgra_SSE2, VP8YuvToBgra32_SSE2, WebPYuv444ToBgra_C, 4)
#if !defined(WEBP_REDUCE_CSP)
YUV444_FUNC(Yuv444ToRgb_SSE2, VP8YuvToRgb32_SSE2, WebPYuv444ToRgb_C, 3)
YUV444_FUNC(Yuv444ToBgr_SSE2, VP8YuvToBgr32_SSE2, WebPYuv444ToBgr_C, 3)
YUV444_FUNC(Yuv444ToArgb_SSE2, VP8YuvToArgb32_SSE2, WebPYuv444ToArgb_C, 4)
YUV444_FUNC(Yuv444ToRgba4444_SSE2, VP8YuvToRgba444432_SSE2,
WebPYuv444ToRgba4444_C, 2)
YUV444_FUNC(Yuv444ToRgb565_SSE2, VP8YuvToRgb56532_SSE2, WebPYuv444ToRgb565_C, 2)
#endif // WEBP_REDUCE_CSP
WEBP_TSAN_IGNORE_FUNCTION void WebPInitYUV444ConvertersSSE2(void) {
WebPYUV444Converters[MODE_RGBA] = Yuv444ToRgba_SSE2;
WebPYUV444Converters[MODE_BGRA] = Yuv444ToBgra_SSE2;
WebPYUV444Converters[MODE_rgbA] = Yuv444ToRgba_SSE2;
WebPYUV444Converters[MODE_bgrA] = Yuv444ToBgra_SSE2;
#if !defined(WEBP_REDUCE_CSP)
WebPYUV444Converters[MODE_RGB] = Yuv444ToRgb_SSE2;
WebPYUV444Converters[MODE_BGR] = Yuv444ToBgr_SSE2;
WebPYUV444Converters[MODE_ARGB] = Yuv444ToArgb_SSE2;
WebPYUV444Converters[MODE_RGBA_4444] = Yuv444ToRgba4444_SSE2;
WebPYUV444Converters[MODE_RGB_565] = Yuv444ToRgb565_SSE2;
WebPYUV444Converters[MODE_Argb] = Yuv444ToArgb_SSE2;
WebPYUV444Converters[MODE_rgbA_4444] = Yuv444ToRgba4444_SSE2;
#endif // WEBP_REDUCE_CSP
}
#else
WEBP_DSP_INIT_STUB(WebPInitYUV444ConvertersSSE2)
#endif // WEBP_USE_SSE2
#if !(defined(FANCY_UPSAMPLING) && defined(WEBP_USE_SSE2))
WEBP_DSP_INIT_STUB(WebPInitUpsamplersSSE2)
#endif