row_gcc.cc 236 KB
Newer Older
1
// VERSION 2
mikhal@webrtc.org's avatar
mikhal@webrtc.org committed
2
/*
3
 *  Copyright 2011 The LibYuv Project Authors. All rights reserved.
mikhal@webrtc.org's avatar
mikhal@webrtc.org committed
4 5 6 7
 *
 *  Use of this source code is governed by a BSD-style license
 *  that can be found in the LICENSE file in the root of the source
 *  tree. An additional intellectual property rights grant can be found
8
 *  in the file PATENTS. All contributing project authors may
mikhal@webrtc.org's avatar
mikhal@webrtc.org committed
9 10 11
 *  be found in the AUTHORS file in the root of the source tree.
 */

12
#include "libyuv/row.h"
mikhal@webrtc.org's avatar
mikhal@webrtc.org committed
13

14 15
#ifdef __cplusplus
namespace libyuv {
mikhal@webrtc.org's avatar
mikhal@webrtc.org committed
16
extern "C" {
17
#endif
mikhal@webrtc.org's avatar
mikhal@webrtc.org committed
18

19
// This module is for GCC x86 and x64.
20 21
#if !defined(LIBYUV_DISABLE_X86) && \
    (defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER)))
22

23
#if defined(HAS_ARGBTOYROW_SSSE3) || defined(HAS_ARGBGRAYROW_SSSE3)
24 25

// Constants for ARGB
26 27
static vec8 kARGBToY = {13, 65, 33, 0, 13, 65, 33, 0,
                        13, 65, 33, 0, 13, 65, 33, 0};
28

29
// JPeg full range.
30 31
static vec8 kARGBToYJ = {15, 75, 38, 0, 15, 75, 38, 0,
                         15, 75, 38, 0, 15, 75, 38, 0};
32 33 34
#endif  // defined(HAS_ARGBTOYROW_SSSE3) || defined(HAS_ARGBGRAYROW_SSSE3)

#if defined(HAS_ARGBTOYROW_SSSE3) || defined(HAS_I422TOARGBROW_SSSE3)
35

36 37
static vec8 kARGBToU = {112, -74, -38, 0, 112, -74, -38, 0,
                        112, -74, -38, 0, 112, -74, -38, 0};
38

39 40
static vec8 kARGBToUJ = {127, -84, -43, 0, 127, -84, -43, 0,
                         127, -84, -43, 0, 127, -84, -43, 0};
41

42
static vec8 kARGBToV = {
43
    -18, -94, 112, 0, -18, -94, 112, 0, -18, -94, 112, 0, -18, -94, 112, 0,
44
};
45

46 47
static vec8 kARGBToVJ = {-20, -107, 127, 0, -20, -107, 127, 0,
                         -20, -107, 127, 0, -20, -107, 127, 0};
48

49
// Constants for BGRA
50 51
static vec8 kBGRAToY = {0, 33, 65, 13, 0, 33, 65, 13,
                        0, 33, 65, 13, 0, 33, 65, 13};
52

53 54
static vec8 kBGRAToU = {0, -38, -74, 112, 0, -38, -74, 112,
                        0, -38, -74, 112, 0, -38, -74, 112};
55

56 57
static vec8 kBGRAToV = {0, 112, -94, -18, 0, 112, -94, -18,
                        0, 112, -94, -18, 0, 112, -94, -18};
58 59

// Constants for ABGR
60 61
static vec8 kABGRToY = {33, 65, 13, 0, 33, 65, 13, 0,
                        33, 65, 13, 0, 33, 65, 13, 0};
62

63 64
static vec8 kABGRToU = {-38, -74, 112, 0, -38, -74, 112, 0,
                        -38, -74, 112, 0, -38, -74, 112, 0};
65

66 67
static vec8 kABGRToV = {112, -94, -18, 0, 112, -94, -18, 0,
                        112, -94, -18, 0, 112, -94, -18, 0};
68

69
// Constants for RGBA.
70 71
static vec8 kRGBAToY = {0, 13, 65, 33, 0, 13, 65, 33,
                        0, 13, 65, 33, 0, 13, 65, 33};
72

73 74
static vec8 kRGBAToU = {0, 112, -74, -38, 0, 112, -74, -38,
                        0, 112, -74, -38, 0, 112, -74, -38};
75

76 77
static vec8 kRGBAToV = {0, -18, -94, 112, 0, -18, -94, 112,
                        0, -18, -94, 112, 0, -18, -94, 112};
78

79 80
static uvec8 kAddY16 = {16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u,
                        16u, 16u, 16u, 16u, 16u, 16u, 16u, 16u};
81

82
// 7 bit fixed point 0.5.
83
static vec16 kAddYJ64 = {64, 64, 64, 64, 64, 64, 64, 64};
84

85 86
static uvec8 kAddUV128 = {128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u,
                          128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u};
87

88 89
static uvec16 kAddUVJ128 = {0x8080u, 0x8080u, 0x8080u, 0x8080u,
                            0x8080u, 0x8080u, 0x8080u, 0x8080u};
90 91 92
#endif  // defined(HAS_ARGBTOYROW_SSSE3) || defined(HAS_I422TOARGBROW_SSSE3)

#ifdef HAS_RGB24TOARGBROW_SSSE3
93

94
// Shuffle table for converting RGB24 to ARGB.
95 96
static uvec8 kShuffleMaskRGB24ToARGB = {0u, 1u, 2u, 12u, 3u, 4u,  5u,  13u,
                                        6u, 7u, 8u, 14u, 9u, 10u, 11u, 15u};
97 98

// Shuffle table for converting RAW to ARGB.
99 100
static uvec8 kShuffleMaskRAWToARGB = {2u, 1u, 0u, 12u, 5u,  4u,  3u, 13u,
                                      8u, 7u, 6u, 14u, 11u, 10u, 9u, 15u};
101

102 103
// Shuffle table for converting RAW to RGB24.  First 8.
static const uvec8 kShuffleMaskRAWToRGB24_0 = {
104 105
    2u,   1u,   0u,   5u,   4u,   3u,   8u,   7u,
    128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u};
106 107 108

// Shuffle table for converting RAW to RGB24.  Middle 8.
static const uvec8 kShuffleMaskRAWToRGB24_1 = {
109 110
    2u,   7u,   6u,   5u,   10u,  9u,   8u,   13u,
    128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u};
111 112 113

// Shuffle table for converting RAW to RGB24.  Last 8.
static const uvec8 kShuffleMaskRAWToRGB24_2 = {
114 115
    8u,   7u,   12u,  11u,  10u,  15u,  14u,  13u,
    128u, 128u, 128u, 128u, 128u, 128u, 128u, 128u};
116

fbarchard@google.com's avatar
fbarchard@google.com committed
117
// Shuffle table for converting ARGB to RGB24.
118
static uvec8 kShuffleMaskARGBToRGB24 = {
119
    0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 10u, 12u, 13u, 14u, 128u, 128u, 128u, 128u};
fbarchard@google.com's avatar
fbarchard@google.com committed
120 121

// Shuffle table for converting ARGB to RAW.
122
static uvec8 kShuffleMaskARGBToRAW = {
123
    2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 8u, 14u, 13u, 12u, 128u, 128u, 128u, 128u};
fbarchard@google.com's avatar
fbarchard@google.com committed
124

125
// Shuffle table for converting ARGBToRGB24 for I422ToRGB24.  First 8 + next 4
126
static uvec8 kShuffleMaskARGBToRGB24_0 = {
127
    0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 128u, 128u, 128u, 128u, 10u, 12u, 13u, 14u};
128

129
// YUY2 shuf 16 Y to 32 Y.
130 131 132
static const lvec8 kShuffleYUY2Y = {0,  0,  2,  2,  4,  4,  6,  6,  8,  8, 10,
                                    10, 12, 12, 14, 14, 0,  0,  2,  2,  4, 4,
                                    6,  6,  8,  8,  10, 10, 12, 12, 14, 14};
133 134

// YUY2 shuf 8 UV to 16 UV.
135 136 137
static const lvec8 kShuffleYUY2UV = {1,  3,  1,  3,  5,  7,  5,  7,  9,  11, 9,
                                     11, 13, 15, 13, 15, 1,  3,  1,  3,  5,  7,
                                     5,  7,  9,  11, 9,  11, 13, 15, 13, 15};
138 139

// UYVY shuf 16 Y to 32 Y.
140 141 142
static const lvec8 kShuffleUYVYY = {1,  1,  3,  3,  5,  5,  7,  7,  9,  9, 11,
                                    11, 13, 13, 15, 15, 1,  1,  3,  3,  5, 5,
                                    7,  7,  9,  9,  11, 11, 13, 13, 15, 15};
143 144

// UYVY shuf 8 UV to 16 UV.
145 146 147
static const lvec8 kShuffleUYVYUV = {0,  2,  0,  2,  4,  6,  4,  6,  8,  10, 8,
                                     10, 12, 14, 12, 14, 0,  2,  0,  2,  4,  6,
                                     4,  6,  8,  10, 8,  10, 12, 14, 12, 14};
148 149 150

// NV21 shuf 8 VU to 16 UV.
static const lvec8 kShuffleNV21 = {
151 152
    1, 0, 1, 0, 3, 2, 3, 2, 5, 4, 5, 4, 7, 6, 7, 6,
    1, 0, 1, 0, 3, 2, 3, 2, 5, 4, 5, 4, 7, 6, 7, 6,
153
};
154
#endif  // HAS_RGB24TOARGBROW_SSSE3
155

156
#ifdef HAS_J400TOARGBROW_SSE2
157
void J400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
158
  asm volatile (
159 160
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    "pslld     $0x18,%%xmm5                    \n"
161
    LABELALIGN
162
  "1:                                          \n"
163 164
    "movq      " MEMACCESS(0) ",%%xmm0         \n"
    "lea       " MEMLEA(0x8,0) ",%0            \n"
165 166 167 168 169 170
    "punpcklbw %%xmm0,%%xmm0                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "punpcklwd %%xmm0,%%xmm0                   \n"
    "punpckhwd %%xmm1,%%xmm1                   \n"
    "por       %%xmm5,%%xmm0                   \n"
    "por       %%xmm5,%%xmm1                   \n"
171 172
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
173
    "lea       " MEMLEA(0x20,1) ",%1           \n"
174
    "sub       $0x8,%2                         \n"
175
    "jg        1b                              \n"
176 177
  : "+r"(src_y),     // %0
    "+r"(dst_argb),  // %1
178
    "+r"(width)        // %2
179
  :: "memory", "cc", "xmm0", "xmm1", "xmm5"
fbarchard@google.com's avatar
fbarchard@google.com committed
180
  );
181
}
182
#endif  // HAS_J400TOARGBROW_SSE2
183

184
#ifdef HAS_RGB24TOARGBROW_SSSE3
185
void RGB24ToARGBRow_SSSE3(const uint8* src_rgb24, uint8* dst_argb, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
186
  asm volatile (
187 188 189
    "pcmpeqb   %%xmm5,%%xmm5                   \n"  // generate mask 0xff000000
    "pslld     $0x18,%%xmm5                    \n"
    "movdqa    %3,%%xmm4                       \n"
190
    LABELALIGN
191
  "1:                                          \n"
192 193 194 195
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm3   \n"
    "lea       " MEMLEA(0x30,0) ",%0           \n"
196 197 198 199 200 201
    "movdqa    %%xmm3,%%xmm2                   \n"
    "palignr   $0x8,%%xmm1,%%xmm2              \n"
    "pshufb    %%xmm4,%%xmm2                   \n"
    "por       %%xmm5,%%xmm2                   \n"
    "palignr   $0xc,%%xmm0,%%xmm1              \n"
    "pshufb    %%xmm4,%%xmm0                   \n"
202
    "movdqu    %%xmm2," MEMACCESS2(0x20,1) "   \n"
203 204
    "por       %%xmm5,%%xmm0                   \n"
    "pshufb    %%xmm4,%%xmm1                   \n"
205
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
206 207 208
    "por       %%xmm5,%%xmm1                   \n"
    "palignr   $0x4,%%xmm3,%%xmm3              \n"
    "pshufb    %%xmm4,%%xmm3                   \n"
209
    "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
210
    "por       %%xmm5,%%xmm3                   \n"
211
    "movdqu    %%xmm3," MEMACCESS2(0x30,1) "   \n"
212
    "lea       " MEMLEA(0x40,1) ",%1           \n"
213
    "sub       $0x10,%2                        \n"
214
    "jg        1b                              \n"
215
  : "+r"(src_rgb24),  // %0
216
    "+r"(dst_argb),  // %1
217
    "+r"(width)        // %2
218
  : "m"(kShuffleMaskRGB24ToARGB)  // %3
219
  : "memory", "cc" , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
fbarchard@google.com's avatar
fbarchard@google.com committed
220
  );
221 222
}

223
void RAWToARGBRow_SSSE3(const uint8* src_raw, uint8* dst_argb, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
224
  asm volatile (
225 226 227
    "pcmpeqb   %%xmm5,%%xmm5                   \n"  // generate mask 0xff000000
    "pslld     $0x18,%%xmm5                    \n"
    "movdqa    %3,%%xmm4                       \n"
228
    LABELALIGN
229
  "1:                                          \n"
230 231 232 233
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm3   \n"
    "lea       " MEMLEA(0x30,0) ",%0           \n"
234 235 236 237 238 239
    "movdqa    %%xmm3,%%xmm2                   \n"
    "palignr   $0x8,%%xmm1,%%xmm2              \n"
    "pshufb    %%xmm4,%%xmm2                   \n"
    "por       %%xmm5,%%xmm2                   \n"
    "palignr   $0xc,%%xmm0,%%xmm1              \n"
    "pshufb    %%xmm4,%%xmm0                   \n"
240
    "movdqu    %%xmm2," MEMACCESS2(0x20,1) "   \n"
241 242
    "por       %%xmm5,%%xmm0                   \n"
    "pshufb    %%xmm4,%%xmm1                   \n"
243
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
244 245 246
    "por       %%xmm5,%%xmm1                   \n"
    "palignr   $0x4,%%xmm3,%%xmm3              \n"
    "pshufb    %%xmm4,%%xmm3                   \n"
247
    "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
248
    "por       %%xmm5,%%xmm3                   \n"
249
    "movdqu    %%xmm3," MEMACCESS2(0x30,1) "   \n"
250
    "lea       " MEMLEA(0x40,1) ",%1           \n"
251
    "sub       $0x10,%2                        \n"
252
    "jg        1b                              \n"
253 254
  : "+r"(src_raw),   // %0
    "+r"(dst_argb),  // %1
255
    "+r"(width)        // %2
256
  : "m"(kShuffleMaskRAWToARGB)  // %3
257
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
fbarchard@google.com's avatar
fbarchard@google.com committed
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
void RAWToRGB24Row_SSSE3(const uint8* src_raw, uint8* dst_rgb24, int width) {
  asm volatile (
   "movdqa     %3,%%xmm3                       \n"
   "movdqa     %4,%%xmm4                       \n"
   "movdqa     %5,%%xmm5                       \n"
    LABELALIGN
  "1:                                          \n"
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x4,0) ",%%xmm1    \n"
    "movdqu    " MEMACCESS2(0x8,0) ",%%xmm2    \n"
    "lea       " MEMLEA(0x18,0) ",%0           \n"
    "pshufb    %%xmm3,%%xmm0                   \n"
    "pshufb    %%xmm4,%%xmm1                   \n"
    "pshufb    %%xmm5,%%xmm2                   \n"
    "movq      %%xmm0," MEMACCESS(1) "         \n"
    "movq      %%xmm1," MEMACCESS2(0x8,1) "    \n"
    "movq      %%xmm2," MEMACCESS2(0x10,1) "   \n"
    "lea       " MEMLEA(0x18,1) ",%1           \n"
    "sub       $0x8,%2                         \n"
    "jg        1b                              \n"
  : "+r"(src_raw),    // %0
    "+r"(dst_rgb24),  // %1
    "+r"(width)       // %2
  : "m"(kShuffleMaskRAWToRGB24_0),  // %3
    "m"(kShuffleMaskRAWToRGB24_1),  // %4
    "m"(kShuffleMaskRAWToRGB24_2)   // %5
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
  );
}

291
void RGB565ToARGBRow_SSE2(const uint8* src, uint8* dst, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
292
  asm volatile (
fbarchard@google.com's avatar
fbarchard@google.com committed
293 294 295
    "mov       $0x1080108,%%eax                \n"
    "movd      %%eax,%%xmm5                    \n"
    "pshufd    $0x0,%%xmm5,%%xmm5              \n"
296
    "mov       $0x20802080,%%eax               \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
297 298 299 300 301 302 303 304 305 306 307
    "movd      %%eax,%%xmm6                    \n"
    "pshufd    $0x0,%%xmm6,%%xmm6              \n"
    "pcmpeqb   %%xmm3,%%xmm3                   \n"
    "psllw     $0xb,%%xmm3                     \n"
    "pcmpeqb   %%xmm4,%%xmm4                   \n"
    "psllw     $0xa,%%xmm4                     \n"
    "psrlw     $0x5,%%xmm4                     \n"
    "pcmpeqb   %%xmm7,%%xmm7                   \n"
    "psllw     $0x8,%%xmm7                     \n"
    "sub       %0,%1                           \n"
    "sub       %0,%1                           \n"
308
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
309
  "1:                                          \n"
310
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
311 312 313 314 315 316 317 318 319 320 321 322 323 324
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm0,%%xmm2                   \n"
    "pand      %%xmm3,%%xmm1                   \n"
    "psllw     $0xb,%%xmm2                     \n"
    "pmulhuw   %%xmm5,%%xmm1                   \n"
    "pmulhuw   %%xmm5,%%xmm2                   \n"
    "psllw     $0x8,%%xmm1                     \n"
    "por       %%xmm2,%%xmm1                   \n"
    "pand      %%xmm4,%%xmm0                   \n"
    "pmulhuw   %%xmm6,%%xmm0                   \n"
    "por       %%xmm7,%%xmm0                   \n"
    "movdqa    %%xmm1,%%xmm2                   \n"
    "punpcklbw %%xmm0,%%xmm1                   \n"
    "punpckhbw %%xmm0,%%xmm2                   \n"
325 326
    MEMOPMEM(movdqu,xmm1,0x00,1,0,2)           //  movdqu  %%xmm1,(%1,%0,2)
    MEMOPMEM(movdqu,xmm2,0x10,1,0,2)           //  movdqu  %%xmm2,0x10(%1,%0,2)
327
    "lea       " MEMLEA(0x10,0) ",%0           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
328
    "sub       $0x8,%2                         \n"
329
    "jg        1b                              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
330 331
  : "+r"(src),  // %0
    "+r"(dst),  // %1
332
    "+r"(width)   // %2
fbarchard@google.com's avatar
fbarchard@google.com committed
333
  :
334 335
  : "memory", "cc", "eax", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
fbarchard@google.com's avatar
fbarchard@google.com committed
336 337 338
  );
}

339
void ARGB1555ToARGBRow_SSE2(const uint8* src, uint8* dst, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
340
  asm volatile (
fbarchard@google.com's avatar
fbarchard@google.com committed
341 342 343 344 345 346 347 348 349 350 351 352 353 354
    "mov       $0x1080108,%%eax                \n"
    "movd      %%eax,%%xmm5                    \n"
    "pshufd    $0x0,%%xmm5,%%xmm5              \n"
    "mov       $0x42004200,%%eax               \n"
    "movd      %%eax,%%xmm6                    \n"
    "pshufd    $0x0,%%xmm6,%%xmm6              \n"
    "pcmpeqb   %%xmm3,%%xmm3                   \n"
    "psllw     $0xb,%%xmm3                     \n"
    "movdqa    %%xmm3,%%xmm4                   \n"
    "psrlw     $0x6,%%xmm4                     \n"
    "pcmpeqb   %%xmm7,%%xmm7                   \n"
    "psllw     $0x8,%%xmm7                     \n"
    "sub       %0,%1                           \n"
    "sub       %0,%1                           \n"
355
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
356
  "1:                                          \n"
357
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm0,%%xmm2                   \n"
    "psllw     $0x1,%%xmm1                     \n"
    "psllw     $0xb,%%xmm2                     \n"
    "pand      %%xmm3,%%xmm1                   \n"
    "pmulhuw   %%xmm5,%%xmm2                   \n"
    "pmulhuw   %%xmm5,%%xmm1                   \n"
    "psllw     $0x8,%%xmm1                     \n"
    "por       %%xmm2,%%xmm1                   \n"
    "movdqa    %%xmm0,%%xmm2                   \n"
    "pand      %%xmm4,%%xmm0                   \n"
    "psraw     $0x8,%%xmm2                     \n"
    "pmulhuw   %%xmm6,%%xmm0                   \n"
    "pand      %%xmm7,%%xmm2                   \n"
    "por       %%xmm2,%%xmm0                   \n"
    "movdqa    %%xmm1,%%xmm2                   \n"
    "punpcklbw %%xmm0,%%xmm1                   \n"
    "punpckhbw %%xmm0,%%xmm2                   \n"
376 377
    MEMOPMEM(movdqu,xmm1,0x00,1,0,2)           //  movdqu  %%xmm1,(%1,%0,2)
    MEMOPMEM(movdqu,xmm2,0x10,1,0,2)           //  movdqu  %%xmm2,0x10(%1,%0,2)
378
    "lea       " MEMLEA(0x10,0) ",%0           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
379
    "sub       $0x8,%2                         \n"
380
    "jg        1b                              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
381 382
  : "+r"(src),  // %0
    "+r"(dst),  // %1
383
    "+r"(width)   // %2
fbarchard@google.com's avatar
fbarchard@google.com committed
384
  :
385 386
  : "memory", "cc", "eax", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
fbarchard@google.com's avatar
fbarchard@google.com committed
387 388 389
  );
}

390
void ARGB4444ToARGBRow_SSE2(const uint8* src, uint8* dst, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
391
  asm volatile (
fbarchard@google.com's avatar
fbarchard@google.com committed
392 393 394 395 396 397 398
    "mov       $0xf0f0f0f,%%eax                \n"
    "movd      %%eax,%%xmm4                    \n"
    "pshufd    $0x0,%%xmm4,%%xmm4              \n"
    "movdqa    %%xmm4,%%xmm5                   \n"
    "pslld     $0x4,%%xmm5                     \n"
    "sub       %0,%1                           \n"
    "sub       %0,%1                           \n"
399
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
400
  "1:                                          \n"
401
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
402 403 404 405 406 407 408 409 410 411 412 413
    "movdqa    %%xmm0,%%xmm2                   \n"
    "pand      %%xmm4,%%xmm0                   \n"
    "pand      %%xmm5,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm2,%%xmm3                   \n"
    "psllw     $0x4,%%xmm1                     \n"
    "psrlw     $0x4,%%xmm3                     \n"
    "por       %%xmm1,%%xmm0                   \n"
    "por       %%xmm3,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "punpcklbw %%xmm2,%%xmm0                   \n"
    "punpckhbw %%xmm2,%%xmm1                   \n"
414 415
    MEMOPMEM(movdqu,xmm0,0x00,1,0,2)           //  movdqu  %%xmm0,(%1,%0,2)
    MEMOPMEM(movdqu,xmm1,0x10,1,0,2)           //  movdqu  %%xmm1,0x10(%1,%0,2)
416
    "lea       " MEMLEA(0x10,0) ",%0           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
417
    "sub       $0x8,%2                         \n"
418
    "jg        1b                              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
419 420
  : "+r"(src),  // %0
    "+r"(dst),  // %1
421
    "+r"(width)   // %2
fbarchard@google.com's avatar
fbarchard@google.com committed
422
  :
423 424
  : "memory", "cc", "eax", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
fbarchard@google.com's avatar
fbarchard@google.com committed
425
  );
426 427
}

428
void ARGBToRGB24Row_SSSE3(const uint8* src, uint8* dst, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
429
  asm volatile (
fbarchard@google.com's avatar
fbarchard@google.com committed
430
    "movdqa    %3,%%xmm6                       \n"
431
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
432
  "1:                                          \n"
433 434 435 436 437
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm3   \n"
    "lea       " MEMLEA(0x40,0) ",%0           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
438 439 440 441 442 443 444 445 446 447
    "pshufb    %%xmm6,%%xmm0                   \n"
    "pshufb    %%xmm6,%%xmm1                   \n"
    "pshufb    %%xmm6,%%xmm2                   \n"
    "pshufb    %%xmm6,%%xmm3                   \n"
    "movdqa    %%xmm1,%%xmm4                   \n"
    "psrldq    $0x4,%%xmm1                     \n"
    "pslldq    $0xc,%%xmm4                     \n"
    "movdqa    %%xmm2,%%xmm5                   \n"
    "por       %%xmm4,%%xmm0                   \n"
    "pslldq    $0x8,%%xmm5                     \n"
448
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
449 450 451 452
    "por       %%xmm5,%%xmm1                   \n"
    "psrldq    $0x8,%%xmm2                     \n"
    "pslldq    $0x4,%%xmm3                     \n"
    "por       %%xmm3,%%xmm2                   \n"
453 454 455
    "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
    "movdqu    %%xmm2," MEMACCESS2(0x20,1) "   \n"
    "lea       " MEMLEA(0x30,1) ",%1           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
456
    "sub       $0x10,%2                        \n"
457
    "jg        1b                              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
458 459
  : "+r"(src),  // %0
    "+r"(dst),  // %1
460
    "+r"(width)   // %2
fbarchard@google.com's avatar
fbarchard@google.com committed
461
  : "m"(kShuffleMaskARGBToRGB24)  // %3
462
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
fbarchard@google.com's avatar
fbarchard@google.com committed
463 464 465
  );
}

466
void ARGBToRAWRow_SSSE3(const uint8* src, uint8* dst, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
467
  asm volatile (
fbarchard@google.com's avatar
fbarchard@google.com committed
468
    "movdqa    %3,%%xmm6                       \n"
469
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
470
  "1:                                          \n"
471 472 473 474 475
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm3   \n"
    "lea       " MEMLEA(0x40,0) ",%0           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
476 477 478 479 480 481 482 483 484 485
    "pshufb    %%xmm6,%%xmm0                   \n"
    "pshufb    %%xmm6,%%xmm1                   \n"
    "pshufb    %%xmm6,%%xmm2                   \n"
    "pshufb    %%xmm6,%%xmm3                   \n"
    "movdqa    %%xmm1,%%xmm4                   \n"
    "psrldq    $0x4,%%xmm1                     \n"
    "pslldq    $0xc,%%xmm4                     \n"
    "movdqa    %%xmm2,%%xmm5                   \n"
    "por       %%xmm4,%%xmm0                   \n"
    "pslldq    $0x8,%%xmm5                     \n"
486
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
487 488 489 490
    "por       %%xmm5,%%xmm1                   \n"
    "psrldq    $0x8,%%xmm2                     \n"
    "pslldq    $0x4,%%xmm3                     \n"
    "por       %%xmm3,%%xmm2                   \n"
491 492 493
    "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
    "movdqu    %%xmm2," MEMACCESS2(0x20,1) "   \n"
    "lea       " MEMLEA(0x30,1) ",%1           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
494
    "sub       $0x10,%2                        \n"
495
    "jg        1b                              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
496 497
  : "+r"(src),  // %0
    "+r"(dst),  // %1
498
    "+r"(width)   // %2
fbarchard@google.com's avatar
fbarchard@google.com committed
499
  : "m"(kShuffleMaskARGBToRAW)  // %3
500
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
fbarchard@google.com's avatar
fbarchard@google.com committed
501 502 503
  );
}

504
void ARGBToRGB565Row_SSE2(const uint8* src, uint8* dst, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
505
  asm volatile (
fbarchard@google.com's avatar
fbarchard@google.com committed
506 507 508 509 510 511 512
    "pcmpeqb   %%xmm3,%%xmm3                   \n"
    "psrld     $0x1b,%%xmm3                    \n"
    "pcmpeqb   %%xmm4,%%xmm4                   \n"
    "psrld     $0x1a,%%xmm4                    \n"
    "pslld     $0x5,%%xmm4                     \n"
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    "pslld     $0xb,%%xmm5                     \n"
513
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
514
  "1:                                          \n"
515
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
516 517 518 519 520 521 522 523 524 525 526 527
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm0,%%xmm2                   \n"
    "pslld     $0x8,%%xmm0                     \n"
    "psrld     $0x3,%%xmm1                     \n"
    "psrld     $0x5,%%xmm2                     \n"
    "psrad     $0x10,%%xmm0                    \n"
    "pand      %%xmm3,%%xmm1                   \n"
    "pand      %%xmm4,%%xmm2                   \n"
    "pand      %%xmm5,%%xmm0                   \n"
    "por       %%xmm2,%%xmm1                   \n"
    "por       %%xmm1,%%xmm0                   \n"
    "packssdw  %%xmm0,%%xmm0                   \n"
528 529 530
    "lea       " MEMLEA(0x10,0) ",%0           \n"
    "movq      %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x8,1) ",%1            \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
531
    "sub       $0x4,%2                         \n"
532
    "jg        1b                              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
533 534
  : "+r"(src),  // %0
    "+r"(dst),  // %1
535
    "+r"(width)   // %2
536
  :: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
fbarchard@google.com's avatar
fbarchard@google.com committed
537
  );
538 539
}

540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584
void ARGBToRGB565DitherRow_SSE2(const uint8* src,
                                uint8* dst,
                                const uint32 dither4,
                                int width) {
  asm volatile(
      "movd       %3,%%xmm6                      \n"
      "punpcklbw  %%xmm6,%%xmm6                  \n"
      "movdqa     %%xmm6,%%xmm7                  \n"
      "punpcklwd  %%xmm6,%%xmm6                  \n"
      "punpckhwd  %%xmm7,%%xmm7                  \n"
      "pcmpeqb    %%xmm3,%%xmm3                  \n"
      "psrld      $0x1b,%%xmm3                   \n"
      "pcmpeqb    %%xmm4,%%xmm4                  \n"
      "psrld      $0x1a,%%xmm4                   \n"
      "pslld      $0x5,%%xmm4                    \n"
      "pcmpeqb    %%xmm5,%%xmm5                  \n"
      "pslld      $0xb,%%xmm5                    \n"

      LABELALIGN
      "1:                                          \n"
      "movdqu     (%0),%%xmm0                    \n"
      "paddusb    %%xmm6,%%xmm0                  \n"
      "movdqa     %%xmm0,%%xmm1                  \n"
      "movdqa     %%xmm0,%%xmm2                  \n"
      "pslld      $0x8,%%xmm0                    \n"
      "psrld      $0x3,%%xmm1                    \n"
      "psrld      $0x5,%%xmm2                    \n"
      "psrad      $0x10,%%xmm0                   \n"
      "pand       %%xmm3,%%xmm1                  \n"
      "pand       %%xmm4,%%xmm2                  \n"
      "pand       %%xmm5,%%xmm0                  \n"
      "por        %%xmm2,%%xmm1                  \n"
      "por        %%xmm1,%%xmm0                  \n"
      "packssdw   %%xmm0,%%xmm0                  \n"
      "lea        0x10(%0),%0                    \n"
      "movq       %%xmm0,(%1)                    \n"
      "lea        0x8(%1),%1                     \n"
      "sub        $0x4,%2                        \n"
      "jg        1b                              \n"
      : "+r"(src),    // %0
        "+r"(dst),    // %1
        "+r"(width)   // %2
      : "m"(dither4)  // %3
      : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
        "xmm7");
fbarchard@google.com's avatar
fbarchard@google.com committed
585 586
}

587
#ifdef HAS_ARGBTORGB565DITHERROW_AVX2
588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629
void ARGBToRGB565DitherRow_AVX2(const uint8* src,
                                uint8* dst,
                                const uint32 dither4,
                                int width) {
  asm volatile(
      "vbroadcastss %3,%%xmm6                    \n"
      "vpunpcklbw %%xmm6,%%xmm6,%%xmm6           \n"
      "vpermq     $0xd8,%%ymm6,%%ymm6            \n"
      "vpunpcklwd %%ymm6,%%ymm6,%%ymm6           \n"
      "vpcmpeqb   %%ymm3,%%ymm3,%%ymm3           \n"
      "vpsrld     $0x1b,%%ymm3,%%ymm3            \n"
      "vpcmpeqb   %%ymm4,%%ymm4,%%ymm4           \n"
      "vpsrld     $0x1a,%%ymm4,%%ymm4            \n"
      "vpslld     $0x5,%%ymm4,%%ymm4             \n"
      "vpslld     $0xb,%%ymm3,%%ymm5             \n"

      LABELALIGN
      "1:                                          \n"
      "vmovdqu    (%0),%%ymm0                    \n"
      "vpaddusb   %%ymm6,%%ymm0,%%ymm0           \n"
      "vpsrld     $0x5,%%ymm0,%%ymm2             \n"
      "vpsrld     $0x3,%%ymm0,%%ymm1             \n"
      "vpsrld     $0x8,%%ymm0,%%ymm0             \n"
      "vpand      %%ymm4,%%ymm2,%%ymm2           \n"
      "vpand      %%ymm3,%%ymm1,%%ymm1           \n"
      "vpand      %%ymm5,%%ymm0,%%ymm0           \n"
      "vpor       %%ymm2,%%ymm1,%%ymm1           \n"
      "vpor       %%ymm1,%%ymm0,%%ymm0           \n"
      "vpackusdw  %%ymm0,%%ymm0,%%ymm0           \n"
      "vpermq     $0xd8,%%ymm0,%%ymm0            \n"
      "lea        0x20(%0),%0                    \n"
      "vmovdqu    %%xmm0,(%1)                    \n"
      "lea        0x10(%1),%1                    \n"
      "sub        $0x8,%2                        \n"
      "jg         1b                             \n"
      "vzeroupper                                \n"
      : "+r"(src),    // %0
        "+r"(dst),    // %1
        "+r"(width)   // %2
      : "m"(dither4)  // %3
      : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
        "xmm7");
630 631 632
}
#endif  // HAS_ARGBTORGB565DITHERROW_AVX2

633
void ARGBToARGB1555Row_SSE2(const uint8* src, uint8* dst, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
634
  asm volatile (
fbarchard@google.com's avatar
fbarchard@google.com committed
635 636 637 638 639 640 641 642
    "pcmpeqb   %%xmm4,%%xmm4                   \n"
    "psrld     $0x1b,%%xmm4                    \n"
    "movdqa    %%xmm4,%%xmm5                   \n"
    "pslld     $0x5,%%xmm5                     \n"
    "movdqa    %%xmm4,%%xmm6                   \n"
    "pslld     $0xa,%%xmm6                     \n"
    "pcmpeqb   %%xmm7,%%xmm7                   \n"
    "pslld     $0xf,%%xmm7                     \n"
643
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
644
  "1:                                          \n"
645
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
646 647 648 649 650 651 652 653 654 655 656 657 658 659 660
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm0,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm3                   \n"
    "psrad     $0x10,%%xmm0                    \n"
    "psrld     $0x3,%%xmm1                     \n"
    "psrld     $0x6,%%xmm2                     \n"
    "psrld     $0x9,%%xmm3                     \n"
    "pand      %%xmm7,%%xmm0                   \n"
    "pand      %%xmm4,%%xmm1                   \n"
    "pand      %%xmm5,%%xmm2                   \n"
    "pand      %%xmm6,%%xmm3                   \n"
    "por       %%xmm1,%%xmm0                   \n"
    "por       %%xmm3,%%xmm2                   \n"
    "por       %%xmm2,%%xmm0                   \n"
    "packssdw  %%xmm0,%%xmm0                   \n"
661 662
    "lea       " MEMLEA(0x10,0) ",%0           \n"
    "movq      %%xmm0," MEMACCESS(1) "         \n"
663
    "lea       " MEMLEA(0x8,1) ",%1            \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
664
    "sub       $0x4,%2                         \n"
665
    "jg        1b                              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
666 667
  : "+r"(src),  // %0
    "+r"(dst),  // %1
668
    "+r"(width)   // %2
669 670
  :: "memory", "cc",
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
fbarchard@google.com's avatar
fbarchard@google.com committed
671 672 673
  );
}

674
void ARGBToARGB4444Row_SSE2(const uint8* src, uint8* dst, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
675
  asm volatile (
fbarchard@google.com's avatar
fbarchard@google.com committed
676 677 678 679
    "pcmpeqb   %%xmm4,%%xmm4                   \n"
    "psllw     $0xc,%%xmm4                     \n"
    "movdqa    %%xmm4,%%xmm3                   \n"
    "psrlw     $0x8,%%xmm3                     \n"
680
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
681
  "1:                                          \n"
682
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
683 684 685 686 687 688 689
    "movdqa    %%xmm0,%%xmm1                   \n"
    "pand      %%xmm3,%%xmm0                   \n"
    "pand      %%xmm4,%%xmm1                   \n"
    "psrlq     $0x4,%%xmm0                     \n"
    "psrlq     $0x8,%%xmm1                     \n"
    "por       %%xmm1,%%xmm0                   \n"
    "packuswb  %%xmm0,%%xmm0                   \n"
690 691 692
    "lea       " MEMLEA(0x10,0) ",%0           \n"
    "movq      %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x8,1) ",%1            \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
693
    "sub       $0x4,%2                         \n"
694
    "jg        1b                              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
695 696
  : "+r"(src),  // %0
    "+r"(dst),  // %1
697
    "+r"(width)   // %2
698
  :: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4"
fbarchard@google.com's avatar
fbarchard@google.com committed
699 700
  );
}
701
#endif  // HAS_RGB24TOARGBROW_SSSE3
fbarchard@google.com's avatar
fbarchard@google.com committed
702

703
#ifdef HAS_ARGBTOYROW_SSSE3
704
// Convert 16 ARGB pixels (64 bytes) to 16 Y values.
705
void ARGBToYRow_SSSE3(const uint8* src_argb, uint8* dst_y, int width) {
706
  asm volatile (
707
    "movdqa    %3,%%xmm4                       \n"
708
    "movdqa    %4,%%xmm5                       \n"
709
    LABELALIGN
710
  "1:                                          \n"
711 712 713 714
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm3   \n"
715 716 717 718
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm1                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm4,%%xmm3                   \n"
719
    "lea       " MEMLEA(0x40,0) ",%0           \n"
720 721 722 723 724
    "phaddw    %%xmm1,%%xmm0                   \n"
    "phaddw    %%xmm3,%%xmm2                   \n"
    "psrlw     $0x7,%%xmm0                     \n"
    "psrlw     $0x7,%%xmm2                     \n"
    "packuswb  %%xmm2,%%xmm0                   \n"
725
    "paddb     %%xmm5,%%xmm0                   \n"
726 727
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
728
    "sub       $0x10,%2                        \n"
729 730 731
    "jg        1b                              \n"
  : "+r"(src_argb),  // %0
    "+r"(dst_y),     // %1
732
    "+r"(width)        // %2
733 734
  : "m"(kARGBToY),   // %3
    "m"(kAddY16)     // %4
735
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
736 737
  );
}
738
#endif  // HAS_ARGBTOYROW_SSSE3
739

740
#ifdef HAS_ARGBTOYJROW_SSSE3
741 742
// Convert 16 ARGB pixels (64 bytes) to 16 YJ values.
// Same as ARGBToYRow but different coefficients, no add 16, but do rounding.
743
void ARGBToYJRow_SSSE3(const uint8* src_argb, uint8* dst_y, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
744
  asm volatile (
745
    "movdqa    %3,%%xmm4                       \n"
746
    "movdqa    %4,%%xmm5                       \n"
747
    LABELALIGN
748
  "1:                                          \n"
749 750 751 752
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm3   \n"
753 754 755 756
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm1                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm4,%%xmm3                   \n"
757
    "lea       " MEMLEA(0x40,0) ",%0           \n"
758 759
    "phaddw    %%xmm1,%%xmm0                   \n"
    "phaddw    %%xmm3,%%xmm2                   \n"
760 761
    "paddw     %%xmm5,%%xmm0                   \n"
    "paddw     %%xmm5,%%xmm2                   \n"
762 763 764
    "psrlw     $0x7,%%xmm0                     \n"
    "psrlw     $0x7,%%xmm2                     \n"
    "packuswb  %%xmm2,%%xmm0                   \n"
765
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
766
    "lea       " MEMLEA(0x10,1) ",%1           \n"
767
    "sub       $0x10,%2                        \n"
768
    "jg        1b                              \n"
769 770
  : "+r"(src_argb),  // %0
    "+r"(dst_y),     // %1
771
    "+r"(width)        // %2
772 773
  : "m"(kARGBToYJ),  // %3
    "m"(kAddYJ64)    // %4
774
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
fbarchard@google.com's avatar
fbarchard@google.com committed
775
  );
776
}
777
#endif  // HAS_ARGBTOYJROW_SSSE3
778

779 780
#ifdef HAS_ARGBTOYROW_AVX2
// vpermd for vphaddw + vpackuswb vpermd.
781
static const lvec32 kPermdARGBToY_AVX = {0, 4, 1, 5, 2, 6, 3, 7};
782

783
// Convert 32 ARGB pixels (128 bytes) to 32 Y values.
784
void ARGBToYRow_AVX2(const uint8* src_argb, uint8* dst_y, int width) {
785 786 787 788 789 790 791
  asm volatile (
    "vbroadcastf128 %3,%%ymm4                  \n"
    "vbroadcastf128 %4,%%ymm5                  \n"
    "vmovdqu    %5,%%ymm6                      \n"
    LABELALIGN
  "1:                                          \n"
    "vmovdqu    " MEMACCESS(0) ",%%ymm0        \n"
792 793 794
    "vmovdqu    " MEMACCESS2(0x20,0) ",%%ymm1  \n"
    "vmovdqu    " MEMACCESS2(0x40,0) ",%%ymm2  \n"
    "vmovdqu    " MEMACCESS2(0x60,0) ",%%ymm3  \n"
795 796 797 798 799
    "vpmaddubsw %%ymm4,%%ymm0,%%ymm0           \n"
    "vpmaddubsw %%ymm4,%%ymm1,%%ymm1           \n"
    "vpmaddubsw %%ymm4,%%ymm2,%%ymm2           \n"
    "vpmaddubsw %%ymm4,%%ymm3,%%ymm3           \n"
    "lea       " MEMLEA(0x80,0) ",%0           \n"
800
    "vphaddw    %%ymm1,%%ymm0,%%ymm0           \n"  // mutates.
801 802 803
    "vphaddw    %%ymm3,%%ymm2,%%ymm2           \n"
    "vpsrlw     $0x7,%%ymm0,%%ymm0             \n"
    "vpsrlw     $0x7,%%ymm2,%%ymm2             \n"
804 805 806
    "vpackuswb  %%ymm2,%%ymm0,%%ymm0           \n"  // mutates.
    "vpermd     %%ymm0,%%ymm6,%%ymm0           \n"  // unmutate.
    "vpaddb     %%ymm5,%%ymm0,%%ymm0           \n"  // add 16 for Y
807 808 809 810 811 812 813
    "vmovdqu    %%ymm0," MEMACCESS(1) "        \n"
    "lea       " MEMLEA(0x20,1) ",%1           \n"
    "sub       $0x20,%2                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_argb),  // %0
    "+r"(dst_y),     // %1
814
    "+r"(width)        // %2
815 816 817
  : "m"(kARGBToY),   // %3
    "m"(kAddY16),    // %4
    "m"(kPermdARGBToY_AVX)  // %5
818
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
819 820 821 822 823
  );
}
#endif  // HAS_ARGBTOYROW_AVX2

#ifdef HAS_ARGBTOYJROW_AVX2
824
// Convert 32 ARGB pixels (128 bytes) to 32 Y values.
825
void ARGBToYJRow_AVX2(const uint8* src_argb, uint8* dst_y, int width) {
826 827 828 829 830 831 832
  asm volatile (
    "vbroadcastf128 %3,%%ymm4                  \n"
    "vbroadcastf128 %4,%%ymm5                  \n"
    "vmovdqu    %5,%%ymm6                      \n"
    LABELALIGN
  "1:                                          \n"
    "vmovdqu    " MEMACCESS(0) ",%%ymm0        \n"
833 834 835
    "vmovdqu    " MEMACCESS2(0x20,0) ",%%ymm1  \n"
    "vmovdqu    " MEMACCESS2(0x40,0) ",%%ymm2  \n"
    "vmovdqu    " MEMACCESS2(0x60,0) ",%%ymm3  \n"
836 837 838 839 840
    "vpmaddubsw %%ymm4,%%ymm0,%%ymm0           \n"
    "vpmaddubsw %%ymm4,%%ymm1,%%ymm1           \n"
    "vpmaddubsw %%ymm4,%%ymm2,%%ymm2           \n"
    "vpmaddubsw %%ymm4,%%ymm3,%%ymm3           \n"
    "lea       " MEMLEA(0x80,0) ",%0           \n"
841
    "vphaddw    %%ymm1,%%ymm0,%%ymm0           \n"  // mutates.
842
    "vphaddw    %%ymm3,%%ymm2,%%ymm2           \n"
843 844
    "vpaddw     %%ymm5,%%ymm0,%%ymm0           \n"  // Add .5 for rounding.
    "vpaddw     %%ymm5,%%ymm2,%%ymm2           \n"
845 846
    "vpsrlw     $0x7,%%ymm0,%%ymm0             \n"
    "vpsrlw     $0x7,%%ymm2,%%ymm2             \n"
847 848
    "vpackuswb  %%ymm2,%%ymm0,%%ymm0           \n"  // mutates.
    "vpermd     %%ymm0,%%ymm6,%%ymm0           \n"  // unmutate.
849 850 851 852 853 854 855
    "vmovdqu    %%ymm0," MEMACCESS(1) "        \n"
    "lea       " MEMLEA(0x20,1) ",%1           \n"
    "sub       $0x20,%2                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_argb),  // %0
    "+r"(dst_y),     // %1
856
    "+r"(width)        // %2
857
  : "m"(kARGBToYJ),   // %3
858
    "m"(kAddYJ64),    // %4
859
    "m"(kPermdARGBToY_AVX)  // %5
860
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
861 862 863 864
  );
}
#endif  // HAS_ARGBTOYJROW_AVX2

865
#ifdef HAS_ARGBTOUVROW_SSSE3
866 867 868 869 870
void ARGBToUVRow_SSSE3(const uint8* src_argb0,
                       int src_stride_argb,
                       uint8* dst_u,
                       uint8* dst_v,
                       int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
871
  asm volatile (
872 873 874
    "movdqa    %5,%%xmm3                       \n"
    "movdqa    %6,%%xmm4                       \n"
    "movdqa    %7,%%xmm5                       \n"
875
    "sub       %1,%2                           \n"
876
    LABELALIGN
877
  "1:                                          \n"
878
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
879 880
    MEMOPREG(movdqu,0x00,0,4,1,xmm7)            //  movdqu (%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm0                   \n"
881
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
882 883
    MEMOPREG(movdqu,0x10,0,4,1,xmm7)            //  movdqu 0x10(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm1                   \n"
884
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
885 886
    MEMOPREG(movdqu,0x20,0,4,1,xmm7)            //  movdqu 0x20(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm2                   \n"
887
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm6   \n"
888 889 890
    MEMOPREG(movdqu,0x30,0,4,1,xmm7)            //  movdqu 0x30(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm6                   \n"

891
    "lea       " MEMLEA(0x40,0) ",%0           \n"
892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911
    "movdqa    %%xmm0,%%xmm7                   \n"
    "shufps    $0x88,%%xmm1,%%xmm0             \n"
    "shufps    $0xdd,%%xmm1,%%xmm7             \n"
    "pavgb     %%xmm7,%%xmm0                   \n"
    "movdqa    %%xmm2,%%xmm7                   \n"
    "shufps    $0x88,%%xmm6,%%xmm2             \n"
    "shufps    $0xdd,%%xmm6,%%xmm7             \n"
    "pavgb     %%xmm7,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm2,%%xmm6                   \n"
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm3,%%xmm1                   \n"
    "pmaddubsw %%xmm3,%%xmm6                   \n"
    "phaddw    %%xmm2,%%xmm0                   \n"
    "phaddw    %%xmm6,%%xmm1                   \n"
    "psraw     $0x8,%%xmm0                     \n"
    "psraw     $0x8,%%xmm1                     \n"
    "packsswb  %%xmm1,%%xmm0                   \n"
    "paddb     %%xmm5,%%xmm0                   \n"
912
    "movlps    %%xmm0," MEMACCESS(1) "         \n"
913
    MEMOPMEM(movhps,xmm0,0x00,1,2,1)           //  movhps    %%xmm0,(%1,%2,1)
914
    "lea       " MEMLEA(0x8,1) ",%1            \n"
915
    "sub       $0x10,%3                        \n"
916
    "jg        1b                              \n"
917 918 919 920
  : "+r"(src_argb0),       // %0
    "+r"(dst_u),           // %1
    "+r"(dst_v),           // %2
    "+rm"(width)           // %3
921 922 923 924
  : "r"((intptr_t)(src_stride_argb)), // %4
    "m"(kARGBToV),  // %5
    "m"(kARGBToU),  // %6
    "m"(kAddUV128)  // %7
925 926
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm6", "xmm7"
fbarchard@google.com's avatar
fbarchard@google.com committed
927
  );
928
}
929
#endif  // HAS_ARGBTOUVROW_SSSE3
930

931 932 933
#ifdef HAS_ARGBTOUVROW_AVX2
// vpshufb for vphaddw + vpackuswb packed to shorts.
static const lvec8 kShufARGBToUV_AVX = {
934 935 936 937 938 939 940
    0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15,
    0, 1, 8, 9, 2, 3, 10, 11, 4, 5, 12, 13, 6, 7, 14, 15};
void ARGBToUVRow_AVX2(const uint8* src_argb0,
                      int src_stride_argb,
                      uint8* dst_u,
                      uint8* dst_v,
                      int width) {
941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962
  asm volatile (
    "vbroadcastf128 %5,%%ymm5                  \n"
    "vbroadcastf128 %6,%%ymm6                  \n"
    "vbroadcastf128 %7,%%ymm7                  \n"
    "sub       %1,%2                           \n"
    LABELALIGN
  "1:                                          \n"
    "vmovdqu    " MEMACCESS(0) ",%%ymm0        \n"
    "vmovdqu    " MEMACCESS2(0x20,0) ",%%ymm1  \n"
    "vmovdqu    " MEMACCESS2(0x40,0) ",%%ymm2  \n"
    "vmovdqu    " MEMACCESS2(0x60,0) ",%%ymm3  \n"
    VMEMOPREG(vpavgb,0x00,0,4,1,ymm0,ymm0)     // vpavgb (%0,%4,1),%%ymm0,%%ymm0
    VMEMOPREG(vpavgb,0x20,0,4,1,ymm1,ymm1)
    VMEMOPREG(vpavgb,0x40,0,4,1,ymm2,ymm2)
    VMEMOPREG(vpavgb,0x60,0,4,1,ymm3,ymm3)
    "lea       " MEMLEA(0x80,0) ",%0           \n"
    "vshufps    $0x88,%%ymm1,%%ymm0,%%ymm4     \n"
    "vshufps    $0xdd,%%ymm1,%%ymm0,%%ymm0     \n"
    "vpavgb     %%ymm4,%%ymm0,%%ymm0           \n"
    "vshufps    $0x88,%%ymm3,%%ymm2,%%ymm4     \n"
    "vshufps    $0xdd,%%ymm3,%%ymm2,%%ymm2     \n"
    "vpavgb     %%ymm4,%%ymm2,%%ymm2           \n"
963

964 965 966 967 968 969 970 971 972 973 974 975
    "vpmaddubsw %%ymm7,%%ymm0,%%ymm1           \n"
    "vpmaddubsw %%ymm7,%%ymm2,%%ymm3           \n"
    "vpmaddubsw %%ymm6,%%ymm0,%%ymm0           \n"
    "vpmaddubsw %%ymm6,%%ymm2,%%ymm2           \n"
    "vphaddw    %%ymm3,%%ymm1,%%ymm1           \n"
    "vphaddw    %%ymm2,%%ymm0,%%ymm0           \n"
    "vpsraw     $0x8,%%ymm1,%%ymm1             \n"
    "vpsraw     $0x8,%%ymm0,%%ymm0             \n"
    "vpacksswb  %%ymm0,%%ymm1,%%ymm0           \n"
    "vpermq     $0xd8,%%ymm0,%%ymm0            \n"
    "vpshufb    %8,%%ymm0,%%ymm0               \n"
    "vpaddb     %%ymm5,%%ymm0,%%ymm0           \n"
976

977
    "vextractf128 $0x0,%%ymm0," MEMACCESS(1) " \n"
978
    VEXTOPMEM(vextractf128,1,ymm0,0x0,1,2,1) // vextractf128 $1,%%ymm0,(%1,%2,1)
979 980 981 982 983 984 985 986 987
    "lea       " MEMLEA(0x10,1) ",%1           \n"
    "sub       $0x20,%3                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_argb0),       // %0
    "+r"(dst_u),           // %1
    "+r"(dst_v),           // %2
    "+rm"(width)           // %3
  : "r"((intptr_t)(src_stride_argb)), // %4
988 989 990
    "m"(kAddUV128),  // %5
    "m"(kARGBToV),   // %6
    "m"(kARGBToU),   // %7
991
    "m"(kShufARGBToUV_AVX)  // %8
992 993
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
994 995 996 997
  );
}
#endif  // HAS_ARGBTOUVROW_AVX2

998
#ifdef HAS_ARGBTOUVJROW_AVX2
999 1000 1001 1002 1003
void ARGBToUVJRow_AVX2(const uint8* src_argb0,
                       int src_stride_argb,
                       uint8* dst_u,
                       uint8* dst_v,
                       int width) {
1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061
  asm volatile (
    "vbroadcastf128 %5,%%ymm5                  \n"
    "vbroadcastf128 %6,%%ymm6                  \n"
    "vbroadcastf128 %7,%%ymm7                  \n"
    "sub       %1,%2                           \n"
    LABELALIGN
  "1:                                          \n"
    "vmovdqu    " MEMACCESS(0) ",%%ymm0        \n"
    "vmovdqu    " MEMACCESS2(0x20,0) ",%%ymm1  \n"
    "vmovdqu    " MEMACCESS2(0x40,0) ",%%ymm2  \n"
    "vmovdqu    " MEMACCESS2(0x60,0) ",%%ymm3  \n"
    VMEMOPREG(vpavgb,0x00,0,4,1,ymm0,ymm0)     // vpavgb (%0,%4,1),%%ymm0,%%ymm0
    VMEMOPREG(vpavgb,0x20,0,4,1,ymm1,ymm1)
    VMEMOPREG(vpavgb,0x40,0,4,1,ymm2,ymm2)
    VMEMOPREG(vpavgb,0x60,0,4,1,ymm3,ymm3)
    "lea       " MEMLEA(0x80,0) ",%0           \n"
    "vshufps    $0x88,%%ymm1,%%ymm0,%%ymm4     \n"
    "vshufps    $0xdd,%%ymm1,%%ymm0,%%ymm0     \n"
    "vpavgb     %%ymm4,%%ymm0,%%ymm0           \n"
    "vshufps    $0x88,%%ymm3,%%ymm2,%%ymm4     \n"
    "vshufps    $0xdd,%%ymm3,%%ymm2,%%ymm2     \n"
    "vpavgb     %%ymm4,%%ymm2,%%ymm2           \n"

    "vpmaddubsw %%ymm7,%%ymm0,%%ymm1           \n"
    "vpmaddubsw %%ymm7,%%ymm2,%%ymm3           \n"
    "vpmaddubsw %%ymm6,%%ymm0,%%ymm0           \n"
    "vpmaddubsw %%ymm6,%%ymm2,%%ymm2           \n"
    "vphaddw    %%ymm3,%%ymm1,%%ymm1           \n"
    "vphaddw    %%ymm2,%%ymm0,%%ymm0           \n"
    "vpaddw     %%ymm5,%%ymm0,%%ymm0           \n"
    "vpaddw     %%ymm5,%%ymm1,%%ymm1           \n"
    "vpsraw     $0x8,%%ymm1,%%ymm1             \n"
    "vpsraw     $0x8,%%ymm0,%%ymm0             \n"
    "vpacksswb  %%ymm0,%%ymm1,%%ymm0           \n"
    "vpermq     $0xd8,%%ymm0,%%ymm0            \n"
    "vpshufb    %8,%%ymm0,%%ymm0               \n"

    "vextractf128 $0x0,%%ymm0," MEMACCESS(1) " \n"
    VEXTOPMEM(vextractf128,1,ymm0,0x0,1,2,1) // vextractf128 $1,%%ymm0,(%1,%2,1)
    "lea       " MEMLEA(0x10,1) ",%1           \n"
    "sub       $0x20,%3                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_argb0),       // %0
    "+r"(dst_u),           // %1
    "+r"(dst_v),           // %2
    "+rm"(width)           // %3
  : "r"((intptr_t)(src_stride_argb)), // %4
    "m"(kAddUVJ128),  // %5
    "m"(kARGBToVJ),  // %6
    "m"(kARGBToUJ),  // %7
    "m"(kShufARGBToUV_AVX)  // %8
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
  );
}
#endif  // HAS_ARGBTOUVJROW_AVX2

1062
#ifdef HAS_ARGBTOUVJROW_SSSE3
1063 1064 1065 1066 1067
void ARGBToUVJRow_SSSE3(const uint8* src_argb0,
                        int src_stride_argb,
                        uint8* dst_u,
                        uint8* dst_v,
                        int width) {
1068
  asm volatile (
1069 1070 1071
    "movdqa    %5,%%xmm3                       \n"
    "movdqa    %6,%%xmm4                       \n"
    "movdqa    %7,%%xmm5                       \n"
1072
    "sub       %1,%2                           \n"
1073
    LABELALIGN
1074
  "1:                                          \n"
1075
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
1076 1077
    MEMOPREG(movdqu,0x00,0,4,1,xmm7)            //  movdqu (%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm0                   \n"
1078
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
1079 1080
    MEMOPREG(movdqu,0x10,0,4,1,xmm7)            //  movdqu 0x10(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm1                   \n"
1081
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
1082 1083
    MEMOPREG(movdqu,0x20,0,4,1,xmm7)            //  movdqu 0x20(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm2                   \n"
1084
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm6   \n"
1085 1086 1087
    MEMOPREG(movdqu,0x30,0,4,1,xmm7)            //  movdqu 0x30(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm6                   \n"

1088
    "lea       " MEMLEA(0x40,0) ",%0           \n"
1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109
    "movdqa    %%xmm0,%%xmm7                   \n"
    "shufps    $0x88,%%xmm1,%%xmm0             \n"
    "shufps    $0xdd,%%xmm1,%%xmm7             \n"
    "pavgb     %%xmm7,%%xmm0                   \n"
    "movdqa    %%xmm2,%%xmm7                   \n"
    "shufps    $0x88,%%xmm6,%%xmm2             \n"
    "shufps    $0xdd,%%xmm6,%%xmm7             \n"
    "pavgb     %%xmm7,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm2,%%xmm6                   \n"
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm3,%%xmm1                   \n"
    "pmaddubsw %%xmm3,%%xmm6                   \n"
    "phaddw    %%xmm2,%%xmm0                   \n"
    "phaddw    %%xmm6,%%xmm1                   \n"
    "paddw     %%xmm5,%%xmm0                   \n"
    "paddw     %%xmm5,%%xmm1                   \n"
    "psraw     $0x8,%%xmm0                     \n"
    "psraw     $0x8,%%xmm1                     \n"
    "packsswb  %%xmm1,%%xmm0                   \n"
1110
    "movlps    %%xmm0," MEMACCESS(1) "         \n"
1111
    MEMOPMEM(movhps,xmm0,0x00,1,2,1)           //  movhps  %%xmm0,(%1,%2,1)
1112
    "lea       " MEMLEA(0x8,1) ",%1            \n"
1113
    "sub       $0x10,%3                        \n"
1114 1115 1116 1117 1118
    "jg        1b                              \n"
  : "+r"(src_argb0),       // %0
    "+r"(dst_u),           // %1
    "+r"(dst_v),           // %2
    "+rm"(width)           // %3
1119 1120 1121 1122
  : "r"((intptr_t)(src_stride_argb)), // %4
    "m"(kARGBToVJ),  // %5
    "m"(kARGBToUJ),  // %6
    "m"(kAddUVJ128)  // %7
1123 1124
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm6", "xmm7"
1125 1126
  );
}
1127
#endif  // HAS_ARGBTOUVJROW_SSSE3
1128

1129
#ifdef HAS_ARGBTOUV444ROW_SSSE3
1130 1131 1132
void ARGBToUV444Row_SSSE3(const uint8* src_argb,
                          uint8* dst_u,
                          uint8* dst_v,
1133 1134
                          int width) {
  asm volatile (
1135 1136 1137
    "movdqa    %4,%%xmm3                       \n"
    "movdqa    %5,%%xmm4                       \n"
    "movdqa    %6,%%xmm5                       \n"
1138
    "sub       %1,%2                           \n"
1139
    LABELALIGN
1140
  "1:                                          \n"
1141 1142 1143 1144
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm6   \n"
1145 1146 1147 1148 1149 1150
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm1                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm4,%%xmm6                   \n"
    "phaddw    %%xmm1,%%xmm0                   \n"
    "phaddw    %%xmm6,%%xmm2                   \n"
1151 1152 1153
    "psraw     $0x8,%%xmm0                     \n"
    "psraw     $0x8,%%xmm2                     \n"
    "packsswb  %%xmm2,%%xmm0                   \n"
1154
    "paddb     %%xmm5,%%xmm0                   \n"
1155 1156 1157 1158 1159
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm6   \n"
1160 1161 1162 1163 1164 1165
    "pmaddubsw %%xmm3,%%xmm0                   \n"
    "pmaddubsw %%xmm3,%%xmm1                   \n"
    "pmaddubsw %%xmm3,%%xmm2                   \n"
    "pmaddubsw %%xmm3,%%xmm6                   \n"
    "phaddw    %%xmm1,%%xmm0                   \n"
    "phaddw    %%xmm6,%%xmm2                   \n"
1166 1167 1168
    "psraw     $0x8,%%xmm0                     \n"
    "psraw     $0x8,%%xmm2                     \n"
    "packsswb  %%xmm2,%%xmm0                   \n"
1169
    "paddb     %%xmm5,%%xmm0                   \n"
1170
    "lea       " MEMLEA(0x40,0) ",%0           \n"
1171
    MEMOPMEM(movdqu,xmm0,0x00,1,2,1)           //  movdqu  %%xmm0,(%1,%2,1)
1172
    "lea       " MEMLEA(0x10,1) ",%1           \n"
1173
    "sub       $0x10,%3                        \n"
1174 1175 1176 1177 1178
    "jg        1b                              \n"
  : "+r"(src_argb),        // %0
    "+r"(dst_u),           // %1
    "+r"(dst_v),           // %2
    "+rm"(width)           // %3
1179 1180 1181
  : "m"(kARGBToV),  // %4
    "m"(kARGBToU),  // %5
    "m"(kAddUV128)  // %6
1182 1183
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm6"
1184 1185
  );
}
1186
#endif  // HAS_ARGBTOUV444ROW_SSSE3
1187

1188
void BGRAToYRow_SSSE3(const uint8* src_bgra, uint8* dst_y, int width) {
1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210
  asm volatile (
    "movdqa    %4,%%xmm5                       \n"
    "movdqa    %3,%%xmm4                       \n"
    LABELALIGN
  "1:                                          \n"
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm3   \n"
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm1                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm4,%%xmm3                   \n"
    "lea       " MEMLEA(0x40,0) ",%0           \n"
    "phaddw    %%xmm1,%%xmm0                   \n"
    "phaddw    %%xmm3,%%xmm2                   \n"
    "psrlw     $0x7,%%xmm0                     \n"
    "psrlw     $0x7,%%xmm2                     \n"
    "packuswb  %%xmm2,%%xmm0                   \n"
    "paddb     %%xmm5,%%xmm0                   \n"
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
1211
    "sub       $0x10,%2                        \n"
1212 1213 1214
    "jg        1b                              \n"
  : "+r"(src_bgra),  // %0
    "+r"(dst_y),     // %1
1215
    "+r"(width)        // %2
1216 1217
  : "m"(kBGRAToY),   // %3
    "m"(kAddY16)     // %4
1218
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1219 1220 1221
  );
}

1222 1223 1224 1225 1226
void BGRAToUVRow_SSSE3(const uint8* src_bgra0,
                       int src_stride_bgra,
                       uint8* dst_u,
                       uint8* dst_v,
                       int width) {
1227
  asm volatile (
1228 1229 1230
    "movdqa    %5,%%xmm3                       \n"
    "movdqa    %6,%%xmm4                       \n"
    "movdqa    %7,%%xmm5                       \n"
1231
    "sub       %1,%2                           \n"
1232
    LABELALIGN
1233
  "1:                                          \n"
1234
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
1235 1236
    MEMOPREG(movdqu,0x00,0,4,1,xmm7)            //  movdqu (%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm0                   \n"
1237
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
1238 1239
    MEMOPREG(movdqu,0x10,0,4,1,xmm7)            //  movdqu 0x10(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm1                   \n"
1240
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
1241 1242
    MEMOPREG(movdqu,0x20,0,4,1,xmm7)            //  movdqu 0x20(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm2                   \n"
1243
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm6   \n"
1244 1245 1246
    MEMOPREG(movdqu,0x30,0,4,1,xmm7)            //  movdqu 0x30(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm6                   \n"

1247
    "lea       " MEMLEA(0x40,0) ",%0           \n"
1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267
    "movdqa    %%xmm0,%%xmm7                   \n"
    "shufps    $0x88,%%xmm1,%%xmm0             \n"
    "shufps    $0xdd,%%xmm1,%%xmm7             \n"
    "pavgb     %%xmm7,%%xmm0                   \n"
    "movdqa    %%xmm2,%%xmm7                   \n"
    "shufps    $0x88,%%xmm6,%%xmm2             \n"
    "shufps    $0xdd,%%xmm6,%%xmm7             \n"
    "pavgb     %%xmm7,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm2,%%xmm6                   \n"
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm3,%%xmm1                   \n"
    "pmaddubsw %%xmm3,%%xmm6                   \n"
    "phaddw    %%xmm2,%%xmm0                   \n"
    "phaddw    %%xmm6,%%xmm1                   \n"
    "psraw     $0x8,%%xmm0                     \n"
    "psraw     $0x8,%%xmm1                     \n"
    "packsswb  %%xmm1,%%xmm0                   \n"
    "paddb     %%xmm5,%%xmm0                   \n"
1268
    "movlps    %%xmm0," MEMACCESS(1) "         \n"
1269
    MEMOPMEM(movhps,xmm0,0x00,1,2,1)           //  movhps  %%xmm0,(%1,%2,1)
1270
    "lea       " MEMLEA(0x8,1) ",%1            \n"
1271
    "sub       $0x10,%3                        \n"
1272
    "jg        1b                              \n"
1273
  : "+r"(src_bgra0),       // %0
1274 1275 1276
    "+r"(dst_u),           // %1
    "+r"(dst_v),           // %2
    "+rm"(width)           // %3
1277 1278 1279 1280
  : "r"((intptr_t)(src_stride_bgra)), // %4
    "m"(kBGRAToV),  // %5
    "m"(kBGRAToU),  // %6
    "m"(kAddUV128)  // %7
1281 1282
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm6", "xmm7"
1283 1284 1285
  );
}

1286
void ABGRToYRow_SSSE3(const uint8* src_abgr, uint8* dst_y, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
1287
  asm volatile (
1288 1289
    "movdqa    %4,%%xmm5                       \n"
    "movdqa    %3,%%xmm4                       \n"
1290
    LABELALIGN
1291
  "1:                                          \n"
1292 1293 1294 1295
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm3   \n"
1296 1297 1298 1299
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm1                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm4,%%xmm3                   \n"
1300
    "lea       " MEMLEA(0x40,0) ",%0           \n"
1301 1302 1303 1304 1305 1306
    "phaddw    %%xmm1,%%xmm0                   \n"
    "phaddw    %%xmm3,%%xmm2                   \n"
    "psrlw     $0x7,%%xmm0                     \n"
    "psrlw     $0x7,%%xmm2                     \n"
    "packuswb  %%xmm2,%%xmm0                   \n"
    "paddb     %%xmm5,%%xmm0                   \n"
1307
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
1308
    "lea       " MEMLEA(0x10,1) ",%1           \n"
1309
    "sub       $0x10,%2                        \n"
1310
    "jg        1b                              \n"
1311
  : "+r"(src_abgr),  // %0
1312
    "+r"(dst_y),     // %1
1313
    "+r"(width)        // %2
1314
  : "m"(kABGRToY),   // %3
1315
    "m"(kAddY16)     // %4
1316
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1317 1318 1319
  );
}

1320
void RGBAToYRow_SSSE3(const uint8* src_rgba, uint8* dst_y, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
1321
  asm volatile (
1322 1323
    "movdqa    %4,%%xmm5                       \n"
    "movdqa    %3,%%xmm4                       \n"
1324
    LABELALIGN
1325
  "1:                                          \n"
1326 1327 1328 1329
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm3   \n"
1330 1331 1332 1333
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm1                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm4,%%xmm3                   \n"
1334
    "lea       " MEMLEA(0x40,0) ",%0           \n"
1335 1336 1337 1338 1339 1340
    "phaddw    %%xmm1,%%xmm0                   \n"
    "phaddw    %%xmm3,%%xmm2                   \n"
    "psrlw     $0x7,%%xmm0                     \n"
    "psrlw     $0x7,%%xmm2                     \n"
    "packuswb  %%xmm2,%%xmm0                   \n"
    "paddb     %%xmm5,%%xmm0                   \n"
1341 1342
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
1343
    "sub       $0x10,%2                        \n"
1344
    "jg        1b                              \n"
1345 1346
  : "+r"(src_rgba),  // %0
    "+r"(dst_y),     // %1
1347
    "+r"(width)        // %2
1348 1349
  : "m"(kRGBAToY),   // %3
    "m"(kAddY16)     // %4
1350
  : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1351 1352
  );
}
1353

1354 1355 1356 1357 1358
void ABGRToUVRow_SSSE3(const uint8* src_abgr0,
                       int src_stride_abgr,
                       uint8* dst_u,
                       uint8* dst_v,
                       int width) {
1359
  asm volatile (
1360 1361 1362
    "movdqa    %5,%%xmm3                       \n"
    "movdqa    %6,%%xmm4                       \n"
    "movdqa    %7,%%xmm5                       \n"
1363
    "sub       %1,%2                           \n"
1364
    LABELALIGN
1365
  "1:                                          \n"
1366
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
1367 1368
    MEMOPREG(movdqu,0x00,0,4,1,xmm7)            //  movdqu (%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm0                   \n"
1369
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
1370 1371
    MEMOPREG(movdqu,0x10,0,4,1,xmm7)            //  movdqu 0x10(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm1                   \n"
1372
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
1373 1374
    MEMOPREG(movdqu,0x20,0,4,1,xmm7)            //  movdqu 0x20(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm2                   \n"
1375
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm6   \n"
1376 1377 1378
    MEMOPREG(movdqu,0x30,0,4,1,xmm7)            //  movdqu 0x30(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm6                   \n"

1379
    "lea       " MEMLEA(0x40,0) ",%0           \n"
1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399
    "movdqa    %%xmm0,%%xmm7                   \n"
    "shufps    $0x88,%%xmm1,%%xmm0             \n"
    "shufps    $0xdd,%%xmm1,%%xmm7             \n"
    "pavgb     %%xmm7,%%xmm0                   \n"
    "movdqa    %%xmm2,%%xmm7                   \n"
    "shufps    $0x88,%%xmm6,%%xmm2             \n"
    "shufps    $0xdd,%%xmm6,%%xmm7             \n"
    "pavgb     %%xmm7,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm2,%%xmm6                   \n"
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm3,%%xmm1                   \n"
    "pmaddubsw %%xmm3,%%xmm6                   \n"
    "phaddw    %%xmm2,%%xmm0                   \n"
    "phaddw    %%xmm6,%%xmm1                   \n"
    "psraw     $0x8,%%xmm0                     \n"
    "psraw     $0x8,%%xmm1                     \n"
    "packsswb  %%xmm1,%%xmm0                   \n"
    "paddb     %%xmm5,%%xmm0                   \n"
1400
    "movlps    %%xmm0," MEMACCESS(1) "         \n"
1401
    MEMOPMEM(movhps,xmm0,0x00,1,2,1)           //  movhps  %%xmm0,(%1,%2,1)
1402
    "lea       " MEMLEA(0x8,1) ",%1            \n"
1403
    "sub       $0x10,%3                        \n"
1404
    "jg        1b                              \n"
1405
  : "+r"(src_abgr0),       // %0
1406 1407 1408
    "+r"(dst_u),           // %1
    "+r"(dst_v),           // %2
    "+rm"(width)           // %3
1409 1410 1411 1412
  : "r"((intptr_t)(src_stride_abgr)), // %4
    "m"(kABGRToV),  // %5
    "m"(kABGRToU),  // %6
    "m"(kAddUV128)  // %7
1413 1414
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm6", "xmm7"
1415 1416 1417
  );
}

1418 1419 1420 1421 1422
void RGBAToUVRow_SSSE3(const uint8* src_rgba0,
                       int src_stride_rgba,
                       uint8* dst_u,
                       uint8* dst_v,
                       int width) {
1423
  asm volatile (
1424 1425 1426
    "movdqa    %5,%%xmm3                       \n"
    "movdqa    %6,%%xmm4                       \n"
    "movdqa    %7,%%xmm5                       \n"
1427
    "sub       %1,%2                           \n"
1428
    LABELALIGN
1429
  "1:                                          \n"
1430
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
1431 1432
    MEMOPREG(movdqu,0x00,0,4,1,xmm7)            //  movdqu (%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm0                   \n"
1433
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
1434 1435
    MEMOPREG(movdqu,0x10,0,4,1,xmm7)            //  movdqu 0x10(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm1                   \n"
1436
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
1437 1438
    MEMOPREG(movdqu,0x20,0,4,1,xmm7)            //  movdqu 0x20(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm2                   \n"
1439
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm6   \n"
1440 1441 1442
    MEMOPREG(movdqu,0x30,0,4,1,xmm7)            //  movdqu 0x30(%0,%4,1),%%xmm7
    "pavgb     %%xmm7,%%xmm6                   \n"

1443
    "lea       " MEMLEA(0x40,0) ",%0           \n"
1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463
    "movdqa    %%xmm0,%%xmm7                   \n"
    "shufps    $0x88,%%xmm1,%%xmm0             \n"
    "shufps    $0xdd,%%xmm1,%%xmm7             \n"
    "pavgb     %%xmm7,%%xmm0                   \n"
    "movdqa    %%xmm2,%%xmm7                   \n"
    "shufps    $0x88,%%xmm6,%%xmm2             \n"
    "shufps    $0xdd,%%xmm6,%%xmm7             \n"
    "pavgb     %%xmm7,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm2,%%xmm6                   \n"
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm2                   \n"
    "pmaddubsw %%xmm3,%%xmm1                   \n"
    "pmaddubsw %%xmm3,%%xmm6                   \n"
    "phaddw    %%xmm2,%%xmm0                   \n"
    "phaddw    %%xmm6,%%xmm1                   \n"
    "psraw     $0x8,%%xmm0                     \n"
    "psraw     $0x8,%%xmm1                     \n"
    "packsswb  %%xmm1,%%xmm0                   \n"
    "paddb     %%xmm5,%%xmm0                   \n"
1464
    "movlps    %%xmm0," MEMACCESS(1) "         \n"
1465
    MEMOPMEM(movhps,xmm0,0x00,1,2,1)           //  movhps  %%xmm0,(%1,%2,1)
1466
    "lea       " MEMLEA(0x8,1) ",%1            \n"
1467
    "sub       $0x10,%3                        \n"
1468 1469 1470 1471 1472
    "jg        1b                              \n"
  : "+r"(src_rgba0),       // %0
    "+r"(dst_u),           // %1
    "+r"(dst_v),           // %2
    "+rm"(width)           // %3
1473 1474 1475 1476
  : "r"((intptr_t)(src_stride_rgba)), // %4
    "m"(kRGBAToV),  // %5
    "m"(kRGBAToU),  // %6
    "m"(kAddUV128)  // %7
1477 1478
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm6", "xmm7"
1479 1480
  );
}
1481

1482
#if defined(HAS_I422TOARGBROW_SSSE3) || defined(HAS_I422TOARGBROW_AVX2)
1483

1484
// Read 8 UV from 444
1485 1486
#define READYUV444 \
  "movq       " MEMACCESS([u_buf]) ",%%xmm0                   \n"            \
1487 1488
    MEMOPREG(movq, 0x00, [u_buf], [v_buf], 1, xmm1)                            \
    "lea        " MEMLEA(0x8, [u_buf]) ",%[u_buf]               \n"            \
1489 1490
    "punpcklbw  %%xmm1,%%xmm0                                   \n"            \
    "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
1491
    "punpcklbw  %%xmm4,%%xmm4                                   \n"            \
1492
    "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"
1493

1494
// Read 4 UV from 422, upsample to 8 UV
1495 1496
#define READYUV422 \
  "movd       " MEMACCESS([u_buf]) ",%%xmm0                   \n"            \
1497
    MEMOPREG(movd, 0x00, [u_buf], [v_buf], 1, xmm1)                            \
1498
    "lea        " MEMLEA(0x4, [u_buf]) ",%[u_buf]               \n"            \
1499
    "punpcklbw  %%xmm1,%%xmm0                                   \n"            \
1500 1501
    "punpcklwd  %%xmm0,%%xmm0                                   \n"            \
    "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
1502
    "punpcklbw  %%xmm4,%%xmm4                                   \n"            \
1503
    "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"
1504

1505
// Read 4 UV from 422, upsample to 8 UV.  With 8 Alpha.
1506 1507
#define READYUVA422 \
  "movd       " MEMACCESS([u_buf]) ",%%xmm0                   \n"            \
1508 1509 1510 1511 1512 1513 1514 1515 1516 1517
    MEMOPREG(movd, 0x00, [u_buf], [v_buf], 1, xmm1)                            \
    "lea        " MEMLEA(0x4, [u_buf]) ",%[u_buf]               \n"            \
    "punpcklbw  %%xmm1,%%xmm0                                   \n"            \
    "punpcklwd  %%xmm0,%%xmm0                                   \n"            \
    "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
    "punpcklbw  %%xmm4,%%xmm4                                   \n"            \
    "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"            \
    "movq       " MEMACCESS([a_buf]) ",%%xmm5                   \n"            \
    "lea        " MEMLEA(0x8, [a_buf]) ",%[a_buf]               \n"

1518
// Read 4 UV from NV12, upsample to 8 UV
1519 1520
#define READNV12 \
  "movq       " MEMACCESS([uv_buf]) ",%%xmm0                  \n"            \
1521
    "lea        " MEMLEA(0x8, [uv_buf]) ",%[uv_buf]             \n"            \
1522 1523
    "punpcklwd  %%xmm0,%%xmm0                                   \n"            \
    "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
1524 1525 1526
    "punpcklbw  %%xmm4,%%xmm4                                   \n"            \
    "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"

1527
// Read 4 VU from NV21, upsample to 8 UV
1528 1529
#define READNV21 \
  "movq       " MEMACCESS([vu_buf]) ",%%xmm0                  \n"            \
1530 1531 1532 1533 1534 1535
    "lea        " MEMLEA(0x8, [vu_buf]) ",%[vu_buf]             \n"            \
    "pshufb     %[kShuffleNV21], %%xmm0                         \n"            \
    "movq       " MEMACCESS([y_buf]) ",%%xmm4                   \n"            \
    "punpcklbw  %%xmm4,%%xmm4                                   \n"            \
    "lea        " MEMLEA(0x8, [y_buf]) ",%[y_buf]               \n"

1536
// Read 4 YUY2 with 8 Y and update 4 UV to 8 UV.
1537 1538
#define READYUY2 \
  "movdqu     " MEMACCESS([yuy2_buf]) ",%%xmm4                \n"            \
1539 1540 1541 1542 1543 1544
    "pshufb     %[kShuffleYUY2Y], %%xmm4                        \n"            \
    "movdqu     " MEMACCESS([yuy2_buf]) ",%%xmm0                \n"            \
    "pshufb     %[kShuffleYUY2UV], %%xmm0                       \n"            \
    "lea        " MEMLEA(0x10, [yuy2_buf]) ",%[yuy2_buf]        \n"

// Read 4 UYVY with 8 Y and update 4 UV to 8 UV.
1545 1546
#define READUYVY \
  "movdqu     " MEMACCESS([uyvy_buf]) ",%%xmm4                \n"            \
1547 1548 1549 1550
    "pshufb     %[kShuffleUYVYY], %%xmm4                        \n"            \
    "movdqu     " MEMACCESS([uyvy_buf]) ",%%xmm0                \n"            \
    "pshufb     %[kShuffleUYVYUV], %%xmm0                       \n"            \
    "lea        " MEMLEA(0x10, [uyvy_buf]) ",%[uyvy_buf]        \n"
1551

1552
#if defined(__x86_64__)
1553 1554
#define YUVTORGB_SETUP(yuvconstants) \
  "movdqa     " MEMACCESS([yuvconstants]) ",%%xmm8            \n"            \
1555 1556 1557 1558 1559 1560 1561
    "movdqa     " MEMACCESS2(32, [yuvconstants]) ",%%xmm9       \n"            \
    "movdqa     " MEMACCESS2(64, [yuvconstants]) ",%%xmm10      \n"            \
    "movdqa     " MEMACCESS2(96, [yuvconstants]) ",%%xmm11      \n"            \
    "movdqa     " MEMACCESS2(128, [yuvconstants]) ",%%xmm12     \n"            \
    "movdqa     " MEMACCESS2(160, [yuvconstants]) ",%%xmm13     \n"            \
    "movdqa     " MEMACCESS2(192, [yuvconstants]) ",%%xmm14     \n"
// Convert 8 pixels: 8 UV and 8 Y
1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584
#define YUVTORGB(yuvconstants)                                    \
  "movdqa     %%xmm0,%%xmm1                                   \n" \
  "movdqa     %%xmm0,%%xmm2                                   \n" \
  "movdqa     %%xmm0,%%xmm3                                   \n" \
  "movdqa     %%xmm11,%%xmm0                                  \n" \
  "pmaddubsw  %%xmm8,%%xmm1                                   \n" \
  "psubw      %%xmm1,%%xmm0                                   \n" \
  "movdqa     %%xmm12,%%xmm1                                  \n" \
  "pmaddubsw  %%xmm9,%%xmm2                                   \n" \
  "psubw      %%xmm2,%%xmm1                                   \n" \
  "movdqa     %%xmm13,%%xmm2                                  \n" \
  "pmaddubsw  %%xmm10,%%xmm3                                  \n" \
  "psubw      %%xmm3,%%xmm2                                   \n" \
  "pmulhuw    %%xmm14,%%xmm4                                  \n" \
  "paddsw     %%xmm4,%%xmm0                                   \n" \
  "paddsw     %%xmm4,%%xmm1                                   \n" \
  "paddsw     %%xmm4,%%xmm2                                   \n" \
  "psraw      $0x6,%%xmm0                                     \n" \
  "psraw      $0x6,%%xmm1                                     \n" \
  "psraw      $0x6,%%xmm2                                     \n" \
  "packuswb   %%xmm0,%%xmm0                                   \n" \
  "packuswb   %%xmm1,%%xmm1                                   \n" \
  "packuswb   %%xmm2,%%xmm2                                   \n"
1585
#define YUVTORGB_REGS \
1586
  "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14",
1587 1588 1589

#else
#define YUVTORGB_SETUP(yuvconstants)
1590
// Convert 8 pixels: 8 UV and 8 Y
1591 1592
#define YUVTORGB(yuvconstants) \
  "movdqa     %%xmm0,%%xmm1                                   \n"            \
1593
    "movdqa     %%xmm0,%%xmm2                                   \n"            \
1594
    "movdqa     %%xmm0,%%xmm3                                   \n"            \
1595 1596
    "movdqa     " MEMACCESS2(96, [yuvconstants]) ",%%xmm0       \n"            \
    "pmaddubsw  " MEMACCESS([yuvconstants]) ",%%xmm1            \n"            \
1597
    "psubw      %%xmm1,%%xmm0                                   \n"            \
1598 1599
    "movdqa     " MEMACCESS2(128, [yuvconstants]) ",%%xmm1      \n"            \
    "pmaddubsw  " MEMACCESS2(32, [yuvconstants]) ",%%xmm2       \n"            \
1600
    "psubw      %%xmm2,%%xmm1                                   \n"            \
1601 1602
    "movdqa     " MEMACCESS2(160, [yuvconstants]) ",%%xmm2      \n"            \
    "pmaddubsw  " MEMACCESS2(64, [yuvconstants]) ",%%xmm3       \n"            \
1603
    "psubw      %%xmm3,%%xmm2                                   \n"            \
1604 1605 1606 1607
    "pmulhuw    " MEMACCESS2(192, [yuvconstants]) ",%%xmm4      \n"            \
    "paddsw     %%xmm4,%%xmm0                                   \n"            \
    "paddsw     %%xmm4,%%xmm1                                   \n"            \
    "paddsw     %%xmm4,%%xmm2                                   \n"            \
1608 1609 1610 1611 1612
    "psraw      $0x6,%%xmm0                                     \n"            \
    "psraw      $0x6,%%xmm1                                     \n"            \
    "psraw      $0x6,%%xmm2                                     \n"            \
    "packuswb   %%xmm0,%%xmm0                                   \n"            \
    "packuswb   %%xmm1,%%xmm1                                   \n"            \
1613
    "packuswb   %%xmm2,%%xmm2                                   \n"
1614 1615
#define YUVTORGB_REGS
#endif
1616

1617
// Store 8 ARGB values.
1618 1619
#define STOREARGB \
  "punpcklbw  %%xmm1,%%xmm0                                    \n"           \
1620 1621 1622 1623 1624
    "punpcklbw  %%xmm5,%%xmm2                                    \n"           \
    "movdqa     %%xmm0,%%xmm1                                    \n"           \
    "punpcklwd  %%xmm2,%%xmm0                                    \n"           \
    "punpckhwd  %%xmm2,%%xmm1                                    \n"           \
    "movdqu     %%xmm0," MEMACCESS([dst_argb]) "                 \n"           \
1625 1626
    "movdqu     %%xmm1," MEMACCESS2(0x10, [dst_argb]) "          \n"           \
    "lea        " MEMLEA(0x20, [dst_argb]) ", %[dst_argb]        \n"
1627

1628
// Store 8 RGBA values.
1629 1630
#define STORERGBA \
  "pcmpeqb   %%xmm5,%%xmm5                                     \n"           \
1631 1632 1633 1634 1635 1636
    "punpcklbw %%xmm2,%%xmm1                                     \n"           \
    "punpcklbw %%xmm0,%%xmm5                                     \n"           \
    "movdqa    %%xmm5,%%xmm0                                     \n"           \
    "punpcklwd %%xmm1,%%xmm5                                     \n"           \
    "punpckhwd %%xmm1,%%xmm0                                     \n"           \
    "movdqu    %%xmm5," MEMACCESS([dst_rgba]) "                  \n"           \
1637 1638
    "movdqu    %%xmm0," MEMACCESS2(0x10, [dst_rgba]) "           \n"           \
    "lea       " MEMLEA(0x20, [dst_rgba]) ",%[dst_rgba]          \n"
1639

1640 1641 1642 1643
void OMITFP I444ToARGBRow_SSSE3(const uint8* y_buf,
                                const uint8* u_buf,
                                const uint8* v_buf,
                                uint8* dst_argb,
1644
                                const struct YuvConstants* yuvconstants,
1645
                                int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
1646
  asm volatile (
1647
    YUVTORGB_SETUP(yuvconstants)
1648
    "sub       %[u_buf],%[v_buf]               \n"
1649
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
1650
    LABELALIGN
1651
  "1:                                          \n"
1652
    READYUV444
1653
    YUVTORGB(yuvconstants)
1654
    STOREARGB
1655
    "sub       $0x8,%[width]                   \n"
1656
    "jg        1b                              \n"
1657 1658 1659
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [u_buf]"+r"(u_buf),    // %[u_buf]
    [v_buf]"+r"(v_buf),    // %[v_buf]
1660
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
1661
    [width]"+rm"(width)    // %[width]
1662
  : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
1663
  : "memory", "cc", NACL_R14 YUVTORGB_REGS
1664
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1665 1666 1667
  );
}

1668 1669
void OMITFP I422ToRGB24Row_SSSE3(const uint8* y_buf,
                                 const uint8* u_buf,
1670 1671
                                 const uint8* v_buf,
                                 uint8* dst_rgb24,
1672
                                 const struct YuvConstants* yuvconstants,
1673 1674
                                 int width) {
  asm volatile (
1675
    YUVTORGB_SETUP(yuvconstants)
1676 1677 1678
    "movdqa    %[kShuffleMaskARGBToRGB24_0],%%xmm5 \n"
    "movdqa    %[kShuffleMaskARGBToRGB24],%%xmm6   \n"
    "sub       %[u_buf],%[v_buf]               \n"
1679
    LABELALIGN
1680
  "1:                                          \n"
1681
    READYUV422
1682
    YUVTORGB(yuvconstants)
1683
    "punpcklbw %%xmm1,%%xmm0                   \n"
1684
    "punpcklbw %%xmm2,%%xmm2                   \n"
1685 1686 1687
    "movdqa    %%xmm0,%%xmm1                   \n"
    "punpcklwd %%xmm2,%%xmm0                   \n"
    "punpckhwd %%xmm2,%%xmm1                   \n"
1688 1689 1690 1691 1692 1693
    "pshufb    %%xmm5,%%xmm0                   \n"
    "pshufb    %%xmm6,%%xmm1                   \n"
    "palignr   $0xc,%%xmm0,%%xmm1              \n"
    "movq      %%xmm0," MEMACCESS([dst_rgb24]) "\n"
    "movdqu    %%xmm1," MEMACCESS2(0x8,[dst_rgb24]) "\n"
    "lea       " MEMLEA(0x18,[dst_rgb24]) ",%[dst_rgb24] \n"
1694
    "subl      $0x8,%[width]                   \n"
1695
    "jg        1b                              \n"
1696
  : [y_buf]"+r"(y_buf),    // %[y_buf]
1697 1698 1699
    [u_buf]"+r"(u_buf),    // %[u_buf]
    [v_buf]"+r"(v_buf),    // %[v_buf]
    [dst_rgb24]"+r"(dst_rgb24),  // %[dst_rgb24]
1700
#if defined(__i386__)
1701
    [width]"+m"(width)     // %[width]
1702
#else
1703
    [width]"+rm"(width)    // %[width]
1704
#endif
1705
  : [yuvconstants]"r"(yuvconstants),  // %[yuvconstants]
1706 1707
    [kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0),
    [kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24)
1708
  : "memory", "cc", NACL_R14 YUVTORGB_REGS
1709
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
1710 1711 1712
  );
}

1713 1714 1715 1716
void OMITFP I422ToARGBRow_SSSE3(const uint8* y_buf,
                                const uint8* u_buf,
                                const uint8* v_buf,
                                uint8* dst_argb,
1717
                                const struct YuvConstants* yuvconstants,
1718
                                int width) {
1719
  asm volatile (
1720
    YUVTORGB_SETUP(yuvconstants)
1721
    "sub       %[u_buf],%[v_buf]               \n"
1722
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
1723
    LABELALIGN
1724
  "1:                                          \n"
1725
    READYUV422
1726
    YUVTORGB(yuvconstants)
1727
    STOREARGB
1728
    "sub       $0x8,%[width]                   \n"
1729
    "jg        1b                              \n"
1730 1731 1732
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [u_buf]"+r"(u_buf),    // %[u_buf]
    [v_buf]"+r"(v_buf),    // %[v_buf]
1733
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
1734
    [width]"+rm"(width)    // %[width]
1735
  : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
1736
  : "memory", "cc", NACL_R14 YUVTORGB_REGS
1737
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1738 1739 1740
  );
}

1741
#ifdef HAS_I422ALPHATOARGBROW_SSSE3
1742 1743 1744 1745 1746
void OMITFP I422AlphaToARGBRow_SSSE3(const uint8* y_buf,
                                     const uint8* u_buf,
                                     const uint8* v_buf,
                                     const uint8* a_buf,
                                     uint8* dst_argb,
1747
                                     const struct YuvConstants* yuvconstants,
1748
                                     int width) {
1749
  // clang-format off
1750
  asm volatile (
1751
    YUVTORGB_SETUP(yuvconstants)
1752 1753 1754 1755 1756 1757
    "sub       %[u_buf],%[v_buf]               \n"
    LABELALIGN
  "1:                                          \n"
    READYUVA422
    YUVTORGB(yuvconstants)
    STOREARGB
1758
    "subl      $0x8,%[width]                   \n"
1759 1760 1761 1762 1763 1764
    "jg        1b                              \n"
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [u_buf]"+r"(u_buf),    // %[u_buf]
    [v_buf]"+r"(v_buf),    // %[v_buf]
    [a_buf]"+r"(a_buf),    // %[a_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
1765
#if defined(__i386__)
1766 1767
    [width]"+m"(width)     // %[width]
#else
1768
    [width]"+rm"(width)    // %[width]
1769
#endif
1770
  : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
1771
  : "memory", "cc", NACL_R14 YUVTORGB_REGS
1772 1773
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
  );
1774
  // clang-format on
1775
}
1776
#endif  // HAS_I422ALPHATOARGBROW_SSSE3
1777

1778 1779 1780
void OMITFP NV12ToARGBRow_SSSE3(const uint8* y_buf,
                                const uint8* uv_buf,
                                uint8* dst_argb,
1781
                                const struct YuvConstants* yuvconstants,
1782
                                int width) {
1783
  // clang-format off
1784
  asm volatile (
1785
    YUVTORGB_SETUP(yuvconstants)
1786
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
1787
    LABELALIGN
1788 1789
  "1:                                          \n"
    READNV12
1790
    YUVTORGB(yuvconstants)
1791
    STOREARGB
1792 1793 1794
    "sub       $0x8,%[width]                   \n"
    "jg        1b                              \n"
  : [y_buf]"+r"(y_buf),    // %[y_buf]
1795 1796
    [uv_buf]"+r"(uv_buf),    // %[uv_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
1797
    [width]"+rm"(width)    // %[width]
1798
  : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
1799 1800
    : "memory", "cc", YUVTORGB_REGS  // Does not use r14.
      "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1801
  );
1802
  // clang-format on
1803 1804
}

1805 1806 1807
void OMITFP NV21ToARGBRow_SSSE3(const uint8* y_buf,
                                const uint8* vu_buf,
                                uint8* dst_argb,
1808
                                const struct YuvConstants* yuvconstants,
1809
                                int width) {
1810
  // clang-format off
1811
  asm volatile (
1812
    YUVTORGB_SETUP(yuvconstants)
1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    LABELALIGN
  "1:                                          \n"
    READNV21
    YUVTORGB(yuvconstants)
    STOREARGB
    "sub       $0x8,%[width]                   \n"
    "jg        1b                              \n"
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [vu_buf]"+r"(vu_buf),    // %[vu_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
    [width]"+rm"(width)    // %[width]
  : [yuvconstants]"r"(yuvconstants), // %[yuvconstants]
    [kShuffleNV21]"m"(kShuffleNV21)
1827 1828
    : "memory", "cc", YUVTORGB_REGS  // Does not use r14.
      "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1829
  );
1830
  // clang-format on
1831 1832
}

1833 1834
void OMITFP YUY2ToARGBRow_SSSE3(const uint8* yuy2_buf,
                                uint8* dst_argb,
1835
                                const struct YuvConstants* yuvconstants,
1836
                                int width) {
1837
  // clang-format off
1838
  asm volatile (
1839
    YUVTORGB_SETUP(yuvconstants)
1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    LABELALIGN
  "1:                                          \n"
    READYUY2
    YUVTORGB(yuvconstants)
    STOREARGB
    "sub       $0x8,%[width]                   \n"
    "jg        1b                              \n"
  : [yuy2_buf]"+r"(yuy2_buf),    // %[yuy2_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
    [width]"+rm"(width)    // %[width]
  : [yuvconstants]"r"(yuvconstants), // %[yuvconstants]
    [kShuffleYUY2Y]"m"(kShuffleYUY2Y),
    [kShuffleYUY2UV]"m"(kShuffleYUY2UV)
1854 1855
    : "memory", "cc", YUVTORGB_REGS  // Does not use r14.
      "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1856
  );
1857
  // clang-format on
1858 1859 1860 1861
}

void OMITFP UYVYToARGBRow_SSSE3(const uint8* uyvy_buf,
                                uint8* dst_argb,
1862
                                const struct YuvConstants* yuvconstants,
1863
                                int width) {
1864
  // clang-format off
1865
  asm volatile (
1866
    YUVTORGB_SETUP(yuvconstants)
1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    LABELALIGN
  "1:                                          \n"
    READUYVY
    YUVTORGB(yuvconstants)
    STOREARGB
    "sub       $0x8,%[width]                   \n"
    "jg        1b                              \n"
  : [uyvy_buf]"+r"(uyvy_buf),    // %[uyvy_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
    [width]"+rm"(width)    // %[width]
  : [yuvconstants]"r"(yuvconstants), // %[yuvconstants]
    [kShuffleUYVYY]"m"(kShuffleUYVYY),
    [kShuffleUYVYUV]"m"(kShuffleUYVYUV)
1881 1882
    : "memory", "cc", YUVTORGB_REGS  // Does not use r14.
      "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1883
  );
1884
  // clang-format on
1885 1886
}

1887 1888 1889
void OMITFP I422ToRGBARow_SSSE3(const uint8* y_buf,
                                const uint8* u_buf,
                                const uint8* v_buf,
1890
                                uint8* dst_rgba,
1891
                                const struct YuvConstants* yuvconstants,
1892 1893
                                int width) {
  asm volatile (
1894
    YUVTORGB_SETUP(yuvconstants)
1895 1896
    "sub       %[u_buf],%[v_buf]               \n"
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
1897
    LABELALIGN
1898 1899
  "1:                                          \n"
    READYUV422
1900
    YUVTORGB(yuvconstants)
1901
    STORERGBA
1902 1903 1904 1905 1906
    "sub       $0x8,%[width]                   \n"
    "jg        1b                              \n"
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [u_buf]"+r"(u_buf),    // %[u_buf]
    [v_buf]"+r"(v_buf),    // %[v_buf]
1907
    [dst_rgba]"+r"(dst_rgba),  // %[dst_rgba]
1908
    [width]"+rm"(width)    // %[width]
1909
  : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
1910
  : "memory", "cc", NACL_R14 YUVTORGB_REGS
1911
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1912 1913 1914
  );
}

1915
#endif  // HAS_I422TOARGBROW_SSSE3
1916

1917
// Read 16 UV from 444
1918 1919
#define READYUV444_AVX2 \
  "vmovdqu    " MEMACCESS([u_buf]) ",%%xmm0                       \n"        \
1920 1921 1922 1923 1924 1925 1926 1927 1928 1929
    MEMOPREG(vmovdqu, 0x00, [u_buf], [v_buf], 1, xmm1)                         \
    "lea        " MEMLEA(0x10, [u_buf]) ",%[u_buf]                  \n"        \
    "vpermq     $0xd8,%%ymm0,%%ymm0                                 \n"        \
    "vpermq     $0xd8,%%ymm1,%%ymm1                                 \n"        \
    "vpunpcklbw %%ymm1,%%ymm0,%%ymm0                                \n"        \
    "vmovdqu    " MEMACCESS([y_buf]) ",%%xmm4                       \n"        \
    "vpermq     $0xd8,%%ymm4,%%ymm4                                 \n"        \
    "vpunpcklbw %%ymm4,%%ymm4,%%ymm4                                \n"        \
    "lea        " MEMLEA(0x10, [y_buf]) ",%[y_buf]                  \n"

1930
// Read 8 UV from 422, upsample to 16 UV.
1931 1932
#define READYUV422_AVX2 \
  "vmovq      " MEMACCESS([u_buf]) ",%%xmm0                       \n"        \
1933 1934 1935 1936
    MEMOPREG(vmovq, 0x00, [u_buf], [v_buf], 1, xmm1)                           \
    "lea        " MEMLEA(0x8, [u_buf]) ",%[u_buf]                   \n"        \
    "vpunpcklbw %%ymm1,%%ymm0,%%ymm0                                \n"        \
    "vpermq     $0xd8,%%ymm0,%%ymm0                                 \n"        \
1937
    "vpunpcklwd %%ymm0,%%ymm0,%%ymm0                                \n"        \
1938 1939 1940 1941 1942
    "vmovdqu    " MEMACCESS([y_buf]) ",%%xmm4                       \n"        \
    "vpermq     $0xd8,%%ymm4,%%ymm4                                 \n"        \
    "vpunpcklbw %%ymm4,%%ymm4,%%ymm4                                \n"        \
    "lea        " MEMLEA(0x10, [y_buf]) ",%[y_buf]                  \n"

1943
// Read 8 UV from 422, upsample to 16 UV.  With 16 Alpha.
1944 1945
#define READYUVA422_AVX2 \
  "vmovq      " MEMACCESS([u_buf]) ",%%xmm0                       \n"        \
1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958
    MEMOPREG(vmovq, 0x00, [u_buf], [v_buf], 1, xmm1)                           \
    "lea        " MEMLEA(0x8, [u_buf]) ",%[u_buf]                   \n"        \
    "vpunpcklbw %%ymm1,%%ymm0,%%ymm0                                \n"        \
    "vpermq     $0xd8,%%ymm0,%%ymm0                                 \n"        \
    "vpunpcklwd %%ymm0,%%ymm0,%%ymm0                                \n"        \
    "vmovdqu    " MEMACCESS([y_buf]) ",%%xmm4                       \n"        \
    "vpermq     $0xd8,%%ymm4,%%ymm4                                 \n"        \
    "vpunpcklbw %%ymm4,%%ymm4,%%ymm4                                \n"        \
    "lea        " MEMLEA(0x10, [y_buf]) ",%[y_buf]                  \n"        \
    "vmovdqu    " MEMACCESS([a_buf]) ",%%xmm5                       \n"        \
    "vpermq     $0xd8,%%ymm5,%%ymm5                                 \n"        \
    "lea        " MEMLEA(0x10, [a_buf]) ",%[a_buf]                  \n"

1959
// Read 8 UV from NV12, upsample to 16 UV.
1960 1961
#define READNV12_AVX2 \
  "vmovdqu    " MEMACCESS([uv_buf]) ",%%xmm0                      \n"        \
1962
    "lea        " MEMLEA(0x10, [uv_buf]) ",%[uv_buf]                \n"        \
1963 1964 1965 1966 1967 1968
    "vpermq     $0xd8,%%ymm0,%%ymm0                                 \n"        \
    "vpunpcklwd %%ymm0,%%ymm0,%%ymm0                                \n"        \
    "vmovdqu    " MEMACCESS([y_buf]) ",%%xmm4                       \n"        \
    "vpermq     $0xd8,%%ymm4,%%ymm4                                 \n"        \
    "vpunpcklbw %%ymm4,%%ymm4,%%ymm4                                \n"        \
    "lea        " MEMLEA(0x10, [y_buf]) ",%[y_buf]                  \n"
1969

1970
// Read 8 VU from NV21, upsample to 16 UV.
1971 1972
#define READNV21_AVX2 \
  "vmovdqu    " MEMACCESS([vu_buf]) ",%%xmm0                      \n"        \
1973 1974 1975 1976 1977 1978 1979 1980
    "lea        " MEMLEA(0x10, [vu_buf]) ",%[vu_buf]                \n"        \
    "vpermq     $0xd8,%%ymm0,%%ymm0                                 \n"        \
    "vpshufb     %[kShuffleNV21], %%ymm0, %%ymm0                    \n"        \
    "vmovdqu    " MEMACCESS([y_buf]) ",%%xmm4                       \n"        \
    "vpermq     $0xd8,%%ymm4,%%ymm4                                 \n"        \
    "vpunpcklbw %%ymm4,%%ymm4,%%ymm4                                \n"        \
    "lea        " MEMLEA(0x10, [y_buf]) ",%[y_buf]                  \n"

1981
// Read 8 YUY2 with 16 Y and upsample 8 UV to 16 UV.
1982 1983
#define READYUY2_AVX2 \
  "vmovdqu    " MEMACCESS([yuy2_buf]) ",%%ymm4                    \n"        \
1984 1985 1986 1987 1988
    "vpshufb    %[kShuffleYUY2Y], %%ymm4, %%ymm4                    \n"        \
    "vmovdqu    " MEMACCESS([yuy2_buf]) ",%%ymm0                    \n"        \
    "vpshufb    %[kShuffleYUY2UV], %%ymm0, %%ymm0                   \n"        \
    "lea        " MEMLEA(0x20, [yuy2_buf]) ",%[yuy2_buf]            \n"

1989
// Read 8 UYVY with 16 Y and upsample 8 UV to 16 UV.
1990 1991
#define READUYVY_AVX2 \
  "vmovdqu     " MEMACCESS([uyvy_buf]) ",%%ymm4                   \n"        \
1992 1993 1994 1995
    "vpshufb     %[kShuffleUYVYY], %%ymm4, %%ymm4                   \n"        \
    "vmovdqu     " MEMACCESS([uyvy_buf]) ",%%ymm0                   \n"        \
    "vpshufb     %[kShuffleUYVYUV], %%ymm0, %%ymm0                  \n"        \
    "lea        " MEMLEA(0x20, [uyvy_buf]) ",%[uyvy_buf]            \n"
1996

1997
#if defined(__x86_64__)
1998 1999
#define YUVTORGB_SETUP_AVX2(yuvconstants) \
  "vmovdqa     " MEMACCESS([yuvconstants]) ",%%ymm8            \n"           \
2000 2001 2002 2003 2004 2005
    "vmovdqa     " MEMACCESS2(32, [yuvconstants]) ",%%ymm9       \n"           \
    "vmovdqa     " MEMACCESS2(64, [yuvconstants]) ",%%ymm10      \n"           \
    "vmovdqa     " MEMACCESS2(96, [yuvconstants]) ",%%ymm11      \n"           \
    "vmovdqa     " MEMACCESS2(128, [yuvconstants]) ",%%ymm12     \n"           \
    "vmovdqa     " MEMACCESS2(160, [yuvconstants]) ",%%ymm13     \n"           \
    "vmovdqa     " MEMACCESS2(192, [yuvconstants]) ",%%ymm14     \n"
2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022
#define YUVTORGB_AVX2(yuvconstants)                                   \
  "vpmaddubsw  %%ymm10,%%ymm0,%%ymm2                              \n" \
  "vpmaddubsw  %%ymm9,%%ymm0,%%ymm1                               \n" \
  "vpmaddubsw  %%ymm8,%%ymm0,%%ymm0                               \n" \
  "vpsubw      %%ymm2,%%ymm13,%%ymm2                              \n" \
  "vpsubw      %%ymm1,%%ymm12,%%ymm1                              \n" \
  "vpsubw      %%ymm0,%%ymm11,%%ymm0                              \n" \
  "vpmulhuw    %%ymm14,%%ymm4,%%ymm4                              \n" \
  "vpaddsw     %%ymm4,%%ymm0,%%ymm0                               \n" \
  "vpaddsw     %%ymm4,%%ymm1,%%ymm1                               \n" \
  "vpaddsw     %%ymm4,%%ymm2,%%ymm2                               \n" \
  "vpsraw      $0x6,%%ymm0,%%ymm0                                 \n" \
  "vpsraw      $0x6,%%ymm1,%%ymm1                                 \n" \
  "vpsraw      $0x6,%%ymm2,%%ymm2                                 \n" \
  "vpackuswb   %%ymm0,%%ymm0,%%ymm0                               \n" \
  "vpackuswb   %%ymm1,%%ymm1,%%ymm1                               \n" \
  "vpackuswb   %%ymm2,%%ymm2,%%ymm2                               \n"
2023
#define YUVTORGB_REGS_AVX2 \
2024
  "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14",
2025
#else  // Convert 16 pixels: 16 UV and 16 Y.
2026
#define YUVTORGB_SETUP_AVX2(yuvconstants)
2027 2028
#define YUVTORGB_AVX2(yuvconstants) \
  "vpmaddubsw  " MEMACCESS2(64, [yuvconstants]) ",%%ymm0,%%ymm2   \n"        \
2029 2030 2031
    "vpmaddubsw  " MEMACCESS2(32, [yuvconstants]) ",%%ymm0,%%ymm1   \n"        \
    "vpmaddubsw  " MEMACCESS([yuvconstants]) ",%%ymm0,%%ymm0        \n"        \
    "vmovdqu     " MEMACCESS2(160, [yuvconstants]) ",%%ymm3         \n"        \
2032
    "vpsubw      %%ymm2,%%ymm3,%%ymm2                               \n"        \
2033
    "vmovdqu     " MEMACCESS2(128, [yuvconstants]) ",%%ymm3         \n"        \
2034
    "vpsubw      %%ymm1,%%ymm3,%%ymm1                               \n"        \
2035
    "vmovdqu     " MEMACCESS2(96, [yuvconstants]) ",%%ymm3          \n"        \
2036
    "vpsubw      %%ymm0,%%ymm3,%%ymm0                               \n"        \
2037
    "vpmulhuw    " MEMACCESS2(192, [yuvconstants]) ",%%ymm4,%%ymm4  \n"        \
2038 2039 2040 2041 2042 2043 2044 2045 2046
    "vpaddsw     %%ymm4,%%ymm0,%%ymm0                               \n"        \
    "vpaddsw     %%ymm4,%%ymm1,%%ymm1                               \n"        \
    "vpaddsw     %%ymm4,%%ymm2,%%ymm2                               \n"        \
    "vpsraw      $0x6,%%ymm0,%%ymm0                                 \n"        \
    "vpsraw      $0x6,%%ymm1,%%ymm1                                 \n"        \
    "vpsraw      $0x6,%%ymm2,%%ymm2                                 \n"        \
    "vpackuswb   %%ymm0,%%ymm0,%%ymm0                               \n"        \
    "vpackuswb   %%ymm1,%%ymm1,%%ymm1                               \n"        \
    "vpackuswb   %%ymm2,%%ymm2,%%ymm2                               \n"
2047 2048
#define YUVTORGB_REGS_AVX2
#endif
2049

2050
// Store 16 ARGB values.
2051 2052
#define STOREARGB_AVX2 \
  "vpunpcklbw %%ymm1,%%ymm0,%%ymm0                                \n"        \
2053 2054 2055 2056 2057 2058
    "vpermq     $0xd8,%%ymm0,%%ymm0                                 \n"        \
    "vpunpcklbw %%ymm5,%%ymm2,%%ymm2                                \n"        \
    "vpermq     $0xd8,%%ymm2,%%ymm2                                 \n"        \
    "vpunpcklwd %%ymm2,%%ymm0,%%ymm1                                \n"        \
    "vpunpckhwd %%ymm2,%%ymm0,%%ymm0                                \n"        \
    "vmovdqu    %%ymm1," MEMACCESS([dst_argb]) "                    \n"        \
2059 2060
    "vmovdqu    %%ymm0," MEMACCESS2(0x20, [dst_argb]) "             \n"        \
    "lea       " MEMLEA(0x40, [dst_argb]) ", %[dst_argb]            \n"
2061

2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073
#ifdef HAS_I444TOARGBROW_AVX2
// 16 pixels
// 16 UV values with 16 Y producing 16 ARGB (64 bytes).
void OMITFP I444ToARGBRow_AVX2(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* dst_argb,
                               const struct YuvConstants* yuvconstants,
                               int width) {
  asm volatile (
    YUVTORGB_SETUP_AVX2(yuvconstants)
    "sub       %[u_buf],%[v_buf]               \n"
2074
    "vpcmpeqb  %%ymm5,%%ymm5,%%ymm5            \n"
2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094
    LABELALIGN
  "1:                                          \n"
    READYUV444_AVX2
    YUVTORGB_AVX2(yuvconstants)
    STOREARGB_AVX2
    "sub       $0x10,%[width]                  \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [u_buf]"+r"(u_buf),    // %[u_buf]
    [v_buf]"+r"(v_buf),    // %[v_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
    [width]"+rm"(width)    // %[width]
  : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
  : "memory", "cc", NACL_R14 YUVTORGB_REGS_AVX2
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
  );
}
#endif  // HAS_I444TOARGBROW_AVX2

2095
#if defined(HAS_I422TOARGBROW_AVX2)
2096 2097
// 16 pixels
// 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes).
2098 2099 2100 2101
void OMITFP I422ToARGBRow_AVX2(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* dst_argb,
2102
                               const struct YuvConstants* yuvconstants,
2103
                               int width) {
2104
  asm volatile (
2105
    YUVTORGB_SETUP_AVX2(yuvconstants)
2106
    "sub       %[u_buf],%[v_buf]               \n"
2107
    "vpcmpeqb  %%ymm5,%%ymm5,%%ymm5            \n"
2108 2109 2110
    LABELALIGN
  "1:                                          \n"
    READYUV422_AVX2
2111
    YUVTORGB_AVX2(yuvconstants)
2112
    STOREARGB_AVX2
2113 2114
    "sub       $0x10,%[width]                  \n"
    "jg        1b                              \n"
2115

2116 2117 2118 2119 2120 2121
    "vzeroupper                                \n"
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [u_buf]"+r"(u_buf),    // %[u_buf]
    [v_buf]"+r"(v_buf),    // %[v_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
    [width]"+rm"(width)    // %[width]
2122
  : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
2123
  : "memory", "cc", NACL_R14 YUVTORGB_REGS_AVX2
2124
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
2125 2126
  );
}
2127
#endif  // HAS_I422TOARGBROW_AVX2
2128

2129 2130 2131 2132
#if defined(HAS_I422ALPHATOARGBROW_AVX2)
// 16 pixels
// 8 UV values upsampled to 16 UV, mixed with 16 Y and 16 A producing 16 ARGB.
void OMITFP I422AlphaToARGBRow_AVX2(const uint8* y_buf,
2133 2134 2135 2136 2137 2138 2139
                                    const uint8* u_buf,
                                    const uint8* v_buf,
                                    const uint8* a_buf,
                                    uint8* dst_argb,
                                    const struct YuvConstants* yuvconstants,
                                    int width) {
  // clang-format off
2140
  asm volatile (
2141
    YUVTORGB_SETUP_AVX2(yuvconstants)
2142 2143 2144 2145 2146 2147
    "sub       %[u_buf],%[v_buf]               \n"
    LABELALIGN
  "1:                                          \n"
    READYUVA422_AVX2
    YUVTORGB_AVX2(yuvconstants)
    STOREARGB_AVX2
2148
    "subl      $0x10,%[width]                  \n"
2149 2150 2151 2152 2153 2154 2155
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [u_buf]"+r"(u_buf),    // %[u_buf]
    [v_buf]"+r"(v_buf),    // %[v_buf]
    [a_buf]"+r"(a_buf),    // %[a_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
2156
#if defined(__i386__)
2157 2158
    [width]"+m"(width)     // %[width]
#else
2159
    [width]"+rm"(width)    // %[width]
2160
#endif
2161
  : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
2162
  : "memory", "cc", NACL_R14 YUVTORGB_REGS_AVX2
2163 2164
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
  );
2165
  // clang-format on
2166 2167 2168
}
#endif  // HAS_I422ALPHATOARGBROW_AVX2

2169 2170 2171 2172 2173 2174 2175
#if defined(HAS_I422TORGBAROW_AVX2)
// 16 pixels
// 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 RGBA (64 bytes).
void OMITFP I422ToRGBARow_AVX2(const uint8* y_buf,
                               const uint8* u_buf,
                               const uint8* v_buf,
                               uint8* dst_argb,
2176
                               const struct YuvConstants* yuvconstants,
2177 2178
                               int width) {
  asm volatile (
2179
    YUVTORGB_SETUP_AVX2(yuvconstants)
2180 2181 2182 2183 2184
    "sub       %[u_buf],%[v_buf]               \n"
    "vpcmpeqb   %%ymm5,%%ymm5,%%ymm5           \n"
    LABELALIGN
  "1:                                          \n"
    READYUV422_AVX2
2185
    YUVTORGB_AVX2(yuvconstants)
2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204

    // Step 3: Weave into RGBA
    "vpunpcklbw %%ymm2,%%ymm1,%%ymm1           \n"
    "vpermq     $0xd8,%%ymm1,%%ymm1            \n"
    "vpunpcklbw %%ymm0,%%ymm5,%%ymm2           \n"
    "vpermq     $0xd8,%%ymm2,%%ymm2            \n"
    "vpunpcklwd %%ymm1,%%ymm2,%%ymm0           \n"
    "vpunpckhwd %%ymm1,%%ymm2,%%ymm1           \n"
    "vmovdqu    %%ymm0," MEMACCESS([dst_argb]) "\n"
    "vmovdqu    %%ymm1," MEMACCESS2(0x20,[dst_argb]) "\n"
    "lea       " MEMLEA(0x40,[dst_argb]) ",%[dst_argb] \n"
    "sub       $0x10,%[width]                  \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [u_buf]"+r"(u_buf),    // %[u_buf]
    [v_buf]"+r"(v_buf),    // %[v_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
    [width]"+rm"(width)    // %[width]
2205
  : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
2206
  : "memory", "cc", NACL_R14 YUVTORGB_REGS_AVX2
2207
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
2208 2209 2210
  );
}
#endif  // HAS_I422TORGBAROW_AVX2
2211

2212 2213 2214 2215 2216 2217
#if defined(HAS_NV12TOARGBROW_AVX2)
// 16 pixels.
// 8 UV values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes).
void OMITFP NV12ToARGBRow_AVX2(const uint8* y_buf,
                               const uint8* uv_buf,
                               uint8* dst_argb,
2218
                               const struct YuvConstants* yuvconstants,
2219
                               int width) {
2220
  // clang-format off
2221
  asm volatile (
2222
    YUVTORGB_SETUP_AVX2(yuvconstants)
2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236
    "vpcmpeqb   %%ymm5,%%ymm5,%%ymm5           \n"
    LABELALIGN
  "1:                                          \n"
    READNV12_AVX2
    YUVTORGB_AVX2(yuvconstants)
    STOREARGB_AVX2
    "sub       $0x10,%[width]                  \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [uv_buf]"+r"(uv_buf),    // %[uv_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
    [width]"+rm"(width)    // %[width]
  : [yuvconstants]"r"(yuvconstants)  // %[yuvconstants]
2237 2238
    : "memory", "cc", YUVTORGB_REGS_AVX2  // Does not use r14.
    "xmm0", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
2239
  );
2240
  // clang-format on
2241
}
2242
#endif  // HAS_NV12TOARGBROW_AVX2
2243

2244 2245 2246 2247 2248 2249
#if defined(HAS_NV21TOARGBROW_AVX2)
// 16 pixels.
// 8 VU values upsampled to 16 UV, mixed with 16 Y producing 16 ARGB (64 bytes).
void OMITFP NV21ToARGBRow_AVX2(const uint8* y_buf,
                               const uint8* vu_buf,
                               uint8* dst_argb,
2250
                               const struct YuvConstants* yuvconstants,
2251
                               int width) {
2252
  // clang-format off
2253
  asm volatile (
2254
    YUVTORGB_SETUP_AVX2(yuvconstants)
2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269
    "vpcmpeqb   %%ymm5,%%ymm5,%%ymm5           \n"
    LABELALIGN
  "1:                                          \n"
    READNV21_AVX2
    YUVTORGB_AVX2(yuvconstants)
    STOREARGB_AVX2
    "sub       $0x10,%[width]                  \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : [y_buf]"+r"(y_buf),    // %[y_buf]
    [vu_buf]"+r"(vu_buf),    // %[vu_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
    [width]"+rm"(width)    // %[width]
  : [yuvconstants]"r"(yuvconstants), // %[yuvconstants]
    [kShuffleNV21]"m"(kShuffleNV21)
2270 2271
    : "memory", "cc", YUVTORGB_REGS_AVX2  // Does not use r14.
      "xmm0", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
2272
  );
2273
  // clang-format on
2274 2275
}
#endif  // HAS_NV21TOARGBROW_AVX2
2276

2277 2278 2279 2280 2281
#if defined(HAS_YUY2TOARGBROW_AVX2)
// 16 pixels.
// 8 YUY2 values with 16 Y and 8 UV producing 16 ARGB (64 bytes).
void OMITFP YUY2ToARGBRow_AVX2(const uint8* yuy2_buf,
                               uint8* dst_argb,
2282
                               const struct YuvConstants* yuvconstants,
2283
                               int width) {
2284
  // clang-format off
2285
  asm volatile (
2286
    YUVTORGB_SETUP_AVX2(yuvconstants)
2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301
    "vpcmpeqb   %%ymm5,%%ymm5,%%ymm5           \n"
    LABELALIGN
  "1:                                          \n"
    READYUY2_AVX2
    YUVTORGB_AVX2(yuvconstants)
    STOREARGB_AVX2
    "sub       $0x10,%[width]                  \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : [yuy2_buf]"+r"(yuy2_buf),    // %[yuy2_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
    [width]"+rm"(width)    // %[width]
  : [yuvconstants]"r"(yuvconstants), // %[yuvconstants]
    [kShuffleYUY2Y]"m"(kShuffleYUY2Y),
    [kShuffleYUY2UV]"m"(kShuffleYUY2UV)
2302 2303
    : "memory", "cc", YUVTORGB_REGS_AVX2  // Does not use r14.
      "xmm0", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
2304
  );
2305
  // clang-format on
2306 2307 2308 2309 2310 2311 2312 2313
}
#endif  // HAS_YUY2TOARGBROW_AVX2

#if defined(HAS_UYVYTOARGBROW_AVX2)
// 16 pixels.
// 8 UYVY values with 16 Y and 8 UV producing 16 ARGB (64 bytes).
void OMITFP UYVYToARGBRow_AVX2(const uint8* uyvy_buf,
                               uint8* dst_argb,
2314
                               const struct YuvConstants* yuvconstants,
2315
                               int width) {
2316
  // clang-format off
2317
  asm volatile (
2318
    YUVTORGB_SETUP_AVX2(yuvconstants)
2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333
    "vpcmpeqb   %%ymm5,%%ymm5,%%ymm5           \n"
    LABELALIGN
  "1:                                          \n"
    READUYVY_AVX2
    YUVTORGB_AVX2(yuvconstants)
    STOREARGB_AVX2
    "sub       $0x10,%[width]                  \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : [uyvy_buf]"+r"(uyvy_buf),    // %[uyvy_buf]
    [dst_argb]"+r"(dst_argb),  // %[dst_argb]
    [width]"+rm"(width)    // %[width]
  : [yuvconstants]"r"(yuvconstants), // %[yuvconstants]
    [kShuffleUYVYY]"m"(kShuffleUYVYY),
    [kShuffleUYVYUV]"m"(kShuffleUYVYUV)
2334 2335
    : "memory", "cc", YUVTORGB_REGS_AVX2  // Does not use r14.
      "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
2336
  );
2337
  // clang-format on
2338 2339 2340
}
#endif  // HAS_UYVYTOARGBROW_AVX2

2341 2342
#ifdef HAS_I400TOARGBROW_SSE2
void I400ToARGBRow_SSE2(const uint8* y_buf, uint8* dst_argb, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
2343
  asm volatile (
2344
    "mov       $0x4a354a35,%%eax               \n"  // 4a35 = 18997 = 1.164
2345 2346
    "movd      %%eax,%%xmm2                    \n"
    "pshufd    $0x0,%%xmm2,%%xmm2              \n"
2347 2348 2349 2350 2351
    "mov       $0x04880488,%%eax               \n"  // 0488 = 1160 = 1.164 * 16
    "movd      %%eax,%%xmm3                    \n"
    "pshufd    $0x0,%%xmm3,%%xmm3              \n"
    "pcmpeqb   %%xmm4,%%xmm4                   \n"
    "pslld     $0x18,%%xmm4                    \n"
2352
    LABELALIGN
2353
  "1:                                          \n"
2354
    // Step 1: Scale Y contribution to 8 G values. G = (y - 16) * 1.164
2355 2356
    "movq      " MEMACCESS(0) ",%%xmm0         \n"
    "lea       " MEMLEA(0x8,0) ",%0            \n"
2357 2358
    "punpcklbw %%xmm0,%%xmm0                   \n"
    "pmulhuw   %%xmm2,%%xmm0                   \n"
2359
    "psubusw   %%xmm3,%%xmm0                   \n"
2360
    "psrlw     $6, %%xmm0                      \n"
2361
    "packuswb  %%xmm0,%%xmm0                   \n"
2362 2363

    // Step 2: Weave into ARGB
2364 2365 2366 2367 2368 2369
    "punpcklbw %%xmm0,%%xmm0                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "punpcklwd %%xmm0,%%xmm0                   \n"
    "punpckhwd %%xmm1,%%xmm1                   \n"
    "por       %%xmm4,%%xmm0                   \n"
    "por       %%xmm4,%%xmm1                   \n"
2370 2371
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
2372
    "lea       " MEMLEA(0x20,1) ",%1           \n"
2373 2374

    "sub       $0x8,%2                         \n"
2375
    "jg        1b                              \n"
2376 2377 2378
  : "+r"(y_buf),     // %0
    "+r"(dst_argb),  // %1
    "+rm"(width)     // %2
2379 2380 2381
  :
  : "memory", "cc", "eax"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4"
2382
  );
2383
}
2384
#endif  // HAS_I400TOARGBROW_SSE2
mikhal@webrtc.org's avatar
mikhal@webrtc.org committed
2385

2386
#ifdef HAS_I400TOARGBROW_AVX2
2387 2388
// 16 pixels of Y converted to 16 pixels of ARGB (64 bytes).
// note: vpunpcklbw mutates and vpackuswb unmutates.
2389
void I400ToARGBRow_AVX2(const uint8* y_buf, uint8* dst_argb, int width) {
2390 2391 2392 2393 2394 2395 2396 2397 2398
  asm volatile (
    "mov        $0x4a354a35,%%eax              \n" // 0488 = 1160 = 1.164 * 16
    "vmovd      %%eax,%%xmm2                   \n"
    "vbroadcastss %%xmm2,%%ymm2                \n"
    "mov        $0x4880488,%%eax               \n" // 4a35 = 18997 = 1.164
    "vmovd      %%eax,%%xmm3                   \n"
    "vbroadcastss %%xmm3,%%ymm3                \n"
    "vpcmpeqb   %%ymm4,%%ymm4,%%ymm4           \n"
    "vpslld     $0x18,%%ymm4,%%ymm4            \n"
2399

2400 2401 2402
    LABELALIGN
  "1:                                          \n"
    // Step 1: Scale Y contribution to 16 G values. G = (y - 16) * 1.164
2403 2404
    "vmovdqu    " MEMACCESS(0) ",%%xmm0        \n"
    "lea        " MEMLEA(0x10,0) ",%0          \n"
2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416
    "vpermq     $0xd8,%%ymm0,%%ymm0            \n"
    "vpunpcklbw %%ymm0,%%ymm0,%%ymm0           \n"
    "vpmulhuw   %%ymm2,%%ymm0,%%ymm0           \n"
    "vpsubusw   %%ymm3,%%ymm0,%%ymm0           \n"
    "vpsrlw     $0x6,%%ymm0,%%ymm0             \n"
    "vpackuswb  %%ymm0,%%ymm0,%%ymm0           \n"
    "vpunpcklbw %%ymm0,%%ymm0,%%ymm1           \n"
    "vpermq     $0xd8,%%ymm1,%%ymm1            \n"
    "vpunpcklwd %%ymm1,%%ymm1,%%ymm0           \n"
    "vpunpckhwd %%ymm1,%%ymm1,%%ymm1           \n"
    "vpor       %%ymm4,%%ymm0,%%ymm0           \n"
    "vpor       %%ymm4,%%ymm1,%%ymm1           \n"
2417 2418 2419
    "vmovdqu    %%ymm0," MEMACCESS(1) "        \n"
    "vmovdqu    %%ymm1," MEMACCESS2(0x20,1) "  \n"
    "lea       " MEMLEA(0x40,1) ",%1           \n"
2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430
    "sub        $0x10,%2                       \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(y_buf),     // %0
    "+r"(dst_argb),  // %1
    "+rm"(width)     // %2
  :
  : "memory", "cc", "eax"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4"
  );
}
2431
#endif  // HAS_I400TOARGBROW_AVX2
2432

2433
#ifdef HAS_MIRRORROW_SSSE3
2434
// Shuffle table for reversing the bytes.
2435 2436
static uvec8 kShuffleMirror = {15u, 14u, 13u, 12u, 11u, 10u, 9u, 8u,
                               7u,  6u,  5u,  4u,  3u,  2u,  1u, 0u};
2437

2438
void MirrorRow_SSSE3(const uint8* src, uint8* dst, int width) {
2439
  intptr_t temp_width = (intptr_t)(width);
fbarchard@google.com's avatar
fbarchard@google.com committed
2440
  asm volatile (
2441
    "movdqa    %3,%%xmm5                       \n"
2442
    LABELALIGN
2443
  "1:                                          \n"
2444
    MEMOPREG(movdqu,-0x10,0,2,1,xmm0)          //  movdqu -0x10(%0,%2),%%xmm0
2445
    "pshufb    %%xmm5,%%xmm0                   \n"
2446
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
2447
    "lea       " MEMLEA(0x10,1) ",%1           \n"
2448
    "sub       $0x10,%2                        \n"
2449
    "jg        1b                              \n"
2450 2451 2452
  : "+r"(src),  // %0
    "+r"(dst),  // %1
    "+r"(temp_width)  // %2
2453
  : "m"(kShuffleMirror) // %3
2454 2455
  : "memory", "cc", NACL_R14
    "xmm0", "xmm5"
2456 2457
  );
}
2458
#endif  // HAS_MIRRORROW_SSSE3
2459

2460 2461 2462 2463 2464 2465 2466
#ifdef HAS_MIRRORROW_AVX2
void MirrorRow_AVX2(const uint8* src, uint8* dst, int width) {
  intptr_t temp_width = (intptr_t)(width);
  asm volatile (
    "vbroadcastf128 %3,%%ymm5                  \n"
    LABELALIGN
  "1:                                          \n"
2467
    MEMOPREG(vmovdqu,-0x20,0,2,1,ymm0)         //  vmovdqu -0x20(%0,%2),%%ymm0
2468 2469 2470 2471
    "vpshufb    %%ymm5,%%ymm0,%%ymm0           \n"
    "vpermq     $0x4e,%%ymm0,%%ymm0            \n"
    "vmovdqu    %%ymm0," MEMACCESS(1) "        \n"
    "lea       " MEMLEA(0x20,1) ",%1           \n"
2472
    "sub       $0x20,%2                        \n"
2473 2474 2475 2476 2477 2478
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src),  // %0
    "+r"(dst),  // %1
    "+r"(temp_width)  // %2
  : "m"(kShuffleMirror) // %3
2479 2480
  : "memory", "cc", NACL_R14
    "xmm0", "xmm5"
2481 2482 2483 2484
  );
}
#endif  // HAS_MIRRORROW_AVX2

2485
#ifdef HAS_MIRRORUVROW_SSSE3
2486
// Shuffle table for reversing the bytes of UV channels.
2487 2488 2489 2490 2491
static uvec8 kShuffleMirrorUV = {14u, 12u, 10u, 8u, 6u, 4u, 2u, 0u,
                                 15u, 13u, 11u, 9u, 7u, 5u, 3u, 1u};
void MirrorUVRow_SSSE3(const uint8* src,
                       uint8* dst_u,
                       uint8* dst_v,
2492
                       int width) {
2493
  intptr_t temp_width = (intptr_t)(width);
fbarchard@google.com's avatar
fbarchard@google.com committed
2494
  asm volatile (
2495
    "movdqa    %4,%%xmm1                       \n"
2496
    "lea       " MEMLEA4(-0x10,0,3,2) ",%0     \n"
2497
    "sub       %1,%2                           \n"
2498
    LABELALIGN
2499
  "1:                                          \n"
2500
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
2501
    "lea       " MEMLEA(-0x10,0) ",%0          \n"
2502
    "pshufb    %%xmm1,%%xmm0                   \n"
2503 2504 2505
    "movlpd    %%xmm0," MEMACCESS(1) "         \n"
    MEMOPMEM(movhpd,xmm0,0x00,1,2,1)           //  movhpd    %%xmm0,(%1,%2)
    "lea       " MEMLEA(0x8,1) ",%1            \n"
2506
    "sub       $8,%3                           \n"
2507
    "jg        1b                              \n"
2508 2509 2510 2511
  : "+r"(src),      // %0
    "+r"(dst_u),    // %1
    "+r"(dst_v),    // %2
    "+r"(temp_width)  // %3
2512
  : "m"(kShuffleMirrorUV)  // %4
2513 2514
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1"
2515 2516
  );
}
2517
#endif  // HAS_MIRRORUVROW_SSSE3
2518

2519
#ifdef HAS_ARGBMIRRORROW_SSE2
2520

2521
void ARGBMirrorRow_SSE2(const uint8* src, uint8* dst, int width) {
2522
  intptr_t temp_width = (intptr_t)(width);
2523
  asm volatile (
2524
    "lea       " MEMLEA4(-0x10,0,2,4) ",%0     \n"
2525
    LABELALIGN
2526
  "1:                                          \n"
2527
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
2528
    "pshufd    $0x1b,%%xmm0,%%xmm0             \n"
2529
    "lea       " MEMLEA(-0x10,0) ",%0          \n"
2530
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
2531
    "lea       " MEMLEA(0x10,1) ",%1           \n"
2532
    "sub       $0x4,%2                         \n"
2533 2534 2535 2536
    "jg        1b                              \n"
  : "+r"(src),  // %0
    "+r"(dst),  // %1
    "+r"(temp_width)  // %2
2537
  :
2538
  : "memory", "cc"
2539
    , "xmm0"
2540 2541
  );
}
2542
#endif  // HAS_ARGBMIRRORROW_SSE2
2543

2544 2545
#ifdef HAS_ARGBMIRRORROW_AVX2
// Shuffle table for reversing the bytes.
2546
static const ulvec32 kARGBShuffleMirror_AVX2 = {7u, 6u, 5u, 4u, 3u, 2u, 1u, 0u};
2547 2548 2549
void ARGBMirrorRow_AVX2(const uint8* src, uint8* dst, int width) {
  intptr_t temp_width = (intptr_t)(width);
  asm volatile (
2550
    "vmovdqu    %3,%%ymm5                      \n"
2551 2552 2553 2554
    LABELALIGN
  "1:                                          \n"
    VMEMOPREG(vpermd,-0x20,0,2,4,ymm5,ymm0) // vpermd -0x20(%0,%2,4),ymm5,ymm0
    "vmovdqu    %%ymm0," MEMACCESS(1) "        \n"
2555
    "lea        " MEMLEA(0x20,1) ",%1          \n"
2556
    "sub        $0x8,%2                        \n"
2557
    "jg         1b                             \n"
2558 2559 2560 2561 2562
    "vzeroupper                                \n"
  : "+r"(src),  // %0
    "+r"(dst),  // %1
    "+r"(temp_width)  // %2
  : "m"(kARGBShuffleMirror_AVX2) // %3
2563 2564
  : "memory", "cc", NACL_R14
    "xmm0", "xmm5"
2565 2566 2567 2568
  );
}
#endif  // HAS_ARGBMIRRORROW_AVX2

2569
#ifdef HAS_SPLITUVROW_AVX2
2570 2571 2572
void SplitUVRow_AVX2(const uint8* src_uv,
                     uint8* dst_u,
                     uint8* dst_v,
2573
                     int width) {
2574
  asm volatile (
2575 2576
    "vpcmpeqb   %%ymm5,%%ymm5,%%ymm5             \n"
    "vpsrlw     $0x8,%%ymm5,%%ymm5               \n"
2577 2578
    "sub        %1,%2                            \n"
    LABELALIGN
Frank Barchard's avatar
Frank Barchard committed
2579
  "1:                                          \n"
2580 2581 2582
    "vmovdqu    " MEMACCESS(0) ",%%ymm0          \n"
    "vmovdqu    " MEMACCESS2(0x20,0) ",%%ymm1    \n"
    "lea        " MEMLEA(0x40,0) ",%0            \n"
2583 2584 2585 2586 2587 2588 2589 2590
    "vpsrlw     $0x8,%%ymm0,%%ymm2               \n"
    "vpsrlw     $0x8,%%ymm1,%%ymm3               \n"
    "vpand      %%ymm5,%%ymm0,%%ymm0             \n"
    "vpand      %%ymm5,%%ymm1,%%ymm1             \n"
    "vpackuswb  %%ymm1,%%ymm0,%%ymm0             \n"
    "vpackuswb  %%ymm3,%%ymm2,%%ymm2             \n"
    "vpermq     $0xd8,%%ymm0,%%ymm0              \n"
    "vpermq     $0xd8,%%ymm2,%%ymm2              \n"
2591 2592 2593 2594 2595
    "vmovdqu    %%ymm0," MEMACCESS(1) "          \n"
    MEMOPMEM(vmovdqu,ymm2,0x00,1,2,1)             //  vmovdqu %%ymm2,(%1,%2)
    "lea        " MEMLEA(0x20,1) ",%1            \n"
    "sub        $0x20,%3                         \n"
    "jg         1b                               \n"
2596
    "vzeroupper                                  \n"
2597 2598 2599
  : "+r"(src_uv),     // %0
    "+r"(dst_u),      // %1
    "+r"(dst_v),      // %2
2600
    "+r"(width)         // %3
2601
  :
2602 2603
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
2604 2605 2606 2607
  );
}
#endif  // HAS_SPLITUVROW_AVX2

2608
#ifdef HAS_SPLITUVROW_SSE2
2609 2610 2611
void SplitUVRow_SSE2(const uint8* src_uv,
                     uint8* dst_u,
                     uint8* dst_v,
2612
                     int width) {
2613 2614 2615 2616
  asm volatile (
    "pcmpeqb    %%xmm5,%%xmm5                    \n"
    "psrlw      $0x8,%%xmm5                      \n"
    "sub        %1,%2                            \n"
2617
    LABELALIGN
Frank Barchard's avatar
Frank Barchard committed
2618
  "1:                                          \n"
2619 2620 2621
    "movdqu     " MEMACCESS(0) ",%%xmm0          \n"
    "movdqu     " MEMACCESS2(0x10,0) ",%%xmm1    \n"
    "lea        " MEMLEA(0x20,0) ",%0            \n"
2622 2623 2624 2625 2626 2627 2628 2629
    "movdqa     %%xmm0,%%xmm2                    \n"
    "movdqa     %%xmm1,%%xmm3                    \n"
    "pand       %%xmm5,%%xmm0                    \n"
    "pand       %%xmm5,%%xmm1                    \n"
    "packuswb   %%xmm1,%%xmm0                    \n"
    "psrlw      $0x8,%%xmm2                      \n"
    "psrlw      $0x8,%%xmm3                      \n"
    "packuswb   %%xmm3,%%xmm2                    \n"
2630 2631 2632
    "movdqu     %%xmm0," MEMACCESS(1) "          \n"
    MEMOPMEM(movdqu,xmm2,0x00,1,2,1)             //  movdqu     %%xmm2,(%1,%2)
    "lea        " MEMLEA(0x10,1) ",%1            \n"
2633 2634 2635 2636 2637
    "sub        $0x10,%3                         \n"
    "jg         1b                               \n"
  : "+r"(src_uv),     // %0
    "+r"(dst_u),      // %1
    "+r"(dst_v),      // %2
2638
    "+r"(width)         // %3
fbarchard@google.com's avatar
fbarchard@google.com committed
2639
  :
2640 2641
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
2642 2643
  );
}
2644
#endif  // HAS_SPLITUVROW_SSE2
2645

2646
#ifdef HAS_MERGEUVROW_AVX2
2647 2648 2649
void MergeUVRow_AVX2(const uint8* src_u,
                     const uint8* src_v,
                     uint8* dst_uv,
2650 2651 2652 2653
                     int width) {
  asm volatile (
    "sub       %0,%1                             \n"
    LABELALIGN
Frank Barchard's avatar
Frank Barchard committed
2654
  "1:                                          \n"
2655 2656 2657 2658 2659
    "vmovdqu   " MEMACCESS(0) ",%%ymm0           \n"
    MEMOPREG(vmovdqu,0x00,0,1,1,ymm1)             //  vmovdqu (%0,%1,1),%%ymm1
    "lea       " MEMLEA(0x20,0) ",%0             \n"
    "vpunpcklbw %%ymm1,%%ymm0,%%ymm2             \n"
    "vpunpckhbw %%ymm1,%%ymm0,%%ymm0             \n"
2660 2661 2662 2663
    "vextractf128 $0x0,%%ymm2," MEMACCESS(2) "   \n"
    "vextractf128 $0x0,%%ymm0," MEMACCESS2(0x10,2) "\n"
    "vextractf128 $0x1,%%ymm2," MEMACCESS2(0x20,2) "\n"
    "vextractf128 $0x1,%%ymm0," MEMACCESS2(0x30,2) "\n"
2664 2665 2666 2667 2668 2669 2670 2671 2672
    "lea       " MEMLEA(0x40,2) ",%2             \n"
    "sub       $0x20,%3                          \n"
    "jg        1b                                \n"
    "vzeroupper                                  \n"
  : "+r"(src_u),     // %0
    "+r"(src_v),     // %1
    "+r"(dst_uv),    // %2
    "+r"(width)      // %3
  :
2673 2674
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2"
2675 2676 2677 2678
  );
}
#endif  // HAS_MERGEUVROW_AVX2

2679
#ifdef HAS_MERGEUVROW_SSE2
2680 2681 2682
void MergeUVRow_SSE2(const uint8* src_u,
                     const uint8* src_v,
                     uint8* dst_uv,
2683
                     int width) {
2684 2685
  asm volatile (
    "sub       %0,%1                             \n"
2686
    LABELALIGN
Frank Barchard's avatar
Frank Barchard committed
2687
  "1:                                          \n"
2688
    "movdqu    " MEMACCESS(0) ",%%xmm0           \n"
2689
    MEMOPREG(movdqu,0x00,0,1,1,xmm1)             //  movdqu    (%0,%1,1),%%xmm1
2690
    "lea       " MEMLEA(0x10,0) ",%0             \n"
2691 2692 2693
    "movdqa    %%xmm0,%%xmm2                     \n"
    "punpcklbw %%xmm1,%%xmm0                     \n"
    "punpckhbw %%xmm1,%%xmm2                     \n"
2694 2695
    "movdqu    %%xmm0," MEMACCESS(2) "           \n"
    "movdqu    %%xmm2," MEMACCESS2(0x10,2) "     \n"
2696
    "lea       " MEMLEA(0x20,2) ",%2             \n"
2697 2698 2699 2700 2701 2702
    "sub       $0x10,%3                          \n"
    "jg        1b                                \n"
  : "+r"(src_u),     // %0
    "+r"(src_v),     // %1
    "+r"(dst_uv),    // %2
    "+r"(width)      // %3
2703
  :
2704 2705
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2"
2706 2707
  );
}
2708
#endif  // HAS_MERGEUVROW_SSE2
2709

2710 2711
#ifdef HAS_COPYROW_SSE2
void CopyRow_SSE2(const uint8* src, uint8* dst, int count) {
fbarchard@google.com's avatar
fbarchard@google.com committed
2712
  asm volatile (
2713 2714 2715 2716
    "test       $0xf,%0                        \n"
    "jne        2f                             \n"
    "test       $0xf,%1                        \n"
    "jne        2f                             \n"
2717
    LABELALIGN
2718
  "1:                                          \n"
2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729
    "movdqa    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqa    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "lea       " MEMLEA(0x20,0) ",%0           \n"
    "movdqa    %%xmm0," MEMACCESS(1) "         \n"
    "movdqa    %%xmm1," MEMACCESS2(0x10,1) "   \n"
    "lea       " MEMLEA(0x20,1) ",%1           \n"
    "sub       $0x20,%2                        \n"
    "jg        1b                              \n"
    "jmp       9f                              \n"
    LABELALIGN
  "2:                                          \n"
2730 2731
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
2732
    "lea       " MEMLEA(0x20,0) ",%0           \n"
2733 2734
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
2735
    "lea       " MEMLEA(0x20,1) ",%1           \n"
2736
    "sub       $0x20,%2                        \n"
2737 2738
    "jg        2b                              \n"
  "9:                                          \n"
2739 2740 2741 2742 2743 2744 2745 2746 2747 2748
  : "+r"(src),   // %0
    "+r"(dst),   // %1
    "+r"(count)  // %2
  :
  : "memory", "cc"
    , "xmm0", "xmm1"
  );
}
#endif  // HAS_COPYROW_SSE2

2749 2750 2751 2752 2753
#ifdef HAS_COPYROW_AVX
void CopyRow_AVX(const uint8* src, uint8* dst, int count) {
  asm volatile (
    LABELALIGN
  "1:                                          \n"
2754 2755
    "vmovdqu   " MEMACCESS(0) ",%%ymm0         \n"
    "vmovdqu   " MEMACCESS2(0x20,0) ",%%ymm1   \n"
2756
    "lea       " MEMLEA(0x40,0) ",%0           \n"
2757 2758
    "vmovdqu   %%ymm0," MEMACCESS(1) "         \n"
    "vmovdqu   %%ymm1," MEMACCESS2(0x20,1) "   \n"
2759 2760 2761 2762 2763 2764 2765 2766
    "lea       " MEMLEA(0x40,1) ",%1           \n"
    "sub       $0x40,%2                        \n"
    "jg        1b                              \n"
  : "+r"(src),   // %0
    "+r"(dst),   // %1
    "+r"(count)  // %2
  :
  : "memory", "cc"
2767
    , "xmm0", "xmm1"
2768 2769 2770 2771
  );
}
#endif  // HAS_COPYROW_AVX

2772
#ifdef HAS_COPYROW_ERMS
2773
// Multiple of 1.
2774
void CopyRow_ERMS(const uint8* src, uint8* dst, int width) {
2775
  size_t width_tmp = (size_t)(width);
2776 2777 2778 2779 2780 2781
  asm volatile("rep movsb " MEMMOVESTRING(0, 1) "          \n"
               : "+S"(src),       // %0
                 "+D"(dst),       // %1
                 "+c"(width_tmp)  // %2
               :
               : "memory", "cc");
2782
}
2783
#endif  // HAS_COPYROW_ERMS
2784

2785 2786 2787 2788 2789 2790 2791 2792
#ifdef HAS_ARGBCOPYALPHAROW_SSE2
// width in pixels
void ARGBCopyAlphaRow_SSE2(const uint8* src, uint8* dst, int width) {
  asm volatile (
    "pcmpeqb   %%xmm0,%%xmm0                   \n"
    "pslld     $0x18,%%xmm0                    \n"
    "pcmpeqb   %%xmm1,%%xmm1                   \n"
    "psrld     $0x8,%%xmm1                     \n"
2793
    LABELALIGN
2794
  "1:                                          \n"
2795 2796
    "movdqu    " MEMACCESS(0) ",%%xmm2         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm3   \n"
2797
    "lea       " MEMLEA(0x20,0) ",%0           \n"
2798 2799
    "movdqu    " MEMACCESS(1) ",%%xmm4         \n"
    "movdqu    " MEMACCESS2(0x10,1) ",%%xmm5   \n"
2800 2801 2802 2803 2804 2805
    "pand      %%xmm0,%%xmm2                   \n"
    "pand      %%xmm0,%%xmm3                   \n"
    "pand      %%xmm1,%%xmm4                   \n"
    "pand      %%xmm1,%%xmm5                   \n"
    "por       %%xmm4,%%xmm2                   \n"
    "por       %%xmm5,%%xmm3                   \n"
2806 2807
    "movdqu    %%xmm2," MEMACCESS(1) "         \n"
    "movdqu    %%xmm3," MEMACCESS2(0x10,1) "   \n"
2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826
    "lea       " MEMLEA(0x20,1) ",%1           \n"
    "sub       $0x8,%2                         \n"
    "jg        1b                              \n"
  : "+r"(src),   // %0
    "+r"(dst),   // %1
    "+r"(width)  // %2
  :
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
  );
}
#endif  // HAS_ARGBCOPYALPHAROW_SSE2

#ifdef HAS_ARGBCOPYALPHAROW_AVX2
// width in pixels
void ARGBCopyAlphaRow_AVX2(const uint8* src, uint8* dst, int width) {
  asm volatile (
    "vpcmpeqb  %%ymm0,%%ymm0,%%ymm0            \n"
    "vpsrld    $0x8,%%ymm0,%%ymm0              \n"
2827
    LABELALIGN
2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849
  "1:                                          \n"
    "vmovdqu   " MEMACCESS(0) ",%%ymm1         \n"
    "vmovdqu   " MEMACCESS2(0x20,0) ",%%ymm2   \n"
    "lea       " MEMLEA(0x40,0) ",%0           \n"
    "vpblendvb %%ymm0," MEMACCESS(1) ",%%ymm1,%%ymm1        \n"
    "vpblendvb %%ymm0," MEMACCESS2(0x20,1) ",%%ymm2,%%ymm2  \n"
    "vmovdqu   %%ymm1," MEMACCESS(1) "         \n"
    "vmovdqu   %%ymm2," MEMACCESS2(0x20,1) "   \n"
    "lea       " MEMLEA(0x40,1) ",%1           \n"
    "sub       $0x10,%2                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src),   // %0
    "+r"(dst),   // %1
    "+r"(width)  // %2
  :
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2"
  );
}
#endif  // HAS_ARGBCOPYALPHAROW_AVX2

2850 2851 2852
#ifdef HAS_ARGBEXTRACTALPHAROW_SSE2
// width in pixels
void ARGBExtractAlphaRow_SSE2(const uint8* src_argb, uint8* dst_a, int width) {
2853
  asm volatile (
2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876
    LABELALIGN
  "1:                                          \n"
    "movdqu    " MEMACCESS(0) ", %%xmm0        \n"
    "movdqu    " MEMACCESS2(0x10, 0) ", %%xmm1 \n"
    "lea       " MEMLEA(0x20, 0) ", %0         \n"
    "psrld     $0x18, %%xmm0                   \n"
    "psrld     $0x18, %%xmm1                   \n"
    "packssdw  %%xmm1, %%xmm0                  \n"
    "packuswb  %%xmm0, %%xmm0                  \n"
    "movq      %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x8, 1) ", %1          \n"
    "sub       $0x8, %2                        \n"
    "jg        1b                              \n"
  : "+r"(src_argb),  // %0
    "+r"(dst_a),     // %1
    "+rm"(width)     // %2
  :
  : "memory", "cc"
    , "xmm0", "xmm1"
  );
}
#endif  // HAS_ARGBEXTRACTALPHAROW_SSE2

2877 2878
#ifdef HAS_ARGBEXTRACTALPHAROW_AVX2
static const uvec8 kShuffleAlphaShort_AVX2 = {
2879 2880
    3u,  128u, 128u, 128u, 7u,  128u, 128u, 128u,
    11u, 128u, 128u, 128u, 15u, 128u, 128u, 128u};
2881 2882

void ARGBExtractAlphaRow_AVX2(const uint8* src_argb, uint8* dst_a, int width) {
2883
  asm volatile (
2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916
    "vmovdqa    %3,%%ymm4                      \n"
    "vbroadcastf128 %4,%%ymm5                  \n"
    LABELALIGN
  "1:                                          \n"
    "vmovdqu   " MEMACCESS(0) ", %%ymm0        \n"
    "vmovdqu   " MEMACCESS2(0x20, 0) ", %%ymm1 \n"
    "vpshufb    %%ymm5,%%ymm0,%%ymm0           \n" // vpsrld $0x18, %%ymm0
    "vpshufb    %%ymm5,%%ymm1,%%ymm1           \n"
    "vmovdqu   " MEMACCESS2(0x40, 0) ", %%ymm2 \n"
    "vmovdqu   " MEMACCESS2(0x60, 0) ", %%ymm3 \n"
    "lea       " MEMLEA(0x80, 0) ", %0         \n"
    "vpackssdw  %%ymm1, %%ymm0, %%ymm0         \n"  // mutates
    "vpshufb    %%ymm5,%%ymm2,%%ymm2           \n"
    "vpshufb    %%ymm5,%%ymm3,%%ymm3           \n"
    "vpackssdw  %%ymm3, %%ymm2, %%ymm2         \n"  // mutates
    "vpackuswb  %%ymm2,%%ymm0,%%ymm0           \n"  // mutates.
    "vpermd     %%ymm0,%%ymm4,%%ymm0           \n"  // unmutate.
    "vmovdqu    %%ymm0," MEMACCESS(1) "        \n"
    "lea       " MEMLEA(0x20,1) ",%1           \n"
    "sub        $0x20, %2                      \n"
    "jg         1b                             \n"
    "vzeroupper                                \n"
  : "+r"(src_argb),  // %0
    "+r"(dst_a),     // %1
    "+rm"(width)     // %2
  : "m"(kPermdARGBToY_AVX),  // %3
    "m"(kShuffleAlphaShort_AVX2)  // %4
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
  );
}
#endif  // HAS_ARGBEXTRACTALPHAROW_AVX2

2917 2918 2919 2920 2921 2922 2923 2924
#ifdef HAS_ARGBCOPYYTOALPHAROW_SSE2
// width in pixels
void ARGBCopyYToAlphaRow_SSE2(const uint8* src, uint8* dst, int width) {
  asm volatile (
    "pcmpeqb   %%xmm0,%%xmm0                   \n"
    "pslld     $0x18,%%xmm0                    \n"
    "pcmpeqb   %%xmm1,%%xmm1                   \n"
    "psrld     $0x8,%%xmm1                     \n"
2925
    LABELALIGN
2926 2927 2928 2929 2930 2931
  "1:                                          \n"
    "movq      " MEMACCESS(0) ",%%xmm2         \n"
    "lea       " MEMLEA(0x8,0) ",%0            \n"
    "punpcklbw %%xmm2,%%xmm2                   \n"
    "punpckhwd %%xmm2,%%xmm3                   \n"
    "punpcklwd %%xmm2,%%xmm2                   \n"
2932 2933
    "movdqu    " MEMACCESS(1) ",%%xmm4         \n"
    "movdqu    " MEMACCESS2(0x10,1) ",%%xmm5   \n"
2934 2935 2936 2937 2938 2939
    "pand      %%xmm0,%%xmm2                   \n"
    "pand      %%xmm0,%%xmm3                   \n"
    "pand      %%xmm1,%%xmm4                   \n"
    "pand      %%xmm1,%%xmm5                   \n"
    "por       %%xmm4,%%xmm2                   \n"
    "por       %%xmm5,%%xmm3                   \n"
2940 2941
    "movdqu    %%xmm2," MEMACCESS(1) "         \n"
    "movdqu    %%xmm3," MEMACCESS2(0x10,1) "   \n"
2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960
    "lea       " MEMLEA(0x20,1) ",%1           \n"
    "sub       $0x8,%2                         \n"
    "jg        1b                              \n"
  : "+r"(src),   // %0
    "+r"(dst),   // %1
    "+r"(width)  // %2
  :
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
  );
}
#endif  // HAS_ARGBCOPYYTOALPHAROW_SSE2

#ifdef HAS_ARGBCOPYYTOALPHAROW_AVX2
// width in pixels
void ARGBCopyYToAlphaRow_AVX2(const uint8* src, uint8* dst, int width) {
  asm volatile (
    "vpcmpeqb  %%ymm0,%%ymm0,%%ymm0            \n"
    "vpsrld    $0x8,%%ymm0,%%ymm0              \n"
2961
    LABELALIGN
2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985
  "1:                                          \n"
    "vpmovzxbd " MEMACCESS(0) ",%%ymm1         \n"
    "vpmovzxbd " MEMACCESS2(0x8,0) ",%%ymm2    \n"
    "lea       " MEMLEA(0x10,0) ",%0           \n"
    "vpslld    $0x18,%%ymm1,%%ymm1             \n"
    "vpslld    $0x18,%%ymm2,%%ymm2             \n"
    "vpblendvb %%ymm0," MEMACCESS(1) ",%%ymm1,%%ymm1        \n"
    "vpblendvb %%ymm0," MEMACCESS2(0x20,1) ",%%ymm2,%%ymm2  \n"
    "vmovdqu   %%ymm1," MEMACCESS(1) "         \n"
    "vmovdqu   %%ymm2," MEMACCESS2(0x20,1) "   \n"
    "lea       " MEMLEA(0x40,1) ",%1           \n"
    "sub       $0x10,%2                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src),   // %0
    "+r"(dst),   // %1
    "+r"(width)  // %2
  :
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2"
  );
}
#endif  // HAS_ARGBCOPYYTOALPHAROW_AVX2

2986
#ifdef HAS_SETROW_X86
2987 2988
void SetRow_X86(uint8* dst, uint8 v8, int width) {
  size_t width_tmp = (size_t)(width >> 2);
2989
  const uint32 v32 = v8 * 0x01010101u;  // Duplicate byte to all bytes.
2990 2991 2992 2993 2994
  asm volatile("rep stosl " MEMSTORESTRING(eax, 0) "       \n"
               : "+D"(dst),       // %0
                 "+c"(width_tmp)  // %1
               : "a"(v32)         // %2
               : "memory", "cc");
2995 2996
}

2997 2998
void SetRow_ERMS(uint8* dst, uint8 v8, int width) {
  size_t width_tmp = (size_t)(width);
2999 3000 3001 3002 3003
  asm volatile("rep stosb " MEMSTORESTRING(al, 0) "        \n"
               : "+D"(dst),       // %0
                 "+c"(width_tmp)  // %1
               : "a"(v8)          // %2
               : "memory", "cc");
3004 3005 3006 3007
}

void ARGBSetRow_X86(uint8* dst_argb, uint32 v32, int width) {
  size_t width_tmp = (size_t)(width);
3008 3009 3010 3011 3012
  asm volatile("rep stosl " MEMSTORESTRING(eax, 0) "       \n"
               : "+D"(dst_argb),  // %0
                 "+c"(width_tmp)  // %1
               : "a"(v32)         // %2
               : "memory", "cc");
3013 3014 3015
}
#endif  // HAS_SETROW_X86

3016
#ifdef HAS_YUY2TOYROW_SSE2
3017
void YUY2ToYRow_SSE2(const uint8* src_yuy2, uint8* dst_y, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
3018
  asm volatile (
3019 3020
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    "psrlw     $0x8,%%xmm5                     \n"
3021
    LABELALIGN
3022
  "1:                                          \n"
3023 3024
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
3025
    "lea       " MEMLEA(0x20,0) ",%0           \n"
3026 3027 3028
    "pand      %%xmm5,%%xmm0                   \n"
    "pand      %%xmm5,%%xmm1                   \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
3029
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
3030
    "lea       " MEMLEA(0x10,1) ",%1           \n"
3031
    "sub       $0x10,%2                        \n"
3032
    "jg        1b                              \n"
3033 3034
  : "+r"(src_yuy2),  // %0
    "+r"(dst_y),     // %1
3035
    "+r"(width)        // %2
3036 3037 3038 3039 3040 3041
  :
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm5"
  );
}

3042 3043 3044 3045 3046
void YUY2ToUVRow_SSE2(const uint8* src_yuy2,
                      int stride_yuy2,
                      uint8* dst_u,
                      uint8* dst_v,
                      int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
3047
  asm volatile (
3048 3049 3050
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    "psrlw     $0x8,%%xmm5                     \n"
    "sub       %1,%2                           \n"
3051
    LABELALIGN
3052
  "1:                                          \n"
3053 3054 3055 3056
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    MEMOPREG(movdqu,0x00,0,4,1,xmm2)           //  movdqu  (%0,%4,1),%%xmm2
    MEMOPREG(movdqu,0x10,0,4,1,xmm3)           //  movdqu  0x10(%0,%4,1),%%xmm3
3057
    "lea       " MEMLEA(0x20,0) ",%0           \n"
3058 3059 3060 3061 3062 3063 3064 3065 3066 3067
    "pavgb     %%xmm2,%%xmm0                   \n"
    "pavgb     %%xmm3,%%xmm1                   \n"
    "psrlw     $0x8,%%xmm0                     \n"
    "psrlw     $0x8,%%xmm1                     \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "pand      %%xmm5,%%xmm0                   \n"
    "packuswb  %%xmm0,%%xmm0                   \n"
    "psrlw     $0x8,%%xmm1                     \n"
    "packuswb  %%xmm1,%%xmm1                   \n"
3068
    "movq      %%xmm0," MEMACCESS(1) "         \n"
3069
    MEMOPMEM(movq,xmm1,0x00,1,2,1)             //  movq    %%xmm1,(%1,%2)
3070
    "lea       " MEMLEA(0x8,1) ",%1            \n"
3071
    "sub       $0x10,%3                        \n"
3072
    "jg        1b                              \n"
3073 3074
  : "+r"(src_yuy2),    // %0
    "+r"(dst_u),       // %1
3075
    "+r"(dst_v),       // %2
3076
    "+r"(width)          // %3
3077
  : "r"((intptr_t)(stride_yuy2))  // %4
3078 3079
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
3080 3081 3082
  );
}

3083
void YUY2ToUV422Row_SSE2(const uint8* src_yuy2,
3084 3085 3086
                         uint8* dst_u,
                         uint8* dst_v,
                         int width) {
3087 3088 3089 3090
  asm volatile (
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    "psrlw     $0x8,%%xmm5                     \n"
    "sub       %1,%2                           \n"
3091
    LABELALIGN
3092
  "1:                                          \n"
3093 3094
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
3095
    "lea       " MEMLEA(0x20,0) ",%0           \n"
3096 3097 3098 3099 3100 3101 3102 3103
    "psrlw     $0x8,%%xmm0                     \n"
    "psrlw     $0x8,%%xmm1                     \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "pand      %%xmm5,%%xmm0                   \n"
    "packuswb  %%xmm0,%%xmm0                   \n"
    "psrlw     $0x8,%%xmm1                     \n"
    "packuswb  %%xmm1,%%xmm1                   \n"
3104
    "movq      %%xmm0," MEMACCESS(1) "         \n"
3105
    MEMOPMEM(movq,xmm1,0x00,1,2,1)             //  movq    %%xmm1,(%1,%2)
3106
    "lea       " MEMLEA(0x8,1) ",%1            \n"
3107 3108
    "sub       $0x10,%3                        \n"
    "jg        1b                              \n"
3109
  : "+r"(src_yuy2),    // %0
3110 3111
    "+r"(dst_u),       // %1
    "+r"(dst_v),       // %2
3112
    "+r"(width)          // %3
3113
  :
3114 3115
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm5"
3116 3117 3118
  );
}

3119
void UYVYToYRow_SSE2(const uint8* src_uyvy, uint8* dst_y, int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
3120
  asm volatile (
3121
    LABELALIGN
3122
  "1:                                          \n"
3123 3124 3125
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "lea       " MEMLEA(0x20,0) ",%0           \n"
3126 3127 3128
    "psrlw     $0x8,%%xmm0                     \n"
    "psrlw     $0x8,%%xmm1                     \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
3129 3130
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
3131
    "sub       $0x10,%2                        \n"
3132
    "jg        1b                              \n"
3133 3134
  : "+r"(src_uyvy),  // %0
    "+r"(dst_y),     // %1
3135
    "+r"(width)        // %2
3136 3137 3138 3139 3140 3141
  :
  : "memory", "cc"
    , "xmm0", "xmm1"
  );
}

3142 3143 3144 3145 3146
void UYVYToUVRow_SSE2(const uint8* src_uyvy,
                      int stride_uyvy,
                      uint8* dst_u,
                      uint8* dst_v,
                      int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
3147
  asm volatile (
3148 3149 3150
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    "psrlw     $0x8,%%xmm5                     \n"
    "sub       %1,%2                           \n"
3151
    LABELALIGN
3152
  "1:                                          \n"
3153 3154
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
3155 3156
    MEMOPREG(movdqu,0x00,0,4,1,xmm2)           //  movdqu  (%0,%4,1),%%xmm2
    MEMOPREG(movdqu,0x10,0,4,1,xmm3)           //  movdqu  0x10(%0,%4,1),%%xmm3
3157
    "lea       " MEMLEA(0x20,0) ",%0           \n"
3158 3159 3160 3161 3162 3163 3164 3165 3166 3167
    "pavgb     %%xmm2,%%xmm0                   \n"
    "pavgb     %%xmm3,%%xmm1                   \n"
    "pand      %%xmm5,%%xmm0                   \n"
    "pand      %%xmm5,%%xmm1                   \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "pand      %%xmm5,%%xmm0                   \n"
    "packuswb  %%xmm0,%%xmm0                   \n"
    "psrlw     $0x8,%%xmm1                     \n"
    "packuswb  %%xmm1,%%xmm1                   \n"
3168
    "movq      %%xmm0," MEMACCESS(1) "         \n"
3169
    MEMOPMEM(movq,xmm1,0x00,1,2,1)             //  movq    %%xmm1,(%1,%2)
3170
    "lea       " MEMLEA(0x8,1) ",%1            \n"
3171
    "sub       $0x10,%3                        \n"
3172
    "jg        1b                              \n"
3173 3174
  : "+r"(src_uyvy),    // %0
    "+r"(dst_u),       // %1
3175
    "+r"(dst_v),       // %2
3176
    "+r"(width)          // %3
3177
  : "r"((intptr_t)(stride_uyvy))  // %4
3178 3179
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
3180 3181
  );
}
3182

3183
void UYVYToUV422Row_SSE2(const uint8* src_uyvy,
3184 3185 3186
                         uint8* dst_u,
                         uint8* dst_v,
                         int width) {
3187 3188 3189 3190
  asm volatile (
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    "psrlw     $0x8,%%xmm5                     \n"
    "sub       %1,%2                           \n"
3191
    LABELALIGN
3192
  "1:                                          \n"
3193 3194 3195
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "lea       " MEMLEA(0x20,0) ",%0           \n"
3196 3197 3198 3199 3200 3201 3202 3203
    "pand      %%xmm5,%%xmm0                   \n"
    "pand      %%xmm5,%%xmm1                   \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "pand      %%xmm5,%%xmm0                   \n"
    "packuswb  %%xmm0,%%xmm0                   \n"
    "psrlw     $0x8,%%xmm1                     \n"
    "packuswb  %%xmm1,%%xmm1                   \n"
3204
    "movq      %%xmm0," MEMACCESS(1) "         \n"
3205
    MEMOPMEM(movq,xmm1,0x00,1,2,1)             //  movq    %%xmm1,(%1,%2)
3206
    "lea       " MEMLEA(0x8,1) ",%1            \n"
3207 3208 3209 3210 3211
    "sub       $0x10,%3                        \n"
    "jg        1b                              \n"
  : "+r"(src_uyvy),    // %0
    "+r"(dst_u),       // %1
    "+r"(dst_v),       // %2
3212
    "+r"(width)          // %3
3213
  :
3214 3215
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm5"
3216 3217
  );
}
3218
#endif  // HAS_YUY2TOYROW_SSE2
3219

3220
#ifdef HAS_YUY2TOYROW_AVX2
3221
void YUY2ToYRow_AVX2(const uint8* src_yuy2, uint8* dst_y, int width) {
3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235
  asm volatile (
    "vpcmpeqb  %%ymm5,%%ymm5,%%ymm5            \n"
    "vpsrlw    $0x8,%%ymm5,%%ymm5              \n"
    LABELALIGN
  "1:                                          \n"
    "vmovdqu   " MEMACCESS(0) ",%%ymm0         \n"
    "vmovdqu   " MEMACCESS2(0x20,0) ",%%ymm1   \n"
    "lea       " MEMLEA(0x40,0) ",%0           \n"
    "vpand     %%ymm5,%%ymm0,%%ymm0            \n"
    "vpand     %%ymm5,%%ymm1,%%ymm1            \n"
    "vpackuswb %%ymm1,%%ymm0,%%ymm0            \n"
    "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
    "vmovdqu   %%ymm0," MEMACCESS(1) "         \n"
    "lea      " MEMLEA(0x20,1) ",%1            \n"
3236
    "sub       $0x20,%2                        \n"
3237
    "jg        1b                              \n"
3238
    "vzeroupper                                \n"
3239 3240
  : "+r"(src_yuy2),  // %0
    "+r"(dst_y),     // %1
3241
    "+r"(width)        // %2
3242 3243 3244 3245 3246 3247
  :
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm5"
  );
}

3248 3249 3250 3251 3252
void YUY2ToUVRow_AVX2(const uint8* src_yuy2,
                      int stride_yuy2,
                      uint8* dst_u,
                      uint8* dst_v,
                      int width) {
3253 3254 3255 3256 3257 3258 3259 3260 3261 3262
  asm volatile (
    "vpcmpeqb  %%ymm5,%%ymm5,%%ymm5            \n"
    "vpsrlw    $0x8,%%ymm5,%%ymm5              \n"
    "sub       %1,%2                           \n"
    LABELALIGN
  "1:                                          \n"
    "vmovdqu   " MEMACCESS(0) ",%%ymm0         \n"
    "vmovdqu   " MEMACCESS2(0x20,0) ",%%ymm1   \n"
    VMEMOPREG(vpavgb,0x00,0,4,1,ymm0,ymm0)     // vpavgb (%0,%4,1),%%ymm0,%%ymm0
    VMEMOPREG(vpavgb,0x20,0,4,1,ymm1,ymm1)
3263
    "lea       " MEMLEA(0x40,0) ",%0           \n"
3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274
    "vpsrlw    $0x8,%%ymm0,%%ymm0              \n"
    "vpsrlw    $0x8,%%ymm1,%%ymm1              \n"
    "vpackuswb %%ymm1,%%ymm0,%%ymm0            \n"
    "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
    "vpand     %%ymm5,%%ymm0,%%ymm1            \n"
    "vpsrlw    $0x8,%%ymm0,%%ymm0              \n"
    "vpackuswb %%ymm1,%%ymm1,%%ymm1            \n"
    "vpackuswb %%ymm0,%%ymm0,%%ymm0            \n"
    "vpermq    $0xd8,%%ymm1,%%ymm1             \n"
    "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
    "vextractf128 $0x0,%%ymm1," MEMACCESS(1) " \n"
3275
    VEXTOPMEM(vextractf128,0,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
3276
    "lea      " MEMLEA(0x10,1) ",%1            \n"
3277 3278 3279 3280 3281 3282
    "sub       $0x20,%3                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_yuy2),    // %0
    "+r"(dst_u),       // %1
    "+r"(dst_v),       // %2
3283
    "+r"(width)          // %3
3284
  : "r"((intptr_t)(stride_yuy2))  // %4
3285 3286
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm5"
3287 3288 3289 3290
  );
}

void YUY2ToUV422Row_AVX2(const uint8* src_yuy2,
3291 3292 3293
                         uint8* dst_u,
                         uint8* dst_v,
                         int width) {
3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313
  asm volatile (
    "vpcmpeqb  %%ymm5,%%ymm5,%%ymm5            \n"
    "vpsrlw    $0x8,%%ymm5,%%ymm5              \n"
    "sub       %1,%2                           \n"
    LABELALIGN
  "1:                                          \n"
    "vmovdqu   " MEMACCESS(0) ",%%ymm0         \n"
    "vmovdqu   " MEMACCESS2(0x20,0) ",%%ymm1   \n"
    "lea       " MEMLEA(0x40,0) ",%0           \n"
    "vpsrlw    $0x8,%%ymm0,%%ymm0              \n"
    "vpsrlw    $0x8,%%ymm1,%%ymm1              \n"
    "vpackuswb %%ymm1,%%ymm0,%%ymm0            \n"
    "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
    "vpand     %%ymm5,%%ymm0,%%ymm1            \n"
    "vpsrlw    $0x8,%%ymm0,%%ymm0              \n"
    "vpackuswb %%ymm1,%%ymm1,%%ymm1            \n"
    "vpackuswb %%ymm0,%%ymm0,%%ymm0            \n"
    "vpermq    $0xd8,%%ymm1,%%ymm1             \n"
    "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
    "vextractf128 $0x0,%%ymm1," MEMACCESS(1) " \n"
3314
    VEXTOPMEM(vextractf128,0,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
3315
    "lea      " MEMLEA(0x10,1) ",%1            \n"
3316 3317 3318 3319 3320 3321
    "sub       $0x20,%3                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_yuy2),    // %0
    "+r"(dst_u),       // %1
    "+r"(dst_v),       // %2
3322
    "+r"(width)          // %3
3323
  :
3324 3325
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm5"
3326 3327 3328
  );
}

3329
void UYVYToYRow_AVX2(const uint8* src_uyvy, uint8* dst_y, int width) {
3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341
  asm volatile (
    LABELALIGN
  "1:                                          \n"
    "vmovdqu   " MEMACCESS(0) ",%%ymm0         \n"
    "vmovdqu   " MEMACCESS2(0x20,0) ",%%ymm1   \n"
    "lea       " MEMLEA(0x40,0) ",%0           \n"
    "vpsrlw    $0x8,%%ymm0,%%ymm0              \n"
    "vpsrlw    $0x8,%%ymm1,%%ymm1              \n"
    "vpackuswb %%ymm1,%%ymm0,%%ymm0            \n"
    "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
    "vmovdqu   %%ymm0," MEMACCESS(1) "         \n"
    "lea      " MEMLEA(0x20,1) ",%1            \n"
3342
    "sub       $0x20,%2                        \n"
3343 3344 3345 3346
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_uyvy),  // %0
    "+r"(dst_y),     // %1
3347
    "+r"(width)        // %2
3348 3349 3350 3351 3352
  :
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm5"
  );
}
3353 3354 3355 3356 3357
void UYVYToUVRow_AVX2(const uint8* src_uyvy,
                      int stride_uyvy,
                      uint8* dst_u,
                      uint8* dst_v,
                      int width) {
3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368
  asm volatile (
    "vpcmpeqb  %%ymm5,%%ymm5,%%ymm5            \n"
    "vpsrlw    $0x8,%%ymm5,%%ymm5              \n"
    "sub       %1,%2                           \n"

    LABELALIGN
  "1:                                          \n"
    "vmovdqu   " MEMACCESS(0) ",%%ymm0         \n"
    "vmovdqu   " MEMACCESS2(0x20,0) ",%%ymm1   \n"
    VMEMOPREG(vpavgb,0x00,0,4,1,ymm0,ymm0)     // vpavgb (%0,%4,1),%%ymm0,%%ymm0
    VMEMOPREG(vpavgb,0x20,0,4,1,ymm1,ymm1)
3369
    "lea       " MEMLEA(0x40,0) ",%0           \n"
3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380
    "vpand     %%ymm5,%%ymm0,%%ymm0            \n"
    "vpand     %%ymm5,%%ymm1,%%ymm1            \n"
    "vpackuswb %%ymm1,%%ymm0,%%ymm0            \n"
    "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
    "vpand     %%ymm5,%%ymm0,%%ymm1            \n"
    "vpsrlw    $0x8,%%ymm0,%%ymm0              \n"
    "vpackuswb %%ymm1,%%ymm1,%%ymm1            \n"
    "vpackuswb %%ymm0,%%ymm0,%%ymm0            \n"
    "vpermq    $0xd8,%%ymm1,%%ymm1             \n"
    "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
    "vextractf128 $0x0,%%ymm1," MEMACCESS(1) " \n"
3381
    VEXTOPMEM(vextractf128,0,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
3382
    "lea      " MEMLEA(0x10,1) ",%1            \n"
3383 3384 3385 3386 3387 3388
    "sub       $0x20,%3                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_uyvy),    // %0
    "+r"(dst_u),       // %1
    "+r"(dst_v),       // %2
3389
    "+r"(width)          // %3
3390
  : "r"((intptr_t)(stride_uyvy))  // %4
3391 3392
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm5"
3393 3394 3395 3396
  );
}

void UYVYToUV422Row_AVX2(const uint8* src_uyvy,
3397 3398 3399
                         uint8* dst_u,
                         uint8* dst_v,
                         int width) {
3400
  asm volatile (
3401 3402
    "vpcmpeqb   %%ymm5,%%ymm5,%%ymm5           \n"
    "vpsrlw     $0x8,%%ymm5,%%ymm5             \n"
3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419
    "sub       %1,%2                           \n"
    LABELALIGN
  "1:                                          \n"
    "vmovdqu   " MEMACCESS(0) ",%%ymm0         \n"
    "vmovdqu   " MEMACCESS2(0x20,0) ",%%ymm1   \n"
    "lea       " MEMLEA(0x40,0) ",%0           \n"
    "vpand     %%ymm5,%%ymm0,%%ymm0            \n"
    "vpand     %%ymm5,%%ymm1,%%ymm1            \n"
    "vpackuswb %%ymm1,%%ymm0,%%ymm0            \n"
    "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
    "vpand     %%ymm5,%%ymm0,%%ymm1            \n"
    "vpsrlw    $0x8,%%ymm0,%%ymm0              \n"
    "vpackuswb %%ymm1,%%ymm1,%%ymm1            \n"
    "vpackuswb %%ymm0,%%ymm0,%%ymm0            \n"
    "vpermq    $0xd8,%%ymm1,%%ymm1             \n"
    "vpermq    $0xd8,%%ymm0,%%ymm0             \n"
    "vextractf128 $0x0,%%ymm1," MEMACCESS(1) " \n"
3420
    VEXTOPMEM(vextractf128,0,ymm0,0x00,1,2,1) // vextractf128 $0x0,%%ymm0,(%1,%2,1)
3421
    "lea      " MEMLEA(0x10,1) ",%1            \n"
3422 3423 3424 3425 3426 3427
    "sub       $0x20,%3                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_uyvy),    // %0
    "+r"(dst_u),       // %1
    "+r"(dst_v),       // %2
3428
    "+r"(width)          // %3
3429
  :
3430 3431
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm5"
3432 3433
  );
}
3434 3435
#endif  // HAS_YUY2TOYROW_AVX2

3436
#ifdef HAS_ARGBBLENDROW_SSSE3
3437
// Shuffle table for isolating alpha.
3438 3439
static uvec8 kShuffleAlpha = {3u,  0x80, 3u,  0x80, 7u,  0x80, 7u,  0x80,
                              11u, 0x80, 11u, 0x80, 15u, 0x80, 15u, 0x80};
3440 3441

// Blend 8 pixels at a time
3442 3443 3444 3445
void ARGBBlendRow_SSSE3(const uint8* src_argb0,
                        const uint8* src_argb1,
                        uint8* dst_argb,
                        int width) {
3446 3447 3448 3449 3450 3451 3452 3453 3454
  asm volatile (
    "pcmpeqb   %%xmm7,%%xmm7                   \n"
    "psrlw     $0xf,%%xmm7                     \n"
    "pcmpeqb   %%xmm6,%%xmm6                   \n"
    "psrlw     $0x8,%%xmm6                     \n"
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    "psllw     $0x8,%%xmm5                     \n"
    "pcmpeqb   %%xmm4,%%xmm4                   \n"
    "pslld     $0x18,%%xmm4                    \n"
3455
    "sub       $0x4,%3                         \n"
3456 3457
    "jl        49f                             \n"

3458
    // 4 pixel loop.
3459
    LABELALIGN
3460
  "40:                                         \n"
3461
    "movdqu    " MEMACCESS(0) ",%%xmm3         \n"
3462
    "lea       " MEMLEA(0x10,0) ",%0           \n"
3463 3464
    "movdqa    %%xmm3,%%xmm0                   \n"
    "pxor      %%xmm4,%%xmm3                   \n"
3465
    "movdqu    " MEMACCESS(1) ",%%xmm2         \n"
3466 3467 3468 3469
    "pshufb    %4,%%xmm3                       \n"
    "pand      %%xmm6,%%xmm2                   \n"
    "paddw     %%xmm7,%%xmm3                   \n"
    "pmullw    %%xmm3,%%xmm2                   \n"
3470
    "movdqu    " MEMACCESS(1) ",%%xmm1         \n"
3471
    "lea       " MEMLEA(0x10,1) ",%1           \n"
3472 3473 3474 3475 3476 3477 3478
    "psrlw     $0x8,%%xmm1                     \n"
    "por       %%xmm4,%%xmm0                   \n"
    "pmullw    %%xmm3,%%xmm1                   \n"
    "psrlw     $0x8,%%xmm2                     \n"
    "paddusb   %%xmm2,%%xmm0                   \n"
    "pand      %%xmm5,%%xmm1                   \n"
    "paddusb   %%xmm1,%%xmm0                   \n"
3479
    "movdqu    %%xmm0," MEMACCESS(2) "         \n"
3480
    "lea       " MEMLEA(0x10,2) ",%2           \n"
3481
    "sub       $0x4,%3                         \n"
3482
    "jge       40b                             \n"
3483

3484 3485 3486
  "49:                                         \n"
    "add       $0x3,%3                         \n"
    "jl        99f                             \n"
3487

3488
    // 1 pixel loop.
3489
  "91:                                         \n"
3490 3491
    "movd      " MEMACCESS(0) ",%%xmm3         \n"
    "lea       " MEMLEA(0x4,0) ",%0            \n"
3492 3493
    "movdqa    %%xmm3,%%xmm0                   \n"
    "pxor      %%xmm4,%%xmm3                   \n"
3494
    "movd      " MEMACCESS(1) ",%%xmm2         \n"
3495 3496 3497 3498
    "pshufb    %4,%%xmm3                       \n"
    "pand      %%xmm6,%%xmm2                   \n"
    "paddw     %%xmm7,%%xmm3                   \n"
    "pmullw    %%xmm3,%%xmm2                   \n"
3499 3500
    "movd      " MEMACCESS(1) ",%%xmm1         \n"
    "lea       " MEMLEA(0x4,1) ",%1            \n"
3501 3502 3503 3504 3505 3506 3507
    "psrlw     $0x8,%%xmm1                     \n"
    "por       %%xmm4,%%xmm0                   \n"
    "pmullw    %%xmm3,%%xmm1                   \n"
    "psrlw     $0x8,%%xmm2                     \n"
    "paddusb   %%xmm2,%%xmm0                   \n"
    "pand      %%xmm5,%%xmm1                   \n"
    "paddusb   %%xmm1,%%xmm0                   \n"
3508 3509
    "movd      %%xmm0," MEMACCESS(2) "         \n"
    "lea       " MEMLEA(0x4,2) ",%2            \n"
3510
    "sub       $0x1,%3                         \n"
3511 3512
    "jge       91b                             \n"
  "99:                                         \n"
3513 3514 3515 3516 3517 3518 3519 3520 3521
  : "+r"(src_argb0),    // %0
    "+r"(src_argb1),    // %1
    "+r"(dst_argb),     // %2
    "+r"(width)         // %3
  : "m"(kShuffleAlpha)  // %4
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
  );
}
3522
#endif  // HAS_ARGBBLENDROW_SSSE3
3523

3524 3525
#ifdef HAS_BLENDPLANEROW_SSSE3
// Blend 8 pixels at a time.
3526 3527 3528 3529
// unsigned version of math
// =((A2*C2)+(B2*(255-C2))+255)/256
// signed version of math
// =(((A2-128)*C2)+((B2-128)*(255-C2))+32768+127)/256
3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572
void BlendPlaneRow_SSSE3(const uint8* src0,
                         const uint8* src1,
                         const uint8* alpha,
                         uint8* dst,
                         int width) {
  asm volatile(
      "pcmpeqb    %%xmm5,%%xmm5                  \n"
      "psllw      $0x8,%%xmm5                    \n"
      "mov        $0x80808080,%%eax              \n"
      "movd       %%eax,%%xmm6                   \n"
      "pshufd     $0x0,%%xmm6,%%xmm6             \n"
      "mov        $0x807f807f,%%eax              \n"
      "movd       %%eax,%%xmm7                   \n"
      "pshufd     $0x0,%%xmm7,%%xmm7             \n"
      "sub        %2,%0                          \n"
      "sub        %2,%1                          \n"
      "sub        %2,%3                          \n"

      // 8 pixel loop.
      LABELALIGN
      "1:                                          \n"
      "movq       (%2),%%xmm0                    \n"
      "punpcklbw  %%xmm0,%%xmm0                  \n"
      "pxor       %%xmm5,%%xmm0                  \n"
      "movq       (%0,%2,1),%%xmm1               \n"
      "movq       (%1,%2,1),%%xmm2               \n"
      "punpcklbw  %%xmm2,%%xmm1                  \n"
      "psubb      %%xmm6,%%xmm1                  \n"
      "pmaddubsw  %%xmm1,%%xmm0                  \n"
      "paddw      %%xmm7,%%xmm0                  \n"
      "psrlw      $0x8,%%xmm0                    \n"
      "packuswb   %%xmm0,%%xmm0                  \n"
      "movq       %%xmm0,(%3,%2,1)               \n"
      "lea        0x8(%2),%2                     \n"
      "sub        $0x8,%4                        \n"
      "jg        1b                              \n"
      : "+r"(src0),   // %0
        "+r"(src1),   // %1
        "+r"(alpha),  // %2
        "+r"(dst),    // %3
        "+rm"(width)  // %4
        ::"memory",
        "cc", "eax", "xmm0", "xmm1", "xmm2", "xmm5", "xmm6", "xmm7");
3573 3574 3575
}
#endif  // HAS_BLENDPLANEROW_SSSE3

3576
#ifdef HAS_BLENDPLANEROW_AVX2
3577 3578 3579 3580 3581
// Blend 32 pixels at a time.
// unsigned version of math
// =((A2*C2)+(B2*(255-C2))+255)/256
// signed version of math
// =(((A2-128)*C2)+((B2-128)*(255-C2))+32768+127)/256
3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633
void BlendPlaneRow_AVX2(const uint8* src0,
                        const uint8* src1,
                        const uint8* alpha,
                        uint8* dst,
                        int width) {
  asm volatile(
      "vpcmpeqb   %%ymm5,%%ymm5,%%ymm5           \n"
      "vpsllw     $0x8,%%ymm5,%%ymm5             \n"
      "mov        $0x80808080,%%eax              \n"
      "vmovd      %%eax,%%xmm6                   \n"
      "vbroadcastss %%xmm6,%%ymm6                \n"
      "mov        $0x807f807f,%%eax              \n"
      "vmovd      %%eax,%%xmm7                   \n"
      "vbroadcastss %%xmm7,%%ymm7                \n"
      "sub        %2,%0                          \n"
      "sub        %2,%1                          \n"
      "sub        %2,%3                          \n"

      // 32 pixel loop.
      LABELALIGN
      "1:                                          \n"
      "vmovdqu    (%2),%%ymm0                    \n"
      "vpunpckhbw %%ymm0,%%ymm0,%%ymm3           \n"
      "vpunpcklbw %%ymm0,%%ymm0,%%ymm0           \n"
      "vpxor      %%ymm5,%%ymm3,%%ymm3           \n"
      "vpxor      %%ymm5,%%ymm0,%%ymm0           \n"
      "vmovdqu    (%0,%2,1),%%ymm1               \n"
      "vmovdqu    (%1,%2,1),%%ymm2               \n"
      "vpunpckhbw %%ymm2,%%ymm1,%%ymm4           \n"
      "vpunpcklbw %%ymm2,%%ymm1,%%ymm1           \n"
      "vpsubb     %%ymm6,%%ymm4,%%ymm4           \n"
      "vpsubb     %%ymm6,%%ymm1,%%ymm1           \n"
      "vpmaddubsw %%ymm4,%%ymm3,%%ymm3           \n"
      "vpmaddubsw %%ymm1,%%ymm0,%%ymm0           \n"
      "vpaddw     %%ymm7,%%ymm3,%%ymm3           \n"
      "vpaddw     %%ymm7,%%ymm0,%%ymm0           \n"
      "vpsrlw     $0x8,%%ymm3,%%ymm3             \n"
      "vpsrlw     $0x8,%%ymm0,%%ymm0             \n"
      "vpackuswb  %%ymm3,%%ymm0,%%ymm0           \n"
      "vmovdqu    %%ymm0,(%3,%2,1)               \n"
      "lea        0x20(%2),%2                    \n"
      "sub        $0x20,%4                       \n"
      "jg        1b                              \n"
      "vzeroupper                                \n"
      : "+r"(src0),   // %0
        "+r"(src1),   // %1
        "+r"(alpha),  // %2
        "+r"(dst),    // %3
        "+rm"(width)  // %4
        ::"memory",
        "cc", "eax", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
        "xmm7");
3634 3635
}
#endif  // HAS_BLENDPLANEROW_AVX2
3636

3637
#ifdef HAS_ARGBATTENUATEROW_SSSE3
3638
// Shuffle table duplicating alpha
3639 3640 3641 3642
static uvec8 kShuffleAlpha0 = {3u, 3u, 3u, 3u, 3u, 3u, 128u, 128u,
                               7u, 7u, 7u, 7u, 7u, 7u, 128u, 128u};
static uvec8 kShuffleAlpha1 = {11u, 11u, 11u, 11u, 11u, 11u, 128u, 128u,
                               15u, 15u, 15u, 15u, 15u, 15u, 128u, 128u};
3643 3644 3645 3646 3647 3648 3649 3650
// Attenuate 4 pixels at a time.
void ARGBAttenuateRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width) {
  asm volatile (
    "pcmpeqb   %%xmm3,%%xmm3                   \n"
    "pslld     $0x18,%%xmm3                    \n"
    "movdqa    %3,%%xmm4                       \n"
    "movdqa    %4,%%xmm5                       \n"

3651
    // 4 pixel loop.
3652
    LABELALIGN
3653
  "1:                                          \n"
3654
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
3655
    "pshufb    %%xmm4,%%xmm0                   \n"
3656
    "movdqu    " MEMACCESS(0) ",%%xmm1         \n"
3657 3658
    "punpcklbw %%xmm1,%%xmm1                   \n"
    "pmulhuw   %%xmm1,%%xmm0                   \n"
3659
    "movdqu    " MEMACCESS(0) ",%%xmm1         \n"
3660
    "pshufb    %%xmm5,%%xmm1                   \n"
3661
    "movdqu    " MEMACCESS(0) ",%%xmm2         \n"
3662 3663
    "punpckhbw %%xmm2,%%xmm2                   \n"
    "pmulhuw   %%xmm2,%%xmm1                   \n"
3664 3665
    "movdqu    " MEMACCESS(0) ",%%xmm2         \n"
    "lea       " MEMLEA(0x10,0) ",%0           \n"
3666 3667 3668 3669 3670
    "pand      %%xmm3,%%xmm2                   \n"
    "psrlw     $0x8,%%xmm0                     \n"
    "psrlw     $0x8,%%xmm1                     \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
    "por       %%xmm2,%%xmm0                   \n"
3671 3672
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
3673
    "sub       $0x4,%2                         \n"
3674 3675 3676 3677 3678 3679 3680 3681 3682 3683
    "jg        1b                              \n"
  : "+r"(src_argb),    // %0
    "+r"(dst_argb),    // %1
    "+r"(width)        // %2
  : "m"(kShuffleAlpha0),  // %3
    "m"(kShuffleAlpha1)  // %4
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
  );
}
3684
#endif  // HAS_ARGBATTENUATEROW_SSSE3
3685

3686 3687
#ifdef HAS_ARGBATTENUATEROW_AVX2
// Shuffle table duplicating alpha.
3688 3689 3690
static const uvec8 kShuffleAlpha_AVX2 = {6u,   7u,   6u,   7u,  6u,  7u,
                                         128u, 128u, 14u,  15u, 14u, 15u,
                                         14u,  15u,  128u, 128u};
3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715
// Attenuate 8 pixels at a time.
void ARGBAttenuateRow_AVX2(const uint8* src_argb, uint8* dst_argb, int width) {
  asm volatile (
    "vbroadcastf128 %3,%%ymm4                  \n"
    "vpcmpeqb   %%ymm5,%%ymm5,%%ymm5           \n"
    "vpslld     $0x18,%%ymm5,%%ymm5            \n"
    "sub        %0,%1                          \n"

    // 8 pixel loop.
    LABELALIGN
  "1:                                          \n"
    "vmovdqu    " MEMACCESS(0) ",%%ymm6        \n"
    "vpunpcklbw %%ymm6,%%ymm6,%%ymm0           \n"
    "vpunpckhbw %%ymm6,%%ymm6,%%ymm1           \n"
    "vpshufb    %%ymm4,%%ymm0,%%ymm2           \n"
    "vpshufb    %%ymm4,%%ymm1,%%ymm3           \n"
    "vpmulhuw   %%ymm2,%%ymm0,%%ymm0           \n"
    "vpmulhuw   %%ymm3,%%ymm1,%%ymm1           \n"
    "vpand      %%ymm5,%%ymm6,%%ymm6           \n"
    "vpsrlw     $0x8,%%ymm0,%%ymm0             \n"
    "vpsrlw     $0x8,%%ymm1,%%ymm1             \n"
    "vpackuswb  %%ymm1,%%ymm0,%%ymm0           \n"
    "vpor       %%ymm6,%%ymm0,%%ymm0           \n"
    MEMOPMEM(vmovdqu,ymm0,0x00,0,1,1)          //  vmovdqu %%ymm0,(%0,%1)
    "lea       " MEMLEA(0x20,0) ",%0           \n"
3716
    "sub        $0x8,%2                        \n"
3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_argb),    // %0
    "+r"(dst_argb),    // %1
    "+r"(width)        // %2
  : "m"(kShuffleAlpha_AVX2)  // %3
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
  );
}
#endif  // HAS_ARGBATTENUATEROW_AVX2

3729
#ifdef HAS_ARGBUNATTENUATEROW_SSE2
3730
// Unattenuate 4 pixels at a time.
3731 3732
void ARGBUnattenuateRow_SSE2(const uint8* src_argb,
                             uint8* dst_argb,
3733
                             int width) {
3734
  uintptr_t alpha;
3735
  asm volatile (
3736
    // 4 pixel loop.
3737
    LABELALIGN
3738
  "1:                                          \n"
3739 3740
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movzb     " MEMACCESS2(0x03,0) ",%3       \n"
3741
    "punpcklbw %%xmm0,%%xmm0                   \n"
3742
    MEMOPREG(movd,0x00,4,3,4,xmm2)             //  movd      0x0(%4,%3,4),%%xmm2
3743
    "movzb     " MEMACCESS2(0x07,0) ",%3       \n"
3744
    MEMOPREG(movd,0x00,4,3,4,xmm3)             //  movd      0x0(%4,%3,4),%%xmm3
3745 3746
    "pshuflw   $0x40,%%xmm2,%%xmm2             \n"
    "pshuflw   $0x40,%%xmm3,%%xmm3             \n"
3747 3748
    "movlhps   %%xmm3,%%xmm2                   \n"
    "pmulhuw   %%xmm2,%%xmm0                   \n"
3749 3750
    "movdqu    " MEMACCESS(0) ",%%xmm1         \n"
    "movzb     " MEMACCESS2(0x0b,0) ",%3       \n"
3751
    "punpckhbw %%xmm1,%%xmm1                   \n"
3752
    MEMOPREG(movd,0x00,4,3,4,xmm2)             //  movd      0x0(%4,%3,4),%%xmm2
3753
    "movzb     " MEMACCESS2(0x0f,0) ",%3       \n"
3754
    MEMOPREG(movd,0x00,4,3,4,xmm3)             //  movd      0x0(%4,%3,4),%%xmm3
3755 3756
    "pshuflw   $0x40,%%xmm2,%%xmm2             \n"
    "pshuflw   $0x40,%%xmm3,%%xmm3             \n"
3757 3758
    "movlhps   %%xmm3,%%xmm2                   \n"
    "pmulhuw   %%xmm2,%%xmm1                   \n"
3759
    "lea       " MEMLEA(0x10,0) ",%0           \n"
3760
    "packuswb  %%xmm1,%%xmm0                   \n"
3761 3762
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
3763
    "sub       $0x4,%2                         \n"
3764
    "jg        1b                              \n"
3765 3766 3767 3768
  : "+r"(src_argb),     // %0
    "+r"(dst_argb),     // %1
    "+r"(width),        // %2
    "=&r"(alpha)        // %3
3769
  : "r"(fixed_invtbl8)  // %4
3770 3771
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
3772 3773
  );
}
3774
#endif  // HAS_ARGBUNATTENUATEROW_SSE2
3775

3776 3777 3778
#ifdef HAS_ARGBUNATTENUATEROW_AVX2
// Shuffle table duplicating alpha.
static const uvec8 kUnattenShuffleAlpha_AVX2 = {
3779
    0u, 1u, 0u, 1u, 0u, 1u, 6u, 7u, 8u, 9u, 8u, 9u, 8u, 9u, 14u, 15u};
3780
// Unattenuate 8 pixels at a time.
3781 3782
void ARGBUnattenuateRow_AVX2(const uint8* src_argb,
                             uint8* dst_argb,
3783
                             int width) {
3784
  uintptr_t alpha;
3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829
  asm volatile (
    "sub        %0,%1                          \n"
    "vbroadcastf128 %5,%%ymm5                  \n"

    // 8 pixel loop.
    LABELALIGN
  "1:                                          \n"
    // replace VPGATHER
    "movzb     " MEMACCESS2(0x03,0) ",%3       \n"
    MEMOPREG(vmovd,0x00,4,3,4,xmm0)             //  vmovd 0x0(%4,%3,4),%%xmm0
    "movzb     " MEMACCESS2(0x07,0) ",%3       \n"
    MEMOPREG(vmovd,0x00,4,3,4,xmm1)             //  vmovd 0x0(%4,%3,4),%%xmm1
    "movzb     " MEMACCESS2(0x0b,0) ",%3       \n"
    "vpunpckldq %%xmm1,%%xmm0,%%xmm6           \n"
    MEMOPREG(vmovd,0x00,4,3,4,xmm2)             //  vmovd 0x0(%4,%3,4),%%xmm2
    "movzb     " MEMACCESS2(0x0f,0) ",%3       \n"
    MEMOPREG(vmovd,0x00,4,3,4,xmm3)             //  vmovd 0x0(%4,%3,4),%%xmm3
    "movzb     " MEMACCESS2(0x13,0) ",%3       \n"
    "vpunpckldq %%xmm3,%%xmm2,%%xmm7           \n"
    MEMOPREG(vmovd,0x00,4,3,4,xmm0)             //  vmovd 0x0(%4,%3,4),%%xmm0
    "movzb     " MEMACCESS2(0x17,0) ",%3       \n"
    MEMOPREG(vmovd,0x00,4,3,4,xmm1)             //  vmovd 0x0(%4,%3,4),%%xmm1
    "movzb     " MEMACCESS2(0x1b,0) ",%3       \n"
    "vpunpckldq %%xmm1,%%xmm0,%%xmm0           \n"
    MEMOPREG(vmovd,0x00,4,3,4,xmm2)             //  vmovd 0x0(%4,%3,4),%%xmm2
    "movzb     " MEMACCESS2(0x1f,0) ",%3       \n"
    MEMOPREG(vmovd,0x00,4,3,4,xmm3)             //  vmovd 0x0(%4,%3,4),%%xmm3
    "vpunpckldq %%xmm3,%%xmm2,%%xmm2           \n"
    "vpunpcklqdq %%xmm7,%%xmm6,%%xmm3          \n"
    "vpunpcklqdq %%xmm2,%%xmm0,%%xmm0          \n"
    "vinserti128 $0x1,%%xmm0,%%ymm3,%%ymm3     \n"
    // end of VPGATHER

    "vmovdqu    " MEMACCESS(0) ",%%ymm6        \n"
    "vpunpcklbw %%ymm6,%%ymm6,%%ymm0           \n"
    "vpunpckhbw %%ymm6,%%ymm6,%%ymm1           \n"
    "vpunpcklwd %%ymm3,%%ymm3,%%ymm2           \n"
    "vpunpckhwd %%ymm3,%%ymm3,%%ymm3           \n"
    "vpshufb    %%ymm5,%%ymm2,%%ymm2           \n"
    "vpshufb    %%ymm5,%%ymm3,%%ymm3           \n"
    "vpmulhuw   %%ymm2,%%ymm0,%%ymm0           \n"
    "vpmulhuw   %%ymm3,%%ymm1,%%ymm1           \n"
    "vpackuswb  %%ymm1,%%ymm0,%%ymm0           \n"
    MEMOPMEM(vmovdqu,ymm0,0x00,0,1,1)          //  vmovdqu %%ymm0,(%0,%1)
    "lea       " MEMLEA(0x20,0) ",%0           \n"
3830
    "sub        $0x8,%2                        \n"
3831 3832
    "jg        1b                              \n"
    "vzeroupper                                \n"
3833 3834 3835 3836
  : "+r"(src_argb),      // %0
    "+r"(dst_argb),      // %1
    "+r"(width),         // %2
    "=&r"(alpha)         // %3
3837 3838
  : "r"(fixed_invtbl8),  // %4
    "m"(kUnattenShuffleAlpha_AVX2)  // %5
3839 3840
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
3841 3842 3843 3844
  );
}
#endif  // HAS_ARGBUNATTENUATEROW_AVX2

3845 3846
#ifdef HAS_ARGBGRAYROW_SSSE3
// Convert 8 ARGB pixels (64 bytes) to 8 Gray ARGB pixels
3847
void ARGBGrayRow_SSSE3(const uint8* src_argb, uint8* dst_argb, int width) {
3848
  asm volatile (
3849
    "movdqa    %3,%%xmm4                       \n"
3850
    "movdqa    %4,%%xmm5                       \n"
3851 3852

    // 8 pixel loop.
3853
    LABELALIGN
3854
  "1:                                          \n"
3855 3856
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
3857 3858 3859
    "pmaddubsw %%xmm4,%%xmm0                   \n"
    "pmaddubsw %%xmm4,%%xmm1                   \n"
    "phaddw    %%xmm1,%%xmm0                   \n"
3860
    "paddw     %%xmm5,%%xmm0                   \n"
3861 3862
    "psrlw     $0x7,%%xmm0                     \n"
    "packuswb  %%xmm0,%%xmm0                   \n"
3863 3864
    "movdqu    " MEMACCESS(0) ",%%xmm2         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm3   \n"
3865
    "lea       " MEMLEA(0x20,0) ",%0           \n"
3866 3867 3868 3869 3870
    "psrld     $0x18,%%xmm2                    \n"
    "psrld     $0x18,%%xmm3                    \n"
    "packuswb  %%xmm3,%%xmm2                   \n"
    "packuswb  %%xmm2,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm3                   \n"
3871
    "punpcklbw %%xmm0,%%xmm0                   \n"
3872
    "punpcklbw %%xmm2,%%xmm3                   \n"
3873
    "movdqa    %%xmm0,%%xmm1                   \n"
3874 3875
    "punpcklwd %%xmm3,%%xmm0                   \n"
    "punpckhwd %%xmm3,%%xmm1                   \n"
3876 3877
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
3878
    "lea       " MEMLEA(0x20,1) ",%1           \n"
3879
    "sub       $0x8,%2                         \n"
3880
    "jg        1b                              \n"
3881 3882 3883
  : "+r"(src_argb),   // %0
    "+r"(dst_argb),   // %1
    "+r"(width)       // %2
3884 3885
  : "m"(kARGBToYJ),   // %3
    "m"(kAddYJ64)     // %4
3886
  : "memory", "cc"
3887
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
3888 3889 3890
  );
}
#endif  // HAS_ARGBGRAYROW_SSSE3
3891 3892 3893 3894 3895 3896

#ifdef HAS_ARGBSEPIAROW_SSSE3
//    b = (r * 35 + g * 68 + b * 17) >> 7
//    g = (r * 45 + g * 88 + b * 22) >> 7
//    r = (r * 50 + g * 98 + b * 24) >> 7
// Constant for ARGB color to sepia tone
3897 3898
static vec8 kARGBToSepiaB = {17, 68, 35, 0, 17, 68, 35, 0,
                             17, 68, 35, 0, 17, 68, 35, 0};
3899

3900 3901
static vec8 kARGBToSepiaG = {22, 88, 45, 0, 22, 88, 45, 0,
                             22, 88, 45, 0, 22, 88, 45, 0};
3902

3903 3904
static vec8 kARGBToSepiaR = {24, 98, 50, 0, 24, 98, 50, 0,
                             24, 98, 50, 0, 24, 98, 50, 0};
3905

3906
// Convert 8 ARGB pixels (32 bytes) to 8 Sepia ARGB pixels.
3907 3908 3909 3910 3911
void ARGBSepiaRow_SSSE3(uint8* dst_argb, int width) {
  asm volatile (
    "movdqa    %2,%%xmm2                       \n"
    "movdqa    %3,%%xmm3                       \n"
    "movdqa    %4,%%xmm4                       \n"
3912 3913

    // 8 pixel loop.
3914
    LABELALIGN
3915
  "1:                                          \n"
3916 3917
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm6   \n"
3918 3919 3920 3921 3922
    "pmaddubsw %%xmm2,%%xmm0                   \n"
    "pmaddubsw %%xmm2,%%xmm6                   \n"
    "phaddw    %%xmm6,%%xmm0                   \n"
    "psrlw     $0x7,%%xmm0                     \n"
    "packuswb  %%xmm0,%%xmm0                   \n"
3923 3924
    "movdqu    " MEMACCESS(0) ",%%xmm5         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
3925 3926 3927 3928 3929 3930
    "pmaddubsw %%xmm3,%%xmm5                   \n"
    "pmaddubsw %%xmm3,%%xmm1                   \n"
    "phaddw    %%xmm1,%%xmm5                   \n"
    "psrlw     $0x7,%%xmm5                     \n"
    "packuswb  %%xmm5,%%xmm5                   \n"
    "punpcklbw %%xmm5,%%xmm0                   \n"
3931 3932
    "movdqu    " MEMACCESS(0) ",%%xmm5         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
3933 3934 3935 3936 3937
    "pmaddubsw %%xmm4,%%xmm5                   \n"
    "pmaddubsw %%xmm4,%%xmm1                   \n"
    "phaddw    %%xmm1,%%xmm5                   \n"
    "psrlw     $0x7,%%xmm5                     \n"
    "packuswb  %%xmm5,%%xmm5                   \n"
3938 3939
    "movdqu    " MEMACCESS(0) ",%%xmm6         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
3940 3941 3942 3943 3944 3945 3946 3947
    "psrld     $0x18,%%xmm6                    \n"
    "psrld     $0x18,%%xmm1                    \n"
    "packuswb  %%xmm1,%%xmm6                   \n"
    "packuswb  %%xmm6,%%xmm6                   \n"
    "punpcklbw %%xmm6,%%xmm5                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "punpcklwd %%xmm5,%%xmm0                   \n"
    "punpckhwd %%xmm5,%%xmm1                   \n"
3948 3949
    "movdqu    %%xmm0," MEMACCESS(0) "         \n"
    "movdqu    %%xmm1," MEMACCESS2(0x10,0) "   \n"
3950
    "lea       " MEMLEA(0x20,0) ",%0           \n"
3951
    "sub       $0x8,%1                         \n"
3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963
    "jg        1b                              \n"
  : "+r"(dst_argb),      // %0
    "+r"(width)          // %1
  : "m"(kARGBToSepiaB),  // %2
    "m"(kARGBToSepiaG),  // %3
    "m"(kARGBToSepiaR)   // %4
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
  );
}
#endif  // HAS_ARGBSEPIAROW_SSSE3

3964 3965 3966
#ifdef HAS_ARGBCOLORMATRIXROW_SSSE3
// Tranform 8 ARGB pixels (32 bytes) with color matrix.
// Same as Sepia except matrix is provided.
3967 3968 3969 3970
void ARGBColorMatrixRow_SSSE3(const uint8* src_argb,
                              uint8* dst_argb,
                              const int8* matrix_argb,
                              int width) {
3971
  asm volatile (
3972 3973 3974 3975 3976
    "movdqu    " MEMACCESS(3) ",%%xmm5         \n"
    "pshufd    $0x00,%%xmm5,%%xmm2             \n"
    "pshufd    $0x55,%%xmm5,%%xmm3             \n"
    "pshufd    $0xaa,%%xmm5,%%xmm4             \n"
    "pshufd    $0xff,%%xmm5,%%xmm5             \n"
3977 3978

    // 8 pixel loop.
3979
    LABELALIGN
3980
  "1:                                          \n"
3981 3982
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm7   \n"
3983
    "pmaddubsw %%xmm2,%%xmm0                   \n"
3984
    "pmaddubsw %%xmm2,%%xmm7                   \n"
3985 3986
    "movdqu    " MEMACCESS(0) ",%%xmm6         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
3987
    "pmaddubsw %%xmm3,%%xmm6                   \n"
3988
    "pmaddubsw %%xmm3,%%xmm1                   \n"
3989 3990 3991 3992
    "phaddsw   %%xmm7,%%xmm0                   \n"
    "phaddsw   %%xmm1,%%xmm6                   \n"
    "psraw     $0x6,%%xmm0                     \n"
    "psraw     $0x6,%%xmm6                     \n"
3993
    "packuswb  %%xmm0,%%xmm0                   \n"
3994 3995
    "packuswb  %%xmm6,%%xmm6                   \n"
    "punpcklbw %%xmm6,%%xmm0                   \n"
3996 3997
    "movdqu    " MEMACCESS(0) ",%%xmm1         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm7   \n"
3998
    "pmaddubsw %%xmm4,%%xmm1                   \n"
3999 4000
    "pmaddubsw %%xmm4,%%xmm7                   \n"
    "phaddsw   %%xmm7,%%xmm1                   \n"
4001 4002
    "movdqu    " MEMACCESS(0) ",%%xmm6         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm7   \n"
4003 4004 4005 4006 4007 4008
    "pmaddubsw %%xmm5,%%xmm6                   \n"
    "pmaddubsw %%xmm5,%%xmm7                   \n"
    "phaddsw   %%xmm7,%%xmm6                   \n"
    "psraw     $0x6,%%xmm1                     \n"
    "psraw     $0x6,%%xmm6                     \n"
    "packuswb  %%xmm1,%%xmm1                   \n"
4009
    "packuswb  %%xmm6,%%xmm6                   \n"
4010 4011 4012 4013
    "punpcklbw %%xmm6,%%xmm1                   \n"
    "movdqa    %%xmm0,%%xmm6                   \n"
    "punpcklwd %%xmm1,%%xmm0                   \n"
    "punpckhwd %%xmm1,%%xmm6                   \n"
4014 4015
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "movdqu    %%xmm6," MEMACCESS2(0x10,1) "   \n"
4016 4017
    "lea       " MEMLEA(0x20,0) ",%0           \n"
    "lea       " MEMLEA(0x20,1) ",%1           \n"
4018
    "sub       $0x8,%2                         \n"
4019
    "jg        1b                              \n"
4020 4021 4022 4023
  : "+r"(src_argb),      // %0
    "+r"(dst_argb),      // %1
    "+r"(width)          // %2
  : "r"(matrix_argb)     // %3
4024
  : "memory", "cc"
4025
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
4026 4027 4028 4029
  );
}
#endif  // HAS_ARGBCOLORMATRIXROW_SSSE3

4030 4031
#ifdef HAS_ARGBQUANTIZEROW_SSE2
// Quantize 4 ARGB pixels (16 bytes).
4032 4033 4034 4035 4036
void ARGBQuantizeRow_SSE2(uint8* dst_argb,
                          int scale,
                          int interval_size,
                          int interval_offset,
                          int width) {
4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051
  asm volatile (
    "movd      %2,%%xmm2                       \n"
    "movd      %3,%%xmm3                       \n"
    "movd      %4,%%xmm4                       \n"
    "pshuflw   $0x40,%%xmm2,%%xmm2             \n"
    "pshufd    $0x44,%%xmm2,%%xmm2             \n"
    "pshuflw   $0x40,%%xmm3,%%xmm3             \n"
    "pshufd    $0x44,%%xmm3,%%xmm3             \n"
    "pshuflw   $0x40,%%xmm4,%%xmm4             \n"
    "pshufd    $0x44,%%xmm4,%%xmm4             \n"
    "pxor      %%xmm5,%%xmm5                   \n"
    "pcmpeqb   %%xmm6,%%xmm6                   \n"
    "pslld     $0x18,%%xmm6                    \n"

    // 4 pixel loop.
4052
    LABELALIGN
4053
  "1:                                          \n"
4054
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
4055 4056
    "punpcklbw %%xmm5,%%xmm0                   \n"
    "pmulhuw   %%xmm2,%%xmm0                   \n"
4057
    "movdqu    " MEMACCESS(0) ",%%xmm1         \n"
4058 4059 4060
    "punpckhbw %%xmm5,%%xmm1                   \n"
    "pmulhuw   %%xmm2,%%xmm1                   \n"
    "pmullw    %%xmm3,%%xmm0                   \n"
4061
    "movdqu    " MEMACCESS(0) ",%%xmm7         \n"
4062 4063 4064 4065 4066 4067
    "pmullw    %%xmm3,%%xmm1                   \n"
    "pand      %%xmm6,%%xmm7                   \n"
    "paddw     %%xmm4,%%xmm0                   \n"
    "paddw     %%xmm4,%%xmm1                   \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
    "por       %%xmm7,%%xmm0                   \n"
4068
    "movdqu    %%xmm0," MEMACCESS(0) "         \n"
4069
    "lea       " MEMLEA(0x10,0) ",%0           \n"
4070
    "sub       $0x4,%1                         \n"
4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082
    "jg        1b                              \n"
  : "+r"(dst_argb),       // %0
    "+r"(width)           // %1
  : "r"(scale),           // %2
    "r"(interval_size),   // %3
    "r"(interval_offset)  // %4
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
  );
}
#endif  // HAS_ARGBQUANTIZEROW_SSE2

4083 4084
#ifdef HAS_ARGBSHADEROW_SSE2
// Shade 4 pixels at a time by specified value.
4085 4086 4087
void ARGBShadeRow_SSE2(const uint8* src_argb,
                       uint8* dst_argb,
                       int width,
4088 4089 4090 4091 4092 4093 4094
                       uint32 value) {
  asm volatile (
    "movd      %3,%%xmm2                       \n"
    "punpcklbw %%xmm2,%%xmm2                   \n"
    "punpcklqdq %%xmm2,%%xmm2                  \n"

    // 4 pixel loop.
4095
    LABELALIGN
4096
  "1:                                          \n"
4097
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
4098
    "lea       " MEMLEA(0x10,0) ",%0           \n"
4099 4100 4101 4102 4103 4104 4105 4106
    "movdqa    %%xmm0,%%xmm1                   \n"
    "punpcklbw %%xmm0,%%xmm0                   \n"
    "punpckhbw %%xmm1,%%xmm1                   \n"
    "pmulhuw   %%xmm2,%%xmm0                   \n"
    "pmulhuw   %%xmm2,%%xmm1                   \n"
    "psrlw     $0x8,%%xmm0                     \n"
    "psrlw     $0x8,%%xmm1                     \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
4107
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
4108
    "lea       " MEMLEA(0x10,1) ",%1           \n"
4109
    "sub       $0x4,%2                         \n"
4110
    "jg        1b                              \n"
4111 4112 4113 4114
  : "+r"(src_argb),  // %0
    "+r"(dst_argb),  // %1
    "+r"(width)      // %2
  : "r"(value)       // %3
4115 4116 4117 4118 4119 4120
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2"
  );
}
#endif  // HAS_ARGBSHADEROW_SSE2

4121
#ifdef HAS_ARGBMULTIPLYROW_SSE2
4122
// Multiply 2 rows of ARGB pixels together, 4 pixels at a time.
4123 4124 4125 4126
void ARGBMultiplyRow_SSE2(const uint8* src_argb0,
                          const uint8* src_argb1,
                          uint8* dst_argb,
                          int width) {
4127
  asm volatile (
4128
    "pxor      %%xmm5,%%xmm5                  \n"
4129 4130

    // 4 pixel loop.
4131
    LABELALIGN
4132
  "1:                                          \n"
4133 4134 4135 4136
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "lea       " MEMLEA(0x10,0) ",%0           \n"
    "movdqu    " MEMACCESS(1) ",%%xmm2         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
4137 4138
    "movdqu    %%xmm0,%%xmm1                   \n"
    "movdqu    %%xmm2,%%xmm3                   \n"
4139 4140 4141 4142 4143 4144 4145
    "punpcklbw %%xmm0,%%xmm0                   \n"
    "punpckhbw %%xmm1,%%xmm1                   \n"
    "punpcklbw %%xmm5,%%xmm2                   \n"
    "punpckhbw %%xmm5,%%xmm3                   \n"
    "pmulhuw   %%xmm2,%%xmm0                   \n"
    "pmulhuw   %%xmm3,%%xmm1                   \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
4146 4147
    "movdqu    %%xmm0," MEMACCESS(2) "         \n"
    "lea       " MEMLEA(0x10,2) ",%2           \n"
4148
    "sub       $0x4,%3                         \n"
4149
    "jg        1b                              \n"
4150 4151
  : "+r"(src_argb0),  // %0
    "+r"(src_argb1),  // %1
4152 4153
    "+r"(dst_argb),   // %2
    "+r"(width)       // %3
4154 4155 4156 4157 4158 4159 4160
  :
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
  );
}
#endif  // HAS_ARGBMULTIPLYROW_SSE2

4161 4162
#ifdef HAS_ARGBMULTIPLYROW_AVX2
// Multiply 2 rows of ARGB pixels together, 8 pixels at a time.
4163 4164 4165 4166
void ARGBMultiplyRow_AVX2(const uint8* src_argb0,
                          const uint8* src_argb1,
                          uint8* dst_argb,
                          int width) {
4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201
  asm volatile (
    "vpxor      %%ymm5,%%ymm5,%%ymm5           \n"

    // 4 pixel loop.
    LABELALIGN
  "1:                                          \n"
    "vmovdqu    " MEMACCESS(0) ",%%ymm1        \n"
    "lea        " MEMLEA(0x20,0) ",%0          \n"
    "vmovdqu    " MEMACCESS(1) ",%%ymm3        \n"
    "lea        " MEMLEA(0x20,1) ",%1          \n"
    "vpunpcklbw %%ymm1,%%ymm1,%%ymm0           \n"
    "vpunpckhbw %%ymm1,%%ymm1,%%ymm1           \n"
    "vpunpcklbw %%ymm5,%%ymm3,%%ymm2           \n"
    "vpunpckhbw %%ymm5,%%ymm3,%%ymm3           \n"
    "vpmulhuw   %%ymm2,%%ymm0,%%ymm0           \n"
    "vpmulhuw   %%ymm3,%%ymm1,%%ymm1           \n"
    "vpackuswb  %%ymm1,%%ymm0,%%ymm0           \n"
    "vmovdqu    %%ymm0," MEMACCESS(2) "        \n"
    "lea       " MEMLEA(0x20,2) ",%2           \n"
    "sub        $0x8,%3                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_argb0),  // %0
    "+r"(src_argb1),  // %1
    "+r"(dst_argb),   // %2
    "+r"(width)       // %3
  :
  : "memory", "cc"
#if defined(__AVX2__)
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
#endif
  );
}
#endif  // HAS_ARGBMULTIPLYROW_AVX2

4202 4203
#ifdef HAS_ARGBADDROW_SSE2
// Add 2 rows of ARGB pixels together, 4 pixels at a time.
4204 4205 4206 4207
void ARGBAddRow_SSE2(const uint8* src_argb0,
                     const uint8* src_argb1,
                     uint8* dst_argb,
                     int width) {
4208 4209
  asm volatile (
    // 4 pixel loop.
4210
    LABELALIGN
4211
  "1:                                          \n"
4212 4213 4214 4215
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "lea       " MEMLEA(0x10,0) ",%0           \n"
    "movdqu    " MEMACCESS(1) ",%%xmm1         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
4216
    "paddusb   %%xmm1,%%xmm0                   \n"
4217 4218
    "movdqu    %%xmm0," MEMACCESS(2) "         \n"
    "lea       " MEMLEA(0x10,2) ",%2           \n"
4219
    "sub       $0x4,%3                         \n"
4220 4221 4222 4223 4224 4225 4226
    "jg        1b                              \n"
  : "+r"(src_argb0),  // %0
    "+r"(src_argb1),  // %1
    "+r"(dst_argb),   // %2
    "+r"(width)       // %3
  :
  : "memory", "cc"
4227
    , "xmm0", "xmm1"
4228 4229 4230 4231
  );
}
#endif  // HAS_ARGBADDROW_SSE2

4232 4233
#ifdef HAS_ARGBADDROW_AVX2
// Add 2 rows of ARGB pixels together, 4 pixels at a time.
4234 4235 4236 4237
void ARGBAddRow_AVX2(const uint8* src_argb0,
                     const uint8* src_argb1,
                     uint8* dst_argb,
                     int width) {
4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261
  asm volatile (
    // 4 pixel loop.
    LABELALIGN
  "1:                                          \n"
    "vmovdqu    " MEMACCESS(0) ",%%ymm0        \n"
    "lea        " MEMLEA(0x20,0) ",%0          \n"
    "vpaddusb   " MEMACCESS(1) ",%%ymm0,%%ymm0 \n"
    "lea        " MEMLEA(0x20,1) ",%1          \n"
    "vmovdqu    %%ymm0," MEMACCESS(2) "        \n"
    "lea        " MEMLEA(0x20,2) ",%2          \n"
    "sub        $0x8,%3                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_argb0),  // %0
    "+r"(src_argb1),  // %1
    "+r"(dst_argb),   // %2
    "+r"(width)       // %3
  :
  : "memory", "cc"
    , "xmm0"
  );
}
#endif  // HAS_ARGBADDROW_AVX2

4262 4263
#ifdef HAS_ARGBSUBTRACTROW_SSE2
// Subtract 2 rows of ARGB pixels, 4 pixels at a time.
4264 4265 4266 4267
void ARGBSubtractRow_SSE2(const uint8* src_argb0,
                          const uint8* src_argb1,
                          uint8* dst_argb,
                          int width) {
4268 4269
  asm volatile (
    // 4 pixel loop.
4270
    LABELALIGN
4271
  "1:                                          \n"
4272 4273 4274 4275
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "lea       " MEMLEA(0x10,0) ",%0           \n"
    "movdqu    " MEMACCESS(1) ",%%xmm1         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
4276
    "psubusb   %%xmm1,%%xmm0                   \n"
4277 4278
    "movdqu    %%xmm0," MEMACCESS(2) "         \n"
    "lea       " MEMLEA(0x10,2) ",%2           \n"
4279
    "sub       $0x4,%3                         \n"
4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291
    "jg        1b                              \n"
  : "+r"(src_argb0),  // %0
    "+r"(src_argb1),  // %1
    "+r"(dst_argb),   // %2
    "+r"(width)       // %3
  :
  : "memory", "cc"
    , "xmm0", "xmm1"
  );
}
#endif  // HAS_ARGBSUBTRACTROW_SSE2

4292 4293
#ifdef HAS_ARGBSUBTRACTROW_AVX2
// Subtract 2 rows of ARGB pixels, 8 pixels at a time.
4294 4295 4296 4297
void ARGBSubtractRow_AVX2(const uint8* src_argb0,
                          const uint8* src_argb1,
                          uint8* dst_argb,
                          int width) {
4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321
  asm volatile (
    // 4 pixel loop.
    LABELALIGN
  "1:                                          \n"
    "vmovdqu    " MEMACCESS(0) ",%%ymm0        \n"
    "lea        " MEMLEA(0x20,0) ",%0          \n"
    "vpsubusb   " MEMACCESS(1) ",%%ymm0,%%ymm0 \n"
    "lea        " MEMLEA(0x20,1) ",%1          \n"
    "vmovdqu    %%ymm0," MEMACCESS(2) "        \n"
    "lea        " MEMLEA(0x20,2) ",%2          \n"
    "sub        $0x8,%3                        \n"
    "jg        1b                              \n"
    "vzeroupper                                \n"
  : "+r"(src_argb0),  // %0
    "+r"(src_argb1),  // %1
    "+r"(dst_argb),   // %2
    "+r"(width)       // %3
  :
  : "memory", "cc"
    , "xmm0"
  );
}
#endif  // HAS_ARGBSUBTRACTROW_AVX2

4322
#ifdef HAS_SOBELXROW_SSE2
4323 4324 4325 4326
// SobelX as a matrix is
// -1  0  1
// -2  0  2
// -1  0  1
4327 4328 4329 4330 4331
void SobelXRow_SSE2(const uint8* src_y0,
                    const uint8* src_y1,
                    const uint8* src_y2,
                    uint8* dst_sobelx,
                    int width) {
4332 4333 4334 4335 4336 4337 4338
  asm volatile (
    "sub       %0,%1                           \n"
    "sub       %0,%2                           \n"
    "sub       %0,%3                           \n"
    "pxor      %%xmm5,%%xmm5                   \n"

    // 8 pixel loop.
4339
    LABELALIGN
4340
  "1:                                          \n"
4341 4342
    "movq      " MEMACCESS(0) ",%%xmm0         \n"
    "movq      " MEMACCESS2(0x2,0) ",%%xmm1    \n"
4343 4344 4345
    "punpcklbw %%xmm5,%%xmm0                   \n"
    "punpcklbw %%xmm5,%%xmm1                   \n"
    "psubw     %%xmm1,%%xmm0                   \n"
4346 4347
    MEMOPREG(movq,0x00,0,1,1,xmm1)             //  movq      (%0,%1,1),%%xmm1
    MEMOPREG(movq,0x02,0,1,1,xmm2)             //  movq      0x2(%0,%1,1),%%xmm2
4348 4349 4350
    "punpcklbw %%xmm5,%%xmm1                   \n"
    "punpcklbw %%xmm5,%%xmm2                   \n"
    "psubw     %%xmm2,%%xmm1                   \n"
4351 4352
    MEMOPREG(movq,0x00,0,2,1,xmm2)             //  movq      (%0,%2,1),%%xmm2
    MEMOPREG(movq,0x02,0,2,1,xmm3)             //  movq      0x2(%0,%2,1),%%xmm3
4353 4354 4355 4356 4357 4358
    "punpcklbw %%xmm5,%%xmm2                   \n"
    "punpcklbw %%xmm5,%%xmm3                   \n"
    "psubw     %%xmm3,%%xmm2                   \n"
    "paddw     %%xmm2,%%xmm0                   \n"
    "paddw     %%xmm1,%%xmm0                   \n"
    "paddw     %%xmm1,%%xmm0                   \n"
4359 4360 4361
    "pxor      %%xmm1,%%xmm1                   \n"
    "psubw     %%xmm0,%%xmm1                   \n"
    "pmaxsw    %%xmm1,%%xmm0                   \n"
4362
    "packuswb  %%xmm0,%%xmm0                   \n"
4363
    MEMOPMEM(movq,xmm0,0x00,0,3,1)             //  movq      %%xmm0,(%0,%3,1)
4364
    "lea       " MEMLEA(0x8,0) ",%0            \n"
4365
    "sub       $0x8,%4                         \n"
4366 4367 4368 4369 4370 4371 4372
    "jg        1b                              \n"
  : "+r"(src_y0),      // %0
    "+r"(src_y1),      // %1
    "+r"(src_y2),      // %2
    "+r"(dst_sobelx),  // %3
    "+r"(width)        // %4
  :
4373 4374
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
4375 4376
  );
}
4377
#endif  // HAS_SOBELXROW_SSE2
4378

4379
#ifdef HAS_SOBELYROW_SSE2
4380 4381 4382 4383
// SobelY as a matrix is
// -1 -2 -1
//  0  0  0
//  1  2  1
4384 4385 4386 4387
void SobelYRow_SSE2(const uint8* src_y0,
                    const uint8* src_y1,
                    uint8* dst_sobely,
                    int width) {
4388 4389 4390 4391 4392 4393
  asm volatile (
    "sub       %0,%1                           \n"
    "sub       %0,%2                           \n"
    "pxor      %%xmm5,%%xmm5                   \n"

    // 8 pixel loop.
4394
    LABELALIGN
4395
  "1:                                          \n"
4396
    "movq      " MEMACCESS(0) ",%%xmm0         \n"
4397
    MEMOPREG(movq,0x00,0,1,1,xmm1)             //  movq      (%0,%1,1),%%xmm1
4398 4399 4400
    "punpcklbw %%xmm5,%%xmm0                   \n"
    "punpcklbw %%xmm5,%%xmm1                   \n"
    "psubw     %%xmm1,%%xmm0                   \n"
4401
    "movq      " MEMACCESS2(0x1,0) ",%%xmm1    \n"
4402
    MEMOPREG(movq,0x01,0,1,1,xmm2)             //  movq      0x1(%0,%1,1),%%xmm2
4403 4404 4405
    "punpcklbw %%xmm5,%%xmm1                   \n"
    "punpcklbw %%xmm5,%%xmm2                   \n"
    "psubw     %%xmm2,%%xmm1                   \n"
4406
    "movq      " MEMACCESS2(0x2,0) ",%%xmm2    \n"
4407
    MEMOPREG(movq,0x02,0,1,1,xmm3)             //  movq      0x2(%0,%1,1),%%xmm3
4408 4409 4410 4411 4412 4413
    "punpcklbw %%xmm5,%%xmm2                   \n"
    "punpcklbw %%xmm5,%%xmm3                   \n"
    "psubw     %%xmm3,%%xmm2                   \n"
    "paddw     %%xmm2,%%xmm0                   \n"
    "paddw     %%xmm1,%%xmm0                   \n"
    "paddw     %%xmm1,%%xmm0                   \n"
4414 4415 4416
    "pxor      %%xmm1,%%xmm1                   \n"
    "psubw     %%xmm0,%%xmm1                   \n"
    "pmaxsw    %%xmm1,%%xmm0                   \n"
4417
    "packuswb  %%xmm0,%%xmm0                   \n"
4418
    MEMOPMEM(movq,xmm0,0x00,0,2,1)             //  movq      %%xmm0,(%0,%2,1)
4419
    "lea       " MEMLEA(0x8,0) ",%0            \n"
4420
    "sub       $0x8,%3                         \n"
4421 4422 4423 4424 4425 4426
    "jg        1b                              \n"
  : "+r"(src_y0),      // %0
    "+r"(src_y1),      // %1
    "+r"(dst_sobely),  // %2
    "+r"(width)        // %3
  :
4427 4428
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
4429 4430
  );
}
4431
#endif  // HAS_SOBELYROW_SSE2
4432 4433 4434 4435 4436 4437 4438

#ifdef HAS_SOBELROW_SSE2
// Adds Sobel X and Sobel Y and stores Sobel into ARGB.
// A = 255
// R = Sobel
// G = Sobel
// B = Sobel
4439 4440 4441 4442
void SobelRow_SSE2(const uint8* src_sobelx,
                   const uint8* src_sobely,
                   uint8* dst_argb,
                   int width) {
4443 4444 4445 4446 4447 4448
  asm volatile (
    "sub       %0,%1                           \n"
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    "pslld     $0x18,%%xmm5                    \n"

    // 8 pixel loop.
4449
    LABELALIGN
4450
  "1:                                          \n"
4451 4452
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    MEMOPREG(movdqu,0x00,0,1,1,xmm1)           //  movdqu    (%0,%1,1),%%xmm1
4453
    "lea       " MEMLEA(0x10,0) ",%0           \n"
4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467
    "paddusb   %%xmm1,%%xmm0                   \n"
    "movdqa    %%xmm0,%%xmm2                   \n"
    "punpcklbw %%xmm0,%%xmm2                   \n"
    "punpckhbw %%xmm0,%%xmm0                   \n"
    "movdqa    %%xmm2,%%xmm1                   \n"
    "punpcklwd %%xmm2,%%xmm1                   \n"
    "punpckhwd %%xmm2,%%xmm2                   \n"
    "por       %%xmm5,%%xmm1                   \n"
    "por       %%xmm5,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm3                   \n"
    "punpcklwd %%xmm0,%%xmm3                   \n"
    "punpckhwd %%xmm0,%%xmm0                   \n"
    "por       %%xmm5,%%xmm3                   \n"
    "por       %%xmm5,%%xmm0                   \n"
4468 4469 4470 4471
    "movdqu    %%xmm1," MEMACCESS(2) "         \n"
    "movdqu    %%xmm2," MEMACCESS2(0x10,2) "   \n"
    "movdqu    %%xmm3," MEMACCESS2(0x20,2) "   \n"
    "movdqu    %%xmm0," MEMACCESS2(0x30,2) "   \n"
4472
    "lea       " MEMLEA(0x40,2) ",%2           \n"
4473
    "sub       $0x10,%3                        \n"
4474 4475 4476 4477 4478 4479
    "jg        1b                              \n"
  : "+r"(src_sobelx),  // %0
    "+r"(src_sobely),  // %1
    "+r"(dst_argb),    // %2
    "+r"(width)        // %3
  :
4480 4481
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
4482 4483 4484 4485
  );
}
#endif  // HAS_SOBELROW_SSE2

4486 4487
#ifdef HAS_SOBELTOPLANEROW_SSE2
// Adds Sobel X and Sobel Y and stores Sobel into a plane.
4488 4489 4490 4491
void SobelToPlaneRow_SSE2(const uint8* src_sobelx,
                          const uint8* src_sobely,
                          uint8* dst_y,
                          int width) {
4492 4493 4494 4495 4496 4497
  asm volatile (
    "sub       %0,%1                           \n"
    "pcmpeqb   %%xmm5,%%xmm5                   \n"
    "pslld     $0x18,%%xmm5                    \n"

    // 8 pixel loop.
4498
    LABELALIGN
4499
  "1:                                          \n"
4500 4501
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    MEMOPREG(movdqu,0x00,0,1,1,xmm1)           //  movdqu    (%0,%1,1),%%xmm1
4502
    "lea       " MEMLEA(0x10,0) ",%0           \n"
4503
    "paddusb   %%xmm1,%%xmm0                   \n"
4504
    "movdqu    %%xmm0," MEMACCESS(2) "         \n"
4505
    "lea       " MEMLEA(0x10,2) ",%2           \n"
4506
    "sub       $0x10,%3                        \n"
4507 4508 4509 4510 4511 4512
    "jg        1b                              \n"
  : "+r"(src_sobelx),  // %0
    "+r"(src_sobely),  // %1
    "+r"(dst_y),       // %2
    "+r"(width)        // %3
  :
4513 4514
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1"
4515 4516 4517 4518
  );
}
#endif  // HAS_SOBELTOPLANEROW_SSE2

4519 4520 4521 4522 4523 4524
#ifdef HAS_SOBELXYROW_SSE2
// Mixes Sobel X, Sobel Y and Sobel into ARGB.
// A = 255
// R = Sobel X
// G = Sobel
// B = Sobel Y
4525 4526 4527 4528
void SobelXYRow_SSE2(const uint8* src_sobelx,
                     const uint8* src_sobely,
                     uint8* dst_argb,
                     int width) {
4529 4530 4531 4532 4533
  asm volatile (
    "sub       %0,%1                           \n"
    "pcmpeqb   %%xmm5,%%xmm5                   \n"

    // 8 pixel loop.
4534
    LABELALIGN
4535
  "1:                                          \n"
4536 4537
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    MEMOPREG(movdqu,0x00,0,1,1,xmm1)           //  movdqu    (%0,%1,1),%%xmm1
4538
    "lea       " MEMLEA(0x10,0) ",%0           \n"
4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552
    "movdqa    %%xmm0,%%xmm2                   \n"
    "paddusb   %%xmm1,%%xmm2                   \n"
    "movdqa    %%xmm0,%%xmm3                   \n"
    "punpcklbw %%xmm5,%%xmm3                   \n"
    "punpckhbw %%xmm5,%%xmm0                   \n"
    "movdqa    %%xmm1,%%xmm4                   \n"
    "punpcklbw %%xmm2,%%xmm4                   \n"
    "punpckhbw %%xmm2,%%xmm1                   \n"
    "movdqa    %%xmm4,%%xmm6                   \n"
    "punpcklwd %%xmm3,%%xmm6                   \n"
    "punpckhwd %%xmm3,%%xmm4                   \n"
    "movdqa    %%xmm1,%%xmm7                   \n"
    "punpcklwd %%xmm0,%%xmm7                   \n"
    "punpckhwd %%xmm0,%%xmm1                   \n"
4553 4554 4555 4556
    "movdqu    %%xmm6," MEMACCESS(2) "         \n"
    "movdqu    %%xmm4," MEMACCESS2(0x10,2) "   \n"
    "movdqu    %%xmm7," MEMACCESS2(0x20,2) "   \n"
    "movdqu    %%xmm1," MEMACCESS2(0x30,2) "   \n"
4557
    "lea       " MEMLEA(0x40,2) ",%2           \n"
4558
    "sub       $0x10,%3                        \n"
4559 4560 4561 4562 4563 4564
    "jg        1b                              \n"
  : "+r"(src_sobelx),  // %0
    "+r"(src_sobely),  // %1
    "+r"(dst_argb),    // %2
    "+r"(width)        // %3
  :
4565 4566
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
4567 4568 4569 4570
  );
}
#endif  // HAS_SOBELXYROW_SSE2

fbarchard@google.com's avatar
fbarchard@google.com committed
4571 4572 4573
#ifdef HAS_COMPUTECUMULATIVESUMROW_SSE2
// Creates a table of cumulative sums where each value is a sum of all values
// above and to the left of the value, inclusive of the value.
4574 4575 4576 4577
void ComputeCumulativeSumRow_SSE2(const uint8* row,
                                  int32* cumsum,
                                  const int32* previous_cumsum,
                                  int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
4578 4579 4580 4581 4582 4583 4584 4585 4586
  asm volatile (
    "pxor      %%xmm0,%%xmm0                   \n"
    "pxor      %%xmm1,%%xmm1                   \n"
    "sub       $0x4,%3                         \n"
    "jl        49f                             \n"
    "test      $0xf,%1                         \n"
    "jne       49f                             \n"

  // 4 pixel loop                              \n"
4587
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
4588
  "40:                                         \n"
4589 4590
    "movdqu    " MEMACCESS(0) ",%%xmm2         \n"
    "lea       " MEMLEA(0x10,0) ",%0           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4591 4592 4593 4594 4595 4596 4597 4598 4599 4600
    "movdqa    %%xmm2,%%xmm4                   \n"
    "punpcklbw %%xmm1,%%xmm2                   \n"
    "movdqa    %%xmm2,%%xmm3                   \n"
    "punpcklwd %%xmm1,%%xmm2                   \n"
    "punpckhwd %%xmm1,%%xmm3                   \n"
    "punpckhbw %%xmm1,%%xmm4                   \n"
    "movdqa    %%xmm4,%%xmm5                   \n"
    "punpcklwd %%xmm1,%%xmm4                   \n"
    "punpckhwd %%xmm1,%%xmm5                   \n"
    "paddd     %%xmm2,%%xmm0                   \n"
4601
    "movdqu    " MEMACCESS(2) ",%%xmm2         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4602 4603
    "paddd     %%xmm0,%%xmm2                   \n"
    "paddd     %%xmm3,%%xmm0                   \n"
4604
    "movdqu    " MEMACCESS2(0x10,2) ",%%xmm3   \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4605 4606
    "paddd     %%xmm0,%%xmm3                   \n"
    "paddd     %%xmm4,%%xmm0                   \n"
4607
    "movdqu    " MEMACCESS2(0x20,2) ",%%xmm4   \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4608 4609
    "paddd     %%xmm0,%%xmm4                   \n"
    "paddd     %%xmm5,%%xmm0                   \n"
4610
    "movdqu    " MEMACCESS2(0x30,2) ",%%xmm5   \n"
4611
    "lea       " MEMLEA(0x40,2) ",%2           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4612
    "paddd     %%xmm0,%%xmm5                   \n"
4613 4614 4615 4616
    "movdqu    %%xmm2," MEMACCESS(1) "         \n"
    "movdqu    %%xmm3," MEMACCESS2(0x10,1) "   \n"
    "movdqu    %%xmm4," MEMACCESS2(0x20,1) "   \n"
    "movdqu    %%xmm5," MEMACCESS2(0x30,1) "   \n"
4617
    "lea       " MEMLEA(0x40,1) ",%1           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4618 4619 4620 4621 4622 4623 4624 4625
    "sub       $0x4,%3                         \n"
    "jge       40b                             \n"

  "49:                                         \n"
    "add       $0x3,%3                         \n"
    "jl        19f                             \n"

  // 1 pixel loop                              \n"
4626
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
4627
  "10:                                         \n"
4628 4629
    "movd      " MEMACCESS(0) ",%%xmm2         \n"
    "lea       " MEMLEA(0x4,0) ",%0            \n"
4630 4631
    "punpcklbw %%xmm1,%%xmm2                   \n"
    "punpcklwd %%xmm1,%%xmm2                   \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4632
    "paddd     %%xmm2,%%xmm0                   \n"
4633 4634
    "movdqu    " MEMACCESS(2) ",%%xmm2         \n"
    "lea       " MEMLEA(0x10,2) ",%2           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4635
    "paddd     %%xmm0,%%xmm2                   \n"
4636 4637
    "movdqu    %%xmm2," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652
    "sub       $0x1,%3                         \n"
    "jge       10b                             \n"

  "19:                                         \n"
  : "+r"(row),  // %0
    "+r"(cumsum),  // %1
    "+r"(previous_cumsum),  // %2
    "+r"(width)  // %3
  :
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
  );
}
#endif  // HAS_COMPUTECUMULATIVESUMROW_SSE2

4653
#ifdef HAS_CUMULATIVESUMTOAVERAGEROW_SSE2
4654 4655 4656 4657 4658
void CumulativeSumToAverageRow_SSE2(const int32* topleft,
                                    const int32* botleft,
                                    int width,
                                    int area,
                                    uint8* dst,
4659
                                    int count) {
fbarchard@google.com's avatar
fbarchard@google.com committed
4660
  asm volatile (
4661 4662 4663
    "movd      %5,%%xmm5                       \n"
    "cvtdq2ps  %%xmm5,%%xmm5                   \n"
    "rcpss     %%xmm5,%%xmm4                   \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4664 4665 4666
    "pshufd    $0x0,%%xmm4,%%xmm4              \n"
    "sub       $0x4,%3                         \n"
    "jl        49f                             \n"
4667 4668 4669
    "cmpl      $0x80,%5                        \n"
    "ja        40f                             \n"

4670 4671 4672 4673 4674
    "pshufd    $0x0,%%xmm5,%%xmm5              \n"
    "pcmpeqb   %%xmm6,%%xmm6                   \n"
    "psrld     $0x10,%%xmm6                    \n"
    "cvtdq2ps  %%xmm6,%%xmm6                   \n"
    "addps     %%xmm6,%%xmm5                   \n"
4675 4676 4677 4678 4679
    "mulps     %%xmm4,%%xmm5                   \n"
    "cvtps2dq  %%xmm5,%%xmm5                   \n"
    "packssdw  %%xmm5,%%xmm5                   \n"

  // 4 pixel small loop                        \n"
4680
    LABELALIGN
4681
  "4:                                         \n"
4682 4683 4684 4685
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm3   \n"
4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709
    MEMOPREG(psubd,0x00,0,4,4,xmm0)            // psubd    0x00(%0,%4,4),%%xmm0
    MEMOPREG(psubd,0x10,0,4,4,xmm1)            // psubd    0x10(%0,%4,4),%%xmm1
    MEMOPREG(psubd,0x20,0,4,4,xmm2)            // psubd    0x20(%0,%4,4),%%xmm2
    MEMOPREG(psubd,0x30,0,4,4,xmm3)            // psubd    0x30(%0,%4,4),%%xmm3
    "lea       " MEMLEA(0x40,0) ",%0           \n"
    "psubd     " MEMACCESS(1) ",%%xmm0         \n"
    "psubd     " MEMACCESS2(0x10,1) ",%%xmm1   \n"
    "psubd     " MEMACCESS2(0x20,1) ",%%xmm2   \n"
    "psubd     " MEMACCESS2(0x30,1) ",%%xmm3   \n"
    MEMOPREG(paddd,0x00,1,4,4,xmm0)            // paddd    0x00(%1,%4,4),%%xmm0
    MEMOPREG(paddd,0x10,1,4,4,xmm1)            // paddd    0x10(%1,%4,4),%%xmm1
    MEMOPREG(paddd,0x20,1,4,4,xmm2)            // paddd    0x20(%1,%4,4),%%xmm2
    MEMOPREG(paddd,0x30,1,4,4,xmm3)            // paddd    0x30(%1,%4,4),%%xmm3
    "lea       " MEMLEA(0x40,1) ",%1           \n"
    "packssdw  %%xmm1,%%xmm0                   \n"
    "packssdw  %%xmm3,%%xmm2                   \n"
    "pmulhuw   %%xmm5,%%xmm0                   \n"
    "pmulhuw   %%xmm5,%%xmm2                   \n"
    "packuswb  %%xmm2,%%xmm0                   \n"
    "movdqu    %%xmm0," MEMACCESS(2) "         \n"
    "lea       " MEMLEA(0x10,2) ",%2           \n"
    "sub       $0x4,%3                         \n"
    "jge       4b                              \n"
    "jmp       49f                             \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4710 4711

  // 4 pixel loop                              \n"
4712
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
4713
  "40:                                         \n"
4714 4715 4716 4717
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
    "movdqu    " MEMACCESS2(0x20,0) ",%%xmm2   \n"
    "movdqu    " MEMACCESS2(0x30,0) ",%%xmm3   \n"
4718 4719 4720 4721
    MEMOPREG(psubd,0x00,0,4,4,xmm0)            // psubd    0x00(%0,%4,4),%%xmm0
    MEMOPREG(psubd,0x10,0,4,4,xmm1)            // psubd    0x10(%0,%4,4),%%xmm1
    MEMOPREG(psubd,0x20,0,4,4,xmm2)            // psubd    0x20(%0,%4,4),%%xmm2
    MEMOPREG(psubd,0x30,0,4,4,xmm3)            // psubd    0x30(%0,%4,4),%%xmm3
4722 4723 4724 4725 4726
    "lea       " MEMLEA(0x40,0) ",%0           \n"
    "psubd     " MEMACCESS(1) ",%%xmm0         \n"
    "psubd     " MEMACCESS2(0x10,1) ",%%xmm1   \n"
    "psubd     " MEMACCESS2(0x20,1) ",%%xmm2   \n"
    "psubd     " MEMACCESS2(0x30,1) ",%%xmm3   \n"
4727 4728 4729 4730
    MEMOPREG(paddd,0x00,1,4,4,xmm0)            // paddd    0x00(%1,%4,4),%%xmm0
    MEMOPREG(paddd,0x10,1,4,4,xmm1)            // paddd    0x10(%1,%4,4),%%xmm1
    MEMOPREG(paddd,0x20,1,4,4,xmm2)            // paddd    0x20(%1,%4,4),%%xmm2
    MEMOPREG(paddd,0x30,1,4,4,xmm3)            // paddd    0x30(%1,%4,4),%%xmm3
4731
    "lea       " MEMLEA(0x40,1) ",%1           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746
    "cvtdq2ps  %%xmm0,%%xmm0                   \n"
    "cvtdq2ps  %%xmm1,%%xmm1                   \n"
    "mulps     %%xmm4,%%xmm0                   \n"
    "mulps     %%xmm4,%%xmm1                   \n"
    "cvtdq2ps  %%xmm2,%%xmm2                   \n"
    "cvtdq2ps  %%xmm3,%%xmm3                   \n"
    "mulps     %%xmm4,%%xmm2                   \n"
    "mulps     %%xmm4,%%xmm3                   \n"
    "cvtps2dq  %%xmm0,%%xmm0                   \n"
    "cvtps2dq  %%xmm1,%%xmm1                   \n"
    "cvtps2dq  %%xmm2,%%xmm2                   \n"
    "cvtps2dq  %%xmm3,%%xmm3                   \n"
    "packssdw  %%xmm1,%%xmm0                   \n"
    "packssdw  %%xmm3,%%xmm2                   \n"
    "packuswb  %%xmm2,%%xmm0                   \n"
4747 4748
    "movdqu    %%xmm0," MEMACCESS(2) "         \n"
    "lea       " MEMLEA(0x10,2) ",%2           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4749 4750 4751 4752 4753 4754 4755 4756
    "sub       $0x4,%3                         \n"
    "jge       40b                             \n"

  "49:                                         \n"
    "add       $0x3,%3                         \n"
    "jl        19f                             \n"

  // 1 pixel loop                              \n"
4757
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
4758
  "10:                                         \n"
4759
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
4760
    MEMOPREG(psubd,0x00,0,4,4,xmm0)            // psubd    0x00(%0,%4,4),%%xmm0
4761 4762
    "lea       " MEMLEA(0x10,0) ",%0           \n"
    "psubd     " MEMACCESS(1) ",%%xmm0         \n"
4763
    MEMOPREG(paddd,0x00,1,4,4,xmm0)            // paddd    0x00(%1,%4,4),%%xmm0
4764
    "lea       " MEMLEA(0x10,1) ",%1           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4765 4766 4767 4768 4769
    "cvtdq2ps  %%xmm0,%%xmm0                   \n"
    "mulps     %%xmm4,%%xmm0                   \n"
    "cvtps2dq  %%xmm0,%%xmm0                   \n"
    "packssdw  %%xmm0,%%xmm0                   \n"
    "packuswb  %%xmm0,%%xmm0                   \n"
4770 4771
    "movd      %%xmm0," MEMACCESS(2) "         \n"
    "lea       " MEMLEA(0x4,2) ",%2            \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
4772 4773 4774 4775 4776 4777 4778
    "sub       $0x1,%3                         \n"
    "jge       10b                             \n"
  "19:                                         \n"
  : "+r"(topleft),  // %0
    "+r"(botleft),  // %1
    "+r"(dst),      // %2
    "+rm"(count)    // %3
4779
  : "r"((intptr_t)(width)),  // %4
fbarchard@google.com's avatar
fbarchard@google.com committed
4780
    "rm"(area)     // %5
4781 4782
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
fbarchard@google.com's avatar
fbarchard@google.com committed
4783 4784
  );
}
4785
#endif  // HAS_CUMULATIVESUMTOAVERAGEROW_SSE2
fbarchard@google.com's avatar
fbarchard@google.com committed
4786

4787 4788
#ifdef HAS_ARGBAFFINEROW_SSE2
// Copy ARGB pixels from source image with slope to a row of destination.
4789
LIBYUV_API
4790 4791 4792 4793 4794
void ARGBAffineRow_SSE2(const uint8* src_argb,
                        int src_argb_stride,
                        uint8* dst_argb,
                        const float* src_dudv,
                        int width) {
4795
  intptr_t src_argb_stride_temp = src_argb_stride;
4796
  intptr_t temp;
4797
  asm volatile (
4798 4799
    "movq      " MEMACCESS(3) ",%%xmm2         \n"
    "movq      " MEMACCESS2(0x08,3) ",%%xmm7   \n"
4800 4801
    "shl       $0x10,%1                        \n"
    "add       $0x4,%1                         \n"
4802 4803 4804 4805 4806 4807
    "movd      %1,%%xmm5                       \n"
    "sub       $0x4,%4                         \n"
    "jl        49f                             \n"

    "pshufd    $0x44,%%xmm7,%%xmm7             \n"
    "pshufd    $0x0,%%xmm5,%%xmm5              \n"
4808
    "movdqa    %%xmm2,%%xmm0                   \n"
4809
    "addps     %%xmm7,%%xmm0                   \n"
4810
    "movlhps   %%xmm0,%%xmm2                   \n"
4811 4812 4813 4814 4815
    "movdqa    %%xmm7,%%xmm4                   \n"
    "addps     %%xmm4,%%xmm4                   \n"
    "movdqa    %%xmm2,%%xmm3                   \n"
    "addps     %%xmm4,%%xmm3                   \n"
    "addps     %%xmm4,%%xmm4                   \n"
4816

4817
  // 4 pixel loop                              \n"
4818
    LABELALIGN
4819
  "40:                                         \n"
4820 4821 4822 4823 4824
    "cvttps2dq %%xmm2,%%xmm0                   \n"  // x, y float to int first 2
    "cvttps2dq %%xmm3,%%xmm1                   \n"  // x, y float to int next 2
    "packssdw  %%xmm1,%%xmm0                   \n"  // x, y as 8 shorts
    "pmaddwd   %%xmm5,%%xmm0                   \n"  // off = x * 4 + y * stride
    "movd      %%xmm0,%k1                      \n"
4825
    "pshufd    $0x39,%%xmm0,%%xmm0             \n"
4826
    "movd      %%xmm0,%k5                      \n"
4827
    "pshufd    $0x39,%%xmm0,%%xmm0             \n"
4828 4829
    MEMOPREG(movd,0x00,0,1,1,xmm1)             //  movd      (%0,%1,1),%%xmm1
    MEMOPREG(movd,0x00,0,5,1,xmm6)             //  movd      (%0,%5,1),%%xmm6
4830 4831
    "punpckldq %%xmm6,%%xmm1                   \n"
    "addps     %%xmm4,%%xmm2                   \n"
4832
    "movq      %%xmm1," MEMACCESS(2) "         \n"
4833
    "movd      %%xmm0,%k1                      \n"
4834
    "pshufd    $0x39,%%xmm0,%%xmm0             \n"
4835 4836 4837
    "movd      %%xmm0,%k5                      \n"
    MEMOPREG(movd,0x00,0,1,1,xmm0)             //  movd      (%0,%1,1),%%xmm0
    MEMOPREG(movd,0x00,0,5,1,xmm6)             //  movd      (%0,%5,1),%%xmm6
4838 4839
    "punpckldq %%xmm6,%%xmm0                   \n"
    "addps     %%xmm4,%%xmm3                   \n"
4840 4841
    "movq      %%xmm0," MEMACCESS2(0x08,2) "   \n"
    "lea       " MEMLEA(0x10,2) ",%2           \n"
4842
    "sub       $0x4,%4                         \n"
4843 4844 4845 4846
    "jge       40b                             \n"

  "49:                                         \n"
    "add       $0x3,%4                         \n"
4847 4848 4849
    "jl        19f                             \n"

  // 1 pixel loop                              \n"
4850
    LABELALIGN
4851
  "10:                                         \n"
4852 4853 4854 4855
    "cvttps2dq %%xmm2,%%xmm0                   \n"
    "packssdw  %%xmm0,%%xmm0                   \n"
    "pmaddwd   %%xmm5,%%xmm0                   \n"
    "addps     %%xmm7,%%xmm2                   \n"
4856 4857
    "movd      %%xmm0,%k1                      \n"
    MEMOPREG(movd,0x00,0,1,1,xmm0)             //  movd      (%0,%1,1),%%xmm0
4858 4859
    "movd      %%xmm0," MEMACCESS(2) "         \n"
    "lea       " MEMLEA(0x04,2) ",%2           \n"
4860
    "sub       $0x1,%4                         \n"
4861 4862 4863 4864 4865
    "jge       10b                             \n"
  "19:                                         \n"
  : "+r"(src_argb),  // %0
    "+r"(src_argb_stride_temp),  // %1
    "+r"(dst_argb),  // %2
4866
    "+r"(src_dudv),  // %3
4867
    "+rm"(width),    // %4
4868
    "=&r"(temp)      // %5
4869
  :
4870 4871
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
4872 4873 4874 4875
  );
}
#endif  // HAS_ARGBAFFINEROW_SSE2

4876
#ifdef HAS_INTERPOLATEROW_SSSE3
4877
// Bilinear filter 16x2 -> 16x1
4878 4879 4880 4881
void InterpolateRow_SSSE3(uint8* dst_ptr,
                          const uint8* src_ptr,
                          ptrdiff_t src_stride,
                          int dst_width,
4882
                          int source_y_fraction) {
4883 4884 4885
  asm volatile (
    "sub       %1,%0                           \n"
    "cmp       $0x0,%3                         \n"
4886
    "je        100f                            \n"
4887
    "cmp       $0x80,%3                        \n"
4888 4889
    "je        50f                             \n"

4890 4891
    "movd      %3,%%xmm0                       \n"
    "neg       %3                              \n"
4892
    "add       $0x100,%3                       \n"
4893 4894 4895 4896
    "movd      %3,%%xmm5                       \n"
    "punpcklbw %%xmm0,%%xmm5                   \n"
    "punpcklwd %%xmm5,%%xmm5                   \n"
    "pshufd    $0x0,%%xmm5,%%xmm5              \n"
4897
    "mov       $0x80808080,%%eax               \n"
4898 4899
    "movd      %%eax,%%xmm4                    \n"
    "pshufd    $0x0,%%xmm4,%%xmm4              \n"
4900 4901

    // General purpose row blend.
4902
    LABELALIGN
4903
  "1:                                          \n"
4904 4905
    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
    MEMOPREG(movdqu,0x00,1,4,1,xmm2)
4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920
    "movdqa     %%xmm0,%%xmm1                  \n"
    "punpcklbw  %%xmm2,%%xmm0                  \n"
    "punpckhbw  %%xmm2,%%xmm1                  \n"
    "psubb      %%xmm4,%%xmm0                  \n"
    "psubb      %%xmm4,%%xmm1                  \n"
    "movdqa     %%xmm5,%%xmm2                  \n"
    "movdqa     %%xmm5,%%xmm3                  \n"
    "pmaddubsw  %%xmm0,%%xmm2                  \n"
    "pmaddubsw  %%xmm1,%%xmm3                  \n"
    "paddw      %%xmm4,%%xmm2                  \n"
    "paddw      %%xmm4,%%xmm3                  \n"
    "psrlw      $0x8,%%xmm2                    \n"
    "psrlw      $0x8,%%xmm3                    \n"
    "packuswb   %%xmm3,%%xmm2                  \n"
    MEMOPMEM(movdqu,xmm2,0x00,1,0,1)
4921
    "lea       " MEMLEA(0x10,1) ",%1           \n"
4922
    "sub       $0x10,%2                        \n"
4923
    "jg        1b                              \n"
4924 4925 4926
    "jmp       99f                             \n"

    // Blend 50 / 50.
4927
    LABELALIGN
4928
  "50:                                         \n"
4929 4930
    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
    MEMOPREG(movdqu,0x00,1,4,1,xmm1)
4931
    "pavgb     %%xmm1,%%xmm0                   \n"
4932
    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)
4933
    "lea       " MEMLEA(0x10,1) ",%1           \n"
4934
    "sub       $0x10,%2                        \n"
4935 4936 4937 4938
    "jg        50b                             \n"
    "jmp       99f                             \n"

    // Blend 100 / 0 - Copy row unchanged.
4939
    LABELALIGN
4940
  "100:                                        \n"
4941 4942
    "movdqu    " MEMACCESS(1) ",%%xmm0         \n"
    MEMOPMEM(movdqu,xmm0,0x00,1,0,1)
4943
    "lea       " MEMLEA(0x10,1) ",%1           \n"
4944
    "sub       $0x10,%2                        \n"
4945 4946 4947
    "jg        100b                            \n"

  "99:                                         \n"
4948 4949 4950
  : "+r"(dst_ptr),     // %0
    "+r"(src_ptr),     // %1
    "+rm"(dst_width),  // %2
4951
    "+r"(source_y_fraction)  // %3
4952
  : "r"((intptr_t)(src_stride))  // %4
4953
  : "memory", "cc", "eax", NACL_R14
4954
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
4955 4956
  );
}
4957
#endif  // HAS_INTERPOLATEROW_SSSE3
4958

4959 4960
#ifdef HAS_INTERPOLATEROW_AVX2
// Bilinear filter 32x2 -> 32x1
4961 4962 4963 4964
void InterpolateRow_AVX2(uint8* dst_ptr,
                         const uint8* src_ptr,
                         ptrdiff_t src_stride,
                         int dst_width,
4965 4966 4967 4968 4969
                         int source_y_fraction) {
  asm volatile (
    "cmp       $0x0,%3                         \n"
    "je        100f                            \n"
    "sub       %1,%0                           \n"
4970
    "cmp       $0x80,%3                        \n"
4971 4972 4973 4974
    "je        50f                             \n"

    "vmovd      %3,%%xmm0                      \n"
    "neg        %3                             \n"
4975
    "add        $0x100,%3                      \n"
4976 4977 4978
    "vmovd      %3,%%xmm5                      \n"
    "vpunpcklbw %%xmm0,%%xmm5,%%xmm5           \n"
    "vpunpcklwd %%xmm5,%%xmm5,%%xmm5           \n"
4979 4980
    "vbroadcastss %%xmm5,%%ymm5                \n"
    "mov        $0x80808080,%%eax              \n"
4981 4982
    "vmovd      %%eax,%%xmm4                   \n"
    "vbroadcastss %%xmm4,%%ymm4                \n"
4983 4984 4985 4986 4987 4988 4989 4990

    // General purpose row blend.
    LABELALIGN
  "1:                                          \n"
    "vmovdqu    " MEMACCESS(1) ",%%ymm0        \n"
    MEMOPREG(vmovdqu,0x00,1,4,1,ymm2)
    "vpunpckhbw %%ymm2,%%ymm0,%%ymm1           \n"
    "vpunpcklbw %%ymm2,%%ymm0,%%ymm0           \n"
4991 4992 4993 4994
    "vpsubb     %%ymm4,%%ymm1,%%ymm1           \n"
    "vpsubb     %%ymm4,%%ymm0,%%ymm0           \n"
    "vpmaddubsw %%ymm1,%%ymm5,%%ymm1           \n"
    "vpmaddubsw %%ymm0,%%ymm5,%%ymm0           \n"
4995
    "vpaddw     %%ymm4,%%ymm1,%%ymm1           \n"
4996 4997 4998
    "vpaddw     %%ymm4,%%ymm0,%%ymm0           \n"
    "vpsrlw     $0x8,%%ymm1,%%ymm1             \n"
    "vpsrlw     $0x8,%%ymm0,%%ymm0             \n"
4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027
    "vpackuswb  %%ymm1,%%ymm0,%%ymm0           \n"
    MEMOPMEM(vmovdqu,ymm0,0x00,1,0,1)
    "lea       " MEMLEA(0x20,1) ",%1           \n"
    "sub       $0x20,%2                        \n"
    "jg        1b                              \n"
    "jmp       99f                             \n"

    // Blend 50 / 50.
    LABELALIGN
  "50:                                         \n"
    "vmovdqu    " MEMACCESS(1) ",%%ymm0        \n"
    VMEMOPREG(vpavgb,0x00,1,4,1,ymm0,ymm0)     // vpavgb (%1,%4,1),%%ymm0,%%ymm0
    MEMOPMEM(vmovdqu,ymm0,0x00,1,0,1)
    "lea       " MEMLEA(0x20,1) ",%1           \n"
    "sub       $0x20,%2                        \n"
    "jg        50b                             \n"
    "jmp       99f                             \n"

    // Blend 100 / 0 - Copy row unchanged.
    LABELALIGN
  "100:                                        \n"
    "rep movsb " MEMMOVESTRING(1,0) "          \n"
    "jmp       999f                            \n"

  "99:                                         \n"
    "vzeroupper                                \n"
  "999:                                        \n"
  : "+D"(dst_ptr),    // %0
    "+S"(src_ptr),    // %1
5028
    "+cm"(dst_width),  // %2
5029 5030
    "+r"(source_y_fraction)  // %3
  : "r"((intptr_t)(src_stride))  // %4
5031 5032
  : "memory", "cc", "eax", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm4", "xmm5"
5033 5034 5035 5036
  );
}
#endif  // HAS_INTERPOLATEROW_AVX2

5037
#ifdef HAS_ARGBSHUFFLEROW_SSSE3
fbarchard@google.com's avatar
fbarchard@google.com committed
5038
// For BGRAToARGB, ABGRToARGB, RGBAToARGB, and ARGBToRGBA.
5039 5040 5041 5042
void ARGBShuffleRow_SSSE3(const uint8* src_argb,
                          uint8* dst_argb,
                          const uint8* shuffler,
                          int width) {
fbarchard@google.com's avatar
fbarchard@google.com committed
5043
  asm volatile (
5044
    "movdqu    " MEMACCESS(3) ",%%xmm5         \n"
5045
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
5046
  "1:                                          \n"
5047 5048
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "movdqu    " MEMACCESS2(0x10,0) ",%%xmm1   \n"
5049
    "lea       " MEMLEA(0x20,0) ",%0           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5050 5051
    "pshufb    %%xmm5,%%xmm0                   \n"
    "pshufb    %%xmm5,%%xmm1                   \n"
5052
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
5053 5054
    "movdqu    %%xmm1," MEMACCESS2(0x10,1) "   \n"
    "lea       " MEMLEA(0x20,1) ",%1           \n"
5055
    "sub       $0x8,%2                         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5056 5057 5058
    "jg        1b                              \n"
  : "+r"(src_argb),  // %0
    "+r"(dst_argb),  // %1
5059
    "+r"(width)        // %2
fbarchard@google.com's avatar
fbarchard@google.com committed
5060 5061 5062 5063 5064
  : "r"(shuffler)    // %3
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm5"
  );
}
5065
#endif  // HAS_ARGBSHUFFLEROW_SSSE3
fbarchard@google.com's avatar
fbarchard@google.com committed
5066

5067 5068
#ifdef HAS_ARGBSHUFFLEROW_AVX2
// For BGRAToARGB, ABGRToARGB, RGBAToARGB, and ARGBToRGBA.
5069 5070 5071 5072
void ARGBShuffleRow_AVX2(const uint8* src_argb,
                         uint8* dst_argb,
                         const uint8* shuffler,
                         int width) {
5073
  asm volatile (
5074
    "vbroadcastf128 " MEMACCESS(3) ",%%ymm5    \n"
5075
    LABELALIGN
5076
  "1:                                          \n"
5077 5078 5079
    "vmovdqu   " MEMACCESS(0) ",%%ymm0         \n"
    "vmovdqu   " MEMACCESS2(0x20,0) ",%%ymm1   \n"
    "lea       " MEMLEA(0x40,0) ",%0           \n"
5080 5081
    "vpshufb   %%ymm5,%%ymm0,%%ymm0            \n"
    "vpshufb   %%ymm5,%%ymm1,%%ymm1            \n"
5082 5083 5084
    "vmovdqu   %%ymm0," MEMACCESS(1) "         \n"
    "vmovdqu   %%ymm1," MEMACCESS2(0x20,1) "   \n"
    "lea       " MEMLEA(0x40,1) ",%1           \n"
5085
    "sub       $0x10,%2                        \n"
5086
    "jg        1b                              \n"
5087
    "vzeroupper                                \n"
5088 5089
  : "+r"(src_argb),  // %0
    "+r"(dst_argb),  // %1
5090
    "+r"(width)        // %2
5091 5092 5093 5094 5095 5096 5097
  : "r"(shuffler)    // %3
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm5"
  );
}
#endif  // HAS_ARGBSHUFFLEROW_AVX2

5098 5099
#ifdef HAS_ARGBSHUFFLEROW_SSE2
// For BGRAToARGB, ABGRToARGB, RGBAToARGB, and ARGBToRGBA.
5100 5101 5102 5103
void ARGBShuffleRow_SSE2(const uint8* src_argb,
                         uint8* dst_argb,
                         const uint8* shuffler,
                         int width) {
5104
  uintptr_t pixel_temp;
5105 5106
  asm volatile (
    "pxor      %%xmm5,%%xmm5                   \n"
5107
    "mov       " MEMACCESS(4) ",%k2            \n"
5108 5109 5110 5111 5112 5113 5114 5115 5116
    "cmp       $0x3000102,%k2                  \n"
    "je        3012f                           \n"
    "cmp       $0x10203,%k2                    \n"
    "je        123f                            \n"
    "cmp       $0x30201,%k2                    \n"
    "je        321f                            \n"
    "cmp       $0x2010003,%k2                  \n"
    "je        2103f                           \n"

5117
    LABELALIGN
5118
  "1:                                          \n"
5119
    "movzb     " MEMACCESS(4) ",%2             \n"
5120
    MEMOPARG(movzb,0x00,0,2,1,2) "             \n"  //  movzb     (%0,%2,1),%2
5121 5122
    "mov       %b2," MEMACCESS(1) "            \n"
    "movzb     " MEMACCESS2(0x1,4) ",%2        \n"
5123
    MEMOPARG(movzb,0x00,0,2,1,2) "             \n"  //  movzb     (%0,%2,1),%2
5124 5125
    "mov       %b2," MEMACCESS2(0x1,1) "       \n"
    "movzb     " MEMACCESS2(0x2,4) ",%2        \n"
5126
    MEMOPARG(movzb,0x00,0,2,1,2) "             \n"  //  movzb     (%0,%2,1),%2
5127 5128
    "mov       %b2," MEMACCESS2(0x2,1) "       \n"
    "movzb     " MEMACCESS2(0x3,4) ",%2        \n"
5129
    MEMOPARG(movzb,0x00,0,2,1,2) "             \n"  //  movzb     (%0,%2,1),%2
5130 5131 5132
    "mov       %b2," MEMACCESS2(0x3,1) "       \n"
    "lea       " MEMLEA(0x4,0) ",%0            \n"
    "lea       " MEMLEA(0x4,1) ",%1            \n"
5133 5134 5135 5136
    "sub       $0x1,%3                         \n"
    "jg        1b                              \n"
    "jmp       99f                             \n"

5137
    LABELALIGN
5138
  "123:                                        \n"
5139 5140
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "lea       " MEMLEA(0x10,0) ",%0           \n"
5141 5142 5143 5144 5145 5146 5147 5148
    "movdqa    %%xmm0,%%xmm1                   \n"
    "punpcklbw %%xmm5,%%xmm0                   \n"
    "punpckhbw %%xmm5,%%xmm1                   \n"
    "pshufhw   $0x1b,%%xmm0,%%xmm0             \n"
    "pshuflw   $0x1b,%%xmm0,%%xmm0             \n"
    "pshufhw   $0x1b,%%xmm1,%%xmm1             \n"
    "pshuflw   $0x1b,%%xmm1,%%xmm1             \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
5149 5150
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
5151
    "sub       $0x4,%3                         \n"
5152 5153 5154
    "jg        123b                            \n"
    "jmp       99f                             \n"

5155
    LABELALIGN
5156
  "321:                                        \n"
5157 5158
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "lea       " MEMLEA(0x10,0) ",%0           \n"
5159 5160 5161 5162 5163 5164 5165 5166
    "movdqa    %%xmm0,%%xmm1                   \n"
    "punpcklbw %%xmm5,%%xmm0                   \n"
    "punpckhbw %%xmm5,%%xmm1                   \n"
    "pshufhw   $0x39,%%xmm0,%%xmm0             \n"
    "pshuflw   $0x39,%%xmm0,%%xmm0             \n"
    "pshufhw   $0x39,%%xmm1,%%xmm1             \n"
    "pshuflw   $0x39,%%xmm1,%%xmm1             \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
5167 5168
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
5169
    "sub       $0x4,%3                         \n"
5170 5171 5172
    "jg        321b                            \n"
    "jmp       99f                             \n"

5173
    LABELALIGN
5174
  "2103:                                       \n"
5175 5176
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "lea       " MEMLEA(0x10,0) ",%0           \n"
5177 5178 5179 5180 5181 5182 5183 5184
    "movdqa    %%xmm0,%%xmm1                   \n"
    "punpcklbw %%xmm5,%%xmm0                   \n"
    "punpckhbw %%xmm5,%%xmm1                   \n"
    "pshufhw   $0x93,%%xmm0,%%xmm0             \n"
    "pshuflw   $0x93,%%xmm0,%%xmm0             \n"
    "pshufhw   $0x93,%%xmm1,%%xmm1             \n"
    "pshuflw   $0x93,%%xmm1,%%xmm1             \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
5185 5186
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
5187
    "sub       $0x4,%3                         \n"
5188 5189 5190
    "jg        2103b                           \n"
    "jmp       99f                             \n"

5191
    LABELALIGN
5192
  "3012:                                       \n"
5193 5194
    "movdqu    " MEMACCESS(0) ",%%xmm0         \n"
    "lea       " MEMLEA(0x10,0) ",%0           \n"
5195 5196 5197 5198 5199 5200 5201 5202
    "movdqa    %%xmm0,%%xmm1                   \n"
    "punpcklbw %%xmm5,%%xmm0                   \n"
    "punpckhbw %%xmm5,%%xmm1                   \n"
    "pshufhw   $0xc6,%%xmm0,%%xmm0             \n"
    "pshuflw   $0xc6,%%xmm0,%%xmm0             \n"
    "pshufhw   $0xc6,%%xmm1,%%xmm1             \n"
    "pshuflw   $0xc6,%%xmm1,%%xmm1             \n"
    "packuswb  %%xmm1,%%xmm0                   \n"
5203 5204
    "movdqu    %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x10,1) ",%1           \n"
5205
    "sub       $0x4,%3                         \n"
5206 5207 5208
    "jg        3012b                           \n"

  "99:                                         \n"
5209 5210 5211
  : "+r"(src_argb),     // %0
    "+r"(dst_argb),     // %1
    "=&d"(pixel_temp),  // %2
5212
    "+r"(width)         // %3
5213
  : "r"(shuffler)       // %4
5214 5215
  : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm5"
5216 5217 5218 5219
  );
}
#endif  // HAS_ARGBSHUFFLEROW_SSE2

5220
#ifdef HAS_I422TOYUY2ROW_SSE2
fbarchard@google.com's avatar
fbarchard@google.com committed
5221 5222 5223
void I422ToYUY2Row_SSE2(const uint8* src_y,
                        const uint8* src_u,
                        const uint8* src_v,
5224 5225 5226
                        uint8* dst_frame,
                        int width) {
  asm volatile (
fbarchard@google.com's avatar
fbarchard@google.com committed
5227
    "sub       %1,%2                             \n"
5228
    LABELALIGN
Frank Barchard's avatar
Frank Barchard committed
5229
  "1:                                          \n"
5230 5231 5232
    "movq      " MEMACCESS(1) ",%%xmm2           \n"
    MEMOPREG(movq,0x00,1,2,1,xmm3)               //  movq    (%1,%2,1),%%xmm3
    "lea       " MEMLEA(0x8,1) ",%1              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5233
    "punpcklbw %%xmm3,%%xmm2                     \n"
5234 5235
    "movdqu    " MEMACCESS(0) ",%%xmm0           \n"
    "lea       " MEMLEA(0x10,0) ",%0             \n"
5236
    "movdqa    %%xmm0,%%xmm1                     \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5237 5238
    "punpcklbw %%xmm2,%%xmm0                     \n"
    "punpckhbw %%xmm2,%%xmm1                     \n"
5239 5240 5241
    "movdqu    %%xmm0," MEMACCESS(3) "           \n"
    "movdqu    %%xmm1," MEMACCESS2(0x10,3) "     \n"
    "lea       " MEMLEA(0x20,3) ",%3             \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5242 5243 5244 5245 5246 5247 5248 5249
    "sub       $0x10,%4                          \n"
    "jg         1b                               \n"
    : "+r"(src_y),  // %0
      "+r"(src_u),  // %1
      "+r"(src_v),  // %2
      "+r"(dst_frame),  // %3
      "+rm"(width)  // %4
    :
5250 5251
    : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3"
fbarchard@google.com's avatar
fbarchard@google.com committed
5252 5253
  );
}
5254
#endif  // HAS_I422TOYUY2ROW_SSE2
fbarchard@google.com's avatar
fbarchard@google.com committed
5255

5256
#ifdef HAS_I422TOUYVYROW_SSE2
fbarchard@google.com's avatar
fbarchard@google.com committed
5257 5258 5259
void I422ToUYVYRow_SSE2(const uint8* src_y,
                        const uint8* src_u,
                        const uint8* src_v,
5260 5261 5262
                        uint8* dst_frame,
                        int width) {
  asm volatile (
fbarchard@google.com's avatar
fbarchard@google.com committed
5263
    "sub        %1,%2                            \n"
5264
    LABELALIGN
Frank Barchard's avatar
Frank Barchard committed
5265
  "1:                                          \n"
5266 5267 5268
    "movq      " MEMACCESS(1) ",%%xmm2           \n"
    MEMOPREG(movq,0x00,1,2,1,xmm3)               //  movq    (%1,%2,1),%%xmm3
    "lea       " MEMLEA(0x8,1) ",%1              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5269
    "punpcklbw %%xmm3,%%xmm2                     \n"
5270
    "movdqu    " MEMACCESS(0) ",%%xmm0           \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5271
    "movdqa    %%xmm2,%%xmm1                     \n"
5272
    "lea       " MEMLEA(0x10,0) ",%0             \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5273 5274
    "punpcklbw %%xmm0,%%xmm1                     \n"
    "punpckhbw %%xmm0,%%xmm2                     \n"
5275 5276 5277
    "movdqu    %%xmm1," MEMACCESS(3) "           \n"
    "movdqu    %%xmm2," MEMACCESS2(0x10,3) "     \n"
    "lea       " MEMLEA(0x20,3) ",%3             \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5278 5279 5280 5281 5282 5283 5284 5285
    "sub       $0x10,%4                          \n"
    "jg         1b                               \n"
    : "+r"(src_y),  // %0
      "+r"(src_u),  // %1
      "+r"(src_v),  // %2
      "+r"(dst_frame),  // %3
      "+rm"(width)  // %4
    :
5286 5287
    : "memory", "cc", NACL_R14
    "xmm0", "xmm1", "xmm2", "xmm3"
fbarchard@google.com's avatar
fbarchard@google.com committed
5288 5289
  );
}
5290
#endif  // HAS_I422TOUYVYROW_SSE2
fbarchard@google.com's avatar
fbarchard@google.com committed
5291

5292 5293
#ifdef HAS_ARGBPOLYNOMIALROW_SSE2
void ARGBPolynomialRow_SSE2(const uint8* src_argb,
5294 5295
                            uint8* dst_argb,
                            const float* poly,
5296 5297 5298 5299 5300
                            int width) {
  asm volatile (
    "pxor      %%xmm3,%%xmm3                   \n"

    // 2 pixel loop.
5301
    LABELALIGN
5302
  "1:                                          \n"
5303 5304
    "movq      " MEMACCESS(0) ",%%xmm0         \n"
    "lea       " MEMLEA(0x8,0) ",%0            \n"
5305 5306 5307 5308 5309 5310 5311 5312
    "punpcklbw %%xmm3,%%xmm0                   \n"
    "movdqa    %%xmm0,%%xmm4                   \n"
    "punpcklwd %%xmm3,%%xmm0                   \n"
    "punpckhwd %%xmm3,%%xmm4                   \n"
    "cvtdq2ps  %%xmm0,%%xmm0                   \n"
    "cvtdq2ps  %%xmm4,%%xmm4                   \n"
    "movdqa    %%xmm0,%%xmm1                   \n"
    "movdqa    %%xmm4,%%xmm5                   \n"
5313 5314 5315 5316
    "mulps     " MEMACCESS2(0x10,3) ",%%xmm0   \n"
    "mulps     " MEMACCESS2(0x10,3) ",%%xmm4   \n"
    "addps     " MEMACCESS(3) ",%%xmm0         \n"
    "addps     " MEMACCESS(3) ",%%xmm4         \n"
5317 5318 5319 5320 5321 5322
    "movdqa    %%xmm1,%%xmm2                   \n"
    "movdqa    %%xmm5,%%xmm6                   \n"
    "mulps     %%xmm1,%%xmm2                   \n"
    "mulps     %%xmm5,%%xmm6                   \n"
    "mulps     %%xmm2,%%xmm1                   \n"
    "mulps     %%xmm6,%%xmm5                   \n"
5323 5324 5325 5326
    "mulps     " MEMACCESS2(0x20,3) ",%%xmm2   \n"
    "mulps     " MEMACCESS2(0x20,3) ",%%xmm6   \n"
    "mulps     " MEMACCESS2(0x30,3) ",%%xmm1   \n"
    "mulps     " MEMACCESS2(0x30,3) ",%%xmm5   \n"
5327 5328 5329 5330 5331 5332 5333 5334
    "addps     %%xmm2,%%xmm0                   \n"
    "addps     %%xmm6,%%xmm4                   \n"
    "addps     %%xmm1,%%xmm0                   \n"
    "addps     %%xmm5,%%xmm4                   \n"
    "cvttps2dq %%xmm0,%%xmm0                   \n"
    "cvttps2dq %%xmm4,%%xmm4                   \n"
    "packuswb  %%xmm4,%%xmm0                   \n"
    "packuswb  %%xmm0,%%xmm0                   \n"
5335 5336
    "movq      %%xmm0," MEMACCESS(1) "         \n"
    "lea       " MEMLEA(0x8,1) ",%1            \n"
5337
    "sub       $0x2,%2                         \n"
5338 5339 5340 5341 5342 5343 5344 5345 5346 5347 5348
    "jg        1b                              \n"
  : "+r"(src_argb),  // %0
    "+r"(dst_argb),  // %1
    "+r"(width)      // %2
  : "r"(poly)        // %3
  : "memory", "cc"
    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
  );
}
#endif  // HAS_ARGBPOLYNOMIALROW_SSE2

5349 5350
#ifdef HAS_ARGBPOLYNOMIALROW_AVX2
void ARGBPolynomialRow_AVX2(const uint8* src_argb,
5351 5352
                            uint8* dst_argb,
                            const float* poly,
5353 5354
                            int width) {
  asm volatile (
5355 5356 5357 5358
    "vbroadcastf128 " MEMACCESS(3) ",%%ymm4     \n"
    "vbroadcastf128 " MEMACCESS2(0x10,3) ",%%ymm5 \n"
    "vbroadcastf128 " MEMACCESS2(0x20,3) ",%%ymm6 \n"
    "vbroadcastf128 " MEMACCESS2(0x30,3) ",%%ymm7 \n"
5359 5360

    // 2 pixel loop.
5361
    LABELALIGN
5362
  "1:                                          \n"
5363 5364
    "vpmovzxbd   " MEMACCESS(0) ",%%ymm0       \n"  // 2 ARGB pixels
    "lea         " MEMLEA(0x8,0) ",%0          \n"
5365 5366 5367 5368 5369 5370 5371 5372 5373 5374
    "vcvtdq2ps   %%ymm0,%%ymm0                 \n"  // X 8 floats
    "vmulps      %%ymm0,%%ymm0,%%ymm2          \n"  // X * X
    "vmulps      %%ymm7,%%ymm0,%%ymm3          \n"  // C3 * X
    "vfmadd132ps %%ymm5,%%ymm4,%%ymm0          \n"  // result = C0 + C1 * X
    "vfmadd231ps %%ymm6,%%ymm2,%%ymm0          \n"  // result += C2 * X * X
    "vfmadd231ps %%ymm3,%%ymm2,%%ymm0          \n"  // result += C3 * X * X * X
    "vcvttps2dq  %%ymm0,%%ymm0                 \n"
    "vpackusdw   %%ymm0,%%ymm0,%%ymm0          \n"
    "vpermq      $0xd8,%%ymm0,%%ymm0           \n"
    "vpackuswb   %%xmm0,%%xmm0,%%xmm0          \n"
5375 5376
    "vmovq       %%xmm0," MEMACCESS(1) "       \n"
    "lea         " MEMLEA(0x8,1) ",%1          \n"
5377
    "sub         $0x2,%2                       \n"
5378 5379 5380 5381 5382 5383
    "jg          1b                            \n"
    "vzeroupper                                \n"
  : "+r"(src_argb),  // %0
    "+r"(dst_argb),  // %1
    "+r"(width)      // %2
  : "r"(poly)        // %3
5384 5385
  : "memory", "cc",
    "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
5386 5387 5388 5389
  );
}
#endif  // HAS_ARGBPOLYNOMIALROW_AVX2

5390
#ifdef HAS_HALFFLOATROW_SSE2
5391
static float kScaleBias = 1.9259299444e-34f;
5392 5393
void HalfFloatRow_SSE2(const uint16* src, uint16* dst, float scale, int width) {
  asm volatile (
5394
    "pshufd      $0x0,%3,%%xmm4                \n"
5395 5396 5397 5398 5399
    "pxor        %%xmm5,%%xmm5                 \n"

    // 16 pixel loop.
    LABELALIGN
  "1:                                          \n"
5400
    "movdqu      " MEMACCESS(0) ",%%xmm2       \n"  // 8 shorts
5401
    "lea         " MEMLEA(0x10,0) ",%0         \n"
5402 5403 5404 5405 5406 5407 5408 5409 5410 5411 5412 5413
    "movdqa      %%xmm2,%%xmm3                 \n"
    "punpcklwd   %%xmm5,%%xmm2                 \n"  // 8 ints in xmm2/1
    "cvtdq2ps    %%xmm2,%%xmm2                 \n"  // 8 floats
    "punpckhwd   %%xmm5,%%xmm3                 \n"
    "cvtdq2ps    %%xmm3,%%xmm3                 \n"
    "mulps       %%xmm4,%%xmm2                 \n"
    "mulps       %%xmm4,%%xmm3                 \n"
    "psrld       $0xd,%%xmm2                   \n"
    "psrld       $0xd,%%xmm3                   \n"
    "packssdw    %%xmm3,%%xmm2                 \n"
    "movdqu      %%xmm2," MEMACCESS(1) "       \n"
    "lea         " MEMLEA(0x10,1) ",%1         \n"
5414 5415
    "sub         $0x8,%2                       \n"
    "jg          1b                            \n"
5416 5417 5418 5419
  : "+r"(src),    // %0
    "+r"(dst),    // %1
    "+r"(width)   // %2
  : "x"(scale * kScaleBias)   // %3
5420
  : "memory", "cc",
5421
    "xmm2", "xmm3", "xmm4", "xmm5"
5422 5423 5424 5425
  );
}
#endif  // HAS_HALFFLOATROW_SSE2

5426
#ifdef HAS_HALFFLOATROW_AVX2
5427
// TODO(fbarchard): consider vadddw instead of vmulps
5428
void HalfFloatRow_AVX2(const uint16* src, uint16* dst, float scale, int width) {
5429 5430 5431 5432 5433 5434 5435
  asm volatile (
    "vbroadcastss  %3, %%ymm4                  \n"
    "vpxor      %%ymm5,%%ymm5,%%ymm5           \n"

    // 16 pixel loop.
    LABELALIGN
  "1:                                          \n"
5436
    "vmovdqu    " MEMACCESS(0) ",%%ymm2        \n"  // 16 shorts
5437
    "lea        " MEMLEA(0x20,0) ",%0          \n"
5438 5439
    "vpunpckhwd %%ymm5,%%ymm2,%%ymm3           \n"  // mutates
    "vpunpcklwd %%ymm5,%%ymm2,%%ymm2           \n"
5440 5441 5442 5443 5444 5445
    "vcvtdq2ps  %%ymm3,%%ymm3                  \n"
    "vcvtdq2ps  %%ymm2,%%ymm2                  \n"
    "vmulps     %%ymm3,%%ymm4,%%ymm3           \n"
    "vmulps     %%ymm2,%%ymm4,%%ymm2           \n"
    "vpsrld     $0xd,%%ymm3,%%ymm3             \n"
    "vpsrld     $0xd,%%ymm2,%%ymm2             \n"
5446
    "vpackssdw  %%ymm3, %%ymm2, %%ymm2         \n"  // unmutates
5447 5448 5449 5450 5451 5452 5453 5454 5455 5456 5457 5458 5459 5460 5461
    "vmovdqu    %%ymm2," MEMACCESS(1) "        \n"
    "lea        " MEMLEA(0x20,1) ",%1          \n"
    "sub        $0x10,%2                       \n"
    "jg         1b                             \n"
    "vzeroupper                                \n"
  : "+r"(src),    // %0
    "+r"(dst),    // %1
    "+r"(width)   // %2
  : "x"(scale * kScaleBias)   // %3
  : "memory", "cc",
    "xmm2", "xmm3", "xmm4", "xmm5"
  );
}
#endif  // HAS_HALFFLOATROW_AVX2

5462
//#ifdef HAS_HALFFLOATROW_F16C
5463
void HalfFloatRow_F16C(const uint16* src, uint16* dst, float scale, int width) {
5464 5465 5466 5467 5468 5469
  asm volatile (
   "vbroadcastss  %3, %%ymm4                  \n"

    // 16 pixel loop.
    LABELALIGN
  "1:                                          \n"
5470 5471
    "vpmovzxwd   " MEMACCESS(0) ",%%ymm2       \n"  // 16 shorts -> 16 ints
    "vpmovzxwd   " MEMACCESS2(0x10,0) ",%%ymm3 \n"
5472
    "lea         " MEMLEA(0x20,0) ",%0         \n"
5473 5474 5475 5476 5477 5478 5479 5480
    "vcvtdq2ps   %%ymm2,%%ymm2                 \n"
    "vcvtdq2ps   %%ymm3,%%ymm3                 \n"
    "vmulps      %%ymm2,%%ymm4,%%ymm2          \n"
    "vmulps      %%ymm3,%%ymm4,%%ymm3          \n"
    "vcvtps2ph   $3, %%ymm2, %%xmm2            \n"
    "vcvtps2ph   $3, %%ymm3, %%xmm3            \n"
    "vmovdqu     %%xmm2," MEMACCESS(1) "       \n"
    "vmovdqu     %%xmm3," MEMACCESS2(0x10,1) " \n"
5481 5482 5483
    "lea         " MEMLEA(0x20,1) ",%1         \n"
    "sub         $0x10,%2                      \n"
    "jg          1b                            \n"
5484

5485 5486 5487 5488 5489 5490
    "vzeroupper                                \n"
  : "+r"(src),   // %0
    "+r"(dst),   // %1
    "+r"(width)  // %2
  : "x"(scale)   // %3
  : "memory", "cc",
5491
    "xmm2", "xmm3", "xmm4"
5492 5493
  );
}
5494
//#endif  // HAS_HALFFLOATROW_F16C
5495

5496 5497 5498 5499 5500 5501 5502 5503 5504 5505 5506 5507 5508 5509 5510 5511 5512 5513 5514 5515 5516 5517 5518 5519 5520 5521 5522 5523 5524 5525
#ifdef HAS_HALFFLOATROW_F16C
void HalfFloat1Row_F16C(const uint16* src, uint16* dst, float, int width) {
  asm volatile (
    // 16 pixel loop.
    LABELALIGN
  "1:                                          \n"
    "vpmovzxwd   " MEMACCESS(0) ",%%ymm2       \n"  // 16 shorts -> 16 ints
    "vpmovzxwd   " MEMACCESS2(0x10,0) ",%%ymm3 \n"
    "lea         " MEMLEA(0x20,0) ",%0         \n"
    "vcvtdq2ps   %%ymm2,%%ymm2                 \n"
    "vcvtdq2ps   %%ymm3,%%ymm3                 \n"
    "vcvtps2ph   $3, %%ymm2, %%xmm2            \n"
    "vcvtps2ph   $3, %%ymm3, %%xmm3            \n"
    "vmovdqu     %%xmm2," MEMACCESS(1) "       \n"
    "vmovdqu     %%xmm3," MEMACCESS2(0x10,1) " \n"
    "lea         " MEMLEA(0x20,1) ",%1         \n"
    "sub         $0x10,%2                      \n"
    "jg          1b                            \n"

    "vzeroupper                                \n"
  : "+r"(src),   // %0
    "+r"(dst),   // %1
    "+r"(width)  // %2
  :
  : "memory", "cc",
    "xmm2", "xmm3"
  );
}
#endif  // HAS_HALFFLOATROW_F16C

5526 5527
#ifdef HAS_ARGBCOLORTABLEROW_X86
// Tranform ARGB pixels with color table.
5528 5529
void ARGBColorTableRow_X86(uint8* dst_argb,
                           const uint8* table_argb,
5530
                           int width) {
5531
  uintptr_t pixel_temp;
5532 5533
  asm volatile (
    // 1 pixel loop.
5534
    LABELALIGN
5535
  "1:                                          \n"
5536 5537
    "movzb     " MEMACCESS(0) ",%1             \n"
    "lea       " MEMLEA(0x4,0) ",%0            \n"
5538
    MEMOPARG(movzb,0x00,3,1,4,1) "             \n"  // movzb (%3,%1,4),%1
5539 5540
    "mov       %b1," MEMACCESS2(-0x4,0) "      \n"
    "movzb     " MEMACCESS2(-0x3,0) ",%1       \n"
5541
    MEMOPARG(movzb,0x01,3,1,4,1) "             \n"  // movzb 0x1(%3,%1,4),%1
5542 5543
    "mov       %b1," MEMACCESS2(-0x3,0) "      \n"
    "movzb     " MEMACCESS2(-0x2,0) ",%1       \n"
5544
    MEMOPARG(movzb,0x02,3,1,4,1) "             \n"  // movzb 0x2(%3,%1,4),%1
5545 5546
    "mov       %b1," MEMACCESS2(-0x2,0) "      \n"
    "movzb     " MEMACCESS2(-0x1,0) ",%1       \n"
5547
    MEMOPARG(movzb,0x03,3,1,4,1) "             \n"  // movzb 0x3(%3,%1,4),%1
5548
    "mov       %b1," MEMACCESS2(-0x1,0) "      \n"
5549 5550
    "dec       %2                              \n"
    "jg        1b                              \n"
5551 5552 5553 5554
  : "+r"(dst_argb),     // %0
    "=&d"(pixel_temp),  // %1
    "+r"(width)         // %2
  : "r"(table_argb)     // %3
5555 5556 5557 5558 5559 5560 5561
  : "memory", "cc");
}
#endif  // HAS_ARGBCOLORTABLEROW_X86

#ifdef HAS_RGBCOLORTABLEROW_X86
// Tranform RGB pixels with color table.
void RGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb, int width) {
5562
  uintptr_t pixel_temp;
5563 5564
  asm volatile (
    // 1 pixel loop.
5565
    LABELALIGN
5566
  "1:                                          \n"
5567 5568
    "movzb     " MEMACCESS(0) ",%1             \n"
    "lea       " MEMLEA(0x4,0) ",%0            \n"
5569
    MEMOPARG(movzb,0x00,3,1,4,1) "             \n"  // movzb (%3,%1,4),%1
5570 5571
    "mov       %b1," MEMACCESS2(-0x4,0) "      \n"
    "movzb     " MEMACCESS2(-0x3,0) ",%1       \n"
5572
    MEMOPARG(movzb,0x01,3,1,4,1) "             \n"  // movzb 0x1(%3,%1,4),%1
5573 5574
    "mov       %b1," MEMACCESS2(-0x3,0) "      \n"
    "movzb     " MEMACCESS2(-0x2,0) ",%1       \n"
5575
    MEMOPARG(movzb,0x02,3,1,4,1) "             \n"  // movzb 0x2(%3,%1,4),%1
5576
    "mov       %b1," MEMACCESS2(-0x2,0) "      \n"
5577 5578
    "dec       %2                              \n"
    "jg        1b                              \n"
5579 5580 5581 5582
  : "+r"(dst_argb),     // %0
    "=&d"(pixel_temp),  // %1
    "+r"(width)         // %2
  : "r"(table_argb)     // %3
5583 5584 5585 5586
  : "memory", "cc");
}
#endif  // HAS_RGBCOLORTABLEROW_X86

fbarchard@google.com's avatar
fbarchard@google.com committed
5587 5588
#ifdef HAS_ARGBLUMACOLORTABLEROW_SSSE3
// Tranform RGB pixels with luma table.
5589 5590
void ARGBLumaColorTableRow_SSSE3(const uint8* src_argb,
                                 uint8* dst_argb,
5591
                                 int width,
5592 5593
                                 const uint8* luma,
                                 uint32 lumacoeff) {
5594 5595
  uintptr_t pixel_temp;
  uintptr_t table_temp;
fbarchard@google.com's avatar
fbarchard@google.com committed
5596
  asm volatile (
5597
    "movd      %6,%%xmm3                       \n"
5598
    "pshufd    $0x0,%%xmm3,%%xmm3              \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5599 5600 5601 5602 5603
    "pcmpeqb   %%xmm4,%%xmm4                   \n"
    "psllw     $0x8,%%xmm4                     \n"
    "pxor      %%xmm5,%%xmm5                   \n"

    // 4 pixel loop.
5604
    LABELALIGN
fbarchard@google.com's avatar
fbarchard@google.com committed
5605
  "1:                                          \n"
5606
    "movdqu    " MEMACCESS(2) ",%%xmm0         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5607 5608 5609 5610 5611 5612 5613 5614
    "pmaddubsw %%xmm3,%%xmm0                   \n"
    "phaddw    %%xmm0,%%xmm0                   \n"
    "pand      %%xmm4,%%xmm0                   \n"
    "punpcklwd %%xmm5,%%xmm0                   \n"
    "movd      %%xmm0,%k1                      \n"  // 32 bit offset
    "add       %5,%1                           \n"
    "pshufd    $0x39,%%xmm0,%%xmm0             \n"

5615
    "movzb     " MEMACCESS(2) ",%0             \n"
5616
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5617 5618
    "mov       %b0," MEMACCESS(3) "            \n"
    "movzb     " MEMACCESS2(0x1,2) ",%0        \n"
5619
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5620 5621
    "mov       %b0," MEMACCESS2(0x1,3) "       \n"
    "movzb     " MEMACCESS2(0x2,2) ",%0        \n"
5622
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5623 5624 5625
    "mov       %b0," MEMACCESS2(0x2,3) "       \n"
    "movzb     " MEMACCESS2(0x3,2) ",%0        \n"
    "mov       %b0," MEMACCESS2(0x3,3) "       \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5626 5627 5628 5629 5630

    "movd      %%xmm0,%k1                      \n"  // 32 bit offset
    "add       %5,%1                           \n"
    "pshufd    $0x39,%%xmm0,%%xmm0             \n"

5631
    "movzb     " MEMACCESS2(0x4,2) ",%0        \n"
5632
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5633 5634
    "mov       %b0," MEMACCESS2(0x4,3) "       \n"
    "movzb     " MEMACCESS2(0x5,2) ",%0        \n"
5635
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5636 5637
    "mov       %b0," MEMACCESS2(0x5,3) "       \n"
    "movzb     " MEMACCESS2(0x6,2) ",%0        \n"
5638
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5639 5640 5641
    "mov       %b0," MEMACCESS2(0x6,3) "       \n"
    "movzb     " MEMACCESS2(0x7,2) ",%0        \n"
    "mov       %b0," MEMACCESS2(0x7,3) "       \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5642 5643 5644 5645 5646

    "movd      %%xmm0,%k1                      \n"  // 32 bit offset
    "add       %5,%1                           \n"
    "pshufd    $0x39,%%xmm0,%%xmm0             \n"

5647
    "movzb     " MEMACCESS2(0x8,2) ",%0        \n"
5648
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5649 5650
    "mov       %b0," MEMACCESS2(0x8,3) "       \n"
    "movzb     " MEMACCESS2(0x9,2) ",%0        \n"
5651
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5652 5653
    "mov       %b0," MEMACCESS2(0x9,3) "       \n"
    "movzb     " MEMACCESS2(0xa,2) ",%0        \n"
5654
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5655 5656 5657
    "mov       %b0," MEMACCESS2(0xa,3) "       \n"
    "movzb     " MEMACCESS2(0xb,2) ",%0        \n"
    "mov       %b0," MEMACCESS2(0xb,3) "       \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5658 5659 5660 5661

    "movd      %%xmm0,%k1                      \n"  // 32 bit offset
    "add       %5,%1                           \n"

5662
    "movzb     " MEMACCESS2(0xc,2) ",%0        \n"
5663
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5664 5665
    "mov       %b0," MEMACCESS2(0xc,3) "       \n"
    "movzb     " MEMACCESS2(0xd,2) ",%0        \n"
5666
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5667 5668
    "mov       %b0," MEMACCESS2(0xd,3) "       \n"
    "movzb     " MEMACCESS2(0xe,2) ",%0        \n"
5669
    MEMOPARG(movzb,0x00,1,0,1,0) "             \n"  // movzb     (%1,%0,1),%0
5670 5671 5672 5673 5674
    "mov       %b0," MEMACCESS2(0xe,3) "       \n"
    "movzb     " MEMACCESS2(0xf,2) ",%0        \n"
    "mov       %b0," MEMACCESS2(0xf,3) "       \n"
    "lea       " MEMLEA(0x10,2) ",%2           \n"
    "lea       " MEMLEA(0x10,3) ",%3           \n"
5675
    "sub       $0x4,%4                         \n"
fbarchard@google.com's avatar
fbarchard@google.com committed
5676
    "jg        1b                              \n"
5677 5678 5679 5680 5681 5682 5683
  : "=&d"(pixel_temp),  // %0
    "=&a"(table_temp),  // %1
    "+r"(src_argb),     // %2
    "+r"(dst_argb),     // %3
    "+rm"(width)        // %4
  : "r"(luma),          // %5
    "rm"(lumacoeff)     // %6
5684
  : "memory", "cc", "xmm0", "xmm3", "xmm4", "xmm5"
5685
  );
fbarchard@google.com's avatar
fbarchard@google.com committed
5686 5687 5688
}
#endif  // HAS_ARGBLUMACOLORTABLEROW_SSSE3

5689 5690
#endif  // defined(__x86_64__) || defined(__i386__)

5691
#ifdef __cplusplus
mikhal@webrtc.org's avatar
mikhal@webrtc.org committed
5692
}  // extern "C"
5693 5694
}  // namespace libyuv
#endif