1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libyuv/source/scale_posix.cc Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,1315 @@
1.4 +/*
1.5 + * Copyright 2013 The LibYuv Project Authors. All rights reserved.
1.6 + *
1.7 + * Use of this source code is governed by a BSD-style license
1.8 + * that can be found in the LICENSE file in the root of the source
1.9 + * tree. An additional intellectual property rights grant can be found
1.10 + * in the file PATENTS. All contributing project authors may
1.11 + * be found in the AUTHORS file in the root of the source tree.
1.12 + */
1.13 +
1.14 +#include "libyuv/row.h"
1.15 +
1.16 +#ifdef __cplusplus
1.17 +namespace libyuv {
1.18 +extern "C" {
1.19 +#endif
1.20 +
1.21 +// This module is for GCC x86 and x64.
1.22 +#if !defined(LIBYUV_DISABLE_X86) && (defined(__x86_64__) || defined(__i386__))
1.23 +
1.24 +// Offsets for source bytes 0 to 9
1.25 +static uvec8 kShuf0 =
1.26 + { 0, 1, 3, 4, 5, 7, 8, 9, 128, 128, 128, 128, 128, 128, 128, 128 };
1.27 +
1.28 +// Offsets for source bytes 11 to 20 with 8 subtracted = 3 to 12.
1.29 +static uvec8 kShuf1 =
1.30 + { 3, 4, 5, 7, 8, 9, 11, 12, 128, 128, 128, 128, 128, 128, 128, 128 };
1.31 +
1.32 +// Offsets for source bytes 21 to 31 with 16 subtracted = 5 to 31.
1.33 +static uvec8 kShuf2 =
1.34 + { 5, 7, 8, 9, 11, 12, 13, 15, 128, 128, 128, 128, 128, 128, 128, 128 };
1.35 +
1.36 +// Offsets for source bytes 0 to 10
1.37 +static uvec8 kShuf01 =
1.38 + { 0, 1, 1, 2, 2, 3, 4, 5, 5, 6, 6, 7, 8, 9, 9, 10 };
1.39 +
1.40 +// Offsets for source bytes 10 to 21 with 8 subtracted = 3 to 13.
1.41 +static uvec8 kShuf11 =
1.42 + { 2, 3, 4, 5, 5, 6, 6, 7, 8, 9, 9, 10, 10, 11, 12, 13 };
1.43 +
1.44 +// Offsets for source bytes 21 to 31 with 16 subtracted = 5 to 31.
1.45 +static uvec8 kShuf21 =
1.46 + { 5, 6, 6, 7, 8, 9, 9, 10, 10, 11, 12, 13, 13, 14, 14, 15 };
1.47 +
1.48 +// Coefficients for source bytes 0 to 10
1.49 +static uvec8 kMadd01 =
1.50 + { 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2 };
1.51 +
1.52 +// Coefficients for source bytes 10 to 21
1.53 +static uvec8 kMadd11 =
1.54 + { 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1 };
1.55 +
1.56 +// Coefficients for source bytes 21 to 31
1.57 +static uvec8 kMadd21 =
1.58 + { 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3 };
1.59 +
1.60 +// Coefficients for source bytes 21 to 31
1.61 +static vec16 kRound34 =
1.62 + { 2, 2, 2, 2, 2, 2, 2, 2 };
1.63 +
1.64 +static uvec8 kShuf38a =
1.65 + { 0, 3, 6, 8, 11, 14, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 };
1.66 +
1.67 +static uvec8 kShuf38b =
1.68 + { 128, 128, 128, 128, 128, 128, 0, 3, 6, 8, 11, 14, 128, 128, 128, 128 };
1.69 +
1.70 +// Arrange words 0,3,6 into 0,1,2
1.71 +static uvec8 kShufAc =
1.72 + { 0, 1, 6, 7, 12, 13, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 };
1.73 +
1.74 +// Arrange words 0,3,6 into 3,4,5
1.75 +static uvec8 kShufAc3 =
1.76 + { 128, 128, 128, 128, 128, 128, 0, 1, 6, 7, 12, 13, 128, 128, 128, 128 };
1.77 +
1.78 +// Scaling values for boxes of 3x3 and 2x3
1.79 +static uvec16 kScaleAc33 =
1.80 + { 65536 / 9, 65536 / 9, 65536 / 6, 65536 / 9, 65536 / 9, 65536 / 6, 0, 0 };
1.81 +
1.82 +// Arrange first value for pixels 0,1,2,3,4,5
1.83 +static uvec8 kShufAb0 =
1.84 + { 0, 128, 3, 128, 6, 128, 8, 128, 11, 128, 14, 128, 128, 128, 128, 128 };
1.85 +
1.86 +// Arrange second value for pixels 0,1,2,3,4,5
1.87 +static uvec8 kShufAb1 =
1.88 + { 1, 128, 4, 128, 7, 128, 9, 128, 12, 128, 15, 128, 128, 128, 128, 128 };
1.89 +
1.90 +// Arrange third value for pixels 0,1,2,3,4,5
1.91 +static uvec8 kShufAb2 =
1.92 + { 2, 128, 5, 128, 128, 128, 10, 128, 13, 128, 128, 128, 128, 128, 128, 128 };
1.93 +
1.94 +// Scaling values for boxes of 3x2 and 2x2
1.95 +static uvec16 kScaleAb2 =
1.96 + { 65536 / 3, 65536 / 3, 65536 / 2, 65536 / 3, 65536 / 3, 65536 / 2, 0, 0 };
1.97 +
1.98 +// GCC versions of row functions are verbatim conversions from Visual C.
1.99 +// Generated using gcc disassembly on Visual C object file:
1.100 +// objdump -D yuvscaler.obj >yuvscaler.txt
1.101 +
1.102 +void ScaleRowDown2_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
1.103 + uint8* dst_ptr, int dst_width) {
1.104 + asm volatile (
1.105 + LABELALIGN
1.106 + "1: \n"
1.107 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.108 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.109 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.110 + "psrlw $0x8,%%xmm0 \n"
1.111 + "psrlw $0x8,%%xmm1 \n"
1.112 + "packuswb %%xmm1,%%xmm0 \n"
1.113 + "movdqa %%xmm0," MEMACCESS(1) " \n"
1.114 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.115 + "sub $0x10,%2 \n"
1.116 + "jg 1b \n"
1.117 + : "+r"(src_ptr), // %0
1.118 + "+r"(dst_ptr), // %1
1.119 + "+r"(dst_width) // %2
1.120 + :
1.121 + : "memory", "cc"
1.122 +#if defined(__SSE2__)
1.123 + , "xmm0", "xmm1"
1.124 +#endif
1.125 + );
1.126 +}
1.127 +
1.128 +void ScaleRowDown2Linear_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
1.129 + uint8* dst_ptr, int dst_width) {
1.130 + asm volatile (
1.131 + "pcmpeqb %%xmm5,%%xmm5 \n"
1.132 + "psrlw $0x8,%%xmm5 \n"
1.133 +
1.134 + LABELALIGN
1.135 + "1: \n"
1.136 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.137 + "movdqa " MEMACCESS2(0x10, 0) ",%%xmm1 \n"
1.138 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.139 + "movdqa %%xmm0,%%xmm2 \n"
1.140 + "psrlw $0x8,%%xmm0 \n"
1.141 + "movdqa %%xmm1,%%xmm3 \n"
1.142 + "psrlw $0x8,%%xmm1 \n"
1.143 + "pand %%xmm5,%%xmm2 \n"
1.144 + "pand %%xmm5,%%xmm3 \n"
1.145 + "pavgw %%xmm2,%%xmm0 \n"
1.146 + "pavgw %%xmm3,%%xmm1 \n"
1.147 + "packuswb %%xmm1,%%xmm0 \n"
1.148 + "movdqa %%xmm0," MEMACCESS(1) " \n"
1.149 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.150 + "sub $0x10,%2 \n"
1.151 + "jg 1b \n"
1.152 + : "+r"(src_ptr), // %0
1.153 + "+r"(dst_ptr), // %1
1.154 + "+r"(dst_width) // %2
1.155 + :
1.156 + : "memory", "cc"
1.157 +#if defined(__SSE2__)
1.158 + , "xmm0", "xmm1", "xmm5"
1.159 +#endif
1.160 + );
1.161 +}
1.162 +
1.163 +void ScaleRowDown2Box_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
1.164 + uint8* dst_ptr, int dst_width) {
1.165 + asm volatile (
1.166 + "pcmpeqb %%xmm5,%%xmm5 \n"
1.167 + "psrlw $0x8,%%xmm5 \n"
1.168 +
1.169 + LABELALIGN
1.170 + "1: \n"
1.171 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.172 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.173 + MEMOPREG(movdqa,0x00,0,3,1,xmm2) // movdqa (%0,%3,1),%%xmm2
1.174 + BUNDLEALIGN
1.175 + MEMOPREG(movdqa,0x10,0,3,1,xmm3) // movdqa 0x10(%0,%3,1),%%xmm3
1.176 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.177 + "pavgb %%xmm2,%%xmm0 \n"
1.178 + "pavgb %%xmm3,%%xmm1 \n"
1.179 + "movdqa %%xmm0,%%xmm2 \n"
1.180 + "psrlw $0x8,%%xmm0 \n"
1.181 + "movdqa %%xmm1,%%xmm3 \n"
1.182 + "psrlw $0x8,%%xmm1 \n"
1.183 + "pand %%xmm5,%%xmm2 \n"
1.184 + "pand %%xmm5,%%xmm3 \n"
1.185 + "pavgw %%xmm2,%%xmm0 \n"
1.186 + "pavgw %%xmm3,%%xmm1 \n"
1.187 + "packuswb %%xmm1,%%xmm0 \n"
1.188 + "movdqa %%xmm0," MEMACCESS(1) " \n"
1.189 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.190 + "sub $0x10,%2 \n"
1.191 + "jg 1b \n"
1.192 + : "+r"(src_ptr), // %0
1.193 + "+r"(dst_ptr), // %1
1.194 + "+r"(dst_width) // %2
1.195 + : "r"((intptr_t)(src_stride)) // %3
1.196 + : "memory", "cc"
1.197 +#if defined(__native_client__) && defined(__x86_64__)
1.198 + , "r14"
1.199 +#endif
1.200 +#if defined(__SSE2__)
1.201 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
1.202 +#endif
1.203 + );
1.204 +}
1.205 +
1.206 +void ScaleRowDown2_Unaligned_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
1.207 + uint8* dst_ptr, int dst_width) {
1.208 + asm volatile (
1.209 + LABELALIGN
1.210 + "1: \n"
1.211 + "movdqu " MEMACCESS(0) ",%%xmm0 \n"
1.212 + "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.213 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.214 + "psrlw $0x8,%%xmm0 \n"
1.215 + "psrlw $0x8,%%xmm1 \n"
1.216 + "packuswb %%xmm1,%%xmm0 \n"
1.217 + "movdqu %%xmm0," MEMACCESS(1) " \n"
1.218 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.219 + "sub $0x10,%2 \n"
1.220 + "jg 1b \n"
1.221 + : "+r"(src_ptr), // %0
1.222 + "+r"(dst_ptr), // %1
1.223 + "+r"(dst_width) // %2
1.224 + :
1.225 + : "memory", "cc"
1.226 +#if defined(__SSE2__)
1.227 + , "xmm0", "xmm1"
1.228 +#endif
1.229 + );
1.230 +}
1.231 +
1.232 +void ScaleRowDown2Linear_Unaligned_SSE2(const uint8* src_ptr,
1.233 + ptrdiff_t src_stride,
1.234 + uint8* dst_ptr, int dst_width) {
1.235 + asm volatile (
1.236 + "pcmpeqb %%xmm5,%%xmm5 \n"
1.237 + "psrlw $0x8,%%xmm5 \n"
1.238 +
1.239 + LABELALIGN
1.240 + "1: \n"
1.241 + "movdqu " MEMACCESS(0) ",%%xmm0 \n"
1.242 + "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.243 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.244 + "movdqa %%xmm0,%%xmm2 \n"
1.245 + "psrlw $0x8,%%xmm0 \n"
1.246 + "movdqa %%xmm1,%%xmm3 \n"
1.247 + "psrlw $0x8,%%xmm1 \n"
1.248 + "pand %%xmm5,%%xmm2 \n"
1.249 + "pand %%xmm5,%%xmm3 \n"
1.250 + "pavgw %%xmm2,%%xmm0 \n"
1.251 + "pavgw %%xmm3,%%xmm1 \n"
1.252 + "packuswb %%xmm1,%%xmm0 \n"
1.253 + "movdqu %%xmm0," MEMACCESS(1) " \n"
1.254 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.255 + "sub $0x10,%2 \n"
1.256 + "jg 1b \n"
1.257 + : "+r"(src_ptr), // %0
1.258 + "+r"(dst_ptr), // %1
1.259 + "+r"(dst_width) // %2
1.260 + :
1.261 + : "memory", "cc"
1.262 +#if defined(__SSE2__)
1.263 + , "xmm0", "xmm1", "xmm5"
1.264 +#endif
1.265 + );
1.266 +}
1.267 +
1.268 +void ScaleRowDown2Box_Unaligned_SSE2(const uint8* src_ptr,
1.269 + ptrdiff_t src_stride,
1.270 + uint8* dst_ptr, int dst_width) {
1.271 + asm volatile (
1.272 + "pcmpeqb %%xmm5,%%xmm5 \n"
1.273 + "psrlw $0x8,%%xmm5 \n"
1.274 +
1.275 + LABELALIGN
1.276 + "1: \n"
1.277 + "movdqu " MEMACCESS(0) ",%%xmm0 \n"
1.278 + "movdqu " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.279 + MEMOPREG(movdqu,0x00,0,3,1,xmm2) // movdqu (%0,%3,1),%%xmm2
1.280 + BUNDLEALIGN
1.281 + MEMOPREG(movdqu,0x10,0,3,1,xmm3) // movdqu 0x10(%0,%3,1),%%xmm3
1.282 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.283 + "pavgb %%xmm2,%%xmm0 \n"
1.284 + "pavgb %%xmm3,%%xmm1 \n"
1.285 + "movdqa %%xmm0,%%xmm2 \n"
1.286 + "psrlw $0x8,%%xmm0 \n"
1.287 + "movdqa %%xmm1,%%xmm3 \n"
1.288 + "psrlw $0x8,%%xmm1 \n"
1.289 + "pand %%xmm5,%%xmm2 \n"
1.290 + "pand %%xmm5,%%xmm3 \n"
1.291 + "pavgw %%xmm2,%%xmm0 \n"
1.292 + "pavgw %%xmm3,%%xmm1 \n"
1.293 + "packuswb %%xmm1,%%xmm0 \n"
1.294 + "movdqu %%xmm0," MEMACCESS(1) " \n"
1.295 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.296 + "sub $0x10,%2 \n"
1.297 + "jg 1b \n"
1.298 + : "+r"(src_ptr), // %0
1.299 + "+r"(dst_ptr), // %1
1.300 + "+r"(dst_width) // %2
1.301 + : "r"((intptr_t)(src_stride)) // %3
1.302 + : "memory", "cc"
1.303 +#if defined(__native_client__) && defined(__x86_64__)
1.304 + , "r14"
1.305 +#endif
1.306 +#if defined(__SSE2__)
1.307 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
1.308 +#endif
1.309 + );
1.310 +}
1.311 +
1.312 +void ScaleRowDown4_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
1.313 + uint8* dst_ptr, int dst_width) {
1.314 + asm volatile (
1.315 + "pcmpeqb %%xmm5,%%xmm5 \n"
1.316 + "psrld $0x18,%%xmm5 \n"
1.317 + "pslld $0x10,%%xmm5 \n"
1.318 +
1.319 + LABELALIGN
1.320 + "1: \n"
1.321 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.322 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.323 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.324 + "pand %%xmm5,%%xmm0 \n"
1.325 + "pand %%xmm5,%%xmm1 \n"
1.326 + "packuswb %%xmm1,%%xmm0 \n"
1.327 + "psrlw $0x8,%%xmm0 \n"
1.328 + "packuswb %%xmm0,%%xmm0 \n"
1.329 + "movq %%xmm0," MEMACCESS(1) " \n"
1.330 + "lea " MEMLEA(0x8,1) ",%1 \n"
1.331 + "sub $0x8,%2 \n"
1.332 + "jg 1b \n"
1.333 + : "+r"(src_ptr), // %0
1.334 + "+r"(dst_ptr), // %1
1.335 + "+r"(dst_width) // %2
1.336 + :
1.337 + : "memory", "cc"
1.338 +#if defined(__SSE2__)
1.339 + , "xmm0", "xmm1", "xmm5"
1.340 +#endif
1.341 + );
1.342 +}
1.343 +
1.344 +void ScaleRowDown4Box_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
1.345 + uint8* dst_ptr, int dst_width) {
1.346 + intptr_t stridex3 = 0;
1.347 + asm volatile (
1.348 + "pcmpeqb %%xmm7,%%xmm7 \n"
1.349 + "psrlw $0x8,%%xmm7 \n"
1.350 + "lea " MEMLEA4(0x00,4,4,2) ",%3 \n"
1.351 +
1.352 + LABELALIGN
1.353 + "1: \n"
1.354 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.355 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.356 + MEMOPREG(movdqa,0x00,0,4,1,xmm2) // movdqa (%0,%4,1),%%xmm2
1.357 + BUNDLEALIGN
1.358 + MEMOPREG(movdqa,0x10,0,4,1,xmm3) // movdqa 0x10(%0,%4,1),%%xmm3
1.359 + "pavgb %%xmm2,%%xmm0 \n"
1.360 + "pavgb %%xmm3,%%xmm1 \n"
1.361 + MEMOPREG(movdqa,0x00,0,4,2,xmm2) // movdqa (%0,%4,2),%%xmm2
1.362 + BUNDLEALIGN
1.363 + MEMOPREG(movdqa,0x10,0,4,2,xmm3) // movdqa 0x10(%0,%4,2),%%xmm3
1.364 + MEMOPREG(movdqa,0x00,0,3,1,xmm4) // movdqa (%0,%3,1),%%xmm4
1.365 + MEMOPREG(movdqa,0x10,0,3,1,xmm5) // movdqa 0x10(%0,%3,1),%%xmm5
1.366 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.367 + "pavgb %%xmm4,%%xmm2 \n"
1.368 + "pavgb %%xmm2,%%xmm0 \n"
1.369 + "pavgb %%xmm5,%%xmm3 \n"
1.370 + "pavgb %%xmm3,%%xmm1 \n"
1.371 + "movdqa %%xmm0,%%xmm2 \n"
1.372 + "psrlw $0x8,%%xmm0 \n"
1.373 + "movdqa %%xmm1,%%xmm3 \n"
1.374 + "psrlw $0x8,%%xmm1 \n"
1.375 + "pand %%xmm7,%%xmm2 \n"
1.376 + "pand %%xmm7,%%xmm3 \n"
1.377 + "pavgw %%xmm2,%%xmm0 \n"
1.378 + "pavgw %%xmm3,%%xmm1 \n"
1.379 + "packuswb %%xmm1,%%xmm0 \n"
1.380 + "movdqa %%xmm0,%%xmm2 \n"
1.381 + "psrlw $0x8,%%xmm0 \n"
1.382 + "pand %%xmm7,%%xmm2 \n"
1.383 + "pavgw %%xmm2,%%xmm0 \n"
1.384 + "packuswb %%xmm0,%%xmm0 \n"
1.385 + "movq %%xmm0," MEMACCESS(1) " \n"
1.386 + "lea " MEMLEA(0x8,1) ",%1 \n"
1.387 + "sub $0x8,%2 \n"
1.388 + "jg 1b \n"
1.389 + : "+r"(src_ptr), // %0
1.390 + "+r"(dst_ptr), // %1
1.391 + "+r"(dst_width), // %2
1.392 + "+r"(stridex3) // %3
1.393 + : "r"((intptr_t)(src_stride)) // %4
1.394 + : "memory", "cc"
1.395 +#if defined(__native_client__) && defined(__x86_64__)
1.396 + , "r14"
1.397 +#endif
1.398 +#if defined(__SSE2__)
1.399 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm7"
1.400 +#endif
1.401 + );
1.402 +}
1.403 +
1.404 +void ScaleRowDown34_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
1.405 + uint8* dst_ptr, int dst_width) {
1.406 + asm volatile (
1.407 + "movdqa %0,%%xmm3 \n"
1.408 + "movdqa %1,%%xmm4 \n"
1.409 + "movdqa %2,%%xmm5 \n"
1.410 + :
1.411 + : "m"(kShuf0), // %0
1.412 + "m"(kShuf1), // %1
1.413 + "m"(kShuf2) // %2
1.414 + );
1.415 + asm volatile (
1.416 + LABELALIGN
1.417 + "1: \n"
1.418 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.419 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm2 \n"
1.420 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.421 + "movdqa %%xmm2,%%xmm1 \n"
1.422 + "palignr $0x8,%%xmm0,%%xmm1 \n"
1.423 + "pshufb %%xmm3,%%xmm0 \n"
1.424 + "pshufb %%xmm4,%%xmm1 \n"
1.425 + "pshufb %%xmm5,%%xmm2 \n"
1.426 + "movq %%xmm0," MEMACCESS(1) " \n"
1.427 + "movq %%xmm1," MEMACCESS2(0x8,1) " \n"
1.428 + "movq %%xmm2," MEMACCESS2(0x10,1) " \n"
1.429 + "lea " MEMLEA(0x18,1) ",%1 \n"
1.430 + "sub $0x18,%2 \n"
1.431 + "jg 1b \n"
1.432 + : "+r"(src_ptr), // %0
1.433 + "+r"(dst_ptr), // %1
1.434 + "+r"(dst_width) // %2
1.435 + :
1.436 + : "memory", "cc"
1.437 +#if defined(__SSE2__)
1.438 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
1.439 +#endif
1.440 + );
1.441 +}
1.442 +
1.443 +void ScaleRowDown34_1_Box_SSSE3(const uint8* src_ptr,
1.444 + ptrdiff_t src_stride,
1.445 + uint8* dst_ptr, int dst_width) {
1.446 + asm volatile (
1.447 + "movdqa %0,%%xmm2 \n" // kShuf01
1.448 + "movdqa %1,%%xmm3 \n" // kShuf11
1.449 + "movdqa %2,%%xmm4 \n" // kShuf21
1.450 + :
1.451 + : "m"(kShuf01), // %0
1.452 + "m"(kShuf11), // %1
1.453 + "m"(kShuf21) // %2
1.454 + );
1.455 + asm volatile (
1.456 + "movdqa %0,%%xmm5 \n" // kMadd01
1.457 + "movdqa %1,%%xmm0 \n" // kMadd11
1.458 + "movdqa %2,%%xmm1 \n" // kRound34
1.459 + :
1.460 + : "m"(kMadd01), // %0
1.461 + "m"(kMadd11), // %1
1.462 + "m"(kRound34) // %2
1.463 + );
1.464 + asm volatile (
1.465 + LABELALIGN
1.466 + "1: \n"
1.467 + "movdqa " MEMACCESS(0) ",%%xmm6 \n"
1.468 + MEMOPREG(movdqa,0x00,0,3,1,xmm7) // movdqa (%0,%3),%%xmm7
1.469 + "pavgb %%xmm7,%%xmm6 \n"
1.470 + "pshufb %%xmm2,%%xmm6 \n"
1.471 + "pmaddubsw %%xmm5,%%xmm6 \n"
1.472 + "paddsw %%xmm1,%%xmm6 \n"
1.473 + "psrlw $0x2,%%xmm6 \n"
1.474 + "packuswb %%xmm6,%%xmm6 \n"
1.475 + "movq %%xmm6," MEMACCESS(1) " \n"
1.476 + "movdqu " MEMACCESS2(0x8,0) ",%%xmm6 \n"
1.477 + MEMOPREG(movdqu,0x8,0,3,1,xmm7) // movdqu 0x8(%0,%3),%%xmm7
1.478 + "pavgb %%xmm7,%%xmm6 \n"
1.479 + "pshufb %%xmm3,%%xmm6 \n"
1.480 + "pmaddubsw %%xmm0,%%xmm6 \n"
1.481 + "paddsw %%xmm1,%%xmm6 \n"
1.482 + "psrlw $0x2,%%xmm6 \n"
1.483 + "packuswb %%xmm6,%%xmm6 \n"
1.484 + "movq %%xmm6," MEMACCESS2(0x8,1) " \n"
1.485 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm6 \n"
1.486 + BUNDLEALIGN
1.487 + MEMOPREG(movdqa,0x10,0,3,1,xmm7) // movdqa 0x10(%0,%3),%%xmm7
1.488 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.489 + "pavgb %%xmm7,%%xmm6 \n"
1.490 + "pshufb %%xmm4,%%xmm6 \n"
1.491 + "pmaddubsw %4,%%xmm6 \n"
1.492 + "paddsw %%xmm1,%%xmm6 \n"
1.493 + "psrlw $0x2,%%xmm6 \n"
1.494 + "packuswb %%xmm6,%%xmm6 \n"
1.495 + "movq %%xmm6," MEMACCESS2(0x10,1) " \n"
1.496 + "lea " MEMLEA(0x18,1) ",%1 \n"
1.497 + "sub $0x18,%2 \n"
1.498 + "jg 1b \n"
1.499 + : "+r"(src_ptr), // %0
1.500 + "+r"(dst_ptr), // %1
1.501 + "+r"(dst_width) // %2
1.502 + : "r"((intptr_t)(src_stride)), // %3
1.503 + "m"(kMadd21) // %4
1.504 + : "memory", "cc"
1.505 +#if defined(__native_client__) && defined(__x86_64__)
1.506 + , "r14"
1.507 +#endif
1.508 +#if defined(__SSE2__)
1.509 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
1.510 +#endif
1.511 + );
1.512 +}
1.513 +
1.514 +void ScaleRowDown34_0_Box_SSSE3(const uint8* src_ptr,
1.515 + ptrdiff_t src_stride,
1.516 + uint8* dst_ptr, int dst_width) {
1.517 + asm volatile (
1.518 + "movdqa %0,%%xmm2 \n" // kShuf01
1.519 + "movdqa %1,%%xmm3 \n" // kShuf11
1.520 + "movdqa %2,%%xmm4 \n" // kShuf21
1.521 + :
1.522 + : "m"(kShuf01), // %0
1.523 + "m"(kShuf11), // %1
1.524 + "m"(kShuf21) // %2
1.525 + );
1.526 + asm volatile (
1.527 + "movdqa %0,%%xmm5 \n" // kMadd01
1.528 + "movdqa %1,%%xmm0 \n" // kMadd11
1.529 + "movdqa %2,%%xmm1 \n" // kRound34
1.530 + :
1.531 + : "m"(kMadd01), // %0
1.532 + "m"(kMadd11), // %1
1.533 + "m"(kRound34) // %2
1.534 + );
1.535 +
1.536 + asm volatile (
1.537 + LABELALIGN
1.538 + "1: \n"
1.539 + "movdqa " MEMACCESS(0) ",%%xmm6 \n"
1.540 + MEMOPREG(movdqa,0x00,0,3,1,xmm7) // movdqa (%0,%3,1),%%xmm7
1.541 + "pavgb %%xmm6,%%xmm7 \n"
1.542 + "pavgb %%xmm7,%%xmm6 \n"
1.543 + "pshufb %%xmm2,%%xmm6 \n"
1.544 + "pmaddubsw %%xmm5,%%xmm6 \n"
1.545 + "paddsw %%xmm1,%%xmm6 \n"
1.546 + "psrlw $0x2,%%xmm6 \n"
1.547 + "packuswb %%xmm6,%%xmm6 \n"
1.548 + "movq %%xmm6," MEMACCESS(1) " \n"
1.549 + "movdqu " MEMACCESS2(0x8,0) ",%%xmm6 \n"
1.550 + MEMOPREG(movdqu,0x8,0,3,1,xmm7) // movdqu 0x8(%0,%3,1),%%xmm7
1.551 + "pavgb %%xmm6,%%xmm7 \n"
1.552 + "pavgb %%xmm7,%%xmm6 \n"
1.553 + "pshufb %%xmm3,%%xmm6 \n"
1.554 + "pmaddubsw %%xmm0,%%xmm6 \n"
1.555 + "paddsw %%xmm1,%%xmm6 \n"
1.556 + "psrlw $0x2,%%xmm6 \n"
1.557 + "packuswb %%xmm6,%%xmm6 \n"
1.558 + "movq %%xmm6," MEMACCESS2(0x8,1) " \n"
1.559 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm6 \n"
1.560 + MEMOPREG(movdqa,0x10,0,3,1,xmm7) // movdqa 0x10(%0,%3,1),%%xmm7
1.561 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.562 + "pavgb %%xmm6,%%xmm7 \n"
1.563 + "pavgb %%xmm7,%%xmm6 \n"
1.564 + "pshufb %%xmm4,%%xmm6 \n"
1.565 + "pmaddubsw %4,%%xmm6 \n"
1.566 + "paddsw %%xmm1,%%xmm6 \n"
1.567 + "psrlw $0x2,%%xmm6 \n"
1.568 + "packuswb %%xmm6,%%xmm6 \n"
1.569 + "movq %%xmm6," MEMACCESS2(0x10,1) " \n"
1.570 + "lea " MEMLEA(0x18,1) ",%1 \n"
1.571 + "sub $0x18,%2 \n"
1.572 + "jg 1b \n"
1.573 + : "+r"(src_ptr), // %0
1.574 + "+r"(dst_ptr), // %1
1.575 + "+r"(dst_width) // %2
1.576 + : "r"((intptr_t)(src_stride)), // %3
1.577 + "m"(kMadd21) // %4
1.578 + : "memory", "cc"
1.579 +#if defined(__native_client__) && defined(__x86_64__)
1.580 + , "r14"
1.581 +#endif
1.582 +#if defined(__SSE2__)
1.583 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
1.584 +#endif
1.585 + );
1.586 +}
1.587 +
1.588 +void ScaleRowDown38_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
1.589 + uint8* dst_ptr, int dst_width) {
1.590 + asm volatile (
1.591 + "movdqa %3,%%xmm4 \n"
1.592 + "movdqa %4,%%xmm5 \n"
1.593 +
1.594 + LABELALIGN
1.595 + "1: \n"
1.596 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.597 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.598 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.599 + "pshufb %%xmm4,%%xmm0 \n"
1.600 + "pshufb %%xmm5,%%xmm1 \n"
1.601 + "paddusb %%xmm1,%%xmm0 \n"
1.602 + "movq %%xmm0," MEMACCESS(1) " \n"
1.603 + "movhlps %%xmm0,%%xmm1 \n"
1.604 + "movd %%xmm1," MEMACCESS2(0x8,1) " \n"
1.605 + "lea " MEMLEA(0xc,1) ",%1 \n"
1.606 + "sub $0xc,%2 \n"
1.607 + "jg 1b \n"
1.608 + : "+r"(src_ptr), // %0
1.609 + "+r"(dst_ptr), // %1
1.610 + "+r"(dst_width) // %2
1.611 + : "m"(kShuf38a), // %3
1.612 + "m"(kShuf38b) // %4
1.613 + : "memory", "cc"
1.614 +#if defined(__SSE2__)
1.615 + , "xmm0", "xmm1", "xmm4", "xmm5"
1.616 +#endif
1.617 + );
1.618 +}
1.619 +
1.620 +void ScaleRowDown38_2_Box_SSSE3(const uint8* src_ptr,
1.621 + ptrdiff_t src_stride,
1.622 + uint8* dst_ptr, int dst_width) {
1.623 + asm volatile (
1.624 + "movdqa %0,%%xmm2 \n"
1.625 + "movdqa %1,%%xmm3 \n"
1.626 + "movdqa %2,%%xmm4 \n"
1.627 + "movdqa %3,%%xmm5 \n"
1.628 + :
1.629 + : "m"(kShufAb0), // %0
1.630 + "m"(kShufAb1), // %1
1.631 + "m"(kShufAb2), // %2
1.632 + "m"(kScaleAb2) // %3
1.633 + );
1.634 + asm volatile (
1.635 + LABELALIGN
1.636 + "1: \n"
1.637 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.638 + MEMOPREG(pavgb,0x00,0,3,1,xmm0) // pavgb (%0,%3,1),%%xmm0
1.639 + "lea " MEMLEA(0x10,0) ",%0 \n"
1.640 + "movdqa %%xmm0,%%xmm1 \n"
1.641 + "pshufb %%xmm2,%%xmm1 \n"
1.642 + "movdqa %%xmm0,%%xmm6 \n"
1.643 + "pshufb %%xmm3,%%xmm6 \n"
1.644 + "paddusw %%xmm6,%%xmm1 \n"
1.645 + "pshufb %%xmm4,%%xmm0 \n"
1.646 + "paddusw %%xmm0,%%xmm1 \n"
1.647 + "pmulhuw %%xmm5,%%xmm1 \n"
1.648 + "packuswb %%xmm1,%%xmm1 \n"
1.649 + "sub $0x6,%2 \n"
1.650 + "movd %%xmm1," MEMACCESS(1) " \n"
1.651 + "psrlq $0x10,%%xmm1 \n"
1.652 + "movd %%xmm1," MEMACCESS2(0x2,1) " \n"
1.653 + "lea " MEMLEA(0x6,1) ",%1 \n"
1.654 + "jg 1b \n"
1.655 + : "+r"(src_ptr), // %0
1.656 + "+r"(dst_ptr), // %1
1.657 + "+r"(dst_width) // %2
1.658 + : "r"((intptr_t)(src_stride)) // %3
1.659 + : "memory", "cc"
1.660 +#if defined(__native_client__) && defined(__x86_64__)
1.661 + , "r14"
1.662 +#endif
1.663 +#if defined(__SSE2__)
1.664 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
1.665 +#endif
1.666 + );
1.667 +}
1.668 +
1.669 +void ScaleRowDown38_3_Box_SSSE3(const uint8* src_ptr,
1.670 + ptrdiff_t src_stride,
1.671 + uint8* dst_ptr, int dst_width) {
1.672 + asm volatile (
1.673 + "movdqa %0,%%xmm2 \n"
1.674 + "movdqa %1,%%xmm3 \n"
1.675 + "movdqa %2,%%xmm4 \n"
1.676 + "pxor %%xmm5,%%xmm5 \n"
1.677 + :
1.678 + : "m"(kShufAc), // %0
1.679 + "m"(kShufAc3), // %1
1.680 + "m"(kScaleAc33) // %2
1.681 + );
1.682 + asm volatile (
1.683 + LABELALIGN
1.684 + "1: \n"
1.685 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.686 + MEMOPREG(movdqa,0x00,0,3,1,xmm6) // movdqa (%0,%3,1),%%xmm6
1.687 + "movhlps %%xmm0,%%xmm1 \n"
1.688 + "movhlps %%xmm6,%%xmm7 \n"
1.689 + "punpcklbw %%xmm5,%%xmm0 \n"
1.690 + "punpcklbw %%xmm5,%%xmm1 \n"
1.691 + "punpcklbw %%xmm5,%%xmm6 \n"
1.692 + "punpcklbw %%xmm5,%%xmm7 \n"
1.693 + "paddusw %%xmm6,%%xmm0 \n"
1.694 + "paddusw %%xmm7,%%xmm1 \n"
1.695 + MEMOPREG(movdqa,0x00,0,3,2,xmm6) // movdqa (%0,%3,2),%%xmm6
1.696 + "lea " MEMLEA(0x10,0) ",%0 \n"
1.697 + "movhlps %%xmm6,%%xmm7 \n"
1.698 + "punpcklbw %%xmm5,%%xmm6 \n"
1.699 + "punpcklbw %%xmm5,%%xmm7 \n"
1.700 + "paddusw %%xmm6,%%xmm0 \n"
1.701 + "paddusw %%xmm7,%%xmm1 \n"
1.702 + "movdqa %%xmm0,%%xmm6 \n"
1.703 + "psrldq $0x2,%%xmm0 \n"
1.704 + "paddusw %%xmm0,%%xmm6 \n"
1.705 + "psrldq $0x2,%%xmm0 \n"
1.706 + "paddusw %%xmm0,%%xmm6 \n"
1.707 + "pshufb %%xmm2,%%xmm6 \n"
1.708 + "movdqa %%xmm1,%%xmm7 \n"
1.709 + "psrldq $0x2,%%xmm1 \n"
1.710 + "paddusw %%xmm1,%%xmm7 \n"
1.711 + "psrldq $0x2,%%xmm1 \n"
1.712 + "paddusw %%xmm1,%%xmm7 \n"
1.713 + "pshufb %%xmm3,%%xmm7 \n"
1.714 + "paddusw %%xmm7,%%xmm6 \n"
1.715 + "pmulhuw %%xmm4,%%xmm6 \n"
1.716 + "packuswb %%xmm6,%%xmm6 \n"
1.717 + "sub $0x6,%2 \n"
1.718 + "movd %%xmm6," MEMACCESS(1) " \n"
1.719 + "psrlq $0x10,%%xmm6 \n"
1.720 + "movd %%xmm6," MEMACCESS2(0x2,1) " \n"
1.721 + "lea " MEMLEA(0x6,1) ",%1 \n"
1.722 + "jg 1b \n"
1.723 + : "+r"(src_ptr), // %0
1.724 + "+r"(dst_ptr), // %1
1.725 + "+r"(dst_width) // %2
1.726 + : "r"((intptr_t)(src_stride)) // %3
1.727 + : "memory", "cc"
1.728 +#if defined(__native_client__) && defined(__x86_64__)
1.729 + , "r14"
1.730 +#endif
1.731 +#if defined(__SSE2__)
1.732 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
1.733 +#endif
1.734 + );
1.735 +}
1.736 +
1.737 +void ScaleAddRows_SSE2(const uint8* src_ptr, ptrdiff_t src_stride,
1.738 + uint16* dst_ptr, int src_width, int src_height) {
1.739 + int tmp_height = 0;
1.740 + intptr_t tmp_src = 0;
1.741 + asm volatile (
1.742 + "pxor %%xmm4,%%xmm4 \n"
1.743 + "sub $0x1,%5 \n"
1.744 +
1.745 + LABELALIGN
1.746 + "1: \n"
1.747 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.748 + "mov %0,%3 \n"
1.749 + "add %6,%0 \n"
1.750 + "movdqa %%xmm0,%%xmm1 \n"
1.751 + "punpcklbw %%xmm4,%%xmm0 \n"
1.752 + "punpckhbw %%xmm4,%%xmm1 \n"
1.753 + "mov %5,%2 \n"
1.754 + "test %2,%2 \n"
1.755 + "je 3f \n"
1.756 +
1.757 + LABELALIGN
1.758 + "2: \n"
1.759 + "movdqa " MEMACCESS(0) ",%%xmm2 \n"
1.760 + "add %6,%0 \n"
1.761 + "movdqa %%xmm2,%%xmm3 \n"
1.762 + "punpcklbw %%xmm4,%%xmm2 \n"
1.763 + "punpckhbw %%xmm4,%%xmm3 \n"
1.764 + "paddusw %%xmm2,%%xmm0 \n"
1.765 + "paddusw %%xmm3,%%xmm1 \n"
1.766 + "sub $0x1,%2 \n"
1.767 + "jg 2b \n"
1.768 +
1.769 + LABELALIGN
1.770 + "3: \n"
1.771 + "movdqa %%xmm0," MEMACCESS(1) " \n"
1.772 + "movdqa %%xmm1," MEMACCESS2(0x10,1) " \n"
1.773 + "lea " MEMLEA(0x10,3) ",%0 \n"
1.774 + "lea " MEMLEA(0x20,1) ",%1 \n"
1.775 + "sub $0x10,%4 \n"
1.776 + "jg 1b \n"
1.777 + : "+r"(src_ptr), // %0
1.778 + "+r"(dst_ptr), // %1
1.779 + "+r"(tmp_height), // %2
1.780 + "+r"(tmp_src), // %3
1.781 + "+r"(src_width), // %4
1.782 + "+rm"(src_height) // %5
1.783 + : "rm"((intptr_t)(src_stride)) // %6
1.784 + : "memory", "cc"
1.785 +#if defined(__SSE2__)
1.786 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4"
1.787 +#endif
1.788 + );
1.789 +}
1.790 +
1.791 +// Bilinear column filtering. SSSE3 version.
1.792 +void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
1.793 + int dst_width, int x, int dx) {
1.794 + intptr_t x0 = 0, x1 = 0, temp_pixel = 0;
1.795 + asm volatile (
1.796 + "movd %6,%%xmm2 \n"
1.797 + "movd %7,%%xmm3 \n"
1.798 + "movl $0x04040000,%k2 \n"
1.799 + "movd %k2,%%xmm5 \n"
1.800 + "pcmpeqb %%xmm6,%%xmm6 \n"
1.801 + "psrlw $0x9,%%xmm6 \n"
1.802 + "pextrw $0x1,%%xmm2,%k3 \n"
1.803 + "subl $0x2,%5 \n"
1.804 + "jl 29f \n"
1.805 + "movdqa %%xmm2,%%xmm0 \n"
1.806 + "paddd %%xmm3,%%xmm0 \n"
1.807 + "punpckldq %%xmm0,%%xmm2 \n"
1.808 + "punpckldq %%xmm3,%%xmm3 \n"
1.809 + "paddd %%xmm3,%%xmm3 \n"
1.810 + "pextrw $0x3,%%xmm2,%k4 \n"
1.811 +
1.812 + LABELALIGN
1.813 + "2: \n"
1.814 + "movdqa %%xmm2,%%xmm1 \n"
1.815 + "paddd %%xmm3,%%xmm2 \n"
1.816 + MEMOPARG(movzwl,0x00,1,3,1,k2) // movzwl (%1,%3,1),%k2
1.817 + "movd %k2,%%xmm0 \n"
1.818 + "psrlw $0x9,%%xmm1 \n"
1.819 + BUNDLEALIGN
1.820 + MEMOPARG(movzwl,0x00,1,4,1,k2) // movzwl (%1,%4,1),%k2
1.821 + "movd %k2,%%xmm4 \n"
1.822 + "pshufb %%xmm5,%%xmm1 \n"
1.823 + "punpcklwd %%xmm4,%%xmm0 \n"
1.824 + "pxor %%xmm6,%%xmm1 \n"
1.825 + "pmaddubsw %%xmm1,%%xmm0 \n"
1.826 + "pextrw $0x1,%%xmm2,%k3 \n"
1.827 + "pextrw $0x3,%%xmm2,%k4 \n"
1.828 + "psrlw $0x7,%%xmm0 \n"
1.829 + "packuswb %%xmm0,%%xmm0 \n"
1.830 + "movd %%xmm0,%k2 \n"
1.831 + "mov %w2," MEMACCESS(0) " \n"
1.832 + "lea " MEMLEA(0x2,0) ",%0 \n"
1.833 + "sub $0x2,%5 \n"
1.834 + "jge 2b \n"
1.835 +
1.836 + LABELALIGN
1.837 + "29: \n"
1.838 + "addl $0x1,%5 \n"
1.839 + "jl 99f \n"
1.840 + MEMOPARG(movzwl,0x00,1,3,1,k2) // movzwl (%1,%3,1),%k2
1.841 + "movd %k2,%%xmm0 \n"
1.842 + "psrlw $0x9,%%xmm2 \n"
1.843 + "pshufb %%xmm5,%%xmm2 \n"
1.844 + "pxor %%xmm6,%%xmm2 \n"
1.845 + "pmaddubsw %%xmm2,%%xmm0 \n"
1.846 + "psrlw $0x7,%%xmm0 \n"
1.847 + "packuswb %%xmm0,%%xmm0 \n"
1.848 + "movd %%xmm0,%k2 \n"
1.849 + "mov %b2," MEMACCESS(0) " \n"
1.850 + "99: \n"
1.851 + : "+r"(dst_ptr), // %0
1.852 + "+r"(src_ptr), // %1
1.853 + "+a"(temp_pixel), // %2
1.854 + "+r"(x0), // %3
1.855 + "+r"(x1), // %4
1.856 + "+rm"(dst_width) // %5
1.857 + : "rm"(x), // %6
1.858 + "rm"(dx) // %7
1.859 + : "memory", "cc"
1.860 +#if defined(__native_client__) && defined(__x86_64__)
1.861 + , "r14"
1.862 +#endif
1.863 +#if defined(__SSE2__)
1.864 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
1.865 +#endif
1.866 + );
1.867 +}
1.868 +
1.869 +// Reads 4 pixels, duplicates them and writes 8 pixels.
1.870 +// Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned.
1.871 +void ScaleColsUp2_SSE2(uint8* dst_ptr, const uint8* src_ptr,
1.872 + int dst_width, int x, int dx) {
1.873 + asm volatile (
1.874 + LABELALIGN
1.875 + "1: \n"
1.876 + "movdqa " MEMACCESS(1) ",%%xmm0 \n"
1.877 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.878 + "movdqa %%xmm0,%%xmm1 \n"
1.879 + "punpcklbw %%xmm0,%%xmm0 \n"
1.880 + "punpckhbw %%xmm1,%%xmm1 \n"
1.881 + "sub $0x20,%2 \n"
1.882 + "movdqa %%xmm0," MEMACCESS(0) " \n"
1.883 + "movdqa %%xmm1," MEMACCESS2(0x10,0) " \n"
1.884 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.885 + "jg 1b \n"
1.886 +
1.887 + : "+r"(dst_ptr), // %0
1.888 + "+r"(src_ptr), // %1
1.889 + "+r"(dst_width) // %2
1.890 + :
1.891 + : "memory", "cc"
1.892 +#if defined(__SSE2__)
1.893 + , "xmm0", "xmm1"
1.894 +#endif
1.895 + );
1.896 +}
1.897 +
1.898 +void ScaleARGBRowDown2_SSE2(const uint8* src_argb,
1.899 + ptrdiff_t src_stride,
1.900 + uint8* dst_argb, int dst_width) {
1.901 + asm volatile (
1.902 + LABELALIGN
1.903 + "1: \n"
1.904 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.905 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.906 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.907 + "shufps $0xdd,%%xmm1,%%xmm0 \n"
1.908 + "sub $0x4,%2 \n"
1.909 + "movdqa %%xmm0," MEMACCESS(1) " \n"
1.910 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.911 + "jg 1b \n"
1.912 + : "+r"(src_argb), // %0
1.913 + "+r"(dst_argb), // %1
1.914 + "+r"(dst_width) // %2
1.915 + :
1.916 + : "memory", "cc"
1.917 +#if defined(__SSE2__)
1.918 + , "xmm0", "xmm1"
1.919 +#endif
1.920 + );
1.921 +}
1.922 +
1.923 +void ScaleARGBRowDown2Linear_SSE2(const uint8* src_argb,
1.924 + ptrdiff_t src_stride,
1.925 + uint8* dst_argb, int dst_width) {
1.926 + asm volatile (
1.927 + LABELALIGN
1.928 + "1: \n"
1.929 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.930 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.931 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.932 + "movdqa %%xmm0,%%xmm2 \n"
1.933 + "shufps $0x88,%%xmm1,%%xmm0 \n"
1.934 + "shufps $0xdd,%%xmm1,%%xmm2 \n"
1.935 + "pavgb %%xmm2,%%xmm0 \n"
1.936 + "sub $0x4,%2 \n"
1.937 + "movdqa %%xmm0," MEMACCESS(1) " \n"
1.938 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.939 + "jg 1b \n"
1.940 + : "+r"(src_argb), // %0
1.941 + "+r"(dst_argb), // %1
1.942 + "+r"(dst_width) // %2
1.943 + :
1.944 + : "memory", "cc"
1.945 +#if defined(__SSE2__)
1.946 + , "xmm0", "xmm1"
1.947 +#endif
1.948 + );
1.949 +}
1.950 +
1.951 +void ScaleARGBRowDown2Box_SSE2(const uint8* src_argb,
1.952 + ptrdiff_t src_stride,
1.953 + uint8* dst_argb, int dst_width) {
1.954 + asm volatile (
1.955 + LABELALIGN
1.956 + "1: \n"
1.957 + "movdqa " MEMACCESS(0) ",%%xmm0 \n"
1.958 + "movdqa " MEMACCESS2(0x10,0) ",%%xmm1 \n"
1.959 + BUNDLEALIGN
1.960 + MEMOPREG(movdqa,0x00,0,3,1,xmm2) // movdqa (%0,%3,1),%%xmm2
1.961 + MEMOPREG(movdqa,0x10,0,3,1,xmm3) // movdqa 0x10(%0,%3,1),%%xmm3
1.962 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.963 + "pavgb %%xmm2,%%xmm0 \n"
1.964 + "pavgb %%xmm3,%%xmm1 \n"
1.965 + "movdqa %%xmm0,%%xmm2 \n"
1.966 + "shufps $0x88,%%xmm1,%%xmm0 \n"
1.967 + "shufps $0xdd,%%xmm1,%%xmm2 \n"
1.968 + "pavgb %%xmm2,%%xmm0 \n"
1.969 + "sub $0x4,%2 \n"
1.970 + "movdqa %%xmm0," MEMACCESS(1) " \n"
1.971 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.972 + "jg 1b \n"
1.973 + : "+r"(src_argb), // %0
1.974 + "+r"(dst_argb), // %1
1.975 + "+r"(dst_width) // %2
1.976 + : "r"((intptr_t)(src_stride)) // %3
1.977 + : "memory", "cc"
1.978 +#if defined(__native_client__) && defined(__x86_64__)
1.979 + , "r14"
1.980 +#endif
1.981 +#if defined(__SSE2__)
1.982 + , "xmm0", "xmm1", "xmm2", "xmm3"
1.983 +#endif
1.984 + );
1.985 +}
1.986 +
1.987 +// Reads 4 pixels at a time.
1.988 +// Alignment requirement: dst_argb 16 byte aligned.
1.989 +void ScaleARGBRowDownEven_SSE2(const uint8* src_argb, ptrdiff_t src_stride,
1.990 + int src_stepx,
1.991 + uint8* dst_argb, int dst_width) {
1.992 + intptr_t src_stepx_x4 = (intptr_t)(src_stepx);
1.993 + intptr_t src_stepx_x12 = 0;
1.994 + asm volatile (
1.995 + "lea " MEMLEA3(0x00,1,4) ",%1 \n"
1.996 + "lea " MEMLEA4(0x00,1,1,2) ",%4 \n"
1.997 + LABELALIGN
1.998 + "1: \n"
1.999 + "movd " MEMACCESS(0) ",%%xmm0 \n"
1.1000 + MEMOPREG(movd,0x00,0,1,1,xmm1) // movd (%0,%1,1),%%xmm1
1.1001 + "punpckldq %%xmm1,%%xmm0 \n"
1.1002 + BUNDLEALIGN
1.1003 + MEMOPREG(movd,0x00,0,1,2,xmm2) // movd (%0,%1,2),%%xmm2
1.1004 + MEMOPREG(movd,0x00,0,4,1,xmm3) // movd (%0,%4,1),%%xmm3
1.1005 + "lea " MEMLEA4(0x00,0,1,4) ",%0 \n"
1.1006 + "punpckldq %%xmm3,%%xmm2 \n"
1.1007 + "punpcklqdq %%xmm2,%%xmm0 \n"
1.1008 + "sub $0x4,%3 \n"
1.1009 + "movdqa %%xmm0," MEMACCESS(2) " \n"
1.1010 + "lea " MEMLEA(0x10,2) ",%2 \n"
1.1011 + "jg 1b \n"
1.1012 + : "+r"(src_argb), // %0
1.1013 + "+r"(src_stepx_x4), // %1
1.1014 + "+r"(dst_argb), // %2
1.1015 + "+r"(dst_width), // %3
1.1016 + "+r"(src_stepx_x12) // %4
1.1017 + :
1.1018 + : "memory", "cc"
1.1019 +#if defined(__native_client__) && defined(__x86_64__)
1.1020 + , "r14"
1.1021 +#endif
1.1022 +#if defined(__SSE2__)
1.1023 + , "xmm0", "xmm1", "xmm2", "xmm3"
1.1024 +#endif
1.1025 + );
1.1026 +}
1.1027 +
1.1028 +// Blends four 2x2 to 4x1.
1.1029 +// Alignment requirement: dst_argb 16 byte aligned.
1.1030 +void ScaleARGBRowDownEvenBox_SSE2(const uint8* src_argb,
1.1031 + ptrdiff_t src_stride, int src_stepx,
1.1032 + uint8* dst_argb, int dst_width) {
1.1033 + intptr_t src_stepx_x4 = (intptr_t)(src_stepx);
1.1034 + intptr_t src_stepx_x12 = 0;
1.1035 + intptr_t row1 = (intptr_t)(src_stride);
1.1036 + asm volatile (
1.1037 + "lea " MEMLEA3(0x00,1,4) ",%1 \n"
1.1038 + "lea " MEMLEA4(0x00,1,1,2) ",%4 \n"
1.1039 + "lea " MEMLEA4(0x00,0,5,1) ",%5 \n"
1.1040 +
1.1041 + LABELALIGN
1.1042 + "1: \n"
1.1043 + "movq " MEMACCESS(0) ",%%xmm0 \n"
1.1044 + MEMOPREG(movhps,0x00,0,1,1,xmm0) // movhps (%0,%1,1),%%xmm0
1.1045 + MEMOPREG(movq,0x00,0,1,2,xmm1) // movq (%0,%1,2),%%xmm1
1.1046 + BUNDLEALIGN
1.1047 + MEMOPREG(movhps,0x00,0,4,1,xmm1) // movhps (%0,%4,1),%%xmm1
1.1048 + "lea " MEMLEA4(0x00,0,1,4) ",%0 \n"
1.1049 + "movq " MEMACCESS(5) ",%%xmm2 \n"
1.1050 + BUNDLEALIGN
1.1051 + MEMOPREG(movhps,0x00,5,1,1,xmm2) // movhps (%5,%1,1),%%xmm2
1.1052 + MEMOPREG(movq,0x00,5,1,2,xmm3) // movq (%5,%1,2),%%xmm3
1.1053 + MEMOPREG(movhps,0x00,5,4,1,xmm3) // movhps (%5,%4,1),%%xmm3
1.1054 + "lea " MEMLEA4(0x00,5,1,4) ",%5 \n"
1.1055 + "pavgb %%xmm2,%%xmm0 \n"
1.1056 + "pavgb %%xmm3,%%xmm1 \n"
1.1057 + "movdqa %%xmm0,%%xmm2 \n"
1.1058 + "shufps $0x88,%%xmm1,%%xmm0 \n"
1.1059 + "shufps $0xdd,%%xmm1,%%xmm2 \n"
1.1060 + "pavgb %%xmm2,%%xmm0 \n"
1.1061 + "sub $0x4,%3 \n"
1.1062 + "movdqa %%xmm0," MEMACCESS(2) " \n"
1.1063 + "lea " MEMLEA(0x10,2) ",%2 \n"
1.1064 + "jg 1b \n"
1.1065 + : "+r"(src_argb), // %0
1.1066 + "+r"(src_stepx_x4), // %1
1.1067 + "+r"(dst_argb), // %2
1.1068 + "+rm"(dst_width), // %3
1.1069 + "+r"(src_stepx_x12), // %4
1.1070 + "+r"(row1) // %5
1.1071 + :
1.1072 + : "memory", "cc"
1.1073 +#if defined(__native_client__) && defined(__x86_64__)
1.1074 + , "r14"
1.1075 +#endif
1.1076 +#if defined(__SSE2__)
1.1077 + , "xmm0", "xmm1", "xmm2", "xmm3"
1.1078 +#endif
1.1079 + );
1.1080 +}
1.1081 +
1.1082 +void ScaleARGBCols_SSE2(uint8* dst_argb, const uint8* src_argb,
1.1083 + int dst_width, int x, int dx) {
1.1084 + intptr_t x0 = 0, x1 = 0;
1.1085 + asm volatile (
1.1086 + "movd %5,%%xmm2 \n"
1.1087 + "movd %6,%%xmm3 \n"
1.1088 + "pshufd $0x0,%%xmm2,%%xmm2 \n"
1.1089 + "pshufd $0x11,%%xmm3,%%xmm0 \n"
1.1090 + "paddd %%xmm0,%%xmm2 \n"
1.1091 + "paddd %%xmm3,%%xmm3 \n"
1.1092 + "pshufd $0x5,%%xmm3,%%xmm0 \n"
1.1093 + "paddd %%xmm0,%%xmm2 \n"
1.1094 + "paddd %%xmm3,%%xmm3 \n"
1.1095 + "pshufd $0x0,%%xmm3,%%xmm3 \n"
1.1096 + "pextrw $0x1,%%xmm2,%k0 \n"
1.1097 + "pextrw $0x3,%%xmm2,%k1 \n"
1.1098 + "cmp $0x0,%4 \n"
1.1099 + "jl 99f \n"
1.1100 + "sub $0x4,%4 \n"
1.1101 + "jl 49f \n"
1.1102 +
1.1103 + LABELALIGN
1.1104 + "40: \n"
1.1105 + MEMOPREG(movd,0x00,3,0,4,xmm0) // movd (%3,%0,4),%%xmm0
1.1106 + MEMOPREG(movd,0x00,3,1,4,xmm1) // movd (%3,%1,4),%%xmm1
1.1107 + "pextrw $0x5,%%xmm2,%k0 \n"
1.1108 + "pextrw $0x7,%%xmm2,%k1 \n"
1.1109 + "paddd %%xmm3,%%xmm2 \n"
1.1110 + "punpckldq %%xmm1,%%xmm0 \n"
1.1111 + MEMOPREG(movd,0x00,3,0,4,xmm1) // movd (%3,%0,4),%%xmm1
1.1112 + MEMOPREG(movd,0x00,3,1,4,xmm4) // movd (%3,%1,4),%%xmm4
1.1113 + "pextrw $0x1,%%xmm2,%k0 \n"
1.1114 + "pextrw $0x3,%%xmm2,%k1 \n"
1.1115 + "punpckldq %%xmm4,%%xmm1 \n"
1.1116 + "punpcklqdq %%xmm1,%%xmm0 \n"
1.1117 + "sub $0x4,%4 \n"
1.1118 + "movdqu %%xmm0," MEMACCESS(2) " \n"
1.1119 + "lea " MEMLEA(0x10,2) ",%2 \n"
1.1120 + "jge 40b \n"
1.1121 +
1.1122 + "49: \n"
1.1123 + "test $0x2,%4 \n"
1.1124 + "je 29f \n"
1.1125 + BUNDLEALIGN
1.1126 + MEMOPREG(movd,0x00,3,0,4,xmm0) // movd (%3,%0,4),%%xmm0
1.1127 + MEMOPREG(movd,0x00,3,1,4,xmm1) // movd (%3,%1,4),%%xmm1
1.1128 + "pextrw $0x5,%%xmm2,%k0 \n"
1.1129 + "punpckldq %%xmm1,%%xmm0 \n"
1.1130 + "movq %%xmm0," MEMACCESS(2) " \n"
1.1131 + "lea " MEMLEA(0x8,2) ",%2 \n"
1.1132 + "29: \n"
1.1133 + "test $0x1,%4 \n"
1.1134 + "je 99f \n"
1.1135 + MEMOPREG(movd,0x00,3,0,4,xmm0) // movd (%3,%0,4),%%xmm0
1.1136 + "movd %%xmm0," MEMACCESS(2) " \n"
1.1137 + "99: \n"
1.1138 + : "+a"(x0), // %0
1.1139 + "+d"(x1), // %1
1.1140 + "+r"(dst_argb), // %2
1.1141 + "+r"(src_argb), // %3
1.1142 + "+r"(dst_width) // %4
1.1143 + : "rm"(x), // %5
1.1144 + "rm"(dx) // %6
1.1145 + : "memory", "cc"
1.1146 +#if defined(__native_client__) && defined(__x86_64__)
1.1147 + , "r14"
1.1148 +#endif
1.1149 +#if defined(__SSE2__)
1.1150 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4"
1.1151 +#endif
1.1152 + );
1.1153 +}
1.1154 +
1.1155 +// Reads 4 pixels, duplicates them and writes 8 pixels.
1.1156 +// Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned.
1.1157 +void ScaleARGBColsUp2_SSE2(uint8* dst_argb, const uint8* src_argb,
1.1158 + int dst_width, int x, int dx) {
1.1159 + asm volatile (
1.1160 + LABELALIGN
1.1161 + "1: \n"
1.1162 + "movdqa " MEMACCESS(1) ",%%xmm0 \n"
1.1163 + "lea " MEMLEA(0x10,1) ",%1 \n"
1.1164 + "movdqa %%xmm0,%%xmm1 \n"
1.1165 + "punpckldq %%xmm0,%%xmm0 \n"
1.1166 + "punpckhdq %%xmm1,%%xmm1 \n"
1.1167 + "sub $0x8,%2 \n"
1.1168 + "movdqa %%xmm0," MEMACCESS(0) " \n"
1.1169 + "movdqa %%xmm1," MEMACCESS2(0x10,0) " \n"
1.1170 + "lea " MEMLEA(0x20,0) ",%0 \n"
1.1171 + "jg 1b \n"
1.1172 +
1.1173 + : "+r"(dst_argb), // %0
1.1174 + "+r"(src_argb), // %1
1.1175 + "+r"(dst_width) // %2
1.1176 + :
1.1177 + : "memory", "cc"
1.1178 +#if defined(__native_client__) && defined(__x86_64__)
1.1179 + , "r14"
1.1180 +#endif
1.1181 +#if defined(__SSE2__)
1.1182 + , "xmm0", "xmm1"
1.1183 +#endif
1.1184 + );
1.1185 +}
1.1186 +
1.1187 +// Shuffle table for arranging 2 pixels into pairs for pmaddubsw
1.1188 +static uvec8 kShuffleColARGB = {
1.1189 + 0u, 4u, 1u, 5u, 2u, 6u, 3u, 7u, // bbggrraa 1st pixel
1.1190 + 8u, 12u, 9u, 13u, 10u, 14u, 11u, 15u // bbggrraa 2nd pixel
1.1191 +};
1.1192 +
1.1193 +// Shuffle table for duplicating 2 fractions into 8 bytes each
1.1194 +static uvec8 kShuffleFractions = {
1.1195 + 0u, 0u, 0u, 0u, 0u, 0u, 0u, 0u, 4u, 4u, 4u, 4u, 4u, 4u, 4u, 4u,
1.1196 +};
1.1197 +
1.1198 +// Bilinear row filtering combines 4x2 -> 4x1. SSSE3 version
1.1199 +void ScaleARGBFilterCols_SSSE3(uint8* dst_argb, const uint8* src_argb,
1.1200 + int dst_width, int x, int dx) {
1.1201 + intptr_t x0 = 0, x1 = 0;
1.1202 + asm volatile (
1.1203 + "movdqa %0,%%xmm4 \n"
1.1204 + "movdqa %1,%%xmm5 \n"
1.1205 + :
1.1206 + : "m"(kShuffleColARGB), // %0
1.1207 + "m"(kShuffleFractions) // %1
1.1208 + );
1.1209 +
1.1210 + asm volatile (
1.1211 + "movd %5,%%xmm2 \n"
1.1212 + "movd %6,%%xmm3 \n"
1.1213 + "pcmpeqb %%xmm6,%%xmm6 \n"
1.1214 + "psrlw $0x9,%%xmm6 \n"
1.1215 + "pextrw $0x1,%%xmm2,%k3 \n"
1.1216 + "sub $0x2,%2 \n"
1.1217 + "jl 29f \n"
1.1218 + "movdqa %%xmm2,%%xmm0 \n"
1.1219 + "paddd %%xmm3,%%xmm0 \n"
1.1220 + "punpckldq %%xmm0,%%xmm2 \n"
1.1221 + "punpckldq %%xmm3,%%xmm3 \n"
1.1222 + "paddd %%xmm3,%%xmm3 \n"
1.1223 + "pextrw $0x3,%%xmm2,%k4 \n"
1.1224 +
1.1225 + LABELALIGN
1.1226 + "2: \n"
1.1227 + "movdqa %%xmm2,%%xmm1 \n"
1.1228 + "paddd %%xmm3,%%xmm2 \n"
1.1229 + MEMOPREG(movq,0x00,1,3,4,xmm0) // movq (%1,%3,4),%%xmm0
1.1230 + "psrlw $0x9,%%xmm1 \n"
1.1231 + BUNDLEALIGN
1.1232 + MEMOPREG(movhps,0x00,1,4,4,xmm0) // movhps (%1,%4,4),%%xmm0
1.1233 + "pshufb %%xmm5,%%xmm1 \n"
1.1234 + "pshufb %%xmm4,%%xmm0 \n"
1.1235 + "pxor %%xmm6,%%xmm1 \n"
1.1236 + "pmaddubsw %%xmm1,%%xmm0 \n"
1.1237 + "psrlw $0x7,%%xmm0 \n"
1.1238 + "pextrw $0x1,%%xmm2,%k3 \n"
1.1239 + "pextrw $0x3,%%xmm2,%k4 \n"
1.1240 + "packuswb %%xmm0,%%xmm0 \n"
1.1241 + "movq %%xmm0," MEMACCESS(0) " \n"
1.1242 + "lea " MEMLEA(0x8,0) ",%0 \n"
1.1243 + "sub $0x2,%2 \n"
1.1244 + "jge 2b \n"
1.1245 +
1.1246 + LABELALIGN
1.1247 + "29: \n"
1.1248 + "add $0x1,%2 \n"
1.1249 + "jl 99f \n"
1.1250 + "psrlw $0x9,%%xmm2 \n"
1.1251 + BUNDLEALIGN
1.1252 + MEMOPREG(movq,0x00,1,3,4,xmm0) // movq (%1,%3,4),%%xmm0
1.1253 + "pshufb %%xmm5,%%xmm2 \n"
1.1254 + "pshufb %%xmm4,%%xmm0 \n"
1.1255 + "pxor %%xmm6,%%xmm2 \n"
1.1256 + "pmaddubsw %%xmm2,%%xmm0 \n"
1.1257 + "psrlw $0x7,%%xmm0 \n"
1.1258 + "packuswb %%xmm0,%%xmm0 \n"
1.1259 + "movd %%xmm0," MEMACCESS(0) " \n"
1.1260 +
1.1261 + LABELALIGN
1.1262 + "99: \n"
1.1263 + : "+r"(dst_argb), // %0
1.1264 + "+r"(src_argb), // %1
1.1265 + "+rm"(dst_width), // %2
1.1266 + "+r"(x0), // %3
1.1267 + "+r"(x1) // %4
1.1268 + : "rm"(x), // %5
1.1269 + "rm"(dx) // %6
1.1270 + : "memory", "cc"
1.1271 +#if defined(__native_client__) && defined(__x86_64__)
1.1272 + , "r14"
1.1273 +#endif
1.1274 +#if defined(__SSE2__)
1.1275 + , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
1.1276 +#endif
1.1277 + );
1.1278 +}
1.1279 +
1.1280 +// Divide num by div and return as 16.16 fixed point result.
1.1281 +int FixedDiv_X86(int num, int div) {
1.1282 + asm volatile (
1.1283 + "cdq \n"
1.1284 + "shld $0x10,%%eax,%%edx \n"
1.1285 + "shl $0x10,%%eax \n"
1.1286 + "idiv %1 \n"
1.1287 + "mov %0, %%eax \n"
1.1288 + : "+a"(num) // %0
1.1289 + : "c"(div) // %1
1.1290 + : "memory", "cc", "edx"
1.1291 + );
1.1292 + return num;
1.1293 +}
1.1294 +
1.1295 +// Divide num - 1 by div - 1 and return as 16.16 fixed point result.
1.1296 +int FixedDiv1_X86(int num, int div) {
1.1297 + asm volatile (
1.1298 + "cdq \n"
1.1299 + "shld $0x10,%%eax,%%edx \n"
1.1300 + "shl $0x10,%%eax \n"
1.1301 + "sub $0x10001,%%eax \n"
1.1302 + "sbb $0x0,%%edx \n"
1.1303 + "sub $0x1,%1 \n"
1.1304 + "idiv %1 \n"
1.1305 + "mov %0, %%eax \n"
1.1306 + : "+a"(num) // %0
1.1307 + : "c"(div) // %1
1.1308 + : "memory", "cc", "edx"
1.1309 + );
1.1310 + return num;
1.1311 +}
1.1312 +
1.1313 +#endif // defined(__x86_64__) || defined(__i386__)
1.1314 +
1.1315 +#ifdef __cplusplus
1.1316 +} // extern "C"
1.1317 +} // namespace libyuv
1.1318 +#endif
