| OLD | NEW |
| 1 /* | 1 /* |
| 2 * Copyright 2013 The LibYuv Project Authors. All rights reserved. | 2 * Copyright 2013 The LibYuv Project Authors. All rights reserved. |
| 3 * | 3 * |
| 4 * Use of this source code is governed by a BSD-style license | 4 * Use of this source code is governed by a BSD-style license |
| 5 * that can be found in the LICENSE file in the root of the source | 5 * that can be found in the LICENSE file in the root of the source |
| 6 * tree. An additional intellectual property rights grant can be found | 6 * tree. An additional intellectual property rights grant can be found |
| 7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
| 8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
| 9 */ | 9 */ |
| 10 | 10 |
| (...skipping 860 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 871 "paddusb %%xmm7,%%xmm1 \n" | 871 "paddusb %%xmm7,%%xmm1 \n" |
| 872 "pmaddubsw %%xmm0,%%xmm1 \n" | 872 "pmaddubsw %%xmm0,%%xmm1 \n" |
| 873 "pextrw $0x1,%%xmm2,%k3 \n" | 873 "pextrw $0x1,%%xmm2,%k3 \n" |
| 874 "pextrw $0x3,%%xmm2,%k4 \n" | 874 "pextrw $0x3,%%xmm2,%k4 \n" |
| 875 "paddw %9,%%xmm1 \n" // make pixels unsigned. | 875 "paddw %9,%%xmm1 \n" // make pixels unsigned. |
| 876 "psrlw $0x7,%%xmm1 \n" | 876 "psrlw $0x7,%%xmm1 \n" |
| 877 "packuswb %%xmm1,%%xmm1 \n" | 877 "packuswb %%xmm1,%%xmm1 \n" |
| 878 "movd %%xmm1,%k2 \n" | 878 "movd %%xmm1,%k2 \n" |
| 879 "mov %w2," MEMACCESS(0) " \n" | 879 "mov %w2," MEMACCESS(0) " \n" |
| 880 "lea " MEMLEA(0x2,0) ",%0 \n" | 880 "lea " MEMLEA(0x2,0) ",%0 \n" |
| 881 "sub $0x2,%5 \n" | 881 "subl $0x2,%5 \n" |
| 882 "jge 2b \n" | 882 "jge 2b \n" |
| 883 | 883 |
| 884 LABELALIGN | 884 LABELALIGN |
| 885 "29: \n" | 885 "29: \n" |
| 886 "addl $0x1,%5 \n" | 886 "addl $0x1,%5 \n" |
| 887 "jl 99f \n" | 887 "jl 99f \n" |
| 888 MEMOPARG(movzwl,0x00,1,3,1,k2) // movzwl (%1,%3,1),%k2 | 888 MEMOPARG(movzwl,0x00,1,3,1,k2) // movzwl (%1,%3,1),%k2 |
| 889 "movd %k2,%%xmm0 \n" | 889 "movd %k2,%%xmm0 \n" |
| 890 "psrlw $0x9,%%xmm2 \n" | 890 "psrlw $0x9,%%xmm2 \n" |
| 891 "pshufb %%xmm5,%%xmm2 \n" | 891 "pshufb %%xmm5,%%xmm2 \n" |
| 892 "psubb %8,%%xmm0 \n" // make pixels signed. | 892 "psubb %8,%%xmm0 \n" // make pixels signed. |
| 893 "pxor %%xmm6,%%xmm2 \n" | 893 "pxor %%xmm6,%%xmm2 \n" |
| 894 "paddusb %%xmm7,%%xmm2 \n" | 894 "paddusb %%xmm7,%%xmm2 \n" |
| 895 "pmaddubsw %%xmm0,%%xmm2 \n" | 895 "pmaddubsw %%xmm0,%%xmm2 \n" |
| 896 "paddw %9,%%xmm2 \n" // make pixels unsigned. | 896 "paddw %9,%%xmm2 \n" // make pixels unsigned. |
| 897 "psrlw $0x7,%%xmm2 \n" | 897 "psrlw $0x7,%%xmm2 \n" |
| 898 "packuswb %%xmm2,%%xmm2 \n" | 898 "packuswb %%xmm2,%%xmm2 \n" |
| 899 "movd %%xmm2,%k2 \n" | 899 "movd %%xmm2,%k2 \n" |
| 900 "mov %b2," MEMACCESS(0) " \n" | 900 "mov %b2," MEMACCESS(0) " \n" |
| 901 "99: \n" | 901 "99: \n" |
| 902 : "+r"(dst_ptr), // %0 | 902 : "+r"(dst_ptr), // %0 |
| 903 "+r"(src_ptr), // %1 | 903 "+r"(src_ptr), // %1 |
| 904 "=&a"(temp_pixel), // %2 | 904 "=&a"(temp_pixel), // %2 |
| 905 "=&r"(x0), // %3 | 905 "=&r"(x0), // %3 |
| 906 "=&r"(x1), // %4 | 906 "=&r"(x1), // %4 |
| 907 #if defined(__x86_64__) |
| 907 "+rm"(dst_width) // %5 | 908 "+rm"(dst_width) // %5 |
| 909 #else |
| 910 "+m"(dst_width) // %5 |
| 911 #endif |
| 908 : "rm"(x), // %6 | 912 : "rm"(x), // %6 |
| 909 "rm"(dx), // %7 | 913 "rm"(dx), // %7 |
| 910 #if defined(__x86_64__) | 914 #if defined(__x86_64__) |
| 911 "x"(kFsub80), // %8 | 915 "x"(kFsub80), // %8 |
| 912 "x"(kFadd40) // %9 | 916 "x"(kFadd40) // %9 |
| 913 #else | 917 #else |
| 914 "m"(kFsub80), // %8 | 918 "m"(kFsub80), // %8 |
| 915 "m"(kFadd40) // %9 | 919 "m"(kFadd40) // %9 |
| 916 #endif | 920 #endif |
| 917 : "memory", "cc", NACL_R14 | 921 : "memory", "cc", NACL_R14 |
| (...skipping 391 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 1309 ); | 1313 ); |
| 1310 return num; | 1314 return num; |
| 1311 } | 1315 } |
| 1312 | 1316 |
| 1313 #endif // defined(__x86_64__) || defined(__i386__) | 1317 #endif // defined(__x86_64__) || defined(__i386__) |
| 1314 | 1318 |
| 1315 #ifdef __cplusplus | 1319 #ifdef __cplusplus |
| 1316 } // extern "C" | 1320 } // extern "C" |
| 1317 } // namespace libyuv | 1321 } // namespace libyuv |
| 1318 #endif | 1322 #endif |
| OLD | NEW |