| 1 | /* |
| 2 | * SIMD-optimized halfpel functions are compiled twice for rnd/no_rnd |
| 3 | * Copyright (c) 2000, 2001 Fabrice Bellard |
| 4 | * Copyright (c) 2003-2004 Michael Niedermayer <michaelni@gmx.at> |
| 5 | * |
| 6 | * MMX optimization by Nick Kurshev <nickols_k@mail.ru> |
| 7 | * mostly rewritten by Michael Niedermayer <michaelni@gmx.at> |
| 8 | * and improved by Zdenek Kabelac <kabi@users.sf.net> |
| 9 | * |
| 10 | * This file is part of FFmpeg. |
| 11 | * |
| 12 | * FFmpeg is free software; you can redistribute it and/or |
| 13 | * modify it under the terms of the GNU Lesser General Public |
| 14 | * License as published by the Free Software Foundation; either |
| 15 | * version 2.1 of the License, or (at your option) any later version. |
| 16 | * |
| 17 | * FFmpeg is distributed in the hope that it will be useful, |
| 18 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 19 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 20 | * Lesser General Public License for more details. |
| 21 | * |
| 22 | * You should have received a copy of the GNU Lesser General Public |
| 23 | * License along with FFmpeg; if not, write to the Free Software |
| 24 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 25 | */ |
| 26 | |
| 27 | #include <stddef.h> |
| 28 | #include <stdint.h> |
| 29 | |
| 30 | #include "inline_asm.h" |
| 31 | |
| 32 | // put_pixels |
| 33 | STATIC void DEF(put, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, |
| 34 | ptrdiff_t line_size, int h) |
| 35 | { |
| 36 | MOVQ_ZERO(mm7); |
| 37 | SET_RND(mm6); // =2 for rnd and =1 for no_rnd version |
| 38 | __asm__ volatile( |
| 39 | "movq (%1), %%mm0 \n\t" |
| 40 | "movq 1(%1), %%mm4 \n\t" |
| 41 | "movq %%mm0, %%mm1 \n\t" |
| 42 | "movq %%mm4, %%mm5 \n\t" |
| 43 | "punpcklbw %%mm7, %%mm0 \n\t" |
| 44 | "punpcklbw %%mm7, %%mm4 \n\t" |
| 45 | "punpckhbw %%mm7, %%mm1 \n\t" |
| 46 | "punpckhbw %%mm7, %%mm5 \n\t" |
| 47 | "paddusw %%mm0, %%mm4 \n\t" |
| 48 | "paddusw %%mm1, %%mm5 \n\t" |
| 49 | "xor %%"REG_a", %%"REG_a" \n\t" |
| 50 | "add %3, %1 \n\t" |
| 51 | ".p2align 3 \n\t" |
| 52 | "1: \n\t" |
| 53 | "movq (%1, %%"REG_a"), %%mm0 \n\t" |
| 54 | "movq 1(%1, %%"REG_a"), %%mm2 \n\t" |
| 55 | "movq %%mm0, %%mm1 \n\t" |
| 56 | "movq %%mm2, %%mm3 \n\t" |
| 57 | "punpcklbw %%mm7, %%mm0 \n\t" |
| 58 | "punpcklbw %%mm7, %%mm2 \n\t" |
| 59 | "punpckhbw %%mm7, %%mm1 \n\t" |
| 60 | "punpckhbw %%mm7, %%mm3 \n\t" |
| 61 | "paddusw %%mm2, %%mm0 \n\t" |
| 62 | "paddusw %%mm3, %%mm1 \n\t" |
| 63 | "paddusw %%mm6, %%mm4 \n\t" |
| 64 | "paddusw %%mm6, %%mm5 \n\t" |
| 65 | "paddusw %%mm0, %%mm4 \n\t" |
| 66 | "paddusw %%mm1, %%mm5 \n\t" |
| 67 | "psrlw $2, %%mm4 \n\t" |
| 68 | "psrlw $2, %%mm5 \n\t" |
| 69 | "packuswb %%mm5, %%mm4 \n\t" |
| 70 | "movq %%mm4, (%2, %%"REG_a") \n\t" |
| 71 | "add %3, %%"REG_a" \n\t" |
| 72 | |
| 73 | "movq (%1, %%"REG_a"), %%mm2 \n\t" // 0 <-> 2 1 <-> 3 |
| 74 | "movq 1(%1, %%"REG_a"), %%mm4 \n\t" |
| 75 | "movq %%mm2, %%mm3 \n\t" |
| 76 | "movq %%mm4, %%mm5 \n\t" |
| 77 | "punpcklbw %%mm7, %%mm2 \n\t" |
| 78 | "punpcklbw %%mm7, %%mm4 \n\t" |
| 79 | "punpckhbw %%mm7, %%mm3 \n\t" |
| 80 | "punpckhbw %%mm7, %%mm5 \n\t" |
| 81 | "paddusw %%mm2, %%mm4 \n\t" |
| 82 | "paddusw %%mm3, %%mm5 \n\t" |
| 83 | "paddusw %%mm6, %%mm0 \n\t" |
| 84 | "paddusw %%mm6, %%mm1 \n\t" |
| 85 | "paddusw %%mm4, %%mm0 \n\t" |
| 86 | "paddusw %%mm5, %%mm1 \n\t" |
| 87 | "psrlw $2, %%mm0 \n\t" |
| 88 | "psrlw $2, %%mm1 \n\t" |
| 89 | "packuswb %%mm1, %%mm0 \n\t" |
| 90 | "movq %%mm0, (%2, %%"REG_a") \n\t" |
| 91 | "add %3, %%"REG_a" \n\t" |
| 92 | |
| 93 | "subl $2, %0 \n\t" |
| 94 | "jnz 1b \n\t" |
| 95 | :"+g"(h), "+S"(pixels) |
| 96 | :"D"(block), "r"((x86_reg)line_size) |
| 97 | :REG_a, "memory"); |
| 98 | } |
| 99 | |
| 100 | // avg_pixels |
| 101 | // this routine is 'slightly' suboptimal but mostly unused |
| 102 | STATIC void DEF(avg, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, |
| 103 | ptrdiff_t line_size, int h) |
| 104 | { |
| 105 | MOVQ_ZERO(mm7); |
| 106 | SET_RND(mm6); // =2 for rnd and =1 for no_rnd version |
| 107 | __asm__ volatile( |
| 108 | "movq (%1), %%mm0 \n\t" |
| 109 | "movq 1(%1), %%mm4 \n\t" |
| 110 | "movq %%mm0, %%mm1 \n\t" |
| 111 | "movq %%mm4, %%mm5 \n\t" |
| 112 | "punpcklbw %%mm7, %%mm0 \n\t" |
| 113 | "punpcklbw %%mm7, %%mm4 \n\t" |
| 114 | "punpckhbw %%mm7, %%mm1 \n\t" |
| 115 | "punpckhbw %%mm7, %%mm5 \n\t" |
| 116 | "paddusw %%mm0, %%mm4 \n\t" |
| 117 | "paddusw %%mm1, %%mm5 \n\t" |
| 118 | "xor %%"REG_a", %%"REG_a" \n\t" |
| 119 | "add %3, %1 \n\t" |
| 120 | ".p2align 3 \n\t" |
| 121 | "1: \n\t" |
| 122 | "movq (%1, %%"REG_a"), %%mm0 \n\t" |
| 123 | "movq 1(%1, %%"REG_a"), %%mm2 \n\t" |
| 124 | "movq %%mm0, %%mm1 \n\t" |
| 125 | "movq %%mm2, %%mm3 \n\t" |
| 126 | "punpcklbw %%mm7, %%mm0 \n\t" |
| 127 | "punpcklbw %%mm7, %%mm2 \n\t" |
| 128 | "punpckhbw %%mm7, %%mm1 \n\t" |
| 129 | "punpckhbw %%mm7, %%mm3 \n\t" |
| 130 | "paddusw %%mm2, %%mm0 \n\t" |
| 131 | "paddusw %%mm3, %%mm1 \n\t" |
| 132 | "paddusw %%mm6, %%mm4 \n\t" |
| 133 | "paddusw %%mm6, %%mm5 \n\t" |
| 134 | "paddusw %%mm0, %%mm4 \n\t" |
| 135 | "paddusw %%mm1, %%mm5 \n\t" |
| 136 | "psrlw $2, %%mm4 \n\t" |
| 137 | "psrlw $2, %%mm5 \n\t" |
| 138 | "movq (%2, %%"REG_a"), %%mm3 \n\t" |
| 139 | "packuswb %%mm5, %%mm4 \n\t" |
| 140 | "pcmpeqd %%mm2, %%mm2 \n\t" |
| 141 | "paddb %%mm2, %%mm2 \n\t" |
| 142 | PAVGB_MMX(%%mm3, %%mm4, %%mm5, %%mm2) |
| 143 | "movq %%mm5, (%2, %%"REG_a") \n\t" |
| 144 | "add %3, %%"REG_a" \n\t" |
| 145 | |
| 146 | "movq (%1, %%"REG_a"), %%mm2 \n\t" // 0 <-> 2 1 <-> 3 |
| 147 | "movq 1(%1, %%"REG_a"), %%mm4 \n\t" |
| 148 | "movq %%mm2, %%mm3 \n\t" |
| 149 | "movq %%mm4, %%mm5 \n\t" |
| 150 | "punpcklbw %%mm7, %%mm2 \n\t" |
| 151 | "punpcklbw %%mm7, %%mm4 \n\t" |
| 152 | "punpckhbw %%mm7, %%mm3 \n\t" |
| 153 | "punpckhbw %%mm7, %%mm5 \n\t" |
| 154 | "paddusw %%mm2, %%mm4 \n\t" |
| 155 | "paddusw %%mm3, %%mm5 \n\t" |
| 156 | "paddusw %%mm6, %%mm0 \n\t" |
| 157 | "paddusw %%mm6, %%mm1 \n\t" |
| 158 | "paddusw %%mm4, %%mm0 \n\t" |
| 159 | "paddusw %%mm5, %%mm1 \n\t" |
| 160 | "psrlw $2, %%mm0 \n\t" |
| 161 | "psrlw $2, %%mm1 \n\t" |
| 162 | "movq (%2, %%"REG_a"), %%mm3 \n\t" |
| 163 | "packuswb %%mm1, %%mm0 \n\t" |
| 164 | "pcmpeqd %%mm2, %%mm2 \n\t" |
| 165 | "paddb %%mm2, %%mm2 \n\t" |
| 166 | PAVGB_MMX(%%mm3, %%mm0, %%mm1, %%mm2) |
| 167 | "movq %%mm1, (%2, %%"REG_a") \n\t" |
| 168 | "add %3, %%"REG_a" \n\t" |
| 169 | |
| 170 | "subl $2, %0 \n\t" |
| 171 | "jnz 1b \n\t" |
| 172 | :"+g"(h), "+S"(pixels) |
| 173 | :"D"(block), "r"((x86_reg)line_size) |
| 174 | :REG_a, "memory"); |
| 175 | } |