1/* 2 * iWMMXt optimized DSP utils 3 * Copyright (c) 2004 AGAWA Koji 4 * 5 * This file is part of Libav. 6 * 7 * Libav is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU Lesser General Public 9 * License as published by the Free Software Foundation; either 10 * version 2.1 of the License, or (at your option) any later version. 11 * 12 * Libav is distributed in the hope that it will be useful, 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 15 * Lesser General Public License for more details. 16 * 17 * You should have received a copy of the GNU Lesser General Public 18 * License along with Libav; if not, write to the Free Software 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 20 */ 21 22#include "libavutil/cpu.h" 23#include "libavcodec/dsputil.h" 24 25#define DEF(x, y) x ## _no_rnd_ ## y ##_iwmmxt 26#define SET_RND(regd) __asm__ volatile ("mov r12, #1 \n\t tbcsth " #regd ", r12":::"r12"); 27#define WAVG2B "wavg2b" 28#include "dsputil_iwmmxt_rnd_template.c" 29#undef DEF 30#undef SET_RND 31#undef WAVG2B 32 33#define DEF(x, y) x ## _ ## y ##_iwmmxt 34#define SET_RND(regd) __asm__ volatile ("mov r12, #2 \n\t tbcsth " #regd ", r12":::"r12"); 35#define WAVG2B "wavg2br" 36#include "dsputil_iwmmxt_rnd_template.c" 37#undef DEF 38#undef SET_RND 39#undef WAVG2BR 40 41// need scheduling 42#define OP(AVG) \ 43 __asm__ volatile ( \ 44 /* alignment */ \ 45 "and r12, %[pixels], #7 \n\t" \ 46 "bic %[pixels], %[pixels], #7 \n\t" \ 47 "tmcr wcgr1, r12 \n\t" \ 48 \ 49 "wldrd wr0, [%[pixels]] \n\t" \ 50 "wldrd wr1, [%[pixels], #8] \n\t" \ 51 "add %[pixels], %[pixels], %[line_size] \n\t" \ 52 "walignr1 wr4, wr0, wr1 \n\t" \ 53 \ 54 "1: \n\t" \ 55 \ 56 "wldrd wr2, [%[pixels]] \n\t" \ 57 "wldrd wr3, [%[pixels], #8] \n\t" \ 58 "add %[pixels], %[pixels], %[line_size] \n\t" \ 59 "pld [%[pixels]] \n\t" \ 60 "walignr1 wr5, wr2, wr3 \n\t" \ 61 AVG " wr6, wr4, wr5 \n\t" \ 62 "wstrd wr6, [%[block]] \n\t" \ 63 "add %[block], %[block], %[line_size] \n\t" \ 64 \ 65 "wldrd wr0, [%[pixels]] \n\t" \ 66 "wldrd wr1, [%[pixels], #8] \n\t" \ 67 "add %[pixels], %[pixels], %[line_size] \n\t" \ 68 "walignr1 wr4, wr0, wr1 \n\t" \ 69 "pld [%[pixels]] \n\t" \ 70 AVG " wr6, wr4, wr5 \n\t" \ 71 "wstrd wr6, [%[block]] \n\t" \ 72 "add %[block], %[block], %[line_size] \n\t" \ 73 \ 74 "subs %[h], %[h], #2 \n\t" \ 75 "bne 1b \n\t" \ 76 : [block]"+r"(block), [pixels]"+r"(pixels), [h]"+r"(h) \ 77 : [line_size]"r"(line_size) \ 78 : "memory", "r12"); 79void put_pixels8_y2_iwmmxt(uint8_t *block, const uint8_t *pixels, const int line_size, int h) 80{ 81 OP("wavg2br"); 82} 83void put_no_rnd_pixels8_y2_iwmmxt(uint8_t *block, const uint8_t *pixels, const int line_size, int h) 84{ 85 OP("wavg2b"); 86} 87#undef OP 88 89void add_pixels_clamped_iwmmxt(const DCTELEM *block, uint8_t *pixels, int line_size) 90{ 91 uint8_t *pixels2 = pixels + line_size; 92 93 __asm__ volatile ( 94 "mov r12, #4 \n\t" 95 "1: \n\t" 96 "pld [%[pixels], %[line_size2]] \n\t" 97 "pld [%[pixels2], %[line_size2]] \n\t" 98 "wldrd wr4, [%[pixels]] \n\t" 99 "wldrd wr5, [%[pixels2]] \n\t" 100 "pld [%[block], #32] \n\t" 101 "wunpckelub wr6, wr4 \n\t" 102 "wldrd wr0, [%[block]] \n\t" 103 "wunpckehub wr7, wr4 \n\t" 104 "wldrd wr1, [%[block], #8] \n\t" 105 "wunpckelub wr8, wr5 \n\t" 106 "wldrd wr2, [%[block], #16] \n\t" 107 "wunpckehub wr9, wr5 \n\t" 108 "wldrd wr3, [%[block], #24] \n\t" 109 "add %[block], %[block], #32 \n\t" 110 "waddhss wr10, wr0, wr6 \n\t" 111 "waddhss wr11, wr1, wr7 \n\t" 112 "waddhss wr12, wr2, wr8 \n\t" 113 "waddhss wr13, wr3, wr9 \n\t" 114 "wpackhus wr14, wr10, wr11 \n\t" 115 "wpackhus wr15, wr12, wr13 \n\t" 116 "wstrd wr14, [%[pixels]] \n\t" 117 "add %[pixels], %[pixels], %[line_size2] \n\t" 118 "subs r12, r12, #1 \n\t" 119 "wstrd wr15, [%[pixels2]] \n\t" 120 "add %[pixels2], %[pixels2], %[line_size2] \n\t" 121 "bne 1b \n\t" 122 : [block]"+r"(block), [pixels]"+r"(pixels), [pixels2]"+r"(pixels2) 123 : [line_size2]"r"(line_size << 1) 124 : "cc", "memory", "r12"); 125} 126 127static void clear_blocks_iwmmxt(DCTELEM *blocks) 128{ 129 __asm__ volatile( 130 "wzero wr0 \n\t" 131 "mov r1, #(128 * 6 / 32) \n\t" 132 "1: \n\t" 133 "wstrd wr0, [%0] \n\t" 134 "wstrd wr0, [%0, #8] \n\t" 135 "wstrd wr0, [%0, #16] \n\t" 136 "wstrd wr0, [%0, #24] \n\t" 137 "subs r1, r1, #1 \n\t" 138 "add %0, %0, #32 \n\t" 139 "bne 1b \n\t" 140 : "+r"(blocks) 141 : 142 : "r1" 143 ); 144} 145 146static void nop(uint8_t *block, const uint8_t *pixels, int line_size, int h) 147{ 148 return; 149} 150 151/* A run time test is not simple. If this file is compiled in 152 * then we should install the functions 153 */ 154 155void ff_dsputil_init_iwmmxt(DSPContext* c, AVCodecContext *avctx) 156{ 157 int mm_flags = AV_CPU_FLAG_IWMMXT; /* multimedia extension flags */ 158 const int high_bit_depth = avctx->bits_per_raw_sample > 8; 159 160 if (avctx->dsp_mask) { 161 if (avctx->dsp_mask & AV_CPU_FLAG_FORCE) 162 mm_flags |= (avctx->dsp_mask & 0xffff); 163 else 164 mm_flags &= ~(avctx->dsp_mask & 0xffff); 165 } 166 167 if (!(mm_flags & AV_CPU_FLAG_IWMMXT)) return; 168 169 c->add_pixels_clamped = add_pixels_clamped_iwmmxt; 170 171 if (!high_bit_depth) { 172 c->clear_blocks = clear_blocks_iwmmxt; 173 174 c->put_pixels_tab[0][0] = put_pixels16_iwmmxt; 175 c->put_pixels_tab[0][1] = put_pixels16_x2_iwmmxt; 176 c->put_pixels_tab[0][2] = put_pixels16_y2_iwmmxt; 177 c->put_pixels_tab[0][3] = put_pixels16_xy2_iwmmxt; 178 c->put_no_rnd_pixels_tab[0][0] = put_pixels16_iwmmxt; 179 c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_iwmmxt; 180 c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_iwmmxt; 181 c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_iwmmxt; 182 183 c->put_pixels_tab[1][0] = put_pixels8_iwmmxt; 184 c->put_pixels_tab[1][1] = put_pixels8_x2_iwmmxt; 185 c->put_pixels_tab[1][2] = put_pixels8_y2_iwmmxt; 186 c->put_pixels_tab[1][3] = put_pixels8_xy2_iwmmxt; 187 c->put_no_rnd_pixels_tab[1][0] = put_pixels8_iwmmxt; 188 c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_iwmmxt; 189 c->put_no_rnd_pixels_tab[1][2] = put_no_rnd_pixels8_y2_iwmmxt; 190 c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_iwmmxt; 191 192 c->avg_pixels_tab[0][0] = avg_pixels16_iwmmxt; 193 c->avg_pixels_tab[0][1] = avg_pixels16_x2_iwmmxt; 194 c->avg_pixels_tab[0][2] = avg_pixels16_y2_iwmmxt; 195 c->avg_pixels_tab[0][3] = avg_pixels16_xy2_iwmmxt; 196 c->avg_no_rnd_pixels_tab[0][0] = avg_pixels16_iwmmxt; 197 c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_iwmmxt; 198 c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_iwmmxt; 199 c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_iwmmxt; 200 201 c->avg_pixels_tab[1][0] = avg_pixels8_iwmmxt; 202 c->avg_pixels_tab[1][1] = avg_pixels8_x2_iwmmxt; 203 c->avg_pixels_tab[1][2] = avg_pixels8_y2_iwmmxt; 204 c->avg_pixels_tab[1][3] = avg_pixels8_xy2_iwmmxt; 205 c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_iwmmxt; 206 c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_iwmmxt; 207 c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_iwmmxt; 208 c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_iwmmxt; 209 } 210} 211