1/* 2 * copyright (c) 2004 AGAWA Koji 3 * 4 * This file is part of Libav. 5 * 6 * Libav is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU Lesser General Public 8 * License as published by the Free Software Foundation; either 9 * version 2.1 of the License, or (at your option) any later version. 10 * 11 * Libav is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * Lesser General Public License for more details. 15 * 16 * You should have received a copy of the GNU Lesser General Public 17 * License along with Libav; if not, write to the Free Software 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 19 */ 20 21#include "libavutil/cpu.h" 22#include "libavcodec/avcodec.h" 23#include "libavcodec/dsputil.h" 24#include "libavcodec/mpegvideo.h" 25#include "mpegvideo_arm.h" 26 27static void dct_unquantize_h263_intra_iwmmxt(MpegEncContext *s, 28 DCTELEM *block, int n, int qscale) 29{ 30 int level, qmul, qadd; 31 int nCoeffs; 32 DCTELEM *block_orig = block; 33 34 assert(s->block_last_index[n]>=0); 35 36 qmul = qscale << 1; 37 38 if (!s->h263_aic) { 39 if (n < 4) 40 level = block[0] * s->y_dc_scale; 41 else 42 level = block[0] * s->c_dc_scale; 43 qadd = (qscale - 1) | 1; 44 }else{ 45 qadd = 0; 46 level = block[0]; 47 } 48 if(s->ac_pred) 49 nCoeffs=63; 50 else 51 nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ]; 52 53 __asm__ volatile ( 54/* "movd %1, %%mm6 \n\t" //qmul */ 55/* "packssdw %%mm6, %%mm6 \n\t" */ 56/* "packssdw %%mm6, %%mm6 \n\t" */ 57 "tbcsth wr6, %[qmul] \n\t" 58/* "movd %2, %%mm5 \n\t" //qadd */ 59/* "packssdw %%mm5, %%mm5 \n\t" */ 60/* "packssdw %%mm5, %%mm5 \n\t" */ 61 "tbcsth wr5, %[qadd] \n\t" 62 "wzero wr7 \n\t" /* "pxor %%mm7, %%mm7 \n\t" */ 63 "wzero wr4 \n\t" /* "pxor %%mm4, %%mm4 \n\t" */ 64 "wsubh wr7, wr5, wr7 \n\t" /* "psubw %%mm5, %%mm7 \n\t" */ 65 "1: \n\t" 66 "wldrd wr2, [%[block]] \n\t" /* "movq (%0, %3), %%mm0 \n\t" */ 67 "wldrd wr3, [%[block], #8] \n\t" /* "movq 8(%0, %3), %%mm1 \n\t" */ 68 "wmulsl wr0, wr6, wr2 \n\t" /* "pmullw %%mm6, %%mm0 \n\t" */ 69 "wmulsl wr1, wr6, wr3 \n\t" /* "pmullw %%mm6, %%mm1 \n\t" */ 70/* "movq (%0, %3), %%mm2 \n\t" */ 71/* "movq 8(%0, %3), %%mm3 \n\t" */ 72 "wcmpgtsh wr2, wr4, wr2 \n\t" /* "pcmpgtw %%mm4, %%mm2 \n\t" // block[i] < 0 ? -1 : 0 */ 73 "wcmpgtsh wr3, wr4, wr2 \n\t" /* "pcmpgtw %%mm4, %%mm3 \n\t" // block[i] < 0 ? -1 : 0 */ 74 "wxor wr0, wr2, wr0 \n\t" /* "pxor %%mm2, %%mm0 \n\t" */ 75 "wxor wr1, wr3, wr1 \n\t" /* "pxor %%mm3, %%mm1 \n\t" */ 76 "waddh wr0, wr7, wr0 \n\t" /* "paddw %%mm7, %%mm0 \n\t" */ 77 "waddh wr1, wr7, wr1 \n\t" /* "paddw %%mm7, %%mm1 \n\t" */ 78 "wxor wr2, wr0, wr2 \n\t" /* "pxor %%mm0, %%mm2 \n\t" */ 79 "wxor wr3, wr1, wr3 \n\t" /* "pxor %%mm1, %%mm3 \n\t" */ 80 "wcmpeqh wr0, wr7, wr0 \n\t" /* "pcmpeqw %%mm7, %%mm0 \n\t" // block[i] == 0 ? -1 : 0 */ 81 "wcmpeqh wr1, wr7, wr1 \n\t" /* "pcmpeqw %%mm7, %%mm1 \n\t" // block[i] == 0 ? -1 : 0 */ 82 "wandn wr0, wr2, wr0 \n\t" /* "pandn %%mm2, %%mm0 \n\t" */ 83 "wandn wr1, wr3, wr1 \n\t" /* "pandn %%mm3, %%mm1 \n\t" */ 84 "wstrd wr0, [%[block]] \n\t" /* "movq %%mm0, (%0, %3) \n\t" */ 85 "wstrd wr1, [%[block], #8] \n\t" /* "movq %%mm1, 8(%0, %3) \n\t" */ 86 "add %[block], %[block], #16 \n\t" /* "addl $16, %3 \n\t" */ 87 "subs %[i], %[i], #1 \n\t" 88 "bne 1b \n\t" /* "jng 1b \n\t" */ 89 :[block]"+r"(block) 90 :[i]"r"((nCoeffs + 8) / 8), [qmul]"r"(qmul), [qadd]"r"(qadd) 91 :"memory"); 92 93 block_orig[0] = level; 94} 95 96void MPV_common_init_iwmmxt(MpegEncContext *s) 97{ 98 if (!(mm_flags & AV_CPU_FLAG_IWMMXT)) return; 99 100 s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_iwmmxt; 101} 102