yading@10: /* yading@10: * BlackFin MPEGVIDEO OPTIMIZATIONS yading@10: * yading@10: * Copyright (C) 2007 Marc Hoffman yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: #include "libavutil/attributes.h" yading@10: #include "libavcodec/avcodec.h" yading@10: #include "libavcodec/mpegvideo.h" yading@10: #include "dsputil_bfin.h" yading@10: yading@10: static int dct_quantize_bfin (MpegEncContext *s, yading@10: int16_t *block, int n, yading@10: int qscale, int *overflow) yading@10: { yading@10: int last_non_zero, q, start_i; yading@10: const short *qmat; yading@10: short *bias; yading@10: const uint8_t *scantable= s->intra_scantable.scantable; yading@10: short dc; yading@10: int max=0; yading@10: yading@10: PROF("fdct",0); yading@10: s->dsp.fdct(block); yading@10: EPROF(); yading@10: yading@10: PROF("denoise",1); yading@10: if(s->dct_error_sum) yading@10: s->denoise_dct(s, block); yading@10: EPROF(); yading@10: yading@10: PROF("quant-init",2); yading@10: if (s->mb_intra) { yading@10: if (!s->h263_aic) { yading@10: if (n < 4) yading@10: q = s->y_dc_scale; yading@10: else yading@10: q = s->c_dc_scale; yading@10: q = q << 3; yading@10: } else yading@10: /* For AIC we skip quant/dequant of INTRADC */ yading@10: q = 1 << 3; yading@10: yading@10: /* note: block[0] is assumed to be positive */ yading@10: dc = block[0] = (block[0] + (q >> 1)) / q; yading@10: start_i = 1; yading@10: last_non_zero = 0; yading@10: if(n<4){ yading@10: bias = s->q_intra_matrix16[qscale][1]; yading@10: qmat = s->q_intra_matrix16[qscale][0]; yading@10: }else{ yading@10: bias = s->q_chroma_intra_matrix16[qscale][1]; yading@10: qmat = s->q_chroma_intra_matrix16[qscale][0]; yading@10: } yading@10: yading@10: } else { yading@10: start_i = 0; yading@10: last_non_zero = -1; yading@10: bias = s->q_inter_matrix16[qscale][1]; yading@10: qmat = s->q_inter_matrix16[qscale][0]; yading@10: yading@10: } yading@10: EPROF(); yading@10: yading@10: PROF("quantize",4); yading@10: yading@10: /* for(i=start_i; i<64; i++) { */ yading@10: /* sign = (block[i]>>15)|1; */ yading@10: /* level = ((abs(block[i])+bias[0])*qmat[i])>>16; */ yading@10: /* if (level < 0) level = 0; */ yading@10: /* max |= level; */ yading@10: /* level = level * sign; */ yading@10: /* block[i] = level; */ yading@10: /* } */ yading@10: yading@10: __asm__ volatile yading@10: ("i2=%1;\n\t" yading@10: "r1=[%1++]; \n\t" yading@10: "r0=r1>>>15 (v); \n\t" yading@10: "lsetup (0f,1f) lc0=%3; \n\t" yading@10: "0: r0=r0|%4; \n\t" yading@10: " r1=abs r1 (v) || r2=[%2++];\n\t" yading@10: " r1=r1+|+%5; \n\t" yading@10: " r1=max(r1,%6) (v); \n\t" yading@10: " r1.h=(a1 =r1.h*r2.h), r1.l=(a0 =r1.l*r2.l) (tfu); \n\t" yading@10: " %0=%0|r1; \n\t" yading@10: " r0.h=(a1 =r1.h*r0.h), r0.l=(a0 =r1.l*r0.l) (is) || r1=[%1++];\n\t" yading@10: "1: r0=r1>>>15 (v) || [i2++]=r0;\n\t" yading@10: "r1=%0>>16; \n\t" yading@10: "%0=%0|r1; \n\t" yading@10: "%0.h=0; \n\t" yading@10: : "=&d" (max) yading@10: : "b" (block), "b" (qmat), "a" (32), "d" (0x00010001), "d" (bias[0]*0x10001), "d" (0) yading@10: : "R0","R1","R2", "I2"); yading@10: if (start_i == 1) block[0] = dc; yading@10: yading@10: EPROF(); yading@10: yading@10: yading@10: PROF("zzscan",5); yading@10: yading@10: __asm__ volatile yading@10: ("r0=b[%1--] (x); \n\t" yading@10: "lsetup (0f,1f) lc0=%3; \n\t" /* for(i=63; i>=start_i; i--) { */ yading@10: "0: p0=r0; \n\t" /* j = scantable[i]; */ yading@10: " p0=%2+(p0<<1); \n\t" /* if (block[j]) { */ yading@10: " r0=w[p0]; \n\t" /* last_non_zero = i; */ yading@10: " cc=r0==0; \n\t" /* break; */ yading@10: " if !cc jump 2f; \n\t" /* } */ yading@10: "1: r0=b[%1--] (x); \n\t" /* } */ yading@10: " %0=%4; \n\t" yading@10: " jump 3f; \n\t" yading@10: "2: %0=lc0; \n\t" yading@10: "3:\n\t" yading@10: yading@10: : "=d" (last_non_zero) yading@10: : "a" (scantable+63), "a" (block), "a" (63), "d" (last_non_zero) yading@10: : "P0","R0"); yading@10: yading@10: EPROF(); yading@10: yading@10: *overflow= s->max_qcoeff < max; //overflow might have happened yading@10: yading@10: bfprof(); yading@10: yading@10: /* we need this permutation so that we correct the IDCT, we only permute the !=0 elements */ yading@10: if (s->dsp.idct_permutation_type != FF_NO_IDCT_PERM) yading@10: ff_block_permute(block, s->dsp.idct_permutation, scantable, last_non_zero); yading@10: yading@10: return last_non_zero; yading@10: } yading@10: yading@10: av_cold void ff_MPV_common_init_bfin (MpegEncContext *s) yading@10: { yading@10: /* s->dct_quantize= dct_quantize_bfin; */ yading@10: }