h264_template_altivec.c

ffmpeg的完整源代码和作者自己写的文档。不但有在Linux的工程哦

/*
 * Copyright (c) 2004 Romain Dolbeau <[e-mail]>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

//#define DEBUG_ALIGNMENT
#ifdef DEBUG_ALIGNMENT
#define ASSERT_ALIGNED(ptr) assert(((unsigned long)ptr&0x0000000F));
#else
#define ASSERT_ALIGNED(ptr) ;
#endif

/* this code assume that stride % 16 == 0 */
void PREFIX_h264_chroma_mc8_altivec(uint8_t * dst, uint8_t * src, int stride, int h, int x, int y) {
  POWERPC_PERF_DECLARE(PREFIX_h264_chroma_mc8_num, 1);
    DECLARE_ALIGNED_16(signed int, ABCD[4]) =
                        {((8 - x) * (8 - y)),
                          ((x) * (8 - y)),
                          ((8 - x) * (y)),
                          ((x) * (y))};
    register int i;
    vec_u8_t fperm;
    const vec_s32_t vABCD = vec_ld(0, ABCD);
    const vec_s16_t vA = vec_splat((vec_s16_t)vABCD, 1);
    const vec_s16_t vB = vec_splat((vec_s16_t)vABCD, 3);
    const vec_s16_t vC = vec_splat((vec_s16_t)vABCD, 5);
    const vec_s16_t vD = vec_splat((vec_s16_t)vABCD, 7);
    LOAD_ZERO;
    const vec_s16_t v32ss = vec_sl(vec_splat_s16(1),vec_splat_u16(5));
    const vec_u16_t v6us = vec_splat_u16(6);
    register int loadSecond = (((unsigned long)src) % 16) <= 7 ? 0 : 1;
    register int reallyBadAlign = (((unsigned long)src) % 16) == 15 ? 1 : 0;

    vec_u8_t vsrcAuc, vsrcBuc, vsrcperm0, vsrcperm1;
    vec_u8_t vsrc0uc, vsrc1uc;
    vec_s16_t vsrc0ssH, vsrc1ssH;
    vec_u8_t vsrcCuc, vsrc2uc, vsrc3uc;
    vec_s16_t vsrc2ssH, vsrc3ssH, psum;
    vec_u8_t vdst, ppsum, vfdst, fsum;

  POWERPC_PERF_START_COUNT(PREFIX_h264_chroma_mc8_num, 1);

    if (((unsigned long)dst) % 16 == 0) {
      fperm = (vec_u8_t)AVV(0x10, 0x11, 0x12, 0x13,
                            0x14, 0x15, 0x16, 0x17,
                            0x08, 0x09, 0x0A, 0x0B,
                            0x0C, 0x0D, 0x0E, 0x0F);
    } else {
      fperm = (vec_u8_t)AVV(0x00, 0x01, 0x02, 0x03,
                            0x04, 0x05, 0x06, 0x07,
                            0x18, 0x19, 0x1A, 0x1B,
                            0x1C, 0x1D, 0x1E, 0x1F);
    }

    vsrcAuc = vec_ld(0, src);

    if (loadSecond)
      vsrcBuc = vec_ld(16, src);
    vsrcperm0 = vec_lvsl(0, src);
    vsrcperm1 = vec_lvsl(1, src);

    vsrc0uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm0);
    if (reallyBadAlign)
      vsrc1uc = vsrcBuc;
    else
      vsrc1uc = vec_perm(vsrcAuc, vsrcBuc, vsrcperm1);

    vsrc0ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc0uc);
    vsrc1ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc1uc);

    if (!loadSecond) {// -> !reallyBadAlign
      for (i = 0 ; i < h ; i++) {


        vsrcCuc = vec_ld(stride + 0, src);

        vsrc2uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm0);
        vsrc3uc = vec_perm(vsrcCuc, vsrcCuc, vsrcperm1);

        vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc2uc);
        vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc3uc);

        psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
        psum = vec_mladd(vB, vsrc1ssH, psum);
        psum = vec_mladd(vC, vsrc2ssH, psum);
        psum = vec_mladd(vD, vsrc3ssH, psum);
        psum = vec_add(v32ss, psum);
        psum = vec_sra(psum, v6us);

        vdst = vec_ld(0, dst);
        ppsum = (vec_u8_t)vec_packsu(psum, psum);
        vfdst = vec_perm(vdst, ppsum, fperm);

        OP_U8_ALTIVEC(fsum, vfdst, vdst);

        vec_st(fsum, 0, dst);

        vsrc0ssH = vsrc2ssH;
        vsrc1ssH = vsrc3ssH;

        dst += stride;
        src += stride;
      }
    } else {
        vec_u8_t vsrcDuc;
      for (i = 0 ; i < h ; i++) {
        vsrcCuc = vec_ld(stride + 0, src);
        vsrcDuc = vec_ld(stride + 16, src);

        vsrc2uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm0);
        if (reallyBadAlign)
          vsrc3uc = vsrcDuc;
        else
          vsrc3uc = vec_perm(vsrcCuc, vsrcDuc, vsrcperm1);

        vsrc2ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc2uc);
        vsrc3ssH = (vec_s16_t)vec_mergeh(zero_u8v,(vec_u8_t)vsrc3uc);

        psum = vec_mladd(vA, vsrc0ssH, vec_splat_s16(0));
        psum = vec_mladd(vB, vsrc1ssH, psum);
        psum = vec_mladd(vC, vsrc2ssH, psum);
        psum = vec_mladd(vD, vsrc3ssH, psum);
        psum = vec_add(v32ss, psum);
        psum = vec_sr(psum, v6us);

        vdst = vec_ld(0, dst);
        ppsum = (vec_u8_t)vec_pack(psum, psum);
        vfdst = vec_perm(vdst, ppsum, fperm);

        OP_U8_ALTIVEC(fsum, vfdst, vdst);

        vec_st(fsum, 0, dst);

        vsrc0ssH = vsrc2ssH;
        vsrc1ssH = vsrc3ssH;

        dst += stride;
        src += stride;
      }
    }
    POWERPC_PERF_STOP_COUNT(PREFIX_h264_chroma_mc8_num, 1);
}

/* this code assume stride % 16 == 0 */
static void PREFIX_h264_qpel16_h_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
  POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_h_lowpass_num, 1);
  register int i;

  LOAD_ZERO;
  const vec_u8_t permM2 = vec_lvsl(-2, src);
  const vec_u8_t permM1 = vec_lvsl(-1, src);
  const vec_u8_t permP0 = vec_lvsl(+0, src);
  const vec_u8_t permP1 = vec_lvsl(+1, src);
  const vec_u8_t permP2 = vec_lvsl(+2, src);
  const vec_u8_t permP3 = vec_lvsl(+3, src);
  const vec_s16_t v5ss = vec_splat_s16(5);
  const vec_u16_t v5us = vec_splat_u16(5);
  const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
  const vec_s16_t v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4));

  vec_u8_t srcM2, srcM1, srcP0, srcP1, srcP2, srcP3;

  register int align = ((((unsigned long)src) - 2) % 16);

  vec_s16_t srcP0A, srcP0B, srcP1A, srcP1B,
                      srcP2A, srcP2B, srcP3A, srcP3B,
                      srcM1A, srcM1B, srcM2A, srcM2B,
                      sum1A, sum1B, sum2A, sum2B, sum3A, sum3B,
                      pp1A, pp1B, pp2A, pp2B, pp3A, pp3B,
                      psumA, psumB, sumA, sumB;

  vec_u8_t sum, vdst, fsum;

  POWERPC_PERF_START_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);

  for (i = 0 ; i < 16 ; i ++) {
    vec_u8_t srcR1 = vec_ld(-2, src);
    vec_u8_t srcR2 = vec_ld(14, src);

    switch (align) {
    default: {
      srcM2 = vec_perm(srcR1, srcR2, permM2);
      srcM1 = vec_perm(srcR1, srcR2, permM1);
      srcP0 = vec_perm(srcR1, srcR2, permP0);
      srcP1 = vec_perm(srcR1, srcR2, permP1);
      srcP2 = vec_perm(srcR1, srcR2, permP2);
      srcP3 = vec_perm(srcR1, srcR2, permP3);
    } break;
    case 11: {
      srcM2 = vec_perm(srcR1, srcR2, permM2);
      srcM1 = vec_perm(srcR1, srcR2, permM1);
      srcP0 = vec_perm(srcR1, srcR2, permP0);
      srcP1 = vec_perm(srcR1, srcR2, permP1);
      srcP2 = vec_perm(srcR1, srcR2, permP2);
      srcP3 = srcR2;
    } break;
    case 12: {
      vec_u8_t srcR3 = vec_ld(30, src);
      srcM2 = vec_perm(srcR1, srcR2, permM2);
      srcM1 = vec_perm(srcR1, srcR2, permM1);
      srcP0 = vec_perm(srcR1, srcR2, permP0);
      srcP1 = vec_perm(srcR1, srcR2, permP1);
      srcP2 = srcR2;
      srcP3 = vec_perm(srcR2, srcR3, permP3);
    } break;
    case 13: {
      vec_u8_t srcR3 = vec_ld(30, src);
      srcM2 = vec_perm(srcR1, srcR2, permM2);
      srcM1 = vec_perm(srcR1, srcR2, permM1);
      srcP0 = vec_perm(srcR1, srcR2, permP0);
      srcP1 = srcR2;
      srcP2 = vec_perm(srcR2, srcR3, permP2);
      srcP3 = vec_perm(srcR2, srcR3, permP3);
    } break;
    case 14: {
      vec_u8_t srcR3 = vec_ld(30, src);
      srcM2 = vec_perm(srcR1, srcR2, permM2);
      srcM1 = vec_perm(srcR1, srcR2, permM1);
      srcP0 = srcR2;
      srcP1 = vec_perm(srcR2, srcR3, permP1);
      srcP2 = vec_perm(srcR2, srcR3, permP2);
      srcP3 = vec_perm(srcR2, srcR3, permP3);
    } break;
    case 15: {
      vec_u8_t srcR3 = vec_ld(30, src);
      srcM2 = vec_perm(srcR1, srcR2, permM2);
      srcM1 = srcR2;
      srcP0 = vec_perm(srcR2, srcR3, permP0);
      srcP1 = vec_perm(srcR2, srcR3, permP1);
      srcP2 = vec_perm(srcR2, srcR3, permP2);
      srcP3 = vec_perm(srcR2, srcR3, permP3);
    } break;
    }

    srcP0A = (vec_s16_t) vec_mergeh(zero_u8v, srcP0);
    srcP0B = (vec_s16_t) vec_mergel(zero_u8v, srcP0);
    srcP1A = (vec_s16_t) vec_mergeh(zero_u8v, srcP1);
    srcP1B = (vec_s16_t) vec_mergel(zero_u8v, srcP1);

    srcP2A = (vec_s16_t) vec_mergeh(zero_u8v, srcP2);
    srcP2B = (vec_s16_t) vec_mergel(zero_u8v, srcP2);
    srcP3A = (vec_s16_t) vec_mergeh(zero_u8v, srcP3);
    srcP3B = (vec_s16_t) vec_mergel(zero_u8v, srcP3);

    srcM1A = (vec_s16_t) vec_mergeh(zero_u8v, srcM1);
    srcM1B = (vec_s16_t) vec_mergel(zero_u8v, srcM1);
    srcM2A = (vec_s16_t) vec_mergeh(zero_u8v, srcM2);
    srcM2B = (vec_s16_t) vec_mergel(zero_u8v, srcM2);

    sum1A = vec_adds(srcP0A, srcP1A);
    sum1B = vec_adds(srcP0B, srcP1B);
    sum2A = vec_adds(srcM1A, srcP2A);
    sum2B = vec_adds(srcM1B, srcP2B);
    sum3A = vec_adds(srcM2A, srcP3A);
    sum3B = vec_adds(srcM2B, srcP3B);

    pp1A = vec_mladd(sum1A, v20ss, v16ss);
    pp1B = vec_mladd(sum1B, v20ss, v16ss);

    pp2A = vec_mladd(sum2A, v5ss, zero_s16v);
    pp2B = vec_mladd(sum2B, v5ss, zero_s16v);

    pp3A = vec_add(sum3A, pp1A);
    pp3B = vec_add(sum3B, pp1B);

    psumA = vec_sub(pp3A, pp2A);
    psumB = vec_sub(pp3B, pp2B);

    sumA = vec_sra(psumA, v5us);
    sumB = vec_sra(psumB, v5us);

    sum = vec_packsu(sumA, sumB);

    ASSERT_ALIGNED(dst);
    vdst = vec_ld(0, dst);

    OP_U8_ALTIVEC(fsum, sum, vdst);

    vec_st(fsum, 0, dst);

    src += srcStride;
    dst += dstStride;
  }
POWERPC_PERF_STOP_COUNT(PREFIX_h264_qpel16_h_lowpass_num, 1);
}

/* this code assume stride % 16 == 0 */
static void PREFIX_h264_qpel16_v_lowpass_altivec(uint8_t * dst, uint8_t * src, int dstStride, int srcStride) {
  POWERPC_PERF_DECLARE(PREFIX_h264_qpel16_v_lowpass_num, 1);

  register int i;

  LOAD_ZERO;
  const vec_u8_t perm = vec_lvsl(0, src);
  const vec_s16_t v20ss = vec_sl(vec_splat_s16(5),vec_splat_u16(2));
  const vec_u16_t v5us = vec_splat_u16(5);
  const vec_s16_t v5ss = vec_splat_s16(5);
  const vec_s16_t v16ss = vec_sl(vec_splat_s16(1),vec_splat_u16(4));

  uint8_t *srcbis = src - (srcStride * 2);

  const vec_u8_t srcM2a = vec_ld(0, srcbis);
  const vec_u8_t srcM2b = vec_ld(16, srcbis);
  const vec_u8_t srcM2 = vec_perm(srcM2a, srcM2b, perm);
//  srcbis += srcStride;
  const vec_u8_t srcM1a = vec_ld(0, srcbis += srcStride);
  const vec_u8_t srcM1b = vec_ld(16, srcbis);
  const vec_u8_t srcM1 = vec_perm(srcM1a, srcM1b, perm);
//  srcbis += srcStride;