view ppc/gmc_altivec.c @ 11032:01bd040f8607 libavcodec

Unroll main loop so the edge==0 case is seperate. This allows many things to be simplified away. h264 decoder is overall 1% faster with a mbaff sample and 0.1% slower with the cathedral sample, probably because the slow loop filter code must be loaded into the code cache for each first MB of each row but isnt used for the following MBs.

/*
 * GMC (Global Motion Compensation)
 * AltiVec-enabled
 * Copyright (c) 2003 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
 */

#include "libavcodec/dsputil.h"
#include "dsputil_ppc.h"
#include "util_altivec.h"
#include "types_altivec.h"

/*
  altivec-enhanced gmc1. ATM this code assume stride is a multiple of 8,
  to preserve proper dst alignment.
*/
#define GMC1_PERF_COND (h==8)
void gmc1_altivec(uint8_t *dst /* align 8 */, uint8_t *src /* align1 */, int stride, int h, int x16, int y16, int rounder)
{
POWERPC_PERF_DECLARE(altivec_gmc1_num, GMC1_PERF_COND);
    const DECLARE_ALIGNED_16(unsigned short, rounder_a) = rounder;
    const DECLARE_ALIGNED_16(unsigned short, ABCD)[8] =
        {
            (16-x16)*(16-y16), /* A */
            (   x16)*(16-y16), /* B */
            (16-x16)*(   y16), /* C */
            (   x16)*(   y16), /* D */
            0, 0, 0, 0         /* padding */
        };
    register const vector unsigned char vczero = (const vector unsigned char)vec_splat_u8(0);
    register const vector unsigned short vcsr8 = (const vector unsigned short)vec_splat_u16(8);
    register vector unsigned char dstv, dstv2, src_0, src_1, srcvA, srcvB, srcvC, srcvD;
    register vector unsigned short Av, Bv, Cv, Dv, rounderV, tempA, tempB, tempC, tempD;
    int i;
    unsigned long dst_odd = (unsigned long)dst & 0x0000000F;
    unsigned long src_really_odd = (unsigned long)src & 0x0000000F;


POWERPC_PERF_START_COUNT(altivec_gmc1_num, GMC1_PERF_COND);

    tempA = vec_ld(0, (unsigned short*)ABCD);
    Av = vec_splat(tempA, 0);
    Bv = vec_splat(tempA, 1);
    Cv = vec_splat(tempA, 2);
    Dv = vec_splat(tempA, 3);

    rounderV = vec_splat((vec_u16)vec_lde(0, &rounder_a), 0);

    // we'll be able to pick-up our 9 char elements
    // at src from those 32 bytes
    // we load the first batch here, as inside the loop
    // we can re-use 'src+stride' from one iteration
    // as the 'src' of the next.
    src_0 = vec_ld(0, src);
    src_1 = vec_ld(16, src);
    srcvA = vec_perm(src_0, src_1, vec_lvsl(0, src));

    if (src_really_odd != 0x0000000F) {
        // if src & 0xF == 0xF, then (src+1) is properly aligned
        // on the second vector.
        srcvB = vec_perm(src_0, src_1, vec_lvsl(1, src));
    } else {
        srcvB = src_1;
    }
    srcvA = vec_mergeh(vczero, srcvA);
    srcvB = vec_mergeh(vczero, srcvB);

    for(i=0; i<h; i++) {
        dst_odd = (unsigned long)dst & 0x0000000F;
        src_really_odd = (((unsigned long)src) + stride) & 0x0000000F;

        dstv = vec_ld(0, dst);

        // we we'll be able to pick-up our 9 char elements
        // at src + stride from those 32 bytes
        // then reuse the resulting 2 vectors srvcC and srcvD
        // as the next srcvA and srcvB
        src_0 = vec_ld(stride + 0, src);
        src_1 = vec_ld(stride + 16, src);
        srcvC = vec_perm(src_0, src_1, vec_lvsl(stride + 0, src));

        if (src_really_odd != 0x0000000F) {
            // if src & 0xF == 0xF, then (src+1) is properly aligned
            // on the second vector.
            srcvD = vec_perm(src_0, src_1, vec_lvsl(stride + 1, src));
        } else {
            srcvD = src_1;
        }

        srcvC = vec_mergeh(vczero, srcvC);
        srcvD = vec_mergeh(vczero, srcvD);


        // OK, now we (finally) do the math :-)
        // those four instructions replaces 32 int muls & 32 int adds.
        // isn't AltiVec nice ?
        tempA = vec_mladd((vector unsigned short)srcvA, Av, rounderV);
        tempB = vec_mladd((vector unsigned short)srcvB, Bv, tempA);
        tempC = vec_mladd((vector unsigned short)srcvC, Cv, tempB);
        tempD = vec_mladd((vector unsigned short)srcvD, Dv, tempC);

        srcvA = srcvC;
        srcvB = srcvD;

        tempD = vec_sr(tempD, vcsr8);

        dstv2 = vec_pack(tempD, (vector unsigned short)vczero);

        if (dst_odd) {
            dstv2 = vec_perm(dstv, dstv2, vcprm(0,1,s0,s1));
        } else {
            dstv2 = vec_perm(dstv, dstv2, vcprm(s0,s1,2,3));
        }

        vec_st(dstv2, 0, dst);

        dst += stride;
        src += stride;
    }

POWERPC_PERF_STOP_COUNT(altivec_gmc1_num, GMC1_PERF_COND);
}