annotate alpha/simple_idct_alpha.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.

/*
 * Simple IDCT (Alpha optimized)
 *
 * Copyright (c) 2001 Michael Niedermayer <[e-mail]>
 *
 * based upon some outcommented C code from mpeg2dec (idct_mmx.c
 * written by Aaron Holtzman <[e-mail]>)
 *
 * Alpha optimizations by Måns Rullgård <[e-mail]>
 *                     and Falk Hueffner <[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 "asm.h"

extern void (*put_pixels_clamped_axp_p)(const DCTELEM *block, uint8_t *pixels,
                                        int line_size);
extern void (*add_pixels_clamped_axp_p)(const DCTELEM *block, uint8_t *pixels,
                                        int line_size);

// cos(i * M_PI / 16) * sqrt(2) * (1 << 14)
// W4 is actually exactly 16384, but using 16383 works around
// accumulating rounding errors for some encoders
#define W1 ((int_fast32_t) 22725)
#define W2 ((int_fast32_t) 21407)
#define W3 ((int_fast32_t) 19266)
#define W4 ((int_fast32_t) 16383)
#define W5 ((int_fast32_t) 12873)
#define W6 ((int_fast32_t)  8867)
#define W7 ((int_fast32_t)  4520)
#define ROW_SHIFT 11
#define COL_SHIFT 20

/* 0: all entries 0, 1: only first entry nonzero, 2: otherwise  */
static inline int idct_row(DCTELEM *row)
{
    int_fast32_t a0, a1, a2, a3, b0, b1, b2, b3, t;
    uint64_t l, r, t2;
    l = ldq(row);
    r = ldq(row + 4);

    if (l == 0 && r == 0)
        return 0;

    a0 = W4 * sextw(l) + (1 << (ROW_SHIFT - 1));

    if (((l & ~0xffffUL) | r) == 0) {
        a0 >>= ROW_SHIFT;
        t2 = (uint16_t) a0;
        t2 |= t2 << 16;
        t2 |= t2 << 32;

        stq(t2, row);
        stq(t2, row + 4);
        return 1;
    }

    a1 = a0;
    a2 = a0;
    a3 = a0;

    t = extwl(l, 4);            /* row[2] */
    if (t != 0) {
        t = sextw(t);
        a0 += W2 * t;
        a1 += W6 * t;
        a2 -= W6 * t;
        a3 -= W2 * t;
    }

    t = extwl(r, 0);            /* row[4] */
    if (t != 0) {
        t = sextw(t);
        a0 += W4 * t;
        a1 -= W4 * t;
        a2 -= W4 * t;
        a3 += W4 * t;
    }

    t = extwl(r, 4);            /* row[6] */
    if (t != 0) {
        t = sextw(t);
        a0 += W6 * t;
        a1 -= W2 * t;
        a2 += W2 * t;
        a3 -= W6 * t;
    }

    t = extwl(l, 2);            /* row[1] */
    if (t != 0) {
        t = sextw(t);
        b0 = W1 * t;
        b1 = W3 * t;
        b2 = W5 * t;
        b3 = W7 * t;
    } else {
        b0 = 0;
        b1 = 0;
        b2 = 0;
        b3 = 0;
    }

    t = extwl(l, 6);            /* row[3] */
    if (t) {
        t = sextw(t);
        b0 += W3 * t;
        b1 -= W7 * t;
        b2 -= W1 * t;
        b3 -= W5 * t;
    }


    t = extwl(r, 2);            /* row[5] */
    if (t) {
        t = sextw(t);
        b0 += W5 * t;
        b1 -= W1 * t;
        b2 += W7 * t;
        b3 += W3 * t;
    }

    t = extwl(r, 6);            /* row[7] */
    if (t) {
        t = sextw(t);
        b0 += W7 * t;
        b1 -= W5 * t;
        b2 += W3 * t;
        b3 -= W1 * t;
    }

    row[0] = (a0 + b0) >> ROW_SHIFT;
    row[1] = (a1 + b1) >> ROW_SHIFT;
    row[2] = (a2 + b2) >> ROW_SHIFT;
    row[3] = (a3 + b3) >> ROW_SHIFT;
    row[4] = (a3 - b3) >> ROW_SHIFT;
    row[5] = (a2 - b2) >> ROW_SHIFT;
    row[6] = (a1 - b1) >> ROW_SHIFT;
    row[7] = (a0 - b0) >> ROW_SHIFT;

    return 2;
}

static inline void idct_col(DCTELEM *col)
{
    int_fast32_t a0, a1, a2, a3, b0, b1, b2, b3;

    col[0] += (1 << (COL_SHIFT - 1)) / W4;

    a0 = W4 * col[8 * 0];
    a1 = W4 * col[8 * 0];
    a2 = W4 * col[8 * 0];
    a3 = W4 * col[8 * 0];

    if (col[8 * 2]) {
        a0 += W2 * col[8 * 2];
        a1 += W6 * col[8 * 2];
        a2 -= W6 * col[8 * 2];
        a3 -= W2 * col[8 * 2];
    }

    if (col[8 * 4]) {
        a0 += W4 * col[8 * 4];
        a1 -= W4 * col[8 * 4];
        a2 -= W4 * col[8 * 4];
        a3 += W4 * col[8 * 4];
    }

    if (col[8 * 6]) {
        a0 += W6 * col[8 * 6];
        a1 -= W2 * col[8 * 6];
        a2 += W2 * col[8 * 6];
        a3 -= W6 * col[8 * 6];
    }

    if (col[8 * 1]) {
        b0 = W1 * col[8 * 1];
        b1 = W3 * col[8 * 1];
        b2 = W5 * col[8 * 1];
        b3 = W7 * col[8 * 1];
    } else {
        b0 = 0;
        b1 = 0;
        b2 = 0;
        b3 = 0;
    }

    if (col[8 * 3]) {
        b0 += W3 * col[8 * 3];
        b1 -= W7 * col[8 * 3];
        b2 -= W1 * col[8 * 3];
        b3 -= W5 * col[8 * 3];
    }

    if (col[8 * 5]) {
        b0 += W5 * col[8 * 5];
        b1 -= W1 * col[8 * 5];
        b2 += W7 * col[8 * 5];
        b3 += W3 * col[8 * 5];
    }

    if (col[8 * 7]) {
        b0 += W7 * col[8 * 7];
        b1 -= W5 * col[8 * 7];
        b2 += W3 * col[8 * 7];
        b3 -= W1 * col[8 * 7];
    }

    col[8 * 0] = (a0 + b0) >> COL_SHIFT;
    col[8 * 7] = (a0 - b0) >> COL_SHIFT;
    col[8 * 1] = (a1 + b1) >> COL_SHIFT;
    col[8 * 6] = (a1 - b1) >> COL_SHIFT;
    col[8 * 2] = (a2 + b2) >> COL_SHIFT;
    col[8 * 5] = (a2 - b2) >> COL_SHIFT;
    col[8 * 3] = (a3 + b3) >> COL_SHIFT;
    col[8 * 4] = (a3 - b3) >> COL_SHIFT;
}

/* If all rows but the first one are zero after row transformation,
   all rows will be identical after column transformation.  */
static inline void idct_col2(DCTELEM *col)
{
    int i;
    uint64_t l, r;

    for (i = 0; i < 8; ++i) {
        int_fast32_t a0 = col[i] + (1 << (COL_SHIFT - 1)) / W4;

        a0 *= W4;
        col[i] = a0 >> COL_SHIFT;
    }

    l = ldq(col + 0 * 4); r = ldq(col + 1 * 4);
    stq(l, col +  2 * 4); stq(r, col +  3 * 4);
    stq(l, col +  4 * 4); stq(r, col +  5 * 4);
    stq(l, col +  6 * 4); stq(r, col +  7 * 4);
    stq(l, col +  8 * 4); stq(r, col +  9 * 4);
    stq(l, col + 10 * 4); stq(r, col + 11 * 4);
    stq(l, col + 12 * 4); stq(r, col + 13 * 4);
    stq(l, col + 14 * 4); stq(r, col + 15 * 4);
}

void ff_simple_idct_axp(DCTELEM *block)
{

    int i;
    int rowsZero = 1;           /* all rows except row 0 zero */
    int rowsConstant = 1;       /* all rows consist of a constant value */

    for (i = 0; i < 8; i++) {
        int sparseness = idct_row(block + 8 * i);

        if (i > 0 && sparseness > 0)
            rowsZero = 0;
        if (sparseness == 2)
            rowsConstant = 0;
    }

    if (rowsZero) {
        idct_col2(block);
    } else if (rowsConstant) {
        idct_col(block);
        for (i = 0; i < 8; i += 2) {
            uint64_t v = (uint16_t) block[0];
            uint64_t w = (uint16_t) block[8];

            v |= v << 16;
            w |= w << 16;
            v |= v << 32;
            w |= w << 32;
            stq(v, block + 0 * 4);
            stq(v, block + 1 * 4);
            stq(w, block + 2 * 4);
            stq(w, block + 3 * 4);
            block += 4 * 4;
        }
    } else {
        for (i = 0; i < 8; i++)
            idct_col(block + i);
    }
}

void ff_simple_idct_put_axp(uint8_t *dest, int line_size, DCTELEM *block)
{
    ff_simple_idct_axp(block);
    put_pixels_clamped_axp_p(block, dest, line_size);
}

void ff_simple_idct_add_axp(uint8_t *dest, int line_size, DCTELEM *block)
{
    ff_simple_idct_axp(block);
    add_pixels_clamped_axp_p(block, dest, line_size);
}