vc1dsp_mmx.c.svn-base

mediastreamer2是开源的网络传输媒体流的库

/*
 * VC-1 and WMV3 - DSP functions MMX-optimized
 * Copyright (c) 2007 Christophe GISQUET <[e-mail]>
 *
 * Permission is hereby granted, free of charge, to any person
 * obtaining a copy of this software and associated documentation
 * files (the "Software"), to deal in the Software without
 * restriction, including without limitation the rights to use,
 * copy, modify, merge, publish, distribute, sublicense, and/or sell
 * copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following
 * conditions:
 *
 * The above copyright notice and this permission notice shall be
 * included in all copies or substantial portions of the Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
 * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
 * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
 * OTHER DEALINGS IN THE SOFTWARE.
 */

#include "dsputil.h"
#include "dsputil_mmx.h"
#include "x86_cpu.h"

/** Add rounder from mm7 to mm3 and pack result at destination */
#define NORMALIZE_MMX(SHIFT)                                    \
     "paddw     %%mm7, %%mm3           \n\t" /* +bias-r */      \
     "paddw     %%mm7, %%mm4           \n\t" /* +bias-r */      \
     "psraw     "SHIFT", %%mm3         \n\t"                    \
     "psraw     "SHIFT", %%mm4         \n\t"

#define TRANSFER_DO_PACK                        \
     "packuswb  %%mm4, %%mm3           \n\t"    \
     "movq      %%mm3, (%2)            \n\t"

#define TRANSFER_DONT_PACK                      \
     "movq      %%mm3, 0(%2)           \n\t"    \
     "movq      %%mm4, 8(%2)           \n\t"

/** @see MSPEL_FILTER13_CORE for use as UNPACK macro */
#define DO_UNPACK(reg)  "punpcklbw %%mm0, " reg "\n\t"
#define DONT_UNPACK(reg)

/** Compute the rounder 32-r or 8-r and unpacks it to mm7 */
#define LOAD_ROUNDER_MMX(ROUND)                 \
     "movd      "ROUND", %%mm7         \n\t"    \
     "punpcklwd %%mm7, %%mm7           \n\t"    \
     "punpckldq %%mm7, %%mm7           \n\t"

#define SHIFT2_LINE(OFF, R0,R1,R2,R3)           \
    "paddw     %%mm"#R2", %%mm"#R1"    \n\t"    \
    "movd      (%0,%3), %%mm"#R0"      \n\t"    \
    "pmullw    %%mm6, %%mm"#R1"        \n\t"    \
    "punpcklbw %%mm0, %%mm"#R0"        \n\t"    \
    "movd      (%0,%2), %%mm"#R3"      \n\t"    \
    "psubw     %%mm"#R0", %%mm"#R1"    \n\t"    \
    "punpcklbw %%mm0, %%mm"#R3"        \n\t"    \
    "paddw     %%mm7, %%mm"#R1"        \n\t"    \
    "psubw     %%mm"#R3", %%mm"#R1"    \n\t"    \
    "psraw     %4, %%mm"#R1"           \n\t"    \
    "movq      %%mm"#R1", "#OFF"(%1)   \n\t"    \
    "add       %2, %0                  \n\t"

DECLARE_ALIGNED_16(const uint64_t, ff_pw_9) = 0x0009000900090009ULL;

/** Sacrifying mm6 allows to pipeline loads from src */
static void vc1_put_ver_16b_shift2_mmx(int16_t *dst,
                                       const uint8_t *src, long int stride,
                                       int rnd, int64_t shift)
{
    asm volatile(
        "mov       $3, %%"REG_c"           \n\t"
        LOAD_ROUNDER_MMX("%5")
        "movq      "MANGLE(ff_pw_9)", %%mm6 \n\t"
        "1:                                \n\t"
        "movd      (%0), %%mm2             \n\t"
        "add       %2, %0                  \n\t"
        "movd      (%0), %%mm3             \n\t"
        "punpcklbw %%mm0, %%mm2            \n\t"
        "punpcklbw %%mm0, %%mm3            \n\t"
        SHIFT2_LINE(  0, 1, 2, 3, 4)
        SHIFT2_LINE( 24, 2, 3, 4, 1)
        SHIFT2_LINE( 48, 3, 4, 1, 2)
        SHIFT2_LINE( 72, 4, 1, 2, 3)
        SHIFT2_LINE( 96, 1, 2, 3, 4)
        SHIFT2_LINE(120, 2, 3, 4, 1)
        SHIFT2_LINE(144, 3, 4, 1, 2)
        SHIFT2_LINE(168, 4, 1, 2, 3)
        "sub       %6, %0                  \n\t"
        "add       $8, %1                  \n\t"
        "dec       %%"REG_c"               \n\t"
        "jnz 1b                            \n\t"
        : "+r"(src), "+r"(dst)
        : "r"(stride), "r"(-2*stride),
          "m"(shift), "m"(rnd), "r"(9*stride-4)
        : "%"REG_c, "memory"
    );
}

/**
 * Data is already unpacked, so some operations can directly be made from
 * memory.
 */
static void vc1_put_hor_16b_shift2_mmx(uint8_t *dst, long int stride,
                                       const int16_t *src, int rnd)
{
    int h = 8;

    src -= 1;
    rnd -= (-1+9+9-1)*1024; /* Add -1024 bias */
    asm volatile(
        LOAD_ROUNDER_MMX("%4")
        "movq      "MANGLE(ff_pw_128)", %%mm6\n\t"
        "movq      "MANGLE(ff_pw_9)", %%mm5 \n\t"
        "1:                                \n\t"
        "movq      2*0+0(%1), %%mm1        \n\t"
        "movq      2*0+8(%1), %%mm2        \n\t"
        "movq      2*1+0(%1), %%mm3        \n\t"
        "movq      2*1+8(%1), %%mm4        \n\t"
        "paddw     2*3+0(%1), %%mm1        \n\t"
        "paddw     2*3+8(%1), %%mm2        \n\t"
        "paddw     2*2+0(%1), %%mm3        \n\t"
        "paddw     2*2+8(%1), %%mm4        \n\t"
        "pmullw    %%mm5, %%mm3            \n\t"
        "pmullw    %%mm5, %%mm4            \n\t"
        "psubw     %%mm1, %%mm3            \n\t"
        "psubw     %%mm2, %%mm4            \n\t"
        NORMALIZE_MMX("$7")
        /* Remove bias */
        "paddw     %%mm6, %%mm3            \n\t"
        "paddw     %%mm6, %%mm4            \n\t"
        TRANSFER_DO_PACK
        "add       $24, %1                 \n\t"
        "add       %3, %2                  \n\t"
        "decl      %0                      \n\t"
        "jnz 1b                            \n\t"
        : "+r"(h), "+r" (src),  "+r" (dst)
        : "r"(stride), "m"(rnd)
        : "memory"
    );
}


/**
 * Purely vertical or horizontal 1/2 shift interpolation.
 * Sacrify mm6 for *9 factor.
 */
static void vc1_put_shift2_mmx(uint8_t *dst, const uint8_t *src,
                               long int stride, int rnd, long int offset)
{
    rnd = 8-rnd;
    asm volatile(
        "mov       $8, %%"REG_c"           \n\t"
        LOAD_ROUNDER_MMX("%5")
        "movq      "MANGLE(ff_pw_9)", %%mm6\n\t"
        "1:                                \n\t"
        "movd      0(%0   ), %%mm3         \n\t"
        "movd      4(%0   ), %%mm4         \n\t"
        "movd      0(%0,%2), %%mm1         \n\t"
        "movd      4(%0,%2), %%mm2         \n\t"
        "add       %2, %0                  \n\t"
        "punpcklbw %%mm0, %%mm3            \n\t"
        "punpcklbw %%mm0, %%mm4            \n\t"
        "punpcklbw %%mm0, %%mm1            \n\t"
        "punpcklbw %%mm0, %%mm2            \n\t"
        "paddw     %%mm1, %%mm3            \n\t"
        "paddw     %%mm2, %%mm4            \n\t"
        "movd      0(%0,%3), %%mm1         \n\t"
        "movd      4(%0,%3), %%mm2         \n\t"
        "pmullw    %%mm6, %%mm3            \n\t" /* 0,9,9,0*/
        "pmullw    %%mm6, %%mm4            \n\t" /* 0,9,9,0*/
        "punpcklbw %%mm0, %%mm1            \n\t"
        "punpcklbw %%mm0, %%mm2            \n\t"
        "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,0*/
        "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,0*/
        "movd      0(%0,%2), %%mm1         \n\t"
        "movd      4(%0,%2), %%mm2         \n\t"
        "punpcklbw %%mm0, %%mm1            \n\t"
        "punpcklbw %%mm0, %%mm2            \n\t"
        "psubw     %%mm1, %%mm3            \n\t" /*-1,9,9,-1*/
        "psubw     %%mm2, %%mm4            \n\t" /*-1,9,9,-1*/
        NORMALIZE_MMX("$4")
        "packuswb  %%mm4, %%mm3            \n\t"
        "movq      %%mm3, (%1)             \n\t"
        "add       %6, %0                  \n\t"
        "add       %4, %1                  \n\t"
        "dec       %%"REG_c"               \n\t"
        "jnz 1b                            \n\t"
        : "+r"(src),  "+r"(dst)
        : "r"(offset), "r"(-2*offset), "g"(stride), "m"(rnd),
          "g"(stride-offset)
        : "%"REG_c, "memory"
    );
}

/**
 * Filter coefficients made global to allow access by all 1 or 3 quarter shift
 * interpolation functions.
 */
DECLARE_ALIGNED_16(const uint64_t, ff_pw_53) = 0x0035003500350035ULL;
DECLARE_ALIGNED_16(const uint64_t, ff_pw_18) = 0x0012001200120012ULL;

/**
 * Core of the 1/4 and 3/4 shift bicubic interpolation.
 *
 * @param UNPACK  Macro unpacking arguments from 8 to 16bits (can be empty).
 * @param MOVQ    "movd 1" or "movq 2", if data read is already unpacked.
 * @param A1      Address of 1st tap (beware of unpacked/packed).
 * @param A2      Address of 2nd tap
 * @param A3      Address of 3rd tap
 * @param A4      Address of 4th tap
 */
#define MSPEL_FILTER13_CORE(UNPACK, MOVQ, A1, A2, A3, A4)       \
     MOVQ "*0+"A1", %%mm1       \n\t"                           \
     MOVQ "*4+"A1", %%mm2       \n\t"                           \
     UNPACK("%%mm1")                                            \
     UNPACK("%%mm2")                                            \
     "pmullw    "MANGLE(ff_pw_3)", %%mm1\n\t"                   \
     "pmullw    "MANGLE(ff_pw_3)", %%mm2\n\t"                   \
     MOVQ "*0+"A2", %%mm3       \n\t"                           \
     MOVQ "*4+"A2", %%mm4       \n\t"                           \
     UNPACK("%%mm3")                                            \
     UNPACK("%%mm4")                                            \
     "pmullw    %%mm6, %%mm3    \n\t" /* *18 */                 \
     "pmullw    %%mm6, %%mm4    \n\t" /* *18 */                 \
     "psubw     %%mm1, %%mm3    \n\t" /* 18,-3 */               \
     "psubw     %%mm2, %%mm4    \n\t" /* 18,-3 */               \
     MOVQ "*0+"A4", %%mm1       \n\t"                           \
     MOVQ "*4+"A4", %%mm2       \n\t"                           \
     UNPACK("%%mm1")                                            \
     UNPACK("%%mm2")                                            \
     "psllw     $2, %%mm1       \n\t" /* 4* */                  \
     "psllw     $2, %%mm2       \n\t" /* 4* */                  \
     "psubw     %%mm1, %%mm3    \n\t" /* -4,18,-3 */            \
     "psubw     %%mm2, %%mm4    \n\t" /* -4,18,-3 */            \
     MOVQ "*0+"A3", %%mm1       \n\t"                           \
     MOVQ "*4+"A3", %%mm2       \n\t"                           \
     UNPACK("%%mm1")                                            \
     UNPACK("%%mm2")                                            \
     "pmullw    %%mm5, %%mm1    \n\t" /* *53 */                 \