comparison x86/fdct_mmx.c @ 11508:7be32921237f libavcodec

Replace remaining uses of ATTR_ALIGNED with DECLARE_ALIGNED

11507:25e209f9153a

 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/common.h"
#include "libavcodec/dsputil.h"

#define ATTR_ALIGN(align) __attribute__ ((__aligned__ (align)))

//////////////////////////////////////////////////////////////////////
//
// constants for the forward DCT
// -----------------------------

//#define RND_FRW_COL    (1 << (SHIFT_FRW_COL-1))

#define X8(x) x,x,x,x,x,x,x,x

//concatenated table, for forward DCT transformation
static const int16_t fdct_tg_all_16[24] ATTR_ALIGN(16) = {
    X8(13036),  // tg * (2<<16) + 0.5
    X8(27146),  // tg * (2<<16) + 0.5
    X8(-21746)  // tg * (2<<16) + 0.5
};

static const int16_t ocos_4_16[8] ATTR_ALIGN(16) = {
    X8(23170)   //cos * (2<<15) + 0.5
};

static const int16_t fdct_one_corr[8] ATTR_ALIGN(16) = { X8(1) };

static const int32_t fdct_r_row[2] ATTR_ALIGN(8) = {RND_FRW_ROW, RND_FRW_ROW };

static struct
{
 const int32_t fdct_r_row_sse2[4] ATTR_ALIGN(16);
} fdct_r_row_sse2 ATTR_ALIGN(16)=
{{
 RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW
}};
//static const long fdct_r_row_sse2[4] ATTR_ALIGN(16) = {RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW};

static const int16_t tab_frw_01234567[] ATTR_ALIGN(8) = {  // forward_dct coeff table
  16384,   16384,   22725,   19266,
  16384,   16384,   12873,    4520,
  21407,    8867,   19266,   -4520,
  -8867,  -21407,  -22725,  -12873,
  16384,  -16384,   12873,  -22725,

  29692,  -12299,   26722,  -31521,
};

static struct
{
 const int16_t tab_frw_01234567_sse2[256] ATTR_ALIGN(16);
} tab_frw_01234567_sse2 ATTR_ALIGN(16) =
{{
//static const int16_t tab_frw_01234567_sse2[] ATTR_ALIGN(16) = {  // forward_dct coeff table
#define TABLE_SSE2 C4,  C4,  C1,  C3, -C6, -C2, -C1, -C5, \
                   C4,  C4,  C5,  C7,  C2,  C6,  C3, -C7, \
                  -C4,  C4,  C7,  C3,  C6, -C2,  C7, -C5, \
                   C4, -C4,  C5, -C1,  C2, -C6,  C3, -C1,
// c1..c7 * cos(pi/4) * 2^15

        : "r" (in), "r" (table), "r" (fdct_r_row), "r" (out));
}

void ff_fdct_mmx(int16_t *block)
{
    int64_t align_tmp[16] ATTR_ALIGN(8);
    int16_t * block1= (int16_t*)align_tmp;
    const int16_t *table= tab_frw_01234567;
    int i;

    fdct_col_mmx(block, block1, 0);

    }
}

void ff_fdct_mmx2(int16_t *block)
{
    int64_t align_tmp[16] ATTR_ALIGN(8);
    int16_t *block1= (int16_t*)align_tmp;
    const int16_t *table= tab_frw_01234567;
    int i;

    fdct_col_mmx(block, block1, 0);

    }
}

void ff_fdct_sse2(int16_t *block)
{
    int64_t align_tmp[16] ATTR_ALIGN(16);
    int16_t * const block1= (int16_t*)align_tmp;

    fdct_col_sse2(block, block1, 0);
    fdct_row_sse2(block1, block);
}

11508:7be32921237f

 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/common.h"
#include "libavcodec/dsputil.h"

//////////////////////////////////////////////////////////////////////
//
// constants for the forward DCT
// -----------------------------

//#define RND_FRW_COL    (1 << (SHIFT_FRW_COL-1))

#define X8(x) x,x,x,x,x,x,x,x

//concatenated table, for forward DCT transformation
DECLARE_ALIGNED(16, static const int16_t, fdct_tg_all_16)[24] = {
    X8(13036),  // tg * (2<<16) + 0.5
    X8(27146),  // tg * (2<<16) + 0.5
    X8(-21746)  // tg * (2<<16) + 0.5
};

DECLARE_ALIGNED(16, static const int16_t, ocos_4_16)[8] = {
    X8(23170)   //cos * (2<<15) + 0.5
};

DECLARE_ALIGNED(16, static const int16_t, fdct_one_corr)[8] = { X8(1) };

DECLARE_ALIGNED(8, static const int32_t, fdct_r_row)[2] = {RND_FRW_ROW, RND_FRW_ROW };

static struct
{
 DECLARE_ALIGNED(16, const int32_t, fdct_r_row_sse2)[4];
} fdct_r_row_sse2 =
{{
 RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW
}};
//DECLARE_ALIGNED(16, static const long, fdct_r_row_sse2)[4] = {RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW, RND_FRW_ROW};

DECLARE_ALIGNED(8, static const int16_t, tab_frw_01234567)[] = {  // forward_dct coeff table
  16384,   16384,   22725,   19266,
  16384,   16384,   12873,    4520,
  21407,    8867,   19266,   -4520,
  -8867,  -21407,  -22725,  -12873,
  16384,  -16384,   12873,  -22725,

  29692,  -12299,   26722,  -31521,
};

static struct
{
 DECLARE_ALIGNED(16, const int16_t, tab_frw_01234567_sse2)[256];
} tab_frw_01234567_sse2 =
{{
//DECLARE_ALIGNED(16, static const int16_t, tab_frw_01234567_sse2)[] = {  // forward_dct coeff table
#define TABLE_SSE2 C4,  C4,  C1,  C3, -C6, -C2, -C1, -C5, \
                   C4,  C4,  C5,  C7,  C2,  C6,  C3, -C7, \
                  -C4,  C4,  C7,  C3,  C6, -C2,  C7, -C5, \
                   C4, -C4,  C5, -C1,  C2, -C6,  C3, -C1,
// c1..c7 * cos(pi/4) * 2^15

        : "r" (in), "r" (table), "r" (fdct_r_row), "r" (out));
}

void ff_fdct_mmx(int16_t *block)
{
    DECLARE_ALIGNED(8, int64_t, align_tmp)[16];
    int16_t * block1= (int16_t*)align_tmp;
    const int16_t *table= tab_frw_01234567;
    int i;

    fdct_col_mmx(block, block1, 0);

    }
}

void ff_fdct_mmx2(int16_t *block)
{
    DECLARE_ALIGNED(8, int64_t, align_tmp)[16];
    int16_t *block1= (int16_t*)align_tmp;
    const int16_t *table= tab_frw_01234567;
    int i;

    fdct_col_mmx(block, block1, 0);

    }
}

void ff_fdct_sse2(int16_t *block)
{
    DECLARE_ALIGNED(16, int64_t, align_tmp)[16];
    int16_t * const block1= (int16_t*)align_tmp;

    fdct_col_sse2(block, block1, 0);
    fdct_row_sse2(block1, block);
}