vp3dsp_sse2.c

这是著名的TCPMP播放器在WINDWOWS,和WINCE下编译通过的源程序.笔者对其中的LIBMAD库做了针对ARM MPU的优化. 并增加了词幕功能.

/*
 * Copyright (C) 2004 the ffmpeg project
 *
 * This library 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 of the License, or (at your option) any later version.
 *
 * This library 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 this library; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */

/**
 * @file vp3dsp_sse2.c
 * SSE2-optimized functions cribbed from the original VP3 source code.
 */

#include "../dsputil.h"
#include "mmx.h"

static const unsigned short __align16 SSE2_dequant_const[] =
{
    0,65535,65535,0,0,0,0,0,    // 0x0000 0000 0000 0000 0000 FFFF FFFF 0000
    0,0,0,0,65535,65535,0,0,    // 0x0000 0000 FFFF FFFF 0000 0000 0000 0000
    65535,65535,65535,0,0,0,0,0,// 0x0000 0000 0000 0000 0000 FFFF FFFF FFFF
    0,0,0,65535,0,0,0,0,        // 0x0000 0000 0000 0000 FFFF 0000 0000 0000
    0,0,0,65535,65535,0,0,0,    // 0x0000 0000 0000 FFFF FFFF 0000 0000 0000
    65535,0,0,0,0,65535,0,0,    // 0x0000 0000 FFFF 0000 0000 0000 0000 FFFF
    0,0,65535,65535, 0,0,0,0    // 0x0000 0000 0000 0000 FFFF FFFF 0000 0000
};

static const unsigned int __align16 eight_data[] =
{ 
    0x00080008, 
    0x00080008,
    0x00080008, 
    0x00080008 
}; 

static const unsigned short __align16 SSE2_idct_data[7 * 8] =
{
    64277,64277,64277,64277,64277,64277,64277,64277, 
    60547,60547,60547,60547,60547,60547,60547,60547, 
    54491,54491,54491,54491,54491,54491,54491,54491, 
    46341,46341,46341,46341,46341,46341,46341,46341, 
    36410,36410,36410,36410,36410,36410,36410,36410, 
    25080,25080,25080,25080,25080,25080,25080,25080, 
    12785,12785,12785,12785,12785,12785,12785,12785
};


#define SSE2_Column_IDCT() {        \
    \
    movdqu_m2r(*I(3), xmm2);     /* xmm2 = i3 */ \
    movdqu_m2r(*C(3), xmm6);     /* xmm6 = c3 */ \
    \
    movdqu_r2r(xmm2, xmm4);      /* xmm4 = i3 */ \
    movdqu_m2r(*I(5), xmm7);     /* xmm7 = i5 */ \
    \
    pmulhw_r2r(xmm6, xmm4);      /* xmm4 = c3 * i3 - i3 */ \
    movdqu_m2r(*C(5), xmm1);     /* xmm1 = c5 */ \
    \
    pmulhw_r2r(xmm7, xmm6);      /* xmm6 = c3 * i5 - i5 */ \
    movdqu_r2r(xmm1, xmm5);      /* xmm5 = c5 */ \
    \
    pmulhw_r2r(xmm2, xmm1);      /* xmm1 = c5 * i3 - i3 */ \
    movdqu_m2r(*I(1), xmm3);     /* xmm3 = i1 */ \
    \
    pmulhw_r2r(xmm7, xmm5);      /* xmm5 = c5 * i5 - i5 */ \
    movdqu_m2r(*C(1), xmm0);     /* xmm0 = c1 */ \
    \
    /* all registers are in use */ \
    \
    paddw_r2r(xmm2, xmm4);       /* xmm4 = c3 * i3 */ \
    paddw_r2r(xmm7, xmm6);       /* xmm6 = c3 * i5 */ \
    \
    paddw_r2r(xmm1, xmm2);       /* xmm2 = c5 * i3 */ \
    movdqu_m2r(*I(7), xmm1);     /* xmm1 = i7 */ \
    \
    paddw_r2r(xmm5, xmm7);       /* xmm7 = c5 * i5 */ \
    movdqu_r2r(xmm0, xmm5);      /* xmm5 = c1 */ \
    \
    pmulhw_r2r(xmm3, xmm0);      /* xmm0 = c1 * i1 - i1 */ \
    paddsw_r2r(xmm7, xmm4);      /* xmm4 = c3 * i3 + c5 * i5 = C */ \
    \
    pmulhw_r2r(xmm1, xmm5);      /* xmm5 = c1 * i7 - i7 */ \
    movdqu_m2r(*C(7), xmm7);     /* xmm7 = c7 */ \
    \
    psubsw_r2r(xmm2, xmm6);      /* xmm6 = c3 * i5 - c5 * i3 = D */ \
    paddw_r2r(xmm3, xmm0);       /* xmm0 = c1 * i1 */ \
    \
    pmulhw_r2r(xmm7, xmm3);      /* xmm3 = c7 * i1 */ \
    movdqu_m2r(*I(2), xmm2);     /* xmm2 = i2 */ \
    \
    pmulhw_r2r(xmm1, xmm7);      /* xmm7 = c7 * i7 */ \
    paddw_r2r(xmm1, xmm5);       /* xmm5 = c1 * i7 */ \
    \
    movdqu_r2r(xmm2, xmm1);      /* xmm1 = i2 */ \
    pmulhw_m2r(*C(2), xmm2);     /* xmm2 = i2 * c2 -i2 */ \
    \
    psubsw_r2r(xmm5, xmm3);      /* xmm3 = c7 * i1 - c1 * i7 = B */ \
    movdqu_m2r(*I(6), xmm5);     /* xmm5 = i6 */ \
    \
    paddsw_r2r(xmm7, xmm0);      /* xmm0 = c1 * i1 + c7 * i7 = A */ \
    movdqu_r2r(xmm5, xmm7);      /* xmm7 = i6 */ \
    \
    psubsw_r2r(xmm4, xmm0);      /* xmm0 = A - C */ \
    pmulhw_m2r(*C(2), xmm5);     /* xmm5 = c2 * i6 - i6 */ \
    \
    paddw_r2r(xmm1, xmm2);       /* xmm2 = i2 * c2 */ \
    pmulhw_m2r(*C(6), xmm1);     /* xmm1 = c6 * i2 */ \
    \
    paddsw_r2r(xmm4, xmm4);      /* xmm4 = C + C */ \
    paddsw_r2r(xmm0, xmm4);      /* xmm4 = A + C = C. */ \
    \
    psubsw_r2r(xmm6, xmm3);      /* xmm3 = B - D */ \
    paddw_r2r(xmm7, xmm5);       /* xmm5 = c2 * i6 */ \
    \
    paddsw_r2r(xmm6, xmm6);      /* xmm6 = D + D */ \
    pmulhw_m2r(*C(6), xmm7);     /* xmm7 = c6 * i6 */ \
    \
    paddsw_r2r(xmm3, xmm6);      /* xmm6 = B + D = D. */ \
    movdqu_r2m(xmm4, *I(1));     /* Save C. at I(1) */ \
    \
    psubsw_r2r(xmm5, xmm1);      /* xmm1 = c6 * i2 - c2 * i6 = H */ \
    movdqu_m2r(*C(4), xmm4);     /* xmm4 = c4 */ \
    \
    movdqu_r2r(xmm3, xmm5);      /* xmm5 = B - D */ \
    pmulhw_r2r(xmm4, xmm3);      /* xmm3 = ( c4 -1 ) * ( B - D ) */ \
    \
    paddsw_r2r(xmm2, xmm7);      /* xmm7 = c2 * i2 + c6 * i6 = G */ \
    movdqu_r2m(xmm6, *I(2));     /* Save D. at I(2) */ \
    \
    movdqu_r2r(xmm0, xmm2);      /* xmm2 = A - C */ \
    movdqu_m2r(*I(0), xmm6);     /* xmm6 = i0 */ \
    \
    pmulhw_r2r(xmm4, xmm0);      /* xmm0 = ( c4 - 1 ) * ( A - C ) = A. */ \
    paddw_r2r(xmm3, xmm5);       /* xmm5 = c4 * ( B - D ) = B. */ \
    \
    movdqu_m2r(*I(4), xmm3);     /* xmm3 = i4 */ \
    psubsw_r2r(xmm1, xmm5);      /* xmm5 = B. - H = B.. */ \
    \
    paddw_r2r(xmm0, xmm2);       /* xmm2 = c4 * ( A - C) = A. */ \
    psubsw_r2r(xmm3, xmm6);      /* xmm6 = i0 - i4 */ \
    \
    movdqu_r2r(xmm6, xmm0);      /* xmm0 = i0 - i4 */ \
    pmulhw_r2r(xmm4, xmm6);      /* xmm6 = (c4 - 1) * (i0 - i4) = F */ \
    \
    paddsw_r2r(xmm3, xmm3);      /* xmm3 = i4 + i4 */ \
    paddsw_r2r(xmm1, xmm1);      /* xmm1 = H + H */ \
    \
    paddsw_r2r(xmm0, xmm3);      /* xmm3 = i0 + i4 */ \
    paddsw_r2r(xmm5, xmm1);      /* xmm1 = B. + H = H. */ \
    \
    pmulhw_r2r(xmm3, xmm4);      /* xmm4 = ( c4 - 1 ) * ( i0 + i4 )  */ \
    paddw_r2r(xmm0, xmm6);       /* xmm6 = c4 * ( i0 - i4 ) */ \
    \
    psubsw_r2r(xmm2, xmm6);      /* xmm6 = F - A. = F. */ \
    paddsw_r2r(xmm2, xmm2);      /* xmm2 = A. + A. */ \
    \
    movdqu_m2r(*I(1), xmm0);     /* Load        C. from I(1) */ \
    paddsw_r2r(xmm6, xmm2);      /* xmm2 = F + A. = A.. */ \
    \
    paddw_r2r(xmm3, xmm4);       /* xmm4 = c4 * ( i0 + i4 ) = 3 */ \
    psubsw_r2r(xmm1, xmm2);      /* xmm2 = A.. - H. = R2 */ \
    \
    paddsw_m2r(*Eight, xmm2);    /* Adjust R2 and R1 before shifting */ \
    paddsw_r2r(xmm1, xmm1);      /* xmm1 = H. + H. */ \
    \
    paddsw_r2r(xmm2, xmm1);      /* xmm1 = A.. + H. = R1 */ \
    psraw_i2r(4, xmm2);          /* xmm2 = op2 */ \
    \
    psubsw_r2r(xmm7, xmm4);      /* xmm4 = E - G = E. */ \
    psraw_i2r(4, xmm1);          /* xmm1 = op1 */ \
    \
    movdqu_m2r(*I(2), xmm3);     /* Load D. from I(2) */ \
    paddsw_r2r(xmm7, xmm7);      /* xmm7 = G + G */ \
    \
    movdqu_r2m(xmm2, *O(2));     /* Write out op2 */ \
    paddsw_r2r(xmm4, xmm7);      /* xmm7 = E + G = G. */ \
    \
    movdqu_r2m(xmm1, *O(1));     /* Write out op1 */ \
    psubsw_r2r(xmm3, xmm4);      /* xmm4 = E. - D. = R4 */ \
    \
    paddsw_m2r(*Eight, xmm4);    /* Adjust R4 and R3 before shifting */ \
    paddsw_r2r(xmm3, xmm3);      /* xmm3 = D. + D. */ \
    \
    paddsw_r2r(xmm4, xmm3);      /* xmm3 = E. + D. = R3 */ \
    psraw_i2r(4, xmm4);          /* xmm4 = op4 */ \
    \
    psubsw_r2r(xmm5, xmm6);      /* xmm6 = F. - B..= R6 */ \
    psraw_i2r(4, xmm3);          /* xmm3 = op3 */ \
    \
    paddsw_m2r(*Eight, xmm6);    /* Adjust R6 and R5 before shifting */ \
    paddsw_r2r(xmm5, xmm5);      /* xmm5 = B.. + B.. */ \
    \
    paddsw_r2r(xmm6, xmm5);      /* xmm5 = F. + B.. = R5 */ \
    psraw_i2r(4, xmm6);          /* xmm6 = op6 */ \
    \
    movdqu_r2m(xmm4, *O(4));     /* Write out op4 */ \
    psraw_i2r(4, xmm5);          /* xmm5 = op5 */ \
    \
    movdqu_r2m(xmm3, *O(3));     /* Write out op3 */ \
    psubsw_r2r(xmm0, xmm7);      /* xmm7 = G. - C. = R7 */ \
    \
    paddsw_m2r(*Eight, xmm7);    /* Adjust R7 and R0 before shifting */ \
    paddsw_r2r(xmm0, xmm0);      /* xmm0 = C. + C. */ \
    \
    paddsw_r2r(xmm7, xmm0);      /* xmm0 = G. + C. */ \
    psraw_i2r(4, xmm7);          /* xmm7 = op7 */ \
    \
    movdqu_r2m(xmm6, *O(6));     /* Write out op6 */ \
    psraw_i2r(4, xmm0);          /* xmm0 = op0 */ \
    \
    movdqu_r2m(xmm5, *O(5));     /* Write out op5 */ \
    movdqu_r2m(xmm7, *O(7));     /* Write out op7 */ \
    \
    movdqu_r2m(xmm0, *O(0));     /* Write out op0 */ \
    \
} /* End of SSE2_Column_IDCT macro */


#define SSE2_Row_IDCT() {        \
    \
    movdqu_m2r(*I(3), xmm2);     /* xmm2 = i3 */ \
    movdqu_m2r(*C(3), xmm6);     /* xmm6 = c3 */ \
    \
    movdqu_r2r(xmm2, xmm4);      /* xmm4 = i3 */ \
    movdqu_m2r(*I(5), xmm7);     /* xmm7 = i5 */ \
    \
    pmulhw_r2r(xmm6, xmm4);      /* xmm4 = c3 * i3 - i3 */ \
    movdqu_m2r(*C(5), xmm1);     /* xmm1 = c5 */ \
    \
    pmulhw_r2r(xmm7, xmm6);      /* xmm6 = c3 * i5 - i5 */ \
    movdqu_r2r(xmm1, xmm5);      /* xmm5 = c5 */ \
    \
    pmulhw_r2r(xmm2, xmm1);      /* xmm1 = c5 * i3 - i3 */ \
    movdqu_m2r(*I(1), xmm3);     /* xmm3 = i1 */ \
    \
    pmulhw_r2r(xmm7, xmm5);      /* xmm5 = c5 * i5 - i5 */ \
    movdqu_m2r(*C(1), xmm0);     /* xmm0 = c1 */ \
    \
    /* all registers are in use */ \
    \
    paddw_r2r(xmm2, xmm4);      /* xmm4 = c3 * i3 */ \
    paddw_r2r(xmm7, xmm6);      /* xmm6 = c3 * i5 */ \
    \
    paddw_r2r(xmm1, xmm2);      /* xmm2 = c5 * i3 */ \
    movdqu_m2r(*I(7), xmm1);    /* xmm1 = i7 */ \
    \
    paddw_r2r(xmm5, xmm7);      /* xmm7 = c5 * i5 */ \
    movdqu_r2r(xmm0, xmm5);     /* xmm5 = c1 */ \
    \
    pmulhw_r2r(xmm3, xmm0);     /* xmm0 = c1 * i1 - i1 */ \
    paddsw_r2r(xmm7, xmm4);     /* xmm4 = c3 * i3 + c5 * i5 = C */ \
    \
    pmulhw_r2r(xmm1, xmm5);     /* xmm5 = c1 * i7 - i7 */ \
    movdqu_m2r(*C(7), xmm7);    /* xmm7 = c7 */ \
    \
    psubsw_r2r(xmm2, xmm6);     /* xmm6 = c3 * i5 - c5 * i3 = D */ \
    paddw_r2r(xmm3, xmm0);      /* xmm0 = c1 * i1 */ \
    \
    pmulhw_r2r(xmm7, xmm3);     /* xmm3 = c7 * i1 */ \
    movdqu_m2r(*I(2), xmm2);    /* xmm2 = i2 */ \
    \
    pmulhw_r2r(xmm1, xmm7);     /* xmm7 = c7 * i7 */ \
    paddw_r2r(xmm1, xmm5);      /* xmm5 = c1 * i7 */ \
    \
    movdqu_r2r(xmm2, xmm1);     /* xmm1 = i2 */ \
    pmulhw_m2r(*C(2), xmm2);    /* xmm2 = i2 * c2 -i2 */ \