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.

    \
    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_r2r(xmm1, xmm1);     /* xmm1 = H. + H. */ \
    paddsw_r2r(xmm2, xmm1);     /* xmm1 = A.. + H. = R1 */ \
    \
    psubsw_r2r(xmm7, xmm4);     /* xmm4 = E - G = E. */ \
    \
    movdqu_m2r(*I(2), xmm3);    /* Load D. from I(2) */ \
    paddsw_r2r(xmm7, xmm7);     /* xmm7 = G + G */ \
    \
    movdqu_r2m(xmm2, *I(2));    /* Write out op2 */ \
    paddsw_r2r(xmm4, xmm7);     /* xmm7 = E + G = G. */ \
    \
    movdqu_r2m(xmm1, *I(1));    /* Write out op1 */ \
    psubsw_r2r(xmm3, xmm4);     /* xmm4 = E. - D. = R4 */ \
    \
    paddsw_r2r(xmm3, xmm3);     /* xmm3 = D. + D. */ \
    \
    paddsw_r2r(xmm4, xmm3);     /* xmm3 = E. + D. = R3 */ \
    \
    psubsw_r2r(xmm5, xmm6);     /* xmm6 = F. - B..= R6 */ \
    \
    paddsw_r2r(xmm5, xmm5);     /* xmm5 = B.. + B.. */ \
    \
    paddsw_r2r(xmm6, xmm5);     /* xmm5 = F. + B.. = R5 */ \
    \
    movdqu_r2m(xmm4, *I(4));    /* Write out op4 */ \
    \
    movdqu_r2m(xmm3, *I(3));    /* Write out op3 */ \
    psubsw_r2r(xmm0, xmm7);     /* xmm7 = G. - C. = R7 */ \
    \
    paddsw_r2r(xmm0, xmm0);     /* xmm0 = C. + C. */ \
    \
    paddsw_r2r(xmm7, xmm0);     /* xmm0 = G. + C. */ \
    \
    movdqu_r2m(xmm6, *I(6));    /* Write out op6 */ \
    \
    movdqu_r2m(xmm5, *I(5));    /* Write out op5 */ \
    movdqu_r2m(xmm7, *I(7));    /* Write out op7 */ \
    \
    movdqu_r2m(xmm0, *I(0));    /* Write out op0 */ \
    \
} /* End of SSE2_Row_IDCT macro */


#define SSE2_Transpose() {    \
    \
    movdqu_m2r(*I(4), xmm4);    /* xmm4=e7e6e5e4e3e2e1e0 */ \
    movdqu_m2r(*I(5), xmm0);    /* xmm4=f7f6f5f4f3f2f1f0 */ \
    \
    movdqu_r2r(xmm4, xmm5);     /* make a copy */ \
    punpcklwd_r2r(xmm0, xmm4);  /* xmm4=f3e3f2e2f1e1f0e0 */ \
    \
    punpckhwd_r2r(xmm0, xmm5);  /* xmm5=f7e7f6e6f5e5f4e4 */ \
    movdqu_m2r(*I(6), xmm6);    /* xmm6=g7g6g5g4g3g2g1g0 */ \
    \
    movdqu_m2r(*I(7), xmm0);    /* xmm0=h7h6h5h4h3h2h1h0 */ \
    movdqu_r2r(xmm6, xmm7);     /* make a copy */ \
    \
    punpcklwd_r2r(xmm0, xmm6);  /* xmm6=h3g3h3g2h1g1h0g0 */ \
    punpckhwd_r2r(xmm0, xmm7);  /* xmm7=h7g7h6g6h5g5h4g4 */ \
    \
    movdqu_r2r(xmm4, xmm3);     /* make a copy */ \
    punpckldq_r2r(xmm6, xmm4);  /* xmm4=h1g1f1e1h0g0f0e0 */ \
    \
    punpckhdq_r2r(xmm6, xmm3);  /* xmm3=h3g3g3e3h2g2f2e2 */ \
    movdqu_r2m(xmm3, *I(6));    /* save h3g3g3e3h2g2f2e2 */ \
    /* Free xmm6 */ \
    movdqu_r2r(xmm5, xmm6);     /* make a copy */ \
    punpckldq_r2r(xmm7, xmm5);  /* xmm5=h5g5f5e5h4g4f4e4 */ \
    \
    punpckhdq_r2r(xmm7, xmm6);  /* xmm6=h7g7f7e7h6g6f6e6 */ \
    movdqu_m2r(*I(0), xmm0);    /* xmm0=a7a6a5a4a3a2a1a0 */ \
    /* Free xmm7 */ \
    movdqu_m2r(*I(1), xmm1);    /* xmm1=b7b6b5b4b3b2b1b0 */ \
    movdqu_r2r(xmm0, xmm7);     /* make a copy */ \
    \
    punpcklwd_r2r(xmm1, xmm0);  /* xmm0=b3a3b2a2b1a1b0a0 */ \
    punpckhwd_r2r(xmm1, xmm7);  /* xmm7=b7a7b6a6b5a5b4a4 */ \
    /* Free xmm1 */ \
    movdqu_m2r(*I(2), xmm2);    /* xmm2=c7c6c5c4c3c2c1c0 */ \
    movdqu_m2r(*I(3), xmm3);    /* xmm3=d7d6d5d4d3d2d1d0 */ \
    \
    movdqu_r2r(xmm2, xmm1);     /* make a copy */ \
    punpcklwd_r2r(xmm3, xmm2);  /* xmm2=d3c3d2c2d1c1d0c0 */ \
    \
    punpckhwd_r2r(xmm3, xmm1);  /* xmm1=d7c7d6c6d5c5d4c4 */ \
    movdqu_r2r(xmm0, xmm3);     /* make a copy        */ \
    \
    punpckldq_r2r(xmm2, xmm0);  /* xmm0=d1c1b1a1d0c0b0a0 */ \
    punpckhdq_r2r(xmm2, xmm3);  /* xmm3=d3c3b3a3d2c2b2a2 */ \
    /* Free xmm2 */ \
    movdqu_r2r(xmm7, xmm2);     /* make a copy */ \
    punpckldq_r2r(xmm1, xmm2);  /* xmm2=d5c5b5a5d4c4b4a4 */ \
    \
    punpckhdq_r2r(xmm1, xmm7);  /* xmm7=d7c7b7a7d6c6b6a6 */ \
    movdqu_r2r(xmm0, xmm1);     /* make a copy */ \
    \
    punpcklqdq_r2r(xmm4, xmm0); /* xmm0=h0g0f0e0d0c0b0a0 */ \
    punpckhqdq_r2r(xmm4, xmm1); /* xmm1=h1g1g1e1d1c1b1a1 */ \
    \
    movdqu_r2m(xmm0, *I(0));    /* save I(0) */ \
    movdqu_r2m(xmm1, *I(1));    /* save I(1) */ \
    \
    movdqu_m2r(*I(6), xmm0);    /* load h3g3g3e3h2g2f2e2 */ \
    movdqu_r2r(xmm3, xmm1);     /* make a copy */ \
    \
    punpcklqdq_r2r(xmm0, xmm1); /* xmm1=h2g2f2e2d2c2b2a2 */ \
    punpckhqdq_r2r(xmm0, xmm3); /* xmm3=h3g3f3e3d3c3b3a3 */ \
    \
    movdqu_r2r(xmm2, xmm4);     /* make a copy */ \
    punpcklqdq_r2r(xmm5, xmm4); /* xmm4=h4g4f4e4d4c4b4a4 */ \
    \
    punpckhqdq_r2r(xmm5, xmm2); /* xmm2=h5g5f5e5d5c5b5a5 */ \
    movdqu_r2m(xmm1, *I(2));    /* save I(2) */ \
    \
    movdqu_r2m(xmm3, *I(3));    /* save I(3) */ \
    movdqu_r2m(xmm4, *I(4));    /* save I(4) */ \
    \
    movdqu_r2m(xmm2, *I(5));    /* save I(5) */ \
    movdqu_r2r(xmm7, xmm5);     /* make a copy */ \
    \
    punpcklqdq_r2r(xmm6, xmm5); /* xmm5=h6g6f6e6d6c6b6a6 */ \
    punpckhqdq_r2r(xmm6, xmm7); /* xmm7=h7g7f7e7d7c7b7a7 */ \
    \
    movdqu_r2m(xmm5, *I(6));    /* save I(6) */ \
    movdqu_r2m(xmm7, *I(7));    /* save I(7) */ \
    \
} /* End of Transpose Macro */


#define SSE2_Dequantize() {        \
    movdqu_m2r(*(eax), xmm0);     \
    \
    pmullw_m2r(*(ebx), xmm0);          /* xmm0 = 07 06 05 04 03 02 01 00 */ \
    movdqu_m2r(*(eax + 16), xmm1);     \
    \
    pmullw_m2r(*(ebx + 16), xmm1);     /* xmm1 = 17 16 15 14 13 12 11 10 */ \
    pshuflw_r2r(xmm0, xmm3, 0x078);    /* xmm3 = 07 06 05 04 01 03 02 00 */ \
    \
    movdqu_r2r(xmm1, xmm2);            /* xmm2 = 17 16 15 14 13 12 11 10 */ \
    movdqu_m2r(*(ecx), xmm7);          /* xmm7 = -- -- -- -- -- FF FF -- */ \
    \
    movdqu_m2r(*(eax + 32), xmm4);     \
    movdqu_m2r(*(eax + 64), xmm5);     \
    \
    pmullw_m2r(*(ebx + 32), xmm4);     /* xmm4 = 27 26 25 24 23 22 21 20 */ \
    pmullw_m2r(*(ebx + 64), xmm5);     /* xmm5 = 47 46 45 44 43 42 41 40 */ \
    \
    movdqu_m2r(*(ecx + 16), xmm6);     /* xmm6 = -- -- FF FF -- -- -- -- */ \
    pand_r2r(xmm2, xmm7);              /* xmm7 = -- -- -- -- -- 12 11 -- */ \
    \
    pand_r2r(xmm4, xmm6);              /* xmm6 = -- -- 25 24 -- -- -- -- */ \
    pxor_r2r(xmm7, xmm2);              /* xmm2 = 17 16 15 14 13 -- -- 10 */ \
    \
    pxor_r2r(xmm6, xmm4);              /* xmm4 = 27 26 -- -- 23 22 21 20 */ \
    pslldq_i2r(4, xmm7);               /* xmm7 = -- -- -- 12 11 -- -- -- */ \
    \
    pslldq_i2r(2, xmm6);               /* xmm6 = -- 25 24 -- -- -- -- -- */ \
    por_r2r(xmm6, xmm7);               /* xmm7 = -- 25 24 12 11 -- -- -- */ \
    \
    movdqu_m2r(*(ecx + 32), xmm0);     /* xmm0 = -- -- -- -- -- FF FF FF */ \
    movdqu_m2r(*(ecx + 48), xmm6);     /* xmm6 = -- -- -- -- FF -- -- -- */ \
    \
    pand_r2r(xmm3, xmm0);              /* xmm0 = -- -- -- -- -- 03 02 00 */ \
    pand_r2r(xmm5, xmm6);              /* xmm6 = -- -- -- -- 43 -- -- -- */ \
    \
    pxor_r2r(xmm0, xmm3);              /* xmm3 = 07 06 05 04 01 -- -- -- */ \
    pxor_r2r(xmm6, xmm5);              /* xmm5 = 47 46 45 44 -- 42 41 40 */ \
    \
    por_r2r(xmm7, xmm0);               /* xmm0 = -- 25 24 12 11 03 02 00 */ \
    pslldq_i2r(8, xmm6);               /* xmm6 = 43 -- -- -- -- -- -- -- */ \
    \
    por_r2r(xmm6, xmm0);               /* xmm0 = 43 25 24 12 11 03 02 00 */ \
    /* 02345 in use */ \
    \
    movdqu_m2r(*(ecx + 64 ), xmm1);    /* xmm1 = -- -- -- FF FF -- -- -- */ \
    pshuflw_r2r(xmm5, xmm5, 0x0B4);    /* xmm5 = 47 46 45 44 42 -- 41 40 */ \
    \
    movdqu_r2r(xmm1, xmm7);            /* xmm7 = -- -- -- FF FF -- -- -- */ \
    movdqu_r2r(xmm1, xmm6);            /* xmm6 = -- -- -- FF FF -- -- -- */ \
    \
    movdqu_r2m(xmm0, *(eax));          /* write  43 25 24 12 11 03 02 00 */ \
    pshufhw_r2r(xmm4, xmm4, 0x0C2);    /* xmm4 = 27 -- -- 26 23 22 21 20 */ \
    \
    pand_r2r(xmm4, xmm7);              /* xmm7 = -- -- -- 26 23 -- -- -- */ \
    pand_r2r(xmm5, xmm1);              /* xmm1 = -- -- -- 44 42 -- -- -- */ \
    \
    pxor_r2r(xmm7, xmm4);              /* xmm4 = 27 -- -- -- -- 22 21 20 */ \
    pxor_r2r(xmm1, xmm5);              /* xmm5 = 47 46 45 -- -- -- 41 40 */ \
    \
    pshuflw_r2r(xmm2, xmm2, 0x0C6);    /* xmm2 = 17 16 15 14 13 10 -- -- */ \
    movdqu_r2r(xmm6, xmm0);            /* xmm0 = -- -- -- FF FF -- -- -- */ \
    \
    pslldq_i2r(2, xmm7);               /* xmm7 = -- -- 26 23 -- -- -- -- */ \
    pslldq_i2r(6, xmm1);               /* xmm1 = 44 42 -- -- -- -- -- -- */ \
    \
    psrldq_i2r(2, xmm0);               /* xmm0 = -- -- -- -- FF FF -- -- */ \
    pand_r2r(xmm3, xmm6);              /* xmm6 = -- -- -- 04 01 -- -- -- */ \
    \
    pand_r2r(xmm2, xmm0);              /* xmm0 = -- -- -- -- 13 10 -- -- */ \
    pxor_r2r(xmm6, xmm3);              /* xmm3 = 07 06 05 -- -- -- -- -- */ \
    \
    pxor_r2r(xmm0, xmm2);              /* xmm2 = 17 16 15 14 -- -- -- -- */ \
    psrldq_i2r(6, xmm6);               /* xmm0 = -- -- -- -- -- -- 04 01 */ \
    \
    por_r2r(xmm7, xmm1);               /* xmm1 = 44 42 26 23 -- -- -- -- */ \
    por_r2r(xmm6, xmm0);               /* xmm1 = -- -- -- -- 13 10 04 01 */ \
    /* 12345 in use */    \
    por_r2r(xmm0, xmm1);               /* xmm1 = 44 42 26 23 13 10 04 01 */ \
    pshuflw_r2r(xmm4, xmm4, 0x093);    /* xmm4 = 27 -- -- -- 22 21 20 -- */ \