generic_mmx.h

DirectFB-1.0.1可用于linux的嵌入式GUI

     static const u32 einser[] = { 0x01000100, 0x01000100 };
     static const u32 zeros[]  = { 0, 0 };

     __asm__ __volatile__ (
	       "movq     %3, %%mm7\n\t"
               "cmp      $0, %2\n\t"
               "jne      3f\n\t"
               "movq     %4, %%mm6\n\t"
               "movd     %5, %%mm0\n\t"
               "punpcklbw %%mm6, %%mm0\n\t" /* mm0 = 00aa 00rr 00gg 00bb */
               "punpcklwd %%mm0, %%mm0\n\t" /* mm0 = 00aa 00aa xxxx xxxx */
               "movq      %%mm7, %%mm1\n\t"
               "punpckldq %%mm0, %%mm0\n\t" /* mm0 = 00aa 00aa 00aa 00aa */
               "psubw     %%mm0, %%mm1\n\t"

               ".align   16\n"
               "4:\n\t"                 /* blend from color */
               "testw    $0xF000, 6(%0)\n\t"
               "jnz      1f\n\t"
               "movq     (%0), %%mm0\n\t"
               "pmullw   %%mm1, %%mm0\n\t"
               "psrlw    $8, %%mm0\n\t"
               "movq     %%mm0, (%0)\n"
               "1:\n\t"
	       "add      $8, %0\n\t"
               "dec      %1\n\t"
               "jnz      4b\n\t"
               "jmp      2f\n\t"

               ".align   16\n"
               "3:\n\t"                      /* blend from Sacc */
               "testw    $0xF000, 6(%0)\n\t"
               "jnz      1f\n\t"
               "movq     (%2), %%mm2\n\t"
               "movq     (%0), %%mm0\n\t"
	       "punpckhwd %%mm2, %%mm2\n\t" /* mm2 = 00aa 00aa xxxx xxxx */
               "movq	  %%mm7, %%mm1\n\t"
               "punpckhdq %%mm2, %%mm2\n\t" /* mm2 = 00aa 00aa 00aa 00aa */
               "psubw    %%mm2, %%mm1\n\t"
               "pmullw   %%mm1, %%mm0\n\t"
               "psrlw    $8, %%mm0\n\t"
               "movq     %%mm0, (%0)\n"
               "1:\n\t"
	       "add      $8, %2\n\t"
	       "add      $8, %0\n\t"
               "dec      %1\n\t"
               "jnz      3b\n\t"
               "2:\n\t"
               "emms"
               : /* no outputs */
               : "D" (gfxs->Xacc), "c" (gfxs->length), "S" (gfxs->Sacc),
                 "m" (*einser), "m" (*zeros), "m" (gfxs->color)
               : "%st", "memory");
}

static void Xacc_blend_srcalpha_MMX( GenefxState *gfxs )
{
     static const u32 ones[]  = { 0x00010001, 0x00010001 };
     static const u32 zeros[] = { 0, 0 };

     __asm__ __volatile__ (
	       "movq     %3, %%mm7\n\t"
               "cmp      $0, %2\n\t"
               "jne      3f\n\t"
               "movq     %4, %%mm6\n\t"
               "movd     %5, %%mm0\n\t"
               "punpcklbw %%mm6, %%mm0\n\t" /* mm0 = 00aa 00rr 00gg 00bb */
               "punpcklwd %%mm0, %%mm0\n\t" /* mm0 = 00aa 00aa xxxx xxxx */
               "punpckldq %%mm0, %%mm0\n\t" /* mm0 = 00aa 00aa 00aa 00aa */
               "paddw     %%mm7, %%mm0\n\t"

               ".align   16\n"
               "4:\n\t"                 /* blend from color */
               "testw    $0xF000, 6(%0)\n\t"
               "jnz      1f\n\t"
               "movq     (%0), %%mm1\n\t"
               "pmullw   %%mm0, %%mm1\n\t"
               "psrlw    $8, %%mm1\n\t"
               "movq     %%mm1, (%0)\n"
               "1:\n\t"
	       "add      $8, %0\n\t"
               "dec      %1\n\t"
               "jnz      4b\n\t"
               "jmp      2f\n\t"

               ".align   16\n"
               "3:\n\t"                      /* blend from Sacc */
               "testw    $0xF000, 6(%0)\n\t"
               "jnz      1f\n\t"
               "movq     (%2), %%mm0\n\t"
               "movq     (%0), %%mm1\n\t"
	       "punpckhwd %%mm0, %%mm0\n\t" /* mm2 = 00aa 00aa xxxx xxxx */
               "punpckhdq %%mm0, %%mm0\n\t" /* mm2 = 00aa 00aa 00aa 00aa */
               "paddw    %%mm7, %%mm0\n\t"
               "pmullw   %%mm0, %%mm1\n\t"
               "psrlw    $8, %%mm1\n\t"
               "movq     %%mm1, (%0)\n"
               "1:\n\t"
	       "add      $8, %2\n\t"
	       "add      $8, %0\n\t"
               "dec      %1\n\t"
               "jnz      3b\n\t"
               "2:\n\t"
               "emms"
               : /* no outputs */
               : "D" (gfxs->Xacc), "c" (gfxs->length), "S" (gfxs->Sacc),
                 "m" (*ones), "m" (*zeros), "m" (gfxs->color)
               : "%st", "memory");
}

static void Dacc_YCbCr_to_RGB_MMX( GenefxState *gfxs )
{  
     static const u16 __aligned(8) sub0[4] = {  16,  16,  16,  16 }; 
     static const u16 __aligned(8) sub1[4] = { 128, 128, 128, 128 };
     static const s16 __aligned(8) mul[20] = {
                     0x253F,  0x253F,  0x253F,  0x253F, // Y       Coeff.
                     0x3312,  0x3312,  0x3312,  0x3312, // V Red   Coeff.
                     0x4093,  0x4093,  0x4093,  0x4093, // U Blue  Coeff.
                    -0x1A04, -0x1A04, -0x1A04, -0x1A04, // V Green Coeff.
                    -0x0C83, -0x0C83, -0x0C83, -0x0C83  // U Green Coeff.
     };

     int                w = gfxs->length & 3;
     GenefxAccumulator *D = gfxs->Dacc;
     
     __asm__ __volatile__ (
          "shrl          $2,    %1\n\t"
          "jz            2f\n\t"
          "pxor       %%mm7, %%mm7\n\t"
               ".align 16\n"
               "1:\n\t"
               "movq        (%0), %%mm0\n\t" // 00 a0 00 y0 00 v0 00 u0
               "movq       8(%0), %%mm1\n\t" // 00 a1 00 y1 00 v1 00 u1
               "movq      16(%0), %%mm2\n\t" // 00 a2 00 y2 00 v2 00 u2
               "movq      24(%0), %%mm3\n\t" // 00 a3 00 y3 00 v3 00 u3
               "movq       %%mm0, %%mm4\n\t" // 00 a0 00 y0 00 v0 00 u0
               "movq       %%mm2, %%mm5\n\t" // 00 a2 00 y2 00 v2 00 u2
               "punpcklwd  %%mm1, %%mm0\n\t" // 00 v1 00 v0 00 u1 00 u0
               "punpcklwd  %%mm3, %%mm2\n\t" // 00 v3 00 v2 00 u3 00 u2
               "punpckhwd  %%mm1, %%mm4\n\t" // 00 a1 00 a0 00 y1 00 y0
               "punpckhwd  %%mm3, %%mm5\n\t" // 00 a3 00 a2 00 y3 00 y2
               "movq       %%mm0, %%mm1\n\t" // 00 v1 00 v1 00 u1 00 u0
               "movq       %%mm4, %%mm3\n\t" // 00 a1 00 a0 00 y1 00 y0
               "punpckldq  %%mm2, %%mm0\n\t" // 00 u3 00 u2 00 u1 00 u0
               "punpckldq  %%mm5, %%mm3\n\t" // 00 y3 00 y2 00 y1 00 y0
               "punpckhdq  %%mm2, %%mm1\n\t" // 00 v3 00 v2 00 v1 00 v0
               "punpckhdq  %%mm5, %%mm4\n\t" // 00 a3 00 a2 00 a1 00 a0
               /* mm0 = u, mm1 = v, mm3 = y, mm4 = a */
               "psubw         %2, %%mm3\n\t" // y -= 16
               "psllw         $3, %%mm3\n\t" // precision
               "pmulhw      (%4), %%mm3\n\t"
               "psubw         %3, %%mm1\n\t" // v -= 128 
               "psllw         $3, %%mm1\n\t" // precision
               "movq       %%mm1, %%mm2\n\t" // 00 v3 00 v2 00 v1 00 v0
               "pmulhw     8(%4), %%mm2\n\t" // vr
               "psubw         %3, %%mm0\n\t" // u -= 128 
               "psllw         $3, %%mm0\n\t" // precision
               "movq       %%mm0, %%mm5\n\t" // 00 u3 00 u2 00 u1 00 u0
               "pmulhw    16(%4), %%mm5\n\t" // ub
               "paddw      %%mm3, %%mm2\n\t" // 00 r3 00 r2 00 r1 00 r0
               "paddw      %%mm3, %%mm5\n\t" // 00 b3 00 b2 00 b1 00 b0
               "pmulhw    24(%4), %%mm1\n\t" // vg
               "packuswb   %%mm2, %%mm2\n\t" // r3 r2 r1 r0 r3 r2 r1 r0
               "packuswb   %%mm5, %%mm5\n\t" // b3 b2 b1 b0 b3 b2 b1 b0
               "pmulhw    32(%4), %%mm0\n\t" // ug
               "punpcklbw  %%mm7, %%mm2\n\t" // 00 r3 00 r2 00 r1 00 r0
               "punpcklbw  %%mm7, %%mm5\n\t" // 00 b3 00 b2 00 b1 00 b0
               "paddw      %%mm1, %%mm3\n\t" // y + vg
               "paddw      %%mm0, %%mm3\n\t" // 00 g3 00 g2 00 g1 00 g0
               "packuswb   %%mm3, %%mm3\n\t" // g3 g2 g1 g0 g3 g2 g1 g0
               "punpcklbw  %%mm7, %%mm3\n\t" // 00 g3 00 g2 00 g1 00 g0
               /* mm5 = b, mm3 = g, mm2 = r, mm4 = a */
               "movq       %%mm5, %%mm0\n\t" // 00 b3 00 b2 00 b1 00 b0
               "movq       %%mm3, %%mm1\n\t" // 00 g3 00 g2 00 g1 00 g0
               "punpcklwd  %%mm2, %%mm0\n\t" // 00 r1 00 b1 00 r0 00 b0
               "punpcklwd  %%mm4, %%mm1\n\t" // 00 a1 00 g1 00 a0 00 g0
               "punpckhwd  %%mm2, %%mm5\n\t" // 00 r3 00 b3 00 r2 00 b2
               "punpckhwd  %%mm4, %%mm3\n\t" // 00 a3 00 g3 00 a2 00 g2
               "movq       %%mm0, %%mm2\n\t" // 00 r1 00 b1 00 r0 00 b0
               "movq       %%mm5, %%mm4\n\t" // 00 r3 00 b3 00 r2 00 b2
               "punpcklwd  %%mm1, %%mm0\n\t" // 00 a0 00 r0 00 g0 00 b0
               "punpcklwd  %%mm3, %%mm5\n\t" // 00 a2 00 r2 00 g2 00 b2
               "punpckhwd  %%mm1, %%mm2\n\t" // 00 a1 00 r1 00 g1 00 b1
               "punpckhwd  %%mm3, %%mm4\n\t" // 00 a3 00 r3 00 g3 00 b3
               "movq       %%mm0,  (%0)\n\t"
               "movq       %%mm2, 8(%0)\n\t"
               "movq       %%mm5,16(%0)\n\t"
               "movq       %%mm4,24(%0)\n\t"
               "add          $32,    %0\n\t"
               "decl          %1\n\t"
               "jnz           1b\n\t"
          "emms\n\t"
          "2:"     
          : "=&D" (D)
          : "c" (gfxs->length), "m" (*sub0), "m" (*sub1), "r" (mul), "0" (D)
          : "memory" );

     while (w) {
          if (!(D->YUV.a & 0xF000)) {
               u16 y, cb, cr;
               s16 r, g, b;

               y  = y_for_rgb[D->YUV.y];
               cb = D->YUV.u;
               cr = D->YUV.v;
               r  = y + cr_for_r[cr];
               g  = y + cr_for_g[cr] + cb_for_g[cb];
               b  = y                + cb_for_b[cb];
               
               D->RGB.r = (r < 0) ? 0 : r;
               D->RGB.g = (g < 0) ? 0 : g;
               D->RGB.b = (b < 0) ? 0 : b;
          }

          D++;
          w--;
     }
}

static void Dacc_RGB_to_YCbCr_MMX( GenefxState *gfxs )
{
     static const u16 __aligned(8) add0[4] = { 128, 128, 128, 128 };
     static const u16 __aligned(8) add1[4] = {  16,  16,  16,  16 };
     static const u16 __aligned(8) mul[24] = {
                    0x03A5, 0x03A5, 0x03A5, 0x03A5, // Eb
                    0x12C8, 0x12C8, 0x12C8, 0x12C8, // Eg
                    0x0991, 0x0991, 0x0991, 0x0991, // Er
                    0x0FE1, 0x0FE1, 0x0FE1, 0x0FE1, // Cb
                    0x140A, 0x140A, 0x140A, 0x140A, // Cr
                    0x1B7B, 0x1B7B, 0x1B7B, 0x1B7B  // Y
     };
      
     int                w = gfxs->length & 3;
     GenefxAccumulator *D = gfxs->Dacc;
     
     __asm__ __volatile__(
          "shrl          $2,    %1\n\t"
          "jz            2f\n\t" 
          "pxor       %%mm7, %%mm7\n\t"     
               ".align 16\n"
               "1:\n\t"
               "movq        (%0), %%mm0\n\t" // 00 a0 00 r0 00 g0 00 b0
               "movq       8(%0), %%mm1\n\t" // 00 a1 00 r1 00 g1 00 b1
               "movq      16(%0), %%mm2\n\t" // 00 a2 00 r2 00 g2 00 b2
               "movq      24(%0), %%mm3\n\t" // 00 a3 00 r3 00 g3 00 b3
               "movq       %%mm0, %%mm4\n\t" // 00 a0 00 r0 00 g0 00 b0
               "movq       %%mm2, %%mm6\n\t" // 00 a2 00 r2 00 g2 00 b2
               "punpcklwd  %%mm1, %%mm0\n\t" // 00 g1 00 g0 00 b1 00 b0
               "punpcklwd  %%mm3, %%mm2\n\t" // 00 g3 00 g2 00 b3 00 b2
               "movq       %%mm0, %%mm5\n\t" // 00 g1 00 g0 00 b1 00 b0
               "punpckldq  %%mm2, %%mm0\n\t" // 00 b3 00 b2 00 b1 00 b0
               "punpckhdq  %%mm2, %%mm5\n\t" // 00 g3 00 g2 00 g1 00 g0
               "punpckhwd  %%mm1, %%mm4\n\t" // 00 a1 00 a0 00 r1 00 r0
               "punpckhwd  %%mm3, %%mm6\n\t" // 00 a3 00 a2 00 r3 00 r2
               "movq       %%mm4, %%mm3\n\t" // 00 a1 00 a0 00 r1 00 r0
               "punpckldq  %%mm6, %%mm4\n\t" // 00 r3 00 r2 00 r1 00 r0
               "punpckhdq  %%mm6, %%mm3\n\t" // 00 a3 00 a2 00 a1 00 a0
               /* mm0 = b, mm5 = g, mm4 = r, mm3 = a */
               "movq       %%mm0, %%mm1\n\t" // save b
               "psllw         $3, %%mm0\n\t"
               "pmulhw      (%2), %%mm0\n\t"
               "movq       %%mm4, %%mm2\n\t" // save r
               "psllw         $3, %%mm5\n\t"
               "pmulhw     8(%2), %%mm5\n\t"
               "psllw         $3, %%mm4\n\t"
               "pmulhw    16(%2), %%mm4\n\t"
               "paddw      %%mm5, %%mm0\n\t"
               "paddw      %%mm4, %%mm0\n\t" // ey
               "psubw      %%mm0, %%mm1\n\t" // b - ey
               "psllw         $3, %%mm1\n\t"
               "pmulhw    24(%2), %%mm1\n\t" // 00 u3 00 u2 00 u1 00 u0
               "psubw      %%mm0, %%mm2\n\t" // r - ey
               "psllw         $3, %%mm2\n\t"
               "pmulhw    32(%2), %%mm2\n\t" // 00 v3 00 v2 00 v1 00 v0
               "paddw         %3, %%mm1\n\t" // Cb + 128
               "packuswb   %%mm1, %%mm1\n\t" // u3 u2 u1 u0 u3 u2 u1 u0
               "psllw         $3, %%mm0\n\t"
               "pmulhw    40(%2), %%mm0\n\t" // 00 y3 00 y2 00 y1 00 y0
               "paddw         %3, %%mm2\n\t" // Cr + 128
               "packuswb   %%mm2, %%mm2\n\t" // v3 v2 v1 v0 v3 v2 v1 v0  
               "paddw         %4, %%mm0\n\t" // Y + 16
               "packuswb   %%mm0, %%mm0\n\t" // y3 y2 y1 y0 y3 y2 y1 y0
               "punpcklbw  %%mm7, %%mm2\n\t" // 00 v3 00 v2 00 v1 00 v0
               "punpcklbw  %%mm7, %%mm1\n\t" // 00 u3 00 y2 00 u1 00 u0
               "punpcklbw  %%mm7, %%mm0\n\t" // 00 y3 00 y2 00 y1 00 y0 
               /* mm1 = u, mm2 = v, mm0 = y, mm3 = a */
               "movq       %%mm2, %%mm4\n\t" // 00 v3 00 v2 00 v1 00 v0
               "movq       %%mm1, %%mm5\n\t" // 00 u3 00 y2 00 u1 00 u0
               "punpcklwd  %%mm3, %%mm2\n\t" // 00 a1 00 v1 00 a0 00 v0
               "punpcklwd  %%mm0, %%mm1\n\t" // 00 y1 00 u1 00 y0 00 u0
               "punpckhwd  %%mm3, %%mm4\n\t" // 00 a3 00 v3 00 a2 00 v2
               "punpckhwd  %%mm0, %%mm5\n\t" // 00 y3 00 u3 00 y2 00 u2
               "movq       %%mm1, %%mm3\n\t" // 00 y1 00 u1 00 y0 00 u0
               "movq       %%mm5, %%mm6\n\t" // 00 y3 00 u3 00 y2 00 u2
               "punpcklwd  %%mm2, %%mm1\n\t" // 00 a0 00 y0 00 v0 00 u0
               "punpcklwd  %%mm4, %%mm5\n\t" // 00 a2 00 y2 00 v2 00 u2
               "punpckhwd  %%mm2, %%mm3\n\t" // 00 a1 00 y1 00 v1 00 u1
               "punpckhwd  %%mm4, %%mm6\n\t" // 00 a3 00 y3 00 v3 00 u3
               "movq       %%mm1,  (%0)\n\t"
               "movq       %%mm3, 8(%0)\n\t"
               "movq       %%mm5,16(%0)\n\t"
               "movq       %%mm6,24(%0)\n\t"
               "add          $32, %0\n\t"
               "decl          %1\n\t"
               "jnz           1b\n\t"
          "emms\n\t"
          "2:"
          : "=&D" (D)
          : "c" (gfxs->length), "r" (mul), "m" (*add0), "m" (*add1), "0" (D)
          : "memory" );

     while (w) {
          if (!(D->RGB.a & 0xF000)) {
               u32 r, g, b, ey;

               r = D->RGB.r; g = D->RGB.g; b = D->RGB.b;
               ey = (19595 * r + 38469 * g + 7471 * b) >> 16;

               D->YUV.y = y_from_ey[ey];
               D->YUV.u = cb_from_bey[b-ey];
               D->YUV.v = cr_from_rey[r-ey];
          }

          D++;
          w--;
     }
}