void
yuv411planar_to_rgb_mmx (const unsigned char *yuv, unsigned char *rgb,
unsigned int w, unsigned int h)
{
unsigned int xx, yy;
register const unsigned char *yp1, *up, *vp;
unsigned char *dp1;
/* plane pointers */
yp1 = yuv;
up = yuv + (w * h);
vp = up + (w * (h / 4));
/* destination pointers */
dp1 = rgb;
yp1 = yuv;
up = yuv + (w * h);
vp = up + ((w / 2) * (h / 2));
dp1 = rgb;
for (yy = 0; yy < h; yy++)
{
for (xx = 0; xx < w; xx += 8)
{
movq_m2r(*yp1, mm0);
movq_r2r(mm0, mm1);
psrlw_i2r(8, mm0);
psllw_i2r(8, mm1);
psrlw_i2r(8, mm1);
pxor_r2r(mm7, mm7);
movd_m2r(*up, mm3);
movd_m2r(*vp, mm2);
punpcklbw_r2r(mm7, mm2);
punpcklbw_r2r(mm7, mm3);
movq_m2r(CONST_16, mm4);
psubsw_r2r(mm4, mm0);
psubsw_r2r(mm4, mm1);
movq_m2r(CONST_128, mm5);
psubsw_r2r(mm5, mm2);
psubsw_r2r(mm5, mm3);
movq_m2r(CONST_YMUL, mm4);
pmullw_r2r(mm4, mm0);
pmullw_r2r(mm4, mm1);
movq_m2r(CONST_CRVCRV, mm7);
pmullw_r2r(mm3, mm7);
movq_m2r(CONST_CBUCBU, mm6);
pmullw_r2r(mm2, mm6);
movq_m2r(CONST_CGUCGU, mm5);
pmullw_r2r(mm2, mm5);
movq_m2r(CONST_CGVCGV, mm4);
pmullw_r2r(mm3, mm4);
movq_r2r(mm0, mm2);
paddsw_r2r(mm7, mm2);
paddsw_r2r(mm1, mm7);
psraw_i2r(RES, mm2);
psraw_i2r(RES, mm7);
packuswb_r2r(mm7, mm2);
pxor_r2r(mm7, mm7);
movq_r2r(mm2, mm3);
punpckhbw_r2r(mm7, mm2);
punpcklbw_r2r(mm3, mm7);
por_r2r(mm7, mm2);
movq_r2r(mm0, mm3);
psubsw_r2r(mm5, mm3);
psubsw_r2r(mm4, mm3);
paddsw_m2r(CONST_32, mm3);
movq_r2r(mm1, mm7);
psubsw_r2r(mm5, mm7);
psubsw_r2r(mm4, mm7);
paddsw_m2r(CONST_32, mm7);
psraw_i2r(RES, mm3);
psraw_i2r(RES, mm7);
packuswb_r2r(mm7, mm3);
pxor_r2r(mm7, mm7);
movq_r2r(mm3, mm4);
punpckhbw_r2r(mm7, mm3);
punpcklbw_r2r(mm4, mm7);
por_r2r(mm7, mm3);
movq_m2r(CONST_32, mm4);
paddsw_r2r(mm6, mm0);
paddsw_r2r(mm6, mm1);
paddsw_r2r(mm4, mm0);
paddsw_r2r(mm4, mm1);
psraw_i2r(RES, mm0);
psraw_i2r(RES, mm1);
packuswb_r2r(mm1, mm0);
pxor_r2r(mm7, mm7);
movq_r2r(mm0, mm5);
punpckhbw_r2r(mm7, mm0);
punpcklbw_r2r(mm5, mm7);
por_r2r(mm7, mm0);
pxor_r2r(mm1, mm1);
movq_r2r(mm0, mm5);
movq_r2r(mm3, mm6);
movq_r2r(mm2, mm7);
punpckhbw_r2r(mm3, mm2);
punpcklbw_r2r(mm6, mm7);
punpckhbw_r2r(mm1, mm0);
punpcklbw_r2r(mm1, mm5);
movq_r2r(mm7, mm1);
punpckhwd_r2r(mm5, mm7);
punpcklwd_r2r(mm5, mm1);
movq_r2r(mm2, mm4);
punpckhwd_r2r(mm0, mm2);
punpcklwd_r2r(mm0, mm4);
movntq_r2m(mm1, *(dp1));
movntq_r2m(mm7, *(dp1 + 8));
movntq_r2m(mm4, *(dp1 + 16));
movntq_r2m(mm2, *(dp1 + 24));
yp1 += 8;
up += 4;
vp += 4;
dp1 += 8 * 4;
}
if (yy & 0x1)
{
up -= w / 2;
vp -= w / 2;
}
}
emms();
}
static void
_evas_yv12torgb_sse(unsigned char **yuv, unsigned char *rgb, int w, int h)
{
#ifdef BUILD_MMX
int xx, yy;
register unsigned char *yp1, *up, *vp;
unsigned char *dp1;
/* destination pointers */
dp1 = rgb;
for (yy = 0; yy < h; yy++)
{
/* plane pointers */
yp1 = yuv[yy];
up = yuv[h + (yy / 2)];
vp = yuv[h + (h / 2) + (yy / 2)];
for (xx = 0; xx < (w - 7); xx += 8)
{
movd_m2r(*up, mm3);
movd_m2r(*vp, mm2);
movq_m2r(*yp1, mm0);
pxor_r2r(mm7, mm7);
punpcklbw_r2r(mm7, mm2);
punpcklbw_r2r(mm7, mm3);
movq_r2r(mm0, mm1);
psrlw_i2r(8, mm0);
psllw_i2r(8, mm1);
psrlw_i2r(8, mm1);
movq_m2r(CONST_16, mm4);
psubsw_r2r(mm4, mm0);
psubsw_r2r(mm4, mm1);
movq_m2r(CONST_128, mm5);
psubsw_r2r(mm5, mm2);
psubsw_r2r(mm5, mm3);
movq_m2r(CONST_YMUL, mm4);
pmullw_r2r(mm4, mm0);
pmullw_r2r(mm4, mm1);
movq_m2r(CONST_CRVCRV, mm7);
pmullw_r2r(mm3, mm7);
movq_m2r(CONST_CBUCBU, mm6);
pmullw_r2r(mm2, mm6);
movq_m2r(CONST_CGUCGU, mm5);
pmullw_r2r(mm2, mm5);
movq_m2r(CONST_CGVCGV, mm4);
pmullw_r2r(mm3, mm4);
movq_r2r(mm0, mm2);
paddsw_r2r(mm7, mm2);
paddsw_r2r(mm1, mm7);
psraw_i2r(RES, mm2);
psraw_i2r(RES, mm7);
packuswb_r2r(mm7, mm2);
pxor_r2r(mm7, mm7);
movq_r2r(mm2, mm3);
punpckhbw_r2r(mm7, mm2);
punpcklbw_r2r(mm3, mm7);
por_r2r(mm7, mm2);
movq_r2r(mm0, mm3);
psubsw_r2r(mm5, mm3);
psubsw_r2r(mm4, mm3);
paddsw_m2r(CONST_32, mm3);
movq_r2r(mm1, mm7);
psubsw_r2r(mm5, mm7);
psubsw_r2r(mm4, mm7);
paddsw_m2r(CONST_32, mm7);
psraw_i2r(RES, mm3);
psraw_i2r(RES, mm7);
packuswb_r2r(mm7, mm3);
pxor_r2r(mm7, mm7);
movq_r2r(mm3, mm4);
punpckhbw_r2r(mm7, mm3);
punpcklbw_r2r(mm4, mm7);
por_r2r(mm7, mm3);
movq_m2r(CONST_32, mm4);
paddsw_r2r(mm6, mm0);
paddsw_r2r(mm6, mm1);
paddsw_r2r(mm4, mm0);
paddsw_r2r(mm4, mm1);
psraw_i2r(RES, mm0);
psraw_i2r(RES, mm1);
packuswb_r2r(mm1, mm0);
pxor_r2r(mm7, mm7);
movq_r2r(mm0, mm5);
punpckhbw_r2r(mm7, mm0);
punpcklbw_r2r(mm5, mm7);
por_r2r(mm7, mm0);
movq_m2r(CONST_FF, mm1);
movq_r2r(mm0, mm5);
movq_r2r(mm3, mm6);
movq_r2r(mm2, mm7);
punpckhbw_r2r(mm3, mm2);
punpcklbw_r2r(mm6, mm7);
punpckhbw_r2r(mm1, mm0);
punpcklbw_r2r(mm1, mm5);
movq_r2r(mm7, mm1);
punpckhwd_r2r(mm5, mm7);
punpcklwd_r2r(mm5, mm1);
movq_r2r(mm2, mm4);
punpckhwd_r2r(mm0, mm2);
punpcklwd_r2r(mm0, mm4);
movntq_r2m(mm1, *(dp1));
movntq_r2m(mm7, *(dp1 + 8));
movntq_r2m(mm4, *(dp1 + 16));
movntq_r2m(mm2, *(dp1 + 24));
yp1 += 8;
up += 4;
vp += 4;
dp1 += 8 * 4;
}
/* cleanup pixles that arent a multiple of 8 pixels wide */
if (xx < w)
{
int y, u, v, r, g, b;
for (; xx < w; xx += 2)
{
u = (*up++) - 128;
v = (*vp++) - 128;
y = RZ(YMUL) * ((*yp1++) - 16);
r = LUT_CLIP((y + (_crv * v)) >> RES);
g = LUT_CLIP((y - (_cgu * u) - (_cgv * v) + RZ(OFF)) >> RES);
b = LUT_CLIP((y + (_cbu * u) + RZ(OFF)) >> RES);
*((DATA32 *) dp1) = 0xff000000 + RGB_JOIN(r,g,b);
dp1 += 4;
y = RZ(YMUL) * ((*yp1++) - 16);
r = LUT_CLIP((y + (_crv * v)) >> RES);
g = LUT_CLIP((y - (_cgu * u) - (_cgv * v) + RZ(OFF)) >> RES);
b = LUT_CLIP((y + (_cbu * u) + RZ(OFF)) >> RES);
*((DATA32 *) dp1) = 0xff000000 + RGB_JOIN(r,g,b);
dp1 += 4;
}
}
}
void deinterlace_bob_yuv_mmx(uint8_t *pdst, uint8_t *psrc,
int width, int height )
{
int Line;
long long* YVal1;
long long* YVal2;
long long* YVal3;
long long* Dest;
uint8_t* pEvenLines = psrc;
uint8_t* pOddLines = psrc+width;
int LineLength = width;
int Pitch = width * 2;
int IsOdd = 1;
long EdgeDetect = 625;
long JaggieThreshold = 73;
int n;
unsigned long long qwEdgeDetect;
unsigned long long qwThreshold;
const unsigned long long Mask = 0xfefefefefefefefeULL;
const unsigned long long YMask = 0x00ff00ff00ff00ffULL;
qwEdgeDetect = EdgeDetect;
qwEdgeDetect += (qwEdgeDetect << 48) + (qwEdgeDetect << 32) + (qwEdgeDetect << 16);
qwThreshold = JaggieThreshold;
qwThreshold += (qwThreshold << 48) + (qwThreshold << 32) + (qwThreshold << 16);
// copy first even line no matter what, and the first odd line if we're
// processing an odd field.
ac_memcpy(pdst, pEvenLines, LineLength);
if (IsOdd)
ac_memcpy(pdst + LineLength, pOddLines, LineLength);
height = height / 2;
for (Line = 0; Line < height - 1; ++Line)
{
if (IsOdd)
{
YVal1 = (long long *)(pOddLines + Line * Pitch);
YVal2 = (long long *)(pEvenLines + (Line + 1) * Pitch);
YVal3 = (long long *)(pOddLines + (Line + 1) * Pitch);
Dest = (long long *)(pdst + (Line * 2 + 2) * LineLength);
}
else
{
YVal1 = (long long *)(pEvenLines + Line * Pitch);
YVal2 = (long long *)(pOddLines + Line * Pitch);
YVal3 = (long long *)(pEvenLines + (Line + 1) * Pitch);
Dest = (long long *)(pdst + (Line * 2 + 1) * LineLength);
}
// For ease of reading, the comments below assume that we're operating on an odd
// field (i.e., that bIsOdd is true). The exact same processing is done when we
// operate on an even field, but the roles of the odd and even fields are reversed.
// It's just too cumbersome to explain the algorithm in terms of "the next odd
// line if we're doing an odd field, or the next even line if we're doing an
// even field" etc. So wherever you see "odd" or "even" below, keep in mind that
// half the time this function is called, those words' meanings will invert.
// Copy the odd line to the overlay verbatim.
ac_memcpy((char *)Dest + LineLength, YVal3, LineLength);
n = LineLength >> 3;
while( n-- )
{
movq_m2r (*YVal1++, mm0);
movq_m2r (*YVal2++, mm1);
movq_m2r (*YVal3++, mm2);
// get intensities in mm3 - 4
movq_r2r ( mm0, mm3 );
movq_r2r ( mm1, mm4 );
movq_r2r ( mm2, mm5 );
pand_m2r ( *&YMask, mm3 );
pand_m2r ( *&YMask, mm4 );
pand_m2r ( *&YMask, mm5 );
// get average in mm0
pand_m2r ( *&Mask, mm0 );
pand_m2r ( *&Mask, mm2 );
psrlw_i2r ( 01, mm0 );
psrlw_i2r ( 01, mm2 );
paddw_r2r ( mm2, mm0 );
// work out (O1 - E) * (O2 - E) / 2 - EdgeDetect * (O1 - O2) ^ 2 >> 12
// result will be in mm6
psrlw_i2r ( 01, mm3 );
psrlw_i2r ( 01, mm4 );
psrlw_i2r ( 01, mm5 );
movq_r2r ( mm3, mm6 );
psubw_r2r ( mm4, mm6 ); //mm6 = O1 - E
movq_r2r ( mm5, mm7 );
psubw_r2r ( mm4, mm7 ); //mm7 = O2 - E
pmullw_r2r ( mm7, mm6 ); // mm6 = (O1 - E) * (O2 - E)
movq_r2r ( mm3, mm7 );
psubw_r2r ( mm5, mm7 ); // mm7 = (O1 - O2)
pmullw_r2r ( mm7, mm7 ); // mm7 = (O1 - O2) ^ 2
psrlw_i2r ( 12, mm7 ); // mm7 = (O1 - O2) ^ 2 >> 12
pmullw_m2r ( *&qwEdgeDetect, mm7 );// mm7 = EdgeDetect * (O1 - O2) ^ 2 >> 12
psubw_r2r ( mm7, mm6 ); // mm6 is what we want
pcmpgtw_m2r ( *&qwThreshold, mm6 );
movq_r2r ( mm6, mm7 );
pand_r2r ( mm6, mm0 );
pandn_r2r ( mm1, mm7 );
por_r2r ( mm0, mm7 );
movq_r2m ( mm7, *Dest++ );
}
}
// Copy last odd line if we're processing an even field.
if (! IsOdd)
{
ac_memcpy(pdst + (height * 2 - 1) * LineLength,
pOddLines + (height - 1) * Pitch,
LineLength);
}
// clear out the MMX registers ready for doing floating point
// again
emms();
}
