/*
* Do a shift right on the 4*4 block in the MMX registers
*
*/
static __inline__ void shift_blk(const uint32_t shift)
{
psrlq_i2r( shift,mm0);
psrlq_i2r( shift,mm1);
psrlq_i2r( shift,mm2);
psrlq_i2r( shift,mm3);
}
static inline void mmx_interp_average_2_U8 (uint8_t * dest,
const uint8_t * src1,
const uint8_t * src2)
{
/* *dest = (*dest + (*src1 + *src2 + 1)/ 2 + 1)/ 2; */
movq_m2r (*dest, mm1); /* load 8 dest bytes */
movq_r2r (mm1, mm2); /* copy 8 dest bytes */
movq_m2r (*src1, mm3); /* load 8 src1 bytes */
movq_r2r (mm3, mm4); /* copy 8 src1 bytes */
movq_m2r (*src2, mm5); /* load 8 src2 bytes */
movq_r2r (mm5, mm6); /* copy 8 src2 bytes */
pxor_r2r (mm3, mm5); /* xor src1 and src2 */
pand_m2r (mask1, mm5); /* mask lower bits */
psrlq_i2r (1, mm5); /* /2 */
por_r2r (mm4, mm6); /* or src1 and src2 */
psubb_r2r (mm5, mm6); /* subtract subresults */
movq_r2r (mm6, mm5); /* copy subresult */
pxor_r2r (mm1, mm5); /* xor srcavg and dest */
pand_m2r (mask1, mm5); /* mask lower bits */
psrlq_i2r (1, mm5); /* /2 */
por_r2r (mm2, mm6); /* or srcavg and dest */
psubb_r2r (mm5, mm6); /* subtract subresults */
movq_r2m (mm6, *dest); /* store result in dest */
}
static inline void mmx_unpack_16rgb (uint8_t * image, const int cpu)
{
static mmx_t mmx_bluemask = {0xf8f8f8f8f8f8f8f8LL};
static mmx_t mmx_greenmask = {0xfcfcfcfcfcfcfcfcLL};
static mmx_t mmx_redmask = {0xf8f8f8f8f8f8f8f8LL};
/*
* convert RGB plane to RGB 16 bits
* mm0 -> B, mm1 -> R, mm2 -> G
* mm4 -> GB, mm5 -> AR pixel 4-7
* mm6 -> GB, mm7 -> AR pixel 0-3
*/
pand_m2r (mmx_bluemask, mm0); /* mm0 = b7b6b5b4b3______ */
pand_m2r (mmx_greenmask, mm2); /* mm2 = g7g6g5g4g3g2____ */
pand_m2r (mmx_redmask, mm1); /* mm1 = r7r6r5r4r3______ */
psrlq_i2r (3, mm0); /* mm0 = ______b7b6b5b4b3 */
pxor_r2r (mm4, mm4); /* mm4 = 0 */
movq_r2r (mm0, mm5); /* mm5 = ______b7b6b5b4b3 */
movq_r2r (mm2, mm7); /* mm7 = g7g6g5g4g3g2____ */
punpcklbw_r2r (mm4, mm2);
punpcklbw_r2r (mm1, mm0);
psllq_i2r (3, mm2);
por_r2r (mm2, mm0);
movntq (mm0, *image);
punpckhbw_r2r (mm4, mm7);
punpckhbw_r2r (mm1, mm5);
psllq_i2r (3, mm7);
por_r2r (mm7, mm5);
movntq (mm5, *(image+8));
}
static __inline__ void
mmx_sum_4_word_accs( mmx_t *accs, int32_t *res )
{
movq_m2r( *accs, mm1 );
movq_r2r( mm1, mm3 );
movq_r2r( mm1, mm2 );
/* Generate sign extensions for mm1 words! */
psraw_i2r( 15, mm3 );
punpcklwd_r2r( mm3, mm1 );
punpckhwd_r2r( mm3, mm2 );
paddd_r2r( mm1, mm2 );
movq_r2r( mm2, mm3);
psrlq_i2r( 32, mm2);
paddd_r2r( mm2, mm3);
movd_r2m( mm3, *res );
}
static void vfilter_chroma_332_packed422_scanline_mmx( uint8_t *output, int width,
uint8_t *m, uint8_t *t, uint8_t *b )
{
int i;
const mmx_t ymask = { 0x00ff00ff00ff00ffULL };
const mmx_t cmask = { 0xff00ff00ff00ff00ULL };
// Get width in bytes.
width *= 2;
i = width / 8;
width -= i * 8;
movq_m2r( ymask, mm7 );
movq_m2r( cmask, mm6 );
while ( i-- )
{
movq_m2r( *t, mm0 );
movq_m2r( *b, mm1 );
movq_m2r( *m, mm2 );
movq_r2r ( mm2, mm3 );
pand_r2r ( mm7, mm3 );
pand_r2r ( mm6, mm0 );
pand_r2r ( mm6, mm1 );
pand_r2r ( mm6, mm2 );
psrlq_i2r( 8, mm0 );
psrlq_i2r( 7, mm1 );
psrlq_i2r( 8, mm2 );
movq_r2r ( mm0, mm4 );
psllw_i2r( 1, mm4 );
paddw_r2r( mm4, mm0 );
movq_r2r ( mm2, mm4 );
psllw_i2r( 1, mm4 );
paddw_r2r( mm4, mm2 );
paddw_r2r( mm0, mm2 );
paddw_r2r( mm1, mm2 );
psllw_i2r( 5, mm2 );
pand_r2r( mm6, mm2 );
por_r2r ( mm3, mm2 );
movq_r2m( mm2, *output );
output += 8;
t += 8;
b += 8;
m += 8;
}
output++; t++; b++; m++;
while ( width-- )
{
*output = (3 * *t + 3 * *m + 2 * *b) >> 3;
output +=2; t+=2; b+=2; m+=2;
}
emms();
}
static inline void mmx_interp_average_4_U8 (uint8_t * dest,
const uint8_t * src1,
const uint8_t * src2,
const uint8_t * src3,
const uint8_t * src4)
{
/* *dest = (*dest + (*src1 + *src2 + *src3 + *src4 + 2)/ 4 + 1)/ 2; */
movq_m2r (*src1, mm1); /* load 8 src1 bytes */
movq_r2r (mm1, mm2); /* copy 8 src1 bytes */
punpcklbw_r2r (mm0, mm1); /* unpack low src1 bytes */
punpckhbw_r2r (mm0, mm2); /* unpack high src1 bytes */
movq_m2r (*src2, mm3); /* load 8 src2 bytes */
movq_r2r (mm3, mm4); /* copy 8 src2 bytes */
punpcklbw_r2r (mm0, mm3); /* unpack low src2 bytes */
punpckhbw_r2r (mm0, mm4); /* unpack high src2 bytes */
paddw_r2r (mm3, mm1); /* add lows */
paddw_r2r (mm4, mm2); /* add highs */
/* now have partials in mm1 and mm2 */
movq_m2r (*src3, mm3); /* load 8 src3 bytes */
movq_r2r (mm3, mm4); /* copy 8 src3 bytes */
punpcklbw_r2r (mm0, mm3); /* unpack low src3 bytes */
punpckhbw_r2r (mm0, mm4); /* unpack high src3 bytes */
paddw_r2r (mm3, mm1); /* add lows */
paddw_r2r (mm4, mm2); /* add highs */
movq_m2r (*src4, mm5); /* load 8 src4 bytes */
movq_r2r (mm5, mm6); /* copy 8 src4 bytes */
punpcklbw_r2r (mm0, mm5); /* unpack low src4 bytes */
punpckhbw_r2r (mm0, mm6); /* unpack high src4 bytes */
paddw_r2r (mm5, mm1); /* add lows */
paddw_r2r (mm6, mm2); /* add highs */
paddw_m2r (round4, mm1);
psraw_i2r (2, mm1); /* /4 */
paddw_m2r (round4, mm2);
psraw_i2r (2, mm2); /* /4 */
/* now have subtotal/4 in mm1 and mm2 */
movq_m2r (*dest, mm3); /* load 8 dest bytes */
movq_r2r (mm3, mm4); /* copy 8 dest bytes */
packuswb_r2r (mm2, mm1); /* pack (w/ saturation) */
movq_r2r (mm1,mm2); /* copy subresult */
pxor_r2r (mm1, mm3); /* xor srcavg and dest */
pand_m2r (mask1, mm3); /* mask lower bits */
psrlq_i2r (1, mm3); /* /2 */
por_r2r (mm2, mm4); /* or srcavg and dest */
psubb_r2r (mm3, mm4); /* subtract subresults */
movq_r2m (mm4, *dest); /* store result in dest */
}
static __inline__ int qblock_sad_mmx(uint8_t *refblk,
uint32_t h,
uint32_t rowstride)
{
int res;
pxor_r2r (mm4,mm4);
movq_r2r (mm0,mm5); /* First row */
movd_m2r (*refblk, mm6);
pxor_r2r ( mm7, mm7);
refblk += rowstride;
punpcklbw_r2r ( mm7, mm5);
punpcklbw_r2r ( mm7, mm6);
movq_r2r ( mm5, mm7);
psubusw_r2r ( mm6, mm5);
psubusw_r2r ( mm7, mm6);
paddw_r2r ( mm5, mm4);
paddw_r2r ( mm6, mm4 );
movq_r2r (mm1,mm5); /* Second row */
movd_m2r (*refblk, mm6);
pxor_r2r ( mm7, mm7);
refblk += rowstride;
punpcklbw_r2r ( mm7, mm5);
punpcklbw_r2r ( mm7, mm6);
movq_r2r ( mm5, mm7);
psubusw_r2r ( mm6, mm5);
psubusw_r2r ( mm7, mm6);
paddw_r2r ( mm5, mm4);
paddw_r2r ( mm6, mm4 );
if( h == 4 )
{
movq_r2r (mm2,mm5); /* Third row */
movd_m2r (*refblk, mm6);
pxor_r2r ( mm7, mm7);
refblk += rowstride;
punpcklbw_r2r ( mm7, mm5);
punpcklbw_r2r ( mm7, mm6);
movq_r2r ( mm5, mm7);
psubusw_r2r ( mm6, mm5);
psubusw_r2r ( mm7, mm6);
paddw_r2r ( mm5, mm4);
paddw_r2r ( mm6, mm4 );
movq_r2r (mm3,mm5); /* Fourth row */
movd_m2r (*refblk, mm6);
pxor_r2r ( mm7, mm7);
punpcklbw_r2r ( mm7, mm5);
punpcklbw_r2r ( mm7, mm6);
movq_r2r ( mm5, mm7);
psubusw_r2r ( mm6, mm5);
psubusw_r2r ( mm7, mm6);
paddw_r2r ( mm5, mm4);
paddw_r2r ( mm6, mm4 );
}
movq_r2r ( mm4, mm5 );
psrlq_i2r ( 32, mm5 );
paddw_r2r ( mm5, mm4 );
movq_r2r ( mm4, mm6 );
psrlq_i2r ( 16, mm6 );
paddw_r2r ( mm6, mm4 );
movd_r2m ( mm4, res );
return res & 0xffff;
}
int bsad_mmx(uint8_t *pf, uint8_t *pb, uint8_t *p2, int lx, int hxf, int hyf, int hxb, int hyb, int h)
{
uint8_t *pfa,*pfb,*pfc,*pba,*pbb,*pbc;
int s, s1, s2;
pfa = pf + hxf;
pfb = pf + lx * hyf;
pfc = pfb + hxf;
pba = pb + hxb;
pbb = pb + lx * hyb;
pbc = pbb + hxb;
s = 0; /* the accumulator */
if (h > 0)
{
pxor_r2r(mm7, mm7);
pxor_r2r(mm6, mm6);
pcmpeqw_r2r(mm5, mm5);
psubw_r2r(mm5, mm6);
psllw_i2r(1, mm6);
do {
BSAD_LOAD(pf[0],mm0,mm1);
BSAD_LOAD_ACC(pfa[0],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pfb[0],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pfc[0],mm2,mm3,mm0,mm1);
paddw_r2r(mm6, mm0);
paddw_r2r(mm6, mm1);
psrlw_i2r(2, mm0);
psrlw_i2r(2, mm1);
BSAD_LOAD(pb[0],mm2,mm3);
BSAD_LOAD_ACC(pba[0],mm4,mm5,mm2,mm3);
BSAD_LOAD_ACC(pbb[0],mm4,mm5,mm2,mm3);
BSAD_LOAD_ACC(pbc[0],mm4,mm5,mm2,mm3);
paddw_r2r(mm6, mm2);
paddw_r2r(mm6, mm3);
psrlw_i2r(2, mm2);
psrlw_i2r(2, mm3);
paddw_r2r(mm2, mm0);
paddw_r2r(mm3, mm1);
psrlw_i2r(1, mm6);
paddw_r2r(mm6, mm0);
paddw_r2r(mm6, mm1);
psllw_i2r(1, mm6);
psrlw_i2r(1, mm0);
psrlw_i2r(1, mm1);
packuswb_r2r(mm1, mm0);
movq_m2r(p2[0], mm1);
movq_r2r(mm0, mm2);
psubusb_r2r(mm1, mm0);
psubusb_r2r(mm2, mm1);
por_r2r(mm1, mm0);
movq_r2r(mm0, mm1);
punpcklbw_r2r(mm7, mm0);
punpckhbw_r2r(mm7, mm1);
paddw_r2r(mm1, mm0);
movq_r2r(mm0, mm1);
punpcklwd_r2r(mm7, mm0);
punpckhwd_r2r(mm7, mm1);
paddd_r2r(mm1, mm0);
movd_r2g(mm0, s1);
psrlq_i2r(32, mm0);
movd_r2g(mm0, s2);
s += s1 + s2;
BSAD_LOAD(pf[8],mm0,mm1);
BSAD_LOAD_ACC(pfa[8],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pfb[8],mm2,mm3,mm0,mm1);
BSAD_LOAD_ACC(pfc[8],mm2,mm3,mm0,mm1);
paddw_r2r(mm6, mm0);
paddw_r2r(mm6, mm1);
psrlw_i2r(2, mm0);
psrlw_i2r(2, mm1);
BSAD_LOAD(pb[8],mm2,mm3);
BSAD_LOAD_ACC(pba[8],mm4,mm5,mm2,mm3);
BSAD_LOAD_ACC(pbb[8],mm4,mm5,mm2,mm3);
BSAD_LOAD_ACC(pbc[8],mm4,mm5,mm2,mm3);
paddw_r2r(mm6, mm2);
paddw_r2r(mm6, mm3);
psrlw_i2r(2, mm2);
psrlw_i2r(2, mm3);
paddw_r2r(mm2, mm0);
paddw_r2r(mm3, mm1);
psrlw_i2r(1, mm6);
paddw_r2r(mm6, mm0);
paddw_r2r(mm6, mm1);
psllw_i2r(1, mm6);
psrlw_i2r(1, mm0);
psrlw_i2r(1, mm1);
packuswb_r2r(mm1, mm0);
movq_m2r(p2[8], mm1);
movq_r2r(mm0, mm2);
psubusb_r2r(mm1, mm0);
psubusb_r2r(mm2, mm1);
por_r2r(mm1, mm0);
movq_r2r(mm0, mm1);
punpcklbw_r2r(mm7, mm0);
punpckhbw_r2r(mm7, mm1);
paddw_r2r(mm1, mm0);
movq_r2r(mm0, mm1);
punpcklwd_r2r(mm7, mm0);
punpckhwd_r2r(mm7, mm1);
paddd_r2r(mm1, mm0);
movd_r2g(mm0, s1);
psrlq_i2r(32, mm0);
movd_r2g(mm0, s2);
s += s1 + s2;
p2 += lx;
pf += lx;
pfa += lx;
pfb += lx;
pfc += lx;
pb += lx;
pba += lx;
pbb += lx;
pbc += lx;
h--;
} while (h > 0);
}
emms();
return s;
}
static inline int XDeint8x8DetectMMXEXT( uint8_t *src, int i_src )
{
int y, x;
int32_t ff, fr;
int fc;
/* Detect interlacing */
fc = 0;
pxor_r2r( mm7, mm7 );
for( y = 0; y < 9; y += 2 )
{
ff = fr = 0;
pxor_r2r( mm5, mm5 );
pxor_r2r( mm6, mm6 );
for( x = 0; x < 8; x+=4 )
{
movd_m2r( src[ x], mm0 );
movd_m2r( src[1*i_src+x], mm1 );
movd_m2r( src[2*i_src+x], mm2 );
movd_m2r( src[3*i_src+x], mm3 );
punpcklbw_r2r( mm7, mm0 );
punpcklbw_r2r( mm7, mm1 );
punpcklbw_r2r( mm7, mm2 );
punpcklbw_r2r( mm7, mm3 );
movq_r2r( mm0, mm4 );
psubw_r2r( mm1, mm0 );
psubw_r2r( mm2, mm4 );
psubw_r2r( mm1, mm2 );
psubw_r2r( mm1, mm3 );
pmaddwd_r2r( mm0, mm0 );
pmaddwd_r2r( mm4, mm4 );
pmaddwd_r2r( mm2, mm2 );
pmaddwd_r2r( mm3, mm3 );
paddd_r2r( mm0, mm2 );
paddd_r2r( mm4, mm3 );
paddd_r2r( mm2, mm5 );
paddd_r2r( mm3, mm6 );
}
movq_r2r( mm5, mm0 );
psrlq_i2r( 32, mm0 );
paddd_r2r( mm0, mm5 );
movd_r2m( mm5, fr );
movq_r2r( mm6, mm0 );
psrlq_i2r( 32, mm0 );
paddd_r2r( mm0, mm6 );
movd_r2m( mm6, ff );
if( ff < 6*fr/8 && fr > 32 )
fc++;
src += 2*i_src;
}
return fc;
}