static inline void mean8(unsigned char *refpix,unsigned char *pixel,int radius_count,int row_stride,int threshold,int8_t *diff,unsigned char *count)
{
int a,b;
pxor_r2r(mm6,mm6); // mm6 (aka count) = 0
pxor_r2r(mm7,mm7); // mm7 (aka diff) = 0
movq_m2r(*refpix,mm3); // mm3 = refpix[0]
movd_g2r(0x80808080,mm4); // mm4 = 128
punpcklbw_r2r(mm4,mm4);
pxor_r2r(mm4,mm3); // mm3 = refpix[0]-128
movd_g2r(threshold,mm5); // mm5 = threshold
punpcklbw_r2r(mm5,mm5);
punpcklbw_r2r(mm5,mm5);
punpcklbw_r2r(mm5,mm5);
for( b=0; b<radius_count; b++ ) {
for( a=0; a<radius_count; a++ ) {
movq_m2r(*pixel,mm0); // mm0 = pixel[0]
pxor_r2r(mm4,mm0); // mm0 = pixel[0]-128
movq_r2r(mm3,mm2); // mm2 = refpix[0]-128
psubsb_r2r(mm0,mm2); // mm2 = refpix[0]-pixel[0]
psubsb_r2r(mm3,mm0); // mm0 = pixel[0]-refpix[0]
pminub_r2r(mm0,mm2); // mm2 = abs(pixel[0]-refpix[0])
movq_r2r(mm5,mm1); // mm1 = threshold
pcmpgtb_r2r(mm2,mm1); // mm1 = (threshold > abs(pixel[0]-refpix[0])) ? -1 : 0
psubb_r2r(mm1,mm6); // mm6 += (threshold > abs(pixel[0]-refpix[0]))
pand_r2r(mm1,mm0); // mm0 = (threshold > abs(pixel[0]-refpix[0])) ? pixel[0]-refpix[0] : 0
paddb_r2r(mm0,mm7); // mm7 += (threshold > abs(pixel[0]-refpix[0])) ? pixel[0]-refpix[0] : 0
++pixel;
}
pixel += row_stride - radius_count;
}
movq_r2m(mm6,*count);
movq_r2m(mm7,*diff);
emms();
}
static void frame_f2i_sse(float *src,u_char *dst,int l)
{
int i;
// put 128 in all 4 words of mm7
movd_g2r(128,mm7);
punpcklwd_r2r(mm7,mm7);
punpckldq_r2r(mm7,mm7);
// put 128 in all 8 bytes of mm6
movd_g2r(128,mm6);
punpcklbw_r2r(mm6,mm6);
punpcklwd_r2r(mm6,mm6);
punpckldq_r2r(mm6,mm6);
for( i=0; i<l; i+=8 ) {
movaps_m2r(src[0],xmm0);
movaps_m2r(src[4],xmm2);
movhlps_r2r(xmm0,xmm1);
cvtps2pi_r2r(xmm0,mm0);
cvtps2pi_r2r(xmm1,mm1);
movhlps_r2r(xmm2,xmm3);
cvtps2pi_r2r(xmm2,mm2);
cvtps2pi_r2r(xmm3,mm3);
packssdw_r2r(mm1,mm0);
packssdw_r2r(mm3,mm2);
psubw_r2r(mm7,mm0);
psubw_r2r(mm7,mm2);
packsswb_r2r(mm2, mm0);
paddb_r2r(mm6, mm0);
movq_r2m(mm0,dst[0]);
src+=8;
dst+=8;
}
emms();
}
static void
deinterlace_scanline_linear_mmx (GstDeinterlaceSimpleMethod * self,
guint8 * out, const guint8 * bot, const guint8 * top, gint size)
{
const mmx_t shiftmask = { 0xfefffefffefffeffULL }; /* To avoid shifting chroma to luma. */
int i;
for (i = size / 32; i; --i) {
movq_m2r (*bot, mm0);
movq_m2r (*top, mm1);
movq_m2r (*(bot + 8), mm2);
movq_m2r (*(top + 8), mm3);
movq_m2r (*(bot + 16), mm4);
movq_m2r (*(top + 16), mm5);
movq_m2r (*(bot + 24), mm6);
movq_m2r (*(top + 24), mm7);
pand_m2r (shiftmask, mm0);
pand_m2r (shiftmask, mm1);
pand_m2r (shiftmask, mm2);
pand_m2r (shiftmask, mm3);
pand_m2r (shiftmask, mm4);
pand_m2r (shiftmask, mm5);
pand_m2r (shiftmask, mm6);
pand_m2r (shiftmask, mm7);
psrlw_i2r (1, mm0);
psrlw_i2r (1, mm1);
psrlw_i2r (1, mm2);
psrlw_i2r (1, mm3);
psrlw_i2r (1, mm4);
psrlw_i2r (1, mm5);
psrlw_i2r (1, mm6);
psrlw_i2r (1, mm7);
paddb_r2r (mm1, mm0);
paddb_r2r (mm3, mm2);
paddb_r2r (mm5, mm4);
paddb_r2r (mm7, mm6);
movq_r2m (mm0, *out);
movq_r2m (mm2, *(out + 8));
movq_r2m (mm4, *(out + 16));
movq_r2m (mm6, *(out + 24));
out += 32;
top += 32;
bot += 32;
}
size = (size & 0x1f);
for (i = size / 8; i; --i) {
movq_m2r (*bot, mm0);
movq_m2r (*top, mm1);
pand_m2r (shiftmask, mm0);
pand_m2r (shiftmask, mm1);
psrlw_i2r (1, mm0);
psrlw_i2r (1, mm1);
paddb_r2r (mm1, mm0);
movq_r2m (mm0, *out);
out += 8;
top += 8;
bot += 8;
}
emms ();
size = size & 0xf;
/* Handle last few pixels. */
for (i = size; i; --i) {
*out++ = ((*top++) + (*bot++)) >> 1;
}
}
static void
deinterlace_scanline_linear_mmx (GstDeinterlaceMethod * self,
GstDeinterlace * parent, guint8 * out,
GstDeinterlaceScanlineData * scanlines, gint width)
{
const mmx_t shiftmask = { 0xfefffefffefffeffULL }; /* To avoid shifting chroma to luma. */
int i;
guint8 *bot = scanlines->b0, *top = scanlines->t0;
for (i = width / 16; i; --i) {
movq_m2r (*bot, mm0);
movq_m2r (*top, mm1);
movq_m2r (*(bot + 8), mm2);
movq_m2r (*(top + 8), mm3);
movq_m2r (*(bot + 16), mm4);
movq_m2r (*(top + 16), mm5);
movq_m2r (*(bot + 24), mm6);
movq_m2r (*(top + 24), mm7);
pand_m2r (shiftmask, mm0);
pand_m2r (shiftmask, mm1);
pand_m2r (shiftmask, mm2);
pand_m2r (shiftmask, mm3);
pand_m2r (shiftmask, mm4);
pand_m2r (shiftmask, mm5);
pand_m2r (shiftmask, mm6);
pand_m2r (shiftmask, mm7);
psrlw_i2r (1, mm0);
psrlw_i2r (1, mm1);
psrlw_i2r (1, mm2);
psrlw_i2r (1, mm3);
psrlw_i2r (1, mm4);
psrlw_i2r (1, mm5);
psrlw_i2r (1, mm6);
psrlw_i2r (1, mm7);
paddb_r2r (mm1, mm0);
paddb_r2r (mm3, mm2);
paddb_r2r (mm5, mm4);
paddb_r2r (mm7, mm6);
movq_r2m (mm0, *out);
movq_r2m (mm2, *(out + 8));
movq_r2m (mm4, *(out + 16));
movq_r2m (mm6, *(out + 24));
out += 32;
top += 32;
bot += 32;
}
width = (width & 0xf);
for (i = width / 4; i; --i) {
movq_m2r (*bot, mm0);
movq_m2r (*top, mm1);
pand_m2r (shiftmask, mm0);
pand_m2r (shiftmask, mm1);
psrlw_i2r (1, mm0);
psrlw_i2r (1, mm1);
paddb_r2r (mm1, mm0);
movq_r2m (mm0, *out);
out += 8;
top += 8;
bot += 8;
}
width = width & 0x7;
/* Handle last few pixels. */
for (i = width * 2; i; --i) {
*out++ = ((*top++) + (*bot++)) >> 1;
}
emms ();
}
static void interpolate_packed422_scanline_mmx( uint8_t *output, uint8_t *top,
uint8_t *bot, int width )
{
const mmx_t shiftmask = { 0xfefffefffefffeffULL }; /* To avoid shifting chroma to luma. */
int i;
for( i = width/16; i; --i ) {
movq_m2r( *bot, mm0 );
movq_m2r( *top, mm1 );
movq_m2r( *(bot + 8), mm2 );
movq_m2r( *(top + 8), mm3 );
movq_m2r( *(bot + 16), mm4 );
movq_m2r( *(top + 16), mm5 );
movq_m2r( *(bot + 24), mm6 );
movq_m2r( *(top + 24), mm7 );
pand_m2r( shiftmask, mm0 );
pand_m2r( shiftmask, mm1 );
pand_m2r( shiftmask, mm2 );
pand_m2r( shiftmask, mm3 );
pand_m2r( shiftmask, mm4 );
pand_m2r( shiftmask, mm5 );
pand_m2r( shiftmask, mm6 );
pand_m2r( shiftmask, mm7 );
psrlw_i2r( 1, mm0 );
psrlw_i2r( 1, mm1 );
psrlw_i2r( 1, mm2 );
psrlw_i2r( 1, mm3 );
psrlw_i2r( 1, mm4 );
psrlw_i2r( 1, mm5 );
psrlw_i2r( 1, mm6 );
psrlw_i2r( 1, mm7 );
paddb_r2r( mm1, mm0 );
paddb_r2r( mm3, mm2 );
paddb_r2r( mm5, mm4 );
paddb_r2r( mm7, mm6 );
movq_r2m( mm0, *output );
movq_r2m( mm2, *(output + 8) );
movq_r2m( mm4, *(output + 16) );
movq_r2m( mm6, *(output + 24) );
output += 32;
top += 32;
bot += 32;
}
width = (width & 0xf);
for( i = width/4; i; --i ) {
movq_m2r( *bot, mm0 );
movq_m2r( *top, mm1 );
pand_m2r( shiftmask, mm0 );
pand_m2r( shiftmask, mm1 );
psrlw_i2r( 1, mm0 );
psrlw_i2r( 1, mm1 );
paddb_r2r( mm1, mm0 );
movq_r2m( mm0, *output );
output += 8;
top += 8;
bot += 8;
}
width = width & 0x7;
/* Handle last few pixels. */
for( i = width * 2; i; --i ) {
*output++ = ((*top++) + (*bot++)) >> 1;
}
emms();
}
