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();
}
VLC_MMX
static int CalculateInterlaceScoreMMX( const picture_t* p_pic_top,
const picture_t* p_pic_bot )
{
assert( p_pic_top->i_planes == p_pic_bot->i_planes );
/* Amount of bits must be known for MMX, thus int32_t.
Doesn't hurt the C implementation. */
int32_t i_score_mmx = 0; /* this must be divided by 255 when finished */
int32_t i_score_c = 0; /* this counts as-is (used for non-MMX parts) */
pxor_r2r( mm7, mm7 ); /* we will keep score in mm7 */
for( int i_plane = 0 ; i_plane < p_pic_top->i_planes ; ++i_plane )
{
/* Sanity check */
if( p_pic_top->p[i_plane].i_visible_lines !=
p_pic_bot->p[i_plane].i_visible_lines )
return -1;
const int i_lasty = p_pic_top->p[i_plane].i_visible_lines-1;
const int w = FFMIN( p_pic_top->p[i_plane].i_visible_pitch,
p_pic_bot->p[i_plane].i_visible_pitch );
const int wm8 = w % 8; /* remainder */
const int w8 = w - wm8; /* part of width that is divisible by 8 */
/* Current line / neighbouring lines picture pointers */
const picture_t *cur = p_pic_bot;
const picture_t *ngh = p_pic_top;
int wc = cur->p[i_plane].i_pitch;
int wn = ngh->p[i_plane].i_pitch;
/* Transcode 1.1.5 only checks every other line. Checking every line
works better for anime, which may contain horizontal,
one pixel thick cartoon outlines.
*/
for( int y = 1; y < i_lasty; ++y )
{
uint8_t *p_c = &cur->p[i_plane].p_pixels[y*wc]; /* this line */
uint8_t *p_p = &ngh->p[i_plane].p_pixels[(y-1)*wn]; /* prev line */
uint8_t *p_n = &ngh->p[i_plane].p_pixels[(y+1)*wn]; /* next line */
int x = 0;
/* Easy-to-read C version further below.
Assumptions: 0 < T < 127
# of pixels < (2^32)/255
Note: calculates score * 255
*/
static alignas (8) const mmx_t b0 = {
.uq = 0x0000000000000000ULL };
static alignas (8) const mmx_t b128 = {
.uq = 0x8080808080808080ULL };
static alignas (8) const mmx_t bT = {
.ub = { T, T, T, T, T, T, T, T } };
for( ; x < w8; x += 8 )
{
movq_m2r( *((int64_t*)p_c), mm0 );
movq_m2r( *((int64_t*)p_p), mm1 );
movq_m2r( *((int64_t*)p_n), mm2 );
psubb_m2r( b128, mm0 );
psubb_m2r( b128, mm1 );
psubb_m2r( b128, mm2 );
psubsb_r2r( mm0, mm1 );
psubsb_r2r( mm0, mm2 );
pxor_r2r( mm3, mm3 );
pxor_r2r( mm4, mm4 );
pxor_r2r( mm5, mm5 );
pxor_r2r( mm6, mm6 );
punpcklbw_r2r( mm1, mm3 );
punpcklbw_r2r( mm2, mm4 );
punpckhbw_r2r( mm1, mm5 );
punpckhbw_r2r( mm2, mm6 );
pmulhw_r2r( mm3, mm4 );
pmulhw_r2r( mm5, mm6 );
packsswb_r2r(mm4, mm6);
pcmpgtb_m2r( bT, mm6 );
psadbw_m2r( b0, mm6 );
paddd_r2r( mm6, mm7 );
p_c += 8;
p_p += 8;
p_n += 8;
}
for( ; x < w; ++x )
{
/* Worst case: need 17 bits for "comb". */
int_fast32_t C = *p_c;
int_fast32_t P = *p_p;
int_fast32_t N = *p_n;
/* Comments in Transcode's filter_ivtc.c attribute this
combing metric to Gunnar Thalin.
The idea is that if the picture is interlaced, both
expressions will have the same sign, and this comes
up positive. The value T = 100 has been chosen such
that a pixel difference of 10 (on average) will
trigger the detector.
*/
int_fast32_t comb = (P - C) * (N - C);
if( comb > T )
++i_score_c;
++p_c;
++p_p;
++p_n;
}
/* Now the other field - swap current and neighbour pictures */
const picture_t *tmp = cur;
cur = ngh;
ngh = tmp;
int tmp_pitch = wc;
wc = wn;
wn = tmp_pitch;
}
}
movd_r2m( mm7, i_score_mmx );
emms();
return i_score_mmx/255 + i_score_c;
}
#endif
/* See header for function doc. */
int CalculateInterlaceScore( const picture_t* p_pic_top,
const picture_t* p_pic_bot )
{
/*
We use the comb metric from the IVTC filter of Transcode 1.1.5.
This was found to work better for the particular purpose of IVTC
than RenderX()'s comb metric.
Note that we *must not* subsample at all in order to catch interlacing
in telecined frames with localized motion (e.g. anime with characters
talking, where only mouths move and everything else stays still.)
*/
assert( p_pic_top != NULL );
assert( p_pic_bot != NULL );
if( p_pic_top->i_planes != p_pic_bot->i_planes )
return -1;
#ifdef CAN_COMPILE_MMXEXT
if (vlc_CPU_MMXEXT())
return CalculateInterlaceScoreMMX( p_pic_top, p_pic_bot );
#endif
int32_t i_score = 0;
for( int i_plane = 0 ; i_plane < p_pic_top->i_planes ; ++i_plane )
{
/* Sanity check */
if( p_pic_top->p[i_plane].i_visible_lines !=
p_pic_bot->p[i_plane].i_visible_lines )
return -1;
const int i_lasty = p_pic_top->p[i_plane].i_visible_lines-1;
const int w = FFMIN( p_pic_top->p[i_plane].i_visible_pitch,
p_pic_bot->p[i_plane].i_visible_pitch );
/* Current line / neighbouring lines picture pointers */
const picture_t *cur = p_pic_bot;
const picture_t *ngh = p_pic_top;
int wc = cur->p[i_plane].i_pitch;
int wn = ngh->p[i_plane].i_pitch;
/* Transcode 1.1.5 only checks every other line. Checking every line
works better for anime, which may contain horizontal,
one pixel thick cartoon outlines.
*/
for( int y = 1; y < i_lasty; ++y )
{
uint8_t *p_c = &cur->p[i_plane].p_pixels[y*wc]; /* this line */
uint8_t *p_p = &ngh->p[i_plane].p_pixels[(y-1)*wn]; /* prev line */
uint8_t *p_n = &ngh->p[i_plane].p_pixels[(y+1)*wn]; /* next line */
for( int x = 0; x < w; ++x )
{
/* Worst case: need 17 bits for "comb". */
int_fast32_t C = *p_c;
int_fast32_t P = *p_p;
int_fast32_t N = *p_n;
/* Comments in Transcode's filter_ivtc.c attribute this
combing metric to Gunnar Thalin.
The idea is that if the picture is interlaced, both
expressions will have the same sign, and this comes
up positive. The value T = 100 has been chosen such
that a pixel difference of 10 (on average) will
trigger the detector.
*/
int_fast32_t comb = (P - C) * (N - C);
if( comb > T )
++i_score;
++p_c;
++p_p;
++p_n;
}
/* Now the other field - swap current and neighbour pictures */
const picture_t *tmp = cur;
cur = ngh;
ngh = tmp;
int tmp_pitch = wc;
wc = wn;
wn = tmp_pitch;
}
}
return i_score;
}
//VLC_MMX // sunqueen delete
static int CalculateInterlaceScoreMMX( const picture_t* p_pic_top,
const picture_t* p_pic_bot )
{
assert( p_pic_top->i_planes == p_pic_bot->i_planes );
/* Amount of bits must be known for MMX, thus int32_t.
Doesn't hurt the C implementation. */
int32_t i_score_mmx = 0; /* this must be divided by 255 when finished */
int32_t i_score_c = 0; /* this counts as-is (used for non-MMX parts) */
pxor_r2r( mm7, mm7 ); /* we will keep score in mm7 */
for( int i_plane = 0 ; i_plane < p_pic_top->i_planes ; ++i_plane )
{
/* Sanity check */
if( p_pic_top->p[i_plane].i_visible_lines !=
p_pic_bot->p[i_plane].i_visible_lines )
return -1;
const int i_lasty = p_pic_top->p[i_plane].i_visible_lines-1;
const int w = FFMIN( p_pic_top->p[i_plane].i_visible_pitch,
p_pic_bot->p[i_plane].i_visible_pitch );
const int wm8 = w % 8; /* remainder */
const int w8 = w - wm8; /* part of width that is divisible by 8 */
/* Current line / neighbouring lines picture pointers */
const picture_t *cur = p_pic_bot;
const picture_t *ngh = p_pic_top;
int wc = cur->p[i_plane].i_pitch;
int wn = ngh->p[i_plane].i_pitch;
/* Transcode 1.1.5 only checks every other line. Checking every line
works better for anime, which may contain horizontal,
one pixel thick cartoon outlines.
*/
for( int y = 1; y < i_lasty; ++y )
{
uint8_t *p_c = &cur->p[i_plane].p_pixels[y*wc]; /* this line */
uint8_t *p_p = &ngh->p[i_plane].p_pixels[(y-1)*wn]; /* prev line */
uint8_t *p_n = &ngh->p[i_plane].p_pixels[(y+1)*wn]; /* next line */
int64_t i_p_c, i_p_p, i_p_n; // sunqueen add
int x = 0;
/* Easy-to-read C version further below.
Assumptions: 0 < T < 127
# of pixels < (2^32)/255
Note: calculates score * 255
*/
// static const mmx_t b0 = { .uq = 0x0000000000000000ULL };
__declspec(align(8)) static const mmx_t b0 = { /*.uq =*/ 0x0000000000000000ULL }; // sunqueen modify
// static const mmx_t b128 = { .uq = 0x8080808080808080ULL };
__declspec(align(8)) static const mmx_t b128 = { /*.uq =*/ 0x8080808080808080ULL }; // sunqueen modify
// static const mmx_t bT = { .ub = { T, T, T, T, T, T, T, T } };
__declspec(align(8)) static const mmx_t bT = { 0x6464646464646464ULL }; // sunqueen modify
for( ; x < w8; x += 8 )
{
// sunqueen add start
i_p_c = *((int64_t*)p_c);
movq_m2r( i_p_c, mm0 );
i_p_p = *((int64_t*)p_p);
movq_m2r( i_p_p, mm1 );
i_p_n = *((int64_t*)p_n);
movq_m2r( i_p_n, mm2 );
// sunqueen add end
#if 0 // sunqueen delete start
movq_m2r( *((int64_t*)p_c), mm0 );
movq_m2r( *((int64_t*)p_p), mm1 );
movq_m2r( *((int64_t*)p_n), mm2 );
#endif // sunqueen delete end
psubb_m2r( b128, mm0 );
psubb_m2r( b128, mm1 );
psubb_m2r( b128, mm2 );
psubsb_r2r( mm0, mm1 );
psubsb_r2r( mm0, mm2 );
pxor_r2r( mm3, mm3 );
pxor_r2r( mm4, mm4 );
pxor_r2r( mm5, mm5 );
pxor_r2r( mm6, mm6 );
punpcklbw_r2r( mm1, mm3 );
punpcklbw_r2r( mm2, mm4 );
punpckhbw_r2r( mm1, mm5 );
punpckhbw_r2r( mm2, mm6 );
pmulhw_r2r( mm3, mm4 );
pmulhw_r2r( mm5, mm6 );
packsswb_r2r(mm4, mm6);
pcmpgtb_m2r( bT, mm6 );
psadbw_m2r( b0, mm6 );
paddd_r2r( mm6, mm7 );
p_c += 8;
p_p += 8;
p_n += 8;
}
for( ; x < w; ++x )
{
/* Worst case: need 17 bits for "comb". */
int_fast32_t C = *p_c;
int_fast32_t P = *p_p;
int_fast32_t N = *p_n;
/* Comments in Transcode's filter_ivtc.c attribute this
combing metric to Gunnar Thalin.
The idea is that if the picture is interlaced, both
expressions will have the same sign, and this comes
up positive. The value T = 100 has been chosen such
that a pixel difference of 10 (on average) will
trigger the detector.
*/
int_fast32_t comb = (P - C) * (N - C);
if( comb > T )
++i_score_c;
++p_c;
++p_p;
++p_n;
}
/* Now the other field - swap current and neighbour pictures */
const picture_t *tmp = cur;
cur = ngh;
ngh = tmp;
int tmp_pitch = wc;
wc = wn;
wn = tmp_pitch;
}
}
movd_r2m( mm7, i_score_mmx );
emms();
return i_score_mmx/255 + i_score_c;
}