* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111, USA.
*****************************************************************************/
-#include <stdio.h>
-#include <string.h>
-
#include "common.h"
#include "clip1.h"
-#ifdef HAVE_MMXEXT
+#ifdef HAVE_MMX
#include "i386/mc.h"
#endif
#ifdef ARCH_PPC
}
}
+static void hpel_filter( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,
+ int stride, int width, int height )
+{
+ mc_hh( src, stride, dsth, stride, width, height );
+ mc_hv( src, stride, dstv, stride, width, height );
+ mc_hc( src, stride, dstc, stride, width, height );
+}
+
+static const int hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
+static const int hpel_ref1[16] = {0,0,0,0,2,2,3,2,2,2,3,2,2,2,3,2};
+
static void mc_luma( uint8_t *src[4], int i_src_stride,
uint8_t *dst, int i_dst_stride,
- int mvx,int mvy,
+ int mvx, int mvy,
int i_width, int i_height )
{
- uint8_t *src1, *src2;
-
- int correction = (mvx&1) && (mvy&1) && ((mvx&2) ^ (mvy&2));
- int hpel1x = mvx>>1;
- int hpel1y = (mvy+1-correction)>>1;
- int filter1 = (hpel1x & 1) + ( (hpel1y & 1) << 1 );
+ int qpel_idx = ((mvy&3)<<2) + (mvx&3);
+ int offset = (mvy>>2)*i_src_stride + (mvx>>2);
+ uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
- src1 = src[filter1] + (hpel1y >> 1) * i_src_stride + (hpel1x >> 1);
-
- if ( (mvx|mvy) & 1 ) /* qpel interpolation needed */
+ if( qpel_idx & 5 ) /* qpel interpolation needed */
{
- int hpel2x = (mvx+1)>>1;
- int hpel2y = (mvy+correction)>>1;
- int filter2 = (hpel2x & 1) + ( (hpel2y & 1) <<1 );
-
- src2 = src[filter2] + (hpel2y >> 1) * i_src_stride + (hpel2x >> 1);
-
+ uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
pixel_avg( dst, i_dst_stride, src1, i_src_stride,
src2, i_src_stride, i_width, i_height );
}
}
static uint8_t *get_ref( uint8_t *src[4], int i_src_stride,
- uint8_t *dst, int * i_dst_stride,
- int mvx,int mvy,
+ uint8_t *dst, int *i_dst_stride,
+ int mvx, int mvy,
int i_width, int i_height )
{
- uint8_t *src1, *src2;
-
- int correction = (mvx&1) && (mvy&1) && ((mvx&2) ^ (mvy&2));
- int hpel1x = mvx>>1;
- int hpel1y = (mvy+1-correction)>>1;
- int filter1 = (hpel1x & 1) + ( (hpel1y & 1) << 1 );
-
- src1 = src[filter1] + (hpel1y >> 1) * i_src_stride + (hpel1x >> 1);
+ int qpel_idx = ((mvy&3)<<2) + (mvx&3);
+ int offset = (mvy>>2)*i_src_stride + (mvx>>2);
+ uint8_t *src1 = src[hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
- if ( (mvx|mvy) & 1 ) /* qpel interpolation needed */
+ if( qpel_idx & 5 ) /* qpel interpolation needed */
{
- int hpel2x = (mvx+1)>>1;
- int hpel2y = (mvy+correction)>>1;
- int filter2 = (hpel2x & 1) + ( (hpel2y & 1) <<1 );
-
- src2 = src[filter2] + (hpel2y >> 1) * i_src_stride + (hpel2x >> 1);
-
+ uint8_t *src2 = src[hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
pixel_avg( dst, *i_dst_stride, src1, i_src_stride,
src2, i_src_stride, i_width, i_height );
-
return dst;
}
else
}
}
-#ifdef HAVE_MMXEXT
-static void motion_compensation_chroma_sse( uint8_t *src, int i_src_stride,
- uint8_t *dst, int i_dst_stride,
- int mvx, int mvy,
- int i_width, int i_height )
+#define MC_COPY(W) \
+static void mc_copy_w##W( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int i_height ) \
+{ \
+ mc_copy( src, i_src, dst, i_dst, W, i_height ); \
+}
+MC_COPY( 16 )
+MC_COPY( 8 )
+MC_COPY( 4 )
+
+static void plane_copy( uint8_t *dst, int i_dst,
+ uint8_t *src, int i_src, int w, int h)
{
- if (i_width == 2) {
- motion_compensation_chroma(src, i_src_stride, dst, i_dst_stride,
- mvx, mvy, i_width, i_height);
- } else {
- const int d8x = mvx&0x07;
- const int d8y = mvy&0x07;
-
- src += (mvy >> 3) * i_src_stride + (mvx >> 3);
-
- x264_mc_chroma_sse(src, i_src_stride, dst, i_dst_stride,
- d8x, d8y, i_height, i_width);
+ while( h-- )
+ {
+ memcpy( dst, src, w );
+ dst += i_dst;
+ src += i_src;
}
}
-#endif
+
+void prefetch_fenc_null( uint8_t *pix_y, int stride_y,
+ uint8_t *pix_uv, int stride_uv, int mb_x )
+{}
+
+void prefetch_ref_null( uint8_t *pix, int stride, int parity )
+{}
void x264_mc_init( int cpu, x264_mc_functions_t *pf )
{
pf->avg_weight[PIXEL_2x4] = pixel_avg_weight_2x4;
pf->avg_weight[PIXEL_2x2] = pixel_avg_weight_2x2;
-#ifdef HAVE_MMXEXT
+ pf->copy[PIXEL_16x16] = mc_copy_w16;
+ pf->copy[PIXEL_8x8] = mc_copy_w8;
+ pf->copy[PIXEL_4x4] = mc_copy_w4;
+
+ pf->plane_copy = plane_copy;
+ pf->hpel_filter = hpel_filter;
+
+ pf->prefetch_fenc = prefetch_fenc_null;
+ pf->prefetch_ref = prefetch_ref_null;
+
+#ifdef HAVE_MMX
if( cpu&X264_CPU_MMXEXT ) {
x264_mc_mmxext_init( pf );
- pf->mc_chroma = motion_compensation_chroma_sse;
+ pf->mc_chroma = x264_mc_chroma_mmxext;
}
-#endif
-#ifdef HAVE_SSE2
if( cpu&X264_CPU_SSE2 )
x264_mc_sse2_init( pf );
#endif
#endif
}
-extern void x264_horizontal_filter_mmxext( uint8_t *dst, int i_dst_stride,
- uint8_t *src, int i_src_stride,
- int i_width, int i_height );
-extern void x264_center_filter_mmxext( uint8_t *dst1, int i_dst1_stride,
- uint8_t *dst2, int i_dst2_stride,
- uint8_t *src, int i_src_stride,
- int i_width, int i_height );
-
-void x264_frame_filter( int cpu, x264_frame_t *frame )
+void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
{
- const int x_inc = 16, y_inc = 16;
- const int stride = frame->i_stride[0];
+ const int b_interlaced = h->sh.b_mbaff;
+ const int stride = frame->i_stride[0] << b_interlaced;
+ const int width = frame->i_width[0];
+ int start = (mb_y*16 >> b_interlaced) - 8;
+ int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
+ int offs = start*stride - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
int x, y;
- pf_mc_t int_h = mc_hh;
- pf_mc_t int_v = mc_hv;
- pf_mc_t int_hv = mc_hc;
+ if( mb_y & b_interlaced )
+ return;
-#ifdef HAVE_MMXEXT
- if ( cpu & X264_CPU_MMXEXT )
+ for( y=0; y<=b_interlaced; y++, offs+=frame->i_stride[0] )
{
- x264_horizontal_filter_mmxext(frame->filtered[1] - 8 * stride - 8, stride,
- frame->plane[0] - 8 * stride - 8, stride,
- stride - 48, frame->i_lines[0] + 16);
- x264_center_filter_mmxext(frame->filtered[2] - 8 * stride - 8, stride,
- frame->filtered[3] - 8 * stride - 8, stride,
- frame->plane[0] - 8 * stride - 8, stride,
- stride - 48, frame->i_lines[0] + 16);
- }
- else
-#endif
- {
- for( y = -8; y < frame->i_lines[0]+8; y += y_inc )
- {
- uint8_t *p_in = frame->plane[0] + y * stride - 8;
- uint8_t *p_h = frame->filtered[1] + y * stride - 8;
- uint8_t *p_v = frame->filtered[2] + y * stride - 8;
- uint8_t *p_hv = frame->filtered[3] + y * stride - 8;
- for( x = -8; x < stride - 64 + 8; x += x_inc )
- {
- int_h( p_in, stride, p_h, stride, x_inc, y_inc );
- int_v( p_in, stride, p_v, stride, x_inc, y_inc );
- int_hv( p_in, stride, p_hv, stride, x_inc, y_inc );
-
- p_h += x_inc;
- p_v += x_inc;
- p_hv += x_inc;
- p_in += x_inc;
- }
- }
+ h->mc.hpel_filter(
+ frame->filtered[1] + offs,
+ frame->filtered[2] + offs,
+ frame->filtered[3] + offs,
+ frame->plane[0] + offs,
+ stride, width + 16, height - start );
}
/* generate integral image:
- * each entry in frame->integral is the sum of all luma samples above and
- * to the left of its location (inclusive).
- * this allows us to calculate the DC of any rectangle by looking only
- * at the corner entries.
- * individual entries will overflow 16 bits, but that's ok:
- * we only need the differences between entries, and those will be correct
- * as long as we don't try to evaluate a rectangle bigger than 16x16.
- * likewise, we don't really have to init the edges to 0, leaving garbage
- * there wouldn't affect the results.*/
+ * frame->integral contains 2 planes. in the upper plane, each element is
+ * the sum of an 8x8 pixel region with top-left corner on that point.
+ * in the lower plane, 4x4 sums (needed only with --partitions p4x4). */
if( frame->integral )
{
- memset( frame->integral - 32 * stride - 32, 0, stride * sizeof(uint16_t) );
- for( y = -31; y < frame->i_lines[0] + 32; y++ )
+ if( start < 0 )
+ {
+ memset( frame->integral - PADV * stride - PADH, 0, stride * sizeof(uint16_t) );
+ start = -PADV;
+ }
+ if( b_end )
+ height += PADV-8;
+ for( y = start; y < height; y++ )
{
- uint8_t *ref = frame->plane[0] + y * stride - 32;
- uint16_t *line = frame->integral + y * stride - 32;
+ uint8_t *ref = frame->plane[0] + y * stride - PADH;
+ uint16_t *line = frame->integral + (y+1) * stride - PADH + 1;
uint16_t v = line[0] = 0;
- for( x = 1; x < stride; x++ )
+ for( x = 0; x < stride-1; x++ )
line[x] = v += ref[x] + line[x-stride] - line[x-stride-1];
+ line -= 8*stride;
+ if( y >= 9-PADV )
+ {
+ uint16_t *sum4 = line + stride * (frame->i_lines[0] + PADV*2);
+ for( x = 1; x < stride-8; x++, line++, sum4++ )
+ {
+ sum4[0] = line[4+4*stride] - line[4] - line[4*stride] + line[0];
+ line[0] += line[8+8*stride] - line[8] - line[8*stride];
+ }
+ }
}
}
}
-void x264_frame_init_lowres( int cpu, x264_frame_t *frame )
+void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
{
// FIXME: tapfilter?
const int i_stride = frame->i_stride[0];
const int i_stride2 = frame->i_stride_lowres;
- const int i_width2 = i_stride2 - 64;
+ const int i_width2 = frame->i_width_lowres;
int x, y, i;
for( y = 0; y < frame->i_lines_lowres - 1; y++ )
{
projects / x264 / blobdiff