#ifdef ARCH_PPC
#include "ppc/mc.h"
#endif
+#ifdef ARCH_ARM
+#include "arm/mc.h"
+#endif
static inline void pixel_avg( uint8_t *dst, int i_dst_stride,
}
}
-static inline void pixel_avg_wxh( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int width, int height )
+static inline void pixel_avg_wxh( uint8_t *dst, int i_dst, uint8_t *src1, int i_src1, uint8_t *src2, int i_src2, int width, int height )
{
int x, y;
for( y = 0; y < height; y++ )
{
for( x = 0; x < width; x++ )
{
- dst[x] = ( dst[x] + src[x] + 1 ) >> 1;
+ dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
}
+ src1 += i_src1;
+ src2 += i_src2;
dst += i_dst;
- src += i_src;
}
}
-#define PIXEL_AVG_C( name, width, height ) \
-static void name( uint8_t *pix1, int i_stride_pix1, \
- uint8_t *pix2, int i_stride_pix2 ) \
-{ \
- pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ); \
-}
-PIXEL_AVG_C( pixel_avg_16x16, 16, 16 )
-PIXEL_AVG_C( pixel_avg_16x8, 16, 8 )
-PIXEL_AVG_C( pixel_avg_8x16, 8, 16 )
-PIXEL_AVG_C( pixel_avg_8x8, 8, 8 )
-PIXEL_AVG_C( pixel_avg_8x4, 8, 4 )
-PIXEL_AVG_C( pixel_avg_4x8, 4, 8 )
-PIXEL_AVG_C( pixel_avg_4x4, 4, 4 )
-PIXEL_AVG_C( pixel_avg_4x2, 4, 2 )
-PIXEL_AVG_C( pixel_avg_2x4, 2, 4 )
-PIXEL_AVG_C( pixel_avg_2x2, 2, 2 )
-
-
/* Implicit weighted bipred only:
* assumes log2_denom = 5, offset = 0, weight1 + weight2 = 64 */
-#define op_scale2(x) dst[x] = x264_clip_uint8( (dst[x]*i_weight1 + src[x]*i_weight2 + (1<<5)) >> 6 )
-static inline void pixel_avg_weight_wxh( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int width, int height, int i_weight1 ){
+#define op_scale2(x) dst[x] = x264_clip_uint8( (src1[x]*i_weight1 + src2[x]*i_weight2 + (1<<5)) >> 6 )
+static inline void pixel_avg_weight_wxh( uint8_t *dst, int i_dst, uint8_t *src1, int i_src1, uint8_t *src2, int i_src2, int width, int height, int i_weight1 )
+{
int y;
const int i_weight2 = 64 - i_weight1;
- for(y=0; y<height; y++, dst += i_dst, src += i_src){
+ for( y = 0; y<height; y++, dst += i_dst, src1 += i_src1, src2 += i_src2 )
+ {
op_scale2(0);
op_scale2(1);
if(width==2) continue;
op_scale2(15);
}
}
+#undef op_scale2
-#define PIXEL_AVG_WEIGHT_C( width, height ) \
-static void pixel_avg_weight_##width##x##height( \
- uint8_t *pix1, int i_stride_pix1, \
- uint8_t *pix2, int i_stride_pix2, int i_weight1 ) \
+#define PIXEL_AVG_C( name, width, height ) \
+static void name( uint8_t *pix1, int i_stride_pix1, \
+ uint8_t *pix2, int i_stride_pix2, \
+ uint8_t *pix3, int i_stride_pix3, int weight ) \
{ \
- pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height, i_weight1 ); \
+ if( weight == 32 )\
+ pixel_avg_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height ); \
+ else\
+ pixel_avg_weight_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, pix3, i_stride_pix3, width, height, weight ); \
}
-
-PIXEL_AVG_WEIGHT_C(16,16)
-PIXEL_AVG_WEIGHT_C(16,8)
-PIXEL_AVG_WEIGHT_C(8,16)
-PIXEL_AVG_WEIGHT_C(8,8)
-PIXEL_AVG_WEIGHT_C(8,4)
-PIXEL_AVG_WEIGHT_C(4,8)
-PIXEL_AVG_WEIGHT_C(4,4)
-PIXEL_AVG_WEIGHT_C(4,2)
-PIXEL_AVG_WEIGHT_C(2,4)
-PIXEL_AVG_WEIGHT_C(2,2)
-#undef op_scale2
-#undef PIXEL_AVG_WEIGHT_C
+PIXEL_AVG_C( pixel_avg_16x16, 16, 16 )
+PIXEL_AVG_C( pixel_avg_16x8, 16, 8 )
+PIXEL_AVG_C( pixel_avg_8x16, 8, 16 )
+PIXEL_AVG_C( pixel_avg_8x8, 8, 8 )
+PIXEL_AVG_C( pixel_avg_8x4, 8, 4 )
+PIXEL_AVG_C( pixel_avg_4x8, 4, 8 )
+PIXEL_AVG_C( pixel_avg_4x4, 4, 4 )
+PIXEL_AVG_C( pixel_avg_4x2, 4, 2 )
+PIXEL_AVG_C( pixel_avg_2x4, 2, 4 )
+PIXEL_AVG_C( pixel_avg_2x2, 2, 2 )
static void mc_copy( uint8_t *src, int i_src_stride, uint8_t *dst, int i_dst_stride, int i_width, int i_height )
{
#define TAPFILTER(pix, d) ((pix)[x-2*d] + (pix)[x+3*d] - 5*((pix)[x-d] + (pix)[x+2*d]) + 20*((pix)[x] + (pix)[x+d]))
static void hpel_filter( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,
- int stride, int width, int height )
+ int stride, int width, int height, int16_t *buf )
{
- int16_t *buf = x264_malloc((width+5)*sizeof(int16_t));
int x, y;
for( y=0; y<height; y++ )
{
dstc += stride;
src += stride;
}
- x264_free(buf);
}
static const int hpel_ref0[16] = {0,1,1,1,0,1,1,1,2,3,3,3,0,1,1,1};
memset( dst, 0, n );
}
+static void integral_init4h( uint16_t *sum, uint8_t *pix, int stride )
+{
+ int x, v = pix[0]+pix[1]+pix[2]+pix[3];
+ for( x=0; x<stride-4; x++ )
+ {
+ sum[x] = v + sum[x-stride];
+ v += pix[x+4] - pix[x];
+ }
+}
+
+static void integral_init8h( uint16_t *sum, uint8_t *pix, int stride )
+{
+ int x, v = pix[0]+pix[1]+pix[2]+pix[3]+pix[4]+pix[5]+pix[6]+pix[7];
+ for( x=0; x<stride-8; x++ )
+ {
+ sum[x] = v + sum[x-stride];
+ v += pix[x+8] - pix[x];
+ }
+}
+
+static void integral_init4v( uint16_t *sum8, uint16_t *sum4, int stride )
+{
+ int x;
+ for( x=0; x<stride-8; x++ )
+ sum4[x] = sum8[x+4*stride] - sum8[x];
+ for( x=0; x<stride-8; x++ )
+ sum8[x] = sum8[x+8*stride] + sum8[x+8*stride+4] - sum8[x] - sum8[x+4];
+}
+
+static void integral_init8v( uint16_t *sum8, int stride )
+{
+ int x;
+ for( x=0; x<stride-8; x++ )
+ sum8[x] = sum8[x+8*stride] - sum8[x];
+}
+
void x264_frame_init_lowres( x264_t *h, x264_frame_t *frame )
{
uint8_t *src = frame->plane[0];
// duplicate last row and column so that their interpolation doesn't have to be special-cased
for( y=0; y<i_height; y++ )
src[i_width+y*i_stride] = src[i_width-1+y*i_stride];
- h->mc.memcpy_aligned( src+i_stride*i_height, src+i_stride*(i_height-1), i_width );
+ memcpy( src+i_stride*i_height, src+i_stride*(i_height-1), i_width );
h->mc.frame_init_lowres_core( src, frame->lowres[0], frame->lowres[1], frame->lowres[2], frame->lowres[3],
i_stride, frame->i_stride_lowres, frame->i_width_lowres, frame->i_lines_lowres );
x264_frame_expand_border_lowres( frame );
for( x = 0; x < h->param.i_bframe + 2; x++ )
for( y = 0; y < h->param.i_bframe + 2; y++ )
frame->i_row_satds[y][x][0] = -1;
+
+ for( y = 0; y <= !!h->param.i_bframe; y++ )
+ for( x = 0; x <= h->param.i_bframe; x++ )
+ frame->lowres_mvs[y][x][0][0] = 0x7FFF;
}
static void frame_init_lowres_core( uint8_t *src0, uint8_t *dst0, uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,
}
}
+#if defined(__GNUC__) && (defined(ARCH_X86) || defined(ARCH_X86_64))
+// gcc isn't smart enough to use the "idiv" instruction
+static ALWAYS_INLINE int32_t div_64_32(int64_t x, int32_t y) {
+ int32_t quotient, remainder;
+ asm("idiv %4"
+ :"=a"(quotient), "=d"(remainder)
+ :"a"((uint32_t)x), "d"((int32_t)(x>>32)), "r"(y)
+ );
+ return quotient;
+}
+#else
+#define div_64_32(x,y) ((x)/(y))
+#endif
+
+/* Estimate the total amount of influence on future quality that could be had if we
+ * were to improve the reference samples used to inter predict any given macroblock. */
+static void mbtree_propagate_cost( int *dst, uint16_t *propagate_in, uint16_t *intra_costs,
+ uint16_t *inter_costs, uint16_t *inv_qscales, int len )
+{
+ int i;
+ for( i=0; i<len; i++ )
+ {
+ int propagate_amount = propagate_in[i] + ((intra_costs[i] * inv_qscales[i] + 128)>>8);
+ dst[i] = div_64_32((int64_t)propagate_amount * (intra_costs[i] - inter_costs[i]), intra_costs[i]);
+ }
+}
+
void x264_mc_init( int cpu, x264_mc_functions_t *pf )
{
pf->mc_luma = mc_luma;
pf->avg[PIXEL_2x4] = pixel_avg_2x4;
pf->avg[PIXEL_2x2] = pixel_avg_2x2;
- pf->avg_weight[PIXEL_16x16]= pixel_avg_weight_16x16;
- pf->avg_weight[PIXEL_16x8] = pixel_avg_weight_16x8;
- pf->avg_weight[PIXEL_8x16] = pixel_avg_weight_8x16;
- pf->avg_weight[PIXEL_8x8] = pixel_avg_weight_8x8;
- pf->avg_weight[PIXEL_8x4] = pixel_avg_weight_8x4;
- pf->avg_weight[PIXEL_4x8] = pixel_avg_weight_4x8;
- pf->avg_weight[PIXEL_4x4] = pixel_avg_weight_4x4;
- pf->avg_weight[PIXEL_4x2] = pixel_avg_weight_4x2;
- pf->avg_weight[PIXEL_2x4] = pixel_avg_weight_2x4;
- pf->avg_weight[PIXEL_2x2] = pixel_avg_weight_2x2;
-
+ pf->copy_16x16_unaligned = mc_copy_w16;
pf->copy[PIXEL_16x16] = mc_copy_w16;
pf->copy[PIXEL_8x8] = mc_copy_w8;
pf->copy[PIXEL_4x4] = mc_copy_w4;
pf->memzero_aligned = memzero_aligned;
pf->frame_init_lowres_core = frame_init_lowres_core;
+ pf->integral_init4h = integral_init4h;
+ pf->integral_init8h = integral_init8h;
+ pf->integral_init4v = integral_init4v;
+ pf->integral_init8v = integral_init8v;
+
+ pf->mbtree_propagate_cost = mbtree_propagate_cost;
+
#ifdef HAVE_MMX
x264_mc_init_mmx( cpu, pf );
#endif
if( cpu&X264_CPU_ALTIVEC )
x264_mc_altivec_init( pf );
#endif
+#ifdef HAVE_ARMV6
+ x264_mc_init_arm( cpu, pf );
+#endif
}
void x264_frame_filter( x264_t *h, x264_frame_t *frame, int mb_y, int b_end )
int start = (mb_y*16 >> b_interlaced) - 8; // buffer = 4 for deblock + 3 for 6tap, rounded to 8
int height = ((b_end ? frame->i_lines[0] : mb_y*16) >> b_interlaced) + 8;
int offs = start*stride - 8; // buffer = 3 for 6tap, aligned to 8 for simd
- int x, y;
+ int y;
if( mb_y & b_interlaced )
return;
frame->filtered[2] + offs,
frame->filtered[3] + offs,
frame->plane[0] + offs,
- stride, width + 16, height - start );
+ stride, width + 16, height - start,
+ h->scratch_buffer );
}
/* generate integral image:
start = -PADV;
}
if( b_end )
- height += PADV-8;
+ height += PADV-9;
for( y = start; y < height; y++ )
{
- 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-1; x++ )
- line[x] = v += ref[x] + line[x-stride] - line[x-stride-1];
- line -= 8*stride;
- if( y >= 9-PADV )
+ uint8_t *pix = frame->plane[0] + y * stride - PADH;
+ uint16_t *sum8 = frame->integral + (y+1) * stride - PADH;
+ uint16_t *sum4;
+ if( h->frames.b_have_sub8x8_esa )
+ {
+ h->mc.integral_init4h( sum8, pix, stride );
+ sum8 -= 8*stride;
+ sum4 = sum8 + stride * (frame->i_lines[0] + PADV*2);
+ if( y >= 8-PADV )
+ h->mc.integral_init4v( sum8, sum4, stride );
+ }
+ else
{
- 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];
- }
+ h->mc.integral_init8h( sum8, pix, stride );
+ if( y >= 8-PADV )
+ h->mc.integral_init8v( sum8-8*stride, stride );
}
}
}