Commit 1d22dd50 authored by Daniel Kang's avatar Daniel Kang Committed by Fiona Glaser

MMX version of high bit depth plane_copy

And various cosmetics.

Patch from Google Code-In
parent 341b6147
......@@ -287,13 +287,13 @@ int x264_frame_copy_picture( x264_t *h, x264_frame_t *dst, x264_picture_t *src )
uint8_t *pix[3];
int stride[3];
get_plane_ptr( h, src, &pix[0], &stride[0], 0, 0, 0 );
h->mc.plane_copy( dst->plane[0], dst->i_stride[0], pix[0], stride[0],
h->param.i_width, h->param.i_height );
h->mc.plane_copy( dst->plane[0], dst->i_stride[0], (pixel*)pix[0],
stride[0]/sizeof(pixel), h->param.i_width, h->param.i_height );
if( i_csp == X264_CSP_NV12 )
{
get_plane_ptr( h, src, &pix[1], &stride[1], 1, 0, 1 );
h->mc.plane_copy( dst->plane[1], dst->i_stride[1], pix[1], stride[1],
h->param.i_width, h->param.i_height>>1 );
h->mc.plane_copy( dst->plane[1], dst->i_stride[1], (pixel*)pix[1],
stride[1]/sizeof(pixel), h->param.i_width, h->param.i_height>>1 );
}
else
{
......
......@@ -280,7 +280,7 @@ MC_COPY( 8 )
MC_COPY( 4 )
void x264_plane_copy_c( pixel *dst, int i_dst,
uint8_t *src, int i_src, int w, int h )
pixel *src, int i_src, int w, int h )
{
while( h-- )
{
......
......@@ -88,7 +88,7 @@ typedef struct
void (*load_deinterleave_8x8x2_fdec)( pixel *dst, pixel *src, int i_src );
void (*plane_copy)( pixel *dst, int i_dst,
uint8_t *src, int i_src, int w, int h );
pixel *src, int i_src, int w, int h );
void (*plane_copy_interleave)( pixel *dst, int i_dst,
pixel *srcu, int i_srcu,
pixel *srcv, int i_srcv, int w, int h );
......
......@@ -825,11 +825,13 @@ HPEL ssse3
%endif ; !HIGH_BIT_DEPTH
;-----------------------------------------------------------------------------
; void plane_copy_core( uint8_t *dst, int i_dst,
; uint8_t *src, int i_src, int w, int h)
; void plane_copy_core( pixel *dst, int i_dst,
; pixel *src, int i_src, int w, int h)
;-----------------------------------------------------------------------------
; assumes i_dst and w are multiples of 16, and i_dst>w
INIT_MMX
cglobal plane_copy_core_mmxext, 6,7
FIX_STRIDES r1d, r3d, r4d
movsxdifnidn r1, r1d
movsxdifnidn r3, r3d
movsxdifnidn r4, r4d
......@@ -840,22 +842,22 @@ cglobal plane_copy_core_mmxext, 6,7
sub r6d, 63
.loopx:
prefetchnta [r2+256]
movq mm0, [r2 ]
movq mm1, [r2+ 8]
movntq [r0 ], mm0
movntq [r0+ 8], mm1
movq mm2, [r2+16]
movq mm3, [r2+24]
movntq [r0+16], mm2
movntq [r0+24], mm3
movq mm4, [r2+32]
movq mm5, [r2+40]
movntq [r0+32], mm4
movntq [r0+40], mm5
movq mm6, [r2+48]
movq mm7, [r2+56]
movntq [r0+48], mm6
movntq [r0+56], mm7
movq m0, [r2 ]
movq m1, [r2+ 8]
movntq [r0 ], m0
movntq [r0+ 8], m1
movq m2, [r2+16]
movq m3, [r2+24]
movntq [r0+16], m2
movntq [r0+24], m3
movq m4, [r2+32]
movq m5, [r2+40]
movntq [r0+32], m4
movntq [r0+40], m5
movq m6, [r2+48]
movq m7, [r2+56]
movntq [r0+48], m6
movntq [r0+56], m7
add r2, 64
add r0, 64
sub r6d, 64
......@@ -864,10 +866,10 @@ cglobal plane_copy_core_mmxext, 6,7
add r6d, 63
jle .end16
.loop16:
movq mm0, [r2 ]
movq mm1, [r2+8]
movntq [r0 ], mm0
movntq [r0+8], mm1
movq m0, [r2 ]
movq m1, [r2+8]
movntq [r0 ], m0
movntq [r0+8], m1
add r2, 16
add r0, 16
sub r6d, 16
......
......@@ -84,8 +84,8 @@ void x264_mc_copy_w16_sse3( uint8_t *, int, uint8_t *, int, int );
void x264_mc_copy_w16_aligned_sse2( pixel *, int, pixel *, int, int );
void x264_prefetch_fenc_mmxext( uint8_t *, int, uint8_t *, int, int );
void x264_prefetch_ref_mmxext( uint8_t *, int, int );
void x264_plane_copy_core_mmxext( uint8_t *, int, uint8_t *, int, int w, int h);
void x264_plane_copy_c( uint8_t *, int, uint8_t *, int, int w, int h );
void x264_plane_copy_core_mmxext( pixel *, int, pixel *, int, int w, int h);
void x264_plane_copy_c( pixel *, int, pixel *, int, int w, int h );
void x264_plane_copy_interleave_core_mmxext( pixel *dst, int i_dst,
pixel *srcu, int i_srcu,
pixel *srcv, int i_srcv, int w, int h );
......@@ -426,23 +426,24 @@ HPEL(16, sse2, sse2, sse2, sse2)
HPEL(16, ssse3, ssse3, ssse3, ssse3)
#endif
HPEL(16, sse2_misalign, sse2, sse2_misalign, sse2)
#endif // HIGH_BIT_DEPTH
static void x264_plane_copy_mmxext( uint8_t *dst, int i_dst, uint8_t *src, int i_src, int w, int h )
static void x264_plane_copy_mmxext( pixel *dst, int i_dst, pixel *src, int i_src, int w, int h )
{
int c_w = 16/sizeof(pixel) - 1;
if( w < 256 ) { // tiny resolutions don't want non-temporal hints. dunno the exact threshold.
x264_plane_copy_c( dst, i_dst, src, i_src, w, h );
} else if( !(w&15) ) {
} else if( !(w&c_w) ) {
x264_plane_copy_core_mmxext( dst, i_dst, src, i_src, w, h );
} else if( i_src > 0 ) {
// have to use plain memcpy on the last line (in memory order) to avoid overreading src
x264_plane_copy_core_mmxext( dst, i_dst, src, i_src, (w+15)&~15, h-1 );
memcpy( dst+i_dst*(h-1), src+i_src*(h-1), w );
x264_plane_copy_core_mmxext( dst, i_dst, src, i_src, (w+c_w)&~c_w, h-1 );
memcpy( dst+i_dst*(h-1), src+i_src*(h-1), w*sizeof(pixel) );
} else {
memcpy( dst, src, w );
x264_plane_copy_core_mmxext( dst+i_dst, i_dst, src+i_src, i_src, (w+15)&~15, h-1 );
memcpy( dst, src, w*sizeof(pixel) );
x264_plane_copy_core_mmxext( dst+i_dst, i_dst, src+i_src, i_src, (w+c_w)&~c_w, h-1 );
}
}
#endif // HIGH_BIT_DEPTH
#define PLANE_INTERLEAVE(cpu) \
static void x264_plane_copy_interleave_##cpu( pixel *dst, int i_dst,\
......@@ -483,12 +484,13 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
pf->memzero_aligned = x264_memzero_aligned_mmx;
pf->integral_init4v = x264_integral_init4v_mmx;
pf->integral_init8v = x264_integral_init8v_mmx;
pf->store_interleave_8x8x2 = x264_store_interleave_8x8x2_mmxext;
if( !(cpu&X264_CPU_MMXEXT) )
return;
pf->plane_copy = x264_plane_copy_mmxext;
pf->plane_copy_interleave = x264_plane_copy_interleave_mmxext;
pf->store_interleave_8x8x2 = x264_store_interleave_8x8x2_mmxext;
pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_mmxext;
pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_mmxext;
......@@ -558,13 +560,6 @@ void x264_mc_init_mmx( int cpu, x264_mc_functions_t *pf )
if( (cpu&X264_CPU_SHUFFLE_IS_FAST) && !(cpu&X264_CPU_SLOW_ATOM) )
pf->integral_init4v = x264_integral_init4v_ssse3;
#else // !HIGH_BIT_DEPTH
pf->store_interleave_8x8x2 = x264_store_interleave_8x8x2_mmxext;
pf->load_deinterleave_8x8x2_fenc = x264_load_deinterleave_8x8x2_fenc_mmx;
pf->load_deinterleave_8x8x2_fdec = x264_load_deinterleave_8x8x2_fdec_mmx;
pf->plane_copy = x264_plane_copy_mmxext;
pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_mmx;
pf->frame_init_lowres_core = x264_frame_init_lowres_core_mmxext;
pf->prefetch_fenc = x264_prefetch_fenc_mmxext;
......
......@@ -1080,7 +1080,7 @@ static int check_mc( int cpu_ref, int cpu_new )
int src_stride = plane_specs[i].src_stride;
int dst_stride = (w + 127) & ~63;
assert( dst_stride * h <= 0x1000 );
uint8_t *src1 = buf1 + X264_MAX(0, -src_stride) * (h-1);
pixel *src1 = pbuf1 + X264_MAX(0, -src_stride) * (h-1);
memset( pbuf3, 0, 0x1000*sizeof(pixel) );
memset( pbuf4, 0, 0x1000*sizeof(pixel) );
call_c( mc_c.plane_copy, pbuf3, dst_stride, src1, src_stride, w, h );
......@@ -1106,7 +1106,7 @@ static int check_mc( int cpu_ref, int cpu_new )
int src_stride = (plane_specs[i].src_stride + 1) >> 1;
int dst_stride = (2*w + 127) & ~63;
assert( dst_stride * h <= 0x1000 );
uint8_t *src1 = buf1 + X264_MAX(0, -src_stride) * (h-1);
pixel *src1 = pbuf1 + X264_MAX(0, -src_stride) * (h-1);
memset( pbuf3, 0, 0x1000*sizeof(pixel) );
memset( pbuf4, 0, 0x1000*sizeof(pixel) );
call_c( mc_c.plane_copy_interleave, pbuf3, dst_stride, src1, src_stride, src1+1024, src_stride+16, w, h );
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment