1 /*****************************************************************************
2 * mc-c.c: arm motion compensation
3 *****************************************************************************
4 * Copyright (C) 2009-2015 x264 project
6 * Authors: David Conrad <lessen42@gmail.com>
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
13 * This program is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02111, USA.
22 * This program is also available under a commercial proprietary license.
23 * For more information, contact us at licensing@x264.com.
24 *****************************************************************************/
26 #include "common/common.h"
29 void x264_prefetch_ref_arm( uint8_t *, intptr_t, int );
30 void x264_prefetch_fenc_arm( uint8_t *, intptr_t, uint8_t *, intptr_t, int );
32 void *x264_memcpy_aligned_neon( void *dst, const void *src, size_t n );
33 void x264_memzero_aligned_neon( void *dst, size_t n );
35 void x264_pixel_avg_16x16_neon( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
36 void x264_pixel_avg_16x8_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
37 void x264_pixel_avg_8x16_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
38 void x264_pixel_avg_8x8_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
39 void x264_pixel_avg_8x4_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
40 void x264_pixel_avg_4x16_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
41 void x264_pixel_avg_4x8_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
42 void x264_pixel_avg_4x4_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
43 void x264_pixel_avg_4x2_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
45 void x264_pixel_avg2_w4_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, int );
46 void x264_pixel_avg2_w8_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, int );
47 void x264_pixel_avg2_w16_neon( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, int );
48 void x264_pixel_avg2_w20_neon( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, int );
50 void x264_plane_copy_neon( pixel *dst, intptr_t i_dst,
51 pixel *src, intptr_t i_src, int w, int h );
52 void x264_plane_copy_deinterleave_neon( pixel *dstu, intptr_t i_dstu,
53 pixel *dstv, intptr_t i_dstv,
54 pixel *src, intptr_t i_src, int w, int h );
55 void x264_plane_copy_deinterleave_rgb_neon( pixel *dsta, intptr_t i_dsta,
56 pixel *dstb, intptr_t i_dstb,
57 pixel *dstc, intptr_t i_dstc,
58 pixel *src, intptr_t i_src, int pw, int w, int h );
59 void x264_plane_copy_interleave_neon( pixel *dst, intptr_t i_dst,
60 pixel *srcu, intptr_t i_srcu,
61 pixel *srcv, intptr_t i_srcv, int w, int h );
62 void x264_plane_copy_swap_neon( pixel *dst, intptr_t i_dst,
63 pixel *src, intptr_t i_src, int w, int h );
65 void x264_store_interleave_chroma_neon( pixel *dst, intptr_t i_dst, pixel *srcu, pixel *srcv, int height );
66 void x264_load_deinterleave_chroma_fdec_neon( pixel *dst, pixel *src, intptr_t i_src, int height );
67 void x264_load_deinterleave_chroma_fenc_neon( pixel *dst, pixel *src, intptr_t i_src, int height );
69 #define MC_WEIGHT(func)\
70 void x264_mc_weight_w20##func##_neon( uint8_t *, intptr_t, uint8_t *, intptr_t, const x264_weight_t *, int );\
71 void x264_mc_weight_w16##func##_neon( uint8_t *, intptr_t, uint8_t *, intptr_t, const x264_weight_t *, int );\
72 void x264_mc_weight_w8##func##_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, const x264_weight_t *, int );\
73 void x264_mc_weight_w4##func##_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, const x264_weight_t *, int );\
75 static weight_fn_t x264_mc##func##_wtab_neon[6] =\
77 x264_mc_weight_w4##func##_neon,\
78 x264_mc_weight_w4##func##_neon,\
79 x264_mc_weight_w8##func##_neon,\
80 x264_mc_weight_w16##func##_neon,\
81 x264_mc_weight_w16##func##_neon,\
82 x264_mc_weight_w20##func##_neon,\
90 void x264_mc_copy_w4_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, int );
91 void x264_mc_copy_w8_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, int );
92 void x264_mc_copy_w16_neon( uint8_t *, intptr_t, uint8_t *, intptr_t, int );
93 void x264_mc_copy_w16_aligned_neon( uint8_t *, intptr_t, uint8_t *, intptr_t, int );
95 void x264_mc_chroma_neon( uint8_t *, uint8_t *, intptr_t, uint8_t *, intptr_t, int, int, int, int );
96 void x264_frame_init_lowres_core_neon( uint8_t *, uint8_t *, uint8_t *, uint8_t *, uint8_t *, intptr_t, intptr_t, int, int );
98 void x264_hpel_filter_v_neon( uint8_t *, uint8_t *, int16_t *, intptr_t, int );
99 void x264_hpel_filter_c_neon( uint8_t *, int16_t *, int );
100 void x264_hpel_filter_h_neon( uint8_t *, uint8_t *, int );
102 void integral_init4h_neon( uint16_t *, uint8_t *, intptr_t );
103 void integral_init4v_neon( uint16_t *, uint16_t *, intptr_t );
104 void integral_init8h_neon( uint16_t *, uint8_t *, intptr_t );
105 void integral_init8v_neon( uint16_t *, intptr_t );
108 static void x264_weight_cache_neon( x264_t *h, x264_weight_t *w )
110 if( w->i_scale == 1<<w->i_denom )
112 if( w->i_offset < 0 )
114 w->weightfn = x264_mc_offsetsub_wtab_neon;
115 w->cachea[0] = -w->i_offset;
119 w->weightfn = x264_mc_offsetadd_wtab_neon;
120 w->cachea[0] = w->i_offset;
123 else if( !w->i_denom )
124 w->weightfn = x264_mc_nodenom_wtab_neon;
126 w->weightfn = x264_mc_wtab_neon;
129 static void (* const x264_pixel_avg_wtab_neon[6])( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, int ) =
132 x264_pixel_avg2_w4_neon,
133 x264_pixel_avg2_w8_neon,
134 x264_pixel_avg2_w16_neon, // no slower than w12, so no point in a separate function
135 x264_pixel_avg2_w16_neon,
136 x264_pixel_avg2_w20_neon,
139 static void (* const x264_mc_copy_wtab_neon[5])( uint8_t *, intptr_t, uint8_t *, intptr_t, int ) =
142 x264_mc_copy_w4_neon,
143 x264_mc_copy_w8_neon,
145 x264_mc_copy_w16_neon,
148 static void mc_luma_neon( uint8_t *dst, intptr_t i_dst_stride,
149 uint8_t *src[4], intptr_t i_src_stride,
151 int i_width, int i_height, const x264_weight_t *weight )
153 int qpel_idx = ((mvy&3)<<2) + (mvx&3);
154 intptr_t offset = (mvy>>2)*i_src_stride + (mvx>>2);
155 uint8_t *src1 = src[x264_hpel_ref0[qpel_idx]] + offset;
156 if ( (mvy&3) == 3 ) // explict if() to force conditional add
157 src1 += i_src_stride;
159 if( qpel_idx & 5 ) /* qpel interpolation needed */
161 uint8_t *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
162 x264_pixel_avg_wtab_neon[i_width>>2](
163 dst, i_dst_stride, src1, i_src_stride,
165 if( weight->weightfn )
166 weight->weightfn[i_width>>2]( dst, i_dst_stride, dst, i_dst_stride, weight, i_height );
168 else if( weight->weightfn )
169 weight->weightfn[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, weight, i_height );
171 x264_mc_copy_wtab_neon[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, i_height );
174 static uint8_t *get_ref_neon( uint8_t *dst, intptr_t *i_dst_stride,
175 uint8_t *src[4], intptr_t i_src_stride,
177 int i_width, int i_height, const x264_weight_t *weight )
179 int qpel_idx = ((mvy&3)<<2) + (mvx&3);
180 intptr_t offset = (mvy>>2)*i_src_stride + (mvx>>2);
181 uint8_t *src1 = src[x264_hpel_ref0[qpel_idx]] + offset;
182 if ( (mvy&3) == 3 ) // explict if() to force conditional add
183 src1 += i_src_stride;
185 if( qpel_idx & 5 ) /* qpel interpolation needed */
187 uint8_t *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
188 x264_pixel_avg_wtab_neon[i_width>>2](
189 dst, *i_dst_stride, src1, i_src_stride,
191 if( weight->weightfn )
192 weight->weightfn[i_width>>2]( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_height );
195 else if( weight->weightfn )
197 weight->weightfn[i_width>>2]( dst, *i_dst_stride, src1, i_src_stride, weight, i_height );
202 *i_dst_stride = i_src_stride;
207 static void hpel_filter_neon( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,
208 intptr_t stride, int width, int height, int16_t *buf )
210 intptr_t realign = (intptr_t)src & 15;
218 x264_hpel_filter_v_neon( dstv, src, buf+8, stride, width );
219 x264_hpel_filter_c_neon( dstc, buf+8, width );
220 x264_hpel_filter_h_neon( dsth, src, width );
227 #endif // !HIGH_BIT_DEPTH
229 void x264_mc_init_arm( int cpu, x264_mc_functions_t *pf )
231 if( !(cpu&X264_CPU_ARMV6) )
235 pf->prefetch_fenc_420 = x264_prefetch_fenc_arm;
236 pf->prefetch_fenc_422 = x264_prefetch_fenc_arm; /* FIXME */
237 pf->prefetch_ref = x264_prefetch_ref_arm;
238 #endif // !HIGH_BIT_DEPTH
240 if( !(cpu&X264_CPU_NEON) )
244 pf->copy_16x16_unaligned = x264_mc_copy_w16_neon;
245 pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_neon;
246 pf->copy[PIXEL_8x8] = x264_mc_copy_w8_neon;
247 pf->copy[PIXEL_4x4] = x264_mc_copy_w4_neon;
249 pf->plane_copy = x264_plane_copy_neon;
250 pf->plane_copy_deinterleave = x264_plane_copy_deinterleave_neon;
251 pf->plane_copy_deinterleave_rgb = x264_plane_copy_deinterleave_rgb_neon;
252 pf->plane_copy_interleave = x264_plane_copy_interleave_neon;
253 pf->plane_copy_swap = x264_plane_copy_swap_neon;
255 pf->store_interleave_chroma = x264_store_interleave_chroma_neon;
256 pf->load_deinterleave_chroma_fdec = x264_load_deinterleave_chroma_fdec_neon;
257 pf->load_deinterleave_chroma_fenc = x264_load_deinterleave_chroma_fenc_neon;
259 pf->avg[PIXEL_16x16] = x264_pixel_avg_16x16_neon;
260 pf->avg[PIXEL_16x8] = x264_pixel_avg_16x8_neon;
261 pf->avg[PIXEL_8x16] = x264_pixel_avg_8x16_neon;
262 pf->avg[PIXEL_8x8] = x264_pixel_avg_8x8_neon;
263 pf->avg[PIXEL_8x4] = x264_pixel_avg_8x4_neon;
264 pf->avg[PIXEL_4x16] = x264_pixel_avg_4x16_neon;
265 pf->avg[PIXEL_4x8] = x264_pixel_avg_4x8_neon;
266 pf->avg[PIXEL_4x4] = x264_pixel_avg_4x4_neon;
267 pf->avg[PIXEL_4x2] = x264_pixel_avg_4x2_neon;
269 pf->weight = x264_mc_wtab_neon;
270 pf->offsetadd = x264_mc_offsetadd_wtab_neon;
271 pf->offsetsub = x264_mc_offsetsub_wtab_neon;
272 pf->weight_cache = x264_weight_cache_neon;
274 pf->mc_chroma = x264_mc_chroma_neon;
275 pf->mc_luma = mc_luma_neon;
276 pf->get_ref = get_ref_neon;
277 pf->hpel_filter = hpel_filter_neon;
278 pf->frame_init_lowres_core = x264_frame_init_lowres_core_neon;
280 pf->integral_init4h = integral_init4h_neon;
281 pf->integral_init8h = integral_init8h_neon;
282 pf->integral_init4v = integral_init4v_neon;
283 pf->integral_init8v = integral_init8v_neon;
284 #endif // !HIGH_BIT_DEPTH
286 // Apple's gcc stupidly cannot align stack variables, and ALIGNED_ARRAY can't work on structs
288 pf->memcpy_aligned = x264_memcpy_aligned_neon;
290 pf->memzero_aligned = x264_memzero_aligned_neon;