1 /*****************************************************************************
2 * deblock.S: arm deblocking
3 *****************************************************************************
4 * Copyright (C) 2009-2014 x264 project
6 * Authors: Mans Rullgard <mans@mansr.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 *****************************************************************************/
28 .macro h264_loop_filter_start
32 and ip, ip, ip, lsl #16
33 ands ip, ip, ip, lsl #8
37 .macro align_push_regs
41 vst1.64 {d12-d15}, [sp,:128]
43 vst1.64 {d8-d11}, [sp,:128]
47 vld1.64 {d8-d11}, [sp,:128]!
48 vld1.64 {d12-d15}, [sp,:128], ip
51 .macro h264_loop_filter_luma
52 vdup.8 q11, r2 @ alpha
54 vabd.u8 q6, q8, q0 @ abs(p0 - q0)
56 vabd.u8 q14, q9, q8 @ abs(p1 - p0)
58 vabd.u8 q15, q1, q0 @ abs(q1 - q0)
60 vclt.u8 q6, q6, q11 @ < alpha
63 vclt.u8 q14, q14, q11 @ < beta
64 vclt.u8 q15, q15, q11 @ < beta
66 vabd.u8 q4, q10, q8 @ abs(p2 - p0)
68 vabd.u8 q5, q2, q0 @ abs(q2 - q0)
69 vclt.u8 q4, q4, q11 @ < beta
71 vclt.u8 q5, q5, q11 @ < beta
78 vhadd.u8 q10, q10, q14
91 vsubw.u8 q10, q10, d17
95 vaddw.u8 q10, q10, d19
99 vrshrn.i16 d5, q10, #3
109 vaddw.s8 q14, q14, d4
111 vsubw.s8 q11, q11, d4
112 vsubw.s8 q12, q12, d5
119 function x264_deblock_v_luma_neon
120 h264_loop_filter_start
122 vld1.64 {d0, d1}, [r0,:128], r1
123 vld1.64 {d2, d3}, [r0,:128], r1
124 vld1.64 {d4, d5}, [r0,:128], r1
125 sub r0, r0, r1, lsl #2
126 sub r0, r0, r1, lsl #1
127 vld1.64 {d20,d21}, [r0,:128], r1
128 vld1.64 {d18,d19}, [r0,:128], r1
129 vld1.64 {d16,d17}, [r0,:128], r1
133 h264_loop_filter_luma
135 sub r0, r0, r1, lsl #1
136 vst1.64 {d8, d9}, [r0,:128], r1
137 vst1.64 {d16,d17}, [r0,:128], r1
138 vst1.64 {d0, d1}, [r0,:128], r1
139 vst1.64 {d10,d11}, [r0,:128]
145 function x264_deblock_h_luma_neon
146 h264_loop_filter_start
149 vld1.64 {d6}, [r0], r1
150 vld1.64 {d20}, [r0], r1
151 vld1.64 {d18}, [r0], r1
152 vld1.64 {d16}, [r0], r1
153 vld1.64 {d0}, [r0], r1
154 vld1.64 {d2}, [r0], r1
155 vld1.64 {d4}, [r0], r1
156 vld1.64 {d26}, [r0], r1
157 vld1.64 {d7}, [r0], r1
158 vld1.64 {d21}, [r0], r1
159 vld1.64 {d19}, [r0], r1
160 vld1.64 {d17}, [r0], r1
161 vld1.64 {d1}, [r0], r1
162 vld1.64 {d3}, [r0], r1
163 vld1.64 {d5}, [r0], r1
164 vld1.64 {d27}, [r0], r1
166 TRANSPOSE8x8 q3, q10, q9, q8, q0, q1, q2, q13
170 h264_loop_filter_luma
172 TRANSPOSE4x4 q4, q8, q0, q5
174 sub r0, r0, r1, lsl #4
176 vst1.32 {d8[0]}, [r0], r1
177 vst1.32 {d16[0]}, [r0], r1
178 vst1.32 {d0[0]}, [r0], r1
179 vst1.32 {d10[0]}, [r0], r1
180 vst1.32 {d8[1]}, [r0], r1
181 vst1.32 {d16[1]}, [r0], r1
182 vst1.32 {d0[1]}, [r0], r1
183 vst1.32 {d10[1]}, [r0], r1
184 vst1.32 {d9[0]}, [r0], r1
185 vst1.32 {d17[0]}, [r0], r1
186 vst1.32 {d1[0]}, [r0], r1
187 vst1.32 {d11[0]}, [r0], r1
188 vst1.32 {d9[1]}, [r0], r1
189 vst1.32 {d17[1]}, [r0], r1
190 vst1.32 {d1[1]}, [r0], r1
191 vst1.32 {d11[1]}, [r0], r1
197 .macro h264_loop_filter_chroma
198 vdup.8 q11, r2 // alpha
200 vabd.u8 q13, q8, q0 // abs(p0 - q0)
201 vabd.u8 q14, q9, q8 // abs(p1 - p0)
207 vabd.u8 q15, q1, q0 // abs(q1 - q0)
210 vclt.u8 q13, q13, q11 // < alpha
213 vdup.8 q11, r3 // beta
215 vrshrn.i16 d4, q2, #3
216 vrshrn.i16 d5, q3, #3
217 vclt.u8 q14, q14, q11 // < beta
219 vclt.u8 q15, q15, q11 // < beta
230 vaddw.s8 q14, q14, d4
231 vaddw.s8 q15, q15, d5
232 vsubw.s8 q11, q11, d4
233 vsubw.s8 q12, q12, d5
240 function x264_deblock_v_chroma_neon
241 h264_loop_filter_start
243 sub r0, r0, r1, lsl #1
244 vld2.8 {d18,d19}, [r0,:128], r1
245 vld2.8 {d16,d17}, [r0,:128], r1
246 vld2.8 {d0, d1}, [r0,:128], r1
247 vld2.8 {d2, d3}, [r0,:128]
249 h264_loop_filter_chroma
251 sub r0, r0, r1, lsl #1
252 vst2.8 {d16,d17}, [r0,:128], r1
253 vst2.8 {d0, d1}, [r0,:128], r1
258 function x264_deblock_h_chroma_neon
259 h264_loop_filter_start
262 vld1.8 {d18}, [r0], r1
263 vld1.8 {d16}, [r0], r1
264 vld1.8 {d0}, [r0], r1
265 vld1.8 {d2}, [r0], r1
266 vld1.8 {d19}, [r0], r1
267 vld1.8 {d17}, [r0], r1
268 vld1.8 {d1}, [r0], r1
269 vld1.8 {d3}, [r0], r1
281 h264_loop_filter_chroma
293 sub r0, r0, r1, lsl #3
294 vst1.8 {d18}, [r0], r1
295 vst1.8 {d16}, [r0], r1
296 vst1.8 {d0}, [r0], r1
297 vst1.8 {d2}, [r0], r1
298 vst1.8 {d19}, [r0], r1
299 vst1.8 {d17}, [r0], r1
300 vst1.8 {d1}, [r0], r1
301 vst1.8 {d3}, [r0], r1
306 function x264_deblock_strength_neon
311 sub ip, ip, #(1<<8)-3
323 vext.8 q3, q0, q1, #15
324 vext.8 q0, q0, q2, #15
327 vext.8 q1, q15, q2, #12
334 vld1.16 {q11}, [r2,:128]! @ mv + 0x10
335 vld1.16 {q3}, [r2,:128]! @ mv + 0x20
336 vld1.16 {q12}, [r2,:128]! @ mv + 0x30
337 vld1.16 {q2}, [r2,:128]! @ mv + 0x40
338 vld1.16 {q13}, [r2,:128]! @ mv + 0x50
339 vext.8 q3, q3, q12, #12
340 vext.8 q2, q2, q13, #12
342 vld1.16 {q3}, [r2,:128]! @ mv + 0x60
344 vld1.16 {q14}, [r2,:128]! @ mv + 0x70
346 vld1.16 {q2}, [r2,:128]! @ mv + 0x80
347 vld1.16 {q15}, [r2,:128]! @ mv + 0x90
349 vext.8 q3, q3, q14, #12
350 vext.8 q2, q2, q15, #12
361 vabd.s16 q1, q12, q13
364 vabd.s16 q0, q11, q12
365 vabd.s16 q2, q13, q14
366 vabd.s16 q3, q14, q15
386 vext.8 q3, q0, q1, #15
387 vext.8 q0, q0, q2, #15
390 vext.8 q1, q15, q2, #12
397 vmin.u8 q8, q8, q10 @ mv ? 1 : 0
399 vadd.u8 q0, q0, q0 @ nnz ? 2 : 0
404 vst1.8 {q9}, [r3,:128], ip @ bs[1]
408 vst1.8 {q8}, [r3,:128] @ bs[0]