1 /*****************************************************************************
2 * deblock.S: h264 encoder
3 *****************************************************************************
4 * Copyright (C) 2009 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.
21 *****************************************************************************/
27 .macro h264_loop_filter_start
31 and ip, ip, ip, lsl #16
32 ands ip, ip, ip, lsl #8
36 .macro align_push_regs
40 vst1.64 {d12-d15}, [sp,:128]
42 vst1.64 {d8-d11}, [sp,:128]
46 vld1.64 {d8-d11}, [sp,:128]!
47 vld1.64 {d12-d15}, [sp,:128], ip
50 .macro h264_loop_filter_luma
51 vdup.8 q11, r2 @ alpha
53 vabd.u8 q6, q8, q0 @ abs(p0 - q0)
55 vabd.u8 q14, q9, q8 @ abs(p1 - p0)
57 vabd.u8 q15, q1, q0 @ abs(q1 - q0)
59 vclt.u8 q6, q6, q11 @ < alpha
62 vclt.u8 q14, q14, q11 @ < beta
63 vclt.u8 q15, q15, q11 @ < beta
65 vabd.u8 q4, q10, q8 @ abs(p2 - p0)
67 vabd.u8 q5, q2, q0 @ abs(q2 - q0)
68 vclt.u8 q4, q4, q11 @ < beta
70 vclt.u8 q5, q5, q11 @ < beta
77 vhadd.u8 q10, q10, q14
90 vsubw.u8 q10, q10, d17
94 vaddw.u8 q10, q10, d19
98 vrshrn.i16 d5, q10, #3
108 vaddw.s8 q14, q14, d4
110 vsubw.s8 q11, q11, d4
111 vsubw.s8 q12, q12, d5
118 function x264_deblock_v_luma_neon
119 h264_loop_filter_start
121 vld1.64 {d0, d1}, [r0,:128], r1
122 vld1.64 {d2, d3}, [r0,:128], r1
123 vld1.64 {d4, d5}, [r0,:128], r1
124 sub r0, r0, r1, lsl #2
125 sub r0, r0, r1, lsl #1
126 vld1.64 {d20,d21}, [r0,:128], r1
127 vld1.64 {d18,d19}, [r0,:128], r1
128 vld1.64 {d16,d17}, [r0,:128], r1
132 h264_loop_filter_luma
134 sub r0, r0, r1, lsl #1
135 vst1.64 {d8, d9}, [r0,:128], r1
136 vst1.64 {d16,d17}, [r0,:128], r1
137 vst1.64 {d0, d1}, [r0,:128], r1
138 vst1.64 {d10,d11}, [r0,:128]
144 function x264_deblock_h_luma_neon
145 h264_loop_filter_start
148 vld1.64 {d6}, [r0], r1
149 vld1.64 {d20}, [r0], r1
150 vld1.64 {d18}, [r0], r1
151 vld1.64 {d16}, [r0], r1
152 vld1.64 {d0}, [r0], r1
153 vld1.64 {d2}, [r0], r1
154 vld1.64 {d4}, [r0], r1
155 vld1.64 {d26}, [r0], r1
156 vld1.64 {d7}, [r0], r1
157 vld1.64 {d21}, [r0], r1
158 vld1.64 {d19}, [r0], r1
159 vld1.64 {d17}, [r0], r1
160 vld1.64 {d1}, [r0], r1
161 vld1.64 {d3}, [r0], r1
162 vld1.64 {d5}, [r0], r1
163 vld1.64 {d27}, [r0], r1
165 TRANSPOSE8x8 q3, q10, q9, q8, q0, q1, q2, q13
169 h264_loop_filter_luma
171 TRANSPOSE4x4 q4, q8, q0, q5
173 sub r0, r0, r1, lsl #4
175 vst1.32 {d8[0]}, [r0], r1
176 vst1.32 {d16[0]}, [r0], r1
177 vst1.32 {d0[0]}, [r0], r1
178 vst1.32 {d10[0]}, [r0], r1
179 vst1.32 {d8[1]}, [r0], r1
180 vst1.32 {d16[1]}, [r0], r1
181 vst1.32 {d0[1]}, [r0], r1
182 vst1.32 {d10[1]}, [r0], r1
183 vst1.32 {d9[0]}, [r0], r1
184 vst1.32 {d17[0]}, [r0], r1
185 vst1.32 {d1[0]}, [r0], r1
186 vst1.32 {d11[0]}, [r0], r1
187 vst1.32 {d9[1]}, [r0], r1
188 vst1.32 {d17[1]}, [r0], r1
189 vst1.32 {d1[1]}, [r0], r1
190 vst1.32 {d11[1]}, [r0], r1
196 .macro h264_loop_filter_chroma
197 vdup.8 d22, r2 // alpha
199 vabd.u8 d26, d16, d0 // abs(p0 - q0)
201 vabd.u8 d28, d18, d16 // abs(p1 - p0)
205 vabd.u8 d30, d2, d0 // abs(q1 - q0)
207 vclt.u8 d26, d26, d22 // < alpha
209 vdup.8 d22, r3 // beta
211 vrshrn.i16 d4, q2, #3
212 vclt.u8 d28, d28, d22 // < beta
214 vclt.u8 d30, d30, d22 // < beta
223 vaddw.s8 q14, q14, d4
224 vsubw.s8 q11, q11, d4
229 function x264_deblock_v_chroma_neon
230 h264_loop_filter_start
232 sub r0, r0, r1, lsl #1
233 vld1.64 {d18}, [r0,:64], r1
234 vld1.64 {d16}, [r0,:64], r1
235 vld1.64 {d0}, [r0,:64], r1
236 vld1.64 {d2}, [r0,:64]
238 h264_loop_filter_chroma
240 sub r0, r0, r1, lsl #1
241 vst1.64 {d16}, [r0,:64], r1
242 vst1.64 {d0}, [r0,:64], r1
247 function x264_deblock_h_chroma_neon
248 h264_loop_filter_start
251 vld1.32 {d18[]}, [r0], r1
252 vld1.32 {d16[]}, [r0], r1
253 vld1.32 {d0[]}, [r0], r1
254 vld1.32 {d2[]}, [r0], r1
255 vld1.32 {d18[1]}, [r0], r1
256 vld1.32 {d16[1]}, [r0], r1
257 vld1.32 {d0[1]}, [r0], r1
258 vld1.32 {d2[1]}, [r0], r1
265 h264_loop_filter_chroma
272 sub r0, r0, r1, lsl #3
273 vst1.32 {d18[0]}, [r0], r1
274 vst1.32 {d16[0]}, [r0], r1
275 vst1.32 {d0[0]}, [r0], r1
276 vst1.32 {d2[0]}, [r0], r1
277 vst1.32 {d18[1]}, [r0], r1
278 vst1.32 {d16[1]}, [r0], r1
279 vst1.32 {d0[1]}, [r0], r1
280 vst1.32 {d2[1]}, [r0], r1