1/* 2 * Copyright (c) 2017 Meng Wang <wangmeng.kids@bytedance.com> 3 * 4 * This file is part of FFmpeg. 5 * 6 * FFmpeg is free software; you can redistribute it and/or 7 * modify it under the terms of the GNU Lesser General Public 8 * License as published by the Free Software Foundation; either 9 * version 2.1 of the License, or (at your option) any later version. 10 * 11 * FFmpeg is distributed in the hope that it will be useful, 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of 13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 14 * Lesser General Public License for more details. 15 * 16 * You should have received a copy of the GNU Lesser General Public 17 * License along with FFmpeg; if not, write to the Free Software 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 19 */ 20 21 22#include "libavutil/arm/asm.S" 23#include "neon.S" 24 25function ff_hevc_sao_band_filter_neon_8, export=1 26 push {r4-r10} 27 ldr r5, [sp, #28] // width 28 ldr r4, [sp, #32] // height 29 ldr r8, [sp, #36] // offset_table 30 vpush {d8-d15} 31 mov r12, r4 // r12 = height 32 mov r6, r0 // r6 = r0 = dst 33 mov r7, r1 // r7 = r1 = src 34 vldm r8, {q0-q3} 35 vmov.u16 q15, #1 36 vmov.u8 q14, #32 370: pld [r1] 38 cmp r5, #4 39 beq 4f 408: subs r4, #1 41 vld1.8 {d16}, [r1], r3 42 vshr.u8 d17, d16, #3 // index = [src>>3] 43 vshll.u8 q9, d17, #1 // lowIndex = 2*index 44 vadd.u16 q11, q9, q15 // highIndex = (2*index+1) << 8 45 vshl.u16 q10, q11, #8 // q10: highIndex; q9: lowIndex; 46 vadd.u16 q10, q9 // combine high and low index; 47 // Look-up Table Round 1; index range: 0-15 48 vtbx.8 d24, {q0-q1}, d20 49 vtbx.8 d25, {q0-q1}, d21 50 // Look-up Table Round 2; index range: 16-31 51 vsub.u8 q10, q14 // Look-up with 8bit 52 vtbx.8 d24, {q2-q3}, d20 53 vtbx.8 d25, {q2-q3}, d21 54 vaddw.u8 q13, q12, d16 55 vqmovun.s16 d8, q13 56 vst1.8 d8, [r0], r2 57 bne 8b 58 subs r5, #8 59 beq 99f 60 mov r4, r12 61 add r6, #8 62 mov r0, r6 63 add r7, #8 64 mov r1, r7 65 b 0b 664: subs r4, #1 67 vld1.32 {d16[0]}, [r1], r3 68 vshr.u8 d17, d16, #3 // src>>3 69 vshll.u8 q9, d17, #1 // lowIndex = 2*index 70 vadd.u16 q11, q9, q15 // highIndex = (2*index+1) << 8 71 vshl.u16 q10, q11, #8 // q10: highIndex; q9: lowIndex; 72 vadd.u16 q10, q9 // combine high and low index; 73 // Look-up Table Round 1; index range: 0-15 74 vtbx.8 d24, {q0-q1}, d20 75 vtbx.8 d25, {q0-q1}, d21 76 // Look-up Table Round 2; index range: 16-32 77 vsub.u8 q10, q14 // Look-up with 8bit 78 vtbx.8 d24, {q2-q3}, d20 79 vtbx.8 d25, {q2-q3}, d21 80 vaddw.u8 q13, q12, d16 81 vqmovun.s16 d14, q13 82 vst1.32 d14[0], [r0], r2 83 bne 4b 84 b 99f 8599: 86 vpop {d8-d15} 87 pop {r4-r10} 88 bx lr 89endfunc 90 91function ff_hevc_sao_edge_filter_neon_8, export=1 92 push {r4-r11} 93 ldr r5, [sp, #32] // width 94 ldr r4, [sp, #36] // height 95 ldr r8, [sp, #40] // a_stride 96 ldr r9, [sp, #44] // b_stride 97 ldr r10, [sp, #48] // sao_offset_val 98 ldr r11, [sp, #52] // edge_idx 99 vpush {d8-d15} 100 mov r12, r4 // r12 = height 101 mov r6, r0 // r6 = r0 = dst 102 mov r7, r1 // r7 = r1 = src 103 vld1.8 {d0}, [r11] // edge_idx tabel load in d0 5x8bit 104 vld1.16 {q1}, [r10] // sao_offset_val table load in q1, 5x16bit 105 vmov.u8 d1, #2 106 vmov.u16 q2, #1 1070: mov r10, r1 108 add r10, r8 // src[x + a_stride] 109 mov r11, r1 110 add r11, r9 // src[x + b_stride] 111 pld [r1] 112 cmp r5, #4 113 beq 4f 1148: subs r4, #1 115 vld1.8 {d16}, [r1], r3 // src[x] 8x8bit 116 vld1.8 {d17}, [r10], r3 // src[x + a_stride] 117 vld1.8 {d18}, [r11], r3 // src[x + b_stride] 118 vcgt.u8 d8, d16, d17 119 vshr.u8 d9, d8, #7 120 vclt.u8 d8, d16, d17 121 vadd.u8 d8, d9 // diff0 122 vcgt.u8 d10, d16, d18 123 vshr.u8 d11, d10, #7 124 vclt.u8 d10, d16, d18 125 vadd.u8 d10, d11 // diff1 126 vadd.s8 d8, d10 127 vadd.s8 d8, d1 128 vtbx.8 d9, {d0}, d8 // offset_val 129 vshll.u8 q6, d9, #1 // lowIndex 130 vadd.u16 q7, q6, q2 131 vshl.u16 q10, q7, #8 // highIndex 132 vadd.u16 q10, q6 // combine lowIndex and highIndex, offset_val 133 vtbx.8 d22, {q1}, d20 134 vtbx.8 d23, {q1}, d21 135 vaddw.u8 q12, q11, d16 136 vqmovun.s16 d26, q12 137 vst1.8 d26, [r0], r2 138 bne 8b 139 subs r5, #8 140 beq 99f 141 mov r4, r12 142 add r6, #8 143 mov r0, r6 144 add r7, #8 145 mov r1, r7 146 b 0b 1474: subs r4, #1 148 vld1.32 {d16[0]}, [r1], r3 149 vld1.32 {d17[0]}, [r10], r3 // src[x + a_stride] 150 vld1.32 {d18[0]}, [r11], r3 // src[x + b_stride] 151 vcgt.u8 d8, d16, d17 152 vshr.u8 d9, d8, #7 153 vclt.u8 d8, d16, d17 154 vadd.u8 d8, d9 // diff0 155 vcgt.u8 d10, d16, d18 156 vshr.u8 d11, d10, #7 157 vclt.u8 d10, d16, d18 158 vadd.u8 d10, d11 // diff1 159 vadd.s8 d8, d10 160 vadd.s8 d8, d1 161 vtbx.8 d9, {d0}, d8 // offset_val 162 vshll.u8 q6, d9, #1 // lowIndex 163 vadd.u16 q7, q6, q2 164 vshl.u16 q10, q7, #8 // highIndex 165 vadd.u16 q10, q6 // combine lowIndex and highIndex, offset_val 166 vtbx.8 d22, {q1}, d20 167 vtbx.8 d23, {q1}, d21 168 vaddw.u8 q12, q11, d16 169 vqmovun.s16 d26, q12 170 vst1.32 d26[0], [r0], r2 171 bne 4b 172 b 99f 17399: 174 vpop {d8-d15} 175 pop {r4-r11} 176 bx lr 177endfunc 178