1 @/***************************************************************************** 2 @* 3 @* Copyright (C) 2012 Ittiam Systems Pvt Ltd, Bangalore 4 @* 5 @* Licensed under the Apache License, Version 2.0 (the "License"); 6 @* you may not use this file except in compliance with the License. 7 @* You may obtain a copy of the License at: 8 @* 9 @* http://www.apache.org/licenses/LICENSE-2.0 10 @* 11 @* Unless required by applicable law or agreed to in writing, software 12 @* distributed under the License is distributed on an "AS IS" BASIS, 13 @* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 14 @* See the License for the specific language governing permissions and 15 @* limitations under the License. 16 @* 17 @*****************************************************************************/ 18 @/******************************************************************************* 19 @* @file 20 @* ihevc_deblk_luma_horz.s 21 @* 22 @* @brief 23 @* contains function definitions for inter prediction interpolation. 24 @* functions are coded using neon intrinsics and can be compiled using 25 26 @* rvct 27 @* 28 @* @author 29 @* anand s 30 @* 31 @* @par list of functions: 32 @* 33 @* 34 @* @remarks 35 @* none 36 @* 37 @*******************************************************************************/ 38 39 .text 40 .align 4 41 42 43 44 45 .extern gai4_ihevc_qp_table 46 .extern gai4_ihevc_tc_table 47 .globl ihevc_deblk_chroma_horz_a9q 48 49 gai4_ihevc_qp_table_addr: 50 .long gai4_ihevc_qp_table - ulbl1 - 8 51 52 gai4_ihevc_tc_table_addr: 53 .long gai4_ihevc_tc_table - ulbl2 - 8 54 55 .type ihevc_deblk_chroma_horz_a9q, %function 56 57 ihevc_deblk_chroma_horz_a9q: 58 push {r4-r12,lr} 59 sub r12,r0,r1 60 vld1.8 {d0},[r0] 61 sub r5,r12,r1 62 add r6,r0,r1 63 add r1,r2,r3 64 vmovl.u8 q0,d0 65 ldr r10,[sp,#0x28] 66 vld1.8 {d2},[r12] 67 add r2,r1,#1 68 ldr r4,[sp,#0x30] 69 vld1.8 {d4},[r5] 70 ldr r8,[sp,#0x34] 71 vld1.8 {d16},[r6] 72 ldr r9,[sp,#0x38] 73 adds r1,r10,r2,asr #1 74 vmovl.u8 q1,d2 75 ldr r7,[sp,#0x2c] 76 ldr r3,gai4_ihevc_qp_table_addr 77 ulbl1: 78 add r3, r3, pc 79 bmi l1.3312 80 cmp r1,#0x39 81 ldrle r1,[r3,r1,lsl #2] 82 subgt r1,r1,#6 83 l1.3312: 84 adds r2,r7,r2,asr #1 85 vmovl.u8 q2,d4 86 bmi l1.3332 87 cmp r2,#0x39 88 ldrle r2,[r3,r2,lsl #2] 89 subgt r2,r2,#6 90 l1.3332: 91 add r1,r1,r4,lsl #1 92 vsub.i16 q3,q0,q1 93 add r3,r1,#2 94 cmp r3,#0x35 95 movgt r1,#0x35 96 vshl.i16 q3,q3,#2 97 vmovl.u8 q8,d16 98 bgt l1.3368 99 adds r3,r1,#2 100 addpl r1,r1,#2 101 movmi r1,#0 102 l1.3368: 103 ldr r3,gai4_ihevc_tc_table_addr 104 ulbl2: 105 add r3, r3, pc 106 vadd.i16 q2,q3,q2 107 add r2,r2,r4,lsl #1 108 vsub.i16 q3,q2,q8 109 add r4,r2,#2 110 ldr r1,[r3,r1,lsl #2] 111 cmp r4,#0x35 112 movgt r2,#0x35 113 bgt l1.3412 114 adds r4,r2,#2 115 addpl r2,r2,#2 116 movmi r2,#0 117 l1.3412: 118 119 120 ldr r2,[r3,r2,lsl #2] 121 cmp r8,#0 122 vdup.16 q8,r2 123 vdup.16 q2,r1 124 rsb r1,r1,#0 125 vrshr.s16 q3,q3,#3 126 vdup.16 q9,r1 127 rsb r1,r2,#0 128 vzip.16 q2,q8 129 vdup.16 q10,r1 130 131 vzip.16 q9,q10 132 133 vmin.s16 q8,q3,q2 134 vmax.s16 q2,q9,q8 135 vadd.i16 q1,q1,q2 136 vsub.i16 q0,q0,q2 137 vqmovun.s16 d2,q1 138 vqmovun.s16 d0,q0 139 beq l1.3528 140 vst1.8 {d2},[r12] 141 l1.3528: 142 cmp r9,#0 143 beq l1.3540 144 vst1.8 {d0},[r0] 145 l1.3540: 146 pop {r4-r12,pc} 147 148 149
