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//*******************************************************************************
20//* ,:file
21//*  ihevc_sao_edge_offset_class3.s
22//*
23//* ,:brief
24//*  Contains function definitions for inter prediction  interpolation.
25//* Functions are coded using NEON  intrinsics and can be compiled using@ ARM
26//* RVCT
27//*
28//* ,:author
29//*  Parthiban V
30//*
31//* ,:par List of Functions:
32//*
33//*
34//* ,:remarks
35//*  None
36//*
37//*******************************************************************************
38//*/
39//void ihevc_sao_edge_offset_class3(UWORD8 *pu1_src,
40//                              WORD32 src_strd,
41//                              UWORD8 *pu1_src_left,
42//                              UWORD8 *pu1_src_top,
43//                              UWORD8 *pu1_src_top_left,
44//                              UWORD8 *pu1_src_top_right,
45//                              UWORD8 *pu1_src_bot_left,
46//                              UWORD8 *pu1_avail,
47//                              WORD8 *pi1_sao_offset,
48//                              WORD32 wd,
49//                              WORD32 ht)
50//**************Variables Vs Registers*****************************************
51//x0 =>    *pu1_src
52//x1 =>    src_strd
53//x2 =>    *pu1_src_left
54//x3 =>    *pu1_src_top
55//x4    =>    *pu1_src_top_left
56//x5    =>    *pu1_avail
57//x6    =>    *pi1_sao_offset
58//x7    =>    wd
59//x8=>    ht
60
61.text
62.p2align 2
63
64.include "ihevc_neon_macros.s"
65
66.globl gi1_table_edge_idx
67.globl ihevc_sao_edge_offset_class3_av8
68
69ihevc_sao_edge_offset_class3_av8:
70
71
72    // STMFD sp!,{x4-x12,x14}            //stack stores the values of the arguments
73    stp         x19, x20,[sp,#-16]!
74    stp         x21, x22,[sp,#-16]!
75    stp         x23, x24,[sp,#-16]!
76
77    MOV         x19,x0                      //Store pu1_src in sp
78    MOV         x21,x6                      //Store pu1_src_left in sp
79    MOV         x22,x3                      //Store pu1_src_top in sp
80    MOV         x23,x7                      //Store pu1_avail in sp
81    MOV         x24,x4                      //Store pu1_src_top_left in sp
82    MOV         x20,x5                      //Store pu1_src_top_right in sp
83    MOV         x13,x6                      //Store pu1_src_bot_left in sp
84
85    MOV         x5,x7                       //Loads pu1_avail
86
87    LDR         x6,[sp,#48]                 //Loads pi1_sao_offset
88    LDR         w7,[sp,#56]                 //Loads wd
89    LDR         w8,[sp,#64]                 //Loads ht
90
91    MOV         x16,x7 // wd
92    MOV         x17,x8 // ht
93
94    SUB         x9,x7,#1                    //wd - 1
95
96    LDRB        w10,[x3,x9]                 //pu1_src_top[wd - 1]
97
98    MOV         x9,x7                       //Move width to x9 for loop count
99
100    SUB         sp,sp,#0xA0                 //Decrement the stack pointer to store some temp arr values
101
102    STRB        w10,[sp]                    //u1_src_top_left_tmp = pu1_src_top[wd - 1]
103    SUB         x10,x8,#1                   //ht-1
104    madd        x11, x10, x1, x0            //pu1_src[(ht - 1) * src_strd + col]
105    ADD         x12,sp,#0x02                //temp array
106
107AU1_SRC_TOP_LOOP:
108    LD1         {v0.8b},[x11],#8            //pu1_src[(ht - 1) * src_strd + col]
109    SUBS        x9,x9,#8                    //Decrement the loop count by 8
110    ST1         {v0.8b},[x12],#8            //au1_src_top_tmp[col] = pu1_src[(ht - 1) * src_strd + col]
111    BNE         AU1_SRC_TOP_LOOP
112
113PU1_AVAIL_5_LOOP:
114    LDRB        w9,[x5,#5]                  //pu1_avail[5]
115    CMP         x9,#0
116    SUB         x10,x7,#1                   //[wd - 1]
117    LDRB        w9,[x0,x10]                 //u1_pos_0_0_tmp = pu1_src[wd - 1]
118    BEQ         PU1_AVAIL_6_LOOP
119
120    MOV         x11,x20                     //Load pu1_src_top_right from sp
121    SUB         x10,x10,#1                  //[wd - 1 - 1]
122
123    LDRB        w11,[x11]                   //pu1_src_top_right[0]
124    SUB         x12,x9,x11                  //pu1_src[wd - 1] - pu1_src_top_right[0]
125
126    ADD         x11,x0,x1                   //pu1_src + src_strd
127
128    LDRB        w14,[x11,x10]               //pu1_src[wd - 1 - 1 + src_strd]
129    CMP         x12,#0
130    movn        x20,#0
131    csel        x12, x20, x12,LT
132    SUB         x11,x9,x14                  //pu1_src[wd - 1] - pu1_src[wd - 1 - 1 + src_strd]
133
134    MOV         x20,#1
135    csel        x12, x20, x12,GT            //SIGN(pu1_src[wd - 1] - pu1_src_top_right[0])
136    CMP         x11,#0
137    movn        x20,#0
138    csel        x11, x20, x11,LT
139    MOV         x20,#1
140    csel        x11, x20, x11,GT            //SIGN(pu1_src[wd - 1] - pu1_src[wd - 1 - 1 + src_strd])
141    ADRP        x14, :got:gi1_table_edge_idx //table pointer
142    LDR         x14, [x14, #:got_lo12:gi1_table_edge_idx]
143    ADD         x11,x12,x11                 //SIGN(pu1_src[wd - 1] - pu1_src_top_right[0]) +  SIGN(pu1_src[wd - 1] - pu1_src[wd - 1 - 1 + src_strd])
144    ADD         x11,x11,#2                  //edge_idx
145
146    LDRSB       x12,[x14,x11]               //edge_idx = gi1_table_edge_idx[edge_idx]
147    CMP         x12,#0                      //0 != edge_idx
148    BEQ         PU1_AVAIL_6_LOOP
149    LDRSB       x10,[x6,x12]                //pi1_sao_offset[edge_idx]
150    ADD         x9,x9,x10                   //pu1_src[0] + pi1_sao_offset[edge_idx]
151    mov         x20,#255
152    cmp         x9,x20
153    csel        x9, x20, x9, ge             //u1_pos_0_0_tmp = CLIP3(pu1_src[0] + pi1_sao_offset[edge_idx], 0, (1 << bit_depth) - 1)
154
155PU1_AVAIL_6_LOOP:
156    LDRB        w10,[x5,#6]                 //pu1_avail[6]
157    SUB         x11,x8,#1                   //ht - 1
158
159    CMP         x10,#0
160    madd        x12, x11, x1, x0            //pu1_src[(ht - 1) * src_strd]
161
162    LDRB        w10,[x12]                   //u1_pos_wd_ht_tmp = pu1_src[(ht - 1) * src_strd]
163    BEQ         PU1_AVAIL_3_LOOP
164
165    MOV         x14,x13                     //Load pu1_src_bot_left from sp
166    SUB         x11,x12,x1                  //pu1_src[(ht - 1) * src_strd) - src_strd]
167
168    LDRB        w14,[x14]                   //Load pu1_src_bot_left[0]
169    ADD         x11,x11,#1                  //pu1_src[(ht - 1) * src_strd + 1 - src_strd]
170
171    LDRB        w11,[x11]                   //Load pu1_src[(ht - 1) * src_strd + 1 - src_strd]
172    SUB         x14,x10,x14                 //pu1_src[(ht - 1) * src_strd] - pu1_src_bot_left[0]
173
174    SUB         x11,x10,x11                 //pu1_src[(ht - 1) * src_strd] - pu1_src[(ht - 1) * src_strd + 1 - src_strd]
175    CMP         x11,#0
176    movn        x20,#0
177    csel        x11, x20, x11,LT
178    MOV         x20,#1
179    csel        x11, x20, x11,GT            //SIGN(pu1_src[(ht - 1) * src_strd] - pu1_src[(ht - 1) * src_strd + 1 - src_strd])
180
181    CMP         x14,#0
182    movn        x20,#0
183    csel        x14, x20, x14,LT
184    MOV         x20,#1
185    csel        x14, x20, x14,GT            //SIGN(pu1_src[(ht - 1) * src_strd] - pu1_src_bot_left[0])
186
187    ADD         x11,x11,x14                 //Add 2 sign value
188
189    ADRP        x14, :got:gi1_table_edge_idx //table pointer
190    LDR         x14, [x14, #:got_lo12:gi1_table_edge_idx]
191    ADD         x11,x11,#2                  //edge_idx
192
193    LDRSB       x12,[x14,x11]               //edge_idx = gi1_table_edge_idx[edge_idx]
194    CMP         x12,#0
195    BEQ         PU1_AVAIL_3_LOOP
196    LDRSB       x11,[x6,x12]                //pi1_sao_offset[edge_idx]
197    ADD         x10,x10,x11                 //pu1_src[(ht - 1) * src_strd] + pi1_sao_offset[edge_idx]
198    mov         x20,#255
199    cmp         x10,x20
200    csel        x10, x20, x10, ge           //u1_pos_wd_ht_tmp = CLIP3(pu1_src[(ht - 1) * src_strd] + pi1_sao_offset[edge_idx], 0, (1 << bit_depth) - 1)
201
202PU1_AVAIL_3_LOOP:
203    MOV         x21,x2
204    MOV         x12,x8                      //Move ht
205
206    MOV         x14,x2                      //Move pu1_src_left to pu1_src_left_cpy
207    movi        v0.16b, #2                  //const_2 = vdupq_n_s8(2)
208    LDRB        w11,[x5,#3]                 //pu1_avail[3]
209
210    CMP         x11,#0
211    movi        v2.8h, #0                   //const_min_clip = vdupq_n_s16(0)
212    SUB         x20,x12,#1                  //ht_tmp--
213    csel        x12, x20, x12,EQ
214
215    LDRB        w5,[x5,#2]                  //pu1_avail[2]
216    movi        v4.8h, #255                 //const_max_clip = vdupq_n_u16((1 << bit_depth) - 1)
217    CMP         x5,#0
218
219    ADD         x20,x0,x1                   //pu1_src += src_strd
220    csel        x0, x20, x0,EQ
221    LD1         {v7.8b},[x6]                //offset_tbl = vld1_s8(pi1_sao_offset)
222    SUB         x20,x12,#1                  //ht_tmp--
223    csel        x12, x20, x12,EQ
224
225    ADRP        x6, :got:gi1_table_edge_idx //table pointer
226    LDR         x6, [x6, #:got_lo12:gi1_table_edge_idx]
227
228    movi        v1.16b, #0xFF               //au1_mask = vdupq_n_s8(-1)
229    ADD         x20,x14,#1                  //pu1_src_left_cpy += 1
230    csel        x14, x20, x14,EQ
231
232    MOV         x15,x0                      //Store pu1_src in sp
233    LD1         {v6.8b},[x6]                //edge_idx_tbl = vld1_s8(gi1_table_edge_idx)
234    MOV         x6,x7                       //move wd to x6 loop_count
235
236    CMP         x7,#16                      //Compare wd with 16
237    BLT         WIDTH_RESIDUE               //If not jump to WIDTH_RESIDUE where loop is unrolled for 8 case
238    CMP         x8,#4                       //Compare ht with 4
239    BLE         WD_16_HT_4_LOOP             //If jump to WD_16_HT_4_LOOP
240
241WIDTH_LOOP_16:
242    MOV         x7,x16                      //Loads wd
243
244    MOV         x5,x23                      //Loads pu1_avail
245    CMP         x6,x7                       //col == wd
246    LDRb        w20, [x5]                   //pu1_avail[0]
247    csel        w8,w20,w8,EQ
248    MOV         x20,#-1
249    csel        x8, x20, x8,NE
250    mov         v1.b[0], w8                 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0)
251
252    CMP         x6,#16                      //if(col == 16)
253    BNE         SKIP_AU1_MASK_VAL
254    LDRB        w8,[x5,#1]                  //pu1_avail[1]
255    mov         v1.b[15], w8                //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15)
256
257SKIP_AU1_MASK_VAL:
258    LDRB        w8,[x5,#2]                  //pu1_avail[2]
259    CMP         x8,#0
260
261    MOV         x4,x17                      //Loads ht
262    SUB         x20,x0,x1                   //pu1_src - src_strd
263    csel        x8, x20, x8,EQ
264
265    csel        x8, x3, x8,NE
266    ADD         x5,sp,#0x42                 //*au1_src_left_tmp
267
268    MOV         x7,x16                      //Loads wd
269    ADD         x8,x8,#1                    //pu1_src - src_strd + 1
270
271    SUB         x7,x7,x6                    //(wd - col)
272    LD1         {v3.16b},[x8]               //pu1_top_row = vld1q_u8(pu1_src - src_strd + 1)
273    ADD         x3,x3,#16
274
275    MOV         x8,x19                      //Loads *pu1_src
276    LD1         {v5.16b},[x0]               //pu1_cur_row = vld1q_u8(pu1_src)
277    ADD         x7,x7,#15                   //15 + (wd - col)
278
279    ADD         x7,x8,x7                    //pu1_src[0 * src_strd + 15 + (wd - col)]
280    cmhi        v17.16b,  v5.16b ,  v3.16b  //vcgtq_u8(pu1_cur_row, pu1_top_row)
281    SUB         x5,x5,#1
282
283AU1_SRC_LEFT_LOOP:
284    LDRB        w8,[x7]                     //load the value and increment by src_strd
285    ADD         x7,x7,x1
286    SUBS        x4,x4,#1                    //decrement the loop count
287    STRB        w8,[x5,#1]!                 //store it in the stack pointer
288    BNE         AU1_SRC_LEFT_LOOP
289
290    movi        v18.16b, #0
291    cmhi        v16.16b,  v3.16b ,  v5.16b  //vcltq_u8(pu1_cur_row, pu1_top_row)
292
293    ADD         x8,x0,x1                    //I *pu1_src + src_strd
294    SUB         v17.16b,  v16.16b ,  v17.16b //sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
295    MOV         x7,x12                      //row count, move ht_tmp to x7
296
297    SUB         x5,x12,x7                   //I ht_tmp - row
298    LD1         {v16.16b},[x8]              //I pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd)
299    ADD         x8,x14,x5                   //I pu1_src_left_cpy[ht_tmp - row]
300
301    ADD         x8,x8,#1                    //I pu1_src_left_cpy[ht_tmp - row + 1]
302    LDRB        w8,[x8]
303
304    MOV         x5,x23                      //I Loads pu1_avail
305    mov         v18.b[15], w8               //I vsetq_lane_u8
306    LDRB        w5,[x5,#2]                  //I pu1_avail[2]
307
308    EXT         v18.16b,  v18.16b ,  v16.16b,#15 //I pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15)
309    CMP         x5,#0                       //I
310    BNE         SIGN_UP_CHANGE_DONE         //I
311
312SIGN_UP_CHANGE:
313    LDRB        w8,[x0,#15]                 //I pu1_src_cpy[15]
314    SUB         x5,x0,x1                    //I pu1_src_cpy[16 - src_strd]
315
316    LDRB        w5,[x5,#16]                 //I load the value
317    SUB         x8,x8,x5                    //I pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]
318    CMP         x8,#0                       //I
319    movn        x20,#0
320    csel        x8, x20, x8,LT              //I
321    MOV         x20,#1
322    csel        x8, x20, x8,GT              //I SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd])
323    mov         v17.b[15], w8               //I sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15)
324
325SIGN_UP_CHANGE_DONE:
326    cmhi        v3.16b,  v5.16b ,  v18.16b  //I vcgtq_u8(pu1_cur_row, pu1_next_row_tmp)
327    cmhi        v18.16b,  v18.16b ,  v5.16b //I vcltq_u8(pu1_cur_row, pu1_next_row_tmp)
328    SUB         v3.16b,  v18.16b ,  v3.16b  //I sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
329
330    ADD         v18.16b,  v0.16b ,  v17.16b //I edge_idx = vaddq_s8(const_2, sign_up)
331    ADD         v18.16b,  v18.16b ,  v3.16b //I edge_idx = vaddq_s8(edge_idx, sign_down)
332    TBL         v18.16b, {v6.16b},v18.16b   //I vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx))
333    NEG         v17.16b, v3.16b             //I sign_up = vnegq_s8(sign_down)
334
335    EXT         v17.16b,  v17.16b ,  v17.16b,#1 //I sign_up = vextq_s8(sign_up, sign_up, 1)
336//  TBL v19.8b, {v6.16b},v19.8b                //I vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx))
337
338    Uxtl        v20.8h, v5.8b               //I pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row)))
339    AND         v18.16b,  v18.16b ,  v1.16b //I edge_idx = vandq_s8(edge_idx, au1_mask)
340
341    TBL         v3.16b, {v7.16b},v18.16b    //I offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx))
342
343    Uxtl2       v22.8h, v5.16b              //I pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row)))
344    SADDW       v20.8h,  v20.8h ,  v3.8b    //I pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset)
345
346    SMAX        v20.8h,  v20.8h ,  v2.8h    //I pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip)
347//  TBL v11.8b, {v7.16b},v19.8b                    //I offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx))
348    UMIN        v20.8h,  v20.8h ,  v4.8h    //I pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip))
349
350    mov         v5.16b, v16.16b
351    SADDW2      v22.8h,  v22.8h ,  v3.16b   //I pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset)
352
353    SMAX        v22.8h,  v22.8h ,  v2.8h    //I pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip)
354    UMIN        v22.8h,  v22.8h ,  v4.8h    //I pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip))
355
356    SUB         x7,x7,#1                    //I Decrement the ht_tmp loop count by 1
357
358PU1_SRC_LOOP:
359    ADD         x8,x0,x1,LSL #1             //II *pu1_src + src_strd
360    xtn         v20.8b,  v20.8h             //I vmovn_s16(pi2_tmp_cur_row.val[0])
361    SUB         x5,x12,x7                   //II ht_tmp - row
362
363    ADD         x4,x0,x1                    //II pu1_src_cpy[16 - src_strd]
364    xtn2        v20.16b,  v22.8h            //I vmovn_s16(pi2_tmp_cur_row.val[1])
365    ADD         x2,x8,x1                    //III *pu1_src + src_strd
366
367    LDRB        w11,[x4,#15]                //II pu1_src_cpy[15]
368    LD1         {v16.16b},[x8]              //II pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd)
369    SUB         x7,x7,#1                    //II Decrement the ht_tmp loop count by 1
370
371    ADD         x8,x14,x5                   //II pu1_src_left_cpy[ht_tmp - row]
372    LD1         {v30.16b},[x2]              //III pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd)
373    LDRB        w8,[x8,#1]
374
375    LDRB        w4,[x0,#16]                 //II load the value
376    mov         v18.b[15], w8               //II vsetq_lane_u8
377    SUB         x11,x11,x4                  //II pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]
378
379    CMP         x11,#0                      //II
380    ST1         { v20.16b},[x0],x1          //I vst1q_u8(pu1_src_cpy, pu1_cur_row)
381    SUB         x5,x12,x7                   //III ht_tmp - row
382
383    movn        x20,#0
384    csel        x11, x20, x11,LT            //II
385    EXT         v18.16b,  v18.16b ,  v16.16b,#15 //II pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15)
386    MOV         x20,#1
387    csel        x11, x20, x11,GT            //II SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd])
388
389    ADD         x8,x14,x5                   //III pu1_src_left_cpy[ht_tmp - row]
390    mov         v17.b[15], w11              //II sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15)
391    CMP         x7,#1                       //III
392
393    BNE         NEXT_ROW_ELSE_2             //III
394    MOV         x5,x23                      //III Loads pu1_avail
395    LDRB        w5,[x5,#3]                  //III pu1_avail[3]
396    CMP         x5,#0                       //III
397    SUB         x20,x2,#2                   //III pu1_src_cpy[src_strd - 1]
398    csel        x8, x20, x8,NE
399
400NEXT_ROW_ELSE_2:
401    LDRB        w8,[x8,#1]                  //III
402    cmhi        v24.16b,  v5.16b ,  v18.16b //II vcgtq_u8(pu1_cur_row, pu1_next_row_tmp)
403    ADD         x5,x0,x1
404
405    LDRB        w2,[x5,#15]                 //III pu1_src_cpy[15]
406    cmhi        v26.16b,  v18.16b ,  v5.16b //II vcltq_u8(pu1_cur_row, pu1_next_row_tmp)
407    LDRB        w5,[x0,#16]                 //III load the value
408
409    SUB         x2,x2,x5                    //III pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]
410    SUB         v24.16b,  v26.16b ,  v24.16b //II sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
411    CMP         x2,#0                       //III
412
413    movn        x20,#0
414    csel        x2, x20, x2,LT              //III
415    mov         v18.b[15], w8               //III vsetq_lane_u8
416    MOV         x20,#1
417    csel        x2, x20, x2,GT              //III SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd])
418
419    SUB         x7,x7,#1                    //III Decrement the ht_tmp loop count by 1
420    ADD         v26.16b,  v0.16b ,  v17.16b //II edge_idx = vaddq_s8(const_2, sign_up)
421
422    NEG         v17.16b, v24.16b            //II sign_up = vnegq_s8(sign_down)
423    EXT         v18.16b,  v18.16b ,  v30.16b,#15 //III pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15)
424
425    ADD         v26.16b,  v26.16b ,  v24.16b //II edge_idx = vaddq_s8(edge_idx, sign_down)
426
427    EXT         v17.16b,  v17.16b ,  v17.16b,#1 //II sign_up = vextq_s8(sign_up, sign_up, 1)
428    TBL         v26.16b, {v6.16b},v26.16b   //II vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx))
429    cmhi        v3.16b,  v16.16b ,  v18.16b //III vcgtq_u8(pu1_cur_row, pu1_next_row_tmp)
430
431    mov         v17.b[15], w2               //III sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15)
432//  TBL v27.8b, {v6.16b},v27.8b                //II vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx))
433    cmhi        v18.16b,  v18.16b ,  v16.16b //III vcltq_u8(pu1_cur_row, pu1_next_row_tmp)
434
435    Uxtl        v28.8h, v5.8b               //II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row)))
436    AND         v26.16b,  v26.16b ,  v1.16b //II edge_idx = vandq_s8(edge_idx, au1_mask)
437
438    SUB         v3.16b,  v18.16b ,  v3.16b  //III sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
439    TBL         v24.16b, {v7.16b},v26.16b   //II offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx))
440    ADD         v18.16b,  v0.16b ,  v17.16b //III edge_idx = vaddq_s8(const_2, sign_up)
441
442    ADD         v18.16b,  v18.16b ,  v3.16b //III edge_idx = vaddq_s8(edge_idx, sign_down)
443//  TBL v25.8b, {v7.16b},v27.8b                    //II offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx))
444    NEG         v17.16b, v3.16b             //III sign_up = vnegq_s8(sign_down)
445
446    SADDW       v28.8h,  v28.8h ,  v24.8b   //II pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset)
447    TBL         v18.16b, {v6.16b},v18.16b   //III vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx))
448    SMAX        v28.8h,  v28.8h ,  v2.8h    //II pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip)
449
450    EXT         v17.16b,  v17.16b ,  v17.16b,#1 //III sign_up = vextq_s8(sign_up, sign_up, 1)
451//  TBL v19.8b, {v6.16b},v19.8b                //III vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx))
452    UMIN        v28.8h,  v28.8h ,  v4.8h    //II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip))
453
454    Uxtl2       v26.8h, v5.16b              //II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row)))
455    AND         v18.16b,  v18.16b ,  v1.16b //III edge_idx = vandq_s8(edge_idx, au1_mask)
456
457    SADDW2      v26.8h,  v26.8h ,  v24.16b  //II pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset)
458    TBL         v3.16b, {v7.16b},v18.16b    //III offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx))
459    SMAX        v26.8h,  v26.8h ,  v2.8h    //II pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip)
460
461    Uxtl        v20.8h, v16.8b              //III pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row)))
462    UMIN        v26.8h,  v26.8h ,  v4.8h    //II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip))
463
464    SADDW       v20.8h,  v20.8h ,  v3.8b    //III pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset)
465//  TBL v11.8b, {v7.16b},v19.8b                    //III offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx))
466    SMAX        v20.8h,  v20.8h ,  v2.8h    //III pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip)
467
468    Uxtl2       v22.8h, v16.16b             //III pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row)))
469    UMIN        v20.8h,  v20.8h ,  v4.8h    //III pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip))
470
471    xtn         v28.8b,  v28.8h             //II vmovn_s16(pi2_tmp_cur_row.val[0])
472    SADDW2      v22.8h,  v22.8h ,  v3.16b   //III pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset)
473
474    xtn2        v28.16b,  v26.8h            //II vmovn_s16(pi2_tmp_cur_row.val[1])
475    SMAX        v22.8h,  v22.8h ,  v2.8h    //III pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip)
476
477    mov         v5.16b, v30.16b             //II pu1_cur_row = pu1_next_row
478    UMIN        v22.8h,  v22.8h ,  v4.8h    //III pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip))
479
480    CMP         x7,#1                       //III
481    ST1         { v28.16b},[x0],x1          //II vst1q_u8(pu1_src_cpy, pu1_cur_row)
482    BGT         PU1_SRC_LOOP                //If not equal jump to PU1_SRC_LOOP
483    BLT         INNER_LOOP_DONE
484
485    ADD         x8,x0,x1,LSL #1             //*pu1_src + src_strd
486    xtn         v20.8b,  v20.8h             //III vmovn_s16(pi2_tmp_cur_row.val[0])
487    MOV         x5,x23                      //Loads pu1_avail
488
489    LDRB        w5,[x5,#3]                  //pu1_avail[3]
490    xtn2        v20.16b,  v22.8h            //III vmovn_s16(pi2_tmp_cur_row.val[1])
491    CMP         x5,#0
492
493    ADD         x4,x0,x1                    //pu1_src_cpy[16 - src_strd]
494    LD1         {v16.16b},[x8]              //pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd)
495    LDRB        w5,[x0,#16]                 //load the value
496
497    BEQ         NEXT_ROW_ELSE_3
498    SUB         x8,x8,#1
499    LDRB        w8,[x8]                     //pu1_src_cpy[src_strd - 1]
500    B           NEXT_ROW_POINTER_ASSIGNED_3
501NEXT_ROW_ELSE_3:
502    SUB         x11,x12,x7                  //ht_tmp - row
503    ADD         x8,x14,x11                  //pu1_src_left_cpy[ht_tmp - row]
504    ADD         x8,x8,#1                    //pu1_src_left_cpy[ht_tmp - row + 1]
505    LDRB        w8,[x8]
506
507NEXT_ROW_POINTER_ASSIGNED_3:
508    LDRB        w11,[x4,#15]                //pu1_src_cpy[15]
509    mov         v18.b[15], w8               //vsetq_lane_u8
510    SUB         x8,x11,x5                   //pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]
511
512    CMP         x8,#0
513    EXT         v18.16b,  v18.16b ,  v16.16b,#15 //pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15)
514    movn        x20,#0
515    csel        x8, x20, x8,LT
516
517    ST1         { v20.16b},[x0],x1          //III vst1q_u8(pu1_src_cpy, pu1_cur_row)
518    cmhi        v24.16b,  v5.16b ,  v18.16b //vcgtq_u8(pu1_cur_row, pu1_next_row_tmp)
519
520    MOV         x20,#1
521    csel        x8, x20, x8,GT              //SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd])
522    cmhi        v26.16b,  v18.16b ,  v5.16b //vcltq_u8(pu1_cur_row, pu1_next_row_tmp)
523
524    mov         v17.b[15], w8               //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15)
525    SUB         v24.16b,  v26.16b ,  v24.16b //sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
526
527    Uxtl        v20.8h, v5.8b               //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row)))
528    ADD         v26.16b,  v0.16b ,  v17.16b //edge_idx = vaddq_s8(const_2, sign_up)
529
530    Uxtl2       v22.8h, v5.16b              //pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row)))
531    ADD         v26.16b,  v26.16b ,  v24.16b //edge_idx = vaddq_s8(edge_idx, sign_down)
532
533    TBL         v26.16b, {v6.16b},v26.16b   //vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx))
534//  TBL v27.8b, {v6.16b},v27.8b                //vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx))
535
536    AND         v26.16b,  v26.16b ,  v1.16b //edge_idx = vandq_s8(edge_idx, au1_mask)
537
538    TBL         v24.16b, {v7.16b},v26.16b   //offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx))
539
540    SADDW       v20.8h,  v20.8h ,  v24.8b   //pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset)
541//  TBL v25.8b, {v7.16b},v27.8b                    //offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx))
542    SMAX        v20.8h,  v20.8h ,  v2.8h    //pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip)
543
544    UMIN        v20.8h,  v20.8h ,  v4.8h    //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip))
545
546    SADDW2      v22.8h,  v22.8h ,  v24.16b  //pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset)
547    SMAX        v22.8h,  v22.8h ,  v2.8h    //pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip)
548    UMIN        v22.8h,  v22.8h ,  v4.8h    //pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip))
549
550INNER_LOOP_DONE:
551    xtn         v20.8b,  v20.8h             //vmovn_s16(pi2_tmp_cur_row.val[0])
552    MOV         x8,x17                      //Loads ht
553
554    xtn2        v20.16b,  v22.8h            //vmovn_s16(pi2_tmp_cur_row.val[1])
555    ADD         x5,sp,#0x42                 //*au1_src_left_tmp
556
557    ST1         { v20.16b},[x0],x1          //vst1q_u8(pu1_src_cpy, pu1_cur_row)
558    MOV         x2,x21                      //Loads *pu1_src_left
559SRC_LEFT_LOOP:
560    LDR         w7,[x5],#4                  //au1_src_left_tmp[row]
561    SUBS        x8,x8,#4
562    STR         w7,[x2],#4                  //pu1_src_left[row] = au1_src_left_tmp[row]
563    BNE         SRC_LEFT_LOOP
564
565    SUBS        x6,x6,#16                   //Decrement the wd loop count by 16
566    CMP         x6,#8                       //Check whether residue remains
567    BLT         RE_ASSINING_LOOP            //Jump to re-assigning loop
568    MOV         x7,x16                      //Loads wd
569    MOV         x0,x15                      //Loads *pu1_src
570    SUB         x7,x7,x6
571    ADD         x0,x0,x7
572    BGT         WIDTH_LOOP_16               //If not equal jump to width_loop
573    BEQ         WIDTH_RESIDUE               //If residue remains jump to residue loop
574
575
576
577WD_16_HT_4_LOOP:
578    MOV         x5,x23                      //Loads pu1_avail
579    MOV         x7,x16                      //Loads wd
580    CMP         x6,x7                       //col == wd
581    LDRb        w20, [x5]                   //pu1_avail[0]
582    csel        w8,w20,w8,EQ
583    MOV         x20,#-1
584    csel        x8, x20, x8,NE
585    mov         v1.b[0], w8                 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0)
586
587    CMP         x6,#16                      //if(col == 16)
588    BNE         SKIP_AU1_MASK_VAL_WD_16_HT_4
589    LDRB        w8,[x5,#1]                  //pu1_avail[1]
590    mov         v1.b[15], w8                //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15)
591
592SKIP_AU1_MASK_VAL_WD_16_HT_4:
593    LDRB        w8,[x5,#2]                  //pu1_avail[2]
594    CMP         x8,#0
595
596    SUB         x20,x0,x1                   //pu1_src - src_strd
597    csel        x8, x20, x8,EQ
598    csel        x8, x3, x8,NE
599    ADD         x8,x8,#1                    //pu1_src - src_strd + 1
600    LD1         {v3.16b},[x8]               //pu1_top_row = vld1q_u8(pu1_src - src_strd + 1)
601
602    ADD         x3,x3,#16
603    ADD         x5,sp,#0x42                 //*au1_src_left_tmp
604    MOV         x4,x17                      //Loads ht
605    MOV         x7,x16                      //Loads wd
606    SUB         x7,x7,x6                    //(wd - col)
607    ADD         x7,x7,#15                   //15 + (wd - col)
608    MOV         x8,x19                      //Loads *pu1_src
609    ADD         x7,x8,x7                    //pu1_src[0 * src_strd + 15 + (wd - col)]
610    SUB         x5,x5,#1
611
612AU1_SRC_LEFT_LOOP_WD_16_HT_4:
613    LDRB        w8,[x7]                     //load the value and increment by src_strd
614    ADD         x7,x7,x1
615    STRB        w8,[x5,#1]!                 //store it in the stack pointer
616    SUBS        x4,x4,#1                    //decrement the loop count
617    BNE         AU1_SRC_LEFT_LOOP_WD_16_HT_4
618
619    LD1         {v5.16b},[x0]               //pu1_cur_row = vld1q_u8(pu1_src)
620
621    cmhi        v17.16b,  v5.16b ,  v3.16b  //vcgtq_u8(pu1_cur_row, pu1_top_row)
622    cmhi        v16.16b,  v3.16b ,  v5.16b  //vcltq_u8(pu1_cur_row, pu1_top_row)
623    SUB         v17.16b,  v16.16b ,  v17.16b //sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
624    movi        v18.16b, #0
625    MOV         x7,x12                      //row count, move ht_tmp to x7
626
627PU1_SRC_LOOP_WD_16_HT_4:
628    ADD         x8,x0,x1                    //*pu1_src + src_strd
629    LD1         {v16.16b},[x8]              //pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd)
630    MOV         x5,x23                      //Loads pu1_avail
631    LDRB        w5,[x5,#3]                  //pu1_avail[3]
632    CMP         x5,#0
633    BEQ         NEXT_ROW_ELSE_WD_16_HT_4
634    CMP         x7,#1
635    SUB         x8,x8,#1
636    LDRb        w20, [x8]                   //pu1_src_cpy[src_strd - 1]
637    csel        w8,w20,w8,EQ
638    BEQ         NEXT_ROW_POINTER_ASSIGNED_WD_16_HT_4
639NEXT_ROW_ELSE_WD_16_HT_4:
640    SUB         x5,x12,x7                   //ht_tmp - row
641    ADD         x8,x14,x5                   //pu1_src_left_cpy[ht_tmp - row]
642    ADD         x8,x8,#1                    //pu1_src_left_cpy[ht_tmp - row + 1]
643    LDRB        w8,[x8]
644
645NEXT_ROW_POINTER_ASSIGNED_WD_16_HT_4:
646    mov         v18.b[15], w8               //vsetq_lane_u8
647    EXT         v18.16b,  v18.16b ,  v16.16b,#15 //pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15)
648
649    CMP         x7,x12
650    BNE         SIGN_UP_CHANGE_WD_16_HT_4
651    MOV         x5,x23                      //Loads pu1_avail
652    LDRB        w5,[x5,#2]                  //pu1_avail[2]
653    CMP         x5,#0
654    BNE         SIGN_UP_CHANGE_DONE_WD_16_HT_4
655
656SIGN_UP_CHANGE_WD_16_HT_4:
657    LDRB        w8,[x0,#15]                 //pu1_src_cpy[15]
658    ADD         x5,x0,#16                   //pu1_src_cpy[16]
659    SUB         x5,x5,x1                    //pu1_src_cpy[16 - src_strd]
660    LDRB        w5,[x5]                     //load the value
661    SUB         x8,x8,x5                    //pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]
662    CMP         x8,#0
663    movn        x20,#0
664    csel        x8, x20, x8,LT
665    MOV         x20,#1
666    csel        x8, x20, x8,GT              //SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd])
667    mov         v17.b[15], w8               //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15)
668
669SIGN_UP_CHANGE_DONE_WD_16_HT_4:
670    cmhi        v20.16b,  v5.16b ,  v18.16b //vcgtq_u8(pu1_cur_row, pu1_next_row_tmp)
671    cmhi        v22.16b,  v18.16b ,  v5.16b //vcltq_u8(pu1_cur_row, pu1_next_row_tmp)
672    SUB         v24.16b,  v22.16b ,  v20.16b //sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
673
674    ADD         v26.16b,  v0.16b ,  v17.16b //edge_idx = vaddq_s8(const_2, sign_up)
675    ADD         v26.16b,  v26.16b ,  v24.16b //edge_idx = vaddq_s8(edge_idx, sign_down)
676    TBL         v26.16b, {v6.16b},v26.16b   //vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx))
677//  TBL v27.8b, {v6.16b},v27.8b                //vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx))
678
679    AND         v26.16b,  v26.16b ,  v1.16b //edge_idx = vandq_s8(edge_idx, au1_mask)
680
681    NEG         v17.16b, v24.16b            //sign_up = vnegq_s8(sign_down)
682    EXT         v17.16b,  v17.16b ,  v17.16b,#1 //sign_up = vextq_s8(sign_up, sign_up, 1)
683
684    TBL         v24.16b, {v7.16b},v26.16b   //offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx))
685    Uxtl        v28.8h, v5.8b               //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row)))
686    SADDW       v28.8h,  v28.8h ,  v24.8b   //pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset)
687    SMAX        v28.8h,  v28.8h ,  v2.8h    //pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip)
688    UMIN        v28.8h,  v28.8h ,  v4.8h    //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip))
689
690//  TBL v25.8b, {v7.16b},v27.8b                    //offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx))
691    Uxtl2       v30.8h, v5.16b              //pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row)))
692    SADDW2      v30.8h,  v30.8h ,  v24.16b  //pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset)
693    SMAX        v30.8h,  v30.8h ,  v2.8h    //pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip)
694    UMIN        v30.8h,  v30.8h ,  v4.8h    //pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip))
695
696    xtn         v28.8b,  v28.8h             //vmovn_s16(pi2_tmp_cur_row.val[0])
697    xtn2        v28.16b,  v30.8h            //vmovn_s16(pi2_tmp_cur_row.val[1])
698
699    ST1         { v28.16b},[x0],x1          //vst1q_u8(pu1_src_cpy, pu1_cur_row)
700
701    mov         v5.16b, v16.16b             //pu1_cur_row = pu1_next_row
702    SUBS        x7,x7,#1                    //Decrement the ht_tmp loop count by 1
703    BNE         PU1_SRC_LOOP_WD_16_HT_4     //If not equal jump to PU1_SRC_LOOP_WD_16_HT_4
704
705    MOV         x8,x17                      //Loads ht
706    ADD         x5,sp,#0x42                 //*au1_src_left_tmp
707    MOV         x2,x21                      //Loads *pu1_src_left
708SRC_LEFT_LOOP_WD_16_HT_4:
709    LDR         w7,[x5],#4                  //au1_src_left_tmp[row]
710    STR         w7,[x2],#4                  //pu1_src_left[row] = au1_src_left_tmp[row]
711    SUBS        x8,x8,#4
712    BNE         SRC_LEFT_LOOP_WD_16_HT_4
713
714    SUBS        x6,x6,#16                   //Decrement the wd loop count by 16
715    BLE         RE_ASSINING_LOOP            //Jump to re-assigning loop
716    BGT         WD_16_HT_4_LOOP             //If not equal jump to width_loop
717
718
719WIDTH_RESIDUE:
720    MOV         x7,x16                      //Loads wd
721    MOV         x5,x23                      //Loads pu1_avail
722    CMP         x6,x7                       //wd_residue == wd
723    LDRb        w20, [x5]                   //pu1_avail[0]
724    csel        w8,w20,w8,EQ
725
726    MOV         x20,#-1
727    csel        x8, x20, x8,NE
728    mov         v1.b[0], w8                 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0)
729
730    LDRB        w8,[x5,#1]                  //pu1_avail[1]
731    mov         v1.b[7], w8                 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15)
732
733PU1_AVAIL_2_RESIDUE:
734    LDRB        w8,[x5,#2]                  //pu1_avail[2]
735    CMP         x8,#0
736
737    SUB         x20,x0,x1                   //pu1_src - src_strd
738    csel        x8, x20, x8,EQ
739    csel        x8, x3, x8,NE
740    ADD         x8,x8,#1                    //pu1_src - src_strd + 1
741    LD1         {v3.16b},[x8]               //pu1_top_row = vld1q_u8(pu1_src - src_strd + 1)
742
743
744    ADD         x5,sp,#0x42                 //*au1_src_left_tmp
745    MOV         x4,x17                      //Loads ht
746    MOV         x7,x16                      //Loads wd
747    MOV         x8,x19                      //Loads *pu1_src
748    SUB         x7,x7,#1                    //(wd - 1)
749    ADD         x7,x8,x7                    //pu1_src[0 * src_strd + (wd - 1)]
750    SUB         x5,x5,#1
751
752AU1_SRC_LEFT_LOOP_RESIDUE:
753    LDRB        w8,[x7]                     //load the value and increment by src_strd
754    ADD         x7,x7,x1
755    STRB        w8,[x5,#1]!                 //store it in the stack pointer
756    SUBS        x4,x4,#1                    //decrement the loop count
757    BNE         AU1_SRC_LEFT_LOOP_RESIDUE
758
759    LD1         {v5.16b},[x0]               //pu1_cur_row = vld1q_u8(pu1_src)
760
761    cmhi        v17.16b,  v5.16b ,  v3.16b  //vcgtq_u8(pu1_cur_row, pu1_top_row)
762    cmhi        v16.16b,  v3.16b ,  v5.16b  //vcltq_u8(pu1_cur_row, pu1_top_row)
763    SUB         v17.16b,  v16.16b ,  v17.16b //sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
764    MOV         x7,x12                      //row count, move ht_tmp to x7
765
766PU1_SRC_LOOP_RESIDUE:
767    movi        v18.16b, #0
768    ADD         x8,x0,x1                    //*pu1_src + src_strd
769    LD1         {v16.16b},[x8]              //pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd)
770    MOV         x5,x23                      //Loads pu1_avail
771    LDRB        w5,[x5,#3]                  //pu1_avail[3]
772    CMP         x5,#0
773    BEQ         NEXT_ROW_ELSE_RESIDUE
774    CMP         x7,#1
775    SUB         x8,x8,#1
776    LDRb        w20, [x8]                   //pu1_src_cpy[src_strd - 1]
777    csel        w8,w20,w8,EQ
778    BEQ         NEXT_ROW_POINTER_ASSIGNED_RESIDUE
779NEXT_ROW_ELSE_RESIDUE:
780    SUB         x5,x12,x7                   //ht_tmp - row
781    ADD         x8,x14,x5                   //pu1_src_left_cpy[ht_tmp - row]
782    ADD         x8,x8,#1                    //pu1_src_left_cpy[ht_tmp - row + 1]
783    LDRB        w8,[x8]
784
785NEXT_ROW_POINTER_ASSIGNED_RESIDUE:
786    mov         v18.b[15], w8               //vsetq_lane_u8
787    EXT         v18.16b,  v18.16b ,  v16.16b,#15 //pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15)
788
789    CMP         x7,x12
790    BNE         SIGN_UP_CHANGE_RESIDUE
791    MOV         x5,x23                      //Loads pu1_avail
792    LDRB        w5,[x5,#2]                  //pu1_avail[2]
793    CMP         x5,#0
794    BNE         SIGN_UP_CHANGE_DONE_RESIDUE
795
796SIGN_UP_CHANGE_RESIDUE:
797    LDRB        w8,[x0,#15]                 //pu1_src_cpy[15]
798    ADD         x5,x0,#16                   //pu1_src_cpy[16]
799    SUB         x5,x5,x1                    //pu1_src_cpy[16 - src_strd]
800    LDRB        w5,[x5]                     //load the value
801    SUB         x8,x8,x5                    //pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]
802    CMP         x8,#0
803    movn        x20,#0
804    csel        x8, x20, x8,LT
805    MOV         x20,#1
806    csel        x8, x20, x8,GT              //SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd])
807    mov         v17.b[15], w8               //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15)
808
809SIGN_UP_CHANGE_DONE_RESIDUE:
810    cmhi        v20.16b,  v5.16b ,  v18.16b //vcgtq_u8(pu1_cur_row, pu1_next_row_tmp)
811    cmhi        v22.16b,  v18.16b ,  v5.16b //vcltq_u8(pu1_cur_row, pu1_next_row_tmp)
812    SUB         v24.16b,  v22.16b ,  v20.16b //sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
813
814    ADD         v26.16b,  v0.16b ,  v17.16b //edge_idx = vaddq_s8(const_2, sign_up)
815    ADD         v26.16b,  v26.16b ,  v24.16b //edge_idx = vaddq_s8(edge_idx, sign_down)
816    TBL         v26.16b, {v6.16b},v26.16b   //vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx))
817//  TBL v27.8b, {v6.16b},v27.8b                //vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx))
818
819    AND         v26.16b,  v26.16b ,  v1.16b //edge_idx = vandq_s8(edge_idx, au1_mask)
820
821    NEG         v17.16b, v24.16b            //sign_up = vnegq_s8(sign_down)
822    EXT         v17.16b,  v17.16b ,  v17.16b,#1 //sign_up = vextq_s8(sign_up, sign_up, 1)
823
824    TBL         v24.8b, {v7.16b},v26.8b     //offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx))
825    Uxtl        v28.8h, v5.8b               //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row)))
826    SADDW       v28.8h,  v28.8h ,  v24.8b   //pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset)
827    SMAX        v28.8h,  v28.8h ,  v2.8h    //pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip)
828    UMIN        v28.8h,  v28.8h ,  v4.8h    //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip))
829
830    xtn         v30.8b,  v28.8h             //vmovn_s16(pi2_tmp_cur_row.val[0])
831
832    ST1         {v30.8b},[x0],x1            //vst1q_u8(pu1_src_cpy, pu1_cur_row)
833    mov         v5.16b, v16.16b             //pu1_cur_row = pu1_next_row
834    SUBS        x7,x7,#1
835    BNE         PU1_SRC_LOOP_RESIDUE
836
837    MOV         x8,x17                      //Loads ht
838    MOV         x2,x21                      //Loads *pu1_src_left
839    ADD         x5,sp,#0x42                 //*au1_src_left_tmp
840
841SRC_LEFT_LOOP_RESIDUE:
842    LDR         w7,[x5],#4                  //au1_src_left_tmp[row]
843    SUBS        x8,x8,#4
844    STR         w7,[x2],#4                  //pu1_src_left[row] = au1_src_left_tmp[row]
845    BNE         SRC_LEFT_LOOP_RESIDUE
846
847
848RE_ASSINING_LOOP:
849    MOV         x7,x16                      //Loads wd
850    MOV         x0,x19                      //Loads *pu1_src
851
852    MOV         x11,x17                     //Loads ht
853    ADD         x8,x0,x7                    //pu1_src[wd]
854
855    MOV         x4,x24                      //Loads pu1_src_top_left
856    SUB         x11,x11,#1                  //ht - 1
857
858    SUB         x8,x8,#1
859    STRB        w9,[x8]                     //pu1_src_org[wd - 1] = u1_pos_wd_0_tmp
860    ADD         x8,x8,#1
861    madd        x6, x11, x1, x0             //pu1_src_org[(ht - 1) * src_strd]
862
863    LDRB        w8,[sp]                     //load u1_src_top_left_tmp from stack pointer
864    ADD         x12,sp,#0x02
865
866    STRB        w10,[x6]                    //pu1_src_org[wd - 1 + (ht - 1) * src_strd] = u1_pos_wd_ht_tmp
867    STRB        w8,[x4]                     //*pu1_src_top_left = u1_src_top_left_tmp
868    MOV         x3,x22                      //Loads pu1_src_top
869
870SRC_TOP_LOOP:
871    LD1         {v0.8b},[x12],#8            //pu1_src_top[col] = au1_src_top_tmp[col]
872    SUBS        x7,x7,#8                    //Decrement the width
873    ST1         {v0.8b},[x3],#8             //pu1_src_top[col] = au1_src_top_tmp[col]
874    BNE         SRC_TOP_LOOP
875
876END_LOOPS:
877    ADD         sp,sp,#0xA0
878    // LDMFD sp!,{x4-x12,x15}             //Reload the registers from SP
879    ldp         x23, x24,[sp], #16
880    ldp         x21, x22,[sp], #16
881    ldp         x19, x20,[sp], #16
882    ret
883
884
885
886