1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
|
///*****************************************************************************
//*
//* Copyright (C) 2012 Ittiam Systems Pvt Ltd, Bangalore
//*
//* Licensed under the Apache License, Version 2.0 (the "License");
//* you may not use this file except in compliance with the License.
//* You may obtain a copy of the License at:
//*
//* http://www.apache.org/licenses/LICENSE-2.0
//*
//* Unless required by applicable law or agreed to in writing, software
//* distributed under the License is distributed on an "AS IS" BASIS,
//* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
//* See the License for the specific language governing permissions and
//* limitations under the License.
//*
//*****************************************************************************/
///**
//*******************************************************************************
//* //file
//* ihevc_inter_pred_chroma_vert_neon.s
//*
//* //brief
//* contains function definitions for inter prediction interpolation.
//* functions are coded using neon intrinsics and can be compiled using
//* rvct
//*
//* //author
//* yogeswaran rs
//*
//* //par list of functions:
//*
//*
//* //remarks
//* none
//*
//*******************************************************************************
//*/
///**
///**
//*******************************************************************************
//*
//* //brief
//* chroma interprediction filter for vertical input
//*
//* //par description:
//* applies a vertical filter with coefficients pointed to by 'pi1_coeff' to
//* the elements pointed by 'pu1_src' and writes to the location pointed by
//* 'pu1_dst' the output is down shifted by 6 and clipped to 8 bits
//* assumptions : the function is optimized considering the fact width is
//* multiple of 2,4 or 8. and also considering height should be multiple of 2
//* width 4,8 is optimized further
//*
//* //param[in] pu1_src
//* uword8 pointer to the source
//*
//* //param[out] pu1_dst
//* uword8 pointer to the destination
//*
//* //param[in] src_strd
//* integer source stride
//*
//* //param[in] dst_strd
//* integer destination stride
//*
//* //param[in] pi1_coeff
//* word8 pointer to the filter coefficients
//*
//* //param[in] ht
//* integer height of the array
//*
//* //param[in] wd
//* integer width of the array
//*
//* //returns
//*
//* //remarks
//* none
//*
//*******************************************************************************
//*/
//void ihevc_inter_pred_chroma_vert(uword8 *pu1_src,
// uword8 *pu1_dst,
// word32 src_strd,
// word32 dst_strd,
// word8 *pi1_coeff,
// word32 ht,
// word32 wd)
//**************variables vs registers*****************************************
//x0 => *pu1_src
//x1 => *pi2_dst
//x2 => src_strd
//x3 => dst_strd
.text
.align 4
.include "ihevc_neon_macros.s"
.globl ihevc_inter_pred_chroma_vert_av8
.type ihevc_inter_pred_chroma_vert_av8, %function
ihevc_inter_pred_chroma_vert_av8:
// stmfd sp!,{x4-x12,x14} //stack stores the values of the arguments
push_v_regs
stp x19, x20,[sp,#-16]!
mov x15,x4 // pi1_coeff
mov x16,x5 // ht
mov x17,x6 // wd
mov x4,x16 //loads ht
mov x12,x15 //loads pi1_coeff
cmp x4,#0 //checks ht == 0
mov x6,x17 //loads wd
sub x0,x0,x2 //pu1_src - src_strd
ld1 {v0.8b},[x12] //loads pi1_coeff
ble end_loops //jumps to end
tst x6,#3 //checks (wd & 3)
abs v3.8b, v0.8b //vabs_s8(coeff)
lsl x10,x6,#1 //2*wd
dup v0.8b, v3.8b[0] //coeffabs_0
dup v1.8b, v3.8b[1] //coeffabs_1
dup v2.8b, v3.8b[2] //coeffabs_2
dup v3.8b, v3.8b[3] //coeffabs_3
bgt outer_loop_wd_2 //jumps to loop handling wd ==2
tst x4,#7 //checks ht for mul of 8
beq core_loop_ht_8 //when height is multiple of 8
lsl x7,x3,#1 //2*dst_strd
sub x9,x7,x10 //2*dst_strd - 2wd
lsl x12,x2,#1 //2*src_strd
sub x8,x12,x10 //2*src_strd - 2wd
mov x5,x10 //2wd
inner_loop_ht_2: //called when wd is multiple of 4 and ht is 4,2
add x6,x0,x2 //pu1_src +src_strd
ld1 {v9.8b},[x6],x2 //loads pu1_src
subs x5,x5,#8 //2wd - 8
ld1 {v5.8b},[x0],#8 //loads src
umull v6.8h, v9.8b, v1.8b //vmull_u8(vreinterpret_u8_u32(src_tmp2), coeffabs_1)
ld1 {v4.8b},[x6],x2 //loads incremented src
umlsl v6.8h, v5.8b, v0.8b //vmlsl_u8(mul_res1, vreinterpret_u8_u32(src_tmp1), coeffabs_0)
ld1 {v8.8b},[x6],x2 //loads incremented src
umlal v6.8h, v4.8b, v2.8b //vmlal_u8(mul_res1, vreinterpret_u8_u32(src_tmp3), coeffabs_2)
umull v4.8h, v4.8b, v1.8b
umlsl v6.8h, v8.8b, v3.8b
umlsl v4.8h, v9.8b, v0.8b
ld1 {v10.8b},[x6] //loads the incremented src
umlal v4.8h, v8.8b, v2.8b
sqrshrun v6.8b, v6.8h,#6 //shifts right
umlsl v4.8h, v10.8b, v3.8b
add x6,x1,x3 //pu1_dst + dst_strd
sqrshrun v4.8b, v4.8h,#6 //shifts right
st1 {v6.8b},[x1],#8 //stores the loaded value
st1 {v4.8b},[x6] //stores the loaded value
bgt inner_loop_ht_2 //inner loop again
subs x4,x4,#2 //ht - 2
add x1,x1,x9 //pu1_dst += (2*dst_strd - 2wd)
mov x5,x10 //2wd
add x0,x0,x8 //pu1_src += (2*src_strd - 2wd)
bgt inner_loop_ht_2 //loop again
b end_loops //jumps to end
outer_loop_wd_2: //called when width is multiple of 2
lsl x5,x3,#1 //2*dst_strd
mov x12,x10 //2wd
sub x9,x5,x10 //2*dst_strd - 2wd
lsl x7,x2,#1 //2*src_strd
sub x8,x7,x10 //2*src_strd - 2wd
inner_loop_wd_2:
add x6,x0,x2 //pu1_src + src_strd
ld1 {v6.s}[0],[x0] //vld1_lane_u32((uint32_t *)pu1_src_tmp, src_tmp1, 0
subs x12,x12,#4 //2wd - 4
add x0,x0,#4 //pu1_src + 4
ld1 {v6.s}[1],[x6],x2 //loads pu1_src_tmp
dup v7.2s, v6.2s[1]
ld1 {v7.s}[1],[x6],x2 //loads pu1_src_tmp
umull v4.8h, v7.8b, v1.8b //vmull_u8(vreinterpret_u8_u32(src_tmp2), coeffabs_1)
dup v7.2s, v7.2s[1]
ld1 {v7.s}[1],[x6],x2
umlsl v4.8h, v6.8b, v0.8b
umlal v4.8h, v7.8b, v2.8b
dup v7.2s, v7.2s[1]
ld1 {v7.s}[1],[x6]
add x6,x1,x3 //pu1_dst + dst_strd
umlsl v4.8h, v7.8b, v3.8b
sqrshrun v4.8b, v4.8h,#6 //vrshrq_n_s16(vreinterpretq_s16_u16(mul_res1),6)
st1 {v4.s}[0],[x1] //stores the loaded value
add x1,x1,#4 //pu1_dst += 4
st1 {v4.s}[1],[x6] //stores the loaded value
bgt inner_loop_wd_2 //inner loop again
//inner loop ends
subs x4,x4,#2 //ht - 2
add x1,x1,x9 //pu1_dst += 2*dst_strd - 2*wd
mov x12,x10 //2wd
add x0,x0,x8 //pu1_src += 2*src_strd - 2*wd
bgt inner_loop_wd_2 //loop again
b end_loops //jumps to end
core_loop_ht_8: //when wd & ht is multiple of 8
lsl x12,x3,#2 //4*dst_strd
sub x8,x12,x10 //4*dst_strd - 2wd
lsl x12,x2,#2 //4*src_strd
sub x9,x12,x10 //4*src_strd - 2wd
bic x5,x10,#7 //x5 ->wd
lsr x14, x10, #3 //divide by 8
mul x12, x4 , x14 //multiply height by width
sub x12, x12,#4 //subtract by one for epilog
prolog:
add x6,x0,x2 //pu1_src + src_strd
ld1 {v5.8b},[x6],x2 //loads pu1_src
subs x5,x5,#8 //2wd - 8
ld1 {v4.8b},[x0],#8 //loads the source
ld1 {v6.8b},[x6],x2 //load and increment
umull v30.8h, v5.8b, v1.8b //mul with coeff 1
ld1 {v7.8b},[x6],x2 //load and increment
umlsl v30.8h, v4.8b, v0.8b
add x7,x1,x3 //pu1_dst
umlal v30.8h, v6.8b, v2.8b
umlsl v30.8h, v7.8b, v3.8b
ld1 {v8.8b},[x6],x2 //load and increment
umull v28.8h, v6.8b, v1.8b //mul_res 2
add x20,x0,x9 //pu1_dst += 4*dst_strd - 2*wd
csel x0, x20, x0,le
umlsl v28.8h, v5.8b, v0.8b
bic x20,x10,#7 //x5 ->wd
csel x5, x20, x5,le
umlal v28.8h, v7.8b, v2.8b
ld1 {v9.8b},[x6],x2
umlsl v28.8h, v8.8b, v3.8b
sqrshrun v30.8b, v30.8h,#6
ld1 {v10.8b},[x6],x2
umull v26.8h, v7.8b, v1.8b
add x6,x0,x2 //pu1_src + src_strd
umlsl v26.8h, v6.8b, v0.8b
st1 {v30.8b},[x1],#8 //stores the loaded value
umlal v26.8h, v8.8b, v2.8b
ld1 {v4.8b},[x0],#8 //loads the source
umlsl v26.8h, v9.8b, v3.8b
sqrshrun v28.8b, v28.8h,#6
add x20,x1,x8 //pu1_src += 4*src_strd - 2*wd
csel x1, x20, x1,le
umull v24.8h, v8.8b, v1.8b
ld1 {v5.8b},[x6],x2 //loads pu1_src
umlsl v24.8h, v7.8b, v0.8b
subs x12,x12,#4
ld1 {v6.8b},[x6],x2 //load and increment
umlal v24.8h, v9.8b, v2.8b
ld1 {v7.8b},[x6],x2 //load and increment
umlsl v24.8h, v10.8b, v3.8b
lsl x11,x2,#2
st1 {v28.8b},[x7],x3 //stores the loaded value
sqrshrun v26.8b, v26.8h,#6
sub x20,x2,x2,lsl #3
neg x11, x20
add x14,x2,x2,lsl #1
add x14,x14,x11
ble epilog //jumps to epilog
kernel_8:
umull v30.8h, v5.8b, v1.8b //mul with coeff 1
subs x5,x5,#8 //2wd - 8
umlsl v30.8h, v4.8b, v0.8b
add x20,x0,x9 //pu1_dst += 4*dst_strd - 2*wd
csel x0, x20, x0,le
umlal v30.8h, v6.8b, v2.8b
lsl x20,x2,#3
sub x20,x20,x2
csel x11,x20,x11,le
//rsble x11,x2,x2,lsl #3
umlsl v30.8h, v7.8b, v3.8b
st1 {v26.8b},[x7],x3 //stores the loaded value
sqrshrun v24.8b, v24.8h,#6
ld1 {v8.8b},[x6],x2 //load and increment
umull v28.8h, v6.8b, v1.8b //mul_res 2
bic x20,x10,#7 //x5 ->wd
csel x5, x20, x5,le
umlsl v28.8h, v5.8b, v0.8b
st1 {v24.8b},[x7],x3 //stores the loaded value
umlal v28.8h, v7.8b, v2.8b
ld1 {v9.8b},[x6],x2
sqrshrun v30.8b, v30.8h,#6
umlsl v28.8h, v8.8b, v3.8b
ld1 {v10.8b},[x6],x2
add x7,x1,x3 //pu1_dst
umull v26.8h, v7.8b, v1.8b
add x6,x0,x2 //pu1_src + src_strd
add x20,x0, x11
prfm PLDL1KEEP,[x20]
umlsl v26.8h, v6.8b, v0.8b
ld1 {v4.8b},[x0],#8 //loads the source
umlal v26.8h, v8.8b, v2.8b
st1 {v30.8b},[x1],#8 //stores the loaded value
umlsl v26.8h, v9.8b, v3.8b
ld1 {v5.8b},[x6],x2 //loads pu1_src
add x11,x11,x2
sqrshrun v28.8b, v28.8h,#6
umull v24.8h, v8.8b, v1.8b
ld1 {v6.8b},[x6],x2 //load and increment
add x20,x1,x8 //pu1_src += 4*src_strd - 2*wd
csel x1, x20, x1,le
cmp x11,x14
lsl x20,x2,#3
sub x20,x20,x2
csel x11,x20,x11,gt
//rsbgt x11,x2,x2,lsl #3
umlsl v24.8h, v7.8b, v0.8b
subs x12,x12,#4
umlal v24.8h, v9.8b, v2.8b
ld1 {v7.8b},[x6],x2 //load and increment
umlsl v24.8h, v10.8b, v3.8b
st1 {v28.8b},[x7],x3 //stores the loaded value
sqrshrun v26.8b, v26.8h,#6
bgt kernel_8 //jumps to kernel_8
epilog:
umull v30.8h, v5.8b, v1.8b //mul with coeff 1
umlsl v30.8h, v4.8b, v0.8b
umlal v30.8h, v6.8b, v2.8b
umlsl v30.8h, v7.8b, v3.8b
st1 {v26.8b},[x7],x3 //stores the loaded value
sqrshrun v24.8b, v24.8h,#6
ld1 {v8.8b},[x6],x2 //load and increment
umull v28.8h, v6.8b, v1.8b //mul_res 2
umlsl v28.8h, v5.8b, v0.8b
umlal v28.8h, v7.8b, v2.8b
umlsl v28.8h, v8.8b, v3.8b
st1 {v24.8b},[x7],x3 //stores the loaded value
sqrshrun v30.8b, v30.8h,#6
ld1 {v9.8b},[x6],x2
umull v26.8h, v7.8b, v1.8b
add x7,x1,x3 //pu1_dst
umlsl v26.8h, v6.8b, v0.8b
st1 {v30.8b},[x1],#8 //stores the loaded value
sqrshrun v28.8b, v28.8h,#6
umlal v26.8h, v8.8b, v2.8b
ld1 {v10.8b},[x6],x2
umlsl v26.8h, v9.8b, v3.8b
umull v24.8h, v8.8b, v1.8b
sqrshrun v26.8b, v26.8h,#6
st1 {v28.8b},[x7],x3 //stores the loaded value
umlsl v24.8h, v7.8b, v0.8b
umlal v24.8h, v9.8b, v2.8b
st1 {v26.8b},[x7],x3 //stores the loaded value
umlsl v24.8h, v10.8b, v3.8b
sqrshrun v24.8b, v24.8h,#6
st1 {v24.8b},[x7],x3 //stores the loaded value
end_loops:
// ldmfd sp!,{x4-x12,x15} //reload the registers from sp
ldp x19, x20,[sp],#16
pop_v_regs
ret
|
