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_class0.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_class0(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@
51@**************Variables Vs Registers*****************************************
52@r0 =>  *pu1_src
53@r1 =>  src_strd
54@r2 =>  *pu1_src_left
55@r3 =>  *pu1_src_top
56@r4 =>  *pu1_src_top_left
57@r7 =>  *pu1_avail
58@r8 =>  *pi1_sao_offset
59@r9 =>  wd
60@r10=>  ht
61
62.text
63.p2align 2
64
65.extern gi1_table_edge_idx
66.globl ihevc_sao_edge_offset_class0_a9q
67
68gi1_table_edge_idx_addr:
69.long gi1_table_edge_idx - ulbl1 - 8
70
71ihevc_sao_edge_offset_class0_a9q:
72
73
74    STMFD       sp!, {r4-r12, r14}          @stack stores the values of the arguments
75    LDR         r9,[sp,#60]                 @Loads wd
76
77    LDR         r4,[sp,#40]                 @Loads pu1_src_top_left
78    VMOV.I8     Q1,#2                       @const_2 = vdupq_n_s8(2)
79    ADD         r11,r3,r9                   @pu1_src_top[wd]
80
81    LDR         r10,[sp,#64]                @Loads ht
82    VMOV.I16    Q2,#0                       @const_min_clip = vdupq_n_s16(0)
83    LDRB        r12,[r11,#-1]               @pu1_src_top[wd - 1]
84
85    LDR         r7,[sp,#52]                 @Loads pu1_avail
86    VMOV.I16    Q3,#255                     @const_max_clip = vdupq_n_u16((1 << bit_depth) - 1)
87    LDR         r14, gi1_table_edge_idx_addr @table pointer
88ulbl1:
89    add         r14,r14,pc
90
91    LDR         r8,[sp,#56]                 @Loads pi1_sao_offset
92    VMOV.S8     Q4,#0xFF                    @au1_mask = vdupq_n_s8(-1)
93    STRB        r12,[r4]                    @*pu1_src_top_left = pu1_src_top[wd - 1]
94
95    MOV         r6,r0                       @pu1_src_org
96    VLD1.8      D10,[r14]                   @edge_idx_tbl = vld1_s8(gi1_table_edge_idx)
97    SUB         r4,r10,#1                   @(ht - 1)
98
99    MOV         r12,r9                      @Move wd to r12 for loop count
100    VLD1.8      D11,[r8]                    @offset_tbl = vld1_s8(pi1_sao_offset)
101    MUL         r4,r4,r1                    @(ht - 1) * src_strd
102
103    ADD         r4,r4,r0                    @pu1_src[(ht - 1) * src_strd]
104
105SRC_TOP_LOOP:                               @wd is always multiple of 8
106    VLD1.8      D0,[r4]!                    @Load pu1_src[(ht - 1) * src_strd + col]
107    SUBS        r12,r12,#8                  @Decrement the loop counter by 8
108    VST1.8      D0,[r3]!                    @Store to pu1_src_top[col]
109    BNE         SRC_TOP_LOOP
110    ADD         r6,r6,#15                   @pu1_src_org[16 - 1]
111
112    CMP         r9,#16                      @Compare wd with 16
113    MOV         r3,r2                       @pu1_src_left backup to reload later
114    BLT         WIDTH_RESIDUE               @If not jump to WIDTH_RESIDUE where loop is unrolled for 8 case
115
116    MOV         r8,r9                       @move wd to r8 for loop count
117
118WIDTH_LOOP_16:
119    CMP         r8,r9                       @if(col == wd)
120    BNE         AU1_MASK_FF                 @jump to else part
121    LDRB        r12,[r7]                    @pu1_avail[0]
122    VMOV.8      D8[0],r12                   @vsetq_lane_s8(pu1_avail[0], au1_mask, 0)
123    B           SKIP_AU1_MASK_FF            @Skip the else part
124
125AU1_MASK_FF:
126    MOV         r12,#0xFF                   @move -1 to r12
127    VMOV.8      D8[0],r12                   @au1_mask = vsetq_lane_s8(-1, au1_mask, 0)
128
129SKIP_AU1_MASK_FF:
130    CMP         r8,#16                      @If col == 16
131    BNE         SKIP_MASKING_IF_NOT16       @If not skip masking
132    LDRB        r12,[r7,#1]                 @pu1_avail[1]
133    VMOV.8      D9[7],r12                   @au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15)
134
135SKIP_MASKING_IF_NOT16:
136    MOV         r12,r0                      @pu1_src_cpy = pu1_src
137    MOV         r4,r10                      @move ht to r4 for loop count
138
139PU1_SRC_LOOP:
140    LDRB        r11,[r2]                    @load pu1_src_left since ht - row =0 when it comes first pu1_src_left is incremented later
141    VLD1.8      D12,[r12]!                  @pu1_cur_row = vld1q_u8(pu1_src_cpy)
142    VLD1.8      D13,[r12], r1               @pu1_cur_row = vld1q_u8(pu1_src_cpy)
143    SUB         r12,#8
144    SUB         r5,r9,r8                    @wd - col
145
146    SUB         r14,r10,r4                  @ht - row
147    VMOV.8      D15[7],r11                  @vsetq_lane_u8(pu1_src_left[ht - row], pu1_cur_row_tmp, 15)
148    MUL         r14,r14,r1                  @(ht - row) * src_strd
149
150    VLD1.8      D26,[r12]!                  @II Iteration pu1_cur_row = vld1q_u8(pu1_src_cpy)
151    VLD1.8      D27,[r12]                   @II Iteration pu1_cur_row = vld1q_u8(pu1_src_cpy)
152    SUB         r12,#8
153    VEXT.8      Q7,Q7,Q6,#15                @pu1_cur_row_tmp = vextq_u8(pu1_cur_row_tmp, pu1_cur_row, 15)
154    ADD         r5,r14,r5                   @(ht - row) * src_strd + (wd - col)
155
156    LDRB        r11,[r2, #1]                @II Iteration load pu1_src_left since ht - row + 1 =1
157    VCGT.U8     Q8,Q6,Q7                    @vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp)
158    LDRB        r14,[r6,r5]                 @pu1_src_org[(ht - row) * src_strd + 16 - 1 + (wd - col)]
159
160    SUB         r4,r4,#1
161    VMOV.8      D29[7],r11                  @II Iteration vsetq_lane_u8(pu1_src_left[ht - row], pu1_cur_row_tmp, 15)
162    VCLT.U8     Q9,Q6,Q7                    @vcltq_u8(pu1_cur_row, pu1_cur_row_tmp)
163
164    SUB         r12,r12,r1                  @Decrement the pu1_src pointer by src_strd
165    VSUB.I8     Q10,Q9,Q8                   @sign_left = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
166    STRB        r14,[r2],#1                 @pu1_src_left[(ht - row)] = au1_src_left_tmp[(ht - row)]
167
168    LDRB        r11,[r12,#16]               @pu1_src_cpy[16]
169    VEXT.8      Q14,Q14,Q13,#15             @II Iteration pu1_cur_row_tmp = vextq_u8(pu1_cur_row_tmp, pu1_cur_row, 15)
170    SUB         r5,r9,r8                    @II wd - col
171
172    ADD         r12,r12,r1                  @Increment the pu1_src pointer by src_strd
173    VMOV.8      D14[0],r11                  @pu1_cur_row_tmp = vsetq_lane_u8(pu1_src_cpy[16], pu1_cur_row_tmp, 0)
174    VCGT.U8     Q15,Q13,Q14                 @II vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp)
175
176    LDRB        r11,[r12,#16]               @II pu1_src_cpy[16]
177    VEXT.8      Q7,Q6,Q7,#1                 @pu1_cur_row_tmp = vextq_u8(pu1_cur_row, pu1_cur_row_tmp, 1)
178    SUB         r14,r10,r4                  @II ht - row
179
180    VCLT.U8     Q0,Q13,Q14                  @II vcltq_u8(pu1_cur_row, pu1_cur_row_tmp)
181    VMOV.8      D28[0],r11                  @II pu1_cur_row_tmp = vsetq_lane_u8(pu1_src_cpy[16], pu1_cur_row_tmp, 0)
182    SUB         r12,r12,r1                  @Decrement the pu1_src pointer by src_strd
183
184    MUL         r14,r14,r1                  @II (ht - row) * src_strd
185    VCGT.U8     Q8,Q6,Q7                    @vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp)
186    ADD         r5,r14,r5                   @II (ht - row) * src_strd + (wd - col)
187
188    VCLT.U8     Q9,Q6,Q7                    @vcltq_u8(pu1_cur_row, pu1_cur_row_tmp)
189    VEXT.8      Q14,Q13,Q14,#1              @II pu1_cur_row_tmp = vextq_u8(pu1_cur_row, pu1_cur_row_tmp, 1)
190
191    LDRB        r14,[r6,r5]                 @II pu1_src_org[(ht - row) * src_strd + 16 - 1 + (wd - col)]
192    VSUB.I8     Q11,Q9,Q8                   @sign_right = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
193    SUBS        r4,r4,#1                    @Decrement row by 1
194
195    VADD.I8     Q7,Q1,Q10                   @edge_idx = vaddq_s8(const_2, sign_left)
196    STRB        r14,[r2],#1                 @II pu1_src_left[(ht - row)] = au1_src_left_tmp[(ht - row)]
197
198    VADD.I8     Q7,Q7,Q11                   @edge_idx = vaddq_s8(edge_idx, sign_right)
199    VMOVL.U8    Q9,D12                      @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row)))
200
201    VSUB.I8     Q10,Q0,Q15                  @II sign_left = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
202    VTBL.8      D14,{D10},D14               @vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx))
203    VCGT.U8     Q15,Q13,Q14                 @II vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp)
204
205    VCLT.U8     Q0,Q13,Q14                  @II vcltq_u8(pu1_cur_row, pu1_cur_row_tmp)
206    VTBL.8      D15,{D10},D15               @vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx))
207    VSUB.I8     Q11,Q0,Q15                  @II sign_right = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
208
209    VAND        Q7,Q7,Q4                    @edge_idx = vandq_s8(edge_idx, au1_mask)
210    VTBL.8      D16,{D11},D14               @offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx))
211    VMOVL.U8    Q0,D26                      @II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row)))
212
213    VADD.I8     Q14,Q1,Q10                  @II edge_idx = vaddq_s8(const_2, sign_left)
214    VADD.I8     Q14,Q14,Q11                 @II edge_idx = vaddq_s8(edge_idx, sign_right)
215
216    VADDW.S8    Q9,Q9,D16                   @pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset)
217    VTBL.8      D28,{D10},D28               @II vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx))
218    VMAX.S16    Q9,Q9,Q2                    @pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip)
219
220    VTBL.8      D29,{D10},D29               @II vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx))
221    VMIN.U16    Q9,Q9,Q3                    @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip))
222
223    VAND        Q14,Q14,Q4                  @II edge_idx = vandq_s8(edge_idx, au1_mask)
224    VTBL.8      D17,{D11},D15               @offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx))
225
226    VMOVL.U8    Q7,D13                      @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row)))
227    VTBL.8      D30,{D11},D28               @II offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx))
228    VADDW.S8    Q7,Q7,D17                   @pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset)
229
230    VMAX.S16    Q7,Q7,Q2                    @pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip)
231    VTBL.8      D31,{D11},D29               @II offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx))
232    VMIN.U16    Q7,Q7,Q3                    @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip))
233
234    VMOVN.I16   D18,Q9                      @vmovn_s16(pi2_tmp_cur_row.val[0])
235    VADDW.S8    Q0,Q0,D30                   @II pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset)
236
237    VMOVN.I16   D19,Q7                      @vmovn_s16(pi2_tmp_cur_row.val[1])
238    VMAX.S16    Q0,Q0,Q2                    @II pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip)
239
240    VMOVL.U8    Q14,D27                     @II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row)))
241    VMIN.U16    Q0,Q0,Q3                    @II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip))
242
243    VMOVN.I16   D0,Q0                       @II vmovn_s16(pi2_tmp_cur_row.val[0])
244    VADDW.S8    Q14,Q14,D31                 @II pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset)
245
246    VMAX.S16    Q14,Q14,Q2                  @II pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip)
247    VST1.8      {D18,D19},[r12],r1          @vst1q_u8(pu1_src_cpy, pu1_cur_row)
248    VMIN.U16    Q14,Q14,Q3                  @II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip))
249
250    VMOVN.I16   D1,Q14                      @II vmovn_s16(pi2_tmp_cur_row.val[1])
251
252    VST1.8      {D0,D1},[r12],r1            @II vst1q_u8(pu1_src_cpy, pu1_cur_row)
253
254    BNE         PU1_SRC_LOOP                @If not equal jump to the inner loop
255
256    ADD         r0,r0,#16                   @pu1_src += 16
257
258    SUBS        r8,r8,#16                   @Decrement column by 16
259    CMP         r8,#8                       @Check whether residue remains
260    MOV         r2,r3                       @Reload pu1_src_left
261    BEQ         WIDTH_RESIDUE               @If residue remains jump to residue loop
262    BGT         WIDTH_LOOP_16               @If not equal jump to width_loop
263    BLT         END_LOOPS                   @Jump to end function
264
265WIDTH_RESIDUE:
266    SUB         r6,r6,#15
267    AND         r8,r9,#0xF                  @wd_rem = wd & 0xF
268    CMP         r8,#0                       @Residue check
269    BEQ         END_LOOPS                   @No Residue jump to end function
270
271    CMP         r8,r9                       @if(wd_rem == wd)
272    BNE         AU1_MASK_FF_RESIDUE         @jump to else part
273    LDRB        r12,[r7]                    @pu1_avail[0]
274    VMOV.8      D8[0],r12                   @vsetq_lane_s8(pu1_avail[0], au1_mask, 0)
275    B           SKIP_AU1_MASK_FF_RESIDUE    @Skip the else part
276
277AU1_MASK_FF_RESIDUE:
278    MOV         r12,#0xFF                   @move -s to r12
279    VMOV.8      D8[0],r12                   @au1_mask = vsetq_lane_s8(-1, au1_mask, 0)
280
281SKIP_AU1_MASK_FF_RESIDUE:
282    LDRB        r11,[r7,#1]                 @pu1_avail[1]
283    SUB         r5,r9,#1                    @wd - 1
284
285    MOV         r4,r10                      @move ht to r4 for loop count
286    VMOV.8      D8[7],r11                   @au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15)
287    MOV         r12,r0                      @pu1_src_cpy = pu1_src
288
289PU1_SRC_LOOP_RESIDUE:
290    VLD1.8      D12,[r12]!                  @pu1_cur_row = vld1q_u8(pu1_src_cpy)
291    VLD1.8      D13,[r12]                   @pu1_cur_row = vld1q_u8(pu1_src_cpy)
292    SUB         r12,#8
293    LDRB        r11,[r2]                    @load pu1_src_left
294    VMOV.8      D15[7],r11                  @vsetq_lane_u8(pu1_src_left[ht - row], pu1_cur_row_tmp, 15)
295    VEXT.8      Q7,Q7,Q6,#15                @pu1_cur_row_tmp = vextq_u8(pu1_cur_row_tmp, pu1_cur_row, 15)
296
297    VCGT.U8     Q8,Q6,Q7                    @vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp)
298    VCLT.U8     Q9,Q6,Q7                    @vcltq_u8(pu1_cur_row, pu1_cur_row_tmp)
299    VSUB.I8     Q10,Q9,Q8                   @sign_left = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
300
301    LDRB        r11,[r12,#16]               @pu1_src_cpy[16]
302    VMOV.8      D14[0],r11                  @pu1_cur_row_tmp = vsetq_lane_u8(pu1_src_cpy[16], pu1_cur_row_tmp, 0)
303    VEXT.8      Q7,Q6,Q7,#1                 @pu1_cur_row_tmp = vextq_u8(pu1_cur_row, pu1_cur_row_tmp, 1)
304
305    VCGT.U8     Q8,Q6,Q7                    @vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp)
306    VCLT.U8     Q9,Q6,Q7                    @vcltq_u8(pu1_cur_row, pu1_cur_row_tmp)
307    VSUB.I8     Q11,Q9,Q8                   @sign_right = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt))
308
309    VADD.I8     Q12,Q1,Q10                  @edge_idx = vaddq_s8(const_2, sign_left)
310    VADD.I8     Q12,Q12,Q11                 @edge_idx = vaddq_s8(edge_idx, sign_right)
311
312    VTBL.8      D24,{D10},D24               @vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx))
313    VTBL.8      D25,{D10},D25               @vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx))
314
315    VAND        Q12,Q12,Q4                  @edge_idx = vandq_s8(edge_idx, au1_mask)
316
317    VNEG.S8     Q10,Q11                     @sign_left = vnegq_s8(sign_right)
318    VEXT.8      Q10,Q10,Q11,#15             @sign_left = vextq_s8(sign_left, sign_left, 15)
319
320    VTBL.8      D26,{D11},D24               @offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx))
321    VMOVL.U8    Q14,D12                     @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row)))
322    VADDW.S8    Q14,Q14,D26                 @pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset)
323    VMAX.S16    Q14,Q14,Q2                  @pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip)
324    VMIN.U16    Q14,Q14,Q3                  @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip))
325
326    VMOVN.I16   D28,Q14                     @vmovn_s16(pi2_tmp_cur_row.val[0])
327
328    SUB         r14,r10,r4                  @ht - row
329    MUL         r14,r14,r1                  @(ht - row) * src_strd
330    ADD         r11,r14,r5                  @(ht - row) * src_strd + (wd - 1)
331    LDRB        r14,[r6, r11]               @pu1_src_org[(ht - row) * src_strd + (wd - 1)]
332    STRB        r14,[r2],#1                 @pu1_src_left[(ht - row)] = au1_src_left_tmp[(ht - row)]
333
334    VST1.8      {D28},[r12],r1              @vst1q_u8(pu1_src_cpy, pu1_cur_row)
335
336    SUBS        r4,r4,#1                    @Decrement row by 1
337    BNE         PU1_SRC_LOOP_RESIDUE        @If not equal jump to the pu1_src loop
338
339END_LOOPS:
340    LDMFD       sp!,{r4-r12,r15}            @Reload the registers from SP
341
342
343
344
345