@/***************************************************************************** @* @* 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_sao_edge_offset_class3.s @* @* ,:brief @* Contains function definitions for inter prediction interpolation. @* Functions are coded using NEON intrinsics and can be compiled using@ ARM @* RVCT @* @* ,:author @* Parthiban V @* @* ,:par List of Functions: @* @* @* ,:remarks @* None @* @******************************************************************************* @*/ @void ihevc_sao_edge_offset_class3(UWORD8 *pu1_src, @ WORD32 src_strd, @ UWORD8 *pu1_src_left, @ UWORD8 *pu1_src_top, @ UWORD8 *pu1_src_top_left, @ UWORD8 *pu1_src_top_right, @ UWORD8 *pu1_src_bot_left, @ UWORD8 *pu1_avail, @ WORD8 *pi1_sao_offset, @ WORD32 wd, @ WORD32 ht) @**************Variables Vs Registers***************************************** @r0 => *pu1_src @r1 => src_strd @r2 => *pu1_src_left @r3 => *pu1_src_top @r4 => *pu1_src_top_left @r5 => *pu1_avail @r6 => *pi1_sao_offset @r7 => wd @r8=> ht .equ pu1_src_top_left_offset, 264 .equ pu1_src_top_right_offset, 268 .equ pu1_src_bot_left_offset, 272 .equ pu1_avail_offset, 276 .equ pi1_sao_offset, 280 .equ wd_offset, 284 .equ ht_offset, 288 .text .syntax unified .p2align 2 .extern gi1_table_edge_idx .globl ihevc_sao_edge_offset_class3_a9q gi1_table_edge_idx_addr_1: .long gi1_table_edge_idx - ulbl1 - 8 gi1_table_edge_idx_addr_2: .long gi1_table_edge_idx - ulbl2 - 8 gi1_table_edge_idx_addr_3: .long gi1_table_edge_idx - ulbl3 - 8 ihevc_sao_edge_offset_class3_a9q: STMFD sp!,{r4-r12,r14} @stack stores the values of the arguments vpush {d8 - d15} SUB sp,sp,#160 @Decrement the stack pointer to store some temp arr values LDR r7,[sp,#wd_offset] @Loads wd LDR r8,[sp,#ht_offset] @Loads ht SUB r9,r7,#1 @wd - 1 LDR r4,[sp,#pu1_src_top_left_offset] @Loads pu1_src_top_left LDRB r10,[r3,r9] @pu1_src_top[wd - 1] MOV r9,r7 @Move width to r9 for loop count LDR r5,[sp,#pu1_avail_offset] @Loads pu1_avail LDR r6,[sp,#pi1_sao_offset] @Loads pi1_sao_offset STR r3,[sp,#156] @Store pu1_src_top in sp STRB r10,[sp] @u1_src_top_left_tmp = pu1_src_top[wd - 1] SUB r10,r8,#1 @ht-1 MLA r11,r10,r1,r0 @pu1_src[(ht - 1) * src_strd + col] ADD r12,sp,#2 @temp array AU1_SRC_TOP_LOOP: VLD1.8 D0,[r11]! @pu1_src[(ht - 1) * src_strd + col] SUBS r9,r9,#8 @Decrement the loop count by 8 VST1.8 D0,[r12]! @au1_src_top_tmp[col] = pu1_src[(ht - 1) * src_strd + col] BNE AU1_SRC_TOP_LOOP PU1_AVAIL_5_LOOP: LDRB r9,[r5,#5] @pu1_avail[5] CMP r9,#0 SUB r10,r7,#1 @[wd - 1] LDRB r9,[r0,r10] @u1_pos_0_0_tmp = pu1_src[wd - 1] BEQ PU1_AVAIL_6_LOOP LDR r11,[sp,#pu1_src_top_right_offset] @Load pu1_src_top_right from sp SUB r10,r10,#1 @[wd - 1 - 1] LDRB r11,[r11] @pu1_src_top_right[0] SUB r12,r9,r11 @pu1_src[wd - 1] - pu1_src_top_right[0] ADD r11,r0,r1 @pu1_src + src_strd LDRB r14,[r11,r10] @pu1_src[wd - 1 - 1 + src_strd] CMP r12,#0 MVNLT r12,#0 SUB r11,r9,r14 @pu1_src[wd - 1] - pu1_src[wd - 1 - 1 + src_strd] MOVGT r12,#1 @SIGN(pu1_src[wd - 1] - pu1_src_top_right[0]) CMP r11,#0 MVNLT r11,#0 MOVGT r11,#1 @SIGN(pu1_src[wd - 1] - pu1_src[wd - 1 - 1 + src_strd]) LDR r14, gi1_table_edge_idx_addr_1 @table pointer ulbl1: add r14,r14,pc ADD r11,r12,r11 @SIGN(pu1_src[wd - 1] - pu1_src_top_right[0]) + SIGN(pu1_src[wd - 1] - pu1_src[wd - 1 - 1 + src_strd]) ADD r11,r11,#2 @edge_idx LDRSB r12,[r14,r11] @edge_idx = gi1_table_edge_idx[edge_idx] CMP r12,#0 @0 != edge_idx BEQ PU1_AVAIL_6_LOOP LDRSB r10,[r6,r12] @pi1_sao_offset[edge_idx] ADD r9,r9,r10 @pu1_src[0] + pi1_sao_offset[edge_idx] USAT r9,#8,r9 @u1_pos_0_0_tmp = CLIP3(pu1_src[0] + pi1_sao_offset[edge_idx], 0, (1 << bit_depth) - 1) PU1_AVAIL_6_LOOP: LDRB r10,[r5,#6] @pu1_avail[6] SUB r11,r8,#1 @ht - 1 CMP r10,#0 STR r0,[sp,#148] @Store pu1_src in sp MLA r12,r11,r1,r0 @pu1_src[(ht - 1) * src_strd] LDRB r10,[r12] @u1_pos_wd_ht_tmp = pu1_src[(ht - 1) * src_strd] BEQ PU1_AVAIL_3_LOOP LDR r14,[sp,#pu1_src_bot_left_offset] @Load pu1_src_bot_left from sp SUB r11,r12,r1 @pu1_src[(ht - 1) * src_strd) - src_strd] LDRB r14,[r14] @Load pu1_src_bot_left[0] ADD r11,r11,#1 @pu1_src[(ht - 1) * src_strd + 1 - src_strd] LDRB r11,[r11] @Load pu1_src[(ht - 1) * src_strd + 1 - src_strd] SUB r14,r10,r14 @pu1_src[(ht - 1) * src_strd] - pu1_src_bot_left[0] SUB r11,r10,r11 @pu1_src[(ht - 1) * src_strd] - pu1_src[(ht - 1) * src_strd + 1 - src_strd] CMP r11,#0 MVNLT r11,#0 MOVGT r11,#1 @SIGN(pu1_src[(ht - 1) * src_strd] - pu1_src[(ht - 1) * src_strd + 1 - src_strd]) CMP r14,#0 MVNLT r14,#0 MOVGT r14,#1 @SIGN(pu1_src[(ht - 1) * src_strd] - pu1_src_bot_left[0]) ADD r11,r11,r14 @Add 2 sign value LDR r14, gi1_table_edge_idx_addr_2 @table pointer ulbl2: add r14,r14,pc ADD r11,r11,#2 @edge_idx LDRSB r12,[r14,r11] @edge_idx = gi1_table_edge_idx[edge_idx] CMP r12,#0 BEQ PU1_AVAIL_3_LOOP LDRSB r11,[r6,r12] @pi1_sao_offset[edge_idx] ADD r10,r10,r11 @pu1_src[(ht - 1) * src_strd] + pi1_sao_offset[edge_idx] USAT r10,#8,r10 @u1_pos_wd_ht_tmp = CLIP3(pu1_src[(ht - 1) * src_strd] + pi1_sao_offset[edge_idx], 0, (1 << bit_depth) - 1) PU1_AVAIL_3_LOOP: STR r2,[sp,#152] @Store pu1_src_left in sp MOV r12,r8 @Move ht MOV r14,r2 @Move pu1_src_left to pu1_src_left_cpy VMOV.I8 Q0,#2 @const_2 = vdupq_n_s8(2) LDRB r11,[r5,#3] @pu1_avail[3] CMP r11,#0 VMOV.I16 Q1,#0 @const_min_clip = vdupq_n_s16(0) SUBEQ r12,r12,#1 @ht_tmp-- LDRB r5,[r5,#2] @pu1_avail[2] VMOV.I16 Q2,#255 @const_max_clip = vdupq_n_u16((1 << bit_depth) - 1) CMP r5,#0 ADDEQ r0,r0,r1 @pu1_src += src_strd VLD1.8 D7,[r6] @offset_tbl = vld1_s8(pi1_sao_offset) SUBEQ r12,r12,#1 @ht_tmp-- LDR r6, gi1_table_edge_idx_addr_3 @table pointer ulbl3: add r6,r6,pc VMOV.S8 Q4,#0xFF @au1_mask = vdupq_n_s8(-1) ADDEQ r14,r14,#1 @pu1_src_left_cpy += 1 STR r0,[sp,#144] @Store pu1_src in sp VLD1.8 D6,[r6] @edge_idx_tbl = vld1_s8(gi1_table_edge_idx) MOV r6,r7 @move wd to r6 loop_count CMP r7,#16 @Compare wd with 16 BLT WIDTH_RESIDUE @If not jump to WIDTH_RESIDUE where loop is unrolled for 8 case CMP r8,#4 @Compare ht with 4 BLE WD_16_HT_4_LOOP @If jump to WD_16_HT_4_LOOP WIDTH_LOOP_16: LDR r7,[sp,#wd_offset] @Loads wd LDR r5,[sp,#pu1_avail_offset] @Loads pu1_avail CMP r6,r7 @col == wd LDRBEQ r8,[r5] @pu1_avail[0] MOVNE r8,#-1 VMOV.8 d8[0],r8 @au1_mask = vsetq_lane_s8(-1, au1_mask, 0) CMP r6,#16 @if(col == 16) BNE SKIP_AU1_MASK_VAL LDRB r8,[r5,#1] @pu1_avail[1] VMOV.8 d9[7],r8 @au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) SKIP_AU1_MASK_VAL: LDRB r8,[r5,#2] @pu1_avail[2] CMP r8,#0 LDR r4,[sp,#ht_offset] @Loads ht SUBEQ r8,r0,r1 @pu1_src - src_strd MOVNE r8,r3 ADD r5,sp,#66 @*au1_src_left_tmp LDR r7,[sp,#wd_offset] @Loads wd ADD r8,r8,#1 @pu1_src - src_strd + 1 SUB r7,r7,r6 @(wd - col) VLD1.8 D10,[r8]! @pu1_top_row = vld1q_u8(pu1_src - src_strd + 1) VLD1.8 D11,[r8] @pu1_top_row = vld1q_u8(pu1_src - src_strd + 1) SUB r8,#8 ADD r3,r3,#16 LDR r8,[sp,#148] @Loads *pu1_src VLD1.8 D12,[r0]! @pu1_cur_row = vld1q_u8(pu1_src) VLD1.8 D13,[r0] @pu1_cur_row = vld1q_u8(pu1_src) SUB r0,#8 ADD r7,r7,#15 @15 + (wd - col) ADD r7,r8,r7 @pu1_src[0 * src_strd + 15 + (wd - col)] VCGT.U8 Q7,Q6,Q5 @vcgtq_u8(pu1_cur_row, pu1_top_row) SUB r5,r5,#1 AU1_SRC_LEFT_LOOP: LDRB r8,[r7],r1 @load the value and increment by src_strd SUBS r4,r4,#1 @decrement the loop count STRB r8,[r5,#1]! @store it in the stack pointer BNE AU1_SRC_LEFT_LOOP VMOV.I8 Q9,#0 VCLT.U8 Q8,Q6,Q5 @vcltq_u8(pu1_cur_row, pu1_top_row) ADD r8,r0,r1 @I *pu1_src + src_strd VSUB.U8 Q7,Q8,Q7 @sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) MOV r7,r12 @row count, move ht_tmp to r7 SUB r5,r12,r7 @I ht_tmp - row VLD1.8 D16,[r8]! @I pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) VLD1.8 D17,[r8] @I pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) SUB r8,#8 ADD r8,r14,r5 @I pu1_src_left_cpy[ht_tmp - row] ADD r8,r8,#1 @I pu1_src_left_cpy[ht_tmp - row + 1] LDRB r8,[r8] LDR r5,[sp,#pu1_avail_offset] @I Loads pu1_avail VMOV.8 D19[7],r8 @I vsetq_lane_u8 LDRB r5,[r5,#2] @I pu1_avail[2] VEXT.8 Q9,Q9,Q8,#15 @I pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15) CMP r5,#0 @I BNE SIGN_UP_CHANGE_DONE @I SIGN_UP_CHANGE: LDRB r8,[r0,#15] @I pu1_src_cpy[15] SUB r5,r0,r1 @I pu1_src_cpy[16 - src_strd] LDRB r5,[r5,#16] @I load the value SUB r8,r8,r5 @I pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd] CMP r8,#0 @I MVNLT r8,#0 @I MOVGT r8,#1 @I SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]) VMOV.8 D15[7],r8 @I sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15) SIGN_UP_CHANGE_DONE: VCGT.U8 Q5,Q6,Q9 @I vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) VCLT.U8 Q9,Q6,Q9 @I vcltq_u8(pu1_cur_row, pu1_next_row_tmp) VSUB.U8 Q5,Q9,Q5 @I sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) VADD.I8 Q9,Q0,Q7 @I edge_idx = vaddq_s8(const_2, sign_up) VADD.I8 Q9,Q9,Q5 @I edge_idx = vaddq_s8(edge_idx, sign_down) VTBL.8 D18,{D6},D18 @I vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) VNEG.S8 Q7,Q5 @I sign_up = vnegq_s8(sign_down) VEXT.8 Q7,Q7,Q7,#1 @I sign_up = vextq_s8(sign_up, sign_up, 1) VTBL.8 D19,{D6},D19 @I vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) VMOVL.U8 Q10,D12 @I pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) VAND Q9,Q9,Q4 @I edge_idx = vandq_s8(edge_idx, au1_mask) VTBL.8 D10,{D7},D18 @I offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) VMOVL.U8 Q11,D13 @I pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) VADDW.S8 Q10,Q10,D10 @I pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) VMAX.S16 Q10,Q10,Q1 @I pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) VTBL.8 D11,{D7},D19 @I offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) VMIN.U16 Q10,Q10,Q2 @I pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) VMOV Q6,Q8 VADDW.S8 Q11,Q11,D11 @I pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) VMAX.S16 Q11,Q11,Q1 @I pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) VMIN.U16 Q11,Q11,Q2 @I pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip)) SUB r7,r7,#1 @I Decrement the ht_tmp loop count by 1 PU1_SRC_LOOP: ADD r8,r0,r1,LSL #1 @II *pu1_src + src_strd VMOVN.I16 D20,Q10 @I vmovn_s16(pi2_tmp_cur_row.val[0]) SUB r5,r12,r7 @II ht_tmp - row ADD r4,r0,r1 @II pu1_src_cpy[16 - src_strd] VMOVN.I16 D21,Q11 @I vmovn_s16(pi2_tmp_cur_row.val[1]) ADD r2,r8,r1 @III *pu1_src + src_strd LDRB r11,[r4,#15] @II pu1_src_cpy[15] VLD1.8 D16,[r8]! @II pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) VLD1.8 D17,[r8] @II pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) SUB r8,#8 SUB r7,r7,#1 @II Decrement the ht_tmp loop count by 1 ADD r8,r14,r5 @II pu1_src_left_cpy[ht_tmp - row] VLD1.8 D30,[r2]! @III pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) VLD1.8 D31,[r2] @III pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) SUB r2,#8 LDRB r8,[r8,#1] LDRB r4,[r0,#16] @II load the value VMOV.8 D19[7],r8 @II vsetq_lane_u8 SUB r11,r11,r4 @II pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd] CMP r11,#0 @II VST1.8 {Q10},[r0],r1 @I vst1q_u8(pu1_src_cpy, pu1_cur_row) SUB r5,r12,r7 @III ht_tmp - row MVNLT r11,#0 @II VEXT.8 Q9,Q9,Q8,#15 @II pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15) MOVGT r11,#1 @II SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]) ADD r8,r14,r5 @III pu1_src_left_cpy[ht_tmp - row] VMOV.8 D15[7],r11 @II sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15) CMP r7,#1 @III BNE NEXT_ROW_ELSE_2 @III LDR r5,[sp,#pu1_avail_offset] @III Loads pu1_avail LDRB r5,[r5,#3] @III pu1_avail[3] CMP r5,#0 @III SUBNE r8,r2,#2 @III pu1_src_cpy[src_strd - 1] NEXT_ROW_ELSE_2: LDRB r8,[r8,#1] @III VCGT.U8 Q12,Q6,Q9 @II vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) ADD r5,r0,r1 LDRB r2,[r5,#15] @III pu1_src_cpy[15] VCLT.U8 Q13,Q6,Q9 @II vcltq_u8(pu1_cur_row, pu1_next_row_tmp) LDRB r5,[r0,#16] @III load the value SUB r2,r2,r5 @III pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd] VSUB.U8 Q12,Q13,Q12 @II sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) CMP r2,#0 @III MVNLT r2,#0 @III VMOV.8 D19[7],r8 @III vsetq_lane_u8 MOVGT r2,#1 @III SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]) SUB r7,r7,#1 @III Decrement the ht_tmp loop count by 1 VADD.I8 Q13,Q0,Q7 @II edge_idx = vaddq_s8(const_2, sign_up) VNEG.S8 Q7,Q12 @II sign_up = vnegq_s8(sign_down) VEXT.8 Q9,Q9,Q15,#15 @III pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15) VADD.I8 Q13,Q13,Q12 @II edge_idx = vaddq_s8(edge_idx, sign_down) VEXT.8 Q7,Q7,Q7,#1 @II sign_up = vextq_s8(sign_up, sign_up, 1) VTBL.8 D26,{D6},D26 @II vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) VCGT.U8 Q5,Q8,Q9 @III vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) VMOV.8 D15[7],r2 @III sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15) VTBL.8 D27,{D6},D27 @II vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) VCLT.U8 Q9,Q8,Q9 @III vcltq_u8(pu1_cur_row, pu1_next_row_tmp) VMOVL.U8 Q14,D12 @II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) VAND Q13,Q13,Q4 @II edge_idx = vandq_s8(edge_idx, au1_mask) VSUB.U8 Q5,Q9,Q5 @III sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) VTBL.8 D24,{D7},D26 @II offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) VADD.I8 Q9,Q0,Q7 @III edge_idx = vaddq_s8(const_2, sign_up) VADD.I8 Q9,Q9,Q5 @III edge_idx = vaddq_s8(edge_idx, sign_down) VTBL.8 D25,{D7},D27 @II offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) VNEG.S8 Q7,Q5 @III sign_up = vnegq_s8(sign_down) VADDW.S8 Q14,Q14,D24 @II pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) VTBL.8 D18,{D6},D18 @III vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) VMAX.S16 Q14,Q14,Q1 @II pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) VEXT.8 Q7,Q7,Q7,#1 @III sign_up = vextq_s8(sign_up, sign_up, 1) VTBL.8 D19,{D6},D19 @III vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) VMIN.U16 Q14,Q14,Q2 @II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) VMOVL.U8 Q13,D13 @II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) VAND Q9,Q9,Q4 @III edge_idx = vandq_s8(edge_idx, au1_mask) VADDW.S8 Q13,Q13,D25 @II pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) VTBL.8 D10,{D7},D18 @III offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) VMAX.S16 Q13,Q13,Q1 @II pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) VMOVL.U8 Q10,D16 @III pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) VMIN.U16 Q13,Q13,Q2 @II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip)) VADDW.S8 Q10,Q10,D10 @III pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) VTBL.8 D11,{D7},D19 @III offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) VMAX.S16 Q10,Q10,Q1 @III pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) VMOVL.U8 Q11,D17 @III pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) VMIN.U16 Q10,Q10,Q2 @III pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) VMOVN.I16 D28,Q14 @II vmovn_s16(pi2_tmp_cur_row.val[0]) VADDW.S8 Q11,Q11,D11 @III pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) VMOVN.I16 D29,Q13 @II vmovn_s16(pi2_tmp_cur_row.val[1]) VMAX.S16 Q11,Q11,Q1 @III pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) VMOV Q6,Q15 @II pu1_cur_row = pu1_next_row VMIN.U16 Q11,Q11,Q2 @III pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip)) CMP r7,#1 @III VST1.8 {Q14},[r0],r1 @II vst1q_u8(pu1_src_cpy, pu1_cur_row) BGT PU1_SRC_LOOP @If not equal jump to PU1_SRC_LOOP BLT INNER_LOOP_DONE ADD r8,r0,r1,LSL #1 @*pu1_src + src_strd VMOVN.I16 D20,Q10 @III vmovn_s16(pi2_tmp_cur_row.val[0]) LDR r5,[sp,#pu1_avail_offset] @Loads pu1_avail LDRB r5,[r5,#3] @pu1_avail[3] VMOVN.I16 D21,Q11 @III vmovn_s16(pi2_tmp_cur_row.val[1]) CMP r5,#0 ADD r4,r0,r1 @pu1_src_cpy[16 - src_strd] VLD1.8 D16,[r8]! @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) VLD1.8 D17,[r8] @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) SUB r8,#8 LDRB r5,[r0,#16] @load the value BEQ NEXT_ROW_ELSE_3 LDRB r8,[r8,#-1] @pu1_src_cpy[src_strd - 1] B NEXT_ROW_POINTER_ASSIGNED_3 NEXT_ROW_ELSE_3: SUB r11,r12,r7 @ht_tmp - row ADD r8,r14,r11 @pu1_src_left_cpy[ht_tmp - row] ADD r8,r8,#1 @pu1_src_left_cpy[ht_tmp - row + 1] LDRB r8,[r8] NEXT_ROW_POINTER_ASSIGNED_3: LDRB r11,[r4,#15] @pu1_src_cpy[15] VMOV.8 D19[7],r8 @vsetq_lane_u8 SUB r8,r11,r5 @pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd] CMP r8,#0 VEXT.8 Q9,Q9,Q8,#15 @pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15) MVNLT r8,#0 VST1.8 {Q10},[r0],r1 @III vst1q_u8(pu1_src_cpy, pu1_cur_row) VCGT.U8 Q12,Q6,Q9 @vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) MOVGT r8,#1 @SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]) VCLT.U8 Q13,Q6,Q9 @vcltq_u8(pu1_cur_row, pu1_next_row_tmp) VMOV.8 D15[7],r8 @sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15) VSUB.U8 Q12,Q13,Q12 @sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) VMOVL.U8 Q10,D12 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) VADD.I8 Q13,Q0,Q7 @edge_idx = vaddq_s8(const_2, sign_up) VMOVL.U8 Q11,D13 @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) VADD.I8 Q13,Q13,Q12 @edge_idx = vaddq_s8(edge_idx, sign_down) VTBL.8 D26,{D6},D26 @vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) VTBL.8 D27,{D6},D27 @vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) VAND Q13,Q13,Q4 @edge_idx = vandq_s8(edge_idx, au1_mask) VTBL.8 D24,{D7},D26 @offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) VADDW.S8 Q10,Q10,D24 @pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) VTBL.8 D25,{D7},D27 @offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) VMAX.S16 Q10,Q10,Q1 @pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) VMIN.U16 Q10,Q10,Q2 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) VADDW.S8 Q11,Q11,D25 @pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) VMAX.S16 Q11,Q11,Q1 @pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) VMIN.U16 Q11,Q11,Q2 @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip)) INNER_LOOP_DONE: VMOVN.I16 D20,Q10 @vmovn_s16(pi2_tmp_cur_row.val[0]) LDR r8,[sp,#ht_offset] @Loads ht VMOVN.I16 D21,Q11 @vmovn_s16(pi2_tmp_cur_row.val[1]) ADD r5,sp,#66 @*au1_src_left_tmp VST1.8 {Q10},[r0],r1 @vst1q_u8(pu1_src_cpy, pu1_cur_row) LDR r2,[sp,#152] @Loads *pu1_src_left SRC_LEFT_LOOP: LDR r7,[r5],#4 @au1_src_left_tmp[row] SUBS r8,r8,#4 STR r7,[r2],#4 @pu1_src_left[row] = au1_src_left_tmp[row] BNE SRC_LEFT_LOOP SUBS r6,r6,#16 @Decrement the wd loop count by 16 CMP r6,#8 @Check whether residue remains BLT RE_ASSINING_LOOP @Jump to re-assigning loop LDR r7,[sp,#wd_offset] @Loads wd LDR r0,[sp,#144] @Loads *pu1_src SUB r7,r7,r6 ADD r0,r0,r7 BGT WIDTH_LOOP_16 @If not equal jump to width_loop BEQ WIDTH_RESIDUE @If residue remains jump to residue loop WD_16_HT_4_LOOP: LDR r5,[sp,#pu1_avail_offset] @Loads pu1_avail LDR r7,[sp,#wd_offset] @Loads wd CMP r6,r7 @col == wd LDRBEQ r8,[r5] @pu1_avail[0] MOVNE r8,#-1 VMOV.8 d8[0],r8 @au1_mask = vsetq_lane_s8(-1, au1_mask, 0) CMP r6,#16 @if(col == 16) BNE SKIP_AU1_MASK_VAL_WD_16_HT_4 LDRB r8,[r5,#1] @pu1_avail[1] VMOV.8 d9[7],r8 @au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) SKIP_AU1_MASK_VAL_WD_16_HT_4: LDRB r8,[r5,#2] @pu1_avail[2] CMP r8,#0 SUBEQ r8,r0,r1 @pu1_src - src_strd MOVNE r8,r3 ADD r8,r8,#1 @pu1_src - src_strd + 1 VLD1.8 D10,[r8]! @pu1_top_row = vld1q_u8(pu1_src - src_strd + 1) VLD1.8 D11,[r8] @pu1_top_row = vld1q_u8(pu1_src - src_strd + 1) SUB r8,#8 ADD r3,r3,#16 ADD r5,sp,#66 @*au1_src_left_tmp LDR r4,[sp,#ht_offset] @Loads ht LDR r7,[sp,#wd_offset] @Loads wd SUB r7,r7,r6 @(wd - col) ADD r7,r7,#15 @15 + (wd - col) LDR r8,[sp,#148] @Loads *pu1_src ADD r7,r8,r7 @pu1_src[0 * src_strd + 15 + (wd - col)] SUB r5,r5,#1 AU1_SRC_LEFT_LOOP_WD_16_HT_4: LDRB r8,[r7],r1 @load the value and increment by src_strd STRB r8,[r5,#1]! @store it in the stack pointer SUBS r4,r4,#1 @decrement the loop count BNE AU1_SRC_LEFT_LOOP_WD_16_HT_4 VLD1.8 D12,[r0]! @pu1_cur_row = vld1q_u8(pu1_src) VLD1.8 D13,[r0] @pu1_cur_row = vld1q_u8(pu1_src) SUB r0,#8 VCGT.U8 Q7,Q6,Q5 @vcgtq_u8(pu1_cur_row, pu1_top_row) VCLT.U8 Q8,Q6,Q5 @vcltq_u8(pu1_cur_row, pu1_top_row) VSUB.U8 Q7,Q8,Q7 @sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) VMOV.I8 Q9,#0 MOV r7,r12 @row count, move ht_tmp to r7 PU1_SRC_LOOP_WD_16_HT_4: ADD r8,r0,r1 @*pu1_src + src_strd VLD1.8 D16,[r8]! @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) VLD1.8 D17,[r8] @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) SUB r8,#8 LDR r5,[sp,#pu1_avail_offset] @Loads pu1_avail LDRB r5,[r5,#3] @pu1_avail[3] CMP r5,#0 BEQ NEXT_ROW_ELSE_WD_16_HT_4 CMP r7,#1 LDRBEQ r8,[r8,#-1] @pu1_src_cpy[src_strd - 1] BEQ NEXT_ROW_POINTER_ASSIGNED_WD_16_HT_4 NEXT_ROW_ELSE_WD_16_HT_4: SUB r5,r12,r7 @ht_tmp - row ADD r8,r14,r5 @pu1_src_left_cpy[ht_tmp - row] ADD r8,r8,#1 @pu1_src_left_cpy[ht_tmp - row + 1] LDRB r8,[r8] NEXT_ROW_POINTER_ASSIGNED_WD_16_HT_4: VMOV.8 D19[7],r8 @vsetq_lane_u8 VEXT.8 Q9,Q9,Q8,#15 @pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15) CMP r7,r12 BNE SIGN_UP_CHANGE_WD_16_HT_4 LDR r5,[sp,#pu1_avail_offset] @Loads pu1_avail LDRB r5,[r5,#2] @pu1_avail[2] CMP r5,#0 BNE SIGN_UP_CHANGE_DONE_WD_16_HT_4 SIGN_UP_CHANGE_WD_16_HT_4: LDRB r8,[r0,#15] @pu1_src_cpy[15] ADD r5,r0,#16 @pu1_src_cpy[16] SUB r5,r5,r1 @pu1_src_cpy[16 - src_strd] LDRB r5,[r5] @load the value SUB r8,r8,r5 @pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd] CMP r8,#0 MVNLT r8,#0 MOVGT r8,#1 @SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]) VMOV.8 D15[7],r8 @sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15) SIGN_UP_CHANGE_DONE_WD_16_HT_4: VCGT.U8 Q10,Q6,Q9 @vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) VCLT.U8 Q11,Q6,Q9 @vcltq_u8(pu1_cur_row, pu1_next_row_tmp) VSUB.U8 Q12,Q11,Q10 @sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) VADD.I8 Q13,Q0,Q7 @edge_idx = vaddq_s8(const_2, sign_up) VADD.I8 Q13,Q13,Q12 @edge_idx = vaddq_s8(edge_idx, sign_down) VTBL.8 D26,{D6},D26 @vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) VTBL.8 D27,{D6},D27 @vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) VAND Q13,Q13,Q4 @edge_idx = vandq_s8(edge_idx, au1_mask) VNEG.S8 Q7,Q12 @sign_up = vnegq_s8(sign_down) VEXT.8 Q7,Q7,Q7,#1 @sign_up = vextq_s8(sign_up, sign_up, 1) VTBL.8 D24,{D7},D26 @offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) VMOVL.U8 Q14,D12 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) VADDW.S8 Q14,Q14,D24 @pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) VMAX.S16 Q14,Q14,Q1 @pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) VMIN.U16 Q14,Q14,Q2 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) VTBL.8 D25,{D7},D27 @offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) VMOVL.U8 Q15,D13 @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) VADDW.S8 Q15,Q15,D25 @pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) VMAX.S16 Q15,Q15,Q1 @pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) VMIN.U16 Q15,Q15,Q2 @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[1]), const_max_clip)) VMOVN.I16 D28,Q14 @vmovn_s16(pi2_tmp_cur_row.val[0]) VMOVN.I16 D29,Q15 @vmovn_s16(pi2_tmp_cur_row.val[1]) VST1.8 {Q14},[r0],r1 @vst1q_u8(pu1_src_cpy, pu1_cur_row) VMOV Q6,Q8 @pu1_cur_row = pu1_next_row SUBS r7,r7,#1 @Decrement the ht_tmp loop count by 1 BNE PU1_SRC_LOOP_WD_16_HT_4 @If not equal jump to PU1_SRC_LOOP_WD_16_HT_4 LDR r8,[sp,#ht_offset] @Loads ht ADD r5,sp,#66 @*au1_src_left_tmp LDR r2,[sp,#152] @Loads *pu1_src_left SRC_LEFT_LOOP_WD_16_HT_4: LDR r7,[r5],#4 @au1_src_left_tmp[row] STR r7,[r2],#4 @pu1_src_left[row] = au1_src_left_tmp[row] SUBS r8,r8,#4 BNE SRC_LEFT_LOOP_WD_16_HT_4 SUBS r6,r6,#16 @Decrement the wd loop count by 16 BLE RE_ASSINING_LOOP @Jump to re-assigning loop LDR r7,[sp,#wd_offset] @Loads wd LDR r0,[sp,#144] @Loads *pu1_src SUB r7,r7,r6 ADD r0,r0,r7 BGT WD_16_HT_4_LOOP @If not equal jump to width_loop WIDTH_RESIDUE: LDR r7,[sp,#wd_offset] @Loads wd LDR r5,[sp,#pu1_avail_offset] @Loads pu1_avail CMP r6,r7 @wd_residue == wd LDRBEQ r8,[r5] @pu1_avail[0] MOVNE r8,#-1 VMOV.8 d8[0],r8 @au1_mask = vsetq_lane_s8(-1, au1_mask, 0) LDRB r8,[r5,#1] @pu1_avail[1] VMOV.8 d8[7],r8 @au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) PU1_AVAIL_2_RESIDUE: LDRB r8,[r5,#2] @pu1_avail[2] CMP r8,#0 SUBEQ r8,r0,r1 @pu1_src - src_strd MOVNE r8,r3 ADD r8,r8,#1 @pu1_src - src_strd + 1 VLD1.8 D10,[r8]! @pu1_top_row = vld1q_u8(pu1_src - src_strd + 1) VLD1.8 D11,[r8] @pu1_top_row = vld1q_u8(pu1_src - src_strd + 1) SUB r8,#8 ADD r5,sp,#66 @*au1_src_left_tmp LDR r4,[sp,#ht_offset] @Loads ht LDR r7,[sp,#wd_offset] @Loads wd LDR r8,[sp,#148] @Loads *pu1_src SUB r7,r7,#1 @(wd - 1) ADD r7,r8,r7 @pu1_src[0 * src_strd + (wd - 1)] SUB r5,r5,#1 AU1_SRC_LEFT_LOOP_RESIDUE: LDRB r8,[r7],r1 @load the value and increment by src_strd STRB r8,[r5,#1]! @store it in the stack pointer SUBS r4,r4,#1 @decrement the loop count BNE AU1_SRC_LEFT_LOOP_RESIDUE VLD1.8 D12,[r0]! @pu1_cur_row = vld1q_u8(pu1_src) VLD1.8 D13,[r0] @pu1_cur_row = vld1q_u8(pu1_src) SUB r0,#8 VCGT.U8 Q7,Q6,Q5 @vcgtq_u8(pu1_cur_row, pu1_top_row) VCLT.U8 Q8,Q6,Q5 @vcltq_u8(pu1_cur_row, pu1_top_row) VSUB.U8 Q7,Q8,Q7 @sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) MOV r7,r12 @row count, move ht_tmp to r7 PU1_SRC_LOOP_RESIDUE: VMOV.I8 Q9,#0 ADD r8,r0,r1 @*pu1_src + src_strd VLD1.8 D16,[r8]! @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) VLD1.8 D17,[r8] @pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) SUB r8,#8 LDR r5,[sp,#pu1_avail_offset] @Loads pu1_avail LDRB r5,[r5,#3] @pu1_avail[3] CMP r5,#0 BEQ NEXT_ROW_ELSE_RESIDUE CMP r7,#1 LDRBEQ r8,[r8,#-1] @pu1_src_cpy[src_strd - 1] BEQ NEXT_ROW_POINTER_ASSIGNED_RESIDUE NEXT_ROW_ELSE_RESIDUE: SUB r5,r12,r7 @ht_tmp - row ADD r8,r14,r5 @pu1_src_left_cpy[ht_tmp - row] ADD r8,r8,#1 @pu1_src_left_cpy[ht_tmp - row + 1] LDRB r8,[r8] NEXT_ROW_POINTER_ASSIGNED_RESIDUE: VMOV.8 D19[7],r8 @vsetq_lane_u8 VEXT.8 Q9,Q9,Q8,#15 @pu1_next_row_tmp = vextq_u8(pu1_next_row_tmp, pu1_next_row, 15) CMP r7,r12 BNE SIGN_UP_CHANGE_RESIDUE LDR r5,[sp,#pu1_avail_offset] @Loads pu1_avail LDRB r5,[r5,#2] @pu1_avail[2] CMP r5,#0 BNE SIGN_UP_CHANGE_DONE_RESIDUE SIGN_UP_CHANGE_RESIDUE: LDRB r8,[r0,#15] @pu1_src_cpy[15] ADD r5,r0,#16 @pu1_src_cpy[16] SUB r5,r5,r1 @pu1_src_cpy[16 - src_strd] LDRB r5,[r5] @load the value SUB r8,r8,r5 @pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd] CMP r8,#0 MVNLT r8,#0 MOVGT r8,#1 @SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]) VMOV.8 D15[7],r8 @sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[15] - pu1_src_cpy[16 - src_strd]), sign_up, 15) SIGN_UP_CHANGE_DONE_RESIDUE: VCGT.U8 Q10,Q6,Q9 @vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) VCLT.U8 Q11,Q6,Q9 @vcltq_u8(pu1_cur_row, pu1_next_row_tmp) VSUB.U8 Q12,Q11,Q10 @sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) VADD.I8 Q13,Q0,Q7 @edge_idx = vaddq_s8(const_2, sign_up) VADD.I8 Q13,Q13,Q12 @edge_idx = vaddq_s8(edge_idx, sign_down) VTBL.8 D26,{D6},D26 @vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) VTBL.8 D27,{D6},D27 @vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) VAND Q13,Q13,Q4 @edge_idx = vandq_s8(edge_idx, au1_mask) VNEG.S8 Q7,Q12 @sign_up = vnegq_s8(sign_down) VEXT.8 Q7,Q7,Q7,#1 @sign_up = vextq_s8(sign_up, sign_up, 1) VTBL.8 D24,{D7},D26 @offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) VMOVL.U8 Q14,D12 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) VADDW.S8 Q14,Q14,D24 @pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) VMAX.S16 Q14,Q14,Q1 @pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) VMIN.U16 Q14,Q14,Q2 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vminq_u16(vreinterpretq_u16_s16(pi2_tmp_cur_row.val[0]), const_max_clip)) VMOVN.I16 D30,Q14 @vmovn_s16(pi2_tmp_cur_row.val[0]) VST1.8 {D30},[r0],r1 @vst1q_u8(pu1_src_cpy, pu1_cur_row) VMOV Q6,Q8 @pu1_cur_row = pu1_next_row SUBS r7,r7,#1 BNE PU1_SRC_LOOP_RESIDUE LDR r8,[sp,#ht_offset] @Loads ht LDR r2,[sp,#152] @Loads *pu1_src_left ADD r5,sp,#66 @*au1_src_left_tmp SRC_LEFT_LOOP_RESIDUE: LDR r7,[r5],#4 @au1_src_left_tmp[row] SUBS r8,r8,#4 STR r7,[r2],#4 @pu1_src_left[row] = au1_src_left_tmp[row] BNE SRC_LEFT_LOOP_RESIDUE RE_ASSINING_LOOP: LDR r7,[sp,#wd_offset] @Loads wd LDR r0,[sp,#148] @Loads *pu1_src LDR r11,[sp,#ht_offset] @Loads ht ADD r8,r0,r7 @pu1_src[wd] LDR r4,[sp,#pu1_src_top_left_offset] @Loads pu1_src_top_left SUB r11,r11,#1 @ht - 1 STRB r9,[r8,#-1] @pu1_src_org[wd - 1] = u1_pos_wd_0_tmp MLA r6,r11,r1,r0 @pu1_src_org[(ht - 1) * src_strd] LDRB r8,[sp] @load u1_src_top_left_tmp from stack pointer ADD r12,sp,#2 STRB r10,[r6] @pu1_src_org[wd - 1 + (ht - 1) * src_strd] = u1_pos_wd_ht_tmp STRB r8,[r4] @*pu1_src_top_left = u1_src_top_left_tmp LDR r3,[sp,#156] @Loads pu1_src_top SRC_TOP_LOOP: VLD1.8 D0,[r12]! @pu1_src_top[col] = au1_src_top_tmp[col] SUBS r7,r7,#8 @Decrement the width VST1.8 D0,[r3]! @pu1_src_top[col] = au1_src_top_tmp[col] BNE SRC_TOP_LOOP END_LOOPS: ADD sp,sp,#160 vpop {d8 - d15} LDMFD sp!,{r4-r12,r15} @Reload the registers from SP