@/***************************************************************************** @* @* 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_class0.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_class0(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 @r7 => *pu1_avail @r8 => *pi1_sao_offset @r9 => wd @r10=> ht .text .p2align 2 .extern gi1_table_edge_idx .globl ihevc_sao_edge_offset_class0_a9q gi1_table_edge_idx_addr: .long gi1_table_edge_idx - ulbl1 - 8 ihevc_sao_edge_offset_class0_a9q: STMFD sp!, {r4-r12, r14} @stack stores the values of the arguments LDR r9,[sp,#60] @Loads wd LDR r4,[sp,#40] @Loads pu1_src_top_left VMOV.I8 Q1,#2 @const_2 = vdupq_n_s8(2) ADD r11,r3,r9 @pu1_src_top[wd] LDR r10,[sp,#64] @Loads ht VMOV.I16 Q2,#0 @const_min_clip = vdupq_n_s16(0) LDRB r12,[r11,#-1] @pu1_src_top[wd - 1] LDR r7,[sp,#52] @Loads pu1_avail VMOV.I16 Q3,#255 @const_max_clip = vdupq_n_u16((1 << bit_depth) - 1) LDR r14, gi1_table_edge_idx_addr @table pointer ulbl1: add r14,r14,pc LDR r8,[sp,#56] @Loads pi1_sao_offset VMOV.S8 Q4,#0xFF @au1_mask = vdupq_n_s8(-1) STRB r12,[r4] @*pu1_src_top_left = pu1_src_top[wd - 1] MOV r6,r0 @pu1_src_org VLD1.8 D10,[r14] @edge_idx_tbl = vld1_s8(gi1_table_edge_idx) SUB r4,r10,#1 @(ht - 1) MOV r12,r9 @Move wd to r12 for loop count VLD1.8 D11,[r8] @offset_tbl = vld1_s8(pi1_sao_offset) MUL r4,r4,r1 @(ht - 1) * src_strd ADD r4,r4,r0 @pu1_src[(ht - 1) * src_strd] SRC_TOP_LOOP: @wd is always multiple of 8 VLD1.8 D0,[r4]! @Load pu1_src[(ht - 1) * src_strd + col] SUBS r12,r12,#8 @Decrement the loop counter by 8 VST1.8 D0,[r3]! @Store to pu1_src_top[col] BNE SRC_TOP_LOOP ADD r6,r6,#15 @pu1_src_org[16 - 1] CMP r9,#16 @Compare wd with 16 MOV r3,r2 @pu1_src_left backup to reload later BLT WIDTH_RESIDUE @If not jump to WIDTH_RESIDUE where loop is unrolled for 8 case MOV r8,r9 @move wd to r8 for loop count WIDTH_LOOP_16: CMP r8,r9 @if(col == wd) BNE AU1_MASK_FF @jump to else part LDRB r12,[r7] @pu1_avail[0] VMOV.8 D8[0],r12 @vsetq_lane_s8(pu1_avail[0], au1_mask, 0) B SKIP_AU1_MASK_FF @Skip the else part AU1_MASK_FF: MOV r12,#0xFF @move -1 to r12 VMOV.8 D8[0],r12 @au1_mask = vsetq_lane_s8(-1, au1_mask, 0) SKIP_AU1_MASK_FF: CMP r8,#16 @If col == 16 BNE SKIP_MASKING_IF_NOT16 @If not skip masking LDRB r12,[r7,#1] @pu1_avail[1] VMOV.8 D9[7],r12 @au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) SKIP_MASKING_IF_NOT16: MOV r12,r0 @pu1_src_cpy = pu1_src MOV r4,r10 @move ht to r4 for loop count PU1_SRC_LOOP: LDRB r11,[r2] @load pu1_src_left since ht - row =0 when it comes first pu1_src_left is incremented later VLD1.8 D12,[r12]! @pu1_cur_row = vld1q_u8(pu1_src_cpy) VLD1.8 D13,[r12], r1 @pu1_cur_row = vld1q_u8(pu1_src_cpy) SUB r12,#8 SUB r5,r9,r8 @wd - col SUB r14,r10,r4 @ht - row VMOV.8 D15[7],r11 @vsetq_lane_u8(pu1_src_left[ht - row], pu1_cur_row_tmp, 15) MUL r14,r14,r1 @(ht - row) * src_strd VLD1.8 D26,[r12]! @II Iteration pu1_cur_row = vld1q_u8(pu1_src_cpy) VLD1.8 D27,[r12] @II Iteration pu1_cur_row = vld1q_u8(pu1_src_cpy) SUB r12,#8 VEXT.8 Q7,Q7,Q6,#15 @pu1_cur_row_tmp = vextq_u8(pu1_cur_row_tmp, pu1_cur_row, 15) ADD r5,r14,r5 @(ht - row) * src_strd + (wd - col) LDRB r11,[r2, #1] @II Iteration load pu1_src_left since ht - row + 1 =1 VCGT.U8 Q8,Q6,Q7 @vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp) LDRB r14,[r6,r5] @pu1_src_org[(ht - row) * src_strd + 16 - 1 + (wd - col)] SUB r4,r4,#1 VMOV.8 D29[7],r11 @II Iteration vsetq_lane_u8(pu1_src_left[ht - row], pu1_cur_row_tmp, 15) VCLT.U8 Q9,Q6,Q7 @vcltq_u8(pu1_cur_row, pu1_cur_row_tmp) SUB r12,r12,r1 @Decrement the pu1_src pointer by src_strd VSUB.I8 Q10,Q9,Q8 @sign_left = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) STRB r14,[r2],#1 @pu1_src_left[(ht - row)] = au1_src_left_tmp[(ht - row)] LDRB r11,[r12,#16] @pu1_src_cpy[16] VEXT.8 Q14,Q14,Q13,#15 @II Iteration pu1_cur_row_tmp = vextq_u8(pu1_cur_row_tmp, pu1_cur_row, 15) SUB r5,r9,r8 @II wd - col ADD r12,r12,r1 @Increment the pu1_src pointer by src_strd VMOV.8 D14[0],r11 @pu1_cur_row_tmp = vsetq_lane_u8(pu1_src_cpy[16], pu1_cur_row_tmp, 0) VCGT.U8 Q15,Q13,Q14 @II vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp) LDRB r11,[r12,#16] @II pu1_src_cpy[16] VEXT.8 Q7,Q6,Q7,#1 @pu1_cur_row_tmp = vextq_u8(pu1_cur_row, pu1_cur_row_tmp, 1) SUB r14,r10,r4 @II ht - row VCLT.U8 Q0,Q13,Q14 @II vcltq_u8(pu1_cur_row, pu1_cur_row_tmp) VMOV.8 D28[0],r11 @II pu1_cur_row_tmp = vsetq_lane_u8(pu1_src_cpy[16], pu1_cur_row_tmp, 0) SUB r12,r12,r1 @Decrement the pu1_src pointer by src_strd MUL r14,r14,r1 @II (ht - row) * src_strd VCGT.U8 Q8,Q6,Q7 @vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp) ADD r5,r14,r5 @II (ht - row) * src_strd + (wd - col) VCLT.U8 Q9,Q6,Q7 @vcltq_u8(pu1_cur_row, pu1_cur_row_tmp) VEXT.8 Q14,Q13,Q14,#1 @II pu1_cur_row_tmp = vextq_u8(pu1_cur_row, pu1_cur_row_tmp, 1) LDRB r14,[r6,r5] @II pu1_src_org[(ht - row) * src_strd + 16 - 1 + (wd - col)] VSUB.I8 Q11,Q9,Q8 @sign_right = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) SUBS r4,r4,#1 @Decrement row by 1 VADD.I8 Q7,Q1,Q10 @edge_idx = vaddq_s8(const_2, sign_left) STRB r14,[r2],#1 @II pu1_src_left[(ht - row)] = au1_src_left_tmp[(ht - row)] VADD.I8 Q7,Q7,Q11 @edge_idx = vaddq_s8(edge_idx, sign_right) VMOVL.U8 Q9,D12 @pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) VSUB.I8 Q10,Q0,Q15 @II sign_left = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) VTBL.8 D14,{D10},D14 @vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) VCGT.U8 Q15,Q13,Q14 @II vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp) VCLT.U8 Q0,Q13,Q14 @II vcltq_u8(pu1_cur_row, pu1_cur_row_tmp) VTBL.8 D15,{D10},D15 @vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) VSUB.I8 Q11,Q0,Q15 @II sign_right = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) VAND Q7,Q7,Q4 @edge_idx = vandq_s8(edge_idx, au1_mask) VTBL.8 D16,{D11},D14 @offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) VMOVL.U8 Q0,D26 @II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) VADD.I8 Q14,Q1,Q10 @II edge_idx = vaddq_s8(const_2, sign_left) VADD.I8 Q14,Q14,Q11 @II edge_idx = vaddq_s8(edge_idx, sign_right) VADDW.S8 Q9,Q9,D16 @pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) VTBL.8 D28,{D10},D28 @II vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) VMAX.S16 Q9,Q9,Q2 @pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) VTBL.8 D29,{D10},D29 @II vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) 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)) VAND Q14,Q14,Q4 @II edge_idx = vandq_s8(edge_idx, au1_mask) VTBL.8 D17,{D11},D15 @offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) VMOVL.U8 Q7,D13 @pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) VTBL.8 D30,{D11},D28 @II offset = vtbl1_s8(offset_tbl, vget_low_s8(edge_idx)) VADDW.S8 Q7,Q7,D17 @pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) VMAX.S16 Q7,Q7,Q2 @pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) VTBL.8 D31,{D11},D29 @II offset = vtbl1_s8(offset_tbl, vget_high_s8(edge_idx)) 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)) VMOVN.I16 D18,Q9 @vmovn_s16(pi2_tmp_cur_row.val[0]) VADDW.S8 Q0,Q0,D30 @II pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) VMOVN.I16 D19,Q7 @vmovn_s16(pi2_tmp_cur_row.val[1]) VMAX.S16 Q0,Q0,Q2 @II pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) VMOVL.U8 Q14,D27 @II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) 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)) VMOVN.I16 D0,Q0 @II vmovn_s16(pi2_tmp_cur_row.val[0]) VADDW.S8 Q14,Q14,D31 @II pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) VMAX.S16 Q14,Q14,Q2 @II pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) VST1.8 {D18,D19},[r12],r1 @vst1q_u8(pu1_src_cpy, pu1_cur_row) 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)) VMOVN.I16 D1,Q14 @II vmovn_s16(pi2_tmp_cur_row.val[1]) VST1.8 {D0,D1},[r12],r1 @II vst1q_u8(pu1_src_cpy, pu1_cur_row) BNE PU1_SRC_LOOP @If not equal jump to the inner loop ADD r0,r0,#16 @pu1_src += 16 SUBS r8,r8,#16 @Decrement column by 16 CMP r8,#8 @Check whether residue remains MOV r2,r3 @Reload pu1_src_left BEQ WIDTH_RESIDUE @If residue remains jump to residue loop BGT WIDTH_LOOP_16 @If not equal jump to width_loop BLT END_LOOPS @Jump to end function WIDTH_RESIDUE: SUB r6,r6,#15 AND r8,r9,#0xF @wd_rem = wd & 0xF CMP r8,#0 @Residue check BEQ END_LOOPS @No Residue jump to end function CMP r8,r9 @if(wd_rem == wd) BNE AU1_MASK_FF_RESIDUE @jump to else part LDRB r12,[r7] @pu1_avail[0] VMOV.8 D8[0],r12 @vsetq_lane_s8(pu1_avail[0], au1_mask, 0) B SKIP_AU1_MASK_FF_RESIDUE @Skip the else part AU1_MASK_FF_RESIDUE: MOV r12,#0xFF @move -s to r12 VMOV.8 D8[0],r12 @au1_mask = vsetq_lane_s8(-1, au1_mask, 0) SKIP_AU1_MASK_FF_RESIDUE: LDRB r11,[r7,#1] @pu1_avail[1] SUB r5,r9,#1 @wd - 1 MOV r4,r10 @move ht to r4 for loop count VMOV.8 D8[7],r11 @au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) MOV r12,r0 @pu1_src_cpy = pu1_src PU1_SRC_LOOP_RESIDUE: VLD1.8 D12,[r12]! @pu1_cur_row = vld1q_u8(pu1_src_cpy) VLD1.8 D13,[r12] @pu1_cur_row = vld1q_u8(pu1_src_cpy) SUB r12,#8 LDRB r11,[r2] @load pu1_src_left VMOV.8 D15[7],r11 @vsetq_lane_u8(pu1_src_left[ht - row], pu1_cur_row_tmp, 15) VEXT.8 Q7,Q7,Q6,#15 @pu1_cur_row_tmp = vextq_u8(pu1_cur_row_tmp, pu1_cur_row, 15) VCGT.U8 Q8,Q6,Q7 @vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp) VCLT.U8 Q9,Q6,Q7 @vcltq_u8(pu1_cur_row, pu1_cur_row_tmp) VSUB.I8 Q10,Q9,Q8 @sign_left = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) LDRB r11,[r12,#16] @pu1_src_cpy[16] VMOV.8 D14[0],r11 @pu1_cur_row_tmp = vsetq_lane_u8(pu1_src_cpy[16], pu1_cur_row_tmp, 0) VEXT.8 Q7,Q6,Q7,#1 @pu1_cur_row_tmp = vextq_u8(pu1_cur_row, pu1_cur_row_tmp, 1) VCGT.U8 Q8,Q6,Q7 @vcgtq_u8(pu1_cur_row, pu1_cur_row_tmp) VCLT.U8 Q9,Q6,Q7 @vcltq_u8(pu1_cur_row, pu1_cur_row_tmp) VSUB.I8 Q11,Q9,Q8 @sign_right = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) VADD.I8 Q12,Q1,Q10 @edge_idx = vaddq_s8(const_2, sign_left) VADD.I8 Q12,Q12,Q11 @edge_idx = vaddq_s8(edge_idx, sign_right) VTBL.8 D24,{D10},D24 @vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) VTBL.8 D25,{D10},D25 @vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) VAND Q12,Q12,Q4 @edge_idx = vandq_s8(edge_idx, au1_mask) VNEG.S8 Q10,Q11 @sign_left = vnegq_s8(sign_right) VEXT.8 Q10,Q10,Q11,#15 @sign_left = vextq_s8(sign_left, sign_left, 15) VTBL.8 D26,{D11},D24 @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,D26 @pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) VMAX.S16 Q14,Q14,Q2 @pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) 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)) VMOVN.I16 D28,Q14 @vmovn_s16(pi2_tmp_cur_row.val[0]) SUB r14,r10,r4 @ht - row MUL r14,r14,r1 @(ht - row) * src_strd ADD r11,r14,r5 @(ht - row) * src_strd + (wd - 1) LDRB r14,[r6, r11] @pu1_src_org[(ht - row) * src_strd + (wd - 1)] STRB r14,[r2],#1 @pu1_src_left[(ht - row)] = au1_src_left_tmp[(ht - row)] VST1.8 {D28},[r12],r1 @vst1q_u8(pu1_src_cpy, pu1_cur_row) SUBS r4,r4,#1 @Decrement row by 1 BNE PU1_SRC_LOOP_RESIDUE @If not equal jump to the pu1_src loop END_LOOPS: LDMFD sp!,{r4-r12,r15} @Reload the registers from SP