diff options
Diffstat (limited to 'common/arm64/ihevc_sao_edge_offset_class2_chroma.s')
-rw-r--r-- | common/arm64/ihevc_sao_edge_offset_class2_chroma.s | 174 |
1 files changed, 87 insertions, 87 deletions
diff --git a/common/arm64/ihevc_sao_edge_offset_class2_chroma.s b/common/arm64/ihevc_sao_edge_offset_class2_chroma.s index 2fa7c22..8e286b4 100644 --- a/common/arm64/ihevc_sao_edge_offset_class2_chroma.s +++ b/common/arm64/ihevc_sao_edge_offset_class2_chroma.s @@ -76,7 +76,7 @@ ihevc_sao_edge_offset_class2_chroma_av8: ldr x9,[sp,#8] ldr w10,[sp,#16] ldr w11,[sp,#24] - push_v_regs + // STMFD sp!, {x4-x12, x14} //stack stores the values of the arguments @@ -322,7 +322,7 @@ PU1_AVAIL_3_LOOP: LDR x2, [x2, #:got_lo12:gi1_table_edge_idx] MOV x6,x7 //move wd to x6 loop_count - movi v8.16b, #0XFF //au1_mask = vdupq_n_s8(-1) + movi v1.16b, #0XFF //au1_mask = vdupq_n_s8(-1) CMP x7,#16 //Compare wd with 16 BLT WIDTH_RESIDUE //If not jump to WIDTH_RESIDUE where loop is unrolled for 8 case @@ -338,19 +338,19 @@ WIDTH_LOOP_16: MOV x20,#-1 csel x8, x20, x8,NE - mov v8.8b[0], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) + mov v1.8b[0], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) CMP x6,#16 //if(col == 16) - mov v8.8b[1], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) + mov v1.8b[1], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) BNE SKIP_AU1_MASK_VAL LDRB w8,[x5,#1] //pu1_avail[1] - mov v8.16b[14], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) - mov v8.16b[15], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) + mov v1.16b[14], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) + mov v1.16b[15], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) SKIP_AU1_MASK_VAL: LDRB w9,[x5,#2] //pu1_avail[2] - LD1 {v12.16b},[x0] //pu1_cur_row = vld1q_u8(pu1_src) + LD1 {v5.16b},[x0] //pu1_cur_row = vld1q_u8(pu1_src) //LD1 {v13.8b},[x0] //pu1_cur_row = vld1q_u8(pu1_src) //SUB x0, x0,#8 CMP x9,#0 @@ -366,17 +366,17 @@ SKIP_AU1_MASK_VAL: ADD x3,x3,#16 ADD x5,sp,#0x4B //*au1_src_left_tmp - LD1 {v10.16b},[x8] //pu1_top_row = vld1q_u8(pu1_src - src_strd - 2) || vld1q_u8(pu1_src_top_cpy - 2) + LD1 {v3.16b},[x8] //pu1_top_row = vld1q_u8(pu1_src - src_strd - 2) || vld1q_u8(pu1_src_top_cpy - 2) //LD1 {v11.8b},[x8] //pu1_top_row = vld1q_u8(pu1_src - src_strd - 2) || vld1q_u8(pu1_src_top_cpy - 2) //SUB x8, x8,#8 SUB x7,x7,x6 //(wd - col) ADD x7,x7,#14 //15 + (wd - col) - cmhi v14.16b, v12.16b , v10.16b //vcgtq_u8(pu1_cur_row, pu1_top_row) + cmhi v17.16b, v5.16b , v3.16b //vcgtq_u8(pu1_cur_row, pu1_top_row) mov x8, x26 //Loads *pu1_src ADD x7,x8,x7 //pu1_src[0 * src_strd + 15 + (wd - col)] - cmhi v16.16b, v10.16b , v12.16b //vcltq_u8(pu1_cur_row, pu1_top_row) + cmhi v16.16b, v3.16b , v5.16b //vcltq_u8(pu1_cur_row, pu1_top_row) AU1_SRC_LEFT_LOOP: LDRH w8,[x7] //load the value and increment by src_strd @@ -388,7 +388,7 @@ AU1_SRC_LEFT_LOOP: BNE AU1_SRC_LEFT_LOOP ADD x8,x0,x1 //I *pu1_src + src_strd - SUB v14.16b, v16.16b , v14.16b //sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) + SUB v17.16b, v16.16b , v17.16b //sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) MOV x7,x12 //row count, move ht_tmp to x7 LD1 {v16.16b},[x8] //I pu1_next_row = vld1q_u8(pu1_src_cpy + src_strd) @@ -430,35 +430,35 @@ AU1_SRC_LEFT_LOOP: csel x8, x20, x8,GT //I SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]) CMP x4,#0 //I - mov v14.8b[0], w8 //I sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) + mov v17.8b[0], w8 //I sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) movn x20,#0 csel x4, x20, x4,LT //I MOV x20,#1 csel x4, x20, x4,GT //I SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]) - mov v14.8b[1], w4 //I sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) + mov v17.8b[1], w4 //I sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) SIGN_UP_CHANGE_DONE: LD1 {v30.8b},[x2] //edge_idx_tbl = vld1_s8(gi1_table_edge_idx) - cmhi v20.16b, v12.16b , v18.16b //I vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) + cmhi v20.16b, v5.16b , v18.16b //I vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) - cmhi v22.16b, v18.16b , v12.16b //I vcltq_u8(pu1_cur_row, pu1_next_row_tmp) + cmhi v22.16b, v18.16b , v5.16b //I vcltq_u8(pu1_cur_row, pu1_next_row_tmp) SUB v22.16b, v22.16b , v20.16b //I sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) - ADD v18.16b, v0.16b , v14.16b //I edge_idx = vaddq_s8(const_2, sign_up) + ADD v18.16b, v0.16b , v17.16b //I edge_idx = vaddq_s8(const_2, sign_up) ADD v18.16b, v18.16b , v22.16b //I edge_idx = vaddq_s8(edge_idx, sign_down) TBL v18.16b, {v30.16b},v18.16b //I vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) - NEG v14.16b, v22.16b //I sign_up = vnegq_s8(sign_down) + NEG v17.16b, v22.16b //I sign_up = vnegq_s8(sign_down) //TBL v19.8b, {v30.16b},v19.8b //I vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) - EXT v14.16b, v14.16b , v14.16b,#14 //I sign_up = vextq_s8(sign_up, sign_up, 14) + EXT v17.16b, v17.16b , v17.16b,#14 //I sign_up = vextq_s8(sign_up, sign_up, 14) - Uxtl v20.8h, v12.8b //I pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) - AND v22.16b, v18.16b , v8.16b //I edge_idx = vandq_s8(edge_idx, au1_mask) + Uxtl v20.8h, v5.8b //I pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) + AND v22.16b, v18.16b , v1.16b //I edge_idx = vandq_s8(edge_idx, au1_mask) mov v23.d[0],v22.d[1] - Uxtl2 v18.8h, v12.16b //I pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) + Uxtl2 v18.8h, v5.16b //I pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) UZP1 v31.8b, v22.8b, v23.8b UZP2 v23.8b, v22.8b, v23.8b //I mov v22.8b,v31.8b @@ -469,7 +469,7 @@ SIGN_UP_CHANGE_DONE: ZIP2 v23.8b, v22.8b, v23.8b //I mov v22.8b,v31.8b - mov v12.16b, v16.16b //I pu1_cur_row = pu1_next_row + mov v5.16b, v16.16b //I pu1_cur_row = pu1_next_row SADDW v20.8h, v20.8h , v22.8b //I pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) 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) @@ -521,17 +521,17 @@ PU1_SRC_LOOP: movn x20,#0 csel x8, x20, x8,LT //II - cmhi v22.16b, v12.16b , v28.16b //II vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) + cmhi v22.16b, v5.16b , v28.16b //II vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) MOV x20,#1 csel x8, x20, x8,GT //II SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]) sub x13,x9,#1 LDRB w5,[x13] //II load the value - mov v14.8b[0], w8 //II sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) + mov v17.8b[0], w8 //II sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) SUB x7,x7,#1 //II Decrement the ht_tmp loop count by 1 SUB x11,x11,x5 //II pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1] - cmhi v24.16b, v28.16b , v12.16b //II vcltq_u8(pu1_cur_row, pu1_next_row_tmp) + cmhi v24.16b, v28.16b , v5.16b //II vcltq_u8(pu1_cur_row, pu1_next_row_tmp) CMP x11,#0 //II movn x20,#0 @@ -545,11 +545,11 @@ PU1_SRC_LOOP: SUB x5,x12,x7 //III ht_tmp - row ADD x10,x0,x1 - mov v14.8b[1], w11 //II sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) + mov v17.8b[1], w11 //II sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) LSL x5,x5,#1 //III (ht_tmp - row) * 2 ADD x9,x14,x5 //III pu1_src_left_cpy[(ht_tmp - row) * 2] - ADD v26.16b, v0.16b , v14.16b //II edge_idx = vaddq_s8(const_2, sign_up) + ADD v26.16b, v0.16b , v17.16b //II edge_idx = vaddq_s8(const_2, sign_up) LDRB w10,[x10,#1] //III pu1_src_cpy[0] sub x13,x9,#2 @@ -562,24 +562,24 @@ PU1_SRC_LOOP: sub x13,x9,#1 LDRB w9,[x13] //III load the value TBL v26.16b, {v22.16b},v26.16b //II vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) - NEG v14.16b, v24.16b //II sign_up = vnegq_s8(sign_down) + NEG v17.16b, v24.16b //II sign_up = vnegq_s8(sign_down) movn x20,#0 csel x4, x20, x4,LT //III SUB x10,x10,x9 //III pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1] //TBL v27.8b, {v22.16b},v27.8b //II vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) - EXT v14.16b, v14.16b , v14.16b,#14 //II sign_up = vextq_s8(sign_up, sign_up, 14) + EXT v17.16b, v17.16b , v17.16b,#14 //II sign_up = vextq_s8(sign_up, sign_up, 14) MOV x20,#1 csel x4, x20, x4,GT //III SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]) - AND v26.16b, v26.16b , v8.16b //II edge_idx = vandq_s8(edge_idx, au1_mask) + AND v26.16b, v26.16b , v1.16b //II edge_idx = vandq_s8(edge_idx, au1_mask) CMP x10,#0 //III mov v27.d[0],v26.d[1] UZP1 v31.8b, v26.8b, v27.8b UZP2 v27.8b, v26.8b, v27.8b //II mov v26.8b,v31.8b - mov v14.8b[0], w4 //III sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) + mov v17.8b[0], w4 //III sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) movn x20,#0 csel x10, x20, x10,LT //III @@ -592,13 +592,13 @@ PU1_SRC_LOOP: TBL v25.8b, {v7.16b},v27.8b //II SUB v22.16b, v22.16b , v20.16b //III sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) - mov v14.8b[1], w10 //III sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) + mov v17.8b[1], w10 //III sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) ZIP1 v31.8b, v24.8b, v25.8b ZIP2 v25.8b, v24.8b, v25.8b //II mov v24.8b,v31.8b - Uxtl v28.8h, v12.8b //II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) - ADD v18.16b, v0.16b , v14.16b //III edge_idx = vaddq_s8(const_2, sign_up) + Uxtl v28.8h, v5.8b //II pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) + ADD v18.16b, v0.16b , v17.16b //III edge_idx = vaddq_s8(const_2, sign_up) LD1 {v20.8b},[x2] //edge_idx_tbl = vld1_s8(gi1_table_edge_idx) SADDW v28.8h, v28.8h , v24.8b //II pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) @@ -608,13 +608,13 @@ PU1_SRC_LOOP: 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)) TBL v18.16b, {v20.16b},v18.16b //III vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) - NEG v14.16b, v22.16b //III sign_up = vnegq_s8(sign_down) + NEG v17.16b, v22.16b //III sign_up = vnegq_s8(sign_down) //TBL v19.8b, {v20.16b},v19.8b //III vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) - EXT v14.16b, v14.16b , v14.16b,#14 //III sign_up = vextq_s8(sign_up, sign_up, 14) + EXT v17.16b, v17.16b , v17.16b,#14 //III sign_up = vextq_s8(sign_up, sign_up, 14) - Uxtl2 v26.8h, v12.16b //II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) - AND v18.16b, v18.16b , v8.16b //III edge_idx = vandq_s8(edge_idx, au1_mask) + Uxtl2 v26.8h, v5.16b //II pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) + AND v18.16b, v18.16b , v1.16b //III edge_idx = vandq_s8(edge_idx, au1_mask) mov v19.d[0],v18.d[1] UZP1 v31.8b, v18.8b, v19.8b @@ -623,7 +623,7 @@ PU1_SRC_LOOP: TBL v22.8b, {v6.16b},v18.8b //III SADDW v26.8h, v26.8h , v25.8b //II pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) - mov v12.16b, v30.16b //III pu1_cur_row = pu1_next_row + mov v5.16b, v30.16b //III pu1_cur_row = pu1_next_row TBL v23.8b, {v7.16b},v19.8b //III 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) @@ -686,35 +686,35 @@ PU1_SRC_LOOP: LD1 {v30.8b},[x2] //edge_idx_tbl = vld1_s8(gi1_table_edge_idx) LDRB w11,[x0,#1] //pu1_src_cpy[0] - mov v14.8b[0], w8 //sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) + mov v17.8b[0], w8 //sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) sub x13,x9,#1 LDRB w5,[x13] //load the value SUB x4,x11,x5 //pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1] - cmhi v22.16b, v12.16b , v18.16b //vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) + cmhi v22.16b, v5.16b , v18.16b //vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) CMP x4,#0 movn x20,#0 csel x4, x20, x4,LT - cmhi v24.16b, v18.16b , v12.16b //vcltq_u8(pu1_cur_row, pu1_next_row_tmp) + cmhi v24.16b, v18.16b , v5.16b //vcltq_u8(pu1_cur_row, pu1_next_row_tmp) MOV x20,#1 csel x4, x20, x4,GT //SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]) - mov v14.8b[1], w4 //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) + mov v17.8b[1], w4 //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) SUB v24.16b, v24.16b , v22.16b //sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) - ADD v26.16b, v0.16b , v14.16b //edge_idx = vaddq_s8(const_2, sign_up) + ADD v26.16b, v0.16b , v17.16b //edge_idx = vaddq_s8(const_2, sign_up) ADD v26.16b, v26.16b , v24.16b //edge_idx = vaddq_s8(edge_idx, sign_down) mov v30.d[1],v30.d[0] TBL v26.16b, {v30.16b},v26.16b //vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) //TBL v27.8b, {v30.16b},v27.8b //vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) - Uxtl v20.8h, v12.8b //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) - AND v26.16b, v26.16b , v8.16b //edge_idx = vandq_s8(edge_idx, au1_mask) + Uxtl v20.8h, v5.8b //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) + AND v26.16b, v26.16b , v1.16b //edge_idx = vandq_s8(edge_idx, au1_mask) mov v27.d[0],v26.d[1] - Uxtl2 v18.8h, v12.16b //pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) + Uxtl2 v18.8h, v5.16b //pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) UZP1 v31.8b, v26.8b, v27.8b UZP2 v27.8b, v26.8b, v27.8b mov v26.8b,v31.8b @@ -771,14 +771,14 @@ WD_16_HT_4_LOOP: MOV x20,#-1 csel x8, x20, x8,NE - mov v8.8b[0], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) - mov v8.8b[1], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) + mov v1.8b[0], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) + mov v1.8b[1], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) CMP x6,#16 //if(col == 16) BNE SKIP_AU1_MASK_VAL_WD_16_HT_4 LDRB w8,[x5,#1] //pu1_avail[1] - mov v8.16b[14], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) - mov v8.16b[15], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) + mov v1.16b[14], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) + mov v1.16b[15], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) SKIP_AU1_MASK_VAL_WD_16_HT_4: LDRB w8,[x5,#2] //pu1_avail[2] @@ -788,7 +788,7 @@ SKIP_AU1_MASK_VAL_WD_16_HT_4: csel x8, x20, x8,EQ csel x8, x3, x8,NE //pu1_src_top_cpy SUB x8,x8,#2 //pu1_src - src_strd - 2 - LD1 {v10.16b},[x8] //pu1_top_row = vld1q_u8(pu1_src - src_strd - 2) || vld1q_u8(pu1_src_top_cpy - 2) + LD1 {v3.16b},[x8] //pu1_top_row = vld1q_u8(pu1_src - src_strd - 2) || vld1q_u8(pu1_src_top_cpy - 2) //LD1 {v11.8b},[x8] //pu1_top_row = vld1q_u8(pu1_src - src_strd - 2) || vld1q_u8(pu1_src_top_cpy - 2) //SUB x8, x8,#8 @@ -809,13 +809,13 @@ AU1_SRC_LEFT_LOOP_WD_16_HT_4: SUBS x4,x4,#1 //decrement the loop count BNE AU1_SRC_LEFT_LOOP_WD_16_HT_4 - LD1 {v12.16b},[x0] //pu1_cur_row = vld1q_u8(pu1_src) + LD1 {v5.16b},[x0] //pu1_cur_row = vld1q_u8(pu1_src) //LD1 {v13.8b},[x0] //pu1_cur_row = vld1q_u8(pu1_src) //SUB x0, x0,#8 - cmhi v14.16b, v12.16b , v10.16b //vcgtq_u8(pu1_cur_row, pu1_top_row) - cmhi v16.16b, v10.16b , v12.16b //vcltq_u8(pu1_cur_row, pu1_top_row) - SUB v14.16b, v16.16b , v14.16b //sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) + cmhi v17.16b, v5.16b , v3.16b //vcgtq_u8(pu1_cur_row, pu1_top_row) + cmhi v16.16b, v3.16b , v5.16b //vcltq_u8(pu1_cur_row, pu1_top_row) + SUB v17.16b, v16.16b , v17.16b //sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) movi v18.16b, #0 MOV x7,x12 //row count, move ht_tmp to x7 @@ -851,7 +851,7 @@ SIGN_UP_CHANGE_WD_16_HT_4: csel x8, x20, x8,LT MOV x20,#1 csel x8, x20, x8,GT //SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]) - mov v14.8b[0], w8 //sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) + mov v17.8b[0], w8 //sign_up = sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) LDRB w8,[x0,#1] //pu1_src_cpy[0] sub x13,x9,#1 @@ -862,25 +862,25 @@ SIGN_UP_CHANGE_WD_16_HT_4: csel x8, x20, x8,LT MOV x20,#1 csel x8, x20, x8,GT //SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]) - mov v14.8b[1], w8 //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) + mov v17.8b[1], w8 //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) SIGN_UP_CHANGE_DONE_WD_16_HT_4: - cmhi v22.16b, v12.16b , v18.16b //vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) - cmhi v24.16b, v18.16b , v12.16b //vcltq_u8(pu1_cur_row, pu1_next_row_tmp) + cmhi v22.16b, v5.16b , v18.16b //vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) + cmhi v24.16b, v18.16b , v5.16b //vcltq_u8(pu1_cur_row, pu1_next_row_tmp) SUB v24.16b, v24.16b , v22.16b //sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) - ADD v26.16b, v0.16b , v14.16b //edge_idx = vaddq_s8(const_2, sign_up) + ADD v26.16b, v0.16b , v17.16b //edge_idx = vaddq_s8(const_2, sign_up) ADD v26.16b, v26.16b , v24.16b //edge_idx = vaddq_s8(edge_idx, sign_down) LD1 {v22.8b},[x2] //edge_idx_tbl = vld1_s8(gi1_table_edge_idx) TBL v26.16b, {v22.16b},v26.16b //vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) //TBL v27.8b, {v22.16b},v27.8b //vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) - AND v26.16b, v26.16b , v8.16b //edge_idx = vandq_s8(edge_idx, au1_mask) + AND v26.16b, v26.16b , v1.16b //edge_idx = vandq_s8(edge_idx, au1_mask) mov v27.d[0],v26.d[1] - NEG v14.16b, v24.16b //sign_up = vnegq_s8(sign_down) - EXT v14.16b, v14.16b , v14.16b,#14 //sign_up = vextq_s8(sign_up, sign_up, 14) + NEG v17.16b, v24.16b //sign_up = vnegq_s8(sign_down) + EXT v17.16b, v17.16b , v17.16b,#14 //sign_up = vextq_s8(sign_up, sign_up, 14) UZP1 v31.8b, v26.8b, v27.8b UZP2 v27.8b, v26.8b, v27.8b @@ -891,12 +891,12 @@ SIGN_UP_CHANGE_DONE_WD_16_HT_4: ZIP2 v25.8b, v24.8b, v25.8b mov v24.8b,v31.8b - Uxtl v28.8h, v12.8b //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) + Uxtl v28.8h, v5.8b //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) SADDW v28.8h, v28.8h , v24.8b //pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) SMAX v28.8h, v28.8h , v2.8h //pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) 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)) - Uxtl2 v26.8h, v12.16b //pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) + Uxtl2 v26.8h, v5.16b //pi2_tmp_cur_row.val[1] = vreinterpretq_s16_u16(vmovl_u8(vget_high_u8(pu1_cur_row))) SADDW v26.8h, v26.8h , v25.8b //pi2_tmp_cur_row.val[1] = vaddw_s8(pi2_tmp_cur_row.val[1], offset) SMAX v26.8h, v26.8h , v2.8h //pi2_tmp_cur_row.val[1] = vmaxq_s16(pi2_tmp_cur_row.val[1], const_min_clip) UMIN v26.8h, v26.8h , 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)) @@ -906,7 +906,7 @@ SIGN_UP_CHANGE_DONE_WD_16_HT_4: ST1 { v28.16b},[x0],x1 //vst1q_u8(pu1_src_cpy, pu1_cur_row) - mov v12.16b, v16.16b //pu1_cur_row = pu1_next_row + mov v5.16b, v16.16b //pu1_cur_row = pu1_next_row SUBS x7,x7,#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 @@ -936,12 +936,12 @@ WIDTH_RESIDUE: MOV x20,#-1 csel x8, x20, x8,NE - mov v8.8b[0], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) - mov v8.8b[1], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) + mov v1.8b[0], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) + mov v1.8b[1], w8 //au1_mask = vsetq_lane_s8(-1, au1_mask, 0) LDRB w8,[x5,#1] //pu1_avail[1] - mov v8.8b[6], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) - mov v8.8b[7], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) + mov v1.8b[6], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) + mov v1.8b[7], w8 //au1_mask = vsetq_lane_s8(pu1_avail[1], au1_mask, 15) LDRB w8,[x5,#2] //pu1_avail[2] CMP x8,#0 @@ -950,7 +950,7 @@ WIDTH_RESIDUE: csel x8, x20, x8,EQ csel x8, x3, x8,NE SUB x8,x8,#2 //pu1_src - src_strd - 2 - LD1 {v10.16b},[x8] //pu1_top_row = vld1q_u8(pu1_src - src_strd - 2) + LD1 {v3.16b},[x8] //pu1_top_row = vld1q_u8(pu1_src - src_strd - 2) //LD1 {v11.8b},[x8] //pu1_top_row = vld1q_u8(pu1_src - src_strd - 2) //SUB x8, x8,#8 @@ -968,13 +968,13 @@ AU1_SRC_LEFT_LOOP_RESIDUE: SUBS x4,x4,#1 //decrement the loop count BNE AU1_SRC_LEFT_LOOP_RESIDUE - LD1 {v12.16b},[x0] //pu1_cur_row = vld1q_u8(pu1_src) + LD1 {v5.16b},[x0] //pu1_cur_row = vld1q_u8(pu1_src) //LD1 {v13.8b},[x0] //pu1_cur_row = vld1q_u8(pu1_src) //SUB x0, x0,#8 - cmhi v14.16b, v12.16b , v10.16b //vcgtq_u8(pu1_cur_row, pu1_top_row) - cmhi v16.16b, v10.16b , v12.16b //vcltq_u8(pu1_cur_row, pu1_top_row) - SUB v14.16b, v16.16b , v14.16b //sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) + cmhi v17.16b, v5.16b , v3.16b //vcgtq_u8(pu1_cur_row, pu1_top_row) + cmhi v16.16b, v3.16b , v5.16b //vcltq_u8(pu1_cur_row, pu1_top_row) + SUB v17.16b, v16.16b , v17.16b //sign_up = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) MOV x7,x12 //row count, move ht_tmp to x7 PU1_SRC_LOOP_RESIDUE: @@ -1009,7 +1009,7 @@ SIGN_UP_CHANGE_RESIDUE: csel x8, x20, x8,LT MOV x20,#1 csel x8, x20, x8,GT //SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]) - mov v14.8b[0], w8 //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) + mov v17.8b[0], w8 //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[0] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]), sign_up, 0) LDRB w8,[x0,#1] //pu1_src_cpy[0] sub x13,x9,#1 @@ -1020,14 +1020,14 @@ SIGN_UP_CHANGE_RESIDUE: csel x8, x20, x8,LT MOV x20,#1 csel x8, x20, x8,GT //SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2]) - mov v14.8b[1], w8 //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) + mov v17.8b[1], w8 //sign_up = vsetq_lane_s8(SIGN(pu1_src_cpy[1] - pu1_src_left_cpy[(ht_tmp - 1 - row) * 2 + 1]), sign_up, 1) SIGN_UP_CHANGE_DONE_RESIDUE: - cmhi v22.16b, v12.16b , v18.16b //vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) - cmhi v24.16b, v18.16b , v12.16b //vcltq_u8(pu1_cur_row, pu1_next_row_tmp) + cmhi v22.16b, v5.16b , v18.16b //vcgtq_u8(pu1_cur_row, pu1_next_row_tmp) + cmhi v24.16b, v18.16b , v5.16b //vcltq_u8(pu1_cur_row, pu1_next_row_tmp) SUB v24.16b, v24.16b , v22.16b //sign_down = vreinterpretq_s8_u8(vsubq_u8(cmp_lt, cmp_gt)) - ADD v26.16b, v0.16b , v14.16b //edge_idx = vaddq_s8(const_2, sign_up) + ADD v26.16b, v0.16b , v17.16b //edge_idx = vaddq_s8(const_2, sign_up) ADD v26.16b, v26.16b , v24.16b //edge_idx = vaddq_s8(edge_idx, sign_down) LD1 {v22.8b},[x2] //edge_idx_tbl = vld1_s8(gi1_table_edge_idx) @@ -1035,11 +1035,11 @@ SIGN_UP_CHANGE_DONE_RESIDUE: TBL v26.16b, {v22.16b},v26.16b //vtbl1_s8(edge_idx_tbl, vget_low_s8(edge_idx)) //TBL v27.8b, {v22.16b},v27.8b //vtbl1_s8(edge_idx_tbl, vget_high_s8(edge_idx)) - AND v26.16b, v26.16b , v8.16b //edge_idx = vandq_s8(edge_idx, au1_mask) + AND v26.16b, v26.16b , v1.16b //edge_idx = vandq_s8(edge_idx, au1_mask) mov v27.d[0],v26.d[1] - NEG v14.16b, v24.16b //sign_up = vnegq_s8(sign_down) - EXT v14.16b, v14.16b , v14.16b,#14 //sign_up = vextq_s8(sign_up, sign_up, 14) + NEG v17.16b, v24.16b //sign_up = vnegq_s8(sign_down) + EXT v17.16b, v17.16b , v17.16b,#14 //sign_up = vextq_s8(sign_up, sign_up, 14) UZP1 v31.8b, v26.8b, v27.8b UZP2 v27.8b, v26.8b, v27.8b @@ -1050,7 +1050,7 @@ SIGN_UP_CHANGE_DONE_RESIDUE: ZIP2 v25.8b, v24.8b, v25.8b mov v24.8b,v31.8b - Uxtl v28.8h, v12.8b //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) + Uxtl v28.8h, v5.8b //pi2_tmp_cur_row.val[0] = vreinterpretq_s16_u16(vmovl_u8(vget_low_u8(pu1_cur_row))) SADDW v28.8h, v28.8h , v24.8b //pi2_tmp_cur_row.val[0] = vaddw_s8(pi2_tmp_cur_row.val[0], offset) SMAX v28.8h, v28.8h , v2.8h //pi2_tmp_cur_row.val[0] = vmaxq_s16(pi2_tmp_cur_row.val[0], const_min_clip) 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)) @@ -1059,7 +1059,7 @@ SIGN_UP_CHANGE_DONE_RESIDUE: ST1 {v28.8b},[x0],x1 //vst1q_u8(pu1_src_cpy, pu1_cur_row) - mov v12.16b, v16.16b //pu1_cur_row = pu1_next_row + mov v5.16b, v16.16b //pu1_cur_row = pu1_next_row SUBS x7,x7,#1 //Decrement the ht_tmp loop count by 1 BNE PU1_SRC_LOOP_RESIDUE //If not equal jump to PU1_SRC_LOOP @@ -1113,7 +1113,7 @@ END_LOOPS: ldp x23, x24,[sp],#16 ldp x21, x22,[sp],#16 ldp x19, x20,[sp],#16 - pop_v_regs + ret |