//****************************************************************************** //* //* Copyright (C) 2015 The Android Open Source Project //* //* 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. //* //***************************************************************************** //* Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore //*/ ///* ////---------------------------------------------------------------------------- //// File Name : impeg2_inter_pred.s //// //// Description : This file has motion compensation related //// interpolation functions on Neon + CortexA-8 platform //// //// Reference Document : //// //// Revision History : //// Date Author Detail Description //// ------------ ---------------- ---------------------------------- //// 18 jun 2010 S Hamsalekha Created //// ////------------------------------------------------------------------------- //*/ ///* //// ---------------------------------------------------------------------------- //// Include Files //// ---------------------------------------------------------------------------- //*/ // PRESERVE8 .text .include "impeg2_neon_macros.s" ///* //// ---------------------------------------------------------------------------- //// Struct/Union Types and Define //// ---------------------------------------------------------------------------- //*/ ///* //// ---------------------------------------------------------------------------- //// Static Global Data section variables //// ---------------------------------------------------------------------------- //*/ //// -------------------------- NONE -------------------------------------------- ///* //// ---------------------------------------------------------------------------- //// Static Prototype Functions //// ---------------------------------------------------------------------------- //*/ //// -------------------------- NONE -------------------------------------------- ///* //// ---------------------------------------------------------------------------- //// Exported functions //// ---------------------------------------------------------------------------- //*/ ///* ////--------------------------------------------------------------------------- //// Function Name : impeg2_copy_mb_av8() //// //// Detail Description : Copies one MB worth of data from src to the dst //// //// Inputs : x0 - pointer to src //// x1 - pointer to dst //// x2 - source width //// x3 - destination width //// Registers Used : v0, v1 //// //// Stack Usage : 64 bytes //// //// Outputs : //// //// Return Data : None //// //// Programming Note : ////----------------------------------------------------------------------------- //*/ .global impeg2_copy_mb_av8 impeg2_copy_mb_av8: //STMFD x13!,{x4,x5,x12,x14} push_v_regs ldr x4, [x0] //src->y ldr x5, [x1] //dst->y //Read one row of data from the src ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ////Repeat 15 times for y ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b, v1.8b}, [x4], x2 //Load and increment src st1 {v0.8b, v1.8b}, [x5], x3 //Store and increment dst lsr x2, x2, #1 //src_offset /= 2 lsr x3, x3, #1 //dst_offset /= 2 ldr x4, [x0, #8] //src->u ldr x5, [x1, #8] //dst->u //Read one row of data from the src ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ////Repeat 7 times for u ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ldr x4, [x0, #16] //src->v ldr x5, [x1, #16] //dst->v //Read one row of data from the src ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ////Repeat 7 times for v ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst ld1 {v0.8b}, [x4], x2 //Load and increment src st1 {v0.8b}, [x5], x3 //Store and increment dst //LDMFD x13!,{x4,x5,x12,PC} pop_v_regs ret ///* ////--------------------------------------------------------------------------- //// Function Name : impeg2_mc_fullx_halfy_8x8_av8() //// //// Detail Description : This function pastes the reference block in the //// current frame buffer.This function is called for //// blocks that are not coded and have motion vectors //// with a half pel resolution. //// //// Inputs : x0 - out : Current Block Pointer //// x1 - ref : Refernce Block Pointer //// x2 - ref_wid : Refernce Block Width //// x3 - out_wid @ Current Block Width //// //// Registers Used : x14, D0-D9 //// //// Stack Usage : 64 bytes //// //// Outputs : The Motion Compensated Block //// //// Return Data : None //// //// Programming Note : ////----------------------------------------------------------------------------- //*/ .global impeg2_mc_fullx_halfy_8x8_av8 impeg2_mc_fullx_halfy_8x8_av8: //STMFD x13!,{x12,x14} push_v_regs add x14, x1, x2 lsl x2, x2, #1 ///* Load 8 + 1 rows from reference block */ ///* Do the addition with out rounding off as rounding value is 1 */ ld1 {v0.8b}, [x1], x2 //// first row hence x1 = D0 ld1 {v2.8b}, [x14], x2 //// second row hence x2 = D2 ld1 {v4.8b}, [x1], x2 //// third row hence x3 = D4 ld1 {v6.8b}, [x14], x2 //// fourth row hence x4 = D6 ld1 {v1.8b}, [x1], x2 //// fifth row hence x5 = D1 ld1 {v3.8b}, [x14], x2 //// sixth row hence x6 = D3 urhadd v9.8b, v1.8b , v6.8b //// estimated row 4 = D9 ld1 {v5.8b}, [x1], x2 //// seventh row hence x7 = D5 urhadd v0.16b, v0.16b , v2.16b //// estimated row 1 = D0, row 5 = D1 urhadd v1.16b, v1.16b , v3.16b //// estimated row 1 = D0, row 5 = D1 ld1 {v7.8b}, [x14], x2 //// eighth row hence x8 = D7 urhadd v2.16b, v2.16b , v4.16b //// estimated row 2 = D2, row 6 = D3 urhadd v3.16b, v3.16b , v5.16b //// estimated row 2 = D2, row 6 = D3 ld1 {v8.8b}, [x1], x2 //// ninth row hence x9 = D8 urhadd v4.16b, v4.16b , v6.16b //// estimated row 3 = D4, row 7 = D5 urhadd v5.16b, v5.16b , v7.16b //// estimated row 3 = D4, row 7 = D5 add x14, x0, x3 lsl x3, x3, #1 ///* Store the eight rows calculated above */ st1 {v2.8b}, [x14], x3 //// second row hence D2 urhadd v7.8b, v7.8b , v8.8b //// estimated row 8 = D7 st1 {v0.8b}, [x0], x3 //// first row hence D0 st1 {v9.8b}, [x14], x3 //// fourth row hence D9 st1 {v4.8b}, [x0], x3 //// third row hence D4 st1 {v3.8b}, [x14], x3 //// sixth row hence x6 = D3 st1 {v1.8b}, [x0], x3 //// fifth row hence x5 = D1 st1 {v7.8b}, [x14], x3 //// eighth row hence x8 = D7 st1 {v5.8b}, [x0], x3 //// seventh row hence x7 = D5 // LDMFD sp!,{x12,pc} pop_v_regs ret ///* ////--------------------------------------------------------------------------- //// Function Name : impeg2_mc_halfx_fully_8x8_av8() //// //// Detail Description : This function pastes the reference block in the //// current frame buffer.This function is called for //// blocks that are not coded and have motion vectors //// with a half pel resolutionand VopRoundingType is 0 .. //// //// Inputs : x0 - out : Current Block Pointer //// x1 - ref : Refernce Block Pointer //// x2 - ref_wid : Refernce Block Width //// x3 - out_wid @ Current Block Width //// //// Registers Used : x12, x14, v0-v10, v12-v14, v16-v18, v20-v22 //// //// Stack Usage : 64 bytes //// //// Outputs : The Motion Compensated Block //// //// Return Data : None //// //// Programming Note : ////----------------------------------------------------------------------------- //*/ .global impeg2_mc_halfx_fully_8x8_av8 impeg2_mc_halfx_fully_8x8_av8: // STMFD sp!,{x12,x14} push_v_regs add x14, x1, x2, lsl #2 add x12, x0, x3, lsl#2 ld1 {v0.8b, v1.8b}, [x1], x2 //load 16 pixels of row1 ld1 {v2.8b, v3.8b}, [x14], x2 // row5 ld1 {v4.8b, v5.8b}, [x1], x2 //load 16 pixels row2 ld1 {v6.8b, v7.8b}, [x14], x2 //row6 ext v8.8b, v0.8b , v1.8b , #1 ext v12.8b, v2.8b , v3.8b , #1 ext v16.8b, v4.8b , v5.8b , #1 ext v20.8b, v6.8b , v7.8b , #1 ld1 {v9.8b, v10.8b}, [x1], x2 //load row3 ld1 {v13.8b, v14.8b}, [x14], x2 //load row7 ld1 {v17.8b, v18.8b}, [x1], x2 //load row4 ld1 {v21.8b, v22.8b}, [x14], x2 //load row8 ext v1.8b, v9.8b , v10.8b , #1 ext v3.8b, v13.8b , v14.8b , #1 ext v5.8b, v17.8b , v18.8b , #1 ext v7.8b, v21.8b , v22.8b , #1 urhadd v0.16b, v0.16b , v8.16b //operate on row1 and row3 urhadd v1.16b, v1.16b , v9.16b //operate on row1 and row3 urhadd v2.16b, v2.16b , v12.16b //operate on row5 and row7 urhadd v3.16b, v3.16b , v13.16b //operate on row5 and row7 urhadd v4.16b, v4.16b , v16.16b //operate on row2 and row4 urhadd v5.16b, v5.16b , v17.16b //operate on row2 and row4 urhadd v6.16b, v6.16b , v20.16b //operate on row6 and row8 urhadd v7.16b, v7.16b , v21.16b //operate on row6 and row8 st1 {v0.8b}, [x0], x3 //store row1 st1 {v2.8b}, [x12], x3 //store row5 st1 {v4.8b}, [x0], x3 //store row2 st1 {v6.8b}, [x12], x3 //store row6 st1 {v1.8b}, [x0], x3 //store row3 st1 {v3.8b}, [x12], x3 //store row7 st1 {v5.8b}, [x0], x3 //store row4 st1 {v7.8b}, [x12], x3 //store row8 // LDMFD sp!,{x12,pc} pop_v_regs ret ///* ////--------------------------------------------------------------------------- //// Function Name : impeg2_mc_halfx_halfy_8x8_av8() //// //// Detail Description : This function pastes the reference block in the //// current frame buffer.This function is called for //// blocks that are not coded and have motion vectors //// with a half pel resolutionand VopRoundingType is 0 .. //// //// Inputs : x0 - out : Current Block Pointer //// x1 - ref : Refernce Block Pointer //// x2 - ref_wid : Refernce Block Width //// x3 - out_wid @ Current Block Width //// //// Registers Used : x14, v0-v18, v22, v24, v26, v28, v30 //// //// Stack Usage : 64 bytes //// //// Outputs : The Motion Compensated Block //// //// Return Data : None //// //// Programming Note : ////----------------------------------------------------------------------------- //*/ .global impeg2_mc_halfx_halfy_8x8_av8 impeg2_mc_halfx_halfy_8x8_av8: // STMFD sp!,{x12,x14} push_v_regs add x14, x1, x2, lsl #2 ld1 {v0.8b, v1.8b}, [x1], x2 //load 16 pixels of row1 ld1 {v2.8b, v3.8b}, [x14], x2 // row5 ld1 {v4.8b, v5.8b}, [x1], x2 //load 16 pixels row2 ld1 {v6.8b, v7.8b}, [x14], x2 //row6 ext v1.8b, v0.8b , v1.8b , #1 ext v3.8b, v2.8b , v3.8b , #1 ext v5.8b, v4.8b , v5.8b , #1 ext v7.8b, v6.8b , v7.8b , #1 ld1 {v8.8b, v9.8b}, [x1], x2 //load row3 ld1 {v10.8b, v11.8b}, [x14], x2 //load row7 ld1 {v12.8b, v13.8b}, [x1], x2 //load row4 ld1 {v14.8b, v15.8b}, [x14], x2 //load row8 ext v9.8b, v8.8b , v9.8b , #1 ld1 {v16.8b, v17.8b}, [x14], x2 //load row9 ext v11.8b, v10.8b , v11.8b , #1 ext v13.8b, v12.8b , v13.8b , #1 ext v15.8b, v14.8b , v15.8b , #1 ext v17.8b, v16.8b , v17.8b , #1 //interpolation in x direction uaddl v0.8h, v0.8b, v1.8b //operate row1 uaddl v2.8h, v2.8b, v3.8b //operate row5 uaddl v4.8h, v4.8b, v5.8b //operate row2 uaddl v6.8h, v6.8b, v7.8b //operate row6 uaddl v8.8h, v8.8b, v9.8b //operate row3 uaddl v10.8h, v10.8b, v11.8b //operate row7 uaddl v12.8h, v12.8b, v13.8b //operate row4 uaddl v14.8h, v14.8b, v15.8b //operate row8 uaddl v16.8h, v16.8b, v17.8b //operate row9 //interpolation in y direction add x14, x0, x3, lsl #2 add v18.8h, v0.8h , v4.8h //operate row1 and row2 add v26.8h, v2.8h , v6.8h //operate row5 and row6 add v20.8h, v4.8h , v8.8h //operate row2 and row3 add v28.8h, v6.8h , v10.8h //operate row6 and row7 rshrn v18.8b, v18.8h, #2 //row1 rshrn v26.8b, v26.8h, #2 //row5 rshrn v20.8b, v20.8h, #2 //row2 rshrn v28.8b, v28.8h, #2 //row6 add v22.8h, v8.8h , v12.8h //operate row3 and row4 st1 {v18.8b}, [x0], x3 //store row1 add v30.8h, v10.8h , v14.8h //operate row7 and row8 st1 {v26.8b}, [x14], x3 //store row5 add v24.8h, v12.8h , v2.8h //operate row4 and row5 st1 {v20.8b}, [x0], x3 //store row2 add v14.8h, v14.8h , v16.8h //operate row8 and row9 st1 {v28.8b}, [x14], x3 //store row6 rshrn v22.8b, v22.8h, #2 //row3 rshrn v30.8b, v30.8h, #2 //row7 rshrn v24.8b, v24.8h, #2 //row4 rshrn v14.8b, v14.8h, #2 //row8 st1 {v22.8b}, [x0], x3 //store row3 st1 {v30.8b}, [x14], x3 //store row7 st1 {v24.8b}, [x0], x3 //store row4 st1 {v14.8b}, [x14], x3 //store row8 // LDMFD sp!,{x12,pc} pop_v_regs ret ///* ////--------------------------------------------------------------------------- //// Function Name : impeg2_mc_fullx_fully_8x8_av8() //// //// Detail Description : This function pastes the reference block in the //// current frame buffer.This function is called for //// blocks that are not coded and have motion vectors //// with a half pel resolutionand .. //// //// Inputs : x0 - out : Current Block Pointer //// x1 - ref : Refernce Block Pointer //// x2 - ref_wid : Refernce Block Width //// x3 - out_wid @ Current Block Width //// //// Registers Used : x12, x14, v0-v3 //// //// Stack Usage : 64 bytes //// //// Outputs : The Motion Compensated Block //// //// Return Data : None //// //// Programming Note : ////----------------------------------------------------------------------------- //*/ .global impeg2_mc_fullx_fully_8x8_av8 impeg2_mc_fullx_fully_8x8_av8: // STMFD sp!,{x12,x14} push_v_regs add x14, x1, x2, lsl #2 add x12, x0, x3, lsl #2 ld1 {v0.8b}, [x1], x2 //load row1 ld1 {v1.8b}, [x14], x2 //load row4 ld1 {v2.8b}, [x1], x2 //load row2 ld1 {v3.8b}, [x14], x2 //load row5 st1 {v0.8b}, [x0], x3 //store row1 st1 {v1.8b}, [x12], x3 //store row4 st1 {v2.8b}, [x0], x3 //store row2 st1 {v3.8b}, [x12], x3 //store row5 ld1 {v0.8b}, [x1], x2 //load row3 ld1 {v1.8b}, [x14], x2 //load row6 ld1 {v2.8b}, [x1], x2 //load row4 ld1 {v3.8b}, [x14], x2 //load row8 st1 {v0.8b}, [x0], x3 //store row3 st1 {v1.8b}, [x12], x3 //store row6 st1 {v2.8b}, [x0], x3 //store row4 st1 {v3.8b}, [x12], x3 //store row8 // LDMFD sp!,{x12,pc} pop_v_regs ret ///* ////--------------------------------------------------------------------------- //// Function Name : impeg2_interpolate_av8() //// //// Detail Description : interpolates two buffers and adds pred //// //// Inputs : x0 - pointer to src1 //// x1 - pointer to src2 //// x2 - dest buf //// x3 - dst stride //// Registers Used : x12, v0-v15 //// //// Stack Usage : 64 bytes //// //// Outputs : The Motion Compensated Block //// //// Return Data : None //// //// Programming Note : ////----------------------------------------------------------------------------- //*/ .global impeg2_interpolate_av8 impeg2_interpolate_av8: //STMFD x13!,{x4-x7,x12,x14} push_v_regs ldr x4, [x0, #0] //ptr_y src1 ldr x5, [x1, #0] //ptr_y src2 ldr x7, [x2, #0] //ptr_y dst buf mov x12, #4 //counter for number of blocks interp_lumablocks_stride: ld1 {v0.16b}, [x4], #16 //row1 src1 ld1 {v2.16b}, [x4], #16 //row2 src1 ld1 {v4.16b}, [x4], #16 //row3 src1 ld1 {v6.16b}, [x4], #16 //row4 src1 ld1 {v8.16b}, [x5], #16 //row1 src2 ld1 {v10.16b}, [x5], #16 //row2 src2 ld1 {v12.16b}, [x5], #16 //row3 src2 ld1 {v14.16b}, [x5], #16 //row4 src2 urhadd v0.16b, v0.16b , v8.16b //operate on row1 urhadd v2.16b, v2.16b , v10.16b //operate on row2 urhadd v4.16b, v4.16b , v12.16b //operate on row3 urhadd v6.16b, v6.16b , v14.16b //operate on row4 st1 {v0.16b}, [x7], x3 //row1 st1 {v2.16b}, [x7], x3 //row2 st1 {v4.16b}, [x7], x3 //row3 st1 {v6.16b}, [x7], x3 //row4 subs x12, x12, #1 bne interp_lumablocks_stride lsr x3, x3, #1 //stride >> 1 ldr x4, [x0, #8] //ptr_u src1 ldr x5, [x1, #8] //ptr_u src2 ldr x7 , [x2, #8] //ptr_u dst buf mov x12, #2 //counter for number of blocks //chroma blocks interp_chromablocks_stride: ld1 {v0.8b, v1.8b}, [x4], #16 //row1 & 2 src1 ld1 {v2.8b, v3.8b}, [x4], #16 //row3 & 4 src1 ld1 {v4.8b, v5.8b}, [x4], #16 //row5 & 6 src1 ld1 {v6.8b, v7.8b}, [x4], #16 //row7 & 8 src1 ld1 {v8.8b, v9.8b}, [x5], #16 //row1 & 2 src2 ld1 {v10.8b, v11.8b}, [x5], #16 //row3 & 4 src2 ld1 {v12.8b, v13.8b}, [x5], #16 //row5 & 6 src2 ld1 {v14.8b, v15.8b}, [x5], #16 //row7 & 8 src2 urhadd v0.16b, v0.16b , v8.16b //operate on row1 & 2 urhadd v1.16b, v1.16b , v9.16b //operate on row1 & 2 urhadd v2.16b, v2.16b , v10.16b //operate on row3 & 4 urhadd v3.16b, v3.16b , v11.16b //operate on row3 & 4 urhadd v4.16b, v4.16b , v12.16b //operate on row5 & 6 urhadd v5.16b, v5.16b , v13.16b //operate on row5 & 6 urhadd v6.16b, v6.16b , v14.16b //operate on row7 & 8 urhadd v7.16b, v7.16b , v15.16b //operate on row7 & 8 st1 {v0.8b}, [x7], x3 //row1 st1 {v1.8b}, [x7], x3 //row2 st1 {v2.8b}, [x7], x3 //row3 st1 {v3.8b}, [x7], x3 //row4 st1 {v4.8b}, [x7], x3 //row5 st1 {v5.8b}, [x7], x3 //row6 st1 {v6.8b}, [x7], x3 //row7 st1 {v7.8b}, [x7], x3 //row8 ldr x4, [x0, #16] //ptr_v src1 ldr x5, [x1, #16] //ptr_v src2 ldr x7, [x2, #16] //ptr_v dst buf subs x12, x12, #1 bne interp_chromablocks_stride //LDMFD x13!,{x4-x7,x12,PC} pop_v_regs ret