You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

1241 lines
33 KiB

///*****************************************************************************
//*
//* 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_itrans_recon_8x8_neon.s
// *
// * @brief
// * contains function definitions for single stage inverse transform
// *
// * @author
// * anand s
// *
// * @par list of functions:
// * - ihevc_itrans_recon_16x16()
// *
// * @remarks
// * none
// *
// *******************************************************************************
//*/
///**
// *******************************************************************************
// *
// * @brief
// * this function performs inverse transform and reconstruction for 8x8
// * input block
// *
// * @par description:
// * performs inverse transform and adds the prediction data and clips output
// * to 8 bit
// *
// * @param[in] pi2_src
// * input 16x16 coefficients
// *
// * @param[in] pi2_tmp
// * temporary 16x16 buffer for storing inverse
// *
// * transform
// * 1st stage output
// *
// * @param[in] pu1_pred
// * prediction 16x16 block
// *
// * @param[out] pu1_dst
// * output 8x8 block
// *
// * @param[in] src_strd
// * input stride
// *
// * @param[in] pred_strd
// * prediction stride
// *
// * @param[in] dst_strd
// * output stride
// *
// * @param[in] shift
// * output shift
// *
// * @param[in] x12
// * zero columns in pi2_src
// *
// * @returns void
// *
// * @remarks
// * none
// *
// *******************************************************************************
// */
//void ihevc_itrans_recon_16x16(word16 *pi2_src,
// word16 *pi2_tmp,
// uword8 *pu1_pred,
// uword8 *pu1_dst,
// word32 src_strd,
// word32 pred_strd,
// word32 dst_strd,
// word32 x12
// word32 x11 )
//**************variables vs registers*************************
// x0 => *pi2_src
// x1 => *pi2_tmp
// x2 => *pu1_pred
// x3 => *pu1_dst
// src_strd
// pred_strd
// dst_strd
// x12
// x11
.text
.align 4
.include "ihevc_neon_macros.s"
.set shift_stage1_idct , 7
.set shift_stage2_idct , 12
//#define zero_cols x12
//#define zero_rows x11
.globl ihevc_itrans_recon_16x16_av8
.extern g_ai2_ihevc_trans_16_transpose
.type ihevc_itrans_recon_16x16_av8, %function
ihevc_itrans_recon_16x16_av8:
ldr w11, [sp]
// stmfd sp!,{x4-x12,x14}
push_v_regs
stp x19, x20,[sp,#-16]!
stp x5, x6,[sp,#-16]!
// add sp,sp,#40
// ldr x8,[sp,#4] @ prediction stride
// ldr x7,[sp,#8] @ destination stride
mov x6, x4 // src stride
mov x12, x7
adrp x14, :got:g_ai2_ihevc_trans_16_transpose
ldr x14, [x14, #:got_lo12:g_ai2_ihevc_trans_16_transpose]
ld1 {v0.4h, v1.4h, v2.4h, v3.4h},[x14] ////d0,d1 are used for storing the constant data
mov x7,#0xffff
and x12,x12,x7
and x11,x11,x7
lsl x6, x6, #1 // x sizeof(word16)
add x9,x0,x6, lsl #1 // 2 rows
add x10,x6,x6, lsl #1 // 3 rows
add x5,x6,x6,lsl #2
mov x7,#0xfff0
cmp x12,x7
bge zero_12cols_decision
mov x19,#0xff00
cmp x12,x19
bge zero_8cols_decision
mov x14,#4
cmp x11,x7
sub x20,x6,#0
neg x20, x20
csel x10,x20,x10,ge
mov x19,#0xff00
cmp x11,x19
csel x8, x5, x8,ge
sub x20,x8,#0
neg x20, x20
csel x8,x20,x8,ge
csel x8, x10, x8,lt
add x5,x5,x6,lsl #3
sub x20,x5,#0
neg x5, x20
b first_stage_top_four_bottom_four
zero_12cols_decision:
mov x14,#1
mov x19,#0xff00
cmp x11,x19
csel x8, x5, x8,ge
csel x8, x10, x8,lt
add x5,x5,x6,lsl #3
sub x20,x5,#0
neg x5, x20
b first_stage_top_four_bottom_four
zero_8cols_decision:
mov x14,#2
mov x8,x5
sub x20,x8,#0
neg x8, x20
mov x19,#0xff00
cmp x11,x19
csel x8, x10, x8,lt
add x5,x5,x6,lsl #3
sub x20,x5,#0
neg x5, x20
cmp x11,x7
sub x20,x6,#0
neg x20, x20
csel x10,x20,x10,ge
b first_stage_top_four_bottom_four
//d0[0]= 64 d2[0]=64
//d0[1]= 90 d2[1]=57
//d0[2]= 89 d2[2]=50
//d0[3]= 87 d2[3]=43
//d1[0]= 83 d3[0]=36
//d1[1]= 80 d3[1]=25
//d1[2]= 75 d3[2]=18
//d1[3]= 70 d3[3]=9
first_stage:
add x0,x0,#8
add x9,x9,#8
first_stage_top_four_bottom_four:
ld1 {v10.4h},[x0],x6
ld1 {v11.4h},[x9],x6
ld1 {v6.4h},[x0],x10
ld1 {v7.4h},[x9],x10
cmp x11,x7
bge skip_load4rows
ld1 {v4.4h},[x0],x6
ld1 {v5.4h},[x9],x6
ld1 {v8.4h},[x0],x8
ld1 {v9.4h},[x9],x8
// registers used: q0,q1,q3,q5,q2,q4
// d10 =x0
//d6= x1
//d11=x2
//d7=x3
skip_load4rows:
smull v24.4s, v6.4h, v0.h[1] //// y1 * cos1(part of b0)
smull v26.4s, v6.4h, v0.h[3] //// y1 * cos3(part of b1)
smull v28.4s, v6.4h, v1.h[1] //// y1 * sin3(part of b2)
smull v30.4s, v6.4h, v1.h[3] //// y1 * sin1(part of b3)
smlal v24.4s, v7.4h, v0.h[3] //// y1 * cos1 + y3 * cos3(part of b0)
smlal v26.4s, v7.4h, v2.h[1] //// y1 * cos3 - y3 * sin1(part of b1)
smlal v28.4s, v7.4h, v3.h[3] //// y1 * sin3 - y3 * cos1(part of b2)
smlsl v30.4s, v7.4h, v2.h[3] //// y1 * sin1 - y3 * sin3(part of b3)
smull v12.4s, v10.4h, v0.h[0]
smlal v12.4s, v11.4h, v0.h[2]
smull v14.4s, v10.4h, v0.h[0]
smlal v14.4s, v11.4h, v1.h[2]
smull v16.4s, v10.4h, v0.h[0]
smlal v16.4s, v11.4h, v2.h[2]
smull v18.4s, v10.4h, v0.h[0]
smlal v18.4s, v11.4h, v3.h[2]
bge skip_last12rows_kernel1
smlal v24.4s, v8.4h, v1.h[1]
smlal v26.4s, v8.4h, v3.h[3]
smlsl v28.4s, v8.4h, v1.h[3]
smlsl v30.4s, v8.4h, v0.h[3]
smlal v24.4s, v9.4h, v1.h[3]
smlsl v26.4s, v9.4h, v2.h[3]
smlsl v28.4s, v9.4h, v0.h[3]
smlal v30.4s, v9.4h, v3.h[3]
smlal v12.4s, v4.4h, v1.h[0]
smlal v12.4s, v5.4h, v1.h[2]
smlal v14.4s, v4.4h, v3.h[0]
smlsl v14.4s, v5.4h, v3.h[2]
smlsl v16.4s, v4.4h, v3.h[0]
smlsl v16.4s, v5.4h, v0.h[2]
smlsl v18.4s, v4.4h, v1.h[0]
smlsl v18.4s, v5.4h, v2.h[2]
//d0[0]= 64 d2[0]=64
//d0[1]= 90 d2[1]=57
//d0[2]= 89 d2[2]=50
//d0[3]= 87 d2[3]=43
//d1[0]= 83 d3[0]=36
//d1[1]= 80 d3[1]=25
//d1[2]= 75 d3[2]=18
//d1[3]= 70 d3[3]=9
mov x19,#0xff00
cmp x11,x19
bge skip_last12rows_kernel1
ld1 {v10.4h},[x0],x6
ld1 {v11.4h},[x9],x6
ld1 {v6.4h},[x0],x10
ld1 {v7.4h},[x9],x10
ld1 {v4.4h},[x0],x6
ld1 {v5.4h},[x9],x6
ld1 {v8.4h},[x0],x5
ld1 {v9.4h},[x9],x5
smlal v24.4s, v6.4h, v2.h[1] //// y1 * cos1(part of b0)
smlsl v26.4s, v6.4h, v1.h[1] //// y1 * cos3(part of b1)
smlsl v28.4s, v6.4h, v3.h[1] //// y1 * sin3(part of b2)
smlal v30.4s, v6.4h, v0.h[1] //// y1 * sin1(part of b3)
smlal v24.4s, v7.4h, v2.h[3] //// y1 * cos1 + y3 * cos3(part of b0)
smlsl v26.4s, v7.4h, v0.h[1] //// y1 * cos3 - y3 * sin1(part of b1)
smlal v28.4s, v7.4h, v2.h[1] //// y1 * sin3 - y3 * cos1(part of b2)
smlal v30.4s, v7.4h, v3.h[1] //// y1 * sin1 - y3 * sin3(part of b3)
smlal v24.4s, v8.4h, v3.h[1]
smlsl v26.4s, v8.4h, v1.h[3]
smlal v28.4s, v8.4h, v0.h[1]
smlsl v30.4s, v8.4h, v1.h[1]
smlal v24.4s, v9.4h, v3.h[3]
smlsl v26.4s, v9.4h, v3.h[1]
smlal v28.4s, v9.4h, v2.h[3]
smlsl v30.4s, v9.4h, v2.h[1]
smlal v12.4s, v10.4h, v0.h[0]
smlal v12.4s, v11.4h, v2.h[2]
smlal v12.4s, v4.4h, v3.h[0]
smlal v12.4s, v5.4h, v3.h[2]
smlsl v14.4s, v10.4h, v0.h[0]
smlsl v14.4s, v11.4h, v0.h[2]
smlsl v14.4s, v4.4h, v1.h[0]
smlsl v14.4s, v5.4h, v2.h[2]
smlsl v16.4s, v10.4h, v0.h[0]
smlal v16.4s, v11.4h, v3.h[2]
smlal v16.4s, v4.4h, v1.h[0]
smlal v16.4s, v5.4h, v1.h[2]
smlal v18.4s, v10.4h, v0.h[0]
smlal v18.4s, v11.4h, v1.h[2]
smlsl v18.4s, v4.4h, v3.h[0]
smlsl v18.4s, v5.4h, v0.h[2]
skip_last12rows_kernel1:
add v20.4s, v12.4s , v24.4s
sub v22.4s, v12.4s , v24.4s
add v12.4s, v14.4s , v26.4s
sub v24.4s, v14.4s , v26.4s
add v14.4s, v16.4s , v28.4s
sub v26.4s, v16.4s , v28.4s
add v16.4s, v18.4s , v30.4s
sub v28.4s, v18.4s , v30.4s
sqrshrn v30.4h, v20.4s,#shift_stage1_idct //// x0 = (a0 + b0 + rnd) >> 7(shift_stage1_idct)
sqrshrn v19.4h, v22.4s,#shift_stage1_idct //// x7 = (a0 - b0 + rnd) >> 7(shift_stage1_idct)
sqrshrn v31.4h, v14.4s,#shift_stage1_idct //// x2 = (a2 + b2 + rnd) >> 7(shift_stage1_idct)
sqrshrn v18.4h, v26.4s,#shift_stage1_idct //// x5 = (a2 - b2 + rnd) >> 7(shift_stage1_idct)
sqrshrn v12.4h, v12.4s,#shift_stage1_idct //// x1 = (a1 + b1 + rnd) >> 7(shift_stage1_idct)
sqrshrn v15.4h, v24.4s,#shift_stage1_idct //// x6 = (a1 - b1 + rnd) >> 7(shift_stage1_idct)
sqrshrn v13.4h, v16.4s,#shift_stage1_idct //// x3 = (a3 + b3 + rnd) >> 7(shift_stage1_idct)
sqrshrn v14.4h, v28.4s,#shift_stage1_idct //// x4 = (a3 - b3 + rnd) >> 7(shift_stage1_idct)
st1 {v30.4h, v31.4h},[x1],#16
st1 {v18.4h, v19.4h},[x1],#16
sub x1,x1,#32
bge skip_stage1_kernel_load
first_stage_middle_eight:
ld1 {v10.4h},[x0],x6
ld1 {v11.4h},[x9],x6
ld1 {v6.4h},[x0],x10
ld1 {v7.4h},[x9],x10
ld1 {v4.4h},[x0],x6
ld1 {v5.4h},[x9],x6
ld1 {v8.4h},[x0],x8
ld1 {v9.4h},[x9],x8
skip_stage1_kernel_load:
smull v24.4s, v6.4h, v2.h[1] //// y1 * cos1(part of b0)
smull v26.4s, v6.4h, v2.h[3] //// y1 * cos3(part of b1)
smull v28.4s, v6.4h, v3.h[1] //// y1 * sin3(part of b2)
smull v30.4s, v6.4h, v3.h[3] //// y1 * sin1(part of b3)
smlsl v24.4s, v7.4h, v1.h[1] //// y1 * cos1 + y3 * cos3(part of b0)
smlsl v26.4s, v7.4h, v0.h[1] //// y1 * cos3 - y3 * sin1(part of b1)
smlsl v28.4s, v7.4h, v1.h[3] //// y1 * sin3 - y3 * cos1(part of b2)
smlsl v30.4s, v7.4h, v3.h[1] //// y1 * sin1 - y3 * sin3(part of b3)
smull v22.4s, v10.4h, v0.h[0]
smlsl v22.4s, v11.4h, v3.h[2]
smull v20.4s, v10.4h, v0.h[0]
smlsl v20.4s, v11.4h, v2.h[2]
smull v16.4s, v10.4h, v0.h[0]
smlsl v16.4s, v11.4h, v1.h[2]
smull v18.4s, v10.4h, v0.h[0]
smlsl v18.4s, v11.4h, v0.h[2]
cmp x11,x7
bge skip_last12rows_kernel2
smlsl v24.4s, v8.4h, v3.h[1]
smlal v26.4s, v8.4h, v2.h[1]
smlal v28.4s, v8.4h, v0.h[1]
smlal v30.4s, v8.4h, v2.h[3]
smlal v24.4s, v9.4h, v0.h[1]
smlal v26.4s, v9.4h, v3.h[1]
smlsl v28.4s, v9.4h, v1.h[1]
smlsl v30.4s, v9.4h, v2.h[1]
smlsl v22.4s, v4.4h, v1.h[0]
smlal v22.4s, v5.4h, v2.h[2]
smlsl v20.4s, v4.4h, v3.h[0]
smlal v20.4s, v5.4h, v0.h[2]
smlal v16.4s, v4.4h, v3.h[0]
smlal v16.4s, v5.4h, v3.h[2]
smlal v18.4s, v4.4h, v1.h[0]
smlsl v18.4s, v5.4h, v1.h[2]
//d0[0]= 64 d2[0]=64
//d0[1]= 90 d2[1]=57
//d0[2]= 89 d2[2]=50
//d0[3]= 87 d2[3]=43
//d1[0]= 83 d3[0]=36
//d1[1]= 80 d3[1]=25
//d1[2]= 75 d3[2]=18
//d1[3]= 70 d3[3]=9
mov x19,#0xff00
cmp x11,x19
bge skip_last12rows_kernel2
ld1 {v10.4h},[x0],x6
ld1 {v11.4h},[x9],x6
ld1 {v6.4h},[x0],x10
ld1 {v7.4h},[x9],x10
ld1 {v4.4h},[x0],x6
ld1 {v5.4h},[x9],x6
ld1 {v8.4h},[x0],x5
ld1 {v9.4h},[x9],x5
smlsl v24.4s, v6.4h, v3.h[3] //// y1 * cos1(part of b0)
smlsl v26.4s, v6.4h, v0.h[3] //// y1 * cos3(part of b1)
smlal v28.4s, v6.4h, v2.h[3] //// y1 * sin3(part of b2)
smlal v30.4s, v6.4h, v1.h[3] //// y1 * sin1(part of b3)
smlsl v24.4s, v7.4h, v0.h[3] //// y1 * cos1 + y3 * cos3(part of b0)
smlal v26.4s, v7.4h, v1.h[3] //// y1 * cos3 - y3 * sin1(part of b1)
smlal v28.4s, v7.4h, v3.h[3] //// y1 * sin3 - y3 * cos1(part of b2)
smlsl v30.4s, v7.4h, v1.h[1] //// y1 * sin1 - y3 * sin3(part of b3)
smlal v24.4s, v8.4h, v2.h[3]
smlal v26.4s, v8.4h, v3.h[3]
smlsl v28.4s, v8.4h, v2.h[1]
smlal v30.4s, v8.4h, v0.h[3]
smlal v24.4s, v9.4h, v1.h[3]
smlsl v26.4s, v9.4h, v1.h[1]
smlal v28.4s, v9.4h, v0.h[3]
smlsl v30.4s, v9.4h, v0.h[1]
smlal v22.4s, v10.4h, v0.h[0]
smlsl v22.4s, v11.4h, v1.h[2]
smlsl v22.4s, v4.4h, v3.h[0]
smlal v22.4s, v5.4h, v0.h[2]
smlsl v20.4s, v10.4h, v0.h[0]
smlsl v20.4s, v11.4h, v3.h[2]
smlal v20.4s, v4.4h, v1.h[0]
smlsl v20.4s, v5.4h, v1.h[2]
smlsl v16.4s, v10.4h, v0.h[0]
smlal v16.4s, v11.4h, v0.h[2]
smlsl v16.4s, v4.4h, v1.h[0]
smlal v16.4s, v5.4h, v2.h[2]
smlal v18.4s, v10.4h, v0.h[0]
smlsl v18.4s, v11.4h, v2.h[2]
smlal v18.4s, v4.4h, v3.h[0]
smlsl v18.4s, v5.4h, v3.h[2]
skip_last12rows_kernel2:
add v4.4s, v22.4s , v24.4s
sub v22.4s, v22.4s , v24.4s
add v6.4s, v20.4s , v26.4s
sub v24.4s, v20.4s , v26.4s
add v10.4s, v16.4s , v28.4s
sub v26.4s, v16.4s , v28.4s
add v16.4s, v18.4s , v30.4s
sub v28.4s, v18.4s , v30.4s
sqrshrn v18.4h, v4.4s,#shift_stage1_idct //// x0 = (a0 + b0 + rnd) >> 7(shift_stage1_idct)
sqrshrn v31.4h, v22.4s,#shift_stage1_idct //// x7 = (a0 - b0 + rnd) >> 7(shift_stage1_idct)
sqrshrn v19.4h, v10.4s,#shift_stage1_idct //// x2 = (a2 + b2 + rnd) >> 7(shift_stage1_idct)
sqrshrn v30.4h, v26.4s,#shift_stage1_idct //// x5 = (a2 - b2 + rnd) >> 7(shift_stage1_idct)
sqrshrn v20.4h, v6.4s,#shift_stage1_idct //// x1 = (a1 + b1 + rnd) >> 7(shift_stage1_idct)
sqrshrn v23.4h, v24.4s,#shift_stage1_idct //// x6 = (a1 - b1 + rnd) >> 7(shift_stage1_idct)
sqrshrn v21.4h, v16.4s,#shift_stage1_idct //// x3 = (a3 + b3 + rnd) >> 7(shift_stage1_idct)
sqrshrn v22.4h, v28.4s,#shift_stage1_idct //// x4 = (a3 - b3 + rnd) >> 7(shift_stage1_idct)
// registers used: {q2,q4,q6,q7}, {q9,q15,q10,q11}
ld1 {v4.4h, v5.4h},[x1],#16
ld1 {v8.4h, v9.4h},[x1],#16
sub x1,x1,#32
//d4=x0
//d12=x1
//d5=x2
//d13=x3
//d18=x4
//d20=x5
//d19=x6
//d21=x7
//d22=x8
//d30=x9
//d23=x10
//d31=x11
//d14=x12
//d8=x13
//d15=x14
//d9=x15
umov x15,v26.d[0]
umov x16,v27.d[0]
umov x19,v28.d[0]
umov x20,v29.d[0]
trn1 v26.4h, v4.4h, v12.4h
trn2 v27.4h, v4.4h, v12.4h
trn1 v28.4h, v5.4h, v13.4h
trn2 v29.4h, v5.4h, v13.4h
trn1 v4.2s, v26.2s, v28.2s
trn2 v5.2s, v26.2s, v28.2s
trn1 v12.2s, v27.2s, v29.2s
trn2 v13.2s, v27.2s, v29.2s
trn1 v26.4h, v18.4h, v20.4h
trn2 v27.4h, v18.4h, v20.4h
trn1 v28.4h, v19.4h, v21.4h
trn2 v29.4h, v19.4h, v21.4h
trn1 v18.2s, v26.2s, v28.2s
trn2 v19.2s, v26.2s, v28.2s
trn1 v20.2s, v27.2s, v29.2s
trn2 v21.2s, v27.2s, v29.2s
trn1 v26.4h, v22.4h, v30.4h
trn2 v27.4h, v22.4h, v30.4h
trn1 v28.4h, v23.4h, v31.4h
trn2 v29.4h, v23.4h, v31.4h
trn1 v22.2s, v26.2s, v28.2s
trn2 v23.2s, v26.2s, v28.2s
trn1 v30.2s, v27.2s, v29.2s
trn2 v31.2s, v27.2s, v29.2s
trn1 v26.4h, v14.4h, v8.4h
trn2 v27.4h, v14.4h, v8.4h
trn1 v28.4h, v15.4h, v9.4h
trn2 v29.4h, v15.4h, v9.4h
trn1 v14.2s, v26.2s, v28.2s
trn2 v15.2s, v26.2s, v28.2s
trn1 v8.2s, v27.2s, v29.2s
trn2 v9.2s, v27.2s, v29.2s
mov v26.d[0],x15
mov v27.d[0],x16
mov v28.d[0],x19
mov v29.d[0],x20
// d4 =x0 1- 4 values
// d5 =x2 1- 4 values
// d12=x1 1- 4 values
// d13=x3 1- 4 values
// d18 =x0 5- 8 values
// d19 =x2 5- 8 values
// d20=x1 5- 8 values
// d21=x3 5- 8 values
// d22 =x0 9- 12 values
// d23 =x2 9- 12 values
// d30=x1 9- 12 values
// d31=x3 9- 12 values
// d14 =x0 13-16 values
// d15 =x2 13- 16 values
// d8=x1 13- 16 values
// d9=x3 13- 16 values
st1 { v4.4h, v5.4h},[x1],#16
st1 { v12.4h, v13.4h},[x1],#16
st1 { v18.4h, v19.4h},[x1],#16
st1 { v20.4h, v21.4h},[x1],#16
st1 { v22.4h, v23.4h},[x1],#16
st1 { v30.4h, v31.4h},[x1],#16
st1 { v14.4h, v15.4h},[x1],#16
st1 { v8.4h, v9.4h},[x1],#16
subs x14,x14,#1
bne first_stage
mov x6,x7
ldp x8, x7,[sp],#16
mov x10,#16
cmp x12,x6
sub x20,x1,#128
csel x1, x20, x1,ge
bge label1
mov x19,#0xff00
cmp x12,x19
sub x20,x1,#256
csel x1, x20, x1,ge
bge label_2
sub x1,x1,#512
sub x20,x10,#0
neg x10, x20
label_2:
add x9,x1,#128
add x11,x9,#128
add x0,x11,#128
label1:
// mov x6,x1
mov x14,#4
add x4,x2,x8, lsl #1 // x4 = x2 + pred_strd * 2 => x4 points to 3rd row of pred data
add x5,x8,x8, lsl #1 //
// add x0,x3,x7, lsl #1 @ x0 points to 3rd row of dest data
// add x10,x7,x7, lsl #1 @
second_stage:
ld1 {v10.4h, v11.4h},[x1],#16
ld1 {v6.4h, v7.4h},[x1],x10
cmp x12,x6
bge second_stage_process
ld1 {v4.4h, v5.4h},[x9],#16
ld1 {v8.4h, v9.4h},[x9],x10
second_stage_process:
smull v24.4s, v6.4h, v0.h[1] //// y1 * cos1(part of b0)
smull v26.4s, v6.4h, v0.h[3] //// y1 * cos3(part of b1)
smull v28.4s, v6.4h, v1.h[1] //// y1 * sin3(part of b2)
smull v30.4s, v6.4h, v1.h[3] //// y1 * sin1(part of b3)
smlal v24.4s, v7.4h, v0.h[3] //// y1 * cos1 + y3 * cos3(part of b0)
smlal v26.4s, v7.4h, v2.h[1] //// y1 * cos3 - y3 * sin1(part of b1)
smlal v28.4s, v7.4h, v3.h[3] //// y1 * sin3 - y3 * cos1(part of b2)
smlsl v30.4s, v7.4h, v2.h[3] //// y1 * sin1 - y3 * sin3(part of b3)
smull v12.4s, v10.4h, v0.h[0]
smlal v12.4s, v11.4h, v0.h[2]
smull v14.4s, v10.4h, v0.h[0]
smlal v14.4s, v11.4h, v1.h[2]
smull v16.4s, v10.4h, v0.h[0]
smlal v16.4s, v11.4h, v2.h[2]
smull v18.4s, v10.4h, v0.h[0]
smlal v18.4s, v11.4h, v3.h[2]
bge skip_last8rows_stage2_kernel1
smlal v24.4s, v8.4h, v1.h[1]
smlal v26.4s, v8.4h, v3.h[3]
smlsl v28.4s, v8.4h, v1.h[3]
smlsl v30.4s, v8.4h, v0.h[3]
smlal v24.4s, v9.4h, v1.h[3]
smlsl v26.4s, v9.4h, v2.h[3]
smlsl v28.4s, v9.4h, v0.h[3]
smlal v30.4s, v9.4h, v3.h[3]
smlal v12.4s, v4.4h, v1.h[0]
smlal v12.4s, v5.4h, v1.h[2]
smlal v14.4s, v4.4h, v3.h[0]
smlsl v14.4s, v5.4h, v3.h[2]
smlsl v16.4s, v4.4h, v3.h[0]
smlsl v16.4s, v5.4h, v0.h[2]
smlsl v18.4s, v4.4h, v1.h[0]
smlsl v18.4s, v5.4h, v2.h[2]
mov x19,#0xff00
cmp x12,x19
bge skip_last8rows_stage2_kernel1
ld1 {v10.4h, v11.4h},[x11],#16
ld1 {v6.4h, v7.4h},[x11],x10
ld1 {v4.4h, v5.4h},[x0],#16
ld1 {v8.4h, v9.4h},[x0],x10
smlal v24.4s, v6.4h, v2.h[1] //// y1 * cos1(part of b0)
smlsl v26.4s, v6.4h, v1.h[1] //// y1 * cos3(part of b1)
smlsl v28.4s, v6.4h, v3.h[1] //// y1 * sin3(part of b2)
smlal v30.4s, v6.4h, v0.h[1] //// y1 * sin1(part of b3)
smlal v24.4s, v7.4h, v2.h[3] //// y1 * cos1 + y3 * cos3(part of b0)
smlsl v26.4s, v7.4h, v0.h[1] //// y1 * cos3 - y3 * sin1(part of b1)
smlal v28.4s, v7.4h, v2.h[1] //// y1 * sin3 - y3 * cos1(part of b2)
smlal v30.4s, v7.4h, v3.h[1] //// y1 * sin1 - y3 * sin3(part of b3)
smlal v24.4s, v8.4h, v3.h[1]
smlsl v26.4s, v8.4h, v1.h[3]
smlal v28.4s, v8.4h, v0.h[1]
smlsl v30.4s, v8.4h, v1.h[1]
smlal v24.4s, v9.4h, v3.h[3]
smlsl v26.4s, v9.4h, v3.h[1]
smlal v28.4s, v9.4h, v2.h[3]
smlsl v30.4s, v9.4h, v2.h[1]
smlal v12.4s, v10.4h, v0.h[0]
smlal v12.4s, v11.4h, v2.h[2]
smlal v12.4s, v4.4h, v3.h[0]
smlal v12.4s, v5.4h, v3.h[2]
smlsl v14.4s, v10.4h, v0.h[0]
smlsl v14.4s, v11.4h, v0.h[2]
smlsl v14.4s, v4.4h, v1.h[0]
smlsl v14.4s, v5.4h, v2.h[2]
smlsl v16.4s, v10.4h, v0.h[0]
smlal v16.4s, v11.4h, v3.h[2]
smlal v16.4s, v4.4h, v1.h[0]
smlal v16.4s, v5.4h, v1.h[2]
smlal v18.4s, v10.4h, v0.h[0]
smlal v18.4s, v11.4h, v1.h[2]
smlsl v18.4s, v4.4h, v3.h[0]
smlsl v18.4s, v5.4h, v0.h[2]
skip_last8rows_stage2_kernel1:
add v20.4s, v12.4s , v24.4s
sub v22.4s, v12.4s , v24.4s
add v12.4s, v14.4s , v26.4s
sub v24.4s, v14.4s , v26.4s
add v14.4s, v16.4s , v28.4s
sub v26.4s, v16.4s , v28.4s
add v16.4s, v18.4s , v30.4s
sub v28.4s, v18.4s , v30.4s
sqrshrn v30.4h, v20.4s,#shift_stage2_idct //// x0 = (a0 + b0 + rnd) >> 7(shift_stage1_idct)
sqrshrn v19.4h, v22.4s,#shift_stage2_idct //// x7 = (a0 - b0 + rnd) >> 7(shift_stage1_idct)
sqrshrn v31.4h, v14.4s,#shift_stage2_idct //// x2 = (a2 + b2 + rnd) >> 7(shift_stage1_idct)
sqrshrn v18.4h, v26.4s,#shift_stage2_idct //// x5 = (a2 - b2 + rnd) >> 7(shift_stage1_idct)
sqrshrn v12.4h, v12.4s,#shift_stage2_idct //// x1 = (a1 + b1 + rnd) >> 7(shift_stage1_idct)
sqrshrn v15.4h, v24.4s,#shift_stage2_idct //// x6 = (a1 - b1 + rnd) >> 7(shift_stage1_idct)
sqrshrn v13.4h, v16.4s,#shift_stage2_idct //// x3 = (a3 + b3 + rnd) >> 7(shift_stage1_idct)
sqrshrn v14.4h, v28.4s,#shift_stage2_idct //// x4 = (a3 - b3 + rnd) >> 7(shift_stage1_idct)
bge skip_stage2_kernel_load
//q2,q4,q6,q7 is used
ld1 {v10.4h, v11.4h},[x1],#16
ld1 {v6.4h, v7.4h},[x1],#16
ld1 {v4.4h, v5.4h},[x9],#16
ld1 {v8.4h, v9.4h},[x9],#16
skip_stage2_kernel_load:
sub x1,x1,#32
st1 {v30.4h, v31.4h},[x1],#16
st1 {v18.4h, v19.4h},[x1],#16
sub x1,x1,#32
smull v24.4s, v6.4h, v2.h[1] //// y1 * cos1(part of b0)
smull v26.4s, v6.4h, v2.h[3] //// y1 * cos3(part of b1)
smull v28.4s, v6.4h, v3.h[1] //// y1 * sin3(part of b2)
smull v30.4s, v6.4h, v3.h[3] //// y1 * sin1(part of b3)
smlsl v24.4s, v7.4h, v1.h[1] //// y1 * cos1 + y3 * cos3(part of b0)
smlsl v26.4s, v7.4h, v0.h[1] //// y1 * cos3 - y3 * sin1(part of b1)
smlsl v28.4s, v7.4h, v1.h[3] //// y1 * sin3 - y3 * cos1(part of b2)
smlsl v30.4s, v7.4h, v3.h[1] //// y1 * sin1 - y3 * sin3(part of b3)
smull v22.4s, v10.4h, v0.h[0]
smlsl v22.4s, v11.4h, v3.h[2]
smull v20.4s, v10.4h, v0.h[0]
smlsl v20.4s, v11.4h, v2.h[2]
smull v16.4s, v10.4h, v0.h[0]
smlsl v16.4s, v11.4h, v1.h[2]
smull v18.4s, v10.4h, v0.h[0]
smlsl v18.4s, v11.4h, v0.h[2]
cmp x12,x6
bge skip_last8rows_stage2_kernel2
smlsl v24.4s, v8.4h, v3.h[1]
smlal v26.4s, v8.4h, v2.h[1]
smlal v28.4s, v8.4h, v0.h[1]
smlal v30.4s, v8.4h, v2.h[3]
smlal v24.4s, v9.4h, v0.h[1]
smlal v26.4s, v9.4h, v3.h[1]
smlsl v28.4s, v9.4h, v1.h[1]
smlsl v30.4s, v9.4h, v2.h[1]
smlsl v22.4s, v4.4h, v1.h[0]
smlal v22.4s, v5.4h, v2.h[2]
smlsl v20.4s, v4.4h, v3.h[0]
smlal v20.4s, v5.4h, v0.h[2]
smlal v16.4s, v4.4h, v3.h[0]
smlal v16.4s, v5.4h, v3.h[2]
smlal v18.4s, v4.4h, v1.h[0]
smlsl v18.4s, v5.4h, v1.h[2]
mov x19,#0xff00
cmp x12,x19
bge skip_last8rows_stage2_kernel2
ld1 {v10.4h, v11.4h},[x11],#16
ld1 {v6.4h, v7.4h},[x11],#16
ld1 {v4.4h, v5.4h},[x0],#16
ld1 {v8.4h, v9.4h},[x0],#16
smlsl v24.4s, v6.4h, v3.h[3] //// y1 * cos1(part of b0)
smlsl v26.4s, v6.4h, v0.h[3] //// y1 * cos3(part of b1)
smlal v28.4s, v6.4h, v2.h[3] //// y1 * sin3(part of b2)
smlal v30.4s, v6.4h, v1.h[3] //// y1 * sin1(part of b3)
smlsl v24.4s, v7.4h, v0.h[3] //// y1 * cos1 + y3 * cos3(part of b0)
smlal v26.4s, v7.4h, v1.h[3] //// y1 * cos3 - y3 * sin1(part of b1)
smlal v28.4s, v7.4h, v3.h[3] //// y1 * sin3 - y3 * cos1(part of b2)
smlsl v30.4s, v7.4h, v1.h[1] //// y1 * sin1 - y3 * sin3(part of b3)
smlal v24.4s, v8.4h, v2.h[3]
smlal v26.4s, v8.4h, v3.h[3]
smlsl v28.4s, v8.4h, v2.h[1]
smlal v30.4s, v8.4h, v0.h[3]
smlal v24.4s, v9.4h, v1.h[3]
smlsl v26.4s, v9.4h, v1.h[1]
smlal v28.4s, v9.4h, v0.h[3]
smlsl v30.4s, v9.4h, v0.h[1]
smlal v22.4s, v10.4h, v0.h[0]
smlsl v22.4s, v11.4h, v1.h[2]
smlsl v22.4s, v4.4h, v3.h[0]
smlal v22.4s, v5.4h, v0.h[2]
smlsl v20.4s, v10.4h, v0.h[0]
smlsl v20.4s, v11.4h, v3.h[2]
smlal v20.4s, v4.4h, v1.h[0]
smlsl v20.4s, v5.4h, v1.h[2]
smlsl v16.4s, v10.4h, v0.h[0]
smlal v16.4s, v11.4h, v0.h[2]
smlsl v16.4s, v4.4h, v1.h[0]
smlal v16.4s, v5.4h, v2.h[2]
smlal v18.4s, v10.4h, v0.h[0]
smlsl v18.4s, v11.4h, v2.h[2]
smlal v18.4s, v4.4h, v3.h[0]
smlsl v18.4s, v5.4h, v3.h[2]
skip_last8rows_stage2_kernel2:
add v4.4s, v22.4s , v24.4s
sub v22.4s, v22.4s , v24.4s
add v6.4s, v20.4s , v26.4s
sub v24.4s, v20.4s , v26.4s
add v10.4s, v16.4s , v28.4s
sub v26.4s, v16.4s , v28.4s
add v16.4s, v18.4s , v30.4s
sub v28.4s, v18.4s , v30.4s
sqrshrn v18.4h, v4.4s,#shift_stage2_idct //// x0 = (a0 + b0 + rnd) >> 7(shift_stage1_idct)
sqrshrn v31.4h, v22.4s,#shift_stage2_idct //// x7 = (a0 - b0 + rnd) >> 7(shift_stage1_idct)
sqrshrn v19.4h, v10.4s,#shift_stage2_idct //// x2 = (a2 + b2 + rnd) >> 7(shift_stage1_idct)
sqrshrn v30.4h, v26.4s,#shift_stage2_idct //// x5 = (a2 - b2 + rnd) >> 7(shift_stage1_idct)
sqrshrn v20.4h, v6.4s,#shift_stage2_idct //// x1 = (a1 + b1 + rnd) >> 7(shift_stage1_idct)
sqrshrn v23.4h, v24.4s,#shift_stage2_idct //// x6 = (a1 - b1 + rnd) >> 7(shift_stage1_idct)
sqrshrn v21.4h, v16.4s,#shift_stage2_idct //// x3 = (a3 + b3 + rnd) >> 7(shift_stage1_idct)
sqrshrn v22.4h, v28.4s,#shift_stage2_idct //// x4 = (a3 - b3 + rnd) >> 7(shift_stage1_idct)
ld1 {v4.4h, v5.4h},[x1],#16
ld1 {v8.4h, v9.4h},[x1],#16
// registers used: {q2,q4,q6,q7}, {q9,q15,q10,q11}
//d4=x0
//d12=x1
//d5=x2
//d13=x3
//d18=x4
//d20=x5
//d19=x6
//d21=x7
//d22=x8
//d30=x9
//d23=x10
//d31=x11
//d14=x12
//d8=x13
//d15=x14
//d9=x15
umov x15,v26.d[0]
umov x16,v27.d[0]
umov x19,v28.d[0]
umov x20,v29.d[0]
trn1 v26.4h, v4.4h, v12.4h
trn2 v27.4h, v4.4h, v12.4h
trn1 v28.4h, v5.4h, v13.4h
trn2 v29.4h, v5.4h, v13.4h
trn1 v4.2s, v26.2s, v28.2s
trn2 v5.2s, v26.2s, v28.2s
trn1 v12.2s, v27.2s, v29.2s
trn2 v13.2s, v27.2s, v29.2s
trn1 v26.4h, v18.4h, v20.4h
trn2 v27.4h, v18.4h, v20.4h
trn1 v28.4h, v19.4h, v21.4h
trn2 v29.4h, v19.4h, v21.4h
trn1 v18.2s, v26.2s, v28.2s
trn2 v19.2s, v26.2s, v28.2s
trn1 v20.2s, v27.2s, v29.2s
trn2 v21.2s, v27.2s, v29.2s
trn1 v26.4h, v22.4h, v30.4h
trn2 v27.4h, v22.4h, v30.4h
trn1 v28.4h, v23.4h, v31.4h
trn2 v29.4h, v23.4h, v31.4h
trn1 v22.2s, v26.2s, v28.2s
trn2 v23.2s, v26.2s, v28.2s
trn1 v30.2s, v27.2s, v29.2s
trn2 v31.2s, v27.2s, v29.2s
trn1 v26.4h, v14.4h, v8.4h
trn2 v27.4h, v14.4h, v8.4h
trn1 v28.4h, v15.4h, v9.4h
trn2 v29.4h, v15.4h, v9.4h
trn1 v14.2s, v26.2s, v28.2s
trn2 v15.2s, v26.2s, v28.2s
trn1 v8.2s, v27.2s, v29.2s
trn2 v9.2s, v27.2s, v29.2s
mov v26.d[0],x15
mov v27.d[0],x16
mov v28.d[0],x19
mov v29.d[0],x20
// d4 =x0 1- 4 values
// d5 =x2 1- 4 values
// d12=x1 1- 4 values
// d13=x3 1- 4 values
// d18 =x0 5- 8 values
// d19 =x2 5- 8 values
// d20=x1 5- 8 values
// d21=x3 5- 8 values
// d22 =x0 9- 12 values
// d23 =x2 9- 12 values
// d30=x1 9- 12 values
// d31=x3 9- 12 values
// d14 =x0 13-16 values
// d15 =x2 13- 16 values
// d8=x1 13- 16 values
// d9=x3 13- 16 values
// swapping v5 and v15
mov v5.d[1],v5.d[0]
mov v5.d[0],v18.d[0]
mov v18.d[0],v5.d[1]
// swapping v23 and v14
mov v23.d[1],v23.d[0]
mov v23.d[0],v14.d[0]
mov v14.d[0],v23.d[1]
// swapping v13 and v20
mov v13.d[1],v13.d[0]
mov v13.d[0],v20.d[0]
mov v20.d[0],v13.d[1]
// swapping v31 and v8
mov v31.d[1],v31.d[0]
mov v31.d[0],v8.d[0]
mov v8.d[0],v31.d[1]
// q2: x0 1-8 values
// q11: x0 9-16 values
// q9 : x2 1-8 values
// q7 : x2 9-16 values
// q6 : x1 1- 8 values
// q10: x3 1-8 values
// q15: x1 9-16 values
// q4: x3 9-16 values
// registers free: q8,q14,q12,q13
ld1 {v16.8b, v17.8b},[x2],x8
ld1 {v28.8b, v29.8b},[x2],x5
ld1 {v24.8b, v25.8b},[x4],x8
ld1 {v26.8b, v27.8b},[x4],x5
mov v4.d[1] ,v5.d[0]
mov v22.d[1] ,v23.d[0]
mov v12.d[1] ,v13.d[0]
mov v30.d[1] ,v31.d[0]
mov v18.d[1] ,v19.d[0]
mov v14.d[1] ,v15.d[0]
mov v20.d[1] ,v21.d[0]
mov v8.d[1] ,v9.d[0]
uaddw v4.8h, v4.8h , v16.8b
uaddw v22.8h, v22.8h , v17.8b
uaddw v12.8h, v12.8h , v28.8b
uaddw v30.8h, v30.8h , v29.8b
uaddw v18.8h, v18.8h , v24.8b
uaddw v14.8h, v14.8h , v25.8b
uaddw v20.8h, v20.8h , v26.8b
uaddw v8.8h, v8.8h , v27.8b
sqxtun v16.8b, v4.8h
sqxtun v17.8b, v22.8h
sqxtun v28.8b, v12.8h
sqxtun v29.8b, v30.8h
sqxtun v24.8b, v18.8h
sqxtun v25.8b, v14.8h
sqxtun v26.8b, v20.8h
sqxtun v27.8b, v8.8h
st1 {v16.8b, v17.8b},[x3],x7
st1 {v28.8b, v29.8b},[x3],x7
st1 {v24.8b, v25.8b},[x3],x7
st1 {v26.8b, v27.8b},[x3],x7
subs x14,x14,#1
bne second_stage
// sub sp,sp,#40
// ldmfd sp!,{x4-x12,pc}
ldp x19, x20,[sp],#16
pop_v_regs
ret