@/*****************************************************************************
@*
@* 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_intra_pred_filters_dc.s
@*
@* @brief
@* contains function definitions for intra prediction dc filtering.
@* functions are coded using neon intrinsics and can be compiled using
@* rvct
@*
@* @author
@* akshaya mukund
@*
@* @par list of functions:
@*
@*
@* @remarks
@* none
@*
@*******************************************************************************
@*/
@/**
@*******************************************************************************
@*
@* @brief
@* luma intraprediction filter for dc input
@*
@* @par description:
@*
@* @param[in] pu1_ref
@* uword8 pointer to the source
@*
@* @param[out] pu1_dst
@* uword8 pointer to the destination
@*
@* @param[in] src_strd
@* integer source stride
@*
@* @param[in] dst_strd
@* integer destination stride
@*
@* @param[in] pi1_coeff
@* word8 pointer to the planar coefficients
@*
@* @param[in] nt
@* size of tranform block
@*
@* @param[in] mode
@* type of filtering
@*
@* @returns
@*
@* @remarks
@* none
@*
@*******************************************************************************
@*/
@void ihevc_intra_pred_luma_dc(uword8 *pu1_ref,
@ word32 src_strd,
@ uword8 *pu1_dst,
@ word32 dst_strd,
@ word32 nt,
@ word32 mode)
@
@**************variables vs registers*****************************************
@r0 => *pu1_ref
@r1 => src_strd
@r2 => *pu1_dst
@r3 => dst_strd
@stack contents from #104
@ nt
@ mode
@ pi1_coeff
.equ nt_offset, 104
.text
.align 4
.globl ihevc_intra_pred_luma_dc_a9q
.type ihevc_intra_pred_luma_dc_a9q, %function
ihevc_intra_pred_luma_dc_a9q:
stmfd sp!, {r4-r12, r14} @stack stores the values of the arguments
vpush {d8 - d15}
ldr r4,[sp,#nt_offset] @loads nt
@********** testing
@mov r6, #128
@b prologue_cpy_32
@********** testing
mov r11, #2 @mov #2 to r11 (to be used to add to 2dc_val & 3dc_val)
mov r9, #0
vmov d17, r11, r9
clz r5, r4
add r6, r0, r4 @&src[nt]
rsb r5, r5, #32 @log2nt
add r7, r0, r4, lsl #1 @&src[2nt]
add r8, r7, #1 @&src[2nt+1]
mvn r5, r5
add r5, r5, #1
vdup.32 d8, r5
ldrb r14, [r8]
vshl.i64 d8, d8, #32
sub r9, r7, #1 @&src[2nt-1]
vshr.s64 d8, d8, #32
mov r7, r8 @r7 also stores 2nt+1
ldrb r12, [r9]
add r14, r14, r12 @src[2nt+1] + src[2nt-1]
add r14, r14, r11 @src[2nt+1] + src[2nt-1] + 2
cmp r4, #4
beq dc_4
mov r10, r4 @nt
add_loop:
vld1.s8 d0, [r6]! @load from src[nt]
mov r5, #0 @
vld1.s8 d1, [r8]! @load from src[2nt+1]
vpaddl.u8 d2, d0
vmov d6, r4, r5 @store nt to accumulate
vpaddl.u8 d3, d1
vld1.s8 d0, [r6]! @load from src[nt] (extra load for 8)
vld1.s8 d1, [r8]! @load from src[2nt+1] (extra load for 8)
vadd.u16 d4, d2, d3
vpaddl.u16 d5, d4
vpadal.u32 d6, d5 @accumulate all inp into d6 (end for nt==8)
subs r10, #8
beq epil_add_loop
core_loop_add:
vpaddl.u8 d2, d0
subs r10, #8
vpaddl.u8 d3, d1
vadd.u16 d4, d2, d3
vld1.s8 d0, [r6]! @load from src[nt] (extra load for 16)
vpaddl.u16 d5, d4
vld1.s8 d1, [r8]! @load from src[2nt+1] (extra load for 16)
vpadal.u32 d6, d5 @accumulate all inp into d6
bne core_loop_add
epil_add_loop:
vshl.s64 d9, d6, d8 @(dc_val) shr by log2nt+1
cmp r4, #32
vmov d28, r14, r5 @src[2nt+1]+2+src[2nt-1] moved to d28
moveq r6, #128
vdup.8 d16, d9[0] @dc_val
vshl.s64 d13, d9, #1 @2*dc
beq prologue_cpy_32
vadd.i64 d14, d13, d28 @src[2nt+1]+2+src[2nt-1]+2dc_val
movne r6, #0 @nt
vshr.u16 d15, d14, #2 @final dst[0]'s value in d15[0]
movne r10, r4
vadd.i64 d11, d13, d9 @3*dc
sub r12, r3, r3, lsl #3 @-7*strd
vadd.i64 d11, d11, d17 @3*dc + 2
add r12, r12, #8 @offset after one 8x8 block (-7*strd + 8)
vdup.16 q12, d11[0] @3*dc + 2 (moved to all lanes)
sub r0, r3, r4 @strd - nt
prologue_col:
@0th column and 0-7 rows done here
@r8 and r9 (2nt+1+col 2nt-1-row)
mov r8, r7 @&src[2nt+1]
add r0, r0, #8 @strd - nt + 8
vld1.s8 d0, [r8]! @col 1::7 load (prol)
sub r9, r9, #7 @&src[2nt-1-row]
vld1.s8 d1, [r9] @row 7::1 (0 also) load (prol)
sub r9, r9, #8
vmovl.u8 q10, d0
vld1.s8 d6, [r8] @col 8::15 load (prol extra)
vadd.i16 q10, q10, q12 @col 1::7 add 3dc+2 (prol)
vmovl.u8 q11, d1
vqshrun.s16 d2, q10, #2 @columns shr2 movn (prol)
vmovl.u8 q13, d6
vadd.i16 q11, q11, q12 @row 1::7 add 3dc+2 (prol)
vmov.i64 d19, #0x00000000000000ff @
vqshrun.s16 d3, q11, #2 @rows shr2 movn (prol)
vbsl d19, d15, d2 @first row with dst[0]
vadd.i16 q13, q13, q12 @col 8::15 add 3dc+2 (prol extra)
vrev64.8 d3, d3
vst1.8 d19, [r2], r3 @store row 0 (prol)
vshr.s64 d3, d3, #8 @row 0 shift (prol) (first value to be ignored)
vmov.i64 d20, #0x00000000000000ff @byte mask row 1 (prol)
loop_again_col_row:
vbsl d20, d3, d16 @row 1 (prol)
vmov.i64 d21, #0x00000000000000ff @byte mask row 2 (prol)
vshr.s64 d3, d3, #8 @row 1 shift (prol)
vst1.8 d20, [r2], r3 @store row 1 (prol)
vqshrun.s16 d4, q13, #2 @columns shr2 movn (prol extra)
vbsl d21, d3, d16 @row 2 (prol)
vmov.i64 d20, #0x00000000000000ff @byte mask row 3 (prol)
vshr.s64 d3, d3, #8 @row 2 shift (prol)
vst1.8 d21, [r2], r3 @store row 2 (prol)
vbsl d20, d3, d16 @row 3 (prol)
vmov.i64 d21, #0x00000000000000ff @byte mask row 4 (prol)
vshr.s64 d3, d3, #8 @row 3 shift (prol)
vst1.8 d20, [r2], r3 @store row 3 (prol)
vbsl d21, d3, d16 @row 4 (prol)
vmov.i64 d20, #0x00000000000000ff @byte mask row 5 (prol)
vshr.s64 d3, d3, #8 @row 4 shift (prol)
vst1.8 d21, [r2], r3 @store row 4 (prol)
vbsl d20, d3, d16 @row 5 (prol)
vmov.i64 d21, #0x00000000000000ff @byte mask row 6 (prol)
vshr.s64 d3, d3, #8 @row 5 shift (prol)
vst1.8 d20, [r2], r3 @store row 5 (prol)
vld1.s8 d1, [r9] @row 8::15 load (prol extra)
vbsl d21, d3, d16 @row 6 (prol)
vmovl.u8 q11, d1
vmov.i64 d20, #0x00000000000000ff @byte mask row 7 (prol)
vshr.s64 d3, d3, #8 @row 6 shift (prol)
vst1.8 d21, [r2], r3 @store row 6 (prol)
vbsl d20, d3, d16 @row 7 (prol)
vadd.i16 q11, q11, q12 @row 8::15 add 3dc+2 (prol extra)
vshr.s64 d3, d3, #8 @row 7 shift (prol)
vst1.8 d20, [r2], r12 @store row 7 (prol)
subs r10, r10, #8 @counter for cols
beq end_func
blt copy_16
vmov.i64 d20, #0x00000000000000ff @byte mask row 9 (prol)
vqshrun.s16 d3, q11, #2 @rows shr2 movn (prol)
vrev64.8 d3, d3
vst1.8 d4, [r2], r3 @store 2nd col (for 16x16)
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r0 @go to next row for 16
vbsl d20, d3, d16 @row 9 (prol)
subs r10, r10, #8
vst1.8 d20, [r2], r3 @store row 9 (prol)
vshr.s64 d3, d3, #8 @row 9 shift (prol)
vmov.i64 d20, #0x00000000000000ff @byte mask row 9 (prol)
b loop_again_col_row
copy_16:
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2], r3
vst1.8 d16, [r2]
b end_func
prologue_cpy_32:
mov r9, #128
@sub r7, r3, #-24
add r5, r2, r3
add r8, r5, r3
add r10, r8, r3
vdup.8 q10, d16[0]
lsl r6, r3, #2
add r6, r6, #0xfffffff0
vst1.8 {d20,d21}, [r2]!
vst1.8 {d20,d21}, [r5]!
vst1.8 {d20,d21}, [r8]!
vst1.8 {d20,d21}, [r10]!
vst1.8 {d20,d21}, [r2], r6
vst1.8 {d20,d21}, [r5], r6
vst1.8 {d20,d21}, [r8], r6
vst1.8 {d20,d21}, [r10], r6
sub r9, r9, #32 @32x32 prol/epil counter dec
kernel_copy:
vst1.8 {d20,d21}, [r2]!
vst1.8 {d20,d21}, [r5]!
vst1.8 {d20,d21}, [r8]!
vst1.8 {d20,d21}, [r10]!
vst1.8 {d20,d21}, [r2], r6
vst1.8 {d20,d21}, [r5], r6
vst1.8 {d20,d21}, [r8], r6
vst1.8 {d20,d21}, [r10], r6
subs r9, r9, #32
vst1.8 {d20,d21}, [r2]!
vst1.8 {d20,d21}, [r5]!
vst1.8 {d20,d21}, [r8]!
vst1.8 {d20,d21}, [r10]!
vst1.8 {d20,d21}, [r2], r6
vst1.8 {d20,d21}, [r5], r6
vst1.8 {d20,d21}, [r8], r6
vst1.8 {d20,d21}, [r10], r6
bne kernel_copy
epilogue_copy:
vst1.8 {d20,d21}, [r2]!
vst1.8 {d20,d21}, [r5]!
vst1.8 {d20,d21}, [r8]!
vst1.8 {d20,d21}, [r10]!
vst1.8 {d20,d21}, [r2]
vst1.8 {d20,d21}, [r5]
vst1.8 {d20,d21}, [r8]
vst1.8 {d20,d21}, [r10]
b end_func
dc_4:
vld1.s8 d0, [r6]! @load from src[nt]
vld1.s8 d1, [r8]! @load from src[2nt+1]
vpaddl.u8 d2, d0
mov r5, #0 @
vmov d6, r4, r5 @store nt to accumulate
vpaddl.u8 d3, d1
vadd.u16 d4, d2, d3
vpaddl.u16 d5, d4
vmov.i64 d30, #0x00000000ffffffff
vand d5, d5, d30
vmov d28, r14, r5 @src[2nt+1]+2+src[2nt-1] moved to d28
vadd.i64 d6, d6, d5 @accumulate all inp into d6 (end for nt==8)
vshl.s64 d9, d6, d8 @(dc_val) shr by log2nt+1
mov r8, r7 @&src[2nt+1]
vshl.s64 d13, d9, #1 @2*dc
sub r9, r9, #3 @&src[2nt-1-row]
vdup.8 d16, d9[0] @dc_val
vadd.i64 d14, d13, d28 @src[2nt+1]+2+src[2nt-1]+2dc_val
vshr.u16 d15, d14, #2 @final dst[0]'s value in d15[0]
sub r12, r3, r3, lsl #2 @-3*strd
vadd.i64 d11, d13, d9 @3*dc
vadd.i64 d11, d11, d17 @3*dc + 2
add r12, r12, #4 @offset after one 4x4 block (-3*strd + 4)
vdup.16 q12, d11[0] @3*dc + 2 (moved to all lanes)
sub r0, r3, r4 @strd - nt
vld1.s8 d0, [r8] @col 1::3 load (prol)
vld1.s8 d1, [r9] @row 3::1 (0 also) load (prol)
vmovl.u8 q10, d0
vmovl.u8 q11, d1
vadd.i16 q10, q10, q12 @col 1::7 add 3dc+2 (prol)
vadd.i16 q11, q11, q12 @row 1::7 add 3dc+2 (prol)
vmov.i64 d19, #0x00000000000000ff @
vqshrun.s16 d2, q10, #2 @columns shr2 movn (prol)
vmov.i64 d20, #0x00000000000000ff @byte mask row 1 (prol)
vqshrun.s16 d3, q11, #2 @rows shr2 movn (prol)
vbsl d19, d15, d2 @first row with dst[0]
vrev64.8 d3, d3
vst1.32 d19[0], [r2], r3 @store row 0 (prol)
vshr.s64 d3, d3, #40 @row 0 shift (prol) (first value to be ignored)
vmov.i64 d21, #0x00000000000000ff @byte mask row 2 (prol)
vbsl d20, d3, d16 @row 1 (prol)
vshr.s64 d3, d3, #8 @row 1 shift (prol)
vst1.32 d20[0], [r2], r3 @store row 1 (prol)
vbsl d21, d3, d16 @row 2 (prol)
vmov.i64 d20, #0x00000000000000ff @byte mask row 3 (prol)
vshr.s64 d3, d3, #8 @row 2 shift (prol)
vst1.32 d21[0], [r2], r3 @store row 2 (prol)
vbsl d20, d3, d16 @row 3 (prol)
vst1.32 d20[0], [r2] @store row 3 (prol)
epilogue_end:
end_func:
vpop {d8 - d15}
ldmfd sp!,{r4-r12,r15} @reload the registers from sp