ref: a893e2231152216d5a6cd262a153f17d2ee1eb50
dir: /codec/processing/src/arm/vaa_calc_neon.S/
/*!
* \copy
* Copyright (c) 2013, Cisco Systems
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
*
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
* COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
* ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*
*/
#ifdef HAVE_NEON
.text
#include "arm_arch_common_macro.S"
#ifdef __APPLE__
.macro ABS_SUB_SUM_16BYTES
vld1.32 {q15}, [$0], $2
vld1.32 {q14}, [$1], $2
vabal.u8 $3, d30, d28
vabal.u8 $4, d31, d29
.endm
.macro ABS_SUB_SUM_8x16BYTES
vld1.32 {q15}, [$0], $2
vld1.32 {q14}, [$1], $2
vabdl.u8 $3, d30, d28
vabdl.u8 $4, d31, d29
ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
.endm
.macro SAD_8X16BITS
vadd.u16 d31, $0, $1
vpaddl.u16 d31, d31
vpaddl.u32 $2, d31
.endm
#else
.macro ABS_SUB_SUM_16BYTES arg0, arg1, arg2, arg3, arg4
vld1.32 {q15}, [\arg0], \arg2
vld1.32 {q14}, [\arg1], \arg2
vabal.u8 \arg3, d30, d28
vabal.u8 \arg4, d31, d29
.endm
.macro ABS_SUB_SUM_8x16BYTES arg0, arg1, arg2, arg3, arg4
vld1.32 {q15}, [\arg0], \arg2
vld1.32 {q14}, [\arg1], \arg2
vabdl.u8 \arg3, d30, d28
vabdl.u8 \arg4, d31, d29
ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
.endm
.macro SAD_8X16BITS arg0, arg1, arg2
vadd.u16 d31, \arg0, \arg1
vpaddl.u16 d31, d31
vpaddl.u32 \arg2, d31
.endm
#endif
WELS_ASM_FUNC_BEGIN VAACalcSad_neon
stmdb sp!, {r4-r8}
ldr r4, [sp, #20] //load pic_stride
ldr r5, [sp, #28] //load psad8x8
//Initial the Q8 register for save the "psadframe"
vmov.s64 q8, #0
//Get the jump distance to use on loop codes
lsl r8, r4, #4
sub r7, r8, #16 //R7 keep the 16*pic_stride-16
sub r8, r2 //R8 keep the 16*pic_stride-pic_width
vaa_calc_sad_loop0:
//R6 keep the pic_width
mov r6, r2
vaa_calc_sad_loop1:
//Process the 16x16 bytes
ABS_SUB_SUM_8x16BYTES r0, r1, r4, q0, q1
ABS_SUB_SUM_8x16BYTES r0, r1, r4, q2, q3
//Do the SAD
SAD_8X16BITS d0, d1, d0
SAD_8X16BITS d2, d3, d1
SAD_8X16BITS d4, d5, d2
SAD_8X16BITS d6, d7, d3
//Write to "psad8x8" buffer
vst4.32 {d0[0],d1[0],d2[0],d3[0]}, [r5]!
//Adjust the input address
sub r0, r7
sub r1, r7
subs r6, #16
//Save to calculate "psadframe"
vadd.u32 q0, q1
vadd.u32 q8, q0
bne vaa_calc_sad_loop1
//Adjust the input address
add r0, r8
add r1, r8
subs r3, #16
bne vaa_calc_sad_loop0
ldr r6, [sp, #24] //load psadframe
vadd.u32 d16, d17
vst1.32 {d16[0]}, [r6]
ldmia sp!, {r4-r8}
WELS_ASM_FUNC_END
#ifdef __APPLE__
.macro SAD_SD_MAD_16BYTES
vld1.32 {q0}, [$0], $2
vld1.32 {q1}, [$1], $2
vpadal.u8 $3, q0
vpadal.u8 $4, q1
vabd.u8 q0, q0, q1
vmax.u8 $5, q0
vpadal.u8 $6, q0
.endm
.macro SAD_SD_MAD_8x16BYTES
vld1.32 {q0}, [$0], $2
vld1.32 {q1}, [$1], $2
vpaddl.u8 q2, q0
vpaddl.u8 q3, q1
vabd.u8 $3, q0, q1
vpaddl.u8 $4, $3 //abs_diff
SAD_SD_MAD_16BYTES $0,$1,$2,q2,q3,$3,$4
SAD_SD_MAD_16BYTES $0,$1,$2,q2,q3,$3,$4
SAD_SD_MAD_16BYTES $0,$1,$2,q2,q3,$3,$4
SAD_SD_MAD_16BYTES $0,$1,$2,q2,q3,$3,$4
SAD_SD_MAD_16BYTES $0,$1,$2,q2,q3,$3,$4
SAD_SD_MAD_16BYTES $0,$1,$2,q2,q3,$3,$4
SAD_SD_MAD_16BYTES $0,$1,$2,q2,q3,$3,$4
vsub.u16 $5, q2, q3
.endm
.macro SAD_SD_MAD_CALC
vpmax.u8 d0, $0, $1 //8bytes
vpmax.u8 d0, d0, d0 //4bytes
vpmax.u8 $2, d0, d0 //2bytes
vpaddl.u16 $3, $3
vpaddl.u32 $3, $3
vpaddl.s16 $4, $4
vpaddl.s32 $4, $4
.endm
#else
.macro SAD_SD_MAD_16BYTES arg0, arg1, arg2, arg3, arg4, arg5, arg6
vld1.32 {q0}, [\arg0], \arg2
vld1.32 {q1}, [\arg1], \arg2
vpadal.u8 \arg3, q0
vpadal.u8 \arg4, q1
vabd.u8 q0, q0, q1
vmax.u8 \arg5, q0
vpadal.u8 \arg6, q0
.endm
.macro SAD_SD_MAD_8x16BYTES arg0, arg1, arg2, arg3, arg4, arg5
vld1.32 {q0}, [\arg0], \arg2
vld1.32 {q1}, [\arg1], \arg2
vpaddl.u8 q2, q0
vpaddl.u8 q3, q1
vabd.u8 \arg3, q0, q1
vpaddl.u8 \arg4, \arg3 //abs_diff
SAD_SD_MAD_16BYTES \arg0,\arg1,\arg2,q2,q3,\arg3,\arg4
SAD_SD_MAD_16BYTES \arg0,\arg1,\arg2,q2,q3,\arg3,\arg4
SAD_SD_MAD_16BYTES \arg0,\arg1,\arg2,q2,q3,\arg3,\arg4
SAD_SD_MAD_16BYTES \arg0,\arg1,\arg2,q2,q3,\arg3,\arg4
SAD_SD_MAD_16BYTES \arg0,\arg1,\arg2,q2,q3,\arg3,\arg4
SAD_SD_MAD_16BYTES \arg0,\arg1,\arg2,q2,q3,\arg3,\arg4
SAD_SD_MAD_16BYTES \arg0,\arg1,\arg2,q2,q3,\arg3,\arg4
vsub.u16 \arg5, q2, q3
.endm
.macro SAD_SD_MAD_CALC arg0, arg1, arg2, arg3, arg4
vpmax.u8 d0, \arg0, \arg1 //8bytes
vpmax.u8 d0, d0, d0 //4bytes
vpmax.u8 \arg2, d0, d0 //2bytes
vpaddl.u16 \arg3, \arg3
vpaddl.u32 \arg3, \arg3
vpaddl.s16 \arg4, \arg4
vpaddl.s32 \arg4, \arg4
.endm
#endif
WELS_ASM_FUNC_BEGIN VAACalcSadBgd_neon
stmdb sp!, {r4-r10}
ldr r4, [sp, #28] //load pic_stride
ldr r5, [sp, #36] //load psad8x8
ldr r6, [sp, #40] //load psd8x8
ldr r7, [sp, #44] //load pmad8x8
//Initial the Q4 register for save the "psadframe"
vmov.s64 q15, #0
//Get the jump distance to use on loop codes
lsl r10, r4, #4
sub r9, r10, #16 //R9 keep the 16*pic_stride-16
sub r10, r2 //R10 keep the 16*pic_stride-pic_width
vaa_calc_sad_bgd_loop0:
//R6 keep the pic_width
mov r8, r2
vaa_calc_sad_bgd_loop1:
//Process the 16x16 bytes pmad psad psd
SAD_SD_MAD_8x16BYTES r0, r1, r4, q13, q11, q9
SAD_SD_MAD_8x16BYTES r0, r1, r4, q14, q12, q10
SAD_SD_MAD_CALC d26, d27, d16, q11, q9
SAD_SD_MAD_CALC d28, d29, d17, q12, q10
//Write to "psad8x8" buffer
vst4.32 {d22[0],d23[0],d24[0],d25[0]}, [r5]!
//Adjust the input address
sub r0, r9
sub r1, r9
//Write to "psd8x8" buffer
vst4.32 {d18[0],d19[0],d20[0],d21[0]}, [r6]!
subs r8, #16
//Write to "pmad8x8" buffer
vst2.16 {d16[0],d17[0]}, [r7]!
//Save to calculate "psadframe"
vadd.u32 q11, q12
vadd.u32 q15, q11
bne vaa_calc_sad_bgd_loop1
//Adjust the input address
add r0, r10
add r1, r10
subs r3, #16
bne vaa_calc_sad_bgd_loop0
ldr r8, [sp, #32] //load psadframe
vadd.u32 d30, d31
vst1.32 {d30[0]}, [r8]
ldmia sp!, {r4-r10}
WELS_ASM_FUNC_END
#ifdef __APPLE__
.macro SSD_MUL_SUM_16BYTES_RESET
vmull.u8 $3, $0, $0
vpaddl.u16 $2, $3
vmull.u8 $3, $1, $1
vpadal.u16 $2, $3
.endm
.macro SSD_MUL_SUM_16BYTES
vmull.u8 $3, $0, $0
vpadal.u16 $2, $3
vmull.u8 $3, $1, $1
vpadal.u16 $2, $3
.endm
.macro SAD_SSD_BGD_16
vld1.8 {q0}, [$0], $2 //load cur_row
vpadal.u8 q3, q0 //add cur_row together
vpadal.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
vpadal.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
vld1.8 {q1}, [$1], $2 //load ref_row
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
//the last row of a 16x16 block
.macro SAD_SSD_BGD_16_end
vld1.8 {q0}, [$0], $1 //load cur_row
vpadal.u8 q3, q0 //add cur_row together
vpadal.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
vpadal.u8 $2, q2 //l_sad for 16 bytes reset for every 8x16
SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
//for the begin of a 8x16 block, use some instructions to reset the register
.macro SAD_SSD_BGD_16_RESET_8x8
vld1.8 {q0}, [$0], $2 //load cur_row
vpaddl.u8 q3, q0 //add cur_row together
vpaddl.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
vld1.8 {q1}, [$1], $2 //load ref_row
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
//for the begin of a 16x16 block, use some instructions to reset the register
.macro SAD_SSD_BGD_16_RESET_16x16
vld1.8 {q0}, [$0], $2 //load cur_row
vld1.8 {q1}, [$1], $2 //load ref_row
vpaddl.u8 q3, q0 //add cur_row together
vpaddl.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
vld1.8 {q1}, [$1], $2 //load ref_row
vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES_RESET d0,d1,q10,q11 //q10 for lsqsum reset for every 16x16
.endm
//for each 8x16 block
.macro SAD_SSD_BGD_CALC_8x16
vpmax.u8 d10, d10, d11 //4 numbers
vpmax.u8 d10, d10, d10 //2 numbers
vpmax.u8 d10, d10, d10 //1 number1
vmov $0, d10 //d26 d27 keeps the l_mad
//p_sd8x8 fix me
vpaddl.u16 q3, q3
vpaddl.u16 q4, q4
vsub.i32 $1, q3, q4
vpaddl.u32 $1, $1
//psad8x8
vpaddl.u16 $2, $2
vpaddl.u32 $2, $2
//psadframe
vadd.i32 q12, $2
.endm
.macro SAD_SSD_BGD_16x16
//for one 8x16
SAD_SSD_BGD_16_RESET_16x16 $0, $1, $2, q6
SAD_SSD_BGD_16 $0, $1, $2, q6
SAD_SSD_BGD_16 $0, $1, $2, q6
SAD_SSD_BGD_16 $0, $1, $2, q6
SAD_SSD_BGD_16 $0, $1, $2, q6
SAD_SSD_BGD_16 $0, $1, $2, q6
SAD_SSD_BGD_16 $0, $1, $2, q6
SAD_SSD_BGD_16 $0, $1, $2, q6
SAD_SSD_BGD_CALC_8x16 d26, q14, q6
//for another 8x16
SAD_SSD_BGD_16_RESET_8x8 $0, $1, $2, q7
SAD_SSD_BGD_16 $0, $1, $2, q7
SAD_SSD_BGD_16 $0, $1, $2, q7
SAD_SSD_BGD_16 $0, $1, $2, q7
SAD_SSD_BGD_16 $0, $1, $2, q7
SAD_SSD_BGD_16 $0, $1, $2, q7
SAD_SSD_BGD_16 $0, $1, $2, q7
SAD_SSD_BGD_16_end $0, $2, q7
SAD_SSD_BGD_CALC_8x16 d27, q15, q7
.endm
.macro SSD_SAD_SD_MAD_PADDL
vpaddl.s16 $0, $0
vpaddl.s32 $0, $0
vadd.i32 $1, $1, $2
.endm
#else
.macro SSD_MUL_SUM_16BYTES_RESET arg0, arg1, arg2, arg3
vmull.u8 \arg3, \arg0, \arg0
vpaddl.u16 \arg2, \arg3
vmull.u8 \arg3, \arg1, \arg1
vpadal.u16 \arg2, \arg3
.endm
.macro SSD_MUL_SUM_16BYTES arg0, arg1, arg2, arg3
vmull.u8 \arg3, \arg0, \arg0
vpadal.u16 \arg2, \arg3
vmull.u8 \arg3, \arg1, \arg1
vpadal.u16 \arg2, \arg3
.endm
.macro SAD_SSD_BGD_16 arg0, arg1, arg2, arg3
vld1.8 {q0}, [\arg0], \arg2 //load cur_row
vpadal.u8 q3, q0 //add cur_row together
vpadal.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
vpadal.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
vld1.8 {q1}, [\arg1], \arg2 //load ref_row
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
//the last row of a 16x16 block
.macro SAD_SSD_BGD_16_end arg0, arg1, arg2
vld1.8 {q0}, [\arg0], \arg1 //load cur_row
vpadal.u8 q3, q0 //add cur_row together
vpadal.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
vpadal.u8 \arg2, q2 //l_sad for 16 bytes reset for every 8x16
SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
//for the begin of a 8x16 block, use some instructions to reset the register
.macro SAD_SSD_BGD_16_RESET_8x8 arg0, arg1, arg2, arg3
vld1.8 {q0}, [\arg0], \arg2 //load cur_row
vpaddl.u8 q3, q0 //add cur_row together
vpaddl.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
vld1.8 {q1}, [\arg1], \arg2 //load ref_row
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
//for the begin of a 16x16 block, use some instructions to reset the register
.macro SAD_SSD_BGD_16_RESET_16x16 arg0, arg1, arg2, arg3
vld1.8 {q0}, [\arg0], \arg2 //load cur_row
vld1.8 {q1}, [\arg1], \arg2 //load ref_row
vpaddl.u8 q3, q0 //add cur_row together
vpaddl.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
vld1.8 {q1}, [\arg1], \arg2 //load ref_row
vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES_RESET d0,d1,q10,q11 //q10 for lsqsum reset for every 16x16
.endm
//for each 8x16 block
.macro SAD_SSD_BGD_CALC_8x16 arg0, arg1, arg2
vpmax.u8 d10, d10, d11 //4 numbers
vpmax.u8 d10, d10, d10 //2 numbers
vpmax.u8 d10, d10, d10 //1 number1
vmov \arg0, d10 //d26 d27 keeps the l_mad
//p_sd8x8
vpaddl.u16 q3, q3
vpaddl.u16 q4, q4
vsub.i32 \arg1, q3, q4
vpaddl.u32 \arg1, \arg1
//psad8x8
vpaddl.u16 \arg2, \arg2
vpaddl.u32 \arg2, \arg2
//psadframe
vadd.i32 q12, \arg2
.endm
.macro SAD_SSD_BGD_16x16 arg0, arg1, arg2
//for one 8x16
SAD_SSD_BGD_16_RESET_16x16 \arg0, \arg1, \arg2, q6
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_BGD_CALC_8x16 d26, q14, q6
//for another 8x16
SAD_SSD_BGD_16_RESET_8x8 \arg0, \arg1, \arg2, q7
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_BGD_16_end \arg0, \arg2, q7
SAD_SSD_BGD_CALC_8x16 d27, q15, q7
.endm
.macro SSD_SAD_SD_MAD_PADDL arg0, arg1, arg2
vpaddl.s16 \arg0, \arg0
vpaddl.s32 \arg0, \arg0
vadd.i32 \arg1, \arg1, \arg2
.endm
#endif
WELS_ASM_FUNC_BEGIN VAACalcSadSsdBgd_neon
stmdb sp!, {r0-r12, r14}
vpush {q4-q7}
ldr r4, [sp, #120] //r4 keeps the pic_stride
sub r5, r4, #1
lsl r5, r5, #4 //r5 keeps the little step
lsl r6, r4, #4
sub r6, r2, r6 //r6 keeps the big step
ldr r8, [sp, #128]//psad8x8
ldr r9, [sp, #132]//psum16x16
ldr r10, [sp, #136]//psqsum16x16
ldr r11, [sp, #140]//psqdiff16x16
ldr r12, [sp, #144]//p_sd8x8
ldr r14, [sp, #148]//p_mad8x8
vmov.i8 q12, #0
vaa_calc_sad_ssd_bgd_height_loop:
mov r7, r2
vaa_calc_sad_ssd_bgd_width_loop:
//l_sd q14&q15, l_mad q13, l_sad q6 & q7, l_sqdiff q8, l_sum q9, l_sqsum q10
SAD_SSD_BGD_16x16 r0,r1,r4
//psad8x8
vst4.32 {d12[0], d13[0], d14[0], d15[0]}, [r8]!
sub r0, r0, r5 //jump to next 16x16
sub r1, r1, r5 //jump to next 16x16
//p_sd8x8
vst4.32 {d28[0], d29[0],d30[0], d31[0]}, [r12]!
//p_mad8x8
vst2.16 {d26[0], d27[0]}, [r14]!
//psqdiff16x16
vpaddl.s32 q8, q8
vadd.i32 d16, d16, d17
vst1.32 {d16[0]}, [r11]! //psqdiff16x16
//psum16x16
SSD_SAD_SD_MAD_PADDL q9, d18, d19
vst1.32 {d18[0]}, [r9]! //psum16x16
//psqsum16x16
vpaddl.s32 q10, q10
vadd.i32 d20, d20, d21
vst1.32 {d20[0]}, [r10]! //psqsum16x16
subs r7, #16
bne vaa_calc_sad_ssd_bgd_width_loop
sub r0, r0, r6 //jump to next 16 x width
sub r1, r1, r6 //jump to next 16 x width
subs r3, #16
bne vaa_calc_sad_ssd_bgd_height_loop
//psadframe
ldr r7, [sp, #124]//psadframe
vadd.i32 d24, d24, d25
vst1.32 {d24[0]}, [r7]
vpop {q4-q7}
ldmia sp!, {r0-r12, r14}
WELS_ASM_FUNC_END
#ifdef __APPLE__
.macro SAD_VAR_16
vld1.8 {q0}, [$0], $2 //load cur_row
vpadal.u8 q3, q0 //add cur_row together
vpadal.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vpadal.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
vld1.8 {q1}, [$1], $2
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
.macro SAD_VAR_16_END
vld1.8 {q0}, [$0], $1 //load cur_row
vpadal.u8 q3, q0 //add cur_row together
vpadal.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vpadal.u8 $2, q2 //l_sad for 16 bytes reset for every 8x16
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
.macro SAD_VAR_16_RESET_16x16
vld1.8 {q0}, [$0], $2 //load cur_row
vld1.8 {q1}, [$1], $2
vpaddl.u8 q3, q0 //add cur_row together
vpaddl.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
vld1.8 {q1}, [$1], $2
vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES_RESET d0,d1, q10, q11
.endm
.macro SAD_VAR_16_RESET_8x8
vld1.8 {q0}, [$0], $2 //load cur_row
vpaddl.u8 q3, q0 //add cur_row together
vpaddl.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
vld1.8 {q1}, [$1], $2
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
.macro SAD_VAR_16x16
//for one 8x16
SAD_VAR_16_RESET_16x16 $0, $1, $2, q6
SAD_VAR_16 $0, $1, $2, q6
SAD_VAR_16 $0, $1, $2, q6
SAD_VAR_16 $0, $1, $2, q6
SAD_VAR_16 $0, $1, $2, q6
SAD_VAR_16 $0, $1, $2, q6
SAD_VAR_16 $0, $1, $2, q6
SAD_VAR_16 $0, $1, $2, q6
vpaddl.u16 q6, q6
vpaddl.u32 q6, q6
vadd.i32 q12, q6
//for another 8x16
SAD_VAR_16_RESET_8x8 $0, $1, $2, q7
SAD_VAR_16 $0, $1, $2, q7
SAD_VAR_16 $0, $1, $2, q7
SAD_VAR_16 $0, $1, $2, q7
SAD_VAR_16 $0, $1, $2, q7
SAD_VAR_16 $0, $1, $2, q7
SAD_VAR_16 $0, $1, $2, q7
SAD_VAR_16_END $0, $2, q7
vpaddl.u16 q7, q7
vpaddl.u32 q7, q7
vadd.i32 q12, q7
.endm
#else
.macro SAD_VAR_16 arg0, arg1, arg2, arg3
vld1.8 {q0}, [\arg0], \arg2 //load cur_row
vpadal.u8 q3, q0 //add cur_row together
vpadal.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vpadal.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
vld1.8 {q1}, [\arg1], \arg2
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
.macro SAD_VAR_16_END arg0, arg1, arg2
vld1.8 {q0}, [\arg0], \arg1 //load cur_row
vpadal.u8 q3, q0 //add cur_row together
vpadal.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vpadal.u8 \arg2, q2 //l_sad for 16 bytes reset for every 8x16
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
.macro SAD_VAR_16_RESET_16x16 arg0, arg1, arg2, arg3
vld1.8 {q0}, [\arg0], \arg2 //load cur_row
vld1.8 {q1}, [\arg1], \arg2
vpaddl.u8 q3, q0 //add cur_row together
vpaddl.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
vld1.8 {q1}, [\arg1], \arg2
vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES_RESET d0,d1, q10, q11
.endm
.macro SAD_VAR_16_RESET_8x8 arg0, arg1, arg2, arg3
vld1.8 {q0}, [\arg0], \arg2 //load cur_row
vpaddl.u8 q3, q0 //add cur_row together
vpaddl.u8 q4, q1 //add ref_row together
vabd.u8 q2, q0, q1 //abs_diff
vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
vld1.8 {q1}, [\arg1], \arg2
vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
.endm
.macro SAD_VAR_16x16 arg0, arg1, arg2
//for one 8x16
SAD_VAR_16_RESET_16x16 \arg0, \arg1, \arg2, q6
SAD_VAR_16 \arg0, \arg1, \arg2, q6
SAD_VAR_16 \arg0, \arg1, \arg2, q6
SAD_VAR_16 \arg0, \arg1, \arg2, q6
SAD_VAR_16 \arg0, \arg1, \arg2, q6
SAD_VAR_16 \arg0, \arg1, \arg2, q6
SAD_VAR_16 \arg0, \arg1, \arg2, q6
SAD_VAR_16 \arg0, \arg1, \arg2, q6
vpaddl.u16 q6, q6
vpaddl.u32 q6, q6
vadd.i32 q12, q6
//for another 8x16
SAD_VAR_16_RESET_8x8 \arg0, \arg1, \arg2, q7
SAD_VAR_16 \arg0, \arg1, \arg2, q7
SAD_VAR_16 \arg0, \arg1, \arg2, q7
SAD_VAR_16 \arg0, \arg1, \arg2, q7
SAD_VAR_16 \arg0, \arg1, \arg2, q7
SAD_VAR_16 \arg0, \arg1, \arg2, q7
SAD_VAR_16 \arg0, \arg1, \arg2, q7
SAD_VAR_16_END \arg0, \arg2, q7
vpaddl.u16 q7, q7
vpaddl.u32 q7, q7
vadd.i32 q12, q7
.endm
#endif
WELS_ASM_FUNC_BEGIN VAACalcSadVar_neon
stmdb sp!, {r4-r11}
vpush {q4}
vpush {q6-q7}
ldr r4, [sp, #80] //r4 keeps the pic_stride
sub r5, r4, #1
lsl r5, r5, #4 //r5 keeps the little step
lsl r6, r4, #4
sub r6, r2, r6 //r6 keeps the big step
ldr r7, [sp, #84] //psadframe
ldr r8, [sp, #88] //psad8x8
ldr r9, [sp, #92] //psum16x16
ldr r10, [sp, #96] //psqsum16x16
vmov.i8 q12, #0
vaa_calc_sad_var_height_loop:
mov r11, r2
vaa_calc_sad_var_width_loop:
SAD_VAR_16x16 r0,r1,r4
//psad8x8
vst4.32 {d12[0], d13[0], d14[0], d15[0]}, [r8]!
sub r0, r0, r5 //jump to next 16x16
sub r1, r1, r5 //jump to next 16x16
//psum16x16
SSD_SAD_SD_MAD_PADDL q9, d18, d19
vst1.32 {d18[0]}, [r9]! //psum16x16
//psqsum16x16
vpaddl.s32 q10, q10
subs r11, #16
vadd.i32 d20, d20, d21
vst1.32 {d20[0]}, [r10]! //psqsum16x16
bne vaa_calc_sad_var_width_loop
sub r0, r0, r6 //jump to next 16 x width
sub r1, r1, r6 //jump to next 16 x width
subs r3, #16
bne vaa_calc_sad_var_height_loop
vadd.i32 d24, d24, d25
vst1.32 {d24[0]}, [r7]
vpop {q6-q7}
vpop {q4}
ldmia sp!, {r4-r11}
WELS_ASM_FUNC_END
#ifdef __APPLE__
.macro SAD_SSD_16
SAD_VAR_16 $0, $1, $2, $3
SSD_MUL_SUM_16BYTES d4,d5,q8, q11
.endm
.macro SAD_SSD_16_END
SAD_VAR_16_END $0, $1, $2
SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
.endm
.macro SAD_SSD_16_RESET_16x16
SAD_VAR_16_RESET_16x16 $0, $1, $2, $3
SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
.endm
.macro SAD_SSD_16_RESET_8x8
SAD_VAR_16_RESET_8x8 $0, $1, $2, $3
SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
.endm
.macro SAD_SSD_16x16
//for one 8x16
SAD_SSD_16_RESET_16x16 $0, $1, $2, q6
SAD_SSD_16 $0, $1, $2, q6
SAD_SSD_16 $0, $1, $2, q6
SAD_SSD_16 $0, $1, $2, q6
SAD_SSD_16 $0, $1, $2, q6
SAD_SSD_16 $0, $1, $2, q6
SAD_SSD_16 $0, $1, $2, q6
SAD_SSD_16 $0, $1, $2, q6
vpaddl.u16 q6, q6
vpaddl.u32 q6, q6
vadd.i32 q12, q6
//for another 8x16
SAD_SSD_16_RESET_8x8 $0, $1, $2, q7
SAD_SSD_16 $0, $1, $2, q7
SAD_SSD_16 $0, $1, $2, q7
SAD_SSD_16 $0, $1, $2, q7
SAD_SSD_16 $0, $1, $2, q7
SAD_SSD_16 $0, $1, $2, q7
SAD_SSD_16 $0, $1, $2, q7
SAD_SSD_16_END $0, $2, q7
vpaddl.u16 q7, q7
vpaddl.u32 q7, q7
vadd.i32 q12, q7
.endm
#else
.macro SAD_SSD_16 arg0, arg1, arg2, arg3
SAD_VAR_16 \arg0, \arg1, \arg2, \arg3
SSD_MUL_SUM_16BYTES d4,d5,q8, q11
.endm
.macro SAD_SSD_16_END arg0, arg1, arg2
SAD_VAR_16_END \arg0, \arg1, \arg2
SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
.endm
.macro SAD_SSD_16_RESET_16x16 arg0, arg1, arg2, arg3
SAD_VAR_16_RESET_16x16 \arg0, \arg1, \arg2, \arg3
SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
.endm
.macro SAD_SSD_16_RESET_8x8 arg0, arg1, arg2, arg3
SAD_VAR_16_RESET_8x8 \arg0, \arg1, \arg2, \arg3
SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
.endm
.macro SAD_SSD_16x16 arg0, arg1, arg2
//for one 8x16
SAD_SSD_16_RESET_16x16 \arg0, \arg1, \arg2, q6
SAD_SSD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_16 \arg0, \arg1, \arg2, q6
SAD_SSD_16 \arg0, \arg1, \arg2, q6
vpaddl.u16 q6, q6
vpaddl.u32 q6, q6
vadd.i32 q12, q6
//for another 8x16
SAD_SSD_16_RESET_8x8 \arg0, \arg1, \arg2, q7
SAD_SSD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_16 \arg0, \arg1, \arg2, q7
SAD_SSD_16_END \arg0, \arg2, q7
vpaddl.u16 q7, q7
vpaddl.u32 q7, q7
vadd.i32 q12, q7
.endm
#endif
WELS_ASM_FUNC_BEGIN VAACalcSadSsd_neon
stmdb sp!, {r4-r12}
vpush {q4}
vpush {q6-q7}
ldr r4, [sp, #84] //r4 keeps the pic_stride
sub r5, r4, #1
lsl r5, r5, #4 //r5 keeps the little step
lsl r6, r4, #4
sub r6, r2, r6 //r6 keeps the big step
ldr r7, [sp, #88] //psadframe
ldr r8, [sp, #92] //psad8x8
ldr r9, [sp, #96] //psum16x16
ldr r10, [sp, #100] //psqsum16x16
ldr r11, [sp, #104] //psqdiff16x16
vmov.i8 q12, #0
vaa_calc_sad_ssd_height_loop:
mov r12, r2
vaa_calc_sad_ssd_width_loop:
SAD_SSD_16x16 r0,r1,r4
//psad8x8
vst4.32 {d12[0], d13[0], d14[0], d15[0]}, [r8]!
sub r0, r0, r5 //jump to next 16x16
sub r1, r1, r5 //jump to next 16x16
//psum16x16
vpaddl.s16 q9, q9
vpaddl.s32 q9, q9
vadd.i32 d18, d18, d19
vst1.32 {d18[0]}, [r9]! //psum16x16
//psqsum16x16
vpaddl.s32 q10, q10
vadd.i32 d20, d20, d21
vst1.32 {d20[0]}, [r10]! //psqsum16x16
//psqdiff16x16
vpaddl.s32 q8, q8
vadd.i32 d16, d16, d17
subs r12, #16
vst1.32 {d16[0]}, [r11]! //psqdiff16x16
bne vaa_calc_sad_ssd_width_loop
sub r0, r0, r6 //jump to next 16 x width
sub r1, r1, r6 //jump to next 16 x width
subs r3, #16
bne vaa_calc_sad_ssd_height_loop
vadd.i32 d24, d24, d25
vst1.32 {d24[0]}, [r7]
vpop {q6-q7}
vpop {q4}
ldmia sp!, {r4-r12}
WELS_ASM_FUNC_END
#endif