ref: 26e1f94bce19cbd01184159cf5a8c6b9a97bad47
dir: /codec/processing/src/arm64/vaa_calc_aarch64_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_AARCH64
#include "arm_arch64_common_macro.S"
.macro ABS_SUB_SUM_16BYTES arg0, arg1
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabal \arg0, v0.8b, v1.8b
uabal2 \arg1, v0.16b,v1.16b
.endm
.macro ABS_SUB_SUM_8x16BYTES arg0, arg1
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabdl \arg0, v0.8b, v1.8b
uabdl2 \arg1, v0.16b,v1.16b
ABS_SUB_SUM_16BYTES \arg0, \arg1
ABS_SUB_SUM_16BYTES \arg0, \arg1
ABS_SUB_SUM_16BYTES \arg0, \arg1
ABS_SUB_SUM_16BYTES \arg0, \arg1
ABS_SUB_SUM_16BYTES \arg0, \arg1
ABS_SUB_SUM_16BYTES \arg0, \arg1
ABS_SUB_SUM_16BYTES \arg0, \arg1
.endm
/*
* void vaa_calc_sad_neon(uint8_t *cur_data, uint8_t *ref_data, int32_t pic_width, int32_t pic_height, int32_t pic_stride,
* int32_t *psadframe, int32_t *psad8x8)
*/
WELS_ASM_AARCH64_FUNC_BEGIN VAACalcSad_AArch64_neon
eor v31.16b, v31.16b, v31.16b
SIGN_EXTENSION x4, w4
lsl x9, x4, #4
sub x10, x9, #16 //x10 keep the 16*pic_stride-16
sub x9, x9, x2 //x9 keep the 16*pic_stride-pic_width
vaa_calc_sad_loop0:
mov w11, w2
vaa_calc_sad_loop1:
ABS_SUB_SUM_8x16BYTES v2.8h, v3.8h
ABS_SUB_SUM_8x16BYTES v4.8h, v5.8h
uaddlv s2, v2.8h
uaddlv s3, v3.8h
uaddlv s4, v4.8h
uaddlv s5, v5.8h
st4 {v2.s, v3.s, v4.s, v5.s}[0], [x6], #16
sub x0, x0, x10
sub x1, x1, x10
sub w11, w11, #16
add v6.2s, v2.2s, v3.2s
add v7.2s, v4.2s, v5.2s
add v6.2s, v6.2s, v7.2s
add v31.2s, v31.2s, v6.2s
cbnz w11, vaa_calc_sad_loop1
add x0, x0, x9
add x1, x1, x9
sub w3, w3, #16
cbnz w3, vaa_calc_sad_loop0
str s31, [x5]
WELS_ASM_AARCH64_FUNC_END
.macro SAD_SD_MAD_8x16BYTES
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v31.16b, v0.16b, v1.16b
uaddlp v2.8h, v31.16b
uaddlp v4.8h, v0.16b
uaddlp v5.8h, v1.16b
.rept 7
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v30.16b, v0.16b, v1.16b
umax v31.16b, v31.16b,v30.16b
uadalp v2.8h, v30.16b
uadalp v4.8h, v0.16b
uadalp v5.8h, v1.16b
.endr
.endm
/*
* void vaa_calc_sad_bgd_neon(uint8_t *cur_data, uint8_t *ref_data, int32_t pic_width, int32_t pic_height, int32_t pic_stride,
* int32_t *psadframe, int32_t *psad8x8, int32_t *p_sd8x8, uint8_t *p_mad8x8)
*/
WELS_ASM_AARCH64_FUNC_BEGIN VAACalcSadBgd_AArch64_neon
ldr x15, [sp, #0]
eor v28.16b, v28.16b, v28.16b
SIGN_EXTENSION x4, w4
lsl x9, x4, #4
sub x10, x9, #16 //x10 keep the 16*pic_stride-16
sub x9, x9, x2 //x9 keep the 16*pic_stride-pic_width
vaa_calc_sad_bgd_loop0:
mov w11, w2
vaa_calc_sad_bgd_loop1:
SAD_SD_MAD_8x16BYTES
umaxv b24, v31.8b
ins v31.d[0], v31.d[1]
umaxv b25, v31.8b
uaddlv s20, v2.4h
ins v2.d[0], v2.d[1]
uaddlv s21, v2.4h
usubl v6.4s, v4.4h, v5.4h
usubl2 v7.4s, v4.8h, v5.8h
addv s16, v6.4s
addv s17, v7.4s
SAD_SD_MAD_8x16BYTES
umaxv b26, v31.8b
ins v31.d[0], v31.d[1]
umaxv b27, v31.8b
uaddlv s22, v2.4h
ins v2.d[0], v2.d[1]
uaddlv s23, v2.4h
usubl v6.4s, v4.4h, v5.4h
usubl2 v7.4s, v4.8h, v5.8h
addv s18, v6.4s
addv s19, v7.4s
st4 {v20.s, v21.s, v22.s, v23.s}[0], [x6], #16
sub x0, x0, x10
sub x1, x1, x10
st4 {v16.s, v17.s, v18.s, v19.s}[0], [x7], #16
sub w11, w11, #16
st4 {v24.b, v25.b, v26.b, v27.b}[0], [x15], #4
add v29.2s, v20.2s, v21.2s
add v30.2s, v22.2s, v23.2s
add v29.2s, v29.2s, v30.2s
add v28.2s, v28.2s, v29.2s
cbnz w11, vaa_calc_sad_bgd_loop1
add x0, x0, x9
add x1, x1, x9
sub w3, w3, #16
cbnz w3, vaa_calc_sad_bgd_loop0
str s28, [x5]
WELS_ASM_AARCH64_FUNC_END
.macro SAD_SSD_BGD_8x16BYTES_1
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v31.16b, v0.16b, v1.16b
umull v30.8h, v31.8b, v31.8b
uaddlp v29.4s, v30.8h
umull2 v30.8h, v31.16b, v31.16b
uadalp v29.4s, v30.8h // p_sqdiff
uaddlp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uaddlp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
uaddlp v2.8h, v31.16b // p_sad
uaddlp v4.8h, v0.16b
uaddlp v5.8h, v1.16b
.rept 7
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v3.16b, v0.16b, v1.16b
umax v31.16b, v31.16b,v3.16b //p_mad
umull v30.8h, v3.8b, v3.8b
uadalp v29.4s, v30.8h
umull2 v30.8h, v3.16b, v3.16b
uadalp v29.4s, v30.8h // p_sqdiff
uadalp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uadalp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
uadalp v2.8h, v3.16b //p_sad
uadalp v4.8h, v0.16b
uadalp v5.8h, v1.16b //p_sd
.endr
.endm
.macro SAD_SSD_BGD_8x16BYTES_2
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v26.16b, v0.16b, v1.16b
umull v30.8h, v26.8b, v26.8b
uadalp v29.4s, v30.8h
umull2 v30.8h, v26.16b, v26.16b
uadalp v29.4s, v30.8h // p_sqdiff
uadalp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uadalp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
uaddlp v16.8h,v26.16b // p_sad
uaddlp v6.8h, v0.16b
uaddlp v7.8h, v1.16b
.rept 7
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v3.16b, v0.16b, v1.16b
umax v26.16b, v26.16b,v3.16b //p_mad
umull v30.8h, v3.8b, v3.8b
uadalp v29.4s, v30.8h
umull2 v30.8h, v3.16b, v3.16b
uadalp v29.4s, v30.8h // p_sqdiff
uadalp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uadalp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
uadalp v16.8h, v3.16b //p_sad
uadalp v6.8h, v0.16b
uadalp v7.8h, v1.16b //p_sd
.endr
.endm
/*
* void vaa_calc_sad_ssd_bgd_c(uint8_t *cur_data, uint8_t *ref_data, int32_t pic_width, int32_t pic_height, int32_t pic_stride,
* int32_t *psadframe,int32_t *psad8x8, int32_t *psum16x16, int32_t *psqsum16x16, int32_t *psqdiff16x16, int32_t *p_sd8x8, uint8_t *p_mad8x8)
*/
WELS_ASM_AARCH64_FUNC_BEGIN VAACalcSadSsdBgd_AArch64_neon
ldr x12, [sp, #0] //psqsum16x16
ldr x13, [sp, #8] //psqdiff16x16
ldr x14, [sp, #16] //p_sd8x8
ldr x15, [sp, #24] //p_mad8x8
eor v17.16b, v17.16b, v17.16b
SIGN_EXTENSION x4, w4
lsl x9, x4, #4
sub x10, x9, #16 //x10 keep the 16*pic_stride-16
sub x9, x9, x2 //x9 keep the 16*pic_stride-pic_width
vaa_calc_sad_ssd_bgd_height_loop:
mov w11, w2
vaa_calc_sad_ssd_bgd_width_loop:
SAD_SSD_BGD_8x16BYTES_1 //psad:v2, v16, psum:v28, psqsum:v27, psqdiff:v29, psd:v4, v5, v6, v7, pmad:v31, v26
SAD_SSD_BGD_8x16BYTES_2
umaxv b22, v31.8b
ins v31.d[0], v31.d[1]
umaxv b23, v31.8b
umaxv b24, v26.8b
ins v26.d[0], v26.d[1]
umaxv b25, v26.8b
st4 {v22.b, v23.b, v24.b, v25.b}[0], [x15], #4
usubl v20.4s, v4.4h, v5.4h
usubl2 v21.4s, v4.8h, v5.8h
addv s20, v20.4s
addv s21, v21.4s
usubl v22.4s, v6.4h, v7.4h
usubl2 v23.4s, v6.8h, v7.8h
addv s22, v22.4s
addv s23, v23.4s
st4 {v20.s, v21.s, v22.s, v23.s}[0], [x14], #16
uaddlv s20, v2.4h
ins v2.d[0], v2.d[1]
uaddlv s21, v2.4h
uaddlv s22, v16.4h
ins v16.d[0], v16.d[1]
uaddlv s23, v16.4h
st4 {v20.s, v21.s, v22.s, v23.s}[0], [x6], #16
uaddlv s28, v28.8h
str s28, [x7], #4
addv s27, v27.4s
str s27, [x12], #4
addv s29, v29.4s
str s29, [x13], #4
sub x0, x0, x10
sub x1, x1, x10
sub w11, w11, #16
add v29.2s, v20.2s, v21.2s
add v30.2s, v22.2s, v23.2s
add v29.2s, v29.2s, v30.2s
add v17.2s, v17.2s, v29.2s
cbnz w11, vaa_calc_sad_ssd_bgd_width_loop
add x0, x0, x9
add x1, x1, x9
sub w3, w3, #16
cbnz w3, vaa_calc_sad_ssd_bgd_height_loop
str s17, [x5]
WELS_ASM_AARCH64_FUNC_END
.macro SAD_SSD_8x16BYTES_1
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v31.16b, v0.16b, v1.16b
umull v30.8h, v31.8b, v31.8b
uaddlp v29.4s, v30.8h
umull2 v30.8h, v31.16b, v31.16b
uadalp v29.4s, v30.8h // p_sqdiff
uaddlp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uaddlp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
uaddlp v2.8h, v31.16b // p_sad
.rept 7
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v3.16b, v0.16b, v1.16b
umull v30.8h, v3.8b, v3.8b
uadalp v29.4s, v30.8h
umull2 v30.8h, v3.16b, v3.16b
uadalp v29.4s, v30.8h // p_sqdiff
uadalp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uadalp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
uadalp v2.8h, v3.16b //p_sad
.endr
.endm
.macro SAD_SSD_8x16BYTES_2
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v26.16b, v0.16b, v1.16b
umull v30.8h, v26.8b, v26.8b
uadalp v29.4s, v30.8h
umull2 v30.8h, v26.16b, v26.16b
uadalp v29.4s, v30.8h // p_sqdiff
uadalp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uadalp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
uaddlp v16.8h,v26.16b // p_sad
uaddlp v6.8h, v0.16b
uaddlp v7.8h, v1.16b
.rept 7
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v3.16b, v0.16b, v1.16b
umull v30.8h, v3.8b, v3.8b
uadalp v29.4s, v30.8h
umull2 v30.8h, v3.16b, v3.16b
uadalp v29.4s, v30.8h // p_sqdiff
uadalp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uadalp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
uadalp v16.8h, v3.16b //p_sad
.endr
.endm
/*
* void vaa_calc_sad_ssd_c(uint8_t *cur_data, uint8_t *ref_data, int32_t pic_width, int32_t pic_height, int32_t pic_stride,
* int32_t *psadframe,int32_t *psad8x8, int32_t *psum16x16, int32_t *psqsum16x16, int32_t *psqdiff16x16)
*/
WELS_ASM_AARCH64_FUNC_BEGIN VAACalcSadSsd_AArch64_neon
ldr x12, [sp, #0] //psqsum16x16
ldr x13, [sp, #8] //psqdiff16x16
eor v17.16b, v17.16b, v17.16b
SIGN_EXTENSION x4, w4
lsl x9, x4, #4
sub x10, x9, #16 //x10 keep the 16*pic_stride-16
sub x9, x9, x2 //x9 keep the 16*pic_stride-pic_width
vaa_calc_sad_ssd_height_loop:
mov w11, w2
vaa_calc_sad_ssd_width_loop:
SAD_SSD_8x16BYTES_1 //psad:v2, v16, psum:v28, psqsum:v27, psqdiff:v29
SAD_SSD_8x16BYTES_2
uaddlv s20, v2.4h
ins v2.d[0], v2.d[1]
uaddlv s21, v2.4h
uaddlv s22, v16.4h
ins v16.d[0], v16.d[1]
uaddlv s23, v16.4h
st4 {v20.s, v21.s, v22.s, v23.s}[0], [x6], #16
uaddlv s28, v28.8h
str s28, [x7], #4
addv s27, v27.4s
str s27, [x12], #4
addv s29, v29.4s
str s29, [x13], #4
sub x0, x0, x10
sub x1, x1, x10
sub w11, w11, #16
add v29.2s, v20.2s, v21.2s
add v30.2s, v22.2s, v23.2s
add v29.2s, v29.2s, v30.2s
add v17.2s, v17.2s, v29.2s
cbnz w11, vaa_calc_sad_ssd_width_loop
add x0, x0, x9
add x1, x1, x9
sub w3, w3, #16
cbnz w3, vaa_calc_sad_ssd_height_loop
str s17, [x5]
WELS_ASM_AARCH64_FUNC_END
.macro SAD_VAR_8x16BYTES_1
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v31.16b, v0.16b, v1.16b
uaddlp v2.8h, v31.16b // p_sad
uaddlp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uaddlp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
.rept 7
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v3.16b, v0.16b, v1.16b
uadalp v2.8h, v3.16b //p_sad
uadalp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uadalp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
.endr
.endm
.macro SAD_VAR_8x16BYTES_2
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v26.16b, v0.16b, v1.16b
uaddlp v16.8h,v26.16b // p_sad
uadalp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uadalp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
.rept 7
ld1 {v0.16b}, [x0], x4
ld1 {v1.16b}, [x1], x4
uabd v3.16b, v0.16b, v1.16b
uadalp v16.8h, v3.16b //p_sad
uadalp v28.8h, v0.16b // p_sum
umull v30.8h, v0.8b, v0.8b
uadalp v27.4s, v30.8h
umull2 v30.8h, v0.16b, v0.16b
uadalp v27.4s, v30.8h // p_sqsum
.endr
.endm
/*
* void vaa_calc_sad_var_c(uint8_t *cur_data, uint8_t *ref_data, int32_t pic_width, int32_t pic_height, int32_t pic_stride,
* int32_t *psadframe,int32_t *psad8x8, int32_t *psum16x16, int32_t *psqsum16x16)
*/
WELS_ASM_AARCH64_FUNC_BEGIN VAACalcSadVar_AArch64_neon
ldr x12, [sp, #0] //psqsum16x16
eor v17.16b, v17.16b, v17.16b
SIGN_EXTENSION x4, w4
lsl x9, x4, #4
sub x10, x9, #16 //x10 keep the 16*pic_stride-16
sub x9, x9, x2 //x9 keep the 16*pic_stride-pic_width
vaa_calc_sad_var_height_loop:
mov w11, w2
vaa_calc_sad_var_width_loop:
SAD_VAR_8x16BYTES_1 //psad:v2, v16, psum:v28, psqsum:v27
SAD_VAR_8x16BYTES_2
uaddlv s20, v2.4h
ins v2.d[0], v2.d[1]
uaddlv s21, v2.4h
uaddlv s22, v16.4h
ins v16.d[0], v16.d[1]
uaddlv s23, v16.4h
st4 {v20.s, v21.s, v22.s, v23.s}[0], [x6], #16
uaddlv s28, v28.8h
str s28, [x7], #4
addv s27, v27.4s
str s27, [x12], #4
sub x0, x0, x10
sub x1, x1, x10
sub w11, w11, #16
add v29.2s, v20.2s, v21.2s
add v30.2s, v22.2s, v23.2s
add v29.2s, v29.2s, v30.2s
add v17.2s, v17.2s, v29.2s
cbnz w11, vaa_calc_sad_var_width_loop
add x0, x0, x9
add x1, x1, x9
sub w3, w3, #16
cbnz w3, vaa_calc_sad_var_height_loop
str s17, [x5]
WELS_ASM_AARCH64_FUNC_END
#endif