har0ke/FFmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
FFmpeg/libavcodec/loongarch/h264dsp_lasx.c
Hao Chen 7845b5ecd6
avcodec/la: Add LSX optimization for loop filter. 
Replaced function(LSX is sufficient for these functions):
ff_h264_v_lpf_chroma_8_lasx
ff_h264_h_lpf_chroma_8_lasx
ff_h264_v_lpf_chroma_intra_8_lasx
ff_h264_h_lpf_chroma_intra_8_lasx
ff_weight_h264_pixels4_8_lasx
ff_biweight_h264_pixels4_8_lasx

./configure --disable-lasx
ffmpeg -i 1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -y /dev/null -an
before: 161fps
after:  199fps

Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn>
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2023-05-25 21:04:43 +02:00

783 lines
36 KiB
C
Raw Blame History

/*
* Loongson LASX optimized h264dsp
*
* Copyright (c) 2021 Loongson Technology Corporation Limited
* Contributed by Shiyou Yin <yinshiyou-hf@loongson.cn>
* Xiwei Gu <guxiwei-hf@loongson.cn>
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "libavutil/loongarch/loongson_intrinsics.h"
#include "h264dsp_loongarch.h"
#define AVC_LPF_P1_OR_Q1(p0_or_q0_org_in, q0_or_p0_org_in, \
p1_or_q1_org_in, p2_or_q2_org_in, \
neg_tc_in, tc_in, p1_or_q1_out) \
{ \
__m256i clip3, temp; \
\
clip3 = __lasx_xvavgr_hu(p0_or_q0_org_in, \
q0_or_p0_org_in); \
temp = __lasx_xvslli_h(p1_or_q1_org_in, 1); \
clip3 = __lasx_xvsub_h(clip3, temp); \
clip3 = __lasx_xvavg_h(p2_or_q2_org_in, clip3); \
clip3 = __lasx_xvclip_h(clip3, neg_tc_in, tc_in); \
p1_or_q1_out = __lasx_xvadd_h(p1_or_q1_org_in, clip3); \
}
#define AVC_LPF_P0Q0(q0_or_p0_org_in, p0_or_q0_org_in, \
p1_or_q1_org_in, q1_or_p1_org_in, \
neg_threshold_in, threshold_in, \
p0_or_q0_out, q0_or_p0_out) \
{ \
__m256i q0_sub_p0, p1_sub_q1, delta; \
\
q0_sub_p0 = __lasx_xvsub_h(q0_or_p0_org_in, \
p0_or_q0_org_in); \
p1_sub_q1 = __lasx_xvsub_h(p1_or_q1_org_in, \
q1_or_p1_org_in); \
q0_sub_p0 = __lasx_xvslli_h(q0_sub_p0, 2); \
p1_sub_q1 = __lasx_xvaddi_hu(p1_sub_q1, 4); \
delta = __lasx_xvadd_h(q0_sub_p0, p1_sub_q1); \
delta = __lasx_xvsrai_h(delta, 3); \
delta = __lasx_xvclip_h(delta, neg_threshold_in, \
threshold_in); \
p0_or_q0_out = __lasx_xvadd_h(p0_or_q0_org_in, delta); \
q0_or_p0_out = __lasx_xvsub_h(q0_or_p0_org_in, delta); \
\
p0_or_q0_out = __lasx_xvclip255_h(p0_or_q0_out); \
q0_or_p0_out = __lasx_xvclip255_h(q0_or_p0_out); \
}
void ff_h264_h_lpf_luma_8_lasx(uint8_t *data, ptrdiff_t img_width,
int alpha_in, int beta_in, int8_t *tc)
{
int img_width_2x = img_width << 1;
int img_width_4x = img_width << 2;
int img_width_8x = img_width << 3;
int img_width_3x = img_width_2x + img_width;
__m256i tmp_vec0, bs_vec;
__m256i tc_vec = {0x0101010100000000, 0x0303030302020202,
0x0101010100000000, 0x0303030302020202};
tmp_vec0 = __lasx_xvldrepl_w((uint32_t*)tc, 0);
tc_vec = __lasx_xvshuf_b(tmp_vec0, tmp_vec0, tc_vec);
bs_vec = __lasx_xvslti_b(tc_vec, 0);
bs_vec = __lasx_xvxori_b(bs_vec, 255);
bs_vec = __lasx_xvandi_b(bs_vec, 1);
if (__lasx_xbnz_v(bs_vec)) {
uint8_t *src = data - 4;
__m256i p3_org, p2_org, p1_org, p0_org, q0_org, q1_org, q2_org, q3_org;
__m256i p0_asub_q0, p1_asub_p0, q1_asub_q0, alpha, beta;
__m256i is_less_than, is_less_than_beta, is_less_than_alpha;
__m256i is_bs_greater_than0;
__m256i zero = __lasx_xvldi(0);
is_bs_greater_than0 = __lasx_xvslt_bu(zero, bs_vec);
{
uint8_t *src_tmp = src + img_width_8x;
__m256i row0, row1, row2, row3, row4, row5, row6, row7;
__m256i row8, row9, row10, row11, row12, row13, row14, row15;
DUP4_ARG2(__lasx_xvldx, src, 0, src, img_width, src, img_width_2x,
src, img_width_3x, row0, row1, row2, row3);
src += img_width_4x;
DUP4_ARG2(__lasx_xvldx, src, 0, src, img_width, src, img_width_2x,
src, img_width_3x, row4, row5, row6, row7);
src -= img_width_4x;
DUP4_ARG2(__lasx_xvldx, src_tmp, 0, src_tmp, img_width, src_tmp,
img_width_2x, src_tmp, img_width_3x,
row8, row9, row10, row11);
src_tmp += img_width_4x;
DUP4_ARG2(__lasx_xvldx, src_tmp, 0, src_tmp, img_width, src_tmp,
img_width_2x, src_tmp, img_width_3x,
row12, row13, row14, row15);
src_tmp -= img_width_4x;
LASX_TRANSPOSE16x8_B(row0, row1, row2, row3, row4, row5, row6,
row7, row8, row9, row10, row11,
row12, row13, row14, row15,
p3_org, p2_org, p1_org, p0_org,
q0_org, q1_org, q2_org, q3_org);
}
p0_asub_q0 = __lasx_xvabsd_bu(p0_org, q0_org);
p1_asub_p0 = __lasx_xvabsd_bu(p1_org, p0_org);
q1_asub_q0 = __lasx_xvabsd_bu(q1_org, q0_org);
alpha = __lasx_xvreplgr2vr_b(alpha_in);
beta = __lasx_xvreplgr2vr_b(beta_in);
is_less_than_alpha = __lasx_xvslt_bu(p0_asub_q0, alpha);
is_less_than_beta = __lasx_xvslt_bu(p1_asub_p0, beta);
is_less_than = is_less_than_alpha & is_less_than_beta;
is_less_than_beta = __lasx_xvslt_bu(q1_asub_q0, beta);
is_less_than = is_less_than_beta & is_less_than;
is_less_than = is_less_than & is_bs_greater_than0;
if (__lasx_xbnz_v(is_less_than)) {
__m256i neg_tc_h, tc_h, p1_org_h, p0_org_h, q0_org_h, q1_org_h;
__m256i p2_asub_p0, q2_asub_q0;
neg_tc_h = __lasx_xvneg_b(tc_vec);
neg_tc_h = __lasx_vext2xv_h_b(neg_tc_h);
tc_h = __lasx_vext2xv_hu_bu(tc_vec);
p1_org_h = __lasx_vext2xv_hu_bu(p1_org);
p0_org_h = __lasx_vext2xv_hu_bu(p0_org);
q0_org_h = __lasx_vext2xv_hu_bu(q0_org);
p2_asub_p0 = __lasx_xvabsd_bu(p2_org, p0_org);
is_less_than_beta = __lasx_xvslt_bu(p2_asub_p0, beta);
is_less_than_beta = is_less_than_beta & is_less_than;
if (__lasx_xbnz_v(is_less_than_beta)) {
__m256i p2_org_h, p1_h;
p2_org_h = __lasx_vext2xv_hu_bu(p2_org);
AVC_LPF_P1_OR_Q1(p0_org_h, q0_org_h, p1_org_h, p2_org_h,
neg_tc_h, tc_h, p1_h);
p1_h = __lasx_xvpickev_b(p1_h, p1_h);
p1_h = __lasx_xvpermi_d(p1_h, 0xd8);
p1_org = __lasx_xvbitsel_v(p1_org, p1_h, is_less_than_beta);
is_less_than_beta = __lasx_xvandi_b(is_less_than_beta, 1);
tc_vec = __lasx_xvadd_b(tc_vec, is_less_than_beta);
}
q2_asub_q0 = __lasx_xvabsd_bu(q2_org, q0_org);
is_less_than_beta = __lasx_xvslt_bu(q2_asub_q0, beta);
is_less_than_beta = is_less_than_beta & is_less_than;
q1_org_h = __lasx_vext2xv_hu_bu(q1_org);
if (__lasx_xbnz_v(is_less_than_beta)) {
__m256i q2_org_h, q1_h;
q2_org_h = __lasx_vext2xv_hu_bu(q2_org);
AVC_LPF_P1_OR_Q1(p0_org_h, q0_org_h, q1_org_h, q2_org_h,
neg_tc_h, tc_h, q1_h);
q1_h = __lasx_xvpickev_b(q1_h, q1_h);
q1_h = __lasx_xvpermi_d(q1_h, 0xd8);
q1_org = __lasx_xvbitsel_v(q1_org, q1_h, is_less_than_beta);
is_less_than_beta = __lasx_xvandi_b(is_less_than_beta, 1);
tc_vec = __lasx_xvadd_b(tc_vec, is_less_than_beta);
}
{
__m256i neg_thresh_h, p0_h, q0_h;
neg_thresh_h = __lasx_xvneg_b(tc_vec);
neg_thresh_h = __lasx_vext2xv_h_b(neg_thresh_h);
tc_h = __lasx_vext2xv_hu_bu(tc_vec);
AVC_LPF_P0Q0(q0_org_h, p0_org_h, p1_org_h, q1_org_h,
neg_thresh_h, tc_h, p0_h, q0_h);
DUP2_ARG2(__lasx_xvpickev_b, p0_h, p0_h, q0_h, q0_h,
p0_h, q0_h);
DUP2_ARG2(__lasx_xvpermi_d, p0_h, 0xd8, q0_h, 0xd8,
p0_h, q0_h);
p0_org = __lasx_xvbitsel_v(p0_org, p0_h, is_less_than);
q0_org = __lasx_xvbitsel_v(q0_org, q0_h, is_less_than);
}
{
__m256i row0, row1, row2, row3, row4, row5, row6, row7;
__m256i control = {0x0000000400000000, 0x0000000500000001,
0x0000000600000002, 0x0000000700000003};
DUP4_ARG3(__lasx_xvpermi_q, p0_org, q3_org, 0x02, p1_org,
q2_org, 0x02, p2_org, q1_org, 0x02, p3_org,
q0_org, 0x02, p0_org, p1_org, p2_org, p3_org);
DUP2_ARG2(__lasx_xvilvl_b, p1_org, p3_org, p0_org, p2_org,
row0, row2);
DUP2_ARG2(__lasx_xvilvh_b, p1_org, p3_org, p0_org, p2_org,
row1, row3);
DUP2_ARG2(__lasx_xvilvl_b, row2, row0, row3, row1, row4, row6);
DUP2_ARG2(__lasx_xvilvh_b, row2, row0, row3, row1, row5, row7);
DUP4_ARG2(__lasx_xvperm_w, row4, control, row5, control, row6,
control, row7, control, row4, row5, row6, row7);
__lasx_xvstelm_d(row4, src, 0, 0);
__lasx_xvstelm_d(row4, src + img_width, 0, 1);
src += img_width_2x;
__lasx_xvstelm_d(row4, src, 0, 2);
__lasx_xvstelm_d(row4, src + img_width, 0, 3);
src += img_width_2x;
__lasx_xvstelm_d(row5, src, 0, 0);
__lasx_xvstelm_d(row5, src + img_width, 0, 1);
src += img_width_2x;
__lasx_xvstelm_d(row5, src, 0, 2);
__lasx_xvstelm_d(row5, src + img_width, 0, 3);
src += img_width_2x;
__lasx_xvstelm_d(row6, src, 0, 0);
__lasx_xvstelm_d(row6, src + img_width, 0, 1);
src += img_width_2x;
__lasx_xvstelm_d(row6, src, 0, 2);
__lasx_xvstelm_d(row6, src + img_width, 0, 3);
src += img_width_2x;
__lasx_xvstelm_d(row7, src, 0, 0);
__lasx_xvstelm_d(row7, src + img_width, 0, 1);
src += img_width_2x;
__lasx_xvstelm_d(row7, src, 0, 2);
__lasx_xvstelm_d(row7, src + img_width, 0, 3);
}
}
}
}
void ff_h264_v_lpf_luma_8_lasx(uint8_t *data, ptrdiff_t img_width,
int alpha_in, int beta_in, int8_t *tc)
{
int img_width_2x = img_width << 1;
int img_width_3x = img_width + img_width_2x;
__m256i tmp_vec0, bs_vec;
__m256i tc_vec = {0x0101010100000000, 0x0303030302020202,
0x0101010100000000, 0x0303030302020202};
tmp_vec0 = __lasx_xvldrepl_w((uint32_t*)tc, 0);
tc_vec = __lasx_xvshuf_b(tmp_vec0, tmp_vec0, tc_vec);
bs_vec = __lasx_xvslti_b(tc_vec, 0);
bs_vec = __lasx_xvxori_b(bs_vec, 255);
bs_vec = __lasx_xvandi_b(bs_vec, 1);
if (__lasx_xbnz_v(bs_vec)) {
__m256i p2_org, p1_org, p0_org, q0_org, q1_org, q2_org;
__m256i p0_asub_q0, p1_asub_p0, q1_asub_q0, alpha, beta;
__m256i is_less_than, is_less_than_beta, is_less_than_alpha;
__m256i p1_org_h, p0_org_h, q0_org_h, q1_org_h;
__m256i is_bs_greater_than0;
__m256i zero = __lasx_xvldi(0);
alpha = __lasx_xvreplgr2vr_b(alpha_in);
beta = __lasx_xvreplgr2vr_b(beta_in);
DUP2_ARG2(__lasx_xvldx, data, -img_width_3x, data, -img_width_2x,
p2_org, p1_org);
p0_org = __lasx_xvldx(data, -img_width);
DUP2_ARG2(__lasx_xvldx, data, 0, data, img_width, q0_org, q1_org);
is_bs_greater_than0 = __lasx_xvslt_bu(zero, bs_vec);
p0_asub_q0 = __lasx_xvabsd_bu(p0_org, q0_org);
p1_asub_p0 = __lasx_xvabsd_bu(p1_org, p0_org);
q1_asub_q0 = __lasx_xvabsd_bu(q1_org, q0_org);
is_less_than_alpha = __lasx_xvslt_bu(p0_asub_q0, alpha);
is_less_than_beta = __lasx_xvslt_bu(p1_asub_p0, beta);
is_less_than = is_less_than_alpha & is_less_than_beta;
is_less_than_beta = __lasx_xvslt_bu(q1_asub_q0, beta);
is_less_than = is_less_than_beta & is_less_than;
is_less_than = is_less_than & is_bs_greater_than0;
if (__lasx_xbnz_v(is_less_than)) {
__m256i neg_tc_h, tc_h, p2_asub_p0, q2_asub_q0;
q2_org = __lasx_xvldx(data, img_width_2x);
neg_tc_h = __lasx_xvneg_b(tc_vec);
neg_tc_h = __lasx_vext2xv_h_b(neg_tc_h);
tc_h = __lasx_vext2xv_hu_bu(tc_vec);
p1_org_h = __lasx_vext2xv_hu_bu(p1_org);
p0_org_h = __lasx_vext2xv_hu_bu(p0_org);
q0_org_h = __lasx_vext2xv_hu_bu(q0_org);
p2_asub_p0 = __lasx_xvabsd_bu(p2_org, p0_org);
is_less_than_beta = __lasx_xvslt_bu(p2_asub_p0, beta);
is_less_than_beta = is_less_than_beta & is_less_than;
if (__lasx_xbnz_v(is_less_than_beta)) {
__m256i p1_h, p2_org_h;
p2_org_h = __lasx_vext2xv_hu_bu(p2_org);
AVC_LPF_P1_OR_Q1(p0_org_h, q0_org_h, p1_org_h, p2_org_h,
neg_tc_h, tc_h, p1_h);
p1_h = __lasx_xvpickev_b(p1_h, p1_h);
p1_h = __lasx_xvpermi_d(p1_h, 0xd8);
p1_h = __lasx_xvbitsel_v(p1_org, p1_h, is_less_than_beta);
p1_org = __lasx_xvpermi_q(p1_org, p1_h, 0x30);
__lasx_xvst(p1_org, data - img_width_2x, 0);
is_less_than_beta = __lasx_xvandi_b(is_less_than_beta, 1);
tc_vec = __lasx_xvadd_b(tc_vec, is_less_than_beta);
}
q2_asub_q0 = __lasx_xvabsd_bu(q2_org, q0_org);
is_less_than_beta = __lasx_xvslt_bu(q2_asub_q0, beta);
is_less_than_beta = is_less_than_beta & is_less_than;
q1_org_h = __lasx_vext2xv_hu_bu(q1_org);
if (__lasx_xbnz_v(is_less_than_beta)) {
__m256i q1_h, q2_org_h;
q2_org_h = __lasx_vext2xv_hu_bu(q2_org);
AVC_LPF_P1_OR_Q1(p0_org_h, q0_org_h, q1_org_h, q2_org_h,
neg_tc_h, tc_h, q1_h);
q1_h = __lasx_xvpickev_b(q1_h, q1_h);
q1_h = __lasx_xvpermi_d(q1_h, 0xd8);
q1_h = __lasx_xvbitsel_v(q1_org, q1_h, is_less_than_beta);
q1_org = __lasx_xvpermi_q(q1_org, q1_h, 0x30);
__lasx_xvst(q1_org, data + img_width, 0);
is_less_than_beta = __lasx_xvandi_b(is_less_than_beta, 1);
tc_vec = __lasx_xvadd_b(tc_vec, is_less_than_beta);
}
{
__m256i neg_thresh_h, p0_h, q0_h;
neg_thresh_h = __lasx_xvneg_b(tc_vec);
neg_thresh_h = __lasx_vext2xv_h_b(neg_thresh_h);
tc_h = __lasx_vext2xv_hu_bu(tc_vec);
AVC_LPF_P0Q0(q0_org_h, p0_org_h, p1_org_h, q1_org_h,
neg_thresh_h, tc_h, p0_h, q0_h);
DUP2_ARG2(__lasx_xvpickev_b, p0_h, p0_h, q0_h, q0_h,
p0_h, q0_h);
DUP2_ARG2(__lasx_xvpermi_d, p0_h, 0Xd8, q0_h, 0xd8,
p0_h, q0_h);
p0_h = __lasx_xvbitsel_v(p0_org, p0_h, is_less_than);
q0_h = __lasx_xvbitsel_v(q0_org, q0_h, is_less_than);
p0_org = __lasx_xvpermi_q(p0_org, p0_h, 0x30);
q0_org = __lasx_xvpermi_q(q0_org, q0_h, 0x30);
__lasx_xvst(p0_org, data - img_width, 0);
__lasx_xvst(q0_org, data, 0);
}
}
}
}
#define AVC_LPF_P0P1P2_OR_Q0Q1Q2(p3_or_q3_org_in, p0_or_q0_org_in, \
q3_or_p3_org_in, p1_or_q1_org_in, \
p2_or_q2_org_in, q1_or_p1_org_in, \
p0_or_q0_out, p1_or_q1_out, p2_or_q2_out) \
{ \
__m256i threshold; \
__m256i const2, const3 = __lasx_xvldi(0); \
\
const2 = __lasx_xvaddi_hu(const3, 2); \
const3 = __lasx_xvaddi_hu(const3, 3); \
threshold = __lasx_xvadd_h(p0_or_q0_org_in, q3_or_p3_org_in); \
threshold = __lasx_xvadd_h(p1_or_q1_org_in, threshold); \
\
p0_or_q0_out = __lasx_xvslli_h(threshold, 1); \
p0_or_q0_out = __lasx_xvadd_h(p0_or_q0_out, p2_or_q2_org_in); \
p0_or_q0_out = __lasx_xvadd_h(p0_or_q0_out, q1_or_p1_org_in); \
p0_or_q0_out = __lasx_xvsrar_h(p0_or_q0_out, const3); \
\
p1_or_q1_out = __lasx_xvadd_h(p2_or_q2_org_in, threshold); \
p1_or_q1_out = __lasx_xvsrar_h(p1_or_q1_out, const2); \
\
p2_or_q2_out = __lasx_xvmul_h(p2_or_q2_org_in, const3); \
p2_or_q2_out = __lasx_xvadd_h(p2_or_q2_out, p3_or_q3_org_in); \
p2_or_q2_out = __lasx_xvadd_h(p2_or_q2_out, p3_or_q3_org_in); \
p2_or_q2_out = __lasx_xvadd_h(p2_or_q2_out, threshold); \
p2_or_q2_out = __lasx_xvsrar_h(p2_or_q2_out, const3); \
}
/* data[-u32_img_width] = (uint8_t)((2 * p1 + p0 + q1 + 2) >> 2); */
#define AVC_LPF_P0_OR_Q0(p0_or_q0_org_in, q1_or_p1_org_in, \
p1_or_q1_org_in, p0_or_q0_out) \
{ \
__m256i const2 = __lasx_xvldi(0); \
const2 = __lasx_xvaddi_hu(const2, 2); \
p0_or_q0_out = __lasx_xvadd_h(p0_or_q0_org_in, q1_or_p1_org_in); \
p0_or_q0_out = __lasx_xvadd_h(p0_or_q0_out, p1_or_q1_org_in); \
p0_or_q0_out = __lasx_xvadd_h(p0_or_q0_out, p1_or_q1_org_in); \
p0_or_q0_out = __lasx_xvsrar_h(p0_or_q0_out, const2); \
}
void ff_h264_h_lpf_luma_intra_8_lasx(uint8_t *data, ptrdiff_t img_width,
int alpha_in, int beta_in)
{
int img_width_2x = img_width << 1;
int img_width_4x = img_width << 2;
int img_width_3x = img_width_2x + img_width;
uint8_t *src = data - 4;
__m256i p0_asub_q0, p1_asub_p0, q1_asub_q0, alpha, beta;
__m256i is_less_than, is_less_than_beta, is_less_than_alpha;
__m256i p3_org, p2_org, p1_org, p0_org, q0_org, q1_org, q2_org, q3_org;
__m256i zero = __lasx_xvldi(0);
{
__m256i row0, row1, row2, row3, row4, row5, row6, row7;
__m256i row8, row9, row10, row11, row12, row13, row14, row15;
DUP4_ARG2(__lasx_xvldx, src, 0, src, img_width, src, img_width_2x,
src, img_width_3x, row0, row1, row2, row3);
src += img_width_4x;
DUP4_ARG2(__lasx_xvldx, src, 0, src, img_width, src, img_width_2x,
src, img_width_3x, row4, row5, row6, row7);
src += img_width_4x;
DUP4_ARG2(__lasx_xvldx, src, 0, src, img_width, src, img_width_2x,
src, img_width_3x, row8, row9, row10, row11);
src += img_width_4x;
DUP4_ARG2(__lasx_xvldx, src, 0, src, img_width, src, img_width_2x,
src, img_width_3x, row12, row13, row14, row15);
src += img_width_4x;
LASX_TRANSPOSE16x8_B(row0, row1, row2, row3,
row4, row5, row6, row7,
row8, row9, row10, row11,
row12, row13, row14, row15,
p3_org, p2_org, p1_org, p0_org,
q0_org, q1_org, q2_org, q3_org);
}
alpha = __lasx_xvreplgr2vr_b(alpha_in);
beta = __lasx_xvreplgr2vr_b(beta_in);
p0_asub_q0 = __lasx_xvabsd_bu(p0_org, q0_org);
p1_asub_p0 = __lasx_xvabsd_bu(p1_org, p0_org);
q1_asub_q0 = __lasx_xvabsd_bu(q1_org, q0_org);
is_less_than_alpha = __lasx_xvslt_bu(p0_asub_q0, alpha);
is_less_than_beta = __lasx_xvslt_bu(p1_asub_p0, beta);
is_less_than = is_less_than_beta & is_less_than_alpha;
is_less_than_beta = __lasx_xvslt_bu(q1_asub_q0, beta);
is_less_than = is_less_than_beta & is_less_than;
is_less_than = __lasx_xvpermi_q(zero, is_less_than, 0x30);
if (__lasx_xbnz_v(is_less_than)) {
__m256i p2_asub_p0, q2_asub_q0, p0_h, q0_h, negate_is_less_than_beta;
__m256i p1_org_h, p0_org_h, q0_org_h, q1_org_h;
__m256i less_alpha_shift2_add2 = __lasx_xvsrli_b(alpha, 2);
less_alpha_shift2_add2 = __lasx_xvaddi_bu(less_alpha_shift2_add2, 2);
less_alpha_shift2_add2 = __lasx_xvslt_bu(p0_asub_q0,
less_alpha_shift2_add2);
p1_org_h = __lasx_vext2xv_hu_bu(p1_org);
p0_org_h = __lasx_vext2xv_hu_bu(p0_org);
q0_org_h = __lasx_vext2xv_hu_bu(q0_org);
q1_org_h = __lasx_vext2xv_hu_bu(q1_org);
p2_asub_p0 = __lasx_xvabsd_bu(p2_org, p0_org);
is_less_than_beta = __lasx_xvslt_bu(p2_asub_p0, beta);
is_less_than_beta = is_less_than_beta & less_alpha_shift2_add2;
negate_is_less_than_beta = __lasx_xvxori_b(is_less_than_beta, 0xff);
is_less_than_beta = is_less_than_beta & is_less_than;
negate_is_less_than_beta = negate_is_less_than_beta & is_less_than;
/* combine and store */
if (__lasx_xbnz_v(is_less_than_beta)) {
__m256i p2_org_h, p3_org_h, p1_h, p2_h;
p2_org_h = __lasx_vext2xv_hu_bu(p2_org);
p3_org_h = __lasx_vext2xv_hu_bu(p3_org);
AVC_LPF_P0P1P2_OR_Q0Q1Q2(p3_org_h, p0_org_h, q0_org_h, p1_org_h,
p2_org_h, q1_org_h, p0_h, p1_h, p2_h);
p0_h = __lasx_xvpickev_b(p0_h, p0_h);
p0_h = __lasx_xvpermi_d(p0_h, 0xd8);
DUP2_ARG2(__lasx_xvpickev_b, p1_h, p1_h, p2_h, p2_h, p1_h, p2_h);
DUP2_ARG2(__lasx_xvpermi_d, p1_h, 0xd8, p2_h, 0xd8, p1_h, p2_h);
p0_org = __lasx_xvbitsel_v(p0_org, p0_h, is_less_than_beta);
p1_org = __lasx_xvbitsel_v(p1_org, p1_h, is_less_than_beta);
p2_org = __lasx_xvbitsel_v(p2_org, p2_h, is_less_than_beta);
}
AVC_LPF_P0_OR_Q0(p0_org_h, q1_org_h, p1_org_h, p0_h);
/* combine */
p0_h = __lasx_xvpickev_b(p0_h, p0_h);
p0_h = __lasx_xvpermi_d(p0_h, 0xd8);
p0_org = __lasx_xvbitsel_v(p0_org, p0_h, negate_is_less_than_beta);
/* if (tmpFlag && (unsigned)ABS(q2-q0) < thresholds->beta_in) */
q2_asub_q0 = __lasx_xvabsd_bu(q2_org, q0_org);
is_less_than_beta = __lasx_xvslt_bu(q2_asub_q0, beta);
is_less_than_beta = is_less_than_beta & less_alpha_shift2_add2;
negate_is_less_than_beta = __lasx_xvxori_b(is_less_than_beta, 0xff);
is_less_than_beta = is_less_than_beta & is_less_than;
negate_is_less_than_beta = negate_is_less_than_beta & is_less_than;
/* combine and store */
if (__lasx_xbnz_v(is_less_than_beta)) {
__m256i q2_org_h, q3_org_h, q1_h, q2_h;
q2_org_h = __lasx_vext2xv_hu_bu(q2_org);
q3_org_h = __lasx_vext2xv_hu_bu(q3_org);
AVC_LPF_P0P1P2_OR_Q0Q1Q2(q3_org_h, q0_org_h, p0_org_h, q1_org_h,
q2_org_h, p1_org_h, q0_h, q1_h, q2_h);
q0_h = __lasx_xvpickev_b(q0_h, q0_h);
q0_h = __lasx_xvpermi_d(q0_h, 0xd8);
DUP2_ARG2(__lasx_xvpickev_b, q1_h, q1_h, q2_h, q2_h, q1_h, q2_h);
DUP2_ARG2(__lasx_xvpermi_d, q1_h, 0xd8, q2_h, 0xd8, q1_h, q2_h);
q0_org = __lasx_xvbitsel_v(q0_org, q0_h, is_less_than_beta);
q1_org = __lasx_xvbitsel_v(q1_org, q1_h, is_less_than_beta);
q2_org = __lasx_xvbitsel_v(q2_org, q2_h, is_less_than_beta);
}
AVC_LPF_P0_OR_Q0(q0_org_h, p1_org_h, q1_org_h, q0_h);
/* combine */
q0_h = __lasx_xvpickev_b(q0_h, q0_h);
q0_h = __lasx_xvpermi_d(q0_h, 0xd8);
q0_org = __lasx_xvbitsel_v(q0_org, q0_h, negate_is_less_than_beta);
/* transpose and store */
{
__m256i row0, row1, row2, row3, row4, row5, row6, row7;
__m256i control = {0x0000000400000000, 0x0000000500000001,
0x0000000600000002, 0x0000000700000003};
DUP4_ARG3(__lasx_xvpermi_q, p0_org, q3_org, 0x02, p1_org, q2_org,
0x02, p2_org, q1_org, 0x02, p3_org, q0_org, 0x02,
p0_org, p1_org, p2_org, p3_org);
DUP2_ARG2(__lasx_xvilvl_b, p1_org, p3_org, p0_org, p2_org,
row0, row2);
DUP2_ARG2(__lasx_xvilvh_b, p1_org, p3_org, p0_org, p2_org,
row1, row3);
DUP2_ARG2(__lasx_xvilvl_b, row2, row0, row3, row1, row4, row6);
DUP2_ARG2(__lasx_xvilvh_b, row2, row0, row3, row1, row5, row7);
DUP4_ARG2(__lasx_xvperm_w, row4, control, row5, control, row6,
control, row7, control, row4, row5, row6, row7);
src = data - 4;
__lasx_xvstelm_d(row4, src, 0, 0);
__lasx_xvstelm_d(row4, src + img_width, 0, 1);
src += img_width_2x;
__lasx_xvstelm_d(row4, src, 0, 2);
__lasx_xvstelm_d(row4, src + img_width, 0, 3);
src += img_width_2x;
__lasx_xvstelm_d(row5, src, 0, 0);
__lasx_xvstelm_d(row5, src + img_width, 0, 1);
src += img_width_2x;
__lasx_xvstelm_d(row5, src, 0, 2);
__lasx_xvstelm_d(row5, src + img_width, 0, 3);
src += img_width_2x;
__lasx_xvstelm_d(row6, src, 0, 0);
__lasx_xvstelm_d(row6, src + img_width, 0, 1);
src += img_width_2x;
__lasx_xvstelm_d(row6, src, 0, 2);
__lasx_xvstelm_d(row6, src + img_width, 0, 3);
src += img_width_2x;
__lasx_xvstelm_d(row7, src, 0, 0);
__lasx_xvstelm_d(row7, src + img_width, 0, 1);
src += img_width_2x;
__lasx_xvstelm_d(row7, src, 0, 2);
__lasx_xvstelm_d(row7, src + img_width, 0, 3);
}
}
}
void ff_h264_v_lpf_luma_intra_8_lasx(uint8_t *data, ptrdiff_t img_width,
int alpha_in, int beta_in)
{
int img_width_2x = img_width << 1;
int img_width_3x = img_width_2x + img_width;
uint8_t *src = data - img_width_2x;
__m256i p0_asub_q0, p1_asub_p0, q1_asub_q0, alpha, beta;
__m256i is_less_than, is_less_than_beta, is_less_than_alpha;
__m256i p1_org, p0_org, q0_org, q1_org;
__m256i zero = __lasx_xvldi(0);
DUP4_ARG2(__lasx_xvldx, src, 0, src, img_width, src, img_width_2x,
src, img_width_3x, p1_org, p0_org, q0_org, q1_org);
alpha = __lasx_xvreplgr2vr_b(alpha_in);
beta = __lasx_xvreplgr2vr_b(beta_in);
p0_asub_q0 = __lasx_xvabsd_bu(p0_org, q0_org);
p1_asub_p0 = __lasx_xvabsd_bu(p1_org, p0_org);
q1_asub_q0 = __lasx_xvabsd_bu(q1_org, q0_org);
is_less_than_alpha = __lasx_xvslt_bu(p0_asub_q0, alpha);
is_less_than_beta = __lasx_xvslt_bu(p1_asub_p0, beta);
is_less_than = is_less_than_beta & is_less_than_alpha;
is_less_than_beta = __lasx_xvslt_bu(q1_asub_q0, beta);
is_less_than = is_less_than_beta & is_less_than;
is_less_than = __lasx_xvpermi_q(zero, is_less_than, 0x30);
if (__lasx_xbnz_v(is_less_than)) {
__m256i p2_asub_p0, q2_asub_q0, p0_h, q0_h, negate_is_less_than_beta;
__m256i p1_org_h, p0_org_h, q0_org_h, q1_org_h;
__m256i p2_org = __lasx_xvldx(src, -img_width);
__m256i q2_org = __lasx_xvldx(data, img_width_2x);
__m256i less_alpha_shift2_add2 = __lasx_xvsrli_b(alpha, 2);
less_alpha_shift2_add2 = __lasx_xvaddi_bu(less_alpha_shift2_add2, 2);
less_alpha_shift2_add2 = __lasx_xvslt_bu(p0_asub_q0,
less_alpha_shift2_add2);
p1_org_h = __lasx_vext2xv_hu_bu(p1_org);
p0_org_h = __lasx_vext2xv_hu_bu(p0_org);
q0_org_h = __lasx_vext2xv_hu_bu(q0_org);
q1_org_h = __lasx_vext2xv_hu_bu(q1_org);
p2_asub_p0 = __lasx_xvabsd_bu(p2_org, p0_org);
is_less_than_beta = __lasx_xvslt_bu(p2_asub_p0, beta);
is_less_than_beta = is_less_than_beta & less_alpha_shift2_add2;
negate_is_less_than_beta = __lasx_xvxori_b(is_less_than_beta, 0xff);
is_less_than_beta = is_less_than_beta & is_less_than;
negate_is_less_than_beta = negate_is_less_than_beta & is_less_than;
/* combine and store */
if (__lasx_xbnz_v(is_less_than_beta)) {
__m256i p2_org_h, p3_org_h, p1_h, p2_h;
__m256i p3_org = __lasx_xvldx(src, -img_width_2x);
p2_org_h = __lasx_vext2xv_hu_bu(p2_org);
p3_org_h = __lasx_vext2xv_hu_bu(p3_org);
AVC_LPF_P0P1P2_OR_Q0Q1Q2(p3_org_h, p0_org_h, q0_org_h, p1_org_h,
p2_org_h, q1_org_h, p0_h, p1_h, p2_h);
p0_h = __lasx_xvpickev_b(p0_h, p0_h);
p0_h = __lasx_xvpermi_d(p0_h, 0xd8);
DUP2_ARG2(__lasx_xvpickev_b, p1_h, p1_h, p2_h, p2_h, p1_h, p2_h);
DUP2_ARG2(__lasx_xvpermi_d, p1_h, 0xd8, p2_h, 0xd8, p1_h, p2_h);
p0_org = __lasx_xvbitsel_v(p0_org, p0_h, is_less_than_beta);
p1_org = __lasx_xvbitsel_v(p1_org, p1_h, is_less_than_beta);
p2_org = __lasx_xvbitsel_v(p2_org, p2_h, is_less_than_beta);
__lasx_xvst(p1_org, src, 0);
__lasx_xvst(p2_org, src - img_width, 0);
}
AVC_LPF_P0_OR_Q0(p0_org_h, q1_org_h, p1_org_h, p0_h);
/* combine */
p0_h = __lasx_xvpickev_b(p0_h, p0_h);
p0_h = __lasx_xvpermi_d(p0_h, 0xd8);
p0_org = __lasx_xvbitsel_v(p0_org, p0_h, negate_is_less_than_beta);
__lasx_xvst(p0_org, data - img_width, 0);
/* if (tmpFlag && (unsigned)ABS(q2-q0) < thresholds->beta_in) */
q2_asub_q0 = __lasx_xvabsd_bu(q2_org, q0_org);
is_less_than_beta = __lasx_xvslt_bu(q2_asub_q0, beta);
is_less_than_beta = is_less_than_beta & less_alpha_shift2_add2;
negate_is_less_than_beta = __lasx_xvxori_b(is_less_than_beta, 0xff);
is_less_than_beta = is_less_than_beta & is_less_than;
negate_is_less_than_beta = negate_is_less_than_beta & is_less_than;
/* combine and store */
if (__lasx_xbnz_v(is_less_than_beta)) {
__m256i q2_org_h, q3_org_h, q1_h, q2_h;
__m256i q3_org = __lasx_xvldx(data, img_width_2x + img_width);
q2_org_h = __lasx_vext2xv_hu_bu(q2_org);
q3_org_h = __lasx_vext2xv_hu_bu(q3_org);
AVC_LPF_P0P1P2_OR_Q0Q1Q2(q3_org_h, q0_org_h, p0_org_h, q1_org_h,
q2_org_h, p1_org_h, q0_h, q1_h, q2_h);
q0_h = __lasx_xvpickev_b(q0_h, q0_h);
q0_h = __lasx_xvpermi_d(q0_h, 0xd8);
DUP2_ARG2(__lasx_xvpickev_b, q1_h, q1_h, q2_h, q2_h, q1_h, q2_h);
DUP2_ARG2(__lasx_xvpermi_d, q1_h, 0xd8, q2_h, 0xd8, q1_h, q2_h);
q0_org = __lasx_xvbitsel_v(q0_org, q0_h, is_less_than_beta);
q1_org = __lasx_xvbitsel_v(q1_org, q1_h, is_less_than_beta);
q2_org = __lasx_xvbitsel_v(q2_org, q2_h, is_less_than_beta);
__lasx_xvst(q1_org, data + img_width, 0);
__lasx_xvst(q2_org, data + img_width_2x, 0);
}
AVC_LPF_P0_OR_Q0(q0_org_h, p1_org_h, q1_org_h, q0_h);
/* combine */
q0_h = __lasx_xvpickev_b(q0_h, q0_h);
q0_h = __lasx_xvpermi_d(q0_h, 0xd8);
q0_org = __lasx_xvbitsel_v(q0_org, q0_h, negate_is_less_than_beta);
__lasx_xvst(q0_org, data, 0);
}
}
void ff_h264_add_pixels4_8_lasx(uint8_t *_dst, int16_t *_src, int stride)
{
__m256i src0, dst0, dst1, dst2, dst3, zero;
__m256i tmp0, tmp1;
uint8_t* _dst1 = _dst + stride;
uint8_t* _dst2 = _dst1 + stride;
uint8_t* _dst3 = _dst2 + stride;
src0 = __lasx_xvld(_src, 0);
dst0 = __lasx_xvldrepl_w(_dst, 0);
dst1 = __lasx_xvldrepl_w(_dst1, 0);
dst2 = __lasx_xvldrepl_w(_dst2, 0);
dst3 = __lasx_xvldrepl_w(_dst3, 0);
tmp0 = __lasx_xvilvl_w(dst1, dst0);
tmp1 = __lasx_xvilvl_w(dst3, dst2);
dst0 = __lasx_xvilvl_d(tmp1, tmp0);
tmp0 = __lasx_vext2xv_hu_bu(dst0);
zero = __lasx_xvldi(0);
tmp1 = __lasx_xvadd_h(src0, tmp0);
dst0 = __lasx_xvpickev_b(tmp1, tmp1);
__lasx_xvstelm_w(dst0, _dst, 0, 0);
__lasx_xvstelm_w(dst0, _dst1, 0, 1);
__lasx_xvstelm_w(dst0, _dst2, 0, 4);
__lasx_xvstelm_w(dst0, _dst3, 0, 5);
__lasx_xvst(zero, _src, 0);
}
void ff_h264_add_pixels8_8_lasx(uint8_t *_dst, int16_t *_src, int stride)
{
__m256i src0, src1, src2, src3;
__m256i dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
__m256i tmp0, tmp1, tmp2, tmp3;
__m256i zero = __lasx_xvldi(0);
uint8_t *_dst1 = _dst + stride;
uint8_t *_dst2 = _dst1 + stride;
uint8_t *_dst3 = _dst2 + stride;
uint8_t *_dst4 = _dst3 + stride;
uint8_t *_dst5 = _dst4 + stride;
uint8_t *_dst6 = _dst5 + stride;
uint8_t *_dst7 = _dst6 + stride;
src0 = __lasx_xvld(_src, 0);
src1 = __lasx_xvld(_src, 32);
src2 = __lasx_xvld(_src, 64);
src3 = __lasx_xvld(_src, 96);
dst0 = __lasx_xvldrepl_d(_dst, 0);
dst1 = __lasx_xvldrepl_d(_dst1, 0);
dst2 = __lasx_xvldrepl_d(_dst2, 0);
dst3 = __lasx_xvldrepl_d(_dst3, 0);
dst4 = __lasx_xvldrepl_d(_dst4, 0);
dst5 = __lasx_xvldrepl_d(_dst5, 0);
dst6 = __lasx_xvldrepl_d(_dst6, 0);
dst7 = __lasx_xvldrepl_d(_dst7, 0);
tmp0 = __lasx_xvilvl_d(dst1, dst0);
tmp1 = __lasx_xvilvl_d(dst3, dst2);
tmp2 = __lasx_xvilvl_d(dst5, dst4);
tmp3 = __lasx_xvilvl_d(dst7, dst6);
dst0 = __lasx_vext2xv_hu_bu(tmp0);
dst1 = __lasx_vext2xv_hu_bu(tmp1);
dst1 = __lasx_vext2xv_hu_bu(tmp1);
dst2 = __lasx_vext2xv_hu_bu(tmp2);
dst3 = __lasx_vext2xv_hu_bu(tmp3);
tmp0 = __lasx_xvadd_h(src0, dst0);
tmp1 = __lasx_xvadd_h(src1, dst1);
tmp2 = __lasx_xvadd_h(src2, dst2);
tmp3 = __lasx_xvadd_h(src3, dst3);
dst1 = __lasx_xvpickev_b(tmp1, tmp0);
dst2 = __lasx_xvpickev_b(tmp3, tmp2);
__lasx_xvst(zero, _src, 0);
__lasx_xvst(zero, _src, 32);
__lasx_xvst(zero, _src, 64);
__lasx_xvst(zero, _src, 96);
__lasx_xvstelm_d(dst1, _dst, 0, 0);
__lasx_xvstelm_d(dst1, _dst1, 0, 2);
__lasx_xvstelm_d(dst1, _dst2, 0, 1);
__lasx_xvstelm_d(dst1, _dst3, 0, 3);
__lasx_xvstelm_d(dst2, _dst4, 0, 0);
__lasx_xvstelm_d(dst2, _dst5, 0, 2);
__lasx_xvstelm_d(dst2, _dst6, 0, 1);
__lasx_xvstelm_d(dst2, _dst7, 0, 3);
}
Reference in New Issue View Git Blame Copy Permalink