[FFmpeg-cvslog] avcodec: [loongarch] Optimize h264_deblock with LASX.
Jin Bo
git at videolan.org
Wed Dec 15 19:44:17 EET 2021
ffmpeg | branch: master | Jin Bo <jinbo at loongson.cn> | Wed Dec 15 11:51:09 2021 +0800| [1ccc45896096b39b9a0d5650618d384df30d15e4] | committer: Michael Niedermayer
avcodec: [loongarch] Optimize h264_deblock with LASX.
./ffmpeg -i ../1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -y /dev/null -an
before:293
after :295
Change-Id: I5ff6cba4eaca0c4218c0c97b880ca500e35f9c87
Signed-off-by: Hao Chen <chenhao at loongson.cn>
Reviewed-by: Shiyou Yin <yinshiyou-hf at loongson.cn>
Reviewed-by: guxiwei <guxiwei-hf at loongson.cn>
Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>
> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=1ccc45896096b39b9a0d5650618d384df30d15e4
---
libavcodec/loongarch/Makefile | 3 +-
libavcodec/loongarch/h264_deblock_lasx.c | 147 ++++++++++++++++++++++++++
libavcodec/loongarch/h264dsp_init_loongarch.c | 2 +
libavcodec/loongarch/h264dsp_lasx.h | 6 ++
4 files changed, 157 insertions(+), 1 deletion(-)
diff --git a/libavcodec/loongarch/Makefile b/libavcodec/loongarch/Makefile
index 242a2be290..1e1fe3fd48 100644
--- a/libavcodec/loongarch/Makefile
+++ b/libavcodec/loongarch/Makefile
@@ -4,4 +4,5 @@ OBJS-$(CONFIG_H264DSP) += loongarch/h264dsp_init_loongarch.o
LASX-OBJS-$(CONFIG_H264CHROMA) += loongarch/h264chroma_lasx.o
LASX-OBJS-$(CONFIG_H264QPEL) += loongarch/h264qpel_lasx.o
LASX-OBJS-$(CONFIG_H264DSP) += loongarch/h264dsp_lasx.o \
- loongarch/h264idct_lasx.o
+ loongarch/h264idct_lasx.o \
+ loongarch/h264_deblock_lasx.o
diff --git a/libavcodec/loongarch/h264_deblock_lasx.c b/libavcodec/loongarch/h264_deblock_lasx.c
new file mode 100644
index 0000000000..c89bea9a84
--- /dev/null
+++ b/libavcodec/loongarch/h264_deblock_lasx.c
@@ -0,0 +1,147 @@
+/*
+ * Copyright (c) 2021 Loongson Technology Corporation Limited
+ * Contributed by Xiwei Gu <guxiwei-hf at 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 "libavcodec/bit_depth_template.c"
+#include "h264dsp_lasx.h"
+#include "libavutil/loongarch/loongson_intrinsics.h"
+
+#define H264_LOOP_FILTER_STRENGTH_ITERATION_LASX(edges, step, mask_mv, dir, \
+ d_idx, mask_dir) \
+do { \
+ int b_idx = 0; \
+ int step_x4 = step << 2; \
+ int d_idx_12 = d_idx + 12; \
+ int d_idx_52 = d_idx + 52; \
+ int d_idx_x4 = d_idx << 2; \
+ int d_idx_x4_48 = d_idx_x4 + 48; \
+ int dir_x32 = dir * 32; \
+ uint8_t *ref_t = (uint8_t*)ref; \
+ uint8_t *mv_t = (uint8_t*)mv; \
+ uint8_t *nnz_t = (uint8_t*)nnz; \
+ uint8_t *bS_t = (uint8_t*)bS; \
+ mask_mv <<= 3; \
+ for (; b_idx < edges; b_idx += step) { \
+ out &= mask_dir; \
+ if (!(mask_mv & b_idx)) { \
+ if (bidir) { \
+ ref2 = __lasx_xvldx(ref_t, d_idx_12); \
+ ref3 = __lasx_xvldx(ref_t, d_idx_52); \
+ ref0 = __lasx_xvld(ref_t, 12); \
+ ref1 = __lasx_xvld(ref_t, 52); \
+ ref2 = __lasx_xvilvl_w(ref3, ref2); \
+ ref0 = __lasx_xvilvl_w(ref0, ref0); \
+ ref1 = __lasx_xvilvl_w(ref1, ref1); \
+ ref3 = __lasx_xvshuf4i_w(ref2, 0xB1); \
+ ref0 = __lasx_xvsub_b(ref0, ref2); \
+ ref1 = __lasx_xvsub_b(ref1, ref3); \
+ ref0 = __lasx_xvor_v(ref0, ref1); \
+\
+ tmp2 = __lasx_xvldx(mv_t, d_idx_x4_48); \
+ tmp3 = __lasx_xvld(mv_t, 48); \
+ tmp4 = __lasx_xvld(mv_t, 208); \
+ tmp5 = __lasx_xvld(mv_t + d_idx_x4, 208); \
+ DUP2_ARG3(__lasx_xvpermi_q, tmp2, tmp2, 0x20, tmp5, tmp5, \
+ 0x20, tmp2, tmp5); \
+ tmp3 = __lasx_xvpermi_q(tmp4, tmp3, 0x20); \
+ tmp2 = __lasx_xvsub_h(tmp2, tmp3); \
+ tmp5 = __lasx_xvsub_h(tmp5, tmp3); \
+ DUP2_ARG2(__lasx_xvsat_h, tmp2, 7, tmp5, 7, tmp2, tmp5); \
+ tmp0 = __lasx_xvpickev_b(tmp5, tmp2); \
+ tmp0 = __lasx_xvpermi_d(tmp0, 0xd8); \
+ tmp0 = __lasx_xvadd_b(tmp0, cnst_1); \
+ tmp0 = __lasx_xvssub_bu(tmp0, cnst_0); \
+ tmp0 = __lasx_xvsat_h(tmp0, 7); \
+ tmp0 = __lasx_xvpickev_b(tmp0, tmp0); \
+ tmp0 = __lasx_xvpermi_d(tmp0, 0xd8); \
+ tmp1 = __lasx_xvpickod_d(tmp0, tmp0); \
+ out = __lasx_xvor_v(ref0, tmp0); \
+ tmp1 = __lasx_xvshuf4i_w(tmp1, 0xB1); \
+ out = __lasx_xvor_v(out, tmp1); \
+ tmp0 = __lasx_xvshuf4i_w(out, 0xB1); \
+ out = __lasx_xvmin_bu(out, tmp0); \
+ } else { \
+ ref0 = __lasx_xvldx(ref_t, d_idx_12); \
+ ref3 = __lasx_xvld(ref_t, 12); \
+ tmp2 = __lasx_xvldx(mv_t, d_idx_x4_48); \
+ tmp3 = __lasx_xvld(mv_t, 48); \
+ tmp4 = __lasx_xvsub_h(tmp3, tmp2); \
+ tmp1 = __lasx_xvsat_h(tmp4, 7); \
+ tmp1 = __lasx_xvpickev_b(tmp1, tmp1); \
+ tmp1 = __lasx_xvadd_b(tmp1, cnst_1); \
+ out = __lasx_xvssub_bu(tmp1, cnst_0); \
+ out = __lasx_xvsat_h(out, 7); \
+ out = __lasx_xvpickev_b(out, out); \
+ ref0 = __lasx_xvsub_b(ref3, ref0); \
+ out = __lasx_xvor_v(out, ref0); \
+ } \
+ } \
+ tmp0 = __lasx_xvld(nnz_t, 12); \
+ tmp1 = __lasx_xvldx(nnz_t, d_idx_12); \
+ tmp0 = __lasx_xvor_v(tmp0, tmp1); \
+ tmp0 = __lasx_xvmin_bu(tmp0, cnst_2); \
+ out = __lasx_xvmin_bu(out, cnst_2); \
+ tmp0 = __lasx_xvslli_h(tmp0, 1); \
+ tmp0 = __lasx_xvmax_bu(out, tmp0); \
+ tmp0 = __lasx_vext2xv_hu_bu(tmp0); \
+ __lasx_xvstelm_d(tmp0, bS_t + dir_x32, 0, 0); \
+ ref_t += step; \
+ mv_t += step_x4; \
+ nnz_t += step; \
+ bS_t += step; \
+ } \
+} while(0)
+
+void ff_h264_loop_filter_strength_lasx(int16_t bS[2][4][4], uint8_t nnz[40],
+ int8_t ref[2][40], int16_t mv[2][40][2],
+ int bidir, int edges, int step,
+ int mask_mv0, int mask_mv1, int field)
+{
+ __m256i out;
+ __m256i ref0, ref1, ref2, ref3;
+ __m256i tmp0, tmp1;
+ __m256i tmp2, tmp3, tmp4, tmp5;
+ __m256i cnst_0, cnst_1, cnst_2;
+ __m256i zero = __lasx_xvldi(0);
+ __m256i one = __lasx_xvnor_v(zero, zero);
+ int64_t cnst3 = 0x0206020602060206, cnst4 = 0x0103010301030103;
+ if (field) {
+ cnst_0 = __lasx_xvreplgr2vr_d(cnst3);
+ cnst_1 = __lasx_xvreplgr2vr_d(cnst4);
+ cnst_2 = __lasx_xvldi(0x01);
+ } else {
+ DUP2_ARG1(__lasx_xvldi, 0x06, 0x03, cnst_0, cnst_1);
+ cnst_2 = __lasx_xvldi(0x01);
+ }
+ step <<= 3;
+ edges <<= 3;
+
+ H264_LOOP_FILTER_STRENGTH_ITERATION_LASX(edges, step, mask_mv1,
+ 1, -8, zero);
+ H264_LOOP_FILTER_STRENGTH_ITERATION_LASX(32, 8, mask_mv0, 0, -1, one);
+
+ DUP2_ARG2(__lasx_xvld, (int8_t*)bS, 0, (int8_t*)bS, 16, tmp0, tmp1);
+ DUP2_ARG2(__lasx_xvilvh_d, tmp0, tmp0, tmp1, tmp1, tmp2, tmp3);
+ LASX_TRANSPOSE4x4_H(tmp0, tmp2, tmp1, tmp3, tmp2, tmp3, tmp4, tmp5);
+ __lasx_xvstelm_d(tmp2, (int8_t*)bS, 0, 0);
+ __lasx_xvstelm_d(tmp3, (int8_t*)bS + 8, 0, 0);
+ __lasx_xvstelm_d(tmp4, (int8_t*)bS + 16, 0, 0);
+ __lasx_xvstelm_d(tmp5, (int8_t*)bS + 24, 0, 0);
+}
diff --git a/libavcodec/loongarch/h264dsp_init_loongarch.c b/libavcodec/loongarch/h264dsp_init_loongarch.c
index 0985c2fe8a..37633c3e51 100644
--- a/libavcodec/loongarch/h264dsp_init_loongarch.c
+++ b/libavcodec/loongarch/h264dsp_init_loongarch.c
@@ -29,6 +29,8 @@ av_cold void ff_h264dsp_init_loongarch(H264DSPContext *c, const int bit_depth,
int cpu_flags = av_get_cpu_flags();
if (have_lasx(cpu_flags)) {
+ if (chroma_format_idc <= 1)
+ c->h264_loop_filter_strength = ff_h264_loop_filter_strength_lasx;
if (bit_depth == 8) {
c->h264_add_pixels4_clear = ff_h264_add_pixels4_8_lasx;
c->h264_add_pixels8_clear = ff_h264_add_pixels8_8_lasx;
diff --git a/libavcodec/loongarch/h264dsp_lasx.h b/libavcodec/loongarch/h264dsp_lasx.h
index bfd567fffa..4cf813750b 100644
--- a/libavcodec/loongarch/h264dsp_lasx.h
+++ b/libavcodec/loongarch/h264dsp_lasx.h
@@ -88,4 +88,10 @@ void ff_h264_idct_add16_intra_lasx(uint8_t *dst, const int32_t *blk_offset,
const uint8_t nzc[15 * 8]);
void ff_h264_deq_idct_luma_dc_lasx(int16_t *dst, int16_t *src,
int32_t de_qval);
+
+void ff_h264_loop_filter_strength_lasx(int16_t bS[2][4][4], uint8_t nnz[40],
+ int8_t ref[2][40], int16_t mv[2][40][2],
+ int bidir, int edges, int step,
+ int mask_mv0, int mask_mv1, int field);
+
#endif // #ifndef AVCODEC_LOONGARCH_H264DSP_LASX_H
More information about the ffmpeg-cvslog
mailing list