From: Shiyou Yin <yinshiyou-hf@loongson.cn>
To: FFmpeg development discussions and patches <ffmpeg-devel@ffmpeg.org>
Cc: andreas.rheinhardt@outlook.com
Subject: Re: [FFmpeg-devel] [PATCH v1 3/3] swscale/la: Add output_lasx.c file.
Date: Tue, 6 Sep 2022 16:12:21 +0800
Message-ID: <ABB32204-033B-410E-8DA7-A41064845E48@loongson.cn> (raw)
In-Reply-To: <AS8P250MB0744718695EE00DD178861ED8F769@AS8P250MB0744.EURP250.PROD.OUTLOOK.COM>
> 2022年8月29日 20:30，Andreas Rheinhardt <andreas.rheinhardt@outlook.com> 写道：
>
> Hao Chen:
>> ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt
>> rgb24 -y /dev/null -an
>> before: 150fps
>> after: 183fps
>>
>> Signed-off-by: Hao Chen <chenhao@loongson.cn>
>> ---
>> libswscale/loongarch/Makefile | 3 +-
>> libswscale/loongarch/output_lasx.c | 1982 +++++++++++++++++
>> libswscale/loongarch/swscale_init_loongarch.c | 3 +
>> libswscale/loongarch/swscale_loongarch.h | 6 +
>> 4 files changed, 1993 insertions(+), 1 deletion(-)
>> create mode 100644 libswscale/loongarch/output_lasx.c
>>
>> diff --git a/libswscale/loongarch/Makefile b/libswscale/loongarch/Makefile
>> index 4345971514..54d48b3de0 100644
>> --- a/libswscale/loongarch/Makefile
>> +++ b/libswscale/loongarch/Makefile
>> @@ -2,4 +2,5 @@ OBJS-$(CONFIG_SWSCALE) += loongarch/swscale_init_loongarch.o
>> LASX-OBJS-$(CONFIG_SWSCALE) += loongarch/swscale_lasx.o \
>> loongarch/input_lasx.o \
>> loongarch/yuv2rgb_lasx.o \
>> - loongarch/rgb2rgb_lasx.o
>> + loongarch/rgb2rgb_lasx.o \
>> + loongarch/output_lasx.o
>> diff --git a/libswscale/loongarch/output_lasx.c b/libswscale/loongarch/output_lasx.c
>> new file mode 100644
>> index 0000000000..19f82692ff
>> --- /dev/null
>> +++ b/libswscale/loongarch/output_lasx.c
>> @@ -0,0 +1,1982 @@
>> +/*
>> + * Copyright (C) 2022 Loongson Technology Corporation Limited
>> + * Contributed by Hao Chen(chenhao@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 "swscale_loongarch.h"
>> +#include "libavutil/loongarch/loongson_intrinsics.h"
>> +
>> +void ff_yuv2planeX_8_lasx(const int16_t *filter, int filterSize,
>> + const int16_t **src, uint8_t *dest, int dstW,
>> + const uint8_t *dither, int offset)
>> +{
>> + int i;
>> + int len = dstW - 15;
>> + __m256i mask = {0x1C0C180814041000, 0x1C1814100C080400,
>> + 0x1C0C180814041000, 0x1C1814100C080400};
>> + __m256i val1, val2, val3;
>> + uint8_t dither0 = dither[offset & 7];
>> + uint8_t dither1 = dither[(offset + 1) & 7];
>> + uint8_t dither2 = dither[(offset + 2) & 7];
>> + uint8_t dither3 = dither[(offset + 3) & 7];
>> + uint8_t dither4 = dither[(offset + 4) & 7];
>> + uint8_t dither5 = dither[(offset + 5) & 7];
>> + uint8_t dither6 = dither[(offset + 6) & 7];
>> + uint8_t dither7 = dither[(offset + 7) & 7];
>> + int val_1[8] = {dither0, dither2, dither4, dither6,
>> + dither0, dither2, dither4, dither6};
>> + int val_2[8] = {dither1, dither3, dither5, dither7,
>> + dither1, dither3, dither5, dither7};
>> + int val_3[8] = {dither0, dither1, dither2, dither3,
>> + dither4, dither5, dither6, dither7};
>> +
>> + DUP2_ARG2(__lasx_xvld, val_1, 0, val_2, 0, val1, val2);
>> + val3 = __lasx_xvld(val_3, 0);
>> +
>> + for (i = 0; i < len; i += 16) {
>> + int j;
>> + __m256i src0, filter0, val;
>> + __m256i val_ev, val_od;
>> +
>> + val_ev = __lasx_xvslli_w(val1, 12);
>> + val_od = __lasx_xvslli_w(val2, 12);
>> +
>> + for (j = 0; j < filterSize; j++) {
>> + src0 = __lasx_xvld(src[j]+ i, 0);
>> + filter0 = __lasx_xvldrepl_h((filter + j), 0);
>> + val_ev = __lasx_xvmaddwev_w_h(val_ev, src0, filter0);
>> + val_od = __lasx_xvmaddwod_w_h(val_od, src0, filter0);
>> + }
>> + val_ev = __lasx_xvsrai_w(val_ev, 19);
>> + val_od = __lasx_xvsrai_w(val_od, 19);
>> + val_ev = __lasx_xvclip255_w(val_ev);
>> + val_od = __lasx_xvclip255_w(val_od);
>> + val = __lasx_xvshuf_b(val_od, val_ev, mask);
>> + __lasx_xvstelm_d(val, (dest + i), 0, 0);
>> + __lasx_xvstelm_d(val, (dest + i), 8, 2);
>> + }
>> + if (dstW - i >= 8){
>> + int j;
>> + __m256i src0, filter0, val_h;
>> + __m256i val_l;
>> +
>> + val_l = __lasx_xvslli_w(val3, 12);
>> +
>> + for (j = 0; j < filterSize; j++) {
>> + src0 = __lasx_xvld(src[j] + i, 0);
>> + src0 = __lasx_vext2xv_w_h(src0);
>> + filter0 = __lasx_xvldrepl_h((filter + j), 0);
>> + filter0 = __lasx_vext2xv_w_h(filter0);
>> + val_l = __lasx_xvmadd_w(val_l, src0, filter0);
>> + }
>> + val_l = __lasx_xvsrai_w(val_l, 19);
>> + val_l = __lasx_xvclip255_w(val_l);
>> + val_h = __lasx_xvpermi_d(val_l, 0x4E);
>> + val_l = __lasx_xvshuf_b(val_h, val_l, mask);
>> + __lasx_xvstelm_d(val_l, (dest + i), 0, 1);
>> + i += 8;
>> + }
>> + for (; i < dstW; i++) {
>> + int val = dither[(i + offset) & 7] << 12;
>> + int j;
>> + for (j = 0; j< filterSize; j++)
>> + val += src[j][i] * filter[j];
>> +
>> + dest[i] = av_clip_uint8(val >> 19);
>> + }
>> +}
>> +
>> +/*Copy from libswscale/output.c*/
>> +static av_always_inline void
>> +yuv2rgb_write(uint8_t *_dest, int i, int Y1, int Y2,
>> + unsigned A1, unsigned A2,
>> + const void *_r, const void *_g, const void *_b, int y,
>> + enum AVPixelFormat target, int hasAlpha)
>> +{
>> + if (target == AV_PIX_FMT_ARGB || target == AV_PIX_FMT_RGBA ||
>> + target == AV_PIX_FMT_ABGR || target == AV_PIX_FMT_BGRA) {
>> + uint32_t *dest = (uint32_t *) _dest;
>> + const uint32_t *r = (const uint32_t *) _r;
>> + const uint32_t *g = (const uint32_t *) _g;
>> + const uint32_t *b = (const uint32_t *) _b;
>> +
>> +#if CONFIG_SMALL
>> + dest[i * 2 + 0] = r[Y1] + g[Y1] + b[Y1];
>> + dest[i * 2 + 1] = r[Y2] + g[Y2] + b[Y2];
>> +#else
>> +#if defined(ASSERT_LEVEL) && ASSERT_LEVEL > 1
>> + int sh = (target == AV_PIX_FMT_RGB32_1 ||
>> + target == AV_PIX_FMT_BGR32_1) ? 0 : 24;
>> + av_assert2((((r[Y1] + g[Y1] + b[Y1]) >> sh) & 0xFF) == 0xFF);
>> +#endif
>> + dest[i * 2 + 0] = r[Y1] + g[Y1] + b[Y1];
>> + dest[i * 2 + 1] = r[Y2] + g[Y2] + b[Y2];
>> +#endif
>> + } else if (target == AV_PIX_FMT_RGB24 || target == AV_PIX_FMT_BGR24) {
>> + uint8_t *dest = (uint8_t *) _dest;
>> + const uint8_t *r = (const uint8_t *) _r;
>> + const uint8_t *g = (const uint8_t *) _g;
>> + const uint8_t *b = (const uint8_t *) _b;
>> +
>> +#define r_b ((target == AV_PIX_FMT_RGB24) ? r : b)
>> +#define b_r ((target == AV_PIX_FMT_RGB24) ? b : r)
>> +
>> + dest[i * 6 + 0] = r_b[Y1];
>> + dest[i * 6 + 1] = g[Y1];
>> + dest[i * 6 + 2] = b_r[Y1];
>> + dest[i * 6 + 3] = r_b[Y2];
>> + dest[i * 6 + 4] = g[Y2];
>> + dest[i * 6 + 5] = b_r[Y2];
>> +#undef r_b
>> +#undef b_r
>> + } else if (target == AV_PIX_FMT_RGB565 || target == AV_PIX_FMT_BGR565 ||
>> + target == AV_PIX_FMT_RGB555 || target == AV_PIX_FMT_BGR555 ||
>> + target == AV_PIX_FMT_RGB444 || target == AV_PIX_FMT_BGR444) {
>> + uint16_t *dest = (uint16_t *) _dest;
>> + const uint16_t *r = (const uint16_t *) _r;
>> + const uint16_t *g = (const uint16_t *) _g;
>> + const uint16_t *b = (const uint16_t *) _b;
>> + int dr1, dg1, db1, dr2, dg2, db2;
>> +
>> + if (target == AV_PIX_FMT_RGB565 || target == AV_PIX_FMT_BGR565) {
>> + dr1 = ff_dither_2x2_8[ y & 1 ][0];
>> + dg1 = ff_dither_2x2_4[ y & 1 ][0];
>> + db1 = ff_dither_2x2_8[(y & 1) ^ 1][0];
>> + dr2 = ff_dither_2x2_8[ y & 1 ][1];
>> + dg2 = ff_dither_2x2_4[ y & 1 ][1];
>> + db2 = ff_dither_2x2_8[(y & 1) ^ 1][1];
>> + } else if (target == AV_PIX_FMT_RGB555 || target == AV_PIX_FMT_BGR555) {
>> + dr1 = ff_dither_2x2_8[ y & 1 ][0];
>> + dg1 = ff_dither_2x2_8[ y & 1 ][1];
>> + db1 = ff_dither_2x2_8[(y & 1) ^ 1][0];
>> + dr2 = ff_dither_2x2_8[ y & 1 ][1];
>> + dg2 = ff_dither_2x2_8[ y & 1 ][0];
>> + db2 = ff_dither_2x2_8[(y & 1) ^ 1][1];
>> + } else {
>> + dr1 = ff_dither_4x4_16[ y & 3 ][0];
>> + dg1 = ff_dither_4x4_16[ y & 3 ][1];
>> + db1 = ff_dither_4x4_16[(y & 3) ^ 3][0];
>> + dr2 = ff_dither_4x4_16[ y & 3 ][1];
>> + dg2 = ff_dither_4x4_16[ y & 3 ][0];
>> + db2 = ff_dither_4x4_16[(y & 3) ^ 3][1];
>> + }
>> +
>> + dest[i * 2 + 0] = r[Y1 + dr1] + g[Y1 + dg1] + b[Y1 + db1];
>> + dest[i * 2 + 1] = r[Y2 + dr2] + g[Y2 + dg2] + b[Y2 + db2];
>> + } else /* 8/4 bits */ {
>> + uint8_t *dest = (uint8_t *) _dest;
>> + const uint8_t *r = (const uint8_t *) _r;
>> + const uint8_t *g = (const uint8_t *) _g;
>> + const uint8_t *b = (const uint8_t *) _b;
>> + int dr1, dg1, db1, dr2, dg2, db2;
>> +
>> + if (target == AV_PIX_FMT_RGB8 || target == AV_PIX_FMT_BGR8) {
>> + const uint8_t * const d64 = ff_dither_8x8_73[y & 7];
>> + const uint8_t * const d32 = ff_dither_8x8_32[y & 7];
>> + dr1 = dg1 = d32[(i * 2 + 0) & 7];
>> + db1 = d64[(i * 2 + 0) & 7];
>> + dr2 = dg2 = d32[(i * 2 + 1) & 7];
>> + db2 = d64[(i * 2 + 1) & 7];
>> + } else {
>> + const uint8_t * const d64 = ff_dither_8x8_73 [y & 7];
>> + const uint8_t * const d128 = ff_dither_8x8_220[y & 7];
>> + dr1 = db1 = d128[(i * 2 + 0) & 7];
>> + dg1 = d64[(i * 2 + 0) & 7];
>> + dr2 = db2 = d128[(i * 2 + 1) & 7];
>> + dg2 = d64[(i * 2 + 1) & 7];
>> + }
>> +
>> + if (target == AV_PIX_FMT_RGB4 || target == AV_PIX_FMT_BGR4) {
>> + dest[i] = r[Y1 + dr1] + g[Y1 + dg1] + b[Y1 + db1] +
>> + ((r[Y2 + dr2] + g[Y2 + dg2] + b[Y2 + db2]) << 4);
>> + } else {
>> + dest[i * 2 + 0] = r[Y1 + dr1] + g[Y1 + dg1] + b[Y1 + db1];
>> + dest[i * 2 + 1] = r[Y2 + dr2] + g[Y2 + dg2] + b[Y2 + db2];
>> + }
>> + }
>> +}
>> +
>> +#define WRITE_YUV2RGB(vec_y1, vec_y2, vec_u, vec_v, t1, t2, t3, t4) \
>> +{ \
>> + Y1 = __lasx_xvpickve2gr_w(vec_y1, t1); \
>> + Y2 = __lasx_xvpickve2gr_w(vec_y2, t2); \
>> + U = __lasx_xvpickve2gr_w(vec_u, t3); \
>> + V = __lasx_xvpickve2gr_w(vec_v, t4); \
>> + r = c->table_rV[V]; \
>> + g = (c->table_gU[U] + c->table_gV[V]); \
>> + b = c->table_bU[U]; \
>> + yuv2rgb_write(dest, count, Y1, Y2, 0, 0, \
>> + r, g, b, y, target, 0); \
>> + count++; \
>> +}
>> +
>> +static void
>> +yuv2rgb_X_template_lasx(SwsContext *c, const int16_t *lumFilter,
>> + const int16_t **lumSrc, int lumFilterSize,
>> + const int16_t *chrFilter, const int16_t **chrUSrc,
>> + const int16_t **chrVSrc, int chrFilterSize,
>> + const int16_t **alpSrc, uint8_t *dest, int dstW,
>> + int y, enum AVPixelFormat target, int hasAlpha)
>> +{
>> + int i, j;
>> + int count = 0;
>> + int t = 1 << 18;
>> + int len = dstW >> 6;
>> + int res = dstW & 63;
>> + int len_count = (dstW + 1) >> 1;
>> + const void *r, *g, *b;
>> + int head = YUVRGB_TABLE_HEADROOM;
>> + __m256i headroom = __lasx_xvreplgr2vr_w(head);
>> +
>> + for (i = 0; i < len; i++) {
>> + int Y1, Y2, U, V, count_lum = count << 1;
>> + __m256i l_src1, l_src2, l_src3, l_src4, u_src1, u_src2, v_src1, v_src2;
>> + __m256i yl1_ev, yl1_od, yh1_ev, yh1_od, yl2_ev, yl2_od, yh2_ev, yh2_od;
>> + __m256i u1_ev, u1_od, v1_ev, v1_od, u2_ev, u2_od, v2_ev, v2_od, temp;
>> +
>> + yl1_ev = __lasx_xvldrepl_w(&t, 0);
>> + yl1_od = yl1_ev;
>> + yh1_ev = yl1_ev;
>> + yh1_od = yl1_ev;
>> + u1_ev = yl1_ev;
>> + v1_ev = yl1_ev;
>> + u1_od = yl1_ev;
>> + v1_od = yl1_ev;
>> + yl2_ev = yl1_ev;
>> + yl2_od = yl1_ev;
>> + yh2_ev = yl1_ev;
>> + yh2_od = yl1_ev;
>> + u2_ev = yl1_ev;
>> + v2_ev = yl1_ev;
>> + u2_od = yl1_ev;
>> + v2_od = yl1_ev;
>> + for (j = 0; j < lumFilterSize; j++) {
>> + int16_t *src_lum = lumSrc[j] + count_lum;
>> + temp = __lasx_xvldrepl_h((lumFilter + j), 0);
>> + DUP4_ARG2(__lasx_xvld, src_lum, 0, src_lum, 32, src_lum, 64,
>> + src_lum, 96, l_src1, l_src2, l_src3, l_src4);
>> +
>> + yl1_ev = __lasx_xvmaddwev_w_h(yl1_ev, temp, l_src1);
>> + yl1_od = __lasx_xvmaddwod_w_h(yl1_od, temp, l_src1);
>> + yh1_ev = __lasx_xvmaddwev_w_h(yh1_ev, temp, l_src2);
>> + yh1_od = __lasx_xvmaddwod_w_h(yh1_od, temp, l_src2);
>> + yl2_ev = __lasx_xvmaddwev_w_h(yl2_ev, temp, l_src3);
>> + yl2_od = __lasx_xvmaddwod_w_h(yl2_od, temp, l_src3);
>> + yh2_ev = __lasx_xvmaddwev_w_h(yh2_ev, temp, l_src4);
>> + yh2_od = __lasx_xvmaddwod_w_h(yh2_od, temp, l_src4);
>> + }
>> + for (j = 0; j < chrFilterSize; j++) {
>> + DUP2_ARG2(__lasx_xvld, chrUSrc[j] + count, 0, chrUSrc[j] + count, 32,
>> + u_src1, u_src2);
>> + DUP2_ARG2(__lasx_xvld, chrVSrc[j] + count, 0, chrVSrc[j] + count, 32,
>> + v_src1, v_src2);
>> + temp = __lasx_xvldrepl_h((chrFilter + j), 0);
>> + u1_ev = __lasx_xvmaddwev_w_h(u1_ev, temp, u_src1);
>> + u1_od = __lasx_xvmaddwod_w_h(u1_od, temp, u_src1);
>> + v1_ev = __lasx_xvmaddwev_w_h(v1_ev, temp, v_src1);
>> + v1_od = __lasx_xvmaddwod_w_h(v1_od, temp, v_src1);
>> + u2_ev = __lasx_xvmaddwev_w_h(u2_ev, temp, u_src2);
>> + u2_od = __lasx_xvmaddwod_w_h(u2_od, temp, u_src2);
>> + v2_ev = __lasx_xvmaddwev_w_h(v2_ev, temp, v_src2);
>> + v2_od = __lasx_xvmaddwod_w_h(v2_od, temp, v_src2);
>> + }
>> + yl1_ev = __lasx_xvsrai_w(yl1_ev, 19);
>> + yh1_ev = __lasx_xvsrai_w(yh1_ev, 19);
>> + yl1_od = __lasx_xvsrai_w(yl1_od, 19);
>> + yh1_od = __lasx_xvsrai_w(yh1_od, 19);
>> + u1_ev = __lasx_xvsrai_w(u1_ev, 19);
>> + v1_ev = __lasx_xvsrai_w(v1_ev, 19);
>> + u1_od = __lasx_xvsrai_w(u1_od, 19);
>> + v1_od = __lasx_xvsrai_w(v1_od, 19);
>> + yl2_ev = __lasx_xvsrai_w(yl2_ev, 19);
>> + yh2_ev = __lasx_xvsrai_w(yh2_ev, 19);
>> + yl2_od = __lasx_xvsrai_w(yl2_od, 19);
>> + yh2_od = __lasx_xvsrai_w(yh2_od, 19);
>> + u2_ev = __lasx_xvsrai_w(u2_ev, 19);
>> + v2_ev = __lasx_xvsrai_w(v2_ev, 19);
>> + u2_od = __lasx_xvsrai_w(u2_od, 19);
>> + v2_od = __lasx_xvsrai_w(v2_od, 19);
>> + u1_ev = __lasx_xvadd_w(u1_ev, headroom);
>> + v1_ev = __lasx_xvadd_w(v1_ev, headroom);
>> + u1_od = __lasx_xvadd_w(u1_od, headroom);
>> + v1_od = __lasx_xvadd_w(v1_od, headroom);
>> + u2_ev = __lasx_xvadd_w(u2_ev, headroom);
>> + v2_ev = __lasx_xvadd_w(v2_ev, headroom);
>> + u2_od = __lasx_xvadd_w(u2_od, headroom);
>> + v2_od = __lasx_xvadd_w(v2_od, headroom);
>> + WRITE_YUV2RGB(yl1_ev, yl1_od, u1_ev, v1_ev, 0, 0, 0, 0);
>> + WRITE_YUV2RGB(yl1_ev, yl1_od, u1_od, v1_od, 1, 1, 0, 0);
>> + WRITE_YUV2RGB(yl1_ev, yl1_od, u1_ev, v1_ev, 2, 2, 1, 1);
>> + WRITE_YUV2RGB(yl1_ev, yl1_od, u1_od, v1_od, 3, 3, 1, 1);
>> + WRITE_YUV2RGB(yl1_ev, yl1_od, u1_ev, v1_ev, 4, 4, 2, 2);
>> + WRITE_YUV2RGB(yl1_ev, yl1_od, u1_od, v1_od, 5, 5, 2, 2);
>> + WRITE_YUV2RGB(yl1_ev, yl1_od, u1_ev, v1_ev, 6, 6, 3, 3);
>> + WRITE_YUV2RGB(yl1_ev, yl1_od, u1_od, v1_od, 7, 7, 3, 3);
>> + WRITE_YUV2RGB(yh1_ev, yh1_od, u1_ev, v1_ev, 0, 0, 4, 4);
>> + WRITE_YUV2RGB(yh1_ev, yh1_od, u1_od, v1_od, 1, 1, 4, 4);
>> + WRITE_YUV2RGB(yh1_ev, yh1_od, u1_ev, v1_ev, 2, 2, 5, 5);
>> + WRITE_YUV2RGB(yh1_ev, yh1_od, u1_od, v1_od, 3, 3, 5, 5);
>> + WRITE_YUV2RGB(yh1_ev, yh1_od, u1_ev, v1_ev, 4, 4, 6, 6);
>> + WRITE_YUV2RGB(yh1_ev, yh1_od, u1_od, v1_od, 5, 5, 6, 6);
>> + WRITE_YUV2RGB(yh1_ev, yh1_od, u1_ev, v1_ev, 6, 6, 7, 7);
>> + WRITE_YUV2RGB(yh1_ev, yh1_od, u1_od, v1_od, 7, 7, 7, 7);
>> + WRITE_YUV2RGB(yl2_ev, yl2_od, u2_ev, v2_ev, 0, 0, 0, 0);
>> + WRITE_YUV2RGB(yl2_ev, yl2_od, u2_od, v2_od, 1, 1, 0, 0);
>> + WRITE_YUV2RGB(yl2_ev, yl2_od, u2_ev, v2_ev, 2, 2, 1, 1);
>> + WRITE_YUV2RGB(yl2_ev, yl2_od, u2_od, v2_od, 3, 3, 1, 1);
>> + WRITE_YUV2RGB(yl2_ev, yl2_od, u2_ev, v2_ev, 4, 4, 2, 2);
>> + WRITE_YUV2RGB(yl2_ev, yl2_od, u2_od, v2_od, 5, 5, 2, 2);
>> + WRITE_YUV2RGB(yl2_ev, yl2_od, u2_ev, v2_ev, 6, 6, 3, 3);
>> + WRITE_YUV2RGB(yl2_ev, yl2_od, u2_od, v2_od, 7, 7, 3, 3);
>> + WRITE_YUV2RGB(yh2_ev, yh2_od, u2_ev, v2_ev, 0, 0, 4, 4);
>> + WRITE_YUV2RGB(yh2_ev, yh2_od, u2_od, v2_od, 1, 1, 4, 4);
>> + WRITE_YUV2RGB(yh2_ev, yh2_od, u2_ev, v2_ev, 2, 2, 5, 5);
>> + WRITE_YUV2RGB(yh2_ev, yh2_od, u2_od, v2_od, 3, 3, 5, 5);
>> + WRITE_YUV2RGB(yh2_ev, yh2_od, u2_ev, v2_ev, 4, 4, 6, 6);
>> + WRITE_YUV2RGB(yh2_ev, yh2_od, u2_od, v2_od, 5, 5, 6, 6);
>> + WRITE_YUV2RGB(yh2_ev, yh2_od, u2_ev, v2_ev, 6, 6, 7, 7);
>> + WRITE_YUV2RGB(yh2_ev, yh2_od, u2_od, v2_od, 7, 7, 7, 7);
>> + }
>> + if (res >= 32) {
>> + int Y1, Y2, U, V, count_lum = count << 1;
>> + __m256i l_src1, l_src2, u_src, v_src;
>> + __m256i yl_ev, yl_od, yh_ev, yh_od;
>> + __m256i u_ev, u_od, v_ev, v_od, temp;
>> +
>> + yl_ev = __lasx_xvldrepl_w(&t, 0);
>> + yl_od = yl_ev;
>> + yh_ev = yl_ev;
>> + yh_od = yl_ev;
>> + u_ev = yl_ev;
>> + v_ev = yl_ev;
>> + u_od = yl_ev;
>> + v_od = yl_ev;
>> + for (j = 0; j < lumFilterSize; j++) {
>> + temp = __lasx_xvldrepl_h((lumFilter + j), 0);
>> + DUP2_ARG2(__lasx_xvld, lumSrc[j] + count_lum, 0, lumSrc[j] + count_lum,
>> + 32, l_src1, l_src2);
>> + yl_ev = __lasx_xvmaddwev_w_h(yl_ev, temp, l_src1);
>> + yl_od = __lasx_xvmaddwod_w_h(yl_od, temp, l_src1);
>> + yh_ev = __lasx_xvmaddwev_w_h(yh_ev, temp, l_src2);
>> + yh_od = __lasx_xvmaddwod_w_h(yh_od, temp, l_src2);
>> + }
>> + for (j = 0; j < chrFilterSize; j++) {
>> + DUP2_ARG2(__lasx_xvld, chrUSrc[j] + count, 0, chrVSrc[j] + count, 0,
>> + u_src, v_src);
>> + temp = __lasx_xvldrepl_h((chrFilter + j), 0);
>> + u_ev = __lasx_xvmaddwev_w_h(u_ev, temp, u_src);
>> + u_od = __lasx_xvmaddwod_w_h(u_od, temp, u_src);
>> + v_ev = __lasx_xvmaddwev_w_h(v_ev, temp, v_src);
>> + v_od = __lasx_xvmaddwod_w_h(v_od, temp, v_src);
>> + }
>> + yl_ev = __lasx_xvsrai_w(yl_ev, 19);
>> + yh_ev = __lasx_xvsrai_w(yh_ev, 19);
>> + yl_od = __lasx_xvsrai_w(yl_od, 19);
>> + yh_od = __lasx_xvsrai_w(yh_od, 19);
>> + u_ev = __lasx_xvsrai_w(u_ev, 19);
>> + v_ev = __lasx_xvsrai_w(v_ev, 19);
>> + u_od = __lasx_xvsrai_w(u_od, 19);
>> + v_od = __lasx_xvsrai_w(v_od, 19);
>> + u_ev = __lasx_xvadd_w(u_ev, headroom);
>> + v_ev = __lasx_xvadd_w(v_ev, headroom);
>> + u_od = __lasx_xvadd_w(u_od, headroom);
>> + v_od = __lasx_xvadd_w(v_od, headroom);
>> + WRITE_YUV2RGB(yl_ev, yl_od, u_ev, v_ev, 0, 0, 0, 0);
>> + WRITE_YUV2RGB(yl_ev, yl_od, u_od, v_od, 1, 1, 0, 0);
>> + WRITE_YUV2RGB(yl_ev, yl_od, u_ev, v_ev, 2, 2, 1, 1);
>> + WRITE_YUV2RGB(yl_ev, yl_od, u_od, v_od, 3, 3, 1, 1);
>> + WRITE_YUV2RGB(yl_ev, yl_od, u_ev, v_ev, 4, 4, 2, 2);
>> + WRITE_YUV2RGB(yl_ev, yl_od, u_od, v_od, 5, 5, 2, 2);
>> + WRITE_YUV2RGB(yl_ev, yl_od, u_ev, v_ev, 6, 6, 3, 3);
>> + WRITE_YUV2RGB(yl_ev, yl_od, u_od, v_od, 7, 7, 3, 3);
>> + WRITE_YUV2RGB(yh_ev, yh_od, u_ev, v_ev, 0, 0, 4, 4);
>> + WRITE_YUV2RGB(yh_ev, yh_od, u_od, v_od, 1, 1, 4, 4);
>> + WRITE_YUV2RGB(yh_ev, yh_od, u_ev, v_ev, 2, 2, 5, 5);
>> + WRITE_YUV2RGB(yh_ev, yh_od, u_od, v_od, 3, 3, 5, 5);
>> + WRITE_YUV2RGB(yh_ev, yh_od, u_ev, v_ev, 4, 4, 6, 6);
>> + WRITE_YUV2RGB(yh_ev, yh_od, u_od, v_od, 5, 5, 6, 6);
>> + WRITE_YUV2RGB(yh_ev, yh_od, u_ev, v_ev, 6, 6, 7, 7);
>> + WRITE_YUV2RGB(yh_ev, yh_od, u_od, v_od, 7, 7, 7, 7);
>> + res -= 32;
>> + }
>> + if (res >= 16) {
>> + int Y1, Y2, U, V;
>> + int count_lum = count << 1;
>> + __m256i l_src, u_src, v_src;
>> + __m256i y_ev, y_od, u, v, temp;
>> +
>> + y_ev = __lasx_xvldrepl_w(&t, 0);
>> + y_od = y_ev;
>> + u = y_ev;
>> + v = y_ev;
>> + for (j = 0; j < lumFilterSize; j++) {
>> + temp = __lasx_xvldrepl_h((lumFilter + j), 0);
>> + l_src = __lasx_xvld(lumSrc[j] + count_lum, 0);
>> + y_ev = __lasx_xvmaddwev_w_h(y_ev, temp, l_src);
>> + y_od = __lasx_xvmaddwod_w_h(y_od, temp, l_src);
>> + }
>> + for (j = 0; j < chrFilterSize; j++) {
>> + DUP2_ARG2(__lasx_xvld, chrUSrc[j] + count, 0, chrVSrc[j] + count,
>> + 0, u_src, v_src);
>> + temp = __lasx_xvldrepl_h((chrFilter + j), 0);
>> + u_src = __lasx_vext2xv_w_h(u_src);
>> + v_src = __lasx_vext2xv_w_h(v_src);
>> + u = __lasx_xvmaddwev_w_h(u, temp, u_src);
>> + v = __lasx_xvmaddwev_w_h(v, temp, v_src);
>> + }
>> + y_ev = __lasx_xvsrai_w(y_ev, 19);
>> + y_od = __lasx_xvsrai_w(y_od, 19);
>> + u = __lasx_xvsrai_w(u, 19);
>> + v = __lasx_xvsrai_w(v, 19);
>> + u = __lasx_xvadd_w(u, headroom);
>> + v = __lasx_xvadd_w(v, headroom);
>> + WRITE_YUV2RGB(y_ev, y_od, u, v, 0, 0, 0, 0);
>> + WRITE_YUV2RGB(y_ev, y_od, u, v, 1, 1, 1, 1);
>> + WRITE_YUV2RGB(y_ev, y_od, u, v, 2, 2, 2, 2);
>> + WRITE_YUV2RGB(y_ev, y_od, u, v, 3, 3, 3, 3);
>> + WRITE_YUV2RGB(y_ev, y_od, u, v, 4, 4, 4, 4);
>> + WRITE_YUV2RGB(y_ev, y_od, u, v, 5, 5, 5, 5);
>> + WRITE_YUV2RGB(y_ev, y_od, u, v, 6, 6, 6, 6);
>> + WRITE_YUV2RGB(y_ev, y_od, u, v, 7, 7, 7, 7);
>> + res -= 16;
>> + }
>> + if (res >= 8) {
>> + int Y1, Y2, U, V;
>> + int count_lum = count << 1;
>> + __m256i l_src, u_src, v_src;
>> + __m256i y_ev, uv, temp;
>> +
>> + y_ev = __lasx_xvldrepl_w(&t, 0);
>> + uv = y_ev;
>> + for (j = 0; j < lumFilterSize; j++) {
>> + temp = __lasx_xvldrepl_h((lumFilter + j), 0);
>> + l_src = __lasx_xvld(lumSrc[j] + count_lum, 0);
>> + l_src = __lasx_vext2xv_w_h(l_src);
>> + y_ev = __lasx_xvmaddwev_w_h(y_ev, temp, l_src);
>> + }
>> + for (j = 0; j < chrFilterSize; j++) {
>> + u_src = __lasx_xvldrepl_d((chrUSrc[j] + count), 0);
>> + v_src = __lasx_xvldrepl_d((chrVSrc[j] + count), 0);
>> + temp = __lasx_xvldrepl_h((chrFilter + j), 0);
>> + u_src = __lasx_xvilvl_d(v_src, u_src);
>> + u_src = __lasx_vext2xv_w_h(u_src);
>> + uv = __lasx_xvmaddwev_w_h(uv, temp, u_src);
>> + }
>> + y_ev = __lasx_xvsrai_w(y_ev, 19);
>> + uv = __lasx_xvsrai_w(uv, 19);
>> + uv = __lasx_xvadd_w(uv, headroom);
>> + WRITE_YUV2RGB(y_ev, y_ev, uv, uv, 0, 1, 0, 4);
>> + WRITE_YUV2RGB(y_ev, y_ev, uv, uv, 2, 3, 1, 5);
>> + WRITE_YUV2RGB(y_ev, y_ev, uv, uv, 4, 5, 2, 6);
>> + WRITE_YUV2RGB(y_ev, y_ev, uv, uv, 6, 7, 3, 7);
>> + }
>> + for (; count < len_count; count++) {
>> + int Y1 = 1 << 18;
>> + int Y2 = Y1;
>> + int U = Y1;
>> + int V = Y1;
>> +
>> + for (j = 0; j < lumFilterSize; j++) {
>> + Y1 += lumSrc[j][count * 2] * lumFilter[j];
>> + Y2 += lumSrc[j][count * 2 + 1] * lumFilter[j];
>> + }
>> + for (j = 0; j < chrFilterSize; j++) {
>> + U += chrUSrc[j][count] * chrFilter[j];
>> + V += chrVSrc[j][count] * chrFilter[j];
>> + }
>> + Y1 >>= 19;
>> + Y2 >>= 19;
>> + U >>= 19;
>> + V >>= 19;
>> + r = c->table_rV[V + YUVRGB_TABLE_HEADROOM];
>> + g = (c->table_gU[U + YUVRGB_TABLE_HEADROOM] +
>> + c->table_gV[V + YUVRGB_TABLE_HEADROOM]);
>> + b = c->table_bU[U + YUVRGB_TABLE_HEADROOM];
>> +
>> + yuv2rgb_write(dest, count, Y1, Y2, 0, 0,
>> + r, g, b, y, target, 0);
>> + }
>> +}
>> +
>> +static void
>> +yuv2rgb_2_template_lasx(SwsContext *c, const int16_t *buf[2],
>> + const int16_t *ubuf[2], const int16_t *vbuf[2],
>> + const int16_t *abuf[2], uint8_t *dest, int dstW,
>> + int yalpha, int uvalpha, int y,
>> + enum AVPixelFormat target, int hasAlpha)
>> +{
>> + const int16_t *buf0 = buf[0], *buf1 = buf[1],
>> + *ubuf0 = ubuf[0], *ubuf1 = ubuf[1],
>> + *vbuf0 = vbuf[0], *vbuf1 = vbuf[1];
>> + int yalpha1 = 4096 - yalpha;
>> + int uvalpha1 = 4096 - uvalpha;
>> + int i, count = 0;
>> + int len = dstW - 15;
>> + int len_count = (dstW + 1) >> 1;
>> + const void *r, *g, *b;
>> + int head = YUVRGB_TABLE_HEADROOM;
>> + __m256i v_yalpha1 = __lasx_xvreplgr2vr_w(yalpha1);
>> + __m256i v_uvalpha1 = __lasx_xvreplgr2vr_w(uvalpha1);
>> + __m256i v_yalpha = __lasx_xvreplgr2vr_w(yalpha);
>> + __m256i v_uvalpha = __lasx_xvreplgr2vr_w(uvalpha);
>> + __m256i headroom = __lasx_xvreplgr2vr_w(head);
>> +
>> + for (i = 0; i < len; i += 16) {
>> + int Y1, Y2, U, V;
>> + int i_dex = i << 1;
>> + int c_dex = count << 1;
>> + __m256i y0_h, y0_l, y0, u0, v0;
>> + __m256i y1_h, y1_l, y1, u1, v1;
>> + __m256i y_l, y_h, u, v;
>> +
>> + DUP4_ARG2(__lasx_xvldx, buf0, i_dex, ubuf0, c_dex, vbuf0, c_dex,
>> + buf1, i_dex, y0, u0, v0, y1);
>> + DUP2_ARG2(__lasx_xvldx, ubuf1, c_dex, vbuf1, c_dex, u1, v1);
>> + DUP2_ARG2(__lasx_xvsllwil_w_h, y0, 0, y1, 0, y0_l, y1_l);
>> + DUP2_ARG1(__lasx_xvexth_w_h, y0, y1, y0_h, y1_h);
>> + DUP4_ARG1(__lasx_vext2xv_w_h, u0, u1, v0, v1, u0, u1, v0, v1);
>> + y0_l = __lasx_xvmul_w(y0_l, v_yalpha1);
>> + y0_h = __lasx_xvmul_w(y0_h, v_yalpha1);
>> + u0 = __lasx_xvmul_w(u0, v_uvalpha1);
>> + v0 = __lasx_xvmul_w(v0, v_uvalpha1);
>> + y_l = __lasx_xvmadd_w(y0_l, v_yalpha, y1_l);
>> + y_h = __lasx_xvmadd_w(y0_h, v_yalpha, y1_h);
>> + u = __lasx_xvmadd_w(u0, v_uvalpha, u1);
>> + v = __lasx_xvmadd_w(v0, v_uvalpha, v1);
>> + y_l = __lasx_xvsrai_w(y_l, 19);
>> + y_h = __lasx_xvsrai_w(y_h, 19);
>> + u = __lasx_xvsrai_w(u, 19);
>> + v = __lasx_xvsrai_w(v, 19);
>> + u = __lasx_xvadd_w(u, headroom);
>> + v = __lasx_xvadd_w(v, headroom);
>> + WRITE_YUV2RGB(y_l, y_l, u, v, 0, 1, 0, 0);
>> + WRITE_YUV2RGB(y_l, y_l, u, v, 2, 3, 1, 1);
>> + WRITE_YUV2RGB(y_h, y_h, u, v, 0, 1, 2, 2);
>> + WRITE_YUV2RGB(y_h, y_h, u, v, 2, 3, 3, 3);
>> + WRITE_YUV2RGB(y_l, y_l, u, v, 4, 5, 4, 4);
>> + WRITE_YUV2RGB(y_l, y_l, u, v, 6, 7, 5, 5);
>> + WRITE_YUV2RGB(y_h, y_h, u, v, 4, 5, 6, 6);
>> + WRITE_YUV2RGB(y_h, y_h, u, v, 6, 7, 7, 7);
>> + }
>> + if (dstW - i >= 8) {
>> + int Y1, Y2, U, V;
>> + int i_dex = i << 1;
>> + __m256i y0_l, y0, u0, v0;
>> + __m256i y1_l, y1, u1, v1;
>> + __m256i y_l, u, v;
>> +
>> + y0 = __lasx_xvldx(buf0, i_dex);
>
> 1. Not long ago, I tried to constify the src pointer of several asm
> functions and noticed that they produced new warnings for loongarch
> (according to patchwork:
> https://patchwork.ffmpeg.org/project/ffmpeg/patch/DB6PR0101MB2214178D3E6B8DCA5B86F8198F9A9@DB6PR0101MB2214.eurprd01.prod.exchangelabs.com/),
> even though I was sure that the code is const-correct. After finding
> (via https://github.com/opencv/opencv/pull/21833) a toolchain
> (https://gitee.com/wenux/cross-compiler-la-on-x86) that can build the
> lasx and lsx code (upstream GCC seems to be lacking lsx and lasx support
> at the moment; at least, my self-compiled loongarch-GCC did not support
> lsx and lasx) the issue was clear: lsxintrin.h and lasxintrin.h do not
> use const at all, even for functions that only read data (I presume the
> vl in __lsx_vldx stands for "vector load"?).
> So I sent another iteration
> https://ffmpeg.org/pipermail/ffmpeg-devel/2022-August/299562.html of
> that patchset that now added wrappers for __lsx_vldx() and
> __lasx_xvldx() and cc'ed you and some other developers from loongson to
> alert you of the issue in the hope that you fix the headers, so that my
> wrappers wouldn't need to be applied. That didn't work, as my mails
> could not be delivered to you. So I applied the patchset.
> 2. You use __lasx_xvldx() to read from a const int16_t. This will give
> new warnings unless the above issue has been fixed. Has it?
> 3. I don't know whether it has, as patchwork's fate tests don't work for
> a few days already. Given that the mails I receive from patchwork when
> it doesn't like a commit message arrive from "Patchwork
> <yinshiyou-hf@loongson.cn>" I presume that loongson is now somehow
> running patchwork, so you should be able to inform the right people to
> fix it.
> 4. If you fixed the const-issue, can you please make an updated
> toolchain with lsx and lasx support enabled available to us?
>
> - Andreas
>
Hi Andreas,
Sorry for the late reply.
This issue will be fixed by using const for v1 of __lsx_vldx, and I will update toolchain of LoongArch patchwork runner ASAP.
Thank you very much for analyzing this problem and giving suggestion.
Thanks,
Shiyou
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
next prev parent reply other threads:[~2022-09-06 8:12 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-08-29 11:26 [FFmpeg-devel] Add loongarch SIMD optimization in swscale lib Hao Chen
2022-08-29 11:26 ` [FFmpeg-devel] [PATCH v1 1/3] swscale/la: Optimize hscale functions with lasx Hao Chen
2022-08-29 11:26 ` [FFmpeg-devel] [PATCH v1 2/3] swscale/la: Add yuv2rgb_lasx.c and rgb2rgb_lasx.c files Hao Chen
2022-08-29 11:26 ` [FFmpeg-devel] [PATCH v1 3/3] swscale/la: Add output_lasx.c file Hao Chen
2022-08-29 12:30 ` Andreas Rheinhardt
2022-09-06 8:12 ` Shiyou Yin [this message]
2022-09-09 1:22 ` Shiyou Yin
2022-09-09 13:11 ` Andreas Rheinhardt
2022-09-11 2:06 ` Shiyou Yin
2022-09-21 3:03 ` Shiyou Yin
2022-09-21 10:50 ` Andreas Rheinhardt
2022-08-29 11:51 [FFmpeg-devel] Add LoongArch SIMD optimization in swscale lib Hao Chen
2022-08-29 11:51 ` [FFmpeg-devel] [PATCH v1 3/3] swscale/la: Add output_lasx.c file Hao Chen
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=ABB32204-033B-410E-8DA7-A41064845E48@loongson.cn \
--to=yinshiyou-hf@loongson.cn \
--cc=andreas.rheinhardt@outlook.com \
--cc=ffmpeg-devel@ffmpeg.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
This inbox may be cloned and mirrored by anyone:
git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git
# If you have public-inbox 1.1+ installed, you may
# initialize and index your mirror using the following commands:
public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
ffmpegdev@gitmailbox.com
public-inbox-index ffmpegdev
Example config snippet for mirrors.
AGPL code for this site: git clone https://public-inbox.org/public-inbox.git