Chiebot-Mirror
/
FFmpeg
mirror of https://github.com/FFmpeg/FFmpeg.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
117966 Commits
37 Branches
400 Tags
476 MiB
Tag: Branch: Tree: 4ec767abcc
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from '4ec767abcc'
${ noResults }
FFmpeg/libswscale/loongarch/output_lsx.c

1849 lines
76 KiB
Raw Normal View History Unescape

swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
/*
* Copyright (C) 2023 Loongson Technology Corporation Limited
* Contributed by Lu Wang <wanglu@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"
/*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_LSX(vec_y1, vec_y2, vec_u, vec_v, t1, t2, t3, t4) \
{ \
Y1 = __lsx_vpickve2gr_w(vec_y1, t1); \
Y2 = __lsx_vpickve2gr_w(vec_y2, t2); \
U = __lsx_vpickve2gr_w(vec_u, t3); \
V = __lsx_vpickve2gr_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
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
yuv2rgb_X_template_lsx(SwsInternal *c, const int16_t *lumFilter,
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
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 >> 5;
int res = dstW & 31;
int len_count = (dstW + 1) >> 1;
const void *r, *g, *b;
int head = YUVRGB_TABLE_HEADROOM;
__m128i headroom = __lsx_vreplgr2vr_w(head);
for (i = 0; i < len; i++) {
int Y1, Y2, U, V, count_lum = count << 1;
__m128i l_src1, l_src2, l_src3, l_src4, u_src1, u_src2, v_src1, v_src2;
__m128i yl_ev, yl_ev1, yl_ev2, yl_od1, yl_od2, yh_ev1, yh_ev2, yh_od1, yh_od2;
__m128i u_ev1, u_ev2, u_od1, u_od2, v_ev1, v_ev2, v_od1, v_od2, temp;
yl_ev = __lsx_vldrepl_w(&t, 0);
yl_ev1 = yl_ev;
yl_od1 = yl_ev;
yh_ev1 = yl_ev;
yh_od1 = yl_ev;
u_ev1 = yl_ev;
v_ev1 = yl_ev;
u_od1 = yl_ev;
v_od1 = yl_ev;
yl_ev2 = yl_ev;
yl_od2 = yl_ev;
yh_ev2 = yl_ev;
yh_od2 = yl_ev;
u_ev2 = yl_ev;
v_ev2 = yl_ev;
u_od2 = yl_ev;
v_od2 = yl_ev;
for (j = 0; j < lumFilterSize; j++) {
temp = __lsx_vldrepl_h((lumFilter + j), 0);
DUP2_ARG2(__lsx_vld, lumSrc[j] + count_lum, 0, lumSrc[j] + count_lum,
16, l_src1, l_src2);
DUP2_ARG2(__lsx_vld, lumSrc[j] + count_lum, 32, lumSrc[j] + count_lum,
48, l_src3, l_src4);
yl_ev1 = __lsx_vmaddwev_w_h(yl_ev1, temp, l_src1);
yl_od1 = __lsx_vmaddwod_w_h(yl_od1, temp, l_src1);
yh_ev1 = __lsx_vmaddwev_w_h(yh_ev1, temp, l_src3);
yh_od1 = __lsx_vmaddwod_w_h(yh_od1, temp, l_src3);
yl_ev2 = __lsx_vmaddwev_w_h(yl_ev2, temp, l_src2);
yl_od2 = __lsx_vmaddwod_w_h(yl_od2, temp, l_src2);
yh_ev2 = __lsx_vmaddwev_w_h(yh_ev2, temp, l_src4);
yh_od2 = __lsx_vmaddwod_w_h(yh_od2, temp, l_src4);
}
for (j = 0; j < chrFilterSize; j++) {
DUP2_ARG2(__lsx_vld, chrUSrc[j] + count, 0, chrVSrc[j] + count, 0,
u_src1, v_src1);
DUP2_ARG2(__lsx_vld, chrUSrc[j] + count, 16, chrVSrc[j] + count, 16,
u_src2, v_src2);
temp = __lsx_vldrepl_h((chrFilter + j), 0);
u_ev1 = __lsx_vmaddwev_w_h(u_ev1, temp, u_src1);
u_od1 = __lsx_vmaddwod_w_h(u_od1, temp, u_src1);
v_ev1 = __lsx_vmaddwev_w_h(v_ev1, temp, v_src1);
v_od1 = __lsx_vmaddwod_w_h(v_od1, temp, v_src1);
u_ev2 = __lsx_vmaddwev_w_h(u_ev2, temp, u_src2);
u_od2 = __lsx_vmaddwod_w_h(u_od2, temp, u_src2);
v_ev2 = __lsx_vmaddwev_w_h(v_ev2, temp, v_src2);
v_od2 = __lsx_vmaddwod_w_h(v_od2, temp, v_src2);
}
yl_ev1 = __lsx_vsrai_w(yl_ev1, 19);
yh_ev1 = __lsx_vsrai_w(yh_ev1, 19);
yl_od1 = __lsx_vsrai_w(yl_od1, 19);
yh_od1 = __lsx_vsrai_w(yh_od1, 19);
u_ev1 = __lsx_vsrai_w(u_ev1, 19);
v_ev1 = __lsx_vsrai_w(v_ev1, 19);
u_od1 = __lsx_vsrai_w(u_od1, 19);
v_od1 = __lsx_vsrai_w(v_od1, 19);
yl_ev2 = __lsx_vsrai_w(yl_ev2, 19);
yh_ev2 = __lsx_vsrai_w(yh_ev2, 19);
yl_od2 = __lsx_vsrai_w(yl_od2, 19);
yh_od2 = __lsx_vsrai_w(yh_od2, 19);
u_ev2 = __lsx_vsrai_w(u_ev2, 19);
v_ev2 = __lsx_vsrai_w(v_ev2, 19);
u_od2 = __lsx_vsrai_w(u_od2, 19);
v_od2 = __lsx_vsrai_w(v_od2, 19);
u_ev1 = __lsx_vadd_w(u_ev1, headroom);
v_ev1 = __lsx_vadd_w(v_ev1, headroom);
u_od1 = __lsx_vadd_w(u_od1, headroom);
v_od1 = __lsx_vadd_w(v_od1, headroom);
u_ev2 = __lsx_vadd_w(u_ev2, headroom);
v_ev2 = __lsx_vadd_w(v_ev2, headroom);
u_od2 = __lsx_vadd_w(u_od2, headroom);
v_od2 = __lsx_vadd_w(v_od2, headroom);
WRITE_YUV2RGB_LSX(yl_ev1, yl_od1, u_ev1, v_ev1, 0, 0, 0, 0);
WRITE_YUV2RGB_LSX(yl_ev1, yl_od1, u_od1, v_od1, 1, 1, 0, 0);
WRITE_YUV2RGB_LSX(yl_ev1, yl_od1, u_ev1, v_ev1, 2, 2, 1, 1);
WRITE_YUV2RGB_LSX(yl_ev1, yl_od1, u_od1, v_od1, 3, 3, 1, 1);
WRITE_YUV2RGB_LSX(yl_ev2, yl_od2, u_ev1, v_ev1, 0, 0, 2, 2);
WRITE_YUV2RGB_LSX(yl_ev2, yl_od2, u_od1, v_od1, 1, 1, 2, 2);
WRITE_YUV2RGB_LSX(yl_ev2, yl_od2, u_ev1, v_ev1, 2, 2, 3, 3);
WRITE_YUV2RGB_LSX(yl_ev2, yl_od2, u_od1, v_od1, 3, 3, 3, 3);
WRITE_YUV2RGB_LSX(yh_ev1, yh_od1, u_ev2, v_ev2, 0, 0, 0, 0);
WRITE_YUV2RGB_LSX(yh_ev1, yh_od1, u_od2, v_od2, 1, 1, 0, 0);
WRITE_YUV2RGB_LSX(yh_ev1, yh_od1, u_ev2, v_ev2, 2, 2, 1, 1);
WRITE_YUV2RGB_LSX(yh_ev1, yh_od1, u_od2, v_od2, 3, 3, 1, 1);
WRITE_YUV2RGB_LSX(yh_ev2, yh_od2, u_ev2, v_ev2, 0, 0, 2, 2);
WRITE_YUV2RGB_LSX(yh_ev2, yh_od2, u_od2, v_od2, 1, 1, 2, 2);
WRITE_YUV2RGB_LSX(yh_ev2, yh_od2, u_ev2, v_ev2, 2, 2, 3, 3);
WRITE_YUV2RGB_LSX(yh_ev2, yh_od2, u_od2, v_od2, 3, 3, 3, 3);
}
if (res >= 16) {
int Y1, Y2, U, V, count_lum = count << 1;
__m128i l_src1, l_src2, u_src1, v_src1;
__m128i yl_ev, yl_ev1, yl_ev2, yl_od1, yl_od2;
__m128i u_ev1, u_od1, v_ev1, v_od1, temp;
yl_ev = __lsx_vldrepl_w(&t, 0);
yl_ev1 = yl_ev;
yl_od1 = yl_ev;
u_ev1 = yl_ev;
v_ev1 = yl_ev;
u_od1 = yl_ev;
v_od1 = yl_ev;
yl_ev2 = yl_ev;
yl_od2 = yl_ev;
for (j = 0; j < lumFilterSize; j++) {
temp = __lsx_vldrepl_h((lumFilter + j), 0);
DUP2_ARG2(__lsx_vld, lumSrc[j] + count_lum, 0, lumSrc[j] + count_lum,
16, l_src1, l_src2);
yl_ev1 = __lsx_vmaddwev_w_h(yl_ev1, temp, l_src1);
yl_od1 = __lsx_vmaddwod_w_h(yl_od1, temp, l_src1);
yl_ev2 = __lsx_vmaddwev_w_h(yl_ev2, temp, l_src2);
yl_od2 = __lsx_vmaddwod_w_h(yl_od2, temp, l_src2);
}
for (j = 0; j < chrFilterSize; j++) {
DUP2_ARG2(__lsx_vld, chrUSrc[j] + count, 0, chrVSrc[j] + count, 0,
u_src1, v_src1);
temp = __lsx_vldrepl_h((chrFilter + j), 0);
u_ev1 = __lsx_vmaddwev_w_h(u_ev1, temp, u_src1);
u_od1 = __lsx_vmaddwod_w_h(u_od1, temp, u_src1);
v_ev1 = __lsx_vmaddwev_w_h(v_ev1, temp, v_src1);
v_od1 = __lsx_vmaddwod_w_h(v_od1, temp, v_src1);
}
yl_ev1 = __lsx_vsrai_w(yl_ev1, 19);
yl_od1 = __lsx_vsrai_w(yl_od1, 19);
u_ev1 = __lsx_vsrai_w(u_ev1, 19);
v_ev1 = __lsx_vsrai_w(v_ev1, 19);
u_od1 = __lsx_vsrai_w(u_od1, 19);
v_od1 = __lsx_vsrai_w(v_od1, 19);
yl_ev2 = __lsx_vsrai_w(yl_ev2, 19);
yl_od2 = __lsx_vsrai_w(yl_od2, 19);
u_ev1 = __lsx_vadd_w(u_ev1, headroom);
v_ev1 = __lsx_vadd_w(v_ev1, headroom);
u_od1 = __lsx_vadd_w(u_od1, headroom);
v_od1 = __lsx_vadd_w(v_od1, headroom);
WRITE_YUV2RGB_LSX(yl_ev1, yl_od1, u_ev1, v_ev1, 0, 0, 0, 0);
WRITE_YUV2RGB_LSX(yl_ev1, yl_od1, u_od1, v_od1, 1, 1, 0, 0);
WRITE_YUV2RGB_LSX(yl_ev1, yl_od1, u_ev1, v_ev1, 2, 2, 1, 1);
WRITE_YUV2RGB_LSX(yl_ev1, yl_od1, u_od1, v_od1, 3, 3, 1, 1);
WRITE_YUV2RGB_LSX(yl_ev2, yl_od2, u_ev1, v_ev1, 0, 0, 2, 2);
WRITE_YUV2RGB_LSX(yl_ev2, yl_od2, u_od1, v_od1, 1, 1, 2, 2);
WRITE_YUV2RGB_LSX(yl_ev2, yl_od2, u_ev1, v_ev1, 2, 2, 3, 3);
WRITE_YUV2RGB_LSX(yl_ev2, yl_od2, u_od1, v_od1, 3, 3, 3, 3);
res -= 16;
}
if (res >= 8) {
int Y1, Y2, U, V, count_lum = count << 1;
__m128i l_src1, u_src, v_src;
__m128i yl_ev, yl_od;
__m128i u_ev, u_od, v_ev, v_od, temp;
yl_ev = __lsx_vldrepl_w(&t, 0);
yl_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 = __lsx_vldrepl_h((lumFilter + j), 0);
l_src1 = __lsx_vld(lumSrc[j] + count_lum, 0);
yl_ev = __lsx_vmaddwev_w_h(yl_ev, temp, l_src1);
yl_od = __lsx_vmaddwod_w_h(yl_od, temp, l_src1);
}
for (j = 0; j < chrFilterSize; j++) {
DUP2_ARG2(__lsx_vld, chrUSrc[j] + count, 0, chrVSrc[j] + count, 0,
u_src, v_src);
temp = __lsx_vldrepl_h((chrFilter + j), 0);
u_ev = __lsx_vmaddwev_w_h(u_ev, temp, u_src);
u_od = __lsx_vmaddwod_w_h(u_od, temp, u_src);
v_ev = __lsx_vmaddwev_w_h(v_ev, temp, v_src);
v_od = __lsx_vmaddwod_w_h(v_od, temp, v_src);
}
yl_ev = __lsx_vsrai_w(yl_ev, 19);
yl_od = __lsx_vsrai_w(yl_od, 19);
u_ev = __lsx_vsrai_w(u_ev, 19);
v_ev = __lsx_vsrai_w(v_ev, 19);
u_od = __lsx_vsrai_w(u_od, 19);
v_od = __lsx_vsrai_w(v_od, 19);
u_ev = __lsx_vadd_w(u_ev, headroom);
v_ev = __lsx_vadd_w(v_ev, headroom);
u_od = __lsx_vadd_w(u_od, headroom);
v_od = __lsx_vadd_w(v_od, headroom);
WRITE_YUV2RGB_LSX(yl_ev, yl_od, u_ev, v_ev, 0, 0, 0, 0);
WRITE_YUV2RGB_LSX(yl_ev, yl_od, u_od, v_od, 1, 1, 0, 0);
WRITE_YUV2RGB_LSX(yl_ev, yl_od, u_ev, v_ev, 2, 2, 1, 1);
WRITE_YUV2RGB_LSX(yl_ev, yl_od, u_od, v_od, 3, 3, 1, 1);
res -= 8;
}
if (res >= 4) {
int Y1, Y2, U, V, count_lum = count << 1;
__m128i l_src1, u_src, v_src;
__m128i yl_ev, yl_od;
__m128i u_ev, u_od, v_ev, v_od, temp;
yl_ev = __lsx_vldrepl_w(&t, 0);
yl_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 = __lsx_vldrepl_h((lumFilter + j), 0);
l_src1 = __lsx_vld(lumSrc[j] + count_lum, 0);
yl_ev = __lsx_vmaddwev_w_h(yl_ev, temp, l_src1);
yl_od = __lsx_vmaddwod_w_h(yl_od, temp, l_src1);
}
for (j = 0; j < chrFilterSize; j++) {
DUP2_ARG2(__lsx_vld, chrUSrc[j] + count, 0, chrVSrc[j] + count, 0,
u_src, v_src);
temp = __lsx_vldrepl_h((chrFilter + j), 0);
u_ev = __lsx_vmaddwev_w_h(u_ev, temp, u_src);
u_od = __lsx_vmaddwod_w_h(u_od, temp, u_src);
v_ev = __lsx_vmaddwev_w_h(v_ev, temp, v_src);
v_od = __lsx_vmaddwod_w_h(v_od, temp, v_src);
}
yl_ev = __lsx_vsrai_w(yl_ev, 19);
yl_od = __lsx_vsrai_w(yl_od, 19);
u_ev = __lsx_vsrai_w(u_ev, 19);
v_ev = __lsx_vsrai_w(v_ev, 19);
u_od = __lsx_vsrai_w(u_od, 19);
v_od = __lsx_vsrai_w(v_od, 19);
u_ev = __lsx_vadd_w(u_ev, headroom);
v_ev = __lsx_vadd_w(v_ev, headroom);
u_od = __lsx_vadd_w(u_od, headroom);
v_od = __lsx_vadd_w(v_od, headroom);
WRITE_YUV2RGB_LSX(yl_ev, yl_od, u_ev, v_ev, 0, 0, 0, 0);
WRITE_YUV2RGB_LSX(yl_ev, yl_od, u_od, v_od, 1, 1, 0, 0);
res -= 4;
}
if (res >= 2) {
int Y1, Y2, U, V, count_lum = count << 1;
__m128i l_src1, u_src, v_src;
__m128i yl_ev, yl_od;
__m128i u_ev, u_od, v_ev, v_od, temp;
yl_ev = __lsx_vldrepl_w(&t, 0);
yl_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 = __lsx_vldrepl_h((lumFilter + j), 0);
l_src1 = __lsx_vld(lumSrc[j] + count_lum, 0);
yl_ev = __lsx_vmaddwev_w_h(yl_ev, temp, l_src1);
yl_od = __lsx_vmaddwod_w_h(yl_od, temp, l_src1);
}
for (j = 0; j < chrFilterSize; j++) {
DUP2_ARG2(__lsx_vld, chrUSrc[j] + count, 0, chrVSrc[j] + count, 0,
u_src, v_src);
temp = __lsx_vldrepl_h((chrFilter + j), 0);
u_ev = __lsx_vmaddwev_w_h(u_ev, temp, u_src);
u_od = __lsx_vmaddwod_w_h(u_od, temp, u_src);
v_ev = __lsx_vmaddwev_w_h(v_ev, temp, v_src);
v_od = __lsx_vmaddwod_w_h(v_od, temp, v_src);
}
yl_ev = __lsx_vsrai_w(yl_ev, 19);
yl_od = __lsx_vsrai_w(yl_od, 19);
u_ev = __lsx_vsrai_w(u_ev, 19);
v_ev = __lsx_vsrai_w(v_ev, 19);
u_od = __lsx_vsrai_w(u_od, 19);
v_od = __lsx_vsrai_w(v_od, 19);
u_ev = __lsx_vadd_w(u_ev, headroom);
v_ev = __lsx_vadd_w(v_ev, headroom);
u_od = __lsx_vadd_w(u_od, headroom);
v_od = __lsx_vadd_w(v_od, headroom);
WRITE_YUV2RGB_LSX(yl_ev, yl_od, u_ev, v_ev, 0, 0, 0, 0);
res -= 2;
}
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
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
yuv2rgb_2_template_lsx(SwsInternal *c, const int16_t *buf[2],
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
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 - 7;
int len_count = (dstW + 1) >> 1;
const void *r, *g, *b;
int head = YUVRGB_TABLE_HEADROOM;
__m128i v_yalpha1 = __lsx_vreplgr2vr_w(yalpha1);
__m128i v_uvalpha1 = __lsx_vreplgr2vr_w(uvalpha1);
__m128i v_yalpha = __lsx_vreplgr2vr_w(yalpha);
__m128i v_uvalpha = __lsx_vreplgr2vr_w(uvalpha);
__m128i headroom = __lsx_vreplgr2vr_w(head);
__m128i zero = __lsx_vldi(0);
for (i = 0; i < len; i += 8) {
int Y1, Y2, U, V;
int i_dex = i << 1;
int c_dex = count << 1;
__m128i y0_h, y0_l, y0, u0, v0;
__m128i y1_h, y1_l, y1, u1, v1;
__m128i y_l, y_h, u, v;
DUP4_ARG2(__lsx_vldx, buf0, i_dex, ubuf0, c_dex, vbuf0, c_dex,
buf1, i_dex, y0, u0, v0, y1);
DUP2_ARG2(__lsx_vldx, ubuf1, c_dex, vbuf1, c_dex, u1, v1);
DUP2_ARG2(__lsx_vsllwil_w_h, y0, 0, y1, 0, y0_l, y1_l);
DUP2_ARG1(__lsx_vexth_w_h, y0, y1, y0_h, y1_h);
DUP4_ARG2(__lsx_vilvl_h, zero, u0, zero, u1, zero, v0, zero, v1,
u0, u1, v0, v1);
y0_l = __lsx_vmul_w(y0_l, v_yalpha1);
y0_h = __lsx_vmul_w(y0_h, v_yalpha1);
u0 = __lsx_vmul_w(u0, v_uvalpha1);
v0 = __lsx_vmul_w(v0, v_uvalpha1);
y_l = __lsx_vmadd_w(y0_l, v_yalpha, y1_l);
y_h = __lsx_vmadd_w(y0_h, v_yalpha, y1_h);
u = __lsx_vmadd_w(u0, v_uvalpha, u1);
v = __lsx_vmadd_w(v0, v_uvalpha, v1);
y_l = __lsx_vsrai_w(y_l, 19);
y_h = __lsx_vsrai_w(y_h, 19);
u = __lsx_vsrai_w(u, 19);
v = __lsx_vsrai_w(v, 19);
u = __lsx_vadd_w(u, headroom);
v = __lsx_vadd_w(v, headroom);
WRITE_YUV2RGB_LSX(y_l, y_l, u, v, 0, 1, 0, 0);
WRITE_YUV2RGB_LSX(y_l, y_l, u, v, 2, 3, 1, 1);
WRITE_YUV2RGB_LSX(y_h, y_h, u, v, 0, 1, 2, 2);
WRITE_YUV2RGB_LSX(y_h, y_h, u, v, 2, 3, 3, 3);
}
if (dstW - i >= 4) {
int Y1, Y2, U, V;
int i_dex = i << 1;
__m128i y0_l, y0, u0, v0;
__m128i y1_l, y1, u1, v1;
__m128i y_l, u, v;
y0 = __lsx_vldx(buf0, i_dex);
u0 = __lsx_vldrepl_d((ubuf0 + count), 0);
v0 = __lsx_vldrepl_d((vbuf0 + count), 0);
y1 = __lsx_vldx(buf1, i_dex);
u1 = __lsx_vldrepl_d((ubuf1 + count), 0);
v1 = __lsx_vldrepl_d((vbuf1 + count), 0);
DUP2_ARG2(__lsx_vilvl_h, zero, y0, zero, y1, y0_l, y1_l);
DUP4_ARG2(__lsx_vilvl_h, zero, u0, zero, u1, zero, v0, zero, v1,
u0, u1, v0, v1);
y0_l = __lsx_vmul_w(y0_l, v_yalpha1);
u0 = __lsx_vmul_w(u0, v_uvalpha1);
v0 = __lsx_vmul_w(v0, v_uvalpha1);
y_l = __lsx_vmadd_w(y0_l, v_yalpha, y1_l);
u = __lsx_vmadd_w(u0, v_uvalpha, u1);
v = __lsx_vmadd_w(v0, v_uvalpha, v1);
y_l = __lsx_vsrai_w(y_l, 19);
u = __lsx_vsrai_w(u, 19);
v = __lsx_vsrai_w(v, 19);
u = __lsx_vadd_w(u, headroom);
v = __lsx_vadd_w(v, headroom);
WRITE_YUV2RGB_LSX(y_l, y_l, u, v, 0, 1, 0, 0);
WRITE_YUV2RGB_LSX(y_l, y_l, u, v, 2, 3, 1, 1);
i += 4;
}
for (; count < len_count; count++) {
int Y1 = (buf0[count * 2] * yalpha1 +
buf1[count * 2] * yalpha) >> 19;
int Y2 = (buf0[count * 2 + 1] * yalpha1 +
buf1[count * 2 + 1] * yalpha) >> 19;
int U = (ubuf0[count] * uvalpha1 + ubuf1[count] * uvalpha) >> 19;
int V = (vbuf0[count] * uvalpha1 + vbuf1[count] * uvalpha) >> 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
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
yuv2rgb_1_template_lsx(SwsInternal *c, const int16_t *buf0,
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
const int16_t *ubuf[2], const int16_t *vbuf[2],
const int16_t *abuf0, uint8_t *dest, int dstW,
int uvalpha, int y, enum AVPixelFormat target,
int hasAlpha)
{
const int16_t *ubuf0 = ubuf[0], *vbuf0 = vbuf[0];
int i;
int len = (dstW - 7);
int len_count = (dstW + 1) >> 1;
const void *r, *g, *b;
if (uvalpha < 2048) {
int count = 0;
int head = YUVRGB_TABLE_HEADROOM;
__m128i headroom = __lsx_vreplgr2vr_h(head);
for (i = 0; i < len; i += 8) {
int Y1, Y2, U, V;
int i_dex = i << 1;
int c_dex = count << 1;
__m128i src_y, src_u, src_v;
__m128i u, v, uv, y_l, y_h;
src_y = __lsx_vldx(buf0, i_dex);
DUP2_ARG2(__lsx_vldx, ubuf0, c_dex, vbuf0, c_dex, src_u, src_v);
src_y = __lsx_vsrari_h(src_y, 7);
src_u = __lsx_vsrari_h(src_u, 7);
src_v = __lsx_vsrari_h(src_v, 7);
y_l = __lsx_vsllwil_w_h(src_y, 0);
y_h = __lsx_vexth_w_h(src_y);
uv = __lsx_vilvl_h(src_v, src_u);
u = __lsx_vaddwev_w_h(uv, headroom);
v = __lsx_vaddwod_w_h(uv, headroom);
WRITE_YUV2RGB_LSX(y_l, y_l, u, v, 0, 1, 0, 0);
WRITE_YUV2RGB_LSX(y_l, y_l, u, v, 2, 3, 1, 1);
WRITE_YUV2RGB_LSX(y_h, y_h, u, v, 0, 1, 2, 2);
WRITE_YUV2RGB_LSX(y_h, y_h, u, v, 2, 3, 3, 3);
}
if (dstW - i >= 4){
int Y1, Y2, U, V;
int i_dex = i << 1;
__m128i src_y, src_u, src_v;
__m128i y_l, u, v, uv;
src_y = __lsx_vldx(buf0, i_dex);
src_u = __lsx_vldrepl_d((ubuf0 + count), 0);
src_v = __lsx_vldrepl_d((vbuf0 + count), 0);
y_l = __lsx_vsrari_h(src_y, 7);
y_l = __lsx_vsllwil_w_h(y_l, 0);
uv = __lsx_vilvl_h(src_v, src_u);
uv = __lsx_vsrari_h(uv, 7);
u = __lsx_vaddwev_w_h(uv, headroom);
v = __lsx_vaddwod_w_h(uv, headroom);
WRITE_YUV2RGB_LSX(y_l, y_l, u, v, 0, 1, 0, 0);
WRITE_YUV2RGB_LSX(y_l, y_l, u, v, 2, 3, 1, 1);
i += 4;
}
for (; count < len_count; count++) {
int Y1 = (buf0[count * 2 ] + 64) >> 7;
int Y2 = (buf0[count * 2 + 1] + 64) >> 7;
int U = (ubuf0[count] + 64) >> 7;
int V = (vbuf0[count] + 64) >> 7;
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);
}
} else {
const int16_t *ubuf1 = ubuf[1], *vbuf1 = vbuf[1];
int count = 0;
int HEADROOM = YUVRGB_TABLE_HEADROOM;
__m128i headroom = __lsx_vreplgr2vr_w(HEADROOM);
for (i = 0; i < len; i += 8) {
int Y1, Y2, U, V;
int i_dex = i << 1;
int c_dex = count << 1;
__m128i src_y, src_u0, src_v0, src_u1, src_v1;
__m128i y_l, y_h, u1, u2, v1, v2;
DUP4_ARG2(__lsx_vldx, buf0, i_dex, ubuf0, c_dex, vbuf0, c_dex,
ubuf1, c_dex, src_y, src_u0, src_v0, src_u1);
src_v1 = __lsx_vldx(vbuf1, c_dex);
src_y = __lsx_vsrari_h(src_y, 7);
u1 = __lsx_vaddwev_w_h(src_u0, src_u1);
v1 = __lsx_vaddwod_w_h(src_u0, src_u1);
u2 = __lsx_vaddwev_w_h(src_v0, src_v1);
v2 = __lsx_vaddwod_w_h(src_v0, src_v1);
y_l = __lsx_vsllwil_w_h(src_y, 0);
y_h = __lsx_vexth_w_h(src_y);
u1 = __lsx_vsrari_w(u1, 8);
v1 = __lsx_vsrari_w(v1, 8);
u2 = __lsx_vsrari_w(u2, 8);
v2 = __lsx_vsrari_w(v2, 8);
u1 = __lsx_vadd_w(u1, headroom);
v1 = __lsx_vadd_w(v1, headroom);
u2 = __lsx_vadd_w(u2, headroom);
v2 = __lsx_vadd_w(v2, headroom);
WRITE_YUV2RGB_LSX(y_l, y_l, u1, v1, 0, 1, 0, 0);
WRITE_YUV2RGB_LSX(y_l, y_l, u2, v2, 2, 3, 0, 0);
WRITE_YUV2RGB_LSX(y_h, y_h, u1, v1, 0, 1, 1, 1);
WRITE_YUV2RGB_LSX(y_h, y_h, u2, v2, 2, 3, 1, 1);
}
if (dstW - i >= 4) {
int Y1, Y2, U, V;
int i_dex = i << 1;
__m128i src_y, src_u0, src_v0, src_u1, src_v1;
__m128i uv;
src_y = __lsx_vldx(buf0, i_dex);
src_u0 = __lsx_vldrepl_d((ubuf0 + count), 0);
src_v0 = __lsx_vldrepl_d((vbuf0 + count), 0);
src_u1 = __lsx_vldrepl_d((ubuf1 + count), 0);
src_v1 = __lsx_vldrepl_d((vbuf1 + count), 0);
src_u0 = __lsx_vilvl_h(src_u1, src_u0);
src_v0 = __lsx_vilvl_h(src_v1, src_v0);
src_y = __lsx_vsrari_h(src_y, 7);
src_y = __lsx_vsllwil_w_h(src_y, 0);
uv = __lsx_vilvl_h(src_v0, src_u0);
uv = __lsx_vhaddw_w_h(uv, uv);
uv = __lsx_vsrari_w(uv, 8);
uv = __lsx_vadd_w(uv, headroom);
WRITE_YUV2RGB_LSX(src_y, src_y, uv, uv, 0, 1, 0, 1);
WRITE_YUV2RGB_LSX(src_y, src_y, uv, uv, 2, 3, 2, 3);
i += 4;
}
for (; count < len_count; count++) {
int Y1 = (buf0[count * 2 ] + 64) >> 7;
int Y2 = (buf0[count * 2 + 1] + 64) >> 7;
int U = (ubuf0[count] + ubuf1[count] + 128) >> 8;
int V = (vbuf0[count] + vbuf1[count] + 128) >> 8;
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);
}
}
}
#define YUV2RGBWRAPPERX(name, base, ext, fmt, hasAlpha) \
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
static void name ## ext ## _X_lsx(SwsInternal *c, const int16_t *lumFilter, \
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
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) \
{ \
name ## base ## _X_template_lsx(c, lumFilter, lumSrc, lumFilterSize, \
chrFilter, chrUSrc, chrVSrc, chrFilterSize, \
alpSrc, dest, dstW, y, fmt, hasAlpha); \
}
#define YUV2RGBWRAPPERX2(name, base, ext, fmt, hasAlpha) \
YUV2RGBWRAPPERX(name, base, ext, fmt, hasAlpha) \
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
static void name ## ext ## _2_lsx(SwsInternal *c, const int16_t *buf[2], \
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
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) \
{ \
name ## base ## _2_template_lsx(c, buf, ubuf, vbuf, abuf, dest, \
dstW, yalpha, uvalpha, y, fmt, hasAlpha); \
}
#define YUV2RGBWRAPPER(name, base, ext, fmt, hasAlpha) \
YUV2RGBWRAPPERX2(name, base, ext, fmt, hasAlpha) \
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
static void name ## ext ## _1_lsx(SwsInternal *c, const int16_t *buf0, \
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
const int16_t *ubuf[2], const int16_t *vbuf[2], \
const int16_t *abuf0, uint8_t *dest, int dstW, \
int uvalpha, int y) \
{ \
name ## base ## _1_template_lsx(c, buf0, ubuf, vbuf, abuf0, dest, \
dstW, uvalpha, y, fmt, hasAlpha); \
}
#if CONFIG_SMALL
#else
#if CONFIG_SWSCALE_ALPHA
#endif
YUV2RGBWRAPPER(yuv2rgb,, x32_1, AV_PIX_FMT_RGB32_1, 0)
YUV2RGBWRAPPER(yuv2rgb,, x32, AV_PIX_FMT_RGB32, 0)
#endif
YUV2RGBWRAPPER(yuv2, rgb, rgb24, AV_PIX_FMT_RGB24, 0)
YUV2RGBWRAPPER(yuv2, rgb, bgr24, AV_PIX_FMT_BGR24, 0)
YUV2RGBWRAPPER(yuv2rgb,, 16, AV_PIX_FMT_RGB565, 0)
YUV2RGBWRAPPER(yuv2rgb,, 15, AV_PIX_FMT_RGB555, 0)
YUV2RGBWRAPPER(yuv2rgb,, 12, AV_PIX_FMT_RGB444, 0)
YUV2RGBWRAPPER(yuv2rgb,, 8, AV_PIX_FMT_RGB8, 0)
YUV2RGBWRAPPER(yuv2rgb,, 4, AV_PIX_FMT_RGB4, 0)
YUV2RGBWRAPPER(yuv2rgb,, 4b, AV_PIX_FMT_RGB4_BYTE, 0)
// This function is copied from libswscale/output.c
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
static av_always_inline void yuv2rgb_write_full(SwsInternal *c,
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
uint8_t *dest, int i, int R, int A, int G, int B,
int y, enum AVPixelFormat target, int hasAlpha, int err[4])
{
int isrgb8 = target == AV_PIX_FMT_BGR8 || target == AV_PIX_FMT_RGB8;
if ((R | G | B) & 0xC0000000) {
R = av_clip_uintp2(R, 30);
G = av_clip_uintp2(G, 30);
B = av_clip_uintp2(B, 30);
}
switch(target) {
case AV_PIX_FMT_ARGB:
dest[0] = hasAlpha ? A : 255;
dest[1] = R >> 22;
dest[2] = G >> 22;
dest[3] = B >> 22;
break;
case AV_PIX_FMT_RGB24:
dest[0] = R >> 22;
dest[1] = G >> 22;
dest[2] = B >> 22;
break;
case AV_PIX_FMT_RGBA:
dest[0] = R >> 22;
dest[1] = G >> 22;
dest[2] = B >> 22;
dest[3] = hasAlpha ? A : 255;
break;
case AV_PIX_FMT_ABGR:
dest[0] = hasAlpha ? A : 255;
dest[1] = B >> 22;
dest[2] = G >> 22;
dest[3] = R >> 22;
break;
case AV_PIX_FMT_BGR24:
dest[0] = B >> 22;
dest[1] = G >> 22;
dest[2] = R >> 22;
break;
case AV_PIX_FMT_BGRA:
dest[0] = B >> 22;
dest[1] = G >> 22;
dest[2] = R >> 22;
dest[3] = hasAlpha ? A : 255;
break;
case AV_PIX_FMT_BGR4_BYTE:
case AV_PIX_FMT_RGB4_BYTE:
case AV_PIX_FMT_BGR8:
case AV_PIX_FMT_RGB8:
{
int r,g,b;
swscale/internal: group user-facing options together This is a preliminary step to separating these into a new struct. This commit contains no functional changes, it is a pure search-and-replace. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
switch (c->opts.dither) {
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
default:
case SWS_DITHER_AUTO:
case SWS_DITHER_ED:
R >>= 22;
G >>= 22;
B >>= 22;
R += (7*err[0] + 1*c->dither_error[0][i] + 5*c->dither_error[0][i+1] + 3*c->dither_error[0][i+2])>>4;
G += (7*err[1] + 1*c->dither_error[1][i] + 5*c->dither_error[1][i+1] + 3*c->dither_error[1][i+2])>>4;
B += (7*err[2] + 1*c->dither_error[2][i] + 5*c->dither_error[2][i+1] + 3*c->dither_error[2][i+2])>>4;
c->dither_error[0][i] = err[0];
c->dither_error[1][i] = err[1];
c->dither_error[2][i] = err[2];
r = R >> (isrgb8 ? 5 : 7);
g = G >> (isrgb8 ? 5 : 6);
b = B >> (isrgb8 ? 6 : 7);
r = av_clip(r, 0, isrgb8 ? 7 : 1);
g = av_clip(g, 0, isrgb8 ? 7 : 3);
b = av_clip(b, 0, isrgb8 ? 3 : 1);
err[0] = R - r*(isrgb8 ? 36 : 255);
err[1] = G - g*(isrgb8 ? 36 : 85);
err[2] = B - b*(isrgb8 ? 85 : 255);
break;
case SWS_DITHER_A_DITHER:
if (isrgb8) {
/* see http://pippin.gimp.org/a_dither/ for details/origin */
#define A_DITHER(u,v) (((((u)+((v)*236))*119)&0xff))
r = (((R >> 19) + A_DITHER(i,y) -96)>>8);
g = (((G >> 19) + A_DITHER(i + 17,y) - 96)>>8);
b = (((B >> 20) + A_DITHER(i + 17*2,y) -96)>>8);
r = av_clip_uintp2(r, 3);
g = av_clip_uintp2(g, 3);
b = av_clip_uintp2(b, 2);
} else {
r = (((R >> 21) + A_DITHER(i,y)-256)>>8);
g = (((G >> 19) + A_DITHER(i + 17,y)-256)>>8);
b = (((B >> 21) + A_DITHER(i + 17*2,y)-256)>>8);
r = av_clip_uintp2(r, 1);
g = av_clip_uintp2(g, 2);
b = av_clip_uintp2(b, 1);
}
break;
case SWS_DITHER_X_DITHER:
if (isrgb8) {
/* see http://pippin.gimp.org/a_dither/ for details/origin */
#define X_DITHER(u,v) (((((u)^((v)*237))*181)&0x1ff)/2)
r = (((R >> 19) + X_DITHER(i,y) - 96)>>8);
g = (((G >> 19) + X_DITHER(i + 17,y) - 96)>>8);
b = (((B >> 20) + X_DITHER(i + 17*2,y) - 96)>>8);
r = av_clip_uintp2(r, 3);
g = av_clip_uintp2(g, 3);
b = av_clip_uintp2(b, 2);
} else {
r = (((R >> 21) + X_DITHER(i,y)-256)>>8);
g = (((G >> 19) + X_DITHER(i + 17,y)-256)>>8);
b = (((B >> 21) + X_DITHER(i + 17*2,y)-256)>>8);
r = av_clip_uintp2(r, 1);
g = av_clip_uintp2(g, 2);
b = av_clip_uintp2(b, 1);
}
break;
}
if(target == AV_PIX_FMT_BGR4_BYTE) {
dest[0] = r + 2*g + 8*b;
} else if(target == AV_PIX_FMT_RGB4_BYTE) {
dest[0] = b + 2*g + 8*r;
} else if(target == AV_PIX_FMT_BGR8) {
dest[0] = r + 8*g + 64*b;
} else if(target == AV_PIX_FMT_RGB8) {
dest[0] = b + 4*g + 32*r;
} else
av_assert2(0);
break; }
}
}
#define YUVTORGB_SETUP_LSX \
int y_offset = c->yuv2rgb_y_offset; \
int y_coeff = c->yuv2rgb_y_coeff; \
int v2r_coe = c->yuv2rgb_v2r_coeff; \
int v2g_coe = c->yuv2rgb_v2g_coeff; \
int u2g_coe = c->yuv2rgb_u2g_coeff; \
int u2b_coe = c->yuv2rgb_u2b_coeff; \
__m128i offset = __lsx_vreplgr2vr_w(y_offset); \
__m128i coeff = __lsx_vreplgr2vr_w(y_coeff); \
__m128i v2r = __lsx_vreplgr2vr_w(v2r_coe); \
__m128i v2g = __lsx_vreplgr2vr_w(v2g_coe); \
__m128i u2g = __lsx_vreplgr2vr_w(u2g_coe); \
__m128i u2b = __lsx_vreplgr2vr_w(u2b_coe); \
#define YUVTORGB_LSX(y, u, v, R, G, B, offset, coeff, \
y_temp, v2r, v2g, u2g, u2b) \
{ \
y = __lsx_vsub_w(y, offset); \
y = __lsx_vmul_w(y, coeff); \
y = __lsx_vadd_w(y, y_temp); \
R = __lsx_vmadd_w(y, v, v2r); \
v = __lsx_vmadd_w(y, v, v2g); \
G = __lsx_vmadd_w(v, u, u2g); \
B = __lsx_vmadd_w(y, u, u2b); \
}
#define WRITE_FULL_A_LSX(r, g, b, a, t1, s) \
{ \
R = __lsx_vpickve2gr_w(r, t1); \
G = __lsx_vpickve2gr_w(g, t1); \
B = __lsx_vpickve2gr_w(b, t1); \
A = __lsx_vpickve2gr_w(a, t1); \
if (A & 0x100) \
A = av_clip_uint8(A); \
yuv2rgb_write_full(c, dest, i + s, R, A, G, B, y, target, hasAlpha, err);\
dest += step; \
}
#define WRITE_FULL_LSX(r, g, b, t1, s) \
{ \
R = __lsx_vpickve2gr_w(r, t1); \
G = __lsx_vpickve2gr_w(g, t1); \
B = __lsx_vpickve2gr_w(b, t1); \
yuv2rgb_write_full(c, dest, i + s, R, 0, G, B, y, target, hasAlpha, err); \
dest += step; \
}
static void
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
yuv2rgb_full_X_template_lsx(SwsInternal *c, const int16_t *lumFilter,
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
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, B, G, R, A;
int step = (target == AV_PIX_FMT_RGB24 ||
target == AV_PIX_FMT_BGR24) ? 3 : 4;
int err[4] = {0};
int a_temp = 1 << 18;
int templ = 1 << 9;
int tempc = templ - (128 << 19);
int ytemp = 1 << 21;
int len = dstW - 7;
__m128i y_temp = __lsx_vreplgr2vr_w(ytemp);
YUVTORGB_SETUP_LSX
if( target == AV_PIX_FMT_BGR4_BYTE || target == AV_PIX_FMT_RGB4_BYTE
|| target == AV_PIX_FMT_BGR8 || target == AV_PIX_FMT_RGB8)
step = 1;
for (i = 0; i < len; i += 8) {
__m128i l_src, u_src, v_src;
__m128i y_ev, y_od, u_ev, u_od, v_ev, v_od, temp;
__m128i R_ev, R_od, G_ev, G_od, B_ev, B_od;
int n = i << 1;
y_ev = y_od = __lsx_vreplgr2vr_w(templ);
u_ev = u_od = v_ev = v_od = __lsx_vreplgr2vr_w(tempc);
for (j = 0; j < lumFilterSize; j++) {
temp = __lsx_vldrepl_h((lumFilter + j), 0);
l_src = __lsx_vldx(lumSrc[j], n);
y_ev = __lsx_vmaddwev_w_h(y_ev, l_src, temp);
y_od = __lsx_vmaddwod_w_h(y_od, l_src, temp);
}
for (j = 0; j < chrFilterSize; j++) {
temp = __lsx_vldrepl_h((chrFilter + j), 0);
DUP2_ARG2(__lsx_vldx, chrUSrc[j], n, chrVSrc[j], n,
u_src, v_src);
DUP2_ARG3(__lsx_vmaddwev_w_h, u_ev, u_src, temp, v_ev,
v_src, temp, u_ev, v_ev);
DUP2_ARG3(__lsx_vmaddwod_w_h, u_od, u_src, temp, v_od,
v_src, temp, u_od, v_od);
}
y_ev = __lsx_vsrai_w(y_ev, 10);
y_od = __lsx_vsrai_w(y_od, 10);
u_ev = __lsx_vsrai_w(u_ev, 10);
u_od = __lsx_vsrai_w(u_od, 10);
v_ev = __lsx_vsrai_w(v_ev, 10);
v_od = __lsx_vsrai_w(v_od, 10);
YUVTORGB_LSX(y_ev, u_ev, v_ev, R_ev, G_ev, B_ev, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
YUVTORGB_LSX(y_od, u_od, v_od, R_od, G_od, B_od, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
if (hasAlpha) {
__m128i a_src, a_ev, a_od;
a_ev = a_od = __lsx_vreplgr2vr_w(a_temp);
for (j = 0; j < lumFilterSize; j++) {
temp = __lsx_vldrepl_h(lumFilter + j, 0);
a_src = __lsx_vldx(alpSrc[j], n);
a_ev = __lsx_vmaddwev_w_h(a_ev, a_src, temp);
a_od = __lsx_vmaddwod_w_h(a_od, a_src, temp);
}
a_ev = __lsx_vsrai_w(a_ev, 19);
a_od = __lsx_vsrai_w(a_od, 19);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 0, 0);
WRITE_FULL_A_LSX(R_od, G_od, B_od, a_od, 0, 1);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 1, 2);
WRITE_FULL_A_LSX(R_od, G_od, B_od, a_od, 1, 3);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 2, 4);
WRITE_FULL_A_LSX(R_od, G_od, B_od, a_od, 2, 5);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 3, 6);
WRITE_FULL_A_LSX(R_od, G_od, B_od, a_od, 3, 7);
} else {
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 0, 0);
WRITE_FULL_LSX(R_od, G_od, B_od, 0, 1);
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 1, 2);
WRITE_FULL_LSX(R_od, G_od, B_od, 1, 3);
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 2, 4);
WRITE_FULL_LSX(R_od, G_od, B_od, 2, 5);
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 3, 6);
WRITE_FULL_LSX(R_od, G_od, B_od, 3, 7);
}
}
if (dstW - i >= 4) {
__m128i l_src, u_src, v_src;
__m128i y_ev, u_ev, v_ev, uv, temp;
__m128i R_ev, G_ev, B_ev;
int n = i << 1;
y_ev = __lsx_vreplgr2vr_w(templ);
u_ev = v_ev = __lsx_vreplgr2vr_w(tempc);
for (j = 0; j < lumFilterSize; j++) {
temp = __lsx_vldrepl_h((lumFilter + j), 0);
l_src = __lsx_vldx(lumSrc[j], n);
l_src = __lsx_vilvl_h(l_src, l_src);
y_ev = __lsx_vmaddwev_w_h(y_ev, l_src, temp);
}
for (j = 0; j < chrFilterSize; j++) {
temp = __lsx_vldrepl_h((chrFilter + j), 0);
DUP2_ARG2(__lsx_vldx, chrUSrc[j], n, chrVSrc[j], n, u_src, v_src);
uv = __lsx_vilvl_h(v_src, u_src);
u_ev = __lsx_vmaddwev_w_h(u_ev, uv, temp);
v_ev = __lsx_vmaddwod_w_h(v_ev, uv, temp);
}
y_ev = __lsx_vsrai_w(y_ev, 10);
u_ev = __lsx_vsrai_w(u_ev, 10);
v_ev = __lsx_vsrai_w(v_ev, 10);
YUVTORGB_LSX(y_ev, u_ev, v_ev, R_ev, G_ev, B_ev, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
if (hasAlpha) {
__m128i a_src, a_ev;
a_ev = __lsx_vreplgr2vr_w(a_temp);
for (j = 0; j < lumFilterSize; j++) {
temp = __lsx_vldrepl_h(lumFilter + j, 0);
a_src = __lsx_vldx(alpSrc[j], n);
a_src = __lsx_vilvl_h(a_src, a_src);
a_ev = __lsx_vmaddwev_w_h(a_ev, a_src, temp);
}
a_ev = __lsx_vsrai_w(a_ev, 19);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 0, 0);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 1, 1);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 2, 2);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 3, 3);
} else {
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 0, 0);
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 1, 1);
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 2, 2);
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 3, 3);
}
i += 4;
}
for (; i < dstW; i++) {
int Y = templ;
int V, U = V = tempc;
A = 0;
for (j = 0; j < lumFilterSize; j++) {
Y += lumSrc[j][i] * lumFilter[j];
}
for (j = 0; j < chrFilterSize; j++) {
U += chrUSrc[j][i] * chrFilter[j];
V += chrVSrc[j][i] * chrFilter[j];
}
Y >>= 10;
U >>= 10;
V >>= 10;
if (hasAlpha) {
A = 1 << 18;
for (j = 0; j < lumFilterSize; j++) {
A += alpSrc[j][i] * lumFilter[j];
}
A >>= 19;
if (A & 0x100)
A = av_clip_uint8(A);
}
Y -= y_offset;
Y *= y_coeff;
Y += ytemp;
R = (unsigned)Y + V * v2r_coe;
G = (unsigned)Y + V * v2g_coe + U * u2g_coe;
B = (unsigned)Y + U * u2b_coe;
yuv2rgb_write_full(c, dest, i, R, A, G, B, y, target, hasAlpha, err);
dest += step;
}
c->dither_error[0][i] = err[0];
c->dither_error[1][i] = err[1];
c->dither_error[2][i] = err[2];
}
static void
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
yuv2rgb_full_2_template_lsx(SwsInternal *c, const int16_t *buf[2],
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
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],
*abuf0 = hasAlpha ? abuf[0] : NULL,
*abuf1 = hasAlpha ? abuf[1] : NULL;
int yalpha1 = 4096 - yalpha;
int uvalpha1 = 4096 - uvalpha;
int uvtemp = 128 << 19;
int atemp = 1 << 18;
int err[4] = {0};
int ytemp = 1 << 21;
int len = dstW - 7;
int i, R, G, B, A;
int step = (target == AV_PIX_FMT_RGB24 ||
target == AV_PIX_FMT_BGR24) ? 3 : 4;
__m128i v_uvalpha1 = __lsx_vreplgr2vr_w(uvalpha1);
__m128i v_yalpha1 = __lsx_vreplgr2vr_w(yalpha1);
__m128i v_uvalpha = __lsx_vreplgr2vr_w(uvalpha);
__m128i v_yalpha = __lsx_vreplgr2vr_w(yalpha);
__m128i uv = __lsx_vreplgr2vr_w(uvtemp);
__m128i a_bias = __lsx_vreplgr2vr_w(atemp);
__m128i y_temp = __lsx_vreplgr2vr_w(ytemp);
YUVTORGB_SETUP_LSX
av_assert2(yalpha <= 4096U);
av_assert2(uvalpha <= 4096U);
if( target == AV_PIX_FMT_BGR4_BYTE || target == AV_PIX_FMT_RGB4_BYTE
|| target == AV_PIX_FMT_BGR8 || target == AV_PIX_FMT_RGB8)
step = 1;
for (i = 0; i < len; i += 8) {
__m128i b0, b1, ub0, ub1, vb0, vb1;
__m128i y0_l, y0_h, y1_l, y1_h, u0_l, u0_h;
__m128i v0_l, v0_h, u1_l, u1_h, v1_l, v1_h;
__m128i y_l, y_h, v_l, v_h, u_l, u_h;
__m128i R_l, R_h, G_l, G_h, B_l, B_h;
int n = i << 1;
DUP4_ARG2(__lsx_vldx, buf0, n, buf1, n, ubuf0,
n, ubuf1, n, b0, b1, ub0, ub1);
DUP2_ARG2(__lsx_vldx, vbuf0, n, vbuf1, n, vb0 , vb1);
DUP2_ARG2(__lsx_vsllwil_w_h, b0, 0, b1, 0, y0_l, y1_l);
DUP4_ARG2(__lsx_vsllwil_w_h, ub0, 0, ub1, 0, vb0, 0, vb1, 0,
u0_l, u1_l, v0_l, v1_l);
DUP2_ARG1(__lsx_vexth_w_h, b0, b1, y0_h, y1_h);
DUP4_ARG1(__lsx_vexth_w_h, ub0, ub1, vb0, vb1,
u0_h, u1_h, v0_h, v1_h);
y0_l = __lsx_vmul_w(y0_l, v_yalpha1);
y0_h = __lsx_vmul_w(y0_h, v_yalpha1);
u0_l = __lsx_vmul_w(u0_l, v_uvalpha1);
u0_h = __lsx_vmul_w(u0_h, v_uvalpha1);
v0_l = __lsx_vmul_w(v0_l, v_uvalpha1);
v0_h = __lsx_vmul_w(v0_h, v_uvalpha1);
y_l = __lsx_vmadd_w(y0_l, v_yalpha, y1_l);
y_h = __lsx_vmadd_w(y0_h, v_yalpha, y1_h);
u_l = __lsx_vmadd_w(u0_l, v_uvalpha, u1_l);
u_h = __lsx_vmadd_w(u0_h, v_uvalpha, u1_h);
v_l = __lsx_vmadd_w(v0_l, v_uvalpha, v1_l);
v_h = __lsx_vmadd_w(v0_h, v_uvalpha, v1_h);
u_l = __lsx_vsub_w(u_l, uv);
u_h = __lsx_vsub_w(u_h, uv);
v_l = __lsx_vsub_w(v_l, uv);
v_h = __lsx_vsub_w(v_h, uv);
y_l = __lsx_vsrai_w(y_l, 10);
y_h = __lsx_vsrai_w(y_h, 10);
u_l = __lsx_vsrai_w(u_l, 10);
u_h = __lsx_vsrai_w(u_h, 10);
v_l = __lsx_vsrai_w(v_l, 10);
v_h = __lsx_vsrai_w(v_h, 10);
YUVTORGB_LSX(y_l, u_l, v_l, R_l, G_l, B_l, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
YUVTORGB_LSX(y_h, u_h, v_h, R_h, G_h, B_h, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
if (hasAlpha) {
__m128i a0, a1, a0_l, a0_h;
__m128i a_l, a_h, a1_l, a1_h;
DUP2_ARG2(__lsx_vldx, abuf0, n, abuf1, n, a0, a1);
DUP2_ARG2(__lsx_vsllwil_w_h, a0, 0, a1, 0, a0_l, a1_l);
DUP2_ARG1(__lsx_vexth_w_h, a0, a1, a0_h, a1_h);
a_l = __lsx_vmadd_w(a_bias, a0_l, v_yalpha1);
a_h = __lsx_vmadd_w(a_bias, a0_h, v_yalpha1);
a_l = __lsx_vmadd_w(a_l, v_yalpha, a1_l);
a_h = __lsx_vmadd_w(a_h, v_yalpha, a1_h);
a_l = __lsx_vsrai_w(a_l, 19);
a_h = __lsx_vsrai_w(a_h, 19);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 0, 0);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 1, 1);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 2, 2);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 3, 3);
WRITE_FULL_A_LSX(R_h, G_h, B_h, a_h, 0, 4);
WRITE_FULL_A_LSX(R_h, G_h, B_h, a_h, 1, 5);
WRITE_FULL_A_LSX(R_h, G_h, B_h, a_h, 2, 6);
WRITE_FULL_A_LSX(R_h, G_h, B_h, a_h, 3, 7);
} else {
WRITE_FULL_LSX(R_l, G_l, B_l, 0, 0);
WRITE_FULL_LSX(R_l, G_l, B_l, 1, 1);
WRITE_FULL_LSX(R_l, G_l, B_l, 2, 2);
WRITE_FULL_LSX(R_l, G_l, B_l, 3, 3);
WRITE_FULL_LSX(R_h, G_h, B_h, 0, 4);
WRITE_FULL_LSX(R_h, G_h, B_h, 1, 5);
WRITE_FULL_LSX(R_h, G_h, B_h, 2, 6);
WRITE_FULL_LSX(R_h, G_h, B_h, 3, 7);
}
}
if (dstW - i >= 4) {
__m128i b0, b1, ub0, ub1, vb0, vb1;
__m128i y0_l, y1_l, u0_l;
__m128i v0_l, u1_l, v1_l;
__m128i y_l, u_l, v_l;
__m128i R_l, G_l, B_l;
int n = i << 1;
DUP4_ARG2(__lsx_vldx, buf0, n, buf1, n, ubuf0, n,
ubuf1, n, b0, b1, ub0, ub1);
DUP2_ARG2(__lsx_vldx, vbuf0, n, vbuf1, n, vb0, vb1);
DUP2_ARG2(__lsx_vsllwil_w_h, b0, 0, b1, 0, y0_l, y1_l);
DUP4_ARG2(__lsx_vsllwil_w_h, ub0, 0, ub1, 0, vb0, 0, vb1, 0,
u0_l, u1_l, v0_l, v1_l);
y0_l = __lsx_vmul_w(y0_l, v_yalpha1);
u0_l = __lsx_vmul_w(u0_l, v_uvalpha1);
v0_l = __lsx_vmul_w(v0_l, v_uvalpha1);
y_l = __lsx_vmadd_w(y0_l, v_yalpha, y1_l);
u_l = __lsx_vmadd_w(u0_l, v_uvalpha, u1_l);
v_l = __lsx_vmadd_w(v0_l, v_uvalpha, v1_l);
u_l = __lsx_vsub_w(u_l, uv);
v_l = __lsx_vsub_w(v_l, uv);
y_l = __lsx_vsrai_w(y_l, 10);
u_l = __lsx_vsrai_w(u_l, 10);
v_l = __lsx_vsrai_w(v_l, 10);
YUVTORGB_LSX(y_l, u_l, v_l, R_l, G_l, B_l, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
if (hasAlpha) {
__m128i a0, a1, a0_l;
__m128i a_l, a1_l;
DUP2_ARG2(__lsx_vldx, abuf0, n, abuf1, n, a0, a1);
DUP2_ARG2(__lsx_vsllwil_w_h, a0, 0, a1, 0, a0_l, a1_l);
a_l = __lsx_vmadd_w(a_bias, a0_l, v_yalpha1);
a_l = __lsx_vmadd_w(a_l, v_yalpha, a1_l);
a_l = __lsx_vsrai_w(a_l, 19);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 0, 0);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 1, 1);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 2, 2);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 3, 3);
} else {
WRITE_FULL_LSX(R_l, G_l, B_l, 0, 0);
WRITE_FULL_LSX(R_l, G_l, B_l, 1, 1);
WRITE_FULL_LSX(R_l, G_l, B_l, 2, 2);
WRITE_FULL_LSX(R_l, G_l, B_l, 3, 3);
}
i += 4;
}
for (; i < dstW; i++){
int Y = ( buf0[i] * yalpha1 + buf1[i] * yalpha ) >> 10;
int U = (ubuf0[i] * uvalpha1 + ubuf1[i] * uvalpha- uvtemp) >> 10;
int V = (vbuf0[i] * uvalpha1 + vbuf1[i] * uvalpha- uvtemp) >> 10;
A = 0;
if (hasAlpha){
A = (abuf0[i] * yalpha1 + abuf1[i] * yalpha + atemp) >> 19;
if (A & 0x100)
A = av_clip_uint8(A);
}
Y -= y_offset;
Y *= y_coeff;
Y += ytemp;
R = (unsigned)Y + V * v2r_coe;
G = (unsigned)Y + V * v2g_coe + U * u2g_coe;
B = (unsigned)Y + U * u2b_coe;
yuv2rgb_write_full(c, dest, i, R, A, G, B, y, target, hasAlpha, err);
dest += step;
}
c->dither_error[0][i] = err[0];
c->dither_error[1][i] = err[1];
c->dither_error[2][i] = err[2];
}
static void
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
yuv2rgb_full_1_template_lsx(SwsInternal *c, const int16_t *buf0,
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
const int16_t *ubuf[2], const int16_t *vbuf[2],
const int16_t *abuf0, uint8_t *dest, int dstW,
int uvalpha, int y, enum AVPixelFormat target,
int hasAlpha)
{
const int16_t *ubuf0 = ubuf[0], *vbuf0 = vbuf[0];
int i, B, G, R, A;
int step = (target == AV_PIX_FMT_RGB24 || target == AV_PIX_FMT_BGR24) ? 3 : 4;
int err[4] = {0};
int ytemp = 1 << 21;
int bias_int = 64;
int len = dstW - 7;
__m128i y_temp = __lsx_vreplgr2vr_w(ytemp);
YUVTORGB_SETUP_LSX
if( target == AV_PIX_FMT_BGR4_BYTE || target == AV_PIX_FMT_RGB4_BYTE
|| target == AV_PIX_FMT_BGR8 || target == AV_PIX_FMT_RGB8)
step = 1;
if (uvalpha < 2048) {
int uvtemp = 128 << 7;
__m128i uv = __lsx_vreplgr2vr_w(uvtemp);
__m128i bias = __lsx_vreplgr2vr_w(bias_int);
for (i = 0; i < len; i += 8) {
__m128i b, ub, vb, ub_l, ub_h, vb_l, vb_h;
__m128i y_l, y_h, u_l, u_h, v_l, v_h;
__m128i R_l, R_h, G_l, G_h, B_l, B_h;
int n = i << 1;
DUP2_ARG2(__lsx_vldx, buf0, n, ubuf0, n, b, ub);
vb = __lsx_vldx(vbuf0, n);
y_l = __lsx_vsllwil_w_h(b, 2);
y_h = __lsx_vexth_w_h(b);
DUP2_ARG2(__lsx_vsllwil_w_h, ub, 0, vb, 0, ub_l, vb_l);
DUP2_ARG1(__lsx_vexth_w_h, ub, vb, ub_h, vb_h);
y_h = __lsx_vslli_w(y_h, 2);
u_l = __lsx_vsub_w(ub_l, uv);
u_h = __lsx_vsub_w(ub_h, uv);
v_l = __lsx_vsub_w(vb_l, uv);
v_h = __lsx_vsub_w(vb_h, uv);
u_l = __lsx_vslli_w(u_l, 2);
u_h = __lsx_vslli_w(u_h, 2);
v_l = __lsx_vslli_w(v_l, 2);
v_h = __lsx_vslli_w(v_h, 2);
YUVTORGB_LSX(y_l, u_l, v_l, R_l, G_l, B_l, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
YUVTORGB_LSX(y_h, u_h, v_h, R_h, G_h, B_h, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
if(hasAlpha) {
__m128i a_src;
__m128i a_l, a_h;
a_src = __lsx_vld(abuf0 + i, 0);
a_l = __lsx_vsllwil_w_h(a_src, 0);
a_h = __lsx_vexth_w_h(a_src);
a_l = __lsx_vadd_w(a_l, bias);
a_h = __lsx_vadd_w(a_h, bias);
a_l = __lsx_vsrai_w(a_l, 7);
a_h = __lsx_vsrai_w(a_h, 7);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 0, 0);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 1, 1);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 2, 2);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 3, 3);
WRITE_FULL_A_LSX(R_h, G_h, B_h, a_h, 0, 4);
WRITE_FULL_A_LSX(R_h, G_h, B_h, a_h, 1, 5);
WRITE_FULL_A_LSX(R_h, G_h, B_h, a_h, 2, 6);
WRITE_FULL_A_LSX(R_h, G_h, B_h, a_h, 3, 7);
} else {
WRITE_FULL_LSX(R_l, G_l, B_l, 0, 0);
WRITE_FULL_LSX(R_l, G_l, B_l, 1, 1);
WRITE_FULL_LSX(R_l, G_l, B_l, 2, 2);
WRITE_FULL_LSX(R_l, G_l, B_l, 3, 3);
WRITE_FULL_LSX(R_h, G_h, B_h, 0, 4);
WRITE_FULL_LSX(R_h, G_h, B_h, 1, 5);
WRITE_FULL_LSX(R_h, G_h, B_h, 2, 6);
WRITE_FULL_LSX(R_h, G_h, B_h, 3, 7);
}
}
if (dstW - i >= 4) {
__m128i b, ub, vb, ub_l, vb_l;
__m128i y_l, u_l, v_l;
__m128i R_l, G_l, B_l;
int n = i << 1;
DUP2_ARG2(__lsx_vldx, buf0, n, ubuf0, n, b, ub);
vb = __lsx_vldx(vbuf0, n);
y_l = __lsx_vsllwil_w_h(b, 0);
DUP2_ARG2(__lsx_vsllwil_w_h, ub, 0, vb, 0, ub_l, vb_l);
y_l = __lsx_vslli_w(y_l, 2);
u_l = __lsx_vsub_w(ub_l, uv);
v_l = __lsx_vsub_w(vb_l, uv);
u_l = __lsx_vslli_w(u_l, 2);
v_l = __lsx_vslli_w(v_l, 2);
YUVTORGB_LSX(y_l, u_l, v_l, R_l, G_l, B_l, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
if(hasAlpha) {
__m128i a_src, a_l;
a_src = __lsx_vldx(abuf0, n);
a_src = __lsx_vsllwil_w_h(a_src, 0);
a_l = __lsx_vadd_w(bias, a_src);
a_l = __lsx_vsrai_w(a_l, 7);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 0, 0);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 1, 1);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 2, 2);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 3, 3);
} else {
WRITE_FULL_LSX(R_l, G_l, B_l, 0, 0);
WRITE_FULL_LSX(R_l, G_l, B_l, 1, 1);
WRITE_FULL_LSX(R_l, G_l, B_l, 2, 2);
WRITE_FULL_LSX(R_l, G_l, B_l, 3, 3);
}
i += 4;
}
for (; i < dstW; i++) {
int Y = buf0[i] << 2;
int U = (ubuf0[i] - uvtemp) << 2;
int V = (vbuf0[i] - uvtemp) << 2;
A = 0;
if(hasAlpha) {
A = (abuf0[i] + 64) >> 7;
if (A & 0x100)
A = av_clip_uint8(A);
}
Y -= y_offset;
Y *= y_coeff;
Y += ytemp;
R = (unsigned)Y + V * v2r_coe;
G = (unsigned)Y + V * v2g_coe + U * u2g_coe;
B = (unsigned)Y + U * u2b_coe;
yuv2rgb_write_full(c, dest, i, R, A, G, B, y, target, hasAlpha, err);
dest += step;
}
} else {
const int16_t *ubuf1 = ubuf[1], *vbuf1 = vbuf[1];
int uvtemp = 128 << 8;
__m128i uv = __lsx_vreplgr2vr_w(uvtemp);
__m128i zero = __lsx_vldi(0);
__m128i bias = __lsx_vreplgr2vr_h(bias_int);
for (i = 0; i < len; i += 8) {
__m128i b, ub0, ub1, vb0, vb1;
__m128i y_ev, y_od, u_ev, u_od, v_ev, v_od;
__m128i R_ev, R_od, G_ev, G_od, B_ev, B_od;
int n = i << 1;
DUP4_ARG2(__lsx_vldx, buf0, n, ubuf0, n, vbuf0, n,
ubuf1, n, b, ub0, vb0, ub1);
vb1 = __lsx_vldx(vbuf, n);
y_ev = __lsx_vaddwev_w_h(b, zero);
y_od = __lsx_vaddwod_w_h(b, zero);
DUP2_ARG2(__lsx_vaddwev_w_h, ub0, vb0, ub1, vb1, u_ev, v_ev);
DUP2_ARG2(__lsx_vaddwod_w_h, ub0, vb0, ub1, vb1, u_od, v_od);
DUP2_ARG2(__lsx_vslli_w, y_ev, 2, y_od, 2, y_ev, y_od);
DUP4_ARG2(__lsx_vsub_w, u_ev, uv, u_od, uv, v_ev, uv, v_od, uv,
u_ev, u_od, v_ev, v_od);
DUP4_ARG2(__lsx_vslli_w, u_ev, 1, u_od, 1, v_ev, 1, v_od, 1,
u_ev, u_od, v_ev, v_od);
YUVTORGB_LSX(y_ev, u_ev, v_ev, R_ev, G_ev, B_ev, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
YUVTORGB_LSX(y_od, u_od, v_od, R_od, G_od, B_od, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
if(hasAlpha) {
__m128i a_src;
__m128i a_ev, a_od;
a_src = __lsx_vld(abuf0 + i, 0);
a_ev = __lsx_vaddwev_w_h(bias, a_src);
a_od = __lsx_vaddwod_w_h(bias, a_src);
a_ev = __lsx_vsrai_w(a_ev, 7);
a_od = __lsx_vsrai_w(a_od, 7);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 0, 0);
WRITE_FULL_A_LSX(R_od, G_od, B_od, a_od, 0, 1);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 1, 2);
WRITE_FULL_A_LSX(R_od, G_od, B_od, a_od, 1, 3);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 2, 4);
WRITE_FULL_A_LSX(R_od, G_od, B_od, a_od, 2, 5);
WRITE_FULL_A_LSX(R_ev, G_ev, B_ev, a_ev, 3, 6);
WRITE_FULL_A_LSX(R_od, G_od, B_od, a_od, 3, 7);
} else {
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 0, 0);
WRITE_FULL_LSX(R_od, G_od, B_od, 0, 1);
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 1, 2);
WRITE_FULL_LSX(R_od, G_od, B_od, 1, 3);
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 2, 4);
WRITE_FULL_LSX(R_od, G_od, B_od, 2, 5);
WRITE_FULL_LSX(R_ev, G_ev, B_ev, 3, 6);
WRITE_FULL_LSX(R_od, G_od, B_od, 3, 7);
}
}
if (dstW - i >= 4) {
__m128i b, ub0, ub1, vb0, vb1;
__m128i y_l, u_l, v_l;
__m128i R_l, G_l, B_l;
int n = i << 1;
DUP4_ARG2(__lsx_vldx, buf0, n, ubuf0, n, vbuf0, n,
ubuf1, n, b, ub0, vb0, ub1);
vb1 = __lsx_vldx(vbuf1, n);
y_l = __lsx_vsllwil_w_h(b, 0);
y_l = __lsx_vslli_w(y_l, 2);
DUP4_ARG2(__lsx_vsllwil_w_h, ub0, 0, vb0, 0, ub1, 0, vb1, 0,
ub0, vb0, ub1, vb1);
DUP2_ARG2(__lsx_vadd_w, ub0, ub1, vb0, vb1, u_l, v_l);
u_l = __lsx_vsub_w(u_l, uv);
v_l = __lsx_vsub_w(v_l, uv);
u_l = __lsx_vslli_w(u_l, 1);
v_l = __lsx_vslli_w(v_l, 1);
YUVTORGB_LSX(y_l, u_l, v_l, R_l, G_l, B_l, offset, coeff,
y_temp, v2r, v2g, u2g, u2b);
if(hasAlpha) {
__m128i a_src;
__m128i a_l;
a_src = __lsx_vld(abuf0 + i, 0);
a_src = __lsx_vilvl_h(a_src, a_src);
a_l = __lsx_vaddwev_w_h(bias, a_l);
a_l = __lsx_vsrai_w(a_l, 7);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 0, 0);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 1, 1);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 2, 2);
WRITE_FULL_A_LSX(R_l, G_l, B_l, a_l, 3, 3);
} else {
WRITE_FULL_LSX(R_l, G_l, B_l, 0, 0);
WRITE_FULL_LSX(R_l, G_l, B_l, 1, 1);
WRITE_FULL_LSX(R_l, G_l, B_l, 2, 2);
WRITE_FULL_LSX(R_l, G_l, B_l, 3, 3);
}
i += 4;
}
for (; i < dstW; i++) {
int Y = buf0[i] << 2;
int U = (ubuf0[i] + ubuf1[i] - uvtemp) << 1;
int V = (vbuf0[i] + vbuf1[i] - uvtemp) << 1;
A = 0;
if(hasAlpha) {
A = (abuf0[i] + 64) >> 7;
if (A & 0x100)
A = av_clip_uint8(A);
}
Y -= y_offset;
Y *= y_coeff;
Y += ytemp;
R = (unsigned)Y + V * v2r_coe;
G = (unsigned)Y + V * v2g_coe + U * u2g_coe;
B = (unsigned)Y + U * u2b_coe;
yuv2rgb_write_full(c, dest, i, R, A, G, B, y, target, hasAlpha, err);
dest += step;
}
}
c->dither_error[0][i] = err[0];
c->dither_error[1][i] = err[1];
c->dither_error[2][i] = err[2];
}
#if CONFIG_SMALL
YUV2RGBWRAPPER(yuv2, rgb_full, bgra32_full, AV_PIX_FMT_BGRA,
CONFIG_SWSCALE_ALPHA && c->needAlpha)
YUV2RGBWRAPPER(yuv2, rgb_full, abgr32_full, AV_PIX_FMT_ABGR,
CONFIG_SWSCALE_ALPHA && c->needAlpha)
YUV2RGBWRAPPER(yuv2, rgb_full, rgba32_full, AV_PIX_FMT_RGBA,
CONFIG_SWSCALE_ALPHA && c->needAlpha)
YUV2RGBWRAPPER(yuv2, rgb_full, argb32_full, AV_PIX_FMT_ARGB,
CONFIG_SWSCALE_ALPHA && c->needAlpha)
#else
#if CONFIG_SWSCALE_ALPHA
YUV2RGBWRAPPER(yuv2, rgb_full, bgra32_full, AV_PIX_FMT_BGRA, 1)
YUV2RGBWRAPPER(yuv2, rgb_full, abgr32_full, AV_PIX_FMT_ABGR, 1)
YUV2RGBWRAPPER(yuv2, rgb_full, rgba32_full, AV_PIX_FMT_RGBA, 1)
YUV2RGBWRAPPER(yuv2, rgb_full, argb32_full, AV_PIX_FMT_ARGB, 1)
#endif
YUV2RGBWRAPPER(yuv2, rgb_full, bgrx32_full, AV_PIX_FMT_BGRA, 0)
YUV2RGBWRAPPER(yuv2, rgb_full, xbgr32_full, AV_PIX_FMT_ABGR, 0)
YUV2RGBWRAPPER(yuv2, rgb_full, rgbx32_full, AV_PIX_FMT_RGBA, 0)
YUV2RGBWRAPPER(yuv2, rgb_full, xrgb32_full, AV_PIX_FMT_ARGB, 0)
#endif
YUV2RGBWRAPPER(yuv2, rgb_full, bgr24_full, AV_PIX_FMT_BGR24, 0)
YUV2RGBWRAPPER(yuv2, rgb_full, rgb24_full, AV_PIX_FMT_RGB24, 0)
YUV2RGBWRAPPER(yuv2, rgb_full, bgr4_byte_full, AV_PIX_FMT_BGR4_BYTE, 0)
YUV2RGBWRAPPER(yuv2, rgb_full, rgb4_byte_full, AV_PIX_FMT_RGB4_BYTE, 0)
YUV2RGBWRAPPER(yuv2, rgb_full, bgr8_full, AV_PIX_FMT_BGR8, 0)
YUV2RGBWRAPPER(yuv2, rgb_full, rgb8_full, AV_PIX_FMT_RGB8, 0)
swscale: rename SwsContext to SwsInternal And preserve the public SwsContext as separate name. The motivation here is that I want to turn SwsContext into a public struct, while keeping the internal implementation hidden. Additionally, I also want to be able to use multiple internal implementations, e.g. for GPU devices. This commit does not include any functional changes. For the most part, it is a simple rename. The only complications arise from the public facing API functions, which preserve their current type (and hence require an additional unwrapping step internally), and the checkasm test framework, which directly accesses SwsInternal. For consistency, the affected functions that need to maintain a distionction have generally been changed to refer to the SwsContext as *sws, and the SwsInternal as *c. In an upcoming commit, I will provide a backing definition for the public SwsContext, and update `sws_internal()` to dereference the internal struct instead of merely casting it. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
av_cold void ff_sws_init_output_lsx(SwsInternal *c,
swscale: [LA] Optimize yuv2plane1_8_c. Reviewed-by: colleague of Shiyou Yin Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
10 months ago
yuv2planar1_fn *yuv2plane1,
yuv2planarX_fn *yuv2planeX,
yuv2interleavedX_fn *yuv2nv12cX,
yuv2packed1_fn *yuv2packed1,
yuv2packed2_fn *yuv2packed2,
yuv2packedX_fn *yuv2packedX,
yuv2anyX_fn *yuv2anyX)
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
{
swscale/internal: group user-facing options together This is a preliminary step to separating these into a new struct. This commit contains no functional changes, it is a pure search-and-replace. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
enum AVPixelFormat dstFormat = c->opts.dst_format;
swscale: [LA] Optimize yuv2plane1_8_c. Reviewed-by: colleague of Shiyou Yin Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
10 months ago
/* Add initialization once optimized */
if (isSemiPlanarYUV(dstFormat) && isDataInHighBits(dstFormat)) {
} else if (is16BPS(dstFormat)) {
} else if (isNBPS(dstFormat)) {
} else if (dstFormat == AV_PIX_FMT_GRAYF32BE) {
} else if (dstFormat == AV_PIX_FMT_GRAYF32LE) {
} else {
*yuv2plane1 = yuv2plane1_8_lsx;
*yuv2planeX = yuv2planeX_8_lsx;
}
swscale/internal: group user-facing options together This is a preliminary step to separating these into a new struct. This commit contains no functional changes, it is a pure search-and-replace. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
if(c->opts.flags & SWS_FULL_CHR_H_INT) {
switch (c->opts.dst_format) {
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
case AV_PIX_FMT_RGBA:
#if CONFIG_SMALL
c->yuv2packedX = yuv2rgba32_full_X_lsx;
c->yuv2packed2 = yuv2rgba32_full_2_lsx;
c->yuv2packed1 = yuv2rgba32_full_1_lsx;
#else
#if CONFIG_SWSCALE_ALPHA
if (c->needAlpha) {
c->yuv2packedX = yuv2rgba32_full_X_lsx;
c->yuv2packed2 = yuv2rgba32_full_2_lsx;
c->yuv2packed1 = yuv2rgba32_full_1_lsx;
} else
#endif /* CONFIG_SWSCALE_ALPHA */
{
c->yuv2packedX = yuv2rgbx32_full_X_lsx;
c->yuv2packed2 = yuv2rgbx32_full_2_lsx;
c->yuv2packed1 = yuv2rgbx32_full_1_lsx;
}
#endif /* !CONFIG_SMALL */
break;
case AV_PIX_FMT_ARGB:
#if CONFIG_SMALL
c->yuv2packedX = yuv2argb32_full_X_lsx;
c->yuv2packed2 = yuv2argb32_full_2_lsx;
c->yuv2packed1 = yuv2argb32_full_1_lsx;
#else
#if CONFIG_SWSCALE_ALPHA
if (c->needAlpha) {
c->yuv2packedX = yuv2argb32_full_X_lsx;
c->yuv2packed2 = yuv2argb32_full_2_lsx;
c->yuv2packed1 = yuv2argb32_full_1_lsx;
} else
#endif /* CONFIG_SWSCALE_ALPHA */
{
c->yuv2packedX = yuv2xrgb32_full_X_lsx;
c->yuv2packed2 = yuv2xrgb32_full_2_lsx;
c->yuv2packed1 = yuv2xrgb32_full_1_lsx;
}
#endif /* !CONFIG_SMALL */
break;
case AV_PIX_FMT_BGRA:
#if CONFIG_SMALL
c->yuv2packedX = yuv2bgra32_full_X_lsx;
c->yuv2packed2 = yuv2bgra32_full_2_lsx;
c->yuv2packed1 = yuv2bgra32_full_1_lsx;
#else
#if CONFIG_SWSCALE_ALPHA
if (c->needAlpha) {
c->yuv2packedX = yuv2bgra32_full_X_lsx;
c->yuv2packed2 = yuv2bgra32_full_2_lsx;
c->yuv2packed1 = yuv2bgra32_full_1_lsx;
} else
#endif /* CONFIG_SWSCALE_ALPHA */
{
c->yuv2packedX = yuv2bgrx32_full_X_lsx;
c->yuv2packed2 = yuv2bgrx32_full_2_lsx;
c->yuv2packed1 = yuv2bgrx32_full_1_lsx;
}
#endif /* !CONFIG_SMALL */
break;
case AV_PIX_FMT_ABGR:
#if CONFIG_SMALL
c->yuv2packedX = yuv2abgr32_full_X_lsx;
c->yuv2packed2 = yuv2abgr32_full_2_lsx;
c->yuv2packed1 = yuv2abgr32_full_1_lsx;
#else
#if CONFIG_SWSCALE_ALPHA
if (c->needAlpha) {
c->yuv2packedX = yuv2abgr32_full_X_lsx;
c->yuv2packed2 = yuv2abgr32_full_2_lsx;
c->yuv2packed1 = yuv2abgr32_full_1_lsx;
} else
#endif /* CONFIG_SWSCALE_ALPHA */
{
c->yuv2packedX = yuv2xbgr32_full_X_lsx;
c->yuv2packed2 = yuv2xbgr32_full_2_lsx;
c->yuv2packed1 = yuv2xbgr32_full_1_lsx;
}
#endif /* !CONFIG_SMALL */
break;
case AV_PIX_FMT_RGB24:
c->yuv2packedX = yuv2rgb24_full_X_lsx;
c->yuv2packed2 = yuv2rgb24_full_2_lsx;
c->yuv2packed1 = yuv2rgb24_full_1_lsx;
break;
case AV_PIX_FMT_BGR24:
c->yuv2packedX = yuv2bgr24_full_X_lsx;
c->yuv2packed2 = yuv2bgr24_full_2_lsx;
c->yuv2packed1 = yuv2bgr24_full_1_lsx;
break;
case AV_PIX_FMT_BGR4_BYTE:
c->yuv2packedX = yuv2bgr4_byte_full_X_lsx;
c->yuv2packed2 = yuv2bgr4_byte_full_2_lsx;
c->yuv2packed1 = yuv2bgr4_byte_full_1_lsx;
break;
case AV_PIX_FMT_RGB4_BYTE:
c->yuv2packedX = yuv2rgb4_byte_full_X_lsx;
c->yuv2packed2 = yuv2rgb4_byte_full_2_lsx;
c->yuv2packed1 = yuv2rgb4_byte_full_1_lsx;
break;
case AV_PIX_FMT_BGR8:
c->yuv2packedX = yuv2bgr8_full_X_lsx;
c->yuv2packed2 = yuv2bgr8_full_2_lsx;
c->yuv2packed1 = yuv2bgr8_full_1_lsx;
break;
case AV_PIX_FMT_RGB8:
c->yuv2packedX = yuv2rgb8_full_X_lsx;
c->yuv2packed2 = yuv2rgb8_full_2_lsx;
c->yuv2packed1 = yuv2rgb8_full_1_lsx;
break;
}
} else {
swscale/internal: group user-facing options together This is a preliminary step to separating these into a new struct. This commit contains no functional changes, it is a pure search-and-replace. Sponsored-by: Sovereign Tech Fund Signed-off-by: Niklas Haas <git@haasn.dev>
3 months ago
switch (c->opts.dst_format) {
swscale/la: Optimize the functions of the swscale series with lsx. ./configure --disable-lasx ffmpeg -i ~/media/1_h264_1080p_30fps_3Mbps.mp4 -f rawvideo -s 640x480 -pix_fmt bgra -y /dev/null -an before: 91fps after: 160fps Reviewed-by: Shiyou Yin <yinshiyou-hf@loongson.cn> Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
2 years ago
case AV_PIX_FMT_RGB32:
case AV_PIX_FMT_BGR32:
#if CONFIG_SMALL
#else
#if CONFIG_SWSCALE_ALPHA
if (c->needAlpha) {
} else
#endif /* CONFIG_SWSCALE_ALPHA */
{
c->yuv2packed1 = yuv2rgbx32_1_lsx;
c->yuv2packed2 = yuv2rgbx32_2_lsx;
c->yuv2packedX = yuv2rgbx32_X_lsx;
}
#endif /* !CONFIG_SMALL */
break;
case AV_PIX_FMT_RGB32_1:
case AV_PIX_FMT_BGR32_1:
#if CONFIG_SMALL
#else
#if CONFIG_SWSCALE_ALPHA
if (c->needAlpha) {
} else
#endif /* CONFIG_SWSCALE_ALPHA */
{
c->yuv2packed1 = yuv2rgbx32_1_1_lsx;
c->yuv2packed2 = yuv2rgbx32_1_2_lsx;
c->yuv2packedX = yuv2rgbx32_1_X_lsx;
}
#endif /* !CONFIG_SMALL */
break;
case AV_PIX_FMT_RGB24:
c->yuv2packed1 = yuv2rgb24_1_lsx;
c->yuv2packed2 = yuv2rgb24_2_lsx;
c->yuv2packedX = yuv2rgb24_X_lsx;
break;
case AV_PIX_FMT_BGR24:
c->yuv2packed1 = yuv2bgr24_1_lsx;
c->yuv2packed2 = yuv2bgr24_2_lsx;
c->yuv2packedX = yuv2bgr24_X_lsx;
break;
case AV_PIX_FMT_RGB565LE:
case AV_PIX_FMT_RGB565BE:
case AV_PIX_FMT_BGR565LE:
case AV_PIX_FMT_BGR565BE:
c->yuv2packed1 = yuv2rgb16_1_lsx;
c->yuv2packed2 = yuv2rgb16_2_lsx;
c->yuv2packedX = yuv2rgb16_X_lsx;
break;
case AV_PIX_FMT_RGB555LE:
case AV_PIX_FMT_RGB555BE:
case AV_PIX_FMT_BGR555LE:
case AV_PIX_FMT_BGR555BE:
c->yuv2packed1 = yuv2rgb15_1_lsx;
c->yuv2packed2 = yuv2rgb15_2_lsx;
c->yuv2packedX = yuv2rgb15_X_lsx;
break;
case AV_PIX_FMT_RGB444LE:
case AV_PIX_FMT_RGB444BE:
case AV_PIX_FMT_BGR444LE:
case AV_PIX_FMT_BGR444BE:
c->yuv2packed1 = yuv2rgb12_1_lsx;
c->yuv2packed2 = yuv2rgb12_2_lsx;
c->yuv2packedX = yuv2rgb12_X_lsx;
break;
case AV_PIX_FMT_RGB8:
case AV_PIX_FMT_BGR8:
c->yuv2packed1 = yuv2rgb8_1_lsx;
c->yuv2packed2 = yuv2rgb8_2_lsx;
c->yuv2packedX = yuv2rgb8_X_lsx;
break;
case AV_PIX_FMT_RGB4:
case AV_PIX_FMT_BGR4:
c->yuv2packed1 = yuv2rgb4_1_lsx;
c->yuv2packed2 = yuv2rgb4_2_lsx;
c->yuv2packedX = yuv2rgb4_X_lsx;
break;
case AV_PIX_FMT_RGB4_BYTE:
case AV_PIX_FMT_BGR4_BYTE:
c->yuv2packed1 = yuv2rgb4b_1_lsx;
c->yuv2packed2 = yuv2rgb4b_2_lsx;
c->yuv2packedX = yuv2rgb4b_X_lsx;
break;
}
}
}