Commit 5740dc27 authored by Paul B Mahol's avatar Paul B Mahol

avfilter/vf_w3fdif: add x86 SIMD

Signed-off-by: 's avatarPaul B Mahol <onemda@gmail.com>
parent 0281ef90
...@@ -29,6 +29,7 @@ ...@@ -29,6 +29,7 @@
#include "formats.h" #include "formats.h"
#include "internal.h" #include "internal.h"
#include "video.h" #include "video.h"
#include "w3fdif.h"
typedef struct W3FDIFContext { typedef struct W3FDIFContext {
const AVClass *class; const AVClass *class;
...@@ -42,6 +43,8 @@ typedef struct W3FDIFContext { ...@@ -42,6 +43,8 @@ typedef struct W3FDIFContext {
AVFrame *prev, *cur, *next; ///< previous, current, next frames AVFrame *prev, *cur, *next; ///< previous, current, next frames
int32_t **work_line; ///< lines we are calculating int32_t **work_line; ///< lines we are calculating
int nb_threads; int nb_threads;
W3FDIFDSPContext dsp;
} W3FDIFContext; } W3FDIFContext;
#define OFFSET(x) offsetof(W3FDIFContext, x) #define OFFSET(x) offsetof(W3FDIFContext, x)
...@@ -81,6 +84,78 @@ static int query_formats(AVFilterContext *ctx) ...@@ -81,6 +84,78 @@ static int query_formats(AVFilterContext *ctx)
return ff_set_common_formats(ctx, fmts_list); return ff_set_common_formats(ctx, fmts_list);
} }
static void filter_simple_low(int32_t *work_line,
uint8_t *in_lines_cur[2],
const int16_t *coef, int linesize)
{
int i;
for (i = 0; i < linesize; i++) {
*work_line = *in_lines_cur[0]++ * coef[0];
*work_line++ += *in_lines_cur[1]++ * coef[1];
}
}
static void filter_complex_low(int32_t *work_line,
uint8_t *in_lines_cur[4],
const int16_t *coef, int linesize)
{
int i;
for (i = 0; i < linesize; i++) {
*work_line = *in_lines_cur[0]++ * coef[0];
*work_line += *in_lines_cur[1]++ * coef[1];
*work_line += *in_lines_cur[2]++ * coef[2];
*work_line++ += *in_lines_cur[3]++ * coef[3];
}
}
static void filter_simple_high(int32_t *work_line,
uint8_t *in_lines_cur[3],
uint8_t *in_lines_adj[3],
const int16_t *coef, int linesize)
{
int i;
for (i = 0; i < linesize; i++) {
*work_line += *in_lines_cur[0]++ * coef[0];
*work_line += *in_lines_adj[0]++ * coef[0];
*work_line += *in_lines_cur[1]++ * coef[1];
*work_line += *in_lines_adj[1]++ * coef[1];
*work_line += *in_lines_cur[2]++ * coef[2];
*work_line++ += *in_lines_adj[2]++ * coef[2];
}
}
static void filter_complex_high(int32_t *work_line,
uint8_t *in_lines_cur[5],
uint8_t *in_lines_adj[5],
const int16_t *coef, int linesize)
{
int i;
for (i = 0; i < linesize; i++) {
*work_line += *in_lines_cur[0]++ * coef[0];
*work_line += *in_lines_adj[0]++ * coef[0];
*work_line += *in_lines_cur[1]++ * coef[1];
*work_line += *in_lines_adj[1]++ * coef[1];
*work_line += *in_lines_cur[2]++ * coef[2];
*work_line += *in_lines_adj[2]++ * coef[2];
*work_line += *in_lines_cur[3]++ * coef[3];
*work_line += *in_lines_adj[3]++ * coef[3];
*work_line += *in_lines_cur[4]++ * coef[4];
*work_line++ += *in_lines_adj[4]++ * coef[4];
}
}
static void filter_scale(uint8_t *out_pixel, const int32_t *work_pixel, int linesize)
{
int j;
for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
*out_pixel = av_clip(*work_pixel, 0, 255 * 256 * 128) >> 15;
}
static int config_input(AVFilterLink *inlink) static int config_input(AVFilterLink *inlink)
{ {
AVFilterContext *ctx = inlink->dst; AVFilterContext *ctx = inlink->dst;
...@@ -101,11 +176,20 @@ static int config_input(AVFilterLink *inlink) ...@@ -101,11 +176,20 @@ static int config_input(AVFilterLink *inlink)
return AVERROR(ENOMEM); return AVERROR(ENOMEM);
for (i = 0; i < s->nb_threads; i++) { for (i = 0; i < s->nb_threads; i++) {
s->work_line[i] = av_calloc(s->linesize[0], sizeof(*s->work_line[0])); s->work_line[i] = av_calloc(FFALIGN(s->linesize[0], 32), sizeof(*s->work_line[0]));
if (!s->work_line[i]) if (!s->work_line[i])
return AVERROR(ENOMEM); return AVERROR(ENOMEM);
} }
s->dsp.filter_simple_low = filter_simple_low;
s->dsp.filter_complex_low = filter_complex_low;
s->dsp.filter_simple_high = filter_simple_high;
s->dsp.filter_complex_high = filter_complex_high;
s->dsp.filter_scale = filter_scale;
if (ARCH_X86)
ff_w3fdif_init_x86(&s->dsp);
return 0; return 0;
} }
...@@ -163,7 +247,7 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_ ...@@ -163,7 +247,7 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_
const int dst_line_stride = out->linesize[plane]; const int dst_line_stride = out->linesize[plane];
const int start = (height * jobnr) / nb_jobs; const int start = (height * jobnr) / nb_jobs;
const int end = (height * (jobnr+1)) / nb_jobs; const int end = (height * (jobnr+1)) / nb_jobs;
int i, j, y_in, y_out; int j, y_in, y_out;
/* copy unchanged the lines of the field */ /* copy unchanged the lines of the field */
y_out = start + (s->field == cur->top_field_first) - (start & 1); y_out = start + (s->field == cur->top_field_first) - (start & 1);
...@@ -184,9 +268,6 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_ ...@@ -184,9 +268,6 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_
out_line = dst_data + (y_out * dst_line_stride); out_line = dst_data + (y_out * dst_line_stride);
while (y_out < end) { while (y_out < end) {
/* clear workspace */
memset(s->work_line[jobnr], 0, sizeof(*s->work_line[jobnr]) * linesize);
/* get low vertical frequencies from current field */ /* get low vertical frequencies from current field */
for (j = 0; j < n_coef_lf[filter]; j++) { for (j = 0; j < n_coef_lf[filter]; j++) {
y_in = (y_out + 1) + (j * 2) - n_coef_lf[filter]; y_in = (y_out + 1) + (j * 2) - n_coef_lf[filter];
...@@ -202,18 +283,12 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_ ...@@ -202,18 +283,12 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_
work_line = s->work_line[jobnr]; work_line = s->work_line[jobnr];
switch (n_coef_lf[filter]) { switch (n_coef_lf[filter]) {
case 2: case 2:
for (i = 0; i < linesize; i++) { s->dsp.filter_simple_low(work_line, in_lines_cur,
*work_line += *in_lines_cur[0]++ * coef_lf[filter][0]; coef_lf[filter], linesize);
*work_line++ += *in_lines_cur[1]++ * coef_lf[filter][1];
}
break; break;
case 4: case 4:
for (i = 0; i < linesize; i++) { s->dsp.filter_complex_low(work_line, in_lines_cur,
*work_line += *in_lines_cur[0]++ * coef_lf[filter][0]; coef_lf[filter], linesize);
*work_line += *in_lines_cur[1]++ * coef_lf[filter][1];
*work_line += *in_lines_cur[2]++ * coef_lf[filter][2];
*work_line++ += *in_lines_cur[3]++ * coef_lf[filter][3];
}
} }
/* get high vertical frequencies from adjacent fields */ /* get high vertical frequencies from adjacent fields */
...@@ -232,36 +307,19 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_ ...@@ -232,36 +307,19 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_
work_line = s->work_line[jobnr]; work_line = s->work_line[jobnr];
switch (n_coef_hf[filter]) { switch (n_coef_hf[filter]) {
case 3: case 3:
for (i = 0; i < linesize; i++) { s->dsp.filter_simple_high(work_line, in_lines_cur, in_lines_adj,
*work_line += *in_lines_cur[0]++ * coef_hf[filter][0]; coef_hf[filter], linesize);
*work_line += *in_lines_adj[0]++ * coef_hf[filter][0];
*work_line += *in_lines_cur[1]++ * coef_hf[filter][1];
*work_line += *in_lines_adj[1]++ * coef_hf[filter][1];
*work_line += *in_lines_cur[2]++ * coef_hf[filter][2];
*work_line++ += *in_lines_adj[2]++ * coef_hf[filter][2];
}
break; break;
case 5: case 5:
for (i = 0; i < linesize; i++) { s->dsp.filter_complex_high(work_line, in_lines_cur, in_lines_adj,
*work_line += *in_lines_cur[0]++ * coef_hf[filter][0]; coef_hf[filter], linesize);
*work_line += *in_lines_adj[0]++ * coef_hf[filter][0];
*work_line += *in_lines_cur[1]++ * coef_hf[filter][1];
*work_line += *in_lines_adj[1]++ * coef_hf[filter][1];
*work_line += *in_lines_cur[2]++ * coef_hf[filter][2];
*work_line += *in_lines_adj[2]++ * coef_hf[filter][2];
*work_line += *in_lines_cur[3]++ * coef_hf[filter][3];
*work_line += *in_lines_adj[3]++ * coef_hf[filter][3];
*work_line += *in_lines_cur[4]++ * coef_hf[filter][4];
*work_line++ += *in_lines_adj[4]++ * coef_hf[filter][4];
}
} }
/* save scaled result to the output frame, scaling down by 256 * 128 */ /* save scaled result to the output frame, scaling down by 256 * 128 */
work_pixel = s->work_line[jobnr]; work_pixel = s->work_line[jobnr];
out_pixel = out_line; out_pixel = out_line;
for (j = 0; j < linesize; j++, out_pixel++, work_pixel++) s->dsp.filter_scale(out_pixel, work_pixel, linesize);
*out_pixel = av_clip(*work_pixel, 0, 255 * 256 * 128) >> 15;
/* move on to next line */ /* move on to next line */
y_out += 2; y_out += 2;
......
/*
* Copyright (c) 2015 Paul B Mahol
*
* 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
*/
#ifndef AVFILTER_W3FDIF_H
#define AVFILTER_W3FDIF_H
#include <stddef.h>
#include <stdint.h>
typedef struct W3FDIFDSPContext {
void (*filter_simple_low)(int32_t *work_line,
uint8_t *in_lines_cur[2],
const int16_t *coef, int linesize);
void (*filter_complex_low)(int32_t *work_line,
uint8_t *in_lines_cur[4],
const int16_t *coef, int linesize);
void (*filter_simple_high)(int32_t *work_line,
uint8_t *in_lines_cur[3],
uint8_t *in_lines_adj[3],
const int16_t *coef, int linesize);
void (*filter_complex_high)(int32_t *work_line,
uint8_t *in_lines_cur[5],
uint8_t *in_lines_adj[5],
const int16_t *coef, int linesize);
void (*filter_scale)(uint8_t *out_pixel, const int32_t *work_pixel, int linesize);
} W3FDIFDSPContext;
void ff_w3fdif_init_x86(W3FDIFDSPContext *dsp);
#endif /* AVFILTER_W3FDIF_H */
...@@ -17,6 +17,7 @@ OBJS-$(CONFIG_STEREO3D_FILTER) += x86/vf_stereo3d_init.o ...@@ -17,6 +17,7 @@ OBJS-$(CONFIG_STEREO3D_FILTER) += x86/vf_stereo3d_init.o
OBJS-$(CONFIG_TBLEND_FILTER) += x86/vf_blend_init.o OBJS-$(CONFIG_TBLEND_FILTER) += x86/vf_blend_init.o
OBJS-$(CONFIG_TINTERLACE_FILTER) += x86/vf_tinterlace_init.o OBJS-$(CONFIG_TINTERLACE_FILTER) += x86/vf_tinterlace_init.o
OBJS-$(CONFIG_VOLUME_FILTER) += x86/af_volume_init.o OBJS-$(CONFIG_VOLUME_FILTER) += x86/af_volume_init.o
OBJS-$(CONFIG_W3FDIF_FILTER) += x86/vf_w3fdif_init.o
OBJS-$(CONFIG_YADIF_FILTER) += x86/vf_yadif_init.o OBJS-$(CONFIG_YADIF_FILTER) += x86/vf_yadif_init.o
YASM-OBJS-$(CONFIG_BLEND_FILTER) += x86/vf_blend.o YASM-OBJS-$(CONFIG_BLEND_FILTER) += x86/vf_blend.o
...@@ -37,4 +38,5 @@ YASM-OBJS-$(CONFIG_STEREO3D_FILTER) += x86/vf_stereo3d.o ...@@ -37,4 +38,5 @@ YASM-OBJS-$(CONFIG_STEREO3D_FILTER) += x86/vf_stereo3d.o
YASM-OBJS-$(CONFIG_TBLEND_FILTER) += x86/vf_blend.o YASM-OBJS-$(CONFIG_TBLEND_FILTER) += x86/vf_blend.o
YASM-OBJS-$(CONFIG_TINTERLACE_FILTER) += x86/vf_interlace.o YASM-OBJS-$(CONFIG_TINTERLACE_FILTER) += x86/vf_interlace.o
YASM-OBJS-$(CONFIG_VOLUME_FILTER) += x86/af_volume.o YASM-OBJS-$(CONFIG_VOLUME_FILTER) += x86/af_volume.o
YASM-OBJS-$(CONFIG_W3FDIF_FILTER) += x86/vf_w3fdif.o
YASM-OBJS-$(CONFIG_YADIF_FILTER) += x86/vf_yadif.o x86/yadif-16.o x86/yadif-10.o YASM-OBJS-$(CONFIG_YADIF_FILTER) += x86/vf_yadif.o x86/yadif-16.o x86/yadif-10.o
This diff is collapsed.
/*
* Copyright (C) 2015 Paul B Mahol
*
* This file is part of FFmpeg.
*
* FFmpeg is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* FFmpeg is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
#include "libavutil/mem.h"
#include "libavutil/x86/asm.h"
#include "libavutil/x86/cpu.h"
#include "libavfilter/w3fdif.h"
void ff_w3fdif_simple_low_sse2(int32_t *work_line,
uint8_t *in_lines_cur[2],
const int16_t *coef, int linesize);
void ff_w3fdif_simple_high_sse2(int32_t *work_line,
uint8_t *in_lines_cur[3],
uint8_t *in_lines_adj[3],
const int16_t *coef, int linesize);
void ff_w3fdif_complex_low_sse2(int32_t *work_line,
uint8_t *in_lines_cur[4],
const int16_t *coef, int linesize);
void ff_w3fdif_complex_high_sse2(int32_t *work_line,
uint8_t *in_lines_cur[5],
uint8_t *in_lines_adj[5],
const int16_t *coef, int linesize);
void ff_w3fdif_scale_sse2(uint8_t *out_pixel, const int32_t *work_pixel, int linesize);
av_cold void ff_w3fdif_init_x86(W3FDIFDSPContext *dsp)
{
int cpu_flags = av_get_cpu_flags();
if (EXTERNAL_SSE2(cpu_flags)) {
dsp->filter_simple_low = ff_w3fdif_simple_low_sse2;
dsp->filter_complex_low = ff_w3fdif_complex_low_sse2;
dsp->filter_scale = ff_w3fdif_scale_sse2;
}
if (ARCH_X86_64 && EXTERNAL_SSE2(cpu_flags)) {
dsp->filter_simple_high = ff_w3fdif_simple_high_sse2;
dsp->filter_complex_high = ff_w3fdif_complex_high_sse2;
}
}
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment