[FFmpeg-devel] [PATCH] avfilter/vf_w3fdif: add x86 SIMD

Paul B Mahol onemda at gmail.com
Wed Oct 7 22:27:03 CEST 2015


Signed-off-by: Paul B Mahol <onemda at gmail.com>
---
 libavfilter/vf_w3fdif.c          | 139 +++++++++++++++--------
 libavfilter/w3fdif.h             |  61 +++++++++++
 libavfilter/x86/Makefile         |   2 +
 libavfilter/x86/vf_w3fdif.asm    | 230 +++++++++++++++++++++++++++++++++++++++
 libavfilter/x86/vf_w3fdif_init.c |  56 ++++++++++
 5 files changed, 441 insertions(+), 47 deletions(-)
 create mode 100644 libavfilter/w3fdif.h
 create mode 100644 libavfilter/x86/vf_w3fdif.asm
 create mode 100644 libavfilter/x86/vf_w3fdif_init.c

diff --git a/libavfilter/vf_w3fdif.c b/libavfilter/vf_w3fdif.c
index 5b105f1..265443a 100644
--- a/libavfilter/vf_w3fdif.c
+++ b/libavfilter/vf_w3fdif.c
@@ -29,20 +29,7 @@
 #include "formats.h"
 #include "internal.h"
 #include "video.h"
-
-typedef struct W3FDIFContext {
-    const AVClass *class;
-    int filter;           ///< 0 is simple, 1 is more complex
-    int deint;            ///< which frames to deinterlace
-    int linesize[4];      ///< bytes of pixel data per line for each plane
-    int planeheight[4];   ///< height of each plane
-    int field;            ///< which field are we on, 0 or 1
-    int eof;
-    int nb_planes;
-    AVFrame *prev, *cur, *next;  ///< previous, current, next frames
-    int32_t **work_line;  ///< lines we are calculating
-    int nb_threads;
-} W3FDIFContext;
+#include "w3fdif.h"
 
 #define OFFSET(x) offsetof(W3FDIFContext, x)
 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
@@ -81,6 +68,78 @@ static int query_formats(AVFilterContext *ctx)
     return ff_set_common_formats(ctx, fmts_list);
 }
 
+static void filter_simple_low(int32_t *work_line,
+                              uint8_t *in_lines_cur[2],
+                              const int32_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 int32_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 int32_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 int32_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 * 256) >> 16;
+}
+
 static int config_input(AVFilterLink *inlink)
 {
     AVFilterContext *ctx = inlink->dst;
@@ -106,6 +165,15 @@ static int config_input(AVFilterLink *inlink)
             return AVERROR(ENOMEM);
     }
 
+    s->filter_simple_low = filter_simple_low;
+    s->filter_complex_low = filter_complex_low;
+    s->filter_simple_high = filter_simple_high;
+    s->filter_complex_high = filter_complex_high;
+    s->filter_scale = filter_scale;
+
+    if (ARCH_X86)
+        ff_w3fdif_init_x86(s);
+
     return 0;
 }
 
@@ -163,7 +231,7 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_
     const int dst_line_stride = out->linesize[plane];
     const int start = (height * jobnr) / 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 */
     y_out = start + (s->field == cur->top_field_first) - (start & 1);
@@ -202,18 +270,12 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_
         work_line = s->work_line[jobnr];
         switch (n_coef_lf[filter]) {
         case 2:
-            for (i = 0; i < linesize; i++) {
-                *work_line   += *in_lines_cur[0]++ * coef_lf[filter][0];
-                *work_line++ += *in_lines_cur[1]++ * coef_lf[filter][1];
-            }
+            s->filter_simple_low(work_line, in_lines_cur,
+                                 coef_lf[filter], linesize);
             break;
         case 4:
-            for (i = 0; i < linesize; i++) {
-                *work_line   += *in_lines_cur[0]++ * coef_lf[filter][0];
-                *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];
-            }
+            s->filter_complex_low(work_line, in_lines_cur,
+                                  coef_lf[filter], linesize);
         }
 
         /* get high vertical frequencies from adjacent fields */
@@ -232,36 +294,19 @@ static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_
         work_line = s->work_line[jobnr];
         switch (n_coef_hf[filter]) {
         case 3:
-            for (i = 0; i < linesize; i++) {
-                *work_line   += *in_lines_cur[0]++ * coef_hf[filter][0];
-                *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];
-            }
+            s->filter_simple_high(work_line, in_lines_cur, in_lines_adj,
+                                  coef_hf[filter], linesize);
             break;
         case 5:
-            for (i = 0; i < linesize; i++) {
-                *work_line   += *in_lines_cur[0]++ * coef_hf[filter][0];
-                *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];
-            }
+            s->filter_complex_high(work_line, in_lines_cur, in_lines_adj,
+                                   coef_hf[filter], linesize);
         }
 
         /* save scaled result to the output frame, scaling down by 256 * 256 */
         work_pixel = s->work_line[jobnr];
         out_pixel = out_line;
 
-        for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
-             *out_pixel = av_clip(*work_pixel, 0, 255 * 256 * 256) >> 16;
+        s->filter_scale(out_pixel, work_pixel, linesize);
 
         /* move on to next line */
         y_out += 2;
diff --git a/libavfilter/w3fdif.h b/libavfilter/w3fdif.h
new file mode 100644
index 0000000..76902cf
--- /dev/null
+++ b/libavfilter/w3fdif.h
@@ -0,0 +1,61 @@
+/*
+ * Copyright (C) 2012 British Broadcasting Corporation, All Rights Reserved
+ * Author of de-interlace algorithm: Jim Easterbrook for BBC R&D
+ * Based on the process described by Martin Weston for BBC R&D
+ * Author of FFmpeg filter: Mark Himsley for BBC Broadcast Systems Development
+ *
+ * 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 "avfilter.h"
+
+typedef struct W3FDIFContext {
+    const AVClass *class;
+    int filter;           ///< 0 is simple, 1 is more complex
+    int deint;            ///< which frames to deinterlace
+    int linesize[4];      ///< bytes of pixel data per line for each plane
+    int planeheight[4];   ///< height of each plane
+    int field;            ///< which field are we on, 0 or 1
+    int eof;
+    int nb_planes;
+    AVFrame *prev, *cur, *next;  ///< previous, current, next frames
+    int32_t **work_line;  ///< lines we are calculating
+    int nb_threads;
+
+    void (*filter_simple_low)(int32_t *work_line,
+                              uint8_t *in_lines_cur[2],
+                              const int32_t *coef, int linesize);
+    void (*filter_complex_low)(int32_t *work_line,
+                               uint8_t *in_lines_cur[4],
+                               const int32_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 int32_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 int32_t *coef, int linesize);
+    void (*filter_scale)(uint8_t *out_pixel, const int32_t *work_pixel, int linesize);
+} W3FDIFContext;
+
+void ff_w3fdif_init_x86(W3FDIFContext *s);
+
+#endif /* AVFILTER_W3FDIF_H */
diff --git a/libavfilter/x86/Makefile b/libavfilter/x86/Makefile
index c024204..33de380 100644
--- a/libavfilter/x86/Makefile
+++ b/libavfilter/x86/Makefile
@@ -17,6 +17,7 @@ OBJS-$(CONFIG_STEREO3D_FILTER)               += x86/vf_stereo3d_init.o
 OBJS-$(CONFIG_TBLEND_FILTER)                 += x86/vf_blend_init.o
 OBJS-$(CONFIG_TINTERLACE_FILTER)             += x86/vf_tinterlace_init.o
 OBJS-$(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
 
 YASM-OBJS-$(CONFIG_BLEND_FILTER)             += x86/vf_blend.o
@@ -37,4 +38,5 @@ YASM-OBJS-$(CONFIG_STEREO3D_FILTER)          += x86/vf_stereo3d.o
 YASM-OBJS-$(CONFIG_TBLEND_FILTER)            += x86/vf_blend.o
 YASM-OBJS-$(CONFIG_TINTERLACE_FILTER)        += x86/vf_interlace.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
diff --git a/libavfilter/x86/vf_w3fdif.asm b/libavfilter/x86/vf_w3fdif.asm
new file mode 100644
index 0000000..3d746f8
--- /dev/null
+++ b/libavfilter/x86/vf_w3fdif.asm
@@ -0,0 +1,230 @@
+;*****************************************************************************
+;* x86-optimized functions for w3fdif filter
+;*
+;* 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/x86/x86util.asm"
+
+SECTION_RODATA
+
+pb_unpck : db 0,-1,-1,-1,1,-1,-1,-1,2,-1,-1,-1,3,-1,-1,-1
+
+SECTION .text
+
+INIT_XMM sse4
+cglobal w3fdif_simple_low, 4, 6, 5, 0, work_line, in_lines_cur0, coef, linesize
+    movd                  m0, [coefq+0]
+    movd                  m1, [coefq+4]
+    mova                  m4, [pb_unpck]
+    SPLATD                m0
+    SPLATD                m1
+    shr            linesized, 2
+    mov                  r4q, 0
+    mov                  r5q, [in_lines_cur0q + gprsize]
+    mov       in_lines_cur0q, [in_lines_cur0q]
+    %define   in_lines_cur1q  r5q
+
+    .loop
+    movd                    m2, [in_lines_cur0q + r4q]
+    movd                    m3, [in_lines_cur1q + r4q]
+    pshufb                  m2, m4
+    pshufb                  m3, m4
+    pmulld                  m2, m0
+    pmulld                  m3, m1
+    paddd                   m2, m3
+    mova  [work_lineq + r4q*4], m2
+    add                    r4q, 4
+    sub              linesized, 1
+    jg .loop
+REP_RET
+
+cglobal w3fdif_simple_high, 5, 10, 8, 0, work_line, in_lines_cur0, in_lines_adj0, coef, linesize
+    movd                  m0, [coefq+0]
+    movd                  m1, [coefq+4]
+    movd                  m2, [coefq+8]
+    mova                  m7, [pb_unpck]
+    SPLATD                m0
+    SPLATD                m1
+    SPLATD                m2
+    shr            linesized, 2
+    mov                  r5q, 0
+    mov                  r7q, [in_lines_cur0q + gprsize*2]
+    mov                  r6q, [in_lines_cur0q + gprsize]
+    mov       in_lines_cur0q, [in_lines_cur0q]
+    %define   in_lines_cur1q  r6q
+    %define   in_lines_cur2q  r7q
+    mov                  r9q, [in_lines_adj0q + gprsize*2]
+    mov                  r8q, [in_lines_adj0q + gprsize]
+    mov       in_lines_adj0q, [in_lines_adj0q]
+    %define   in_lines_adj1q  r8q
+    %define   in_lines_adj2q  r9q
+
+    .loop
+    movd                    m3, [in_lines_cur0q + r5q]
+    movd                    m4, [in_lines_cur1q + r5q]
+    movd                    m5, [in_lines_cur2q + r5q]
+    pshufb                  m3, m7
+    pshufb                  m4, m7
+    pshufb                  m5, m7
+    pmulld                  m3, m0
+    pmulld                  m4, m1
+    pmulld                  m5, m2
+    paddd                   m3, m4
+    paddd                   m3, m5
+    movd                    m4, [in_lines_adj0q + r5q]
+    movd                    m5, [in_lines_adj1q + r5q]
+    movd                    m6, [in_lines_adj2q + r5q]
+    pshufb                  m4, m7
+    pshufb                  m5, m7
+    pshufb                  m6, m7
+    pmulld                  m4, m0
+    pmulld                  m5, m1
+    pmulld                  m6, m2
+    paddd                   m4, m5
+    paddd                   m4, m6
+    paddd                   m3, m4
+    movu                    m4, [work_lineq + r5q*4]
+    paddd                   m3, m4
+    mova  [work_lineq + r5q*4], m3
+    add                    r5q, 4
+    sub              linesized, 1
+    jg .loop
+REP_RET
+
+cglobal w3fdif_complex_low, 4, 9, 8, 0, work_line, in_lines_cur0, coef, linesize
+    movd                  m0, [coefq+0]
+    movd                  m1, [coefq+4]
+    movd                  m2, [coefq+8]
+    movd                  m3, [coefq+12]
+    SPLATD                m0
+    SPLATD                m1
+    SPLATD                m2
+    SPLATD                m3
+    shr            linesized, 2
+    mov                  r4q, 0
+    mov                  r7q, [in_lines_cur0q + gprsize*3]
+    mov                  r6q, [in_lines_cur0q + gprsize*2]
+    mov                  r5q, [in_lines_cur0q + gprsize]
+    mov       in_lines_cur0q, [in_lines_cur0q]
+    %define   in_lines_cur1q  r5q
+    %define   in_lines_cur2q  r6q
+    %define   in_lines_cur3q  r7q
+
+    .loop
+    movd                    m4, [in_lines_cur0q + r4q]
+    movd                    m5, [in_lines_cur1q + r4q]
+    pshufb                  m4, [pb_unpck]
+    pshufb                  m5, [pb_unpck]
+    pmulld                  m4, m0
+    pmulld                  m5, m1
+    paddd                   m4, m5
+    movd                    m5, [in_lines_cur2q + r4q]
+    movd                    m6, [in_lines_cur3q + r4q]
+    pshufb                  m5, [pb_unpck]
+    pshufb                  m6, [pb_unpck]
+    pmulld                  m5, m2
+    pmulld                  m6, m3
+    paddd                   m5, m6
+    paddd                   m4, m5
+    mova  [work_lineq + r4q*4], m4
+    add                    r4q, 4
+    sub              linesized, 1
+    jg .loop
+REP_RET
+
+cglobal w3fdif_complex_high, 5, 14, 8, 0, work_line, in_lines_cur0, in_lines_adj0, coef, linesize
+    movd                  m0, [coefq+0]
+    movd                  m1, [coefq+4]
+    movd                  m2, [coefq+8]
+    movd                  m3, [coefq+12]
+    movd                  m4, [coefq+16]
+    SPLATD                m0
+    SPLATD                m1
+    SPLATD                m2
+    SPLATD                m3
+    SPLATD                m4
+    shr            linesized, 2
+    mov                  r5q, 0
+    mov                  r9q, [in_lines_cur0q + gprsize*4]
+    mov                  r8q, [in_lines_cur0q + gprsize*3]
+    mov                  r7q, [in_lines_cur0q + gprsize*2]
+    mov                  r6q, [in_lines_cur0q + gprsize]
+    mov       in_lines_cur0q, [in_lines_cur0q]
+    %define   in_lines_cur1q  r6q
+    %define   in_lines_cur2q  r7q
+    %define   in_lines_cur3q  r8q
+    %define   in_lines_cur4q  r9q
+    mov                 r13q, [in_lines_adj0q + gprsize*4]
+    mov                 r12q, [in_lines_adj0q + gprsize*3]
+    mov                 r11q, [in_lines_adj0q + gprsize*2]
+    mov                 r10q, [in_lines_adj0q + gprsize]
+    mov       in_lines_adj0q, [in_lines_adj0q]
+    %define   in_lines_adj1q  r10q
+    %define   in_lines_adj2q  r11q
+    %define   in_lines_adj3q  r12q
+    %define   in_lines_adj4q  r13q
+
+    .loop
+    movd                    m5, [in_lines_cur0q + r5q]
+    movd                    m6, [in_lines_cur1q + r5q]
+    pshufb                  m5, [pb_unpck]
+    pshufb                  m6, [pb_unpck]
+    pmulld                  m5, m0
+    pmulld                  m6, m1
+    paddd                   m5, m6
+    movd                    m6, [in_lines_cur2q + r5q]
+    movd                    m7, [in_lines_cur3q + r5q]
+    pshufb                  m6, [pb_unpck]
+    pshufb                  m7, [pb_unpck]
+    pmulld                  m6, m2
+    pmulld                  m7, m3
+    paddd                   m5, m6
+    paddd                   m5, m7
+    movd                    m6, [in_lines_adj0q + r5q]
+    movd                    m7, [in_lines_adj1q + r5q]
+    pshufb                  m6, [pb_unpck]
+    pshufb                  m7, [pb_unpck]
+    pmulld                  m6, m0
+    pmulld                  m7, m1
+    paddd                   m5, m6
+    paddd                   m5, m7
+    movd                    m6, [in_lines_adj2q + r5q]
+    movd                    m7, [in_lines_adj3q + r5q]
+    pshufb                  m6, [pb_unpck]
+    pshufb                  m7, [pb_unpck]
+    pmulld                  m6, m2
+    pmulld                  m7, m3
+    paddd                   m5, m6
+    paddd                   m5, m7
+    movd                    m6, [in_lines_cur4q + r5q]
+    movd                    m7, [in_lines_adj4q + r5q]
+    pshufb                  m6, [pb_unpck]
+    pshufb                  m7, [pb_unpck]
+    pmulld                  m6, m4
+    pmulld                  m7, m4
+    paddd                   m5, m6
+    paddd                   m5, m7
+    movu                    m6, [work_lineq + r5q*4]
+    paddd                   m5, m6
+    mova  [work_lineq + r5q*4], m5
+    add                    r5q, 4
+    sub              linesized, 1
+    jg .loop
+REP_RET
diff --git a/libavfilter/x86/vf_w3fdif_init.c b/libavfilter/x86/vf_w3fdif_init.c
new file mode 100644
index 0000000..07a44fb
--- /dev/null
+++ b/libavfilter/x86/vf_w3fdif_init.c
@@ -0,0 +1,56 @@
+/*
+ * 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_sse4(int32_t *work_line,
+                               uint8_t *in_lines_cur[2],
+                               const int32_t *coef, int linesize);
+
+void ff_w3fdif_simple_high_sse4(int32_t *work_line,
+                                uint8_t *in_lines_cur[3],
+                                uint8_t *in_lines_adj[3],
+                                const int32_t *coef, int linesize);
+
+void ff_w3fdif_complex_low_sse4(int32_t *work_line,
+                                uint8_t *in_lines_cur[4],
+                                const int32_t *coef, int linesize);
+
+void ff_w3fdif_complex_high_sse4(int32_t *work_line,
+                                 uint8_t *in_lines_cur[5],
+                                 uint8_t *in_lines_adj[5],
+                                 const int32_t *coef, int linesize);
+
+av_cold void ff_w3fdif_init_x86(W3FDIFContext *s)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (EXTERNAL_SSE4(cpu_flags)) {
+        s->filter_simple_low   = ff_w3fdif_simple_low_sse4;
+        s->filter_simple_high  = ff_w3fdif_simple_high_sse4;
+        s->filter_complex_low  = ff_w3fdif_complex_low_sse4;
+        s->filter_complex_high = ff_w3fdif_complex_high_sse4;
+    }
+}
-- 
1.9.1



More information about the ffmpeg-devel mailing list