[FFmpeg-devel] [PATCH 1/2] VP56: move vp56_edge_filter to new VP56DSPContext
Mans Rullgard
mans
Fri Apr 23 16:24:33 CEST 2010
Using macro templates allows the vp[56]_adjust functions to be
inlined instead of called through function pointers. The new
function pointers enable optimised implementations of the filters.
4% faster VP6 decoding on Cortex-A8.
---
libavcodec/Makefile | 4 +-
libavcodec/vp5.c | 18 ----------
libavcodec/vp56.c | 20 ++----------
libavcodec/vp56.h | 5 +--
libavcodec/vp56dsp.c | 86 ++++++++++++++++++++++++++++++++++++++++++++++++++
libavcodec/vp56dsp.h | 33 +++++++++++++++++++
libavcodec/vp6.c | 14 --------
7 files changed, 126 insertions(+), 54 deletions(-)
create mode 100644 libavcodec/vp56dsp.c
create mode 100644 libavcodec/vp56dsp.h
diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 68b42b4..064eee7 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -357,9 +357,9 @@ OBJS-$(CONFIG_VORBIS_DECODER) += vorbis_dec.o vorbis.o \
OBJS-$(CONFIG_VORBIS_ENCODER) += vorbis_enc.o vorbis.o \
vorbis_data.o
OBJS-$(CONFIG_VP3_DECODER) += vp3.o vp3dsp.o
-OBJS-$(CONFIG_VP5_DECODER) += vp5.o vp56.o vp56data.o \
+OBJS-$(CONFIG_VP5_DECODER) += vp5.o vp56.o vp56data.o vp56dsp.o \
vp3dsp.o
-OBJS-$(CONFIG_VP6_DECODER) += vp6.o vp56.o vp56data.o \
+OBJS-$(CONFIG_VP6_DECODER) += vp6.o vp56.o vp56data.o vp56dsp.o \
vp3dsp.o vp6dsp.o huffman.o
OBJS-$(CONFIG_VQA_DECODER) += vqavideo.o
OBJS-$(CONFIG_WAVPACK_DECODER) += wavpack.o
diff --git a/libavcodec/vp5.c b/libavcodec/vp5.c
index f7177ca..1479344 100644
--- a/libavcodec/vp5.c
+++ b/libavcodec/vp5.c
@@ -69,23 +69,6 @@ static int vp5_parse_header(VP56Context *s, const uint8_t *buf, int buf_size,
return 1;
}
-/* Gives very similar result than the vp6 version except in a few cases */
-static int vp5_adjust(int v, int t)
-{
- int s2, s1 = v >> 31;
- v ^= s1;
- v -= s1;
- v *= v < 2*t;
- v -= t;
- s2 = v >> 31;
- v ^= s2;
- v -= s2;
- v = t - v;
- v += s1;
- v ^= s1;
- return v;
-}
-
static void vp5_parse_vector_adjustment(VP56Context *s, VP56mv *vect)
{
VP56RangeCoder *c = &s->c;
@@ -274,7 +257,6 @@ static av_cold int vp5_decode_init(AVCodecContext *avctx)
vp56_init(avctx, 1, 0);
s->vp56_coord_div = vp5_coord_div;
s->parse_vector_adjustment = vp5_parse_vector_adjustment;
- s->adjust = vp5_adjust;
s->parse_coeff = vp5_parse_coeff;
s->default_models_init = vp5_default_models_init;
s->parse_vector_models = vp5_parse_vector_models;
diff --git a/libavcodec/vp56.c b/libavcodec/vp56.c
index 5ce838a..74fe5ff 100644
--- a/libavcodec/vp56.c
+++ b/libavcodec/vp56.c
@@ -300,27 +300,12 @@ static void vp56_add_predictors_dc(VP56Context *s, VP56Frame ref_frame)
}
}
-static void vp56_edge_filter(VP56Context *s, uint8_t *yuv,
- int pix_inc, int line_inc, int t)
-{
- int pix2_inc = 2 * pix_inc;
- int i, v;
-
- for (i=0; i<12; i++) {
- v = (yuv[-pix2_inc] + 3*(yuv[0]-yuv[-pix_inc]) - yuv[pix_inc] + 4) >>3;
- v = s->adjust(v, t);
- yuv[-pix_inc] = av_clip_uint8(yuv[-pix_inc] + v);
- yuv[0] = av_clip_uint8(yuv[0] - v);
- yuv += line_inc;
- }
-}
-
static void vp56_deblock_filter(VP56Context *s, uint8_t *yuv,
int stride, int dx, int dy)
{
int t = vp56_filter_threshold[s->quantizer];
- if (dx) vp56_edge_filter(s, yuv + 10-dx , 1, stride, t);
- if (dy) vp56_edge_filter(s, yuv + stride*(10-dy), stride, 1, t);
+ if (dx) s->vp56dsp.edge_filter_hor(yuv + 10-dx , stride, t);
+ if (dy) s->vp56dsp.edge_filter_ver(yuv + stride*(10-dy), stride, t);
}
static void vp56_mc(VP56Context *s, int b, int plane, uint8_t *src,
@@ -665,6 +650,7 @@ av_cold void vp56_init(AVCodecContext *avctx, int flip, int has_alpha)
if (avctx->idct_algo == FF_IDCT_AUTO)
avctx->idct_algo = FF_IDCT_VP3;
dsputil_init(&s->dsp, avctx);
+ ff_vp56dsp_init(&s->vp56dsp, avctx->codec->id);
ff_init_scantable(s->dsp.idct_permutation, &s->scantable,ff_zigzag_direct);
for (i=0; i<4; i++)
diff --git a/libavcodec/vp56.h b/libavcodec/vp56.h
index 2d9e2d8..89eba05 100644
--- a/libavcodec/vp56.h
+++ b/libavcodec/vp56.h
@@ -28,14 +28,13 @@
#include "dsputil.h"
#include "get_bits.h"
#include "bytestream.h"
-
+#include "vp56dsp.h"
typedef struct vp56_context VP56Context;
typedef struct vp56_mv VP56mv;
typedef void (*VP56ParseVectorAdjustment)(VP56Context *s,
VP56mv *vect);
-typedef int (*VP56Adjust)(int v, int t);
typedef void (*VP56Filter)(VP56Context *s, uint8_t *dst, uint8_t *src,
int offset1, int offset2, int stride,
VP56mv mv, int mask, int select, int luma);
@@ -90,6 +89,7 @@ typedef struct {
struct vp56_context {
AVCodecContext *avctx;
DSPContext dsp;
+ VP56DSPContext vp56dsp;
ScanTable scantable;
AVFrame frames[4];
AVFrame *framep[6];
@@ -149,7 +149,6 @@ struct vp56_context {
const uint8_t *vp56_coord_div;
VP56ParseVectorAdjustment parse_vector_adjustment;
- VP56Adjust adjust;
VP56Filter filter;
VP56ParseCoeff parse_coeff;
VP56DefaultModelsInit default_models_init;
diff --git a/libavcodec/vp56dsp.c b/libavcodec/vp56dsp.c
new file mode 100644
index 0000000..991c90a
--- /dev/null
+++ b/libavcodec/vp56dsp.c
@@ -0,0 +1,86 @@
+/*
+ * Copyright (c) 2006 Aurelien Jacobs <aurel at gnuage.org>
+ * Copyright (c) 2010 Mans Rullgard <mans at mansr.com>
+ *
+ * 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 <stdint.h>
+#include "avcodec.h"
+#include "vp56dsp.h"
+
+/* Gives very similar result than the vp6 version except in a few cases */
+static int vp5_adjust(int v, int t)
+{
+ int s2, s1 = v >> 31;
+ v ^= s1;
+ v -= s1;
+ v *= v < 2*t;
+ v -= t;
+ s2 = v >> 31;
+ v ^= s2;
+ v -= s2;
+ v = t - v;
+ v += s1;
+ v ^= s1;
+ return v;
+}
+
+static int vp6_adjust(int v, int t)
+{
+ int V = v, s = v >> 31;
+ V ^= s;
+ V -= s;
+ if (V-t-1 >= (unsigned)(t-1))
+ return v;
+ V = 2*t - V;
+ V += s;
+ V ^= s;
+ return V;
+}
+
+
+#define VP56_EDGE_FILTER(pfx, suf, pix_inc, line_inc, adjust) \
+static void pfx##_edge_filter_##suf(uint8_t *yuv, int stride, int t) \
+{ \
+ int pix2_inc = 2 * pix_inc; \
+ int i, v; \
+ \
+ for (i=0; i<12; i++) { \
+ v = (yuv[-pix2_inc] + 3*(yuv[0]-yuv[-pix_inc]) - yuv[pix_inc] + 4)>>3;\
+ v = adjust(v, t); \
+ yuv[-pix_inc] = av_clip_uint8(yuv[-pix_inc] + v); \
+ yuv[0] = av_clip_uint8(yuv[0] - v); \
+ yuv += line_inc; \
+ } \
+}
+
+VP56_EDGE_FILTER(vp5, hor, 1, stride, vp5_adjust)
+VP56_EDGE_FILTER(vp5, ver, stride, 1, vp5_adjust)
+VP56_EDGE_FILTER(vp6, hor, 1, stride, vp6_adjust)
+VP56_EDGE_FILTER(vp6, ver, stride, 1, vp6_adjust)
+
+void ff_vp56dsp_init(VP56DSPContext *s, enum CodecID codec)
+{
+ if (codec == CODEC_ID_VP5) {
+ s->edge_filter_hor = vp5_edge_filter_hor;
+ s->edge_filter_ver = vp5_edge_filter_ver;
+ } else {
+ s->edge_filter_hor = vp6_edge_filter_hor;
+ s->edge_filter_ver = vp6_edge_filter_ver;
+ }
+}
diff --git a/libavcodec/vp56dsp.h b/libavcodec/vp56dsp.h
new file mode 100644
index 0000000..dbc84b1
--- /dev/null
+++ b/libavcodec/vp56dsp.h
@@ -0,0 +1,33 @@
+/*
+ * Copyright (c) 2010 Mans Rullgard <mans at mansr.com>
+ *
+ * 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 AVCODEC_VP56DSP_H
+#define AVCODEC_VP56DSP_H
+
+#include <stdint.h>
+
+typedef struct VP56DSPContext {
+ void (*edge_filter_hor)(uint8_t *yuv, int stride, int t);
+ void (*edge_filter_ver)(uint8_t *yuv, int stride, int t);
+} VP56DSPContext;
+
+void ff_vp56dsp_init(VP56DSPContext *s, enum CodecID codec);
+
+#endif /* AVCODEC_VP56DSP_H */
diff --git a/libavcodec/vp6.c b/libavcodec/vp6.c
index e4adb17..58c31f9 100644
--- a/libavcodec/vp6.c
+++ b/libavcodec/vp6.c
@@ -481,19 +481,6 @@ static void vp6_parse_coeff(VP56Context *s)
}
}
-static int vp6_adjust(int v, int t)
-{
- int V = v, s = v >> 31;
- V ^= s;
- V -= s;
- if (V-t-1 >= (unsigned)(t-1))
- return v;
- V = 2*t - V;
- V += s;
- V ^= s;
- return V;
-}
-
static int vp6_block_variance(uint8_t *src, int stride)
{
int sum = 0, square_sum = 0;
@@ -592,7 +579,6 @@ static av_cold int vp6_decode_init(AVCodecContext *avctx)
avctx->codec->id == CODEC_ID_VP6A);
s->vp56_coord_div = vp6_coord_div;
s->parse_vector_adjustment = vp6_parse_vector_adjustment;
- s->adjust = vp6_adjust;
s->filter = vp6_filter;
s->default_models_init = vp6_default_models_init;
s->parse_vector_models = vp6_parse_vector_models;
--
1.7.0.4
More information about the ffmpeg-devel
mailing list