lavfi: add spp filter.
authorClément Bœsch <ubitux@gmail.com>
Sat, 11 May 2013 10:03:38 +0000 (12:03 +0200)
committerClément Bœsch <ubitux@gmail.com>
Thu, 13 Jun 2013 23:27:22 +0000 (01:27 +0200)
Changelog
LICENSE
configure
doc/filters.texi
libavfilter/Makefile
libavfilter/allfilters.c
libavfilter/version.h
libavfilter/vf_spp.c [new file with mode: 0644]
libavfilter/vf_spp.h [new file with mode: 0644]
libavfilter/x86/Makefile
libavfilter/x86/vf_spp.c [new file with mode: 0644]

index 2a52317..be3cf4c 100644 (file)
--- a/Changelog
+++ b/Changelog
@@ -67,6 +67,7 @@ version <next>:
 - ffprobe -show_chapters option
 - WavPack encoding through libwavpack
 - rotate filter
+- spp filter ported from libmpcodecs
 
 
 version 1.2:
diff --git a/LICENSE b/LICENSE
index 28cb294..99f6478 100644 (file)
--- a/LICENSE
+++ b/LICENSE
@@ -42,6 +42,7 @@ Specifically, the GPL parts of FFmpeg are
     - vf_pp.c
     - vf_sab.c
     - vf_smartblur.c
+    - vf_spp.c
     - vf_stereo3d.c
     - vf_super2xsai.c
     - vf_tinterlace.c
index ff35919..5659d86 100755 (executable)
--- a/configure
+++ b/configure
@@ -2175,6 +2175,7 @@ sab_filter_deps="gpl swscale"
 scale_filter_deps="swscale"
 smartblur_filter_deps="gpl swscale"
 showspectrum_filter_deps="avcodec rdft"
+spp_filter_deps="gpl avcodec fft"
 stereo3d_filter_deps="gpl"
 subtitles_filter_deps="avformat avcodec libass"
 super2xsai_filter_deps="gpl"
index c0bd281..98bdaa0 100644 (file)
@@ -6476,6 +6476,42 @@ stereo3d=abl:sbsr
 @end example
 @end itemize
 
+@section spp
+
+Apply a simple postprocessing filter that compresses and decompresses the image
+at several (or - in the case of @option{quality} level @code{6} - all) shifts
+and average the results.
+
+The filter accepts the following options:
+
+@table @option
+@item quality
+Set quality. This option defines the number of levels for averaging. It accepts
+an integer in the range 0-6. If set to @code{0}, the filter will have no
+effect. A value of @code{6} means the higher quality. For each increment of
+that value the speed drops by a factor of approximately 2.  Default value is
+@code{3}.
+
+@item qp
+Force a constant quantization parameter. If not set, the filter will use the QP
+from the video stream (if available).
+
+@item mode
+Set thresholding mode. Available modes are:
+
+@table @samp
+@item hard
+Set hard thresholding (default).
+@item soft
+Set soft thresholding (better de-ringing effect, but likely blurrier).
+@end table
+
+@item use_bframe_qp
+Enable the use of the QP from the B-Frames if set to @code{1}. Using this
+option may cause flicker since the B-Frames have often larger QP. Default is
+@code{0} (not enabled).
+@end table
+
 @anchor{subtitles}
 @section subtitles
 
index 9746bbb..bd0a135 100644 (file)
@@ -182,6 +182,7 @@ OBJS-$(CONFIG_SETTB_FILTER)                  += f_settb.o
 OBJS-$(CONFIG_SHOWINFO_FILTER)               += vf_showinfo.o
 OBJS-$(CONFIG_SMARTBLUR_FILTER)              += vf_smartblur.o
 OBJS-$(CONFIG_SPLIT_FILTER)                  += split.o
+OBJS-$(CONFIG_SPP_FILTER)                    += vf_spp.o
 OBJS-$(CONFIG_STEREO3D_FILTER)               += vf_stereo3d.o
 OBJS-$(CONFIG_SUBTITLES_FILTER)              += vf_subtitles.o
 OBJS-$(CONFIG_SUPER2XSAI_FILTER)             += vf_super2xsai.o
index e802601..85a793f 100644 (file)
@@ -177,6 +177,7 @@ void avfilter_register_all(void)
     REGISTER_FILTER(SHOWINFO,       showinfo,       vf);
     REGISTER_FILTER(SMARTBLUR,      smartblur,      vf);
     REGISTER_FILTER(SPLIT,          split,          vf);
+    REGISTER_FILTER(SPP,            spp,            vf);
     REGISTER_FILTER(STEREO3D,       stereo3d,       vf);
     REGISTER_FILTER(SUBTITLES,      subtitles,      vf);
     REGISTER_FILTER(SUPER2XSAI,     super2xsai,     vf);
index 4485a7f..8ad033c 100644 (file)
@@ -30,8 +30,8 @@
 #include "libavutil/avutil.h"
 
 #define LIBAVFILTER_VERSION_MAJOR  3
-#define LIBAVFILTER_VERSION_MINOR  76
-#define LIBAVFILTER_VERSION_MICRO 101
+#define LIBAVFILTER_VERSION_MINOR  77
+#define LIBAVFILTER_VERSION_MICRO 100
 
 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \
                                                LIBAVFILTER_VERSION_MINOR, \
diff --git a/libavfilter/vf_spp.c b/libavfilter/vf_spp.c
new file mode 100644 (file)
index 0000000..b84cfb7
--- /dev/null
@@ -0,0 +1,437 @@
+/*
+ * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
+ * Copyright (c) 2013 Clément Bœsch <ubitux@gmail.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 General Public License as published by
+ * the Free Software Foundation; either version 2 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU 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.
+ */
+
+/**
+ * @file
+ * Simple post processing filter
+ *
+ * This implementation is based on an algorithm described in
+ * "Aria Nosratinia Embedded Post-Processing for
+ * Enhancement of Compressed Images (1999)"
+ *
+ * Originally written by Michael Niedermayer for the MPlayer project, and
+ * ported by Clément Bœsch for FFmpeg.
+ */
+
+#include "libavcodec/dsputil.h"
+#include "libavutil/avassert.h"
+#include "libavutil/imgutils.h"
+#include "libavutil/opt.h"
+#include "libavutil/pixdesc.h"
+#include "internal.h"
+#include "vf_spp.h"
+
+enum mode {
+    MODE_HARD,
+    MODE_SOFT,
+    NB_MODES
+};
+
+#define OFFSET(x) offsetof(SPPContext, x)
+#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
+static const AVOption spp_options[] = {
+    { "quality", "set quality", OFFSET(log2_count), AV_OPT_TYPE_INT, {.i64 = 3}, 0, MAX_LEVEL, FLAGS },
+    { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 63, FLAGS },
+    { "mode", "set thresholding mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64 = MODE_HARD}, 0, NB_MODES - 1, FLAGS, "mode" },
+        { "hard", "hard thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_HARD}, INT_MIN, INT_MAX, FLAGS, "mode" },
+        { "soft", "soft thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_SOFT}, INT_MIN, INT_MAX, FLAGS, "mode" },
+    { "use_bframe_qp", "use B-frames' QP", OFFSET(use_bframe_qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 1, FLAGS },
+    { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(spp);
+
+// XXX: share between filters?
+DECLARE_ALIGNED(8, static const uint8_t, ldither)[8][8] = {
+    {  0,  48,  12,  60,   3,  51,  15,  63 },
+    { 32,  16,  44,  28,  35,  19,  47,  31 },
+    {  8,  56,   4,  52,  11,  59,   7,  55 },
+    { 40,  24,  36,  20,  43,  27,  39,  23 },
+    {  2,  50,  14,  62,   1,  49,  13,  61 },
+    { 34,  18,  46,  30,  33,  17,  45,  29 },
+    { 10,  58,   6,  54,   9,  57,   5,  53 },
+    { 42,  26,  38,  22,  41,  25,  37,  21 },
+};
+
+static const uint8_t offset[127][2] = {
+    {0,0},
+    {0,0}, {4,4},                                           // quality = 1
+    {0,0}, {2,2}, {6,4}, {4,6},                             // quality = 2
+    {0,0}, {5,1}, {2,2}, {7,3}, {4,4}, {1,5}, {6,6}, {3,7}, // quality = 3
+
+    {0,0}, {4,0}, {1,1}, {5,1}, {3,2}, {7,2}, {2,3}, {6,3}, // quality = 4
+    {0,4}, {4,4}, {1,5}, {5,5}, {3,6}, {7,6}, {2,7}, {6,7},
+
+    {0,0}, {0,2}, {0,4}, {0,6}, {1,1}, {1,3}, {1,5}, {1,7}, // quality = 5
+    {2,0}, {2,2}, {2,4}, {2,6}, {3,1}, {3,3}, {3,5}, {3,7},
+    {4,0}, {4,2}, {4,4}, {4,6}, {5,1}, {5,3}, {5,5}, {5,7},
+    {6,0}, {6,2}, {6,4}, {6,6}, {7,1}, {7,3}, {7,5}, {7,7},
+
+    {0,0}, {4,4}, {0,4}, {4,0}, {2,2}, {6,6}, {2,6}, {6,2}, // quality = 6
+    {0,2}, {4,6}, {0,6}, {4,2}, {2,0}, {6,4}, {2,4}, {6,0},
+    {1,1}, {5,5}, {1,5}, {5,1}, {3,3}, {7,7}, {3,7}, {7,3},
+    {1,3}, {5,7}, {1,7}, {5,3}, {3,1}, {7,5}, {3,5}, {7,1},
+    {0,1}, {4,5}, {0,5}, {4,1}, {2,3}, {6,7}, {2,7}, {6,3},
+    {0,3}, {4,7}, {0,7}, {4,3}, {2,1}, {6,5}, {2,5}, {6,1},
+    {1,0}, {5,4}, {1,4}, {5,0}, {3,2}, {7,6}, {3,6}, {7,2},
+    {1,2}, {5,6}, {1,6}, {5,2}, {3,0}, {7,4}, {3,4}, {7,0},
+};
+
+static void hardthresh_c(int16_t dst[64], const int16_t src[64],
+                         int qp, const uint8_t *permutation)
+{
+    int i;
+    int bias = 0; // FIXME
+
+    unsigned threshold1 = qp * ((1<<4) - bias) - 1;
+    unsigned threshold2 = threshold1 << 1;
+
+    memset(dst, 0, 64 * sizeof(dst[0]));
+    dst[0] = (src[0] + 4) >> 3;
+
+    for (i = 1; i < 64; i++) {
+        int level = src[i];
+        if (((unsigned)(level + threshold1)) > threshold2) {
+            const int j = permutation[i];
+            dst[j] = (level + 4) >> 3;
+        }
+    }
+}
+
+static void softthresh_c(int16_t dst[64], const int16_t src[64],
+                         int qp, const uint8_t *permutation)
+{
+    int i;
+    int bias = 0; //FIXME
+
+    unsigned threshold1 = qp * ((1<<4) - bias) - 1;
+    unsigned threshold2 = threshold1 << 1;
+
+    memset(dst, 0, 64 * sizeof(dst[0]));
+    dst[0] = (src[0] + 4) >> 3;
+
+    for (i = 1; i < 64; i++) {
+        int level = src[i];
+        if (((unsigned)(level + threshold1)) > threshold2) {
+            const int j = permutation[i];
+            if (level > 0) dst[j] = (level - threshold1 + 4) >> 3;
+            else           dst[j] = (level + threshold1 + 4) >> 3;
+        }
+    }
+}
+
+static void store_slice_c(uint8_t *dst, const int16_t *src,
+                          int dst_linesize, int src_linesize,
+                          int width, int height, int log2_scale,
+                          const uint8_t dither[8][8])
+{
+    int y, x;
+
+#define STORE(pos) do {                                                     \
+    temp = ((src[x + y*src_linesize + pos] << log2_scale) + d[pos]) >> 6;   \
+    if (temp & 0x100)                                                       \
+        temp = ~(temp >> 31);                                               \
+    dst[x + y*dst_linesize + pos] = temp;                                   \
+} while (0)
+
+    for (y = 0; y < height; y++) {
+        const uint8_t *d = dither[y];
+        for (x = 0; x < width; x += 8) {
+            int temp;
+            STORE(0);
+            STORE(1);
+            STORE(2);
+            STORE(3);
+            STORE(4);
+            STORE(5);
+            STORE(6);
+            STORE(7);
+        }
+    }
+}
+
+static inline void add_block(int16_t *dst, int linesize, const int16_t block[64])
+{
+    int y;
+
+    for (y = 0; y < 8; y++) {
+        *(uint32_t *)&dst[0 + y*linesize] += *(uint32_t *)&block[0 + y*8];
+        *(uint32_t *)&dst[2 + y*linesize] += *(uint32_t *)&block[2 + y*8];
+        *(uint32_t *)&dst[4 + y*linesize] += *(uint32_t *)&block[4 + y*8];
+        *(uint32_t *)&dst[6 + y*linesize] += *(uint32_t *)&block[6 + y*8];
+    }
+}
+
+// XXX: export the function?
+static inline int norm_qscale(int qscale, int type)
+{
+    switch (type) {
+    case FF_QSCALE_TYPE_MPEG1: return qscale;
+    case FF_QSCALE_TYPE_MPEG2: return qscale >> 1;
+    case FF_QSCALE_TYPE_H264:  return qscale >> 2;
+    case FF_QSCALE_TYPE_VP56:  return (63 - qscale + 2) >> 2;
+    }
+    return qscale;
+}
+
+static void filter(SPPContext *p, uint8_t *dst, uint8_t *src,
+                   int dst_linesize, int src_linesize, int width, int height,
+                   const uint8_t *qp_table, int qp_stride, int is_luma)
+{
+    int x, y, i;
+    const int count = 1 << p->log2_count;
+    const int linesize = is_luma ? p->temp_linesize : FFALIGN(width+16, 16);
+    DECLARE_ALIGNED(16, uint64_t, block_align)[32];
+    int16_t *block  = (int16_t *)block_align;
+    int16_t *block2 = (int16_t *)(block_align + 16);
+
+    for (y = 0; y < height; y++) {
+        int index = 8 + 8*linesize + y*linesize;
+        memcpy(p->src + index, src + y*src_linesize, width);
+        for (x = 0; x < 8; x++) {
+            p->src[index         - x - 1] = p->src[index +         x    ];
+            p->src[index + width + x    ] = p->src[index + width - x - 1];
+        }
+    }
+    for (y = 0; y < 8; y++) {
+        memcpy(p->src + (       7-y)*linesize, p->src + (       y+8)*linesize, linesize);
+        memcpy(p->src + (height+8+y)*linesize, p->src + (height-y+7)*linesize, linesize);
+    }
+
+    for (y = 0; y < height + 8; y += 8) {
+        memset(p->temp + (8 + y) * linesize, 0, 8 * linesize * sizeof(*p->temp));
+        for (x = 0; x < width + 8; x += 8) {
+            int qp;
+
+            if (p->qp) {
+                qp = p->qp;
+            } else{
+                const int qps = 3 + is_luma;
+                qp = qp_table[(FFMIN(x, width - 1) >> qps) + (FFMIN(y, height - 1) >> qps) * qp_stride];
+                qp = FFMAX(1, norm_qscale(qp, p->qscale_type));
+            }
+            for (i = 0; i < count; i++) {
+                const int x1 = x + offset[i + count - 1][0];
+                const int y1 = y + offset[i + count - 1][1];
+                const int index = x1 + y1*linesize;
+                p->dsp.get_pixels(block, p->src + index, linesize);
+                p->dsp.fdct(block);
+                p->requantize(block2, block, qp, p->dsp.idct_permutation);
+                p->dsp.idct(block2);
+                add_block(p->temp + index, linesize, block2);
+            }
+        }
+        if (y)
+            p->store_slice(dst + (y - 8) * dst_linesize, p->temp + 8 + y*linesize,
+                           dst_linesize, linesize, width,
+                           FFMIN(8, height + 8 - y), MAX_LEVEL - p->log2_count,
+                           ldither);
+    }
+}
+
+static int query_formats(AVFilterContext *ctx)
+{
+    static const enum PixelFormat pix_fmts[] = {
+        AV_PIX_FMT_YUV444P,  AV_PIX_FMT_YUV422P,
+        AV_PIX_FMT_YUV420P,  AV_PIX_FMT_YUV411P,
+        AV_PIX_FMT_YUV410P,  AV_PIX_FMT_YUV440P,
+        AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ422P,
+        AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ440P,
+        AV_PIX_FMT_NONE
+    };
+    ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
+    return 0;
+}
+
+static int config_input(AVFilterLink *inlink)
+{
+    SPPContext *spp = inlink->dst->priv;
+    const int h = FFALIGN(inlink->h + 16, 16);
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
+
+    spp->hsub = desc->log2_chroma_w;
+    spp->vsub = desc->log2_chroma_h;
+    spp->temp_linesize = FFALIGN(inlink->w + 16, 16);
+    spp->temp = av_malloc(spp->temp_linesize * h * sizeof(*spp->temp));
+    spp->src  = av_malloc(spp->temp_linesize * h * sizeof(*spp->src));
+    if (!spp->use_bframe_qp) {
+        /* we are assuming here the qp blocks will not be smaller that 16x16 */
+        spp->non_b_qp_alloc_size = FF_CEIL_RSHIFT(inlink->w, 4) * FF_CEIL_RSHIFT(inlink->h, 4);
+        spp->non_b_qp_table = av_calloc(spp->non_b_qp_alloc_size, sizeof(*spp->non_b_qp_table));
+        if (!spp->non_b_qp_table)
+            return AVERROR(ENOMEM);
+    }
+    if (!spp->temp || !spp->src)
+        return AVERROR(ENOMEM);
+    return 0;
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFrame *in)
+{
+    AVFilterContext *ctx = inlink->dst;
+    SPPContext *spp = ctx->priv;
+    AVFilterLink *outlink = ctx->outputs[0];
+    AVFrame *out = in;
+    int qp_stride = 0;
+    const int8_t *qp_table = NULL;
+
+    /* if we are not in a constant user quantizer mode and we don't want to use
+     * the quantizers from the B-frames (B-frames often have a higher QP), we
+     * need to save the qp table from the last non B-frame; this is what the
+     * following code block does */
+    if (!spp->qp) {
+        qp_table = av_frame_get_qp_table(in, &qp_stride, &spp->qscale_type);
+
+        if (qp_table && !spp->use_bframe_qp && in->pict_type != AV_PICTURE_TYPE_B) {
+            int w, h;
+
+            /* if the qp stride is not set, it means the QP are only defined on
+             * a line basis */
+            if (!qp_stride) {
+                w = FF_CEIL_RSHIFT(inlink->w, 4);
+                h = 1;
+            } else {
+                w = FF_CEIL_RSHIFT(qp_stride, 4);
+                h = FF_CEIL_RSHIFT(inlink->h, 4);
+            }
+            av_assert0(w * h <= spp->non_b_qp_alloc_size);
+            memcpy(spp->non_b_qp_table, qp_table, w * h);
+        }
+    }
+
+    if (spp->log2_count && !ctx->is_disabled) {
+        if (!spp->use_bframe_qp && spp->non_b_qp_table)
+            qp_table = spp->non_b_qp_table;
+
+        if (qp_table || spp->qp) {
+            const int cw = FF_CEIL_RSHIFT(inlink->w, spp->hsub);
+            const int ch = FF_CEIL_RSHIFT(inlink->h, spp->vsub);
+
+            /* get a new frame if in-place is not possible or if the dimensions
+             * are not multiple of 8 */
+            if (!av_frame_is_writable(in) || (inlink->w & 7) || (inlink->h & 7)) {
+                const int aligned_w = FFALIGN(inlink->w, 8);
+                const int aligned_h = FFALIGN(inlink->h, 8);
+
+                out = ff_get_video_buffer(outlink, aligned_w, aligned_h);
+                if (!out) {
+                    av_frame_free(&in);
+                    return AVERROR(ENOMEM);
+                }
+                av_frame_copy_props(out, in);
+                out->width  = in->width;
+                out->height = in->height;
+            }
+
+            filter(spp, out->data[0], in->data[0], out->linesize[0], in->linesize[0], inlink->w, inlink->h, qp_table, qp_stride, 1);
+            filter(spp, out->data[1], in->data[1], out->linesize[1], in->linesize[1], cw,        ch,        qp_table, qp_stride, 0);
+            filter(spp, out->data[2], in->data[2], out->linesize[2], in->linesize[2], cw,        ch,        qp_table, qp_stride, 0);
+            emms_c();
+        }
+    }
+
+    if (in != out) {
+        if (in->data[3])
+            av_image_copy_plane(out->data[3], out->linesize[3],
+                                in ->data[3], in ->linesize[3],
+                                inlink->w, inlink->h);
+        av_frame_free(&in);
+    }
+    return ff_filter_frame(outlink, out);
+}
+
+static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
+                           char *res, int res_len, int flags)
+{
+    SPPContext *spp = ctx->priv;
+
+    if (!strcmp(cmd, "level")) {
+        if (!strcmp(args, "max"))
+            spp->log2_count = MAX_LEVEL;
+        else
+            spp->log2_count = av_clip(strtol(args, NULL, 10), 0, MAX_LEVEL);
+        return 0;
+    }
+    return AVERROR(ENOSYS);
+}
+
+static av_cold int init(AVFilterContext *ctx)
+{
+    SPPContext *spp = ctx->priv;
+
+    spp->avctx = avcodec_alloc_context3(NULL);
+    if (!spp->avctx)
+        return AVERROR(ENOMEM);
+    avpriv_dsputil_init(&spp->dsp, spp->avctx);
+    spp->store_slice = store_slice_c;
+    switch (spp->mode) {
+    case MODE_HARD: spp->requantize = hardthresh_c; break;
+    case MODE_SOFT: spp->requantize = softthresh_c; break;
+    }
+    if (ARCH_X86)
+        ff_spp_init_x86(spp);
+    return 0;
+}
+
+static av_cold void uninit(AVFilterContext *ctx)
+{
+    SPPContext *spp = ctx->priv;
+
+    av_freep(&spp->temp);
+    av_freep(&spp->src);
+    if (spp->avctx) {
+        avcodec_close(spp->avctx);
+        av_freep(&spp->avctx);
+    }
+    av_freep(&spp->non_b_qp_table);
+}
+
+static const AVFilterPad spp_inputs[] = {
+    {
+        .name         = "default",
+        .type         = AVMEDIA_TYPE_VIDEO,
+        .config_props = config_input,
+        .filter_frame = filter_frame,
+    },
+    { NULL }
+};
+
+static const AVFilterPad spp_outputs[] = {
+    {
+        .name = "default",
+        .type = AVMEDIA_TYPE_VIDEO,
+    },
+    { NULL }
+};
+
+AVFilter avfilter_vf_spp = {
+    .name            = "spp",
+    .description     = NULL_IF_CONFIG_SMALL("XXX"),
+    .priv_size       = sizeof(SPPContext),
+    .init            = init,
+    .uninit          = uninit,
+    .query_formats   = query_formats,
+    .inputs          = spp_inputs,
+    .outputs         = spp_outputs,
+    .process_command = process_command,
+    .priv_class      = &spp_class,
+    .flags           = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
+};
diff --git a/libavfilter/vf_spp.h b/libavfilter/vf_spp.h
new file mode 100644 (file)
index 0000000..c4293b6
--- /dev/null
@@ -0,0 +1,59 @@
+/*
+ * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
+ * Copyright (c) 2013 Clément Bœsch
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU 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_SPP_H
+#define AVFILTER_SPP_H
+
+#include "libavcodec/avcodec.h"
+#include "libavcodec/dsputil.h"
+#include "avfilter.h"
+
+#define MAX_LEVEL 6 /* quality levels */
+
+typedef struct {
+    const AVClass *av_class;
+
+    int log2_count;
+    int qp;
+    int mode;
+    int qscale_type;
+    int temp_linesize;
+    uint8_t *src;
+    int16_t *temp;
+    AVCodecContext *avctx;
+    DSPContext dsp;
+    int8_t *non_b_qp_table;
+    int non_b_qp_alloc_size;
+    int use_bframe_qp;
+    int hsub, vsub;
+
+    void (*store_slice)(uint8_t *dst, const int16_t *src,
+                        int dst_stride, int src_stride,
+                        int width, int height, int log2_scale,
+                        const uint8_t dither[8][8]);
+
+    void (*requantize)(int16_t dst[64], const int16_t src[64],
+                       int qp, const uint8_t *permutation);
+} SPPContext;
+
+void ff_spp_init_x86(SPPContext *s);
+
+#endif /* AVFILTER_SPP_H */
index cd97347..0b19c4b 100644 (file)
@@ -1,5 +1,6 @@
 OBJS-$(CONFIG_GRADFUN_FILTER)                += x86/vf_gradfun.o
 OBJS-$(CONFIG_HQDN3D_FILTER)                 += x86/vf_hqdn3d_init.o
+OBJS-$(CONFIG_SPP_FILTER)                    += x86/vf_spp.o
 OBJS-$(CONFIG_VOLUME_FILTER)                 += x86/af_volume_init.o
 OBJS-$(CONFIG_YADIF_FILTER)                  += x86/vf_yadif_init.o
 
diff --git a/libavfilter/x86/vf_spp.c b/libavfilter/x86/vf_spp.c
new file mode 100644 (file)
index 0000000..eb46ddc
--- /dev/null
@@ -0,0 +1,233 @@
+/*
+ * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU 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 "libavfilter/vf_spp.h"
+
+#if HAVE_MMX_INLINE
+static void hardthresh_mmx(int16_t dst[64], const int16_t src[64],
+                           int qp, const uint8_t *permutation)
+{
+    int bias = 0; //FIXME
+    unsigned int threshold1;
+
+    threshold1 = qp * ((1<<4) - bias) - 1;
+
+#define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3)    \
+    "movq " #src0 ", %%mm0      \n"                                     \
+    "movq " #src1 ", %%mm1      \n"                                     \
+    "movq " #src2 ", %%mm2      \n"                                     \
+    "movq " #src3 ", %%mm3      \n"                                     \
+    "psubw %%mm4, %%mm0         \n"                                     \
+    "psubw %%mm4, %%mm1         \n"                                     \
+    "psubw %%mm4, %%mm2         \n"                                     \
+    "psubw %%mm4, %%mm3         \n"                                     \
+    "paddusw %%mm5, %%mm0       \n"                                     \
+    "paddusw %%mm5, %%mm1       \n"                                     \
+    "paddusw %%mm5, %%mm2       \n"                                     \
+    "paddusw %%mm5, %%mm3       \n"                                     \
+    "paddw %%mm6, %%mm0         \n"                                     \
+    "paddw %%mm6, %%mm1         \n"                                     \
+    "paddw %%mm6, %%mm2         \n"                                     \
+    "paddw %%mm6, %%mm3         \n"                                     \
+    "psubusw %%mm6, %%mm0       \n"                                     \
+    "psubusw %%mm6, %%mm1       \n"                                     \
+    "psubusw %%mm6, %%mm2       \n"                                     \
+    "psubusw %%mm6, %%mm3       \n"                                     \
+    "psraw $3, %%mm0            \n"                                     \
+    "psraw $3, %%mm1            \n"                                     \
+    "psraw $3, %%mm2            \n"                                     \
+    "psraw $3, %%mm3            \n"                                     \
+                                                                        \
+    "movq %%mm0, %%mm7          \n"                                     \
+    "punpcklwd %%mm2, %%mm0     \n" /*A*/                               \
+    "punpckhwd %%mm2, %%mm7     \n" /*C*/                               \
+    "movq %%mm1, %%mm2          \n"                                     \
+    "punpcklwd %%mm3, %%mm1     \n" /*B*/                               \
+    "punpckhwd %%mm3, %%mm2     \n" /*D*/                               \
+    "movq %%mm0, %%mm3          \n"                                     \
+    "punpcklwd %%mm1, %%mm0     \n" /*A*/                               \
+    "punpckhwd %%mm7, %%mm3     \n" /*C*/                               \
+    "punpcklwd %%mm2, %%mm7     \n" /*B*/                               \
+    "punpckhwd %%mm2, %%mm1     \n" /*D*/                               \
+                                                                        \
+    "movq %%mm0, " #dst0 "      \n"                                     \
+    "movq %%mm7, " #dst1 "      \n"                                     \
+    "movq %%mm3, " #dst2 "      \n"                                     \
+    "movq %%mm1, " #dst3 "      \n"
+
+    __asm__ volatile(
+        "movd %2, %%mm4             \n"
+        "movd %3, %%mm5             \n"
+        "movd %4, %%mm6             \n"
+        "packssdw %%mm4, %%mm4      \n"
+        "packssdw %%mm5, %%mm5      \n"
+        "packssdw %%mm6, %%mm6      \n"
+        "packssdw %%mm4, %%mm4      \n"
+        "packssdw %%mm5, %%mm5      \n"
+        "packssdw %%mm6, %%mm6      \n"
+        REQUANT_CORE(  (%1),  8(%1), 16(%1), 24(%1),  (%0), 8(%0), 64(%0), 72(%0))
+        REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0))
+        REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0))
+        REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0))
+        : : "r" (src), "r" (dst), "g" (threshold1+1), "g" (threshold1+5), "g" (threshold1-4) //FIXME maybe more accurate then needed?
+    );
+    dst[0] = (src[0] + 4) >> 3;
+}
+
+static void softthresh_mmx(int16_t dst[64], const int16_t src[64],
+                           int qp, const uint8_t *permutation)
+{
+    int bias = 0; //FIXME
+    unsigned int threshold1;
+
+    threshold1 = qp*((1<<4) - bias) - 1;
+
+#undef REQUANT_CORE
+#define REQUANT_CORE(dst0, dst1, dst2, dst3, src0, src1, src2, src3)    \
+    "movq " #src0 ", %%mm0      \n"                                     \
+    "movq " #src1 ", %%mm1      \n"                                     \
+    "pxor %%mm6, %%mm6          \n"                                     \
+    "pxor %%mm7, %%mm7          \n"                                     \
+    "pcmpgtw %%mm0, %%mm6       \n"                                     \
+    "pcmpgtw %%mm1, %%mm7       \n"                                     \
+    "pxor %%mm6, %%mm0          \n"                                     \
+    "pxor %%mm7, %%mm1          \n"                                     \
+    "psubusw %%mm4, %%mm0       \n"                                     \
+    "psubusw %%mm4, %%mm1       \n"                                     \
+    "pxor %%mm6, %%mm0          \n"                                     \
+    "pxor %%mm7, %%mm1          \n"                                     \
+    "movq " #src2 ", %%mm2      \n"                                     \
+    "movq " #src3 ", %%mm3      \n"                                     \
+    "pxor %%mm6, %%mm6          \n"                                     \
+    "pxor %%mm7, %%mm7          \n"                                     \
+    "pcmpgtw %%mm2, %%mm6       \n"                                     \
+    "pcmpgtw %%mm3, %%mm7       \n"                                     \
+    "pxor %%mm6, %%mm2          \n"                                     \
+    "pxor %%mm7, %%mm3          \n"                                     \
+    "psubusw %%mm4, %%mm2       \n"                                     \
+    "psubusw %%mm4, %%mm3       \n"                                     \
+    "pxor %%mm6, %%mm2          \n"                                     \
+    "pxor %%mm7, %%mm3          \n"                                     \
+                                                                        \
+    "paddsw %%mm5, %%mm0        \n"                                     \
+    "paddsw %%mm5, %%mm1        \n"                                     \
+    "paddsw %%mm5, %%mm2        \n"                                     \
+    "paddsw %%mm5, %%mm3        \n"                                     \
+    "psraw $3, %%mm0            \n"                                     \
+    "psraw $3, %%mm1            \n"                                     \
+    "psraw $3, %%mm2            \n"                                     \
+    "psraw $3, %%mm3            \n"                                     \
+                                                                        \
+    "movq %%mm0, %%mm7          \n"                                     \
+    "punpcklwd %%mm2, %%mm0     \n" /*A*/                               \
+    "punpckhwd %%mm2, %%mm7     \n" /*C*/                               \
+    "movq %%mm1, %%mm2          \n"                                     \
+    "punpcklwd %%mm3, %%mm1     \n" /*B*/                               \
+    "punpckhwd %%mm3, %%mm2     \n" /*D*/                               \
+    "movq %%mm0, %%mm3          \n"                                     \
+    "punpcklwd %%mm1, %%mm0     \n" /*A*/                               \
+    "punpckhwd %%mm7, %%mm3     \n" /*C*/                               \
+    "punpcklwd %%mm2, %%mm7     \n" /*B*/                               \
+    "punpckhwd %%mm2, %%mm1     \n" /*D*/                               \
+                                                                        \
+    "movq %%mm0, " #dst0 "      \n"                                     \
+    "movq %%mm7, " #dst1 "      \n"                                     \
+    "movq %%mm3, " #dst2 "      \n"                                     \
+    "movq %%mm1, " #dst3 "      \n"
+
+    __asm__ volatile(
+        "movd %2, %%mm4             \n"
+        "movd %3, %%mm5             \n"
+        "packssdw %%mm4, %%mm4      \n"
+        "packssdw %%mm5, %%mm5      \n"
+        "packssdw %%mm4, %%mm4      \n"
+        "packssdw %%mm5, %%mm5      \n"
+        REQUANT_CORE(  (%1),  8(%1), 16(%1), 24(%1),  (%0), 8(%0), 64(%0), 72(%0))
+        REQUANT_CORE(32(%1), 40(%1), 48(%1), 56(%1),16(%0),24(%0), 48(%0), 56(%0))
+        REQUANT_CORE(64(%1), 72(%1), 80(%1), 88(%1),32(%0),40(%0), 96(%0),104(%0))
+        REQUANT_CORE(96(%1),104(%1),112(%1),120(%1),80(%0),88(%0),112(%0),120(%0))
+        : : "r" (src), "r" (dst), "g" (threshold1), "rm" (4) //FIXME maybe more accurate then needed?
+    );
+
+    dst[0] = (src[0] + 4) >> 3;
+}
+
+static void store_slice_mmx(uint8_t *dst, const int16_t *src,
+                            int dst_stride, int src_stride,
+                            int width, int height, int log2_scale,
+                            const uint8_t dither[8][8])
+{
+    int y;
+
+    for (y = 0; y < height; y++) {
+        uint8_t *dst1 = dst;
+        const int16_t *src1 = src;
+        __asm__ volatile(
+            "movq (%3), %%mm3           \n"
+            "movq (%3), %%mm4           \n"
+            "movd %4, %%mm2             \n"
+            "pxor %%mm0, %%mm0          \n"
+            "punpcklbw %%mm0, %%mm3     \n"
+            "punpckhbw %%mm0, %%mm4     \n"
+            "psraw %%mm2, %%mm3         \n"
+            "psraw %%mm2, %%mm4         \n"
+            "movd %5, %%mm2             \n"
+            "1:                         \n"
+            "movq (%0), %%mm0           \n"
+            "movq 8(%0), %%mm1          \n"
+            "paddw %%mm3, %%mm0         \n"
+            "paddw %%mm4, %%mm1         \n"
+            "psraw %%mm2, %%mm0         \n"
+            "psraw %%mm2, %%mm1         \n"
+            "packuswb %%mm1, %%mm0      \n"
+            "movq %%mm0, (%1)           \n"
+            "add $16, %0                \n"
+            "add $8, %1                 \n"
+            "cmp %2, %1                 \n"
+            " jb 1b                     \n"
+            : "+r" (src1), "+r"(dst1)
+            : "r"(dst + width), "r"(dither[y]), "g"(log2_scale), "g"(MAX_LEVEL - log2_scale)
+        );
+        src += src_stride;
+        dst += dst_stride;
+    }
+}
+
+#endif /* HAVE_MMX_INLINE */
+
+av_cold void ff_spp_init_x86(SPPContext *s)
+{
+#if HAVE_MMX_INLINE
+    int cpu_flags = av_get_cpu_flags();
+
+    if (cpu_flags & AV_CPU_FLAG_MMX) {
+        s->store_slice = store_slice_mmx;
+        switch (s->mode) {
+        case 0: s->requantize = hardthresh_mmx; break;
+        case 1: s->requantize = softthresh_mmx; break;
+        }
+    }
+#endif
+}