lavfi: port mp=pp7 to libavfilter

The only difference with mp=pp7 is that default mode is "medium", as stated
in the MPlayer docs, rather than "hard".

Signed-off-by: Stefano Sabatini <stefasab@gmail.com>

7 files changed, 537 insertions, 1 deletions

diff --git a/libavfilter/Makefile b/libavfilter/Makefile
index 0ead6c5..3f8016a 100644
--- a/libavfilter/Makefile
+++ b/libavfilter/Makefile
@@ -164,6 +164,7 @@ OBJS-$(CONFIG_PERSPECTIVE_FILTER)            += vf_perspective.o
 OBJS-$(CONFIG_PHASE_FILTER)                  += vf_phase.o
 OBJS-$(CONFIG_PIXDESCTEST_FILTER)            += vf_pixdesctest.o
 OBJS-$(CONFIG_PP_FILTER)                     += vf_pp.o
+OBJS-$(CONFIG_PP7_FILTER)                    += vf_pp7.o
 OBJS-$(CONFIG_PSNR_FILTER)                   += vf_psnr.o dualinput.o framesync.o
 OBJS-$(CONFIG_PULLUP_FILTER)                 += vf_pullup.o
 OBJS-$(CONFIG_REMOVELOGO_FILTER)             += bbox.o lswsutils.o lavfutils.o vf_removelogo.o
diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c
index 6543629..a330eb9 100644
--- a/libavfilter/allfilters.c
+++ b/libavfilter/allfilters.c
@@ -179,6 +179,7 @@ void avfilter_register_all(void)
     REGISTER_FILTER(PHASE,          phase,          vf);
     REGISTER_FILTER(PIXDESCTEST,    pixdesctest,    vf);
     REGISTER_FILTER(PP,             pp,             vf);
+    REGISTER_FILTER(PP7,            pp7,            vf);
     REGISTER_FILTER(PSNR,           psnr,           vf);
     REGISTER_FILTER(PULLUP,         pullup,         vf);
     REGISTER_FILTER(REMOVELOGO,     removelogo,     vf);
diff --git a/libavfilter/version.h b/libavfilter/version.h
index 50967f4..b931aa9 100644
--- a/libavfilter/version.h
+++ b/libavfilter/version.h
@@ -30,7 +30,7 @@
 #include "libavutil/version.h"
 
 #define LIBAVFILTER_VERSION_MAJOR  5
-#define LIBAVFILTER_VERSION_MINOR  6
+#define LIBAVFILTER_VERSION_MINOR  7
 #define LIBAVFILTER_VERSION_MICRO 100
 
 #define LIBAVFILTER_VERSION_INT AV_VERSION_INT(LIBAVFILTER_VERSION_MAJOR, \
diff --git a/libavfilter/vf_pp7.c b/libavfilter/vf_pp7.c
new file mode 100644
index 0000000..2f071f2
--- /dev/null
+++ b/libavfilter/vf_pp7.c
@@ -0,0 +1,419 @@
+/*
+ * Copyright (c) 2005 Michael Niedermayer <michaelni@gmx.at>
+ * Copyright (c) 2014 Arwa Arif <arwaarif1994@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
+ * Postprocessing filter - 7
+ *
+ * Originally written by Michael Niedermayer for the MPlayer
+ * project, and ported by Arwa Arif for FFmpeg.
+ */
+
+#include "libavutil/avassert.h"
+#include "libavutil/imgutils.h"
+#include "libavutil/opt.h"
+#include "libavutil/pixdesc.h"
+#include "internal.h"
+#include "vf_pp7.h"
+#include "libavcodec/avcodec.h"
+
+enum mode {
+    MODE_HARD,
+    MODE_SOFT,
+    MODE_MEDIUM
+};
+
+#define OFFSET(x) offsetof(PP7Context, x)
+#define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
+static const AVOption pp7_options[] = {
+    { "qp", "force a constant quantizer parameter", OFFSET(qp), AV_OPT_TYPE_INT, {.i64 = 0}, 0, 64, FLAGS },
+    { "mode", "set thresholding mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64 = MODE_MEDIUM}, 0, 2, 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" },
+        { "medium", "medium thresholding", 0, AV_OPT_TYPE_CONST, {.i64 = MODE_MEDIUM}, INT_MIN, INT_MAX, FLAGS, "mode" },
+    { NULL }
+};
+
+AVFILTER_DEFINE_CLASS(pp7);
+
+DECLARE_ALIGNED(8, static const uint8_t, dither)[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, },
+};
+
+#define N0 4
+#define N1 5
+#define N2 10
+#define SN0 2
+#define SN1 2.2360679775
+#define SN2 3.16227766017
+#define N (1 << 16)
+
+static const int factor[16] = {
+    N / (N0 * N0), N / (N0 * N1), N / (N0 * N0), N / (N0 * N2),
+    N / (N1 * N0), N / (N1 * N1), N / (N1 * N0), N / (N1 * N2),
+    N / (N0 * N0), N / (N0 * N1), N / (N0 * N0), N / (N0 * N2),
+    N / (N2 * N0), N / (N2 * N1), N / (N2 * N0), N / (N2 * N2),
+};
+
+static const int thres[16] = {
+    N / (SN0 * SN0), N / (SN0 * SN2), N / (SN0 * SN0), N / (SN0 * SN2),
+    N / (SN2 * SN0), N / (SN2 * SN2), N / (SN2 * SN0), N / (SN2 * SN2),
+    N / (SN0 * SN0), N / (SN0 * SN2), N / (SN0 * SN0), N / (SN0 * SN2),
+    N / (SN2 * SN0), N / (SN2 * SN2), N / (SN2 * SN0), N / (SN2 * SN2),
+};
+
+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 init_thres2(PP7Context *p)
+{
+    int qp, i;
+    int bias = 0; //FIXME
+
+    for (qp = 0; qp < 99; qp++) {
+        for (i = 0; i < 16; i++) {
+            p->thres2[qp][i] = ((i&1) ? SN2 : SN0) * ((i&4) ? SN2 : SN0) * FFMAX(1, qp) * (1<<2) - 1 - bias;
+        }
+    }
+}
+
+static inline void dctA_c(int16_t *dst, uint8_t *src, int stride)
+{
+    int i;
+
+    for (i = 0; i < 4; i++) {
+        int s0 = src[0 * stride] + src[6 * stride];
+        int s1 = src[1 * stride] + src[5 * stride];
+        int s2 = src[2 * stride] + src[4 * stride];
+        int s3 = src[3 * stride];
+        int s = s3 + s3;
+        s3 = s  - s0;
+        s0 = s  + s0;
+        s  = s2 + s1;
+        s2 = s2 - s1;
+        dst[0] = s0 + s;
+        dst[2] = s0 - s;
+        dst[1] = 2 * s3 +     s2;
+        dst[3] =     s3 - 2 * s2;
+        src++;
+        dst += 4;
+    }
+}
+
+static void dctB_c(int16_t *dst, int16_t *src)
+{
+    int i;
+
+    for (i = 0; i < 4; i++) {
+        int s0 = src[0 * 4] + src[6 * 4];
+        int s1 = src[1 * 4] + src[5 * 4];
+        int s2 = src[2 * 4] + src[4 * 4];
+        int s3 = src[3 * 4];
+        int s = s3 + s3;
+        s3 = s  - s0;
+        s0 = s  + s0;
+        s  = s2 + s1;
+        s2 = s2 - s1;
+        dst[0 * 4] = s0 + s;
+        dst[2 * 4] = s0 - s;
+        dst[1 * 4] = 2 * s3 +     s2;
+        dst[3 * 4] =     s3 - 2 * s2;
+        src++;
+        dst++;
+    }
+}
+
+static int hardthresh_c(PP7Context *p, int16_t *src, int qp)
+{
+    int i;
+    int a;
+
+    a = src[0] * factor[0];
+    for (i = 1; i < 16; i++) {
+        unsigned int threshold1 = p->thres2[qp][i];
+        unsigned int threshold2 = threshold1 << 1;
+        int level = src[i];
+        if (((unsigned)(level + threshold1)) > threshold2)
+            a += level * factor[i];
+    }
+    return (a + (1 << 11)) >> 12;
+}
+
+static int mediumthresh_c(PP7Context *p, int16_t *src, int qp)
+{
+    int i;
+    int a;
+
+    a = src[0] * factor[0];
+    for (i = 1; i < 16; i++) {
+        unsigned int threshold1 = p->thres2[qp][i];
+        unsigned int threshold2 = threshold1 << 1;
+        int level = src[i];
+        if (((unsigned)(level + threshold1)) > threshold2) {
+            if (((unsigned)(level + 2 * threshold1)) > 2 * threshold2)
+                a += level * factor[i];
+            else {
+                if (level > 0)
+                    a += 2 * (level - (int)threshold1) * factor[i];
+                else
+                    a += 2 * (level + (int)threshold1) * factor[i];
+            }
+        }
+    }
+    return (a + (1 << 11)) >> 12;
+}
+
+static int softthresh_c(PP7Context *p, int16_t *src, int qp)
+{
+    int i;
+    int a;
+
+    a = src[0] * factor[0];
+    for (i = 1; i < 16; i++) {
+        unsigned int threshold1 = p->thres2[qp][i];
+        unsigned int threshold2 = threshold1 << 1;
+        int level = src[i];
+        if (((unsigned)(level + threshold1)) > threshold2) {
+            if (level > 0)
+                a += (level - (int)threshold1) * factor[i];
+            else
+                a += (level + (int)threshold1) * factor[i];
+        }
+    }
+    return (a + (1 << 11)) >> 12;
+}
+
+static void filter(PP7Context *p, uint8_t *dst, uint8_t *src,
+                   int dst_stride, int src_stride,
+                   int width, int height,
+                   uint8_t *qp_store, int qp_stride, int is_luma)
+{
+    int x, y;
+    const int stride = is_luma ? p->temp_stride : ((width + 16 + 15) & (~15));
+    uint8_t *p_src = p->src + 8 * stride;
+    int16_t *block = (int16_t *)p->src;
+    int16_t *temp  = (int16_t *)(p->src + 32);
+
+    if (!src || !dst) return;
+    for (y = 0; y < height; y++) {
+        int index = 8 + 8 * stride + y * stride;
+        memcpy(p_src + index, src + y * src_stride, 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     ) * stride, p_src + (    y + 8     ) * stride, stride);
+        memcpy(p_src + (height + 8 + y) * stride, p_src + (height - y + 7) * stride, stride);
+    }
+    //FIXME (try edge emu)
+
+    for (y = 0; y < height; y++) {
+        for (x = -8; x < 0; x += 4) {
+            const int index = x + y * stride + (8 - 3) * (1 + stride) + 8; //FIXME silly offset
+            uint8_t *src  = p_src + index;
+            int16_t *tp   = temp + 4 * x;
+
+            dctA_c(tp + 4 * 8, src, stride);
+        }
+        for (x = 0; x < width; ) {
+            const int qps = 3 + is_luma;
+            int qp;
+            int end = FFMIN(x + 8, width);
+
+            if (p->qp)
+                qp = p->qp;
+            else {
+                qp = qp_store[ (FFMIN(x, width - 1) >> qps) + (FFMIN(y, height - 1) >> qps) * qp_stride];
+                qp = norm_qscale(qp, p->qscale_type);
+            }
+            for (; x < end; x++) {
+                const int index = x + y * stride + (8 - 3) * (1 + stride) + 8; //FIXME silly offset
+                uint8_t *src = p_src + index;
+                int16_t *tp  = temp + 4 * x;
+                int v;
+
+                if ((x & 3) == 0)
+                    dctA_c(tp + 4 * 8, src, stride);
+
+                p->dctB(block, tp);
+
+                v = p->requantize(p, block, qp);
+                v = (v + dither[y & 7][x & 7]) >> 6;
+                if ((unsigned)v > 255)
+                    v = (-v) >> 31;
+                dst[x + y * dst_stride] = v;
+            }
+        }
+    }
+}
+
+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_GRAY8,    AV_PIX_FMT_NONE
+    };
+    ff_set_common_formats(ctx, ff_make_format_list(pix_fmts));
+    return 0;
+}
+
+static int config_input(AVFilterLink *inlink)
+{
+    AVFilterContext *ctx = inlink->dst;
+    PP7Context *pp7 = ctx->priv;
+    const int h = FFALIGN(inlink->h + 16, 16);
+    const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
+
+    pp7->hsub = desc->log2_chroma_w;
+    pp7->vsub = desc->log2_chroma_h;
+
+    pp7->temp_stride = FFALIGN(inlink->w + 16, 16);
+    pp7->src = av_malloc(pp7->temp_stride * (h + 8) * sizeof(uint8_t));
+
+    if (!pp7->src)
+        return AVERROR(ENOMEM);
+
+    init_thres2(pp7);
+
+    switch (pp7->mode) {
+        case 0: pp7->requantize = hardthresh_c; break;
+        case 1: pp7->requantize = softthresh_c; break;
+        default:
+        case 2: pp7->requantize = mediumthresh_c; break;
+    }
+
+    pp7->dctB = dctB_c;
+
+    if (ARCH_X86)
+        ff_pp7_init_x86(pp7);
+
+    return 0;
+}
+
+static int filter_frame(AVFilterLink *inlink, AVFrame *in)
+{
+    AVFilterContext *ctx = inlink->dst;
+    PP7Context *pp7 = ctx->priv;
+    AVFilterLink *outlink = ctx->outputs[0];
+    AVFrame *out = in;
+
+    int qp_stride = 0;
+    uint8_t *qp_table = NULL;
+
+    if (!pp7->qp)
+        qp_table = av_frame_get_qp_table(in, &qp_stride, &pp7->qscale_type);
+
+    if (!ctx->is_disabled) {
+        const int cw = FF_CEIL_RSHIFT(inlink->w, pp7->hsub);
+        const int ch = FF_CEIL_RSHIFT(inlink->h, pp7->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);
+        }
+
+        if (qp_table || pp7->qp) {
+
+            filter(pp7, out->data[0], in->data[0], out->linesize[0], in->linesize[0],
+                   inlink->w, inlink->h, qp_table, qp_stride, 1);
+            filter(pp7, out->data[1], in->data[1], out->linesize[1], in->linesize[1],
+                   cw,        ch,        qp_table, qp_stride, 0);
+            filter(pp7, 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 av_cold void uninit(AVFilterContext *ctx)
+{
+    PP7Context *pp7 = ctx->priv;
+    av_freep(&pp7->src);
+}
+
+static const AVFilterPad pp7_inputs[] = {
+    {
+        .name         = "default",
+        .type         = AVMEDIA_TYPE_VIDEO,
+        .config_props = config_input,
+        .filter_frame = filter_frame,
+    },
+    { NULL }
+};
+
+static const AVFilterPad pp7_outputs[] = {
+    {
+        .name = "default",
+        .type = AVMEDIA_TYPE_VIDEO,
+    },
+    { NULL }
+};
+
+AVFilter ff_vf_pp7 = {
+    .name            = "pp7",
+    .description     = NULL_IF_CONFIG_SMALL("Apply Postprocessing 7 filter."),
+    .priv_size       = sizeof(PP7Context),
+    .uninit          = uninit,
+    .query_formats   = query_formats,
+    .inputs          = pp7_inputs,
+    .outputs         = pp7_outputs,
+    .priv_class      = &pp7_class,
+    .flags           = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
+};
diff --git a/libavfilter/vf_pp7.h b/libavfilter/vf_pp7.h
new file mode 100644
index 0000000..9aa8d73
--- /dev/null
+++ b/libavfilter/vf_pp7.h
@@ -0,0 +1,46 @@
+/*
+ * Copyright (c) 2005 Michael Niedermayer <michaelni@gmx.at>
+ * Copyright (c) 2014 Arwa Arif <arwaarif1994@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.
+ */
+
+#ifndef AVFILTER_PP7_H
+#define AVFILTER_PP7_H
+
+#include "avfilter.h"
+
+typedef struct PP7Context {
+    AVClass *class;
+    int thres2[99][16];
+
+    int qp;
+    int mode;
+    int qscale_type;
+    int hsub;
+    int vsub;
+    int temp_stride;
+    uint8_t *src;
+
+    int (*requantize)(struct PP7Context *p, int16_t *src, int qp);
+    void (*dctB)(int16_t *dst, int16_t *src);
+
+} PP7Context;
+
+void ff_pp7_init_x86(PP7Context *pp7);
+
+#endif /* AVFILTER_PP7_H */
diff --git a/libavfilter/x86/Makefile b/libavfilter/x86/Makefile
index d9265c9..a472a72 100644
--- a/libavfilter/x86/Makefile
+++ b/libavfilter/x86/Makefile
@@ -4,6 +4,7 @@ OBJS-$(CONFIG_HQDN3D_FILTER)                 += x86/vf_hqdn3d_init.o
 OBJS-$(CONFIG_IDET_FILTER)                   += x86/vf_idet_init.o
 OBJS-$(CONFIG_INTERLACE_FILTER)              += x86/vf_interlace_init.o
 OBJS-$(CONFIG_NOISE_FILTER)                  += x86/vf_noise.o
+OBJS-$(CONFIG_PP7_FILTER)                    += x86/vf_pp7.o
 OBJS-$(CONFIG_PULLUP_FILTER)                 += x86/vf_pullup_init.o
 OBJS-$(CONFIG_SPP_FILTER)                    += x86/vf_spp.o
 OBJS-$(CONFIG_TINTERLACE_FILTER)             += x86/vf_tinterlace_init.o
diff --git a/libavfilter/x86/vf_pp7.c b/libavfilter/x86/vf_pp7.c
new file mode 100644
index 0000000..64c8618
--- /dev/null
+++ b/libavfilter/x86/vf_pp7.c
@@ -0,0 +1,68 @@
+/*
+ * Copyright (c) 2005 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_pp7.h"
+
+static void dctB_mmx(int16_t *dst, int16_t *src)
+{
+#if HAVE_MMX_INLINE
+    __asm__ volatile (
+        "movq  (%0), %%mm0      \n\t"
+        "movq  1*4*2(%0), %%mm1 \n\t"
+        "paddw 6*4*2(%0), %%mm0 \n\t"
+        "paddw 5*4*2(%0), %%mm1 \n\t"
+        "movq  2*4*2(%0), %%mm2 \n\t"
+        "movq  3*4*2(%0), %%mm3 \n\t"
+        "paddw 4*4*2(%0), %%mm2 \n\t"
+        "paddw %%mm3, %%mm3     \n\t" //s
+        "movq %%mm3, %%mm4      \n\t" //s
+        "psubw %%mm0, %%mm3     \n\t" //s-s0
+        "paddw %%mm0, %%mm4     \n\t" //s+s0
+        "movq %%mm2, %%mm0      \n\t" //s2
+        "psubw %%mm1, %%mm2     \n\t" //s2-s1
+        "paddw %%mm1, %%mm0     \n\t" //s2+s1
+        "movq %%mm4, %%mm1      \n\t" //s0'
+        "psubw %%mm0, %%mm4     \n\t" //s0'-s'
+        "paddw %%mm0, %%mm1     \n\t" //s0'+s'
+        "movq %%mm3, %%mm0      \n\t" //s3'
+        "psubw %%mm2, %%mm3     \n\t"
+        "psubw %%mm2, %%mm3     \n\t"
+        "paddw %%mm0, %%mm2     \n\t"
+        "paddw %%mm0, %%mm2     \n\t"
+        "movq %%mm1, (%1)       \n\t"
+        "movq %%mm4, 2*4*2(%1)  \n\t"
+        "movq %%mm2, 1*4*2(%1)  \n\t"
+        "movq %%mm3, 3*4*2(%1)  \n\t"
+        :: "r" (src), "r"(dst)
+    );
+#endif
+}
+
+av_cold void ff_pp7_init_x86(PP7Context *p)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (HAVE_MMX_INLINE && cpu_flags & AV_CPU_FLAG_MMX)
+        p->dctB = dctB_mmx;
+}