From: Ronald S. Bultje <rsbultje@gmail.com>
Date: Thu, 29 Mar 2012 16:29:03 +0000 (-0700)
Subject: truemotion2: handle out-of-frame motion vectors through edge extension.
X-Git-Tag: v9_beta1~2167
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=bf39d3b59d85e5734babe48b61b8d92d18188185;p=platform%2Fupstream%2Flibav.git

truemotion2: handle out-of-frame motion vectors through edge extension.

Found-by: Mateusz "j00ru" Jurczyk and Gynvael Coldwind
CC: libav-stable@libav.org
---

diff --git a/libavcodec/truemotion2.c b/libavcodec/truemotion2.c
index a2170fc..646c222 100644
--- a/libavcodec/truemotion2.c
+++ b/libavcodec/truemotion2.c
@@ -57,7 +57,9 @@ typedef struct TM2Context{
     int *clast;
 
     /* data for current and previous frame */
+    int *Y1_base, *U1_base, *V1_base, *Y2_base, *U2_base, *V2_base;
     int *Y1, *U1, *V1, *Y2, *U2, *V2;
+    int y_stride, uv_stride;
     int cur;
 } TM2Context;
 
@@ -348,9 +350,9 @@ static inline int GET_TOK(TM2Context *ctx,int type) {
     int *Y, *U, *V;\
     int Ystride, Ustride, Vstride;\
 \
-    Ystride = ctx->avctx->width;\
-    Vstride = (ctx->avctx->width + 1) >> 1;\
-    Ustride = (ctx->avctx->width + 1) >> 1;\
+    Ystride = ctx->y_stride;\
+    Vstride = ctx->uv_stride;\
+    Ustride = ctx->uv_stride;\
     Y = (ctx->cur?ctx->Y2:ctx->Y1) + by * 4 * Ystride + bx * 4;\
     V = (ctx->cur?ctx->V2:ctx->V1) + by * 2 * Vstride + bx * 2;\
     U = (ctx->cur?ctx->U2:ctx->U1) + by * 2 * Ustride + bx * 2;\
@@ -638,6 +640,8 @@ static inline void tm2_motion_block(TM2Context *ctx, AVFrame *pic, int bx, int b
 
     mx = GET_TOK(ctx, TM2_MOT);
     my = GET_TOK(ctx, TM2_MOT);
+    mx = av_clip(mx, -(bx * 4 + 4), ctx->avctx->width  - bx * 4);
+    my = av_clip(my, -(by * 4 + 4), ctx->avctx->height - by * 4);
 
     Yo += my * oYstride + mx;
     Uo += (my >> 1) * oUstride + (mx >> 1);
@@ -678,15 +682,12 @@ static inline void tm2_motion_block(TM2Context *ctx, AVFrame *pic, int bx, int b
 static int tm2_decode_blocks(TM2Context *ctx, AVFrame *p)
 {
     int i, j;
-    int bw, bh;
+    int w = ctx->avctx->width, h = ctx->avctx->height, bw = w >> 2, bh = h >> 2, cw = w >> 1;
     int type;
     int keyframe = 1;
     int *Y, *U, *V;
     uint8_t *dst;
 
-    bw = ctx->avctx->width >> 2;
-    bh = ctx->avctx->height >> 2;
-
     for(i = 0; i < TM2_NUM_STREAMS; i++)
         ctx->tok_ptrs[i] = 0;
 
@@ -739,17 +740,54 @@ static int tm2_decode_blocks(TM2Context *ctx, AVFrame *p)
     U = (ctx->cur?ctx->U2:ctx->U1);
     V = (ctx->cur?ctx->V2:ctx->V1);
     dst = p->data[0];
-    for(j = 0; j < ctx->avctx->height; j++){
-        for(i = 0; i < ctx->avctx->width; i++){
+    for(j = 0; j < h; j++){
+        for(i = 0; i < w; i++){
             int y = Y[i], u = U[i >> 1], v = V[i >> 1];
             dst[3*i+0] = av_clip_uint8(y + v);
             dst[3*i+1] = av_clip_uint8(y);
             dst[3*i+2] = av_clip_uint8(y + u);
         }
-        Y += ctx->avctx->width;
+
+        /* horizontal edge extension */
+        Y[-4]    = Y[-3]    = Y[-2]    = Y[-1] = Y[0];
+        Y[w + 3] = Y[w + 2] = Y[w + 1] = Y[w]  = Y[w - 1];
+
+        /* vertical edge extension */
+        if (j == 0) {
+            memcpy(Y - 4 - 1 * ctx->y_stride, Y - 4, ctx->y_stride);
+            memcpy(Y - 4 - 2 * ctx->y_stride, Y - 4, ctx->y_stride);
+            memcpy(Y - 4 - 3 * ctx->y_stride, Y - 4, ctx->y_stride);
+            memcpy(Y - 4 - 4 * ctx->y_stride, Y - 4, ctx->y_stride);
+        } else if (j == h - 1) {
+            memcpy(Y - 4 + 1 * ctx->y_stride, Y - 4, ctx->y_stride);
+            memcpy(Y - 4 + 2 * ctx->y_stride, Y - 4, ctx->y_stride);
+            memcpy(Y - 4 + 3 * ctx->y_stride, Y - 4, ctx->y_stride);
+            memcpy(Y - 4 + 4 * ctx->y_stride, Y - 4, ctx->y_stride);
+        }
+
+        Y += ctx->y_stride;
         if (j & 1) {
-            U += ctx->avctx->width >> 1;
-            V += ctx->avctx->width >> 1;
+            /* horizontal edge extension */
+            U[-2]     = U[-1] = U[0];
+            V[-2]     = V[-1] = V[0];
+            U[cw + 1] = U[cw] = U[cw - 1];
+            V[cw + 1] = V[cw] = V[cw - 1];
+
+            /* vertical edge extension */
+            if (j == 1) {
+                memcpy(U - 2 - 1 * ctx->uv_stride, U - 2, ctx->uv_stride);
+                memcpy(V - 2 - 1 * ctx->uv_stride, V - 2, ctx->uv_stride);
+                memcpy(U - 2 - 2 * ctx->uv_stride, U - 2, ctx->uv_stride);
+                memcpy(V - 2 - 2 * ctx->uv_stride, V - 2, ctx->uv_stride);
+            } else if (j == h - 1) {
+                memcpy(U - 2 + 1 * ctx->uv_stride, U - 2, ctx->uv_stride);
+                memcpy(V - 2 + 1 * ctx->uv_stride, V - 2, ctx->uv_stride);
+                memcpy(U - 2 + 2 * ctx->uv_stride, U - 2, ctx->uv_stride);
+                memcpy(V - 2 + 2 * ctx->uv_stride, V - 2, ctx->uv_stride);
+            }
+
+            U += ctx->uv_stride;
+            V += ctx->uv_stride;
         }
         dst += p->linesize[0];
     }
@@ -821,7 +859,7 @@ static int decode_frame(AVCodecContext *avctx,
 
 static av_cold int decode_init(AVCodecContext *avctx){
     TM2Context * const l = avctx->priv_data;
-    int i;
+    int i, w = avctx->width, h = avctx->height;
 
     if((avctx->width & 3) || (avctx->height & 3)){
         av_log(avctx, AV_LOG_ERROR, "Width and height must be multiple of 4\n");
@@ -834,21 +872,46 @@ static av_cold int decode_init(AVCodecContext *avctx){
 
     ff_dsputil_init(&l->dsp, avctx);
 
-    l->last = av_malloc(4 * sizeof(int) * (avctx->width >> 2));
-    l->clast = av_malloc(4 * sizeof(int) * (avctx->width >> 2));
+    l->last  = av_malloc(4 * sizeof(*l->last)  * (w >> 2));
+    l->clast = av_malloc(4 * sizeof(*l->clast) * (w >> 2));
 
     for(i = 0; i < TM2_NUM_STREAMS; i++) {
         l->tokens[i] = NULL;
         l->tok_lens[i] = 0;
     }
 
-    l->Y1 = av_malloc(sizeof(int) * avctx->width * avctx->height);
-    l->U1 = av_malloc(sizeof(int) * ((avctx->width + 1) >> 1) * ((avctx->height + 1) >> 1));
-    l->V1 = av_malloc(sizeof(int) * ((avctx->width + 1) >> 1) * ((avctx->height + 1) >> 1));
-    l->Y2 = av_malloc(sizeof(int) * avctx->width * avctx->height);
-    l->U2 = av_malloc(sizeof(int) * ((avctx->width + 1) >> 1) * ((avctx->height + 1) >> 1));
-    l->V2 = av_malloc(sizeof(int) * ((avctx->width + 1) >> 1) * ((avctx->height + 1) >> 1));
+    w += 8;
+    h += 8;
+    l->Y1_base = av_malloc(sizeof(*l->Y1_base) * w * h);
+    l->Y2_base = av_malloc(sizeof(*l->Y2_base) * w * h);
+    l->y_stride = w;
+    w = (w + 1) >> 1;
+    h = (h + 1) >> 1;
+    l->U1_base = av_malloc(sizeof(*l->U1_base) * w * h);
+    l->V1_base = av_malloc(sizeof(*l->V1_base) * w * h);
+    l->U2_base = av_malloc(sizeof(*l->U2_base) * w * h);
+    l->V2_base = av_malloc(sizeof(*l->V1_base) * w * h);
+    l->uv_stride = w;
     l->cur = 0;
+    if (!l->Y1_base || !l->Y2_base || !l->U1_base ||
+        !l->V1_base || !l->U2_base || !l->V2_base ||
+        !l->last    || !l->clast) {
+        av_freep(l->Y1_base);
+        av_freep(l->Y2_base);
+        av_freep(l->U1_base);
+        av_freep(l->U2_base);
+        av_freep(l->V1_base);
+        av_freep(l->V2_base);
+        av_freep(l->last);
+        av_freep(l->clast);
+        return AVERROR(ENOMEM);
+    }
+    l->Y1 = l->Y1_base + l->y_stride  * 4 + 4;
+    l->Y2 = l->Y2_base + l->y_stride  * 4 + 4;
+    l->U1 = l->U1_base + l->uv_stride * 2 + 2;
+    l->U2 = l->U2_base + l->uv_stride * 2 + 2;
+    l->V1 = l->V1_base + l->uv_stride * 2 + 2;
+    l->V2 = l->V2_base + l->uv_stride * 2 + 2;
 
     return 0;
 }
@@ -863,12 +926,12 @@ static av_cold int decode_end(AVCodecContext *avctx){
     for(i = 0; i < TM2_NUM_STREAMS; i++)
         av_free(l->tokens[i]);
     if(l->Y1){
-        av_free(l->Y1);
-        av_free(l->U1);
-        av_free(l->V1);
-        av_free(l->Y2);
-        av_free(l->U2);
-        av_free(l->V2);
+        av_free(l->Y1_base);
+        av_free(l->U1_base);
+        av_free(l->V1_base);
+        av_free(l->Y2_base);
+        av_free(l->U2_base);
+        av_free(l->V2_base);
     }
 
     if (pic->data[0])