X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fdsputil.c;h=80efd07825bf026558abc0f16052592002ddde17;hb=bbdfa06d437a1cb8dd18169bc0c3db129aaa25d2;hp=9e69bf7ae3c983ca229fe5370e8bf70ae0a688f7;hpb=4c79b95c32ec9d44af48a5a52c7390b84c52928c;p=ffmpeg

diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index 9e69bf7ae3c..80efd07825b 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -29,7 +29,6 @@
 
 #include "avcodec.h"
 #include "dsputil.h"
-#include "mpegvideo.h"
 #include "simple_idct.h"
 #include "faandct.h"
 #include "faanidct.h"
@@ -152,6 +151,8 @@ static const uint8_t simple_mmx_permutation[64]={
         0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
 };
 
+static const uint8_t idct_sse2_row_perm[8] = {0, 4, 1, 5, 2, 6, 3, 7};
+
 void ff_init_scantable(uint8_t *permutation, ScanTable *st, const uint8_t *src_scantable){
     int i;
     int end;
@@ -457,6 +458,77 @@ static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w)
     }
 }
 
+/**
+ * Copies a rectangular area of samples to a temporary buffer and replicates the boarder samples.
+ * @param buf destination buffer
+ * @param src source buffer
+ * @param linesize number of bytes between 2 vertically adjacent samples in both the source and destination buffers
+ * @param block_w width of block
+ * @param block_h height of block
+ * @param src_x x coordinate of the top left sample of the block in the source buffer
+ * @param src_y y coordinate of the top left sample of the block in the source buffer
+ * @param w width of the source buffer
+ * @param h height of the source buffer
+ */
+void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, int block_w, int block_h,
+                                    int src_x, int src_y, int w, int h){
+    int x, y;
+    int start_y, start_x, end_y, end_x;
+
+    if(src_y>= h){
+        src+= (h-1-src_y)*linesize;
+        src_y=h-1;
+    }else if(src_y<=-block_h){
+        src+= (1-block_h-src_y)*linesize;
+        src_y=1-block_h;
+    }
+    if(src_x>= w){
+        src+= (w-1-src_x);
+        src_x=w-1;
+    }else if(src_x<=-block_w){
+        src+= (1-block_w-src_x);
+        src_x=1-block_w;
+    }
+
+    start_y= FFMAX(0, -src_y);
+    start_x= FFMAX(0, -src_x);
+    end_y= FFMIN(block_h, h-src_y);
+    end_x= FFMIN(block_w, w-src_x);
+
+    // copy existing part
+    for(y=start_y; y<end_y; y++){
+        for(x=start_x; x<end_x; x++){
+            buf[x + y*linesize]= src[x + y*linesize];
+        }
+    }
+
+    //top
+    for(y=0; y<start_y; y++){
+        for(x=start_x; x<end_x; x++){
+            buf[x + y*linesize]= buf[x + start_y*linesize];
+        }
+    }
+
+    //bottom
+    for(y=end_y; y<block_h; y++){
+        for(x=start_x; x<end_x; x++){
+            buf[x + y*linesize]= buf[x + (end_y-1)*linesize];
+        }
+    }
+
+    for(y=0; y<block_h; y++){
+       //left
+        for(x=0; x<start_x; x++){
+            buf[x + y*linesize]= buf[start_x + y*linesize];
+        }
+
+       //right
+        for(x=end_x; x<block_w; x++){
+            buf[x + y*linesize]= buf[end_x - 1 + y*linesize];
+        }
+    }
+}
+
 static void get_pixels_c(DCTELEM *restrict block, const uint8_t *pixels, int line_size)
 {
     int i;
@@ -3626,7 +3698,7 @@ static int rd8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int
     DECLARE_ALIGNED_8 (uint64_t, aligned_bak[stride]);
     DCTELEM * const temp= (DCTELEM*)aligned_temp;
     uint8_t * const bak= (uint8_t*)aligned_bak;
-    int i, last, run, bits, level, distoration, start_i;
+    int i, last, run, bits, level, distortion, start_i;
     const int esc_length= s->ac_esc_length;
     uint8_t * length;
     uint8_t * last_length;
@@ -3693,9 +3765,9 @@ static int rd8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int
 
     s->dsp.idct_add(bak, stride, temp);
 
-    distoration= s->dsp.sse[1](NULL, bak, src1, stride, 8);
+    distortion= s->dsp.sse[1](NULL, bak, src1, stride, 8);
 
-    return distoration + ((bits*s->qscale*s->qscale*109 + 64)>>7);
+    return distortion + ((bits*s->qscale*s->qscale*109 + 64)>>7);
 }
 
 static int bit8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride, int h){
@@ -4406,6 +4478,10 @@ void dsputil_init(DSPContext* c, AVCodecContext *avctx)
         for(i=0; i<64; i++)
             c->idct_permutation[i]= (i&0x24) | ((i&3)<<3) | ((i>>3)&3);
         break;
+    case FF_SSE2_IDCT_PERM:
+        for(i=0; i<64; i++)
+            c->idct_permutation[i]= (i&0x38) | idct_sse2_row_perm[i&7];
+        break;
     default:
         av_log(avctx, AV_LOG_ERROR, "Internal error, IDCT permutation not set\n");
     }