ffmpeg roll of source to mar 9 version...

patched-ffmpeg-mt/libswscale/swscale_template.c

 /*
  * Copyright (C) 2001-2003 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.
  *
@@ skipping 2633 matching lines @@
     lastDstY= dstY;
 
     for (;dstY < dstH; dstY++) {
         unsigned char *dest =dst[0]+dstStride[0]*dstY;
         const int chrDstY= dstY>>c->chrDstVSubSample;
         unsigned char *uDest=dst[1]+dstStride[1]*chrDstY;
         unsigned char *vDest=dst[2]+dstStride[2]*chrDstY;
         unsigned char *aDest=(CONFIG_SWSCALE_ALPHA && alpPixBuf) ? dst[3]+dstStride[3]*dstY : NULL;
 
         const int firstLumSrcY= vLumFilterPos[dstY]; //First line needed as input
+        const int firstLumSrcY2= vLumFilterPos[FFMIN(dstY | ((1<<c->chrDstVSubSample) - 1), dstH-1)];
         const int firstChrSrcY= vChrFilterPos[chrDstY]; //First line needed as input
         int lastLumSrcY= firstLumSrcY + vLumFilterSize -1; // Last line needed as input
+        int lastLumSrcY2=firstLumSrcY2+ vLumFilterSize -1; // Last line needed as input
         int lastChrSrcY= firstChrSrcY + vChrFilterSize -1; // Last line needed as input
         int enough_lines;
 
         //handle holes (FAST_BILINEAR & weird filters)
         if (firstLumSrcY > lastInLumBuf) lastInLumBuf= firstLumSrcY-1;
         if (firstChrSrcY > lastInChrBuf) lastInChrBuf= firstChrSrcY-1;
         assert(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);
         assert(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);
 
-        // Do we have enough lines in this slice to output the dstY line
-        enough_lines = lastLumSrcY < srcSliceY + srcSliceH && lastChrSrcY < -((-srcSliceY - srcSliceH)>>c->chrSrcVSubSample);
-        if (!enough_lines) {
-            lastLumSrcY = srcSliceY + srcSliceH - 1;
-            lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
-        }
-
         DEBUG_BUFFERS("dstY: %d\n", dstY);
         DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
                          firstLumSrcY,    lastLumSrcY,    lastInLumBuf);
         DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
                          firstChrSrcY,    lastChrSrcY,    lastInChrBuf);
 
+        // Do we have enough lines in this slice to output the dstY line
+        enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH && lastChrSrcY < -((-srcSliceY - srcSliceH)>>c->chrSrcVSubSample);
+
+        if (!enough_lines) {
+            lastLumSrcY = srcSliceY + srcSliceH - 1;
+            lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
+            DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
+                                            lastLumSrcY, lastChrSrcY);
+        }
+
         //Do horizontal scaling
         while(lastInLumBuf < lastLumSrcY) {
             const uint8_t *src1= src[0]+(lastInLumBuf + 1 - srcSliceY)*srcStride[0];
             const uint8_t *src2= src[3]+(lastInLumBuf + 1 - srcSliceY)*srcStride[3];
             lumBufIndex++;
-            DEBUG_BUFFERS("\t\tlumBufIndex %d: lastInLumBuf: %d\n",
-                               lumBufIndex,    lastInLumBuf);
             assert(lumBufIndex < 2*vLumBufSize);
             assert(lastInLumBuf + 1 - srcSliceY < srcSliceH);
             assert(lastInLumBuf + 1 - srcSliceY >= 0);
             RENAME(hyscale)(c, lumPixBuf[ lumBufIndex ], dstW, src1, srcW, lumXInc,
                             hLumFilter, hLumFilterPos, hLumFilterSize,
                             formatConvBuffer,
                             pal, 0);
             if (CONFIG_SWSCALE_ALPHA && alpPixBuf)
                 RENAME(hyscale)(c, alpPixBuf[ lumBufIndex ], dstW, src2, srcW, lumXInc,
                                 hLumFilter, hLumFilterPos, hLumFilterSize,
                                 formatConvBuffer,
                                 pal, 1);
             lastInLumBuf++;
+            DEBUG_BUFFERS("\t\tlumBufIndex %d: lastInLumBuf: %d\n",
+                               lumBufIndex,    lastInLumBuf);
         }
         while(lastInChrBuf < lastChrSrcY) {
             const uint8_t *src1= src[1]+(lastInChrBuf + 1 - chrSrcSliceY)*srcStride[1];
             const uint8_t *src2= src[2]+(lastInChrBuf + 1 - chrSrcSliceY)*srcStride[2];
             chrBufIndex++;
-            DEBUG_BUFFERS("\t\tchrBufIndex %d: lastInChrBuf: %d\n",
-                               chrBufIndex,    lastInChrBuf);
             assert(chrBufIndex < 2*vChrBufSize);
             assert(lastInChrBuf + 1 - chrSrcSliceY < (chrSrcSliceH));
             assert(lastInChrBuf + 1 - chrSrcSliceY >= 0);
             //FIXME replace parameters through context struct (some at least)
 
             if (c->needs_hcscale)
                 RENAME(hcscale)(c, chrPixBuf[ chrBufIndex ], chrDstW, src1, src2, chrSrcW, chrXInc,
                                 hChrFilter, hChrFilterPos, hChrFilterSize,
                                 formatConvBuffer,
                                 pal);
             lastInChrBuf++;
+            DEBUG_BUFFERS("\t\tchrBufIndex %d: lastInChrBuf: %d\n",
+                               chrBufIndex,    lastInChrBuf);
         }
         //wrap buf index around to stay inside the ring buffer
         if (lumBufIndex >= vLumBufSize) lumBufIndex-= vLumBufSize;
         if (chrBufIndex >= vChrBufSize) chrBufIndex-= vChrBufSize;
         if (!enough_lines)
             break; //we can't output a dstY line so let's try with the next slice
 
 #if COMPILE_TEMPLATE_MMX
         c->blueDither= ff_dither8[dstY&1];
         if (c->dstFormat == PIX_FMT_RGB555 || c->dstFormat == PIX_FMT_BGR555)
@@ skipping 315 matching lines @@
         c->lumSrcOffset = ALT32_CORR;
         c->chrSrcOffset = ALT32_CORR;
         break;
     case PIX_FMT_RGB48LE:
         c->lumSrcOffset = 1;
         c->chrSrcOffset = 1;
         c->alpSrcOffset = 1;
         break;
     }
 
-    if (c->srcRange != c->dstRange && !(isRGB(c->dstFormat) || isBGR(c->dstFormat))) {
+    if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
         if (c->srcRange) {
             c->lumConvertRange = RENAME(lumRangeFromJpeg);
             c->chrConvertRange = RENAME(chrRangeFromJpeg);
         } else {
             c->lumConvertRange = RENAME(lumRangeToJpeg);
             c->chrConvertRange = RENAME(chrRangeToJpeg);
         }
     }
 
     if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
           srcFormat == PIX_FMT_MONOBLACK || srcFormat == PIX_FMT_MONOWHITE))
         c->needs_hcscale = 1;
 }