* 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 postprocess_template.c
"paddb " #a ", " #b " \n\t"
#endif
-//FIXME? |255-0| = 1 (shouldnt be a problem ...)
+//FIXME? |255-0| = 1 (should not be a problem ...)
#ifdef HAVE_MMX
/**
* Check if the middle 8x8 Block in the given 8x16 block is flat
* Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar
* values are correctly clipped (MMX2)
* values are wraparound (C)
- * conclusion: its fast, but introduces ugly horizontal patterns if there is a continious gradient
+ * Conclusion: It is fast, but introduces ugly horizontal patterns
+ * if there is a continuous gradient.
0 8 16 24
x = 8
x/2 = 4
* Experimental Filter 1
* will not damage linear gradients
* Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
- * can only smooth blocks at the expected locations (it cant smooth them if they did move)
- * MMX2 version does correct clipping C version doesnt
+ * can only smooth blocks at the expected locations (it cannot smooth them if they did move)
+ * MMX2 version does correct clipping C version does not
*/
static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
{
}
#endif //HAVE_MMX
-static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
- QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
+static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
+ const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
/**
- * Copies a block from src to dst and fixes the blacklevel
- * levelFix == 0 -> dont touch the brighness & contrast
+ * Copies a block from src to dst and fixes the blacklevel.
+ * levelFix == 0 -> do not touch the brighness & contrast
*/
#undef SCALED_CPY
-static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, uint8_t src[], int srcStride,
+static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,
int levelFix, int64_t *packedOffsetAndScale)
{
#ifndef HAVE_MMX
/**
* Filters array of bytes (Y or U or V values)
*/
-static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
- QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
+static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
+ const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
{
- PPContext __attribute__((aligned(8))) c= *c2; //copy to stack for faster access
+ DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
int x,y;
#ifdef COMPILE_TIME_MODE
const int mode= COMPILE_TIME_MODE;
sum+= yHistogram[i];
}
- /* we allways get a completly black picture first */
+ /* We always get a completely black picture first. */
maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
clipped= sum;
/* copy & deinterlace first row of blocks */
y=-BLOCK_SIZE;
{
- uint8_t *srcBlock= &(src[y*srcStride]);
+ const uint8_t *srcBlock= &(src[y*srcStride]);
uint8_t *dstBlock= tempDst + dstStride;
- // From this point on it is guranteed that we can read and write 16 lines downward
+ // From this point on it is guaranteed that we can read and write 16 lines downward
// finish 1 block before the next otherwise we might have a problem
// with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
for(x=0; x<width; x+=BLOCK_SIZE)
);
#elif defined(HAVE_3DNOW)
-//FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...
+//FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
/* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
for(y=0; y<height; y+=BLOCK_SIZE)
{
//1% speedup if these are here instead of the inner loop
- uint8_t *srcBlock= &(src[y*srcStride]);
+ const uint8_t *srcBlock= &(src[y*srcStride]);
uint8_t *dstBlock= &(dst[y*dstStride]);
#ifdef HAVE_MMX
uint8_t *tempBlock1= c.tempBlocks;
uint8_t *tempBlock2= c.tempBlocks + 8;
#endif
- int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
+ const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
int QP=0;
/* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
srcBlock= tempSrc;
}
- // From this point on it is guranteed that we can read and write 16 lines downward
+ // From this point on it is guaranteed that we can read and write 16 lines downward
// finish 1 block before the next otherwise we might have a problem
// with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
for(x=0; x<width; x+=BLOCK_SIZE)
);
#elif defined(HAVE_3DNOW)
-//FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...
+//FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
/* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
else if(mode & H_DEBLOCK)
{
#ifdef HAVE_ALTIVEC
- unsigned char __attribute__ ((aligned(16))) tempBlock[272];
+ DECLARE_ALIGNED(16, unsigned char, tempBlock[272]);
transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
const int t=vertClassify_altivec(tempBlock-48, 16, &c);
projects / ffmpeg.git / blobdiff