comparison postproc/postprocess_template.c @ 4403:f3dc8bf8383a

top row bugfix

4402:67abbf501b02

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, struct PPMode *ppMode);

/**
 * Copies a block from src to dst and fixes the blacklevel
 * numLines must be a multiple of 4
 * levelFix == 0 -> dont touch the brighness & contrast
 */
static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, uint8_t src[], int srcStride,
	int levelFix)
{

						&(src[srcStride*i]), BLOCK_SIZE);
#endif
	}
}


/**
 * 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,

	}

	/* copy & deinterlace first row of blocks */
	y=-BLOCK_SIZE;
	{
		//1% speedup if these are here instead of the inner loop
		uint8_t *srcBlock= &(src[y*srcStride]);
		uint8_t *dstBlock= &(dst[y*dstStride]);

		dstBlock= tempDst + dstStride;

		// From this point on it is guranteed that we can read and write 16 lines downward
		// finish 1 block before the next otherwise weŽll 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)

			prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
			prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
*/
#endif

			RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
				srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX);

			if(mode & LINEAR_IPOL_DEINT_FILTER)
				RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
			else if(mode & LINEAR_BLEND_DEINT_FILTER)
				RENAME(deInterlaceBlendLinear)(dstBlock, dstStride);

				RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
*/
			dstBlock+=8;
			srcBlock+=8;
		}
		memcpy(&(dst[y*dstStride]) + 8*dstStride, tempDst + 9*dstStride, copyAhead*dstStride );
	}

	for(y=0; y<height; y+=BLOCK_SIZE)
	{
		//1% speedup if these are here instead of the inner loop

4403:f3dc8bf8383a

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, struct PPMode *ppMode);

/**
 * Copies a block from src to dst and fixes the blacklevel
 * levelFix == 0 -> dont touch the brighness & contrast
 */
static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, uint8_t src[], int srcStride,
	int levelFix)
{

						&(src[srcStride*i]), BLOCK_SIZE);
#endif
	}
}

/**
 * Duplicates the given 8 src pixels ? times upward
 */
static inline void RENAME(duplicate)(uint8_t src[], int stride)
{
#ifdef HAVE_MMX
	asm volatile(
		"movq (%0), %%mm0		\n\t"
		"addl %1, %0			\n\t"
		"movq %%mm0, (%0)		\n\t"
		"movq %%mm0, (%0, %1)		\n\t"
		"movq %%mm0, (%0, %1, 2)	\n\t"
		: "+r" (src)
		: "r" (-stride)
	);
#else
	int i;
	uint8_t *p=src;
	for(i=0; i<3; i++)
	{
		p-= stride;
		memcpy(p, src, 8);
	}
#endif
}

/**
 * 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,

	}

	/* copy & deinterlace first row of blocks */
	y=-BLOCK_SIZE;
	{
		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
		// finish 1 block before the next otherwise weŽll 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)

			prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
			prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
*/
#endif

			RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
				srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX);

			RENAME(duplicate)(dstBlock + dstStride*8, dstStride);

			if(mode & LINEAR_IPOL_DEINT_FILTER)
				RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
			else if(mode & LINEAR_BLEND_DEINT_FILTER)
				RENAME(deInterlaceBlendLinear)(dstBlock, dstStride);

				RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
*/
			dstBlock+=8;
			srcBlock+=8;
		}
		memcpy(dst, tempDst + 9*dstStride, copyAhead*dstStride );
	}

	for(y=0; y<height; y+=BLOCK_SIZE)
	{
		//1% speedup if these are here instead of the inner loop