comparison postprocess_template.c @ 49:cdfe8fc50090 libpostproc

misc typo fixes

48:72675332ed82

#if 0
/**
 * 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
        0 8 16 24
        x = 8
        x/2 = 4
        x/8 = 1
        1 12 12 23

                        "g" ((long)x), "g" ((long)copyAhead)
                        : "%"REG_a, "%"REG_d
                        );

#elif defined(HAVE_3DNOW)
//FIXME check if this is faster on an 3dnow chip or if its 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);
                        prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
*/

                         "g" ((long)x), "g" ((long)copyAhead)
                        : "%"REG_a, "%"REG_d
                        );

#elif defined(HAVE_3DNOW)
//FIXME check if this is faster on an 3dnow chip or if its 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);
                        prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
*/

49:cdfe8fc50090

#if 0
/**
 * Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar
 * values are correctly clipped (MMX2)
 * values are wraparound (C)
 * Conclusion: It is fast, but introduces ugly horizontal patterns
 * if there is a continuous gradient.
        0 8 16 24
        x = 8
        x/2 = 4
        x/8 = 1
        1 12 12 23

                        "g" ((long)x), "g" ((long)copyAhead)
                        : "%"REG_a, "%"REG_d
                        );

#elif defined(HAVE_3DNOW)
//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);
                        prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
*/

                         "g" ((long)x), "g" ((long)copyAhead)
                        : "%"REG_a, "%"REG_d
                        );

#elif defined(HAVE_3DNOW)
//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);
                        prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
*/