comparison postprocess_template.c @ 29:2e855f3f4e0b libpostproc

printf --> av_log with some help from Oded

28:b55400a067f0

                        }
                        if(y==4) bias+=d;
                        num++;
                        if(num%1000000 == 0)
                        {
                                printf(" %d %d %d %d\n", num, sum, max, bias);
                        }
                }
        }
}
*/

                                        }
                                        errorSum+= error;

                                        if(1024LL*1024LL*1024LL % numSkiped == 0)
                                        {
                                                printf( "sum:%1.3f, skip:%d, wQP:%d, "
                                                        "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
                                                        (float)errorSum/numSkiped, numSkiped, worstQP, worstRange,
                                                        worstDiff, (float)numSkiped/numPixels);
                                        }
                                }

                "4:                                     \n\t"

                :: "r" (src), "r" (tempBlured), "r"((long)stride), "m" (tempBluredPast)
                : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
                );
//printf("%d\n", test);
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
{
        int y;
        int d=0;
//        int sysd=0;

                +(*(tempBluredPast+256))
                +4)>>3;
        *tempBluredPast=i;
//        ((*tempBluredPast)*3 + d + 2)>>2;

//printf("%d %d %d\n", maxNoise[0], maxNoise[1], maxNoise[2]);
/*
Switch between
 1  0  0  0  0  0  0  (0)
64 32 16  8  4  2  1  (1)
64 48 36 27 20 15 11 (33) (approx)

                if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;

                for(i=0; i<256; i++)
                {
                        sum+= yHistogram[i];
//                        printf("%d ", yHistogram[i]);
                }
//                printf("\n\n");

                /* we allways get a completly black picture first */
                maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);

                clipped= sum;

                                memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
                        }
                }
        }

//printf("\n");
        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]);
                uint8_t *dstBlock= &(dst[y*dstStride]);

                                memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), ABS(dstStride));

                        dstBlock= tempDst + dstStride;
                        srcBlock= tempSrc;
                }
//printf("\n");

                // From this point on it is guranteed 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)

29:2e855f3f4e0b

                        }
                        if(y==4) bias+=d;
                        num++;
                        if(num%1000000 == 0)
                        {
                                av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
                        }
                }
        }
}
*/

                                        }
                                        errorSum+= error;

                                        if(1024LL*1024LL*1024LL % numSkiped == 0)
                                        {
                                                av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
                                                        "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
                                                        (float)errorSum/numSkiped, numSkiped, worstQP, worstRange,
                                                        worstDiff, (float)numSkiped/numPixels);
                                        }
                                }

                "4:                                     \n\t"

                :: "r" (src), "r" (tempBlured), "r"((long)stride), "m" (tempBluredPast)
                : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
                );
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
{
        int y;
        int d=0;
//        int sysd=0;

                +(*(tempBluredPast+256))
                +4)>>3;
        *tempBluredPast=i;
//        ((*tempBluredPast)*3 + d + 2)>>2;

/*
Switch between
 1  0  0  0  0  0  0  (0)
64 32 16  8  4  2  1  (1)
64 48 36 27 20 15 11 (33) (approx)

                if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;

                for(i=0; i<256; i++)
                {
                        sum+= yHistogram[i];
                }

                /* we allways get a completly black picture first */
                maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);

                clipped= sum;

                                memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
                        }
                }
        }

        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]);
                uint8_t *dstBlock= &(dst[y*dstStride]);

                                memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), ABS(dstStride));

                        dstBlock= tempDst + dstStride;
                        srcBlock= tempSrc;
                }

                // From this point on it is guranteed 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)