阈值图像的c ++优化

Question

我正在编写自定义阈值功能。我有两个在性能上有奇怪结果的实现。

const uchar *ptr;

const uchar *ptr0;
const uchar *ptr1;
const uchar *ptr2;
const uchar *ptr3;
const uchar *ptr4;
const uchar *ptr5;
const uchar *ptr6;
const uchar *ptr7;

uchar *ptrR;

for (int y = inset; y < height; y++) {
    // setup ptrs

    for (int x = inset; x < width; x++) {

        threshVal = *ptr + 8;

        *ptrR = (((*ptr0 - threshVal) >> 31) + 
                ((*ptr1 - threshVal) >> 31) + 
                ((*ptr2 - threshVal) >> 31) + 
                ((*ptr3 - threshVal) >> 31) + 
                ((*ptr4 - threshVal) >> 31) + 
                ((*ptr5 - threshVal) >> 31) + 
                ((*ptr6 - threshVal) >> 31) + 
                ((*ptr7 - threshVal) >> 31) + 8) * 20;
        // increase ptrs
    }
}

我重写了这个解决方案并且性能更差（here is idea）

static uchar threshValues[600];
static uchar *threshValuesOffset;

// call this once
memset(threshValues, 0, sizeof(uchar)*300);
memset(threshValues+300, 20, sizeof(uchar)*300);
threshValuesOffset = threshValues + 300;

for (int y = inset; y < height; y++) {
// setup ptrs

    for (int x = inset; x < width; x++) {

        threshVal = *ptr + 8;
        uchar *checkPtr = threshValuesOffset - threshVal;

        *ptrR = checkPtr[*ptr0] + 
                checkPtr[*ptr1] + 
                checkPtr[*ptr2] + 
                checkPtr[*ptr3] + 
                checkPtr[*ptr4] + 
                checkPtr[*ptr5] + 
                checkPtr[*ptr6] + 
                checkPtr[*ptr7];

        // increase ptrs
    }
}

结果相同，但表现差于25％。如何通过这个问题获得最佳性能？