//shuipin for(int Y = 0; Y < Radius; Y++){ int Stride = Y * Width; for(int X = 0; X < Width; X++){ colsumPtr[X] += colsumPtr[Stride + X]; } }
should be

//shuipin for(int Y = 0; Y < Radius; Y++){ int Stride = Y * Width; for(int X = 0; X < Width; X++){ colsumPtr[X] += cachePtr[Stride + X]; } }

In BoxFilterNeonIntrinsics function:
for(int Y = 0; Y < Radius; Y++){ int Stride = Y * Width; float* tmpColSumPtr = colsumPtr; float* tmpCachePtr = cachePtr;

for(int Y = 0; Y < Radius; Y++){ int Stride = Y * Width; float* tmpColSumPtr = colsumPtr; float* tmpCachePtr = cachePtr + stride;

where is:
boxFilter.cpp,
Function:
void BoxFilterCache(..)..
//shuipin for(int Y=0;Y Radius;Y++){ int Stride =Y * Width; for(int X = 0;X<Width;X++){ colsumPtr[X]+=colsumPtr[Stride+X； } }
It is actually unused.