解决方法
@H_301_6@// N number of ints to be added
// a,b input array
// c sum array
// nReg number of required vector registers
const unsigned nReg = N*sizeof(uint32_t)/sizeof(__v4si);
__v4si a[nReg],b[nReg],c[nReg];
for (unsigned i=0; i<nReg; ++i)
c[i] = _mm_add_epi32(a[i],b[i]);
// in c++ simply
for (unsigned i=0; i<nReg; ++i)
c[i] = a[i] + b[i];
根据需要展开循环和预取元素.建议进行性能分析.用__v16qi,__ v8hi,__ v2di替换__v4si为8,16,64位整数.
