Question

我们尝试做一些SSE操作，但是，在add_sse函数结束时，我们试图读回刚刚计算的值，它会给我们一个seg错误。但是，如果我们只是在for循环中打印该值，结果就可以了。也可以只读取每个数组中的元素0。读取元素1及其后将导致seg错误。

任何人都可以帮我们确定问题吗？我们尝试了一切，但仍然不明白我们会有一个段错误。谢谢

void main()
{
    ResultCounter *c_sse=(ResultCounter *)memalign(16,sizeof(ResultCounter)*4);    

    resetCounter (c_sse);  //initial struct to all 0

    add_sse (1,2, 3,4, c_sse);
}

void add_sse (unsigned int first, unsigned int second, unsigned int third, unsigned int fourth, ResultCounter *c)
{
    __attribute__((align(16))) int m_intarray[4] = {first, second, third,fourth};

    __attribute__((align(16))) int m_Larray[4] = {c[0].L, c[1].L, c[2].L,c[3].L};
    __attribute__((align(16))) int m_Marray[4] = {c[0].M, c[1].M, c[2].M,c[3].M};
    __attribute__((align(16))) int m_Harray[4] = {c[0].H, c[1].H, c[2].H,c[3].H};

    __m128i N = _mm_load_si128(&m_intarray[0]);
    __m128i L = _mm_load_si128(&m_Larray[0]);
    __m128i M = _mm_load_si128(&m_Marray[0]);
    __m128i H = _mm_load_si128(&m_Harray[0]);

    __m128i Lcarry = _mm_and_si128 (L, N);

    L = _mm_xor_si128 (L, N);

    __m128i Mcarry = _mm_and_si128 (M, Lcarry); 

    M = _mm_xor_si128 (M, Lcarry);

    H = _mm_or_si128 (H,Mcarry);

    _mm_store_si128(&m_Larray[0], L);
    _mm_store_si128(&m_Marray[0], M);
    _mm_store_si128(&m_Harray[0], H);

    for(i = 0; i < 4; i++) {
        //printf ("L:%d,addr=%u,M:%u,addr=%u,H:%u,addr=%u\n",m_Larray[i],&m_Larray[i],m_Marray[i],&m_Marray[i],m_Harray[i],&m_Harray[i]);
        c[i].L=m_Larray[i];
        c[i].M=m_Marray[i];
        c[i].H=m_Harray[i];
    }
}

//The struct used in main function.
typedef struct
{
    unsigned int L;
    unsigned int M;
    unsigned int H;
} ResultCounter;

Answer 1

问题是ResultCounter结构的大小是12个字节，所以尽管数组的第一个元素c[0]是16字节对齐的，但第二个元素c[1]是不。现在最快/最简单的修复方法是向此结构添加4个字节的填充，例如另外一个未使用的int：

typedef struct
{
    unsigned int L;
    unsigned int M;
    unsigned int H;
    unsigned int unused;
} ResultCounter;

Answer 2

_mm_load_si128需要16字节对齐的数据。

SSE中的Seg Fault，不确定是什么原因造成的

2 个答案: