CODEWITHSUNDEEP
x86ssesimd
user3646608
user3646608

Reputation: 1

SIMD matrix multiply segmentation error

__m128d c1,c2,c3,c4,a1,a2,b1,b2;

int ida = 2;

for(int i = 0; i<n; i++) {
  b1 = _mm_load_pd(b+i*n);
  b2 = _mm_load_pd(b+i*n+ida);
  for(int j = 0; j<n/2; j++) {
    a1 = _mm_load_pd(a+i+j*2*n);
    a2 = _mm_load_pd(a+i+j*2*n+n);
    c1 = _mm_load_pd(c+j*2*n);
    c2 = _mm_load_pd(c+j*2*n+n);
    c3 = _mm_load_pd(c+j*2*n+ida);
    c4 = _mm_load_pd(c+j*2*n+n+ida);
    c1 = _mm_add_pd(c1, _mm_mul_pd(a1, b1));
    c2 = _mm_add_pd(c2, _mm_mul_pd(a2, b1));
    c3 = _mm_add_pd(c3, _mm_mul_pd(a1, b2));
    c4 = _mm_add_pd(c4, _mm_mul_pd(a2, b2));
    _mm_store_pd(c+j*2*n, c1);
    _mm_store_pd(c+j*2*n+n, c2);
    _mm_store_pd(c+j*2*n+ida, c3);
    _mm_store_pd(c+j*2*n+n+ida, c4);
  }
}

I have a segmentation error but I don't know why this error occurred.

The matrix is like:

a1  a2  a3  a4
a5  a6 ...

I want to make n*n matrix multiply.

Upvotes: 0

Views: 157

Answers (1)

Paul R
Paul R

Reputation: 213160

It looks like your loads will be misaligned in at least some cases, e.g. when i = 1. Change all instances of _mm_load_pd_ to _mm_loadu_pd in order to handle misaligned cases.

Upvotes: 1

Related Questions