Auto vectorization of loops

Explain
,
1

Hi! I’m trying to understand why do I get loop unrolling instead of vectorization in this code (Compiler Explorer)? I definitely know that this C code will be vectorized:

void sqr(const int* restrict src, int* restrict dest, unsigned int len) {
  for(int i = 0; i < len; i++){
    dest[i] = src[i] * src[i];
  }
}

Could somebody explain where am I wrong, please?

1 Like
2

Loop-Vectorization-Disabled-to-Work-Around-Regression

3 Likes
3

Thank you!

4

Note that the issue has been fixed on master with the upgrade to LLVM 22

5

As noted in the linked release notes, that is incorrect:

This has been reported and fixed upstream, however at time of writing the fix has not been cherry-picked into LLVM’s 22.x release branch, therefore we expect this performance regression to affect not only Zig 0.16.x but also 0.17.x, finally resolved in 0.18.x.

1 Like
6

Oops my bad, I should have double checked.

Got confused cause I’m using a patched version of LLVM.