Incorrectly compiled vectorized code

[igogo-x86]

The following code:

#include <stdio.h>

__attribute__((noinline)) void func(double *__restrict__ cme, double *__restrict__ qn, double *__restrict__ qsp, double *__restrict__ cwat, double *__restrict__ cmeres, double deltat, unsigned long long N) {
    for (unsigned long long i = 0; i < N; ++i) {
        if (qn[i] > 2.0)
            cme[i] = cme[i] + qn[i] * deltat;

       cme[i] = cme[i] + cmeres[i];
    }
}

int main() {
    double zero[] = {0.0, 0.0, 0.0, 0.0};
    double cme[] = {1.0, 1.0, 1.0, 1.0};
    double deltat = 1.0;
    double qn[] = {2.0, 2.0, 2.0, 2.0};
    double qsp[] = {2.0, 2.0, 2.0, 2.0};
    double cwat[] = {1.0, 1.0, 1.0, 1.0};

    func(cme, qn, /*qsp*/ qsp, /*cwat*/ zero, /* cmeres */zero, deltat, /*N*/4);
    
    printf("%f\n", cme[0]);
    return 0;
}

gives different results when vectorised and not-vectorised:

$ clang++ -mcpu=neoverse-v1 -march=armv8+sve -Ofast bug.c && ./a.out
2.000000
$ clang++ -mcpu=neoverse-v1 -march=armv8+sve -O0 bug.c && ./a.out
1.000000

[llvmbot]

@llvm/issue-subscribers-bug

[igogo-x86]

Pre-commit tests - https://reviews.llvm.org/D157094
Fix - https://reviews.llvm.org/D157095

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

[tru]

@igogo-x86 should this be fixed in the release branch as well?

[igogo-x86]

@tru, yes, definitely. The final solution has two pre-commit with tests, and I hope this is the right way to cherry-pick it.

[igogo-x86]

/cherry-pick b560d5c
/cherry-pick 7542477
/cherry-pick 84d444f

[llvmbot]

Failed to cherry-pick: 84d444f

https://github.com/llvm/llvm-project/actions/runs/5806988783

Please manually backport the fix and push it to your github fork. Once this is done, please add a comment like this:

/branch <user>/<repo>/<branch>

[llvmbot]

/branch llvm/llvm-project-release-prs/issue64419

[llvmbot]

/pull-request llvm/llvm-project-release-prs#558

[nikic]

/cherry-pick b560d5c 7542477 84d444f

[llvmbot]

/branch llvm/llvm-project-release-prs/issue64419