[Bug c/112457] New: Possible better vectorization of different reduction min/max reduction

[juzhe.zhong at rivai dot ai via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112457

            Bug ID: 112457
           Summary: Possible better vectorization of different reduction
                    min/max reduction
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: juzhe.zhong at rivai dot ai
  Target Milestone: ---

Hi, Richard.

GCC-14 almost has all features of RVV.

I am planning to participate on improving GCC loop vectorizer in GCC-15.

Fix FAILs of TSVC is one of my plan.

Currently we can vectorize this following case:

int idx = 0;
int max = 0;
void foo (int n, int * __restrict a){
for (int i = 0; i < n; ++i) {
  max = max < a[i] ? a[i] : max;
}
}

However, if we change this case it failed:

void foo2 (int n, int * __restrict a){
for (int i = 0; i < n; ++i) {
  if (max < a[i]) {
    max = a[i];
  } else
    max = max;
}
}

Now, I notice another interesting and possible vectorization enhancement which
inspired by this patch of LLVM:
https://reviews.llvm.org/D143465

And more advance case is which is case from LLVM patch:
which is vectorization reduction with index:

void foo3 (int n, int * __restrict a){
for (int i = 0; i < n; ++i) {
  if (max < a[i]) {
    idx = i;
    max = a[i];
  }
}
}

I wonder it is a valuable optimization ? If yes, it would be one of my TODO
list.

Thanks.