https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126301
Bug ID: 126301
Summary: Incorrect loop control for loops where latch preceeds
loop
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: alfierichards at gcc dot gnu.org
Target Milestone: ---
Created attachment 65074
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=65074&action=edit
Graph dump of vect pass output
For AArch64 with partial loops (and riscv I think) when dealing with a loop
such as
> int a;
> char b[];
> int c() {
> unsigned d = 0;
> for (; d < a; ++d)
> if (b[0] + b[d + 1])
> return 0;
> }
where the loop latch is at the beginning of the loop, the loop control logic
checks if the mask for the "next mask" is not empty, but that is the mask for
the iteration after the one coming.
This is overly pessimistic and guarantees at least a full VF iteration of
scalar fallback code.
https://godbolt.org/z/YGv3M9vEx
