https://bugs.llvm.org/show_bug.cgi?id=43553

            Bug ID: 43553
           Summary: SLP vectorizer doesn't obey
                    llvm.loop.vectorize.width=1
           Product: libraries
           Version: 9.0
          Hardware: Other
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Loop Optimizer
          Assignee: [email protected]
          Reporter: [email protected]
                CC: [email protected]

Follow up to bug 43539. Code example is there.

The SLP vectorizer (plus any other passes that emit SIMD code) should always
obey the llvm.loop.vectorize.width=1 (emitted by #pragma clang loop
vectorize(disable)) or llvm.loop.vectorize.enable=false.

Currently, what happens on non-trivial loops is that the loop optimizer will
unroll (and/or interleave) the loop, and SLP will vectorize the unrolled loop.

Enabled or disabled, the loop ends up being vectorized, making this
annotation/pragma useless.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs

Reply via email to