On 9/18/26 7:56 AM, Robin Dapp wrote:
Compared to v1, the patch now removes the early VLS rejection
completely, and considers scenarios where `this` has spills while
`other` does not, which can happen when VLS is legal for unknown-iters
loops. It is also different from the early implementation in that we
now don't trigger the early return when both main loops have spills. It
has caused several vect dump expectation changes in some testcases, but
the final assembly for them remains identical.
OK.
It looked independent of patch #2 in the series, so I went ahead and
pushed it.
Thanks,
Jeff
