[llvm] DAG: Handle load in SimplifyDemandedVectorElts (PR #122671)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> I'm still looking at the x86 mess - but something I've hit is that the hasOneUse() checks on the shouldReduceLoadWidth callback are often getting confused by extra uses of the load nodes's chain - is there anything we can do is clean that up? (See also #126764)

SDNode::hasOneUse checks are rarely the correct thing over SDValue::hasOneUse. We should probably rename the SDNode version; it's too easy to mix up N->hasOneUse vs. N.hasOneUse 

https://github.com/llvm/llvm-project/pull/122671