[llvm] [AMDGPU] Fix overlapping insert crash during rewrite-agpr-copy-mfma (PR #205962)

[Dhruva Chakrabarti via llvm-commits]

================
@@ -610,8 +610,34 @@ void AMDGPURewriteAGPRCopyMFMAImpl::eliminateSpillsOfReassignedVGPRs() const {
 
     ArrayRef<MCPhysReg> AllocOrder = RegClassInfo.getOrder(RC);
 
+    // The stack slot's LiveInterval may be discontiguous: a slot can be live
+    // in memory around a spill store and around a much later reload. Once
+    // we unspill the slot into a register, however, the value must reside in
+    // that register continuously from its first reference to its last. Checking
+    // interference against the slot's discontiguous interval could let us pick
+    // a PhysReg that is busy inside a gap, corrupting it. Instead, check
+    // interference over the contiguous hull the replacement register will
+    // occupy.
+    //
+    // The index-based checkInterference only consults the assigned-vreg matrix;
+    // it does not account for fixed (reg-unit) or regmask interference. Build a
+    // hull LiveInterval so we can additionally query those, ensuring we never
+    // reassign into a register clobbered by a fixed def or a call inside the
+    // hull's gap. Avoid calling checkInterference with the hull interval, as
+    // it may return stale results when a temporary interval is reused across
+    // slots.
+    LiveInterval HullLI(LI->reg(), LI->weight());
----------------
dhruvachak wrote:

The temporary LiveInterval is gone now.

https://github.com/llvm/llvm-project/pull/205962