[llvm] [PHIElimination] Handle subranges in LiveInterval updates (PR #69429)

[Carl Ritson via llvm-commits]

================
@@ -400,24 +400,50 @@ void PHIElimination::LowerPHINode(MachineBasicBlock &MBB,
     }
 
     LiveInterval &DestLI = LIS->getInterval(DestReg);
-    assert(!DestLI.empty() && "PHIs should have nonempty LiveIntervals.");
-    if (DestLI.endIndex().isDead()) {
-      // A dead PHI's live range begins and ends at the start of the MBB, but
-      // the lowered copy, which will still be dead, needs to begin and end at
-      // the copy instruction.
-      VNInfo *OrigDestVNI = DestLI.getVNInfoAt(MBBStartIndex);
-      assert(OrigDestVNI && "PHI destination should be live at block entry.");
-      DestLI.removeSegment(MBBStartIndex, MBBStartIndex.getDeadSlot());
-      DestLI.createDeadDef(DestCopyIndex.getRegSlot(),
-                           LIS->getVNInfoAllocator());
-      DestLI.removeValNo(OrigDestVNI);
-    } else {
-      // Otherwise, remove the region from the beginning of MBB to the copy
-      // instruction from DestReg's live interval.
-      DestLI.removeSegment(MBBStartIndex, DestCopyIndex.getRegSlot());
-      VNInfo *DestVNI = DestLI.getVNInfoAt(DestCopyIndex.getRegSlot());
+    assert(!DestLI.empty() && "PHIs should have non-empty LiveIntervals.");
+
+    SlotIndex NewStart = DestCopyIndex.getRegSlot();
+
+    SmallVector<LiveRange *> ToUpdate;
+    ToUpdate.push_back(&DestLI);
+    for (auto &SR : DestLI.subranges())
+      ToUpdate.push_back(&SR);
+
+    for (auto LR : ToUpdate) {
+      auto DestSegment = LR->find(MBBStartIndex);
+      assert(DestSegment != LR->end() &&
+             "PHI destination must be live in block");
+
+      if (LR->endIndex().isDead()) {
+        // A dead PHI's live range begins and ends at the start of the MBB, but
+        // the lowered copy, which will still be dead, needs to begin and end at
+        // the copy instruction.
+        VNInfo *OrigDestVNI = LR->getVNInfoAt(DestSegment->start);
+        assert(OrigDestVNI && "PHI destination should be live at block entry.");
+        LR->removeSegment(DestSegment->start, DestSegment->start.getDeadSlot());
+        LR->createDeadDef(NewStart, LIS->getVNInfoAllocator());
+        LR->removeValNo(OrigDestVNI);
+        continue;
+      }
+
+      if (DestSegment->start > NewStart) {
+        // With a single PHI removed from block the index of the copy may be
+        // lower than the original PHI. Extend live range backward to cover
+        // the copy.
----------------
perlfu wrote:

I have moved and reworded this comment for clarity.

There are two details here that matter to understanding this:
* LiveIntervals of PHIs destinations start at the PHI slot index, not the block start.
* Destination copies are not inserted in the order of the PHIs they replace.

Taking the example of the attached test.
Original PHIs look like this:
```
320B      %phi0:vgpr_32 = PHI %s0a:vgpr_32, %bb.0, %s0c:vgpr_32, %bb.1
336B      %phi1:vgpr_32 = PHI %s0b:vgpr_32, %bb.0, %s0d:vgpr_32, %bb.1
```

With intervals as follows:
```
%6 [112r,208r:0) 0 at 112r  weight:0.000000e+00 -- s0a
%7 [128r,224r:0) 0 at 128r  weight:0.000000e+00 -- s0b
%8 [208r,256B:0) 0 at 208r  weight:0.000000e+00 -- s0c
%9 [224r,256B:0) 0 at 224r  weight:0.000000e+00 -- s0d
%10 [320r,384r:0) 0 at 320r  weight:0.000000e+00 -- phi0
%11 [336r,384r:0) 0 at 336r  weight:0.000000e+00 -- phi1
```

phi0 is lowered, giving the following IR:
```
336B      %phi1:vgpr_32 = PHI %s0b:vgpr_32, %bb.0, %s0d:vgpr_32, %bb.1
344B      %phi0:vgpr_32 = COPY %12:vgpr_32
```

phi1 is lowered, giving:
```
312B      %phi1:vgpr_32 = COPY %13:vgpr_32
344B      %phi0:vgpr_32 = COPY %12:vgpr_32
```

Hence COPY for phi0 has index after original PHI, but phi1 has index before original PHI.

https://github.com/llvm/llvm-project/pull/69429