[llvm] [PHIElimination] Reuse existing COPY in predecessor basic block (PR #131837)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,35 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc -run-pass=phi-node-elimination -mtriple=aarch64-linux-gnu -o - %s | FileCheck %s
+
+# Verify that the original COPY in bb.1 is reappropriated as the PHI source in bb.2,
+# instead of creating a new COPY with the same source register.
+
+---
+name: test
+tracksRegLiveness: true
+body: |
+  ; CHECK-LABEL: name: test
+  ; CHECK: bb.0:
+  ; CHECK-NEXT:   successors: %bb.2(0x40000000), %bb.1(0x40000000)
+  ; CHECK-NEXT:   liveins: $nzcv, $wzr
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:gpr32 = IMPLICIT_DEF
+  ; CHECK-NEXT:   Bcc 8, %bb.2, implicit $nzcv
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.1:
+  ; CHECK-NEXT:   successors: %bb.2(0x80000000)
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT:   [[DEF:%[0-9]+]]:gpr32 = COPY $wzr
+  ; CHECK-NEXT: {{  $}}
+  ; CHECK-NEXT: bb.2:
+  ; CHECK-NEXT:   %y:gpr32 = COPY [[DEF]]
+  ; CHECK-NEXT:   $wzr = COPY %y
+  bb.0:
+    liveins: $nzcv, $wzr
+    Bcc 8, %bb.2, implicit $nzcv
+  bb.1:
+    %x:gpr32 = COPY $wzr
----------------
arsenm wrote:

Can you add a variant of the test where the source is a virtual register, and one that uses a not-special-case-constant register? I suspect the coalescer deals with this in the ordinary virtual register source case, and could maybe deal with constant physregs? 

https://github.com/llvm/llvm-project/pull/131837