[llvm] [GISel] Fix ShuffleVector assert (PR #139769)

[Matt Arsenault via llvm-commits]

================
@@ -135,3 +135,27 @@ body:             |
     SI_RETURN
 ...
 
+
+---
+name: shuffle_vector_to_copy
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; CHECK-LABEL: name: shuffle_vector_to_copy
+    ; CHECK: liveins: $vgpr0, $vgpr1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(p3) = COPY $vgpr0
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(p3) = COPY $vgpr1
+    ; CHECK-NEXT: [[LOAD:%[0-9]+]]:_(<8 x s16>) = G_LOAD [[COPY]](p3) :: (load (<8 x s16>), align 8, addrspace 3)
+    ; CHECK-NEXT: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16), [[UV4:%[0-9]+]]:_(s16), [[UV5:%[0-9]+]]:_(s16), [[UV6:%[0-9]+]]:_(s16), [[UV7:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[LOAD]](<8 x s16>)
+    ; CHECK-NEXT: G_STORE [[UV4]](s16), [[COPY1]](p3) :: (store (s16), addrspace 3)
+    ; CHECK-NEXT: SI_RETURN
+    %0:_(p3) = COPY $vgpr0
+    %1:_(p3) = COPY $vgpr1
+    %12:_(<8 x s16>) = G_IMPLICIT_DEF
+    %10:_(<8 x s16>) = G_LOAD %0(p3) :: (load (<8 x s16>), align 8, addrspace 3)
+    %11:_(s16) = G_SHUFFLE_VECTOR %10(<8 x s16>), %12, shufflemask(4)
----------------
arsenm wrote:

Should we even permit this in the MIR verifier? What produced this shuffle? 

https://github.com/llvm/llvm-project/pull/139769