[AMDGCN] Allow unscheduling of bundled insns

This is a patch arising from AMD's fuzzing project.

In the test case, the scheduling algorithm decides to undo an attempted
schedule, but is unprepared to handle bundled instructions at that
point -- and those can arise via the expansion of intrinsics earlier
in compilation.  The fix is to use the splice method instead of
remove/insert, since that can handle bundles properly.

Author: jtb20

llvm/lib/Target/AMDGPU/GCNSchedStrategy.cpp

@@ -1567,8 +1567,7 @@ void GCNSchedStage::revertScheduling() {
     }
 
     if (MI->getIterator() != DAG.RegionEnd) {
-      DAG.BB->remove(MI);
-      DAG.BB->insert(DAG.RegionEnd, MI);
+      DAG.BB->splice(DAG.RegionEnd, DAG.BB, MI);
       if (!MI->isDebugInstr())
         DAG.LIS->handleMove(*MI, true);
     }

llvm/test/CodeGen/AMDGPU/sema-v-unsched-bundle.ll

@@ -0,0 +1,18 @@
+; REQUIRES: asserts
+; RUN: llc -mtriple=amdgcn -O1 -mcpu=gfx90a -debug-only=machine-scheduler -filetype=null < %s 2>&1 | FileCheck --check-prefix=DEBUG %s
+
+; DEBUG: Attempting to revert scheduling.
+
+@G = global <32 x i8> splat (i8 1)
+@G.1 = global <32 x i8> splat (i8 127)
+
+define amdgpu_kernel void @gws_sema_v_offset0(i32 %val, <32 x i1>* %inp) {
+  %LGV1 = load <32 x i8>, ptr @G.1, align 32
+  %LGV = load <32 x i8>, ptr @G, align 32
+  call void @llvm.amdgcn.ds.gws.sema.v(i32 0)
+  %C = icmp ne <32 x i8> %LGV, %LGV1
+  store <32 x i1> %C, ptr %inp, align 4
+  ret void
+}
+
+declare void @llvm.amdgcn.ds.gws.sema.v(i32)