[FIRRTL][CheckCombLoops] don't crash on force+rwprobeop, workaround. (l…

…lvm#6821)

Unbreak this, at cost of missing loops through force, while
the better fix is being worked out.

cc llvm#6820.

lib/Dialect/FIRRTL/Transforms/CheckCombLoops.cpp

@@ -327,6 +327,13 @@ class DiscoverLoops {
     if (leader == rwProbeClasses.member_end())
       return;
     auto iter = rwProbeRefersTo.find(*leader);
+
+    // This should be found, but for now may not be due to needing
+    // RWProbeOp support.  May cause missed loops involving force for now.
+    // https://github.com/llvm/circt/issues/6820
+    if (iter == rwProbeRefersTo.end())
+      return;
+
     assert(iter != rwProbeRefersTo.end());
     if (iter->second != dstNode)
       drivenBy[iter->second].second.push_back(getOrAddNode(srcVal));

test/Dialect/FIRRTL/check-comb-cycles.mlir

@@ -1086,3 +1086,23 @@ firrtl.circuit "OutsideDialectSource" {
     firrtl.strictconnect %b, %a : !firrtl.uint<32>
   }
 }
+
+// -----
+
+// Check RWProbeOp + force doesn't crash.
+// No loop here.
+firrtl.circuit "Issue6820" {
+  firrtl.module private @Foo(in %clock: !firrtl.clock sym @sym, out %clockProbe_bore: !firrtl.rwprobe<clock>) {
+    %0 = firrtl.ref.rwprobe <@Foo::@sym> : !firrtl.rwprobe<clock>
+    firrtl.ref.define %clockProbe_bore, %0 : !firrtl.rwprobe<clock>
+  }
+  firrtl.module @Issue6820(in %clock: !firrtl.clock, out %clockProbe: !firrtl.rwprobe<clock>) attributes {convention = #firrtl<convention scalarized>} {
+    %foo_clock, %foo_clockProbe_bore = firrtl.instance foo @Foo(in clock: !firrtl.clock, out clockProbe_bore: !firrtl.rwprobe<clock>)
+    firrtl.strictconnect %foo_clock, %clock : !firrtl.clock
+    firrtl.ref.define %clockProbe, %foo_clockProbe_bore : !firrtl.rwprobe<clock>
+    %c1_ui1 = firrtl.constant 1 : !firrtl.uint<1>
+    %c0_ui1 = firrtl.constant 0 : !firrtl.uint<1>
+    %0 = firrtl.asClock %c0_ui1 : (!firrtl.uint<1>) -> !firrtl.clock
+    firrtl.ref.force %clock, %c1_ui1, %clockProbe, %0 : !firrtl.clock, !firrtl.uint<1>, !firrtl.clock
+  }
+}