[OpenACC] Fix infinite loop bug with the device_type checking

I noticed while writing a test that we ended up infinite looping thanks
to an early exit not correctly setting the next step of the loop.

Author: erichkeane

clang/lib/Sema/SemaOpenACCClause.cpp

@@ -403,8 +403,10 @@ class SemaOpenACCClauseVisitor {
 
       // There are no clauses of the current kind between these device_types, so
       // continue.
-      if (CurClauseKindItr == CurDevTypeItr)
+      if (CurClauseKindItr == CurDevTypeItr) {
+        PrevDeviceTypeItr = CurDevTypeItr;
         continue;
+      }
 
       // At this point, we know that this device_type region has a collapse.  So
       // diagnose if the two device_types have any overlap in their

clang/test/SemaOpenACC/loop-construct-collapse-clause.cpp

@@ -572,4 +572,9 @@ void no_dupes_since_last_device_type() {
 #pragma acc loop device_type(radeon) collapse(1) device_type(nvidia, radeon) collapse(2)
   for(unsigned i = 0; i < 5; ++i)
     for(unsigned j = 0; j < 5; ++j);
+
+  int NotConstexpr;
+  // expected-error@+1 3{{OpenACC 'collapse' clause loop count must be a constant expression}}
+#pragma acc loop collapse(NotConstexpr) device_type(radeon, nvidia) collapse(NotConstexpr) device_type(host) collapse(NotConstexpr)
+  for(unsigned j = 0; j < 5; ++j);
 }