Fix non concurrent mark sweep ergonomics.

Previously we would continue to do sticky GC until the sticky GC did
not free enough memory for the allocation, this was excessive since it
could do one sticky GC per allocation. The new logic uses the next GC
type before trying all the GCs in the plan.

Before memalloc benchmark (non concurrent mark sweep):
Total time spent in GC: 11.212701s
Score: 7790

After:
Total time spent in GC: 9.422676s
Score: 6870

Change-Id: Iba75b70ea825ef3fd4b3e064d4f12c2fe5a3b176
diff --git a/runtime/gc/heap.cc b/runtime/gc/heap.cc
index 90ee955..1a32a9a 100644
--- a/runtime/gc/heap.cc
+++ b/runtime/gc/heap.cc
@@ -1157,13 +1157,29 @@
     ptr = TryToAllocate<true, false>(self, allocator, alloc_size, bytes_allocated, usable_size);
   }
 
+  collector::GcType tried_type = next_gc_type_;
+  if (ptr == nullptr) {
+    const bool gc_ran =
+        CollectGarbageInternal(tried_type, kGcCauseForAlloc, false) != collector::kGcTypeNone;
+    if (was_default_allocator && allocator != GetCurrentAllocator()) {
+      *klass = sirt_klass.get();
+      return nullptr;
+    }
+    if (gc_ran) {
+      ptr = TryToAllocate<true, false>(self, allocator, alloc_size, bytes_allocated, usable_size);
+    }
+  }
+
   // Loop through our different Gc types and try to Gc until we get enough free memory.
   for (collector::GcType gc_type : gc_plan_) {
     if (ptr != nullptr) {
       break;
     }
+    if (gc_type == tried_type) {
+      continue;
+    }
     // Attempt to run the collector, if we succeed, re-try the allocation.
-    bool gc_ran =
+    const bool gc_ran =
         CollectGarbageInternal(gc_type, kGcCauseForAlloc, false) != collector::kGcTypeNone;
     if (was_default_allocator && allocator != GetCurrentAllocator()) {
       *klass = sirt_klass.get();