promoteToSharedBelow: disallow promotion anywhere below thread mapping

Shared memory is accessible from all threads.  If promotion is requested
below the thread mapping, the scoped tensor reference groups are
specific to threads, guaranteeing no reuse between different threads.
It does not make sense to use shared memory in such cases.  Furthermore,
if one attempted to use shared memory below thread mapping, e.g., to
hide global memory latency, one would have to account for the
pre-existing thread mapping when emitting shared<->global memory copies.

Author: ftynse

tc/core/polyhedral/cuda/memory_promotion_heuristic.cc

@@ -440,6 +440,17 @@ void promoteToSharedBelow(
     size_t& remainingMemory) {
   auto root = scop.scheduleRoot();
 
+  // Promotion to shared below threads does not make sense because the computed
+  // groups would be specific to threads thus not benefiting from coalescing or
+  // inter-thread communication through shared memory (use registers instead).
+  auto ancestors = node->ancestors(root);
+  ancestors.push_back(node);
+  for (auto ancestor : ancestors) {
+    if (isMappingTo<mapping::ThreadId>(ancestor)) {
+      throw promotion::IncorrectScope(
+          "shared memory promotion below thread mapping");
+    }
+  }
   // Children of a sequence/set band must be filters, but promotion would
   // insert an extension node.
   if (node->as<detail::ScheduleTreeSequence>() ||

test/test_cuda_mapper_memory_promotion.cc

@@ -439,12 +439,10 @@ TEST_F(MapperMemoryPromotionRAW, fitAtOuterDepths) {
 
 TEST_F(MapperMemoryPromotionRAW, throwIfCopiesBelowThreads) {
   EXPECT_THROW(
-      makeWithSharedGreedy(42, 40, 64, 64, 3, 8192),
-      promotion::PromotionBelowThreadsException);
+      makeWithSharedGreedy(42, 40, 64, 64, 3, 8192), promotion::IncorrectScope);
 
   EXPECT_THROW(
-      makeWithSharedGreedy(42, 40, 64, 64, 4, 8192),
-      promotion::PromotionBelowThreadsException);
+      makeWithSharedGreedy(42, 40, 64, 64, 4, 8192), promotion::IncorrectScope);
 }
 
 class MatMulBias : public TestMapper {