[MetaSchedule] Add logging of usage of tensor intrinsics

src/meta_schedule/schedule_rule/multi_level_tiling.cc

@@ -87,6 +87,7 @@ void MultiLevelTilingNode::InitializeWithTuneContext(const TuneContext& context)
       TVM_PY_LOG(INFO, context->logging_func) << "'thread_warp_size' is not defined in the target";
     }
   }
+  logging_func = context->logging_func;
 }
 
 // Entry of the mega rule; Inherited from ScheduleRuleNode

src/meta_schedule/schedule_rule/multi_level_tiling.h

@@ -186,6 +186,8 @@ class MultiLevelTilingNode : public ScheduleRuleNode {
   int thread_warp_size_;
   /*! \brief The maximum number of threads to be used size of a thread warp */
   int max_threads_per_block_;
+  /*! \brief The logging function */
+  PackedFunc logging_func;
 
   void VisitAttrs(tvm::AttrVisitor* v) {
     v->Visit("structure", &structure);

src/meta_schedule/schedule_rule/multi_level_tiling_tensor_core.cc

@@ -198,9 +198,12 @@ Array<Schedule> MultiLevelTilingTensorCoreNode::Apply(const Schedule& sch,
   }
   Array<Schedule> results;
   for (auto&& state : ApplySubRules(initial_states)) {
+    TVM_PY_LOG(INFO, logging_func) << "Sketch " << results.size() << ": tensorizing with "
+                                   << state.as<TensorCoreStateNode>()->intrin_group.compute_intrin;
     results.push_back(std::move(state->sch));
   }
   if (results.empty()) {
+    TVM_PY_LOG(INFO, logging_func) << "The workload cannot be tensorized.";
     return {original_sch};
   }
   return results;
@@ -276,8 +279,8 @@ std::vector<State> MultiLevelTilingTensorCoreNode::AddReadReuseTensorCore(
     } else if (dtype.is_int() && dtype.bits() == 8) {
       sch->StorageAlign(cache_read, 0, -2, 32, 16);
     } else {
-      LOG(WARNING) << "StorageAlign is not applied for data type " << dtype
-                   << ", shared memory accesses might be inefficient.";
+      TVM_PY_LOG(WARNING, logging_func) << "StorageAlign is not applied for data type " << dtype
+                                        << ", shared memory accesses might be inefficient.";
     }
   }
   return {state};

src/meta_schedule/schedule_rule/multi_level_tiling_with_intrin.cc

@@ -49,14 +49,17 @@ class MultiLevelTilingWithIntrinNode : public MultiLevelTilingNode {
   Array<tir::Schedule> Apply(const tir::Schedule& sch, const tir::BlockRV& block_rv) final {
     auto desc_func = tir::TensorIntrin::Get(intrin_name)->desc;
     if (!CheckAutoTensorizeApplicable(sch, block_rv, desc_func)) {
+      TVM_PY_LOG(INFO, logging_func) << "The workload cannot be tensorized.";
       return {sch};
     }
 
     auto res = MultiLevelTilingNode::Apply(sch->Copy(), block_rv);
 
     if (res.empty()) {
+      TVM_PY_LOG(INFO, logging_func) << "The workload cannot be tensorized.";
       return {sch};
     }
+    TVM_PY_LOG(INFO, logging_func) << "Tensorizing with " << intrin_name;
     return res;
   }