minor fixes

alibaba · May 23, 2024 · ccb2aaa · ccb2aaa
1 parent 6613d88
commit ccb2aaa
Show file tree

Hide file tree

Showing 2 changed files with 13 additions and 11 deletions.
diff --git a/tao_compiler/mlir/disc/transforms/disc_op_schedule.cc b/tao_compiler/mlir/disc/transforms/disc_op_schedule.cc
@@ -354,7 +354,7 @@ class ScheduleGraph {
   explicit ScheduleGraph(std::vector<Operation*>& post_order_instructions,
                          LatencyEstimator* latency_estimator,
                          AsyncTracker* async_tracker) {
-    InitilizeGrpahTopology(post_order_instructions, latency_estimator,
+    InitilizeGraphTopology(post_order_instructions, latency_estimator,
                            async_tracker);
     InitializeGraphAnalysis(latency_estimator, async_tracker);
   }
@@ -497,7 +497,7 @@ class ScheduleGraph {
     }
   }
 
-  void InitilizeGrpahTopology(std::vector<Operation*>& post_order_instructions,
+  void InitilizeGraphTopology(std::vector<Operation*>& post_order_instructions,
                               LatencyEstimator* latency_estimator,
                               AsyncTracker* async_tracker) {
     original_order_ = post_order_instructions;
@@ -957,7 +957,6 @@ struct DiscOpSchedulePass : public DiscOpSchedulePassBase<DiscOpSchedulePass> {
       return;
     }
 
-    bool need_schedule = false;
     // Initialization
     latency_estimator_ = new LatencyEstimator();
     async_tracker_ = new AsyncTracker(scheduler_config_);
@@ -966,20 +965,21 @@ struct DiscOpSchedulePass : public DiscOpSchedulePassBase<DiscOpSchedulePass> {
     for (auto& block : main_func.getBody()) {
       for (auto& op : block) {
         original_op_sequence.push_back(&op);
-        need_schedule =
-            need_schedule || async_tracker_->IsSupportedAsyncDone(&op);
       }
     }
 
-    if (!need_schedule) {
-      return;
-    }
-
     scheduler_core_ = new SchedulerCore(latency_estimator_, async_tracker_,
                                         scheduler_config_);
     auto scheduled_op_sequence =
         scheduler_core_->ScheduleComputation(original_op_sequence);
 
+    if (scheduled_op_sequence.size() != original_op_sequence.size()) {
+      main_func.emitError(
+          "Schedule Pass error, scheduled_op_sequence should have the same num "
+          "of ops with original_op_sequence");
+      signalPassFailure();
+      return;
+    }
     for (auto& block : main_func.getBody()) {
       for (int op_idx = 0; op_idx < scheduled_op_sequence.size(); op_idx++) {
         scheduled_op_sequence[op_idx]->moveBefore(&block.front());

diff --git a/tao_compiler/mlir/disc/transforms/lhlo_legalize_roots_to_loops.cc b/tao_compiler/mlir/disc/transforms/lhlo_legalize_roots_to_loops.cc
@@ -5712,9 +5712,11 @@ struct DiscLhloLegalizeRootsToParallelLoops
     // TODO: We should put even single nodes into a fusion by fusion pass
     // Revisit this and walk lmhlo::FusionOp only after the revision done.
     func.walk([&](lmhlo::LmhloOp op) {
-      // Skip the embedded ops in lmhlo.fusion or lmhlo.reduce/scatter
+      // Skip the embedded ops in lmhlo.fusion or lmhlo.reduce/scatter or
+      // lmhlo_disc.args_mutation
       lmhlo::LmhloOp parent = op->getParentOfType<lmhlo::LmhloOp>();
-      if (parent && !isa<lmhlo::FusionOp>(op)) {
+      if (isa<lmhlo_disc::ArgsMutationOp>(op) ||
+          parent && !isa<lmhlo::FusionOp>(op)) {
         return;
       }
       if (isFusionType<FusionType::kStitch>(op) &&