[MachinePipeliner] Limit the number of stores in BB (#154940)

The dependency analysis in MachinePipeliner checks dependencies for
every pair of store instructions in the target basic block. This means
the time complexity of the analysis is `O(N^2)`, where `N` is the number
of store instructions. Therefore, compilation time can become
significantly long when there are too many store instructions.

To mitigate it, this patch introduces logic to count the number of store
instructions at the beginning of the pipeliner and bail out if it
exceeds the threshold. The default value if the threshold should be
large enough. Thus, in most practical cases where the pipeliner is
beneficial, this patch should not cause any performance regression.

Related issue: #150262

Author: kasuga-fj

llvm/lib/CodeGen/MachinePipeliner.cpp

@@ -110,6 +110,7 @@ STATISTIC(NumFailZeroMII, "Pipeliner abort due to zero MII");
 STATISTIC(NumFailNoSchedule, "Pipeliner abort due to no schedule found");
 STATISTIC(NumFailZeroStage, "Pipeliner abort due to zero stage");
 STATISTIC(NumFailLargeMaxStage, "Pipeliner abort due to too many stages");
+STATISTIC(NumFailTooManyStores, "Pipeliner abort due to too many stores");
 
 /// A command line option to turn software pipelining on or off.
 static cl::opt<bool> EnableSWP("enable-pipeliner", cl::Hidden, cl::init(true),
@@ -193,6 +194,13 @@ static cl::opt<bool>
     MVECodeGen("pipeliner-mve-cg", cl::Hidden, cl::init(false),
                cl::desc("Use the MVE code generator for software pipelining"));
 
+/// A command line argument to limit the number of store instructions in the
+/// target basic block.
+static cl::opt<unsigned> SwpMaxNumStores(
+    "pipeliner-max-num-stores",
+    cl::desc("Maximum number of stores allwed in the target loop."), cl::Hidden,
+    cl::init(200));
+
 namespace llvm {
 
 // A command line option to enable the CopyToPhi DAG mutation.
@@ -544,6 +552,23 @@ bool MachinePipeliner::canPipelineLoop(MachineLoop &L) {
     return false;
   }
 
+  unsigned NumStores = 0;
+  for (MachineInstr &MI : *L.getHeader())
+    if (MI.mayStore())
+      ++NumStores;
+  if (NumStores > SwpMaxNumStores) {
+    LLVM_DEBUG(dbgs() << "Too many stores\n");
+    NumFailTooManyStores++;
+    ORE->emit([&]() {
+      return MachineOptimizationRemarkAnalysis(DEBUG_TYPE, "canPipelineLoop",
+                                               L.getStartLoc(), L.getHeader())
+             << "Too many store instructions in the loop: "
+             << ore::NV("NumStores", NumStores) << " > "
+             << ore::NV("SwpMaxNumStores", SwpMaxNumStores) << ".";
+    });
+    return false;
+  }
+
   // Remove any subregisters from inputs to phi nodes.
   preprocessPhiNodes(*L.getHeader());
   return true;