Stitch in loop passes in codegen pipeline

Author: optimisan

llvm/include/llvm/Passes/CodeGenPassBuilder.h

@@ -110,13 +110,16 @@
 #include "llvm/Target/CGPassBuilderOption.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Transforms/CFGuard.h"
+#include "llvm/Transforms/ObjCARC.h"
 #include "llvm/Transforms/Scalar/ConstantHoisting.h"
 #include "llvm/Transforms/Scalar/LoopPassManager.h"
 #include "llvm/Transforms/Scalar/LoopStrengthReduce.h"
+#include "llvm/Transforms/Scalar/LoopTermFold.h"
 #include "llvm/Transforms/Scalar/LowerConstantIntrinsics.h"
 #include "llvm/Transforms/Scalar/MergeICmps.h"
 #include "llvm/Transforms/Scalar/PartiallyInlineLibCalls.h"
 #include "llvm/Transforms/Scalar/ScalarizeMaskedMemIntrin.h"
+#include "llvm/Transforms/Utils/CanonicalizeFreezeInLoops.h"
 #include "llvm/Transforms/Utils/EntryExitInstrumenter.h"
 #include "llvm/Transforms/Utils/LowerInvoke.h"
 #include <cassert>
@@ -691,7 +694,12 @@ void CodeGenPassBuilder<Derived, TargetMachineT>::addIRPasses(
 
   // Run loop strength reduction before anything else.
   if (getOptLevel() != CodeGenOptLevel::None && !Opt.DisableLSR) {
-    addPass(createFunctionToLoopPassAdaptor(LoopStrengthReducePass(),
+    LoopPassManager LPM;
+    LPM.addPass(CanonicalizeFreezeInLoopsPass());
+    LPM.addPass(LoopStrengthReducePass());
+    if (Opt.EnableLoopTermFold)
+      LPM.addPass(LoopTermFoldPass());
+    addPass(createFunctionToLoopPassAdaptor(std::move(LPM),
                                             /*UseMemorySSA=*/true));
   }
 
@@ -736,7 +744,8 @@ void CodeGenPassBuilder<Derived, TargetMachineT>::addIRPasses(
   addPass(ScalarizeMaskedMemIntrinPass());
 
   // Expand reduction intrinsics into shuffle sequences if the target wants to.
-  addPass(ExpandReductionsPass());
+  if (!Opt.DisableExpandReductions)
+    addPass(ExpandReductionsPass());
 
   // Convert conditional moves to conditional jumps when profitable.
   if (getOptLevel() != CodeGenOptLevel::None && !Opt.DisableSelectOptimize)
@@ -811,6 +820,9 @@ void CodeGenPassBuilder<Derived, TargetMachineT>::addISelPrepare(
     AddIRPass &addPass) const {
   derived().addPreISel(addPass);
 
+  if (getOptLevel() != CodeGenOptLevel::None)
+    addPass(ObjCARCContractPass());
+
   addPass(CallBrPreparePass());
   // Add both the safe stack and the stack protection passes: each of them will
   // only protect functions that have corresponding attributes.

llvm/include/llvm/Target/CGPassBuilderOption.h

@@ -51,6 +51,8 @@ struct CGPassBuilderOption {
   bool EnableMachineFunctionSplitter = false;
   bool EnableSinkAndFold = false;
   bool EnableTailMerge = true;
+  /// Enable LoopTermFold immediately after LSR
+  bool EnableLoopTermFold = false;
   bool MISchedPostRA = false;
   bool EarlyLiveIntervals = false;
   bool GCEmptyBlocks = false;

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

@@ -101,7 +101,9 @@
 #include "llvm/Transforms/Scalar/FlattenCFG.h"
 #include "llvm/Transforms/Scalar/GVN.h"
 #include "llvm/Transforms/Scalar/InferAddressSpaces.h"
+#include "llvm/Transforms/Scalar/LICM.h"
 #include "llvm/Transforms/Scalar/LoopDataPrefetch.h"
+#include "llvm/Transforms/Scalar/LoopPassManager.h"
 #include "llvm/Transforms/Scalar/NaryReassociate.h"
 #include "llvm/Transforms/Scalar/SeparateConstOffsetFromGEP.h"
 #include "llvm/Transforms/Scalar/Sink.h"
@@ -2021,7 +2023,11 @@ void AMDGPUCodeGenPassBuilder::addIRPasses(AddIRPass &addPass) const {
     // TODO: May want to move later or split into an early and late one.
     addPass(AMDGPUCodeGenPreparePass(TM));
 
-    // TODO: LICM
+    // Try to hoist loop invariant parts of divisions AMDGPUCodeGenPrepare may
+    // have expanded.
+    if (TM.getOptLevel() > CodeGenOptLevel::Less)
+      addPass(createFunctionToLoopPassAdaptor(LICMPass(LICMOptions()),
+                                              /*UseMemorySSA=*/true));
   }
 
   Base::addIRPasses(addPass);