[LoopIdiom] Basic OptimizationRemarkEmitter handling

[lebedev.ri]

Summary

I'm adding ORE to memset/memcpy formation, with tests,
but mainly this is split off from D61144.

[lebedev.ri]

Actually generate check lines

[thegameg]

lib/Transforms/Scalar/LoopIdiomRecognize.cpp

980

Can we make the function name an argument of the remark? Something like (haven't tried to see if it actually works):

ORE.emit([&]() {
    return OptimizationRemark(DEBUG_TYPE, "ProcessLoopStridedStore",
                              NewCall->getDebugLoc(), Preheader)
           << "Transformed loop-strided store into a call to "
            << ore::NV(NewCall->getCalledFunction()) << "() function";
  });

And I guess the same applies to the other remark.

This way, the diagnostic will still show up in the same way but the serialized version will have the function (with its debug loc if available) as a separate argument.

[lebedev.ri]

Use ore::NV() to report the newly called function

[thegameg]

LGTM, thanks!

[lebedev.ri]

In D62631#1522166, @thegameg wrote:

LGTM, thanks!

Hm, that was fast, thank you for the review!
I'll land this tomorrow.

lib/Transforms/Scalar/LoopIdiomRecognize.cpp

 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/Analysis/LoopAccessAnalysis.h"
 #include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Analysis/LoopPass.h"
 #include "llvm/Analysis/MemoryLocation.h"
+#include "llvm/Analysis/OptimizationRemarkEmitter.h"
 #include "llvm/Analysis/ScalarEvolution.h"
 #include "llvm/Analysis/ScalarEvolutionExpander.h"
 #include "llvm/Analysis/ScalarEvolutionExpressions.h"
 #include "llvm/Analysis/TargetLibraryInfo.h"
 #include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/IR/Attributes.h"
 #include "llvm/IR/BasicBlock.h"
class LoopIdiomRecognize {
   Loop *CurLoop = nullptr;
   AliasAnalysis *AA;
   DominatorTree *DT;
   LoopInfo *LI;
   ScalarEvolution *SE;
   TargetLibraryInfo *TLI;
   const TargetTransformInfo *TTI;
   const DataLayout *DL;
+  OptimizationRemarkEmitter &ORE;
   bool ApplyCodeSizeHeuristics;
 
 public:
   explicit LoopIdiomRecognize(AliasAnalysis *AA, DominatorTree *DT,
                               LoopInfo *LI, ScalarEvolution *SE,
                               TargetLibraryInfo *TLI,
                               const TargetTransformInfo *TTI,
-                              const DataLayout *DL)
-      : AA(AA), DT(DT), LI(LI), SE(SE), TLI(TLI), TTI(TTI), DL(DL) {}
+                              const DataLayout *DL,
+                              OptimizationRemarkEmitter &ORE)
+      : AA(AA), DT(DT), LI(LI), SE(SE), TLI(TLI), TTI(TTI), DL(DL), ORE(ORE) {}
 
   bool runOnLoop(Loop *L);
 
 private:
   using StoreList = SmallVector<StoreInst *, 8>;
   using StoreListMap = MapVector<Value *, StoreList>;
 
   StoreListMap StoreRefsForMemset;
bool runOnLoop(Loop *L, LPPassManager &LPM) override {
     ScalarEvolution *SE = &getAnalysis<ScalarEvolutionWrapperPass>().getSE();
     TargetLibraryInfo *TLI =
         &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI();
     const TargetTransformInfo *TTI =
         &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(
             *L->getHeader()->getParent());
     const DataLayout *DL = &L->getHeader()->getModule()->getDataLayout();
 
-    LoopIdiomRecognize LIR(AA, DT, LI, SE, TLI, TTI, DL);
+    // For the old PM, we can't use OptimizationRemarkEmitter as an analysis
+    // pass.  Function analyses need to be preserved across loop transformations
+    // but ORE cannot be preserved (see comment before the pass definition).
+    OptimizationRemarkEmitter ORE(L->getHeader()->getParent());
+
+    LoopIdiomRecognize LIR(AA, DT, LI, SE, TLI, TTI, DL, ORE);
     return LIR.runOnLoop(L);
   }
 
   /// This transformation requires natural loop information & requires that
   /// loop preheaders be inserted into the CFG.
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.addRequired<TargetLibraryInfoWrapperPass>();
     AU.addRequired<TargetTransformInfoWrapperPass>();
     getLoopAnalysisUsage(AU);
   }
 };
 
 } // end anonymous namespace
 
 char LoopIdiomRecognizeLegacyPass::ID = 0;
 
 PreservedAnalyses LoopIdiomRecognizePass::run(Loop &L, LoopAnalysisManager &AM,
                                               LoopStandardAnalysisResults &AR,
                                               LPMUpdater &) {
   const auto *DL = &L.getHeader()->getModule()->getDataLayout();
 
-  LoopIdiomRecognize LIR(&AR.AA, &AR.DT, &AR.LI, &AR.SE, &AR.TLI, &AR.TTI, DL);
+  const auto &FAM =
+      AM.getResult<FunctionAnalysisManagerLoopProxy>(L, AR).getManager();
+  Function *F = L.getHeader()->getParent();
+
+  auto *ORE = FAM.getCachedResult<OptimizationRemarkEmitterAnalysis>(*F);
+  // FIXME: This should probably be optional rather than required.
+  if (!ORE)
+    report_fatal_error(
+        "LoopIdiomRecognizePass: OptimizationRemarkEmitterAnalysis not cached "
+        "at a higher level");
+
+  LoopIdiomRecognize LIR(&AR.AA, &AR.DT, &AR.LI, &AR.SE, &AR.TLI, &AR.TTI, DL,
+                         *ORE);
   if (!LIR.runOnLoop(&L))
     return PreservedAnalyses::all();
 
   return getLoopPassPreservedAnalyses();
 }
 
 INITIALIZE_PASS_BEGIN(LoopIdiomRecognizeLegacyPass, "loop-idiom",
                       "Recognize loop idioms", false, false)
if (SplatValue) {
     NewCall = Builder.CreateCall(MSP, {BasePtr, PatternPtr, NumBytes});
   }
 
   LLVM_DEBUG(dbgs() << "  Formed memset: " << *NewCall << "\n"
                     << "    from store to: " << *Ev << " at: " << *TheStore
                     << "\n");
   NewCall->setDebugLoc(TheStore->getDebugLoc());
 
+  ORE.emit([&]() {
+    return OptimizationRemark(DEBUG_TYPE, "ProcessLoopStridedStore",
+                              NewCall->getDebugLoc(), Preheader)
+           << (Twine("Transformed loop-strided store into a call to ") +
+               NewCall->getCalledFunction()->getName() + "() function")
+                  .str();
+  });
+
   // Okay, the memset has been formed.  Zap the original store and anything that
   // feeds into it.
   for (auto *I : Stores)
     deleteDeadInstruction(I);
   ++NumMemSet;
   return true;
 }
 
NewCall = Builder.CreateElementUnorderedAtomicMemCpy(
         NumBytes, StoreSize);
   }
   NewCall->setDebugLoc(SI->getDebugLoc());
 
   LLVM_DEBUG(dbgs() << "  Formed memcpy: " << *NewCall << "\n"
                     << "    from load ptr=" << *LoadEv << " at: " << *LI << "\n"
                     << "    from store ptr=" << *StoreEv << " at: " << *SI
                     << "\n");
+  ORE.emit([&]() {
+    return OptimizationRemark(DEBUG_TYPE, "ProcessLoopStoreOfLoopLoad",
+                              NewCall->getDebugLoc(), Preheader)
+           << (Twine("Formed a call to ") +
+               NewCall->getCalledFunction()->getName() + "() function")
+                  .str();
+  });
 
   // Okay, the memcpy has been formed.  Zap the original store and anything that
   // feeds into it.
   deleteDeadInstruction(SI);
   ++NumMemCpy;
   return true;
 }
 

test/Transforms/LoopIdiom/memcpy-debugify-remarks.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -basicaa -debugify -loop-idiom -pass-remarks=loop-idiom -pass-remarks-analysis=loop-idiom -verify -verify-each -verify-dom-info -verify-loop-info  < %s -S 2>&1 | FileCheck %s
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; Check that everything still works when debuginfo is present, and that it is reasonably propagated.
+
+; CHECK: remark: <stdin>:6:1: Formed a call to llvm.memcpy.p0i8.p0i8.i64() function
+
+define void @test6_dest_align(i32* noalias align 1 %Base, i32* noalias align 4 %Dest, i64 %Size) nounwind ssp {
+bb.nph:
+  br label %for.body
+
+for.body:                                         ; preds = %bb.nph, %for.body
+  %indvar = phi i64 [ 0, %bb.nph ], [ %indvar.next, %for.body ]
+  %I.0.014 = getelementptr i32, i32* %Base, i64 %indvar
+  %DestI = getelementptr i32, i32* %Dest, i64 %indvar
+  %V = load i32, i32* %I.0.014, align 1
+  store i32 %V, i32* %DestI, align 4
+  %indvar.next = add i64 %indvar, 1
+  %exitcond = icmp eq i64 %indvar.next, %Size
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}

test/Transforms/LoopIdiom/memset-debugify-remarks.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -basicaa -debugify -loop-idiom -pass-remarks=loop-idiom -pass-remarks-analysis=loop-idiom -verify -verify-each -verify-dom-info -verify-loop-info  < %s -S 2>&1 | FileCheck %s
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; Check that everything still works when debuginfo is present, and that it is reasonably propagated.
+
+; void my_basic_memset(char* begin, char* end, char value) {
+;   for( ; begin != end; ++begin)
+;     *begin = value;
+; }
+
+; CHECK: remark: <stdin>:4:1: Transformed loop-strided store into a call to llvm.memset.p0i8.i64() function
+
+define void @_Z15my_basic_memsetPcS_c(i8* %ptr, i8* %end, i8 %value) {
+; CHECK-LABEL: @_Z15my_basic_memsetPcS_c(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[PTR1:%.*]] = ptrtoint i8* [[PTR:%.*]] to i64
+; CHECK-NEXT:    [[CMP3:%.*]] = icmp eq i8* [[PTR]], [[END:%.*]], !dbg !15
+; CHECK-NEXT:    call void @llvm.dbg.value(metadata i1 [[CMP3]], metadata !9, metadata !DIExpression()), !dbg !15
+; CHECK-NEXT:    br i1 [[CMP3]], label [[FOR_END:%.*]], label [[FOR_BODY_PREHEADER:%.*]], !dbg !16
+; CHECK:       for.body.preheader:
+; CHECK-NEXT:    [[TMP0:%.*]] = sub i64 0, [[PTR1]], !dbg !17
+; CHECK-NEXT:    [[SCEVGEP:%.*]] = getelementptr i8, i8* [[END]], i64 [[TMP0]], !dbg !17
+; CHECK-NEXT:    [[SCEVGEP2:%.*]] = ptrtoint i8* [[SCEVGEP]] to i64
+; CHECK-NEXT:    call void @llvm.memset.p0i8.i64(i8* align 1 [[PTR]], i8 [[VALUE:%.*]], i64 [[SCEVGEP2]], i1 false), !dbg !17
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]], !dbg !17
+; CHECK:       for.body:
+; CHECK-NEXT:    [[PTR_ADDR_04:%.*]] = phi i8* [ [[INCDEC_PTR:%.*]], [[FOR_BODY]] ], [ [[PTR]], [[FOR_BODY_PREHEADER]] ], !dbg !18
+; CHECK-NEXT:    call void @llvm.dbg.value(metadata i8* [[PTR_ADDR_04]], metadata !11, metadata !DIExpression()), !dbg !18
+; CHECK-NEXT:    [[INCDEC_PTR]] = getelementptr inbounds i8, i8* [[PTR_ADDR_04]], i64 1, !dbg !19
+; CHECK-NEXT:    call void @llvm.dbg.value(metadata i8* [[INCDEC_PTR]], metadata !13, metadata !DIExpression()), !dbg !19
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i8* [[INCDEC_PTR]], [[END]], !dbg !20
+; CHECK-NEXT:    call void @llvm.dbg.value(metadata i1 [[CMP]], metadata !14, metadata !DIExpression()), !dbg !20
+; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_END_LOOPEXIT:%.*]], label [[FOR_BODY]], !dbg !21
+; CHECK:       for.end.loopexit:
+; CHECK-NEXT:    br label [[FOR_END]], !dbg !22
+; CHECK:       for.end:
+; CHECK-NEXT:    ret void, !dbg !22
+;
+entry:
+  %cmp3 = icmp eq i8* %ptr, %end
+  br i1 %cmp3, label %for.end, label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %ptr.addr.04 = phi i8* [ %incdec.ptr, %for.body ], [ %ptr, %entry ]
+  store i8 %value, i8* %ptr.addr.04, align 1
+  %incdec.ptr = getelementptr inbounds i8, i8* %ptr.addr.04, i64 1
+  %cmp = icmp eq i8* %incdec.ptr, %end
+  br i1 %cmp, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}

test/Transforms/LoopIdiom/nontemporal_store.ll

 ; RUN: opt -loop-idiom < %s -S | FileCheck %s
-; RUN: opt -aa-pipeline=basic-aa -passes='require<aa>,require<targetir>,require<scalar-evolution>,loop(loop-idiom)' < %s -S | FileCheck %s
+; RUN: opt -aa-pipeline=basic-aa -passes='require<aa>,require<targetir>,require<scalar-evolution>,require<opt-remark-emit>,loop(loop-idiom)' < %s -S | FileCheck %s
 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "x86_64-unknown-linux-gnu"
 
 %struct.bigBlock_t = type { [256 x <4 x float>] }
 
 ; CHECK-LABEL: @test(
 ; CHECK-NOT: llvm.memset
 define void @test(%struct.bigBlock_t* %p) {