[llvm] [OpenMP] [IR Builder] Changes to Support Scan Operation (PR #136035)

[Michael Kruse via llvm-commits]

================
@@ -5360,6 +5362,144 @@ TEST_F(OpenMPIRBuilderTest, CreateReductions) {
   EXPECT_TRUE(findGEPZeroOne(ReductionFn->getArg(1), FirstRHS, SecondRHS));
 }
 
+static void createScan(llvm::Value *scanVar, llvm::Type *scanType,
+                       OpenMPIRBuilder &OMPBuilder, IRBuilder<> &Builder,
+                       OpenMPIRBuilder::LocationDescription Loc,
+                       OpenMPIRBuilder::InsertPointTy &allocaIP,
+                       ScanInfo *&ScanRedInfo) {
+  using InsertPointTy = OpenMPIRBuilder::InsertPointTy;
+  ASSERT_EXPECTED_INIT(InsertPointTy, retIp,
+                       OMPBuilder.createScan(Loc, allocaIP, {scanVar},
+                                             {scanType}, true, ScanRedInfo));
+  Builder.restoreIP(retIp);
+}
+/*
+ Following is the pseudocode of the code generated by the test case
+ <declare pointer to buffer> ptr
+  size num_iters = 100
+  // temp buffer allocation
+  omp masked {
+    buff = malloc(num_iters*scanvarstype)
+    *ptr = buff
+  }
+ barrier;
+  // input phase loop
+  for (i: 0..<num_iters>) {
+    <input phase>;
+    buffer = *ptr;
+    buffer[i] = red;
+  }
+  // scan reduction
+  omp masked
+  {
+    for (int k = 0; k != ceil(log2(num_iters)); ++k) {
+      i=pow(2,k)
+      for (size cnt = last_iter; cnt >= i; --cnt) {
+        buffer = *ptr;
+        buffer[cnt] op= buffer[cnt-i];
+      }
+    }
+  }
+ barrier;
+ // scan phase loop
+  for (0..<num_iters>) {
+    buffer = *ptr;
+    red = buffer[i] ;
+    <scan phase>;
+  }
+  // temp buffer deletion
+  omp masked {
+    free(*ptr)
+  }
+  barrier;
+*/
+TEST_F(OpenMPIRBuilderTest, ScanReduction) {
+  using InsertPointTy = OpenMPIRBuilder::InsertPointTy;
+  OpenMPIRBuilder OMPBuilder(*M);
+  OMPBuilder.initialize();
+  IRBuilder<> Builder(BB);
+  OpenMPIRBuilder::LocationDescription Loc({Builder.saveIP(), DL});
+  Value *TripCount = F->getArg(0);
+  Type *LCTy = TripCount->getType();
+  Value *StartVal = ConstantInt::get(LCTy, 1);
+  Value *StopVal = ConstantInt::get(LCTy, 100);
+  Value *Step = ConstantInt::get(LCTy, 1);
+  auto AllocaIP = Builder.saveIP();
+
+  llvm::Value *ScanVar = Builder.CreateAlloca(Builder.getFloatTy());
+  llvm::Value *OrigVar = Builder.CreateAlloca(Builder.getFloatTy());
+  unsigned NumBodiesGenerated = 0;
+  ScanInfo *ScanRedInfo;
+  ASSERT_EXPECTED_INIT(ScanInfo *, ScanInformation,
+                       OMPBuilder.scanInfoInitialize());
+  ScanRedInfo = ScanInformation;
+  auto LoopBodyGenCB = [&](InsertPointTy CodeGenIP, llvm::Value *LC) {
+    NumBodiesGenerated += 1;
+    Builder.restoreIP(CodeGenIP);
+    createScan(ScanVar, Builder.getFloatTy(), OMPBuilder, Builder, Loc,
+               AllocaIP, ScanRedInfo);
+    return Error::success();
+  };
+  llvm::SmallVector<CanonicalLoopInfo *> loops;
+  ASSERT_EXPECTED_INIT(llvm::SmallVector<CanonicalLoopInfo *>, loopvec,
+                       OMPBuilder.createCanonicalScanLoops(
+                           Loc, LoopBodyGenCB, StartVal, StopVal, Step, false,
+                           false, Builder.saveIP(), "scan", ScanRedInfo));
+  loops = loopvec;
+  CanonicalLoopInfo *InputLoop = loops.front();
+  CanonicalLoopInfo *ScanLoop = loops.back();
+  Builder.restoreIP(ScanLoop->getAfterIP());
+  InputLoop->assertOK();
+  ScanLoop->assertOK();
+
+  EXPECT_EQ(ScanLoop->getAfter(), Builder.GetInsertBlock());
+  EXPECT_EQ(NumBodiesGenerated, 2U);
+  SmallVector<OpenMPIRBuilder::ReductionInfo> ReductionInfos = {
+      {Builder.getFloatTy(), OrigVar, ScanVar,
+       /*EvaluationKind=*/OpenMPIRBuilder::EvalKind::Scalar, sumReduction,
+       /*ReductionGenClang=*/nullptr, sumAtomicReduction}};
+  OpenMPIRBuilder::LocationDescription RedLoc({InputLoop->getAfterIP(), DL});
+  llvm::BasicBlock *Cont = splitBB(Builder, false, "omp.scan.loop.cont");
+  ASSERT_EXPECTED_INIT(
+      InsertPointTy, retIp,
+      OMPBuilder.emitScanReduction(RedLoc, ReductionInfos, ScanRedInfo));
+  Builder.restoreIP(retIp);
+  Builder.CreateBr(Cont);
+  Builder.SetInsertPoint(Cont);
+  unsigned NumMallocs = 0;
+  unsigned NumFrees = 0;
+  unsigned NumMasked = 0;
+  unsigned NumEndMasked = 0;
+  unsigned NumLog = 0;
+  unsigned NumCeil = 0;
+  for (Instruction &I : instructions(F)) {
+    if (isa<CallInst>(I)) {
+      CallInst *Call = dyn_cast<CallInst>(&I);
+      auto Name = Call->getCalledFunction()->getName();
----------------
Meinersbur wrote:

No almost-always-auto

https://github.com/llvm/llvm-project/pull/136035