JlmToMlirConverterTests.cpp

Go to the documentation of this file.

/*
 * Copyright 2024 Louis Maurin <louis7maurin@gmail.com>
 * Copyright 2024 Magnus Själander <work@sjalander.com>
 * See COPYING for terms of redistribution.
 */
 
#include <gtest/gtest.h>
 
#include <jlm/llvm/ir/operators/lambda.hpp>
#include <jlm/llvm/ir/RvsdgModule.hpp>
#include <jlm/llvm/ir/types.hpp>
#include <jlm/mlir/backend/JlmToMlirConverter.hpp>
#include <jlm/rvsdg/bitstring/comparison.hpp>
#include <jlm/rvsdg/bitstring/constant.hpp>
 
TEST(JlmToMlirConverterTests, TestLambda)
{
  using namespace jlm::llvm;
  using namespace mlir::rvsdg;
 
  auto rvsdgModule = LlvmRvsdgModule::Create(jlm::util::FilePath(""), "", "");
  auto graph = &rvsdgModule->Rvsdg();
 
  {
    // Setup the function
    std::cout << "Function Setup" << std::endl;
    auto functionType = jlm::rvsdg::FunctionType::Create(
        { IOStateType::Create(), MemoryStateType::Create() },
        { jlm::rvsdg::BitType::Create(32), IOStateType::Create(), MemoryStateType::Create() });
 
    auto lambda = jlm::rvsdg::LambdaNode::Create(
        graph->GetRootRegion(),
        LlvmLambdaOperation::Create(functionType, "test", Linkage::externalLinkage));
    auto iOStateArgument = lambda->GetFunctionArguments()[0];
    auto memoryStateArgument = lambda->GetFunctionArguments()[1];
 
    auto constant = &jlm::rvsdg::BitConstantOperation::create(*lambda->subregion(), { 32, 4 });
 
    lambda->finalize({ constant, iOStateArgument, memoryStateArgument });
 
    // Convert the RVSDG to MLIR
    std::cout << "Convert to MLIR" << std::endl;
    jlm::mlir::JlmToMlirConverter mlirgen;
    auto omega = mlirgen.ConvertModule(*rvsdgModule);
 
    // Validate the generated MLIR
    std::cout << "Validate MLIR" << std::endl;
    auto & omegaRegion = omega.getRegion();
    EXPECT_EQ(omegaRegion.getBlocks().size(), 1u);
    auto & omegaBlock = omegaRegion.front();
    // Lamda + terminating operation
    EXPECT_EQ(omegaBlock.getOperations().size(), 2u);
    auto & mlirLambda = omegaBlock.front();
    EXPECT_TRUE(
        mlirLambda.getName().getStringRef().equals(mlir::rvsdg::LambdaNode::getOperationName()));
 
    // Verify function name
    std::cout << "Verify function name" << std::endl;
    auto functionNameAttribute = mlirLambda.getAttr(::llvm::StringRef("sym_name"));
    auto * functionName = static_cast<mlir::StringAttr *>(&functionNameAttribute);
    auto string = functionName->getValue().str();
    EXPECT_EQ(string, "test");
 
    // Verify function signature
    std::cout << "Verify function signature" << std::endl;
 
    auto result = mlirLambda.getResult(0).getType();
    EXPECT_EQ(result.getTypeID(), mlir::FunctionType::getTypeID());
 
    auto lambdaOp = ::mlir::dyn_cast<::mlir::rvsdg::LambdaNode>(&mlirLambda);
 
    auto lamdbaTerminator = lambdaOp.getRegion().front().getTerminator();
    auto lambdaResult = mlir::dyn_cast<mlir::rvsdg::LambdaResult>(lamdbaTerminator);
    EXPECT_NE(lambdaResult, nullptr);
    lambdaResult->dump();
 
    std::vector<mlir::Type> arguments;
    for (auto argument : lambdaOp->getRegion(0).getArguments())
    {
      arguments.push_back(argument.getType());
    }
    EXPECT_EQ(arguments[0].getTypeID(), IOStateEdgeType::getTypeID());
    EXPECT_EQ(arguments[1].getTypeID(), MemStateEdgeType::getTypeID());
    std::vector<mlir::Type> results;
    for (auto returnType : lambdaResult->getOperandTypes())
    {
      results.push_back(returnType);
    }
    EXPECT_TRUE(results[0].isa<mlir::IntegerType>());
    EXPECT_TRUE(results[1].isa<mlir::rvsdg::IOStateEdgeType>());
    EXPECT_TRUE(results[2].isa<mlir::rvsdg::MemStateEdgeType>());
 
    auto & lambdaRegion = mlirLambda.getRegion(0);
    auto & lambdaBlock = lambdaRegion.front();
    // Bitconstant + terminating operation
    EXPECT_EQ(lambdaBlock.getOperations().size(), 2u);
    EXPECT_TRUE(lambdaBlock.front().getName().getStringRef().equals(
        mlir::arith::ConstantIntOp::getOperationName()));
 
    omega->destroy();
  }
}
 
static void
useChainsUpTraverse(mlir::Operation * operation, std::vector<llvm::StringRef> definingOperations)
{
  if (definingOperations.empty())
    return;
  std::cout << "Checking if operation: "
            << operation->getOperand(0).getDefiningOp()->getName().getStringRef().data()
            << " is equal to: " << definingOperations.back().data() << std::endl;
  EXPECT_TRUE(operation->getOperand(0).getDefiningOp()->getName().getStringRef().equals(
      definingOperations.back()));
  definingOperations.pop_back();
  useChainsUpTraverse(operation->getOperand(0).getDefiningOp(), definingOperations);
}
 
TEST(JlmToMlirConverterTests, TestAddOperation)
{
  using namespace jlm::llvm;
  using namespace mlir::rvsdg;
 
  auto rvsdgModule = LlvmRvsdgModule::Create(jlm::util::FilePath(""), "", "");
  auto graph = &rvsdgModule->Rvsdg();
 
  {
    // Setup the function
    std::cout << "Function Setup" << std::endl;
    auto functionType = jlm::rvsdg::FunctionType::Create(
        { IOStateType::Create(), MemoryStateType::Create() },
        { jlm::rvsdg::BitType::Create(32), IOStateType::Create(), MemoryStateType::Create() });
 
    auto lambda = jlm::rvsdg::LambdaNode::Create(
        graph->GetRootRegion(),
        LlvmLambdaOperation::Create(functionType, "test", Linkage::externalLinkage));
    auto iOStateArgument = lambda->GetFunctionArguments()[0];
    auto memoryStateArgument = lambda->GetFunctionArguments()[1];
 
    // Create add operation
    std::cout << "Add Operation" << std::endl;
    auto constant1 = &jlm::rvsdg::BitConstantOperation::create(*lambda->subregion(), { 32, 4 });
    auto constant2 = &jlm::rvsdg::BitConstantOperation::create(*lambda->subregion(), { 32, 5 });
    auto add = jlm::rvsdg::bitadd_op::create(32, constant1, constant2);
 
    lambda->finalize({ add, iOStateArgument, memoryStateArgument });
 
    // Convert the RVSDG to MLIR
    std::cout << "Convert to MLIR" << std::endl;
    jlm::mlir::JlmToMlirConverter mlirgen;
    auto omega = mlirgen.ConvertModule(*rvsdgModule);
 
    // Checking blocks and operations count
    std::cout << "Checking blocks and operations count" << std::endl;
    auto & omegaRegion = omega.getRegion();
    EXPECT_EQ(omegaRegion.getBlocks().size(), 1u);
    auto & omegaBlock = omegaRegion.front();
    // Lamda + terminating operation
    EXPECT_EQ(omegaBlock.getOperations().size(), 2u);
 
    // Checking lambda block operations
    std::cout << "Checking lambda block operations" << std::endl;
    auto & mlirLambda = omegaBlock.front();
    auto & lambdaRegion = mlirLambda.getRegion(0);
    auto & lambdaBlock = lambdaRegion.front();
    // 2 Bits contants + add + terminating operation
    EXPECT_EQ(lambdaBlock.getOperations().size(), 4u);
 
    // Checking lambda block operations types
    std::cout << "Checking lambda block operations types" << std::endl;
    std::vector<mlir::Operation *> operations;
    for (auto & operation : lambdaBlock.getOperations())
    {
      operations.push_back(&operation);
    }
 
    unsigned constCount = 0;
    for (auto & operation : operations)
    {
      if (operation->getName().getStringRef().equals(mlir::rvsdg::LambdaResult::getOperationName()))
        continue;
      if (operation->getName().getStringRef().equals(
              mlir::arith::ConstantIntOp::getOperationName()))
      {
        constCount++;
        continue;
      }
      // Checking add operation
      std::cout << "Checking add operation" << std::endl;
      EXPECT_TRUE(operation->getName().getStringRef().equals(
          mlir::arith::AddIOp::getOperationName())); // Last remaining operation is the add
                                                     // operation
      EXPECT_EQ(operation->getNumOperands(), 2u);
      auto addOperand1 = operation->getOperand(0);
      auto addOperand2 = operation->getOperand(1);
      EXPECT_TRUE(addOperand1.getType().isInteger(32));
      EXPECT_TRUE(addOperand2.getType().isInteger(32));
    }
    EXPECT_EQ(constCount, 2u);
 
    useChainsUpTraverse(
        &lambdaBlock.getOperations().back(),
        { mlir::arith::ConstantIntOp::getOperationName(),
          mlir::arith::AddIOp::getOperationName() });
 
    omega->destroy();
  }
}
 
TEST(JlmToMlirConverterTests, TestComZeroExt)
{
  using namespace jlm::llvm;
  using namespace mlir::rvsdg;
 
  auto rvsdgModule = LlvmRvsdgModule::Create(jlm::util::FilePath(""), "", "");
  auto graph = &rvsdgModule->Rvsdg();
 
  {
    // Setup the function
    std::cout << "Function Setup" << std::endl;
    auto functionType = jlm::rvsdg::FunctionType::Create(
        { IOStateType::Create(), MemoryStateType::Create() },
        { jlm::rvsdg::BitType::Create(1), IOStateType::Create(), MemoryStateType::Create() });
 
    auto lambda = jlm::rvsdg::LambdaNode::Create(
        graph->GetRootRegion(),
        LlvmLambdaOperation::Create(functionType, "test", Linkage::externalLinkage));
    auto iOStateArgument = lambda->GetFunctionArguments()[0];
    auto memoryStateArgument = lambda->GetFunctionArguments()[1];
 
    // Create add operation
    std::cout << "Add Operation" << std::endl;
    auto & constant1 = jlm::rvsdg::BitConstantOperation::create(*lambda->subregion(), { 8, 4 });
    jlm::rvsdg::BitConstantOperation::create(*lambda->subregion(), { 16, 5 }); // Unused constant
    jlm::rvsdg::BitConstantOperation::create(*lambda->subregion(), { 16, 6 }); // Unused constant
 
    // zero extension of constant1
    auto & zeroExt = ZExtOperation::create(16, constant1);
 
    auto mul = jlm::rvsdg::bitmul_op::create(16, &zeroExt, &zeroExt);
 
    auto comp = jlm::rvsdg::bitsgt_op::create(16, mul, mul);
 
    lambda->finalize({ comp, iOStateArgument, memoryStateArgument });
 
    // Convert the RVSDG to MLIR
    std::cout << "Convert to MLIR" << std::endl;
    jlm::mlir::JlmToMlirConverter mlirgen;
    auto omega = mlirgen.ConvertModule(*rvsdgModule);
 
    // Checking blocks and operations count
    std::cout << "Checking blocks and operations count" << std::endl;
    auto & omegaRegion = omega.getRegion();
    EXPECT_EQ(omegaRegion.getBlocks().size(), 1u);
    auto & omegaBlock = omegaRegion.front();
    // Lamda + terminating operation
    EXPECT_EQ(omegaBlock.getOperations().size(), 2u);
 
    // Checking lambda block operations
    std::cout << "Checking lambda block operations" << std::endl;
    auto & mlirLambda = omegaBlock.front();
    auto & lambdaRegion = mlirLambda.getRegion(0);
    auto & lambdaBlock = lambdaRegion.front();
    // 3 Bits contants + ZeroExt + Mul + Comp + terminating operation
    EXPECT_EQ(lambdaBlock.getOperations().size(), 7u);
 
    // Checking lambda block operations types
    std::cout << "Checking lambda block operations types" << std::endl;
    std::vector<mlir::Operation *> operations;
    for (auto & operation : lambdaBlock.getOperations())
    {
      std::cout << "Operation: " << operation.getName().getStringRef().data() << std::endl;
      operations.push_back(&operation);
    }
 
    unsigned constCount = 0;
    unsigned extCount = 0;
    unsigned mulCount = 0;
    unsigned compCount = 0;
    for (auto & operation : operations)
    {
      if (operation->getName().getStringRef().equals(mlir::rvsdg::LambdaResult::getOperationName()))
        continue;
      if (operation->getName().getStringRef().equals(
              mlir::arith::ConstantIntOp::getOperationName()))
      {
        EXPECT_TRUE(
            operation->getResult(0).getType().isInteger(8)
            || operation->getResult(0).getType().isInteger(16));
        constCount++;
        continue;
      }
      if (operation->getName().getStringRef().equals(mlir::arith::ExtUIOp::getOperationName()))
      {
        EXPECT_EQ(operation->getNumOperands(), 1u);
        EXPECT_TRUE(operation->getOperand(0).getType().isInteger(8));
        EXPECT_EQ(operation->getNumResults(), 1u);
        EXPECT_TRUE(operation->getResult(0).getType().isInteger(16));
        extCount++;
        continue;
      }
      if (operation->getName().getStringRef().equals(mlir::arith::MulIOp::getOperationName()))
      {
        EXPECT_EQ(operation->getNumOperands(), 2u);
        EXPECT_TRUE(operation->getOperand(0).getType().isInteger(16));
        EXPECT_TRUE(operation->getOperand(1).getType().isInteger(16));
        EXPECT_EQ(operation->getNumResults(), 1u);
        EXPECT_TRUE(operation->getResult(0).getType().isInteger(16));
        mulCount++;
        continue;
      }
      if (operation->getName().getStringRef().equals(mlir::arith::CmpIOp::getOperationName()))
      {
        auto comparisonOp = mlir::cast<mlir::arith::CmpIOp>(operation);
        EXPECT_EQ(comparisonOp.getPredicate(), mlir::arith::CmpIPredicate::sgt);
        EXPECT_EQ(operation->getNumOperands(), 2u);
        EXPECT_TRUE(operation->getOperand(0).getType().isInteger(16));
        EXPECT_TRUE(operation->getOperand(1).getType().isInteger(16));
        EXPECT_EQ(operation->getNumResults(), 1u);
        compCount++;
        continue;
      }
      FAIL();
    }
 
    // Check counts
    std::cout << "Checking counts" << std::endl;
    EXPECT_EQ(constCount, 3u);
    EXPECT_EQ(extCount, 1u);
    EXPECT_EQ(mulCount, 1u);
    EXPECT_EQ(compCount, 1u);
 
    useChainsUpTraverse(
        &lambdaBlock.getOperations().back(),
        { mlir::arith::ConstantIntOp::getOperationName(),
          mlir::arith::ExtUIOp::getOperationName(),
          mlir::arith::MulIOp::getOperationName(),
          mlir::arith::CmpIOp::getOperationName() });
 
    omega->destroy();
  }
}
 
TEST(JlmToMlirConverterTests, TestMatch)
{
  using namespace jlm::llvm;
  using namespace mlir::rvsdg;
 
  auto rvsdgModule = LlvmRvsdgModule::Create(jlm::util::FilePath(""), "", "");
  auto graph = &rvsdgModule->Rvsdg();
 
  {
    // Setup the function
    std::cout << "Function Setup" << std::endl;
    auto functionType = jlm::rvsdg::FunctionType::Create(
        { IOStateType::Create(), MemoryStateType::Create() },
        { jlm::rvsdg::ControlType::Create(2), IOStateType::Create(), MemoryStateType::Create() });
 
    auto lambda = jlm::rvsdg::LambdaNode::Create(
        graph->GetRootRegion(),
        LlvmLambdaOperation::Create(functionType, "test", Linkage::externalLinkage));
    auto iOStateArgument = lambda->GetFunctionArguments()[0];
    auto memoryStateArgument = lambda->GetFunctionArguments()[1];
 
    // Create a match operation
    std::cout << "Match Operation" << std::endl;
    auto predicateConst = &jlm::rvsdg::BitConstantOperation::create(*lambda->subregion(), { 8, 4 });
 
    auto match =
        jlm::rvsdg::MatchOperation::Create(*predicateConst, { { 4, 0 }, { 5, 1 }, { 6, 1 } }, 2, 2);
 
    lambda->finalize({ match, iOStateArgument, memoryStateArgument });
 
    // Convert the RVSDG to MLIR
    std::cout << "Convert to MLIR" << std::endl;
    jlm::mlir::JlmToMlirConverter mlirgen;
    auto omega = mlirgen.ConvertModule(*rvsdgModule);
 
    // Checking blocks and operations count
    std::cout << "Checking blocks and operations count" << std::endl;
    auto & omegaRegion = omega.getRegion();
    EXPECT_EQ(omegaRegion.getBlocks().size(), 1u);
    auto & omegaBlock = omegaRegion.front();
    // Lamda + terminating operation
    EXPECT_EQ(omegaBlock.getOperations().size(), 2u);
 
    // Checking lambda block operations
    std::cout << "Checking lambda block operations" << std::endl;
    auto & mlirLambda = omegaBlock.front();
    auto & lambdaRegion = mlirLambda.getRegion(0);
    auto & lambdaBlock = lambdaRegion.front();
    // 1 Bits contants + Match + terminating operation
    EXPECT_EQ(lambdaBlock.getOperations().size(), 3u);
 
    bool matchFound = false;
    for (auto & operation : lambdaBlock.getOperations())
    {
      if (mlir::isa<mlir::rvsdg::Match>(operation))
      {
        matchFound = true;
        std::cout << "Checking match operation" << std::endl;
        auto matchOp = mlir::cast<mlir::rvsdg::Match>(operation);
 
        EXPECT_TRUE(mlir::isa<mlir::arith::ConstantIntOp>(matchOp.getInput().getDefiningOp()));
        auto constant = mlir::cast<mlir::arith::ConstantIntOp>(matchOp.getInput().getDefiningOp());
        EXPECT_EQ(constant.value(), 4u);
        EXPECT_TRUE(constant.getType().isInteger(8));
 
        auto mapping = matchOp.getMapping();
        mapping.dump();
        // 3 alternatives + default
        EXPECT_EQ(mapping.size(), 4u);
 
        // ** region check alternatives *$
        for (auto & attr : mapping)
        {
          EXPECT_TRUE(attr.isa<::mlir::rvsdg::MatchRuleAttr>());
          auto matchRuleAttr = attr.cast<::mlir::rvsdg::MatchRuleAttr>();
          if (matchRuleAttr.isDefault())
          {
            EXPECT_EQ(matchRuleAttr.getIndex(), 2u);
            EXPECT_TRUE(matchRuleAttr.getValues().empty());
            continue;
          }
 
          EXPECT_EQ(matchRuleAttr.getValues().size(), 1u);
 
          const int64_t value = matchRuleAttr.getValues().front();
 
          EXPECT_TRUE(
              (matchRuleAttr.getIndex() == 0 && value == 4)
              || (matchRuleAttr.getIndex() == 1 && (value == 5 || value == 6)));
        }
        // ** endregion check alternatives **
      }
    }
    EXPECT_TRUE(matchFound);
 
    omega->destroy();
  }
}
 
TEST(JlmToMlirConverterTests, TestGamma)
{
  using namespace jlm::llvm;
  using namespace mlir::rvsdg;
 
  auto rvsdgModule = LlvmRvsdgModule::Create(jlm::util::FilePath(""), "", "");
  auto graph = &rvsdgModule->Rvsdg();
 
  {
 
    // Create a gamma operation
    std::cout << "Gamma Operation" << std::endl;
    auto CtrlConstant = &jlm::rvsdg::ControlConstantOperation::create(graph->GetRootRegion(), 3, 1);
    auto entryvar1 = &jlm::rvsdg::BitConstantOperation::create(graph->GetRootRegion(), { 32, 5 });
    auto entryvar2 = &jlm::rvsdg::BitConstantOperation::create(graph->GetRootRegion(), { 32, 6 });
    auto rvsdgGammaNode = jlm::rvsdg::GammaNode::create(
        CtrlConstant, // predicate
        3             // nalternatives
    );
 
    rvsdgGammaNode->AddEntryVar(entryvar1);
    rvsdgGammaNode->AddEntryVar(entryvar2);
 
    std::vector<jlm::rvsdg::Output *> exitvars1;
    std::vector<jlm::rvsdg::Output *> exitvars2;
    for (int i = 0; i < 3; i++)
    {
      exitvars1.push_back(
          &jlm::rvsdg::BitConstantOperation::create(*rvsdgGammaNode->subregion(i), { 32, i + 1 }));
      exitvars2.push_back(&jlm::rvsdg::BitConstantOperation::create(
          *rvsdgGammaNode->subregion(i),
          { 32, 10 * (i + 1) }));
    }
 
    rvsdgGammaNode->AddExitVar(exitvars1);
    rvsdgGammaNode->AddExitVar(exitvars2);
 
    // Convert the RVSDG to MLIR
    std::cout << "Convert to MLIR" << std::endl;
    jlm::mlir::JlmToMlirConverter mlirgen;
    auto omega = mlirgen.ConvertModule(*rvsdgModule);
 
    // Checking blocks and operations count
    std::cout << "Checking blocks and operations count" << std::endl;
    auto & omegaRegion = omega.getRegion();
    EXPECT_EQ(omegaRegion.getBlocks().size(), 1u);
    auto & omegaBlock = omegaRegion.front();
    // 1 control + 2 constants + gamma + terminating operation
    EXPECT_EQ(omegaBlock.getOperations().size(), 5u);
 
    bool gammaFound = false;
    for (auto & operation : omegaBlock.getOperations())
    {
      if (mlir::isa<mlir::rvsdg::GammaNode>(operation))
      {
        gammaFound = true;
        std::cout << "Checking gamma operation" << std::endl;
        auto gammaOp = mlir::cast<mlir::rvsdg::GammaNode>(operation);
        EXPECT_EQ(gammaOp.getNumRegions(), 3u);
        // 1 predicate + 2 entryVars
        EXPECT_EQ(gammaOp.getNumOperands(), 3u);
        EXPECT_EQ(gammaOp.getNumResults(), 2u);
 
        std::cout << "Checking gamma predicate" << std::endl;
        EXPECT_TRUE(mlir::isa<mlir::rvsdg::ConstantCtrl>(gammaOp.getPredicate().getDefiningOp()));
        auto controlConstant =
            mlir::cast<mlir::rvsdg::ConstantCtrl>(gammaOp.getPredicate().getDefiningOp());
        EXPECT_EQ(controlConstant.getValue(), 1u);
        EXPECT_TRUE(mlir::isa<mlir::rvsdg::RVSDG_CTRLType>(controlConstant.getType()));
        auto ctrlType = mlir::cast<mlir::rvsdg::RVSDG_CTRLType>(controlConstant.getType());
        EXPECT_EQ(ctrlType.getNumOptions(), 3u);
 
        std::cout << "Checking gamma entryVars" << std::endl;
        auto entryVars = gammaOp.getInputs();
        EXPECT_EQ(entryVars.size(), 2u);
        EXPECT_TRUE(mlir::isa<mlir::arith::ConstantIntOp>(entryVars[0].getDefiningOp()));
        EXPECT_TRUE(mlir::isa<mlir::arith::ConstantIntOp>(entryVars[1].getDefiningOp()));
        auto entryVar1 = mlir::cast<mlir::arith::ConstantIntOp>(entryVars[0].getDefiningOp());
        auto entryVar2 = mlir::cast<mlir::arith::ConstantIntOp>(entryVars[1].getDefiningOp());
        EXPECT_EQ(entryVar1.value(), 5u);
        EXPECT_EQ(entryVar2.value(), 6u);
 
        std::cout << "Checking gamma subregions" << std::endl;
        for (size_t i = 0; i < gammaOp.getNumRegions(); i++)
        {
          EXPECT_EQ(gammaOp.getRegion(i).getBlocks().size(), 1u);
          auto & gammaBlock = gammaOp.getRegion(i).front();
          // 2 bit constants + gamma result
          EXPECT_EQ(gammaBlock.getOperations().size(), 3u);
 
          std::cout << "Checking gamma exitVars" << std::endl;
          auto gammaResult = gammaBlock.getTerminator();
          EXPECT_TRUE(mlir::isa<mlir::rvsdg::GammaResult>(gammaResult));
          auto gammaResultOp = mlir::cast<mlir::rvsdg::GammaResult>(gammaResult);
          EXPECT_EQ(gammaResultOp.getNumOperands(), 2u);
          for (size_t j = 0; j < gammaResultOp.getNumOperands(); j++)
          {
            EXPECT_TRUE(
                mlir::isa<mlir::arith::ConstantIntOp>(gammaResultOp.getOperand(j).getDefiningOp()));
            auto constant =
                mlir::cast<mlir::arith::ConstantIntOp>(gammaResultOp.getOperand(j).getDefiningOp());
            EXPECT_EQ(static_cast<size_t>(constant.value()), (1 - j) * (i + 1) + 10 * (i + 1) * j);
          }
        }
      }
    }
    EXPECT_TRUE(gammaFound);
    omega->destroy();
  }
}
 
TEST(JlmToMlirConverterTests, TestTheta)
{
  using namespace jlm::llvm;
  using namespace mlir::rvsdg;
 
  auto rvsdgModule = LlvmRvsdgModule::Create(jlm::util::FilePath(""), "", "");
  auto graph = &rvsdgModule->Rvsdg();
 
  {
    // Create a theta operation
    std::cout << "Theta Operation" << std::endl;
    auto entryvar1 = &jlm::rvsdg::BitConstantOperation::create(graph->GetRootRegion(), { 32, 5 });
    auto entryvar2 = &jlm::rvsdg::BitConstantOperation::create(graph->GetRootRegion(), { 32, 6 });
    jlm::rvsdg::ThetaNode * rvsdgThetaNode = jlm::rvsdg::ThetaNode::create(&graph->GetRootRegion());
 
    auto predicate =
        &jlm::rvsdg::ControlConstantOperation::create(*rvsdgThetaNode->subregion(), 2, 0);
 
    rvsdgThetaNode->AddLoopVar(entryvar1);
    rvsdgThetaNode->AddLoopVar(entryvar2);
    rvsdgThetaNode->set_predicate(predicate);
 
    // Convert the RVSDG to MLIR
    std::cout << "Convert to MLIR" << std::endl;
    jlm::mlir::JlmToMlirConverter mlirgen;
    auto omega = mlirgen.ConvertModule(*rvsdgModule);
 
    // Checking blocks and operations count
    std::cout << "Checking blocks and operations count" << std::endl;
    auto & omegaRegion = omega.getRegion();
    EXPECT_EQ(omegaRegion.getBlocks().size(), 1u);
    auto & omegaBlock = omegaRegion.front();
    // 1 theta + 1 predicate + 2 constants
    EXPECT_EQ(omegaBlock.getOperations().size(), 4u);
 
    bool thetaFound = false;
    for (auto & operation : omegaBlock.getOperations())
    {
      if (mlir::isa<mlir::rvsdg::ThetaNode>(operation))
      {
        thetaFound = true;
        std::cout << "Checking theta operation" << std::endl;
        auto thetaOp = mlir::cast<mlir::rvsdg::ThetaNode>(operation);
        // 2 loop vars
        EXPECT_EQ(thetaOp.getNumOperands(), 2u);
        EXPECT_EQ(thetaOp.getNumResults(), 2u);
 
        auto & thetaBlock = thetaOp.getRegion().front();
        auto thetaResult = thetaBlock.getTerminator();
 
        EXPECT_TRUE(mlir::isa<mlir::rvsdg::ThetaResult>(thetaResult));
        auto thetaResultOp = mlir::cast<mlir::rvsdg::ThetaResult>(thetaResult);
 
        std::cout << "Checking theta predicate" << std::endl;
 
        EXPECT_TRUE(
            mlir::isa<mlir::rvsdg::ConstantCtrl>(thetaResultOp.getPredicate().getDefiningOp()));
        auto controlConstant =
            mlir::cast<mlir::rvsdg::ConstantCtrl>(thetaResultOp.getPredicate().getDefiningOp());
 
        EXPECT_EQ(controlConstant.getValue(), 0u);
 
        EXPECT_TRUE(mlir::isa<mlir::rvsdg::RVSDG_CTRLType>(controlConstant.getType()));
        auto ctrlType = mlir::cast<mlir::rvsdg::RVSDG_CTRLType>(controlConstant.getType());
        EXPECT_EQ(ctrlType.getNumOptions(), 2u);
 
        std::cout << "Checking theta loop vars" << std::endl;
        auto loopVars = thetaOp.getInputs();
        EXPECT_EQ(loopVars.size(), 2u);
        EXPECT_TRUE(mlir::isa<mlir::arith::ConstantIntOp>(loopVars[0].getDefiningOp()));
        EXPECT_TRUE(mlir::isa<mlir::arith::ConstantIntOp>(loopVars[1].getDefiningOp()));
        auto loopVar1 = mlir::cast<mlir::arith::ConstantIntOp>(loopVars[0].getDefiningOp());
        auto loopVar2 = mlir::cast<mlir::arith::ConstantIntOp>(loopVars[1].getDefiningOp());
        EXPECT_EQ(loopVar1.value(), 5u);
        EXPECT_EQ(loopVar2.value(), 6u);
 
        // Theta result, constant control predicate
        EXPECT_EQ(thetaBlock.getOperations().size(), 2u);
 
        std::cout << "Checking loop exitVars" << std::endl;
        std::cout << thetaResultOp.getNumOperands() << std::endl;
 
        std::cout << "Checking theta subregion" << std::endl;
 
        // Two arguments and predicate
        EXPECT_EQ(thetaResultOp.getNumOperands(), 3u);
      }
    }
    // }
    EXPECT_TRUE(thetaFound);
    omega->destroy();
  }
}