MooreToCore.cpp

Go to the documentation of this file.

//===- MooreToCore.cpp - Moore To Core Conversion Pass --------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This is the main Moore to Core Conversion Pass Implementation.
//
//===----------------------------------------------------------------------===//
 
#include "circt/Conversion/MooreToCore.h"
#include "circt/Dialect/Comb/CombOps.h"
#include "circt/Dialect/Debug/DebugOps.h"
#include "circt/Dialect/HW/HWOps.h"
#include "circt/Dialect/LLHD/IR/LLHDOps.h"
#include "circt/Dialect/Moore/MooreOps.h"
#include "circt/Dialect/Sim/SimOps.h"
#include "circt/Dialect/Verif/VerifOps.h"
#include "mlir/Dialect/ControlFlow/IR/ControlFlowOps.h"
#include "mlir/Dialect/Func/IR/FuncOps.h"
#include "mlir/Dialect/SCF/IR/SCF.h"
#include "mlir/IR/BuiltinDialect.h"
#include "mlir/IR/Iterators.h"
#include "mlir/Interfaces/SideEffectInterfaces.h"
#include "mlir/Pass/Pass.h"
#include "mlir/Transforms/DialectConversion.h"
#include "mlir/Transforms/RegionUtils.h"
#include "llvm/ADT/TypeSwitch.h"
 
namespace circt {
#define GEN_PASS_DEF_CONVERTMOORETOCORE
#include "circt/Conversion/Passes.h.inc"
} // namespace circt
 
using namespace mlir;
using namespace circt;
using namespace moore;
 
using comb::ICmpPredicate;
using llvm::SmallDenseSet;
 
namespace {
 
/// Returns the passed value if the integer width is already correct.
/// Zero-extends if it is too narrow.
/// Truncates if the integer is too wide and the truncated part is zero, if it
/// is not zero it returns the max value integer of target-width.
static Value adjustIntegerWidth(OpBuilder &builder, Value value,
                                uint32_t targetWidth, Location loc) {
  uint32_t intWidth = value.getType().getIntOrFloatBitWidth();
  if (intWidth == targetWidth)
    return value;
 
  if (intWidth < targetWidth) {
    Value zeroExt = builder.create<hw::ConstantOp>(
        loc, builder.getIntegerType(targetWidth - intWidth), 0);
    return builder.create<comb::ConcatOp>(loc, ValueRange{zeroExt, value});
  }
 
  Value hi = builder.create<comb::ExtractOp>(loc, value, targetWidth,
                                             intWidth - targetWidth);
  Value zero = builder.create<hw::ConstantOp>(
      loc, builder.getIntegerType(intWidth - targetWidth), 0);
  Value isZero = builder.create<comb::ICmpOp>(loc, comb::ICmpPredicate::eq, hi,
                                              zero, false);
  Value lo = builder.create<comb::ExtractOp>(loc, value, 0, targetWidth);
  Value max = builder.create<hw::ConstantOp>(
      loc, builder.getIntegerType(targetWidth), -1);
  return builder.create<comb::MuxOp>(loc, isZero, lo, max, false);
}
 
/// Get the ModulePortInfo from a SVModuleOp.
static hw::ModulePortInfo getModulePortInfo(const TypeConverter &typeConverter,
                                            SVModuleOp op) {
  size_t inputNum = 0;
  size_t resultNum = 0;
  auto moduleTy = op.getModuleType();
  SmallVector<hw::PortInfo> inputs, outputs;
  inputs.reserve(moduleTy.getNumInputs());
  outputs.reserve(moduleTy.getNumOutputs());
 
  for (auto port : moduleTy.getPorts()) {
    Type portTy = typeConverter.convertType(port.type);
    if (auto ioTy = dyn_cast_or_null<hw::InOutType>(portTy)) {
      inputs.push_back(hw::PortInfo(
          {{port.name, ioTy.getElementType(), hw::ModulePort::InOut},
           inputNum++,
           {}}));
      continue;
    }
 
    if (port.dir == hw::ModulePort::Direction::Output) {
      outputs.push_back(
          hw::PortInfo({{port.name, portTy, port.dir}, resultNum++, {}}));
    } else {
      // FIXME: Once we support net<...>, ref<...> type to represent type of
      // special port like inout or ref port which is not a input or output
      // port. It can change to generate corresponding types for direction of
      // port or do specified operation to it. Now inout and ref port is treated
      // as input port.
      inputs.push_back(
          hw::PortInfo({{port.name, portTy, port.dir}, inputNum++, {}}));
    }
  }
 
  return hw::ModulePortInfo(inputs, outputs);
}
 
//===----------------------------------------------------------------------===//
// Structural Conversion
//===----------------------------------------------------------------------===//
 
struct SVModuleOpConversion : public OpConversionPattern<SVModuleOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(SVModuleOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.setInsertionPoint(op);
 
    // Create the hw.module to replace moore.module
    auto hwModuleOp =
        rewriter.create<hw::HWModuleOp>(op.getLoc(), op.getSymNameAttr(),
                                        getModulePortInfo(*typeConverter, op));
    // Make hw.module have the same visibility as the moore.module.
    // The entry/top level module is public, otherwise is private.
    SymbolTable::setSymbolVisibility(hwModuleOp,
                                     SymbolTable::getSymbolVisibility(op));
    rewriter.eraseBlock(hwModuleOp.getBodyBlock());
    if (failed(
            rewriter.convertRegionTypes(&op.getBodyRegion(), *typeConverter)))
      return failure();
    rewriter.inlineRegionBefore(op.getBodyRegion(), hwModuleOp.getBodyRegion(),
                                hwModuleOp.getBodyRegion().end());
 
    // Erase the original op
    rewriter.eraseOp(op);
    return success();
  }
};
 
struct OutputOpConversion : public OpConversionPattern<OutputOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(OutputOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<hw::OutputOp>(op, adaptor.getOperands());
    return success();
  }
};
 
struct InstanceOpConversion : public OpConversionPattern<InstanceOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(InstanceOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    auto instName = op.getInstanceNameAttr();
    auto moduleName = op.getModuleNameAttr();
 
    // Create the new hw instanceOp to replace the original one.
    rewriter.setInsertionPoint(op);
    auto instOp = rewriter.create<hw::InstanceOp>(
        op.getLoc(), op.getResultTypes(), instName, moduleName, op.getInputs(),
        op.getInputNamesAttr(), op.getOutputNamesAttr(),
        /*Parameter*/ rewriter.getArrayAttr({}), /*InnerSymbol*/ nullptr,
        /*doNotPrint*/ nullptr);
 
    // Replace uses chain and erase the original op.
    op.replaceAllUsesWith(instOp.getResults());
    rewriter.eraseOp(op);
    return success();
  }
};
 
static void getValuesToObserve(Region *region,
                               function_ref<void(Value)> setInsertionPoint,
                               const TypeConverter *typeConverter,
                               ConversionPatternRewriter &rewriter,
                               SmallVector<Value> &observeValues) {
  SmallDenseSet<Value> alreadyObserved;
  Location loc = region->getLoc();
 
  auto probeIfSignal = [&](Value value) -> Value {
    if (!isa<hw::InOutType>(value.getType()))
      return value;
    return rewriter.create<llhd::PrbOp>(loc, value);
  };
 
  region->getParentOp()->walk<WalkOrder::PreOrder, ForwardDominanceIterator<>>(
      [&](Operation *operation) {
        for (auto value : operation->getOperands()) {
          if (region->isAncestor(value.getParentRegion()))
            continue;
          if (auto *defOp = value.getDefiningOp();
              defOp && defOp->hasTrait<OpTrait::ConstantLike>())
            continue;
          if (!alreadyObserved.insert(value).second)
            continue;
 
          OpBuilder::InsertionGuard g(rewriter);
          if (auto remapped = rewriter.getRemappedValue(value)) {
            setInsertionPoint(remapped);
            observeValues.push_back(probeIfSignal(remapped));
          } else {
            setInsertionPoint(value);
            auto type = typeConverter->convertType(value.getType());
            auto converted = typeConverter->materializeTargetConversion(
                rewriter, loc, type, value);
            observeValues.push_back(probeIfSignal(converted));
          }
        }
      });
}
 
struct ProcedureOpConversion : public OpConversionPattern<ProcedureOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(ProcedureOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    // Collect values to observe before we do any modifications to the region.
    SmallVector<Value> observedValues;
    if (op.getKind() == ProcedureKind::AlwaysComb ||
        op.getKind() == ProcedureKind::AlwaysLatch) {
      auto setInsertionPoint = [&](Value value) {
        rewriter.setInsertionPoint(op);
      };
      getValuesToObserve(&op.getBody(), setInsertionPoint, typeConverter,
                         rewriter, observedValues);
    }
 
    auto loc = op.getLoc();
    if (failed(rewriter.convertRegionTypes(&op.getBody(), *typeConverter)))
      return failure();
 
    // Handle initial and final procedures. These lower to a corresponding
    // `llhd.process` or `llhd.final` op that executes the body and then halts.
    if (op.getKind() == ProcedureKind::Initial ||
        op.getKind() == ProcedureKind::Final) {
      Operation *newOp;
      if (op.getKind() == ProcedureKind::Initial)
        newOp = rewriter.create<llhd::ProcessOp>(loc);
      else
        newOp = rewriter.create<llhd::FinalOp>(loc);
      auto &body = newOp->getRegion(0);
      rewriter.inlineRegionBefore(op.getBody(), body, body.end());
      for (auto returnOp :
           llvm::make_early_inc_range(body.getOps<ReturnOp>())) {
        rewriter.setInsertionPoint(returnOp);
        rewriter.replaceOpWithNewOp<llhd::HaltOp>(returnOp);
      }
      rewriter.eraseOp(op);
      return success();
    }
 
    // All other procedures lower to a an `llhd.process`.
    auto newOp = rewriter.create<llhd::ProcessOp>(loc);
 
    // We need to add an empty entry block because it is not allowed in MLIR to
    // branch back to the entry block. Instead we put the logic in the second
    // block and branch to that.
    rewriter.createBlock(&newOp.getBody());
    auto *block = &op.getBody().front();
    rewriter.create<cf::BranchOp>(loc, block);
    rewriter.inlineRegionBefore(op.getBody(), newOp.getBody(),
                                newOp.getBody().end());
 
    // Add special handling for `always_comb` and `always_latch` procedures.
    // These run once at simulation startup and then implicitly wait for any of
    // the values they access to change before running again. To implement this,
    // we create another basic block that contains the implicit wait, and make
    // all `moore.return` ops branch to that wait block instead of immediately
    // jumping back up to the body.
    if (op.getKind() == ProcedureKind::AlwaysComb ||
        op.getKind() == ProcedureKind::AlwaysLatch) {
      Block *waitBlock = rewriter.createBlock(&newOp.getBody());
      rewriter.create<llhd::WaitOp>(loc, observedValues, Value(), ValueRange{},
                                    block);
      block = waitBlock;
    }
 
    // Make all `moore.return` ops branch back up to the beginning of the
    // process, or the wait block created above for `always_comb` and
    // `always_latch` procedures.
    for (auto returnOp : llvm::make_early_inc_range(newOp.getOps<ReturnOp>())) {
      rewriter.setInsertionPoint(returnOp);
      rewriter.create<cf::BranchOp>(loc, block);
      rewriter.eraseOp(returnOp);
    }
 
    rewriter.eraseOp(op);
    return success();
  }
};
 
struct WaitEventOpConversion : public OpConversionPattern<WaitEventOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(WaitEventOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    // In order to convert the `wait_event` op we need to create three separate
    // blocks at the location of the op:
    //
    // - A "wait" block that reads the current state of any values used to
    //   detect events and then waits until any of those values change. When a
    //   change occurs, control transfers to the "check" block.
    // - A "check" block which is executed after any interesting signal has
    //   changed. This is where any `detect_event` ops read the current state of
    //   interesting values and compare them against their state before the wait
    //   in order to detect an event. If any events were detected, control
    //   transfers to the "resume" block; otherwise control goes back to the
    //   "wait" block.
    // - A "resume" block which holds any ops after the `wait_event` op. This is
    //   where control is expected to resume after an event has happened.
    //
    // Block structure before:
    //     opA
    //     moore.wait_event { ... }
    //     opB
    //
    // Block structure after:
    //     opA
    //     cf.br ^wait
    // ^wait:
    //     <read "before" values>
    //     llhd.wait ^check, ...
    // ^check:
    //     <read "after" values>
    //     <detect edges>
    //     cf.cond_br %event, ^resume, ^wait
    // ^resume:
    //     opB
    auto *resumeBlock =
        rewriter.splitBlock(op->getBlock(), ++Block::iterator(op));
    auto *waitBlock = rewriter.createBlock(resumeBlock);
    auto *checkBlock = rewriter.createBlock(resumeBlock);
 
    auto loc = op.getLoc();
    rewriter.setInsertionPoint(op);
    rewriter.create<cf::BranchOp>(loc, waitBlock);
 
    // We need to inline two copies of the `wait_event`'s body region: one is
    // used to determine the values going into `detect_event` ops before the
    // `llhd.wait`, and one will do the actual event detection after the
    // `llhd.wait`.
    //
    // Create a copy of the entire `wait_event` op in the wait block, which also
    // creates a copy of its region. Take note of all inputs to `detect_event`
    // ops and delete the `detect_event` ops in this copy.
    SmallVector<Value> valuesBefore;
    rewriter.setInsertionPointToEnd(waitBlock);
    auto clonedOp = cast<WaitEventOp>(rewriter.clone(*op));
    for (auto detectOp :
         llvm::make_early_inc_range(clonedOp.getOps<DetectEventOp>())) {
      valuesBefore.push_back(detectOp.getInput());
      rewriter.eraseOp(detectOp);
    }
 
    // Determine the values used during event detection that are defined outside
    // the `wait_event`'s body region. We want to wait for a change on these
    // signals before we check if any interesting event happened.
    SmallVector<Value> observeValues;
    auto setInsertionPointAfterDef = [&](Value value) {
      if (auto *op = value.getDefiningOp())
        rewriter.setInsertionPointAfter(op);
      if (auto arg = dyn_cast<BlockArgument>(value))
        rewriter.setInsertionPointToStart(value.getParentBlock());
    };
 
    getValuesToObserve(&clonedOp.getBody(), setInsertionPointAfterDef,
                       typeConverter, rewriter, observeValues);
 
    // Create the `llhd.wait` op that suspends the current process and waits for
    // a change in the interesting values listed in `observeValues`. When a
    // change is detected, execution resumes in the "check" block.
    auto waitOp = rewriter.create<llhd::WaitOp>(loc, observeValues, Value(),
                                                ValueRange{}, checkBlock);
    rewriter.inlineBlockBefore(&clonedOp.getBody().front(), waitOp);
    rewriter.eraseOp(clonedOp);
 
    // Collect a list of all detect ops and inline the `wait_event` body into
    // the check block.
    SmallVector<DetectEventOp> detectOps(op.getBody().getOps<DetectEventOp>());
    rewriter.inlineBlockBefore(&op.getBody().front(), checkBlock,
                               checkBlock->end());
    rewriter.eraseOp(op);
 
    // Helper function to detect if a certain change occurred between a value
    // before the `llhd.wait` and after.
    auto computeTrigger = [&](Value before, Value after, Edge edge) -> Value {
      assert(before.getType() == after.getType() &&
             "mismatched types after clone op");
      auto beforeType = cast<IntType>(before.getType());
 
      // 9.4.2 IEEE 1800-2017: An edge event shall be detected only on the LSB
      // of the expression
      if (beforeType.getWidth() != 1 && edge != Edge::AnyChange) {
        constexpr int LSB = 0;
        beforeType =
            IntType::get(rewriter.getContext(), 1, beforeType.getDomain());
        before =
            rewriter.create<moore::ExtractOp>(loc, beforeType, before, LSB);
        after = rewriter.create<moore::ExtractOp>(loc, beforeType, after, LSB);
      }
 
      auto intType = rewriter.getIntegerType(beforeType.getWidth());
      before = typeConverter->materializeTargetConversion(rewriter, loc,
                                                          intType, before);
      after = typeConverter->materializeTargetConversion(rewriter, loc, intType,
                                                         after);
 
      if (edge == Edge::AnyChange)
        return rewriter.create<comb::ICmpOp>(loc, ICmpPredicate::ne, before,
                                             after, true);
 
      SmallVector<Value> disjuncts;
      Value trueVal = rewriter.create<hw::ConstantOp>(loc, APInt(1, 1));
 
      if (edge == Edge::PosEdge || edge == Edge::BothEdges) {
        Value notOldVal =
            rewriter.create<comb::XorOp>(loc, before, trueVal, true);
        Value posedge =
            rewriter.create<comb::AndOp>(loc, notOldVal, after, true);
        disjuncts.push_back(posedge);
      }
 
      if (edge == Edge::NegEdge || edge == Edge::BothEdges) {
        Value notCurrVal =
            rewriter.create<comb::XorOp>(loc, after, trueVal, true);
        Value posedge =
            rewriter.create<comb::AndOp>(loc, before, notCurrVal, true);
        disjuncts.push_back(posedge);
      }
 
      return rewriter.createOrFold<comb::OrOp>(loc, disjuncts, true);
    };
 
    // Convert all `detect_event` ops into a check for the corresponding event
    // between the value before and after the `llhd.wait`. The "before" value
    // has been collected into `valuesBefore` in the "wait" block; the "after"
    // value corresponds to the detect op's input.
    SmallVector<Value> triggers;
    for (auto [detectOp, before] : llvm::zip(detectOps, valuesBefore)) {
      if (!isa<IntType>(before.getType()))
        return detectOp->emitError() << "requires int operand";
 
      rewriter.setInsertionPoint(detectOp);
      auto trigger =
          computeTrigger(before, detectOp.getInput(), detectOp.getEdge());
      if (detectOp.getCondition()) {
        auto condition = typeConverter->materializeTargetConversion(
            rewriter, loc, rewriter.getI1Type(), detectOp.getCondition());
        trigger = rewriter.create<comb::AndOp>(loc, trigger, condition, true);
      }
      triggers.push_back(trigger);
      rewriter.eraseOp(detectOp);
    }
 
    // If any `detect_event` op detected an event, branch to the "resume" block
    // which contains any code after the `wait_event` op. If no events were
    // detected, branch back to the "wait" block to wait for the next change on
    // the interesting signals.
    rewriter.setInsertionPointToEnd(checkBlock);
    if (!triggers.empty()) {
      auto triggered = rewriter.createOrFold<comb::OrOp>(loc, triggers, true);
      rewriter.create<cf::CondBranchOp>(loc, triggered, resumeBlock, waitBlock);
    } else {
      rewriter.create<cf::BranchOp>(loc, waitBlock);
    }
 
    return success();
  }
};
 
//===----------------------------------------------------------------------===//
// Declaration Conversion
//===----------------------------------------------------------------------===//
 
struct VariableOpConversion : public OpConversionPattern<VariableOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(VariableOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Location loc = op.getLoc();
    Type resultType = typeConverter->convertType(op.getResult().getType());
    if (!resultType)
      return rewriter.notifyMatchFailure(op.getLoc(), "invalid variable type");
 
    // Determine the initial value of the signal.
    Value init = adaptor.getInitial();
    if (!init) {
      Type elementType = cast<hw::InOutType>(resultType).getElementType();
      int64_t width = hw::getBitWidth(elementType);
      if (width == -1)
        return failure();
 
      // TODO: Once the core dialects support four-valued integers, this code
      // will additionally need to generate an all-X value for four-valued
      // variables.
      Value constZero = rewriter.create<hw::ConstantOp>(loc, APInt(width, 0));
      init = rewriter.createOrFold<hw::BitcastOp>(loc, elementType, constZero);
    }
 
    rewriter.replaceOpWithNewOp<llhd::SignalOp>(op, resultType,
                                                op.getNameAttr(), init);
    return success();
  }
};
 
struct NetOpConversion : public OpConversionPattern<NetOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(NetOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    auto loc = op.getLoc();
    if (op.getKind() != NetKind::Wire)
      return rewriter.notifyMatchFailure(loc, "only wire nets supported");
 
    auto resultType = typeConverter->convertType(op.getResult().getType());
    if (!resultType)
      return rewriter.notifyMatchFailure(loc, "invalid net type");
 
    // TODO: Once the core dialects support four-valued integers, this code
    // will additionally need to generate an all-X value for four-valued nets.
    auto elementType = cast<hw::InOutType>(resultType).getElementType();
    int64_t width = hw::getBitWidth(elementType);
    if (width == -1)
      return failure();
    auto constZero = rewriter.create<hw::ConstantOp>(loc, APInt(width, 0));
    auto init =
        rewriter.createOrFold<hw::BitcastOp>(loc, elementType, constZero);
 
    auto signal = rewriter.replaceOpWithNewOp<llhd::SignalOp>(
        op, resultType, op.getNameAttr(), init);
 
    if (auto assignedValue = adaptor.getAssignment()) {
      auto timeAttr = llhd::TimeAttr::get(resultType.getContext(), 0U,
                                          llvm::StringRef("ns"), 0, 1);
      auto time = rewriter.create<llhd::ConstantTimeOp>(loc, timeAttr);
      rewriter.create<llhd::DrvOp>(loc, signal, assignedValue, time, Value{});
    }
 
    return success();
  }
};
 
//===----------------------------------------------------------------------===//
// Expression Conversion
//===----------------------------------------------------------------------===//
 
struct ConstantOpConv : public OpConversionPattern<ConstantOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(ConstantOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    // FIXME: Discard unknown bits and map them to 0 for now.
    auto value = op.getValue().toAPInt(false);
    auto type = rewriter.getIntegerType(value.getBitWidth());
    rewriter.replaceOpWithNewOp<hw::ConstantOp>(
        op, type, rewriter.getIntegerAttr(type, value));
    return success();
  }
};
 
struct StringConstantOpConv : public OpConversionPattern<StringConstantOp> {
  using OpConversionPattern::OpConversionPattern;
  LogicalResult
  matchAndRewrite(moore::StringConstantOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    const auto str = op.getValue();
    const unsigned byteWidth = str.size() * 8;
    const auto resultType =
        typeConverter->convertType(op.getResult().getType());
    if (const auto intType = mlir::dyn_cast<IntegerType>(resultType)) {
      if (intType.getWidth() < byteWidth) {
        return rewriter.notifyMatchFailure(op,
                                           "invalid string constant type size");
      }
    } else {
      return rewriter.notifyMatchFailure(op, "invalid string constant type");
    }
    APInt value(byteWidth, 0);
    for (size_t i = 0; i < str.size(); ++i) {
      const auto asciiChar = static_cast<uint8_t>(str[i]);
      value |= APInt(byteWidth, asciiChar) << (8 * (str.size() - 1 - i));
    }
    rewriter.replaceOpWithNewOp<hw::ConstantOp>(
        op, resultType, rewriter.getIntegerAttr(resultType, value));
    return success();
  }
};
 
struct ConcatOpConversion : public OpConversionPattern<ConcatOp> {
  using OpConversionPattern::OpConversionPattern;
  LogicalResult
  matchAndRewrite(ConcatOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<comb::ConcatOp>(op, adaptor.getValues());
    return success();
  }
};
 
struct ReplicateOpConversion : public OpConversionPattern<ReplicateOp> {
  using OpConversionPattern::OpConversionPattern;
  LogicalResult
  matchAndRewrite(ReplicateOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getResult().getType());
 
    rewriter.replaceOpWithNewOp<comb::ReplicateOp>(op, resultType,
                                                   adaptor.getValue());
    return success();
  }
};
 
struct ExtractOpConversion : public OpConversionPattern<ExtractOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(ExtractOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    // TODO: properly handle out-of-bounds accesses
    Type resultType = typeConverter->convertType(op.getResult().getType());
    Type inputType = adaptor.getInput().getType();
 
    if (isa<IntegerType>(inputType)) {
      rewriter.replaceOpWithNewOp<comb::ExtractOp>(
          op, resultType, adaptor.getInput(), adaptor.getLowBit());
      return success();
    }
 
    if (auto arrTy = dyn_cast<hw::ArrayType>(inputType)) {
      int64_t width = llvm::Log2_64_Ceil(arrTy.getNumElements());
      Value idx = rewriter.create<hw::ConstantOp>(
          op.getLoc(), rewriter.getIntegerType(width), adaptor.getLowBit());
      if (isa<hw::ArrayType>(resultType)) {
        rewriter.replaceOpWithNewOp<hw::ArraySliceOp>(op, resultType,
                                                      adaptor.getInput(), idx);
        return success();
      }
 
      // Otherwise, it has to be the array's element type
      rewriter.replaceOpWithNewOp<hw::ArrayGetOp>(op, adaptor.getInput(), idx);
      return success();
    }
 
    return failure();
  }
};
 
struct ExtractRefOpConversion : public OpConversionPattern<ExtractRefOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(ExtractRefOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    // TODO: properly handle out-of-bounds accesses
    Type resultType = typeConverter->convertType(op.getResult().getType());
    Type inputType =
        cast<hw::InOutType>(adaptor.getInput().getType()).getElementType();
 
    if (auto intType = dyn_cast<IntegerType>(inputType)) {
      int64_t width = hw::getBitWidth(inputType);
      if (width == -1)
        return failure();
 
      Value lowBit = rewriter.create<hw::ConstantOp>(
          op.getLoc(), rewriter.getIntegerType(llvm::Log2_64_Ceil(width)),
          adaptor.getLowBit());
      rewriter.replaceOpWithNewOp<llhd::SigExtractOp>(
          op, resultType, adaptor.getInput(), lowBit);
      return success();
    }
 
    if (auto arrType = dyn_cast<hw::ArrayType>(inputType)) {
      Value lowBit = rewriter.create<hw::ConstantOp>(
          op.getLoc(),
          rewriter.getIntegerType(llvm::Log2_64_Ceil(arrType.getNumElements())),
          adaptor.getLowBit());
 
      if (isa<hw::ArrayType>(
              cast<hw::InOutType>(resultType).getElementType())) {
        rewriter.replaceOpWithNewOp<llhd::SigArraySliceOp>(
            op, resultType, adaptor.getInput(), lowBit);
        return success();
      }
 
      rewriter.replaceOpWithNewOp<llhd::SigArrayGetOp>(op, adaptor.getInput(),
                                                       lowBit);
      return success();
    }
 
    return failure();
  }
};
 
struct DynExtractOpConversion : public OpConversionPattern<DynExtractOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(DynExtractOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getResult().getType());
    Type inputType = adaptor.getInput().getType();
 
    if (auto intType = dyn_cast<IntegerType>(inputType)) {
      Value amount = adjustIntegerWidth(rewriter, adaptor.getLowBit(),
                                        intType.getWidth(), op->getLoc());
      Value value = rewriter.create<comb::ShrUOp>(op->getLoc(),
                                                  adaptor.getInput(), amount);
 
      rewriter.replaceOpWithNewOp<comb::ExtractOp>(op, resultType, value, 0);
      return success();
    }
 
    if (auto arrType = dyn_cast<hw::ArrayType>(inputType)) {
      unsigned idxWidth = llvm::Log2_64_Ceil(arrType.getNumElements());
      Value idx = adjustIntegerWidth(rewriter, adaptor.getLowBit(), idxWidth,
                                     op->getLoc());
 
      if (isa<hw::ArrayType>(resultType)) {
        rewriter.replaceOpWithNewOp<hw::ArraySliceOp>(op, resultType,
                                                      adaptor.getInput(), idx);
        return success();
      }
 
      rewriter.replaceOpWithNewOp<hw::ArrayGetOp>(op, adaptor.getInput(), idx);
      return success();
    }
 
    return failure();
  }
};
 
struct DynExtractRefOpConversion : public OpConversionPattern<DynExtractRefOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(DynExtractRefOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    // TODO: properly handle out-of-bounds accesses
    Type resultType = typeConverter->convertType(op.getResult().getType());
    Type inputType =
        cast<hw::InOutType>(adaptor.getInput().getType()).getElementType();
 
    if (auto intType = dyn_cast<IntegerType>(inputType)) {
      int64_t width = hw::getBitWidth(inputType);
      if (width == -1)
        return failure();
 
      Value amount =
          adjustIntegerWidth(rewriter, adaptor.getLowBit(),
                             llvm::Log2_64_Ceil(width), op->getLoc());
      rewriter.replaceOpWithNewOp<llhd::SigExtractOp>(
          op, resultType, adaptor.getInput(), amount);
      return success();
    }
 
    if (auto arrType = dyn_cast<hw::ArrayType>(inputType)) {
      Value idx = adjustIntegerWidth(
          rewriter, adaptor.getLowBit(),
          llvm::Log2_64_Ceil(arrType.getNumElements()), op->getLoc());
 
      if (isa<hw::ArrayType>(
              cast<hw::InOutType>(resultType).getElementType())) {
        rewriter.replaceOpWithNewOp<llhd::SigArraySliceOp>(
            op, resultType, adaptor.getInput(), idx);
        return success();
      }
 
      rewriter.replaceOpWithNewOp<llhd::SigArrayGetOp>(op, adaptor.getInput(),
                                                       idx);
      return success();
    }
 
    return failure();
  }
};
 
struct StructCreateOpConversion : public OpConversionPattern<StructCreateOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(StructCreateOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getResult().getType());
    rewriter.replaceOpWithNewOp<hw::StructCreateOp>(op, resultType,
                                                    adaptor.getFields());
    return success();
  }
};
 
struct StructExtractOpConversion : public OpConversionPattern<StructExtractOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(StructExtractOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<hw::StructExtractOp>(
        op, adaptor.getInput(), adaptor.getFieldNameAttr());
    return success();
  }
};
 
struct StructExtractRefOpConversion
    : public OpConversionPattern<StructExtractRefOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(StructExtractRefOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<llhd::SigStructExtractOp>(
        op, adaptor.getInput(), adaptor.getFieldNameAttr());
    return success();
  }
};
 
struct ReduceAndOpConversion : public OpConversionPattern<ReduceAndOp> {
  using OpConversionPattern::OpConversionPattern;
  LogicalResult
  matchAndRewrite(ReduceAndOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getInput().getType());
    Value max = rewriter.create<hw::ConstantOp>(op->getLoc(), resultType, -1);
 
    rewriter.replaceOpWithNewOp<comb::ICmpOp>(op, comb::ICmpPredicate::eq,
                                              adaptor.getInput(), max);
    return success();
  }
};
 
struct ReduceOrOpConversion : public OpConversionPattern<ReduceOrOp> {
  using OpConversionPattern::OpConversionPattern;
  LogicalResult
  matchAndRewrite(ReduceOrOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getInput().getType());
    Value zero = rewriter.create<hw::ConstantOp>(op->getLoc(), resultType, 0);
 
    rewriter.replaceOpWithNewOp<comb::ICmpOp>(op, comb::ICmpPredicate::ne,
                                              adaptor.getInput(), zero);
    return success();
  }
};
 
struct ReduceXorOpConversion : public OpConversionPattern<ReduceXorOp> {
  using OpConversionPattern::OpConversionPattern;
  LogicalResult
  matchAndRewrite(ReduceXorOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
 
    rewriter.replaceOpWithNewOp<comb::ParityOp>(op, adaptor.getInput());
    return success();
  }
};
 
struct BoolCastOpConversion : public OpConversionPattern<BoolCastOp> {
  using OpConversionPattern::OpConversionPattern;
  LogicalResult
  matchAndRewrite(BoolCastOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getInput().getType());
    if (isa_and_nonnull<IntegerType>(resultType)) {
      Value zero = rewriter.create<hw::ConstantOp>(op->getLoc(), resultType, 0);
      rewriter.replaceOpWithNewOp<comb::ICmpOp>(op, comb::ICmpPredicate::ne,
                                                adaptor.getInput(), zero);
      return success();
    }
    return failure();
  }
};
 
struct NotOpConversion : public OpConversionPattern<NotOp> {
  using OpConversionPattern::OpConversionPattern;
  LogicalResult
  matchAndRewrite(NotOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType =
        ConversionPattern::typeConverter->convertType(op.getResult().getType());
    Value max = rewriter.create<hw::ConstantOp>(op.getLoc(), resultType, -1);
 
    rewriter.replaceOpWithNewOp<comb::XorOp>(op, adaptor.getInput(), max);
    return success();
  }
};
 
struct NegOpConversion : public OpConversionPattern<NegOp> {
  using OpConversionPattern::OpConversionPattern;
  LogicalResult
  matchAndRewrite(NegOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType =
        ConversionPattern::typeConverter->convertType(op.getResult().getType());
    Value zero = rewriter.create<hw::ConstantOp>(op.getLoc(), resultType, 0);
 
    rewriter.replaceOpWithNewOp<comb::SubOp>(op, zero, adaptor.getInput());
    return success();
  }
};
 
template <typename SourceOp, typename TargetOp>
struct BinaryOpConversion : public OpConversionPattern<SourceOp> {
  using OpConversionPattern<SourceOp>::OpConversionPattern;
  using OpAdaptor = typename SourceOp::Adaptor;
 
  LogicalResult
  matchAndRewrite(SourceOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<TargetOp>(op, adaptor.getLhs(),
                                          adaptor.getRhs(), false);
    return success();
  }
};
 
template <typename SourceOp, ICmpPredicate pred>
struct ICmpOpConversion : public OpConversionPattern<SourceOp> {
  using OpConversionPattern<SourceOp>::OpConversionPattern;
  using OpAdaptor = typename SourceOp::Adaptor;
 
  LogicalResult
  matchAndRewrite(SourceOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType =
        ConversionPattern::typeConverter->convertType(op.getResult().getType());
 
    rewriter.replaceOpWithNewOp<comb::ICmpOp>(
        op, resultType, pred, adaptor.getLhs(), adaptor.getRhs());
    return success();
  }
};
 
template <typename SourceOp, bool withoutX>
struct CaseXZEqOpConversion : public OpConversionPattern<SourceOp> {
  using OpConversionPattern<SourceOp>::OpConversionPattern;
  using OpAdaptor = typename SourceOp::Adaptor;
 
  LogicalResult
  matchAndRewrite(SourceOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    // Check each operand if it is a known constant and extract the X and/or Z
    // bits to be ignored.
    // TODO: Once the core dialects support four-valued integers, we will have
    // to create ops that extract X and Z bits from the operands, since we also
    // have to do the right casez/casex comparison on non-constant inputs.
    unsigned bitWidth = op.getLhs().getType().getWidth();
    auto ignoredBits = APInt::getZero(bitWidth);
    auto detectIgnoredBits = [&](Value value) {
      auto constOp = value.getDefiningOp<ConstantOp>();
      if (!constOp)
        return;
      auto constValue = constOp.getValue();
      if (withoutX)
        ignoredBits |= constValue.getZBits();
      else
        ignoredBits |= constValue.getUnknownBits();
    };
    detectIgnoredBits(op.getLhs());
    detectIgnoredBits(op.getRhs());
 
    // If we have detected any bits to be ignored, mask them in the operands for
    // the comparison.
    Value lhs = adaptor.getLhs();
    Value rhs = adaptor.getRhs();
    if (!ignoredBits.isZero()) {
      ignoredBits.flipAllBits();
      auto maskOp = rewriter.create<hw::ConstantOp>(op.getLoc(), ignoredBits);
      lhs = rewriter.createOrFold<comb::AndOp>(op.getLoc(), lhs, maskOp);
      rhs = rewriter.createOrFold<comb::AndOp>(op.getLoc(), rhs, maskOp);
    }
 
    rewriter.replaceOpWithNewOp<comb::ICmpOp>(op, ICmpPredicate::ceq, lhs, rhs);
    return success();
  }
};
 
//===----------------------------------------------------------------------===//
// Conversions
//===----------------------------------------------------------------------===//
 
struct ConversionOpConversion : public OpConversionPattern<ConversionOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(ConversionOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Location loc = op.getLoc();
    Type resultType = typeConverter->convertType(op.getResult().getType());
    int64_t inputBw = hw::getBitWidth(adaptor.getInput().getType());
    int64_t resultBw = hw::getBitWidth(resultType);
    if (inputBw == -1 || resultBw == -1)
      return failure();
 
    Value input = rewriter.createOrFold<hw::BitcastOp>(
        loc, rewriter.getIntegerType(inputBw), adaptor.getInput());
    Value amount = adjustIntegerWidth(rewriter, input, resultBw, loc);
 
    Value result =
        rewriter.createOrFold<hw::BitcastOp>(loc, resultType, amount);
    rewriter.replaceOp(op, result);
    return success();
  }
};
 
struct TruncOpConversion : public OpConversionPattern<TruncOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(TruncOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<comb::ExtractOp>(op, adaptor.getInput(), 0,
                                                 op.getType().getWidth());
    return success();
  }
};
 
struct ZExtOpConversion : public OpConversionPattern<ZExtOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(ZExtOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    auto targetWidth = op.getType().getWidth();
    auto inputWidth = op.getInput().getType().getWidth();
 
    auto zeroExt = rewriter.create<hw::ConstantOp>(
        op.getLoc(), rewriter.getIntegerType(targetWidth - inputWidth), 0);
 
    rewriter.replaceOpWithNewOp<comb::ConcatOp>(
        op, ValueRange{zeroExt, adaptor.getInput()});
    return success();
  }
};
 
struct SExtOpConversion : public OpConversionPattern<SExtOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(SExtOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    auto type = typeConverter->convertType(op.getType());
    auto value =
        comb::createOrFoldSExt(op.getLoc(), adaptor.getInput(), type, rewriter);
    rewriter.replaceOp(op, value);
    return success();
  }
};
 
//===----------------------------------------------------------------------===//
// Statement Conversion
//===----------------------------------------------------------------------===//
 
struct HWInstanceOpConversion : public OpConversionPattern<hw::InstanceOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(hw::InstanceOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    SmallVector<Type> convResTypes;
    if (typeConverter->convertTypes(op.getResultTypes(), convResTypes).failed())
      return failure();
 
    rewriter.replaceOpWithNewOp<hw::InstanceOp>(
        op, convResTypes, op.getInstanceName(), op.getModuleName(),
        adaptor.getOperands(), op.getArgNames(),
        op.getResultNames(), /*Parameter*/
        rewriter.getArrayAttr({}), /*InnerSymbol*/ nullptr);
 
    return success();
  }
};
 
struct ReturnOpConversion : public OpConversionPattern<func::ReturnOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(func::ReturnOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<func::ReturnOp>(op, adaptor.getOperands());
    return success();
  }
};
 
struct CondBranchOpConversion : public OpConversionPattern<cf::CondBranchOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(cf::CondBranchOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<cf::CondBranchOp>(
        op, adaptor.getCondition(), adaptor.getTrueDestOperands(),
        adaptor.getFalseDestOperands(), op.getTrueDest(), op.getFalseDest());
    return success();
  }
};
 
struct BranchOpConversion : public OpConversionPattern<cf::BranchOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(cf::BranchOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<cf::BranchOp>(op, op.getDest(),
                                              adaptor.getDestOperands());
    return success();
  }
};
 
struct CallOpConversion : public OpConversionPattern<func::CallOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(func::CallOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    SmallVector<Type> convResTypes;
    if (typeConverter->convertTypes(op.getResultTypes(), convResTypes).failed())
      return failure();
    rewriter.replaceOpWithNewOp<func::CallOp>(
        op, adaptor.getCallee(), convResTypes, adaptor.getOperands());
    return success();
  }
};
 
struct UnrealizedConversionCastConversion
    : public OpConversionPattern<UnrealizedConversionCastOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(UnrealizedConversionCastOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    SmallVector<Type> convResTypes;
    if (typeConverter->convertTypes(op.getResultTypes(), convResTypes).failed())
      return failure();
 
    // Drop the cast if the operand and result types agree after type
    // conversion.
    if (convResTypes == adaptor.getOperands().getTypes()) {
      rewriter.replaceOp(op, adaptor.getOperands());
      return success();
    }
 
    rewriter.replaceOpWithNewOp<UnrealizedConversionCastOp>(
        op, convResTypes, adaptor.getOperands());
    return success();
  }
};
 
struct ShlOpConversion : public OpConversionPattern<ShlOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(ShlOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getResult().getType());
 
    // Comb shift operations require the same bit-width for value and amount
    Value amount =
        adjustIntegerWidth(rewriter, adaptor.getAmount(),
                           resultType.getIntOrFloatBitWidth(), op->getLoc());
    rewriter.replaceOpWithNewOp<comb::ShlOp>(op, resultType, adaptor.getValue(),
                                             amount, false);
    return success();
  }
};
 
struct ShrOpConversion : public OpConversionPattern<ShrOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(ShrOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getResult().getType());
 
    // Comb shift operations require the same bit-width for value and amount
    Value amount =
        adjustIntegerWidth(rewriter, adaptor.getAmount(),
                           resultType.getIntOrFloatBitWidth(), op->getLoc());
    rewriter.replaceOpWithNewOp<comb::ShrUOp>(
        op, resultType, adaptor.getValue(), amount, false);
    return success();
  }
};
 
struct PowUOpConversion : public OpConversionPattern<PowUOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(PowUOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getResult().getType());
 
    Location loc = op.getLoc();
    auto intType = cast<IntType>(op.getRhs().getType());
 
    // transform a ** b into scf.for 0 to b step 1 { init *= a }, init = 1
    Type integerType = rewriter.getIntegerType(intType.getWidth());
    Value lowerBound = rewriter.create<hw::ConstantOp>(loc, integerType, 0);
    Value upperBound =
        rewriter.create<ConversionOp>(loc, integerType, op.getRhs());
    Value step = rewriter.create<hw::ConstantOp>(loc, integerType, 1);
 
    Value initVal = rewriter.create<hw::ConstantOp>(loc, resultType, 1);
    Value lhsVal = rewriter.create<ConversionOp>(loc, resultType, op.getLhs());
 
    auto forOp = rewriter.create<scf::ForOp>(
        loc, lowerBound, upperBound, step, ValueRange(initVal),
        [&](OpBuilder &builder, Location loc, Value i, ValueRange iterArgs) {
          Value loopVar = iterArgs.front();
          Value mul = rewriter.create<comb::MulOp>(loc, lhsVal, loopVar);
          rewriter.create<scf::YieldOp>(loc, ValueRange(mul));
        });
 
    rewriter.replaceOp(op, forOp.getResult(0));
 
    return success();
  }
};
 
struct PowSOpConversion : public OpConversionPattern<PowSOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(PowSOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getResult().getType());
 
    Location loc = op.getLoc();
    auto intType = cast<IntType>(op.getRhs().getType());
    // transform a ** b into scf.for 0 to b step 1 { init *= a }, init = 1
    Type integerType = rewriter.getIntegerType(intType.getWidth());
    Value lhsVal = rewriter.create<ConversionOp>(loc, resultType, op.getLhs());
    Value rhsVal = rewriter.create<ConversionOp>(loc, integerType, op.getRhs());
    Value constZero = rewriter.create<hw::ConstantOp>(loc, integerType, 0);
    Value constZeroResult = rewriter.create<hw::ConstantOp>(loc, resultType, 0);
    Value isNegative = rewriter.create<comb::ICmpOp>(loc, ICmpPredicate::slt,
                                                     rhsVal, constZero);
 
    // if the exponent is negative, return 0
    lhsVal =
        rewriter.create<comb::MuxOp>(loc, isNegative, constZeroResult, lhsVal);
    Value upperBound =
        rewriter.create<comb::MuxOp>(loc, isNegative, constZero, rhsVal);
 
    Value lowerBound = constZero;
    Value step = rewriter.create<hw::ConstantOp>(loc, integerType, 1);
    Value initVal = rewriter.create<hw::ConstantOp>(loc, resultType, 1);
 
    auto forOp = rewriter.create<scf::ForOp>(
        loc, lowerBound, upperBound, step, ValueRange(initVal),
        [&](OpBuilder &builder, Location loc, Value i, ValueRange iterArgs) {
          auto loopVar = iterArgs.front();
          auto mul = rewriter.create<comb::MulOp>(loc, lhsVal, loopVar);
          rewriter.create<scf::YieldOp>(loc, ValueRange(mul));
        });
 
    rewriter.replaceOp(op, forOp.getResult(0));
 
    return success();
  }
};
 
struct AShrOpConversion : public OpConversionPattern<AShrOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(AShrOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    Type resultType = typeConverter->convertType(op.getResult().getType());
 
    // Comb shift operations require the same bit-width for value and amount
    Value amount =
        adjustIntegerWidth(rewriter, adaptor.getAmount(),
                           resultType.getIntOrFloatBitWidth(), op->getLoc());
    rewriter.replaceOpWithNewOp<comb::ShrSOp>(
        op, resultType, adaptor.getValue(), amount, false);
    return success();
  }
};
 
struct ReadOpConversion : public OpConversionPattern<ReadOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(ReadOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<llhd::PrbOp>(op, adaptor.getInput());
    return success();
  }
};
 
struct AssignedVariableOpConversion
    : public OpConversionPattern<AssignedVariableOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(AssignedVariableOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<hw::WireOp>(op, adaptor.getInput(),
                                            adaptor.getNameAttr());
    return success();
  }
};
 
template <typename OpTy, unsigned DeltaTime, unsigned EpsilonTime>
struct AssignOpConversion : public OpConversionPattern<OpTy> {
  using OpConversionPattern<OpTy>::OpConversionPattern;
  using OpAdaptor = typename OpTy::Adaptor;
 
  LogicalResult
  matchAndRewrite(OpTy op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    // TODO: When we support delay control in Moore dialect, we need to update
    // this conversion.
    auto timeAttr = llhd::TimeAttr::get(
        op->getContext(), 0U, llvm::StringRef("ns"), DeltaTime, EpsilonTime);
    auto time = rewriter.create<llhd::ConstantTimeOp>(op->getLoc(), timeAttr);
    rewriter.replaceOpWithNewOp<llhd::DrvOp>(op, adaptor.getDst(),
                                             adaptor.getSrc(), time, Value{});
    return success();
  }
};
 
struct ConditionalOpConversion : public OpConversionPattern<ConditionalOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(ConditionalOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    // TODO: This lowering is only correct if the condition is two-valued. If
    // the condition is X or Z, both branches of the conditional must be
    // evaluated and merged with the appropriate lookup table. See documentation
    // for `ConditionalOp`.
    auto type = typeConverter->convertType(op.getType());
 
    auto hasNoWriteEffect = [](Region &region) {
      auto result = region.walk([](Operation *operation) {
        if (auto memOp = dyn_cast<MemoryEffectOpInterface>(operation))
          if (!memOp.hasEffect<MemoryEffects::Write>() &&
              !memOp.hasEffect<MemoryEffects::Free>())
            return WalkResult::advance();
 
        return WalkResult::interrupt();
      });
      return !result.wasInterrupted();
    };
 
    if (hasNoWriteEffect(op.getTrueRegion()) &&
        hasNoWriteEffect(op.getFalseRegion())) {
      Operation *trueTerm = op.getTrueRegion().front().getTerminator();
      Operation *falseTerm = op.getFalseRegion().front().getTerminator();
 
      rewriter.inlineBlockBefore(&op.getTrueRegion().front(), op);
      rewriter.inlineBlockBefore(&op.getFalseRegion().front(), op);
 
      Value convTrueVal = typeConverter->materializeTargetConversion(
          rewriter, op.getLoc(), type, trueTerm->getOperand(0));
      Value convFalseVal = typeConverter->materializeTargetConversion(
          rewriter, op.getLoc(), type, falseTerm->getOperand(0));
 
      rewriter.eraseOp(trueTerm);
      rewriter.eraseOp(falseTerm);
 
      rewriter.replaceOpWithNewOp<comb::MuxOp>(op, adaptor.getCondition(),
                                               convTrueVal, convFalseVal);
      return success();
    }
 
    auto ifOp =
        rewriter.create<scf::IfOp>(op.getLoc(), type, adaptor.getCondition());
    rewriter.inlineRegionBefore(op.getTrueRegion(), ifOp.getThenRegion(),
                                ifOp.getThenRegion().end());
    rewriter.inlineRegionBefore(op.getFalseRegion(), ifOp.getElseRegion(),
                                ifOp.getElseRegion().end());
    rewriter.replaceOp(op, ifOp);
    return success();
  }
};
 
struct YieldOpConversion : public OpConversionPattern<YieldOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(YieldOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<scf::YieldOp>(op, adaptor.getResult());
    return success();
  }
};
 
template <typename SourceOp>
struct InPlaceOpConversion : public OpConversionPattern<SourceOp> {
  using OpConversionPattern<SourceOp>::OpConversionPattern;
  using OpAdaptor = typename SourceOp::Adaptor;
 
  LogicalResult
  matchAndRewrite(SourceOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.modifyOpInPlace(op,
                             [&]() { op->setOperands(adaptor.getOperands()); });
    return success();
  }
};
 
template <typename MooreOpTy, typename VerifOpTy>
struct AssertLikeOpConversion : public OpConversionPattern<MooreOpTy> {
  using OpConversionPattern<MooreOpTy>::OpConversionPattern;
  using OpAdaptor = typename MooreOpTy::Adaptor;
 
  LogicalResult
  matchAndRewrite(MooreOpTy op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    StringAttr label =
        op.getLabel().has_value()
            ? StringAttr::get(op->getContext(), op.getLabel().value())
            : StringAttr::get(op->getContext());
    rewriter.replaceOpWithNewOp<VerifOpTy>(op, adaptor.getCond(), mlir::Value(),
                                           label);
    return success();
  }
};
 
//===----------------------------------------------------------------------===//
// Format String Conversion
//===----------------------------------------------------------------------===//
 
struct FormatLiteralOpConversion : public OpConversionPattern<FormatLiteralOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(FormatLiteralOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<sim::FormatLitOp>(op, adaptor.getLiteral());
    return success();
  }
};
 
struct FormatConcatOpConversion : public OpConversionPattern<FormatConcatOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(FormatConcatOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<sim::FormatStringConcatOp>(op,
                                                           adaptor.getInputs());
    return success();
  }
};
 
struct FormatIntOpConversion : public OpConversionPattern<FormatIntOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(FormatIntOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    // TODO: These should honor the width, alignment, and padding.
    switch (op.getFormat()) {
    case IntFormat::Decimal:
      rewriter.replaceOpWithNewOp<sim::FormatDecOp>(op, adaptor.getValue());
      return success();
    case IntFormat::Binary:
      rewriter.replaceOpWithNewOp<sim::FormatBinOp>(op, adaptor.getValue());
      return success();
    case IntFormat::HexLower:
    case IntFormat::HexUpper:
      rewriter.replaceOpWithNewOp<sim::FormatHexOp>(op, adaptor.getValue());
      return success();
    default:
      return rewriter.notifyMatchFailure(op, "unsupported int format");
    }
  }
};
 
struct DisplayBIOpConversion : public OpConversionPattern<DisplayBIOp> {
  using OpConversionPattern::OpConversionPattern;
 
  LogicalResult
  matchAndRewrite(DisplayBIOp op, OpAdaptor adaptor,
                  ConversionPatternRewriter &rewriter) const override {
    rewriter.replaceOpWithNewOp<sim::PrintFormattedProcOp>(
        op, adaptor.getMessage());
    return success();
  }
};
 
} // namespace
 
//===----------------------------------------------------------------------===//
// Conversion Infrastructure
//===----------------------------------------------------------------------===//
 
static void populateLegality(ConversionTarget &target,
                             const TypeConverter &converter) {
  target.addIllegalDialect<MooreDialect>();
  target.addLegalDialect<comb::CombDialect>();
  target.addLegalDialect<hw::HWDialect>();
  target.addLegalDialect<llhd::LLHDDialect>();
  target.addLegalDialect<mlir::BuiltinDialect>();
  target.addLegalDialect<sim::SimDialect>();
  target.addLegalDialect<verif::VerifDialect>();
 
  target.addLegalOp<debug::ScopeOp>();
 
  target.addDynamicallyLegalOp<
      cf::CondBranchOp, cf::BranchOp, scf::IfOp, scf::ForOp, scf::YieldOp,
      func::CallOp, func::ReturnOp, UnrealizedConversionCastOp, hw::OutputOp,
      hw::InstanceOp, debug::ArrayOp, debug::StructOp, debug::VariableOp>(
      [&](Operation *op) { return converter.isLegal(op); });
 
  target.addDynamicallyLegalOp<func::FuncOp>([&](func::FuncOp op) {
    return converter.isSignatureLegal(op.getFunctionType()) &&
           converter.isLegal(&op.getFunctionBody());
  });
 
  target.addDynamicallyLegalOp<hw::HWModuleOp>([&](hw::HWModuleOp op) {
    return converter.isSignatureLegal(op.getModuleType().getFuncType()) &&
           converter.isLegal(&op.getBody());
  });
}
 
static void populateTypeConversion(TypeConverter &typeConverter) {
  typeConverter.addConversion([&](IntType type) {
    return IntegerType::get(type.getContext(), type.getWidth());
  });
 
  typeConverter.addConversion([&](FormatStringType type) {
    return sim::FormatStringType::get(type.getContext());
  });
 
  typeConverter.addConversion([&](ArrayType type) -> std::optional<Type> {
    if (auto elementType = typeConverter.convertType(type.getElementType()))
      return hw::ArrayType::get(elementType, type.getSize());
    return {};
  });
 
  // FIXME: Unpacked arrays support more element types than their packed
  // variants, and as such, mapping them to hw::Array is somewhat naive. See
  // also the analogous note below concerning unpacked struct type conversion.
  typeConverter.addConversion(
      [&](UnpackedArrayType type) -> std::optional<Type> {
        if (auto elementType = typeConverter.convertType(type.getElementType()))
          return hw::ArrayType::get(elementType, type.getSize());
        return {};
      });
 
  typeConverter.addConversion([&](StructType type) -> std::optional<Type> {
    SmallVector<hw::StructType::FieldInfo> fields;
    for (auto field : type.getMembers()) {
      hw::StructType::FieldInfo info;
      info.type = typeConverter.convertType(field.type);
      if (!info.type)
        return {};
      info.name = field.name;
      fields.push_back(info);
    }
    return hw::StructType::get(type.getContext(), fields);
  });
 
  // FIXME: Mapping unpacked struct type to struct type in hw dialect may be a
  // plain solution. The packed and unpacked data structures have some
  // differences though they look similarily. The packed data structure is
  // contiguous in memory but another is opposite. The differences will affect
  // data layout and granularity of event tracking in simulation.
  typeConverter.addConversion(
      [&](UnpackedStructType type) -> std::optional<Type> {
        SmallVector<hw::StructType::FieldInfo> fields;
        for (auto field : type.getMembers()) {
          hw::StructType::FieldInfo info;
          info.type = typeConverter.convertType(field.type);
          if (!info.type)
            return {};
          info.name = field.name;
          fields.push_back(info);
        }
        return hw::StructType::get(type.getContext(), fields);
      });
 
  typeConverter.addConversion([&](RefType type) -> std::optional<Type> {
    if (auto innerType = typeConverter.convertType(type.getNestedType()))
      return hw::InOutType::get(innerType);
    return {};
  });
 
  // Valid target types.
  typeConverter.addConversion([](IntegerType type) { return type; });
  typeConverter.addConversion([](debug::ArrayType type) { return type; });
  typeConverter.addConversion([](debug::ScopeType type) { return type; });
  typeConverter.addConversion([](debug::StructType type) { return type; });
 
  typeConverter.addConversion([&](hw::InOutType type) -> std::optional<Type> {
    if (auto innerType = typeConverter.convertType(type.getElementType()))
      return hw::InOutType::get(innerType);
    return {};
  });
 
  typeConverter.addConversion([&](hw::ArrayType type) -> std::optional<Type> {
    if (auto elementType = typeConverter.convertType(type.getElementType()))
      return hw::ArrayType::get(elementType, type.getNumElements());
    return {};
  });
 
  typeConverter.addConversion([&](hw::StructType type) -> std::optional<Type> {
    SmallVector<hw::StructType::FieldInfo> fields;
    for (auto field : type.getElements()) {
      hw::StructType::FieldInfo info;
      info.type = typeConverter.convertType(field.type);
      if (!info.type)
        return {};
      info.name = field.name;
      fields.push_back(info);
    }
    return hw::StructType::get(type.getContext(), fields);
  });
 
  typeConverter.addTargetMaterialization(
      [&](mlir::OpBuilder &builder, mlir::Type resultType,
          mlir::ValueRange inputs, mlir::Location loc) -> mlir::Value {
        if (inputs.size() != 1 || !inputs[0])
          return Value();
        return builder
            .create<UnrealizedConversionCastOp>(loc, resultType, inputs[0])
            .getResult(0);
      });
 
  typeConverter.addSourceMaterialization(
      [&](mlir::OpBuilder &builder, mlir::Type resultType,
          mlir::ValueRange inputs, mlir::Location loc) -> mlir::Value {
        if (inputs.size() != 1)
          return Value();
        return builder
            .create<UnrealizedConversionCastOp>(loc, resultType, inputs[0])
            ->getResult(0);
      });
}
 
static void populateOpConversion(RewritePatternSet &patterns,
                                 TypeConverter &typeConverter) {
  auto *context = patterns.getContext();
  // clang-format off
  patterns.add<
    // Patterns of declaration operations.
    VariableOpConversion,
    NetOpConversion,
 
    // Patterns for conversion operations.
    ConversionOpConversion,
    TruncOpConversion,
    ZExtOpConversion,
    SExtOpConversion,
 
    // Patterns of miscellaneous operations.
    ConstantOpConv, ConcatOpConversion, ReplicateOpConversion,
    ExtractOpConversion, DynExtractOpConversion, DynExtractRefOpConversion,
    ReadOpConversion,
    StructExtractOpConversion, StructExtractRefOpConversion,
    ExtractRefOpConversion, StructCreateOpConversion, ConditionalOpConversion,
    YieldOpConversion, OutputOpConversion, StringConstantOpConv,
 
    // Patterns of unary operations.
    ReduceAndOpConversion, ReduceOrOpConversion, ReduceXorOpConversion,
    BoolCastOpConversion, NotOpConversion, NegOpConversion,
 
    // Patterns of binary operations.
    BinaryOpConversion<AddOp, comb::AddOp>,
    BinaryOpConversion<SubOp, comb::SubOp>,
    BinaryOpConversion<MulOp, comb::MulOp>,
    BinaryOpConversion<DivUOp, comb::DivUOp>,
    BinaryOpConversion<DivSOp, comb::DivSOp>,
    BinaryOpConversion<ModUOp, comb::ModUOp>,
    BinaryOpConversion<ModSOp, comb::ModSOp>,
    BinaryOpConversion<AndOp, comb::AndOp>,
    BinaryOpConversion<OrOp, comb::OrOp>,
    BinaryOpConversion<XorOp, comb::XorOp>,
 
    // Patterns of power operations.
    PowUOpConversion, PowSOpConversion,
 
    // Patterns of relational operations.
    ICmpOpConversion<UltOp, ICmpPredicate::ult>,
    ICmpOpConversion<SltOp, ICmpPredicate::slt>,
    ICmpOpConversion<UleOp, ICmpPredicate::ule>,
    ICmpOpConversion<SleOp, ICmpPredicate::sle>,
    ICmpOpConversion<UgtOp, ICmpPredicate::ugt>,
    ICmpOpConversion<SgtOp, ICmpPredicate::sgt>,
    ICmpOpConversion<UgeOp, ICmpPredicate::uge>,
    ICmpOpConversion<SgeOp, ICmpPredicate::sge>,
    ICmpOpConversion<EqOp, ICmpPredicate::eq>,
    ICmpOpConversion<NeOp, ICmpPredicate::ne>,
    ICmpOpConversion<CaseEqOp, ICmpPredicate::ceq>,
    ICmpOpConversion<CaseNeOp, ICmpPredicate::cne>,
    ICmpOpConversion<WildcardEqOp, ICmpPredicate::weq>,
    ICmpOpConversion<WildcardNeOp, ICmpPredicate::wne>,
    CaseXZEqOpConversion<CaseZEqOp, true>,
    CaseXZEqOpConversion<CaseXZEqOp, false>,
 
    // Patterns of structural operations.
    SVModuleOpConversion, InstanceOpConversion, ProcedureOpConversion, WaitEventOpConversion,
 
    // Patterns of shifting operations.
    ShrOpConversion, ShlOpConversion, AShrOpConversion,
 
    // Patterns of assignment operations.
    AssignOpConversion<ContinuousAssignOp, 0, 1>,
    AssignOpConversion<BlockingAssignOp, 0, 1>,
    AssignOpConversion<NonBlockingAssignOp, 1, 0>,
    AssignedVariableOpConversion,
 
    // Patterns of branch operations.
    CondBranchOpConversion, BranchOpConversion,
 
    // Patterns of other operations outside Moore dialect.
    HWInstanceOpConversion, ReturnOpConversion,
    CallOpConversion, UnrealizedConversionCastConversion,
    InPlaceOpConversion<debug::ArrayOp>,
    InPlaceOpConversion<debug::StructOp>,
    InPlaceOpConversion<debug::VariableOp>,
 
    // Patterns of assert-like operations
    AssertLikeOpConversion<AssertOp, verif::AssertOp>,
    AssertLikeOpConversion<AssumeOp, verif::AssumeOp>,
    AssertLikeOpConversion<CoverOp, verif::CoverOp>,
 
    // Format strings.
    FormatLiteralOpConversion,
    FormatConcatOpConversion,
    FormatIntOpConversion,
    DisplayBIOpConversion
  >(typeConverter, context);
  // clang-format on
 
  mlir::populateAnyFunctionOpInterfaceTypeConversionPattern(patterns,
                                                            typeConverter);
  hw::populateHWModuleLikeTypeConversionPattern(
      hw::HWModuleOp::getOperationName(), patterns, typeConverter);
}
 
//===----------------------------------------------------------------------===//
// Moore to Core Conversion Pass
//===----------------------------------------------------------------------===//
 
namespace {
struct MooreToCorePass
    : public circt::impl::ConvertMooreToCoreBase<MooreToCorePass> {
  void runOnOperation() override;
};
} // namespace
 
/// Create a Moore to core dialects conversion pass.
std::unique_ptr<OperationPass<ModuleOp>> circt::createConvertMooreToCorePass() {
  return std::make_unique<MooreToCorePass>();
}
 
/// This is the main entrypoint for the Moore to Core conversion pass.
void MooreToCorePass::runOnOperation() {
  MLIRContext &context = getContext();
  ModuleOp module = getOperation();
 
  IRRewriter rewriter(module);
  (void)mlir::eraseUnreachableBlocks(rewriter, module->getRegions());
 
  ConversionTarget target(context);
  TypeConverter typeConverter;
  RewritePatternSet patterns(&context);
  populateTypeConversion(typeConverter);
  populateLegality(target, typeConverter);
  populateOpConversion(patterns, typeConverter);
 
  if (failed(applyFullConversion(module, target, std::move(patterns))))
    signalPassFailure();
}