hello-mlir/lib/LowerToLLVMPass.cpp

//
// Created by ricardo on 07/06/25.
//

#include <mlir/Dialect/Arith/IR/Arith.h>
#include <mlir/Dialect/LLVMIR/LLVMTypes.h>
#include <mlir/Dialect/LLVMIR/LLVMDialect.h>
#include <mlir/Dialect/MemRef/IR/MemRef.h>
#include <mlir/Dialect/SCF/IR/SCF.h>
#include <mlir/Pass/Pass.h>
#include <mlir/Transforms/DialectConversion.h>
#include <mlir/Conversion/LLVMCommon/ConversionTarget.h>
#include <mlir/Conversion/LLVMCommon/TypeConverter.h>
#include <mlir/Conversion/AffineToStandard/AffineToStandard.h>
#include <mlir/Conversion/ArithToLLVM/ArithToLLVM.h>
#include <mlir/Conversion/ControlFlowToLLVM/ControlFlowToLLVM.h>
#include <mlir/Conversion/FuncToLLVM/ConvertFuncToLLVM.h>
#include <mlir/Conversion/FuncToLLVM/ConvertFuncToLLVMPass.h>
#include <mlir/Conversion/MemRefToLLVM/MemRefToLLVM.h>
#include <mlir/Conversion/SCFToControlFlow/SCFToControlFlow.h>

#include "Dialect.h"
#include "Passes.h"

namespace
{
    class PrintOpLoweringPattern : public mlir::ConversionPattern
    {
    public:
        explicit PrintOpLoweringPattern(mlir::MLIRContext* context) : ConversionPattern(
            mlir::hello::PrintOp::getOperationName(), 1, context)
        {
        }

        mlir::LogicalResult matchAndRewrite(mlir::Operation* op, llvm::ArrayRef<mlir::Value> operands,
                                            mlir::ConversionPatternRewriter& rewriter) const final
        {
            auto* context = rewriter.getContext();
            auto memRefType = llvm::cast<mlir::MemRefType>(op->getOperandTypes().front());
            auto memRefShape = memRefType.getShape();
            auto location = op->getLoc();

            auto parentModule = op->getParentOfType<mlir::ModuleOp>();

            // Get the `printf` function declaration.
            auto printfRef = getOrInsertPrintf(rewriter, parentModule);

            // Create the format string in C format.
            mlir::Value formatSpecifierString = getOrCreateGlobalString(location, rewriter, "format_specifier",
                                                                        "%f \0",
                                                                        parentModule);
            // Create the LF format string in C format.
            mlir::Value newLineString = getOrCreateGlobalString(location, rewriter, "new_line", "\n\0", parentModule);

            // Create a loop to print the value in the tensor.
            llvm::SmallVector<mlir::Value, 4> loopIterators;
            for (const auto i : llvm::seq<size_t>(0, memRefShape.size()))
            {
                auto lowerBound = rewriter.create<mlir::arith::ConstantIndexOp>(location, 0);
                auto upperBound = rewriter.create<mlir::arith::ConstantIndexOp>(location, memRefShape[i]);
                auto step = rewriter.create<mlir::arith::ConstantIndexOp>(location, 1);

                auto loop = rewriter.create<mlir::scf::ForOp>(location, lowerBound, upperBound, step);
                // FIXME: Remove the nested operation in loop, Why?
                for (mlir::Operation& nestedOperation : *loop.getBody())
                {
                    rewriter.eraseOp(&nestedOperation);
                }

                loopIterators.push_back(loop.getInductionVar());

                // Place the new line output and terminator in the end of loop.
                rewriter.setInsertionPointToEnd(loop.getBody());

                if (i != memRefShape.size() - 1)
                {
                    // Add change line when in a row.
                    rewriter.create<mlir::LLVM::CallOp>(location, getPrintfFunctionType(context), printfRef,
                                                        newLineString);
                }
                rewriter.create<mlir::scf::YieldOp>(location);

                // Then place the rewriter at the start of the loop, so when finishing once,
                // ths rewriter will at the start of newly created loop.
                rewriter.setInsertionPointToStart(loop.getBody());
            }

            auto printOperation = llvm::cast<mlir::hello::PrintOp>(op);
            auto loadedElement = rewriter.create<mlir::memref::LoadOp>(location, printOperation.getInput(),
                                                                       loopIterators);
            rewriter.create<mlir::LLVM::CallOp>(location, getPrintfFunctionType(context), printfRef,
                                                llvm::ArrayRef<mlir::Value>({formatSpecifierString, loadedElement}));

            // At last remove the print operation.
            rewriter.eraseOp(printOperation);
            return mlir::success();
        }

    private:
        static mlir::LLVM::LLVMFunctionType getPrintfFunctionType(mlir::MLIRContext* context)
        {
            const auto llvmInteger32Type = mlir::IntegerType::get(context, 32);
            const auto llvmPointerType = mlir::LLVM::LLVMPointerType::get(context);
            // The `printf` is `int printf(char *, ...)`.
            const auto llvmFunctionType = mlir::LLVM::LLVMFunctionType::get(llvmInteger32Type, llvmPointerType, true);

            return llvmFunctionType;
        }

        static mlir::FlatSymbolRefAttr getOrInsertPrintf(mlir::PatternRewriter& rewriter, mlir::ModuleOp& module)
        {
            auto* context = module.getContext();
            if (module.lookupSymbol<mlir::LLVM::LLVMFuncOp>("printf"))
            {
                return mlir::SymbolRefAttr::get(context, "printf");
            }

            // Insert the `printf` declarations in the body of the parent module.
            mlir::PatternRewriter::InsertionGuard guard(rewriter);
            rewriter.setInsertionPointToStart(module.getBody());

            rewriter.create<mlir::LLVM::LLVMFuncOp>(module.getLoc(), "printf", getPrintfFunctionType(context));
            return mlir::SymbolRefAttr::get(context, "printf");
        }

        /// Create or get a global string used for print.
        static mlir::Value getOrCreateGlobalString(mlir::Location& loc, mlir::OpBuilder& builder, llvm::StringRef name,
                                                   llvm::StringRef value, mlir::ModuleOp& module)
        {
            auto global = module.lookupSymbol<mlir::LLVM::GlobalOp>(name);
            if (!global)
            {
                // Failed to find the global value, create it.
                mlir::OpBuilder::InsertionGuard guard(builder);
                builder.setInsertionPointToStart(module.getBody());

                auto stringType = mlir::LLVM::LLVMArrayType::get(
                    mlir::IntegerType::get(builder.getContext(), 8), value.size());
                global = builder.create<mlir::LLVM::GlobalOp>(loc, stringType, true, mlir::LLVM::Linkage::Internal,
                                                              name, builder.getStringAttr(value), 0);
            }

            // Get the pointer to the first character in the global string.
            mlir::Value globalPointer = builder.create<mlir::LLVM::AddressOfOp>(loc, global);
            mlir::Value constantZero = builder.create<mlir::LLVM::ConstantOp>(
                loc, builder.getI64Type(), builder.getIndexAttr(0));

            return builder.create<mlir::LLVM::GEPOp>(loc, mlir::LLVM::LLVMPointerType::get(builder.getContext()),
                                                     global.getType(),
                                                     globalPointer, llvm::ArrayRef({
                                                         constantZero, constantZero
                                                     }));
        }
    };

    struct HelloToLLVMLoweringPass : mlir::PassWrapper<HelloToLLVMLoweringPass, mlir::OperationPass<mlir::ModuleOp>>
    {
        MLIR_DEFINE_EXPLICIT_INTERNAL_INLINE_TYPE_ID(HelloToLLVMLoweringPass)

        llvm::StringRef getArgument() const final
        {
            return "hello-to-llvm";
        }

        void getDependentDialects(mlir::DialectRegistry& registry) const final
        {
            registry.insert<mlir::LLVM::LLVMDialect, mlir::scf::SCFDialect>();
        }

        void runOnOperation() final;
    };
}

void HelloToLLVMLoweringPass::runOnOperation()
{
    mlir::LLVMConversionTarget target(getContext());
    target.addLegalDialect<mlir::LLVM::LLVMDialect>();
    target.addLegalOp<mlir::ModuleOp>();

    mlir::LLVMTypeConverter typeConverter(&getContext());

    mlir::RewritePatternSet patterns(&getContext());

    // The lower path is a little bit complicated.
    // Convert affine dialect to standard dialect.
    mlir::populateAffineToStdConversionPatterns(patterns);
    // Convert scf dialect to cf dialect.
    mlir::populateSCFToControlFlowConversionPatterns(patterns);
    // Convert arith to llvm dialect.
    mlir::arith::populateArithToLLVMConversionPatterns(typeConverter, patterns);
    mlir::populateFinalizeMemRefToLLVMConversionPatterns(typeConverter, patterns);
    mlir::cf::populateControlFlowToLLVMConversionPatterns(typeConverter, patterns);
    mlir::populateFuncToLLVMConversionPatterns(typeConverter, patterns);

    patterns.add<PrintOpLoweringPattern>(&getContext());

    // As we need to convert all operations to LLVM dialect, so
    // we perform a full convertion, which will permit the legal opertions in result.
    if (const auto module = getOperation();
        mlir::failed(mlir::applyFullConversion(module, target, std::move(patterns))))
    {
        signalPassFailure();
    }
}

std::unique_ptr<mlir::Pass> mlir::hello::createLowerToLLVMPass()
{
    return std::make_unique<HelloToLLVMLoweringPass>();
}