[CIR][Lowering] Partially lower variadic builtins

Implement lowering steps for va_start, va_end, and va_copy. The va_arg was not implemented because it requires ABI-specific lowering. ghstack-source-id: 1ab2923027143aa28bb7361b884a5c8ee04cfbc9 Pull Request resolved: #95
llvm · Dec 20, 2023 · fa1511f · fa1511f
1 parent 5a977c8
commit fa1511f
Show file tree

Hide file tree

Showing 2 changed files with 113 additions and 1 deletion.
diff --git a/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp b/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
@@ -49,6 +49,7 @@
 #include "llvm/ADT/Sequence.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/Support/Casting.h"
+#include "llvm/Support/ErrorHandling.h"
 #include <optional>
 
 using namespace cir;
@@ -544,6 +545,66 @@ class CIRConstantLowering
   }
 };
 
+class CIRVAStartLowering
+    : public mlir::OpConversionPattern<mlir::cir::VAStartOp> {
+public:
+  using OpConversionPattern<mlir::cir::VAStartOp>::OpConversionPattern;
+
+  mlir::LogicalResult
+  matchAndRewrite(mlir::cir::VAStartOp op, OpAdaptor adaptor,
+                  mlir::ConversionPatternRewriter &rewriter) const override {
+    auto i8PtrTy = mlir::LLVM::LLVMPointerType::get(getContext());
+    auto vaList = rewriter.create<mlir::LLVM::BitcastOp>(
+        op.getLoc(), i8PtrTy, adaptor.getOperands().front());
+    rewriter.replaceOpWithNewOp<mlir::LLVM::VaStartOp>(op, vaList);
+    return mlir::success();
+  }
+};
+
+class CIRVAEndLowering : public mlir::OpConversionPattern<mlir::cir::VAEndOp> {
+public:
+  using OpConversionPattern<mlir::cir::VAEndOp>::OpConversionPattern;
+
+  mlir::LogicalResult
+  matchAndRewrite(mlir::cir::VAEndOp op, OpAdaptor adaptor,
+                  mlir::ConversionPatternRewriter &rewriter) const override {
+    auto i8PtrTy = mlir::LLVM::LLVMPointerType::get(getContext());
+    auto vaList = rewriter.create<mlir::LLVM::BitcastOp>(
+        op.getLoc(), i8PtrTy, adaptor.getOperands().front());
+    rewriter.replaceOpWithNewOp<mlir::LLVM::VaEndOp>(op, vaList);
+    return mlir::success();
+  }
+};
+
+class CIRVACopyLowering
+    : public mlir::OpConversionPattern<mlir::cir::VACopyOp> {
+public:
+  using OpConversionPattern<mlir::cir::VACopyOp>::OpConversionPattern;
+
+  mlir::LogicalResult
+  matchAndRewrite(mlir::cir::VACopyOp op, OpAdaptor adaptor,
+                  mlir::ConversionPatternRewriter &rewriter) const override {
+    auto i8PtrTy = mlir::LLVM::LLVMPointerType::get(getContext());
+    auto dstList = rewriter.create<mlir::LLVM::BitcastOp>(
+        op.getLoc(), i8PtrTy, adaptor.getOperands().front());
+    auto srcList = rewriter.create<mlir::LLVM::BitcastOp>(
+        op.getLoc(), i8PtrTy, adaptor.getOperands().back());
+    rewriter.replaceOpWithNewOp<mlir::LLVM::VaCopyOp>(op, dstList, srcList);
+    return mlir::success();
+  }
+};
+
+class CIRVAArgLowering : public mlir::OpConversionPattern<mlir::cir::VAArgOp> {
+public:
+  using OpConversionPattern<mlir::cir::VAArgOp>::OpConversionPattern;
+
+  mlir::LogicalResult
+  matchAndRewrite(mlir::cir::VAArgOp op, OpAdaptor adaptor,
+                  mlir::ConversionPatternRewriter &rewriter) const override {
+    return op.emitError("cir.vaarg lowering is NYI");
+  }
+};
+
 class CIRFuncLowering : public mlir::OpConversionPattern<mlir::cir::FuncOp> {
 public:
   using OpConversionPattern<mlir::cir::FuncOp>::OpConversionPattern;
@@ -997,7 +1058,8 @@ void populateCIRToLLVMConversionPatterns(mlir::RewritePatternSet &patterns,
                CIRBinOpLowering, CIRLoadLowering, CIRConstantLowering,
                CIRStoreLowering, CIRAllocaLowering, CIRFuncLowering,
                CIRScopeOpLowering, CIRCastOpLowering, CIRIfLowering,
-               CIRGlobalOpLowering, CIRGetGlobalOpLowering>(
+               CIRGlobalOpLowering, CIRGetGlobalOpLowering, CIRVAStartLowering,
+               CIRVAEndLowering, CIRVACopyLowering, CIRVAArgLowering>(
       converter, patterns.getContext());
 }
 
@@ -1018,6 +1080,16 @@ void prepareTypeConverter(mlir::LLVMTypeConverter &converter) {
     // LLVM doesn't work with signed types, so we drop the CIR signs here.
     return mlir::IntegerType::get(type.getContext(), type.getWidth());
   });
+  converter.addConversion([&](mlir::cir::StructType type) -> mlir::Type {
+    llvm::SmallVector<mlir::Type> llvmMembers;
+    for (auto ty : type.getMembers())
+      llvmMembers.push_back(converter.convertType(ty));
+    auto llvmStruct = mlir::LLVM::LLVMStructType::getIdentified(
+        type.getContext(), type.getTypeName());
+    if (llvmStruct.setBody(llvmMembers, /*isPacked=*/type.getPacked()).failed())
+      llvm_unreachable("Failed to set body of struct");
+    return llvmStruct;
+  });
 }
 } // namespace
 

diff --git a/clang/test/CIR/Lowering/variadics.cir b/clang/test/CIR/Lowering/variadics.cir
@@ -0,0 +1,40 @@
+// RUN: cir-tool %s -cir-to-llvm -o %t.cir
+// RUN: FileCheck --input-file=%t.cir %s -check-prefix=MLIR
+
+!s32i = !cir.int<s, 32>
+!u32i = !cir.int<u, 32>
+!u8i = !cir.int<u, 8>
+
+!ty_22struct2E__va_list_tag22 = !cir.struct<"struct.__va_list_tag", !u32i, !u32i, !cir.ptr<!u8i>, !cir.ptr<!u8i>, #cir.recdecl.ast>
+
+module {
+  cir.func @average(%arg0: !s32i, ...) -> !s32i {
+    %0 = cir.alloca !s32i, cir.ptr <!s32i>, ["count", init] {alignment = 4 : i64}
+    %1 = cir.alloca !s32i, cir.ptr <!s32i>, ["__retval"] {alignment = 4 : i64}
+    %2 = cir.alloca !cir.array<!ty_22struct2E__va_list_tag22 x 1>, cir.ptr <!cir.array<!ty_22struct2E__va_list_tag22 x 1>>, ["args"] {alignment = 16 : i64}
+    %3 = cir.alloca !cir.array<!ty_22struct2E__va_list_tag22 x 1>, cir.ptr <!cir.array<!ty_22struct2E__va_list_tag22 x 1>>, ["args_copy"] {alignment = 16 : i64}
+    cir.store %arg0, %0 : !s32i, cir.ptr <!s32i>
+    %4 = cir.cast(array_to_ptrdecay, %2 : !cir.ptr<!cir.array<!ty_22struct2E__va_list_tag22 x 1>>), !cir.ptr<!ty_22struct2E__va_list_tag22>
+    cir.va.start %4 : !cir.ptr<!ty_22struct2E__va_list_tag22>
+    //      MLIR: %{{[0-9]+}} = llvm.getelementptr %{{[0-9]+}}[0] : (!llvm.ptr) -> !llvm.ptr
+    // MLIR-NEXT: %{{[0-9]+}} = llvm.bitcast %{{[0-9]+}} : !llvm.ptr to !llvm.ptr
+    // MLIR-NEXT: llvm.intr.vastart %{{[0-9]+}} : !llvm.ptr
+    %5 = cir.cast(array_to_ptrdecay, %3 : !cir.ptr<!cir.array<!ty_22struct2E__va_list_tag22 x 1>>), !cir.ptr<!ty_22struct2E__va_list_tag22>
+    %6 = cir.cast(array_to_ptrdecay, %2 : !cir.ptr<!cir.array<!ty_22struct2E__va_list_tag22 x 1>>), !cir.ptr<!ty_22struct2E__va_list_tag22>
+    cir.va.copy %6 to %5 : !cir.ptr<!ty_22struct2E__va_list_tag22>, !cir.ptr<!ty_22struct2E__va_list_tag22>
+    //      MLIR: %{{[0-9]+}} = llvm.getelementptr %{{[0-9]+}}[0] : (!llvm.ptr) -> !llvm.ptr
+    // MLIR-NEXT: %{{[0-9]+}} = llvm.getelementptr %{{[0-9]+}}[0] : (!llvm.ptr) -> !llvm.ptr
+    // MLIR-NEXT: %{{[0-9]+}} = llvm.bitcast %{{[0-9]+}} : !llvm.ptr to !llvm.ptr
+    // MLIR-NEXT: %{{[0-9]+}} = llvm.bitcast %{{[0-9]+}} : !llvm.ptr to !llvm.ptr
+    // MLIR-NEXT: llvm.intr.vacopy %13 to %{{[0-9]+}} : !llvm.ptr, !llvm.ptr
+    %7 = cir.cast(array_to_ptrdecay, %2 : !cir.ptr<!cir.array<!ty_22struct2E__va_list_tag22 x 1>>), !cir.ptr<!ty_22struct2E__va_list_tag22>
+    cir.va.end %7 : !cir.ptr<!ty_22struct2E__va_list_tag22>
+    //      MLIR: %{{[0-9]+}} = llvm.getelementptr %{{[0-9]+}}[0] : (!llvm.ptr) -> !llvm.ptr
+    // MLIR-NEXT: %{{[0-9]+}} = llvm.bitcast %{{[0-9]+}} : !llvm.ptr to !llvm.ptr
+    // MLIR-NEXT: llvm.intr.vaend %{{[0-9]+}} : !llvm.ptr
+    %8 = cir.const(#cir.int<0> : !s32i) : !s32i
+    cir.store %8, %1 : !s32i, cir.ptr <!s32i>
+    %9 = cir.load %1 : cir.ptr <!s32i>, !s32i
+    cir.return %9 : !s32i
+  }
+}