[CIR][CIRGen] Support __builtin_memset_inline

clang/include/clang/CIR/Dialect/IR/CIROps.td

@@ -4147,6 +4147,37 @@ def MemSetOp : CIR_Op<"libc.memset"> {
   let hasVerifier = 0;
 }
 
+//===----------------------------------------------------------------------===//
+// MemSetInlineOp
+//===----------------------------------------------------------------------===//
+
+def MemSetInlineOp : CIR_Op<"memset_inline"> {
+  let arguments = (ins Arg<VoidPtr, "", [MemWrite]>:$dst,
+                       SInt32:$val,
+                       I64Attr:$len);
+  let summary = "Fill a block of memory with constant length without calling"
+                "any external function";
+  let description = [{
+    Given the CIR pointer, `dst`, `cir.memset_inline` will set the first `len`
+    bytes of the memory pointed by `dst` to the specified `val`.
+
+    The `len` argument must be a constant integer argument specifying the number
+    of bytes to fill.
+
+    Examples:
+
+    ```mlir
+      // Set 2 bytes from a struct to 0
+      cir.memset_inline 2 bytes from %struct set to %zero : !cir.ptr<!void>, !s32i
+    ```
+  }];
+
+  let assemblyFormat = [{
+    $len `bytes` `from` $dst `set` `to` $val attr-dict
+    `:` qualified(type($dst)) `,` type($val)
+  }];
+  let hasVerifier = 0;
+}
 //===----------------------------------------------------------------------===//
 // MemChrOp
 //===----------------------------------------------------------------------===//

clang/lib/CIR/CodeGen/CIRGenBuilder.h

@@ -623,6 +623,13 @@ class CIRGenBuilderTy : public cir::CIRBaseBuilderTy {
     return create<cir::MemSetOp>(loc, dst, val, len);
   }
 
+  cir::MemSetInlineOp createMemSetInline(mlir::Location loc, mlir::Value dst,
+                                         mlir::Value val,
+                                         mlir::IntegerAttr len) {
+    val = createIntCast(val, cir::IntType::get(getContext(), 32, true));
+    return create<cir::MemSetInlineOp>(loc, dst, val, len);
+  }
+
   mlir::Value createNeg(mlir::Value value) {
 
     if (auto intTy = mlir::dyn_cast<cir::IntType>(value.getType())) {

clang/lib/CIR/CodeGen/CIRGenBuiltin.cpp

@@ -1554,8 +1554,20 @@ RValue CIRGenFunction::emitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
     return RValue::get(Dest.getPointer());
   }
 
-  case Builtin::BI__builtin_memset_inline:
-    llvm_unreachable("BI__builtin_memset_inline NYI");
+  case Builtin::BI__builtin_memset_inline: {
+    Address Dest = emitPointerWithAlignment(E->getArg(0));
+    mlir::Value ByteVal = emitScalarExpr(E->getArg(1));
+    uint64_t size =
+        E->getArg(2)->EvaluateKnownConstInt(getContext()).getZExtValue();
+    emitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
+                        E->getArg(0)->getExprLoc(), FD, 0);
+    builder.createMemSetInline(
+        getLoc(E->getSourceRange()), Dest.getPointer(), ByteVal,
+        mlir::IntegerAttr::get(mlir::IntegerType::get(builder.getContext(), 64),
+                               size));
+    // __builtin_memset_inline has no return value
+    return RValue::get(nullptr);
+  }
   case Builtin::BI__builtin___memset_chk: {
     // fold __builtin_memset_chk(x, y, cst1, cst2) to memset iff cst1<=cst2.
     llvm::APSInt size;

clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp

@@ -777,6 +777,18 @@ mlir::LogicalResult CIRToLLVMMemSetOpLowering::matchAndRewrite(
   return mlir::success();
 }
 
+mlir::LogicalResult CIRToLLVMMemSetInlineOpLowering::matchAndRewrite(
+    cir::MemSetInlineOp op, OpAdaptor adaptor,
+    mlir::ConversionPatternRewriter &rewriter) const {
+  auto converted = rewriter.create<mlir::LLVM::TruncOp>(
+      op.getLoc(), mlir::IntegerType::get(op.getContext(), 8),
+      adaptor.getVal());
+  rewriter.replaceOpWithNewOp<mlir::LLVM::MemsetInlineOp>(
+      op, adaptor.getDst(), converted, adaptor.getLenAttr(),
+      /*isVolatile=*/false);
+  return mlir::success();
+}
+
 static mlir::Value getLLVMIntCast(mlir::ConversionPatternRewriter &rewriter,
                                   mlir::Value llvmSrc, mlir::Type llvmDstIntTy,
                                   bool isUnsigned, uint64_t cirSrcWidth,
@@ -1851,8 +1863,8 @@ mlir::LogicalResult CIRToLLVMVAArgOpLowering::matchAndRewrite(
   return op.emitError("cir.vaarg lowering is NYI");
 }
 
-  /// Returns the name used for the linkage attribute. This *must* correspond
-  /// to the name of the attribute in ODS.
+/// Returns the name used for the linkage attribute. This *must* correspond
+/// to the name of the attribute in ODS.
 StringRef CIRToLLVMFuncOpLowering::getLinkageAttrNameString() {
   return "linkage";
 }
@@ -1886,8 +1898,8 @@ void CIRToLLVMFuncOpLowering::lowerFuncAttributes(
   }
 }
 
-  /// When do module translation, we can only translate LLVM-compatible types.
-  /// Here we lower possible OpenCLKernelMetadataAttr to use the converted type.
+/// When do module translation, we can only translate LLVM-compatible types.
+/// Here we lower possible OpenCLKernelMetadataAttr to use the converted type.
 void CIRToLLVMFuncOpLowering::lowerFuncOpenCLKernelMetadata(
     mlir::NamedAttribute &extraAttrsEntry) const {
   const auto attrKey = cir::OpenCLKernelMetadataAttr::getMnemonic();
@@ -2100,8 +2112,8 @@ mlir::LogicalResult CIRToLLVMSwitchFlatOpLowering::matchAndRewrite(
   return mlir::success();
 }
 
-  /// Replace CIR global with a region initialized LLVM global and update
-  /// insertion point to the end of the initializer block.
+/// Replace CIR global with a region initialized LLVM global and update
+/// insertion point to the end of the initializer block.
 void CIRToLLVMGlobalOpLowering::setupRegionInitializedLLVMGlobalOp(
     cir::GlobalOp op, mlir::ConversionPatternRewriter &rewriter) const {
   const auto llvmType = getTypeConverter()->convertType(op.getSymType());
@@ -3890,8 +3902,9 @@ void populateCIRToLLVMConversionPatterns(
       CIRToLLVMBaseClassAddrOpLowering, CIRToLLVMDerivedClassAddrOpLowering,
       CIRToLLVMVTTAddrPointOpLowering, CIRToLLVMIsFPClassOpLowering,
       CIRToLLVMAbsOpLowering, CIRToLLVMMemMoveOpLowering,
-      CIRToLLVMMemSetOpLowering, CIRToLLVMMemCpyInlineOpLowering,
-      CIRToLLVMSignBitOpLowering, CIRToLLVMPtrMaskOpLowering
+      CIRToLLVMMemSetOpLowering, CIRToLLVMMemSetInlineOpLowering,
+      CIRToLLVMMemCpyInlineOpLowering, CIRToLLVMSignBitOpLowering,
+      CIRToLLVMPtrMaskOpLowering
 #define GET_BUILTIN_LOWERING_LIST
 #include "clang/CIR/Dialect/IR/CIRBuiltinsLowering.inc"
 #undef GET_BUILTIN_LOWERING_LIST

clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.h

@@ -125,6 +125,16 @@ class CIRToLLVMMemSetOpLowering
                   mlir::ConversionPatternRewriter &) const override;
 };
 
+class CIRToLLVMMemSetInlineOpLowering
+    : public mlir::OpConversionPattern<cir::MemSetInlineOp> {
+public:
+  using mlir::OpConversionPattern<cir::MemSetInlineOp>::OpConversionPattern;
+
+  mlir::LogicalResult
+  matchAndRewrite(cir::MemSetInlineOp op, OpAdaptor adaptor,
+                  mlir::ConversionPatternRewriter &rewriter) const override;
+};
+
 class CIRToLLVMPtrStrideOpLowering
     : public mlir::OpConversionPattern<cir::PtrStrideOp> {
 public:

clang/test/CIR/CodeGen/builtins-memory.c

@@ -210,3 +210,23 @@ void test_memcpy_inline_aligned_buffers(unsigned long long *dst, const unsigned
   // COM: LLVM: call void @llvm.memcpy.inline.p0.p0.i64(ptr align 8 {{%.*}}, ptr align 8 {{%.*}}, i64 4, i1 false)
   __builtin_memcpy_inline(dst, src, 4);
 }
+
+void test_memset_inline(void *dst, int val) {
+
+  // CIR-LABEL: test_memset_inline
+  // CIR: cir.memset_inline 0 bytes from {{%.*}} set to {{%.*}} : !cir.ptr<!void>, !s32i
+
+  // LLVM-LABEL: test_memset_inline
+  // LLVM: call void @llvm.memset.inline.p0.i64(ptr {{%.*}}, i8 {{%.*}}, i64 0, i1 false)
+  __builtin_memset_inline(dst, val, 0);
+
+  // CIR: cir.memset_inline 1 bytes from {{%.*}} set to {{%.*}} : !cir.ptr<!void>, !s32i
+
+  // LLVM: call void @llvm.memset.inline.p0.i64(ptr {{%.*}}, i8 {{%.*}}, i64 1, i1 false)
+  __builtin_memset_inline(dst, val, 1);
+
+  // CIR: cir.memset_inline 4 bytes from {{%.*}} set to {{%.*}} : !cir.ptr<!void>, !s32i
+
+  // LLVM: call void @llvm.memset.inline.p0.i64(ptr {{%.*}}, i8 {{%.*}}, i64 4, i1 false)
+  __builtin_memset_inline(dst, val, 4);
+}