[CIR] Upstream X86 builtin _mm_prefetch and _mm_clflush

HendrikHuebner · HendrikHuebner · commit 539eebe34fb2 · 2025-11-10T23:04:51.000+01:00
diff --git a/clang/include/clang/CIR/Dialect/IR/CIROps.td b/clang/include/clang/CIR/Dialect/IR/CIROps.td
@@ -413,6 +413,18 @@ def CIR_ConstantOp : CIR_Op<"const", [
 
     template <typename T>
     T getValueAttr() { return mlir::dyn_cast<T>(getValue()); }
+
+    llvm::APInt getIntValue() {
+      if (const auto intAttr = getValueAttr<cir::IntAttr>())
+        return intAttr.getValue();
+      llvm_unreachable("Expected an IntAttr in ConstantOp");
+    }
+
+    bool getBoolValue() {
+      if (const auto boolAttr = getValueAttr<cir::BoolAttr>())
+        return boolAttr.getValue();
+      llvm_unreachable("Expected a BoolAttr in ConstantOp");
+    }
   }];
 
   let hasFolder = 1;
@@ -2579,6 +2591,39 @@ def CIR_FuncOp : CIR_Op<"func", [
   }];
 }
 
+//===----------------------------------------------------------------------===//
+// LLVMIntrinsicCallOp
+//===----------------------------------------------------------------------===//
+
+def CIR_LLVMIntrinsicCallOp : CIR_Op<"llvm.intrinsic"> {
+  let summary = "Call to llvm intrinsic functions that is not defined in CIR";
+  let description = [{
+    `cir.llvm.intrinsic` operation represents a call-like expression which has
+    return type and arguments that maps directly to a llvm intrinsic.
+    It only records intrinsic `intrinsic_name`.
+  }];
+
+  let results = (outs Optional<CIR_AnyType>:$result);
+  let arguments = (ins
+                   StrAttr:$intrinsic_name, Variadic<CIR_AnyType>:$arg_ops);
+
+  let skipDefaultBuilders = 1;
+
+  let assemblyFormat = [{
+    $intrinsic_name $arg_ops `:` functional-type($arg_ops, $result) attr-dict
+  }];
+
+  let builders = [
+    OpBuilder<(ins "mlir::StringAttr":$intrinsic_name, "mlir::Type":$resType,
+              CArg<"mlir::ValueRange", "{}">:$operands), [{
+      $_state.addAttribute("intrinsic_name", intrinsic_name);
+      $_state.addOperands(operands);
+      if (resType)
+        $_state.addTypes(resType);
+    }]>,
+  ];
+}
+
 //===----------------------------------------------------------------------===//
 // CallOp
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/CIR/CodeGen/CIRGenBuiltinX86.cpp b/clang/lib/CIR/CodeGen/CIRGenBuiltinX86.cpp
@@ -21,6 +21,49 @@
 using namespace clang;
 using namespace clang::CIRGen;
 
+/// Get integer from a mlir::Value that is an int constant or a constant op.
+static int64_t getIntValueFromConstOp(mlir::Value val) {
+  return val.getDefiningOp<cir::ConstantOp>().getIntValue().getSExtValue();
+}
+
+static mlir::Value emitClFlush(CIRGenFunction& cgf,
+                               const CallExpr* e,
+                               mlir::Value& op) {
+    mlir::Type voidTy = cir::VoidType::get(&cgf.getMLIRContext());
+    mlir::Location location = cgf.getLoc(e->getExprLoc());
+    return cgf.getBuilder()
+        .create<cir::LLVMIntrinsicCallOp>(
+            location, cgf.getBuilder().getStringAttr("x86.sse2.clflush"),
+            voidTy, op)
+        .getResult();
+}
+
+static mlir::Value emitPrefetch(CIRGenFunction& cgf,
+                                const CallExpr* e,
+                                mlir::Value& addr,
+                                int64_t hint) {
+  CIRGenBuilderTy& builder = cgf.getBuilder();
+  mlir::Type voidTy = cir::VoidType::get(&cgf.getMLIRContext());
+  mlir::Type sInt32Ty = cir::IntType::get(&cgf.getMLIRContext(), 32, true);
+  mlir::Value address = builder.createPtrBitcast(addr, voidTy);
+  mlir::Location location = cgf.getLoc(e->getExprLoc());
+  mlir::Value rw =
+      cir::ConstantOp::create(builder, location,
+                              cir::IntAttr::get(sInt32Ty, (hint >> 2) & 0x1));
+  mlir::Value locality =
+      cir::ConstantOp::create(builder, location,
+                              cir::IntAttr::get(sInt32Ty, hint & 0x3));
+  mlir::Value data = cir::ConstantOp::create(builder, location,
+                                             cir::IntAttr::get(sInt32Ty, 1));
+
+  return cir::LLVMIntrinsicCallOp::create(
+             builder, location,
+             builder.getStringAttr("prefetch"), voidTy,
+             mlir::ValueRange{address, rw, locality, data})
+      .getResult();
+}
+
+
 mlir::Value CIRGenFunction::emitX86BuiltinExpr(unsigned builtinID,
                                                const CallExpr *e) {
   if (builtinID == Builtin::BI__builtin_cpu_is) {
@@ -43,11 +86,28 @@ mlir::Value CIRGenFunction::emitX86BuiltinExpr(unsigned builtinID,
   // Find out if any arguments are required to be integer constant expressions.
   assert(!cir::MissingFeatures::handleBuiltinICEArguments());
 
+  // The operands of the builtin call
+  llvm::SmallVector<mlir::Value, 4> ops;
+
+  // `ICEArguments` is a bitmap indicating whether the argument at the i-th bit
+  // is required to be a constant integer expression.
+  unsigned ICEArguments = 0;
+  ASTContext::GetBuiltinTypeError error;
+  getContext().GetBuiltinType(builtinID, error, &ICEArguments);
+  assert(error == ASTContext::GE_None && "Error while getting builtin type.");
+
+  const unsigned numArgs = e->getNumArgs();
+  for (unsigned i = 0; i != numArgs; i++) {
+    ops.push_back(emitScalarOrConstFoldImmArg(ICEArguments, i, e));
+  }
+
   switch (builtinID) {
   default:
     return {};
   case X86::BI_mm_prefetch:
+    return emitPrefetch(*this, e, ops[0], getIntValueFromConstOp(ops[1]));
   case X86::BI_mm_clflush:
+    return emitClFlush(*this, e, ops[0]);
   case X86::BI_mm_lfence:
   case X86::BI_mm_pause:
   case X86::BI_mm_mfence:
diff --git a/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp b/clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp
@@ -1430,6 +1430,28 @@ mlir::Value CIRGenFunction::emitPromotedScalarExpr(const Expr *e,
   return ScalarExprEmitter(*this, builder).Visit(const_cast<Expr *>(e));
 }
 
+mlir::Value CIRGenFunction::emitScalarOrConstFoldImmArg(unsigned ICEArguments,
+                                                        unsigned index,
+                                                        const CallExpr *e) {
+  mlir::Value arg{};
+
+  // The bit at the specified index indicates whether the argument is required
+  // to be a constant integer expression.
+  bool isArgRequiredToBeConstant = (ICEArguments & (1 << index));
+
+  if (!isArgRequiredToBeConstant) {
+    arg = emitScalarExpr(e->getArg(index));
+  } else {
+    // If this is required to be a constant, constant fold it so that we
+    // know that the generated intrinsic gets a ConstantInt.
+    std::optional<llvm::APSInt> result =
+        e->getArg(index)->getIntegerConstantExpr(getContext());
+    assert(result && "Expected argument to be a constant");
+    arg = builder.getConstInt(getLoc(e->getSourceRange()), *result);
+  }
+  return arg;
+}
+
 [[maybe_unused]] static bool mustVisitNullValue(const Expr *e) {
   // If a null pointer expression's type is the C++0x nullptr_t and
   // the expression is not a simple literal, it must be evaluated
diff --git a/clang/lib/CIR/CodeGen/CIRGenFunction.h b/clang/lib/CIR/CodeGen/CIRGenFunction.h
@@ -1529,6 +1529,10 @@ class CIRGenFunction : public CIRGenTypeCache {
   mlir::Value emitScalarExpr(const clang::Expr *e,
                              bool ignoreResultAssign = false);
 
+  mlir::Value emitScalarOrConstFoldImmArg(unsigned ICEArguments,
+                                          unsigned index,
+                                          const CallExpr *e);
+
   mlir::Value emitScalarPrePostIncDec(const UnaryOperator *e, LValue lv,
                                       cir::UnaryOpKind kind, bool isPre);
 
diff --git a/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp b/clang/lib/CIR/Lowering/DirectToLLVM/LowerToLLVM.cpp
@@ -320,6 +320,20 @@ static mlir::LLVM::CallIntrinsicOp replaceOpWithCallLLVMIntrinsicOp(
   return callIntrinOp;
 }
 
+mlir::LogicalResult CIRToLLVMLLVMIntrinsicCallOpLowering::matchAndRewrite(
+    cir::LLVMIntrinsicCallOp op,
+    OpAdaptor adaptor,
+    mlir::ConversionPatternRewriter &rewriter) const {
+  mlir::Type llvmResTy =
+      getTypeConverter()->convertType(op->getResultTypes()[0]);
+  if (!llvmResTy)
+    return op.emitError("expected LLVM result type");
+  StringRef name = op.getIntrinsicName();
+  replaceOpWithCallLLVMIntrinsicOp(rewriter, op, "llvm." + name, llvmResTy,
+                                   adaptor.getOperands());
+  return mlir::success();
+}
+
 /// IntAttr visitor.
 mlir::Value CIRAttrToValue::visitCirAttr(cir::IntAttr intAttr) {
   mlir::Location loc = parentOp->getLoc();
diff --git a/clang/test/CIR/CodeGen/X86/sse-builtins.c b/clang/test/CIR/CodeGen/X86/sse-builtins.c
@@ -0,0 +1,23 @@
+// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse -fclangir -emit-cir -o %t.cir -Wall -Werror
+// RUN: FileCheck --check-prefix=CIR --input-file=%t.cir %s
+// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse -fclangir -emit-llvm -o %t.ll -Wall -Werror
+// RUN: FileCheck --check-prefixes=LLVM --input-file=%t.ll %s
+
+// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse -fno-signed-char -fclangir -emit-cir -o %t.cir -Wall -Werror
+// RUN: FileCheck --check-prefix=CIR --input-file=%t.cir %s
+// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse -fclangir -emit-llvm -o %t.ll -Wall -Werror
+// RUN: FileCheck --check-prefixes=LLVM --input-file=%t.ll %s
+
+// This test mimics clang/test/CodeGen/X86/sse-builtins.c, which eventually
+// CIR shall be able to support fully.
+
+#include <immintrin.h>
+
+
+void test_mm_prefetch(char const* p) {
+  // CIR-LABEL: test_mm_prefetch
+  // LLVM-LABEL: test_mm_prefetch
+  _mm_prefetch(p, 0);
+  // CIR: cir.prefetch read locality(0) %{{.*}} : !cir.ptr<!void>
+  // LLVM: call void @llvm.prefetch.p0(ptr {{.*}}, i32 0, i32 0, i32 1)
+}
diff --git a/clang/test/CIR/CodeGen/X86/sse2-builtins.c b/clang/test/CIR/CodeGen/X86/sse2-builtins.c
@@ -0,0 +1,23 @@
+// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fclangir -emit-cir -o %t.cir -Wall -Werror
+// RUN: FileCheck --check-prefixes=CIR-CHECK --input-file=%t.cir %s
+// RUN: %clang_cc1 -x c -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fno-signed-char -fclangir -emit-cir -o %t.cir -Wall -Werror
+// RUN: FileCheck --check-prefixes=CIR-CHECK --input-file=%t.cir %s
+
+// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fclangir -emit-llvm -o %t.ll -Wall -Werror
+// RUN: FileCheck --check-prefixes=LLVM-CHECK --input-file=%t.ll %s
+// RUN: %clang_cc1 -x c++ -flax-vector-conversions=none -ffreestanding %s -triple=x86_64-unknown-linux -target-feature +sse2 -fno-signed-char -fclangir -emit-llvm -o %t.ll -Wall -Werror
+// RUN: FileCheck --check-prefixes=LLVM-CHECK --input-file=%t.ll %s
+
+// This test mimics clang/test/CodeGen/X86/sse2-builtins.c, which eventually
+// CIR shall be able to support fully.
+
+#include <immintrin.h>
+
+
+void test_mm_clflush(void* A) {
+  // CIR-LABEL: test_mm_clflush
+  // LLVM-LABEL: teh
+  _mm_clflush(A);
+  // CIR-CHECK: {{%.*}} = cir.llvm.intrinsic "x86.sse2.clflush" {{%.*}} : (!cir.ptr<!void>) -> !void
+  // LLVM-CHECK: call void @llvm.x86.sse2.clflush(ptr {{%.*}})
+}
