[clang] [CIR] Add X86 prefetch builtins (PR #168051)

[Bruno Cardoso Lopes via cfe-commits]

================
@@ -33,6 +38,32 @@ static mlir::Value emitIntrinsicCallOp(CIRGenFunction &cgf, const CallExpr *e,
       .getResult();
 }
 
+static mlir::Value emitPrefetch(CIRGenFunction &cgf, unsigned builtinID,
+                                const CallExpr *e,
+                                const SmallVector<mlir::Value> &ops) {
+  CIRGenBuilderTy &builder = cgf.getBuilder();
+  mlir::Location location = cgf.getLoc(e->getExprLoc());
+  mlir::Type voidTy = builder.getVoidTy();
+  mlir::Value address = builder.createPtrBitcast(ops[0], voidTy);
+  bool isWrite{};
+  int locality{};
+
+  assert(builtinID == X86::BI_mm_prefetch || builtinID == X86::BI_m_prefetchw ||
+         builtinID == X86::BI_m_prefetch && "Expected prefetch builtin");
+
+  if (builtinID == X86::BI_mm_prefetch) {
+    int hint = getIntValueFromConstOp(ops[1]);
+    isWrite = (hint >> 2) & 0x1;
+    locality = hint & 0x3;
+  } else {
+    isWrite = (builtinID == X86::BI_m_prefetchw);
+    locality = 0x3;
+  }
+
+  cir::PrefetchOp::create(builder, location, address, locality, isWrite);
+  return {};
----------------
bcardosolopes wrote:

Why not emit a PoisonValue/UndefValue here? The result is not going to be used anyways and will be later eliminated. A comment explaining probably helps. I'm against using optional because `mlir::Value` type already incorporates that semantics and is a weird shortcut we'd be taking.

https://github.com/llvm/llvm-project/pull/168051