[clang] [CIR][AMDGPU] Route printf in device code through the OpenCL runtime (PR #226435)

[Mariya Podchishchaeva via cfe-commits]

================
@@ -1096,3 +1096,45 @@ CIRGenFunction::emitAMDGPUBuiltinExpr(unsigned builtinId,
     return std::nullopt;
   }
 }
+
+// Emit AMDGPU printf CIR stand-in function call. This stand-in function call is
+// lowered to the appropriate call structure during LLVM IR lowering.
+mlir::Value
+CIRGenFunction::emitAMDGPUDevicePrintfCallExpr(const CallExpr *expr) {
+  assert(cgm.getTriple().isAMDGCN() ||
+         (cgm.getTriple().isSPIRV() &&
+          cgm.getTriple().getVendor() == llvm::Triple::AMD));
+  assert(expr->getBuiltinCallee() == Builtin::BIprintf ||
+         expr->getBuiltinCallee() == Builtin::BI__builtin_printf);
+  assert(expr->getNumArgs() >= 1);
+
+  const FunctionProtoType *funcPrototype =
+      expr->getDirectCallee()->getType()->getAs<FunctionProtoType>();
+  CallArgList args;
+  emitCallArgs(args, funcPrototype, expr->arguments(), expr->getDirectCallee());
+
+  mlir::Location loc = getLoc(expr->getBeginLoc());
+
+  // We don't know how to emit non-scalar varargs.
+  bool hasNonScalar = llvm::any_of(args, [&](const CallArg &a) {
+    return a.hasLValue() || !a.getKnownRValue().isScalar();
+  });
+  if (hasNonScalar) {
+    cgm.errorUnsupported(expr, "non-scalar args to printf");
+    return builder.getConstInt(loc, builder.getSInt32Ty(), 0);
+  }
+
+  llvm::SmallVector<mlir::Value, 8> callArgs;
+  for (const CallArg &a : args)
+    callArgs.push_back(a.getKnownRValue().getValue());
+
+  // int __cir_amdgpu_printf(char *format, ...);
+  auto fnTy = cir::FuncType::get({cir::PointerType::get(builder.getSInt8Ty())},
+                                 builder.getSInt32Ty(),
+                                 /*isVarArg=*/true);
+  cir::FuncOp fn = cgm.createRuntimeFunction(fnTy, "__cir_amdgpu_printf");
----------------
Fznamznon wrote:

Maybe that shouldn't be AMDGPU specific. Can it be just printf op? Or it can be "offloading printf", I'm sure other offloading targets might perform some workarounds for printf support. For example SPIR-V

https://github.com/llvm/llvm-project/pull/226435