aboutsummaryrefslogtreecommitdiff
path: root/tools
diff options
context:
space:
mode:
Diffstat (limited to 'tools')
-rw-r--r--tools/driver/NativeCompiler.cc197
-rw-r--r--tools/driver/NativeCompiler.h9
-rw-r--r--tools/driver/prepare-native-runtime.py168
3 files changed, 374 insertions, 0 deletions
diff --git a/tools/driver/NativeCompiler.cc b/tools/driver/NativeCompiler.cc
new file mode 100644
index 0000000..f2ea3dc
--- /dev/null
+++ b/tools/driver/NativeCompiler.cc
@@ -0,0 +1,197 @@
+#include "NativeCompiler.h"
+#include "cps/CodeGen.h"
+#include "clang/Basic/Diagnostic.h"
+#include "clang/Basic/DiagnosticOptions.h"
+#include "clang/Driver/Compilation.h"
+#include "clang/Driver/Driver.h"
+#include "clang/Driver/Job.h"
+#include "lld/Common/Driver.h"
+#include "llvm/IR/LegacyPassManager.h"
+#include "llvm/InitializePasses.h"
+#include "llvm/PassRegistry.h"
+#include "llvm/MC/TargetRegistry.h"
+#include "llvm/Support/Compression.h"
+#include "llvm/Support/FileSystem.h"
+#include "llvm/Support/FileUtilities.h"
+#include "llvm/Support/JSON.h"
+#include "llvm/Support/Path.h"
+#include "llvm/Support/TargetSelect.h"
+#include "llvm/Support/VirtualFileSystem.h"
+#include "llvm/Target/TargetMachine.h"
+#include "llvm/TargetParser/Triple.h"
+#include "llvm/Transforms/Utils/Cloning.h"
+#include <cstdlib>
+
+#ifdef _WIN32
+LLD_HAS_DRIVER(mingw)
+LLD_HAS_DRIVER(coff)
+#elif defined(__APPLE__)
+LLD_HAS_DRIVER(macho)
+#else
+LLD_HAS_DRIVER(elf)
+#endif
+
+namespace {
+
+#include "NativeRuntime.inc"
+
+struct TemporaryDirectory {
+ llvm::SmallString<256> Path;
+ ~TemporaryDirectory() {
+ if (!Path.empty()) llvm::sys::fs::remove_directories(Path);
+ }
+};
+
+bool fail(const llvm::Twine &Message) {
+ llvm::errs() << "cpsc: fatal error: " << Message << "\n";
+ return false;
+}
+
+class DriverDiagnostics : public clang::DiagnosticConsumer {
+ void HandleDiagnostic(clang::DiagnosticsEngine::Level Level,
+ const clang::Diagnostic &Info) override {
+ clang::DiagnosticConsumer::HandleDiagnostic(Level, Info);
+ llvm::SmallString<256> Message;
+ Info.FormatDiagnostic(Message);
+ llvm::errs() << "cpsc: " << (Level >= clang::DiagnosticsEngine::Error
+ ? "error: " : "warning: ") << Message << "\n";
+ }
+};
+
+} // namespace
+
+bool compileNative(const cps::CodeGen &CG, llvm::StringRef OutputPath) {
+ TemporaryDirectory Runtime;
+ if (auto EC = llvm::sys::fs::createUniqueDirectory("cpsc-native", Runtime.Path))
+ return fail("cannot create native compilation directory: " + EC.message());
+
+ auto Manifest = llvm::json::parse(NativeRuntimeManifest);
+ if (!Manifest) return fail(llvm::toString(Manifest.takeError()));
+ auto &Metadata = *Manifest->getAsObject();
+ llvm::SmallVector<uint8_t, 0> Data;
+ if (auto Error = llvm::compression::zlib::decompress(
+ llvm::ArrayRef(reinterpret_cast<const uint8_t *>(NativeRuntimeData),
+ sizeof(NativeRuntimeData) - 1), Data, NativeRuntimeSize))
+ return fail("cannot unpack native link resources: " + llvm::toString(std::move(Error)));
+
+ auto relocate = [&](llvm::StringRef Text) {
+ std::string Result = Text.str();
+ const std::string Marker = "@CPSC_RUNTIME@";
+ size_t Position = 0;
+ while ((Position = Result.find(Marker, Position)) != std::string::npos) {
+ Result.replace(Position, Marker.size(), Runtime.Path.str());
+ Position += Runtime.Path.size();
+ }
+ return Result;
+ };
+ for (const auto &Entry : *Metadata.getArray("files")) {
+ const auto &File = *Entry.getAsObject();
+ llvm::SmallString<256> Path(Runtime.Path);
+ llvm::sys::path::append(Path, *File.getString("path"));
+ if (auto EC = llvm::sys::fs::create_directories(llvm::sys::path::parent_path(Path)))
+ return fail("cannot prepare native link resources: " + EC.message());
+ std::error_code EC;
+ llvm::raw_fd_ostream Output(Path, EC);
+ if (EC) return fail("cannot open native link resource: " + EC.message());
+ llvm::StringRef Contents(reinterpret_cast<const char *>(Data.data()) +
+ *File.getInteger("offset"), *File.getInteger("size"));
+ if (*File.getBoolean("script")) Output << relocate(Contents);
+ else Output << Contents;
+ Output.close();
+ if (Output.has_error()) {
+ auto Error = Output.error();
+ Output.clear_error();
+ return fail("cannot write native link resource: " + Error.message());
+ }
+ }
+
+ if (llvm::InitializeNativeTarget() || llvm::InitializeNativeTargetAsmPrinter())
+ return fail("cannot initialize the native LLVM target");
+ auto &Registry = *llvm::PassRegistry::getPassRegistry();
+ llvm::initializeCore(Registry);
+ llvm::initializeCodeGen(Registry);
+ llvm::initializeScalarOpts(Registry);
+ llvm::initializeVectorization(Registry);
+ llvm::initializeTransformUtils(Registry);
+ llvm::initializeInstCombine(Registry);
+ llvm::initializeTarget(Registry);
+ llvm::Triple Triple(*Metadata.getString("target"));
+ std::string Error;
+ const auto *Target = llvm::TargetRegistry::lookupTarget(Triple, Error);
+ if (!Target) return fail(Error);
+ std::unique_ptr<llvm::TargetMachine> Machine(Target->createTargetMachine(
+ Triple, "generic", "", llvm::TargetOptions(), llvm::Reloc::PIC_));
+ if (!Machine) return fail("cannot create the native LLVM target machine");
+ auto Module = llvm::CloneModule(CG.getModule());
+ Module->setTargetTriple(Triple);
+ Module->setDataLayout(Machine->createDataLayout());
+ llvm::SmallString<256> ObjectPath(Runtime.Path);
+ llvm::sys::path::append(ObjectPath, "program.o");
+ std::error_code EC;
+ llvm::raw_fd_ostream ObjectOutput(ObjectPath, EC);
+ if (EC) return fail("cannot open native object file: " + EC.message());
+ llvm::legacy::PassManager Passes;
+ if (Machine->addPassesToEmitFile(Passes, ObjectOutput, nullptr,
+ llvm::CodeGenFileType::ObjectFile))
+ return fail("native target cannot emit object files");
+ Passes.run(*Module);
+ ObjectOutput.close();
+ if (ObjectOutput.has_error()) {
+ auto Error = ObjectOutput.error();
+ ObjectOutput.clear_error();
+ return fail("cannot write native object file: " + Error.message());
+ }
+
+ // Link beside the destination so the final rename stays on one filesystem.
+ llvm::SmallString<256> StagedPath;
+ int FD;
+ if (auto EC = llvm::sys::fs::createUniqueFile(OutputPath + ".tmp-%%%%%%", FD, StagedPath))
+ return fail("cannot open native output file: " + EC.message());
+ llvm::FileRemover StagedOutput(StagedPath);
+ llvm::raw_fd_ostream CloseFD(FD, true);
+ CloseFD.close();
+
+ // All link inputs are explicit. Clang's driver provides job validation and
+ // platform command construction; no compiler subprocess is executed.
+ clang::DiagnosticOptions Options;
+ DriverDiagnostics Printer;
+ clang::DiagnosticsEngine Diagnostics(new clang::DiagnosticIDs, Options, &Printer, false);
+ clang::driver::Driver Driver("cpsc", Triple.str(), Diagnostics);
+ // The standard linker flavor constructs a job without checking for an
+ // installed linker. Its platform arguments are passed to embedded LLD.
+ std::vector<std::string> Args = {"cpsc", "--driver-mode=g++", "-nostdlib", "-nostdlib++",
+ "-fuse-ld=ld", "-o", std::string(StagedPath), std::string(ObjectPath)};
+ for (const auto &Arg : *Metadata.getArray("arguments")) {
+ Args.push_back("-Xlinker");
+ Args.push_back(relocate(*Arg.getAsString()));
+ }
+ std::vector<const char *> Argv;
+ for (const auto &Arg : Args) Argv.push_back(Arg.c_str());
+ Driver.setCheckInputsExist(false);
+ std::unique_ptr<clang::driver::Compilation> Compilation(Driver.BuildCompilation(Argv));
+ if (!Compilation || Diagnostics.hasErrorOccurred()) return false;
+ if (Compilation->getJobs().size() != 1)
+ return fail("expected one embedded linker job");
+ const auto &Job = *Compilation->getJobs().begin();
+ llvm::SmallVector<const char *, 64> LinkArgs;
+#ifdef _WIN32
+ LinkArgs.push_back("ld.lld");
+ const lld::DriverDef Linkers[] = {{lld::MinGW, &lld::mingw::link}, {lld::WinLink, &lld::coff::link}};
+#elif defined(__APPLE__)
+ LinkArgs.push_back("ld64.lld");
+ const lld::DriverDef Linkers[] = {{lld::Darwin, &lld::macho::link}};
+#else
+ LinkArgs.push_back("ld.lld");
+ const lld::DriverDef Linkers[] = {{lld::Gnu, &lld::elf::link}};
+#endif
+ LinkArgs.append(Job.getArguments().begin(), Job.getArguments().end());
+ auto Result = lld::lldMain(LinkArgs, llvm::outs(), llvm::errs(), Linkers);
+ if (!Result.canRunAgain) std::exit(1);
+ if (Result.retCode != 0) return fail("native linking failed");
+ if (auto EC = llvm::sys::fs::setPermissions(StagedPath,
+ llvm::sys::fs::all_read | llvm::sys::fs::owner_write | llvm::sys::fs::all_exe))
+ return fail("cannot set executable permissions: " + EC.message());
+ if (auto EC = llvm::sys::fs::rename(StagedPath, OutputPath))
+ return fail("cannot save native output: " + EC.message());
+ return true;
+}
diff --git a/tools/driver/NativeCompiler.h b/tools/driver/NativeCompiler.h
new file mode 100644
index 0000000..17933e9
--- /dev/null
+++ b/tools/driver/NativeCompiler.h
@@ -0,0 +1,9 @@
+#pragma once
+
+#include "llvm/ADT/StringRef.h"
+
+namespace cps { class CodeGen; }
+
+// Generate a host executable without an external compiler, linker or SDK.
+// Failure leaves any existing output intact and reports a diagnostic.
+bool compileNative(const cps::CodeGen &CG, llvm::StringRef OutputPath);
diff --git a/tools/driver/prepare-native-runtime.py b/tools/driver/prepare-native-runtime.py
new file mode 100644
index 0000000..8463eba
--- /dev/null
+++ b/tools/driver/prepare-native-runtime.py
@@ -0,0 +1,168 @@
+#!/usr/bin/env python3
+"""Freeze the build toolchain's native link inputs into the compiler."""
+
+import argparse
+import json
+from pathlib import Path
+import re
+import shlex
+import subprocess
+import zlib
+
+
+def main():
+ parser = argparse.ArgumentParser(description=__doc__)
+ parser.add_argument("--compiler", required=True)
+ parser.add_argument("--output", required=True)
+ parser.add_argument("--platform", choices=("linux", "windows", "macos"), required=True)
+ parser.add_argument("--flag", action="append", default=[])
+ args = parser.parse_args()
+ output = Path(args.output)
+ output.parent.mkdir(parents=True, exist_ok=True)
+
+ def compiler(*options):
+ return subprocess.run([args.compiler, *args.flag, *options], check=True,
+ capture_output=True, text=True)
+
+ # An actual object lets the driver validate its inputs without linking a probe.
+ source = output.parent / "native-probe.cc"
+ obj = output.parent / "native-probe.o"
+ binary = output.parent / "native-probe.out"
+ source.write_text("int main() { return 0; }\n")
+ compiler("-c", str(source), "-o", str(obj))
+ static = [] if args.platform == "macos" else (
+ ["-static"] if args.platform == "windows" else
+ ["-static-libstdc++", "-static-libgcc"])
+ trace = compiler("-###", "-fuse-ld=lld", *static, str(obj), "-o", str(binary))
+ commands = [shlex.split(line.strip()) for line in trace.stderr.splitlines()
+ if line.lstrip().startswith('"')]
+ if len(commands) != 1:
+ raise RuntimeError("Expected one native linker command: " + trace.stderr)
+ command = commands[0][1:]
+ directories = [Path(arg[2:]) for arg in command if arg.startswith("-L")]
+ target = compiler("-dumpmachine").stdout.strip()
+ files = {}
+ scripts = set()
+ arguments = []
+ root = "@CPSC_RUNTIME@"
+
+ def resolve(name, static_only=False):
+ if name.startswith(":"):
+ names = [name[1:]]
+ elif args.platform == "macos":
+ names = ["lib" + name + suffix for suffix in (".tbd", ".dylib", ".a")]
+ elif static_only:
+ names = ["lib" + name + ".a"]
+ else:
+ names = ["lib" + name + suffix for suffix in (".dll.a", ".so", ".a")]
+ for directory in directories:
+ for filename in names:
+ path = directory / filename
+ if path.is_file():
+ return path
+ for filename in names:
+ found = Path(compiler("-print-file-name=" + filename).stdout.strip())
+ if found.is_file():
+ return found
+ raise RuntimeError("Cannot find native link input " + name)
+
+ sdk = None
+ if args.platform == "macos":
+ if "-syslibroot" not in command:
+ raise RuntimeError("A macOS SDK is required to prepare native linking")
+ sdk = Path(command[command.index("-syslibroot") + 1])
+ directories += [sdk / "usr/lib"]
+ # System dylibs live in the dyld cache on users' machines. Embed their
+ # SDK link interfaces, not runtime dylibs or SDK headers.
+ for path in sorted((sdk / "usr/lib").rglob("*.tbd")):
+ files["sdk/" + path.relative_to(sdk).as_posix()] = path.read_bytes()
+
+ mapped = {}
+
+ def resource(path):
+ path = path.resolve()
+ if path in mapped:
+ return root + "/" + mapped[path]
+ if sdk and path.is_relative_to(sdk.resolve()):
+ name = "sdk/" + path.relative_to(sdk.resolve()).as_posix()
+ else:
+ name = "files/" + str(len(mapped)) + "/" + path.name
+ mapped[path] = name
+ data = path.read_bytes()
+ if b"GNU ld script" in data[:256]:
+ text = re.sub(r"/\*.*?\*/", "", data.decode(), flags=re.S)
+ # GNU libc/libm linker scripts contain absolute paths and nested
+ # GROUP/AS_NEEDED inputs. Relocate every referenced link input.
+ pattern = r"(?:/[^\s()]+|lib[\w.+-]+\.(?:so(?:\.[\d]+)*|a))"
+
+ def relocate(match):
+ token = match.group()
+ dependency = Path(token)
+ if not dependency.is_file():
+ dependency = resolve(":" + token)
+ return resource(dependency)
+
+ data = re.sub(pattern, relocate, text).encode()
+ scripts.add(name)
+ files[name] = data
+ return root + "/" + name
+
+ static_only = args.platform == "windows"
+ i = 0
+ while i < len(command):
+ arg = command[i]
+ i += 1
+ if arg == "-o":
+ i += 1
+ elif arg == str(obj):
+ pass
+ elif arg == "-L":
+ i += 1
+ elif arg.startswith("-L"):
+ pass
+ elif arg in ("-dynamic-linker", "--dynamic-linker"):
+ # The interpreter must remain an operating-system path.
+ arguments += [arg, command[i]]
+ i += 1
+ elif arg == "-syslibroot":
+ arguments += [arg, root + "/sdk"]
+ i += 1
+ elif arg in ("-Bstatic", "-Bdynamic"):
+ static_only = arg == "-Bstatic"
+ arguments.append(arg)
+ elif arg.startswith("-l"):
+ arguments.append(resource(resolve(arg[2:], static_only)))
+ elif Path(arg).is_file():
+ arguments.append(resource(Path(arg)))
+ else:
+ arguments.append(arg)
+
+ blob = bytearray()
+ manifest = {"target": target, "arguments": arguments, "files": []}
+ for name, data in files.items():
+ manifest["files"].append({"path": name, "offset": len(blob),
+ "size": len(data), "script": name in scripts})
+ blob.extend(data)
+ compressed = zlib.compress(blob, 9)
+ # Small separate literals avoid MSVC's individual string literal limit.
+ literals = "\n".join('"' + "".join(f"\\x{b:02x}" for b in compressed[i:i+1024]) + '"'
+ for i in range(0, len(compressed), 1024))
+ content = ("// Generated native link resources; do not edit.\n"
+ f"static constexpr size_t NativeRuntimeSize = {len(blob)};\n"
+ "static const char NativeRuntimeData[] =\n" + literals + ";\n"
+ 'static const char NativeRuntimeManifest[] = R"CPSC(\n' +
+ json.dumps(manifest) + '\n)CPSC";\n')
+ if not output.exists() or output.read_text() != content:
+ output.write_text(content)
+ # Track the actual toolchain inputs, not only the generator script.
+ dependencies = [str(path) for path in mapped]
+ if sdk:
+ dependencies += [str(path) for path in (sdk / "usr/lib").rglob("*.tbd")]
+ (output.parent / "native-runtime-inputs.cmake").write_text(
+ "set_property(DIRECTORY APPEND PROPERTY CMAKE_CONFIGURE_DEPENDS\n" +
+ "\n".join(' [==[' + path + ']==]' for path in dependencies) + ")\n")
+ print(f"Embedded {len(files)} native link inputs ({len(compressed)} compressed bytes)")
+
+
+if __name__ == "__main__":
+ main()
generated by cgit v1.3.1-4-g156e (git 2.54.0) at 2026-11-15 09:21:19 +0000